blob: 93899953c603da3d7e424a0fb77546f12749bec0 [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_mount.h"
Nathan Scotta844f452005-11-02 14:38:42 +110028#include "xfs_buf_item.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_trans_priv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include "xfs_error.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000031#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032
33
34kmem_zone_t *xfs_buf_item_zone;
35
36#ifdef XFS_TRANS_DEBUG
37/*
38 * This function uses an alternate strategy for tracking the bytes
39 * that the user requests to be logged. This can then be used
40 * in conjunction with the bli_orig array in the buf log item to
41 * catch bugs in our callers' code.
42 *
43 * We also double check the bits set in xfs_buf_item_log using a
44 * simple algorithm to check that every byte is accounted for.
45 */
46STATIC void
47xfs_buf_item_log_debug(
48 xfs_buf_log_item_t *bip,
49 uint first,
50 uint last)
51{
52 uint x;
53 uint byte;
54 uint nbytes;
55 uint chunk_num;
56 uint word_num;
57 uint bit_num;
58 uint bit_set;
59 uint *wordp;
60
61 ASSERT(bip->bli_logged != NULL);
62 byte = first;
63 nbytes = last - first + 1;
64 bfset(bip->bli_logged, first, nbytes);
65 for (x = 0; x < nbytes; x++) {
Dave Chinnerc1155412010-05-07 11:05:19 +100066 chunk_num = byte >> XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 word_num = chunk_num >> BIT_TO_WORD_SHIFT;
68 bit_num = chunk_num & (NBWORD - 1);
69 wordp = &(bip->bli_format.blf_data_map[word_num]);
70 bit_set = *wordp & (1 << bit_num);
71 ASSERT(bit_set);
72 byte++;
73 }
74}
75
76/*
77 * This function is called when we flush something into a buffer without
78 * logging it. This happens for things like inodes which are logged
79 * separately from the buffer.
80 */
81void
82xfs_buf_item_flush_log_debug(
83 xfs_buf_t *bp,
84 uint first,
85 uint last)
86{
87 xfs_buf_log_item_t *bip;
88 uint nbytes;
89
90 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
91 if ((bip == NULL) || (bip->bli_item.li_type != XFS_LI_BUF)) {
92 return;
93 }
94
95 ASSERT(bip->bli_logged != NULL);
96 nbytes = last - first + 1;
97 bfset(bip->bli_logged, first, nbytes);
98}
99
100/*
Nathan Scottc41564b2006-03-29 08:55:14 +1000101 * This function is called to verify that our callers have logged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102 * all the bytes that they changed.
103 *
104 * It does this by comparing the original copy of the buffer stored in
105 * the buf log item's bli_orig array to the current copy of the buffer
Nathan Scottc41564b2006-03-29 08:55:14 +1000106 * and ensuring that all bytes which mismatch are set in the bli_logged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107 * array of the buf log item.
108 */
109STATIC void
110xfs_buf_item_log_check(
111 xfs_buf_log_item_t *bip)
112{
113 char *orig;
114 char *buffer;
115 int x;
116 xfs_buf_t *bp;
117
118 ASSERT(bip->bli_orig != NULL);
119 ASSERT(bip->bli_logged != NULL);
120
121 bp = bip->bli_buf;
122 ASSERT(XFS_BUF_COUNT(bp) > 0);
123 ASSERT(XFS_BUF_PTR(bp) != NULL);
124 orig = bip->bli_orig;
125 buffer = XFS_BUF_PTR(bp);
126 for (x = 0; x < XFS_BUF_COUNT(bp); x++) {
127 if (orig[x] != buffer[x] && !btst(bip->bli_logged, x))
128 cmn_err(CE_PANIC,
129 "xfs_buf_item_log_check bip %x buffer %x orig %x index %d",
130 bip, bp, orig, x);
131 }
132}
133#else
134#define xfs_buf_item_log_debug(x,y,z)
135#define xfs_buf_item_log_check(x)
136#endif
137
138STATIC void xfs_buf_error_relse(xfs_buf_t *bp);
139STATIC void xfs_buf_do_callbacks(xfs_buf_t *bp, xfs_log_item_t *lip);
140
141/*
142 * This returns the number of log iovecs needed to log the
143 * given buf log item.
144 *
145 * It calculates this as 1 iovec for the buf log format structure
146 * and 1 for each stretch of non-contiguous chunks to be logged.
147 * Contiguous chunks are logged in a single iovec.
148 *
149 * If the XFS_BLI_STALE flag has been set, then log nothing.
150 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000151STATIC uint
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152xfs_buf_item_size(
153 xfs_buf_log_item_t *bip)
154{
155 uint nvecs;
156 int next_bit;
157 int last_bit;
158 xfs_buf_t *bp;
159
160 ASSERT(atomic_read(&bip->bli_refcount) > 0);
161 if (bip->bli_flags & XFS_BLI_STALE) {
162 /*
163 * The buffer is stale, so all we need to log
164 * is the buf log format structure with the
165 * cancel flag in it.
