blob: 88492916c3dc0ed9b3bdd1aead549c8b00d87bbc [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{
Christoph Hellwigadadbee2011-07-13 13:43:49 +020093 xfs_buf_log_item_t *bip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 uint nbytes;
95
Christoph Hellwigadadbee2011-07-13 13:43:49 +020096 if (bip == NULL || (bip->bli_item.li_type != XFS_LI_BUF))
Linus Torvalds1da177e2005-04-16 15:20:36 -070097 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -070098
99 ASSERT(bip->bli_logged != NULL);
100 nbytes = last - first + 1;
101 bfset(bip->bli_logged, first, nbytes);
102}
103
104/*
Nathan Scottc41564b2006-03-29 08:55:14 +1000105 * This function is called to verify that our callers have logged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106 * all the bytes that they changed.
107 *
108 * It does this by comparing the original copy of the buffer stored in
109 * the buf log item's bli_orig array to the current copy of the buffer
Nathan Scottc41564b2006-03-29 08:55:14 +1000110 * and ensuring that all bytes which mismatch are set in the bli_logged
Linus Torvalds1da177e2005-04-16 15:20:36 -0700111 * array of the buf log item.
112 */
113STATIC void
114xfs_buf_item_log_check(
115 xfs_buf_log_item_t *bip)
116{
117 char *orig;
118 char *buffer;
119 int x;
120 xfs_buf_t *bp;
121
122 ASSERT(bip->bli_orig != NULL);
123 ASSERT(bip->bli_logged != NULL);
124
125 bp = bip->bli_buf;
126 ASSERT(XFS_BUF_COUNT(bp) > 0);
127 ASSERT(XFS_BUF_PTR(bp) != NULL);
128 orig = bip->bli_orig;
129 buffer = XFS_BUF_PTR(bp);
130 for (x = 0; x < XFS_BUF_COUNT(bp); x++) {
Dave Chinner0b932cc2011-03-07 10:08:35 +1100131 if (orig[x] != buffer[x] && !btst(bip->bli_logged, x)) {
132 xfs_emerg(bp->b_mount,
133 "%s: bip %x buffer %x orig %x index %d",
134 __func__, bip, bp, orig, x);
135 ASSERT(0);
136 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 }
138}
139#else
140#define xfs_buf_item_log_debug(x,y,z)
141#define xfs_buf_item_log_check(x)
142#endif
143
Dave Chinnerc90821a2010-12-03 17:00:52 +1100144STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
146/*
147 * This returns the number of log iovecs needed to log the
148 * given buf log item.
149 *
150 * It calculates this as 1 iovec for the buf log format structure
151 * and 1 for each stretch of non-contiguous chunks to be logged.
152 * Contiguous chunks are logged in a single iovec.
153 *
154 * If the XFS_BLI_STALE flag has been set, then log nothing.
155 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000156STATIC uint
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157xfs_buf_item_size(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000158 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000160 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
161 struct xfs_buf *bp = bip->bli_buf;
162 uint nvecs;
163 int next_bit;
164 int last_bit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165
166 ASSERT(atomic_read(&bip->bli_refcount) > 0);
167 if (bip->bli_flags & XFS_BLI_STALE) {
168 /*
169 * The buffer is stale, so all we need to log
170 * is the buf log format structure with the
171 * cancel flag in it.
172 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000173 trace_xfs_buf_item_size_stale(bip);
Dave Chinnerc1155412010-05-07 11:05:19 +1000174 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 return 1;
176 }
177
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 ASSERT(bip->bli_flags & XFS_BLI_LOGGED);
179 nvecs = 1;
180 last_bit = xfs_next_bit(bip->bli_format.blf_data_map,
181 bip->bli_format.blf_map_size, 0);
182 ASSERT(last_bit != -1);
183 nvecs++;
184 while (last_bit != -1) {
185 /*
186 * This takes the bit number to start looking from and
187 * returns the next set bit from there. It returns -1
188 * if there are no more bits set or the start bit is
189 * beyond the end of the bitmap.
190 */
191 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
192 bip->bli_format.blf_map_size,
193 last_bit + 1);
194 /*
195 * If we run out of bits, leave the loop,
196 * else if we find a new set of bits bump the number of vecs,
197 * else keep scanning the current set of bits.
198 */
199 if (next_bit == -1) {
200 last_bit = -1;
201 } else if (next_bit != last_bit + 1) {
202 last_bit = next_bit;
203 nvecs++;
Dave Chinnerc1155412010-05-07 11:05:19 +1000204 } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) !=
205 (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) +
206 XFS_BLF_CHUNK)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207 last_bit = next_bit;
208 nvecs++;
209 } else {
210 last_bit++;
211 }
212 }
213
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000214 trace_xfs_buf_item_size(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 return nvecs;
216}
217
218/*
219 * This is called to fill in the vector of log iovecs for the
220 * given log buf item. It fills the first entry with a buf log
221 * format structure, and the rest point to contiguous chunks
222 * within the buffer.
