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 | ||
4afe9785 | 96 | if (jh->b_jlist == BJ_None && !buffer_locked(bh) && |
9f818b4a | 97 | !buffer_dirty(bh) && !buffer_write_io_error(bh)) { |
1da177e4 | 98 | JBUFFER_TRACE(jh, "remove from checkpoint list"); |
78ce89c9 | 99 | ret = __journal_remove_checkpoint(jh) + 1; |
1da177e4 LT |
100 | jbd_unlock_bh_state(bh); |
101 | journal_remove_journal_head(bh); | |
102 | BUFFER_TRACE(bh, "release"); | |
103 | __brelse(bh); | |
1da177e4 LT |
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; | |
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); | |
2c68ee75 | 126 | mutex_lock(&journal->j_checkpoint_mutex); |
1da177e4 LT |
127 | |
128 | /* | |
129 | * Test again, another process may have checkpointed while we | |
130 | * were waiting for the checkpoint lock | |
131 | */ | |
132 | spin_lock(&journal->j_state_lock); | |
133 | nblocks = jbd_space_needed(journal); | |
134 | if (__log_space_left(journal) < nblocks) { | |
135 | spin_unlock(&journal->j_state_lock); | |
136 | log_do_checkpoint(journal); | |
137 | spin_lock(&journal->j_state_lock); | |
138 | } | |
2c68ee75 | 139 | mutex_unlock(&journal->j_checkpoint_mutex); |
1da177e4 LT |
140 | } |
141 | } | |
142 | ||
143 | /* | |
144 | * We were unable to perform jbd_trylock_bh_state() inside j_list_lock. | |
145 | * The caller must restart a list walk. Wait for someone else to run | |
146 | * jbd_unlock_bh_state(). | |
147 | */ | |
148 | static void jbd_sync_bh(journal_t *journal, struct buffer_head *bh) | |
e7ab8d65 | 149 | __releases(journal->j_list_lock) |
1da177e4 LT |
150 | { |
151 | get_bh(bh); | |
152 | spin_unlock(&journal->j_list_lock); | |
153 | jbd_lock_bh_state(bh); | |
154 | jbd_unlock_bh_state(bh); | |
155 | put_bh(bh); | |
156 | } | |
157 | ||
158 | /* | |
78ce89c9 JK |
159 | * Clean up transaction's list of buffers submitted for io. |
160 | * We wait for any pending IO to complete and remove any clean | |
161 | * buffers. Note that we take the buffers in the opposite ordering | |
162 | * from the one in which they were submitted for IO. | |
1da177e4 | 163 | * |
4afe9785 HK |
164 | * Return 0 on success, and return <0 if some buffers have failed |
165 | * to be written out. | |
166 | * | |
1da177e4 LT |
167 | * Called with j_list_lock held. |
168 | */ | |
4afe9785 | 169 | static int __wait_cp_io(journal_t *journal, transaction_t *transaction) |
1da177e4 | 170 | { |
78ce89c9 | 171 | struct journal_head *jh; |
1da177e4 | 172 | struct buffer_head *bh; |
78ce89c9 JK |
173 | tid_t this_tid; |
174 | int released = 0; | |
4afe9785 | 175 | int ret = 0; |
78ce89c9 JK |
176 | |
177 | this_tid = transaction->t_tid; | |
178 | restart: | |
179 | /* Did somebody clean up the transaction in the meanwhile? */ | |
180 | if (journal->j_checkpoint_transactions != transaction || | |
181 | transaction->t_tid != this_tid) | |
4afe9785 | 182 | return ret; |
78ce89c9 JK |
183 | while (!released && transaction->t_checkpoint_io_list) { |
184 | jh = transaction->t_checkpoint_io_list; | |
1da177e4 | 185 | bh = jh2bh(jh); |
78ce89c9 JK |
186 | if (!jbd_trylock_bh_state(bh)) { |
187 | jbd_sync_bh(journal, bh); | |
188 | spin_lock(&journal->j_list_lock); | |
189 | goto restart; | |
190 | } | |
1da177e4 LT |
191 | if (buffer_locked(bh)) { |
192 | atomic_inc(&bh->b_count); | |
193 | spin_unlock(&journal->j_list_lock); | |
78ce89c9 | 194 | jbd_unlock_bh_state(bh); |
1da177e4 LT |
195 | wait_on_buffer(bh); |
