Loading...
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_daddr(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 */
115int
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 struct xfs_buf **bpp)
123{
124 struct xfs_buf *bp;
125 struct xfs_buf_log_item *bip;
126 int error;
127
128 *bpp = NULL;
129 if (!tp)
130 return xfs_buf_get_map(target, map, nmaps, flags, bpp);
131
132 /*
133 * If we find the buffer in the cache with this transaction
134 * pointer in its b_fsprivate2 field, then we know we already
135 * have it locked. In this case we just increment the lock
136 * recursion count and return the buffer to the caller.
137 */
138 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
139 if (bp != NULL) {
140 ASSERT(xfs_buf_islocked(bp));
141 if (xfs_is_shutdown(tp->t_mountp)) {
142 xfs_buf_stale(bp);
143 bp->b_flags |= XBF_DONE;
144 }
145
146 ASSERT(bp->b_transp == tp);
147 bip = bp->b_log_item;
148 ASSERT(bip != NULL);
149 ASSERT(atomic_read(&bip->bli_refcount) > 0);
150 bip->bli_recur++;
151 trace_xfs_trans_get_buf_recur(bip);
152 *bpp = bp;
153 return 0;
154 }
155
156 error = xfs_buf_get_map(target, map, nmaps, flags, &bp);
157 if (error)
158 return error;
159
160 ASSERT(!bp->b_error);
161
162 _xfs_trans_bjoin(tp, bp, 1);
163 trace_xfs_trans_get_buf(bp->b_log_item);
164 *bpp = bp;
165 return 0;
166}
167
168/*
169 * Get and lock the superblock buffer for the given transaction.
170 */
171struct xfs_buf *
172xfs_trans_getsb(
173 struct xfs_trans *tp)
174{
175 struct xfs_buf *bp = tp->t_mountp->m_sb_bp;
176
177 /*
178 * Just increment the lock recursion count if the buffer is already
179 * attached to this transaction.
180 */
181 if (bp->b_transp == tp) {
182 struct xfs_buf_log_item *bip = bp->b_log_item;
183
184 ASSERT(bip != NULL);
185 ASSERT(atomic_read(&bip->bli_refcount) > 0);
186 bip->bli_recur++;
187
188 trace_xfs_trans_getsb_recur(bip);
189 } else {
190 xfs_buf_lock(bp);
191 xfs_buf_hold(bp);
192 _xfs_trans_bjoin(tp, bp, 1);
193
194 trace_xfs_trans_getsb(bp->b_log_item);
195 }
196
197 return bp;
198}
199
200/*
201 * Get and lock the buffer for the caller if it is not already
202 * locked within the given transaction. If it has not yet been
203 * read in, read it from disk. If it is already locked
204 * within the transaction and already read in, just increment its
205 * lock recursion count and return a pointer to it.
206 *
207 * If the transaction pointer is NULL, make this just a normal
208 * read_buf() call.
209 */
210int
211xfs_trans_read_buf_map(
212 struct xfs_mount *mp,
213 struct xfs_trans *tp,
214 struct xfs_buftarg *target,
215 struct xfs_buf_map *map,
216 int nmaps,
217 xfs_buf_flags_t flags,
218 struct xfs_buf **bpp,
219 const struct xfs_buf_ops *ops)
220{
221 struct xfs_buf *bp = NULL;
222 struct xfs_buf_log_item *bip;
223 int error;
224
225 *bpp = NULL;
226 /*
227 * If we find the buffer in the cache with this transaction
228 * pointer in its b_fsprivate2 field, then we know we already
229 * have it locked. If it is already read in we just increment
230 * the lock recursion count and return the buffer to the caller.
231 * If the buffer is not yet read in, then we read it in, increment
232 * the lock recursion count, and return it to the caller.
233 */
234 if (tp)
235 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
236 if (bp) {
237 ASSERT(xfs_buf_islocked(bp));
238 ASSERT(bp->b_transp == tp);
239 ASSERT(bp->b_log_item != NULL);
240 ASSERT(!bp->b_error);
241 ASSERT(bp->b_flags & XBF_DONE);
242
243 /*
244 * We never locked this buf ourselves, so we shouldn't
245 * brelse it either. Just get out.
246 */
247 if (xfs_is_shutdown(mp)) {
248 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
249 return -EIO;
250 }
251
252 /*
253 * Check if the caller is trying to read a buffer that is
254 * already attached to the transaction yet has no buffer ops
255 * assigned. Ops are usually attached when the buffer is
256 * attached to the transaction, or by the read caller if
257 * special circumstances. That didn't happen, which is not
258 * how this is supposed to go.
259 *
260 * If the buffer passes verification we'll let this go, but if
261 * not we have to shut down. Let the transaction cleanup code
262 * release this buffer when it kills the tranaction.
263 */
264 ASSERT(bp->b_ops != NULL);
265 error = xfs_buf_reverify(bp, ops);
266 if (error) {
267 xfs_buf_ioerror_alert(bp, __return_address);
268
269 if (tp->t_flags & XFS_TRANS_DIRTY)
270 xfs_force_shutdown(tp->t_mountp,
271 SHUTDOWN_META_IO_ERROR);
272
273 /* bad CRC means corrupted metadata */
274 if (error == -EFSBADCRC)
275 error = -EFSCORRUPTED;
276 return error;
277 }
278
279 bip = bp->b_log_item;
280 bip->bli_recur++;
281
282 ASSERT(atomic_read(&bip->bli_refcount) > 0);
283 trace_xfs_trans_read_buf_recur(bip);
284 ASSERT(bp->b_ops != NULL || ops == NULL);
285 *bpp = bp;
286 return 0;
287 }
288
289 error = xfs_buf_read_map(target, map, nmaps, flags, &bp, ops,
290 __return_address);
291 switch (error) {
292 case 0:
293 break;
294 default:
295 if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
296 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
297 fallthrough;
298 case -ENOMEM:
299 case -EAGAIN:
300 return error;
301 }
302
303 if (xfs_is_shutdown(mp)) {
304 xfs_buf_relse(bp);
305 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
306 return -EIO;
307 }
308
309 if (tp) {
310 _xfs_trans_bjoin(tp, bp, 1);
311 trace_xfs_trans_read_buf(bp->b_log_item);
312 }
313 ASSERT(bp->b_ops != NULL || ops == NULL);
314 *bpp = bp;
315 return 0;
316
317}
318
319/* Has this buffer been dirtied by anyone? */
320bool
321xfs_trans_buf_is_dirty(
322 struct xfs_buf *bp)
323{
324 struct xfs_buf_log_item *bip = bp->b_log_item;
325
326 if (!bip)
327 return false;
328 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
329 return test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
330}
331
332/*
333 * Release a buffer previously joined to the transaction. If the buffer is
334 * modified within this transaction, decrement the recursion count but do not
335 * release the buffer even if the count goes to 0. If the buffer is not modified
336 * within the transaction, decrement the recursion count and release the buffer
337 * if the recursion count goes to 0.
