]>
Commit | Line | Data |
---|---|---|
0b61f8a4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 | 2 | /* |
7b718769 NS |
3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
4 | * All Rights Reserved. | |
1da177e4 | 5 | */ |
1da177e4 | 6 | #include "xfs.h" |
a844f451 | 7 | #include "xfs_fs.h" |
5467b34b | 8 | #include "xfs_shared.h" |
4fb6e8ad | 9 | #include "xfs_format.h" |
239880ef DC |
10 | #include "xfs_log_format.h" |
11 | #include "xfs_trans_resv.h" | |
a844f451 | 12 | #include "xfs_bit.h" |
1da177e4 | 13 | #include "xfs_mount.h" |
239880ef | 14 | #include "xfs_trans.h" |
a844f451 | 15 | #include "xfs_buf_item.h" |
aac855ab DC |
16 | #include "xfs_inode.h" |
17 | #include "xfs_inode_item.h" | |
6f5de180 DC |
18 | #include "xfs_quota.h" |
19 | #include "xfs_dquot_item.h" | |
20 | #include "xfs_dquot.h" | |
1da177e4 | 21 | #include "xfs_trans_priv.h" |
0b1b213f | 22 | #include "xfs_trace.h" |
239880ef | 23 | #include "xfs_log.h" |
1da177e4 LT |
24 | |
25 | ||
26 | kmem_zone_t *xfs_buf_item_zone; | |
27 | ||
7bfa31d8 CH |
28 | static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip) |
29 | { | |
30 | return container_of(lip, struct xfs_buf_log_item, bli_item); | |
31 | } | |
32 | ||
c90821a2 | 33 | STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp); |
1da177e4 | 34 | |
8a6453a8 DW |
35 | /* Is this log iovec plausibly large enough to contain the buffer log format? */ |
36 | bool | |
37 | xfs_buf_log_check_iovec( | |
38 | struct xfs_log_iovec *iovec) | |
39 | { | |
40 | struct xfs_buf_log_format *blfp = iovec->i_addr; | |
41 | char *bmp_end; | |
42 | char *item_end; | |
43 | ||
44 | if (offsetof(struct xfs_buf_log_format, blf_data_map) > iovec->i_len) | |
45 | return false; | |
46 | ||
47 | item_end = (char *)iovec->i_addr + iovec->i_len; | |
48 | bmp_end = (char *)&blfp->blf_data_map[blfp->blf_map_size]; | |
49 | return bmp_end <= item_end; | |
50 | } | |
51 | ||
166d1368 DC |
52 | static inline int |
53 | xfs_buf_log_format_size( | |
54 | struct xfs_buf_log_format *blfp) | |
55 | { | |
56 | return offsetof(struct xfs_buf_log_format, blf_data_map) + | |
57 | (blfp->blf_map_size * sizeof(blfp->blf_data_map[0])); | |
58 | } | |
59 | ||
1da177e4 LT |
60 | /* |
61 | * This returns the number of log iovecs needed to log the | |
62 | * given buf log item. | |
63 | * | |
64 | * It calculates this as 1 iovec for the buf log format structure | |
65 | * and 1 for each stretch of non-contiguous chunks to be logged. | |
66 | * Contiguous chunks are logged in a single iovec. | |
67 | * | |
68 | * If the XFS_BLI_STALE flag has been set, then log nothing. | |
69 | */ | |
166d1368 | 70 | STATIC void |
372cc85e | 71 | xfs_buf_item_size_segment( |
70a20655 CM |
72 | struct xfs_buf_log_item *bip, |
73 | struct xfs_buf_log_format *blfp, | |
74 | int *nvecs, | |
75 | int *nbytes) | |
1da177e4 | 76 | { |
70a20655 CM |
77 | struct xfs_buf *bp = bip->bli_buf; |
78 | int next_bit; | |
79 | int last_bit; | |
1da177e4 | 80 | |
372cc85e DC |
81 | last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
82 | if (last_bit == -1) | |
166d1368 | 83 | return; |
372cc85e DC |
84 | |
85 | /* | |
86 | * initial count for a dirty buffer is 2 vectors - the format structure | |
87 | * and the first dirty region. | |
88 | */ | |
166d1368 DC |
89 | *nvecs += 2; |
90 | *nbytes += xfs_buf_log_format_size(blfp) + XFS_BLF_CHUNK; | |
1da177e4 | 91 | |
1da177e4 LT |
92 | while (last_bit != -1) { |
93 | /* | |
94 | * This takes the bit number to start looking from and | |
95 | * returns the next set bit from there. It returns -1 | |
96 | * if there are no more bits set or the start bit is | |
97 | * beyond the end of the bitmap. | |
98 | */ | |
372cc85e DC |
99 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
100 | last_bit + 1); | |
1da177e4 LT |
101 | /* |
102 | * If we run out of bits, leave the loop, | |
103 | * else if we find a new set of bits bump the number of vecs, | |
104 | * else keep scanning the current set of bits. | |
105 | */ | |
106 | if (next_bit == -1) { | |
372cc85e | 107 | break; |
1da177e4 LT |
108 | } else if (next_bit != last_bit + 1) { |
109 | last_bit = next_bit; | |
166d1368 | 110 | (*nvecs)++; |
c1155410 DC |
111 | } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) != |
112 | (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) + | |
113 | XFS_BLF_CHUNK)) { | |
1da177e4 | 114 | last_bit = next_bit; |
166d1368 | 115 | (*nvecs)++; |
1da177e4 LT |
116 | } else { |
117 | last_bit++; | |
118 | } | |
166d1368 | 119 | *nbytes += XFS_BLF_CHUNK; |
1da177e4 | 120 | } |
1da177e4 LT |
121 | } |
122 | ||
123 | /* | |
372cc85e DC |
124 | * This returns the number of log iovecs needed to log the given buf log item. |
125 | * | |
126 | * It calculates this as 1 iovec for the buf log format structure and 1 for each | |
127 | * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged | |
128 | * in a single iovec. | |
129 | * | |
130 | * Discontiguous buffers need a format structure per region that that is being | |
131 | * logged. This makes the changes in the buffer appear to log recovery as though | |
132 | * they came from separate buffers, just like would occur if multiple buffers | |
133 | * were used instead of a single discontiguous buffer. This enables | |
134 | * discontiguous buffers to be in-memory constructs, completely transparent to | |
135 | * what ends up on disk. | |
136 | * | |
137 | * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log | |
138 | * format structures. | |
1da177e4 | 139 | */ |
166d1368 | 140 | STATIC void |
372cc85e | 141 | xfs_buf_item_size( |
166d1368 DC |
142 | struct xfs_log_item *lip, |
143 | int *nvecs, | |
144 | int *nbytes) | |
1da177e4 | 145 | { |
7bfa31d8 | 146 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
372cc85e DC |
147 | int i; |
148 | ||
149 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
150 | if (bip->bli_flags & XFS_BLI_STALE) { | |
