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// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_mount.h"
13#include "xfs_trans.h"
14#include "xfs_buf_item.h"
15#include "xfs_trans_priv.h"
16#include "xfs_trace.h"
17
18/*
19 * Check to see if a buffer matching the given parameters is already
20 * a part of the given transaction.
21 */
22STATIC struct xfs_buf *
23xfs_trans_buf_item_match(
24 struct xfs_trans *tp,
25 struct xfs_buftarg *target,
26 struct xfs_buf_map *map,
27 int nmaps)
28{
29 struct xfs_log_item *lip;
30 struct xfs_buf_log_item *blip;
31 int len = 0;
32 int i;
33
34 for (i = 0; i < nmaps; i++)
35 len += map[i].bm_len;
36
37 list_for_each_entry(lip, &tp->t_items, li_trans) {
38 blip = (struct xfs_buf_log_item *)lip;
39 if (blip->bli_item.li_type == XFS_LI_BUF &&
40 blip->bli_buf->b_target == target &&
41 XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn &&
42 blip->bli_buf->b_length == len) {
43 ASSERT(blip->bli_buf->b_map_count == nmaps);
44 return blip->bli_buf;
45 }
46 }
47
48 return NULL;
49}
50
51/*
52 * Add the locked buffer to the transaction.
53 *
54 * The buffer must be locked, and it cannot be associated with any
55 * transaction.
56 *
57 * If the buffer does not yet have a buf log item associated with it,
58 * then allocate one for it. Then add the buf item to the transaction.
59 */
60STATIC void
61_xfs_trans_bjoin(
62 struct xfs_trans *tp,
63 struct xfs_buf *bp,
64 int reset_recur)
65{
66 struct xfs_buf_log_item *bip;
67
68 ASSERT(bp->b_transp == NULL);
69
70 /*
71 * The xfs_buf_log_item pointer is stored in b_log_item. If
72 * it doesn't have one yet, then allocate one and initialize it.
73 * The checks to see if one is there are in xfs_buf_item_init().
74 */
75 xfs_buf_item_init(bp, tp->t_mountp);
76 bip = bp->b_log_item;
77 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
78 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
79 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
80 if (reset_recur)
81 bip->bli_recur = 0;
82
83 /*
84 * Take a reference for this transaction on the buf item.
85 */
86 atomic_inc(&bip->bli_refcount);
87
88 /*
89 * Attach the item to the transaction so we can find it in
90 * xfs_trans_get_buf() and friends.
91 */
92 xfs_trans_add_item(tp, &bip->bli_item);
93 bp->b_transp = tp;
94
95}
96
97void
98xfs_trans_bjoin(
99 struct xfs_trans *tp,
100 struct xfs_buf *bp)
101{
102 _xfs_trans_bjoin(tp, bp, 0);
103 trace_xfs_trans_bjoin(bp->b_log_item);
104}
105
106/*
107 * Get and lock the buffer for the caller if it is not already
108 * locked within the given transaction. If it is already locked
109 * within the transaction, just increment its lock recursion count
110 * and return a pointer to it.
111 *
112 * If the transaction pointer is NULL, make this just a normal
113 * get_buf() call.
114 */
115struct xfs_buf *
116xfs_trans_get_buf_map(
117 struct xfs_trans *tp,
118 struct xfs_buftarg *target,
119 struct xfs_buf_map *map,
120 int nmaps,
121 xfs_buf_flags_t flags)
122{
123 xfs_buf_t *bp;
124 struct xfs_buf_log_item *bip;
125
126 if (!tp)
127 return xfs_buf_get_map(target, map, nmaps, flags);
128
129 /*
130 * If we find the buffer in the cache with this transaction
131 * pointer in its b_fsprivate2 field, then we know we already
132 * have it locked. In this case we just increment the lock
133 * recursion count and return the buffer to the caller.
134 */
135 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
136 if (bp != NULL) {
137 ASSERT(xfs_buf_islocked(bp));
138 if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
139 xfs_buf_stale(bp);
140 bp->b_flags |= XBF_DONE;
141 }
142
143 ASSERT(bp->b_transp == tp);
144 bip = bp->b_log_item;
145 ASSERT(bip != NULL);
146 ASSERT(atomic_read(&bip->bli_refcount) > 0);
147 bip->bli_recur++;
148 trace_xfs_trans_get_buf_recur(bip);
149 return bp;
150 }
151
152 bp = xfs_buf_get_map(target, map, nmaps, flags);
153 if (bp == NULL) {
154 return NULL;
155 }
156
157 ASSERT(!bp->b_error);
158
159 _xfs_trans_bjoin(tp, bp, 1);
160 trace_xfs_trans_get_buf(bp->b_log_item);
161 return bp;
162}
163
164/*
165 * Get and lock the superblock buffer of this file system for the
166 * given transaction.
