]>
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 | */ | |
30 | static inline void __buffer_unlink_first(struct journal_head *jh) | |
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 | ||
43 | /* | |
44 | * Unlink a buffer from a transaction checkpoint(io) list. | |
45 | * | |
46 | * Called with j_list_lock held. | |
47 | */ | |
48 | static inline void __buffer_unlink(struct journal_head *jh) | |
49 | { | |
50 | transaction_t *transaction = jh->b_cp_transaction; | |
51 | ||
52 | __buffer_unlink_first(jh); | |
53 | if (transaction->t_checkpoint_io_list == jh) { | |
54 | transaction->t_checkpoint_io_list = jh->b_cpnext; | |
55 | if (transaction->t_checkpoint_io_list == jh) | |
56 | transaction->t_checkpoint_io_list = NULL; | |
57 | } | |
58 | } | |
59 | ||
60 | /* | |
61 | * Move a buffer from the checkpoint list to the checkpoint io list | |
62 | * | |
63 | * Called with j_list_lock held | |
64 | */ | |
65 | static inline void __buffer_relink_io(struct journal_head *jh) | |
66 | { | |
67 | transaction_t *transaction = jh->b_cp_transaction; | |
68 | ||
69 | __buffer_unlink_first(jh); | |
70 | ||
71 | if (!transaction->t_checkpoint_io_list) { | |
72 | jh->b_cpnext = jh->b_cpprev = jh; | |
73 | } else { | |
74 | jh->b_cpnext = transaction->t_checkpoint_io_list; | |
75 | jh->b_cpprev = transaction->t_checkpoint_io_list->b_cpprev; | |
76 | jh->b_cpprev->b_cpnext = jh; | |
77 | jh->b_cpnext->b_cpprev = jh; | |
78 | } | |
79 | transaction->t_checkpoint_io_list = jh; | |
80 | } | |
81 | ||
82 | /* | |
83 | * Try to release a checkpointed buffer from its transaction. | |
84 | * Returns 1 if we released it and 2 if we also released the | |
85 | * whole transaction. | |
86 | * | |
87 | * Requires j_list_lock | |
470decc6 DK |
88 | */ |
89 | static int __try_to_free_cp_buf(struct journal_head *jh) | |
90 | { | |
91 | int ret = 0; | |
92 | struct buffer_head *bh = jh2bh(jh); | |
93 | ||
932bb305 | 94 | if (jh->b_transaction == NULL && !buffer_locked(bh) && |
44519faf | 95 | !buffer_dirty(bh) && !buffer_write_io_error(bh)) { |
470decc6 | 96 | JBUFFER_TRACE(jh, "remove from checkpoint list"); |
f7f4bccb | 97 | ret = __jbd2_journal_remove_checkpoint(jh) + 1; |
470decc6 DK |
98 | } |
99 | return ret; | |
100 | } | |
101 | ||
102 | /* | |
f7f4bccb | 103 | * __jbd2_log_wait_for_space: wait until there is space in the journal. |
470decc6 DK |
104 | * |
105 | * Called under j-state_lock *only*. It will be unlocked if we have to wait | |
106 | * for a checkpoint to free up some space in the log. | |
107 | */ | |
f7f4bccb | 108 | void __jbd2_log_wait_for_space(journal_t *journal) |
05d5233d TT |
109 | __acquires(&journal->j_state_lock) |
110 | __releases(&journal->j_state_lock) | |
470decc6 | 111 | { |
8c3f25d8 | 112 | int nblocks, space_left; |
a931da6a | 113 | /* assert_spin_locked(&journal->j_state_lock); */ |
470decc6 | 114 | |
77444ac4 | 115 | nblocks = journal->j_max_transaction_buffers; |
76c39904 | 116 | while (jbd2_log_space_left(journal) < nblocks) { |
a931da6a | 117 | write_unlock(&journal->j_state_lock); |
53cf9784 | 118 | mutex_lock_io(&journal->j_checkpoint_mutex); |
470decc6 DK |
119 | |
120 | /* | |
121 | * Test again, another process may have checkpointed while we | |
23f8b79e | 122 | * were waiting for the checkpoint lock. If there are no |