196 | /* the journal_head may have gone by now */ | |
197 | BUFFER_TRACE(bh, "brelse"); | |
198 | __brelse(bh); | |
78ce89c9 JK |
199 | spin_lock(&journal->j_list_lock); |
200 | goto restart; | |
1da177e4 | 201 | } |
9f818b4a | 202 | if (unlikely(buffer_write_io_error(bh))) |
4afe9785 HK |
203 | ret = -EIO; |
204 | ||
1da177e4 | 205 | /* |
78ce89c9 JK |
206 | * Now in whatever state the buffer currently is, we know that |
207 | * it has been written out and so we can drop it from the list | |
1da177e4 | 208 | */ |
78ce89c9 JK |
209 | released = __journal_remove_checkpoint(jh); |
210 | jbd_unlock_bh_state(bh); | |
211 | journal_remove_journal_head(bh); | |
212 | __brelse(bh); | |
213 | } | |
4afe9785 HK |
214 | |
215 | return ret; | |
1da177e4 LT |
216 | } |
217 | ||
218 | #define NR_BATCH 64 | |
219 | ||
220 | static void | |
221 | __flush_batch(journal_t *journal, struct buffer_head **bhs, int *batch_count) | |
222 | { | |
223 | int i; | |
224 | ||
26707699 | 225 | ll_rw_block(SWRITE, *batch_count, bhs); |
1da177e4 LT |
226 | for (i = 0; i < *batch_count; i++) { |
227 | struct buffer_head *bh = bhs[i]; | |
228 | clear_buffer_jwrite(bh); | |
229 | BUFFER_TRACE(bh, "brelse"); | |
230 | __brelse(bh); | |
231 | } | |
232 | *batch_count = 0; | |
233 | } | |
234 | ||
235 | /* | |
236 | * Try to flush one buffer from the checkpoint list to disk. | |
237 | * | |
238 | * Return 1 if something happened which requires us to abort the current | |
4afe9785 HK |
239 | * scan of the checkpoint list. Return <0 if the buffer has failed to |
240 | * be written out. | |
1da177e4 | 241 | * |
78ce89c9 | 242 | * Called with j_list_lock held and drops it if 1 is returned |
1da177e4 LT |
243 | * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it |
244 | */ | |
78ce89c9 JK |
245 | static int __process_buffer(journal_t *journal, struct journal_head *jh, |
246 | struct buffer_head **bhs, int *batch_count) | |
1da177e4 LT |
247 | { |
248 | struct buffer_head *bh = jh2bh(jh); | |
249 | int ret = 0; | |
250 | ||
78ce89c9 JK |
251 | if (buffer_locked(bh)) { |
252 | atomic_inc(&bh->b_count); | |
253 | spin_unlock(&journal->j_list_lock); | |
254 | jbd_unlock_bh_state(bh); | |
255 | wait_on_buffer(bh); | |
256 | /* the journal_head may have gone by now */ | |
257 | BUFFER_TRACE(bh, "brelse"); | |
258 | __brelse(bh); | |
259 | ret = 1; | |
260 | } else if (jh->b_transaction != NULL) { | |
261 | transaction_t *t = jh->b_transaction; | |
262 | tid_t tid = t->t_tid; | |
1da177e4 | 263 | |
78ce89c9 JK |
264 | spin_unlock(&journal->j_list_lock); |
265 | jbd_unlock_bh_state(bh); | |
266 | log_start_commit(journal, tid); | |
267 | log_wait_commit(journal, tid); | |
268 | ret = 1; | |
269 | } else if (!buffer_dirty(bh)) { | |
4afe9785 | 270 | ret = 1; |
9f818b4a | 271 | if (unlikely(buffer_write_io_error(bh))) |
4afe9785 | 272 | ret = -EIO; |
78ce89c9 JK |
273 | J_ASSERT_JH(jh, !buffer_jbddirty(bh)); |
274 | BUFFER_TRACE(bh, "remove from checkpoint"); | |
275 | __journal_remove_checkpoint(jh); | |
276 | spin_unlock(&journal->j_list_lock); | |
277 | jbd_unlock_bh_state(bh); | |
278 | journal_remove_journal_head(bh); | |
279 | __brelse(bh); | |
78ce89c9 | 280 | } else { |
1da177e4 LT |
281 | /* |
282 | * Important: we are about to write the buffer, and | |
283 | * possibly block, while still holding the journal lock. | |
284 | * We cannot afford to let the transaction logic start | |
285 | * messing around with this buffer before we write it to | |
