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" |
1da177e4 LT |
32 | |
33 | ||
34 | kmem_zone_t *xfs_buf_item_zone; | |
35 | ||
7bfa31d8 CH |
36 | static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip) |
37 | { | |
38 | return container_of(lip, struct xfs_buf_log_item, bli_item); | |
39 | } | |
40 | ||
c90821a2 | 41 | STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp); |
1da177e4 | 42 | |
166d1368 DC |
43 | static inline int |
44 | xfs_buf_log_format_size( | |
45 | struct xfs_buf_log_format *blfp) | |
46 | { | |
47 | return offsetof(struct xfs_buf_log_format, blf_data_map) + | |
48 | (blfp->blf_map_size * sizeof(blfp->blf_data_map[0])); | |
49 | } | |
50 | ||
1da177e4 LT |
51 | /* |
52 | * This returns the number of log iovecs needed to log the | |
53 | * given buf log item. | |
54 | * | |
55 | * It calculates this as 1 iovec for the buf log format structure | |
56 | * and 1 for each stretch of non-contiguous chunks to be logged. | |
57 | * Contiguous chunks are logged in a single iovec. | |
58 | * | |
59 | * If the XFS_BLI_STALE flag has been set, then log nothing. | |
60 | */ | |
166d1368 | 61 | STATIC void |
372cc85e DC |
62 | xfs_buf_item_size_segment( |
63 | struct xfs_buf_log_item *bip, | |
166d1368 DC |
64 | struct xfs_buf_log_format *blfp, |
65 | int *nvecs, | |
66 | int *nbytes) | |
1da177e4 | 67 | { |
7bfa31d8 | 68 | struct xfs_buf *bp = bip->bli_buf; |
7bfa31d8 CH |
69 | int next_bit; |
70 | int last_bit; | |
1da177e4 | 71 | |
372cc85e DC |
72 | last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
73 | if (last_bit == -1) | |
166d1368 | 74 | return; |
372cc85e DC |
75 | |
76 | /* | |
77 | * initial count for a dirty buffer is 2 vectors - the format structure | |
78 | * and the first dirty region. | |
79 | */ | |
166d1368 DC |
80 | *nvecs += 2; |
81 | *nbytes += xfs_buf_log_format_size(blfp) + XFS_BLF_CHUNK; | |
1da177e4 | 82 | |
1da177e4 LT |
83 | while (last_bit != -1) { |
84 | /* | |
85 | * This takes the bit number to start looking from and | |
86 | * returns the next set bit from there. It returns -1 | |
87 | * if there are no more bits set or the start bit is | |
88 | * beyond the end of the bitmap. | |
89 | */ | |
372cc85e DC |
90 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
91 | last_bit + 1); | |
1da177e4 LT |
92 | /* |
93 | * If we run out of bits, leave the loop, | |
94 | * else if we find a new set of bits bump the number of vecs, | |
95 | * else keep scanning the current set of bits. | |
96 | */ | |
97 | if (next_bit == -1) { | |
372cc85e | 98 | break; |
1da177e4 LT |
99 | } else if (next_bit != last_bit + 1) { |
100 | last_bit = next_bit; | |
166d1368 | 101 | (*nvecs)++; |
c1155410 DC |
102 | } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) != |
103 | (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) + | |
104 | XFS_BLF_CHUNK)) { | |
1da177e4 | 105 | last_bit = next_bit; |
166d1368 | 106 | (*nvecs)++; |
1da177e4 LT |
107 | } else { |
108 | last_bit++; | |
109 | } | |
166d1368 | 110 | *nbytes += XFS_BLF_CHUNK; |
1da177e4 | 111 | } |
1da177e4 LT |
112 | } |
113 | ||
114 | /* | |
372cc85e DC |
115 | * This returns the number of log iovecs needed to log the given buf log item. |
116 | * | |
117 | * It calculates this as 1 iovec for the buf log format structure and 1 for each | |
118 | * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged | |
119 | * in a single iovec. | |
120 | * | |
121 | * Discontiguous buffers need a format structure per region that that is being | |
122 | * logged. This makes the changes in the buffer appear to log recovery as though | |
123 | * they came from separate buffers, just like would occur if multiple buffers | |
124 | * were used instead of a single discontiguous buffer. This enables | |
125 | * discontiguous buffers to be in-memory constructs, completely transparent to | |
126 | * what ends up on disk. | |
127 | * | |
128 | * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log | |
129 | * format structures. | |
1da177e4 | 130 | */ |
166d1368 | 131 | STATIC void |
372cc85e | 132 | xfs_buf_item_size( |
166d1368 DC |
133 | struct xfs_log_item *lip, |
134 | int *nvecs, | |
135 | int *nbytes) | |
1da177e4 | 136 | { |
7bfa31d8 | 137 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
372cc85e DC |
138 | int i; |
139 | ||
140 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
141 | if (bip->bli_flags & XFS_BLI_STALE) { | |
142 | /* | |
143 | * The buffer is stale, so all we need to log | |
144 | * is the buf log format structure with the | |
145 | * cancel flag in it. | |
146 | */ | |
147 | trace_xfs_buf_item_size_stale(bip); | |
b9438173 | 148 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
166d1368 DC |
149 | *nvecs += bip->bli_format_count; |
150 | for (i = 0; i < bip->bli_format_count; i++) { | |
151 | *nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]); | |
152 | } | |
153 | return; | |
372cc85e DC |
154 | } |
155 | ||
156 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); | |
157 | ||
5f6bed76 DC |
158 | if (bip->bli_flags & XFS_BLI_ORDERED) { |
159 | /* | |
160 | * The buffer has been logged just to order it. | |
161 | * It is not being included in the transaction | |
162 | * commit, so no vectors are used at all. | |
163 | */ | |
164 | trace_xfs_buf_item_size_ordered(bip); | |
166d1368 DC |
165 | *nvecs = XFS_LOG_VEC_ORDERED; |
166 | return; | |
5f6bed76 DC |
167 | } |
168 | ||
372cc85e DC |
169 | /* |
170 | * the vector count is based on the number of buffer vectors we have | |
171 | * dirty bits in. This will only be greater than one when we have a | |
172 | * compound buffer with more than one segment dirty. Hence for compound | |
173 | * buffers we need to track which segment the dirty bits correspond to, | |
174 | * and when we move from one segment to the next increment the vector | |
175 | * count for the extra buf log format structure that will need to be | |
