blob: cb0c5839154bf8017c86d52251993c3d1ac75f18 [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)
David Chinnera3f74ff2008-03-06 13:43:42 +1100307 return (flags & XFS_BUF_TRYLOCK) ?
308 EAGAIN : XFS_ERROR(ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309
310 if ((bp != NULL) && (XFS_BUF_GETERROR(bp) != 0)) {
311 xfs_ioerror_alert("xfs_trans_read_buf", mp,
312 bp, blkno);
313 error = XFS_BUF_GETERROR(bp);
314 xfs_buf_relse(bp);
315 return error;
316 }
317#ifdef DEBUG
318 if (xfs_do_error && (bp != NULL)) {
319 if (xfs_error_target == target) {
320 if (((xfs_req_num++) % xfs_error_mod) == 0) {
321 xfs_buf_relse(bp);
Nathan Scottb6574522006-06-09 15:29:40 +1000322 cmn_err(CE_DEBUG, "Returning error!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 return XFS_ERROR(EIO);
324 }
325 }
326 }
327#endif
328 if (XFS_FORCED_SHUTDOWN(mp))
329 goto shutdown_abort;
330 *bpp = bp;
331 return 0;
332 }
333
334 /*
335 * If we find the buffer in the cache with this transaction
336 * pointer in its b_fsprivate2 field, then we know we already
337 * have it locked. If it is already read in we just increment
338 * the lock recursion count and return the buffer to the caller.
339 * If the buffer is not yet read in, then we read it in, increment
340 * the lock recursion count, and return it to the caller.
341 */
342 if (tp->t_items.lic_next == NULL) {
343 bp = xfs_trans_buf_item_match(tp, target, blkno, len);
344 } else {
345 bp = xfs_trans_buf_item_match_all(tp, target, blkno, len);
346 }
347 if (bp != NULL) {
348 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
349 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
350 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
351 ASSERT((XFS_BUF_ISERROR(bp)) == 0);
352 if (!(XFS_BUF_ISDONE(bp))) {
353 xfs_buftrace("READ_BUF_INCORE !DONE", bp);
354 ASSERT(!XFS_BUF_ISASYNC(bp));
355 XFS_BUF_READ(bp);
356 xfsbdstrat(tp->t_mountp, bp);
David Chinnerd64e31a2008-04-10 12:22:17 +1000357 error = xfs_iowait(bp);
358 if (error) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 xfs_ioerror_alert("xfs_trans_read_buf", mp,
360 bp, blkno);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 xfs_buf_relse(bp);
362 /*
David Chinnerd64e31a2008-04-10 12:22:17 +1000363 * We can gracefully recover from most read
364 * errors. Ones we can't are those that happen
365 * after the transaction's already dirty.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700366 */
367 if (tp->t_flags & XFS_TRANS_DIRTY)
368 xfs_force_shutdown(tp->t_mountp,
Nathan Scott7d04a332006-06-09 14:58:38 +1000369 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 return error;
371 }
372 }
373 /*
374 * We never locked this buf ourselves, so we shouldn't
375 * brelse it either. Just get out.
376 */
377 if (XFS_FORCED_SHUTDOWN(mp)) {
378 xfs_buftrace("READ_BUF_INCORE XFSSHUTDN", bp);
379 *bpp = NULL;
380 return XFS_ERROR(EIO);
381 }
382
383
384 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
385 bip->bli_recur++;
386
387 ASSERT(atomic_read(&bip->bli_refcount) > 0);
388 xfs_buf_item_trace("READ RECUR", bip);
389 *bpp = bp;
390 return 0;
391 }
392
393 /*
394 * We always specify the BUF_BUSY flag within a transaction so
395 * that get_buf does not try to push out a delayed write buffer
396 * which might cause another transaction to take place (if the
397 * buffer was delayed alloc). Such recursive transactions can
398 * easily deadlock with our current transaction as well as cause
399 * us to run out of stack space.
