Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * 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. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * 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 | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 LT |
22 | #include "xfs_log.h" |
23 | #include "xfs_trans.h" | |
1da177e4 | 24 | #include "xfs_sb.h" |
da353b0d | 25 | #include "xfs_ag.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
a844f451 | 27 | #include "xfs_buf_item.h" |
1da177e4 | 28 | #include "xfs_trans_priv.h" |
1da177e4 | 29 | #include "xfs_error.h" |
0b1b213f | 30 | #include "xfs_trace.h" |
1da177e4 LT |
31 | |
32 | ||
33 | kmem_zone_t *xfs_buf_item_zone; | |
34 | ||
7bfa31d8 CH |
35 | static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip) |
36 | { | |
37 | return container_of(lip, struct xfs_buf_log_item, bli_item); | |
38 | } | |
39 | ||
c90821a2 | 40 | STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp); |
1da177e4 | 41 | |
166d1368 DC |
42 | static inline int |
43 | xfs_buf_log_format_size( | |
44 | struct xfs_buf_log_format *blfp) | |
45 | { | |
46 | return offsetof(struct xfs_buf_log_format, blf_data_map) + | |
47 | (blfp->blf_map_size * sizeof(blfp->blf_data_map[0])); | |
48 | } | |
49 | ||
1da177e4 LT |
50 | /* |
51 | * This returns the number of log iovecs needed to log the | |
52 | * given buf log item. | |
53 | * | |
54 | * It calculates this as 1 iovec for the buf log format structure | |
55 | * and 1 for each stretch of non-contiguous chunks to be logged. | |
56 | * Contiguous chunks are logged in a single iovec. | |
57 | * | |
58 | * If the XFS_BLI_STALE flag has been set, then log nothing. | |
59 | */ | |
166d1368 | 60 | STATIC void |
372cc85e DC |
61 | xfs_buf_item_size_segment( |
62 | struct xfs_buf_log_item *bip, | |
166d1368 DC |
63 | struct xfs_buf_log_format *blfp, |
64 | int *nvecs, | |
65 | int *nbytes) | |
1da177e4 | 66 | { |
7bfa31d8 | 67 | struct xfs_buf *bp = bip->bli_buf; |
7bfa31d8 CH |
68 | int next_bit; |
69 | int last_bit; | |
1da177e4 | 70 | |
372cc85e DC |
71 | last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
72 | if (last_bit == -1) | |
166d1368 | 73 | return; |
372cc85e DC |
74 | |
75 | /* | |
76 | * initial count for a dirty buffer is 2 vectors - the format structure | |
77 | * and the first dirty region. | |
78 | */ | |
166d1368 DC |
79 | *nvecs += 2; |
80 | *nbytes += xfs_buf_log_format_size(blfp) + XFS_BLF_CHUNK; | |
1da177e4 | 81 | |
1da177e4 LT |
82 | while (last_bit != -1) { |
83 | /* | |
84 | * This takes the bit number to start looking from and | |
85 | * returns the next set bit from there. It returns -1 | |
86 | * if there are no more bits set or the start bit is | |
87 | * beyond the end of the bitmap. | |
88 | */ | |
372cc85e DC |
89 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
90 | last_bit + 1); | |
1da177e4 LT |
91 | /* |
92 | * If we run out of bits, leave the loop, | |
93 | * else if we find a new set of bits bump the number of vecs, | |
94 | * else keep scanning the current set of bits. | |
95 | */ | |
96 | if (next_bit == -1) { | |
372cc85e | 97 | break; |
1da177e4 LT |
98 | } else if (next_bit != last_bit + 1) { |
99 | last_bit = next_bit; | |
166d1368 | 100 | (*nvecs)++; |
c1155410 DC |
101 | } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) != |
102 | (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) + | |
103 | XFS_BLF_CHUNK)) { | |
1da177e4 | 104 | last_bit = next_bit; |
166d1368 | 105 | (*nvecs)++; |
1da177e4 LT |
106 | } else { |
107 | last_bit++; | |
108 | } | |
166d1368 | 109 | *nbytes += XFS_BLF_CHUNK; |
1da177e4 | 110 | } |
1da177e4 LT |
111 | } |
112 | ||
113 | /* | |
372cc85e DC |
114 | * This returns the number of log iovecs needed to log the given buf log item. |
115 | * | |
116 | * It calculates this as 1 iovec for the buf log format structure and 1 for each | |
117 | * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged | |
118 | * in a single iovec. | |
119 | * | |
120 | * Discontiguous buffers need a format structure per region that that is being | |
121 | * logged. This makes the changes in the buffer appear to log recovery as though | |
122 | * they came from separate buffers, just like would occur if multiple buffers | |
123 | * were used instead of a single discontiguous buffer. This enables | |
124 | * discontiguous buffers to be in-memory constructs, completely transparent to | |
125 | * what ends up on disk. | |
126 | * | |
127 | * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log | |
128 | * format structures. | |
1da177e4 | 129 | */ |
166d1368 | 130 | STATIC void |
372cc85e | 131 | xfs_buf_item_size( |
166d1368 DC |
132 | struct xfs_log_item *lip, |
133 | int *nvecs, | |
134 | int *nbytes) | |
1da177e4 | 135 | { |
7bfa31d8 | 136 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
372cc85e DC |
137 | int i; |
138 | ||
139 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
140 | if (bip->bli_flags & XFS_BLI_STALE) { | |
141 | /* | |
142 | * The buffer is stale, so all we need to log | |
143 | * is the buf log format structure with the | |
144 | * cancel flag in it. | |
145 | */ | |
146 | trace_xfs_buf_item_size_stale(bip); | |
b9438173 | 147 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
166d1368 DC |
148 | *nvecs += bip->bli_format_count; |
149 | for (i = 0; i < bip->bli_format_count; i++) { | |
150 | *nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]); | |
151 | } | |
152 | return; | |
372cc85e DC |
153 | } |
154 | ||
155 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); | |
156 | ||
5f6bed76 DC |
157 | if (bip->bli_flags & XFS_BLI_ORDERED) { |
158 | /* | |