223 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000224STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225xfs_buf_item_format(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000226 struct xfs_log_item *lip,
227 struct xfs_log_iovec *vecp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000229 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
230 struct xfs_buf *bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 uint base_size;
232 uint nvecs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 int first_bit;
234 int last_bit;
235 int next_bit;
236 uint nbits;
237 uint buffer_offset;
238
239 ASSERT(atomic_read(&bip->bli_refcount) > 0);
240 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
241 (bip->bli_flags & XFS_BLI_STALE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
243 /*
244 * The size of the base structure is the size of the
245 * declared structure plus the space for the extra words
246 * of the bitmap. We subtract one from the map size, because
247 * the first element of the bitmap is accounted for in the
248 * size of the base structure.
249 */
250 base_size =
251 (uint)(sizeof(xfs_buf_log_format_t) +
252 ((bip->bli_format.blf_map_size - 1) * sizeof(uint)));
Christoph Hellwig4e0d5f92010-06-23 18:11:15 +1000253 vecp->i_addr = &bip->bli_format;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 vecp->i_len = base_size;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000255 vecp->i_type = XLOG_REG_TYPE_BFORMAT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 vecp++;
257 nvecs = 1;
258
Dave Chinnerccf7c232010-05-20 23:19:42 +1000259 /*
260 * If it is an inode buffer, transfer the in-memory state to the
261 * format flags and clear the in-memory state. We do not transfer
262 * this state if the inode buffer allocation has not yet been committed
263 * to the log as setting the XFS_BLI_INODE_BUF flag will prevent
264 * correct replay of the inode allocation.
265 */
266 if (bip->bli_flags & XFS_BLI_INODE_BUF) {
267 if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) &&
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000268 xfs_log_item_in_current_chkpt(lip)))
Dave Chinnerccf7c232010-05-20 23:19:42 +1000269 bip->bli_format.blf_flags |= XFS_BLF_INODE_BUF;
270 bip->bli_flags &= ~XFS_BLI_INODE_BUF;
271 }
272
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 if (bip->bli_flags & XFS_BLI_STALE) {
274 /*
275 * The buffer is stale, so all we need to log
276 * is the buf log format structure with the
277 * cancel flag in it.
278 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000279 trace_xfs_buf_item_format_stale(bip);
Dave Chinnerc1155412010-05-07 11:05:19 +1000280 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 bip->bli_format.blf_size = nvecs;
282 return;
283 }
284
285 /*
286 * Fill in an iovec for each set of contiguous chunks.
287 */
288 first_bit = xfs_next_bit(bip->bli_format.blf_data_map,
289 bip->bli_format.blf_map_size, 0);
290 ASSERT(first_bit != -1);
291 last_bit = first_bit;
292 nbits = 1;
293 for (;;) {
294 /*
295 * This takes the bit number to start looking from and
296 * returns the next set bit from there. It returns -1
297 * if there are no more bits set or the start bit is
298 * beyond the end of the bitmap.
299 */
300 next_bit = xfs_next_bit(bip->bli_format.blf_data_map,
301 bip->bli_format.blf_map_size,
302 (uint)last_bit + 1);
303 /*
304 * If we run out of bits fill in the last iovec and get
305 * out of the loop.
306 * Else if we start a new set of bits then fill in the
307 * iovec for the series we were looking at and start
308 * counting the bits in the new one.
309 * Else we're still in the same set of bits so just
310 * keep counting and scanning.
