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" |
1da177e4 | 16 | #include "xfs_trans_priv.h" |
0b1b213f | 17 | #include "xfs_trace.h" |
239880ef | 18 | #include "xfs_log.h" |
1da177e4 LT |
19 | |
20 | ||
21 | kmem_zone_t *xfs_buf_item_zone; | |
22 | ||
7bfa31d8 CH |
23 | static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip) |
24 | { | |
25 | return container_of(lip, struct xfs_buf_log_item, bli_item); | |
26 | } | |
27 | ||
c90821a2 | 28 | STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp); |
1da177e4 | 29 | |
166d1368 DC |
30 | static inline int |
31 | xfs_buf_log_format_size( | |
32 | struct xfs_buf_log_format *blfp) | |
33 | { | |
34 | return offsetof(struct xfs_buf_log_format, blf_data_map) + | |
35 | (blfp->blf_map_size * sizeof(blfp->blf_data_map[0])); | |
36 | } | |
37 | ||
1da177e4 LT |
38 | /* |
39 | * This returns the number of log iovecs needed to log the | |
40 | * given buf log item. | |
41 | * | |
42 | * It calculates this as 1 iovec for the buf log format structure | |
43 | * and 1 for each stretch of non-contiguous chunks to be logged. | |
44 | * Contiguous chunks are logged in a single iovec. | |
45 | * | |
46 | * If the XFS_BLI_STALE flag has been set, then log nothing. | |
47 | */ | |
166d1368 | 48 | STATIC void |
372cc85e | 49 | xfs_buf_item_size_segment( |
70a20655 CM |
50 | struct xfs_buf_log_item *bip, |
51 | struct xfs_buf_log_format *blfp, | |
52 | int *nvecs, | |
53 | int *nbytes) | |
1da177e4 | 54 | { |
70a20655 CM |
55 | struct xfs_buf *bp = bip->bli_buf; |
56 | int next_bit; | |
57 | int last_bit; | |
1da177e4 | 58 | |
372cc85e DC |
59 | last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
60 | if (last_bit == -1) | |
166d1368 | 61 | return; |
372cc85e DC |
62 | |
63 | /* | |
64 | * initial count for a dirty buffer is 2 vectors - the format structure | |
65 | * and the first dirty region. | |
66 | */ | |
166d1368 DC |
67 | *nvecs += 2; |
68 | *nbytes += xfs_buf_log_format_size(blfp) + XFS_BLF_CHUNK; | |
1da177e4 | 69 | |
1da177e4 LT |
70 | while (last_bit != -1) { |
71 | /* | |
72 | * This takes the bit number to start looking from and | |
73 | * returns the next set bit from there. It returns -1 | |
74 | * if there are no more bits set or the start bit is | |
75 | * beyond the end of the bitmap. | |
76 | */ | |
372cc85e DC |
77 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
78 | last_bit + 1); | |
1da177e4 LT |
79 | /* |
80 | * If we run out of bits, leave the loop, | |
81 | * else if we find a new set of bits bump the number of vecs, | |
82 | * else keep scanning the current set of bits. | |
83 | */ | |
84 | if (next_bit == -1) { | |
372cc85e | 85 | break; |
1da177e4 LT |
86 | } else if (next_bit != last_bit + 1) { |
87 | last_bit = next_bit; | |
166d1368 | 88 | (*nvecs)++; |
c1155410 DC |
89 | } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) != |
90 | (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) + | |
91 | XFS_BLF_CHUNK)) { | |
1da177e4 | 92 | last_bit = next_bit; |
166d1368 | 93 | (*nvecs)++; |
1da177e4 LT |
94 | } else { |
95 | last_bit++; | |
96 | } | |
166d1368 | 97 | *nbytes += XFS_BLF_CHUNK; |
1da177e4 | 98 | } |
1da177e4 LT |
99 | } |
100 | ||
101 | /* | |
372cc85e DC |
102 | * This returns the number of log iovecs needed to log the given buf log item. |
103 | * | |
104 | * It calculates this as 1 iovec for the buf log format structure and 1 for each | |
105 | * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged | |
106 | * in a single iovec. | |
107 | * | |
108 | * Discontiguous buffers need a format structure per region that that is being | |
109 | * logged. This makes the changes in the buffer appear to log recovery as though | |
110 | * they came from separate buffers, just like would occur if multiple buffers | |
111 | * were used instead of a single discontiguous buffer. This enables | |
112 | * discontiguous buffers to be in-memory constructs, completely transparent to | |
113 | * what ends up on disk. | |
114 | * | |
115 | * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log | |
116 | * format structures. | |
1da177e4 | 117 | */ |
166d1368 | 118 | STATIC void |
372cc85e | 119 | xfs_buf_item_size( |
166d1368 DC |
120 | struct xfs_log_item *lip, |
121 | int *nvecs, | |
122 | int *nbytes) | |
1da177e4 | 123 | { |
7bfa31d8 | 124 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
372cc85e DC |
125 | int i; |
126 | ||
127 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
128 | if (bip->bli_flags & XFS_BLI_STALE) { | |
129 | /* | |
130 | * The buffer is stale, so all we need to log | |
131 | * is the buf log format structure with the | |
132 | * cancel flag in it. | |
133 | */ | |
134 | trace_xfs_buf_item_size_stale(bip); | |
b9438173 | 135 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
166d1368 DC |
136 | *nvecs += bip->bli_format_count; |
137 | for (i = 0; i < bip->bli_format_count; i++) { | |
138 | *nbytes += xfs_buf_log_format_size(&bip->bli_formats[i]); | |
139 | } | |
140 | return; | |
372cc85e DC |
141 | } |
142 | ||
143 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); | |
144 | ||
5f6bed76 DC |
145 | if (bip->bli_flags & XFS_BLI_ORDERED) { |
146 | /* | |
147 | * The buffer has been logged just to order it. | |
148 | * It is not being included in the transaction | |
149 | * commit, so no vectors are used at all. | |
150 | */ | |
151 | trace_xfs_buf_item_size_ordered(bip); | |
166d1368 DC |
152 | *nvecs = XFS_LOG_VEC_ORDERED; |
153 | return; | |
5f6bed76 DC |
154 | } |
155 | ||
372cc85e DC |
156 | /* |
157 | * the vector count is based on the number of buffer vectors we have | |
158 | * dirty bits in. This will only be greater than one when we have a | |
159 | * compound buffer with more than one segment dirty. Hence for compound | |
160 | * buffers we need to track which segment the dirty bits correspond to, | |
161 | * and when we move from one segment to the next increment the vector | |
162 | * count for the extra buf log format structure that will need to be | |
163 | * written. | |
164 | */ | |
372cc85e | 165 | for (i = 0; i < bip->bli_format_count; i++) { |
166d1368 DC |
166 | xfs_buf_item_size_segment(bip, &bip->bli_formats[i], |
167 | nvecs, nbytes); | |
372cc85e | 168 | } |
372cc85e | 169 | trace_xfs_buf_item_size(bip); |
372cc85e DC |
170 | } |
171 | ||
1234351c | 172 | static inline void |
7aeb7222 | 173 | xfs_buf_item_copy_iovec( |
bde7cff6 | 174 | struct xfs_log_vec *lv, |
1234351c | 175 | struct xfs_log_iovec **vecp, |
7aeb7222 CH |