8c3f25d8 TT |
123 | * transactions ready to be checkpointed, try to recover |
124 | * journal space by calling cleanup_journal_tail(), and if | |
125 | * that doesn't work, by waiting for the currently committing | |
126 | * transaction to complete. If there is absolutely no way | |
127 | * to make progress, this is either a BUG or corrupted | |
128 | * filesystem, so abort the journal and leave a stack | |
129 | * trace for forensic evidence. | |
470decc6 | 130 | */ |
a931da6a | 131 | write_lock(&journal->j_state_lock); |
1245799f DM |
132 | if (journal->j_flags & JBD2_ABORT) { |
133 | mutex_unlock(&journal->j_checkpoint_mutex); | |
134 | return; | |
135 | } | |
23f8b79e | 136 | spin_lock(&journal->j_list_lock); |
76c39904 | 137 | space_left = jbd2_log_space_left(journal); |
8c3f25d8 | 138 | if (space_left < nblocks) { |
23f8b79e | 139 | int chkpt = journal->j_checkpoint_transactions != NULL; |
8c3f25d8 | 140 | tid_t tid = 0; |
23f8b79e | 141 | |
8c3f25d8 TT |
142 | if (journal->j_committing_transaction) |
143 | tid = journal->j_committing_transaction->t_tid; | |
23f8b79e | 144 | spin_unlock(&journal->j_list_lock); |
a931da6a | 145 | write_unlock(&journal->j_state_lock); |
23f8b79e DG |
146 | if (chkpt) { |
147 | jbd2_log_do_checkpoint(journal); | |
8c3f25d8 TT |
148 | } else if (jbd2_cleanup_journal_tail(journal) == 0) { |
149 | /* We were able to recover space; yay! */ | |
150 | ; | |
151 | } else if (tid) { | |
0ef54180 PG |
152 | /* |
153 | * jbd2_journal_commit_transaction() may want | |
154 | * to take the checkpoint_mutex if JBD2_FLUSHED | |
155 | * is set. So we need to temporarily drop it. | |
156 | */ | |
157 | mutex_unlock(&journal->j_checkpoint_mutex); | |
8c3f25d8 | 158 | jbd2_log_wait_commit(journal, tid); |
0ef54180 PG |
159 | write_lock(&journal->j_state_lock); |
160 | continue; | |
23f8b79e | 161 | } else { |
8c3f25d8 TT |
162 | printk(KERN_ERR "%s: needed %d blocks and " |
163 | "only had %d space available\n", | |
164 | __func__, nblocks, space_left); | |
165 | printk(KERN_ERR "%s: no way to get more " | |
166 | "journal space in %s\n", __func__, | |
167 | journal->j_devname); | |
168 | WARN_ON(1); | |
51f57b01 | 169 | jbd2_journal_abort(journal, -EIO); |
23f8b79e | 170 | } |
a931da6a | 171 | write_lock(&journal->j_state_lock); |
23f8b79e DG |
172 | } else { |
173 | spin_unlock(&journal->j_list_lock); | |
470decc6 DK |
174 | } |
175 | mutex_unlock(&journal->j_checkpoint_mutex); | |
176 | } | |
177 | } | |
178 | ||
470decc6 | 179 | static void |
1a0d3786 | 180 | __flush_batch(journal_t *journal, int *batch_count) |
470decc6 DK |
181 | { |
182 | int i; | |
d3ad8434 | 183 | struct blk_plug plug; |
470decc6 | 184 | |
d3ad8434 | 185 | blk_start_plug(&plug); |
9cb569d6 | 186 | for (i = 0; i < *batch_count; i++) |
70fd7614 | 187 | write_dirty_buffer(journal->j_chkpt_bhs[i], REQ_SYNC); |
d3ad8434 | 188 | blk_finish_plug(&plug); |
9cb569d6 | 189 | |
470decc6 | 190 | for (i = 0; i < *batch_count; i++) { |
1a0d3786 | 191 | struct buffer_head *bh = journal->j_chkpt_bhs[i]; |
470decc6 DK |
192 | BUFFER_TRACE(bh, "brelse"); |
193 | __brelse(bh); | |
194 | } | |
195 | *batch_count = 0; | |
196 | } | |
197 | ||
470decc6 DK |
198 | /* |
199 | * Perform an actual checkpoint. We take the first transaction on the | |
