]>
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" |
4fb6e8ad | 20 | #include "xfs_format.h" |
239880ef DC |
21 | #include "xfs_log_format.h" |
22 | #include "xfs_trans_resv.h" | |
a844f451 | 23 | #include "xfs_bit.h" |
1da177e4 | 24 | #include "xfs_sb.h" |
1da177e4 | 25 | #include "xfs_mount.h" |
239880ef | 26 | #include "xfs_trans.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" |
239880ef | 31 | #include "xfs_log.h" |
1618ea8c | 32 | #include "xfs_inode.h" |
1da177e4 LT |
33 | |
34 | ||
35 | kmem_zone_t *xfs_buf_item_zone; | |
36 | ||
7bfa31d8 CH |
37 | static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip) |
38 | { | |
39 | return container_of(lip, struct xfs_buf_log_item, bli_item); | |
40 | } | |
41 | ||
c90821a2 | 42 | STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp); |
1da177e4 | 43 | |
166d1368 DC |
44 | static inline int |
45 | xfs_buf_log_format_size( | |
46 | struct xfs_buf_log_format *blfp) | |
47 | { | |
48 | return offsetof(struct xfs_buf_log_format, blf_data_map) + | |
49 | (blfp->blf_map_size * sizeof(blfp->blf_data_map[0])); | |
50 | } | |
51 | ||
1da177e4 LT |
52 | /* |
53 | * This returns the number of log iovecs needed to log the | |
54 | * given buf log item. | |
55 | * | |
56 | * It calculates this as 1 iovec for the buf log format structure | |
57 | * and 1 for each stretch of non-contiguous chunks to be logged. | |
58 | * Contiguous chunks are logged in a single iovec. | |
59 | * | |
60 | * If the XFS_BLI_STALE flag has been set, then log nothing. | |
61 | */ | |
166d1368 | 62 | STATIC void |
372cc85e DC |
63 | xfs_buf_item_size_segment( |
64 | struct xfs_buf_log_item *bip, | |
166d1368 DC |
65 | struct xfs_buf_log_format *blfp, |
66 | int *nvecs, | |
67 | int *nbytes) | |
1da177e4 | 68 | { |
7bfa31d8 | 69 | struct xfs_buf *bp = bip->bli_buf; |
7bfa31d8 CH |
70 | int next_bit; |
71 | int last_bit; | |
1da177e4 | 72 | |
372cc85e DC |
73 | last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
74 | if (last_bit == -1) | |
166d1368 | 75 | return; |
372cc85e DC |
76 | |
77 | /* | |
78 | * initial count for a dirty buffer is 2 vectors - the format structure | |
79 | * and the first dirty region. | |
80 | */ | |
166d1368 DC |
81 | *nvecs += 2; |
82 | *nbytes += xfs_buf_log_format_size(blfp) + XFS_BLF_CHUNK; | |
1da177e4 | 83 | |
1da177e4 LT |
84 | while (last_bit != -1) { |
85 | /* | |
86 | * This takes the bit number to start looking from and | |
87 | * returns the next set bit from there. It returns -1 | |
88 | * if there are no more bits set or the start bit is | |
89 | * beyond the end of the bitmap. | |
90 | */ | |
372cc85e DC |
91 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
92 | last_bit + 1); | |
1da177e4 LT |
93 | /* |
94 | * If we run out of bits, leave the loop, | |
95 | * else if we find a new set of bits bump the number of vecs, | |
96 | * else keep scanning the current set of bits. | |
97 | */ | |
98 | if (next_bit == -1) { | |
372cc85e | 99 | break; |
1da177e4 LT |
100 | } else if (next_bit != last_bit + 1) { |
101 | last_bit = next_bit; | |
166d1368 | 102 | (*nvecs)++; |
c1155410 DC |
103 | } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) != |
104 | (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) + | |
105 | XFS_BLF_CHUNK)) { | |
1da177e4 | 106 | last_bit = next_bit; |
166d1368 | 107 | (*nvecs)++; |
1da177e4 LT |
108 | } else { |
109 | last_bit++; | |
110 | } | |
166d1368 | 111 | *nbytes += XFS_BLF_CHUNK; |
1da177e4 | 112 | } |
1da177e4 LT |
113 | } |
114 | ||
115 | /* | |
372cc85e DC |
116 | * This returns the number of log iovecs needed to log the given buf log item. |
117 | * | |
118 | * It calculates this as 1 iovec for the buf log format structure and 1 for each | |
119 | * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged | |
120 | * in a single iovec. | |
121 | * | |
122 | * Discontiguous buffers need a format structure per region that that is being | |
123 | * logged. This makes the changes in the buffer appear to log recovery as though | |
124 | * they came from separate buffers, just like would occur if multiple buffers | |
125 | * were used instead of a single discontiguous buffer. This enables | |
126 | * discontiguous buffers to be in-memory constructs, completely transparent to | |
127 | * what ends up on disk. | |
128 | * | |
129 | * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log | |
130 | * format structures. | |
1da177e4 | 131 | */ |
166d1368 | 132 | STATIC void |
372cc85e | 133 | xfs_buf_item_size( |
166d1368 DC |
134 | struct xfs_log_item *lip, |
135 | int *nvecs, | |
136 | int *nbytes) | |
1da177e4 | 137 | { |
7bfa31d8 | 138 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
372cc85e DC |
139 | int i; |
140 | ||
141 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
142 | if (bip->bli_flags & XFS_BLI_STALE) { | |
143 | /* | |
144 | * The buffer is stale, so all we need to log | |
145 | * is the buf log format structure with the | |
146 | * cancel flag in it. | |
147 | */ | |
148 | trace_xfs_buf_item_size_stale(bip); | |
b9438173 | 149 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
166d1368 DC |
150 | *nvecs += bip->bli_format_count; |
151 | for (i = 0; i < bip->bli_format_count; i++) { | |
152 | *nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]); | |
153 | } | |
154 | return; | |
372cc85e DC |
155 | } |
156 | ||
157 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); | |
158 | ||
5f6bed76 DC |
159 | if (bip->bli_flags & XFS_BLI_ORDERED) { |
160 | /* | |
161 | * The buffer has been logged just to order it. | |
162 | * It is not being included in the transaction | |
163 | * commit, so no vectors are used at all. | |
164 | */ | |
165 | trace_xfs_buf_item_size_ordered(bip); | |
166d1368 DC |
166 | *nvecs = XFS_LOG_VEC_ORDERED; |
167 | return; | |
5f6bed76 DC |
168 | } |
169 | ||
372cc85e DC |
170 | /* |
171 | * the vector count is based on the number of buffer vectors we have | |
172 | * dirty bits in. This will only be greater than one when we have a | |
173 | * compound buffer with more than one segment dirty. Hence for compound | |
174 | * buffers we need to track which segment the dirty bits correspond to, | |
175 | * and when we move from one segment to the next increment the vector | |
