blob: a30f7e9eb2b96df6ab35aeb349430d4ddcc3a77a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 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"
David Chinnerda353b02007-08-28 14:00:13 +100026#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_dmapi.h"
28#include "xfs_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110029#include "xfs_buf_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_error.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000032#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033
34
35kmem_zone_t *xfs_buf_item_zone;
36
37#ifdef XFS_TRANS_DEBUG
38/*
39 * This function uses an alternate strategy for tracking the bytes
40 * that the user requests to be logged. This can then be used
41 * in conjunction with the bli_orig array in the buf log item to
42 * catch bugs in our callers' code.
43 *
44 * We also double check the bits set in xfs_buf_item_log using a
45 * simple algorithm to check that every byte is accounted for.
46 */
47STATIC void
48xfs_buf_item_log_debug(
49 xfs_buf_log_item_t *bip,
50 uint first,
51 uint last)
52{
53 uint x;
54 uint byte;
55 uint nbytes;
56 uint chunk_num;
57 uint word_num;
58 uint bit_num;
59 uint bit_set;
60 uint *wordp;
61
62 ASSERT(bip->bli_logged != NULL);
63 byte = first;
64 nbytes = last - first + 1;
65 bfset(bip->bli_logged, first, nbytes);
66 for (x = 0; x < nbytes; x++) {
67 chunk_num = byte >> XFS_BLI_SHIFT;
68 word_num = chunk_num >> BIT_TO_WORD_SHIFT;
69 bit_num = chunk_num & (NBWORD - 1);
70 wordp = &(bip->bli_format.blf_data_map[word_num]);
71 bit_set = *wordp & (1 << bit_num);
72 ASSERT(bit_set);
73 byte++;
74 }
75}
76
77/*
78 * This function is called when we flush something into a buffer without
79 * logging it. This happens for things like inodes which are logged
80 * separately from the buffer.
81 */
82void
83xfs_buf_item_flush_log_debug(
84 xfs_buf_t *bp,
85 uint first,
86 uint last)
87{
88 xfs_buf_log_item_t *bip;
89 uint nbytes;
90
91 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
92 if ((bip == NULL) || (bip->bli_item.li_type != XFS_LI_BUF)) {
93 return;
94 }
95
96 ASSERT(bip->bli_logged != NULL);
97 nbytes = last - first + 1;
98 bfset(bip->bli_logged, first, nbytes);
99}
100
101/*
Nathan Scottc41564b2006-03-29 08:55:14 +1000102 * This function is called to verify that our callers have logged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 * all the bytes that they changed.
104 *
105 * It does this by comparing the original copy of the buffer stored in
106 * the buf log item's bli_orig array to the current copy of the buffer
Nathan Scottc41564b2006-03-29 08:55:14 +1000107 * and ensuring that all bytes which mismatch are set in the bli_logged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * array of the buf log item.
109 */
110STATIC void
111xfs_buf_item_log_check(
112 xfs_buf_log_item_t *bip)
113{
114 char *orig;
115 char *buffer;
116 int x;
117 xfs_buf_t *bp;
118
119 ASSERT(bip->bli_orig != NULL);
120 ASSERT(bip->bli_logged != NULL);
121
122 bp = bip->bli_buf;
123 ASSERT(XFS_BUF_COUNT(bp) > 0);
124 ASSERT(XFS_BUF_PTR(bp) != NULL);
125 orig = bip->bli_orig;
126 buffer = XFS_BUF_PTR(bp);
127 for (x = 0; x < XFS_BUF_COUNT(bp); x++) {
128 if (orig[x] != buffer[x] && !btst(bip->bli_logged, x))
129 cmn_err(CE_PANIC,
130 "xfs_buf_item_log_check bip %x buffer %x orig %x index %d",
131 bip, bp, orig, x);
132 }
133}
134#else
135#define xfs_buf_item_log_debug(x,y,z)
136#define xfs_buf_item_log_check(x)
137#endif
138
139STATIC void xfs_buf_error_relse(xfs_buf_t *bp);
140STATIC void xfs_buf_do_callbacks(xfs_buf_t *bp, xfs_log_item_t *lip);
141
142/*
143 * This returns the number of log iovecs needed to log the
144 * given buf log item.
145 *
146 * It calculates this as 1 iovec for the buf log format structure
147 * and 1 for each stretch of non-contiguous chunks to be logged.
148 * Contiguous chunks are logged in a single iovec.
149 *
150 * If the XFS_BLI_STALE flag has been set, then log nothing.
151 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000152STATIC uint
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153xfs_buf_item_size(
154 xfs_buf_log_item_t *bip)
155{
156 uint nvecs;
157 int next_bit;
158 int last_bit;
159 xfs_buf_t *bp;
160
161 ASSERT(atomic_read(&bip->bli_refcount) > 0);
162 if (bip->bli_flags & XFS_BLI_STALE) {
163 /*
164 * The buffer is stale, so all we need to log
165 * is the buf log format structure with the
166 * cancel flag in it.
