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_sb.h"
25#include "xfs_ag.h"
26#include "xfs_mount.h"
27#include "xfs_inode.h"
28#include "xfs_trans.h"
29#include "xfs_buf_item.h"
30#include "xfs_trans_priv.h"
31#include "xfs_error.h"
32#include "xfs_trace.h"
33
34/*
35 * Check to see if a buffer matching the given parameters is already
36 * a part of the given transaction.
37 */
38STATIC struct xfs_buf *
39xfs_trans_buf_item_match(
40 struct xfs_trans *tp,
41 struct xfs_buftarg *target,
42 struct xfs_buf_map *map,
43 int nmaps)
44{
45 struct xfs_log_item_desc *lidp;
46 struct xfs_buf_log_item *blip;
47 int len = 0;
48 int i;
49
50 for (i = 0; i < nmaps; i++)
51 len += map[i].bm_len;
52
53 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
54 blip = (struct xfs_buf_log_item *)lidp->lid_item;
55 if (blip->bli_item.li_type == XFS_LI_BUF &&
56 blip->bli_buf->b_target == target &&
57 XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn &&
58 blip->bli_buf->b_length == len) {
59 ASSERT(blip->bli_buf->b_map_count == nmaps);
60 return blip->bli_buf;
61 }
62 }
63
64 return NULL;
65}
66
67/*
68 * Add the locked buffer to the transaction.
69 *
70 * The buffer must be locked, and it cannot be associated with any
71 * transaction.
72 *
73 * If the buffer does not yet have a buf log item associated with it,
74 * then allocate one for it. Then add the buf item to the transaction.
75 */
76STATIC void
77_xfs_trans_bjoin(
78 struct xfs_trans *tp,
79 struct xfs_buf *bp,
80 int reset_recur)
81{
82 struct xfs_buf_log_item *bip;
83
84 ASSERT(bp->b_transp == NULL);
85
86 /*
87 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
88 * it doesn't have one yet, then allocate one and initialize it.
89 * The checks to see if one is there are in xfs_buf_item_init().
90 */
91 xfs_buf_item_init(bp, tp->t_mountp);
92 bip = bp->b_fspriv;
93 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
94 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
95 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
96 if (reset_recur)
97 bip->bli_recur = 0;
98
99 /*
100 * Take a reference for this transaction on the buf item.
101 */
102 atomic_inc(&bip->bli_refcount);
103
104 /*
105 * Get a log_item_desc to point at the new item.
106 */
107 xfs_trans_add_item(tp, &bip->bli_item);
108
109 /*
110 * Initialize b_fsprivate2 so we can find it with incore_match()
111 * in xfs_trans_get_buf() and friends above.
112 */
113 bp->b_transp = tp;
114
115}
116
117void
118xfs_trans_bjoin(
119 struct xfs_trans *tp,
120 struct xfs_buf *bp)
121{
122 _xfs_trans_bjoin(tp, bp, 0);
123 trace_xfs_trans_bjoin(bp->b_fspriv);
124}
125
126/*
127 * Get and lock the buffer for the caller if it is not already
128 * locked within the given transaction. If it is already locked
129 * within the transaction, just increment its lock recursion count
130 * and return a pointer to it.
131 *
132 * If the transaction pointer is NULL, make this just a normal
133 * get_buf() call.
134 */
135struct xfs_buf *
136xfs_trans_get_buf_map(
137 struct xfs_trans *tp,
138 struct xfs_buftarg *target,
139 struct xfs_buf_map *map,
140 int nmaps,
141 xfs_buf_flags_t flags)
142{
143 xfs_buf_t *bp;
144 xfs_buf_log_item_t *bip;
145
146 if (!tp)
147 return xfs_buf_get_map(target, map, nmaps, flags);
148
149 /*
150 * If we find the buffer in the cache with this transaction
151 * pointer in its b_fsprivate2 field, then we know we already
152 * have it locked. In this case we just increment the lock
153 * recursion count and return the buffer to the caller.
