]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/fs/jbd/checkpoint.c | |
3 | * | |
4 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1999 | |
5 | * | |
6 | * Copyright 1999 Red Hat Software --- All Rights Reserved | |
7 | * | |
8 | * This file is part of the Linux kernel and is made available under | |
9 | * the terms of the GNU General Public License, version 2, or at your | |
10 | * option, any later version, incorporated herein by reference. | |
11 | * | |
12 | * Checkpoint routines for the generic filesystem journaling code. | |
13 | * Part of the ext2fs journaling system. | |
14 | * | |
15 | * Checkpointing is the process of ensuring that a section of the log is | |
16 | * committed fully to disk, so that that portion of the log can be | |
17 | * reused. | |
18 | */ | |
19 | ||
20 | #include <linux/time.h> | |
21 | #include <linux/fs.h> | |
22 | #include <linux/jbd.h> | |
23 | #include <linux/errno.h> | |
24 | #include <linux/slab.h> | |
25 | ||
26 | /* | |
27 | * Unlink a buffer from a transaction checkpoint list. | |
28 | * | |
29 | * Called with j_list_lock held. | |
30 | */ | |
31 | static inline void __buffer_unlink_first(struct journal_head *jh) | |
32 | { | |
33 | transaction_t *transaction = jh->b_cp_transaction; | |
34 | ||
35 | jh->b_cpnext->b_cpprev = jh->b_cpprev; | |
36 | jh->b_cpprev->b_cpnext = jh->b_cpnext; | |
37 | if (transaction->t_checkpoint_list == jh) { | |
38 | transaction->t_checkpoint_list = jh->b_cpnext; | |
39 | if (transaction->t_checkpoint_list == jh) | |
40 | transaction->t_checkpoint_list = NULL; | |
41 | } | |
42 | } | |
43 | ||
44 | /* | |
45 | * Unlink a buffer from a transaction checkpoint(io) list. | |
46 | * | |
47 | * Called with j_list_lock held. | |
48 | */ | |
49 | static inline void __buffer_unlink(struct journal_head *jh) | |
50 | { | |
51 | transaction_t *transaction = jh->b_cp_transaction; | |
52 | ||
53 | __buffer_unlink_first(jh); | |
54 | if (transaction->t_checkpoint_io_list == jh) { | |
55 | transaction->t_checkpoint_io_list = jh->b_cpnext; | |
56 | if (transaction->t_checkpoint_io_list == jh) | |
57 | transaction->t_checkpoint_io_list = NULL; | |
58 | } | |
59 | } | |
60 | ||
61 | /* | |
62 | * Move a buffer from the checkpoint list to the checkpoint io list | |
63 | * | |
64 | * Called with j_list_lock held | |
65 | */ | |
66 | static inline void __buffer_relink_io(struct journal_head *jh) | |
67 | { | |
68 | transaction_t *transaction = jh->b_cp_transaction; | |
69 | ||
70 | __buffer_unlink_first(jh); | |
71 | ||
72 | if (!transaction->t_checkpoint_io_list) { | |
73 | jh->b_cpnext = jh->b_cpprev = jh; | |
74 | } else { | |
75 | jh->b_cpnext = transaction->t_checkpoint_io_list; | |
76 | jh->b_cpprev = transaction->t_checkpoint_io_list->b_cpprev; | |
77 | jh->b_cpprev->b_cpnext = jh; | |
78 | jh->b_cpnext->b_cpprev = jh; | |
79 | } | |
80 | transaction->t_checkpoint_io_list = jh; | |
81 | } | |
82 | ||
83 | /* | |
84 | * Try to release a checkpointed buffer from its transaction. | |
85 | * Returns 1 if we released it and 2 if we also released the | |
86 | * whole transaction. | |
87 | * | |
88 | * Requires j_list_lock | |
89 | * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it | |
90 | */ | |
91 | static int __try_to_free_cp_buf(struct journal_head *jh) | |
92 | { | |
93 | int ret = 0; | |
94 | struct buffer_head *bh = jh2bh(jh); | |
95 | ||
96 | if (jh->b_jlist == BJ_None && !buffer_locked(bh) && | |
97 | !buffer_dirty(bh) && !buffer_write_io_error(bh)) { | |
