ARM: 7409/1: Do not call flush_cache_user_range with mmap_sem held
[linux/fpc-iii.git] / fs / xfs / xfs_buf_item.c
blob7888a75630788e3bec732c1fcf00d5b974bca800
1 /*
2 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_types.h"
21 #include "xfs_bit.h"
22 #include "xfs_log.h"
23 #include "xfs_inum.h"
24 #include "xfs_trans.h"
25 #include "xfs_sb.h"
26 #include "xfs_ag.h"
27 #include "xfs_mount.h"
28 #include "xfs_buf_item.h"
29 #include "xfs_trans_priv.h"
30 #include "xfs_error.h"
31 #include "xfs_trace.h"
34 kmem_zone_t *xfs_buf_item_zone;
36 static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip)
38 return container_of(lip, struct xfs_buf_log_item, bli_item);
42 #ifdef XFS_TRANS_DEBUG
44 * This function uses an alternate strategy for tracking the bytes
45 * that the user requests to be logged. This can then be used
46 * in conjunction with the bli_orig array in the buf log item to
47 * catch bugs in our callers' code.
49 * We also double check the bits set in xfs_buf_item_log using a
50 * simple algorithm to check that every byte is accounted for.
52 STATIC void
53 xfs_buf_item_log_debug(
54 xfs_buf_log_item_t *bip,
55 uint first,
56 uint last)
58 uint x;
59 uint byte;
60 uint nbytes;
61 uint chunk_num;
62 uint word_num;
63 uint bit_num;
64 uint bit_set;
65 uint *wordp;
67 ASSERT(bip->bli_logged != NULL);
68 byte = first;
69 nbytes = last - first + 1;
70 bfset(bip->bli_logged, first, nbytes);
71 for (x = 0; x < nbytes; x++) {
72 chunk_num = byte >> XFS_BLF_SHIFT;
73 word_num = chunk_num >> BIT_TO_WORD_SHIFT;
74 bit_num = chunk_num & (NBWORD - 1);
75 wordp = &(bip->bli_format.blf_data_map[word_num]);
76 bit_set = *wordp & (1 << bit_num);
77 ASSERT(bit_set);
78 byte++;
83 * This function is called when we flush something into a buffer without
84 * logging it. This happens for things like inodes which are logged
85 * separately from the buffer.
87 void
88 xfs_buf_item_flush_log_debug(
89 xfs_buf_t *bp,
90 uint first,
91 uint last)
93 xfs_buf_log_item_t *bip;
94 uint nbytes;
96 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
97 if ((bip == NULL) || (bip->bli_item.li_type != XFS_LI_BUF)) {
98 return;
101 ASSERT(bip->bli_logged != NULL);
102 nbytes = last - first + 1;
103 bfset(bip->bli_logged, first, nbytes);
107 * This function is called to verify that our callers have logged
108 * all the bytes that they changed.
110 * It does this by comparing the original copy of the buffer stored in
111 * the buf log item's bli_orig array to the current copy of the buffer
112 * and ensuring that all bytes which mismatch are set in the bli_logged
113 * array of the buf log item.
115 STATIC void
116 xfs_buf_item_log_check(
117 xfs_buf_log_item_t *bip)
119 char *orig;
120 char *buffer;
121 int x;
122 xfs_buf_t *bp;
124 ASSERT(bip->bli_orig != NULL);
125 ASSERT(bip->bli_logged != NULL);
127 bp = bip->bli_buf;
128 ASSERT(XFS_BUF_COUNT(bp) > 0);
129 ASSERT(XFS_BUF_PTR(bp) != NULL);
130 orig = bip->bli_orig;
131 buffer = XFS_BUF_PTR(bp);
132 for (x = 0; x < XFS_BUF_COUNT(bp); x++) {
133 if (orig[x] != buffer[x] && !btst(bip->bli_logged, x)) {
134 xfs_emerg(bp->b_mount,
135 "%s: bip %x buffer %x orig %x index %d",
136 __func__, bip, bp, orig, x);
137 ASSERT(0);
141 #else
142 #define xfs_buf_item_log_debug(x,y,z)
143 #define xfs_buf_item_log_check(x)
144 #endif
146 STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp);
149 * This returns the number of log iovecs needed to log the
150 * given buf log item.