176 | * written. | |
177 | */ | |
372cc85e | 178 | for (i = 0; i < bip->bli_format_count; i++) { |
166d1368 DC |
179 | xfs_buf_item_size_segment(bip, &bip->bli_formats[i], |
180 | nvecs, nbytes); | |
372cc85e | 181 | } |
372cc85e | 182 | trace_xfs_buf_item_size(bip); |
372cc85e DC |
183 | } |
184 | ||
1234351c | 185 | static inline void |
7aeb7222 | 186 | xfs_buf_item_copy_iovec( |
bde7cff6 | 187 | struct xfs_log_vec *lv, |
1234351c | 188 | struct xfs_log_iovec **vecp, |
7aeb7222 CH |
189 | struct xfs_buf *bp, |
190 | uint offset, | |
191 | int first_bit, | |
192 | uint nbits) | |
193 | { | |
194 | offset += first_bit * XFS_BLF_CHUNK; | |
bde7cff6 | 195 | xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK, |
1234351c CH |
196 | xfs_buf_offset(bp, offset), |
197 | nbits * XFS_BLF_CHUNK); | |
7aeb7222 CH |
198 | } |
199 | ||
200 | static inline bool | |
201 | xfs_buf_item_straddle( | |
202 | struct xfs_buf *bp, | |
203 | uint offset, | |
204 | int next_bit, | |
205 | int last_bit) | |
206 | { | |
207 | return xfs_buf_offset(bp, offset + (next_bit << XFS_BLF_SHIFT)) != | |
208 | (xfs_buf_offset(bp, offset + (last_bit << XFS_BLF_SHIFT)) + | |
209 | XFS_BLF_CHUNK); | |
210 | } | |
211 | ||
1234351c | 212 | static void |
372cc85e DC |
213 | xfs_buf_item_format_segment( |
214 | struct xfs_buf_log_item *bip, | |
bde7cff6 | 215 | struct xfs_log_vec *lv, |
1234351c | 216 | struct xfs_log_iovec **vecp, |
372cc85e DC |
217 | uint offset, |
218 | struct xfs_buf_log_format *blfp) | |
219 | { | |
7bfa31d8 | 220 | struct xfs_buf *bp = bip->bli_buf; |
1da177e4 | 221 | uint base_size; |
1da177e4 LT |
222 | int first_bit; |
223 | int last_bit; | |
224 | int next_bit; | |
225 | uint nbits; | |
1da177e4 | 226 | |
372cc85e | 227 | /* copy the flags across from the base format item */ |
b9438173 | 228 | blfp->blf_flags = bip->__bli_format.blf_flags; |
1da177e4 LT |
229 | |
230 | /* | |
77c1a08f DC |
231 | * Base size is the actual size of the ondisk structure - it reflects |
232 | * the actual size of the dirty bitmap rather than the size of the in | |
233 | * memory structure. | |
1da177e4 | 234 | */ |
166d1368 | 235 | base_size = xfs_buf_log_format_size(blfp); |
820a554f | 236 | |
820a554f MT |
237 | first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
238 | if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) { | |
239 | /* | |
240 | * If the map is not be dirty in the transaction, mark | |
241 | * the size as zero and do not advance the vector pointer. | |
242 | */ | |
bde7cff6 | 243 | return; |
820a554f MT |
244 | } |
245 | ||
bde7cff6 CH |
246 | blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size); |
247 | blfp->blf_size = 1; | |
1da177e4 LT |
248 | |
249 | if (bip->bli_flags & XFS_BLI_STALE) { | |
250 | /* | |
251 | * The buffer is stale, so all we need to log | |
252 | * is the buf log format structure with the | |
253 | * cancel flag in it. | |
254 | */ | |
0b1b213f | 255 | trace_xfs_buf_item_format_stale(bip); |
372cc85e | 256 | ASSERT(blfp->blf_flags & XFS_BLF_CANCEL); |
bde7cff6 | 257 | return; |
1da177e4 LT |
258 | } |
259 | ||
5f6bed76 | 260 | |
1da177e4 LT |
261 | /* |
262 | * Fill in an iovec for each set of contiguous chunks. | |
263 | */ | |
1da177e4 LT |
264 | last_bit = first_bit; |
265 | nbits = 1; | |
266 | for (;;) { | |
267 | /* | |
268 | * This takes the bit number to start looking from and | |
269 | * returns the next set bit from there. It returns -1 | |
270 | * if there are no more bits set or the start bit is | |
271 | * beyond the end of the bitmap. | |
272 | */ | |
372cc85e DC |
273 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
274 | (uint)last_bit + 1); | |
1da177e4 | 275 | /* |
7aeb7222 CH |
276 | * If we run out of bits fill in the last iovec and get out of |
277 | * the loop. Else if we start a new set of bits then fill in | |
278 | * the iovec for the series we were looking at and start | |
279 | * counting the bits in the new one. Else we're still in the | |
280 | * same set of bits so just keep counting and scanning. | |
1da177e4 LT |
281 | */ |
282 | if (next_bit == -1) { | |
bde7cff6 | 283 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
7aeb7222 | 284 | first_bit, nbits); |
bde7cff6 | 285 | blfp->blf_size++; |
1da177e4 | 286 | break; |
7aeb7222 CH |
287 | } else if (next_bit != last_bit + 1 || |
288 | xfs_buf_item_straddle(bp, offset, next_bit, last_bit)) { | |
bde7cff6 | 289 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
1234351c | 290 | first_bit, nbits); |
bde7cff6 | 291 | blfp->blf_size++; |
1da177e4 LT |
292 | first_bit = next_bit; |
293 | last_bit = next_bit; | |
294 | nbits = 1; | |
295 | } else { | |
296 | last_bit++; | |
297 | nbits++; | |
298 | } | |
299 | } | |
372cc85e DC |
300 | } |
301 | ||
302 | /* | |
303 | * This is called to fill in the vector of log iovecs for the | |
304 | * given log buf item. It fills the first entry with a buf log | |
305 | * format structure, and the rest point to contiguous chunks | |
306 | * within the buffer. | |
307 | */ | |
308 | STATIC void | |
309 | xfs_buf_item_format( | |
310 | struct xfs_log_item *lip, | |
bde7cff6 | 311 | struct xfs_log_vec *lv) |
372cc85e DC |
312 | { |
313 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); | |
314 | struct xfs_buf *bp = bip->bli_buf; | |
bde7cff6 | 315 | struct xfs_log_iovec *vecp = NULL; |
372cc85e DC |
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)); | |
0d612fb5 DC |
322 | ASSERT((bip->bli_flags & XFS_BLI_STALE) || |
323 | (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF | |
324 | && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF)); | |
325 | ||
372cc85e DC |
326 | |
327 | /* | |
328 | * If it is an inode buffer, transfer the in-memory state to the | |
ddf6ad01 DC |
329 | * format flags and clear the in-memory state. |
330 | * | |
331 | * For buffer based inode allocation, we do not transfer | |
