Root/
1 | /* |
2 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
3 | * All Rights Reserved. |
4 | * |
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 |
7 | * published by the Free Software Foundation. |
8 | * |
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. |
13 | * |
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 |
17 | */ |
18 | #include "xfs.h" |
19 | #include "xfs_fs.h" |
20 | #include "xfs_types.h" |
21 | #include "xfs_bit.h" |
22 | #include "xfs_log.h" |
23 | #include "xfs_inum.h" |
24 | #include "xfs_trans.h" |
25 | #include "xfs_sb.h" |
26 | #include "xfs_ag.h" |
27 | #include "xfs_mount.h" |
28 | #include "xfs_buf_item.h" |
29 | #include "xfs_trans_priv.h" |
30 | #include "xfs_error.h" |
31 | #include "xfs_trace.h" |
32 | |
33 | |
34 | kmem_zone_t *xfs_buf_item_zone; |
35 | |
36 | static inline struct xfs_buf_log_item *BUF_ITEM(struct xfs_log_item *lip) |
37 | { |
38 | return container_of(lip, struct xfs_buf_log_item, bli_item); |
39 | } |
40 | |
41 | |
42 | #ifdef XFS_TRANS_DEBUG |
43 | /* |
44 | * This function uses an alternate strategy for tracking the bytes |
45 | * that the user requests to be logged. This can then be used |
46 | * in conjunction with the bli_orig array in the buf log item to |
47 | * catch bugs in our callers' code. |
48 | * |
49 | * We also double check the bits set in xfs_buf_item_log using a |
50 | * simple algorithm to check that every byte is accounted for. |
51 | */ |
52 | STATIC void |
53 | xfs_buf_item_log_debug( |
54 | xfs_buf_log_item_t *bip, |
55 | uint first, |
56 | uint last) |
57 | { |
58 | uint x; |
59 | uint byte; |
60 | uint nbytes; |
61 | uint chunk_num; |
62 | uint word_num; |
63 | uint bit_num; |
64 | uint bit_set; |
65 | uint *wordp; |
66 | |
67 | ASSERT(bip->bli_logged != NULL); |
68 | byte = first; |
69 | nbytes = last - first + 1; |
70 | bfset(bip->bli_logged, first, nbytes); |
71 | for (x = 0; x < nbytes; x++) { |
72 | chunk_num = byte >> XFS_BLF_SHIFT; |
73 | word_num = chunk_num >> BIT_TO_WORD_SHIFT; |
74 | bit_num = chunk_num & (NBWORD - 1); |
75 | wordp = &(bip->bli_format.blf_data_map[word_num]); |
76 | bit_set = *wordp & (1 << bit_num); |
77 | ASSERT(bit_set); |
78 | byte++; |
79 | } |
80 | } |
81 | |
82 | /* |
83 | * This function is called when we flush something into a buffer without |
84 | * logging it. This happens for things like inodes which are logged |
85 | * separately from the buffer. |
86 | */ |
87 | void |
88 | xfs_buf_item_flush_log_debug( |
89 | xfs_buf_t *bp, |
90 | uint first, |
91 | uint last) |
92 | { |
93 | xfs_buf_log_item_t *bip; |
94 | uint nbytes; |
95 | |
96 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); |
97 | if ((bip == NULL) || (bip->bli_item.li_type != XFS_LI_BUF)) { |
98 | return; |
99 | } |
100 | |
101 | ASSERT(bip->bli_logged != NULL); |
102 | nbytes = last - first + 1; |
103 | bfset(bip->bli_logged, first, nbytes); |
104 | } |
105 | |
106 | /* |
107 | * This function is called to verify that our callers have logged |
108 | * all the bytes that they changed. |
109 | * |
110 | * It does this by comparing the original copy of the buffer stored in |
111 | * the buf log item's bli_orig array to the current copy of the buffer |
112 | * and ensuring that all bytes which mismatch are set in the bli_logged |
113 | * array of the buf log item. |
114 | */ |
115 | STATIC void |
116 | xfs_buf_item_log_check( |
117 | xfs_buf_log_item_t *bip) |
118 | { |
119 | char *orig; |
120 | char *buffer; |
121 | int x; |
122 | xfs_buf_t *bp; |
123 | |
124 | ASSERT(bip->bli_orig != NULL); |
125 | ASSERT(bip->bli_logged != NULL); |
126 | |
127 | bp = bip->bli_buf; |
128 | ASSERT(XFS_BUF_COUNT(bp) > 0); |
129 | ASSERT(XFS_BUF_PTR(bp) != NULL); |
130 | orig = bip->bli_orig; |
131 | buffer = XFS_BUF_PTR(bp); |
132 | for (x = 0; x < XFS_BUF_COUNT(bp); x++) { |
133 | if (orig[x] != buffer[x] && !btst(bip->bli_logged, x)) { |
134 | xfs_emerg(bp->b_mount, |
135 | "%s: bip %x buffer %x orig %x index %d", |
136 | __func__, bip, bp, orig, x); |
137 | ASSERT(0); |
138 | } |
139 | } |
140 | } |
141 | #else |
142 | #define xfs_buf_item_log_debug(x,y,z) |
143 | #define xfs_buf_item_log_check(x) |
144 | #endif |
145 | |
146 | STATIC void xfs_buf_do_callbacks(struct xfs_buf *bp); |
147 | |
148 | /* |
149 | * This returns the number of log iovecs needed to log the |
150 | * given buf log item. |
151 | * |
152 | * It calculates this as 1 iovec for the buf log format structure |
153 | * and 1 for each stretch of non-contiguous chunks to be logged. |
154 | * Contiguous chunks are logged in a single iovec. |
155 | * |
156 | * If the XFS_BLI_STALE flag has been set, then log nothing. |