152 * It calculates this as 1 iovec for the buf log format structure
153 * and 1 for each stretch of non-contiguous chunks to be logged.
154 * Contiguous chunks are logged in a single iovec.
156 * If the XFS_BLI_STALE flag has been set, then log nothing.
158 STATIC uint
159 xfs_buf_item_size(
160 struct xfs_log_item *lip)
162 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
163 struct xfs_buf *bp = bip->bli_buf;
164 uint nvecs;
165 int next_bit;
166 int last_bit;
168 ASSERT(atomic_read(&bip->bli_refcount) > 0);
169 if (bip->bli_flags & XFS_BLI_STALE) {
171 * The buffer is stale, so all we need to log
172 * is the buf log format structure with the
173 * cancel flag in it.
175 trace_xfs_buf_item_size_stale(bip);
176 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
177 return 1;
180 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
181 nvecs = 1;
182 last_bit = xfs_next_bit(bip->bli_format.blf_data_map,
183 bip->bli_format.blf_map_size, 0);
184 ASSERT(last_bit != -1);
185 nvecs++;
186 while (last_bit != -1) {
188 * This takes the bit number to start looking from and
189 * returns the next set bit from there. It returns -1
190 * if there are no more bits set or the start bit is
191 * beyond the end of the bitmap.
193 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
194 bip->bli_format.blf_map_size,
195 last_bit + 1);
197 * If we run out of bits, leave the loop,
198 * else if we find a new set of bits bump the number of vecs,
199 * else keep scanning the current set of bits.
201 if (next_bit == -1) {
202 last_bit = -1;
203 } else if (next_bit != last_bit + 1) {
204 last_bit = next_bit;
205 nvecs++;
206 } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
207 (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
208 XFS_BLF_CHUNK)) {
209 last_bit = next_bit;
210 nvecs++;
211 } else {
212 last_bit++;
216 trace_xfs_buf_item_size(bip);
217 return nvecs;
221 * This is called to fill in the vector of log iovecs for the
222 * given log buf item. It fills the first entry with a buf log
223 * format structure, and the rest point to contiguous chunks
224 * within the buffer.
226 STATIC void
227 xfs_buf_item_format(
228 struct xfs_log_item *lip,
229 struct xfs_log_iovec *vecp)
231 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
232 struct xfs_buf *bp = bip->bli_buf;
233 uint base_size;
234 uint nvecs;
235 int first_bit;
236 int last_bit;
237 int next_bit;
238 uint nbits;
239 uint buffer_offset;
241 ASSERT(atomic_read(&bip->bli_refcount) > 0);
242 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
243 (bip->bli_flags & XFS_BLI_STALE));
246 * The size of the base structure is the size of the
247 * declared structure plus the space for the extra words
248 * of the bitmap. We subtract one from the map size, because
249 * the first element of the bitmap is accounted for in the
250 * size of the base structure.
252 base_size =
253 (uint)(sizeof(xfs_buf_log_format_t) +
254 ((bip->bli_format.blf_map_size - 1) * sizeof(uint)));
255 vecp->i_addr = &bip->bli_format;
256 vecp->i_len = base_size;
257 vecp->i_type = XLOG_REG_TYPE_BFORMAT;
258 vecp++;
259 nvecs = 1;
262 * If it is an inode buffer, transfer the in-memory state to the
263 * format flags and clear the in-memory state. We do not transfer
264 * this state if the inode buffer allocation has not yet been committed
265 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
266 * correct replay of the inode allocation.
268 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
269 if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
270 xfs_log_item_in_current_chkpt(lip)))
271 bip->bli_format.blf_flags |= XFS_BLF_INODE_BUF;
272 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
275 if (bip->bli_flags & XFS_BLI_STALE) {
277 * The buffer is stale, so all we need to log
278 * is the buf log format structure with the
279 * cancel flag in it.
