Commit | Line | Data |
---|---|---|
f5166768 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
470decc6 | 2 | /* |
58862699 | 3 | * linux/fs/jbd2/checkpoint.c |
470decc6 DK |
4 | * |
5 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1999 | |
6 | * | |
7 | * Copyright 1999 Red Hat Software --- All Rights Reserved | |
8 | * | |
470decc6 DK |
9 | * Checkpoint routines for the generic filesystem journaling code. |
10 | * Part of the ext2fs journaling system. | |
11 | * | |
12 | * Checkpointing is the process of ensuring that a section of the log is | |
13 | * committed fully to disk, so that that portion of the log can be | |
14 | * reused. | |
15 | */ | |
16 | ||
17 | #include <linux/time.h> | |
18 | #include <linux/fs.h> | |
f7f4bccb | 19 | #include <linux/jbd2.h> |
470decc6 DK |
20 | #include <linux/errno.h> |
21 | #include <linux/slab.h> | |
cc3e1bea | 22 | #include <linux/blkdev.h> |
879c5e6b | 23 | #include <trace/events/jbd2.h> |
470decc6 DK |
24 | |
25 | /* | |
26 | * Unlink a buffer from a transaction checkpoint list. | |
27 | * | |
28 | * Called with j_list_lock held. | |
29 | */ | |
be222553 | 30 | static inline void __buffer_unlink(struct journal_head *jh) |
470decc6 DK |
31 | { |
32 | transaction_t *transaction = jh->b_cp_transaction; | |
33 | ||
34 | jh->b_cpnext->b_cpprev = jh->b_cpprev; | |
35 | jh->b_cpprev->b_cpnext = jh->b_cpnext; | |
36 | if (transaction->t_checkpoint_list == jh) { | |
37 | transaction->t_checkpoint_list = jh->b_cpnext; | |
38 | if (transaction->t_checkpoint_list == jh) | |
39 | transaction->t_checkpoint_list = NULL; | |
40 | } | |
41 | } | |
42 | ||
470decc6 | 43 | /* |
f7f4bccb | 44 | * __jbd2_log_wait_for_space: wait until there is space in the journal. |
470decc6 DK |
45 | * |
46 | * Called under j-state_lock *only*. It will be unlocked if we have to wait | |
47 | * for a checkpoint to free up some space in the log. | |
48 | */ | |
f7f4bccb | 49 | void __jbd2_log_wait_for_space(journal_t *journal) |
05d5233d TT |
50 | __acquires(&journal->j_state_lock) |
51 | __releases(&journal->j_state_lock) | |
470decc6 | 52 | { |
8c3f25d8 | 53 | int nblocks, space_left; |
a931da6a | 54 | /* assert_spin_locked(&journal->j_state_lock); */ |
470decc6 | 55 | |
77444ac4 | 56 | nblocks = journal->j_max_transaction_buffers; |
76c39904 | 57 | while (jbd2_log_space_left(journal) < nblocks) { |
a931da6a | 58 | write_unlock(&journal->j_state_lock); |
53cf9784 | 59 | mutex_lock_io(&journal->j_checkpoint_mutex); |
470decc6 DK |
60 | |
61 | /* | |
62 | * Test again, another process may have checkpointed while we | |
23f8b79e | 63 | * were waiting for the checkpoint lock. If there are no |
8c3f25d8 TT |
64 | * transactions ready to be checkpointed, try to recover |
65 | * journal space by calling cleanup_journal_tail(), and if | |
66 | * that doesn't work, by waiting for the currently committing | |
67 | * transaction to complete. If there is absolutely no way | |
68 | * to make progress, this is either a BUG or corrupted | |
69 | * filesystem, so abort the journal and leave a stack | |
70 | * trace for forensic evidence. | |
470decc6 | 71 | */ |
a931da6a | 72 | write_lock(&journal->j_state_lock); |
1245799f DM |
73 | if (journal->j_flags & JBD2_ABORT) { |
74 | mutex_unlock(&journal->j_checkpoint_mutex); | |
75 | return; | |
76 | } | |
23f8b79e | 77 | spin_lock(&journal->j_list_lock); |
76c39904 | 78 | space_left = jbd2_log_space_left(journal); |
8c3f25d8 | 79 | if (space_left < nblocks) { |
23f8b79e | 80 | int chkpt = journal->j_checkpoint_transactions != NULL; |
8c3f25d8 | 81 | tid_t tid = 0; |
23f8b79e | 82 | |
8c3f25d8 TT |
