]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/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 | /* | |
78ce89c9 | 27 | * Unlink a buffer from a transaction checkpoint list. |
1da177e4 LT |
28 | * |
29 | * Called with j_list_lock held. | |
30 | */ | |
78ce89c9 | 31 | static inline void __buffer_unlink_first(struct journal_head *jh) |
1da177e4 | 32 | { |
78ce89c9 | 33 | transaction_t *transaction = jh->b_cp_transaction; |
1da177e4 LT |
34 | |
35 | jh->b_cpnext->b_cpprev = jh->b_cpprev; | |
36 | jh->b_cpprev->b_cpnext = jh->b_cpnext; | |
78ce89c9 | 37 | if (transaction->t_checkpoint_list == jh) { |
1da177e4 | 38 | transaction->t_checkpoint_list = jh->b_cpnext; |
78ce89c9 JK |
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; | |
1da177e4 LT |
81 | } |
82 | ||
83 | /* | |
84 | * Try to release a checkpointed buffer from its transaction. | |
78ce89c9 JK |
85 | * Returns 1 if we released it and 2 if we also released the |
86 | * whole transaction. | |
87 | * | |
1da177e4 LT |
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) && !buffer_dirty(bh)) { | |
97 | JBUFFER_TRACE(jh, "remove from checkpoint list"); | |
78ce89c9 | 98 | ret = __journal_remove_checkpoint(jh) + 1; |
1da177e4 LT |
99 | jbd_unlock_bh_state(bh); |
100 | journal_remove_journal_head(bh); | |
101 | BUFFER_TRACE(bh, "release"); | |
102 | __brelse(bh); | |
1da177e4 LT |
103 | } else { |
104 | jbd_unlock_bh_state(bh); | |
105 | } | |
106 | return ret; | |
107 | } | |
108 | ||
109 | /* | |
110 | * __log_wait_for_space: wait until there is space in the journal. | |
111 | * | |
112 | * Called under j-state_lock *only*. It will be unlocked if we have to wait | |
113 | * for a checkpoint to free up some space in the log. | |
114 | */ | |
115 | void __log_wait_for_space(journal_t *journal) | |
116 | { | |
117 | int nblocks; | |
118 | assert_spin_locked(&journal->j_state_lock); | |
119 | ||
120 | nblocks = jbd_space_needed(journal); | |
121 | while (__log_space_left(journal) < nblocks) { | |
122 | if (journal->j_flags & JFS_ABORT) | |
123 | return; | |
124 | spin_unlock(&journal->j_state_lock); | |
2c68ee75 | 125 | mutex_lock(&journal->j_checkpoint_mutex); |
1da177e4 LT |
126 | |
127 | /* | |
128 | * Test again, another process may have checkpointed while we | |
129 | * were waiting for the checkpoint lock | |
130 | */ | |
131 | spin_lock(&journal->j_state_lock); | |
132 | nblocks = jbd_space_needed(journal); | |
133 | if (__log_space_left(journal) < nblocks) { | |
134 | spin_unlock(&journal->j_state_lock); | |
135 | log_do_checkpoint(journal); | |
136 | spin_lock(&journal->j_state_lock); | |
137 | } | |
2c68ee75 | 138 | mutex_unlock(&journal->j_checkpoint_mutex); |
1da177e4 LT |
139 | } |
140 | } | |
141 | ||
142 | /* | |
143 | * We were unable to perform jbd_trylock_bh_state() inside j_list_lock. | |
144 | * The caller must restart a list walk. Wait for someone else to run | |
145 | * jbd_unlock_bh_state(). | |
146 | */ | |
147 | static void jbd_sync_bh(journal_t *journal, struct buffer_head *bh) | |
148 | { | |
149 | get_bh(bh); | |
150 | spin_unlock(&journal->j_list_lock); | |
151 | jbd_lock_bh_state(bh); | |
152 | jbd_unlock_bh_state(bh); | |
153 | put_bh(bh); | |
154 | } | |
155 | ||
156 | /* | |
78ce89c9 JK |
157 | * Clean up transaction's list of buffers submitted for io. |
158 | * We wait for any pending IO to complete and remove any clean | |
159 | * buffers. Note that we take the buffers in the opposite ordering | |
