Commit | Line | Data |
---|---|---|
1da177e4 | 1 | /* |
58862699 | 2 | * linux/fs/jbd/checkpoint.c |
ae6ddcc5 | 3 | * |
1da177e4 LT |
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 | * | |
ae6ddcc5 MC |
12 | * Checkpoint routines for the generic filesystem journaling code. |
13 | * Part of the ext2fs journaling system. | |
1da177e4 LT |
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) | |
e7ab8d65 | 148 | __releases(journal->j_list_lock) |
1da177e4 LT |
149 | { |
150 | get_bh(bh); | |
151 | spin_unlock(&journal->j_list_lock); | |
152 | jbd_lock_bh_state(bh); | |
153 | jbd_unlock_bh_state(bh); | |
154 | put_bh(bh); | |
155 | } | |
156 | ||
157 | /* | |
78ce89c9 JK |
158 | * Clean up transaction's list of buffers submitted for io. |
159 | * We wait for any pending IO to complete and remove any clean | |
160 | * buffers. Note that we take the buffers in the opposite ordering | |
161 | * from the one in which they were submitted for IO. | |
1da177e4 LT |
162 | * |
163 | * Called with j_list_lock held. | |
164 | */ | |
78ce89c9 | 165 | static void __wait_cp_io(journal_t *journal, transaction_t *transaction) |
1da177e4 | 166 | { |
78ce89c9 | 167 | struct journal_head *jh; |
1da177e4 | 168 | struct buffer_head *bh; |
78ce89c9 JK |
169 | tid_t this_tid; |
170 | int released = 0; | |
171 | ||
172 | this_tid = transaction->t_tid; | |
173 | restart: | |
174 | /* Did somebody clean up the transaction in the meanwhile? */ | |
175 | if (journal->j_checkpoint_transactions != transaction || | |
176 | transaction->t_tid != this_tid) | |
177 | return; | |
178 | while (!released && transaction->t_checkpoint_io_list) { | |
179 | jh = transaction->t_checkpoint_io_list; | |
1da177e4 | 180 | bh = jh2bh(jh); |
78ce89c9 JK |
181 | if (!jbd_trylock_bh_state(bh)) { |
182 | jbd_sync_bh(journal, bh); | |
183 | spin_lock(&journal->j_list_lock); | |
184 | goto restart; | |
185 | } | |
1da177e4 LT |
186 | if (buffer_locked(bh)) { |
187 | atomic_inc(&bh->b_count); | |
188 | spin_unlock(&journal->j_list_lock); | |
78ce89c9 | 189 | jbd_unlock_bh_state(bh); |
1da177e4 LT |
190 | wait_on_buffer(bh); |
191 | /* the journal_head may have gone by now */ | |
192 | BUFFER_TRACE(bh, "brelse"); | |
193 | __brelse(bh); | |
78ce89c9 JK |
194 | spin_lock(&journal->j_list_lock); |
195 | goto restart; | |
1da177e4 | 196 | } |
1da177e4 | 197 | /* |
78ce89c9 JK |
198 | * Now in whatever state the buffer currently is, we know that |
199 | * it has been written out and so we can drop it from the list | |
1da177e4 | 200 | */ |
78ce89c9 JK |
201 | released = __journal_remove_checkpoint(jh); |
202 | jbd_unlock_bh_state(bh); | |
203 | journal_remove_journal_head(bh); | |
204 | __brelse(bh); | |
205 | } | |
1da177e4 LT |
206 | } |
207 | ||
208 | #define NR_BATCH 64 | |
209 | ||
210 | static void | |
211 | __flush_batch(journal_t *journal, struct buffer_head **bhs, int *batch_count) | |
212 | { | |
213 | int i; | |
214 | ||
26707699 | 215 | ll_rw_block(SWRITE, *batch_count, bhs); |
1da177e4 LT |
216 | for (i = 0; i < *batch_count; i++) { |
217 | struct buffer_head *bh = bhs[i]; | |
218 | clear_buffer_jwrite(bh); | |
219 | BUFFER_TRACE(bh, "brelse"); | |
220 | __brelse(bh); | |
221 | } | |
222 | *batch_count = 0; | |
223 | } | |
224 | ||
225 | /* | |
226 | * Try to flush one buffer from the checkpoint list to disk. | |
227 | * | |
228 | * Return 1 if something happened which requires us to abort the current | |
ae6ddcc5 | 229 | * scan of the checkpoint list. |
1da177e4 | 230 | * |
78ce89c9 | 231 | * Called with j_list_lock held and drops it if 1 is returned |
1da177e4 LT |
232 | * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it |
