xfs_buf_item.c 35.3 KB
Newer Older
L
Linus Torvalds 已提交
1
/*
2 3
 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
 * All Rights Reserved.
L
Linus Torvalds 已提交
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
L
Linus Torvalds 已提交
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.
L
Linus Torvalds 已提交
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
L
Linus Torvalds 已提交
17 18
 */
#include "xfs.h"
19
#include "xfs_fs.h"
20
#include "xfs_format.h"
21 22
#include "xfs_log_format.h"
#include "xfs_trans_resv.h"
23
#include "xfs_bit.h"
L
Linus Torvalds 已提交
24 25
#include "xfs_sb.h"
#include "xfs_mount.h"
26
#include "xfs_trans.h"
27
#include "xfs_buf_item.h"
L
Linus Torvalds 已提交
28 29
#include "xfs_trans_priv.h"
#include "xfs_error.h"
C
Christoph Hellwig 已提交
30
#include "xfs_trace.h"
31
#include "xfs_log.h"
32
#include "xfs_inode.h"
L
Linus Torvalds 已提交
33 34 35 36


kmem_zone_t	*xfs_buf_item_zone;

37 38 39 40 41
static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip)
{
	return container_of(lip, struct xfs_buf_log_item, bli_item);
}

42
STATIC void	xfs_buf_do_callbacks(struct xfs_buf *bp);
L
Linus Torvalds 已提交
43

44 45 46 47 48 49 50 51
static inline int
xfs_buf_log_format_size(
	struct xfs_buf_log_format *blfp)
{
	return offsetof(struct xfs_buf_log_format, blf_data_map) +
			(blfp->blf_map_size * sizeof(blfp->blf_data_map[0]));
}

L
Linus Torvalds 已提交
52 53 54 55 56 57 58 59 60 61
/*
 * This returns the number of log iovecs needed to log the
 * given buf log item.
 *
 * It calculates this as 1 iovec for the buf log format structure
 * and 1 for each stretch of non-contiguous chunks to be logged.
 * Contiguous chunks are logged in a single iovec.
 *
 * If the XFS_BLI_STALE flag has been set, then log nothing.
 */
62
STATIC void
63
xfs_buf_item_size_segment(
64 65 66 67
	struct xfs_buf_log_item		*bip,
	struct xfs_buf_log_format	*blfp,
	int				*nvecs,
	int				*nbytes)
L
Linus Torvalds 已提交
68
{
69 70 71
	struct xfs_buf			*bp = bip->bli_buf;
	int				next_bit;
	int				last_bit;
L
Linus Torvalds 已提交
72

73 74
	last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
	if (last_bit == -1)
75
		return;
76 77 78 79 80

	/*
	 * initial count for a dirty buffer is 2 vectors - the format structure
	 * and the first dirty region.
	 */
81 82
	*nvecs += 2;
	*nbytes += xfs_buf_log_format_size(blfp) + XFS_BLF_CHUNK;
L
Linus Torvalds 已提交
83 84 85 86 87 88 89 90

	while (last_bit != -1) {
		/*
		 * This takes the bit number to start looking from and
		 * returns the next set bit from there.  It returns -1
		 * if there are no more bits set or the start bit is
		 * beyond the end of the bitmap.
		 */
91 92
		next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
					last_bit + 1);
L
Linus Torvalds 已提交
93 94 95 96 97 98
		/*
		 * If we run out of bits, leave the loop,
		 * else if we find a new set of bits bump the number of vecs,
		 * else keep scanning the current set of bits.
		 */
		if (next_bit == -1) {
99
			break;
L
Linus Torvalds 已提交
100 101
		} else if (next_bit != last_bit + 1) {
			last_bit = next_bit;
102
			(*nvecs)++;
103 104 105
		} else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
			   (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
			    XFS_BLF_CHUNK)) {
L
Linus Torvalds 已提交
106
			last_bit = next_bit;
107
			(*nvecs)++;
L
Linus Torvalds 已提交
108 109 110
		} else {
			last_bit++;
		}
111
		*nbytes += XFS_BLF_CHUNK;
L
Linus Torvalds 已提交
112 113 114 115
	}
}

/*
116 117 118 119 120 121 122 123 124 125 126 127 128 129 130
 * This returns the number of log iovecs needed to log the given buf log item.
 *
 * It calculates this as 1 iovec for the buf log format structure and 1 for each
 * stretch of non-contiguous chunks to be logged.  Contiguous chunks are logged
 * in a single iovec.
 *
 * Discontiguous buffers need a format structure per region that that is being
 * logged. This makes the changes in the buffer appear to log recovery as though
 * they came from separate buffers, just like would occur if multiple buffers
 * were used instead of a single discontiguous buffer. This enables
 * discontiguous buffers to be in-memory constructs, completely transparent to
 * what ends up on disk.
 *
 * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log
 * format structures.
L
Linus Torvalds 已提交
131
 */
132
STATIC void
133
xfs_buf_item_size(
134 135 136
	struct xfs_log_item	*lip,
	int			*nvecs,
	int			*nbytes)
L
Linus Torvalds 已提交
137
{
138
	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
139 140 141 142 143 144 145 146 147 148
	int			i;

	ASSERT(atomic_read(&bip->bli_refcount) > 0);
	if (bip->bli_flags & XFS_BLI_STALE) {
		/*
		 * The buffer is stale, so all we need to log
		 * is the buf log format structure with the
		 * cancel flag in it.
		 */
		trace_xfs_buf_item_size_stale(bip);
149
		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
150 151 152 153 154
		*nvecs += bip->bli_format_count;
		for (i = 0; i < bip->bli_format_count; i++) {
			*nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]);
		}
		return;
155 156 157 158
	}

	ASSERT(bip->bli_flags & XFS_BLI_LOGGED);

159 160 161 162 163 164 165
	if (bip->bli_flags & XFS_BLI_ORDERED) {
		/*
		 * The buffer has been logged just to order it.
		 * It is not being included in the transaction
		 * commit, so no vectors are used at all.
		 */
		trace_xfs_buf_item_size_ordered(bip);
166 167
		*nvecs = XFS_LOG_VEC_ORDERED;
		return;
168 169
	}

170 171 172 173 174 175 176 177 178 179
	/*
	 * the vector count is based on the number of buffer vectors we have
	 * dirty bits in. This will only be greater than one when we have a
	 * compound buffer with more than one segment dirty. Hence for compound
	 * buffers we need to track which segment the dirty bits correspond to,
	 * and when we move from one segment to the next increment the vector
	 * count for the extra buf log format structure that will need to be
	 * written.
	 */
	for (i = 0; i < bip->bli_format_count; i++) {
180 181
		xfs_buf_item_size_segment(bip, &bip->bli_formats[i],
					  nvecs, nbytes);
182 183 184 185
	}
	trace_xfs_buf_item_size(bip);
}

