Skip to content

C
ccan
Commit 7581be1b authored by Rusty Russell's avatar Rusty Russell
Browse files
Options

tdb: delete from CCAN. 

This was an early experiment in putting code into CCAN, but TDB is a public
library distributed as part of the Samba sources; there is no reason to
publish it here (especially now the unit tests are also in Samba).
parent 98b8ada2
Hide whitespace changes
Inline Side-by-side
Showing with 0 additions and 13453 deletions
ccan/tdb/LICENSE deleted 120000 → 0
View file @ 98b8ada2
../../licenses/LGPL-3
\ No newline at end of file
ccan/tdb/_info deleted 100644 → 0
View file @ 98b8ada2
#include <string.h>
#include <stdio.h>
/**
* tdb - The trivial (transactional) database
*
* The tdb module provides an efficient keyword data mapping (usually
* within a file). It supports transactions, so the contents of the
* database is reliable even across crashes.
*
* Example:
* #include <ccan/tdb/tdb.h>
* #include <ccan/str/str.h>
* #include <err.h>
* #include <stdio.h>
*
* static void usage(const char *argv0)
* {
* errx(1, "Usage: %s fetch <dbfile> <key>\n"
* "OR %s store <dbfile> <key> <data>", argv0, argv0);
* }
*
* int main(int argc, char *argv[])
* {
* struct tdb_context *tdb;
* TDB_DATA key, value;
*
* if (argc < 4)
* usage(argv[0]);
*
* tdb = tdb_open(argv[2], 1024, TDB_DEFAULT, O_CREAT|O_RDWR,
* 0600);
* if (!tdb)
* err(1, "Opening %s", argv[2]);
*
* key.dptr = (void *)argv[3];
* key.dsize = strlen(argv[3]);
*
* if (streq(argv[1], "fetch")) {
* if (argc != 4)
* usage(argv[0]);
* value = tdb_fetch(tdb, key);
* if (!value.dptr)
* errx(1, "fetch %s: %s",
* argv[3], tdb_errorstr(tdb));
* printf("%.*s\n", value.dsize, (char *)value.dptr);
* free(value.dptr);
* } else if (streq(argv[1], "store")) {
* if (argc != 5)
* usage(argv[0]);
* value.dptr = (void *)argv[4];
* value.dsize = strlen(argv[4]);
* if (tdb_store(tdb, key, value, 0) != 0)
* errx(1, "store %s: %s",
* argv[3], tdb_errorstr(tdb));
* } else
* usage(argv[0]);
*
* return 0;
* }
*
* Maintainer: Rusty Russell <rusty@rustcorp.com.au>
*
* Author: Andrew Tridgell, Jeremy Allison, Rusty Russell
*
* License: LGPL (v3 or any later version)
*
* Ccanlint:
* // valgrind breaks fcntl locks.
* tests_pass_valgrind test/run-open-during-transaction.c:FAIL
* tests_pass_valgrind_noleaks test/run-die-during-transaction.c:FAIL
*/
int main(int argc, char *argv[])
{
if (argc != 2)
return 1;
if (strcmp(argv[1], "depends") == 0) {
printf("ccan/compiler\n");
return 0;
}
return 1;
}
ccan/tdb/check.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
trivial database library
Copyright (C) Rusty Russell 2009
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
#include <limits.h>
/* Since we opened it, these shouldn't fail unless it's recent corruption. */
static bool tdb_check_header(struct tdb_context *tdb, tdb_off_t *recovery)
{
struct tdb_header hdr;
uint32_t h1, h2;
if (tdb->methods->tdb_read(tdb, 0, &hdr, sizeof(hdr), 0) == -1)
return false;
if (strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0)
goto corrupt;
CONVERT(hdr);
if (hdr.version != TDB_VERSION)
goto corrupt;
if (hdr.rwlocks != 0 && hdr.rwlocks != TDB_HASH_RWLOCK_MAGIC)
goto corrupt;
tdb_header_hash(tdb, &h1, &h2);
if (hdr.magic1_hash && hdr.magic2_hash &&
(hdr.magic1_hash != h1 || hdr.magic2_hash != h2))
goto corrupt;
if (hdr.hash_size == 0)
goto corrupt;
if (hdr.hash_size != tdb->header.hash_size)
goto corrupt;
if (hdr.recovery_start != 0 &&
hdr.recovery_start < TDB_DATA_START(tdb->header.hash_size))
goto corrupt;
*recovery = hdr.recovery_start;
return true;
corrupt:
tdb->ecode = TDB_ERR_CORRUPT;
TDB_LOG((tdb, TDB_DEBUG_ERROR, "Header is corrupt\n"));
return false;
}
/* Generic record header check. */
static bool tdb_check_record(struct tdb_context *tdb,
tdb_off_t off,
const struct tdb_record *rec)
{
tdb_off_t tailer;
/* Check rec->next: 0 or points to record offset, aligned. */
if (rec->next > 0 && rec->next < TDB_DATA_START(tdb->header.hash_size)){
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Record offset %d too small next %d\n",
off, rec->next));
goto corrupt;
}
if (rec->next + sizeof(*rec) < rec->next) {
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Record offset %d too large next %d\n",
off, rec->next));
goto corrupt;
}
if ((rec->next % TDB_ALIGNMENT) != 0) {
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Record offset %d misaligned next %d\n",
off, rec->next));
goto corrupt;
}
if (tdb->methods->tdb_oob(tdb, rec->next+sizeof(*rec), 0))
goto corrupt;
/* Check rec_len: similar to rec->next, implies next record. */
if ((rec->rec_len % TDB_ALIGNMENT) != 0) {
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Record offset %d misaligned length %d\n",
off, rec->rec_len));
goto corrupt;
}
/* Must fit tailer. */
if (rec->rec_len < sizeof(tailer)) {
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Record offset %d too short length %d\n",
off, rec->rec_len));
goto corrupt;
}
/* OOB allows "right at the end" access, so this works for last rec. */
if (tdb->methods->tdb_oob(tdb, off+sizeof(*rec)+rec->rec_len, 0))
goto corrupt;
/* Check tailer. */
if (tdb_ofs_read(tdb, off+sizeof(*rec)+rec->rec_len-sizeof(tailer),
&tailer) == -1)
goto corrupt;
if (tailer != sizeof(*rec) + rec->rec_len) {
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Record offset %d invalid tailer\n", off));
goto corrupt;
}
return true;
corrupt:
tdb->ecode = TDB_ERR_CORRUPT;
return false;
}
/* Grab some bytes: may copy if can't use mmap.
Caller has already done bounds check. */
static TDB_DATA get_bytes(struct tdb_context *tdb,
tdb_off_t off, tdb_len_t len)
{
TDB_DATA d;
d.dsize = len;
if (tdb->transaction == NULL && tdb->map_ptr != NULL)
d.dptr = (unsigned char *)tdb->map_ptr + off;
else
d.dptr = tdb_alloc_read(tdb, off, d.dsize);
return d;
}
/* Frees data if we're not able to simply use mmap. */
static void put_bytes(struct tdb_context *tdb, TDB_DATA d)
{
if (tdb->transaction == NULL && tdb->map_ptr != NULL)
return;
free(d.dptr);
}
/* We use the excellent Jenkins lookup3 hash; this is based on hash_word2.
* See: http://burtleburtle.net/bob/c/lookup3.c
*/
#define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
static void hash(uint32_t key, uint32_t *pc, uint32_t *pb)
{
uint32_t a,b,c;
/* Set up the internal state */
a = b = c = 0xdeadbeef + *pc;
c += *pb;
a += key;
c ^= b; c -= rot(b,14);
a ^= c; a -= rot(c,11);
b ^= a; b -= rot(a,25);
c ^= b; c -= rot(b,16);
a ^= c; a -= rot(c,4);
b ^= a; b -= rot(a,14);
c ^= b; c -= rot(b,24);
*pc=c; *pb=b;
}
/*
We want to check that all free records are in the free list
(only once), and all free list entries are free records. Similarly
for each hash chain of used records.
Doing that naively (without walking hash chains, since we want to be
linear) means keeping a list of records which have been seen in each
hash chain, and another of records pointed to (ie. next pointers
from records and the initial hash chain heads). These two lists
should be equal. This will take 8 bytes per record, and require
sorting at the end.
So instead, we record each offset in a bitmap such a way that
recording it twice will cancel out. Since each offset should appear
exactly twice, the bitmap should be zero at the end.
The approach was inspired by Bloom Filters (see Wikipedia). For
each value, we flip K bits in a bitmap of size N. The number of
distinct arrangements is:
N! / (K! * (N-K)!)
Of course, not all arrangements are actually distinct, but testing
shows this formula to be close enough.
So, if K == 8 and N == 256, the probability of two things flipping the same
bits is 1 in 409,663,695,276,000.
Given that ldb uses a hash size of 10000, using 32 bytes per hash chain
(320k) seems reasonable.
*/
#define NUM_HASHES 8
#define BITMAP_BITS 256
static void bit_flip(unsigned char bits[], unsigned int idx)
{
bits[idx / CHAR_BIT] ^= (1 << (idx % CHAR_BIT));
}
/* We record offsets in a bitmap for the particular chain it should be in. */
static void record_offset(unsigned char bits[], tdb_off_t off)
{
uint32_t h1 = off, h2 = 0;
unsigned int i;
/* We get two good hash values out of jhash2, so we use both. Then
* we keep going to produce further hash values. */
for (i = 0; i < NUM_HASHES / 2; i++) {
hash(off, &h1, &h2);
bit_flip(bits, h1 % BITMAP_BITS);
bit_flip(bits, h2 % BITMAP_BITS);
h2++;
}
}
/* Check that an in-use record is valid. */
static bool tdb_check_used_record(struct tdb_context *tdb,
tdb_off_t off,
const struct tdb_record *rec,
unsigned char **hashes,
int (*check)(TDB_DATA, TDB_DATA, void *),
void *private_data)
{
TDB_DATA key, data;
if (!tdb_check_record(tdb, off, rec))
return false;
/* key + data + tailer must fit in record */
if (rec->key_len + rec->data_len + sizeof(tdb_off_t) > rec->rec_len) {
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Record offset %d too short for contents\n", off));
return false;
}
key = get_bytes(tdb, off + sizeof(*rec), rec->key_len);
if (!key.dptr)
return false;
if (tdb->hash_fn(&key) != rec->full_hash) {
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Record offset %d has incorrect hash\n", off));
goto fail_put_key;
}
/* Mark this offset as a known value for this hash bucket. */
record_offset(hashes[BUCKET(rec->full_hash)+1], off);
/* And similarly if the next pointer is valid. */
if (rec->next)
record_offset(hashes[BUCKET(rec->full_hash)+1], rec->next);
/* If they supply a check function and this record isn't dead,
get data and feed it. */
if (check && rec->magic != TDB_DEAD_MAGIC) {
data = get_bytes(tdb, off + sizeof(*rec) + rec->key_len,
rec->data_len);
if (!data.dptr)
goto fail_put_key;
if (check(key, data, private_data) == -1)
goto fail_put_data;
put_bytes(tdb, data);
}
put_bytes(tdb, key);
return true;
fail_put_data:
put_bytes(tdb, data);
fail_put_key:
put_bytes(tdb, key);
return false;
}
/* Check that an unused record is valid. */
static bool tdb_check_free_record(struct tdb_context *tdb,
tdb_off_t off,
const struct tdb_record *rec,
unsigned char **hashes)
{
if (!tdb_check_record(tdb, off, rec))
return false;
/* Mark this offset as a known value for the free list. */
record_offset(hashes[0], off);
/* And similarly if the next pointer is valid. */
if (rec->next)
record_offset(hashes[0], rec->next);
return true;
}
/* Slow, but should be very rare. */
size_t tdb_dead_space(struct tdb_context *tdb, tdb_off_t off)
{
size_t len;
for (len = 0; off + len < tdb->map_size; len++) {
char c;
if (tdb->methods->tdb_read(tdb, off, &c, 1, 0))
return 0;
if (c != 0 && c != 0x42)
break;
}
return len;
}
int tdb_check(struct tdb_context *tdb,
int (*check)(TDB_DATA key, TDB_DATA data, void *private_data),
void *private_data)
{
unsigned int h;
unsigned char **hashes;
tdb_off_t off, recovery_start;
struct tdb_record rec;
bool found_recovery = false;
tdb_len_t dead;
bool locked;
/* Read-only databases use no locking at all: it's best-effort.
* We may have a write lock already, so skip that case too. */
if (tdb->read_only || tdb->allrecord_lock.count != 0) {
locked = false;
} else {
if (tdb_lockall_read(tdb) == -1)
return -1;
locked = true;
}
/* Make sure we know true size of the underlying file. */
tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
/* Header must be OK: also gets us the recovery ptr, if any. */
if (!tdb_check_header(tdb, &recovery_start))
goto unlock;
/* We should have the whole header, too. */
if (tdb->map_size < TDB_DATA_START(tdb->header.hash_size)) {
tdb->ecode = TDB_ERR_CORRUPT;
TDB_LOG((tdb, TDB_DEBUG_ERROR, "File too short for hashes\n"));
goto unlock;
}
/* One big malloc: pointers then bit arrays. */
hashes = (unsigned char **)calloc(
1, sizeof(hashes[0]) * (1+tdb->header.hash_size)
+ BITMAP_BITS / CHAR_BIT * (1+tdb->header.hash_size));
if (!hashes) {
tdb->ecode = TDB_ERR_OOM;
goto unlock;
}
/* Initialize pointers */
hashes[0] = (unsigned char *)(&hashes[1+tdb->header.hash_size]);
for (h = 1; h < 1+tdb->header.hash_size; h++)
hashes[h] = hashes[h-1] + BITMAP_BITS / CHAR_BIT;
/* Freelist and hash headers are all in a row: read them. */
for (h = 0; h < 1+tdb->header.hash_size; h++) {
if (tdb_ofs_read(tdb, FREELIST_TOP + h*sizeof(tdb_off_t),
&off) == -1)
goto free;
if (off)
record_offset(hashes[h], off);
}
/* For each record, read it in and check it's ok. */
for (off = TDB_DATA_START(tdb->header.hash_size);
off < tdb->map_size;
off += sizeof(rec) + rec.rec_len) {
if (tdb->methods->tdb_read(tdb, off, &rec, sizeof(rec),
DOCONV()) == -1)
goto free;
switch (rec.magic) {
case TDB_MAGIC:
case TDB_DEAD_MAGIC:
if (!tdb_check_used_record(tdb, off, &rec, hashes,
check, private_data))
goto free;
break;
case TDB_FREE_MAGIC:
if (!tdb_check_free_record(tdb, off, &rec, hashes))
goto free;
break;
/* If we crash after ftruncate, we can get zeroes or fill. */
case TDB_RECOVERY_INVALID_MAGIC:
case 0x42424242:
if (recovery_start == off) {
found_recovery = true;
break;
}
dead = tdb_dead_space(tdb, off);
if (dead < sizeof(rec))
goto corrupt;
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Dead space at %d-%d (of %u)\n",
off, off + dead, tdb->map_size));
rec.rec_len = dead - sizeof(rec);
break;
case TDB_RECOVERY_MAGIC:
if (recovery_start != off) {
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Unexpected recovery record at offset %d\n",
off));
goto free;
}
found_recovery = true;
break;
default: ;
corrupt:
tdb->ecode = TDB_ERR_CORRUPT;
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Bad magic 0x%x at offset %d\n",
rec.magic, off));
goto free;
}
}
/* Now, hashes should all be empty: each record exists and is referred
* to by one other. */
for (h = 0; h < 1+tdb->header.hash_size; h++) {
unsigned int i;
for (i = 0; i < BITMAP_BITS / CHAR_BIT; i++) {
if (hashes[h][i] != 0) {
tdb->ecode = TDB_ERR_CORRUPT;
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Hashes do not match records\n"));
goto free;
}
}
}
/* We must have found recovery area if there was one. */
if (recovery_start != 0 && !found_recovery) {
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Expected a recovery area at %u\n",
recovery_start));
goto free;
}
free(hashes);
if (locked) {
tdb_unlockall_read(tdb);
}
return 0;
free:
free(hashes);
unlock:
if (locked) {
tdb_unlockall_read(tdb);
}
return -1;
}
ccan/tdb/dump.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
trivial database library
Copyright (C) Andrew Tridgell 1999-2005
Copyright (C) Paul `Rusty' Russell 2000
Copyright (C) Jeremy Allison 2000-2003
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
static tdb_off_t tdb_dump_record(struct tdb_context *tdb, int hash,
tdb_off_t offset)
{
struct tdb_record rec;
tdb_off_t tailer_ofs, tailer;
if (tdb->methods->tdb_read(tdb, offset, (char *)&rec,
sizeof(rec), DOCONV()) == -1) {
printf("ERROR: failed to read record at %u\n", offset);
return 0;
}
printf(" rec: hash=%d offset=0x%08x next=0x%08x rec_len=%d "
"key_len=%d data_len=%d full_hash=0x%x magic=0x%x\n",
hash, offset, rec.next, rec.rec_len, rec.key_len, rec.data_len,
rec.full_hash, rec.magic);
tailer_ofs = offset + sizeof(rec) + rec.rec_len - sizeof(tdb_off_t);
if (tdb_ofs_read(tdb, tailer_ofs, &tailer) == -1) {
printf("ERROR: failed to read tailer at %u\n", tailer_ofs);
return rec.next;
}
if (tailer != rec.rec_len + sizeof(rec)) {
printf("ERROR: tailer does not match record! tailer=%u totalsize=%u\n",
(unsigned int)tailer, (unsigned int)(rec.rec_len + sizeof(rec)));
}
return rec.next;
}
static int tdb_dump_chain(struct tdb_context *tdb, int i)
{
tdb_off_t rec_ptr, top;
top = TDB_HASH_TOP(i);
if (tdb_lock(tdb, i, F_WRLCK) != 0)
return -1;
if (tdb_ofs_read(tdb, top, &rec_ptr) == -1)
return tdb_unlock(tdb, i, F_WRLCK);
if (rec_ptr)
printf("hash=%d\n", i);
while (rec_ptr) {
rec_ptr = tdb_dump_record(tdb, i, rec_ptr);
}
return tdb_unlock(tdb, i, F_WRLCK);
}
void tdb_dump_all(struct tdb_context *tdb)
{
int i;
for (i=0;i<tdb->header.hash_size;i++) {
tdb_dump_chain(tdb, i);
}
printf("freelist:\n");
tdb_dump_chain(tdb, -1);
}
int tdb_printfreelist(struct tdb_context *tdb)
{
int ret;
long total_free = 0;
tdb_off_t offset, rec_ptr;
struct tdb_record rec;
if ((ret = tdb_lock(tdb, -1, F_WRLCK)) != 0)
return ret;
offset = FREELIST_TOP;
/* read in the freelist top */
if (tdb_ofs_read(tdb, offset, &rec_ptr) == -1) {
tdb_unlock(tdb, -1, F_WRLCK);
return 0;
}
printf("freelist top=[0x%08x]\n", rec_ptr );
while (rec_ptr) {
if (tdb->methods->tdb_read(tdb, rec_ptr, (char *)&rec,
sizeof(rec), DOCONV()) == -1) {
tdb_unlock(tdb, -1, F_WRLCK);
return -1;
}
if (rec.magic != TDB_FREE_MAGIC) {
printf("bad magic 0x%08x in free list\n", rec.magic);
tdb_unlock(tdb, -1, F_WRLCK);
return -1;
}
printf("entry offset=[0x%08x], rec.rec_len = [0x%08x (%d)] (end = 0x%08x)\n",
rec_ptr, rec.rec_len, rec.rec_len, rec_ptr + rec.rec_len);
total_free += rec.rec_len;
/* move to the next record */
rec_ptr = rec.next;
}
printf("total rec_len = [0x%08x (%d)]\n", (int)total_free,
(int)total_free);
return tdb_unlock(tdb, -1, F_WRLCK);
}
ccan/tdb/error.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
trivial database library
Copyright (C) Andrew Tridgell 1999-2005
Copyright (C) Paul `Rusty' Russell 2000
Copyright (C) Jeremy Allison 2000-2003
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
enum TDB_ERROR tdb_error(struct tdb_context *tdb)
{
return tdb->ecode;
}
static struct tdb_errname {
enum TDB_ERROR ecode; const char *estring;
} emap[] = { {TDB_SUCCESS, "Success"},
{TDB_ERR_CORRUPT, "Corrupt database"},
{TDB_ERR_IO, "IO Error"},
{TDB_ERR_LOCK, "Locking error"},
{TDB_ERR_OOM, "Out of memory"},
{TDB_ERR_EXISTS, "Record exists"},
{TDB_ERR_NOLOCK, "Lock exists on other keys"},
{TDB_ERR_EINVAL, "Invalid parameter"},
{TDB_ERR_NOEXIST, "Record does not exist"},
{TDB_ERR_RDONLY, "write not permitted"} };
/* Error string for the last tdb error */
const char *tdb_errorstr(struct tdb_context *tdb)
{
uint32_t i;
for (i = 0; i < sizeof(emap) / sizeof(struct tdb_errname); i++)
if (tdb->ecode == emap[i].ecode)
return emap[i].estring;
return "Invalid error code";
}
ccan/tdb/freelist.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
trivial database library
Copyright (C) Andrew Tridgell 1999-2005
Copyright (C) Paul `Rusty' Russell 2000
Copyright (C) Jeremy Allison 2000-2003
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
/* 'right' merges can involve O(n^2) cost when combined with a
traverse, so they are disabled until we find a way to do them in
O(1) time
*/
#define USE_RIGHT_MERGES 0
/* read a freelist record and check for simple errors */
int tdb_rec_free_read(struct tdb_context *tdb, tdb_off_t off, struct tdb_record *rec)
{
if (tdb->methods->tdb_read(tdb, off, rec, sizeof(*rec),DOCONV()) == -1)
return -1;
if (rec->magic == TDB_MAGIC) {
/* this happens when a app is showdown while deleting a record - we should
not completely fail when this happens */
TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_rec_free_read non-free magic 0x%x at offset=%d - fixing\n",
rec->magic, off));
rec->magic = TDB_FREE_MAGIC;
if (tdb->methods->tdb_write(tdb, off, rec, sizeof(*rec)) == -1)
return -1;
}
if (rec->magic != TDB_FREE_MAGIC) {
/* Ensure ecode is set for log fn. */
tdb->ecode = TDB_ERR_CORRUPT;
TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_rec_free_read bad magic 0x%x at offset=%d\n",
rec->magic, off));
return -1;
}
if (tdb->methods->tdb_oob(tdb, rec->next+sizeof(*rec), 0) != 0)
return -1;
return 0;
}
#if USE_RIGHT_MERGES
/* Remove an element from the freelist. Must have alloc lock. */
static int remove_from_freelist(struct tdb_context *tdb, tdb_off_t off, tdb_off_t next)
{
tdb_off_t last_ptr, i;
/* read in the freelist top */
last_ptr = FREELIST_TOP;
while (tdb_ofs_read(tdb, last_ptr, &i) != -1 && i != 0) {
if (i == off) {
/* We've found it! */
return tdb_ofs_write(tdb, last_ptr, &next);
}
/* Follow chain (next offset is at start of record) */
last_ptr = i;
}
tdb->ecode = TDB_ERR_CORRUPT;
TDB_LOG((tdb, TDB_DEBUG_FATAL,"remove_from_freelist: not on list at off=%d\n", off));
return -1;
}
#endif
/* update a record tailer (must hold allocation lock) */
static int update_tailer(struct tdb_context *tdb, tdb_off_t offset,
const struct tdb_record *rec)
{
tdb_off_t totalsize;
/* Offset of tailer from record header */
totalsize = sizeof(*rec) + rec->rec_len;
return tdb_ofs_write(tdb, offset + totalsize - sizeof(tdb_off_t),
&totalsize);
}
/* Add an element into the freelist. Merge adjacent records if
necessary. */
int tdb_free(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
{
/* Allocation and tailer lock */
if (tdb_lock(tdb, -1, F_WRLCK) != 0)
return -1;
/* set an initial tailer, so if we fail we don't leave a bogus record */
if (update_tailer(tdb, offset, rec) != 0) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: update_tailer failed!\n"));
goto fail;
}
#if USE_RIGHT_MERGES
/* Look right first (I'm an Australian, dammit) */
if (offset + sizeof(*rec) + rec->rec_len + sizeof(*rec) <= tdb->map_size) {
tdb_off_t right = offset + sizeof(*rec) + rec->rec_len;
struct tdb_record r;
if (tdb->methods->tdb_read(tdb, right, &r, sizeof(r), DOCONV()) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: right read failed at %u\n", right));
goto left;
}
/* If it's free, expand to include it. */
if (r.magic == TDB_FREE_MAGIC) {
if (remove_from_freelist(tdb, right, r.next) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: right free failed at %u\n", right));
goto left;
}
rec->rec_len += sizeof(r) + r.rec_len;
if (update_tailer(tdb, offset, rec) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: update_tailer failed at %u\n", offset));
goto fail;
}
}
}
left:
#endif
/* Look left */
if (offset - sizeof(tdb_off_t) > TDB_DATA_START(tdb->header.hash_size)) {
tdb_off_t left = offset - sizeof(tdb_off_t);
struct tdb_record l;
tdb_off_t leftsize;
/* Read in tailer and jump back to header */
if (tdb_ofs_read(tdb, left, &leftsize) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: left offset read failed at %u\n", left));
goto update;
}
/* it could be uninitialised data */
if (leftsize == 0 || leftsize == TDB_PAD_U32) {
goto update;
}
left = offset - leftsize;
if (leftsize > offset ||
left < TDB_DATA_START(tdb->header.hash_size)) {
goto update;
}
/* Now read in the left record */
if (tdb->methods->tdb_read(tdb, left, &l, sizeof(l), DOCONV()) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: left read failed at %u (%u)\n", left, leftsize));
goto update;
}
/* If it's free, expand to include it. */
if (l.magic == TDB_FREE_MAGIC) {
/* we now merge the new record into the left record, rather than the other
way around. This makes the operation O(1) instead of O(n). This change
prevents traverse from being O(n^2) after a lot of deletes */
l.rec_len += sizeof(*rec) + rec->rec_len;
if (tdb_rec_write(tdb, left, &l) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: update_left failed at %u\n", left));
goto fail;
}
if (update_tailer(tdb, left, &l) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free: update_tailer failed at %u\n", offset));
goto fail;
}
tdb_unlock(tdb, -1, F_WRLCK);
return 0;
}
}
update:
/* Now, prepend to free list */
rec->magic = TDB_FREE_MAGIC;
if (tdb_ofs_read(tdb, FREELIST_TOP, &rec->next) == -1 ||
tdb_rec_write(tdb, offset, rec) == -1 ||
tdb_ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_free record write failed at offset=%d\n", offset));
goto fail;
}
/* And we're done. */
tdb_unlock(tdb, -1, F_WRLCK);
return 0;
fail:
tdb_unlock(tdb, -1, F_WRLCK);
return -1;
}
/*
the core of tdb_allocate - called when we have decided which
free list entry to use
Note that we try to allocate by grabbing data from the end of an existing record,
not the beginning. This is so the left merge in a free is more likely to be
able to free up the record without fragmentation
*/
static tdb_off_t tdb_allocate_ofs(struct tdb_context *tdb,
tdb_len_t length, tdb_off_t rec_ptr,
struct tdb_record *rec, tdb_off_t last_ptr)
{
#define MIN_REC_SIZE (sizeof(struct tdb_record) + sizeof(tdb_off_t) + 8)
if (rec->rec_len < length + MIN_REC_SIZE) {
/* we have to grab the whole record */
/* unlink it from the previous record */
if (tdb_ofs_write(tdb, last_ptr, &rec->next) == -1) {
return 0;
}
/* mark it not free */
rec->magic = TDB_MAGIC;
if (tdb_rec_write(tdb, rec_ptr, rec) == -1) {
return 0;
}
return rec_ptr;
}
/* we're going to just shorten the existing record */
rec->rec_len -= (length + sizeof(*rec));
if (tdb_rec_write(tdb, rec_ptr, rec) == -1) {
return 0;
}
if (update_tailer(tdb, rec_ptr, rec) == -1) {
return 0;
}
/* and setup the new record */
rec_ptr += sizeof(*rec) + rec->rec_len;
memset(rec, '\0', sizeof(*rec));
rec->rec_len = length;
rec->magic = TDB_MAGIC;
if (tdb_rec_write(tdb, rec_ptr, rec) == -1) {
return 0;
}
if (update_tailer(tdb, rec_ptr, rec) == -1) {
return 0;
}
return rec_ptr;
}
/* allocate some space from the free list. The offset returned points
to a unconnected tdb_record within the database with room for at
least length bytes of total data
0 is returned if the space could not be allocated
*/
tdb_off_t tdb_allocate(struct tdb_context *tdb, tdb_len_t length, struct tdb_record *rec)
{
tdb_off_t rec_ptr, last_ptr, newrec_ptr;
struct {
tdb_off_t rec_ptr, last_ptr;
tdb_len_t rec_len;
} bestfit;
float multiplier = 1.0;
if (tdb_lock(tdb, -1, F_WRLCK) == -1)
return 0;
/* over-allocate to reduce fragmentation */
length *= 1.25;
/* Extra bytes required for tailer */
length += sizeof(tdb_off_t);
length = TDB_ALIGN(length, TDB_ALIGNMENT);
again:
last_ptr = FREELIST_TOP;
/* read in the freelist top */
if (tdb_ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1)
goto fail;
bestfit.rec_ptr = 0;
bestfit.last_ptr = 0;
bestfit.rec_len = 0;
/*
this is a best fit allocation strategy. Originally we used
a first fit strategy, but it suffered from massive fragmentation
issues when faced with a slowly increasing record size.
*/
while (rec_ptr) {
if (tdb_rec_free_read(tdb, rec_ptr, rec) == -1) {
goto fail;
}
if (rec->rec_len >= length) {
if (bestfit.rec_ptr == 0 ||
rec->rec_len < bestfit.rec_len) {
bestfit.rec_len = rec->rec_len;
bestfit.rec_ptr = rec_ptr;
bestfit.last_ptr = last_ptr;
}
}
/* move to the next record */
last_ptr = rec_ptr;
rec_ptr = rec->next;
/* if we've found a record that is big enough, then
stop searching if its also not too big. The
definition of 'too big' changes as we scan
through */
if (bestfit.rec_len > 0 &&
bestfit.rec_len < length * multiplier) {
break;
}
/* this multiplier means we only extremely rarely
search more than 50 or so records. At 50 records we
accept records up to 11 times larger than what we
want */
multiplier *= 1.05;
}
if (bestfit.rec_ptr != 0) {
if (tdb_rec_free_read(tdb, bestfit.rec_ptr, rec) == -1) {
goto fail;
}
newrec_ptr = tdb_allocate_ofs(tdb, length, bestfit.rec_ptr,
rec, bestfit.last_ptr);
tdb_unlock(tdb, -1, F_WRLCK);
return newrec_ptr;
}
/* we didn't find enough space. See if we can expand the
database and if we can then try again */
if (tdb_expand(tdb, length + sizeof(*rec)) == 0)
goto again;
fail:
tdb_unlock(tdb, -1, F_WRLCK);
return 0;
}
/*
return the size of the freelist - used to decide if we should repack
*/
int tdb_freelist_size(struct tdb_context *tdb)
{
tdb_off_t ptr;
int count=0;
if (tdb_lock(tdb, -1, F_RDLCK) == -1) {
return -1;
}
ptr = FREELIST_TOP;
while (tdb_ofs_read(tdb, ptr, &ptr) == 0 && ptr != 0) {
count++;
}
tdb_unlock(tdb, -1, F_RDLCK);
return count;
}
ccan/tdb/freelistcheck.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
trivial database library
Copyright (C) Jeremy Allison 2006
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
/* Check the freelist is good and contains no loops.
Very memory intensive - only do this as a consistency
checker. Heh heh - uses an in memory tdb as the storage
for the "seen" record list. For some reason this strikes
me as extremely clever as I don't have to write another tree
data structure implementation :-).
*/
static int seen_insert(struct tdb_context *mem_tdb, tdb_off_t rec_ptr)
{
TDB_DATA key, data;
memset(&data, '\0', sizeof(data));
key.dptr = (unsigned char *)&rec_ptr;
key.dsize = sizeof(rec_ptr);
return tdb_store(mem_tdb, key, data, TDB_INSERT);
}
int tdb_validate_freelist(struct tdb_context *tdb, int *pnum_entries)
{
struct tdb_context *mem_tdb = NULL;
struct tdb_record rec;
tdb_off_t rec_ptr;
int ret = -1;
*pnum_entries = 0;
mem_tdb = tdb_open("flval", tdb->header.hash_size,
TDB_INTERNAL, O_RDWR, 0600);
if (!mem_tdb) {
return -1;
}
if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
tdb_close(mem_tdb);
return 0;
}
/* Store the FREELIST_TOP record. */
if (seen_insert(mem_tdb, FREELIST_TOP) == -1) {
tdb->ecode = TDB_ERR_CORRUPT;
ret = -1;
goto fail;
}
/* read in the freelist top */
if (tdb_ofs_read(tdb, FREELIST_TOP, &rec_ptr) == -1) {
goto fail;
}
while (rec_ptr) {
/* If we can't store this record (we've seen it
before) then the free list has a loop and must
be corrupt. */
if (seen_insert(mem_tdb, rec_ptr)) {
tdb->ecode = TDB_ERR_CORRUPT;
ret = -1;
goto fail;
}
if (tdb_rec_free_read(tdb, rec_ptr, &rec) == -1) {
goto fail;
}
/* move to the next record */
rec_ptr = rec.next;
*pnum_entries += 1;
}
ret = 0;
fail:
tdb_close(mem_tdb);
tdb_unlock(tdb, -1, F_WRLCK);
return ret;
}
ccan/tdb/hash.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
trivial database library
Copyright (C) Rusty Russell 2010
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
/* This is based on the hash algorithm from gdbm */
unsigned int tdb_old_hash(TDB_DATA *key)
{
uint32_t value; /* Used to compute the hash value. */
uint32_t i; /* Used to cycle through random values. */
/* Set the initial value from the key size. */
for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
value = (value + (key->dptr[i] << (i*5 % 24)));
return (1103515243 * value + 12345);
}
#if HAVE_LITTLE_ENDIAN
# define HASH_LITTLE_ENDIAN 1
# define HASH_BIG_ENDIAN 0
#elif HAVE_BIG_ENDIAN
# define HASH_LITTLE_ENDIAN 0
# define HASH_BIG_ENDIAN 1
#else
# error Unknown endian
#endif
/*
-------------------------------------------------------------------------------
lookup3.c, by Bob Jenkins, May 2006, Public Domain.
These are functions for producing 32-bit hashes for hash table lookup.
hash_word(), hashlittle(), hashlittle2(), hashbig(), mix(), and final()
are externally useful functions. Routines to test the hash are included
if SELF_TEST is defined. You can use this free for any purpose. It's in
the public domain. It has no warranty.
You probably want to use hashlittle(). hashlittle() and hashbig()
hash byte arrays. hashlittle() is is faster than hashbig() on
little-endian machines. Intel and AMD are little-endian machines.
On second thought, you probably want hashlittle2(), which is identical to
hashlittle() except it returns two 32-bit hashes for the price of one.
You could implement hashbig2() if you wanted but I haven't bothered here.
If you want to find a hash of, say, exactly 7 integers, do
a = i1; b = i2; c = i3;
mix(a,b,c);
a += i4; b += i5; c += i6;
mix(a,b,c);
a += i7;
final(a,b,c);
then use c as the hash value. If you have a variable length array of
4-byte integers to hash, use hash_word(). If you have a byte array (like
a character string), use hashlittle(). If you have several byte arrays, or
a mix of things, see the comments above hashlittle().
Why is this so big? I read 12 bytes at a time into 3 4-byte integers,
then mix those integers. This is fast (you can do a lot more thorough
mixing with 12*3 instructions on 3 integers than you can with 3 instructions
on 1 byte), but shoehorning those bytes into integers efficiently is messy.
*/
#define hashsize(n) ((uint32_t)1<<(n))
#define hashmask(n) (hashsize(n)-1)
#define rot(x,k) (((x)<<(k)) | ((x)>>(32-(k))))
/*
-------------------------------------------------------------------------------
mix -- mix 3 32-bit values reversibly.
This is reversible, so any information in (a,b,c) before mix() is
still in (a,b,c) after mix().
If four pairs of (a,b,c) inputs are run through mix(), or through
mix() in reverse, there are at least 32 bits of the output that
are sometimes the same for one pair and different for another pair.
This was tested for:
* pairs that differed by one bit, by two bits, in any combination
of top bits of (a,b,c), or in any combination of bottom bits of
(a,b,c).
* "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
is commonly produced by subtraction) look like a single 1-bit
difference.
* the base values were pseudorandom, all zero but one bit set, or
all zero plus a counter that starts at zero.
Some k values for my "a-=c; a^=rot(c,k); c+=b;" arrangement that
satisfy this are
4 6 8 16 19 4
9 15 3 18 27 15
14 9 3 7 17 3
Well, "9 15 3 18 27 15" didn't quite get 32 bits diffing
for "differ" defined as + with a one-bit base and a two-bit delta. I
used http://burtleburtle.net/bob/hash/avalanche.html to choose
the operations, constants, and arrangements of the variables.
This does not achieve avalanche. There are input bits of (a,b,c)
that fail to affect some output bits of (a,b,c), especially of a. The
most thoroughly mixed value is c, but it doesn't really even achieve
avalanche in c.
This allows some parallelism. Read-after-writes are good at doubling
the number of bits affected, so the goal of mixing pulls in the opposite
direction as the goal of parallelism. I did what I could. Rotates
seem to cost as much as shifts on every machine I could lay my hands
on, and rotates are much kinder to the top and bottom bits, so I used
rotates.
-------------------------------------------------------------------------------
*/
#define mix(a,b,c) \
{ \
a -= c; a ^= rot(c, 4); c += b; \
b -= a; b ^= rot(a, 6); a += c; \
c -= b; c ^= rot(b, 8); b += a; \
a -= c; a ^= rot(c,16); c += b; \
b -= a; b ^= rot(a,19); a += c; \
c -= b; c ^= rot(b, 4); b += a; \
}
/*
-------------------------------------------------------------------------------
final -- final mixing of 3 32-bit values (a,b,c) into c
Pairs of (a,b,c) values differing in only a few bits will usually
produce values of c that look totally different. This was tested for
* pairs that differed by one bit, by two bits, in any combination
of top bits of (a,b,c), or in any combination of bottom bits of
(a,b,c).
* "differ" is defined as +, -, ^, or ~^. For + and -, I transformed
the output delta to a Gray code (a^(a>>1)) so a string of 1's (as
is commonly produced by subtraction) look like a single 1-bit
difference.
* the base values were pseudorandom, all zero but one bit set, or
all zero plus a counter that starts at zero.
These constants passed:
14 11 25 16 4 14 24
12 14 25 16 4 14 24
and these came close:
4 8 15 26 3 22 24
10 8 15 26 3 22 24
11 8 15 26 3 22 24
-------------------------------------------------------------------------------
*/
#define final(a,b,c) \
{ \
c ^= b; c -= rot(b,14); \
a ^= c; a -= rot(c,11); \
b ^= a; b -= rot(a,25); \
c ^= b; c -= rot(b,16); \
a ^= c; a -= rot(c,4); \
b ^= a; b -= rot(a,14); \
c ^= b; c -= rot(b,24); \
}
/*
-------------------------------------------------------------------------------
hashlittle() -- hash a variable-length key into a 32-bit value
k : the key (the unaligned variable-length array of bytes)
length : the length of the key, counting by bytes
val2 : IN: can be any 4-byte value OUT: second 32 bit hash.
Returns a 32-bit value. Every bit of the key affects every bit of
the return value. Two keys differing by one or two bits will have
totally different hash values. Note that the return value is better
mixed than val2, so use that first.
The best hash table sizes are powers of 2. There is no need to do
mod a prime (mod is sooo slow!). If you need less than 32 bits,
use a bitmask. For example, if you need only 10 bits, do
h = (h & hashmask(10));
In which case, the hash table should have hashsize(10) elements.
If you are hashing n strings (uint8_t **)k, do it like this:
for (i=0, h=0; i<n; ++i) h = hashlittle( k[i], len[i], h);
By Bob Jenkins, 2006. bob_jenkins@burtleburtle.net. You may use this
code any way you wish, private, educational, or commercial. It's free.
Use for hash table lookup, or anything where one collision in 2^^32 is
acceptable. Do NOT use for cryptographic purposes.
-------------------------------------------------------------------------------
*/
static uint32_t hashlittle( const void *key, size_t length )
{
uint32_t a,b,c; /* internal state */
union { const void *ptr; size_t i; } u; /* needed for Mac Powerbook G4 */
/* Set up the internal state */
a = b = c = 0xdeadbeef + ((uint32_t)length);
u.ptr = key;
if (HASH_LITTLE_ENDIAN && ((u.i & 0x3) == 0)) {
const uint32_t *k = (const uint32_t *)key; /* read 32-bit chunks */
#ifdef VALGRIND
const uint8_t *k8;
#endif
/*------ all but last block: aligned reads and affect 32 bits of (a,b,c) */
while (length > 12)
{
a += k[0];
b += k[1];
c += k[2];
mix(a,b,c);
length -= 12;
k += 3;
}
/*----------------------------- handle the last (probably partial) block */
/*
* "k[2]&0xffffff" actually reads beyond the end of the string, but
* then masks off the part it's not allowed to read. Because the
* string is aligned, the masked-off tail is in the same word as the
* rest of the string. Every machine with memory protection I've seen
* does it on word boundaries, so is OK with this. But VALGRIND will
* still catch it and complain. The masking trick does make the hash
* noticably faster for short strings (like English words).
*/
#ifndef VALGRIND
switch(length)
{
case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
case 11: c+=k[2]&0xffffff; b+=k[1]; a+=k[0]; break;
case 10: c+=k[2]&0xffff; b+=k[1]; a+=k[0]; break;
case 9 : c+=k[2]&0xff; b+=k[1]; a+=k[0]; break;
case 8 : b+=k[1]; a+=k[0]; break;
case 7 : b+=k[1]&0xffffff; a+=k[0]; break;
case 6 : b+=k[1]&0xffff; a+=k[0]; break;
case 5 : b+=k[1]&0xff; a+=k[0]; break;
case 4 : a+=k[0]; break;
case 3 : a+=k[0]&0xffffff; break;
case 2 : a+=k[0]&0xffff; break;
case 1 : a+=k[0]&0xff; break;
case 0 : return c; /* zero length strings require no mixing */
}
#else /* make valgrind happy */
k8 = (const uint8_t *)k;
switch(length)
{
case 12: c+=k[2]; b+=k[1]; a+=k[0]; break;
case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
case 10: c+=((uint32_t)k8[9])<<8; /* fall through */
case 9 : c+=k8[8]; /* fall through */
case 8 : b+=k[1]; a+=k[0]; break;
case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
case 6 : b+=((uint32_t)k8[5])<<8; /* fall through */
case 5 : b+=k8[4]; /* fall through */
case 4 : a+=k[0]; break;
case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
case 2 : a+=((uint32_t)k8[1])<<8; /* fall through */
case 1 : a+=k8[0]; break;
case 0 : return c;
}
#endif /* !valgrind */
} else if (HASH_LITTLE_ENDIAN && ((u.i & 0x1) == 0)) {
const uint16_t *k = (const uint16_t *)key; /* read 16-bit chunks */
const uint8_t *k8;
/*--------------- all but last block: aligned reads and different mixing */
while (length > 12)
{
a += k[0] + (((uint32_t)k[1])<<16);
b += k[2] + (((uint32_t)k[3])<<16);
c += k[4] + (((uint32_t)k[5])<<16);
mix(a,b,c);
length -= 12;
k += 6;
}
/*----------------------------- handle the last (probably partial) block */
k8 = (const uint8_t *)k;
switch(length)
{
case 12: c+=k[4]+(((uint32_t)k[5])<<16);
b+=k[2]+(((uint32_t)k[3])<<16);
a+=k[0]+(((uint32_t)k[1])<<16);
break;
case 11: c+=((uint32_t)k8[10])<<16; /* fall through */
case 10: c+=k[4];
b+=k[2]+(((uint32_t)k[3])<<16);
a+=k[0]+(((uint32_t)k[1])<<16);
break;
case 9 : c+=k8[8]; /* fall through */
case 8 : b+=k[2]+(((uint32_t)k[3])<<16);
a+=k[0]+(((uint32_t)k[1])<<16);
break;
case 7 : b+=((uint32_t)k8[6])<<16; /* fall through */
case 6 : b+=k[2];
a+=k[0]+(((uint32_t)k[1])<<16);
break;
case 5 : b+=k8[4]; /* fall through */
case 4 : a+=k[0]+(((uint32_t)k[1])<<16);
break;
case 3 : a+=((uint32_t)k8[2])<<16; /* fall through */
case 2 : a+=k[0];
break;
case 1 : a+=k8[0];
break;
case 0 : return c; /* zero length requires no mixing */
}
} else { /* need to read the key one byte at a time */
const uint8_t *k = (const uint8_t *)key;
/*--------------- all but the last block: affect some 32 bits of (a,b,c) */
while (length > 12)
{
a += k[0];
a += ((uint32_t)k[1])<<8;
a += ((uint32_t)k[2])<<16;
a += ((uint32_t)k[3])<<24;
b += k[4];
b += ((uint32_t)k[5])<<8;
b += ((uint32_t)k[6])<<16;
b += ((uint32_t)k[7])<<24;
c += k[8];
c += ((uint32_t)k[9])<<8;
c += ((uint32_t)k[10])<<16;
c += ((uint32_t)k[11])<<24;
mix(a,b,c);
length -= 12;
k += 12;
}
/*-------------------------------- last block: affect all 32 bits of (c) */
switch(length) /* all the case statements fall through */
{
case 12: c+=((uint32_t)k[11])<<24;
case 11: c+=((uint32_t)k[10])<<16;
case 10: c+=((uint32_t)k[9])<<8;
case 9 : c+=k[8];
case 8 : b+=((uint32_t)k[7])<<24;
case 7 : b+=((uint32_t)k[6])<<16;
case 6 : b+=((uint32_t)k[5])<<8;
case 5 : b+=k[4];
case 4 : a+=((uint32_t)k[3])<<24;
case 3 : a+=((uint32_t)k[2])<<16;
case 2 : a+=((uint32_t)k[1])<<8;
case 1 : a+=k[0];
break;
case 0 : return c;
}
}
final(a,b,c);
return c;
}
unsigned int tdb_jenkins_hash(TDB_DATA *key)
{
return hashlittle(key->dptr, key->dsize);
}
ccan/tdb/io.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
trivial database library
Copyright (C) Andrew Tridgell 1999-2005
Copyright (C) Paul `Rusty' Russell 2000
Copyright (C) Jeremy Allison 2000-2003
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
#ifndef MAX
#define MAX(a,b) ((a) > (b) ? (a) : (b))
#endif
/* check for an out of bounds access - if it is out of bounds then
see if the database has been expanded by someone else and expand
if necessary
note that "len" is the minimum length needed for the db
*/
static int tdb_oob(struct tdb_context *tdb, tdb_off_t len, int probe)
{
struct stat st;
if (len <= tdb->map_size)
return 0;
if (tdb->flags & TDB_INTERNAL) {
if (!probe) {
/* Ensure ecode is set for log fn. */
tdb->ecode = TDB_ERR_IO;
TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_oob len %d beyond internal malloc size %d\n",
(int)len, (int)tdb->map_size));
}
return -1;
}
if (fstat(tdb->fd, &st) == -1) {
tdb->ecode = TDB_ERR_IO;
return -1;
}
if (st.st_size < (size_t)len) {
if (!probe) {
/* Ensure ecode is set for log fn. */
tdb->ecode = TDB_ERR_IO;
TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_oob len %d beyond eof at %d\n",
(int)len, (int)st.st_size));
}
return -1;
}
/* Unmap, update size, remap */
if (tdb_munmap(tdb) == -1) {
tdb->ecode = TDB_ERR_IO;
return -1;
}
tdb->map_size = st.st_size;
tdb_mmap(tdb);
return 0;
}
/* write a lump of data at a specified offset */
static int tdb_write(struct tdb_context *tdb, tdb_off_t off,
const void *buf, tdb_len_t len)
{
if (len == 0) {
return 0;
}
if (tdb->read_only || tdb->traverse_read) {
tdb->ecode = TDB_ERR_RDONLY;
return -1;
}
if (tdb->methods->tdb_oob(tdb, off + len, 0) != 0)
return -1;
if (tdb->map_ptr) {
memcpy(off + (char *)tdb->map_ptr, buf, len);
} else {
ssize_t written = pwrite(tdb->fd, buf, len, off);
if ((written != (ssize_t)len) && (written != -1)) {
/* try once more */
tdb->ecode = TDB_ERR_IO;
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_write: wrote only "
"%d of %d bytes at %d, trying once more\n",
(int)written, len, off));
written = pwrite(tdb->fd, (const char *)buf+written,
len-written,
off+written);
}
if (written == -1) {
/* Ensure ecode is set for log fn. */
tdb->ecode = TDB_ERR_IO;
TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_write failed at %d "
"len=%d (%s)\n", off, len, strerror(errno)));
return -1;
} else if (written != (ssize_t)len) {
tdb->ecode = TDB_ERR_IO;
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_write: failed to "
"write %d bytes at %d in two attempts\n",
len, off));
return -1;
}
}
return 0;
}
/* Endian conversion: we only ever deal with 4 byte quantities */
void *tdb_convert(void *buf, uint32_t size)
{
uint32_t i;
unsigned char *p = buf, tmp;
for (i = 0; i < size; i += 4) {
tmp = p[i];
p[i] = p[i+3];
p[i+3] = tmp;
tmp = p[i+1];
p[i+1] = p[i+2];
p[i+2] = tmp;
}
return buf;
}
/* read a lump of data at a specified offset, maybe convert */
static int tdb_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
tdb_len_t len, int cv)
{
if (tdb->methods->tdb_oob(tdb, off + len, 0) != 0) {
return -1;
}
if (tdb->map_ptr) {
memcpy(buf, off + (char *)tdb->map_ptr, len);
} else {
ssize_t ret = pread(tdb->fd, buf, len, off);
if (ret != (ssize_t)len) {
/* Ensure ecode is set for log fn. */
tdb->ecode = TDB_ERR_IO;
TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_read failed at %d "
"len=%d ret=%d (%s) map_size=%d\n",
(int)off, (int)len, (int)ret, strerror(errno),
(int)tdb->map_size));
return -1;
}
}
if (cv) {
tdb_convert(buf, len);
}
return 0;
}
/*
do an unlocked scan of the hash table heads to find the next non-zero head. The value
will then be confirmed with the lock held
*/
static void tdb_next_hash_chain(struct tdb_context *tdb, uint32_t *chain)
{
uint32_t h = *chain;
if (tdb->map_ptr) {
for (;h < tdb->header.hash_size;h++) {
if (0 != *(uint32_t *)(TDB_HASH_TOP(h) + (unsigned char *)tdb->map_ptr)) {
break;
}
}
} else {
uint32_t off=0;
for (;h < tdb->header.hash_size;h++) {
if (tdb_ofs_read(tdb, TDB_HASH_TOP(h), &off) != 0 || off != 0) {
break;
}
}
}
(*chain) = h;
}
int tdb_munmap(struct tdb_context *tdb)
{
if (tdb->flags & TDB_INTERNAL)
return 0;
#if HAVE_MMAP
if (tdb->map_ptr) {
int ret;
ret = munmap(tdb->map_ptr, tdb->map_size);
if (ret != 0)
return ret;
}
#endif
tdb->map_ptr = NULL;
return 0;
}
void tdb_mmap(struct tdb_context *tdb)
{
if (tdb->flags & TDB_INTERNAL)
return;
#if HAVE_MMAP
if (!(tdb->flags & TDB_NOMMAP)) {
tdb->map_ptr = mmap(NULL, tdb->map_size,
PROT_READ|(tdb->read_only? 0:PROT_WRITE),
MAP_SHARED, tdb->fd, 0);
/*
* NB. When mmap fails it returns MAP_FAILED *NOT* NULL !!!!
