mm/memunmap: don't access uninitialized memmap in memunmap_pages()
[linux-2.6-block.git] / fs / xfs / xfs_trans_buf.c
CommitLineData
0b61f8a4 1// SPDX-License-Identifier: GPL-2.0
1da177e4 2/*
7b718769
NS
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
1da177e4 5 */
1da177e4 6#include "xfs.h"
a844f451 7#include "xfs_fs.h"
70a9883c 8#include "xfs_shared.h"
a4fbe6ab 9#include "xfs_format.h"
239880ef
DC
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
1da177e4 12#include "xfs_mount.h"
239880ef 13#include "xfs_trans.h"
a844f451 14#include "xfs_buf_item.h"
1da177e4 15#include "xfs_trans_priv.h"
0b1b213f 16#include "xfs_trace.h"
1da177e4 17
4a5224d7
CH
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,
de2a4f59
DC
26 struct xfs_buf_map *map,
27 int nmaps)
4a5224d7 28{
e6631f85 29 struct xfs_log_item *lip;
e98c414f 30 struct xfs_buf_log_item *blip;
de2a4f59
DC
31 int len = 0;
32 int i;
33
34 for (i = 0; i < nmaps; i++)
35 len += map[i].bm_len;
1da177e4 36
e6631f85
DC
37 list_for_each_entry(lip, &tp->t_items, li_trans) {
38 blip = (struct xfs_buf_log_item *)lip;
e98c414f 39 if (blip->bli_item.li_type == XFS_LI_BUF &&
49074c06 40 blip->bli_buf->b_target == target &&
de2a4f59
DC
41 XFS_BUF_ADDR(blip->bli_buf) == map[0].bm_bn &&
42 blip->bli_buf->b_length == len) {
43 ASSERT(blip->bli_buf->b_map_count == nmaps);
e98c414f 44 return blip->bli_buf;
de2a4f59 45 }
4a5224d7
CH
46 }
47
48 return NULL;
49}
1da177e4 50
d7e84f41
CH
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
bf9d9013 68 ASSERT(bp->b_transp == NULL);
d7e84f41
CH
69
70 /*
fb1755a6 71 * The xfs_buf_log_item pointer is stored in b_log_item. If
d7e84f41
CH
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);
fb1755a6 76 bip = bp->b_log_item;
d7e84f41 77 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
0f22f9d0 78 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
d7e84f41
CH
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 /*
e6631f85
DC
89 * Attach the item to the transaction so we can find it in
90 * xfs_trans_get_buf() and friends.
d7e84f41 91 */
e98c414f 92 xfs_trans_add_item(tp, &bip->bli_item);
bf9d9013 93 bp->b_transp = tp;
d7e84f41
CH
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);
fb1755a6 103 trace_xfs_trans_bjoin(bp->b_log_item);
d7e84f41 104}
1da177e4
LT
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 *
1da177e4
LT
112 * If the transaction pointer is NULL, make this just a normal
113 * get_buf() call.
114 */
de2a4f59
DC
115struct xfs_buf *
116xfs_trans_get_buf_map(
117 struct xfs_trans *tp,
118 struct xfs_buftarg *target,
119 struct xfs_buf_map *map,
120 int nmaps,
121 xfs_buf_flags_t flags)
1da177e4
LT
122{
123 xfs_buf_t *bp;
70a20655 124 struct xfs_buf_log_item *bip;
1da177e4 125
de2a4f59
DC
126 if (!tp)
127 return xfs_buf_get_map(target, map, nmaps, flags);
1da177e4
LT
128
129 /*
130 * If we find the buffer in the cache with this transaction
131 * pointer in its b_fsprivate2 field, then we know we already
132 * have it locked. In this case we just increment the lock
133 * recursion count and return the buffer to the caller.