154 */
155 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
156 if (bp != NULL) {
157 ASSERT(xfs_buf_islocked(bp));
158 if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
159 xfs_buf_stale(bp);
160 XFS_BUF_DONE(bp);
161 }
162
163 ASSERT(bp->b_transp == tp);
164 bip = bp->b_fspriv;
165 ASSERT(bip != NULL);
166 ASSERT(atomic_read(&bip->bli_refcount) > 0);
167 bip->bli_recur++;
168 trace_xfs_trans_get_buf_recur(bip);
169 return (bp);
170 }
171
172 bp = xfs_buf_get_map(target, map, nmaps, flags);
173 if (bp == NULL) {
174 return NULL;
175 }
176
177 ASSERT(!bp->b_error);
178
179 _xfs_trans_bjoin(tp, bp, 1);
180 trace_xfs_trans_get_buf(bp->b_fspriv);
181 return (bp);
182}
183
184/*
185 * Get and lock the superblock buffer of this file system for the
186 * given transaction.
187 *
188 * We don't need to use incore_match() here, because the superblock
189 * buffer is a private buffer which we keep a pointer to in the
190 * mount structure.
191 */
192xfs_buf_t *
193xfs_trans_getsb(xfs_trans_t *tp,
194 struct xfs_mount *mp,
195 int flags)
196{
197 xfs_buf_t *bp;
198 xfs_buf_log_item_t *bip;
199
200 /*
201 * Default to just trying to lock the superblock buffer
202 * if tp is NULL.
203 */
204 if (tp == NULL) {
205 return (xfs_getsb(mp, flags));
206 }
207
208 /*
209 * If the superblock buffer already has this transaction
210 * pointer in its b_fsprivate2 field, then we know we already
211 * have it locked. In this case we just increment the lock
212 * recursion count and return the buffer to the caller.
213 */
214 bp = mp->m_sb_bp;
215 if (bp->b_transp == tp) {
216 bip = bp->b_fspriv;
217 ASSERT(bip != NULL);
218 ASSERT(atomic_read(&bip->bli_refcount) > 0);
219 bip->bli_recur++;
220 trace_xfs_trans_getsb_recur(bip);
221 return (bp);
222 }
223
224 bp = xfs_getsb(mp, flags);
225 if (bp == NULL)
226 return NULL;
227
228 _xfs_trans_bjoin(tp, bp, 1);
229 trace_xfs_trans_getsb(bp->b_fspriv);
230 return (bp);
231}
232
233#ifdef DEBUG
234xfs_buftarg_t *xfs_error_target;
235int xfs_do_error;
236int xfs_req_num;
237int xfs_error_mod = 33;
238#endif
239
240/*
241 * Get and lock the buffer for the caller if it is not already
242 * locked within the given transaction. If it has not yet been
243 * read in, read it from disk. If it is already locked
244 * within the transaction and already read in, just increment its
245 * lock recursion count and return a pointer to it.
246 *
247 * If the transaction pointer is NULL, make this just a normal
248 * read_buf() call.
249 */
250int
251xfs_trans_read_buf_map(
252 struct xfs_mount *mp,
253 struct xfs_trans *tp,
254 struct xfs_buftarg *target,
255 struct xfs_buf_map *map,
256 int nmaps,
257 xfs_buf_flags_t flags,
258 struct xfs_buf **bpp,
259 const struct xfs_buf_ops *ops)
260{
261 xfs_buf_t *bp;
262 xfs_buf_log_item_t *bip;
263 int error;
264
265 *bpp = NULL;
266 if (!tp) {
267 bp = xfs_buf_read_map(target, map, nmaps, flags, ops);
268 if (!bp)
269 return (flags & XBF_TRYLOCK) ?