372cc85e DC |
332 | * this state if the inode buffer allocation has not yet been committed |
333 | * to the log as setting the XFS_BLI_INODE_BUF flag will prevent | |
334 | * correct replay of the inode allocation. | |
ddf6ad01 DC |
335 | * |
336 | * For icreate item based inode allocation, the buffers aren't written | |
337 | * to the journal during allocation, and hence we should always tag the | |
338 | * buffer as an inode buffer so that the correct unlinked list replay | |
339 | * occurs during recovery. | |
372cc85e DC |
340 | */ |
341 | if (bip->bli_flags & XFS_BLI_INODE_BUF) { | |
ddf6ad01 DC |
342 | if (xfs_sb_version_hascrc(&lip->li_mountp->m_sb) || |
343 | !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && | |
372cc85e | 344 | xfs_log_item_in_current_chkpt(lip))) |
b9438173 | 345 | bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF; |
372cc85e DC |
346 | bip->bli_flags &= ~XFS_BLI_INODE_BUF; |
347 | } | |
348 | ||
5f6bed76 DC |
349 | if ((bip->bli_flags & (XFS_BLI_ORDERED|XFS_BLI_STALE)) == |
350 | XFS_BLI_ORDERED) { | |
351 | /* | |
352 | * The buffer has been logged just to order it. It is not being | |
353 | * included in the transaction commit, so don't format it. | |
354 | */ | |
355 | trace_xfs_buf_item_format_ordered(bip); | |
356 | return; | |
357 | } | |
358 | ||
372cc85e | 359 | for (i = 0; i < bip->bli_format_count; i++) { |
bde7cff6 | 360 | xfs_buf_item_format_segment(bip, lv, &vecp, offset, |
1234351c | 361 | &bip->bli_formats[i]); |
372cc85e DC |
362 | offset += bp->b_maps[i].bm_len; |
363 | } | |
1da177e4 LT |
364 | |
365 | /* | |
366 | * Check to make sure everything is consistent. | |
367 | */ | |
0b1b213f | 368 | trace_xfs_buf_item_format(bip); |
1da177e4 LT |
369 | } |
370 | ||
371 | /* | |
64fc35de | 372 | * This is called to pin the buffer associated with the buf log item in memory |
4d16e924 | 373 | * so it cannot be written out. |
64fc35de DC |
374 | * |
375 | * We also always take a reference to the buffer log item here so that the bli | |
376 | * is held while the item is pinned in memory. This means that we can | |
377 | * unconditionally drop the reference count a transaction holds when the | |
378 | * transaction is completed. | |
1da177e4 | 379 | */ |
ba0f32d4 | 380 | STATIC void |
1da177e4 | 381 | xfs_buf_item_pin( |
7bfa31d8 | 382 | struct xfs_log_item *lip) |
1da177e4 | 383 | { |
7bfa31d8 | 384 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
1da177e4 | 385 | |
1da177e4 LT |
386 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
387 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
5f6bed76 | 388 | (bip->bli_flags & XFS_BLI_ORDERED) || |
1da177e4 | 389 | (bip->bli_flags & XFS_BLI_STALE)); |
7bfa31d8 | 390 | |
0b1b213f | 391 | trace_xfs_buf_item_pin(bip); |
4d16e924 CH |
392 | |
393 | atomic_inc(&bip->bli_refcount); | |
394 | atomic_inc(&bip->bli_buf->b_pin_count); | |
1da177e4 LT |
395 | } |
396 | ||
1da177e4 LT |
397 | /* |
398 | * This is called to unpin the buffer associated with the buf log | |
399 | * item which was previously pinned with a call to xfs_buf_item_pin(). | |
1da177e4 LT |
400 | * |
401 | * Also drop the reference to the buf item for the current transaction. | |
402 | * If the XFS_BLI_STALE flag is set and we are the last reference, | |
403 | * then free up the buf log item and unlock the buffer. | |
9412e318 CH |
404 | * |
405 | * If the remove flag is set we are called from uncommit in the | |
406 | * forced-shutdown path. If that is true and the reference count on | |
407 | * the log item is going to drop to zero we need to free the item's | |
408 | * descriptor in the transaction. | |
1da177e4 | 409 | */ |
ba0f32d4 | 410 | STATIC void |
1da177e4 | 411 | xfs_buf_item_unpin( |
7bfa31d8 | 412 | struct xfs_log_item *lip, |
9412e318 | 413 | int remove) |
1da177e4 | 414 | { |
7bfa31d8 | 415 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
9412e318 | 416 | xfs_buf_t *bp = bip->bli_buf; |
7bfa31d8 | 417 | struct xfs_ail *ailp = lip->li_ailp; |
8e123850 | 418 | int stale = bip->bli_flags & XFS_BLI_STALE; |
7bfa31d8 | 419 | int freed; |
1da177e4 | 420 | |
adadbeef | 421 | ASSERT(bp->b_fspriv == bip); |
1da177e4 | 422 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
9412e318 | 423 | |
0b1b213f | 424 | trace_xfs_buf_item_unpin(bip); |
1da177e4 LT |
425 | |
426 | freed = atomic_dec_and_test(&bip->bli_refcount); | |
4d16e924 CH |
427 | |
428 | if (atomic_dec_and_test(&bp->b_pin_count)) | |
429 | wake_up_all(&bp->b_waiters); | |
7bfa31d8 | 430 | |
1da177e4 LT |
431 | if (freed && stale) { |
432 | ASSERT(bip->bli_flags & XFS_BLI_STALE); | |
0c842ad4 | 433 | ASSERT(xfs_buf_islocked(bp)); |
5cfd28b6 | 434 | ASSERT(bp->b_flags & XBF_STALE); |
b9438173 | 435 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
9412e318 | 436 | |
0b1b213f CH |
437 | trace_xfs_buf_item_unpin_stale(bip); |
438 | ||
9412e318 CH |
439 | if (remove) { |
440 | /* | |
e34a314c DC |
441 | * If we are in a transaction context, we have to |
442 | * remove the log item from the transaction as we are | |
443 | * about to release our reference to the buffer. If we | |
444 | * don't, the unlock that occurs later in | |
445 | * xfs_trans_uncommit() will try to reference the | |
9412e318 CH |
446 | * buffer which we no longer have a hold on. |
447 | */ | |
e34a314c DC |
448 | if (lip->li_desc) |
449 | xfs_trans_del_item(lip); | |
9412e318 CH |
450 | |
451 | /* | |
452 | * Since the transaction no longer refers to the buffer, | |
453 | * the buffer should no longer refer to the transaction. | |
454 | */ | |
bf9d9013 | 455 | bp->b_transp = NULL; |
9412e318 CH |
456 | } |
457 | ||
1da177e4 LT |
458 | /* |
459 | * If we get called here because of an IO error, we may | |
783a2f65 | 460 | * or may not have the item on the AIL. xfs_trans_ail_delete() |
1da177e4 | 461 | * will take care of that situation. |
783a2f65 | 462 | * xfs_trans_ail_delete() drops the AIL lock. |