151 | /* | |
152 | * The buffer is stale, so all we need to log | |
153 | * is the buf log format structure with the | |
154 | * cancel flag in it. | |
155 | */ | |
156 | trace_xfs_buf_item_size_stale(bip); | |
b9438173 | 157 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
166d1368 DC |
158 | *nvecs += bip->bli_format_count; |
159 | for (i = 0; i < bip->bli_format_count; i++) { | |
160 | *nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]); | |
161 | } | |
162 | return; | |
372cc85e DC |
163 | } |
164 | ||
165 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); | |
166 | ||
5f6bed76 DC |
167 | if (bip->bli_flags & XFS_BLI_ORDERED) { |
168 | /* | |
169 | * The buffer has been logged just to order it. | |
170 | * It is not being included in the transaction | |
171 | * commit, so no vectors are used at all. | |
172 | */ | |
173 | trace_xfs_buf_item_size_ordered(bip); | |
166d1368 DC |
174 | *nvecs = XFS_LOG_VEC_ORDERED; |
175 | return; | |
5f6bed76 DC |
176 | } |
177 | ||
372cc85e DC |
178 | /* |
179 | * the vector count is based on the number of buffer vectors we have | |
180 | * dirty bits in. This will only be greater than one when we have a | |
181 | * compound buffer with more than one segment dirty. Hence for compound | |
182 | * buffers we need to track which segment the dirty bits correspond to, | |
183 | * and when we move from one segment to the next increment the vector | |
184 | * count for the extra buf log format structure that will need to be | |
185 | * written. | |
186 | */ | |
372cc85e | 187 | for (i = 0; i < bip->bli_format_count; i++) { |
166d1368 DC |
188 | xfs_buf_item_size_segment(bip, &bip->bli_formats[i], |
189 | nvecs, nbytes); | |
372cc85e | 190 | } |
372cc85e | 191 | trace_xfs_buf_item_size(bip); |
372cc85e DC |
192 | } |
193 | ||
1234351c | 194 | static inline void |
7aeb7222 | 195 | xfs_buf_item_copy_iovec( |
bde7cff6 | 196 | struct xfs_log_vec *lv, |
1234351c | 197 | struct xfs_log_iovec **vecp, |
7aeb7222 CH |
198 | struct xfs_buf *bp, |
199 | uint offset, | |
200 | int first_bit, | |
201 | uint nbits) | |
202 | { | |
203 | offset += first_bit * XFS_BLF_CHUNK; | |
bde7cff6 | 204 | xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK, |
1234351c CH |
205 | xfs_buf_offset(bp, offset), |
206 | nbits * XFS_BLF_CHUNK); | |
7aeb7222 CH |
207 | } |
208 | ||
209 | static inline bool | |
210 | xfs_buf_item_straddle( | |
211 | struct xfs_buf *bp, | |
212 | uint offset, | |
213 | int next_bit, | |
214 | int last_bit) | |
215 | { | |
216 | return xfs_buf_offset(bp, offset + (next_bit << XFS_BLF_SHIFT)) != | |
217 | (xfs_buf_offset(bp, offset + (last_bit << XFS_BLF_SHIFT)) + | |
218 | XFS_BLF_CHUNK); | |
219 | } | |
220 | ||
1234351c | 221 | static void |
372cc85e DC |
222 | xfs_buf_item_format_segment( |
223 | struct xfs_buf_log_item *bip, | |
bde7cff6 | 224 | struct xfs_log_vec *lv, |
1234351c | 225 | struct xfs_log_iovec **vecp, |
372cc85e DC |
226 | uint offset, |
227 | struct xfs_buf_log_format *blfp) | |
228 | { | |
70a20655 CM |
229 | struct xfs_buf *bp = bip->bli_buf; |
230 | uint base_size; | |
231 | int first_bit; | |
232 | int last_bit; | |
233 | int next_bit; | |
234 | uint nbits; | |
1da177e4 | 235 | |
372cc85e | 236 | /* copy the flags across from the base format item */ |
b9438173 | 237 | blfp->blf_flags = bip->__bli_format.blf_flags; |
1da177e4 LT |
238 | |
239 | /* | |
77c1a08f DC |
240 | * Base size is the actual size of the ondisk structure - it reflects |
241 | * the actual size of the dirty bitmap rather than the size of the in | |
242 | * memory structure. | |
1da177e4 | 243 | */ |
166d1368 | 244 | base_size = xfs_buf_log_format_size(blfp); |
820a554f | 245 | |
820a554f MT |
246 | first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
247 | if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) { | |
248 | /* | |
249 | * If the map is not be dirty in the transaction, mark | |
250 | * the size as zero and do not advance the vector pointer. | |
251 | */ | |
bde7cff6 | 252 | return; |
820a554f MT |
253 | } |
254 | ||
bde7cff6 CH |
255 | blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size); |
256 | blfp->blf_size = 1; | |
1da177e4 LT |
257 | |
258 | if (bip->bli_flags & XFS_BLI_STALE) { | |
259 | /* | |
260 | * The buffer is stale, so all we need to log | |
261 | * is the buf log format structure with the | |
262 | * cancel flag in it. | |
263 | */ | |
0b1b213f | 264 | trace_xfs_buf_item_format_stale(bip); |
372cc85e | 265 | ASSERT(blfp->blf_flags & XFS_BLF_CANCEL); |
bde7cff6 | 266 | return; |
1da177e4 LT |
267 | } |
268 | ||
5f6bed76 | 269 | |
1da177e4 LT |
270 | /* |
271 | * Fill in an iovec for each set of contiguous chunks. | |
272 | */ | |
1da177e4 LT |
273 | last_bit = first_bit; |
274 | nbits = 1; | |
275 | for (;;) { | |
276 | /* | |
277 | * This takes the bit number to start looking from and | |
278 | * returns the next set bit from there. It returns -1 | |
279 | * if there are no more bits set or the start bit is | |
280 | * beyond the end of the bitmap. | |
281 | */ | |
372cc85e DC |
282 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
283 | (uint)last_bit + 1); | |
1da177e4 | 284 | /* |
7aeb7222 CH |
285 | * If we run out of bits fill in the last iovec and get out of |
286 | * the loop. Else if we start a new set of bits then fill in | |
287 | * the iovec for the series we were looking at and start | |
288 | * counting the bits in the new one. Else we're still in the | |
289 | * same set of bits so just keep counting and scanning. | |
1da177e4 LT |
290 | */ |
291 | if (next_bit == -1) { | |
bde7cff6 | 292 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
7aeb7222 | 293 | first_bit, nbits); |
bde7cff6 | 294 | blfp->blf_size++; |
1da177e4 | 295 | break; |
7aeb7222 CH |
296 | } else if (next_bit != last_bit + 1 || |
297 | xfs_buf_item_straddle(bp, offset, next_bit, last_bit)) { | |
bde7cff6 | 298 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