134 */
de2a4f59 135 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
1da177e4 136 if (bp != NULL) {
0c842ad4 137 ASSERT(xfs_buf_islocked(bp));
c867cb61
CH
138 if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
139 xfs_buf_stale(bp);
b0388bf1 140 bp->b_flags |= XBF_DONE;
c867cb61 141 }
0b1b213f 142
bf9d9013 143 ASSERT(bp->b_transp == tp);
fb1755a6 144 bip = bp->b_log_item;
1da177e4
LT
145 ASSERT(bip != NULL);
146 ASSERT(atomic_read(&bip->bli_refcount) > 0);
147 bip->bli_recur++;
0b1b213f 148 trace_xfs_trans_get_buf_recur(bip);
d99831ff 149 return bp;
1da177e4
LT
150 }
151
de2a4f59 152 bp = xfs_buf_get_map(target, map, nmaps, flags);
1da177e4
LT
153 if (bp == NULL) {
154 return NULL;
155 }
156
5a52c2a5 157 ASSERT(!bp->b_error);
1da177e4 158
d7e84f41 159 _xfs_trans_bjoin(tp, bp, 1);
fb1755a6 160 trace_xfs_trans_get_buf(bp->b_log_item);
d99831ff 161 return bp;
1da177e4
LT
162}
163
164/*
165 * Get and lock the superblock buffer of this file system for the
166 * given transaction.
167 *
168 * We don't need to use incore_match() here, because the superblock
169 * buffer is a private buffer which we keep a pointer to in the
170 * mount structure.
171 */
172xfs_buf_t *
70a20655
CM
173xfs_trans_getsb(
174 xfs_trans_t *tp,
8c9ce2f7 175 struct xfs_mount *mp)
1da177e4
LT
176{
177 xfs_buf_t *bp;
70a20655 178 struct xfs_buf_log_item *bip;
1da177e4
LT
179
180 /*
181 * Default to just trying to lock the superblock buffer
182 * if tp is NULL.
183 */
d99831ff 184 if (tp == NULL)
8c9ce2f7 185 return xfs_getsb(mp);
1da177e4
LT
186
187 /*
188 * If the superblock buffer already has this transaction
189 * pointer in its b_fsprivate2 field, then we know we already
190 * have it locked. In this case we just increment the lock
191 * recursion count and return the buffer to the caller.
192 */
193 bp = mp->m_sb_bp;
bf9d9013 194 if (bp->b_transp == tp) {
fb1755a6 195 bip = bp->b_log_item;
1da177e4
LT
196 ASSERT(bip != NULL);
197 ASSERT(atomic_read(&bip->bli_refcount) > 0);
198 bip->bli_recur++;
0b1b213f 199 trace_xfs_trans_getsb_recur(bip);
d99831ff 200 return bp;
1da177e4
LT
201 }
202
8c9ce2f7 203 bp = xfs_getsb(mp);
d7e84f41 204 if (bp == NULL)
1da177e4 205 return NULL;
1da177e4 206
d7e84f41 207 _xfs_trans_bjoin(tp, bp, 1);
fb1755a6 208 trace_xfs_trans_getsb(bp->b_log_item);
d99831ff 209 return bp;
1da177e4
LT
210}
211
1da177e4
LT
212/*
213 * Get and lock the buffer for the caller if it is not already
214 * locked within the given transaction. If it has not yet been
215 * read in, read it from disk. If it is already locked
216 * within the transaction and already read in, just increment its
217 * lock recursion count and return a pointer to it.
218 *
1da177e4
LT
219 * If the transaction pointer is NULL, make this just a normal
220 * read_buf() call.
221 */
222int
de2a4f59
DC
223xfs_trans_read_buf_map(
224 struct xfs_mount *mp,
225 struct xfs_trans *tp,
226 struct xfs_buftarg *target,
227 struct xfs_buf_map *map,
228 int nmaps,
229 xfs_buf_flags_t flags,
c3f8fc73 230 struct xfs_buf **bpp,
1813dd64 231 const struct xfs_buf_ops *ops)
1da177e4 232{
2d3d0c53
DC
233 struct xfs_buf *bp = NULL;
234 struct xfs_buf_log_item *bip;
1da177e4
LT
235 int error;
236
7ca790a5 237 *bpp = NULL;
1da177e4
LT
238 /*
239 * If we find the buffer in the cache with this transaction
240 * pointer in its b_fsprivate2 field, then we know we already
241 * have it locked. If it is already read in we just increment
242 * the lock recursion count and return the buffer to the caller.