167 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000168 trace_xfs_buf_item_size_stale(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
170 return 1;
171 }
172
173 bp = bip->bli_buf;
174 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
175 nvecs = 1;
176 last_bit = xfs_next_bit(bip->bli_format.blf_data_map,
177 bip->bli_format.blf_map_size, 0);
178 ASSERT(last_bit != -1);
179 nvecs++;
180 while (last_bit != -1) {
181 /*
182 * This takes the bit number to start looking from and
183 * returns the next set bit from there. It returns -1
184 * if there are no more bits set or the start bit is
185 * beyond the end of the bitmap.
186 */
187 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
188 bip->bli_format.blf_map_size,
189 last_bit + 1);
190 /*
191 * If we run out of bits, leave the loop,
192 * else if we find a new set of bits bump the number of vecs,
193 * else keep scanning the current set of bits.
194 */
195 if (next_bit == -1) {
196 last_bit = -1;
197 } else if (next_bit != last_bit + 1) {
198 last_bit = next_bit;
199 nvecs++;
200 } else if (xfs_buf_offset(bp, next_bit * XFS_BLI_CHUNK) !=
201 (xfs_buf_offset(bp, last_bit * XFS_BLI_CHUNK) +
202 XFS_BLI_CHUNK)) {
203 last_bit = next_bit;
204 nvecs++;
205 } else {
206 last_bit++;
207 }
208 }
209
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000210 trace_xfs_buf_item_size(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700211 return nvecs;
212}
213
214/*
215 * This is called to fill in the vector of log iovecs for the
216 * given log buf item. It fills the first entry with a buf log
217 * format structure, and the rest point to contiguous chunks
218 * within the buffer.
219 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000220STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221xfs_buf_item_format(
222 xfs_buf_log_item_t *bip,
223 xfs_log_iovec_t *log_vector)
224{
225 uint base_size;
226 uint nvecs;
227 xfs_log_iovec_t *vecp;
228 xfs_buf_t *bp;
229 int first_bit;
230 int last_bit;
231 int next_bit;
232 uint nbits;
233 uint buffer_offset;
234
235 ASSERT(atomic_read(&bip->bli_refcount) > 0);
236 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
237 (bip->bli_flags & XFS_BLI_STALE));
238 bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 vecp = log_vector;
240
241 /*
242 * The size of the base structure is the size of the
243 * declared structure plus the space for the extra words
244 * of the bitmap. We subtract one from the map size, because
245 * the first element of the bitmap is accounted for in the
246 * size of the base structure.
247 */
248 base_size =
249 (uint)(sizeof(xfs_buf_log_format_t) +
250 ((bip->bli_format.blf_map_size - 1) * sizeof(uint)));
251 vecp->i_addr = (xfs_caddr_t)&bip->bli_format;
252 vecp->i_len = base_size;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000253 XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_BFORMAT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 vecp++;
255 nvecs = 1;
256
257 if (bip->bli_flags & XFS_BLI_STALE) {
258 /*
259 * The buffer is stale, so all we need to log
260 * is the buf log format structure with the
261 * cancel flag in it.
262 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000263 trace_xfs_buf_item_format_stale(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
265 bip->bli_format.blf_size = nvecs;
266 return;
267 }
268
269 /*
270 * Fill in an iovec for each set of contiguous chunks.
271 */
272 first_bit = xfs_next_bit(bip->bli_format.blf_data_map,
273 bip->bli_format.blf_map_size, 0);
274 ASSERT(first_bit != -1);
275 last_bit = first_bit;
276 nbits = 1;
277 for (;;) {
278 /*
279 * This takes the bit number to start looking from and
280 * returns the next set bit from there. It returns -1
281 * if there are no more bits set or the start bit is
282 * beyond the end of the bitmap.
283 */
284 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
285 bip->bli_format.blf_map_size,
286 (uint)last_bit + 1);
287 /*
288 * If we run out of bits fill in the last iovec and get
289 * out of the loop.
290 * Else if we start a new set of bits then fill in the
291 * iovec for the series we were looking at and start
292 * counting the bits in the new one.
293 * Else we're still in the same set of bits so just
294 * keep counting and scanning.
