blob: 9ffb19cf376b47cc57eafd38b67a1d534048827f [file] [log] [blame]
Dave Kleikamp470decc2006-10-11 01:20:57 -07001/*
Uwe Kleine-König58862692007-05-09 07:51:49 +02002 * linux/fs/jbd2/checkpoint.c
Dave Kleikamp470decc2006-10-11 01:20:57 -07003 *
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
5 *
6 * Copyright 1999 Red Hat Software --- All Rights Reserved
7 *
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
11 *
12 * Checkpoint routines for the generic filesystem journaling code.
13 * Part of the ext2fs journaling system.
14 *
15 * Checkpointing is the process of ensuring that a section of the log is
16 * committed fully to disk, so that that portion of the log can be
17 * reused.
18 */
19
20#include <linux/time.h>
21#include <linux/fs.h>
Mingming Caof7f4bcc2006-10-11 01:20:59 -070022#include <linux/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070023#include <linux/errno.h>
24#include <linux/slab.h>
Theodore Ts'occ3e1be2009-12-23 06:52:08 -050025#include <linux/blkdev.h>
Theodore Ts'o879c5e62009-06-17 11:47:48 -040026#include <trace/events/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070027
28/*
29 * Unlink a buffer from a transaction checkpoint list.
30 *
31 * Called with j_list_lock held.
32 */
33static inline void __buffer_unlink_first(struct journal_head *jh)
34{
35 transaction_t *transaction = jh->b_cp_transaction;
36
37 jh->b_cpnext->b_cpprev = jh->b_cpprev;
38 jh->b_cpprev->b_cpnext = jh->b_cpnext;
39 if (transaction->t_checkpoint_list == jh) {
40 transaction->t_checkpoint_list = jh->b_cpnext;
41 if (transaction->t_checkpoint_list == jh)
42 transaction->t_checkpoint_list = NULL;
43 }
44}
45
46/*
47 * Unlink a buffer from a transaction checkpoint(io) list.
48 *
49 * Called with j_list_lock held.
50 */
51static inline void __buffer_unlink(struct journal_head *jh)
52{
53 transaction_t *transaction = jh->b_cp_transaction;
54
55 __buffer_unlink_first(jh);
56 if (transaction->t_checkpoint_io_list == jh) {
57 transaction->t_checkpoint_io_list = jh->b_cpnext;
58 if (transaction->t_checkpoint_io_list == jh)
59 transaction->t_checkpoint_io_list = NULL;
60 }
61}
62
63/*
64 * Move a buffer from the checkpoint list to the checkpoint io list
65 *
66 * Called with j_list_lock held
67 */
68static inline void __buffer_relink_io(struct journal_head *jh)
69{
70 transaction_t *transaction = jh->b_cp_transaction;
71
72 __buffer_unlink_first(jh);
73
74 if (!transaction->t_checkpoint_io_list) {
75 jh->b_cpnext = jh->b_cpprev = jh;
76 } else {
77 jh->b_cpnext = transaction->t_checkpoint_io_list;
78 jh->b_cpprev = transaction->t_checkpoint_io_list->b_cpprev;
79 jh->b_cpprev->b_cpnext = jh;
80 jh->b_cpnext->b_cpprev = jh;
81 }
82 transaction->t_checkpoint_io_list = jh;
83}
84
85/*
86 * Try to release a checkpointed buffer from its transaction.
87 * Returns 1 if we released it and 2 if we also released the
88 * whole transaction.
89 *
90 * Requires j_list_lock
Dave Kleikamp470decc2006-10-11 01:20:57 -070091 */
92static int __try_to_free_cp_buf(struct journal_head *jh)
93{
94 int ret = 0;
95 struct buffer_head *bh = jh2bh(jh);
96
Jan Kara932bb302012-03-13 22:45:25 -040097 if (jh->b_transaction == NULL && !buffer_locked(bh) &&
Hidehiro Kawai44519fa2008-10-10 20:29:13 -040098 !buffer_dirty(bh) && !buffer_write_io_error(bh)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -070099 JBUFFER_TRACE(jh, "remove from checkpoint list");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700100 ret = __jbd2_journal_remove_checkpoint(jh) + 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700101 }
102 return ret;
103}
104
105/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700106 * __jbd2_log_wait_for_space: wait until there is space in the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700107 *
108 * Called under j-state_lock *only*. It will be unlocked if we have to wait
109 * for a checkpoint to free up some space in the log.