270 EAGAIN : XFS_ERROR(ENOMEM);
271
272 if (bp->b_error) {
273 error = bp->b_error;
274 xfs_buf_ioerror_alert(bp, __func__);
275 XFS_BUF_UNDONE(bp);
276 xfs_buf_stale(bp);
277 xfs_buf_relse(bp);
278
279 /* bad CRC means corrupted metadata */
280 if (error == EFSBADCRC)
281 error = EFSCORRUPTED;
282 return error;
283 }
284#ifdef DEBUG
285 if (xfs_do_error) {
286 if (xfs_error_target == target) {
287 if (((xfs_req_num++) % xfs_error_mod) == 0) {
288 xfs_buf_relse(bp);
289 xfs_debug(mp, "Returning error!");
290 return XFS_ERROR(EIO);
291 }
292 }
293 }
294#endif
295 if (XFS_FORCED_SHUTDOWN(mp))
296 goto shutdown_abort;
297 *bpp = bp;
298 return 0;
299 }
300
301 /*
302 * If we find the buffer in the cache with this transaction
303 * pointer in its b_fsprivate2 field, then we know we already
304 * have it locked. If it is already read in we just increment
305 * the lock recursion count and return the buffer to the caller.
306 * If the buffer is not yet read in, then we read it in, increment
307 * the lock recursion count, and return it to the caller.
308 */
309 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
310 if (bp != NULL) {
311 ASSERT(xfs_buf_islocked(bp));
312 ASSERT(bp->b_transp == tp);
313 ASSERT(bp->b_fspriv != NULL);
314 ASSERT(!bp->b_error);
315 if (!(XFS_BUF_ISDONE(bp))) {
316 trace_xfs_trans_read_buf_io(bp, _RET_IP_);
317 ASSERT(!XFS_BUF_ISASYNC(bp));
318 ASSERT(bp->b_iodone == NULL);
319 XFS_BUF_READ(bp);
320 bp->b_ops = ops;
321
322 /*
323 * XXX(hch): clean up the error handling here to be less
324 * of a mess..
325 */
326 if (XFS_FORCED_SHUTDOWN(mp)) {
327 trace_xfs_bdstrat_shut(bp, _RET_IP_);
328 xfs_bioerror_relse(bp);
329 } else {
330 xfs_buf_iorequest(bp);
331 }
332
333 error = xfs_buf_iowait(bp);
334 if (error) {
335 xfs_buf_ioerror_alert(bp, __func__);
336 xfs_buf_relse(bp);
337 /*
338 * We can gracefully recover from most read
339 * errors. Ones we can't are those that happen
340 * after the transaction's already dirty.
341 */
342 if (tp->t_flags & XFS_TRANS_DIRTY)
343 xfs_force_shutdown(tp->t_mountp,
344 SHUTDOWN_META_IO_ERROR);
345 /* bad CRC means corrupted metadata */
346 if (error == EFSBADCRC)
347 error = EFSCORRUPTED;
348 return error;
349 }
350 }
351 /*
352 * We never locked this buf ourselves, so we shouldn't
353 * brelse it either. Just get out.
354 */
355 if (XFS_FORCED_SHUTDOWN(mp)) {
356 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
357 *bpp = NULL;
358 return XFS_ERROR(EIO);
359 }
360
361
362 bip = bp->b_fspriv;
363 bip->bli_recur++;
364
365 ASSERT(atomic_read(&bip->bli_refcount) > 0);
366 trace_xfs_trans_read_buf_recur(bip);
367 *bpp = bp;
368 return 0;
369 }
370
371 bp = xfs_buf_read_map(target, map, nmaps, flags, ops);
372 if (bp == NULL) {
373 *bpp = NULL;
374 return (flags & XBF_TRYLOCK) ?