C
Christoph Hellwig 已提交
186
static inline void
187
xfs_buf_item_copy_iovec(
188
	struct xfs_log_vec	*lv,
C
Christoph Hellwig 已提交
189
	struct xfs_log_iovec	**vecp,
190 191 192 193 194 195
	struct xfs_buf		*bp,
	uint			offset,
	int			first_bit,
	uint			nbits)
{
	offset += first_bit * XFS_BLF_CHUNK;
196
	xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK,
C
Christoph Hellwig 已提交
197 198
			xfs_buf_offset(bp, offset),
			nbits * XFS_BLF_CHUNK);
199 200 201 202 203 204 205 206 207 208 209 210 211 212
}

static inline bool
xfs_buf_item_straddle(
	struct xfs_buf		*bp,
	uint			offset,
	int			next_bit,
	int			last_bit)
{
	return xfs_buf_offset(bp, offset + (next_bit << XFS_BLF_SHIFT)) !=
		(xfs_buf_offset(bp, offset + (last_bit << XFS_BLF_SHIFT)) +
		 XFS_BLF_CHUNK);
}

C
Christoph Hellwig 已提交
213
static void
214 215
xfs_buf_item_format_segment(
	struct xfs_buf_log_item	*bip,
216
	struct xfs_log_vec	*lv,
C
Christoph Hellwig 已提交
217
	struct xfs_log_iovec	**vecp,
218 219 220
	uint			offset,
	struct xfs_buf_log_format *blfp)
{
221 222 223 224 225 226
	struct xfs_buf		*bp = bip->bli_buf;
	uint			base_size;
	int			first_bit;
	int			last_bit;
	int			next_bit;
	uint			nbits;
L
Linus Torvalds 已提交
227

228
	/* copy the flags across from the base format item */
229
	blfp->blf_flags = bip->__bli_format.blf_flags;
L
Linus Torvalds 已提交
230 231

	/*
232 233 234
	 * Base size is the actual size of the ondisk structure - it reflects
	 * the actual size of the dirty bitmap rather than the size of the in
	 * memory structure.
L
Linus Torvalds 已提交
235
	 */
236
	base_size = xfs_buf_log_format_size(blfp);
237 238 239 240 241 242 243

	first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0);
	if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) {
		/*
		 * If the map is not be dirty in the transaction, mark
		 * the size as zero and do not advance the vector pointer.
		 */
244
		return;
245 246
	}

247 248
	blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size);
	blfp->blf_size = 1;
L
Linus Torvalds 已提交
249 250 251 252 253 254 255

	if (bip->bli_flags & XFS_BLI_STALE) {
		/*
		 * The buffer is stale, so all we need to log
		 * is the buf log format structure with the
		 * cancel flag in it.
		 */
C
Christoph Hellwig 已提交
256
		trace_xfs_buf_item_format_stale(bip);
257
		ASSERT(blfp->blf_flags & XFS_BLF_CANCEL);
258
		return;
L
Linus Torvalds 已提交
259 260
	}

261

L
Linus Torvalds 已提交
262 263 264 265 266 267 268 269 270 271 272 273
	/*
	 * Fill in an iovec for each set of contiguous chunks.
	 */
	last_bit = first_bit;
	nbits = 1;
	for (;;) {
		/*
		 * This takes the bit number to start looking from and
		 * returns the next set bit from there.  It returns -1
		 * if there are no more bits set or the start bit is
		 * beyond the end of the bitmap.
		 */
274 275
		next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size,
					(uint)last_bit + 1);
L
Linus Torvalds 已提交
276
		/*
277 278 279 280 281
		 * If we run out of bits fill in the last iovec and get out of
		 * the loop.  Else if we start a new set of bits then fill in
		 * the iovec for the series we were looking at and start
		 * counting the bits in the new one.  Else we're still in the
		 * same set of bits so just keep counting and scanning.
L
Linus Torvalds 已提交
282 283
		 */
		if (next_bit == -1) {
284
			xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
285
						first_bit, nbits);
286
			blfp->blf_size++;
L
Linus Torvalds 已提交
287
			break;
288 289
		} else if (next_bit != last_bit + 1 ||
		           xfs_buf_item_straddle(bp, offset, next_bit, last_bit)) {
290
			xfs_buf_item_copy_iovec(lv, vecp, bp, offset,
C
Christoph Hellwig 已提交
291
						first_bit, nbits);
292
			blfp->blf_size++;
L
Linus Torvalds 已提交
293 294 295 296 297 298 299 300
			first_bit = next_bit;
			last_bit = next_bit;
			nbits = 1;
		} else {
			last_bit++;
			nbits++;
		}
	}
301 302 303 304 305 306 307 308 309 310 311
}

/*
 * This is called to fill in the vector of log iovecs for the
 * given log buf item.  It fills the first entry with a buf log
 * format structure, and the rest point to contiguous chunks
 * within the buffer.
 */
STATIC void
xfs_buf_item_format(
	struct xfs_log_item	*lip,
312
	struct xfs_log_vec	*lv)
313 314 315
{
	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
	struct xfs_buf		*bp = bip->bli_buf;
316
	struct xfs_log_iovec	*vecp = NULL;
317 318 319 320 321 322
	uint			offset = 0;
	int			i;

	ASSERT(atomic_read(&bip->bli_refcount) > 0);
	ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
	       (bip->bli_flags & XFS_BLI_STALE));
323 324 325
	ASSERT((bip->bli_flags & XFS_BLI_STALE) ||
	       (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF
	        && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF));
326 327
	ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED) ||
	       (bip->bli_flags & XFS_BLI_STALE));
328