311 */
312 if (next_bit == -1) {
Dave Chinnerc1155412010-05-07 11:05:19 +1000313 buffer_offset = first_bit * XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
Dave Chinnerc1155412010-05-07 11:05:19 +1000315 vecp->i_len = nbits * XFS_BLF_CHUNK;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000316 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 nvecs++;
318 break;
319 } else if (next_bit != last_bit + 1) {
Dave Chinnerc1155412010-05-07 11:05:19 +1000320 buffer_offset = first_bit * XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
Dave Chinnerc1155412010-05-07 11:05:19 +1000322 vecp->i_len = nbits * XFS_BLF_CHUNK;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000323 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 nvecs++;
325 vecp++;
326 first_bit = next_bit;
327 last_bit = next_bit;
328 nbits = 1;
Dave Chinnerc1155412010-05-07 11:05:19 +1000329 } else if (xfs_buf_offset(bp, next_bit << XFS_BLF_SHIFT) !=
330 (xfs_buf_offset(bp, last_bit << XFS_BLF_SHIFT) +
331 XFS_BLF_CHUNK)) {
332 buffer_offset = first_bit * XFS_BLF_CHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 vecp->i_addr = xfs_buf_offset(bp, buffer_offset);
Dave Chinnerc1155412010-05-07 11:05:19 +1000334 vecp->i_len = nbits * XFS_BLF_CHUNK;
Christoph Hellwig4139b3b2010-01-19 09:56:45 +0000335 vecp->i_type = XLOG_REG_TYPE_BCHUNK;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336/* You would think we need to bump the nvecs here too, but we do not
337 * this number is used by recovery, and it gets confused by the boundary
338 * split here
339 * nvecs++;
340 */
341 vecp++;
342 first_bit = next_bit;
343 last_bit = next_bit;
344 nbits = 1;
345 } else {
346 last_bit++;
347 nbits++;
348 }
349 }
350 bip->bli_format.blf_size = nvecs;
351
352 /*
353 * Check to make sure everything is consistent.
354 */
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000355 trace_xfs_buf_item_format(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 xfs_buf_item_log_check(bip);
357}
358
359/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000360 * This is called to pin the buffer associated with the buf log item in memory
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000361 * so it cannot be written out.
Dave Chinner64fc35d2010-05-07 11:04:34 +1000362 *
363 * We also always take a reference to the buffer log item here so that the bli
364 * is held while the item is pinned in memory. This means that we can
365 * unconditionally drop the reference count a transaction holds when the
366 * transaction is completed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000368STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369xfs_buf_item_pin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000370 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000372 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000374 ASSERT(XFS_BUF_ISBUSY(bip->bli_buf));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 ASSERT(atomic_read(&bip->bli_refcount) > 0);
376 ASSERT((bip->bli_flags & XFS_BLI_LOGGED) ||
377 (bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000378
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000379 trace_xfs_buf_item_pin(bip);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000380
381 atomic_inc(&bip->bli_refcount);
382 atomic_inc(&bip->bli_buf->b_pin_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
384
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385/*
386 * This is called to unpin the buffer associated with the buf log
387 * item which was previously pinned with a call to xfs_buf_item_pin().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 *
389 * Also drop the reference to the buf item for the current transaction.
390 * If the XFS_BLI_STALE flag is set and we are the last reference,
391 * then free up the buf log item and unlock the buffer.
Christoph Hellwig9412e312010-06-23 18:11:15 +1000392 *
393 * If the remove flag is set we are called from uncommit in the
394 * forced-shutdown path. If that is true and the reference count on
395 * the log item is going to drop to zero we need to free the item's
396 * descriptor in the transaction.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000398STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399xfs_buf_item_unpin(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000400 struct xfs_log_item *lip,
Christoph Hellwig9412e312010-06-23 18:11:15 +1000401 int remove)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000403 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000404 xfs_buf_t *bp = bip->bli_buf;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000405 struct xfs_ail *ailp = lip->li_ailp;
Dave Chinner8e123852010-03-08 11:26:03 +1100406 int stale = bip->bli_flags & XFS_BLI_STALE;
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000407 int freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200409 ASSERT(bp->b_fspriv == bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410 ASSERT(atomic_read(&bip->bli_refcount) > 0);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000411
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000412 trace_xfs_buf_item_unpin(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413
414 freed = atomic_dec_and_test(&bip->bli_refcount);
Christoph Hellwig4d16e922010-06-23 18:11:15 +1000415
416 if (atomic_dec_and_test(&bp->b_pin_count))
417 wake_up_all(&bp->b_waiters);
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000418
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 if (freed && stale) {
420 ASSERT(bip->bli_flags & XFS_BLI_STALE);
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200421 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
423 ASSERT(XFS_BUF_ISSTALE(bp));
Dave Chinnerc1155412010-05-07 11:05:19 +1000424 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000425
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000426 trace_xfs_buf_item_unpin_stale(bip);
427
Christoph Hellwig9412e312010-06-23 18:11:15 +1000428 if (remove) {
429 /*
Dave Chinnere34a3142011-01-27 12:13:35 +1100430 * If we are in a transaction context, we have to
431 * remove the log item from the transaction as we are
432 * about to release our reference to the buffer. If we
433 * don't, the unlock that occurs later in
434 * xfs_trans_uncommit() will try to reference the
Christoph Hellwig9412e312010-06-23 18:11:15 +1000435 * buffer which we no longer have a hold on.