1234351c | 299 | first_bit, nbits); |
bde7cff6 | 300 | blfp->blf_size++; |
1da177e4 LT |
301 | first_bit = next_bit; |
302 | last_bit = next_bit; | |
303 | nbits = 1; | |
304 | } else { | |
305 | last_bit++; | |
306 | nbits++; | |
307 | } | |
308 | } | |
372cc85e DC |
309 | } |
310 | ||
311 | /* | |
312 | * This is called to fill in the vector of log iovecs for the | |
313 | * given log buf item. It fills the first entry with a buf log | |
314 | * format structure, and the rest point to contiguous chunks | |
315 | * within the buffer. | |
316 | */ | |
317 | STATIC void | |
318 | xfs_buf_item_format( | |
319 | struct xfs_log_item *lip, | |
bde7cff6 | 320 | struct xfs_log_vec *lv) |
372cc85e DC |
321 | { |
322 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); | |
323 | struct xfs_buf *bp = bip->bli_buf; | |
bde7cff6 | 324 | struct xfs_log_iovec *vecp = NULL; |
372cc85e DC |
325 | uint offset = 0; |
326 | int i; | |
327 | ||
328 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
329 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
330 | (bip->bli_flags & XFS_BLI_STALE)); | |
0d612fb5 DC |
331 | ASSERT((bip->bli_flags & XFS_BLI_STALE) || |
332 | (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF | |
333 | && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF)); | |
e9385cc6 BF |
334 | ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED) || |
335 | (bip->bli_flags & XFS_BLI_STALE)); | |
0d612fb5 | 336 | |
372cc85e DC |
337 | |
338 | /* | |
339 | * If it is an inode buffer, transfer the in-memory state to the | |
ddf6ad01 DC |
340 | * format flags and clear the in-memory state. |
341 | * | |
342 | * For buffer based inode allocation, we do not transfer | |
372cc85e DC |
343 | * this state if the inode buffer allocation has not yet been committed |
344 | * to the log as setting the XFS_BLI_INODE_BUF flag will prevent | |
345 | * correct replay of the inode allocation. | |
ddf6ad01 DC |
346 | * |
347 | * For icreate item based inode allocation, the buffers aren't written | |
348 | * to the journal during allocation, and hence we should always tag the | |
349 | * buffer as an inode buffer so that the correct unlinked list replay | |
350 | * occurs during recovery. | |
372cc85e DC |
351 | */ |
352 | if (bip->bli_flags & XFS_BLI_INODE_BUF) { | |
b81b79f4 | 353 | if (xfs_sb_version_has_v3inode(&lip->li_mountp->m_sb) || |
ddf6ad01 | 354 | !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && |
372cc85e | 355 | xfs_log_item_in_current_chkpt(lip))) |
b9438173 | 356 | bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF; |
372cc85e DC |
357 | bip->bli_flags &= ~XFS_BLI_INODE_BUF; |
358 | } | |
359 | ||
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); |
70a20655 | 417 | xfs_buf_t *bp = bip->bli_buf; |
70a20655 CM |
418 | int stale = bip->bli_flags & XFS_BLI_STALE; |
419 | int freed; | |
1da177e4 | 420 | |
fb1755a6 | 421 | ASSERT(bp->b_log_item == 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 | */ | |
e6631f85 | 448 | if (!list_empty(&lip->li_trans)) |
e34a314c | 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 | 458 | /* |
849274c1 BF |
459 | * If we get called here because of an IO error, we may or may |
460 | * not have the item on the AIL. xfs_trans_ail_delete() will | |
461 | * take care of that situation. xfs_trans_ail_delete() drops | |
462 | * the AIL lock. | |
1da177e4 LT |
463 | */ |
464 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { | |
aac855ab DC |
465 | lip->li_cb(bp, lip); |
466 | xfs_iflush_done(bp); | |
fb1755a6 | 467 | bp->b_log_item = NULL; |
1da177e4 | 468 | } else { |
849274c1 | 469 | xfs_trans_ail_delete(lip, SHUTDOWN_LOG_IO_ERROR); |
1da177e4 | 470 | xfs_buf_item_relse(bp); |
fb1755a6 | 471 | ASSERT(bp->b_log_item == NULL); |
1da177e4 LT |
472 | } |
473 | xfs_buf_relse(bp); | |
960c60af | 474 | } else if (freed && remove) { |
137fff09 | 475 | /* |
54b3b1f6 BF |
476 | * The buffer must be locked and held by the caller to simulate |
477 | * an async I/O failure. | |
137fff09 | 478 | */ |
960c60af | 479 | xfs_buf_lock(bp); |
137fff09 DC |
480 | xfs_buf_hold(bp); |
481 | bp->b_flags |= XBF_ASYNC; | |
54b3b1f6 | 482 | xfs_buf_ioend_fail(bp); |
1da177e4 LT |
483 | } |
484 | } | |
485 | ||
ba0f32d4 | 486 | STATIC uint |
43ff2122 CH |
487 | xfs_buf_item_push( |
488 | struct xfs_log_item *lip, | |
489 | struct list_head *buffer_list) | |
1da177e4 | 490 | { |
7bfa31d8 CH |
491 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
492 | struct xfs_buf *bp = bip->bli_buf; | |
43ff2122 | 493 | uint rval = XFS_ITEM_SUCCESS; |
1da177e4 | 494 | |
811e64c7 | 495 | if (xfs_buf_ispinned(bp)) |
1da177e4 | 496 | return XFS_ITEM_PINNED; |
5337fe9b BF |
497 | if (!xfs_buf_trylock(bp)) { |
498 | /* | |
499 | * If we have just raced with a buffer being pinned and it has | |
500 | * been marked stale, we could end up stalling until someone else | |
501 | * issues a log force to unpin the stale buffer. Check for the | |
502 | * race condition here so xfsaild recognizes the buffer is pinned | |
503 | * and queues a log force to move it along. | |
504 | */ | |
505 | if (xfs_buf_ispinned(bp)) | |
506 | return XFS_ITEM_PINNED; | |
1da177e4 | 507 | return XFS_ITEM_LOCKED; |
5337fe9b | 508 | } |
1da177e4 | 509 | |
1da177e4 | 510 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
43ff2122 CH |
511 | |
512 | trace_xfs_buf_item_push(bip); | |
513 | ||
ac8809f9 | 514 | /* has a previous flush failed due to IO errors? */ |
f9bccfcc BF |
515 | if (bp->b_flags & XBF_WRITE_FAIL) { |
516 | xfs_buf_alert_ratelimited(bp, "XFS: Failing async write", | |
517 | "Failing async write on buffer block 0x%llx. Retrying async write.", | |
518 | (long long)bp->b_bn); | |
ac8809f9 DC |
519 | } |
520 | ||
43ff2122 CH |
521 | if (!xfs_buf_delwri_queue(bp, buffer_list)) |
522 | rval = XFS_ITEM_FLUSHING; | |
523 | xfs_buf_unlock(bp); | |
524 | return rval; | |
1da177e4 LT |