233 | */ | |
78ce89c9 JK |
234 | static int __process_buffer(journal_t *journal, struct journal_head *jh, |
235 | struct buffer_head **bhs, int *batch_count) | |
1da177e4 LT |
236 | { |
237 | struct buffer_head *bh = jh2bh(jh); | |
238 | int ret = 0; | |
239 | ||
78ce89c9 JK |
240 | if (buffer_locked(bh)) { |
241 | atomic_inc(&bh->b_count); | |
242 | spin_unlock(&journal->j_list_lock); | |
243 | jbd_unlock_bh_state(bh); | |
244 | wait_on_buffer(bh); | |
245 | /* the journal_head may have gone by now */ | |
246 | BUFFER_TRACE(bh, "brelse"); | |
247 | __brelse(bh); | |
248 | ret = 1; | |
249 | } else if (jh->b_transaction != NULL) { | |
250 | transaction_t *t = jh->b_transaction; | |
251 | tid_t tid = t->t_tid; | |
1da177e4 | 252 | |
78ce89c9 JK |
253 | spin_unlock(&journal->j_list_lock); |
254 | jbd_unlock_bh_state(bh); | |
255 | log_start_commit(journal, tid); | |
256 | log_wait_commit(journal, tid); | |
257 | ret = 1; | |
258 | } else if (!buffer_dirty(bh)) { | |
259 | J_ASSERT_JH(jh, !buffer_jbddirty(bh)); | |
260 | BUFFER_TRACE(bh, "remove from checkpoint"); | |
261 | __journal_remove_checkpoint(jh); | |
262 | spin_unlock(&journal->j_list_lock); | |
263 | jbd_unlock_bh_state(bh); | |
264 | journal_remove_journal_head(bh); | |
265 | __brelse(bh); | |
266 | ret = 1; | |
267 | } else { | |
1da177e4 LT |
268 | /* |
269 | * Important: we are about to write the buffer, and | |
270 | * possibly block, while still holding the journal lock. | |
271 | * We cannot afford to let the transaction logic start | |
272 | * messing around with this buffer before we write it to | |
ae6ddcc5 | 273 | * disk, as that would break recoverability. |
1da177e4 LT |
274 | */ |
275 | BUFFER_TRACE(bh, "queue"); | |
276 | get_bh(bh); | |
277 | J_ASSERT_BH(bh, !buffer_jwrite(bh)); | |
278 | set_buffer_jwrite(bh); | |
279 | bhs[*batch_count] = bh; | |
78ce89c9 | 280 | __buffer_relink_io(jh); |
1da177e4 LT |
281 | jbd_unlock_bh_state(bh); |
282 | (*batch_count)++; | |
283 | if (*batch_count == NR_BATCH) { | |
78ce89c9 | 284 | spin_unlock(&journal->j_list_lock); |
1da177e4 LT |
285 | __flush_batch(journal, bhs, batch_count); |
286 | ret = 1; | |
287 | } | |
1da177e4 LT |
288 | } |
289 | return ret; | |
290 | } | |
291 | ||
292 | /* | |
78ce89c9 JK |
293 | * Perform an actual checkpoint. We take the first transaction on the |
294 | * list of transactions to be checkpointed and send all its buffers | |
295 | * to disk. We submit larger chunks of data at once. | |
ae6ddcc5 | 296 | * |
1da177e4 LT |
297 | * The journal should be locked before calling this function. |
298 | */ | |
299 | int log_do_checkpoint(journal_t *journal) | |
300 | { | |
78ce89c9 JK |
301 | transaction_t *transaction; |
302 | tid_t this_tid; | |
1da177e4 | 303 | int result; |
1da177e4 LT |
304 | |
305 | jbd_debug(1, "Start checkpoint\n"); | |
306 | ||
ae6ddcc5 | 307 | /* |
1da177e4 LT |
308 | * First thing: if there are any transactions in the log which |
309 | * don't need checkpointing, just eliminate them from the | |
ae6ddcc5 | 310 | * journal straight away. |
1da177e4 LT |
311 | */ |
312 | result = cleanup_journal_tail(journal); | |
313 | jbd_debug(1, "cleanup_journal_tail returned %d\n", result); | |
314 | if (result <= 0) | |
315 | return result; | |
316 | ||
317 | /* | |
78ce89c9 JK |
318 | * OK, we need to start writing disk blocks. Take one transaction |
319 | * and write it. | |
1da177e4 | 320 | */ |
78ce89c9 JK |
321 | spin_lock(&journal->j_list_lock); |
322 | if (!journal->j_checkpoint_transactions) | |
323 | goto out; | |
324 | transaction = journal->j_checkpoint_transactions; | |
325 | this_tid = transaction->t_tid; | |
326 | restart: | |