176 | struct xfs_buf *bp, |
177 | uint offset, | |
178 | int first_bit, | |
179 | uint nbits) | |
180 | { | |
181 | offset += first_bit * XFS_BLF_CHUNK; | |
bde7cff6 | 182 | xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BCHUNK, |
1234351c CH |
183 | xfs_buf_offset(bp, offset), |
184 | nbits * XFS_BLF_CHUNK); | |
7aeb7222 CH |
185 | } |
186 | ||
187 | static inline bool | |
188 | xfs_buf_item_straddle( | |
189 | struct xfs_buf *bp, | |
190 | uint offset, | |
191 | int next_bit, | |
192 | int last_bit) | |
193 | { | |
194 | return xfs_buf_offset(bp, offset + (next_bit << XFS_BLF_SHIFT)) != | |
195 | (xfs_buf_offset(bp, offset + (last_bit << XFS_BLF_SHIFT)) + | |
196 | XFS_BLF_CHUNK); | |
197 | } | |
198 | ||
1234351c | 199 | static void |
372cc85e DC |
200 | xfs_buf_item_format_segment( |
201 | struct xfs_buf_log_item *bip, | |
bde7cff6 | 202 | struct xfs_log_vec *lv, |
1234351c | 203 | struct xfs_log_iovec **vecp, |
372cc85e DC |
204 | uint offset, |
205 | struct xfs_buf_log_format *blfp) | |
206 | { | |
70a20655 CM |
207 | struct xfs_buf *bp = bip->bli_buf; |
208 | uint base_size; | |
209 | int first_bit; | |
210 | int last_bit; | |
211 | int next_bit; | |
212 | uint nbits; | |
1da177e4 | 213 | |
372cc85e | 214 | /* copy the flags across from the base format item */ |
b9438173 | 215 | blfp->blf_flags = bip->__bli_format.blf_flags; |
1da177e4 LT |
216 | |
217 | /* | |
77c1a08f DC |
218 | * Base size is the actual size of the ondisk structure - it reflects |
219 | * the actual size of the dirty bitmap rather than the size of the in | |
220 | * memory structure. | |
1da177e4 | 221 | */ |
166d1368 | 222 | base_size = xfs_buf_log_format_size(blfp); |
820a554f | 223 | |
820a554f MT |
224 | first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
225 | if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) { | |
226 | /* | |
227 | * If the map is not be dirty in the transaction, mark | |
228 | * the size as zero and do not advance the vector pointer. | |
229 | */ | |
bde7cff6 | 230 | return; |
820a554f MT |
231 | } |
232 | ||
bde7cff6 CH |
233 | blfp = xlog_copy_iovec(lv, vecp, XLOG_REG_TYPE_BFORMAT, blfp, base_size); |
234 | blfp->blf_size = 1; | |
1da177e4 LT |
235 | |
236 | if (bip->bli_flags & XFS_BLI_STALE) { | |
237 | /* | |
238 | * The buffer is stale, so all we need to log | |
239 | * is the buf log format structure with the | |
240 | * cancel flag in it. | |
241 | */ | |
0b1b213f | 242 | trace_xfs_buf_item_format_stale(bip); |
372cc85e | 243 | ASSERT(blfp->blf_flags & XFS_BLF_CANCEL); |
bde7cff6 | 244 | return; |
1da177e4 LT |
245 | } |
246 | ||
5f6bed76 | 247 | |
1da177e4 LT |
248 | /* |
249 | * Fill in an iovec for each set of contiguous chunks. | |
250 | */ | |
1da177e4 LT |
251 | last_bit = first_bit; |
252 | nbits = 1; | |
253 | for (;;) { | |
254 | /* | |
255 | * This takes the bit number to start looking from and | |
256 | * returns the next set bit from there. It returns -1 | |
257 | * if there are no more bits set or the start bit is | |
258 | * beyond the end of the bitmap. | |
259 | */ | |
372cc85e DC |
260 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
261 | (uint)last_bit + 1); | |
1da177e4 | 262 | /* |
7aeb7222 CH |
263 | * If we run out of bits fill in the last iovec and get out of |
264 | * the loop. Else if we start a new set of bits then fill in | |
265 | * the iovec for the series we were looking at and start | |
266 | * counting the bits in the new one. Else we're still in the | |
267 | * same set of bits so just keep counting and scanning. | |
1da177e4 LT |
268 | */ |
269 | if (next_bit == -1) { | |
bde7cff6 | 270 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
7aeb7222 | 271 | first_bit, nbits); |
bde7cff6 | 272 | blfp->blf_size++; |
1da177e4 | 273 | break; |
7aeb7222 CH |
274 | } else if (next_bit != last_bit + 1 || |
275 | xfs_buf_item_straddle(bp, offset, next_bit, last_bit)) { | |
bde7cff6 | 276 | xfs_buf_item_copy_iovec(lv, vecp, bp, offset, |
1234351c | 277 | first_bit, nbits); |
bde7cff6 | 278 | blfp->blf_size++; |
1da177e4 LT |
279 | first_bit = next_bit; |
280 | last_bit = next_bit; | |
281 | nbits = 1; | |
282 | } else { | |
283 | last_bit++; | |
284 | nbits++; | |
285 | } | |
286 | } | |
372cc85e DC |
287 | } |
288 | ||
289 | /* | |
290 | * This is called to fill in the vector of log iovecs for the | |
291 | * given log buf item. It fills the first entry with a buf log | |
292 | * format structure, and the rest point to contiguous chunks | |
293 | * within the buffer. | |
294 | */ | |
295 | STATIC void | |
296 | xfs_buf_item_format( | |
297 | struct xfs_log_item *lip, | |
bde7cff6 | 298 | struct xfs_log_vec *lv) |
372cc85e DC |
299 | { |
300 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); | |
301 | struct xfs_buf *bp = bip->bli_buf; | |
bde7cff6 | 302 | struct xfs_log_iovec *vecp = NULL; |
372cc85e DC |
303 | uint offset = 0; |
304 | int i; | |
305 | ||
306 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
307 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
308 | (bip->bli_flags & XFS_BLI_STALE)); | |
0d612fb5 DC |
309 | ASSERT((bip->bli_flags & XFS_BLI_STALE) || |
310 | (xfs_blft_from_flags(&bip->__bli_format) > XFS_BLFT_UNKNOWN_BUF | |
311 | && xfs_blft_from_flags(&bip->__bli_format) < XFS_BLFT_MAX_BUF)); | |
e9385cc6 BF |
312 | ASSERT(!(bip->bli_flags & XFS_BLI_ORDERED) || |
313 | (bip->bli_flags & XFS_BLI_STALE)); | |
0d612fb5 | 314 | |
372cc85e DC |
315 | |
316 | /* | |
317 | * If it is an inode buffer, transfer the in-memory state to the | |
ddf6ad01 DC |
318 | * format flags and clear the in-memory state. |
319 | * | |
320 | * For buffer based inode allocation, we do not transfer | |
372cc85e DC |
321 | * this state if the inode buffer allocation has not yet been committed |
322 | * to the log as setting the XFS_BLI_INODE_BUF flag will prevent | |
323 | * correct replay of the inode allocation. | |
ddf6ad01 DC |
324 | * |
325 | * For icreate item based inode allocation, the buffers aren't written | |
326 | * to the journal during allocation, and hence we should always tag the | |
327 | * buffer as an inode buffer so that the correct unlinked list replay | |
328 | * occurs during recovery. | |
372cc85e DC |
329 | */ |
330 | if (bip->bli_flags & XFS_BLI_INODE_BUF) { | |
ddf6ad01 DC |
331 | if (xfs_sb_version_hascrc(&lip->li_mountp->m_sb) || |
332 | !((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && | |
372cc85e | 333 | xfs_log_item_in_current_chkpt(lip))) |
b9438173 | 334 | bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF; |