338 *
339 * If the buffer is to be released and it was not already dirty before this
340 * transaction began, then also free the buf_log_item associated with it.
341 *
342 * If the transaction pointer is NULL, this is a normal xfs_buf_relse() call.
343 */
344void
345xfs_trans_brelse(
346 struct xfs_trans *tp,
347 struct xfs_buf *bp)
348{
349 struct xfs_buf_log_item *bip = bp->b_log_item;
350
351 ASSERT(bp->b_transp == tp);
352
353 if (!tp) {
354 xfs_buf_relse(bp);
355 return;
356 }
357
358 trace_xfs_trans_brelse(bip);
359 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
360 ASSERT(atomic_read(&bip->bli_refcount) > 0);
361
362 /*
363 * If the release is for a recursive lookup, then decrement the count
364 * and return.
365 */
366 if (bip->bli_recur > 0) {
367 bip->bli_recur--;
368 return;
369 }
370
371 /*
372 * If the buffer is invalidated or dirty in this transaction, we can't
373 * release it until we commit.
374 */
375 if (test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags))
376 return;
377 if (bip->bli_flags & XFS_BLI_STALE)
378 return;
379
380 /*
381 * Unlink the log item from the transaction and clear the hold flag, if
382 * set. We wouldn't want the next user of the buffer to get confused.
383 */
384 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
385 xfs_trans_del_item(&bip->bli_item);
386 bip->bli_flags &= ~XFS_BLI_HOLD;
387
388 /* drop the reference to the bli */
389 xfs_buf_item_put(bip);
390
391 bp->b_transp = NULL;
392 xfs_buf_relse(bp);
393}
394
395/*
396 * Mark the buffer as not needing to be unlocked when the buf item's
397 * iop_committing() routine is called. The buffer must already be locked
398 * and associated with the given transaction.
399 */
400/* ARGSUSED */
401void
402xfs_trans_bhold(
403 xfs_trans_t *tp,
404 struct xfs_buf *bp)
405{
406 struct xfs_buf_log_item *bip = bp->b_log_item;
407
408 ASSERT(bp->b_transp == tp);
409 ASSERT(bip != NULL);
410 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
411 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
412 ASSERT(atomic_read(&bip->bli_refcount) > 0);
413
414 bip->bli_flags |= XFS_BLI_HOLD;
415 trace_xfs_trans_bhold(bip);
416}
417
418/*
419 * Cancel the previous buffer hold request made on this buffer
420 * for this transaction.
421 */
422void
423xfs_trans_bhold_release(
424 xfs_trans_t *tp,
425 struct xfs_buf *bp)
426{
427 struct xfs_buf_log_item *bip = bp->b_log_item;
428
429 ASSERT(bp->b_transp == tp);
430 ASSERT(bip != NULL);
431 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
432 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
433 ASSERT(atomic_read(&bip->bli_refcount) > 0);
434 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
435
436 bip->bli_flags &= ~XFS_BLI_HOLD;
437 trace_xfs_trans_bhold_release(bip);
438}
439
440/*
441 * Mark a buffer dirty in the transaction.
442 */
443void
444xfs_trans_dirty_buf(
445 struct xfs_trans *tp,
446 struct xfs_buf *bp)
447{
448 struct xfs_buf_log_item *bip = bp->b_log_item;
449
450 ASSERT(bp->b_transp == tp);
451 ASSERT(bip != NULL);
452
453 /*
454 * Mark the buffer as needing to be written out eventually,
455 * and set its iodone function to remove the buffer's buf log
456 * item from the AIL and free it when the buffer is flushed
457 * to disk.
458 */
459 bp->b_flags |= XBF_DONE;
460
461 ASSERT(atomic_read(&bip->bli_refcount) > 0);
462
463 /*
464 * If we invalidated the buffer within this transaction, then
465 * cancel the invalidation now that we're dirtying the buffer
466 * again. There are no races with the code in xfs_buf_item_unpin(),
467 * because we have a reference to the buffer this entire time.
468 */
469 if (bip->bli_flags & XFS_BLI_STALE) {
470 bip->bli_flags &= ~XFS_BLI_STALE;
471 ASSERT(bp->b_flags & XBF_STALE);
472 bp->b_flags &= ~XBF_STALE;
473 bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
474 }
475 bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
476
477 tp->t_flags |= XFS_TRANS_DIRTY;
478 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
479}
480
481/*
482 * This is called to mark bytes first through last inclusive of the given
483 * buffer as needing to be logged when the transaction is committed.
484 * The buffer must already be associated with the given transaction.
485 *
486 * First and last are numbers relative to the beginning of this buffer,
487 * so the first byte in the buffer is numbered 0 regardless of the
488 * value of b_blkno.
489 */
490void
491xfs_trans_log_buf(
492 struct xfs_trans *tp,
493 struct xfs_buf *bp,
494 uint first,
495 uint last)
496{
497 struct xfs_buf_log_item *bip = bp->b_log_item;
498
499 ASSERT(first <= last && last < BBTOB(bp->b_length));
500 ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
501
502 xfs_trans_dirty_buf(tp, bp);
503
504 trace_xfs_trans_log_buf(bip);
505 xfs_buf_item_log(bip, first, last);
506}
507
508
509/*
510 * Invalidate a buffer that is being used within a transaction.
511 *
512 * Typically this is because the blocks in the buffer are being freed, so we
513 * need to prevent it from being written out when we're done. Allowing it
514 * to be written again might overwrite data in the free blocks if they are
515 * reallocated to a file.