159 | * The buffer has been logged just to order it. | |
160 | * It is not being included in the transaction | |
161 | * commit, so no vectors are used at all. | |
162 | */ | |
163 | trace_xfs_buf_item_size_ordered(bip); | |
166d1368 DC |
164 | *nvecs = XFS_LOG_VEC_ORDERED; |
165 | return; | |
5f6bed76 DC |
166 | } |
167 | ||
372cc85e DC |
168 | /* |
169 | * the vector count is based on the number of buffer vectors we have | |
170 | * dirty bits in. This will only be greater than one when we have a | |
171 | * compound buffer with more than one segment dirty. Hence for compound | |
172 | * buffers we need to track which segment the dirty bits correspond to, | |
173 | * and when we move from one segment to the next increment the vector | |
174 | * count for the extra buf log format structure that will need to be | |
175 | * written. | |
176 | */ | |
372cc85e | 177 | for (i = 0; i < bip->bli_format_count; i++) { |
166d1368 DC |
178 | xfs_buf_item_size_segment(bip, &bip->bli_formats[i], |
179 | nvecs, nbytes); | |
372cc85e | 180 | } |
372cc85e | 181 | trace_xfs_buf_item_size(bip); |
372cc85e DC |
182 | } |
183 | ||
184 | static struct xfs_log_iovec * | |
185 | xfs_buf_item_format_segment( | |
186 | struct xfs_buf_log_item *bip, | |
187 | struct xfs_log_iovec *vecp, | |
188 | uint offset, | |
189 | struct xfs_buf_log_format *blfp) | |
190 | { | |
7bfa31d8 | 191 | struct xfs_buf *bp = bip->bli_buf; |
1da177e4 LT |
192 | uint base_size; |
193 | uint nvecs; | |
1da177e4 LT |
194 | int first_bit; |
195 | int last_bit; | |
196 | int next_bit; | |
197 | uint nbits; | |
198 | uint buffer_offset; | |
199 | ||
372cc85e | 200 | /* copy the flags across from the base format item */ |
b9438173 | 201 | blfp->blf_flags = bip->__bli_format.blf_flags; |
1da177e4 LT |
202 | |
203 | /* | |
77c1a08f DC |
204 | * Base size is the actual size of the ondisk structure - it reflects |
205 | * the actual size of the dirty bitmap rather than the size of the in | |
206 | * memory structure. | |
1da177e4 | 207 | */ |
166d1368 | 208 | base_size = xfs_buf_log_format_size(blfp); |
820a554f MT |
209 | |
210 | nvecs = 0; | |
211 | first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); | |
212 | if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) { | |
213 | /* | |
214 | * If the map is not be dirty in the transaction, mark | |
215 | * the size as zero and do not advance the vector pointer. | |
216 | */ | |
217 | goto out; | |
218 | } | |
219 | ||
372cc85e | 220 | vecp->i_addr = blfp; |
1da177e4 | 221 | vecp->i_len = base_size; |
4139b3b3 | 222 | vecp->i_type = XLOG_REG_TYPE_BFORMAT; |
1da177e4 LT |
223 | vecp++; |
224 | nvecs = 1; | |
225 | ||
226 | if (bip->bli_flags & XFS_BLI_STALE) { | |
227 | /* | |
228 | * The buffer is stale, so all we need to log | |
229 | * is the buf log format structure with the | |
230 | * cancel flag in it. | |
231 | */ | |
0b1b213f | 232 | trace_xfs_buf_item_format_stale(bip); |
372cc85e | 233 | ASSERT(blfp->blf_flags & XFS_BLF_CANCEL); |
820a554f | 234 | goto out; |
1da177e4 LT |
235 | } |
236 | ||
5f6bed76 | 237 | |
1da177e4 LT |
238 | /* |
239 | * Fill in an iovec for each set of contiguous chunks. | |
240 | */ | |
820a554f | 241 | |
1da177e4 LT |
242 | last_bit = first_bit; |
243 | nbits = 1; | |
244 | for (;;) { | |
245 | /* | |
246 | * This takes the bit number to start looking from and | |
247 | * returns the next set bit from there. It returns -1 | |
248 | * if there are no more bits set or the start bit is | |
249 | * beyond the end of the bitmap. | |
250 | */ | |
372cc85e DC |
251 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
252 | (uint)last_bit + 1); | |
1da177e4 LT |
253 | /* |
254 | * If we run out of bits fill in the last iovec and get | |
255 | * out of the loop. | |
256 | * Else if we start a new set of bits then fill in the | |
257 | * iovec for the series we were looking at and start | |
258 | * counting the bits in the new one. | |
259 | * Else we're still in the same set of bits so just | |
260 | * keep counting and scanning. | |
261 | */ | |
262 | if (next_bit == -1) { | |
372cc85e | 263 | buffer_offset = offset + first_bit * XFS_BLF_CHUNK; |
1da177e4 | 264 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
c1155410 | 265 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
4139b3b3 | 266 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
1da177e4 LT |
267 | nvecs++; |
268 | break; | |
269 | } else if (next_bit != last_bit + 1) { | |
372cc85e | 270 | buffer_offset = offset + first_bit * XFS_BLF_CHUNK; |
1da177e4 | 271 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
c1155410 | 272 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
4139b3b3 | 273 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
1da177e4 LT |
274 | nvecs++; |
275 | vecp++; | |
276 | first_bit = next_bit; | |
277 | last_bit = next_bit; | |
278 | nbits = 1; | |
372cc85e DC |
279 | } else if (xfs_buf_offset(bp, offset + |
280 | (next_bit << XFS_BLF_SHIFT)) != | |
281 | (xfs_buf_offset(bp, offset + | |
282 | (last_bit << XFS_BLF_SHIFT)) + | |
c1155410 | 283 | XFS_BLF_CHUNK)) { |
372cc85e | 284 | buffer_offset = offset + first_bit * XFS_BLF_CHUNK; |
1da177e4 | 285 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
c1155410 | 286 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
4139b3b3 | 287 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
709da6a6 | 288 | nvecs++; |
1da177e4 LT |
289 | vecp++; |
290 | first_bit = next_bit; | |
291 | last_bit = next_bit; | |
292 | nbits = 1; | |
293 | } else { | |
294 | last_bit++; | |
295 | nbits++; | |
296 | } | |
297 | } | |
820a554f MT |
298 | out: |
299 | blfp->blf_size = nvecs; | |
372cc85e DC |