157 | */ |
158 | STATIC uint |
159 | xfs_buf_item_size( |
160 | struct xfs_log_item *lip) |
161 | { |
162 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
163 | struct xfs_buf *bp = bip->bli_buf; |
164 | uint nvecs; |
165 | int next_bit; |
166 | int last_bit; |
167 | |
168 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
169 | if (bip->bli_flags & XFS_BLI_STALE) { |
170 | /* |
171 | * The buffer is stale, so all we need to log |
172 | * is the buf log format structure with the |
173 | * cancel flag in it. |
174 | */ |
175 | trace_xfs_buf_item_size_stale(bip); |
176 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); |
177 | return 1; |
178 | } |
179 | |
180 | ASSERT(bip->bli_flags & XFS_BLI_LOGGED); |
181 | nvecs = 1; |
182 | last_bit = xfs_next_bit(bip->bli_format.blf_data_map, |
183 | bip->bli_format.blf_map_size, 0); |
184 | ASSERT(last_bit != -1); |
185 | nvecs++; |
186 | while (last_bit != -1) { |
187 | /* |
188 | * This takes the bit number to start looking from and |
189 | * returns the next set bit from there. It returns -1 |
190 | * if there are no more bits set or the start bit is |
191 | * beyond the end of the bitmap. |
192 | */ |
193 | next_bit = xfs_next_bit(bip->bli_format.blf_data_map, |
194 | bip->bli_format.blf_map_size, |
195 | last_bit + 1); |
196 | /* |
197 | * If we run out of bits, leave the loop, |
198 | * else if we find a new set of bits bump the number of vecs, |
199 | * else keep scanning the current set of bits. |
200 | */ |
201 | if (next_bit == -1) { |
202 | last_bit = -1; |
203 | } else if (next_bit != last_bit + 1) { |
204 | last_bit = next_bit; |
205 | nvecs++; |
206 | } else if (xfs_buf_offset(bp, next_bit * XFS_BLF_CHUNK) != |
207 | (xfs_buf_offset(bp, last_bit * XFS_BLF_CHUNK) + |
208 | XFS_BLF_CHUNK)) { |
209 | last_bit = next_bit; |
210 | nvecs++; |
211 | } else { |
212 | last_bit++; |
213 | } |
214 | } |
215 | |
216 | trace_xfs_buf_item_size(bip); |
217 | return nvecs; |
218 | } |
219 | |
220 | /* |
221 | * This is called to fill in the vector of log iovecs for the |
222 | * given log buf item. It fills the first entry with a buf log |
223 | * format structure, and the rest point to contiguous chunks |
224 | * within the buffer. |
225 | */ |
226 | STATIC void |
227 | xfs_buf_item_format( |
228 | struct xfs_log_item *lip, |
229 | struct xfs_log_iovec *vecp) |
230 | { |
231 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
232 | struct xfs_buf *bp = bip->bli_buf; |
233 | uint base_size; |
234 | uint nvecs; |
235 | int first_bit; |
236 | int last_bit; |
237 | int next_bit; |
238 | uint nbits; |
239 | uint buffer_offset; |
240 | |
241 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
242 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || |
243 | (bip->bli_flags & XFS_BLI_STALE)); |
244 | |
245 | /* |
246 | * The size of the base structure is the size of the |
247 | * declared structure plus the space for the extra words |
248 | * of the bitmap. We subtract one from the map size, because |
249 | * the first element of the bitmap is accounted for in the |
250 | * size of the base structure. |
251 | */ |
252 | base_size = |
253 | (uint)(sizeof(xfs_buf_log_format_t) + |
254 | ((bip->bli_format.blf_map_size - 1) * sizeof(uint))); |
255 | vecp->i_addr = &bip->bli_format; |
256 | vecp->i_len = base_size; |
257 | vecp->i_type = XLOG_REG_TYPE_BFORMAT; |
258 | vecp++; |
259 | nvecs = 1; |
260 | |
261 | /* |
262 | * If it is an inode buffer, transfer the in-memory state to the |
263 | * format flags and clear the in-memory state. We do not transfer |
264 | * this state if the inode buffer allocation has not yet been committed |
265 | * to the log as setting the XFS_BLI_INODE_BUF flag will prevent |
266 | * correct replay of the inode allocation. |
267 | */ |
268 | if (bip->bli_flags & XFS_BLI_INODE_BUF) { |
269 | if (!((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && |
270 | xfs_log_item_in_current_chkpt(lip))) |
271 | bip->bli_format.blf_flags |= XFS_BLF_INODE_BUF; |
272 | bip->bli_flags &= ~XFS_BLI_INODE_BUF; |
273 | } |
274 | |
275 | if (bip->bli_flags & XFS_BLI_STALE) { |
276 | /* |
277 | * The buffer is stale, so all we need to log |
278 | * is the buf log format structure with the |
279 | * cancel flag in it. |
280 | */ |
281 | trace_xfs_buf_item_format_stale(bip); |
282 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); |
283 | bip->bli_format.blf_size = nvecs; |
284 | return; |
285 | } |
286 | |
287 | /* |
288 | * Fill in an iovec for each set of contiguous chunks. |
289 | */ |
290 | first_bit = xfs_next_bit(bip->bli_format.blf_data_map, |