372cc85e DC |
335 | bip->bli_flags &= ~XFS_BLI_INODE_BUF; |
336 | } | |
337 | ||
338 | for (i = 0; i < bip->bli_format_count; i++) { | |
bde7cff6 | 339 | xfs_buf_item_format_segment(bip, lv, &vecp, offset, |
1234351c | 340 | &bip->bli_formats[i]); |
a3916e52 | 341 | offset += BBTOB(bp->b_maps[i].bm_len); |
372cc85e | 342 | } |
1da177e4 LT |
343 | |
344 | /* | |
345 | * Check to make sure everything is consistent. | |
346 | */ | |
0b1b213f | 347 | trace_xfs_buf_item_format(bip); |
1da177e4 LT |
348 | } |
349 | ||
350 | /* | |
64fc35de | 351 | * This is called to pin the buffer associated with the buf log item in memory |
4d16e924 | 352 | * so it cannot be written out. |
64fc35de DC |
353 | * |
354 | * We also always take a reference to the buffer log item here so that the bli | |
355 | * is held while the item is pinned in memory. This means that we can | |
356 | * unconditionally drop the reference count a transaction holds when the | |
357 | * transaction is completed. | |
1da177e4 | 358 | */ |
ba0f32d4 | 359 | STATIC void |
1da177e4 | 360 | xfs_buf_item_pin( |
7bfa31d8 | 361 | struct xfs_log_item *lip) |
1da177e4 | 362 | { |
7bfa31d8 | 363 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
1da177e4 | 364 | |
1da177e4 LT |
365 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
366 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
5f6bed76 | 367 | (bip->bli_flags & XFS_BLI_ORDERED) || |
1da177e4 | 368 | (bip->bli_flags & XFS_BLI_STALE)); |
7bfa31d8 | 369 | |
0b1b213f | 370 | trace_xfs_buf_item_pin(bip); |
4d16e924 CH |
371 | |
372 | atomic_inc(&bip->bli_refcount); | |
373 | atomic_inc(&bip->bli_buf->b_pin_count); | |
1da177e4 LT |
374 | } |
375 | ||
1da177e4 LT |
376 | /* |
377 | * This is called to unpin the buffer associated with the buf log | |
378 | * item which was previously pinned with a call to xfs_buf_item_pin(). | |
1da177e4 LT |
379 | * |
380 | * Also drop the reference to the buf item for the current transaction. | |
381 | * If the XFS_BLI_STALE flag is set and we are the last reference, | |
382 | * then free up the buf log item and unlock the buffer. | |
9412e318 CH |
383 | * |
384 | * If the remove flag is set we are called from uncommit in the | |
385 | * forced-shutdown path. If that is true and the reference count on | |
386 | * the log item is going to drop to zero we need to free the item's | |
387 | * descriptor in the transaction. | |
1da177e4 | 388 | */ |
ba0f32d4 | 389 | STATIC void |
1da177e4 | 390 | xfs_buf_item_unpin( |
7bfa31d8 | 391 | struct xfs_log_item *lip, |
9412e318 | 392 | int remove) |
1da177e4 | 393 | { |
7bfa31d8 | 394 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
70a20655 CM |
395 | xfs_buf_t *bp = bip->bli_buf; |
396 | struct xfs_ail *ailp = lip->li_ailp; | |
397 | int stale = bip->bli_flags & XFS_BLI_STALE; | |
398 | int freed; | |
1da177e4 | 399 | |
fb1755a6 | 400 | ASSERT(bp->b_log_item == bip); |
1da177e4 | 401 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
9412e318 | 402 | |
0b1b213f | 403 | trace_xfs_buf_item_unpin(bip); |
1da177e4 LT |
404 | |
405 | freed = atomic_dec_and_test(&bip->bli_refcount); | |
4d16e924 CH |
406 | |
407 | if (atomic_dec_and_test(&bp->b_pin_count)) | |
408 | wake_up_all(&bp->b_waiters); | |
7bfa31d8 | 409 | |
1da177e4 LT |
410 | if (freed && stale) { |
411 | ASSERT(bip->bli_flags & XFS_BLI_STALE); | |
0c842ad4 | 412 | ASSERT(xfs_buf_islocked(bp)); |
5cfd28b6 | 413 | ASSERT(bp->b_flags & XBF_STALE); |
b9438173 | 414 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
9412e318 | 415 | |
0b1b213f CH |
416 | trace_xfs_buf_item_unpin_stale(bip); |
417 | ||
9412e318 CH |
418 | if (remove) { |
419 | /* | |
e34a314c DC |
420 | * If we are in a transaction context, we have to |
421 | * remove the log item from the transaction as we are | |
422 | * about to release our reference to the buffer. If we | |
423 | * don't, the unlock that occurs later in | |
424 | * xfs_trans_uncommit() will try to reference the | |
9412e318 CH |
425 | * buffer which we no longer have a hold on. |
426 | */ | |
e6631f85 | 427 | if (!list_empty(&lip->li_trans)) |
e34a314c | 428 | xfs_trans_del_item(lip); |
9412e318 CH |
429 | |
430 | /* | |
431 | * Since the transaction no longer refers to the buffer, | |
432 | * the buffer should no longer refer to the transaction. | |
433 | */ | |
bf9d9013 | 434 | bp->b_transp = NULL; |
9412e318 CH |
435 | } |
436 | ||
1da177e4 LT |
437 | /* |
438 | * If we get called here because of an IO error, we may | |
783a2f65 | 439 | * or may not have the item on the AIL. xfs_trans_ail_delete() |
1da177e4 | 440 | * will take care of that situation. |
783a2f65 | 441 | * xfs_trans_ail_delete() drops the AIL lock. |
1da177e4 LT |
442 | */ |
443 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { | |
c90821a2 | 444 | xfs_buf_do_callbacks(bp); |
fb1755a6 | 445 | bp->b_log_item = NULL; |
643c8c05 | 446 | list_del_init(&bp->b_li_list); |
cb669ca5 | 447 | bp->b_iodone = NULL; |
1da177e4 | 448 | } else { |
57e80956 | 449 | spin_lock(&ailp->ail_lock); |
04913fdd | 450 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR); |
1da177e4 | 451 | xfs_buf_item_relse(bp); |
fb1755a6 | 452 | ASSERT(bp->b_log_item == NULL); |
1da177e4 LT |
453 | } |
454 | xfs_buf_relse(bp); | |
960c60af | 455 | } else if (freed && remove) { |
137fff09 DC |
456 | /* |
457 | * There are currently two references to the buffer - the active | |
458 | * LRU reference and the buf log item. What we are about to do | |
459 | * here - simulate a failed IO completion - requires 3 | |
460 | * references. | |
461 | * | |
462 | * The LRU reference is removed by the xfs_buf_stale() call. The | |
463 | * buf item reference is removed by the xfs_buf_iodone() | |
464 | * callback that is run by xfs_buf_do_callbacks() during ioend | |
465 | * processing (via the bp->b_iodone callback), and then finally | |
466 | * the ioend processing will drop the IO reference if the buffer | |
467 | * is marked XBF_ASYNC. | |
468 | * | |
469 | * Hence we need to take an additional reference here so that IO | |
470 | * completion processing doesn't free the buffer prematurely. | |
471 | */ | |
960c60af | 472 | xfs_buf_lock(bp); |
137fff09 DC |
473 | xfs_buf_hold(bp); |
474 | bp->b_flags |= XBF_ASYNC; | |
2451337d | 475 | xfs_buf_ioerror(bp, -EIO); |
b0388bf1 | 476 | bp->b_flags &= ~XBF_DONE; |
960c60af | 477 | xfs_buf_stale(bp); |
e8aaba9a | 478 | xfs_buf_ioend(bp); |
1da177e4 LT |
479 | } |
480 | } | |
481 | ||
ac8809f9 DC |
482 | /* |
