blob: 75a15f371b00c6e19a5e953e25b68be0ea391f03 [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)) {
Jan Karade1b7942011-06-13 15:38:22 -040099 /*
100 * Get our reference so that bh cannot be freed before
101 * we unlock it
102 */
103 get_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700104 JBUFFER_TRACE(jh, "remove from checkpoint list");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700105 ret = __jbd2_journal_remove_checkpoint(jh) + 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700106 BUFFER_TRACE(bh, "release");
107 __brelse(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700108 }
109 return ret;
110}
111
112/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700113 * __jbd2_log_wait_for_space: wait until there is space in the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700114 *
115 * Called under j-state_lock *only*. It will be unlocked if we have to wait
116 * for a checkpoint to free up some space in the log.
117 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700118void __jbd2_log_wait_for_space(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700119{
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500120 int nblocks, space_left;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400121 /* assert_spin_locked(&journal->j_state_lock); */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700122
Jan Kara76c39902013-06-04 12:12:57 -0400123 nblocks = jbd2_space_needed(journal);
124 while (jbd2_log_space_left(journal) < nblocks) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700125 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700126 return;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400127 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700128 mutex_lock(&journal->j_checkpoint_mutex);
129
130 /*
131 * Test again, another process may have checkpointed while we
Duane Griffin23f8b792008-10-08 23:28:31 -0400132 * were waiting for the checkpoint lock. If there are no
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500133 * transactions ready to be checkpointed, try to recover
134 * journal space by calling cleanup_journal_tail(), and if
135 * that doesn't work, by waiting for the currently committing
136 * transaction to complete. If there is absolutely no way
137 * to make progress, this is either a BUG or corrupted
138 * filesystem, so abort the journal and leave a stack
139 * trace for forensic evidence.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700140 */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400141 write_lock(&journal->j_state_lock);
Duane Griffin23f8b792008-10-08 23:28:31 -0400142 spin_lock(&journal->j_list_lock);
Jan Kara76c39902013-06-04 12:12:57 -0400143 nblocks = jbd2_space_needed(journal);
144 space_left = jbd2_log_space_left(journal);
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500145 if (space_left < nblocks) {
Duane Griffin23f8b792008-10-08 23:28:31 -0400146 int chkpt = journal->j_checkpoint_transactions != NULL;
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500147 tid_t tid = 0;
Duane Griffin23f8b792008-10-08 23:28:31 -0400148
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500149 if (journal->j_committing_transaction)
150 tid = journal->j_committing_transaction->t_tid;
Duane Griffin23f8b792008-10-08 23:28:31 -0400151 spin_unlock(&journal->j_list_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400152 write_unlock(&journal->j_state_lock);
Duane Griffin23f8b792008-10-08 23:28:31 -0400153 if (chkpt) {
154 jbd2_log_do_checkpoint(journal);
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500155 } else if (jbd2_cleanup_journal_tail(journal) == 0) {
156 /* We were able to recover space; yay! */
157 ;
158 } else if (tid) {
159 jbd2_log_wait_commit(journal, tid);
Duane Griffin23f8b792008-10-08 23:28:31 -0400160 } else {
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500161 printk(KERN_ERR "%s: needed %d blocks and "
162 "only had %d space available\n",
163 __func__, nblocks, space_left);
164 printk(KERN_ERR "%s: no way to get more "
165 "journal space in %s\n", __func__,
166 journal->j_devname);
167 WARN_ON(1);
Duane Griffin23f8b792008-10-08 23:28:31 -0400168 jbd2_journal_abort(journal, 0);
169 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400170 write_lock(&journal->j_state_lock);
Duane Griffin23f8b792008-10-08 23:28:31 -0400171 } else {
172 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700173 }
174 mutex_unlock(&journal->j_checkpoint_mutex);
175 }
176}
177
178/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700179 * Clean up transaction's list of buffers submitted for io.
180 * We wait for any pending IO to complete and remove any clean
181 * buffers. Note that we take the buffers in the opposite ordering
182 * from the one in which they were submitted for IO.