83 | if (journal->j_committing_transaction) |
84 | tid = journal->j_committing_transaction->t_tid; | |
23f8b79e | 85 | spin_unlock(&journal->j_list_lock); |
a931da6a | 86 | write_unlock(&journal->j_state_lock); |
23f8b79e DG |
87 | if (chkpt) { |
88 | jbd2_log_do_checkpoint(journal); | |
8c3f25d8 TT |
89 | } else if (jbd2_cleanup_journal_tail(journal) == 0) { |
90 | /* We were able to recover space; yay! */ | |
91 | ; | |
92 | } else if (tid) { | |
0ef54180 PG |
93 | /* |
94 | * jbd2_journal_commit_transaction() may want | |
95 | * to take the checkpoint_mutex if JBD2_FLUSHED | |
96 | * is set. So we need to temporarily drop it. | |
97 | */ | |
98 | mutex_unlock(&journal->j_checkpoint_mutex); | |
8c3f25d8 | 99 | jbd2_log_wait_commit(journal, tid); |
0ef54180 PG |
100 | write_lock(&journal->j_state_lock); |
101 | continue; | |
23f8b79e | 102 | } else { |
8c3f25d8 TT |
103 | printk(KERN_ERR "%s: needed %d blocks and " |
104 | "only had %d space available\n", | |
105 | __func__, nblocks, space_left); | |
106 | printk(KERN_ERR "%s: no way to get more " | |
107 | "journal space in %s\n", __func__, | |
108 | journal->j_devname); | |
109 | WARN_ON(1); | |
51f57b01 | 110 | jbd2_journal_abort(journal, -EIO); |
23f8b79e | 111 | } |
a931da6a | 112 | write_lock(&journal->j_state_lock); |
23f8b79e DG |
113 | } else { |
114 | spin_unlock(&journal->j_list_lock); | |
470decc6 DK |
115 | } |
116 | mutex_unlock(&journal->j_checkpoint_mutex); | |
117 | } | |
118 | } | |
119 | ||
470decc6 | 120 | static void |
1a0d3786 | 121 | __flush_batch(journal_t *journal, int *batch_count) |
470decc6 DK |
122 | { |
123 | int i; | |
d3ad8434 | 124 | struct blk_plug plug; |
470decc6 | 125 | |
d3ad8434 | 126 | blk_start_plug(&plug); |
9cb569d6 | 127 | for (i = 0; i < *batch_count; i++) |
70fd7614 | 128 | write_dirty_buffer(journal->j_chkpt_bhs[i], REQ_SYNC); |
d3ad8434 | 129 | blk_finish_plug(&plug); |
9cb569d6 | 130 | |
470decc6 | 131 | for (i = 0; i < *batch_count; i++) { |
1a0d3786 | 132 | struct buffer_head *bh = journal->j_chkpt_bhs[i]; |
470decc6 DK |
133 | BUFFER_TRACE(bh, "brelse"); |
134 | __brelse(bh); | |
c2d6fd9d | 135 | journal->j_chkpt_bhs[i] = NULL; |
470decc6 DK |
136 | } |
137 | *batch_count = 0; | |
138 | } | |
139 | ||
470decc6 DK |
140 | /* |
141 | * Perform an actual checkpoint. We take the first transaction on the | |
142 | * list of transactions to be checkpointed and send all its buffers | |
143 | * to disk. We submit larger chunks of data at once. | |
144 | * | |
145 | * The journal should be locked before calling this function. | |
44519faf | 146 | * Called with j_checkpoint_mutex held. |
470decc6 | 147 | */ |
f7f4bccb | 148 | int jbd2_log_do_checkpoint(journal_t *journal) |
470decc6 | 149 | { |
be1158cc TT |
150 | struct journal_head *jh; |
151 | struct buffer_head *bh; | |
152 | transaction_t *transaction; | |
153 | tid_t this_tid; | |
dc6e8d66 | 154 | int result, batch_count = 0; |
470decc6 | 155 | |
cb3b3bf2 | 156 | jbd2_debug(1, "Start checkpoint\n"); |
470decc6 DK |
157 | |
158 | /* | |
159 | * First thing: if there are any transactions in the log which | |
160 | * don't need checkpointing, just eliminate them from the | |
161 | * journal straight away. | |
162 | */ | |
f7f4bccb | 163 | result = jbd2_cleanup_journal_tail(journal); |
879c5e6b | 164 | trace_jbd2_checkpoint(journal, result); |
cb3b3bf2 | 165 | jbd2_debug(1, "cleanup_journal_tail returned %d\n", result); |
470decc6 DK |
166 | if (result <= 0) |
167 | return result; | |
168 | ||
169 | /* | |
170 | * OK, we need to start writing disk blocks. Take one transaction | |
171 | * and write it. | |
