Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
7b718769 NS |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. | |
1da177e4 | 4 | * |
7b718769 NS |
5 | * This program is free software; you can redistribute it and/or |
6 | * modify it under the terms of the GNU General Public License as | |
1da177e4 LT |
7 | * published by the Free Software Foundation. |
8 | * | |
7b718769 NS |
9 | * This program is distributed in the hope that it would be useful, |
10 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
11 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
12 | * GNU General Public License for more details. | |
1da177e4 | 13 | * |
7b718769 NS |
14 | * You should have received a copy of the GNU General Public License |
15 | * along with this program; if not, write the Free Software Foundation, | |
16 | * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
1da177e4 | 17 | */ |
1da177e4 | 18 | #include "xfs.h" |
a844f451 | 19 | #include "xfs_fs.h" |
1da177e4 | 20 | #include "xfs_types.h" |
a844f451 | 21 | #include "xfs_bit.h" |
1da177e4 LT |
22 | #include "xfs_log.h" |
23 | #include "xfs_trans.h" | |
1da177e4 | 24 | #include "xfs_sb.h" |
da353b0d | 25 | #include "xfs_ag.h" |
1da177e4 | 26 | #include "xfs_mount.h" |
a844f451 | 27 | #include "xfs_buf_item.h" |
1da177e4 | 28 | #include "xfs_trans_priv.h" |
1da177e4 | 29 | #include "xfs_error.h" |
0b1b213f | 30 | #include "xfs_trace.h" |
1da177e4 LT |
31 | |
32 | ||
33 | kmem_zone_t *xfs_buf_item_zone; | |
34 | ||
7bfa31d8 CH |
35 | static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip) |
36 | { | |
37 | return container_of(lip, struct xfs_buf_log_item, bli_item); | |
38 | } | |
39 | ||
c90821a2 | 40 | STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp); |
1da177e4 LT |
41 | |
42 | /* | |
43 | * This returns the number of log iovecs needed to log the | |
44 | * given buf log item. | |
45 | * | |
46 | * It calculates this as 1 iovec for the buf log format structure | |
47 | * and 1 for each stretch of non-contiguous chunks to be logged. | |
48 | * Contiguous chunks are logged in a single iovec. | |
49 | * | |
50 | * If the XFS_BLI_STALE flag has been set, then log nothing. | |
51 | */ | |
ba0f32d4 | 52 | STATIC uint |
372cc85e DC |
53 | xfs_buf_item_size_segment( |
54 | struct xfs_buf_log_item *bip, | |
55 | struct xfs_buf_log_format *blfp) | |
1da177e4 | 56 | { |
7bfa31d8 CH |
57 | struct xfs_buf *bp = bip->bli_buf; |
58 | uint nvecs; | |
59 | int next_bit; | |
60 | int last_bit; | |
1da177e4 | 61 | |
372cc85e DC |
62 | last_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); |
63 | if (last_bit == -1) | |
64 | return 0; | |
65 | ||
66 | /* | |
67 | * initial count for a dirty buffer is 2 vectors - the format structure | |
68 | * and the first dirty region. | |
69 | */ | |
70 | nvecs = 2; | |
1da177e4 | 71 | |
1da177e4 LT |
72 | while (last_bit != -1) { |
73 | /* | |
74 | * This takes the bit number to start looking from and | |
75 | * returns the next set bit from there. It returns -1 | |
76 | * if there are no more bits set or the start bit is | |
77 | * beyond the end of the bitmap. | |
78 | */ | |
372cc85e DC |
79 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
80 | last_bit + 1); | |
1da177e4 LT |
81 | /* |
82 | * If we run out of bits, leave the loop, | |
83 | * else if we find a new set of bits bump the number of vecs, | |
84 | * else keep scanning the current set of bits. | |
85 | */ | |
86 | if (next_bit == -1) { | |
372cc85e | 87 | break; |
1da177e4 LT |
88 | } else if (next_bit != last_bit + 1) { |
89 | last_bit = next_bit; | |
90 | nvecs++; | |
c1155410 DC |
91 | } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) != |
92 | (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) + | |
93 | XFS_BLF_CHUNK)) { | |
1da177e4 LT |
94 | last_bit = next_bit; |
95 | nvecs++; | |
96 | } else { | |
97 | last_bit++; | |
98 | } | |
99 | } | |
100 | ||
1da177e4 LT |
101 | return nvecs; |
102 | } | |
103 | ||
104 | /* | |
372cc85e DC |
105 | * This returns the number of log iovecs needed to log the given buf log item. |
106 | * | |
107 | * It calculates this as 1 iovec for the buf log format structure and 1 for each | |
108 | * stretch of non-contiguous chunks to be logged. Contiguous chunks are logged | |
109 | * in a single iovec. | |
110 | * | |
111 | * Discontiguous buffers need a format structure per region that that is being | |
112 | * logged. This makes the changes in the buffer appear to log recovery as though | |
113 | * they came from separate buffers, just like would occur if multiple buffers | |
114 | * were used instead of a single discontiguous buffer. This enables | |
115 | * discontiguous buffers to be in-memory constructs, completely transparent to | |
116 | * what ends up on disk. | |
117 | * | |
118 | * If the XFS_BLI_STALE flag has been set, then log nothing but the buf log | |
119 | * format structures. | |
1da177e4 | 120 | */ |
372cc85e DC |
121 | STATIC uint |
122 | xfs_buf_item_size( | |
123 | struct xfs_log_item *lip) | |
1da177e4 | 124 | { |
7bfa31d8 | 125 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
372cc85e DC |
126 | uint nvecs; |
127 | int i; | |
128 | ||
129 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
130 | if (bip->bli_flags & XFS_BLI_STALE) { | |
131 | /* | |
132 | * The buffer is stale, so all we need to log | |
133 | * is the buf log format structure with the | |
134 | * cancel flag in it. | |
135 | */ | |
136 | trace_xfs_buf_item_size_stale(bip); | |