483 | * Buffer IO error rate limiting. Limit it to no more than 10 messages per 30 | |
484 | * seconds so as to not spam logs too much on repeated detection of the same | |
485 | * buffer being bad.. | |
486 | */ | |
487 | ||
02cc1876 | 488 | static DEFINE_RATELIMIT_STATE(xfs_buf_write_fail_rl_state, 30 * HZ, 10); |
ac8809f9 | 489 | |
ba0f32d4 | 490 | STATIC uint |
43ff2122 CH |
491 | xfs_buf_item_push( |
492 | struct xfs_log_item *lip, | |
493 | struct list_head *buffer_list) | |
1da177e4 | 494 | { |
7bfa31d8 CH |
495 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
496 | struct xfs_buf *bp = bip->bli_buf; | |
43ff2122 | 497 | uint rval = XFS_ITEM_SUCCESS; |
1da177e4 | 498 | |
811e64c7 | 499 | if (xfs_buf_ispinned(bp)) |
1da177e4 | 500 | return XFS_ITEM_PINNED; |
5337fe9b BF |
501 | if (!xfs_buf_trylock(bp)) { |
502 | /* | |
503 | * If we have just raced with a buffer being pinned and it has | |
504 | * been marked stale, we could end up stalling until someone else | |
505 | * issues a log force to unpin the stale buffer. Check for the | |
506 | * race condition here so xfsaild recognizes the buffer is pinned | |
507 | * and queues a log force to move it along. | |
508 | */ | |
509 | if (xfs_buf_ispinned(bp)) | |
510 | return XFS_ITEM_PINNED; | |
1da177e4 | 511 | return XFS_ITEM_LOCKED; |
5337fe9b | 512 | } |
1da177e4 | 513 | |
1da177e4 | 514 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
43ff2122 CH |
515 | |
516 | trace_xfs_buf_item_push(bip); | |
517 | ||
ac8809f9 DC |
518 | /* has a previous flush failed due to IO errors? */ |
519 | if ((bp->b_flags & XBF_WRITE_FAIL) && | |
fdadf267 | 520 | ___ratelimit(&xfs_buf_write_fail_rl_state, "XFS: Failing async write")) { |
dbd329f1 | 521 | xfs_warn(bp->b_mount, |
fdadf267 | 522 | "Failing async write on buffer block 0x%llx. Retrying async write.", |
ac8809f9 DC |
523 | (long long)bp->b_bn); |
524 | } | |
525 | ||
43ff2122 CH |
526 | if (!xfs_buf_delwri_queue(bp, buffer_list)) |
527 | rval = XFS_ITEM_FLUSHING; | |
528 | xfs_buf_unlock(bp); | |
529 | return rval; | |
1da177e4 LT |
530 | } |
531 | ||
95808459 BF |
532 | /* |
533 | * Drop the buffer log item refcount and take appropriate action. This helper | |
534 | * determines whether the bli must be freed or not, since a decrement to zero | |
535 | * does not necessarily mean the bli is unused. | |
536 | * | |
537 | * Return true if the bli is freed, false otherwise. | |
538 | */ | |
539 | bool | |
540 | xfs_buf_item_put( | |
541 | struct xfs_buf_log_item *bip) | |
542 | { | |
543 | struct xfs_log_item *lip = &bip->bli_item; | |
544 | bool aborted; | |
545 | bool dirty; | |
546 | ||
547 | /* drop the bli ref and return if it wasn't the last one */ | |
548 | if (!atomic_dec_and_test(&bip->bli_refcount)) | |
549 | return false; | |
550 | ||
551 | /* | |
552 | * We dropped the last ref and must free the item if clean or aborted. | |
553 | * If the bli is dirty and non-aborted, the buffer was clean in the | |
554 | * transaction but still awaiting writeback from previous changes. In | |
555 | * that case, the bli is freed on buffer writeback completion. | |
556 | */ | |
557 | aborted = test_bit(XFS_LI_ABORTED, &lip->li_flags) || | |
558 | XFS_FORCED_SHUTDOWN(lip->li_mountp); | |
559 | dirty = bip->bli_flags & XFS_BLI_DIRTY; | |
560 | if (dirty && !aborted) | |
561 | return false; | |
562 | ||
563 | /* | |
564 | * The bli is aborted or clean. An aborted item may be in the AIL | |
565 | * regardless of dirty state. For example, consider an aborted | |
566 | * transaction that invalidated a dirty bli and cleared the dirty | |
567 | * state. | |
568 | */ | |
569 | if (aborted) | |
570 | xfs_trans_ail_remove(lip, SHUTDOWN_LOG_IO_ERROR); | |
571 | xfs_buf_item_relse(bip->bli_buf); | |
572 | return true; | |
573 | } | |
574 | ||
1da177e4 | 575 | /* |
64fc35de DC |
576 | * Release the buffer associated with the buf log item. If there is no dirty |
577 | * logged data associated with the buffer recorded in the buf log item, then | |
578 | * free the buf log item and remove the reference to it in the buffer. | |
1da177e4 | 579 | * |
64fc35de DC |
580 | * This call ignores the recursion count. It is only called when the buffer |
581 | * should REALLY be unlocked, regardless of the recursion count. | |
1da177e4 | 582 | * |
64fc35de DC |
583 | * We unconditionally drop the transaction's reference to the log item. If the |
584 | * item was logged, then another reference was taken when it was pinned, so we | |
585 | * can safely drop the transaction reference now. This also allows us to avoid | |
586 | * potential races with the unpin code freeing the bli by not referencing the | |
587 | * bli after we've dropped the reference count. | |
588 | * | |
589 | * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item | |
590 | * if necessary but do not unlock the buffer. This is for support of | |
591 | * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't | |
592 | * free the item. | |
1da177e4 | 593 | */ |
ba0f32d4 | 594 | STATIC void |
ddf92053 | 595 | xfs_buf_item_release( |
7bfa31d8 | 596 | struct xfs_log_item *lip) |
1da177e4 | 597 | { |
7bfa31d8 CH |
598 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
599 | struct xfs_buf *bp = bip->bli_buf; | |
95808459 | 600 | bool released; |
d9183105 | 601 | bool hold = bip->bli_flags & XFS_BLI_HOLD; |
d9183105 | 602 | bool stale = bip->bli_flags & XFS_BLI_STALE; |
7bf7a193 | 603 | #if defined(DEBUG) || defined(XFS_WARN) |
d9183105 | 604 | bool ordered = bip->bli_flags & XFS_BLI_ORDERED; |
95808459 | 605 | bool dirty = bip->bli_flags & XFS_BLI_DIRTY; |
4d09807f BF |
606 | bool aborted = test_bit(XFS_LI_ABORTED, |
607 | &lip->li_flags); | |
7bf7a193 | 608 | #endif |
1da177e4 | 609 | |
ddf92053 | 610 | trace_xfs_buf_item_release(bip); |
1da177e4 LT |
611 | |
612 | /* | |
6453c65d BF |
613 | * The bli dirty state should match whether the blf has logged segments |
614 | * except for ordered buffers, where only the bli should be dirty. | |
1da177e4 | 615 | */ |
6453c65d BF |
616 | ASSERT((!ordered && dirty == xfs_buf_item_dirty_format(bip)) || |
617 | (ordered && dirty && !xfs_buf_item_dirty_format(bip))); | |
d9183105 BF |
618 | ASSERT(!stale || (bip->__bli_format.blf_flags & XFS_BLF_CANCEL)); |
619 | ||
46f9d2eb | 620 | /* |
d9183105 BF |
621 | * Clear the buffer's association with this transaction and |
622 | * per-transaction state from the bli, which has been copied above. | |
623 | */ | |
624 | bp->b_transp = NULL; | |