200 | * list of transactions to be checkpointed and send all its buffers | |
201 | * to disk. We submit larger chunks of data at once. | |
202 | * | |
203 | * The journal should be locked before calling this function. | |
44519faf | 204 | * Called with j_checkpoint_mutex held. |
470decc6 | 205 | */ |
f7f4bccb | 206 | int jbd2_log_do_checkpoint(journal_t *journal) |
470decc6 | 207 | { |
be1158cc TT |
208 | struct journal_head *jh; |
209 | struct buffer_head *bh; | |
210 | transaction_t *transaction; | |
211 | tid_t this_tid; | |
dc6e8d66 | 212 | int result, batch_count = 0; |
470decc6 DK |
213 | |
214 | jbd_debug(1, "Start checkpoint\n"); | |
215 | ||
216 | /* | |
217 | * First thing: if there are any transactions in the log which | |
218 | * don't need checkpointing, just eliminate them from the | |
219 | * journal straight away. | |
220 | */ | |
f7f4bccb | 221 | result = jbd2_cleanup_journal_tail(journal); |
879c5e6b | 222 | trace_jbd2_checkpoint(journal, result); |
470decc6 DK |
223 | jbd_debug(1, "cleanup_journal_tail returned %d\n", result); |
224 | if (result <= 0) | |
225 | return result; | |
226 | ||
227 | /* | |
228 | * OK, we need to start writing disk blocks. Take one transaction | |
229 | * and write it. | |
230 | */ | |
44519faf | 231 | result = 0; |
470decc6 DK |
232 | spin_lock(&journal->j_list_lock); |
233 | if (!journal->j_checkpoint_transactions) | |
234 | goto out; | |
235 | transaction = journal->j_checkpoint_transactions; | |
8e85fb3f JL |
236 | if (transaction->t_chp_stats.cs_chp_time == 0) |
237 | transaction->t_chp_stats.cs_chp_time = jiffies; | |
470decc6 DK |
238 | this_tid = transaction->t_tid; |
239 | restart: | |
240 | /* | |
241 | * If someone cleaned up this transaction while we slept, we're | |
242 | * done (maybe it's a new transaction, but it fell at the same | |
243 | * address). | |
244 | */ | |
be1158cc TT |
245 | if (journal->j_checkpoint_transactions != transaction || |
246 | transaction->t_tid != this_tid) | |
247 | goto out; | |
470decc6 | 248 | |
be1158cc TT |
249 | /* checkpoint all of the transaction's buffers */ |
250 | while (transaction->t_checkpoint_list) { | |
251 | jh = transaction->t_checkpoint_list; | |
252 | bh = jh2bh(jh); | |
253 | ||
254 | if (buffer_locked(bh)) { | |
be1158cc | 255 | get_bh(bh); |
ccd3c437 | 256 | spin_unlock(&journal->j_list_lock); |
be1158cc TT |
257 | wait_on_buffer(bh); |
258 | /* the journal_head may have gone by now */ | |
259 | BUFFER_TRACE(bh, "brelse"); | |
260 | __brelse(bh); | |
261 | goto retry; | |
470decc6 | 262 | } |
be1158cc TT |
263 | if (jh->b_transaction != NULL) { |
264 | transaction_t *t = jh->b_transaction; | |
265 | tid_t tid = t->t_tid; | |
470decc6 | 266 | |
be1158cc TT |
267 | transaction->t_chp_stats.cs_forced_to_close++; |
268 | spin_unlock(&journal->j_list_lock); | |
269 | if (unlikely(journal->j_flags & JBD2_UNMOUNT)) | |
270 | /* | |
271 | * The journal thread is dead; so | |
272 | * starting and waiting for a commit | |
273 | * to finish will cause us to wait for | |
274 | * a _very_ long time. | |
275 | */ | |
276 | printk(KERN_ERR | |
277 | "JBD2: %s: Waiting for Godot: block %llu\n", | |
278 | journal->j_devname, (unsigned long long) bh->b_blocknr); | |
279 | ||
53cf9784 XW |
280 | if (batch_count) |
281 | __flush_batch(journal, &batch_count); | |
