blob: 7b7e005e3dccd3620ddf66ff334dfe4f6e94bc03 [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
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +100036static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip)
37{
38 return container_of(lip, struct xfs_buf_log_item, bli_item);
39}
40
41
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#ifdef XFS_TRANS_DEBUG
43/*
44 * This function uses an alternate strategy for tracking the bytes
45 * that the user requests to be logged. This can then be used
46 * in conjunction with the bli_orig array in the buf log item to
47 * catch bugs in our callers' code.
48 *
49 * We also double check the bits set in xfs_buf_item_log using a
50 * simple algorithm to check that every byte is accounted for.
51 */
52STATIC void
53xfs_buf_item_log_debug(
54 xfs_buf_log_item_t *bip,
55 uint first,
56 uint last)
57{
58 uint x;
59 uint byte;
60 uint nbytes;
61 uint chunk_num;
62 uint word_num;
63 uint bit_num;
64 uint bit_set;
65 uint *wordp;
66
67 ASSERT(bip->bli_logged != NULL);
68 byte = first;
69 nbytes = last - first + 1;
70 bfset(bip->bli_logged, first, nbytes);
71 for (x = 0; x < nbytes; x++) {
Dave Chinnerc1155412010-05-07 11:05:19 +100072 chunk_num = byte >> XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 word_num = chunk_num >> BIT_TO_WORD_SHIFT;
74 bit_num = chunk_num & (NBWORD - 1);
75 wordp = &(bip->bli_format.blf_data_map[word_num]);
76 bit_set = *wordp & (1 << bit_num);
77 ASSERT(bit_set);
78 byte++;
79 }
80}
81
82/*
83 * This function is called when we flush something into a buffer without
84 * logging it. This happens for things like inodes which are logged
85 * separately from the buffer.
86 */
87void
88xfs_buf_item_flush_log_debug(
89 xfs_buf_t *bp,
90 uint first,
91 uint last)
92{
93 xfs_buf_log_item_t *bip;
94 uint nbytes;
95
96 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
97 if ((bip == NULL) || (bip->bli_item.li_type != XFS_LI_BUF)) {
98 return;
99 }
100
101 ASSERT(bip->bli_logged != NULL);
102 nbytes = last - first + 1;
103 bfset(bip->bli_logged, first, nbytes);
104}
105
106/*
Nathan Scottc41564b2006-03-29 08:55:14 +1000107 * This function is called to verify that our callers have logged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 * all the bytes that they changed.
109 *
110 * It does this by comparing the original copy of the buffer stored in
111 * the buf log item's bli_orig array to the current copy of the buffer
Nathan Scottc41564b2006-03-29 08:55:14 +1000112 * and ensuring that all bytes which mismatch are set in the bli_logged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113 * array of the buf log item.
114 */
115STATIC void
116xfs_buf_item_log_check(
117 xfs_buf_log_item_t *bip)
118{
119 char *orig;
120 char *buffer;
121 int x;
122 xfs_buf_t *bp;
123
124 ASSERT(bip->bli_orig != NULL);
125 ASSERT(bip->bli_logged != NULL);
126
127 bp = bip->bli_buf;
128 ASSERT(XFS_BUF_COUNT(bp) > 0);
129 ASSERT(XFS_BUF_PTR(bp) != NULL);
130 orig = bip->bli_orig;
131 buffer = XFS_BUF_PTR(bp);
132 for (x = 0; x < XFS_BUF_COUNT(bp); x++) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100133 if (orig[x] != buffer[x] && !btst(bip->bli_logged, x)) {
134 xfs_emerg(bp->b_mount,
135 "%s: bip %x buffer %x orig %x index %d",
136 __func__, bip, bp, orig, x);
137 ASSERT(0);
138 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 }
140}
141#else
142#define xfs_buf_item_log_debug(x,y,z)
143#define xfs_buf_item_log_check(x)
144#endif
145
Dave Chinnerc90821a2010-12-03 17:00:52 +1100146STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147
148/*
149 * This returns the number of log iovecs needed to log the
150 * given buf log item.
151 *
152 * It calculates this as 1 iovec for the buf log format structure
153 * and 1 for each stretch of non-contiguous chunks to be logged.
154 * Contiguous chunks are logged in a single iovec.
155 *
156 * If the XFS_BLI_STALE flag has been set, then log nothing.
157 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000158STATIC uint
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159xfs_buf_item_size(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000160 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000162 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
163 struct xfs_buf *bp = bip->bli_buf;
164 uint nvecs;
165 int next_bit;
166 int last_bit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
168 ASSERT(atomic_read(&bip->bli_refcount) > 0);
169 if (bip->bli_flags & XFS_BLI_STALE) {
170 /*
171 * The buffer is stale, so all we need to log
172 * is the buf log format structure with the
173 * cancel flag in it.
