aboutsummaryrefslogtreecommitdiffstats
path: root/fs/xfs/xfs_buf_item.c
diff options
context:
space:
mode:
Diffstat (limited to 'fs/xfs/xfs_buf_item.c')
-rw-r--r--fs/xfs/xfs_buf_item.c1221
1 files changed, 1221 insertions, 0 deletions
diff --git a/fs/xfs/xfs_buf_item.c b/fs/xfs/xfs_buf_item.c
new file mode 100644
index 000000000000..9ab0039f07df
--- /dev/null
+++ b/fs/xfs/xfs_buf_item.c
@@ -0,0 +1,1221 @@
1/*
2 * Copyright (c) 2000-2004 Silicon Graphics, Inc. All Rights Reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11 *
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
18 *
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22 *
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
25 *
26 * http://www.sgi.com
27 *
28 * For further information regarding this notice, see:
29 *
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31 */
32
33/*
34 * This file contains the implementation of the xfs_buf_log_item.
35 * It contains the item operations used to manipulate the buf log
36 * items as well as utility routines used by the buffer specific
37 * transaction routines.
38 */
39
40#include "xfs.h"
41
42#include "xfs_macros.h"
43#include "xfs_types.h"
44#include "xfs_inum.h"
45#include "xfs_log.h"
46#include "xfs_trans.h"
47#include "xfs_buf_item.h"
48#include "xfs_sb.h"
49#include "xfs_dir.h"
50#include "xfs_dmapi.h"
51#include "xfs_mount.h"
52#include "xfs_trans_priv.h"
53#include "xfs_rw.h"
54#include "xfs_bit.h"
55#include "xfs_error.h"
56
57
58kmem_zone_t *xfs_buf_item_zone;
59
60#ifdef XFS_TRANS_DEBUG
61/*
62 * This function uses an alternate strategy for tracking the bytes
63 * that the user requests to be logged. This can then be used
64 * in conjunction with the bli_orig array in the buf log item to
65 * catch bugs in our callers' code.
66 *
67 * We also double check the bits set in xfs_buf_item_log using a
68 * simple algorithm to check that every byte is accounted for.
69 */
70STATIC void
71xfs_buf_item_log_debug(
72 xfs_buf_log_item_t *bip,
73 uint first,
74 uint last)
75{
76 uint x;
77 uint byte;
78 uint nbytes;
79 uint chunk_num;
80 uint word_num;
81 uint bit_num;
82 uint bit_set;
83 uint *wordp;
84
85 ASSERT(bip->bli_logged != NULL);
86 byte = first;
87 nbytes = last - first + 1;
88 bfset(bip->bli_logged, first, nbytes);
89 for (x = 0; x < nbytes; x++) {
90 chunk_num = byte >> XFS_BLI_SHIFT;
91 word_num = chunk_num >> BIT_TO_WORD_SHIFT;
92 bit_num = chunk_num & (NBWORD - 1);
93 wordp = &(bip->bli_format.blf_data_map[word_num]);
94 bit_set = *wordp & (1 << bit_num);
95 ASSERT(bit_set);
96 byte++;
97 }
98}
99
100/*
101 * This function is called when we flush something into a buffer without
102 * logging it. This happens for things like inodes which are logged
103 * separately from the buffer.
104 */
105void
106xfs_buf_item_flush_log_debug(
107 xfs_buf_t *bp,
108 uint first,
109 uint last)
110{
111 xfs_buf_log_item_t *bip;
112 uint nbytes;
113
114 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
115 if ((bip == NULL) || (bip->bli_item.li_type != XFS_LI_BUF)) {
116 return;
117 }
118
119 ASSERT(bip->bli_logged != NULL);
120 nbytes = last - first + 1;
121 bfset(bip->bli_logged, first, nbytes);
122}
123
124/*
125 * This function is called to verify that our caller's have logged
126 * all the bytes that they changed.
127 *
128 * It does this by comparing the original copy of the buffer stored in
129 * the buf log item's bli_orig array to the current copy of the buffer
130 * and ensuring that all bytes which miscompare are set in the bli_logged
131 * array of the buf log item.
132 */
133STATIC void
134xfs_buf_item_log_check(
135 xfs_buf_log_item_t *bip)
136{
137 char *orig;
138 char *buffer;
139 int x;
140 xfs_buf_t *bp;
141
142 ASSERT(bip->bli_orig != NULL);
143 ASSERT(bip->bli_logged != NULL);
144
145 bp = bip->bli_buf;
146 ASSERT(XFS_BUF_COUNT(bp) > 0);
147 ASSERT(XFS_BUF_PTR(bp) != NULL);
148 orig = bip->bli_orig;
149 buffer = XFS_BUF_PTR(bp);
150 for (x = 0; x < XFS_BUF_COUNT(bp); x++) {
151 if (orig[x] != buffer[x] && !btst(bip->bli_logged, x))
152 cmn_err(CE_PANIC,
153 "xfs_buf_item_log_check bip %x buffer %x orig %x index %d",
154 bip, bp, orig, x);
155 }
156}
157#else
158#define xfs_buf_item_log_debug(x,y,z)
159#define xfs_buf_item_log_check(x)
160#endif
161
162STATIC void xfs_buf_error_relse(xfs_buf_t *bp);
163STATIC void xfs_buf_do_callbacks(xfs_buf_t *bp, xfs_log_item_t *lip);
164
165/*
166 * This returns the number of log iovecs needed to log the
167 * given buf log item.