167 *
168 * We don't need to use incore_match() here, because the superblock
169 * buffer is a private buffer which we keep a pointer to in the
170 * mount structure.
171 */
172xfs_buf_t *
173xfs_trans_getsb(
174 xfs_trans_t *tp,
175 struct xfs_mount *mp)
176{
177 xfs_buf_t *bp;
178 struct xfs_buf_log_item *bip;
179
180 /*
181 * Default to just trying to lock the superblock buffer
182 * if tp is NULL.
183 */
184 if (tp == NULL)
185 return xfs_getsb(mp);
186
187 /*
188 * If the superblock buffer already has this transaction
189 * pointer in its b_fsprivate2 field, then we know we already
190 * have it locked. In this case we just increment the lock
191 * recursion count and return the buffer to the caller.
192 */
193 bp = mp->m_sb_bp;
194 if (bp->b_transp == tp) {
195 bip = bp->b_log_item;
196 ASSERT(bip != NULL);
197 ASSERT(atomic_read(&bip->bli_refcount) > 0);
198 bip->bli_recur++;
199 trace_xfs_trans_getsb_recur(bip);
200 return bp;
201 }
202
203 bp = xfs_getsb(mp);
204 if (bp == NULL)
205 return NULL;
206
207 _xfs_trans_bjoin(tp, bp, 1);
208 trace_xfs_trans_getsb(bp->b_log_item);
209 return bp;
210}
211
212/*
213 * Get and lock the buffer for the caller if it is not already
214 * locked within the given transaction. If it has not yet been
215 * read in, read it from disk. If it is already locked
216 * within the transaction and already read in, just increment its
217 * lock recursion count and return a pointer to it.
218 *
219 * If the transaction pointer is NULL, make this just a normal
220 * read_buf() call.
221 */
222int
223xfs_trans_read_buf_map(
224 struct xfs_mount *mp,
225 struct xfs_trans *tp,
226 struct xfs_buftarg *target,
227 struct xfs_buf_map *map,
228 int nmaps,
229 xfs_buf_flags_t flags,
230 struct xfs_buf **bpp,
231 const struct xfs_buf_ops *ops)
232{
233 struct xfs_buf *bp = NULL;
234 struct xfs_buf_log_item *bip;
235 int error;
236
237 *bpp = NULL;
238 /*
239 * If we find the buffer in the cache with this transaction
240 * pointer in its b_fsprivate2 field, then we know we already
241 * have it locked. If it is already read in we just increment
242 * the lock recursion count and return the buffer to the caller.
243 * If the buffer is not yet read in, then we read it in, increment
244 * the lock recursion count, and return it to the caller.
245 */
246 if (tp)
247 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
248 if (bp) {
249 ASSERT(xfs_buf_islocked(bp));
250 ASSERT(bp->b_transp == tp);
251 ASSERT(bp->b_log_item != NULL);
252 ASSERT(!bp->b_error);
253 ASSERT(bp->b_flags & XBF_DONE);
254
255 /*
256 * We never locked this buf ourselves, so we shouldn't
257 * brelse it either. Just get out.
258 */
259 if (XFS_FORCED_SHUTDOWN(mp)) {
260 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
261 return -EIO;
262 }
263
264 /*
265 * Check if the caller is trying to read a buffer that is
266 * already attached to the transaction yet has no buffer ops
267 * assigned. Ops are usually attached when the buffer is
268 * attached to the transaction, or by the read caller if
269 * special circumstances. That didn't happen, which is not
270 * how this is supposed to go.
271 *
272 * If the buffer passes verification we'll let this go, but if
273 * not we have to shut down. Let the transaction cleanup code
274 * release this buffer when it kills the tranaction.
