blob: 6f8c21ce0d6d95fd8c45a5d3eea4ef0240fad1fd [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++) {
133 if (orig[x] != buffer[x] && !btst(bip->bli_logged, x))
134 cmn_err(CE_PANIC,
135 "xfs_buf_item_log_check bip %x buffer %x orig %x index %d",
136 bip, bp, orig, x);
137 }
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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip);
410 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);
421 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
422 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 */
444 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
445 }
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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455 XFS_BUF_SET_FSPRIVATE(bp, NULL);
456 XFS_BUF_CLR_IODONE_FUNC(bp);
457 } 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);
461 ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL);
462 }
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;
Dave Chinnerd808f612010-02-02 10:13:42 +1100484 if (!XFS_BUF_CPSEMA(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. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 XFS_BUF_SET_FSPRIVATE2(bp, NULL);
527
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{
685 xfs_log_item_t *lip;
686 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);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
698 lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
699 if (lip->li_type == XFS_LI_BUF) {
700 return;
701 }
702 }
703
704 /*
Dave Chinnerc1155412010-05-07 11:05:19 +1000705 * chunks is the number of XFS_BLF_CHUNK size pieces
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 * the buffer can be divided into. Make sure not to
707 * truncate any pieces. map_size is the size of the
708 * bitmap needed to describe the chunks of the buffer.
709 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000710 chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLF_CHUNK - 1)) >> XFS_BLF_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711 map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT);
712
713 bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone,
714 KM_SLEEP);
Dave Chinner43f5efc2010-03-23 10:10:00 +1100715 xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 bip->bli_buf = bp;
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000717 xfs_buf_hold(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 bip->bli_format.blf_type = XFS_LI_BUF;
719 bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp);
720 bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp));
721 bip->bli_format.blf_map_size = map_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
723#ifdef XFS_TRANS_DEBUG
724 /*
725 * Allocate the arrays for tracking what needs to be logged
726 * and what our callers request to be logged. bli_orig
727 * holds a copy of the original, clean buffer for comparison
728 * against, and bli_logged keeps a 1 bit flag per byte in
729 * the buffer to indicate which bytes the callers have asked
730 * to have logged.
731 */
732 bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP);
733 memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp));
734 bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP);
735#endif
736
737 /*
738 * Put the buf item into the list of items attached to the
739 * buffer at the front.
740 */
741 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
742 bip->bli_item.li_bio_list =
743 XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
744 }
745 XFS_BUF_SET_FSPRIVATE(bp, bip);
746}
747
748
749/*
750 * Mark bytes first through last inclusive as dirty in the buf
751 * item's bitmap.
752 */
753void
754xfs_buf_item_log(
755 xfs_buf_log_item_t *bip,
756 uint first,
757 uint last)
758{
759 uint first_bit;
760 uint last_bit;
761 uint bits_to_set;
762 uint bits_set;
763 uint word_num;
764 uint *wordp;
765 uint bit;
766 uint end_bit;
767 uint mask;
768
769 /*
770 * Mark the item as having some dirty data for
771 * quick reference in xfs_buf_item_dirty.
772 */
773 bip->bli_flags |= XFS_BLI_DIRTY;
774
775 /*
776 * Convert byte offsets to bit numbers.
777 */
Dave Chinnerc1155412010-05-07 11:05:19 +1000778 first_bit = first >> XFS_BLF_SHIFT;
779 last_bit = last >> XFS_BLF_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780
781 /*
782 * Calculate the total number of bits to be set.
783 */
784 bits_to_set = last_bit - first_bit + 1;
785
786 /*
787 * Get a pointer to the first word in the bitmap
788 * to set a bit in.
789 */
790 word_num = first_bit >> BIT_TO_WORD_SHIFT;
791 wordp = &(bip->bli_format.blf_data_map[word_num]);
792
793 /*
794 * Calculate the starting bit in the first word.
795 */
796 bit = first_bit & (uint)(NBWORD - 1);
797
798 /*
799 * First set any bits in the first word of our range.
800 * If it starts at bit 0 of the word, it will be
801 * set below rather than here. That is what the variable
802 * bit tells us. The variable bits_set tracks the number
803 * of bits that have been set so far. End_bit is the number
804 * of the last bit to be set in this word plus one.
805 */
806 if (bit) {
807 end_bit = MIN(bit + bits_to_set, (uint)NBWORD);
808 mask = ((1 << (end_bit - bit)) - 1) << bit;
809 *wordp |= mask;
810 wordp++;
811 bits_set = end_bit - bit;
812 } else {
813 bits_set = 0;
814 }
815
816 /*
817 * Now set bits a whole word at a time that are between
818 * first_bit and last_bit.
819 */
820 while ((bits_to_set - bits_set) >= NBWORD) {
821 *wordp |= 0xffffffff;
822 bits_set += NBWORD;
823 wordp++;
824 }
825
826 /*
827 * Finally, set any bits left to be set in one last partial word.