98 | JBUFFER_TRACE(jh, "remove from checkpoint list"); | |
99 | ret = __journal_remove_checkpoint(jh) + 1; | |
100 | jbd_unlock_bh_state(bh); | |
101 | journal_remove_journal_head(bh); | |
102 | BUFFER_TRACE(bh, "release"); | |
103 | __brelse(bh); | |
104 | } else { | |
105 | jbd_unlock_bh_state(bh); | |
106 | } | |
107 | return ret; | |
108 | } | |
109 | ||
110 | /* | |
111 | * __log_wait_for_space: wait until there is space in the journal. | |
112 | * | |
113 | * Called under j-state_lock *only*. It will be unlocked if we have to wait | |
114 | * for a checkpoint to free up some space in the log. | |
115 | */ | |
116 | void __log_wait_for_space(journal_t *journal) | |
117 | { | |
118 | int nblocks, space_left; | |
119 | assert_spin_locked(&journal->j_state_lock); | |
120 | ||
121 | nblocks = jbd_space_needed(journal); | |
122 | while (__log_space_left(journal) < nblocks) { | |
123 | if (journal->j_flags & JFS_ABORT) | |
124 | return; | |
125 | spin_unlock(&journal->j_state_lock); | |
126 | mutex_lock(&journal->j_checkpoint_mutex); | |
127 | ||
128 | /* | |
129 | * Test again, another process may have checkpointed while we | |
130 | * were waiting for the checkpoint lock. If there are no | |
131 | * transactions ready to be checkpointed, try to recover | |
132 | * journal space by calling cleanup_journal_tail(), and if | |
133 | * that doesn't work, by waiting for the currently committing | |
134 | * transaction to complete. If there is absolutely no way | |
135 | * to make progress, this is either a BUG or corrupted | |
136 | * filesystem, so abort the journal and leave a stack | |
137 | * trace for forensic evidence. | |
138 | */ | |
139 | spin_lock(&journal->j_state_lock); | |
140 | spin_lock(&journal->j_list_lock); | |
141 | nblocks = jbd_space_needed(journal); | |
142 | space_left = __log_space_left(journal); | |
143 | if (space_left < nblocks) { | |
144 | int chkpt = journal->j_checkpoint_transactions != NULL; | |
145 | tid_t tid = 0; | |
146 | ||
147 | if (journal->j_committing_transaction) | |
148 | tid = journal->j_committing_transaction->t_tid; | |
149 | spin_unlock(&journal->j_list_lock); | |
150 | spin_unlock(&journal->j_state_lock); | |
151 | if (chkpt) { | |
152 | log_do_checkpoint(journal); | |
153 | } else if (cleanup_journal_tail(journal) == 0) { | |
154 | /* We were able to recover space; yay! */ | |
155 | ; | |
156 | } else if (tid) { | |
157 | log_wait_commit(journal, tid); | |
158 | } else { | |
159 | printk(KERN_ERR "%s: needed %d blocks and " | |
160 | "only had %d space available\n", | |
161 | __func__, nblocks, space_left); | |
162 | printk(KERN_ERR "%s: no way to get more " | |
163 | "journal space\n", __func__); | |
164 | WARN_ON(1); | |
165 | journal_abort(journal, 0); | |
166 | } | |
167 | spin_lock(&journal->j_state_lock); | |
168 | } else { | |
169 | spin_unlock(&journal->j_list_lock); | |
170 | } | |
171 | mutex_unlock(&journal->j_checkpoint_mutex); | |
172 | } | |
173 | } | |
174 | ||
175 | /* | |
176 | * We were unable to perform jbd_trylock_bh_state() inside j_list_lock. | |
177 | * The caller must restart a list walk. Wait for someone else to run | |
178 | * jbd_unlock_bh_state(). | |
179 | */ | |
180 | static void jbd_sync_bh(journal_t *journal, struct buffer_head *bh) | |
181 | __releases(journal->j_list_lock) | |
182 | { | |
183 | get_bh(bh); | |
184 | spin_unlock(&journal->j_list_lock); | |
185 | jbd_lock_bh_state(bh); | |
186 | jbd_unlock_bh_state(bh); | |