516 *
517 * We prevent the buffer from being written out by marking it stale. We can't
518 * get rid of the buf log item at this point because the buffer may still be
519 * pinned by another transaction. If that is the case, then we'll wait until
520 * the buffer is committed to disk for the last time (we can tell by the ref
521 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
522 * keep the buffer locked so that the buffer and buf log item are not reused.
523 *
524 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
525 * the buf item. This will be used at recovery time to determine that copies
526 * of the buffer in the log before this should not be replayed.
527 *
528 * We mark the item descriptor and the transaction dirty so that we'll hold
529 * the buffer until after the commit.
530 *
531 * Since we're invalidating the buffer, we also clear the state about which
532 * parts of the buffer have been logged. We also clear the flag indicating
533 * that this is an inode buffer since the data in the buffer will no longer
534 * be valid.
535 *
536 * We set the stale bit in the buffer as well since we're getting rid of it.
537 */
538void
539xfs_trans_binval(
540 xfs_trans_t *tp,
541 struct xfs_buf *bp)
542{
543 struct xfs_buf_log_item *bip = bp->b_log_item;
544 int i;
545
546 ASSERT(bp->b_transp == tp);
547 ASSERT(bip != NULL);
548 ASSERT(atomic_read(&bip->bli_refcount) > 0);
549
550 trace_xfs_trans_binval(bip);
551
552 if (bip->bli_flags & XFS_BLI_STALE) {
553 /*
554 * If the buffer is already invalidated, then
555 * just return.
556 */
557 ASSERT(bp->b_flags & XBF_STALE);
558 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
559 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
560 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
561 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
562 ASSERT(test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
563 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
564 return;
565 }
566
567 xfs_buf_stale(bp);
568
569 bip->bli_flags |= XFS_BLI_STALE;
570 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
571 bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
572 bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
573 bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
574 for (i = 0; i < bip->bli_format_count; i++) {
575 memset(bip->bli_formats[i].blf_data_map, 0,
576 (bip->bli_formats[i].blf_map_size * sizeof(uint)));
577 }
578 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
579 tp->t_flags |= XFS_TRANS_DIRTY;
580}
581
582/*
583 * This call is used to indicate that the buffer contains on-disk inodes which
584 * must be handled specially during recovery. They require special handling
585 * because only the di_next_unlinked from the inodes in the buffer should be
586 * recovered. The rest of the data in the buffer is logged via the inodes
587 * themselves.
588 *
589 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
590 * transferred to the buffer's log format structure so that we'll know what to
591 * do at recovery time.
592 */
593void
594xfs_trans_inode_buf(
595 xfs_trans_t *tp,
596 struct xfs_buf *bp)
597{
598 struct xfs_buf_log_item *bip = bp->b_log_item;
599
600 ASSERT(bp->b_transp == tp);
601 ASSERT(bip != NULL);
602 ASSERT(atomic_read(&bip->bli_refcount) > 0);
603
604 bip->bli_flags |= XFS_BLI_INODE_BUF;
605 bp->b_flags |= _XBF_INODES;
606 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
607}
608
609/*
610 * This call is used to indicate that the buffer is going to
611 * be staled and was an inode buffer. This means it gets
612 * special processing during unpin - where any inodes
613 * associated with the buffer should be removed from ail.
614 * There is also special processing during recovery,
615 * any replay of the inodes in the buffer needs to be
616 * prevented as the buffer may have been reused.
617 */
618void
619xfs_trans_stale_inode_buf(
620 xfs_trans_t *tp,
621 struct xfs_buf *bp)
622{
623 struct xfs_buf_log_item *bip = bp->b_log_item;
624
625 ASSERT(bp->b_transp == tp);
626 ASSERT(bip != NULL);
627 ASSERT(atomic_read(&bip->bli_refcount) > 0);
628
629 bip->bli_flags |= XFS_BLI_STALE_INODE;
630 bp->b_flags |= _XBF_INODES;
631 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
632}
633
634/*
635 * Mark the buffer as being one which contains newly allocated
636 * inodes. We need to make sure that even if this buffer is
637 * relogged as an 'inode buf' we still recover all of the inode
638 * images in the face of a crash. This works in coordination with
639 * xfs_buf_item_committed() to ensure that the buffer remains in the
640 * AIL at its original location even after it has been relogged.
641 */
642/* ARGSUSED */
643void
644xfs_trans_inode_alloc_buf(
645 xfs_trans_t *tp,
646 struct xfs_buf *bp)
647{
648 struct xfs_buf_log_item *bip = bp->b_log_item;
649
650 ASSERT(bp->b_transp == tp);
651 ASSERT(bip != NULL);
652 ASSERT(atomic_read(&bip->bli_refcount) > 0);
653
654 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
655 bp->b_flags |= _XBF_INODES;
656 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
657}
658
659/*
660 * Mark the buffer as ordered for this transaction. This means that the contents
661 * of the buffer are not recorded in the transaction but it is tracked in the
662 * AIL as though it was. This allows us to record logical changes in
663 * transactions rather than the physical changes we make to the buffer without
664 * changing writeback ordering constraints of metadata buffers.
665 */
666bool
667xfs_trans_ordered_buf(
668 struct xfs_trans *tp,
669 struct xfs_buf *bp)
670{
671 struct xfs_buf_log_item *bip = bp->b_log_item;
672
673 ASSERT(bp->b_transp == tp);
674 ASSERT(bip != NULL);
675 ASSERT(atomic_read(&bip->bli_refcount) > 0);
676
677 if (xfs_buf_item_dirty_format(bip))
678 return false;
679
680 bip->bli_flags |= XFS_BLI_ORDERED;
681 trace_xfs_buf_item_ordered(bip);
682
683 /*
684 * We don't log a dirty range of an ordered buffer but it still needs
685 * to be marked dirty and that it has been logged.
686 */
687 xfs_trans_dirty_buf(tp, bp);
688 return true;
689}
690
691/*
692 * Set the type of the buffer for log recovery so that it can correctly identify
693 * and hence attach the correct buffer ops to the buffer after replay.