183 *
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400184 * Return 0 on success, and return <0 if some buffers have failed
185 * to be written out.
186 *
Dave Kleikamp470decc2006-10-11 01:20:57 -0700187 * Called with j_list_lock held.
188 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400189static int __wait_cp_io(journal_t *journal, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700190{
191 struct journal_head *jh;
192 struct buffer_head *bh;
193 tid_t this_tid;
194 int released = 0;
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400195 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700196
197 this_tid = transaction->t_tid;
198restart:
199 /* Did somebody clean up the transaction in the meanwhile? */
200 if (journal->j_checkpoint_transactions != transaction ||
201 transaction->t_tid != this_tid)
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400202 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700203 while (!released && transaction->t_checkpoint_io_list) {
204 jh = transaction->t_checkpoint_io_list;
205 bh = jh2bh(jh);
Jan Karade1b7942011-06-13 15:38:22 -0400206 get_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700207 if (buffer_locked(bh)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700208 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700209 wait_on_buffer(bh);
210 /* the journal_head may have gone by now */
211 BUFFER_TRACE(bh, "brelse");
212 __brelse(bh);
213 spin_lock(&journal->j_list_lock);
214 goto restart;
215 }
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400216 if (unlikely(buffer_write_io_error(bh)))
217 ret = -EIO;
218
Dave Kleikamp470decc2006-10-11 01:20:57 -0700219 /*
220 * Now in whatever state the buffer currently is, we know that
221 * it has been written out and so we can drop it from the list
222 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700223 released = __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700224 __brelse(bh);
225 }
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400226
227 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700228}
229
Dave Kleikamp470decc2006-10-11 01:20:57 -0700230static void
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500231__flush_batch(journal_t *journal, int *batch_count)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700232{
233 int i;
Tao Mad3ad8432011-06-27 12:36:29 -0400234 struct blk_plug plug;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700235
Tao Mad3ad8432011-06-27 12:36:29 -0400236 blk_start_plug(&plug);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200237 for (i = 0; i < *batch_count; i++)
Tao Mad3ad8432011-06-27 12:36:29 -0400238 write_dirty_buffer(journal->j_chkpt_bhs[i], WRITE_SYNC);
239 blk_finish_plug(&plug);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200240
Dave Kleikamp470decc2006-10-11 01:20:57 -0700241 for (i = 0; i < *batch_count; i++) {
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500242 struct buffer_head *bh = journal->j_chkpt_bhs[i];
Dave Kleikamp470decc2006-10-11 01:20:57 -0700243 BUFFER_TRACE(bh, "brelse");
244 __brelse(bh);
245 }
246 *batch_count = 0;
247}
248
249/*
250 * Try to flush one buffer from the checkpoint list to disk.
251 *
252 * Return 1 if something happened which requires us to abort the current
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400253 * scan of the checkpoint list. Return <0 if the buffer has failed to
254 * be written out.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700255 *
256 * Called with j_list_lock held and drops it if 1 is returned
Dave Kleikamp470decc2006-10-11 01:20:57 -0700257 */
258static int __process_buffer(journal_t *journal, struct journal_head *jh,
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500259 int *batch_count, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700260{
261 struct buffer_head *bh = jh2bh(jh);
262 int ret = 0;
263
264 if (buffer_locked(bh)) {
Jan Karade1b7942011-06-13 15:38:22 -0400265 get_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700266 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700267 wait_on_buffer(bh);
268 /* the journal_head may have gone by now */
269 BUFFER_TRACE(bh, "brelse");
270 __brelse(bh);
271 ret = 1;
272 } else if (jh->b_transaction != NULL) {
273 transaction_t *t = jh->b_transaction;
274 tid_t tid = t->t_tid;
275
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500276 transaction->t_chp_stats.cs_forced_to_close++;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700277 spin_unlock(&journal->j_list_lock);
Theodore Ts'o5c2178e2010-10-27 21:30:04 -0400278 if (unlikely(journal->j_flags & JBD2_UNMOUNT))
279 /*
280 * The journal thread is dead; so starting and
281 * waiting for a commit to finish will cause
282 * us to wait for a _very_ long time.