187 | put_bh(bh); | |
188 | } | |
189 | ||
190 | /* | |
191 | * Clean up transaction's list of buffers submitted for io. | |
192 | * We wait for any pending IO to complete and remove any clean | |
193 | * buffers. Note that we take the buffers in the opposite ordering | |
194 | * from the one in which they were submitted for IO. | |
195 | * | |
196 | * Return 0 on success, and return <0 if some buffers have failed | |
197 | * to be written out. | |
198 | * | |
199 | * Called with j_list_lock held. | |
200 | */ | |
201 | static int __wait_cp_io(journal_t *journal, transaction_t *transaction) | |
202 | { | |
203 | struct journal_head *jh; | |
204 | struct buffer_head *bh; | |
205 | tid_t this_tid; | |
206 | int released = 0; | |
207 | int ret = 0; | |
208 | ||
209 | this_tid = transaction->t_tid; | |
210 | restart: | |
211 | /* Did somebody clean up the transaction in the meanwhile? */ | |
212 | if (journal->j_checkpoint_transactions != transaction || | |
213 | transaction->t_tid != this_tid) | |
214 | return ret; | |
215 | while (!released && transaction->t_checkpoint_io_list) { | |
216 | jh = transaction->t_checkpoint_io_list; | |
217 | bh = jh2bh(jh); | |
218 | if (!jbd_trylock_bh_state(bh)) { | |
219 | jbd_sync_bh(journal, bh); | |
220 | spin_lock(&journal->j_list_lock); | |
221 | goto restart; | |
222 | } | |
223 | if (buffer_locked(bh)) { | |
224 | atomic_inc(&bh->b_count); | |
225 | spin_unlock(&journal->j_list_lock); | |
226 | jbd_unlock_bh_state(bh); | |
227 | wait_on_buffer(bh); | |
228 | /* the journal_head may have gone by now */ | |
229 | BUFFER_TRACE(bh, "brelse"); | |
230 | __brelse(bh); | |
231 | spin_lock(&journal->j_list_lock); | |
232 | goto restart; | |
233 | } | |
234 | if (unlikely(buffer_write_io_error(bh))) | |
235 | ret = -EIO; | |
236 | ||
237 | /* | |
238 | * Now in whatever state the buffer currently is, we know that | |
239 | * it has been written out and so we can drop it from the list | |
240 | */ | |
241 | released = __journal_remove_checkpoint(jh); | |
242 | jbd_unlock_bh_state(bh); | |
243 | journal_remove_journal_head(bh); | |
244 | __brelse(bh); | |
245 | } | |
246 | ||
247 | return ret; | |
248 | } | |
249 | ||
250 | #define NR_BATCH 64 | |
251 | ||
252 | static void | |
253 | __flush_batch(journal_t *journal, struct buffer_head **bhs, int *batch_count) | |
254 | { | |
255 | int i; | |
256 | ||
257 | for (i = 0; i < *batch_count; i++) | |
258 | write_dirty_buffer(bhs[i], WRITE); | |
259 | ||
260 | for (i = 0; i < *batch_count; i++) { | |
261 | struct buffer_head *bh = bhs[i]; | |
262 | clear_buffer_jwrite(bh); | |
263 | BUFFER_TRACE(bh, "brelse"); | |
264 | __brelse(bh); | |
265 | } | |
266 | *batch_count = 0; | |
267 | } | |
268 | ||
269 | /* | |
270 | * Try to flush one buffer from the checkpoint list to disk. | |
271 | * | |
272 | * Return 1 if something happened which requires us to abort the current | |
273 | * scan of the checkpoint list. Return <0 if the buffer has failed to | |
274 | * be written out. | |
275 | * | |
276 | * Called with j_list_lock held and drops it if 1 is returned | |
277 | * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it | |
278 | */ | |
279 | static int __process_buffer(journal_t *journal, struct journal_head *jh, | |
280 | struct buffer_head **bhs, int *batch_count) | |
281 | { | |
282 | struct buffer_head *bh = jh2bh(jh); | |
283 | int ret = 0; | |
284 | ||
285 | if (buffer_locked(bh)) { | |
286 | atomic_inc(&bh->b_count); | |