110 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700111void __jbd2_log_wait_for_space(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700112{
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500113 int nblocks, space_left;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400114 /* assert_spin_locked(&journal->j_state_lock); */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700115
Jan Kara76c39902013-06-04 12:12:57 -0400116 nblocks = jbd2_space_needed(journal);
117 while (jbd2_log_space_left(journal) < nblocks) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700118 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700119 return;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400120 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700121 mutex_lock(&journal->j_checkpoint_mutex);
122
123 /*
124 * Test again, another process may have checkpointed while we
Duane Griffin23f8b792008-10-08 23:28:31 -0400125 * were waiting for the checkpoint lock. If there are no
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500126 * transactions ready to be checkpointed, try to recover
127 * journal space by calling cleanup_journal_tail(), and if
128 * that doesn't work, by waiting for the currently committing
129 * transaction to complete. If there is absolutely no way
130 * to make progress, this is either a BUG or corrupted
131 * filesystem, so abort the journal and leave a stack
132 * trace for forensic evidence.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700133 */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400134 write_lock(&journal->j_state_lock);
Duane Griffin23f8b792008-10-08 23:28:31 -0400135 spin_lock(&journal->j_list_lock);
Jan Kara76c39902013-06-04 12:12:57 -0400136 nblocks = jbd2_space_needed(journal);
137 space_left = jbd2_log_space_left(journal);
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500138 if (space_left < nblocks) {
Duane Griffin23f8b792008-10-08 23:28:31 -0400139 int chkpt = journal->j_checkpoint_transactions != NULL;
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500140 tid_t tid = 0;
Duane Griffin23f8b792008-10-08 23:28:31 -0400141
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500142 if (journal->j_committing_transaction)
143 tid = journal->j_committing_transaction->t_tid;
Duane Griffin23f8b792008-10-08 23:28:31 -0400144 spin_unlock(&journal->j_list_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400145 write_unlock(&journal->j_state_lock);
Duane Griffin23f8b792008-10-08 23:28:31 -0400146 if (chkpt) {
147 jbd2_log_do_checkpoint(journal);
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500148 } else if (jbd2_cleanup_journal_tail(journal) == 0) {
149 /* We were able to recover space; yay! */
150 ;
151 } else if (tid) {
Paul Gortmaker0ef54182013-06-12 22:47:35 -0400152 /*
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);
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500158 jbd2_log_wait_commit(journal, tid);
Paul Gortmaker0ef54182013-06-12 22:47:35 -0400159 write_lock(&journal->j_state_lock);
160 continue;
Duane Griffin23f8b792008-10-08 23:28:31 -0400161 } else {
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500162 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);
Duane Griffin23f8b792008-10-08 23:28:31 -0400169 jbd2_journal_abort(journal, 0);
170 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400171 write_lock(&journal->j_state_lock);
Duane Griffin23f8b792008-10-08 23:28:31 -0400172 } else {
173 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700174 }
175 mutex_unlock(&journal->j_checkpoint_mutex);
176 }
177}
178
Dave Kleikamp470decc2006-10-11 01:20:57 -0700179static void
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500180__flush_batch(journal_t *journal, int *batch_count)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700181{
182 int i;
Tao Mad3ad8432011-06-27 12:36:29 -0400183 struct blk_plug plug;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700184
Tao Mad3ad8432011-06-27 12:36:29 -0400185 blk_start_plug(&plug);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200186 for (i = 0; i < *batch_count; i++)
Tao Mad3ad8432011-06-27 12:36:29 -0400187 write_dirty_buffer(journal->j_chkpt_bhs[i], WRITE_SYNC);
188 blk_finish_plug(&plug);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200189
Dave Kleikamp470decc2006-10-11 01:20:57 -0700190 for (i = 0; i < *batch_count; i++) {
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500191 struct buffer_head *bh = journal->j_chkpt_bhs[i];
Dave Kleikamp470decc2006-10-11 01:20:57 -0700192 BUFFER_TRACE(bh, "brelse");
193 __brelse(bh);
194 }
195 *batch_count = 0;
196}
197
198/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700199 * 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.
