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 * Forcibly detach a buffer previously joined to the transaction. The caller
397 * will retain its locked reference to the buffer after this function returns.
398 * The buffer must be completely clean and must not be held to the transaction.
399 */
400void
401xfs_trans_bdetach(
402 struct xfs_trans *tp,
403 struct xfs_buf *bp)
404{
405 struct xfs_buf_log_item *bip = bp->b_log_item;
406
407 ASSERT(tp != NULL);
408 ASSERT(bp->b_transp == tp);
409 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
410 ASSERT(atomic_read(&bip->bli_refcount) > 0);
411
412 trace_xfs_trans_bdetach(bip);
413
414 /*
415 * Erase all recursion count, since we're removing this buffer from the
416 * transaction.
417 */
418 bip->bli_recur = 0;
419
420 /*
421 * The buffer must be completely clean. Specifically, it had better
422 * not be dirty, stale, logged, ordered, or held to the transaction.
423 */
424 ASSERT(!test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
425 ASSERT(!(bip->bli_flags & XFS_BLI_DIRTY));
426 ASSERT(!(bip->bli_flags & XFS_BLI_HOLD));
427 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
428 ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
429 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
430
431 /* Unlink the log item from the transaction and drop the log item. */
432 xfs_trans_del_item(&bip->bli_item);
433 xfs_buf_item_put(bip);
434 bp->b_transp = NULL;
435}
436
437/*
438 * Mark the buffer as not needing to be unlocked when the buf item's
439 * iop_committing() routine is called. The buffer must already be locked
440 * and associated with the given transaction.
441 */
442/* ARGSUSED */
443void
444xfs_trans_bhold(
445 xfs_trans_t *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 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
453 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
454 ASSERT(atomic_read(&bip->bli_refcount) > 0);
455
456 bip->bli_flags |= XFS_BLI_HOLD;
457 trace_xfs_trans_bhold(bip);
458}
459
460/*
461 * Cancel the previous buffer hold request made on this buffer
462 * for this transaction.
463 */
464void
465xfs_trans_bhold_release(
466 xfs_trans_t *tp,
467 struct xfs_buf *bp)
468{
469 struct xfs_buf_log_item *bip = bp->b_log_item;
470
471 ASSERT(bp->b_transp == tp);
472 ASSERT(bip != NULL);
473 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
474 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
475 ASSERT(atomic_read(&bip->bli_refcount) > 0);
476 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
477
478 bip->bli_flags &= ~XFS_BLI_HOLD;
479 trace_xfs_trans_bhold_release(bip);
480}
481
482/*
483 * Mark a buffer dirty in the transaction.
484 */
485void
486xfs_trans_dirty_buf(
487 struct xfs_trans *tp,
488 struct xfs_buf *bp)
489{
490 struct xfs_buf_log_item *bip = bp->b_log_item;
491
492 ASSERT(bp->b_transp == tp);
493 ASSERT(bip != NULL);
494
495 /*
496 * Mark the buffer as needing to be written out eventually,
497 * and set its iodone function to remove the buffer's buf log
498 * item from the AIL and free it when the buffer is flushed
499 * to disk.
500 */
501 bp->b_flags |= XBF_DONE;
502
503 ASSERT(atomic_read(&bip->bli_refcount) > 0);
504
505 /*
506 * If we invalidated the buffer within this transaction, then
507 * cancel the invalidation now that we're dirtying the buffer
508 * again. There are no races with the code in xfs_buf_item_unpin(),
509 * because we have a reference to the buffer this entire time.
510 */
511 if (bip->bli_flags & XFS_BLI_STALE) {
512 bip->bli_flags &= ~XFS_BLI_STALE;
513 ASSERT(bp->b_flags & XBF_STALE);
514 bp->b_flags &= ~XBF_STALE;
515 bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
516 }
517 bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
518
519 tp->t_flags |= XFS_TRANS_DIRTY;
520 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
521}
522
523/*
524 * This is called to mark bytes first through last inclusive of the given
525 * buffer as needing to be logged when the transaction is committed.
526 * The buffer must already be associated with the given transaction.
527 *
528 * First and last are numbers relative to the beginning of this buffer,
529 * so the first byte in the buffer is numbered 0 regardless of the
530 * value of b_blkno.