329 330 331

	/*
	 * If it is an inode buffer, transfer the in-memory state to the
D
Dave Chinner 已提交
332 333 334
	 * format flags and clear the in-memory state.
	 *
	 * For buffer based inode allocation, we do not transfer
335 336 337
	 * this state if the inode buffer allocation has not yet been committed
	 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
	 * correct replay of the inode allocation.
D
Dave Chinner 已提交
338 339 340 341 342
	 *
	 * For icreate item based inode allocation, the buffers aren't written
	 * to the journal during allocation, and hence we should always tag the
	 * buffer as an inode buffer so that the correct unlinked list replay
	 * occurs during recovery.
343 344
	 */
	if (bip->bli_flags & XFS_BLI_INODE_BUF) {
D
Dave Chinner 已提交
345 346
		if (xfs_sb_version_hascrc(&lip->li_mountp->m_sb) ||
		    !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
347
		      xfs_log_item_in_current_chkpt(lip)))
348
			bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF;
349 350 351 352
		bip->bli_flags &= ~XFS_BLI_INODE_BUF;
	}

	for (i = 0; i < bip->bli_format_count; i++) {
353
		xfs_buf_item_format_segment(bip, lv, &vecp, offset,
C
Christoph Hellwig 已提交
354
					    &bip->bli_formats[i]);
355
		offset += BBTOB(bp->b_maps[i].bm_len);
356
	}
L
Linus Torvalds 已提交
357 358 359 360

	/*
	 * Check to make sure everything is consistent.
	 */
C
Christoph Hellwig 已提交
361
	trace_xfs_buf_item_format(bip);
L
Linus Torvalds 已提交
362 363 364
}

/*
365
 * This is called to pin the buffer associated with the buf log item in memory
C
Christoph Hellwig 已提交
366
 * so it cannot be written out.
367 368 369 370 371
 *
 * We also always take a reference to the buffer log item here so that the bli
 * is held while the item is pinned in memory. This means that we can
 * unconditionally drop the reference count a transaction holds when the
 * transaction is completed.
L
Linus Torvalds 已提交
372
 */
373
STATIC void
L
Linus Torvalds 已提交
374
xfs_buf_item_pin(
375
	struct xfs_log_item	*lip)
L
Linus Torvalds 已提交
376
{
377
	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
L
Linus Torvalds 已提交
378 379 380

	ASSERT(atomic_read(&bip->bli_refcount) > 0);
	ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
381
	       (bip->bli_flags & XFS_BLI_ORDERED) ||
L
Linus Torvalds 已提交
382
	       (bip->bli_flags & XFS_BLI_STALE));
383

C
Christoph Hellwig 已提交
384
	trace_xfs_buf_item_pin(bip);
C
Christoph Hellwig 已提交
385 386 387

	atomic_inc(&bip->bli_refcount);
	atomic_inc(&bip->bli_buf->b_pin_count);
L
Linus Torvalds 已提交
388 389 390 391 392 393 394 395 396
}

/*
 * This is called to unpin the buffer associated with the buf log
 * item which was previously pinned with a call to xfs_buf_item_pin().
 *
 * Also drop the reference to the buf item for the current transaction.
 * If the XFS_BLI_STALE flag is set and we are the last reference,
 * then free up the buf log item and unlock the buffer.
397 398 399 400 401
 *
 * If the remove flag is set we are called from uncommit in the
 * forced-shutdown path.  If that is true and the reference count on
 * the log item is going to drop to zero we need to free the item's
 * descriptor in the transaction.
L
Linus Torvalds 已提交
402
 */
403
STATIC void
L
Linus Torvalds 已提交
404
xfs_buf_item_unpin(
405
	struct xfs_log_item	*lip,
406
	int			remove)
L
Linus Torvalds 已提交
407
{
408
	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
409 410 411 412
	xfs_buf_t		*bp = bip->bli_buf;
	struct xfs_ail		*ailp = lip->li_ailp;
	int			stale = bip->bli_flags & XFS_BLI_STALE;
	int			freed;
L
Linus Torvalds 已提交
413

C
Carlos Maiolino 已提交
414
	ASSERT(bp->b_log_item == bip);
L
Linus Torvalds 已提交
415
	ASSERT(atomic_read(&bip->bli_refcount) > 0);
416

C
Christoph Hellwig 已提交
417
	trace_xfs_buf_item_unpin(bip);
L
Linus Torvalds 已提交
418 419

	freed = atomic_dec_and_test(&bip->bli_refcount);
C
Christoph Hellwig 已提交
420 421 422

	if (atomic_dec_and_test(&bp->b_pin_count))
		wake_up_all(&bp->b_waiters);
423

L
Linus Torvalds 已提交
424 425
	if (freed && stale) {
		ASSERT(bip->bli_flags & XFS_BLI_STALE);
426
		ASSERT(xfs_buf_islocked(bp));
427
		ASSERT(bp->b_flags & XBF_STALE);
428
		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
429

C
Christoph Hellwig 已提交
430 431
		trace_xfs_buf_item_unpin_stale(bip);

432 433
		if (remove) {
			/*
434 435 436 437 438
			 * If we are in a transaction context, we have to
			 * remove the log item from the transaction as we are
			 * about to release our reference to the buffer.  If we
			 * don't, the unlock that occurs later in
			 * xfs_trans_uncommit() will try to reference the
439 440
			 * buffer which we no longer have a hold on.
			 */
441 442
			if (lip->li_desc)
				xfs_trans_del_item(lip);
443 444 445 446 447

			/*
			 * Since the transaction no longer refers to the buffer,
			 * the buffer should no longer refer to the transaction.
			 */
448
			bp->b_transp = NULL;
449 450
		}

L
Linus Torvalds 已提交
451 452
		/*
		 * If we get called here because of an IO error, we may
453
		 * or may not have the item on the AIL. xfs_trans_ail_delete()
L
Linus Torvalds 已提交
454
		 * will take care of that situation.
455
		 * xfs_trans_ail_delete() drops the AIL lock.
L
Linus Torvalds 已提交
456 457
		 */
		if (bip->bli_flags & XFS_BLI_STALE_INODE) {
458
			xfs_buf_do_callbacks(bp);
C
Carlos Maiolino 已提交
459
			bp->b_log_item = NULL;
460
			list_del_init(&bp->b_li_list);
461
			bp->b_iodone = NULL;
L
Linus Torvalds 已提交
462
		} else {
463
			spin_lock(&ailp->ail_lock);
464
			xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR);
L
Linus Torvalds 已提交
465
			xfs_buf_item_relse(bp);
C
Carlos Maiolino 已提交
466
			ASSERT(bp->b_log_item == NULL);
L
Linus Torvalds 已提交
467 468
		}
		xfs_buf_relse(bp);
469
	} else if (freed && remove) {
470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485
		/*
		 * There are currently two references to the buffer - the active
		 * LRU reference and the buf log item. What we are about to do
		 * here - simulate a failed IO completion - requires 3
		 * references.
		 *
		 * The LRU reference is removed by the xfs_buf_stale() call. The
		 * buf item reference is removed by the xfs_buf_iodone()
		 * callback that is run by xfs_buf_do_callbacks() during ioend
		 * processing (via the bp->b_iodone callback), and then finally
		 * the ioend processing will drop the IO reference if the buffer
		 * is marked XBF_ASYNC.
		 *
		 * Hence we need to take an additional reference here so that IO
		 * completion processing doesn't free the buffer prematurely.
		 */
486
		xfs_buf_lock(bp);
487 488
		xfs_buf_hold(bp);
		bp->b_flags |= XBF_ASYNC;
D
Dave Chinner 已提交
489
		xfs_buf_ioerror(bp, -EIO);
490
		bp->b_flags &= ~XBF_DONE;
491
		xfs_buf_stale(bp);
492
		xfs_buf_ioend(bp);
L
Linus Torvalds 已提交
493 494 495
	}
}