243 * If the buffer is not yet read in, then we read it in, increment
244 * the lock recursion count, and return it to the caller.
245 */
2d3d0c53
DC
246 if (tp)
247 bp = xfs_trans_buf_item_match(tp, target, map, nmaps);
248 if (bp) {
0c842ad4 249 ASSERT(xfs_buf_islocked(bp));
bf9d9013 250 ASSERT(bp->b_transp == tp);
fb1755a6 251 ASSERT(bp->b_log_item != NULL);
5a52c2a5 252 ASSERT(!bp->b_error);
2d3d0c53
DC
253 ASSERT(bp->b_flags & XBF_DONE);
254
1da177e4
LT
255 /*
256 * We never locked this buf ourselves, so we shouldn't
257 * brelse it either. Just get out.
258 */
259 if (XFS_FORCED_SHUTDOWN(mp)) {
0b1b213f 260 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
2451337d 261 return -EIO;
1da177e4
LT
262 }
263
1aff5696
DW
264 /*
265 * Check if the caller is trying to read a buffer that is
266 * already attached to the transaction yet has no buffer ops
267 * assigned. Ops are usually attached when the buffer is
268 * attached to the transaction, or by the read caller if
269 * special circumstances. That didn't happen, which is not
270 * how this is supposed to go.
271 *
272 * If the buffer passes verification we'll let this go, but if
273 * not we have to shut down. Let the transaction cleanup code
274 * release this buffer when it kills the tranaction.
275 */
276 ASSERT(bp->b_ops != NULL);
75d02303 277 error = xfs_buf_reverify(bp, ops);
1aff5696
DW
278 if (error) {
279 xfs_buf_ioerror_alert(bp, __func__);
280
281 if (tp->t_flags & XFS_TRANS_DIRTY)
282 xfs_force_shutdown(tp->t_mountp,
283 SHUTDOWN_META_IO_ERROR);
284
285 /* bad CRC means corrupted metadata */
286 if (error == -EFSBADCRC)
287 error = -EFSCORRUPTED;
288 return error;
289 }
290
fb1755a6 291 bip = bp->b_log_item;
1da177e4
LT
292 bip->bli_recur++;
293
294 ASSERT(atomic_read(&bip->bli_refcount) > 0);
0b1b213f 295 trace_xfs_trans_read_buf_recur(bip);
1aff5696 296 ASSERT(bp->b_ops != NULL || ops == NULL);
1da177e4
LT
297 *bpp = bp;
298 return 0;
299 }
300
1813dd64 301 bp = xfs_buf_read_map(target, map, nmaps, flags, ops);
2d3d0c53
DC
302 if (!bp) {
303 if (!(flags & XBF_TRYLOCK))
304 return -ENOMEM;
305 return tp ? 0 : -EAGAIN;
1da177e4 306 }
2d3d0c53
DC
307
308 /*
309 * If we've had a read error, then the contents of the buffer are
310 * invalid and should not be used. To ensure that a followup read tries
311 * to pull the buffer from disk again, we clear the XBF_DONE flag and
312 * mark the buffer stale. This ensures that anyone who has a current
313 * reference to the buffer will interpret it's contents correctly and
314 * future cache lookups will also treat it as an empty, uninitialised
315 * buffer.