291 | bip->bli_format.blf_map_size, 0); |
292 | ASSERT(first_bit != -1); |
293 | last_bit = first_bit; |
294 | nbits = 1; |
295 | for (;;) { |
296 | /* |
297 | * This takes the bit number to start looking from and |
298 | * returns the next set bit from there. It returns -1 |
299 | * if there are no more bits set or the start bit is |
300 | * beyond the end of the bitmap. |
301 | */ |
302 | next_bit = xfs_next_bit(bip->bli_format.blf_data_map, |
303 | bip->bli_format.blf_map_size, |
304 | (uint)last_bit + 1); |
305 | /* |
306 | * If we run out of bits fill in the last iovec and get |
307 | * out of the loop. |
308 | * Else if we start a new set of bits then fill in the |
309 | * iovec for the series we were looking at and start |
310 | * counting the bits in the new one. |
311 | * Else we're still in the same set of bits so just |
312 | * keep counting and scanning. |
313 | */ |
314 | if (next_bit == -1) { |
315 | buffer_offset = first_bit * XFS_BLF_CHUNK; |
316 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
317 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
318 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
319 | nvecs++; |
320 | break; |
321 | } else if (next_bit != last_bit + 1) { |
322 | buffer_offset = first_bit * XFS_BLF_CHUNK; |
323 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
324 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
325 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
326 | nvecs++; |
327 | vecp++; |
328 | first_bit = next_bit; |
329 | last_bit = next_bit; |
330 | nbits = 1; |
331 | } else if (xfs_buf_offset(bp, next_bit << XFS_BLF_SHIFT) != |
332 | (xfs_buf_offset(bp, last_bit << XFS_BLF_SHIFT) + |
333 | XFS_BLF_CHUNK)) { |
334 | buffer_offset = first_bit * XFS_BLF_CHUNK; |
335 | vecp->i_addr = xfs_buf_offset(bp, buffer_offset); |
336 | vecp->i_len = nbits * XFS_BLF_CHUNK; |
337 | vecp->i_type = XLOG_REG_TYPE_BCHUNK; |
338 | /* You would think we need to bump the nvecs here too, but we do not |
339 | * this number is used by recovery, and it gets confused by the boundary |
340 | * split here |
341 | * nvecs++; |
342 | */ |
343 | vecp++; |
344 | first_bit = next_bit; |
345 | last_bit = next_bit; |
346 | nbits = 1; |
347 | } else { |
348 | last_bit++; |
349 | nbits++; |
350 | } |
351 | } |
352 | bip->bli_format.blf_size = nvecs; |
353 | |
354 | /* |
355 | * Check to make sure everything is consistent. |
356 | */ |
357 | trace_xfs_buf_item_format(bip); |
358 | xfs_buf_item_log_check(bip); |
359 | } |
360 | |
361 | /* |
362 | * This is called to pin the buffer associated with the buf log item in memory |
363 | * so it cannot be written out. |
364 | * |
365 | * We also always take a reference to the buffer log item here so that the bli |
366 | * is held while the item is pinned in memory. This means that we can |
367 | * unconditionally drop the reference count a transaction holds when the |
368 | * transaction is completed. |
369 | */ |
370 | STATIC void |
371 | xfs_buf_item_pin( |
372 | struct xfs_log_item *lip) |
373 | { |
374 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
375 | |
376 | ASSERT(XFS_BUF_ISBUSY(bip->bli_buf)); |
377 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
378 | ASSERT((bip->bli_flags & XFS_BLI_LOGGED) || |
379 | (bip->bli_flags & XFS_BLI_STALE)); |
380 | |
381 | trace_xfs_buf_item_pin(bip); |
382 | |
383 | atomic_inc(&bip->bli_refcount); |
384 | atomic_inc(&bip->bli_buf->b_pin_count); |
385 | } |
386 | |
387 | /* |
388 | * This is called to unpin the buffer associated with the buf log |
389 | * item which was previously pinned with a call to xfs_buf_item_pin(). |
390 | * |
391 | * Also drop the reference to the buf item for the current transaction. |
392 | * If the XFS_BLI_STALE flag is set and we are the last reference, |
393 | * then free up the buf log item and unlock the buffer. |
394 | * |
395 | * If the remove flag is set we are called from uncommit in the |
396 | * forced-shutdown path. If that is true and the reference count on |
397 | * the log item is going to drop to zero we need to free the item's |
398 | * descriptor in the transaction. |
399 | */ |
400 | STATIC void |
401 | xfs_buf_item_unpin( |
402 | struct xfs_log_item *lip, |
403 | int remove) |
404 | { |
405 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
406 | xfs_buf_t *bp = bip->bli_buf; |
407 | struct xfs_ail *ailp = lip->li_ailp; |
408 | int stale = bip->bli_flags & XFS_BLI_STALE; |
409 | int freed; |
410 | |
411 | ASSERT(XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t *) == bip); |
412 | ASSERT(atomic_read(&bip->bli_refcount) > 0); |
413 | |
414 | trace_xfs_buf_item_unpin(bip); |