694 */
695void
696xfs_trans_buf_set_type(
697 struct xfs_trans *tp,
698 struct xfs_buf *bp,
699 enum xfs_blft type)
700{
701 struct xfs_buf_log_item *bip = bp->b_log_item;
702
703 if (!tp)
704 return;
705
706 ASSERT(bp->b_transp == tp);
707 ASSERT(bip != NULL);
708 ASSERT(atomic_read(&bip->bli_refcount) > 0);
709
710 xfs_blft_to_flags(&bip->__bli_format, type);
711}
712
713void
714xfs_trans_buf_copy_type(
715 struct xfs_buf *dst_bp,
716 struct xfs_buf *src_bp)
717{
718 struct xfs_buf_log_item *sbip = src_bp->b_log_item;
719 struct xfs_buf_log_item *dbip = dst_bp->b_log_item;
720 enum xfs_blft type;
721
722 type = xfs_blft_from_flags(&sbip->__bli_format);
723 xfs_blft_to_flags(&dbip->__bli_format, type);
724}
725
726/*
727 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
728 * dquots. However, unlike in inode buffer recovery, dquot buffers get
729 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
730 * The only thing that makes dquot buffers different from regular
731 * buffers is that we must not replay dquot bufs when recovering
732 * if a _corresponding_ quotaoff has happened. We also have to distinguish
733 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
734 * can be turned off independently.
735 */
736/* ARGSUSED */
737void
738xfs_trans_dquot_buf(
739 xfs_trans_t *tp,
740 struct xfs_buf *bp,
741 uint type)
742{
743 struct xfs_buf_log_item *bip = bp->b_log_item;
744
745 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
746 type == XFS_BLF_PDQUOT_BUF ||
747 type == XFS_BLF_GDQUOT_BUF);
748
749 bip->__bli_format.blf_flags |= type;
750
751 switch (type) {
752 case XFS_BLF_UDQUOT_BUF:
753 type = XFS_BLFT_UDQUOT_BUF;
754 break;
755 case XFS_BLF_PDQUOT_BUF:
756 type = XFS_BLFT_PDQUOT_BUF;
757 break;
758 case XFS_BLF_GDQUOT_BUF:
759 type = XFS_BLFT_GDQUOT_BUF;
760 break;
761 default:
762 type = XFS_BLFT_UNKNOWN_BUF;
763 break;
764 }
765
766 bp->b_flags |= _XBF_DQUOTS;
767 xfs_trans_buf_set_type(tp, bp, type);
768}
1/*
2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
17 */
18#include "xfs.h"
19#include "xfs_fs.h"
20#include "xfs_types.h"
21#include "xfs_bit.h"
22#include "xfs_log.h"
23#include "xfs_inum.h"
24#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
27#include "xfs_mount.h"
28#include "xfs_bmap_btree.h"
29#include "xfs_alloc_btree.h"
30#include "xfs_ialloc_btree.h"
31#include "xfs_dinode.h"
32#include "xfs_inode.h"
33#include "xfs_buf_item.h"
34#include "xfs_trans_priv.h"
35#include "xfs_error.h"
36#include "xfs_rw.h"
37#include "xfs_trace.h"
38
39/*
40 * Check to see if a buffer matching the given parameters is already
41 * a part of the given transaction.
42 */
43STATIC struct xfs_buf *
44xfs_trans_buf_item_match(
45 struct xfs_trans *tp,
46 struct xfs_buftarg *target,
47 xfs_daddr_t blkno,
48 int len)
49{
50 struct xfs_log_item_desc *lidp;
51 struct xfs_buf_log_item *blip;
52
53 len = BBTOB(len);
54 list_for_each_entry(lidp, &tp->t_items, lid_trans) {
55 blip = (struct xfs_buf_log_item *)lidp->lid_item;
56 if (blip->bli_item.li_type == XFS_LI_BUF &&
57 blip->bli_buf->b_target == target &&
58 XFS_BUF_ADDR(blip->bli_buf) == blkno &&
59 XFS_BUF_COUNT(blip->bli_buf) == len)
60 return blip->bli_buf;
61 }
62
63 return NULL;
64}
65
66/*
67 * Add the locked buffer to the transaction.
68 *
69 * The buffer must be locked, and it cannot be associated with any
70 * transaction.
71 *
72 * If the buffer does not yet have a buf log item associated with it,
73 * then allocate one for it. Then add the buf item to the transaction.
74 */
75STATIC void
76_xfs_trans_bjoin(
77 struct xfs_trans *tp,
78 struct xfs_buf *bp,
79 int reset_recur)
80{
81 struct xfs_buf_log_item *bip;
82
83 ASSERT(bp->b_transp == NULL);
84
85 /*
86 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
87 * it doesn't have one yet, then allocate one and initialize it.
88 * The checks to see if one is there are in xfs_buf_item_init().
89 */
90 xfs_buf_item_init(bp, tp->t_mountp);
91 bip = bp->b_fspriv;
92 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
93 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
94 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
95 if (reset_recur)
96 bip->bli_recur = 0;
97
98 /*
99 * Take a reference for this transaction on the buf item.
100 */
101 atomic_inc(&bip->bli_refcount);
102
103 /*
104 * Get a log_item_desc to point at the new item.
105 */
106 xfs_trans_add_item(tp, &bip->bli_item);
107
108 /*
109 * Initialize b_fsprivate2 so we can find it with incore_match()
110 * in xfs_trans_get_buf() and friends above.
111 */
112 bp->b_transp = tp;
113
114}
115
116void
117xfs_trans_bjoin(
118 struct xfs_trans *tp,
119 struct xfs_buf *bp)
120{
121 _xfs_trans_bjoin(tp, bp, 0);
122 trace_xfs_trans_bjoin(bp->b_fspriv);
123}
124
125/*
126 * Get and lock the buffer for the caller if it is not already
127 * locked within the given transaction. If it is already locked
128 * within the transaction, just increment its lock recursion count
129 * and return a pointer to it.
130 *
131 * If the transaction pointer is NULL, make this just a normal
132 * get_buf() call.
