Loading...
1/*
2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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.
8 *
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.
13 *
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
17 */
18#include "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_shared.h"
21#include "xfs_format.h"
22#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
24#include "xfs_mount.h"
25#include "xfs_inode.h"
26#include "xfs_trans.h"
27#include "xfs_buf_item.h"
28#include "xfs_trans_priv.h"
29#include "xfs_error.h"
30#include "xfs_trace.h"
31
32/*
33 * Check to see if a buffer matching the given parameters is already
34 * a part of the given transaction.
35 */
36STATIC struct xfs_buf *
37xfs_trans_buf_item_match(
38 struct xfs_trans *tp,
39 struct xfs_buftarg *target,
40 struct xfs_buf_map *map,
41 int nmaps)
42{
43 struct xfs_log_item_desc *lidp;
44 struct xfs_buf_log_item *blip;
45 int len = 0;
46 int i;
47
48 for (i = 0; i < nmaps; i++)
49 len += map[i].bm_len;
50
51 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
52 blip = (struct xfs_buf_log_item *)lidp->lid_item;
53 if (blip->bli_item.li_type == XFS_LI_BUF &&
54 blip->bli_buf->b_target == target &&
55 XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn &&
56 blip->bli_buf->b_length == len) {
57 ASSERT(blip->bli_buf->b_map_count == nmaps);
58 return blip->bli_buf;
59 }
60 }
61
62 return NULL;
63}
64
65/*
66 * Add the locked buffer to the transaction.
67 *
68 * The buffer must be locked, and it cannot be associated with any
69 * transaction.
70 *
71 * If the buffer does not yet have a buf log item associated with it,
72 * then allocate one for it. Then add the buf item to the transaction.
73 */
74STATIC void
75_xfs_trans_bjoin(
76 struct xfs_trans *tp,
77 struct xfs_buf *bp,
78 int reset_recur)
79{
80 struct xfs_buf_log_item *bip;
81
82 ASSERT(bp->b_transp == NULL);
83
84 /*
85 * The xfs_buf_log_item pointer is stored in b_log_item. If
86 * it doesn't have one yet, then allocate one and initialize it.
87 * The checks to see if one is there are in xfs_buf_item_init().
88 */
89 xfs_buf_item_init(bp, tp->t_mountp);
90 bip = bp->b_log_item;
91 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
92 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
93 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
94 if (reset_recur)
95 bip->bli_recur = 0;
96
97 /*
98 * Take a reference for this transaction on the buf item.
99 */
100 atomic_inc(&bip->bli_refcount);
101
102 /*
103 * Get a log_item_desc to point at the new item.
104 */
105 xfs_trans_add_item(tp, &bip->bli_item);
106
107 /*
108 * Initialize b_fsprivate2 so we can find it with incore_match()
109 * in xfs_trans_get_buf() and friends above.
110 */
111 bp->b_transp = tp;
112
113}
114
115void
116xfs_trans_bjoin(
117 struct xfs_trans *tp,
118 struct xfs_buf *bp)
119{
120 _xfs_trans_bjoin(tp, bp, 0);
121 trace_xfs_trans_bjoin(bp->b_log_item);
122}
123
124/*
125 * Get and lock the buffer for the caller if it is not already
126 * locked within the given transaction. If it is already locked
127 * within the transaction, just increment its lock recursion count
128 * and return a pointer to it.
129 *
130 * If the transaction pointer is NULL, make this just a normal
131 * get_buf() call.
132 */
133struct xfs_buf *
134xfs_trans_get_buf_map(
135 struct xfs_trans *tp,
136 struct xfs_buftarg *target,
137 struct xfs_buf_map *map,
138 int nmaps,
139 xfs_buf_flags_t flags)
140{
141 xfs_buf_t *bp;
142 struct xfs_buf_log_item *bip;
143
144 if (!tp)
145 return xfs_buf_get_map(target, map, nmaps, flags);
146
147 /*
148 * If we find the buffer in the cache with this transaction
149 * pointer in its b_fsprivate2 field, then we know we already
150 * have it locked. In this case we just increment the lock
151 * recursion count and return the buffer to the caller.
152 */
153 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
154 if (bp != NULL) {
155 ASSERT(xfs_buf_islocked(bp));
156 if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
157 xfs_buf_stale(bp);
158 bp->b_flags |= XBF_DONE;
159 }
160
161 ASSERT(bp->b_transp == tp);
162 bip = bp->b_log_item;
163 ASSERT(bip != NULL);
164 ASSERT(atomic_read(&bip->bli_refcount) > 0);
165 bip->bli_recur++;
166 trace_xfs_trans_get_buf_recur(bip);
167 return bp;
168 }
169
170 bp = xfs_buf_get_map(target, map, nmaps, flags);
171 if (bp == NULL) {
172 return NULL;
173 }
174
175 ASSERT(!bp->b_error);
176
177 _xfs_trans_bjoin(tp, bp, 1);
178 trace_xfs_trans_get_buf(bp->b_log_item);
179 return bp;
180}
181
182/*
183 * Get and lock the superblock buffer of this file system for the
184 * given transaction.
185 *
186 * We don't need to use incore_match() here, because the superblock
187 * buffer is a private buffer which we keep a pointer to in the
188 * mount structure.
189 */
190xfs_buf_t *
191xfs_trans_getsb(
192 xfs_trans_t *tp,
193 struct xfs_mount *mp,
194 int flags)
195{
196 xfs_buf_t *bp;
197 struct xfs_buf_log_item *bip;
198
199 /*
200 * Default to just trying to lock the superblock buffer
201 * if tp is NULL.
202 */
203 if (tp == NULL)
204 return xfs_getsb(mp, flags);
205
206 /*
207 * If the superblock buffer already has this transaction
208 * pointer in its b_fsprivate2 field, then we know we already
209 * have it locked. In this case we just increment the lock
210 * recursion count and return the buffer to the caller.
211 */
212 bp = mp->m_sb_bp;
213 if (bp->b_transp == tp) {
214 bip = bp->b_log_item;
215 ASSERT(bip != NULL);
216 ASSERT(atomic_read(&bip->bli_refcount) > 0);
217 bip->bli_recur++;
218 trace_xfs_trans_getsb_recur(bip);
219 return bp;
220 }
221
222 bp = xfs_getsb(mp, flags);
223 if (bp == NULL)
224 return NULL;
225
226 _xfs_trans_bjoin(tp, bp, 1);
227 trace_xfs_trans_getsb(bp->b_log_item);
228 return bp;
229}
230
231/*
232 * Get and lock the buffer for the caller if it is not already
233 * locked within the given transaction. If it has not yet been
234 * read in, read it from disk. If it is already locked
235 * within the transaction and already read in, just increment its
236 * lock recursion count and return a pointer to it.
