Commit afd64673 authored by Andrew Morton's avatar Andrew Morton Committed by Linus Torvalds

[PATCH] reiserfs: reiserfs_file_write implementation

From: Oleg Drokin <green@namesys.com>

With the current 'one block at a time' algorithm, writes past the end of a
file are slow because each new file block is separately added into the tree
causing shifting of other items which is CPU expensive.

With this new implementation if you write into file with big enough chunks,
it uses half as much CPU.  Also this version is more SMP friendly than the
current one.

There are some known-bad applications that break with this patch (ie.  start
to work very slow or even hang).

This is because the filesystem returns a large value in the stat.st_blocksize
hint (128k instead of 4k).  This tickles a small number of application bugs.
One is KDE's kmail 3.04 (fixed by upgrading to 3.1+) and the other is
sleepycat's database from before 1997.

If you hit a slowdown problem that you believe is related to the increased
"recommended i/o size" value, try to mount your fs with nolargeio=1 mount
option (remount should work too).

This patch exports block_commit_write(), generic_osync_inode() and
remove_suid() to modules.
parent a61638bc
......@@ -9,6 +9,7 @@
#include <linux/errno.h>
#include <linux/buffer_head.h>
#include <linux/kernel.h>
#include <linux/pagemap.h>
#include <linux/reiserfs_fs_sb.h>
#include <linux/reiserfs_fs_i.h>
......@@ -733,7 +734,7 @@ static inline int allocate_without_wrapping_disk (reiserfs_blocknr_hint_t * hint
int rest = amount_needed;
int nr_allocated;
while (rest > 0) {
while (rest > 0 && start <= finish) {
nr_allocated = scan_bitmap (hint->th, &start, finish, 1,
rest + prealloc_size, !hint->formatted_node,
hint->block);
......@@ -879,7 +880,9 @@ void reiserfs_claim_blocks_to_be_allocated(
if ( !blocks )
return;
spin_lock(&REISERFS_SB(sb)->bitmap_lock);
REISERFS_SB(sb)->reserved_blocks += blocks;
spin_unlock(&REISERFS_SB(sb)->bitmap_lock);
}
/* Unreserve @blocks amount of blocks in fs pointed by @sb */
......@@ -896,6 +899,22 @@ void reiserfs_release_claimed_blocks(
if ( !blocks )
return;
spin_lock(&REISERFS_SB(sb)->bitmap_lock);
REISERFS_SB(sb)->reserved_blocks -= blocks;
spin_unlock(&REISERFS_SB(sb)->bitmap_lock);
RFALSE( REISERFS_SB(sb)->reserved_blocks < 0, "amount of blocks reserved became zero?");
}
/* This function estimates how much pages we will be able to write to FS
used for reiserfs_file_write() purposes for now. */
int reiserfs_can_fit_pages ( struct super_block *sb /* superblock of filesystem
to estimate space */ )
{
unsigned long space;
spin_lock(&REISERFS_SB(sb)->bitmap_lock);
space = (SB_FREE_BLOCKS(sb) - REISERFS_SB(sb)->reserved_blocks) >> ( PAGE_CACHE_SHIFT - sb->s_blocksize_bits);
spin_unlock(&REISERFS_SB(sb)->bitmap_lock);
return space;
}
......@@ -6,6 +6,8 @@
#include <linux/time.h>
#include <linux/reiserfs_fs.h>
#include <linux/smp_lock.h>
#include <asm/uaccess.h>
#include <linux/pagemap.h>
/*
** We pack the tails of files on file close, not at the time they are written.
......@@ -140,9 +142,1018 @@ static int reiserfs_setattr(struct dentry *dentry, struct iattr *attr) {
return error ;
}
/* I really do not want to play with memory shortage right now, so
to simplify the code, we are not going to write more than this much pages at
a time. This still should considerably improve performance compared to 4k
at a time case. This is 32 pages of 4k size. */
#define REISERFS_WRITE_PAGES_AT_A_TIME (128 * 1024) / PAGE_CACHE_SIZE
/* Allocates blocks for a file to fulfil write request.
Maps all unmapped but prepared pages from the list.
Updates metadata with newly allocated blocknumbers as needed */
int reiserfs_allocate_blocks_for_region(
struct inode *inode, /* Inode we work with */
loff_t pos, /* Writing position */
int num_pages, /* number of pages write going
to touch */
int write_bytes, /* amount of bytes to write */
struct page **prepared_pages, /* array of
prepared pages
*/
int blocks_to_allocate /* Amount of blocks we
need to allocate to
fit the data into file
*/
)
{
struct cpu_key key; // cpu key of item that we are going to deal with
struct item_head *ih; // pointer to item head that we are going to deal with
struct buffer_head *bh; // Buffer head that contains items that we are going to deal with
struct reiserfs_transaction_handle th; // transaction handle for transaction we are going to create.
__u32 * item; // pointer to item we are going to deal with
INITIALIZE_PATH(path); // path to item, that we are going to deal with.
b_blocknr_t allocated_blocks[blocks_to_allocate]; // Pointer to a place where allocated blocknumbers would be stored. Right now statically allocated, later that will change.
reiserfs_blocknr_hint_t hint; // hint structure for block allocator.
size_t res; // return value of various functions that we call.
int curr_block; // current block used to keep track of unmapped blocks.
int i; // loop counter
int itempos; // position in item
unsigned int from = (pos & (PAGE_CACHE_SIZE - 1)); // writing position in
// first page
unsigned int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1; /* last modified byte offset in last page */
__u64 hole_size ; // amount of blocks for a file hole, if it needed to be created.
int modifying_this_item = 0; // Flag for items traversal code to keep track
// of the fact that we already prepared
// current block for journal
RFALSE(!blocks_to_allocate, "green-9004: tried to allocate zero blocks?");
/* First we compose a key to point at the writing position, we want to do
that outside of any locking region. */
make_cpu_key (&key, inode, pos+1, TYPE_ANY, 3/*key length*/);
/* If we came here, it means we absolutely need to open a transaction,
since we need to allocate some blocks */
reiserfs_write_lock(inode->i_sb); // Journaling stuff and we need that.