281 trace_xfs_buf_item_format_stale(bip);
282 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
283 bip->bli_format.blf_size = nvecs;
284 return;
288 * Fill in an iovec for each set of contiguous chunks.
290 first_bit = xfs_next_bit(bip->bli_format.blf_data_map,
291 bip->bli_format.blf_map_size, 0);
292 ASSERT(first_bit != -1);
293 last_bit = first_bit;
294 nbits = 1;
295 for (;;) {
297 * This takes the bit number to start looking from and
298 * returns the next set bit from there. It returns -1
299 * if there are no more bits set or the start bit is
300 * beyond the end of the bitmap.
302 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
303 bip->bli_format.blf_map_size,
304 (uint)last_bit + 1);
306 * If we run out of bits fill in the last iovec and get
307 * out of the loop.
308 * Else if we start a new set of bits then fill in the
309 * iovec for the series we were looking at and start
310 * counting the bits in the new one.
311 * Else we're still in the same set of bits so just
312 * keep counting and scanning.
314 if (next_bit == -1) {
315 buffer_offset = first_bit * XFS_BLF_CHUNK;
316 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
317 vecp->i_len = nbits * XFS_BLF_CHUNK;
318 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
319 nvecs++;
320 break;
321 } else if (next_bit != last_bit + 1) {
322 buffer_offset = first_bit * XFS_BLF_CHUNK;
323 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
324 vecp->i_len = nbits * XFS_BLF_CHUNK;
325 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
326 nvecs++;
327 vecp++;
328 first_bit = next_bit;
329 last_bit = next_bit;
330 nbits = 1;
331 } else if (xfs_buf_offset(bp, next_bit << XFS_BLF_SHIFT) !=
332 (xfs_buf_offset(bp, last_bit << XFS_BLF_SHIFT) +
333 XFS_BLF_CHUNK)) {
334 buffer_offset = first_bit * XFS_BLF_CHUNK;
335 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
336 vecp->i_len = nbits * XFS_BLF_CHUNK;
337 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
338 /* You would think we need to bump the nvecs here too, but we do not
339 * this number is used by recovery, and it gets confused by the boundary
340 * split here
341 * nvecs++;
343 vecp++;
344 first_bit = next_bit;
345 last_bit = next_bit;
346 nbits = 1;
347 } else {
348 last_bit++;
349 nbits++;
352 bip->bli_format.blf_size = nvecs;
355 * Check to make sure everything is consistent.
357 trace_xfs_buf_item_format(bip);
358 xfs_buf_item_log_check(bip);
362 * This is called to pin the buffer associated with the buf log item in memory
363 * so it cannot be written out.
365 * We also always take a reference to the buffer log item here so that the bli
366 * is held while the item is pinned in memory. This means that we can
367 * unconditionally drop the reference count a transaction holds when the
368 * transaction is completed.
370 STATIC void
371 xfs_buf_item_pin(
372 struct xfs_log_item *lip)
374 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
376 ASSERT(XFS_BUF_ISBUSY(bip->bli_buf));
377 ASSERT(atomic_read(&bip->bli_refcount) > 0);
378 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
379 (bip->bli_flags & XFS_BLI_STALE));
381 trace_xfs_buf_item_pin(bip);
383 atomic_inc(&bip->bli_refcount);
384 atomic_inc(&bip->bli_buf->b_pin_count);
388 * This is called to unpin the buffer associated with the buf log
389 * item which was previously pinned with a call to xfs_buf_item_pin().
391 * Also drop the reference to the buf item for the current transaction.
392 * If the XFS_BLI_STALE flag is set and we are the last reference,
393 * then free up the buf log item and unlock the buffer.
395 * If the remove flag is set we are called from uncommit in the
396 * forced-shutdown path. If that is true and the reference count on
397 * the log item is going to drop to zero we need to free the item's
398 * descriptor in the transaction.