166 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000167 trace_xfs_buf_item_size_stale(bip);
Dave Chinnerc1155412010-05-07 11:05:19 +1000168 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 return 1;
170 }
171
172 bp = bip->bli_buf;
173 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
174 nvecs = 1;
175 last_bit = xfs_next_bit(bip->bli_format.blf_data_map,
176 bip->bli_format.blf_map_size, 0);
177 ASSERT(last_bit != -1);
178 nvecs++;
179 while (last_bit != -1) {
180 /*
181 * This takes the bit number to start looking from and
182 * returns the next set bit from there. It returns -1
183 * if there are no more bits set or the start bit is
184 * beyond the end of the bitmap.
185 */
186 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
187 bip->bli_format.blf_map_size,
188 last_bit + 1);
189 /*
190 * If we run out of bits, leave the loop,
191 * else if we find a new set of bits bump the number of vecs,
192 * else keep scanning the current set of bits.
193 */
194 if (next_bit == -1) {
195 last_bit = -1;
196 } else if (next_bit != last_bit + 1) {
197 last_bit = next_bit;
198 nvecs++;
Dave Chinnerc1155412010-05-07 11:05:19 +1000199 } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
200 (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
201 XFS_BLF_CHUNK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 last_bit = next_bit;
203 nvecs++;
204 } else {
205 last_bit++;
206 }
207 }
208
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000209 trace_xfs_buf_item_size(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210 return nvecs;
211}
212
213/*
214 * This is called to fill in the vector of log iovecs for the
215 * given log buf item. It fills the first entry with a buf log
216 * format structure, and the rest point to contiguous chunks
217 * within the buffer.
218 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000219STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220xfs_buf_item_format(
221 xfs_buf_log_item_t *bip,
222 xfs_log_iovec_t *log_vector)
223{
224 uint base_size;
225 uint nvecs;
226 xfs_log_iovec_t *vecp;
227 xfs_buf_t *bp;
228 int first_bit;
229 int last_bit;
230 int next_bit;
231 uint nbits;
232 uint buffer_offset;
233
234 ASSERT(atomic_read(&bip->bli_refcount) > 0);
235 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
236 (bip->bli_flags & XFS_BLI_STALE));
237 bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238 vecp = log_vector;
239
240 /*
241 * The size of the base structure is the size of the
242 * declared structure plus the space for the extra words
243 * of the bitmap. We subtract one from the map size, because
244 * the first element of the bitmap is accounted for in the
245 * size of the base structure.
246 */
247 base_size =
248 (uint)(sizeof(xfs_buf_log_format_t) +
249 ((bip->bli_format.blf_map_size - 1) * sizeof(uint)));
250 vecp->i_addr = (xfs_caddr_t)&bip->bli_format;
251 vecp->i_len = base_size;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000252 vecp->i_type = XLOG_REG_TYPE_BFORMAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253 vecp++;
254 nvecs = 1;
255
Dave Chinnerccf7c232010-05-20 23:19:42 +1000256 /*
257 * If it is an inode buffer, transfer the in-memory state to the
258 * format flags and clear the in-memory state. We do not transfer
259 * this state if the inode buffer allocation has not yet been committed
260 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
261 * correct replay of the inode allocation.
262 */
263 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
264 if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
265 xfs_log_item_in_current_chkpt(&bip->bli_item)))
266 bip->bli_format.blf_flags |= XFS_BLF_INODE_BUF;
267 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
268 }
269
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 if (bip->bli_flags & XFS_BLI_STALE) {
271 /*
272 * The buffer is stale, so all we need to log
273 * is the buf log format structure with the
274 * cancel flag in it.
275 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000276 trace_xfs_buf_item_format_stale(bip);
Dave Chinnerc1155412010-05-07 11:05:19 +1000277 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700278 bip->bli_format.blf_size = nvecs;
279 return;
280 }
281
282 /*
283 * Fill in an iovec for each set of contiguous chunks.
284 */
285 first_bit = xfs_next_bit(bip->bli_format.blf_data_map,
286 bip->bli_format.blf_map_size, 0);
287 ASSERT(first_bit != -1);
288 last_bit = first_bit;
289 nbits = 1;
290 for (;;) {
291 /*
292 * This takes the bit number to start looking from and
293 * returns the next set bit from there. It returns -1
294 * if there are no more bits set or the start bit is
295 * beyond the end of the bitmap.
296 */
297 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
298 bip->bli_format.blf_map_size,
299 (uint)last_bit + 1);
300 /*
301 * If we run out of bits fill in the last iovec and get
302 * out of the loop.
303 * Else if we start a new set of bits then fill in the
304 * iovec for the series we were looking at and start
305 * counting the bits in the new one.
306 * Else we're still in the same set of bits so just
307 * keep counting and scanning.