*/
if (tdb->map_ptr == MAP_FAILED) {
tdb->map_ptr = NULL;
TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_mmap failed for size %d (%s)\n",
tdb->map_size, strerror(errno)));
}
} else {
tdb->map_ptr = NULL;
}
#else
tdb->map_ptr = NULL;
#endif
}
/* expand a file. we prefer to use ftruncate, as that is what posix
says to use for mmap expansion */
static int tdb_expand_file(struct tdb_context *tdb, tdb_off_t size, tdb_off_t addition)
{
char buf[8192];
if (tdb->read_only || tdb->traverse_read) {
tdb->ecode = TDB_ERR_RDONLY;
return -1;
}
if (ftruncate(tdb->fd, size+addition) == -1) {
char b = 0;
ssize_t written = pwrite(tdb->fd, &b, 1, (size+addition) - 1);
if (written == 0) {
/* try once more, potentially revealing errno */
written = pwrite(tdb->fd, &b, 1, (size+addition) - 1);
}
if (written == 0) {
/* again - give up, guessing errno */
errno = ENOSPC;
}
if (written != 1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file to %d failed (%s)\n",
size+addition, strerror(errno)));
return -1;
}
}
/* now fill the file with something. This ensures that the
file isn't sparse, which would be very bad if we ran out of
disk. This must be done with write, not via mmap */
memset(buf, TDB_PAD_BYTE, sizeof(buf));
while (addition) {
size_t n = addition>sizeof(buf)?sizeof(buf):addition;
ssize_t written = pwrite(tdb->fd, buf, n, size);
if (written == 0) {
/* prevent infinite loops: try _once_ more */
written = pwrite(tdb->fd, buf, n, size);
}
if (written == 0) {
/* give up, trying to provide a useful errno */
TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file write "
"returned 0 twice: giving up!\n"));
errno = ENOSPC;
return -1;
} else if (written == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file write of "
"%d bytes failed (%s)\n", (int)n,
strerror(errno)));
return -1;
} else if (written != n) {
TDB_LOG((tdb, TDB_DEBUG_WARNING, "expand_file: wrote "
"only %d of %d bytes - retrying\n", (int)written,
(int)n));
}
addition -= written;
size += written;
}
return 0;
}
/* You need 'size', this tells you how much you should expand by. */
tdb_off_t tdb_expand_adjust(tdb_off_t map_size, tdb_off_t size, int page_size)
{
tdb_off_t new_size, top_size;
/* limit size in order to avoid using up huge amounts of memory for
* in memory tdbs if an oddball huge record creeps in */
if (size > 100 * 1024) {
top_size = map_size + size * 2;
} else {
top_size = map_size + size * 100;
}
/* always make room for at least top_size more records, and at
least 25% more space. if the DB is smaller than 100MiB,
otherwise grow it by 10% only. */
if (map_size > 100 * 1024 * 1024) {
new_size = map_size * 1.10;
} else {
new_size = map_size * 1.25;
}
/* Round the database up to a multiple of the page size */
new_size = MAX(top_size, new_size);
return TDB_ALIGN(new_size, page_size) - map_size;
}
/* expand the database at least size bytes by expanding the underlying
file and doing the mmap again if necessary */
int tdb_expand(struct tdb_context *tdb, tdb_off_t size)
{
struct tdb_record rec;
tdb_off_t offset;
if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "lock failed in tdb_expand\n"));
return -1;
}
/* must know about any previous expansions by another process */
tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
size = tdb_expand_adjust(tdb->map_size, size, tdb->page_size);
if (!(tdb->flags & TDB_INTERNAL))
tdb_munmap(tdb);
/*
* We must ensure the file is unmapped before doing this
* to ensure consistency with systems like OpenBSD where
* writes and mmaps are not consistent.
*/
/* expand the file itself */
if (!(tdb->flags & TDB_INTERNAL)) {
if (tdb->methods->tdb_expand_file(tdb, tdb->map_size, size) != 0)
goto fail;
}
tdb->map_size += size;
if (tdb->flags & TDB_INTERNAL) {
char *new_map_ptr = (char *)realloc(tdb->map_ptr,
tdb->map_size);
if (!new_map_ptr) {
tdb->map_size -= size;
goto fail;
}
tdb->map_ptr = new_map_ptr;
} else {
/*
* We must ensure the file is remapped before adding the space
* to ensure consistency with systems like OpenBSD where
* writes and mmaps are not consistent.
*/
/* We're ok if the mmap fails as we'll fallback to read/write */
tdb_mmap(tdb);
}
/* form a new freelist record */
memset(&rec,'\0',sizeof(rec));
rec.rec_len = size - sizeof(rec);
/* link it into the free list */
offset = tdb->map_size - size;
if (tdb_free(tdb, offset, &rec) == -1)
goto fail;
tdb_unlock(tdb, -1, F_WRLCK);
return 0;
fail:
tdb_unlock(tdb, -1, F_WRLCK);
return -1;
}
/* read/write a tdb_off_t */
int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
{
return tdb->methods->tdb_read(tdb, offset, (char*)d, sizeof(*d), DOCONV());
}
int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d)
{
tdb_off_t off = *d;
return tdb->methods->tdb_write(tdb, offset, CONVERT(off), sizeof(*d));
}
/* read a lump of data, allocating the space for it */
unsigned char *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len)
{
unsigned char *buf;
/* some systems don't like zero length malloc */
if (!(buf = (unsigned char *)malloc(len ? len : 1))) {
/* Ensure ecode is set for log fn. */
tdb->ecode = TDB_ERR_OOM;
TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_alloc_read malloc failed len=%d (%s)\n",
len, strerror(errno)));
return NULL;
}
if (tdb->methods->tdb_read(tdb, offset, buf, len, 0) == -1) {
SAFE_FREE(buf);
return NULL;
}
return buf;
}
/* Give a piece of tdb data to a parser */
int tdb_parse_data(struct tdb_context *tdb, TDB_DATA key,
tdb_off_t offset, tdb_len_t len,
int (*parser)(TDB_DATA key, TDB_DATA data,
void *private_data),
void *private_data)
{
TDB_DATA data;
int result;
data.dsize = len;
if ((tdb->transaction == NULL) && (tdb->map_ptr != NULL)) {
/*
* Optimize by avoiding the malloc/memcpy/free, point the
* parser directly at the mmap area.
*/
if (tdb->methods->tdb_oob(tdb, offset+len, 0) != 0) {
return -1;
}
data.dptr = offset + (unsigned char *)tdb->map_ptr;
return parser(key, data, private_data);
}
if (!(data.dptr = tdb_alloc_read(tdb, offset, len))) {
return -1;
}
result = parser(key, data, private_data);
free(data.dptr);
return result;
}
/* read/write a record */
int tdb_rec_read(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
{
if (tdb->methods->tdb_read(tdb, offset, rec, sizeof(*rec),DOCONV()) == -1)
return -1;
if (TDB_BAD_MAGIC(rec)) {
/* Ensure ecode is set for log fn. */
tdb->ecode = TDB_ERR_CORRUPT;
TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_rec_read bad magic 0x%x at offset=%d\n", rec->magic, offset));
return -1;
}
return tdb->methods->tdb_oob(tdb, rec->next+sizeof(*rec), 0);
}
int tdb_rec_write(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec)
{
struct tdb_record r = *rec;
return tdb->methods->tdb_write(tdb, offset, CONVERT(r), sizeof(r));
}
static const struct tdb_methods io_methods = {
tdb_read,
tdb_write,
tdb_next_hash_chain,
tdb_oob,
tdb_expand_file,
};
/*
initialise the default methods table
*/
void tdb_io_init(struct tdb_context *tdb)
{
tdb->methods = &io_methods;
}
ccan/tdb/lock.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
trivial database library
Copyright (C) Andrew Tridgell 1999-2005
Copyright (C) Paul `Rusty' Russell 2000
Copyright (C) Jeremy Allison 2000-2003
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
void tdb_setalarm_sigptr(struct tdb_context *tdb, volatile sig_atomic_t *ptr)
{
tdb->interrupt_sig_ptr = ptr;
}
static int fcntl_lock(struct tdb_context *tdb,
int rw, off_t off, off_t len, bool waitflag)
{
struct flock fl;
fl.l_type = rw;
fl.l_whence = SEEK_SET;
fl.l_start = off;
fl.l_len = len;
fl.l_pid = 0;
if (waitflag)
return fcntl(tdb->fd, F_SETLKW, &fl);
else
return fcntl(tdb->fd, F_SETLK, &fl);
}
static int fcntl_unlock(struct tdb_context *tdb, int rw, off_t off, off_t len)
{
struct flock fl;
#if 0 /* Check they matched up locks and unlocks correctly. */
char line[80];
FILE *locks;
bool found = false;
locks = fopen("/proc/locks", "r");
while (fgets(line, 80, locks)) {
char *p;
int type, start, l;
/* eg. 1: FLOCK ADVISORY WRITE 2440 08:01:2180826 0 EOF */
p = strchr(line, ':') + 1;
if (strncmp(p, " POSIX ADVISORY ", strlen(" POSIX ADVISORY ")))
continue;
p += strlen(" FLOCK ADVISORY ");
if (strncmp(p, "READ ", strlen("READ ")) == 0)
type = F_RDLCK;
else if (strncmp(p, "WRITE ", strlen("WRITE ")) == 0)
type = F_WRLCK;
else
abort();
p += 6;
if (atoi(p) != getpid())
continue;
p = strchr(strchr(p, ' ') + 1, ' ') + 1;
start = atoi(p);
p = strchr(p, ' ') + 1;
if (strncmp(p, "EOF", 3) == 0)
l = 0;
else
l = atoi(p) - start + 1;
if (off == start) {
if (len != l) {
fprintf(stderr, "Len %u should be %u: %s",
(int)len, l, line);
abort();
}
if (type != rw) {
fprintf(stderr, "Type %s wrong: %s",
rw == F_RDLCK ? "READ" : "WRITE", line);
abort();
}
found = true;
break;
}
}
if (!found) {
fprintf(stderr, "Unlock on %u@%u not found!\n",
(int)off, (int)len);
abort();
}
fclose(locks);
#endif
fl.l_type = F_UNLCK;
fl.l_whence = SEEK_SET;
fl.l_start = off;
fl.l_len = len;
fl.l_pid = 0;
return fcntl(tdb->fd, F_SETLKW, &fl);
}
/* list -1 is the alloc list, otherwise a hash chain. */
static tdb_off_t lock_offset(int list)
{
return FREELIST_TOP + 4*list;
}
/* a byte range locking function - return 0 on success
this functions locks/unlocks 1 byte at the specified offset.
On error, errno is also set so that errors are passed back properly
through tdb_open().
note that a len of zero means lock to end of file
*/
int tdb_brlock(struct tdb_context *tdb,
int rw_type, tdb_off_t offset, size_t len,
enum tdb_lock_flags flags)
{
int ret;
if (tdb->flags & TDB_NOLOCK) {
return 0;
}
if (flags & TDB_LOCK_MARK_ONLY) {
return 0;
}
if ((rw_type == F_WRLCK) && (tdb->read_only || tdb->traverse_read)) {
tdb->ecode = TDB_ERR_RDONLY;
return -1;
}
do {
ret = fcntl_lock(tdb, rw_type, offset, len,
flags & TDB_LOCK_WAIT);
/* Check for a sigalarm break. */
if (ret == -1 && errno == EINTR &&
tdb->interrupt_sig_ptr &&
*tdb->interrupt_sig_ptr) {
break;
}
} while (ret == -1 && errno == EINTR);
if (ret == -1) {
tdb->ecode = TDB_ERR_LOCK;
/* Generic lock error. errno set by fcntl.
* EAGAIN is an expected return from non-blocking
* locks. */
if (!(flags & TDB_LOCK_PROBE) && errno != EAGAIN) {
TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brlock failed (fd=%d) at offset %d rw_type=%d flags=%d len=%d\n",
tdb->fd, offset, rw_type, flags, (int)len));
}
return -1;
}
return 0;
}
int tdb_brunlock(struct tdb_context *tdb,
int rw_type, tdb_off_t offset, size_t len)
{
int ret;
if (tdb->flags & TDB_NOLOCK) {
return 0;
}
do {
ret = fcntl_unlock(tdb, rw_type, offset, len);
} while (ret == -1 && errno == EINTR);
if (ret == -1) {
TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_brunlock failed (fd=%d) at offset %d rw_type=%d len=%d\n",
tdb->fd, offset, rw_type, (int)len));
}
return ret;
}
/*
upgrade a read lock to a write lock. This needs to be handled in a
special way as some OSes (such as solaris) have too conservative
deadlock detection and claim a deadlock when progress can be
made. For those OSes we may loop for a while.
*/
int tdb_allrecord_upgrade(struct tdb_context *tdb)
{
int count = 1000;
if (tdb->allrecord_lock.count != 1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"tdb_allrecord_upgrade failed: count %u too high\n",
tdb->allrecord_lock.count));
return -1;
}
if (tdb->allrecord_lock.off != 1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"tdb_allrecord_upgrade failed: already upgraded?\n"));
return -1;
}
while (count--) {
struct timeval tv;
if (tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0,
TDB_LOCK_WAIT|TDB_LOCK_PROBE) == 0) {
tdb->allrecord_lock.ltype = F_WRLCK;
tdb->allrecord_lock.off = 0;
return 0;
}
if (errno != EDEADLK) {
break;
}
/* sleep for as short a time as we can - more portable than usleep() */
tv.tv_sec = 0;
tv.tv_usec = 1;
select(0, NULL, NULL, NULL, &tv);
}
TDB_LOG((tdb, TDB_DEBUG_TRACE,"tdb_allrecord_upgrade failed\n"));
return -1;
}
static struct tdb_lock_type *find_nestlock(struct tdb_context *tdb,
tdb_off_t offset)
{
unsigned int i;
for (i=0; i<tdb->num_lockrecs; i++) {
if (tdb->lockrecs[i].off == offset) {
return &tdb->lockrecs[i];
}
}
return NULL;
}
/* lock an offset in the database. */
int tdb_nest_lock(struct tdb_context *tdb, uint32_t offset, int ltype,
enum tdb_lock_flags flags)
{
struct tdb_lock_type *new_lck;
if (offset >= lock_offset(tdb->header.hash_size)) {
tdb->ecode = TDB_ERR_LOCK;
TDB_LOG((tdb, TDB_DEBUG_ERROR,"tdb_lock: invalid offset %u for ltype=%d\n",
offset, ltype));
return -1;
}
if (tdb->flags & TDB_NOLOCK)
return 0;
new_lck = find_nestlock(tdb, offset);
if (new_lck) {
/*
* Just increment the in-memory struct, posix locks
* don't stack.
*/
new_lck->count++;
return 0;
}
new_lck = (struct tdb_lock_type *)realloc(
tdb->lockrecs,
sizeof(*tdb->lockrecs) * (tdb->num_lockrecs+1));
if (new_lck == NULL) {
errno = ENOMEM;
return -1;
}
tdb->lockrecs = new_lck;
/* Since fcntl locks don't nest, we do a lock for the first one,
and simply bump the count for future ones */
if (tdb_brlock(tdb, ltype, offset, 1, flags)) {
return -1;
}
tdb->lockrecs[tdb->num_lockrecs].off = offset;
tdb->lockrecs[tdb->num_lockrecs].count = 1;
tdb->lockrecs[tdb->num_lockrecs].ltype = ltype;
tdb->num_lockrecs++;
return 0;
}
static int tdb_lock_and_recover(struct tdb_context *tdb)
{
int ret;
/* We need to match locking order in transaction commit. */
if (tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0, TDB_LOCK_WAIT)) {
return -1;
}
if (tdb_brlock(tdb, F_WRLCK, OPEN_LOCK, 1, TDB_LOCK_WAIT)) {
tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
return -1;
}
ret = tdb_transaction_recover(tdb);
tdb_brunlock(tdb, F_WRLCK, OPEN_LOCK, 1);
tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
return ret;
}
static bool have_data_locks(const struct tdb_context *tdb)
{
unsigned int i;
for (i = 0; i < tdb->num_lockrecs; i++) {
if (tdb->lockrecs[i].off >= lock_offset(-1))
return true;
}
return false;
}
static int tdb_lock_list(struct tdb_context *tdb, int list, int ltype,
enum tdb_lock_flags waitflag)
{
int ret;
bool check = false;
/* a allrecord lock allows us to avoid per chain locks */
if (tdb->allrecord_lock.count &&
(ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
return 0;
}
if (tdb->allrecord_lock.count) {
tdb->ecode = TDB_ERR_LOCK;
ret = -1;
} else {
/* Only check when we grab first data lock. */
check = !have_data_locks(tdb);
ret = tdb_nest_lock(tdb, lock_offset(list), ltype, waitflag);
if (ret == 0 && check && tdb_needs_recovery(tdb)) {
tdb_nest_unlock(tdb, lock_offset(list), ltype, false);
if (tdb_lock_and_recover(tdb) == -1) {
return -1;
}
return tdb_lock_list(tdb, list, ltype, waitflag);
}
}
return ret;
}
/* lock a list in the database. list -1 is the alloc list */
int tdb_lock(struct tdb_context *tdb, int list, int ltype)
{
int ret;
ret = tdb_lock_list(tdb, list, ltype, TDB_LOCK_WAIT);
if (ret) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_lock failed on list %d "
"ltype=%d (%s)\n", list, ltype, strerror(errno)));
}
return ret;
}
/* lock a list in the database. list -1 is the alloc list. non-blocking lock */
int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype)
{
return tdb_lock_list(tdb, list, ltype, TDB_LOCK_NOWAIT);
}
int tdb_nest_unlock(struct tdb_context *tdb, uint32_t offset, int ltype,
bool mark_lock)
{
int ret = -1;
struct tdb_lock_type *lck;
if (tdb->flags & TDB_NOLOCK)
return 0;
/* Sanity checks */
if (offset >= lock_offset(tdb->header.hash_size)) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: offset %u invalid (%d)\n", offset, tdb->header.hash_size));
return ret;
}
lck = find_nestlock(tdb, offset);
if ((lck == NULL) || (lck->count == 0)) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: count is 0\n"));
return -1;
}
if (lck->count > 1) {
lck->count--;
return 0;
}
/*
* This lock has count==1 left, so we need to unlock it in the
* kernel. We don't bother with decrementing the in-memory array
* element, we're about to overwrite it with the last array element
* anyway.
*/
if (mark_lock) {
ret = 0;
} else {
ret = tdb_brunlock(tdb, ltype, offset, 1);
}
/*
* Shrink the array by overwriting the element just unlocked with the
* last array element.
*/
*lck = tdb->lockrecs[--tdb->num_lockrecs];
/*
* We don't bother with realloc when the array shrinks, but if we have
* a completely idle tdb we should get rid of the locked array.
*/
if (tdb->num_lockrecs == 0) {
SAFE_FREE(tdb->lockrecs);
}
if (ret)
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlock: An error occurred unlocking!\n"));
return ret;
}
int tdb_unlock(struct tdb_context *tdb, int list, int ltype)
{
/* a global lock allows us to avoid per chain locks */
if (tdb->allrecord_lock.count &&
(ltype == tdb->allrecord_lock.ltype || ltype == F_RDLCK)) {
return 0;
}
if (tdb->allrecord_lock.count) {
tdb->ecode = TDB_ERR_LOCK;
return -1;
}
return tdb_nest_unlock(tdb, lock_offset(list), ltype, false);
}
/*
get the transaction lock
*/
int tdb_transaction_lock(struct tdb_context *tdb, int ltype)
{
return tdb_nest_lock(tdb, TRANSACTION_LOCK, ltype, TDB_LOCK_WAIT);
}
/*
release the transaction lock
*/
int tdb_transaction_unlock(struct tdb_context *tdb, int ltype)
{
return tdb_nest_unlock(tdb, TRANSACTION_LOCK, ltype, false);
}
/* Returns 0 if all done, -1 if error, 1 if ok. */
static int tdb_allrecord_check(struct tdb_context *tdb, int ltype,
enum tdb_lock_flags flags, bool upgradable)
{
/* There are no locks on read-only dbs */
if (tdb->read_only || tdb->traverse_read) {
tdb->ecode = TDB_ERR_LOCK;
return -1;
}
if (tdb->allrecord_lock.count && tdb->allrecord_lock.ltype == ltype) {
tdb->allrecord_lock.count++;
return 0;
}
if (tdb->allrecord_lock.count) {
/* a global lock of a different type exists */
tdb->ecode = TDB_ERR_LOCK;
return -1;
}
if (tdb_have_extra_locks(tdb)) {
/* can't combine global and chain locks */
tdb->ecode = TDB_ERR_LOCK;
return -1;
}
if (upgradable && ltype != F_RDLCK) {
/* tdb error: you can't upgrade a write lock! */
tdb->ecode = TDB_ERR_LOCK;
return -1;
}
return 1;
}
/* We only need to lock individual bytes, but Linux merges consecutive locks
* so we lock in contiguous ranges. */
static int tdb_chainlock_gradual(struct tdb_context *tdb,
int ltype, enum tdb_lock_flags flags,
size_t off, size_t len)
{
int ret;
enum tdb_lock_flags nb_flags = (flags & ~TDB_LOCK_WAIT);
if (len <= 4) {
/* Single record. Just do blocking lock. */
return tdb_brlock(tdb, ltype, off, len, flags);
}
/* First we try non-blocking. */
ret = tdb_brlock(tdb, ltype, off, len, nb_flags);
if (ret == 0) {
return 0;
}
/* Try locking first half, then second. */
ret = tdb_chainlock_gradual(tdb, ltype, flags, off, len / 2);
if (ret == -1)
return -1;
ret = tdb_chainlock_gradual(tdb, ltype, flags,
off + len / 2, len - len / 2);
if (ret == -1) {
tdb_brunlock(tdb, ltype, off, len / 2);
return -1;
}
return 0;
}
/* lock/unlock entire database. It can only be upgradable if you have some
* other way of guaranteeing exclusivity (ie. transaction write lock).
* We do the locking gradually to avoid being starved by smaller locks. */
int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
enum tdb_lock_flags flags, bool upgradable)
{
switch (tdb_allrecord_check(tdb, ltype, flags, upgradable)) {
case -1:
return -1;
case 0:
return 0;
}
/* We cover two kinds of locks:
* 1) Normal chain locks. Taken for almost all operations.
* 3) Individual records locks. Taken after normal or free
* chain locks.
*
* It is (1) which cause the starvation problem, so we're only
* gradual for that. */
if (tdb_chainlock_gradual(tdb, ltype, flags, FREELIST_TOP,
tdb->header.hash_size * 4) == -1) {
return -1;
}
/* Grab individual record locks. */
if (tdb_brlock(tdb, ltype, lock_offset(tdb->header.hash_size), 0,
flags) == -1) {
tdb_brunlock(tdb, ltype, FREELIST_TOP,
tdb->header.hash_size * 4);
return -1;
}
tdb->allrecord_lock.count = 1;
/* If it's upgradable, it's actually exclusive so we can treat
* it as a write lock. */
tdb->allrecord_lock.ltype = upgradable ? F_WRLCK : ltype;
tdb->allrecord_lock.off = upgradable;
if (tdb_needs_recovery(tdb)) {
bool mark = flags & TDB_LOCK_MARK_ONLY;
tdb_allrecord_unlock(tdb, ltype, mark);
if (mark) {
tdb->ecode = TDB_ERR_LOCK;
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"tdb_lockall_mark cannot do recovery\n"));
return -1;
}
if (tdb_lock_and_recover(tdb) == -1) {
return -1;
}
return tdb_allrecord_lock(tdb, ltype, flags, upgradable);
}
return 0;
}
/* unlock entire db */
int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype, bool mark_lock)
{
/* There are no locks on read-only dbs */
if (tdb->read_only || tdb->traverse_read) {
tdb->ecode = TDB_ERR_LOCK;
return -1;
}
if (tdb->allrecord_lock.count == 0) {
tdb->ecode = TDB_ERR_LOCK;
return -1;
}
/* Upgradable locks are marked as write locks. */
if (tdb->allrecord_lock.ltype != ltype
&& (!tdb->allrecord_lock.off || ltype != F_RDLCK)) {
tdb->ecode = TDB_ERR_LOCK;
return -1;
}
if (tdb->allrecord_lock.count > 1) {
tdb->allrecord_lock.count--;
return 0;
}
if (!mark_lock && tdb_brunlock(tdb, ltype, FREELIST_TOP, 0)) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_unlockall failed (%s)\n", strerror(errno)));
return -1;
}
tdb->allrecord_lock.count = 0;
tdb->allrecord_lock.ltype = 0;
return 0;
}
/* lock entire database with write lock */
int tdb_lockall(struct tdb_context *tdb)
{
tdb_trace(tdb, "tdb_lockall");
return tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_WAIT, false);
}
/* lock entire database with write lock - mark only */
int tdb_lockall_mark(struct tdb_context *tdb)
{
tdb_trace(tdb, "tdb_lockall_mark");
return tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_MARK_ONLY, false);
}
/* unlock entire database with write lock - unmark only */
int tdb_lockall_unmark(struct tdb_context *tdb)
{
tdb_trace(tdb, "tdb_lockall_unmark");
return tdb_allrecord_unlock(tdb, F_WRLCK, true);
}
/* lock entire database with write lock - nonblocking varient */
int tdb_lockall_nonblock(struct tdb_context *tdb)
{
int ret = tdb_allrecord_lock(tdb, F_WRLCK, TDB_LOCK_NOWAIT, false);
tdb_trace_ret(tdb, "tdb_lockall_nonblock", ret);
return ret;
}
/* unlock entire database with write lock */
int tdb_unlockall(struct tdb_context *tdb)
{
tdb_trace(tdb, "tdb_unlockall");
return tdb_allrecord_unlock(tdb, F_WRLCK, false);
}
/* lock entire database with read lock */
int tdb_lockall_read(struct tdb_context *tdb)
{
tdb_trace(tdb, "tdb_lockall_read");
return tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false);
}
/* lock entire database with read lock - nonblock varient */
int tdb_lockall_read_nonblock(struct tdb_context *tdb)
{
int ret = tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_NOWAIT, false);
tdb_trace_ret(tdb, "tdb_lockall_read_nonblock", ret);
return ret;
}
/* unlock entire database with read lock */
int tdb_unlockall_read(struct tdb_context *tdb)
{
tdb_trace(tdb, "tdb_unlockall_read");
return tdb_allrecord_unlock(tdb, F_RDLCK, false);
}
/* lock/unlock one hash chain. This is meant to be used to reduce
contention - it cannot guarantee how many records will be locked */
int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key)
{
int ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
tdb_trace_1rec(tdb, "tdb_chainlock", key);
return ret;
}
/* lock/unlock one hash chain, non-blocking. This is meant to be used
to reduce contention - it cannot guarantee how many records will be
locked */
int tdb_chainlock_nonblock(struct tdb_context *tdb, TDB_DATA key)
{
int ret = tdb_lock_nonblock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
tdb_trace_1rec_ret(tdb, "tdb_chainlock_nonblock", key, ret);
return ret;
}
/* mark a chain as locked without actually locking it. Warning! use with great caution! */
int tdb_chainlock_mark(struct tdb_context *tdb, TDB_DATA key)
{
int ret = tdb_nest_lock(tdb, lock_offset(BUCKET(tdb->hash_fn(&key))),
F_WRLCK, TDB_LOCK_MARK_ONLY);
tdb_trace_1rec(tdb, "tdb_chainlock_mark", key);
return ret;
}
/* unmark a chain as locked without actually locking it. Warning! use with great caution! */
int tdb_chainlock_unmark(struct tdb_context *tdb, TDB_DATA key)
{
tdb_trace_1rec(tdb, "tdb_chainlock_unmark", key);
return tdb_nest_unlock(tdb, lock_offset(BUCKET(tdb->hash_fn(&key))),
F_WRLCK, true);
}
int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key)
{
tdb_trace_1rec(tdb, "tdb_chainunlock", key);
return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_WRLCK);
}
int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key)
{
int ret;
ret = tdb_lock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
tdb_trace_1rec(tdb, "tdb_chainlock_read", key);
return ret;
}
int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key)
{
tdb_trace_1rec(tdb, "tdb_chainunlock_read", key);
return tdb_unlock(tdb, BUCKET(tdb->hash_fn(&key)), F_RDLCK);
}
/* record lock stops delete underneath */
int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off)
{
if (tdb->allrecord_lock.count) {
return 0;
}
return off ? tdb_brlock(tdb, F_RDLCK, off, 1, TDB_LOCK_WAIT) : 0;
}
/*
Write locks override our own fcntl readlocks, so check it here.
Note this is meant to be F_SETLK, *not* F_SETLKW, as it's not
an error to fail to get the lock here.
*/
int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off)
{
struct tdb_traverse_lock *i;
for (i = &tdb->travlocks; i; i = i->next)
if (i->off == off)
return -1;
if (tdb->allrecord_lock.count) {
if (tdb->allrecord_lock.ltype == F_WRLCK) {
return 0;
}
return -1;
}
return tdb_brlock(tdb, F_WRLCK, off, 1, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
}
int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off)
{
if (tdb->allrecord_lock.count) {
return 0;
}
return tdb_brunlock(tdb, F_WRLCK, off, 1);
}
/* fcntl locks don't stack: avoid unlocking someone else's */
int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off)
{
struct tdb_traverse_lock *i;
uint32_t count = 0;
if (tdb->allrecord_lock.count) {
return 0;
}
if (off == 0)
return 0;
for (i = &tdb->travlocks; i; i = i->next)
if (i->off == off)
count++;
return (count == 1 ? tdb_brunlock(tdb, F_RDLCK, off, 1) : 0);
}
bool tdb_have_extra_locks(struct tdb_context *tdb)
{
unsigned int extra = tdb->num_lockrecs;
/* A transaction holds the lock for all records. */
if (!tdb->transaction && tdb->allrecord_lock.count) {
return true;
}
/* We always hold the active lock if CLEAR_IF_FIRST. */
if (find_nestlock(tdb, ACTIVE_LOCK)) {
extra--;
}
/* In a transaction, we expect to hold the transaction lock */
if (tdb->transaction && find_nestlock(tdb, TRANSACTION_LOCK)) {
extra--;
}
return extra;
}
/* The transaction code uses this to remove all locks. Note that this
may include OPEN_LOCK. */
void tdb_release_extra_locks(struct tdb_context *tdb)
{
unsigned int i, extra = 0;
if (tdb->allrecord_lock.count != 0) {
tdb_brunlock(tdb, tdb->allrecord_lock.ltype, FREELIST_TOP, 0);
tdb->allrecord_lock.count = 0;
}
for (i=0;i<tdb->num_lockrecs;i++) {
struct tdb_lock_type *lck = &tdb->lockrecs[i];
/* Don't release transaction or active locks! */
if (tdb->transaction && lck->off == TRANSACTION_LOCK) {
tdb->lockrecs[extra++] = *lck;
} else if (lck->off == ACTIVE_LOCK) {
tdb->lockrecs[extra++] = *lck;
} else {
tdb_brunlock(tdb, lck->ltype, lck->off, 1);
}
}
tdb->num_lockrecs = extra;
if (tdb->num_lockrecs == 0) {
SAFE_FREE(tdb->lockrecs);
}
}
ccan/tdb/open.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
trivial database library
Copyright (C) Andrew Tridgell 1999-2005
Copyright (C) Paul `Rusty' Russell 2000
Copyright (C) Jeremy Allison 2000-2003
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
/* all contexts, to ensure no double-opens (fcntl locks don't nest!) */
static struct tdb_context *tdbs = NULL;
/* We use two hashes to double-check they're using the right hash function. */
void tdb_header_hash(struct tdb_context *tdb,
uint32_t *magic1_hash, uint32_t *magic2_hash)
{
TDB_DATA hash_key;
uint32_t tdb_magic = TDB_MAGIC;
hash_key.dptr = (unsigned char *)TDB_MAGIC_FOOD;
hash_key.dsize = sizeof(TDB_MAGIC_FOOD);
*magic1_hash = tdb->hash_fn(&hash_key);
hash_key.dptr = CONVERT(tdb_magic);
hash_key.dsize = sizeof(tdb_magic);
*magic2_hash = tdb->hash_fn(&hash_key);
/* Make sure at least one hash is non-zero! */
if (*magic1_hash == 0 && *magic2_hash == 0)
*magic1_hash = 1;
}
/* initialise a new database with a specified hash size */
static int tdb_new_database(struct tdb_context *tdb, int hash_size)
{
struct tdb_header *newdb;
size_t size;
int ret = -1;
ssize_t written;
/* We make it up in memory, then write it out if not internal */
size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
if (!(newdb = (struct tdb_header *)calloc(size, 1))) {
tdb->ecode = TDB_ERR_OOM;
return -1;
}
/* Fill in the header */
newdb->version = TDB_VERSION;
newdb->hash_size = hash_size;
tdb_header_hash(tdb, &newdb->magic1_hash, &newdb->magic2_hash);
/* Make sure older tdbs (which don't check the magic hash fields)
* will refuse to open this TDB. */
if (tdb->flags & TDB_INCOMPATIBLE_HASH)
newdb->rwlocks = TDB_HASH_RWLOCK_MAGIC;
if (tdb->flags & TDB_INTERNAL) {
tdb->map_size = size;
tdb->map_ptr = (char *)newdb;
memcpy(&tdb->header, newdb, sizeof(tdb->header));
/* Convert the `ondisk' version if asked. */
CONVERT(*newdb);
return 0;
}
if (lseek(tdb->fd, 0, SEEK_SET) == -1)
goto fail;
if (ftruncate(tdb->fd, 0) == -1)
goto fail;
/* This creates an endian-converted header, as if read from disk */
CONVERT(*newdb);
memcpy(&tdb->header, newdb, sizeof(tdb->header));
/* Don't endian-convert the magic food! */
memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
/* we still have "ret == -1" here */
written = write(tdb->fd, newdb, size);
if (written == size) {
ret = 0;
} else if (written != -1) {
/* call write once again, this usually should return -1 and
* set errno appropriately */
size -= written;
written = write(tdb->fd, newdb+written, size);
if (written == size) {
ret = 0;
} else if (written >= 0) {
/* a second incomplete write - we give up.
* guessing the errno... */
errno = ENOSPC;
}
}
fail:
SAFE_FREE(newdb);
return ret;
}
static int tdb_already_open(dev_t device,
ino_t ino)
{
struct tdb_context *i;
for (i = tdbs; i; i = i->next) {
if (i->device == device && i->inode == ino) {
return 1;
}
}
return 0;
}
/* open the database, creating it if necessary
The open_flags and mode are passed straight to the open call on the
database file. A flags value of O_WRONLY is invalid. The hash size
is advisory, use zero for a default value.
Return is NULL on error, in which case errno is also set. Don't
try to call tdb_error or tdb_errname, just do strerror(errno).
@param name may be NULL for internal databases. */
struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
int open_flags, mode_t mode)
{
return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL, NULL);
}
/* a default logging function */
static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...) PRINTF_FMT(3, 4);
static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
{
}
static bool check_header_hash(struct tdb_context *tdb,
bool default_hash, uint32_t *m1, uint32_t *m2)
{
tdb_header_hash(tdb, m1, m2);
if (tdb->header.magic1_hash == *m1 &&
tdb->header.magic2_hash == *m2) {
return true;
}
/* If they explicitly set a hash, always respect it. */
if (!default_hash)
return false;
/* Otherwise, try the other inbuilt hash. */
if (tdb->hash_fn == tdb_old_hash)
tdb->hash_fn = tdb_jenkins_hash;
else
tdb->hash_fn = tdb_old_hash;
return check_header_hash(tdb, false, m1, m2);
}
struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
int open_flags, mode_t mode,
const struct tdb_logging_context *log_ctx,
tdb_hash_func hash_fn)
{
struct tdb_context *tdb;
struct stat st;
int rev = 0, locked = 0;
unsigned char *vp;
uint32_t vertest;
unsigned v;
const char *hash_alg;
uint32_t magic1, magic2;
if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
/* Can't log this */
errno = ENOMEM;
goto fail;
}
tdb_io_init(tdb);
tdb->fd = -1;
#ifdef TDB_TRACE
tdb->tracefd = -1;
#endif
tdb->name = NULL;
tdb->map_ptr = NULL;
tdb->flags = tdb_flags;
tdb->open_flags = open_flags;
if (log_ctx) {
tdb->log = *log_ctx;
} else {
tdb->log.log_fn = null_log_fn;
tdb->log.log_private = NULL;
}
if (hash_fn) {
tdb->hash_fn = hash_fn;
hash_alg = "the user defined";
} else {
/* This controls what we use when creating a tdb. */
if (tdb->flags & TDB_INCOMPATIBLE_HASH) {
tdb->hash_fn = tdb_jenkins_hash;
} else {
tdb->hash_fn = tdb_old_hash;
}
hash_alg = "either default";
}
/* cache the page size */
tdb->page_size = getpagesize();
if (tdb->page_size <= 0) {
tdb->page_size = 0x2000;
}
tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
if ((open_flags & O_ACCMODE) == O_WRONLY) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't open tdb %s write-only\n",
name));
errno = EINVAL;
goto fail;
}
if (hash_size == 0)
hash_size = DEFAULT_HASH_SIZE;
if ((open_flags & O_ACCMODE) == O_RDONLY) {
tdb->read_only = 1;
/* read only databases don't do locking or clear if first */
tdb->flags |= TDB_NOLOCK;
tdb->flags &= ~TDB_CLEAR_IF_FIRST;
}
if ((tdb->flags & TDB_ALLOW_NESTING) &&
(tdb->flags & TDB_DISALLOW_NESTING)) {
tdb->ecode = TDB_ERR_NESTING;
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
"allow_nesting and disallow_nesting are not allowed together!"));
errno = EINVAL;
goto fail;
}
/*
* TDB_DISALLOW_NESTING is the default behavior.
*/
if (!(tdb->flags & TDB_ALLOW_NESTING)) {
tdb->flags |= TDB_DISALLOW_NESTING;
}
/* internal databases don't mmap or lock, and start off cleared */
if (tdb->flags & TDB_INTERNAL) {
tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
tdb->flags &= ~TDB_CLEAR_IF_FIRST;
if (tdb_new_database(tdb, hash_size) != 0) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
goto fail;
}
goto internal;
}
if ((tdb->fd = open(name, open_flags, mode)) == -1) {
TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_open_ex: could not open file %s: %s\n",
name, strerror(errno)));
goto fail; /* errno set by open(2) */
}
/* on exec, don't inherit the fd */
v = fcntl(tdb->fd, F_GETFD, 0);
fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
/* ensure there is only one process initialising at once */
if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get open lock on %s: %s\n",
name, strerror(errno)));
goto fail; /* errno set by tdb_brlock */
}
/* we need to zero database if we are the only one with it open */
if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
(!tdb->read_only) &&
(locked = (tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE) == 0))) {
open_flags |= O_CREAT;
if (ftruncate(tdb->fd, 0) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
"failed to truncate %s: %s\n",
name, strerror(errno)));
goto fail; /* errno set by ftruncate */
}
}
errno = 0;
if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
|| strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
if (errno == 0) {
errno = EIO; /* ie bad format or something */
}
goto fail;
}
rev = (tdb->flags & TDB_CONVERT);
} else if (tdb->header.version != TDB_VERSION
&& !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION)))) {
/* wrong version */
errno = EIO;
goto fail;
}
vp = (unsigned char *)&tdb->header.version;
vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
(((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
if (!rev)
tdb->flags &= ~TDB_CONVERT;
else {
tdb->flags |= TDB_CONVERT;
tdb_convert(&tdb->header, sizeof(tdb->header));
}
if (fstat(tdb->fd, &st) == -1)
goto fail;
if (tdb->header.rwlocks != 0 &&
tdb->header.rwlocks != TDB_HASH_RWLOCK_MAGIC) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
goto fail;
}
if ((tdb->header.magic1_hash == 0) && (tdb->header.magic2_hash == 0)) {
/* older TDB without magic hash references */
tdb->hash_fn = tdb_old_hash;
} else if (!check_header_hash(tdb, !hash_fn, &magic1, &magic2)) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
"%s was not created with %s hash function we are using\n"
"magic1_hash[0x%08X %s 0x%08X] "
"magic2_hash[0x%08X %s 0x%08X]\n",
name, hash_alg,
tdb->header.magic1_hash,
(tdb->header.magic1_hash == magic1) ? "==" : "!=",
magic1,
tdb->header.magic2_hash,
(tdb->header.magic2_hash == magic2) ? "==" : "!=",
magic2));
errno = EINVAL;
goto fail;
}
/* Is it already in the open list? If so, fail. */
if (tdb_already_open(st.st_dev, st.st_ino)) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
"%s (%d,%d) is already open in this process\n",
name, (int)st.st_dev, (int)st.st_ino));
errno = EBUSY;
goto fail;
}
if (!(tdb->name = (char *)strdup(name))) {
errno = ENOMEM;
goto fail;
}
tdb->map_size = st.st_size;
tdb->device = st.st_dev;
tdb->inode = st.st_ino;
tdb_mmap(tdb);
if (locked) {
if (tdb_nest_unlock(tdb, ACTIVE_LOCK, F_WRLCK, false) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
"failed to take ACTIVE_LOCK on %s: %s\n",
name, strerror(errno)));
goto fail;
}
}
/* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
we didn't get the initial exclusive lock as we need to let all other
users know we're using it. */
if (tdb_flags & TDB_CLEAR_IF_FIRST) {
/* leave this lock in place to indicate it's in use */
if (tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
goto fail;
}
}
/* if needed, run recovery */
if (tdb_transaction_recover(tdb) == -1) {
goto fail;
}
#ifdef TDB_TRACE
{
char tracefile[strlen(name) + 32];
snprintf(tracefile, sizeof(tracefile),
"%s.trace.%li", name, (long)getpid());
tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
if (tdb->tracefd >= 0) {
tdb_enable_seqnum(tdb);
tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
open_flags);
} else
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
}
#endif
internal:
/* Internal (memory-only) databases skip all the code above to
* do with disk files, and resume here by releasing their
* open lock and hooking into the active list. */
if (tdb_nest_unlock(tdb, OPEN_LOCK, F_WRLCK, false) == -1) {
goto fail;
}
tdb->next = tdbs;
tdbs = tdb;
return tdb;
fail:
{ int save_errno = errno;
if (!tdb)
return NULL;
#ifdef TDB_TRACE
close(tdb->tracefd);
#endif
if (tdb->map_ptr) {
if (tdb->flags & TDB_INTERNAL)
SAFE_FREE(tdb->map_ptr);
else
tdb_munmap(tdb);
}
SAFE_FREE(tdb->name);
if (tdb->fd != -1)
if (close(tdb->fd) != 0)
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
SAFE_FREE(tdb->lockrecs);
SAFE_FREE(tdb);
errno = save_errno;
return NULL;
}
}
/*
* Set the maximum number of dead records per hash chain
*/
void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
{
tdb->max_dead_records = max_dead;
}
/**
* Close a database.