journal_begin(&th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 3 + 1); // Wish I know if this number enough
reiserfs_update_inode_transaction(inode) ;
/* Look for the in-tree position of our write, need path for block allocator */
res = search_for_position_by_key(inode->i_sb, &key, &path);
if ( res == IO_ERROR ) {
res = -EIO;
goto error_exit;
}
/* Allocate blocks */
/* First fill in "hint" structure for block allocator */
hint.th = &th; // transaction handle.
hint.path = &path; // Path, so that block allocator can determine packing locality or whatever it needs to determine.
hint.inode = inode; // Inode is needed by block allocator too.
hint.search_start = 0; // We have no hint on where to search free blocks for block allocator.
hint.key = key.on_disk_key; // on disk key of file.
hint.block = inode->i_blocks>>(inode->i_sb->s_blocksize_bits-9); // Number of disk blocks this file occupies already.
hint.formatted_node = 0; // We are allocating blocks for unformatted node.
hint.preallocate = 0; // We do not do any preallocation for now.
/* Call block allocator to allocate blocks */
res = reiserfs_allocate_blocknrs(&hint, allocated_blocks, blocks_to_allocate, blocks_to_allocate);
if ( res != CARRY_ON ) {
if ( res == NO_DISK_SPACE ) {
/* We flush the transaction in case of no space. This way some
blocks might become free */
SB_JOURNAL(inode->i_sb)->j_must_wait = 1;
restart_transaction(&th, inode, &path);
/* We might have scheduled, so search again */
res = search_for_position_by_key(inode->i_sb, &key, &path);
if ( res == IO_ERROR ) {
res = -EIO;
goto error_exit;
}
/* update changed info for hint structure. */
res = reiserfs_allocate_blocknrs(&hint, allocated_blocks, blocks_to_allocate, blocks_to_allocate);
if ( res != CARRY_ON ) {
res = -ENOSPC;
pathrelse(&path);
goto error_exit;
}
} else {
res = -ENOSPC;
pathrelse(&path);
goto error_exit;
}
}
#ifdef __BIG_ENDIAN
// Too bad, I have not found any way to convert a given region from
// cpu format to little endian format
{
int i;
for ( i = 0; i < blocks_to_allocate ; i++)
allocated_blocks[i]=cpu_to_le32(allocated_blocks[i]);
}
#endif
/* Blocks allocating well might have scheduled and tree might have changed,
let's search the tree again */
/* find where in the tree our write should go */
res = search_for_position_by_key(inode->i_sb, &key, &path);
if ( res == IO_ERROR ) {
res = -EIO;
goto error_exit_free_blocks;
}
bh = get_last_bh( &path ); // Get a bufferhead for last element in path.
ih = get_ih( &path ); // Get a pointer to last item head in path.
item = get_item( &path ); // Get a pointer to last item in path
/* Let's see what we have found */
if ( res != POSITION_FOUND ) { /* position not found, this means that we
might need to append file with holes
first */
// Since we are writing past the file's end, we need to find out if
// there is a hole that needs to be inserted before our writing
// position, and how many blocks it is going to cover (we need to
// populate pointers to file blocks representing the hole with zeros)
hole_size = (pos + 1 - (le_key_k_offset( get_inode_item_key_version(inode), &(ih->ih_key))+op_bytes_number(ih, inode->i_sb->s_blocksize))) >> inode->i_sb->s_blocksize_bits;
if ( hole_size > 0 ) {
int to_paste = min_t(__u64, hole_size, MAX_ITEM_LEN(inode->i_sb->s_blocksize)/UNFM_P_SIZE ); // How much data to insert first time.
/* area filled with zeroes, to supply as list of zero blocknumbers
We allocate it outside of loop just in case loop would spin for
several iterations. */
char *zeros = kmalloc(to_paste*UNFM_P_SIZE, GFP_ATOMIC); // We cannot insert more than MAX_ITEM_LEN bytes anyway.
if ( !zeros ) {
res = -ENOMEM;
goto error_exit_free_blocks;
}
memset ( zeros, 0, to_paste*UNFM_P_SIZE);
do {
to_paste = min_t(__u64, hole_size, MAX_ITEM_LEN(inode->i_sb->s_blocksize)/UNFM_P_SIZE );
if ( is_indirect_le_ih(ih) ) {
/* Ok, there is existing indirect item already. Need to append it */
/* Calculate position past inserted item */
make_cpu_key( &key, inode, le_key_k_offset( get_inode_item_key_version(inode), &(ih->ih_key)) + op_bytes_number(ih, inode->i_sb->s_blocksize), TYPE_INDIRECT, 3);
res = reiserfs_paste_into_item( &th, &path, &key, (char *)zeros, UNFM_P_SIZE*to_paste);
if ( res ) {
kfree(zeros);
goto error_exit_free_blocks;
}
} else if ( is_statdata_le_ih(ih) ) {
/* No existing item, create it */
/* item head for new item */
struct item_head ins_ih;
/* create a key for our new item */
make_cpu_key( &key, inode, 1, TYPE_INDIRECT, 3);
/* Create new item head for our new item */
make_le_item_head (&ins_ih, &key, key.version, 1,
TYPE_INDIRECT, to_paste*UNFM_P_SIZE,
0 /* free space */);
/* Find where such item should live in the tree */
res = search_item (inode->i_sb, &key, &path);
if ( res != ITEM_NOT_FOUND ) {
/* item should not exist, otherwise we have error */
if ( res != -ENOSPC ) {
reiserfs_warning ("green-9008: search_by_key (%K) returned %d\n",
&key, res);
}
res = -EIO;
kfree(zeros);
goto error_exit_free_blocks;
}
res = reiserfs_insert_item( &th, &path, &key, &ins_ih, (char *)zeros);
} else {
reiserfs_panic(inode->i_sb, "green-9011: Unexpected key type %K\n", &key);
}
if ( res ) {
kfree(zeros);
goto error_exit_free_blocks;
}
/* Now we want to check if transaction is too full, and if it is
we restart it. This will also free the path. */
if (journal_transaction_should_end(&th, th.t_blocks_allocated))
restart_transaction(&th, inode, &path);
/* Well, need to recalculate path and stuff */
set_cpu_key_k_offset( &key, cpu_key_k_offset(&key) + (to_paste << inode->i_blkbits));
res = search_for_position_by_key(inode->i_sb, &key, &path);
if ( res == IO_ERROR ) {
res = -EIO;
kfree(zeros);
goto error_exit_free_blocks;
}
bh=get_last_bh(&path);
ih=get_ih(&path);
item = get_item(&path);
hole_size -= to_paste;
} while ( hole_size );
kfree(zeros);
}
}
// Go through existing indirect items first
// replace all zeroes with blocknumbers from list
// Note that if no corresponding item was found, by previous search,
// it means there are no existing in-tree representation for file area
// we are going to overwrite, so there is nothing to scan through for holes.