531 */
532void
533xfs_trans_log_buf(
534 struct xfs_trans *tp,
535 struct xfs_buf *bp,
536 uint first,
537 uint last)
538{
539 struct xfs_buf_log_item *bip = bp->b_log_item;
540
541 ASSERT(first <= last && last < BBTOB(bp->b_length));
542 ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
543
544 xfs_trans_dirty_buf(tp, bp);
545
546 trace_xfs_trans_log_buf(bip);
547 xfs_buf_item_log(bip, first, last);
548}
549
550
551/*
552 * Invalidate a buffer that is being used within a transaction.
553 *
554 * Typically this is because the blocks in the buffer are being freed, so we
555 * need to prevent it from being written out when we're done. Allowing it
556 * to be written again might overwrite data in the free blocks if they are
557 * reallocated to a file.
558 *
559 * We prevent the buffer from being written out by marking it stale. We can't
560 * get rid of the buf log item at this point because the buffer may still be
561 * pinned by another transaction. If that is the case, then we'll wait until
562 * the buffer is committed to disk for the last time (we can tell by the ref
563 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
564 * keep the buffer locked so that the buffer and buf log item are not reused.
565 *
566 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
567 * the buf item. This will be used at recovery time to determine that copies
568 * of the buffer in the log before this should not be replayed.
569 *
570 * We mark the item descriptor and the transaction dirty so that we'll hold
571 * the buffer until after the commit.
572 *
573 * Since we're invalidating the buffer, we also clear the state about which
574 * parts of the buffer have been logged. We also clear the flag indicating
575 * that this is an inode buffer since the data in the buffer will no longer
576 * be valid.
577 *
578 * We set the stale bit in the buffer as well since we're getting rid of it.
579 */
580void
581xfs_trans_binval(
582 xfs_trans_t *tp,
583 struct xfs_buf *bp)
584{
585 struct xfs_buf_log_item *bip = bp->b_log_item;
586 int i;
587
588 ASSERT(bp->b_transp == tp);
589 ASSERT(bip != NULL);
590 ASSERT(atomic_read(&bip->bli_refcount) > 0);
591
592 trace_xfs_trans_binval(bip);
593
594 if (bip->bli_flags & XFS_BLI_STALE) {
595 /*
596 * If the buffer is already invalidated, then
597 * just return.
598 */
599 ASSERT(bp->b_flags & XBF_STALE);
600 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
601 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
602 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
603 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
604 ASSERT(test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
605 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
606 return;
607 }
608
609 xfs_buf_stale(bp);
610
611 bip->bli_flags |= XFS_BLI_STALE;
612 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
613 bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
614 bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
615 bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
616 for (i = 0; i < bip->bli_format_count; i++) {
617 memset(bip->bli_formats[i].blf_data_map, 0,
618 (bip->bli_formats[i].blf_map_size * sizeof(uint)));
619 }
620 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
621 tp->t_flags |= XFS_TRANS_DIRTY;
622}
623
624/*
625 * This call is used to indicate that the buffer contains on-disk inodes which
626 * must be handled specially during recovery. They require special handling
627 * because only the di_next_unlinked from the inodes in the buffer should be
628 * recovered. The rest of the data in the buffer is logged via the inodes
629 * themselves.
630 *
631 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
632 * transferred to the buffer's log format structure so that we'll know what to
633 * do at recovery time.
634 */
635void
636xfs_trans_inode_buf(
637 xfs_trans_t *tp,
638 struct xfs_buf *bp)
639{
640 struct xfs_buf_log_item *bip = bp->b_log_item;
641
642 ASSERT(bp->b_transp == tp);
643 ASSERT(bip != NULL);
644 ASSERT(atomic_read(&bip->bli_refcount) > 0);
645
646 bip->bli_flags |= XFS_BLI_INODE_BUF;
647 bp->b_flags |= _XBF_INODES;
648 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
649}
650
651/*
652 * This call is used to indicate that the buffer is going to
653 * be staled and was an inode buffer. This means it gets
654 * special processing during unpin - where any inodes
655 * associated with the buffer should be removed from ail.
656 * There is also special processing during recovery,
657 * any replay of the inodes in the buffer needs to be
658 * prevented as the buffer may have been reused.