172 | */ | |
173 | spin_lock(&journal->j_list_lock); | |
174 | if (!journal->j_checkpoint_transactions) | |
175 | goto out; | |
176 | transaction = journal->j_checkpoint_transactions; | |
8e85fb3f JL |
177 | if (transaction->t_chp_stats.cs_chp_time == 0) |
178 | transaction->t_chp_stats.cs_chp_time = jiffies; | |
470decc6 DK |
179 | this_tid = transaction->t_tid; |
180 | restart: | |
181 | /* | |
182 | * If someone cleaned up this transaction while we slept, we're | |
183 | * done (maybe it's a new transaction, but it fell at the same | |
184 | * address). | |
185 | */ | |
be1158cc TT |
186 | if (journal->j_checkpoint_transactions != transaction || |
187 | transaction->t_tid != this_tid) | |
188 | goto out; | |
470decc6 | 189 | |
be1158cc TT |
190 | /* checkpoint all of the transaction's buffers */ |
191 | while (transaction->t_checkpoint_list) { | |
192 | jh = transaction->t_checkpoint_list; | |
193 | bh = jh2bh(jh); | |
194 | ||
be1158cc TT |
195 | if (jh->b_transaction != NULL) { |
196 | transaction_t *t = jh->b_transaction; | |
197 | tid_t tid = t->t_tid; | |
470decc6 | 198 | |
be1158cc TT |
199 | transaction->t_chp_stats.cs_forced_to_close++; |
200 | spin_unlock(&journal->j_list_lock); | |
201 | if (unlikely(journal->j_flags & JBD2_UNMOUNT)) | |
202 | /* | |
203 | * The journal thread is dead; so | |
204 | * starting and waiting for a commit | |
205 | * to finish will cause us to wait for | |
206 | * a _very_ long time. | |
207 | */ | |
208 | printk(KERN_ERR | |
209 | "JBD2: %s: Waiting for Godot: block %llu\n", | |
210 | journal->j_devname, (unsigned long long) bh->b_blocknr); | |
211 | ||
53cf9784 XW |
212 | if (batch_count) |
213 | __flush_batch(journal, &batch_count); | |
be1158cc | 214 | jbd2_log_start_commit(journal, tid); |
53cf9784 XW |
215 | /* |
216 | * jbd2_journal_commit_transaction() may want | |
217 | * to take the checkpoint_mutex if JBD2_FLUSHED | |
218 | * is set, jbd2_update_log_tail() called by | |
219 | * jbd2_journal_commit_transaction() may also take | |
220 | * checkpoint_mutex. So we need to temporarily | |
221 | * drop it. | |
222 | */ | |
223 | mutex_unlock(&journal->j_checkpoint_mutex); | |
be1158cc | 224 | jbd2_log_wait_commit(journal, tid); |
53cf9784 XW |
225 | mutex_lock_io(&journal->j_checkpoint_mutex); |
226 | spin_lock(&journal->j_list_lock); | |
227 | goto restart; | |
be1158cc | 228 | } |
e34c8dd2 ZC |
229 | if (!trylock_buffer(bh)) { |
230 | /* | |
231 | * The buffer is locked, it may be writing back, or | |
232 | * flushing out in the last couple of cycles, or | |
233 | * re-adding into a new transaction, need to check | |
234 | * it again until it's unlocked. | |
235 | */ | |
236 | get_bh(bh); | |
237 | spin_unlock(&journal->j_list_lock); | |
238 | wait_on_buffer(bh); | |
239 | /* the journal_head may have gone by now */ | |
240 | BUFFER_TRACE(bh, "brelse"); | |
241 | __brelse(bh); | |
242 | goto retry; | |
243 | } else if (!buffer_dirty(bh)) { | |
244 | unlock_buffer(bh); | |
be1158cc | 245 | BUFFER_TRACE(bh, "remove from checkpoint"); |
c2d6fd9d ZY |
246 | /* |
247 | * If the transaction was released or the checkpoint | |
248 | * list was empty, we're done. | |
249 | */ | |
250 | if (__jbd2_journal_remove_checkpoint(jh) || | |
251 | !transaction->t_checkpoint_list) | |
0e5ecf0a | 252 | goto out; |
c2d6fd9d | 253 | } else { |
e34c8dd2 | 254 | unlock_buffer(bh); |
c2d6fd9d ZY |
255 | /* |
256 | * We are about to write the buffer, it could be | |
257 | * raced by some other transaction shrink or buffer | |
258 | * re-log logic once we release the j_list_lock, | |
259 | * leave it on the checkpoint list and check status | |