for ( curr_block = 0, itempos = path.pos_in_item ; curr_block < blocks_to_allocate && res == POSITION_FOUND ; ) {
if ( itempos >= ih_item_len(ih)/UNFM_P_SIZE ) {
/* We run out of data in this indirect item, let's look for another
one. */
/* First if we are already modifying current item, log it */
if ( modifying_this_item ) {
journal_mark_dirty (&th, inode->i_sb, bh);
modifying_this_item = 0;
}
/* Then set the key to look for a new indirect item (offset of old
item is added to old item length */
set_cpu_key_k_offset( &key, le_key_k_offset( get_inode_item_key_version(inode), &(ih->ih_key)) + op_bytes_number(ih, inode->i_sb->s_blocksize));
/* Search ofor position of new key in the tree. */
res = search_for_position_by_key(inode->i_sb, &key, &path);
if ( res == IO_ERROR) {
res = -EIO;
goto error_exit_free_blocks;
}
bh=get_last_bh(&path);
ih=get_ih(&path);
item = get_item(&path);
itempos = path.pos_in_item;
continue; // loop to check all kinds of conditions and so on.
}
/* Ok, we have correct position in item now, so let's see if it is
representing file hole (blocknumber is zero) and fill it if needed */
if ( !item[itempos] ) {
/* Ok, a hole. Now we need to check if we already prepared this
block to be journaled */
while ( !modifying_this_item ) { // loop until succeed
/* Well, this item is not journaled yet, so we must prepare
it for journal first, before we can change it */
struct item_head tmp_ih; // We copy item head of found item,
// here to detect if fs changed under
// us while we were preparing for
// journal.
int fs_gen; // We store fs generation here to find if someone
// changes fs under our feet
copy_item_head (&tmp_ih, ih); // Remember itemhead
fs_gen = get_generation (inode->i_sb); // remember fs generation
reiserfs_prepare_for_journal(inode->i_sb, bh, 1); // Prepare a buffer within which indirect item is stored for changing.
if (fs_changed (fs_gen, inode->i_sb) && item_moved (&tmp_ih, &path)) {
// Sigh, fs was changed under us, we need to look for new
// location of item we are working with
/* unmark prepaerd area as journaled and search for it's
new position */
reiserfs_restore_prepared_buffer(inode->i_sb, bh);
res = search_for_position_by_key(inode->i_sb, &key, &path);
if ( res == IO_ERROR) {
res = -EIO;
goto error_exit_free_blocks;
}
bh=get_last_bh(&path);
ih=get_ih(&path);
item = get_item(&path);
// Itempos is still the same
continue;
}
modifying_this_item = 1;
}
item[itempos] = allocated_blocks[curr_block]; // Assign new block
curr_block++;
}
itempos++;
}
if ( modifying_this_item ) { // We need to log last-accessed block, if it
// was modified, but not logged yet.
journal_mark_dirty (&th, inode->i_sb, bh);
}
if ( curr_block < blocks_to_allocate ) {
// Oh, well need to append to indirect item, or to create indirect item
// if there weren't any
if ( is_indirect_le_ih(ih) ) {
// Existing indirect item - append. First calculate key for append
// position. We do not need to recalculate path as it should
// already point to correct place.
make_cpu_key( &key, inode, le_key_k_offset( get_inode_item_key_version(inode), &(ih->ih_key)) + op_bytes_number(ih, inode->i_sb->s_blocksize), TYPE_INDIRECT, 3);
res = reiserfs_paste_into_item( &th, &path, &key, (char *)(allocated_blocks+curr_block), UNFM_P_SIZE*(blocks_to_allocate-curr_block));
if ( res ) {
goto error_exit_free_blocks;
}
} else if (is_statdata_le_ih(ih) ) {
// Last found item was statdata. That means we need to create indirect item.
struct item_head ins_ih; /* itemhead for new item */
/* create a key for our new item */
make_cpu_key( &key, inode, 1, TYPE_INDIRECT, 3); // Position one,
// because that's
// where first
// indirect item
// begins
/* Create new item head for our new item */
make_le_item_head (&ins_ih, &key, key.version, 1, TYPE_INDIRECT,
(blocks_to_allocate-curr_block)*UNFM_P_SIZE,
0 /* free space */);
/* Find where such item should live in the tree */
res = search_item (inode->i_sb, &key, &path);
if ( res != ITEM_NOT_FOUND ) {
/* Well, if we have found such item already, or some error
occured, we need to warn user and return error */
if ( res != -ENOSPC ) {
reiserfs_warning ("green-9009: search_by_key (%K) returned %d\n",
&key, res);
}
res = -EIO;
goto error_exit_free_blocks;
}
/* Insert item into the tree with the data as its body */
res = reiserfs_insert_item( &th, &path, &key, &ins_ih, (char *)(allocated_blocks+curr_block));
} else {
reiserfs_panic(inode->i_sb, "green-9010: unexpected item type for key %K\n",&key);
}
}
/* Now the final thing, if we have grew the file, we must update it's size*/
if ( pos + write_bytes > inode->i_size) {
inode->i_size = pos + write_bytes; // Set new size
}
/* Amount of on-disk blocks used by file have changed, update it */
inode->i_blocks += blocks_to_allocate << (inode->i_blkbits - 9);
reiserfs_update_sd(&th, inode); // And update on-disk metadata
// finish all journal stuff now, We are not going to play with metadata
// anymore.