308 */
309 if (next_bit == -1) {
Dave Chinnerc1155412010-05-07 11:05:19 +1000310 buffer_offset = first_bit * XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
Dave Chinnerc1155412010-05-07 11:05:19 +1000312 vecp->i_len = nbits * XFS_BLF_CHUNK;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000313 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 nvecs++;
315 break;
316 } else if (next_bit != last_bit + 1) {
Dave Chinnerc1155412010-05-07 11:05:19 +1000317 buffer_offset = first_bit * XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
Dave Chinnerc1155412010-05-07 11:05:19 +1000319 vecp->i_len = nbits * XFS_BLF_CHUNK;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000320 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 nvecs++;
322 vecp++;
323 first_bit = next_bit;
324 last_bit = next_bit;
325 nbits = 1;
Dave Chinnerc1155412010-05-07 11:05:19 +1000326 } else if (xfs_buf_offset(bp, next_bit << XFS_BLF_SHIFT) !=
327 (xfs_buf_offset(bp, last_bit << XFS_BLF_SHIFT) +
328 XFS_BLF_CHUNK)) {
329 buffer_offset = first_bit * XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
Dave Chinnerc1155412010-05-07 11:05:19 +1000331 vecp->i_len = nbits * XFS_BLF_CHUNK;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000332 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333/* You would think we need to bump the nvecs here too, but we do not
334 * this number is used by recovery, and it gets confused by the boundary
335 * split here
336 * nvecs++;
337 */
338 vecp++;
339 first_bit = next_bit;
340 last_bit = next_bit;
341 nbits = 1;
342 } else {
343 last_bit++;
344 nbits++;
345 }
346 }
347 bip->bli_format.blf_size = nvecs;
348
349 /*
350 * Check to make sure everything is consistent.
351 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000352 trace_xfs_buf_item_format(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 xfs_buf_item_log_check(bip);
354}
355
356/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000357 * This is called to pin the buffer associated with the buf log item in memory
358 * so it cannot be written out. Simply call bpin() on the buffer to do this.
359 *
360 * We also always take a reference to the buffer log item here so that the bli
361 * is held while the item is pinned in memory. This means that we can
362 * unconditionally drop the reference count a transaction holds when the
363 * transaction is completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 */
Dave Chinner64fc35d2010-05-07 11:04:34 +1000365
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000366STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367xfs_buf_item_pin(
368 xfs_buf_log_item_t *bip)
369{
370 xfs_buf_t *bp;
371
372 bp = bip->bli_buf;
373 ASSERT(XFS_BUF_ISBUSY(bp));
374 ASSERT(atomic_read(&bip->bli_refcount) > 0);
375 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
376 (bip->bli_flags & XFS_BLI_STALE));
Dave Chinner64fc35d2010-05-07 11:04:34 +1000377 atomic_inc(&bip->bli_refcount);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000378 trace_xfs_buf_item_pin(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379 xfs_bpin(bp);
380}
381
382
383/*
384 * This is called to unpin the buffer associated with the buf log
385 * item which was previously pinned with a call to xfs_buf_item_pin().
386 * Just call bunpin() on the buffer to do this.
387 *
388 * Also drop the reference to the buf item for the current transaction.
389 * If the XFS_BLI_STALE flag is set and we are the last reference,
390 * then free up the buf log item and unlock the buffer.
Christoph Hellwig9412e312010-06-23 18:11:15 +1000391 *
392 * If the remove flag is set we are called from uncommit in the
393 * forced-shutdown path. If that is true and the reference count on
394 * the log item is going to drop to zero we need to free the item's
395 * descriptor in the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000397STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398xfs_buf_item_unpin(
Christoph Hellwig9412e312010-06-23 18:11:15 +1000399 xfs_buf_log_item_t *bip,
400 int remove)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401{
David Chinner783a2f62008-10-30 17:39:58 +1100402 struct xfs_ail *ailp;
Christoph Hellwig9412e312010-06-23 18:11:15 +1000403 xfs_buf_t *bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404 int freed;
Dave Chinner8e123852010-03-08 11:26:03 +1100405 int stale = bip->bli_flags & XFS_BLI_STALE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip);
408 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000409
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000410 trace_xfs_buf_item_unpin(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411
412 freed = atomic_dec_and_test(&bip->bli_refcount);
David Chinner783a2f62008-10-30 17:39:58 +1100413 ailp = bip->bli_item.li_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 xfs_bunpin(bp);
415 if (freed && stale) {
416 ASSERT(bip->bli_flags & XFS_BLI_STALE);
417 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
418 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
419 ASSERT(XFS_BUF_ISSTALE(bp));
Dave Chinnerc1155412010-05-07 11:05:19 +1000420 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000421
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000422 trace_xfs_buf_item_unpin_stale(bip);
423
Christoph Hellwig9412e312010-06-23 18:11:15 +1000424 if (remove) {
425 /*
426 * We have to remove the log item from the transaction
427 * as we are about to release our reference to the
428 * buffer. If we don't, the unlock that occurs later
429 * in xfs_trans_uncommit() will ry to reference the
430 * buffer which we no longer have a hold on.