295 */
296 if (next_bit == -1) {
297 buffer_offset = first_bit * XFS_BLI_CHUNK;
298 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
299 vecp->i_len = nbits * XFS_BLI_CHUNK;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000300 XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_BCHUNK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301 nvecs++;
302 break;
303 } else if (next_bit != last_bit + 1) {
304 buffer_offset = first_bit * XFS_BLI_CHUNK;
305 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
306 vecp->i_len = nbits * XFS_BLI_CHUNK;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000307 XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_BCHUNK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308 nvecs++;
309 vecp++;
310 first_bit = next_bit;
311 last_bit = next_bit;
312 nbits = 1;
313 } else if (xfs_buf_offset(bp, next_bit << XFS_BLI_SHIFT) !=
314 (xfs_buf_offset(bp, last_bit << XFS_BLI_SHIFT) +
315 XFS_BLI_CHUNK)) {
316 buffer_offset = first_bit * XFS_BLI_CHUNK;
317 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
318 vecp->i_len = nbits * XFS_BLI_CHUNK;
Tim Shimmin7e9c6392005-09-02 16:42:05 +1000319 XLOG_VEC_SET_TYPE(vecp, XLOG_REG_TYPE_BCHUNK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320/* You would think we need to bump the nvecs here too, but we do not
321 * this number is used by recovery, and it gets confused by the boundary
322 * split here
323 * nvecs++;
324 */
325 vecp++;
326 first_bit = next_bit;
327 last_bit = next_bit;
328 nbits = 1;
329 } else {
330 last_bit++;
331 nbits++;
332 }
333 }
334 bip->bli_format.blf_size = nvecs;
335
336 /*
337 * Check to make sure everything is consistent.
338 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000339 trace_xfs_buf_item_format(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 xfs_buf_item_log_check(bip);
341}
342
343/*
344 * This is called to pin the buffer associated with the buf log
345 * item in memory so it cannot be written out. Simply call bpin()
346 * on the buffer to do this.
347 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000348STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349xfs_buf_item_pin(
350 xfs_buf_log_item_t *bip)
351{
352 xfs_buf_t *bp;
353
354 bp = bip->bli_buf;
355 ASSERT(XFS_BUF_ISBUSY(bp));
356 ASSERT(atomic_read(&bip->bli_refcount) > 0);
357 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
358 (bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000359 trace_xfs_buf_item_pin(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 xfs_bpin(bp);
361}
362
363
364/*
365 * This is called to unpin the buffer associated with the buf log
366 * item which was previously pinned with a call to xfs_buf_item_pin().
367 * Just call bunpin() on the buffer to do this.
368 *
369 * Also drop the reference to the buf item for the current transaction.
370 * If the XFS_BLI_STALE flag is set and we are the last reference,
371 * then free up the buf log item and unlock the buffer.
372 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000373STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374xfs_buf_item_unpin(
375 xfs_buf_log_item_t *bip,
376 int stale)
377{
David Chinner783a2f62008-10-30 17:39:58 +1100378 struct xfs_ail *ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 xfs_buf_t *bp;
380 int freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
382 bp = bip->bli_buf;
383 ASSERT(bp != NULL);
384 ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip);
385 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000386 trace_xfs_buf_item_unpin(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387
388 freed = atomic_dec_and_test(&bip->bli_refcount);
David Chinner783a2f62008-10-30 17:39:58 +1100389 ailp = bip->bli_item.li_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 xfs_bunpin(bp);
391 if (freed && stale) {
392 ASSERT(bip->bli_flags & XFS_BLI_STALE);
393 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
394 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
395 ASSERT(XFS_BUF_ISSTALE(bp));
396 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000397 trace_xfs_buf_item_unpin_stale(bip);
398
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 /*
400 * If we get called here because of an IO error, we may
David Chinner783a2f62008-10-30 17:39:58 +1100401 * or may not have the item on the AIL. xfs_trans_ail_delete()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402 * will take care of that situation.
David Chinner783a2f62008-10-30 17:39:58 +1100403 * xfs_trans_ail_delete() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 */
405 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
406 xfs_buf_do_callbacks(bp, (xfs_log_item_t *)bip);
407 XFS_BUF_SET_FSPRIVATE(bp, NULL);
408 XFS_BUF_CLR_IODONE_FUNC(bp);
409 } else {
David Chinner783a2f62008-10-30 17:39:58 +1100410 spin_lock(&ailp->xa_lock);
411 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 xfs_buf_item_relse(bp);
413 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL);
414 }
415 xfs_buf_relse(bp);
416 }
417}
418
419/*
420 * this is called from uncommit in the forced-shutdown path.
421 * we need to check to see if the reference count on the log item
422 * is going to drop to zero. If so, unpin will free the log item
423 * so we need to free the item's descriptor (that points to the item)
424 * in the transaction.
425 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000426STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427xfs_buf_item_unpin_remove(
428 xfs_buf_log_item_t *bip,
429 xfs_trans_t *tp)
430{
431 xfs_buf_t *bp;
432 xfs_log_item_desc_t *lidp;
433 int stale = 0;
434
435 bp = bip->bli_buf;
436 /*
437 * will xfs_buf_item_unpin() call xfs_buf_item_relse()?