b9438173 | 137 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
372cc85e DC |
138 | return bip->bli_format_count; |
139 | } | |
140 | ||
141 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); | |
142 | ||
143 | /* | |
144 | * the vector count is based on the number of buffer vectors we have | |
145 | * dirty bits in. This will only be greater than one when we have a | |
146 | * compound buffer with more than one segment dirty. Hence for compound | |
147 | * buffers we need to track which segment the dirty bits correspond to, | |
148 | * and when we move from one segment to the next increment the vector | |
149 | * count for the extra buf log format structure that will need to be | |
150 | * written. | |
151 | */ | |
152 | nvecs = 0; | |
153 | for (i = 0; i < bip->bli_format_count; i++) { | |
154 | nvecs += xfs_buf_item_size_segment(bip, &bip->bli_formats[i]); | |
155 | } | |
156 | ||
157 | trace_xfs_buf_item_size(bip); | |
158 | return nvecs; | |
159 | } | |
160 | ||
161 | static struct xfs_log_iovec * | |
162 | xfs_buf_item_format_segment( | |
163 | struct xfs_buf_log_item *bip, | |
164 | struct xfs_log_iovec *vecp, | |
165 | uint offset, | |
166 | struct xfs_buf_log_format *blfp) | |
167 | { | |
7bfa31d8 | 168 | struct xfs_buf *bp = bip->bli_buf; |
1da177e4 LT |
169 | uint base_size; |
170 | uint nvecs; | |
1da177e4 LT |
171 | int first_bit; |
172 | int last_bit; | |
173 | int next_bit; | |
174 | uint nbits; | |
175 | uint buffer_offset; | |
176 | ||
372cc85e | 177 | /* copy the flags across from the base format item */ |
b9438173 | 178 | blfp->blf_flags = bip->__bli_format.blf_flags; |
1da177e4 LT |
179 | |
180 | /* | |
77c1a08f DC |
181 | * Base size is the actual size of the ondisk structure - it reflects |
182 | * the actual size of the dirty bitmap rather than the size of the in | |
183 | * memory structure. | |
1da177e4 | 184 | */ |
77c1a08f | 185 | base_size = offsetof(struct xfs_buf_log_format, blf_data_map) + |
372cc85e | 186 | (blfp->blf_map_size * sizeof(blfp->blf_data_map[0])); |
820a554f MT |
187 | |
188 | nvecs = 0; | |
189 | first_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, 0); | |
190 | if (!(bip->bli_flags & XFS_BLI_STALE) && first_bit == -1) { | |
191 | /* | |
192 | * If the map is not be dirty in the transaction, mark | |
193 | * the size as zero and do not advance the vector pointer. | |
194 | */ | |
195 | goto out; | |
196 | } | |
197 | ||
372cc85e | 198 | vecp->i_addr = blfp; |
1da177e4 | 199 | vecp->i_len = base_size; |
4139b3b3 | 200 | vecp->i_type = XLOG_REG_TYPE_BFORMAT; |
1da177e4 LT |
201 | vecp++; |
202 | nvecs = 1; | |
203 | ||
204 | if (bip->bli_flags & XFS_BLI_STALE) { | |
205 | /* | |
206 | * The buffer is stale, so all we need to log | |
207 | * is the buf log format structure with the | |
208 | * cancel flag in it. | |
209 | */ | |
0b1b213f | 210 | trace_xfs_buf_item_format_stale(bip); |
372cc85e | 211 | ASSERT(blfp->blf_flags & XFS_BLF_CANCEL); |
820a554f | 212 | goto out; |
1da177e4 LT |
213 | } |
214 | ||
215 | /* | |
216 | * Fill in an iovec for each set of contiguous chunks. | |
217 | */ | |
820a554f | 218 | |
1da177e4 LT |
219 | last_bit = first_bit; |
220 | nbits = 1; | |
221 | for (;;) { | |
222 | /* | |
223 | * This takes the bit number to start looking from and | |
224 | * returns the next set bit from there. It returns -1 | |
225 | * if there are no more bits set or the start bit is | |
226 | * beyond the end of the bitmap. | |
227 | */ | |
372cc85e DC |
228 | next_bit = xfs_next_bit(blfp->blf_data_map, blfp->blf_map_size, |
229 | (uint)last_bit + 1); | |
1da177e4 LT |
230 | /* |
231 | * If we run out of bits fill in the last iovec and get | |
232 | * out of the loop. | |
233 | * Else if we start a new set of bits then fill in the | |
234 | * iovec for the series we were looking at and start | |
235 | * counting the bits in the new one. | |
236 | * Else we're still in the same set of bits so just | |
237 | * keep counting and scanning. | |
238 | */ | |
239 | if (next_bit == -1) { | |
372cc85e | 240 | buffer_offset = offset + first_bit * XFS_BLF_CHUNK; |
1da177e4 | 241 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
c1155410 | 242 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
4139b3b3 | 243 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
1da177e4 LT |
244 | nvecs++; |
245 | break; | |
246 | } else if (next_bit != last_bit + 1) { | |
372cc85e | 247 | buffer_offset = offset + first_bit * XFS_BLF_CHUNK; |
1da177e4 | 248 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
c1155410 | 249 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
4139b3b3 | 250 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
1da177e4 LT |
251 | nvecs++; |
252 | vecp++; | |
253 | first_bit = next_bit; | |
254 | last_bit = next_bit; | |
255 | nbits = 1; | |
372cc85e DC |
256 | } else if (xfs_buf_offset(bp, offset + |
257 | (next_bit << XFS_BLF_SHIFT)) != | |
258 | (xfs_buf_offset(bp, offset + | |
259 | (last_bit << XFS_BLF_SHIFT)) + | |
c1155410 | 260 | XFS_BLF_CHUNK)) { |
372cc85e | 261 | buffer_offset = offset + first_bit * XFS_BLF_CHUNK; |
1da177e4 | 262 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
c1155410 | 263 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
4139b3b3 | 264 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
709da6a6 | 265 | nvecs++; |
1da177e4 LT |
266 | vecp++; |
267 | first_bit = next_bit; | |
268 | last_bit = next_bit; | |
269 | nbits = 1; | |
270 | } else { | |
271 | last_bit++; | |
