xfs_log.c 113 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
1
/*
2 3
 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
 * All Rights Reserved.
Linus Torvalds's avatar
Linus Torvalds committed
4
 *
5 6
 * This program is free software; you can redistribute it and/or
 * modify it under the terms of the GNU General Public License as
Linus Torvalds's avatar
Linus Torvalds committed
7 8
 * published by the Free Software Foundation.
 *
9 10 11 12
 * This program is distributed in the hope that it would 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.
Linus Torvalds's avatar
Linus Torvalds committed
13
 *
14 15 16
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write the Free Software Foundation,
 * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
Linus Torvalds's avatar
Linus Torvalds committed
17 18
 */
#include "xfs.h"
19
#include "xfs_fs.h"
20
#include "xfs_shared.h"
21
#include "xfs_format.h"
22 23
#include "xfs_log_format.h"
#include "xfs_trans_resv.h"
Linus Torvalds's avatar
Linus Torvalds committed
24 25
#include "xfs_mount.h"
#include "xfs_error.h"
26 27 28
#include "xfs_trans.h"
#include "xfs_trans_priv.h"
#include "xfs_log.h"
Linus Torvalds's avatar
Linus Torvalds committed
29 30
#include "xfs_log_priv.h"
#include "xfs_log_recover.h"
31
#include "xfs_inode.h"
32
#include "xfs_trace.h"
33
#include "xfs_fsops.h"
34
#include "xfs_cksum.h"
35
#include "xfs_sysfs.h"
36
#include "xfs_sb.h"
Linus Torvalds's avatar
Linus Torvalds committed
37

38
kmem_zone_t	*xfs_log_ticket_zone;
Linus Torvalds's avatar
Linus Torvalds committed
39 40

/* Local miscellaneous function prototypes */
41 42 43 44 45 46 47
STATIC int
xlog_commit_record(
	struct xlog		*log,
	struct xlog_ticket	*ticket,
	struct xlog_in_core	**iclog,
	xfs_lsn_t		*commitlsnp);

Mark Tinguely's avatar
Mark Tinguely committed
48 49 50 51 52 53
STATIC struct xlog *
xlog_alloc_log(
	struct xfs_mount	*mp,
	struct xfs_buftarg	*log_target,
	xfs_daddr_t		blk_offset,
	int			num_bblks);
54 55 56 57
STATIC int
xlog_space_left(
	struct xlog		*log,
	atomic64_t		*head);
Mark Tinguely's avatar
Mark Tinguely committed
58 59 60 61 62 63 64
STATIC int
xlog_sync(
	struct xlog		*log,
	struct xlog_in_core	*iclog);
STATIC void
xlog_dealloc_log(
	struct xlog		*log);
Linus Torvalds's avatar
Linus Torvalds committed
65 66 67

/* local state machine functions */
STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int);
Mark Tinguely's avatar
Mark Tinguely committed
68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93
STATIC void
xlog_state_do_callback(
	struct xlog		*log,
	int			aborted,
	struct xlog_in_core	*iclog);
STATIC int
xlog_state_get_iclog_space(
	struct xlog		*log,
	int			len,
	struct xlog_in_core	**iclog,
	struct xlog_ticket	*ticket,
	int			*continued_write,
	int			*logoffsetp);
STATIC int
xlog_state_release_iclog(
	struct xlog		*log,
	struct xlog_in_core	*iclog);
STATIC void
xlog_state_switch_iclogs(
	struct xlog		*log,
	struct xlog_in_core	*iclog,
	int			eventual_size);
STATIC void
xlog_state_want_sync(
	struct xlog		*log,
	struct xlog_in_core	*iclog);
Linus Torvalds's avatar
Linus Torvalds committed
94

95 96
STATIC void
xlog_grant_push_ail(
Mark Tinguely's avatar
Mark Tinguely committed
97 98 99 100 101 102 103 104 105 106
	struct xlog		*log,
	int			need_bytes);
STATIC void
xlog_regrant_reserve_log_space(
	struct xlog		*log,
	struct xlog_ticket	*ticket);
STATIC void
xlog_ungrant_log_space(
	struct xlog		*log,
	struct xlog_ticket	*ticket);
Linus Torvalds's avatar
Linus Torvalds committed
107

108
#if defined(DEBUG)
Mark Tinguely's avatar
Mark Tinguely committed
109 110 111
STATIC void
xlog_verify_dest_ptr(
	struct xlog		*log,
112
	void			*ptr);
113 114
STATIC void
xlog_verify_grant_tail(
Mark Tinguely's avatar
Mark Tinguely committed
115 116 117 118 119 120
	struct xlog *log);
STATIC void
xlog_verify_iclog(
	struct xlog		*log,
	struct xlog_in_core	*iclog,
	int			count,
121
	bool                    syncing);
Mark Tinguely's avatar
Mark Tinguely committed
122 123 124 125 126
STATIC void
xlog_verify_tail_lsn(
	struct xlog		*log,
	struct xlog_in_core	*iclog,
	xfs_lsn_t		tail_lsn);
Linus Torvalds's avatar
Linus Torvalds committed
127 128
#else
#define xlog_verify_dest_ptr(a,b)
129
#define xlog_verify_grant_tail(a)
Linus Torvalds's avatar
Linus Torvalds committed
130 131 132 133
#define xlog_verify_iclog(a,b,c,d)
#define xlog_verify_tail_lsn(a,b,c)
#endif

Mark Tinguely's avatar
Mark Tinguely committed
134 135 136
STATIC int
xlog_iclogs_empty(
	struct xlog		*log);
Linus Torvalds's avatar
Linus Torvalds committed
137

138
static void
139
xlog_grant_sub_space(
140 141 142
	struct xlog		*log,
	atomic64_t		*head,
	int			bytes)
143
{
144 145
	int64_t	head_val = atomic64_read(head);
	int64_t new, old;
146

147 148
	do {
		int	cycle, space;
149

150
		xlog_crack_grant_head_val(head_val, &cycle, &space);
151

152 153 154 155 156 157 158 159 160 161
		space -= bytes;
		if (space < 0) {
			space += log->l_logsize;
			cycle--;
		}

		old = head_val;
		new = xlog_assign_grant_head_val(cycle, space);
		head_val = atomic64_cmpxchg(head, old, new);
	} while (head_val != old);
162 163 164
}

static void
165
xlog_grant_add_space(
166 167 168
	struct xlog		*log,
	atomic64_t		*head,
	int			bytes)
169
{
170 171
	int64_t	head_val = atomic64_read(head);
	int64_t new, old;
172

173 174 175
	do {
		int		tmp;
		int		cycle, space;
176

177
		xlog_crack_grant_head_val(head_val, &cycle, &space);
178

179 180 181 182 183 184 185 186 187 188 189 190
		tmp = log->l_logsize - space;
		if (tmp > bytes)
			space += bytes;
		else {
			space = bytes - tmp;
			cycle++;
		}

		old = head_val;
		new = xlog_assign_grant_head_val(cycle, space);
		head_val = atomic64_cmpxchg(head, old, new);
	} while (head_val != old);
191
}
192

193 194 195 196 197 198 199 200 201
STATIC void
xlog_grant_head_init(
	struct xlog_grant_head	*head)
{
	xlog_assign_grant_head(&head->grant, 1, 0);
	INIT_LIST_HEAD(&head->waiters);
	spin_lock_init(&head->lock);
}

202 203 204 205 206 207 208 209 210 211 212 213
STATIC void
xlog_grant_head_wake_all(
	struct xlog_grant_head	*head)
{
	struct xlog_ticket	*tic;

	spin_lock(&head->lock);
	list_for_each_entry(tic, &head->waiters, t_queue)
		wake_up_process(tic->t_task);
	spin_unlock(&head->lock);
}

214 215
static inline int
xlog_ticket_reservation(
216
	struct xlog		*log,
217 218
	struct xlog_grant_head	*head,
	struct xlog_ticket	*tic)
219
{
220 221 222 223
	if (head == &log->l_write_head) {
		ASSERT(tic->t_flags & XLOG_TIC_PERM_RESERV);
		return tic->t_unit_res;
	} else {
224
		if (tic->t_flags & XLOG_TIC_PERM_RESERV)
225
			return tic->t_unit_res * tic->t_cnt;
226
		else
227
			return tic->t_unit_res;
228 229 230 231
	}
}

STATIC bool
232
xlog_grant_head_wake(
233
	struct xlog		*log,
234
	struct xlog_grant_head	*head,
235 236 237 238 239
	int			*free_bytes)
{
	struct xlog_ticket	*tic;
	int			need_bytes;

240 241
	list_for_each_entry(tic, &head->waiters, t_queue) {
		need_bytes = xlog_ticket_reservation(log, head, tic);
242 243 244
		if (*free_bytes < need_bytes)
			return false;

245 246
		*free_bytes -= need_bytes;
		trace_xfs_log_grant_wake_up(log, tic);
247
		wake_up_process(tic->t_task);
248 249 250 251 252 253
	}

	return true;
}

STATIC int
254
xlog_grant_head_wait(
255
	struct xlog		*log,
256
	struct xlog_grant_head	*head,
257
	struct xlog_ticket	*tic,
258 259
	int			need_bytes) __releases(&head->lock)
					    __acquires(&head->lock)
260
{
261
	list_add_tail(&tic->t_queue, &head->waiters);
262 263 264 265 266 267

	do {
		if (XLOG_FORCED_SHUTDOWN(log))
			goto shutdown;
		xlog_grant_push_ail(log, need_bytes);

268
		__set_current_state(TASK_UNINTERRUPTIBLE);
269
		spin_unlock(&head->lock);
270

271
		XFS_STATS_INC(log->l_mp, xs_sleep_logspace);
272

273 274
		trace_xfs_log_grant_sleep(log, tic);
		schedule();
275 276
		trace_xfs_log_grant_wake(log, tic);

277
		spin_lock(&head->lock);
278 279
		if (XLOG_FORCED_SHUTDOWN(log))
			goto shutdown;
280
	} while (xlog_space_left(log, &head->grant) < need_bytes);
281 282 283 284 285

	list_del_init(&tic->t_queue);
	return 0;
shutdown:
	list_del_init(&tic->t_queue);
286
	return -EIO;
287 288
}

289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307
/*
 * Atomically get the log space required for a log ticket.
 *
 * Once a ticket gets put onto head->waiters, it will only return after the
 * needed reservation is satisfied.
 *
 * This function is structured so that it has a lock free fast path. This is
 * necessary because every new transaction reservation will come through this
 * path. Hence any lock will be globally hot if we take it unconditionally on
 * every pass.
 *
 * As tickets are only ever moved on and off head->waiters under head->lock, we
 * only need to take that lock if we are going to add the ticket to the queue
 * and sleep. We can avoid taking the lock if the ticket was never added to
 * head->waiters because the t_queue list head will be empty and we hold the
 * only reference to it so it can safely be checked unlocked.
 */
STATIC int
xlog_grant_head_check(
308
	struct xlog		*log,
309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342
	struct xlog_grant_head	*head,
	struct xlog_ticket	*tic,
	int			*need_bytes)
{
	int			free_bytes;
	int			error = 0;

	ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));

	/*
	 * If there are other waiters on the queue then give them a chance at
	 * logspace before us.  Wake up the first waiters, if we do not wake
	 * up all the waiters then go to sleep waiting for more free space,
	 * otherwise try to get some space for this transaction.
	 */
	*need_bytes = xlog_ticket_reservation(log, head, tic);
	free_bytes = xlog_space_left(log, &head->grant);
	if (!list_empty_careful(&head->waiters)) {
		spin_lock(&head->lock);
		if (!xlog_grant_head_wake(log, head, &free_bytes) ||
		    free_bytes < *need_bytes) {
			error = xlog_grant_head_wait(log, head, tic,
						     *need_bytes);
		}
		spin_unlock(&head->lock);
	} else if (free_bytes < *need_bytes) {
		spin_lock(&head->lock);
		error = xlog_grant_head_wait(log, head, tic, *need_bytes);
		spin_unlock(&head->lock);
	}

	return error;
}

343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365
static void
xlog_tic_reset_res(xlog_ticket_t *tic)
{
	tic->t_res_num = 0;
	tic->t_res_arr_sum = 0;
	tic->t_res_num_ophdrs = 0;
}

static void
xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type)
{
	if (tic->t_res_num == XLOG_TIC_LEN_MAX) {
		/* add to overflow and start again */
		tic->t_res_o_flow += tic->t_res_arr_sum;
		tic->t_res_num = 0;
		tic->t_res_arr_sum = 0;
	}

	tic->t_res_arr[tic->t_res_num].r_len = len;
	tic->t_res_arr[tic->t_res_num].r_type = type;
	tic->t_res_arr_sum += len;
	tic->t_res_num++;
}
366

367 368 369 370 371 372 373 374
/*
 * Replenish the byte reservation required by moving the grant write head.
 */
int
xfs_log_regrant(
	struct xfs_mount	*mp,
	struct xlog_ticket	*tic)
{
375
	struct xlog		*log = mp->m_log;
376 377 378 379
	int			need_bytes;
	int			error = 0;

	if (XLOG_FORCED_SHUTDOWN(log))
380
		return -EIO;
381

382
	XFS_STATS_INC(mp, xs_try_logspace);
383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437

	/*
	 * This is a new transaction on the ticket, so we need to change the
	 * transaction ID so that the next transaction has a different TID in
	 * the log. Just add one to the existing tid so that we can see chains
	 * of rolling transactions in the log easily.
	 */
	tic->t_tid++;

	xlog_grant_push_ail(log, tic->t_unit_res);

	tic->t_curr_res = tic->t_unit_res;
	xlog_tic_reset_res(tic);

	if (tic->t_cnt > 0)
		return 0;

	trace_xfs_log_regrant(log, tic);

	error = xlog_grant_head_check(log, &log->l_write_head, tic,
				      &need_bytes);
	if (error)
		goto out_error;

	xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
	trace_xfs_log_regrant_exit(log, tic);
	xlog_verify_grant_tail(log);
	return 0;

out_error:
	/*
	 * If we are failing, make sure the ticket doesn't have any current
	 * reservations.  We don't want to add this back when the ticket/
	 * transaction gets cancelled.
	 */
	tic->t_curr_res = 0;
	tic->t_cnt = 0;	/* ungrant will give back unit_res * t_cnt. */
	return error;
}

/*
 * Reserve log space and return a ticket corresponding the reservation.
 *
 * Each reservation is going to reserve extra space for a log record header.
 * When writes happen to the on-disk log, we don't subtract the length of the
 * log record header from any reservation.  By wasting space in each
 * reservation, we prevent over allocation problems.
 */
int
xfs_log_reserve(
	struct xfs_mount	*mp,
	int		 	unit_bytes,
	int		 	cnt,
	struct xlog_ticket	**ticp,
	__uint8_t	 	client,
438
	bool			permanent)
439
{
440
	struct xlog		*log = mp->m_log;
441 442 443 444 445 446 447
	struct xlog_ticket	*tic;
	int			need_bytes;
	int			error = 0;

	ASSERT(client == XFS_TRANSACTION || client == XFS_LOG);

	if (XLOG_FORCED_SHUTDOWN(log))
448
		return -EIO;
449

450
	XFS_STATS_INC(mp, xs_try_logspace);
451 452 453 454 455

	ASSERT(*ticp == NULL);
	tic = xlog_ticket_alloc(log, unit_bytes, cnt, client, permanent,
				KM_SLEEP | KM_MAYFAIL);
	if (!tic)
456
		return -ENOMEM;
457 458 459

	*ticp = tic;

460 461
	xlog_grant_push_ail(log, tic->t_cnt ? tic->t_unit_res * tic->t_cnt
					    : tic->t_unit_res);
462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487

	trace_xfs_log_reserve(log, tic);

	error = xlog_grant_head_check(log, &log->l_reserve_head, tic,
				      &need_bytes);
	if (error)
		goto out_error;

	xlog_grant_add_space(log, &log->l_reserve_head.grant, need_bytes);
	xlog_grant_add_space(log, &log->l_write_head.grant, need_bytes);
	trace_xfs_log_reserve_exit(log, tic);
	xlog_verify_grant_tail(log);
	return 0;

out_error:
	/*
	 * If we are failing, make sure the ticket doesn't have any current
	 * reservations.  We don't want to add this back when the ticket/
	 * transaction gets cancelled.
	 */
	tic->t_curr_res = 0;
	tic->t_cnt = 0;	/* ungrant will give back unit_res * t_cnt. */
	return error;
}


Linus Torvalds's avatar
Linus Torvalds committed
488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509
/*
 * NOTES:
 *
 *	1. currblock field gets updated at startup and after in-core logs
 *		marked as with WANT_SYNC.
 */

/*
 * This routine is called when a user of a log manager ticket is done with
 * the reservation.  If the ticket was ever used, then a commit record for
 * the associated transaction is written out as a log operation header with
 * no data.  The flag XLOG_TIC_INITED is set when the first write occurs with
 * a given ticket.  If the ticket was one with a permanent reservation, then
 * a few operations are done differently.  Permanent reservation tickets by
 * default don't release the reservation.  They just commit the current
 * transaction with the belief that the reservation is still needed.  A flag
 * must be passed in before permanent reservations are actually released.
 * When these type of tickets are not released, they need to be set into
 * the inited state again.  By doing this, a start record will be written
 * out when the next write occurs.
 */
xfs_lsn_t
510 511 512 513
xfs_log_done(
	struct xfs_mount	*mp,
	struct xlog_ticket	*ticket,
	struct xlog_in_core	**iclog,
514
	bool			regrant)
Linus Torvalds's avatar
Linus Torvalds committed
515
{
516
	struct xlog		*log = mp->m_log;
517
	xfs_lsn_t		lsn = 0;
Linus Torvalds's avatar
Linus Torvalds committed
518 519 520 521 522 523 524

	if (XLOG_FORCED_SHUTDOWN(log) ||
	    /*
	     * If nothing was ever written, don't write out commit record.
	     * If we get an error, just continue and give back the log ticket.
	     */
	    (((ticket->t_flags & XLOG_TIC_INITED) == 0) &&
525
	     (xlog_commit_record(log, ticket, iclog, &lsn)))) {
Linus Torvalds's avatar
Linus Torvalds committed
526
		lsn = (xfs_lsn_t) -1;
527
		regrant = false;
Linus Torvalds's avatar
Linus Torvalds committed
528 529 530
	}


531
	if (!regrant) {
532 533
		trace_xfs_log_done_nonperm(log, ticket);

Linus Torvalds's avatar
Linus Torvalds committed
534
		/*
535
		 * Release ticket if not permanent reservation or a specific
Linus Torvalds's avatar
Linus Torvalds committed
536 537 538 539
		 * request has been made to release a permanent reservation.
		 */
		xlog_ungrant_log_space(log, ticket);
	} else {
540 541
		trace_xfs_log_done_perm(log, ticket);

Linus Torvalds's avatar
Linus Torvalds committed
542
		xlog_regrant_reserve_log_space(log, ticket);
543 544 545 546
		/* If this ticket was a permanent reservation and we aren't
		 * trying to release it, reset the inited flags; so next time
		 * we write, a start record will be written out.
		 */
Linus Torvalds's avatar
Linus Torvalds committed
547
		ticket->t_flags |= XLOG_TIC_INITED;
548
	}
Linus Torvalds's avatar
Linus Torvalds committed
549

550
	xfs_log_ticket_put(ticket);
Linus Torvalds's avatar
Linus Torvalds committed
551
	return lsn;
552
}
Linus Torvalds's avatar
Linus Torvalds committed
553 554 555 556 557 558 559 560

/*
 * Attaches a new iclog I/O completion callback routine during
 * transaction commit.  If the log is in error state, a non-zero
 * return code is handed back and the caller is responsible for
 * executing the callback at an appropriate time.
 */
int
561 562 563 564
xfs_log_notify(
	struct xfs_mount	*mp,
	struct xlog_in_core	*iclog,
	xfs_log_callback_t	*cb)
Linus Torvalds's avatar
Linus Torvalds committed
565
{
Eric Sandeen's avatar
Eric Sandeen committed
566
	int	abortflg;
Linus Torvalds's avatar
Linus Torvalds committed
567

568
	spin_lock(&iclog->ic_callback_lock);
Linus Torvalds's avatar
Linus Torvalds committed
569 570 571 572 573 574 575 576
	abortflg = (iclog->ic_state & XLOG_STATE_IOERROR);
	if (!abortflg) {
		ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) ||
			      (iclog->ic_state == XLOG_STATE_WANT_SYNC));
		cb->cb_next = NULL;
		*(iclog->ic_callback_tail) = cb;
		iclog->ic_callback_tail = &(cb->cb_next);
	}
577
	spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds's avatar
Linus Torvalds committed
578
	return abortflg;
579
}
Linus Torvalds's avatar
Linus Torvalds committed
580 581

int
582 583 584
xfs_log_release_iclog(
	struct xfs_mount	*mp,
	struct xlog_in_core	*iclog)
Linus Torvalds's avatar
Linus Torvalds committed
585
{
586
	if (xlog_state_release_iclog(mp->m_log, iclog)) {
587
		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
588
		return -EIO;
Linus Torvalds's avatar
Linus Torvalds committed
589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604
	}

	return 0;
}

/*
 * Mount a log filesystem
 *
 * mp		- ubiquitous xfs mount point structure
 * log_target	- buftarg of on-disk log device
 * blk_offset	- Start block # where block size is 512 bytes (BBSIZE)
 * num_bblocks	- Number of BBSIZE blocks in on-disk log
 *
 * Return error or zero.
 */
int
605 606 607 608 609
xfs_log_mount(
	xfs_mount_t	*mp,
	xfs_buftarg_t	*log_target,
	xfs_daddr_t	blk_offset,
	int		num_bblks)
Linus Torvalds's avatar
Linus Torvalds committed
610
{
611 612
	int		error = 0;
	int		min_logfsbs;
613

614 615 616 617
	if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
		xfs_notice(mp, "Mounting V%d Filesystem",
			   XFS_SB_VERSION_NUM(&mp->m_sb));
	} else {
618
		xfs_notice(mp,
619 620
"Mounting V%d filesystem in no-recovery mode. Filesystem will be inconsistent.",
			   XFS_SB_VERSION_NUM(&mp->m_sb));
621
		ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds's avatar
Linus Torvalds committed
622 623 624
	}

	mp->m_log = xlog_alloc_log(mp, log_target, blk_offset, num_bblks);
625
	if (IS_ERR(mp->m_log)) {
626
		error = PTR_ERR(mp->m_log);
627 628
		goto out;
	}
Linus Torvalds's avatar
Linus Torvalds committed
629