237 *
238 * If the transaction pointer is NULL, make this just a normal
239 * read_buf() call.
240 */
241int
242xfs_trans_read_buf_map(
243 struct xfs_mount *mp,
244 struct xfs_trans *tp,
245 struct xfs_buftarg *target,
246 struct xfs_buf_map *map,
247 int nmaps,
248 xfs_buf_flags_t flags,
249 struct xfs_buf **bpp,
250 const struct xfs_buf_ops *ops)
251{
252 struct xfs_buf *bp = NULL;
253 struct xfs_buf_log_item *bip;
254 int error;
255
256 *bpp = NULL;
257 /*
258 * If we find the buffer in the cache with this transaction
259 * pointer in its b_fsprivate2 field, then we know we already
260 * have it locked. If it is already read in we just increment
261 * the lock recursion count and return the buffer to the caller.
262 * If the buffer is not yet read in, then we read it in, increment
263 * the lock recursion count, and return it to the caller.
264 */
265 if (tp)
266 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
267 if (bp) {
268 ASSERT(xfs_buf_islocked(bp));
269 ASSERT(bp->b_transp == tp);
270 ASSERT(bp->b_log_item != NULL);
271 ASSERT(!bp->b_error);
272 ASSERT(bp->b_flags & XBF_DONE);
273
274 /*
275 * We never locked this buf ourselves, so we shouldn't
276 * brelse it either. Just get out.
277 */
278 if (XFS_FORCED_SHUTDOWN(mp)) {
279 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
280 return -EIO;
281 }
282
283 bip = bp->b_log_item;
284 bip->bli_recur++;
285
286 ASSERT(atomic_read(&bip->bli_refcount) > 0);
287 trace_xfs_trans_read_buf_recur(bip);
288 *bpp = bp;
289 return 0;
290 }
291
292 bp = xfs_buf_read_map(target, map, nmaps, flags, ops);
293 if (!bp) {
294 if (!(flags & XBF_TRYLOCK))
295 return -ENOMEM;
296 return tp ? 0 : -EAGAIN;
297 }
298
299 /*
300 * If we've had a read error, then the contents of the buffer are
301 * invalid and should not be used. To ensure that a followup read tries
302 * to pull the buffer from disk again, we clear the XBF_DONE flag and
303 * mark the buffer stale. This ensures that anyone who has a current
304 * reference to the buffer will interpret it's contents correctly and
305 * future cache lookups will also treat it as an empty, uninitialised
306 * buffer.
307 */
308 if (bp->b_error) {
309 error = bp->b_error;
310 if (!XFS_FORCED_SHUTDOWN(mp))
311 xfs_buf_ioerror_alert(bp, __func__);
312 bp->b_flags &= ~XBF_DONE;
313 xfs_buf_stale(bp);
314
315 if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
316 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
317 xfs_buf_relse(bp);
318
319 /* bad CRC means corrupted metadata */
320 if (error == -EFSBADCRC)
321 error = -EFSCORRUPTED;
322 return error;
323 }
324
325 if (XFS_FORCED_SHUTDOWN(mp)) {
326 xfs_buf_relse(bp);
327 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
328 return -EIO;
329 }
330
331 if (tp) {
332 _xfs_trans_bjoin(tp, bp, 1);
333 trace_xfs_trans_read_buf(bp->b_log_item);
334 }
335 *bpp = bp;
336 return 0;
337
338}
339
340/*
341 * Release the buffer bp which was previously acquired with one of the
342 * xfs_trans_... buffer allocation routines if the buffer has not
343 * been modified within this transaction. If the buffer is modified
344 * within this transaction, do decrement the recursion count but do
345 * not release the buffer even if the count goes to 0. If the buffer is not
346 * modified within the transaction, decrement the recursion count and
347 * release the buffer if the recursion count goes to 0.
348 *
349 * If the buffer is to be released and it was not modified before
350 * this transaction began, then free the buf_log_item associated with it.
351 *
352 * If the transaction pointer is NULL, make this just a normal
353 * brelse() call.
354 */
355void
356xfs_trans_brelse(
357 xfs_trans_t *tp,
358 xfs_buf_t *bp)
359{
360 struct xfs_buf_log_item *bip;
361 int freed;
362
363 /*
364 * Default to a normal brelse() call if the tp is NULL.
365 */
366 if (tp == NULL) {
367 ASSERT(bp->b_transp == NULL);
368 xfs_buf_relse(bp);
369 return;
370 }
371
372 ASSERT(bp->b_transp == tp);
373 bip = bp->b_log_item;
374 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
375 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
376 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
377 ASSERT(atomic_read(&bip->bli_refcount) > 0);
378
379 trace_xfs_trans_brelse(bip);
380
381 /*
382 * If the release is just for a recursive lock,
383 * then decrement the count and return.
384 */
385 if (bip->bli_recur > 0) {
386 bip->bli_recur--;
387 return;
388 }
389
390 /*
391 * If the buffer is dirty within this transaction, we can't
392 * release it until we commit.
393 */
394 if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
395 return;
396
397 /*
398 * If the buffer has been invalidated, then we can't release
399 * it until the transaction commits to disk unless it is re-dirtied
400 * as part of this transaction. This prevents us from pulling
401 * the item from the AIL before we should.
402 */
403 if (bip->bli_flags & XFS_BLI_STALE)
404 return;
405
406 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
407
408 /*
409 * Free up the log item descriptor tracking the released item.
410 */
411 xfs_trans_del_item(&bip->bli_item);
412
413 /*
414 * Clear the hold flag in the buf log item if it is set.
415 * We wouldn't want the next user of the buffer to
416 * get confused.
417 */
418 if (bip->bli_flags & XFS_BLI_HOLD) {
419 bip->bli_flags &= ~XFS_BLI_HOLD;
420 }
421
422 /*
423 * Drop our reference to the buf log item.
424 */
425 freed = atomic_dec_and_test(&bip->bli_refcount);
426
427 /*
428 * If the buf item is not tracking data in the log, then we must free it
429 * before releasing the buffer back to the free pool.
430 *
431 * If the fs has shutdown and we dropped the last reference, it may fall
432 * on us to release a (possibly dirty) bli if it never made it to the
433 * AIL (e.g., the aborted unpin already happened and didn't release it
434 * due to our reference). Since we're already shutdown and need
435 * ail_lock, just force remove from the AIL and release the bli here.
436 */
437 if (XFS_FORCED_SHUTDOWN(tp->t_mountp) && freed) {
438 xfs_trans_ail_remove(&bip->bli_item, SHUTDOWN_LOG_IO_ERROR);
439 xfs_buf_item_relse(bp);
440 } else if (!(bip->bli_flags & XFS_BLI_DIRTY)) {
441/***
442 ASSERT(bp->b_pincount == 0);
443***/
444 ASSERT(atomic_read(&bip->bli_refcount) == 0);
445 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
446 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
447 xfs_buf_item_relse(bp);
448 }
449
450 bp->b_transp = NULL;
451 xfs_buf_relse(bp);
452}
453
454/*
455 * Mark the buffer as not needing to be unlocked when the buf item's
456 * iop_unlock() routine is called. The buffer must already be locked
457 * and associated with the given transaction.