400 */
401 bp = xfs_buf_read_flags(target, blkno, len, flags | BUF_BUSY);
402 if (bp == NULL) {
403 *bpp = NULL;
404 return 0;
405 }
406 if (XFS_BUF_GETERROR(bp) != 0) {
407 XFS_BUF_SUPER_STALE(bp);
408 xfs_buftrace("READ ERROR", bp);
409 error = XFS_BUF_GETERROR(bp);
410
411 xfs_ioerror_alert("xfs_trans_read_buf", mp,
412 bp, blkno);
413 if (tp->t_flags & XFS_TRANS_DIRTY)
Nathan Scott7d04a332006-06-09 14:58:38 +1000414 xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 xfs_buf_relse(bp);
416 return error;
417 }
418#ifdef DEBUG
419 if (xfs_do_error && !(tp->t_flags & XFS_TRANS_DIRTY)) {
420 if (xfs_error_target == target) {
421 if (((xfs_req_num++) % xfs_error_mod) == 0) {
422 xfs_force_shutdown(tp->t_mountp,
Nathan Scott7d04a332006-06-09 14:58:38 +1000423 SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 xfs_buf_relse(bp);
Nathan Scottb6574522006-06-09 15:29:40 +1000425 cmn_err(CE_DEBUG, "Returning trans error!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700426 return XFS_ERROR(EIO);
427 }
428 }
429 }
430#endif
431 if (XFS_FORCED_SHUTDOWN(mp))
432 goto shutdown_abort;
433
434 /*
435 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
436 * it doesn't have one yet, then allocate one and initialize it.
437 * The checks to see if one is there are in xfs_buf_item_init().
438 */
439 xfs_buf_item_init(bp, tp->t_mountp);
440
441 /*
442 * Set the recursion count for the buffer within this transaction
443 * to 0.
444 */
445 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
446 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
447 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
448 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
449 bip->bli_recur = 0;
450
451 /*
452 * Take a reference for this transaction on the buf item.
453 */
454 atomic_inc(&bip->bli_refcount);
455
456 /*
457 * Get a log_item_desc to point at the new item.
458 */
459 (void) xfs_trans_add_item(tp, (xfs_log_item_t*)bip);
460
461 /*
462 * Initialize b_fsprivate2 so we can find it with incore_match()
463 * above.
464 */
465 XFS_BUF_SET_FSPRIVATE2(bp, tp);
466
467 xfs_buftrace("TRANS READ", bp);
468 xfs_buf_item_trace("READ", bip);
469 *bpp = bp;
470 return 0;
471
472shutdown_abort:
473 /*
474 * the theory here is that buffer is good but we're
475 * bailing out because the filesystem is being forcibly
476 * shut down. So we should leave the b_flags alone since
477 * the buffer's not staled and just get out.
478 */
479#if defined(DEBUG)
480 if (XFS_BUF_ISSTALE(bp) && XFS_BUF_ISDELAYWRITE(bp))
481 cmn_err(CE_NOTE, "about to pop assert, bp == 0x%p", bp);
482#endif
483 ASSERT((XFS_BUF_BFLAGS(bp) & (XFS_B_STALE|XFS_B_DELWRI)) !=
484 (XFS_B_STALE|XFS_B_DELWRI));
485
486 xfs_buftrace("READ_BUF XFSSHUTDN", bp);
487 xfs_buf_relse(bp);
488 *bpp = NULL;
489 return XFS_ERROR(EIO);
490}
491
492
493/*
494 * Release the buffer bp which was previously acquired with one of the
495 * xfs_trans_... buffer allocation routines if the buffer has not
496 * been modified within this transaction. If the buffer is modified
497 * within this transaction, do decrement the recursion count but do
498 * not release the buffer even if the count goes to 0. If the buffer is not
499 * modified within the transaction, decrement the recursion count and
500 * release the buffer if the recursion count goes to 0.
501 *
502 * If the buffer is to be released and it was not modified before
503 * this transaction began, then free the buf_log_item associated with it.
504 *
505 * If the transaction pointer is NULL, make this just a normal
506 * brelse() call.
507 */
508void
509xfs_trans_brelse(xfs_trans_t *tp,
510 xfs_buf_t *bp)
511{
512 xfs_buf_log_item_t *bip;
513 xfs_log_item_t *lip;
514 xfs_log_item_desc_t *lidp;
515
516 /*
517 * Default to a normal brelse() call if the tp is NULL.
518 */
519 if (tp == NULL) {
520 ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL);
521 /*
522 * If there's a buf log item attached to the buffer,
523 * then let the AIL know that the buffer is being
524 * unlocked.