415 | |
416 | freed = atomic_dec_and_test(&bip->bli_refcount); |
417 | |
418 | if (atomic_dec_and_test(&bp->b_pin_count)) |
419 | wake_up_all(&bp->b_waiters); |
420 | |
421 | if (freed && stale) { |
422 | ASSERT(bip->bli_flags & XFS_BLI_STALE); |
423 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); |
424 | ASSERT(!(XFS_BUF_ISDELAYWRITE(bp))); |
425 | ASSERT(XFS_BUF_ISSTALE(bp)); |
426 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); |
427 | |
428 | trace_xfs_buf_item_unpin_stale(bip); |
429 | |
430 | if (remove) { |
431 | /* |
432 | * If we are in a transaction context, we have to |
433 | * remove the log item from the transaction as we are |
434 | * about to release our reference to the buffer. If we |
435 | * don't, the unlock that occurs later in |
436 | * xfs_trans_uncommit() will try to reference the |
437 | * buffer which we no longer have a hold on. |
438 | */ |
439 | if (lip->li_desc) |
440 | xfs_trans_del_item(lip); |
441 | |
442 | /* |
443 | * Since the transaction no longer refers to the buffer, |
444 | * the buffer should no longer refer to the transaction. |
445 | */ |
446 | XFS_BUF_SET_FSPRIVATE2(bp, NULL); |
447 | } |
448 | |
449 | /* |
450 | * If we get called here because of an IO error, we may |
451 | * or may not have the item on the AIL. xfs_trans_ail_delete() |
452 | * will take care of that situation. |
453 | * xfs_trans_ail_delete() drops the AIL lock. |
454 | */ |
455 | if (bip->bli_flags & XFS_BLI_STALE_INODE) { |
456 | xfs_buf_do_callbacks(bp); |
457 | XFS_BUF_SET_FSPRIVATE(bp, NULL); |
458 | XFS_BUF_CLR_IODONE_FUNC(bp); |
459 | } else { |
460 | spin_lock(&ailp->xa_lock); |
461 | xfs_trans_ail_delete(ailp, (xfs_log_item_t *)bip); |
462 | xfs_buf_item_relse(bp); |
463 | ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL); |
464 | } |
465 | xfs_buf_relse(bp); |
466 | } |
467 | } |
468 | |
469 | /* |
470 | * This is called to attempt to lock the buffer associated with this |
471 | * buf log item. Don't sleep on the buffer lock. If we can't get |
472 | * the lock right away, return 0. If we can get the lock, take a |
473 | * reference to the buffer. If this is a delayed write buffer that |
474 | * needs AIL help to be written back, invoke the pushbuf routine |
475 | * rather than the normal success path. |
476 | */ |
477 | STATIC uint |
478 | xfs_buf_item_trylock( |
479 | struct xfs_log_item *lip) |
480 | { |
481 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
482 | struct xfs_buf *bp = bip->bli_buf; |
483 | |
484 | if (XFS_BUF_ISPINNED(bp)) |
485 | return XFS_ITEM_PINNED; |
486 | if (!XFS_BUF_CPSEMA(bp)) |
487 | return XFS_ITEM_LOCKED; |
488 | |
489 | /* take a reference to the buffer. */ |
490 | XFS_BUF_HOLD(bp); |
491 | |
492 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
493 | trace_xfs_buf_item_trylock(bip); |
494 | if (XFS_BUF_ISDELAYWRITE(bp)) |
495 | return XFS_ITEM_PUSHBUF; |
496 | return XFS_ITEM_SUCCESS; |
497 | } |
498 | |
499 | /* |
500 | * Release the buffer associated with the buf log item. If there is no dirty |
501 | * logged data associated with the buffer recorded in the buf log item, then |
502 | * free the buf log item and remove the reference to it in the buffer. |
503 | * |
504 | * This call ignores the recursion count. It is only called when the buffer |
505 | * should REALLY be unlocked, regardless of the recursion count. |
506 | * |
507 | * We unconditionally drop the transaction's reference to the log item. If the |
508 | * item was logged, then another reference was taken when it was pinned, so we |
509 | * can safely drop the transaction reference now. This also allows us to avoid |
510 | * potential races with the unpin code freeing the bli by not referencing the |
511 | * bli after we've dropped the reference count. |
512 | * |
513 | * If the XFS_BLI_HOLD flag is set in the buf log item, then free the log item |
514 | * if necessary but do not unlock the buffer. This is for support of |
515 | * xfs_trans_bhold(). Make sure the XFS_BLI_HOLD field is cleared if we don't |
516 | * free the item. |
517 | */ |
518 | STATIC void |
519 | xfs_buf_item_unlock( |
520 | struct xfs_log_item *lip) |
521 | { |
522 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
523 | struct xfs_buf *bp = bip->bli_buf; |
524 | int aborted; |
525 | uint hold; |
526 | |
527 | /* Clear the buffer's association with this transaction. */ |
528 | XFS_BUF_SET_FSPRIVATE2(bp, NULL); |
529 | |
530 | /* |
531 | * If this is a transaction abort, don't return early. Instead, allow |
532 | * the brelse to happen. Normally it would be done for stale |
533 | * (cancelled) buffers at unpin time, but we'll never go through the |