272 | nbits++; | |
273 | } | |
274 | } | |
820a554f MT |
275 | out: |
276 | blfp->blf_size = nvecs; | |
372cc85e DC |
277 | return vecp; |
278 | } | |
279 | ||
280 | /* | |
281 | * This is called to fill in the vector of log iovecs for the | |
282 | * given log buf item. It fills the first entry with a buf log | |
283 | * format structure, and the rest point to contiguous chunks | |
284 | * within the buffer. | |
285 | */ | |
286 | STATIC void | |
287 | xfs_buf_item_format( | |
288 | struct xfs_log_item *lip, | |
289 | struct xfs_log_iovec *vecp) | |
290 | { | |
291 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); | |
292 | struct xfs_buf *bp = bip->bli_buf; | |
293 | uint offset = 0; | |
294 | int i; | |
295 | ||
296 | ASSERT(atomic_read(&bip->bli_refcount) > 0); | |
297 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
298 | (bip->bli_flags & XFS_BLI_STALE)); | |
299 | ||
300 | /* | |
301 | * If it is an inode buffer, transfer the in-memory state to the | |
302 | * format flags and clear the in-memory state. We do not transfer | |
303 | * this state if the inode buffer allocation has not yet been committed | |
304 | * to the log as setting the XFS_BLI_INODE_BUF flag will prevent | |
305 | * correct replay of the inode allocation. | |
306 | */ | |
307 | if (bip->bli_flags & XFS_BLI_INODE_BUF) { | |
308 | if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && | |
309 | xfs_log_item_in_current_chkpt(lip))) | |
b9438173 | 310 | bip->__bli_format.blf_flags |= XFS_BLF_INODE_BUF; |
372cc85e DC |
311 | bip->bli_flags &= ~XFS_BLI_INODE_BUF; |
312 | } | |
313 | ||
314 | for (i = 0; i < bip->bli_format_count; i++) { | |
315 | vecp = xfs_buf_item_format_segment(bip, vecp, offset, | |
316 | &bip->bli_formats[i]); | |
317 | offset += bp->b_maps[i].bm_len; | |
318 | } | |
1da177e4 LT |
319 | |
320 | /* | |
321 | * Check to make sure everything is consistent. | |
322 | */ | |
0b1b213f | 323 | trace_xfs_buf_item_format(bip); |
1da177e4 LT |
324 | } |
325 | ||
326 | /* | |
64fc35de | 327 | * This is called to pin the buffer associated with the buf log item in memory |
4d16e924 | 328 | * so it cannot be written out. |
64fc35de DC |
329 | * |
330 | * We also always take a reference to the buffer log item here so that the bli | |
331 | * is held while the item is pinned in memory. This means that we can | |
332 | * unconditionally drop the reference count a transaction holds when the | |
333 | * transaction is completed. | |
1da177e4 | 334 | */ |
ba0f32d4 | 335 | STATIC void |
1da177e4 | 336 | xfs_buf_item_pin( |
7bfa31d8 | 337 | struct xfs_log_item *lip) |
1da177e4 | 338 | { |
7bfa31d8 | 339 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
1da177e4 | 340 | |
1da177e4 LT |
341 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
342 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || | |
343 | (bip->bli_flags & XFS_BLI_STALE)); | |
7bfa31d8 | 344 | |
0b1b213f | 345 | trace_xfs_buf_item_pin(bip); |
4d16e924 CH |
346 | |
347 | atomic_inc(&bip->bli_refcount); | |
348 | atomic_inc(&bip->bli_buf->b_pin_count); | |
1da177e4 LT |
349 | } |
350 | ||
1da177e4 LT |
351 | /* |
352 | * This is called to unpin the buffer associated with the buf log | |
353 | * item which was previously pinned with a call to xfs_buf_item_pin(). | |
1da177e4 LT |
354 | * |
355 | * Also drop the reference to the buf item for the current transaction. | |
356 | * If the XFS_BLI_STALE flag is set and we are the last reference, | |
357 | * then free up the buf log item and unlock the buffer. | |
9412e318 CH |
358 | * |
359 | * If the remove flag is set we are called from uncommit in the | |
360 | * forced-shutdown path. If that is true and the reference count on | |
361 | * the log item is going to drop to zero we need to free the item's | |
362 | * descriptor in the transaction. | |
1da177e4 | 363 | */ |
ba0f32d4 | 364 | STATIC void |
1da177e4 | 365 | xfs_buf_item_unpin( |
7bfa31d8 | 366 | struct xfs_log_item *lip, |
9412e318 | 367 | int remove) |
1da177e4 | 368 | { |
7bfa31d8 | 369 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
9412e318 | 370 | xfs_buf_t *bp = bip->bli_buf; |
7bfa31d8 | 371 | struct xfs_ail *ailp = lip->li_ailp; |
8e123850 | 372 | int stale = bip->bli_flags & XFS_BLI_STALE; |
7bfa31d8 | 373 | int freed; |
1da177e4 | 374 | |
adadbeef | 375 | ASSERT(bp->b_fspriv == bip); |
1da177e4 | 376 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
9412e318 | 377 | |
0b1b213f | 378 | trace_xfs_buf_item_unpin(bip); |
1da177e4 LT |
379 | |
380 | freed = atomic_dec_and_test(&bip->bli_refcount); | |
4d16e924 CH |
381 | |
382 | if (atomic_dec_and_test(&bp->b_pin_count)) | |
383 | wake_up_all(&bp->b_waiters); | |
7bfa31d8 | 384 | |
1da177e4 LT |
385 | if (freed && stale) { |
386 | ASSERT(bip->bli_flags & XFS_BLI_STALE); | |
0c842ad4 | 387 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 | 388 | ASSERT(XFS_BUF_ISSTALE(bp)); |
b9438173 | 389 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
9412e318 | 390 | |
0b1b213f CH |
391 | trace_xfs_buf_item_unpin_stale(bip); |
392 | ||
9412e318 CH |
393 | if (remove) { |
394 | /* | |
e34a314c DC |
395 | * If we are in a transaction context, we have to |
396 | * remove the log item from the transaction as we are | |
397 | * about to release our reference to the buffer. If we | |
398 | * don't, the unlock that occurs later in | |
399 | * xfs_trans_uncommit() will try to reference the | |
9412e318 CH |
400 | * buffer which we no longer have a hold on. |