525 */
526 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
527 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
528 if (lip->li_type == XFS_LI_BUF) {
529 bip = XFS_BUF_FSPRIVATE(bp,xfs_buf_log_item_t*);
530 xfs_trans_unlocked_item(
531 bip->bli_item.li_mountp,
532 lip);
533 }
534 }
535 xfs_buf_relse(bp);
536 return;
537 }
538
539 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
540 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
541 ASSERT(bip->bli_item.li_type == XFS_LI_BUF);
542 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
543 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
544 ASSERT(atomic_read(&bip->bli_refcount) > 0);
545
546 /*
547 * Find the item descriptor pointing to this buffer's
548 * log item. It must be there.
549 */
550 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
551 ASSERT(lidp != NULL);
552
553 /*
554 * If the release is just for a recursive lock,
555 * then decrement the count and return.
556 */
557 if (bip->bli_recur > 0) {
558 bip->bli_recur--;
559 xfs_buf_item_trace("RELSE RECUR", bip);
560 return;
561 }
562
563 /*
564 * If the buffer is dirty within this transaction, we can't
565 * release it until we commit.
566 */
567 if (lidp->lid_flags & XFS_LID_DIRTY) {
568 xfs_buf_item_trace("RELSE DIRTY", bip);
569 return;
570 }
571
572 /*
573 * If the buffer has been invalidated, then we can't release
574 * it until the transaction commits to disk unless it is re-dirtied
575 * as part of this transaction. This prevents us from pulling
576 * the item from the AIL before we should.
577 */
578 if (bip->bli_flags & XFS_BLI_STALE) {
579 xfs_buf_item_trace("RELSE STALE", bip);
580 return;
581 }
582
583 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
584 xfs_buf_item_trace("RELSE", bip);
585
586 /*
587 * Free up the log item descriptor tracking the released item.
588 */
589 xfs_trans_free_item(tp, lidp);
590
591 /*
592 * Clear the hold flag in the buf log item if it is set.
593 * We wouldn't want the next user of the buffer to
594 * get confused.
595 */
596 if (bip->bli_flags & XFS_BLI_HOLD) {
597 bip->bli_flags &= ~XFS_BLI_HOLD;
598 }
599
600 /*
601 * Drop our reference to the buf log item.
602 */
603 atomic_dec(&bip->bli_refcount);
604
605 /*
606 * If the buf item is not tracking data in the log, then
607 * we must free it before releasing the buffer back to the
608 * free pool. Before releasing the buffer to the free pool,
609 * clear the transaction pointer in b_fsprivate2 to dissolve
610 * its relation to this transaction.
611 */
612 if (!xfs_buf_item_dirty(bip)) {
613/***
614 ASSERT(bp->b_pincount == 0);
615***/
616 ASSERT(atomic_read(&bip->bli_refcount) == 0);
617 ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL));
618 ASSERT(!(bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF));
619 xfs_buf_item_relse(bp);
620 bip = NULL;
621 }
622 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
623
624 /*
625 * If we've still got a buf log item on the buffer, then
626 * tell the AIL that the buffer is being unlocked.
627 */
628 if (bip != NULL) {
629 xfs_trans_unlocked_item(bip->bli_item.li_mountp,
630 (xfs_log_item_t*)bip);
631 }
632
633 xfs_buf_relse(bp);
634 return;
635}
636
637/*
638 * Add the locked buffer to the transaction.
639 * The buffer must be locked, and it cannot be associated with any
640 * transaction.
641 *
642 * If the buffer does not yet have a buf log item associated with it,
643 * then allocate one for it. Then add the buf item to the transaction.
644 */
645void
646xfs_trans_bjoin(xfs_trans_t *tp,
647 xfs_buf_t *bp)
648{
649 xfs_buf_log_item_t *bip;
650
651 ASSERT(XFS_BUF_ISBUSY(bp));
652 ASSERT(XFS_BUF_FSPRIVATE2(bp, void *) == NULL);
653
654 /*
655 * The xfs_buf_log_item pointer is stored in b_fsprivate. If
656 * it doesn't have one yet, then allocate one and initialize it.
657 * The checks to see if one is there are in xfs_buf_item_init().