625 | bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD | XFS_BLI_ORDERED); | |
626 | ||
627 | /* | |
95808459 BF |
628 | * Unref the item and unlock the buffer unless held or stale. Stale |
629 | * buffers remain locked until final unpin unless the bli is freed by | |
630 | * the unref call. The latter implies shutdown because buffer | |
631 | * invalidation dirties the bli and transaction. | |
46f9d2eb | 632 | */ |
95808459 BF |
633 | released = xfs_buf_item_put(bip); |
634 | if (hold || (stale && !released)) | |
d9183105 | 635 | return; |
4d09807f | 636 | ASSERT(!stale || aborted); |
95808459 | 637 | xfs_buf_relse(bp); |
1da177e4 LT |
638 | } |
639 | ||
ddf92053 CH |
640 | STATIC void |
641 | xfs_buf_item_committing( | |
642 | struct xfs_log_item *lip, | |
643 | xfs_lsn_t commit_lsn) | |
644 | { | |
645 | return xfs_buf_item_release(lip); | |
646 | } | |
647 | ||
1da177e4 LT |
648 | /* |
649 | * This is called to find out where the oldest active copy of the | |
650 | * buf log item in the on disk log resides now that the last log | |
651 | * write of it completed at the given lsn. | |
652 | * We always re-log all the dirty data in a buffer, so usually the | |
653 | * latest copy in the on disk log is the only one that matters. For | |
654 | * those cases we simply return the given lsn. | |
655 | * | |
656 | * The one exception to this is for buffers full of newly allocated | |
657 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF | |
658 | * flag set, indicating that only the di_next_unlinked fields from the | |
659 | * inodes in the buffers will be replayed during recovery. If the | |
660 | * original newly allocated inode images have not yet been flushed | |
661 | * when the buffer is so relogged, then we need to make sure that we | |
662 | * keep the old images in the 'active' portion of the log. We do this | |
663 | * by returning the original lsn of that transaction here rather than | |
664 | * the current one. | |
665 | */ | |
ba0f32d4 | 666 | STATIC xfs_lsn_t |
1da177e4 | 667 | xfs_buf_item_committed( |
7bfa31d8 | 668 | struct xfs_log_item *lip, |
1da177e4 LT |
669 | xfs_lsn_t lsn) |
670 | { | |
7bfa31d8 CH |
671 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
672 | ||
0b1b213f CH |
673 | trace_xfs_buf_item_committed(bip); |
674 | ||
7bfa31d8 CH |
675 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0) |
676 | return lip->li_lsn; | |
677 | return lsn; | |
1da177e4 LT |
678 | } |
679 | ||
272e42b2 | 680 | static const struct xfs_item_ops xfs_buf_item_ops = { |
7bfa31d8 CH |
681 | .iop_size = xfs_buf_item_size, |
682 | .iop_format = xfs_buf_item_format, | |
683 | .iop_pin = xfs_buf_item_pin, | |
684 | .iop_unpin = xfs_buf_item_unpin, | |
ddf92053 CH |
685 | .iop_release = xfs_buf_item_release, |
686 | .iop_committing = xfs_buf_item_committing, | |
7bfa31d8 CH |
687 | .iop_committed = xfs_buf_item_committed, |
688 | .iop_push = xfs_buf_item_push, | |
1da177e4 LT |
689 | }; |
690 | ||
372cc85e DC |
691 | STATIC int |
692 | xfs_buf_item_get_format( | |
693 | struct xfs_buf_log_item *bip, | |
694 | int count) | |
695 | { | |
696 | ASSERT(bip->bli_formats == NULL); | |
697 | bip->bli_format_count = count; | |
698 | ||
699 | if (count == 1) { | |
b9438173 | 700 | bip->bli_formats = &bip->__bli_format; |
372cc85e DC |
701 | return 0; |
702 | } | |
703 | ||
704 | bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format), | |
707e0dda | 705 | 0); |
372cc85e | 706 | if (!bip->bli_formats) |
2451337d | 707 | return -ENOMEM; |
372cc85e DC |
708 | return 0; |
709 | } | |
710 | ||
711 | STATIC void | |
712 | xfs_buf_item_free_format( | |
713 | struct xfs_buf_log_item *bip) | |
714 | { | |
b9438173 | 715 | if (bip->bli_formats != &bip->__bli_format) { |
372cc85e DC |
716 | kmem_free(bip->bli_formats); |
717 | bip->bli_formats = NULL; | |
718 | } | |
719 | } | |
1da177e4 LT |
720 | |
721 | /* | |
722 | * Allocate a new buf log item to go with the given buffer. | |
fb1755a6 CM |
723 | * Set the buffer's b_log_item field to point to the new |
724 | * buf log item. | |
1da177e4 | 725 | */ |
f79af0b9 | 726 | int |
1da177e4 | 727 | xfs_buf_item_init( |
f79af0b9 DC |
728 | struct xfs_buf *bp, |
729 | struct xfs_mount *mp) | |
1da177e4 | 730 | { |
fb1755a6 | 731 | struct xfs_buf_log_item *bip = bp->b_log_item; |
1da177e4 LT |
732 | int chunks; |
733 | int map_size; | |
372cc85e DC |
734 | int error; |
735 | int i; | |
1da177e4 LT |
736 | |
737 | /* | |
738 | * Check to see if there is already a buf log item for | |
fb1755a6 | 739 | * this buffer. If we do already have one, there is |
1da177e4 LT |
740 | * nothing to do here so return. |
741 | */ | |
dbd329f1 | 742 | ASSERT(bp->b_mount == mp); |
1a2ebf83 | 743 | if (bip) { |
fb1755a6 | 744 | ASSERT(bip->bli_item.li_type == XFS_LI_BUF); |
1a2ebf83 DC |
745 | ASSERT(!bp->b_transp); |
746 | ASSERT(bip->bli_buf == bp); | |
f79af0b9 | 747 | return 0; |
fb1755a6 | 748 | } |
1da177e4 | 749 | |
707e0dda | 750 | bip = kmem_zone_zalloc(xfs_buf_item_zone, 0); |
43f5efc5 | 751 | xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops); |
1da177e4 | 752 | bip->bli_buf = bp; |
372cc85e DC |
753 | |
754 | /* | |
755 | * chunks is the number of XFS_BLF_CHUNK size pieces the buffer | |
756 | * can be divided into. Make sure not to truncate any pieces. | |
757 | * map_size is the size of the bitmap needed to describe the | |
758 | * chunks of the buffer. | |
759 | * | |
760 | * Discontiguous buffer support follows the layout of the underlying | |
761 | * buffer. This makes the implementation as simple as possible. | |
762 | */ | |
763 | error = xfs_buf_item_get_format(bip, bp->b_map_count); | |
764 | ASSERT(error == 0); | |
f79af0b9 DC |
765 | if (error) { /* to stop gcc throwing set-but-unused warnings */ |
766 | kmem_zone_free(xfs_buf_item_zone, bip); | |
767 | return error; | |
768 | } | |
769 | ||
372cc85e DC |
770 | |
771 | for (i = 0; i < bip->bli_format_count; i++) { | |
772 | chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len), | |
773 | XFS_BLF_CHUNK); | |
774 | map_size = DIV_ROUND_UP(chunks, NBWORD); | |
775 | ||
776 | bip->bli_formats[i].blf_type = XFS_LI_BUF; | |
777 | bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn; | |
778 | bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len; | |
779 | bip->bli_formats[i].blf_map_size = map_size; | |
780 | } | |
1da177e4 | 781 | |
fb1755a6 | 782 | bp->b_log_item = bip; |
f79af0b9 DC |
783 | xfs_buf_hold(bp); |
784 | return 0; | |
1da177e4 LT |
785 | } |
786 | ||
787 | ||
788 | /* | |