174 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000175 trace_xfs_buf_item_size_stale(bip);
Dave Chinnerc1155412010-05-07 11:05:19 +1000176 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 return 1;
178 }
179
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
181 nvecs = 1;
182 last_bit = xfs_next_bit(bip->bli_format.blf_data_map,
183 bip->bli_format.blf_map_size, 0);
184 ASSERT(last_bit != -1);
185 nvecs++;
186 while (last_bit != -1) {
187 /*
188 * This takes the bit number to start looking from and
189 * returns the next set bit from there. It returns -1
190 * if there are no more bits set or the start bit is
191 * beyond the end of the bitmap.
192 */
193 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
194 bip->bli_format.blf_map_size,
195 last_bit + 1);
196 /*
197 * If we run out of bits, leave the loop,
198 * else if we find a new set of bits bump the number of vecs,
199 * else keep scanning the current set of bits.
200 */
201 if (next_bit == -1) {
202 last_bit = -1;
203 } else if (next_bit != last_bit + 1) {
204 last_bit = next_bit;
205 nvecs++;
Dave Chinnerc1155412010-05-07 11:05:19 +1000206 } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
207 (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
208 XFS_BLF_CHUNK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 last_bit = next_bit;
210 nvecs++;
211 } else {
212 last_bit++;
213 }
214 }
215
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000216 trace_xfs_buf_item_size(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 return nvecs;
218}
219
220/*
221 * This is called to fill in the vector of log iovecs for the
222 * given log buf item. It fills the first entry with a buf log
223 * format structure, and the rest point to contiguous chunks
224 * within the buffer.
225 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000226STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227xfs_buf_item_format(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000228 struct xfs_log_item *lip,
229 struct xfs_log_iovec *vecp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000231 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
232 struct xfs_buf *bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 uint base_size;
234 uint nvecs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 int first_bit;
236 int last_bit;
237 int next_bit;
238 uint nbits;
239 uint buffer_offset;
240
241 ASSERT(atomic_read(&bip->bli_refcount) > 0);
242 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
243 (bip->bli_flags & XFS_BLI_STALE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244
245 /*
246 * The size of the base structure is the size of the
247 * declared structure plus the space for the extra words
248 * of the bitmap. We subtract one from the map size, because
249 * the first element of the bitmap is accounted for in the
250 * size of the base structure.
251 */
252 base_size =
253 (uint)(sizeof(xfs_buf_log_format_t) +
254 ((bip->bli_format.blf_map_size - 1) * sizeof(uint)));
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000255 vecp->i_addr = &bip->bli_format;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 vecp->i_len = base_size;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000257 vecp->i_type = XLOG_REG_TYPE_BFORMAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 vecp++;
259 nvecs = 1;
260
Dave Chinnerccf7c232010-05-20 23:19:42 +1000261 /*
262 * If it is an inode buffer, transfer the in-memory state to the
263 * format flags and clear the in-memory state. We do not transfer
264 * this state if the inode buffer allocation has not yet been committed
265 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
266 * correct replay of the inode allocation.
267 */
268 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
269 if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000270 xfs_log_item_in_current_chkpt(lip)))
Dave Chinnerccf7c232010-05-20 23:19:42 +1000271 bip->bli_format.blf_flags |= XFS_BLF_INODE_BUF;
272 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
273 }
274
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 if (bip->bli_flags & XFS_BLI_STALE) {
276 /*
277 * The buffer is stale, so all we need to log
278 * is the buf log format structure with the
279 * cancel flag in it.
280 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000281 trace_xfs_buf_item_format_stale(bip);
Dave Chinnerc1155412010-05-07 11:05:19 +1000282 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 bip->bli_format.blf_size = nvecs;
284 return;
285 }
286
287 /*
288 * Fill in an iovec for each set of contiguous chunks.
289 */
290 first_bit = xfs_next_bit(bip->bli_format.blf_data_map,
291 bip->bli_format.blf_map_size, 0);
292 ASSERT(first_bit != -1);
293 last_bit = first_bit;
294 nbits = 1;
295 for (;;) {
296 /*
297 * This takes the bit number to start looking from and
298 * returns the next set bit from there. It returns -1
299 * if there are no more bits set or the start bit is
300 * beyond the end of the bitmap.
301 */
302 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
303 bip->bli_format.blf_map_size,
304 (uint)last_bit + 1);
305 /*
306 * If we run out of bits fill in the last iovec and get
307 * out of the loop.