287 | spin_unlock(&journal->j_list_lock); | |
288 | jbd_unlock_bh_state(bh); | |
289 | wait_on_buffer(bh); | |
290 | /* the journal_head may have gone by now */ | |
291 | BUFFER_TRACE(bh, "brelse"); | |
292 | __brelse(bh); | |
293 | ret = 1; | |
294 | } else if (jh->b_transaction != NULL) { | |
295 | transaction_t *t = jh->b_transaction; | |
296 | tid_t tid = t->t_tid; | |
297 | ||
298 | spin_unlock(&journal->j_list_lock); | |
299 | jbd_unlock_bh_state(bh); | |
300 | log_start_commit(journal, tid); | |
301 | log_wait_commit(journal, tid); | |
302 | ret = 1; | |
303 | } else if (!buffer_dirty(bh)) { | |
304 | ret = 1; | |
305 | if (unlikely(buffer_write_io_error(bh))) | |
306 | ret = -EIO; | |
307 | J_ASSERT_JH(jh, !buffer_jbddirty(bh)); | |
308 | BUFFER_TRACE(bh, "remove from checkpoint"); | |
309 | __journal_remove_checkpoint(jh); | |
310 | spin_unlock(&journal->j_list_lock); | |
311 | jbd_unlock_bh_state(bh); | |
312 | journal_remove_journal_head(bh); | |
313 | __brelse(bh); | |
314 | } else { | |
315 | /* | |
316 | * Important: we are about to write the buffer, and | |
317 | * possibly block, while still holding the journal lock. | |
318 | * We cannot afford to let the transaction logic start | |
319 | * messing around with this buffer before we write it to | |
320 | * disk, as that would break recoverability. | |
321 | */ | |
322 | BUFFER_TRACE(bh, "queue"); | |
323 | get_bh(bh); | |
324 | J_ASSERT_BH(bh, !buffer_jwrite(bh)); | |
325 | set_buffer_jwrite(bh); | |
326 | bhs[*batch_count] = bh; | |
327 | __buffer_relink_io(jh); | |
328 | jbd_unlock_bh_state(bh); | |
329 | (*batch_count)++; | |
330 | if (*batch_count == NR_BATCH) { | |
331 | spin_unlock(&journal->j_list_lock); | |
332 | __flush_batch(journal, bhs, batch_count); | |
333 | ret = 1; | |
334 | } | |
335 | } | |
336 | return ret; | |
337 | } | |
338 | ||
339 | /* | |
340 | * Perform an actual checkpoint. We take the first transaction on the | |
341 | * list of transactions to be checkpointed and send all its buffers | |
342 | * to disk. We submit larger chunks of data at once. | |
343 | * | |
344 | * The journal should be locked before calling this function. | |
345 | * Called with j_checkpoint_mutex held. | |
346 | */ | |
347 | int log_do_checkpoint(journal_t *journal) | |
348 | { | |
349 | transaction_t *transaction; | |
350 | tid_t this_tid; | |
351 | int result; | |
352 | ||
353 | jbd_debug(1, "Start checkpoint\n"); | |
354 | ||
355 | /* | |
356 | * First thing: if there are any transactions in the log which | |
357 | * don't need checkpointing, just eliminate them from the | |
358 | * journal straight away. | |
359 | */ | |
360 | result = cleanup_journal_tail(journal); | |
361 | jbd_debug(1, "cleanup_journal_tail returned %d\n", result); | |
362 | if (result <= 0) | |
363 | return result; | |
364 | ||
365 | /* | |
366 | * OK, we need to start writing disk blocks. Take one transaction | |
367 | * and write it. | |
368 | */ | |
369 | result = 0; | |
370 | spin_lock(&journal->j_list_lock); | |
371 | if (!journal->j_checkpoint_transactions) | |
372 | goto out; | |
373 | transaction = journal->j_checkpoint_transactions; | |
374 | this_tid = transaction->t_tid; | |
375 | restart: | |
376 | /* | |
377 | * If someone cleaned up this transaction while we slept, we're | |
378 | * done (maybe it's a new transaction, but it fell at the same | |
379 | * address). | |
380 | */ | |
381 | if (journal->j_checkpoint_transactions == transaction && | |