160 | * from the one in which they were submitted for IO. | |
1da177e4 LT |
161 | * |
162 | * Called with j_list_lock held. | |
163 | */ | |
78ce89c9 | 164 | static void __wait_cp_io(journal_t *journal, transaction_t *transaction) |
1da177e4 | 165 | { |
78ce89c9 | 166 | struct journal_head *jh; |
1da177e4 | 167 | struct buffer_head *bh; |
78ce89c9 JK |
168 | tid_t this_tid; |
169 | int released = 0; | |
170 | ||
171 | this_tid = transaction->t_tid; | |
172 | restart: | |
173 | /* Did somebody clean up the transaction in the meanwhile? */ | |
174 | if (journal->j_checkpoint_transactions != transaction || | |
175 | transaction->t_tid != this_tid) | |
176 | return; | |
177 | while (!released && transaction->t_checkpoint_io_list) { | |
178 | jh = transaction->t_checkpoint_io_list; | |
1da177e4 | 179 | bh = jh2bh(jh); |
78ce89c9 JK |
180 | if (!jbd_trylock_bh_state(bh)) { |
181 | jbd_sync_bh(journal, bh); | |
182 | spin_lock(&journal->j_list_lock); | |
183 | goto restart; | |
184 | } | |
1da177e4 LT |
185 | if (buffer_locked(bh)) { |
186 | atomic_inc(&bh->b_count); | |
187 | spin_unlock(&journal->j_list_lock); | |
78ce89c9 | 188 | jbd_unlock_bh_state(bh); |
1da177e4 LT |
189 | wait_on_buffer(bh); |
190 | /* the journal_head may have gone by now */ | |
191 | BUFFER_TRACE(bh, "brelse"); | |
192 | __brelse(bh); | |
78ce89c9 JK |
193 | spin_lock(&journal->j_list_lock); |
194 | goto restart; | |
1da177e4 | 195 | } |
1da177e4 | 196 | /* |
78ce89c9 JK |
197 | * Now in whatever state the buffer currently is, we know that |
198 | * it has been written out and so we can drop it from the list | |
1da177e4 | 199 | */ |
78ce89c9 JK |
200 | released = __journal_remove_checkpoint(jh); |
201 | jbd_unlock_bh_state(bh); | |
202 | journal_remove_journal_head(bh); | |
203 | __brelse(bh); | |
204 | } | |
1da177e4 LT |
205 | } |
206 | ||
207 | #define NR_BATCH 64 | |
208 | ||
209 | static void | |
210 | __flush_batch(journal_t *journal, struct buffer_head **bhs, int *batch_count) | |
211 | { | |
212 | int i; | |
213 | ||
26707699 | 214 | ll_rw_block(SWRITE, *batch_count, bhs); |
1da177e4 LT |
215 | for (i = 0; i < *batch_count; i++) { |
216 | struct buffer_head *bh = bhs[i]; | |
217 | clear_buffer_jwrite(bh); | |
218 | BUFFER_TRACE(bh, "brelse"); | |
219 | __brelse(bh); | |
220 | } | |
221 | *batch_count = 0; | |
222 | } | |
223 | ||
224 | /* | |
225 | * Try to flush one buffer from the checkpoint list to disk. | |
226 | * | |
227 | * Return 1 if something happened which requires us to abort the current | |
228 | * scan of the checkpoint list. | |
229 | * | |
78ce89c9 | 230 | * Called with j_list_lock held and drops it if 1 is returned |
1da177e4 LT |
231 | * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it |
232 | */ | |
78ce89c9 JK |
233 | static int __process_buffer(journal_t *journal, struct journal_head *jh, |
234 | struct buffer_head **bhs, int *batch_count) | |
1da177e4 LT |
235 | { |
236 | struct buffer_head *bh = jh2bh(jh); | |
237 | int ret = 0; | |
238 | ||
78ce89c9 JK |
239 | if (buffer_locked(bh)) { |
240 | atomic_inc(&bh->b_count); | |
241 | spin_unlock(&journal->j_list_lock); | |
242 | jbd_unlock_bh_state(bh); | |
243 | wait_on_buffer(bh); | |
244 | /* the journal_head may have gone by now */ | |
245 | BUFFER_TRACE(bh, "brelse"); | |
246 | __brelse(bh); | |
247 | ret = 1; | |
248 | } else if (jh->b_transaction != NULL) { | |
249 | transaction_t *t = jh->b_transaction; | |