1da177e4 LT |
463 | */ |
464 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { | |
c90821a2 | 465 | xfs_buf_do_callbacks(bp); |
adadbeef | 466 | bp->b_fspriv = NULL; |
cb669ca5 | 467 | bp->b_iodone = NULL; |
1da177e4 | 468 | } else { |
783a2f65 | 469 | spin_lock(&ailp->xa_lock); |
04913fdd | 470 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR); |
1da177e4 | 471 | xfs_buf_item_relse(bp); |
adadbeef | 472 | ASSERT(bp->b_fspriv == NULL); |
1da177e4 LT |
473 | } |
474 | xfs_buf_relse(bp); | |
960c60af | 475 | } else if (freed && remove) { |
137fff09 DC |
476 | /* |
477 | * There are currently two references to the buffer - the active | |
478 | * LRU reference and the buf log item. What we are about to do | |
479 | * here - simulate a failed IO completion - requires 3 | |
480 | * references. | |
481 | * | |
482 | * The LRU reference is removed by the xfs_buf_stale() call. The | |
483 | * buf item reference is removed by the xfs_buf_iodone() | |
484 | * callback that is run by xfs_buf_do_callbacks() during ioend | |
485 | * processing (via the bp->b_iodone callback), and then finally | |
486 | * the ioend processing will drop the IO reference if the buffer | |
487 | * is marked XBF_ASYNC. | |
488 | * | |
489 | * Hence we need to take an additional reference here so that IO | |
490 | * completion processing doesn't free the buffer prematurely. | |
491 | */ | |
960c60af | 492 | xfs_buf_lock(bp); |
137fff09 DC |
493 | xfs_buf_hold(bp); |
494 | bp->b_flags |= XBF_ASYNC; | |
2451337d | 495 | xfs_buf_ioerror(bp, -EIO); |
b0388bf1 | 496 | bp->b_flags &= ~XBF_DONE; |
960c60af | 497 | xfs_buf_stale(bp); |
e8aaba9a | 498 | xfs_buf_ioend(bp); |
1da177e4 LT |
499 | } |
500 | } | |
501 | ||
ac8809f9 DC |
502 | /* |
503 | * Buffer IO error rate limiting. Limit it to no more than 10 messages per 30 | |
504 | * seconds so as to not spam logs too much on repeated detection of the same | |
505 | * buffer being bad.. | |
506 | */ | |
507 | ||
02cc1876 | 508 | static DEFINE_RATELIMIT_STATE(xfs_buf_write_fail_rl_state, 30 * HZ, 10); |
ac8809f9 | 509 | |
ba0f32d4 | 510 | STATIC uint |
43ff2122 CH |
511 | xfs_buf_item_push( |
512 | struct xfs_log_item *lip, | |
513 | struct list_head *buffer_list) | |
1da177e4 | 514 | { |
7bfa31d8 CH |
515 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
516 | struct xfs_buf *bp = bip->bli_buf; | |
43ff2122 | 517 | uint rval = XFS_ITEM_SUCCESS; |
1da177e4 | 518 | |
811e64c7 | 519 | if (xfs_buf_ispinned(bp)) |
1da177e4 | 520 | return XFS_ITEM_PINNED; |
5337fe9b BF |
521 | if (!xfs_buf_trylock(bp)) { |
522 | /* | |
523 | * If we have just raced with a buffer being pinned and it has | |
524 | * been marked stale, we could end up stalling until someone else | |
525 | * issues a log force to unpin the stale buffer. Check for the | |
526 | * race condition here so xfsaild recognizes the buffer is pinned | |
527 | * and queues a log force to move it along. | |
528 | */ | |
529 | if (xfs_buf_ispinned(bp)) | |
530 | return XFS_ITEM_PINNED; | |
1da177e4 | 531 | return XFS_ITEM_LOCKED; |
5337fe9b | 532 | } |
1da177e4 | 533 | |
1da177e4 | 534 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
43ff2122 CH |
535 | |
536 | trace_xfs_buf_item_push(bip); | |
537 | ||
ac8809f9 DC |
538 | /* has a previous flush failed due to IO errors? */ |
539 | if ((bp->b_flags & XBF_WRITE_FAIL) && | |
fdadf267 | 540 | ___ratelimit(&xfs_buf_write_fail_rl_state, "XFS: Failing async write")) { |
ac8809f9 | 541 | xfs_warn(bp->b_target->bt_mount, |
fdadf267 | 542 | "Failing async write on buffer block 0x%llx. Retrying async write.", |
ac8809f9 DC |
543 | (long long)bp->b_bn); |
544 | } | |
545 | ||
43ff2122 CH |
546 | if (!xfs_buf_delwri_queue(bp, buffer_list)) |
547 | rval = XFS_ITEM_FLUSHING; | |
548 | xfs_buf_unlock(bp); | |
549 | return rval; | |
1da177e4 LT |
550 | } |
551 | ||
552 | /* | |
64fc35de DC |
553 | * Release the buffer associated with the buf log item. If there is no dirty |
554 | * logged data associated with the buffer recorded in the buf log item, then | |
555 | * free the buf log item and remove the reference to it in the buffer. | |
1da177e4 | 556 | * |
64fc35de DC |
557 | * This call ignores the recursion count. It is only called when the buffer |
558 | * should REALLY be unlocked, regardless of the recursion count. | |
1da177e4 | 559 | * |
64fc35de DC |
560 | * We unconditionally drop the transaction's reference to the log item. If the |
561 | * item was logged, then another reference was taken when it was pinned, so we | |
562 | * can safely drop the transaction reference now. This also allows us to avoid | |
563 | * potential races with the unpin code freeing the bli by not referencing the | |
564 | * bli after we've dropped the reference count. | |
565 | * | |
566 | * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item | |
567 | * if necessary but do not unlock the buffer. This is for support of | |
568 | * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't | |
569 | * free the item. | |
1da177e4 | 570 | */ |
ba0f32d4 | 571 | STATIC void |
1da177e4 | 572 | xfs_buf_item_unlock( |
7bfa31d8 | 573 | struct xfs_log_item *lip) |
1da177e4 | 574 | { |
7bfa31d8 CH |
575 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
576 | struct xfs_buf *bp = bip->bli_buf; | |
5f6bed76 DC |
577 | bool clean; |
578 | bool aborted; | |
579 | int flags; | |
1da177e4 | 580 | |
64fc35de | 581 | /* Clear the buffer's association with this transaction. */ |
bf9d9013 | 582 | bp->b_transp = NULL; |
1da177e4 LT |
583 | |
584 | /* | |
64fc35de DC |
585 | * If this is a transaction abort, don't return early. Instead, allow |
586 | * the brelse to happen. Normally it would be done for stale | |
587 | * (cancelled) buffers at unpin time, but we'll never go through the | |
588 | * pin/unpin cycle if we abort inside commit. | |
1da177e4 | 589 | */ |