176 | * count for the extra buf log format structure that will need to be | |
177 | * written. | |
178 | */ | |
372cc85e | 179 | for (i = 0; i < bip->bli_format_count; i++) { |
166d1368 DC |
180 | xfs_buf_item_size_segment(bip, &bip->bli_formats[i], |
181 | nvecs, nbytes); | |
372cc85e | 182 | } |
372cc85e | 183 | trace_xfs_buf_item_size(bip); |
372cc85e DC |
184 | } |
185 | ||
1234351c | 186 | static inline void |
7aeb7222 | 187 | xfs_buf_item_copy_iovec( |
bde7cff6 | 188 | struct xfs_log_vec *lv, |
1234351c | 189 | struct xfs_log_iovec **vecp, |
7aeb7222 CH |
190 | struct xfs_buf *bp, |
191 | uint offset, | |
192 | int first_bit, | |
193 | uint nbits) | |
194 | { | |
195 | offset += first_bit * XFS_BLF_CHUNK; | |
bde7cff6 | 196 | xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK, |
1234351c CH |
197 | xfs_buf_offset(bp, offset), |
198 | nbits * XFS_BLF_CHUNK); | |
7aeb7222 CH |
199 | } |
200 | ||
201 | static inline bool | |
202 | xfs_buf_item_straddle( | |
203 | struct xfs_buf *bp, | |
204 | uint offset, | |
205 | int next_bit, | |
206 | int last_bit) | |
207 | { | |
208 | return xfs_buf_offset(bp, offset + (next_bit << XFS_BLF_SHIFT)) != | |
209 | (xfs_buf_offset(bp, offset + (last_bit << XFS_BLF_SHIFT)) + | |
210 | XFS_BLF_CHUNK); | |
211 | } | |
212 | ||
1234351c | 213 | static void |
372cc85e DC |
214 | xfs_buf_item_format_segment( |
215 | struct xfs_buf_log_item *bip, | |
bde7cff6 | 216 | struct xfs_log_vec *lv, |
1234351c | 217 | struct xfs_log_iovec **vecp, |
372cc85e DC |
218 | uint offset, |
219 | struct xfs_buf_log_format *blfp) | |
220 | { | |
7bfa31d8 | 221 | struct xfs_buf *bp = bip->bli_buf; |
1da177e4 | 222 | uint base_size; |
1da177e4 LT |
223 | int first_bit; |
224 | int last_bit; | |
225 | int next_bit; | |
226 | uint nbits; | |
1da177e4 | 227 | |
372cc85e | 228 | /* copy the flags across from the base format item */ |
b9438173 | 229 | blfp->blf_flags = bip->__bli_format.blf_flags; |
1da177e4 LT |
230 | |
231 | /* | |
77c1a08f DC |
232 | * Base size is the actual size of the ondisk structure - it reflects |
233 | * the actual size of the dirty bitmap rather than the size of the in | |
234 | * memory structure. | |
1da177e4 | 235 | */ |
166d1368 | 236 | base_size = xfs_buf_log_format_size(blfp); |
820a554f | 237 | |
820a554f MT |
238 | first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
239 | if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) { | |
240 | /* | |
241 | * If the map is not be dirty in the transaction, mark | |
242 | * the size as zero and do not advance the vector pointer. | |
243 | */ | |
bde7cff6 | 244 | return; |
820a554f MT |
245 | } |
246 | ||
bde7cff6 CH |
247 | blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size); |
248 | blfp->blf_size = 1; | |
1da177e4 LT |
249 | |
250 | if (bip->bli_flags & XFS_BLI_STALE) { | |
251 | /* | |
252 | * The buffer is stale, so all we need to log | |
253 | * is the buf log format structure with the | |
254 | * cancel flag in it. | |
255 | */ | |
0b1b213f | 256 | trace_xfs_buf_item_format_stale(bip); |
372cc85e | 257 | ASSERT(blfp->blf_flags & XFS_BLF_CANCEL); |
bde7cff6 | 258 | return; |
1da177e4 LT |
259 | } |
260 | ||
5f6bed76 | 261 | |
1da177e4 LT |
262 | /* |
263 | * Fill in an iovec for each set of contiguous chunks. | |
264 | */ | |
1da177e4 LT |
265 | last_bit = first_bit; |
266 | nbits = 1; | |
267 | for (;;) { | |
268 | /* | |
269 | * This takes the bit number to start looking from and | |
270 | * returns the next set bit from there. It returns -1 | |
271 | * if there are no more bits set or the start bit is | |
272 | * beyond the end of the bitmap. | |
273 | */ | |
372cc85e DC |
274 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
275 | (uint)last_bit + 1); | |
1da177e4 | 276 | /* |
7aeb7222 CH |
277 | * If we run out of bits fill in the last iovec and get out of |
278 | * the loop. Else if we start a new set of bits then fill in | |
279 | * the iovec for the series we were looking at and start | |
280 | * counting the bits in the new one. Else we're still in the | |
281 | * same set of bits so just keep counting and scanning. | |
1da177e4 LT |
282 | */ |
283 | if (next_bit == -1) { | |
bde7cff6 | 284 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
7aeb7222 | 285 | first_bit, nbits); |
bde7cff6 | 286 | blfp->blf_size++; |
1da177e4 | 287 | break; |
7aeb7222 CH |
288 | } else if (next_bit != last_bit + 1 || |
289 | xfs_buf_item_straddle(bp, offset, next_bit, last_bit)) { | |
bde7cff6 | 290 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
1234351c | 291 | first_bit, nbits); |
bde7cff6 | 292 | blfp->blf_size++; |
1da177e4 LT |
293 | first_bit = next_bit; |
294 | last_bit = next_bit; | |
295 | nbits = 1; | |
296 | } else { | |
297 | last_bit++; | |
298 | nbits++; | |
299 | } | |
300 | } | |
372cc85e DC |
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, | |
bde7cff6 | 312 | struct xfs_log_vec *lv) |
372cc85e DC |
313 | { |
314 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); | |
315 | struct xfs_buf *bp = bip->bli_buf; | |
bde7cff6 | 316 | struct xfs_log_iovec *vecp = NULL; |
372cc85e DC |
317 | uint offset = 0; |
318 | int i; | |
319 | ||
320 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
321 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
322 | (bip->bli_flags & XFS_BLI_STALE)); | |
0d612fb5 DC |
323 | ASSERT((bip->bli_flags & XFS_BLI_STALE) || |
324 | (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF | |
325 | && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF)); | |
326 | ||
372cc85e DC |
327 | |
328 | /* | |
329 | * If it is an inode buffer, transfer the in-memory state to the | |
ddf6ad01 DC |
330 | * format flags and clear the in-memory state. |
331 | * | |
332 | * For buffer based inode allocation, we do not transfer | |
372cc85e DC |
333 | * this state if the inode buffer allocation has not yet been committed |
334 | * to the log as setting the XFS_BLI_INODE_BUF flag will prevent | |
335 | * correct replay of the inode allocation. | |
ddf6ad01 DC |
336 | * |
337 | * For icreate item based inode allocation, the buffers aren't written | |
338 | * to the journal during allocation, and hence we should always tag the | |
339 | * buffer as an inode buffer so that the correct unlinked list replay | |