250 | tid_t tid = t->t_tid; | |
1da177e4 | 251 | |
78ce89c9 JK |
252 | spin_unlock(&journal->j_list_lock); |
253 | jbd_unlock_bh_state(bh); | |
254 | log_start_commit(journal, tid); | |
255 | log_wait_commit(journal, tid); | |
256 | ret = 1; | |
257 | } else if (!buffer_dirty(bh)) { | |
258 | J_ASSERT_JH(jh, !buffer_jbddirty(bh)); | |
259 | BUFFER_TRACE(bh, "remove from checkpoint"); | |
260 | __journal_remove_checkpoint(jh); | |
261 | spin_unlock(&journal->j_list_lock); | |
262 | jbd_unlock_bh_state(bh); | |
263 | journal_remove_journal_head(bh); | |
264 | __brelse(bh); | |
265 | ret = 1; | |
266 | } else { | |
1da177e4 LT |
267 | /* |
268 | * Important: we are about to write the buffer, and | |
269 | * possibly block, while still holding the journal lock. | |
270 | * We cannot afford to let the transaction logic start | |
271 | * messing around with this buffer before we write it to | |
272 | * disk, as that would break recoverability. | |
273 | */ | |
274 | BUFFER_TRACE(bh, "queue"); | |
275 | get_bh(bh); | |
276 | J_ASSERT_BH(bh, !buffer_jwrite(bh)); | |
277 | set_buffer_jwrite(bh); | |
278 | bhs[*batch_count] = bh; | |
78ce89c9 | 279 | __buffer_relink_io(jh); |
1da177e4 LT |
280 | jbd_unlock_bh_state(bh); |
281 | (*batch_count)++; | |
282 | if (*batch_count == NR_BATCH) { | |
78ce89c9 | 283 | spin_unlock(&journal->j_list_lock); |
1da177e4 LT |
284 | __flush_batch(journal, bhs, batch_count); |
285 | ret = 1; | |
286 | } | |
1da177e4 LT |
287 | } |
288 | return ret; | |
289 | } | |
290 | ||
291 | /* | |
78ce89c9 JK |
292 | * Perform an actual checkpoint. We take the first transaction on the |
293 | * list of transactions to be checkpointed and send all its buffers | |
294 | * to disk. We submit larger chunks of data at once. | |
1da177e4 | 295 | * |
1da177e4 LT |
296 | * The journal should be locked before calling this function. |
297 | */ | |
298 | int log_do_checkpoint(journal_t *journal) | |
299 | { | |
78ce89c9 JK |
300 | transaction_t *transaction; |
301 | tid_t this_tid; | |
1da177e4 | 302 | int result; |
1da177e4 LT |
303 | |
304 | jbd_debug(1, "Start checkpoint\n"); | |
305 | ||
306 | /* | |
307 | * First thing: if there are any transactions in the log which | |
308 | * don't need checkpointing, just eliminate them from the | |
309 | * journal straight away. | |
310 | */ | |
311 | result = cleanup_journal_tail(journal); | |
312 | jbd_debug(1, "cleanup_journal_tail returned %d\n", result); | |
313 | if (result <= 0) | |
314 | return result; | |
315 | ||
316 | /* | |
78ce89c9 JK |
317 | * OK, we need to start writing disk blocks. Take one transaction |
318 | * and write it. | |
1da177e4 | 319 | */ |
78ce89c9 JK |
320 | spin_lock(&journal->j_list_lock); |
321 | if (!journal->j_checkpoint_transactions) | |
322 | goto out; | |
323 | transaction = journal->j_checkpoint_transactions; | |
324 | this_tid = transaction->t_tid; | |
325 | restart: | |
1da177e4 | 326 | /* |
78ce89c9 JK |
327 | * If someone cleaned up this transaction while we slept, we're |
328 | * done (maybe it's a new transaction, but it fell at the same | |
329 | * address). | |
1da177e4 | 330 | */ |
78ce89c9 JK |
331 | if (journal->j_checkpoint_transactions == transaction && |
332 | transaction->t_tid == this_tid) { | |
333 | int batch_count = 0; | |
334 | struct buffer_head *bhs[NR_BATCH]; | |
335 | struct journal_head *jh; | |
336 | int retry = 0; | |
337 | ||
338 | while (!retry && transaction->t_checkpoint_list) { | |
1da177e4 LT |