436 */
Dave Chinnere34a3142011-01-27 12:13:35 +1100437 if (lip->li_desc)
438 xfs_trans_del_item(lip);
Christoph Hellwig9412e312010-06-23 18:11:15 +1000439
440 /*
441 * Since the transaction no longer refers to the buffer,
442 * the buffer should no longer refer to the transaction.
443 */
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200444 bp->b_transp = NULL;
Christoph Hellwig9412e312010-06-23 18:11:15 +1000445 }
446
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 /*
448 * If we get called here because of an IO error, we may
David Chinner783a2f62008-10-30 17:39:58 +1100449 * or may not have the item on the AIL. xfs_trans_ail_delete()
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 * will take care of that situation.
David Chinner783a2f62008-10-30 17:39:58 +1100451 * xfs_trans_ail_delete() drops the AIL lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 */
453 if (bip->bli_flags & XFS_BLI_STALE_INODE) {
Dave Chinnerc90821a2010-12-03 17:00:52 +1100454 xfs_buf_do_callbacks(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200455 bp->b_fspriv = NULL;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200456 bp->b_iodone = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 } else {
David Chinner783a2f62008-10-30 17:39:58 +1100458 spin_lock(&ailp->xa_lock);
459 xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 xfs_buf_item_relse(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200461 ASSERT(bp->b_fspriv == NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 }
463 xfs_buf_relse(bp);
464 }
465}
466
467/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468 * 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
Dave Chinnerd808f612010-02-02 10:13:42 +1100470 * the lock right away, return 0. If we can get the lock, take a
471 * reference to the buffer. If this is a delayed write buffer that
472 * needs AIL help to be written back, invoke the pushbuf routine
473 * rather than the normal success path.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000475STATIC uint
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476xfs_buf_item_trylock(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000477 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000479 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
480 struct xfs_buf *bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481
Dave Chinnerd808f612010-02-02 10:13:42 +1100482 if (XFS_BUF_ISPINNED(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 return XFS_ITEM_PINNED;
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200484 if (!xfs_buf_trylock(bp))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485 return XFS_ITEM_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486
Dave Chinnerd808f612010-02-02 10:13:42 +1100487 /* take a reference to the buffer. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 XFS_BUF_HOLD(bp);
489
490 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000491 trace_xfs_buf_item_trylock(bip);
Dave Chinnerd808f612010-02-02 10:13:42 +1100492 if (XFS_BUF_ISDELAYWRITE(bp))
493 return XFS_ITEM_PUSHBUF;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494 return XFS_ITEM_SUCCESS;
495}
496
497/*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000498 * Release the buffer associated with the buf log item. If there is no dirty
499 * logged data associated with the buffer recorded in the buf log item, then
500 * free the buf log item and remove the reference to it in the buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000502 * This call ignores the recursion count. It is only called when the buffer
503 * should REALLY be unlocked, regardless of the recursion count.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 *
Dave Chinner64fc35d2010-05-07 11:04:34 +1000505 * We unconditionally drop the transaction's reference to the log item. If the
506 * item was logged, then another reference was taken when it was pinned, so we
507 * can safely drop the transaction reference now. This also allows us to avoid
508 * potential races with the unpin code freeing the bli by not referencing the
509 * bli after we've dropped the reference count.
510 *
511 * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item
512 * if necessary but do not unlock the buffer. This is for support of
513 * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't
514 * free the item.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000516STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700517xfs_buf_item_unlock(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000518 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700519{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000520 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
521 struct xfs_buf *bp = bip->bli_buf;
522 int aborted;
523 uint hold;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524
Dave Chinner64fc35d2010-05-07 11:04:34 +1000525 /* Clear the buffer's association with this transaction. */
Christoph Hellwigbf9d9012011-07-13 13:43:49 +0200526 bp->b_transp = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000529 * If this is a transaction abort, don't return early. Instead, allow
530 * the brelse to happen. Normally it would be done for stale
531 * (cancelled) buffers at unpin time, but we'll never go through the
532 * pin/unpin cycle if we abort inside commit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 */
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000534 aborted = (lip->li_flags & XFS_LI_ABORTED) != 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700535
536 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537 * Before possibly freeing the buf item, determine if we should
538 * release the buffer at the end of this routine.
539 */
540 hold = bip->bli_flags & XFS_BLI_HOLD;
Dave Chinner64fc35d2010-05-07 11:04:34 +1000541
542 /* Clear the per transaction state. */
543 bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD);
544
545 /*
546 * If the buf item is marked stale, then don't do anything. We'll
547 * unlock the buffer and free the buf item when the buffer is unpinned
548 * for the last time.
