blob: 60b6b898022bcb1be98f96f2a21615e1bff5e131 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_types.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_sb.h"
26#include "xfs_ag.h"
Nathan Scotta844f452005-11-02 14:38:42 +110027#include "xfs_dir2.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_dmapi.h"
29#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_bmap_btree.h"
31#include "xfs_alloc_btree.h"
32#include "xfs_ialloc_btree.h"
Nathan Scotta844f452005-11-02 14:38:42 +110033#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"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include "xfs_trans_priv.h"
39#include "xfs_error.h"
40#include "xfs_rw.h"
41
42
43STATIC xfs_buf_t *xfs_trans_buf_item_match(xfs_trans_t *, xfs_buftarg_t *,
44 xfs_daddr_t, int);
45STATIC xfs_buf_t *xfs_trans_buf_item_match_all(xfs_trans_t *, xfs_buftarg_t *,
46 xfs_daddr_t, int);
47
48
49/*
50 * Get and lock the buffer for the caller if it is not already
51 * locked within the given transaction. If it is already locked
52 * within the transaction, just increment its lock recursion count
53 * and return a pointer to it.
54 *
55 * Use the fast path function xfs_trans_buf_item_match() or the buffer
56 * cache routine incore_match() to find the buffer
57 * if it is already owned by this transaction.
58 *
59 * If we don't already own the buffer, use get_buf() to get it.
60 * If it doesn't yet have an associated xfs_buf_log_item structure,
61 * then allocate one and add the item to this transaction.
62 *
63 * If the transaction pointer is NULL, make this just a normal
64 * get_buf() call.
65 */
66xfs_buf_t *
67xfs_trans_get_buf(xfs_trans_t *tp,
68 xfs_buftarg_t *target_dev,
69 xfs_daddr_t blkno,
70 int len,
71 uint flags)
72{
73 xfs_buf_t *bp;
74 xfs_buf_log_item_t *bip;
75
76 if (flags == 0)
77 flags = XFS_BUF_LOCK | XFS_BUF_MAPPED;
78
79 /*
80 * Default to a normal get_buf() call if the tp is NULL.
81 */
82 if (tp == NULL) {
83 bp = xfs_buf_get_flags(target_dev, blkno, len,
84 flags | BUF_BUSY);
85 return(bp);
86 }
87
88 /*
89 * If we find the buffer in the cache with this transaction
90 * pointer in its b_fsprivate2 field, then we know we already
91 * have it locked. In this case we just increment the lock
92 * recursion count and return the buffer to the caller.
93 */
94 if (tp->t_items.lic_next == NULL) {
95 bp = xfs_trans_buf_item_match(tp, target_dev, blkno, len);
96 } else {
97 bp = xfs_trans_buf_item_match_all(tp, target_dev, blkno, len);
98 }
99 if (bp != NULL) {
100 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
101 if (XFS_FORCED_SHUTDOWN(tp->t_mountp)) {
102 xfs_buftrace("TRANS GET RECUR SHUT", bp);
103 XFS_BUF_SUPER_STALE(bp);
104 }
105 /*
106 * If the buffer is stale then it was binval'ed
107 * since last read. This doesn't matter since the
108 * caller isn't allowed to use the data anyway.
109 */
110 else if (XFS_BUF_ISSTALE(bp)) {
111 xfs_buftrace("TRANS GET RECUR STALE", bp);
112 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
113 }
114 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
115 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
116 ASSERT(bip != NULL);
117 ASSERT(atomic_read(&bip->bli_refcount) > 0);
118 bip->bli_recur++;
119 xfs_buftrace("TRANS GET RECUR", bp);
120 xfs_buf_item_trace("GET RECUR", bip);
121 return (bp);
122 }
123
124 /*
125 * We always specify the BUF_BUSY flag within a transaction so
126 * that get_buf does not try to push out a delayed write buffer
127 * which might cause another transaction to take place (if the
128 * buffer was delayed alloc). Such recursive transactions can
129 * easily deadlock with our current transaction as well as cause
130 * us to run out of stack space.
131 */
132 bp = xfs_buf_get_flags(target_dev, blkno, len, flags | BUF_BUSY);
133 if (bp == NULL) {
134 return NULL;
135 }
136
137 ASSERT(!XFS_BUF_GETERROR(bp));
138
139 /*
140 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
141 * it doesn't have one yet, then allocate one and initialize it.
142 * The checks to see if one is there are in xfs_buf_item_init().
143 */
144 xfs_buf_item_init(bp, tp->t_mountp);
145
146 /*
147 * Set the recursion count for the buffer within this transaction
148 * to 0.