168 *
169 * It calculates this as 1 iovec for the buf log format structure
170 * and 1 for each stretch of non-contiguous chunks to be logged.
171 * Contiguous chunks are logged in a single iovec.
172 *
173 * If the XFS_BLI_STALE flag has been set, then log nothing.
174 */
175uint
176xfs_buf_item_size(
177 xfs_buf_log_item_t *bip)
178{
179 uint nvecs;
180 int next_bit;
181 int last_bit;
182 xfs_buf_t *bp;
183
184 ASSERT(atomic_read(&bip->bli_refcount) > 0);
185 if (bip->bli_flags & XFS_BLI_STALE) {
186 /*
187 * The buffer is stale, so all we need to log
188 * is the buf log format structure with the
189 * cancel flag in it.
190 */
191 xfs_buf_item_trace("SIZE STALE", bip);
192 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
193 return 1;
194 }
195
196 bp = bip->bli_buf;
197 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
198 nvecs = 1;
199 last_bit = xfs_next_bit(bip->bli_format.blf_data_map,
200 bip->bli_format.blf_map_size, 0);
201 ASSERT(last_bit != -1);
202 nvecs++;
203 while (last_bit != -1) {
204 /*
205 * This takes the bit number to start looking from and
206 * returns the next set bit from there. It returns -1
207 * if there are no more bits set or the start bit is
208 * beyond the end of the bitmap.
209 */
210 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
211 bip->bli_format.blf_map_size,
212 last_bit + 1);
213 /*
214 * If we run out of bits, leave the loop,
215 * else if we find a new set of bits bump the number of vecs,
216 * else keep scanning the current set of bits.
217 */
218 if (next_bit == -1) {
219 last_bit = -1;
220 } else if (next_bit != last_bit + 1) {
221 last_bit = next_bit;
222 nvecs++;
223 } else if (xfs_buf_offset(bp, next_bit * XFS_BLI_CHUNK) !=
224 (xfs_buf_offset(bp, last_bit * XFS_BLI_CHUNK) +
225 XFS_BLI_CHUNK)) {
226 last_bit = next_bit;
227 nvecs++;
228 } else {
229 last_bit++;
230 }
231 }
232
233 xfs_buf_item_trace("SIZE NORM", bip);
234 return nvecs;
235}
236
237/*
238 * This is called to fill in the vector of log iovecs for the
239 * given log buf item. It fills the first entry with a buf log
240 * format structure, and the rest point to contiguous chunks
241 * within the buffer.
242 */
243void
244xfs_buf_item_format(
245 xfs_buf_log_item_t *bip,
246 xfs_log_iovec_t *log_vector)
247{
248 uint base_size;
249 uint nvecs;
250 xfs_log_iovec_t *vecp;
251 xfs_buf_t *bp;
252 int first_bit;
253 int last_bit;
254 int next_bit;
255 uint nbits;
256 uint buffer_offset;
257
258 ASSERT(atomic_read(&bip->bli_refcount) > 0);
259 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
260 (bip->bli_flags & XFS_BLI_STALE));
261 bp = bip->bli_buf;
262 ASSERT(XFS_BUF_BP_ISMAPPED(bp));
263 vecp = log_vector;
264
265 /*
266 * The size of the base structure is the size of the
267 * declared structure plus the space for the extra words
268 * of the bitmap. We subtract one from the map size, because
269 * the first element of the bitmap is accounted for in the
270 * size of the base structure.
271 */
272 base_size =
273 (uint)(sizeof(xfs_buf_log_format_t) +
274 ((bip->bli_format.blf_map_size - 1) * sizeof(uint)));
275 vecp->i_addr = (xfs_caddr_t)&bip->bli_format;
276 vecp->i_len = base_size;
277 vecp++;
278 nvecs = 1;
279
280 if (bip->bli_flags & XFS_BLI_STALE) {
281 /*
282 * The buffer is stale, so all we need to log
283 * is the buf log format structure with the
284 * cancel flag in it.
285 */
286 xfs_buf_item_trace("FORMAT STALE", bip);
287 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
288 bip->bli_format.blf_size = nvecs;
289 return;
290 }
291
292 /*
293 * Fill in an iovec for each set of contiguous chunks.
294 */
295 first_bit = xfs_next_bit(bip->bli_format.blf_data_map,
296 bip->bli_format.blf_map_size, 0);
297 ASSERT(first_bit != -1);
298 last_bit = first_bit;
299 nbits = 1;
300 for (;;) {
301 /*
302 * This takes the bit number to start looking from and
303 * returns the next set bit from there. It returns -1
304 * if there are no more bits set or the start bit is
305 * beyond the end of the bitmap.
306 */
307 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
308 bip->bli_format.blf_map_size,
309 (uint)last_bit + 1);
310 /*
311 * If we run out of bits fill in the last iovec and get
312 * out of the loop.
313 * Else if we start a new set of bits then fill in the
314 * iovec for the series we were looking at and start
315 * counting the bits in the new one.
316 * Else we're still in the same set of bits so just
317 * keep counting and scanning.