828 */
829 end_bit = bits_to_set - bits_set;
830 if (end_bit) {
831 mask = (1 << end_bit) - 1;
832 *wordp |= mask;
833 }
834
835 xfs_buf_item_log_debug(bip, first, last);
836}
837
838
839/*
840 * Return 1 if the buffer has some data that has been logged (at any
841 * point, not just the current transaction) and 0 if not.
842 */
843uint
844xfs_buf_item_dirty(
845 xfs_buf_log_item_t *bip)
846{
847 return (bip->bli_flags & XFS_BLI_DIRTY);
848}
849
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000850STATIC void
851xfs_buf_item_free(
852 xfs_buf_log_item_t *bip)
853{
854#ifdef XFS_TRANS_DEBUG
855 kmem_free(bip->bli_orig);
856 kmem_free(bip->bli_logged);
857#endif /* XFS_TRANS_DEBUG */
858
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000859 kmem_zone_free(xfs_buf_item_zone, bip);
860}
861
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862/*
863 * This is called when the buf log item is no longer needed. It should
864 * free the buf log item associated with the given buffer and clear
865 * the buffer's pointer to the buf log item. If there are no more
866 * items in the list, clear the b_iodone field of the buffer (see
867 * xfs_buf_attach_iodone() below).
868 */
869void
870xfs_buf_item_relse(
871 xfs_buf_t *bp)
872{
873 xfs_buf_log_item_t *bip;
874
Christoph Hellwig0b1b2132009-12-14 23:14:59 +0000875 trace_xfs_buf_item_relse(bp, _RET_IP_);
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*);
878 XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list);
879 if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) &&
880 (XFS_BUF_IODONE_FUNC(bp) != NULL)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 XFS_BUF_CLR_IODONE_FUNC(bp);
882 }
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +1000883 xfs_buf_rele(bp);
884 xfs_buf_item_free(bip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885}
886
887
888/*
889 * Add the given log item with its callback to the list of callbacks
890 * to be called when the buffer's I/O completes. If it is not set
891 * already, set the buffer's b_iodone() routine to be
892 * xfs_buf_iodone_callbacks() and link the log item into the list of
893 * items rooted at b_fsprivate. Items are always added as the second
894 * entry in the list if there is a first, because the buf item code
895 * assumes that the buf log item is first.
896 */
897void
898xfs_buf_attach_iodone(
899 xfs_buf_t *bp,
900 void (*cb)(xfs_buf_t *, xfs_log_item_t *),
901 xfs_log_item_t *lip)
902{
903 xfs_log_item_t *head_lip;
904
905 ASSERT(XFS_BUF_ISBUSY(bp));
906 ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
907
908 lip->li_cb = cb;
909 if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) {
910 head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *);
911 lip->li_bio_list = head_lip->li_bio_list;
912 head_lip->li_bio_list = lip;
913 } else {
914 XFS_BUF_SET_FSPRIVATE(bp, lip);
915 }
916
917 ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) ||
918 (XFS_BUF_IODONE_FUNC(bp) == NULL));
919 XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks);
920}
921
Dave Chinnerc90821a2010-12-03 17:00:52 +1100922/*
923 * We can have many callbacks on a buffer. Running the callbacks individually
924 * can cause a lot of contention on the AIL lock, so we allow for a single
925 * callback to be able to scan the remaining lip->li_bio_list for other items
926 * of the same type and callback to be processed in the first call.
927 *
928 * As a result, the loop walking the callback list below will also modify the
929 * list. it removes the first item from the list and then runs the callback.
930 * The loop then restarts from the new head of the list. This allows the
931 * callback to scan and modify the list attached to the buffer and we don't
932 * have to care about maintaining a next item pointer.
933 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934STATIC void
935xfs_buf_do_callbacks(
Dave Chinnerc90821a2010-12-03 17:00:52 +1100936 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937{
Dave Chinnerc90821a2010-12-03 17:00:52 +1100938 struct xfs_log_item *lip;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
Dave Chinnerc90821a2010-12-03 17:00:52 +1100940 while ((lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *)) != NULL) {
941 XFS_BUF_SET_FSPRIVATE(bp, lip->li_bio_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 ASSERT(lip->li_cb != NULL);
943 /*
944 * Clear the next pointer so we don't have any
945 * confusion if the item is added to another buf.
946 * Don't touch the log item after calling its
947 * callback, because it could have freed itself.
948 */
949 lip->li_bio_list = NULL;
950 lip->li_cb(bp, lip);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 }
952}
953
954/*
955 * This is the iodone() function for buffers which have had callbacks
956 * attached to them by xfs_buf_attach_iodone(). It should remove each
957 * log item from the buffer's list and call the callback of each in turn.