431 */
432 xfs_trans_del_item(&bip->bli_item);
433
434 /*
435 * Since the transaction no longer refers to the buffer,
436 * the buffer should no longer refer to the transaction.
437 */
438 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
439 }
440
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 /*
442 * If we get called here because of an IO error, we may
David Chinner783a2f62008-10-30 17:39:58 +1100443 * or may not have the item on the AIL. xfs_trans_ail_delete()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700444 * will take care of that situation.
David Chinner783a2f62008-10-30 17:39:58 +1100445 * xfs_trans_ail_delete() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446 */
447 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
448 xfs_buf_do_callbacks(bp, (xfs_log_item_t *)bip);
449 XFS_BUF_SET_FSPRIVATE(bp, NULL);
450 XFS_BUF_CLR_IODONE_FUNC(bp);
451 } else {
David Chinner783a2f62008-10-30 17:39:58 +1100452 spin_lock(&ailp->xa_lock);
453 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 xfs_buf_item_relse(bp);
455 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL);
456 }
457 xfs_buf_relse(bp);
458 }
459}
460
461/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 * This is called to attempt to lock the buffer associated with this
463 * buf log item. Don't sleep on the buffer lock. If we can't get
Dave Chinnerd808f612010-02-02 10:13:42 +1100464 * the lock right away, return 0. If we can get the lock, take a
465 * reference to the buffer. If this is a delayed write buffer that
466 * needs AIL help to be written back, invoke the pushbuf routine
467 * rather than the normal success path.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000469STATIC uint
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470xfs_buf_item_trylock(
471 xfs_buf_log_item_t *bip)
472{
473 xfs_buf_t *bp;
474
475 bp = bip->bli_buf;
Dave Chinnerd808f612010-02-02 10:13:42 +1100476 if (XFS_BUF_ISPINNED(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 return XFS_ITEM_PINNED;
Dave Chinnerd808f612010-02-02 10:13:42 +1100478 if (!XFS_BUF_CPSEMA(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479 return XFS_ITEM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
Dave Chinnerd808f612010-02-02 10:13:42 +1100481 /* take a reference to the buffer. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 XFS_BUF_HOLD(bp);
483
484 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000485 trace_xfs_buf_item_trylock(bip);
Dave Chinnerd808f612010-02-02 10:13:42 +1100486 if (XFS_BUF_ISDELAYWRITE(bp))
487 return XFS_ITEM_PUSHBUF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 return XFS_ITEM_SUCCESS;
489}
490
491/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000492 * Release the buffer associated with the buf log item. If there is no dirty
493 * logged data associated with the buffer recorded in the buf log item, then
494 * free the buf log item and remove the reference to it in the buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700495 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000496 * This call ignores the recursion count. It is only called when the buffer
497 * should REALLY be unlocked, regardless of the recursion count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000499 * We unconditionally drop the transaction's reference to the log item. If the
500 * item was logged, then another reference was taken when it was pinned, so we
501 * can safely drop the transaction reference now. This also allows us to avoid
502 * potential races with the unpin code freeing the bli by not referencing the
503 * bli after we've dropped the reference count.
504 *
505 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
506 * if necessary but do not unlock the buffer. This is for support of
507 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
508 * free the item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000510STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511xfs_buf_item_unlock(
512 xfs_buf_log_item_t *bip)
513{
514 int aborted;
515 xfs_buf_t *bp;
516 uint hold;
517
518 bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519
Dave Chinner64fc35d2010-05-07 11:04:34 +1000520 /* Clear the buffer's association with this transaction. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
522
523 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000524 * If this is a transaction abort, don't return early. Instead, allow
525 * the brelse to happen. Normally it would be done for stale
526 * (cancelled) buffers at unpin time, but we'll never go through the
527 * pin/unpin cycle if we abort inside commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 */
529 aborted = (bip->bli_item.li_flags & XFS_LI_ABORTED) != 0;
530
531 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 * Before possibly freeing the buf item, determine if we should
533 * release the buffer at the end of this routine.
534 */
535 hold = bip->bli_flags & XFS_BLI_HOLD;
Dave Chinner64fc35d2010-05-07 11:04:34 +1000536
537 /* Clear the per transaction state. */
538 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD);
539
540 /*
541 * If the buf item is marked stale, then don't do anything. We'll
542 * unlock the buffer and free the buf item when the buffer is unpinned
543 * for the last time.