pathrelse(&path);
journal_end(&th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 3 + 1);
reiserfs_write_unlock(inode->i_sb);
// go through all the pages/buffers and map the buffers to newly allocated
// blocks (so that system knows where to write these pages later).
curr_block = 0;
for ( i = 0; i < num_pages ; i++ ) {
struct page *page=prepared_pages[i]; //current page
struct buffer_head *head = page_buffers(page);// first buffer for a page
int block_start, block_end; // in-page offsets for buffers.
if (!page_buffers(page))
reiserfs_panic(inode->i_sb, "green-9005: No buffers for prepared page???");
/* For each buffer in page */
for(bh = head, block_start = 0; bh != head || !block_start;
block_start=block_end, bh = bh->b_this_page) {
if (!bh)
reiserfs_panic(inode->i_sb, "green-9006: Allocated but absent buffer for a page?");
block_end = block_start+inode->i_sb->s_blocksize;
if (i == 0 && block_end <= from )
/* if this buffer is before requested data to map, skip it */
continue;
if (i == num_pages - 1 && block_start >= to)
/* If this buffer is after requested data to map, abort
processing of current page */
break;
if ( !buffer_mapped(bh) ) { // Ok, unmapped buffer, need to map it
map_bh( bh, inode->i_sb, le32_to_cpu(allocated_blocks[curr_block]));
curr_block++;
}
}
}
RFALSE( curr_block > blocks_to_allocate, "green-9007: Used too many blocks? weird");
return 0;
// Need to deal with transaction here.
error_exit_free_blocks:
pathrelse(&path);
// free blocks
for( i = 0; i < blocks_to_allocate; i++ )
reiserfs_free_block( &th, le32_to_cpu(allocated_blocks[i]));
error_exit:
journal_end(&th, inode->i_sb, JOURNAL_PER_BALANCE_CNT * 3 + 1);
reiserfs_write_unlock(inode->i_sb);
return res;
}
/* Unlock pages prepared by reiserfs_prepare_file_region_for_write */
void reiserfs_unprepare_pages(struct page **prepared_pages, /* list of locked pages */
int num_pages /* amount of pages */) {
int i; // loop counter
for (i=0; i < num_pages ; i++) {
struct page *page = prepared_pages[i];
try_to_free_buffers(page);
kunmap(page);
unlock_page(page);
page_cache_release(page);
}
}
/* This function will copy data from userspace to specified pages within
supplied byte range */
int reiserfs_copy_from_user_to_file_region(
loff_t pos, /* In-file position */
int num_pages, /* Number of pages affected */
int write_bytes, /* Amount of bytes to write */
struct page **prepared_pages, /* pointer to
array to
prepared pages
*/
const char *buf /* Pointer to user-supplied
data*/
)
{
long page_fault=0; // status of copy_from_user.
int i; // loop counter.
int offset; // offset in page
for ( i = 0, offset = (pos & (PAGE_CACHE_SIZE-1)); i < num_pages ; i++,offset=0) {
int count = min_t(int,PAGE_CACHE_SIZE-offset,write_bytes); // How much of bytes to write to this page
struct page *page=prepared_pages[i]; // Current page we process.
fault_in_pages_readable( buf, count);
/* Copy data from userspace to the current page */
kmap(page);
page_fault = __copy_from_user(page_address(page)+offset, buf, count); // Copy the data.
/* Flush processor's dcache for this page */
flush_dcache_page(page);
kunmap(page);
buf+=count;
write_bytes-=count;
if (page_fault)
break; // Was there a fault? abort.
}
return page_fault?-EFAULT:0;
}
/* Submit pages for write. This was separated from actual file copying
because we might want to allocate block numbers in-between.
This function assumes that caller will adjust file size to correct value. */
int reiserfs_submit_file_region_for_write(
loff_t pos, /* Writing position offset */
int num_pages, /* Number of pages to write */
int write_bytes, /* number of bytes to write */
struct page **prepared_pages /* list of pages */
)
{
int status; // return status of block_commit_write.
int retval = 0; // Return value we are going to return.
int i; // loop counter
int offset; // Writing offset in page.
for ( i = 0, offset = (pos & (PAGE_CACHE_SIZE-1)); i < num_pages ; i++,offset=0) {
int count = min_t(int,PAGE_CACHE_SIZE-offset,write_bytes); // How much of bytes to write to this page
struct page *page=prepared_pages[i]; // Current page we process.
status = block_commit_write(page, offset, offset+count);
if ( status )
retval = status; // To not overcomplicate matters We are going to
// submit all the pages even if there was error.
// we only remember error status to report it on
// exit.
write_bytes-=count;
SetPageReferenced(page);
unlock_page(page); // We unlock the page as it was locked by earlier call
// to grab_cache_page
page_cache_release(page);
}
return retval;
}
/* Look if passed writing region is going to touch file's tail
(if it is present). And if it is, convert the tail to unformatted node */
int reiserfs_check_for_tail_and_convert( struct inode *inode, /* inode to deal with */
loff_t pos, /* Writing position */
int write_bytes /* amount of bytes to write */
)
{
INITIALIZE_PATH(path); // needed for search_for_position
struct cpu_key key; // Key that would represent last touched writing byte.
struct item_head *ih; // item header of found block;
int res; // Return value of various functions we call.
int cont_expand_offset; // We will put offset for generic_cont_expand here
// This can be int just because tails are created
// only for small files.
/* this embodies a dependency on a particular tail policy */
if ( inode->i_size >= inode->i_sb->s_blocksize*4 ) {
/* such a big files do not have tails, so we won't bother ourselves
to look for tails, simply return */
return 0;
}
reiserfs_write_lock(inode->i_sb);
/* find the item containing the last byte to be written, or if
* writing past the end of the file then the last item of the
* file (and then we check its type). */
make_cpu_key (&key, inode, pos+write_bytes+1, TYPE_ANY, 3/*key length*/);
res = search_for_position_by_key(inode->i_sb, &key, &path);
if ( res == IO_ERROR ) {
reiserfs_write_unlock(inode->i_sb);
return -EIO;
}
ih = get_ih(&path);
res = 0;
if ( is_direct_le_ih(ih) ) {
/* Ok, closest item is file tail (tails are stored in "direct"
* items), so we need to unpack it. */
/* To not overcomplicate matters, we just call generic_cont_expand
which will in turn call other stuff and finally will boil down to
reiserfs_get_block() that would do necessary conversion. */
cont_expand_offset = le_key_k_offset(get_inode_item_key_version(inode), &(ih->ih_key));
pathrelse(&path);
res = generic_cont_expand( inode, cont_expand_offset);
} else
pathrelse(&path);
reiserfs_write_unlock(inode->i_sb);
return res;
}
/* This function locks pages starting from @pos for @inode.