283 */
284 printk(KERN_ERR "JBD2: %s: "
285 "Waiting for Godot: block %llu\n",
286 journal->j_devname,
287 (unsigned long long) bh->b_blocknr);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700288 jbd2_log_start_commit(journal, tid);
289 jbd2_log_wait_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700290 ret = 1;
291 } else if (!buffer_dirty(bh)) {
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400292 ret = 1;
293 if (unlikely(buffer_write_io_error(bh)))
294 ret = -EIO;
Jan Karade1b7942011-06-13 15:38:22 -0400295 get_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700296 BUFFER_TRACE(bh, "remove from checkpoint");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700297 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700298 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700299 __brelse(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700300 } else {
301 /*
302 * Important: we are about to write the buffer, and
303 * possibly block, while still holding the journal lock.
304 * We cannot afford to let the transaction logic start
305 * messing around with this buffer before we write it to
306 * disk, as that would break recoverability.
307 */
308 BUFFER_TRACE(bh, "queue");
309 get_bh(bh);
310 J_ASSERT_BH(bh, !buffer_jwrite(bh));
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500311 journal->j_chkpt_bhs[*batch_count] = bh;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700312 __buffer_relink_io(jh);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500313 transaction->t_chp_stats.cs_written++;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700314 (*batch_count)++;
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500315 if (*batch_count == JBD2_NR_BATCH) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700316 spin_unlock(&journal->j_list_lock);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500317 __flush_batch(journal, batch_count);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700318 ret = 1;
319 }
320 }
321 return ret;
322}
323
324/*
325 * Perform an actual checkpoint. We take the first transaction on the
326 * list of transactions to be checkpointed and send all its buffers
327 * to disk. We submit larger chunks of data at once.
328 *
329 * The journal should be locked before calling this function.
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400330 * Called with j_checkpoint_mutex held.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700331 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700332int jbd2_log_do_checkpoint(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700333{
334 transaction_t *transaction;
335 tid_t this_tid;
336 int result;
337
338 jbd_debug(1, "Start checkpoint\n");
339
340 /*
341 * First thing: if there are any transactions in the log which
342 * don't need checkpointing, just eliminate them from the
343 * journal straight away.
344 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700345 result = jbd2_cleanup_journal_tail(journal);
Theodore Ts'o879c5e62009-06-17 11:47:48 -0400346 trace_jbd2_checkpoint(journal, result);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700347 jbd_debug(1, "cleanup_journal_tail returned %d\n", result);
348 if (result <= 0)
349 return result;
350
351 /*
352 * OK, we need to start writing disk blocks. Take one transaction
353 * and write it.
354 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400355 result = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700356 spin_lock(&journal->j_list_lock);
357 if (!journal->j_checkpoint_transactions)
358 goto out;
359 transaction = journal->j_checkpoint_transactions;
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500360 if (transaction->t_chp_stats.cs_chp_time == 0)
361 transaction->t_chp_stats.cs_chp_time = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700362 this_tid = transaction->t_tid;
363restart:
364 /*
365 * If someone cleaned up this transaction while we slept, we're
366 * done (maybe it's a new transaction, but it fell at the same
367 * address).