401 | */ | |
e34a314c DC |
402 | if (lip->li_desc) |
403 | xfs_trans_del_item(lip); | |
9412e318 CH |
404 | |
405 | /* | |
406 | * Since the transaction no longer refers to the buffer, | |
407 | * the buffer should no longer refer to the transaction. | |
408 | */ | |
bf9d9013 | 409 | bp->b_transp = NULL; |
9412e318 CH |
410 | } |
411 | ||
1da177e4 LT |
412 | /* |
413 | * If we get called here because of an IO error, we may | |
783a2f65 | 414 | * or may not have the item on the AIL. xfs_trans_ail_delete() |
1da177e4 | 415 | * will take care of that situation. |
783a2f65 | 416 | * xfs_trans_ail_delete() drops the AIL lock. |
1da177e4 LT |
417 | */ |
418 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { | |
c90821a2 | 419 | xfs_buf_do_callbacks(bp); |
adadbeef | 420 | bp->b_fspriv = NULL; |
cb669ca5 | 421 | bp->b_iodone = NULL; |
1da177e4 | 422 | } else { |
783a2f65 | 423 | spin_lock(&ailp->xa_lock); |
04913fdd | 424 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_LOG_IO_ERROR); |
1da177e4 | 425 | xfs_buf_item_relse(bp); |
adadbeef | 426 | ASSERT(bp->b_fspriv == NULL); |
1da177e4 LT |
427 | } |
428 | xfs_buf_relse(bp); | |
960c60af | 429 | } else if (freed && remove) { |
137fff09 DC |
430 | /* |
431 | * There are currently two references to the buffer - the active | |
432 | * LRU reference and the buf log item. What we are about to do | |
433 | * here - simulate a failed IO completion - requires 3 | |
434 | * references. | |
435 | * | |
436 | * The LRU reference is removed by the xfs_buf_stale() call. The | |
437 | * buf item reference is removed by the xfs_buf_iodone() | |
438 | * callback that is run by xfs_buf_do_callbacks() during ioend | |
439 | * processing (via the bp->b_iodone callback), and then finally | |
440 | * the ioend processing will drop the IO reference if the buffer | |
441 | * is marked XBF_ASYNC. | |
442 | * | |
443 | * Hence we need to take an additional reference here so that IO | |
444 | * completion processing doesn't free the buffer prematurely. | |
445 | */ | |
960c60af | 446 | xfs_buf_lock(bp); |
137fff09 DC |
447 | xfs_buf_hold(bp); |
448 | bp->b_flags |= XBF_ASYNC; | |
960c60af CH |
449 | xfs_buf_ioerror(bp, EIO); |
450 | XFS_BUF_UNDONE(bp); | |
451 | xfs_buf_stale(bp); | |
452 | xfs_buf_ioend(bp, 0); | |
1da177e4 LT |
453 | } |
454 | } | |
455 | ||
ba0f32d4 | 456 | STATIC uint |
43ff2122 CH |
457 | xfs_buf_item_push( |
458 | struct xfs_log_item *lip, | |
459 | struct list_head *buffer_list) | |
1da177e4 | 460 | { |
7bfa31d8 CH |
461 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
462 | struct xfs_buf *bp = bip->bli_buf; | |
43ff2122 | 463 | uint rval = XFS_ITEM_SUCCESS; |
1da177e4 | 464 | |
811e64c7 | 465 | if (xfs_buf_ispinned(bp)) |
1da177e4 | 466 | return XFS_ITEM_PINNED; |
5337fe9b BF |
467 | if (!xfs_buf_trylock(bp)) { |
468 | /* | |
469 | * If we have just raced with a buffer being pinned and it has | |
470 | * been marked stale, we could end up stalling until someone else | |
471 | * issues a log force to unpin the stale buffer. Check for the | |
472 | * race condition here so xfsaild recognizes the buffer is pinned | |
473 | * and queues a log force to move it along. | |
474 | */ | |
475 | if (xfs_buf_ispinned(bp)) | |
476 | return XFS_ITEM_PINNED; | |
1da177e4 | 477 | return XFS_ITEM_LOCKED; |
5337fe9b | 478 | } |
1da177e4 | 479 | |
1da177e4 | 480 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
43ff2122 CH |
481 | |
482 | trace_xfs_buf_item_push(bip); | |
483 | ||
484 | if (!xfs_buf_delwri_queue(bp, buffer_list)) | |
485 | rval = XFS_ITEM_FLUSHING; | |
486 | xfs_buf_unlock(bp); | |
487 | return rval; | |
1da177e4 LT |
488 | } |
489 | ||
490 | /* | |
64fc35de DC |
491 | * Release the buffer associated with the buf log item. If there is no dirty |
492 | * logged data associated with the buffer recorded in the buf log item, then | |
493 | * free the buf log item and remove the reference to it in the buffer. | |
1da177e4 | 494 | * |
64fc35de DC |
495 | * This call ignores the recursion count. It is only called when the buffer |
496 | * should REALLY be unlocked, regardless of the recursion count. | |
1da177e4 | 497 | * |
64fc35de DC |
498 | * We unconditionally drop the transaction's reference to the log item. If the |
499 | * item was logged, then another reference was taken when it was pinned, so we | |
500 | * can safely drop the transaction reference now. This also allows us to avoid | |
501 | * potential races with the unpin code freeing the bli by not referencing the | |
502 | * bli after we've dropped the reference count. | |
503 | * | |
504 | * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item | |
505 | * if necessary but do not unlock the buffer. This is for support of | |
506 | * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't | |
507 | * free the item. | |
1da177e4 | 508 | */ |
ba0f32d4 | 509 | STATIC void |
1da177e4 | 510 | xfs_buf_item_unlock( |
7bfa31d8 | 511 | struct xfs_log_item *lip) |
1da177e4 | 512 | { |
7bfa31d8 CH |
513 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
514 | struct xfs_buf *bp = bip->bli_buf; | |
c883d0c4 | 515 | int aborted, clean, i; |
7bfa31d8 | 516 | uint hold; |
1da177e4 | 517 | |
64fc35de | 518 | /* Clear the buffer's association with this transaction. */ |
bf9d9013 | 519 | bp->b_transp = NULL; |
1da177e4 LT |
520 | |
521 | /* | |
64fc35de DC |
522 | * If this is a transaction abort, don't return early. Instead, allow |
523 | * the brelse to happen. Normally it would be done for stale | |