496 497 498 499 500 501
/*
 * Buffer IO error rate limiting. Limit it to no more than 10 messages per 30
 * seconds so as to not spam logs too much on repeated detection of the same
 * buffer being bad..
 */

502
static DEFINE_RATELIMIT_STATE(xfs_buf_write_fail_rl_state, 30 * HZ, 10);
503

504
STATIC uint
505 506 507
xfs_buf_item_push(
	struct xfs_log_item	*lip,
	struct list_head	*buffer_list)
L
Linus Torvalds 已提交
508
{
509 510
	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
	struct xfs_buf		*bp = bip->bli_buf;
511
	uint			rval = XFS_ITEM_SUCCESS;
L
Linus Torvalds 已提交
512

513
	if (xfs_buf_ispinned(bp))
L
Linus Torvalds 已提交
514
		return XFS_ITEM_PINNED;
515 516 517 518 519 520 521 522 523 524
	if (!xfs_buf_trylock(bp)) {
		/*
		 * If we have just raced with a buffer being pinned and it has
		 * been marked stale, we could end up stalling until someone else
		 * issues a log force to unpin the stale buffer. Check for the
		 * race condition here so xfsaild recognizes the buffer is pinned
		 * and queues a log force to move it along.
		 */
		if (xfs_buf_ispinned(bp))
			return XFS_ITEM_PINNED;
L
Linus Torvalds 已提交
525
		return XFS_ITEM_LOCKED;
526
	}
L
Linus Torvalds 已提交
527 528

	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
529 530 531

	trace_xfs_buf_item_push(bip);

532 533
	/* has a previous flush failed due to IO errors? */
	if ((bp->b_flags & XBF_WRITE_FAIL) &&
534
	    ___ratelimit(&xfs_buf_write_fail_rl_state, "XFS: Failing async write")) {
535
		xfs_warn(bp->b_target->bt_mount,
536
"Failing async write on buffer block 0x%llx. Retrying async write.",
537 538 539
			 (long long)bp->b_bn);
	}

540 541 542 543
	if (!xfs_buf_delwri_queue(bp, buffer_list))
		rval = XFS_ITEM_FLUSHING;
	xfs_buf_unlock(bp);
	return rval;
L
Linus Torvalds 已提交
544 545 546
}

/*
547 548 549
 * Release the buffer associated with the buf log item.  If there is no dirty
 * logged data associated with the buffer recorded in the buf log item, then
 * free the buf log item and remove the reference to it in the buffer.
L
Linus Torvalds 已提交
550
 *
551 552
 * This call ignores the recursion count.  It is only called when the buffer
 * should REALLY be unlocked, regardless of the recursion count.
L
Linus Torvalds 已提交
553
 *
554 555 556 557 558 559 560 561 562 563
 * We unconditionally drop the transaction's reference to the log item. If the
 * item was logged, then another reference was taken when it was pinned, so we
 * can safely drop the transaction reference now.  This also allows us to avoid
 * potential races with the unpin code freeing the bli by not referencing the
 * bli after we've dropped the reference count.
 *
 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
 * if necessary but do not unlock the buffer.  This is for support of
 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
 * free the item.
L
Linus Torvalds 已提交
564
 */
565
STATIC void
L
Linus Torvalds 已提交
566
xfs_buf_item_unlock(
567
	struct xfs_log_item	*lip)
L
Linus Torvalds 已提交
568
{
569 570
	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);
	struct xfs_buf		*bp = bip->bli_buf;
571 572 573
	bool			aborted = !!(lip->li_flags & XFS_LI_ABORTED);
	bool			hold = !!(bip->bli_flags & XFS_BLI_HOLD);
	bool			dirty = !!(bip->bli_flags & XFS_BLI_DIRTY);
D
Darrick J. Wong 已提交
574
#if defined(DEBUG) || defined(XFS_WARN)
575
	bool			ordered = !!(bip->bli_flags & XFS_BLI_ORDERED);
D
Darrick J. Wong 已提交
576
#endif
L
Linus Torvalds 已提交
577

578
	/* Clear the buffer's association with this transaction. */
579
	bp->b_transp = NULL;
L
Linus Torvalds 已提交
580 581

	/*
582 583
	 * The per-transaction state has been copied above so clear it from the
	 * bli.
L
Linus Torvalds 已提交
584
	 */
585
	bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED);
586 587 588 589 590

	/*
	 * If the buf item is marked stale, then don't do anything.  We'll
	 * unlock the buffer and free the buf item when the buffer is unpinned
	 * for the last time.
L
Linus Torvalds 已提交
591
	 */
592
	if (bip->bli_flags & XFS_BLI_STALE) {
C
Christoph Hellwig 已提交
593
		trace_xfs_buf_item_unlock_stale(bip);
594
		ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
595 596
		if (!aborted) {
			atomic_dec(&bip->bli_refcount);
L
Linus Torvalds 已提交
597
			return;
598
		}
L
Linus Torvalds 已提交
599 600
	}

C
Christoph Hellwig 已提交
601
	trace_xfs_buf_item_unlock(bip);
L
Linus Torvalds 已提交
602 603

	/*
604
	 * If the buf item isn't tracking any data, free it, otherwise drop the
605 606 607 608
	 * reference we hold to it. If we are aborting the transaction, this may
	 * be the only reference to the buf item, so we free it anyway
	 * regardless of whether it is dirty or not. A dirty abort implies a
	 * shutdown, anyway.
609
	 *
610 611
	 * The bli dirty state should match whether the blf has logged segments
	 * except for ordered buffers, where only the bli should be dirty.
L
Linus Torvalds 已提交
612
	 */
613 614
	ASSERT((!ordered && dirty == xfs_buf_item_dirty_format(bip)) ||
	       (ordered && dirty && !xfs_buf_item_dirty_format(bip)));
615 616 617