525 | } |
526 | ||
95808459 BF |
527 | /* |
528 | * Drop the buffer log item refcount and take appropriate action. This helper | |
529 | * determines whether the bli must be freed or not, since a decrement to zero | |
530 | * does not necessarily mean the bli is unused. | |
531 | * | |
532 | * Return true if the bli is freed, false otherwise. | |
533 | */ | |
534 | bool | |
535 | xfs_buf_item_put( | |
536 | struct xfs_buf_log_item *bip) | |
537 | { | |
538 | struct xfs_log_item *lip = &bip->bli_item; | |
539 | bool aborted; | |
540 | bool dirty; | |
541 | ||
542 | /* drop the bli ref and return if it wasn't the last one */ | |
543 | if (!atomic_dec_and_test(&bip->bli_refcount)) | |
544 | return false; | |
545 | ||
546 | /* | |
547 | * We dropped the last ref and must free the item if clean or aborted. | |
548 | * If the bli is dirty and non-aborted, the buffer was clean in the | |
549 | * transaction but still awaiting writeback from previous changes. In | |
550 | * that case, the bli is freed on buffer writeback completion. | |
551 | */ | |
552 | aborted = test_bit(XFS_LI_ABORTED, &lip->li_flags) || | |
553 | XFS_FORCED_SHUTDOWN(lip->li_mountp); | |
554 | dirty = bip->bli_flags & XFS_BLI_DIRTY; | |
555 | if (dirty && !aborted) | |
556 | return false; | |
557 | ||
558 | /* | |
559 | * The bli is aborted or clean. An aborted item may be in the AIL | |
560 | * regardless of dirty state. For example, consider an aborted | |
561 | * transaction that invalidated a dirty bli and cleared the dirty | |
562 | * state. | |
563 | */ | |
564 | if (aborted) | |
2b3cf093 | 565 | xfs_trans_ail_delete(lip, 0); |
95808459 BF |
566 | xfs_buf_item_relse(bip->bli_buf); |
567 | return true; | |
568 | } | |
569 | ||
1da177e4 | 570 | /* |
64fc35de DC |
571 | * Release the buffer associated with the buf log item. If there is no dirty |
572 | * logged data associated with the buffer recorded in the buf log item, then | |
573 | * free the buf log item and remove the reference to it in the buffer. | |
1da177e4 | 574 | * |
64fc35de DC |
575 | * This call ignores the recursion count. It is only called when the buffer |
576 | * should REALLY be unlocked, regardless of the recursion count. | |
1da177e4 | 577 | * |
64fc35de DC |
578 | * We unconditionally drop the transaction's reference to the log item. If the |
579 | * item was logged, then another reference was taken when it was pinned, so we | |
580 | * can safely drop the transaction reference now. This also allows us to avoid | |
581 | * potential races with the unpin code freeing the bli by not referencing the | |
582 | * bli after we've dropped the reference count. | |
583 | * | |
584 | * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item | |
585 | * if necessary but do not unlock the buffer. This is for support of | |
586 | * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't | |
587 | * free the item. | |
1da177e4 | 588 | */ |
ba0f32d4 | 589 | STATIC void |
ddf92053 | 590 | xfs_buf_item_release( |
7bfa31d8 | 591 | struct xfs_log_item *lip) |
1da177e4 | 592 | { |
7bfa31d8 CH |
593 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
594 | struct xfs_buf *bp = bip->bli_buf; | |
95808459 | 595 | bool released; |
d9183105 | 596 | bool hold = bip->bli_flags & XFS_BLI_HOLD; |
d9183105 | 597 | bool stale = bip->bli_flags & XFS_BLI_STALE; |
7bf7a193 | 598 | #if defined(DEBUG) || defined(XFS_WARN) |
d9183105 | 599 | bool ordered = bip->bli_flags & XFS_BLI_ORDERED; |
95808459 | 600 | bool dirty = bip->bli_flags & XFS_BLI_DIRTY; |
4d09807f BF |
601 | bool aborted = test_bit(XFS_LI_ABORTED, |
602 | &lip->li_flags); | |
7bf7a193 | 603 | #endif |
1da177e4 | 604 | |
ddf92053 | 605 | trace_xfs_buf_item_release(bip); |
1da177e4 LT |
606 | |
607 | /* | |
6453c65d BF |
608 | * The bli dirty state should match whether the blf has logged segments |
609 | * except for ordered buffers, where only the bli should be dirty. | |
1da177e4 | 610 | */ |
6453c65d BF |
611 | ASSERT((!ordered && dirty == xfs_buf_item_dirty_format(bip)) || |
612 | (ordered && dirty && !xfs_buf_item_dirty_format(bip))); | |
d9183105 BF |
613 | ASSERT(!stale || (bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
614 | ||
46f9d2eb | 615 | /* |
d9183105 BF |
616 | * Clear the buffer's association with this transaction and |
617 | * per-transaction state from the bli, which has been copied above. | |
618 | */ | |
619 | bp->b_transp = NULL; | |
620 | bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED); | |
621 | ||
622 | /* | |
95808459 BF |
623 | * Unref the item and unlock the buffer unless held or stale. Stale |
624 | * buffers remain locked until final unpin unless the bli is freed by | |
625 | * the unref call. The latter implies shutdown because buffer | |
626 | * invalidation dirties the bli and transaction. | |
46f9d2eb | 627 | */ |
95808459 BF |
628 | released = xfs_buf_item_put(bip); |
629 | if (hold || (stale && !released)) | |
d9183105 | 630 | return; |
4d09807f | 631 | ASSERT(!stale || aborted); |
95808459 | 632 | xfs_buf_relse(bp); |
1da177e4 LT |
633 | } |
634 | ||
ddf92053 CH |
635 | STATIC void |
636 | xfs_buf_item_committing( | |
637 | struct xfs_log_item *lip, | |
638 | xfs_lsn_t commit_lsn) | |
639 | { | |
640 | return xfs_buf_item_release(lip); | |
641 | } | |
642 | ||
1da177e4 LT |
643 | /* |
644 | * This is called to find out where the oldest active copy of the | |
645 | * buf log item in the on disk log resides now that the last log | |
646 | * write of it completed at the given lsn. | |
647 | * We always re-log all the dirty data in a buffer, so usually the | |
648 | * latest copy in the on disk log is the only one that matters. For | |
649 | * those cases we simply return the given lsn. | |
650 | * | |
651 | * The one exception to this is for buffers full of newly allocated | |
652 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF | |
653 | * flag set, indicating that only the di_next_unlinked fields from the | |