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400204 * Called with j_checkpoint_mutex held.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700205 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700206int jbd2_log_do_checkpoint(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700207{
Theodore Ts'obe1158c2014-09-01 21:19:01 -0400208 struct journal_head *jh;
209 struct buffer_head *bh;
210 transaction_t *transaction;
211 tid_t this_tid;
Theodore Ts'odc6e8d62014-09-04 18:09:22 -0400212 int result, batch_count = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700213
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 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700221 result = jbd2_cleanup_journal_tail(journal);
Theodore Ts'o879c5e62009-06-17 11:47:48 -0400222 trace_jbd2_checkpoint(journal, result);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700223 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 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400231 result = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700232 spin_lock(&journal->j_list_lock);
233 if (!journal->j_checkpoint_transactions)
234 goto out;
235 transaction = journal->j_checkpoint_transactions;
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500236 if (transaction->t_chp_stats.cs_chp_time == 0)
237 transaction->t_chp_stats.cs_chp_time = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700238 this_tid = transaction->t_tid;
239restart:
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 */
Theodore Ts'obe1158c2014-09-01 21:19:01 -0400245 if (journal->j_checkpoint_transactions != transaction ||
246 transaction->t_tid != this_tid)
247 goto out;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700248
Theodore Ts'obe1158c2014-09-01 21:19:01 -0400249 /* 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)) {
255 spin_unlock(&journal->j_list_lock);
256 get_bh(bh);
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;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700262 }
Theodore Ts'obe1158c2014-09-01 21:19:01 -0400263 if (jh->b_transaction != NULL) {
264 transaction_t *t = jh->b_transaction;
265 tid_t tid = t->t_tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700266
Theodore Ts'obe1158c2014-09-01 21:19:01 -0400267 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
280 jbd2_log_start_commit(journal, tid);
281 jbd2_log_wait_commit(journal, tid);
282 goto retry;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700283 }
Theodore Ts'obe1158c2014-09-01 21:19:01 -0400284 if (!buffer_dirty(bh)) {
285 if (unlikely(buffer_write_io_error(bh)) && !result)
286 result = -EIO;
Theodore Ts'obe1158c2014-09-01 21:19:01 -0400287 BUFFER_TRACE(bh, "remove from checkpoint");
Jan Kara0e5ecf0a2014-09-04 18:09:29 -0400288 if (__jbd2_journal_remove_checkpoint(jh))
289 /* The transaction was released; we're done */
290 goto out;
291 continue;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700292 }
293 /*
Theodore Ts'obe1158c2014-09-01 21:19:01 -0400294 * Important: we are about to write the buffer, and
295 * possibly block, while still holding the journal
296 * lock. We cannot afford to let the transaction
297 * logic start messing around with this buffer before
298 * we write it to disk, as that would break
299 * recoverability.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700300 */
Theodore Ts'obe1158c2014-09-01 21:19:01 -0400301 BUFFER_TRACE(bh, "queue");
302 get_bh(bh);
303 J_ASSERT_BH(bh, !buffer_jwrite(bh));
304 journal->j_chkpt_bhs[batch_count++] = bh;
305 __buffer_relink_io(jh);
306 transaction->t_chp_stats.cs_written++;
307 if ((batch_count == JBD2_NR_BATCH) ||
308 need_resched() ||
309 spin_needbreak(&journal->j_list_lock))
310 goto unlock_and_flush;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700311 }
Theodore Ts'obe1158c2014-09-01 21:19:01 -0400312
313 if (batch_count) {
314 unlock_and_flush:
315 spin_unlock(&journal->j_list_lock);
316 retry:
317 if (batch_count)
318 __flush_batch(journal, &batch_count);
319 spin_lock(&journal->j_list_lock);
320 goto restart;
321 }
322
323 /*
324 * Now we issued all of the transaction's buffers, let's deal
325 * with the buffers that are out for I/O.