149 */
150 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
151 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
152 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
153 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
154 bip->bli_recur = 0;
155
156 /*
157 * Take a reference for this transaction on the buf item.
158 */
159 atomic_inc(&bip->bli_refcount);
160
161 /*
162 * Get a log_item_desc to point at the new item.
163 */
164 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
165
166 /*
167 * Initialize b_fsprivate2 so we can find it with incore_match()
168 * above.
169 */
170 XFS_BUF_SET_FSPRIVATE2(bp, tp);
171
172 xfs_buftrace("TRANS GET", bp);
173 xfs_buf_item_trace("GET", bip);
174 return (bp);
175}
176
177/*
178 * Get and lock the superblock buffer of this file system for the
179 * given transaction.
180 *
181 * We don't need to use incore_match() here, because the superblock
182 * buffer is a private buffer which we keep a pointer to in the
183 * mount structure.
184 */
185xfs_buf_t *
186xfs_trans_getsb(xfs_trans_t *tp,
187 struct xfs_mount *mp,
188 int flags)
189{
190 xfs_buf_t *bp;
191 xfs_buf_log_item_t *bip;
192
193 /*
194 * Default to just trying to lock the superblock buffer
195 * if tp is NULL.
196 */
197 if (tp == NULL) {
198 return (xfs_getsb(mp, flags));
199 }
200
201 /*
202 * If the superblock buffer already has this transaction
203 * pointer in its b_fsprivate2 field, then we know we already
204 * have it locked. In this case we just increment the lock
205 * recursion count and return the buffer to the caller.
206 */
207 bp = mp->m_sb_bp;
208 if (XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp) {
209 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
210 ASSERT(bip != NULL);
211 ASSERT(atomic_read(&bip->bli_refcount) > 0);
212 bip->bli_recur++;
213 xfs_buf_item_trace("GETSB RECUR", bip);
214 return (bp);
215 }
216
217 bp = xfs_getsb(mp, flags);
218 if (bp == NULL) {
219 return NULL;
220 }
221
222 /*
223 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
224 * it doesn't have one yet, then allocate one and initialize it.
225 * The checks to see if one is there are in xfs_buf_item_init().
226 */
227 xfs_buf_item_init(bp, mp);
228
229 /*
230 * Set the recursion count for the buffer within this transaction
231 * to 0.
232 */
233 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
234 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
235 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
236 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
237 bip->bli_recur = 0;
238
239 /*
240 * Take a reference for this transaction on the buf item.
241 */
242 atomic_inc(&bip->bli_refcount);
243
244 /*
245 * Get a log_item_desc to point at the new item.
246 */
247 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
248
249 /*
250 * Initialize b_fsprivate2 so we can find it with incore_match()
251 * above.
252 */
253 XFS_BUF_SET_FSPRIVATE2(bp, tp);
254
255 xfs_buf_item_trace("GETSB", bip);
256 return (bp);
257}
258
259#ifdef DEBUG
260xfs_buftarg_t *xfs_error_target;
261int xfs_do_error;
262int xfs_req_num;
263int xfs_error_mod = 33;
264#endif
265
266/*
267 * Get and lock the buffer for the caller if it is not already
268 * locked within the given transaction. If it has not yet been
269 * read in, read it from disk. If it is already locked
270 * within the transaction and already read in, just increment its
271 * lock recursion count and return a pointer to it.
272 *
273 * Use the fast path function xfs_trans_buf_item_match() or the buffer
274 * cache routine incore_match() to find the buffer
275 * if it is already owned by this transaction.
276 *
277 * If we don't already own the buffer, use read_buf() to get it.
278 * If it doesn't yet have an associated xfs_buf_log_item structure,
279 * then allocate one and add the item to this transaction.
280 *
281 * If the transaction pointer is NULL, make this just a normal
282 * read_buf() call.
283 */
284int
285xfs_trans_read_buf(
286 xfs_mount_t *mp,
287 xfs_trans_t *tp,
288 xfs_buftarg_t *target,
289 xfs_daddr_t blkno,
290 int len,
291 uint flags,
292 xfs_buf_t **bpp)
293{
294 xfs_buf_t *bp;
295 xfs_buf_log_item_t *bip;
296 int error;
297
298 if (flags == 0)
299 flags = XFS_BUF_LOCK | XFS_BUF_MAPPED;
300
301 /*
302 * Default to a normal get_buf() call if the tp is NULL.