260 | * again to make sure it's clean. | |
261 | */ | |
262 | BUFFER_TRACE(bh, "queue"); | |
263 | get_bh(bh); | |
264 | J_ASSERT_BH(bh, !buffer_jwrite(bh)); | |
265 | journal->j_chkpt_bhs[batch_count++] = bh; | |
266 | transaction->t_chp_stats.cs_written++; | |
267 | transaction->t_checkpoint_list = jh->b_cpnext; | |
470decc6 | 268 | } |
c2d6fd9d | 269 | |
be1158cc | 270 | if ((batch_count == JBD2_NR_BATCH) || |
c2d6fd9d ZY |
271 | need_resched() || spin_needbreak(&journal->j_list_lock) || |
272 | jh2bh(transaction->t_checkpoint_list) == journal->j_chkpt_bhs[0]) | |
be1158cc | 273 | goto unlock_and_flush; |
470decc6 | 274 | } |
be1158cc TT |
275 | |
276 | if (batch_count) { | |
277 | unlock_and_flush: | |
278 | spin_unlock(&journal->j_list_lock); | |
279 | retry: | |
280 | if (batch_count) | |
281 | __flush_batch(journal, &batch_count); | |
282 | spin_lock(&journal->j_list_lock); | |
283 | goto restart; | |
284 | } | |
285 | ||
470decc6 DK |
286 | out: |
287 | spin_unlock(&journal->j_list_lock); | |
214eb5a4 | 288 | result = jbd2_cleanup_journal_tail(journal); |
44519faf HK |
289 | |
290 | return (result < 0) ? result : 0; | |
470decc6 DK |
291 | } |
292 | ||
293 | /* | |
294 | * Check the list of checkpoint transactions for the journal to see if | |
295 | * we have already got rid of any since the last update of the log tail | |
296 | * in the journal superblock. If so, we can instantly roll the | |
297 | * superblock forward to remove those transactions from the log. | |
298 | * | |
299 | * Return <0 on error, 0 on success, 1 if there was nothing to clean up. | |
300 | * | |
301 | * Called with the journal lock held. | |
302 | * | |
303 | * This is the only part of the journaling code which really needs to be | |
304 | * aware of transaction aborts. Checkpointing involves writing to the | |
305 | * main filesystem area rather than to the journal, so it can proceed | |
44519faf HK |
306 | * even in abort state, but we must not update the super block if |
307 | * checkpointing may have failed. Otherwise, we would lose some metadata | |
308 | * buffers which should be written-back to the filesystem. | |
470decc6 DK |
309 | */ |
310 | ||
f7f4bccb | 311 | int jbd2_cleanup_journal_tail(journal_t *journal) |
470decc6 | 312 | { |
470decc6 | 313 | tid_t first_tid; |
79feb521 | 314 | unsigned long blocknr; |
470decc6 | 315 | |
44519faf | 316 | if (is_journal_aborted(journal)) |
6f6a6fda | 317 | return -EIO; |
44519faf | 318 | |
79feb521 | 319 | if (!jbd2_journal_get_log_tail(journal, &first_tid, &blocknr)) |
470decc6 | 320 | return 1; |
79feb521 | 321 | J_ASSERT(blocknr != 0); |
cc3e1bea TT |
322 | |
323 | /* | |
79feb521 JK |
324 | * We need to make sure that any blocks that were recently written out |
325 | * --- perhaps by jbd2_log_do_checkpoint() --- are flushed out before | |
326 | * we drop the transactions from the journal. It's unlikely this will | |
327 | * be necessary, especially with an appropriately sized journal, but we | |
328 | * need this to guarantee correctness. Fortunately | |
329 | * jbd2_cleanup_journal_tail() doesn't get called all that often. | |
cc3e1bea | 330 | */ |
79feb521 | 331 | if (journal->j_flags & JBD2_BARRIER) |
c6bf3f0e | 332 | blkdev_issue_flush(journal->j_fs_dev); |
79feb521 | 333 | |
6f6a6fda | 334 | return __jbd2_update_log_tail(journal, first_tid, blocknr); |
470decc6 DK |
335 | } |
336 | ||
337 | ||
338 | /* Checkpoint list management */ | |
339 | ||
373ac521 ZY |
340 | enum shrink_type {SHRINK_DESTROY, SHRINK_BUSY_STOP, SHRINK_BUSY_SKIP}; |
341 | ||
4ba3fcdd ZY |
342 | /* |