	/*
	 * Clean buffers, by definition, cannot be in the AIL. However, aborted
618 619 620 621 622 623 624 625 626
	 * buffers may be in the AIL regardless of dirty state. An aborted
	 * transaction that invalidates a buffer already in the AIL may have
	 * marked it stale and cleared the dirty state, for example.
	 *
	 * Therefore if we are aborting a buffer and we've just taken the last
	 * reference away, we have to check if it is in the AIL before freeing
	 * it. We need to free it in this case, because an aborted transaction
	 * has already shut the filesystem down and this is the last chance we
	 * will have to do so.
627 628
	 */
	if (atomic_dec_and_test(&bip->bli_refcount)) {
629
		if (aborted) {
630
			ASSERT(XFS_FORCED_SHUTDOWN(lip->li_mountp));
631
			xfs_trans_ail_remove(lip, SHUTDOWN_LOG_IO_ERROR);
632
			xfs_buf_item_relse(bp);
633
		} else if (!dirty)
634
			xfs_buf_item_relse(bp);
635
	}
L
Linus Torvalds 已提交
636

637
	if (!hold)
L
Linus Torvalds 已提交
638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658
		xfs_buf_relse(bp);
}

/*
 * This is called to find out where the oldest active copy of the
 * buf log item in the on disk log resides now that the last log
 * write of it completed at the given lsn.
 * We always re-log all the dirty data in a buffer, so usually the
 * latest copy in the on disk log is the only one that matters.  For
 * those cases we simply return the given lsn.
 *
 * The one exception to this is for buffers full of newly allocated
 * inodes.  These buffers are only relogged with the XFS_BLI_INODE_BUF
 * flag set, indicating that only the di_next_unlinked fields from the
 * inodes in the buffers will be replayed during recovery.  If the
 * original newly allocated inode images have not yet been flushed
 * when the buffer is so relogged, then we need to make sure that we
 * keep the old images in the 'active' portion of the log.  We do this
 * by returning the original lsn of that transaction here rather than
 * the current one.
 */
659
STATIC xfs_lsn_t
L
Linus Torvalds 已提交
660
xfs_buf_item_committed(
661
	struct xfs_log_item	*lip,
L
Linus Torvalds 已提交
662 663
	xfs_lsn_t		lsn)
{
664 665
	struct xfs_buf_log_item	*bip = BUF_ITEM(lip);

C
Christoph Hellwig 已提交
666 667
	trace_xfs_buf_item_committed(bip);

668 669 670
	if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
		return lip->li_lsn;
	return lsn;
L
Linus Torvalds 已提交
671 672
}

673
STATIC void
674 675 676
xfs_buf_item_committing(
	struct xfs_log_item	*lip,
	xfs_lsn_t		commit_lsn)
L
Linus Torvalds 已提交
677 678 679 680 681 682
{
}

/*
 * This is the ops vector shared by all buf log items.
 */
C
Christoph Hellwig 已提交
683
static const struct xfs_item_ops xfs_buf_item_ops = {
684 685 686 687 688 689 690 691
	.iop_size	= xfs_buf_item_size,
	.iop_format	= xfs_buf_item_format,
	.iop_pin	= xfs_buf_item_pin,
	.iop_unpin	= xfs_buf_item_unpin,
	.iop_unlock	= xfs_buf_item_unlock,
	.iop_committed	= xfs_buf_item_committed,
	.iop_push	= xfs_buf_item_push,
	.iop_committing = xfs_buf_item_committing
L
Linus Torvalds 已提交
692 693
};

694 695 696 697 698 699 700 701 702
STATIC int
xfs_buf_item_get_format(
	struct xfs_buf_log_item	*bip,
	int			count)
{
	ASSERT(bip->bli_formats == NULL);
	bip->bli_format_count = count;

	if (count == 1) {
703
		bip->bli_formats = &bip->__bli_format;
704 705 706 707 708 709
		return 0;
	}

	bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format),
				KM_SLEEP);
	if (!bip->bli_formats)
D
Dave Chinner 已提交
710
		return -ENOMEM;
711 712 713 714 715 716 717
	return 0;
}

STATIC void
xfs_buf_item_free_format(
	struct xfs_buf_log_item	*bip)
{
718
	if (bip->bli_formats != &bip->__bli_format) {
719 720 721 722
		kmem_free(bip->bli_formats);
		bip->bli_formats = NULL;
	}
}
L
Linus Torvalds 已提交
723 724 725

/*
 * Allocate a new buf log item to go with the given buffer.
C
Carlos Maiolino 已提交
726 727
 * Set the buffer's b_log_item field to point to the new
 * buf log item.
L
Linus Torvalds 已提交
728
 */
D
Dave Chinner 已提交
729
int
L
Linus Torvalds 已提交
730
xfs_buf_item_init(
D
Dave Chinner 已提交
731 732
	struct xfs_buf	*bp,
	struct xfs_mount *mp)
L
Linus Torvalds 已提交
733
{
C
Carlos Maiolino 已提交
734
	struct xfs_buf_log_item	*bip = bp->b_log_item;
L
Linus Torvalds 已提交
735 736
	int			chunks;
	int			map_size;
737 738
	int			error;
	int			i;
L
Linus Torvalds 已提交
739 740 741

	/*
	 * Check to see if there is already a buf log item for
C
Carlos Maiolino 已提交
742
	 * this buffer. If we do already have one, there is
L
Linus Torvalds 已提交
743 744
	 * nothing to do here so return.
	 */
745
	ASSERT(bp->b_target->bt_mount == mp);
C
Carlos Maiolino 已提交
746 747
	if (bip != NULL) {
		ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
D
Dave Chinner 已提交
748
		return 0;
C
Carlos Maiolino 已提交
749
	}
L
Linus Torvalds 已提交
750

751
	bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP);
752
	xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
L
Linus Torvalds 已提交
753
	bip->bli_buf = bp;
754 755 756 757 758 759 760 761 762 763 764 765

	/*
	 * chunks is the number of XFS_BLF_CHUNK size pieces the buffer
	 * can be divided into. Make sure not to truncate any pieces.
	 * map_size is the size of the bitmap needed to describe the
	 * chunks of the buffer.
	 *
	 * Discontiguous buffer support follows the layout of the underlying
	 * buffer. This makes the implementation as simple as possible.
	 */
	error = xfs_buf_item_get_format(bip, bp->b_map_count);
	ASSERT(error == 0);
D
Dave Chinner 已提交
766 767 768 769 770
	if (error) {	/* to stop gcc throwing set-but-unused warnings */
		kmem_zone_free(xfs_buf_item_zone, bip);
		return error;
	}