303 */
304 if (tp == NULL) {
305 bp = xfs_buf_read_flags(target, blkno, len, flags | BUF_BUSY);
306 if (!bp)
307 return XFS_ERROR(ENOMEM);
308
309 if ((bp != NULL) && (XFS_BUF_GETERROR(bp) != 0)) {
310 xfs_ioerror_alert("xfs_trans_read_buf", mp,
311 bp, blkno);
312 error = XFS_BUF_GETERROR(bp);
313 xfs_buf_relse(bp);
314 return error;
315 }
316#ifdef DEBUG
317 if (xfs_do_error && (bp != NULL)) {
318 if (xfs_error_target == target) {
319 if (((xfs_req_num++) % xfs_error_mod) == 0) {
320 xfs_buf_relse(bp);
Nathan Scottb6574522006-06-09 15:29:40 +1000321 cmn_err(CE_DEBUG, "Returning error!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 return XFS_ERROR(EIO);
323 }
324 }
325 }
326#endif
327 if (XFS_FORCED_SHUTDOWN(mp))
328 goto shutdown_abort;
329 *bpp = bp;
330 return 0;
331 }
332
333 /*
334 * If we find the buffer in the cache with this transaction
335 * pointer in its b_fsprivate2 field, then we know we already
336 * have it locked. If it is already read in we just increment
337 * the lock recursion count and return the buffer to the caller.
338 * If the buffer is not yet read in, then we read it in, increment
339 * the lock recursion count, and return it to the caller.
340 */
341 if (tp->t_items.lic_next == NULL) {
342 bp = xfs_trans_buf_item_match(tp, target, blkno, len);
343 } else {
344 bp = xfs_trans_buf_item_match_all(tp, target, blkno, len);
345 }
346 if (bp != NULL) {
347 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
348 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
349 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
350 ASSERT((XFS_BUF_ISERROR(bp)) == 0);
351 if (!(XFS_BUF_ISDONE(bp))) {
352 xfs_buftrace("READ_BUF_INCORE !DONE", bp);
353 ASSERT(!XFS_BUF_ISASYNC(bp));
354 XFS_BUF_READ(bp);
355 xfsbdstrat(tp->t_mountp, bp);
356 xfs_iowait(bp);
357 if (XFS_BUF_GETERROR(bp) != 0) {
358 xfs_ioerror_alert("xfs_trans_read_buf", mp,
359 bp, blkno);
360 error = XFS_BUF_GETERROR(bp);
361 xfs_buf_relse(bp);
362 /*
363 * We can gracefully recover from most
364 * read errors. Ones we can't are those
365 * that happen after the transaction's
366 * already dirty.
367 */
368 if (tp->t_flags & XFS_TRANS_DIRTY)
369 xfs_force_shutdown(tp->t_mountp,
Nathan Scott7d04a332006-06-09 14:58:38 +1000370 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 return error;
372 }
373 }
374 /*
375 * We never locked this buf ourselves, so we shouldn't
376 * brelse it either. Just get out.
377 */
378 if (XFS_FORCED_SHUTDOWN(mp)) {
379 xfs_buftrace("READ_BUF_INCORE XFSSHUTDN", bp);
380 *bpp = NULL;
381 return XFS_ERROR(EIO);
382 }
383
384
385 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
386 bip->bli_recur++;
387
388 ASSERT(atomic_read(&bip->bli_refcount) > 0);
389 xfs_buf_item_trace("READ RECUR", bip);
390 *bpp = bp;
391 return 0;
392 }
393
394 /*
395 * We always specify the BUF_BUSY flag within a transaction so
396 * that get_buf does not try to push out a delayed write buffer
397 * which might cause another transaction to take place (if the
398 * buffer was delayed alloc). Such recursive transactions can
399 * easily deadlock with our current transaction as well as cause
400 * us to run out of stack space.
401 */
402 bp = xfs_buf_read_flags(target, blkno, len, flags | BUF_BUSY);
403 if (bp == NULL) {
404 *bpp = NULL;
405 return 0;
406 }
407 if (XFS_BUF_GETERROR(bp) != 0) {
408 XFS_BUF_SUPER_STALE(bp);
409 xfs_buftrace("READ ERROR", bp);
410 error = XFS_BUF_GETERROR(bp);
411
412 xfs_ioerror_alert("xfs_trans_read_buf", mp,
413 bp, blkno);
414 if (tp->t_flags & XFS_TRANS_DIRTY)
Nathan Scott7d04a332006-06-09 14:58:38 +1000415 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 xfs_buf_relse(bp);
417 return error;
418 }
419#ifdef DEBUG
420 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
421 if (xfs_error_target == target) {
422 if (((xfs_req_num++) % xfs_error_mod) == 0) {
423 xfs_force_shutdown(tp->t_mountp,
Nathan Scott7d04a332006-06-09 14:58:38 +1000424 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 xfs_buf_relse(bp);
Nathan Scottb6574522006-06-09 15:29:40 +1000426 cmn_err(CE_DEBUG, "Returning trans error!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 return XFS_ERROR(EIO);
428 }
429 }
430 }
431#endif
432 if (XFS_FORCED_SHUTDOWN(mp))
433 goto shutdown_abort;
434
435 /*
436 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
437 * it doesn't have one yet, then allocate one and initialize it.