524 | * (cancelled) buffers at unpin time, but we'll never go through the | |
525 | * pin/unpin cycle if we abort inside commit. | |
1da177e4 | 526 | */ |
7bfa31d8 | 527 | aborted = (lip->li_flags & XFS_LI_ABORTED) != 0; |
1da177e4 LT |
528 | |
529 | /* | |
64fc35de DC |
530 | * Before possibly freeing the buf item, determine if we should |
531 | * release the buffer at the end of this routine. | |
532 | */ | |
533 | hold = bip->bli_flags & XFS_BLI_HOLD; | |
534 | ||
535 | /* Clear the per transaction state. */ | |
536 | bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD); | |
537 | ||
538 | /* | |
539 | * If the buf item is marked stale, then don't do anything. We'll | |
540 | * unlock the buffer and free the buf item when the buffer is unpinned | |
541 | * for the last time. | |
1da177e4 LT |
542 | */ |
543 | if (bip->bli_flags & XFS_BLI_STALE) { | |
0b1b213f | 544 | trace_xfs_buf_item_unlock_stale(bip); |
b9438173 | 545 | ASSERT(bip->__bli_format.blf_flags & XFS_BLF_CANCEL); |
64fc35de DC |
546 | if (!aborted) { |
547 | atomic_dec(&bip->bli_refcount); | |
1da177e4 | 548 | return; |
64fc35de | 549 | } |
1da177e4 LT |
550 | } |
551 | ||
0b1b213f | 552 | trace_xfs_buf_item_unlock(bip); |
1da177e4 LT |
553 | |
554 | /* | |
64fc35de | 555 | * If the buf item isn't tracking any data, free it, otherwise drop the |
3b19034d DC |
556 | * reference we hold to it. If we are aborting the transaction, this may |
557 | * be the only reference to the buf item, so we free it anyway | |
558 | * regardless of whether it is dirty or not. A dirty abort implies a | |
559 | * shutdown, anyway. | |
1da177e4 | 560 | */ |
c883d0c4 MT |
561 | clean = 1; |
562 | for (i = 0; i < bip->bli_format_count; i++) { | |
563 | if (!xfs_bitmap_empty(bip->bli_formats[i].blf_data_map, | |
564 | bip->bli_formats[i].blf_map_size)) { | |
565 | clean = 0; | |
566 | break; | |
567 | } | |
568 | } | |
569 | if (clean) | |
1da177e4 | 570 | xfs_buf_item_relse(bp); |
3b19034d DC |
571 | else if (aborted) { |
572 | if (atomic_dec_and_test(&bip->bli_refcount)) { | |
573 | ASSERT(XFS_FORCED_SHUTDOWN(lip->li_mountp)); | |
574 | xfs_buf_item_relse(bp); | |
575 | } | |
576 | } else | |
64fc35de | 577 | atomic_dec(&bip->bli_refcount); |
1da177e4 | 578 | |
64fc35de | 579 | if (!hold) |
1da177e4 | 580 | xfs_buf_relse(bp); |
1da177e4 LT |
581 | } |
582 | ||
583 | /* | |
584 | * This is called to find out where the oldest active copy of the | |
585 | * buf log item in the on disk log resides now that the last log | |
586 | * write of it completed at the given lsn. | |
587 | * We always re-log all the dirty data in a buffer, so usually the | |
588 | * latest copy in the on disk log is the only one that matters. For | |
589 | * those cases we simply return the given lsn. | |
590 | * | |
591 | * The one exception to this is for buffers full of newly allocated | |
592 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF | |
593 | * flag set, indicating that only the di_next_unlinked fields from the | |
594 | * inodes in the buffers will be replayed during recovery. If the | |
595 | * original newly allocated inode images have not yet been flushed | |
596 | * when the buffer is so relogged, then we need to make sure that we | |
597 | * keep the old images in the 'active' portion of the log. We do this | |
598 | * by returning the original lsn of that transaction here rather than | |
599 | * the current one. | |
600 | */ | |
ba0f32d4 | 601 | STATIC xfs_lsn_t |
1da177e4 | 602 | xfs_buf_item_committed( |
7bfa31d8 | 603 | struct xfs_log_item *lip, |
1da177e4 LT |
604 | xfs_lsn_t lsn) |
605 | { | |
7bfa31d8 CH |
606 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
607 | ||
0b1b213f CH |
608 | trace_xfs_buf_item_committed(bip); |
609 | ||
7bfa31d8 CH |
610 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0) |
611 | return lip->li_lsn; | |
612 | return lsn; | |
1da177e4 LT |
613 | } |
614 | ||
ba0f32d4 | 615 | STATIC void |
7bfa31d8 CH |
616 | xfs_buf_item_committing( |
617 | struct xfs_log_item *lip, | |
618 | xfs_lsn_t commit_lsn) | |
1da177e4 LT |
619 | { |
620 | } | |
621 | ||
622 | /* | |
623 | * This is the ops vector shared by all buf log items. | |
624 | */ | |
272e42b2 | 625 | static const struct xfs_item_ops xfs_buf_item_ops = { |
7bfa31d8 CH |
626 | .iop_size = xfs_buf_item_size, |
627 | .iop_format = xfs_buf_item_format, | |
628 | .iop_pin = xfs_buf_item_pin, | |
629 | .iop_unpin = xfs_buf_item_unpin, | |
7bfa31d8 CH |
630 | .iop_unlock = xfs_buf_item_unlock, |
631 | .iop_committed = xfs_buf_item_committed, | |
632 | .iop_push = xfs_buf_item_push, | |
7bfa31d8 | 633 | .iop_committing = xfs_buf_item_committing |
1da177e4 LT |
634 | }; |
635 | ||
372cc85e DC |
636 | STATIC int |
637 | xfs_buf_item_get_format( | |
638 | struct xfs_buf_log_item *bip, | |
639 | int count) | |
640 | { | |
641 | ASSERT(bip->bli_formats == NULL); | |
642 | bip->bli_format_count = count; | |
643 | ||
644 | if (count == 1) { | |
b9438173 | 645 | bip->bli_formats = &bip->__bli_format; |
372cc85e DC |
646 | return 0; |
647 | } | |
648 | ||
649 | bip->bli_formats = kmem_zalloc(count * sizeof(struct xfs_buf_log_format), | |
650 | KM_SLEEP); | |
651 | if (!bip->bli_formats) | |
652 | return ENOMEM; | |
653 | return 0; | |
654 | } | |
655 | ||
656 | STATIC void | |
657 | xfs_buf_item_free_format( | |
658 | struct xfs_buf_log_item *bip) | |
659 | { | |
b9438173 | 660 | if (bip->bli_formats != &bip->__bli_format) { |
372cc85e DC |