@num_pages pages are locked and stored in
@prepared_pages array. Also buffers are allocated for these pages.
First and last page of the region is read if it is overwritten only
partially. If last page did not exist before write (file hole or file
append), it is zeroed, then.
Returns number of unallocated blocks that should be allocated to cover
new file data.*/
int reiserfs_prepare_file_region_for_write(
struct inode *inode /* Inode of the file */,
loff_t pos, /* position in the file */
int num_pages, /* number of pages to
prepare */
int write_bytes, /* Amount of bytes to be
overwritten from
@pos */
struct page **prepared_pages /* pointer to array
where to store
prepared pages */
)
{
int res=0; // Return values of different functions we call.
unsigned long index = pos >> PAGE_CACHE_SHIFT; // Offset in file in pages.
int from = (pos & (PAGE_CACHE_SIZE - 1)); // Writing offset in first page
int to = ((pos + write_bytes - 1) & (PAGE_CACHE_SIZE - 1)) + 1;
/* offset of last modified byte in last
page */
struct address_space *mapping = inode->i_mapping; // Pages are mapped here.
int i; // Simple counter
int blocks = 0; /* Return value (blocks that should be allocated) */
struct buffer_head *bh, *head; // Current bufferhead and first bufferhead
// of a page.
unsigned block_start, block_end; // Starting and ending offsets of current
// buffer in the page.
struct buffer_head *wait[2], **wait_bh=wait; // Buffers for page, if
// Page appeared to be not up
// to date. Note how we have
// at most 2 buffers, this is
// because we at most may
// partially overwrite two
// buffers for one page. One at // the beginning of write area
// and one at the end.
// Everything inthe middle gets // overwritten totally.
struct cpu_key key; // cpu key of item that we are going to deal with
struct item_head *ih = NULL; // pointer to item head that we are going to deal with
struct buffer_head *itembuf=NULL; // Buffer head that contains items that we are going to deal with
INITIALIZE_PATH(path); // path to item, that we are going to deal with.
__u32 * item=0; // pointer to item we are going to deal with
if ( num_pages < 1 ) {
reiserfs_warning("green-9001: reiserfs_prepare_file_region_for_write called with zero number of pages to process\n");
return -EFAULT;
}
/* We have 2 loops for pages. In first loop we grab and lock the pages, so
that nobody would touch these until we release the pages. Then
we'd start to deal with mapping buffers to blocks. */
for ( i = 0; i < num_pages; i++) {
prepared_pages[i] = grab_cache_page(mapping, index + i); // locks the page
if ( !prepared_pages[i]) {
res = -ENOMEM;
goto failed_page_grabbing;
}
if (!page_has_buffers(prepared_pages[i]))
create_empty_buffers(prepared_pages[i], inode->i_sb->s_blocksize, 0);
}
/* Let's count amount of blocks for a case where all the blocks
overwritten are new (we will substract already allocated blocks later)*/
if ( num_pages > 2 )
/* These are full-overwritten pages so we count all the blocks in
these pages are counted as needed to be allocated */
blocks = (num_pages - 2) << (PAGE_CACHE_SHIFT - inode->i_blkbits);
/* count blocks needed for first page (possibly partially written) */
blocks += ((PAGE_CACHE_SIZE - from) >> inode->i_blkbits) +
!!(from & (inode->i_sb->s_blocksize-1)); /* roundup */
/* Now we account for last page. If last page == first page (we
overwrite only one page), we substract all the blocks past the
last writing position in a page out of already calculated number
of blocks */
blocks += ((num_pages > 1) << (PAGE_CACHE_SHIFT-inode->i_blkbits)) -
((PAGE_CACHE_SIZE - to) >> inode->i_blkbits);
/* Note how we do not roundup here since partial blocks still
should be allocated */
/* Now if all the write area lies past the file end, no point in
maping blocks, since there is none, so we just zero out remaining
parts of first and last pages in write area (if needed) */
if ( (pos & ~(PAGE_CACHE_SIZE - 1)) > inode->i_size ) {
if ( from != 0 ) {/* First page needs to be partially zeroed */
char *kaddr = kmap_atomic(prepared_pages[0], KM_USER0);
memset(kaddr, 0, from);
kunmap_atomic( kaddr, KM_USER0);
}
if ( to != PAGE_CACHE_SIZE ) { /* Last page needs to be partially zeroed */
char *kaddr = kmap_atomic(prepared_pages[num_pages-1], KM_USER0);
memset(kaddr+to, 0, PAGE_CACHE_SIZE - to);
kunmap_atomic( kaddr, KM_USER0);
}
/* Since all blocks are new - use already calculated value */
return blocks;
}
/* Well, since we write somewhere into the middle of a file, there is
possibility we are writing over some already allocated blocks, so
let's map these blocks and substract number of such blocks out of blocks
we need to allocate (calculated above) */
/* Mask write position to start on blocksize, we do it out of the
loop for performance reasons */
pos &= ~(inode->i_sb->s_blocksize - 1);
/* Set cpu key to the starting position in a file (on left block boundary)*/
make_cpu_key (&key, inode, 1 + ((pos) & ~(inode->i_sb->s_blocksize - 1)), TYPE_ANY, 3/*key length*/);
reiserfs_write_lock(inode->i_sb); // We need that for at least search_by_key()
for ( i = 0; i < num_pages ; i++ ) {
int item_pos=-1; /* Position in indirect item */
head = page_buffers(prepared_pages[i]);
/* For each buffer in the page */
for(bh = head, block_start = 0; bh != head || !block_start;
block_start=block_end, bh = bh->b_this_page) {
if (!bh)
reiserfs_panic(inode->i_sb, "green-9002: Allocated but absent buffer for a page?");
/* Find where this buffer ends */
block_end = block_start+inode->i_sb->s_blocksize;
if (i == 0 && block_end <= from )
/* if this buffer is before requested data to map, skip it*/
continue;
if (i == num_pages - 1 && block_start >= to) {
/* If this buffer is after requested data to map, abort
processing of current page */
break;
}
if ( buffer_mapped(bh) && bh->b_blocknr !=0 ) {
/* This is optimisation for a case where buffer is mapped
and have blocknumber assigned. In case significant amount
of such buffers are present, we may avoid some amount
of search_by_key calls.