368 */
369 if (journal->j_checkpoint_transactions == transaction &&
370 transaction->t_tid == this_tid) {
371 int batch_count = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700372 struct journal_head *jh;
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400373 int retry = 0, err;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700374
375 while (!retry && transaction->t_checkpoint_list) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700376 jh = transaction->t_checkpoint_list;
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500377 retry = __process_buffer(journal, jh, &batch_count,
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500378 transaction);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400379 if (retry < 0 && !result)
380 result = retry;
Nick Piggin95c354f2008-01-30 13:31:20 +0100381 if (!retry && (need_resched() ||
382 spin_needbreak(&journal->j_list_lock))) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700383 spin_unlock(&journal->j_list_lock);
384 retry = 1;
385 break;
386 }
387 }
388
389 if (batch_count) {
390 if (!retry) {
391 spin_unlock(&journal->j_list_lock);
392 retry = 1;
393 }
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500394 __flush_batch(journal, &batch_count);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700395 }
396
397 if (retry) {
398 spin_lock(&journal->j_list_lock);
399 goto restart;
400 }
401 /*
402 * Now we have cleaned up the first transaction's checkpoint
403 * list. Let's clean up the second one
404 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400405 err = __wait_cp_io(journal, transaction);
406 if (!result)
407 result = err;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700408 }
409out:
410 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700411 if (result < 0)
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400412 jbd2_journal_abort(journal, result);
413 else
414 result = jbd2_cleanup_journal_tail(journal);
415
416 return (result < 0) ? result : 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700417}
418
419/*
420 * Check the list of checkpoint transactions for the journal to see if
421 * we have already got rid of any since the last update of the log tail
422 * in the journal superblock. If so, we can instantly roll the
423 * superblock forward to remove those transactions from the log.
424 *
425 * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
426 *
427 * Called with the journal lock held.
428 *
429 * This is the only part of the journaling code which really needs to be
430 * aware of transaction aborts. Checkpointing involves writing to the
431 * main filesystem area rather than to the journal, so it can proceed
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400432 * even in abort state, but we must not update the super block if
433 * checkpointing may have failed. Otherwise, we would lose some metadata
434 * buffers which should be written-back to the filesystem.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700435 */
436
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700437int jbd2_cleanup_journal_tail(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700438{
Dave Kleikamp470decc2006-10-11 01:20:57 -0700439 tid_t first_tid;
Jan Kara79feb522012-03-13 22:22:54 -0400440 unsigned long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700441
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400442 if (is_journal_aborted(journal))
443 return 1;
444
Jan Kara79feb522012-03-13 22:22:54 -0400445 if (!jbd2_journal_get_log_tail(journal, &first_tid, &blocknr))
446 return 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700447 J_ASSERT(blocknr != 0);
448
Theodore Ts'occ3e1be2009-12-23 06:52:08 -0500449 /*
Jan Kara79feb522012-03-13 22:22:54 -0400450 * We need to make sure that any blocks that were recently written out
451 * --- perhaps by jbd2_log_do_checkpoint() --- are flushed out before
452 * we drop the transactions from the journal. It's unlikely this will
453 * be necessary, especially with an appropriately sized journal, but we
454 * need this to guarantee correctness. Fortunately
455 * jbd2_cleanup_journal_tail() doesn't get called all that often.
Theodore Ts'occ3e1be2009-12-23 06:52:08 -0500456 */
Jan Kara79feb522012-03-13 22:22:54 -0400457 if (journal->j_flags & JBD2_BARRIER)
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200458 blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
Jan Kara79feb522012-03-13 22:22:54 -0400459
460 __jbd2_update_log_tail(journal, first_tid, blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700461 return 0;
462}
463
464
465/* Checkpoint list management */
466
467/*
468 * journal_clean_one_cp_list
469 *
Jan Karade1b7942011-06-13 15:38:22 -0400470 * Find all the written-back checkpoint buffers in the given list and
471 * release them.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700472 *
473 * Called with the journal locked.
474 * Called with j_list_lock held.
Paul Bolle90802ed2011-12-05 13:00:34 +0100475 * Returns number of buffers reaped (for debug)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700476 */
477
478static int journal_clean_one_cp_list(struct journal_head *jh, int *released)
479{
480 struct journal_head *last_jh;
481 struct journal_head *next_jh = jh;
482 int ret, freed = 0;
483
484 *released = 0;
485 if (!jh)
486 return 0;
487
488 last_jh = jh->b_cpprev;
489 do {
490 jh = next_jh;
491 next_jh = jh->b_cpnext;
Jan Kara932bb302012-03-13 22:45:25 -0400492 ret = __try_to_free_cp_buf(jh);
493 if (ret) {
494 freed++;
495 if (ret == 2) {
496 *released = 1;
497 return freed;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700498 }
499 }
500 /*
501 * This function only frees up some memory
502 * if possible so we dont have an obligation
503 * to finish processing. Bail out if preemption
504 * requested:
505 */
506 if (need_resched())
507 return freed;
508 } while (jh != last_jh);
509
510 return freed;
511}
512
513/*
514 * journal_clean_checkpoint_list
515 *
516 * Find all the written-back checkpoint buffers in the journal and release them.