400 STATIC void
401 xfs_buf_item_unpin(
402 struct xfs_log_item *lip,
403 int remove)
405 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
406 xfs_buf_t *bp = bip->bli_buf;
407 struct xfs_ail *ailp = lip->li_ailp;
408 int stale = bip->bli_flags & XFS_BLI_STALE;
409 int freed;
411 ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip);
412 ASSERT(atomic_read(&bip->bli_refcount) > 0);
414 trace_xfs_buf_item_unpin(bip);
416 freed = atomic_dec_and_test(&bip->bli_refcount);
418 if (atomic_dec_and_test(&bp->b_pin_count))
419 wake_up_all(&bp->b_waiters);
421 if (freed && stale) {
422 ASSERT(bip->bli_flags & XFS_BLI_STALE);
423 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
424 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
425 ASSERT(XFS_BUF_ISSTALE(bp));
426 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
428 trace_xfs_buf_item_unpin_stale(bip);
430 if (remove) {
432 * If we are in a transaction context, we have to
433 * remove the log item from the transaction as we are
434 * about to release our reference to the buffer. If we
435 * don't, the unlock that occurs later in
436 * xfs_trans_uncommit() will try to reference the
437 * buffer which we no longer have a hold on.
439 if (lip->li_desc)
440 xfs_trans_del_item(lip);
443 * Since the transaction no longer refers to the buffer,
444 * the buffer should no longer refer to the transaction.
446 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
450 * If we get called here because of an IO error, we may
451 * or may not have the item on the AIL. xfs_trans_ail_delete()
452 * will take care of that situation.
453 * xfs_trans_ail_delete() drops the AIL lock.
455 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
456 xfs_buf_do_callbacks(bp);
457 XFS_BUF_SET_FSPRIVATE(bp, NULL);
458 XFS_BUF_CLR_IODONE_FUNC(bp);
459 } else {
460 spin_lock(&ailp->xa_lock);
461 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip);
462 xfs_buf_item_relse(bp);
463 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL);
465 xfs_buf_relse(bp);
470 * This is called to attempt to lock the buffer associated with this
471 * buf log item. Don't sleep on the buffer lock. If we can't get
472 * the lock right away, return 0. If we can get the lock, take a
473 * reference to the buffer. If this is a delayed write buffer that
474 * needs AIL help to be written back, invoke the pushbuf routine
475 * rather than the normal success path.
477 STATIC uint
478 xfs_buf_item_trylock(
479 struct xfs_log_item *lip)
481 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
482 struct xfs_buf *bp = bip->bli_buf;
484 if (XFS_BUF_ISPINNED(bp))
485 return XFS_ITEM_PINNED;
486 if (!XFS_BUF_CPSEMA(bp))
487 return XFS_ITEM_LOCKED;
489 /* take a reference to the buffer. */
490 XFS_BUF_HOLD(bp);
492 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
493 trace_xfs_buf_item_trylock(bip);
494 if (XFS_BUF_ISDELAYWRITE(bp))
495 return XFS_ITEM_PUSHBUF;
496 return XFS_ITEM_SUCCESS;
500 * Release the buffer associated with the buf log item. If there is no dirty
501 * logged data associated with the buffer recorded in the buf log item, then
502 * free the buf log item and remove the reference to it in the buffer.
504 * This call ignores the recursion count. It is only called when the buffer
505 * should REALLY be unlocked, regardless of the recursion count.
507 * We unconditionally drop the transaction's reference to the log item. If the
508 * item was logged, then another reference was taken when it was pinned, so we
509 * can safely drop the transaction reference now. This also allows us to avoid
510 * potential races with the unpin code freeing the bli by not referencing the
511 * bli after we've dropped the reference count.
513 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
514 * if necessary but do not unlock the buffer. This is for support of
515 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
516 * free the item.
518 STATIC void
519 xfs_buf_item_unlock(
520 struct xfs_log_item *lip)
522 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
523 struct xfs_buf *bp = bip->bli_buf;
524 int aborted;
525 uint hold;
527 /* Clear the buffer's association with this transaction. */
528 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
531 * If this is a transaction abort, don't return early. Instead, allow
532 * the brelse to happen. Normally it would be done for stale
533 * (cancelled) buffers at unpin time, but we'll never go through the
534 * pin/unpin cycle if we abort inside commit.