654 | * inodes in the buffers will be replayed during recovery. If the | |
655 | * original newly allocated inode images have not yet been flushed | |
656 | * when the buffer is so relogged, then we need to make sure that we | |
657 | * keep the old images in the 'active' portion of the log. We do this | |
658 | * by returning the original lsn of that transaction here rather than | |
659 | * the current one. | |
660 | */ | |
ba0f32d4 | 661 | STATIC xfs_lsn_t |
1da177e4 | 662 | xfs_buf_item_committed( |
7bfa31d8 | 663 | struct xfs_log_item *lip, |
1da177e4 LT |
664 | xfs_lsn_t lsn) |
665 | { | |
7bfa31d8 CH |
666 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
667 | ||
0b1b213f CH |
668 | trace_xfs_buf_item_committed(bip); |
669 | ||
7bfa31d8 CH |
670 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0) |
671 | return lip->li_lsn; | |
672 | return lsn; | |
1da177e4 LT |
673 | } |
674 | ||
272e42b2 | 675 | static const struct xfs_item_ops xfs_buf_item_ops = { |
7bfa31d8 CH |
676 | .iop_size = xfs_buf_item_size, |
677 | .iop_format = xfs_buf_item_format, | |
678 | .iop_pin = xfs_buf_item_pin, | |
679 | .iop_unpin = xfs_buf_item_unpin, | |
ddf92053 CH |
680 | .iop_release = xfs_buf_item_release, |
681 | .iop_committing = xfs_buf_item_committing, | |
7bfa31d8 CH |
682 | .iop_committed = xfs_buf_item_committed, |
683 | .iop_push = xfs_buf_item_push, | |
1da177e4 LT |
684 | }; |
685 | ||
c64dd49b | 686 | STATIC void |
372cc85e DC |
687 | xfs_buf_item_get_format( |
688 | struct xfs_buf_log_item *bip, | |
689 | int count) | |
690 | { | |
691 | ASSERT(bip->bli_formats == NULL); | |
692 | bip->bli_format_count = count; | |
693 | ||
694 | if (count == 1) { | |
b9438173 | 695 | bip->bli_formats = &bip->__bli_format; |
c64dd49b | 696 | return; |
372cc85e DC |
697 | } |
698 | ||
699 | bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format), | |
707e0dda | 700 | 0); |
372cc85e DC |
701 | } |
702 | ||
703 | STATIC void | |
704 | xfs_buf_item_free_format( | |
705 | struct xfs_buf_log_item *bip) | |
706 | { | |
b9438173 | 707 | if (bip->bli_formats != &bip->__bli_format) { |
372cc85e DC |
708 | kmem_free(bip->bli_formats); |
709 | bip->bli_formats = NULL; | |
710 | } | |
711 | } | |
1da177e4 LT |
712 | |
713 | /* | |
714 | * Allocate a new buf log item to go with the given buffer. | |
fb1755a6 CM |
715 | * Set the buffer's b_log_item field to point to the new |
716 | * buf log item. | |
1da177e4 | 717 | */ |
f79af0b9 | 718 | int |
1da177e4 | 719 | xfs_buf_item_init( |
f79af0b9 DC |
720 | struct xfs_buf *bp, |
721 | struct xfs_mount *mp) | |
1da177e4 | 722 | { |
fb1755a6 | 723 | struct xfs_buf_log_item *bip = bp->b_log_item; |
1da177e4 LT |
724 | int chunks; |
725 | int map_size; | |
372cc85e | 726 | int i; |
1da177e4 LT |
727 | |
728 | /* | |
729 | * Check to see if there is already a buf log item for | |
fb1755a6 | 730 | * this buffer. If we do already have one, there is |
1da177e4 LT |
731 | * nothing to do here so return. |
732 | */ | |
dbd329f1 | 733 | ASSERT(bp->b_mount == mp); |
1a2ebf83 | 734 | if (bip) { |
fb1755a6 | 735 | ASSERT(bip->bli_item.li_type == XFS_LI_BUF); |
1a2ebf83 DC |
736 | ASSERT(!bp->b_transp); |
737 | ASSERT(bip->bli_buf == bp); | |
f79af0b9 | 738 | return 0; |
fb1755a6 | 739 | } |
1da177e4 | 740 | |
707e0dda | 741 | bip = kmem_zone_zalloc(xfs_buf_item_zone, 0); |
43f5efc5 | 742 | xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops); |
1da177e4 | 743 | bip->bli_buf = bp; |
372cc85e DC |
744 | |
745 | /* | |
746 | * chunks is the number of XFS_BLF_CHUNK size pieces the buffer | |
747 | * can be divided into. Make sure not to truncate any pieces. | |
748 | * map_size is the size of the bitmap needed to describe the | |
749 | * chunks of the buffer. | |
750 | * | |
751 | * Discontiguous buffer support follows the layout of the underlying | |
752 | * buffer. This makes the implementation as simple as possible. | |
753 | */ | |
c64dd49b | 754 | xfs_buf_item_get_format(bip, bp->b_map_count); |
372cc85e DC |
755 | |
756 | for (i = 0; i < bip->bli_format_count; i++) { | |
757 | chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len), | |
758 | XFS_BLF_CHUNK); | |
759 | map_size = DIV_ROUND_UP(chunks, NBWORD); | |
760 | ||
c3d5f0c2 DW |
761 | if (map_size > XFS_BLF_DATAMAP_SIZE) { |
762 | kmem_cache_free(xfs_buf_item_zone, bip); | |
763 | xfs_err(mp, | |
764 | "buffer item dirty bitmap (%u uints) too small to reflect %u bytes!", | |
765 | map_size, | |
766 | BBTOB(bp->b_maps[i].bm_len)); | |
767 | return -EFSCORRUPTED; | |
768 | } | |
769 | ||
372cc85e DC |
770 | bip->bli_formats[i].blf_type = XFS_LI_BUF; |
771 | bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn; | |
772 | bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len; | |
773 | bip->bli_formats[i].blf_map_size = map_size; | |
774 | } | |
1da177e4 | 775 | |
fb1755a6 | 776 | bp->b_log_item = bip; |
f79af0b9 DC |
777 | xfs_buf_hold(bp); |
778 | return 0; | |
1da177e4 LT |
779 | } |
780 | ||
781 | ||
782 | /* | |
783 | * Mark bytes first through last inclusive as dirty in the buf | |
784 | * item's bitmap. | |
785 | */ | |
632b89e8 | 786 | static void |
372cc85e | 787 | xfs_buf_item_log_segment( |
1da177e4 | 788 | uint first, |
372cc85e DC |
789 | uint last, |
790 | uint *map) | |
1da177e4 LT |
791 | { |
792 | uint first_bit; | |
793 | uint last_bit; | |
794 | uint bits_to_set; | |
795 | uint bits_set; | |
796 | uint word_num; | |
797 | uint *wordp; | |
798 | uint bit; | |
799 | uint end_bit; | |
800 | uint mask; | |
801 | ||
c3d5f0c2 DW |
802 | ASSERT(first < XFS_BLF_DATAMAP_SIZE * XFS_BLF_CHUNK * NBWORD); |
803 | ASSERT(last < XFS_BLF_DATAMAP_SIZE * XFS_BLF_CHUNK * NBWORD); | |
804 | ||
1da177e4 LT |
805 | /* |
806 | * Convert byte offsets to bit numbers. | |
807 | */ | |
c1155410 DC |
808 | first_bit = first >> XFS_BLF_SHIFT; |
809 | last_bit = last >> XFS_BLF_SHIFT; | |
1da177e4 LT |
