blob: 7f34f4716165311f66b62b81dd25a32e899dcf80 [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) {
Paul Gortmaker0ef54182013-06-12 22:47:35 -0400159 /*
160 * jbd2_journal_commit_transaction() may want
161 * to take the checkpoint_mutex if JBD2_FLUSHED
162 * is set. So we need to temporarily drop it.
163 */
164 mutex_unlock(&journal->j_checkpoint_mutex);
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500165 jbd2_log_wait_commit(journal, tid);
Paul Gortmaker0ef54182013-06-12 22:47:35 -0400166 write_lock(&journal->j_state_lock);
167 continue;
Duane Griffin23f8b792008-10-08 23:28:31 -0400168 } else {
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500169 printk(KERN_ERR "%s: needed %d blocks and "
170 "only had %d space available\n",
171 __func__, nblocks, space_left);
172 printk(KERN_ERR "%s: no way to get more "
173 "journal space in %s\n", __func__,
174 journal->j_devname);
175 WARN_ON(1);
Duane Griffin23f8b792008-10-08 23:28:31 -0400176 jbd2_journal_abort(journal, 0);
177 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400178 write_lock(&journal->j_state_lock);
Duane Griffin23f8b792008-10-08 23:28:31 -0400179 } else {
180 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700181 }
182 mutex_unlock(&journal->j_checkpoint_mutex);
183 }
184}
185
186/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700187 * Clean up transaction's list of buffers submitted for io.
188 * We wait for any pending IO to complete and remove any clean
189 * buffers. Note that we take the buffers in the opposite ordering
190 * from the one in which they were submitted for IO.
191 *
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400192 * Return 0 on success, and return <0 if some buffers have failed
193 * to be written out.
194 *
Dave Kleikamp470decc2006-10-11 01:20:57 -0700195 * Called with j_list_lock held.
196 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400197static int __wait_cp_io(journal_t *journal, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700198{
199 struct journal_head *jh;
200 struct buffer_head *bh;
201 tid_t this_tid;
202 int released = 0;
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400203 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700204
205 this_tid = transaction->t_tid;
206restart:
207 /* Did somebody clean up the transaction in the meanwhile? */
208 if (journal->j_checkpoint_transactions != transaction ||
209 transaction->t_tid != this_tid)
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400210 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700211 while (!released && transaction->t_checkpoint_io_list) {
212 jh = transaction->t_checkpoint_io_list;
213 bh = jh2bh(jh);
Jan Karade1b7942011-06-13 15:38:22 -0400214 get_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700215 if (buffer_locked(bh)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700216 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700217 wait_on_buffer(bh);
218 /* the journal_head may have gone by now */
219 BUFFER_TRACE(bh, "brelse");
220 __brelse(bh);
221 spin_lock(&journal->j_list_lock);
222 goto restart;
223 }
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400224 if (unlikely(buffer_write_io_error(bh)))
225 ret = -EIO;
226
Dave Kleikamp470decc2006-10-11 01:20:57 -0700227 /*
228 * Now in whatever state the buffer currently is, we know that
229 * it has been written out and so we can drop it from the list
230 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700231 released = __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700232 __brelse(bh);
233 }
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400234
235 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700236}
237
Dave Kleikamp470decc2006-10-11 01:20:57 -0700238static void
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500239__flush_batch(journal_t *journal, int *batch_count)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700240{
241 int i;
Tao Mad3ad8432011-06-27 12:36:29 -0400242 struct blk_plug plug;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700243
Tao Mad3ad8432011-06-27 12:36:29 -0400244 blk_start_plug(&plug);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200245 for (i = 0; i < *batch_count; i++)
Tao Mad3ad8432011-06-27 12:36:29 -0400246 write_dirty_buffer(journal->j_chkpt_bhs[i], WRITE_SYNC);
247 blk_finish_plug(&plug);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200248
Dave Kleikamp470decc2006-10-11 01:20:57 -0700249 for (i = 0; i < *batch_count; i++) {
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500250 struct buffer_head *bh = journal->j_chkpt_bhs[i];
Dave Kleikamp470decc2006-10-11 01:20:57 -0700251 BUFFER_TRACE(bh, "brelse");
252 __brelse(bh);
253 }
254 *batch_count = 0;
255}
256
257/*
258 * Try to flush one buffer from the checkpoint list to disk.