318 */
319 if (next_bit == -1) {
320 buffer_offset = first_bit * XFS_BLI_CHUNK;
321 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
322 vecp->i_len = nbits * XFS_BLI_CHUNK;
323 nvecs++;
324 break;
325 } else if (next_bit != last_bit + 1) {
326 buffer_offset = first_bit * XFS_BLI_CHUNK;
327 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
328 vecp->i_len = nbits * XFS_BLI_CHUNK;
329 nvecs++;
330 vecp++;
331 first_bit = next_bit;
332 last_bit = next_bit;
333 nbits = 1;
334 } else if (xfs_buf_offset(bp, next_bit << XFS_BLI_SHIFT) !=
335 (xfs_buf_offset(bp, last_bit << XFS_BLI_SHIFT) +
336 XFS_BLI_CHUNK)) {
337 buffer_offset = first_bit * XFS_BLI_CHUNK;
338 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
339 vecp->i_len = nbits * XFS_BLI_CHUNK;
340/* You would think we need to bump the nvecs here too, but we do not
341 * this number is used by recovery, and it gets confused by the boundary
342 * split here
343 * nvecs++;
344 */
345 vecp++;
346 first_bit = next_bit;
347 last_bit = next_bit;
348 nbits = 1;
349 } else {
350 last_bit++;
351 nbits++;
352 }
353 }
354 bip->bli_format.blf_size = nvecs;
355
356 /*
357 * Check to make sure everything is consistent.
358 */
359 xfs_buf_item_trace("FORMAT NORM", bip);
360 xfs_buf_item_log_check(bip);
361}
362
363/*
364 * This is called to pin the buffer associated with the buf log
365 * item in memory so it cannot be written out. Simply call bpin()
366 * on the buffer to do this.
367 */
368void
369xfs_buf_item_pin(
370 xfs_buf_log_item_t *bip)
371{
372 xfs_buf_t *bp;
373
374 bp = bip->bli_buf;
375 ASSERT(XFS_BUF_ISBUSY(bp));
376 ASSERT(atomic_read(&bip->bli_refcount) > 0);
377 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
378 (bip->bli_flags & XFS_BLI_STALE));
379 xfs_buf_item_trace("PIN", bip);
380 xfs_buftrace("XFS_PIN", bp);
381 xfs_bpin(bp);
382}
383
384
385/*
386 * This is called to unpin the buffer associated with the buf log
387 * item which was previously pinned with a call to xfs_buf_item_pin().
388 * Just call bunpin() on the buffer to do this.
389 *
390 * Also drop the reference to the buf item for the current transaction.
391 * If the XFS_BLI_STALE flag is set and we are the last reference,
392 * then free up the buf log item and unlock the buffer.
393 */
394void
395xfs_buf_item_unpin(
396 xfs_buf_log_item_t *bip,
397 int stale)
398{
399 xfs_mount_t *mp;
400 xfs_buf_t *bp;
401 int freed;
402 SPLDECL(s);
403
404 bp = bip->bli_buf;
405 ASSERT(bp != NULL);
406 ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip);
407 ASSERT(atomic_read(&bip->bli_refcount) > 0);
408 xfs_buf_item_trace("UNPIN", bip);
409 xfs_buftrace("XFS_UNPIN", bp);
410
411 freed = atomic_dec_and_test(&bip->bli_refcount);
412 mp = bip->bli_item.li_mountp;
413 xfs_bunpin(bp);
414 if (freed && stale) {
415 ASSERT(bip->bli_flags & XFS_BLI_STALE);
416 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
417 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
418 ASSERT(XFS_BUF_ISSTALE(bp));
419 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
420 xfs_buf_item_trace("UNPIN STALE", bip);
421 xfs_buftrace("XFS_UNPIN STALE", bp);
422 /*
423 * If we get called here because of an IO error, we may
424 * or may not have the item on the AIL. xfs_trans_delete_ail()
425 * will take care of that situation.
426 * xfs_trans_delete_ail() drops the AIL lock.
427 */
428 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
429 xfs_buf_do_callbacks(bp, (xfs_log_item_t *)bip);
430 XFS_BUF_SET_FSPRIVATE(bp, NULL);
431 XFS_BUF_CLR_IODONE_FUNC(bp);
432 } else {
433 AIL_LOCK(mp,s);
434 xfs_trans_delete_ail(mp, (xfs_log_item_t *)bip, s);
435 xfs_buf_item_relse(bp);
436 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL);
437 }
438 xfs_buf_relse(bp);
439 }
440}
441
442/*
443 * this is called from uncommit in the forced-shutdown path.
444 * we need to check to see if the reference count on the log item
445 * is going to drop to zero. If so, unpin will free the log item
446 * so we need to free the item's descriptor (that points to the item)
447 * in the transaction.
448 */
449void
450xfs_buf_item_unpin_remove(
451 xfs_buf_log_item_t *bip,
452 xfs_trans_t *tp)
453{
454 xfs_buf_t *bp;
455 xfs_log_item_desc_t *lidp;
456 int stale = 0;
457
458 bp = bip->bli_buf;
459 /*
460 * will xfs_buf_item_unpin() call xfs_buf_item_relse()?
461 */
462 if ((atomic_read(&bip->bli_refcount) == 1) &&
463 (bip->bli_flags & XFS_BLI_STALE)) {
464 ASSERT(XFS_BUF_VALUSEMA(bip->bli_buf) <= 0);
465 xfs_buf_item_trace("UNPIN REMOVE", bip);
466 xfs_buftrace("XFS_UNPIN_REMOVE", bp);
467 /*
468 * yes -- clear the xaction descriptor in-use flag
469 * and free the chunk if required. We can safely
470 * do some work here and then call buf_item_unpin
471 * to do the rest because if the if is true, then
472 * we are holding the buffer locked so no one else
473 * will be able to bump up the refcount.