810 | |
811 | /* | |
812 | * Calculate the total number of bits to be set. | |
813 | */ | |
814 | bits_to_set = last_bit - first_bit + 1; | |
815 | ||
816 | /* | |
817 | * Get a pointer to the first word in the bitmap | |
818 | * to set a bit in. | |
819 | */ | |
820 | word_num = first_bit >> BIT_TO_WORD_SHIFT; | |
372cc85e | 821 | wordp = &map[word_num]; |
1da177e4 LT |
822 | |
823 | /* | |
824 | * Calculate the starting bit in the first word. | |
825 | */ | |
826 | bit = first_bit & (uint)(NBWORD - 1); | |
827 | ||
828 | /* | |
829 | * First set any bits in the first word of our range. | |
830 | * If it starts at bit 0 of the word, it will be | |
831 | * set below rather than here. That is what the variable | |
832 | * bit tells us. The variable bits_set tracks the number | |
833 | * of bits that have been set so far. End_bit is the number | |
834 | * of the last bit to be set in this word plus one. | |
835 | */ | |
836 | if (bit) { | |
9bb54cb5 | 837 | end_bit = min(bit + bits_to_set, (uint)NBWORD); |
79c350e4 | 838 | mask = ((1U << (end_bit - bit)) - 1) << bit; |
1da177e4 LT |
839 | *wordp |= mask; |
840 | wordp++; | |
841 | bits_set = end_bit - bit; | |
842 | } else { | |
843 | bits_set = 0; | |
844 | } | |
845 | ||
846 | /* | |
847 | * Now set bits a whole word at a time that are between | |
848 | * first_bit and last_bit. | |
849 | */ | |
850 | while ((bits_to_set - bits_set) >= NBWORD) { | |
12025460 | 851 | *wordp = 0xffffffff; |
1da177e4 LT |
852 | bits_set += NBWORD; |
853 | wordp++; | |
854 | } | |
855 | ||
856 | /* | |
857 | * Finally, set any bits left to be set in one last partial word. | |
858 | */ | |
859 | end_bit = bits_to_set - bits_set; | |
860 | if (end_bit) { | |
79c350e4 | 861 | mask = (1U << end_bit) - 1; |
1da177e4 LT |
862 | *wordp |= mask; |
863 | } | |
1da177e4 LT |
864 | } |
865 | ||
372cc85e DC |
866 | /* |
867 | * Mark bytes first through last inclusive as dirty in the buf | |
868 | * item's bitmap. | |
869 | */ | |
870 | void | |
871 | xfs_buf_item_log( | |
70a20655 | 872 | struct xfs_buf_log_item *bip, |
372cc85e DC |
873 | uint first, |
874 | uint last) | |
875 | { | |
876 | int i; | |
877 | uint start; | |
878 | uint end; | |
879 | struct xfs_buf *bp = bip->bli_buf; | |
880 | ||
372cc85e DC |
881 | /* |
882 | * walk each buffer segment and mark them dirty appropriately. | |
883 | */ | |
884 | start = 0; | |
885 | for (i = 0; i < bip->bli_format_count; i++) { | |
886 | if (start > last) | |
887 | break; | |
a3916e52 BF |
888 | end = start + BBTOB(bp->b_maps[i].bm_len) - 1; |
889 | ||
890 | /* skip to the map that includes the first byte to log */ | |
372cc85e DC |
891 | if (first > end) { |
892 | start += BBTOB(bp->b_maps[i].bm_len); | |
893 | continue; | |
894 | } | |
a3916e52 BF |
895 | |
896 | /* | |
897 | * Trim the range to this segment and mark it in the bitmap. | |
898 | * Note that we must convert buffer offsets to segment relative | |
899 | * offsets (e.g., the first byte of each segment is byte 0 of | |
900 | * that segment). | |
901 | */ | |
372cc85e DC |
902 | if (first < start) |
903 | first = start; | |
904 | if (end > last) | |
905 | end = last; | |
a3916e52 | 906 | xfs_buf_item_log_segment(first - start, end - start, |
372cc85e DC |
907 | &bip->bli_formats[i].blf_data_map[0]); |
908 | ||
a3916e52 | 909 | start += BBTOB(bp->b_maps[i].bm_len); |
372cc85e DC |
910 | } |
911 | } | |
912 | ||
1da177e4 | 913 | |
6453c65d BF |
914 | /* |
915 | * Return true if the buffer has any ranges logged/dirtied by a transaction, | |
916 | * false otherwise. | |
917 | */ | |
918 | bool | |
919 | xfs_buf_item_dirty_format( | |
920 | struct xfs_buf_log_item *bip) | |
921 | { | |
922 | int i; | |
923 | ||
924 | for (i = 0; i < bip->bli_format_count; i++) { | |
925 | if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map, | |
926 | bip->bli_formats[i].blf_map_size)) | |
927 | return true; | |
928 | } | |
929 | ||
930 | return false; | |
931 | } | |
932 | ||
e1f5dbd7 LM |
933 | STATIC void |
934 | xfs_buf_item_free( | |
70a20655 | 935 | struct xfs_buf_log_item *bip) |
e1f5dbd7 | 936 | { |
372cc85e | 937 | xfs_buf_item_free_format(bip); |
b1c5ebb2 | 938 | kmem_free(bip->bli_item.li_lv_shadow); |
377bcd5f | 939 | kmem_cache_free(xfs_buf_item_zone, bip); |
e1f5dbd7 LM |
940 | } |
941 | ||
1da177e4 | 942 | /* |
b01d1461 | 943 | * xfs_buf_item_relse() is called when the buf log item is no longer needed. |
1da177e4 LT |
944 | */ |
945 | void | |
946 | xfs_buf_item_relse( | |
947 | xfs_buf_t *bp) | |
948 | { | |
fb1755a6 | 949 | struct xfs_buf_log_item *bip = bp->b_log_item; |
1da177e4 | 950 | |
0b1b213f | 951 | trace_xfs_buf_item_relse(bp, _RET_IP_); |
826f7e34 | 952 | ASSERT(!test_bit(XFS_LI_IN_AIL, &bip->bli_item.li_flags)); |
0b1b213f | 953 | |
fb1755a6 | 954 | bp->b_log_item = NULL; |
e1f5dbd7 LM |
955 | xfs_buf_rele(bp); |
956 | xfs_buf_item_free(bip); | |
1da177e4 LT |
957 | } |
958 | ||
959 | ||
960 | /* | |
961 | * Add the given log item with its callback to the list of callbacks | |
b01d1461 | 962 | * to be called when the buffer's I/O completes. |
1da177e4 LT |
963 | */ |
964 | void | |
965 | xfs_buf_attach_iodone( | |
efe2330f CH |
966 | struct xfs_buf *bp, |
967 | void (*cb)(struct xfs_buf *, struct xfs_log_item *), | |
968 | struct xfs_log_item *lip) | |
1da177e4 | 969 | { |
0c842ad4 | 970 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
971 | |
972 | lip->li_cb = cb; | |
643c8c05 | 973 | list_add_tail(&lip->li_bio_list, &bp->b_li_list); |
1da177e4 LT |
974 | } |
975 | ||
c90821a2 DC |
976 | /* |
977 | * We can have many callbacks on a buffer. Running the callbacks individually | |
978 | * can cause a lot of contention on the AIL lock, so we allow for a single | |
643c8c05 CM |
979 | * callback to be able to scan the remaining items in bp->b_li_list for other |
980 | * items of the same type and callback to be processed in the first call. | |
c90821a2 DC |