659 */
660void
661xfs_trans_stale_inode_buf(
662 xfs_trans_t *tp,
663 struct xfs_buf *bp)
664{
665 struct xfs_buf_log_item *bip = bp->b_log_item;
666
667 ASSERT(bp->b_transp == tp);
668 ASSERT(bip != NULL);
669 ASSERT(atomic_read(&bip->bli_refcount) > 0);
670
671 bip->bli_flags |= XFS_BLI_STALE_INODE;
672 bp->b_flags |= _XBF_INODES;
673 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
674}
675
676/*
677 * Mark the buffer as being one which contains newly allocated
678 * inodes. We need to make sure that even if this buffer is
679 * relogged as an 'inode buf' we still recover all of the inode
680 * images in the face of a crash. This works in coordination with
681 * xfs_buf_item_committed() to ensure that the buffer remains in the
682 * AIL at its original location even after it has been relogged.
683 */
684/* ARGSUSED */
685void
686xfs_trans_inode_alloc_buf(
687 xfs_trans_t *tp,
688 struct xfs_buf *bp)
689{
690 struct xfs_buf_log_item *bip = bp->b_log_item;
691
692 ASSERT(bp->b_transp == tp);
693 ASSERT(bip != NULL);
694 ASSERT(atomic_read(&bip->bli_refcount) > 0);
695
696 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
697 bp->b_flags |= _XBF_INODES;
698 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
699}
700
701/*
702 * Mark the buffer as ordered for this transaction. This means that the contents
703 * of the buffer are not recorded in the transaction but it is tracked in the
704 * AIL as though it was. This allows us to record logical changes in
705 * transactions rather than the physical changes we make to the buffer without
706 * changing writeback ordering constraints of metadata buffers.
707 */
708bool
709xfs_trans_ordered_buf(
710 struct xfs_trans *tp,
711 struct xfs_buf *bp)
712{
713 struct xfs_buf_log_item *bip = bp->b_log_item;
714
715 ASSERT(bp->b_transp == tp);
716 ASSERT(bip != NULL);
717 ASSERT(atomic_read(&bip->bli_refcount) > 0);
718
719 if (xfs_buf_item_dirty_format(bip))
720 return false;
721
722 bip->bli_flags |= XFS_BLI_ORDERED;
723 trace_xfs_buf_item_ordered(bip);
724
725 /*
726 * We don't log a dirty range of an ordered buffer but it still needs
727 * to be marked dirty and that it has been logged.
728 */
729 xfs_trans_dirty_buf(tp, bp);
730 return true;
731}
732
733/*
734 * Set the type of the buffer for log recovery so that it can correctly identify
735 * and hence attach the correct buffer ops to the buffer after replay.
736 */
737void
738xfs_trans_buf_set_type(
739 struct xfs_trans *tp,
740 struct xfs_buf *bp,
741 enum xfs_blft type)
742{
743 struct xfs_buf_log_item *bip = bp->b_log_item;
744
745 if (!tp)
746 return;
747
748 ASSERT(bp->b_transp == tp);
749 ASSERT(bip != NULL);
750 ASSERT(atomic_read(&bip->bli_refcount) > 0);
751
752 xfs_blft_to_flags(&bip->__bli_format, type);
753}
754
755void
756xfs_trans_buf_copy_type(
757 struct xfs_buf *dst_bp,
758 struct xfs_buf *src_bp)
759{
760 struct xfs_buf_log_item *sbip = src_bp->b_log_item;
761 struct xfs_buf_log_item *dbip = dst_bp->b_log_item;
762 enum xfs_blft type;
763
764 type = xfs_blft_from_flags(&sbip->__bli_format);
765 xfs_blft_to_flags(&dbip->__bli_format, type);
766}
767
768/*
769 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
770 * dquots. However, unlike in inode buffer recovery, dquot buffers get
771 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
772 * The only thing that makes dquot buffers different from regular
773 * buffers is that we must not replay dquot bufs when recovering
774 * if a _corresponding_ quotaoff has happened. We also have to distinguish
775 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
776 * can be turned off independently.