474 */
475 lidp = xfs_trans_find_item(tp, (xfs_log_item_t *) bip);
476 stale = lidp->lid_flags & XFS_LID_BUF_STALE;
477 xfs_trans_free_item(tp, lidp);
478 /*
479 * Since the transaction no longer refers to the buffer,
480 * the buffer should no longer refer to the transaction.
481 */
482 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
483 }
484
485 xfs_buf_item_unpin(bip, stale);
486
487 return;
488}
489
490/*
491 * This is called to attempt to lock the buffer associated with this
492 * buf log item. Don't sleep on the buffer lock. If we can't get
493 * the lock right away, return 0. If we can get the lock, pull the
494 * buffer from the free list, mark it busy, and return 1.
495 */
496uint
497xfs_buf_item_trylock(
498 xfs_buf_log_item_t *bip)
499{
500 xfs_buf_t *bp;
501
502 bp = bip->bli_buf;
503
504 if (XFS_BUF_ISPINNED(bp)) {
505 return XFS_ITEM_PINNED;
506 }
507
508 if (!XFS_BUF_CPSEMA(bp)) {
509 return XFS_ITEM_LOCKED;
510 }
511
512 /*
513 * Remove the buffer from the free list. Only do this
514 * if it's on the free list. Private buffers like the
515 * superblock buffer are not.
516 */
517 XFS_BUF_HOLD(bp);
518
519 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
520 xfs_buf_item_trace("TRYLOCK SUCCESS", bip);
521 return XFS_ITEM_SUCCESS;
522}
523
524/*
525 * Release the buffer associated with the buf log item.
526 * If there is no dirty logged data associated with the
527 * buffer recorded in the buf log item, then free the
528 * buf log item and remove the reference to it in the
529 * buffer.
530 *
531 * This call ignores the recursion count. It is only called
532 * when the buffer should REALLY be unlocked, regardless
533 * of the recursion count.
534 *
535 * If the XFS_BLI_HOLD flag is set in the buf log item, then
536 * free the log item if necessary but do not unlock the buffer.
537 * This is for support of xfs_trans_bhold(). Make sure the
538 * XFS_BLI_HOLD field is cleared if we don't free the item.
539 */
540void
541xfs_buf_item_unlock(
542 xfs_buf_log_item_t *bip)
543{
544 int aborted;
545 xfs_buf_t *bp;
546 uint hold;
547
548 bp = bip->bli_buf;
549 xfs_buftrace("XFS_UNLOCK", bp);
550
551 /*
552 * Clear the buffer's association with this transaction.
553 */
554 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
555
556 /*
557 * If this is a transaction abort, don't return early.
558 * Instead, allow the brelse to happen.
559 * Normally it would be done for stale (cancelled) buffers
560 * at unpin time, but we'll never go through the pin/unpin
561 * cycle if we abort inside commit.
562 */
563 aborted = (bip->bli_item.li_flags & XFS_LI_ABORTED) != 0;
564
565 /*
566 * If the buf item is marked stale, then don't do anything.
567 * We'll unlock the buffer and free the buf item when the
568 * buffer is unpinned for the last time.
569 */
570 if (bip->bli_flags & XFS_BLI_STALE) {
571 bip->bli_flags &= ~XFS_BLI_LOGGED;
572 xfs_buf_item_trace("UNLOCK STALE", bip);
573 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
574 if (!aborted)
575 return;
576 }
577
578 /*
579 * Drop the transaction's reference to the log item if
580 * it was not logged as part of the transaction. Otherwise
581 * we'll drop the reference in xfs_buf_item_unpin() when
582 * the transaction is really through with the buffer.
583 */
584 if (!(bip->bli_flags & XFS_BLI_LOGGED)) {
585 atomic_dec(&bip->bli_refcount);
586 } else {
587 /*
588 * Clear the logged flag since this is per
589 * transaction state.
590 */
591 bip->bli_flags &= ~XFS_BLI_LOGGED;
592 }
593
594 /*
595 * Before possibly freeing the buf item, determine if we should
596 * release the buffer at the end of this routine.
597 */
598 hold = bip->bli_flags & XFS_BLI_HOLD;
599 xfs_buf_item_trace("UNLOCK", bip);
600
601 /*
602 * If the buf item isn't tracking any data, free it.
603 * Otherwise, if XFS_BLI_HOLD is set clear it.
604 */
605 if (xfs_count_bits(bip->bli_format.blf_data_map,
606 bip->bli_format.blf_map_size, 0) == 0) {
607 xfs_buf_item_relse(bp);
608 } else if (hold) {
609 bip->bli_flags &= ~XFS_BLI_HOLD;
610 }
611
612 /*
613 * Release the buffer if XFS_BLI_HOLD was not set.
614 */
615 if (!hold) {
616 xfs_buf_relse(bp);
617 }
618}
619
620/*
621 * This is called to find out where the oldest active copy of the
622 * buf log item in the on disk log resides now that the last log
623 * write of it completed at the given lsn.