300 | return vecp; |
301 | } | |
302 | ||
303 | /* | |
304 | * This is called to fill in the vector of log iovecs for the | |
305 | * given log buf item. It fills the first entry with a buf log | |
306 | * format structure, and the rest point to contiguous chunks | |
307 | * within the buffer. | |
308 | */ | |
309 | STATIC void | |
310 | xfs_buf_item_format( | |
311 | struct xfs_log_item *lip, | |
312 | struct xfs_log_iovec *vecp) | |
313 | { | |
314 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); | |
315 | struct xfs_buf *bp = bip->bli_buf; | |
316 | uint offset = 0; | |
317 | int i; | |
318 | ||
319 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
320 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
321 | (bip->bli_flags & XFS_BLI_STALE)); | |
322 | ||
323 | /* | |
324 | * If it is an inode buffer, transfer the in-memory state to the | |
ddf6ad01 DC |
325 | * format flags and clear the in-memory state. |
326 | * | |
327 | * For buffer based inode allocation, we do not transfer | |
372cc85e DC |
328 | * this state if the inode buffer allocation has not yet been committed |
329 | * to the log as setting the XFS_BLI_INODE_BUF flag will prevent | |
330 | * correct replay of the inode allocation. | |
ddf6ad01 DC |
331 | * |
332 | * For icreate item based inode allocation, the buffers aren't written | |
333 | * to the journal during allocation, and hence we should always tag the | |
334 | * buffer as an inode buffer so that the correct unlinked list replay | |
335 | * occurs during recovery. | |
372cc85e DC |
336 | */ |
337 | if (bip->bli_flags & XFS_BLI_INODE_BUF) { | |
ddf6ad01 DC |
338 | if (xfs_sb_version_hascrc(&lip->li_mountp->m_sb) || |
339 | !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && | |
372cc85e | 340 | xfs_log_item_in_current_chkpt(lip))) |
b9438173 | 341 | bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF; |
372cc85e DC |
342 | bip->bli_flags &= ~XFS_BLI_INODE_BUF; |
343 | } | |
344 | ||
5f6bed76 DC |
345 | if ((bip->bli_flags & (XFS_BLI_ORDERED|XFS_BLI_STALE)) == |
346 | XFS_BLI_ORDERED) { | |
347 | /* | |
348 | * The buffer has been logged just to order it. It is not being | |
349 | * included in the transaction commit, so don't format it. | |
350 | */ | |
351 | trace_xfs_buf_item_format_ordered(bip); | |
352 | return; | |
353 | } | |
354 | ||
372cc85e DC |
355 | for (i = 0; i < bip->bli_format_count; i++) { |
356 | vecp = xfs_buf_item_format_segment(bip, vecp, offset, | |
357 | &bip->bli_formats[i]); | |
358 | offset += bp->b_maps[i].bm_len; | |
359 | } | |
1da177e4 LT |
360 | |
361 | /* | |
362 | * Check to make sure everything is consistent. | |
363 | */ | |
0b1b213f | 364 | trace_xfs_buf_item_format(bip); |
1da177e4 LT |
365 | } |
366 | ||
367 | /* | |
64fc35de | 368 | * This is called to pin the buffer associated with the buf log item in memory |
4d16e924 | 369 | * so it cannot be written out. |
64fc35de DC |
370 | * |
371 | * We also always take a reference to the buffer log item here so that the bli | |
372 | * is held while the item is pinned in memory. This means that we can | |
373 | * unconditionally drop the reference count a transaction holds when the | |
374 | * transaction is completed. | |
1da177e4 | 375 | */ |
ba0f32d4 | 376 | STATIC void |
1da177e4 | 377 | xfs_buf_item_pin( |
7bfa31d8 | 378 | struct xfs_log_item *lip) |
1da177e4 | 379 | { |
7bfa31d8 | 380 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
1da177e4 | 381 | |
1da177e4 LT |
382 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
383 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
5f6bed76 | 384 | (bip->bli_flags & XFS_BLI_ORDERED) || |
1da177e4 | 385 | (bip->bli_flags & XFS_BLI_STALE)); |
7bfa31d8 | 386 | |
0b1b213f | 387 | trace_xfs_buf_item_pin(bip); |
4d16e924 CH |
388 | |
389 | atomic_inc(&bip->bli_refcount); | |
390 | atomic_inc(&bip->bli_buf->b_pin_count); | |
1da177e4 LT |
391 | } |
392 | ||
1da177e4 LT |
393 | /* |
394 | * This is called to unpin the buffer associated with the buf log | |
395 | * item which was previously pinned with a call to xfs_buf_item_pin(). | |
1da177e4 LT |
396 | * |
397 | * Also drop the reference to the buf item for the current transaction. | |
398 | * If the XFS_BLI_STALE flag is set and we are the last reference, | |
399 | * then free up the buf log item and unlock the buffer. | |
9412e318 CH |
400 | * |
401 | * If the remove flag is set we are called from uncommit in the | |
402 | * forced-shutdown path. If that is true and the reference count on | |
403 | * the log item is going to drop to zero we need to free the item's | |
404 | * descriptor in the transaction. | |
1da177e4 | 405 | */ |
ba0f32d4 | 406 | STATIC void |
1da177e4 | 407 | xfs_buf_item_unpin( |
7bfa31d8 | 408 | struct xfs_log_item *lip, |
9412e318 | 409 | int remove) |
1da177e4 | 410 | { |
7bfa31d8 | 411 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
9412e318 | 412 | xfs_buf_t *bp = bip->bli_buf; |
7bfa31d8 | 413 | struct xfs_ail *ailp = lip->li_ailp; |
8e123850 | 414 | int stale = bip->bli_flags & XFS_BLI_STALE; |
7bfa31d8 | 415 | int freed; |
1da177e4 | 416 | |
adadbeef | 417 | ASSERT(bp->b_fspriv == bip); |
1da177e4 | 418 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
9412e318 | 419 | |
0b1b213f | 420 | trace_xfs_buf_item_unpin(bip); |
1da177e4 LT |
421 | |
422 | freed = atomic_dec_and_test(&bip->bli_refcount); | |
4d16e924 CH |
423 | |
424 | if (atomic_dec_and_test(&bp->b_pin_count)) | |
425 | wake_up_all(&bp->b_waiters); | |
7bfa31d8 | 426 | |
1da177e4 LT |
427 | if (freed && stale) { |
428 | ASSERT(bip->bli_flags & XFS_BLI_STALE); | |
0c842ad4 | 429 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 430 | ASSERT(XFS_BUF_ISSTALE(bp)); |