777 */
778/* ARGSUSED */
779void
780xfs_trans_dquot_buf(
781 xfs_trans_t *tp,
782 struct xfs_buf *bp,
783 uint type)
784{
785 struct xfs_buf_log_item *bip = bp->b_log_item;
786
787 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
788 type == XFS_BLF_PDQUOT_BUF ||
789 type == XFS_BLF_GDQUOT_BUF);
790
791 bip->__bli_format.blf_flags |= type;
792
793 switch (type) {
794 case XFS_BLF_UDQUOT_BUF:
795 type = XFS_BLFT_UDQUOT_BUF;
796 break;
797 case XFS_BLF_PDQUOT_BUF:
798 type = XFS_BLFT_PDQUOT_BUF;
799 break;
800 case XFS_BLF_GDQUOT_BUF:
801 type = XFS_BLFT_GDQUOT_BUF;
802 break;
803 default:
804 type = XFS_BLFT_UNKNOWN_BUF;
805 break;
806 }
807
808 bp->b_flags |= _XBF_DQUOTS;
809 xfs_trans_buf_set_type(tp, bp, type);
810}
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 */
171static struct xfs_buf *
172__xfs_trans_getsb(
173 struct xfs_trans *tp,
174 struct xfs_buf *bp)
175{
176 /*
177 * Just increment the lock recursion count if the buffer is already
178 * attached to this transaction.
179 */
180 if (bp->b_transp == tp) {
181 struct xfs_buf_log_item *bip = bp->b_log_item;
182
183 ASSERT(bip != NULL);
184 ASSERT(atomic_read(&bip->bli_refcount) > 0);
185 bip->bli_recur++;
186
187 trace_xfs_trans_getsb_recur(bip);
188 } else {
189 xfs_buf_lock(bp);
190 xfs_buf_hold(bp);
191 _xfs_trans_bjoin(tp, bp, 1);
192
193 trace_xfs_trans_getsb(bp->b_log_item);
194 }
195
196 return bp;
197}
198
199struct xfs_buf *
200xfs_trans_getsb(
201 struct xfs_trans *tp)
202{
203 return __xfs_trans_getsb(tp, tp->t_mountp->m_sb_bp);
204}
205
206struct xfs_buf *
207xfs_trans_getrtsb(
208 struct xfs_trans *tp)
209{
210 if (!tp->t_mountp->m_rtsb_bp)
211 return NULL;
212 return __xfs_trans_getsb(tp, tp->t_mountp->m_rtsb_bp);
213}
214
215/*
216 * Get and lock the buffer for the caller if it is not already
217 * locked within the given transaction. If it has not yet been
218 * read in, read it from disk. If it is already locked
219 * within the transaction and already read in, just increment its
220 * lock recursion count and return a pointer to it.
221 *
222 * If the transaction pointer is NULL, make this just a normal
223 * read_buf() call.
224 */
225int
226xfs_trans_read_buf_map(
227 struct xfs_mount *mp,
228 struct xfs_trans *tp,
229 struct xfs_buftarg *target,
230 struct xfs_buf_map *map,
231 int nmaps,
232 xfs_buf_flags_t flags,
233 struct xfs_buf **bpp,
234 const struct xfs_buf_ops *ops)
235{
236 struct xfs_buf *bp = NULL;
237 struct xfs_buf_log_item *bip;
238 int error;
239
240 *bpp = NULL;
241 /*
242 * If we find the buffer in the cache with this transaction
243 * pointer in its b_fsprivate2 field, then we know we already
244 * have it locked. If it is already read in we just increment
245 * the lock recursion count and return the buffer to the caller.
246 * If the buffer is not yet read in, then we read it in, increment
247 * the lock recursion count, and return it to the caller.
248 */
249 if (tp)
250 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
251 if (bp) {
252 ASSERT(xfs_buf_islocked(bp));
253 ASSERT(bp->b_transp == tp);
254 ASSERT(bp->b_log_item != NULL);
255 ASSERT(!bp->b_error);
256 ASSERT(bp->b_flags & XBF_DONE);
257
258 /*
259 * We never locked this buf ourselves, so we shouldn't
260 * brelse it either. Just get out.
261 */
262 if (xfs_is_shutdown(mp)) {
263 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
264 return -EIO;
265 }
266
267 /*
268 * Check if the caller is trying to read a buffer that is
269 * already attached to the transaction yet has no buffer ops
270 * assigned. Ops are usually attached when the buffer is
271 * attached to the transaction, or by the read caller if
272 * special circumstances. That didn't happen, which is not
273 * how this is supposed to go.
274 *
275 * If the buffer passes verification we'll let this go, but if
276 * not we have to shut down. Let the transaction cleanup code
277 * release this buffer when it kills the tranaction.