*
* @returns -1 for error; 0 for success.
**/
int tdb_close(struct tdb_context *tdb)
{
struct tdb_context **i;
int ret = 0;
if (tdb->transaction) {
tdb_transaction_cancel(tdb);
}
tdb_trace(tdb, "tdb_close");
if (tdb->map_ptr) {
if (tdb->flags & TDB_INTERNAL)
SAFE_FREE(tdb->map_ptr);
else
tdb_munmap(tdb);
}
SAFE_FREE(tdb->name);
if (tdb->fd != -1) {
ret = close(tdb->fd);
tdb->fd = -1;
}
SAFE_FREE(tdb->lockrecs);
/* Remove from contexts list */
for (i = &tdbs; *i; i = &(*i)->next) {
if (*i == tdb) {
*i = tdb->next;
break;
}
}
#ifdef TDB_TRACE
close(tdb->tracefd);
#endif
memset(tdb, 0, sizeof(*tdb));
SAFE_FREE(tdb);
return ret;
}
/* register a loging function */
void tdb_set_logging_function(struct tdb_context *tdb,
const struct tdb_logging_context *log_ctx)
{
tdb->log = *log_ctx;
}
void *tdb_get_logging_private(struct tdb_context *tdb)
{
return tdb->log.log_private;
}
static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
{
struct stat st;
if (tdb->flags & TDB_INTERNAL) {
return 0; /* Nothing to do. */
}
if (tdb_have_extra_locks(tdb)) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
goto fail;
}
if (tdb->transaction != 0) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
goto fail;
}
if (tdb_munmap(tdb) != 0) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
goto fail;
}
if (close(tdb->fd) != 0)
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
if (tdb->fd == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
goto fail;
}
if (fstat(tdb->fd, &st) != 0) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
goto fail;
}
if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
goto fail;
}
tdb_mmap(tdb);
/* We may still think we hold the active lock. */
tdb->num_lockrecs = 0;
SAFE_FREE(tdb->lockrecs);
if (active_lock && tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
goto fail;
}
return 0;
fail:
tdb_close(tdb);
return -1;
}
/* reopen a tdb - this can be used after a fork to ensure that we have an independent
seek pointer from our parent and to re-establish locks */
int tdb_reopen(struct tdb_context *tdb)
{
return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
}
/* reopen all tdb's */
int tdb_reopen_all(int parent_longlived)
{
struct tdb_context *tdb;
for (tdb=tdbs; tdb; tdb = tdb->next) {
bool active_lock = (tdb->flags & TDB_CLEAR_IF_FIRST);
/*
* If the parent is longlived (ie. a
* parent daemon architecture), we know
* it will keep it's active lock on a
* tdb opened with CLEAR_IF_FIRST. Thus
* for child processes we don't have to
* add an active lock. This is essential
* to improve performance on systems that
* keep POSIX locks as a non-scalable data
* structure in the kernel.
*/
if (parent_longlived) {
/* Ensure no clear-if-first. */
active_lock = false;
}
if (tdb_reopen_internal(tdb, active_lock) != 0)
return -1;
}
return 0;
}
ccan/tdb/summary.c deleted 100644 → 0
View file @ 98b8ada2
/*
Trivial Database: human-readable summary code
Copyright (C) Rusty Russell 2010
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
#define SUMMARY_FORMAT \
"Size of file/data: %u/%zu\n" \
"Number of records: %zu\n" \
"Smallest/average/largest keys: %zu/%zu/%zu\n" \
"Smallest/average/largest data: %zu/%zu/%zu\n" \
"Smallest/average/largest padding: %zu/%zu/%zu\n" \
"Number of dead records: %zu\n" \
"Smallest/average/largest dead records: %zu/%zu/%zu\n" \
"Number of free records: %zu\n" \
"Smallest/average/largest free records: %zu/%zu/%zu\n" \
"Number of hash chains: %zu\n" \
"Smallest/average/largest hash chains: %zu/%zu/%zu\n" \
"Number of uncoalesced records: %zu\n" \
"Smallest/average/largest uncoalesced runs: %zu/%zu/%zu\n" \
"Percentage keys/data/padding/free/dead/rechdrs&tailers/hashes: %.0f/%.0f/%.0f/%.0f/%.0f/%.0f/%.0f\n"
/* We don't use tally module, to keep upstream happy. */
struct tally {
size_t min, max, total;
size_t num;
};
static void tally_init(struct tally *tally)
{
tally->total = 0;
tally->num = 0;
tally->min = tally->max = 0;
}
static void tally_add(struct tally *tally, size_t len)
{
if (tally->num == 0)
tally->max = tally->min = len;
else if (len > tally->max)
tally->max = len;
else if (len < tally->min)
tally->min = len;
tally->num++;
tally->total += len;
}
static size_t tally_mean(const struct tally *tally)
{
if (!tally->num)
return 0;
return tally->total / tally->num;
}
static size_t get_hash_length(struct tdb_context *tdb, unsigned int i)
{
tdb_off_t rec_ptr;
size_t count = 0;
if (tdb_ofs_read(tdb, TDB_HASH_TOP(i), &rec_ptr) == -1)
return 0;
/* keep looking until we find the right record */
while (rec_ptr) {
struct tdb_record r;
++count;
if (tdb_rec_read(tdb, rec_ptr, &r) == -1)
return 0;
rec_ptr = r.next;
}
return count;
}
char *tdb_summary(struct tdb_context *tdb)
{
tdb_off_t off;
struct tally freet, keys, data, dead, extra, hash, uncoal;
struct tdb_record rec;
char *ret = NULL;
bool locked;
size_t len, unc = 0;
/* Read-only databases use no locking at all: it's best-effort.
* We may have a write lock already, so skip that case too. */
if (tdb->read_only || tdb->allrecord_lock.count != 0) {
locked = false;
} else {
if (tdb_lockall_read(tdb) == -1)
return NULL;
locked = true;
}
tally_init(&freet);
tally_init(&keys);
tally_init(&data);
tally_init(&dead);
tally_init(&extra);
tally_init(&hash);
tally_init(&uncoal);
for (off = TDB_DATA_START(tdb->header.hash_size);
off < tdb->map_size - 1;
off += sizeof(rec) + rec.rec_len) {
if (tdb->methods->tdb_read(tdb, off, &rec, sizeof(rec),
DOCONV()) == -1)
goto unlock;
switch (rec.magic) {
case TDB_MAGIC:
tally_add(&keys, rec.key_len);
tally_add(&data, rec.data_len);
tally_add(&extra, rec.rec_len - (rec.key_len
+ rec.data_len));
if (unc > 1)
tally_add(&uncoal, unc - 1);
unc = 0;
break;
case TDB_FREE_MAGIC:
tally_add(&freet, rec.rec_len);
unc++;
break;
/* If we crash after ftruncate, we can get zeroes or fill. */
case TDB_RECOVERY_INVALID_MAGIC:
case 0x42424242:
unc++;
rec.rec_len = tdb_dead_space(tdb, off) - sizeof(rec);
/* Fall through */
case TDB_DEAD_MAGIC:
tally_add(&dead, rec.rec_len);
break;
default:
TDB_LOG((tdb, TDB_DEBUG_ERROR,
"Unexpected record magic 0x%x at offset %d\n",
rec.magic, off));
goto unlock;
}
}
if (unc > 1)
tally_add(&uncoal, unc - 1);
for (off = 0; off < tdb->header.hash_size; off++)
tally_add(&hash, get_hash_length(tdb, off));
/* 20 is max length of a %zu. */
len = strlen(SUMMARY_FORMAT) + 35*20 + 1;
ret = malloc(len);
if (!ret)
goto unlock;
sprintf(ret, SUMMARY_FORMAT,
tdb->map_size, keys.total+data.total,
keys.num,
keys.min, tally_mean(&keys), keys.max,
data.min, tally_mean(&data), data.max,
extra.min, tally_mean(&extra), extra.max,
dead.num,
dead.min, tally_mean(&dead), dead.max,
freet.num,
freet.min, tally_mean(&freet), freet.max,
hash.num,
hash.min, tally_mean(&hash), hash.max,
uncoal.total,
uncoal.min, tally_mean(&uncoal), uncoal.max,
keys.total * 100.0 / tdb->map_size,
data.total * 100.0 / tdb->map_size,
extra.total * 100.0 / tdb->map_size,
freet.total * 100.0 / tdb->map_size,
dead.total * 100.0 / tdb->map_size,
(keys.num + freet.num + dead.num)
* (sizeof(struct tdb_record) + sizeof(uint32_t))
* 100.0 / tdb->map_size,
tdb->header.hash_size * sizeof(tdb_off_t)
* 100.0 / tdb->map_size);
unlock:
if (locked) {
tdb_unlockall_read(tdb);
}
return ret;
}
ccan/tdb/tdb.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
trivial database library
Copyright (C) Andrew Tridgell 1999-2005
Copyright (C) Paul `Rusty' Russell 2000
Copyright (C) Jeremy Allison 2000-2003
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
TDB_DATA tdb_null;
/*
non-blocking increment of the tdb sequence number if the tdb has been opened using
the TDB_SEQNUM flag
*/
void tdb_increment_seqnum_nonblock(struct tdb_context *tdb)
{
tdb_off_t seqnum=0;
if (!(tdb->flags & TDB_SEQNUM)) {
return;
}
/* we ignore errors from this, as we have no sane way of
dealing with them.
*/
tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
seqnum++;
tdb_ofs_write(tdb, TDB_SEQNUM_OFS, &seqnum);
}
/*
increment the tdb sequence number if the tdb has been opened using
the TDB_SEQNUM flag
*/
static void tdb_increment_seqnum(struct tdb_context *tdb)
{
if (!(tdb->flags & TDB_SEQNUM)) {
return;
}
if (tdb_nest_lock(tdb, TDB_SEQNUM_OFS, F_WRLCK,
TDB_LOCK_WAIT|TDB_LOCK_PROBE) != 0) {
return;
}
tdb_increment_seqnum_nonblock(tdb);
tdb_nest_unlock(tdb, TDB_SEQNUM_OFS, F_WRLCK, false);
}
static int tdb_key_compare(TDB_DATA key, TDB_DATA data, void *private_data)
{
return memcmp(data.dptr, key.dptr, data.dsize);
}
/* Returns 0 on fail. On success, return offset of record, and fills
in rec */
static tdb_off_t tdb_find(struct tdb_context *tdb, TDB_DATA key, uint32_t hash,
struct tdb_record *r)
{
tdb_off_t rec_ptr;
/* read in the hash top */
if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
return 0;
/* keep looking until we find the right record */
while (rec_ptr) {
if (tdb_rec_read(tdb, rec_ptr, r) == -1)
return 0;
if (!TDB_DEAD(r) && hash==r->full_hash
&& key.dsize==r->key_len
&& tdb_parse_data(tdb, key, rec_ptr + sizeof(*r),
r->key_len, tdb_key_compare,
NULL) == 0) {
return rec_ptr;
}
/* detect tight infinite loop */
if (rec_ptr == r->next) {
tdb->ecode = TDB_ERR_CORRUPT;
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_find: loop detected.\n"));
return 0;
}
rec_ptr = r->next;
}
tdb->ecode = TDB_ERR_NOEXIST;
return 0;
}
/* As tdb_find, but if you succeed, keep the lock */
tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, int locktype,
struct tdb_record *rec)
{
uint32_t rec_ptr;
if (tdb_lock(tdb, BUCKET(hash), locktype) == -1)
return 0;
if (!(rec_ptr = tdb_find(tdb, key, hash, rec)))
tdb_unlock(tdb, BUCKET(hash), locktype);
return rec_ptr;
}
static TDB_DATA _tdb_fetch(struct tdb_context *tdb, TDB_DATA key);
/* update an entry in place - this only works if the new data size
is <= the old data size and the key exists.
on failure return -1.
*/
static int tdb_update_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, TDB_DATA dbuf)
{
struct tdb_record rec;
tdb_off_t rec_ptr;
/* find entry */
if (!(rec_ptr = tdb_find(tdb, key, hash, &rec)))
return -1;
/* it could be an exact duplicate of what is there - this is
* surprisingly common (eg. with a ldb re-index). */
if (rec.key_len == key.dsize &&
rec.data_len == dbuf.dsize &&
rec.full_hash == hash) {
TDB_DATA data = _tdb_fetch(tdb, key);
if (data.dsize == dbuf.dsize &&
memcmp(data.dptr, dbuf.dptr, data.dsize) == 0) {
if (data.dptr) {
free(data.dptr);
}
return 0;
}
if (data.dptr) {
free(data.dptr);
}
}
/* must be long enough key, data and tailer */
if (rec.rec_len < key.dsize + dbuf.dsize + sizeof(tdb_off_t)) {
tdb->ecode = TDB_SUCCESS; /* Not really an error */
return -1;
}
if (tdb->methods->tdb_write(tdb, rec_ptr + sizeof(rec) + rec.key_len,
dbuf.dptr, dbuf.dsize) == -1)
return -1;
if (dbuf.dsize != rec.data_len) {
/* update size */
rec.data_len = dbuf.dsize;
return tdb_rec_write(tdb, rec_ptr, &rec);
}
return 0;
}
/* find an entry in the database given a key */
/* If an entry doesn't exist tdb_err will be set to
* TDB_ERR_NOEXIST. If a key has no data attached
* then the TDB_DATA will have zero length but
* a non-zero pointer
*/
static TDB_DATA _tdb_fetch(struct tdb_context *tdb, TDB_DATA key)
{
tdb_off_t rec_ptr;
struct tdb_record rec;
TDB_DATA ret;
uint32_t hash;
/* find which hash bucket it is in */
hash = tdb->hash_fn(&key);
if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec)))
return tdb_null;
ret.dptr = tdb_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len,
rec.data_len);
ret.dsize = rec.data_len;
tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
return ret;
}
TDB_DATA tdb_fetch(struct tdb_context *tdb, TDB_DATA key)
{
TDB_DATA ret = _tdb_fetch(tdb, key);
tdb_trace_1rec_retrec(tdb, "tdb_fetch", key, ret);
return ret;
}
/*
* Find an entry in the database and hand the record's data to a parsing
* function. The parsing function is executed under the chain read lock, so it
* should be fast and should not block on other syscalls.
*
* DON'T CALL OTHER TDB CALLS FROM THE PARSER, THIS MIGHT LEAD TO SEGFAULTS.
*
* For mmapped tdb's that do not have a transaction open it points the parsing
* function directly at the mmap area, it avoids the malloc/memcpy in this
* case. If a transaction is open or no mmap is available, it has to do
* malloc/read/parse/free.
*
* This is interesting for all readers of potentially large data structures in
* the tdb records, ldb indexes being one example.
*/
int tdb_parse_record(struct tdb_context *tdb, TDB_DATA key,
int (*parser)(TDB_DATA key, TDB_DATA data,
void *private_data),
void *private_data)
{
tdb_off_t rec_ptr;
struct tdb_record rec;
int ret;
uint32_t hash;
/* find which hash bucket it is in */
hash = tdb->hash_fn(&key);
if (!(rec_ptr = tdb_find_lock_hash(tdb,key,hash,F_RDLCK,&rec))) {
tdb_trace_1rec_ret(tdb, "tdb_parse_record", key, -1);
tdb->ecode = TDB_ERR_NOEXIST;
return 0;
}
tdb_trace_1rec_ret(tdb, "tdb_parse_record", key, 0);
ret = tdb_parse_data(tdb, key, rec_ptr + sizeof(rec) + rec.key_len,
rec.data_len, parser, private_data);
tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
return ret;
}
/* check if an entry in the database exists
note that 1 is returned if the key is found and 0 is returned if not found
this doesn't match the conventions in the rest of this module, but is
compatible with gdbm
*/
static int tdb_exists_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash)
{
struct tdb_record rec;
if (tdb_find_lock_hash(tdb, key, hash, F_RDLCK, &rec) == 0)
return 0;
tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
return 1;
}
int tdb_exists(struct tdb_context *tdb, TDB_DATA key)
{
uint32_t hash = tdb->hash_fn(&key);
int ret;
ret = tdb_exists_hash(tdb, key, hash);
tdb_trace_1rec_ret(tdb, "tdb_exists", key, ret);
return ret;
}
/* actually delete an entry in the database given the offset */
int tdb_do_delete(struct tdb_context *tdb, tdb_off_t rec_ptr, struct tdb_record *rec)
{
tdb_off_t last_ptr, i;
struct tdb_record lastrec;
if (tdb->read_only || tdb->traverse_read) return -1;
if (((tdb->traverse_write != 0) && (!TDB_DEAD(rec))) ||
tdb_write_lock_record(tdb, rec_ptr) == -1) {
/* Someone traversing here: mark it as dead */
rec->magic = TDB_DEAD_MAGIC;
return tdb_rec_write(tdb, rec_ptr, rec);
}
if (tdb_write_unlock_record(tdb, rec_ptr) != 0)
return -1;
/* find previous record in hash chain */
if (tdb_ofs_read(tdb, TDB_HASH_TOP(rec->full_hash), &i) == -1)
return -1;
for (last_ptr = 0; i != rec_ptr; last_ptr = i, i = lastrec.next)
if (tdb_rec_read(tdb, i, &lastrec) == -1)
return -1;
/* unlink it: next ptr is at start of record. */
if (last_ptr == 0)
last_ptr = TDB_HASH_TOP(rec->full_hash);
if (tdb_ofs_write(tdb, last_ptr, &rec->next) == -1)
return -1;
/* recover the space */
if (tdb_free(tdb, rec_ptr, rec) == -1)
return -1;
return 0;
}
static int tdb_count_dead(struct tdb_context *tdb, uint32_t hash)
{
int res = 0;
tdb_off_t rec_ptr;
struct tdb_record rec;
/* read in the hash top */
if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
return 0;
while (rec_ptr) {
if (tdb_rec_read(tdb, rec_ptr, &rec) == -1)
return 0;
if (rec.magic == TDB_DEAD_MAGIC) {
res += 1;
}
rec_ptr = rec.next;
}
return res;
}
/*
* Purge all DEAD records from a hash chain
*/
static int tdb_purge_dead(struct tdb_context *tdb, uint32_t hash)
{
int res = -1;
struct tdb_record rec;
tdb_off_t rec_ptr;
if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
return -1;
}
/* read in the hash top */
if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
goto fail;
while (rec_ptr) {
tdb_off_t next;
if (tdb_rec_read(tdb, rec_ptr, &rec) == -1) {
goto fail;
}
next = rec.next;
if (rec.magic == TDB_DEAD_MAGIC
&& tdb_do_delete(tdb, rec_ptr, &rec) == -1) {
goto fail;
}
rec_ptr = next;
}
res = 0;
fail:
tdb_unlock(tdb, -1, F_WRLCK);
return res;
}
/* delete an entry in the database given a key */
static int tdb_delete_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash)
{
tdb_off_t rec_ptr;
struct tdb_record rec;
int ret;
if (tdb->max_dead_records != 0) {
/*
* Allow for some dead records per hash chain, mainly for
* tdb's with a very high create/delete rate like locking.tdb.
*/
if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
return -1;
if (tdb_count_dead(tdb, hash) >= tdb->max_dead_records) {
/*
* Don't let the per-chain freelist grow too large,
* delete all existing dead records
*/
tdb_purge_dead(tdb, hash);
}
if (!(rec_ptr = tdb_find(tdb, key, hash, &rec))) {
tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
return -1;
}
/*
* Just mark the record as dead.
*/
rec.magic = TDB_DEAD_MAGIC;
ret = tdb_rec_write(tdb, rec_ptr, &rec);
}
else {
if (!(rec_ptr = tdb_find_lock_hash(tdb, key, hash, F_WRLCK,
&rec)))
return -1;
ret = tdb_do_delete(tdb, rec_ptr, &rec);
}
if (ret == 0) {
tdb_increment_seqnum(tdb);
}
if (tdb_unlock(tdb, BUCKET(rec.full_hash), F_WRLCK) != 0)
TDB_LOG((tdb, TDB_DEBUG_WARNING, "tdb_delete: WARNING tdb_unlock failed!\n"));
return ret;
}
int tdb_delete(struct tdb_context *tdb, TDB_DATA key)
{
uint32_t hash = tdb->hash_fn(&key);
int ret;
ret = tdb_delete_hash(tdb, key, hash);
tdb_trace_1rec_ret(tdb, "tdb_delete", key, ret);
return ret;
}
/*
* See if we have a dead record around with enough space
*/
static tdb_off_t tdb_find_dead(struct tdb_context *tdb, uint32_t hash,
struct tdb_record *r, tdb_len_t length)
{
tdb_off_t rec_ptr;
/* read in the hash top */
if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1)
return 0;
/* keep looking until we find the right record */
while (rec_ptr) {
if (tdb_rec_read(tdb, rec_ptr, r) == -1)
return 0;
if (TDB_DEAD(r) && r->rec_len >= length) {
/*
* First fit for simple coding, TODO: change to best
* fit
*/
return rec_ptr;
}
rec_ptr = r->next;
}
return 0;
}
static int _tdb_store(struct tdb_context *tdb, TDB_DATA key,
TDB_DATA dbuf, int flag, uint32_t hash)
{
struct tdb_record rec;
tdb_off_t rec_ptr;
char *p = NULL;
int ret = -1;
/* check for it existing, on insert. */
if (flag == TDB_INSERT) {
if (tdb_exists_hash(tdb, key, hash)) {
tdb->ecode = TDB_ERR_EXISTS;
goto fail;
}
} else {
/* first try in-place update, on modify or replace. */
if (tdb_update_hash(tdb, key, hash, dbuf) == 0) {
goto done;
}
if (tdb->ecode == TDB_ERR_NOEXIST &&
flag == TDB_MODIFY) {
/* if the record doesn't exist and we are in TDB_MODIFY mode then
we should fail the store */
goto fail;
}
}
/* reset the error code potentially set by the tdb_update() */
tdb->ecode = TDB_SUCCESS;
/* delete any existing record - if it doesn't exist we don't
care. Doing this first reduces fragmentation, and avoids
coalescing with `allocated' block before it's updated. */
if (flag != TDB_INSERT)
tdb_delete_hash(tdb, key, hash);
/* Copy key+value *before* allocating free space in case malloc
fails and we are left with a dead spot in the tdb. */
if (!(p = (char *)malloc(key.dsize + dbuf.dsize))) {
tdb->ecode = TDB_ERR_OOM;
goto fail;
}
memcpy(p, key.dptr, key.dsize);
if (dbuf.dsize)
memcpy(p+key.dsize, dbuf.dptr, dbuf.dsize);
if (tdb->max_dead_records != 0) {
/*
* Allow for some dead records per hash chain, look if we can
* find one that can hold the new record. We need enough space
* for key, data and tailer. If we find one, we don't have to
* consult the central freelist.
*/
rec_ptr = tdb_find_dead(
tdb, hash, &rec,
key.dsize + dbuf.dsize + sizeof(tdb_off_t));
if (rec_ptr != 0) {
rec.key_len = key.dsize;
rec.data_len = dbuf.dsize;
rec.full_hash = hash;
rec.magic = TDB_MAGIC;
if (tdb_rec_write(tdb, rec_ptr, &rec) == -1
|| tdb->methods->tdb_write(
tdb, rec_ptr + sizeof(rec),
p, key.dsize + dbuf.dsize) == -1) {
goto fail;
}
goto done;
}
}
/*
* We have to allocate some space from the freelist, so this means we
* have to lock it. Use the chance to purge all the DEAD records from
* the hash chain under the freelist lock.
*/
if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
goto fail;
}
if ((tdb->max_dead_records != 0)
&& (tdb_purge_dead(tdb, hash) == -1)) {
tdb_unlock(tdb, -1, F_WRLCK);
goto fail;
}
/* we have to allocate some space */
rec_ptr = tdb_allocate(tdb, key.dsize + dbuf.dsize, &rec);
tdb_unlock(tdb, -1, F_WRLCK);
if (rec_ptr == 0) {
goto fail;
}
/* Read hash top into next ptr */
if (tdb_ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
goto fail;
rec.key_len = key.dsize;
rec.data_len = dbuf.dsize;
rec.full_hash = hash;
rec.magic = TDB_MAGIC;
/* write out and point the top of the hash chain at it */
if (tdb_rec_write(tdb, rec_ptr, &rec) == -1
|| tdb->methods->tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1
|| tdb_ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
/* Need to tdb_unallocate() here */
goto fail;
}
done:
ret = 0;
fail:
if (ret == 0) {
tdb_increment_seqnum(tdb);
}
SAFE_FREE(p);
return ret;
}
/* store an element in the database, replacing any existing element
with the same key
return 0 on success, -1 on failure
*/
int tdb_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
{
uint32_t hash;
int ret;
if (tdb->read_only || tdb->traverse_read) {
tdb->ecode = TDB_ERR_RDONLY;
tdb_trace_2rec_flag_ret(tdb, "tdb_store", key, dbuf, flag, -1);
return -1;
}
/* find which hash bucket it is in */
hash = tdb->hash_fn(&key);
if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
return -1;
ret = _tdb_store(tdb, key, dbuf, flag, hash);
tdb_trace_2rec_flag_ret(tdb, "tdb_store", key, dbuf, flag, ret);
tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
return ret;
}
/* Append to an entry. Create if not exist. */
int tdb_append(struct tdb_context *tdb, TDB_DATA key, TDB_DATA new_dbuf)
{
uint32_t hash;
TDB_DATA dbuf;
int ret = -1;
/* find which hash bucket it is in */
hash = tdb->hash_fn(&key);
if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
return -1;
dbuf = _tdb_fetch(tdb, key);
if (dbuf.dptr == NULL) {
dbuf.dptr = (unsigned char *)malloc(new_dbuf.dsize);
} else {
unsigned int new_len = dbuf.dsize + new_dbuf.dsize;
unsigned char *new_dptr;
/* realloc '0' is special: don't do that. */
if (new_len == 0)
new_len = 1;
new_dptr = (unsigned char *)realloc(dbuf.dptr, new_len);
if (new_dptr == NULL) {
free(dbuf.dptr);
}
dbuf.dptr = new_dptr;
}
if (dbuf.dptr == NULL) {
tdb->ecode = TDB_ERR_OOM;
goto failed;
}
memcpy(dbuf.dptr + dbuf.dsize, new_dbuf.dptr, new_dbuf.dsize);
dbuf.dsize += new_dbuf.dsize;
ret = _tdb_store(tdb, key, dbuf, 0, hash);
tdb_trace_2rec_retrec(tdb, "tdb_append", key, new_dbuf, dbuf);
failed:
tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
SAFE_FREE(dbuf.dptr);
return ret;
}
/*
return the name of the current tdb file
useful for external logging functions
*/
const char *tdb_name(struct tdb_context *tdb)
{
return tdb->name;
}
/*
return the underlying file descriptor being used by tdb, or -1
useful for external routines that want to check the device/inode
of the fd
*/
int tdb_fd(struct tdb_context *tdb)
{
return tdb->fd;
}
/*
return the current logging function
useful for external tdb routines that wish to log tdb errors
*/
tdb_log_func tdb_log_fn(struct tdb_context *tdb)
{
return tdb->log.log_fn;
}
/*
get the tdb sequence number. Only makes sense if the writers opened
with TDB_SEQNUM set. Note that this sequence number will wrap quite
quickly, so it should only be used for a 'has something changed'
test, not for code that relies on the count of the number of changes
made. If you want a counter then use a tdb record.
The aim of this sequence number is to allow for a very lightweight
test of a possible tdb change.
*/
int tdb_get_seqnum(struct tdb_context *tdb)
{
tdb_off_t seqnum=0;
tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
return seqnum;
}
int tdb_hash_size(struct tdb_context *tdb)
{
return tdb->header.hash_size;
}
size_t tdb_map_size(struct tdb_context *tdb)
{
return tdb->map_size;
}
int tdb_get_flags(struct tdb_context *tdb)
{
return tdb->flags;
}
void tdb_add_flags(struct tdb_context *tdb, unsigned flags)
{
if ((flags & TDB_ALLOW_NESTING) &&
(flags & TDB_DISALLOW_NESTING)) {
tdb->ecode = TDB_ERR_NESTING;
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_add_flags: "
"allow_nesting and disallow_nesting are not allowed together!"));
return;
}
if (flags & TDB_ALLOW_NESTING) {
tdb->flags &= ~TDB_DISALLOW_NESTING;
}
if (flags & TDB_DISALLOW_NESTING) {
tdb->flags &= ~TDB_ALLOW_NESTING;
}
tdb->flags |= flags;
}
void tdb_remove_flags(struct tdb_context *tdb, unsigned flags)
{
if ((flags & TDB_ALLOW_NESTING) &&
(flags & TDB_DISALLOW_NESTING)) {
tdb->ecode = TDB_ERR_NESTING;
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_remove_flags: "
"allow_nesting and disallow_nesting are not allowed together!"));
return;
}
if (flags & TDB_ALLOW_NESTING) {
tdb->flags |= TDB_DISALLOW_NESTING;
}
if (flags & TDB_DISALLOW_NESTING) {
tdb->flags |= TDB_ALLOW_NESTING;
}
tdb->flags &= ~flags;
}
/*
enable sequence number handling on an open tdb
*/
void tdb_enable_seqnum(struct tdb_context *tdb)
{
tdb->flags |= TDB_SEQNUM;
}
/*
add a region of the file to the freelist. Length is the size of the region in bytes,
which includes the free list header that needs to be added
*/
static int tdb_free_region(struct tdb_context *tdb, tdb_off_t offset, ssize_t length)
{
struct tdb_record rec;
if (length <= sizeof(rec)) {
/* the region is not worth adding */
return 0;
}
if (length + offset > tdb->map_size) {
TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_free_region: adding region beyond end of file\n"));
return -1;
}
memset(&rec,'\0',sizeof(rec));
rec.rec_len = length - sizeof(rec);
if (tdb_free(tdb, offset, &rec) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_free_region: failed to add free record\n"));
return -1;
}
return 0;
}
/*
wipe the entire database, deleting all records. This can be done
very fast by using a allrecord lock. The entire data portion of the
file becomes a single entry in the freelist.
This code carefully steps around the recovery area, leaving it alone
*/
int tdb_wipe_all(struct tdb_context *tdb)
{
int i;
tdb_off_t offset = 0;
ssize_t data_len;
tdb_off_t recovery_head;
tdb_len_t recovery_size = 0;
if (tdb_lockall(tdb) != 0) {
return -1;
}
tdb_trace(tdb, "tdb_wipe_all");
/* see if the tdb has a recovery area, and remember its size
if so. We don't want to lose this as otherwise each
tdb_wipe_all() in a transaction will increase the size of
the tdb by the size of the recovery area */
if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_wipe_all: failed to read recovery head\n"));
goto failed;
}
if (recovery_head != 0) {
struct tdb_record rec;
if (tdb->methods->tdb_read(tdb, recovery_head, &rec, sizeof(rec), DOCONV()) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_wipe_all: failed to read recovery record\n"));
return -1;
}
recovery_size = rec.rec_len + sizeof(rec);
}
/* wipe the hashes */
for (i=0;i<tdb->header.hash_size;i++) {
if (tdb_ofs_write(tdb, TDB_HASH_TOP(i), &offset) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_wipe_all: failed to write hash %d\n", i));
goto failed;
}
}
/* wipe the freelist */
if (tdb_ofs_write(tdb, FREELIST_TOP, &offset) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_wipe_all: failed to write freelist\n"));
goto failed;
}
/* add all the rest of the file to the freelist, possibly leaving a gap
for the recovery area */
if (recovery_size == 0) {
/* the simple case - the whole file can be used as a freelist */
data_len = (tdb->map_size - TDB_DATA_START(tdb->header.hash_size));
if (tdb_free_region(tdb, TDB_DATA_START(tdb->header.hash_size), data_len) != 0) {
goto failed;
}
} else {
/* we need to add two freelist entries - one on either
side of the recovery area
Note that we cannot shift the recovery area during
this operation. Only the transaction.c code may
move the recovery area or we risk subtle data
corruption
*/
data_len = (recovery_head - TDB_DATA_START(tdb->header.hash_size));
if (tdb_free_region(tdb, TDB_DATA_START(tdb->header.hash_size), data_len) != 0) {
goto failed;
}
/* and the 2nd free list entry after the recovery area - if any */
data_len = tdb->map_size - (recovery_head+recovery_size);
if (tdb_free_region(tdb, recovery_head+recovery_size, data_len) != 0) {
goto failed;
}
}
if (tdb_unlockall(tdb) != 0) {
TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_wipe_all: failed to unlock\n"));
goto failed;
}
return 0;
failed:
tdb_unlockall(tdb);
return -1;
}
struct traverse_state {
bool error;
struct tdb_context *dest_db;
};
/*
traverse function for repacking
*/
static int repack_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data, void *private_data)
{
struct traverse_state *state = (struct traverse_state *)private_data;
if (tdb_store(state->dest_db, key, data, TDB_INSERT) != 0) {
state->error = true;
return -1;
}
return 0;
}
/*
repack a tdb
*/
int tdb_repack(struct tdb_context *tdb)
{
struct tdb_context *tmp_db;
struct traverse_state state;
tdb_trace(tdb, "tdb_repack");
if (tdb_transaction_start(tdb) != 0) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to start transaction\n"));
return -1;
}
tmp_db = tdb_open("tmpdb", tdb_hash_size(tdb), TDB_INTERNAL, O_RDWR|O_CREAT, 0);
if (tmp_db == NULL) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to create tmp_db\n"));
tdb_transaction_cancel(tdb);
return -1;
}
state.error = false;
state.dest_db = tmp_db;
if (tdb_traverse_read(tdb, repack_traverse, &state) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to traverse copying out\n"));
tdb_transaction_cancel(tdb);
tdb_close(tmp_db);
return -1;
}
if (state.error) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Error during traversal\n"));
tdb_transaction_cancel(tdb);
tdb_close(tmp_db);
return -1;
}
if (tdb_wipe_all(tdb) != 0) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to wipe database\n"));
tdb_transaction_cancel(tdb);
tdb_close(tmp_db);
return -1;
}
state.error = false;
state.dest_db = tdb;
if (tdb_traverse_read(tmp_db, repack_traverse, &state) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to traverse copying back\n"));
tdb_transaction_cancel(tdb);
tdb_close(tmp_db);
return -1;
}
if (state.error) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Error during second traversal\n"));
tdb_transaction_cancel(tdb);
tdb_close(tmp_db);
return -1;
}
tdb_close(tmp_db);
if (tdb_transaction_commit(tdb) != 0) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, __location__ " Failed to commit\n"));
return -1;
}
return 0;
}
#ifdef TDB_TRACE
static void tdb_trace_write(struct tdb_context *tdb, const char *str)
{
if (write(tdb->tracefd, str, strlen(str)) != strlen(str)) {
close(tdb->tracefd);
tdb->tracefd = -1;
}
}
static void tdb_trace_start(struct tdb_context *tdb)
{
tdb_off_t seqnum=0;
char msg[sizeof(tdb_off_t) * 4 + 1];
tdb_ofs_read(tdb, TDB_SEQNUM_OFS, &seqnum);
snprintf(msg, sizeof(msg), "%u ", seqnum);
tdb_trace_write(tdb, msg);
}
static void tdb_trace_end(struct tdb_context *tdb)
{
tdb_trace_write(tdb, "\n");
}
static void tdb_trace_end_ret(struct tdb_context *tdb, int ret)
{
char msg[sizeof(ret) * 4 + 4];
snprintf(msg, sizeof(msg), " = %i\n", ret);
tdb_trace_write(tdb, msg);
}
static void tdb_trace_record(struct tdb_context *tdb, TDB_DATA rec)
{
char msg[20 + rec.dsize*2], *p;
unsigned int i;
/* We differentiate zero-length records from non-existent ones. */
if (rec.dptr == NULL) {
tdb_trace_write(tdb, " NULL");
return;
}
/* snprintf here is purely cargo-cult programming. */
p = msg;
p += snprintf(p, sizeof(msg), " %zu:", rec.dsize);
for (i = 0; i < rec.dsize; i++)
p += snprintf(p, 2, "%02x", rec.dptr[i]);
tdb_trace_write(tdb, msg);
}
void tdb_trace(struct tdb_context *tdb, const char *op)
{
tdb_trace_start(tdb);
tdb_trace_write(tdb, op);
tdb_trace_end(tdb);
}
void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op)
{
char msg[sizeof(tdb_off_t) * 4 + 1];
snprintf(msg, sizeof(msg), "%u ", seqnum);
tdb_trace_write(tdb, msg);
tdb_trace_write(tdb, op);
tdb_trace_end(tdb);
}
void tdb_trace_open(struct tdb_context *tdb, const char *op,
unsigned hash_size, unsigned tdb_flags, unsigned open_flags)
{
char msg[128];
snprintf(msg, sizeof(msg),
"%s %u 0x%x 0x%x", op, hash_size, tdb_flags, open_flags);
tdb_trace_start(tdb);
tdb_trace_write(tdb, msg);
tdb_trace_end(tdb);
}
void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret)
{
tdb_trace_start(tdb);
tdb_trace_write(tdb, op);
tdb_trace_end_ret(tdb, ret);
}
void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret)
{
tdb_trace_start(tdb);
tdb_trace_write(tdb, op);
tdb_trace_write(tdb, " =");
tdb_trace_record(tdb, ret);
tdb_trace_end(tdb);
}
void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
TDB_DATA rec)
{
tdb_trace_start(tdb);
tdb_trace_write(tdb, op);
tdb_trace_record(tdb, rec);
tdb_trace_end(tdb);
}
void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
TDB_DATA rec, int ret)
{
tdb_trace_start(tdb);
tdb_trace_write(tdb, op);
tdb_trace_record(tdb, rec);
tdb_trace_end_ret(tdb, ret);
}
void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
TDB_DATA rec, TDB_DATA ret)
{
tdb_trace_start(tdb);
tdb_trace_write(tdb, op);
tdb_trace_record(tdb, rec);
tdb_trace_write(tdb, " =");
tdb_trace_record(tdb, ret);
tdb_trace_end(tdb);
}
void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
int ret)
{
char msg[1 + sizeof(ret) * 4];
snprintf(msg, sizeof(msg), " %#x", flag);
tdb_trace_start(tdb);
tdb_trace_write(tdb, op);
tdb_trace_record(tdb, rec1);
tdb_trace_record(tdb, rec2);
tdb_trace_write(tdb, msg);
tdb_trace_end_ret(tdb, ret);
}
void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret)
{
tdb_trace_start(tdb);
tdb_trace_write(tdb, op);
tdb_trace_record(tdb, rec1);
tdb_trace_record(tdb, rec2);
tdb_trace_write(tdb, " =");
tdb_trace_record(tdb, ret);
tdb_trace_end(tdb);
}
#endif
ccan/tdb/tdb.h deleted 100644 → 0
View file @ 98b8ada2
#ifndef __TDB_H__
#define __TDB_H__
/*
Unix SMB/CIFS implementation.
trivial database library
Copyright (C) Andrew Tridgell 1999-2004
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#ifdef __cplusplus
extern "C" {
#endif
#ifndef _SAMBA_BUILD_
/* For mode_t */
#include <sys/types.h>
/* For O_* flags. */
#include <sys/stat.h>
/* For sig_atomic_t. */
#include <signal.h>
#endif
#include <ccan/compiler/compiler.h>
/* flags to tdb_store() */
#define TDB_REPLACE 1 /* Unused */
#define TDB_INSERT 2 /* Don't overwrite an existing entry */
#define TDB_MODIFY 3 /* Don't create an existing entry */
/* flags for tdb_open() */
#define TDB_DEFAULT 0 /* just a readability place holder */
#define TDB_CLEAR_IF_FIRST 1
#define TDB_INTERNAL 2 /* don't store on disk */
#define TDB_NOLOCK 4 /* don't do any locking */
#define TDB_NOMMAP 8 /* don't use mmap */
#define TDB_CONVERT 16 /* convert endian (internal use) */
#define TDB_BIGENDIAN 32 /* header is big-endian (internal use) */
#define TDB_NOSYNC 64 /* don't use synchronous transactions */
#define TDB_SEQNUM 128 /* maintain a sequence number */
#define TDB_VOLATILE 256 /* Activate the per-hashchain freelist, default 5 */
#define TDB_ALLOW_NESTING 512 /* Allow transactions to nest */
#define TDB_DISALLOW_NESTING 1024 /* Disallow transactions to nest */
#define TDB_INCOMPATIBLE_HASH 2048 /* Better hashing: can't be opened by older tdb versions. */
/* error codes */
enum TDB_ERROR {TDB_SUCCESS=0, TDB_ERR_CORRUPT, TDB_ERR_IO, TDB_ERR_LOCK,
TDB_ERR_OOM, TDB_ERR_EXISTS, TDB_ERR_NOLOCK, TDB_ERR_LOCK_TIMEOUT,
TDB_ERR_NOEXIST, TDB_ERR_EINVAL, TDB_ERR_RDONLY,
TDB_ERR_NESTING};
/* debugging uses one of the following levels */
enum tdb_debug_level {TDB_DEBUG_FATAL = 0, TDB_DEBUG_ERROR,
TDB_DEBUG_WARNING, TDB_DEBUG_TRACE};
typedef struct TDB_DATA {
unsigned char *dptr;
size_t dsize;
} TDB_DATA;
/* this is the context structure that is returned from a db open */
typedef struct tdb_context TDB_CONTEXT;
typedef int (*tdb_traverse_func)(struct tdb_context *, TDB_DATA, TDB_DATA, void *);
typedef void (*tdb_log_func)(struct tdb_context *, enum tdb_debug_level, const char *, ...) PRINTF_FMT(3, 4);
typedef unsigned int (*tdb_hash_func)(TDB_DATA *key);
struct tdb_logging_context {
tdb_log_func log_fn;
void *log_private;
};
struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
int open_flags, mode_t mode);
struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
int open_flags, mode_t mode,
const struct tdb_logging_context *log_ctx,
tdb_hash_func hash_fn);
void tdb_set_max_dead(struct tdb_context *tdb, int max_dead);
int tdb_reopen(struct tdb_context *tdb);
int tdb_reopen_all(int parent_longlived);
void tdb_set_logging_function(struct tdb_context *tdb, const struct tdb_logging_context *log_ctx);
enum TDB_ERROR tdb_error(struct tdb_context *tdb);
const char *tdb_errorstr(struct tdb_context *tdb);
TDB_DATA tdb_fetch(struct tdb_context *tdb, TDB_DATA key);
int tdb_parse_record(struct tdb_context *tdb, TDB_DATA key,
int (*parser)(TDB_DATA key, TDB_DATA data,
void *private_data),
void *private_data);
int tdb_delete(struct tdb_context *tdb, TDB_DATA key);
int tdb_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag);
int tdb_append(struct tdb_context *tdb, TDB_DATA key, TDB_DATA new_dbuf);
int tdb_close(struct tdb_context *tdb);
TDB_DATA tdb_firstkey(struct tdb_context *tdb);
TDB_DATA tdb_nextkey(struct tdb_context *tdb, TDB_DATA key);
int tdb_traverse(struct tdb_context *tdb, tdb_traverse_func fn, void *);
int tdb_traverse_read(struct tdb_context *tdb, tdb_traverse_func fn, void *);
int tdb_exists(struct tdb_context *tdb, TDB_DATA key);
int tdb_lockall(struct tdb_context *tdb);
int tdb_lockall_nonblock(struct tdb_context *tdb);
int tdb_unlockall(struct tdb_context *tdb);
int tdb_lockall_read(struct tdb_context *tdb);
int tdb_lockall_read_nonblock(struct tdb_context *tdb);
int tdb_unlockall_read(struct tdb_context *tdb);
int tdb_lockall_mark(struct tdb_context *tdb);
int tdb_lockall_unmark(struct tdb_context *tdb);
const char *tdb_name(struct tdb_context *tdb);
int tdb_fd(struct tdb_context *tdb);
tdb_log_func tdb_log_fn(struct tdb_context *tdb);
void *tdb_get_logging_private(struct tdb_context *tdb);
int tdb_transaction_start(struct tdb_context *tdb);
int tdb_transaction_prepare_commit(struct tdb_context *tdb);
int tdb_transaction_commit(struct tdb_context *tdb);
int tdb_transaction_cancel(struct tdb_context *tdb);
int tdb_transaction_recover(struct tdb_context *tdb);
int tdb_get_seqnum(struct tdb_context *tdb);
int tdb_hash_size(struct tdb_context *tdb);
size_t tdb_map_size(struct tdb_context *tdb);
int tdb_get_flags(struct tdb_context *tdb);
void tdb_add_flags(struct tdb_context *tdb, unsigned flag);
void tdb_remove_flags(struct tdb_context *tdb, unsigned flag);
void tdb_enable_seqnum(struct tdb_context *tdb);
void tdb_increment_seqnum_nonblock(struct tdb_context *tdb);
unsigned int tdb_jenkins_hash(TDB_DATA *key);
int tdb_check(struct tdb_context *tdb,
int (*check)(TDB_DATA key, TDB_DATA data, void *private_data),
void *private_data);
/* Low level locking functions: use with care */
int tdb_chainlock(struct tdb_context *tdb, TDB_DATA key);
int tdb_chainlock_nonblock(struct tdb_context *tdb, TDB_DATA key);
int tdb_chainunlock(struct tdb_context *tdb, TDB_DATA key);
int tdb_chainlock_read(struct tdb_context *tdb, TDB_DATA key);
int tdb_chainunlock_read(struct tdb_context *tdb, TDB_DATA key);
int tdb_chainlock_mark(struct tdb_context *tdb, TDB_DATA key);
int tdb_chainlock_unmark(struct tdb_context *tdb, TDB_DATA key);
void tdb_setalarm_sigptr(struct tdb_context *tdb, volatile sig_atomic_t *sigptr);
/* wipe and repack */
int tdb_wipe_all(struct tdb_context *tdb);
int tdb_repack(struct tdb_context *tdb);
/* Debug functions. Not used in production. */
void tdb_dump_all(struct tdb_context *tdb);
int tdb_printfreelist(struct tdb_context *tdb);
int tdb_validate_freelist(struct tdb_context *tdb, int *pnum_entries);
int tdb_freelist_size(struct tdb_context *tdb);
char *tdb_summary(struct tdb_context *tdb);
extern TDB_DATA tdb_null;
#ifdef __cplusplus
}
#endif
#endif /* tdb.h */
ccan/tdb/tdb_private.h deleted 100644 → 0
View file @ 98b8ada2
#ifndef TDB_PRIVATE_H
#define TDB_PRIVATE_H
/*
Unix SMB/CIFS implementation.