275 */
276 ASSERT(bp->b_ops != NULL);
277 error = xfs_buf_reverify(bp, ops);
278 if (error) {
279 xfs_buf_ioerror_alert(bp, __func__);
280
281 if (tp->t_flags & XFS_TRANS_DIRTY)
282 xfs_force_shutdown(tp->t_mountp,
283 SHUTDOWN_META_IO_ERROR);
284
285 /* bad CRC means corrupted metadata */
286 if (error == -EFSBADCRC)
287 error = -EFSCORRUPTED;
288 return error;
289 }
290
291 bip = bp->b_log_item;
292 bip->bli_recur++;
293
294 ASSERT(atomic_read(&bip->bli_refcount) > 0);
295 trace_xfs_trans_read_buf_recur(bip);
296 ASSERT(bp->b_ops != NULL || ops == NULL);
297 *bpp = bp;
298 return 0;
299 }
300
301 bp = xfs_buf_read_map(target, map, nmaps, flags, ops);
302 if (!bp) {
303 if (!(flags & XBF_TRYLOCK))
304 return -ENOMEM;
305 return tp ? 0 : -EAGAIN;
306 }
307
308 /*
309 * If we've had a read error, then the contents of the buffer are
310 * invalid and should not be used. To ensure that a followup read tries
311 * to pull the buffer from disk again, we clear the XBF_DONE flag and
312 * mark the buffer stale. This ensures that anyone who has a current
313 * reference to the buffer will interpret it's contents correctly and
314 * future cache lookups will also treat it as an empty, uninitialised
315 * buffer.
316 */
317 if (bp->b_error) {
318 error = bp->b_error;
319 if (!XFS_FORCED_SHUTDOWN(mp))
320 xfs_buf_ioerror_alert(bp, __func__);
321 bp->b_flags &= ~XBF_DONE;
322 xfs_buf_stale(bp);
323
324 if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
325 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
326 xfs_buf_relse(bp);
327
328 /* bad CRC means corrupted metadata */
329 if (error == -EFSBADCRC)
330 error = -EFSCORRUPTED;
331 return error;
332 }
333
334 if (XFS_FORCED_SHUTDOWN(mp)) {
335 xfs_buf_relse(bp);
336 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
337 return -EIO;
338 }
339
340 if (tp) {
341 _xfs_trans_bjoin(tp, bp, 1);
342 trace_xfs_trans_read_buf(bp->b_log_item);
343 }
344 ASSERT(bp->b_ops != NULL || ops == NULL);
345 *bpp = bp;
346 return 0;
347
348}
349
350/* Has this buffer been dirtied by anyone? */
351bool
352xfs_trans_buf_is_dirty(
353 struct xfs_buf *bp)
354{
355 struct xfs_buf_log_item *bip = bp->b_log_item;
356
357 if (!bip)
358 return false;
359 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
360 return test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
361}
362
363/*
364 * Release a buffer previously joined to the transaction. If the buffer is
365 * modified within this transaction, decrement the recursion count but do not
366 * release the buffer even if the count goes to 0. If the buffer is not modified
367 * within the transaction, decrement the recursion count and release the buffer
368 * if the recursion count goes to 0.
369 *
370 * If the buffer is to be released and it was not already dirty before this
371 * transaction began, then also free the buf_log_item associated with it.
372 *
373 * If the transaction pointer is NULL, this is a normal xfs_buf_relse() call.
374 */
375void
376xfs_trans_brelse(
377 struct xfs_trans *tp,
378 struct xfs_buf *bp)
379{
380 struct xfs_buf_log_item *bip = bp->b_log_item;
381
382 ASSERT(bp->b_transp == tp);
383
384 if (!tp) {
385 xfs_buf_relse(bp);
386 return;
387 }
388
389 trace_xfs_trans_brelse(bip);
390 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
391 ASSERT(atomic_read(&bip->bli_refcount) > 0);
392
393 /*
394 * If the release is for a recursive lookup, then decrement the count
395 * and return.
396 */
397 if (bip->bli_recur > 0) {
398 bip->bli_recur--;
399 return;
400 }
401
402 /*
403 * If the buffer is invalidated or dirty in this transaction, we can't
404 * release it until we commit.
405 */
406 if (test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags))
407 return;
408 if (bip->bli_flags & XFS_BLI_STALE)
409 return;
410
411 /*
412 * Unlink the log item from the transaction and clear the hold flag, if
413 * set. We wouldn't want the next user of the buffer to get confused.