375 0 : XFS_ERROR(ENOMEM);
376 }
377 if (bp->b_error) {
378 error = bp->b_error;
379 xfs_buf_stale(bp);
380 XFS_BUF_DONE(bp);
381 xfs_buf_ioerror_alert(bp, __func__);
382 if (tp->t_flags & XFS_TRANS_DIRTY)
383 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
384 xfs_buf_relse(bp);
385
386 /* bad CRC means corrupted metadata */
387 if (error == EFSBADCRC)
388 error = EFSCORRUPTED;
389 return error;
390 }
391#ifdef DEBUG
392 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
393 if (xfs_error_target == target) {
394 if (((xfs_req_num++) % xfs_error_mod) == 0) {
395 xfs_force_shutdown(tp->t_mountp,
396 SHUTDOWN_META_IO_ERROR);
397 xfs_buf_relse(bp);
398 xfs_debug(mp, "Returning trans error!");
399 return XFS_ERROR(EIO);
400 }
401 }
402 }
403#endif
404 if (XFS_FORCED_SHUTDOWN(mp))
405 goto shutdown_abort;
406
407 _xfs_trans_bjoin(tp, bp, 1);
408 trace_xfs_trans_read_buf(bp->b_fspriv);
409
410 *bpp = bp;
411 return 0;
412
413shutdown_abort:
414 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
415 xfs_buf_relse(bp);
416 *bpp = NULL;
417 return XFS_ERROR(EIO);
418}
419
420/*
421 * Release the buffer bp which was previously acquired with one of the
422 * xfs_trans_... buffer allocation routines if the buffer has not
423 * been modified within this transaction. If the buffer is modified
424 * within this transaction, do decrement the recursion count but do
425 * not release the buffer even if the count goes to 0. If the buffer is not
426 * modified within the transaction, decrement the recursion count and
427 * release the buffer if the recursion count goes to 0.
428 *
429 * If the buffer is to be released and it was not modified before
430 * this transaction began, then free the buf_log_item associated with it.
431 *
432 * If the transaction pointer is NULL, make this just a normal
433 * brelse() call.
434 */
435void
436xfs_trans_brelse(xfs_trans_t *tp,
437 xfs_buf_t *bp)
438{
439 xfs_buf_log_item_t *bip;
440
441 /*
442 * Default to a normal brelse() call if the tp is NULL.
443 */
444 if (tp == NULL) {
445 ASSERT(bp->b_transp == NULL);
446 xfs_buf_relse(bp);
447 return;
448 }
449
450 ASSERT(bp->b_transp == tp);
451 bip = bp->b_fspriv;
452 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
453 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
454 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
455 ASSERT(atomic_read(&bip->bli_refcount) > 0);
456
457 trace_xfs_trans_brelse(bip);
458
459 /*
460 * If the release is just for a recursive lock,
461 * then decrement the count and return.
462 */
463 if (bip->bli_recur > 0) {
464 bip->bli_recur--;
465 return;
466 }
467
468 /*
469 * If the buffer is dirty within this transaction, we can't
470 * release it until we commit.
471 */
472 if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
473 return;
474
475 /*
476 * If the buffer has been invalidated, then we can't release
477 * it until the transaction commits to disk unless it is re-dirtied
478 * as part of this transaction. This prevents us from pulling
479 * the item from the AIL before we should.
480 */
481 if (bip->bli_flags & XFS_BLI_STALE)
482 return;
483
484 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
485
486 /*
487 * Free up the log item descriptor tracking the released item.
488 */
489 xfs_trans_del_item(&bip->bli_item);
490
491 /*
492 * Clear the hold flag in the buf log item if it is set.
493 * We wouldn't want the next user of the buffer to
494 * get confused.
495 */
496 if (bip->bli_flags & XFS_BLI_HOLD) {
497 bip->bli_flags &= ~XFS_BLI_HOLD;
498 }
499
500 /*
501 * Drop our reference to the buf log item.
502 */
503 atomic_dec(&bip->bli_refcount);
504
505 /*
506 * If the buf item is not tracking data in the log, then
507 * we must free it before releasing the buffer back to the
508 * free pool. Before releasing the buffer to the free pool,
509 * clear the transaction pointer in b_fsprivate2 to dissolve
510 * its relation to this transaction.
511 */
512 if (!xfs_buf_item_dirty(bip)) {
513/***
514 ASSERT(bp->b_pincount == 0);
515***/
516 ASSERT(atomic_read(&bip->bli_refcount) == 0);
517 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
518 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
519 xfs_buf_item_relse(bp);
520 }
521
522 bp->b_transp = NULL;
523 xfs_buf_relse(bp);
524}
525
526/*
527 * Mark the buffer as not needing to be unlocked when the buf item's
528 * iop_unlock() routine is called. The buffer must already be locked
529 * and associated with the given transaction.