Probably it would be possible to move parts of this code
out of BKL, but I afraid that would overcomplicate code
without any noticeable benefit.
*/
item_pos++;
/* Update the key */
set_cpu_key_k_offset( &key, cpu_key_k_offset(&key) + inode->i_sb->s_blocksize);
blocks--; // Decrease the amount of blocks that need to be
// allocated
continue; // Go to the next buffer
}
if ( !itembuf || /* if first iteration */
item_pos >= ih_item_len(ih)/UNFM_P_SIZE)
{ /* or if we progressed past the
current unformatted_item */
/* Try to find next item */
res = search_for_position_by_key(inode->i_sb, &key, &path);
/* Abort if no more items */
if ( res != POSITION_FOUND )
break;
/* Update information about current indirect item */
itembuf = get_last_bh( &path );
ih = get_ih( &path );
item = get_item( &path );
item_pos = path.pos_in_item;
RFALSE( !is_indirect_le_ih (ih), "green-9003: indirect item expected");
}
/* See if there is some block associated with the file
at that position, map the buffer to this block */
if ( get_block_num(item,item_pos) ) {
map_bh(bh, inode->i_sb, get_block_num(item,item_pos));
blocks--; // Decrease the amount of blocks that need to be
// allocated
}
item_pos++;
/* Update the key */
set_cpu_key_k_offset( &key, cpu_key_k_offset(&key) + inode->i_sb->s_blocksize);
}
}
pathrelse(&path); // Free the path
reiserfs_write_unlock(inode->i_sb);
/* Now zero out unmappend buffers for the first and last pages of
write area or issue read requests if page is mapped. */
/* First page, see if it is not uptodate */
if ( !PageUptodate(prepared_pages[0]) ) {
head = page_buffers(prepared_pages[0]);
/* For each buffer in page */
for(bh = head, block_start = 0; bh != head || !block_start;
block_start=block_end, bh = bh->b_this_page) {
if (!bh)
reiserfs_panic(inode->i_sb, "green-9002: Allocated but absent buffer for a page?");
/* Find where this buffer ends */
block_end = block_start+inode->i_sb->s_blocksize;
if ( block_end <= from )
/* if this buffer is before requested data to map, skip it*/
continue;
if ( block_start < from ) { /* Aha, our partial buffer */
if ( buffer_mapped(bh) ) { /* If it is mapped, we need to
issue READ request for it to
not loose data */
ll_rw_block(READ, 1, &bh);
*wait_bh++=bh;
} else { /* Not mapped, zero it */
char *kaddr = kmap_atomic(prepared_pages[0], KM_USER0);
memset(kaddr+block_start, 0, from-block_start);
kunmap_atomic( kaddr, KM_USER0);
set_buffer_uptodate(bh);
}
}
}
}
/* Last page, see if it is not uptodate, or if the last page is past the end of the file. */
if ( !PageUptodate(prepared_pages[num_pages-1]) ||
((pos+write_bytes)>>PAGE_CACHE_SHIFT) > (inode->i_size>>PAGE_CACHE_SHIFT) ) {
head = page_buffers(prepared_pages[num_pages-1]);
/* for each buffer in page */
for(bh = head, block_start = 0; bh != head || !block_start;
block_start=block_end, bh = bh->b_this_page) {
if (!bh)
reiserfs_panic(inode->i_sb, "green-9002: Allocated but absent buffer for a page?");
/* Find where this buffer ends */
block_end = block_start+inode->i_sb->s_blocksize;
if ( block_start >= to )
/* if this buffer is after requested data to map, skip it*/
break;
if ( block_end > to ) { /* Aha, our partial buffer */
if ( buffer_mapped(bh) ) { /* If it is mapped, we need to
issue READ request for it to
not loose data */
ll_rw_block(READ, 1, &bh);
*wait_bh++=bh;
} else { /* Not mapped, zero it */
char *kaddr = kmap_atomic(prepared_pages[num_pages-1], KM_USER0);
memset(kaddr+to, 0, block_end-to);
kunmap_atomic( kaddr, KM_USER0);
set_buffer_uptodate(bh);
}
}
}
}
/* Wait for read requests we made to happen, if necessary */
while(wait_bh > wait) {
wait_on_buffer(*--wait_bh);
if (!buffer_uptodate(*wait_bh)) {
res = -EIO;
goto failed_read;
}
}
return blocks;
failed_page_grabbing:
num_pages = i;
failed_read:
reiserfs_unprepare_pages(prepared_pages, num_pages);
return res;
}
/* Write @count bytes at position @ppos in a file indicated by @file
from the buffer @buf.
generic_file_write() is only appropriate for filesystems that are not seeking to optimize performance and want
something simple that works. It is not for serious use by general purpose filesystems, excepting the one that it was
written for (ext2/3). This is for several reasons:
* It has no understanding of any filesystem specific optimizations.
* It enters the filesystem repeatedly for each page that is written.
* It depends on reiserfs_get_block() function which if implemented by reiserfs performs costly search_by_key
* operation for each page it is supplied with. By contrast reiserfs_file_write() feeds as much as possible at a time
* to reiserfs which allows for fewer tree traversals.
* Each indirect pointer insertion takes a lot of cpu, because it involves memory moves inside of blocks.
* Asking the block allocation code for blocks one at a time is slightly less efficient.
All of these reasons for not using only generic file write were understood back when reiserfs was first miscoded to
use it, but we were in a hurry to make code freeze, and so it couldn't be revised then. This new code should make
things right finally.
