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_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}