789 | * Mark bytes first through last inclusive as dirty in the buf | |
790 | * item's bitmap. | |
791 | */ | |
632b89e8 | 792 | static void |
372cc85e | 793 | xfs_buf_item_log_segment( |
1da177e4 | 794 | uint first, |
372cc85e DC |
795 | uint last, |
796 | uint *map) | |
1da177e4 LT |
797 | { |
798 | uint first_bit; | |
799 | uint last_bit; | |
800 | uint bits_to_set; | |
801 | uint bits_set; | |
802 | uint word_num; | |
803 | uint *wordp; | |
804 | uint bit; | |
805 | uint end_bit; | |
806 | uint mask; | |
807 | ||
1da177e4 LT |
808 | /* |
809 | * Convert byte offsets to bit numbers. | |
810 | */ | |
c1155410 DC |
811 | first_bit = first >> XFS_BLF_SHIFT; |
812 | last_bit = last >> XFS_BLF_SHIFT; | |
1da177e4 LT |
813 | |
814 | /* | |
815 | * Calculate the total number of bits to be set. | |
816 | */ | |
817 | bits_to_set = last_bit - first_bit + 1; | |
818 | ||
819 | /* | |
820 | * Get a pointer to the first word in the bitmap | |
821 | * to set a bit in. | |
822 | */ | |
823 | word_num = first_bit >> BIT_TO_WORD_SHIFT; | |
372cc85e | 824 | wordp = &map[word_num]; |
1da177e4 LT |
825 | |
826 | /* | |
827 | * Calculate the starting bit in the first word. | |
828 | */ | |
829 | bit = first_bit & (uint)(NBWORD - 1); | |
830 | ||
831 | /* | |
832 | * First set any bits in the first word of our range. | |
833 | * If it starts at bit 0 of the word, it will be | |
834 | * set below rather than here. That is what the variable | |
835 | * bit tells us. The variable bits_set tracks the number | |
836 | * of bits that have been set so far. End_bit is the number | |
837 | * of the last bit to be set in this word plus one. | |
838 | */ | |
839 | if (bit) { | |
9bb54cb5 | 840 | end_bit = min(bit + bits_to_set, (uint)NBWORD); |
79c350e4 | 841 | mask = ((1U << (end_bit - bit)) - 1) << bit; |
1da177e4 LT |
842 | *wordp |= mask; |
843 | wordp++; | |
844 | bits_set = end_bit - bit; | |
845 | } else { | |
846 | bits_set = 0; | |
847 | } | |
848 | ||
849 | /* | |
850 | * Now set bits a whole word at a time that are between | |
851 | * first_bit and last_bit. | |
852 | */ | |
853 | while ((bits_to_set - bits_set) >= NBWORD) { | |
854 | *wordp |= 0xffffffff; | |
855 | bits_set += NBWORD; | |
856 | wordp++; | |
857 | } | |
858 | ||
859 | /* | |
860 | * Finally, set any bits left to be set in one last partial word. | |
861 | */ | |
862 | end_bit = bits_to_set - bits_set; | |
863 | if (end_bit) { | |
79c350e4 | 864 | mask = (1U << end_bit) - 1; |
1da177e4 LT |
865 | *wordp |= mask; |
866 | } | |
1da177e4 LT |
867 | } |
868 | ||
372cc85e DC |
869 | /* |
870 | * Mark bytes first through last inclusive as dirty in the buf | |
871 | * item's bitmap. | |
872 | */ | |
873 | void | |
874 | xfs_buf_item_log( | |
70a20655 | 875 | struct xfs_buf_log_item *bip, |
372cc85e DC |
876 | uint first, |
877 | uint last) | |
878 | { | |
879 | int i; | |
880 | uint start; | |
881 | uint end; | |
882 | struct xfs_buf *bp = bip->bli_buf; | |
883 | ||
372cc85e DC |
884 | /* |
885 | * walk each buffer segment and mark them dirty appropriately. | |
886 | */ | |
887 | start = 0; | |
888 | for (i = 0; i < bip->bli_format_count; i++) { | |
889 | if (start > last) | |
890 | break; | |
a3916e52 BF |
891 | end = start + BBTOB(bp->b_maps[i].bm_len) - 1; |
892 | ||
893 | /* skip to the map that includes the first byte to log */ | |
372cc85e DC |
894 | if (first > end) { |
895 | start += BBTOB(bp->b_maps[i].bm_len); | |
896 | continue; | |
897 | } | |
a3916e52 BF |
898 | |
899 | /* | |
900 | * Trim the range to this segment and mark it in the bitmap. | |
901 | * Note that we must convert buffer offsets to segment relative | |
902 | * offsets (e.g., the first byte of each segment is byte 0 of | |
903 | * that segment). | |
904 | */ | |
372cc85e DC |
905 | if (first < start) |
906 | first = start; | |
907 | if (end > last) | |
908 | end = last; | |
a3916e52 | 909 | xfs_buf_item_log_segment(first - start, end - start, |
372cc85e DC |
910 | &bip->bli_formats[i].blf_data_map[0]); |
911 | ||
a3916e52 | 912 | start += BBTOB(bp->b_maps[i].bm_len); |
372cc85e DC |
913 | } |
914 | } | |
915 | ||
1da177e4 | 916 | |
6453c65d BF |
917 | /* |
918 | * Return true if the buffer has any ranges logged/dirtied by a transaction, | |
919 | * false otherwise. | |
920 | */ | |
921 | bool | |
922 | xfs_buf_item_dirty_format( | |
923 | struct xfs_buf_log_item *bip) | |
924 | { | |
925 | int i; | |
926 | ||
927 | for (i = 0; i < bip->bli_format_count; i++) { | |
928 | if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map, | |
929 | bip->bli_formats[i].blf_map_size)) | |
930 | return true; | |
931 | } | |
932 | ||
933 | return false; | |
934 | } | |
935 | ||
e1f5dbd7 LM |
936 | STATIC void |
937 | xfs_buf_item_free( | |
70a20655 | 938 | struct xfs_buf_log_item *bip) |
e1f5dbd7 | 939 | { |
372cc85e | 940 | xfs_buf_item_free_format(bip); |
b1c5ebb2 | 941 | kmem_free(bip->bli_item.li_lv_shadow); |
e1f5dbd7 LM |
942 | kmem_zone_free(xfs_buf_item_zone, bip); |
943 | } | |
944 | ||
1da177e4 LT |
945 | /* |
946 | * This is called when the buf log item is no longer needed. It should | |
947 | * free the buf log item associated with the given buffer and clear | |
948 | * the buffer's pointer to the buf log item. If there are no more | |
949 | * items in the list, clear the b_iodone field of the buffer (see | |
950 | * xfs_buf_attach_iodone() below). | |
951 | */ | |
952 | void | |
953 | xfs_buf_item_relse( | |
954 | xfs_buf_t *bp) | |
955 | { | |
fb1755a6 | 956 | struct xfs_buf_log_item *bip = bp->b_log_item; |
1da177e4 | 957 | |
0b1b213f | 958 | trace_xfs_buf_item_relse(bp, _RET_IP_); |
5f6bed76 | 959 | ASSERT(!(bip->bli_item.li_flags & XFS_LI_IN_AIL)); |
0b1b213f | 960 | |
fb1755a6 | 961 | bp->b_log_item = NULL; |
643c8c05 | 962 | if (list_empty(&bp->b_li_list)) |
cb669ca5 | 963 | bp->b_iodone = NULL; |
adadbeef | 964 | |
e1f5dbd7 LM |
965 | xfs_buf_rele(bp); |
966 | xfs_buf_item_free(bip); | |
1da177e4 LT |
967 | } |
968 | ||
969 | ||
970 | /* | |
971 | * Add the given log item with its callback to the list of callbacks | |
972 | * to be called when the buffer's I/O completes. If it is not set | |
973 | * already, set the buffer's b_iodone() routine to be | |
974 | * xfs_buf_iodone_callbacks() and link the log item into the list of | |
fb1755a6 | 975 | * items rooted at b_li_list. |
1da177e4 LT |
976 | */ |
977 | void | |
978 | xfs_buf_attach_iodone( | |
efe2330f CH |
979 | struct xfs_buf *bp, |
980 | void (*cb)(struct xfs_buf *, struct xfs_log_item *), | |
981 | struct xfs_log_item *lip) | |