trivial database library - private includes
Copyright (C) Andrew Tridgell 2005
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#ifdef _SAMBA_BUILD_
#include "replace.h"
#include "system/filesys.h"
#include "system/time.h"
#include "system/shmem.h"
#include "system/select.h"
#include "system/wait.h"
#else
#include "config.h"
#define _FILE_OFFSET_BITS 64
#include <stdint.h>
#include <stdbool.h>
#include <stdlib.h>
#include <sys/time.h>
#include <sys/mman.h>
#include <unistd.h>
#include <fcntl.h>
#include <string.h>
#include <errno.h>
#include <stdio.h>
#include <utime.h>
#endif
#include "tdb.h"
/* #define TDB_TRACE 1 */
#ifndef __STRING
#define __STRING(x) #x
#endif
#ifndef __STRINGSTRING
#define __STRINGSTRING(x) __STRING(x)
#endif
#ifndef __location__
#define __location__ __FILE__ ":" __STRINGSTRING(__LINE__)
#endif
#if !HAVE_GETPAGESIZE
#define getpagesize() 0x2000
#endif
typedef uint32_t tdb_len_t;
typedef uint32_t tdb_off_t;
#ifndef offsetof
#define offsetof(t,f) ((size_t)&((t *)0)->f)
#endif
#define TDB_MAGIC_FOOD "TDB file\n"
#define TDB_VERSION (0x26011967 + 6)
#define TDB_MAGIC (0x26011999U)
#define TDB_FREE_MAGIC (~TDB_MAGIC)
#define TDB_DEAD_MAGIC (0xFEE1DEAD)
#define TDB_RECOVERY_MAGIC (0xf53bc0e7U)
#define TDB_RECOVERY_INVALID_MAGIC (0x0)
#define TDB_HASH_RWLOCK_MAGIC (0xbad1a51U)
#define TDB_ALIGNMENT 4
#define DEFAULT_HASH_SIZE 131
#define FREELIST_TOP (sizeof(struct tdb_header))
#define TDB_ALIGN(x,a) (((x) + (a)-1) & ~((a)-1))
#define TDB_BYTEREV(x) (((((x)&0xff)<<24)|((x)&0xFF00)<<8)|(((x)>>8)&0xFF00)|((x)>>24))
#define TDB_DEAD(r) ((r)->magic == TDB_DEAD_MAGIC)
#define TDB_BAD_MAGIC(r) ((r)->magic != TDB_MAGIC && !TDB_DEAD(r))
#define TDB_HASH_TOP(hash) (FREELIST_TOP + (BUCKET(hash)+1)*sizeof(tdb_off_t))
#define TDB_HASHTABLE_SIZE(tdb) ((tdb->header.hash_size+1)*sizeof(tdb_off_t))
#define TDB_DATA_START(hash_size) (TDB_HASH_TOP(hash_size-1) + sizeof(tdb_off_t))
#define TDB_RECOVERY_HEAD offsetof(struct tdb_header, recovery_start)
#define TDB_SEQNUM_OFS offsetof(struct tdb_header, sequence_number)
#define TDB_PAD_BYTE 0x42
#define TDB_PAD_U32 0x42424242
/* NB assumes there is a local variable called "tdb" that is the
* current context, also takes doubly-parenthesized print-style
* argument. */
#define TDB_LOG(x) tdb->log.log_fn x
#ifdef TDB_TRACE
void tdb_trace(struct tdb_context *tdb, const char *op);
void tdb_trace_seqnum(struct tdb_context *tdb, uint32_t seqnum, const char *op);
void tdb_trace_open(struct tdb_context *tdb, const char *op,
unsigned hash_size, unsigned tdb_flags, unsigned open_flags);
void tdb_trace_ret(struct tdb_context *tdb, const char *op, int ret);
void tdb_trace_retrec(struct tdb_context *tdb, const char *op, TDB_DATA ret);
void tdb_trace_1rec(struct tdb_context *tdb, const char *op,
TDB_DATA rec);
void tdb_trace_1rec_ret(struct tdb_context *tdb, const char *op,
TDB_DATA rec, int ret);
void tdb_trace_1rec_retrec(struct tdb_context *tdb, const char *op,
TDB_DATA rec, TDB_DATA ret);
void tdb_trace_2rec_flag_ret(struct tdb_context *tdb, const char *op,
TDB_DATA rec1, TDB_DATA rec2, unsigned flag,
int ret);
void tdb_trace_2rec_retrec(struct tdb_context *tdb, const char *op,
TDB_DATA rec1, TDB_DATA rec2, TDB_DATA ret);
#else
#define tdb_trace(tdb, op)
#define tdb_trace_seqnum(tdb, seqnum, op)
#define tdb_trace_open(tdb, op, hash_size, tdb_flags, open_flags)
#define tdb_trace_ret(tdb, op, ret)
#define tdb_trace_retrec(tdb, op, ret)
#define tdb_trace_1rec(tdb, op, rec)
#define tdb_trace_1rec_ret(tdb, op, rec, ret)
#define tdb_trace_1rec_retrec(tdb, op, rec, ret)
#define tdb_trace_2rec_flag_ret(tdb, op, rec1, rec2, flag, ret)
#define tdb_trace_2rec_retrec(tdb, op, rec1, rec2, ret)
#endif /* !TDB_TRACE */
/* lock offsets */
#define OPEN_LOCK 0
#define ACTIVE_LOCK 4
#define TRANSACTION_LOCK 8
/* free memory if the pointer is valid and zero the pointer */
#ifndef SAFE_FREE
#define SAFE_FREE(x) do { if ((x) != NULL) {free(x); (x)=NULL;} } while(0)
#endif
#define BUCKET(hash) ((hash) % tdb->header.hash_size)
#define DOCONV() (tdb->flags & TDB_CONVERT)
#define CONVERT(x) (DOCONV() ? tdb_convert(&x, sizeof(x)) : &x)
/* the body of the database is made of one tdb_record for the free space
plus a separate data list for each hash value */
struct tdb_record {
tdb_off_t next; /* offset of the next record in the list */
tdb_len_t rec_len; /* total byte length of record */
tdb_len_t key_len; /* byte length of key */
tdb_len_t data_len; /* byte length of data */
uint32_t full_hash; /* the full 32 bit hash of the key */
uint32_t magic; /* try to catch errors */
/* the following union is implied:
union {
char record[rec_len];
struct {
char key[key_len];
char data[data_len];
}
uint32_t totalsize; (tailer)
}
*/
};
/* this is stored at the front of every database */
struct tdb_header {
char magic_food[32]; /* for /etc/magic */
uint32_t version; /* version of the code */
uint32_t hash_size; /* number of hash entries */
tdb_off_t rwlocks; /* obsolete - kept to detect old formats */
tdb_off_t recovery_start; /* offset of transaction recovery region */
tdb_off_t sequence_number; /* used when TDB_SEQNUM is set */
uint32_t magic1_hash; /* hash of TDB_MAGIC_FOOD. */
uint32_t magic2_hash; /* hash of TDB_MAGIC. */
tdb_off_t reserved[27];
};
struct tdb_lock_type {
uint32_t off;
uint32_t count;
uint32_t ltype;
};
struct tdb_traverse_lock {
struct tdb_traverse_lock *next;
uint32_t off;
uint32_t hash;
int lock_rw;
};
enum tdb_lock_flags {
/* WAIT == F_SETLKW, NOWAIT == F_SETLK */
TDB_LOCK_NOWAIT = 0,
TDB_LOCK_WAIT = 1,
/* If set, don't log an error on failure. */
TDB_LOCK_PROBE = 2,
/* If set, don't actually lock at all. */
TDB_LOCK_MARK_ONLY = 4,
};
struct tdb_methods {
int (*tdb_read)(struct tdb_context *, tdb_off_t , void *, tdb_len_t , int );
int (*tdb_write)(struct tdb_context *, tdb_off_t, const void *, tdb_len_t);
void (*next_hash_chain)(struct tdb_context *, uint32_t *);
int (*tdb_oob)(struct tdb_context *, tdb_off_t , int );
int (*tdb_expand_file)(struct tdb_context *, tdb_off_t , tdb_off_t );
};
struct tdb_context {
char *name; /* the name of the database */
void *map_ptr; /* where it is currently mapped */
int fd; /* open file descriptor for the database */
tdb_len_t map_size; /* how much space has been mapped */
int read_only; /* opened read-only */
int traverse_read; /* read-only traversal */
int traverse_write; /* read-write traversal */
struct tdb_lock_type allrecord_lock; /* .offset == upgradable */
int num_lockrecs;
struct tdb_lock_type *lockrecs; /* only real locks, all with count>0 */
enum TDB_ERROR ecode; /* error code for last tdb error */
struct tdb_header header; /* a cached copy of the header */
uint32_t flags; /* the flags passed to tdb_open */
struct tdb_traverse_lock travlocks; /* current traversal locks */
struct tdb_context *next; /* all tdbs to avoid multiple opens */
dev_t device; /* uniquely identifies this tdb */
ino_t inode; /* uniquely identifies this tdb */
struct tdb_logging_context log;
unsigned int (*hash_fn)(TDB_DATA *key);
int open_flags; /* flags used in the open - needed by reopen */
const struct tdb_methods *methods;
struct tdb_transaction *transaction;
int page_size;
int max_dead_records;
#ifdef TDB_TRACE
int tracefd;
#endif
volatile sig_atomic_t *interrupt_sig_ptr;
};
/*
internal prototypes
*/
int tdb_munmap(struct tdb_context *tdb);
void tdb_mmap(struct tdb_context *tdb);
int tdb_lock(struct tdb_context *tdb, int list, int ltype);
int tdb_lock_nonblock(struct tdb_context *tdb, int list, int ltype);
int tdb_nest_lock(struct tdb_context *tdb, uint32_t offset, int ltype,
enum tdb_lock_flags flags);
int tdb_nest_unlock(struct tdb_context *tdb, uint32_t offset, int ltype,
bool mark_lock);
bool tdb_have_locks(struct tdb_context *tdb);
int tdb_unlock(struct tdb_context *tdb, int list, int ltype);
int tdb_brlock(struct tdb_context *tdb,
int rw_type, tdb_off_t offset, size_t len,
enum tdb_lock_flags flags);
int tdb_brunlock(struct tdb_context *tdb,
int rw_type, tdb_off_t offset, size_t len);
bool tdb_have_extra_locks(struct tdb_context *tdb);
void tdb_release_extra_locks(struct tdb_context *tdb);
int tdb_transaction_lock(struct tdb_context *tdb, int ltype);
int tdb_transaction_unlock(struct tdb_context *tdb, int ltype);
int tdb_allrecord_lock(struct tdb_context *tdb, int ltype,
enum tdb_lock_flags flags, bool upgradable);
int tdb_allrecord_unlock(struct tdb_context *tdb, int ltype, bool mark_lock);
int tdb_allrecord_upgrade(struct tdb_context *tdb);
int tdb_write_lock_record(struct tdb_context *tdb, tdb_off_t off);
int tdb_write_unlock_record(struct tdb_context *tdb, tdb_off_t off);
int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
void *tdb_convert(void *buf, uint32_t size);
int tdb_free(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
tdb_off_t tdb_allocate(struct tdb_context *tdb, tdb_len_t length, struct tdb_record *rec);
int tdb_ofs_read(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
int tdb_ofs_write(struct tdb_context *tdb, tdb_off_t offset, tdb_off_t *d);
int tdb_lock_record(struct tdb_context *tdb, tdb_off_t off);
int tdb_unlock_record(struct tdb_context *tdb, tdb_off_t off);
bool tdb_needs_recovery(struct tdb_context *tdb);
int tdb_rec_read(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
int tdb_rec_write(struct tdb_context *tdb, tdb_off_t offset, struct tdb_record *rec);
int tdb_do_delete(struct tdb_context *tdb, tdb_off_t rec_ptr, struct tdb_record *rec);
unsigned char *tdb_alloc_read(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t len);
int tdb_parse_data(struct tdb_context *tdb, TDB_DATA key,
tdb_off_t offset, tdb_len_t len,
int (*parser)(TDB_DATA key, TDB_DATA data,
void *private_data),
void *private_data);
tdb_off_t tdb_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, int locktype,
struct tdb_record *rec);
void tdb_io_init(struct tdb_context *tdb);
int tdb_expand(struct tdb_context *tdb, tdb_off_t size);
tdb_off_t tdb_expand_adjust(tdb_off_t map_size, tdb_off_t size, int page_size);
int tdb_rec_free_read(struct tdb_context *tdb, tdb_off_t off,
struct tdb_record *rec);
void tdb_header_hash(struct tdb_context *tdb,
uint32_t *magic1_hash, uint32_t *magic2_hash);
unsigned int tdb_old_hash(TDB_DATA *key);
size_t tdb_dead_space(struct tdb_context *tdb, tdb_off_t off);
#endif
ccan/tdb/test/external-agent.c deleted 100644 → 0
View file @ 98b8ada2
#include "external-agent.h"
#include "lock-tracking.h"
#include "logging.h"
#include <sys/types.h>
#include <sys/wait.h>
#include <unistd.h>
#include <err.h>
#include <fcntl.h>
#include <stdlib.h>
#include <limits.h>
#include <string.h>
#include <errno.h>
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/tdb_private.h>
#include <ccan/tap/tap.h>
#include <stdio.h>
#include <stdarg.h>
static struct tdb_context *tdb;
static enum agent_return do_operation(enum operation op, const char *name)
{
TDB_DATA k;
enum agent_return ret;
TDB_DATA data;
if (op != OPEN && op != OPEN_WITH_CLEAR_IF_FIRST && !tdb) {
diag("external: No tdb open!");
return OTHER_FAILURE;
}
k.dptr = (void *)name;
k.dsize = strlen(name);
locking_would_block = 0;
switch (op) {
case OPEN:
if (tdb) {
diag("Already have tdb %s open", tdb_name(tdb));
return OTHER_FAILURE;
}
tdb = tdb_open_ex(name, 0, TDB_DEFAULT, O_RDWR, 0,
&taplogctx, NULL);
if (!tdb) {
if (!locking_would_block)
diag("Opening tdb gave %s", strerror(errno));
ret = OTHER_FAILURE;
} else
ret = SUCCESS;
break;
case OPEN_WITH_CLEAR_IF_FIRST:
if (tdb)
return OTHER_FAILURE;
tdb = tdb_open_ex(name, 0, TDB_CLEAR_IF_FIRST, O_RDWR, 0,
&taplogctx, NULL);
ret = tdb ? SUCCESS : OTHER_FAILURE;
break;
case TRANSACTION_START:
ret = tdb_transaction_start(tdb) == 0 ? SUCCESS : OTHER_FAILURE;
break;
case FETCH:
data = tdb_fetch(tdb, k);
if (data.dptr == NULL) {
if (tdb_error(tdb) == TDB_ERR_NOEXIST)
ret = FAILED;
else
ret = OTHER_FAILURE;
} else if (data.dsize != k.dsize
|| memcmp(data.dptr, k.dptr, k.dsize) != 0) {
ret = OTHER_FAILURE;
} else {
ret = SUCCESS;
}
free(data.dptr);
break;
case STORE:
ret = tdb_store(tdb, k, k, 0) == 0 ? SUCCESS : OTHER_FAILURE;
break;
case TRANSACTION_COMMIT:
ret = tdb_transaction_commit(tdb)==0 ? SUCCESS : OTHER_FAILURE;
break;
case CHECK:
ret = tdb_check(tdb, NULL, NULL) == 0 ? SUCCESS : OTHER_FAILURE;
break;
case NEEDS_RECOVERY:
ret = tdb_needs_recovery(tdb) ? SUCCESS : FAILED;
break;
case CLOSE:
ret = tdb_close(tdb) == 0 ? SUCCESS : OTHER_FAILURE;
tdb = NULL;
break;
default:
ret = OTHER_FAILURE;
}
if (locking_would_block)
ret = WOULD_HAVE_BLOCKED;
return ret;
}
struct agent {
int cmdfd, responsefd;
};
/* Do this before doing any tdb stuff. Return handle, or NULL. */
struct agent *prepare_external_agent(void)
{
int pid, ret;
int command[2], response[2];
char name[1+PATH_MAX];
if (pipe(command) != 0 || pipe(response) != 0)
return NULL;
pid = fork();
if (pid < 0)
return NULL;
if (pid != 0) {
struct agent *agent = malloc(sizeof(*agent));
close(command[0]);
close(response[1]);
agent->cmdfd = command[1];
agent->responsefd = response[0];
return agent;
}
close(command[1]);
close(response[0]);
/* We want to fail, not block. */
nonblocking_locks = true;
log_prefix = "external: ";
while ((ret = read(command[0], name, sizeof(name))) > 0) {
enum agent_return result;
result = do_operation(name[0], name+1);
if (write(response[1], &result, sizeof(result))
!= sizeof(result))
err(1, "Writing response");
}
exit(0);
}
/* Ask the external agent to try to do an operation. */
enum agent_return external_agent_operation(struct agent *agent,
enum operation op,
const char *name)
{
enum agent_return res;
unsigned int len;
char *string;
if (!name)
name = "";
len = 1 + strlen(name) + 1;
string = malloc(len);
string[0] = op;
strcpy(string+1, name);
if (write(agent->cmdfd, string, len) != len
|| read(agent->responsefd, &res, sizeof(res)) != sizeof(res))
res = AGENT_DIED;
free(string);
return res;
}
const char *agent_return_name(enum agent_return ret)
{
return ret == SUCCESS ? "SUCCESS"
: ret == WOULD_HAVE_BLOCKED ? "WOULD_HAVE_BLOCKED"
: ret == AGENT_DIED ? "AGENT_DIED"
: ret == FAILED ? "FAILED"
: ret == OTHER_FAILURE ? "OTHER_FAILURE"
: "**INVALID**";
}
const char *operation_name(enum operation op)
{
switch (op) {
case OPEN: return "OPEN";
case OPEN_WITH_CLEAR_IF_FIRST: return "OPEN_WITH_CLEAR_IF_FIRST";
case TRANSACTION_START: return "TRANSACTION_START";
case FETCH: return "FETCH";
case STORE: return "STORE";
case TRANSACTION_COMMIT: return "TRANSACTION_COMMIT";
case CHECK: return "CHECK";
case NEEDS_RECOVERY: return "NEEDS_RECOVERY";
case CLOSE: return "CLOSE";
}
return "**INVALID**";
}
ccan/tdb/test/external-agent.h deleted 100644 → 0
View file @ 98b8ada2
#ifndef TDB_TEST_EXTERNAL_AGENT_H
#define TDB_TEST_EXTERNAL_AGENT_H
/* For locking tests, we need a different process to try things at
* various times. */
enum operation {
OPEN,
OPEN_WITH_CLEAR_IF_FIRST,
TRANSACTION_START,
FETCH,
STORE,
TRANSACTION_COMMIT,
CHECK,
NEEDS_RECOVERY,
CLOSE,
};
/* Do this before doing any tdb stuff. Return handle, or -1. */
struct agent *prepare_external_agent(void);
enum agent_return {
SUCCESS,
WOULD_HAVE_BLOCKED,
AGENT_DIED,
FAILED, /* For fetch, or NEEDS_RECOVERY */
OTHER_FAILURE,
};
/* Ask the external agent to try to do an operation.
* name == tdb name for OPEN/OPEN_WITH_CLEAR_IF_FIRST,
* record name for FETCH/STORE (store stores name as data too)
*/
enum agent_return external_agent_operation(struct agent *handle,
enum operation op,
const char *name);
/* Mapping enum -> string. */
const char *agent_return_name(enum agent_return ret);
const char *operation_name(enum operation op);
#endif /* TDB_TEST_EXTERNAL_AGENT_H */
ccan/tdb/test/lock-tracking.c deleted 100644 → 0
View file @ 98b8ada2
/* We save the locks so we can reaquire them. */
#include <unistd.h>
#include <fcntl.h>
#include <stdarg.h>
#include <stdlib.h>
#include <ccan/tap/tap.h>
#include <ccan/tdb/tdb_private.h>
#include "lock-tracking.h"
struct lock {
struct lock *next;
unsigned int off;
unsigned int len;
int type;
};
static struct lock *locks;
int locking_errors = 0;
bool suppress_lockcheck = false;
bool nonblocking_locks;
int locking_would_block = 0;
void (*unlock_callback)(int fd);
int fcntl_with_lockcheck(int fd, int cmd, ... /* arg */ )
{
va_list ap;
int ret, arg3;
struct flock *fl;
bool may_block = false;
if (cmd != F_SETLK && cmd != F_SETLKW) {
/* This may be totally bogus, but we don't know in general. */
va_start(ap, cmd);
arg3 = va_arg(ap, int);
va_end(ap);
return fcntl(fd, cmd, arg3);
}
va_start(ap, cmd);
fl = va_arg(ap, struct flock *);
va_end(ap);
if (cmd == F_SETLKW && nonblocking_locks) {
cmd = F_SETLK;
may_block = true;
}
ret = fcntl(fd, cmd, fl);
/* Detect when we failed, but might have been OK if we waited. */
if (may_block && ret == -1 && (errno == EAGAIN || errno == EACCES)) {
locking_would_block++;
}
if (fl->l_type == F_UNLCK) {
struct lock **l;
struct lock *old = NULL;
for (l = &locks; *l; l = &(*l)->next) {
if ((*l)->off == fl->l_start
&& (*l)->len == fl->l_len) {
if (ret == 0) {
old = *l;
*l = (*l)->next;
free(old);
}
break;
}
}
if (!old && !suppress_lockcheck) {
diag("Unknown unlock %u@%u - %i",
(int)fl->l_len, (int)fl->l_start, ret);
locking_errors++;
}
} else {
struct lock *new, *i;
unsigned int fl_end = fl->l_start + fl->l_len;
if (fl->l_len == 0)
fl_end = (unsigned int)-1;
/* Check for overlaps: we shouldn't do this. */
for (i = locks; i; i = i->next) {
unsigned int i_end = i->off + i->len;
if (i->len == 0)
i_end = (unsigned int)-1;
if (fl->l_start >= i->off && fl->l_start < i_end)
break;
if (fl_end >= i->off && fl_end < i_end)
break;
/* tdb_allrecord_lock does this, handle adjacent: */
if (fl->l_start == i_end && fl->l_type == i->type) {
if (ret == 0) {
i->len = fl->l_len
? i->len + fl->l_len
: 0;
}
goto done;
}
}
if (i) {
/* Special case: upgrade of allrecord lock. */
if (i->type == F_RDLCK && fl->l_type == F_WRLCK
&& i->off == FREELIST_TOP
&& fl->l_start == FREELIST_TOP
&& i->len == 0
&& fl->l_len == 0) {
if (ret == 0)
i->type = F_WRLCK;
goto done;
}
if (!suppress_lockcheck) {
diag("%s lock %u@%u overlaps %u@%u",
fl->l_type == F_WRLCK ? "write" : "read",
(int)fl->l_len, (int)fl->l_start,
i->len, (int)i->off);
locking_errors++;
}
}
if (ret == 0) {
new = malloc(sizeof *new);
new->off = fl->l_start;
new->len = fl->l_len;
new->type = fl->l_type;
new->next = locks;
locks = new;
}
}
done:
if (ret == 0 && fl->l_type == F_UNLCK && unlock_callback)
unlock_callback(fd);
return ret;
}
unsigned int forget_locking(void)
{
unsigned int num = 0;
while (locks) {
struct lock *next = locks->next;
free(locks);
locks = next;
num++;
}
return num;
}
ccan/tdb/test/lock-tracking.h deleted 100644 → 0
View file @ 98b8ada2
#ifndef LOCK_TRACKING_H
#define LOCK_TRACKING_H
#include <stdbool.h>
/* Set this if you want a callback after fnctl unlock. */
extern void (*unlock_callback)(int fd);
/* Replacement fcntl. */
int fcntl_with_lockcheck(int fd, int cmd, ... /* arg */ );
/* Discard locking info: returns number of locks outstanding. */
unsigned int forget_locking(void);
/* Number of errors in locking. */
extern int locking_errors;
/* Suppress lock checking. */
extern bool suppress_lockcheck;
/* Make all locks non-blocking. */
extern bool nonblocking_locks;
/* Number of times we failed a lock because we made it non-blocking. */
extern int locking_would_block;
#endif /* LOCK_TRACKING_H */
ccan/tdb/test/logging.c deleted 100644 → 0
View file @ 98b8ada2
#include "logging.h"
#include <ccan/tap/tap.h>
#include <stdarg.h>
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
bool suppress_logging = false;
const char *log_prefix = "";
/* Turn log messages into tap diag messages. */
static void taplog(struct tdb_context *tdb,
enum tdb_debug_level level,
const char *fmt, ...)
{
va_list ap;
char line[200];
if (suppress_logging)
return;
va_start(ap, fmt);
vsprintf(line, fmt, ap);
va_end(ap);
/* Strip trailing \n: diag adds it. */
if (line[0] && line[strlen(line)-1] == '\n')
diag("%s%.*s", log_prefix, (unsigned)strlen(line)-1, line);
else
diag("%s%s", log_prefix, line);
}
struct tdb_logging_context taplogctx = { taplog, NULL };
ccan/tdb/test/logging.h deleted 100644 → 0
View file @ 98b8ada2
#ifndef TDB_TEST_LOGGING_H
#define TDB_TEST_LOGGING_H
#include <ccan/tdb/tdb.h>
#include <stdbool.h>
extern bool suppress_logging;
extern const char *log_prefix;
extern struct tdb_logging_context taplogctx;
#endif /* TDB_TEST_LOGGING_H */
ccan/tdb/test/run-3G-file.c deleted 100644 → 0
View file @ 98b8ada2
/* We need this otherwise fcntl locking fails. */
#define _FILE_OFFSET_BITS 64
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
#include "logging.h"
static int tdb_expand_file_sparse(struct tdb_context *tdb,
tdb_off_t size,
tdb_off_t addition)
{
if (tdb->read_only || tdb->traverse_read) {
tdb->ecode = TDB_ERR_RDONLY;
return -1;
}
if (ftruncate(tdb->fd, size+addition) == -1) {
char b = 0;
ssize_t written = pwrite(tdb->fd, &b, 1, (size+addition) - 1);
if (written == 0) {
/* try once more, potentially revealing errno */
written = pwrite(tdb->fd, &b, 1, (size+addition) - 1);
}
if (written == 0) {
/* again - give up, guessing errno */
errno = ENOSPC;
}
if (written != 1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "expand_file to %d failed (%s)\n",
size+addition, strerror(errno)));
return -1;
}
}
return 0;
}
static const struct tdb_methods large_io_methods = {
tdb_read,
tdb_write,
tdb_next_hash_chain,
tdb_oob,
tdb_expand_file_sparse
};
static int test_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data,
void *_data)
{
TDB_DATA *expect = _data;
ok1(key.dsize == strlen("hi"));
ok1(memcmp(key.dptr, "hi", strlen("hi")) == 0);
ok1(data.dsize == expect->dsize);
ok1(memcmp(data.dptr, expect->dptr, data.dsize) == 0);
return 0;
}
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
TDB_DATA key, orig_data, data;
uint32_t hash;
tdb_off_t rec_ptr;
struct tdb_record rec;
plan_tests(24);
tdb = tdb_open_ex("run-36-file.tdb", 1024, TDB_CLEAR_IF_FIRST,
O_CREAT|O_TRUNC|O_RDWR, 0600, &taplogctx, NULL);
ok1(tdb);
tdb->methods = &large_io_methods;
/* Enlarge the file (internally multiplies by 100). */
ok1(tdb_expand(tdb, 30000000) == 0);
/* Put an entry in, and check it. */
key.dsize = strlen("hi");
key.dptr = (void *)"hi";
orig_data.dsize = strlen("world");
orig_data.dptr = (void *)"world";
ok1(tdb_store(tdb, key, orig_data, TDB_INSERT) == 0);
data = tdb_fetch(tdb, key);
ok1(data.dsize == strlen("world"));
ok1(memcmp(data.dptr, "world", strlen("world")) == 0);
free(data.dptr);
/* That currently fills at the end, make sure that's true. */
hash = tdb->hash_fn(&key);
rec_ptr = tdb_find_lock_hash(tdb, key, hash, F_RDLCK, &rec);
ok1(rec_ptr);
ok1(rec_ptr > 2U*1024*1024*1024);
tdb_unlock(tdb, BUCKET(rec.full_hash), F_RDLCK);
/* Traverse must work. */
ok1(tdb_traverse(tdb, test_traverse, &orig_data) == 1);
/* Delete should work. */
ok1(tdb_delete(tdb, key) == 0);
ok1(tdb_traverse(tdb, test_traverse, NULL) == 0);
/* Transactions should work. */
ok1(tdb_transaction_start(tdb) == 0);
ok1(tdb_store(tdb, key, orig_data, TDB_INSERT) == 0);
data = tdb_fetch(tdb, key);
ok1(data.dsize == strlen("world"));
ok1(memcmp(data.dptr, "world", strlen("world")) == 0);
free(data.dptr);
ok1(tdb_transaction_commit(tdb) == 0);
ok1(tdb_traverse(tdb, test_traverse, &orig_data) == 1);
tdb_close(tdb);
return exit_status();
}
ccan/tdb/test/run-bad-tdb-header.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
#include "logging.h"
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
struct tdb_header hdr;
int fd;
plan_tests(11);
/* Can open fine if complete crap, as long as O_CREAT. */
fd = open("run-bad-tdb-header.tdb", O_RDWR|O_CREAT|O_TRUNC, 0600);
ok1(fd >= 0);
ok1(write(fd, "hello world", 11) == 11);
close(fd);
tdb = tdb_open_ex("run-bad-tdb-header.tdb", 1024, 0, O_RDWR, 0,
&taplogctx, NULL);
ok1(!tdb);
tdb = tdb_open_ex("run-bad-tdb-header.tdb", 1024, 0, O_CREAT|O_RDWR,
0600, &taplogctx, NULL);
ok1(tdb);
tdb_close(tdb);
/* Now, with wrong version it should *not* overwrite. */
fd = open("run-bad-tdb-header.tdb", O_RDWR);
ok1(fd >= 0);
ok1(read(fd, &hdr, sizeof(hdr)) == sizeof(hdr));
ok1(hdr.version == TDB_VERSION);
hdr.version++;
lseek(fd, 0, SEEK_SET);
ok1(write(fd, &hdr, sizeof(hdr)) == sizeof(hdr));
close(fd);
tdb = tdb_open_ex("run-bad-tdb-header.tdb", 1024, 0, O_RDWR|O_CREAT,
0600, &taplogctx, NULL);
ok1(errno == EIO);
ok1(!tdb);
/* With truncate, will be fine. */
tdb = tdb_open_ex("run-bad-tdb-header.tdb", 1024, 0,
O_RDWR|O_CREAT|O_TRUNC, 0600, &taplogctx, NULL);
ok1(tdb);
tdb_close(tdb);
return exit_status();
}
ccan/tdb/test/run-check.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
#include "logging.h"
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
TDB_DATA key, data;
plan_tests(13);
tdb = tdb_open_ex("run-check.tdb", 1, TDB_CLEAR_IF_FIRST,
O_CREAT|O_TRUNC|O_RDWR, 0600, &taplogctx, NULL);
ok1(tdb);
ok1(tdb_check(tdb, NULL, NULL) == 0);
key.dsize = strlen("hi");
key.dptr = (void *)"hi";
data.dsize = strlen("world");
data.dptr = (void *)"world";
ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
tdb = tdb_open_ex("run-check.tdb", 1024, 0, O_RDWR, 0,
&taplogctx, NULL);
ok1(tdb);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
tdb = tdb_open_ex("test/tdb.corrupt", 1024, 0, O_RDWR, 0,
&taplogctx, NULL);
ok1(tdb);
ok1(tdb_check(tdb, NULL, NULL) == -1);
ok1(tdb_error(tdb) == TDB_ERR_CORRUPT);
tdb_close(tdb);
/* Big and little endian should work! */
tdb = tdb_open_ex("test/old-nohash-le.tdb", 1024, 0, O_RDWR, 0,
&taplogctx, NULL);
ok1(tdb);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
tdb = tdb_open_ex("test/old-nohash-be.tdb", 1024, 0, O_RDWR, 0,
&taplogctx, NULL);
ok1(tdb);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
return exit_status();
}
ccan/tdb/test/run-corrupt.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
#include "logging.h"
static int check(TDB_DATA key, TDB_DATA data, void *private)
{
unsigned int *sizes = private;
if (key.dsize > strlen("hello"))
return -1;
if (memcmp(key.dptr, "hello", key.dsize) != 0)
return -1;
if (data.dsize != strlen("world"))
return -1;
if (memcmp(data.dptr, "world", data.dsize) != 0)
return -1;
sizes[0] += key.dsize;
sizes[1] += data.dsize;
return 0;
}
static void tdb_flip_bit(struct tdb_context *tdb, unsigned int bit)
{
unsigned int off = bit / CHAR_BIT;
unsigned char mask = (1 << (bit % CHAR_BIT));
if (tdb->map_ptr)
((unsigned char *)tdb->map_ptr)[off] ^= mask;
else {
unsigned char c;
if (pread(tdb->fd, &c, 1, off) != 1)
err(1, "pread");
c ^= mask;
if (pwrite(tdb->fd, &c, 1, off) != 1)
err(1, "pwrite");
}
}
static void check_test(struct tdb_context *tdb)
{
TDB_DATA key, data;
unsigned int i, verifiable, corrupt, sizes[2], dsize, ksize;
ok1(tdb_check(tdb, NULL, NULL) == 0);
key.dptr = (void *)"hello";
data.dsize = strlen("world");
data.dptr = (void *)"world";
/* Key and data size respectively. */
dsize = ksize = 0;
/* 5 keys in hash size 2 means we'll have multichains. */
for (key.dsize = 1; key.dsize <= 5; key.dsize++) {
ksize += key.dsize;
dsize += data.dsize;
if (tdb_store(tdb, key, data, TDB_INSERT) != 0)
abort();
}
/* This is how many bytes we expect to be verifiable. */
/* From the file header. */
verifiable = strlen(TDB_MAGIC_FOOD) + 1
+ 2 * sizeof(uint32_t) + 2 * sizeof(tdb_off_t)
+ 2 * sizeof(uint32_t);
/* From the free list chain and hash chains. */
verifiable += 3 * sizeof(tdb_off_t);
/* From the record headers & tailer */
verifiable += 5 * (sizeof(struct tdb_record) + sizeof(uint32_t));
/* The free block: we ignore datalen, keylen, full_hash. */
verifiable += sizeof(struct tdb_record) - 3*sizeof(uint32_t) +
sizeof(uint32_t);
/* Our check function verifies the key and data. */
verifiable += ksize + dsize;
/* Flip one bit at a time, make sure it detects verifiable bytes. */
for (i = 0, corrupt = 0; i < tdb->map_size * CHAR_BIT; i++) {
tdb_flip_bit(tdb, i);
memset(sizes, 0, sizeof(sizes));
if (tdb_check(tdb, check, sizes) != 0)
corrupt++;
else if (sizes[0] != ksize || sizes[1] != dsize)
corrupt++;
tdb_flip_bit(tdb, i);
}
ok(corrupt == verifiable * CHAR_BIT, "corrupt %u should be %u",
corrupt, verifiable * CHAR_BIT);
}
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
plan_tests(4);
/* This should use mmap. */
tdb = tdb_open_ex("run-corrupt.tdb", 2, TDB_CLEAR_IF_FIRST,
O_CREAT|O_TRUNC|O_RDWR, 0600, &taplogctx, NULL);
if (!tdb)
abort();
check_test(tdb);
tdb_close(tdb);
/* This should not. */
tdb = tdb_open_ex("run-corrupt.tdb", 2, TDB_CLEAR_IF_FIRST|TDB_NOMMAP,
O_CREAT|O_TRUNC|O_RDWR, 0600, &taplogctx, NULL);
if (!tdb)
abort();
check_test(tdb);
tdb_close(tdb);
return exit_status();
}
ccan/tdb/test/run-die-during-transaction.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <unistd.h>
#include "lock-tracking.h"
static ssize_t pwrite_check(int fd, const void *buf, size_t count, off_t offset);
static ssize_t write_check(int fd, const void *buf, size_t count);
static int ftruncate_check(int fd, off_t length);
#define pwrite pwrite_check
#define write write_check
#define fcntl fcntl_with_lockcheck
#define ftruncate ftruncate_check
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <stdbool.h>
#include <stdarg.h>
#include <err.h>
#include <setjmp.h>
#include "external-agent.h"
#include "logging.h"
#undef write
#undef pwrite
#undef fcntl
#undef ftruncate
static bool in_transaction;
static int target, current;
static jmp_buf jmpbuf;
#define TEST_DBNAME "run-die-during-transaction.tdb"
#define KEY_STRING "helloworld"
static void maybe_die(int fd)
{
if (in_transaction && current++ == target) {
longjmp(jmpbuf, 1);
}
}
static ssize_t pwrite_check(int fd,
const void *buf, size_t count, off_t offset)
{
ssize_t ret;
maybe_die(fd);
ret = pwrite(fd, buf, count, offset);
if (ret != count)
return ret;
maybe_die(fd);
return ret;
}
static ssize_t write_check(int fd, const void *buf, size_t count)
{
ssize_t ret;
maybe_die(fd);
ret = write(fd, buf, count);
if (ret != count)
return ret;
maybe_die(fd);
return ret;
}
static int ftruncate_check(int fd, off_t length)
{
int ret;
maybe_die(fd);
ret = ftruncate(fd, length);
maybe_die(fd);
return ret;
}
static bool test_death(enum operation op, struct agent *agent)
{
struct tdb_context *tdb = NULL;
TDB_DATA key;
enum agent_return ret;
int needed_recovery = 0;
current = target = 0;
reset:
unlink(TEST_DBNAME);
tdb = tdb_open_ex(TEST_DBNAME, 1024, TDB_NOMMAP,
O_CREAT|O_TRUNC|O_RDWR, 0600, &taplogctx, NULL);
if (setjmp(jmpbuf) != 0) {
/* We're partway through. Simulate our death. */
close(tdb->fd);
forget_locking();
in_transaction = false;
ret = external_agent_operation(agent, NEEDS_RECOVERY, "");
if (ret == SUCCESS)
needed_recovery++;
else if (ret != FAILED) {
diag("Step %u agent NEEDS_RECOVERY = %s", current,
agent_return_name(ret));
return false;
}
ret = external_agent_operation(agent, op, KEY_STRING);
if (ret != SUCCESS) {
diag("Step %u op %s failed = %s", current,
operation_name(op),
agent_return_name(ret));
return false;
}
ret = external_agent_operation(agent, NEEDS_RECOVERY, "");
if (ret != FAILED) {
diag("Still needs recovery after step %u = %s",
current, agent_return_name(ret));
return false;
}
ret = external_agent_operation(agent, CHECK, "");
if (ret != SUCCESS) {
diag("Step %u check failed = %s", current,
agent_return_name(ret));
return false;
}
ret = external_agent_operation(agent, CLOSE, "");
if (ret != SUCCESS) {
diag("Step %u close failed = %s", current,
agent_return_name(ret));
return false;
}
/* Suppress logging as this tries to use closed fd. */
suppress_logging = true;
suppress_lockcheck = true;
tdb_close(tdb);
suppress_logging = false;
suppress_lockcheck = false;
target++;
current = 0;
goto reset;
}
/* Put key for agent to fetch. */
key.dsize = strlen(KEY_STRING);
key.dptr = (void *)KEY_STRING;
if (tdb_store(tdb, key, key, TDB_INSERT) != 0)
return false;
/* This is the key we insert in transaction. */
key.dsize--;
ret = external_agent_operation(agent, OPEN, TEST_DBNAME);
if (ret != SUCCESS)
errx(1, "Agent failed to open: %s", agent_return_name(ret));
ret = external_agent_operation(agent, FETCH, KEY_STRING);
if (ret != SUCCESS)
errx(1, "Agent failed find key: %s", agent_return_name(ret));
in_transaction = true;
if (tdb_transaction_start(tdb) != 0)
return false;
if (tdb_store(tdb, key, key, TDB_INSERT) != 0)
return false;
if (tdb_transaction_commit(tdb) != 0)
return false;
in_transaction = false;
/* We made it! */
diag("Completed %u runs", current);
tdb_close(tdb);
ret = external_agent_operation(agent, CLOSE, "");
if (ret != SUCCESS) {
diag("Step %u close failed = %s", current,
agent_return_name(ret));
return false;
}
ok1(needed_recovery);
ok1(locking_errors == 0);
ok1(forget_locking() == 0);
locking_errors = 0;
return true;
}
int main(int argc, char *argv[])
{
enum operation ops[] = { FETCH, STORE, TRANSACTION_START };
struct agent *agent;
int i;
plan_tests(12);
unlock_callback = maybe_die;
agent = prepare_external_agent();
if (!agent)
err(1, "preparing agent");
for (i = 0; i < sizeof(ops)/sizeof(ops[0]); i++) {
diag("Testing %s after death", operation_name(ops[i]));
ok1(test_death(ops[i], agent));
}
return exit_status();
}
ccan/tdb/test/run-endian.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
#include "logging.h"
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
TDB_DATA key, data;
plan_tests(13);
tdb = tdb_open_ex("run-endian.tdb", 1024,
TDB_CLEAR_IF_FIRST|TDB_CONVERT,
O_CREAT|O_TRUNC|O_RDWR, 0600, &taplogctx, NULL);
ok1(tdb);
key.dsize = strlen("hi");
key.dptr = (void *)"hi";
data.dsize = strlen("world");
data.dptr = (void *)"world";
ok1(tdb_store(tdb, key, data, TDB_MODIFY) < 0);
ok1(tdb_error(tdb) == TDB_ERR_NOEXIST);
ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0);
ok1(tdb_store(tdb, key, data, TDB_INSERT) < 0);
ok1(tdb_error(tdb) == TDB_ERR_EXISTS);
ok1(tdb_store(tdb, key, data, TDB_MODIFY) == 0);
data = tdb_fetch(tdb, key);
ok1(data.dsize == strlen("world"));
ok1(memcmp(data.dptr, "world", strlen("world")) == 0);
free(data.dptr);
key.dsize++;
data = tdb_fetch(tdb, key);
ok1(data.dptr == NULL);
tdb_close(tdb);
/* Reopen: should read it */
tdb = tdb_open_ex("run-endian.tdb", 1024, 0, O_RDWR, 0,
&taplogctx, NULL);
ok1(tdb);
key.dsize = strlen("hi");
key.dptr = (void *)"hi";
data = tdb_fetch(tdb, key);
ok1(data.dsize == strlen("world"));
ok1(memcmp(data.dptr, "world", strlen("world")) == 0);
free(data.dptr);
tdb_close(tdb);
return exit_status();
}
ccan/tdb/test/run-incompatible.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
static unsigned int tdb_dumb_hash(TDB_DATA *key)
{
return key->dsize;
}
static void log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
{
unsigned int *count = tdb_get_logging_private(tdb);
if (strstr(fmt, "hash"))
(*count)++;
}
static unsigned int hdr_rwlocks(const char *fname)
{
struct tdb_header hdr;
int fd = open(fname, O_RDONLY);
if (fd == -1)
return -1;
if (read(fd, &hdr, sizeof(hdr)) != sizeof(hdr))
return -1;
close(fd);
return hdr.rwlocks;
}
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
unsigned int log_count, flags;
TDB_DATA d, r;
struct tdb_logging_context log_ctx = { log_fn, &log_count };
plan_tests(38 * 2);
for (flags = 0; flags <= TDB_CONVERT; flags += TDB_CONVERT) {
unsigned int rwmagic = TDB_HASH_RWLOCK_MAGIC;
if (flags & TDB_CONVERT)
tdb_convert(&rwmagic, sizeof(rwmagic));
/* Create an old-style hash. */
log_count = 0;
tdb = tdb_open_ex("run-incompatible.tdb", 0, flags,
O_CREAT|O_RDWR|O_TRUNC, 0600, &log_ctx,
NULL);
ok1(tdb);
ok1(log_count == 0);
d.dptr = (void *)"Hello";
d.dsize = 5;
ok1(tdb_store(tdb, d, d, TDB_INSERT) == 0);
tdb_close(tdb);
/* Should not have marked rwlocks field. */
ok1(hdr_rwlocks("run-incompatible.tdb") == 0);
/* We can still open any old-style with incompat flag. */
log_count = 0;
tdb = tdb_open_ex("run-incompatible.tdb", 0,
TDB_INCOMPATIBLE_HASH,
O_RDWR, 0600, &log_ctx, NULL);
ok1(tdb);
ok1(log_count == 0);
r = tdb_fetch(tdb, d);
ok1(r.dsize == 5);
free(r.dptr);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
log_count = 0;
tdb = tdb_open_ex("test/jenkins-le-hash.tdb", 0, 0, O_RDONLY,
0, &log_ctx, tdb_jenkins_hash);
ok1(tdb);
ok1(log_count == 0);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
log_count = 0;
tdb = tdb_open_ex("test/jenkins-be-hash.tdb", 0, 0, O_RDONLY,
0, &log_ctx, tdb_jenkins_hash);
ok1(tdb);
ok1(log_count == 0);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
/* OK, now create with incompatible flag, default hash. */
log_count = 0;
tdb = tdb_open_ex("run-incompatible.tdb", 0,
flags|TDB_INCOMPATIBLE_HASH,
O_CREAT|O_RDWR|O_TRUNC, 0600, &log_ctx,
NULL);
ok1(tdb);
ok1(log_count == 0);
d.dptr = (void *)"Hello";
d.dsize = 5;
ok1(tdb_store(tdb, d, d, TDB_INSERT) == 0);
tdb_close(tdb);
/* Should have marked rwlocks field. */
ok1(hdr_rwlocks("run-incompatible.tdb") == rwmagic);
/* Cannot open with old hash. */
log_count = 0;
tdb = tdb_open_ex("run-incompatible.tdb", 0, 0,
O_RDWR, 0600, &log_ctx, tdb_old_hash);
ok1(!tdb);
ok1(log_count == 1);
/* Can open with jenkins hash. */
log_count = 0;
tdb = tdb_open_ex("run-incompatible.tdb", 0, 0,
O_RDWR, 0600, &log_ctx, tdb_jenkins_hash);
ok1(tdb);
ok1(log_count == 0);
r = tdb_fetch(tdb, d);
ok1(r.dsize == 5);
free(r.dptr);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
/* Can open by letting it figure it out itself. */
log_count = 0;
tdb = tdb_open_ex("run-incompatible.tdb", 0, 0,
O_RDWR, 0600, &log_ctx, NULL);
ok1(tdb);
ok1(log_count == 0);
r = tdb_fetch(tdb, d);
ok1(r.dsize == 5);
free(r.dptr);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
/* We can also use incompatible hash with other hashes. */
log_count = 0;
tdb = tdb_open_ex("run-incompatible.tdb", 0,
flags|TDB_INCOMPATIBLE_HASH,
O_CREAT|O_RDWR|O_TRUNC, 0600, &log_ctx,
tdb_dumb_hash);
ok1(tdb);
ok1(log_count == 0);
d.dptr = (void *)"Hello";
d.dsize = 5;
ok1(tdb_store(tdb, d, d, TDB_INSERT) == 0);
tdb_close(tdb);
/* Should have marked rwlocks field. */
ok1(hdr_rwlocks("run-incompatible.tdb") == rwmagic);
/* It should not open if we don't specify. */
log_count = 0;
tdb = tdb_open_ex("run-incompatible.tdb", 0, 0, O_RDWR, 0,
&log_ctx, NULL);
ok1(!tdb);
ok1(log_count == 1);
/* Should reopen with correct hash. */
log_count = 0;
tdb = tdb_open_ex("run-incompatible.tdb", 0, 0, O_RDWR, 0,
&log_ctx, tdb_dumb_hash);
ok1(tdb);
ok1(log_count == 0);
r = tdb_fetch(tdb, d);
ok1(r.dsize == 5);
free(r.dptr);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
}
return exit_status();
}
ccan/tdb/test/run-nested-transactions.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <stdbool.h>
#include <err.h>
#include "logging.h"
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
TDB_DATA key, data;
plan_tests(27);
key.dsize = strlen("hi");
key.dptr = (void *)"hi";
tdb = tdb_open_ex("run-nested-transactions.tdb",
1024, TDB_CLEAR_IF_FIRST,
O_CREAT|O_TRUNC|O_RDWR, 0600, &taplogctx, NULL);
ok1(tdb);
/* No nesting by default. */
ok1(tdb_transaction_start(tdb) == 0);
data.dptr = (void *)"world";
data.dsize = strlen("world");
ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0);
data = tdb_fetch(tdb, key);
ok1(data.dsize == strlen("world"));
ok1(memcmp(data.dptr, "world", strlen("world")) == 0);
free(data.dptr);
ok1(tdb_transaction_start(tdb) != 0);
ok1(tdb_error(tdb) == TDB_ERR_NESTING);
data = tdb_fetch(tdb, key);
ok1(data.dsize == strlen("world"));
ok1(memcmp(data.dptr, "world", strlen("world")) == 0);
free(data.dptr);
ok1(tdb_transaction_commit(tdb) == 0);
data = tdb_fetch(tdb, key);
ok1(data.dsize == strlen("world"));
ok1(memcmp(data.dptr, "world", strlen("world")) == 0);
free(data.dptr);
tdb_close(tdb);
tdb = tdb_open_ex("run-nested-transactions.tdb",
1024, TDB_ALLOW_NESTING, O_RDWR, 0, &taplogctx, NULL);
ok1(tdb);
ok1(tdb_transaction_start(tdb) == 0);
ok1(tdb_transaction_start(tdb) == 0);
ok1(tdb_delete(tdb, key) == 0);
ok1(tdb_transaction_commit(tdb) == 0);
ok1(!tdb_exists(tdb, key));
ok1(tdb_transaction_cancel(tdb) == 0);
/* Surprise! Kills inner "committed" transaction. */
ok1(tdb_exists(tdb, key));
ok1(tdb_transaction_start(tdb) == 0);
ok1(tdb_transaction_start(tdb) == 0);
ok1(tdb_delete(tdb, key) == 0);
ok1(tdb_transaction_commit(tdb) == 0);
ok1(!tdb_exists(tdb, key));
ok1(tdb_transaction_commit(tdb) == 0);
ok1(!tdb_exists(tdb, key));
tdb_close(tdb);
return exit_status();
}
ccan/tdb/test/run-nested-traverse.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include "lock-tracking.h"
#define fcntl fcntl_with_lockcheck
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#undef fcntl
#include <stdlib.h>
#include <stdbool.h>
#include <err.h>
#include "external-agent.h"
#include "logging.h"
static struct agent *agent;
static bool correct_key(TDB_DATA key)
{
return key.dsize == strlen("hi")
&& memcmp(key.dptr, "hi", key.dsize) == 0;
}
static bool correct_data(TDB_DATA data)
{
return data.dsize == strlen("world")
&& memcmp(data.dptr, "world", data.dsize) == 0;
}
static int traverse2(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data,
void *p)
{
ok1(correct_key(key));
ok1(correct_data(data));
return 0;
}
static int traverse1(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data,
void *p)
{
ok1(correct_key(key));
ok1(correct_data(data));
ok1(external_agent_operation(agent, TRANSACTION_START, tdb_name(tdb))
== WOULD_HAVE_BLOCKED);
tdb_traverse(tdb, traverse2, NULL);
/* That should *not* release the transaction lock! */
ok1(external_agent_operation(agent, TRANSACTION_START, tdb_name(tdb))
== WOULD_HAVE_BLOCKED);
return 0;
}
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
TDB_DATA key, data;
plan_tests(17);
agent = prepare_external_agent();
if (!agent)
err(1, "preparing agent");
tdb = tdb_open_ex("run-nested-traverse.tdb", 1024, TDB_CLEAR_IF_FIRST,
O_CREAT|O_TRUNC|O_RDWR, 0600, &taplogctx, NULL);
ok1(tdb);
ok1(external_agent_operation(agent, OPEN, tdb_name(tdb)) == SUCCESS);
ok1(external_agent_operation(agent, TRANSACTION_START, tdb_name(tdb))
== SUCCESS);
ok1(external_agent_operation(agent, TRANSACTION_COMMIT, tdb_name(tdb))
== SUCCESS);
key.dsize = strlen("hi");
key.dptr = (void *)"hi";
data.dptr = (void *)"world";
data.dsize = strlen("world");
ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0);
tdb_traverse(tdb, traverse1, NULL);
tdb_traverse_read(tdb, traverse1, NULL);
tdb_close(tdb);
return exit_status();
}
ccan/tdb/test/run-no-lock-during-traverse.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <unistd.h>
#include "lock-tracking.h"
#define fcntl fcntl_with_lockcheck
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
#include "logging.h"
#undef fcntl
#define NUM_ENTRIES 10
static bool prepare_entries(struct tdb_context *tdb)
{
unsigned int i;
TDB_DATA key, data;
for (i = 0; i < NUM_ENTRIES; i++) {
key.dsize = sizeof(i);
key.dptr = (void *)&i;
data.dsize = strlen("world");
data.dptr = (void *)"world";
if (tdb_store(tdb, key, data, 0) != 0)
return false;
}
return true;
}
static void delete_entries(struct tdb_context *tdb)
{
unsigned int i;
TDB_DATA key;
for (i = 0; i < NUM_ENTRIES; i++) {
key.dsize = sizeof(i);
key.dptr = (void *)&i;
ok1(tdb_delete(tdb, key) == 0);
}
}
/* We don't know how many times this will run. */
static int delete_other(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data,
void *private_data)
{
unsigned int i;
memcpy(&i, key.dptr, 4);
i = (i + 1) % NUM_ENTRIES;
key.dptr = (void *)&i;
if (tdb_delete(tdb, key) != 0)
(*(int *)private_data)++;
return 0;
}
static int delete_self(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data,
void *private_data)
{
ok1(tdb_delete(tdb, key) == 0);
return 0;
}
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
int errors = 0;
plan_tests(41);
tdb = tdb_open_ex("run-no-lock-during-traverse.tdb",
1024, TDB_CLEAR_IF_FIRST, O_CREAT|O_TRUNC|O_RDWR,
0600, &taplogctx, NULL);
ok1(tdb);
ok1(prepare_entries(tdb));
ok1(locking_errors == 0);
ok1(tdb_lockall(tdb) == 0);
ok1(locking_errors == 0);
tdb_traverse(tdb, delete_other, &errors);
ok1(errors == 0);
ok1(locking_errors == 0);
ok1(tdb_unlockall(tdb) == 0);
ok1(prepare_entries(tdb));
ok1(locking_errors == 0);
ok1(tdb_lockall(tdb) == 0);
ok1(locking_errors == 0);
tdb_traverse(tdb, delete_self, NULL);
ok1(locking_errors == 0);
ok1(tdb_unlockall(tdb) == 0);
ok1(prepare_entries(tdb));
ok1(locking_errors == 0);
ok1(tdb_lockall(tdb) == 0);
ok1(locking_errors == 0);
delete_entries(tdb);
ok1(locking_errors == 0);
ok1(tdb_unlockall(tdb) == 0);
ok1(tdb_close(tdb) == 0);
return exit_status();
}
ccan/tdb/test/run-oldhash.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
#include "logging.h"
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
plan_tests(8);
/* Old format (with zeroes in the hash magic fields) should
* open with any hash (since we don't know what hash they used). */
tdb = tdb_open_ex("test/old-nohash-le.tdb", 0, 0, O_RDWR, 0,
&taplogctx, NULL);
ok1(tdb);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
tdb = tdb_open_ex("test/old-nohash-be.tdb", 0, 0, O_RDWR, 0,
&taplogctx, NULL);
ok1(tdb);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
tdb = tdb_open_ex("test/old-nohash-le.tdb", 0, 0, O_RDWR, 0,
&taplogctx, tdb_jenkins_hash);
ok1(tdb);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
tdb = tdb_open_ex("test/old-nohash-be.tdb", 0, 0, O_RDWR, 0,
&taplogctx, tdb_jenkins_hash);
ok1(tdb);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
return exit_status();
}
ccan/tdb/test/run-open-during-transaction.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <unistd.h>
#include "lock-tracking.h"
static ssize_t pwrite_check(int fd, const void *buf, size_t count, off_t offset);
static ssize_t write_check(int fd, const void *buf, size_t count);
static int ftruncate_check(int fd, off_t length);
#define pwrite pwrite_check
#define write write_check
#define fcntl fcntl_with_lockcheck
#define ftruncate ftruncate_check
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <stdbool.h>
#include <stdarg.h>
#include <err.h>
#include "external-agent.h"
#include "logging.h"
static struct agent *agent;
static bool opened;
static int errors = 0;
static bool clear_if_first;
#define TEST_DBNAME "run-open-during-transaction.tdb"
#undef write
#undef pwrite
#undef fcntl
#undef ftruncate
static bool is_same(const char *snapshot, const char *latest, off_t len)
{
unsigned i;
for (i = 0; i < len; i++) {
if (snapshot[i] != latest[i])
return false;
}
return true;
}
static bool compare_file(int fd, const char *snapshot, off_t snapshot_len)
{
char *contents;
bool same;
/* over-length read serves as length check. */
contents = malloc(snapshot_len+1);
same = pread(fd, contents, snapshot_len+1, 0) == snapshot_len
&& is_same(snapshot, contents, snapshot_len);
free(contents);
return same;
}
static void check_file_intact(int fd)
{
enum agent_return ret;
struct stat st;
char *contents;
fstat(fd, &st);
contents = malloc(st.st_size);
if (pread(fd, contents, st.st_size, 0) != st.st_size) {
diag("Read fail");
errors++;
return;
}
/* Ask agent to open file. */
ret = external_agent_operation(agent, clear_if_first ?