316 */
5a52c2a5
CS
317 if (bp->b_error) {
318 error = bp->b_error;
2d3d0c53
DC
319 if (!XFS_FORCED_SHUTDOWN(mp))
320 xfs_buf_ioerror_alert(bp, __func__);
321 bp->b_flags &= ~XBF_DONE;
c867cb61 322 xfs_buf_stale(bp);
2d3d0c53
DC
323
324 if (tp && (tp->t_flags & XFS_TRANS_DIRTY))
7d04a335 325 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
1da177e4 326 xfs_buf_relse(bp);
ac75a1f7
DC
327
328 /* bad CRC means corrupted metadata */
2451337d
DC
329 if (error == -EFSBADCRC)
330 error = -EFSCORRUPTED;
1da177e4
LT
331 return error;
332 }
2d3d0c53
DC
333
334 if (XFS_FORCED_SHUTDOWN(mp)) {
335 xfs_buf_relse(bp);
336 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
337 return -EIO;
1da177e4 338 }
1da177e4 339
e9892d3c 340 if (tp) {
2d3d0c53 341 _xfs_trans_bjoin(tp, bp, 1);
fb1755a6 342 trace_xfs_trans_read_buf(bp->b_log_item);
e9892d3c 343 }
1aff5696 344 ASSERT(bp->b_ops != NULL || ops == NULL);
1da177e4
LT
345 *bpp = bp;
346 return 0;
347
1da177e4
LT
348}
349
38b6238e
DW
350/* Has this buffer been dirtied by anyone? */
351bool
352xfs_trans_buf_is_dirty(
353 struct xfs_buf *bp)
354{
355 struct xfs_buf_log_item *bip = bp->b_log_item;
356
357 if (!bip)
358 return false;
359 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
360 return test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
361}
362
1da177e4 363/*
23420d05
BF
364 * Release a buffer previously joined to the transaction. If the buffer is
365 * modified within this transaction, decrement the recursion count but do not
366 * release the buffer even if the count goes to 0. If the buffer is not modified
367 * within the transaction, decrement the recursion count and release the buffer
368 * if the recursion count goes to 0.
1da177e4 369 *
23420d05
BF
370 * If the buffer is to be released and it was not already dirty before this
371 * transaction began, then also free the buf_log_item associated with it.
1da177e4 372 *
23420d05 373 * If the transaction pointer is NULL, this is a normal xfs_buf_relse() call.
1da177e4
LT
374 */
375void
70a20655 376xfs_trans_brelse(
23420d05
BF
377 struct xfs_trans *tp,
378 struct xfs_buf *bp)
1da177e4 379{
23420d05 380 struct xfs_buf_log_item *bip = bp->b_log_item;
1da177e4 381
23420d05
BF
382 ASSERT(bp->b_transp == tp);
383
384 if (!tp) {
1da177e4
LT
385 xfs_buf_relse(bp);
386 return;
387 }
388
23420d05 389 trace_xfs_trans_brelse(bip);
1da177e4 390 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
1da177e4
LT
391 ASSERT(atomic_read(&bip->bli_refcount) > 0);
392
1da177e4 393 /*
23420d05
BF
394 * If the release is for a recursive lookup, then decrement the count
395 * and return.
1da177e4
LT
396 */
397 if (bip->bli_recur > 0) {
398 bip->bli_recur--;
1da177e4
LT
399 return;
400 }
401
402 /*
23420d05 403 * If the buffer is invalidated or dirty in this transaction, we can't
1da177e4
LT
404 * release it until we commit.
405 */
e6631f85 406 if (test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags))
1da177e4 407 return;
0b1b213f 408 if (bip->bli_flags & XFS_BLI_STALE)
1da177e4 409 return;
1da177e4 410
1da177e4 411 /*
23420d05
BF
412 * Unlink the log item from the transaction and clear the hold flag, if
413 * set. We wouldn't want the next user of the buffer to get confused.
1da177e4 414 */
23420d05 415 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
e98c414f 416 xfs_trans_del_item(&bip->bli_item);
23420d05 417 bip->bli_flags &= ~XFS_BLI_HOLD;
1da177e4 418
95808459
BF
419 /* drop the reference to the bli */
420 xfs_buf_item_put(bip);
1da177e4 421
5b03ff1b 422 bp->b_transp = NULL;
1da177e4 423 xfs_buf_relse(bp);
1da177e4
LT
424}
425
1da177e4
LT
426/*
427 * Mark the buffer as not needing to be unlocked when the buf item's
ddf92053 428 * iop_committing() routine is called. The buffer must already be locked
1da177e4
LT
429 * and associated with the given transaction.