438 * The checks to see if one is there are in xfs_buf_item_init().
439 */
440 xfs_buf_item_init(bp, tp->t_mountp);
441
442 /*
443 * Set the recursion count for the buffer within this transaction
444 * to 0.
445 */
446 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
447 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
448 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
449 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
450 bip->bli_recur = 0;
451
452 /*
453 * Take a reference for this transaction on the buf item.
454 */
455 atomic_inc(&bip->bli_refcount);
456
457 /*
458 * Get a log_item_desc to point at the new item.
459 */
460 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
461
462 /*
463 * Initialize b_fsprivate2 so we can find it with incore_match()
464 * above.
465 */
466 XFS_BUF_SET_FSPRIVATE2(bp, tp);
467
468 xfs_buftrace("TRANS READ", bp);
469 xfs_buf_item_trace("READ", bip);
470 *bpp = bp;
471 return 0;
472
473shutdown_abort:
474 /*
475 * the theory here is that buffer is good but we're
476 * bailing out because the filesystem is being forcibly
477 * shut down. So we should leave the b_flags alone since
478 * the buffer's not staled and just get out.
479 */
480#if defined(DEBUG)
481 if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp))
482 cmn_err(CE_NOTE, "about to pop assert, bp == 0x%p", bp);
483#endif
484 ASSERT((XFS_BUF_BFLAGS(bp) & (XFS_B_STALE|XFS_B_DELWRI)) !=
485 (XFS_B_STALE|XFS_B_DELWRI));
486
487 xfs_buftrace("READ_BUF XFSSHUTDN", bp);
488 xfs_buf_relse(bp);
489 *bpp = NULL;
490 return XFS_ERROR(EIO);
491}
492
493
494/*
495 * Release the buffer bp which was previously acquired with one of the
496 * xfs_trans_... buffer allocation routines if the buffer has not
497 * been modified within this transaction. If the buffer is modified
498 * within this transaction, do decrement the recursion count but do
499 * not release the buffer even if the count goes to 0. If the buffer is not
500 * modified within the transaction, decrement the recursion count and
501 * release the buffer if the recursion count goes to 0.
502 *
503 * If the buffer is to be released and it was not modified before
504 * this transaction began, then free the buf_log_item associated with it.
505 *
506 * If the transaction pointer is NULL, make this just a normal
507 * brelse() call.
508 */
509void
510xfs_trans_brelse(xfs_trans_t *tp,
511 xfs_buf_t *bp)
512{
513 xfs_buf_log_item_t *bip;
514 xfs_log_item_t *lip;
515 xfs_log_item_desc_t *lidp;
516
517 /*
518 * Default to a normal brelse() call if the tp is NULL.
519 */
520 if (tp == NULL) {
521 ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL);
522 /*
523 * If there's a buf log item attached to the buffer,
524 * then let the AIL know that the buffer is being
525 * unlocked.
526 */
527 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
528 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
529 if (lip->li_type == XFS_LI_BUF) {
530 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
531 xfs_trans_unlocked_item(
532 bip->bli_item.li_mountp,
533 lip);
534 }
535 }
536 xfs_buf_relse(bp);
537 return;
538 }
539
540 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
541 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
542 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
543 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
544 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
545 ASSERT(atomic_read(&bip->bli_refcount) > 0);
546
547 /*
548 * Find the item descriptor pointing to this buffer's
549 * log item. It must be there.
550 */
551 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
552 ASSERT(lidp != NULL);
553
554 /*
555 * If the release is just for a recursive lock,
556 * then decrement the count and return.
557 */
558 if (bip->bli_recur > 0) {
559 bip->bli_recur--;
560 xfs_buf_item_trace("RELSE RECUR", bip);
561 return;
562 }
563
564 /*
565 * If the buffer is dirty within this transaction, we can't
566 * release it until we commit.
567 */
568 if (lidp->lid_flags & XFS_LID_DIRTY) {
569 xfs_buf_item_trace("RELSE DIRTY", bip);
570 return;
571 }
572
573 /*
574 * If the buffer has been invalidated, then we can't release
575 * it until the transaction commits to disk unless it is re-dirtied
576 * as part of this transaction. This prevents us from pulling
577 * the item from the AIL before we should.