414 */
415 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
416 xfs_trans_del_item(&bip->bli_item);
417 bip->bli_flags &= ~XFS_BLI_HOLD;
418
419 /* drop the reference to the bli */
420 xfs_buf_item_put(bip);
421
422 bp->b_transp = NULL;
423 xfs_buf_relse(bp);
424}
425
426/*
427 * Mark the buffer as not needing to be unlocked when the buf item's
428 * iop_committing() routine is called. The buffer must already be locked
429 * and associated with the given transaction.
430 */
431/* ARGSUSED */
432void
433xfs_trans_bhold(
434 xfs_trans_t *tp,
435 xfs_buf_t *bp)
436{
437 struct xfs_buf_log_item *bip = bp->b_log_item;
438
439 ASSERT(bp->b_transp == tp);
440 ASSERT(bip != NULL);
441 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
442 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
443 ASSERT(atomic_read(&bip->bli_refcount) > 0);
444
445 bip->bli_flags |= XFS_BLI_HOLD;
446 trace_xfs_trans_bhold(bip);
447}
448
449/*
450 * Cancel the previous buffer hold request made on this buffer
451 * for this transaction.
452 */
453void
454xfs_trans_bhold_release(
455 xfs_trans_t *tp,
456 xfs_buf_t *bp)
457{
458 struct xfs_buf_log_item *bip = bp->b_log_item;
459
460 ASSERT(bp->b_transp == tp);
461 ASSERT(bip != NULL);
462 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
463 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
464 ASSERT(atomic_read(&bip->bli_refcount) > 0);
465 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
466
467 bip->bli_flags &= ~XFS_BLI_HOLD;
468 trace_xfs_trans_bhold_release(bip);
469}
470
471/*
472 * Mark a buffer dirty in the transaction.
473 */
474void
475xfs_trans_dirty_buf(
476 struct xfs_trans *tp,
477 struct xfs_buf *bp)
478{
479 struct xfs_buf_log_item *bip = bp->b_log_item;
480
481 ASSERT(bp->b_transp == tp);
482 ASSERT(bip != NULL);
483 ASSERT(bp->b_iodone == NULL ||
484 bp->b_iodone == xfs_buf_iodone_callbacks);
485
486 /*
487 * Mark the buffer as needing to be written out eventually,
488 * and set its iodone function to remove the buffer's buf log
489 * item from the AIL and free it when the buffer is flushed
490 * to disk. See xfs_buf_attach_iodone() for more details
491 * on li_cb and xfs_buf_iodone_callbacks().
492 * If we end up aborting this transaction, we trap this buffer
493 * inside the b_bdstrat callback so that this won't get written to
494 * disk.
495 */
496 bp->b_flags |= XBF_DONE;
497
498 ASSERT(atomic_read(&bip->bli_refcount) > 0);
499 bp->b_iodone = xfs_buf_iodone_callbacks;
500 bip->bli_item.li_cb = xfs_buf_iodone;
501
502 /*
503 * If we invalidated the buffer within this transaction, then
504 * cancel the invalidation now that we're dirtying the buffer
505 * again. There are no races with the code in xfs_buf_item_unpin(),
506 * because we have a reference to the buffer this entire time.
507 */
508 if (bip->bli_flags & XFS_BLI_STALE) {
509 bip->bli_flags &= ~XFS_BLI_STALE;
510 ASSERT(bp->b_flags & XBF_STALE);
511 bp->b_flags &= ~XBF_STALE;
512 bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
513 }
514 bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
515
516 tp->t_flags |= XFS_TRANS_DIRTY;
517 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
518}
519
520/*
521 * This is called to mark bytes first through last inclusive of the given
522 * buffer as needing to be logged when the transaction is committed.
523 * The buffer must already be associated with the given transaction.
524 *
525 * First and last are numbers relative to the beginning of this buffer,
526 * so the first byte in the buffer is numbered 0 regardless of the
527 * value of b_blkno.
528 */
529void
530xfs_trans_log_buf(
531 struct xfs_trans *tp,
532 struct xfs_buf *bp,
533 uint first,
534 uint last)
535{
536 struct xfs_buf_log_item *bip = bp->b_log_item;
537
538 ASSERT(first <= last && last < BBTOB(bp->b_length));
539 ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
540
541 xfs_trans_dirty_buf(tp, bp);
542
543 trace_xfs_trans_log_buf(bip);
544 xfs_buf_item_log(bip, first, last);
545}
546
547
548/*
549 * Invalidate a buffer that is being used within a transaction.