630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649
	/*
	 * Validate the given log space and drop a critical message via syslog
	 * if the log size is too small that would lead to some unexpected
	 * situations in transaction log space reservation stage.
	 *
	 * Note: we can't just reject the mount if the validation fails.  This
	 * would mean that people would have to downgrade their kernel just to
	 * remedy the situation as there is no way to grow the log (short of
	 * black magic surgery with xfs_db).
	 *
	 * We can, however, reject mounts for CRC format filesystems, as the
	 * mkfs binary being used to make the filesystem should never create a
	 * filesystem with a log that is too small.
	 */
	min_logfsbs = xfs_log_calc_minimum_size(mp);

	if (mp->m_sb.sb_logblocks < min_logfsbs) {
		xfs_warn(mp,
		"Log size %d blocks too small, minimum size is %d blocks",
			 mp->m_sb.sb_logblocks, min_logfsbs);
650
		error = -EINVAL;
651 652 653 654
	} else if (mp->m_sb.sb_logblocks > XFS_MAX_LOG_BLOCKS) {
		xfs_warn(mp,
		"Log size %d blocks too large, maximum size is %lld blocks",
			 mp->m_sb.sb_logblocks, XFS_MAX_LOG_BLOCKS);
655
		error = -EINVAL;
656 657 658 659 660
	} else if (XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks) > XFS_MAX_LOG_BYTES) {
		xfs_warn(mp,
		"log size %lld bytes too large, maximum size is %lld bytes",
			 XFS_FSB_TO_B(mp, mp->m_sb.sb_logblocks),
			 XFS_MAX_LOG_BYTES);
661
		error = -EINVAL;
662 663 664 665 666 667 668
	}
	if (error) {
		if (xfs_sb_version_hascrc(&mp->m_sb)) {
			xfs_crit(mp, "AAIEEE! Log failed size checks. Abort!");
			ASSERT(0);
			goto out_free_log;
		}
669
		xfs_crit(mp, "Log size out of supported range.");
670
		xfs_crit(mp,
671
"Continuing onwards, but if log hangs are experienced then please report this message in the bug report.");
672 673
	}

674 675 676 677 678
	/*
	 * Initialize the AIL now we have a log.
	 */
	error = xfs_trans_ail_init(mp);
	if (error) {
679
		xfs_warn(mp, "AIL initialisation failed: error %d", error);
680
		goto out_free_log;
681
	}
682
	mp->m_log->l_ailp = mp->m_ail;
683

Linus Torvalds's avatar
Linus Torvalds committed
684 685 686 687 688
	/*
	 * skip log recovery on a norecovery mount.  pretend it all
	 * just worked.
	 */
	if (!(mp->m_flags & XFS_MOUNT_NORECOVERY)) {
689
		int	readonly = (mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds's avatar
Linus Torvalds committed
690 691

		if (readonly)
692
			mp->m_flags &= ~XFS_MOUNT_RDONLY;
Linus Torvalds's avatar
Linus Torvalds committed
693

694
		error = xlog_recover(mp->m_log);
Linus Torvalds's avatar
Linus Torvalds committed
695 696

		if (readonly)
697
			mp->m_flags |= XFS_MOUNT_RDONLY;
Linus Torvalds's avatar
Linus Torvalds committed
698
		if (error) {
699 700
			xfs_warn(mp, "log mount/recovery failed: error %d",
				error);
701
			xlog_recover_cancel(mp->m_log);
702
			goto out_destroy_ail;
Linus Torvalds's avatar
Linus Torvalds committed
703 704 705
		}
	}

706 707 708 709 710
	error = xfs_sysfs_init(&mp->m_log->l_kobj, &xfs_log_ktype, &mp->m_kobj,
			       "log");
	if (error)
		goto out_destroy_ail;

Linus Torvalds's avatar
Linus Torvalds committed
711 712 713
	/* Normal transactions can now occur */
	mp->m_log->l_flags &= ~XLOG_ACTIVE_RECOVERY;

714 715 716 717 718 719 720
	/*
	 * Now the log has been fully initialised and we know were our
	 * space grant counters are, we can initialise the permanent ticket
	 * needed for delayed logging to work.
	 */
	xlog_cil_init_post_recovery(mp->m_log);

Linus Torvalds's avatar
Linus Torvalds committed
721
	return 0;
722 723 724 725 726

out_destroy_ail:
	xfs_trans_ail_destroy(mp);
out_free_log:
	xlog_dealloc_log(mp->m_log);
727
out:
728
	return error;
729
}
Linus Torvalds's avatar
Linus Torvalds committed
730 731

/*
732 733 734 735
 * Finish the recovery of the file system.  This is separate from the
 * xfs_log_mount() call, because it depends on the code in xfs_mountfs() to read
 * in the root and real-time bitmap inodes between calling xfs_log_mount() and
 * here.
Linus Torvalds's avatar
Linus Torvalds committed
736
 *
737 738 739
 * If we finish recovery successfully, start the background log work. If we are
 * not doing recovery, then we have a RO filesystem and we don't need to start
 * it.
Linus Torvalds's avatar
Linus Torvalds committed
740 741
 */
int
742 743
xfs_log_mount_finish(
	struct xfs_mount	*mp)
Linus Torvalds's avatar
Linus Torvalds committed
744
{
745
	int	error = 0;
746
	bool	readonly = (mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds's avatar
Linus Torvalds committed
747

748
	if (mp->m_flags & XFS_MOUNT_NORECOVERY) {
749
		ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
750
		return 0;
751 752 753
	} else if (readonly) {
		/* Allow unlinked processing to proceed */
		mp->m_flags &= ~XFS_MOUNT_RDONLY;
Linus Torvalds's avatar
Linus Torvalds committed
754 755
	}

756 757 758 759 760 761 762 763
	/*
	 * During the second phase of log recovery, we need iget and
	 * iput to behave like they do for an active filesystem.
	 * xfs_fs_drop_inode needs to be able to prevent the deletion
	 * of inodes before we're done replaying log items on those
	 * inodes.  Turn it off immediately after recovery finishes
	 * so that we don't leak the quota inodes if subsequent mount
	 * activities fail.
764 765 766 767 768 769 770 771 772 773 774
	 *
	 * We let all inodes involved in redo item processing end up on
	 * the LRU instead of being evicted immediately so that if we do
	 * something to an unlinked inode, the irele won't cause
	 * premature truncation and freeing of the inode, which results
	 * in log recovery failure.  We have to evict the unreferenced
	 * lru inodes after clearing MS_ACTIVE because we don't
	 * otherwise clean up the lru if there's a subsequent failure in
	 * xfs_mountfs, which leads to us leaking the inodes if nothing
	 * else (e.g. quotacheck) references the inodes before the
	 * mount failure occurs.
775 776
	 */
	mp->m_super->s_flags |= MS_ACTIVE;
777 778 779
	error = xlog_recover_finish(mp->m_log);
	if (!error)
		xfs_log_work_queue(mp);
780
	mp->m_super->s_flags &= ~MS_ACTIVE;
781
	evict_inodes(mp->m_super);
782

783 784 785
	if (readonly)
		mp->m_flags |= XFS_MOUNT_RDONLY;

786 787 788 789 790 791 792 793 794 795 796 797 798 799 800
	return error;
}

/*
 * The mount has failed. Cancel the recovery if it hasn't completed and destroy
 * the log.
 */
int
xfs_log_mount_cancel(
	struct xfs_mount	*mp)
{
	int			error;

	error = xlog_recover_cancel(mp->m_log);
	xfs_log_unmount(mp);
801

Linus Torvalds's avatar
Linus Torvalds committed
802 803 804 805 806 807 808 809 810 811 812 813 814 815 816
	return error;
}

/*
 * Final log writes as part of unmount.
 *
 * Mark the filesystem clean as unmount happens.  Note that during relocation
 * this routine needs to be executed as part of source-bag while the
 * deallocation must not be done until source-end.
 */

/*
 * Unmount record used to have a string "Unmount filesystem--" in the
 * data section where the "Un" was really a magic number (XLOG_UNMOUNT_TYPE).
 * We just write the magic number now since that particular field isn't
817
 * currently architecture converted and "Unmount" is a bit foo.
Linus Torvalds's avatar
Linus Torvalds committed
818 819 820
 * As far as I know, there weren't any dependencies on the old behaviour.
 */

821
static int
Linus Torvalds's avatar
Linus Torvalds committed
822 823
xfs_log_unmount_write(xfs_mount_t *mp)
{
Mark Tinguely's avatar
Mark Tinguely committed
824
	struct xlog	 *log = mp->m_log;
Linus Torvalds's avatar
Linus Torvalds committed
825 826 827 828
	xlog_in_core_t	 *iclog;
#ifdef DEBUG
	xlog_in_core_t	 *first_iclog;
#endif
829
	xlog_ticket_t	*tic = NULL;
Linus Torvalds's avatar
Linus Torvalds committed
830 831 832 833
	xfs_lsn_t	 lsn;
	int		 error;

	/*
834
	 * Don't write out unmount record on norecovery mounts or ro devices.
Linus Torvalds's avatar
Linus Torvalds committed
835 836
	 * Or, if we are doing a forced umount (typically because of IO errors).
	 */
837 838 839
	if (mp->m_flags & XFS_MOUNT_NORECOVERY ||
	    xfs_readonly_buftarg(log->l_mp->m_logdev_targp)) {
		ASSERT(mp->m_flags & XFS_MOUNT_RDONLY);
Linus Torvalds's avatar
Linus Torvalds committed
840
		return 0;
841
	}
Linus Torvalds's avatar
Linus Torvalds committed
842

843
	error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
844
	ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
Linus Torvalds's avatar
Linus Torvalds committed
845 846 847 848 849 850 851 852 853 854 855 856

#ifdef DEBUG
	first_iclog = iclog = log->l_iclog;
	do {
		if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
			ASSERT(iclog->ic_state & XLOG_STATE_ACTIVE);
			ASSERT(iclog->ic_offset == 0);
		}
		iclog = iclog->ic_next;
	} while (iclog != first_iclog);
#endif
	if (! (XLOG_FORCED_SHUTDOWN(log))) {
857
		error = xfs_log_reserve(mp, 600, 1, &tic, XFS_LOG, 0);
Linus Torvalds's avatar
Linus Torvalds committed
858
		if (!error) {
859 860 861 862 863 864 865 866 867
			/* the data section must be 32 bit size aligned */
			struct {
			    __uint16_t magic;
			    __uint16_t pad1;
			    __uint32_t pad2; /* may as well make it 64 bits */
			} magic = {
				.magic = XLOG_UNMOUNT_TYPE,
			};
			struct xfs_log_iovec reg = {
868
				.i_addr = &magic,
869 870 871 872 873 874 875 876
				.i_len = sizeof(magic),
				.i_type = XLOG_REG_TYPE_UNMOUNT,
			};
			struct xfs_log_vec vec = {
				.lv_niovecs = 1,
				.lv_iovecp = &reg,
			};

877
			/* remove inited flag, and account for space used */
878
			tic->t_flags = 0;
879
			tic->t_curr_res -= sizeof(magic);
880
			error = xlog_write(log, &vec, tic, &lsn,
Linus Torvalds's avatar
Linus Torvalds committed
881 882 883 884 885 886 887 888
					   NULL, XLOG_UNMOUNT_TRANS);
			/*
			 * At this point, we're umounting anyway,
			 * so there's no point in transitioning log state
			 * to IOERROR. Just continue...
			 */
		}

889 890
		if (error)
			xfs_alert(mp, "%s: unmount record failed", __func__);
Linus Torvalds's avatar
Linus Torvalds committed
891 892


Eric Sandeen's avatar
Eric Sandeen committed
893
		spin_lock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
894
		iclog = log->l_iclog;
895
		atomic_inc(&iclog->ic_refcnt);
Linus Torvalds's avatar
Linus Torvalds committed
896
		xlog_state_want_sync(log, iclog);
897
		spin_unlock(&log->l_icloglock);
898
		error = xlog_state_release_iclog(log, iclog);
Linus Torvalds's avatar
Linus Torvalds committed
899

Eric Sandeen's avatar
Eric Sandeen committed
900
		spin_lock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
901 902 903
		if (!(iclog->ic_state == XLOG_STATE_ACTIVE ||
		      iclog->ic_state == XLOG_STATE_DIRTY)) {
			if (!XLOG_FORCED_SHUTDOWN(log)) {
904 905
				xlog_wait(&iclog->ic_force_wait,
							&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
906
			} else {
Eric Sandeen's avatar
Eric Sandeen committed
907
				spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
908 909
			}
		} else {
Eric Sandeen's avatar
Eric Sandeen committed
910
			spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
911
		}
912
		if (tic) {
913
			trace_xfs_log_umount_write(log, tic);
914
			xlog_ungrant_log_space(log, tic);
915
			xfs_log_ticket_put(tic);
916
		}
Linus Torvalds's avatar
Linus Torvalds committed
917 918 919 920 921 922 923
	} else {
		/*
		 * We're already in forced_shutdown mode, couldn't
		 * even attempt to write out the unmount transaction.
		 *
		 * Go through the motions of sync'ing and releasing
		 * the iclog, even though no I/O will actually happen,
924
		 * we need to wait for other log I/Os that may already
Linus Torvalds's avatar
Linus Torvalds committed
925 926 927 928 929 930
		 * be in progress.  Do this as a separate section of
		 * code so we'll know if we ever get stuck here that
		 * we're in this odd situation of trying to unmount
		 * a file system that went into forced_shutdown as
		 * the result of an unmount..
		 */
Eric Sandeen's avatar
Eric Sandeen committed
931
		spin_lock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
932
		iclog = log->l_iclog;
933
		atomic_inc(&iclog->ic_refcnt);
Linus Torvalds's avatar
Linus Torvalds committed
934 935

		xlog_state_want_sync(log, iclog);
936
		spin_unlock(&log->l_icloglock);
937
		error =  xlog_state_release_iclog(log, iclog);
Linus Torvalds's avatar
Linus Torvalds committed
938

Eric Sandeen's avatar
Eric Sandeen committed
939
		spin_lock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
940 941 942 943 944

		if ( ! (   iclog->ic_state == XLOG_STATE_ACTIVE
			|| iclog->ic_state == XLOG_STATE_DIRTY
			|| iclog->ic_state == XLOG_STATE_IOERROR) ) {

945 946
				xlog_wait(&iclog->ic_force_wait,
							&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
947
		} else {
Eric Sandeen's avatar
Eric Sandeen committed
948
			spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
949 950 951
		}
	}

952
	return error;
Linus Torvalds's avatar
Linus Torvalds committed
953 954 955
}	/* xfs_log_unmount_write */

/*
956
 * Empty the log for unmount/freeze.
957 958 959 960 961
 *
 * To do this, we first need to shut down the background log work so it is not
 * trying to cover the log as we clean up. We then need to unpin all objects in
 * the log so we can then flush them out. Once they have completed their IO and
 * run the callbacks removing themselves from the AIL, we can write the unmount
962
 * record.
Linus Torvalds's avatar
Linus Torvalds committed
963 964
 */
void
965 966
xfs_log_quiesce(
	struct xfs_mount	*mp)
Linus Torvalds's avatar
Linus Torvalds committed
967
{
968
	cancel_delayed_work_sync(&mp->m_log->l_work);
969 970 971 972 973 974 975 976 977 978 979 980 981 982 983
	xfs_log_force(mp, XFS_LOG_SYNC);

	/*
	 * The superblock buffer is uncached and while xfs_ail_push_all_sync()
	 * will push it, xfs_wait_buftarg() will not wait for it. Further,
	 * xfs_buf_iowait() cannot be used because it was pushed with the
	 * XBF_ASYNC flag set, so we need to use a lock/unlock pair to wait for
	 * the IO to complete.
	 */
	xfs_ail_push_all_sync(mp->m_ail);
	xfs_wait_buftarg(mp->m_ddev_targp);
	xfs_buf_lock(mp->m_sb_bp);
	xfs_buf_unlock(mp->m_sb_bp);

	xfs_log_unmount_write(mp);
984 985 986 987 988 989 990 991 992 993 994 995 996 997
}

/*
 * Shut down and release the AIL and Log.
 *
 * During unmount, we need to ensure we flush all the dirty metadata objects
 * from the AIL so that the log is empty before we write the unmount record to
 * the log. Once this is done, we can tear down the AIL and the log.
 */
void
xfs_log_unmount(
	struct xfs_mount	*mp)
{
	xfs_log_quiesce(mp);
998

999
	xfs_trans_ail_destroy(mp);
1000 1001 1002

	xfs_sysfs_del(&mp->m_log->l_kobj);

1003
	xlog_dealloc_log(mp->m_log);
Linus Torvalds's avatar
Linus Torvalds committed
1004 1005
}

1006 1007 1008 1009 1010
void
xfs_log_item_init(
	struct xfs_mount	*mp,
	struct xfs_log_item	*item,
	int			type,
1011
	const struct xfs_item_ops *ops)
1012 1013 1014 1015 1016
{
	item->li_mountp = mp;
	item->li_ailp = mp->m_ail;
	item->li_type = type;
	item->li_ops = ops;
1017 1018 1019 1020
	item->li_lv = NULL;

	INIT_LIST_HEAD(&item->li_ail);
	INIT_LIST_HEAD(&item->li_cil);
1021 1022
}

1023 1024 1025
/*
 * Wake up processes waiting for log space after we have moved the log tail.
 */
Linus Torvalds's avatar
Linus Torvalds committed
1026
void
1027
xfs_log_space_wake(
1028
	struct xfs_mount	*mp)
Linus Torvalds's avatar
Linus Torvalds committed
1029
{
1030
	struct xlog		*log = mp->m_log;
1031
	int			free_bytes;
Linus Torvalds's avatar
Linus Torvalds committed
1032 1033 1034 1035

	if (XLOG_FORCED_SHUTDOWN(log))
		return;

1036
	if (!list_empty_careful(&log->l_write_head.waiters)) {
1037 1038
		ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));

1039 1040
		spin_lock(&log->l_write_head.lock);
		free_bytes = xlog_space_left(log, &log->l_write_head.grant);
1041
		xlog_grant_head_wake(log, &log->l_write_head, &free_bytes);
1042
		spin_unlock(&log->l_write_head.lock);
Linus Torvalds's avatar
Linus Torvalds committed
1043
	}
1044

1045
	if (!list_empty_careful(&log->l_reserve_head.waiters)) {
1046 1047
		ASSERT(!(log->l_flags & XLOG_ACTIVE_RECOVERY));

1048 1049
		spin_lock(&log->l_reserve_head.lock);
		free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
1050
		xlog_grant_head_wake(log, &log->l_reserve_head, &free_bytes);
1051
		spin_unlock(&log->l_reserve_head.lock);
Linus Torvalds's avatar
Linus Torvalds committed
1052
	}
1053
}
Linus Torvalds's avatar
Linus Torvalds committed
1054 1055

/*
1056 1057 1058 1059
 * Determine if we have a transaction that has gone to disk that needs to be
 * covered. To begin the transition to the idle state firstly the log needs to
 * be idle. That means the CIL, the AIL and the iclogs needs to be empty before
 * we start attempting to cover the log.
1060
 *
1061 1062 1063 1064 1065 1066 1067 1068 1069 1070
 * Only if we are then in a state where covering is needed, the caller is
 * informed that dummy transactions are required to move the log into the idle
 * state.
 *
 * If there are any items in the AIl or CIL, then we do not want to attempt to
 * cover the log as we may be in a situation where there isn't log space
 * available to run a dummy transaction and this can lead to deadlocks when the
 * tail of the log is pinned by an item that is modified in the CIL.  Hence
 * there's no point in running a dummy transaction at this point because we
 * can't start trying to idle the log until both the CIL and AIL are empty.
Linus Torvalds's avatar
Linus Torvalds committed
1071
 */
1072
static int
Linus Torvalds's avatar
Linus Torvalds committed
1073 1074
xfs_log_need_covered(xfs_mount_t *mp)
{
Mark Tinguely's avatar
Mark Tinguely committed
1075
	struct xlog	*log = mp->m_log;
1076
	int		needed = 0;
Linus Torvalds's avatar
Linus Torvalds committed
1077

1078
	if (!xfs_fs_writable(mp, SB_FREEZE_WRITE))
Linus Torvalds's avatar
Linus Torvalds committed
1079 1080
		return 0;

1081 1082 1083
	if (!xlog_cil_empty(log))
		return 0;

Eric Sandeen's avatar
Eric Sandeen committed
1084
	spin_lock(&log->l_icloglock);
1085 1086 1087 1088 1089 1090 1091
	switch (log->l_covered_state) {
	case XLOG_STATE_COVER_DONE:
	case XLOG_STATE_COVER_DONE2:
	case XLOG_STATE_COVER_IDLE:
		break;
	case XLOG_STATE_COVER_NEED:
	case XLOG_STATE_COVER_NEED2:
1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
		if (xfs_ail_min_lsn(log->l_ailp))
			break;
		if (!xlog_iclogs_empty(log))
			break;

		needed = 1;
		if (log->l_covered_state == XLOG_STATE_COVER_NEED)
			log->l_covered_state = XLOG_STATE_COVER_DONE;
		else
			log->l_covered_state = XLOG_STATE_COVER_DONE2;
		break;
1103
	default:
Linus Torvalds's avatar
Linus Torvalds committed
1104
		needed = 1;
1105
		break;
Linus Torvalds's avatar
Linus Torvalds committed
1106
	}
Eric Sandeen's avatar
Eric Sandeen committed
1107
	spin_unlock(&log->l_icloglock);
1108
	return needed;
Linus Torvalds's avatar
Linus Torvalds committed
1109 1110
}

1111
/*
Linus Torvalds's avatar
Linus Torvalds committed
1112 1113 1114
 * We may be holding the log iclog lock upon entering this routine.
 */
xfs_lsn_t
1115
xlog_assign_tail_lsn_locked(
1116
	struct xfs_mount	*mp)
Linus Torvalds's avatar
Linus Torvalds committed
1117
{
1118
	struct xlog		*log = mp->m_log;
1119 1120 1121 1122
	struct xfs_log_item	*lip;
	xfs_lsn_t		tail_lsn;

	assert_spin_locked(&mp->m_ail->xa_lock);
Linus Torvalds's avatar
Linus Torvalds committed
1123