Future Features: providing search_by_key with hints.
*/
ssize_t reiserfs_file_write( struct file *file, /* the file we are going to write into */
const char *buf, /* pointer to user supplied data
(in userspace) */
size_t count, /* amount of bytes to write */
loff_t *ppos /* pointer to position in file that we start writing at. Should be updated to
* new current position before returning. */ )
{
size_t already_written = 0; // Number of bytes already written to the file.
loff_t pos; // Current position in the file.
size_t res; // return value of various functions that we call.
struct inode *inode = file->f_dentry->d_inode; // Inode of the file that we are writing to.
struct page * prepared_pages[REISERFS_WRITE_PAGES_AT_A_TIME];
/* To simplify coding at this time, we store
locked pages in array for now */
if ( count <= PAGE_CACHE_SIZE || file->f_flags & O_DIRECT)
return generic_file_write(file, buf, count, ppos);
if ( unlikely((ssize_t) count < 0 ))
return -EINVAL;
if (unlikely(!access_ok(VERIFY_READ, buf, count)))
return -EFAULT;
down(&inode->i_sem); // locks the entire file for just us
pos = *ppos;
/* Check if we can write to specified region of file, file
is not overly big and this kind of stuff. Adjust pos and
count, if needed */
res = generic_write_checks(inode, file, &pos, &count, 0);
if (res)
goto out;
if ( count == 0 )
goto out;
remove_suid(file->f_dentry);
inode_update_time(inode, 1); /* Both mtime and ctime */
// Ok, we are done with all the checks.
// Now we should start real work
/* If we are going to write past the file's packed tail or if we are going
to overwrite part of the tail, we need that tail to be converted into
unformatted node */
res = reiserfs_check_for_tail_and_convert( inode, pos, count);
if (res)
goto out;
while ( count > 0) {
/* This is the main loop in which we running until some error occures
or until we write all of the data. */
int num_pages;/* amount of pages we are going to write this iteration */
int write_bytes; /* amount of bytes to write during this iteration */
int blocks_to_allocate; /* how much blocks we need to allocate for
this iteration */
/* (pos & (PAGE_CACHE_SIZE-1)) is an idiom for offset into a page of pos*/
num_pages = !!((pos+count) & (PAGE_CACHE_SIZE - 1)) + /* round up partial
pages */
((count + (pos & (PAGE_CACHE_SIZE-1))) >> PAGE_CACHE_SHIFT);
/* convert size to amount of
pages */
reiserfs_write_lock(inode->i_sb);
if ( num_pages > REISERFS_WRITE_PAGES_AT_A_TIME
|| num_pages > reiserfs_can_fit_pages(inode->i_sb) ) {
/* If we were asked to write more data than we want to or if there
is not that much space, then we shorten amount of data to write
for this iteration. */
num_pages = min_t(int, REISERFS_WRITE_PAGES_AT_A_TIME, reiserfs_can_fit_pages(inode->i_sb));
/* Also we should not forget to set size in bytes accordingly */
write_bytes = (num_pages << PAGE_CACHE_SHIFT) -
(pos & (PAGE_CACHE_SIZE-1));
/* If position is not on the
start of the page, we need
to substract the offset
within page */
} else
write_bytes = count;
/* reserve the blocks to be allocated later, so that later on
we still have the space to write the blocks to */
reiserfs_claim_blocks_to_be_allocated(inode->i_sb, num_pages << (PAGE_CACHE_SHIFT - inode->i_blkbits));
reiserfs_write_unlock(inode->i_sb);
if ( !num_pages ) { /* If we do not have enough space even for */
res = -ENOSPC; /* single page, return -ENOSPC */
if ( pos > (inode->i_size & (inode->i_sb->s_blocksize-1)))
break; // In case we are writing past the file end, break.
// Otherwise we are possibly overwriting the file, so
// let's set write size to be equal or less than blocksize.
// This way we get it correctly for file holes.
// But overwriting files on absolutelly full volumes would not
// be very efficient. Well, people are not supposed to fill
// 100% of disk space anyway.
write_bytes = min_t(int, count, inode->i_sb->s_blocksize - (pos & (inode->i_sb->s_blocksize - 1)));
num_pages = 1;
}
/* Prepare for writing into the region, read in all the
partially overwritten pages, if needed. And lock the pages,
so that nobody else can access these until we are done.
We get number of actual blocks needed as a result.*/
blocks_to_allocate = reiserfs_prepare_file_region_for_write(inode, pos, num_pages, write_bytes, prepared_pages);
if ( blocks_to_allocate < 0 ) {
res = blocks_to_allocate;
reiserfs_release_claimed_blocks(inode->i_sb, num_pages << (PAGE_CACHE_SHIFT - inode->i_blkbits));
break;
}
/* First we correct our estimate of how many blocks we need */
reiserfs_release_claimed_blocks(inode->i_sb, (num_pages << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits)) - blocks_to_allocate );
if ( blocks_to_allocate > 0) {/*We only allocate blocks if we need to*/
/* Fill in all the possible holes and append the file if needed */
res = reiserfs_allocate_blocks_for_region(inode, pos, num_pages, write_bytes, prepared_pages, blocks_to_allocate);
} else if ( pos + write_bytes > inode->i_size ) {
/* File might have grown even though no new blocks were added */
inode->i_size = pos + write_bytes;
inode->i_sb->s_op->dirty_inode(inode);
}
/* well, we have allocated the blocks, so it is time to free
the reservation we made earlier. */
reiserfs_release_claimed_blocks(inode->i_sb, blocks_to_allocate);
if ( res ) {
reiserfs_unprepare_pages(prepared_pages, num_pages);
break;
}
/* NOTE that allocating blocks and filling blocks can be done in reverse order
and probably we would do that just to get rid of garbage in files after a
crash */
/* Copy data from user-supplied buffer to file's pages */
res = reiserfs_copy_from_user_to_file_region(pos, num_pages, write_bytes, prepared_pages, buf);
if ( res ) {
reiserfs_unprepare_pages(prepared_pages, num_pages);
break;
}
/* Send the pages to disk and unlock them. */
res = reiserfs_submit_file_region_for_write(pos, num_pages, write_bytes, prepared_pages);
if ( res )
break;
already_written += write_bytes;
buf += write_bytes;
*ppos = pos += write_bytes;
count -= write_bytes;
}
if ((file->f_flags & O_SYNC) || IS_SYNC(inode))
res = generic_osync_inode(inode, OSYNC_METADATA|OSYNC_DATA);
up(&inode->i_sem);
return (already_written != 0)?already_written:res;
out:
up(&inode->i_sem); // unlock the file on exit.
return res;
}
struct file_operations reiserfs_file_operations = {
.read = generic_file_read,
.write = generic_file_write,
.write = reiserfs_file_write,
.ioctl = reiserfs_ioctl,
.mmap = generic_file_mmap,
.release = reiserfs_file_release,
......