133 */
134xfs_buf_t *
135xfs_trans_get_buf(xfs_trans_t *tp,
136 xfs_buftarg_t *target_dev,
137 xfs_daddr_t blkno,
138 int len,
139 uint flags)
140{
141 xfs_buf_t *bp;
142 xfs_buf_log_item_t *bip;
143
144 if (flags == 0)
145 flags = XBF_LOCK | XBF_MAPPED;
146
147 /*
148 * Default to a normal get_buf() call if the tp is NULL.
149 */
150 if (tp == NULL)
151 return xfs_buf_get(target_dev, blkno, len,
152 flags | XBF_DONT_BLOCK);
153
154 /*
155 * If we find the buffer in the cache with this transaction
156 * pointer in its b_fsprivate2 field, then we know we already
157 * have it locked. In this case we just increment the lock
158 * recursion count and return the buffer to the caller.
159 */
160 bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
161 if (bp != NULL) {
162 ASSERT(xfs_buf_islocked(bp));
163 if (XFS_FORCED_SHUTDOWN(tp->t_mountp))
164 XFS_BUF_SUPER_STALE(bp);
165
166 /*
167 * If the buffer is stale then it was binval'ed
168 * since last read. This doesn't matter since the
169 * caller isn't allowed to use the data anyway.
170 */
171 else if (XFS_BUF_ISSTALE(bp))
172 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
173
174 ASSERT(bp->b_transp == tp);
175 bip = bp->b_fspriv;
176 ASSERT(bip != NULL);
177 ASSERT(atomic_read(&bip->bli_refcount) > 0);
178 bip->bli_recur++;
179 trace_xfs_trans_get_buf_recur(bip);
180 return (bp);
181 }
182
183 /*
184 * We always specify the XBF_DONT_BLOCK flag within a transaction
185 * so that get_buf does not try to push out a delayed write buffer
186 * which might cause another transaction to take place (if the
187 * buffer was delayed alloc). Such recursive transactions can
188 * easily deadlock with our current transaction as well as cause
189 * us to run out of stack space.
190 */
191 bp = xfs_buf_get(target_dev, blkno, len, flags | XBF_DONT_BLOCK);
192 if (bp == NULL) {
193 return NULL;
194 }
195
196 ASSERT(!bp->b_error);
197
198 _xfs_trans_bjoin(tp, bp, 1);
199 trace_xfs_trans_get_buf(bp->b_fspriv);
200 return (bp);
201}
202
203/*
204 * Get and lock the superblock buffer of this file system for the
205 * given transaction.
206 *
207 * We don't need to use incore_match() here, because the superblock
208 * buffer is a private buffer which we keep a pointer to in the
209 * mount structure.
210 */
211xfs_buf_t *
212xfs_trans_getsb(xfs_trans_t *tp,
213 struct xfs_mount *mp,
214 int flags)
215{
216 xfs_buf_t *bp;
217 xfs_buf_log_item_t *bip;
218
219 /*
220 * Default to just trying to lock the superblock buffer
221 * if tp is NULL.
222 */
223 if (tp == NULL) {
224 return (xfs_getsb(mp, flags));
225 }
226
227 /*
228 * If the superblock buffer already has this transaction
229 * pointer in its b_fsprivate2 field, then we know we already
230 * have it locked. In this case we just increment the lock
231 * recursion count and return the buffer to the caller.
232 */
233 bp = mp->m_sb_bp;
234 if (bp->b_transp == tp) {
235 bip = bp->b_fspriv;
236 ASSERT(bip != NULL);
237 ASSERT(atomic_read(&bip->bli_refcount) > 0);
238 bip->bli_recur++;
239 trace_xfs_trans_getsb_recur(bip);
240 return (bp);
241 }
242
243 bp = xfs_getsb(mp, flags);
244 if (bp == NULL)
245 return NULL;
246
247 _xfs_trans_bjoin(tp, bp, 1);
248 trace_xfs_trans_getsb(bp->b_fspriv);
249 return (bp);
250}
251
252#ifdef DEBUG
253xfs_buftarg_t *xfs_error_target;
254int xfs_do_error;
255int xfs_req_num;
256int xfs_error_mod = 33;
257#endif
258
259/*
260 * Get and lock the buffer for the caller if it is not already
261 * locked within the given transaction. If it has not yet been
262 * read in, read it from disk. If it is already locked
263 * within the transaction and already read in, just increment its
264 * lock recursion count and return a pointer to it.
265 *
266 * If the transaction pointer is NULL, make this just a normal
267 * read_buf() call.
268 */
269int
270xfs_trans_read_buf(
271 xfs_mount_t *mp,
272 xfs_trans_t *tp,
273 xfs_buftarg_t *target,
274 xfs_daddr_t blkno,
275 int len,
276 uint flags,
277 xfs_buf_t **bpp)
278{
279 xfs_buf_t *bp;
280 xfs_buf_log_item_t *bip;
281 int error;
282
283 if (flags == 0)
284 flags = XBF_LOCK | XBF_MAPPED;
285
286 /*
287 * Default to a normal get_buf() call if the tp is NULL.
288 */
289 if (tp == NULL) {
290 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
291 if (!bp)
292 return (flags & XBF_TRYLOCK) ?
293 EAGAIN : XFS_ERROR(ENOMEM);
294
295 if (bp->b_error) {
296 error = bp->b_error;
297 xfs_ioerror_alert("xfs_trans_read_buf", mp,
298 bp, blkno);
299 xfs_buf_relse(bp);
300 return error;
301 }
302#ifdef DEBUG
303 if (xfs_do_error) {
304 if (xfs_error_target == target) {
305 if (((xfs_req_num++) % xfs_error_mod) == 0) {
306 xfs_buf_relse(bp);
307 xfs_debug(mp, "Returning error!");
308 return XFS_ERROR(EIO);
309 }
310 }
311 }
312#endif
313 if (XFS_FORCED_SHUTDOWN(mp))
314 goto shutdown_abort;
315 *bpp = bp;
316 return 0;
317 }
318
319 /*
320 * If we find the buffer in the cache with this transaction
321 * pointer in its b_fsprivate2 field, then we know we already
322 * have it locked. If it is already read in we just increment
323 * the lock recursion count and return the buffer to the caller.
324 * If the buffer is not yet read in, then we read it in, increment
325 * the lock recursion count, and return it to the caller.