308 * Else if we start a new set of bits then fill in the
309 * iovec for the series we were looking at and start
310 * counting the bits in the new one.
311 * Else we're still in the same set of bits so just
312 * keep counting and scanning.
313 */
314 if (next_bit == -1) {
Dave Chinnerc1155412010-05-07 11:05:19 +1000315 buffer_offset = first_bit * XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
Dave Chinnerc1155412010-05-07 11:05:19 +1000317 vecp->i_len = nbits * XFS_BLF_CHUNK;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000318 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 nvecs++;
320 break;
321 } else if (next_bit != last_bit + 1) {
Dave Chinnerc1155412010-05-07 11:05:19 +1000322 buffer_offset = first_bit * XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
Dave Chinnerc1155412010-05-07 11:05:19 +1000324 vecp->i_len = nbits * XFS_BLF_CHUNK;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000325 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 nvecs++;
327 vecp++;
328 first_bit = next_bit;
329 last_bit = next_bit;
330 nbits = 1;
Dave Chinnerc1155412010-05-07 11:05:19 +1000331 } else if (xfs_buf_offset(bp, next_bit << XFS_BLF_SHIFT) !=
332 (xfs_buf_offset(bp, last_bit << XFS_BLF_SHIFT) +
333 XFS_BLF_CHUNK)) {
334 buffer_offset = first_bit * XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
Dave Chinnerc1155412010-05-07 11:05:19 +1000336 vecp->i_len = nbits * XFS_BLF_CHUNK;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000337 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338/* You would think we need to bump the nvecs here too, but we do not
339 * this number is used by recovery, and it gets confused by the boundary
340 * split here
341 * nvecs++;
342 */
343 vecp++;
344 first_bit = next_bit;
345 last_bit = next_bit;
346 nbits = 1;
347 } else {
348 last_bit++;
349 nbits++;
350 }
351 }
352 bip->bli_format.blf_size = nvecs;
353
354 /*
355 * Check to make sure everything is consistent.
356 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000357 trace_xfs_buf_item_format(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 xfs_buf_item_log_check(bip);
359}
360
361/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000362 * This is called to pin the buffer associated with the buf log item in memory
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000363 * so it cannot be written out.
Dave Chinner64fc35d2010-05-07 11:04:34 +1000364 *
365 * We also always take a reference to the buffer log item here so that the bli
366 * is held while the item is pinned in memory. This means that we can
367 * unconditionally drop the reference count a transaction holds when the
368 * transaction is completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000370STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371xfs_buf_item_pin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000372 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000374 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000376 ASSERT(XFS_BUF_ISBUSY(bip->bli_buf));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 ASSERT(atomic_read(&bip->bli_refcount) > 0);
378 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
379 (bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000380
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000381 trace_xfs_buf_item_pin(bip);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000382
383 atomic_inc(&bip->bli_refcount);
384 atomic_inc(&bip->bli_buf->b_pin_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385}
386
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387/*
388 * This is called to unpin the buffer associated with the buf log
389 * item which was previously pinned with a call to xfs_buf_item_pin().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390 *
391 * Also drop the reference to the buf item for the current transaction.
392 * If the XFS_BLI_STALE flag is set and we are the last reference,
393 * then free up the buf log item and unlock the buffer.
Christoph Hellwig9412e312010-06-23 18:11:15 +1000394 *
395 * If the remove flag is set we are called from uncommit in the
396 * forced-shutdown path. If that is true and the reference count on
397 * the log item is going to drop to zero we need to free the item's
398 * descriptor in the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000400STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700401xfs_buf_item_unpin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000402 struct xfs_log_item *lip,
Christoph Hellwig9412e312010-06-23 18:11:15 +1000403 int remove)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000405 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000406 xfs_buf_t *bp = bip->bli_buf;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000407 struct xfs_ail *ailp = lip->li_ailp;
Dave Chinner8e123852010-03-08 11:26:03 +1100408 int stale = bip->bli_flags & XFS_BLI_STALE;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000409 int freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip);
412 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000413
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000414 trace_xfs_buf_item_unpin(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415
416 freed = atomic_dec_and_test(&bip->bli_refcount);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000417
418 if (atomic_dec_and_test(&bp->b_pin_count))
419 wake_up_all(&bp->b_waiters);
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000420
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 if (freed && stale) {
422 ASSERT(bip->bli_flags & XFS_BLI_STALE);
423 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
424 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
425 ASSERT(XFS_BUF_ISSTALE(bp));
Dave Chinnerc1155412010-05-07 11:05:19 +1000426 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000427
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000428 trace_xfs_buf_item_unpin_stale(bip);
429
Christoph Hellwig9412e312010-06-23 18:11:15 +1000430 if (remove) {
431 /*
Dave Chinnere34a3142011-01-27 12:13:35 +1100432 * If we are in a transaction context, we have to
433 * remove the log item from the transaction as we are
434 * about to release our reference to the buffer. If we
435 * don't, the unlock that occurs later in
436 * xfs_trans_uncommit() will try to reference the
Christoph Hellwig9412e312010-06-23 18:11:15 +1000437 * buffer which we no longer have a hold on.