1124 1125 1126
	/*
	 * To make sure we always have a valid LSN for the log tail we keep
	 * track of the last LSN which was committed in log->l_last_sync_lsn,
1127
	 * and use that when the AIL was empty.
1128
	 */
1129 1130 1131 1132
	lip = xfs_ail_min(mp->m_ail);
	if (lip)
		tail_lsn = lip->li_lsn;
	else
1133
		tail_lsn = atomic64_read(&log->l_last_sync_lsn);
Dave Chinner's avatar
Dave Chinner committed
1134
	trace_xfs_log_assign_tail_lsn(log, tail_lsn);
1135
	atomic64_set(&log->l_tail_lsn, tail_lsn);
Linus Torvalds's avatar
Linus Torvalds committed
1136
	return tail_lsn;
1137
}
Linus Torvalds's avatar
Linus Torvalds committed
1138

1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151
xfs_lsn_t
xlog_assign_tail_lsn(
	struct xfs_mount	*mp)
{
	xfs_lsn_t		tail_lsn;

	spin_lock(&mp->m_ail->xa_lock);
	tail_lsn = xlog_assign_tail_lsn_locked(mp);
	spin_unlock(&mp->m_ail->xa_lock);

	return tail_lsn;
}

Linus Torvalds's avatar
Linus Torvalds committed
1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165
/*
 * Return the space in the log between the tail and the head.  The head
 * is passed in the cycle/bytes formal parms.  In the special case where
 * the reserve head has wrapped passed the tail, this calculation is no
 * longer valid.  In this case, just return 0 which means there is no space
 * in the log.  This works for all places where this function is called
 * with the reserve head.  Of course, if the write head were to ever
 * wrap the tail, we should blow up.  Rather than catch this case here,
 * we depend on other ASSERTions in other parts of the code.   XXXmiken
 *
 * This code also handles the case where the reservation head is behind
 * the tail.  The details of this case are described below, but the end
 * result is that we return the size of the log as the amount of space left.
 */
1166
STATIC int
1167
xlog_space_left(
1168
	struct xlog	*log,
1169
	atomic64_t	*head)
Linus Torvalds's avatar
Linus Torvalds committed
1170
{
1171 1172 1173 1174 1175
	int		free_bytes;
	int		tail_bytes;
	int		tail_cycle;
	int		head_cycle;
	int		head_bytes;
Linus Torvalds's avatar
Linus Torvalds committed
1176

1177
	xlog_crack_grant_head(head, &head_cycle, &head_bytes);
1178 1179
	xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_bytes);
	tail_bytes = BBTOB(tail_bytes);
1180 1181 1182
	if (tail_cycle == head_cycle && head_bytes >= tail_bytes)
		free_bytes = log->l_logsize - (head_bytes - tail_bytes);
	else if (tail_cycle + 1 < head_cycle)
Linus Torvalds's avatar
Linus Torvalds committed
1183
		return 0;
1184 1185 1186
	else if (tail_cycle < head_cycle) {
		ASSERT(tail_cycle == (head_cycle - 1));
		free_bytes = tail_bytes - head_bytes;
Linus Torvalds's avatar
Linus Torvalds committed
1187 1188 1189 1190 1191 1192
	} else {
		/*
		 * The reservation head is behind the tail.
		 * In this case we just want to return the size of the
		 * log as the amount of space left.
		 */
1193
		xfs_alert(log->l_mp, "xlog_space_left: head behind tail");
1194
		xfs_alert(log->l_mp,
1195 1196 1197 1198 1199
			  "  tail_cycle = %d, tail_bytes = %d",
			  tail_cycle, tail_bytes);
		xfs_alert(log->l_mp,
			  "  GH   cycle = %d, GH   bytes = %d",
			  head_cycle, head_bytes);
Linus Torvalds's avatar
Linus Torvalds committed
1200 1201 1202 1203
		ASSERT(0);
		free_bytes = log->l_logsize;
	}
	return free_bytes;
1204
}
Linus Torvalds's avatar
Linus Torvalds committed
1205 1206 1207 1208 1209 1210 1211 1212


/*
 * Log function which is called when an io completes.
 *
 * The log manager needs its own routine, in order to control what
 * happens with the buffer after the write completes.
 */
1213
static void
Linus Torvalds's avatar
Linus Torvalds committed
1214 1215
xlog_iodone(xfs_buf_t *bp)
{
Mark Tinguely's avatar
Mark Tinguely committed
1216 1217 1218
	struct xlog_in_core	*iclog = bp->b_fspriv;
	struct xlog		*l = iclog->ic_log;
	int			aborted = 0;
Linus Torvalds's avatar
Linus Torvalds committed
1219 1220

	/*
1221 1222 1223
	 * Race to shutdown the filesystem if we see an error or the iclog is in
	 * IOABORT state. The IOABORT state is only set in DEBUG mode to inject
	 * CRC errors into log recovery.
Linus Torvalds's avatar
Linus Torvalds committed
1224
	 */
1225 1226 1227 1228 1229 1230
	if (XFS_TEST_ERROR(bp->b_error, l->l_mp, XFS_ERRTAG_IODONE_IOERR,
			   XFS_RANDOM_IODONE_IOERR) ||
	    iclog->ic_state & XLOG_STATE_IOABORT) {
		if (iclog->ic_state & XLOG_STATE_IOABORT)
			iclog->ic_state &= ~XLOG_STATE_IOABORT;

1231
		xfs_buf_ioerror_alert(bp, __func__);
1232
		xfs_buf_stale(bp);
1233
		xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR);
Linus Torvalds's avatar
Linus Torvalds committed
1234 1235 1236 1237 1238 1239 1240 1241 1242
		/*
		 * This flag will be propagated to the trans-committed
		 * callback routines to let them know that the log-commit
		 * didn't succeed.
		 */
		aborted = XFS_LI_ABORTED;
	} else if (iclog->ic_state & XLOG_STATE_IOERROR) {
		aborted = XFS_LI_ABORTED;
	}
1243 1244

	/* log I/O is always issued ASYNC */
1245
	ASSERT(bp->b_flags & XBF_ASYNC);
Linus Torvalds's avatar
Linus Torvalds committed
1246
	xlog_state_done_syncing(iclog, aborted);
1247

1248
	/*
1249 1250 1251 1252
	 * drop the buffer lock now that we are done. Nothing references
	 * the buffer after this, so an unmount waiting on this lock can now
	 * tear it down safely. As such, it is unsafe to reference the buffer
	 * (bp) after the unlock as we could race with it being freed.
1253
	 */
1254
	xfs_buf_unlock(bp);
1255
}
Linus Torvalds's avatar
Linus Torvalds committed
1256 1257 1258 1259

/*
 * Return size of each in-core log record buffer.
 *
Malcolm Parsons's avatar
Malcolm Parsons committed
1260
 * All machines get 8 x 32kB buffers by default, unless tuned otherwise.
Linus Torvalds's avatar
Linus Torvalds committed
1261 1262 1263 1264 1265 1266
 *
 * If the filesystem blocksize is too large, we may need to choose a
 * larger size since the directory code currently logs entire blocks.
 */

STATIC void
Mark Tinguely's avatar
Mark Tinguely committed
1267 1268 1269
xlog_get_iclog_buffer_size(
	struct xfs_mount	*mp,
	struct xlog		*log)
Linus Torvalds's avatar
Linus Torvalds committed
1270 1271 1272 1273
{
	int size;
	int xhdrs;

1274 1275 1276
	if (mp->m_logbufs <= 0)
		log->l_iclog_bufs = XLOG_MAX_ICLOGS;
	else
1277
		log->l_iclog_bufs = mp->m_logbufs;
Linus Torvalds's avatar
Linus Torvalds committed
1278 1279 1280 1281

	/*
	 * Buffer size passed in from mount system call.
	 */
1282
	if (mp->m_logbsize > 0) {
Linus Torvalds's avatar
Linus Torvalds committed
1283 1284 1285 1286 1287 1288 1289
		size = log->l_iclog_size = mp->m_logbsize;
		log->l_iclog_size_log = 0;
		while (size != 1) {
			log->l_iclog_size_log++;
			size >>= 1;
		}

1290
		if (xfs_sb_version_haslogv2(&mp->m_sb)) {
Malcolm Parsons's avatar
Malcolm Parsons committed
1291 1292
			/* # headers = size / 32k
			 * one header holds cycles from 32k of data
Linus Torvalds's avatar
Linus Torvalds committed
1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304
			 */

			xhdrs = mp->m_logbsize / XLOG_HEADER_CYCLE_SIZE;
			if (mp->m_logbsize % XLOG_HEADER_CYCLE_SIZE)
				xhdrs++;
			log->l_iclog_hsize = xhdrs << BBSHIFT;
			log->l_iclog_heads = xhdrs;
		} else {
			ASSERT(mp->m_logbsize <= XLOG_BIG_RECORD_BSIZE);
			log->l_iclog_hsize = BBSIZE;
			log->l_iclog_heads = 1;
		}
1305
		goto done;
Linus Torvalds's avatar
Linus Torvalds committed
1306 1307
	}

Malcolm Parsons's avatar
Malcolm Parsons committed
1308
	/* All machines use 32kB buffers by default. */
1309 1310
	log->l_iclog_size = XLOG_BIG_RECORD_BSIZE;
	log->l_iclog_size_log = XLOG_BIG_RECORD_BSHIFT;
Linus Torvalds's avatar
Linus Torvalds committed
1311 1312 1313 1314 1315

	/* the default log size is 16k or 32k which is one header sector */
	log->l_iclog_hsize = BBSIZE;
	log->l_iclog_heads = 1;

1316 1317
done:
	/* are we being asked to make the sizes selected above visible? */
1318 1319 1320 1321
	if (mp->m_logbufs == 0)
		mp->m_logbufs = log->l_iclog_bufs;
	if (mp->m_logbsize == 0)
		mp->m_logbsize = log->l_iclog_size;
Linus Torvalds's avatar
Linus Torvalds committed
1322 1323 1324
}	/* xlog_get_iclog_buffer_size */


1325 1326 1327 1328
void
xfs_log_work_queue(
	struct xfs_mount        *mp)
{
1329
	queue_delayed_work(mp->m_sync_workqueue, &mp->m_log->l_work,
1330 1331 1332 1333 1334 1335 1336 1337
				msecs_to_jiffies(xfs_syncd_centisecs * 10));
}

/*
 * Every sync period we need to unpin all items in the AIL and push them to
 * disk. If there is nothing dirty, then we might need to cover the log to
 * indicate that the filesystem is idle.
 */
1338
static void
1339 1340 1341 1342 1343 1344 1345 1346
xfs_log_worker(
	struct work_struct	*work)
{
	struct xlog		*log = container_of(to_delayed_work(work),
						struct xlog, l_work);
	struct xfs_mount	*mp = log->l_mp;

	/* dgc: errors ignored - not fatal and nowhere to report them */
1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360
	if (xfs_log_need_covered(mp)) {
		/*
		 * Dump a transaction into the log that contains no real change.
		 * This is needed to stamp the current tail LSN into the log
		 * during the covering operation.
		 *
		 * We cannot use an inode here for this - that will push dirty
		 * state back up into the VFS and then periodic inode flushing
		 * will prevent log covering from making progress. Hence we
		 * synchronously log the superblock instead to ensure the
		 * superblock is immediately unpinned and can be written back.
		 */
		xfs_sync_sb(mp, true);
	} else
1361 1362 1363 1364 1365 1366 1367 1368 1369
		xfs_log_force(mp, 0);

	/* start pushing all the metadata that is currently dirty */
	xfs_ail_push_all(mp->m_ail);

	/* queue us up again */
	xfs_log_work_queue(mp);
}

Linus Torvalds's avatar
Linus Torvalds committed
1370 1371 1372 1373 1374
/*
 * This routine initializes some of the log structure for a given mount point.
 * Its primary purpose is to fill in enough, so recovery can occur.  However,
 * some other stuff may be filled in too.
 */
Mark Tinguely's avatar
Mark Tinguely committed
1375 1376 1377 1378 1379 1380
STATIC struct xlog *
xlog_alloc_log(
	struct xfs_mount	*mp,
	struct xfs_buftarg	*log_target,
	xfs_daddr_t		blk_offset,
	int			num_bblks)
Linus Torvalds's avatar
Linus Torvalds committed
1381
{
Mark Tinguely's avatar
Mark Tinguely committed
1382
	struct xlog		*log;
Linus Torvalds's avatar
Linus Torvalds committed
1383 1384 1385 1386 1387
	xlog_rec_header_t	*head;
	xlog_in_core_t		**iclogp;
	xlog_in_core_t		*iclog, *prev_iclog=NULL;
	xfs_buf_t		*bp;
	int			i;
1388
	int			error = -ENOMEM;
1389
	uint			log2_size = 0;
Linus Torvalds's avatar
Linus Torvalds committed
1390

Mark Tinguely's avatar
Mark Tinguely committed
1391
	log = kmem_zalloc(sizeof(struct xlog), KM_MAYFAIL);
1392
	if (!log) {
1393
		xfs_warn(mp, "Log allocation failed: No memory!");
1394 1395
		goto out;
	}
Linus Torvalds's avatar
Linus Torvalds committed
1396 1397 1398 1399 1400 1401 1402 1403

	log->l_mp	   = mp;
	log->l_targ	   = log_target;
	log->l_logsize     = BBTOB(num_bblks);
	log->l_logBBstart  = blk_offset;
	log->l_logBBsize   = num_bblks;
	log->l_covered_state = XLOG_STATE_COVER_IDLE;
	log->l_flags	   |= XLOG_ACTIVE_RECOVERY;
1404
	INIT_DELAYED_WORK(&log->l_work, xfs_log_worker);
Linus Torvalds's avatar
Linus Torvalds committed
1405 1406 1407

	log->l_prev_block  = -1;
	/* log->l_tail_lsn = 0x100000000LL; cycle = 1; current block = 0 */
1408 1409
	xlog_assign_atomic_lsn(&log->l_tail_lsn, 1, 0);
	xlog_assign_atomic_lsn(&log->l_last_sync_lsn, 1, 0);
Linus Torvalds's avatar
Linus Torvalds committed
1410
	log->l_curr_cycle  = 1;	    /* 0 is bad since this is initial value */
1411 1412 1413

	xlog_grant_head_init(&log->l_reserve_head);
	xlog_grant_head_init(&log->l_write_head);
Linus Torvalds's avatar
Linus Torvalds committed
1414

1415
	error = -EFSCORRUPTED;
1416
	if (xfs_sb_version_hassector(&mp->m_sb)) {
1417 1418
	        log2_size = mp->m_sb.sb_logsectlog;
		if (log2_size < BBSHIFT) {
1419 1420
			xfs_warn(mp, "Log sector size too small (0x%x < 0x%x)",
				log2_size, BBSHIFT);
1421 1422 1423
			goto out_free_log;
		}

1424 1425
	        log2_size -= BBSHIFT;
		if (log2_size > mp->m_sectbb_log) {
1426 1427
			xfs_warn(mp, "Log sector size too large (0x%x > 0x%x)",
				log2_size, mp->m_sectbb_log);
1428 1429
			goto out_free_log;
		}
1430 1431 1432 1433

		/* for larger sector sizes, must have v2 or external log */
		if (log2_size && log->l_logBBstart > 0 &&
			    !xfs_sb_version_haslogv2(&mp->m_sb)) {
1434 1435 1436
			xfs_warn(mp,
		"log sector size (0x%x) invalid for configuration.",
				log2_size);
1437 1438
			goto out_free_log;
		}
Linus Torvalds's avatar
Linus Torvalds committed
1439
	}
1440
	log->l_sectBBsize = 1 << log2_size;
Linus Torvalds's avatar
Linus Torvalds committed
1441 1442 1443

	xlog_get_iclog_buffer_size(mp, log);

1444 1445 1446 1447 1448
	/*
	 * Use a NULL block for the extra log buffer used during splits so that
	 * it will trigger errors if we ever try to do IO on it without first
	 * having set it up properly.
	 */
1449
	error = -ENOMEM;
1450
	bp = xfs_buf_alloc(mp->m_logdev_targp, XFS_BUF_DADDR_NULL,
1451
			   BTOBB(log->l_iclog_size), XBF_NO_IOACCT);
1452 1453
	if (!bp)
		goto out_free_log;
1454 1455 1456 1457 1458 1459

	/*
	 * The iclogbuf buffer locks are held over IO but we are not going to do
	 * IO yet.  Hence unlock the buffer so that the log IO path can grab it
	 * when appropriately.
	 */
1460
	ASSERT(xfs_buf_islocked(bp));
1461 1462
	xfs_buf_unlock(bp);

1463 1464
	/* use high priority wq for log I/O completion */
	bp->b_ioend_wq = mp->m_log_workqueue;
1465
	bp->b_iodone = xlog_iodone;
Linus Torvalds's avatar
Linus Torvalds committed
1466 1467
	log->l_xbuf = bp;

Eric Sandeen's avatar
Eric Sandeen committed
1468
	spin_lock_init(&log->l_icloglock);
1469
	init_waitqueue_head(&log->l_flush_wait);
Linus Torvalds's avatar
Linus Torvalds committed
1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480

	iclogp = &log->l_iclog;
	/*
	 * The amount of memory to allocate for the iclog structure is
	 * rather funky due to the way the structure is defined.  It is
	 * done this way so that we can use different sizes for machines
	 * with different amounts of memory.  See the definition of
	 * xlog_in_core_t in xfs_log_priv.h for details.
	 */
	ASSERT(log->l_iclog_size >= 4096);
	for (i=0; i < log->l_iclog_bufs; i++) {
1481 1482 1483 1484
		*iclogp = kmem_zalloc(sizeof(xlog_in_core_t), KM_MAYFAIL);
		if (!*iclogp)
			goto out_free_iclog;

Linus Torvalds's avatar
Linus Torvalds committed
1485 1486 1487
		iclog = *iclogp;
		iclog->ic_prev = prev_iclog;
		prev_iclog = iclog;
1488

1489
		bp = xfs_buf_get_uncached(mp->m_logdev_targp,
1490 1491
					  BTOBB(log->l_iclog_size),
					  XBF_NO_IOACCT);
1492 1493
		if (!bp)
			goto out_free_iclog;
1494

1495 1496 1497
		ASSERT(xfs_buf_islocked(bp));
		xfs_buf_unlock(bp);

1498 1499
		/* use high priority wq for log I/O completion */
		bp->b_ioend_wq = mp->m_log_workqueue;
1500
		bp->b_iodone = xlog_iodone;
1501
		iclog->ic_bp = bp;
1502
		iclog->ic_data = bp->b_addr;
1503
#ifdef DEBUG
1504
		log->l_iclog_bak[i] = &iclog->ic_header;
1505
#endif
Linus Torvalds's avatar
Linus Torvalds committed
1506 1507
		head = &iclog->ic_header;
		memset(head, 0, sizeof(xlog_rec_header_t));
1508 1509
		head->h_magicno = cpu_to_be32(XLOG_HEADER_MAGIC_NUM);
		head->h_version = cpu_to_be32(
1510
			xfs_sb_version_haslogv2(&log->l_mp->m_sb) ? 2 : 1);
1511
		head->h_size = cpu_to_be32(log->l_iclog_size);
Linus Torvalds's avatar
Linus Torvalds committed
1512
		/* new fields */
1513
		head->h_fmt = cpu_to_be32(XLOG_FMT);
Linus Torvalds's avatar
Linus Torvalds committed
1514 1515
		memcpy(&head->h_fs_uuid, &mp->m_sb.sb_uuid, sizeof(uuid_t));

1516
		iclog->ic_size = BBTOB(bp->b_length) - log->l_iclog_hsize;
Linus Torvalds's avatar
Linus Torvalds committed
1517 1518
		iclog->ic_state = XLOG_STATE_ACTIVE;
		iclog->ic_log = log;
1519 1520
		atomic_set(&iclog->ic_refcnt, 0);
		spin_lock_init(&iclog->ic_callback_lock);
Linus Torvalds's avatar
Linus Torvalds committed
1521
		iclog->ic_callback_tail = &(iclog->ic_callback);
1522
		iclog->ic_datap = (char *)iclog->ic_data + log->l_iclog_hsize;
Linus Torvalds's avatar
Linus Torvalds committed
1523

1524 1525
		init_waitqueue_head(&iclog->ic_force_wait);
		init_waitqueue_head(&iclog->ic_write_wait);
Linus Torvalds's avatar
Linus Torvalds committed
1526 1527 1528 1529 1530 1531

		iclogp = &iclog->ic_next;
	}
	*iclogp = log->l_iclog;			/* complete ring */
	log->l_iclog->ic_prev = prev_iclog;	/* re-write 1st prev ptr */

1532 1533 1534
	error = xlog_cil_init(log);
	if (error)
		goto out_free_iclog;
Linus Torvalds's avatar
Linus Torvalds committed
1535
	return log;
1536 1537 1538 1539

out_free_iclog:
	for (iclog = log->l_iclog; iclog; iclog = prev_iclog) {
		prev_iclog = iclog->ic_next;
1540
		if (iclog->ic_bp)
1541 1542
			xfs_buf_free(iclog->ic_bp);
		kmem_free(iclog);
1543 1544
		if (prev_iclog == log->l_iclog)
			break;
1545 1546 1547 1548 1549
	}
	spinlock_destroy(&log->l_icloglock);
	xfs_buf_free(log->l_xbuf);
out_free_log:
	kmem_free(log);
1550
out:
1551
	return ERR_PTR(error);
Linus Torvalds's avatar
Linus Torvalds committed
1552 1553 1554 1555 1556 1557 1558 1559
}	/* xlog_alloc_log */


/*
 * Write out the commit record of a transaction associated with the given
 * ticket.  Return the lsn of the commit record.
 */
STATIC int
1560
xlog_commit_record(
1561
	struct xlog		*log,
1562 1563 1564
	struct xlog_ticket	*ticket,
	struct xlog_in_core	**iclog,
	xfs_lsn_t		*commitlsnp)
Linus Torvalds's avatar
Linus Torvalds committed
1565
{
1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576
	struct xfs_mount *mp = log->l_mp;
	int	error;
	struct xfs_log_iovec reg = {
		.i_addr = NULL,
		.i_len = 0,
		.i_type = XLOG_REG_TYPE_COMMIT,
	};
	struct xfs_log_vec vec = {
		.lv_niovecs = 1,
		.lv_iovecp = &reg,
	};
Linus Torvalds's avatar
Linus Torvalds committed
1577 1578