326 */
327 bp = xfs_trans_buf_item_match(tp, target, blkno, len);
328 if (bp != NULL) {
329 ASSERT(xfs_buf_islocked(bp));
330 ASSERT(bp->b_transp == tp);
331 ASSERT(bp->b_fspriv != NULL);
332 ASSERT(!bp->b_error);
333 if (!(XFS_BUF_ISDONE(bp))) {
334 trace_xfs_trans_read_buf_io(bp, _RET_IP_);
335 ASSERT(!XFS_BUF_ISASYNC(bp));
336 XFS_BUF_READ(bp);
337 xfsbdstrat(tp->t_mountp, bp);
338 error = xfs_buf_iowait(bp);
339 if (error) {
340 xfs_ioerror_alert("xfs_trans_read_buf", mp,
341 bp, blkno);
342 xfs_buf_relse(bp);
343 /*
344 * We can gracefully recover from most read
345 * errors. Ones we can't are those that happen
346 * after the transaction's already dirty.
347 */
348 if (tp->t_flags & XFS_TRANS_DIRTY)
349 xfs_force_shutdown(tp->t_mountp,
350 SHUTDOWN_META_IO_ERROR);
351 return error;
352 }
353 }
354 /*
355 * We never locked this buf ourselves, so we shouldn't
356 * brelse it either. Just get out.
357 */
358 if (XFS_FORCED_SHUTDOWN(mp)) {
359 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
360 *bpp = NULL;
361 return XFS_ERROR(EIO);
362 }
363
364
365 bip = bp->b_fspriv;
366 bip->bli_recur++;
367
368 ASSERT(atomic_read(&bip->bli_refcount) > 0);
369 trace_xfs_trans_read_buf_recur(bip);
370 *bpp = bp;
371 return 0;
372 }
373
374 /*
375 * We always specify the XBF_DONT_BLOCK flag within a transaction
376 * so that get_buf does not try to push out a delayed write buffer
377 * which might cause another transaction to take place (if the
378 * buffer was delayed alloc). Such recursive transactions can
379 * easily deadlock with our current transaction as well as cause
380 * us to run out of stack space.
381 */
382 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
383 if (bp == NULL) {
384 *bpp = NULL;
385 return (flags & XBF_TRYLOCK) ?
386 0 : XFS_ERROR(ENOMEM);
387 }
388 if (bp->b_error) {
389 error = bp->b_error;
390 XFS_BUF_SUPER_STALE(bp);
391 xfs_ioerror_alert("xfs_trans_read_buf", mp,
392 bp, blkno);
393 if (tp->t_flags & XFS_TRANS_DIRTY)
394 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
395 xfs_buf_relse(bp);
396 return error;
397 }
398#ifdef DEBUG
399 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
400 if (xfs_error_target == target) {
401 if (((xfs_req_num++) % xfs_error_mod) == 0) {
402 xfs_force_shutdown(tp->t_mountp,
403 SHUTDOWN_META_IO_ERROR);
404 xfs_buf_relse(bp);
405 xfs_debug(mp, "Returning trans error!");
406 return XFS_ERROR(EIO);
407 }
408 }
409 }
410#endif
411 if (XFS_FORCED_SHUTDOWN(mp))
412 goto shutdown_abort;
413
414 _xfs_trans_bjoin(tp, bp, 1);
415 trace_xfs_trans_read_buf(bp->b_fspriv);
416
417 *bpp = bp;
418 return 0;
419
420shutdown_abort:
421 /*
422 * the theory here is that buffer is good but we're
423 * bailing out because the filesystem is being forcibly
424 * shut down. So we should leave the b_flags alone since
425 * the buffer's not staled and just get out.
426 */
427#if defined(DEBUG)
428 if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp))
429 xfs_notice(mp, "about to pop assert, bp == 0x%p", bp);
430#endif
431 ASSERT((bp->b_flags & (XBF_STALE|XBF_DELWRI)) !=
432 (XBF_STALE|XBF_DELWRI));
433
434 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
435 xfs_buf_relse(bp);
436 *bpp = NULL;
437 return XFS_ERROR(EIO);
438}
439
440
441/*
442 * Release the buffer bp which was previously acquired with one of the
443 * xfs_trans_... buffer allocation routines if the buffer has not
444 * been modified within this transaction. If the buffer is modified
445 * within this transaction, do decrement the recursion count but do
446 * not release the buffer even if the count goes to 0. If the buffer is not
447 * modified within the transaction, decrement the recursion count and
448 * release the buffer if the recursion count goes to 0.
449 *
450 * If the buffer is to be released and it was not modified before
451 * this transaction began, then free the buf_log_item associated with it.
452 *
453 * If the transaction pointer is NULL, make this just a normal
454 * brelse() call.
455 */
456void
457xfs_trans_brelse(xfs_trans_t *tp,
458 xfs_buf_t *bp)
459{
460 xfs_buf_log_item_t *bip;
461
462 /*
463 * Default to a normal brelse() call if the tp is NULL.
464 */
465 if (tp == NULL) {
466 struct xfs_log_item *lip = bp->b_fspriv;
467
468 ASSERT(bp->b_transp == NULL);
469
470 /*
471 * If there's a buf log item attached to the buffer,
472 * then let the AIL know that the buffer is being
473 * unlocked.
474 */
475 if (lip != NULL && lip->li_type == XFS_LI_BUF) {
476 bip = bp->b_fspriv;
477 xfs_trans_unlocked_item(bip->bli_item.li_ailp, lip);
478 }
479 xfs_buf_relse(bp);
480 return;
481 }
482
483 ASSERT(bp->b_transp == tp);
484 bip = bp->b_fspriv;
485 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
486 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
487 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
488 ASSERT(atomic_read(&bip->bli_refcount) > 0);
489
490 trace_xfs_trans_brelse(bip);
491
492 /*
493 * If the release is just for a recursive lock,
494 * then decrement the count and return.
495 */
496 if (bip->bli_recur > 0) {
497 bip->bli_recur--;
498 return;
499 }
500
501 /*
502 * If the buffer is dirty within this transaction, we can't
503 * release it until we commit.