340 | * occurs during recovery. | |
372cc85e DC |
341 | */ |
342 | if (bip->bli_flags & XFS_BLI_INODE_BUF) { | |
ddf6ad01 DC |
343 | if (xfs_sb_version_hascrc(&lip->li_mountp->m_sb) || |
344 | !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && | |
372cc85e | 345 | xfs_log_item_in_current_chkpt(lip))) |
b9438173 | 346 | bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF; |
372cc85e DC |
347 | bip->bli_flags &= ~XFS_BLI_INODE_BUF; |
348 | } | |
349 | ||
5f6bed76 DC |
350 | if ((bip->bli_flags & (XFS_BLI_ORDERED|XFS_BLI_STALE)) == |
351 | XFS_BLI_ORDERED) { | |
352 | /* | |
353 | * The buffer has been logged just to order it. It is not being | |
354 | * included in the transaction commit, so don't format it. | |
355 | */ | |
356 | trace_xfs_buf_item_format_ordered(bip); | |
357 | return; | |
358 | } | |
359 | ||
372cc85e | 360 | for (i = 0; i < bip->bli_format_count; i++) { |
bde7cff6 | 361 | xfs_buf_item_format_segment(bip, lv, &vecp, offset, |
1234351c | 362 | &bip->bli_formats[i]); |
a3916e52 | 363 | offset += BBTOB(bp->b_maps[i].bm_len); |
372cc85e | 364 | } |
1da177e4 LT |
365 | |
366 | /* | |
367 | * Check to make sure everything is consistent. | |
368 | */ | |
0b1b213f | 369 | trace_xfs_buf_item_format(bip); |
1da177e4 LT |
370 | } |
371 | ||
372 | /* | |
64fc35de | 373 | * This is called to pin the buffer associated with the buf log item in memory |
4d16e924 | 374 | * so it cannot be written out. |
64fc35de DC |
375 | * |
376 | * We also always take a reference to the buffer log item here so that the bli | |
377 | * is held while the item is pinned in memory. This means that we can | |
378 | * unconditionally drop the reference count a transaction holds when the | |
379 | * transaction is completed. | |
1da177e4 | 380 | */ |
ba0f32d4 | 381 | STATIC void |
1da177e4 | 382 | xfs_buf_item_pin( |
7bfa31d8 | 383 | struct xfs_log_item *lip) |
1da177e4 | 384 | { |
7bfa31d8 | 385 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
1da177e4 | 386 | |
1da177e4 LT |
387 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
388 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
5f6bed76 | 389 | (bip->bli_flags & XFS_BLI_ORDERED) || |
1da177e4 | 390 | (bip->bli_flags & XFS_BLI_STALE)); |
7bfa31d8 | 391 | |
0b1b213f | 392 | trace_xfs_buf_item_pin(bip); |
4d16e924 CH |
393 | |
394 | atomic_inc(&bip->bli_refcount); | |
395 | atomic_inc(&bip->bli_buf->b_pin_count); | |
1da177e4 LT |
396 | } |
397 | ||
1da177e4 LT |
398 | /* |
399 | * This is called to unpin the buffer associated with the buf log | |
400 | * item which was previously pinned with a call to xfs_buf_item_pin(). | |
1da177e4 LT |
401 | * |
402 | * Also drop the reference to the buf item for the current transaction. | |
403 | * If the XFS_BLI_STALE flag is set and we are the last reference, | |
404 | * then free up the buf log item and unlock the buffer. | |
9412e318 CH |
405 | * |
406 | * If the remove flag is set we are called from uncommit in the | |
407 | * forced-shutdown path. If that is true and the reference count on | |
408 | * the log item is going to drop to zero we need to free the item's | |
409 | * descriptor in the transaction. | |
1da177e4 | 410 | */ |
ba0f32d4 | 411 | STATIC void |
1da177e4 | 412 | xfs_buf_item_unpin( |
7bfa31d8 | 413 | struct xfs_log_item *lip, |
9412e318 | 414 | int remove) |
1da177e4 | 415 | { |
7bfa31d8 | 416 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
9412e318 | 417 | xfs_buf_t *bp = bip->bli_buf; |
7bfa31d8 | 418 | struct xfs_ail *ailp = lip->li_ailp; |
8e123850 | 419 | int stale = bip->bli_flags & XFS_BLI_STALE; |
7bfa31d8 | 420 | int freed; |
1da177e4 | 421 | |
adadbeef | 422 | ASSERT(bp->b_fspriv == bip); |
1da177e4 | 423 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
9412e318 | 424 | |
0b1b213f | 425 | trace_xfs_buf_item_unpin(bip); |
1da177e4 LT |
426 | |
427 | freed = atomic_dec_and_test(&bip->bli_refcount); | |
4d16e924 CH |
428 | |
429 | if (atomic_dec_and_test(&bp->b_pin_count)) | |
430 | wake_up_all(&bp->b_waiters); | |
7bfa31d8 | 431 | |
1da177e4 LT |
432 | if (freed && stale) { |
433 | ASSERT(bip->bli_flags & XFS_BLI_STALE); | |
0c842ad4 | 434 | ASSERT(xfs_buf_islocked(bp)); |
5cfd28b6 | 435 | ASSERT(bp->b_flags & XBF_STALE); |
b9438173 | 436 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
9412e318 | 437 | |
0b1b213f CH |
438 | trace_xfs_buf_item_unpin_stale(bip); |
439 | ||
9412e318 CH |
440 | if (remove) { |
441 | /* | |
e34a314c DC |
442 | * If we are in a transaction context, we have to |
443 | * remove the log item from the transaction as we are | |
444 | * about to release our reference to the buffer. If we | |
445 | * don't, the unlock that occurs later in | |
446 | * xfs_trans_uncommit() will try to reference the | |
9412e318 CH |
447 | * buffer which we no longer have a hold on. |
448 | */ | |
e34a314c DC |
449 | if (lip->li_desc) |
450 | xfs_trans_del_item(lip); | |
9412e318 CH |
451 | |
452 | /* | |
453 | * Since the transaction no longer refers to the buffer, | |
454 | * the buffer should no longer refer to the transaction. | |
455 | */ | |
bf9d9013 | 456 | bp->b_transp = NULL; |
9412e318 CH |
457 | } |
458 | ||
1da177e4 LT |
459 | /* |
460 | * If we get called here because of an IO error, we may | |
783a2f65 | 461 | * or may not have the item on the AIL. xfs_trans_ail_delete() |
1da177e4 | 462 | * will take care of that situation. |
783a2f65 | 463 | * xfs_trans_ail_delete() drops the AIL lock. |
1da177e4 LT |
464 | */ |
465 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { | |
c90821a2 | 466 | xfs_buf_do_callbacks(bp); |
adadbeef | 467 | bp->b_fspriv = NULL; |
cb669ca5 | 468 | bp->b_iodone = NULL; |
1da177e4 | 469 | } else { |
783a2f65 | 470 | spin_lock(&ailp->xa_lock); |
04913fdd | 471 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR); |
1da177e4 | 472 | xfs_buf_item_relse(bp); |
adadbeef | 473 | ASSERT(bp->b_fspriv == NULL); |
1da177e4 LT |
474 | } |
475 | xfs_buf_relse(bp); | |
960c60af | 476 | } else if (freed && remove) { |
137fff09 DC |
477 | /* |
478 | * There are currently two references to the buffer - the active | |
479 | * LRU reference and the buf log item. What we are about to do | |
480 | * here - simulate a failed IO completion - requires 3 | |
481 | * references. | |
482 | * | |
