]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - fs/xfs/xfs_trans_buf.c
xfs: stop passing opaque handles to xfs_log.c routines
[mirror_ubuntu-bionic-kernel.git] / fs / xfs / xfs_trans_buf.c
CommitLineData
1da177e4 1/*
7b718769
NS
2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
1da177e4 4 *
7b718769
NS
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
1da177e4
LT
7 * published by the Free Software Foundation.
8 *
7b718769
NS
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.
1da177e4 13 *
7b718769
NS
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
1da177e4 17 */
1da177e4 18#include "xfs.h"
a844f451 19#include "xfs_fs.h"
1da177e4 20#include "xfs_types.h"
a844f451 21#include "xfs_bit.h"
1da177e4 22#include "xfs_log.h"
a844f451 23#include "xfs_inum.h"
1da177e4 24#include "xfs_trans.h"
1da177e4
LT
25#include "xfs_sb.h"
26#include "xfs_ag.h"
a844f451 27#include "xfs_dir2.h"
1da177e4
LT
28#include "xfs_dmapi.h"
29#include "xfs_mount.h"
a844f451
NS
30#include "xfs_bmap_btree.h"
31#include "xfs_alloc_btree.h"
32#include "xfs_ialloc_btree.h"
a844f451
NS
33#include "xfs_dir2_sf.h"
34#include "xfs_attr_sf.h"
35#include "xfs_dinode.h"
36#include "xfs_inode.h"
37#include "xfs_buf_item.h"
1da177e4
LT
38#include "xfs_trans_priv.h"
39#include "xfs_error.h"
40#include "xfs_rw.h"
0b1b213f 41#include "xfs_trace.h"
1da177e4
LT
42
43
44STATIC xfs_buf_t *xfs_trans_buf_item_match(xfs_trans_t *, xfs_buftarg_t *,
45 xfs_daddr_t, int);
46STATIC xfs_buf_t *xfs_trans_buf_item_match_all(xfs_trans_t *, xfs_buftarg_t *,
47 xfs_daddr_t, int);
48
49
50/*
51 * Get and lock the buffer for the caller if it is not already
52 * locked within the given transaction. If it is already locked
53 * within the transaction, just increment its lock recursion count
54 * and return a pointer to it.
55 *
56 * Use the fast path function xfs_trans_buf_item_match() or the buffer
57 * cache routine incore_match() to find the buffer
58 * if it is already owned by this transaction.
59 *
60 * If we don't already own the buffer, use get_buf() to get it.
61 * If it doesn't yet have an associated xfs_buf_log_item structure,
62 * then allocate one and add the item to this transaction.
63 *
64 * If the transaction pointer is NULL, make this just a normal
65 * get_buf() call.
66 */
67xfs_buf_t *
68xfs_trans_get_buf(xfs_trans_t *tp,
69 xfs_buftarg_t *target_dev,
70 xfs_daddr_t blkno,
71 int len,
72 uint flags)
73{
74 xfs_buf_t *bp;
75 xfs_buf_log_item_t *bip;
76
77 if (flags == 0)
0cadda1c 78 flags = XBF_LOCK | XBF_MAPPED;
1da177e4
LT
79
80 /*
81 * Default to a normal get_buf() call if the tp is NULL.
82 */
6ad112bf 83 if (tp == NULL)
0cadda1c
CH
84 return xfs_buf_get(target_dev, blkno, len,
85 flags | XBF_DONT_BLOCK);
1da177e4
LT
86
87 /*
88 * If we find the buffer in the cache with this transaction
89 * pointer in its b_fsprivate2 field, then we know we already
90 * have it locked. In this case we just increment the lock
91 * recursion count and return the buffer to the caller.
92 */
93 if (tp->t_items.lic_next == NULL) {
94 bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
95 } else {
96 bp = xfs_trans_buf_item_match_all(tp, target_dev, blkno, len);
97 }
98 if (bp != NULL) {
99 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
0b1b213f 100 if (XFS_FORCED_SHUTDOWN(tp->t_mountp))
1da177e4 101 XFS_BUF_SUPER_STALE(bp);
0b1b213f 102
1da177e4
LT
103 /*
104 * If the buffer is stale then it was binval'ed
105 * since last read. This doesn't matter since the
106 * caller isn't allowed to use the data anyway.
107 */
0b1b213f 108 else if (XFS_BUF_ISSTALE(bp))
1da177e4 109 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
0b1b213f 110
1da177e4
LT
111 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
112 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
113 ASSERT(bip != NULL);
114 ASSERT(atomic_read(&bip->bli_refcount) > 0);
115 bip->bli_recur++;
0b1b213f 116 trace_xfs_trans_get_buf_recur(bip);
1da177e4
LT
117 return (bp);
118 }
119
120 /*
0cadda1c
CH
121 * We always specify the XBF_DONT_BLOCK flag within a transaction
122 * so that get_buf does not try to push out a delayed write buffer
1da177e4
LT
123 * which might cause another transaction to take place (if the
124 * buffer was delayed alloc). Such recursive transactions can
125 * easily deadlock with our current transaction as well as cause
126 * us to run out of stack space.