382 | transaction->t_tid == this_tid) { | |
383 | int batch_count = 0; | |
384 | struct buffer_head *bhs[NR_BATCH]; | |
385 | struct journal_head *jh; | |
386 | int retry = 0, err; | |
387 | ||
388 | while (!retry && transaction->t_checkpoint_list) { | |
389 | struct buffer_head *bh; | |
390 | ||
391 | jh = transaction->t_checkpoint_list; | |
392 | bh = jh2bh(jh); | |
393 | if (!jbd_trylock_bh_state(bh)) { | |
394 | jbd_sync_bh(journal, bh); | |
395 | retry = 1; | |
396 | break; | |
397 | } | |
398 | retry = __process_buffer(journal, jh, bhs,&batch_count); | |
399 | if (retry < 0 && !result) | |
400 | result = retry; | |
401 | if (!retry && (need_resched() || | |
402 | spin_needbreak(&journal->j_list_lock))) { | |
403 | spin_unlock(&journal->j_list_lock); | |
404 | retry = 1; | |
405 | break; | |
406 | } | |
407 | } | |
408 | ||
409 | if (batch_count) { | |
410 | if (!retry) { | |
411 | spin_unlock(&journal->j_list_lock); | |
412 | retry = 1; | |
413 | } | |
414 | __flush_batch(journal, bhs, &batch_count); | |
415 | } | |
416 | ||
417 | if (retry) { | |
418 | spin_lock(&journal->j_list_lock); | |
419 | goto restart; | |
420 | } | |
421 | /* | |
422 | * Now we have cleaned up the first transaction's checkpoint | |
423 | * list. Let's clean up the second one | |
424 | */ | |
425 | err = __wait_cp_io(journal, transaction); | |
426 | if (!result) | |
427 | result = err; | |
428 | } | |
429 | out: | |
430 | spin_unlock(&journal->j_list_lock); | |
431 | if (result < 0) | |
432 | journal_abort(journal, result); | |
433 | else | |
434 | result = cleanup_journal_tail(journal); | |
435 | ||
436 | return (result < 0) ? result : 0; | |
437 | } | |
438 | ||
439 | /* | |
440 | * Check the list of checkpoint transactions for the journal to see if | |
441 | * we have already got rid of any since the last update of the log tail | |
442 | * in the journal superblock. If so, we can instantly roll the | |
443 | * superblock forward to remove those transactions from the log. | |
444 | * | |
445 | * Return <0 on error, 0 on success, 1 if there was nothing to clean up. | |
446 | * | |
447 | * Called with the journal lock held. | |
448 | * | |
449 | * This is the only part of the journaling code which really needs to be | |
450 | * aware of transaction aborts. Checkpointing involves writing to the | |
451 | * main filesystem area rather than to the journal, so it can proceed | |
452 | * even in abort state, but we must not update the super block if | |
453 | * checkpointing may have failed. Otherwise, we would lose some metadata | |
454 | * buffers which should be written-back to the filesystem. | |
455 | */ | |
456 | ||
457 | int cleanup_journal_tail(journal_t *journal) | |
458 | { | |
459 | transaction_t * transaction; | |
460 | tid_t first_tid; | |
461 | unsigned int blocknr, freed; | |
462 | ||
463 | if (is_journal_aborted(journal)) | |
464 | return 1; | |
465 | ||
466 | /* OK, work out the oldest transaction remaining in the log, and | |
467 | * the log block it starts at. | |
468 | * | |
469 | * If the log is now empty, we need to work out which is the | |
470 | * next transaction ID we will write, and where it will | |
471 | * start. */ | |
472 | ||
473 | spin_lock(&journal->j_state_lock); | |
474 | spin_lock(&journal->j_list_lock); | |
475 | transaction = journal->j_checkpoint_transactions; | |
476 | if (transaction) { | |
477 | first_tid = transaction->t_tid; | |
478 | blocknr = transaction->t_log_start; | |