549 */
550 if (bip->bli_flags & XFS_BLI_STALE) {
551 trace_xfs_buf_item_unlock_stale(bip);
Dave Chinnerc1155412010-05-07 11:05:19 +1000552 ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000553 if (!aborted) {
554 atomic_dec(&bip->bli_refcount);
555 return;
556 }
557 }
558
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000559 trace_xfs_buf_item_unlock(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
561 /*
Dave Chinner64fc35d2010-05-07 11:04:34 +1000562 * If the buf item isn't tracking any data, free it, otherwise drop the
563 * reference we hold to it.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 */
Eric Sandeen24ad33f2007-06-28 16:43:30 +1000565 if (xfs_bitmap_empty(bip->bli_format.blf_data_map,
Dave Chinner64fc35d2010-05-07 11:04:34 +1000566 bip->bli_format.blf_map_size))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 xfs_buf_item_relse(bp);
Dave Chinner64fc35d2010-05-07 11:04:34 +1000568 else
569 atomic_dec(&bip->bli_refcount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
Dave Chinner64fc35d2010-05-07 11:04:34 +1000571 if (!hold)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573}
574
575/*
576 * This is called to find out where the oldest active copy of the
577 * buf log item in the on disk log resides now that the last log
578 * write of it completed at the given lsn.
579 * We always re-log all the dirty data in a buffer, so usually the
580 * latest copy in the on disk log is the only one that matters. For
581 * those cases we simply return the given lsn.
582 *
583 * The one exception to this is for buffers full of newly allocated
584 * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF
585 * flag set, indicating that only the di_next_unlinked fields from the
586 * inodes in the buffers will be replayed during recovery. If the
587 * original newly allocated inode images have not yet been flushed
588 * when the buffer is so relogged, then we need to make sure that we
589 * keep the old images in the 'active' portion of the log. We do this
590 * by returning the original lsn of that transaction here rather than
591 * the current one.
592 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000593STATIC xfs_lsn_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594xfs_buf_item_committed(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000595 struct xfs_log_item *lip,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 xfs_lsn_t lsn)
597{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000598 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
599
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000600 trace_xfs_buf_item_committed(bip);
601
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000602 if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0)
603 return lip->li_lsn;
604 return lsn;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605}
606
607/*
Dave Chinnerd808f612010-02-02 10:13:42 +1100608 * The buffer is locked, but is not a delayed write buffer. This happens
609 * if we race with IO completion and hence we don't want to try to write it
610 * again. Just release the buffer.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 */
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000612STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613xfs_buf_item_push(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000614 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000616 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
617 struct xfs_buf *bp = bip->bli_buf;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
619 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000620 ASSERT(!XFS_BUF_ISDELAYWRITE(bp));
621
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000622 trace_xfs_buf_item_push(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623
Dave Chinnerd808f612010-02-02 10:13:42 +1100624 xfs_buf_relse(bp);
625}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626
Dave Chinnerd808f612010-02-02 10:13:42 +1100627/*
628 * The buffer is locked and is a delayed write buffer. Promote the buffer
629 * in the delayed write queue as the caller knows that they must invoke
630 * the xfsbufd to get this buffer written. We have to unlock the buffer
631 * to allow the xfsbufd to write it, too.
632 */
633STATIC void
634xfs_buf_item_pushbuf(
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000635 struct xfs_log_item *lip)
Dave Chinnerd808f612010-02-02 10:13:42 +1100636{
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000637 struct xfs_buf_log_item *bip = BUF_ITEM(lip);
638 struct xfs_buf *bp = bip->bli_buf;
Dave Chinnerd808f612010-02-02 10:13:42 +1100639
640 ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000641 ASSERT(XFS_BUF_ISDELAYWRITE(bp));
642
Dave Chinnerd808f612010-02-02 10:13:42 +1100643 trace_xfs_buf_item_pushbuf(bip);
644
Dave Chinnerd808f612010-02-02 10:13:42 +1100645 xfs_buf_delwri_promote(bp);
646 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647}
648
Christoph Hellwigba0f32d2005-06-21 15:36:52 +1000649STATIC void
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000650xfs_buf_item_committing(
651 struct xfs_log_item *lip,
652 xfs_lsn_t commit_lsn)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653{
654}
655
656/*
657 * This is the ops vector shared by all buf log items.
658 */
David Chinner7989cb82007-02-10 18:34:56 +1100659static struct xfs_item_ops xfs_buf_item_ops = {
Christoph Hellwig7bfa31d2010-06-23 18:11:15 +1000660 .iop_size = xfs_buf_item_size,
661 .iop_format = xfs_buf_item_format,
662 .iop_pin = xfs_buf_item_pin,
663 .iop_unpin = xfs_buf_item_unpin,
664 .iop_trylock = xfs_buf_item_trylock,
665 .iop_unlock = xfs_buf_item_unlock,
666 .iop_committed = xfs_buf_item_committed,
667 .iop_push = xfs_buf_item_push,
668 .iop_pushbuf = xfs_buf_item_pushbuf,
669 .iop_committing = xfs_buf_item_committing
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670};
671
672
673/*
674 * Allocate a new buf log item to go with the given buffer.