658 */
659 xfs_buf_item_init(bp, tp->t_mountp);
660 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
661 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
662 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
663 ASSERT(!(bip->bli_flags & XFS_BLI_LOGGED));
664
665 /*
666 * Take a reference for this transaction on the buf item.
667 */
668 atomic_inc(&bip->bli_refcount);
669
670 /*
671 * Get a log_item_desc to point at the new item.
672 */
673 (void) xfs_trans_add_item(tp, (xfs_log_item_t *)bip);
674
675 /*
676 * Initialize b_fsprivate2 so we can find it with incore_match()
677 * in xfs_trans_get_buf() and friends above.
678 */
679 XFS_BUF_SET_FSPRIVATE2(bp, tp);
680
681 xfs_buf_item_trace("BJOIN", bip);
682}
683
684/*
685 * Mark the buffer as not needing to be unlocked when the buf item's
686 * IOP_UNLOCK() routine is called. The buffer must already be locked
687 * and associated with the given transaction.
688 */
689/* ARGSUSED */
690void
691xfs_trans_bhold(xfs_trans_t *tp,
692 xfs_buf_t *bp)
693{
694 xfs_buf_log_item_t *bip;
695
696 ASSERT(XFS_BUF_ISBUSY(bp));
697 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
698 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
699
700 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
701 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
702 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
703 ASSERT(atomic_read(&bip->bli_refcount) > 0);
704 bip->bli_flags |= XFS_BLI_HOLD;
705 xfs_buf_item_trace("BHOLD", bip);
706}
707
708/*
Tim Shimminefa092f2005-09-05 08:29:01 +1000709 * Cancel the previous buffer hold request made on this buffer
710 * for this transaction.
711 */
712void
713xfs_trans_bhold_release(xfs_trans_t *tp,
714 xfs_buf_t *bp)
715{
716 xfs_buf_log_item_t *bip;
717
718 ASSERT(XFS_BUF_ISBUSY(bp));
719 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
720 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
721
722 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
723 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
724 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_CANCEL));
725 ASSERT(atomic_read(&bip->bli_refcount) > 0);
726 ASSERT(bip->bli_flags & XFS_BLI_HOLD);
727 bip->bli_flags &= ~XFS_BLI_HOLD;
728 xfs_buf_item_trace("BHOLD RELEASE", bip);
729}
730
731/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 * This is called to mark bytes first through last inclusive of the given
733 * buffer as needing to be logged when the transaction is committed.
734 * The buffer must already be associated with the given transaction.
735 *
736 * First and last are numbers relative to the beginning of this buffer,
737 * so the first byte in the buffer is numbered 0 regardless of the
738 * value of b_blkno.
739 */
740void
741xfs_trans_log_buf(xfs_trans_t *tp,
742 xfs_buf_t *bp,
743 uint first,
744 uint last)
745{
746 xfs_buf_log_item_t *bip;
747 xfs_log_item_desc_t *lidp;
748
749 ASSERT(XFS_BUF_ISBUSY(bp));
750 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
751 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
752 ASSERT((first <= last) && (last < XFS_BUF_COUNT(bp)));
753 ASSERT((XFS_BUF_IODONE_FUNC(bp) == NULL) ||
754 (XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks));
755
756 /*
757 * Mark the buffer as needing to be written out eventually,
758 * and set its iodone function to remove the buffer's buf log
759 * item from the AIL and free it when the buffer is flushed
760 * to disk. See xfs_buf_attach_iodone() for more details
761 * on li_cb and xfs_buf_iodone_callbacks().
762 * If we end up aborting this transaction, we trap this buffer
763 * inside the b_bdstrat callback so that this won't get written to
764 * disk.
765 */
766 XFS_BUF_DELAYWRITE(bp);
767 XFS_BUF_DONE(bp);
768
769 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
770 ASSERT(atomic_read(&bip->bli_refcount) > 0);
771 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
772 bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*))xfs_buf_iodone;
773
774 /*
775 * If we invalidated the buffer within this transaction, then
776 * cancel the invalidation now that we're dirtying the buffer
777 * again. There are no races with the code in xfs_buf_item_unpin(),
778 * because we have a reference to the buffer this entire time.