343 | * journal_shrink_one_cp_list | |
344 | * | |
b98dba27 | 345 | * Find all the written-back checkpoint buffers in the given list |
4ba3fcdd ZY |
346 | * and try to release them. If the whole transaction is released, set |
347 | * the 'released' parameter. Return the number of released checkpointed | |
348 | * buffers. | |
349 | * | |
350 | * Called with j_list_lock held. | |
351 | */ | |
352 | static unsigned long journal_shrink_one_cp_list(struct journal_head *jh, | |
373ac521 ZY |
353 | enum shrink_type type, |
354 | bool *released) | |
4ba3fcdd ZY |
355 | { |
356 | struct journal_head *last_jh; | |
357 | struct journal_head *next_jh = jh; | |
358 | unsigned long nr_freed = 0; | |
359 | int ret; | |
360 | ||
b98dba27 ZY |
361 | *released = false; |
362 | if (!jh) | |
4ba3fcdd ZY |
363 | return 0; |
364 | ||
365 | last_jh = jh->b_cpprev; | |
366 | do { | |
367 | jh = next_jh; | |
368 | next_jh = jh->b_cpnext; | |
369 | ||
373ac521 | 370 | if (type == SHRINK_DESTROY) { |
46f881b5 ZY |
371 | ret = __jbd2_journal_remove_checkpoint(jh); |
372 | } else { | |
373 | ret = jbd2_journal_try_remove_checkpoint(jh); | |
373ac521 ZY |
374 | if (ret < 0) { |
375 | if (type == SHRINK_BUSY_SKIP) | |
376 | continue; | |
377 | break; | |
378 | } | |
46f881b5 | 379 | } |
4ba3fcdd ZY |
380 | |
381 | nr_freed++; | |
4ba3fcdd ZY |
382 | if (ret) { |
383 | *released = true; | |
384 | break; | |
385 | } | |
386 | ||
387 | if (need_resched()) | |
388 | break; | |
b98dba27 | 389 | } while (jh != last_jh); |
4ba3fcdd ZY |
390 | |
391 | return nr_freed; | |
392 | } | |
393 | ||
394 | /* | |
395 | * jbd2_journal_shrink_checkpoint_list | |
396 | * | |
397 | * Find 'nr_to_scan' written-back checkpoint buffers in the journal | |
398 | * and try to release them. Return the number of released checkpointed | |
399 | * buffers. | |
400 | * | |
401 | * Called with j_list_lock held. | |
402 | */ | |
403 | unsigned long jbd2_journal_shrink_checkpoint_list(journal_t *journal, | |
404 | unsigned long *nr_to_scan) | |
405 | { | |
406 | transaction_t *transaction, *last_transaction, *next_transaction; | |
b98dba27 | 407 | bool __maybe_unused released; |
4ba3fcdd ZY |
408 | tid_t first_tid = 0, last_tid = 0, next_tid = 0; |
409 | tid_t tid = 0; | |
410 | unsigned long nr_freed = 0; | |
b98dba27 | 411 | unsigned long freed; |
4ba3fcdd ZY |
412 | |
413 | again: | |
414 | spin_lock(&journal->j_list_lock); | |
415 | if (!journal->j_checkpoint_transactions) { | |
416 | spin_unlock(&journal->j_list_lock); | |
417 | goto out; | |
418 | } | |
419 | ||
420 | /* | |
421 | * Get next shrink transaction, resume previous scan or start | |
422 | * over again. If some others do checkpoint and drop transaction | |
423 | * from the checkpoint list, we ignore saved j_shrink_transaction | |
424 | * and start over unconditionally. | |
425 | */ | |
426 | if (journal->j_shrink_transaction) | |
427 | transaction = journal->j_shrink_transaction; | |
428 | else | |
429 | transaction = journal->j_checkpoint_transactions; | |
430 | ||
431 | if (!first_tid) | |
432 | first_tid = transaction->t_tid; | |
433 | last_transaction = journal->j_checkpoint_transactions->t_cpprev; | |
434 | next_transaction = transaction; | |
435 | last_tid = last_transaction->t_tid; | |
436 | do { | |
437 | transaction = next_transaction; | |
438 | next_transaction = transaction->t_cpnext; | |
439 | tid = transaction->t_tid; | |
4ba3fcdd | 440 | |
b98dba27 | 441 | freed = journal_shrink_one_cp_list(transaction->t_checkpoint_list, |
373ac521 | 442 | SHRINK_BUSY_SKIP, &released); |
b98dba27 ZY |
443 | nr_freed += freed; |
444 | (*nr_to_scan) -= min(*nr_to_scan, freed); | |