438 */
439 if ((atomic_read(&bip->bli_refcount) == 1) &&
440 (bip->bli_flags & XFS_BLI_STALE)) {
441 ASSERT(XFS_BUF_VALUSEMA(bip->bli_buf) <= 0);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000442 trace_xfs_buf_item_unpin_stale(bip);
443
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 /*
445 * yes -- clear the xaction descriptor in-use flag
446 * and free the chunk if required. We can safely
447 * do some work here and then call buf_item_unpin
448 * to do the rest because if the if is true, then
449 * we are holding the buffer locked so no one else
450 * will be able to bump up the refcount.
451 */
452 lidp = xfs_trans_find_item(tp, (xfs_log_item_t *) bip);
453 stale = lidp->lid_flags & XFS_LID_BUF_STALE;
454 xfs_trans_free_item(tp, lidp);
455 /*
456 * Since the transaction no longer refers to the buffer,
457 * the buffer should no longer refer to the transaction.
458 */
459 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
460 }
461
462 xfs_buf_item_unpin(bip, stale);
463
464 return;
465}
466
467/*
468 * This is called to attempt to lock the buffer associated with this
469 * buf log item. Don't sleep on the buffer lock. If we can't get
470 * the lock right away, return 0. If we can get the lock, pull the
471 * buffer from the free list, mark it busy, and return 1.
472 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000473STATIC uint
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474xfs_buf_item_trylock(
475 xfs_buf_log_item_t *bip)
476{
477 xfs_buf_t *bp;
478
479 bp = bip->bli_buf;
480
481 if (XFS_BUF_ISPINNED(bp)) {
482 return XFS_ITEM_PINNED;
483 }
484
485 if (!XFS_BUF_CPSEMA(bp)) {
486 return XFS_ITEM_LOCKED;
487 }
488
489 /*
490 * Remove the buffer from the free list. Only do this
491 * if it's on the free list. Private buffers like the
492 * superblock buffer are not.
493 */
494 XFS_BUF_HOLD(bp);
495
496 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000497 trace_xfs_buf_item_trylock(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 return XFS_ITEM_SUCCESS;
499}
500
501/*
502 * Release the buffer associated with the buf log item.
503 * If there is no dirty logged data associated with the
504 * buffer recorded in the buf log item, then free the
505 * buf log item and remove the reference to it in the
506 * buffer.
507 *
508 * This call ignores the recursion count. It is only called
509 * when the buffer should REALLY be unlocked, regardless
510 * of the recursion count.
511 *
512 * If the XFS_BLI_HOLD flag is set in the buf log item, then
513 * free the log item if necessary but do not unlock the buffer.
514 * This is for support of xfs_trans_bhold(). Make sure the
515 * XFS_BLI_HOLD field is cleared if we don't free the item.
516 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000517STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518xfs_buf_item_unlock(
519 xfs_buf_log_item_t *bip)
520{
521 int aborted;
522 xfs_buf_t *bp;
523 uint hold;
524
525 bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
527 /*
528 * Clear the buffer's association with this transaction.
529 */
530 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
531
532 /*
533 * If this is a transaction abort, don't return early.
534 * Instead, allow the brelse to happen.
535 * Normally it would be done for stale (cancelled) buffers
536 * at unpin time, but we'll never go through the pin/unpin
537 * cycle if we abort inside commit.
538 */
539 aborted = (bip->bli_item.li_flags & XFS_LI_ABORTED) != 0;
540
541 /*
542 * If the buf item is marked stale, then don't do anything.
543 * We'll unlock the buffer and free the buf item when the
544 * buffer is unpinned for the last time.
545 */
546 if (bip->bli_flags & XFS_BLI_STALE) {
547 bip->bli_flags &= ~XFS_BLI_LOGGED;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000548 trace_xfs_buf_item_unlock_stale(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 ASSERT(bip->bli_format.blf_flags & XFS_BLI_CANCEL);
550 if (!aborted)
551 return;
552 }
553
554 /*
555 * Drop the transaction's reference to the log item if
556 * it was not logged as part of the transaction. Otherwise
557 * we'll drop the reference in xfs_buf_item_unpin() when
558 * the transaction is really through with the buffer.
559 */
560 if (!(bip->bli_flags & XFS_BLI_LOGGED)) {
561 atomic_dec(&bip->bli_refcount);
562 } else {
563 /*
564 * Clear the logged flag since this is per
565 * transaction state.
566 */
567 bip->bli_flags &= ~XFS_BLI_LOGGED;
568 }
569
570 /*
571 * Before possibly freeing the buf item, determine if we should
572 * release the buffer at the end of this routine.
573 */
574 hold = bip->bli_flags & XFS_BLI_HOLD;
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000575 trace_xfs_buf_item_unlock(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576
577 /*
578 * If the buf item isn't tracking any data, free it.