981 | * |
982 | * As a result, the loop walking the callback list below will also modify the | |
983 | * list. it removes the first item from the list and then runs the callback. | |
643c8c05 | 984 | * The loop then restarts from the new first item int the list. This allows the |
c90821a2 DC |
985 | * callback to scan and modify the list attached to the buffer and we don't |
986 | * have to care about maintaining a next item pointer. | |
987 | */ | |
1da177e4 LT |
988 | STATIC void |
989 | xfs_buf_do_callbacks( | |
c90821a2 | 990 | struct xfs_buf *bp) |
1da177e4 | 991 | { |
fb1755a6 | 992 | struct xfs_buf_log_item *blip = bp->b_log_item; |
c90821a2 | 993 | struct xfs_log_item *lip; |
1da177e4 | 994 | |
fb1755a6 CM |
995 | /* If there is a buf_log_item attached, run its callback */ |
996 | if (blip) { | |
997 | lip = &blip->bli_item; | |
998 | lip->li_cb(bp, lip); | |
999 | } | |
1000 | ||
643c8c05 CM |
1001 | while (!list_empty(&bp->b_li_list)) { |
1002 | lip = list_first_entry(&bp->b_li_list, struct xfs_log_item, | |
1003 | li_bio_list); | |
1004 | ||
1da177e4 | 1005 | /* |
643c8c05 | 1006 | * Remove the item from the list, so we don't have any |
1da177e4 LT |
1007 | * confusion if the item is added to another buf. |
1008 | * Don't touch the log item after calling its | |
1009 | * callback, because it could have freed itself. | |
1010 | */ | |
643c8c05 | 1011 | list_del_init(&lip->li_bio_list); |
1da177e4 | 1012 | lip->li_cb(bp, lip); |
1da177e4 LT |
1013 | } |
1014 | } | |
1015 | ||
0b80ae6e CM |
1016 | /* |
1017 | * Invoke the error state callback for each log item affected by the failed I/O. | |
1018 | * | |
1019 | * If a metadata buffer write fails with a non-permanent error, the buffer is | |
1020 | * eventually resubmitted and so the completion callbacks are not run. The error | |
1021 | * state may need to be propagated to the log items attached to the buffer, | |
1022 | * however, so the next AIL push of the item knows hot to handle it correctly. | |
1023 | */ | |
1024 | STATIC void | |
1025 | xfs_buf_do_callbacks_fail( | |
1026 | struct xfs_buf *bp) | |
1027 | { | |
643c8c05 | 1028 | struct xfs_log_item *lip; |
fb1755a6 | 1029 | struct xfs_ail *ailp; |
0b80ae6e | 1030 | |
fb1755a6 CM |
1031 | /* |
1032 | * Buffer log item errors are handled directly by xfs_buf_item_push() | |
1033 | * and xfs_buf_iodone_callback_error, and they have no IO error | |
1034 | * callbacks. Check only for items in b_li_list. | |
1035 | */ | |
643c8c05 | 1036 | if (list_empty(&bp->b_li_list)) |
fb1755a6 CM |
1037 | return; |
1038 | ||
643c8c05 CM |
1039 | lip = list_first_entry(&bp->b_li_list, struct xfs_log_item, |
1040 | li_bio_list); | |
fb1755a6 | 1041 | ailp = lip->li_ailp; |
57e80956 | 1042 | spin_lock(&ailp->ail_lock); |
643c8c05 | 1043 | list_for_each_entry(lip, &bp->b_li_list, li_bio_list) { |
0b80ae6e CM |
1044 | if (lip->li_ops->iop_error) |
1045 | lip->li_ops->iop_error(lip, bp); | |
1046 | } | |
57e80956 | 1047 | spin_unlock(&ailp->ail_lock); |
0b80ae6e CM |
1048 | } |
1049 | ||
df309390 CM |
1050 | static bool |
1051 | xfs_buf_iodone_callback_error( | |
bfc60177 | 1052 | struct xfs_buf *bp) |
1da177e4 | 1053 | { |
fb1755a6 | 1054 | struct xfs_buf_log_item *bip = bp->b_log_item; |
643c8c05 | 1055 | struct xfs_log_item *lip; |
fb1755a6 | 1056 | struct xfs_mount *mp; |
bfc60177 CH |
1057 | static ulong lasttime; |
1058 | static xfs_buftarg_t *lasttarg; | |
df309390 | 1059 | struct xfs_error_cfg *cfg; |
1da177e4 | 1060 | |
fb1755a6 CM |
1061 | /* |
1062 | * The failed buffer might not have a buf_log_item attached or the | |
1063 | * log_item list might be empty. Get the mp from the available | |
1064 | * xfs_log_item | |
1065 | */ | |
643c8c05 CM |
1066 | lip = list_first_entry_or_null(&bp->b_li_list, struct xfs_log_item, |
1067 | li_bio_list); | |
1068 | mp = lip ? lip->li_mountp : bip->bli_item.li_mountp; | |
fb1755a6 | 1069 | |
bfc60177 CH |
1070 | /* |
1071 | * If we've already decided to shutdown the filesystem because of | |
1072 | * I/O errors, there's no point in giving this a retry. | |
1073 | */ | |
df309390 CM |
1074 | if (XFS_FORCED_SHUTDOWN(mp)) |
1075 | goto out_stale; | |
1da177e4 | 1076 | |
49074c06 | 1077 | if (bp->b_target != lasttarg || |
bfc60177 CH |
1078 | time_after(jiffies, (lasttime + 5*HZ))) { |
1079 | lasttime = jiffies; | |
cdbcf82b | 1080 | xfs_buf_ioerror_alert(bp, __this_address); |
bfc60177 | 1081 | } |
49074c06 | 1082 | lasttarg = bp->b_target; |
1da177e4 | 1083 | |
df309390 CM |
1084 | /* synchronous writes will have callers process the error */ |
1085 | if (!(bp->b_flags & XBF_ASYNC)) | |
1086 | goto out_stale; | |
1087 | ||
1088 | trace_xfs_buf_item_iodone_async(bp, _RET_IP_); | |
df309390 | 1089 | |
5539d367 ES |
1090 | cfg = xfs_error_get_cfg(mp, XFS_ERR_METADATA, bp->b_error); |
1091 | ||
bfc60177 | 1092 | /* |
25985edc | 1093 | * If the write was asynchronous then no one will be looking for the |
df309390 CM |
1094 | * error. If this is the first failure of this type, clear the error |
1095 | * state and write the buffer out again. This means we always retry an | |
1096 | * async write failure at least once, but we also need to set the buffer | |
1097 | * up to behave correctly now for repeated failures. | |
bfc60177 | 1098 | */ |
0b4db5df | 1099 | if (!(bp->b_flags & (XBF_STALE | XBF_WRITE_FAIL)) || |
df309390 | 1100 | bp->b_last_error != bp->b_error) { |
0b4db5df | 1101 | bp->b_flags |= (XBF_WRITE | XBF_DONE | XBF_WRITE_FAIL); |
df309390 | 1102 | bp->b_last_error = bp->b_error; |
77169812 ES |
1103 | if (cfg->retry_timeout != XFS_ERR_RETRY_FOREVER && |
1104 | !bp->b_first_retry_time) | |
5539d367 | 1105 | bp->b_first_retry_time = jiffies; |
a5ea70d2 | 1106 | |
df309390 CM |
1107 | xfs_buf_ioerror(bp, 0); |
1108 | xfs_buf_submit(bp); | |
1109 | return true; | |
1110 | } | |
43ff2122 | 1111 | |
df309390 CM |
1112 | /* |
1113 | * Repeated failure on an async write. Take action according to the | |