1da177e4 | 327 | /* |
78ce89c9 JK |
328 | * If someone cleaned up this transaction while we slept, we're |
329 | * done (maybe it's a new transaction, but it fell at the same | |
330 | * address). | |
1da177e4 | 331 | */ |
78ce89c9 JK |
332 | if (journal->j_checkpoint_transactions == transaction && |
333 | transaction->t_tid == this_tid) { | |
334 | int batch_count = 0; | |
335 | struct buffer_head *bhs[NR_BATCH]; | |
336 | struct journal_head *jh; | |
337 | int retry = 0; | |
338 | ||
339 | while (!retry && transaction->t_checkpoint_list) { | |
1da177e4 LT |
340 | struct buffer_head *bh; |
341 | ||
78ce89c9 | 342 | jh = transaction->t_checkpoint_list; |
1da177e4 LT |
343 | bh = jh2bh(jh); |
344 | if (!jbd_trylock_bh_state(bh)) { | |
345 | jbd_sync_bh(journal, bh); | |
1da177e4 LT |
346 | retry = 1; |
347 | break; | |
348 | } | |
78ce89c9 | 349 | retry = __process_buffer(journal, jh, bhs,&batch_count); |
95c354fe NP |
350 | if (!retry && (need_resched() || |
351 | spin_needbreak(&journal->j_list_lock))) { | |
78ce89c9 | 352 | spin_unlock(&journal->j_list_lock); |
1da177e4 LT |
353 | retry = 1; |
354 | break; | |
355 | } | |
78ce89c9 | 356 | } |
1da177e4 | 357 | |
00ea8145 | 358 | if (batch_count) { |
78ce89c9 JK |
359 | if (!retry) { |
360 | spin_unlock(&journal->j_list_lock); | |
361 | retry = 1; | |
362 | } | |
1da177e4 | 363 | __flush_batch(journal, bhs, &batch_count); |
00ea8145 | 364 | } |
1da177e4 | 365 | |
78ce89c9 JK |
366 | if (retry) { |
367 | spin_lock(&journal->j_list_lock); | |
368 | goto restart; | |
369 | } | |
1da177e4 | 370 | /* |
78ce89c9 JK |
371 | * Now we have cleaned up the first transaction's checkpoint |
372 | * list. Let's clean up the second one | |
7c8903f6 | 373 | */ |
78ce89c9 | 374 | __wait_cp_io(journal, transaction); |
1da177e4 | 375 | } |
78ce89c9 | 376 | out: |
1da177e4 LT |
377 | spin_unlock(&journal->j_list_lock); |
378 | result = cleanup_journal_tail(journal); | |
379 | if (result < 0) | |
380 | return result; | |
1da177e4 LT |
381 | return 0; |
382 | } | |
383 | ||
384 | /* | |
385 | * Check the list of checkpoint transactions for the journal to see if | |
386 | * we have already got rid of any since the last update of the log tail | |
387 | * in the journal superblock. If so, we can instantly roll the | |
388 | * superblock forward to remove those transactions from the log. | |
ae6ddcc5 | 389 | * |
1da177e4 | 390 | * Return <0 on error, 0 on success, 1 if there was nothing to clean up. |
ae6ddcc5 | 391 | * |
1da177e4 LT |
392 | * Called with the journal lock held. |
393 | * | |
394 | * This is the only part of the journaling code which really needs to be | |
395 | * aware of transaction aborts. Checkpointing involves writing to the | |
396 | * main filesystem area rather than to the journal, so it can proceed | |
397 | * even in abort state, but we must not update the journal superblock if | |
398 | * we have an abort error outstanding. | |
399 | */ | |
400 | ||
401 | int cleanup_journal_tail(journal_t *journal) | |
402 | { | |
403 | transaction_t * transaction; | |
404 | tid_t first_tid; | |
405 | unsigned long blocknr, freed; | |
406 | ||
407 | /* OK, work out the oldest transaction remaining in the log, and | |
ae6ddcc5 MC |
408 | * the log block it starts at. |
409 | * | |
1da177e4 LT |
410 | * If the log is now empty, we need to work out which is the |
411 | * next transaction ID we will write, and where it will | |
412 | * start. */ | |
413 | ||
414 | spin_lock(&journal->j_state_lock); | |
415 | spin_lock(&journal->j_list_lock); | |
416 | transaction = journal->j_checkpoint_transactions; | |
417 | if (transaction) { | |
418 | first_tid = transaction->t_tid; | |