544 */
545 if (bip->bli_flags & XFS_BLI_STALE) {
546 trace_xfs_buf_item_unlock_stale(bip);
Dave Chinnerc1155412010-05-07 11:05:19 +1000547 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000548 if (!aborted) {
549 atomic_dec(&bip->bli_refcount);
550 return;
551 }
552 }
553
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000554 trace_xfs_buf_item_unlock(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700555
556 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000557 * If the buf item isn't tracking any data, free it, otherwise drop the
558 * reference we hold to it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 */
Eric Sandeen24ad33f2007-06-28 16:43:30 +1000560 if (xfs_bitmap_empty(bip->bli_format.blf_data_map,
Dave Chinner64fc35d2010-05-07 11:04:34 +1000561 bip->bli_format.blf_map_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 xfs_buf_item_relse(bp);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000563 else
564 atomic_dec(&bip->bli_refcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565
Dave Chinner64fc35d2010-05-07 11:04:34 +1000566 if (!hold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568}
569
570/*
571 * This is called to find out where the oldest active copy of the
572 * buf log item in the on disk log resides now that the last log
573 * write of it completed at the given lsn.
574 * We always re-log all the dirty data in a buffer, so usually the
575 * latest copy in the on disk log is the only one that matters. For
576 * those cases we simply return the given lsn.
577 *
578 * The one exception to this is for buffers full of newly allocated
579 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
580 * flag set, indicating that only the di_next_unlinked fields from the
581 * inodes in the buffers will be replayed during recovery. If the
582 * original newly allocated inode images have not yet been flushed
583 * when the buffer is so relogged, then we need to make sure that we
584 * keep the old images in the 'active' portion of the log. We do this
585 * by returning the original lsn of that transaction here rather than
586 * the current one.
587 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000588STATIC xfs_lsn_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589xfs_buf_item_committed(
590 xfs_buf_log_item_t *bip,
591 xfs_lsn_t lsn)
592{
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000593 trace_xfs_buf_item_committed(bip);
594
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
596 (bip->bli_item.li_lsn != 0)) {
597 return bip->bli_item.li_lsn;
598 }
599 return (lsn);
600}
601
602/*
Dave Chinnerd808f612010-02-02 10:13:42 +1100603 * The buffer is locked, but is not a delayed write buffer. This happens
604 * if we race with IO completion and hence we don't want to try to write it
605 * again. Just release the buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000607STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608xfs_buf_item_push(
609 xfs_buf_log_item_t *bip)
610{
611 xfs_buf_t *bp;
612
613 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000614 trace_xfs_buf_item_push(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615
616 bp = bip->bli_buf;
Dave Chinnerd808f612010-02-02 10:13:42 +1100617 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
618 xfs_buf_relse(bp);
619}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Dave Chinnerd808f612010-02-02 10:13:42 +1100621/*
622 * The buffer is locked and is a delayed write buffer. Promote the buffer
623 * in the delayed write queue as the caller knows that they must invoke
624 * the xfsbufd to get this buffer written. We have to unlock the buffer
625 * to allow the xfsbufd to write it, too.
626 */
627STATIC void
628xfs_buf_item_pushbuf(
629 xfs_buf_log_item_t *bip)
630{
631 xfs_buf_t *bp;
632
633 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
634 trace_xfs_buf_item_pushbuf(bip);
635
636 bp = bip->bli_buf;
637 ASSERT(XFS_BUF_ISDELAYWRITE(bp));
638 xfs_buf_delwri_promote(bp);
639 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640}
641
642/* ARGSUSED */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000643STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644xfs_buf_item_committing(xfs_buf_log_item_t *bip, xfs_lsn_t commit_lsn)
645{
646}
647
648/*
649 * This is the ops vector shared by all buf log items.
650 */
David Chinner7989cb82007-02-10 18:34:56 +1100651static struct xfs_item_ops xfs_buf_item_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652 .iop_size = (uint(*)(xfs_log_item_t*))xfs_buf_item_size,
653 .iop_format = (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
654 xfs_buf_item_format,
655 .iop_pin = (void(*)(xfs_log_item_t*))xfs_buf_item_pin,
Christoph Hellwig9412e312010-06-23 18:11:15 +1000656 .iop_unpin = (void(*)(xfs_log_item_t*, int))xfs_buf_item_unpin,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657 .iop_trylock = (uint(*)(xfs_log_item_t*))xfs_buf_item_trylock,
658 .iop_unlock = (void(*)(xfs_log_item_t*))xfs_buf_item_unlock,
659 .iop_committed = (xfs_lsn_t(*)(xfs_log_item_t*, xfs_lsn_t))
660 xfs_buf_item_committed,
661 .iop_push = (void(*)(xfs_log_item_t*))xfs_buf_item_push,
Dave Chinnerd808f612010-02-02 10:13:42 +1100662 .iop_pushbuf = (void(*)(xfs_log_item_t*))xfs_buf_item_pushbuf,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663 .iop_committing = (void(*)(xfs_log_item_t*, xfs_lsn_t))
664 xfs_buf_item_committing
665};
666
667
668/*
669 * Allocate a new buf log item to go with the given buffer.
670 * Set the buffer's b_fsprivate field to point to the new
671 * buf log item. If there are other item's attached to the
672 * buffer (see xfs_buf_attach_iodone() below), then put the
673 * buf log item at the front.