339 | struct buffer_head *bh; |
340 | ||
78ce89c9 | 341 | jh = transaction->t_checkpoint_list; |
1da177e4 LT |
342 | bh = jh2bh(jh); |
343 | if (!jbd_trylock_bh_state(bh)) { | |
344 | jbd_sync_bh(journal, bh); | |
1da177e4 LT |
345 | retry = 1; |
346 | break; | |
347 | } | |
78ce89c9 JK |
348 | retry = __process_buffer(journal, jh, bhs,&batch_count); |
349 | if (!retry && lock_need_resched(&journal->j_list_lock)){ | |
350 | spin_unlock(&journal->j_list_lock); | |
1da177e4 LT |
351 | retry = 1; |
352 | break; | |
353 | } | |
78ce89c9 | 354 | } |
1da177e4 | 355 | |
00ea8145 | 356 | if (batch_count) { |
78ce89c9 JK |
357 | if (!retry) { |
358 | spin_unlock(&journal->j_list_lock); | |
359 | retry = 1; | |
360 | } | |
1da177e4 | 361 | __flush_batch(journal, bhs, &batch_count); |
00ea8145 | 362 | } |
1da177e4 | 363 | |
78ce89c9 JK |
364 | if (retry) { |
365 | spin_lock(&journal->j_list_lock); | |
366 | goto restart; | |
367 | } | |
1da177e4 | 368 | /* |
78ce89c9 JK |
369 | * Now we have cleaned up the first transaction's checkpoint |
370 | * list. Let's clean up the second one | |
7c8903f6 | 371 | */ |
78ce89c9 | 372 | __wait_cp_io(journal, transaction); |
1da177e4 | 373 | } |
78ce89c9 | 374 | out: |
1da177e4 LT |
375 | spin_unlock(&journal->j_list_lock); |
376 | result = cleanup_journal_tail(journal); | |
377 | if (result < 0) | |
378 | return result; | |
1da177e4 LT |
379 | return 0; |
380 | } | |
381 | ||
382 | /* | |
383 | * Check the list of checkpoint transactions for the journal to see if | |
384 | * we have already got rid of any since the last update of the log tail | |
385 | * in the journal superblock. If so, we can instantly roll the | |
386 | * superblock forward to remove those transactions from the log. | |
387 | * | |
388 | * Return <0 on error, 0 on success, 1 if there was nothing to clean up. | |
389 | * | |
390 | * Called with the journal lock held. | |
391 | * | |
392 | * This is the only part of the journaling code which really needs to be | |
393 | * aware of transaction aborts. Checkpointing involves writing to the | |
394 | * main filesystem area rather than to the journal, so it can proceed | |
395 | * even in abort state, but we must not update the journal superblock if | |
396 | * we have an abort error outstanding. | |
397 | */ | |
398 | ||
399 | int cleanup_journal_tail(journal_t *journal) | |
400 | { | |
401 | transaction_t * transaction; | |
402 | tid_t first_tid; | |
403 | unsigned long blocknr, freed; | |
404 | ||
405 | /* OK, work out the oldest transaction remaining in the log, and | |
406 | * the log block it starts at. | |
407 | * | |
408 | * If the log is now empty, we need to work out which is the | |
409 | * next transaction ID we will write, and where it will | |
410 | * start. */ | |
411 | ||
412 | spin_lock(&journal->j_state_lock); | |
413 | spin_lock(&journal->j_list_lock); | |
414 | transaction = journal->j_checkpoint_transactions; | |
415 | if (transaction) { | |
416 | first_tid = transaction->t_tid; | |
417 | blocknr = transaction->t_log_start; | |
418 | } else if ((transaction = journal->j_committing_transaction) != NULL) { | |
419 | first_tid = transaction->t_tid; | |
420 | blocknr = transaction->t_log_start; | |
421 | } else if ((transaction = journal->j_running_transaction) != NULL) { | |
422 | first_tid = transaction->t_tid; | |
423 | blocknr = journal->j_head; | |
424 | } else { | |
425 | first_tid = journal->j_transaction_sequence; | |