ae6ddcc5 | 286 | * disk, as that would break recoverability. |
1da177e4 LT |
287 | */ |
288 | BUFFER_TRACE(bh, "queue"); | |
289 | get_bh(bh); | |
290 | J_ASSERT_BH(bh, !buffer_jwrite(bh)); | |
291 | set_buffer_jwrite(bh); | |
292 | bhs[*batch_count] = bh; | |
78ce89c9 | 293 | __buffer_relink_io(jh); |
1da177e4 LT |
294 | jbd_unlock_bh_state(bh); |
295 | (*batch_count)++; | |
296 | if (*batch_count == NR_BATCH) { | |
78ce89c9 | 297 | spin_unlock(&journal->j_list_lock); |
1da177e4 LT |
298 | __flush_batch(journal, bhs, batch_count); |
299 | ret = 1; | |
300 | } | |
1da177e4 LT |
301 | } |
302 | return ret; | |
303 | } | |
304 | ||
305 | /* | |
78ce89c9 JK |
306 | * Perform an actual checkpoint. We take the first transaction on the |
307 | * list of transactions to be checkpointed and send all its buffers | |
308 | * to disk. We submit larger chunks of data at once. | |
ae6ddcc5 | 309 | * |
1da177e4 | 310 | * The journal should be locked before calling this function. |
4afe9785 | 311 | * Called with j_checkpoint_mutex held. |
1da177e4 LT |
312 | */ |
313 | int log_do_checkpoint(journal_t *journal) | |
314 | { | |
78ce89c9 JK |
315 | transaction_t *transaction; |
316 | tid_t this_tid; | |
1da177e4 | 317 | int result; |
1da177e4 LT |
318 | |
319 | jbd_debug(1, "Start checkpoint\n"); | |
320 | ||
ae6ddcc5 | 321 | /* |
1da177e4 LT |
322 | * First thing: if there are any transactions in the log which |
323 | * don't need checkpointing, just eliminate them from the | |
ae6ddcc5 | 324 | * journal straight away. |
1da177e4 LT |
325 | */ |
326 | result = cleanup_journal_tail(journal); | |
327 | jbd_debug(1, "cleanup_journal_tail returned %d\n", result); | |
328 | if (result <= 0) | |
329 | return result; | |
330 | ||
331 | /* | |
78ce89c9 JK |
332 | * OK, we need to start writing disk blocks. Take one transaction |
333 | * and write it. | |
1da177e4 | 334 | */ |
4afe9785 | 335 | result = 0; |
78ce89c9 JK |
336 | spin_lock(&journal->j_list_lock); |
337 | if (!journal->j_checkpoint_transactions) | |
338 | goto out; | |
339 | transaction = journal->j_checkpoint_transactions; | |
340 | this_tid = transaction->t_tid; | |
341 | restart: | |
1da177e4 | 342 | /* |
78ce89c9 JK |
343 | * If someone cleaned up this transaction while we slept, we're |
344 | * done (maybe it's a new transaction, but it fell at the same | |
345 | * address). | |
1da177e4 | 346 | */ |
78ce89c9 JK |
347 | if (journal->j_checkpoint_transactions == transaction && |
348 | transaction->t_tid == this_tid) { | |
349 | int batch_count = 0; | |
350 | struct buffer_head *bhs[NR_BATCH]; | |
351 | struct journal_head *jh; | |
4afe9785 | 352 | int retry = 0, err; |
78ce89c9 JK |
353 | |
354 | while (!retry && transaction->t_checkpoint_list) { | |
1da177e4 LT |
355 | struct buffer_head *bh; |
356 | ||
78ce89c9 | 357 | jh = transaction->t_checkpoint_list; |
1da177e4 LT |
358 | bh = jh2bh(jh); |
359 | if (!jbd_trylock_bh_state(bh)) { | |
360 | jbd_sync_bh(journal, bh); | |
1da177e4 LT |
361 | retry = 1; |
362 | break; | |
363 | } | |
78ce89c9 | 364 | retry = __process_buffer(journal, jh, bhs,&batch_count); |
4afe9785 HK |
365 | if (retry < 0 && !result) |
366 | result = retry; | |
95c354fe NP |
367 | if (!retry && (need_resched() || |
368 | spin_needbreak(&journal->j_list_lock))) { | |
78ce89c9 | 369 | spin_unlock(&journal->j_list_lock); |
1da177e4 LT |
370 | retry = 1; |
371 | break; | |
372 | } | |
78ce89c9 | 373 | } |