675 * Set the buffer's b_fsprivate field to point to the new
676 * buf log item. If there are other item's attached to the
677 * buffer (see xfs_buf_attach_iodone() below), then put the
678 * buf log item at the front.
679 */
680void
681xfs_buf_item_init(
682 xfs_buf_t *bp,
683 xfs_mount_t *mp)
684{
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200685 xfs_log_item_t *lip = bp->b_fspriv;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 xfs_buf_log_item_t *bip;
687 int chunks;
688 int map_size;
689
690 /*
691 * Check to see if there is already a buf log item for
692 * this buffer. If there is, it is guaranteed to be
693 * the first. If we do already have one, there is
694 * nothing to do here so return.
695 */
Dave Chinnerebad8612010-09-22 10:47:20 +1000696 ASSERT(bp->b_target->bt_mount == mp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200697 if (lip != NULL && lip->li_type == XFS_LI_BUF)
698 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
700 /*
Dave Chinnerc1155412010-05-07 11:05:19 +1000701 * chunks is the number of XFS_BLF_CHUNK size pieces
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 * the buffer can be divided into. Make sure not to
703 * truncate any pieces. map_size is the size of the
704 * bitmap needed to describe the chunks of the buffer.
705 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000706 chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLF_CHUNK - 1)) >> XFS_BLF_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);
708
709 bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
710 KM_SLEEP);
Dave Chinner43f5efc2010-03-23 10:10:00 +1100711 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700712 bip->bli_buf = bp;
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000713 xfs_buf_hold(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 bip->bli_format.blf_type = XFS_LI_BUF;
715 bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp);
716 bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp));
717 bip->bli_format.blf_map_size = map_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
719#ifdef XFS_TRANS_DEBUG
720 /*
721 * Allocate the arrays for tracking what needs to be logged
722 * and what our callers request to be logged. bli_orig
723 * holds a copy of the original, clean buffer for comparison
724 * against, and bli_logged keeps a 1 bit flag per byte in
725 * the buffer to indicate which bytes the callers have asked
726 * to have logged.
727 */
728 bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP);
729 memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp));
730 bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP);
731#endif
732
733 /*
734 * Put the buf item into the list of items attached to the
735 * buffer at the front.
736 */
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200737 if (bp->b_fspriv)
738 bip->bli_item.li_bio_list = bp->b_fspriv;
739 bp->b_fspriv = bip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740}
741
742
743/*
744 * Mark bytes first through last inclusive as dirty in the buf
745 * item's bitmap.
746 */
747void
748xfs_buf_item_log(
749 xfs_buf_log_item_t *bip,
750 uint first,
751 uint last)
752{
753 uint first_bit;
754 uint last_bit;
755 uint bits_to_set;
756 uint bits_set;
757 uint word_num;
758 uint *wordp;
759 uint bit;
760 uint end_bit;
761 uint mask;
762
763 /*
764 * Mark the item as having some dirty data for
765 * quick reference in xfs_buf_item_dirty.
766 */
767 bip->bli_flags |= XFS_BLI_DIRTY;
768
769 /*
770 * Convert byte offsets to bit numbers.
771 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000772 first_bit = first >> XFS_BLF_SHIFT;
773 last_bit = last >> XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774
775 /*
776 * Calculate the total number of bits to be set.
777 */
778 bits_to_set = last_bit - first_bit + 1;
779
780 /*
781 * Get a pointer to the first word in the bitmap
782 * to set a bit in.
783 */
784 word_num = first_bit >> BIT_TO_WORD_SHIFT;
785 wordp = &(bip->bli_format.blf_data_map[word_num]);
786
787 /*
788 * Calculate the starting bit in the first word.
789 */
790 bit = first_bit & (uint)(NBWORD - 1);
791
792 /*
793 * First set any bits in the first word of our range.
794 * If it starts at bit 0 of the word, it will be
795 * set below rather than here. That is what the variable
796 * bit tells us. The variable bits_set tracks the number
797 * of bits that have been set so far. End_bit is the number
798 * of the last bit to be set in this word plus one.
799 */
800 if (bit) {
801 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
802 mask = ((1 << (end_bit - bit)) - 1) << bit;
803 *wordp |= mask;
804 wordp++;
805 bits_set = end_bit - bit;
806 } else {
807 bits_set = 0;
808 }
809
810 /*
811 * Now set bits a whole word at a time that are between
812 * first_bit and last_bit.