771 772 773 774 775 776 777 778 779 780 781

	for (i = 0; i < bip->bli_format_count; i++) {
		chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len),
				      XFS_BLF_CHUNK);
		map_size = DIV_ROUND_UP(chunks, NBWORD);

		bip->bli_formats[i].blf_type = XFS_LI_BUF;
		bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn;
		bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len;
		bip->bli_formats[i].blf_map_size = map_size;
	}
L
Linus Torvalds 已提交
782

C
Carlos Maiolino 已提交
783
	bp->b_log_item = bip;
D
Dave Chinner 已提交
784 785
	xfs_buf_hold(bp);
	return 0;
L
Linus Torvalds 已提交
786 787 788 789 790 791 792
}


/*
 * Mark bytes first through last inclusive as dirty in the buf
 * item's bitmap.
 */
793
static void
794
xfs_buf_item_log_segment(
L
Linus Torvalds 已提交
795
	uint			first,
796 797
	uint			last,
	uint			*map)
L
Linus Torvalds 已提交
798 799 800 801 802 803 804 805 806 807 808 809 810 811
{
	uint		first_bit;
	uint		last_bit;
	uint		bits_to_set;
	uint		bits_set;
	uint		word_num;
	uint		*wordp;
	uint		bit;
	uint		end_bit;
	uint		mask;

	/*
	 * Convert byte offsets to bit numbers.
	 */
812 813
	first_bit = first >> XFS_BLF_SHIFT;
	last_bit = last >> XFS_BLF_SHIFT;
L
Linus Torvalds 已提交
814 815 816 817 818 819 820 821 822 823 824

	/*
	 * Calculate the total number of bits to be set.
	 */
	bits_to_set = last_bit - first_bit + 1;

	/*
	 * Get a pointer to the first word in the bitmap
	 * to set a bit in.
	 */
	word_num = first_bit >> BIT_TO_WORD_SHIFT;
825
	wordp = &map[word_num];
L
Linus Torvalds 已提交
826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841

	/*
	 * Calculate the starting bit in the first word.
	 */
	bit = first_bit & (uint)(NBWORD - 1);

	/*
	 * First set any bits in the first word of our range.
	 * If it starts at bit 0 of the word, it will be
	 * set below rather than here.  That is what the variable
	 * bit tells us. The variable bits_set tracks the number
	 * of bits that have been set so far.  End_bit is the number
	 * of the last bit to be set in this word plus one.
	 */
	if (bit) {
		end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
X
Xie XiuQi 已提交
842
		mask = ((1U << (end_bit - bit)) - 1) << bit;
L
Linus Torvalds 已提交
843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864
		*wordp |= mask;
		wordp++;
		bits_set = end_bit - bit;
	} else {
		bits_set = 0;
	}

	/*
	 * Now set bits a whole word at a time that are between
	 * first_bit and last_bit.
	 */
	while ((bits_to_set - bits_set) >= NBWORD) {
		*wordp |= 0xffffffff;
		bits_set += NBWORD;
		wordp++;
	}

	/*
	 * Finally, set any bits left to be set in one last partial word.
	 */
	end_bit = bits_to_set - bits_set;
	if (end_bit) {
X
Xie XiuQi 已提交
865
		mask = (1U << end_bit) - 1;
L
Linus Torvalds 已提交
866 867 868 869
		*wordp |= mask;
	}
}

870 871 872 873 874 875
/*
 * Mark bytes first through last inclusive as dirty in the buf
 * item's bitmap.
 */
void
xfs_buf_item_log(
876
	struct xfs_buf_log_item	*bip,
877 878 879 880 881 882 883 884 885 886 887 888 889 890 891
	uint			first,
	uint			last)
{
	int			i;
	uint			start;
	uint			end;
	struct xfs_buf		*bp = bip->bli_buf;

	/*
	 * walk each buffer segment and mark them dirty appropriately.
	 */
	start = 0;
	for (i = 0; i < bip->bli_format_count; i++) {
		if (start > last)
			break;
892 893 894
		end = start + BBTOB(bp->b_maps[i].bm_len) - 1;

		/* skip to the map that includes the first byte to log */
895 896 897 898
		if (first > end) {
			start += BBTOB(bp->b_maps[i].bm_len);
			continue;
		}
899 900 901 902 903 904 905

		/*
		 * Trim the range to this segment and mark it in the bitmap.
		 * Note that we must convert buffer offsets to segment relative
		 * offsets (e.g., the first byte of each segment is byte 0 of
		 * that segment).
		 */
906 907 908 909
		if (first < start)
			first = start;
		if (end > last)
			end = last;
910
		xfs_buf_item_log_segment(first - start, end - start,
911 912
					 &bip->bli_formats[i].blf_data_map[0]);

913
		start += BBTOB(bp->b_maps[i].bm_len);
914 915 916
	}
}

L
Linus Torvalds 已提交
917

918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936
/*
 * Return true if the buffer has any ranges logged/dirtied by a transaction,
 * false otherwise.
 */
bool
xfs_buf_item_dirty_format(
	struct xfs_buf_log_item	*bip)
{
	int			i;

	for (i = 0; i < bip->bli_format_count; i++) {
		if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map,
			     bip->bli_formats[i].blf_map_size))
			return true;
	}

	return false;
}

937 938
STATIC void
xfs_buf_item_free(
939
	struct xfs_buf_log_item	*bip)
940
{
941
	xfs_buf_item_free_format(bip);
942
	kmem_free(bip->bli_item.li_lv_shadow);
943 944 945
	kmem_zone_free(xfs_buf_item_zone, bip);
}

L
Linus Torvalds 已提交
946 947 948 949 950 951 952 953 954 955 956
/*
 * This is called when the buf log item is no longer needed.  It should
 * free the buf log item associated with the given buffer and clear
 * the buffer's pointer to the buf log item.  If there are no more
 * items in the list, clear the b_iodone field of the buffer (see
 * xfs_buf_attach_iodone() below).
 */
void
xfs_buf_item_relse(
	xfs_buf_t	*bp)
{
C
Carlos Maiolino 已提交
957
	struct xfs_buf_log_item	*bip = bp->b_log_item;
L
Linus Torvalds 已提交
958

C
Christoph Hellwig 已提交
959
	trace_xfs_buf_item_relse(bp, _RET_IP_);
960
	ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
C
Christoph Hellwig 已提交
961

C
Carlos Maiolino 已提交
962
	bp->b_log_item = NULL;
963
	if (list_empty(&bp->b_li_list))
964
		bp->b_iodone = NULL;
965