483 | * The LRU reference is removed by the xfs_buf_stale() call. The | |
484 | * buf item reference is removed by the xfs_buf_iodone() | |
485 | * callback that is run by xfs_buf_do_callbacks() during ioend | |
486 | * processing (via the bp->b_iodone callback), and then finally | |
487 | * the ioend processing will drop the IO reference if the buffer | |
488 | * is marked XBF_ASYNC. | |
489 | * | |
490 | * Hence we need to take an additional reference here so that IO | |
491 | * completion processing doesn't free the buffer prematurely. | |
492 | */ | |
960c60af | 493 | xfs_buf_lock(bp); |
137fff09 DC |
494 | xfs_buf_hold(bp); |
495 | bp->b_flags |= XBF_ASYNC; | |
2451337d | 496 | xfs_buf_ioerror(bp, -EIO); |
b0388bf1 | 497 | bp->b_flags &= ~XBF_DONE; |
960c60af | 498 | xfs_buf_stale(bp); |
e8aaba9a | 499 | xfs_buf_ioend(bp); |
1da177e4 LT |
500 | } |
501 | } | |
502 | ||
ac8809f9 DC |
503 | /* |
504 | * Buffer IO error rate limiting. Limit it to no more than 10 messages per 30 | |
505 | * seconds so as to not spam logs too much on repeated detection of the same | |
506 | * buffer being bad.. | |
507 | */ | |
508 | ||
02cc1876 | 509 | static DEFINE_RATELIMIT_STATE(xfs_buf_write_fail_rl_state, 30 * HZ, 10); |
ac8809f9 | 510 | |
ba0f32d4 | 511 | STATIC uint |
43ff2122 CH |
512 | xfs_buf_item_push( |
513 | struct xfs_log_item *lip, | |
514 | struct list_head *buffer_list) | |
1da177e4 | 515 | { |
7bfa31d8 CH |
516 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
517 | struct xfs_buf *bp = bip->bli_buf; | |
43ff2122 | 518 | uint rval = XFS_ITEM_SUCCESS; |
1da177e4 | 519 | |
811e64c7 | 520 | if (xfs_buf_ispinned(bp)) |
1da177e4 | 521 | return XFS_ITEM_PINNED; |
5337fe9b BF |
522 | if (!xfs_buf_trylock(bp)) { |
523 | /* | |
524 | * If we have just raced with a buffer being pinned and it has | |
525 | * been marked stale, we could end up stalling until someone else | |
526 | * issues a log force to unpin the stale buffer. Check for the | |
527 | * race condition here so xfsaild recognizes the buffer is pinned | |
528 | * and queues a log force to move it along. | |
529 | */ | |
530 | if (xfs_buf_ispinned(bp)) | |
531 | return XFS_ITEM_PINNED; | |
1da177e4 | 532 | return XFS_ITEM_LOCKED; |
5337fe9b | 533 | } |
1da177e4 | 534 | |
1da177e4 | 535 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
43ff2122 CH |
536 | |
537 | trace_xfs_buf_item_push(bip); | |
538 | ||
ac8809f9 DC |
539 | /* has a previous flush failed due to IO errors? */ |
540 | if ((bp->b_flags & XBF_WRITE_FAIL) && | |
fdadf267 | 541 | ___ratelimit(&xfs_buf_write_fail_rl_state, "XFS: Failing async write")) { |
ac8809f9 | 542 | xfs_warn(bp->b_target->bt_mount, |
fdadf267 | 543 | "Failing async write on buffer block 0x%llx. Retrying async write.", |
ac8809f9 DC |
544 | (long long)bp->b_bn); |
545 | } | |
546 | ||
43ff2122 CH |
547 | if (!xfs_buf_delwri_queue(bp, buffer_list)) |
548 | rval = XFS_ITEM_FLUSHING; | |
549 | xfs_buf_unlock(bp); | |
550 | return rval; | |
1da177e4 LT |
551 | } |
552 | ||
553 | /* | |
64fc35de DC |
554 | * Release the buffer associated with the buf log item. If there is no dirty |
555 | * logged data associated with the buffer recorded in the buf log item, then | |
556 | * free the buf log item and remove the reference to it in the buffer. | |
1da177e4 | 557 | * |
64fc35de DC |
558 | * This call ignores the recursion count. It is only called when the buffer |
559 | * should REALLY be unlocked, regardless of the recursion count. | |
1da177e4 | 560 | * |
64fc35de DC |
561 | * We unconditionally drop the transaction's reference to the log item. If the |
562 | * item was logged, then another reference was taken when it was pinned, so we | |
563 | * can safely drop the transaction reference now. This also allows us to avoid | |
564 | * potential races with the unpin code freeing the bli by not referencing the | |
565 | * bli after we've dropped the reference count. | |
566 | * | |
567 | * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item | |
568 | * if necessary but do not unlock the buffer. This is for support of | |
569 | * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't | |
570 | * free the item. | |
1da177e4 | 571 | */ |
ba0f32d4 | 572 | STATIC void |
1da177e4 | 573 | xfs_buf_item_unlock( |
7bfa31d8 | 574 | struct xfs_log_item *lip) |
1da177e4 | 575 | { |
7bfa31d8 CH |
576 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
577 | struct xfs_buf *bp = bip->bli_buf; | |
5f6bed76 DC |
578 | bool clean; |
579 | bool aborted; | |
580 | int flags; | |
1da177e4 | 581 | |
64fc35de | 582 | /* Clear the buffer's association with this transaction. */ |
bf9d9013 | 583 | bp->b_transp = NULL; |
1da177e4 LT |
584 | |
585 | /* | |
64fc35de DC |
586 | * If this is a transaction abort, don't return early. Instead, allow |
587 | * the brelse to happen. Normally it would be done for stale | |
588 | * (cancelled) buffers at unpin time, but we'll never go through the | |
589 | * pin/unpin cycle if we abort inside commit. | |
1da177e4 | 590 | */ |
5f6bed76 | 591 | aborted = (lip->li_flags & XFS_LI_ABORTED) ? true : false; |
1da177e4 | 592 | /* |
5f6bed76 DC |
593 | * Before possibly freeing the buf item, copy the per-transaction state |
594 | * so we can reference it safely later after clearing it from the | |
595 | * buffer log item. | |
64fc35de | 596 | */ |
5f6bed76 DC |
597 | flags = bip->bli_flags; |
598 | bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED); | |
64fc35de DC |
599 | |
600 | /* | |
601 | * If the buf item is marked stale, then don't do anything. We'll | |
602 | * unlock the buffer and free the buf item when the buffer is unpinned | |
603 | * for the last time. | |
1da177e4 | 604 | */ |
5f6bed76 | 605 | if (flags & XFS_BLI_STALE) { |
0b1b213f | 606 | trace_xfs_buf_item_unlock_stale(bip); |
b9438173 | 607 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
64fc35de DC |
608 | if (!aborted) { |
609 | atomic_dec(&bip->bli_refcount); | |
1da177e4 | 610 | return; |
64fc35de | 611 | } |
1da177e4 LT |
612 | } |
613 | ||
0b1b213f | 614 | trace_xfs_buf_item_unlock(bip); |
1da177e4 LT |
615 | |
616 | /* | |
64fc35de | 617 | * If the buf item isn't tracking any data, free it, otherwise drop the |
3b19034d DC |
618 | * reference we hold to it. If we are aborting the transaction, this may |