550 *
551 * Typically this is because the blocks in the buffer are being freed, so we
552 * need to prevent it from being written out when we're done. Allowing it
553 * to be written again might overwrite data in the free blocks if they are
554 * reallocated to a file.
555 *
556 * We prevent the buffer from being written out by marking it stale. We can't
557 * get rid of the buf log item at this point because the buffer may still be
558 * pinned by another transaction. If that is the case, then we'll wait until
559 * the buffer is committed to disk for the last time (we can tell by the ref
560 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
561 * keep the buffer locked so that the buffer and buf log item are not reused.
562 *
563 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
564 * the buf item. This will be used at recovery time to determine that copies
565 * of the buffer in the log before this should not be replayed.
566 *
567 * We mark the item descriptor and the transaction dirty so that we'll hold
568 * the buffer until after the commit.
569 *
570 * Since we're invalidating the buffer, we also clear the state about which
571 * parts of the buffer have been logged. We also clear the flag indicating
572 * that this is an inode buffer since the data in the buffer will no longer
573 * be valid.
574 *
575 * We set the stale bit in the buffer as well since we're getting rid of it.
576 */
577void
578xfs_trans_binval(
579 xfs_trans_t *tp,
580 xfs_buf_t *bp)
581{
582 struct xfs_buf_log_item *bip = bp->b_log_item;
583 int i;
584
585 ASSERT(bp->b_transp == tp);
586 ASSERT(bip != NULL);
587 ASSERT(atomic_read(&bip->bli_refcount) > 0);
588
589 trace_xfs_trans_binval(bip);
590
591 if (bip->bli_flags & XFS_BLI_STALE) {
592 /*
593 * If the buffer is already invalidated, then
594 * just return.
595 */
596 ASSERT(bp->b_flags & XBF_STALE);
597 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
598 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
599 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
600 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
601 ASSERT(test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
602 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
603 return;
604 }
605
606 xfs_buf_stale(bp);
607
608 bip->bli_flags |= XFS_BLI_STALE;
609 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
610 bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
611 bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
612 bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
613 for (i = 0; i < bip->bli_format_count; i++) {
614 memset(bip->bli_formats[i].blf_data_map, 0,
615 (bip->bli_formats[i].blf_map_size * sizeof(uint)));
616 }
617 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
618 tp->t_flags |= XFS_TRANS_DIRTY;
619}
620
621/*
622 * This call is used to indicate that the buffer contains on-disk inodes which
623 * must be handled specially during recovery. They require special handling
624 * because only the di_next_unlinked from the inodes in the buffer should be
625 * recovered. The rest of the data in the buffer is logged via the inodes
626 * themselves.
627 *
628 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
629 * transferred to the buffer's log format structure so that we'll know what to
630 * do at recovery time.
631 */
632void
633xfs_trans_inode_buf(
634 xfs_trans_t *tp,
635 xfs_buf_t *bp)
636{
637 struct xfs_buf_log_item *bip = bp->b_log_item;
638
639 ASSERT(bp->b_transp == tp);
640 ASSERT(bip != NULL);
641 ASSERT(atomic_read(&bip->bli_refcount) > 0);
642
643 bip->bli_flags |= XFS_BLI_INODE_BUF;
644 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
645}
646
647/*
648 * This call is used to indicate that the buffer is going to
649 * be staled and was an inode buffer. This means it gets
650 * special processing during unpin - where any inodes
651 * associated with the buffer should be removed from ail.
652 * There is also special processing during recovery,
653 * any replay of the inodes in the buffer needs to be
654 * prevented as the buffer may have been reused.
655 */
656void
657xfs_trans_stale_inode_buf(
658 xfs_trans_t *tp,
659 xfs_buf_t *bp)
660{
661 struct xfs_buf_log_item *bip = bp->b_log_item;
662
663 ASSERT(bp->b_transp == tp);
664 ASSERT(bip != NULL);
665 ASSERT(atomic_read(&bip->bli_refcount) > 0);
666
667 bip->bli_flags |= XFS_BLI_STALE_INODE;
668 bip->bli_item.li_cb = xfs_buf_iodone;
669 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
670}
671
672/*
673 * Mark the buffer as being one which contains newly allocated
674 * inodes. We need to make sure that even if this buffer is
675 * relogged as an 'inode buf' we still recover all of the inode
676 * images in the face of a crash. This works in coordination with
677 * xfs_buf_item_committed() to ensure that the buffer remains in the
678 * AIL at its original location even after it has been relogged.