b9438173 | 431 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
9412e318 | 432 | |
0b1b213f CH |
433 | trace_xfs_buf_item_unpin_stale(bip); |
434 | ||
9412e318 CH |
435 | if (remove) { |
436 | /* | |
e34a314c DC |
437 | * If we are in a transaction context, we have to |
438 | * remove the log item from the transaction as we are | |
439 | * about to release our reference to the buffer. If we | |
440 | * don't, the unlock that occurs later in | |
441 | * xfs_trans_uncommit() will try to reference the | |
9412e318 CH |
442 | * buffer which we no longer have a hold on. |
443 | */ | |
e34a314c DC |
444 | if (lip->li_desc) |
445 | xfs_trans_del_item(lip); | |
9412e318 CH |
446 | |
447 | /* | |
448 | * Since the transaction no longer refers to the buffer, | |
449 | * the buffer should no longer refer to the transaction. | |
450 | */ | |
bf9d9013 | 451 | bp->b_transp = NULL; |
9412e318 CH |
452 | } |
453 | ||
1da177e4 LT |
454 | /* |
455 | * If we get called here because of an IO error, we may | |
783a2f65 | 456 | * or may not have the item on the AIL. xfs_trans_ail_delete() |
1da177e4 | 457 | * will take care of that situation. |
783a2f65 | 458 | * xfs_trans_ail_delete() drops the AIL lock. |
1da177e4 LT |
459 | */ |
460 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { | |
c90821a2 | 461 | xfs_buf_do_callbacks(bp); |
adadbeef | 462 | bp->b_fspriv = NULL; |
cb669ca5 | 463 | bp->b_iodone = NULL; |
1da177e4 | 464 | } else { |
783a2f65 | 465 | spin_lock(&ailp->xa_lock); |
04913fdd | 466 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR); |
1da177e4 | 467 | xfs_buf_item_relse(bp); |
adadbeef | 468 | ASSERT(bp->b_fspriv == NULL); |
1da177e4 LT |
469 | } |
470 | xfs_buf_relse(bp); | |
960c60af | 471 | } else if (freed && remove) { |
137fff09 DC |
472 | /* |
473 | * There are currently two references to the buffer - the active | |
474 | * LRU reference and the buf log item. What we are about to do | |
475 | * here - simulate a failed IO completion - requires 3 | |
476 | * references. | |
477 | * | |
478 | * The LRU reference is removed by the xfs_buf_stale() call. The | |
479 | * buf item reference is removed by the xfs_buf_iodone() | |
480 | * callback that is run by xfs_buf_do_callbacks() during ioend | |
481 | * processing (via the bp->b_iodone callback), and then finally | |
482 | * the ioend processing will drop the IO reference if the buffer | |
483 | * is marked XBF_ASYNC. | |
484 | * | |
485 | * Hence we need to take an additional reference here so that IO | |
486 | * completion processing doesn't free the buffer prematurely. | |
487 | */ | |
960c60af | 488 | xfs_buf_lock(bp); |
137fff09 DC |
489 | xfs_buf_hold(bp); |
490 | bp->b_flags |= XBF_ASYNC; | |
960c60af CH |
491 | xfs_buf_ioerror(bp, EIO); |
492 | XFS_BUF_UNDONE(bp); | |
493 | xfs_buf_stale(bp); | |
494 | xfs_buf_ioend(bp, 0); | |
1da177e4 LT |
495 | } |
496 | } | |
497 | ||
ba0f32d4 | 498 | STATIC uint |
43ff2122 CH |
499 | xfs_buf_item_push( |
500 | struct xfs_log_item *lip, | |
501 | struct list_head *buffer_list) | |
1da177e4 | 502 | { |
7bfa31d8 CH |
503 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
504 | struct xfs_buf *bp = bip->bli_buf; | |
43ff2122 | 505 | uint rval = XFS_ITEM_SUCCESS; |
1da177e4 | 506 | |
811e64c7 | 507 | if (xfs_buf_ispinned(bp)) |
1da177e4 | 508 | return XFS_ITEM_PINNED; |
5337fe9b BF |
509 | if (!xfs_buf_trylock(bp)) { |
510 | /* | |
511 | * If we have just raced with a buffer being pinned and it has | |
512 | * been marked stale, we could end up stalling until someone else | |
513 | * issues a log force to unpin the stale buffer. Check for the | |
514 | * race condition here so xfsaild recognizes the buffer is pinned | |
515 | * and queues a log force to move it along. | |
516 | */ | |
517 | if (xfs_buf_ispinned(bp)) | |
518 | return XFS_ITEM_PINNED; | |
1da177e4 | 519 | return XFS_ITEM_LOCKED; |
5337fe9b | 520 | } |
1da177e4 | 521 | |
1da177e4 | 522 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
43ff2122 CH |
523 | |
524 | trace_xfs_buf_item_push(bip); | |
525 | ||
526 | if (!xfs_buf_delwri_queue(bp, buffer_list)) | |
527 | rval = XFS_ITEM_FLUSHING; | |
528 | xfs_buf_unlock(bp); | |
529 | return rval; | |
1da177e4 LT |
530 | } |
531 | ||
532 | /* | |
64fc35de DC |
533 | * Release the buffer associated with the buf log item. If there is no dirty |
534 | * logged data associated with the buffer recorded in the buf log item, then | |
535 | * free the buf log item and remove the reference to it in the buffer. | |
1da177e4 | 536 | * |
64fc35de DC |
537 | * This call ignores the recursion count. It is only called when the buffer |
538 | * should REALLY be unlocked, regardless of the recursion count. | |
1da177e4 | 539 | * |
64fc35de DC |
540 | * We unconditionally drop the transaction's reference to the log item. If the |
541 | * item was logged, then another reference was taken when it was pinned, so we | |
542 | * can safely drop the transaction reference now. This also allows us to avoid | |
543 | * potential races with the unpin code freeing the bli by not referencing the | |
544 | * bli after we've dropped the reference count. | |
545 | * | |
546 | * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item | |
547 | * if necessary but do not unlock the buffer. This is for support of | |
548 | * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't | |
549 | * free the item. | |
1da177e4 | 550 | */ |
ba0f32d4 | 551 | STATIC void |
1da177e4 | 552 | xfs_buf_item_unlock( |
7bfa31d8 | 553 | struct xfs_log_item *lip) |
1da177e4 | 554 | { |
7bfa31d8 CH |
555 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