578 */
579 if (bip->bli_flags & XFS_BLI_STALE) {
580 xfs_buf_item_trace("RELSE STALE", bip);
581 return;
582 }
583
584 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
585 xfs_buf_item_trace("RELSE", bip);
586
587 /*
588 * Free up the log item descriptor tracking the released item.
589 */
590 xfs_trans_free_item(tp, lidp);
591
592 /*
593 * Clear the hold flag in the buf log item if it is set.
594 * We wouldn't want the next user of the buffer to
595 * get confused.
596 */
597 if (bip->bli_flags & XFS_BLI_HOLD) {
598 bip->bli_flags &= ~XFS_BLI_HOLD;
599 }
600
601 /*
602 * Drop our reference to the buf log item.
603 */
604 atomic_dec(&bip->bli_refcount);
605
606 /*
607 * If the buf item is not tracking data in the log, then
608 * we must free it before releasing the buffer back to the
609 * free pool. Before releasing the buffer to the free pool,
610 * clear the transaction pointer in b_fsprivate2 to dissolve
611 * its relation to this transaction.
612 */
613 if (!xfs_buf_item_dirty(bip)) {
614/***
615 ASSERT(bp->b_pincount == 0);
616***/
617 ASSERT(atomic_read(&bip->bli_refcount) == 0);
618 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
619 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
620 xfs_buf_item_relse(bp);
621 bip = NULL;
622 }
623 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
624
625 /*
626 * If we've still got a buf log item on the buffer, then
627 * tell the AIL that the buffer is being unlocked.
628 */
629 if (bip != NULL) {
630 xfs_trans_unlocked_item(bip->bli_item.li_mountp,
631 (xfs_log_item_t*)bip);
632 }
633
634 xfs_buf_relse(bp);
635 return;
636}
637
638/*
639 * Add the locked buffer to the transaction.
640 * The buffer must be locked, and it cannot be associated with any
641 * transaction.
642 *
643 * If the buffer does not yet have a buf log item associated with it,
644 * then allocate one for it. Then add the buf item to the transaction.
645 */
646void
647xfs_trans_bjoin(xfs_trans_t *tp,
648 xfs_buf_t *bp)
649{
650 xfs_buf_log_item_t *bip;
651
652 ASSERT(XFS_BUF_ISBUSY(bp));
653 ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL);
654
655 /*
656 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
657 * it doesn't have one yet, then allocate one and initialize it.
658 * The checks to see if one is there are in xfs_buf_item_init().
659 */
660 xfs_buf_item_init(bp, tp->t_mountp);
661 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
662 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
663 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
664 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
665
666 /*
667 * Take a reference for this transaction on the buf item.
668 */
669 atomic_inc(&bip->bli_refcount);
670
671 /*
672 * Get a log_item_desc to point at the new item.
673 */
674 (void) xfs_trans_add_item(tp, (xfs_log_item_t *)bip);
675
676 /*
677 * Initialize b_fsprivate2 so we can find it with incore_match()
678 * in xfs_trans_get_buf() and friends above.
679 */
680 XFS_BUF_SET_FSPRIVATE2(bp, tp);
681
682 xfs_buf_item_trace("BJOIN", bip);
683}
684
685/*
686 * Mark the buffer as not needing to be unlocked when the buf item's
687 * IOP_UNLOCK() routine is called. The buffer must already be locked
688 * and associated with the given transaction.
689 */
690/* ARGSUSED */
691void
692xfs_trans_bhold(xfs_trans_t *tp,
693 xfs_buf_t *bp)
694{
695 xfs_buf_log_item_t *bip;
696
697 ASSERT(XFS_BUF_ISBUSY(bp));
698 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
699 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
700
701 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
702 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
703 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
704 ASSERT(atomic_read(&bip->bli_refcount) > 0);
705 bip->bli_flags |= XFS_BLI_HOLD;
706 xfs_buf_item_trace("BHOLD", bip);
707}
708
709/*
Tim Shimminefa092f2005-09-05 08:29:01 +1000710 * Cancel the previous buffer hold request made on this buffer
711 * for this transaction.
712 */
713void
714xfs_trans_bhold_release(xfs_trans_t *tp,
715 xfs_buf_t *bp)
716{
717 xfs_buf_log_item_t *bip;
718
719 ASSERT(XFS_BUF_ISBUSY(bp));
720 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
721 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
722
723 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
724 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
725 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
726 ASSERT(atomic_read(&bip->bli_refcount) > 0);
727 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
728 bip->bli_flags &= ~XFS_BLI_HOLD;
729 xfs_buf_item_trace("BHOLD RELEASE", bip);
730}
731
732/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 * This is called to mark bytes first through last inclusive of the given
734 * buffer as needing to be logged when the transaction is committed.