624 * We always re-log all the dirty data in a buffer, so usually the
625 * latest copy in the on disk log is the only one that matters. For
626 * those cases we simply return the given lsn.
627 *
628 * The one exception to this is for buffers full of newly allocated
629 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
630 * flag set, indicating that only the di_next_unlinked fields from the
631 * inodes in the buffers will be replayed during recovery. If the
632 * original newly allocated inode images have not yet been flushed
633 * when the buffer is so relogged, then we need to make sure that we
634 * keep the old images in the 'active' portion of the log. We do this
635 * by returning the original lsn of that transaction here rather than
636 * the current one.
637 */
638xfs_lsn_t
639xfs_buf_item_committed(
640 xfs_buf_log_item_t *bip,
641 xfs_lsn_t lsn)
642{
643 xfs_buf_item_trace("COMMITTED", bip);
644 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
645 (bip->bli_item.li_lsn != 0)) {
646 return bip->bli_item.li_lsn;
647 }
648 return (lsn);
649}
650
651/*
652 * This is called when the transaction holding the buffer is aborted.
653 * Just behave as if the transaction had been cancelled. If we're shutting down
654 * and have aborted this transaction, we'll trap this buffer when it tries to
655 * get written out.
656 */
657void
658xfs_buf_item_abort(
659 xfs_buf_log_item_t *bip)
660{
661 xfs_buf_t *bp;
662
663 bp = bip->bli_buf;
664 xfs_buftrace("XFS_ABORT", bp);
665 XFS_BUF_SUPER_STALE(bp);
666 xfs_buf_item_unlock(bip);
667 return;
668}
669
670/*
671 * This is called to asynchronously write the buffer associated with this
672 * buf log item out to disk. The buffer will already have been locked by
673 * a successful call to xfs_buf_item_trylock(). If the buffer still has
674 * B_DELWRI set, then get it going out to disk with a call to bawrite().
675 * If not, then just release the buffer.
676 */
677void
678xfs_buf_item_push(
679 xfs_buf_log_item_t *bip)
680{
681 xfs_buf_t *bp;
682
683 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
684 xfs_buf_item_trace("PUSH", bip);
685
686 bp = bip->bli_buf;
687
688 if (XFS_BUF_ISDELAYWRITE(bp)) {
689 xfs_bawrite(bip->bli_item.li_mountp, bp);
690 } else {
691 xfs_buf_relse(bp);
692 }
693}
694
695/* ARGSUSED */
696void
697xfs_buf_item_committing(xfs_buf_log_item_t *bip, xfs_lsn_t commit_lsn)
698{
699}
700
701/*
702 * This is the ops vector shared by all buf log items.
703 */
704struct xfs_item_ops xfs_buf_item_ops = {
705 .iop_size = (uint(*)(xfs_log_item_t*))xfs_buf_item_size,
706 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
707 xfs_buf_item_format,
708 .iop_pin = (void(*)(xfs_log_item_t*))xfs_buf_item_pin,
709 .iop_unpin = (void(*)(xfs_log_item_t*, int))xfs_buf_item_unpin,
710 .iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t *))
711 xfs_buf_item_unpin_remove,
712 .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_buf_item_trylock,
713 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_buf_item_unlock,
714 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
715 xfs_buf_item_committed,
716 .iop_push = (void(*)(xfs_log_item_t*))xfs_buf_item_push,
717 .iop_abort = (void(*)(xfs_log_item_t*))xfs_buf_item_abort,
718 .iop_pushbuf = NULL,
719 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
720 xfs_buf_item_committing
721};
722
723
724/*
725 * Allocate a new buf log item to go with the given buffer.
726 * Set the buffer's b_fsprivate field to point to the new
727 * buf log item. If there are other item's attached to the
728 * buffer (see xfs_buf_attach_iodone() below), then put the
729 * buf log item at the front.
730 */
731void
732xfs_buf_item_init(
733 xfs_buf_t *bp,
734 xfs_mount_t *mp)
735{
736 xfs_log_item_t *lip;
737 xfs_buf_log_item_t *bip;
738 int chunks;
739 int map_size;
740
741 /*
742 * Check to see if there is already a buf log item for
743 * this buffer. If there is, it is guaranteed to be
744 * the first. If we do already have one, there is
745 * nothing to do here so return.
746 */
747 if (XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *) != mp)
748 XFS_BUF_SET_FSPRIVATE3(bp, mp);
749 XFS_BUF_SET_BDSTRAT_FUNC(bp, xfs_bdstrat_cb);
750 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
751 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
752 if (lip->li_type == XFS_LI_BUF) {
753 return;
754 }
755 }
756
757 /*
758 * chunks is the number of XFS_BLI_CHUNK size pieces
759 * the buffer can be divided into. Make sure not to
760 * truncate any pieces. map_size is the size of the
761 * bitmap needed to describe the chunks of the buffer.
762 */
763 chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLI_CHUNK - 1)) >> XFS_BLI_SHIFT);
764 map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);
765
766 bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
767 KM_SLEEP);
768 bip->bli_item.li_type = XFS_LI_BUF;
769 bip->bli_item.li_ops = &xfs_buf_item_ops;
770 bip->bli_item.li_mountp = mp;
771 bip->bli_buf = bp;
772 bip->bli_format.blf_type = XFS_LI_BUF;
773 bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp);
774 bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp));
775 bip->bli_format.blf_map_size = map_size;
776#ifdef XFS_BLI_TRACE
777 bip->bli_trace = ktrace_alloc(XFS_BLI_TRACE_SIZE, KM_SLEEP);
778#endif
779
780#ifdef XFS_TRANS_DEBUG
781 /*
782 * Allocate the arrays for tracking what needs to be logged
783 * and what our callers request to be logged. bli_orig
784 * holds a copy of the original, clean buffer for comparison
785 * against, and bli_logged keeps a 1 bit flag per byte in
786 * the buffer to indicate which bytes the callers have asked
787 * to have logged.