661 | kmem_free(bip->bli_formats); |
662 | bip->bli_formats = NULL; | |
663 | } | |
664 | } | |
1da177e4 LT |
665 | |
666 | /* | |
667 | * Allocate a new buf log item to go with the given buffer. | |
668 | * Set the buffer's b_fsprivate field to point to the new | |
669 | * buf log item. If there are other item's attached to the | |
670 | * buffer (see xfs_buf_attach_iodone() below), then put the | |
671 | * buf log item at the front. | |
672 | */ | |
673 | void | |
674 | xfs_buf_item_init( | |
675 | xfs_buf_t *bp, | |
676 | xfs_mount_t *mp) | |
677 | { | |
adadbeef | 678 | xfs_log_item_t *lip = bp->b_fspriv; |
1da177e4 LT |
679 | xfs_buf_log_item_t *bip; |
680 | int chunks; | |
681 | int map_size; | |
372cc85e DC |
682 | int error; |
683 | int i; | |
1da177e4 LT |
684 | |
685 | /* | |
686 | * Check to see if there is already a buf log item for | |
687 | * this buffer. If there is, it is guaranteed to be | |
688 | * the first. If we do already have one, there is | |
689 | * nothing to do here so return. | |
690 | */ | |
ebad861b | 691 | ASSERT(bp->b_target->bt_mount == mp); |
adadbeef CH |
692 | if (lip != NULL && lip->li_type == XFS_LI_BUF) |
693 | return; | |
1da177e4 | 694 | |
372cc85e | 695 | bip = kmem_zone_zalloc(xfs_buf_item_zone, KM_SLEEP); |
43f5efc5 | 696 | xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops); |
1da177e4 | 697 | bip->bli_buf = bp; |
e1f5dbd7 | 698 | xfs_buf_hold(bp); |
372cc85e DC |
699 | |
700 | /* | |
701 | * chunks is the number of XFS_BLF_CHUNK size pieces the buffer | |
702 | * can be divided into. Make sure not to truncate any pieces. | |
703 | * map_size is the size of the bitmap needed to describe the | |
704 | * chunks of the buffer. | |
705 | * | |
706 | * Discontiguous buffer support follows the layout of the underlying | |
707 | * buffer. This makes the implementation as simple as possible. | |
708 | */ | |
709 | error = xfs_buf_item_get_format(bip, bp->b_map_count); | |
710 | ASSERT(error == 0); | |
711 | ||
712 | for (i = 0; i < bip->bli_format_count; i++) { | |
713 | chunks = DIV_ROUND_UP(BBTOB(bp->b_maps[i].bm_len), | |
714 | XFS_BLF_CHUNK); | |
715 | map_size = DIV_ROUND_UP(chunks, NBWORD); | |
716 | ||
717 | bip->bli_formats[i].blf_type = XFS_LI_BUF; | |
718 | bip->bli_formats[i].blf_blkno = bp->b_maps[i].bm_bn; | |
719 | bip->bli_formats[i].blf_len = bp->b_maps[i].bm_len; | |
720 | bip->bli_formats[i].blf_map_size = map_size; | |
721 | } | |
1da177e4 LT |
722 | |
723 | #ifdef XFS_TRANS_DEBUG | |
724 | /* | |
725 | * Allocate the arrays for tracking what needs to be logged | |
726 | * and what our callers request to be logged. bli_orig | |
727 | * holds a copy of the original, clean buffer for comparison | |
728 | * against, and bli_logged keeps a 1 bit flag per byte in | |
729 | * the buffer to indicate which bytes the callers have asked | |
730 | * to have logged. | |
731 | */ | |
aa0e8833 DC |
732 | bip->bli_orig = kmem_alloc(BBTOB(bp->b_length), KM_SLEEP); |
733 | memcpy(bip->bli_orig, bp->b_addr, BBTOB(bp->b_length)); | |
734 | bip->bli_logged = kmem_zalloc(BBTOB(bp->b_length) / NBBY, KM_SLEEP); | |
1da177e4 LT |
735 | #endif |
736 | ||
737 | /* | |
738 | * Put the buf item into the list of items attached to the | |
739 | * buffer at the front. | |
740 | */ | |
adadbeef CH |
741 | if (bp->b_fspriv) |
742 | bip->bli_item.li_bio_list = bp->b_fspriv; | |
743 | bp->b_fspriv = bip; | |
1da177e4 LT |
744 | } |
745 | ||
746 | ||
747 | /* | |
748 | * Mark bytes first through last inclusive as dirty in the buf | |
749 | * item's bitmap. | |
750 | */ | |
751 | void | |
372cc85e DC |
752 | xfs_buf_item_log_segment( |
753 | struct xfs_buf_log_item *bip, | |
1da177e4 | 754 | uint first, |
372cc85e DC |
755 | uint last, |
756 | uint *map) | |
1da177e4 LT |
757 | { |
758 | uint first_bit; | |
759 | uint last_bit; | |
760 | uint bits_to_set; | |
761 | uint bits_set; | |
762 | uint word_num; | |
763 | uint *wordp; | |
764 | uint bit; | |
765 | uint end_bit; | |
766 | uint mask; | |
767 | ||
1da177e4 LT |
768 | /* |
769 | * Convert byte offsets to bit numbers. | |
770 | */ | |
c1155410 DC |
771 | first_bit = first >> XFS_BLF_SHIFT; |
772 | last_bit = last >> XFS_BLF_SHIFT; | |
1da177e4 LT |
773 | |
774 | /* | |
775 | * Calculate the total number of bits to be set. | |
776 | */ | |
777 | bits_to_set = last_bit - first_bit + 1; | |
778 | ||
779 | /* | |
780 | * Get a pointer to the first word in the bitmap | |
781 | * to set a bit in. | |
782 | */ | |
783 | word_num = first_bit >> BIT_TO_WORD_SHIFT; | |
372cc85e | 784 | wordp = &map[word_num]; |
1da177e4 LT |
785 | |
786 | /* | |
787 | * Calculate the starting bit in the first word. | |
788 | */ | |
789 | bit = first_bit & (uint)(NBWORD - 1); | |
790 | ||
791 | /* | |
792 | * First set any bits in the first word of our range. | |
793 | * If it starts at bit 0 of the word, it will be | |
794 | * set below rather than here. That is what the variable | |
795 | * bit tells us. The variable bits_set tracks the number | |
796 | * of bits that have been set so far. End_bit is the number | |
797 | * of the last bit to be set in this word plus one. | |
798 | */ | |
799 | if (bit) { | |
800 | end_bit = MIN(bit + bits_to_set, (uint)NBWORD); | |
801 | mask = ((1 << (end_bit - bit)) - 1) << bit; | |
802 | *wordp |= mask; | |
803 | wordp++; | |
804 | bits_set = end_bit - bit; | |
805 | } else { | |
806 | bits_set = 0; | |
807 | } | |
808 | ||
809 | /* | |
810 | * Now set bits a whole word at a time that are between | |