534 | * pin/unpin cycle if we abort inside commit. |
535 | */ |
536 | aborted = (lip->li_flags & XFS_LI_ABORTED) != 0; |
537 | |
538 | /* |
539 | * Before possibly freeing the buf item, determine if we should |
540 | * release the buffer at the end of this routine. |
541 | */ |
542 | hold = bip->bli_flags & XFS_BLI_HOLD; |
543 | |
544 | /* Clear the per transaction state. */ |
545 | bip->bli_flags &= ~(XFS_BLI_LOGGED | XFS_BLI_HOLD); |
546 | |
547 | /* |
548 | * If the buf item is marked stale, then don't do anything. We'll |
549 | * unlock the buffer and free the buf item when the buffer is unpinned |
550 | * for the last time. |
551 | */ |
552 | if (bip->bli_flags & XFS_BLI_STALE) { |
553 | trace_xfs_buf_item_unlock_stale(bip); |
554 | ASSERT(bip->bli_format.blf_flags & XFS_BLF_CANCEL); |
555 | if (!aborted) { |
556 | atomic_dec(&bip->bli_refcount); |
557 | return; |
558 | } |
559 | } |
560 | |
561 | trace_xfs_buf_item_unlock(bip); |
562 | |
563 | /* |
564 | * If the buf item isn't tracking any data, free it, otherwise drop the |
565 | * reference we hold to it. |
566 | */ |
567 | if (xfs_bitmap_empty(bip->bli_format.blf_data_map, |
568 | bip->bli_format.blf_map_size)) |
569 | xfs_buf_item_relse(bp); |
570 | else |
571 | atomic_dec(&bip->bli_refcount); |
572 | |
573 | if (!hold) |
574 | xfs_buf_relse(bp); |
575 | } |
576 | |
577 | /* |
578 | * This is called to find out where the oldest active copy of the |
579 | * buf log item in the on disk log resides now that the last log |
580 | * write of it completed at the given lsn. |
581 | * We always re-log all the dirty data in a buffer, so usually the |
582 | * latest copy in the on disk log is the only one that matters. For |
583 | * those cases we simply return the given lsn. |
584 | * |
585 | * The one exception to this is for buffers full of newly allocated |
586 | * inodes. These buffers are only relogged with the XFS_BLI_INODE_BUF |
587 | * flag set, indicating that only the di_next_unlinked fields from the |
588 | * inodes in the buffers will be replayed during recovery. If the |
589 | * original newly allocated inode images have not yet been flushed |
590 | * when the buffer is so relogged, then we need to make sure that we |
591 | * keep the old images in the 'active' portion of the log. We do this |
592 | * by returning the original lsn of that transaction here rather than |
593 | * the current one. |
594 | */ |
595 | STATIC xfs_lsn_t |
596 | xfs_buf_item_committed( |
597 | struct xfs_log_item *lip, |
598 | xfs_lsn_t lsn) |
599 | { |
600 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
601 | |
602 | trace_xfs_buf_item_committed(bip); |
603 | |
604 | if ((bip->bli_flags & XFS_BLI_INODE_ALLOC_BUF) && lip->li_lsn != 0) |
605 | return lip->li_lsn; |
606 | return lsn; |
607 | } |
608 | |
609 | /* |
610 | * The buffer is locked, but is not a delayed write buffer. This happens |
611 | * if we race with IO completion and hence we don't want to try to write it |
612 | * again. Just release the buffer. |
613 | */ |
614 | STATIC void |
615 | xfs_buf_item_push( |
616 | struct xfs_log_item *lip) |
617 | { |
618 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
619 | struct xfs_buf *bp = bip->bli_buf; |
620 | |
621 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
622 | ASSERT(!XFS_BUF_ISDELAYWRITE(bp)); |
623 | |
624 | trace_xfs_buf_item_push(bip); |
625 | |
626 | xfs_buf_relse(bp); |
627 | } |
628 | |
629 | /* |
630 | * The buffer is locked and is a delayed write buffer. Promote the buffer |
631 | * in the delayed write queue as the caller knows that they must invoke |
632 | * the xfsbufd to get this buffer written. We have to unlock the buffer |
633 | * to allow the xfsbufd to write it, too. |
634 | */ |
635 | STATIC void |
636 | xfs_buf_item_pushbuf( |
637 | struct xfs_log_item *lip) |
638 | { |
639 | struct xfs_buf_log_item *bip = BUF_ITEM(lip); |
640 | struct xfs_buf *bp = bip->bli_buf; |
641 | |
642 | ASSERT(!(bip->bli_flags & XFS_BLI_STALE)); |
643 | ASSERT(XFS_BUF_ISDELAYWRITE(bp)); |
644 | |
645 | trace_xfs_buf_item_pushbuf(bip); |
646 | |
647 | xfs_buf_delwri_promote(bp); |
648 | xfs_buf_relse(bp); |
649 | } |
650 | |
651 | STATIC void |
652 | xfs_buf_item_committing( |
653 | struct xfs_log_item *lip, |
654 | xfs_lsn_t commit_lsn) |
655 | { |
656 | } |
657 | |
658 | /* |
659 | * This is the ops vector shared by all buf log items. |
660 | */ |
661 | static struct xfs_item_ops xfs_buf_item_ops = { |
662 | .iop_size = xfs_buf_item_size, |
663 | .iop_format = xfs_buf_item_format, |
664 | .iop_pin = xfs_buf_item_pin, |
665 | .iop_unpin = xfs_buf_item_unpin, |