4ba3fcdd ZY |
445 | if (*nr_to_scan == 0) |
446 | break; | |
447 | if (need_resched() || spin_needbreak(&journal->j_list_lock)) | |
448 | break; | |
4ba3fcdd ZY |
449 | } while (transaction != last_transaction); |
450 | ||
451 | if (transaction != last_transaction) { | |
452 | journal->j_shrink_transaction = next_transaction; | |
453 | next_tid = next_transaction->t_tid; | |
454 | } else { | |
455 | journal->j_shrink_transaction = NULL; | |
456 | next_tid = 0; | |
457 | } | |
458 | ||
459 | spin_unlock(&journal->j_list_lock); | |
460 | cond_resched(); | |
461 | ||
462 | if (*nr_to_scan && next_tid) | |
463 | goto again; | |
464 | out: | |
4ba3fcdd | 465 | trace_jbd2_shrink_checkpoint_list(journal, first_tid, tid, last_tid, |
b98dba27 | 466 | nr_freed, next_tid); |
4ba3fcdd ZY |
467 | |
468 | return nr_freed; | |
469 | } | |
470 | ||
470decc6 DK |
471 | /* |
472 | * journal_clean_checkpoint_list | |
473 | * | |
474 | * Find all the written-back checkpoint buffers in the journal and release them. | |
841df7df | 475 | * If 'destroy' is set, release all buffers unconditionally. |
470decc6 | 476 | * |
470decc6 | 477 | * Called with j_list_lock held. |
470decc6 | 478 | */ |
841df7df | 479 | void __jbd2_journal_clean_checkpoint_list(journal_t *journal, bool destroy) |
470decc6 DK |
480 | { |
481 | transaction_t *transaction, *last_transaction, *next_transaction; | |
373ac521 | 482 | enum shrink_type type; |
b98dba27 | 483 | bool released; |
470decc6 DK |
484 | |
485 | transaction = journal->j_checkpoint_transactions; | |
486 | if (!transaction) | |
50849db3 | 487 | return; |
470decc6 | 488 | |
373ac521 | 489 | type = destroy ? SHRINK_DESTROY : SHRINK_BUSY_STOP; |
470decc6 DK |
490 | last_transaction = transaction->t_cpprev; |
491 | next_transaction = transaction; | |
492 | do { | |
493 | transaction = next_transaction; | |
494 | next_transaction = transaction->t_cpnext; | |
b98dba27 | 495 | journal_shrink_one_cp_list(transaction->t_checkpoint_list, |
373ac521 | 496 | type, &released); |
470decc6 DK |
497 | /* |
498 | * This function only frees up some memory if possible so we | |
499 | * dont have an obligation to finish processing. Bail out if | |
500 | * preemption requested: | |
501 | */ | |
50849db3 JK |
502 | if (need_resched()) |
503 | return; | |
50849db3 JK |
504 | /* |
505 | * Stop scanning if we couldn't free the transaction. This | |
506 | * avoids pointless scanning of transactions which still | |
507 | * weren't checkpointed. | |
508 | */ | |
b98dba27 | 509 | if (!released) |
50849db3 | 510 | return; |
470decc6 | 511 | } while (transaction != last_transaction); |
470decc6 DK |
512 | } |
513 | ||
841df7df JK |
514 | /* |
515 | * Remove buffers from all checkpoint lists as journal is aborted and we just | |
516 | * need to free memory | |
517 | */ | |
518 | void jbd2_journal_destroy_checkpoint(journal_t *journal) | |
519 | { | |
520 | /* | |
521 | * We loop because __jbd2_journal_clean_checkpoint_list() may abort | |
522 | * early due to a need of rescheduling. | |
523 | */ | |
524 | while (1) { | |
525 | spin_lock(&journal->j_list_lock); | |
526 | if (!journal->j_checkpoint_transactions) { | |
527 | spin_unlock(&journal->j_list_lock); | |
528 | break; | |
529 | } | |
530 | __jbd2_journal_clean_checkpoint_list(journal, true); | |
531 | spin_unlock(&journal->j_list_lock); | |
532 | cond_resched(); | |
533 | } | |
534 | } | |
535 | ||
470decc6 DK |
536 | /* |
537 | * journal_remove_checkpoint: called after a buffer has been committed | |
538 | * to disk (either by being write-back flushed to disk, or being | |