OPEN_WITH_CLEAR_IF_FIRST :
OPEN,
TEST_DBNAME);
/* It's OK to open it, but it must not have changed! */
if (!compare_file(fd, contents, st.st_size)) {
diag("Agent changed file after opening %s",
agent_return_name(ret));
errors++;
}
if (ret == SUCCESS) {
ret = external_agent_operation(agent, CLOSE, NULL);
if (ret != SUCCESS) {
diag("Agent failed to close tdb: %s",
agent_return_name(ret));
errors++;
}
} else if (ret != WOULD_HAVE_BLOCKED) {
diag("Agent opening file gave %s",
agent_return_name(ret));
errors++;
}
free(contents);
}
static void after_unlock(int fd)
{
if (opened)
check_file_intact(fd);
}
static ssize_t pwrite_check(int fd,
const void *buf, size_t count, off_t offset)
{
if (opened)
check_file_intact(fd);
return pwrite(fd, buf, count, offset);
}
static ssize_t write_check(int fd, const void *buf, size_t count)
{
if (opened)
check_file_intact(fd);
return write(fd, buf, count);
}
static int ftruncate_check(int fd, off_t length)
{
if (opened)
check_file_intact(fd);
return ftruncate(fd, length);
}
int main(int argc, char *argv[])
{
const int flags[] = { TDB_DEFAULT,
TDB_CLEAR_IF_FIRST,
TDB_NOMMAP,
TDB_CLEAR_IF_FIRST | TDB_NOMMAP };
int i;
struct tdb_context *tdb;
TDB_DATA key, data;
plan_tests(20);
agent = prepare_external_agent();
if (!agent)
err(1, "preparing agent");
unlock_callback = after_unlock;
for (i = 0; i < sizeof(flags)/sizeof(flags[0]); i++) {
clear_if_first = (flags[i] & TDB_CLEAR_IF_FIRST);
diag("Test with %s and %s\n",
clear_if_first ? "CLEAR" : "DEFAULT",
(flags[i] & TDB_NOMMAP) ? "no mmap" : "mmap");
unlink(TEST_DBNAME);
tdb = tdb_open_ex(TEST_DBNAME, 1024, flags[i],
O_CREAT|O_TRUNC|O_RDWR, 0600,
&taplogctx, NULL);
ok1(tdb);
opened = true;
ok1(tdb_transaction_start(tdb) == 0);
key.dsize = strlen("hi");
key.dptr = (void *)"hi";
data.dptr = (void *)"world";
data.dsize = strlen("world");
ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0);
ok1(tdb_transaction_commit(tdb) == 0);
ok(!errors, "We had %u open errors", errors);
opened = false;
tdb_close(tdb);
}
return exit_status();
}
ccan/tdb/test/run-readonly-check.c deleted 100644 → 0
View file @ 98b8ada2
/* We should be able to tdb_check a O_RDONLY tdb, and we were previously allowed
* to tdb_check() inside a transaction (though that's paranoia!). */
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
#include "logging.h"
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
TDB_DATA key, data;
plan_tests(11);
tdb = tdb_open_ex("run-readonly-check.tdb", 1024,
TDB_DEFAULT,
O_CREAT|O_TRUNC|O_RDWR, 0600, &taplogctx, NULL);
ok1(tdb);
key.dsize = strlen("hi");
key.dptr = (void *)"hi";
data.dsize = strlen("world");
data.dptr = (void *)"world";
ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0);
ok1(tdb_check(tdb, NULL, NULL) == 0);
/* We are also allowed to do a check inside a transaction. */
ok1(tdb_transaction_start(tdb) == 0);
ok1(tdb_check(tdb, NULL, NULL) == 0);
ok1(tdb_close(tdb) == 0);
tdb = tdb_open_ex("run-readonly-check.tdb", 1024,
TDB_DEFAULT, O_RDONLY, 0, &taplogctx, NULL);
ok1(tdb);
ok1(tdb_store(tdb, key, data, TDB_MODIFY) == -1);
ok1(tdb_error(tdb) == TDB_ERR_RDONLY);
ok1(tdb_check(tdb, NULL, NULL) == 0);
ok1(tdb_close(tdb) == 0);
return exit_status();
}
ccan/tdb/test/run-rwlock-check.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
static void log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
{
unsigned int *count = tdb_get_logging_private(tdb);
if (strstr(fmt, "spinlocks"))
(*count)++;
}
/* The code should barf on TDBs created with rwlocks. */
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
unsigned int log_count;
struct tdb_logging_context log_ctx = { log_fn, &log_count };
plan_tests(4);
/* We should fail to open rwlock-using tdbs of either endian. */
log_count = 0;
tdb = tdb_open_ex("test/rwlock-le.tdb", 0, 0, O_RDWR, 0,
&log_ctx, NULL);
ok1(!tdb);
ok1(log_count == 1);
log_count = 0;
tdb = tdb_open_ex("test/rwlock-be.tdb", 0, 0, O_RDWR, 0,
&log_ctx, NULL);
ok1(!tdb);
ok1(log_count == 1);
return exit_status();
}
ccan/tdb/test/run-summary.c deleted 100644 → 0
View file @ 98b8ada2
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tdb/summary.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
int main(int argc, char *argv[])
{
unsigned int i, j;
struct tdb_context *tdb;
int flags[] = { TDB_INTERNAL, TDB_DEFAULT, TDB_NOMMAP,
TDB_INTERNAL|TDB_CONVERT, TDB_CONVERT,
TDB_NOMMAP|TDB_CONVERT };
TDB_DATA key = { (unsigned char *)&j, sizeof(j) };
TDB_DATA data = { (unsigned char *)&j, sizeof(j) };
char *summary;
plan_tests(sizeof(flags) / sizeof(flags[0]) * 14);
for (i = 0; i < sizeof(flags) / sizeof(flags[0]); i++) {
tdb = tdb_open("run-summary.tdb", 131, flags[i],
O_RDWR|O_CREAT|O_TRUNC, 0600);
ok1(tdb);
if (!tdb)
continue;
/* Put some stuff in there. */
for (j = 0; j < 500; j++) {
/* Make sure padding varies to we get some graphs! */
data.dsize = j % (sizeof(j) + 1);
if (tdb_store(tdb, key, data, TDB_REPLACE) != 0)
fail("Storing in tdb");
}
summary = tdb_summary(tdb);
diag("%s", summary);
ok1(strstr(summary, "Size of file/data: "));
ok1(strstr(summary, "Number of records: 500\n"));
ok1(strstr(summary, "Smallest/average/largest keys: 4/4/4\n"));
ok1(strstr(summary, "Smallest/average/largest data: 0/2/4\n"));
ok1(strstr(summary, "Smallest/average/largest padding: "));
ok1(strstr(summary, "Number of dead records: 0\n"));
ok1(strstr(summary, "Number of free records: 1\n"));
ok1(strstr(summary, "Smallest/average/largest free records: "));
ok1(strstr(summary, "Number of hash chains: 131\n"));
ok1(strstr(summary, "Smallest/average/largest hash chains: "));
ok1(strstr(summary, "Number of uncoalesced records: 0\n"));
ok1(strstr(summary, "Smallest/average/largest uncoalesced runs: 0/0/0\n"));
ok1(strstr(summary, "Percentage keys/data/padding/free/dead/rechdrs&tailers/hashes: "));
free(summary);
tdb_close(tdb);
}
return exit_status();
}
ccan/tdb/test/run-transaction-expand.c deleted 100644 → 0
View file @ 98b8ada2
/* We need this otherwise fcntl locking fails. */
#define _FILE_OFFSET_BITS 64
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb_private.h>
/* Speed up the tests: setting TDB_NOSYNC removed recovery altogether. */
static inline int fake_fsync(int fd)
{
return 0;
}
#define fsync fake_fsync
#ifdef MS_SYNC
static inline int fake_msync(void *addr, size_t length, int flags)
{
return 0;
}
#define msync fake_msync
#endif
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
#include "logging.h"
static void write_record(struct tdb_context *tdb, size_t extra_len,
TDB_DATA *data)
{
TDB_DATA key;
key.dsize = strlen("hi");
key.dptr = (void *)"hi";
data->dsize += extra_len;
tdb_transaction_start(tdb);
tdb_store(tdb, key, *data, TDB_REPLACE);
tdb_transaction_commit(tdb);
}
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
size_t i;
TDB_DATA data;
struct tdb_record rec;
tdb_off_t off;
plan_tests(4);
tdb = tdb_open_ex("run-transaction-expand.tdb",
1024, TDB_CLEAR_IF_FIRST,
O_CREAT|O_TRUNC|O_RDWR, 0600, &taplogctx, NULL);
ok1(tdb);
data.dsize = 0;
data.dptr = calloc(1000, getpagesize());
/* Simulate a slowly growing record. */
for (i = 0; i < 1000; i++)
write_record(tdb, getpagesize(), &data);
tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &off);
tdb_read(tdb, off, &rec, sizeof(rec), DOCONV());
diag("TDB size = %zu, recovery = %u-%u",
(size_t)tdb->map_size, off, off + sizeof(rec) + rec.rec_len);
/* We should only be about 5 times larger than largest record. */
ok1(tdb->map_size < 6 * i * getpagesize());
tdb_close(tdb);
tdb = tdb_open_ex("run-transaction-expand.tdb",
1024, TDB_CLEAR_IF_FIRST,
O_CREAT|O_TRUNC|O_RDWR, 0600, &taplogctx, NULL);
ok1(tdb);
data.dsize = 0;
/* Simulate a slowly growing record, repacking to keep
* recovery area at end. */
for (i = 0; i < 1000; i++) {
write_record(tdb, getpagesize(), &data);
if (i % 10 == 0)
tdb_repack(tdb);
}
tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &off);
tdb_read(tdb, off, &rec, sizeof(rec), DOCONV());
diag("TDB size = %zu, recovery = %u-%u",
(size_t)tdb->map_size, off, off + sizeof(rec) + rec.rec_len);
/* We should only be about 4 times larger than largest record. */
ok1(tdb->map_size < 5 * i * getpagesize());
tdb_close(tdb);
free(data.dptr);
return exit_status();
}
ccan/tdb/test/run-traverse-in-transaction.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include "lock-tracking.h"
#define fcntl fcntl_with_lockcheck
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#undef fcntl_with_lockcheck
#include <stdlib.h>
#include <stdbool.h>
#include <err.h>
#include "external-agent.h"
#include "logging.h"
static struct agent *agent;
static bool correct_key(TDB_DATA key)
{
return key.dsize == strlen("hi")
&& memcmp(key.dptr, "hi", key.dsize) == 0;
}
static bool correct_data(TDB_DATA data)
{
return data.dsize == strlen("world")
&& memcmp(data.dptr, "world", data.dsize) == 0;
}
static int traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA data,
void *p)
{
ok1(correct_key(key));
ok1(correct_data(data));
return 0;
}
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
TDB_DATA key, data;
plan_tests(13);
agent = prepare_external_agent();
if (!agent)
err(1, "preparing agent");
tdb = tdb_open_ex("run-traverse-in-transaction.tdb",
1024, TDB_CLEAR_IF_FIRST, O_CREAT|O_TRUNC|O_RDWR,
0600, &taplogctx, NULL);
ok1(tdb);
key.dsize = strlen("hi");
key.dptr = (void *)"hi";
data.dptr = (void *)"world";
data.dsize = strlen("world");
ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0);
ok1(external_agent_operation(agent, OPEN, tdb_name(tdb)) == SUCCESS);
ok1(tdb_transaction_start(tdb) == 0);
ok1(external_agent_operation(agent, TRANSACTION_START, tdb_name(tdb))
== WOULD_HAVE_BLOCKED);
tdb_traverse(tdb, traverse, NULL);
/* That should *not* release the transaction lock! */
ok1(external_agent_operation(agent, TRANSACTION_START, tdb_name(tdb))
== WOULD_HAVE_BLOCKED);
tdb_traverse_read(tdb, traverse, NULL);
/* That should *not* release the transaction lock! */
ok1(external_agent_operation(agent, TRANSACTION_START, tdb_name(tdb))
== WOULD_HAVE_BLOCKED);
ok1(tdb_transaction_commit(tdb) == 0);
/* Now we should be fine. */
ok1(external_agent_operation(agent, TRANSACTION_START, tdb_name(tdb))
== SUCCESS);
tdb_close(tdb);
return exit_status();
}
ccan/tdb/test/run-wronghash-fail.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
static void log_fn(struct tdb_context *tdb, enum tdb_debug_level level, const char *fmt, ...)
{
unsigned int *count = tdb_get_logging_private(tdb);
if (strstr(fmt, "hash"))
(*count)++;
}
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
unsigned int log_count;
TDB_DATA d;
struct tdb_logging_context log_ctx = { log_fn, &log_count };
plan_tests(28);
/* Create with default hash. */
log_count = 0;
tdb = tdb_open_ex("run-wronghash-fail.tdb", 0, 0,
O_CREAT|O_RDWR|O_TRUNC, 0600, &log_ctx, NULL);
ok1(tdb);
ok1(log_count == 0);
d.dptr = (void *)"Hello";
d.dsize = 5;
ok1(tdb_store(tdb, d, d, TDB_INSERT) == 0);
tdb_close(tdb);
/* Fail to open with different hash. */
tdb = tdb_open_ex("run-wronghash-fail.tdb", 0, 0, O_RDWR, 0,
&log_ctx, tdb_jenkins_hash);
ok1(!tdb);
ok1(log_count == 1);
/* Create with different hash. */
log_count = 0;
tdb = tdb_open_ex("run-wronghash-fail.tdb", 0, 0,
O_CREAT|O_RDWR|O_TRUNC,
0600, &log_ctx, tdb_jenkins_hash);
ok1(tdb);
ok1(log_count == 0);
tdb_close(tdb);
/* Endian should be no problem. */
log_count = 0;
tdb = tdb_open_ex("test/jenkins-le-hash.tdb", 0, 0, O_RDWR, 0,
&log_ctx, tdb_old_hash);
ok1(!tdb);
ok1(log_count == 1);
log_count = 0;
tdb = tdb_open_ex("test/jenkins-be-hash.tdb", 0, 0, O_RDWR, 0,
&log_ctx, tdb_old_hash);
ok1(!tdb);
ok1(log_count == 1);
log_count = 0;
/* Fail to open with old default hash. */
tdb = tdb_open_ex("run-wronghash-fail.tdb", 0, 0, O_RDWR, 0,
&log_ctx, tdb_old_hash);
ok1(!tdb);
ok1(log_count == 1);
log_count = 0;
tdb = tdb_open_ex("test/jenkins-le-hash.tdb", 0, 0, O_RDONLY,
0, &log_ctx, tdb_jenkins_hash);
ok1(tdb);
ok1(log_count == 0);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
log_count = 0;
tdb = tdb_open_ex("test/jenkins-be-hash.tdb", 0, 0, O_RDONLY,
0, &log_ctx, tdb_jenkins_hash);
ok1(tdb);
ok1(log_count == 0);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
/* It should open with jenkins hash if we don't specify. */
log_count = 0;
tdb = tdb_open_ex("test/jenkins-le-hash.tdb", 0, 0, O_RDWR, 0,
&log_ctx, NULL);
ok1(tdb);
ok1(log_count == 0);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
log_count = 0;
tdb = tdb_open_ex("test/jenkins-be-hash.tdb", 0, 0, O_RDWR, 0,
&log_ctx, NULL);
ok1(tdb);
ok1(log_count == 0);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
log_count = 0;
tdb = tdb_open_ex("run-wronghash-fail.tdb", 0, 0, O_RDONLY,
0, &log_ctx, NULL);
ok1(tdb);
ok1(log_count == 0);
ok1(tdb_check(tdb, NULL, NULL) == 0);
tdb_close(tdb);
return exit_status();
}
ccan/tdb/test/run-zero-append.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
#include "logging.h"
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
TDB_DATA key, data;
plan_tests(4);
tdb = tdb_open_ex(NULL, 1024, TDB_INTERNAL, O_CREAT|O_TRUNC|O_RDWR,
0600, &taplogctx, NULL);
ok1(tdb);
/* Tickle bug on appending zero length buffer to zero length buffer. */
key.dsize = strlen("hi");
key.dptr = (void *)"hi";
data.dptr = (void *)"world";
data.dsize = 0;
ok1(tdb_append(tdb, key, data) == 0);
ok1(tdb_append(tdb, key, data) == 0);
data = tdb_fetch(tdb, key);
ok1(data.dsize == 0);
tdb_close(tdb);
free(data.dptr);
return exit_status();
}
ccan/tdb/test/run.c deleted 100644 → 0
View file @ 98b8ada2
#define _XOPEN_SOURCE 500
#include <ccan/tdb/tdb.h>
#include <ccan/tdb/io.c>
#include <ccan/tdb/tdb.c>
#include <ccan/tdb/lock.c>
#include <ccan/tdb/freelist.c>
#include <ccan/tdb/traverse.c>
#include <ccan/tdb/transaction.c>
#include <ccan/tdb/error.c>
#include <ccan/tdb/open.c>
#include <ccan/tdb/check.c>
#include <ccan/tdb/hash.c>
#include <ccan/tap/tap.h>
#include <stdlib.h>
#include <err.h>
#include "logging.h"
int main(int argc, char *argv[])
{
struct tdb_context *tdb;
TDB_DATA key, data;
plan_tests(10);
tdb = tdb_open_ex("run.tdb", 1024, TDB_CLEAR_IF_FIRST,
O_CREAT|O_TRUNC|O_RDWR, 0600, &taplogctx, NULL);
ok1(tdb);
key.dsize = strlen("hi");
key.dptr = (void *)"hi";
data.dsize = strlen("world");
data.dptr = (void *)"world";
ok1(tdb_store(tdb, key, data, TDB_MODIFY) < 0);
ok1(tdb_error(tdb) == TDB_ERR_NOEXIST);
ok1(tdb_store(tdb, key, data, TDB_INSERT) == 0);
ok1(tdb_store(tdb, key, data, TDB_INSERT) < 0);
ok1(tdb_error(tdb) == TDB_ERR_EXISTS);
ok1(tdb_store(tdb, key, data, TDB_MODIFY) == 0);
data = tdb_fetch(tdb, key);
ok1(data.dsize == strlen("world"));
ok1(memcmp(data.dptr, "world", strlen("world")) == 0);
free(data.dptr);
key.dsize++;
data = tdb_fetch(tdb, key);
ok1(data.dptr == NULL);
tdb_close(tdb);
return exit_status();
}
ccan/tdb/tools/Makefile deleted 100644 → 0
View file @ 98b8ada2
LDLIBS:=../../tdb.o ../../tally.o
CFLAGS:=-I../../.. -Wall -O3 #-g -pg
LDFLAGS:=-L../../..
default: replay_trace tdbtorture tdbdump tdbtool starvation mktdb speed
benchmark: replay_trace
@trap "rm -f /tmp/trace.$$$$" 0; for f in benchmarks/*.rz; do if runzip -k $$f -o /tmp/trace.$$$$ && echo -n "$$f": && ./replay_trace --quiet -n 5 replay.tdb /tmp/trace.$$$$ && rm /tmp/trace.$$$$; then rm -f /tmp/trace.$$$$; else exit 1; fi; done
REPLAY_LIBS=$(LDLIBS) ../../str_talloc.o ../../grab_file.o ../../talloc.o ../../noerr.o
replay_trace: replay_trace.c keywords.c $(REPLAY_LIBS)
$(LINK.c) $< $(LOADLIBES) $(REPLAY_LIBS) -o $@
keywords.c: keywords.gperf
gperf $< > $@
check: replay_trace
@rm -f *.reduced_trace
@set -e; for f in tests/*.trace.tar.bz2; do \
tar xvfj $$f; \
./replay_trace replay.tdb *.reduced_trace; \
rm -f *.reduced_trace; \
done
# Usage: make mytest.trace.tar.bz2 TRACEFILES=*.trace
%.trace.tar.bz2: $(patsubst %.trace,%.reduced_trace,$(wildcard $(TRACEFILES)))
tar cvfj $@ $^
%.reduced_trace: %.trace
@sed 's/\(^[0-9]* traverse\) .*/\1fn/' < $^ > $@
clean:
rm -f replay_trace tdbtorture tdbdump tdbtool speed *.o
ccan/tdb/tools/keywords.c deleted 100644 → 0
View file @ 98b8ada2
/* ANSI-C code produced by gperf version 3.0.3 */
/* Command-line: gperf keywords.gperf */
/* Computed positions: -k'5,$' */
#if !((' ' == 32) && ('!' == 33) && ('"' == 34) && ('#' == 35) \
&& ('%' == 37) && ('&' == 38) && ('\'' == 39) && ('(' == 40) \
&& (')' == 41) && ('*' == 42) && ('+' == 43) && (',' == 44) \
&& ('-' == 45) && ('.' == 46) && ('/' == 47) && ('0' == 48) \
&& ('1' == 49) && ('2' == 50) && ('3' == 51) && ('4' == 52) \
&& ('5' == 53) && ('6' == 54) && ('7' == 55) && ('8' == 56) \
&& ('9' == 57) && (':' == 58) && (';' == 59) && ('<' == 60) \
&& ('=' == 61) && ('>' == 62) && ('?' == 63) && ('A' == 65) \
&& ('B' == 66) && ('C' == 67) && ('D' == 68) && ('E' == 69) \
&& ('F' == 70) && ('G' == 71) && ('H' == 72) && ('I' == 73) \
&& ('J' == 74) && ('K' == 75) && ('L' == 76) && ('M' == 77) \
&& ('N' == 78) && ('O' == 79) && ('P' == 80) && ('Q' == 81) \
&& ('R' == 82) && ('S' == 83) && ('T' == 84) && ('U' == 85) \
&& ('V' == 86) && ('W' == 87) && ('X' == 88) && ('Y' == 89) \
&& ('Z' == 90) && ('[' == 91) && ('\\' == 92) && (']' == 93) \
&& ('^' == 94) && ('_' == 95) && ('a' == 97) && ('b' == 98) \
&& ('c' == 99) && ('d' == 100) && ('e' == 101) && ('f' == 102) \
&& ('g' == 103) && ('h' == 104) && ('i' == 105) && ('j' == 106) \
&& ('k' == 107) && ('l' == 108) && ('m' == 109) && ('n' == 110) \
&& ('o' == 111) && ('p' == 112) && ('q' == 113) && ('r' == 114) \
&& ('s' == 115) && ('t' == 116) && ('u' == 117) && ('v' == 118) \
&& ('w' == 119) && ('x' == 120) && ('y' == 121) && ('z' == 122) \
&& ('{' == 123) && ('|' == 124) && ('}' == 125) && ('~' == 126))
/* The character set is not based on ISO-646. */
#error "gperf generated tables don't work with this execution character set. Please report a bug to <bug-gnu-gperf@gnu.org>."
#endif
#line 1 "keywords.gperf"
#line 4 "keywords.gperf"
struct op_table {
const char *name;
enum op_type type;
void (*enhance_op)(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[]);
};
/* maximum key range = 53, duplicates = 0 */
#ifdef __GNUC__
__inline
#else
#ifdef __cplusplus
inline
#endif
#endif
static unsigned int
hash_keyword (register const char *str, register unsigned int len)
{
static const unsigned char asso_values[] =
{
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 45, 61, 30,
5, 0, 0, 5, 5, 61, 61, 0, 0, 0,
20, 61, 20, 61, 25, 0, 5, 0, 61, 0,
61, 5, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61, 61, 61, 61, 61,
61, 61, 61, 61, 61, 61
};
return len + asso_values[(unsigned char)str[4]] + asso_values[(unsigned char)str[len - 1]];
}
#ifdef __GNUC__
__inline
#ifdef __GNUC_STDC_INLINE__
__attribute__ ((__gnu_inline__))
#endif
#endif
const struct op_table *
find_keyword (register const char *str, register unsigned int len)
{
enum
{
TOTAL_KEYWORDS = 35,
MIN_WORD_LENGTH = 8,
MAX_WORD_LENGTH = 30,
MIN_HASH_VALUE = 8,
MAX_HASH_VALUE = 60
};
static const struct op_table wordlist[] =
{
{""}, {""}, {""}, {""}, {""}, {""}, {""}, {""},
#line 44 "keywords.gperf"
{"traverse", OP_TDB_TRAVERSE, op_add_traverse,},
#line 33 "keywords.gperf"
{"tdb_store", OP_TDB_STORE, op_add_store,},
#line 32 "keywords.gperf"
{"tdb_exists", OP_TDB_EXISTS, op_add_key_ret,},
#line 16 "keywords.gperf"
{"tdb_lockall", OP_TDB_LOCKALL, op_add_nothing,},
#line 36 "keywords.gperf"
{"tdb_wipe_all", OP_TDB_WIPE_ALL, op_add_wipe_all,},
#line 20 "keywords.gperf"
{"tdb_unlockall", OP_TDB_UNLOCKALL, op_add_nothing,},
#line 48 "keywords.gperf"
{"tdb_fetch", OP_TDB_FETCH, op_add_key_data,},
#line 49 "keywords.gperf"
{"tdb_delete", OP_TDB_DELETE, op_add_key_ret,},
#line 17 "keywords.gperf"
{"tdb_lockall_mark", OP_TDB_LOCKALL_MARK, op_add_nothing,},
#line 46 "keywords.gperf"
{"tdb_firstkey", OP_TDB_FIRSTKEY, op_add_key,},
#line 18 "keywords.gperf"
{"tdb_lockall_unmark", OP_TDB_LOCKALL_UNMARK, op_add_nothing,},
#line 35 "keywords.gperf"
{"tdb_get_seqnum", OP_TDB_GET_SEQNUM, op_add_seqnum,},
#line 19 "keywords.gperf"
{"tdb_lockall_nonblock", OP_TDB_LOCKALL_NONBLOCK, op_add_nothing,},
#line 21 "keywords.gperf"
{"tdb_lockall_read", OP_TDB_LOCKALL_READ, op_add_nothing,},
{""},
#line 23 "keywords.gperf"
{"tdb_unlockall_read", OP_TDB_UNLOCKALL_READ, op_add_nothing,},
{""},
#line 22 "keywords.gperf"
{"tdb_lockall_read_nonblock", OP_TDB_LOCKALL_READ_NONBLOCK, op_add_nothing,},
#line 43 "keywords.gperf"
{"tdb_traverse_end", OP_TDB_TRAVERSE_END, op_analyze_traverse,},
#line 38 "keywords.gperf"
{"tdb_transaction_cancel", OP_TDB_TRANSACTION_CANCEL, op_analyze_transaction,},
#line 42 "keywords.gperf"
{"tdb_traverse_start", OP_TDB_TRAVERSE_START, op_add_traverse_start,},
{""},
#line 45 "keywords.gperf"
{"traversefn", OP_TDB_TRAVERSE, op_add_traversefn,},
#line 37 "keywords.gperf"
{"tdb_transaction_start", OP_TDB_TRANSACTION_START, op_add_transaction,},
#line 39 "keywords.gperf"
{"tdb_transaction_commit", OP_TDB_TRANSACTION_COMMIT, op_analyze_transaction,},
#line 41 "keywords.gperf"
{"tdb_traverse_read_start", OP_TDB_TRAVERSE_READ_START, op_add_traverse_start,},
{""},
#line 50 "keywords.gperf"
{"tdb_repack", OP_TDB_REPACK, op_add_nothing,},
#line 47 "keywords.gperf"
{"tdb_nextkey", OP_TDB_NEXTKEY, op_add_key_data,},
{""}, {""}, {""},
#line 40 "keywords.gperf"
{"tdb_transaction_prepare_commit", OP_TDB_TRANSACTION_PREPARE_COMMIT, op_add_nothing,},
#line 31 "keywords.gperf"
{"tdb_parse_record", OP_TDB_PARSE_RECORD, op_add_key_ret,},
{""},
#line 24 "keywords.gperf"
{"tdb_chainlock", OP_TDB_CHAINLOCK, op_add_chainlock,},
{""},
#line 28 "keywords.gperf"
{"tdb_chainunlock", OP_TDB_CHAINUNLOCK, op_analyze_chainlock,},
{""}, {""},
#line 26 "keywords.gperf"
{"tdb_chainlock_mark", OP_TDB_CHAINLOCK_MARK, op_add_key,},
{""},
#line 27 "keywords.gperf"
{"tdb_chainlock_unmark", OP_TDB_CHAINLOCK_UNMARK, op_add_key,},
{""},
#line 25 "keywords.gperf"
{"tdb_chainlock_nonblock", OP_TDB_CHAINLOCK_NONBLOCK, op_add_chainlock_ret,},
#line 29 "keywords.gperf"
{"tdb_chainlock_read", OP_TDB_CHAINLOCK_READ, op_add_chainlock,},
{""},
#line 30 "keywords.gperf"
{"tdb_chainunlock_read", OP_TDB_CHAINUNLOCK_READ, op_analyze_chainlock,},
{""}, {""}, {""}, {""},
#line 34 "keywords.gperf"
{"tdb_append", OP_TDB_APPEND, op_add_append,}
};
if (len <= MAX_WORD_LENGTH && len >= MIN_WORD_LENGTH)
{
register int key = hash_keyword (str, len);
if (key <= MAX_HASH_VALUE && key >= 0)
{
register const char *s = wordlist[key].name;
if (*str == *s && !strcmp (str + 1, s + 1))
return &wordlist[key];
}
}
return 0;
}
ccan/tdb/tools/keywords.gperf deleted 100644 → 0
View file @ 98b8ada2
%{
%}
%language=ANSI-C
struct op_table {
const char *name;
enum op_type type;
void (*enhance_op)(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[]);
};
%define hash-function-name hash_keyword
%define lookup-function-name find_keyword
%readonly-tables
%struct-type
%enum
%%
tdb_lockall, OP_TDB_LOCKALL, op_add_nothing,
tdb_lockall_mark, OP_TDB_LOCKALL_MARK, op_add_nothing,
tdb_lockall_unmark, OP_TDB_LOCKALL_UNMARK, op_add_nothing,
tdb_lockall_nonblock, OP_TDB_LOCKALL_NONBLOCK, op_add_nothing,
tdb_unlockall, OP_TDB_UNLOCKALL, op_add_nothing,
tdb_lockall_read, OP_TDB_LOCKALL_READ, op_add_nothing,
tdb_lockall_read_nonblock, OP_TDB_LOCKALL_READ_NONBLOCK, op_add_nothing,
tdb_unlockall_read, OP_TDB_UNLOCKALL_READ, op_add_nothing,
tdb_chainlock, OP_TDB_CHAINLOCK, op_add_chainlock,
tdb_chainlock_nonblock, OP_TDB_CHAINLOCK_NONBLOCK, op_add_chainlock_ret,
tdb_chainlock_mark, OP_TDB_CHAINLOCK_MARK, op_add_key,
tdb_chainlock_unmark, OP_TDB_CHAINLOCK_UNMARK, op_add_key,
tdb_chainunlock, OP_TDB_CHAINUNLOCK, op_analyze_chainlock,
tdb_chainlock_read, OP_TDB_CHAINLOCK_READ, op_add_chainlock,
tdb_chainunlock_read, OP_TDB_CHAINUNLOCK_READ, op_analyze_chainlock,
tdb_parse_record, OP_TDB_PARSE_RECORD, op_add_key_ret,
tdb_exists, OP_TDB_EXISTS, op_add_key_ret,
tdb_store, OP_TDB_STORE, op_add_store,
tdb_append, OP_TDB_APPEND, op_add_append,
tdb_get_seqnum, OP_TDB_GET_SEQNUM, op_add_seqnum,
tdb_wipe_all, OP_TDB_WIPE_ALL, op_add_wipe_all,
tdb_transaction_start, OP_TDB_TRANSACTION_START, op_add_transaction,
tdb_transaction_cancel, OP_TDB_TRANSACTION_CANCEL, op_analyze_transaction,
tdb_transaction_commit, OP_TDB_TRANSACTION_COMMIT, op_analyze_transaction,
tdb_transaction_prepare_commit, OP_TDB_TRANSACTION_PREPARE_COMMIT, op_add_nothing,
tdb_traverse_read_start, OP_TDB_TRAVERSE_READ_START, op_add_traverse_start,
tdb_traverse_start, OP_TDB_TRAVERSE_START, op_add_traverse_start,
tdb_traverse_end, OP_TDB_TRAVERSE_END, op_analyze_traverse,
traverse, OP_TDB_TRAVERSE, op_add_traverse,
traversefn, OP_TDB_TRAVERSE, op_add_traversefn,
tdb_firstkey, OP_TDB_FIRSTKEY, op_add_key,
tdb_nextkey, OP_TDB_NEXTKEY, op_add_key_data,
tdb_fetch, OP_TDB_FETCH, op_add_key_data,
tdb_delete, OP_TDB_DELETE, op_add_key_ret,
tdb_repack, OP_TDB_REPACK, op_add_nothing,
ccan/tdb/tools/mktdb.c deleted 100644 → 0
View file @ 98b8ada2
#include <ccan/tdb/tdb.h>
#include <stdlib.h>
#include <stdio.h>
#include <fcntl.h>
#include <err.h>
int main(int argc, char *argv[])
{
unsigned int i, num_recs;
struct tdb_context *tdb;
if (argc != 3 || (num_recs = atoi(argv[2])) == 0)
errx(1, "Usage: mktdb <tdbfile> <numrecords>");
tdb = tdb_open(argv[1], 10007, TDB_DEFAULT, O_CREAT|O_TRUNC|O_RDWR, 0600);
if (!tdb)
err(1, "Opening %s", argv[1]);
for (i = 0; i < num_recs; i++) {
TDB_DATA d;
d.dptr = (void *)&i;
d.dsize = sizeof(i);
if (tdb_store(tdb, d, d, TDB_INSERT) != 0)
err(1, "Failed to store record %i", i);
}
printf("Done\n");
return 0;
}
ccan/tdb/tools/replay_trace.c deleted 100644 → 0
View file @ 98b8ada2
#include <ccan/tdb/tdb.h>
#include <ccan/grab_file/grab_file.h>
#include <ccan/hash/hash.h>
#include <ccan/talloc/talloc.h>
#include <ccan/str_talloc/str_talloc.h>
#include <ccan/str/str.h>
#include <ccan/list/list.h>
#include <err.h>
#include <ctype.h>
#include <string.h>
#include <unistd.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <sys/time.h>
#include <errno.h>
#include <signal.h>
#include <assert.h>
#include <fcntl.h>
#define STRINGIFY2(x) #x
#define STRINGIFY(x) STRINGIFY2(x)
static bool quiet = false;
/* Avoid mod by zero */
static unsigned int total_keys = 1;
/* All the wipe_all ops. */
static struct op_desc *wipe_alls = NULL;
static unsigned int num_wipe_alls = 0;
/* #define DEBUG_DEPS 1 */
/* Traversals block transactions in the current implementation. */
#define TRAVERSALS_TAKE_TRANSACTION_LOCK 1
struct pipe {
int fd[2];
};
static struct pipe *pipes;
static int backoff_fd = -1;
static void __attribute__((noreturn)) fail(const char *filename,
unsigned int line,
const char *fmt, ...)