813 */
814 while ((bits_to_set - bits_set) >= NBWORD) {
815 *wordp |= 0xffffffff;
816 bits_set += NBWORD;
817 wordp++;
818 }
819
820 /*
821 * Finally, set any bits left to be set in one last partial word.
822 */
823 end_bit = bits_to_set - bits_set;
824 if (end_bit) {
825 mask = (1 << end_bit) - 1;
826 *wordp |= mask;
827 }
828
829 xfs_buf_item_log_debug(bip, first, last);
830}
831
832
833/*
834 * Return 1 if the buffer has some data that has been logged (at any
835 * point, not just the current transaction) and 0 if not.
836 */
837uint
838xfs_buf_item_dirty(
839 xfs_buf_log_item_t *bip)
840{
841 return (bip->bli_flags & XFS_BLI_DIRTY);
842}
843
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000844STATIC void
845xfs_buf_item_free(
846 xfs_buf_log_item_t *bip)
847{
848#ifdef XFS_TRANS_DEBUG
849 kmem_free(bip->bli_orig);
850 kmem_free(bip->bli_logged);
851#endif /* XFS_TRANS_DEBUG */
852
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000853 kmem_zone_free(xfs_buf_item_zone, bip);
854}
855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856/*
857 * This is called when the buf log item is no longer needed. It should
858 * free the buf log item associated with the given buffer and clear
859 * the buffer's pointer to the buf log item. If there are no more
860 * items in the list, clear the b_iodone field of the buffer (see
861 * xfs_buf_attach_iodone() below).
862 */
863void
864xfs_buf_item_relse(
865 xfs_buf_t *bp)
866{
867 xfs_buf_log_item_t *bip;
868
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000869 trace_xfs_buf_item_relse(bp, _RET_IP_);
870
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200871 bip = bp->b_fspriv;
872 bp->b_fspriv = bip->bli_item.li_bio_list;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200873 if (bp->b_fspriv == NULL)
874 bp->b_iodone = NULL;
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200875
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000876 xfs_buf_rele(bp);
877 xfs_buf_item_free(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878}
879
880
881/*
882 * Add the given log item with its callback to the list of callbacks
883 * to be called when the buffer's I/O completes. If it is not set
884 * already, set the buffer's b_iodone() routine to be
885 * xfs_buf_iodone_callbacks() and link the log item into the list of
886 * items rooted at b_fsprivate. Items are always added as the second
887 * entry in the list if there is a first, because the buf item code
888 * assumes that the buf log item is first.
889 */
890void
891xfs_buf_attach_iodone(
892 xfs_buf_t *bp,
893 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
894 xfs_log_item_t *lip)
895{
896 xfs_log_item_t *head_lip;
897
898 ASSERT(XFS_BUF_ISBUSY(bp));
Christoph Hellwig0c842ad2011-07-08 14:36:19 +0200899 ASSERT(xfs_buf_islocked(bp));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900
901 lip->li_cb = cb;
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200902 head_lip = bp->b_fspriv;
903 if (head_lip) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904 lip->li_bio_list = head_lip->li_bio_list;
905 head_lip->li_bio_list = lip;
906 } else {
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200907 bp->b_fspriv = lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908 }
909
Christoph Hellwigcb669ca2011-07-13 13:43:49 +0200910 ASSERT(bp->b_iodone == NULL ||
911 bp->b_iodone == xfs_buf_iodone_callbacks);
912 bp->b_iodone = xfs_buf_iodone_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913}
914
Dave Chinnerc90821a2010-12-03 17:00:52 +1100915/*
916 * We can have many callbacks on a buffer. Running the callbacks individually
917 * can cause a lot of contention on the AIL lock, so we allow for a single
918 * callback to be able to scan the remaining lip->li_bio_list for other items
919 * of the same type and callback to be processed in the first call.
920 *
921 * As a result, the loop walking the callback list below will also modify the
922 * list. it removes the first item from the list and then runs the callback.
923 * The loop then restarts from the new head of the list. This allows the
924 * callback to scan and modify the list attached to the buffer and we don't
925 * have to care about maintaining a next item pointer.
926 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927STATIC void
928xfs_buf_do_callbacks(
Dave Chinnerc90821a2010-12-03 17:00:52 +1100929 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930{
Dave Chinnerc90821a2010-12-03 17:00:52 +1100931 struct xfs_log_item *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
Christoph Hellwigadadbee2011-07-13 13:43:49 +0200933 while ((lip = bp->b_fspriv) != NULL) {
934 bp->b_fspriv = lip->li_bio_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 ASSERT(lip->li_cb != NULL);
936 /*
937 * Clear the next pointer so we don't have any
938 * confusion if the item is added to another buf.