438 */
Dave Chinnere34a3142011-01-27 12:13:35 +1100439 if (lip->li_desc)
440 xfs_trans_del_item(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000441
442 /*
443 * Since the transaction no longer refers to the buffer,
444 * the buffer should no longer refer to the transaction.
445 */
446 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
447 }
448
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 /*
450 * If we get called here because of an IO error, we may
David Chinner783a2f62008-10-30 17:39:58 +1100451 * or may not have the item on the AIL. xfs_trans_ail_delete()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 * will take care of that situation.
David Chinner783a2f62008-10-30 17:39:58 +1100453 * xfs_trans_ail_delete() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 */
455 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
Dave Chinnerc90821a2010-12-03 17:00:52 +1100456 xfs_buf_do_callbacks(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 XFS_BUF_SET_FSPRIVATE(bp, NULL);
458 XFS_BUF_CLR_IODONE_FUNC(bp);
459 } else {
David Chinner783a2f62008-10-30 17:39:58 +1100460 spin_lock(&ailp->xa_lock);
461 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 xfs_buf_item_relse(bp);
463 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL);
464 }
465 xfs_buf_relse(bp);
466 }
467}
468
469/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 * This is called to attempt to lock the buffer associated with this
471 * buf log item. Don't sleep on the buffer lock. If we can't get
Dave Chinnerd808f612010-02-02 10:13:42 +1100472 * the lock right away, return 0. If we can get the lock, take a
473 * reference to the buffer. If this is a delayed write buffer that
474 * needs AIL help to be written back, invoke the pushbuf routine
475 * rather than the normal success path.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000477STATIC uint
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478xfs_buf_item_trylock(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000479 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000481 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
482 struct xfs_buf *bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483
Dave Chinnerd808f612010-02-02 10:13:42 +1100484 if (XFS_BUF_ISPINNED(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 return XFS_ITEM_PINNED;
Dave Chinnerd808f612010-02-02 10:13:42 +1100486 if (!XFS_BUF_CPSEMA(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 return XFS_ITEM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
Dave Chinnerd808f612010-02-02 10:13:42 +1100489 /* take a reference to the buffer. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 XFS_BUF_HOLD(bp);
491
492 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000493 trace_xfs_buf_item_trylock(bip);
Dave Chinnerd808f612010-02-02 10:13:42 +1100494 if (XFS_BUF_ISDELAYWRITE(bp))
495 return XFS_ITEM_PUSHBUF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 return XFS_ITEM_SUCCESS;
497}
498
499/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000500 * Release the buffer associated with the buf log item. If there is no dirty
501 * logged data associated with the buffer recorded in the buf log item, then
502 * free the buf log item and remove the reference to it in the buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000504 * This call ignores the recursion count. It is only called when the buffer
505 * should REALLY be unlocked, regardless of the recursion count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700506 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000507 * We unconditionally drop the transaction's reference to the log item. If the
508 * item was logged, then another reference was taken when it was pinned, so we
509 * can safely drop the transaction reference now. This also allows us to avoid
510 * potential races with the unpin code freeing the bli by not referencing the
511 * bli after we've dropped the reference count.
512 *
513 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
514 * if necessary but do not unlock the buffer. This is for support of
515 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
516 * free the item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000518STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519xfs_buf_item_unlock(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000520 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000522 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
523 struct xfs_buf *bp = bip->bli_buf;
524 int aborted;
525 uint hold;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526
Dave Chinner64fc35d2010-05-07 11:04:34 +1000527 /* Clear the buffer's association with this transaction. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
529
530 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000531 * If this is a transaction abort, don't return early. Instead, allow
532 * the brelse to happen. Normally it would be done for stale
533 * (cancelled) buffers at unpin time, but we'll never go through the
534 * pin/unpin cycle if we abort inside commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535 */
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000536 aborted = (lip->li_flags & XFS_LI_ABORTED) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537
538 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700539 * Before possibly freeing the buf item, determine if we should
540 * release the buffer at the end of this routine.