426 | blocknr = journal->j_head; | |
427 | } | |
428 | spin_unlock(&journal->j_list_lock); | |
429 | J_ASSERT(blocknr != 0); | |
430 | ||
431 | /* If the oldest pinned transaction is at the tail of the log | |
432 | already then there's not much we can do right now. */ | |
433 | if (journal->j_tail_sequence == first_tid) { | |
434 | spin_unlock(&journal->j_state_lock); | |
435 | return 1; | |
436 | } | |
437 | ||
438 | /* OK, update the superblock to recover the freed space. | |
439 | * Physical blocks come first: have we wrapped beyond the end of | |
440 | * the log? */ | |
441 | freed = blocknr - journal->j_tail; | |
442 | if (blocknr < journal->j_tail) | |
443 | freed = freed + journal->j_last - journal->j_first; | |
444 | ||
445 | jbd_debug(1, | |
446 | "Cleaning journal tail from %d to %d (offset %lu), " | |
447 | "freeing %lu\n", | |
448 | journal->j_tail_sequence, first_tid, blocknr, freed); | |
449 | ||
450 | journal->j_free += freed; | |
451 | journal->j_tail_sequence = first_tid; | |
452 | journal->j_tail = blocknr; | |
453 | spin_unlock(&journal->j_state_lock); | |
454 | if (!(journal->j_flags & JFS_ABORT)) | |
455 | journal_update_superblock(journal, 1); | |
456 | return 0; | |
457 | } | |
458 | ||
459 | ||
460 | /* Checkpoint list management */ | |
461 | ||
78ce89c9 JK |
462 | /* |
463 | * journal_clean_one_cp_list | |
464 | * | |
465 | * Find all the written-back checkpoint buffers in the given list and release them. | |
466 | * | |
467 | * Called with the journal locked. | |
468 | * Called with j_list_lock held. | |
469 | * Returns number of bufers reaped (for debug) | |
470 | */ | |
471 | ||
472 | static int journal_clean_one_cp_list(struct journal_head *jh, int *released) | |
473 | { | |
474 | struct journal_head *last_jh; | |
475 | struct journal_head *next_jh = jh; | |
476 | int ret, freed = 0; | |
477 | ||
478 | *released = 0; | |
479 | if (!jh) | |
480 | return 0; | |
481 | ||
482 | last_jh = jh->b_cpprev; | |
483 | do { | |
484 | jh = next_jh; | |
485 | next_jh = jh->b_cpnext; | |
486 | /* Use trylock because of the ranking */ | |
487 | if (jbd_trylock_bh_state(jh2bh(jh))) { | |
488 | ret = __try_to_free_cp_buf(jh); | |
489 | if (ret) { | |
490 | freed++; | |
491 | if (ret == 2) { | |
492 | *released = 1; | |
493 | return freed; | |
494 | } | |
495 | } | |
496 | } | |
497 | /* | |
498 | * This function only frees up some memory | |
499 | * if possible so we dont have an obligation | |
500 | * to finish processing. Bail out if preemption | |
501 | * requested: | |
502 | */ | |
503 | if (need_resched()) | |
504 | return freed; | |
505 | } while (jh != last_jh); | |
506 | ||
507 | return freed; | |
508 | } | |
509 | ||
1da177e4 LT |
510 | /* |
511 | * journal_clean_checkpoint_list | |
512 | * | |
513 | * Find all the written-back checkpoint buffers in the journal and release them. | |
514 | * | |
515 | * Called with the journal locked. | |
516 | * Called with j_list_lock held. | |
78ce89c9 | 517 | * Returns number of buffers reaped (for debug) |
1da177e4 LT |
518 | */ |
519 | ||
520 | int __journal_clean_checkpoint_list(journal_t *journal) | |
521 | { | |
522 | transaction_t *transaction, *last_transaction, *next_transaction; | |
7c8903f6 | 523 | int ret = 0; |
78ce89c9 | 524 | int released; |
1da177e4 LT |
525 | |
526 | transaction = journal->j_checkpoint_transactions; | |
78ce89c9 | 527 | if (!transaction) |
1da177e4 LT |
528 | goto out; |
529 | ||
530 | last_transaction = transaction->t_cpprev; | |