530 */
531/* ARGSUSED */
532void
533xfs_trans_bhold(xfs_trans_t *tp,
534 xfs_buf_t *bp)
535{
536 xfs_buf_log_item_t *bip = bp->b_fspriv;
537
538 ASSERT(bp->b_transp == tp);
539 ASSERT(bip != NULL);
540 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
541 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
542 ASSERT(atomic_read(&bip->bli_refcount) > 0);
543
544 bip->bli_flags |= XFS_BLI_HOLD;
545 trace_xfs_trans_bhold(bip);
546}
547
548/*
549 * Cancel the previous buffer hold request made on this buffer
550 * for this transaction.
551 */
552void
553xfs_trans_bhold_release(xfs_trans_t *tp,
554 xfs_buf_t *bp)
555{
556 xfs_buf_log_item_t *bip = bp->b_fspriv;
557
558 ASSERT(bp->b_transp == tp);
559 ASSERT(bip != NULL);
560 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
561 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
562 ASSERT(atomic_read(&bip->bli_refcount) > 0);
563 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
564
565 bip->bli_flags &= ~XFS_BLI_HOLD;
566 trace_xfs_trans_bhold_release(bip);
567}
568
569/*
570 * This is called to mark bytes first through last inclusive of the given
571 * buffer as needing to be logged when the transaction is committed.
572 * The buffer must already be associated with the given transaction.
573 *
574 * First and last are numbers relative to the beginning of this buffer,
575 * so the first byte in the buffer is numbered 0 regardless of the
576 * value of b_blkno.
577 */
578void
579xfs_trans_log_buf(xfs_trans_t *tp,
580 xfs_buf_t *bp,
581 uint first,
582 uint last)
583{
584 xfs_buf_log_item_t *bip = bp->b_fspriv;
585
586 ASSERT(bp->b_transp == tp);
587 ASSERT(bip != NULL);
588 ASSERT(first <= last && last < BBTOB(bp->b_length));
589 ASSERT(bp->b_iodone == NULL ||
590 bp->b_iodone == xfs_buf_iodone_callbacks);
591
592 /*
593 * Mark the buffer as needing to be written out eventually,
594 * and set its iodone function to remove the buffer's buf log
595 * item from the AIL and free it when the buffer is flushed
596 * to disk. See xfs_buf_attach_iodone() for more details
597 * on li_cb and xfs_buf_iodone_callbacks().
598 * If we end up aborting this transaction, we trap this buffer
599 * inside the b_bdstrat callback so that this won't get written to
600 * disk.
601 */
602 XFS_BUF_DONE(bp);
603
604 ASSERT(atomic_read(&bip->bli_refcount) > 0);
605 bp->b_iodone = xfs_buf_iodone_callbacks;
606 bip->bli_item.li_cb = xfs_buf_iodone;
607
608 trace_xfs_trans_log_buf(bip);
609
610 /*
611 * If we invalidated the buffer within this transaction, then
612 * cancel the invalidation now that we're dirtying the buffer
613 * again. There are no races with the code in xfs_buf_item_unpin(),
614 * because we have a reference to the buffer this entire time.
615 */
616 if (bip->bli_flags & XFS_BLI_STALE) {
617 bip->bli_flags &= ~XFS_BLI_STALE;
618 ASSERT(XFS_BUF_ISSTALE(bp));
619 XFS_BUF_UNSTALE(bp);
620 bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
621 }
622
623 tp->t_flags |= XFS_TRANS_DIRTY;
624 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
625
626 /*
627 * If we have an ordered buffer we are not logging any dirty range but
628 * it still needs to be marked dirty and that it has been logged.
629 */
630 bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
631 if (!(bip->bli_flags & XFS_BLI_ORDERED))
632 xfs_buf_item_log(bip, first, last);
633}
634
635
636/*
637 * Invalidate a buffer that is being used within a transaction.
638 *
639 * Typically this is because the blocks in the buffer are being freed, so we
640 * need to prevent it from being written out when we're done. Allowing it
641 * to be written again might overwrite data in the free blocks if they are
642 * reallocated to a file.