1da177e4 | 374 | |
00ea8145 | 375 | if (batch_count) { |
78ce89c9 JK |
376 | if (!retry) { |
377 | spin_unlock(&journal->j_list_lock); | |
378 | retry = 1; | |
379 | } | |
1da177e4 | 380 | __flush_batch(journal, bhs, &batch_count); |
00ea8145 | 381 | } |
1da177e4 | 382 | |
78ce89c9 JK |
383 | if (retry) { |
384 | spin_lock(&journal->j_list_lock); | |
385 | goto restart; | |
386 | } | |
1da177e4 | 387 | /* |
78ce89c9 JK |
388 | * Now we have cleaned up the first transaction's checkpoint |
389 | * list. Let's clean up the second one | |
7c8903f6 | 390 | */ |
4afe9785 HK |
391 | err = __wait_cp_io(journal, transaction); |
392 | if (!result) | |
393 | result = err; | |
1da177e4 | 394 | } |
78ce89c9 | 395 | out: |
1da177e4 | 396 | spin_unlock(&journal->j_list_lock); |
1da177e4 | 397 | if (result < 0) |
4afe9785 HK |
398 | journal_abort(journal, result); |
399 | else | |
400 | result = cleanup_journal_tail(journal); | |
401 | ||
402 | return (result < 0) ? result : 0; | |
1da177e4 LT |
403 | } |
404 | ||
405 | /* | |
406 | * Check the list of checkpoint transactions for the journal to see if | |
407 | * we have already got rid of any since the last update of the log tail | |
408 | * in the journal superblock. If so, we can instantly roll the | |
409 | * superblock forward to remove those transactions from the log. | |
ae6ddcc5 | 410 | * |
1da177e4 | 411 | * Return <0 on error, 0 on success, 1 if there was nothing to clean up. |
ae6ddcc5 | 412 | * |
1da177e4 LT |
413 | * Called with the journal lock held. |
414 | * | |
415 | * This is the only part of the journaling code which really needs to be | |
416 | * aware of transaction aborts. Checkpointing involves writing to the | |
417 | * main filesystem area rather than to the journal, so it can proceed | |
4afe9785 HK |
418 | * even in abort state, but we must not update the super block if |
419 | * checkpointing may have failed. Otherwise, we would lose some metadata | |
420 | * buffers which should be written-back to the filesystem. | |
1da177e4 LT |
421 | */ |
422 | ||
423 | int cleanup_journal_tail(journal_t *journal) | |
424 | { | |
425 | transaction_t * transaction; | |
426 | tid_t first_tid; | |
427 | unsigned long blocknr, freed; | |
428 | ||
4afe9785 HK |
429 | if (is_journal_aborted(journal)) |
430 | return 1; | |
431 | ||
1da177e4 | 432 | /* OK, work out the oldest transaction remaining in the log, and |
ae6ddcc5 MC |
433 | * the log block it starts at. |
434 | * | |
1da177e4 LT |
435 | * If the log is now empty, we need to work out which is the |
436 | * next transaction ID we will write, and where it will | |
437 | * start. */ | |
438 | ||
439 | spin_lock(&journal->j_state_lock); | |
440 | spin_lock(&journal->j_list_lock); | |
441 | transaction = journal->j_checkpoint_transactions; | |
442 | if (transaction) { | |
443 | first_tid = transaction->t_tid; | |
444 | blocknr = transaction->t_log_start; | |
445 | } else if ((transaction = journal->j_committing_transaction) != NULL) { | |
446 | first_tid = transaction->t_tid; | |
447 | blocknr = transaction->t_log_start; | |
448 | } else if ((transaction = journal->j_running_transaction) != NULL) { | |
449 | first_tid = transaction->t_tid; | |
450 | blocknr = journal->j_head; | |
451 | } else { | |
452 | first_tid = journal->j_transaction_sequence; | |
453 | blocknr = journal->j_head; | |
454 | } | |
455 | spin_unlock(&journal->j_list_lock); | |
456 | J_ASSERT(blocknr != 0); | |
457 | ||