430 */
431/* ARGSUSED */
432void
70a20655
CM
433xfs_trans_bhold(
434 xfs_trans_t *tp,
435 xfs_buf_t *bp)
1da177e4 436{
fb1755a6 437 struct xfs_buf_log_item *bip = bp->b_log_item;
1da177e4 438
bf9d9013 439 ASSERT(bp->b_transp == tp);
adadbeef 440 ASSERT(bip != NULL);
1da177e4 441 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
0f22f9d0 442 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
1da177e4 443 ASSERT(atomic_read(&bip->bli_refcount) > 0);
adadbeef 444
1da177e4 445 bip->bli_flags |= XFS_BLI_HOLD;
0b1b213f 446 trace_xfs_trans_bhold(bip);
1da177e4
LT
447}
448
efa092f3
TS
449/*
450 * Cancel the previous buffer hold request made on this buffer
451 * for this transaction.
452 */
453void
70a20655
CM
454xfs_trans_bhold_release(
455 xfs_trans_t *tp,
456 xfs_buf_t *bp)
efa092f3 457{
fb1755a6 458 struct xfs_buf_log_item *bip = bp->b_log_item;
efa092f3 459
bf9d9013 460 ASSERT(bp->b_transp == tp);
adadbeef 461 ASSERT(bip != NULL);
efa092f3 462 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
0f22f9d0 463 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_CANCEL));
efa092f3
TS
464 ASSERT(atomic_read(&bip->bli_refcount) > 0);
465 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
0b1b213f 466
adadbeef 467 bip->bli_flags &= ~XFS_BLI_HOLD;
0b1b213f 468 trace_xfs_trans_bhold_release(bip);
efa092f3
TS
469}
470
1da177e4 471/*
9684010d 472 * Mark a buffer dirty in the transaction.
1da177e4
LT
473 */
474void
9684010d
BF
475xfs_trans_dirty_buf(
476 struct xfs_trans *tp,
477 struct xfs_buf *bp)
1da177e4 478{
fb1755a6 479 struct xfs_buf_log_item *bip = bp->b_log_item;
1da177e4 480
bf9d9013 481 ASSERT(bp->b_transp == tp);
adadbeef 482 ASSERT(bip != NULL);
cb669ca5
CH
483 ASSERT(bp->b_iodone == NULL ||
484 bp->b_iodone == xfs_buf_iodone_callbacks);
1da177e4
LT
485
486 /*
487 * Mark the buffer as needing to be written out eventually,
488 * and set its iodone function to remove the buffer's buf log
489 * item from the AIL and free it when the buffer is flushed
490 * to disk. See xfs_buf_attach_iodone() for more details
491 * on li_cb and xfs_buf_iodone_callbacks().
492 * If we end up aborting this transaction, we trap this buffer
493 * inside the b_bdstrat callback so that this won't get written to
494 * disk.
495 */
b0388bf1 496 bp->b_flags |= XBF_DONE;
1da177e4 497
1da177e4 498 ASSERT(atomic_read(&bip->bli_refcount) > 0);
cb669ca5 499 bp->b_iodone = xfs_buf_iodone_callbacks;
ca30b2a7 500 bip->bli_item.li_cb = xfs_buf_iodone;
1da177e4
LT
501
502 /*
503 * If we invalidated the buffer within this transaction, then
504 * cancel the invalidation now that we're dirtying the buffer
505 * again. There are no races with the code in xfs_buf_item_unpin(),
506 * because we have a reference to the buffer this entire time.
507 */
508 if (bip->bli_flags & XFS_BLI_STALE) {
1da177e4 509 bip->bli_flags &= ~XFS_BLI_STALE;
5cfd28b6
DC
510 ASSERT(bp->b_flags & XBF_STALE);
511 bp->b_flags &= ~XBF_STALE;
0f22f9d0 512 bip->__bli_format.blf_flags &= ~XFS_BLF_CANCEL;
1da177e4 513 }
9684010d 514 bip->bli_flags |= XFS_BLI_DIRTY | XFS_BLI_LOGGED;
1da177e4 515
1da177e4 516 tp->t_flags |= XFS_TRANS_DIRTY;
e6631f85 517 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
9684010d
BF
518}
519
520/*
521 * This is called to mark bytes first through last inclusive of the given
522 * buffer as needing to be logged when the transaction is committed.
523 * The buffer must already be associated with the given transaction.
524 *
525 * First and last are numbers relative to the beginning of this buffer,
526 * so the first byte in the buffer is numbered 0 regardless of the
527 * value of b_blkno.