278 */
279 ASSERT(bp->b_ops != NULL);
280 error = xfs_buf_reverify(bp, ops);
281 if (error) {
282 xfs_buf_ioerror_alert(bp, __return_address);
283
284 if (tp->t_flags & XFS_TRANS_DIRTY)
285 xfs_force_shutdown(tp->t_mountp,
286 SHUTDOWN_META_IO_ERROR);
287
288 /* bad CRC means corrupted metadata */
289 if (error == -EFSBADCRC)
290 error = -EFSCORRUPTED;
291 return error;
292 }
293
294 bip = bp->b_log_item;
295 bip->bli_recur++;
296
297 ASSERT(atomic_read(&bip->bli_refcount) > 0);
298 trace_xfs_trans_read_buf_recur(bip);
299 ASSERT(bp->b_ops != NULL || ops == NULL);
300 *bpp = bp;
301 return 0;
302 }
303
304 error = xfs_buf_read_map(target, map, nmaps, flags, &bp, ops,
305 __return_address);
306 switch (error) {
307 case 0:
308 break;
309 default:
310 if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
311 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
312 fallthrough;
313 case -ENOMEM:
314 case -EAGAIN:
315 return error;
316 }
317
318 if (xfs_is_shutdown(mp)) {
319 xfs_buf_relse(bp);
320 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
321 return -EIO;
322 }
323
324 if (tp) {
325 _xfs_trans_bjoin(tp, bp, 1);
326 trace_xfs_trans_read_buf(bp->b_log_item);
327 }
328 ASSERT(bp->b_ops != NULL || ops == NULL);
329 *bpp = bp;
330 return 0;
331
332}
333
334/* Has this buffer been dirtied by anyone? */
335bool
336xfs_trans_buf_is_dirty(
337 struct xfs_buf *bp)
338{
339 struct xfs_buf_log_item *bip = bp->b_log_item;
340
341 if (!bip)
342 return false;
343 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
344 return test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
345}
346
347/*
348 * Release a buffer previously joined to the transaction. If the buffer is
349 * modified within this transaction, decrement the recursion count but do not
350 * release the buffer even if the count goes to 0. If the buffer is not modified
351 * within the transaction, decrement the recursion count and release the buffer
352 * if the recursion count goes to 0.
353 *
354 * If the buffer is to be released and it was not already dirty before this
355 * transaction began, then also free the buf_log_item associated with it.
356 *
357 * If the transaction pointer is NULL, this is a normal xfs_buf_relse() call.
358 */
359void
360xfs_trans_brelse(
361 struct xfs_trans *tp,
362 struct xfs_buf *bp)
363{
364 struct xfs_buf_log_item *bip = bp->b_log_item;
365
366 ASSERT(bp->b_transp == tp);
367
368 if (!tp) {
369 xfs_buf_relse(bp);
370 return;
371 }
372
373 trace_xfs_trans_brelse(bip);
374 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
375 ASSERT(atomic_read(&bip->bli_refcount) > 0);
376
377 /*
378 * If the release is for a recursive lookup, then decrement the count
379 * and return.
380 */
381 if (bip->bli_recur > 0) {
382 bip->bli_recur--;
383 return;
384 }
385
386 /*
387 * If the buffer is invalidated or dirty in this transaction, we can't
388 * release it until we commit.
389 */
390 if (test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags))
391 return;
392 if (bip->bli_flags & XFS_BLI_STALE)
393 return;
394
395 /*
396 * Unlink the log item from the transaction and clear the hold flag, if
397 * set. We wouldn't want the next user of the buffer to get confused.
398 */
399 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
400 xfs_trans_del_item(&bip->bli_item);
401 bip->bli_flags &= ~XFS_BLI_HOLD;
402
403 /* drop the reference to the bli */
404 xfs_buf_item_put(bip);
405
406 bp->b_transp = NULL;
407 xfs_buf_relse(bp);
408}
409
410/*
411 * Forcibly detach a buffer previously joined to the transaction. The caller
412 * will retain its locked reference to the buffer after this function returns.
413 * The buffer must be completely clean and must not be held to the transaction.
414 */
415void
416xfs_trans_bdetach(
417 struct xfs_trans *tp,
418 struct xfs_buf *bp)
419{
420 struct xfs_buf_log_item *bip = bp->b_log_item;
421
422 ASSERT(tp != NULL);
423 ASSERT(bp->b_transp == tp);
424 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
425 ASSERT(atomic_read(&bip->bli_refcount) > 0);
426
427 trace_xfs_trans_bdetach(bip);
428
429 /*
430 * Erase all recursion count, since we're removing this buffer from the
431 * transaction.