458 */
459/* ARGSUSED */
460void
461xfs_trans_bhold(
462 xfs_trans_t *tp,
463 xfs_buf_t *bp)
464{
465 struct xfs_buf_log_item *bip = bp->b_log_item;
466
467 ASSERT(bp->b_transp == tp);
468 ASSERT(bip != NULL);
469 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
470 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
471 ASSERT(atomic_read(&bip->bli_refcount) > 0);
472
473 bip->bli_flags |= XFS_BLI_HOLD;
474 trace_xfs_trans_bhold(bip);
475}
476
477/*
478 * Cancel the previous buffer hold request made on this buffer
479 * for this transaction.
480 */
481void
482xfs_trans_bhold_release(
483 xfs_trans_t *tp,
484 xfs_buf_t *bp)
485{
486 struct xfs_buf_log_item *bip = bp->b_log_item;
487
488 ASSERT(bp->b_transp == tp);
489 ASSERT(bip != NULL);
490 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
491 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
492 ASSERT(atomic_read(&bip->bli_refcount) > 0);
493 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
494
495 bip->bli_flags &= ~XFS_BLI_HOLD;
496 trace_xfs_trans_bhold_release(bip);
497}
498
499/*
500 * Mark a buffer dirty in the transaction.
501 */
502void
503xfs_trans_dirty_buf(
504 struct xfs_trans *tp,
505 struct xfs_buf *bp)
506{
507 struct xfs_buf_log_item *bip = bp->b_log_item;
508
509 ASSERT(bp->b_transp == tp);
510 ASSERT(bip != NULL);
511 ASSERT(bp->b_iodone == NULL ||
512 bp->b_iodone == xfs_buf_iodone_callbacks);
513
514 /*
515 * Mark the buffer as needing to be written out eventually,
516 * and set its iodone function to remove the buffer's buf log
517 * item from the AIL and free it when the buffer is flushed
518 * to disk. See xfs_buf_attach_iodone() for more details
519 * on li_cb and xfs_buf_iodone_callbacks().
520 * If we end up aborting this transaction, we trap this buffer
521 * inside the b_bdstrat callback so that this won't get written to
522 * disk.
523 */
524 bp->b_flags |= XBF_DONE;
525
526 ASSERT(atomic_read(&bip->bli_refcount) > 0);
527 bp->b_iodone = xfs_buf_iodone_callbacks;
528 bip->bli_item.li_cb = xfs_buf_iodone;
529
530 /*
531 * If we invalidated the buffer within this transaction, then
532 * cancel the invalidation now that we're dirtying the buffer
533 * again. There are no races with the code in xfs_buf_item_unpin(),
534 * because we have a reference to the buffer this entire time.
535 */
536 if (bip->bli_flags & XFS_BLI_STALE) {
537 bip->bli_flags &= ~XFS_BLI_STALE;
538 ASSERT(bp->b_flags & XBF_STALE);
539 bp->b_flags &= ~XBF_STALE;
540 bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
541 }
542 bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
543
544 tp->t_flags |= XFS_TRANS_DIRTY;
545 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
546}
547
548/*
549 * This is called to mark bytes first through last inclusive of the given
550 * buffer as needing to be logged when the transaction is committed.
551 * The buffer must already be associated with the given transaction.
552 *
553 * First and last are numbers relative to the beginning of this buffer,
554 * so the first byte in the buffer is numbered 0 regardless of the
555 * value of b_blkno.
556 */
557void
558xfs_trans_log_buf(
559 struct xfs_trans *tp,
560 struct xfs_buf *bp,
561 uint first,
562 uint last)
563{
564 struct xfs_buf_log_item *bip = bp->b_log_item;
565
566 ASSERT(first <= last && last < BBTOB(bp->b_length));
567 ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
568
569 xfs_trans_dirty_buf(tp, bp);
570
571 trace_xfs_trans_log_buf(bip);
572 xfs_buf_item_log(bip, first, last);
573}
574
575
576/*
577 * Invalidate a buffer that is being used within a transaction.
578 *
579 * Typically this is because the blocks in the buffer are being freed, so we
580 * need to prevent it from being written out when we're done. Allowing it
581 * to be written again might overwrite data in the free blocks if they are
582 * reallocated to a file.
583 *
584 * We prevent the buffer from being written out by marking it stale. We can't
585 * get rid of the buf log item at this point because the buffer may still be
586 * pinned by another transaction. If that is the case, then we'll wait until
587 * the buffer is committed to disk for the last time (we can tell by the ref
588 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
589 * keep the buffer locked so that the buffer and buf log item are not reused.
590 *
591 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
592 * the buf item. This will be used at recovery time to determine that copies
593 * of the buffer in the log before this should not be replayed.
594 *
595 * We mark the item descriptor and the transaction dirty so that we'll hold
596 * the buffer until after the commit.
597 *
598 * Since we're invalidating the buffer, we also clear the state about which
599 * parts of the buffer have been logged. We also clear the flag indicating
600 * that this is an inode buffer since the data in the buffer will no longer
601 * be valid.
602 *
603 * We set the stale bit in the buffer as well since we're getting rid of it.
604 */
605void
606xfs_trans_binval(
607 xfs_trans_t *tp,
608 xfs_buf_t *bp)
609{
610 struct xfs_buf_log_item *bip = bp->b_log_item;
611 int i;
612
613 ASSERT(bp->b_transp == tp);
614 ASSERT(bip != NULL);
615 ASSERT(atomic_read(&bip->bli_refcount) > 0);
616
617 trace_xfs_trans_binval(bip);
618
619 if (bip->bli_flags & XFS_BLI_STALE) {
620 /*
621 * If the buffer is already invalidated, then
622 * just return.
623 */
624 ASSERT(bp->b_flags & XBF_STALE);
625 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
626 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
627 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
628 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
629 ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
630 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
631 return;
632 }
633
634 xfs_buf_stale(bp);
635
636 bip->bli_flags |= XFS_BLI_STALE;
637 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
638 bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
639 bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
640 bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
641 for (i = 0; i < bip->bli_format_count; i++) {
642 memset(bip->bli_formats[i].blf_data_map, 0,
643 (bip->bli_formats[i].blf_map_size * sizeof(uint)));
644 }
645 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
646 tp->t_flags |= XFS_TRANS_DIRTY;
647}
648
649/*
650 * This call is used to indicate that the buffer contains on-disk inodes which
651 * must be handled specially during recovery. They require special handling
652 * because only the di_next_unlinked from the inodes in the buffer should be
653 * recovered. The rest of the data in the buffer is logged via the inodes
654 * themselves.