	ASSERT_ALWAYS(iclog);
1579 1580 1581
	error = xlog_write(log, &vec, ticket, commitlsnp, iclog,
					XLOG_COMMIT_TRANS);
	if (error)
1582
		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
1583
	return error;
1584
}
Linus Torvalds's avatar
Linus Torvalds committed
1585 1586 1587 1588 1589 1590 1591 1592

/*
 * Push on the buffer cache code if we ever use more than 75% of the on-disk
 * log space.  This code pushes on the lsn which would supposedly free up
 * the 25% which we want to leave free.  We may need to adopt a policy which
 * pushes on an lsn which is further along in the log once we reach the high
 * water mark.  In this manner, we would be creating a low water mark.
 */
1593
STATIC void
1594
xlog_grant_push_ail(
1595
	struct xlog	*log,
1596
	int		need_bytes)
Linus Torvalds's avatar
Linus Torvalds committed
1597
{
1598
	xfs_lsn_t	threshold_lsn = 0;
1599
	xfs_lsn_t	last_sync_lsn;
1600 1601 1602 1603 1604 1605 1606 1607
	int		free_blocks;
	int		free_bytes;
	int		threshold_block;
	int		threshold_cycle;
	int		free_threshold;

	ASSERT(BTOBB(need_bytes) < log->l_logBBsize);

1608
	free_bytes = xlog_space_left(log, &log->l_reserve_head.grant);
1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621
	free_blocks = BTOBBT(free_bytes);

	/*
	 * Set the threshold for the minimum number of free blocks in the
	 * log to the maximum of what the caller needs, one quarter of the
	 * log, and 256 blocks.
	 */
	free_threshold = BTOBB(need_bytes);
	free_threshold = MAX(free_threshold, (log->l_logBBsize >> 2));
	free_threshold = MAX(free_threshold, 256);
	if (free_blocks >= free_threshold)
		return;

1622 1623 1624
	xlog_crack_atomic_lsn(&log->l_tail_lsn, &threshold_cycle,
						&threshold_block);
	threshold_block += free_threshold;
Linus Torvalds's avatar
Linus Torvalds committed
1625
	if (threshold_block >= log->l_logBBsize) {
1626 1627
		threshold_block -= log->l_logBBsize;
		threshold_cycle += 1;
Linus Torvalds's avatar
Linus Torvalds committed
1628
	}
1629 1630 1631 1632
	threshold_lsn = xlog_assign_lsn(threshold_cycle,
					threshold_block);
	/*
	 * Don't pass in an lsn greater than the lsn of the last
1633 1634
	 * log record known to be on disk. Use a snapshot of the last sync lsn
	 * so that it doesn't change between the compare and the set.
Linus Torvalds's avatar
Linus Torvalds committed
1635
	 */
1636 1637 1638
	last_sync_lsn = atomic64_read(&log->l_last_sync_lsn);
	if (XFS_LSN_CMP(threshold_lsn, last_sync_lsn) > 0)
		threshold_lsn = last_sync_lsn;
1639 1640 1641 1642 1643 1644 1645

	/*
	 * Get the transaction layer to kick the dirty buffers out to
	 * disk asynchronously. No point in trying to do this if
	 * the filesystem is shutting down.
	 */
	if (!XLOG_FORCED_SHUTDOWN(log))
1646
		xfs_ail_push(log->l_ailp, threshold_lsn);
1647
}
Linus Torvalds's avatar
Linus Torvalds committed
1648

1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660
/*
 * Stamp cycle number in every block
 */
STATIC void
xlog_pack_data(
	struct xlog		*log,
	struct xlog_in_core	*iclog,
	int			roundoff)
{
	int			i, j, k;
	int			size = iclog->ic_offset + roundoff;
	__be32			cycle_lsn;
1661
	char			*dp;
1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695

	cycle_lsn = CYCLE_LSN_DISK(iclog->ic_header.h_lsn);

	dp = iclog->ic_datap;
	for (i = 0; i < BTOBB(size); i++) {
		if (i >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE))
			break;
		iclog->ic_header.h_cycle_data[i] = *(__be32 *)dp;
		*(__be32 *)dp = cycle_lsn;
		dp += BBSIZE;
	}

	if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
		xlog_in_core_2_t *xhdr = iclog->ic_data;

		for ( ; i < BTOBB(size); i++) {
			j = i / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
			k = i % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
			xhdr[j].hic_xheader.xh_cycle_data[k] = *(__be32 *)dp;
			*(__be32 *)dp = cycle_lsn;
			dp += BBSIZE;
		}

		for (i = 1; i < log->l_iclog_heads; i++)
			xhdr[i].hic_xheader.xh_cycle = cycle_lsn;
	}
}

/*
 * Calculate the checksum for a log buffer.
 *
 * This is a little more complicated than it should be because the various
 * headers and the actual data are non-contiguous.
 */
1696
__le32
1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713
xlog_cksum(
	struct xlog		*log,
	struct xlog_rec_header	*rhead,
	char			*dp,
	int			size)
{
	__uint32_t		crc;

	/* first generate the crc for the record header ... */
	crc = xfs_start_cksum((char *)rhead,
			      sizeof(struct xlog_rec_header),
			      offsetof(struct xlog_rec_header, h_crc));

	/* ... then for additional cycle data for v2 logs ... */
	if (xfs_sb_version_haslogv2(&log->l_mp->m_sb)) {
		union xlog_in_core2 *xhdr = (union xlog_in_core2 *)rhead;
		int		i;
1714
		int		xheads;
1715

1716 1717 1718
		xheads = size / XLOG_HEADER_CYCLE_SIZE;
		if (size % XLOG_HEADER_CYCLE_SIZE)
			xheads++;
1719

1720
		for (i = 1; i < xheads; i++) {
1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731
			crc = crc32c(crc, &xhdr[i].hic_xheader,
				     sizeof(struct xlog_rec_ext_header));
		}
	}

	/* ... and finally for the payload */
	crc = crc32c(crc, dp, size);

	return xfs_end_cksum(crc);
}

1732 1733 1734 1735 1736 1737 1738
/*
 * The bdstrat callback function for log bufs. This gives us a central
 * place to trap bufs in case we get hit by a log I/O error and need to
 * shutdown. Actually, in practice, even when we didn't get a log error,
 * we transition the iclogs to IOERROR state *after* flushing all existing
 * iclogs to disk. This is because we don't want anymore new transactions to be
 * started or completed afterwards.
1739 1740 1741 1742 1743 1744
 *
 * We lock the iclogbufs here so that we can serialise against IO completion
 * during unmount. We might be processing a shutdown triggered during unmount,
 * and that can occur asynchronously to the unmount thread, and hence we need to
 * ensure that completes before tearing down the iclogbufs. Hence we need to
 * hold the buffer lock across the log IO to acheive that.
1745 1746 1747 1748 1749
 */
STATIC int
xlog_bdstrat(
	struct xfs_buf		*bp)
{
1750
	struct xlog_in_core	*iclog = bp->b_fspriv;
1751

1752
	xfs_buf_lock(bp);
1753
	if (iclog->ic_state & XLOG_STATE_IOERROR) {
1754
		xfs_buf_ioerror(bp, -EIO);
1755
		xfs_buf_stale(bp);
1756
		xfs_buf_ioend(bp);
1757 1758 1759
		/*
		 * It would seem logical to return EIO here, but we rely on
		 * the log state machine to propagate I/O errors instead of
1760 1761
		 * doing it here. Similarly, IO completion will unlock the
		 * buffer, so we don't do it here.
1762 1763 1764 1765
		 */
		return 0;
	}

1766
	xfs_buf_submit(bp);
1767 1768
	return 0;
}
Linus Torvalds's avatar
Linus Torvalds committed
1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794

/*
 * Flush out the in-core log (iclog) to the on-disk log in an asynchronous 
 * fashion.  Previously, we should have moved the current iclog
 * ptr in the log to point to the next available iclog.  This allows further
 * write to continue while this code syncs out an iclog ready to go.
 * Before an in-core log can be written out, the data section must be scanned
 * to save away the 1st word of each BBSIZE block into the header.  We replace
 * it with the current cycle count.  Each BBSIZE block is tagged with the
 * cycle count because there in an implicit assumption that drives will
 * guarantee that entire 512 byte blocks get written at once.  In other words,
 * we can't have part of a 512 byte block written and part not written.  By
 * tagging each block, we will know which blocks are valid when recovering
 * after an unclean shutdown.
 *
 * This routine is single threaded on the iclog.  No other thread can be in
 * this routine with the same iclog.  Changing contents of iclog can there-
 * fore be done without grabbing the state machine lock.  Updating the global
 * log will require grabbing the lock though.
 *
 * The entire log manager uses a logical block numbering scheme.  Only
 * log_sync (and then only bwrite()) know about the fact that the log may
 * not start with block zero on a given device.  The log block start offset
 * is added immediately before calling bwrite().
 */

1795
STATIC int
Mark Tinguely's avatar
Mark Tinguely committed
1796 1797 1798
xlog_sync(
	struct xlog		*log,
	struct xlog_in_core	*iclog)
Linus Torvalds's avatar
Linus Torvalds committed
1799 1800
{
	xfs_buf_t	*bp;
1801
	int		i;
Linus Torvalds's avatar
Linus Torvalds committed
1802 1803 1804 1805 1806
	uint		count;		/* byte count of bwrite */
	uint		count_init;	/* initial count before roundup */
	int		roundoff;       /* roundoff to BB or stripe */
	int		split = 0;	/* split write into two regions */
	int		error;
1807
	int		v2 = xfs_sb_version_haslogv2(&log->l_mp->m_sb);
1808
	int		size;
Linus Torvalds's avatar
Linus Torvalds committed
1809

1810
	XFS_STATS_INC(log->l_mp, xs_log_writes);
1811
	ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds's avatar
Linus Torvalds committed
1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831

	/* Add for LR header */
	count_init = log->l_iclog_hsize + iclog->ic_offset;

	/* Round out the log write size */
	if (v2 && log->l_mp->m_sb.sb_logsunit > 1) {
		/* we have a v2 stripe unit to use */
		count = XLOG_LSUNITTOB(log, XLOG_BTOLSUNIT(log, count_init));
	} else {
		count = BBTOB(BTOBB(count_init));
	}
	roundoff = count - count_init;
	ASSERT(roundoff >= 0);
	ASSERT((v2 && log->l_mp->m_sb.sb_logsunit > 1 && 
                roundoff < log->l_mp->m_sb.sb_logsunit)
		|| 
		(log->l_mp->m_sb.sb_logsunit <= 1 && 
		 roundoff < BBTOB(1)));

	/* move grant heads by roundoff in sync */
1832 1833
	xlog_grant_add_space(log, &log->l_reserve_head.grant, roundoff);
	xlog_grant_add_space(log, &log->l_write_head.grant, roundoff);
Linus Torvalds's avatar
Linus Torvalds committed
1834 1835 1836 1837 1838

	/* put cycle number in every block */
	xlog_pack_data(log, iclog, roundoff); 

	/* real byte length */
1839 1840 1841 1842
	size = iclog->ic_offset;
	if (v2)
		size += roundoff;
	iclog->ic_header.h_len = cpu_to_be32(size);
Linus Torvalds's avatar
Linus Torvalds committed
1843

1844
	bp = iclog->ic_bp;
1845
	XFS_BUF_SET_ADDR(bp, BLOCK_LSN(be64_to_cpu(iclog->ic_header.h_lsn)));
Linus Torvalds's avatar
Linus Torvalds committed
1846

1847
	XFS_STATS_ADD(log->l_mp, xs_log_blocks, BTOBB(count));
Linus Torvalds's avatar
Linus Torvalds committed
1848 1849 1850

	/* Do we need to split this write into 2 parts? */
	if (XFS_BUF_ADDR(bp) + BTOBB(count) > log->l_logBBsize) {
1851 1852
		char		*dptr;

Linus Torvalds's avatar
Linus Torvalds committed
1853 1854
		split = count - (BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp)));
		count = BBTOB(log->l_logBBsize - XFS_BUF_ADDR(bp));
1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872
		iclog->ic_bwritecnt = 2;

		/*
		 * Bump the cycle numbers at the start of each block in the
		 * part of the iclog that ends up in the buffer that gets
		 * written to the start of the log.
		 *
		 * Watch out for the header magic number case, though.
		 */
		dptr = (char *)&iclog->ic_header + count;
		for (i = 0; i < split; i += BBSIZE) {
			__uint32_t cycle = be32_to_cpu(*(__be32 *)dptr);
			if (++cycle == XLOG_HEADER_MAGIC_NUM)
				cycle++;
			*(__be32 *)dptr = cpu_to_be32(cycle);

			dptr += BBSIZE;
		}
Linus Torvalds's avatar
Linus Torvalds committed
1873 1874 1875
	} else {
		iclog->ic_bwritecnt = 1;
	}
1876 1877 1878 1879

	/* calculcate the checksum */
	iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header,
					    iclog->ic_datap, size);
1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896
#ifdef DEBUG
	/*
	 * Intentionally corrupt the log record CRC based on the error injection
	 * frequency, if defined. This facilitates testing log recovery in the
	 * event of torn writes. Hence, set the IOABORT state to abort the log
	 * write on I/O completion and shutdown the fs. The subsequent mount
	 * detects the bad CRC and attempts to recover.
	 */
	if (log->l_badcrc_factor &&
	    (prandom_u32() % log->l_badcrc_factor == 0)) {
		iclog->ic_header.h_crc &= 0xAAAAAAAA;
		iclog->ic_state |= XLOG_STATE_IOABORT;
		xfs_warn(log->l_mp,
	"Intentionally corrupted log record at LSN 0x%llx. Shutdown imminent.",
			 be64_to_cpu(iclog->ic_header.h_lsn));
	}
#endif
1897

1898
	bp->b_io_length = BTOBB(count);
1899
	bp->b_fspriv = iclog;
1900
	bp->b_flags &= ~(XBF_FUA | XBF_FLUSH);
1901
	bp->b_flags |= (XBF_ASYNC | XBF_SYNCIO | XBF_WRITE);
1902

1903
	if (log->l_mp->m_flags & XFS_MOUNT_BARRIER) {
1904 1905
		bp->b_flags |= XBF_FUA;

1906
		/*
1907 1908 1909 1910 1911 1912 1913
		 * Flush the data device before flushing the log to make
		 * sure all meta data written back from the AIL actually made
		 * it to disk before stamping the new log tail LSN into the
		 * log buffer.  For an external log we need to issue the
		 * flush explicitly, and unfortunately synchronously here;
		 * for an internal log we can simply use the block layer
		 * state machine for preflushes.
1914 1915 1916
		 */
		if (log->l_mp->m_logdev_targp != log->l_mp->m_ddev_targp)
			xfs_blkdev_issue_flush(log->l_mp->m_ddev_targp);
1917 1918
		else
			bp->b_flags |= XBF_FLUSH;
1919
	}
Linus Torvalds's avatar
Linus Torvalds committed
1920 1921 1922 1923

	ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
	ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);

1924
	xlog_verify_iclog(log, iclog, count, true);
Linus Torvalds's avatar
Linus Torvalds committed
1925 1926 1927

	/* account for log which doesn't start at block #0 */
	XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1928

Linus Torvalds's avatar
Linus Torvalds committed
1929 1930 1931 1932
	/*
	 * Don't call xfs_bwrite here. We do log-syncs even when the filesystem
	 * is shutting down.
	 */
1933 1934 1935
	error = xlog_bdstrat(bp);
	if (error) {
		xfs_buf_ioerror_alert(bp, "xlog_sync");
1936
		return error;
Linus Torvalds's avatar
Linus Torvalds committed
1937 1938
	}
	if (split) {
1939
		bp = iclog->ic_log->l_xbuf;
Linus Torvalds's avatar
Linus Torvalds committed
1940
		XFS_BUF_SET_ADDR(bp, 0);	     /* logical 0 */
1941 1942
		xfs_buf_associate_memory(bp,
				(char *)&iclog->ic_header + count, split);
1943
		bp->b_fspriv = iclog;
1944
		bp->b_flags &= ~(XBF_FUA | XBF_FLUSH);
1945
		bp->b_flags |= (XBF_ASYNC | XBF_SYNCIO | XBF_WRITE);
1946
		if (log->l_mp->m_flags & XFS_MOUNT_BARRIER)
1947
			bp->b_flags |= XBF_FUA;
Linus Torvalds's avatar
Linus Torvalds committed
1948 1949 1950 1951

		ASSERT(XFS_BUF_ADDR(bp) <= log->l_logBBsize-1);
		ASSERT(XFS_BUF_ADDR(bp) + BTOBB(count) <= log->l_logBBsize);

1952
		/* account for internal log which doesn't start at block #0 */
Linus Torvalds's avatar
Linus Torvalds committed
1953
		XFS_BUF_SET_ADDR(bp, XFS_BUF_ADDR(bp) + log->l_logBBstart);
1954 1955 1956
		error = xlog_bdstrat(bp);
		if (error) {
			xfs_buf_ioerror_alert(bp, "xlog_sync (split)");
1957
			return error;
Linus Torvalds's avatar
Linus Torvalds committed
1958 1959
		}
	}
1960
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
1961 1962 1963
}	/* xlog_sync */

/*
1964
 * Deallocate a log structure
Linus Torvalds's avatar
Linus Torvalds committed
1965
 */
1966
STATIC void
Mark Tinguely's avatar
Mark Tinguely committed
1967 1968
xlog_dealloc_log(
	struct xlog	*log)
Linus Torvalds's avatar
Linus Torvalds committed
1969 1970 1971 1972
{
	xlog_in_core_t	*iclog, *next_iclog;
	int		i;

1973 1974
	xlog_cil_destroy(log);

1975
	/*
1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989
	 * Cycle all the iclogbuf locks to make sure all log IO completion
	 * is done before we tear down these buffers.
	 */
	iclog = log->l_iclog;
	for (i = 0; i < log->l_iclog_bufs; i++) {
		xfs_buf_lock(iclog->ic_bp);
		xfs_buf_unlock(iclog->ic_bp);
		iclog = iclog->ic_next;
	}

	/*
	 * Always need to ensure that the extra buffer does not point to memory
	 * owned by another log buffer before we free it. Also, cycle the lock
	 * first to ensure we've completed IO on it.
1990
	 */
1991 1992
	xfs_buf_lock(log->l_xbuf);
	xfs_buf_unlock(log->l_xbuf);
1993
	xfs_buf_set_empty(log->l_xbuf, BTOBB(log->l_iclog_size));
1994 1995
	xfs_buf_free(log->l_xbuf);

Linus Torvalds's avatar
Linus Torvalds committed
1996
	iclog = log->l_iclog;
1997
	for (i = 0; i < log->l_iclog_bufs; i++) {
Linus Torvalds's avatar
Linus Torvalds committed
1998 1999
		xfs_buf_free(iclog->ic_bp);
		next_iclog = iclog->ic_next;
2000
		kmem_free(iclog);
Linus Torvalds's avatar
Linus Torvalds committed
2001 2002 2003 2004 2005
		iclog = next_iclog;
	}
	spinlock_destroy(&log->l_icloglock);

	log->l_mp->m_log = NULL;
2006
	kmem_free(log);
2007
}	/* xlog_dealloc_log */
Linus Torvalds's avatar
Linus Torvalds committed
2008 2009 2010 2011 2012 2013

/*
 * Update counters atomically now that memcpy is done.
 */
/* ARGSUSED */
static inline void
Mark Tinguely's avatar
Mark Tinguely committed
2014 2015 2016 2017 2018
xlog_state_finish_copy(
	struct xlog		*log,
	struct xlog_in_core	*iclog,
	int			record_cnt,
	int			copy_bytes)
Linus Torvalds's avatar
Linus Torvalds committed
2019
{
Eric Sandeen's avatar
Eric Sandeen committed
2020
	spin_lock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
2021

2022
	be32_add_cpu(&iclog->ic_header.h_num_logops, record_cnt);
Linus Torvalds's avatar
Linus Torvalds committed
2023 2024
	iclog->ic_offset += copy_bytes;

Eric Sandeen's avatar
Eric Sandeen committed
2025
	spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
2026 2027 2028 2029 2030
}	/* xlog_state_finish_copy */




2031 2032 2033 2034
/*
 * print out info relating to regions written which consume
 * the reservation
 */
2035 2036 2037 2038
void
xlog_print_tic_res(
	struct xfs_mount	*mp,
	struct xlog_ticket	*ticket)
2039 2040 2041 2042 2043
{
	uint i;
	uint ophdr_spc = ticket->t_res_num_ophdrs * (uint)sizeof(xlog_op_header_t);

	/* match with XLOG_REG_TYPE_* in xfs_log.h */
2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065
#define REG_TYPE_STR(type, str)	[XLOG_REG_TYPE_##type] = str
	static char *res_type_str[XLOG_REG_TYPE_MAX + 1] = {
	    REG_TYPE_STR(BFORMAT, "bformat"),
	    REG_TYPE_STR(BCHUNK, "bchunk"),
	    REG_TYPE_STR(EFI_FORMAT, "efi_format"),
	    REG_TYPE_STR(EFD_FORMAT, "efd_format"),
	    REG_TYPE_STR(IFORMAT, "iformat"),
	    REG_TYPE_STR(ICORE, "icore"),
	    REG_TYPE_STR(IEXT, "iext"),
	    REG_TYPE_STR(IBROOT, "ibroot"),
	    REG_TYPE_STR(ILOCAL, "ilocal"),
	    REG_TYPE_STR(IATTR_EXT, "iattr_ext"),
	    REG_TYPE_STR(IATTR_BROOT, "iattr_broot"),
	    REG_TYPE_STR(IATTR_LOCAL, "iattr_local"),
	    REG_TYPE_STR(QFORMAT, "qformat"),
	    REG_TYPE_STR(DQUOT, "dquot"),
	    REG_TYPE_STR(QUOTAOFF, "quotaoff"),
	    REG_TYPE_STR(LRHEADER, "LR header"),
	    REG_TYPE_STR(UNMOUNT, "unmount"),
	    REG_TYPE_STR(COMMIT, "commit"),
	    REG_TYPE_STR(TRANSHDR, "trans header"),
	    REG_TYPE_STR(ICREATE, "inode create")
2066
	};
2067
#undef REG_TYPE_STR
2068