541 */
542 hold = bip->bli_flags & XFS_BLI_HOLD;
Dave Chinner64fc35d2010-05-07 11:04:34 +1000543
544 /* Clear the per transaction state. */
545 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD);
546
547 /*
548 * If the buf item is marked stale, then don't do anything. We'll
549 * unlock the buffer and free the buf item when the buffer is unpinned
550 * for the last time.
551 */
552 if (bip->bli_flags & XFS_BLI_STALE) {
553 trace_xfs_buf_item_unlock_stale(bip);
Dave Chinnerc1155412010-05-07 11:05:19 +1000554 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000555 if (!aborted) {
556 atomic_dec(&bip->bli_refcount);
557 return;
558 }
559 }
560
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000561 trace_xfs_buf_item_unlock(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562
563 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000564 * If the buf item isn't tracking any data, free it, otherwise drop the
565 * reference we hold to it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 */
Eric Sandeen24ad33f2007-06-28 16:43:30 +1000567 if (xfs_bitmap_empty(bip->bli_format.blf_data_map,
Dave Chinner64fc35d2010-05-07 11:04:34 +1000568 bip->bli_format.blf_map_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 xfs_buf_item_relse(bp);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000570 else
571 atomic_dec(&bip->bli_refcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
Dave Chinner64fc35d2010-05-07 11:04:34 +1000573 if (!hold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700574 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575}
576
577/*
578 * This is called to find out where the oldest active copy of the
579 * buf log item in the on disk log resides now that the last log
580 * write of it completed at the given lsn.
581 * We always re-log all the dirty data in a buffer, so usually the
582 * latest copy in the on disk log is the only one that matters. For
583 * those cases we simply return the given lsn.
584 *
585 * The one exception to this is for buffers full of newly allocated
586 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
587 * flag set, indicating that only the di_next_unlinked fields from the
588 * inodes in the buffers will be replayed during recovery. If the
589 * original newly allocated inode images have not yet been flushed
590 * when the buffer is so relogged, then we need to make sure that we
591 * keep the old images in the 'active' portion of the log. We do this
592 * by returning the original lsn of that transaction here rather than
593 * the current one.
594 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000595STATIC xfs_lsn_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596xfs_buf_item_committed(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000597 struct xfs_log_item *lip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 xfs_lsn_t lsn)
599{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000600 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
601
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000602 trace_xfs_buf_item_committed(bip);
603
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000604 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
605 return lip->li_lsn;
606 return lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607}
608
609/*
Dave Chinnerd808f612010-02-02 10:13:42 +1100610 * The buffer is locked, but is not a delayed write buffer. This happens
611 * if we race with IO completion and hence we don't want to try to write it
612 * again. Just release the buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000614STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615xfs_buf_item_push(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000616 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000618 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
619 struct xfs_buf *bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000622 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
623
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000624 trace_xfs_buf_item_push(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
Dave Chinnerd808f612010-02-02 10:13:42 +1100626 xfs_buf_relse(bp);
627}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
Dave Chinnerd808f612010-02-02 10:13:42 +1100629/*
630 * The buffer is locked and is a delayed write buffer. Promote the buffer
631 * in the delayed write queue as the caller knows that they must invoke
632 * the xfsbufd to get this buffer written. We have to unlock the buffer
633 * to allow the xfsbufd to write it, too.
634 */
635STATIC void
636xfs_buf_item_pushbuf(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000637 struct xfs_log_item *lip)
Dave Chinnerd808f612010-02-02 10:13:42 +1100638{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000639 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
640 struct xfs_buf *bp = bip->bli_buf;
Dave Chinnerd808f612010-02-02 10:13:42 +1100641
642 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000643 ASSERT(XFS_BUF_ISDELAYWRITE(bp));
644
Dave Chinnerd808f612010-02-02 10:13:42 +1100645 trace_xfs_buf_item_pushbuf(bip);
646
Dave Chinnerd808f612010-02-02 10:13:42 +1100647 xfs_buf_delwri_promote(bp);
648 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649}
650
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000651STATIC void
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000652xfs_buf_item_committing(
653 struct xfs_log_item *lip,
654 xfs_lsn_t commit_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655{
656}
657
658/*
659 * This is the ops vector shared by all buf log items.