966 967
	xfs_buf_rele(bp);
	xfs_buf_item_free(bip);
L
Linus Torvalds 已提交
968 969 970 971 972 973 974 975
}


/*
 * Add the given log item with its callback to the list of callbacks
 * to be called when the buffer's I/O completes.  If it is not set
 * already, set the buffer's b_iodone() routine to be
 * xfs_buf_iodone_callbacks() and link the log item into the list of
C
Carlos Maiolino 已提交
976
 * items rooted at b_li_list.
L
Linus Torvalds 已提交
977 978 979 980 981 982 983
 */
void
xfs_buf_attach_iodone(
	xfs_buf_t	*bp,
	void		(*cb)(xfs_buf_t *, xfs_log_item_t *),
	xfs_log_item_t	*lip)
{
984
	ASSERT(xfs_buf_islocked(bp));
L
Linus Torvalds 已提交
985 986

	lip->li_cb = cb;
987
	list_add_tail(&lip->li_bio_list, &bp->b_li_list);
L
Linus Torvalds 已提交
988

989 990 991
	ASSERT(bp->b_iodone == NULL ||
	       bp->b_iodone == xfs_buf_iodone_callbacks);
	bp->b_iodone = xfs_buf_iodone_callbacks;
L
Linus Torvalds 已提交
992 993
}

994 995 996
/*
 * We can have many callbacks on a buffer. Running the callbacks individually
 * can cause a lot of contention on the AIL lock, so we allow for a single
997 998
 * callback to be able to scan the remaining items in bp->b_li_list for other
 * items of the same type and callback to be processed in the first call.
999 1000 1001
 *
 * As a result, the loop walking the callback list below will also modify the
 * list. it removes the first item from the list and then runs the callback.
1002
 * The loop then restarts from the new first item int the list. This allows the
1003 1004 1005
 * callback to scan and modify the list attached to the buffer and we don't
 * have to care about maintaining a next item pointer.
 */
L
Linus Torvalds 已提交
1006 1007
STATIC void
xfs_buf_do_callbacks(
1008
	struct xfs_buf		*bp)
L
Linus Torvalds 已提交
1009
{
C
Carlos Maiolino 已提交
1010
	struct xfs_buf_log_item *blip = bp->b_log_item;
1011
	struct xfs_log_item	*lip;
L
Linus Torvalds 已提交
1012

C
Carlos Maiolino 已提交
1013 1014 1015 1016 1017 1018
	/* If there is a buf_log_item attached, run its callback */
	if (blip) {
		lip = &blip->bli_item;
		lip->li_cb(bp, lip);
	}

1019 1020 1021 1022
	while (!list_empty(&bp->b_li_list)) {
		lip = list_first_entry(&bp->b_li_list, struct xfs_log_item,
				       li_bio_list);

L
Linus Torvalds 已提交
1023
		/*
1024
		 * Remove the item from the list, so we don't have any
L
Linus Torvalds 已提交
1025 1026 1027 1028
		 * confusion if the item is added to another buf.
		 * Don't touch the log item after calling its
		 * callback, because it could have freed itself.
		 */
1029
		list_del_init(&lip->li_bio_list);
L
Linus Torvalds 已提交
1030 1031 1032 1033
		lip->li_cb(bp, lip);
	}
}

1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045
/*
 * Invoke the error state callback for each log item affected by the failed I/O.
 *
 * If a metadata buffer write fails with a non-permanent error, the buffer is
 * eventually resubmitted and so the completion callbacks are not run. The error
 * state may need to be propagated to the log items attached to the buffer,
 * however, so the next AIL push of the item knows hot to handle it correctly.
 */
STATIC void
xfs_buf_do_callbacks_fail(
	struct xfs_buf		*bp)
{
1046
	struct xfs_log_item	*lip;
C
Carlos Maiolino 已提交
1047
	struct xfs_ail		*ailp;
1048

C
Carlos Maiolino 已提交
1049 1050 1051 1052 1053
	/*
	 * Buffer log item errors are handled directly by xfs_buf_item_push()
	 * and xfs_buf_iodone_callback_error, and they have no IO error
	 * callbacks. Check only for items in b_li_list.
	 */
1054
	if (list_empty(&bp->b_li_list))
C
Carlos Maiolino 已提交
1055 1056
		return;

1057 1058
	lip = list_first_entry(&bp->b_li_list, struct xfs_log_item,
			li_bio_list);
C
Carlos Maiolino 已提交
1059
	ailp = lip->li_ailp;
1060
	spin_lock(&ailp->ail_lock);
1061
	list_for_each_entry(lip, &bp->b_li_list, li_bio_list) {
1062 1063 1064
		if (lip->li_ops->iop_error)
			lip->li_ops->iop_error(lip, bp);
	}
1065
	spin_unlock(&ailp->ail_lock);
1066 1067
}

1068 1069
static bool
xfs_buf_iodone_callback_error(
1070
	struct xfs_buf		*bp)
L
Linus Torvalds 已提交
1071
{
C
Carlos Maiolino 已提交
1072
	struct xfs_buf_log_item	*bip = bp->b_log_item;
1073
	struct xfs_log_item	*lip;
C
Carlos Maiolino 已提交
1074
	struct xfs_mount	*mp;
1075 1076
	static ulong		lasttime;
	static xfs_buftarg_t	*lasttarg;
1077
	struct xfs_error_cfg	*cfg;
L
Linus Torvalds 已提交
1078

C
Carlos Maiolino 已提交
1079 1080 1081 1082 1083
	/*
	 * The failed buffer might not have a buf_log_item attached or the
	 * log_item list might be empty. Get the mp from the available
	 * xfs_log_item
	 */
1084 1085 1086
	lip = list_first_entry_or_null(&bp->b_li_list, struct xfs_log_item,
				       li_bio_list);
	mp = lip ? lip->li_mountp : bip->bli_item.li_mountp;
C
Carlos Maiolino 已提交
1087

1088 1089 1090 1091
	/*
	 * If we've already decided to shutdown the filesystem because of
	 * I/O errors, there's no point in giving this a retry.
	 */
1092 1093
	if (XFS_FORCED_SHUTDOWN(mp))
		goto out_stale;
L
Linus Torvalds 已提交
1094

1095
	if (bp->b_target != lasttarg ||
1096 1097
	    time_after(jiffies, (lasttime + 5*HZ))) {
		lasttime = jiffies;
1098
		xfs_buf_ioerror_alert(bp, __func__);
1099
	}
1100
	lasttarg = bp->b_target;
L
Linus Torvalds 已提交
1101