517 *
518 * Called with the journal locked.
519 * Called with j_list_lock held.
520 * Returns number of buffers reaped (for debug)
521 */
522
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700523int __jbd2_journal_clean_checkpoint_list(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700524{
525 transaction_t *transaction, *last_transaction, *next_transaction;
526 int ret = 0;
527 int released;
528
529 transaction = journal->j_checkpoint_transactions;
530 if (!transaction)
531 goto out;
532
533 last_transaction = transaction->t_cpprev;
534 next_transaction = transaction;
535 do {
536 transaction = next_transaction;
537 next_transaction = transaction->t_cpnext;
538 ret += journal_clean_one_cp_list(transaction->
539 t_checkpoint_list, &released);
540 /*
541 * This function only frees up some memory if possible so we
542 * dont have an obligation to finish processing. Bail out if
543 * preemption requested:
544 */
545 if (need_resched())
546 goto out;
547 if (released)
548 continue;
549 /*
550 * It is essential that we are as careful as in the case of
551 * t_checkpoint_list with removing the buffer from the list as
552 * we can possibly see not yet submitted buffers on io_list
553 */
554 ret += journal_clean_one_cp_list(transaction->
555 t_checkpoint_io_list, &released);
556 if (need_resched())
557 goto out;
558 } while (transaction != last_transaction);
559out:
560 return ret;
561}
562
563/*
564 * journal_remove_checkpoint: called after a buffer has been committed
565 * to disk (either by being write-back flushed to disk, or being
566 * committed to the log).
567 *
568 * We cannot safely clean a transaction out of the log until all of the
569 * buffer updates committed in that transaction have safely been stored
570 * elsewhere on disk. To achieve this, all of the buffers in a
571 * transaction need to be maintained on the transaction's checkpoint
572 * lists until they have been rewritten, at which point this function is
573 * called to remove the buffer from the existing transaction's
574 * checkpoint lists.
575 *
576 * The function returns 1 if it frees the transaction, 0 otherwise.
Jan Karade1b7942011-06-13 15:38:22 -0400577 * The function can free jh and bh.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700578 *
Dave Kleikamp470decc2006-10-11 01:20:57 -0700579 * This function is called with j_list_lock held.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700580 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700581int __jbd2_journal_remove_checkpoint(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700582{
Theodore Ts'obf699322009-09-30 00:32:06 -0400583 struct transaction_chp_stats_s *stats;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700584 transaction_t *transaction;
585 journal_t *journal;
586 int ret = 0;
587
588 JBUFFER_TRACE(jh, "entry");
589
590 if ((transaction = jh->b_cp_transaction) == NULL) {
591 JBUFFER_TRACE(jh, "not on transaction");
592 goto out;
593 }
594 journal = transaction->t_journal;
595
Jan Karade1b7942011-06-13 15:38:22 -0400596 JBUFFER_TRACE(jh, "removing from transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700597 __buffer_unlink(jh);
598 jh->b_cp_transaction = NULL;
Jan Karade1b7942011-06-13 15:38:22 -0400599 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700600
601 if (transaction->t_checkpoint_list != NULL ||
602 transaction->t_checkpoint_io_list != NULL)
603 goto out;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700604
605 /*
606 * There is one special case to worry about: if we have just pulled the
Jan Karaf5a7a6b2008-01-28 23:58:27 -0500607 * buffer off a running or committing transaction's checkpoing list,
608 * then even if the checkpoint list is empty, the transaction obviously
609 * cannot be dropped!