{
va_list ap;
va_start(ap, fmt);
fprintf(stderr, "%s:%u: FAIL: ", filename, line);
vfprintf(stderr, fmt, ap);
fprintf(stderr, "\n");
va_end(ap);
exit(1);
}
/* Try or die. */
#define try(expr, expect) \
do { \
int ret = (expr); \
if (ret != (expect)) \
fail(filename[file], i+1, \
STRINGIFY(expr) "= %i", ret); \
} while (0)
/* Try or imitate results. */
#define unreliable(expr, expect, force, undo) \
do { \
int ret = expr; \
if (ret != expect) { \
fprintf(stderr, "%s:%u: %s gave %i not %i", \
filename[file], i+1, STRINGIFY(expr), \
ret, expect); \
if (expect == 0) \
force; \
else \
undo; \
} \
} while (0)
static bool key_eq(TDB_DATA a, TDB_DATA b)
{
if (a.dsize != b.dsize)
return false;
return memcmp(a.dptr, b.dptr, a.dsize) == 0;
}
/* This is based on the hash algorithm from gdbm */
static unsigned int hash_key(TDB_DATA *key)
{
uint32_t value; /* Used to compute the hash value. */
uint32_t i; /* Used to cycle through random values. */
/* Set the initial value from the key size. */
for (value = 0x238F13AF ^ key->dsize, i=0; i < key->dsize; i++)
value = (value + (key->dptr[i] << (i*5 % 24)));
return (1103515243 * value + 12345);
}
enum op_type {
OP_TDB_LOCKALL,
OP_TDB_LOCKALL_MARK,
OP_TDB_LOCKALL_UNMARK,
OP_TDB_LOCKALL_NONBLOCK,
OP_TDB_UNLOCKALL,
OP_TDB_LOCKALL_READ,
OP_TDB_LOCKALL_READ_NONBLOCK,
OP_TDB_UNLOCKALL_READ,
OP_TDB_CHAINLOCK,
OP_TDB_CHAINLOCK_NONBLOCK,
OP_TDB_CHAINLOCK_MARK,
OP_TDB_CHAINLOCK_UNMARK,
OP_TDB_CHAINUNLOCK,
OP_TDB_CHAINLOCK_READ,
OP_TDB_CHAINUNLOCK_READ,
OP_TDB_PARSE_RECORD,
OP_TDB_EXISTS,
OP_TDB_STORE,
OP_TDB_APPEND,
OP_TDB_GET_SEQNUM,
OP_TDB_WIPE_ALL,
OP_TDB_TRANSACTION_START,
OP_TDB_TRANSACTION_CANCEL,
OP_TDB_TRANSACTION_PREPARE_COMMIT,
OP_TDB_TRANSACTION_COMMIT,
OP_TDB_TRAVERSE_READ_START,
OP_TDB_TRAVERSE_START,
OP_TDB_TRAVERSE_END,
OP_TDB_TRAVERSE,
OP_TDB_TRAVERSE_END_EARLY,
OP_TDB_FIRSTKEY,
OP_TDB_NEXTKEY,
OP_TDB_FETCH,
OP_TDB_DELETE,
OP_TDB_REPACK,
};
struct op {
unsigned int seqnum;
enum op_type type;
TDB_DATA key;
TDB_DATA data;
int ret;
/* Who is waiting for us? */
struct list_head post;
/* What are we waiting for? */
struct list_head pre;
/* If I'm part of a group (traverse/transaction) where is
* start? (Otherwise, 0) */
unsigned int group_start;
union {
int flag; /* open and store */
struct { /* append */
TDB_DATA pre;
TDB_DATA post;
} append;
/* transaction/traverse start/chainlock */
unsigned int group_len;
};
};
struct op_desc {
unsigned int file;
unsigned int op_num;
};
static unsigned char hex_char(const char *filename, unsigned int line, char c)
{
c = toupper(c);
if (c >= 'A' && c <= 'F')
return c - 'A' + 10;
if (c >= '0' && c <= '9')
return c - '0';
fail(filename, line, "invalid hex character '%c'", c);
}
/* TDB data is <size>:<%02x>* */
static TDB_DATA make_tdb_data(const void *ctx,
const char *filename, unsigned int line,
const char *word)
{
TDB_DATA data;
unsigned int i;
const char *p;
if (streq(word, "NULL"))
return tdb_null;
data.dsize = atoi(word);
data.dptr = talloc_array(ctx, unsigned char, data.dsize);
p = strchr(word, ':');
if (!p)
fail(filename, line, "invalid tdb data '%s'", word);
p++;
for (i = 0; i < data.dsize; i++)
data.dptr[i] = hex_char(filename, line, p[i*2])*16
+ hex_char(filename, line, p[i*2+1]);
return data;
}
static void add_op(const char *filename, struct op **op, unsigned int i,
unsigned int seqnum, enum op_type type)
{
struct op *new;
*op = talloc_realloc(NULL, *op, struct op, i+1);
new = (*op) + i;
new->type = type;
new->seqnum = seqnum;
new->ret = 0;
new->group_start = 0;
}
static void op_add_nothing(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (words[2])
fail(filename[file], op_num+1, "Expected no arguments");
op[op_num].key = tdb_null;
}
static void op_add_key(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (words[2] == NULL || words[3])
fail(filename[file], op_num+1, "Expected just a key");
op[op_num].key = make_tdb_data(op, filename[file], op_num+1, words[2]);
total_keys++;
}
static void op_add_key_ret(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (!words[2] || !words[3] || !words[4] || words[5]
|| !streq(words[3], "="))
fail(filename[file], op_num+1, "Expected <key> = <ret>");
op[op_num].ret = atoi(words[4]);
op[op_num].key = make_tdb_data(op, filename[file], op_num+1, words[2]);
/* May only be a unique key if it fails */
if (op[op_num].ret != 0)
total_keys++;
}
static void op_add_key_data(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (!words[2] || !words[3] || !words[4] || words[5]
|| !streq(words[3], "="))
fail(filename[file], op_num+1, "Expected <key> = <data>");
op[op_num].key = make_tdb_data(op, filename[file], op_num+1, words[2]);
op[op_num].data = make_tdb_data(op, filename[file], op_num+1, words[4]);
/* Likely only be a unique key if it fails */
if (!op[op_num].data.dptr)
total_keys++;
else if (random() % 2)
total_keys++;
}
/* We don't record the keys or data for a traverse, as we don't use them. */
static void op_add_traverse(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (!words[2] || !words[3] || !words[4] || words[5]
|| !streq(words[3], "="))
fail(filename[file], op_num+1, "Expected <key> = <data>");
op[op_num].key = tdb_null;
}
/* Full traverse info is useful for debugging, but changing it to
* "traversefn" without the data makes the traces *much* smaller! */
static void op_add_traversefn(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (words[2])
fail(filename[file], op_num+1, "Expected no values");
op[op_num].key = tdb_null;
}
/* <seqnum> tdb_store <rec> <rec> <flag> = <ret> */
static void op_add_store(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (!words[2] || !words[3] || !words[4] || !words[5] || !words[6]
|| words[7] || !streq(words[5], "="))
fail(filename[file], op_num+1, "Expect <key> <data> <flag> = <ret>");
op[op_num].flag = strtoul(words[4], NULL, 0);
op[op_num].ret = atoi(words[6]);
op[op_num].key = make_tdb_data(op, filename[file], op_num+1, words[2]);
op[op_num].data = make_tdb_data(op, filename[file], op_num+1, words[3]);
total_keys++;
}
/* <seqnum> tdb_append <rec> <rec> = <rec> */
static void op_add_append(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (!words[2] || !words[3] || !words[4] || !words[5] || words[6]
|| !streq(words[4], "="))
fail(filename[file], op_num+1, "Expect <key> <data> = <rec>");
op[op_num].key = make_tdb_data(op, filename[file], op_num+1, words[2]);
op[op_num].data = make_tdb_data(op, filename[file], op_num+1, words[3]);
op[op_num].append.post
= make_tdb_data(op, filename[file], op_num+1, words[5]);
/* By subtraction, figure out what previous data was. */
op[op_num].append.pre.dptr = op[op_num].append.post.dptr;
op[op_num].append.pre.dsize
= op[op_num].append.post.dsize - op[op_num].data.dsize;
total_keys++;
}
/* <seqnum> tdb_get_seqnum = <ret> */
static void op_add_seqnum(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (!words[2] || !words[3] || words[4] || !streq(words[2], "="))
fail(filename[file], op_num+1, "Expect = <ret>");
op[op_num].key = tdb_null;
op[op_num].ret = atoi(words[3]);
}
static void op_add_traverse_start(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (words[2])
fail(filename[file], op_num+1, "Expect no arguments");
op[op_num].key = tdb_null;
op[op_num].group_len = 0;
}
static void op_add_transaction(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (words[2])
fail(filename[file], op_num+1, "Expect no arguments");
op[op_num].key = tdb_null;
op[op_num].group_len = 0;
}
static void op_add_chainlock(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (words[2] == NULL || words[3])
fail(filename[file], op_num+1, "Expected just a key");
/* A chainlock key isn't a key in the normal sense; it doesn't
* have to be in the db at all. Also, we don't want to hash this op. */
op[op_num].data = make_tdb_data(op, filename[file], op_num+1, words[2]);
op[op_num].key = tdb_null;
op[op_num].group_len = 0;
}
static void op_add_chainlock_ret(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (!words[2] || !words[3] || !words[4] || words[5]
|| !streq(words[3], "="))
fail(filename[file], op_num+1, "Expected <key> = <ret>");
op[op_num].ret = atoi(words[4]);
op[op_num].data = make_tdb_data(op, filename[file], op_num+1, words[2]);
op[op_num].key = tdb_null;
op[op_num].group_len = 0;
total_keys++;
}
static void op_add_wipe_all(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
if (words[2])
fail(filename[file], op_num+1, "Expected no arguments");
op[op_num].key = tdb_null;
wipe_alls = talloc_realloc(NULL, wipe_alls, struct op_desc,
num_wipe_alls+1);
wipe_alls[num_wipe_alls].file = file;
wipe_alls[num_wipe_alls].op_num = op_num;
num_wipe_alls++;
}
static int op_find_start(struct op op[], unsigned int op_num, enum op_type type)
{
unsigned int i;
for (i = op_num-1; i > 0; i--) {
if (op[i].type == type && !op[i].group_len)
return i;
}
return 0;
}
static void op_analyze_transaction(char *filename[], struct op op[],
unsigned file, unsigned op_num,
char *words[])
{
unsigned int start, i;
op[op_num].key = tdb_null;
if (words[2])
fail(filename[file], op_num+1, "Expect no arguments");
start = op_find_start(op, op_num, OP_TDB_TRANSACTION_START);
if (!start)
fail(filename[file], op_num+1, "no transaction start found");
op[start].group_len = op_num - start;
/* This rolls in nested transactions. I think that's right. */
for (i = start; i <= op_num; i++)
op[i].group_start = start;
}
/* We treat chainlocks a lot like transactions, even though that's overkill */
static void op_analyze_chainlock(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
unsigned int i, start;
if (words[2] == NULL || words[3])
fail(filename[file], op_num+1, "Expected just a key");
op[op_num].data = make_tdb_data(op, filename[file], op_num+1, words[2]);
op[op_num].key = tdb_null;
total_keys++;
start = op_find_start(op, op_num, OP_TDB_CHAINLOCK);
if (!start)
start = op_find_start(op, op_num, OP_TDB_CHAINLOCK_READ);
if (!start)
fail(filename[file], op_num+1, "no initial chainlock found");
/* FIXME: We'd have to do something clever to make this work
* vs. deadlock. */
if (!key_eq(op[start].data, op[op_num].data))
fail(filename[file], op_num+1, "nested chainlock calls?");
op[start].group_len = op_num - start;
for (i = start; i <= op_num; i++)
op[i].group_start = start;
}
static void op_analyze_traverse(char *filename[], struct op op[],
unsigned file, unsigned op_num, char *words[])
{
int i, start;
op[op_num].key = tdb_null;
/* = %u means traverse function terminated. */
if (words[2]) {
if (!streq(words[2], "=") || !words[3] || words[4])
fail(filename[file], op_num+1, "expect = <num>");
op[op_num].ret = atoi(words[3]);
} else
op[op_num].ret = 0;
start = op_find_start(op, op_num, OP_TDB_TRAVERSE_START);
if (!start)
start = op_find_start(op, op_num, OP_TDB_TRAVERSE_READ_START);
if (!start)
fail(filename[file], op_num+1, "no traversal start found");
op[start].group_len = op_num - start;
/* Don't roll in nested traverse/chainlock */
for (i = start; i <= op_num; i++)
if (!op[i].group_start)
op[i].group_start = start;
}
/* Keep -Wmissing-declarations happy: */
const struct op_table *
find_keyword (register const char *str, register unsigned int len);
#include "keywords.c"
struct depend {
/* We can have more than one */
struct list_node pre_list;
struct list_node post_list;
struct op_desc needs;
struct op_desc prereq;
};
static void check_deps(const char *filename, struct op op[], unsigned int num)
{
#ifdef DEBUG_DEPS
unsigned int i;
for (i = 1; i < num; i++)
if (!list_empty(&op[i].pre))
fail(filename, i+1, "Still has dependencies");
#endif
}
static void dump_pre(char *filename[], struct op *op[],
unsigned int file, unsigned int i)
{
struct depend *dep;
if (!quiet) {
printf("%s:%u (%u) still waiting for:\n", filename[file], i+1,
op[file][i].seqnum);
list_for_each(&op[file][i].pre, dep, pre_list)
printf(" %s:%u (%u)\n",
filename[dep->prereq.file], dep->prereq.op_num+1,
op[dep->prereq.file][dep->prereq.op_num].seqnum);
}
check_deps(filename[file], op[file], i);
}
/* We simply read/write pointers, since we all are children. */
static bool do_pre(struct tdb_context *tdb,
char *filename[], struct op *op[],
unsigned int file, int pre_fd, unsigned int i,
bool backoff)
{
while (!list_empty(&op[file][i].pre)) {
struct depend *dep;
#if DEBUG_DEPS
printf("%s:%u:waiting for pre\n", filename[file], i+1);
fflush(stdout);
#endif
if (backoff)
alarm(2);
else
alarm(10);
while (read(pre_fd, &dep, sizeof(dep)) != sizeof(dep)) {
if (errno == EINTR) {
if (backoff) {
struct op_desc desc = { file,i };
warnx("%s:%u:avoiding deadlock",
filename[file], i+1);
if (write(backoff_fd, &desc,
sizeof(desc)) != sizeof(desc))
err(1, "writing backoff_fd");
return false;
}
dump_pre(filename, op, file, i);
exit(1);
} else
errx(1, "Reading from pipe");
}
alarm(0);
#if DEBUG_DEPS
printf("%s:%u:got pre %u from %s:%u\n", filename[file], i+1,
dep->needs.op_num+1, filename[dep->prereq.file],
dep->prereq.op_num+1);
fflush(stdout);
#endif
/* This could be any op, not just this one. */
talloc_free(dep);
}
return true;
}
static void do_post(char *filename[], struct op *op[],
unsigned int file, unsigned int i)
{
struct depend *dep;
list_for_each(&op[file][i].post, dep, post_list) {
#if DEBUG_DEPS
printf("%s:%u:sending to file %s:%u\n", filename[file], i+1,
filename[dep->needs.file], dep->needs.op_num+1);
#endif
if (write(pipes[dep->needs.file].fd[1], &dep, sizeof(dep))
!= sizeof(dep))
err(1, "%s:%u failed to tell file %s",
filename[file], i+1, filename[dep->needs.file]);
}
}
static int get_len(TDB_DATA key, TDB_DATA data, void *private_data)
{
return data.dsize;
}
static unsigned run_ops(struct tdb_context *tdb,
int pre_fd,
char *filename[],
struct op *op[],
unsigned int file,
unsigned int start, unsigned int stop,
bool backoff);
struct traverse_info {
struct op **op;
char **filename;
unsigned file;
int pre_fd;
unsigned int start;
unsigned int i;
};
/* More complex. Just do whatever's they did at the n'th entry. */
static int nontrivial_traverse(struct tdb_context *tdb,
TDB_DATA key, TDB_DATA data,
void *_tinfo)
{
struct traverse_info *tinfo = _tinfo;
unsigned int trav_len = tinfo->op[tinfo->file][tinfo->start].group_len;
bool avoid_deadlock = false;
if (tinfo->i == tinfo->start + trav_len) {
/* This can happen if traverse expects to be empty. */
if (trav_len == 1)
return 1;
fail(tinfo->filename[tinfo->file], tinfo->start + 1,
"traverse did not terminate");
}
if (tinfo->op[tinfo->file][tinfo->i].type != OP_TDB_TRAVERSE)
fail(tinfo->filename[tinfo->file], tinfo->start + 1,
"%s:%u:traverse terminated early");
#if TRAVERSALS_TAKE_TRANSACTION_LOCK
avoid_deadlock = true;
#endif
/* Run any normal ops. */
tinfo->i = run_ops(tdb, tinfo->pre_fd, tinfo->filename, tinfo->op,
tinfo->file, tinfo->i+1, tinfo->start + trav_len,
avoid_deadlock);
/* We backed off, or we hit OP_TDB_TRAVERSE_END/EARLY. */
if (tinfo->op[tinfo->file][tinfo->i].type != OP_TDB_TRAVERSE)
return 1;
return 0;
}
static unsigned op_traverse(struct tdb_context *tdb,
int pre_fd,
char *filename[],
unsigned int file,
int (*traversefn)(struct tdb_context *,
tdb_traverse_func, void *),
struct op *op[],
unsigned int start)
{
struct traverse_info tinfo = { op, filename, file, pre_fd,
start, start+1 };
traversefn(tdb, nontrivial_traverse, &tinfo);
/* Traversing in wrong order can have strange effects: eg. if
* original traverse went A (delete A), B, we might do B
* (delete A). So if we have ops left over, we do it now. */
while (tinfo.i != start + op[file][start].group_len) {
if (op[file][tinfo.i].type == OP_TDB_TRAVERSE
|| op[file][tinfo.i].type == OP_TDB_TRAVERSE_END_EARLY)
tinfo.i++;
else
tinfo.i = run_ops(tdb, pre_fd, filename, op, file,
tinfo.i,
start + op[file][start].group_len,
false);
}
return tinfo.i;
}
static void break_out(int sig)
{
}
static __attribute__((noinline))
unsigned run_ops(struct tdb_context *tdb,
int pre_fd,
char *filename[],
struct op *op[],
unsigned int file,
unsigned int start, unsigned int stop,
bool backoff)
{
unsigned int i;
struct sigaction sa;
sa.sa_handler = break_out;
sa.sa_flags = 0;
sigaction(SIGALRM, &sa, NULL);
for (i = start; i < stop; i++) {
if (!do_pre(tdb, filename, op, file, pre_fd, i, backoff))
return i;
switch (op[file][i].type) {
case OP_TDB_LOCKALL:
try(tdb_lockall(tdb), op[file][i].ret);
break;
case OP_TDB_LOCKALL_MARK:
try(tdb_lockall_mark(tdb), op[file][i].ret);
break;
case OP_TDB_LOCKALL_UNMARK:
try(tdb_lockall_unmark(tdb), op[file][i].ret);
break;
case OP_TDB_LOCKALL_NONBLOCK:
unreliable(tdb_lockall_nonblock(tdb), op[file][i].ret,
tdb_lockall(tdb), tdb_unlockall(tdb));
break;
case OP_TDB_UNLOCKALL:
try(tdb_unlockall(tdb), op[file][i].ret);
break;
case OP_TDB_LOCKALL_READ:
try(tdb_lockall_read(tdb), op[file][i].ret);
break;
case OP_TDB_LOCKALL_READ_NONBLOCK:
unreliable(tdb_lockall_read_nonblock(tdb),
op[file][i].ret,
tdb_lockall_read(tdb),
tdb_unlockall_read(tdb));
break;
case OP_TDB_UNLOCKALL_READ:
try(tdb_unlockall_read(tdb), op[file][i].ret);
break;
case OP_TDB_CHAINLOCK:
try(tdb_chainlock(tdb, op[file][i].key),
op[file][i].ret);
break;
case OP_TDB_CHAINLOCK_NONBLOCK:
unreliable(tdb_chainlock_nonblock(tdb, op[file][i].key),
op[file][i].ret,
tdb_chainlock(tdb, op[file][i].key),
tdb_chainunlock(tdb, op[file][i].key));
break;
case OP_TDB_CHAINLOCK_MARK:
try(tdb_chainlock_mark(tdb, op[file][i].key),
op[file][i].ret);
break;
case OP_TDB_CHAINLOCK_UNMARK:
try(tdb_chainlock_unmark(tdb, op[file][i].key),
op[file][i].ret);
break;
case OP_TDB_CHAINUNLOCK:
try(tdb_chainunlock(tdb, op[file][i].key),
op[file][i].ret);
break;
case OP_TDB_CHAINLOCK_READ:
try(tdb_chainlock_read(tdb, op[file][i].key),
op[file][i].ret);
break;
case OP_TDB_CHAINUNLOCK_READ:
try(tdb_chainunlock_read(tdb, op[file][i].key),
op[file][i].ret);
break;
case OP_TDB_PARSE_RECORD:
try(tdb_parse_record(tdb, op[file][i].key, get_len,
NULL),
op[file][i].ret);
break;
case OP_TDB_EXISTS:
try(tdb_exists(tdb, op[file][i].key), op[file][i].ret);
break;
case OP_TDB_STORE:
try(tdb_store(tdb, op[file][i].key, op[file][i].data,
op[file][i].flag),
op[file][i].ret);
break;
case OP_TDB_APPEND:
try(tdb_append(tdb, op[file][i].key, op[file][i].data),
op[file][i].ret);
break;
case OP_TDB_GET_SEQNUM:
try(tdb_get_seqnum(tdb), op[file][i].ret);
break;
case OP_TDB_WIPE_ALL:
try(tdb_wipe_all(tdb), op[file][i].ret);
break;
case OP_TDB_TRANSACTION_START:
try(tdb_transaction_start(tdb), op[file][i].ret);
break;
case OP_TDB_TRANSACTION_CANCEL:
try(tdb_transaction_cancel(tdb), op[file][i].ret);
break;
case OP_TDB_TRANSACTION_PREPARE_COMMIT:
try(tdb_transaction_prepare_commit(tdb),
op[file][i].ret);
break;
case OP_TDB_TRANSACTION_COMMIT:
try(tdb_transaction_commit(tdb), op[file][i].ret);
break;
case OP_TDB_TRAVERSE_READ_START:
i = op_traverse(tdb, pre_fd, filename, file,
tdb_traverse_read, op, i);
break;
case OP_TDB_TRAVERSE_START:
i = op_traverse(tdb, pre_fd, filename, file,
tdb_traverse, op, i);
break;
case OP_TDB_TRAVERSE:
case OP_TDB_TRAVERSE_END_EARLY:
/* Terminate: we're in a traverse, and we've
* done our ops. */
return i;
case OP_TDB_TRAVERSE_END:
fail(filename[file], i+1, "unexpected end traverse");
/* FIXME: These must be treated like traverse. */
case OP_TDB_FIRSTKEY:
if (!key_eq(tdb_firstkey(tdb), op[file][i].data))
fail(filename[file], i+1, "bad firstkey");
break;
case OP_TDB_NEXTKEY:
if (!key_eq(tdb_nextkey(tdb, op[file][i].key),
op[file][i].data))
fail(filename[file], i+1, "bad nextkey");
break;
case OP_TDB_FETCH: {
TDB_DATA f = tdb_fetch(tdb, op[file][i].key);
if (!key_eq(f, op[file][i].data))
fail(filename[file], i+1, "bad fetch %u",
f.dsize);
break;
}
case OP_TDB_DELETE:
try(tdb_delete(tdb, op[file][i].key), op[file][i].ret);
break;
case OP_TDB_REPACK:
/* We do nothing here: the transaction and traverse are
* traced. It's in the trace to mark it, since it
* may become unnecessary in future. */
break;
}
do_post(filename, op, file, i);
}
return i;
}
/* tdbtorture, in particular, can do a tdb_close with a transaction in
* progress. */
static struct op *maybe_cancel_transaction(char *filename[], unsigned int file,
struct op *op, unsigned int *num)
{
unsigned int start = op_find_start(op, *num, OP_TDB_TRANSACTION_START);
if (start) {
char *words[] = { "<unknown>", "tdb_close", NULL };
add_op(filename[file], &op, *num, op[start].seqnum,
OP_TDB_TRANSACTION_CANCEL);
op_analyze_transaction(filename, op, file, *num, words);
(*num)++;
}
return op;
}
static struct op *load_tracefile(char *filename[],
unsigned int file,
unsigned int *num,
unsigned int *hashsize,
unsigned int *tdb_flags,
unsigned int *open_flags)
{
unsigned int i;
struct op *op = talloc_array(NULL, struct op, 1);
char **words;
char **lines;
char *contents;
contents = grab_file(NULL, filename[file], NULL);
if (!contents)
err(1, "Reading %s", filename[file]);
lines = strsplit(contents, contents, "\n");
if (!lines[0])
errx(1, "%s is empty", filename[file]);
words = strsplit(lines, lines[0], " ");
if (!streq(words[1], "tdb_open"))
fail(filename[file], 1, "does not start with tdb_open");
*hashsize = atoi(words[2]);
*tdb_flags = strtoul(words[3], NULL, 0);
*open_flags = strtoul(words[4], NULL, 0);
for (i = 1; lines[i]; i++) {
const struct op_table *opt;
words = strsplit(lines, lines[i], " ");
if (!words[0] || !words[1])
fail(filename[file], i+1,
"Expected seqnum number and op");
opt = find_keyword(words[1], strlen(words[1]));
if (!opt) {
if (streq(words[1], "tdb_close")) {
if (lines[i+1])
fail(filename[file], i+2,
"lines after tdb_close");
*num = i;
talloc_free(lines);
return maybe_cancel_transaction(filename, file,
op, num);
}
fail(filename[file], i+1,
"Unknown operation '%s'", words[1]);
}
add_op(filename[file], &op, i, atoi(words[0]), opt->type);
opt->enhance_op(filename, op, file, i, words);
}
if (!quiet)
fprintf(stderr,
"%s:%u:last operation is not tdb_close: incomplete?",
filename[file], i);
talloc_free(contents);
*num = i - 1;
return maybe_cancel_transaction(filename, file, op, num);
}
/* We remember all the keys we've ever seen, and who has them. */
struct keyinfo {
TDB_DATA key;
unsigned int num_users;
struct op_desc *user;
};
static bool starts_transaction(const struct op *op)
{
return op->type == OP_TDB_TRANSACTION_START;
}
static bool in_transaction(const struct op op[], unsigned int i)
{
return op[i].group_start && starts_transaction(&op[op[i].group_start]);
}
static bool successful_transaction(const struct op *op)
{
return starts_transaction(op)
&& op[op->group_len].type == OP_TDB_TRANSACTION_COMMIT;
}
static bool starts_traverse(const struct op *op)
{
return op->type == OP_TDB_TRAVERSE_START
|| op->type == OP_TDB_TRAVERSE_READ_START;
}
static bool in_traverse(const struct op op[], unsigned int i)
{
return op[i].group_start && starts_traverse(&op[op[i].group_start]);
}
static bool starts_chainlock(const struct op *op)
{
return op->type == OP_TDB_CHAINLOCK_READ
|| op->type == OP_TDB_CHAINLOCK;
}
static bool in_chainlock(const struct op op[], unsigned int i)
{
return op[i].group_start && starts_chainlock(&op[op[i].group_start]);
}
static const TDB_DATA must_not_exist;
static const TDB_DATA must_exist;
static const TDB_DATA not_exists_or_empty;
/* NULL means doesn't care if it exists or not, &must_exist means
* it must exist but we don't care what, &must_not_exist means it must
* not exist, otherwise the data it needs. */
static const TDB_DATA *needs(const TDB_DATA *key, const struct op *op)
{
/* Look through for an op in this transaction which needs this key. */
if (starts_transaction(op) || starts_chainlock(op)) {
unsigned int i;
const TDB_DATA *need = NULL;
for (i = 1; i < op->group_len; i++) {
if (key_eq(op[i].key, *key)
|| op[i].type == OP_TDB_WIPE_ALL) {
need = needs(key, &op[i]);
/* tdb_exists() is special: there might be
* something in the transaction with more
* specific requirements. Other ops don't have
* specific requirements (eg. store or delete),
* but they change the value so we can't get
* more information from future ops. */
if (op[i].type != OP_TDB_EXISTS)
break;
}
}
return need;
}
switch (op->type) {
/* FIXME: Pull forward deps, since we can deadlock */
case OP_TDB_CHAINLOCK:
case OP_TDB_CHAINLOCK_NONBLOCK:
case OP_TDB_CHAINLOCK_MARK:
case OP_TDB_CHAINLOCK_UNMARK:
case OP_TDB_CHAINUNLOCK:
case OP_TDB_CHAINLOCK_READ:
case OP_TDB_CHAINUNLOCK_READ:
return NULL;
case OP_TDB_APPEND:
if (op->append.pre.dsize == 0)
return &not_exists_or_empty;
return &op->append.pre;
case OP_TDB_STORE:
if (op->flag == TDB_INSERT) {
if (op->ret < 0)
return &must_exist;
else
return &must_not_exist;
} else if (op->flag == TDB_MODIFY) {
if (op->ret < 0)
return &must_not_exist;
else
return &must_exist;
}
/* No flags? Don't care */
return NULL;
case OP_TDB_EXISTS:
if (op->ret == 1)
return &must_exist;
else
return &must_not_exist;
case OP_TDB_PARSE_RECORD:
if (op->ret < 0)
return &must_not_exist;
return &must_exist;
/* FIXME: handle these. */
case OP_TDB_WIPE_ALL:
case OP_TDB_FIRSTKEY:
case OP_TDB_NEXTKEY:
case OP_TDB_GET_SEQNUM:
case OP_TDB_TRAVERSE:
case OP_TDB_TRANSACTION_COMMIT:
case OP_TDB_TRANSACTION_CANCEL:
case OP_TDB_TRANSACTION_START:
return NULL;
case OP_TDB_FETCH:
if (!op->data.dptr)
return &must_not_exist;
return &op->data;
case OP_TDB_DELETE:
if (op->ret < 0)
return &must_not_exist;
return &must_exist;
default:
errx(1, "Unexpected op type %i", op->type);
}
}
/* What's the data after this op? pre if nothing changed. */
static const TDB_DATA *gives(const TDB_DATA *key, const TDB_DATA *pre,
const struct op *op)
{
if (starts_transaction(op) || starts_chainlock(op)) {
unsigned int i;
/* Cancelled transactions don't change anything. */
if (op[op->group_len].type == OP_TDB_TRANSACTION_CANCEL)
return pre;
assert(op[op->group_len].type == OP_TDB_TRANSACTION_COMMIT
|| op[op->group_len].type == OP_TDB_CHAINUNLOCK_READ
|| op[op->group_len].type == OP_TDB_CHAINUNLOCK);
for (i = 1; i < op->group_len; i++) {
/* This skips nested transactions, too */
if (key_eq(op[i].key, *key)
|| op[i].type == OP_TDB_WIPE_ALL)
pre = gives(key, pre, &op[i]);
}
return pre;
}
/* Failed ops don't change state of db. */
if (op->ret < 0)
return pre;
if (op->type == OP_TDB_DELETE || op->type == OP_TDB_WIPE_ALL)
return &tdb_null;
if (op->type == OP_TDB_APPEND)
return &op->append.post;
if (op->type == OP_TDB_STORE)
return &op->data;
return pre;
}
static void add_hash_user(struct keyinfo *hash,
unsigned int h,
struct op *op[],
unsigned int file,
unsigned int op_num)
{
hash[h].user = talloc_realloc(hash, hash[h].user,
struct op_desc, hash[h].num_users+1);
/* If it's in a transaction, it's the transaction which
* matters from an analysis POV. */
if (in_transaction(op[file], op_num)
|| in_chainlock(op[file], op_num)) {
unsigned i;
op_num = op[file][op_num].group_start;
/* Don't include twice. */
for (i = 0; i < hash[h].num_users; i++) {
if (hash[h].user[i].file == file
&& hash[h].user[i].op_num == op_num)
return;
}
}
hash[h].user[hash[h].num_users].op_num = op_num;
hash[h].user[hash[h].num_users].file = file;
hash[h].num_users++;
}
static struct keyinfo *hash_ops(struct op *op[], unsigned int num_ops[],
unsigned int num)
{
unsigned int i, j, h;
struct keyinfo *hash;
hash = talloc_zero_array(op[0], struct keyinfo, total_keys*2);
for (i = 0; i < num; i++) {
for (j = 1; j < num_ops[i]; j++) {
/* We can't do this on allocation, due to realloc. */
list_head_init(&op[i][j].post);
list_head_init(&op[i][j].pre);
if (!op[i][j].key.dptr)
continue;
h = hash_key(&op[i][j].key) % (total_keys * 2);
while (!key_eq(hash[h].key, op[i][j].key)) {
if (!hash[h].key.dptr) {
hash[h].key = op[i][j].key;
break;
}
h = (h + 1) % (total_keys * 2);
}
/* Might as well save some memory if we can. */
if (op[i][j].key.dptr != hash[h].key.dptr) {
talloc_free(op[i][j].key.dptr);
op[i][j].key.dptr = hash[h].key.dptr;
}
add_hash_user(hash, h, op, i, j);
}
}
/* Any wipe all entries need adding to all hash entries. */
for (h = 0; h < total_keys*2; h++) {
if (!hash[h].num_users)
continue;
for (i = 0; i < num_wipe_alls; i++)
add_hash_user(hash, h, op,
wipe_alls[i].file, wipe_alls[i].op_num);
}
return hash;
}
static bool satisfies(const TDB_DATA *key, const TDB_DATA *data,
const struct op *op)
{
const TDB_DATA *need = needs(key, op);
/* Don't need anything? Cool. */
if (!need)
return true;
/* This should be tdb_null or a real value. */
assert(data != &must_exist);
assert(data != &must_not_exist);
assert(data != &not_exists_or_empty);
/* Must not exist? data must not exist. */
if (need == &must_not_exist)
return data == &tdb_null;
/* Must exist? */
if (need == &must_exist)
return data != &tdb_null;
/* Either noexist or empty. */
if (need == &not_exists_or_empty)
return data->dsize == 0;
/* Needs something specific. */
return key_eq(*data, *need);
}
static void move_to_front(struct op_desc res[], unsigned off, unsigned elem)
{
if (elem != off) {
struct op_desc tmp = res[elem];
memmove(res + off + 1, res + off, (elem - off)*sizeof(res[0]));
res[off] = tmp;
}
}
static void restore_to_pos(struct op_desc res[], unsigned off, unsigned elem)
{
if (elem != off) {
struct op_desc tmp = res[off];
memmove(res + off, res + off + 1, (elem - off)*sizeof(res[0]));
res[elem] = tmp;
}
}
static bool sort_deps(char *filename[], struct op *op[],
struct op_desc res[],
unsigned off, unsigned num,
const TDB_DATA *key, const TDB_DATA *data,
unsigned num_files, unsigned fuzz)
{
unsigned int i, files_done;
struct op *this_op;
bool done[num_files];
/* None left? We're sorted. */
if (off == num)
return true;
/* Does this make sequence number go backwards? Allow a little fuzz. */
if (off > 0) {
int seqnum1 = op[res[off-1].file][res[off-1].op_num].seqnum;
int seqnum2 = op[res[off].file][res[off].op_num].seqnum;
if (seqnum1 - seqnum2 > (int)fuzz) {
#if DEBUG_DEPS
printf("Seqnum jump too far (%u -> %u)\n",
seqnum1, seqnum2);
#endif
return false;
}
}
memset(done, 0, sizeof(done));
/* Since ops within a trace file are ordered, we just need to figure
* out which file to try next. Since we don't take into account
* inter-key relationships (which exist by virtue of trace file order),
* we minimize the chance of harm by trying to keep in seqnum order. */
for (files_done = 0, i = off; i < num && files_done < num_files; i++) {
if (done[res[i].file])
continue;
this_op = &op[res[i].file][res[i].op_num];
/* Is what we have good enough for this op? */
if (satisfies(key, data, this_op)) {
move_to_front(res, off, i);
if (sort_deps(filename, op, res, off+1, num,
key, gives(key, data, this_op),
num_files, fuzz))
return true;
restore_to_pos(res, off, i);
}
done[res[i].file] = true;
files_done++;
}
/* No combination worked. */
return false;
}
static void check_dep_sorting(struct op_desc user[], unsigned num_users,
unsigned num_files)
{
#if DEBUG_DEPS
unsigned int i;
unsigned minima[num_files];
memset(minima, 0, sizeof(minima));
for (i = 0; i < num_users; i++) {
assert(minima[user[i].file] < user[i].op_num);
minima[user[i].file] = user[i].op_num;
}
#endif
}
/* All these ops happen on the same key. Which comes first?
*
* This can happen both because read ops or failed write ops don't
* change sequence number, and also due to race since we access the
* number unlocked (the race can cause less detectable ordering problems,
* in which case we'll deadlock and report: fix manually in that case).
*/
static bool figure_deps(char *filename[], struct op *op[],
const TDB_DATA *key, const TDB_DATA *data,
struct op_desc user[],
unsigned num_users, unsigned num_files)
{
unsigned int fuzz;
/* We prefer to keep strict seqnum order if possible: it's the
* most likely. We get more lax if that fails. */
for (fuzz = 0; fuzz < 100; fuzz = (fuzz + 1)*2) {
if (sort_deps(filename, op, user, 0, num_users, key, data,
num_files, fuzz))
break;
}
if (fuzz >= 100)
return false;
check_dep_sorting(user, num_users, num_files);
return true;
}
/* We're having trouble sorting out dependencies for this key. Assume that it's
* a pre-existing record in the db, so determine a likely value. */
static const TDB_DATA *preexisting_data(char *filename[], struct op *op[],
const TDB_DATA *key,
struct op_desc *user,
unsigned int num_users)
{
unsigned int i;
const TDB_DATA *data;
for (i = 0; i < num_users; i++) {
data = needs(key, &op[user->file][user->op_num]);
if (data && data != &must_not_exist) {
if (!quiet)
printf("%s:%u: needs pre-existing record\n",
filename[user->file], user->op_num+1);
return data;
}
}
return &tdb_null;
}
static void sort_ops(struct tdb_context *tdb,
struct keyinfo hash[], char *filename[], struct op *op[],
unsigned int num)
{
unsigned int h;
/* Gcc nexted function extension. How cool is this? */
int compare_seqnum(const void *_a, const void *_b)
{
const struct op_desc *a = _a, *b = _b;
/* First, maintain order within any trace file. */
if (a->file == b->file)
return a->op_num - b->op_num;
/* Otherwise, arrange by seqnum order. */
if (op[a->file][a->op_num].seqnum !=
op[b->file][b->op_num].seqnum)
return op[a->file][a->op_num].seqnum
- op[b->file][b->op_num].seqnum;
/* Cancelled transactions are assumed to happen first. */
if (starts_transaction(&op[a->file][a->op_num])
&& !successful_transaction(&op[a->file][a->op_num]))
return -1;
if (starts_transaction(&op[b->file][b->op_num])
&& !successful_transaction(&op[b->file][b->op_num]))
return 1;
/* No idea. */
return 0;
}
/* Now sort into seqnum order. */
for (h = 0; h < total_keys * 2; h++) {
struct op_desc *user = hash[h].user;
qsort(user, hash[h].num_users, sizeof(user[0]), compare_seqnum);
if (!figure_deps(filename, op, &hash[h].key, &tdb_null, user,
hash[h].num_users, num)) {
const TDB_DATA *data;
data = preexisting_data(filename, op, &hash[h].key,
user, hash[h].num_users);
/* Give the first op what it wants: does that help? */
if (!figure_deps(filename, op, &hash[h].key, data, user,
hash[h].num_users, num))
fail(filename[user[0].file], user[0].op_num+1,
"Could not resolve inter-dependencies");
if (tdb_store(tdb, hash[h].key, *data, TDB_INSERT) != 0)
errx(1, "Could not store initial value");
}
}
}
static int destroy_depend(struct depend *dep)
{
list_del(&dep->pre_list);
list_del(&dep->post_list);
return 0;
}
static void add_dependency(void *ctx,
struct op *op[],
char *filename[],
const struct op_desc *needs,
const struct op_desc *prereq)
{
struct depend *dep;
/* We don't depend on ourselves. */
if (needs->file == prereq->file) {
assert(prereq->op_num < needs->op_num);
return;
}
#if DEBUG_DEPS
printf("%s:%u: depends on %s:%u\n",
filename[needs->file], needs->op_num+1,
filename[prereq->file], prereq->op_num+1);
#endif
dep = talloc(ctx, struct depend);
dep->needs = *needs;
dep->prereq = *prereq;
#if TRAVERSALS_TAKE_TRANSACTION_LOCK
/* If something in a traverse depends on something in another
* traverse/transaction, it creates a dependency between the
* two groups. */
if ((in_traverse(op[prereq->file], prereq->op_num)
&& (starts_transaction(&op[needs->file][needs->op_num])
|| starts_traverse(&op[needs->file][needs->op_num])))
|| (in_traverse(op[needs->file], needs->op_num)
&& (starts_transaction(&op[prereq->file][prereq->op_num])
|| starts_traverse(&op[prereq->file][prereq->op_num])))) {
unsigned int start;
/* We are satisfied by end of group. */
start = op[prereq->file][prereq->op_num].group_start;
dep->prereq.op_num = start + op[prereq->file][start].group_len;
/* And we need that done by start of our group. */
dep->needs.op_num = op[needs->file][needs->op_num].group_start;
}
/* There is also this case:
* <traverse> <read foo> ...
* <transaction> ... </transaction> <create foo>
* Where if we start the traverse then wait, we could block
* the transaction and deadlock.
*
* We try to address this by ensuring that where seqnum indicates it's
* possible, we wait for <create foo> before *starting* traverse.