1102 1103 1104 1105 1106 1107 1108
	/* synchronous writes will have callers process the error */
	if (!(bp->b_flags & XBF_ASYNC))
		goto out_stale;

	trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
	ASSERT(bp->b_iodone != NULL);

1109 1110
	cfg = xfs_error_get_cfg(mp, XFS_ERR_METADATA, bp->b_error);

1111
	/*
L
Lucas De Marchi 已提交
1112
	 * If the write was asynchronous then no one will be looking for the
1113 1114 1115 1116
	 * error.  If this is the first failure of this type, clear the error
	 * state and write the buffer out again. This means we always retry an
	 * async write failure at least once, but we also need to set the buffer
	 * up to behave correctly now for repeated failures.
1117
	 */
1118
	if (!(bp->b_flags & (XBF_STALE | XBF_WRITE_FAIL)) ||
1119
	     bp->b_last_error != bp->b_error) {
1120
		bp->b_flags |= (XBF_WRITE | XBF_DONE | XBF_WRITE_FAIL);
1121
		bp->b_last_error = bp->b_error;
1122 1123
		if (cfg->retry_timeout != XFS_ERR_RETRY_FOREVER &&
		    !bp->b_first_retry_time)
1124
			bp->b_first_retry_time = jiffies;
1125

1126 1127 1128 1129
		xfs_buf_ioerror(bp, 0);
		xfs_buf_submit(bp);
		return true;
	}
1130

1131 1132 1133 1134
	/*
	 * Repeated failure on an async write. Take action according to the
	 * error configuration we have been set up to use.
	 */
1135 1136 1137 1138

	if (cfg->max_retries != XFS_ERR_RETRY_FOREVER &&
	    ++bp->b_retries > cfg->max_retries)
			goto permanent_error;
1139
	if (cfg->retry_timeout != XFS_ERR_RETRY_FOREVER &&
1140 1141
	    time_after(jiffies, cfg->retry_timeout + bp->b_first_retry_time))
			goto permanent_error;
1142

1143 1144 1145 1146
	/* At unmount we may treat errors differently */
	if ((mp->m_flags & XFS_MOUNT_UNMOUNTING) && mp->m_fail_unmount)
		goto permanent_error;

1147 1148 1149 1150 1151
	/*
	 * Still a transient error, run IO completion failure callbacks and let
	 * the higher layers retry the buffer.
	 */
	xfs_buf_do_callbacks_fail(bp);
1152 1153 1154
	xfs_buf_ioerror(bp, 0);
	xfs_buf_relse(bp);
	return true;
C
Christoph Hellwig 已提交
1155

1156
	/*
1157 1158
	 * Permanent error - we need to trigger a shutdown if we haven't already
	 * to indicate that inconsistency will result from this action.
1159
	 */
1160 1161 1162
permanent_error:
	xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
out_stale:
1163
	xfs_buf_stale(bp);
1164
	bp->b_flags |= XBF_DONE;
C
Christoph Hellwig 已提交
1165
	trace_xfs_buf_error_relse(bp, _RET_IP_);
1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191
	return false;
}

/*
 * This is the iodone() function for buffers which have had callbacks attached
 * to them by xfs_buf_attach_iodone(). We need to iterate the items on the
 * callback list, mark the buffer as having no more callbacks and then push the
 * buffer through IO completion processing.
 */
void
xfs_buf_iodone_callbacks(
	struct xfs_buf		*bp)
{
	/*
	 * If there is an error, process it. Some errors require us
	 * to run callbacks after failure processing is done so we
	 * detect that and take appropriate action.
	 */
	if (bp->b_error && xfs_buf_iodone_callback_error(bp))
		return;

	/*
	 * Successful IO or permanent error. Either way, we can clear the
	 * retry state here in preparation for the next error that may occur.
	 */
	bp->b_last_error = 0;
1192
	bp->b_retries = 0;
1193
	bp->b_first_retry_time = 0;
C
Christoph Hellwig 已提交
1194

1195
	xfs_buf_do_callbacks(bp);
C
Carlos Maiolino 已提交
1196
	bp->b_log_item = NULL;
1197
	list_del_init(&bp->b_li_list);
1198
	bp->b_iodone = NULL;
1199
	xfs_buf_ioend(bp);
L
Linus Torvalds 已提交
1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210
}

/*
 * This is the iodone() function for buffers which have been
 * logged.  It is called when they are eventually flushed out.
 * It should remove the buf item from the AIL, and free the buf item.
 * It is called by xfs_buf_iodone_callbacks() above which will take
 * care of cleaning up the buffer itself.
 */
void
xfs_buf_iodone(
1211 1212
	struct xfs_buf		*bp,
	struct xfs_log_item	*lip)
L
Linus Torvalds 已提交
1213
{
1214
	struct xfs_ail		*ailp = lip->li_ailp;
L
Linus Torvalds 已提交
1215

1216
	ASSERT(BUF_ITEM(lip)->bli_buf == bp);
L
Linus Torvalds 已提交
1217

1218
	xfs_buf_rele(bp);
L
Linus Torvalds 已提交
1219 1220 1221 1222 1223 1224

	/*
	 * If we are forcibly shutting down, this may well be
	 * off the AIL already. That's because we simulate the
	 * log-committed callbacks to unpin these buffers. Or we may never
	 * have put this item on AIL because of the transaction was
1225
	 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
L
Linus Torvalds 已提交
1226 1227 1228
	 *
	 * Either way, AIL is useless if we're forcing a shutdown.
	 */
1229
	spin_lock(&ailp->ail_lock);
1230
	xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE);
1231
	xfs_buf_item_free(BUF_ITEM(lip));
L
Linus Torvalds 已提交
1232
}
1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243

/*
 * Requeue a failed buffer for writeback
 *
 * Return true if the buffer has been re-queued properly, false otherwise
 */
bool
xfs_buf_resubmit_failed_buffers(
	struct xfs_buf		*bp,
	struct list_head	*buffer_list)
{
1244
	struct xfs_log_item	*lip;
1245 1246 1247 1248

	/*
	 * Clear XFS_LI_FAILED flag from all items before resubmit
	 *
1249
	 * XFS_LI_FAILED set/clear is protected by ail_lock, caller  this
1250 1251
	 * function already have it acquired
	 */
1252
	list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
1253 1254 1255 1256 1257
		xfs_clear_li_failed(lip);

	/* Add this buffer back to the delayed write list */
	return xfs_buf_delwri_queue(bp, buffer_list);
}