5f6bed76 | 590 | aborted = (lip->li_flags & XFS_LI_ABORTED) ? true : false; |
1da177e4 | 591 | /* |
5f6bed76 DC |
592 | * Before possibly freeing the buf item, copy the per-transaction state |
593 | * so we can reference it safely later after clearing it from the | |
594 | * buffer log item. | |
64fc35de | 595 | */ |
5f6bed76 DC |
596 | flags = bip->bli_flags; |
597 | bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED); | |
64fc35de DC |
598 | |
599 | /* | |
600 | * If the buf item is marked stale, then don't do anything. We'll | |
601 | * unlock the buffer and free the buf item when the buffer is unpinned | |
602 | * for the last time. | |
1da177e4 | 603 | */ |
5f6bed76 | 604 | if (flags & XFS_BLI_STALE) { |
0b1b213f | 605 | trace_xfs_buf_item_unlock_stale(bip); |
b9438173 | 606 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
64fc35de DC |
607 | if (!aborted) { |
608 | atomic_dec(&bip->bli_refcount); | |
1da177e4 | 609 | return; |
64fc35de | 610 | } |
1da177e4 LT |
611 | } |
612 | ||
0b1b213f | 613 | trace_xfs_buf_item_unlock(bip); |
1da177e4 LT |
614 | |
615 | /* | |
64fc35de | 616 | * If the buf item isn't tracking any data, free it, otherwise drop the |
3b19034d DC |
617 | * reference we hold to it. If we are aborting the transaction, this may |
618 | * be the only reference to the buf item, so we free it anyway | |
619 | * regardless of whether it is dirty or not. A dirty abort implies a | |
620 | * shutdown, anyway. | |
5f6bed76 DC |
621 | * |
622 | * Ordered buffers are dirty but may have no recorded changes, so ensure | |
623 | * we only release clean items here. | |
1da177e4 | 624 | */ |
5f6bed76 DC |
625 | clean = (flags & XFS_BLI_DIRTY) ? false : true; |
626 | if (clean) { | |
627 | int i; | |
628 | for (i = 0; i < bip->bli_format_count; i++) { | |
629 | if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map, | |
630 | bip->bli_formats[i].blf_map_size)) { | |
631 | clean = false; | |
632 | break; | |
633 | } | |
c883d0c4 MT |
634 | } |
635 | } | |
46f9d2eb DC |
636 | |
637 | /* | |
638 | * Clean buffers, by definition, cannot be in the AIL. However, aborted | |
639 | * buffers may be dirty and hence in the AIL. Therefore if we are | |
640 | * aborting a buffer and we've just taken the last refernce away, we | |
641 | * have to check if it is in the AIL before freeing it. We need to free | |
642 | * it in this case, because an aborted transaction has already shut the | |
643 | * filesystem down and this is the last chance we will have to do so. | |
644 | */ | |
645 | if (atomic_dec_and_test(&bip->bli_refcount)) { | |
646 | if (clean) | |
647 | xfs_buf_item_relse(bp); | |
648 | else if (aborted) { | |
649 | ASSERT(XFS_FORCED_SHUTDOWN(lip->li_mountp)); | |
146e54b7 | 650 | xfs_trans_ail_remove(lip, SHUTDOWN_LOG_IO_ERROR); |
3b19034d DC |
651 | xfs_buf_item_relse(bp); |
652 | } | |
46f9d2eb | 653 | } |
1da177e4 | 654 | |
5f6bed76 | 655 | if (!(flags & XFS_BLI_HOLD)) |
1da177e4 | 656 | xfs_buf_relse(bp); |
1da177e4 LT |
657 | } |
658 | ||
659 | /* | |
660 | * This is called to find out where the oldest active copy of the | |
661 | * buf log item in the on disk log resides now that the last log | |
662 | * write of it completed at the given lsn. | |
663 | * We always re-log all the dirty data in a buffer, so usually the | |
664 | * latest copy in the on disk log is the only one that matters. For | |
665 | * those cases we simply return the given lsn. | |
666 | * | |
667 | * The one exception to this is for buffers full of newly allocated | |
668 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF | |
669 | * flag set, indicating that only the di_next_unlinked fields from the | |
670 | * inodes in the buffers will be replayed during recovery. If the | |
671 | * original newly allocated inode images have not yet been flushed | |
672 | * when the buffer is so relogged, then we need to make sure that we | |
673 | * keep the old images in the 'active' portion of the log. We do this | |
674 | * by returning the original lsn of that transaction here rather than | |
675 | * the current one. | |
676 | */ | |
ba0f32d4 | 677 | STATIC xfs_lsn_t |
1da177e4 | 678 | xfs_buf_item_committed( |
7bfa31d8 | 679 | struct xfs_log_item *lip, |
1da177e4 LT |
680 | xfs_lsn_t lsn) |
681 | { | |
7bfa31d8 CH |
682 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
683 | ||
0b1b213f CH |
684 | trace_xfs_buf_item_committed(bip); |
685 | ||
7bfa31d8 CH |
686 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0) |
687 | return lip->li_lsn; | |
688 | return lsn; | |
1da177e4 LT |
689 | } |
690 | ||
ba0f32d4 | 691 | STATIC void |
7bfa31d8 CH |
692 | xfs_buf_item_committing( |
693 | struct xfs_log_item *lip, | |
694 | xfs_lsn_t commit_lsn) | |
1da177e4 LT |
695 | { |
696 | } | |
697 | ||
698 | /* | |
699 | * This is the ops vector shared by all buf log items. | |
700 | */ | |
272e42b2 | 701 | static const struct xfs_item_ops xfs_buf_item_ops = { |
7bfa31d8 CH |
702 | .iop_size = xfs_buf_item_size, |
703 | .iop_format = xfs_buf_item_format, | |
704 | .iop_pin = xfs_buf_item_pin, | |
705 | .iop_unpin = xfs_buf_item_unpin, | |
7bfa31d8 CH |
706 | .iop_unlock = xfs_buf_item_unlock, |
707 | .iop_committed = xfs_buf_item_committed, | |
708 | .iop_push = xfs_buf_item_push, | |
7bfa31d8 | 709 | .iop_committing = xfs_buf_item_committing |
1da177e4 LT |
710 | }; |
711 | ||
372cc85e DC |
712 | STATIC int |
713 | xfs_buf_item_get_format( | |
714 | struct xfs_buf_log_item *bip, | |
715 | int count) | |
716 | { | |
717 | ASSERT(bip->bli_formats == NULL); | |
718 | bip->bli_format_count = count; | |
719 | ||
720 | if (count == 1) { | |
b9438173 | 721 | bip->bli_formats = &bip->__bli_format; |
372cc85e DC |
722 | return 0; |
723 | } | |
724 | ||
725 | bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format), | |
726 | KM_SLEEP); | |
727 | if (!bip->bli_formats) | |
2451337d | 728 | return -ENOMEM; |
372cc85e DC |
729 | return 0; |