619 | * be the only reference to the buf item, so we free it anyway | |
620 | * regardless of whether it is dirty or not. A dirty abort implies a | |
621 | * shutdown, anyway. | |
5f6bed76 DC |
622 | * |
623 | * Ordered buffers are dirty but may have no recorded changes, so ensure | |
624 | * we only release clean items here. | |
1da177e4 | 625 | */ |
5f6bed76 DC |
626 | clean = (flags & XFS_BLI_DIRTY) ? false : true; |
627 | if (clean) { | |
628 | int i; | |
629 | for (i = 0; i < bip->bli_format_count; i++) { | |
630 | if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map, | |
631 | bip->bli_formats[i].blf_map_size)) { | |
632 | clean = false; | |
633 | break; | |
634 | } | |
c883d0c4 MT |
635 | } |
636 | } | |
46f9d2eb DC |
637 | |
638 | /* | |
639 | * Clean buffers, by definition, cannot be in the AIL. However, aborted | |
3d4b4a3e BF |
640 | * buffers may be in the AIL regardless of dirty state. An aborted |
641 | * transaction that invalidates a buffer already in the AIL may have | |
642 | * marked it stale and cleared the dirty state, for example. | |
643 | * | |
644 | * Therefore if we are aborting a buffer and we've just taken the last | |
645 | * reference away, we have to check if it is in the AIL before freeing | |
646 | * it. We need to free it in this case, because an aborted transaction | |
647 | * has already shut the filesystem down and this is the last chance we | |
648 | * will have to do so. | |
46f9d2eb DC |
649 | */ |
650 | if (atomic_dec_and_test(&bip->bli_refcount)) { | |
3d4b4a3e | 651 | if (aborted) { |
46f9d2eb | 652 | ASSERT(XFS_FORCED_SHUTDOWN(lip->li_mountp)); |
146e54b7 | 653 | xfs_trans_ail_remove(lip, SHUTDOWN_LOG_IO_ERROR); |
3b19034d | 654 | xfs_buf_item_relse(bp); |
3d4b4a3e BF |
655 | } else if (clean) |
656 | xfs_buf_item_relse(bp); | |
46f9d2eb | 657 | } |
1da177e4 | 658 | |
5f6bed76 | 659 | if (!(flags & XFS_BLI_HOLD)) |
1da177e4 | 660 | xfs_buf_relse(bp); |
1da177e4 LT |
661 | } |
662 | ||
663 | /* | |
664 | * This is called to find out where the oldest active copy of the | |
665 | * buf log item in the on disk log resides now that the last log | |
666 | * write of it completed at the given lsn. | |
667 | * We always re-log all the dirty data in a buffer, so usually the | |
668 | * latest copy in the on disk log is the only one that matters. For | |
669 | * those cases we simply return the given lsn. | |
670 | * | |
671 | * The one exception to this is for buffers full of newly allocated | |
672 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF | |
673 | * flag set, indicating that only the di_next_unlinked fields from the | |
674 | * inodes in the buffers will be replayed during recovery. If the | |
675 | * original newly allocated inode images have not yet been flushed | |
676 | * when the buffer is so relogged, then we need to make sure that we | |
677 | * keep the old images in the 'active' portion of the log. We do this | |
678 | * by returning the original lsn of that transaction here rather than | |
679 | * the current one. | |
680 | */ | |
ba0f32d4 | 681 | STATIC xfs_lsn_t |
1da177e4 | 682 | xfs_buf_item_committed( |
7bfa31d8 | 683 | struct xfs_log_item *lip, |
1da177e4 LT |
684 | xfs_lsn_t lsn) |
685 | { | |
7bfa31d8 CH |
686 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
687 | ||
0b1b213f CH |
688 | trace_xfs_buf_item_committed(bip); |
689 | ||
7bfa31d8 CH |
690 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0) |
691 | return lip->li_lsn; | |
692 | return lsn; | |
1da177e4 LT |
693 | } |
694 | ||
ba0f32d4 | 695 | STATIC void |
7bfa31d8 CH |
696 | xfs_buf_item_committing( |
697 | struct xfs_log_item *lip, | |
698 | xfs_lsn_t commit_lsn) | |
1da177e4 LT |
699 | { |
700 | } | |
701 | ||
702 | /* | |
703 | * This is the ops vector shared by all buf log items. | |
704 | */ | |
272e42b2 | 705 | static const struct xfs_item_ops xfs_buf_item_ops = { |
7bfa31d8 CH |
706 | .iop_size = xfs_buf_item_size, |
707 | .iop_format = xfs_buf_item_format, | |
708 | .iop_pin = xfs_buf_item_pin, | |
709 | .iop_unpin = xfs_buf_item_unpin, | |
7bfa31d8 CH |
710 | .iop_unlock = xfs_buf_item_unlock, |
711 | .iop_committed = xfs_buf_item_committed, | |
712 | .iop_push = xfs_buf_item_push, | |
7bfa31d8 | 713 | .iop_committing = xfs_buf_item_committing |
1da177e4 LT |
714 | }; |
715 | ||
372cc85e DC |
716 | STATIC int |
717 | xfs_buf_item_get_format( | |
718 | struct xfs_buf_log_item *bip, | |
719 | int count) | |
720 | { | |
721 | ASSERT(bip->bli_formats == NULL); | |
722 | bip->bli_format_count = count; | |
723 | ||
724 | if (count == 1) { | |
b9438173 | 725 | bip->bli_formats = &bip->__bli_format; |
372cc85e DC |
726 | return 0; |
727 | } | |
728 | ||
729 | bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format), | |
730 | KM_SLEEP); | |
731 | if (!bip->bli_formats) | |
2451337d | 732 | return -ENOMEM; |
372cc85e DC |
733 | return 0; |
734 | } | |
735 | ||
736 | STATIC void | |
737 | xfs_buf_item_free_format( | |
738 | struct xfs_buf_log_item *bip) | |
739 | { | |
b9438173 | 740 | if (bip->bli_formats != &bip->__bli_format) { |
372cc85e DC |
741 | kmem_free(bip->bli_formats); |
742 | bip->bli_formats = NULL; | |
743 | } | |
744 | } | |
1da177e4 LT |
745 | |
746 | /* | |
747 | * Allocate a new buf log item to go with the given buffer. | |
748 | * Set the buffer's b_fsprivate field to point to the new | |
749 | * buf log item. If there are other item's attached to the | |
750 | * buffer (see xfs_buf_attach_iodone() below), then put the | |
751 | * buf log item at the front. | |
752 | */ | |
f79af0b9 | 753 | int |
1da177e4 | 754 | xfs_buf_item_init( |
f79af0b9 DC |
755 | struct xfs_buf *bp, |
756 | struct xfs_mount *mp) | |
1da177e4 | 757 | { |
f79af0b9 DC |
758 | struct xfs_log_item *lip = bp->b_fspriv; |
759 | struct xfs_buf_log_item *bip; | |
1da177e4 LT |
760 | int chunks; |
761 | int map_size; | |
372cc85e DC |
762 | int error; |
763 | int i; | |
1da177e4 LT |
764 | |
765 | /* | |
766 | * Check to see if there is already a buf log item for | |
767 | * this buffer. If there is, it is guaranteed to be | |
768 | * the first. If we do already have one, there is | |
769 | * nothing to do here so return. | |
770 | */ | |
ebad861b | 771 | ASSERT(bp->b_target->bt_mount == mp); |