479 | } else if ((transaction = journal->j_committing_transaction) != NULL) { | |
480 | first_tid = transaction->t_tid; | |
481 | blocknr = transaction->t_log_start; | |
482 | } else if ((transaction = journal->j_running_transaction) != NULL) { | |
483 | first_tid = transaction->t_tid; | |
484 | blocknr = journal->j_head; | |
485 | } else { | |
486 | first_tid = journal->j_transaction_sequence; | |
487 | blocknr = journal->j_head; | |
488 | } | |
489 | spin_unlock(&journal->j_list_lock); | |
490 | J_ASSERT(blocknr != 0); | |
491 | ||
492 | /* If the oldest pinned transaction is at the tail of the log | |
493 | already then there's not much we can do right now. */ | |
494 | if (journal->j_tail_sequence == first_tid) { | |
495 | spin_unlock(&journal->j_state_lock); | |
496 | return 1; | |
497 | } | |
498 | ||
499 | /* OK, update the superblock to recover the freed space. | |
500 | * Physical blocks come first: have we wrapped beyond the end of | |
501 | * the log? */ | |
502 | freed = blocknr - journal->j_tail; | |
503 | if (blocknr < journal->j_tail) | |
504 | freed = freed + journal->j_last - journal->j_first; | |
505 | ||
506 | jbd_debug(1, | |
507 | "Cleaning journal tail from %d to %d (offset %u), " | |
508 | "freeing %u\n", | |
509 | journal->j_tail_sequence, first_tid, blocknr, freed); | |
510 | ||
511 | journal->j_free += freed; | |
512 | journal->j_tail_sequence = first_tid; | |
513 | journal->j_tail = blocknr; | |
514 | spin_unlock(&journal->j_state_lock); | |
515 | if (!(journal->j_flags & JFS_ABORT)) | |
516 | journal_update_superblock(journal, 1); | |
517 | return 0; | |
518 | } | |
519 | ||
520 | ||
521 | /* Checkpoint list management */ | |
522 | ||
523 | /* | |
524 | * journal_clean_one_cp_list | |
525 | * | |
526 | * Find all the written-back checkpoint buffers in the given list and release them. | |
527 | * | |
528 | * Called with the journal locked. | |
529 | * Called with j_list_lock held. | |
530 | * Returns number of bufers reaped (for debug) | |
531 | */ | |
532 | ||
533 | static int journal_clean_one_cp_list(struct journal_head *jh, int *released) | |
534 | { | |
535 | struct journal_head *last_jh; | |
536 | struct journal_head *next_jh = jh; | |
537 | int ret, freed = 0; | |
538 | ||
539 | *released = 0; | |
540 | if (!jh) | |
541 | return 0; | |
542 | ||
543 | last_jh = jh->b_cpprev; | |
544 | do { | |
545 | jh = next_jh; | |
546 | next_jh = jh->b_cpnext; | |
547 | /* Use trylock because of the ranking */ | |
548 | if (jbd_trylock_bh_state(jh2bh(jh))) { | |
549 | ret = __try_to_free_cp_buf(jh); | |
550 | if (ret) { | |
551 | freed++; | |
552 | if (ret == 2) { | |
553 | *released = 1; | |
554 | return freed; | |
555 | } | |
556 | } | |
557 | } | |
558 | /* | |
559 | * This function only frees up some memory | |
560 | * if possible so we dont have an obligation | |
561 | * to finish processing. Bail out if preemption | |
562 | * requested: | |
563 | */ | |
564 | if (need_resched()) | |
565 | return freed; | |
566 | } while (jh != last_jh); | |
567 | ||
568 | return freed; | |
569 | } | |
570 | ||
571 | /* | |
572 | * journal_clean_checkpoint_list | |
573 | * | |
574 | * Find all the written-back checkpoint buffers in the journal and release them. | |
575 | * | |
576 | * Called with the journal locked. | |
577 | * Called with j_list_lock held. | |
578 | * Returns number of buffers reaped (for debug) | |
579 | */ | |
580 | ||
581 | int __journal_clean_checkpoint_list(journal_t *journal) | |