326 */
Theodore Ts'o88fe1ac2014-09-01 21:26:09 -0400327restart2:
328 /* Did somebody clean up the transaction in the meanwhile? */
329 if (journal->j_checkpoint_transactions != transaction ||
330 transaction->t_tid != this_tid)
331 goto out;
332
Theodore Ts'odc6e8d62014-09-04 18:09:22 -0400333 while (transaction->t_checkpoint_io_list) {
Theodore Ts'o88fe1ac2014-09-01 21:26:09 -0400334 jh = transaction->t_checkpoint_io_list;
335 bh = jh2bh(jh);
Theodore Ts'o88fe1ac2014-09-01 21:26:09 -0400336 if (buffer_locked(bh)) {
337 spin_unlock(&journal->j_list_lock);
Theodore Ts'odc6e8d62014-09-04 18:09:22 -0400338 get_bh(bh);
Theodore Ts'o88fe1ac2014-09-01 21:26:09 -0400339 wait_on_buffer(bh);
340 /* the journal_head may have gone by now */
341 BUFFER_TRACE(bh, "brelse");
342 __brelse(bh);
343 spin_lock(&journal->j_list_lock);
344 goto restart2;
345 }
346 if (unlikely(buffer_write_io_error(bh)) && !result)
347 result = -EIO;
348
349 /*
350 * Now in whatever state the buffer currently is, we
351 * know that it has been written out and so we can
352 * drop it from the list
353 */
Theodore Ts'odc6e8d62014-09-04 18:09:22 -0400354 if (__jbd2_journal_remove_checkpoint(jh))
355 break;
Theodore Ts'o88fe1ac2014-09-01 21:26:09 -0400356 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700357out:
358 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700359 if (result < 0)
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400360 jbd2_journal_abort(journal, result);
361 else
362 result = jbd2_cleanup_journal_tail(journal);
363
364 return (result < 0) ? result : 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700365}
366
367/*
368 * Check the list of checkpoint transactions for the journal to see if
369 * we have already got rid of any since the last update of the log tail
370 * in the journal superblock. If so, we can instantly roll the
371 * superblock forward to remove those transactions from the log.
372 *
373 * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
374 *
375 * Called with the journal lock held.
376 *
377 * This is the only part of the journaling code which really needs to be
378 * aware of transaction aborts. Checkpointing involves writing to the
379 * main filesystem area rather than to the journal, so it can proceed
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400380 * even in abort state, but we must not update the super block if
381 * checkpointing may have failed. Otherwise, we would lose some metadata
382 * buffers which should be written-back to the filesystem.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700383 */
384
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700385int jbd2_cleanup_journal_tail(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700386{
Dave Kleikamp470decc2006-10-11 01:20:57 -0700387 tid_t first_tid;
Jan Kara79feb522012-03-13 22:22:54 -0400388 unsigned long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700389
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400390 if (is_journal_aborted(journal))
391 return 1;
392
Jan Kara79feb522012-03-13 22:22:54 -0400393 if (!jbd2_journal_get_log_tail(journal, &first_tid, &blocknr))
394 return 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700395 J_ASSERT(blocknr != 0);
396
Theodore Ts'occ3e1be2009-12-23 06:52:08 -0500397 /*
Jan Kara79feb522012-03-13 22:22:54 -0400398 * We need to make sure that any blocks that were recently written out
399 * --- perhaps by jbd2_log_do_checkpoint() --- are flushed out before
400 * we drop the transactions from the journal. It's unlikely this will
401 * be necessary, especially with an appropriately sized journal, but we
402 * need this to guarantee correctness. Fortunately
403 * jbd2_cleanup_journal_tail() doesn't get called all that often.
Theodore Ts'occ3e1be2009-12-23 06:52:08 -0500404 */
Jan Kara79feb522012-03-13 22:22:54 -0400405 if (journal->j_flags & JBD2_BARRIER)
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200406 blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
Jan Kara79feb522012-03-13 22:22:54 -0400407
408 __jbd2_update_log_tail(journal, first_tid, blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700409 return 0;
410}
411
412
413/* Checkpoint list management */
414
415/*
416 * journal_clean_one_cp_list
417 *
Jan Karade1b7942011-06-13 15:38:22 -0400418 * Find all the written-back checkpoint buffers in the given list and
419 * release them.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700420 *
421 * Called with the journal locked.
422 * Called with j_list_lock held.