660 */
David Chinner7989cb82007-02-10 18:34:56 +1100661static struct xfs_item_ops xfs_buf_item_ops = {
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000662 .iop_size = xfs_buf_item_size,
663 .iop_format = xfs_buf_item_format,
664 .iop_pin = xfs_buf_item_pin,
665 .iop_unpin = xfs_buf_item_unpin,
666 .iop_trylock = xfs_buf_item_trylock,
667 .iop_unlock = xfs_buf_item_unlock,
668 .iop_committed = xfs_buf_item_committed,
669 .iop_push = xfs_buf_item_push,
670 .iop_pushbuf = xfs_buf_item_pushbuf,
671 .iop_committing = xfs_buf_item_committing
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672};
673
674
675/*
676 * Allocate a new buf log item to go with the given buffer.
677 * Set the buffer's b_fsprivate field to point to the new
678 * buf log item. If there are other item's attached to the
679 * buffer (see xfs_buf_attach_iodone() below), then put the
680 * buf log item at the front.
681 */
682void
683xfs_buf_item_init(
684 xfs_buf_t *bp,
685 xfs_mount_t *mp)
686{
687 xfs_log_item_t *lip;
688 xfs_buf_log_item_t *bip;
689 int chunks;
690 int map_size;
691
692 /*
693 * Check to see if there is already a buf log item for
694 * this buffer. If there is, it is guaranteed to be
695 * the first. If we do already have one, there is
696 * nothing to do here so return.
697 */
Dave Chinnerebad8612010-09-22 10:47:20 +1000698 ASSERT(bp->b_target->bt_mount == mp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
700 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
701 if (lip->li_type == XFS_LI_BUF) {
702 return;
703 }
704 }
705
706 /*
Dave Chinnerc1155412010-05-07 11:05:19 +1000707 * chunks is the number of XFS_BLF_CHUNK size pieces
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708 * the buffer can be divided into. Make sure not to
709 * truncate any pieces. map_size is the size of the
710 * bitmap needed to describe the chunks of the buffer.
711 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000712 chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLF_CHUNK - 1)) >> XFS_BLF_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);
714
715 bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
716 KM_SLEEP);
Dave Chinner43f5efc2010-03-23 10:10:00 +1100717 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 bip->bli_buf = bp;
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000719 xfs_buf_hold(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 bip->bli_format.blf_type = XFS_LI_BUF;
721 bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp);
722 bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp));
723 bip->bli_format.blf_map_size = map_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724
725#ifdef XFS_TRANS_DEBUG
726 /*
727 * Allocate the arrays for tracking what needs to be logged
728 * and what our callers request to be logged. bli_orig
729 * holds a copy of the original, clean buffer for comparison
730 * against, and bli_logged keeps a 1 bit flag per byte in
731 * the buffer to indicate which bytes the callers have asked
732 * to have logged.
733 */
734 bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP);
735 memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp));
736 bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP);
737#endif
738
739 /*
740 * Put the buf item into the list of items attached to the
741 * buffer at the front.
742 */
743 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
744 bip->bli_item.li_bio_list =
745 XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
746 }
747 XFS_BUF_SET_FSPRIVATE(bp, bip);
748}
749
750
751/*
752 * Mark bytes first through last inclusive as dirty in the buf
753 * item's bitmap.
754 */
755void
756xfs_buf_item_log(
757 xfs_buf_log_item_t *bip,
758 uint first,
759 uint last)
760{
761 uint first_bit;
762 uint last_bit;
763 uint bits_to_set;
764 uint bits_set;
765 uint word_num;
766 uint *wordp;
767 uint bit;
768 uint end_bit;
769 uint mask;
770
771 /*
772 * Mark the item as having some dirty data for
773 * quick reference in xfs_buf_item_dirty.
774 */
775 bip->bli_flags |= XFS_BLI_DIRTY;
776
777 /*
778 * Convert byte offsets to bit numbers.
779 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000780 first_bit = first >> XFS_BLF_SHIFT;
781 last_bit = last >> XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782
783 /*
784 * Calculate the total number of bits to be set.
785 */
786 bits_to_set = last_bit - first_bit + 1;
787
788 /*
789 * Get a pointer to the first word in the bitmap
790 * to set a bit in.
791 */
792 word_num = first_bit >> BIT_TO_WORD_SHIFT;
793 wordp = &(bip->bli_format.blf_data_map[word_num]);
794
795 /*
796 * Calculate the starting bit in the first word.
797 */
798 bit = first_bit & (uint)(NBWORD - 1);
799
800 /*
801 * First set any bits in the first word of our range.
802 * If it starts at bit 0 of the word, it will be
803 * set below rather than here. That is what the variable
804 * bit tells us. The variable bits_set tracks the number
805 * of bits that have been set so far. End_bit is the number
806 * of the last bit to be set in this word plus one.