788 */
789 bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP);
790 memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp));
791 bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP);
792#endif
793
794 /*
795 * Put the buf item into the list of items attached to the
796 * buffer at the front.
797 */
798 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
799 bip->bli_item.li_bio_list =
800 XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
801 }
802 XFS_BUF_SET_FSPRIVATE(bp, bip);
803}
804
805
806/*
807 * Mark bytes first through last inclusive as dirty in the buf
808 * item's bitmap.
809 */
810void
811xfs_buf_item_log(
812 xfs_buf_log_item_t *bip,
813 uint first,
814 uint last)
815{
816 uint first_bit;
817 uint last_bit;
818 uint bits_to_set;
819 uint bits_set;
820 uint word_num;
821 uint *wordp;
822 uint bit;
823 uint end_bit;
824 uint mask;
825
826 /*
827 * Mark the item as having some dirty data for
828 * quick reference in xfs_buf_item_dirty.
829 */
830 bip->bli_flags |= XFS_BLI_DIRTY;
831
832 /*
833 * Convert byte offsets to bit numbers.
834 */
835 first_bit = first >> XFS_BLI_SHIFT;
836 last_bit = last >> XFS_BLI_SHIFT;
837
838 /*
839 * Calculate the total number of bits to be set.
840 */
841 bits_to_set = last_bit - first_bit + 1;
842
843 /*
844 * Get a pointer to the first word in the bitmap
845 * to set a bit in.
846 */
847 word_num = first_bit >> BIT_TO_WORD_SHIFT;
848 wordp = &(bip->bli_format.blf_data_map[word_num]);
849
850 /*
851 * Calculate the starting bit in the first word.
852 */
853 bit = first_bit & (uint)(NBWORD - 1);
854
855 /*
856 * First set any bits in the first word of our range.
857 * If it starts at bit 0 of the word, it will be
858 * set below rather than here. That is what the variable
859 * bit tells us. The variable bits_set tracks the number
860 * of bits that have been set so far. End_bit is the number
861 * of the last bit to be set in this word plus one.
862 */
863 if (bit) {
864 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
865 mask = ((1 << (end_bit - bit)) - 1) << bit;
866 *wordp |= mask;
867 wordp++;
868 bits_set = end_bit - bit;
869 } else {
870 bits_set = 0;
871 }
872
873 /*
874 * Now set bits a whole word at a time that are between
875 * first_bit and last_bit.
876 */
877 while ((bits_to_set - bits_set) >= NBWORD) {
878 *wordp |= 0xffffffff;
879 bits_set += NBWORD;
880 wordp++;
881 }
882
883 /*
884 * Finally, set any bits left to be set in one last partial word.
885 */
886 end_bit = bits_to_set - bits_set;
887 if (end_bit) {
888 mask = (1 << end_bit) - 1;
889 *wordp |= mask;
890 }
891
892 xfs_buf_item_log_debug(bip, first, last);
893}
894
895
896/*
897 * Return 1 if the buffer has some data that has been logged (at any
898 * point, not just the current transaction) and 0 if not.
899 */
900uint
901xfs_buf_item_dirty(
902 xfs_buf_log_item_t *bip)
903{
904 return (bip->bli_flags & XFS_BLI_DIRTY);
905}
906
907/*
908 * This is called when the buf log item is no longer needed. It should
909 * free the buf log item associated with the given buffer and clear
910 * the buffer's pointer to the buf log item. If there are no more
911 * items in the list, clear the b_iodone field of the buffer (see
912 * xfs_buf_attach_iodone() below).
913 */
914void
915xfs_buf_item_relse(
916 xfs_buf_t *bp)
917{
918 xfs_buf_log_item_t *bip;
919
920 xfs_buftrace("XFS_RELSE", bp);
921 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
922 XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list);
923 if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) &&
924 (XFS_BUF_IODONE_FUNC(bp) != NULL)) {
925 ASSERT((XFS_BUF_ISUNINITIAL(bp)) == 0);
926 XFS_BUF_CLR_IODONE_FUNC(bp);
927 }
928
929#ifdef XFS_TRANS_DEBUG
930 kmem_free(bip->bli_orig, XFS_BUF_COUNT(bp));
931 bip->bli_orig = NULL;
932 kmem_free(bip->bli_logged, XFS_BUF_COUNT(bp) / NBBY);
933 bip->bli_logged = NULL;
934#endif /* XFS_TRANS_DEBUG */
935
936#ifdef XFS_BLI_TRACE
937 ktrace_free(bip->bli_trace);
938#endif
939 kmem_zone_free(xfs_buf_item_zone, bip);
940}
941
942
943/*
944 * Add the given log item with its callback to the list of callbacks
945 * to be called when the buffer's I/O completes. If it is not set
946 * already, set the buffer's b_iodone() routine to be
947 * xfs_buf_iodone_callbacks() and link the log item into the list of
948 * items rooted at b_fsprivate. Items are always added as the second
949 * entry in the list if there is a first, because the buf item code
950 * assumes that the buf log item is first.