432 */
433 bip->bli_recur = 0;
434
435 /*
436 * The buffer must be completely clean. Specifically, it had better
437 * not be dirty, stale, logged, ordered, or held to the transaction.
438 */
439 ASSERT(!test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
440 ASSERT(!(bip->bli_flags & XFS_BLI_DIRTY));
441 ASSERT(!(bip->bli_flags & XFS_BLI_HOLD));
442 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
443 ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
444 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
445
446 /* Unlink the log item from the transaction and drop the log item. */
447 xfs_trans_del_item(&bip->bli_item);
448 xfs_buf_item_put(bip);
449 bp->b_transp = NULL;
450}
451
452/*
453 * Mark the buffer as not needing to be unlocked when the buf item's
454 * iop_committing() routine is called. The buffer must already be locked
455 * and associated with the given transaction.
456 */
457/* ARGSUSED */
458void
459xfs_trans_bhold(
460 xfs_trans_t *tp,
461 struct xfs_buf *bp)
462{
463 struct xfs_buf_log_item *bip = bp->b_log_item;
464
465 ASSERT(bp->b_transp == tp);
466 ASSERT(bip != NULL);
467 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
468 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
469 ASSERT(atomic_read(&bip->bli_refcount) > 0);
470
471 bip->bli_flags |= XFS_BLI_HOLD;
472 trace_xfs_trans_bhold(bip);
473}
474
475/*
476 * Cancel the previous buffer hold request made on this buffer
477 * for this transaction.
478 */
479void
480xfs_trans_bhold_release(
481 xfs_trans_t *tp,
482 struct xfs_buf *bp)
483{
484 struct xfs_buf_log_item *bip = bp->b_log_item;
485
486 ASSERT(bp->b_transp == tp);
487 ASSERT(bip != NULL);
488 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
489 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
490 ASSERT(atomic_read(&bip->bli_refcount) > 0);
491 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
492
493 bip->bli_flags &= ~XFS_BLI_HOLD;
494 trace_xfs_trans_bhold_release(bip);
495}
496
497/*
498 * Mark a buffer dirty in the transaction.
499 */
500void
501xfs_trans_dirty_buf(
502 struct xfs_trans *tp,
503 struct xfs_buf *bp)
504{
505 struct xfs_buf_log_item *bip = bp->b_log_item;
506
507 ASSERT(bp->b_transp == tp);
508 ASSERT(bip != NULL);
509
510 /*
511 * Mark the buffer as needing to be written out eventually,
512 * and set its iodone function to remove the buffer's buf log
513 * item from the AIL and free it when the buffer is flushed
514 * to disk.
515 */
516 bp->b_flags |= XBF_DONE;
517
518 ASSERT(atomic_read(&bip->bli_refcount) > 0);
519
520 /*
521 * If we invalidated the buffer within this transaction, then
522 * cancel the invalidation now that we're dirtying the buffer
523 * again. There are no races with the code in xfs_buf_item_unpin(),
524 * because we have a reference to the buffer this entire time.
525 */
526 if (bip->bli_flags & XFS_BLI_STALE) {
527 bip->bli_flags &= ~XFS_BLI_STALE;
528 ASSERT(bp->b_flags & XBF_STALE);
529 bp->b_flags &= ~XBF_STALE;
530 bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
531 }
532 bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
533
534 tp->t_flags |= XFS_TRANS_DIRTY;
535 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
536}
537
538/*
539 * This is called to mark bytes first through last inclusive of the given
540 * buffer as needing to be logged when the transaction is committed.
541 * The buffer must already be associated with the given transaction.
542 *
543 * First and last are numbers relative to the beginning of this buffer,
544 * so the first byte in the buffer is numbered 0 regardless of the
545 * value of b_blkno.
546 */
547void
548xfs_trans_log_buf(
549 struct xfs_trans *tp,
550 struct xfs_buf *bp,
551 uint first,
552 uint last)
553{
554 struct xfs_buf_log_item *bip = bp->b_log_item;
555
556 ASSERT(first <= last && last < BBTOB(bp->b_length));
557 ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
558
559 xfs_trans_dirty_buf(tp, bp);
560
561 trace_xfs_trans_log_buf(bip);
562 xfs_buf_item_log(bip, first, last);
563}
564
565
566/*
567 * Invalidate a buffer that is being used within a transaction.