556 | struct xfs_buf *bp = bip->bli_buf; | |
5f6bed76 DC |
557 | bool clean; |
558 | bool aborted; | |
559 | int flags; | |
1da177e4 | 560 | |
64fc35de | 561 | /* Clear the buffer's association with this transaction. */ |
bf9d9013 | 562 | bp->b_transp = NULL; |
1da177e4 LT |
563 | |
564 | /* | |
64fc35de DC |
565 | * If this is a transaction abort, don't return early. Instead, allow |
566 | * the brelse to happen. Normally it would be done for stale | |
567 | * (cancelled) buffers at unpin time, but we'll never go through the | |
568 | * pin/unpin cycle if we abort inside commit. | |
1da177e4 | 569 | */ |
5f6bed76 | 570 | aborted = (lip->li_flags & XFS_LI_ABORTED) ? true : false; |
1da177e4 | 571 | /* |
5f6bed76 DC |
572 | * Before possibly freeing the buf item, copy the per-transaction state |
573 | * so we can reference it safely later after clearing it from the | |
574 | * buffer log item. | |
64fc35de | 575 | */ |
5f6bed76 DC |
576 | flags = bip->bli_flags; |
577 | bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED); | |
64fc35de DC |
578 | |
579 | /* | |
580 | * If the buf item is marked stale, then don't do anything. We'll | |
581 | * unlock the buffer and free the buf item when the buffer is unpinned | |
582 | * for the last time. | |
1da177e4 | 583 | */ |
5f6bed76 | 584 | if (flags & XFS_BLI_STALE) { |
0b1b213f | 585 | trace_xfs_buf_item_unlock_stale(bip); |
b9438173 | 586 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
64fc35de DC |
587 | if (!aborted) { |
588 | atomic_dec(&bip->bli_refcount); | |
1da177e4 | 589 | return; |
64fc35de | 590 | } |
1da177e4 LT |
591 | } |
592 | ||
0b1b213f | 593 | trace_xfs_buf_item_unlock(bip); |
1da177e4 LT |
594 | |
595 | /* | |
64fc35de | 596 | * If the buf item isn't tracking any data, free it, otherwise drop the |
3b19034d DC |
597 | * reference we hold to it. If we are aborting the transaction, this may |
598 | * be the only reference to the buf item, so we free it anyway | |
599 | * regardless of whether it is dirty or not. A dirty abort implies a | |
600 | * shutdown, anyway. | |
5f6bed76 DC |
601 | * |
602 | * Ordered buffers are dirty but may have no recorded changes, so ensure | |
603 | * we only release clean items here. | |
1da177e4 | 604 | */ |
5f6bed76 DC |
605 | clean = (flags & XFS_BLI_DIRTY) ? false : true; |
606 | if (clean) { | |
607 | int i; | |
608 | for (i = 0; i < bip->bli_format_count; i++) { | |
609 | if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map, | |
610 | bip->bli_formats[i].blf_map_size)) { | |
611 | clean = false; | |
612 | break; | |
613 | } | |
c883d0c4 MT |
614 | } |
615 | } | |
2ad01f53 | 616 | if (clean || aborted) { |
3b19034d | 617 | if (atomic_dec_and_test(&bip->bli_refcount)) { |
2ad01f53 | 618 | ASSERT(!aborted || XFS_FORCED_SHUTDOWN(lip->li_mountp)); |
3b19034d DC |
619 | xfs_buf_item_relse(bp); |
620 | } | |
621 | } else | |
64fc35de | 622 | atomic_dec(&bip->bli_refcount); |
1da177e4 | 623 | |
5f6bed76 | 624 | if (!(flags & XFS_BLI_HOLD)) |
1da177e4 | 625 | xfs_buf_relse(bp); |
1da177e4 LT |
626 | } |
627 | ||
628 | /* | |
629 | * This is called to find out where the oldest active copy of the | |
630 | * buf log item in the on disk log resides now that the last log | |
631 | * write of it completed at the given lsn. | |
632 | * We always re-log all the dirty data in a buffer, so usually the | |
633 | * latest copy in the on disk log is the only one that matters. For | |
634 | * those cases we simply return the given lsn. | |
635 | * | |
636 | * The one exception to this is for buffers full of newly allocated | |
637 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF | |
638 | * flag set, indicating that only the di_next_unlinked fields from the | |
639 | * inodes in the buffers will be replayed during recovery. If the | |
640 | * original newly allocated inode images have not yet been flushed | |
641 | * when the buffer is so relogged, then we need to make sure that we | |
642 | * keep the old images in the 'active' portion of the log. We do this | |
643 | * by returning the original lsn of that transaction here rather than | |
644 | * the current one. | |
645 | */ | |
ba0f32d4 | 646 | STATIC xfs_lsn_t |
1da177e4 | 647 | xfs_buf_item_committed( |
7bfa31d8 | 648 | struct xfs_log_item *lip, |
1da177e4 LT |
649 | xfs_lsn_t lsn) |
650 | { | |
7bfa31d8 CH |
651 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
652 | ||
0b1b213f CH |
653 | trace_xfs_buf_item_committed(bip); |
654 | ||
7bfa31d8 CH |
655 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0) |
656 | return lip->li_lsn; | |
657 | return lsn; | |
1da177e4 LT |
658 | } |
659 | ||
ba0f32d4 | 660 | STATIC void |
7bfa31d8 CH |
661 | xfs_buf_item_committing( |
662 | struct xfs_log_item *lip, | |
663 | xfs_lsn_t commit_lsn) | |
1da177e4 LT |
664 | { |
665 | } | |
666 | ||
667 | /* | |
668 | * This is the ops vector shared by all buf log items. | |
669 | */ | |
272e42b2 | 670 | static const struct xfs_item_ops xfs_buf_item_ops = { |
7bfa31d8 CH |
671 | .iop_size = xfs_buf_item_size, |
672 | .iop_format = xfs_buf_item_format, | |
673 | .iop_pin = xfs_buf_item_pin, | |
674 | .iop_unpin = xfs_buf_item_unpin, | |
7bfa31d8 CH |
675 | .iop_unlock = xfs_buf_item_unlock, |
676 | .iop_committed = xfs_buf_item_committed, | |
677 | .iop_push = xfs_buf_item_push, | |
7bfa31d8 | 678 | .iop_committing = xfs_buf_item_committing |
1da177e4 LT |
679 | }; |
680 | ||
372cc85e DC |
681 | STATIC int |
682 | xfs_buf_item_get_format( | |
683 | struct xfs_buf_log_item *bip, | |
684 | int count) | |
685 | { | |
686 | ASSERT(bip->bli_formats == NULL); | |
687 | bip->bli_format_count = count; | |
688 | ||
689 | if (count == 1) { | |