*/
else if (in_traverse(op[needs->file], needs->op_num)) {
struct op *need = &op[needs->file][needs->op_num];
if (op[needs->file][need->group_start].seqnum >
op[prereq->file][prereq->op_num].seqnum) {
dep->needs.op_num = need->group_start;
}
}
#endif
/* If you depend on a transaction or chainlock, you actually
* depend on it ending. */
if (starts_transaction(&op[prereq->file][dep->prereq.op_num])
|| starts_chainlock(&op[prereq->file][dep->prereq.op_num])) {
dep->prereq.op_num
+= op[dep->prereq.file][dep->prereq.op_num].group_len;
#if DEBUG_DEPS
printf("-> Actually end of transaction %s:%u\n",
filename[dep->prereq->file], dep->prereq->op_num+1);
#endif
} else
/* We should never create a dependency from middle of
* a transaction. */
assert(!in_transaction(op[prereq->file], dep->prereq.op_num)
|| op[prereq->file][dep->prereq.op_num].type
== OP_TDB_TRANSACTION_COMMIT
|| op[prereq->file][dep->prereq.op_num].type
== OP_TDB_TRANSACTION_CANCEL);
list_add(&op[dep->prereq.file][dep->prereq.op_num].post,
&dep->post_list);
list_add(&op[dep->needs.file][dep->needs.op_num].pre,
&dep->pre_list);
talloc_set_destructor(dep, destroy_depend);
}
static bool changes_db(const TDB_DATA *key, const struct op *op)
{
return gives(key, NULL, op) != NULL;
}
static void depend_on_previous(struct op *op[],
char *filename[],
unsigned int num,
struct op_desc user[],
unsigned int i,
int prev)
{
bool deps[num];
int j;
if (i == 0)
return;
if (prev == i - 1) {
/* Just depend on previous. */
add_dependency(NULL, op, filename, &user[i], &user[prev]);
return;
}
/* We have to wait for the readers. Find last one in *each* file. */
memset(deps, 0, sizeof(deps));
deps[user[i].file] = true;
for (j = i - 1; j > prev; j--) {
if (!deps[user[j].file]) {
add_dependency(NULL, op, filename, &user[i], &user[j]);
deps[user[j].file] = true;
}
}
}
/* This is simple, but not complete. We don't take into account
* indirect dependencies. */
static void optimize_dependencies(struct op *op[], unsigned int num_ops[],
unsigned int num)
{
unsigned int i, j;
/* There can only be one real dependency on each file */
for (i = 0; i < num; i++) {
for (j = 1; j < num_ops[i]; j++) {
struct depend *dep, *next;
struct depend *prev[num];
memset(prev, 0, sizeof(prev));
list_for_each_safe(&op[i][j].pre, dep, next, pre_list) {
if (!prev[dep->prereq.file]) {
prev[dep->prereq.file] = dep;
continue;
}
if (prev[dep->prereq.file]->prereq.op_num
< dep->prereq.op_num) {
talloc_free(prev[dep->prereq.file]);
prev[dep->prereq.file] = dep;
} else
talloc_free(dep);
}
}
}
for (i = 0; i < num; i++) {
int deps[num];
for (j = 0; j < num; j++)
deps[j] = -1;
for (j = 1; j < num_ops[i]; j++) {
struct depend *dep, *next;
list_for_each_safe(&op[i][j].pre, dep, next, pre_list) {
if (deps[dep->prereq.file]
>= (int)dep->prereq.op_num)
talloc_free(dep);
else
deps[dep->prereq.file]
= dep->prereq.op_num;
}
}
}
}
#if TRAVERSALS_TAKE_TRANSACTION_LOCK
/* Force an order among the traversals, so they don't deadlock (as much) */
static void make_traverse_depends(char *filename[],
struct op *op[], unsigned int num_ops[],
unsigned int num)
{
unsigned int i, num_traversals = 0;
int j;
struct op_desc *desc;
/* Sort by which one runs first. */
int compare_traverse_desc(const void *_a, const void *_b)
{
const struct op_desc *da = _a, *db = _b;
const struct op *a = &op[da->file][da->op_num],
*b = &op[db->file][db->op_num];
if (a->seqnum != b->seqnum)
return a->seqnum - b->seqnum;
/* If they have same seqnum, it means one didn't make any
* changes. Thus sort by end in that case. */
return a[a->group_len].seqnum - b[b->group_len].seqnum;
}
desc = talloc_array(NULL, struct op_desc, 1);
/* Count them. */
for (i = 0; i < num; i++) {
for (j = 1; j < num_ops[i]; j++) {
/* Traverse start (ignore those in
* transactions; they're already covered by
* transaction dependencies). */
if (starts_traverse(&op[i][j])
&& !in_transaction(op[i], j)) {
desc = talloc_realloc(NULL, desc,
struct op_desc,
num_traversals+1);
desc[num_traversals].file = i;
desc[num_traversals].op_num = j;
num_traversals++;
}
}
}
qsort(desc, num_traversals, sizeof(desc[0]), compare_traverse_desc);
for (i = 1; i < num_traversals; i++) {
const struct op *prev = &op[desc[i-1].file][desc[i-1].op_num];
const struct op *curr = &op[desc[i].file][desc[i].op_num];
/* Read traverses don't depend on each other (read lock). */
if (prev->type == OP_TDB_TRAVERSE_READ_START
&& curr->type == OP_TDB_TRAVERSE_READ_START)
continue;
/* Only make dependency if it's clear. */
if (compare_traverse_desc(&desc[i], &desc[i-1])) {
/* i depends on end of traverse i-1. */
struct op_desc end = desc[i-1];
end.op_num += prev->group_len;
add_dependency(NULL, op, filename, &desc[i], &end);
}
}
talloc_free(desc);
}
static void set_nonblock(int fd)
{
if (fcntl(fd, F_SETFL, fcntl(fd, F_GETFL)|O_NONBLOCK) != 0)
err(1, "Setting pipe nonblocking");
}
static bool handle_backoff(struct op *op[], int fd)
{
struct op_desc desc;
bool handled = false;
/* Sloppy coding: we assume PIPEBUF never fills. */
while (read(fd, &desc, sizeof(desc)) != -1) {
unsigned int i;
handled = true;
for (i = desc.op_num; i > 0; i--) {
if (op[desc.file][i].type == OP_TDB_TRAVERSE) {
/* We insert a fake end here. */
op[desc.file][i].type
= OP_TDB_TRAVERSE_END_EARLY;
break;
} else if (starts_traverse(&op[desc.file][i])) {
unsigned int start = i;
struct op tmp = op[desc.file][i];
/* Move the ops outside traverse. */
memmove(&op[desc.file][i],
&op[desc.file][i+1],
(desc.op_num-i-1) * sizeof(op[0][0]));
op[desc.file][desc.op_num] = tmp;
while (op[desc.file][i].group_start == start) {
op[desc.file][i++].group_start
= desc.op_num;
}
break;
}
}
}
return handled;
}
#else /* !TRAVERSALS_TAKE_TRANSACTION_LOCK */
static bool handle_backoff(struct op *op[], int fd)
{
return false;
}
#endif
static void derive_dependencies(struct tdb_context *tdb,
char *filename[],
struct op *op[], unsigned int num_ops[],
unsigned int num)
{
struct keyinfo *hash;
unsigned int h, i;
/* Create hash table for faster key lookup. */
hash = hash_ops(op, num_ops, num);
/* Sort them by sequence number. */
sort_ops(tdb, hash, filename, op, num);
/* Create dependencies back to the last change, rather than
* creating false dependencies by naively making each one
* depend on the previous. This has two purposes: it makes
* later optimization simpler, and it also avoids deadlock with
* same sequence number ops inside traversals (if one
* traversal doesn't write anything, two ops can have the same
* sequence number yet we can create a traversal dependency
* the other way). */
for (h = 0; h < total_keys * 2; h++) {
int prev = -1;
if (hash[h].num_users < 2)
continue;
for (i = 0; i < hash[h].num_users; i++) {
if (changes_db(&hash[h].key, &op[hash[h].user[i].file]
[hash[h].user[i].op_num])) {
depend_on_previous(op, filename, num,
hash[h].user, i, prev);
prev = i;
} else if (prev >= 0)
add_dependency(hash, op, filename,
&hash[h].user[i],
&hash[h].user[prev]);
}
}
#if TRAVERSALS_TAKE_TRANSACTION_LOCK
make_traverse_depends(filename, op, num_ops, num);
#endif
optimize_dependencies(op, num_ops, num);
}
static struct timeval run_test(char *argv[],
unsigned int num_ops[],
unsigned int hashsize[],
unsigned int tdb_flags[],
unsigned int open_flags[],
struct op *op[],
int fds[2])
{
unsigned int i;
struct timeval start, end, diff;
bool ok = true;
for (i = 0; argv[i+2]; i++) {
struct tdb_context *tdb;
char c;
switch (fork()) {
case -1:
err(1, "fork failed");
case 0:
close(fds[1]);
tdb = tdb_open(argv[1], hashsize[i],
tdb_flags[i], open_flags[i], 0600);
if (!tdb)
err(1, "Opening tdb %s", argv[1]);
/* This catches parent exiting. */
if (read(fds[0], &c, 1) != 1)
exit(1);
run_ops(tdb, pipes[i].fd[0], argv+2, op, i, 1,
num_ops[i], false);
check_deps(argv[2+i], op[i], num_ops[i]);
exit(0);
default:
break;
}
}
/* Let everything settle. */
sleep(1);
if (!quiet)
printf("Starting run...");
fflush(stdout);
gettimeofday(&start, NULL);
/* Tell them all to go! Any write of sufficient length will do. */
if (write(fds[1], hashsize, i) != i)
err(1, "Writing to wakeup pipe");
for (i = 0; argv[i + 2]; i++) {
int status;
wait(&status);
if (!WIFEXITED(status)) {
warnx("Child died with signal %i", WTERMSIG(status));
ok = false;
} else if (WEXITSTATUS(status) != 0)
/* Assume child spat out error. */
ok = false;
}
if (!ok)
exit(1);
gettimeofday(&end, NULL);
if (!quiet)
printf("done\n");
if (end.tv_usec < start.tv_usec) {
end.tv_usec += 1000000;
end.tv_sec--;
}
diff.tv_sec = end.tv_sec - start.tv_sec;
diff.tv_usec = end.tv_usec - start.tv_usec;
return diff;
}
static void init_tdb(struct tdb_context *master_tdb,
const char *name, unsigned int hashsize)
{
TDB_DATA key, data;
struct tdb_context *tdb;
tdb = tdb_open(name, hashsize, TDB_CLEAR_IF_FIRST|TDB_NOSYNC,
O_CREAT|O_TRUNC|O_RDWR, 0600);
if (!tdb)
errx(1, "opening tdb %s", name);
for (key = tdb_firstkey(master_tdb);
key.dptr;
key = tdb_nextkey(master_tdb, key)) {
data = tdb_fetch(master_tdb, key);
if (tdb_store(tdb, key, data, TDB_INSERT) != 0)
errx(1, "Failed to store initial key");
}
tdb_close(tdb);
}
int main(int argc, char *argv[])
{
struct timeval diff;
unsigned int i, num_ops[argc], hashsize[argc], tdb_flags[argc], open_flags[argc];
struct op *op[argc];
int fds[2];
struct tdb_context *master;
unsigned int runs = 1;
if (argc < 3)
errx(1, "Usage: %s [--quiet] [-n <number>] <tdbfile> <tracefile>...", argv[0]);
if (streq(argv[1], "--quiet")) {
quiet = true;
argv++;
argc--;
}
if (streq(argv[1], "-n")) {
runs = atoi(argv[2]);
argv += 2;
argc -= 2;
}
pipes = talloc_array(NULL, struct pipe, argc - 1);
for (i = 0; i < argc - 2; i++) {
if (!quiet)
printf("Loading tracefile %s...", argv[2+i]);
fflush(stdout);
op[i] = load_tracefile(argv+2, i, &num_ops[i], &hashsize[i],
&tdb_flags[i], &open_flags[i]);
if (pipe(pipes[i].fd) != 0)
err(1, "creating pipe");
/* Don't truncate, or clear if first: we do that. */
open_flags[i] &= ~(O_TRUNC);
tdb_flags[i] &= ~(TDB_CLEAR_IF_FIRST);
/* Open NOSYNC, to save time. */
tdb_flags[i] |= TDB_NOSYNC;
if (!quiet)
printf("done\n");
}
/* Dependency may figure we need to create seed records. */
master = tdb_open(NULL, 0, TDB_INTERNAL, O_RDWR, 0);
if (!quiet) {
printf("Calculating inter-dependencies...");
fflush(stdout);
}
derive_dependencies(master, argv+2, op, num_ops, i);
if (!quiet)
printf("done\n");
for (i = 0; i < runs; i++) {
init_tdb(master, argv[1], hashsize[0]);
/* Don't fork for single arg case: simple debugging. */
if (argc == 3) {
struct timeval start, end;
struct tdb_context *tdb;
tdb = tdb_open(argv[1], hashsize[0], tdb_flags[0],
open_flags[0], 0600);
if (!quiet) {
printf("Single threaded run...");
fflush(stdout);
}
gettimeofday(&start, NULL);
run_ops(tdb, pipes[0].fd[0], argv+2, op, 0, 1,
num_ops[0], false);
gettimeofday(&end, NULL);
if (!quiet)
printf("done\n");
tdb_close(tdb);
check_deps(argv[2], op[0], num_ops[0]);
if (end.tv_usec < start.tv_usec) {
end.tv_usec += 1000000;
end.tv_sec--;
}
diff.tv_sec = end.tv_sec - start.tv_sec;
diff.tv_usec = end.tv_usec - start.tv_usec;
goto print_time;
}
if (pipe(fds) != 0)
err(1, "creating pipe");
#if TRAVERSALS_TAKE_TRANSACTION_LOCK
if (pipe(pipes[argc-2].fd) != 0)
err(1, "creating pipe");
backoff_fd = pipes[argc-2].fd[1];
set_nonblock(pipes[argc-2].fd[1]);
set_nonblock(pipes[argc-2].fd[0]);
#endif
do {
diff = run_test(argv, num_ops, hashsize, tdb_flags,
open_flags, op, fds);
} while (handle_backoff(op, pipes[argc-2].fd[0]));
print_time:
if (!quiet)
printf("Time replaying: ");
printf("%lu usec\n", diff.tv_sec * 1000000UL + diff.tv_usec);
}
exit(0);
}
ccan/tdb/tools/speed.c deleted 100644 → 0
View file @ 98b8ada2
/* Simple speed test for TDB */
#include <err.h>
#include <time.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <sys/time.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <stdbool.h>
#include <ccan/tdb/tdb.h>
/* Nanoseconds per operation */
static size_t normalize(const struct timeval *start,
const struct timeval *stop,
unsigned int num)
{
struct timeval diff;
timersub(stop, start, &diff);
/* Floating point is more accurate here. */
return (double)(diff.tv_sec * 1000000 + diff.tv_usec)
/ num * 1000;
}
static size_t file_size(void)
{
struct stat st;
if (stat("/tmp/speed.tdb", &st) != 0)
return -1;
return st.st_size;
}
static int count_record(struct tdb_context *tdb,
TDB_DATA key, TDB_DATA data, void *p)
{
int *total = p;
*total += *(int *)data.dptr;
return 0;
}
int main(int argc, char *argv[])
{
unsigned int i, j, num = 1000, stage = 0, stopat = -1;
int flags = TDB_DEFAULT;
TDB_DATA key, data;
struct tdb_context *tdb;
struct timeval start, stop;
bool transaction = false;
if (argv[1] && strcmp(argv[1], "--internal") == 0) {
flags = TDB_INTERNAL;
argc--;
argv++;
}
if (argv[1] && strcmp(argv[1], "--transaction") == 0) {
transaction = true;
argc--;
argv++;
}
tdb = tdb_open("/tmp/speed.tdb", 100003, flags, O_RDWR|O_CREAT|O_TRUNC,
0600);
if (!tdb)
err(1, "Opening /tmp/speed.tdb");
key.dptr = (void *)&i;
key.dsize = sizeof(i);
data = key;
if (argv[1]) {
num = atoi(argv[1]);
argv++;
argc--;
}
if (argv[1]) {
stopat = atoi(argv[1]);
argv++;
argc--;
}
if (transaction && tdb_transaction_start(tdb))
errx(1, "starting transaction: %s", tdb_errorstr(tdb));
/* Add 1000 records. */
printf("Adding %u records: ", num); fflush(stdout);
gettimeofday(&start, NULL);
for (i = 0; i < num; i++)
if (tdb_store(tdb, key, data, TDB_INSERT) != 0)
errx(1, "Inserting key %u in tdb: %s",
i, tdb_errorstr(tdb));
gettimeofday(&stop, NULL);
if (transaction && tdb_transaction_commit(tdb))
errx(1, "committing transaction: %s", tdb_errorstr(tdb));
printf(" %zu ns (%zu bytes)\n",
normalize(&start, &stop, num), file_size());
if (++stage == stopat)
exit(0);
if (transaction && tdb_transaction_start(tdb))
errx(1, "starting transaction: %s", tdb_errorstr(tdb));
/* Finding 1000 records. */
printf("Finding %u records: ", num); fflush(stdout);
gettimeofday(&start, NULL);
for (i = 0; i < num; i++) {
int *dptr;
dptr = (int *)tdb_fetch(tdb, key).dptr;
if (!dptr || *dptr != i)
errx(1, "Fetching key %u in tdb gave %u",
i, dptr ? *dptr : -1);
}
gettimeofday(&stop, NULL);
if (transaction && tdb_transaction_commit(tdb))
errx(1, "committing transaction: %s", tdb_errorstr(tdb));
printf(" %zu ns (%zu bytes)\n",
normalize(&start, &stop, num), file_size());
if (++stage == stopat)
exit(0);
if (transaction && tdb_transaction_start(tdb))
errx(1, "starting transaction: %s", tdb_errorstr(tdb));
/* Missing 1000 records. */
printf("Missing %u records: ", num); fflush(stdout);
gettimeofday(&start, NULL);
for (i = num; i < num*2; i++) {
int *dptr;
dptr = (int *)tdb_fetch(tdb, key).dptr;
if (dptr)
errx(1, "Fetching key %u in tdb gave %u", i, *dptr);
}
gettimeofday(&stop, NULL);
if (transaction && tdb_transaction_commit(tdb))
errx(1, "committing transaction: %s", tdb_errorstr(tdb));
printf(" %zu ns (%zu bytes)\n",
normalize(&start, &stop, num), file_size());
if (++stage == stopat)
exit(0);
if (transaction && tdb_transaction_start(tdb))
errx(1, "starting transaction: %s", tdb_errorstr(tdb));
/* Traverse 1000 records. */
printf("Traversing %u records: ", num); fflush(stdout);
i = 0;
gettimeofday(&start, NULL);
if (tdb_traverse(tdb, count_record, &i) != num)
errx(1, "Traverse returned wrong number of records");
if (i != (num - 1) * (num / 2))
errx(1, "Traverse tallied to %u", i);
gettimeofday(&stop, NULL);
if (transaction && tdb_transaction_commit(tdb))
errx(1, "committing transaction: %s", tdb_errorstr(tdb));
printf(" %zu ns (%zu bytes)\n",
normalize(&start, &stop, num), file_size());
if (++stage == stopat)
exit(0);
if (transaction && tdb_transaction_start(tdb))
errx(1, "starting transaction: %s", tdb_errorstr(tdb));
/* Delete 1000 records (not in order). */
printf("Deleting %u records: ", num); fflush(stdout);
gettimeofday(&start, NULL);
for (j = 0; j < num; j++) {
i = (j + 100003) % num;
if (tdb_delete(tdb, key) != 0)
errx(1, "Deleting key %u in tdb: %s",
i, tdb_errorstr(tdb));
}
gettimeofday(&stop, NULL);
if (transaction && tdb_transaction_commit(tdb))
errx(1, "committing transaction: %s", tdb_errorstr(tdb));
printf(" %zu ns (%zu bytes)\n",
normalize(&start, &stop, num), file_size());
if (++stage == stopat)
exit(0);
if (transaction && tdb_transaction_start(tdb))
errx(1, "starting transaction: %s", tdb_errorstr(tdb));
/* Re-add 1000 records (not in order). */
printf("Re-adding %u records: ", num); fflush(stdout);
gettimeofday(&start, NULL);
for (j = 0; j < num; j++) {
i = (j + 100003) % num;
if (tdb_store(tdb, key, data, TDB_INSERT) != 0)
errx(1, "Inserting key %u in tdb: %s",
i, tdb_errorstr(tdb));
}
gettimeofday(&stop, NULL);
if (transaction && tdb_transaction_commit(tdb))
errx(1, "committing transaction: %s", tdb_errorstr(tdb));
printf(" %zu ns (%zu bytes)\n",
normalize(&start, &stop, num), file_size());
if (++stage == stopat)
exit(0);
if (transaction && tdb_transaction_start(tdb))
errx(1, "starting transaction: %s", tdb_errorstr(tdb));
/* Append 1000 records. */
printf("Appending %u records: ", num); fflush(stdout);
gettimeofday(&start, NULL);
for (i = 0; i < num; i++)
if (tdb_append(tdb, key, data) != 0)
errx(1, "Appending key %u in tdb: %s",
i, tdb_errorstr(tdb));
gettimeofday(&stop, NULL);
if (transaction && tdb_transaction_commit(tdb))
errx(1, "committing transaction: %s", tdb_errorstr(tdb));
printf(" %zu ns (%zu bytes)\n",
normalize(&start, &stop, num), file_size());
if (++stage == stopat)
exit(0);
if (transaction && tdb_transaction_start(tdb))
errx(1, "starting transaction: %s", tdb_errorstr(tdb));
/* Churn 1000 records: not in order! */
printf("Churning %u records: ", num); fflush(stdout);
gettimeofday(&start, NULL);
for (j = 0; j < num; j++) {
i = (j + 1000019) % num;
if (tdb_delete(tdb, key) != 0)
errx(1, "Deleting key %u in tdb: %s",
i, tdb_errorstr(tdb));
i += num;
if (tdb_store(tdb, key, data, TDB_INSERT) != 0)
errx(1, "Inserting key %u in tdb: %s",
i, tdb_errorstr(tdb));
}
gettimeofday(&stop, NULL);
if (transaction && tdb_transaction_commit(tdb))
errx(1, "committing transaction: %s", tdb_errorstr(tdb));
printf(" %zu ns (%zu bytes)\n",
normalize(&start, &stop, num), file_size());
return 0;
}
ccan/tdb/tools/starvation.c deleted 100644 → 0
View file @ 98b8ada2
/* Demonstrate starvation of tdb_lockall */
#include <ccan/tdb/tdb.h>
#include <err.h>
#include <unistd.h>
#include <fcntl.h>
#include <time.h>
#include <stdlib.h>
#include <stdbool.h>
#include <stdarg.h>
#include <stdio.h>
#include <string.h>
#include <sys/time.h>
static void usage(const char *extra)
{
errx(1, "%s%s"
"Usage: starvation [lockall|gradual] <num> <worktime-in-ms>\n"
" Each locker holds lock for between 1/2 and 1 1/2 times\n"
" worktime, then sleeps for one second.\n\n"
" Main process tries tdb_lockall or tdb_lockall_gradual.",
extra ? extra : "", extra ? "\n" : "");
}
static void run_and_sleep(struct tdb_context *tdb, int parentfd, unsigned time)
{
char c;
struct timespec hold;
unsigned rand, randtime;
TDB_DATA key;
key.dptr = (void *)&rand;
key.dsize = sizeof(rand);
while (read(parentfd, &c, 1) != 0) {
/* Lock a random key. */
rand = random();
if (tdb_chainlock(tdb, key) != 0)
errx(1, "chainlock failed: %s", tdb_errorstr(tdb));
/* Hold it for some variable time. */
randtime = time / 2 + (random() % time);
hold.tv_sec = randtime / 1000;
hold.tv_nsec = (randtime % 1000) * 1000000;
nanosleep(&hold, NULL);
if (tdb_chainunlock(tdb, key) != 0)
errx(1, "chainunlock failed: %s", tdb_errorstr(tdb));
/* Wait for a second without the lock. */
sleep(1);
}
exit(0);
}
static void logfn(struct tdb_context *tdb,
enum tdb_debug_level level,
const char *fmt, ...)
{
va_list ap;
va_start(ap, fmt);
vfprintf(stderr, fmt, ap);
va_end(ap);
}
int main(int argc, char *argv[])
{
int (*lockall)(struct tdb_context *);
unsigned int num, worktime, i;
int pfd[2];
struct tdb_context *tdb;
struct tdb_logging_context log = { logfn, NULL };
struct timeval start, end, duration;
if (argc != 4)
usage(NULL);
if (strcmp(argv[1], "lockall") == 0)
lockall = tdb_lockall;
else if (strcmp(argv[1], "gradual") == 0) {
#if 0
lockall = tdb_lockall_gradual;
#else
errx(1, "gradual is now the default implementation");
#endif
} else
usage("Arg1 should be 'lockall' or 'gradual'");
num = atoi(argv[2]);
worktime = atoi(argv[3]);
if (!num || !worktime)
usage("Number of threads and worktime must be non-zero");
if (pipe(pfd) != 0)
err(1, "Creating pipe");
tdb = tdb_open_ex("/tmp/starvation.tdb", 10000, TDB_DEFAULT,
O_RDWR|O_CREAT|O_TRUNC, 0600, &log, NULL);
if (!tdb)
err(1, "Opening tdb /tmp/starvation.tdb");
for (i = 0; i < num; i++) {
switch (fork()) {
case 0:
close(pfd[1]);
fcntl(pfd[0], F_SETFL,
fcntl(pfd[0], F_GETFL)|O_NONBLOCK);
srandom(getpid() + i);
if (tdb_reopen(tdb) != 0)
err(1, "Reopening tdb %s", tdb_name(tdb));
run_and_sleep(tdb, pfd[0], worktime);
case -1:
err(1, "forking");
}
/* Stagger the children. */
usleep(random() % (1000000 / num));
}
close(pfd[0]);
sleep(1);
gettimeofday(&start, NULL);
if (lockall(tdb) != 0)
errx(1, "lockall failed: %s", tdb_errorstr(tdb));
gettimeofday(&end, NULL);
duration.tv_sec = end.tv_sec - start.tv_sec;
duration.tv_usec = end.tv_usec - start.tv_usec;
if (duration.tv_usec < 0) {
--duration.tv_sec;
duration.tv_usec += 1000000;
}
if (tdb_unlockall(tdb) != 0)
errx(1, "unlockall failed: %s", tdb_errorstr(tdb));
tdb_close(tdb);
unlink("/tmp/starvation.tdb");
printf("Took %lu.%06lu seconds\n", duration.tv_sec, duration.tv_usec);
return 0;
}
ccan/tdb/tools/tdbdump.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
simple tdb dump util
Copyright (C) Andrew Tridgell 2001
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include <ccan/tdb/tdb.h>
#include <stdlib.h>
#include <getopt.h>
#include <stdarg.h>
#include <stdio.h>
#include <string.h>
#include <unistd.h>
#include <sys/types.h>
#include <fcntl.h>
#include <ctype.h>
static void print_data(TDB_DATA d)
{
unsigned char *p = (unsigned char *)d.dptr;
int len = d.dsize;
while (len--) {
if (isprint(*p) && !strchr("\"\\", *p)) {
fputc(*p, stdout);
} else {
printf("\\%02X", *p);
}
p++;
}
}
static int traverse_fn(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, void *state)
{
printf("{\n");
printf("key(%d) = \"", (int)key.dsize);
print_data(key);
printf("\"\n");
printf("data(%d) = \"", (int)dbuf.dsize);
print_data(dbuf);
printf("\"\n");
printf("}\n");
return 0;
}
static int dump_tdb(const char *fname, const char *keyname)
{
TDB_CONTEXT *tdb;
TDB_DATA key, value;
tdb = tdb_open(fname, 0, 0, O_RDONLY, 0);
if (!tdb) {
printf("Failed to open %s\n", fname);
return 1;
}
if (!keyname) {
tdb_traverse(tdb, traverse_fn, NULL);
} else {
key.dptr = (void *)keyname;
key.dsize = strlen( keyname);
value = tdb_fetch(tdb, key);
if (!value.dptr) {
return 1;
} else {
print_data(value);
free(value.dptr);
}
}
return 0;
}
static void usage( void)
{
printf( "Usage: tdbdump [options] <filename>\n\n");
printf( " -h this help message\n");
printf( " -k keyname dumps value of keyname\n");
}
int main(int argc, char *argv[])
{
char *fname, *keyname=NULL;
int c;
if (argc < 2) {
printf("Usage: tdbdump <fname>\n");
exit(1);
}
while ((c = getopt( argc, argv, "hk:")) != -1) {
switch (c) {
case 'h':
usage();
exit( 0);
case 'k':
keyname = optarg;
break;
default:
usage();
exit( 1);
}
}
fname = argv[optind];
return dump_tdb(fname, keyname);
}
ccan/tdb/tools/tdbtool.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
Samba database functions
Copyright (C) Andrew Tridgell 1999-2000
Copyright (C) Paul `Rusty' Russell 2000
Copyright (C) Jeremy Allison 2000
Copyright (C) Andrew Esh 2001
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include <ccan/tdb/tdb.h>
#include <stdlib.h>
#include <stdio.h>
#include <ctype.h>
#include <sys/time.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>
#include <string.h>
#include <stdarg.h>
static int do_command(void);
const char *cmdname;
char *arg1, *arg2;
size_t arg1len, arg2len;
int bIterate = 0;
char *line;
TDB_DATA iterate_kbuf;
char cmdline[1024];
static int disable_mmap;
enum commands {
CMD_CREATE_TDB,
CMD_OPEN_TDB,
CMD_TRANSACTION_START,
CMD_TRANSACTION_COMMIT,
CMD_TRANSACTION_CANCEL,
CMD_ERASE,
CMD_DUMP,
CMD_INSERT,
CMD_MOVE,
CMD_STORE,
CMD_SHOW,
CMD_KEYS,
CMD_HEXKEYS,
CMD_DELETE,
CMD_LIST_HASH_FREE,
CMD_LIST_FREE,
CMD_INFO,
CMD_MMAP,
CMD_SPEED,
CMD_FIRST,
CMD_NEXT,
CMD_SYSTEM,
CMD_CHECK,
CMD_QUIT,
CMD_HELP
};
typedef struct {
const char *name;
enum commands cmd;
} COMMAND_TABLE;
COMMAND_TABLE cmd_table[] = {
{"create", CMD_CREATE_TDB},
{"open", CMD_OPEN_TDB},
{"transaction_start", CMD_TRANSACTION_START},
{"transaction_commit", CMD_TRANSACTION_COMMIT},
{"transaction_cancel", CMD_TRANSACTION_CANCEL},
{"erase", CMD_ERASE},
{"dump", CMD_DUMP},
{"insert", CMD_INSERT},
{"move", CMD_MOVE},
{"store", CMD_STORE},
{"show", CMD_SHOW},
{"keys", CMD_KEYS},
{"hexkeys", CMD_HEXKEYS},
{"delete", CMD_DELETE},
{"list", CMD_LIST_HASH_FREE},
{"free", CMD_LIST_FREE},
{"info", CMD_INFO},
{"speed", CMD_SPEED},
{"mmap", CMD_MMAP},
{"first", CMD_FIRST},
{"1", CMD_FIRST},
{"next", CMD_NEXT},
{"n", CMD_NEXT},
{"check", CMD_CHECK},
{"quit", CMD_QUIT},
{"q", CMD_QUIT},
{"!", CMD_SYSTEM},
{NULL, CMD_HELP}
};
struct timeval tp1,tp2;
static void _start_timer(void)
{
gettimeofday(&tp1,NULL);
}
static double _end_timer(void)
{
gettimeofday(&tp2,NULL);
return((tp2.tv_sec - tp1.tv_sec) +
(tp2.tv_usec - tp1.tv_usec)*1.0e-6);
}
static void tdb_log(struct tdb_context *tdb, enum tdb_debug_level level, const char *format, ...)
{
va_list ap;
va_start(ap, format);
vfprintf(stderr, format, ap);
va_end(ap);
}
/* a tdb tool for manipulating a tdb database */
static TDB_CONTEXT *tdb;
static int print_rec(TDB_CONTEXT *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state);
static int print_key(TDB_CONTEXT *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state);
static int print_hexkey(TDB_CONTEXT *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state);
static void print_asc(const char *buf,int len)
{
int i;
/* We're probably printing ASCII strings so don't try to display
the trailing NULL character. */
if (buf[len - 1] == 0)
len--;
for (i=0;i<len;i++)
printf("%c",isprint(buf[i])?buf[i]:'.');
}
static void print_data(const char *buf,int len)
{
int i=0;
if (len<=0) return;
printf("[%03X] ",i);
for (i=0;i<len;) {
printf("%02X ",(int)((unsigned char)buf[i]));
i++;
if (i%8 == 0) printf(" ");
if (i%16 == 0) {
print_asc(&buf[i-16],8); printf(" ");
print_asc(&buf[i-8],8); printf("\n");
if (i<len) printf("[%03X] ",i);
}
}
if (i%16) {
int n;
n = 16 - (i%16);
printf(" ");
if (n>8) printf(" ");
while (n--) printf(" ");
n = i%16;
if (n > 8) n = 8;
print_asc(&buf[i-(i%16)],n); printf(" ");
n = (i%16) - n;
if (n>0) print_asc(&buf[i-n],n);
printf("\n");
}
}
static void help(void)
{
printf("\n"
"tdbtool: \n"
" create dbname : create a database\n"
" open dbname : open an existing database\n"
" openjh dbname : open an existing database (jenkins hash)\n"
" transaction_start : start a transaction\n"
" transaction_commit : commit a transaction\n"
" transaction_cancel : cancel a transaction\n"
" erase : erase the database\n"
" dump : dump the database as strings\n"
" keys : dump the database keys as strings\n"
" hexkeys : dump the database keys as hex values\n"
" info : print summary info about the database\n"
" insert key data : insert a record\n"
" move key file : move a record to a destination tdb\n"
" store key data : store a record (replace)\n"
" show key : show a record by key\n"
" delete key : delete a record by key\n"
" list : print the database hash table and freelist\n"
" free : print the database freelist\n"
" check : check the integrity of an opened database\n"
" speed : perform speed tests on the database\n"
" ! command : execute system command\n"
" 1 | first : print the first record\n"
" n | next : print the next record\n"
" q | quit : terminate\n"
" \\n : repeat 'next' command\n"
"\n");
}
static void terror(const char *why)
{
printf("%s\n", why);
}
static void create_tdb(const char *tdbname)
{
struct tdb_logging_context log_ctx;
log_ctx.log_fn = tdb_log;
if (tdb) tdb_close(tdb);
tdb = tdb_open_ex(tdbname, 0, TDB_CLEAR_IF_FIRST | (disable_mmap?TDB_NOMMAP:0),
O_RDWR | O_CREAT | O_TRUNC, 0600, &log_ctx, NULL);
if (!tdb) {
printf("Could not create %s: %s\n", tdbname, strerror(errno));
}
}
static void open_tdb(const char *tdbname)
{
struct tdb_logging_context log_ctx;
log_ctx.log_fn = tdb_log;
if (tdb) tdb_close(tdb);
tdb = tdb_open_ex(tdbname, 0, disable_mmap?TDB_NOMMAP:0, O_RDWR, 0600,
&log_ctx, NULL);
if (!tdb) {
printf("Could not open %s: %s\n", tdbname, strerror(errno));
}
}
static void insert_tdb(char *keyname, size_t keylen, char* data, size_t datalen)
{
TDB_DATA key, dbuf;
if ((keyname == NULL) || (keylen == 0)) {
terror("need key");
return;
}
key.dptr = (unsigned char *)keyname;
key.dsize = keylen;
dbuf.dptr = (unsigned char *)data;
dbuf.dsize = datalen;
if (tdb_store(tdb, key, dbuf, TDB_INSERT) == -1) {
terror("insert failed");
}
}
static void store_tdb(char *keyname, size_t keylen, char* data, size_t datalen)
{
TDB_DATA key, dbuf;
if ((keyname == NULL) || (keylen == 0)) {
terror("need key");
return;
}
if ((data == NULL) || (datalen == 0)) {
terror("need data");
return;
}
key.dptr = (unsigned char *)keyname;
key.dsize = keylen;
dbuf.dptr = (unsigned char *)data;
dbuf.dsize = datalen;
printf("Storing key:\n");
print_rec(tdb, key, dbuf, NULL);
if (tdb_store(tdb, key, dbuf, TDB_REPLACE) == -1) {
terror("store failed");
}
}
static void show_tdb(char *keyname, size_t keylen)
{
TDB_DATA key, dbuf;
if ((keyname == NULL) || (keylen == 0)) {
terror("need key");
return;
}
key.dptr = (unsigned char *)keyname;
key.dsize = keylen;
dbuf = tdb_fetch(tdb, key);
if (!dbuf.dptr) {
terror("fetch failed");
return;
}
print_rec(tdb, key, dbuf, NULL);
free( dbuf.dptr );
return;
}
static void delete_tdb(char *keyname, size_t keylen)
{
TDB_DATA key;
if ((keyname == NULL) || (keylen == 0)) {
terror("need key");
return;
}
key.dptr = (unsigned char *)keyname;
key.dsize = keylen;
if (tdb_delete(tdb, key) != 0) {
terror("delete failed");
}
}
static void move_rec(char *keyname, size_t keylen, char* tdbname)
{
TDB_DATA key, dbuf;
TDB_CONTEXT *dst_tdb;
if ((keyname == NULL) || (keylen == 0)) {
terror("need key");
return;
}
if ( !tdbname ) {
terror("need destination tdb name");
return;
}
key.dptr = (unsigned char *)keyname;
key.dsize = keylen;
dbuf = tdb_fetch(tdb, key);
if (!dbuf.dptr) {
terror("fetch failed");
return;
}
print_rec(tdb, key, dbuf, NULL);
dst_tdb = tdb_open(tdbname, 0, 0, O_RDWR, 0600);
if ( !dst_tdb ) {
terror("unable to open destination tdb");
return;
}
if ( tdb_store( dst_tdb, key, dbuf, TDB_REPLACE ) == -1 ) {
terror("failed to move record");
}
else
printf("record moved\n");
tdb_close( dst_tdb );
return;
}
static int print_rec(TDB_CONTEXT *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state)
{
printf("\nkey %d bytes\n", (int)key.dsize);
print_asc((const char *)key.dptr, key.dsize);
printf("\ndata %d bytes\n", (int)dbuf.dsize);
print_data((const char *)dbuf.dptr, dbuf.dsize);
return 0;
}
static int print_key(TDB_CONTEXT *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state)
{
printf("key %d bytes: ", (int)key.dsize);
print_asc((const char *)key.dptr, key.dsize);
printf("\n");
return 0;
}
static int print_hexkey(TDB_CONTEXT *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state)
{
printf("key %d bytes\n", (int)key.dsize);
print_data((const char *)key.dptr, key.dsize);
printf("\n");
return 0;
}
static int total_bytes;
static int traverse_fn(TDB_CONTEXT *the_tdb, TDB_DATA key, TDB_DATA dbuf, void *state)
{
total_bytes += dbuf.dsize;
return 0;
}
static void info_tdb(void)
{
char *summary = tdb_summary(tdb);
if (!summary) {
printf("Error = %s\n", tdb_errorstr(tdb));
} else {
printf("%s", summary);
free(summary);
}
}
static void speed_tdb(const char *tlimit)
{
unsigned timelimit = tlimit?atoi(tlimit):0;
double t;
int ops;
if (timelimit == 0) timelimit = 5;
ops = 0;
printf("Testing store speed for %u seconds\n", timelimit);
_start_timer();
do {
long int r = random();
TDB_DATA key, dbuf;
key.dptr = (unsigned char *)"store test";
key.dsize = strlen((char *)key.dptr);
dbuf.dptr = (unsigned char *)&r;
dbuf.dsize = sizeof(r);
tdb_store(tdb, key, dbuf, TDB_REPLACE);
t = _end_timer();
ops++;
} while (t < timelimit);
printf("%10.3f ops/sec\n", ops/t);
ops = 0;
printf("Testing fetch speed for %u seconds\n", timelimit);
_start_timer();
do {
long int r = random();
TDB_DATA key, dbuf;
key.dptr = (unsigned char *)"store test";
key.dsize = strlen((char *)key.dptr);
dbuf.dptr = (unsigned char *)&r;
dbuf.dsize = sizeof(r);
tdb_fetch(tdb, key);
t = _end_timer();
ops++;
} while (t < timelimit);
printf("%10.3f ops/sec\n", ops/t);
ops = 0;
printf("Testing transaction speed for %u seconds\n", timelimit);
_start_timer();
do {
long int r = random();
TDB_DATA key, dbuf;
key.dptr = (unsigned char *)"transaction test";
key.dsize = strlen((char *)key.dptr);
dbuf.dptr = (unsigned char *)&r;
dbuf.dsize = sizeof(r);
tdb_transaction_start(tdb);
tdb_store(tdb, key, dbuf, TDB_REPLACE);
tdb_transaction_commit(tdb);
t = _end_timer();
ops++;
} while (t < timelimit);
printf("%10.3f ops/sec\n", ops/t);
ops = 0;
printf("Testing traverse speed for %u seconds\n", timelimit);
_start_timer();
do {
tdb_traverse(tdb, traverse_fn, NULL);
t = _end_timer();
ops++;
} while (t < timelimit);
printf("%10.3f ops/sec\n", ops/t);
}
static void toggle_mmap(void)
{
disable_mmap = !disable_mmap;
if (disable_mmap) {
printf("mmap is disabled\n");
} else {
printf("mmap is enabled\n");
}
}
static char *tdb_getline(const char *prompt)
{
static char thisline[1024];
char *p;
fputs(prompt, stdout);
thisline[0] = 0;
p = fgets(thisline, sizeof(thisline)-1, stdin);
if (p) p = strchr(p, '\n');
if (p) *p = 0;
return p?thisline:NULL;
}
static int do_delete_fn(TDB_CONTEXT *the_tdb, TDB_DATA key, TDB_DATA dbuf,
void *state)
{
return tdb_delete(the_tdb, key);
}
static void first_record(TDB_CONTEXT *the_tdb, TDB_DATA *pkey)
{
TDB_DATA dbuf;
*pkey = tdb_firstkey(the_tdb);
dbuf = tdb_fetch(the_tdb, *pkey);
if (!dbuf.dptr) terror("fetch failed");
else {
print_rec(the_tdb, *pkey, dbuf, NULL);
}
}
static void next_record(TDB_CONTEXT *the_tdb, TDB_DATA *pkey)
{
TDB_DATA dbuf;
*pkey = tdb_nextkey(the_tdb, *pkey);
dbuf = tdb_fetch(the_tdb, *pkey);
if (!dbuf.dptr)
terror("fetch failed");
else
print_rec(the_tdb, *pkey, dbuf, NULL);
}
static void check_db(TDB_CONTEXT *the_tdb)
{
if (!the_tdb) {
printf("Error: No database opened!\n");
} else {
if (tdb_check(the_tdb, NULL, NULL) == -1)
printf("Integrity check for the opened database failed.\n");
else
printf("Database integrity is OK.\n");
}
}
static int do_command(void)
{
COMMAND_TABLE *ctp = cmd_table;
enum commands mycmd = CMD_HELP;
int cmd_len;
if (cmdname && strlen(cmdname) == 0) {
mycmd = CMD_NEXT;
} else {
while (ctp->name) {
cmd_len = strlen(ctp->name);
if (strncmp(ctp->name,cmdname,cmd_len) == 0) {
mycmd = ctp->cmd;
break;
}
ctp++;
}
}
switch (mycmd) {
case CMD_CREATE_TDB:
bIterate = 0;
create_tdb(arg1);
return 0;
case CMD_OPEN_TDB:
bIterate = 0;
open_tdb(arg1);
return 0;
case CMD_SYSTEM:
/* Shell command */
if (system(arg1) == -1) {
terror("system() call failed\n");
}
return 0;
case CMD_QUIT:
return 1;
default:
/* all the rest require a open database */
if (!tdb) {
bIterate = 0;
terror("database not open");
help();
return 0;
}
switch (mycmd) {
case CMD_TRANSACTION_START:
bIterate = 0;
tdb_transaction_start(tdb);
return 0;
case CMD_TRANSACTION_COMMIT:
bIterate = 0;
tdb_transaction_commit(tdb);
return 0;
case CMD_TRANSACTION_CANCEL:
bIterate = 0;
tdb_transaction_cancel(tdb);
return 0;
case CMD_ERASE:
bIterate = 0;
tdb_traverse(tdb, do_delete_fn, NULL);
return 0;
case CMD_DUMP:
bIterate = 0;
tdb_traverse(tdb, print_rec, NULL);
return 0;
case CMD_INSERT:
bIterate = 0;
insert_tdb(arg1, arg1len,arg2,arg2len);
return 0;
case CMD_MOVE:
bIterate = 0;
move_rec(arg1,arg1len,arg2);
return 0;
case CMD_STORE:
bIterate = 0;
store_tdb(arg1,arg1len,arg2,arg2len);
return 0;
case CMD_SHOW:
bIterate = 0;
show_tdb(arg1, arg1len);
return 0;
case CMD_KEYS:
tdb_traverse(tdb, print_key, NULL);
return 0;
case CMD_HEXKEYS:
tdb_traverse(tdb, print_hexkey, NULL);
return 0;
case CMD_DELETE:
bIterate = 0;
delete_tdb(arg1,arg1len);
return 0;
case CMD_LIST_HASH_FREE:
tdb_dump_all(tdb);
return 0;
case CMD_LIST_FREE:
tdb_printfreelist(tdb);
return 0;
case CMD_INFO:
info_tdb();
return 0;
case CMD_SPEED:
speed_tdb(arg1);
return 0;
case CMD_MMAP:
toggle_mmap();
return 0;
case CMD_FIRST:
bIterate = 1;
first_record(tdb, &iterate_kbuf);
return 0;
case CMD_NEXT:
if (bIterate)
next_record(tdb, &iterate_kbuf);
return 0;
case CMD_CHECK:
check_db(tdb);
return 0;
case CMD_HELP:
help();
return 0;
case CMD_CREATE_TDB:
case CMD_OPEN_TDB:
case CMD_SYSTEM:
case CMD_QUIT:
/*
* unhandled commands. cases included here to avoid compiler
* warnings.
*/
return 0;
}
}
return 0;
}
static char *convert_string(char *instring, size_t *sizep)
{
size_t length = 0;
char *outp, *inp;
char temp[3];
outp = inp = instring;
while (*inp) {
if (*inp == '\\') {
inp++;
if (*inp && strchr("0123456789abcdefABCDEF",(int)*inp)) {
temp[0] = *inp++;
temp[1] = '\0';
if (*inp && strchr("0123456789abcdefABCDEF",(int)*inp)) {
temp[1] = *inp++;
temp[2] = '\0';
}
*outp++ = (char)strtol((const char *)temp,NULL,16);
} else {
*outp++ = *inp++;
}
} else {
*outp++ = *inp++;
}
length++;
}
*sizep = length;
return instring;
}
int main(int argc, char *argv[])
{
cmdname = "";
arg1 = NULL;
arg1len = 0;
arg2 = NULL;
arg2len = 0;
if (argv[1]) {
cmdname = "open";
arg1 = argv[1];
do_command();
cmdname = "";
arg1 = NULL;
}
switch (argc) {
case 1:
case 2:
/* Interactive mode */
while ((cmdname = tdb_getline("tdb> "))) {
arg2 = arg1 = NULL;
if ((arg1 = strchr((const char *)cmdname,' ')) != NULL) {
arg1++;
arg2 = arg1;
while (*arg2) {
if (*arg2 == ' ') {
*arg2++ = '\0';
break;
}
if ((*arg2++ == '\\') && (*arg2 == ' ')) {
arg2++;
}
}
}
if (arg1) arg1 = convert_string(arg1,&arg1len);
if (arg2) arg2 = convert_string(arg2,&arg2len);
if (do_command()) break;
}
break;
case 5:
arg2 = convert_string(argv[4],&arg2len);
case 4:
arg1 = convert_string(argv[3],&arg1len);
case 3:
cmdname = argv[2];
default:
do_command();
break;
}
if (tdb) tdb_close(tdb);
return 0;
}
ccan/tdb/tools/tdbtorture.c deleted 100644 → 0
View file @ 98b8ada2
/* this tests tdb by doing lots of ops from several simultaneous
writers - that stresses the locking code.
*/
#include <ccan/tdb/tdb.h>
#include <stdlib.h>
#include <err.h>
#include <getopt.h>
#include <stdarg.h>
#include <stdio.h>
#include <string.h>
#include <errno.h>
#include <unistd.h>
#include <sys/types.h>
#include <fcntl.h>
#include <time.h>
#include <sys/wait.h>
#define REOPEN_PROB 30
#define DELETE_PROB 8
#define STORE_PROB 4
#define APPEND_PROB 6
#define TRANSACTION_PROB 10
#define TRANSACTION_PREPARE_PROB 2
#define LOCKSTORE_PROB 5
#define TRAVERSE_PROB 20
#define TRAVERSE_READ_PROB 20
#define TRAVERSE_MOD_PROB 100
#define TRAVERSE_ABORT_PROB 500
#define CULL_PROB 100
#define KEYLEN 3
#define DATALEN 100
static struct tdb_context *db;
static int in_transaction;
static int in_traverse;
static int error_count;
static int always_transaction = 0;
static int hash_size = 2;
static int loopnum;
static int count_pipe;
static struct tdb_logging_context log_ctx;
#ifdef PRINTF_FMT
static void tdb_log(struct tdb_context *tdb, enum tdb_debug_level level, const char *format, ...) PRINTF_FMT(3,4);
#endif
static void tdb_log(struct tdb_context *tdb, enum tdb_debug_level level, const char *format, ...)
{
va_list ap;
if (level != TDB_DEBUG_TRACE)
error_count++;
va_start(ap, format);
vfprintf(stdout, format, ap);
va_end(ap);
fflush(stdout);
#if 0
if (level != TDB_DEBUG_TRACE) {
char *ptr;
signal(SIGUSR1, SIG_IGN);
asprintf(&ptr,"xterm -e gdb /proc/%d/exe %d", getpid(), getpid());
system(ptr);
free(ptr);
}
#endif
}
static void fatal(const char *why)
{
perror(why);
error_count++;
}
static char *randbuf(int len)
{
char *buf;
int i;
buf = (char *)malloc(len+1);
for (i=0;i<len;i++) {
buf[i] = 'a' + (rand() % 26);
}
buf[i] = 0;
return buf;
}
static void addrec_db(void);
static int modify_traverse(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
void *state)
{
#if CULL_PROB
if (random() % CULL_PROB == 0) {
tdb_delete(tdb, key);
}
#endif
#if TRAVERSE_MOD_PROB
if (random() % TRAVERSE_MOD_PROB == 0) {
addrec_db();
}
#endif
#if TRAVERSE_ABORT_PROB
if (random() % TRAVERSE_ABORT_PROB == 0)
return 1;
#endif
return 0;
}
static void addrec_db(void)
{
int klen, dlen;
char *k, *d;
TDB_DATA key, data;
klen = 1 + (rand() % KEYLEN);
dlen = 1 + (rand() % DATALEN);
k = randbuf(klen);
d = randbuf(dlen);
key.dptr = (unsigned char *)k;
key.dsize = klen+1;
data.dptr = (unsigned char *)d;
data.dsize = dlen+1;
#if REOPEN_PROB
if (in_traverse == 0 && in_transaction == 0 && random() % REOPEN_PROB == 0) {
tdb_reopen_all(0);
goto next;
}
#endif
#if TRANSACTION_PROB
if (in_traverse == 0 && in_transaction == 0 && (always_transaction || random() % TRANSACTION_PROB == 0)) {
if (tdb_transaction_start(db) != 0) {
fatal("tdb_transaction_start failed");
}
in_transaction++;
goto next;
}
if (in_traverse == 0 && in_transaction && random() % TRANSACTION_PROB == 0) {
if (random() % TRANSACTION_PREPARE_PROB == 0) {
if (tdb_transaction_prepare_commit(db) != 0) {
fatal("tdb_transaction_prepare_commit failed");
}
}
if (tdb_transaction_commit(db) != 0) {
fatal("tdb_transaction_commit failed");
}
in_transaction--;
goto next;
}
if (in_traverse == 0 && in_transaction && random() % TRANSACTION_PROB == 0) {
if (tdb_transaction_cancel(db) != 0) {
fatal("tdb_transaction_cancel failed");
}
in_transaction--;
goto next;
}
#endif
#if DELETE_PROB
if (random() % DELETE_PROB == 0) {
tdb_delete(db, key);
goto next;
}
#endif
#if STORE_PROB
if (random() % STORE_PROB == 0) {
if (tdb_store(db, key, data, TDB_REPLACE) != 0) {
fatal("tdb_store failed");
}
goto next;
}
#endif
#if APPEND_PROB
if (random() % APPEND_PROB == 0) {
if (tdb_append(db, key, data) != 0) {
fatal("tdb_append failed");
}
goto next;
}
#endif
#if LOCKSTORE_PROB
if (random() % LOCKSTORE_PROB == 0) {
tdb_chainlock(db, key);
data = tdb_fetch(db, key);
if (tdb_store(db, key, data, TDB_REPLACE) != 0) {
fatal("tdb_store failed");
}
if (data.dptr) free(data.dptr);
tdb_chainunlock(db, key);
goto next;
}
#endif
#if TRAVERSE_PROB
/* FIXME: recursive traverses break transactions? */
if (in_traverse == 0 && random() % TRAVERSE_PROB == 0) {
in_traverse++;
tdb_traverse(db, modify_traverse, NULL);
in_traverse--;
goto next;
}
#endif
#if TRAVERSE_READ_PROB
if (in_traverse == 0 && random() % TRAVERSE_READ_PROB == 0) {
in_traverse++;
tdb_traverse_read(db, NULL, NULL);
in_traverse--;
goto next;
}
#endif
data = tdb_fetch(db, key);
if (data.dptr) free(data.dptr);
next:
free(k);
free(d);
}
static int traverse_fn(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf,
void *state)
{
tdb_delete(tdb, key);
return 0;
}
static void usage(void)
{
printf("Usage: tdbtorture [-t] [-k] [-n NUM_PROCS] [-l NUM_LOOPS] [-s SEED] [-H HASH_SIZE]\n");
exit(0);
}
static void send_count_and_suicide(int sig)
{
/* This ensures our successor can continue where we left off. */
write(count_pipe, &loopnum, sizeof(loopnum));
/* This gives a unique signature. */
kill(getpid(), SIGUSR2);
}
static int run_child(int i, int seed, unsigned num_loops, unsigned start)
{
db = tdb_open_ex("torture.tdb", hash_size, TDB_DEFAULT,
O_RDWR | O_CREAT, 0600, &log_ctx, NULL);
if (!db) {
fatal("db open failed");
}
srand(seed + i);
srandom(seed + i);
/* Set global, then we're ready to handle being killed. */
loopnum = start;
signal(SIGUSR1, send_count_and_suicide);
for (;loopnum<num_loops && error_count == 0;loopnum++) {
addrec_db();
}
if (error_count == 0) {
tdb_traverse_read(db, NULL, NULL);
if (always_transaction) {
while (in_transaction) {
tdb_transaction_cancel(db);
in_transaction--;
}
if (tdb_transaction_start(db) != 0)
fatal("tdb_transaction_start failed");
}
tdb_traverse(db, traverse_fn, NULL);
tdb_traverse(db, traverse_fn, NULL);
if (always_transaction) {
if (tdb_transaction_commit(db) != 0)
fatal("tdb_transaction_commit failed");
}
}
tdb_close(db);
return (error_count < 100 ? error_count : 100);
}
int main(int argc, char * const *argv)
{
int i, seed = -1;
int num_loops = 5000;
int num_procs = 3;
int c, pfds[2];
extern char *optarg;
pid_t *pids;
int kill_random = 0;
int *done;
log_ctx.log_fn = tdb_log;
while ((c = getopt(argc, argv, "n:l:s:H:thk")) != -1) {
switch (c) {
case 'n':
num_procs = strtol(optarg, NULL, 0);
break;
case 'l':
num_loops = strtol(optarg, NULL, 0);
break;
case 'H':
hash_size = strtol(optarg, NULL, 0);
break;
case 's':
seed = strtol(optarg, NULL, 0);
break;
case 't':
always_transaction = 1;
break;
case 'k':
kill_random = 1;
break;
default:
usage();
}
}
unlink("torture.tdb");
if (seed == -1) {
seed = (getpid() + time(NULL)) & 0x7FFFFFFF;
}
if (num_procs == 1 && !kill_random) {
/* Don't fork for this case, makes debugging easier. */
error_count = run_child(0, seed, num_loops, 0);
goto done;
}
pids = (pid_t *)calloc(sizeof(pid_t), num_procs);
done = (int *)calloc(sizeof(int), num_procs);
if (pipe(pfds) != 0) {
perror("Creating pipe");
exit(1);
}
count_pipe = pfds[1];
for (i=0;i<num_procs;i++) {
if ((pids[i]=fork()) == 0) {
close(pfds[0]);
if (i == 0) {
printf("testing with %d processes, %d loops, %d hash_size, seed=%d%s\n",
num_procs, num_loops, hash_size, seed, always_transaction ? " (all within transactions)" : "");
}
exit(run_child(i, seed, num_loops, 0));
}
}
while (num_procs) {
int status, j;
pid_t pid;
if (error_count != 0) {
/* try and stop the test on any failure */
for (j=0;j<num_procs;j++) {
if (pids[j] != 0) {
kill(pids[j], SIGTERM);
}
}
}
pid = waitpid(-1, &status, kill_random ? WNOHANG : 0);
if (pid == 0) {
struct timespec ts;
/* Sleep for 1/10 second. */
ts.tv_sec = 0;
ts.tv_nsec = 100000000;
nanosleep(&ts, NULL);
/* Kill someone. */
kill(pids[random() % num_procs], SIGUSR1);
continue;
}
if (pid == -1) {
perror("failed to wait for child\n");
exit(1);
}
for (j=0;j<num_procs;j++) {
if (pids[j] == pid) break;
}
if (j == num_procs) {
printf("unknown child %d exited!?\n", (int)pid);
exit(1);
}
if (WIFSIGNALED(status)) {
if (WTERMSIG(status) == SIGUSR2
|| WTERMSIG(status) == SIGUSR1) {
/* SIGUSR2 means they wrote to pipe. */
if (WTERMSIG(status) == SIGUSR2) {
read(pfds[0], &done[j],
sizeof(done[j]));
}
pids[j] = fork();
if (pids[j] == 0)
exit(run_child(j, seed, num_loops,
done[j]));
printf("Restarting child %i for %u-%u\n",
j, done[j], num_loops);
continue;
}
printf("child %d exited with signal %d\n",
(int)pid, WTERMSIG(status));
error_count++;
} else {
if (WEXITSTATUS(status) != 0) {
printf("child %d exited with status %d\n",
(int)pid, WEXITSTATUS(status));
error_count++;
}
}
memmove(&pids[j], &pids[j+1],
(num_procs - j - 1)*sizeof(pids[0]));
num_procs--;
}
free(pids);
done:
if (error_count == 0) {
db = tdb_open_ex("torture.tdb", hash_size, TDB_DEFAULT,
O_RDWR, 0, &log_ctx, NULL);
if (!db) {
fatal("db open failed");
}
if (tdb_check(db, NULL, NULL) == -1) {
printf("db check failed");
exit(1);
}
tdb_close(db);
printf("OK\n");
}
return error_count;
}
ccan/tdb/transaction.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
trivial database library
Copyright (C) Andrew Tridgell 2005
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
/*
transaction design:
- only allow a single transaction at a time per database. This makes
using the transaction API simpler, as otherwise the caller would
have to cope with temporary failures in transactions that conflict
with other current transactions
- keep the transaction recovery information in the same file as the
database, using a special 'transaction recovery' record pointed at
by the header. This removes the need for extra journal files as
used by some other databases
- dynamically allocated the transaction recover record, re-using it
for subsequent transactions. If a larger record is needed then
tdb_free() the old record to place it on the normal tdb freelist
before allocating the new record
- during transactions, keep a linked list of writes all that have
been performed by intercepting all tdb_write() calls. The hooked
transaction versions of tdb_read() and tdb_write() check this
linked list and try to use the elements of the list in preference
to the real database.