655 *
656 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
657 * transferred to the buffer's log format structure so that we'll know what to
658 * do at recovery time.
659 */
660void
661xfs_trans_inode_buf(
662 xfs_trans_t *tp,
663 xfs_buf_t *bp)
664{
665 struct xfs_buf_log_item *bip = bp->b_log_item;
666
667 ASSERT(bp->b_transp == tp);
668 ASSERT(bip != NULL);
669 ASSERT(atomic_read(&bip->bli_refcount) > 0);
670
671 bip->bli_flags |= XFS_BLI_INODE_BUF;
672 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
673}
674
675/*
676 * This call is used to indicate that the buffer is going to
677 * be staled and was an inode buffer. This means it gets
678 * special processing during unpin - where any inodes
679 * associated with the buffer should be removed from ail.
680 * There is also special processing during recovery,
681 * any replay of the inodes in the buffer needs to be
682 * prevented as the buffer may have been reused.
683 */
684void
685xfs_trans_stale_inode_buf(
686 xfs_trans_t *tp,
687 xfs_buf_t *bp)
688{
689 struct xfs_buf_log_item *bip = bp->b_log_item;
690
691 ASSERT(bp->b_transp == tp);
692 ASSERT(bip != NULL);
693 ASSERT(atomic_read(&bip->bli_refcount) > 0);
694
695 bip->bli_flags |= XFS_BLI_STALE_INODE;
696 bip->bli_item.li_cb = xfs_buf_iodone;
697 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
698}
699
700/*
701 * Mark the buffer as being one which contains newly allocated
702 * inodes. We need to make sure that even if this buffer is
703 * relogged as an 'inode buf' we still recover all of the inode
704 * images in the face of a crash. This works in coordination with
705 * xfs_buf_item_committed() to ensure that the buffer remains in the
706 * AIL at its original location even after it has been relogged.
707 */
708/* ARGSUSED */
709void
710xfs_trans_inode_alloc_buf(
711 xfs_trans_t *tp,
712 xfs_buf_t *bp)
713{
714 struct xfs_buf_log_item *bip = bp->b_log_item;
715
716 ASSERT(bp->b_transp == tp);
717 ASSERT(bip != NULL);
718 ASSERT(atomic_read(&bip->bli_refcount) > 0);
719
720 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
721 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
722}
723
724/*
725 * Mark the buffer as ordered for this transaction. This means that the contents
726 * of the buffer are not recorded in the transaction but it is tracked in the
727 * AIL as though it was. This allows us to record logical changes in
728 * transactions rather than the physical changes we make to the buffer without
729 * changing writeback ordering constraints of metadata buffers.
730 */
731bool
732xfs_trans_ordered_buf(
733 struct xfs_trans *tp,
734 struct xfs_buf *bp)
735{
736 struct xfs_buf_log_item *bip = bp->b_log_item;
737
738 ASSERT(bp->b_transp == tp);
739 ASSERT(bip != NULL);
740 ASSERT(atomic_read(&bip->bli_refcount) > 0);
741
742 if (xfs_buf_item_dirty_format(bip))
743 return false;
744
745 bip->bli_flags |= XFS_BLI_ORDERED;
746 trace_xfs_buf_item_ordered(bip);
747
748 /*
749 * We don't log a dirty range of an ordered buffer but it still needs
750 * to be marked dirty and that it has been logged.
751 */
752 xfs_trans_dirty_buf(tp, bp);
753 return true;
754}
755
756/*
757 * Set the type of the buffer for log recovery so that it can correctly identify
758 * and hence attach the correct buffer ops to the buffer after replay.
759 */
760void
761xfs_trans_buf_set_type(
762 struct xfs_trans *tp,
763 struct xfs_buf *bp,
764 enum xfs_blft type)
765{
766 struct xfs_buf_log_item *bip = bp->b_log_item;
767
768 if (!tp)
769 return;
770
771 ASSERT(bp->b_transp == tp);
772 ASSERT(bip != NULL);
773 ASSERT(atomic_read(&bip->bli_refcount) > 0);
774
775 xfs_blft_to_flags(&bip->__bli_format, type);
776}
777
778void
779xfs_trans_buf_copy_type(
780 struct xfs_buf *dst_bp,
781 struct xfs_buf *src_bp)
782{
783 struct xfs_buf_log_item *sbip = src_bp->b_log_item;
784 struct xfs_buf_log_item *dbip = dst_bp->b_log_item;
785 enum xfs_blft type;
786
787 type = xfs_blft_from_flags(&sbip->__bli_format);
788 xfs_blft_to_flags(&dbip->__bli_format, type);
789}
790
791/*
792 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
793 * dquots. However, unlike in inode buffer recovery, dquot buffers get
794 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
795 * The only thing that makes dquot buffers different from regular
796 * buffers is that we must not replay dquot bufs when recovering
797 * if a _corresponding_ quotaoff has happened. We also have to distinguish
798 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
799 * can be turned off independently.
800 */
801/* ARGSUSED */
802void
803xfs_trans_dquot_buf(
804 xfs_trans_t *tp,
805 xfs_buf_t *bp,
806 uint type)
807{
808 struct xfs_buf_log_item *bip = bp->b_log_item;
809
810 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
811 type == XFS_BLF_PDQUOT_BUF ||
812 type == XFS_BLF_GDQUOT_BUF);
813
814 bip->__bli_format.blf_flags |= type;
815
816 switch (type) {
817 case XFS_BLF_UDQUOT_BUF:
818 type = XFS_BLFT_UDQUOT_BUF;
819 break;
820 case XFS_BLF_PDQUOT_BUF:
821 type = XFS_BLFT_PDQUOT_BUF;
822 break;
823 case XFS_BLF_GDQUOT_BUF:
824 type = XFS_BLFT_GDQUOT_BUF;
825 break;
826 default:
827 type = XFS_BLFT_UNKNOWN_BUF;
828 break;
829 }
830
831 xfs_trans_buf_set_type(tp, bp, type);
832}
1/*
2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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.
8 *
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.
13 *
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
17 */
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_bmap_btree.h"
29#include "xfs_alloc_btree.h"
30#include "xfs_ialloc_btree.h"
31#include "xfs_dinode.h"
32#include "xfs_inode.h"
33#include "xfs_buf_item.h"
34#include "xfs_trans_priv.h"
35#include "xfs_error.h"
36#include "xfs_rw.h"
37#include "xfs_trace.h"
38
39/*
40 * Check to see if a buffer matching the given parameters is already
41 * a part of the given transaction.