579 * Otherwise, if XFS_BLI_HOLD is set clear it.
580 */
Eric Sandeen24ad33f2007-06-28 16:43:30 +1000581 if (xfs_bitmap_empty(bip->bli_format.blf_data_map,
582 bip->bli_format.blf_map_size)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 xfs_buf_item_relse(bp);
584 } else if (hold) {
585 bip->bli_flags &= ~XFS_BLI_HOLD;
586 }
587
588 /*
589 * Release the buffer if XFS_BLI_HOLD was not set.
590 */
591 if (!hold) {
592 xfs_buf_relse(bp);
593 }
594}
595
596/*
597 * This is called to find out where the oldest active copy of the
598 * buf log item in the on disk log resides now that the last log
599 * write of it completed at the given lsn.
600 * We always re-log all the dirty data in a buffer, so usually the
601 * latest copy in the on disk log is the only one that matters. For
602 * those cases we simply return the given lsn.
603 *
604 * The one exception to this is for buffers full of newly allocated
605 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
606 * flag set, indicating that only the di_next_unlinked fields from the
607 * inodes in the buffers will be replayed during recovery. If the
608 * original newly allocated inode images have not yet been flushed
609 * when the buffer is so relogged, then we need to make sure that we
610 * keep the old images in the 'active' portion of the log. We do this
611 * by returning the original lsn of that transaction here rather than
612 * the current one.
613 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000614STATIC xfs_lsn_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615xfs_buf_item_committed(
616 xfs_buf_log_item_t *bip,
617 xfs_lsn_t lsn)
618{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000619 trace_xfs_buf_item_committed(bip);
620
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
622 (bip->bli_item.li_lsn != 0)) {
623 return bip->bli_item.li_lsn;
624 }
625 return (lsn);
626}
627
628/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 * This is called to asynchronously write the buffer associated with this
630 * buf log item out to disk. The buffer will already have been locked by
631 * a successful call to xfs_buf_item_trylock(). If the buffer still has
632 * B_DELWRI set, then get it going out to disk with a call to bawrite().
633 * If not, then just release the buffer.
634 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000635STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636xfs_buf_item_push(
637 xfs_buf_log_item_t *bip)
638{
639 xfs_buf_t *bp;
640
641 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000642 trace_xfs_buf_item_push(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643
644 bp = bip->bli_buf;
645
646 if (XFS_BUF_ISDELAYWRITE(bp)) {
David Chinnerdb7a19f2008-04-10 12:22:24 +1000647 int error;
648 error = xfs_bawrite(bip->bli_item.li_mountp, bp);
649 if (error)
650 xfs_fs_cmn_err(CE_WARN, bip->bli_item.li_mountp,
651 "xfs_buf_item_push: pushbuf error %d on bip %p, bp %p",
652 error, bip, bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 } else {
654 xfs_buf_relse(bp);
655 }
656}
657
658/* ARGSUSED */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000659STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660xfs_buf_item_committing(xfs_buf_log_item_t *bip, xfs_lsn_t commit_lsn)
661{
662}
663
664/*
665 * This is the ops vector shared by all buf log items.
666 */
David Chinner7989cb82007-02-10 18:34:56 +1100667static struct xfs_item_ops xfs_buf_item_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 .iop_size = (uint(*)(xfs_log_item_t*))xfs_buf_item_size,
669 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
670 xfs_buf_item_format,
671 .iop_pin = (void(*)(xfs_log_item_t*))xfs_buf_item_pin,
672 .iop_unpin = (void(*)(xfs_log_item_t*, int))xfs_buf_item_unpin,
673 .iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t *))
674 xfs_buf_item_unpin_remove,
675 .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_buf_item_trylock,
676 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_buf_item_unlock,
677 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
678 xfs_buf_item_committed,
679 .iop_push = (void(*)(xfs_log_item_t*))xfs_buf_item_push,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 .iop_pushbuf = NULL,
681 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
682 xfs_buf_item_committing
683};
684
685
686/*
687 * Allocate a new buf log item to go with the given buffer.
688 * Set the buffer's b_fsprivate field to point to the new
689 * buf log item. If there are other item's attached to the
690 * buffer (see xfs_buf_attach_iodone() below), then put the
691 * buf log item at the front.
692 */
693void
694xfs_buf_item_init(
695 xfs_buf_t *bp,
696 xfs_mount_t *mp)
697{
698 xfs_log_item_t *lip;
699 xfs_buf_log_item_t *bip;
700 int chunks;
701 int map_size;
702
703 /*
704 * Check to see if there is already a buf log item for
705 * this buffer. If there is, it is guaranteed to be
706 * the first. If we do already have one, there is
707 * nothing to do here so return.