674 */
675void
676xfs_buf_item_init(
677 xfs_buf_t *bp,
678 xfs_mount_t *mp)
679{
680 xfs_log_item_t *lip;
681 xfs_buf_log_item_t *bip;
682 int chunks;
683 int map_size;
684
685 /*
686 * Check to see if there is already a buf log item for
687 * this buffer. If there is, it is guaranteed to be
688 * the first. If we do already have one, there is
689 * nothing to do here so return.
690 */
Christoph Hellwig15ac08a2008-12-09 04:47:30 -0500691 if (bp->b_mount != mp)
692 bp->b_mount = mp;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 XFS_BUF_SET_BDSTRAT_FUNC(bp, xfs_bdstrat_cb);
694 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
695 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
696 if (lip->li_type == XFS_LI_BUF) {
697 return;
698 }
699 }
700
701 /*
Dave Chinnerc1155412010-05-07 11:05:19 +1000702 * chunks is the number of XFS_BLF_CHUNK size pieces
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 * the buffer can be divided into. Make sure not to
704 * truncate any pieces. map_size is the size of the
705 * bitmap needed to describe the chunks of the buffer.
706 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000707 chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLF_CHUNK - 1)) >> XFS_BLF_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);
709
710 bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
711 KM_SLEEP);
Dave Chinner43f5efc2010-03-23 10:10:00 +1100712 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 bip->bli_buf = bp;
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000714 xfs_buf_hold(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 bip->bli_format.blf_type = XFS_LI_BUF;
716 bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp);
717 bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp));
718 bip->bli_format.blf_map_size = map_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719
720#ifdef XFS_TRANS_DEBUG
721 /*
722 * Allocate the arrays for tracking what needs to be logged
723 * and what our callers request to be logged. bli_orig
724 * holds a copy of the original, clean buffer for comparison
725 * against, and bli_logged keeps a 1 bit flag per byte in
726 * the buffer to indicate which bytes the callers have asked
727 * to have logged.
728 */
729 bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP);
730 memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp));
731 bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP);
732#endif
733
734 /*
735 * Put the buf item into the list of items attached to the
736 * buffer at the front.
737 */
738 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
739 bip->bli_item.li_bio_list =
740 XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
741 }
742 XFS_BUF_SET_FSPRIVATE(bp, bip);
743}
744
745
746/*
747 * Mark bytes first through last inclusive as dirty in the buf
748 * item's bitmap.
749 */
750void
751xfs_buf_item_log(
752 xfs_buf_log_item_t *bip,
753 uint first,
754 uint last)
755{
756 uint first_bit;
757 uint last_bit;
758 uint bits_to_set;
759 uint bits_set;
760 uint word_num;
761 uint *wordp;
762 uint bit;
763 uint end_bit;
764 uint mask;
765
766 /*
767 * Mark the item as having some dirty data for
768 * quick reference in xfs_buf_item_dirty.
769 */
770 bip->bli_flags |= XFS_BLI_DIRTY;
771
772 /*
773 * Convert byte offsets to bit numbers.
774 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000775 first_bit = first >> XFS_BLF_SHIFT;
776 last_bit = last >> XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778 /*
779 * Calculate the total number of bits to be set.
780 */
781 bits_to_set = last_bit - first_bit + 1;
782
783 /*
784 * Get a pointer to the first word in the bitmap
785 * to set a bit in.
786 */
787 word_num = first_bit >> BIT_TO_WORD_SHIFT;
788 wordp = &(bip->bli_format.blf_data_map[word_num]);
789
790 /*
791 * Calculate the starting bit in the first word.
792 */
793 bit = first_bit & (uint)(NBWORD - 1);
794
795 /*
796 * First set any bits in the first word of our range.
797 * If it starts at bit 0 of the word, it will be
798 * set below rather than here. That is what the variable
799 * bit tells us. The variable bits_set tracks the number
800 * of bits that have been set so far. End_bit is the number
801 * of the last bit to be set in this word plus one.
802 */
803 if (bit) {
804 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
805 mask = ((1 << (end_bit - bit)) - 1) << bit;
806 *wordp |= mask;
807 wordp++;
808 bits_set = end_bit - bit;
809 } else {
810 bits_set = 0;
811 }
812
813 /*
814 * Now set bits a whole word at a time that are between
815 * first_bit and last_bit.
816 */
817 while ((bits_to_set - bits_set) >= NBWORD) {
818 *wordp |= 0xffffffff;
819 bits_set += NBWORD;
820 wordp++;
821 }
822
823 /*
824 * Finally, set any bits left to be set in one last partial word.
825 */
826 end_bit = bits_to_set - bits_set;
827 if (end_bit) {
828 mask = (1 << end_bit) - 1;
829 *wordp |= mask;
830 }
831
832 xfs_buf_item_log_debug(bip, first, last);
833}
834
835
836/*
837 * Return 1 if the buffer has some data that has been logged (at any
838 * point, not just the current transaction) and 0 if not.