1da177e4 | 982 | { |
0c842ad4 | 983 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
984 | |
985 | lip->li_cb = cb; | |
643c8c05 | 986 | list_add_tail(&lip->li_bio_list, &bp->b_li_list); |
1da177e4 | 987 | |
cb669ca5 CH |
988 | ASSERT(bp->b_iodone == NULL || |
989 | bp->b_iodone == xfs_buf_iodone_callbacks); | |
990 | bp->b_iodone = xfs_buf_iodone_callbacks; | |
1da177e4 LT |
991 | } |
992 | ||
c90821a2 DC |
993 | /* |
994 | * We can have many callbacks on a buffer. Running the callbacks individually | |
995 | * can cause a lot of contention on the AIL lock, so we allow for a single | |
643c8c05 CM |
996 | * callback to be able to scan the remaining items in bp->b_li_list for other |
997 | * items of the same type and callback to be processed in the first call. | |
c90821a2 DC |
998 | * |
999 | * As a result, the loop walking the callback list below will also modify the | |
1000 | * list. it removes the first item from the list and then runs the callback. | |
643c8c05 | 1001 | * The loop then restarts from the new first item int the list. This allows the |
c90821a2 DC |
1002 | * callback to scan and modify the list attached to the buffer and we don't |
1003 | * have to care about maintaining a next item pointer. | |
1004 | */ | |
1da177e4 LT |
1005 | STATIC void |
1006 | xfs_buf_do_callbacks( | |
c90821a2 | 1007 | struct xfs_buf *bp) |
1da177e4 | 1008 | { |
fb1755a6 | 1009 | struct xfs_buf_log_item *blip = bp->b_log_item; |
c90821a2 | 1010 | struct xfs_log_item *lip; |
1da177e4 | 1011 | |
fb1755a6 CM |
1012 | /* If there is a buf_log_item attached, run its callback */ |
1013 | if (blip) { | |
1014 | lip = &blip->bli_item; | |
1015 | lip->li_cb(bp, lip); | |
1016 | } | |
1017 | ||
643c8c05 CM |
1018 | while (!list_empty(&bp->b_li_list)) { |
1019 | lip = list_first_entry(&bp->b_li_list, struct xfs_log_item, | |
1020 | li_bio_list); | |
1021 | ||
1da177e4 | 1022 | /* |
643c8c05 | 1023 | * Remove the item from the list, so we don't have any |
1da177e4 LT |
1024 | * confusion if the item is added to another buf. |
1025 | * Don't touch the log item after calling its | |
1026 | * callback, because it could have freed itself. | |
1027 | */ | |
643c8c05 | 1028 | list_del_init(&lip->li_bio_list); |
1da177e4 | 1029 | lip->li_cb(bp, lip); |
1da177e4 LT |
1030 | } |
1031 | } | |
1032 | ||
0b80ae6e CM |
1033 | /* |
1034 | * Invoke the error state callback for each log item affected by the failed I/O. | |
1035 | * | |
1036 | * If a metadata buffer write fails with a non-permanent error, the buffer is | |
1037 | * eventually resubmitted and so the completion callbacks are not run. The error | |
1038 | * state may need to be propagated to the log items attached to the buffer, | |
1039 | * however, so the next AIL push of the item knows hot to handle it correctly. | |
1040 | */ | |
1041 | STATIC void | |
1042 | xfs_buf_do_callbacks_fail( | |
1043 | struct xfs_buf *bp) | |
1044 | { | |
643c8c05 | 1045 | struct xfs_log_item *lip; |
fb1755a6 | 1046 | struct xfs_ail *ailp; |
0b80ae6e | 1047 | |
fb1755a6 CM |
1048 | /* |
1049 | * Buffer log item errors are handled directly by xfs_buf_item_push() | |
1050 | * and xfs_buf_iodone_callback_error, and they have no IO error | |
1051 | * callbacks. Check only for items in b_li_list. | |
1052 | */ | |
643c8c05 | 1053 | if (list_empty(&bp->b_li_list)) |
fb1755a6 CM |
1054 | return; |
1055 | ||
643c8c05 CM |
1056 | lip = list_first_entry(&bp->b_li_list, struct xfs_log_item, |
1057 | li_bio_list); | |
fb1755a6 | 1058 | ailp = lip->li_ailp; |
57e80956 | 1059 | spin_lock(&ailp->ail_lock); |
643c8c05 | 1060 | list_for_each_entry(lip, &bp->b_li_list, li_bio_list) { |
0b80ae6e CM |
1061 | if (lip->li_ops->iop_error) |
1062 | lip->li_ops->iop_error(lip, bp); | |
1063 | } | |
57e80956 | 1064 | spin_unlock(&ailp->ail_lock); |
0b80ae6e CM |
1065 | } |
1066 | ||
df309390 CM |
1067 | static bool |
1068 | xfs_buf_iodone_callback_error( | |
bfc60177 | 1069 | struct xfs_buf *bp) |
1da177e4 | 1070 | { |
fb1755a6 | 1071 | struct xfs_buf_log_item *bip = bp->b_log_item; |
643c8c05 | 1072 | struct xfs_log_item *lip; |
fb1755a6 | 1073 | struct xfs_mount *mp; |
bfc60177 CH |
1074 | static ulong lasttime; |
1075 | static xfs_buftarg_t *lasttarg; | |
df309390 | 1076 | struct xfs_error_cfg *cfg; |
1da177e4 | 1077 | |
fb1755a6 CM |
1078 | /* |
1079 | * The failed buffer might not have a buf_log_item attached or the | |
1080 | * log_item list might be empty. Get the mp from the available | |
1081 | * xfs_log_item | |
1082 | */ | |
643c8c05 CM |
1083 | lip = list_first_entry_or_null(&bp->b_li_list, struct xfs_log_item, |
1084 | li_bio_list); | |
1085 | mp = lip ? lip->li_mountp : bip->bli_item.li_mountp; | |
fb1755a6 | 1086 | |
bfc60177 CH |
1087 | /* |
1088 | * If we've already decided to shutdown the filesystem because of | |
1089 | * I/O errors, there's no point in giving this a retry. | |
1090 | */ | |
df309390 CM |
1091 | if (XFS_FORCED_SHUTDOWN(mp)) |
1092 | goto out_stale; | |
1da177e4 | 1093 | |
49074c06 | 1094 | if (bp->b_target != lasttarg || |
bfc60177 CH |
1095 | time_after(jiffies, (lasttime + 5*HZ))) { |
1096 | lasttime = jiffies; | |
b38505b0 | 1097 | xfs_buf_ioerror_alert(bp, __func__); |
bfc60177 | 1098 | } |
49074c06 | 1099 | lasttarg = bp->b_target; |
1da177e4 | 1100 | |
df309390 CM |
1101 | /* synchronous writes will have callers process the error */ |
1102 | if (!(bp->b_flags & XBF_ASYNC)) | |
1103 | goto out_stale; | |
1104 | ||
1105 | trace_xfs_buf_item_iodone_async(bp, _RET_IP_); | |
1106 | ASSERT(bp->b_iodone != NULL); | |
1107 | ||
5539d367 ES |
1108 | cfg = xfs_error_get_cfg(mp, XFS_ERR_METADATA, bp->b_error); |
1109 | ||
bfc60177 | 1110 | /* |
25985edc | 1111 | * If the write was asynchronous then no one will be looking for the |
df309390 CM |
1112 | * error. If this is the first failure of this type, clear the error |
1113 | * state and write the buffer out again. This means we always retry an | |
1114 | * async write failure at least once, but we also need to set the buffer | |
1115 | * up to behave correctly now for repeated failures. | |
bfc60177 | 1116 | */ |
0b4db5df | 1117 | if (!(bp->b_flags & (XBF_STALE | XBF_WRITE_FAIL)) || |
df309390 | 1118 | bp->b_last_error != bp->b_error) { |
0b4db5df | 1119 | bp->b_flags |= (XBF_WRITE | XBF_DONE | XBF_WRITE_FAIL); |
df309390 | 1120 | bp->b_last_error = bp->b_error; |
77169812 ES |
1121 | if (cfg->retry_timeout != XFS_ERR_RETRY_FOREVER && |
1122 | !bp->b_first_retry_time) | |
5539d367 | 1123 | bp->b_first_retry_time = jiffies; |