419 | blocknr = transaction->t_log_start; | |
420 | } else if ((transaction = journal->j_committing_transaction) != NULL) { | |
421 | first_tid = transaction->t_tid; | |
422 | blocknr = transaction->t_log_start; | |
423 | } else if ((transaction = journal->j_running_transaction) != NULL) { | |
424 | first_tid = transaction->t_tid; | |
425 | blocknr = journal->j_head; | |
426 | } else { | |
427 | first_tid = journal->j_transaction_sequence; | |
428 | blocknr = journal->j_head; | |
429 | } | |
430 | spin_unlock(&journal->j_list_lock); | |
431 | J_ASSERT(blocknr != 0); | |
432 | ||
433 | /* If the oldest pinned transaction is at the tail of the log | |
434 | already then there's not much we can do right now. */ | |
435 | if (journal->j_tail_sequence == first_tid) { | |
436 | spin_unlock(&journal->j_state_lock); | |
437 | return 1; | |
438 | } | |
439 | ||
440 | /* OK, update the superblock to recover the freed space. | |
441 | * Physical blocks come first: have we wrapped beyond the end of | |
442 | * the log? */ | |
443 | freed = blocknr - journal->j_tail; | |
444 | if (blocknr < journal->j_tail) | |
445 | freed = freed + journal->j_last - journal->j_first; | |
446 | ||
447 | jbd_debug(1, | |
448 | "Cleaning journal tail from %d to %d (offset %lu), " | |
449 | "freeing %lu\n", | |
450 | journal->j_tail_sequence, first_tid, blocknr, freed); | |
451 | ||
452 | journal->j_free += freed; | |
453 | journal->j_tail_sequence = first_tid; | |
454 | journal->j_tail = blocknr; | |
455 | spin_unlock(&journal->j_state_lock); | |
456 | if (!(journal->j_flags & JFS_ABORT)) | |
457 | journal_update_superblock(journal, 1); | |
458 | return 0; | |
459 | } | |
460 | ||
461 | ||
462 | /* Checkpoint list management */ | |
463 | ||
78ce89c9 JK |
464 | /* |
465 | * journal_clean_one_cp_list | |
466 | * | |
467 | * Find all the written-back checkpoint buffers in the given list and release them. | |
468 | * | |
469 | * Called with the journal locked. | |
470 | * Called with j_list_lock held. | |
471 | * Returns number of bufers reaped (for debug) | |
472 | */ | |
473 | ||
474 | static int journal_clean_one_cp_list(struct journal_head *jh, int *released) | |
475 | { | |
476 | struct journal_head *last_jh; | |
477 | struct journal_head *next_jh = jh; | |
478 | int ret, freed = 0; | |
479 | ||
480 | *released = 0; | |
481 | if (!jh) | |
482 | return 0; | |
483 | ||
e9ad5620 | 484 | last_jh = jh->b_cpprev; |
78ce89c9 JK |
485 | do { |
486 | jh = next_jh; | |
487 | next_jh = jh->b_cpnext; | |
488 | /* Use trylock because of the ranking */ | |
489 | if (jbd_trylock_bh_state(jh2bh(jh))) { | |
490 | ret = __try_to_free_cp_buf(jh); | |
491 | if (ret) { | |
492 | freed++; | |
493 | if (ret == 2) { | |
494 | *released = 1; | |
495 | return freed; | |
496 | } | |
497 | } | |
498 | } | |
499 | /* | |
500 | * This function only frees up some memory | |
501 | * if possible so we dont have an obligation | |
502 | * to finish processing. Bail out if preemption | |
503 | * requested: | |
504 | */ | |
505 | if (need_resched()) | |
506 | return freed; | |
507 | } while (jh != last_jh); | |
508 | ||
509 | return freed; | |
510 | } | |
511 | ||
1da177e4 LT |
512 | /* |
513 | * journal_clean_checkpoint_list | |
514 | * | |
515 | * Find all the written-back checkpoint buffers in the journal and release them. | |
516 | * | |
517 | * Called with the journal locked. | |
518 | * Called with j_list_lock held. | |
78ce89c9 | 519 | * Returns number of buffers reaped (for debug) |
1da177e4 LT |
520 | */ |
521 | ||
522 | int __journal_clean_checkpoint_list(journal_t *journal) | |
523 | { | |
524 | transaction_t *transaction, *last_transaction, *next_transaction; | |