be1158cc | 282 | jbd2_log_start_commit(journal, tid); |
53cf9784 XW |
283 | /* |
284 | * jbd2_journal_commit_transaction() may want | |
285 | * to take the checkpoint_mutex if JBD2_FLUSHED | |
286 | * is set, jbd2_update_log_tail() called by | |
287 | * jbd2_journal_commit_transaction() may also take | |
288 | * checkpoint_mutex. So we need to temporarily | |
289 | * drop it. | |
290 | */ | |
291 | mutex_unlock(&journal->j_checkpoint_mutex); | |
be1158cc | 292 | jbd2_log_wait_commit(journal, tid); |
53cf9784 XW |
293 | mutex_lock_io(&journal->j_checkpoint_mutex); |
294 | spin_lock(&journal->j_list_lock); | |
295 | goto restart; | |
be1158cc TT |
296 | } |
297 | if (!buffer_dirty(bh)) { | |
298 | if (unlikely(buffer_write_io_error(bh)) && !result) | |
299 | result = -EIO; | |
be1158cc | 300 | BUFFER_TRACE(bh, "remove from checkpoint"); |
0e5ecf0a JK |
301 | if (__jbd2_journal_remove_checkpoint(jh)) |
302 | /* The transaction was released; we're done */ | |
303 | goto out; | |
304 | continue; | |
470decc6 DK |
305 | } |
306 | /* | |
be1158cc TT |
307 | * Important: we are about to write the buffer, and |
308 | * possibly block, while still holding the journal | |
309 | * lock. We cannot afford to let the transaction | |
310 | * logic start messing around with this buffer before | |
311 | * we write it to disk, as that would break | |
312 | * recoverability. | |
470decc6 | 313 | */ |
be1158cc TT |
314 | BUFFER_TRACE(bh, "queue"); |
315 | get_bh(bh); | |
316 | J_ASSERT_BH(bh, !buffer_jwrite(bh)); | |
317 | journal->j_chkpt_bhs[batch_count++] = bh; | |
318 | __buffer_relink_io(jh); | |
319 | transaction->t_chp_stats.cs_written++; | |
320 | if ((batch_count == JBD2_NR_BATCH) || | |
321 | need_resched() || | |
322 | spin_needbreak(&journal->j_list_lock)) | |
323 | goto unlock_and_flush; | |
470decc6 | 324 | } |
be1158cc TT |
325 | |
326 | if (batch_count) { | |
327 | unlock_and_flush: | |
328 | spin_unlock(&journal->j_list_lock); | |
329 | retry: | |
330 | if (batch_count) | |
331 | __flush_batch(journal, &batch_count); | |
332 | spin_lock(&journal->j_list_lock); | |
333 | goto restart; | |
334 | } | |
335 | ||
336 | /* | |
337 | * Now we issued all of the transaction's buffers, let's deal | |
338 | * with the buffers that are out for I/O. | |
339 | */ | |
88fe1acb TT |
340 | restart2: |
341 | /* Did somebody clean up the transaction in the meanwhile? */ | |
342 | if (journal->j_checkpoint_transactions != transaction || | |
343 | transaction->t_tid != this_tid) | |
344 | goto out; | |
345 | ||
dc6e8d66 | 346 | while (transaction->t_checkpoint_io_list) { |
88fe1acb TT |
347 | jh = transaction->t_checkpoint_io_list; |
348 | bh = jh2bh(jh); | |
88fe1acb | 349 | if (buffer_locked(bh)) { |
dc6e8d66 | 350 | get_bh(bh); |
ccd3c437 | 351 | spin_unlock(&journal->j_list_lock); |
88fe1acb TT |
352 | wait_on_buffer(bh); |
353 | /* the journal_head may have gone by now */ | |
354 | BUFFER_TRACE(bh, "brelse"); | |
355 | __brelse(bh); | |
356 | spin_lock(&journal->j_list_lock); | |
357 | goto restart2; | |
358 | } | |
359 | if (unlikely(buffer_write_io_error(bh)) && !result) | |
360 | result = -EIO; | |
361 | ||
362 | /* | |
363 | * Now in whatever state the buffer currently is, we | |
364 | * know that it has been written out and so we can | |