536 aborted = (lip->li_flags & XFS_LI_ABORTED) != 0;
539 * Before possibly freeing the buf item, determine if we should
540 * release the buffer at the end of this routine.
542 hold = bip->bli_flags & XFS_BLI_HOLD;
544 /* Clear the per transaction state. */
545 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD);
548 * If the buf item is marked stale, then don't do anything. We'll
549 * unlock the buffer and free the buf item when the buffer is unpinned
550 * for the last time.
552 if (bip->bli_flags & XFS_BLI_STALE) {
553 trace_xfs_buf_item_unlock_stale(bip);
554 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
555 if (!aborted) {
556 atomic_dec(&bip->bli_refcount);
557 return;
561 trace_xfs_buf_item_unlock(bip);
564 * If the buf item isn't tracking any data, free it, otherwise drop the
565 * reference we hold to it.
567 if (xfs_bitmap_empty(bip->bli_format.blf_data_map,
568 bip->bli_format.blf_map_size))
569 xfs_buf_item_relse(bp);
570 else
571 atomic_dec(&bip->bli_refcount);
573 if (!hold)
574 xfs_buf_relse(bp);
578 * This is called to find out where the oldest active copy of the
579 * buf log item in the on disk log resides now that the last log
580 * write of it completed at the given lsn.
581 * We always re-log all the dirty data in a buffer, so usually the
582 * latest copy in the on disk log is the only one that matters. For
583 * those cases we simply return the given lsn.
585 * The one exception to this is for buffers full of newly allocated
586 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
587 * flag set, indicating that only the di_next_unlinked fields from the
588 * inodes in the buffers will be replayed during recovery. If the
589 * original newly allocated inode images have not yet been flushed
590 * when the buffer is so relogged, then we need to make sure that we
591 * keep the old images in the 'active' portion of the log. We do this
592 * by returning the original lsn of that transaction here rather than
593 * the current one.
595 STATIC xfs_lsn_t
596 xfs_buf_item_committed(
597 struct xfs_log_item *lip,
598 xfs_lsn_t lsn)
600 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
602 trace_xfs_buf_item_committed(bip);
604 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
605 return lip->li_lsn;
606 return lsn;
610 * The buffer is locked, but is not a delayed write buffer. This happens
611 * if we race with IO completion and hence we don't want to try to write it
612 * again. Just release the buffer.
614 STATIC void
615 xfs_buf_item_push(
616 struct xfs_log_item *lip)
618 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
619 struct xfs_buf *bp = bip->bli_buf;
621 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
622 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
624 trace_xfs_buf_item_push(bip);
626 xfs_buf_relse(bp);
630 * The buffer is locked and is a delayed write buffer. Promote the buffer
631 * in the delayed write queue as the caller knows that they must invoke
632 * the xfsbufd to get this buffer written. We have to unlock the buffer
633 * to allow the xfsbufd to write it, too.
635 STATIC bool
636 xfs_buf_item_pushbuf(
637 struct xfs_log_item *lip)
639 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
640 struct xfs_buf *bp = bip->bli_buf;
642 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
643 ASSERT(XFS_BUF_ISDELAYWRITE(bp));
645 trace_xfs_buf_item_pushbuf(bip);
647 xfs_buf_delwri_promote(bp);
648 xfs_buf_relse(bp);
649 return true;
652 STATIC void
653 xfs_buf_item_committing(
654 struct xfs_log_item *lip,
655 xfs_lsn_t commit_lsn)
660 * This is the ops vector shared by all buf log items.
662 static struct xfs_item_ops xfs_buf_item_ops = {
663 .iop_size = xfs_buf_item_size,
664 .iop_format = xfs_buf_item_format,
665 .iop_pin = xfs_buf_item_pin,
666 .iop_unpin = xfs_buf_item_unpin,
667 .iop_trylock = xfs_buf_item_trylock,
668 .iop_unlock = xfs_buf_item_unlock,
669 .iop_committed = xfs_buf_item_committed,
670 .iop_push = xfs_buf_item_push,
671 .iop_pushbuf = xfs_buf_item_pushbuf,
672 .iop_committing = xfs_buf_item_committing
677 * Allocate a new buf log item to go with the given buffer.