568 *
569 * Typically this is because the blocks in the buffer are being freed, so we
570 * need to prevent it from being written out when we're done. Allowing it
571 * to be written again might overwrite data in the free blocks if they are
572 * reallocated to a file.
573 *
574 * We prevent the buffer from being written out by marking it stale. We can't
575 * get rid of the buf log item at this point because the buffer may still be
576 * pinned by another transaction. If that is the case, then we'll wait until
577 * the buffer is committed to disk for the last time (we can tell by the ref
578 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
579 * keep the buffer locked so that the buffer and buf log item are not reused.
580 *
581 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
582 * the buf item. This will be used at recovery time to determine that copies
583 * of the buffer in the log before this should not be replayed.
584 *
585 * We mark the item descriptor and the transaction dirty so that we'll hold
586 * the buffer until after the commit.
587 *
588 * Since we're invalidating the buffer, we also clear the state about which
589 * parts of the buffer have been logged. We also clear the flag indicating
590 * that this is an inode buffer since the data in the buffer will no longer
591 * be valid.
592 *
593 * We set the stale bit in the buffer as well since we're getting rid of it.
594 */
595void
596xfs_trans_binval(
597 xfs_trans_t *tp,
598 struct xfs_buf *bp)
599{
600 struct xfs_buf_log_item *bip = bp->b_log_item;
601 int i;
602
603 ASSERT(bp->b_transp == tp);
604 ASSERT(bip != NULL);
605 ASSERT(atomic_read(&bip->bli_refcount) > 0);
606
607 trace_xfs_trans_binval(bip);
608
609 if (bip->bli_flags & XFS_BLI_STALE) {
610 /*
611 * If the buffer is already invalidated, then
612 * just return.
613 */
614 ASSERT(bp->b_flags & XBF_STALE);
615 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
616 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
617 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
618 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
619 ASSERT(test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
620 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
621 return;
622 }
623
624 xfs_buf_stale(bp);
625
626 bip->bli_flags |= XFS_BLI_STALE;
627 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
628 bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
629 bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
630 bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
631 for (i = 0; i < bip->bli_format_count; i++) {
632 memset(bip->bli_formats[i].blf_data_map, 0,
633 (bip->bli_formats[i].blf_map_size * sizeof(uint)));
634 }
635 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
636 tp->t_flags |= XFS_TRANS_DIRTY;
637}
638
639/*
640 * This call is used to indicate that the buffer contains on-disk inodes which
641 * must be handled specially during recovery. They require special handling
642 * because only the di_next_unlinked from the inodes in the buffer should be
643 * recovered. The rest of the data in the buffer is logged via the inodes
644 * themselves.
645 *
646 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
647 * transferred to the buffer's log format structure so that we'll know what to
648 * do at recovery time.
649 */
650void
651xfs_trans_inode_buf(
652 xfs_trans_t *tp,
653 struct xfs_buf *bp)
654{
655 struct xfs_buf_log_item *bip = bp->b_log_item;
656
657 ASSERT(bp->b_transp == tp);
658 ASSERT(bip != NULL);
659 ASSERT(atomic_read(&bip->bli_refcount) > 0);
660
661 bip->bli_flags |= XFS_BLI_INODE_BUF;
662 bp->b_flags |= _XBF_INODES;
663 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
664}
665
666/*
667 * This call is used to indicate that the buffer is going to
668 * be staled and was an inode buffer. This means it gets
669 * special processing during unpin - where any inodes
670 * associated with the buffer should be removed from ail.
671 * There is also special processing during recovery,
672 * any replay of the inodes in the buffer needs to be
673 * prevented as the buffer may have been reused.
674 */
675void
676xfs_trans_stale_inode_buf(
677 xfs_trans_t *tp,
678 struct xfs_buf *bp)
679{
680 struct xfs_buf_log_item *bip = bp->b_log_item;
681
682 ASSERT(bp->b_transp == tp);
683 ASSERT(bip != NULL);
684 ASSERT(atomic_read(&bip->bli_refcount) > 0);
685
686 bip->bli_flags |= XFS_BLI_STALE_INODE;
687 bp->b_flags |= _XBF_INODES;
688 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
689}
690
691/*
692 * Mark the buffer as being one which contains newly allocated
693 * inodes. We need to make sure that even if this buffer is
694 * relogged as an 'inode buf' we still recover all of the inode
695 * images in the face of a crash. This works in coordination with
696 * xfs_buf_item_committed() to ensure that the buffer remains in the
697 * AIL at its original location even after it has been relogged.