adadbeef | 772 | if (lip != NULL && lip->li_type == XFS_LI_BUF) |
f79af0b9 | 773 | return 0; |
1da177e4 | 774 | |
372cc85e | 775 | bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP); |
43f5efc5 | 776 | xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops); |
1da177e4 | 777 | bip->bli_buf = bp; |
372cc85e DC |
778 | |
779 | /* | |
780 | * chunks is the number of XFS_BLF_CHUNK size pieces the buffer | |
781 | * can be divided into. Make sure not to truncate any pieces. | |
782 | * map_size is the size of the bitmap needed to describe the | |
783 | * chunks of the buffer. | |
784 | * | |
785 | * Discontiguous buffer support follows the layout of the underlying | |
786 | * buffer. This makes the implementation as simple as possible. | |
787 | */ | |
788 | error = xfs_buf_item_get_format(bip, bp->b_map_count); | |
789 | ASSERT(error == 0); | |
f79af0b9 DC |
790 | if (error) { /* to stop gcc throwing set-but-unused warnings */ |
791 | kmem_zone_free(xfs_buf_item_zone, bip); | |
792 | return error; | |
793 | } | |
794 | ||
372cc85e DC |
795 | |
796 | for (i = 0; i < bip->bli_format_count; i++) { | |
797 | chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len), | |
798 | XFS_BLF_CHUNK); | |
799 | map_size = DIV_ROUND_UP(chunks, NBWORD); | |
800 | ||
801 | bip->bli_formats[i].blf_type = XFS_LI_BUF; | |
802 | bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn; | |
803 | bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len; | |
804 | bip->bli_formats[i].blf_map_size = map_size; | |
805 | } | |
1da177e4 | 806 | |
1da177e4 LT |
807 | /* |
808 | * Put the buf item into the list of items attached to the | |
809 | * buffer at the front. | |
810 | */ | |
adadbeef CH |
811 | if (bp->b_fspriv) |
812 | bip->bli_item.li_bio_list = bp->b_fspriv; | |
813 | bp->b_fspriv = bip; | |
f79af0b9 DC |
814 | xfs_buf_hold(bp); |
815 | return 0; | |
1da177e4 LT |
816 | } |
817 | ||
818 | ||
819 | /* | |
820 | * Mark bytes first through last inclusive as dirty in the buf | |
821 | * item's bitmap. | |
822 | */ | |
632b89e8 | 823 | static void |
372cc85e | 824 | xfs_buf_item_log_segment( |
1da177e4 | 825 | uint first, |
372cc85e DC |
826 | uint last, |
827 | uint *map) | |
1da177e4 LT |
828 | { |
829 | uint first_bit; | |
830 | uint last_bit; | |
831 | uint bits_to_set; | |
832 | uint bits_set; | |
833 | uint word_num; | |
834 | uint *wordp; | |
835 | uint bit; | |
836 | uint end_bit; | |
837 | uint mask; | |
838 | ||
1da177e4 LT |
839 | /* |
840 | * Convert byte offsets to bit numbers. | |
841 | */ | |
c1155410 DC |
842 | first_bit = first >> XFS_BLF_SHIFT; |
843 | last_bit = last >> XFS_BLF_SHIFT; | |
1da177e4 LT |
844 | |
845 | /* | |
846 | * Calculate the total number of bits to be set. | |
847 | */ | |
848 | bits_to_set = last_bit - first_bit + 1; | |
849 | ||
850 | /* | |
851 | * Get a pointer to the first word in the bitmap | |
852 | * to set a bit in. | |
853 | */ | |
854 | word_num = first_bit >> BIT_TO_WORD_SHIFT; | |
372cc85e | 855 | wordp = &map[word_num]; |
1da177e4 LT |
856 | |
857 | /* | |
858 | * Calculate the starting bit in the first word. | |
859 | */ | |
860 | bit = first_bit & (uint)(NBWORD - 1); | |
861 | ||
862 | /* | |
863 | * First set any bits in the first word of our range. | |
864 | * If it starts at bit 0 of the word, it will be | |
865 | * set below rather than here. That is what the variable | |
866 | * bit tells us. The variable bits_set tracks the number | |
867 | * of bits that have been set so far. End_bit is the number | |
868 | * of the last bit to be set in this word plus one. | |
869 | */ | |
870 | if (bit) { | |
871 | end_bit = MIN(bit + bits_to_set, (uint)NBWORD); | |
79c350e4 | 872 | mask = ((1U << (end_bit - bit)) - 1) << bit; |
1da177e4 LT |
873 | *wordp |= mask; |
874 | wordp++; | |
875 | bits_set = end_bit - bit; | |
876 | } else { | |
877 | bits_set = 0; | |
878 | } | |
879 | ||
880 | /* | |
881 | * Now set bits a whole word at a time that are between | |
882 | * first_bit and last_bit. | |
883 | */ | |
884 | while ((bits_to_set - bits_set) >= NBWORD) { | |
885 | *wordp |= 0xffffffff; | |
886 | bits_set += NBWORD; | |
887 | wordp++; | |
888 | } | |
889 | ||
890 | /* | |
891 | * Finally, set any bits left to be set in one last partial word. | |
892 | */ | |
893 | end_bit = bits_to_set - bits_set; | |
894 | if (end_bit) { | |
79c350e4 | 895 | mask = (1U << end_bit) - 1; |
1da177e4 LT |
896 | *wordp |= mask; |
897 | } | |
1da177e4 LT |
898 | } |
899 | ||
372cc85e DC |
900 | /* |
901 | * Mark bytes first through last inclusive as dirty in the buf | |
902 | * item's bitmap. | |
903 | */ | |
904 | void | |
905 | xfs_buf_item_log( | |
906 | xfs_buf_log_item_t *bip, | |
907 | uint first, | |
908 | uint last) | |
909 | { | |
910 | int i; | |
911 | uint start; | |
912 | uint end; | |
913 | struct xfs_buf *bp = bip->bli_buf; | |
914 | ||
372cc85e DC |
915 | /* |
916 | * walk each buffer segment and mark them dirty appropriately. | |
917 | */ | |
918 | start = 0; | |
919 | for (i = 0; i < bip->bli_format_count; i++) { | |
920 | if (start > last) | |
921 | break; | |
a3916e52 BF |
922 | end = start + BBTOB(bp->b_maps[i].bm_len) - 1; |
923 | ||
924 | /* skip to the map that includes the first byte to log */ | |
372cc85e DC |
925 | if (first > end) { |
926 | start += BBTOB(bp->b_maps[i].bm_len); | |
927 | continue; | |
928 | } | |
a3916e52 BF |
929 | |
930 | /* | |
931 | * Trim the range to this segment and mark it in the bitmap. | |
932 | * Note that we must convert buffer offsets to segment relative | |
933 | * offsets (e.g., the first byte of each segment is byte 0 of | |
934 | * that segment). | |
935 | */ | |
372cc85e DC |
936 | if (first < start) |
937 | first = start; | |
938 | if (end > last) | |
939 | end = last; | |
a3916e52 | 940 | xfs_buf_item_log_segment(first - start, end - start, |
372cc85e DC |
941 | &bip->bli_formats[i].blf_data_map[0]); |
942 | ||
a3916e52 | 943 | start += BBTOB(bp->b_maps[i].bm_len); |
372cc85e DC |
944 | } |
945 | } | |
946 | ||
1da177e4 | 947 | |
e1f5dbd7 LM |
948 | STATIC void |
949 | xfs_buf_item_free( | |
950 | xfs_buf_log_item_t *bip) | |
951 | { | |
372cc85e | 952 | xfs_buf_item_free_format(bip); |
b1c5ebb2 | 953 | kmem_free(bip->bli_item.li_lv_shadow); |
e1f5dbd7 LM |