735 * The buffer must already be associated with the given transaction.
736 *
737 * First and last are numbers relative to the beginning of this buffer,
738 * so the first byte in the buffer is numbered 0 regardless of the
739 * value of b_blkno.
740 */
741void
742xfs_trans_log_buf(xfs_trans_t *tp,
743 xfs_buf_t *bp,
744 uint first,
745 uint last)
746{
747 xfs_buf_log_item_t *bip;
748 xfs_log_item_desc_t *lidp;
749
750 ASSERT(XFS_BUF_ISBUSY(bp));
751 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
752 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
753 ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp)));
754 ASSERT((XFS_BUF_IODONE_FUNC(bp) == NULL) ||
755 (XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks));
756
757 /*
758 * Mark the buffer as needing to be written out eventually,
759 * and set its iodone function to remove the buffer's buf log
760 * item from the AIL and free it when the buffer is flushed
761 * to disk. See xfs_buf_attach_iodone() for more details
762 * on li_cb and xfs_buf_iodone_callbacks().
763 * If we end up aborting this transaction, we trap this buffer
764 * inside the b_bdstrat callback so that this won't get written to
765 * disk.
766 */
767 XFS_BUF_DELAYWRITE(bp);
768 XFS_BUF_DONE(bp);
769
770 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
771 ASSERT(atomic_read(&bip->bli_refcount) > 0);
772 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
773 bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*))xfs_buf_iodone;
774
775 /*
776 * If we invalidated the buffer within this transaction, then
777 * cancel the invalidation now that we're dirtying the buffer
778 * again. There are no races with the code in xfs_buf_item_unpin(),
779 * because we have a reference to the buffer this entire time.
780 */
781 if (bip->bli_flags & XFS_BLI_STALE) {
782 xfs_buf_item_trace("BLOG UNSTALE", bip);
783 bip->bli_flags &= ~XFS_BLI_STALE;
784 ASSERT(XFS_BUF_ISSTALE(bp));
785 XFS_BUF_UNSTALE(bp);
786 bip->bli_format.blf_flags &= ~XFS_BLI_CANCEL;
787 }
788
789 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
790 ASSERT(lidp != NULL);
791
792 tp->t_flags |= XFS_TRANS_DIRTY;
793 lidp->lid_flags |= XFS_LID_DIRTY;
794 lidp->lid_flags &= ~XFS_LID_BUF_STALE;
795 bip->bli_flags |= XFS_BLI_LOGGED;
796 xfs_buf_item_log(bip, first, last);
797 xfs_buf_item_trace("BLOG", bip);
798}
799
800
801/*
802 * This called to invalidate a buffer that is being used within
803 * a transaction. Typically this is because the blocks in the
804 * buffer are being freed, so we need to prevent it from being
805 * written out when we're done. Allowing it to be written again
806 * might overwrite data in the free blocks if they are reallocated
807 * to a file.
808 *
809 * We prevent the buffer from being written out by clearing the
810 * B_DELWRI flag. We can't always
811 * get rid of the buf log item at this point, though, because
812 * the buffer may still be pinned by another transaction. If that
813 * is the case, then we'll wait until the buffer is committed to
814 * disk for the last time (we can tell by the ref count) and
815 * free it in xfs_buf_item_unpin(). Until it is cleaned up we
816 * will keep the buffer locked so that the buffer and buf log item
817 * are not reused.
818 */
819void
820xfs_trans_binval(
821 xfs_trans_t *tp,
822 xfs_buf_t *bp)
823{
824 xfs_log_item_desc_t *lidp;
825 xfs_buf_log_item_t *bip;
826
827 ASSERT(XFS_BUF_ISBUSY(bp));
828 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
829 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
830
831 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
832 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
833 ASSERT(lidp != NULL);
834 ASSERT(atomic_read(&bip->bli_refcount) > 0);
835
836 if (bip->bli_flags & XFS_BLI_STALE) {
837 /*
838 * If the buffer is already invalidated, then
839 * just return.