458 | /* If the oldest pinned transaction is at the tail of the log | |
459 | already then there's not much we can do right now. */ | |
460 | if (journal->j_tail_sequence == first_tid) { | |
461 | spin_unlock(&journal->j_state_lock); | |
462 | return 1; | |
463 | } | |
464 | ||
465 | /* OK, update the superblock to recover the freed space. | |
466 | * Physical blocks come first: have we wrapped beyond the end of | |
467 | * the log? */ | |
468 | freed = blocknr - journal->j_tail; | |
469 | if (blocknr < journal->j_tail) | |
470 | freed = freed + journal->j_last - journal->j_first; | |
471 | ||
472 | jbd_debug(1, | |
473 | "Cleaning journal tail from %d to %d (offset %lu), " | |
474 | "freeing %lu\n", | |
475 | journal->j_tail_sequence, first_tid, blocknr, freed); | |
476 | ||
477 | journal->j_free += freed; | |
478 | journal->j_tail_sequence = first_tid; | |
479 | journal->j_tail = blocknr; | |
480 | spin_unlock(&journal->j_state_lock); | |
481 | if (!(journal->j_flags & JFS_ABORT)) | |
482 | journal_update_superblock(journal, 1); | |
483 | return 0; | |
484 | } | |
485 | ||
486 | ||
487 | /* Checkpoint list management */ | |
488 | ||
78ce89c9 JK |
489 | /* |
490 | * journal_clean_one_cp_list | |
491 | * | |
492 | * Find all the written-back checkpoint buffers in the given list and release them. | |
493 | * | |
494 | * Called with the journal locked. | |
495 | * Called with j_list_lock held. | |
496 | * Returns number of bufers reaped (for debug) | |
497 | */ | |
498 | ||
499 | static int journal_clean_one_cp_list(struct journal_head *jh, int *released) | |
500 | { | |
501 | struct journal_head *last_jh; | |
502 | struct journal_head *next_jh = jh; | |
503 | int ret, freed = 0; | |
504 | ||
505 | *released = 0; | |
506 | if (!jh) | |
507 | return 0; | |
508 | ||
e9ad5620 | 509 | last_jh = jh->b_cpprev; |
78ce89c9 JK |
510 | do { |
511 | jh = next_jh; | |
512 | next_jh = jh->b_cpnext; | |
513 | /* Use trylock because of the ranking */ | |
514 | if (jbd_trylock_bh_state(jh2bh(jh))) { | |
515 | ret = __try_to_free_cp_buf(jh); | |
516 | if (ret) { | |
517 | freed++; | |
518 | if (ret == 2) { | |
519 | *released = 1; | |
520 | return freed; | |
521 | } | |
522 | } | |
523 | } | |
524 | /* | |
525 | * This function only frees up some memory | |
526 | * if possible so we dont have an obligation | |
527 | * to finish processing. Bail out if preemption | |
528 | * requested: | |
529 | */ | |
530 | if (need_resched()) | |
531 | return freed; | |
532 | } while (jh != last_jh); | |
533 | ||
534 | return freed; | |
535 | } | |
536 | ||
1da177e4 LT |
537 | /* |
538 | * journal_clean_checkpoint_list | |
539 | * | |
540 | * Find all the written-back checkpoint buffers in the journal and release them. | |
541 | * | |
542 | * Called with the journal locked. | |
543 | * Called with j_list_lock held. | |
78ce89c9 | 544 | * Returns number of buffers reaped (for debug) |
1da177e4 LT |
545 | */ |
546 | ||
547 | int __journal_clean_checkpoint_list(journal_t *journal) | |
548 | { | |
549 | transaction_t *transaction, *last_transaction, *next_transaction; | |
7c8903f6 | 550 | int ret = 0; |
78ce89c9 | 551 | int released; |
1da177e4 LT |
552 | |
553 | transaction = journal->j_checkpoint_transactions; | |
78ce89c9 | 554 | if (!transaction) |
1da177e4 LT |
555 | goto out; |
556 | ||
557 | last_transaction = transaction->t_cpprev; | |
558 | next_transaction = transaction; | |
559 | do { | |
1da177e4 LT |
560 | transaction = next_transaction; |