678 * Set the buffer's b_fsprivate field to point to the new
679 * buf log item. If there are other item's attached to the
680 * buffer (see xfs_buf_attach_iodone() below), then put the
681 * buf log item at the front.
683 void
684 xfs_buf_item_init(
685 xfs_buf_t *bp,
686 xfs_mount_t *mp)
688 xfs_log_item_t *lip;
689 xfs_buf_log_item_t *bip;
690 int chunks;
691 int map_size;
694 * Check to see if there is already a buf log item for
695 * this buffer. If there is, it is guaranteed to be
696 * the first. If we do already have one, there is
697 * nothing to do here so return.
699 ASSERT(bp->b_target->bt_mount == mp);
700 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
701 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
702 if (lip->li_type == XFS_LI_BUF) {
703 return;
708 * chunks is the number of XFS_BLF_CHUNK size pieces
709 * the buffer can be divided into. Make sure not to
710 * truncate any pieces. map_size is the size of the
711 * bitmap needed to describe the chunks of the buffer.
713 chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLF_CHUNK - 1)) >> XFS_BLF_SHIFT);
714 map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);
716 bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
717 KM_SLEEP);
718 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
719 bip->bli_buf = bp;
720 xfs_buf_hold(bp);
721 bip->bli_format.blf_type = XFS_LI_BUF;
722 bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp);
723 bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp));
724 bip->bli_format.blf_map_size = map_size;
726 #ifdef XFS_TRANS_DEBUG
728 * Allocate the arrays for tracking what needs to be logged
729 * and what our callers request to be logged. bli_orig
730 * holds a copy of the original, clean buffer for comparison
731 * against, and bli_logged keeps a 1 bit flag per byte in
732 * the buffer to indicate which bytes the callers have asked
733 * to have logged.
735 bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP);
736 memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp));
737 bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP);
738 #endif
741 * Put the buf item into the list of items attached to the
742 * buffer at the front.
744 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
745 bip->bli_item.li_bio_list =
746 XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
748 XFS_BUF_SET_FSPRIVATE(bp, bip);
753 * Mark bytes first through last inclusive as dirty in the buf
754 * item's bitmap.
756 void
757 xfs_buf_item_log(
758 xfs_buf_log_item_t *bip,
759 uint first,
760 uint last)
762 uint first_bit;
763 uint last_bit;
764 uint bits_to_set;
765 uint bits_set;
766 uint word_num;
767 uint *wordp;
768 uint bit;
769 uint end_bit;
770 uint mask;
773 * Mark the item as having some dirty data for
774 * quick reference in xfs_buf_item_dirty.
776 bip->bli_flags |= XFS_BLI_DIRTY;
779 * Convert byte offsets to bit numbers.
781 first_bit = first >> XFS_BLF_SHIFT;
782 last_bit = last >> XFS_BLF_SHIFT;
785 * Calculate the total number of bits to be set.
787 bits_to_set = last_bit - first_bit + 1;
790 * Get a pointer to the first word in the bitmap
791 * to set a bit in.
793 word_num = first_bit >> BIT_TO_WORD_SHIFT;
794 wordp = &(bip->bli_format.blf_data_map[word_num]);
797 * Calculate the starting bit in the first word.
799 bit = first_bit & (uint)(NBWORD - 1);
802 * First set any bits in the first word of our range.
803 * If it starts at bit 0 of the word, it will be
804 * set below rather than here. That is what the variable
805 * bit tells us. The variable bits_set tracks the number
806 * of bits that have been set so far. End_bit is the number
807 * of the last bit to be set in this word plus one.
809 if (bit) {
810 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
811 mask = ((1 << (end_bit - bit)) - 1) << bit;
812 *wordp |= mask;
813 wordp++;
814 bits_set = end_bit - bit;
815 } else {
816 bits_set = 0;
820 * Now set bits a whole word at a time that are between
821 * first_bit and last_bit.