127 */
0cadda1c 128 bp = xfs_buf_get(target_dev, blkno, len, flags | XBF_DONT_BLOCK);
1da177e4
LT
129 if (bp == NULL) {
130 return NULL;
131 }
132
133 ASSERT(!XFS_BUF_GETERROR(bp));
134
135 /*
136 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
137 * it doesn't have one yet, then allocate one and initialize it.
138 * The checks to see if one is there are in xfs_buf_item_init().
139 */
140 xfs_buf_item_init(bp, tp->t_mountp);
141
142 /*
143 * Set the recursion count for the buffer within this transaction
144 * to 0.
145 */
146 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
147 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
148 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
149 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
150 bip->bli_recur = 0;
151
152 /*
153 * Take a reference for this transaction on the buf item.
154 */
155 atomic_inc(&bip->bli_refcount);
156
157 /*
158 * Get a log_item_desc to point at the new item.
159 */
160 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
161
162 /*
163 * Initialize b_fsprivate2 so we can find it with incore_match()
164 * above.
165 */
166 XFS_BUF_SET_FSPRIVATE2(bp, tp);
167
0b1b213f 168 trace_xfs_trans_get_buf(bip);
1da177e4
LT
169 return (bp);
170}
171
172/*
173 * Get and lock the superblock buffer of this file system for the
174 * given transaction.
175 *
176 * We don't need to use incore_match() here, because the superblock
177 * buffer is a private buffer which we keep a pointer to in the
178 * mount structure.
179 */
180xfs_buf_t *
181xfs_trans_getsb(xfs_trans_t *tp,
182 struct xfs_mount *mp,
183 int flags)
184{
185 xfs_buf_t *bp;
186 xfs_buf_log_item_t *bip;
187
188 /*
189 * Default to just trying to lock the superblock buffer
190 * if tp is NULL.
191 */
192 if (tp == NULL) {
193 return (xfs_getsb(mp, flags));
194 }
195
196 /*
197 * If the superblock buffer already has this transaction
198 * pointer in its b_fsprivate2 field, then we know we already
199 * have it locked. In this case we just increment the lock
200 * recursion count and return the buffer to the caller.
201 */
202 bp = mp->m_sb_bp;
203 if (XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp) {
204 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
205 ASSERT(bip != NULL);
206 ASSERT(atomic_read(&bip->bli_refcount) > 0);
207 bip->bli_recur++;
0b1b213f 208 trace_xfs_trans_getsb_recur(bip);
1da177e4
LT
209 return (bp);
210 }
211
212 bp = xfs_getsb(mp, flags);
213 if (bp == NULL) {
214 return NULL;
215 }
216
217 /*
218 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
219 * it doesn't have one yet, then allocate one and initialize it.
220 * The checks to see if one is there are in xfs_buf_item_init().
221 */
222 xfs_buf_item_init(bp, mp);
223
224 /*
225 * Set the recursion count for the buffer within this transaction
226 * to 0.
227 */
228 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
229 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
230 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
231 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
232 bip->bli_recur = 0;
233
234 /*
235 * Take a reference for this transaction on the buf item.
236 */
237 atomic_inc(&bip->bli_refcount);
238
239 /*
240 * Get a log_item_desc to point at the new item.
241 */
242 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
243
244 /*
245 * Initialize b_fsprivate2 so we can find it with incore_match()
246 * above.
247 */
248 XFS_BUF_SET_FSPRIVATE2(bp, tp);
249
0b1b213f 250 trace_xfs_trans_getsb(bip);
1da177e4
LT
251 return (bp);
252}
253
254#ifdef DEBUG
255xfs_buftarg_t *xfs_error_target;
256int xfs_do_error;
257int xfs_req_num;
258int xfs_error_mod = 33;
259#endif
260
261/*
262 * Get and lock the buffer for the caller if it is not already
263 * locked within the given transaction. If it has not yet been
264 * read in, read it from disk. If it is already locked
265 * within the transaction and already read in, just increment its
266 * lock recursion count and return a pointer to it.
267 *
268 * Use the fast path function xfs_trans_buf_item_match() or the buffer
269 * cache routine incore_match() to find the buffer
270 * if it is already owned by this transaction.
271 *
272 * If we don't already own the buffer, use read_buf() to get it.
273 * If it doesn't yet have an associated xfs_buf_log_item structure,
274 * then allocate one and add the item to this transaction.
275 *
276 * If the transaction pointer is NULL, make this just a normal
277 * read_buf() call.