666 | .iop_trylock = xfs_buf_item_trylock, |
667 | .iop_unlock = xfs_buf_item_unlock, |
668 | .iop_committed = xfs_buf_item_committed, |
669 | .iop_push = xfs_buf_item_push, |
670 | .iop_pushbuf = xfs_buf_item_pushbuf, |
671 | .iop_committing = xfs_buf_item_committing |
672 | }; |
673 | |
674 | |
675 | /* |
676 | * Allocate a new buf log item to go with the given buffer. |
677 | * Set the buffer's b_fsprivate field to point to the new |
678 | * buf log item. If there are other item's attached to the |
679 | * buffer (see xfs_buf_attach_iodone() below), then put the |
680 | * buf log item at the front. |
681 | */ |
682 | void |
683 | xfs_buf_item_init( |
684 | xfs_buf_t *bp, |
685 | xfs_mount_t *mp) |
686 | { |
687 | xfs_log_item_t *lip; |
688 | xfs_buf_log_item_t *bip; |
689 | int chunks; |
690 | int map_size; |
691 | |
692 | /* |
693 | * Check to see if there is already a buf log item for |
694 | * this buffer. If there is, it is guaranteed to be |
695 | * the first. If we do already have one, there is |
696 | * nothing to do here so return. |
697 | */ |
698 | ASSERT(bp->b_target->bt_mount == mp); |
699 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { |
700 | lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); |
701 | if (lip->li_type == XFS_LI_BUF) { |
702 | return; |
703 | } |
704 | } |
705 | |
706 | /* |
707 | * chunks is the number of XFS_BLF_CHUNK size pieces |
708 | * the buffer can be divided into. Make sure not to |
709 | * truncate any pieces. map_size is the size of the |
710 | * bitmap needed to describe the chunks of the buffer. |
711 | */ |
712 | chunks = (int)((XFS_BUF_COUNT(bp) + (XFS_BLF_CHUNK - 1)) >> XFS_BLF_SHIFT); |
713 | map_size = (int)((chunks + NBWORD) >> BIT_TO_WORD_SHIFT); |
714 | |
715 | bip = (xfs_buf_log_item_t*)kmem_zone_zalloc(xfs_buf_item_zone, |
716 | KM_SLEEP); |
717 | xfs_log_item_init(mp, &bip->bli_item, XFS_LI_BUF, &xfs_buf_item_ops); |
718 | bip->bli_buf = bp; |
719 | xfs_buf_hold(bp); |
720 | bip->bli_format.blf_type = XFS_LI_BUF; |
721 | bip->bli_format.blf_blkno = (__int64_t)XFS_BUF_ADDR(bp); |
722 | bip->bli_format.blf_len = (ushort)BTOBB(XFS_BUF_COUNT(bp)); |
723 | bip->bli_format.blf_map_size = map_size; |
724 | |
725 | #ifdef XFS_TRANS_DEBUG |
726 | /* |
727 | * Allocate the arrays for tracking what needs to be logged |
728 | * and what our callers request to be logged. bli_orig |
729 | * holds a copy of the original, clean buffer for comparison |
730 | * against, and bli_logged keeps a 1 bit flag per byte in |
731 | * the buffer to indicate which bytes the callers have asked |
732 | * to have logged. |
733 | */ |
734 | bip->bli_orig = (char *)kmem_alloc(XFS_BUF_COUNT(bp), KM_SLEEP); |
735 | memcpy(bip->bli_orig, XFS_BUF_PTR(bp), XFS_BUF_COUNT(bp)); |
736 | bip->bli_logged = (char *)kmem_zalloc(XFS_BUF_COUNT(bp) / NBBY, KM_SLEEP); |
737 | #endif |
738 | |
739 | /* |
740 | * Put the buf item into the list of items attached to the |
741 | * buffer at the front. |
742 | */ |
743 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { |
744 | bip->bli_item.li_bio_list = |
745 | XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); |
746 | } |
747 | XFS_BUF_SET_FSPRIVATE(bp, bip); |
748 | } |
749 | |
750 | |
751 | /* |
752 | * Mark bytes first through last inclusive as dirty in the buf |
753 | * item's bitmap. |
754 | */ |
755 | void |
756 | xfs_buf_item_log( |
757 | xfs_buf_log_item_t *bip, |
758 | uint first, |
759 | uint last) |
760 | { |
761 | uint first_bit; |
762 | uint last_bit; |
763 | uint bits_to_set; |
764 | uint bits_set; |
765 | uint word_num; |
766 | uint *wordp; |
767 | uint bit; |
768 | uint end_bit; |
769 | uint mask; |
770 | |
771 | /* |
772 | * Mark the item as having some dirty data for |
773 | * quick reference in xfs_buf_item_dirty. |
774 | */ |
775 | bip->bli_flags |= XFS_BLI_DIRTY; |
776 | |
777 | /* |
778 | * Convert byte offsets to bit numbers. |
779 | */ |
780 | first_bit = first >> XFS_BLF_SHIFT; |
781 | last_bit = last >> XFS_BLF_SHIFT; |
782 | |
783 | /* |
784 | * Calculate the total number of bits to be set. |
785 | */ |
786 | bits_to_set = last_bit - first_bit + 1; |
787 | |
788 | /* |
789 | * Get a pointer to the first word in the bitmap |
790 | * to set a bit in. |
791 | */ |
792 | word_num = first_bit >> BIT_TO_WORD_SHIFT; |
793 | wordp = &(bip->bli_format.blf_data_map[word_num]); |
794 | |
795 | /* |
796 | * Calculate the starting bit in the first word. |
797 | */ |
798 | bit = first_bit & (uint)(NBWORD - 1); |
799 | |
800 | /* |
801 | * First set any bits in the first word of our range. |
802 | * If it starts at bit 0 of the word, it will be |