365 | * drop it from the list | |
366 | */ | |
dc6e8d66 TT |
367 | if (__jbd2_journal_remove_checkpoint(jh)) |
368 | break; | |
88fe1acb | 369 | } |
470decc6 DK |
370 | out: |
371 | spin_unlock(&journal->j_list_lock); | |
470decc6 | 372 | if (result < 0) |
44519faf HK |
373 | jbd2_journal_abort(journal, result); |
374 | else | |
375 | result = jbd2_cleanup_journal_tail(journal); | |
376 | ||
377 | return (result < 0) ? result : 0; | |
470decc6 DK |
378 | } |
379 | ||
380 | /* | |
381 | * Check the list of checkpoint transactions for the journal to see if | |
382 | * we have already got rid of any since the last update of the log tail | |
383 | * in the journal superblock. If so, we can instantly roll the | |
384 | * superblock forward to remove those transactions from the log. | |
385 | * | |
386 | * Return <0 on error, 0 on success, 1 if there was nothing to clean up. | |
387 | * | |
388 | * Called with the journal lock held. | |
389 | * | |
390 | * This is the only part of the journaling code which really needs to be | |
391 | * aware of transaction aborts. Checkpointing involves writing to the | |
392 | * main filesystem area rather than to the journal, so it can proceed | |
44519faf HK |
393 | * even in abort state, but we must not update the super block if |
394 | * checkpointing may have failed. Otherwise, we would lose some metadata | |
395 | * buffers which should be written-back to the filesystem. | |
470decc6 DK |
396 | */ |
397 | ||
f7f4bccb | 398 | int jbd2_cleanup_journal_tail(journal_t *journal) |
470decc6 | 399 | { |
470decc6 | 400 | tid_t first_tid; |
79feb521 | 401 | unsigned long blocknr; |
470decc6 | 402 | |
44519faf | 403 | if (is_journal_aborted(journal)) |
6f6a6fda | 404 | return -EIO; |
44519faf | 405 | |
79feb521 | 406 | if (!jbd2_journal_get_log_tail(journal, &first_tid, &blocknr)) |
470decc6 | 407 | return 1; |
79feb521 | 408 | J_ASSERT(blocknr != 0); |
cc3e1bea TT |
409 | |
410 | /* | |
79feb521 JK |
411 | * We need to make sure that any blocks that were recently written out |
412 | * --- perhaps by jbd2_log_do_checkpoint() --- are flushed out before | |
413 | * we drop the transactions from the journal. It's unlikely this will | |
414 | * be necessary, especially with an appropriately sized journal, but we | |
415 | * need this to guarantee correctness. Fortunately | |
416 | * jbd2_cleanup_journal_tail() doesn't get called all that often. | |
cc3e1bea | 417 | */ |
79feb521 | 418 | if (journal->j_flags & JBD2_BARRIER) |
9398554f | 419 | blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS); |
79feb521 | 420 | |
6f6a6fda | 421 | return __jbd2_update_log_tail(journal, first_tid, blocknr); |
470decc6 DK |
422 | } |
423 | ||
424 | ||
425 | /* Checkpoint list management */ | |
426 | ||
427 | /* | |
428 | * journal_clean_one_cp_list | |
429 | * | |
de1b7941 | 430 | * Find all the written-back checkpoint buffers in the given list and |
841df7df | 431 | * release them. If 'destroy' is set, clean all buffers unconditionally. |
470decc6 | 432 | * |
470decc6 | 433 | * Called with j_list_lock held. |
50849db3 | 434 | * Returns 1 if we freed the transaction, 0 otherwise. |
470decc6 | 435 | */ |
841df7df | 436 | static int journal_clean_one_cp_list(struct journal_head *jh, bool destroy) |
470decc6 DK |