939 * Don't touch the log item after calling its
940 * callback, because it could have freed itself.
941 */
942 lip->li_bio_list = NULL;
943 lip->li_cb(bp, lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944 }
945}
946
947/*
948 * This is the iodone() function for buffers which have had callbacks
949 * attached to them by xfs_buf_attach_iodone(). It should remove each
950 * log item from the buffer's list and call the callback of each in turn.
951 * When done, the buffer's fsprivate field is set to NULL and the buffer
952 * is unlocked with a call to iodone().
953 */
954void
955xfs_buf_iodone_callbacks(
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000956 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957{
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000958 struct xfs_log_item *lip = bp->b_fspriv;
959 struct xfs_mount *mp = lip->li_mountp;
960 static ulong lasttime;
961 static xfs_buftarg_t *lasttarg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000963 if (likely(!XFS_BUF_GETERROR(bp)))
964 goto do_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000966 /*
967 * If we've already decided to shutdown the filesystem because of
968 * I/O errors, there's no point in giving this a retry.
969 */
970 if (XFS_FORCED_SHUTDOWN(mp)) {
971 XFS_BUF_SUPER_STALE(bp);
972 trace_xfs_buf_item_iodone(bp, _RET_IP_);
973 goto do_callbacks;
974 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000976 if (XFS_BUF_TARGET(bp) != lasttarg ||
977 time_after(jiffies, (lasttime + 5*HZ))) {
978 lasttime = jiffies;
Dave Chinner0b932cc2011-03-07 10:08:35 +1100979 xfs_alert(mp, "Device %s: metadata write error block 0x%llx",
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000980 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)),
Dave Chinner0b932cc2011-03-07 10:08:35 +1100981 (__uint64_t)XFS_BUF_ADDR(bp));
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000982 }
983 lasttarg = XFS_BUF_TARGET(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000985 /*
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300986 * If the write was asynchronous then no one will be looking for the
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000987 * error. Clear the error state and write the buffer out again.
988 *
989 * During sync or umount we'll write all pending buffers again
990 * synchronous, which will catch these errors if they keep hanging
991 * around.
992 */
993 if (XFS_BUF_ISASYNC(bp)) {
994 XFS_BUF_ERROR(bp, 0); /* errno of 0 unsets the flag */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000996 if (!XFS_BUF_ISSTALE(bp)) {
997 XFS_BUF_DELAYWRITE(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998 XFS_BUF_DONE(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000999 XFS_BUF_SET_START(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 }
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02001001 ASSERT(bp->b_iodone != NULL);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001002 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1003 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 return;
1005 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001006
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001007 /*
1008 * If the write of the buffer was synchronous, we want to make
1009 * sure to return the error to the caller of xfs_bwrite().
1010 */
1011 XFS_BUF_STALE(bp);
1012 XFS_BUF_DONE(bp);
1013 XFS_BUF_UNDELAYWRITE(bp);
1014
1015 trace_xfs_buf_error_relse(bp, _RET_IP_);
1016 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1017
1018do_callbacks:
Dave Chinnerc90821a2010-12-03 17:00:52 +11001019 xfs_buf_do_callbacks(bp);
Christoph Hellwigadadbee2011-07-13 13:43:49 +02001020 bp->b_fspriv = NULL;
Christoph Hellwigcb669ca2011-07-13 13:43:49 +02001021 bp->b_iodone = NULL;
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001022 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001023}
1024
1025/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001026 * This is the iodone() function for buffers which have been
1027 * logged. It is called when they are eventually flushed out.
1028 * It should remove the buf item from the AIL, and free the buf item.
1029 * It is called by xfs_buf_iodone_callbacks() above which will take
1030 * care of cleaning up the buffer itself.
1031 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032void
1033xfs_buf_iodone(
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001034 struct xfs_buf *bp,
1035 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036{
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001037 struct xfs_ail *ailp = lip->li_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001039 ASSERT(BUF_ITEM(lip)->bli_buf == bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +10001041 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
1043 /*
1044 * If we are forcibly shutting down, this may well be
1045 * off the AIL already. That's because we simulate the
1046 * log-committed callbacks to unpin these buffers. Or we may never
1047 * have put this item on AIL because of the transaction was
David Chinner783a2f62008-10-30 17:39:58 +11001048 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 *
1050 * Either way, AIL is useless if we're forcing a shutdown.
1051 */
David Chinnerfc1829f2008-10-30 17:39:46 +11001052 spin_lock(&ailp->xa_lock);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001053 xfs_trans_ail_delete(ailp, lip);
1054 xfs_buf_item_free(BUF_ITEM(lip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055}