278 */
279int
280xfs_trans_read_buf(
281 xfs_mount_t *mp,
282 xfs_trans_t *tp,
283 xfs_buftarg_t *target,
284 xfs_daddr_t blkno,
285 int len,
286 uint flags,
287 xfs_buf_t **bpp)
288{
289 xfs_buf_t *bp;
290 xfs_buf_log_item_t *bip;
291 int error;
292
293 if (flags == 0)
0cadda1c 294 flags = XBF_LOCK | XBF_MAPPED;
1da177e4
LT
295
296 /*
297 * Default to a normal get_buf() call if the tp is NULL.
298 */
299 if (tp == NULL) {
0cadda1c 300 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
1da177e4 301 if (!bp)
0cadda1c 302 return (flags & XBF_TRYLOCK) ?
a3f74ffb 303 EAGAIN : XFS_ERROR(ENOMEM);
1da177e4 304
a0f7bfd3 305 if (XFS_BUF_GETERROR(bp) != 0) {
1da177e4
LT
306 xfs_ioerror_alert("xfs_trans_read_buf", mp,
307 bp, blkno);
308 error = XFS_BUF_GETERROR(bp);
309 xfs_buf_relse(bp);
310 return error;
311 }
312#ifdef DEBUG
a0f7bfd3 313 if (xfs_do_error) {
1da177e4
LT
314 if (xfs_error_target == target) {
315 if (((xfs_req_num++) % xfs_error_mod) == 0) {
316 xfs_buf_relse(bp);
b6574520 317 cmn_err(CE_DEBUG, "Returning error!\n");
1da177e4
LT
318 return XFS_ERROR(EIO);
319 }
320 }
321 }
322#endif
323 if (XFS_FORCED_SHUTDOWN(mp))
324 goto shutdown_abort;
325 *bpp = bp;
326 return 0;
327 }
328
329 /*
330 * If we find the buffer in the cache with this transaction
331 * pointer in its b_fsprivate2 field, then we know we already
332 * have it locked. If it is already read in we just increment
333 * the lock recursion count and return the buffer to the caller.
334 * If the buffer is not yet read in, then we read it in, increment
335 * the lock recursion count, and return it to the caller.
336 */
337 if (tp->t_items.lic_next == NULL) {
338 bp = xfs_trans_buf_item_match(tp, target, blkno, len);
339 } else {
340 bp = xfs_trans_buf_item_match_all(tp, target, blkno, len);
341 }
342 if (bp != NULL) {
343 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
344 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
345 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
346 ASSERT((XFS_BUF_ISERROR(bp)) == 0);
347 if (!(XFS_BUF_ISDONE(bp))) {
0b1b213f 348 trace_xfs_trans_read_buf_io(bp, _RET_IP_);
1da177e4
LT
349 ASSERT(!XFS_BUF_ISASYNC(bp));
350 XFS_BUF_READ(bp);
351 xfsbdstrat(tp->t_mountp, bp);
d64e31a2
DC
352 error = xfs_iowait(bp);
353 if (error) {
1da177e4
LT
354 xfs_ioerror_alert("xfs_trans_read_buf", mp,
355 bp, blkno);
1da177e4
LT
356 xfs_buf_relse(bp);
357 /*
d64e31a2
DC
358 * We can gracefully recover from most read
359 * errors. Ones we can't are those that happen
360 * after the transaction's already dirty.
1da177e4
LT
361 */
362 if (tp->t_flags & XFS_TRANS_DIRTY)
363 xfs_force_shutdown(tp->t_mountp,
7d04a335 364 SHUTDOWN_META_IO_ERROR);
1da177e4
LT
365 return error;
366 }
367 }
368 /*
369 * We never locked this buf ourselves, so we shouldn't
370 * brelse it either. Just get out.
371 */
372 if (XFS_FORCED_SHUTDOWN(mp)) {
0b1b213f 373 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
1da177e4
LT
374 *bpp = NULL;
375 return XFS_ERROR(EIO);
376 }
377
378
379 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
380 bip->bli_recur++;
381
382 ASSERT(atomic_read(&bip->bli_refcount) > 0);
0b1b213f 383 trace_xfs_trans_read_buf_recur(bip);
1da177e4
LT
384 *bpp = bp;
385 return 0;
386 }
387
388 /*
0cadda1c
CH
389 * We always specify the XBF_DONT_BLOCK flag within a transaction
390 * so that get_buf does not try to push out a delayed write buffer
1da177e4
LT
391 * which might cause another transaction to take place (if the
392 * buffer was delayed alloc). Such recursive transactions can
393 * easily deadlock with our current transaction as well as cause
394 * us to run out of stack space.