528 */
529void
530xfs_trans_log_buf(
531 struct xfs_trans *tp,
532 struct xfs_buf *bp,
533 uint first,
534 uint last)
535{
fb1755a6 536 struct xfs_buf_log_item *bip = bp->b_log_item;
9684010d
BF
537
538 ASSERT(first <= last && last < BBTOB(bp->b_length));
8dc518df 539 ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED));
9684010d
BF
540
541 xfs_trans_dirty_buf(tp, bp);
5f6bed76 542
9684010d 543 trace_xfs_trans_log_buf(bip);
8dc518df 544 xfs_buf_item_log(bip, first, last);
1da177e4
LT
545}
546
547
548/*
43ff2122
CH
549 * Invalidate a buffer that is being used within a transaction.
550 *
551 * Typically this is because the blocks in the buffer are being freed, so we
552 * need to prevent it from being written out when we're done. Allowing it
553 * to be written again might overwrite data in the free blocks if they are
554 * reallocated to a file.
1da177e4 555 *
43ff2122
CH
556 * We prevent the buffer from being written out by marking it stale. We can't
557 * get rid of the buf log item at this point because the buffer may still be
558 * pinned by another transaction. If that is the case, then we'll wait until
559 * the buffer is committed to disk for the last time (we can tell by the ref
560 * count) and free it in xfs_buf_item_unpin(). Until that happens we will
561 * keep the buffer locked so that the buffer and buf log item are not reused.
562 *
563 * We also set the XFS_BLF_CANCEL flag in the buf log format structure and log
564 * the buf item. This will be used at recovery time to determine that copies
565 * of the buffer in the log before this should not be replayed.
566 *
567 * We mark the item descriptor and the transaction dirty so that we'll hold
568 * the buffer until after the commit.
569 *
570 * Since we're invalidating the buffer, we also clear the state about which
571 * parts of the buffer have been logged. We also clear the flag indicating
572 * that this is an inode buffer since the data in the buffer will no longer
573 * be valid.
574 *
575 * We set the stale bit in the buffer as well since we're getting rid of it.
1da177e4
LT
576 */
577void
578xfs_trans_binval(
70a20655
CM
579 xfs_trans_t *tp,
580 xfs_buf_t *bp)
1da177e4 581{
fb1755a6 582 struct xfs_buf_log_item *bip = bp->b_log_item;
91e4bac0 583 int i;
1da177e4 584
bf9d9013 585 ASSERT(bp->b_transp == tp);
adadbeef 586 ASSERT(bip != NULL);
1da177e4
LT
587 ASSERT(atomic_read(&bip->bli_refcount) > 0);
588
0b1b213f
CH
589 trace_xfs_trans_binval(bip);
590
1da177e4
LT
591 if (bip->bli_flags & XFS_BLI_STALE) {
592 /*
593 * If the buffer is already invalidated, then
594 * just return.
595 */
5cfd28b6 596 ASSERT(bp->b_flags & XBF_STALE);
1da177e4 597 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
0f22f9d0 598 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLF_INODE_BUF));
61fe135c 599 ASSERT(!(bip->__bli_format.blf_flags & XFS_BLFT_MASK));
0f22f9d0 600 ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL);
e6631f85 601 ASSERT(test_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags));
1da177e4 602 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
1da177e4
LT
603 return;
604 }
605
c867cb61 606 xfs_buf_stale(bp);
43ff2122 607
1da177e4 608 bip->bli_flags |= XFS_BLI_STALE;
ccf7c23f 609 bip->bli_flags &= ~(XFS_BLI_INODE_BUF | XFS_BLI_LOGGED | XFS_BLI_DIRTY);
0f22f9d0
MT
610 bip->__bli_format.blf_flags &= ~XFS_BLF_INODE_BUF;
611 bip->__bli_format.blf_flags |= XFS_BLF_CANCEL;
61fe135c 612 bip->__bli_format.blf_flags &= ~XFS_BLFT_MASK;
91e4bac0
MT
613 for (i = 0; i < bip->bli_format_count; i++) {
614 memset(bip->bli_formats[i].blf_data_map, 0,
615 (bip->bli_formats[i].blf_map_size * sizeof(uint)));
616 }
e6631f85 617 set_bit(XFS_LI_DIRTY, &bip->bli_item.li_flags);
1da177e4 618 tp->t_flags |= XFS_TRANS_DIRTY;
1da177e4
LT
619}
620
621/*
ccf7c23f
DC
622 * This call is used to indicate that the buffer contains on-disk inodes which
623 * must be handled specially during recovery. They require special handling
624 * because only the di_next_unlinked from the inodes in the buffer should be
625 * recovered. The rest of the data in the buffer is logged via the inodes
626 * themselves.