823 while ((bits_to_set - bits_set) >= NBWORD) {
824 *wordp |= 0xffffffff;
825 bits_set += NBWORD;
826 wordp++;
830 * Finally, set any bits left to be set in one last partial word.
832 end_bit = bits_to_set - bits_set;
833 if (end_bit) {
834 mask = (1 << end_bit) - 1;
835 *wordp |= mask;
838 xfs_buf_item_log_debug(bip, first, last);
843 * Return 1 if the buffer has some data that has been logged (at any
844 * point, not just the current transaction) and 0 if not.
846 uint
847 xfs_buf_item_dirty(
848 xfs_buf_log_item_t *bip)
850 return (bip->bli_flags & XFS_BLI_DIRTY);
853 STATIC void
854 xfs_buf_item_free(
855 xfs_buf_log_item_t *bip)
857 #ifdef XFS_TRANS_DEBUG
858 kmem_free(bip->bli_orig);
859 kmem_free(bip->bli_logged);
860 #endif /* XFS_TRANS_DEBUG */
862 kmem_zone_free(xfs_buf_item_zone, bip);
866 * This is called when the buf log item is no longer needed. It should
867 * free the buf log item associated with the given buffer and clear
868 * the buffer's pointer to the buf log item. If there are no more
869 * items in the list, clear the b_iodone field of the buffer (see
870 * xfs_buf_attach_iodone() below).
872 void
873 xfs_buf_item_relse(
874 xfs_buf_t *bp)
876 xfs_buf_log_item_t *bip;
878 trace_xfs_buf_item_relse(bp, _RET_IP_);
880 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
881 XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list);
882 if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) &&
883 (XFS_BUF_IODONE_FUNC(bp) != NULL)) {
884 XFS_BUF_CLR_IODONE_FUNC(bp);
886 xfs_buf_rele(bp);
887 xfs_buf_item_free(bip);
892 * Add the given log item with its callback to the list of callbacks
893 * to be called when the buffer's I/O completes. If it is not set
894 * already, set the buffer's b_iodone() routine to be
895 * xfs_buf_iodone_callbacks() and link the log item into the list of
896 * items rooted at b_fsprivate. Items are always added as the second
897 * entry in the list if there is a first, because the buf item code
898 * assumes that the buf log item is first.
900 void
901 xfs_buf_attach_iodone(
902 xfs_buf_t *bp,
903 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
904 xfs_log_item_t *lip)
906 xfs_log_item_t *head_lip;
908 ASSERT(XFS_BUF_ISBUSY(bp));
909 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
911 lip->li_cb = cb;
912 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
913 head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
914 lip->li_bio_list = head_lip->li_bio_list;
915 head_lip->li_bio_list = lip;
916 } else {
917 XFS_BUF_SET_FSPRIVATE(bp, lip);
920 ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) ||
921 (XFS_BUF_IODONE_FUNC(bp) == NULL));
922 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
926 * We can have many callbacks on a buffer. Running the callbacks individually
927 * can cause a lot of contention on the AIL lock, so we allow for a single
928 * callback to be able to scan the remaining lip->li_bio_list for other items
929 * of the same type and callback to be processed in the first call.
931 * As a result, the loop walking the callback list below will also modify the
932 * list. it removes the first item from the list and then runs the callback.
933 * The loop then restarts from the new head of the list. This allows the
934 * callback to scan and modify the list attached to the buffer and we don't
935 * have to care about maintaining a next item pointer.
937 STATIC void
938 xfs_buf_do_callbacks(
939 struct xfs_buf *bp)
941 struct xfs_log_item *lip;
943 while ((lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *)) != NULL) {
944 XFS_BUF_SET_FSPRIVATE(bp, lip->li_bio_list);
945 ASSERT(lip->li_cb != NULL);
947 * Clear the next pointer so we don't have any
948 * confusion if the item is added to another buf.