708 */
Christoph Hellwig15ac08a2008-12-09 04:47:30 -0500709 if (bp->b_mount != mp)
710 bp->b_mount = mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 XFS_BUF_SET_BDSTRAT_FUNC(bp, xfs_bdstrat_cb);
712 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
713 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
714 if (lip->li_type == XFS_LI_BUF) {
715 return;
716 }
717 }
718
719 /*
720 * chunks is the number of XFS_BLI_CHUNK size pieces
721 * the buffer can be divided into. Make sure not to
722 * truncate any pieces. map_size is the size of the
723 * bitmap needed to describe the chunks of the buffer.
724 */
725 chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLI_CHUNK - 1)) >> XFS_BLI_SHIFT);
726 map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);
727
728 bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
729 KM_SLEEP);
730 bip->bli_item.li_type = XFS_LI_BUF;
731 bip->bli_item.li_ops = &xfs_buf_item_ops;
732 bip->bli_item.li_mountp = mp;
David Chinnerfc1829f2008-10-30 17:39:46 +1100733 bip->bli_item.li_ailp = mp->m_ail;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 bip->bli_buf = bp;
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000735 xfs_buf_hold(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 bip->bli_format.blf_type = XFS_LI_BUF;
737 bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp);
738 bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp));
739 bip->bli_format.blf_map_size = map_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740
741#ifdef XFS_TRANS_DEBUG
742 /*
743 * Allocate the arrays for tracking what needs to be logged
744 * and what our callers request to be logged. bli_orig
745 * holds a copy of the original, clean buffer for comparison
746 * against, and bli_logged keeps a 1 bit flag per byte in
747 * the buffer to indicate which bytes the callers have asked
748 * to have logged.
749 */
750 bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP);
751 memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp));
752 bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP);
753#endif
754
755 /*
756 * Put the buf item into the list of items attached to the
757 * buffer at the front.
758 */
759 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
760 bip->bli_item.li_bio_list =
761 XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
762 }
763 XFS_BUF_SET_FSPRIVATE(bp, bip);
764}
765
766
767/*
768 * Mark bytes first through last inclusive as dirty in the buf
769 * item's bitmap.
770 */
771void
772xfs_buf_item_log(
773 xfs_buf_log_item_t *bip,
774 uint first,
775 uint last)
776{
777 uint first_bit;
778 uint last_bit;
779 uint bits_to_set;
780 uint bits_set;
781 uint word_num;
782 uint *wordp;
783 uint bit;
784 uint end_bit;
785 uint mask;
786
787 /*
788 * Mark the item as having some dirty data for
789 * quick reference in xfs_buf_item_dirty.
790 */
791 bip->bli_flags |= XFS_BLI_DIRTY;
792
793 /*
794 * Convert byte offsets to bit numbers.
795 */
796 first_bit = first >> XFS_BLI_SHIFT;
797 last_bit = last >> XFS_BLI_SHIFT;
798
799 /*
800 * Calculate the total number of bits to be set.
801 */
802 bits_to_set = last_bit - first_bit + 1;
803
804 /*
805 * Get a pointer to the first word in the bitmap
806 * to set a bit in.
807 */
808 word_num = first_bit >> BIT_TO_WORD_SHIFT;
809 wordp = &(bip->bli_format.blf_data_map[word_num]);
810
811 /*
812 * Calculate the starting bit in the first word.
813 */
814 bit = first_bit & (uint)(NBWORD - 1);
815
816 /*
817 * First set any bits in the first word of our range.
818 * If it starts at bit 0 of the word, it will be
819 * set below rather than here. That is what the variable
820 * bit tells us. The variable bits_set tracks the number
821 * of bits that have been set so far. End_bit is the number
822 * of the last bit to be set in this word plus one.
823 */
824 if (bit) {
825 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
826 mask = ((1 << (end_bit - bit)) - 1) << bit;
827 *wordp |= mask;
828 wordp++;
829 bits_set = end_bit - bit;
830 } else {
831 bits_set = 0;
832 }
833
834 /*
835 * Now set bits a whole word at a time that are between
836 * first_bit and last_bit.
837 */
838 while ((bits_to_set - bits_set) >= NBWORD) {
839 *wordp |= 0xffffffff;
840 bits_set += NBWORD;
841 wordp++;
842 }
843
844 /*
845 * Finally, set any bits left to be set in one last partial word.
846 */
847 end_bit = bits_to_set - bits_set;
848 if (end_bit) {
849 mask = (1 << end_bit) - 1;
850 *wordp |= mask;
851 }
852
853 xfs_buf_item_log_debug(bip, first, last);
854}
855
856
857/*
858 * Return 1 if the buffer has some data that has been logged (at any
859 * point, not just the current transaction) and 0 if not.