958 * When done, the buffer's fsprivate field is set to NULL and the buffer
959 * is unlocked with a call to iodone().
960 */
961void
962xfs_buf_iodone_callbacks(
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000963 struct xfs_buf *bp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964{
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000965 struct xfs_log_item *lip = bp->b_fspriv;
966 struct xfs_mount *mp = lip->li_mountp;
967 static ulong lasttime;
968 static xfs_buftarg_t *lasttarg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000970 if (likely(!XFS_BUF_GETERROR(bp)))
971 goto do_callbacks;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000973 /*
974 * If we've already decided to shutdown the filesystem because of
975 * I/O errors, there's no point in giving this a retry.
976 */
977 if (XFS_FORCED_SHUTDOWN(mp)) {
978 XFS_BUF_SUPER_STALE(bp);
979 trace_xfs_buf_item_iodone(bp, _RET_IP_);
980 goto do_callbacks;
981 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000983 if (XFS_BUF_TARGET(bp) != lasttarg ||
984 time_after(jiffies, (lasttime + 5*HZ))) {
985 lasttime = jiffies;
986 cmn_err(CE_ALERT, "Device %s, XFS metadata write error"
987 " block 0x%llx in %s",
988 XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)),
989 (__uint64_t)XFS_BUF_ADDR(bp), mp->m_fsname);
990 }
991 lasttarg = XFS_BUF_TARGET(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
Christoph Hellwigbfc60172011-01-07 13:02:23 +0000993 /*
994 * If the write was asynchronous then noone will be looking for the
995 * error. Clear the error state and write the buffer out again.
996 *
997 * During sync or umount we'll write all pending buffers again
998 * synchronous, which will catch these errors if they keep hanging
999 * around.
1000 */
1001 if (XFS_BUF_ISASYNC(bp)) {
1002 XFS_BUF_ERROR(bp, 0); /* errno of 0 unsets the flag */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001004 if (!XFS_BUF_ISSTALE(bp)) {
1005 XFS_BUF_DELAYWRITE(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 XFS_BUF_DONE(bp);
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001007 XFS_BUF_SET_START(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008 }
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001009 ASSERT(XFS_BUF_IODONE_FUNC(bp));
1010 trace_xfs_buf_item_iodone_async(bp, _RET_IP_);
1011 xfs_buf_relse(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 return;
1013 }
Christoph Hellwig0b1b2132009-12-14 23:14:59 +00001014
Christoph Hellwigbfc60172011-01-07 13:02:23 +00001015 /*
1016 * If the write of the buffer was synchronous, we want to make
1017 * sure to return the error to the caller of xfs_bwrite().
1018 */
1019 XFS_BUF_STALE(bp);
1020 XFS_BUF_DONE(bp);
1021 XFS_BUF_UNDELAYWRITE(bp);
1022
1023 trace_xfs_buf_error_relse(bp, _RET_IP_);
1024 xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
1025
1026do_callbacks:
Dave Chinnerc90821a2010-12-03 17:00:52 +11001027 xfs_buf_do_callbacks(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028 XFS_BUF_SET_FSPRIVATE(bp, NULL);
1029 XFS_BUF_CLR_IODONE_FUNC(bp);
Christoph Hellwig1a1a3e92010-10-06 18:41:18 +00001030 xfs_buf_ioend(bp, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031}
1032
1033/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034 * This is the iodone() function for buffers which have been
1035 * logged. It is called when they are eventually flushed out.
1036 * It should remove the buf item from the AIL, and free the buf item.
1037 * It is called by xfs_buf_iodone_callbacks() above which will take
1038 * care of cleaning up the buffer itself.
1039 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040void
1041xfs_buf_iodone(
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001042 struct xfs_buf *bp,
1043 struct xfs_log_item *lip)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044{
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001045 struct xfs_ail *ailp = lip->li_ailp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001047 ASSERT(BUF_ITEM(lip)->bli_buf == bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
Lachlan McIlroye1f5dbd2008-09-17 16:52:13 +10001049 xfs_buf_rele(bp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050
1051 /*
1052 * If we are forcibly shutting down, this may well be
1053 * off the AIL already. That's because we simulate the
1054 * log-committed callbacks to unpin these buffers. Or we may never
1055 * have put this item on AIL because of the transaction was
David Chinner783a2f62008-10-30 17:39:58 +11001056 * aborted forcibly. xfs_trans_ail_delete() takes care of these.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 *
1058 * Either way, AIL is useless if we're forcing a shutdown.
1059 */
David Chinnerfc1829f2008-10-30 17:39:46 +11001060 spin_lock(&ailp->xa_lock);
Christoph Hellwigca30b2a2010-06-23 18:11:15 +10001061 xfs_trans_ail_delete(ailp, lip);
1062 xfs_buf_item_free(BUF_ITEM(lip));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063}