730 | } | |
731 | ||
732 | STATIC void | |
733 | xfs_buf_item_free_format( | |
734 | struct xfs_buf_log_item *bip) | |
735 | { | |
b9438173 | 736 | if (bip->bli_formats != &bip->__bli_format) { |
372cc85e DC |
737 | kmem_free(bip->bli_formats); |
738 | bip->bli_formats = NULL; | |
739 | } | |
740 | } | |
1da177e4 LT |
741 | |
742 | /* | |
743 | * Allocate a new buf log item to go with the given buffer. | |
744 | * Set the buffer's b_fsprivate field to point to the new | |
745 | * buf log item. If there are other item's attached to the | |
746 | * buffer (see xfs_buf_attach_iodone() below), then put the | |
747 | * buf log item at the front. | |
748 | */ | |
f79af0b9 | 749 | int |
1da177e4 | 750 | xfs_buf_item_init( |
f79af0b9 DC |
751 | struct xfs_buf *bp, |
752 | struct xfs_mount *mp) | |
1da177e4 | 753 | { |
f79af0b9 DC |
754 | struct xfs_log_item *lip = bp->b_fspriv; |
755 | struct xfs_buf_log_item *bip; | |
1da177e4 LT |
756 | int chunks; |
757 | int map_size; | |
372cc85e DC |
758 | int error; |
759 | int i; | |
1da177e4 LT |
760 | |
761 | /* | |
762 | * Check to see if there is already a buf log item for | |
763 | * this buffer. If there is, it is guaranteed to be | |
764 | * the first. If we do already have one, there is | |
765 | * nothing to do here so return. | |
766 | */ | |
ebad861b | 767 | ASSERT(bp->b_target->bt_mount == mp); |
adadbeef | 768 | if (lip != NULL && lip->li_type == XFS_LI_BUF) |
f79af0b9 | 769 | return 0; |
1da177e4 | 770 | |
372cc85e | 771 | bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP); |
43f5efc5 | 772 | xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops); |
1da177e4 | 773 | bip->bli_buf = bp; |
372cc85e DC |
774 | |
775 | /* | |
776 | * chunks is the number of XFS_BLF_CHUNK size pieces the buffer | |
777 | * can be divided into. Make sure not to truncate any pieces. | |
778 | * map_size is the size of the bitmap needed to describe the | |
779 | * chunks of the buffer. | |
780 | * | |
781 | * Discontiguous buffer support follows the layout of the underlying | |
782 | * buffer. This makes the implementation as simple as possible. | |
783 | */ | |
784 | error = xfs_buf_item_get_format(bip, bp->b_map_count); | |
785 | ASSERT(error == 0); | |
f79af0b9 DC |
786 | if (error) { /* to stop gcc throwing set-but-unused warnings */ |
787 | kmem_zone_free(xfs_buf_item_zone, bip); | |
788 | return error; | |
789 | } | |
790 | ||
372cc85e DC |
791 | |
792 | for (i = 0; i < bip->bli_format_count; i++) { | |
793 | chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len), | |
794 | XFS_BLF_CHUNK); | |
795 | map_size = DIV_ROUND_UP(chunks, NBWORD); | |
796 | ||
797 | bip->bli_formats[i].blf_type = XFS_LI_BUF; | |
798 | bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn; | |
799 | bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len; | |
800 | bip->bli_formats[i].blf_map_size = map_size; | |
801 | } | |
1da177e4 | 802 | |
1da177e4 LT |
803 | /* |
804 | * Put the buf item into the list of items attached to the | |
805 | * buffer at the front. | |
806 | */ | |
adadbeef CH |
807 | if (bp->b_fspriv) |
808 | bip->bli_item.li_bio_list = bp->b_fspriv; | |
809 | bp->b_fspriv = bip; | |
f79af0b9 DC |
810 | xfs_buf_hold(bp); |
811 | return 0; | |
1da177e4 LT |
812 | } |
813 | ||
814 | ||
815 | /* | |
816 | * Mark bytes first through last inclusive as dirty in the buf | |
817 | * item's bitmap. | |
818 | */ | |
632b89e8 | 819 | static void |
372cc85e | 820 | xfs_buf_item_log_segment( |
1da177e4 | 821 | uint first, |
372cc85e DC |
822 | uint last, |
823 | uint *map) | |
1da177e4 LT |
824 | { |
825 | uint first_bit; | |
826 | uint last_bit; | |
827 | uint bits_to_set; | |
828 | uint bits_set; | |
829 | uint word_num; | |
830 | uint *wordp; | |
831 | uint bit; | |
832 | uint end_bit; | |
833 | uint mask; | |
834 | ||
1da177e4 LT |
835 | /* |
836 | * Convert byte offsets to bit numbers. | |
837 | */ | |
c1155410 DC |
838 | first_bit = first >> XFS_BLF_SHIFT; |
839 | last_bit = last >> XFS_BLF_SHIFT; | |
1da177e4 LT |
840 | |
841 | /* | |
842 | * Calculate the total number of bits to be set. | |
843 | */ | |
844 | bits_to_set = last_bit - first_bit + 1; | |
845 | ||
846 | /* | |
847 | * Get a pointer to the first word in the bitmap | |
848 | * to set a bit in. | |
849 | */ | |
850 | word_num = first_bit >> BIT_TO_WORD_SHIFT; | |
372cc85e | 851 | wordp = &map[word_num]; |
1da177e4 LT |
852 | |
853 | /* | |
854 | * Calculate the starting bit in the first word. | |
855 | */ | |
856 | bit = first_bit & (uint)(NBWORD - 1); | |
857 | ||
858 | /* | |
859 | * First set any bits in the first word of our range. | |
860 | * If it starts at bit 0 of the word, it will be | |
861 | * set below rather than here. That is what the variable | |
862 | * bit tells us. The variable bits_set tracks the number | |
863 | * of bits that have been set so far. End_bit is the number | |
864 | * of the last bit to be set in this word plus one. | |
865 | */ | |
866 | if (bit) { | |
867 | end_bit = MIN(bit + bits_to_set, (uint)NBWORD); | |
868 | mask = ((1 << (end_bit - bit)) - 1) << bit; | |
869 | *wordp |= mask; | |
870 | wordp++; | |
871 | bits_set = end_bit - bit; | |
872 | } else { | |
873 | bits_set = 0; | |
874 | } | |
875 | ||
876 | /* | |
877 | * Now set bits a whole word at a time that are between | |
878 | * first_bit and last_bit. | |
879 | */ | |
880 | while ((bits_to_set - bits_set) >= NBWORD) { | |
881 | *wordp |= 0xffffffff; | |
882 | bits_set += NBWORD; | |
883 | wordp++; | |
884 | } | |
885 | ||
886 | /* | |
887 | * Finally, set any bits left to be set in one last partial word. | |
888 | */ | |
889 | end_bit = bits_to_set - bits_set; | |
890 | if (end_bit) { | |
891 | mask = (1 << end_bit) - 1; | |
892 | *wordp |= mask; | |
893 | } | |
1da177e4 LT |
894 | } |
895 | ||
372cc85e DC |
896 | /* |
897 | * Mark bytes first through last inclusive as dirty in the buf | |