b9438173 | 690 | bip->bli_formats = &bip->__bli_format; |
372cc85e DC |
691 | return 0; |
692 | } | |
693 | ||
694 | bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format), | |
695 | KM_SLEEP); | |
696 | if (!bip->bli_formats) | |
697 | return ENOMEM; | |
698 | return 0; | |
699 | } | |
700 | ||
701 | STATIC void | |
702 | xfs_buf_item_free_format( | |
703 | struct xfs_buf_log_item *bip) | |
704 | { | |
b9438173 | 705 | if (bip->bli_formats != &bip->__bli_format) { |
372cc85e DC |
706 | kmem_free(bip->bli_formats); |
707 | bip->bli_formats = NULL; | |
708 | } | |
709 | } | |
1da177e4 LT |
710 | |
711 | /* | |
712 | * Allocate a new buf log item to go with the given buffer. | |
713 | * Set the buffer's b_fsprivate field to point to the new | |
714 | * buf log item. If there are other item's attached to the | |
715 | * buffer (see xfs_buf_attach_iodone() below), then put the | |
716 | * buf log item at the front. | |
717 | */ | |
718 | void | |
719 | xfs_buf_item_init( | |
720 | xfs_buf_t *bp, | |
721 | xfs_mount_t *mp) | |
722 | { | |
adadbeef | 723 | xfs_log_item_t *lip = bp->b_fspriv; |
1da177e4 LT |
724 | xfs_buf_log_item_t *bip; |
725 | int chunks; | |
726 | int map_size; | |
372cc85e DC |
727 | int error; |
728 | int i; | |
1da177e4 LT |
729 | |
730 | /* | |
731 | * Check to see if there is already a buf log item for | |
732 | * this buffer. If there is, it is guaranteed to be | |
733 | * the first. If we do already have one, there is | |
734 | * nothing to do here so return. | |
735 | */ | |
ebad861b | 736 | ASSERT(bp->b_target->bt_mount == mp); |
adadbeef CH |
737 | if (lip != NULL && lip->li_type == XFS_LI_BUF) |
738 | return; | |
1da177e4 | 739 | |
372cc85e | 740 | bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP); |
43f5efc5 | 741 | xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops); |
1da177e4 | 742 | bip->bli_buf = bp; |
e1f5dbd7 | 743 | xfs_buf_hold(bp); |
372cc85e DC |
744 | |
745 | /* | |
746 | * chunks is the number of XFS_BLF_CHUNK size pieces the buffer | |
747 | * can be divided into. Make sure not to truncate any pieces. | |
748 | * map_size is the size of the bitmap needed to describe the | |
749 | * chunks of the buffer. | |
750 | * | |
751 | * Discontiguous buffer support follows the layout of the underlying | |
752 | * buffer. This makes the implementation as simple as possible. | |
753 | */ | |
754 | error = xfs_buf_item_get_format(bip, bp->b_map_count); | |
755 | ASSERT(error == 0); | |
756 | ||
757 | for (i = 0; i < bip->bli_format_count; i++) { | |
758 | chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len), | |
759 | XFS_BLF_CHUNK); | |
760 | map_size = DIV_ROUND_UP(chunks, NBWORD); | |
761 | ||
762 | bip->bli_formats[i].blf_type = XFS_LI_BUF; | |
763 | bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn; | |
764 | bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len; | |
765 | bip->bli_formats[i].blf_map_size = map_size; | |
766 | } | |
1da177e4 LT |
767 | |
768 | #ifdef XFS_TRANS_DEBUG | |
769 | /* | |
770 | * Allocate the arrays for tracking what needs to be logged | |
771 | * and what our callers request to be logged. bli_orig | |
772 | * holds a copy of the original, clean buffer for comparison | |
773 | * against, and bli_logged keeps a 1 bit flag per byte in | |
774 | * the buffer to indicate which bytes the callers have asked | |
775 | * to have logged. | |
776 | */ | |
aa0e8833 DC |
777 | bip->bli_orig = kmem_alloc(BBTOB(bp->b_length), KM_SLEEP); |
778 | memcpy(bip->bli_orig, bp->b_addr, BBTOB(bp->b_length)); | |
779 | bip->bli_logged = kmem_zalloc(BBTOB(bp->b_length) / NBBY, KM_SLEEP); | |
1da177e4 LT |
780 | #endif |
781 | ||
782 | /* | |
783 | * Put the buf item into the list of items attached to the | |
784 | * buffer at the front. | |
785 | */ | |
adadbeef CH |
786 | if (bp->b_fspriv) |
787 | bip->bli_item.li_bio_list = bp->b_fspriv; | |
788 | bp->b_fspriv = bip; | |
1da177e4 LT |
789 | } |
790 | ||
791 | ||
792 | /* | |
793 | * Mark bytes first through last inclusive as dirty in the buf | |
794 | * item's bitmap. | |
795 | */ | |
796 | void | |
372cc85e DC |
797 | xfs_buf_item_log_segment( |
798 | struct xfs_buf_log_item *bip, | |
1da177e4 | 799 | uint first, |
372cc85e DC |
800 | uint last, |
801 | uint *map) | |
1da177e4 LT |
802 | { |
803 | uint first_bit; | |
804 | uint last_bit; | |
805 | uint bits_to_set; | |
806 | uint bits_set; | |
807 | uint word_num; | |
808 | uint *wordp; | |
809 | uint bit; | |
810 | uint end_bit; | |
811 | uint mask; | |
812 | ||
1da177e4 LT |
813 | /* |
814 | * Convert byte offsets to bit numbers. | |
815 | */ | |
c1155410 DC |
816 | first_bit = first >> XFS_BLF_SHIFT; |
817 | last_bit = last >> XFS_BLF_SHIFT; | |
1da177e4 LT |
818 | |
819 | /* | |
820 | * Calculate the total number of bits to be set. | |
821 | */ | |
822 | bits_to_set = last_bit - first_bit + 1; | |
823 | ||
824 | /* | |
825 | * Get a pointer to the first word in the bitmap | |
826 | * to set a bit in. | |
827 | */ | |
828 | word_num = first_bit >> BIT_TO_WORD_SHIFT; | |
372cc85e | 829 | wordp = &map[word_num]; |
1da177e4 LT |
830 | |
831 | /* | |
832 | * Calculate the starting bit in the first word. | |
833 | */ | |
834 | bit = first_bit & (uint)(NBWORD - 1); | |
835 | ||
836 | /* | |
837 | * First set any bits in the first word of our range. | |
838 | * If it starts at bit 0 of the word, it will be | |
839 | * set below rather than here. That is what the variable | |
840 | * bit tells us. The variable bits_set tracks the number | |
841 | * of bits that have been set so far. End_bit is the number | |
842 | * of the last bit to be set in this word plus one. | |
843 | */ | |