504 */
505 if (bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY)
506 return;
507
508 /*
509 * If the buffer has been invalidated, then we can't release
510 * it until the transaction commits to disk unless it is re-dirtied
511 * as part of this transaction. This prevents us from pulling
512 * the item from the AIL before we should.
513 */
514 if (bip->bli_flags & XFS_BLI_STALE)
515 return;
516
517 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
518
519 /*
520 * Free up the log item descriptor tracking the released item.
521 */
522 xfs_trans_del_item(&bip->bli_item);
523
524 /*
525 * Clear the hold flag in the buf log item if it is set.
526 * We wouldn't want the next user of the buffer to
527 * get confused.
528 */
529 if (bip->bli_flags & XFS_BLI_HOLD) {
530 bip->bli_flags &= ~XFS_BLI_HOLD;
531 }
532
533 /*
534 * Drop our reference to the buf log item.
535 */
536 atomic_dec(&bip->bli_refcount);
537
538 /*
539 * If the buf item is not tracking data in the log, then
540 * we must free it before releasing the buffer back to the
541 * free pool. Before releasing the buffer to the free pool,
542 * clear the transaction pointer in b_fsprivate2 to dissolve
543 * its relation to this transaction.
544 */
545 if (!xfs_buf_item_dirty(bip)) {
546/***
547 ASSERT(bp->b_pincount == 0);
548***/
549 ASSERT(atomic_read(&bip->bli_refcount) == 0);
550 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
551 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
552 xfs_buf_item_relse(bp);
553 bip = NULL;
554 }
555 bp->b_transp = NULL;
556
557 /*
558 * If we've still got a buf log item on the buffer, then
559 * tell the AIL that the buffer is being unlocked.
560 */
561 if (bip != NULL) {
562 xfs_trans_unlocked_item(bip->bli_item.li_ailp,
563 (xfs_log_item_t*)bip);
564 }
565
566 xfs_buf_relse(bp);
567 return;
568}
569
570/*
571 * Mark the buffer as not needing to be unlocked when the buf item's
572 * IOP_UNLOCK() routine is called. The buffer must already be locked
573 * and associated with the given transaction.
574 */
575/* ARGSUSED */
576void
577xfs_trans_bhold(xfs_trans_t *tp,
578 xfs_buf_t *bp)
579{
580 xfs_buf_log_item_t *bip = bp->b_fspriv;
581
582 ASSERT(bp->b_transp == tp);
583 ASSERT(bip != NULL);
584 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
585 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
586 ASSERT(atomic_read(&bip->bli_refcount) > 0);
587
588 bip->bli_flags |= XFS_BLI_HOLD;
589 trace_xfs_trans_bhold(bip);
590}
591
592/*
593 * Cancel the previous buffer hold request made on this buffer
594 * for this transaction.
595 */
596void
597xfs_trans_bhold_release(xfs_trans_t *tp,
598 xfs_buf_t *bp)
599{
600 xfs_buf_log_item_t *bip = bp->b_fspriv;
601
602 ASSERT(bp->b_transp == tp);
603 ASSERT(bip != NULL);
604 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
605 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_CANCEL));
606 ASSERT(atomic_read(&bip->bli_refcount) > 0);
607 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
608
609 bip->bli_flags &= ~XFS_BLI_HOLD;
610 trace_xfs_trans_bhold_release(bip);
611}
612
613/*
614 * This is called to mark bytes first through last inclusive of the given
615 * buffer as needing to be logged when the transaction is committed.
616 * The buffer must already be associated with the given transaction.
617 *
618 * First and last are numbers relative to the beginning of this buffer,
619 * so the first byte in the buffer is numbered 0 regardless of the
620 * value of b_blkno.
621 */
622void
623xfs_trans_log_buf(xfs_trans_t *tp,
624 xfs_buf_t *bp,
625 uint first,
626 uint last)
627{
628 xfs_buf_log_item_t *bip = bp->b_fspriv;
629
630 ASSERT(bp->b_transp == tp);
631 ASSERT(bip != NULL);
632 ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp)));
633 ASSERT(bp->b_iodone == NULL ||
634 bp->b_iodone == xfs_buf_iodone_callbacks);
635
636 /*
637 * Mark the buffer as needing to be written out eventually,
638 * and set its iodone function to remove the buffer's buf log
639 * item from the AIL and free it when the buffer is flushed
640 * to disk. See xfs_buf_attach_iodone() for more details
641 * on li_cb and xfs_buf_iodone_callbacks().
642 * If we end up aborting this transaction, we trap this buffer
643 * inside the b_bdstrat callback so that this won't get written to
644 * disk.
645 */
646 XFS_BUF_DELAYWRITE(bp);
647 XFS_BUF_DONE(bp);
648
649 ASSERT(atomic_read(&bip->bli_refcount) > 0);
650 bp->b_iodone = xfs_buf_iodone_callbacks;
651 bip->bli_item.li_cb = xfs_buf_iodone;
652
653 trace_xfs_trans_log_buf(bip);
654
655 /*
656 * If we invalidated the buffer within this transaction, then
657 * cancel the invalidation now that we're dirtying the buffer
658 * again. There are no races with the code in xfs_buf_item_unpin(),
659 * because we have a reference to the buffer this entire time.
660 */
661 if (bip->bli_flags & XFS_BLI_STALE) {
662 bip->bli_flags &= ~XFS_BLI_STALE;
663 ASSERT(XFS_BUF_ISSTALE(bp));
664 XFS_BUF_UNSTALE(bp);
665 bip->bli_format.blf_flags &= ~XFS_BLF_CANCEL;
666 }
667
668 tp->t_flags |= XFS_TRANS_DIRTY;
669 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
670 bip->bli_flags |= XFS_BLI_LOGGED;
671 xfs_buf_item_log(bip, first, last);
672}
673
674
675/*
676 * This called to invalidate a buffer that is being used within
677 * a transaction. Typically this is because the blocks in the
678 * buffer are being freed, so we need to prevent it from being
679 * written out when we're done. Allowing it to be written again
680 * might overwrite data in the free blocks if they are reallocated
681 * to a file.