643 *
644 * We prevent the buffer from being written out by marking it stale. We can't
645 * get rid of the buf log item at this point because the buffer may still be
646 * pinned by another transaction. If that is the case, then we'll wait until
647 * the buffer is committed to disk for the last time (we can tell by the ref
648 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
649 * keep the buffer locked so that the buffer and buf log item are not reused.
650 *
651 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
652 * the buf item. This will be used at recovery time to determine that copies
653 * of the buffer in the log before this should not be replayed.
654 *
655 * We mark the item descriptor and the transaction dirty so that we'll hold
656 * the buffer until after the commit.
657 *
658 * Since we're invalidating the buffer, we also clear the state about which
659 * parts of the buffer have been logged. We also clear the flag indicating
660 * that this is an inode buffer since the data in the buffer will no longer
661 * be valid.
662 *
663 * We set the stale bit in the buffer as well since we're getting rid of it.
664 */
665void
666xfs_trans_binval(
667 xfs_trans_t *tp,
668 xfs_buf_t *bp)
669{
670 xfs_buf_log_item_t *bip = bp->b_fspriv;
671 int i;
672
673 ASSERT(bp->b_transp == tp);
674 ASSERT(bip != NULL);
675 ASSERT(atomic_read(&bip->bli_refcount) > 0);
676
677 trace_xfs_trans_binval(bip);
678
679 if (bip->bli_flags & XFS_BLI_STALE) {
680 /*
681 * If the buffer is already invalidated, then
682 * just return.
683 */
684 ASSERT(XFS_BUF_ISSTALE(bp));
685 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
686 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
687 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
688 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
689 ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
690 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
691 return;
692 }
693
694 xfs_buf_stale(bp);
695
696 bip->bli_flags |= XFS_BLI_STALE;
697 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
698 bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
699 bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
700 bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
701 for (i = 0; i < bip->bli_format_count; i++) {
702 memset(bip->bli_formats[i].blf_data_map, 0,
703 (bip->bli_formats[i].blf_map_size * sizeof(uint)));
704 }
705 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
706 tp->t_flags |= XFS_TRANS_DIRTY;
707}
708
709/*
710 * This call is used to indicate that the buffer contains on-disk inodes which
711 * must be handled specially during recovery. They require special handling
712 * because only the di_next_unlinked from the inodes in the buffer should be
713 * recovered. The rest of the data in the buffer is logged via the inodes
714 * themselves.
715 *
716 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
717 * transferred to the buffer's log format structure so that we'll know what to
718 * do at recovery time.
719 */
720void
721xfs_trans_inode_buf(
722 xfs_trans_t *tp,
723 xfs_buf_t *bp)
724{
725 xfs_buf_log_item_t *bip = bp->b_fspriv;
726
727 ASSERT(bp->b_transp == tp);
728 ASSERT(bip != NULL);
729 ASSERT(atomic_read(&bip->bli_refcount) > 0);
730
731 bip->bli_flags |= XFS_BLI_INODE_BUF;
732 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
733}
734
735/*
736 * This call is used to indicate that the buffer is going to
737 * be staled and was an inode buffer. This means it gets
738 * special processing during unpin - where any inodes
739 * associated with the buffer should be removed from ail.
740 * There is also special processing during recovery,
741 * any replay of the inodes in the buffer needs to be
742 * prevented as the buffer may have been reused.
743 */
744void
745xfs_trans_stale_inode_buf(
746 xfs_trans_t *tp,
747 xfs_buf_t *bp)
748{
749 xfs_buf_log_item_t *bip = bp->b_fspriv;
750
751 ASSERT(bp->b_transp == tp);
752 ASSERT(bip != NULL);
753 ASSERT(atomic_read(&bip->bli_refcount) > 0);
754
755 bip->bli_flags |= XFS_BLI_STALE_INODE;
756 bip->bli_item.li_cb = xfs_buf_iodone;
757 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
758}
759
760/*
761 * Mark the buffer as being one which contains newly allocated
762 * inodes. We need to make sure that even if this buffer is
763 * relogged as an 'inode buf' we still recover all of the inode
764 * images in the face of a crash. This works in coordination with
765 * xfs_buf_item_committed() to ensure that the buffer remains in the
766 * AIL at its original location even after it has been relogged.