- don't allow any locks to be held when a transaction starts,
otherwise we can end up with deadlock (plus lack of lock nesting
in posix locks would mean the lock is lost)
- if the caller gains a lock during the transaction but doesn't
release it then fail the commit
- allow for nested calls to tdb_transaction_start(), re-using the
existing transaction record. If the inner transaction is cancelled
then a subsequent commit will fail
- keep a mirrored copy of the tdb hash chain heads to allow for the
fast hash heads scan on traverse, updating the mirrored copy in
the transaction version of tdb_write
- allow callers to mix transaction and non-transaction use of tdb,
although once a transaction is started then an exclusive lock is
gained until the transaction is committed or cancelled
- the commit stategy involves first saving away all modified data
into a linearised buffer in the transaction recovery area, then
marking the transaction recovery area with a magic value to
indicate a valid recovery record. In total 4 fsync/msync calls are
needed per commit to prevent race conditions. It might be possible
to reduce this to 3 or even 2 with some more work.
- check for a valid recovery record on open of the tdb, while the
open lock is held. Automatically recover from the transaction
recovery area if needed, then continue with the open as
usual. This allows for smooth crash recovery with no administrator
intervention.
- if TDB_NOSYNC is passed to flags in tdb_open then transactions are
still available, but no transaction recovery area is used and no
fsync/msync calls are made.
- if TDB_ALLOW_NESTING is passed to flags in tdb open, or added using
tdb_add_flags() transaction nesting is enabled.
It resets the TDB_DISALLOW_NESTING flag, as both cannot be used together.
The default is that transaction nesting is allowed.
Note: this default may change in future versions of tdb.
Beware. when transactions are nested a transaction successfully
completed with tdb_transaction_commit() can be silently unrolled later.
- if TDB_DISALLOW_NESTING is passed to flags in tdb open, or added using
tdb_add_flags() transaction nesting is disabled.
It resets the TDB_ALLOW_NESTING flag, as both cannot be used together.
An attempt create a nested transaction will fail with TDB_ERR_NESTING.
The default is that transaction nesting is allowed.
Note: this default may change in future versions of tdb.
*/
/*
hold the context of any current transaction
*/
struct tdb_transaction {
/* we keep a mirrored copy of the tdb hash heads here so
tdb_next_hash_chain() can operate efficiently */
uint32_t *hash_heads;
/* the original io methods - used to do IOs to the real db */
const struct tdb_methods *io_methods;
/* the list of transaction blocks. When a block is first
written to, it gets created in this list */
uint8_t **blocks;
uint32_t num_blocks;
uint32_t block_size; /* bytes in each block */
uint32_t last_block_size; /* number of valid bytes in the last block */
/* non-zero when an internal transaction error has
occurred. All write operations will then fail until the
transaction is ended */
int transaction_error;
/* when inside a transaction we need to keep track of any
nested tdb_transaction_start() calls, as these are allowed,
but don't create a new transaction */
int nesting;
/* set when a prepare has already occurred */
bool prepared;
tdb_off_t magic_offset;
/* old file size before transaction */
tdb_len_t old_map_size;
/* we should re-pack on commit */
bool need_repack;
};
/*
read while in a transaction. We need to check first if the data is in our list
of transaction elements, then if not do a real read
*/
static int transaction_read(struct tdb_context *tdb, tdb_off_t off, void *buf,
tdb_len_t len, int cv)
{
uint32_t blk;
/* break it down into block sized ops */
while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
if (transaction_read(tdb, off, buf, len2, cv) != 0) {
return -1;
}
len -= len2;
off += len2;
buf = (void *)(len2 + (char *)buf);
}
if (len == 0) {
return 0;
}
blk = off / tdb->transaction->block_size;
/* see if we have it in the block list */
if (tdb->transaction->num_blocks <= blk ||
tdb->transaction->blocks[blk] == NULL) {
/* nope, do a real read */
if (tdb->transaction->io_methods->tdb_read(tdb, off, buf, len, cv) != 0) {
goto fail;
}
return 0;
}
/* it is in the block list. Now check for the last block */
if (blk == tdb->transaction->num_blocks-1) {
if (len > tdb->transaction->last_block_size) {
goto fail;
}
}
/* now copy it out of this block */
memcpy(buf, tdb->transaction->blocks[blk] + (off % tdb->transaction->block_size), len);
if (cv) {
tdb_convert(buf, len);
}
return 0;
fail:
TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_read: failed at off=%d len=%d\n", off, len));
tdb->ecode = TDB_ERR_IO;
tdb->transaction->transaction_error = 1;
return -1;
}
/*
write while in a transaction
*/
static int transaction_write(struct tdb_context *tdb, tdb_off_t off,
const void *buf, tdb_len_t len)
{
uint32_t blk;
/* Only a commit is allowed on a prepared transaction */
if (tdb->transaction->prepared) {
tdb->ecode = TDB_ERR_EINVAL;
TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_write: transaction already prepared, write not allowed\n"));
tdb->transaction->transaction_error = 1;
return -1;
}
/* if the write is to a hash head, then update the transaction
hash heads */
if (len == sizeof(tdb_off_t) && off >= FREELIST_TOP &&
off < FREELIST_TOP+TDB_HASHTABLE_SIZE(tdb)) {
uint32_t chain = (off-FREELIST_TOP) / sizeof(tdb_off_t);
memcpy(&tdb->transaction->hash_heads[chain], buf, len);
}
/* break it up into block sized chunks */
while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
if (transaction_write(tdb, off, buf, len2) != 0) {
return -1;
}
len -= len2;
off += len2;
if (buf != NULL) {
buf = (const void *)(len2 + (const char *)buf);
}
}
if (len == 0) {
return 0;
}
blk = off / tdb->transaction->block_size;
off = off % tdb->transaction->block_size;
if (tdb->transaction->num_blocks <= blk) {
uint8_t **new_blocks;
/* expand the blocks array */
if (tdb->transaction->blocks == NULL) {
new_blocks = (uint8_t **)malloc(
(blk+1)*sizeof(uint8_t *));
} else {
new_blocks = (uint8_t **)realloc(
tdb->transaction->blocks,
(blk+1)*sizeof(uint8_t *));
}
if (new_blocks == NULL) {
tdb->ecode = TDB_ERR_OOM;
goto fail;
}
memset(&new_blocks[tdb->transaction->num_blocks], 0,
(1+(blk - tdb->transaction->num_blocks))*sizeof(uint8_t *));
tdb->transaction->blocks = new_blocks;
tdb->transaction->num_blocks = blk+1;
tdb->transaction->last_block_size = 0;
}
/* allocate and fill a block? */
if (tdb->transaction->blocks[blk] == NULL) {
tdb->transaction->blocks[blk] = (uint8_t *)calloc(tdb->transaction->block_size, 1);
if (tdb->transaction->blocks[blk] == NULL) {
tdb->ecode = TDB_ERR_OOM;
tdb->transaction->transaction_error = 1;
return -1;
}
if (tdb->transaction->old_map_size > blk * tdb->transaction->block_size) {
tdb_len_t len2 = tdb->transaction->block_size;
if (len2 + (blk * tdb->transaction->block_size) > tdb->transaction->old_map_size) {
len2 = tdb->transaction->old_map_size - (blk * tdb->transaction->block_size);
}
if (tdb->transaction->io_methods->tdb_read(tdb, blk * tdb->transaction->block_size,
tdb->transaction->blocks[blk],
len2, 0) != 0) {
SAFE_FREE(tdb->transaction->blocks[blk]);
tdb->ecode = TDB_ERR_IO;
goto fail;
}
if (blk == tdb->transaction->num_blocks-1) {
tdb->transaction->last_block_size = len2;
}
}
}
/* overwrite part of an existing block */
if (buf == NULL) {
memset(tdb->transaction->blocks[blk] + off, 0, len);
} else {
memcpy(tdb->transaction->blocks[blk] + off, buf, len);
}
if (blk == tdb->transaction->num_blocks-1) {
if (len + off > tdb->transaction->last_block_size) {
tdb->transaction->last_block_size = len + off;
}
}
return 0;
fail:
TDB_LOG((tdb, TDB_DEBUG_FATAL, "transaction_write: failed at off=%d len=%d\n",
(blk*tdb->transaction->block_size) + off, len));
tdb->transaction->transaction_error = 1;
return -1;
}
/*
write while in a transaction - this varient never expands the transaction blocks, it only
updates existing blocks. This means it cannot change the recovery size
*/
static int transaction_write_existing(struct tdb_context *tdb, tdb_off_t off,
const void *buf, tdb_len_t len)
{
uint32_t blk;
/* break it up into block sized chunks */
while (len + (off % tdb->transaction->block_size) > tdb->transaction->block_size) {
tdb_len_t len2 = tdb->transaction->block_size - (off % tdb->transaction->block_size);
if (transaction_write_existing(tdb, off, buf, len2) != 0) {
return -1;
}
len -= len2;
off += len2;
if (buf != NULL) {
buf = (const void *)(len2 + (const char *)buf);
}
}
if (len == 0) {
return 0;
}
blk = off / tdb->transaction->block_size;
off = off % tdb->transaction->block_size;
if (tdb->transaction->num_blocks <= blk ||
tdb->transaction->blocks[blk] == NULL) {
return 0;
}
if (blk == tdb->transaction->num_blocks-1 &&
off + len > tdb->transaction->last_block_size) {
if (off >= tdb->transaction->last_block_size) {
return 0;
}
len = tdb->transaction->last_block_size - off;
}
/* overwrite part of an existing block */
memcpy(tdb->transaction->blocks[blk] + off, buf, len);
return 0;
}
/*
accelerated hash chain head search, using the cached hash heads
*/
static void transaction_next_hash_chain(struct tdb_context *tdb, uint32_t *chain)
{
uint32_t h = *chain;
for (;h < tdb->header.hash_size;h++) {
/* the +1 takes account of the freelist */
if (0 != tdb->transaction->hash_heads[h+1]) {
break;
}
}
(*chain) = h;
}
/*
out of bounds check during a transaction
*/
static int transaction_oob(struct tdb_context *tdb, tdb_off_t len, int probe)
{
if (len <= tdb->map_size) {
return 0;
}
tdb->ecode = TDB_ERR_IO;
return -1;
}
/*
transaction version of tdb_expand().
*/
static int transaction_expand_file(struct tdb_context *tdb, tdb_off_t size,
tdb_off_t addition)
{
/* add a write to the transaction elements, so subsequent
reads see the zero data */
if (transaction_write(tdb, size, NULL, addition) != 0) {
return -1;
}
tdb->transaction->need_repack = true;
return 0;
}
static const struct tdb_methods transaction_methods = {
transaction_read,
transaction_write,
transaction_next_hash_chain,
transaction_oob,
transaction_expand_file,
};
/*
sync to disk
*/
static int transaction_sync(struct tdb_context *tdb, tdb_off_t offset, tdb_len_t length)
{
if (tdb->flags & TDB_NOSYNC) {
return 0;
}
if (fsync(tdb->fd) != 0) {
tdb->ecode = TDB_ERR_IO;
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction: fsync failed\n"));
return -1;
}
#ifdef MS_SYNC
if (tdb->map_ptr) {
tdb_off_t moffset = offset & ~(tdb->page_size-1);
if (msync(moffset + (char *)tdb->map_ptr,
length + (offset - moffset), MS_SYNC) != 0) {
tdb->ecode = TDB_ERR_IO;
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction: msync failed - %s\n",
strerror(errno)));
return -1;
}
}
#endif
return 0;
}
static int _tdb_transaction_cancel(struct tdb_context *tdb)
{
int i, ret = 0;
if (tdb->transaction == NULL) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_cancel: no transaction\n"));
return -1;
}
if (tdb->transaction->nesting != 0) {
tdb->transaction->transaction_error = 1;
tdb->transaction->nesting--;
return 0;
}
tdb->map_size = tdb->transaction->old_map_size;
/* free all the transaction blocks */
for (i=0;i<tdb->transaction->num_blocks;i++) {
if (tdb->transaction->blocks[i] != NULL) {
free(tdb->transaction->blocks[i]);
}
}
SAFE_FREE(tdb->transaction->blocks);
if (tdb->transaction->magic_offset) {
const struct tdb_methods *methods = tdb->transaction->io_methods;
uint32_t invalid = TDB_RECOVERY_INVALID_MAGIC;
/* remove the recovery marker */
if (methods->tdb_write(tdb, tdb->transaction->magic_offset, &invalid, 4) == -1 ||
transaction_sync(tdb, tdb->transaction->magic_offset, 4) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_cancel: failed to remove recovery magic\n"));
ret = -1;
}
}
/* This also removes the OPEN_LOCK, if we have it. */
tdb_release_extra_locks(tdb);
/* restore the normal io methods */
tdb->methods = tdb->transaction->io_methods;
tdb_transaction_unlock(tdb, F_WRLCK);
SAFE_FREE(tdb->transaction->hash_heads);
SAFE_FREE(tdb->transaction);
return ret;
}
/*
start a tdb transaction. No token is returned, as only a single
transaction is allowed to be pending per tdb_context
*/
int tdb_transaction_start(struct tdb_context *tdb)
{
/* some sanity checks */
if (tdb->read_only || (tdb->flags & TDB_INTERNAL) || tdb->traverse_read) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction on a read-only or internal db\n"));
tdb->ecode = TDB_ERR_EINVAL;
return -1;
}
/* cope with nested tdb_transaction_start() calls */
if (tdb->transaction != NULL) {
if (!(tdb->flags & TDB_ALLOW_NESTING)) {
tdb->ecode = TDB_ERR_NESTING;
return -1;
}
tdb_trace(tdb, "tdb_transaction_start");
tdb->transaction->nesting++;
TDB_LOG((tdb, TDB_DEBUG_TRACE, "tdb_transaction_start: nesting %d\n",
tdb->transaction->nesting));
return 0;
}
if (tdb_have_extra_locks(tdb)) {
/* the caller must not have any locks when starting a
transaction as otherwise we'll be screwed by lack
of nested locks in posix */
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction with locks held\n"));
tdb->ecode = TDB_ERR_LOCK;
return -1;
}
if (tdb->travlocks.next != NULL) {
/* you cannot use transactions inside a traverse (although you can use
traverse inside a transaction) as otherwise you can end up with
deadlock */
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: cannot start a transaction within a traverse\n"));
tdb->ecode = TDB_ERR_LOCK;
return -1;
}
tdb->transaction = (struct tdb_transaction *)
calloc(sizeof(struct tdb_transaction), 1);
if (tdb->transaction == NULL) {
tdb->ecode = TDB_ERR_OOM;
return -1;
}
/* a page at a time seems like a reasonable compromise between compactness and efficiency */
tdb->transaction->block_size = tdb->page_size;
/* get the transaction write lock. This is a blocking lock. As
discussed with Volker, there are a number of ways we could
make this async, which we will probably do in the future */
if (tdb_transaction_lock(tdb, F_WRLCK) == -1) {
SAFE_FREE(tdb->transaction->blocks);
SAFE_FREE(tdb->transaction);
return -1;
}
/* get a read lock from the freelist to the end of file. This
is upgraded to a write lock during the commit */
if (tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, true) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_start: failed to get hash locks\n"));
goto fail_allrecord_lock;
}
/* setup a copy of the hash table heads so the hash scan in
traverse can be fast */
tdb->transaction->hash_heads = (uint32_t *)
calloc(tdb->header.hash_size+1, sizeof(uint32_t));
if (tdb->transaction->hash_heads == NULL) {
tdb->ecode = TDB_ERR_OOM;
goto fail;
}
if (tdb->methods->tdb_read(tdb, FREELIST_TOP, tdb->transaction->hash_heads,
TDB_HASHTABLE_SIZE(tdb), 0) != 0) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_start: failed to read hash heads\n"));
tdb->ecode = TDB_ERR_IO;
goto fail;
}
/* make sure we know about any file expansions already done by
anyone else */
tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
tdb->transaction->old_map_size = tdb->map_size;
/* finally hook the io methods, replacing them with
transaction specific methods */
tdb->transaction->io_methods = tdb->methods;
tdb->methods = &transaction_methods;
/* Trace at the end, so we get sequence number correct. */
tdb_trace(tdb, "tdb_transaction_start");
return 0;
fail:
tdb_allrecord_unlock(tdb, F_RDLCK, false);
fail_allrecord_lock:
tdb_transaction_unlock(tdb, F_WRLCK);
SAFE_FREE(tdb->transaction->blocks);
SAFE_FREE(tdb->transaction->hash_heads);
SAFE_FREE(tdb->transaction);
return -1;
}
/*
cancel the current transaction
*/
int tdb_transaction_cancel(struct tdb_context *tdb)
{
tdb_trace(tdb, "tdb_transaction_cancel");
return _tdb_transaction_cancel(tdb);
}
/*
work out how much space the linearised recovery data will consume
*/
static tdb_len_t tdb_recovery_size(struct tdb_context *tdb)
{
tdb_len_t recovery_size = 0;
int i;
recovery_size = sizeof(uint32_t);
for (i=0;i<tdb->transaction->num_blocks;i++) {
if (i * tdb->transaction->block_size >= tdb->transaction->old_map_size) {
break;
}
if (tdb->transaction->blocks[i] == NULL) {
continue;
}
recovery_size += 2*sizeof(tdb_off_t);
if (i == tdb->transaction->num_blocks-1) {
recovery_size += tdb->transaction->last_block_size;
} else {
recovery_size += tdb->transaction->block_size;
}
}
return recovery_size;
}
/*
allocate the recovery area, or use an existing recovery area if it is
large enough
*/
static int tdb_recovery_allocate(struct tdb_context *tdb,
tdb_len_t *recovery_size,
tdb_off_t *recovery_offset,
tdb_len_t *recovery_max_size)
{
struct tdb_record rec;
const struct tdb_methods *methods = tdb->transaction->io_methods;
tdb_off_t recovery_head, new_end;
if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to read recovery head\n"));
return -1;
}
rec.rec_len = 0;
if (recovery_head != 0) {
if (methods->tdb_read(tdb, recovery_head, &rec, sizeof(rec), DOCONV()) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to read recovery record\n"));
return -1;
}
/* ignore invalid recovery regions: can happen in crash */
if (rec.magic != TDB_RECOVERY_MAGIC &&
rec.magic != TDB_RECOVERY_INVALID_MAGIC) {
recovery_head = 0;
}
}
*recovery_size = tdb_recovery_size(tdb);
/* Existing recovery area? */
if (recovery_head != 0 && *recovery_size <= rec.rec_len) {
/* it fits in the existing area */
*recovery_max_size = rec.rec_len;
*recovery_offset = recovery_head;
return 0;
}
/* If recovery area in middle of file, we need a new one. */
if (recovery_head == 0
|| recovery_head + sizeof(rec) + rec.rec_len != tdb->map_size) {
/* we need to free up the old recovery area, then allocate a
new one at the end of the file. Note that we cannot use
tdb_allocate() to allocate the new one as that might return
us an area that is being currently used (as of the start of
the transaction) */
if (recovery_head) {
if (tdb_free(tdb, recovery_head, &rec) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL,
"tdb_recovery_allocate: failed to"
" free previous recovery area\n"));
return -1;
}
/* the tdb_free() call might have increased
* the recovery size */
*recovery_size = tdb_recovery_size(tdb);
}
/* New head will be at end of file. */
recovery_head = tdb->map_size;
}
/* Now we know where it will be. */
*recovery_offset = recovery_head;
/* Expand by more than we need, so we don't do it often. */
*recovery_max_size = tdb_expand_adjust(tdb->map_size,
*recovery_size,
tdb->page_size)
- sizeof(rec);
new_end = recovery_head + sizeof(rec) + *recovery_max_size;
if (methods->tdb_expand_file(tdb, tdb->transaction->old_map_size,
new_end - tdb->transaction->old_map_size)
== -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to create recovery area\n"));
return -1;
}
/* remap the file (if using mmap) */
methods->tdb_oob(tdb, tdb->map_size + 1, 1);
/* we have to reset the old map size so that we don't try to expand the file
again in the transaction commit, which would destroy the recovery area */
tdb->transaction->old_map_size = tdb->map_size;
/* write the recovery header offset and sync - we can sync without a race here
as the magic ptr in the recovery record has not been set */
CONVERT(recovery_head);
if (methods->tdb_write(tdb, TDB_RECOVERY_HEAD,
&recovery_head, sizeof(tdb_off_t)) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to write recovery head\n"));
return -1;
}
if (transaction_write_existing(tdb, TDB_RECOVERY_HEAD, &recovery_head, sizeof(tdb_off_t)) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_recovery_allocate: failed to write recovery head\n"));
return -1;
}
return 0;
}
/*
setup the recovery data that will be used on a crash during commit
*/
static int transaction_setup_recovery(struct tdb_context *tdb,
tdb_off_t *magic_offset)
{
tdb_len_t recovery_size;
unsigned char *data, *p;
const struct tdb_methods *methods = tdb->transaction->io_methods;
struct tdb_record *rec;
tdb_off_t recovery_offset, recovery_max_size;
tdb_off_t old_map_size = tdb->transaction->old_map_size;
uint32_t magic, tailer;
int i;
/*
check that the recovery area has enough space
*/
if (tdb_recovery_allocate(tdb, &recovery_size,
&recovery_offset, &recovery_max_size) == -1) {
return -1;
}
data = (unsigned char *)malloc(recovery_size + sizeof(*rec));
if (data == NULL) {
tdb->ecode = TDB_ERR_OOM;
return -1;
}
rec = (struct tdb_record *)data;
memset(rec, 0, sizeof(*rec));
rec->magic = TDB_RECOVERY_INVALID_MAGIC;
rec->data_len = recovery_size;
rec->rec_len = recovery_max_size;
rec->key_len = old_map_size;
CONVERT(rec);
/* build the recovery data into a single blob to allow us to do a single
large write, which should be more efficient */
p = data + sizeof(*rec);
for (i=0;i<tdb->transaction->num_blocks;i++) {
tdb_off_t offset;
tdb_len_t length;
if (tdb->transaction->blocks[i] == NULL) {
continue;
}
offset = i * tdb->transaction->block_size;
length = tdb->transaction->block_size;
if (i == tdb->transaction->num_blocks-1) {
length = tdb->transaction->last_block_size;
}
if (offset >= old_map_size) {
continue;
}
if (offset + length > tdb->transaction->old_map_size) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: transaction data over new region boundary\n"));
free(data);
tdb->ecode = TDB_ERR_CORRUPT;
return -1;
}
memcpy(p, &offset, 4);
memcpy(p+4, &length, 4);
if (DOCONV()) {
tdb_convert(p, 8);
}
/* the recovery area contains the old data, not the
new data, so we have to call the original tdb_read
method to get it */
if (methods->tdb_read(tdb, offset, p + 8, length, 0) != 0) {
free(data);
tdb->ecode = TDB_ERR_IO;
return -1;
}
p += 8 + length;
}
/* and the tailer */
tailer = sizeof(*rec) + recovery_max_size;
memcpy(p, &tailer, 4);
CONVERT(p);
/* write the recovery data to the recovery area */
if (methods->tdb_write(tdb, recovery_offset, data, sizeof(*rec) + recovery_size) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write recovery data\n"));
free(data);
tdb->ecode = TDB_ERR_IO;
return -1;
}
if (transaction_write_existing(tdb, recovery_offset, data, sizeof(*rec) + recovery_size) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write secondary recovery data\n"));
free(data);
tdb->ecode = TDB_ERR_IO;
return -1;
}
/* as we don't have ordered writes, we have to sync the recovery
data before we update the magic to indicate that the recovery
data is present */
if (transaction_sync(tdb, recovery_offset, sizeof(*rec) + recovery_size) == -1) {
free(data);
return -1;
}
free(data);
magic = TDB_RECOVERY_MAGIC;
CONVERT(magic);
*magic_offset = recovery_offset + offsetof(struct tdb_record, magic);
if (methods->tdb_write(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write recovery magic\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
if (transaction_write_existing(tdb, *magic_offset, &magic, sizeof(magic)) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_setup_recovery: failed to write secondary recovery magic\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
/* ensure the recovery magic marker is on disk */
if (transaction_sync(tdb, *magic_offset, sizeof(magic)) == -1) {
return -1;
}
return 0;
}
static int _tdb_transaction_prepare_commit(struct tdb_context *tdb)
{
const struct tdb_methods *methods;
if (tdb->transaction == NULL) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: no transaction\n"));
return -1;
}
if (tdb->transaction->prepared) {
tdb->ecode = TDB_ERR_EINVAL;
_tdb_transaction_cancel(tdb);
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: transaction already prepared\n"));
return -1;
}
if (tdb->transaction->transaction_error) {
tdb->ecode = TDB_ERR_IO;
_tdb_transaction_cancel(tdb);
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: transaction error pending\n"));
return -1;
}
if (tdb->transaction->nesting != 0) {
tdb->transaction->nesting--;
return 0;
}
/* check for a null transaction */
if (tdb->transaction->blocks == NULL) {
return 0;
}
methods = tdb->transaction->io_methods;
/* if there are any locks pending then the caller has not
nested their locks properly, so fail the transaction */
if (tdb_have_extra_locks(tdb)) {
tdb->ecode = TDB_ERR_LOCK;
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: locks pending on commit\n"));
_tdb_transaction_cancel(tdb);
return -1;
}
/* upgrade the main transaction lock region to a write lock */
if (tdb_allrecord_upgrade(tdb) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: failed to upgrade hash locks\n"));
_tdb_transaction_cancel(tdb);
return -1;
}
/* get the open lock - this prevents new users attaching to the database
during the commit */
if (tdb_nest_lock(tdb, OPEN_LOCK, F_WRLCK, TDB_LOCK_WAIT) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_prepare_commit: failed to get open lock\n"));
_tdb_transaction_cancel(tdb);
return -1;
}
if (!(tdb->flags & TDB_NOSYNC)) {
/* write the recovery data to the end of the file */
if (transaction_setup_recovery(tdb, &tdb->transaction->magic_offset) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_prepare_commit: failed to setup recovery data\n"));
_tdb_transaction_cancel(tdb);
return -1;
}
}
tdb->transaction->prepared = true;
/* expand the file to the new size if needed */
if (tdb->map_size != tdb->transaction->old_map_size) {
if (methods->tdb_expand_file(tdb, tdb->transaction->old_map_size,
tdb->map_size -
tdb->transaction->old_map_size) == -1) {
tdb->ecode = TDB_ERR_IO;
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_prepare_commit: expansion failed\n"));
_tdb_transaction_cancel(tdb);
return -1;
}
tdb->map_size = tdb->transaction->old_map_size;
methods->tdb_oob(tdb, tdb->map_size + 1, 1);
}
/* Keep the open lock until the actual commit */
return 0;
}
/*
prepare to commit the current transaction
*/
int tdb_transaction_prepare_commit(struct tdb_context *tdb)
{
tdb_trace(tdb, "tdb_transaction_prepare_commit");
return _tdb_transaction_prepare_commit(tdb);
}
/*
commit the current transaction
*/
int tdb_transaction_commit(struct tdb_context *tdb)
{
const struct tdb_methods *methods;
int i;
bool need_repack;
if (tdb->transaction == NULL) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: no transaction\n"));
return -1;
}
tdb_trace(tdb, "tdb_transaction_commit");
if (tdb->transaction->transaction_error) {
tdb->ecode = TDB_ERR_IO;
tdb_transaction_cancel(tdb);
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_transaction_commit: transaction error pending\n"));
return -1;
}
if (tdb->transaction->nesting != 0) {
tdb->transaction->nesting--;
return 0;
}
/* check for a null transaction */
if (tdb->transaction->blocks == NULL) {
_tdb_transaction_cancel(tdb);
return 0;
}
if (!tdb->transaction->prepared) {
int ret = _tdb_transaction_prepare_commit(tdb);
if (ret)
return ret;
}
methods = tdb->transaction->io_methods;
/* perform all the writes */
for (i=0;i<tdb->transaction->num_blocks;i++) {
tdb_off_t offset;
tdb_len_t length;
if (tdb->transaction->blocks[i] == NULL) {
continue;
}
offset = i * tdb->transaction->block_size;
length = tdb->transaction->block_size;
if (i == tdb->transaction->num_blocks-1) {
length = tdb->transaction->last_block_size;
}
if (methods->tdb_write(tdb, offset, tdb->transaction->blocks[i], length) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed during commit\n"));
/* we've overwritten part of the data and
possibly expanded the file, so we need to
run the crash recovery code */
tdb->methods = methods;
tdb_transaction_recover(tdb);
_tdb_transaction_cancel(tdb);
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_commit: write failed\n"));
return -1;
}
SAFE_FREE(tdb->transaction->blocks[i]);
}
SAFE_FREE(tdb->transaction->blocks);
tdb->transaction->num_blocks = 0;
/* ensure the new data is on disk */
if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
return -1;
}
/*
TODO: maybe write to some dummy hdr field, or write to magic
offset without mmap, before the last sync, instead of the
utime() call
*/
/* on some systems (like Linux 2.6.x) changes via mmap/msync
don't change the mtime of the file, this means the file may
not be backed up (as tdb rounding to block sizes means that
file size changes are quite rare too). The following forces
mtime changes when a transaction completes */
#if HAVE_UTIME
utime(tdb->name, NULL);
#endif
need_repack = tdb->transaction->need_repack;
/* use a transaction cancel to free memory and remove the
transaction locks */
_tdb_transaction_cancel(tdb);
if (need_repack) {
return tdb_repack(tdb);
}
return 0;
}
/*
recover from an aborted transaction. Must be called with exclusive
database write access already established (including the open
lock to prevent new processes attaching)
*/
int tdb_transaction_recover(struct tdb_context *tdb)
{
tdb_off_t recovery_head, recovery_eof;
unsigned char *data, *p;
uint32_t zero = 0;
struct tdb_record rec;
/* find the recovery area */
if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery head\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
if (recovery_head == 0) {
/* we have never allocated a recovery record */
return 0;
}
/* read the recovery record */
if (tdb->methods->tdb_read(tdb, recovery_head, &rec,
sizeof(rec), DOCONV()) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery record\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
if (rec.magic != TDB_RECOVERY_MAGIC) {
/* there is no valid recovery data */
return 0;
}
if (tdb->read_only) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: attempt to recover read only database\n"));
tdb->ecode = TDB_ERR_CORRUPT;
return -1;
}
recovery_eof = rec.key_len;
data = (unsigned char *)malloc(rec.data_len);
if (data == NULL) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to allocate recovery data\n"));
tdb->ecode = TDB_ERR_OOM;
return -1;
}
/* read the full recovery data */
if (tdb->methods->tdb_read(tdb, recovery_head + sizeof(rec), data,
rec.data_len, 0) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to read recovery data\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
/* recover the file data */
p = data;
while (p+8 < data + rec.data_len) {
uint32_t ofs, len;
if (DOCONV()) {
tdb_convert(p, 8);
}
memcpy(&ofs, p, 4);
memcpy(&len, p+4, 4);
if (tdb->methods->tdb_write(tdb, ofs, p+8, len) == -1) {
free(data);
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to recover %d bytes at offset %d\n", len, ofs));
tdb->ecode = TDB_ERR_IO;
return -1;
}
p += 8 + len;
}
free(data);
if (transaction_sync(tdb, 0, tdb->map_size) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to sync recovery\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
/* if the recovery area is after the recovered eof then remove it */
if (recovery_eof <= recovery_head) {
if (tdb_ofs_write(tdb, TDB_RECOVERY_HEAD, &zero) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to remove recovery head\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
}
/* remove the recovery magic */
if (tdb_ofs_write(tdb, recovery_head + offsetof(struct tdb_record, magic),
&zero) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to remove recovery magic\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
if (transaction_sync(tdb, 0, recovery_eof) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_transaction_recover: failed to sync2 recovery\n"));
tdb->ecode = TDB_ERR_IO;
return -1;
}
TDB_LOG((tdb, TDB_DEBUG_TRACE, "tdb_transaction_recover: recovered %d byte database\n",
recovery_eof));
/* all done */
return 0;
}
/* Any I/O failures we say "needs recovery". */
bool tdb_needs_recovery(struct tdb_context *tdb)
{
tdb_off_t recovery_head;
struct tdb_record rec;
/* find the recovery area */
if (tdb_ofs_read(tdb, TDB_RECOVERY_HEAD, &recovery_head) == -1) {
return true;
}
if (recovery_head == 0) {
/* we have never allocated a recovery record */
return false;
}
/* read the recovery record */
if (tdb->methods->tdb_read(tdb, recovery_head, &rec,
sizeof(rec), DOCONV()) == -1) {
return true;
}
return (rec.magic == TDB_RECOVERY_MAGIC);
}
ccan/tdb/traverse.c deleted 100644 → 0
View file @ 98b8ada2
/*
Unix SMB/CIFS implementation.
trivial database library
Copyright (C) Andrew Tridgell 1999-2005
Copyright (C) Paul `Rusty' Russell 2000
Copyright (C) Jeremy Allison 2000-2003
** NOTE! The following LGPL license applies to the tdb
** library. This does NOT imply that all of Samba is released
** under the LGPL
This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
#define TDB_NEXT_LOCK_ERR ((tdb_off_t)-1)
/* Uses traverse lock: 0 = finish, TDB_NEXT_LOCK_ERR = error,
other = record offset */
static tdb_off_t tdb_next_lock(struct tdb_context *tdb, struct tdb_traverse_lock *tlock,
struct tdb_record *rec)
{
int want_next = (tlock->off != 0);
/* Lock each chain from the start one. */
for (; tlock->hash < tdb->header.hash_size; tlock->hash++) {
if (!tlock->off && tlock->hash != 0) {
/* this is an optimization for the common case where
the hash chain is empty, which is particularly
common for the use of tdb with ldb, where large
hashes are used. In that case we spend most of our
time in tdb_brlock(), locking empty hash chains.
To avoid this, we do an unlocked pre-check to see
if the hash chain is empty before starting to look
inside it. If it is empty then we can avoid that
hash chain. If it isn't empty then we can't believe
the value we get back, as we read it without a
lock, so instead we get the lock and re-fetch the
value below.
Notice that not doing this optimization on the
first hash chain is critical. We must guarantee
that we have done at least one fcntl lock at the
start of a search to guarantee that memory is
coherent on SMP systems. If records are added by
others during the search then thats OK, and we
could possibly miss those with this trick, but we
could miss them anyway without this trick, so the
semantics don't change.
With a non-indexed ldb search this trick gains us a
factor of around 80 in speed on a linux 2.6.x
system (testing using ldbtest).
*/
tdb->methods->next_hash_chain(tdb, &tlock->hash);
if (tlock->hash == tdb->header.hash_size) {
continue;
}
}
if (tdb_lock(tdb, tlock->hash, tlock->lock_rw) == -1)
return TDB_NEXT_LOCK_ERR;
/* No previous record? Start at top of chain. */
if (!tlock->off) {
if (tdb_ofs_read(tdb, TDB_HASH_TOP(tlock->hash),
&tlock->off) == -1)
goto fail;
} else {
/* Otherwise unlock the previous record. */
if (tdb_unlock_record(tdb, tlock->off) != 0)
goto fail;
}
if (want_next) {
/* We have offset of old record: grab next */
if (tdb_rec_read(tdb, tlock->off, rec) == -1)
goto fail;
tlock->off = rec->next;
}
/* Iterate through chain */
while( tlock->off) {
tdb_off_t current;
if (tdb_rec_read(tdb, tlock->off, rec) == -1)
goto fail;
/* Detect infinite loops. From "Shlomi Yaakobovich" <Shlomi@exanet.com>. */
if (tlock->off == rec->next) {
tdb->ecode = TDB_ERR_CORRUPT;
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_next_lock: loop detected.\n"));
goto fail;
}
if (!TDB_DEAD(rec)) {
/* Woohoo: we found one! */
if (tdb_lock_record(tdb, tlock->off) != 0)
goto fail;
return tlock->off;
}
/* Try to clean dead ones from old traverses */
current = tlock->off;
tlock->off = rec->next;
if (!(tdb->read_only || tdb->traverse_read) &&
tdb_do_delete(tdb, current, rec) != 0)
goto fail;
}
tdb_unlock(tdb, tlock->hash, tlock->lock_rw);
want_next = 0;
}
/* We finished iteration without finding anything */
tdb->ecode = TDB_SUCCESS;
return 0;
fail:
tlock->off = 0;
if (tdb_unlock(tdb, tlock->hash, tlock->lock_rw) != 0)
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_next_lock: On error unlock failed!\n"));
return TDB_NEXT_LOCK_ERR;
}
/* traverse the entire database - calling fn(tdb, key, data) on each element.
return -1 on error or the record count traversed
if fn is NULL then it is not called
a non-zero return value from fn() indicates that the traversal should stop
*/
static int tdb_traverse_internal(struct tdb_context *tdb,
tdb_traverse_func fn, void *private_data,
struct tdb_traverse_lock *tl)
{
TDB_DATA key, dbuf;
struct tdb_record rec;
int ret = 0, count = 0;
tdb_off_t off;
/* This was in the initializaton, above, but the IRIX compiler
* did not like it. crh
*/
tl->next = tdb->travlocks.next;
/* fcntl locks don't stack: beware traverse inside traverse */
tdb->travlocks.next = tl;
/* tdb_next_lock places locks on the record returned, and its chain */
while ((off = tdb_next_lock(tdb, tl, &rec)) != 0) {
if (off == TDB_NEXT_LOCK_ERR) {
ret = -1;
goto out;
}
count++;
/* now read the full record */
key.dptr = tdb_alloc_read(tdb, tl->off + sizeof(rec),
rec.key_len + rec.data_len);
if (!key.dptr) {
ret = -1;
if (tdb_unlock(tdb, tl->hash, tl->lock_rw) != 0)
goto out;
if (tdb_unlock_record(tdb, tl->off) != 0)
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_traverse: key.dptr == NULL and unlock_record failed!\n"));
goto out;
}
key.dsize = rec.key_len;
dbuf.dptr = key.dptr + rec.key_len;
dbuf.dsize = rec.data_len;
tdb_trace_1rec_retrec(tdb, "traverse", key, dbuf);
/* Drop chain lock, call out */
if (tdb_unlock(tdb, tl->hash, tl->lock_rw) != 0) {
ret = -1;
SAFE_FREE(key.dptr);
goto out;
}
if (fn && fn(tdb, key, dbuf, private_data)) {
/* They want us to terminate traversal */
tdb_trace_ret(tdb, "tdb_traverse_end", count);
if (tdb_unlock_record(tdb, tl->off) != 0) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_traverse: unlock_record failed!\n"));;
ret = -1;
}
SAFE_FREE(key.dptr);
goto out;
}
SAFE_FREE(key.dptr);
}
tdb_trace(tdb, "tdb_traverse_end");
out:
tdb->travlocks.next = tl->next;
if (ret < 0)
return -1;
else
return count;
}
/*
a write style traverse - temporarily marks the db read only
*/
int tdb_traverse_read(struct tdb_context *tdb,
tdb_traverse_func fn, void *private_data)
{
struct tdb_traverse_lock tl = { NULL, 0, 0, F_RDLCK };
int ret;
/* we need to get a read lock on the transaction lock here to
cope with the lock ordering semantics of solaris10 */
if (tdb_transaction_lock(tdb, F_RDLCK)) {
return -1;
}
tdb->traverse_read++;
tdb_trace(tdb, "tdb_traverse_read_start");
ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
tdb->traverse_read--;
tdb_transaction_unlock(tdb, F_RDLCK);
return ret;
}
/*
a write style traverse - needs to get the transaction lock to
prevent deadlocks
WARNING: The data buffer given to the callback fn does NOT meet the
alignment restrictions malloc gives you.
*/
int tdb_traverse(struct tdb_context *tdb,
tdb_traverse_func fn, void *private_data)
{
struct tdb_traverse_lock tl = { NULL, 0, 0, F_WRLCK };
int ret;
if (tdb->read_only || tdb->traverse_read) {
return tdb_traverse_read(tdb, fn, private_data);
}
if (tdb_transaction_lock(tdb, F_WRLCK)) {
return -1;
}
tdb->traverse_write++;
tdb_trace(tdb, "tdb_traverse_start");
ret = tdb_traverse_internal(tdb, fn, private_data, &tl);
tdb->traverse_write--;
tdb_transaction_unlock(tdb, F_WRLCK);
return ret;
}
/* find the first entry in the database and return its key */
TDB_DATA tdb_firstkey(struct tdb_context *tdb)
{
TDB_DATA key;
struct tdb_record rec;
tdb_off_t off;
/* release any old lock */
if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0)
return tdb_null;
tdb->travlocks.off = tdb->travlocks.hash = 0;
tdb->travlocks.lock_rw = F_RDLCK;
/* Grab first record: locks chain and returned record. */
off = tdb_next_lock(tdb, &tdb->travlocks, &rec);
if (off == 0 || off == TDB_NEXT_LOCK_ERR) {
tdb_trace_retrec(tdb, "tdb_firstkey", tdb_null);
return tdb_null;
}
/* now read the key */
key.dsize = rec.key_len;
key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
tdb_trace_retrec(tdb, "tdb_firstkey", key);
/* Unlock the hash chain of the record we just read. */
if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0)
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_firstkey: error occurred while tdb_unlocking!\n"));
return key;
}
/* find the next entry in the database, returning its key */
TDB_DATA tdb_nextkey(struct tdb_context *tdb, TDB_DATA oldkey)
{
uint32_t oldhash;
TDB_DATA key = tdb_null;
struct tdb_record rec;
unsigned char *k = NULL;
tdb_off_t off;
/* Is locked key the old key? If so, traverse will be reliable. */
if (tdb->travlocks.off) {
if (tdb_lock(tdb,tdb->travlocks.hash,tdb->travlocks.lock_rw))
return tdb_null;
if (tdb_rec_read(tdb, tdb->travlocks.off, &rec) == -1
|| !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
rec.key_len))
|| memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
/* No, it wasn't: unlock it and start from scratch */
if (tdb_unlock_record(tdb, tdb->travlocks.off) != 0) {
tdb_trace_1rec_retrec(tdb, "tdb_nextkey",
oldkey, tdb_null);
SAFE_FREE(k);
return tdb_null;
}
if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0) {
SAFE_FREE(k);
return tdb_null;
}
tdb->travlocks.off = 0;
}
SAFE_FREE(k);
}
if (!tdb->travlocks.off) {
/* No previous element: do normal find, and lock record */
tdb->travlocks.off = tdb_find_lock_hash(tdb, oldkey, tdb->hash_fn(&oldkey), tdb->travlocks.lock_rw, &rec);
if (!tdb->travlocks.off) {
tdb_trace_1rec_retrec(tdb, "tdb_nextkey", oldkey, tdb_null);
return tdb_null;
}
tdb->travlocks.hash = BUCKET(rec.full_hash);
if (tdb_lock_record(tdb, tdb->travlocks.off) != 0) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: lock_record failed (%s)!\n", strerror(errno)));
return tdb_null;
}
}
oldhash = tdb->travlocks.hash;
/* Grab next record: locks chain and returned record,
unlocks old record */
off = tdb_next_lock(tdb, &tdb->travlocks, &rec);
if (off != TDB_NEXT_LOCK_ERR && off != 0) {
key.dsize = rec.key_len;
key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
key.dsize);
/* Unlock the chain of this new record */
if (tdb_unlock(tdb, tdb->travlocks.hash, tdb->travlocks.lock_rw) != 0)
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
}
/* Unlock the chain of old record */
if (tdb_unlock(tdb, BUCKET(oldhash), tdb->travlocks.lock_rw) != 0)
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_nextkey: WARNING tdb_unlock failed!\n"));
tdb_trace_1rec_retrec(tdb, "tdb_nextkey", oldkey, key);
return key;
}
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7