2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081
	xfs_warn(mp, "xlog_write: reservation summary:");
	xfs_warn(mp, "  unit res    = %d bytes",
		 ticket->t_unit_res);
	xfs_warn(mp, "  current res = %d bytes",
		 ticket->t_curr_res);
	xfs_warn(mp, "  total reg   = %u bytes (o/flow = %u bytes)",
		 ticket->t_res_arr_sum, ticket->t_res_o_flow);
	xfs_warn(mp, "  ophdrs      = %u (ophdr space = %u bytes)",
		 ticket->t_res_num_ophdrs, ophdr_spc);
	xfs_warn(mp, "  ophdr + reg = %u bytes",
		 ticket->t_res_arr_sum + ticket->t_res_o_flow + ophdr_spc);
	xfs_warn(mp, "  num regions = %u",
		 ticket->t_res_num);
2082 2083

	for (i = 0; i < ticket->t_res_num; i++) {
2084
		uint r_type = ticket->t_res_arr[i].r_type;
2085
		xfs_warn(mp, "region[%u]: %s - %u bytes", i,
2086
			    ((r_type <= 0 || r_type > XLOG_REG_TYPE_MAX) ?
2087
			    "bad-rtype" : res_type_str[r_type]),
2088 2089
			    ticket->t_res_arr[i].r_len);
	}
2090

2091
	xfs_alert_tag(mp, XFS_PTAG_LOGRES,
2092
		"xlog_write: reservation ran out. Need to up reservation");
2093
	xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
2094 2095
}

Dave Chinner's avatar
Dave Chinner committed
2096 2097 2098 2099 2100 2101 2102
/*
 * Calculate the potential space needed by the log vector.  Each region gets
 * its own xlog_op_header_t and may need to be double word aligned.
 */
static int
xlog_write_calc_vec_length(
	struct xlog_ticket	*ticket,
2103
	struct xfs_log_vec	*log_vector)
Dave Chinner's avatar
Dave Chinner committed
2104
{
2105
	struct xfs_log_vec	*lv;
Dave Chinner's avatar
Dave Chinner committed
2106 2107 2108 2109 2110 2111 2112 2113
	int			headers = 0;
	int			len = 0;
	int			i;

	/* acct for start rec of xact */
	if (ticket->t_flags & XLOG_TIC_INITED)
		headers++;

2114
	for (lv = log_vector; lv; lv = lv->lv_next) {
2115 2116 2117 2118
		/* we don't write ordered log vectors */
		if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED)
			continue;

2119 2120 2121 2122
		headers += lv->lv_niovecs;

		for (i = 0; i < lv->lv_niovecs; i++) {
			struct xfs_log_iovec	*vecp = &lv->lv_iovecp[i];
Dave Chinner's avatar
Dave Chinner committed
2123

2124 2125 2126
			len += vecp->i_len;
			xlog_tic_add_region(ticket, vecp->i_len, vecp->i_type);
		}
Dave Chinner's avatar
Dave Chinner committed
2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140
	}

	ticket->t_res_num_ophdrs += headers;
	len += headers * sizeof(struct xlog_op_header);

	return len;
}

/*
 * If first write for transaction, insert start record  We can't be trying to
 * commit if we are inited.  We can't have any "partial_copy" if we are inited.
 */
static int
xlog_write_start_rec(
2141
	struct xlog_op_header	*ophdr,
Dave Chinner's avatar
Dave Chinner committed
2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155 2156 2157 2158 2159
	struct xlog_ticket	*ticket)
{
	if (!(ticket->t_flags & XLOG_TIC_INITED))
		return 0;

	ophdr->oh_tid	= cpu_to_be32(ticket->t_tid);
	ophdr->oh_clientid = ticket->t_clientid;
	ophdr->oh_len = 0;
	ophdr->oh_flags = XLOG_START_TRANS;
	ophdr->oh_res2 = 0;

	ticket->t_flags &= ~XLOG_TIC_INITED;

	return sizeof(struct xlog_op_header);
}

static xlog_op_header_t *
xlog_write_setup_ophdr(
2160
	struct xlog		*log,
2161
	struct xlog_op_header	*ophdr,
Dave Chinner's avatar
Dave Chinner committed
2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182
	struct xlog_ticket	*ticket,
	uint			flags)
{
	ophdr->oh_tid = cpu_to_be32(ticket->t_tid);
	ophdr->oh_clientid = ticket->t_clientid;
	ophdr->oh_res2 = 0;

	/* are we copying a commit or unmount record? */
	ophdr->oh_flags = flags;

	/*
	 * We've seen logs corrupted with bad transaction client ids.  This
	 * makes sure that XFS doesn't generate them on.  Turn this into an EIO
	 * and shut down the filesystem.
	 */
	switch (ophdr->oh_clientid)  {
	case XFS_TRANSACTION:
	case XFS_VOLUME:
	case XFS_LOG:
		break;
	default:
2183
		xfs_warn(log->l_mp,
Dave Chinner's avatar
Dave Chinner committed
2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195
			"Bad XFS transaction clientid 0x%x in ticket 0x%p",
			ophdr->oh_clientid, ticket);
		return NULL;
	}

	return ophdr;
}

/*
 * Set up the parameters of the region copy into the log. This has
 * to handle region write split across multiple log buffers - this
 * state is kept external to this function so that this code can
2196
 * be written in an obvious, self documenting manner.
Dave Chinner's avatar
Dave Chinner committed
2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242
 */
static int
xlog_write_setup_copy(
	struct xlog_ticket	*ticket,
	struct xlog_op_header	*ophdr,
	int			space_available,
	int			space_required,
	int			*copy_off,
	int			*copy_len,
	int			*last_was_partial_copy,
	int			*bytes_consumed)
{
	int			still_to_copy;

	still_to_copy = space_required - *bytes_consumed;
	*copy_off = *bytes_consumed;

	if (still_to_copy <= space_available) {
		/* write of region completes here */
		*copy_len = still_to_copy;
		ophdr->oh_len = cpu_to_be32(*copy_len);
		if (*last_was_partial_copy)
			ophdr->oh_flags |= (XLOG_END_TRANS|XLOG_WAS_CONT_TRANS);
		*last_was_partial_copy = 0;
		*bytes_consumed = 0;
		return 0;
	}

	/* partial write of region, needs extra log op header reservation */
	*copy_len = space_available;
	ophdr->oh_len = cpu_to_be32(*copy_len);
	ophdr->oh_flags |= XLOG_CONTINUE_TRANS;
	if (*last_was_partial_copy)
		ophdr->oh_flags |= XLOG_WAS_CONT_TRANS;
	*bytes_consumed += *copy_len;
	(*last_was_partial_copy)++;

	/* account for new log op header */
	ticket->t_curr_res -= sizeof(struct xlog_op_header);
	ticket->t_res_num_ophdrs++;

	return sizeof(struct xlog_op_header);
}

static int
xlog_write_copy_finish(
2243
	struct xlog		*log,
Dave Chinner's avatar
Dave Chinner committed
2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285
	struct xlog_in_core	*iclog,
	uint			flags,
	int			*record_cnt,
	int			*data_cnt,
	int			*partial_copy,
	int			*partial_copy_len,
	int			log_offset,
	struct xlog_in_core	**commit_iclog)
{
	if (*partial_copy) {
		/*
		 * This iclog has already been marked WANT_SYNC by
		 * xlog_state_get_iclog_space.
		 */
		xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
		*record_cnt = 0;
		*data_cnt = 0;
		return xlog_state_release_iclog(log, iclog);
	}

	*partial_copy = 0;
	*partial_copy_len = 0;

	if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
		/* no more space in this iclog - push it. */
		xlog_state_finish_copy(log, iclog, *record_cnt, *data_cnt);
		*record_cnt = 0;
		*data_cnt = 0;

		spin_lock(&log->l_icloglock);
		xlog_state_want_sync(log, iclog);
		spin_unlock(&log->l_icloglock);

		if (!commit_iclog)
			return xlog_state_release_iclog(log, iclog);
		ASSERT(flags & XLOG_COMMIT_TRANS);
		*commit_iclog = iclog;
	}

	return 0;
}

Linus Torvalds's avatar
Linus Torvalds committed
2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325
/*
 * Write some region out to in-core log
 *
 * This will be called when writing externally provided regions or when
 * writing out a commit record for a given transaction.
 *
 * General algorithm:
 *	1. Find total length of this write.  This may include adding to the
 *		lengths passed in.
 *	2. Check whether we violate the tickets reservation.
 *	3. While writing to this iclog
 *	    A. Reserve as much space in this iclog as can get
 *	    B. If this is first write, save away start lsn
 *	    C. While writing this region:
 *		1. If first write of transaction, write start record
 *		2. Write log operation header (header per region)
 *		3. Find out if we can fit entire region into this iclog
 *		4. Potentially, verify destination memcpy ptr
 *		5. Memcpy (partial) region
 *		6. If partial copy, release iclog; otherwise, continue
 *			copying more regions into current iclog
 *	4. Mark want sync bit (in simulation mode)
 *	5. Release iclog for potential flush to on-disk log.
 *
 * ERRORS:
 * 1.	Panic if reservation is overrun.  This should never happen since
 *	reservation amounts are generated internal to the filesystem.
 * NOTES:
 * 1. Tickets are single threaded data structures.
 * 2. The XLOG_END_TRANS & XLOG_CONTINUE_TRANS flags are passed down to the
 *	syncing routine.  When a single log_write region needs to span
 *	multiple in-core logs, the XLOG_CONTINUE_TRANS bit should be set
 *	on all log operation writes which don't contain the end of the
 *	region.  The XLOG_END_TRANS bit is used for the in-core log
 *	operation which contains the end of the continued log_write region.
 * 3. When xlog_state_get_iclog_space() grabs the rest of the current iclog,
 *	we don't really know exactly how much space will be used.  As a result,
 *	we don't update ic_offset until the end when we know exactly how many
 *	bytes have been written out.
 */
2326
int
2327
xlog_write(
2328
	struct xlog		*log,
2329
	struct xfs_log_vec	*log_vector,
2330 2331 2332 2333
	struct xlog_ticket	*ticket,
	xfs_lsn_t		*start_lsn,
	struct xlog_in_core	**commit_iclog,
	uint			flags)
Linus Torvalds's avatar
Linus Torvalds committed
2334
{
Christoph Hellwig's avatar
Christoph Hellwig committed
2335
	struct xlog_in_core	*iclog = NULL;
2336 2337
	struct xfs_log_iovec	*vecp;
	struct xfs_log_vec	*lv;
Christoph Hellwig's avatar
Christoph Hellwig committed
2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348
	int			len;
	int			index;
	int			partial_copy = 0;
	int			partial_copy_len = 0;
	int			contwr = 0;
	int			record_cnt = 0;
	int			data_cnt = 0;
	int			error;

	*start_lsn = 0;

2349
	len = xlog_write_calc_vec_length(ticket, log_vector);
2350

2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364
	/*
	 * Region headers and bytes are already accounted for.
	 * We only need to take into account start records and
	 * split regions in this function.
	 */
	if (ticket->t_flags & XLOG_TIC_INITED)
		ticket->t_curr_res -= sizeof(xlog_op_header_t);

	/*
	 * Commit record headers need to be accounted for. These
	 * come in as separate writes so are easy to detect.
	 */
	if (flags & (XLOG_COMMIT_TRANS | XLOG_UNMOUNT_TRANS))
		ticket->t_curr_res -= sizeof(xlog_op_header_t);
2365 2366

	if (ticket->t_curr_res < 0)
2367
		xlog_print_tic_res(log->l_mp, ticket);
Linus Torvalds's avatar
Linus Torvalds committed
2368

2369 2370 2371
	index = 0;
	lv = log_vector;
	vecp = lv->lv_iovecp;
2372
	while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) {
2373
		void		*ptr;
Christoph Hellwig's avatar
Christoph Hellwig committed
2374
		int		log_offset;
Linus Torvalds's avatar
Linus Torvalds committed
2375

Christoph Hellwig's avatar
Christoph Hellwig committed
2376 2377 2378 2379
		error = xlog_state_get_iclog_space(log, len, &iclog, ticket,
						   &contwr, &log_offset);
		if (error)
			return error;
Linus Torvalds's avatar
Linus Torvalds committed
2380

Christoph Hellwig's avatar
Christoph Hellwig committed
2381
		ASSERT(log_offset <= iclog->ic_size - 1);
2382
		ptr = iclog->ic_datap + log_offset;
Linus Torvalds's avatar
Linus Torvalds committed
2383

Christoph Hellwig's avatar
Christoph Hellwig committed
2384 2385 2386
		/* start_lsn is the first lsn written to. That's all we need. */
		if (!*start_lsn)
			*start_lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Dave Chinner's avatar
Dave Chinner committed
2387

Christoph Hellwig's avatar
Christoph Hellwig committed
2388 2389 2390 2391
		/*
		 * This loop writes out as many regions as can fit in the amount
		 * of space which was allocated by xlog_state_get_iclog_space().
		 */
2392 2393
		while (lv && (!lv->lv_niovecs || index < lv->lv_niovecs)) {
			struct xfs_log_iovec	*reg;
Christoph Hellwig's avatar
Christoph Hellwig committed
2394 2395 2396 2397
			struct xlog_op_header	*ophdr;
			int			start_rec_copy;
			int			copy_len;
			int			copy_off;
2398 2399 2400 2401 2402 2403 2404 2405
			bool			ordered = false;

			/* ordered log vectors have no regions to write */
			if (lv->lv_buf_len == XFS_LOG_VEC_ORDERED) {
				ASSERT(lv->lv_niovecs == 0);
				ordered = true;
				goto next_lv;
			}
Christoph Hellwig's avatar
Christoph Hellwig committed
2406

2407
			reg = &vecp[index];
2408
			ASSERT(reg->i_len % sizeof(__int32_t) == 0);
2409
			ASSERT((unsigned long)ptr % sizeof(__int32_t) == 0);
Christoph Hellwig's avatar
Christoph Hellwig committed
2410 2411 2412 2413

			start_rec_copy = xlog_write_start_rec(ptr, ticket);
			if (start_rec_copy) {
				record_cnt++;
2414
				xlog_write_adv_cnt(&ptr, &len, &log_offset,
Christoph Hellwig's avatar
Christoph Hellwig committed
2415 2416
						   start_rec_copy);
			}
Dave Chinner's avatar
Dave Chinner committed
2417

Christoph Hellwig's avatar
Christoph Hellwig committed
2418 2419
			ophdr = xlog_write_setup_ophdr(log, ptr, ticket, flags);
			if (!ophdr)
2420
				return -EIO;
Christoph Hellwig's avatar
Christoph Hellwig committed
2421

2422
			xlog_write_adv_cnt(&ptr, &len, &log_offset,
Christoph Hellwig's avatar
Christoph Hellwig committed
2423 2424 2425 2426
					   sizeof(struct xlog_op_header));

			len += xlog_write_setup_copy(ticket, ophdr,
						     iclog->ic_size-log_offset,
2427
						     reg->i_len,
Christoph Hellwig's avatar
Christoph Hellwig committed
2428 2429 2430 2431 2432
						     &copy_off, &copy_len,
						     &partial_copy,
						     &partial_copy_len);
			xlog_verify_dest_ptr(log, ptr);

2433 2434 2435 2436 2437 2438 2439 2440
			/*
			 * Copy region.
			 *
			 * Unmount records just log an opheader, so can have
			 * empty payloads with no data region to copy. Hence we
			 * only copy the payload if the vector says it has data
			 * to copy.
			 */
Christoph Hellwig's avatar
Christoph Hellwig committed
2441
			ASSERT(copy_len >= 0);
2442 2443 2444 2445 2446
			if (copy_len > 0) {
				memcpy(ptr, reg->i_addr + copy_off, copy_len);
				xlog_write_adv_cnt(&ptr, &len, &log_offset,
						   copy_len);
			}
Christoph Hellwig's avatar
Christoph Hellwig committed
2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474
			copy_len += start_rec_copy + sizeof(xlog_op_header_t);
			record_cnt++;
			data_cnt += contwr ? copy_len : 0;

			error = xlog_write_copy_finish(log, iclog, flags,
						       &record_cnt, &data_cnt,
						       &partial_copy,
						       &partial_copy_len,
						       log_offset,
						       commit_iclog);
			if (error)
				return error;

			/*
			 * if we had a partial copy, we need to get more iclog
			 * space but we don't want to increment the region
			 * index because there is still more is this region to
			 * write.
			 *
			 * If we completed writing this region, and we flushed
			 * the iclog (indicated by resetting of the record
			 * count), then we also need to get more log space. If
			 * this was the last record, though, we are done and
			 * can just return.
			 */
			if (partial_copy)
				break;

2475
			if (++index == lv->lv_niovecs) {
2476
next_lv:
2477 2478 2479 2480 2481
				lv = lv->lv_next;
				index = 0;
				if (lv)
					vecp = lv->lv_iovecp;
			}
2482
			if (record_cnt == 0 && ordered == false) {
2483
				if (!lv)
Christoph Hellwig's avatar
Christoph Hellwig committed
2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494
					return 0;
				break;
			}
		}
	}

	ASSERT(len == 0);

	xlog_state_finish_copy(log, iclog, record_cnt, data_cnt);
	if (!commit_iclog)
		return xlog_state_release_iclog(log, iclog);
Linus Torvalds's avatar
Linus Torvalds committed
2495 2496 2497 2498

	ASSERT(flags & XLOG_COMMIT_TRANS);
	*commit_iclog = iclog;
	return 0;
Christoph Hellwig's avatar
Christoph Hellwig committed
2499
}
Linus Torvalds's avatar
Linus Torvalds committed
2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511


/*****************************************************************************
 *
 *		State Machine functions
 *
 *****************************************************************************
 */

/* Clean iclogs starting from the head.  This ordering must be
 * maintained, so an iclog doesn't become ACTIVE beyond one that
 * is SYNCING.  This is also required to maintain the notion that we use
2512
 * a ordered wait queue to hold off would be writers to the log when every
Linus Torvalds's avatar
Linus Torvalds committed
2513 2514 2515 2516
 * iclog is trying to sync to disk.
 *
 * State Change: DIRTY -> ACTIVE
 */
2517
STATIC void
Mark Tinguely's avatar
Mark Tinguely committed
2518 2519
xlog_state_clean_log(
	struct xlog *log)
Linus Torvalds's avatar
Linus Torvalds committed
2520 2521 2522 2523 2524 2525 2526 2527 2528
{
	xlog_in_core_t	*iclog;
	int changed = 0;

	iclog = log->l_iclog;
	do {
		if (iclog->ic_state == XLOG_STATE_DIRTY) {
			iclog->ic_state	= XLOG_STATE_ACTIVE;
			iclog->ic_offset       = 0;
2529
			ASSERT(iclog->ic_callback == NULL);
Linus Torvalds's avatar
Linus Torvalds committed
2530 2531 2532 2533 2534 2535 2536 2537 2538
			/*
			 * If the number of ops in this iclog indicate it just
			 * contains the dummy transaction, we can
			 * change state into IDLE (the second time around).
			 * Otherwise we should change the state into
			 * NEED a dummy.
			 * We don't need to cover the dummy.
			 */
			if (!changed &&
2539 2540
			   (be32_to_cpu(iclog->ic_header.h_num_logops) ==
			   		XLOG_COVER_OPS)) {
Linus Torvalds's avatar
Linus Torvalds committed
2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598
				changed = 1;
			} else {
				/*
				 * We have two dirty iclogs so start over
				 * This could also be num of ops indicates
				 * this is not the dummy going out.
				 */
				changed = 2;
			}
			iclog->ic_header.h_num_logops = 0;
			memset(iclog->ic_header.h_cycle_data, 0,
			      sizeof(iclog->ic_header.h_cycle_data));
			iclog->ic_header.h_lsn = 0;
		} else if (iclog->ic_state == XLOG_STATE_ACTIVE)
			/* do nothing */;
		else
			break;	/* stop cleaning */
		iclog = iclog->ic_next;
	} while (iclog != log->l_iclog);

	/* log is locked when we are called */
	/*
	 * Change state for the dummy log recording.
	 * We usually go to NEED. But we go to NEED2 if the changed indicates
	 * we are done writing the dummy record.
	 * If we are done with the second dummy recored (DONE2), then
	 * we go to IDLE.
	 */
	if (changed) {
		switch (log->l_covered_state) {
		case XLOG_STATE_COVER_IDLE:
		case XLOG_STATE_COVER_NEED:
		case XLOG_STATE_COVER_NEED2:
			log->l_covered_state = XLOG_STATE_COVER_NEED;
			break;

		case XLOG_STATE_COVER_DONE:
			if (changed == 1)
				log->l_covered_state = XLOG_STATE_COVER_NEED2;
			else
				log->l_covered_state = XLOG_STATE_COVER_NEED;
			break;

		case XLOG_STATE_COVER_DONE2:
			if (changed == 1)
				log->l_covered_state = XLOG_STATE_COVER_IDLE;
			else
				log->l_covered_state = XLOG_STATE_COVER_NEED;
			break;

		default:
			ASSERT(0);
		}
	}
}	/* xlog_state_clean_log */

STATIC xfs_lsn_t
xlog_get_lowest_lsn(
Mark Tinguely's avatar
Mark Tinguely committed
2599
	struct xlog	*log)
Linus Torvalds's avatar
Linus Torvalds committed
2600 2601 2602 2603 2604 2605 2606 2607
{
	xlog_in_core_t  *lsn_log;
	xfs_lsn_t	lowest_lsn, lsn;

	lsn_log = log->l_iclog;
	lowest_lsn = 0;
	do {
	    if (!(lsn_log->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY))) {
2608
		lsn = be64_to_cpu(lsn_log->ic_header.h_lsn);
Linus Torvalds's avatar
Linus Torvalds committed
2609 2610 2611 2612 2613 2614 2615
		if ((lsn && !lowest_lsn) ||
		    (XFS_LSN_CMP(lsn, lowest_lsn) < 0)) {
			lowest_lsn = lsn;
		}
	    }
	    lsn_log = lsn_log->ic_next;
	} while (lsn_log != log->l_iclog);
2616
	return lowest_lsn;
Linus Torvalds's avatar
Linus Torvalds committed
2617 2618 2619 2620 2621
}