767 */
768/* ARGSUSED */
769void
770xfs_trans_inode_alloc_buf(
771 xfs_trans_t *tp,
772 xfs_buf_t *bp)
773{
774 xfs_buf_log_item_t *bip = bp->b_fspriv;
775
776 ASSERT(bp->b_transp == tp);
777 ASSERT(bip != NULL);
778 ASSERT(atomic_read(&bip->bli_refcount) > 0);
779
780 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
781 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
782}
783
784/*
785 * Mark the buffer as ordered for this transaction. This means
786 * that the contents of the buffer are not recorded in the transaction
787 * but it is tracked in the AIL as though it was. This allows us
788 * to record logical changes in transactions rather than the physical
789 * changes we make to the buffer without changing writeback ordering
790 * constraints of metadata buffers.
791 */
792void
793xfs_trans_ordered_buf(
794 struct xfs_trans *tp,
795 struct xfs_buf *bp)
796{
797 struct xfs_buf_log_item *bip = bp->b_fspriv;
798
799 ASSERT(bp->b_transp == tp);
800 ASSERT(bip != NULL);
801 ASSERT(atomic_read(&bip->bli_refcount) > 0);
802
803 bip->bli_flags |= XFS_BLI_ORDERED;
804 trace_xfs_buf_item_ordered(bip);
805}
806
807/*
808 * Set the type of the buffer for log recovery so that it can correctly identify
809 * and hence attach the correct buffer ops to the buffer after replay.
810 */
811void
812xfs_trans_buf_set_type(
813 struct xfs_trans *tp,
814 struct xfs_buf *bp,
815 enum xfs_blft type)
816{
817 struct xfs_buf_log_item *bip = bp->b_fspriv;
818
819 if (!tp)
820 return;
821
822 ASSERT(bp->b_transp == tp);
823 ASSERT(bip != NULL);
824 ASSERT(atomic_read(&bip->bli_refcount) > 0);
825
826 xfs_blft_to_flags(&bip->__bli_format, type);
827}
828
829void
830xfs_trans_buf_copy_type(
831 struct xfs_buf *dst_bp,
832 struct xfs_buf *src_bp)
833{
834 struct xfs_buf_log_item *sbip = src_bp->b_fspriv;
835 struct xfs_buf_log_item *dbip = dst_bp->b_fspriv;
836 enum xfs_blft type;
837
838 type = xfs_blft_from_flags(&sbip->__bli_format);
839 xfs_blft_to_flags(&dbip->__bli_format, type);
840}
841
842/*
843 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
844 * dquots. However, unlike in inode buffer recovery, dquot buffers get
845 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
846 * The only thing that makes dquot buffers different from regular
847 * buffers is that we must not replay dquot bufs when recovering
848 * if a _corresponding_ quotaoff has happened. We also have to distinguish
849 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
850 * can be turned off independently.
851 */
852/* ARGSUSED */
853void
854xfs_trans_dquot_buf(
855 xfs_trans_t *tp,
856 xfs_buf_t *bp,
857 uint type)
858{
859 struct xfs_buf_log_item *bip = bp->b_fspriv;
860
861 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
862 type == XFS_BLF_PDQUOT_BUF ||
863 type == XFS_BLF_GDQUOT_BUF);
864
865 bip->__bli_format.blf_flags |= type;
866
867 switch (type) {
868 case XFS_BLF_UDQUOT_BUF:
869 type = XFS_BLFT_UDQUOT_BUF;
870 break;
871 case XFS_BLF_PDQUOT_BUF:
872 type = XFS_BLFT_PDQUOT_BUF;
873 break;
874 case XFS_BLF_GDQUOT_BUF:
875 type = XFS_BLFT_GDQUOT_BUF;
876 break;
877 default:
878 type = XFS_BLFT_UNKNOWN_BUF;
879 break;
880 }
881
882 xfs_trans_buf_set_type(tp, bp, type);
883}
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}