582 | { | |
583 | transaction_t *transaction, *last_transaction, *next_transaction; | |
584 | int ret = 0; | |
585 | int released; | |
586 | ||
587 | transaction = journal->j_checkpoint_transactions; | |
588 | if (!transaction) | |
589 | goto out; | |
590 | ||
591 | last_transaction = transaction->t_cpprev; | |
592 | next_transaction = transaction; | |
593 | do { | |
594 | transaction = next_transaction; | |
595 | next_transaction = transaction->t_cpnext; | |
596 | ret += journal_clean_one_cp_list(transaction-> | |
597 | t_checkpoint_list, &released); | |
598 | /* | |
599 | * This function only frees up some memory if possible so we | |
600 | * dont have an obligation to finish processing. Bail out if | |
601 | * preemption requested: | |
602 | */ | |
603 | if (need_resched()) | |
604 | goto out; | |
605 | if (released) | |
606 | continue; | |
607 | /* | |
608 | * It is essential that we are as careful as in the case of | |
609 | * t_checkpoint_list with removing the buffer from the list as | |
610 | * we can possibly see not yet submitted buffers on io_list | |
611 | */ | |
612 | ret += journal_clean_one_cp_list(transaction-> | |
613 | t_checkpoint_io_list, &released); | |
614 | if (need_resched()) | |
615 | goto out; | |
616 | } while (transaction != last_transaction); | |
617 | out: | |
618 | return ret; | |
619 | } | |
620 | ||
621 | /* | |
622 | * journal_remove_checkpoint: called after a buffer has been committed | |
623 | * to disk (either by being write-back flushed to disk, or being | |
624 | * committed to the log). | |
625 | * | |
626 | * We cannot safely clean a transaction out of the log until all of the | |
627 | * buffer updates committed in that transaction have safely been stored | |
628 | * elsewhere on disk. To achieve this, all of the buffers in a | |
629 | * transaction need to be maintained on the transaction's checkpoint | |
630 | * lists until they have been rewritten, at which point this function is | |
631 | * called to remove the buffer from the existing transaction's | |
632 | * checkpoint lists. | |
633 | * | |
634 | * The function returns 1 if it frees the transaction, 0 otherwise. | |
635 | * | |
636 | * This function is called with the journal locked. | |
637 | * This function is called with j_list_lock held. | |
638 | * This function is called with jbd_lock_bh_state(jh2bh(jh)) | |
639 | */ | |
640 | ||
641 | int __journal_remove_checkpoint(struct journal_head *jh) | |
642 | { | |
643 | transaction_t *transaction; | |
644 | journal_t *journal; | |
645 | int ret = 0; | |
646 | ||
647 | JBUFFER_TRACE(jh, "entry"); | |
648 | ||
649 | if ((transaction = jh->b_cp_transaction) == NULL) { | |
650 | JBUFFER_TRACE(jh, "not on transaction"); | |
651 | goto out; | |
652 | } | |
653 | journal = transaction->t_journal; | |
654 | ||
655 | __buffer_unlink(jh); | |
656 | jh->b_cp_transaction = NULL; | |
657 | ||
658 | if (transaction->t_checkpoint_list != NULL || | |
659 | transaction->t_checkpoint_io_list != NULL) | |
660 | goto out; | |
661 | JBUFFER_TRACE(jh, "transaction has no more buffers"); | |
662 | ||
663 | /* | |
664 | * There is one special case to worry about: if we have just pulled the | |
665 | * buffer off a running or committing transaction's checkpoing list, | |
666 | * then even if the checkpoint list is empty, the transaction obviously | |
667 | * cannot be dropped! | |
668 | * | |
669 | * The locking here around t_state is a bit sleazy. | |
670 | * See the comment at the end of journal_commit_transaction(). | |