949 * Don't touch the log item after calling its
950 * callback, because it could have freed itself.
952 lip->li_bio_list = NULL;
953 lip->li_cb(bp, lip);
958 * This is the iodone() function for buffers which have had callbacks
959 * attached to them by xfs_buf_attach_iodone(). It should remove each
960 * log item from the buffer's list and call the callback of each in turn.
961 * When done, the buffer's fsprivate field is set to NULL and the buffer
962 * is unlocked with a call to iodone().
964 void
965 xfs_buf_iodone_callbacks(
966 struct xfs_buf *bp)
968 struct xfs_log_item *lip = bp->b_fspriv;
969 struct xfs_mount *mp = lip->li_mountp;
970 static ulong lasttime;
971 static xfs_buftarg_t *lasttarg;
973 if (likely(!XFS_BUF_GETERROR(bp)))
974 goto do_callbacks;
977 * If we've already decided to shutdown the filesystem because of
978 * I/O errors, there's no point in giving this a retry.
980 if (XFS_FORCED_SHUTDOWN(mp)) {
981 XFS_BUF_SUPER_STALE(bp);
982 trace_xfs_buf_item_iodone(bp, _RET_IP_);
983 goto do_callbacks;
986 if (XFS_BUF_TARGET(bp) != lasttarg ||
987 time_after(jiffies, (lasttime + 5*HZ))) {
988 lasttime = jiffies;
989 xfs_alert(mp, "Device %s: metadata write error block 0x%llx",
990 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)),
991 (__uint64_t)XFS_BUF_ADDR(bp));
993 lasttarg = XFS_BUF_TARGET(bp);
996 * If the write was asynchronous then no one will be looking for the
997 * error. Clear the error state and write the buffer out again.
999 * During sync or umount we'll write all pending buffers again
1000 * synchronous, which will catch these errors if they keep hanging
1001 * around.
1003 if (XFS_BUF_ISASYNC(bp)) {
1004 XFS_BUF_ERROR(bp, 0); /* errno of 0 unsets the flag */
1006 if (!XFS_BUF_ISSTALE(bp)) {
1007 XFS_BUF_DELAYWRITE(bp);
1008 XFS_BUF_DONE(bp);
1009 XFS_BUF_SET_START(bp);
1011 ASSERT(XFS_BUF_IODONE_FUNC(bp));
1012 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1013 xfs_buf_relse(bp);
1014 return;
1018 * If the write of the buffer was synchronous, we want to make
1019 * sure to return the error to the caller of xfs_bwrite().
1021 XFS_BUF_STALE(bp);
1022 XFS_BUF_DONE(bp);
1023 XFS_BUF_UNDELAYWRITE(bp);
1025 trace_xfs_buf_error_relse(bp, _RET_IP_);
1027 do_callbacks:
1028 xfs_buf_do_callbacks(bp);
1029 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1030 XFS_BUF_CLR_IODONE_FUNC(bp);
1031 xfs_buf_ioend(bp, 0);
1035 * This is the iodone() function for buffers which have been
1036 * logged. It is called when they are eventually flushed out.
1037 * It should remove the buf item from the AIL, and free the buf item.
1038 * It is called by xfs_buf_iodone_callbacks() above which will take
1039 * care of cleaning up the buffer itself.
1041 void
1042 xfs_buf_iodone(
1043 struct xfs_buf *bp,
1044 struct xfs_log_item *lip)
1046 struct xfs_ail *ailp = lip->li_ailp;
1048 ASSERT(BUF_ITEM(lip)->bli_buf == bp);
1050 xfs_buf_rele(bp);
1053 * If we are forcibly shutting down, this may well be
1054 * off the AIL already. That's because we simulate the
1055 * log-committed callbacks to unpin these buffers. Or we may never
1056 * have put this item on AIL because of the transaction was
1057 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
1059 * Either way, AIL is useless if we're forcing a shutdown.
1061 spin_lock(&ailp->xa_lock);
1062 xfs_trans_ail_delete(ailp, lip);
1063 xfs_buf_item_free(BUF_ITEM(lip));