539 | * committed to the log). | |
540 | * | |
541 | * We cannot safely clean a transaction out of the log until all of the | |
542 | * buffer updates committed in that transaction have safely been stored | |
543 | * elsewhere on disk. To achieve this, all of the buffers in a | |
544 | * transaction need to be maintained on the transaction's checkpoint | |
545 | * lists until they have been rewritten, at which point this function is | |
546 | * called to remove the buffer from the existing transaction's | |
547 | * checkpoint lists. | |
548 | * | |
549 | * The function returns 1 if it frees the transaction, 0 otherwise. | |
de1b7941 | 550 | * The function can free jh and bh. |
470decc6 | 551 | * |
470decc6 | 552 | * This function is called with j_list_lock held. |
470decc6 | 553 | */ |
f7f4bccb | 554 | int __jbd2_journal_remove_checkpoint(struct journal_head *jh) |
470decc6 | 555 | { |
bf699327 | 556 | struct transaction_chp_stats_s *stats; |
470decc6 DK |
557 | transaction_t *transaction; |
558 | journal_t *journal; | |
470decc6 DK |
559 | |
560 | JBUFFER_TRACE(jh, "entry"); | |
561 | ||
1866cba8 ZY |
562 | transaction = jh->b_cp_transaction; |
563 | if (!transaction) { | |
470decc6 | 564 | JBUFFER_TRACE(jh, "not on transaction"); |
1866cba8 | 565 | return 0; |
470decc6 DK |
566 | } |
567 | journal = transaction->t_journal; | |
568 | ||
de1b7941 | 569 | JBUFFER_TRACE(jh, "removing from transaction"); |
fcf37549 | 570 | |
470decc6 DK |
571 | __buffer_unlink(jh); |
572 | jh->b_cp_transaction = NULL; | |
0705e8d1 | 573 | percpu_counter_dec(&journal->j_checkpoint_jh_count); |
de1b7941 | 574 | jbd2_journal_put_journal_head(jh); |
470decc6 | 575 | |
1866cba8 | 576 | /* Is this transaction empty? */ |
be222553 | 577 | if (transaction->t_checkpoint_list) |
1866cba8 | 578 | return 0; |
470decc6 DK |
579 | |
580 | /* | |
581 | * There is one special case to worry about: if we have just pulled the | |
f5a7a6b0 JK |
582 | * buffer off a running or committing transaction's checkpoing list, |
583 | * then even if the checkpoint list is empty, the transaction obviously | |
584 | * cannot be dropped! | |
470decc6 | 585 | * |
f5a7a6b0 | 586 | * The locking here around t_state is a bit sleazy. |
f7f4bccb | 587 | * See the comment at the end of jbd2_journal_commit_transaction(). |
470decc6 | 588 | */ |
de1b7941 | 589 | if (transaction->t_state != T_FINISHED) |
1866cba8 | 590 | return 0; |
470decc6 | 591 | |
1866cba8 ZY |
592 | /* |
593 | * OK, that was the last buffer for the transaction, we can now | |
594 | * safely remove this transaction from the log. | |
595 | */ | |
bf699327 TT |
596 | stats = &transaction->t_chp_stats; |
597 | if (stats->cs_chp_time) | |
598 | stats->cs_chp_time = jbd2_time_diff(stats->cs_chp_time, | |
599 | jiffies); | |
600 | trace_jbd2_checkpoint_stats(journal->j_fs_dev->bd_dev, | |
601 | transaction->t_tid, stats); | |
470decc6 | 602 | |
f7f4bccb | 603 | __jbd2_journal_drop_transaction(journal, transaction); |
0c2022ec | 604 | jbd2_journal_free_transaction(transaction); |
1866cba8 | 605 | return 1; |
470decc6 DK |
606 | } |
607 | ||
46f881b5 ZY |
608 | /* |
609 | * Check the checkpoint buffer and try to remove it from the checkpoint | |
610 | * list if it's clean. Returns -EBUSY if it is not clean, returns 1 if | |
611 | * it frees the transaction, 0 otherwise. | |
612 | * | |
613 | * This function is called with j_list_lock held. | |
614 | */ | |
615 | int jbd2_journal_try_remove_checkpoint(struct journal_head *jh) | |
616 | { | |
617 | struct buffer_head *bh = jh2bh(jh); | |
618 | ||
590a809f ZC |
619 | if (jh->b_transaction) |
620 | return -EBUSY; | |