811 | * first_bit and last_bit. | |
812 | */ | |
813 | while ((bits_to_set - bits_set) >= NBWORD) { | |
814 | *wordp |= 0xffffffff; | |
815 | bits_set += NBWORD; | |
816 | wordp++; | |
817 | } | |
818 | ||
819 | /* | |
820 | * Finally, set any bits left to be set in one last partial word. | |
821 | */ | |
822 | end_bit = bits_to_set - bits_set; | |
823 | if (end_bit) { | |
824 | mask = (1 << end_bit) - 1; | |
825 | *wordp |= mask; | |
826 | } | |
1da177e4 LT |
827 | } |
828 | ||
372cc85e DC |
829 | /* |
830 | * Mark bytes first through last inclusive as dirty in the buf | |
831 | * item's bitmap. | |
832 | */ | |
833 | void | |
834 | xfs_buf_item_log( | |
835 | xfs_buf_log_item_t *bip, | |
836 | uint first, | |
837 | uint last) | |
838 | { | |
839 | int i; | |
840 | uint start; | |
841 | uint end; | |
842 | struct xfs_buf *bp = bip->bli_buf; | |
843 | ||
844 | /* | |
845 | * Mark the item as having some dirty data for | |
846 | * quick reference in xfs_buf_item_dirty. | |
847 | */ | |
848 | bip->bli_flags |= XFS_BLI_DIRTY; | |
849 | ||
850 | /* | |
851 | * walk each buffer segment and mark them dirty appropriately. | |
852 | */ | |
853 | start = 0; | |
854 | for (i = 0; i < bip->bli_format_count; i++) { | |
855 | if (start > last) | |
856 | break; | |
857 | end = start + BBTOB(bp->b_maps[i].bm_len); | |
858 | if (first > end) { | |
859 | start += BBTOB(bp->b_maps[i].bm_len); | |
860 | continue; | |
861 | } | |
862 | if (first < start) | |
863 | first = start; | |
864 | if (end > last) | |
865 | end = last; | |
866 | ||
867 | xfs_buf_item_log_segment(bip, first, end, | |
868 | &bip->bli_formats[i].blf_data_map[0]); | |
869 | ||
870 | start += bp->b_maps[i].bm_len; | |
871 | } | |
872 | } | |
873 | ||
1da177e4 LT |
874 | |
875 | /* | |
876 | * Return 1 if the buffer has some data that has been logged (at any | |
877 | * point, not just the current transaction) and 0 if not. | |
878 | */ | |
879 | uint | |
880 | xfs_buf_item_dirty( | |
881 | xfs_buf_log_item_t *bip) | |
882 | { | |
883 | return (bip->bli_flags & XFS_BLI_DIRTY); | |
884 | } | |
885 | ||
e1f5dbd7 LM |
886 | STATIC void |
887 | xfs_buf_item_free( | |
888 | xfs_buf_log_item_t *bip) | |
889 | { | |
890 | #ifdef XFS_TRANS_DEBUG | |
891 | kmem_free(bip->bli_orig); | |
892 | kmem_free(bip->bli_logged); | |
893 | #endif /* XFS_TRANS_DEBUG */ | |
894 | ||
372cc85e | 895 | xfs_buf_item_free_format(bip); |
e1f5dbd7 LM |
896 | kmem_zone_free(xfs_buf_item_zone, bip); |
897 | } | |
898 | ||
1da177e4 LT |
899 | /* |
900 | * This is called when the buf log item is no longer needed. It should | |
901 | * free the buf log item associated with the given buffer and clear | |
902 | * the buffer's pointer to the buf log item. If there are no more | |
903 | * items in the list, clear the b_iodone field of the buffer (see | |
904 | * xfs_buf_attach_iodone() below). | |
905 | */ | |
906 | void | |
907 | xfs_buf_item_relse( | |
908 | xfs_buf_t *bp) | |
909 | { | |
910 | xfs_buf_log_item_t *bip; | |
911 | ||
0b1b213f CH |
912 | trace_xfs_buf_item_relse(bp, _RET_IP_); |
913 | ||
adadbeef CH |
914 | bip = bp->b_fspriv; |
915 | bp->b_fspriv = bip->bli_item.li_bio_list; | |
cb669ca5 CH |
916 | if (bp->b_fspriv == NULL) |
917 | bp->b_iodone = NULL; | |
adadbeef | 918 | |
e1f5dbd7 LM |
919 | xfs_buf_rele(bp); |
920 | xfs_buf_item_free(bip); | |
1da177e4 LT |
921 | } |
922 | ||
923 | ||
924 | /* | |
925 | * Add the given log item with its callback to the list of callbacks | |
926 | * to be called when the buffer's I/O completes. If it is not set | |
927 | * already, set the buffer's b_iodone() routine to be | |
928 | * xfs_buf_iodone_callbacks() and link the log item into the list of | |
929 | * items rooted at b_fsprivate. Items are always added as the second | |
930 | * entry in the list if there is a first, because the buf item code | |
931 | * assumes that the buf log item is first. | |
932 | */ | |
933 | void | |
934 | xfs_buf_attach_iodone( | |
935 | xfs_buf_t *bp, | |
936 | void (*cb)(xfs_buf_t *, xfs_log_item_t *), | |
937 | xfs_log_item_t *lip) | |
938 | { | |
939 | xfs_log_item_t *head_lip; | |
940 | ||
0c842ad4 | 941 | ASSERT(xfs_buf_islocked(bp)); |
1da177e4 LT |
942 | |
943 | lip->li_cb = cb; | |
adadbeef CH |
944 | head_lip = bp->b_fspriv; |
945 | if (head_lip) { | |
1da177e4 LT |
946 | lip->li_bio_list = head_lip->li_bio_list; |
947 | head_lip->li_bio_list = lip; | |
948 | } else { | |
adadbeef | 949 | bp->b_fspriv = lip; |
1da177e4 LT |
950 | } |
951 | ||
cb669ca5 CH |
952 | ASSERT(bp->b_iodone == NULL || |
953 | bp->b_iodone == xfs_buf_iodone_callbacks); | |
954 | bp->b_iodone = xfs_buf_iodone_callbacks; | |
1da177e4 LT |
955 | } |
956 | ||
c90821a2 DC |
957 | /* |
958 | * We can have many callbacks on a buffer. Running the callbacks individually | |
959 | * can cause a lot of contention on the AIL lock, so we allow for a single | |
960 | * callback to be able to scan the remaining lip->li_bio_list for other items | |
961 | * of the same type and callback to be processed in the first call. | |
962 | * | |
963 | * As a result, the loop walking the callback list below will also modify the | |
964 | * list. it removes the first item from the list and then runs the callback. | |
965 | * The loop then restarts from the new head of the list. This allows the | |
966 | * callback to scan and modify the list attached to the buffer and we don't | |
967 | * have to care about maintaining a next item pointer. | |
968 | */ | |
1da177e4 LT |
969 | STATIC void |