844 | if (bit) { | |
845 | end_bit = MIN(bit + bits_to_set, (uint)NBWORD); | |
846 | mask = ((1 << (end_bit - bit)) - 1) << bit; | |
847 | *wordp |= mask; | |
848 | wordp++; | |
849 | bits_set = end_bit - bit; | |
850 | } else { | |
851 | bits_set = 0; | |
852 | } | |
853 | ||
854 | /* | |
855 | * Now set bits a whole word at a time that are between | |
856 | * first_bit and last_bit. | |
857 | */ | |
858 | while ((bits_to_set - bits_set) >= NBWORD) { | |
859 | *wordp |= 0xffffffff; | |
860 | bits_set += NBWORD; | |
861 | wordp++; | |
862 | } | |
863 | ||
864 | /* | |
865 | * Finally, set any bits left to be set in one last partial word. | |
866 | */ | |
867 | end_bit = bits_to_set - bits_set; | |
868 | if (end_bit) { | |
869 | mask = (1 << end_bit) - 1; | |
870 | *wordp |= mask; | |
871 | } | |
1da177e4 LT |
872 | } |
873 | ||
372cc85e DC |
874 | /* |
875 | * Mark bytes first through last inclusive as dirty in the buf | |
876 | * item's bitmap. | |
877 | */ | |
878 | void | |
879 | xfs_buf_item_log( | |
880 | xfs_buf_log_item_t *bip, | |
881 | uint first, | |
882 | uint last) | |
883 | { | |
884 | int i; | |
885 | uint start; | |
886 | uint end; | |
887 | struct xfs_buf *bp = bip->bli_buf; | |
888 | ||
372cc85e DC |
889 | /* |
890 | * walk each buffer segment and mark them dirty appropriately. | |
891 | */ | |
892 | start = 0; | |
893 | for (i = 0; i < bip->bli_format_count; i++) { | |
894 | if (start > last) | |
895 | break; | |
896 | end = start + BBTOB(bp->b_maps[i].bm_len); | |
897 | if (first > end) { | |
898 | start += BBTOB(bp->b_maps[i].bm_len); | |
899 | continue; | |
900 | } | |
901 | if (first < start) | |
902 | first = start; | |
903 | if (end > last) | |
904 | end = last; | |
905 | ||
906 | xfs_buf_item_log_segment(bip, first, end, | |
907 | &bip->bli_formats[i].blf_data_map[0]); | |
908 | ||
909 | start += bp->b_maps[i].bm_len; | |
910 | } | |
911 | } | |
912 | ||
1da177e4 LT |
913 | |
914 | /* | |
5f6bed76 | 915 | * Return 1 if the buffer has been logged or ordered in a transaction (at any |
1da177e4 LT |
916 | * point, not just the current transaction) and 0 if not. |
917 | */ | |
918 | uint | |
919 | xfs_buf_item_dirty( | |
920 | xfs_buf_log_item_t *bip) | |
921 | { | |
922 | return (bip->bli_flags & XFS_BLI_DIRTY); | |
923 | } | |
924 | ||
e1f5dbd7 LM |
925 | STATIC void |
926 | xfs_buf_item_free( | |
927 | xfs_buf_log_item_t *bip) | |
928 | { | |
929 | #ifdef XFS_TRANS_DEBUG | |
930 | kmem_free(bip->bli_orig); | |
931 | kmem_free(bip->bli_logged); | |
932 | #endif /* XFS_TRANS_DEBUG */ | |
933 | ||
372cc85e | 934 | xfs_buf_item_free_format(bip); |
e1f5dbd7 LM |
935 | kmem_zone_free(xfs_buf_item_zone, bip); |
936 | } | |
937 | ||
1da177e4 LT |
938 | /* |
939 | * This is called when the buf log item is no longer needed. It should | |
940 | * free the buf log item associated with the given buffer and clear | |
941 | * the buffer's pointer to the buf log item. If there are no more | |
942 | * items in the list, clear the b_iodone field of the buffer (see | |
943 | * xfs_buf_attach_iodone() below). | |
944 | */ | |
945 | void | |
946 | xfs_buf_item_relse( | |
947 | xfs_buf_t *bp) | |
948 | { | |
5f6bed76 | 949 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 950 | |
0b1b213f | 951 | trace_xfs_buf_item_relse(bp, _RET_IP_); |
5f6bed76 | 952 | ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL)); |
0b1b213f | 953 | |
adadbeef | 954 | bp->b_fspriv = bip->bli_item.li_bio_list; |
cb669ca5 CH |
955 | if (bp->b_fspriv == NULL) |
956 | bp->b_iodone = NULL; | |
adadbeef | 957 | |
e1f5dbd7 LM |
958 | xfs_buf_rele(bp); |
959 | xfs_buf_item_free(bip); | |
1da177e4 LT |
960 | } |
961 | ||
962 | ||
963 | /* | |
964 | * Add the given log item with its callback to the list of callbacks | |
965 | * to be called when the buffer's I/O completes. If it is not set | |
966 | * already, set the buffer's b_iodone() routine to be | |
967 | * xfs_buf_iodone_callbacks() and link the log item into the list of | |
968 | * items rooted at b_fsprivate. Items are always added as the second | |
969 | * entry in the list if there is a first, because the buf item code | |
970 | * assumes that the buf log item is first. | |
971 | */ | |
972 | void | |
973 | xfs_buf_attach_iodone( | |
974 | xfs_buf_t *bp, | |
975 | void (*cb)(xfs_buf_t *, xfs_log_item_t *), | |
976 | xfs_log_item_t *lip) | |
977 | { | |
978 | xfs_log_item_t *head_lip; | |
979 | ||
0c842ad4 | 980 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
981 | |
982 | lip->li_cb = cb; | |
adadbeef CH |
983 | head_lip = bp->b_fspriv; |
984 | if (head_lip) { | |
1da177e4 LT |
985 | lip->li_bio_list = head_lip->li_bio_list; |
986 | head_lip->li_bio_list = lip; | |
987 | } else { | |
adadbeef | 988 | bp->b_fspriv = lip; |
1da177e4 LT |
989 | } |
990 | ||
cb669ca5 CH |
991 | ASSERT(bp->b_iodone == NULL || |
992 | bp->b_iodone == xfs_buf_iodone_callbacks); | |
993 | bp->b_iodone = xfs_buf_iodone_callbacks; | |
1da177e4 LT |
994 | } |
995 | ||
c90821a2 DC |
996 | /* |
997 | * We can have many callbacks on a buffer. Running the callbacks individually | |
998 | * can cause a lot of contention on the AIL lock, so we allow for a single | |
999 | * callback to be able to scan the remaining lip->li_bio_list for other items | |
1000 | * of the same type and callback to be processed in the first call. | |
1001 | * | |
1002 | * As a result, the loop walking the callback list below will also modify the | |
1003 | * list. it removes the first item from the list and then runs the callback. | |
1004 | * The loop then restarts from the new head of the list. This allows the | |
1005 | * callback to scan and modify the list attached to the buffer and we don't | |