839 */
840uint
841xfs_buf_item_dirty(
842 xfs_buf_log_item_t *bip)
843{
844 return (bip->bli_flags & XFS_BLI_DIRTY);
845}
846
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000847STATIC void
848xfs_buf_item_free(
849 xfs_buf_log_item_t *bip)
850{
851#ifdef XFS_TRANS_DEBUG
852 kmem_free(bip->bli_orig);
853 kmem_free(bip->bli_logged);
854#endif /* XFS_TRANS_DEBUG */
855
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000856 kmem_zone_free(xfs_buf_item_zone, bip);
857}
858
Linus Torvalds1da177e2005-04-16 15:20:36 -0700859/*
860 * This is called when the buf log item is no longer needed. It should
861 * free the buf log item associated with the given buffer and clear
862 * the buffer's pointer to the buf log item. If there are no more
863 * items in the list, clear the b_iodone field of the buffer (see
864 * xfs_buf_attach_iodone() below).
865 */
866void
867xfs_buf_item_relse(
868 xfs_buf_t *bp)
869{
870 xfs_buf_log_item_t *bip;
871
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000872 trace_xfs_buf_item_relse(bp, _RET_IP_);
873
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
875 XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list);
876 if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) &&
877 (XFS_BUF_IODONE_FUNC(bp) != NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 XFS_BUF_CLR_IODONE_FUNC(bp);
879 }
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000880 xfs_buf_rele(bp);
881 xfs_buf_item_free(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882}
883
884
885/*
886 * Add the given log item with its callback to the list of callbacks
887 * to be called when the buffer's I/O completes. If it is not set
888 * already, set the buffer's b_iodone() routine to be
889 * xfs_buf_iodone_callbacks() and link the log item into the list of
890 * items rooted at b_fsprivate. Items are always added as the second
891 * entry in the list if there is a first, because the buf item code
892 * assumes that the buf log item is first.
893 */
894void
895xfs_buf_attach_iodone(
896 xfs_buf_t *bp,
897 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
898 xfs_log_item_t *lip)
899{
900 xfs_log_item_t *head_lip;
901
902 ASSERT(XFS_BUF_ISBUSY(bp));
903 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
904
905 lip->li_cb = cb;
906 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
907 head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
908 lip->li_bio_list = head_lip->li_bio_list;
909 head_lip->li_bio_list = lip;
910 } else {
911 XFS_BUF_SET_FSPRIVATE(bp, lip);
912 }
913
914 ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) ||
915 (XFS_BUF_IODONE_FUNC(bp) == NULL));
916 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
917}
918
919STATIC void
920xfs_buf_do_callbacks(
921 xfs_buf_t *bp,
922 xfs_log_item_t *lip)
923{
924 xfs_log_item_t *nlip;
925
926 while (lip != NULL) {
927 nlip = lip->li_bio_list;
928 ASSERT(lip->li_cb != NULL);
929 /*
930 * Clear the next pointer so we don't have any
931 * confusion if the item is added to another buf.
932 * Don't touch the log item after calling its
933 * callback, because it could have freed itself.
934 */
935 lip->li_bio_list = NULL;
936 lip->li_cb(bp, lip);
937 lip = nlip;
938 }
939}
940
941/*
942 * This is the iodone() function for buffers which have had callbacks
943 * attached to them by xfs_buf_attach_iodone(). It should remove each
944 * log item from the buffer's list and call the callback of each in turn.
945 * When done, the buffer's fsprivate field is set to NULL and the buffer
946 * is unlocked with a call to iodone().
947 */
948void
949xfs_buf_iodone_callbacks(
950 xfs_buf_t *bp)
951{
952 xfs_log_item_t *lip;
953 static ulong lasttime;
954 static xfs_buftarg_t *lasttarg;
955 xfs_mount_t *mp;
956
957 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) != NULL);
958 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
959
960 if (XFS_BUF_GETERROR(bp) != 0) {
961 /*
962 * If we've already decided to shutdown the filesystem
963 * because of IO errors, there's no point in giving this
964 * a retry.
965 */
966 mp = lip->li_mountp;
967 if (XFS_FORCED_SHUTDOWN(mp)) {
968 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
969 XFS_BUF_SUPER_STALE(bp);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000970 trace_xfs_buf_item_iodone(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 xfs_buf_do_callbacks(bp, lip);
972 XFS_BUF_SET_FSPRIVATE(bp, NULL);
973 XFS_BUF_CLR_IODONE_FUNC(bp);
Lachlan McIlroy4fdc7782008-12-22 17:52:58 +1100974 xfs_biodone(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 return;
976 }
977
978 if ((XFS_BUF_TARGET(bp) != lasttarg) ||
979 (time_after(jiffies, (lasttime + 5*HZ)))) {
980 lasttime = jiffies;
Nathan Scottb6574522006-06-09 15:29:40 +1000981 cmn_err(CE_ALERT, "Device %s, XFS metadata write error"
982 " block 0x%llx in %s",
983 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 (__uint64_t)XFS_BUF_ADDR(bp), mp->m_fsname);
985 }
986 lasttarg = XFS_BUF_TARGET(bp);
987
988 if (XFS_BUF_ISASYNC(bp)) {
989 /*
990 * If the write was asynchronous then noone will be
991 * looking for the error. Clear the error state
992 * and write the buffer out again delayed write.