42 */
43STATIC struct xfs_buf *
44xfs_trans_buf_item_match(
45 struct xfs_trans *tp,
46 struct xfs_buftarg *target,
47 xfs_daddr_t blkno,
48 int len)
49{
50 struct xfs_log_item_desc *lidp;
51 struct xfs_buf_log_item *blip;
52
53 len = BBTOB(len);
54 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
55 blip = (struct xfs_buf_log_item *)lidp->lid_item;
56 if (blip->bli_item.li_type == XFS_LI_BUF &&
57 blip->bli_buf->b_target == target &&
58 XFS_BUF_ADDR(blip->bli_buf) == blkno &&
59 XFS_BUF_COUNT(blip->bli_buf) == len)
60 return blip->bli_buf;
61 }
62
63 return NULL;
64}
65
66/*
67 * Add the locked buffer to the transaction.
68 *
69 * The buffer must be locked, and it cannot be associated with any
70 * transaction.
71 *
72 * If the buffer does not yet have a buf log item associated with it,
73 * then allocate one for it. Then add the buf item to the transaction.
74 */
75STATIC void
76_xfs_trans_bjoin(
77 struct xfs_trans *tp,
78 struct xfs_buf *bp,
79 int reset_recur)
80{
81 struct xfs_buf_log_item *bip;
82
83 ASSERT(bp->b_transp == NULL);
84
85 /*
86 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
87 * it doesn't have one yet, then allocate one and initialize it.
88 * The checks to see if one is there are in xfs_buf_item_init().
89 */
90 xfs_buf_item_init(bp, tp->t_mountp);
91 bip = bp->b_fspriv;
92 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
93 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
94 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
95 if (reset_recur)
96 bip->bli_recur = 0;
97
98 /*
99 * Take a reference for this transaction on the buf item.
100 */
101 atomic_inc(&bip->bli_refcount);
102
103 /*
104 * Get a log_item_desc to point at the new item.
105 */
106 xfs_trans_add_item(tp, &bip->bli_item);
107
108 /*
109 * Initialize b_fsprivate2 so we can find it with incore_match()
110 * in xfs_trans_get_buf() and friends above.
111 */
112 bp->b_transp = tp;
113
114}
115
116void
117xfs_trans_bjoin(
118 struct xfs_trans *tp,
119 struct xfs_buf *bp)
120{
121 _xfs_trans_bjoin(tp, bp, 0);
122 trace_xfs_trans_bjoin(bp->b_fspriv);
123}
124
125/*
126 * Get and lock the buffer for the caller if it is not already
127 * locked within the given transaction. If it is already locked
128 * within the transaction, just increment its lock recursion count
129 * and return a pointer to it.
130 *
131 * If the transaction pointer is NULL, make this just a normal
132 * get_buf() call.
133 */
134xfs_buf_t *
135xfs_trans_get_buf(xfs_trans_t *tp,
136 xfs_buftarg_t *target_dev,
137 xfs_daddr_t blkno,
138 int len,
139 uint flags)
140{
141 xfs_buf_t *bp;
142 xfs_buf_log_item_t *bip;
143
144 if (flags == 0)
145 flags = XBF_LOCK | XBF_MAPPED;
146
147 /*
148 * Default to a normal get_buf() call if the tp is NULL.
149 */
150 if (tp == NULL)
151 return xfs_buf_get(target_dev, blkno, len,
152 flags | XBF_DONT_BLOCK);
153
154 /*
155 * If we find the buffer in the cache with this transaction
156 * pointer in its b_fsprivate2 field, then we know we already
157 * have it locked. In this case we just increment the lock
158 * recursion count and return the buffer to the caller.
159 */
160 bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
161 if (bp != NULL) {
162 ASSERT(xfs_buf_islocked(bp));
163 if (XFS_FORCED_SHUTDOWN(tp->t_mountp))
164 XFS_BUF_SUPER_STALE(bp);
165
166 /*
167 * If the buffer is stale then it was binval'ed
168 * since last read. This doesn't matter since the
169 * caller isn't allowed to use the data anyway.
170 */
171 else if (XFS_BUF_ISSTALE(bp))
172 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
173
174 ASSERT(bp->b_transp == tp);
175 bip = bp->b_fspriv;
176 ASSERT(bip != NULL);
177 ASSERT(atomic_read(&bip->bli_refcount) > 0);
178 bip->bli_recur++;
179 trace_xfs_trans_get_buf_recur(bip);
180 return (bp);
181 }
182
183 /*
184 * We always specify the XBF_DONT_BLOCK flag within a transaction
185 * so that get_buf does not try to push out a delayed write buffer
186 * which might cause another transaction to take place (if the
187 * buffer was delayed alloc). Such recursive transactions can
188 * easily deadlock with our current transaction as well as cause
189 * us to run out of stack space.
190 */
191 bp = xfs_buf_get(target_dev, blkno, len, flags | XBF_DONT_BLOCK);
192 if (bp == NULL) {
193 return NULL;
194 }
195
196 ASSERT(!bp->b_error);
197
198 _xfs_trans_bjoin(tp, bp, 1);
199 trace_xfs_trans_get_buf(bp->b_fspriv);
200 return (bp);
201}
202
203/*
204 * Get and lock the superblock buffer of this file system for the
205 * given transaction.
206 *
207 * We don't need to use incore_match() here, because the superblock
208 * buffer is a private buffer which we keep a pointer to in the
209 * mount structure.
210 */
211xfs_buf_t *
212xfs_trans_getsb(xfs_trans_t *tp,
213 struct xfs_mount *mp,
214 int flags)
215{
216 xfs_buf_t *bp;
217 xfs_buf_log_item_t *bip;
218
219 /*
220 * Default to just trying to lock the superblock buffer
221 * if tp is NULL.
222 */
223 if (tp == NULL) {
224 return (xfs_getsb(mp, flags));
225 }
226
227 /*
228 * If the superblock buffer already has this transaction
229 * pointer in its b_fsprivate2 field, then we know we already
230 * have it locked. In this case we just increment the lock
231 * recursion count and return the buffer to the caller.
232 */
233 bp = mp->m_sb_bp;
234 if (bp->b_transp == tp) {
235 bip = bp->b_fspriv;
236 ASSERT(bip != NULL);
237 ASSERT(atomic_read(&bip->bli_refcount) > 0);
238 bip->bli_recur++;
239 trace_xfs_trans_getsb_recur(bip);
240 return (bp);
241 }
242
243 bp = xfs_getsb(mp, flags);
244 if (bp == NULL)
245 return NULL;
246
247 _xfs_trans_bjoin(tp, bp, 1);
248 trace_xfs_trans_getsb(bp->b_fspriv);
249 return (bp);
250}
251
252#ifdef DEBUG
253xfs_buftarg_t *xfs_error_target;
254int xfs_do_error;
255int xfs_req_num;
256int xfs_error_mod = 33;
257#endif
258
259/*
260 * Get and lock the buffer for the caller if it is not already
261 * locked within the given transaction. If it has not yet been
262 * read in, read it from disk. If it is already locked
263 * within the transaction and already read in, just increment its
264 * lock recursion count and return a pointer to it.