395 */
0cadda1c 396 bp = xfs_buf_read(target, blkno, len, flags | XBF_DONT_BLOCK);
1da177e4
LT
397 if (bp == NULL) {
398 *bpp = NULL;
399 return 0;
400 }
401 if (XFS_BUF_GETERROR(bp) != 0) {
402 XFS_BUF_SUPER_STALE(bp);
1da177e4
LT
403 error = XFS_BUF_GETERROR(bp);
404
405 xfs_ioerror_alert("xfs_trans_read_buf", mp,
406 bp, blkno);
407 if (tp->t_flags & XFS_TRANS_DIRTY)
7d04a335 408 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
1da177e4
LT
409 xfs_buf_relse(bp);
410 return error;
411 }
412#ifdef DEBUG
413 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
414 if (xfs_error_target == target) {
415 if (((xfs_req_num++) % xfs_error_mod) == 0) {
416 xfs_force_shutdown(tp->t_mountp,
7d04a335 417 SHUTDOWN_META_IO_ERROR);
1da177e4 418 xfs_buf_relse(bp);
b6574520 419 cmn_err(CE_DEBUG, "Returning trans error!\n");
1da177e4
LT
420 return XFS_ERROR(EIO);
421 }
422 }
423 }
424#endif
425 if (XFS_FORCED_SHUTDOWN(mp))
426 goto shutdown_abort;
427
428 /*
429 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
430 * it doesn't have one yet, then allocate one and initialize it.
431 * The checks to see if one is there are in xfs_buf_item_init().
432 */
433 xfs_buf_item_init(bp, tp->t_mountp);
434
435 /*
436 * Set the recursion count for the buffer within this transaction
437 * to 0.
438 */
439 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
440 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
441 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
442 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
443 bip->bli_recur = 0;
444
445 /*
446 * Take a reference for this transaction on the buf item.
447 */
448 atomic_inc(&bip->bli_refcount);
449
450 /*
451 * Get a log_item_desc to point at the new item.
452 */
453 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
454
455 /*
456 * Initialize b_fsprivate2 so we can find it with incore_match()
457 * above.
458 */
459 XFS_BUF_SET_FSPRIVATE2(bp, tp);
460
0b1b213f 461 trace_xfs_trans_read_buf(bip);
1da177e4
LT
462 *bpp = bp;
463 return 0;
464
465shutdown_abort:
466 /*
467 * the theory here is that buffer is good but we're
468 * bailing out because the filesystem is being forcibly
469 * shut down. So we should leave the b_flags alone since
470 * the buffer's not staled and just get out.
471 */
472#if defined(DEBUG)
473 if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp))
474 cmn_err(CE_NOTE, "about to pop assert, bp == 0x%p", bp);
475#endif
0cadda1c
CH
476 ASSERT((XFS_BUF_BFLAGS(bp) & (XBF_STALE|XBF_DELWRI)) !=
477 (XBF_STALE|XBF_DELWRI));
1da177e4 478
0b1b213f 479 trace_xfs_trans_read_buf_shut(bp, _RET_IP_);
1da177e4
LT
480 xfs_buf_relse(bp);
481 *bpp = NULL;
482 return XFS_ERROR(EIO);
483}
484
485
486/*
487 * Release the buffer bp which was previously acquired with one of the
488 * xfs_trans_... buffer allocation routines if the buffer has not
489 * been modified within this transaction. If the buffer is modified
490 * within this transaction, do decrement the recursion count but do
491 * not release the buffer even if the count goes to 0. If the buffer is not
492 * modified within the transaction, decrement the recursion count and
493 * release the buffer if the recursion count goes to 0.
494 *
495 * If the buffer is to be released and it was not modified before
496 * this transaction began, then free the buf_log_item associated with it.
497 *
498 * If the transaction pointer is NULL, make this just a normal
499 * brelse() call.
500 */
501void
502xfs_trans_brelse(xfs_trans_t *tp,
503 xfs_buf_t *bp)
504{
505 xfs_buf_log_item_t *bip;
506 xfs_log_item_t *lip;
507 xfs_log_item_desc_t *lidp;
508
509 /*
510 * Default to a normal brelse() call if the tp is NULL.
511 */
512 if (tp == NULL) {
513 ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL);
514 /*
515 * If there's a buf log item attached to the buffer,
516 * then let the AIL know that the buffer is being
517 * unlocked.
518 */
519 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
520 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
521 if (lip->li_type == XFS_LI_BUF) {
522 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
783a2f65
DC
523 xfs_trans_unlocked_item(bip->bli_item.li_ailp,
524 lip);
1da177e4
LT
525 }
526 }
527 xfs_buf_relse(bp);
528 return;
529 }
530
531 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
532 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
533 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
534 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
535 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
536 ASSERT(atomic_read(&bip->bli_refcount) > 0);
537
538 /*
539 * Find the item descriptor pointing to this buffer's
540 * log item. It must be there.