671 | */ | |
672 | if (transaction->t_state != T_FINISHED) { | |
673 | JBUFFER_TRACE(jh, "belongs to running/committing transaction"); | |
674 | goto out; | |
675 | } | |
676 | ||
677 | /* OK, that was the last buffer for the transaction: we can now | |
678 | safely remove this transaction from the log */ | |
679 | ||
680 | __journal_drop_transaction(journal, transaction); | |
681 | ||
682 | /* Just in case anybody was waiting for more transactions to be | |
683 | checkpointed... */ | |
684 | wake_up(&journal->j_wait_logspace); | |
685 | ret = 1; | |
686 | out: | |
687 | JBUFFER_TRACE(jh, "exit"); | |
688 | return ret; | |
689 | } | |
690 | ||
691 | /* | |
692 | * journal_insert_checkpoint: put a committed buffer onto a checkpoint | |
693 | * list so that we know when it is safe to clean the transaction out of | |
694 | * the log. | |
695 | * | |
696 | * Called with the journal locked. | |
697 | * Called with j_list_lock held. | |
698 | */ | |
699 | void __journal_insert_checkpoint(struct journal_head *jh, | |
700 | transaction_t *transaction) | |
701 | { | |
702 | JBUFFER_TRACE(jh, "entry"); | |
703 | J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh))); | |
704 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
705 | ||
706 | jh->b_cp_transaction = transaction; | |
707 | ||
708 | if (!transaction->t_checkpoint_list) { | |
709 | jh->b_cpnext = jh->b_cpprev = jh; | |
710 | } else { | |
711 | jh->b_cpnext = transaction->t_checkpoint_list; | |
712 | jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev; | |
713 | jh->b_cpprev->b_cpnext = jh; | |
714 | jh->b_cpnext->b_cpprev = jh; | |
715 | } | |
716 | transaction->t_checkpoint_list = jh; | |
717 | } | |
718 | ||
719 | /* | |
720 | * We've finished with this transaction structure: adios... | |
721 | * | |
722 | * The transaction must have no links except for the checkpoint by this | |
723 | * point. | |
724 | * | |
725 | * Called with the journal locked. | |
726 | * Called with j_list_lock held. | |
727 | */ | |
728 | ||
729 | void __journal_drop_transaction(journal_t *journal, transaction_t *transaction) | |
730 | { | |
731 | assert_spin_locked(&journal->j_list_lock); | |
732 | if (transaction->t_cpnext) { | |
733 | transaction->t_cpnext->t_cpprev = transaction->t_cpprev; | |
734 | transaction->t_cpprev->t_cpnext = transaction->t_cpnext; | |
735 | if (journal->j_checkpoint_transactions == transaction) | |
736 | journal->j_checkpoint_transactions = | |
737 | transaction->t_cpnext; | |
738 | if (journal->j_checkpoint_transactions == transaction) | |
739 | journal->j_checkpoint_transactions = NULL; | |
740 | } | |
741 | ||
742 | J_ASSERT(transaction->t_state == T_FINISHED); | |
743 | J_ASSERT(transaction->t_buffers == NULL); | |
744 | J_ASSERT(transaction->t_sync_datalist == NULL); | |
745 | J_ASSERT(transaction->t_forget == NULL); | |
746 | J_ASSERT(transaction->t_iobuf_list == NULL); | |
747 | J_ASSERT(transaction->t_shadow_list == NULL); | |
748 | J_ASSERT(transaction->t_log_list == NULL); | |
749 | J_ASSERT(transaction->t_checkpoint_list == NULL); | |
750 | J_ASSERT(transaction->t_checkpoint_io_list == NULL); | |
751 | J_ASSERT(transaction->t_updates == 0); | |
752 | J_ASSERT(journal->j_committing_transaction != transaction); | |
753 | J_ASSERT(journal->j_running_transaction != transaction); | |
754 | ||
755 | jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid); | |
756 | kfree(transaction); | |
757 | } |