954 | kmem_zone_free(xfs_buf_item_zone, bip); |
955 | } | |
956 | ||
1da177e4 LT |
957 | /* |
958 | * This is called when the buf log item is no longer needed. It should | |
959 | * free the buf log item associated with the given buffer and clear | |
960 | * the buffer's pointer to the buf log item. If there are no more | |
961 | * items in the list, clear the b_iodone field of the buffer (see | |
962 | * xfs_buf_attach_iodone() below). | |
963 | */ | |
964 | void | |
965 | xfs_buf_item_relse( | |
966 | xfs_buf_t *bp) | |
967 | { | |
5f6bed76 | 968 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 969 | |
0b1b213f | 970 | trace_xfs_buf_item_relse(bp, _RET_IP_); |
5f6bed76 | 971 | ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL)); |
0b1b213f | 972 | |
adadbeef | 973 | bp->b_fspriv = bip->bli_item.li_bio_list; |
cb669ca5 CH |
974 | if (bp->b_fspriv == NULL) |
975 | bp->b_iodone = NULL; | |
adadbeef | 976 | |
e1f5dbd7 LM |
977 | xfs_buf_rele(bp); |
978 | xfs_buf_item_free(bip); | |
1da177e4 LT |
979 | } |
980 | ||
981 | ||
982 | /* | |
983 | * Add the given log item with its callback to the list of callbacks | |
984 | * to be called when the buffer's I/O completes. If it is not set | |
985 | * already, set the buffer's b_iodone() routine to be | |
986 | * xfs_buf_iodone_callbacks() and link the log item into the list of | |
987 | * items rooted at b_fsprivate. Items are always added as the second | |
988 | * entry in the list if there is a first, because the buf item code | |
989 | * assumes that the buf log item is first. | |
990 | */ | |
991 | void | |
992 | xfs_buf_attach_iodone( | |
993 | xfs_buf_t *bp, | |
994 | void (*cb)(xfs_buf_t *, xfs_log_item_t *), | |
995 | xfs_log_item_t *lip) | |
996 | { | |
997 | xfs_log_item_t *head_lip; | |
998 | ||
0c842ad4 | 999 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
1000 | |
1001 | lip->li_cb = cb; | |
adadbeef CH |
1002 | head_lip = bp->b_fspriv; |
1003 | if (head_lip) { | |
1da177e4 LT |
1004 | lip->li_bio_list = head_lip->li_bio_list; |
1005 | head_lip->li_bio_list = lip; | |
1006 | } else { | |
adadbeef | 1007 | bp->b_fspriv = lip; |
1da177e4 LT |
1008 | } |
1009 | ||
cb669ca5 CH |
1010 | ASSERT(bp->b_iodone == NULL || |
1011 | bp->b_iodone == xfs_buf_iodone_callbacks); | |
1012 | bp->b_iodone = xfs_buf_iodone_callbacks; | |
1da177e4 LT |
1013 | } |
1014 | ||
c90821a2 DC |
1015 | /* |
1016 | * We can have many callbacks on a buffer. Running the callbacks individually | |
1017 | * can cause a lot of contention on the AIL lock, so we allow for a single | |
1018 | * callback to be able to scan the remaining lip->li_bio_list for other items | |
1019 | * of the same type and callback to be processed in the first call. | |
1020 | * | |
1021 | * As a result, the loop walking the callback list below will also modify the | |
1022 | * list. it removes the first item from the list and then runs the callback. | |
1023 | * The loop then restarts from the new head of the list. This allows the | |
1024 | * callback to scan and modify the list attached to the buffer and we don't | |
1025 | * have to care about maintaining a next item pointer. | |
1026 | */ | |
1da177e4 LT |
1027 | STATIC void |
1028 | xfs_buf_do_callbacks( | |
c90821a2 | 1029 | struct xfs_buf *bp) |
1da177e4 | 1030 | { |
c90821a2 | 1031 | struct xfs_log_item *lip; |
1da177e4 | 1032 | |
adadbeef CH |
1033 | while ((lip = bp->b_fspriv) != NULL) { |
1034 | bp->b_fspriv = lip->li_bio_list; | |
1da177e4 LT |
1035 | ASSERT(lip->li_cb != NULL); |
1036 | /* | |
1037 | * Clear the next pointer so we don't have any | |
1038 | * confusion if the item is added to another buf. | |
1039 | * Don't touch the log item after calling its | |
1040 | * callback, because it could have freed itself. | |
1041 | */ | |
1042 | lip->li_bio_list = NULL; | |
1043 | lip->li_cb(bp, lip); | |
1da177e4 LT |
1044 | } |
1045 | } | |
1046 | ||
1618ea8c CM |
1047 | /* |
1048 | * Invoke the error state callback for each log item affected by the failed I/O. | |
1049 | * | |
1050 | * If a metadata buffer write fails with a non-permanent error, the buffer is | |
1051 | * eventually resubmitted and so the completion callbacks are not run. The error | |
1052 | * state may need to be propagated to the log items attached to the buffer, | |
1053 | * however, so the next AIL push of the item knows hot to handle it correctly. | |
1054 | */ | |
1055 | STATIC void | |
1056 | xfs_buf_do_callbacks_fail( | |
1057 | struct xfs_buf *bp) | |
1058 | { | |
1059 | struct xfs_log_item *next; | |
1060 | struct xfs_log_item *lip = bp->b_fspriv; | |
1061 | struct xfs_ail *ailp = lip->li_ailp; | |
1062 | ||
1063 | spin_lock(&ailp->xa_lock); | |
1064 | for (; lip; lip = next) { | |
1065 | next = lip->li_bio_list; | |
1066 | if (lip->li_ops->iop_error) | |
1067 | lip->li_ops->iop_error(lip, bp); | |
1068 | } | |
1069 | spin_unlock(&ailp->xa_lock); | |
1070 | } | |
1071 | ||
df309390 CM |
1072 | static bool |
1073 | xfs_buf_iodone_callback_error( | |
bfc60177 | 1074 | struct xfs_buf *bp) |
1da177e4 | 1075 | { |
bfc60177 CH |
1076 | struct xfs_log_item *lip = bp->b_fspriv; |
1077 | struct xfs_mount *mp = lip->li_mountp; | |
1078 | static ulong lasttime; | |
1079 | static xfs_buftarg_t *lasttarg; | |
df309390 | 1080 | struct xfs_error_cfg *cfg; |
1da177e4 | 1081 | |
bfc60177 CH |
1082 | /* |
1083 | * If we've already decided to shutdown the filesystem because of | |
1084 | * I/O errors, there's no point in giving this a retry. | |
1085 | */ | |
df309390 CM |
1086 | if (XFS_FORCED_SHUTDOWN(mp)) |
1087 | goto out_stale; | |
1da177e4 | 1088 | |
49074c06 | 1089 | if (bp->b_target != lasttarg || |
bfc60177 CH |
1090 | time_after(jiffies, (lasttime + 5*HZ))) { |
1091 | lasttime = jiffies; | |
b38505b0 | 1092 | xfs_buf_ioerror_alert(bp, __func__); |
bfc60177 | 1093 | } |
49074c06 | 1094 | lasttarg = bp->b_target; |
1da177e4 | 1095 | |
df309390 CM |
1096 | /* synchronous writes will have callers process the error */ |
1097 | if (!(bp->b_flags & XBF_ASYNC)) | |
1098 | goto out_stale; | |
1099 | ||
1100 | trace_xfs_buf_item_iodone_async(bp, _RET_IP_); | |
1101 | ASSERT(bp->b_iodone != NULL); | |
1102 | ||
5539d367 ES |
1103 | cfg = xfs_error_get_cfg(mp, XFS_ERR_METADATA, bp->b_error); |
1104 | ||
bfc60177 | 1105 | /* |