437 | { |
438 | struct journal_head *last_jh; | |
439 | struct journal_head *next_jh = jh; | |
50849db3 | 440 | int ret; |
470decc6 | 441 | |
470decc6 DK |
442 | if (!jh) |
443 | return 0; | |
444 | ||
445 | last_jh = jh->b_cpprev; | |
446 | do { | |
447 | jh = next_jh; | |
448 | next_jh = jh->b_cpnext; | |
841df7df JK |
449 | if (!destroy) |
450 | ret = __try_to_free_cp_buf(jh); | |
451 | else | |
452 | ret = __jbd2_journal_remove_checkpoint(jh) + 1; | |
cc97f1a7 | 453 | if (!ret) |
33d14975 | 454 | return 0; |
50849db3 JK |
455 | if (ret == 2) |
456 | return 1; | |
470decc6 DK |
457 | /* |
458 | * This function only frees up some memory | |
459 | * if possible so we dont have an obligation | |
460 | * to finish processing. Bail out if preemption | |
461 | * requested: | |
462 | */ | |
463 | if (need_resched()) | |
33d14975 | 464 | return 0; |
470decc6 DK |
465 | } while (jh != last_jh); |
466 | ||
33d14975 | 467 | return 0; |
470decc6 DK |
468 | } |
469 | ||
470 | /* | |
471 | * journal_clean_checkpoint_list | |
472 | * | |
473 | * Find all the written-back checkpoint buffers in the journal and release them. | |
841df7df | 474 | * If 'destroy' is set, release all buffers unconditionally. |
470decc6 | 475 | * |
470decc6 | 476 | * Called with j_list_lock held. |
470decc6 | 477 | */ |
841df7df | 478 | void __jbd2_journal_clean_checkpoint_list(journal_t *journal, bool destroy) |
470decc6 DK |
479 | { |
480 | transaction_t *transaction, *last_transaction, *next_transaction; | |
cc97f1a7 | 481 | int ret; |
470decc6 DK |
482 | |
483 | transaction = journal->j_checkpoint_transactions; | |
484 | if (!transaction) | |
50849db3 | 485 | return; |
470decc6 DK |
486 | |
487 | last_transaction = transaction->t_cpprev; | |
488 | next_transaction = transaction; | |
489 | do { | |
490 | transaction = next_transaction; | |
491 | next_transaction = transaction->t_cpnext; | |
841df7df JK |
492 | ret = journal_clean_one_cp_list(transaction->t_checkpoint_list, |
493 | destroy); | |
470decc6 DK |
494 | /* |
495 | * This function only frees up some memory if possible so we | |
496 | * dont have an obligation to finish processing. Bail out if | |
497 | * preemption requested: | |
498 | */ | |
50849db3 JK |
499 | if (need_resched()) |
500 | return; | |
501 | if (ret) | |
470decc6 DK |
502 | continue; |
503 | /* | |
504 | * It is essential that we are as careful as in the case of | |
505 | * t_checkpoint_list with removing the buffer from the list as | |
506 | * we can possibly see not yet submitted buffers on io_list | |
507 | */ | |
50849db3 | 508 | ret = journal_clean_one_cp_list(transaction-> |
841df7df | 509 | t_checkpoint_io_list, destroy); |
50849db3 JK |
510 | if (need_resched()) |
511 | return; | |
512 | /* | |
513 | * Stop scanning if we couldn't free the transaction. This | |
514 | * avoids pointless scanning of transactions which still | |
515 | * weren't checkpointed. | |
516 | */ | |
517 | if (!ret) | |
518 | return; | |
470decc6 | 519 | } while (transaction != last_transaction); |
470decc6 DK |
520 | } |
521 | ||
841df7df JK |
522 | /* |
523 | * Remove buffers from all checkpoint lists as journal is aborted and we just | |
524 | * need to free memory | |
525 | */ | |
526 | void jbd2_journal_destroy_checkpoint(journal_t *journal) | |
527 | { | |