898 | * item's bitmap. | |
899 | */ | |
900 | void | |
901 | xfs_buf_item_log( | |
902 | xfs_buf_log_item_t *bip, | |
903 | uint first, | |
904 | uint last) | |
905 | { | |
906 | int i; | |
907 | uint start; | |
908 | uint end; | |
909 | struct xfs_buf *bp = bip->bli_buf; | |
910 | ||
372cc85e DC |
911 | /* |
912 | * walk each buffer segment and mark them dirty appropriately. | |
913 | */ | |
914 | start = 0; | |
915 | for (i = 0; i < bip->bli_format_count; i++) { | |
916 | if (start > last) | |
917 | break; | |
918 | end = start + BBTOB(bp->b_maps[i].bm_len); | |
919 | if (first > end) { | |
920 | start += BBTOB(bp->b_maps[i].bm_len); | |
921 | continue; | |
922 | } | |
923 | if (first < start) | |
924 | first = start; | |
925 | if (end > last) | |
926 | end = last; | |
927 | ||
72b0636b | 928 | xfs_buf_item_log_segment(first, end, |
372cc85e DC |
929 | &bip->bli_formats[i].blf_data_map[0]); |
930 | ||
931 | start += bp->b_maps[i].bm_len; | |
932 | } | |
933 | } | |
934 | ||
1da177e4 LT |
935 | |
936 | /* | |
5f6bed76 | 937 | * Return 1 if the buffer has been logged or ordered in a transaction (at any |
1da177e4 LT |
938 | * point, not just the current transaction) and 0 if not. |
939 | */ | |
940 | uint | |
941 | xfs_buf_item_dirty( | |
942 | xfs_buf_log_item_t *bip) | |
943 | { | |
944 | return (bip->bli_flags & XFS_BLI_DIRTY); | |
945 | } | |
946 | ||
e1f5dbd7 LM |
947 | STATIC void |
948 | xfs_buf_item_free( | |
949 | xfs_buf_log_item_t *bip) | |
950 | { | |
372cc85e | 951 | xfs_buf_item_free_format(bip); |
e1f5dbd7 LM |
952 | kmem_zone_free(xfs_buf_item_zone, bip); |
953 | } | |
954 | ||
1da177e4 LT |
955 | /* |
956 | * This is called when the buf log item is no longer needed. It should | |
957 | * free the buf log item associated with the given buffer and clear | |
958 | * the buffer's pointer to the buf log item. If there are no more | |
959 | * items in the list, clear the b_iodone field of the buffer (see | |
960 | * xfs_buf_attach_iodone() below). | |
961 | */ | |
962 | void | |
963 | xfs_buf_item_relse( | |
964 | xfs_buf_t *bp) | |
965 | { | |
5f6bed76 | 966 | xfs_buf_log_item_t *bip = bp->b_fspriv; |
1da177e4 | 967 | |
0b1b213f | 968 | trace_xfs_buf_item_relse(bp, _RET_IP_); |
5f6bed76 | 969 | ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL)); |
0b1b213f | 970 | |
adadbeef | 971 | bp->b_fspriv = bip->bli_item.li_bio_list; |
cb669ca5 CH |
972 | if (bp->b_fspriv == NULL) |
973 | bp->b_iodone = NULL; | |
adadbeef | 974 | |
e1f5dbd7 LM |
975 | xfs_buf_rele(bp); |
976 | xfs_buf_item_free(bip); | |
1da177e4 LT |
977 | } |
978 | ||
979 | ||
980 | /* | |
981 | * Add the given log item with its callback to the list of callbacks | |
982 | * to be called when the buffer's I/O completes. If it is not set | |
983 | * already, set the buffer's b_iodone() routine to be | |
984 | * xfs_buf_iodone_callbacks() and link the log item into the list of | |
985 | * items rooted at b_fsprivate. Items are always added as the second | |
986 | * entry in the list if there is a first, because the buf item code | |
987 | * assumes that the buf log item is first. | |
988 | */ | |
989 | void | |
990 | xfs_buf_attach_iodone( | |
991 | xfs_buf_t *bp, | |
992 | void (*cb)(xfs_buf_t *, xfs_log_item_t *), | |
993 | xfs_log_item_t *lip) | |
994 | { | |
995 | xfs_log_item_t *head_lip; | |
996 | ||
0c842ad4 | 997 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
998 | |
999 | lip->li_cb = cb; | |
adadbeef CH |
1000 | head_lip = bp->b_fspriv; |
1001 | if (head_lip) { | |
1da177e4 LT |
1002 | lip->li_bio_list = head_lip->li_bio_list; |
1003 | head_lip->li_bio_list = lip; | |
1004 | } else { | |
adadbeef | 1005 | bp->b_fspriv = lip; |
1da177e4 LT |
1006 | } |
1007 | ||
cb669ca5 CH |
1008 | ASSERT(bp->b_iodone == NULL || |
1009 | bp->b_iodone == xfs_buf_iodone_callbacks); | |
1010 | bp->b_iodone = xfs_buf_iodone_callbacks; | |
1da177e4 LT |
1011 | } |
1012 | ||
c90821a2 DC |
1013 | /* |
1014 | * We can have many callbacks on a buffer. Running the callbacks individually | |
1015 | * can cause a lot of contention on the AIL lock, so we allow for a single | |
1016 | * callback to be able to scan the remaining lip->li_bio_list for other items | |
1017 | * of the same type and callback to be processed in the first call. | |
1018 | * | |
1019 | * As a result, the loop walking the callback list below will also modify the | |
1020 | * list. it removes the first item from the list and then runs the callback. | |
1021 | * The loop then restarts from the new head of the list. This allows the | |
1022 | * callback to scan and modify the list attached to the buffer and we don't | |
1023 | * have to care about maintaining a next item pointer. | |
1024 | */ | |
1da177e4 LT |
1025 | STATIC void |
1026 | xfs_buf_do_callbacks( | |
c90821a2 | 1027 | struct xfs_buf *bp) |
1da177e4 | 1028 | { |
c90821a2 | 1029 | struct xfs_log_item *lip; |
1da177e4 | 1030 | |
adadbeef CH |
1031 | while ((lip = bp->b_fspriv) != NULL) { |
1032 | bp->b_fspriv = lip->li_bio_list; | |
1da177e4 LT |
1033 | ASSERT(lip->li_cb != NULL); |
1034 | /* | |
1035 | * Clear the next pointer so we don't have any | |
1036 | * confusion if the item is added to another buf. | |
1037 | * Don't touch the log item after calling its | |
1038 | * callback, because it could have freed itself. | |
1039 | */ | |
1040 | lip->li_bio_list = NULL; | |
1041 | lip->li_cb(bp, lip); | |
1da177e4 LT |
1042 | } |
1043 | } | |
1044 | ||
df309390 CM |
1045 | static bool |
1046 | xfs_buf_iodone_callback_error( | |
bfc60177 | 1047 | struct xfs_buf *bp) |
1da177e4 | 1048 | { |
bfc60177 CH |
1049 | struct xfs_log_item *lip = bp->b_fspriv; |
1050 | struct xfs_mount *mp = lip->li_mountp; | |
1051 | static ulong lasttime; | |
1052 | static xfs_buftarg_t *lasttarg; | |
df309390 | 1053 | struct xfs_error_cfg *cfg; |
1da177e4 | 1054 | |
bfc60177 CH |
1055 | /* |