STATIC void
xlog_state_do_callback(
Mark Tinguely's avatar
Mark Tinguely committed
2622 2623 2624
	struct xlog		*log,
	int			aborted,
	struct xlog_in_core	*ciclog)
Linus Torvalds's avatar
Linus Torvalds committed
2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637
{
	xlog_in_core_t	   *iclog;
	xlog_in_core_t	   *first_iclog;	/* used to know when we've
						 * processed all iclogs once */
	xfs_log_callback_t *cb, *cb_next;
	int		   flushcnt = 0;
	xfs_lsn_t	   lowest_lsn;
	int		   ioerrors;	/* counter: iclogs with errors */
	int		   loopdidcallbacks; /* flag: inner loop did callbacks*/
	int		   funcdidcallbacks; /* flag: function did callbacks */
	int		   repeats;	/* for issuing console warnings if
					 * looping too many times */

Eric Sandeen's avatar
Eric Sandeen committed
2638
	spin_lock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682
	first_iclog = iclog = log->l_iclog;
	ioerrors = 0;
	funcdidcallbacks = 0;
	repeats = 0;

	do {
		/*
		 * Scan all iclogs starting with the one pointed to by the
		 * log.  Reset this starting point each time the log is
		 * unlocked (during callbacks).
		 *
		 * Keep looping through iclogs until one full pass is made
		 * without running any callbacks.
		 */
		first_iclog = log->l_iclog;
		iclog = log->l_iclog;
		loopdidcallbacks = 0;
		repeats++;

		do {

			/* skip all iclogs in the ACTIVE & DIRTY states */
			if (iclog->ic_state &
			    (XLOG_STATE_ACTIVE|XLOG_STATE_DIRTY)) {
				iclog = iclog->ic_next;
				continue;
			}

			/*
			 * Between marking a filesystem SHUTDOWN and stopping
			 * the log, we do flush all iclogs to disk (if there
			 * wasn't a log I/O error). So, we do want things to
			 * go smoothly in case of just a SHUTDOWN  w/o a
			 * LOG_IO_ERROR.
			 */
			if (!(iclog->ic_state & XLOG_STATE_IOERROR)) {
				/*
				 * Can only perform callbacks in order.  Since
				 * this iclog is not in the DONE_SYNC/
				 * DO_CALLBACK state, we skip the rest and
				 * just try to clean up.  If we set our iclog
				 * to DO_CALLBACK, we will not process it when
				 * we retry since a previous iclog is in the
				 * CALLBACK and the state cannot change since
Eric Sandeen's avatar
Eric Sandeen committed
2683
				 * we are holding the l_icloglock.
Linus Torvalds's avatar
Linus Torvalds committed
2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708
				 */
				if (!(iclog->ic_state &
					(XLOG_STATE_DONE_SYNC |
						 XLOG_STATE_DO_CALLBACK))) {
					if (ciclog && (ciclog->ic_state ==
							XLOG_STATE_DONE_SYNC)) {
						ciclog->ic_state = XLOG_STATE_DO_CALLBACK;
					}
					break;
				}
				/*
				 * We now have an iclog that is in either the
				 * DO_CALLBACK or DONE_SYNC states. The other
				 * states (WANT_SYNC, SYNCING, or CALLBACK were
				 * caught by the above if and are going to
				 * clean (i.e. we aren't doing their callbacks)
				 * see the above if.
				 */

				/*
				 * We will do one more check here to see if we
				 * have chased our tail around.
				 */

				lowest_lsn = xlog_get_lowest_lsn(log);
2709 2710
				if (lowest_lsn &&
				    XFS_LSN_CMP(lowest_lsn,
2711
						be64_to_cpu(iclog->ic_header.h_lsn)) < 0) {
Linus Torvalds's avatar
Linus Torvalds committed
2712 2713 2714 2715 2716 2717 2718 2719
					iclog = iclog->ic_next;
					continue; /* Leave this iclog for
						   * another thread */
				}

				iclog->ic_state = XLOG_STATE_CALLBACK;


2720
				/*
2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733
				 * Completion of a iclog IO does not imply that
				 * a transaction has completed, as transactions
				 * can be large enough to span many iclogs. We
				 * cannot change the tail of the log half way
				 * through a transaction as this may be the only
				 * transaction in the log and moving th etail to
				 * point to the middle of it will prevent
				 * recovery from finding the start of the
				 * transaction. Hence we should only update the
				 * last_sync_lsn if this iclog contains
				 * transaction completion callbacks on it.
				 *
				 * We have to do this before we drop the
2734 2735
				 * icloglock to ensure we are the only one that
				 * can update it.
Linus Torvalds's avatar
Linus Torvalds committed
2736
				 */
2737 2738
				ASSERT(XFS_LSN_CMP(atomic64_read(&log->l_last_sync_lsn),
					be64_to_cpu(iclog->ic_header.h_lsn)) <= 0);
2739 2740 2741
				if (iclog->ic_callback)
					atomic64_set(&log->l_last_sync_lsn,
						be64_to_cpu(iclog->ic_header.h_lsn));
Linus Torvalds's avatar
Linus Torvalds committed
2742

2743
			} else
Linus Torvalds's avatar
Linus Torvalds committed
2744
				ioerrors++;
2745 2746

			spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
2747

2748 2749 2750 2751 2752 2753 2754 2755 2756
			/*
			 * Keep processing entries in the callback list until
			 * we come around and it is empty.  We need to
			 * atomically see that the list is empty and change the
			 * state to DIRTY so that we don't miss any more
			 * callbacks being added.
			 */
			spin_lock(&iclog->ic_callback_lock);
			cb = iclog->ic_callback;
2757
			while (cb) {
Linus Torvalds's avatar
Linus Torvalds committed
2758 2759
				iclog->ic_callback_tail = &(iclog->ic_callback);
				iclog->ic_callback = NULL;
2760
				spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds's avatar
Linus Torvalds committed
2761 2762

				/* perform callbacks in the order given */
2763
				for (; cb; cb = cb_next) {
Linus Torvalds's avatar
Linus Torvalds committed
2764 2765 2766
					cb_next = cb->cb_next;
					cb->cb_func(cb->cb_arg, aborted);
				}
2767
				spin_lock(&iclog->ic_callback_lock);
Linus Torvalds's avatar
Linus Torvalds committed
2768 2769 2770 2771 2772 2773
				cb = iclog->ic_callback;
			}

			loopdidcallbacks++;
			funcdidcallbacks++;

2774
			spin_lock(&log->l_icloglock);
2775
			ASSERT(iclog->ic_callback == NULL);
2776
			spin_unlock(&iclog->ic_callback_lock);
Linus Torvalds's avatar
Linus Torvalds committed
2777 2778 2779 2780 2781 2782 2783 2784 2785 2786
			if (!(iclog->ic_state & XLOG_STATE_IOERROR))
				iclog->ic_state = XLOG_STATE_DIRTY;

			/*
			 * Transition from DIRTY to ACTIVE if applicable.
			 * NOP if STATE_IOERROR.
			 */
			xlog_state_clean_log(log);

			/* wake up threads waiting in xfs_log_force() */
2787
			wake_up_all(&iclog->ic_force_wait);
Linus Torvalds's avatar
Linus Torvalds committed
2788 2789 2790

			iclog = iclog->ic_next;
		} while (first_iclog != iclog);
2791 2792 2793 2794

		if (repeats > 5000) {
			flushcnt += repeats;
			repeats = 0;
2795
			xfs_warn(log->l_mp,
2796
				"%s: possible infinite loop (%d iterations)",
2797
				__func__, flushcnt);
Linus Torvalds's avatar
Linus Torvalds committed
2798 2799 2800
		}
	} while (!ioerrors && loopdidcallbacks);

2801
#ifdef DEBUG
Linus Torvalds's avatar
Linus Torvalds committed
2802
	/*
2803 2804 2805 2806 2807 2808 2809 2810 2811 2812
	 * Make one last gasp attempt to see if iclogs are being left in limbo.
	 * If the above loop finds an iclog earlier than the current iclog and
	 * in one of the syncing states, the current iclog is put into
	 * DO_CALLBACK and the callbacks are deferred to the completion of the
	 * earlier iclog. Walk the iclogs in order and make sure that no iclog
	 * is in DO_CALLBACK unless an earlier iclog is in one of the syncing
	 * states.
	 *
	 * Note that SYNCING|IOABORT is a valid state so we cannot just check
	 * for ic_state == SYNCING.
Linus Torvalds's avatar
Linus Torvalds committed
2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823
	 */
	if (funcdidcallbacks) {
		first_iclog = iclog = log->l_iclog;
		do {
			ASSERT(iclog->ic_state != XLOG_STATE_DO_CALLBACK);
			/*
			 * Terminate the loop if iclogs are found in states
			 * which will cause other threads to clean up iclogs.
			 *
			 * SYNCING - i/o completion will go through logs
			 * DONE_SYNC - interrupt thread should be waiting for
Eric Sandeen's avatar
Eric Sandeen committed
2824
			 *              l_icloglock
Linus Torvalds's avatar
Linus Torvalds committed
2825 2826 2827
			 * IOERROR - give up hope all ye who enter here
			 */
			if (iclog->ic_state == XLOG_STATE_WANT_SYNC ||
2828
			    iclog->ic_state & XLOG_STATE_SYNCING ||
Linus Torvalds's avatar
Linus Torvalds committed
2829 2830 2831 2832 2833 2834 2835 2836
			    iclog->ic_state == XLOG_STATE_DONE_SYNC ||
			    iclog->ic_state == XLOG_STATE_IOERROR )
				break;
			iclog = iclog->ic_next;
		} while (first_iclog != iclog);
	}
#endif

2837
	if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR))
2838
		wake_up_all(&log->l_flush_wait);
2839 2840

	spin_unlock(&log->l_icloglock);
2841
}
Linus Torvalds's avatar
Linus Torvalds committed
2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854


/*
 * Finish transitioning this iclog to the dirty state.
 *
 * Make sure that we completely execute this routine only when this is
 * the last call to the iclog.  There is a good chance that iclog flushes,
 * when we reach the end of the physical log, get turned into 2 separate
 * calls to bwrite.  Hence, one iclog flush could generate two calls to this
 * routine.  By using the reference count bwritecnt, we guarantee that only
 * the second completion goes through.
 *
 * Callbacks could take time, so they are done outside the scope of the
2855
 * global state machine log lock.
Linus Torvalds's avatar
Linus Torvalds committed
2856
 */
2857
STATIC void
Linus Torvalds's avatar
Linus Torvalds committed
2858 2859 2860 2861
xlog_state_done_syncing(
	xlog_in_core_t	*iclog,
	int		aborted)
{
Mark Tinguely's avatar
Mark Tinguely committed
2862
	struct xlog	   *log = iclog->ic_log;
Linus Torvalds's avatar
Linus Torvalds committed
2863

Eric Sandeen's avatar
Eric Sandeen committed
2864
	spin_lock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
2865 2866 2867

	ASSERT(iclog->ic_state == XLOG_STATE_SYNCING ||
	       iclog->ic_state == XLOG_STATE_IOERROR);
2868
	ASSERT(atomic_read(&iclog->ic_refcnt) == 0);
Linus Torvalds's avatar
Linus Torvalds committed
2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879
	ASSERT(iclog->ic_bwritecnt == 1 || iclog->ic_bwritecnt == 2);


	/*
	 * If we got an error, either on the first buffer, or in the case of
	 * split log writes, on the second, we mark ALL iclogs STATE_IOERROR,
	 * and none should ever be attempted to be written to disk
	 * again.
	 */
	if (iclog->ic_state != XLOG_STATE_IOERROR) {
		if (--iclog->ic_bwritecnt == 1) {
Eric Sandeen's avatar
Eric Sandeen committed
2880
			spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
2881 2882 2883 2884 2885 2886 2887 2888 2889 2890
			return;
		}
		iclog->ic_state = XLOG_STATE_DONE_SYNC;
	}

	/*
	 * Someone could be sleeping prior to writing out the next
	 * iclog buffer, we wake them all, one will get to do the
	 * I/O, the others get to wait for the result.
	 */
2891
	wake_up_all(&iclog->ic_write_wait);
Eric Sandeen's avatar
Eric Sandeen committed
2892
	spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
2893 2894 2895 2896 2897 2898
	xlog_state_do_callback(log, aborted, iclog);	/* also cleans log */
}	/* xlog_state_done_syncing */


/*
 * If the head of the in-core log ring is not (ACTIVE or DIRTY), then we must
2899 2900 2901
 * sleep.  We wait on the flush queue on the head iclog as that should be
 * the first iclog to complete flushing. Hence if all iclogs are syncing,
 * we will wait here and all new writes will sleep until a sync completes.
Linus Torvalds's avatar
Linus Torvalds committed
2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914
 *
 * The in-core logs are used in a circular fashion. They are not used
 * out-of-order even when an iclog past the head is free.
 *
 * return:
 *	* log_offset where xlog_write() can start writing into the in-core
 *		log's data space.
 *	* in-core log pointer to which xlog_write() should write.
 *	* boolean indicating this is a continued write to an in-core log.
 *		If this is the last write, then the in-core log's offset field
 *		needs to be incremented, depending on the amount of data which
 *		is copied.
 */
2915
STATIC int
Mark Tinguely's avatar
Mark Tinguely committed
2916 2917 2918 2919 2920 2921 2922
xlog_state_get_iclog_space(
	struct xlog		*log,
	int			len,
	struct xlog_in_core	**iclogp,
	struct xlog_ticket	*ticket,
	int			*continued_write,
	int			*logoffsetp)
Linus Torvalds's avatar
Linus Torvalds committed
2923 2924 2925 2926 2927 2928 2929
{
	int		  log_offset;
	xlog_rec_header_t *head;
	xlog_in_core_t	  *iclog;
	int		  error;

restart:
Eric Sandeen's avatar
Eric Sandeen committed
2930
	spin_lock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
2931
	if (XLOG_FORCED_SHUTDOWN(log)) {
Eric Sandeen's avatar
Eric Sandeen committed
2932
		spin_unlock(&log->l_icloglock);
2933
		return -EIO;
Linus Torvalds's avatar
Linus Torvalds committed
2934 2935 2936
	}

	iclog = log->l_iclog;
2937
	if (iclog->ic_state != XLOG_STATE_ACTIVE) {
2938
		XFS_STATS_INC(log->l_mp, xs_log_noiclogs);
2939 2940

		/* Wait for log writes to have flushed */
2941
		xlog_wait(&log->l_flush_wait, &log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
2942 2943
		goto restart;
	}
2944

Linus Torvalds's avatar
Linus Torvalds committed
2945 2946
	head = &iclog->ic_header;

2947
	atomic_inc(&iclog->ic_refcnt);	/* prevents sync */
Linus Torvalds's avatar
Linus Torvalds committed
2948 2949 2950 2951 2952 2953 2954 2955 2956
	log_offset = iclog->ic_offset;

	/* On the 1st write to an iclog, figure out lsn.  This works
	 * if iclogs marked XLOG_STATE_WANT_SYNC always write out what they are
	 * committing to.  If the offset is set, that's how many blocks
	 * must be written.
	 */
	if (log_offset == 0) {
		ticket->t_curr_res -= log->l_iclog_hsize;
2957
		xlog_tic_add_region(ticket,
2958 2959
				    log->l_iclog_hsize,
				    XLOG_REG_TYPE_LRHEADER);
2960 2961
		head->h_cycle = cpu_to_be32(log->l_curr_cycle);
		head->h_lsn = cpu_to_be64(
2962
			xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block));
Linus Torvalds's avatar
Linus Torvalds committed
2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977
		ASSERT(log->l_curr_block >= 0);
	}

	/* If there is enough room to write everything, then do it.  Otherwise,
	 * claim the rest of the region and make sure the XLOG_STATE_WANT_SYNC
	 * bit is on, so this will get flushed out.  Don't update ic_offset
	 * until you know exactly how many bytes get copied.  Therefore, wait
	 * until later to update ic_offset.
	 *
	 * xlog_write() algorithm assumes that at least 2 xlog_op_header_t's
	 * can fit into remaining data section.
	 */
	if (iclog->ic_size - iclog->ic_offset < 2*sizeof(xlog_op_header_t)) {
		xlog_state_switch_iclogs(log, iclog, iclog->ic_size);

2978 2979 2980 2981 2982 2983 2984 2985 2986
		/*
		 * If I'm the only one writing to this iclog, sync it to disk.
		 * We need to do an atomic compare and decrement here to avoid
		 * racing with concurrent atomic_dec_and_lock() calls in
		 * xlog_state_release_iclog() when there is more than one
		 * reference to the iclog.
		 */
		if (!atomic_add_unless(&iclog->ic_refcnt, -1, 1)) {
			/* we are the only one */
Eric Sandeen's avatar
Eric Sandeen committed
2987
			spin_unlock(&log->l_icloglock);
2988 2989
			error = xlog_state_release_iclog(log, iclog);
			if (error)
2990
				return error;
Linus Torvalds's avatar
Linus Torvalds committed
2991
		} else {
Eric Sandeen's avatar
Eric Sandeen committed
2992
			spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012
		}
		goto restart;
	}

	/* Do we have enough room to write the full amount in the remainder
	 * of this iclog?  Or must we continue a write on the next iclog and
	 * mark this iclog as completely taken?  In the case where we switch
	 * iclogs (to mark it taken), this particular iclog will release/sync
	 * to disk in xlog_write().
	 */
	if (len <= iclog->ic_size - iclog->ic_offset) {
		*continued_write = 0;
		iclog->ic_offset += len;
	} else {
		*continued_write = 1;
		xlog_state_switch_iclogs(log, iclog, iclog->ic_size);
	}
	*iclogp = iclog;

	ASSERT(iclog->ic_offset <= iclog->ic_size);
Eric Sandeen's avatar
Eric Sandeen committed
3013
	spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026

	*logoffsetp = log_offset;
	return 0;
}	/* xlog_state_get_iclog_space */

/* The first cnt-1 times through here we don't need to
 * move the grant write head because the permanent
 * reservation has reserved cnt times the unit amount.
 * Release part of current permanent unit reservation and
 * reset current reservation to be one units worth.  Also
 * move grant reservation head forward.
 */
STATIC void
Mark Tinguely's avatar
Mark Tinguely committed
3027 3028 3029
xlog_regrant_reserve_log_space(
	struct xlog		*log,
	struct xlog_ticket	*ticket)
Linus Torvalds's avatar
Linus Torvalds committed
3030
{
3031 3032
	trace_xfs_log_regrant_reserve_enter(log, ticket);

Linus Torvalds's avatar
Linus Torvalds committed
3033 3034 3035
	if (ticket->t_cnt > 0)
		ticket->t_cnt--;

3036
	xlog_grant_sub_space(log, &log->l_reserve_head.grant,
3037
					ticket->t_curr_res);
3038
	xlog_grant_sub_space(log, &log->l_write_head.grant,
3039
					ticket->t_curr_res);
Linus Torvalds's avatar
Linus Torvalds committed
3040
	ticket->t_curr_res = ticket->t_unit_res;
3041
	xlog_tic_reset_res(ticket);
3042 3043 3044

	trace_xfs_log_regrant_reserve_sub(log, ticket);

Linus Torvalds's avatar
Linus Torvalds committed
3045
	/* just return if we still have some of the pre-reserved space */
3046
	if (ticket->t_cnt > 0)
Linus Torvalds's avatar
Linus Torvalds committed
3047 3048
		return;

3049
	xlog_grant_add_space(log, &log->l_reserve_head.grant,
3050
					ticket->t_unit_res);
3051 3052 3053

	trace_xfs_log_regrant_reserve_exit(log, ticket);

Linus Torvalds's avatar
Linus Torvalds committed
3054
	ticket->t_curr_res = ticket->t_unit_res;
3055
	xlog_tic_reset_res(ticket);
Linus Torvalds's avatar
Linus Torvalds committed
3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073
}	/* xlog_regrant_reserve_log_space */


/*
 * Give back the space left from a reservation.
 *
 * All the information we need to make a correct determination of space left
 * is present.  For non-permanent reservations, things are quite easy.  The
 * count should have been decremented to zero.  We only need to deal with the
 * space remaining in the current reservation part of the ticket.  If the
 * ticket contains a permanent reservation, there may be left over space which
 * needs to be released.  A count of N means that N-1 refills of the current
 * reservation can be done before we need to ask for more space.  The first
 * one goes to fill up the first current reservation.  Once we run out of
 * space, the count will stay at zero and the only space remaining will be
 * in the current reservation field.
 */
STATIC void
Mark Tinguely's avatar
Mark Tinguely committed
3074 3075 3076
xlog_ungrant_log_space(
	struct xlog		*log,
	struct xlog_ticket	*ticket)
Linus Torvalds's avatar
Linus Torvalds committed
3077
{
3078 3079
	int	bytes;

Linus Torvalds's avatar
Linus Torvalds committed
3080 3081 3082
	if (ticket->t_cnt > 0)
		ticket->t_cnt--;

3083 3084
	trace_xfs_log_ungrant_enter(log, ticket);
	trace_xfs_log_ungrant_sub(log, ticket);
Linus Torvalds's avatar
Linus Torvalds committed
3085

3086 3087
	/*
	 * If this is a permanent reservation ticket, we may be able to free
Linus Torvalds's avatar
Linus Torvalds committed
3088 3089
	 * up more space based on the remaining count.
	 */
3090
	bytes = ticket->t_curr_res;
Linus Torvalds's avatar
Linus Torvalds committed
3091 3092
	if (ticket->t_cnt > 0) {
		ASSERT(ticket->t_flags & XLOG_TIC_PERM_RESERV);
3093
		bytes += ticket->t_unit_res*ticket->t_cnt;
Linus Torvalds's avatar
Linus Torvalds committed
3094 3095
	}

3096 3097
	xlog_grant_sub_space(log, &log->l_reserve_head.grant, bytes);
	xlog_grant_sub_space(log, &log->l_write_head.grant, bytes);
3098

3099 3100
	trace_xfs_log_ungrant_exit(log, ticket);

3101
	xfs_log_space_wake(log->l_mp);
3102
}
Linus Torvalds's avatar
Linus Torvalds committed
3103 3104 3105 3106 3107 3108 3109 3110 3111 3112