25985edc | 1106 | * If the write was asynchronous then no one will be looking for the |
df309390 CM |
1107 | * error. If this is the first failure of this type, clear the error |
1108 | * state and write the buffer out again. This means we always retry an | |
1109 | * async write failure at least once, but we also need to set the buffer | |
1110 | * up to behave correctly now for repeated failures. | |
bfc60177 | 1111 | */ |
0b4db5df | 1112 | if (!(bp->b_flags & (XBF_STALE | XBF_WRITE_FAIL)) || |
df309390 | 1113 | bp->b_last_error != bp->b_error) { |
0b4db5df | 1114 | bp->b_flags |= (XBF_WRITE | XBF_DONE | XBF_WRITE_FAIL); |
df309390 | 1115 | bp->b_last_error = bp->b_error; |
77169812 ES |
1116 | if (cfg->retry_timeout != XFS_ERR_RETRY_FOREVER && |
1117 | !bp->b_first_retry_time) | |
5539d367 | 1118 | bp->b_first_retry_time = jiffies; |
a5ea70d2 | 1119 | |
df309390 CM |
1120 | xfs_buf_ioerror(bp, 0); |
1121 | xfs_buf_submit(bp); | |
1122 | return true; | |
1123 | } | |
43ff2122 | 1124 | |
df309390 CM |
1125 | /* |
1126 | * Repeated failure on an async write. Take action according to the | |
1127 | * error configuration we have been set up to use. | |
1128 | */ | |
a5ea70d2 CM |
1129 | |
1130 | if (cfg->max_retries != XFS_ERR_RETRY_FOREVER && | |
1131 | ++bp->b_retries > cfg->max_retries) | |
1132 | goto permanent_error; | |
77169812 | 1133 | if (cfg->retry_timeout != XFS_ERR_RETRY_FOREVER && |
a5ea70d2 CM |
1134 | time_after(jiffies, cfg->retry_timeout + bp->b_first_retry_time)) |
1135 | goto permanent_error; | |
bfc60177 | 1136 | |
e6b3bb78 CM |
1137 | /* At unmount we may treat errors differently */ |
1138 | if ((mp->m_flags & XFS_MOUNT_UNMOUNTING) && mp->m_fail_unmount) | |
1139 | goto permanent_error; | |
1140 | ||
1618ea8c CM |
1141 | /* |
1142 | * Still a transient error, run IO completion failure callbacks and let | |
1143 | * the higher layers retry the buffer. | |
1144 | */ | |
1145 | xfs_buf_do_callbacks_fail(bp); | |
df309390 CM |
1146 | xfs_buf_ioerror(bp, 0); |
1147 | xfs_buf_relse(bp); | |
1148 | return true; | |
0b1b213f | 1149 | |
bfc60177 | 1150 | /* |
df309390 CM |
1151 | * Permanent error - we need to trigger a shutdown if we haven't already |
1152 | * to indicate that inconsistency will result from this action. | |
bfc60177 | 1153 | */ |
df309390 CM |
1154 | permanent_error: |
1155 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1156 | out_stale: | |
c867cb61 | 1157 | xfs_buf_stale(bp); |
b0388bf1 | 1158 | bp->b_flags |= XBF_DONE; |
0b1b213f | 1159 | trace_xfs_buf_error_relse(bp, _RET_IP_); |
df309390 CM |
1160 | return false; |
1161 | } | |
1162 | ||
1163 | /* | |
1164 | * This is the iodone() function for buffers which have had callbacks attached | |
1165 | * to them by xfs_buf_attach_iodone(). We need to iterate the items on the | |
1166 | * callback list, mark the buffer as having no more callbacks and then push the | |
1167 | * buffer through IO completion processing. | |
1168 | */ | |
1169 | void | |
1170 | xfs_buf_iodone_callbacks( | |
1171 | struct xfs_buf *bp) | |
1172 | { | |
1173 | /* | |
1174 | * If there is an error, process it. Some errors require us | |
1175 | * to run callbacks after failure processing is done so we | |
1176 | * detect that and take appropriate action. | |
1177 | */ | |
1178 | if (bp->b_error && xfs_buf_iodone_callback_error(bp)) | |
1179 | return; | |
1180 | ||
1181 | /* | |
1182 | * Successful IO or permanent error. Either way, we can clear the | |
1183 | * retry state here in preparation for the next error that may occur. | |
1184 | */ | |
1185 | bp->b_last_error = 0; | |
a5ea70d2 | 1186 | bp->b_retries = 0; |
4dd2eb63 | 1187 | bp->b_first_retry_time = 0; |
0b1b213f | 1188 | |
c90821a2 | 1189 | xfs_buf_do_callbacks(bp); |
adadbeef | 1190 | bp->b_fspriv = NULL; |
cb669ca5 | 1191 | bp->b_iodone = NULL; |
e8aaba9a | 1192 | xfs_buf_ioend(bp); |
1da177e4 LT |
1193 | } |
1194 | ||
1da177e4 LT |
1195 | /* |
1196 | * This is the iodone() function for buffers which have been | |
1197 | * logged. It is called when they are eventually flushed out. | |
1198 | * It should remove the buf item from the AIL, and free the buf item. | |
1199 | * It is called by xfs_buf_iodone_callbacks() above which will take | |
1200 | * care of cleaning up the buffer itself. | |
1201 | */ | |
1da177e4 LT |
1202 | void |
1203 | xfs_buf_iodone( | |
ca30b2a7 CH |
1204 | struct xfs_buf *bp, |
1205 | struct xfs_log_item *lip) | |
1da177e4 | 1206 | { |
ca30b2a7 | 1207 | struct xfs_ail *ailp = lip->li_ailp; |
1da177e4 | 1208 | |
ca30b2a7 | 1209 | ASSERT(BUF_ITEM(lip)->bli_buf == bp); |
1da177e4 | 1210 | |
e1f5dbd7 | 1211 | xfs_buf_rele(bp); |
1da177e4 LT |
1212 | |
1213 | /* | |
1214 | * If we are forcibly shutting down, this may well be | |
1215 | * off the AIL already. That's because we simulate the | |
1216 | * log-committed callbacks to unpin these buffers. Or we may never | |
1217 | * have put this item on AIL because of the transaction was | |
783a2f65 | 1218 | * aborted forcibly. xfs_trans_ail_delete() takes care of these. |
1da177e4 LT |
1219 | * |
1220 | * Either way, AIL is useless if we're forcing a shutdown. | |
1221 | */ | |
fc1829f3 | 1222 | spin_lock(&ailp->xa_lock); |
04913fdd | 1223 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 | 1224 | xfs_buf_item_free(BUF_ITEM(lip)); |
1da177e4 | 1225 | } |
4f6d4e8f CM |
1226 | |
1227 | /* | |
1228 | * Requeue a failed buffer for writeback | |
1229 | * | |
1230 | * Return true if the buffer has been re-queued properly, false otherwise | |
1231 | */ | |
1232 | bool | |
1233 | xfs_buf_resubmit_failed_buffers( | |
1234 | struct xfs_buf *bp, | |
1235 | struct xfs_log_item *lip, | |
1236 | struct list_head *buffer_list) | |
1237 | { | |
1238 | struct xfs_log_item *next; | |
1239 | ||
1240 | /* | |
1241 | * Clear XFS_LI_FAILED flag from all items before resubmit | |
1242 | * | |
1243 | * XFS_LI_FAILED set/clear is protected by xa_lock, caller this | |
1244 | * function already have it acquired | |
1245 | */ | |
1246 | for (; lip; lip = next) { | |
1247 | next = lip->li_bio_list; | |
1248 | xfs_clear_li_failed(lip); | |
1249 | } | |
1250 | ||
1251 | /* Add this buffer back to the delayed write list */ | |
1252 | return xfs_buf_delwri_queue(bp, buffer_list); | |
1253 | } |