541 */
542 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
543 ASSERT(lidp != NULL);
544
0b1b213f
CH
545 trace_xfs_trans_brelse(bip);
546
1da177e4
LT
547 /*
548 * If the release is just for a recursive lock,
549 * then decrement the count and return.
550 */
551 if (bip->bli_recur > 0) {
552 bip->bli_recur--;
1da177e4
LT
553 return;
554 }
555
556 /*
557 * If the buffer is dirty within this transaction, we can't
558 * release it until we commit.
559 */
0b1b213f 560 if (lidp->lid_flags & XFS_LID_DIRTY)
1da177e4 561 return;
1da177e4
LT
562
563 /*
564 * If the buffer has been invalidated, then we can't release
565 * it until the transaction commits to disk unless it is re-dirtied
566 * as part of this transaction. This prevents us from pulling
567 * the item from the AIL before we should.
568 */
0b1b213f 569 if (bip->bli_flags & XFS_BLI_STALE)
1da177e4 570 return;
1da177e4
LT
571
572 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
1da177e4
LT
573
574 /*
575 * Free up the log item descriptor tracking the released item.
576 */
577 xfs_trans_free_item(tp, lidp);
578
579 /*
580 * Clear the hold flag in the buf log item if it is set.
581 * We wouldn't want the next user of the buffer to
582 * get confused.
583 */
584 if (bip->bli_flags & XFS_BLI_HOLD) {
585 bip->bli_flags &= ~XFS_BLI_HOLD;
586 }
587
588 /*
589 * Drop our reference to the buf log item.
590 */
591 atomic_dec(&bip->bli_refcount);
592
593 /*
594 * If the buf item is not tracking data in the log, then
595 * we must free it before releasing the buffer back to the
596 * free pool. Before releasing the buffer to the free pool,
597 * clear the transaction pointer in b_fsprivate2 to dissolve
598 * its relation to this transaction.
599 */
600 if (!xfs_buf_item_dirty(bip)) {
601/***
602 ASSERT(bp->b_pincount == 0);
603***/
604 ASSERT(atomic_read(&bip->bli_refcount) == 0);
605 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
606 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
607 xfs_buf_item_relse(bp);
608 bip = NULL;
609 }
610 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
611
612 /*
613 * If we've still got a buf log item on the buffer, then
614 * tell the AIL that the buffer is being unlocked.
615 */
616 if (bip != NULL) {
783a2f65 617 xfs_trans_unlocked_item(bip->bli_item.li_ailp,
1da177e4
LT
618 (xfs_log_item_t*)bip);
619 }
620
621 xfs_buf_relse(bp);
622 return;
623}
624
625/*
626 * Add the locked buffer to the transaction.
627 * The buffer must be locked, and it cannot be associated with any
628 * transaction.
629 *
630 * If the buffer does not yet have a buf log item associated with it,
631 * then allocate one for it. Then add the buf item to the transaction.
632 */
633void
634xfs_trans_bjoin(xfs_trans_t *tp,
635 xfs_buf_t *bp)
636{
637 xfs_buf_log_item_t *bip;
638
639 ASSERT(XFS_BUF_ISBUSY(bp));
640 ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL);
641
642 /*
643 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
644 * it doesn't have one yet, then allocate one and initialize it.
645 * The checks to see if one is there are in xfs_buf_item_init().
646 */
647 xfs_buf_item_init(bp, tp->t_mountp);
648 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
649 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
650 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
651 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
652
653 /*
654 * Take a reference for this transaction on the buf item.
655 */
656 atomic_inc(&bip->bli_refcount);
657
658 /*
659 * Get a log_item_desc to point at the new item.
660 */
661 (void) xfs_trans_add_item(tp, (xfs_log_item_t *)bip);
662
663 /*
664 * Initialize b_fsprivate2 so we can find it with incore_match()
665 * in xfs_trans_get_buf() and friends above.
666 */
667 XFS_BUF_SET_FSPRIVATE2(bp, tp);
668
0b1b213f 669 trace_xfs_trans_bjoin(bip);
1da177e4
LT
670}
671
672/*
673 * Mark the buffer as not needing to be unlocked when the buf item's
674 * IOP_UNLOCK() routine is called. The buffer must already be locked
675 * and associated with the given transaction.
676 */
677/* ARGSUSED */
678void
679xfs_trans_bhold(xfs_trans_t *tp,
680 xfs_buf_t *bp)
681{
682 xfs_buf_log_item_t *bip;
683
684 ASSERT(XFS_BUF_ISBUSY(bp));
685 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
686 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
687
688 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
689 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
690 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
691 ASSERT(atomic_read(&bip->bli_refcount) > 0);
692 bip->bli_flags |= XFS_BLI_HOLD;
0b1b213f 693 trace_xfs_trans_bhold(bip);
1da177e4
LT
694}
695
efa092f3
TS
696/*
697 * Cancel the previous buffer hold request made on this buffer
698 * for this transaction.