803 | * set below rather than here. That is what the variable |
804 | * bit tells us. The variable bits_set tracks the number |
805 | * of bits that have been set so far. End_bit is the number |
806 | * of the last bit to be set in this word plus one. |
807 | */ |
808 | if (bit) { |
809 | end_bit = MIN(bit + bits_to_set, (uint)NBWORD); |
810 | mask = ((1 << (end_bit - bit)) - 1) << bit; |
811 | *wordp |= mask; |
812 | wordp++; |
813 | bits_set = end_bit - bit; |
814 | } else { |
815 | bits_set = 0; |
816 | } |
817 | |
818 | /* |
819 | * Now set bits a whole word at a time that are between |
820 | * first_bit and last_bit. |
821 | */ |
822 | while ((bits_to_set - bits_set) >= NBWORD) { |
823 | *wordp |= 0xffffffff; |
824 | bits_set += NBWORD; |
825 | wordp++; |
826 | } |
827 | |
828 | /* |
829 | * Finally, set any bits left to be set in one last partial word. |
830 | */ |
831 | end_bit = bits_to_set - bits_set; |
832 | if (end_bit) { |
833 | mask = (1 << end_bit) - 1; |
834 | *wordp |= mask; |
835 | } |
836 | |
837 | xfs_buf_item_log_debug(bip, first, last); |
838 | } |
839 | |
840 | |
841 | /* |
842 | * Return 1 if the buffer has some data that has been logged (at any |
843 | * point, not just the current transaction) and 0 if not. |
844 | */ |
845 | uint |
846 | xfs_buf_item_dirty( |
847 | xfs_buf_log_item_t *bip) |
848 | { |
849 | return (bip->bli_flags & XFS_BLI_DIRTY); |
850 | } |
851 | |
852 | STATIC void |
853 | xfs_buf_item_free( |
854 | xfs_buf_log_item_t *bip) |
855 | { |
856 | #ifdef XFS_TRANS_DEBUG |
857 | kmem_free(bip->bli_orig); |
858 | kmem_free(bip->bli_logged); |
859 | #endif /* XFS_TRANS_DEBUG */ |
860 | |
861 | kmem_zone_free(xfs_buf_item_zone, bip); |
862 | } |
863 | |
864 | /* |
865 | * This is called when the buf log item is no longer needed. It should |
866 | * free the buf log item associated with the given buffer and clear |
867 | * the buffer's pointer to the buf log item. If there are no more |
868 | * items in the list, clear the b_iodone field of the buffer (see |
869 | * xfs_buf_attach_iodone() below). |
870 | */ |
871 | void |
872 | xfs_buf_item_relse( |
873 | xfs_buf_t *bp) |
874 | { |
875 | xfs_buf_log_item_t *bip; |
876 | |
877 | trace_xfs_buf_item_relse(bp, _RET_IP_); |
878 | |
879 | bip = XFS_BUF_FSPRIVATE(bp, xfs_buf_log_item_t*); |
880 | XFS_BUF_SET_FSPRIVATE(bp, bip->bli_item.li_bio_list); |
881 | if ((XFS_BUF_FSPRIVATE(bp, void *) == NULL) && |
882 | (XFS_BUF_IODONE_FUNC(bp) != NULL)) { |
883 | XFS_BUF_CLR_IODONE_FUNC(bp); |
884 | } |
885 | xfs_buf_rele(bp); |
886 | xfs_buf_item_free(bip); |
887 | } |
888 | |
889 | |
890 | /* |
891 | * Add the given log item with its callback to the list of callbacks |
892 | * to be called when the buffer's I/O completes. If it is not set |
893 | * already, set the buffer's b_iodone() routine to be |
894 | * xfs_buf_iodone_callbacks() and link the log item into the list of |
895 | * items rooted at b_fsprivate. Items are always added as the second |
896 | * entry in the list if there is a first, because the buf item code |
897 | * assumes that the buf log item is first. |
898 | */ |
899 | void |
900 | xfs_buf_attach_iodone( |
901 | xfs_buf_t *bp, |
902 | void (*cb)(xfs_buf_t *, xfs_log_item_t *), |
903 | xfs_log_item_t *lip) |
904 | { |
905 | xfs_log_item_t *head_lip; |
906 | |
907 | ASSERT(XFS_BUF_ISBUSY(bp)); |
908 | ASSERT(XFS_BUF_VALUSEMA(bp) <= 0); |
909 | |
910 | lip->li_cb = cb; |
911 | if (XFS_BUF_FSPRIVATE(bp, void *) != NULL) { |
912 | head_lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *); |
913 | lip->li_bio_list = head_lip->li_bio_list; |
914 | head_lip->li_bio_list = lip; |
915 | } else { |
916 | XFS_BUF_SET_FSPRIVATE(bp, lip); |
917 | } |
918 | |
919 | ASSERT((XFS_BUF_IODONE_FUNC(bp) == xfs_buf_iodone_callbacks) || |
920 | (XFS_BUF_IODONE_FUNC(bp) == NULL)); |
921 | XFS_BUF_SET_IODONE_FUNC(bp, xfs_buf_iodone_callbacks); |
922 | } |
923 | |
924 | /* |
925 | * We can have many callbacks on a buffer. Running the callbacks individually |
926 | * can cause a lot of contention on the AIL lock, so we allow for a single |
927 | * callback to be able to scan the remaining lip->li_bio_list for other items |
928 | * of the same type and callback to be processed in the first call. |
929 | * |
930 | * As a result, the loop walking the callback list below will also modify the |
931 | * list. it removes the first item from the list and then runs the callback. |
932 | * The loop then restarts from the new head of the list. This allows the |
933 | * callback to scan and modify the list attached to the buffer and we don't |
934 | * have to care about maintaining a next item pointer. |