265 *
266 * If the transaction pointer is NULL, make this just a normal
267 * read_buf() call.
268 */
269int
270xfs_trans_read_buf(
271 xfs_mount_t *mp,
272 xfs_trans_t *tp,
273 xfs_buftarg_t *target,
274 xfs_daddr_t blkno,
275 int len,
276 uint flags,
277 xfs_buf_t **bpp)
278{
279 xfs_buf_t *bp;
280 xfs_buf_log_item_t *bip;
281 int error;
282
283 if (flags == 0)
284 flags = XBF_LOCK | XBF_MAPPED;
285
286 /*
287 * Default to a normal get_buf() call if the tp is NULL.
288 */
289 if (tp == NULL) {
290 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
291 if (!bp)
292 return (flags & XBF_TRYLOCK) ?
293 EAGAIN : XFS_ERROR(ENOMEM);
294
295 if (bp->b_error) {
296 error = bp->b_error;
297 xfs_ioerror_alert("xfs_trans_read_buf", mp,
298 bp, blkno);
299 xfs_buf_relse(bp);
300 return error;
301 }
302#ifdef DEBUG
303 if (xfs_do_error) {
304 if (xfs_error_target == target) {
305 if (((xfs_req_num++) % xfs_error_mod) == 0) {
306 xfs_buf_relse(bp);
307 xfs_debug(mp, "Returning error!");
308 return XFS_ERROR(EIO);
309 }
310 }
311 }
312#endif
313 if (XFS_FORCED_SHUTDOWN(mp))
314 goto shutdown_abort;
315 *bpp = bp;
316 return 0;
317 }
318
319 /*
320 * If we find the buffer in the cache with this transaction
321 * pointer in its b_fsprivate2 field, then we know we already
322 * have it locked. If it is already read in we just increment
323 * the lock recursion count and return the buffer to the caller.
324 * If the buffer is not yet read in, then we read it in, increment
325 * the lock recursion count, and return it to the caller.
326 */
327 bp = xfs_trans_buf_item_match(tp, target, blkno, len);
328 if (bp != NULL) {
329 ASSERT(xfs_buf_islocked(bp));
330 ASSERT(bp->b_transp == tp);
331 ASSERT(bp->b_fspriv != NULL);
332 ASSERT(!bp->b_error);
333 if (!(XFS_BUF_ISDONE(bp))) {
334 trace_xfs_trans_read_buf_io(bp, _RET_IP_);
335 ASSERT(!XFS_BUF_ISASYNC(bp));
336 XFS_BUF_READ(bp);
337 xfsbdstrat(tp->t_mountp, bp);
338 error = xfs_buf_iowait(bp);
339 if (error) {
340 xfs_ioerror_alert("xfs_trans_read_buf", mp,
341 bp, blkno);
342 xfs_buf_relse(bp);
343 /*
344 * We can gracefully recover from most read
345 * errors. Ones we can't are those that happen
346 * after the transaction's already dirty.
347 */
348 if (tp->t_flags & XFS_TRANS_DIRTY)
349 xfs_force_shutdown(tp->t_mountp,
350 SHUTDOWN_META_IO_ERROR);
351 return error;
352 }
353 }
354 /*
355 * We never locked this buf ourselves, so we shouldn't
356 * brelse it either. Just get out.
357 */
358 if (XFS_FORCED_SHUTDOWN(mp)) {
359 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
360 *bpp = NULL;
361 return XFS_ERROR(EIO);
362 }
363
364
365 bip = bp->b_fspriv;
366 bip->bli_recur++;
367
368 ASSERT(atomic_read(&bip->bli_refcount) > 0);
369 trace_xfs_trans_read_buf_recur(bip);
370 *bpp = bp;
371 return 0;
372 }
373
374 /*
375 * We always specify the XBF_DONT_BLOCK flag within a transaction
376 * so that get_buf does not try to push out a delayed write buffer
377 * which might cause another transaction to take place (if the
378 * buffer was delayed alloc). Such recursive transactions can
379 * easily deadlock with our current transaction as well as cause
380 * us to run out of stack space.
381 */
382 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
383 if (bp == NULL) {
384 *bpp = NULL;
385 return (flags & XBF_TRYLOCK) ?
386 0 : XFS_ERROR(ENOMEM);
387 }
388 if (bp->b_error) {
389 error = bp->b_error;
390 XFS_BUF_SUPER_STALE(bp);
391 xfs_ioerror_alert("xfs_trans_read_buf", mp,
392 bp, blkno);
393 if (tp->t_flags & XFS_TRANS_DIRTY)
394 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
395 xfs_buf_relse(bp);
396 return error;
397 }
398#ifdef DEBUG
399 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
400 if (xfs_error_target == target) {
401 if (((xfs_req_num++) % xfs_error_mod) == 0) {
402 xfs_force_shutdown(tp->t_mountp,
403 SHUTDOWN_META_IO_ERROR);
404 xfs_buf_relse(bp);
405 xfs_debug(mp, "Returning trans error!");
406 return XFS_ERROR(EIO);
407 }
408 }
409 }
410#endif
411 if (XFS_FORCED_SHUTDOWN(mp))
412 goto shutdown_abort;
413
414 _xfs_trans_bjoin(tp, bp, 1);
415 trace_xfs_trans_read_buf(bp->b_fspriv);
416
417 *bpp = bp;
418 return 0;
419
420shutdown_abort:
421 /*
422 * the theory here is that buffer is good but we're
423 * bailing out because the filesystem is being forcibly
424 * shut down. So we should leave the b_flags alone since
425 * the buffer's not staled and just get out.
426 */
427#if defined(DEBUG)
428 if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp))
429 xfs_notice(mp, "about to pop assert, bp == 0x%p", bp);
430#endif
431 ASSERT((bp->b_flags & (XBF_STALE|XBF_DELWRI)) !=
432 (XBF_STALE|XBF_DELWRI));
433
434 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
435 xfs_buf_relse(bp);
436 *bpp = NULL;
437 return XFS_ERROR(EIO);
438}
439
440
441/*
442 * Release the buffer bp which was previously acquired with one of the
443 * xfs_trans_... buffer allocation routines if the buffer has not
444 * been modified within this transaction. If the buffer is modified
445 * within this transaction, do decrement the recursion count but do
446 * not release the buffer even if the count goes to 0. If the buffer is not
447 * modified within the transaction, decrement the recursion count and
448 * release the buffer if the recursion count goes to 0.
449 *
450 * If the buffer is to be released and it was not modified before
451 * this transaction began, then free the buf_log_item associated with it.