/*
 * Flush iclog to disk if this is the last reference to the given iclog and
 * the WANT_SYNC bit is set.
 *
 * When this function is entered, the iclog is not necessarily in the
 * WANT_SYNC state.  It may be sitting around waiting to get filled.
 *
 *
 */
3113
STATIC int
3114
xlog_state_release_iclog(
Mark Tinguely's avatar
Mark Tinguely committed
3115 3116
	struct xlog		*log,
	struct xlog_in_core	*iclog)
Linus Torvalds's avatar
Linus Torvalds committed
3117 3118 3119
{
	int		sync = 0;	/* do we sync? */

3120
	if (iclog->ic_state & XLOG_STATE_IOERROR)
3121
		return -EIO;
3122 3123 3124 3125 3126

	ASSERT(atomic_read(&iclog->ic_refcnt) > 0);
	if (!atomic_dec_and_lock(&iclog->ic_refcnt, &log->l_icloglock))
		return 0;

Linus Torvalds's avatar
Linus Torvalds committed
3127
	if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeen's avatar
Eric Sandeen committed
3128
		spin_unlock(&log->l_icloglock);
3129
		return -EIO;
Linus Torvalds's avatar
Linus Torvalds committed
3130 3131 3132 3133
	}
	ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE ||
	       iclog->ic_state == XLOG_STATE_WANT_SYNC);

3134
	if (iclog->ic_state == XLOG_STATE_WANT_SYNC) {
3135
		/* update tail before writing to iclog */
3136
		xfs_lsn_t tail_lsn = xlog_assign_tail_lsn(log->l_mp);
Linus Torvalds's avatar
Linus Torvalds committed
3137 3138
		sync++;
		iclog->ic_state = XLOG_STATE_SYNCING;
3139 3140
		iclog->ic_header.h_tail_lsn = cpu_to_be64(tail_lsn);
		xlog_verify_tail_lsn(log, iclog, tail_lsn);
Linus Torvalds's avatar
Linus Torvalds committed
3141 3142
		/* cycle incremented when incrementing curr_block */
	}
Eric Sandeen's avatar
Eric Sandeen committed
3143
	spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3144 3145 3146

	/*
	 * We let the log lock go, so it's possible that we hit a log I/O
3147
	 * error or some other SHUTDOWN condition that marks the iclog
Linus Torvalds's avatar
Linus Torvalds committed
3148 3149 3150 3151
	 * as XLOG_STATE_IOERROR before the bwrite. However, we know that
	 * this iclog has consistent data, so we ignore IOERROR
	 * flags after this point.
	 */
3152
	if (sync)
Linus Torvalds's avatar
Linus Torvalds committed
3153
		return xlog_sync(log, iclog);
3154
	return 0;
Linus Torvalds's avatar
Linus Torvalds committed
3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165
}	/* xlog_state_release_iclog */


/*
 * This routine will mark the current iclog in the ring as WANT_SYNC
 * and move the current iclog pointer to the next iclog in the ring.
 * When this routine is called from xlog_state_get_iclog_space(), the
 * exact size of the iclog has not yet been determined.  All we know is
 * that every data block.  We have run out of space in this log record.
 */
STATIC void
Mark Tinguely's avatar
Mark Tinguely committed
3166 3167 3168 3169
xlog_state_switch_iclogs(
	struct xlog		*log,
	struct xlog_in_core	*iclog,
	int			eventual_size)
Linus Torvalds's avatar
Linus Torvalds committed
3170 3171 3172 3173 3174
{
	ASSERT(iclog->ic_state == XLOG_STATE_ACTIVE);
	if (!eventual_size)
		eventual_size = iclog->ic_offset;
	iclog->ic_state = XLOG_STATE_WANT_SYNC;
3175
	iclog->ic_header.h_prev_block = cpu_to_be32(log->l_prev_block);
Linus Torvalds's avatar
Linus Torvalds committed
3176 3177 3178 3179 3180 3181 3182
	log->l_prev_block = log->l_curr_block;
	log->l_prev_cycle = log->l_curr_cycle;

	/* roll log?: ic_offset changed later */
	log->l_curr_block += BTOBB(eventual_size)+BTOBB(log->l_iclog_hsize);

	/* Round up to next log-sunit */
3183
	if (xfs_sb_version_haslogv2(&log->l_mp->m_sb) &&
Linus Torvalds's avatar
Linus Torvalds committed
3184 3185 3186 3187 3188 3189
	    log->l_mp->m_sb.sb_logsunit > 1) {
		__uint32_t sunit_bb = BTOBB(log->l_mp->m_sb.sb_logsunit);
		log->l_curr_block = roundup(log->l_curr_block, sunit_bb);
	}

	if (log->l_curr_block >= log->l_logBBsize) {
3190 3191 3192 3193 3194 3195 3196 3197 3198 3199
		/*
		 * Rewind the current block before the cycle is bumped to make
		 * sure that the combined LSN never transiently moves forward
		 * when the log wraps to the next cycle. This is to support the
		 * unlocked sample of these fields from xlog_valid_lsn(). Most
		 * other cases should acquire l_icloglock.
		 */
		log->l_curr_block -= log->l_logBBsize;
		ASSERT(log->l_curr_block >= 0);
		smp_wmb();
Linus Torvalds's avatar
Linus Torvalds committed
3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223
		log->l_curr_cycle++;
		if (log->l_curr_cycle == XLOG_HEADER_MAGIC_NUM)
			log->l_curr_cycle++;
	}
	ASSERT(iclog == log->l_iclog);
	log->l_iclog = iclog->ic_next;
}	/* xlog_state_switch_iclogs */

/*
 * Write out all data in the in-core log as of this exact moment in time.
 *
 * Data may be written to the in-core log during this call.  However,
 * we don't guarantee this data will be written out.  A change from past
 * implementation means this routine will *not* write out zero length LRs.
 *
 * Basically, we try and perform an intelligent scan of the in-core logs.
 * If we determine there is no flushable data, we just return.  There is no
 * flushable data if:
 *
 *	1. the current iclog is active and has no data; the previous iclog
 *		is in the active or dirty state.
 *	2. the current iclog is drity, and the previous iclog is in the
 *		active or dirty state.
 *
3224
 * We may sleep if:
Linus Torvalds's avatar
Linus Torvalds committed
3225 3226 3227 3228 3229 3230 3231 3232 3233 3234
 *
 *	1. the current iclog is not in the active nor dirty state.
 *	2. the current iclog dirty, and the previous iclog is not in the
 *		active nor dirty state.
 *	3. the current iclog is active, and there is another thread writing
 *		to this particular iclog.
 *	4. a) the current iclog is active and has no other writers
 *	   b) when we return from flushing out this iclog, it is still
 *		not in the active nor dirty state.
 */
3235 3236 3237 3238 3239
int
_xfs_log_force(
	struct xfs_mount	*mp,
	uint			flags,
	int			*log_flushed)
Linus Torvalds's avatar
Linus Torvalds committed
3240
{
3241
	struct xlog		*log = mp->m_log;
3242 3243 3244
	struct xlog_in_core	*iclog;
	xfs_lsn_t		lsn;

3245
	XFS_STATS_INC(mp, xs_log_force);
Linus Torvalds's avatar
Linus Torvalds committed
3246

3247
	xlog_cil_force(log);
3248

Eric Sandeen's avatar
Eric Sandeen committed
3249
	spin_lock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3250 3251 3252

	iclog = log->l_iclog;
	if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeen's avatar
Eric Sandeen committed
3253
		spin_unlock(&log->l_icloglock);
3254
		return -EIO;
Linus Torvalds's avatar
Linus Torvalds committed
3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269
	}

	/* If the head iclog is not active nor dirty, we just attach
	 * ourselves to the head and go to sleep.
	 */
	if (iclog->ic_state == XLOG_STATE_ACTIVE ||
	    iclog->ic_state == XLOG_STATE_DIRTY) {
		/*
		 * If the head is dirty or (active and empty), then
		 * we need to look at the previous iclog.  If the previous
		 * iclog is active or dirty we are done.  There is nothing
		 * to sync out.  Otherwise, we attach ourselves to the
		 * previous iclog and go to sleep.
		 */
		if (iclog->ic_state == XLOG_STATE_DIRTY ||
3270 3271
		    (atomic_read(&iclog->ic_refcnt) == 0
		     && iclog->ic_offset == 0)) {
Linus Torvalds's avatar
Linus Torvalds committed
3272 3273 3274 3275 3276 3277 3278
			iclog = iclog->ic_prev;
			if (iclog->ic_state == XLOG_STATE_ACTIVE ||
			    iclog->ic_state == XLOG_STATE_DIRTY)
				goto no_sleep;
			else
				goto maybe_sleep;
		} else {
3279
			if (atomic_read(&iclog->ic_refcnt) == 0) {
Linus Torvalds's avatar
Linus Torvalds committed
3280 3281 3282 3283 3284 3285
				/* We are the only one with access to this
				 * iclog.  Flush it out now.  There should
				 * be a roundoff of zero to show that someone
				 * has already taken care of the roundoff from
				 * the previous sync.
				 */
3286
				atomic_inc(&iclog->ic_refcnt);
3287
				lsn = be64_to_cpu(iclog->ic_header.h_lsn);
Linus Torvalds's avatar
Linus Torvalds committed
3288
				xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeen's avatar
Eric Sandeen committed
3289
				spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3290 3291

				if (xlog_state_release_iclog(log, iclog))
3292
					return -EIO;
3293 3294 3295

				if (log_flushed)
					*log_flushed = 1;
Eric Sandeen's avatar
Eric Sandeen committed
3296
				spin_lock(&log->l_icloglock);
3297
				if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
Linus Torvalds's avatar
Linus Torvalds committed
3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321
				    iclog->ic_state != XLOG_STATE_DIRTY)
					goto maybe_sleep;
				else
					goto no_sleep;
			} else {
				/* Someone else is writing to this iclog.
				 * Use its call to flush out the data.  However,
				 * the other thread may not force out this LR,
				 * so we mark it WANT_SYNC.
				 */
				xlog_state_switch_iclogs(log, iclog, 0);
				goto maybe_sleep;
			}
		}
	}

	/* By the time we come around again, the iclog could've been filled
	 * which would give it another lsn.  If we have a new lsn, just
	 * return because the relevant data has been flushed.
	 */
maybe_sleep:
	if (flags & XFS_LOG_SYNC) {
		/*
		 * We must check if we're shutting down here, before
Eric Sandeen's avatar
Eric Sandeen committed
3322
		 * we wait, while we're holding the l_icloglock.
Linus Torvalds's avatar
Linus Torvalds committed
3323 3324 3325 3326
		 * Then we check again after waking up, in case our
		 * sleep was disturbed by a bad news.
		 */
		if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeen's avatar
Eric Sandeen committed
3327
			spin_unlock(&log->l_icloglock);
3328
			return -EIO;
Linus Torvalds's avatar
Linus Torvalds committed
3329
		}
3330
		XFS_STATS_INC(mp, xs_log_force_sleep);
3331
		xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3332 3333 3334 3335 3336 3337
		/*
		 * No need to grab the log lock here since we're
		 * only deciding whether or not to return EIO
		 * and the memory read should be atomic.
		 */
		if (iclog->ic_state & XLOG_STATE_IOERROR)
3338
			return -EIO;
Linus Torvalds's avatar
Linus Torvalds committed
3339 3340 3341
	} else {

no_sleep:
Eric Sandeen's avatar
Eric Sandeen committed
3342
		spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3343 3344
	}
	return 0;
3345
}
Linus Torvalds's avatar
Linus Torvalds committed
3346

3347 3348 3349 3350 3351 3352 3353 3354 3355 3356
/*
 * Wrapper for _xfs_log_force(), to be used when caller doesn't care
 * about errors or whether the log was flushed or not. This is the normal
 * interface to use when trying to unpin items or move the log forward.
 */
void
xfs_log_force(
	xfs_mount_t	*mp,
	uint		flags)
{
3357
	trace_xfs_log_force(mp, 0, _RET_IP_);
3358
	_xfs_log_force(mp, flags, NULL);
3359
}
Linus Torvalds's avatar
Linus Torvalds committed
3360 3361

/*
3362
 * Force the in-core log to disk for a specific LSN.
Linus Torvalds's avatar
Linus Torvalds committed
3363 3364 3365 3366 3367 3368 3369
 *
 * Find in-core log with lsn.
 *	If it is in the DIRTY state, just return.
 *	If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
 *		state and go to sleep or return.
 *	If it is in any other state, go to sleep or return.
 *
3370 3371 3372 3373 3374
 * Synchronous forces are implemented with a signal variable. All callers
 * to force a given lsn to disk will wait on a the sv attached to the
 * specific in-core log.  When given in-core log finally completes its
 * write to disk, that thread will wake up all threads waiting on the
 * sv.
Linus Torvalds's avatar
Linus Torvalds committed
3375
 */
3376 3377 3378 3379 3380 3381
int
_xfs_log_force_lsn(
	struct xfs_mount	*mp,
	xfs_lsn_t		lsn,
	uint			flags,
	int			*log_flushed)
Linus Torvalds's avatar
Linus Torvalds committed
3382
{
3383
	struct xlog		*log = mp->m_log;
3384 3385
	struct xlog_in_core	*iclog;
	int			already_slept = 0;
Linus Torvalds's avatar
Linus Torvalds committed
3386

3387
	ASSERT(lsn != 0);
Linus Torvalds's avatar
Linus Torvalds committed
3388

3389
	XFS_STATS_INC(mp, xs_log_force);
Linus Torvalds's avatar
Linus Torvalds committed
3390

3391 3392 3393
	lsn = xlog_cil_force_lsn(log, lsn);
	if (lsn == NULLCOMMITLSN)
		return 0;
3394

3395 3396 3397 3398
try_again:
	spin_lock(&log->l_icloglock);
	iclog = log->l_iclog;
	if (iclog->ic_state & XLOG_STATE_IOERROR) {
Eric Sandeen's avatar
Eric Sandeen committed
3399
		spin_unlock(&log->l_icloglock);
3400
		return -EIO;
Linus Torvalds's avatar
Linus Torvalds committed
3401 3402
	}

3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437
	do {
		if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
			iclog = iclog->ic_next;
			continue;
		}

		if (iclog->ic_state == XLOG_STATE_DIRTY) {
			spin_unlock(&log->l_icloglock);
			return 0;
		}

		if (iclog->ic_state == XLOG_STATE_ACTIVE) {
			/*
			 * We sleep here if we haven't already slept (e.g.
			 * this is the first time we've looked at the correct
			 * iclog buf) and the buffer before us is going to
			 * be sync'ed. The reason for this is that if we
			 * are doing sync transactions here, by waiting for
			 * the previous I/O to complete, we can allow a few
			 * more transactions into this iclog before we close
			 * it down.
			 *
			 * Otherwise, we mark the buffer WANT_SYNC, and bump
			 * up the refcnt so we can release the log (which
			 * drops the ref count).  The state switch keeps new
			 * transaction commits from using this buffer.  When
			 * the current commits finish writing into the buffer,
			 * the refcount will drop to zero and the buffer will
			 * go out then.
			 */
			if (!already_slept &&
			    (iclog->ic_prev->ic_state &
			     (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) {
				ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));

3438
				XFS_STATS_INC(mp, xs_log_force_sleep);
3439

3440 3441
				xlog_wait(&iclog->ic_prev->ic_write_wait,
							&log->l_icloglock);
3442 3443 3444
				already_slept = 1;
				goto try_again;
			}
3445
			atomic_inc(&iclog->ic_refcnt);
Linus Torvalds's avatar
Linus Torvalds committed
3446
			xlog_state_switch_iclogs(log, iclog, 0);
Eric Sandeen's avatar
Eric Sandeen committed
3447
			spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3448
			if (xlog_state_release_iclog(log, iclog))
3449
				return -EIO;
3450 3451
			if (log_flushed)
				*log_flushed = 1;
Eric Sandeen's avatar
Eric Sandeen committed
3452
			spin_lock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3453 3454
		}

3455 3456 3457 3458 3459 3460 3461 3462 3463
		if ((flags & XFS_LOG_SYNC) && /* sleep */
		    !(iclog->ic_state &
		      (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
			/*
			 * Don't wait on completion if we know that we've
			 * gotten a log write error.
			 */
			if (iclog->ic_state & XLOG_STATE_IOERROR) {
				spin_unlock(&log->l_icloglock);
3464
				return -EIO;
3465
			}
3466
			XFS_STATS_INC(mp, xs_log_force_sleep);
3467
			xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
3468 3469 3470 3471 3472 3473
			/*
			 * No need to grab the log lock here since we're
			 * only deciding whether or not to return EIO
			 * and the memory read should be atomic.
			 */
			if (iclog->ic_state & XLOG_STATE_IOERROR)
3474
				return -EIO;
3475
		} else {		/* just return */
Eric Sandeen's avatar
Eric Sandeen committed
3476
			spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3477 3478
		}

3479 3480
		return 0;
	} while (iclog != log->l_iclog);
Linus Torvalds's avatar
Linus Torvalds committed
3481

3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496
	spin_unlock(&log->l_icloglock);
	return 0;
}

/*
 * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care
 * about errors or whether the log was flushed or not. This is the normal
 * interface to use when trying to unpin items or move the log forward.
 */
void
xfs_log_force_lsn(
	xfs_mount_t	*mp,
	xfs_lsn_t	lsn,
	uint		flags)
{
3497
	trace_xfs_log_force(mp, lsn, _RET_IP_);
3498
	_xfs_log_force_lsn(mp, lsn, flags, NULL);
3499
}
Linus Torvalds's avatar
Linus Torvalds committed
3500 3501 3502 3503 3504

/*
 * Called when we want to mark the current iclog as being ready to sync to
 * disk.
 */
3505
STATIC void
Mark Tinguely's avatar
Mark Tinguely committed
3506 3507 3508
xlog_state_want_sync(
	struct xlog		*log,
	struct xlog_in_core	*iclog)
Linus Torvalds's avatar
Linus Torvalds committed
3509
{
3510
	assert_spin_locked(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3511 3512 3513 3514 3515 3516 3517

	if (iclog->ic_state == XLOG_STATE_ACTIVE) {
		xlog_state_switch_iclogs(log, iclog, 0);
	} else {
		ASSERT(iclog->ic_state &
			(XLOG_STATE_WANT_SYNC|XLOG_STATE_IOERROR));
	}
3518
}
Linus Torvalds's avatar
Linus Torvalds committed
3519 3520 3521 3522 3523 3524 3525 3526 3527 3528


/*****************************************************************************
 *
 *		TICKET functions
 *
 *****************************************************************************
 */

/*
Malcolm Parsons's avatar
Malcolm Parsons committed
3529
 * Free a used ticket when its refcount falls to zero.
Linus Torvalds's avatar
Linus Torvalds committed
3530
 */
3531 3532 3533
void
xfs_log_ticket_put(
	xlog_ticket_t	*ticket)
Linus Torvalds's avatar
Linus Torvalds committed
3534
{
3535
	ASSERT(atomic_read(&ticket->t_ref) > 0);
3536
	if (atomic_dec_and_test(&ticket->t_ref))
3537 3538
		kmem_zone_free(xfs_log_ticket_zone, ticket);
}
Linus Torvalds's avatar
Linus Torvalds committed
3539

3540 3541 3542 3543 3544 3545 3546 3547
xlog_ticket_t *
xfs_log_ticket_get(
	xlog_ticket_t	*ticket)
{
	ASSERT(atomic_read(&ticket->t_ref) > 0);
	atomic_inc(&ticket->t_ref);
	return ticket;
}
Linus Torvalds's avatar
Linus Torvalds committed
3548 3549

/*
3550 3551
 * Figure out the total log space unit (in bytes) that would be
 * required for a log ticket.
Linus Torvalds's avatar
Linus Torvalds committed
3552
 */
3553 3554 3555 3556
int
xfs_log_calc_unit_res(
	struct xfs_mount	*mp,
	int			unit_bytes)
Linus Torvalds's avatar
Linus Torvalds committed
3557
{
3558 3559 3560
	struct xlog		*log = mp->m_log;
	int			iclog_space;
	uint			num_headers;
Linus Torvalds's avatar
Linus Torvalds committed
3561 3562 3563 3564 3565 3566 3567 3568

	/*
	 * Permanent reservations have up to 'cnt'-1 active log operations
	 * in the log.  A unit in this case is the amount of space for one
	 * of these log operations.  Normal reservations have a cnt of 1
	 * and their unit amount is the total amount of space required.
	 *
	 * The following lines of code account for non-transaction data
3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590
	 * which occupy space in the on-disk log.
	 *
	 * Normal form of a transaction is:
	 * <oph><trans-hdr><start-oph><reg1-oph><reg1><reg2-oph>...<commit-oph>
	 * and then there are LR hdrs, split-recs and roundoff at end of syncs.
	 *
	 * We need to account for all the leadup data and trailer data
	 * around the transaction data.
	 * And then we need to account for the worst case in terms of using
	 * more space.
	 * The worst case will happen if:
	 * - the placement of the transaction happens to be such that the
	 *   roundoff is at its maximum
	 * - the transaction data is synced before the commit record is synced
	 *   i.e. <transaction-data><roundoff> | <commit-rec><roundoff>
	 *   Therefore the commit record is in its own Log Record.
	 *   This can happen as the commit record is called with its
	 *   own region to xlog_write().
	 *   This then means that in the worst case, roundoff can happen for
	 *   the commit-rec as well.
	 *   The commit-rec is smaller than padding in this scenario and so it is
	 *   not added separately.
Linus Torvalds's avatar
Linus Torvalds committed
3591 3592
	 */

3593 3594 3595 3596
	/* for trans header */
	unit_bytes += sizeof(xlog_op_header_t);
	unit_bytes += sizeof(xfs_trans_header_t);

Linus Torvalds's avatar
Linus Torvalds committed
3597
	/* for start-rec */
3598 3599
	unit_bytes += sizeof(xlog_op_header_t);

3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628
	/*
	 * for LR headers - the space for data in an iclog is the size minus
	 * the space used for the headers. If we use the iclog size, then we
	 * undercalculate the number of headers required.
	 *
	 * Furthermore - the addition of op headers for split-recs might
	 * increase the space required enough to require more log and op
	 * headers, so take that into account too.
	 *
	 * IMPORTANT: This reservation makes the assumption that if this
	 * transaction is the first in an iclog and hence has the LR headers
	 * accounted to it, then the remaining space in the iclog is
	 * exclusively for this transaction.  i.e. if the transaction is larger
	 * than the iclog, it will be the only thing in that iclog.
	 * Fundamentally, this means we must pass the entire log vector to
	 * xlog_write to guarantee this.
	 */
	iclog_space = log->l_iclog_size - log->l_iclog_hsize;
	num_headers = howmany(unit_bytes, iclog_space);