779 */
780 if (bip->bli_flags & XFS_BLI_STALE) {
781 xfs_buf_item_trace("BLOG UNSTALE", bip);
782 bip->bli_flags &= ~XFS_BLI_STALE;
783 ASSERT(XFS_BUF_ISSTALE(bp));
784 XFS_BUF_UNSTALE(bp);
785 bip->bli_format.blf_flags &= ~XFS_BLI_CANCEL;
786 }
787
788 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
789 ASSERT(lidp != NULL);
790
791 tp->t_flags |= XFS_TRANS_DIRTY;
792 lidp->lid_flags |= XFS_LID_DIRTY;
793 lidp->lid_flags &= ~XFS_LID_BUF_STALE;
794 bip->bli_flags |= XFS_BLI_LOGGED;
795 xfs_buf_item_log(bip, first, last);
796 xfs_buf_item_trace("BLOG", bip);
797}
798
799
800/*
801 * This called to invalidate a buffer that is being used within
802 * a transaction. Typically this is because the blocks in the
803 * buffer are being freed, so we need to prevent it from being
804 * written out when we're done. Allowing it to be written again
805 * might overwrite data in the free blocks if they are reallocated
806 * to a file.
807 *
808 * We prevent the buffer from being written out by clearing the
809 * B_DELWRI flag. We can't always
810 * get rid of the buf log item at this point, though, because
811 * the buffer may still be pinned by another transaction. If that
812 * is the case, then we'll wait until the buffer is committed to
813 * disk for the last time (we can tell by the ref count) and
814 * free it in xfs_buf_item_unpin(). Until it is cleaned up we
815 * will keep the buffer locked so that the buffer and buf log item
816 * are not reused.
817 */
818void
819xfs_trans_binval(
820 xfs_trans_t *tp,
821 xfs_buf_t *bp)
822{
823 xfs_log_item_desc_t *lidp;
824 xfs_buf_log_item_t *bip;
825
826 ASSERT(XFS_BUF_ISBUSY(bp));
827 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
828 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
829
830 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
831 lidp = xfs_trans_find_item(tp, (xfs_log_item_t*)bip);
832 ASSERT(lidp != NULL);
833 ASSERT(atomic_read(&bip->bli_refcount) > 0);
834
835 if (bip->bli_flags & XFS_BLI_STALE) {
836 /*
837 * If the buffer is already invalidated, then
838 * just return.
839 */
840 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
841 ASSERT(XFS_BUF_ISSTALE(bp));
842 ASSERT(!(bip->bli_flags & (XFS_BLI_LOGGED | XFS_BLI_DIRTY)));
843 ASSERT(!(bip->bli_format.blf_flags & XFS_BLI_INODE_BUF));
844 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
845 ASSERT(lidp->lid_flags & XFS_LID_DIRTY);
846 ASSERT(tp->t_flags & XFS_TRANS_DIRTY);
847 xfs_buftrace("XFS_BINVAL RECUR", bp);
848 xfs_buf_item_trace("BINVAL RECUR", bip);
849 return;
850 }
851
852 /*
853 * Clear the dirty bit in the buffer and set the STALE flag
854 * in the buf log item. The STALE flag will be used in
855 * xfs_buf_item_unpin() to determine if it should clean up
856 * when the last reference to the buf item is given up.
857 * We set the XFS_BLI_CANCEL flag in the buf log format structure
858 * and log the buf item. This will be used at recovery time
859 * to determine that copies of the buffer in the log before
860 * this should not be replayed.
861 * We mark the item descriptor and the transaction dirty so
862 * that we'll hold the buffer until after the commit.
863 *
864 * Since we're invalidating the buffer, we also clear the state
865 * about which parts of the buffer have been logged. We also
866 * clear the flag indicating that this is an inode buffer since
867 * the data in the buffer will no longer be valid.
868 *
869 * We set the stale bit in the buffer as well since we're getting
870 * rid of it.