679 */
680/* ARGSUSED */
681void
682xfs_trans_inode_alloc_buf(
683 xfs_trans_t *tp,
684 xfs_buf_t *bp)
685{
686 struct xfs_buf_log_item *bip = bp->b_log_item;
687
688 ASSERT(bp->b_transp == tp);
689 ASSERT(bip != NULL);
690 ASSERT(atomic_read(&bip->bli_refcount) > 0);
691
692 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
693 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
694}
695
696/*
697 * Mark the buffer as ordered for this transaction. This means that the contents
698 * of the buffer are not recorded in the transaction but it is tracked in the
699 * AIL as though it was. This allows us to record logical changes in
700 * transactions rather than the physical changes we make to the buffer without
701 * changing writeback ordering constraints of metadata buffers.
702 */
703bool
704xfs_trans_ordered_buf(
705 struct xfs_trans *tp,
706 struct xfs_buf *bp)
707{
708 struct xfs_buf_log_item *bip = bp->b_log_item;
709
710 ASSERT(bp->b_transp == tp);
711 ASSERT(bip != NULL);
712 ASSERT(atomic_read(&bip->bli_refcount) > 0);
713
714 if (xfs_buf_item_dirty_format(bip))
715 return false;
716
717 bip->bli_flags |= XFS_BLI_ORDERED;
718 trace_xfs_buf_item_ordered(bip);
719
720 /*
721 * We don't log a dirty range of an ordered buffer but it still needs
722 * to be marked dirty and that it has been logged.
723 */
724 xfs_trans_dirty_buf(tp, bp);
725 return true;
726}
727
728/*
729 * Set the type of the buffer for log recovery so that it can correctly identify
730 * and hence attach the correct buffer ops to the buffer after replay.
731 */
732void
733xfs_trans_buf_set_type(
734 struct xfs_trans *tp,
735 struct xfs_buf *bp,
736 enum xfs_blft type)
737{
738 struct xfs_buf_log_item *bip = bp->b_log_item;
739
740 if (!tp)
741 return;
742
743 ASSERT(bp->b_transp == tp);
744 ASSERT(bip != NULL);
745 ASSERT(atomic_read(&bip->bli_refcount) > 0);
746
747 xfs_blft_to_flags(&bip->__bli_format, type);
748}
749
750void
751xfs_trans_buf_copy_type(
752 struct xfs_buf *dst_bp,
753 struct xfs_buf *src_bp)
754{
755 struct xfs_buf_log_item *sbip = src_bp->b_log_item;
756 struct xfs_buf_log_item *dbip = dst_bp->b_log_item;
757 enum xfs_blft type;
758
759 type = xfs_blft_from_flags(&sbip->__bli_format);
760 xfs_blft_to_flags(&dbip->__bli_format, type);
761}
762
763/*
764 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
765 * dquots. However, unlike in inode buffer recovery, dquot buffers get
766 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
767 * The only thing that makes dquot buffers different from regular
768 * buffers is that we must not replay dquot bufs when recovering
769 * if a _corresponding_ quotaoff has happened. We also have to distinguish
770 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
771 * can be turned off independently.
772 */
773/* ARGSUSED */
774void
775xfs_trans_dquot_buf(
776 xfs_trans_t *tp,
777 xfs_buf_t *bp,
778 uint type)
779{
780 struct xfs_buf_log_item *bip = bp->b_log_item;
781
782 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
783 type == XFS_BLF_PDQUOT_BUF ||
784 type == XFS_BLF_GDQUOT_BUF);
785
786 bip->__bli_format.blf_flags |= type;
787
788 switch (type) {
789 case XFS_BLF_UDQUOT_BUF:
790 type = XFS_BLFT_UDQUOT_BUF;
791 break;
792 case XFS_BLF_PDQUOT_BUF:
793 type = XFS_BLFT_PDQUOT_BUF;
794 break;
795 case XFS_BLF_GDQUOT_BUF:
796 type = XFS_BLFT_GDQUOT_BUF;
797 break;
798 default:
799 type = XFS_BLFT_UNKNOWN_BUF;
800 break;
801 }
802
803 xfs_trans_buf_set_type(tp, bp, type);
804}