......@@ -14,6 +14,8 @@
#include <linux/mpage.h>
#include <linux/writeback.h>
extern int reiserfs_default_io_size; /* default io size devuned in super.c */
/* args for the create parameter of reiserfs_get_block */
#define GET_BLOCK_NO_CREATE 0 /* don't create new blocks or convert tails */
#define GET_BLOCK_CREATE 1 /* add anything you need to find block */
......@@ -908,7 +910,7 @@ static void init_inode (struct inode * inode, struct path * path)
copy_key (INODE_PKEY (inode), &(ih->ih_key));
inode->i_blksize = PAGE_SIZE;
inode->i_blksize = reiserfs_default_io_size;
INIT_LIST_HEAD(&(REISERFS_I(inode)->i_prealloc_list ));
REISERFS_I(inode)->i_flags = 0;
......@@ -1598,7 +1600,7 @@ int reiserfs_new_inode (struct reiserfs_transaction_handle *th,
}
// these do not go to on-disk stat data
inode->i_ino = le32_to_cpu (ih.ih_key.k_objectid);
inode->i_blksize = PAGE_SIZE;
inode->i_blksize = reiserfs_default_io_size;
// store in in-core inode the key of stat data and version all
// object items will have (directory items will have old offset
......
......@@ -532,6 +532,11 @@ static const arg_desc_t tails[] = {
{NULL, 0}
};
int reiserfs_default_io_size = 128 * 1024; /* Default recommended I/O size is 128k.
There might be broken applications that are
confused by this. Use nolargeio mount option
to get usual i/o size = PAGE_SIZE.
*/
/* proceed only one option from a list *cur - string containing of mount options
opts - array of options which are accepted
......@@ -657,6 +662,7 @@ for old setups still work */
{"block-allocator", 'a', balloc, -1},
{"resize", 'r', 0, -1},
{"jdev", 'j', 0, -1},
{"nolargeio", 'w', 0, -1},
{NULL, 0, 0, -1}
};
......@@ -688,6 +694,10 @@ for old setups still work */
}
}
if ( c == 'w' ) {
reiserfs_default_io_size = PAGE_SIZE;
}
if (c == 'j') {
if (arg && *arg && jdev_name) {
*jdev_name = arg;
......@@ -1318,6 +1328,7 @@ static int reiserfs_fill_super (struct super_block * s, void * data, int silent)
reiserfs_proc_register( s, "oidmap", reiserfs_oidmap_in_proc );
reiserfs_proc_register( s, "journal", reiserfs_journal_in_proc );
init_waitqueue_head (&(sbi->s_wait));
sbi->bitmap_lock = SPIN_LOCK_UNLOCKED;
return (0);
......
......@@ -1268,6 +1268,7 @@ struct path var = {ILLEGAL_PATH_ELEMENT_OFFSET, }
/* Size of pointer to the unformatted node. */
#define UNFM_P_SIZE (sizeof(unp_t))
#define UNFM_P_SHIFT 2
// in in-core inode key is stored on le form
#define INODE_PKEY(inode) ((struct key *)(REISERFS_I(inode)->i_key))
......@@ -1838,7 +1839,7 @@ void reiserfs_do_truncate (struct reiserfs_transaction_handle *th,
void padd_item (char * item, int total_length, int length);
/* inode.c */
void restart_transaction(struct reiserfs_transaction_handle *th, struct inode *inode, struct path *path);
void reiserfs_read_locked_inode(struct inode * inode, struct reiserfs_iget_args *args) ;
int reiserfs_find_actor(struct inode * inode, void *p) ;
int reiserfs_init_locked_inode(struct inode * inode, void *p) ;
......@@ -2111,6 +2112,7 @@ void reiserfs_discard_all_prealloc (struct reiserfs_transaction_handle *th);
#endif
void reiserfs_claim_blocks_to_be_allocated( struct super_block *sb, int blocks);
void reiserfs_release_claimed_blocks( struct super_block *sb, int blocks);
int reiserfs_can_fit_pages(struct super_block *sb);
/* hashes.c */
__u32 keyed_hash (const signed char *msg, int len);
......
......@@ -397,6 +397,7 @@ struct reiserfs_sb_info
reiserfs_proc_info_data_t s_proc_info_data;
struct proc_dir_entry *procdir;
int reserved_blocks; /* amount of blocks reserved for further allocations */
spinlock_t bitmap_lock; /* this lock on now only used to protect reserved_blocks variable */
};
/* Definitions of reiserfs on-disk properties: */
......
......@@ -223,6 +223,7 @@ EXPORT_SYMBOL(block_sync_page);
EXPORT_SYMBOL(generic_cont_expand);
EXPORT_SYMBOL(cont_prepare_write);
EXPORT_SYMBOL(generic_commit_write);
EXPORT_SYMBOL(block_commit_write);
EXPORT_SYMBOL(block_truncate_page);
EXPORT_SYMBOL(generic_block_bmap);
EXPORT_SYMBOL(generic_file_read);
......@@ -556,6 +557,8 @@ EXPORT_SYMBOL(buffer_insert_list);
EXPORT_SYMBOL(make_bad_inode);
EXPORT_SYMBOL(is_bad_inode);
EXPORT_SYMBOL(__inode_dir_notify);
EXPORT_SYMBOL(generic_osync_inode);
EXPORT_SYMBOL(remove_suid);
#ifdef CONFIG_UID16
EXPORT_SYMBOL(overflowuid);
......
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