1114 | * error configuration we have been set up to use. | |
1115 | */ | |
a5ea70d2 CM |
1116 | |
1117 | if (cfg->max_retries != XFS_ERR_RETRY_FOREVER && | |
1118 | ++bp->b_retries > cfg->max_retries) | |
1119 | goto permanent_error; | |
77169812 | 1120 | if (cfg->retry_timeout != XFS_ERR_RETRY_FOREVER && |
a5ea70d2 CM |
1121 | time_after(jiffies, cfg->retry_timeout + bp->b_first_retry_time)) |
1122 | goto permanent_error; | |
bfc60177 | 1123 | |
e6b3bb78 CM |
1124 | /* At unmount we may treat errors differently */ |
1125 | if ((mp->m_flags & XFS_MOUNT_UNMOUNTING) && mp->m_fail_unmount) | |
1126 | goto permanent_error; | |
1127 | ||
0b80ae6e CM |
1128 | /* |
1129 | * Still a transient error, run IO completion failure callbacks and let | |
1130 | * the higher layers retry the buffer. | |
1131 | */ | |
1132 | xfs_buf_do_callbacks_fail(bp); | |
df309390 CM |
1133 | xfs_buf_ioerror(bp, 0); |
1134 | xfs_buf_relse(bp); | |
1135 | return true; | |
0b1b213f | 1136 | |
bfc60177 | 1137 | /* |
df309390 CM |
1138 | * Permanent error - we need to trigger a shutdown if we haven't already |
1139 | * to indicate that inconsistency will result from this action. | |
bfc60177 | 1140 | */ |
df309390 CM |
1141 | permanent_error: |
1142 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1143 | out_stale: | |
c867cb61 | 1144 | xfs_buf_stale(bp); |
b0388bf1 | 1145 | bp->b_flags |= XBF_DONE; |
0b1b213f | 1146 | trace_xfs_buf_error_relse(bp, _RET_IP_); |
df309390 CM |
1147 | return false; |
1148 | } | |
1149 | ||
aac855ab DC |
1150 | static inline bool |
1151 | xfs_buf_had_callback_errors( | |
df309390 CM |
1152 | struct xfs_buf *bp) |
1153 | { | |
f593bf14 | 1154 | |
df309390 | 1155 | /* |
f593bf14 DC |
1156 | * If there is an error, process it. Some errors require us to run |
1157 | * callbacks after failure processing is done so we detect that and take | |
1158 | * appropriate action. | |
df309390 CM |
1159 | */ |
1160 | if (bp->b_error && xfs_buf_iodone_callback_error(bp)) | |
aac855ab | 1161 | return true; |
df309390 CM |
1162 | |
1163 | /* | |
1164 | * Successful IO or permanent error. Either way, we can clear the | |
1165 | * retry state here in preparation for the next error that may occur. | |
1166 | */ | |
1167 | bp->b_last_error = 0; | |
a5ea70d2 | 1168 | bp->b_retries = 0; |
4dd2eb63 | 1169 | bp->b_first_retry_time = 0; |
aac855ab DC |
1170 | return false; |
1171 | } | |
0b1b213f | 1172 | |
aac855ab DC |
1173 | static void |
1174 | xfs_buf_run_callbacks( | |
1175 | struct xfs_buf *bp) | |
1176 | { | |
1177 | ||
1178 | if (xfs_buf_had_callback_errors(bp)) | |
1179 | return; | |
c90821a2 | 1180 | xfs_buf_do_callbacks(bp); |
fb1755a6 | 1181 | bp->b_log_item = NULL; |
f593bf14 DC |
1182 | } |
1183 | ||
1184 | /* | |
b01d1461 | 1185 | * Inode buffer iodone callback function. |
f593bf14 DC |
1186 | */ |
1187 | void | |
b01d1461 | 1188 | xfs_buf_inode_iodone( |
f593bf14 DC |
1189 | struct xfs_buf *bp) |
1190 | { | |
aac855ab DC |
1191 | struct xfs_buf_log_item *blip = bp->b_log_item; |
1192 | struct xfs_log_item *lip; | |
1193 | ||
1194 | if (xfs_buf_had_callback_errors(bp)) | |
1195 | return; | |
1196 | ||
1197 | /* If there is a buf_log_item attached, run its callback */ | |
1198 | if (blip) { | |
1199 | lip = &blip->bli_item; | |
1200 | lip->li_cb(bp, lip); | |
1201 | bp->b_log_item = NULL; | |
1202 | } | |
1203 | ||
1204 | xfs_iflush_done(bp); | |
b01d1461 | 1205 | xfs_buf_ioend_finish(bp); |
1da177e4 LT |
1206 | } |
1207 | ||
f593bf14 | 1208 | /* |
b01d1461 | 1209 | * Dquot buffer iodone callback function. |
f593bf14 DC |
1210 | */ |
1211 | void | |
b01d1461 | 1212 | xfs_buf_dquot_iodone( |
f593bf14 DC |
1213 | struct xfs_buf *bp) |
1214 | { | |
6f5de180 DC |
1215 | struct xfs_buf_log_item *blip = bp->b_log_item; |
1216 | struct xfs_log_item *lip; | |
1217 | ||
1218 | if (xfs_buf_had_callback_errors(bp)) | |
1219 | return; | |
1220 | ||
1221 | /* a newly allocated dquot buffer might have a log item attached */ | |
1222 | if (blip) { | |
1223 | lip = &blip->bli_item; | |
1224 | lip->li_cb(bp, lip); | |
1225 | bp->b_log_item = NULL; | |
1226 | } | |
1227 | ||
1228 | xfs_dquot_done(bp); | |
f593bf14 DC |
1229 | xfs_buf_ioend_finish(bp); |
1230 | } | |
1231 | ||
0c7e5afb | 1232 | /* |
b01d1461 | 1233 | * Dirty buffer iodone callback function. |
0c7e5afb DC |
1234 | */ |
1235 | void | |
b01d1461 | 1236 | xfs_buf_iodone( |
0c7e5afb DC |
1237 | struct xfs_buf *bp) |
1238 | { | |
1239 | xfs_buf_run_callbacks(bp); | |
1240 | xfs_buf_ioend_finish(bp); | |
1241 | } | |
f593bf14 | 1242 | |
1da177e4 LT |
1243 | /* |
1244 | * This is the iodone() function for buffers which have been | |
1245 | * logged. It is called when they are eventually flushed out. | |
1246 | * It should remove the buf item from the AIL, and free the buf item. | |
1247 | * It is called by xfs_buf_iodone_callbacks() above which will take | |
1248 | * care of cleaning up the buffer itself. | |
1249 | */ | |
1da177e4 | 1250 | void |
b01d1461 | 1251 | xfs_buf_item_iodone( |
ca30b2a7 CH |
1252 | struct xfs_buf *bp, |
1253 | struct xfs_log_item *lip) | |
1da177e4 | 1254 | { |
ca30b2a7 | 1255 | ASSERT(BUF_ITEM(lip)->bli_buf == bp); |
1da177e4 | 1256 | |
e1f5dbd7 | 1257 | xfs_buf_rele(bp); |
1da177e4 LT |
1258 | |
1259 | /* | |
849274c1 BF |
1260 | * If we are forcibly shutting down, this may well be off the AIL |
1261 | * already. That's because we simulate the log-committed callbacks to | |
1262 | * unpin these buffers. Or we may never have put this item on AIL | |
1263 | * because of the transaction was aborted forcibly. | |
1264 | * xfs_trans_ail_delete() takes care of these. | |
1da177e4 LT |
1265 | * |
1266 | * Either way, AIL is useless if we're forcing a shutdown. | |
1267 | */ | |
849274c1 | 1268 | xfs_trans_ail_delete(lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 | 1269 | xfs_buf_item_free(BUF_ITEM(lip)); |
1da177e4 | 1270 | } |