Paul Bolle90802ed2011-12-05 13:00:34 +0100423 * Returns number of buffers reaped (for debug)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700424 */
425
426static int journal_clean_one_cp_list(struct journal_head *jh, int *released)
427{
428 struct journal_head *last_jh;
429 struct journal_head *next_jh = jh;
430 int ret, freed = 0;
431
432 *released = 0;
433 if (!jh)
434 return 0;
435
436 last_jh = jh->b_cpprev;
437 do {
438 jh = next_jh;
439 next_jh = jh->b_cpnext;
Jan Kara932bb302012-03-13 22:45:25 -0400440 ret = __try_to_free_cp_buf(jh);
441 if (ret) {
442 freed++;
443 if (ret == 2) {
444 *released = 1;
445 return freed;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700446 }
447 }
448 /*
449 * This function only frees up some memory
450 * if possible so we dont have an obligation
451 * to finish processing. Bail out if preemption
452 * requested:
453 */
454 if (need_resched())
455 return freed;
456 } while (jh != last_jh);
457
458 return freed;
459}
460
461/*
462 * journal_clean_checkpoint_list
463 *
464 * Find all the written-back checkpoint buffers in the journal and release them.
465 *
466 * Called with the journal locked.
467 * Called with j_list_lock held.
468 * Returns number of buffers reaped (for debug)
469 */
470
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700471int __jbd2_journal_clean_checkpoint_list(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700472{
473 transaction_t *transaction, *last_transaction, *next_transaction;
474 int ret = 0;
475 int released;
476
477 transaction = journal->j_checkpoint_transactions;
478 if (!transaction)
479 goto out;
480
481 last_transaction = transaction->t_cpprev;
482 next_transaction = transaction;
483 do {
484 transaction = next_transaction;
485 next_transaction = transaction->t_cpnext;
486 ret += journal_clean_one_cp_list(transaction->
487 t_checkpoint_list, &released);
488 /*
489 * This function only frees up some memory if possible so we
490 * dont have an obligation to finish processing. Bail out if
491 * preemption requested:
492 */
493 if (need_resched())
494 goto out;
495 if (released)
496 continue;
497 /*
498 * It is essential that we are as careful as in the case of
499 * t_checkpoint_list with removing the buffer from the list as
500 * we can possibly see not yet submitted buffers on io_list
501 */
502 ret += journal_clean_one_cp_list(transaction->
503 t_checkpoint_io_list, &released);
504 if (need_resched())
505 goto out;
506 } while (transaction != last_transaction);
507out:
508 return ret;
509}
510
511/*
512 * journal_remove_checkpoint: called after a buffer has been committed
513 * to disk (either by being write-back flushed to disk, or being
514 * committed to the log).
515 *
516 * We cannot safely clean a transaction out of the log until all of the
517 * buffer updates committed in that transaction have safely been stored
518 * elsewhere on disk. To achieve this, all of the buffers in a
519 * transaction need to be maintained on the transaction's checkpoint
520 * lists until they have been rewritten, at which point this function is
521 * called to remove the buffer from the existing transaction's
522 * checkpoint lists.
523 *
524 * The function returns 1 if it frees the transaction, 0 otherwise.
Jan Karade1b7942011-06-13 15:38:22 -0400525 * The function can free jh and bh.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700526 *
Dave Kleikamp470decc2006-10-11 01:20:57 -0700527 * This function is called with j_list_lock held.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700528 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700529int __jbd2_journal_remove_checkpoint(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700530{
Theodore Ts'obf699322009-09-30 00:32:06 -0400531 struct transaction_chp_stats_s *stats;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700532 transaction_t *transaction;
533 journal_t *journal;
534 int ret = 0;
535
536 JBUFFER_TRACE(jh, "entry");
537
538 if ((transaction = jh->b_cp_transaction) == NULL) {
539 JBUFFER_TRACE(jh, "not on transaction");
540 goto out;
541 }
542 journal = transaction->t_journal;
543
Jan Karade1b7942011-06-13 15:38:22 -0400544 JBUFFER_TRACE(jh, "removing from transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700545 __buffer_unlink(jh);
546 jh->b_cp_transaction = NULL;
Jan Karade1b7942011-06-13 15:38:22 -0400547 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700548
549 if (transaction->t_checkpoint_list != NULL ||
550 transaction->t_checkpoint_io_list != NULL)
551 goto out;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700552
553 /*
554 * There is one special case to worry about: if we have just pulled the
Jan Karaf5a7a6b2008-01-28 23:58:27 -0500555 * buffer off a running or committing transaction's checkpoing list,
556 * then even if the checkpoint list is empty, the transaction obviously
557 * cannot be dropped!