840 */
841 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
842 ASSERT(XFS_BUF_ISSTALE(bp));
843 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
844 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_INODE_BUF));
845 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
846 ASSERT(lidp->lid_flags & XFS_LID_DIRTY);
847 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
848 xfs_buftrace("XFS_BINVAL RECUR", bp);
849 xfs_buf_item_trace("BINVAL RECUR", bip);
850 return;
851 }
852
853 /*
854 * Clear the dirty bit in the buffer and set the STALE flag
855 * in the buf log item. The STALE flag will be used in
856 * xfs_buf_item_unpin() to determine if it should clean up
857 * when the last reference to the buf item is given up.
858 * We set the XFS_BLI_CANCEL flag in the buf log format structure
859 * and log the buf item. This will be used at recovery time
860 * to determine that copies of the buffer in the log before
861 * this should not be replayed.
862 * We mark the item descriptor and the transaction dirty so
863 * that we'll hold the buffer until after the commit.
864 *
865 * Since we're invalidating the buffer, we also clear the state
866 * about which parts of the buffer have been logged. We also
867 * clear the flag indicating that this is an inode buffer since
868 * the data in the buffer will no longer be valid.
869 *
870 * We set the stale bit in the buffer as well since we're getting
871 * rid of it.
872 */
873 XFS_BUF_UNDELAYWRITE(bp);
874 XFS_BUF_STALE(bp);
875 bip->bli_flags |= XFS_BLI_STALE;
876 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_DIRTY);
877 bip->bli_format.blf_flags &= ~XFS_BLI_INODE_BUF;
878 bip->bli_format.blf_flags |= XFS_BLI_CANCEL;
879 memset((char *)(bip->bli_format.blf_data_map), 0,
880 (bip->bli_format.blf_map_size * sizeof(uint)));
881 lidp->lid_flags |= XFS_LID_DIRTY|XFS_LID_BUF_STALE;
882 tp->t_flags |= XFS_TRANS_DIRTY;
883 xfs_buftrace("XFS_BINVAL", bp);
884 xfs_buf_item_trace("BINVAL", bip);
885}
886
887/*
888 * This call is used to indicate that the buffer contains on-disk
889 * inodes which must be handled specially during recovery. They
890 * require special handling because only the di_next_unlinked from
891 * the inodes in the buffer should be recovered. The rest of the
892 * data in the buffer is logged via the inodes themselves.
893 *
894 * All we do is set the XFS_BLI_INODE_BUF flag in the buffer's log
895 * format structure so that we'll know what to do at recovery time.
896 */
897/* ARGSUSED */
898void
899xfs_trans_inode_buf(
900 xfs_trans_t *tp,
901 xfs_buf_t *bp)
902{
903 xfs_buf_log_item_t *bip;
904
905 ASSERT(XFS_BUF_ISBUSY(bp));
906 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
907 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
908
909 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
910 ASSERT(atomic_read(&bip->bli_refcount) > 0);
911
912 bip->bli_format.blf_flags |= XFS_BLI_INODE_BUF;
913}
914
915/*
916 * This call is used to indicate that the buffer is going to
917 * be staled and was an inode buffer. This means it gets
918 * special processing during unpin - where any inodes
919 * associated with the buffer should be removed from ail.
920 * There is also special processing during recovery,
921 * any replay of the inodes in the buffer needs to be
922 * prevented as the buffer may have been reused.
923 */
924void
925xfs_trans_stale_inode_buf(
926 xfs_trans_t *tp,
927 xfs_buf_t *bp)
928{
929 xfs_buf_log_item_t *bip;
930
931 ASSERT(XFS_BUF_ISBUSY(bp));
932 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
933 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
934
935 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
936 ASSERT(atomic_read(&bip->bli_refcount) > 0);
937
938 bip->bli_flags |= XFS_BLI_STALE_INODE;
939 bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*))
940 xfs_buf_iodone;
941}
942
943
944
945/*
946 * Mark the buffer as being one which contains newly allocated
947 * inodes. We need to make sure that even if this buffer is
948 * relogged as an 'inode buf' we still recover all of the inode
949 * images in the face of a crash. This works in coordination with
950 * xfs_buf_item_committed() to ensure that the buffer remains in the
951 * AIL at its original location even after it has been relogged.
952 */
953/* ARGSUSED */
954void
955xfs_trans_inode_alloc_buf(
956 xfs_trans_t *tp,
957 xfs_buf_t *bp)
958{
959 xfs_buf_log_item_t *bip;
960
961 ASSERT(XFS_BUF_ISBUSY(bp));
962 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
963 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
964
965 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
966 ASSERT(atomic_read(&bip->bli_refcount) > 0);
967
968 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
969}
970
971
972/*
973 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
974 * dquots. However, unlike in inode buffer recovery, dquot buffers get
975 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
976 * The only thing that makes dquot buffers different from regular
977 * buffers is that we must not replay dquot bufs when recovering
978 * if a _corresponding_ quotaoff has happened. We also have to distinguish
979 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
980 * can be turned off independently.