807 */
808 if (bit) {
809 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
810 mask = ((1 << (end_bit - bit)) - 1) << bit;
811 *wordp |= mask;
812 wordp++;
813 bits_set = end_bit - bit;
814 } else {
815 bits_set = 0;
816 }
817
818 /*
819 * Now set bits a whole word at a time that are between
820 * first_bit and last_bit.
821 */
822 while ((bits_to_set - bits_set) >= NBWORD) {
823 *wordp |= 0xffffffff;
824 bits_set += NBWORD;
825 wordp++;
826 }
827
828 /*
829 * Finally, set any bits left to be set in one last partial word.
830 */
831 end_bit = bits_to_set - bits_set;
832 if (end_bit) {
833 mask = (1 << end_bit) - 1;
834 *wordp |= mask;
835 }
836
837 xfs_buf_item_log_debug(bip, first, last);
838}
839
840
841/*
842 * Return 1 if the buffer has some data that has been logged (at any
843 * point, not just the current transaction) and 0 if not.
844 */
845uint
846xfs_buf_item_dirty(
847 xfs_buf_log_item_t *bip)
848{
849 return (bip->bli_flags & XFS_BLI_DIRTY);
850}
851
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000852STATIC void
853xfs_buf_item_free(
854 xfs_buf_log_item_t *bip)
855{
856#ifdef XFS_TRANS_DEBUG
857 kmem_free(bip->bli_orig);
858 kmem_free(bip->bli_logged);
859#endif /* XFS_TRANS_DEBUG */
860
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000861 kmem_zone_free(xfs_buf_item_zone, bip);
862}
863
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864/*
865 * This is called when the buf log item is no longer needed. It should
866 * free the buf log item associated with the given buffer and clear
867 * the buffer's pointer to the buf log item. If there are no more
868 * items in the list, clear the b_iodone field of the buffer (see
869 * xfs_buf_attach_iodone() below).
870 */
871void
872xfs_buf_item_relse(
873 xfs_buf_t *bp)
874{
875 xfs_buf_log_item_t *bip;
876
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000877 trace_xfs_buf_item_relse(bp, _RET_IP_);
878
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
880 XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list);
881 if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) &&
882 (XFS_BUF_IODONE_FUNC(bp) != NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 XFS_BUF_CLR_IODONE_FUNC(bp);
884 }
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000885 xfs_buf_rele(bp);
886 xfs_buf_item_free(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887}
888
889
890/*
891 * Add the given log item with its callback to the list of callbacks
892 * to be called when the buffer's I/O completes. If it is not set
893 * already, set the buffer's b_iodone() routine to be
894 * xfs_buf_iodone_callbacks() and link the log item into the list of
895 * items rooted at b_fsprivate. Items are always added as the second
896 * entry in the list if there is a first, because the buf item code
897 * assumes that the buf log item is first.
898 */
899void
900xfs_buf_attach_iodone(
901 xfs_buf_t *bp,
902 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
903 xfs_log_item_t *lip)
904{
905 xfs_log_item_t *head_lip;
906
907 ASSERT(XFS_BUF_ISBUSY(bp));
908 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
909
910 lip->li_cb = cb;
911 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
912 head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
913 lip->li_bio_list = head_lip->li_bio_list;
914 head_lip->li_bio_list = lip;
915 } else {
916 XFS_BUF_SET_FSPRIVATE(bp, lip);
917 }
918
919 ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) ||
920 (XFS_BUF_IODONE_FUNC(bp) == NULL));
921 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
922}
923
Dave Chinnerc90821a2010-12-03 17:00:52 +1100924/*
925 * We can have many callbacks on a buffer. Running the callbacks individually
926 * can cause a lot of contention on the AIL lock, so we allow for a single
927 * callback to be able to scan the remaining lip->li_bio_list for other items
928 * of the same type and callback to be processed in the first call.
929 *
930 * As a result, the loop walking the callback list below will also modify the
931 * list. it removes the first item from the list and then runs the callback.
932 * The loop then restarts from the new head of the list. This allows the
933 * callback to scan and modify the list attached to the buffer and we don't
934 * have to care about maintaining a next item pointer.
935 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936STATIC void
937xfs_buf_do_callbacks(
Dave Chinnerc90821a2010-12-03 17:00:52 +1100938 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939{
Dave Chinnerc90821a2010-12-03 17:00:52 +1100940 struct xfs_log_item *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941
Dave Chinnerc90821a2010-12-03 17:00:52 +1100942 while ((lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *)) != NULL) {
943 XFS_BUF_SET_FSPRIVATE(bp, lip->li_bio_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 ASSERT(lip->li_cb != NULL);
945 /*
946 * Clear the next pointer so we don't have any
947 * confusion if the item is added to another buf.