860 */
861uint
862xfs_buf_item_dirty(
863 xfs_buf_log_item_t *bip)
864{
865 return (bip->bli_flags & XFS_BLI_DIRTY);
866}
867
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000868STATIC void
869xfs_buf_item_free(
870 xfs_buf_log_item_t *bip)
871{
872#ifdef XFS_TRANS_DEBUG
873 kmem_free(bip->bli_orig);
874 kmem_free(bip->bli_logged);
875#endif /* XFS_TRANS_DEBUG */
876
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000877 kmem_zone_free(xfs_buf_item_zone, bip);
878}
879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880/*
881 * This is called when the buf log item is no longer needed. It should
882 * free the buf log item associated with the given buffer and clear
883 * the buffer's pointer to the buf log item. If there are no more
884 * items in the list, clear the b_iodone field of the buffer (see
885 * xfs_buf_attach_iodone() below).
886 */
887void
888xfs_buf_item_relse(
889 xfs_buf_t *bp)
890{
891 xfs_buf_log_item_t *bip;
892
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000893 trace_xfs_buf_item_relse(bp, _RET_IP_);
894
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
896 XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list);
897 if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) &&
898 (XFS_BUF_IODONE_FUNC(bp) != NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899 XFS_BUF_CLR_IODONE_FUNC(bp);
900 }
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000901 xfs_buf_rele(bp);
902 xfs_buf_item_free(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903}
904
905
906/*
907 * Add the given log item with its callback to the list of callbacks
908 * to be called when the buffer's I/O completes. If it is not set
909 * already, set the buffer's b_iodone() routine to be
910 * xfs_buf_iodone_callbacks() and link the log item into the list of
911 * items rooted at b_fsprivate. Items are always added as the second
912 * entry in the list if there is a first, because the buf item code
913 * assumes that the buf log item is first.
914 */
915void
916xfs_buf_attach_iodone(
917 xfs_buf_t *bp,
918 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
919 xfs_log_item_t *lip)
920{
921 xfs_log_item_t *head_lip;
922
923 ASSERT(XFS_BUF_ISBUSY(bp));
924 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
925
926 lip->li_cb = cb;
927 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
928 head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
929 lip->li_bio_list = head_lip->li_bio_list;
930 head_lip->li_bio_list = lip;
931 } else {
932 XFS_BUF_SET_FSPRIVATE(bp, lip);
933 }
934
935 ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) ||
936 (XFS_BUF_IODONE_FUNC(bp) == NULL));
937 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
938}
939
940STATIC void
941xfs_buf_do_callbacks(
942 xfs_buf_t *bp,
943 xfs_log_item_t *lip)
944{
945 xfs_log_item_t *nlip;
946
947 while (lip != NULL) {
948 nlip = lip->li_bio_list;
949 ASSERT(lip->li_cb != NULL);
950 /*
951 * Clear the next pointer so we don't have any
952 * confusion if the item is added to another buf.
953 * Don't touch the log item after calling its
954 * callback, because it could have freed itself.
955 */
956 lip->li_bio_list = NULL;
957 lip->li_cb(bp, lip);
958 lip = nlip;
959 }
960}
961
962/*
963 * This is the iodone() function for buffers which have had callbacks
964 * attached to them by xfs_buf_attach_iodone(). It should remove each
965 * log item from the buffer's list and call the callback of each in turn.
966 * When done, the buffer's fsprivate field is set to NULL and the buffer
967 * is unlocked with a call to iodone().
968 */
969void
970xfs_buf_iodone_callbacks(
971 xfs_buf_t *bp)
972{
973 xfs_log_item_t *lip;
974 static ulong lasttime;
975 static xfs_buftarg_t *lasttarg;
976 xfs_mount_t *mp;
977
978 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
979 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
980
981 if (XFS_BUF_GETERROR(bp) != 0) {
982 /*
983 * If we've already decided to shutdown the filesystem
984 * because of IO errors, there's no point in giving this
985 * a retry.
986 */
987 mp = lip->li_mountp;
988 if (XFS_FORCED_SHUTDOWN(mp)) {
989 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
990 XFS_BUF_SUPER_STALE(bp);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000991 trace_xfs_buf_item_iodone(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 xfs_buf_do_callbacks(bp, lip);
993 XFS_BUF_SET_FSPRIVATE(bp, NULL);
994 XFS_BUF_CLR_IODONE_FUNC(bp);
Lachlan McIlroy4fdc7782008-12-22 17:52:58 +1100995 xfs_biodone(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 return;
997 }
998
999 if ((XFS_BUF_TARGET(bp) != lasttarg) ||
1000 (time_after(jiffies, (lasttime + 5*HZ)))) {
1001 lasttime = jiffies;
Nathan Scottb6574522006-06-09 15:29:40 +10001002 cmn_err(CE_ALERT, "Device %s, XFS metadata write error"
1003 " block 0x%llx in %s",
1004 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)),
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 (__uint64_t)XFS_BUF_ADDR(bp), mp->m_fsname);
1006 }
1007 lasttarg = XFS_BUF_TARGET(bp);
1008
1009 if (XFS_BUF_ISASYNC(bp)) {
1010 /*
1011 * If the write was asynchronous then noone will be
1012 * looking for the error. Clear the error state
1013 * and write the buffer out again delayed write.