981 */
982/* ARGSUSED */
983void
984xfs_trans_dquot_buf(
985 xfs_trans_t *tp,
986 xfs_buf_t *bp,
987 uint type)
988{
989 xfs_buf_log_item_t *bip;
990
991 ASSERT(XFS_BUF_ISBUSY(bp));
992 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
993 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
994 ASSERT(type == XFS_BLI_UDQUOT_BUF ||
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000995 type == XFS_BLI_PDQUOT_BUF ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 type == XFS_BLI_GDQUOT_BUF);
997
998 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
999 ASSERT(atomic_read(&bip->bli_refcount) > 0);
1000
1001 bip->bli_format.blf_flags |= type;
1002}
1003
1004/*
1005 * Check to see if a buffer matching the given parameters is already
1006 * a part of the given transaction. Only check the first, embedded
1007 * chunk, since we don't want to spend all day scanning large transactions.
1008 */
1009STATIC xfs_buf_t *
1010xfs_trans_buf_item_match(
1011 xfs_trans_t *tp,
1012 xfs_buftarg_t *target,
1013 xfs_daddr_t blkno,
1014 int len)
1015{
1016 xfs_log_item_chunk_t *licp;
1017 xfs_log_item_desc_t *lidp;
1018 xfs_buf_log_item_t *blip;
1019 xfs_buf_t *bp;
1020 int i;
1021
1022 bp = NULL;
1023 len = BBTOB(len);
1024 licp = &tp->t_items;
1025 if (!XFS_LIC_ARE_ALL_FREE(licp)) {
1026 for (i = 0; i < licp->lic_unused; i++) {
1027 /*
1028 * Skip unoccupied slots.
1029 */
1030 if (XFS_LIC_ISFREE(licp, i)) {
1031 continue;
1032 }
1033
1034 lidp = XFS_LIC_SLOT(licp, i);
1035 blip = (xfs_buf_log_item_t *)lidp->lid_item;
1036 if (blip->bli_item.li_type != XFS_LI_BUF) {
1037 continue;
1038 }
1039
1040 bp = blip->bli_buf;
1041 if ((XFS_BUF_TARGET(bp) == target) &&
1042 (XFS_BUF_ADDR(bp) == blkno) &&
1043 (XFS_BUF_COUNT(bp) == len)) {
1044 /*
1045 * We found it. Break out and
1046 * return the pointer to the buffer.
1047 */
1048 break;
1049 } else {
1050 bp = NULL;
1051 }
1052 }
1053 }
1054 return bp;
1055}
1056
1057/*
1058 * Check to see if a buffer matching the given parameters is already
1059 * a part of the given transaction. Check all the chunks, we
1060 * want to be thorough.
1061 */
1062STATIC xfs_buf_t *
1063xfs_trans_buf_item_match_all(
1064 xfs_trans_t *tp,
1065 xfs_buftarg_t *target,
1066 xfs_daddr_t blkno,
1067 int len)
1068{
1069 xfs_log_item_chunk_t *licp;
1070 xfs_log_item_desc_t *lidp;
1071 xfs_buf_log_item_t *blip;
1072 xfs_buf_t *bp;
1073 int i;
1074
1075 bp = NULL;
1076 len = BBTOB(len);
1077 for (licp = &tp->t_items; licp != NULL; licp = licp->lic_next) {
1078 if (XFS_LIC_ARE_ALL_FREE(licp)) {
1079 ASSERT(licp == &tp->t_items);
1080 ASSERT(licp->lic_next == NULL);
1081 return NULL;
1082 }
1083 for (i = 0; i < licp->lic_unused; i++) {
1084 /*
1085 * Skip unoccupied slots.
1086 */
1087 if (XFS_LIC_ISFREE(licp, i)) {
1088 continue;
1089 }
1090
1091 lidp = XFS_LIC_SLOT(licp, i);
1092 blip = (xfs_buf_log_item_t *)lidp->lid_item;
1093 if (blip->bli_item.li_type != XFS_LI_BUF) {
1094 continue;
1095 }
1096
1097 bp = blip->bli_buf;
1098 if ((XFS_BUF_TARGET(bp) == target) &&
1099 (XFS_BUF_ADDR(bp) == blkno) &&
1100 (XFS_BUF_COUNT(bp) == len)) {
1101 /*
1102 * We found it. Break out and
1103 * return the pointer to the buffer.
1104 */
1105 return bp;
1106 }
1107 }
1108 }
1109 return NULL;
1110}