a5ea70d2 | 1124 | |
df309390 CM |
1125 | xfs_buf_ioerror(bp, 0); |
1126 | xfs_buf_submit(bp); | |
1127 | return true; | |
1128 | } | |
43ff2122 | 1129 | |
df309390 CM |
1130 | /* |
1131 | * Repeated failure on an async write. Take action according to the | |
1132 | * error configuration we have been set up to use. | |
1133 | */ | |
a5ea70d2 CM |
1134 | |
1135 | if (cfg->max_retries != XFS_ERR_RETRY_FOREVER && | |
1136 | ++bp->b_retries > cfg->max_retries) | |
1137 | goto permanent_error; | |
77169812 | 1138 | if (cfg->retry_timeout != XFS_ERR_RETRY_FOREVER && |
a5ea70d2 CM |
1139 | time_after(jiffies, cfg->retry_timeout + bp->b_first_retry_time)) |
1140 | goto permanent_error; | |
bfc60177 | 1141 | |
e6b3bb78 CM |
1142 | /* At unmount we may treat errors differently */ |
1143 | if ((mp->m_flags & XFS_MOUNT_UNMOUNTING) && mp->m_fail_unmount) | |
1144 | goto permanent_error; | |
1145 | ||
0b80ae6e CM |
1146 | /* |
1147 | * Still a transient error, run IO completion failure callbacks and let | |
1148 | * the higher layers retry the buffer. | |
1149 | */ | |
1150 | xfs_buf_do_callbacks_fail(bp); | |
df309390 CM |
1151 | xfs_buf_ioerror(bp, 0); |
1152 | xfs_buf_relse(bp); | |
1153 | return true; | |
0b1b213f | 1154 | |
bfc60177 | 1155 | /* |
df309390 CM |
1156 | * Permanent error - we need to trigger a shutdown if we haven't already |
1157 | * to indicate that inconsistency will result from this action. | |
bfc60177 | 1158 | */ |
df309390 CM |
1159 | permanent_error: |
1160 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); | |
1161 | out_stale: | |
c867cb61 | 1162 | xfs_buf_stale(bp); |
b0388bf1 | 1163 | bp->b_flags |= XBF_DONE; |
0b1b213f | 1164 | trace_xfs_buf_error_relse(bp, _RET_IP_); |
df309390 CM |
1165 | return false; |
1166 | } | |
1167 | ||
1168 | /* | |
1169 | * This is the iodone() function for buffers which have had callbacks attached | |
1170 | * to them by xfs_buf_attach_iodone(). We need to iterate the items on the | |
1171 | * callback list, mark the buffer as having no more callbacks and then push the | |
1172 | * buffer through IO completion processing. | |
1173 | */ | |
1174 | void | |
1175 | xfs_buf_iodone_callbacks( | |
1176 | struct xfs_buf *bp) | |
1177 | { | |
1178 | /* | |
1179 | * If there is an error, process it. Some errors require us | |
1180 | * to run callbacks after failure processing is done so we | |
1181 | * detect that and take appropriate action. | |
1182 | */ | |
1183 | if (bp->b_error && xfs_buf_iodone_callback_error(bp)) | |
1184 | return; | |
1185 | ||
1186 | /* | |
1187 | * Successful IO or permanent error. Either way, we can clear the | |
1188 | * retry state here in preparation for the next error that may occur. | |
1189 | */ | |
1190 | bp->b_last_error = 0; | |
a5ea70d2 | 1191 | bp->b_retries = 0; |
4dd2eb63 | 1192 | bp->b_first_retry_time = 0; |
0b1b213f | 1193 | |
c90821a2 | 1194 | xfs_buf_do_callbacks(bp); |
fb1755a6 | 1195 | bp->b_log_item = NULL; |
643c8c05 | 1196 | list_del_init(&bp->b_li_list); |
cb669ca5 | 1197 | bp->b_iodone = NULL; |
e8aaba9a | 1198 | xfs_buf_ioend(bp); |
1da177e4 LT |
1199 | } |
1200 | ||
1da177e4 LT |
1201 | /* |
1202 | * This is the iodone() function for buffers which have been | |
1203 | * logged. It is called when they are eventually flushed out. | |
1204 | * It should remove the buf item from the AIL, and free the buf item. | |
1205 | * It is called by xfs_buf_iodone_callbacks() above which will take | |
1206 | * care of cleaning up the buffer itself. | |
1207 | */ | |
1da177e4 LT |
1208 | void |
1209 | xfs_buf_iodone( | |
ca30b2a7 CH |
1210 | struct xfs_buf *bp, |
1211 | struct xfs_log_item *lip) | |
1da177e4 | 1212 | { |
ca30b2a7 | 1213 | struct xfs_ail *ailp = lip->li_ailp; |
1da177e4 | 1214 | |
ca30b2a7 | 1215 | ASSERT(BUF_ITEM(lip)->bli_buf == bp); |
1da177e4 | 1216 | |
e1f5dbd7 | 1217 | xfs_buf_rele(bp); |
1da177e4 LT |
1218 | |
1219 | /* | |
1220 | * If we are forcibly shutting down, this may well be | |
1221 | * off the AIL already. That's because we simulate the | |
1222 | * log-committed callbacks to unpin these buffers. Or we may never | |
1223 | * have put this item on AIL because of the transaction was | |
783a2f65 | 1224 | * aborted forcibly. xfs_trans_ail_delete() takes care of these. |
1da177e4 LT |
1225 | * |
1226 | * Either way, AIL is useless if we're forcing a shutdown. | |
1227 | */ | |
57e80956 | 1228 | spin_lock(&ailp->ail_lock); |
04913fdd | 1229 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 | 1230 | xfs_buf_item_free(BUF_ITEM(lip)); |
1da177e4 | 1231 | } |
d3a304b6 CM |
1232 | |
1233 | /* | |
d43aaf16 | 1234 | * Requeue a failed buffer for writeback. |
d3a304b6 | 1235 | * |
d43aaf16 DC |
1236 | * We clear the log item failed state here as well, but we have to be careful |
1237 | * about reference counts because the only active reference counts on the buffer | |
1238 | * may be the failed log items. Hence if we clear the log item failed state | |
1239 | * before queuing the buffer for IO we can release all active references to | |
1240 | * the buffer and free it, leading to use after free problems in | |
1241 | * xfs_buf_delwri_queue. It makes no difference to the buffer or log items which | |
1242 | * order we process them in - the buffer is locked, and we own the buffer list | |
1243 | * so nothing on them is going to change while we are performing this action. | |
1244 | * | |
1245 | * Hence we can safely queue the buffer for IO before we clear the failed log | |
1246 | * item state, therefore always having an active reference to the buffer and | |
1247 | * avoiding the transient zero-reference state that leads to use-after-free. | |
1248 | * | |
1249 | * Return true if the buffer was added to the buffer list, false if it was | |
1250 | * already on the buffer list. | |
d3a304b6 CM |
1251 | */ |
1252 | bool | |
1253 | xfs_buf_resubmit_failed_buffers( | |
1254 | struct xfs_buf *bp, | |
d3a304b6 CM |
1255 | struct list_head *buffer_list) |
1256 | { | |
643c8c05 | 1257 | struct xfs_log_item *lip; |
d43aaf16 DC |
1258 | bool ret; |
1259 | ||
1260 | ret = xfs_buf_delwri_queue(bp, buffer_list); | |
d3a304b6 CM |
1261 | |
1262 | /* | |
d43aaf16 | 1263 | * XFS_LI_FAILED set/clear is protected by ail_lock, caller of this |
d3a304b6 CM |
1264 | * function already have it acquired |
1265 | */ | |
643c8c05 | 1266 | list_for_each_entry(lip, &bp->b_li_list, li_bio_list) |
d3a304b6 | 1267 | xfs_clear_li_failed(lip); |
d3a304b6 | 1268 | |
d43aaf16 | 1269 | return ret; |
d3a304b6 | 1270 | } |