46f881b5 ZY |
621 | if (!trylock_buffer(bh)) |
622 | return -EBUSY; | |
623 | if (buffer_dirty(bh)) { | |
624 | unlock_buffer(bh); | |
625 | return -EBUSY; | |
626 | } | |
627 | unlock_buffer(bh); | |
628 | ||
629 | /* | |
630 | * Buffer is clean and the IO has finished (we held the buffer | |
631 | * lock) so the checkpoint is done. We can safely remove the | |
632 | * buffer from this transaction. | |
633 | */ | |
634 | JBUFFER_TRACE(jh, "remove from checkpoint list"); | |
635 | return __jbd2_journal_remove_checkpoint(jh); | |
636 | } | |
637 | ||
470decc6 DK |
638 | /* |
639 | * journal_insert_checkpoint: put a committed buffer onto a checkpoint | |
640 | * list so that we know when it is safe to clean the transaction out of | |
641 | * the log. | |
642 | * | |
643 | * Called with the journal locked. | |
644 | * Called with j_list_lock held. | |
645 | */ | |
f7f4bccb | 646 | void __jbd2_journal_insert_checkpoint(struct journal_head *jh, |
470decc6 DK |
647 | transaction_t *transaction) |
648 | { | |
649 | JBUFFER_TRACE(jh, "entry"); | |
650 | J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh))); | |
651 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
652 | ||
de1b7941 JK |
653 | /* Get reference for checkpointing transaction */ |
654 | jbd2_journal_grab_journal_head(jh2bh(jh)); | |
470decc6 DK |
655 | jh->b_cp_transaction = transaction; |
656 | ||
657 | if (!transaction->t_checkpoint_list) { | |
658 | jh->b_cpnext = jh->b_cpprev = jh; | |
659 | } else { | |
660 | jh->b_cpnext = transaction->t_checkpoint_list; | |
661 | jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev; | |
662 | jh->b_cpprev->b_cpnext = jh; | |
663 | jh->b_cpnext->b_cpprev = jh; | |
664 | } | |
665 | transaction->t_checkpoint_list = jh; | |
0705e8d1 | 666 | percpu_counter_inc(&transaction->t_journal->j_checkpoint_jh_count); |
470decc6 DK |
667 | } |
668 | ||
669 | /* | |
670 | * We've finished with this transaction structure: adios... | |
671 | * | |
672 | * The transaction must have no links except for the checkpoint by this | |
673 | * point. | |
674 | * | |
675 | * Called with the journal locked. | |
676 | * Called with j_list_lock held. | |
677 | */ | |
678 | ||
f7f4bccb | 679 | void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction) |
470decc6 DK |
680 | { |
681 | assert_spin_locked(&journal->j_list_lock); | |
4ba3fcdd ZY |
682 | |
683 | journal->j_shrink_transaction = NULL; | |
470decc6 DK |
684 | if (transaction->t_cpnext) { |
685 | transaction->t_cpnext->t_cpprev = transaction->t_cpprev; | |
686 | transaction->t_cpprev->t_cpnext = transaction->t_cpnext; | |
687 | if (journal->j_checkpoint_transactions == transaction) | |
688 | journal->j_checkpoint_transactions = | |
689 | transaction->t_cpnext; | |
690 | if (journal->j_checkpoint_transactions == transaction) | |
691 | journal->j_checkpoint_transactions = NULL; | |
692 | } | |
693 | ||
694 | J_ASSERT(transaction->t_state == T_FINISHED); | |
695 | J_ASSERT(transaction->t_buffers == NULL); | |
470decc6 | 696 | J_ASSERT(transaction->t_forget == NULL); |
470decc6 | 697 | J_ASSERT(transaction->t_shadow_list == NULL); |
470decc6 | 698 | J_ASSERT(transaction->t_checkpoint_list == NULL); |
a51dca9c | 699 | J_ASSERT(atomic_read(&transaction->t_updates) == 0); |
470decc6 DK |
700 | J_ASSERT(journal->j_committing_transaction != transaction); |
701 | J_ASSERT(journal->j_running_transaction != transaction); | |
702 | ||
2201c590 SA |
703 | trace_jbd2_drop_transaction(journal, transaction); |
704 | ||
cb3b3bf2 | 705 | jbd2_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid); |
470decc6 | 706 | } |