935 | */ |
936 | STATIC void |
937 | xfs_buf_do_callbacks( |
938 | struct xfs_buf *bp) |
939 | { |
940 | struct xfs_log_item *lip; |
941 | |
942 | while ((lip = XFS_BUF_FSPRIVATE(bp, xfs_log_item_t *)) != NULL) { |
943 | XFS_BUF_SET_FSPRIVATE(bp, lip->li_bio_list); |
944 | ASSERT(lip->li_cb != NULL); |
945 | /* |
946 | * Clear the next pointer so we don't have any |
947 | * confusion if the item is added to another buf. |
948 | * Don't touch the log item after calling its |
949 | * callback, because it could have freed itself. |
950 | */ |
951 | lip->li_bio_list = NULL; |
952 | lip->li_cb(bp, lip); |
953 | } |
954 | } |
955 | |
956 | /* |
957 | * This is the iodone() function for buffers which have had callbacks |
958 | * attached to them by xfs_buf_attach_iodone(). It should remove each |
959 | * log item from the buffer's list and call the callback of each in turn. |
960 | * When done, the buffer's fsprivate field is set to NULL and the buffer |
961 | * is unlocked with a call to iodone(). |
962 | */ |
963 | void |
964 | xfs_buf_iodone_callbacks( |
965 | struct xfs_buf *bp) |
966 | { |
967 | struct xfs_log_item *lip = bp->b_fspriv; |
968 | struct xfs_mount *mp = lip->li_mountp; |
969 | static ulong lasttime; |
970 | static xfs_buftarg_t *lasttarg; |
971 | |
972 | if (likely(!XFS_BUF_GETERROR(bp))) |
973 | goto do_callbacks; |
974 | |
975 | /* |
976 | * If we've already decided to shutdown the filesystem because of |
977 | * I/O errors, there's no point in giving this a retry. |
978 | */ |
979 | if (XFS_FORCED_SHUTDOWN(mp)) { |
980 | XFS_BUF_SUPER_STALE(bp); |
981 | trace_xfs_buf_item_iodone(bp, _RET_IP_); |
982 | goto do_callbacks; |
983 | } |
984 | |
985 | if (XFS_BUF_TARGET(bp) != lasttarg || |
986 | time_after(jiffies, (lasttime + 5*HZ))) { |
987 | lasttime = jiffies; |
988 | xfs_alert(mp, "Device %s: metadata write error block 0x%llx", |
989 | XFS_BUFTARG_NAME(XFS_BUF_TARGET(bp)), |
990 | (__uint64_t)XFS_BUF_ADDR(bp)); |
991 | } |
992 | lasttarg = XFS_BUF_TARGET(bp); |
993 | |
994 | /* |
995 | * If the write was asynchronous then no one will be looking for the |
996 | * error. Clear the error state and write the buffer out again. |
997 | * |
998 | * During sync or umount we'll write all pending buffers again |
999 | * synchronous, which will catch these errors if they keep hanging |
1000 | * around. |
1001 | */ |
1002 | if (XFS_BUF_ISASYNC(bp)) { |
1003 | XFS_BUF_ERROR(bp, 0); /* errno of 0 unsets the flag */ |
1004 | |
1005 | if (!XFS_BUF_ISSTALE(bp)) { |
1006 | XFS_BUF_DELAYWRITE(bp); |
1007 | XFS_BUF_DONE(bp); |
1008 | XFS_BUF_SET_START(bp); |
1009 | } |
1010 | ASSERT(XFS_BUF_IODONE_FUNC(bp)); |
1011 | trace_xfs_buf_item_iodone_async(bp, _RET_IP_); |
1012 | xfs_buf_relse(bp); |
1013 | return; |
1014 | } |
1015 | |
1016 | /* |
1017 | * If the write of the buffer was synchronous, we want to make |
1018 | * sure to return the error to the caller of xfs_bwrite(). |
1019 | */ |
1020 | XFS_BUF_STALE(bp); |
1021 | XFS_BUF_DONE(bp); |
1022 | XFS_BUF_UNDELAYWRITE(bp); |
1023 | |
1024 | trace_xfs_buf_error_relse(bp, _RET_IP_); |
1025 | xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR); |
1026 | |
1027 | do_callbacks: |
1028 | xfs_buf_do_callbacks(bp); |
1029 | XFS_BUF_SET_FSPRIVATE(bp, NULL); |
1030 | XFS_BUF_CLR_IODONE_FUNC(bp); |
1031 | xfs_buf_ioend(bp, 0); |
1032 | } |
1033 | |
1034 | /* |
1035 | * This is the iodone() function for buffers which have been |
1036 | * logged. It is called when they are eventually flushed out. |
1037 | * It should remove the buf item from the AIL, and free the buf item. |
1038 | * It is called by xfs_buf_iodone_callbacks() above which will take |
1039 | * care of cleaning up the buffer itself. |
1040 | */ |
1041 | void |
1042 | xfs_buf_iodone( |
1043 | struct xfs_buf *bp, |
1044 | struct xfs_log_item *lip) |
1045 | { |
1046 | struct xfs_ail *ailp = lip->li_ailp; |
1047 | |
1048 | ASSERT(BUF_ITEM(lip)->bli_buf == bp); |
1049 | |
1050 | xfs_buf_rele(bp); |
1051 | |
1052 | /* |
1053 | * If we are forcibly shutting down, this may well be |
1054 | * off the AIL already. That's because we simulate the |
1055 | * log-committed callbacks to unpin these buffers. Or we may never |
1056 | * have put this item on AIL because of the transaction was |
1057 | * aborted forcibly. xfs_trans_ail_delete() takes care of these. |
1058 | * |
1059 | * Either way, AIL is useless if we're forcing a shutdown. |
1060 | */ |
1061 | spin_lock(&ailp->xa_lock); |
1062 | xfs_trans_ail_delete(ailp, lip); |
1063 | xfs_buf_item_free(BUF_ITEM(lip)); |
1064 | } |
1065 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9