698 */
699/* ARGSUSED */
700void
701xfs_trans_inode_alloc_buf(
702 xfs_trans_t *tp,
703 struct xfs_buf *bp)
704{
705 struct xfs_buf_log_item *bip = bp->b_log_item;
706
707 ASSERT(bp->b_transp == tp);
708 ASSERT(bip != NULL);
709 ASSERT(atomic_read(&bip->bli_refcount) > 0);
710
711 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
712 bp->b_flags |= _XBF_INODES;
713 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
714}
715
716/*
717 * Mark the buffer as ordered for this transaction. This means that the contents
718 * of the buffer are not recorded in the transaction but it is tracked in the
719 * AIL as though it was. This allows us to record logical changes in
720 * transactions rather than the physical changes we make to the buffer without
721 * changing writeback ordering constraints of metadata buffers.
722 */
723bool
724xfs_trans_ordered_buf(
725 struct xfs_trans *tp,
726 struct xfs_buf *bp)
727{
728 struct xfs_buf_log_item *bip = bp->b_log_item;
729
730 ASSERT(bp->b_transp == tp);
731 ASSERT(bip != NULL);
732 ASSERT(atomic_read(&bip->bli_refcount) > 0);
733
734 if (xfs_buf_item_dirty_format(bip))
735 return false;
736
737 bip->bli_flags |= XFS_BLI_ORDERED;
738 trace_xfs_buf_item_ordered(bip);
739
740 /*
741 * We don't log a dirty range of an ordered buffer but it still needs
742 * to be marked dirty and that it has been logged.
743 */
744 xfs_trans_dirty_buf(tp, bp);
745 return true;
746}
747
748/*
749 * Set the type of the buffer for log recovery so that it can correctly identify
750 * and hence attach the correct buffer ops to the buffer after replay.
751 */
752void
753xfs_trans_buf_set_type(
754 struct xfs_trans *tp,
755 struct xfs_buf *bp,
756 enum xfs_blft type)
757{
758 struct xfs_buf_log_item *bip = bp->b_log_item;
759
760 if (!tp)
761 return;
762
763 ASSERT(bp->b_transp == tp);
764 ASSERT(bip != NULL);
765 ASSERT(atomic_read(&bip->bli_refcount) > 0);
766
767 xfs_blft_to_flags(&bip->__bli_format, type);
768}
769
770void
771xfs_trans_buf_copy_type(
772 struct xfs_buf *dst_bp,
773 struct xfs_buf *src_bp)
774{
775 struct xfs_buf_log_item *sbip = src_bp->b_log_item;
776 struct xfs_buf_log_item *dbip = dst_bp->b_log_item;
777 enum xfs_blft type;
778
779 type = xfs_blft_from_flags(&sbip->__bli_format);
780 xfs_blft_to_flags(&dbip->__bli_format, type);
781}
782
783/*
784 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
785 * dquots. However, unlike in inode buffer recovery, dquot buffers get
786 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
787 * The only thing that makes dquot buffers different from regular
788 * buffers is that we must not replay dquot bufs when recovering
789 * if a _corresponding_ quotaoff has happened. We also have to distinguish
790 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
791 * can be turned off independently.
792 */
793/* ARGSUSED */
794void
795xfs_trans_dquot_buf(
796 xfs_trans_t *tp,
797 struct xfs_buf *bp,
798 uint type)
799{
800 struct xfs_buf_log_item *bip = bp->b_log_item;
801
802 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
803 type == XFS_BLF_PDQUOT_BUF ||
804 type == XFS_BLF_GDQUOT_BUF);
805
806 bip->__bli_format.blf_flags |= type;
807
808 switch (type) {
809 case XFS_BLF_UDQUOT_BUF:
810 type = XFS_BLFT_UDQUOT_BUF;
811 break;
812 case XFS_BLF_PDQUOT_BUF:
813 type = XFS_BLFT_PDQUOT_BUF;
814 break;
815 case XFS_BLF_GDQUOT_BUF:
816 type = XFS_BLFT_GDQUOT_BUF;
817 break;
818 default:
819 type = XFS_BLFT_UNKNOWN_BUF;
820 break;
821 }
822
823 bp->b_flags |= _XBF_DQUOTS;
824 xfs_trans_buf_set_type(tp, bp, type);
825}