699 */
700void
701xfs_trans_bhold_release(xfs_trans_t *tp,
702 xfs_buf_t *bp)
703{
704 xfs_buf_log_item_t *bip;
705
706 ASSERT(XFS_BUF_ISBUSY(bp));
707 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
708 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
709
710 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
711 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
712 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
713 ASSERT(atomic_read(&bip->bli_refcount) > 0);
714 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
715 bip->bli_flags &= ~XFS_BLI_HOLD;
0b1b213f
CH
716
717 trace_xfs_trans_bhold_release(bip);
efa092f3
TS
718}
719
1da177e4
LT
720/*
721 * This is called to mark bytes first through last inclusive of the given
722 * buffer as needing to be logged when the transaction is committed.
723 * The buffer must already be associated with the given transaction.
724 *
725 * First and last are numbers relative to the beginning of this buffer,
726 * so the first byte in the buffer is numbered 0 regardless of the
727 * value of b_blkno.
728 */
729void
730xfs_trans_log_buf(xfs_trans_t *tp,
731 xfs_buf_t *bp,
732 uint first,
733 uint last)
734{
735 xfs_buf_log_item_t *bip;
736 xfs_log_item_desc_t *lidp;
737
738 ASSERT(XFS_BUF_ISBUSY(bp));
739 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
740 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
741 ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp)));
742 ASSERT((XFS_BUF_IODONE_FUNC(bp) == NULL) ||
743 (XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks));
744
745 /*
746 * Mark the buffer as needing to be written out eventually,
747 * and set its iodone function to remove the buffer's buf log
748 * item from the AIL and free it when the buffer is flushed
749 * to disk. See xfs_buf_attach_iodone() for more details
750 * on li_cb and xfs_buf_iodone_callbacks().
751 * If we end up aborting this transaction, we trap this buffer
752 * inside the b_bdstrat callback so that this won't get written to
753 * disk.
754 */
755 XFS_BUF_DELAYWRITE(bp);
756 XFS_BUF_DONE(bp);
757
758 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
759 ASSERT(atomic_read(&bip->bli_refcount) > 0);
760 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
761 bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*))xfs_buf_iodone;
762
0b1b213f
CH
763 trace_xfs_trans_log_buf(bip);
764
1da177e4
LT
765 /*
766 * If we invalidated the buffer within this transaction, then
767 * cancel the invalidation now that we're dirtying the buffer
768 * again. There are no races with the code in xfs_buf_item_unpin(),
769 * because we have a reference to the buffer this entire time.
770 */
771 if (bip->bli_flags & XFS_BLI_STALE) {
1da177e4
LT
772 bip->bli_flags &= ~XFS_BLI_STALE;
773 ASSERT(XFS_BUF_ISSTALE(bp));
774 XFS_BUF_UNSTALE(bp);
775 bip->bli_format.blf_flags &= ~XFS_BLI_CANCEL;
776 }
777
778 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
779 ASSERT(lidp != NULL);
780
781 tp->t_flags |= XFS_TRANS_DIRTY;
782 lidp->lid_flags |= XFS_LID_DIRTY;
783 lidp->lid_flags &= ~XFS_LID_BUF_STALE;
784 bip->bli_flags |= XFS_BLI_LOGGED;
785 xfs_buf_item_log(bip, first, last);
1da177e4
LT
786}
787
788
789/*
790 * This called to invalidate a buffer that is being used within
791 * a transaction. Typically this is because the blocks in the
792 * buffer are being freed, so we need to prevent it from being
793 * written out when we're done. Allowing it to be written again
794 * might overwrite data in the free blocks if they are reallocated
795 * to a file.
796 *
797 * We prevent the buffer from being written out by clearing the
798 * B_DELWRI flag. We can't always
799 * get rid of the buf log item at this point, though, because
800 * the buffer may still be pinned by another transaction. If that
801 * is the case, then we'll wait until the buffer is committed to
802 * disk for the last time (we can tell by the ref count) and
803 * free it in xfs_buf_item_unpin(). Until it is cleaned up we
804 * will keep the buffer locked so that the buffer and buf log item
805 * are not reused.
806 */
807void
808xfs_trans_binval(
809 xfs_trans_t *tp,
810 xfs_buf_t *bp)
811{
812 xfs_log_item_desc_t *lidp;
813 xfs_buf_log_item_t *bip;
814
815 ASSERT(XFS_BUF_ISBUSY(bp));
816 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
817 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
818
819 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
820 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
821 ASSERT(lidp != NULL);
822 ASSERT(atomic_read(&bip->bli_refcount) > 0);
823
0b1b213f
CH
824 trace_xfs_trans_binval(bip);
825
1da177e4
LT
826 if (bip->bli_flags & XFS_BLI_STALE) {
827 /*
828 * If the buffer is already invalidated, then
829 * just return.