528 | /* | |
529 | * We loop because __jbd2_journal_clean_checkpoint_list() may abort | |
530 | * early due to a need of rescheduling. | |
531 | */ | |
532 | while (1) { | |
533 | spin_lock(&journal->j_list_lock); | |
534 | if (!journal->j_checkpoint_transactions) { | |
535 | spin_unlock(&journal->j_list_lock); | |
536 | break; | |
537 | } | |
538 | __jbd2_journal_clean_checkpoint_list(journal, true); | |
539 | spin_unlock(&journal->j_list_lock); | |
540 | cond_resched(); | |
541 | } | |
542 | } | |
543 | ||
470decc6 DK |
544 | /* |
545 | * journal_remove_checkpoint: called after a buffer has been committed | |
546 | * to disk (either by being write-back flushed to disk, or being | |
547 | * committed to the log). | |
548 | * | |
549 | * We cannot safely clean a transaction out of the log until all of the | |
550 | * buffer updates committed in that transaction have safely been stored | |
551 | * elsewhere on disk. To achieve this, all of the buffers in a | |
552 | * transaction need to be maintained on the transaction's checkpoint | |
553 | * lists until they have been rewritten, at which point this function is | |
554 | * called to remove the buffer from the existing transaction's | |
555 | * checkpoint lists. | |
556 | * | |
557 | * The function returns 1 if it frees the transaction, 0 otherwise. | |
de1b7941 | 558 | * The function can free jh and bh. |
470decc6 | 559 | * |
470decc6 | 560 | * This function is called with j_list_lock held. |
470decc6 | 561 | */ |
f7f4bccb | 562 | int __jbd2_journal_remove_checkpoint(struct journal_head *jh) |
470decc6 | 563 | { |
bf699327 | 564 | struct transaction_chp_stats_s *stats; |
470decc6 DK |
565 | transaction_t *transaction; |
566 | journal_t *journal; | |
567 | int ret = 0; | |
568 | ||
569 | JBUFFER_TRACE(jh, "entry"); | |
570 | ||
571 | if ((transaction = jh->b_cp_transaction) == NULL) { | |
572 | JBUFFER_TRACE(jh, "not on transaction"); | |
573 | goto out; | |
574 | } | |
575 | journal = transaction->t_journal; | |
576 | ||
de1b7941 | 577 | JBUFFER_TRACE(jh, "removing from transaction"); |
470decc6 DK |
578 | __buffer_unlink(jh); |
579 | jh->b_cp_transaction = NULL; | |
de1b7941 | 580 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
581 | |
582 | if (transaction->t_checkpoint_list != NULL || | |
583 | transaction->t_checkpoint_io_list != NULL) | |
584 | goto out; | |
470decc6 DK |
585 | |
586 | /* | |
587 | * There is one special case to worry about: if we have just pulled the | |
f5a7a6b0 JK |
588 | * buffer off a running or committing transaction's checkpoing list, |
589 | * then even if the checkpoint list is empty, the transaction obviously | |
590 | * cannot be dropped! | |
470decc6 | 591 | * |
f5a7a6b0 | 592 | * The locking here around t_state is a bit sleazy. |
f7f4bccb | 593 | * See the comment at the end of jbd2_journal_commit_transaction(). |
470decc6 | 594 | */ |
de1b7941 | 595 | if (transaction->t_state != T_FINISHED) |
470decc6 | 596 | goto out; |
470decc6 DK |
597 | |
598 | /* OK, that was the last buffer for the transaction: we can now | |
599 | safely remove this transaction from the log */ | |
bf699327 TT |
600 | stats = &transaction->t_chp_stats; |
601 | if (stats->cs_chp_time) | |
602 | stats->cs_chp_time = jbd2_time_diff(stats->cs_chp_time, | |