561 | next_transaction = transaction->t_cpnext; | |
78ce89c9 JK |
562 | ret += journal_clean_one_cp_list(transaction-> |
563 | t_checkpoint_list, &released); | |
564 | /* | |
565 | * This function only frees up some memory if possible so we | |
566 | * dont have an obligation to finish processing. Bail out if | |
567 | * preemption requested: | |
568 | */ | |
569 | if (need_resched()) | |
570 | goto out; | |
571 | if (released) | |
572 | continue; | |
573 | /* | |
574 | * It is essential that we are as careful as in the case of | |
575 | * t_checkpoint_list with removing the buffer from the list as | |
576 | * we can possibly see not yet submitted buffers on io_list | |
577 | */ | |
578 | ret += journal_clean_one_cp_list(transaction-> | |
579 | t_checkpoint_io_list, &released); | |
580 | if (need_resched()) | |
581 | goto out; | |
1da177e4 LT |
582 | } while (transaction != last_transaction); |
583 | out: | |
584 | return ret; | |
585 | } | |
586 | ||
ae6ddcc5 | 587 | /* |
1da177e4 LT |
588 | * journal_remove_checkpoint: called after a buffer has been committed |
589 | * to disk (either by being write-back flushed to disk, or being | |
590 | * committed to the log). | |
591 | * | |
592 | * We cannot safely clean a transaction out of the log until all of the | |
593 | * buffer updates committed in that transaction have safely been stored | |
594 | * elsewhere on disk. To achieve this, all of the buffers in a | |
595 | * transaction need to be maintained on the transaction's checkpoint | |
78ce89c9 | 596 | * lists until they have been rewritten, at which point this function is |
1da177e4 | 597 | * called to remove the buffer from the existing transaction's |
78ce89c9 JK |
598 | * checkpoint lists. |
599 | * | |
600 | * The function returns 1 if it frees the transaction, 0 otherwise. | |
1da177e4 LT |
601 | * |
602 | * This function is called with the journal locked. | |
603 | * This function is called with j_list_lock held. | |
78ce89c9 | 604 | * This function is called with jbd_lock_bh_state(jh2bh(jh)) |
1da177e4 LT |
605 | */ |
606 | ||
78ce89c9 | 607 | int __journal_remove_checkpoint(struct journal_head *jh) |
1da177e4 LT |
608 | { |
609 | transaction_t *transaction; | |
610 | journal_t *journal; | |
78ce89c9 | 611 | int ret = 0; |
1da177e4 LT |
612 | |
613 | JBUFFER_TRACE(jh, "entry"); | |
614 | ||
615 | if ((transaction = jh->b_cp_transaction) == NULL) { | |
616 | JBUFFER_TRACE(jh, "not on transaction"); | |
617 | goto out; | |
618 | } | |
619 | journal = transaction->t_journal; | |
620 | ||
621 | __buffer_unlink(jh); | |
78ce89c9 | 622 | jh->b_cp_transaction = NULL; |
1da177e4 | 623 | |
78ce89c9 JK |
624 | if (transaction->t_checkpoint_list != NULL || |
625 | transaction->t_checkpoint_io_list != NULL) | |
1da177e4 LT |
626 | goto out; |
627 | JBUFFER_TRACE(jh, "transaction has no more buffers"); | |
628 | ||
629 | /* | |
630 | * There is one special case to worry about: if we have just pulled the | |
d4beaf4a JK |
631 | * buffer off a running or committing transaction's checkpoing list, |
632 | * then even if the checkpoint list is empty, the transaction obviously | |
633 | * cannot be dropped! | |
1da177e4 | 634 | * |
d4beaf4a | 635 | * The locking here around t_state is a bit sleazy. |
1da177e4 LT |
636 | * See the comment at the end of journal_commit_transaction(). |
637 | */ | |
d4beaf4a JK |
638 | if (transaction->t_state != T_FINISHED) { |