	/* for split-recs - ophdrs added when data split over LRs */
	unit_bytes += sizeof(xlog_op_header_t) * num_headers;

	/* add extra header reservations if we overrun */
	while (!num_headers ||
	       howmany(unit_bytes, iclog_space) > num_headers) {
		unit_bytes += sizeof(xlog_op_header_t);
		num_headers++;
	}
3629
	unit_bytes += log->l_iclog_hsize * num_headers;
Linus Torvalds's avatar
Linus Torvalds committed
3630

3631 3632 3633 3634
	/* for commit-rec LR header - note: padding will subsume the ophdr */
	unit_bytes += log->l_iclog_hsize;

	/* for roundoff padding for transaction data and one for commit record */
3635
	if (xfs_sb_version_haslogv2(&mp->m_sb) && mp->m_sb.sb_logsunit > 1) {
Linus Torvalds's avatar
Linus Torvalds committed
3636
		/* log su roundoff */
3637
		unit_bytes += 2 * mp->m_sb.sb_logsunit;
Linus Torvalds's avatar
Linus Torvalds committed
3638 3639
	} else {
		/* BB roundoff */
3640
		unit_bytes += 2 * BBSIZE;
Linus Torvalds's avatar
Linus Torvalds committed
3641 3642
        }

3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666
	return unit_bytes;
}

/*
 * Allocate and initialise a new log ticket.
 */
struct xlog_ticket *
xlog_ticket_alloc(
	struct xlog		*log,
	int			unit_bytes,
	int			cnt,
	char			client,
	bool			permanent,
	xfs_km_flags_t		alloc_flags)
{
	struct xlog_ticket	*tic;
	int			unit_res;

	tic = kmem_zone_zalloc(xfs_log_ticket_zone, alloc_flags);
	if (!tic)
		return NULL;

	unit_res = xfs_log_calc_unit_res(log->l_mp, unit_bytes);

3667
	atomic_set(&tic->t_ref, 1);
3668
	tic->t_task		= current;
3669
	INIT_LIST_HEAD(&tic->t_queue);
3670 3671
	tic->t_unit_res		= unit_res;
	tic->t_curr_res		= unit_res;
Linus Torvalds's avatar
Linus Torvalds committed
3672 3673
	tic->t_cnt		= cnt;
	tic->t_ocnt		= cnt;
3674
	tic->t_tid		= prandom_u32();
Linus Torvalds's avatar
Linus Torvalds committed
3675 3676
	tic->t_clientid		= client;
	tic->t_flags		= XLOG_TIC_INITED;
3677
	if (permanent)
Linus Torvalds's avatar
Linus Torvalds committed
3678 3679
		tic->t_flags |= XLOG_TIC_PERM_RESERV;

3680
	xlog_tic_reset_res(tic);
3681

Linus Torvalds's avatar
Linus Torvalds committed
3682
	return tic;
3683
}
Linus Torvalds's avatar
Linus Torvalds committed
3684 3685 3686 3687 3688 3689 3690 3691


/******************************************************************************
 *
 *		Log debug routines
 *
 ******************************************************************************
 */
3692
#if defined(DEBUG)
Linus Torvalds's avatar
Linus Torvalds committed
3693 3694 3695 3696 3697 3698
/*
 * Make sure that the destination ptr is within the valid data region of
 * one of the iclogs.  This uses backup pointers stored in a different
 * part of the log in case we trash the log structure.
 */
void
3699
xlog_verify_dest_ptr(
3700
	struct xlog	*log,
3701
	void		*ptr)
Linus Torvalds's avatar
Linus Torvalds committed
3702 3703 3704 3705
{
	int i;
	int good_ptr = 0;

3706 3707 3708
	for (i = 0; i < log->l_iclog_bufs; i++) {
		if (ptr >= log->l_iclog_bak[i] &&
		    ptr <= log->l_iclog_bak[i] + log->l_iclog_size)
Linus Torvalds's avatar
Linus Torvalds committed
3709 3710
			good_ptr++;
	}
3711 3712

	if (!good_ptr)
3713
		xfs_emerg(log->l_mp, "%s: invalid ptr", __func__);
3714
}
Linus Torvalds's avatar
Linus Torvalds committed
3715

3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726
/*
 * Check to make sure the grant write head didn't just over lap the tail.  If
 * the cycles are the same, we can't be overlapping.  Otherwise, make sure that
 * the cycles differ by exactly one and check the byte count.
 *
 * This check is run unlocked, so can give false positives. Rather than assert
 * on failures, use a warn-once flag and a panic tag to allow the admin to
 * determine if they want to panic the machine when such an error occurs. For
 * debug kernels this will have the same effect as using an assert but, unlinke
 * an assert, it can be turned off at runtime.
 */
3727 3728
STATIC void
xlog_verify_grant_tail(
3729
	struct xlog	*log)
3730
{
3731
	int		tail_cycle, tail_blocks;
3732
	int		cycle, space;
3733

3734
	xlog_crack_grant_head(&log->l_write_head.grant, &cycle, &space);
3735 3736
	xlog_crack_atomic_lsn(&log->l_tail_lsn, &tail_cycle, &tail_blocks);
	if (tail_cycle != cycle) {
3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749
		if (cycle - 1 != tail_cycle &&
		    !(log->l_flags & XLOG_TAIL_WARN)) {
			xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
				"%s: cycle - 1 != tail_cycle", __func__);
			log->l_flags |= XLOG_TAIL_WARN;
		}

		if (space > BBTOB(tail_blocks) &&
		    !(log->l_flags & XLOG_TAIL_WARN)) {
			xfs_alert_tag(log->l_mp, XFS_PTAG_LOGRES,
				"%s: space > BBTOB(tail_blocks)", __func__);
			log->l_flags |= XLOG_TAIL_WARN;
		}
3750 3751 3752
	}
}

Linus Torvalds's avatar
Linus Torvalds committed
3753 3754
/* check if it will fit */
STATIC void
Mark Tinguely's avatar
Mark Tinguely committed
3755 3756 3757 3758
xlog_verify_tail_lsn(
	struct xlog		*log,
	struct xlog_in_core	*iclog,
	xfs_lsn_t		tail_lsn)
Linus Torvalds's avatar
Linus Torvalds committed
3759 3760 3761 3762 3763 3764 3765
{
    int blocks;

    if (CYCLE_LSN(tail_lsn) == log->l_prev_cycle) {
	blocks =
	    log->l_logBBsize - (log->l_prev_block - BLOCK_LSN(tail_lsn));
	if (blocks < BTOBB(iclog->ic_offset)+BTOBB(log->l_iclog_hsize))
3766
		xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
Linus Torvalds's avatar
Linus Torvalds committed
3767 3768 3769 3770
    } else {
	ASSERT(CYCLE_LSN(tail_lsn)+1 == log->l_prev_cycle);

	if (BLOCK_LSN(tail_lsn) == log->l_prev_block)
3771
		xfs_emerg(log->l_mp, "%s: tail wrapped", __func__);
Linus Torvalds's avatar
Linus Torvalds committed
3772 3773 3774

	blocks = BLOCK_LSN(tail_lsn) - log->l_prev_block;
	if (blocks < BTOBB(iclog->ic_offset) + 1)
3775
		xfs_emerg(log->l_mp, "%s: ran out of log space", __func__);
Linus Torvalds's avatar
Linus Torvalds committed
3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794
    }
}	/* xlog_verify_tail_lsn */

/*
 * Perform a number of checks on the iclog before writing to disk.
 *
 * 1. Make sure the iclogs are still circular
 * 2. Make sure we have a good magic number
 * 3. Make sure we don't have magic numbers in the data
 * 4. Check fields of each log operation header for:
 *	A. Valid client identifier
 *	B. tid ptr value falls in valid ptr space (user space code)
 *	C. Length in log record header is correct according to the
 *		individual operation headers within record.
 * 5. When a bwrite will occur within 5 blocks of the front of the physical
 *	log, check the preceding blocks of the physical log to make sure all
 *	the cycle numbers agree with the current cycle number.
 */
STATIC void
Mark Tinguely's avatar
Mark Tinguely committed
3795 3796 3797 3798
xlog_verify_iclog(
	struct xlog		*log,
	struct xlog_in_core	*iclog,
	int			count,
3799
	bool                    syncing)
Linus Torvalds's avatar
Linus Torvalds committed
3800 3801 3802 3803
{
	xlog_op_header_t	*ophead;
	xlog_in_core_t		*icptr;
	xlog_in_core_2_t	*xhdr;
3804
	void			*base_ptr, *ptr, *p;
3805
	ptrdiff_t		field_offset;
Linus Torvalds's avatar
Linus Torvalds committed
3806 3807 3808 3809 3810
	__uint8_t		clientid;
	int			len, i, j, k, op_len;
	int			idx;

	/* check validity of iclog pointers */
Eric Sandeen's avatar
Eric Sandeen committed
3811
	spin_lock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3812
	icptr = log->l_iclog;
3813 3814 3815
	for (i = 0; i < log->l_iclog_bufs; i++, icptr = icptr->ic_next)
		ASSERT(icptr);

Linus Torvalds's avatar
Linus Torvalds committed
3816
	if (icptr != log->l_iclog)
3817
		xfs_emerg(log->l_mp, "%s: corrupt iclog ring", __func__);
Eric Sandeen's avatar
Eric Sandeen committed
3818
	spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3819 3820

	/* check log magic numbers */
3821
	if (iclog->ic_header.h_magicno != cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
3822
		xfs_emerg(log->l_mp, "%s: invalid magic num", __func__);
Linus Torvalds's avatar
Linus Torvalds committed
3823

3824 3825 3826
	base_ptr = ptr = &iclog->ic_header;
	p = &iclog->ic_header;
	for (ptr += BBSIZE; ptr < base_ptr + count; ptr += BBSIZE) {
3827
		if (*(__be32 *)ptr == cpu_to_be32(XLOG_HEADER_MAGIC_NUM))
3828 3829
			xfs_emerg(log->l_mp, "%s: unexpected magic num",
				__func__);
Linus Torvalds's avatar
Linus Torvalds committed
3830 3831 3832
	}

	/* check fields */
3833
	len = be32_to_cpu(iclog->ic_header.h_num_logops);
3834 3835
	base_ptr = ptr = iclog->ic_datap;
	ophead = ptr;
3836
	xhdr = iclog->ic_data;
Linus Torvalds's avatar
Linus Torvalds committed
3837
	for (i = 0; i < len; i++) {
3838
		ophead = ptr;
Linus Torvalds's avatar
Linus Torvalds committed
3839 3840

		/* clientid is only 1 byte */
3841 3842
		p = &ophead->oh_clientid;
		field_offset = p - base_ptr;
3843
		if (!syncing || (field_offset & 0x1ff)) {
Linus Torvalds's avatar
Linus Torvalds committed
3844 3845
			clientid = ophead->oh_clientid;
		} else {
3846
			idx = BTOBBT((char *)&ophead->oh_clientid - iclog->ic_datap);
Linus Torvalds's avatar
Linus Torvalds committed
3847 3848 3849
			if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
				j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
				k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3850 3851
				clientid = xlog_get_client_id(
					xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds's avatar
Linus Torvalds committed
3852
			} else {
3853 3854
				clientid = xlog_get_client_id(
					iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds's avatar
Linus Torvalds committed
3855 3856 3857
			}
		}
		if (clientid != XFS_TRANSACTION && clientid != XFS_LOG)
3858 3859 3860 3861
			xfs_warn(log->l_mp,
				"%s: invalid clientid %d op 0x%p offset 0x%lx",
				__func__, clientid, ophead,
				(unsigned long)field_offset);
Linus Torvalds's avatar
Linus Torvalds committed
3862 3863

		/* check length */
3864 3865
		p = &ophead->oh_len;
		field_offset = p - base_ptr;
3866
		if (!syncing || (field_offset & 0x1ff)) {
3867
			op_len = be32_to_cpu(ophead->oh_len);
Linus Torvalds's avatar
Linus Torvalds committed
3868
		} else {
3869 3870
			idx = BTOBBT((uintptr_t)&ophead->oh_len -
				    (uintptr_t)iclog->ic_datap);
Linus Torvalds's avatar
Linus Torvalds committed
3871 3872 3873
			if (idx >= (XLOG_HEADER_CYCLE_SIZE / BBSIZE)) {
				j = idx / (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
				k = idx % (XLOG_HEADER_CYCLE_SIZE / BBSIZE);
3874
				op_len = be32_to_cpu(xhdr[j].hic_xheader.xh_cycle_data[k]);
Linus Torvalds's avatar
Linus Torvalds committed
3875
			} else {
3876
				op_len = be32_to_cpu(iclog->ic_header.h_cycle_data[idx]);
Linus Torvalds's avatar
Linus Torvalds committed
3877 3878 3879 3880 3881
			}
		}
		ptr += sizeof(xlog_op_header_t) + op_len;
	}
}	/* xlog_verify_iclog */
3882
#endif
Linus Torvalds's avatar
Linus Torvalds committed
3883 3884

/*
Eric Sandeen's avatar
Eric Sandeen committed
3885
 * Mark all iclogs IOERROR. l_icloglock is held by the caller.
Linus Torvalds's avatar
Linus Torvalds committed
3886 3887 3888
 */
STATIC int
xlog_state_ioerror(
Mark Tinguely's avatar
Mark Tinguely committed
3889
	struct xlog	*log)
Linus Torvalds's avatar
Linus Torvalds committed
3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903
{
	xlog_in_core_t	*iclog, *ic;

	iclog = log->l_iclog;
	if (! (iclog->ic_state & XLOG_STATE_IOERROR)) {
		/*
		 * Mark all the incore logs IOERROR.
		 * From now on, no log flushes will result.
		 */
		ic = iclog;
		do {
			ic->ic_state = XLOG_STATE_IOERROR;
			ic = ic->ic_next;
		} while (ic != iclog);
3904
		return 0;
Linus Torvalds's avatar
Linus Torvalds committed
3905 3906 3907 3908
	}
	/*
	 * Return non-zero, if state transition has already happened.
	 */
3909
	return 1;
Linus Torvalds's avatar
Linus Torvalds committed
3910 3911 3912 3913 3914 3915
}

/*
 * This is called from xfs_force_shutdown, when we're forcibly
 * shutting down the filesystem, typically because of an IO error.
 * Our main objectives here are to make sure that:
3916 3917 3918
 *	a. if !logerror, flush the logs to disk. Anything modified
 *	   after this is ignored.
 *	b. the filesystem gets marked 'SHUTDOWN' for all interested
Linus Torvalds's avatar
Linus Torvalds committed
3919
 *	   parties to find out, 'atomically'.
3920
 *	c. those who're sleeping on log reservations, pinned objects and
Linus Torvalds's avatar
Linus Torvalds committed
3921
 *	    other resources get woken up, and be told the bad news.
3922
 *	d. nothing new gets queued up after (b) and (c) are done.
3923
 *
3924 3925 3926
 * Note: for the !logerror case we need to flush the regions held in memory out
 * to disk first. This needs to be done before the log is marked as shutdown,
 * otherwise the iclog writes will fail.
Linus Torvalds's avatar
Linus Torvalds committed
3927 3928 3929 3930 3931 3932
 */
int
xfs_log_force_umount(
	struct xfs_mount	*mp,
	int			logerror)
{
Mark Tinguely's avatar
Mark Tinguely committed
3933
	struct xlog	*log;
Linus Torvalds's avatar
Linus Torvalds committed
3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944
	int		retval;

	log = mp->m_log;

	/*
	 * If this happens during log recovery, don't worry about
	 * locking; the log isn't open for business yet.
	 */
	if (!log ||
	    log->l_flags & XLOG_ACTIVE_RECOVERY) {
		mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
3945
		if (mp->m_sb_bp)
3946
			mp->m_sb_bp->b_flags |= XBF_DONE;
3947
		return 0;
Linus Torvalds's avatar
Linus Torvalds committed
3948 3949 3950 3951 3952 3953 3954 3955
	}

	/*
	 * Somebody could've already done the hard work for us.
	 * No need to get locks for this.
	 */
	if (logerror && log->l_iclog->ic_state & XLOG_STATE_IOERROR) {
		ASSERT(XLOG_FORCED_SHUTDOWN(log));
3956
		return 1;
Linus Torvalds's avatar
Linus Torvalds committed
3957
	}
3958 3959

	/*
3960 3961 3962 3963 3964
	 * Flush all the completed transactions to disk before marking the log
	 * being shut down. We need to do it in this order to ensure that
	 * completed operations are safely on disk before we shut down, and that
	 * we don't have to issue any buffer IO after the shutdown flags are set
	 * to guarantee this.
3965
	 */
3966
	if (!logerror)
3967
		_xfs_log_force(mp, XFS_LOG_SYNC, NULL);
3968

Linus Torvalds's avatar
Linus Torvalds committed
3969
	/*
3970 3971
	 * mark the filesystem and the as in a shutdown state and wake
	 * everybody up to tell them the bad news.
Linus Torvalds's avatar
Linus Torvalds committed
3972
	 */
Eric Sandeen's avatar
Eric Sandeen committed
3973
	spin_lock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3974
	mp->m_flags |= XFS_MOUNT_FS_SHUTDOWN;
3975
	if (mp->m_sb_bp)
3976
		mp->m_sb_bp->b_flags |= XBF_DONE;
3977

Linus Torvalds's avatar
Linus Torvalds committed
3978
	/*
3979 3980
	 * Mark the log and the iclogs with IO error flags to prevent any
	 * further log IO from being issued or completed.
Linus Torvalds's avatar
Linus Torvalds committed
3981 3982
	 */
	log->l_flags |= XLOG_IO_ERROR;
3983
	retval = xlog_state_ioerror(log);
Eric Sandeen's avatar
Eric Sandeen committed
3984
	spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
3985 3986

	/*
3987 3988 3989 3990
	 * We don't want anybody waiting for log reservations after this. That
	 * means we have to wake up everybody queued up on reserveq as well as
	 * writeq.  In addition, we make sure in xlog_{re}grant_log_space that
	 * we don't enqueue anything once the SHUTDOWN flag is set, and this
3991
	 * action is protected by the grant locks.
Linus Torvalds's avatar
Linus Torvalds committed
3992
	 */
3993 3994
	xlog_grant_head_wake_all(&log->l_reserve_head);
	xlog_grant_head_wake_all(&log->l_write_head);
Linus Torvalds's avatar
Linus Torvalds committed
3995 3996

	/*
3997 3998 3999 4000
	 * Wake up everybody waiting on xfs_log_force. Wake the CIL push first
	 * as if the log writes were completed. The abort handling in the log
	 * item committed callback functions will do this again under lock to
	 * avoid races.
Linus Torvalds's avatar
Linus Torvalds committed
4001
	 */
4002
	spin_lock(&log->l_cilp->xc_push_lock);
4003
	wake_up_all(&log->l_cilp->xc_commit_wait);
4004
	spin_unlock(&log->l_cilp->xc_push_lock);
Linus Torvalds's avatar
Linus Torvalds committed
4005 4006 4007 4008 4009 4010
	xlog_state_do_callback(log, XFS_LI_ABORTED, NULL);

#ifdef XFSERRORDEBUG
	{
		xlog_in_core_t	*iclog;

Eric Sandeen's avatar
Eric Sandeen committed
4011
		spin_lock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
4012 4013 4014 4015 4016
		iclog = log->l_iclog;
		do {
			ASSERT(iclog->ic_callback == 0);
			iclog = iclog->ic_next;
		} while (iclog != log->l_iclog);
Eric Sandeen's avatar
Eric Sandeen committed
4017
		spin_unlock(&log->l_icloglock);
Linus Torvalds's avatar
Linus Torvalds committed
4018 4019 4020
	}
#endif
	/* return non-zero if log IOERROR transition had already happened */
4021
	return retval;
Linus Torvalds's avatar
Linus Torvalds committed
4022 4023
}

4024
STATIC int
Mark Tinguely's avatar
Mark Tinguely committed
4025 4026
xlog_iclogs_empty(
	struct xlog	*log)
Linus Torvalds's avatar
Linus Torvalds committed
4027 4028 4029 4030 4031 4032 4033 4034 4035
{
	xlog_in_core_t	*iclog;

	iclog = log->l_iclog;
	do {
		/* endianness does not matter here, zero is zero in
		 * any language.
		 */
		if (iclog->ic_header.h_num_logops)
4036
			return 0;
Linus Torvalds's avatar
Linus Torvalds committed
4037 4038
		iclog = iclog->ic_next;
	} while (iclog != log->l_iclog);
4039
	return 1;
Linus Torvalds's avatar
Linus Torvalds committed
4040
}
4041

4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083
/*
 * Verify that an LSN stamped into a piece of metadata is valid. This is
 * intended for use in read verifiers on v5 superblocks.
 */
bool
xfs_log_check_lsn(
	struct xfs_mount	*mp,
	xfs_lsn_t		lsn)
{
	struct xlog		*log = mp->m_log;
	bool			valid;

	/*
	 * norecovery mode skips mount-time log processing and unconditionally
	 * resets the in-core LSN. We can't validate in this mode, but
	 * modifications are not allowed anyways so just return true.
	 */
	if (mp->m_flags & XFS_MOUNT_NORECOVERY)
		return true;

	/*
	 * Some metadata LSNs are initialized to NULL (e.g., the agfl). This is
	 * handled by recovery and thus safe to ignore here.
	 */
	if (lsn == NULLCOMMITLSN)
		return true;

	valid = xlog_valid_lsn(mp->m_log, lsn);

	/* warn the user about what's gone wrong before verifier failure */
	if (!valid) {
		spin_lock(&log->l_icloglock);
		xfs_warn(mp,
"Corruption warning: Metadata has LSN (%d:%d) ahead of current LSN (%d:%d). "
"Please unmount and run xfs_repair (>= v4.3) to resolve.",
			 CYCLE_LSN(lsn), BLOCK_LSN(lsn),
			 log->l_curr_cycle, log->l_curr_block);
		spin_unlock(&log->l_icloglock);
	}

	return valid;
}