1056 | * If we've already decided to shutdown the filesystem because of | |
1057 | * I/O errors, there's no point in giving this a retry. | |
1058 | */ | |
df309390 CM |
1059 | if (XFS_FORCED_SHUTDOWN(mp)) |
1060 | goto out_stale; | |
1da177e4 | 1061 | |
49074c06 | 1062 | if (bp->b_target != lasttarg || |
bfc60177 CH |
1063 | time_after(jiffies, (lasttime + 5*HZ))) { |
1064 | lasttime = jiffies; | |
b38505b0 | 1065 | xfs_buf_ioerror_alert(bp, __func__); |
bfc60177 | 1066 | } |
49074c06 | 1067 | lasttarg = bp->b_target; |
1da177e4 | 1068 | |
df309390 CM |
1069 | /* synchronous writes will have callers process the error */ |
1070 | if (!(bp->b_flags & XBF_ASYNC)) | |
1071 | goto out_stale; | |
1072 | ||
1073 | trace_xfs_buf_item_iodone_async(bp, _RET_IP_); | |
1074 | ASSERT(bp->b_iodone != NULL); | |
1075 | ||
bfc60177 | 1076 | /* |
25985edc | 1077 | * If the write was asynchronous then no one will be looking for the |
df309390 CM |
1078 | * error. If this is the first failure of this type, clear the error |
1079 | * state and write the buffer out again. This means we always retry an | |
1080 | * async write failure at least once, but we also need to set the buffer | |
1081 | * up to behave correctly now for repeated failures. | |
bfc60177 | 1082 | */ |
df309390 CM |
1083 | if (!(bp->b_flags & (XBF_STALE|XBF_WRITE_FAIL)) || |
1084 | bp->b_last_error != bp->b_error) { | |
1085 | bp->b_flags |= (XBF_WRITE | XBF_ASYNC | | |
1086 | XBF_DONE | XBF_WRITE_FAIL); | |
1087 | bp->b_last_error = bp->b_error; | |
a5ea70d2 CM |
1088 | bp->b_retries = 0; |
1089 | bp->b_first_retry_time = jiffies; | |
1090 | ||
df309390 CM |
1091 | xfs_buf_ioerror(bp, 0); |
1092 | xfs_buf_submit(bp); | |
1093 | return true; | |
1094 | } | |
43ff2122 | 1095 | |
df309390 CM |
1096 | /* |
1097 | * Repeated failure on an async write. Take action according to the | |
1098 | * error configuration we have been set up to use. | |
1099 | */ | |
1100 | cfg = xfs_error_get_cfg(mp, XFS_ERR_METADATA, bp->b_error); | |
a5ea70d2 CM |
1101 | |
1102 | if (cfg->max_retries != XFS_ERR_RETRY_FOREVER && | |
1103 | ++bp->b_retries > cfg->max_retries) | |
1104 | goto permanent_error; | |
1105 | if (cfg->retry_timeout && | |
1106 | time_after(jiffies, cfg->retry_timeout + bp->b_first_retry_time)) | |
1107 | goto permanent_error; | |
bfc60177 | 1108 | |
e6b3bb78 CM |
1109 | /* At unmount we may treat errors differently */ |
1110 | if ((mp->m_flags & XFS_MOUNT_UNMOUNTING) && mp->m_fail_unmount) | |
1111 | goto permanent_error; | |
1112 | ||
df309390 CM |
1113 | /* still a transient error, higher layers will retry */ |
1114 | xfs_buf_ioerror(bp, 0); | |
1115 | xfs_buf_relse(bp); | |
1116 | return true; | |
0b1b213f | 1117 | |
bfc60177 | 1118 | /* |
df309390 CM |
1119 | * Permanent error - we need to trigger a shutdown if we haven't already |
1120 | * to indicate that inconsistency will result from this action. | |
bfc60177 | 1121 | */ |
df309390 CM |
1122 | permanent_error: |
1123 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1124 | out_stale: | |
c867cb61 | 1125 | xfs_buf_stale(bp); |
b0388bf1 | 1126 | bp->b_flags |= XBF_DONE; |
0b1b213f | 1127 | trace_xfs_buf_error_relse(bp, _RET_IP_); |
df309390 CM |
1128 | return false; |
1129 | } | |
1130 | ||
1131 | /* | |
1132 | * This is the iodone() function for buffers which have had callbacks attached | |
1133 | * to them by xfs_buf_attach_iodone(). We need to iterate the items on the | |
1134 | * callback list, mark the buffer as having no more callbacks and then push the | |
1135 | * buffer through IO completion processing. | |
1136 | */ | |
1137 | void | |
1138 | xfs_buf_iodone_callbacks( | |
1139 | struct xfs_buf *bp) | |
1140 | { | |
1141 | /* | |
1142 | * If there is an error, process it. Some errors require us | |
1143 | * to run callbacks after failure processing is done so we | |
1144 | * detect that and take appropriate action. | |
1145 | */ | |
1146 | if (bp->b_error && xfs_buf_iodone_callback_error(bp)) | |
1147 | return; | |
1148 | ||
1149 | /* | |
1150 | * Successful IO or permanent error. Either way, we can clear the | |
1151 | * retry state here in preparation for the next error that may occur. | |
1152 | */ | |
1153 | bp->b_last_error = 0; | |
a5ea70d2 | 1154 | bp->b_retries = 0; |
0b1b213f | 1155 | |
c90821a2 | 1156 | xfs_buf_do_callbacks(bp); |
adadbeef | 1157 | bp->b_fspriv = NULL; |
cb669ca5 | 1158 | bp->b_iodone = NULL; |
e8aaba9a | 1159 | xfs_buf_ioend(bp); |
1da177e4 LT |
1160 | } |
1161 | ||
1da177e4 LT |
1162 | /* |
1163 | * This is the iodone() function for buffers which have been | |
1164 | * logged. It is called when they are eventually flushed out. | |
1165 | * It should remove the buf item from the AIL, and free the buf item. | |
1166 | * It is called by xfs_buf_iodone_callbacks() above which will take | |
1167 | * care of cleaning up the buffer itself. | |
1168 | */ | |
1da177e4 LT |
1169 | void |
1170 | xfs_buf_iodone( | |
ca30b2a7 CH |
1171 | struct xfs_buf *bp, |
1172 | struct xfs_log_item *lip) | |
1da177e4 | 1173 | { |
ca30b2a7 | 1174 | struct xfs_ail *ailp = lip->li_ailp; |
1da177e4 | 1175 | |
ca30b2a7 | 1176 | ASSERT(BUF_ITEM(lip)->bli_buf == bp); |
1da177e4 | 1177 | |
e1f5dbd7 | 1178 | xfs_buf_rele(bp); |
1da177e4 LT |
1179 | |
1180 | /* | |
1181 | * If we are forcibly shutting down, this may well be | |
1182 | * off the AIL already. That's because we simulate the | |
1183 | * log-committed callbacks to unpin these buffers. Or we may never | |
1184 | * have put this item on AIL because of the transaction was | |
783a2f65 | 1185 | * aborted forcibly. xfs_trans_ail_delete() takes care of these. |
1da177e4 LT |
1186 | * |
1187 | * Either way, AIL is useless if we're forcing a shutdown. | |
1188 | */ | |
fc1829f3 | 1189 | spin_lock(&ailp->xa_lock); |
04913fdd | 1190 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 | 1191 | xfs_buf_item_free(BUF_ITEM(lip)); |
1da177e4 | 1192 | } |