951 */
952void
953xfs_buf_attach_iodone(
954 xfs_buf_t *bp,
955 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
956 xfs_log_item_t *lip)
957{
958 xfs_log_item_t *head_lip;
959
960 ASSERT(XFS_BUF_ISBUSY(bp));
961 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
962
963 lip->li_cb = cb;
964 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
965 head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
966 lip->li_bio_list = head_lip->li_bio_list;
967 head_lip->li_bio_list = lip;
968 } else {
969 XFS_BUF_SET_FSPRIVATE(bp, lip);
970 }
971
972 ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) ||
973 (XFS_BUF_IODONE_FUNC(bp) == NULL));
974 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
975}
976
977STATIC void
978xfs_buf_do_callbacks(
979 xfs_buf_t *bp,
980 xfs_log_item_t *lip)
981{
982 xfs_log_item_t *nlip;
983
984 while (lip != NULL) {
985 nlip = lip->li_bio_list;
986 ASSERT(lip->li_cb != NULL);
987 /*
988 * Clear the next pointer so we don't have any
989 * confusion if the item is added to another buf.
990 * Don't touch the log item after calling its
991 * callback, because it could have freed itself.
992 */
993 lip->li_bio_list = NULL;
994 lip->li_cb(bp, lip);
995 lip = nlip;
996 }
997}
998
999/*
1000 * This is the iodone() function for buffers which have had callbacks
1001 * attached to them by xfs_buf_attach_iodone(). It should remove each
1002 * log item from the buffer's list and call the callback of each in turn.
1003 * When done, the buffer's fsprivate field is set to NULL and the buffer
1004 * is unlocked with a call to iodone().
1005 */
1006void
1007xfs_buf_iodone_callbacks(
1008 xfs_buf_t *bp)
1009{
1010 xfs_log_item_t *lip;
1011 static ulong lasttime;
1012 static xfs_buftarg_t *lasttarg;
1013 xfs_mount_t *mp;
1014
1015 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
1016 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1017
1018 if (XFS_BUF_GETERROR(bp) != 0) {
1019 /*
1020 * If we've already decided to shutdown the filesystem
1021 * because of IO errors, there's no point in giving this
1022 * a retry.
1023 */
1024 mp = lip->li_mountp;
1025 if (XFS_FORCED_SHUTDOWN(mp)) {
1026 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
1027 XFS_BUF_SUPER_STALE(bp);
1028 xfs_buftrace("BUF_IODONE_CB", bp);
1029 xfs_buf_do_callbacks(bp, lip);
1030 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1031 XFS_BUF_CLR_IODONE_FUNC(bp);
1032
1033 /*
1034 * XFS_SHUT flag gets set when we go thru the
1035 * entire buffer cache and deliberately start
1036 * throwing away delayed write buffers.
1037 * Since there's no biowait done on those,
1038 * we should just brelse them.
1039 */
1040 if (XFS_BUF_ISSHUT(bp)) {
1041 XFS_BUF_UNSHUT(bp);
1042 xfs_buf_relse(bp);
1043 } else {
1044 xfs_biodone(bp);
1045 }
1046
1047 return;
1048 }
1049
1050 if ((XFS_BUF_TARGET(bp) != lasttarg) ||
1051 (time_after(jiffies, (lasttime + 5*HZ)))) {
1052 lasttime = jiffies;
1053 prdev("XFS write error in file system meta-data "
1054 "block 0x%llx in %s",
1055 XFS_BUF_TARGET(bp),
1056 (__uint64_t)XFS_BUF_ADDR(bp), mp->m_fsname);
1057 }
1058 lasttarg = XFS_BUF_TARGET(bp);
1059
1060 if (XFS_BUF_ISASYNC(bp)) {
1061 /*
1062 * If the write was asynchronous then noone will be
1063 * looking for the error. Clear the error state
1064 * and write the buffer out again delayed write.
1065 *
1066 * XXXsup This is OK, so long as we catch these
1067 * before we start the umount; we don't want these
1068 * DELWRI metadata bufs to be hanging around.
1069 */
1070 XFS_BUF_ERROR(bp,0); /* errno of 0 unsets the flag */
1071
1072 if (!(XFS_BUF_ISSTALE(bp))) {
1073 XFS_BUF_DELAYWRITE(bp);
1074 XFS_BUF_DONE(bp);
1075 XFS_BUF_SET_START(bp);
1076 }
1077 ASSERT(XFS_BUF_IODONE_FUNC(bp));
1078 xfs_buftrace("BUF_IODONE ASYNC", bp);
1079 xfs_buf_relse(bp);
1080 } else {
1081 /*
1082 * If the write of the buffer was not asynchronous,
1083 * then we want to make sure to return the error
1084 * to the caller of bwrite(). Because of this we
1085 * cannot clear the B_ERROR state at this point.
1086 * Instead we install a callback function that
1087 * will be called when the buffer is released, and
1088 * that routine will clear the error state and
1089 * set the buffer to be written out again after
1090 * some delay.