Dave Kleikamp470decc2006-10-11 01:20:57 -0700610 *
Jan Karaf5a7a6b2008-01-28 23:58:27 -0500611 * The locking here around t_state is a bit sleazy.
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700612 * See the comment at the end of jbd2_journal_commit_transaction().
Dave Kleikamp470decc2006-10-11 01:20:57 -0700613 */
Jan Karade1b7942011-06-13 15:38:22 -0400614 if (transaction->t_state != T_FINISHED)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700615 goto out;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700616
617 /* OK, that was the last buffer for the transaction: we can now
618 safely remove this transaction from the log */
Theodore Ts'obf699322009-09-30 00:32:06 -0400619 stats = &transaction->t_chp_stats;
620 if (stats->cs_chp_time)
621 stats->cs_chp_time = jbd2_time_diff(stats->cs_chp_time,
622 jiffies);
623 trace_jbd2_checkpoint_stats(journal->j_fs_dev->bd_dev,
624 transaction->t_tid, stats);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700625
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700626 __jbd2_journal_drop_transaction(journal, transaction);
Yongqiang Yang0c2022e2012-02-20 17:53:02 -0500627 jbd2_journal_free_transaction(transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700628 ret = 1;
629out:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700630 return ret;
631}
632
633/*
634 * journal_insert_checkpoint: put a committed buffer onto a checkpoint
635 * list so that we know when it is safe to clean the transaction out of
636 * the log.
637 *
638 * Called with the journal locked.
639 * Called with j_list_lock held.
640 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700641void __jbd2_journal_insert_checkpoint(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700642 transaction_t *transaction)
643{
644 JBUFFER_TRACE(jh, "entry");
645 J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh)));
646 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
647
Jan Karade1b7942011-06-13 15:38:22 -0400648 /* Get reference for checkpointing transaction */
649 jbd2_journal_grab_journal_head(jh2bh(jh));
Dave Kleikamp470decc2006-10-11 01:20:57 -0700650 jh->b_cp_transaction = transaction;
651
652 if (!transaction->t_checkpoint_list) {
653 jh->b_cpnext = jh->b_cpprev = jh;
654 } else {
655 jh->b_cpnext = transaction->t_checkpoint_list;
656 jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev;
657 jh->b_cpprev->b_cpnext = jh;
658 jh->b_cpnext->b_cpprev = jh;
659 }
660 transaction->t_checkpoint_list = jh;
661}
662
663/*
664 * We've finished with this transaction structure: adios...
665 *
666 * The transaction must have no links except for the checkpoint by this
667 * point.
668 *
669 * Called with the journal locked.
670 * Called with j_list_lock held.
671 */
672
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700673void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700674{
675 assert_spin_locked(&journal->j_list_lock);
676 if (transaction->t_cpnext) {
677 transaction->t_cpnext->t_cpprev = transaction->t_cpprev;
678 transaction->t_cpprev->t_cpnext = transaction->t_cpnext;
679 if (journal->j_checkpoint_transactions == transaction)
680 journal->j_checkpoint_transactions =
681 transaction->t_cpnext;
682 if (journal->j_checkpoint_transactions == transaction)
683 journal->j_checkpoint_transactions = NULL;
684 }
685
686 J_ASSERT(transaction->t_state == T_FINISHED);
687 J_ASSERT(transaction->t_buffers == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700688 J_ASSERT(transaction->t_forget == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700689 J_ASSERT(transaction->t_shadow_list == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700690 J_ASSERT(transaction->t_checkpoint_list == NULL);
691 J_ASSERT(transaction->t_checkpoint_io_list == NULL);
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400692 J_ASSERT(atomic_read(&transaction->t_updates) == 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700693 J_ASSERT(journal->j_committing_transaction != transaction);
694 J_ASSERT(journal->j_running_transaction != transaction);
695
Seiji Aguchi2201c592012-02-20 17:53:01 -0500696 trace_jbd2_drop_transaction(journal, transaction);
697
Dave Kleikamp470decc2006-10-11 01:20:57 -0700698 jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700699}