830 */
831 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
832 ASSERT(XFS_BUF_ISSTALE(bp));
833 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
834 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_INODE_BUF));
835 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
836 ASSERT(lidp->lid_flags & XFS_LID_DIRTY);
837 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
1da177e4
LT
838 return;
839 }
840
841 /*
842 * Clear the dirty bit in the buffer and set the STALE flag
843 * in the buf log item. The STALE flag will be used in
844 * xfs_buf_item_unpin() to determine if it should clean up
845 * when the last reference to the buf item is given up.
846 * We set the XFS_BLI_CANCEL flag in the buf log format structure
847 * and log the buf item. This will be used at recovery time
848 * to determine that copies of the buffer in the log before
849 * this should not be replayed.
850 * We mark the item descriptor and the transaction dirty so
851 * that we'll hold the buffer until after the commit.
852 *
853 * Since we're invalidating the buffer, we also clear the state
854 * about which parts of the buffer have been logged. We also
855 * clear the flag indicating that this is an inode buffer since
856 * the data in the buffer will no longer be valid.
857 *
858 * We set the stale bit in the buffer as well since we're getting
859 * rid of it.
860 */
861 XFS_BUF_UNDELAYWRITE(bp);
862 XFS_BUF_STALE(bp);
863 bip->bli_flags |= XFS_BLI_STALE;
864 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_DIRTY);
865 bip->bli_format.blf_flags &= ~XFS_BLI_INODE_BUF;
866 bip->bli_format.blf_flags |= XFS_BLI_CANCEL;
867 memset((char *)(bip->bli_format.blf_data_map), 0,
868 (bip->bli_format.blf_map_size * sizeof(uint)));
869 lidp->lid_flags |= XFS_LID_DIRTY|XFS_LID_BUF_STALE;
870 tp->t_flags |= XFS_TRANS_DIRTY;
1da177e4
LT
871}
872
873/*
874 * This call is used to indicate that the buffer contains on-disk
875 * inodes which must be handled specially during recovery. They
876 * require special handling because only the di_next_unlinked from
877 * the inodes in the buffer should be recovered. The rest of the
878 * data in the buffer is logged via the inodes themselves.
879 *
880 * All we do is set the XFS_BLI_INODE_BUF flag in the buffer's log
881 * format structure so that we'll know what to do at recovery time.
882 */
883/* ARGSUSED */
884void
885xfs_trans_inode_buf(
886 xfs_trans_t *tp,
887 xfs_buf_t *bp)
888{
889 xfs_buf_log_item_t *bip;
890
891 ASSERT(XFS_BUF_ISBUSY(bp));
892 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
893 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
894
895 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
896 ASSERT(atomic_read(&bip->bli_refcount) > 0);
897
898 bip->bli_format.blf_flags |= XFS_BLI_INODE_BUF;
899}
900
901/*
902 * This call is used to indicate that the buffer is going to
903 * be staled and was an inode buffer. This means it gets
904 * special processing during unpin - where any inodes
905 * associated with the buffer should be removed from ail.
906 * There is also special processing during recovery,
907 * any replay of the inodes in the buffer needs to be
908 * prevented as the buffer may have been reused.
909 */
910void
911xfs_trans_stale_inode_buf(
912 xfs_trans_t *tp,
913 xfs_buf_t *bp)
914{
915 xfs_buf_log_item_t *bip;
916
917 ASSERT(XFS_BUF_ISBUSY(bp));
918 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
919 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
920
921 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
922 ASSERT(atomic_read(&bip->bli_refcount) > 0);
923
924 bip->bli_flags |= XFS_BLI_STALE_INODE;
925 bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*))
926 xfs_buf_iodone;
927}
928
929
930
931/*
932 * Mark the buffer as being one which contains newly allocated
933 * inodes. We need to make sure that even if this buffer is
934 * relogged as an 'inode buf' we still recover all of the inode
935 * images in the face of a crash. This works in coordination with
936 * xfs_buf_item_committed() to ensure that the buffer remains in the
937 * AIL at its original location even after it has been relogged.
938 */
939/* ARGSUSED */
940void
941xfs_trans_inode_alloc_buf(
942 xfs_trans_t *tp,
943 xfs_buf_t *bp)
944{
945 xfs_buf_log_item_t *bip;
946
947 ASSERT(XFS_BUF_ISBUSY(bp));
948 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
949 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
950
951 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
952 ASSERT(atomic_read(&bip->bli_refcount) > 0);
953
954 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
955}
956
957
958/*
959 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
960 * dquots. However, unlike in inode buffer recovery, dquot buffers get
961 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
962 * The only thing that makes dquot buffers different from regular
963 * buffers is that we must not replay dquot bufs when recovering
964 * if a _corresponding_ quotaoff has happened. We also have to distinguish
965 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
966 * can be turned off independently.