1091 */
1092 /* We actually overwrite the existing b-relse
1093 function at times, but we're gonna be shutting down
1094 anyway. */
1095 XFS_BUF_SET_BRELSE_FUNC(bp,xfs_buf_error_relse);
1096 XFS_BUF_DONE(bp);
1097 XFS_BUF_V_IODONESEMA(bp);
1098 }
1099 return;
1100 }
1101#ifdef XFSERRORDEBUG
1102 xfs_buftrace("XFS BUFCB NOERR", bp);
1103#endif
1104 xfs_buf_do_callbacks(bp, lip);
1105 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1106 XFS_BUF_CLR_IODONE_FUNC(bp);
1107 xfs_biodone(bp);
1108}
1109
1110/*
1111 * This is a callback routine attached to a buffer which gets an error
1112 * when being written out synchronously.
1113 */
1114STATIC void
1115xfs_buf_error_relse(
1116 xfs_buf_t *bp)
1117{
1118 xfs_log_item_t *lip;
1119 xfs_mount_t *mp;
1120
1121 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1122 mp = (xfs_mount_t *)lip->li_mountp;
1123 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
1124
1125 XFS_BUF_STALE(bp);
1126 XFS_BUF_DONE(bp);
1127 XFS_BUF_UNDELAYWRITE(bp);
1128 XFS_BUF_ERROR(bp,0);
1129 xfs_buftrace("BUF_ERROR_RELSE", bp);
1130 if (! XFS_FORCED_SHUTDOWN(mp))
1131 xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
1132 /*
1133 * We have to unpin the pinned buffers so do the
1134 * callbacks.
1135 */
1136 xfs_buf_do_callbacks(bp, lip);
1137 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1138 XFS_BUF_CLR_IODONE_FUNC(bp);
1139 XFS_BUF_SET_BRELSE_FUNC(bp,NULL);
1140 xfs_buf_relse(bp);
1141}
1142
1143
1144/*
1145 * This is the iodone() function for buffers which have been
1146 * logged. It is called when they are eventually flushed out.
1147 * It should remove the buf item from the AIL, and free the buf item.
1148 * It is called by xfs_buf_iodone_callbacks() above which will take
1149 * care of cleaning up the buffer itself.
1150 */
1151/* ARGSUSED */
1152void
1153xfs_buf_iodone(
1154 xfs_buf_t *bp,
1155 xfs_buf_log_item_t *bip)
1156{
1157 struct xfs_mount *mp;
1158 SPLDECL(s);
1159
1160 ASSERT(bip->bli_buf == bp);
1161
1162 mp = bip->bli_item.li_mountp;
1163
1164 /*
1165 * If we are forcibly shutting down, this may well be
1166 * off the AIL already. That's because we simulate the
1167 * log-committed callbacks to unpin these buffers. Or we may never
1168 * have put this item on AIL because of the transaction was
1169 * aborted forcibly. xfs_trans_delete_ail() takes care of these.
1170 *
1171 * Either way, AIL is useless if we're forcing a shutdown.
1172 */
1173 AIL_LOCK(mp,s);
1174 /*
1175 * xfs_trans_delete_ail() drops the AIL lock.
1176 */
1177 xfs_trans_delete_ail(mp, (xfs_log_item_t *)bip, s);
1178
1179#ifdef XFS_TRANS_DEBUG
1180 kmem_free(bip->bli_orig, XFS_BUF_COUNT(bp));
1181 bip->bli_orig = NULL;
1182 kmem_free(bip->bli_logged, XFS_BUF_COUNT(bp) / NBBY);
1183 bip->bli_logged = NULL;
1184#endif /* XFS_TRANS_DEBUG */
1185
1186#ifdef XFS_BLI_TRACE
1187 ktrace_free(bip->bli_trace);
1188#endif
1189 kmem_zone_free(xfs_buf_item_zone, bip);
1190}
1191
1192#if defined(XFS_BLI_TRACE)
1193void
1194xfs_buf_item_trace(
1195 char *id,
1196 xfs_buf_log_item_t *bip)
1197{
1198 xfs_buf_t *bp;
1199 ASSERT(bip->bli_trace != NULL);
1200
1201 bp = bip->bli_buf;
1202 ktrace_enter(bip->bli_trace,
1203 (void *)id,
1204 (void *)bip->bli_buf,
1205 (void *)((unsigned long)bip->bli_flags),
1206 (void *)((unsigned long)bip->bli_recur),
1207 (void *)((unsigned long)atomic_read(&bip->bli_refcount)),
1208 (void *)((unsigned long)
1209 (0xFFFFFFFF & XFS_BUF_ADDR(bp) >> 32)),
1210 (void *)((unsigned long)(0xFFFFFFFF & XFS_BUF_ADDR(bp))),
1211 (void *)((unsigned long)XFS_BUF_COUNT(bp)),
1212 (void *)((unsigned long)XFS_BUF_BFLAGS(bp)),
1213 XFS_BUF_FSPRIVATE(bp, void *),
1214 XFS_BUF_FSPRIVATE2(bp, void *),
1215 (void *)(unsigned long)XFS_BUF_ISPINNED(bp),
1216 (void *)XFS_BUF_IODONE_FUNC(bp),
1217 (void *)((unsigned long)(XFS_BUF_VALUSEMA(bp))),
1218 (void *)bip->bli_item.li_desc,
1219 (void *)((unsigned long)bip->bli_item.li_flags));
1220}
1221#endif /* XFS_BLI_TRACE */
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-15 01:56:27 -0500