1014 *
1015 * XXXsup This is OK, so long as we catch these
1016 * before we start the umount; we don't want these
1017 * DELWRI metadata bufs to be hanging around.
1018 */
1019 XFS_BUF_ERROR(bp,0); /* errno of 0 unsets the flag */
1020
1021 if (!(XFS_BUF_ISSTALE(bp))) {
1022 XFS_BUF_DELAYWRITE(bp);
1023 XFS_BUF_DONE(bp);
1024 XFS_BUF_SET_START(bp);
1025 }
1026 ASSERT(XFS_BUF_IODONE_FUNC(bp));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001027 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 xfs_buf_relse(bp);
1029 } else {
1030 /*
1031 * If the write of the buffer was not asynchronous,
1032 * then we want to make sure to return the error
1033 * to the caller of bwrite(). Because of this we
1034 * cannot clear the B_ERROR state at this point.
1035 * Instead we install a callback function that
1036 * will be called when the buffer is released, and
1037 * that routine will clear the error state and
1038 * set the buffer to be written out again after
1039 * some delay.
1040 */
1041 /* We actually overwrite the existing b-relse
1042 function at times, but we're gonna be shutting down
1043 anyway. */
1044 XFS_BUF_SET_BRELSE_FUNC(bp,xfs_buf_error_relse);
1045 XFS_BUF_DONE(bp);
David Chinnerb4dd3302008-08-13 16:36:11 +10001046 XFS_BUF_FINISH_IOWAIT(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 }
1048 return;
1049 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001050
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 xfs_buf_do_callbacks(bp, lip);
1052 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1053 XFS_BUF_CLR_IODONE_FUNC(bp);
1054 xfs_biodone(bp);
1055}
1056
1057/*
1058 * This is a callback routine attached to a buffer which gets an error
1059 * when being written out synchronously.
1060 */
1061STATIC void
1062xfs_buf_error_relse(
1063 xfs_buf_t *bp)
1064{
1065 xfs_log_item_t *lip;
1066 xfs_mount_t *mp;
1067
1068 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1069 mp = (xfs_mount_t *)lip->li_mountp;
1070 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
1071
1072 XFS_BUF_STALE(bp);
1073 XFS_BUF_DONE(bp);
1074 XFS_BUF_UNDELAYWRITE(bp);
1075 XFS_BUF_ERROR(bp,0);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001076
1077 trace_xfs_buf_error_relse(bp, _RET_IP_);
1078
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 if (! XFS_FORCED_SHUTDOWN(mp))
Nathan Scott7d04a332006-06-09 14:58:38 +10001080 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081 /*
1082 * We have to unpin the pinned buffers so do the
1083 * callbacks.
1084 */
1085 xfs_buf_do_callbacks(bp, lip);
1086 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1087 XFS_BUF_CLR_IODONE_FUNC(bp);
1088 XFS_BUF_SET_BRELSE_FUNC(bp,NULL);
1089 xfs_buf_relse(bp);
1090}
1091
1092
1093/*
1094 * This is the iodone() function for buffers which have been
1095 * logged. It is called when they are eventually flushed out.
1096 * It should remove the buf item from the AIL, and free the buf item.
1097 * It is called by xfs_buf_iodone_callbacks() above which will take
1098 * care of cleaning up the buffer itself.
1099 */
1100/* ARGSUSED */
1101void
1102xfs_buf_iodone(
1103 xfs_buf_t *bp,
1104 xfs_buf_log_item_t *bip)
1105{
David Chinner783a2f62008-10-30 17:39:58 +11001106 struct xfs_ail *ailp = bip->bli_item.li_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107
1108 ASSERT(bip->bli_buf == bp);
1109
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +10001110 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
1112 /*
1113 * If we are forcibly shutting down, this may well be
1114 * off the AIL already. That's because we simulate the
1115 * log-committed callbacks to unpin these buffers. Or we may never
1116 * have put this item on AIL because of the transaction was
David Chinner783a2f62008-10-30 17:39:58 +11001117 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 *
1119 * Either way, AIL is useless if we're forcing a shutdown.
1120 */
David Chinnerfc1829f2008-10-30 17:39:46 +11001121 spin_lock(&ailp->xa_lock);
David Chinner783a2f62008-10-30 17:39:58 +11001122 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip);
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +10001123 xfs_buf_item_free(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124}