1006 | * have to care about maintaining a next item pointer. | |
1007 | */ | |
1da177e4 LT |
1008 | STATIC void |
1009 | xfs_buf_do_callbacks( | |
c90821a2 | 1010 | struct xfs_buf *bp) |
1da177e4 | 1011 | { |
c90821a2 | 1012 | struct xfs_log_item *lip; |
1da177e4 | 1013 | |
adadbeef CH |
1014 | while ((lip = bp->b_fspriv) != NULL) { |
1015 | bp->b_fspriv = lip->li_bio_list; | |
1da177e4 LT |
1016 | ASSERT(lip->li_cb != NULL); |
1017 | /* | |
1018 | * Clear the next pointer so we don't have any | |
1019 | * confusion if the item is added to another buf. | |
1020 | * Don't touch the log item after calling its | |
1021 | * callback, because it could have freed itself. | |
1022 | */ | |
1023 | lip->li_bio_list = NULL; | |
1024 | lip->li_cb(bp, lip); | |
1da177e4 LT |
1025 | } |
1026 | } | |
1027 | ||
1028 | /* | |
1029 | * This is the iodone() function for buffers which have had callbacks | |
1030 | * attached to them by xfs_buf_attach_iodone(). It should remove each | |
1031 | * log item from the buffer's list and call the callback of each in turn. | |
1032 | * When done, the buffer's fsprivate field is set to NULL and the buffer | |
1033 | * is unlocked with a call to iodone(). | |
1034 | */ | |
1035 | void | |
1036 | xfs_buf_iodone_callbacks( | |
bfc60177 | 1037 | struct xfs_buf *bp) |
1da177e4 | 1038 | { |
bfc60177 CH |
1039 | struct xfs_log_item *lip = bp->b_fspriv; |
1040 | struct xfs_mount *mp = lip->li_mountp; | |
1041 | static ulong lasttime; | |
1042 | static xfs_buftarg_t *lasttarg; | |
1da177e4 | 1043 | |
5a52c2a5 | 1044 | if (likely(!xfs_buf_geterror(bp))) |
bfc60177 | 1045 | goto do_callbacks; |
1da177e4 | 1046 | |
bfc60177 CH |
1047 | /* |
1048 | * If we've already decided to shutdown the filesystem because of | |
1049 | * I/O errors, there's no point in giving this a retry. | |
1050 | */ | |
1051 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
c867cb61 | 1052 | xfs_buf_stale(bp); |
c867cb61 | 1053 | XFS_BUF_DONE(bp); |
bfc60177 CH |
1054 | trace_xfs_buf_item_iodone(bp, _RET_IP_); |
1055 | goto do_callbacks; | |
1056 | } | |
1da177e4 | 1057 | |
49074c06 | 1058 | if (bp->b_target != lasttarg || |
bfc60177 CH |
1059 | time_after(jiffies, (lasttime + 5*HZ))) { |
1060 | lasttime = jiffies; | |
b38505b0 | 1061 | xfs_buf_ioerror_alert(bp, __func__); |
bfc60177 | 1062 | } |
49074c06 | 1063 | lasttarg = bp->b_target; |
1da177e4 | 1064 | |
bfc60177 | 1065 | /* |
25985edc | 1066 | * If the write was asynchronous then no one will be looking for the |
bfc60177 CH |
1067 | * error. Clear the error state and write the buffer out again. |
1068 | * | |
43ff2122 CH |
1069 | * XXX: This helps against transient write errors, but we need to find |
1070 | * a way to shut the filesystem down if the writes keep failing. | |
1071 | * | |
1072 | * In practice we'll shut the filesystem down soon as non-transient | |
1073 | * erorrs tend to affect the whole device and a failing log write | |
1074 | * will make us give up. But we really ought to do better here. | |
bfc60177 CH |
1075 | */ |
1076 | if (XFS_BUF_ISASYNC(bp)) { | |
43ff2122 CH |
1077 | ASSERT(bp->b_iodone != NULL); |
1078 | ||
1079 | trace_xfs_buf_item_iodone_async(bp, _RET_IP_); | |
1080 | ||
5a52c2a5 | 1081 | xfs_buf_ioerror(bp, 0); /* errno of 0 unsets the flag */ |
bfc60177 CH |
1082 | |
1083 | if (!XFS_BUF_ISSTALE(bp)) { | |
43ff2122 | 1084 | bp->b_flags |= XBF_WRITE | XBF_ASYNC | XBF_DONE; |
a2dcf5df | 1085 | xfs_buf_iorequest(bp); |
43ff2122 CH |
1086 | } else { |
1087 | xfs_buf_relse(bp); | |
1da177e4 | 1088 | } |
43ff2122 | 1089 | |
1da177e4 LT |
1090 | return; |
1091 | } | |
0b1b213f | 1092 | |
bfc60177 CH |
1093 | /* |
1094 | * If the write of the buffer was synchronous, we want to make | |
1095 | * sure to return the error to the caller of xfs_bwrite(). | |
1096 | */ | |
c867cb61 | 1097 | xfs_buf_stale(bp); |
1da177e4 | 1098 | XFS_BUF_DONE(bp); |
0b1b213f CH |
1099 | |
1100 | trace_xfs_buf_error_relse(bp, _RET_IP_); | |
1101 | ||
bfc60177 | 1102 | do_callbacks: |
c90821a2 | 1103 | xfs_buf_do_callbacks(bp); |
adadbeef | 1104 | bp->b_fspriv = NULL; |
cb669ca5 | 1105 | bp->b_iodone = NULL; |
bfc60177 | 1106 | xfs_buf_ioend(bp, 0); |
1da177e4 LT |
1107 | } |
1108 | ||
1da177e4 LT |
1109 | /* |
1110 | * This is the iodone() function for buffers which have been | |
1111 | * logged. It is called when they are eventually flushed out. | |
1112 | * It should remove the buf item from the AIL, and free the buf item. | |
1113 | * It is called by xfs_buf_iodone_callbacks() above which will take | |
1114 | * care of cleaning up the buffer itself. | |
1115 | */ | |
1da177e4 LT |
1116 | void |
1117 | xfs_buf_iodone( | |
ca30b2a7 CH |
1118 | struct xfs_buf *bp, |
1119 | struct xfs_log_item *lip) | |
1da177e4 | 1120 | { |
ca30b2a7 | 1121 | struct xfs_ail *ailp = lip->li_ailp; |
1da177e4 | 1122 | |
ca30b2a7 | 1123 | ASSERT(BUF_ITEM(lip)->bli_buf == bp); |
1da177e4 | 1124 | |
e1f5dbd7 | 1125 | xfs_buf_rele(bp); |
1da177e4 LT |
1126 | |
1127 | /* | |
1128 | * If we are forcibly shutting down, this may well be | |
1129 | * off the AIL already. That's because we simulate the | |
1130 | * log-committed callbacks to unpin these buffers. Or we may never | |
1131 | * have put this item on AIL because of the transaction was | |
783a2f65 | 1132 | * aborted forcibly. xfs_trans_ail_delete() takes care of these. |
1da177e4 LT |
1133 | * |
1134 | * Either way, AIL is useless if we're forcing a shutdown. | |
1135 | */ | |
fc1829f3 | 1136 | spin_lock(&ailp->xa_lock); |
04913fdd | 1137 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 | 1138 | xfs_buf_item_free(BUF_ITEM(lip)); |
1da177e4 | 1139 | } |