948 * Don't touch the log item after calling its
949 * callback, because it could have freed itself.
950 */
951 lip->li_bio_list = NULL;
952 lip->li_cb(bp, lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953 }
954}
955
956/*
957 * This is the iodone() function for buffers which have had callbacks
958 * attached to them by xfs_buf_attach_iodone(). It should remove each
959 * log item from the buffer's list and call the callback of each in turn.
960 * When done, the buffer's fsprivate field is set to NULL and the buffer
961 * is unlocked with a call to iodone().
962 */
963void
964xfs_buf_iodone_callbacks(
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000965 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966{
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000967 struct xfs_log_item *lip = bp->b_fspriv;
968 struct xfs_mount *mp = lip->li_mountp;
969 static ulong lasttime;
970 static xfs_buftarg_t *lasttarg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000972 if (likely(!XFS_BUF_GETERROR(bp)))
973 goto do_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000975 /*
976 * If we've already decided to shutdown the filesystem because of
977 * I/O errors, there's no point in giving this a retry.
978 */
979 if (XFS_FORCED_SHUTDOWN(mp)) {
980 XFS_BUF_SUPER_STALE(bp);
981 trace_xfs_buf_item_iodone(bp, _RET_IP_);
982 goto do_callbacks;
983 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000985 if (XFS_BUF_TARGET(bp) != lasttarg ||
986 time_after(jiffies, (lasttime + 5*HZ))) {
987 lasttime = jiffies;
Dave Chinner0b932cc2011-03-07 10:08:35 +1100988 xfs_alert(mp, "Device %s: metadata write error block 0x%llx",
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000989 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)),
Dave Chinner0b932cc2011-03-07 10:08:35 +1100990 (__uint64_t)XFS_BUF_ADDR(bp));
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000991 }
992 lasttarg = XFS_BUF_TARGET(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000994 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300995 * If the write was asynchronous then no one will be looking for the
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000996 * error. Clear the error state and write the buffer out again.
997 *
998 * During sync or umount we'll write all pending buffers again
999 * synchronous, which will catch these errors if they keep hanging
1000 * around.
1001 */
1002 if (XFS_BUF_ISASYNC(bp)) {
1003 XFS_BUF_ERROR(bp, 0); /* errno of 0 unsets the flag */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001005 if (!XFS_BUF_ISSTALE(bp)) {
1006 XFS_BUF_DELAYWRITE(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 XFS_BUF_DONE(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001008 XFS_BUF_SET_START(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009 }
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001010 ASSERT(XFS_BUF_IODONE_FUNC(bp));
1011 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1012 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 return;
1014 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001015
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001016 /*
1017 * If the write of the buffer was synchronous, we want to make
1018 * sure to return the error to the caller of xfs_bwrite().
1019 */
1020 XFS_BUF_STALE(bp);
1021 XFS_BUF_DONE(bp);
1022 XFS_BUF_UNDELAYWRITE(bp);
1023
1024 trace_xfs_buf_error_relse(bp, _RET_IP_);
1025 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1026
1027do_callbacks:
Dave Chinnerc90821a2010-12-03 17:00:52 +11001028 xfs_buf_do_callbacks(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1030 XFS_BUF_CLR_IODONE_FUNC(bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001031 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032}
1033
1034/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035 * This is the iodone() function for buffers which have been
1036 * logged. It is called when they are eventually flushed out.
1037 * It should remove the buf item from the AIL, and free the buf item.
1038 * It is called by xfs_buf_iodone_callbacks() above which will take
1039 * care of cleaning up the buffer itself.
1040 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041void
1042xfs_buf_iodone(
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001043 struct xfs_buf *bp,
1044 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045{
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001046 struct xfs_ail *ailp = lip->li_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001048 ASSERT(BUF_ITEM(lip)->bli_buf == bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +10001050 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
1052 /*
1053 * If we are forcibly shutting down, this may well be
1054 * off the AIL already. That's because we simulate the
1055 * log-committed callbacks to unpin these buffers. Or we may never
1056 * have put this item on AIL because of the transaction was
David Chinner783a2f62008-10-30 17:39:58 +11001057 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 *
1059 * Either way, AIL is useless if we're forcing a shutdown.
1060 */
David Chinnerfc1829f2008-10-30 17:39:46 +11001061 spin_lock(&ailp->xa_lock);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001062 xfs_trans_ail_delete(ailp, lip);
1063 xfs_buf_item_free(BUF_ITEM(lip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064}