1da177e4 627 *
ccf7c23f
DC
628 * All we do is set the XFS_BLI_INODE_BUF flag in the items flags so it can be
629 * transferred to the buffer's log format structure so that we'll know what to
630 * do at recovery time.
1da177e4 631 */
1da177e4
LT
632void
633xfs_trans_inode_buf(
70a20655
CM
634 xfs_trans_t *tp,
635 xfs_buf_t *bp)
1da177e4 636{
fb1755a6 637 struct xfs_buf_log_item *bip = bp->b_log_item;
1da177e4 638
bf9d9013 639 ASSERT(bp->b_transp == tp);
adadbeef 640 ASSERT(bip != NULL);
1da177e4
LT
641 ASSERT(atomic_read(&bip->bli_refcount) > 0);
642
ccf7c23f 643 bip->bli_flags |= XFS_BLI_INODE_BUF;
61fe135c 644 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
1da177e4
LT
645}
646
647/*
648 * This call is used to indicate that the buffer is going to
649 * be staled and was an inode buffer. This means it gets
93848a99 650 * special processing during unpin - where any inodes
1da177e4
LT
651 * associated with the buffer should be removed from ail.
652 * There is also special processing during recovery,
653 * any replay of the inodes in the buffer needs to be
654 * prevented as the buffer may have been reused.
655 */
656void
657xfs_trans_stale_inode_buf(
70a20655
CM
658 xfs_trans_t *tp,
659 xfs_buf_t *bp)
1da177e4 660{
fb1755a6 661 struct xfs_buf_log_item *bip = bp->b_log_item;
1da177e4 662
bf9d9013 663 ASSERT(bp->b_transp == tp);
adadbeef 664 ASSERT(bip != NULL);
1da177e4
LT
665 ASSERT(atomic_read(&bip->bli_refcount) > 0);
666
667 bip->bli_flags |= XFS_BLI_STALE_INODE;
ca30b2a7 668 bip->bli_item.li_cb = xfs_buf_iodone;
61fe135c 669 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
1da177e4
LT
670}
671
1da177e4
LT
672/*
673 * Mark the buffer as being one which contains newly allocated
674 * inodes. We need to make sure that even if this buffer is
675 * relogged as an 'inode buf' we still recover all of the inode
676 * images in the face of a crash. This works in coordination with
677 * xfs_buf_item_committed() to ensure that the buffer remains in the
678 * AIL at its original location even after it has been relogged.
679 */
680/* ARGSUSED */
681void
682xfs_trans_inode_alloc_buf(
70a20655
CM
683 xfs_trans_t *tp,
684 xfs_buf_t *bp)
1da177e4 685{
fb1755a6 686 struct xfs_buf_log_item *bip = bp->b_log_item;
1da177e4 687
bf9d9013 688 ASSERT(bp->b_transp == tp);
adadbeef 689 ASSERT(bip != NULL);
1da177e4
LT
690 ASSERT(atomic_read(&bip->bli_refcount) > 0);
691
692 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
61fe135c 693 xfs_trans_buf_set_type(tp, bp, XFS_BLFT_DINO_BUF);
1da177e4
LT
694}
695
5f6bed76 696/*
8dc518df
BF
697 * Mark the buffer as ordered for this transaction. This means that the contents
698 * of the buffer are not recorded in the transaction but it is tracked in the
699 * AIL as though it was. This allows us to record logical changes in
700 * transactions rather than the physical changes we make to the buffer without
701 * changing writeback ordering constraints of metadata buffers.