871 */
872 XFS_BUF_UNDELAYWRITE(bp);
873 XFS_BUF_STALE(bp);
874 bip->bli_flags |= XFS_BLI_STALE;
875 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_DIRTY);
876 bip->bli_format.blf_flags &= ~XFS_BLI_INODE_BUF;
877 bip->bli_format.blf_flags |= XFS_BLI_CANCEL;
878 memset((char *)(bip->bli_format.blf_data_map), 0,
879 (bip->bli_format.blf_map_size * sizeof(uint)));
880 lidp->lid_flags |= XFS_LID_DIRTY|XFS_LID_BUF_STALE;
881 tp->t_flags |= XFS_TRANS_DIRTY;
882 xfs_buftrace("XFS_BINVAL", bp);
883 xfs_buf_item_trace("BINVAL", bip);
884}
885
886/*
887 * This call is used to indicate that the buffer contains on-disk
888 * inodes which must be handled specially during recovery. They
889 * require special handling because only the di_next_unlinked from
890 * the inodes in the buffer should be recovered. The rest of the
891 * data in the buffer is logged via the inodes themselves.
892 *
893 * All we do is set the XFS_BLI_INODE_BUF flag in the buffer's log
894 * format structure so that we'll know what to do at recovery time.
895 */
896/* ARGSUSED */
897void
898xfs_trans_inode_buf(
899 xfs_trans_t *tp,
900 xfs_buf_t *bp)
901{
902 xfs_buf_log_item_t *bip;
903
904 ASSERT(XFS_BUF_ISBUSY(bp));
905 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
906 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
907
908 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
909 ASSERT(atomic_read(&bip->bli_refcount) > 0);
910
911 bip->bli_format.blf_flags |= XFS_BLI_INODE_BUF;
912}
913
914/*
915 * This call is used to indicate that the buffer is going to
916 * be staled and was an inode buffer. This means it gets
917 * special processing during unpin - where any inodes
918 * associated with the buffer should be removed from ail.
919 * There is also special processing during recovery,
920 * any replay of the inodes in the buffer needs to be
921 * prevented as the buffer may have been reused.
922 */
923void
924xfs_trans_stale_inode_buf(
925 xfs_trans_t *tp,
926 xfs_buf_t *bp)
927{
928 xfs_buf_log_item_t *bip;
929
930 ASSERT(XFS_BUF_ISBUSY(bp));
931 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
932 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
933
934 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
935 ASSERT(atomic_read(&bip->bli_refcount) > 0);
936
937 bip->bli_flags |= XFS_BLI_STALE_INODE;
938 bip->bli_item.li_cb = (void(*)(xfs_buf_t*,xfs_log_item_t*))
939 xfs_buf_iodone;
940}
941
942
943
944/*
945 * Mark the buffer as being one which contains newly allocated
946 * inodes. We need to make sure that even if this buffer is
947 * relogged as an 'inode buf' we still recover all of the inode
948 * images in the face of a crash. This works in coordination with
949 * xfs_buf_item_committed() to ensure that the buffer remains in the
950 * AIL at its original location even after it has been relogged.
951 */
952/* ARGSUSED */
953void
954xfs_trans_inode_alloc_buf(
955 xfs_trans_t *tp,
956 xfs_buf_t *bp)
957{
958 xfs_buf_log_item_t *bip;
959
960 ASSERT(XFS_BUF_ISBUSY(bp));
961 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
962 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
963
964 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
965 ASSERT(atomic_read(&bip->bli_refcount) > 0);
966
967 bip->bli_flags |= XFS_BLI_INODE_ALLOC_BUF;
968}
969
970
971/*
972 * Similar to xfs_trans_inode_buf(), this marks the buffer as a cluster of
973 * dquots. However, unlike in inode buffer recovery, dquot buffers get
974 * recovered in their entirety. (Hence, no XFS_BLI_DQUOT_ALLOC_BUF flag).
975 * The only thing that makes dquot buffers different from regular
976 * buffers is that we must not replay dquot bufs when recovering
977 * if a _corresponding_ quotaoff has happened. We also have to distinguish
978 * between usr dquot bufs and grp dquot bufs, because usr and grp quotas
979 * can be turned off independently.