452 *
453 * If the transaction pointer is NULL, make this just a normal
454 * brelse() call.
455 */
456void
457xfs_trans_brelse(xfs_trans_t *tp,
458 xfs_buf_t *bp)
459{
460 xfs_buf_log_item_t *bip;
461
462 /*
463 * Default to a normal brelse() call if the tp is NULL.
464 */
465 if (tp == NULL) {
466 struct xfs_log_item *lip = bp->b_fspriv;
467
468 ASSERT(bp->b_transp == NULL);
469
470 /*
471 * If there's a buf log item attached to the buffer,
472 * then let the AIL know that the buffer is being
473 * unlocked.
474 */
475 if (lip != NULL && lip->li_type == XFS_LI_BUF) {
476 bip = bp->b_fspriv;
477 xfs_trans_unlocked_item(bip->bli_item.li_ailp, lip);
478 }
479 xfs_buf_relse(bp);
480 return;
481 }
482
483 ASSERT(bp->b_transp == tp);
484 bip = bp->b_fspriv;
485 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
486 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
487 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
488 ASSERT(atomic_read(&bip->bli_refcount) > 0);
489
490 trace_xfs_trans_brelse(bip);
491
492 /*
493 * If the release is just for a recursive lock,
494 * then decrement the count and return.
495 */
496 if (bip->bli_recur > 0) {
497 bip->bli_recur--;
498 return;
499 }
500
501 /*
502 * If the buffer is dirty within this transaction, we can't
503 * release it until we commit.
504 */
505 if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
506 return;
507
508 /*
509 * If the buffer has been invalidated, then we can't release
510 * it until the transaction commits to disk unless it is re-dirtied
511 * as part of this transaction. This prevents us from pulling
512 * the item from the AIL before we should.
513 */
514 if (bip->bli_flags & XFS_BLI_STALE)
515 return;
516
517 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
518
519 /*
520 * Free up the log item descriptor tracking the released item.
521 */
522 xfs_trans_del_item(&bip->bli_item);
523
524 /*
525 * Clear the hold flag in the buf log item if it is set.
526 * We wouldn't want the next user of the buffer to
527 * get confused.
528 */
529 if (bip->bli_flags & XFS_BLI_HOLD) {
530 bip->bli_flags &= ~XFS_BLI_HOLD;
531 }
532
533 /*
534 * Drop our reference to the buf log item.
535 */
536 atomic_dec(&bip->bli_refcount);
537
538 /*
539 * If the buf item is not tracking data in the log, then
540 * we must free it before releasing the buffer back to the
541 * free pool. Before releasing the buffer to the free pool,
542 * clear the transaction pointer in b_fsprivate2 to dissolve
543 * its relation to this transaction.
544 */
545 if (!xfs_buf_item_dirty(bip)) {
546/***
547 ASSERT(bp->b_pincount == 0);
548***/
549 ASSERT(atomic_read(&bip->bli_refcount) == 0);
550 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
551 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
552 xfs_buf_item_relse(bp);
553 bip = NULL;
554 }
555 bp->b_transp = NULL;
556
557 /*
558 * If we've still got a buf log item on the buffer, then
559 * tell the AIL that the buffer is being unlocked.
560 */
561 if (bip != NULL) {
562 xfs_trans_unlocked_item(bip->bli_item.li_ailp,
563 (xfs_log_item_t*)bip);
564 }
565
566 xfs_buf_relse(bp);
567 return;
568}
569
570/*
571 * Mark the buffer as not needing to be unlocked when the buf item's
572 * IOP_UNLOCK() routine is called. The buffer must already be locked
573 * and associated with the given transaction.
574 */
575/* ARGSUSED */
576void
577xfs_trans_bhold(xfs_trans_t *tp,
578 xfs_buf_t *bp)
579{
580 xfs_buf_log_item_t *bip = bp->b_fspriv;
581
582 ASSERT(bp->b_transp == tp);
583 ASSERT(bip != NULL);
584 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
585 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
586 ASSERT(atomic_read(&bip->bli_refcount) > 0);
587
588 bip->bli_flags |= XFS_BLI_HOLD;
589 trace_xfs_trans_bhold(bip);
590}
591
592/*
593 * Cancel the previous buffer hold request made on this buffer
594 * for this transaction.
595 */
596void
597xfs_trans_bhold_release(xfs_trans_t *tp,
598 xfs_buf_t *bp)
599{
600 xfs_buf_log_item_t *bip = bp->b_fspriv;
601
602 ASSERT(bp->b_transp == tp);
603 ASSERT(bip != NULL);
604 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
605 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
606 ASSERT(atomic_read(&bip->bli_refcount) > 0);
607 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
608
609 bip->bli_flags &= ~XFS_BLI_HOLD;
610 trace_xfs_trans_bhold_release(bip);
611}
612
613/*
614 * This is called to mark bytes first through last inclusive of the given
615 * buffer as needing to be logged when the transaction is committed.
616 * The buffer must already be associated with the given transaction.
617 *
618 * First and last are numbers relative to the beginning of this buffer,
619 * so the first byte in the buffer is numbered 0 regardless of the
620 * value of b_blkno.
621 */
622void
623xfs_trans_log_buf(xfs_trans_t *tp,
624 xfs_buf_t *bp,
625 uint first,
626 uint last)
627{
628 xfs_buf_log_item_t *bip = bp->b_fspriv;
629
630 ASSERT(bp->b_transp == tp);
631 ASSERT(bip != NULL);
632 ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp)));
633 ASSERT(bp->b_iodone == NULL ||
634 bp->b_iodone == xfs_buf_iodone_callbacks);
635
636 /*
637 * Mark the buffer as needing to be written out eventually,
638 * and set its iodone function to remove the buffer's buf log
639 * item from the AIL and free it when the buffer is flushed
640 * to disk. See xfs_buf_attach_iodone() for more details
641 * on li_cb and xfs_buf_iodone_callbacks().
642 * If we end up aborting this transaction, we trap this buffer
643 * inside the b_bdstrat callback so that this won't get written to
644 * disk.
645 */
646 XFS_BUF_DELAYWRITE(bp);
647 XFS_BUF_DONE(bp);
648
649 ASSERT(atomic_read(&bip->bli_refcount) > 0);
650 bp->b_iodone = xfs_buf_iodone_callbacks;
651 bip->bli_item.li_cb = xfs_buf_iodone;
652
653 trace_xfs_trans_log_buf(bip);
654
655 /*
656 * If we invalidated the buffer within this transaction, then
657 * cancel the invalidation now that we're dirtying the buffer
658 * again. There are no races with the code in xfs_buf_item_unpin(),
659 * because we have a reference to the buffer this entire time.