970 | xfs_buf_do_callbacks( | |
c90821a2 | 971 | struct xfs_buf *bp) |
1da177e4 | 972 | { |
c90821a2 | 973 | struct xfs_log_item *lip; |
1da177e4 | 974 | |
adadbeef CH |
975 | while ((lip = bp->b_fspriv) != NULL) { |
976 | bp->b_fspriv = lip->li_bio_list; | |
1da177e4 LT |
977 | ASSERT(lip->li_cb != NULL); |
978 | /* | |
979 | * Clear the next pointer so we don't have any | |
980 | * confusion if the item is added to another buf. | |
981 | * Don't touch the log item after calling its | |
982 | * callback, because it could have freed itself. | |
983 | */ | |
984 | lip->li_bio_list = NULL; | |
985 | lip->li_cb(bp, lip); | |
1da177e4 LT |
986 | } |
987 | } | |
988 | ||
989 | /* | |
990 | * This is the iodone() function for buffers which have had callbacks | |
991 | * attached to them by xfs_buf_attach_iodone(). It should remove each | |
992 | * log item from the buffer's list and call the callback of each in turn. | |
993 | * When done, the buffer's fsprivate field is set to NULL and the buffer | |
994 | * is unlocked with a call to iodone(). | |
995 | */ | |
996 | void | |
997 | xfs_buf_iodone_callbacks( | |
bfc60177 | 998 | struct xfs_buf *bp) |
1da177e4 | 999 | { |
bfc60177 CH |
1000 | struct xfs_log_item *lip = bp->b_fspriv; |
1001 | struct xfs_mount *mp = lip->li_mountp; | |
1002 | static ulong lasttime; | |
1003 | static xfs_buftarg_t *lasttarg; | |
1da177e4 | 1004 | |
5a52c2a5 | 1005 | if (likely(!xfs_buf_geterror(bp))) |
bfc60177 | 1006 | goto do_callbacks; |
1da177e4 | 1007 | |
bfc60177 CH |
1008 | /* |
1009 | * If we've already decided to shutdown the filesystem because of | |
1010 | * I/O errors, there's no point in giving this a retry. | |
1011 | */ | |
1012 | if (XFS_FORCED_SHUTDOWN(mp)) { | |
c867cb61 | 1013 | xfs_buf_stale(bp); |
c867cb61 | 1014 | XFS_BUF_DONE(bp); |
bfc60177 CH |
1015 | trace_xfs_buf_item_iodone(bp, _RET_IP_); |
1016 | goto do_callbacks; | |
1017 | } | |
1da177e4 | 1018 | |
49074c06 | 1019 | if (bp->b_target != lasttarg || |
bfc60177 CH |
1020 | time_after(jiffies, (lasttime + 5*HZ))) { |
1021 | lasttime = jiffies; | |
b38505b0 | 1022 | xfs_buf_ioerror_alert(bp, __func__); |
bfc60177 | 1023 | } |
49074c06 | 1024 | lasttarg = bp->b_target; |
1da177e4 | 1025 | |
bfc60177 | 1026 | /* |
25985edc | 1027 | * If the write was asynchronous then no one will be looking for the |
bfc60177 CH |
1028 | * error. Clear the error state and write the buffer out again. |
1029 | * | |
43ff2122 CH |
1030 | * XXX: This helps against transient write errors, but we need to find |
1031 | * a way to shut the filesystem down if the writes keep failing. | |
1032 | * | |
1033 | * In practice we'll shut the filesystem down soon as non-transient | |
1034 | * erorrs tend to affect the whole device and a failing log write | |
1035 | * will make us give up. But we really ought to do better here. | |
bfc60177 CH |
1036 | */ |
1037 | if (XFS_BUF_ISASYNC(bp)) { | |
43ff2122 CH |
1038 | ASSERT(bp->b_iodone != NULL); |
1039 | ||
1040 | trace_xfs_buf_item_iodone_async(bp, _RET_IP_); | |
1041 | ||
5a52c2a5 | 1042 | xfs_buf_ioerror(bp, 0); /* errno of 0 unsets the flag */ |
bfc60177 CH |
1043 | |
1044 | if (!XFS_BUF_ISSTALE(bp)) { | |
43ff2122 | 1045 | bp->b_flags |= XBF_WRITE | XBF_ASYNC | XBF_DONE; |
a2dcf5df | 1046 | xfs_buf_iorequest(bp); |
43ff2122 CH |
1047 | } else { |
1048 | xfs_buf_relse(bp); | |
1da177e4 | 1049 | } |
43ff2122 | 1050 | |
1da177e4 LT |
1051 | return; |
1052 | } | |
0b1b213f | 1053 | |
bfc60177 CH |
1054 | /* |
1055 | * If the write of the buffer was synchronous, we want to make | |
1056 | * sure to return the error to the caller of xfs_bwrite(). | |
1057 | */ | |
c867cb61 | 1058 | xfs_buf_stale(bp); |
1da177e4 | 1059 | XFS_BUF_DONE(bp); |
0b1b213f CH |
1060 | |
1061 | trace_xfs_buf_error_relse(bp, _RET_IP_); | |
1062 | ||
bfc60177 | 1063 | do_callbacks: |
c90821a2 | 1064 | xfs_buf_do_callbacks(bp); |
adadbeef | 1065 | bp->b_fspriv = NULL; |
cb669ca5 | 1066 | bp->b_iodone = NULL; |
bfc60177 | 1067 | xfs_buf_ioend(bp, 0); |
1da177e4 LT |
1068 | } |
1069 | ||
1da177e4 LT |
1070 | /* |
1071 | * This is the iodone() function for buffers which have been | |
1072 | * logged. It is called when they are eventually flushed out. | |
1073 | * It should remove the buf item from the AIL, and free the buf item. | |
1074 | * It is called by xfs_buf_iodone_callbacks() above which will take | |
1075 | * care of cleaning up the buffer itself. | |
1076 | */ | |
1da177e4 LT |
1077 | void |
1078 | xfs_buf_iodone( | |
ca30b2a7 CH |
1079 | struct xfs_buf *bp, |
1080 | struct xfs_log_item *lip) | |
1da177e4 | 1081 | { |
ca30b2a7 | 1082 | struct xfs_ail *ailp = lip->li_ailp; |
1da177e4 | 1083 | |
ca30b2a7 | 1084 | ASSERT(BUF_ITEM(lip)->bli_buf == bp); |
1da177e4 | 1085 | |
e1f5dbd7 | 1086 | xfs_buf_rele(bp); |
1da177e4 LT |
1087 | |
1088 | /* | |
1089 | * If we are forcibly shutting down, this may well be | |
1090 | * off the AIL already. That's because we simulate the | |
1091 | * log-committed callbacks to unpin these buffers. Or we may never | |
1092 | * have put this item on AIL because of the transaction was | |
783a2f65 | 1093 | * aborted forcibly. xfs_trans_ail_delete() takes care of these. |
1da177e4 LT |
1094 | * |
1095 | * Either way, AIL is useless if we're forcing a shutdown. | |
1096 | */ | |
fc1829f3 | 1097 | spin_lock(&ailp->xa_lock); |
04913fdd | 1098 | xfs_trans_ail_delete(ailp, lip, SHUTDOWN_CORRUPT_INCORE); |
ca30b2a7 | 1099 | xfs_buf_item_free(BUF_ITEM(lip)); |
1da177e4 | 1100 | } |