967 */
968/* ARGSUSED */
969void
970xfs_trans_dquot_buf(
971 xfs_trans_t *tp,
972 xfs_buf_t *bp,
973 uint type)
974{
975 xfs_buf_log_item_t *bip;
976
977 ASSERT(XFS_BUF_ISBUSY(bp));
978 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
979 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
980 ASSERT(type == XFS_BLI_UDQUOT_BUF ||
c8ad20ff 981 type == XFS_BLI_PDQUOT_BUF ||
1da177e4
LT
982 type == XFS_BLI_GDQUOT_BUF);
983
984 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
985 ASSERT(atomic_read(&bip->bli_refcount) > 0);
986
987 bip->bli_format.blf_flags |= type;
988}
989
990/*
991 * Check to see if a buffer matching the given parameters is already
992 * a part of the given transaction. Only check the first, embedded
993 * chunk, since we don't want to spend all day scanning large transactions.
994 */
995STATIC xfs_buf_t *
996xfs_trans_buf_item_match(
997 xfs_trans_t *tp,
998 xfs_buftarg_t *target,
999 xfs_daddr_t blkno,
1000 int len)
1001{
1002 xfs_log_item_chunk_t *licp;
1003 xfs_log_item_desc_t *lidp;
1004 xfs_buf_log_item_t *blip;
1005 xfs_buf_t *bp;
1006 int i;
1007
1008 bp = NULL;
1009 len = BBTOB(len);
1010 licp = &tp->t_items;
39dab9d7 1011 if (!xfs_lic_are_all_free(licp)) {
1da177e4
LT
1012 for (i = 0; i < licp->lic_unused; i++) {
1013 /*
1014 * Skip unoccupied slots.
1015 */
39dab9d7 1016 if (xfs_lic_isfree(licp, i)) {
1da177e4
LT
1017 continue;
1018 }
1019
39dab9d7 1020 lidp = xfs_lic_slot(licp, i);
1da177e4
LT
1021 blip = (xfs_buf_log_item_t *)lidp->lid_item;
1022 if (blip->bli_item.li_type != XFS_LI_BUF) {
1023 continue;
1024 }
1025
1026 bp = blip->bli_buf;
1027 if ((XFS_BUF_TARGET(bp) == target) &&
1028 (XFS_BUF_ADDR(bp) == blkno) &&
1029 (XFS_BUF_COUNT(bp) == len)) {
1030 /*
1031 * We found it. Break out and
1032 * return the pointer to the buffer.
1033 */
1034 break;
1035 } else {
1036 bp = NULL;
1037 }
1038 }
1039 }
1040 return bp;
1041}
1042
1043/*
1044 * Check to see if a buffer matching the given parameters is already
1045 * a part of the given transaction. Check all the chunks, we
1046 * want to be thorough.
1047 */
1048STATIC xfs_buf_t *
1049xfs_trans_buf_item_match_all(
1050 xfs_trans_t *tp,
1051 xfs_buftarg_t *target,
1052 xfs_daddr_t blkno,
1053 int len)
1054{
1055 xfs_log_item_chunk_t *licp;
1056 xfs_log_item_desc_t *lidp;
1057 xfs_buf_log_item_t *blip;
1058 xfs_buf_t *bp;
1059 int i;
1060
1061 bp = NULL;
1062 len = BBTOB(len);
1063 for (licp = &tp->t_items; licp != NULL; licp = licp->lic_next) {
39dab9d7 1064 if (xfs_lic_are_all_free(licp)) {
1da177e4
LT
1065 ASSERT(licp == &tp->t_items);
1066 ASSERT(licp->lic_next == NULL);
1067 return NULL;
1068 }
1069 for (i = 0; i < licp->lic_unused; i++) {
1070 /*
1071 * Skip unoccupied slots.
1072 */
39dab9d7 1073 if (xfs_lic_isfree(licp, i)) {
1da177e4
LT
1074 continue;
1075 }
1076
39dab9d7 1077 lidp = xfs_lic_slot(licp, i);
1da177e4
LT
1078 blip = (xfs_buf_log_item_t *)lidp->lid_item;
1079 if (blip->bli_item.li_type != XFS_LI_BUF) {
1080 continue;
1081 }
1082
1083 bp = blip->bli_buf;
1084 if ((XFS_BUF_TARGET(bp) == target) &&
1085 (XFS_BUF_ADDR(bp) == blkno) &&
1086 (XFS_BUF_COUNT(bp) == len)) {
1087 /*
1088 * We found it. Break out and
1089 * return the pointer to the buffer.
1090 */
1091 return bp;
1092 }
1093 }
1094 }
1095 return NULL;
1096}