660 */
661 if (bip->bli_flags & XFS_BLI_STALE) {
662 bip->bli_flags &= ~XFS_BLI_STALE;
663 ASSERT(XFS_BUF_ISSTALE(bp));
664 XFS_BUF_UNSTALE(bp);
665 bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL;
666 }
667
668 tp->t_flags |= XFS_TRANS_DIRTY;
669 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
670 bip->bli_flags |= XFS_BLI_LOGGED;
671 xfs_buf_item_log(bip, first, last);
672}
673
674
675/*
676 * This called to invalidate a buffer that is being used within
677 * a transaction. Typically this is because the blocks in the
678 * buffer are being freed, so we need to prevent it from being
679 * written out when we're done. Allowing it to be written again
680 * might overwrite data in the free blocks if they are reallocated
681 * to a file.
682 *
683 * We prevent the buffer from being written out by clearing the
684 * B_DELWRI flag. We can't always
685 * get rid of the buf log item at this point, though, because
686 * the buffer may still be pinned by another transaction. If that
687 * is the case, then we'll wait until the buffer is committed to
688 * disk for the last time (we can tell by the ref count) and
689 * free it in xfs_buf_item_unpin(). Until it is cleaned up we
690 * will keep the buffer locked so that the buffer and buf log item
691 * are not reused.
692 */
693void
694xfs_trans_binval(
695 xfs_trans_t *tp,
696 xfs_buf_t *bp)
697{
698 xfs_buf_log_item_t *bip = bp->b_fspriv;
699
700 ASSERT(bp->b_transp == tp);
701 ASSERT(bip != NULL);
702 ASSERT(atomic_read(&bip->bli_refcount) > 0);
703
704 trace_xfs_trans_binval(bip);
705
706 if (bip->bli_flags & XFS_BLI_STALE) {
707 /*
708 * If the buffer is already invalidated, then
709 * just return.
710 */
711 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
712 ASSERT(XFS_BUF_ISSTALE(bp));
713 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
714 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF));
715 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
716 ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
717 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
718 return;
719 }
720
721 /*
722 * Clear the dirty bit in the buffer and set the STALE flag
723 * in the buf log item. The STALE flag will be used in
724 * xfs_buf_item_unpin() to determine if it should clean up
725 * when the last reference to the buf item is given up.
726 * We set the XFS_BLF_CANCEL flag in the buf log format structure
727 * and log the buf item. This will be used at recovery time
728 * to determine that copies of the buffer in the log before
729 * this should not be replayed.
730 * We mark the item descriptor and the transaction dirty so
731 * that we'll hold the buffer until after the commit.
732 *
733 * Since we're invalidating the buffer, we also clear the state
734 * about which parts of the buffer have been logged. We also
735 * clear the flag indicating that this is an inode buffer since
736 * the data in the buffer will no longer be valid.
737 *
738 * We set the stale bit in the buffer as well since we're getting
739 * rid of it.
740 */
741 XFS_BUF_UNDELAYWRITE(bp);
742 XFS_BUF_STALE(bp);
743 bip->bli_flags |= XFS_BLI_STALE;
744 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
745 bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
746 bip->bli_format.blf_flags |= XFS_BLF_CANCEL;
747 memset((char *)(bip->bli_format.blf_data_map), 0,
748 (bip->bli_format.blf_map_size * sizeof(uint)));
749 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
750 tp->t_flags |= XFS_TRANS_DIRTY;
751}
752
753/*
754 * This call is used to indicate that the buffer contains on-disk inodes which
755 * must be handled specially during recovery. They require special handling
756 * because only the di_next_unlinked from the inodes in the buffer should be
757 * recovered. The rest of the data in the buffer is logged via the inodes
758 * themselves.
759 *
760 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
761 * transferred to the buffer's log format structure so that we'll know what to
762 * do at recovery time.
763 */
764void
765xfs_trans_inode_buf(
766 xfs_trans_t *tp,
767 xfs_buf_t *bp)
768{
769 xfs_buf_log_item_t *bip = bp->b_fspriv;
770
771 ASSERT(bp->b_transp == tp);
772 ASSERT(bip != NULL);
773 ASSERT(atomic_read(&bip->bli_refcount) > 0);
774
775 bip->bli_flags |= XFS_BLI_INODE_BUF;
776}
777
778/*
779 * This call is used to indicate that the buffer is going to
780 * be staled and was an inode buffer. This means it gets
781 * special processing during unpin - where any inodes
782 * associated with the buffer should be removed from ail.
783 * There is also special processing during recovery,
784 * any replay of the inodes in the buffer needs to be
785 * prevented as the buffer may have been reused.
786 */
787void
788xfs_trans_stale_inode_buf(
789 xfs_trans_t *tp,
790 xfs_buf_t *bp)
791{
792 xfs_buf_log_item_t *bip = bp->b_fspriv;
793
794 ASSERT(bp->b_transp == tp);
795 ASSERT(bip != NULL);
796 ASSERT(atomic_read(&bip->bli_refcount) > 0);
797
798 bip->bli_flags |= XFS_BLI_STALE_INODE;
799 bip->bli_item.li_cb = xfs_buf_iodone;
800}
801
802/*
803 * Mark the buffer as being one which contains newly allocated
804 * inodes. We need to make sure that even if this buffer is
805 * relogged as an 'inode buf' we still recover all of the inode
806 * images in the face of a crash. This works in coordination with
807 * xfs_buf_item_committed() to ensure that the buffer remains in the
808 * AIL at its original location even after it has been relogged.
809 */
810/* ARGSUSED */
811void
812xfs_trans_inode_alloc_buf(
813 xfs_trans_t *tp,
814 xfs_buf_t *bp)
815{
816 xfs_buf_log_item_t *bip = bp->b_fspriv;
817
818 ASSERT(bp->b_transp == tp);
819 ASSERT(bip != NULL);
820 ASSERT(atomic_read(&bip->bli_refcount) > 0);
821
822 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
823}
824
825
826/*
827 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
828 * dquots. However, unlike in inode buffer recovery, dquot buffers get
829 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
830 * The only thing that makes dquot buffers different from regular
831 * buffers is that we must not replay dquot bufs when recovering
832 * if a _corresponding_ quotaoff has happened. We also have to distinguish
833 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
834 * can be turned off independently.
835 */
836/* ARGSUSED */
837void
838xfs_trans_dquot_buf(
839 xfs_trans_t *tp,
840 xfs_buf_t *bp,
841 uint type)
842{
843 xfs_buf_log_item_t *bip = bp->b_fspriv;
844
845 ASSERT(bp->b_transp == tp);
846 ASSERT(bip != NULL);
847 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
848 type == XFS_BLF_PDQUOT_BUF ||
849 type == XFS_BLF_GDQUOT_BUF);
850 ASSERT(atomic_read(&bip->bli_refcount) > 0);
851
852 bip->bli_format.blf_flags |= type;
853}