531 | next_transaction = transaction; | |
532 | do { | |
1da177e4 LT |
533 | transaction = next_transaction; |
534 | next_transaction = transaction->t_cpnext; | |
78ce89c9 JK |
535 | ret += journal_clean_one_cp_list(transaction-> |
536 | t_checkpoint_list, &released); | |
537 | /* | |
538 | * This function only frees up some memory if possible so we | |
539 | * dont have an obligation to finish processing. Bail out if | |
540 | * preemption requested: | |
541 | */ | |
542 | if (need_resched()) | |
543 | goto out; | |
544 | if (released) | |
545 | continue; | |
546 | /* | |
547 | * It is essential that we are as careful as in the case of | |
548 | * t_checkpoint_list with removing the buffer from the list as | |
549 | * we can possibly see not yet submitted buffers on io_list | |
550 | */ | |
551 | ret += journal_clean_one_cp_list(transaction-> | |
552 | t_checkpoint_io_list, &released); | |
553 | if (need_resched()) | |
554 | goto out; | |
1da177e4 LT |
555 | } while (transaction != last_transaction); |
556 | out: | |
557 | return ret; | |
558 | } | |
559 | ||
560 | /* | |
561 | * journal_remove_checkpoint: called after a buffer has been committed | |
562 | * to disk (either by being write-back flushed to disk, or being | |
563 | * committed to the log). | |
564 | * | |
565 | * We cannot safely clean a transaction out of the log until all of the | |
566 | * buffer updates committed in that transaction have safely been stored | |
567 | * elsewhere on disk. To achieve this, all of the buffers in a | |
568 | * transaction need to be maintained on the transaction's checkpoint | |
78ce89c9 | 569 | * lists until they have been rewritten, at which point this function is |
1da177e4 | 570 | * called to remove the buffer from the existing transaction's |
78ce89c9 JK |
571 | * checkpoint lists. |
572 | * | |
573 | * The function returns 1 if it frees the transaction, 0 otherwise. | |
1da177e4 LT |
574 | * |
575 | * This function is called with the journal locked. | |
576 | * This function is called with j_list_lock held. | |
78ce89c9 | 577 | * This function is called with jbd_lock_bh_state(jh2bh(jh)) |
1da177e4 LT |
578 | */ |
579 | ||
78ce89c9 | 580 | int __journal_remove_checkpoint(struct journal_head *jh) |
1da177e4 LT |
581 | { |
582 | transaction_t *transaction; | |
583 | journal_t *journal; | |
78ce89c9 | 584 | int ret = 0; |
1da177e4 LT |
585 | |
586 | JBUFFER_TRACE(jh, "entry"); | |
587 | ||
588 | if ((transaction = jh->b_cp_transaction) == NULL) { | |
589 | JBUFFER_TRACE(jh, "not on transaction"); | |
590 | goto out; | |
591 | } | |
592 | journal = transaction->t_journal; | |
593 | ||
594 | __buffer_unlink(jh); | |
78ce89c9 | 595 | jh->b_cp_transaction = NULL; |
1da177e4 | 596 | |
78ce89c9 JK |
597 | if (transaction->t_checkpoint_list != NULL || |
598 | transaction->t_checkpoint_io_list != NULL) | |
1da177e4 LT |
599 | goto out; |
600 | JBUFFER_TRACE(jh, "transaction has no more buffers"); | |
601 | ||
602 | /* | |
603 | * There is one special case to worry about: if we have just pulled the | |
604 | * buffer off a committing transaction's forget list, then even if the | |
605 | * checkpoint list is empty, the transaction obviously cannot be | |
606 | * dropped! | |
607 | * | |
608 | * The locking here around j_committing_transaction is a bit sleazy. | |
609 | * See the comment at the end of journal_commit_transaction(). | |
610 | */ | |
611 | if (transaction == journal->j_committing_transaction) { | |