Dave Kleikamp470decc2006-10-11 01:20:57 -0700558 *
Jan Karaf5a7a6b2008-01-28 23:58:27 -0500559 * The locking here around t_state is a bit sleazy.
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700560 * See the comment at the end of jbd2_journal_commit_transaction().
Dave Kleikamp470decc2006-10-11 01:20:57 -0700561 */
Jan Karade1b7942011-06-13 15:38:22 -0400562 if (transaction->t_state != T_FINISHED)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700563 goto out;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700564
565 /* OK, that was the last buffer for the transaction: we can now
566 safely remove this transaction from the log */
Theodore Ts'obf699322009-09-30 00:32:06 -0400567 stats = &transaction->t_chp_stats;
568 if (stats->cs_chp_time)
569 stats->cs_chp_time = jbd2_time_diff(stats->cs_chp_time,
570 jiffies);
571 trace_jbd2_checkpoint_stats(journal->j_fs_dev->bd_dev,
572 transaction->t_tid, stats);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700573
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700574 __jbd2_journal_drop_transaction(journal, transaction);
Yongqiang Yang0c2022e2012-02-20 17:53:02 -0500575 jbd2_journal_free_transaction(transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700576 ret = 1;
577out:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700578 return ret;
579}
580
581/*
582 * journal_insert_checkpoint: put a committed buffer onto a checkpoint
583 * list so that we know when it is safe to clean the transaction out of
584 * the log.
585 *
586 * Called with the journal locked.
587 * Called with j_list_lock held.
588 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700589void __jbd2_journal_insert_checkpoint(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700590 transaction_t *transaction)
591{
592 JBUFFER_TRACE(jh, "entry");
593 J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh)));
594 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
595
Jan Karade1b7942011-06-13 15:38:22 -0400596 /* Get reference for checkpointing transaction */
597 jbd2_journal_grab_journal_head(jh2bh(jh));
Dave Kleikamp470decc2006-10-11 01:20:57 -0700598 jh->b_cp_transaction = transaction;
599
600 if (!transaction->t_checkpoint_list) {
601 jh->b_cpnext = jh->b_cpprev = jh;
602 } else {
603 jh->b_cpnext = transaction->t_checkpoint_list;
604 jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev;
605 jh->b_cpprev->b_cpnext = jh;
606 jh->b_cpnext->b_cpprev = jh;
607 }
608 transaction->t_checkpoint_list = jh;
609}
610
611/*
612 * We've finished with this transaction structure: adios...
613 *
614 * The transaction must have no links except for the checkpoint by this
615 * point.
616 *
617 * Called with the journal locked.
618 * Called with j_list_lock held.
619 */
620
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700621void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700622{
623 assert_spin_locked(&journal->j_list_lock);
624 if (transaction->t_cpnext) {
625 transaction->t_cpnext->t_cpprev = transaction->t_cpprev;
626 transaction->t_cpprev->t_cpnext = transaction->t_cpnext;
627 if (journal->j_checkpoint_transactions == transaction)
628 journal->j_checkpoint_transactions =
629 transaction->t_cpnext;
630 if (journal->j_checkpoint_transactions == transaction)
631 journal->j_checkpoint_transactions = NULL;
632 }
633
634 J_ASSERT(transaction->t_state == T_FINISHED);
635 J_ASSERT(transaction->t_buffers == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700636 J_ASSERT(transaction->t_forget == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700637 J_ASSERT(transaction->t_shadow_list == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700638 J_ASSERT(transaction->t_checkpoint_list == NULL);
639 J_ASSERT(transaction->t_checkpoint_io_list == NULL);
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400640 J_ASSERT(atomic_read(&transaction->t_updates) == 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700641 J_ASSERT(journal->j_committing_transaction != transaction);
642 J_ASSERT(journal->j_running_transaction != transaction);
643
Seiji Aguchi2201c592012-02-20 17:53:01 -0500644 trace_jbd2_drop_transaction(journal, transaction);
645
Dave Kleikamp470decc2006-10-11 01:20:57 -0700646 jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700647}