7c8903f6 | 525 | int ret = 0; |
78ce89c9 | 526 | int released; |
1da177e4 LT |
527 | |
528 | transaction = journal->j_checkpoint_transactions; | |
78ce89c9 | 529 | if (!transaction) |
1da177e4 LT |
530 | goto out; |
531 | ||
532 | last_transaction = transaction->t_cpprev; | |
533 | next_transaction = transaction; | |
534 | do { | |
1da177e4 LT |
535 | transaction = next_transaction; |
536 | next_transaction = transaction->t_cpnext; | |
78ce89c9 JK |
537 | ret += journal_clean_one_cp_list(transaction-> |
538 | t_checkpoint_list, &released); | |
539 | /* | |
540 | * This function only frees up some memory if possible so we | |
541 | * dont have an obligation to finish processing. Bail out if | |
542 | * preemption requested: | |
543 | */ | |
544 | if (need_resched()) | |
545 | goto out; | |
546 | if (released) | |
547 | continue; | |
548 | /* | |
549 | * It is essential that we are as careful as in the case of | |
550 | * t_checkpoint_list with removing the buffer from the list as | |
551 | * we can possibly see not yet submitted buffers on io_list | |
552 | */ | |
553 | ret += journal_clean_one_cp_list(transaction-> | |
554 | t_checkpoint_io_list, &released); | |
555 | if (need_resched()) | |
556 | goto out; | |
1da177e4 LT |
557 | } while (transaction != last_transaction); |
558 | out: | |
559 | return ret; | |
560 | } | |
561 | ||
ae6ddcc5 | 562 | /* |
1da177e4 LT |
563 | * journal_remove_checkpoint: called after a buffer has been committed |
564 | * to disk (either by being write-back flushed to disk, or being | |
565 | * committed to the log). | |
566 | * | |
567 | * We cannot safely clean a transaction out of the log until all of the | |
568 | * buffer updates committed in that transaction have safely been stored | |
569 | * elsewhere on disk. To achieve this, all of the buffers in a | |
570 | * transaction need to be maintained on the transaction's checkpoint | |
78ce89c9 | 571 | * lists until they have been rewritten, at which point this function is |
1da177e4 | 572 | * called to remove the buffer from the existing transaction's |
78ce89c9 JK |
573 | * checkpoint lists. |
574 | * | |
575 | * The function returns 1 if it frees the transaction, 0 otherwise. | |
1da177e4 LT |
576 | * |
577 | * This function is called with the journal locked. | |
578 | * This function is called with j_list_lock held. | |
78ce89c9 | 579 | * This function is called with jbd_lock_bh_state(jh2bh(jh)) |
1da177e4 LT |
580 | */ |
581 | ||
78ce89c9 | 582 | int __journal_remove_checkpoint(struct journal_head *jh) |
1da177e4 LT |
583 | { |
584 | transaction_t *transaction; | |
585 | journal_t *journal; | |
78ce89c9 | 586 | int ret = 0; |
1da177e4 LT |
587 | |
588 | JBUFFER_TRACE(jh, "entry"); | |
589 | ||
590 | if ((transaction = jh->b_cp_transaction) == NULL) { | |
591 | JBUFFER_TRACE(jh, "not on transaction"); | |
592 | goto out; | |
593 | } | |
594 | journal = transaction->t_journal; | |
595 | ||
596 | __buffer_unlink(jh); | |
78ce89c9 | 597 | jh->b_cp_transaction = NULL; |
1da177e4 | 598 | |
78ce89c9 JK |
599 | if (transaction->t_checkpoint_list != NULL || |
600 | transaction->t_checkpoint_io_list != NULL) | |
1da177e4 LT |
601 | goto out; |
602 | JBUFFER_TRACE(jh, "transaction has no more buffers"); | |
603 | ||
604 | /* | |
605 | * There is one special case to worry about: if we have just pulled the | |
d4beaf4a JK |
606 | * buffer off a running or committing transaction's checkpoing list, |
607 | * then even if the checkpoint list is empty, the transaction obviously | |
608 | * cannot be dropped! | |
1da177e4 | 609 | * |
d4beaf4a | 610 | * The locking here around t_state is a bit sleazy. |