980 */
981/* ARGSUSED */
982void
983xfs_trans_dquot_buf(
984 xfs_trans_t *tp,
985 xfs_buf_t *bp,
986 uint type)
987{
988 xfs_buf_log_item_t *bip;
989
990 ASSERT(XFS_BUF_ISBUSY(bp));
991 ASSERT(XFS_BUF_FSPRIVATE2(bp, xfs_trans_t *) == tp);
992 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
993 ASSERT(type == XFS_BLI_UDQUOT_BUF ||
Nathan Scottc8ad20f2005-06-21 15:38:48 +1000994 type == XFS_BLI_PDQUOT_BUF ||
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995 type == XFS_BLI_GDQUOT_BUF);
996
997 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *);
998 ASSERT(atomic_read(&bip->bli_refcount) > 0);
999
1000 bip->bli_format.blf_flags |= type;
1001}
1002
1003/*
1004 * Check to see if a buffer matching the given parameters is already
1005 * a part of the given transaction. Only check the first, embedded
1006 * chunk, since we don't want to spend all day scanning large transactions.
1007 */
1008STATIC xfs_buf_t *
1009xfs_trans_buf_item_match(
1010 xfs_trans_t *tp,
1011 xfs_buftarg_t *target,
1012 xfs_daddr_t blkno,
1013 int len)
1014{
1015 xfs_log_item_chunk_t *licp;
1016 xfs_log_item_desc_t *lidp;
1017 xfs_buf_log_item_t *blip;
1018 xfs_buf_t *bp;
1019 int i;
1020
1021 bp = NULL;
1022 len = BBTOB(len);
1023 licp = &tp->t_items;
1024 if (!XFS_LIC_ARE_ALL_FREE(licp)) {
1025 for (i = 0; i < licp->lic_unused; i++) {
1026 /*
1027 * Skip unoccupied slots.
1028 */
1029 if (XFS_LIC_ISFREE(licp, i)) {
1030 continue;
1031 }
1032
1033 lidp = XFS_LIC_SLOT(licp, i);
1034 blip = (xfs_buf_log_item_t *)lidp->lid_item;
1035 if (blip->bli_item.li_type != XFS_LI_BUF) {
1036 continue;
1037 }
1038
1039 bp = blip->bli_buf;
1040 if ((XFS_BUF_TARGET(bp) == target) &&
1041 (XFS_BUF_ADDR(bp) == blkno) &&
1042 (XFS_BUF_COUNT(bp) == len)) {
1043 /*
1044 * We found it. Break out and
1045 * return the pointer to the buffer.
1046 */
1047 break;
1048 } else {
1049 bp = NULL;
1050 }
1051 }
1052 }
1053 return bp;
1054}
1055
1056/*
1057 * Check to see if a buffer matching the given parameters is already
1058 * a part of the given transaction. Check all the chunks, we
1059 * want to be thorough.
1060 */
1061STATIC xfs_buf_t *
1062xfs_trans_buf_item_match_all(
1063 xfs_trans_t *tp,
1064 xfs_buftarg_t *target,
1065 xfs_daddr_t blkno,
1066 int len)
1067{
1068 xfs_log_item_chunk_t *licp;
1069 xfs_log_item_desc_t *lidp;
1070 xfs_buf_log_item_t *blip;
1071 xfs_buf_t *bp;
1072 int i;
1073
1074 bp = NULL;
1075 len = BBTOB(len);
1076 for (licp = &tp->t_items; licp != NULL; licp = licp->lic_next) {
1077 if (XFS_LIC_ARE_ALL_FREE(licp)) {
1078 ASSERT(licp == &tp->t_items);
1079 ASSERT(licp->lic_next == NULL);
1080 return NULL;
1081 }
1082 for (i = 0; i < licp->lic_unused; i++) {
1083 /*
1084 * Skip unoccupied slots.
1085 */
1086 if (XFS_LIC_ISFREE(licp, i)) {
1087 continue;
1088 }
1089
1090 lidp = XFS_LIC_SLOT(licp, i);
1091 blip = (xfs_buf_log_item_t *)lidp->lid_item;
1092 if (blip->bli_item.li_type != XFS_LI_BUF) {
1093 continue;
1094 }
1095
1096 bp = blip->bli_buf;
1097 if ((XFS_BUF_TARGET(bp) == target) &&
1098 (XFS_BUF_ADDR(bp) == blkno) &&
1099 (XFS_BUF_COUNT(bp) == len)) {
1100 /*
1101 * We found it. Break out and
1102 * return the pointer to the buffer.
1103 */
1104 return bp;
1105 }
1106 }
1107 }
1108 return NULL;
1109}