612 | JBUFFER_TRACE(jh, "belongs to committing transaction"); | |
613 | goto out; | |
614 | } | |
615 | ||
616 | /* OK, that was the last buffer for the transaction: we can now | |
617 | safely remove this transaction from the log */ | |
618 | ||
619 | __journal_drop_transaction(journal, transaction); | |
620 | ||
621 | /* Just in case anybody was waiting for more transactions to be | |
622 | checkpointed... */ | |
623 | wake_up(&journal->j_wait_logspace); | |
78ce89c9 | 624 | ret = 1; |
1da177e4 LT |
625 | out: |
626 | JBUFFER_TRACE(jh, "exit"); | |
78ce89c9 | 627 | return ret; |
1da177e4 LT |
628 | } |
629 | ||
630 | /* | |
631 | * journal_insert_checkpoint: put a committed buffer onto a checkpoint | |
632 | * list so that we know when it is safe to clean the transaction out of | |
633 | * the log. | |
634 | * | |
635 | * Called with the journal locked. | |
636 | * Called with j_list_lock held. | |
637 | */ | |
638 | void __journal_insert_checkpoint(struct journal_head *jh, | |
639 | transaction_t *transaction) | |
640 | { | |
641 | JBUFFER_TRACE(jh, "entry"); | |
642 | J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh))); | |
643 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
644 | ||
645 | jh->b_cp_transaction = transaction; | |
646 | ||
647 | if (!transaction->t_checkpoint_list) { | |
648 | jh->b_cpnext = jh->b_cpprev = jh; | |
649 | } else { | |
650 | jh->b_cpnext = transaction->t_checkpoint_list; | |
651 | jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev; | |
652 | jh->b_cpprev->b_cpnext = jh; | |
653 | jh->b_cpnext->b_cpprev = jh; | |
654 | } | |
655 | transaction->t_checkpoint_list = jh; | |
656 | } | |
657 | ||
658 | /* | |
659 | * We've finished with this transaction structure: adios... | |
660 | * | |
661 | * The transaction must have no links except for the checkpoint by this | |
662 | * point. | |
663 | * | |
664 | * Called with the journal locked. | |
665 | * Called with j_list_lock held. | |
666 | */ | |
667 | ||
668 | void __journal_drop_transaction(journal_t *journal, transaction_t *transaction) | |
669 | { | |
670 | assert_spin_locked(&journal->j_list_lock); | |
671 | if (transaction->t_cpnext) { | |
672 | transaction->t_cpnext->t_cpprev = transaction->t_cpprev; | |
673 | transaction->t_cpprev->t_cpnext = transaction->t_cpnext; | |
674 | if (journal->j_checkpoint_transactions == transaction) | |
675 | journal->j_checkpoint_transactions = | |
676 | transaction->t_cpnext; | |
677 | if (journal->j_checkpoint_transactions == transaction) | |
678 | journal->j_checkpoint_transactions = NULL; | |
679 | } | |
680 | ||
681 | J_ASSERT(transaction->t_state == T_FINISHED); | |
682 | J_ASSERT(transaction->t_buffers == NULL); | |
683 | J_ASSERT(transaction->t_sync_datalist == NULL); | |
684 | J_ASSERT(transaction->t_forget == NULL); | |
685 | J_ASSERT(transaction->t_iobuf_list == NULL); | |
686 | J_ASSERT(transaction->t_shadow_list == NULL); | |
687 | J_ASSERT(transaction->t_log_list == NULL); | |
688 | J_ASSERT(transaction->t_checkpoint_list == NULL); | |
78ce89c9 | 689 | J_ASSERT(transaction->t_checkpoint_io_list == NULL); |
1da177e4 LT |
690 | J_ASSERT(transaction->t_updates == 0); |
691 | J_ASSERT(journal->j_committing_transaction != transaction); | |
692 | J_ASSERT(journal->j_running_transaction != transaction); | |
693 | ||
694 | jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid); | |
695 | kfree(transaction); | |
696 | } |