993 *
994 * XXXsup This is OK, so long as we catch these
995 * before we start the umount; we don't want these
996 * DELWRI metadata bufs to be hanging around.
997 */
998 XFS_BUF_ERROR(bp,0); /* errno of 0 unsets the flag */
999
1000 if (!(XFS_BUF_ISSTALE(bp))) {
1001 XFS_BUF_DELAYWRITE(bp);
1002 XFS_BUF_DONE(bp);
1003 XFS_BUF_SET_START(bp);
1004 }
1005 ASSERT(XFS_BUF_IODONE_FUNC(bp));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001006 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 xfs_buf_relse(bp);
1008 } else {
1009 /*
1010 * If the write of the buffer was not asynchronous,
1011 * then we want to make sure to return the error
1012 * to the caller of bwrite(). Because of this we
1013 * cannot clear the B_ERROR state at this point.
1014 * Instead we install a callback function that
1015 * will be called when the buffer is released, and
1016 * that routine will clear the error state and
1017 * set the buffer to be written out again after
1018 * some delay.
1019 */
1020 /* We actually overwrite the existing b-relse
1021 function at times, but we're gonna be shutting down
1022 anyway. */
1023 XFS_BUF_SET_BRELSE_FUNC(bp,xfs_buf_error_relse);
1024 XFS_BUF_DONE(bp);
David Chinnerb4dd3302008-08-13 16:36:11 +10001025 XFS_BUF_FINISH_IOWAIT(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 }
1027 return;
1028 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001029
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 xfs_buf_do_callbacks(bp, lip);
1031 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1032 XFS_BUF_CLR_IODONE_FUNC(bp);
1033 xfs_biodone(bp);
1034}
1035
1036/*
1037 * This is a callback routine attached to a buffer which gets an error
1038 * when being written out synchronously.
1039 */
1040STATIC void
1041xfs_buf_error_relse(
1042 xfs_buf_t *bp)
1043{
1044 xfs_log_item_t *lip;
1045 xfs_mount_t *mp;
1046
1047 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
1048 mp = (xfs_mount_t *)lip->li_mountp;
1049 ASSERT(XFS_BUF_TARGET(bp) == mp->m_ddev_targp);
1050
1051 XFS_BUF_STALE(bp);
1052 XFS_BUF_DONE(bp);
1053 XFS_BUF_UNDELAYWRITE(bp);
1054 XFS_BUF_ERROR(bp,0);
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001055
1056 trace_xfs_buf_error_relse(bp, _RET_IP_);
1057
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 if (! XFS_FORCED_SHUTDOWN(mp))
Nathan Scott7d04a332006-06-09 14:58:38 +10001059 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 /*
1061 * We have to unpin the pinned buffers so do the
1062 * callbacks.
1063 */
1064 xfs_buf_do_callbacks(bp, lip);
1065 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1066 XFS_BUF_CLR_IODONE_FUNC(bp);
1067 XFS_BUF_SET_BRELSE_FUNC(bp,NULL);
1068 xfs_buf_relse(bp);
1069}
1070
1071
1072/*
1073 * This is the iodone() function for buffers which have been
1074 * logged. It is called when they are eventually flushed out.
1075 * It should remove the buf item from the AIL, and free the buf item.
1076 * It is called by xfs_buf_iodone_callbacks() above which will take
1077 * care of cleaning up the buffer itself.
1078 */
1079/* ARGSUSED */
1080void
1081xfs_buf_iodone(
1082 xfs_buf_t *bp,
1083 xfs_buf_log_item_t *bip)
1084{
David Chinner783a2f62008-10-30 17:39:58 +11001085 struct xfs_ail *ailp = bip->bli_item.li_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
1087 ASSERT(bip->bli_buf == bp);
1088
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +10001089 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
1091 /*
1092 * If we are forcibly shutting down, this may well be
1093 * off the AIL already. That's because we simulate the
1094 * log-committed callbacks to unpin these buffers. Or we may never
1095 * have put this item on AIL because of the transaction was
David Chinner783a2f62008-10-30 17:39:58 +11001096 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 *
1098 * Either way, AIL is useless if we're forcing a shutdown.
1099 */
David Chinnerfc1829f2008-10-30 17:39:46 +11001100 spin_lock(&ailp->xa_lock);
David Chinner783a2f62008-10-30 17:39:58 +11001101 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip);
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +10001102 xfs_buf_item_free(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103}