603 | jiffies); | |
604 | trace_jbd2_checkpoint_stats(journal->j_fs_dev->bd_dev, | |
605 | transaction->t_tid, stats); | |
470decc6 | 606 | |
f7f4bccb | 607 | __jbd2_journal_drop_transaction(journal, transaction); |
0c2022ec | 608 | jbd2_journal_free_transaction(transaction); |
470decc6 DK |
609 | ret = 1; |
610 | out: | |
470decc6 DK |
611 | return ret; |
612 | } | |
613 | ||
614 | /* | |
615 | * journal_insert_checkpoint: put a committed buffer onto a checkpoint | |
616 | * list so that we know when it is safe to clean the transaction out of | |
617 | * the log. | |
618 | * | |
619 | * Called with the journal locked. | |
620 | * Called with j_list_lock held. | |
621 | */ | |
f7f4bccb | 622 | void __jbd2_journal_insert_checkpoint(struct journal_head *jh, |
470decc6 DK |
623 | transaction_t *transaction) |
624 | { | |
625 | JBUFFER_TRACE(jh, "entry"); | |
626 | J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh))); | |
627 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
628 | ||
de1b7941 JK |
629 | /* Get reference for checkpointing transaction */ |
630 | jbd2_journal_grab_journal_head(jh2bh(jh)); | |
470decc6 DK |
631 | jh->b_cp_transaction = transaction; |
632 | ||
633 | if (!transaction->t_checkpoint_list) { | |
634 | jh->b_cpnext = jh->b_cpprev = jh; | |
635 | } else { | |
636 | jh->b_cpnext = transaction->t_checkpoint_list; | |
637 | jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev; | |
638 | jh->b_cpprev->b_cpnext = jh; | |
639 | jh->b_cpnext->b_cpprev = jh; | |
640 | } | |
641 | transaction->t_checkpoint_list = jh; | |
642 | } | |
643 | ||
644 | /* | |
645 | * We've finished with this transaction structure: adios... | |
646 | * | |
647 | * The transaction must have no links except for the checkpoint by this | |
648 | * point. | |
649 | * | |
650 | * Called with the journal locked. | |
651 | * Called with j_list_lock held. | |
652 | */ | |
653 | ||
f7f4bccb | 654 | void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction) |
470decc6 DK |
655 | { |
656 | assert_spin_locked(&journal->j_list_lock); | |
657 | if (transaction->t_cpnext) { | |
658 | transaction->t_cpnext->t_cpprev = transaction->t_cpprev; | |
659 | transaction->t_cpprev->t_cpnext = transaction->t_cpnext; | |
660 | if (journal->j_checkpoint_transactions == transaction) | |
661 | journal->j_checkpoint_transactions = | |
662 | transaction->t_cpnext; | |
663 | if (journal->j_checkpoint_transactions == transaction) | |
664 | journal->j_checkpoint_transactions = NULL; | |
665 | } | |
666 | ||
667 | J_ASSERT(transaction->t_state == T_FINISHED); | |
668 | J_ASSERT(transaction->t_buffers == NULL); | |
470decc6 | 669 | J_ASSERT(transaction->t_forget == NULL); |
470decc6 | 670 | J_ASSERT(transaction->t_shadow_list == NULL); |
470decc6 DK |
671 | J_ASSERT(transaction->t_checkpoint_list == NULL); |
672 | J_ASSERT(transaction->t_checkpoint_io_list == NULL); | |
a51dca9c | 673 | J_ASSERT(atomic_read(&transaction->t_updates) == 0); |
470decc6 DK |
674 | J_ASSERT(journal->j_committing_transaction != transaction); |
675 | J_ASSERT(journal->j_running_transaction != transaction); | |
676 | ||
2201c590 SA |
677 | trace_jbd2_drop_transaction(journal, transaction); |
678 | ||
470decc6 | 679 | jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid); |
470decc6 | 680 | } |