639 | JBUFFER_TRACE(jh, "belongs to running/committing transaction"); | |
1da177e4 LT |
640 | goto out; |
641 | } | |
642 | ||
643 | /* OK, that was the last buffer for the transaction: we can now | |
644 | safely remove this transaction from the log */ | |
645 | ||
646 | __journal_drop_transaction(journal, transaction); | |
647 | ||
648 | /* Just in case anybody was waiting for more transactions to be | |
649 | checkpointed... */ | |
650 | wake_up(&journal->j_wait_logspace); | |
78ce89c9 | 651 | ret = 1; |
1da177e4 LT |
652 | out: |
653 | JBUFFER_TRACE(jh, "exit"); | |
78ce89c9 | 654 | return ret; |
1da177e4 LT |
655 | } |
656 | ||
657 | /* | |
658 | * journal_insert_checkpoint: put a committed buffer onto a checkpoint | |
659 | * list so that we know when it is safe to clean the transaction out of | |
660 | * the log. | |
661 | * | |
662 | * Called with the journal locked. | |
663 | * Called with j_list_lock held. | |
664 | */ | |
ae6ddcc5 | 665 | void __journal_insert_checkpoint(struct journal_head *jh, |
1da177e4 LT |
666 | transaction_t *transaction) |
667 | { | |
668 | JBUFFER_TRACE(jh, "entry"); | |
669 | J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh))); | |
670 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
671 | ||
672 | jh->b_cp_transaction = transaction; | |
673 | ||
674 | if (!transaction->t_checkpoint_list) { | |
675 | jh->b_cpnext = jh->b_cpprev = jh; | |
676 | } else { | |
677 | jh->b_cpnext = transaction->t_checkpoint_list; | |
678 | jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev; | |
679 | jh->b_cpprev->b_cpnext = jh; | |
680 | jh->b_cpnext->b_cpprev = jh; | |
681 | } | |
682 | transaction->t_checkpoint_list = jh; | |
683 | } | |
684 | ||
685 | /* | |
686 | * We've finished with this transaction structure: adios... | |
ae6ddcc5 | 687 | * |
1da177e4 LT |
688 | * The transaction must have no links except for the checkpoint by this |
689 | * point. | |
690 | * | |
691 | * Called with the journal locked. | |
692 | * Called with j_list_lock held. | |
693 | */ | |
694 | ||
695 | void __journal_drop_transaction(journal_t *journal, transaction_t *transaction) | |
696 | { | |
697 | assert_spin_locked(&journal->j_list_lock); | |
698 | if (transaction->t_cpnext) { | |
699 | transaction->t_cpnext->t_cpprev = transaction->t_cpprev; | |
700 | transaction->t_cpprev->t_cpnext = transaction->t_cpnext; | |
701 | if (journal->j_checkpoint_transactions == transaction) | |
702 | journal->j_checkpoint_transactions = | |
703 | transaction->t_cpnext; | |
704 | if (journal->j_checkpoint_transactions == transaction) | |
705 | journal->j_checkpoint_transactions = NULL; | |
706 | } | |
707 | ||
708 | J_ASSERT(transaction->t_state == T_FINISHED); | |
709 | J_ASSERT(transaction->t_buffers == NULL); | |
710 | J_ASSERT(transaction->t_sync_datalist == NULL); | |
711 | J_ASSERT(transaction->t_forget == NULL); | |
712 | J_ASSERT(transaction->t_iobuf_list == NULL); | |
713 | J_ASSERT(transaction->t_shadow_list == NULL); | |
714 | J_ASSERT(transaction->t_log_list == NULL); | |
715 | J_ASSERT(transaction->t_checkpoint_list == NULL); | |
78ce89c9 | 716 | J_ASSERT(transaction->t_checkpoint_io_list == NULL); |
1da177e4 LT |
717 | J_ASSERT(transaction->t_updates == 0); |
718 | J_ASSERT(journal->j_committing_transaction != transaction); | |
719 | J_ASSERT(journal->j_running_transaction != transaction); | |
720 | ||
721 | jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid); | |
722 | kfree(transaction); | |
723 | } |