1da177e4 LT |
611 | * See the comment at the end of journal_commit_transaction(). |
612 | */ | |
d4beaf4a JK |
613 | if (transaction->t_state != T_FINISHED) { |
614 | JBUFFER_TRACE(jh, "belongs to running/committing transaction"); | |
1da177e4 LT |
615 | goto out; |
616 | } | |
617 | ||
618 | /* OK, that was the last buffer for the transaction: we can now | |
619 | safely remove this transaction from the log */ | |
620 | ||
621 | __journal_drop_transaction(journal, transaction); | |
622 | ||
623 | /* Just in case anybody was waiting for more transactions to be | |
624 | checkpointed... */ | |
625 | wake_up(&journal->j_wait_logspace); | |
78ce89c9 | 626 | ret = 1; |
1da177e4 LT |
627 | out: |
628 | JBUFFER_TRACE(jh, "exit"); | |
78ce89c9 | 629 | return ret; |
1da177e4 LT |
630 | } |
631 | ||
632 | /* | |
633 | * journal_insert_checkpoint: put a committed buffer onto a checkpoint | |
634 | * list so that we know when it is safe to clean the transaction out of | |
635 | * the log. | |
636 | * | |
637 | * Called with the journal locked. | |
638 | * Called with j_list_lock held. | |
639 | */ | |
ae6ddcc5 | 640 | void __journal_insert_checkpoint(struct journal_head *jh, |
1da177e4 LT |
641 | transaction_t *transaction) |
642 | { | |
643 | JBUFFER_TRACE(jh, "entry"); | |
644 | J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh))); | |
645 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
646 | ||
647 | jh->b_cp_transaction = transaction; | |
648 | ||
649 | if (!transaction->t_checkpoint_list) { | |
650 | jh->b_cpnext = jh->b_cpprev = jh; | |
651 | } else { | |
652 | jh->b_cpnext = transaction->t_checkpoint_list; | |
653 | jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev; | |
654 | jh->b_cpprev->b_cpnext = jh; | |
655 | jh->b_cpnext->b_cpprev = jh; | |
656 | } | |
657 | transaction->t_checkpoint_list = jh; | |
658 | } | |
659 | ||
660 | /* | |
661 | * We've finished with this transaction structure: adios... | |
ae6ddcc5 | 662 | * |
1da177e4 LT |
663 | * The transaction must have no links except for the checkpoint by this |
664 | * point. | |
665 | * | |
666 | * Called with the journal locked. | |
667 | * Called with j_list_lock held. | |
668 | */ | |
669 | ||
670 | void __journal_drop_transaction(journal_t *journal, transaction_t *transaction) | |
671 | { | |
672 | assert_spin_locked(&journal->j_list_lock); | |
673 | if (transaction->t_cpnext) { | |
674 | transaction->t_cpnext->t_cpprev = transaction->t_cpprev; | |
675 | transaction->t_cpprev->t_cpnext = transaction->t_cpnext; | |
676 | if (journal->j_checkpoint_transactions == transaction) | |
677 | journal->j_checkpoint_transactions = | |
678 | transaction->t_cpnext; | |
679 | if (journal->j_checkpoint_transactions == transaction) | |
680 | journal->j_checkpoint_transactions = NULL; | |
681 | } | |
682 | ||
683 | J_ASSERT(transaction->t_state == T_FINISHED); | |
684 | J_ASSERT(transaction->t_buffers == NULL); | |
685 | J_ASSERT(transaction->t_sync_datalist == NULL); | |
686 | J_ASSERT(transaction->t_forget == NULL); | |
687 | J_ASSERT(transaction->t_iobuf_list == NULL); | |
688 | J_ASSERT(transaction->t_shadow_list == NULL); | |
689 | J_ASSERT(transaction->t_log_list == NULL); | |
690 | J_ASSERT(transaction->t_checkpoint_list == NULL); | |
78ce89c9 | 691 | J_ASSERT(transaction->t_checkpoint_io_list == NULL); |
1da177e4 LT |
692 | J_ASSERT(transaction->t_updates == 0); |
693 | J_ASSERT(journal->j_committing_transaction != transaction); | |
694 | J_ASSERT(journal->j_running_transaction != transaction); | |
695 | ||
696 | jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid); | |
697 | kfree(transaction); | |
698 | } |