682 *
683 * We prevent the buffer from being written out by clearing the
684 * B_DELWRI flag. We can't always
685 * get rid of the buf log item at this point, though, because
686 * the buffer may still be pinned by another transaction. If that
687 * is the case, then we'll wait until the buffer is committed to
688 * disk for the last time (we can tell by the ref count) and
689 * free it in xfs_buf_item_unpin(). Until it is cleaned up we
690 * will keep the buffer locked so that the buffer and buf log item
691 * are not reused.
692 */
693void
694xfs_trans_binval(
695 xfs_trans_t *tp,
696 xfs_buf_t *bp)
697{
698 xfs_buf_log_item_t *bip = bp->b_fspriv;
699
700 ASSERT(bp->b_transp == tp);
701 ASSERT(bip != NULL);
702 ASSERT(atomic_read(&bip->bli_refcount) > 0);
703
704 trace_xfs_trans_binval(bip);
705
706 if (bip->bli_flags & XFS_BLI_STALE) {
707 /*
708 * If the buffer is already invalidated, then
709 * just return.
710 */
711 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
712 ASSERT(XFS_BUF_ISSTALE(bp));
713 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
714 ASSERT(!(bip->bli_format.blf_flags & XFS_BLF_INODE_BUF));
715 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
716 ASSERT(bip->bli_item.li_desc->lid_flags & XFS_LID_DIRTY);
717 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
718 return;
719 }
720
721 /*
722 * Clear the dirty bit in the buffer and set the STALE flag
723 * in the buf log item. The STALE flag will be used in
724 * xfs_buf_item_unpin() to determine if it should clean up
725 * when the last reference to the buf item is given up.
726 * We set the XFS_BLF_CANCEL flag in the buf log format structure
727 * and log the buf item. This will be used at recovery time
728 * to determine that copies of the buffer in the log before
729 * this should not be replayed.
730 * We mark the item descriptor and the transaction dirty so
731 * that we'll hold the buffer until after the commit.
732 *
733 * Since we're invalidating the buffer, we also clear the state
734 * about which parts of the buffer have been logged. We also
735 * clear the flag indicating that this is an inode buffer since
736 * the data in the buffer will no longer be valid.
737 *
738 * We set the stale bit in the buffer as well since we're getting
739 * rid of it.
740 */
741 XFS_BUF_UNDELAYWRITE(bp);
742 XFS_BUF_STALE(bp);
743 bip->bli_flags |= XFS_BLI_STALE;
744 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
745 bip->bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
746 bip->bli_format.blf_flags |= XFS_BLF_CANCEL;
747 memset((char *)(bip->bli_format.blf_data_map), 0,
748 (bip->bli_format.blf_map_size * sizeof(uint)));
749 bip->bli_item.li_desc->lid_flags |= XFS_LID_DIRTY;
750 tp->t_flags |= XFS_TRANS_DIRTY;
751}
752
753/*
754 * This call is used to indicate that the buffer contains on-disk inodes which
755 * must be handled specially during recovery. They require special handling
756 * because only the di_next_unlinked from the inodes in the buffer should be
757 * recovered. The rest of the data in the buffer is logged via the inodes
758 * themselves.
759 *
760 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
761 * transferred to the buffer's log format structure so that we'll know what to
762 * do at recovery time.
763 */
764void
765xfs_trans_inode_buf(
766 xfs_trans_t *tp,
767 xfs_buf_t *bp)
768{
769 xfs_buf_log_item_t *bip = bp->b_fspriv;
770
771 ASSERT(bp->b_transp == tp);
772 ASSERT(bip != NULL);
773 ASSERT(atomic_read(&bip->bli_refcount) > 0);
774
775 bip->bli_flags |= XFS_BLI_INODE_BUF;
776}
777
778/*
779 * This call is used to indicate that the buffer is going to
780 * be staled and was an inode buffer. This means it gets
781 * special processing during unpin - where any inodes
782 * associated with the buffer should be removed from ail.
783 * There is also special processing during recovery,
784 * any replay of the inodes in the buffer needs to be
785 * prevented as the buffer may have been reused.
786 */
787void
788xfs_trans_stale_inode_buf(
789 xfs_trans_t *tp,
790 xfs_buf_t *bp)
791{
792 xfs_buf_log_item_t *bip = bp->b_fspriv;
793
794 ASSERT(bp->b_transp == tp);
795 ASSERT(bip != NULL);
796 ASSERT(atomic_read(&bip->bli_refcount) > 0);
797
798 bip->bli_flags |= XFS_BLI_STALE_INODE;
799 bip->bli_item.li_cb = xfs_buf_iodone;
800}
801
802/*
803 * Mark the buffer as being one which contains newly allocated
804 * inodes. We need to make sure that even if this buffer is
805 * relogged as an 'inode buf' we still recover all of the inode
806 * images in the face of a crash. This works in coordination with
807 * xfs_buf_item_committed() to ensure that the buffer remains in the
808 * AIL at its original location even after it has been relogged.
809 */
810/* ARGSUSED */
811void
812xfs_trans_inode_alloc_buf(
813 xfs_trans_t *tp,
814 xfs_buf_t *bp)
815{
816 xfs_buf_log_item_t *bip = bp->b_fspriv;
817
818 ASSERT(bp->b_transp == tp);
819 ASSERT(bip != NULL);
820 ASSERT(atomic_read(&bip->bli_refcount) > 0);
821
822 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
823}
824
825
826/*
827 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
828 * dquots. However, unlike in inode buffer recovery, dquot buffers get
829 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
830 * The only thing that makes dquot buffers different from regular
831 * buffers is that we must not replay dquot bufs when recovering
832 * if a _corresponding_ quotaoff has happened. We also have to distinguish
833 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
834 * can be turned off independently.
835 */
836/* ARGSUSED */
837void
838xfs_trans_dquot_buf(
839 xfs_trans_t *tp,
840 xfs_buf_t *bp,
841 uint type)
842{
843 xfs_buf_log_item_t *bip = bp->b_fspriv;
844
845 ASSERT(bp->b_transp == tp);
846 ASSERT(bip != NULL);
847 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
848 type == XFS_BLF_PDQUOT_BUF ||
849 type == XFS_BLF_GDQUOT_BUF);
850 ASSERT(atomic_read(&bip->bli_refcount) > 0);
851
852 bip->bli_format.blf_flags |= type;
853}