5f6bed76 702 */
a5814bce 703bool
5f6bed76
DC
704xfs_trans_ordered_buf(
705 struct xfs_trans *tp,
706 struct xfs_buf *bp)
707{
fb1755a6 708 struct xfs_buf_log_item *bip = bp->b_log_item;
5f6bed76
DC
709
710 ASSERT(bp->b_transp == tp);
711 ASSERT(bip != NULL);
712 ASSERT(atomic_read(&bip->bli_refcount) > 0);
a5814bce
BF
713
714 if (xfs_buf_item_dirty_format(bip))
715 return false;
5f6bed76
DC
716
717 bip->bli_flags |= XFS_BLI_ORDERED;
718 trace_xfs_buf_item_ordered(bip);
8dc518df
BF
719
720 /*
721 * We don't log a dirty range of an ordered buffer but it still needs
722 * to be marked dirty and that it has been logged.
723 */
724 xfs_trans_dirty_buf(tp, bp);
a5814bce 725 return true;
5f6bed76
DC
726}
727
ee1a47ab
CH
728/*
729 * Set the type of the buffer for log recovery so that it can correctly identify
730 * and hence attach the correct buffer ops to the buffer after replay.
731 */
732void
733xfs_trans_buf_set_type(
734 struct xfs_trans *tp,
735 struct xfs_buf *bp,
61fe135c 736 enum xfs_blft type)
ee1a47ab 737{
fb1755a6 738 struct xfs_buf_log_item *bip = bp->b_log_item;
ee1a47ab 739
d75afeb3
DC
740 if (!tp)
741 return;
742
ee1a47ab
CH
743 ASSERT(bp->b_transp == tp);
744 ASSERT(bip != NULL);
745 ASSERT(atomic_read(&bip->bli_refcount) > 0);
ee1a47ab 746
61fe135c 747 xfs_blft_to_flags(&bip->__bli_format, type);
ee1a47ab 748}
1da177e4 749
d75afeb3
DC
750void
751xfs_trans_buf_copy_type(
752 struct xfs_buf *dst_bp,
753 struct xfs_buf *src_bp)
754{
fb1755a6
CM
755 struct xfs_buf_log_item *sbip = src_bp->b_log_item;
756 struct xfs_buf_log_item *dbip = dst_bp->b_log_item;
61fe135c 757 enum xfs_blft type;
d75afeb3 758
61fe135c
DC
759 type = xfs_blft_from_flags(&sbip->__bli_format);
760 xfs_blft_to_flags(&dbip->__bli_format, type);
d75afeb3
DC
761}
762
1da177e4
LT
763/*
764 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
765 * dquots. However, unlike in inode buffer recovery, dquot buffers get
766 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
767 * The only thing that makes dquot buffers different from regular
768 * buffers is that we must not replay dquot bufs when recovering
769 * if a _corresponding_ quotaoff has happened. We also have to distinguish
770 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
771 * can be turned off independently.
772 */
773/* ARGSUSED */
774void
775xfs_trans_dquot_buf(
70a20655
CM
776 xfs_trans_t *tp,
777 xfs_buf_t *bp,
778 uint type)
1da177e4 779{
fb1755a6 780 struct xfs_buf_log_item *bip = bp->b_log_item;
61fe135c 781
c1155410
DC
782 ASSERT(type == XFS_BLF_UDQUOT_BUF ||
783 type == XFS_BLF_PDQUOT_BUF ||
784 type == XFS_BLF_GDQUOT_BUF);
1da177e4 785
61fe135c
DC
786 bip->__bli_format.blf_flags |= type;
787
788 switch (type) {
789 case XFS_BLF_UDQUOT_BUF:
790 type = XFS_BLFT_UDQUOT_BUF;
791 break;
792 case XFS_BLF_PDQUOT_BUF:
793 type = XFS_BLFT_PDQUOT_BUF;
794 break;
795 case XFS_BLF_GDQUOT_BUF:
796 type = XFS_BLFT_GDQUOT_BUF;
797 break;
798 default:
799 type = XFS_BLFT_UNKNOWN_BUF;
800 break;
801 }
802
ee1a47ab 803 xfs_trans_buf_set_type(tp, bp, type);
1da177e4 804}