259 *
260 * Return 1 if something happened which requires us to abort the current
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400261 * scan of the checkpoint list. Return <0 if the buffer has failed to
262 * be written out.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700263 *
264 * Called with j_list_lock held and drops it if 1 is returned
Dave Kleikamp470decc2006-10-11 01:20:57 -0700265 */
266static int __process_buffer(journal_t *journal, struct journal_head *jh,
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500267 int *batch_count, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700268{
269 struct buffer_head *bh = jh2bh(jh);
270 int ret = 0;
271
272 if (buffer_locked(bh)) {
Jan Karade1b7942011-06-13 15:38:22 -0400273 get_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700274 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700275 wait_on_buffer(bh);
276 /* the journal_head may have gone by now */
277 BUFFER_TRACE(bh, "brelse");
278 __brelse(bh);
279 ret = 1;
280 } else if (jh->b_transaction != NULL) {
281 transaction_t *t = jh->b_transaction;
282 tid_t tid = t->t_tid;
283
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500284 transaction->t_chp_stats.cs_forced_to_close++;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700285 spin_unlock(&journal->j_list_lock);
Theodore Ts'o5c2178e2010-10-27 21:30:04 -0400286 if (unlikely(journal->j_flags & JBD2_UNMOUNT))
287 /*
288 * The journal thread is dead; so starting and
289 * waiting for a commit to finish will cause
290 * us to wait for a _very_ long time.
291 */
292 printk(KERN_ERR "JBD2: %s: "
293 "Waiting for Godot: block %llu\n",
294 journal->j_devname,
295 (unsigned long long) bh->b_blocknr);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700296 jbd2_log_start_commit(journal, tid);
297 jbd2_log_wait_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700298 ret = 1;
299 } else if (!buffer_dirty(bh)) {
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400300 ret = 1;
301 if (unlikely(buffer_write_io_error(bh)))
302 ret = -EIO;
Jan Karade1b7942011-06-13 15:38:22 -0400303 get_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700304 BUFFER_TRACE(bh, "remove from checkpoint");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700305 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700306 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700307 __brelse(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700308 } else {
309 /*
310 * Important: we are about to write the buffer, and
311 * possibly block, while still holding the journal lock.
312 * We cannot afford to let the transaction logic start
313 * messing around with this buffer before we write it to
314 * disk, as that would break recoverability.
315 */
316 BUFFER_TRACE(bh, "queue");
317 get_bh(bh);
318 J_ASSERT_BH(bh, !buffer_jwrite(bh));
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500319 journal->j_chkpt_bhs[*batch_count] = bh;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700320 __buffer_relink_io(jh);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500321 transaction->t_chp_stats.cs_written++;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700322 (*batch_count)++;
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500323 if (*batch_count == JBD2_NR_BATCH) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700324 spin_unlock(&journal->j_list_lock);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500325 __flush_batch(journal, batch_count);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700326 ret = 1;
327 }
328 }
329 return ret;
330}
331
332/*
333 * Perform an actual checkpoint. We take the first transaction on the
334 * list of transactions to be checkpointed and send all its buffers
335 * to disk. We submit larger chunks of data at once.
336 *
337 * The journal should be locked before calling this function.
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400338 * Called with j_checkpoint_mutex held.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700339 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700340int jbd2_log_do_checkpoint(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700341{
342 transaction_t *transaction;
343 tid_t this_tid;
344 int result;
345
346 jbd_debug(1, "Start checkpoint\n");
347
348 /*
349 * First thing: if there are any transactions in the log which
350 * don't need checkpointing, just eliminate them from the
351 * journal straight away.
352 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700353 result = jbd2_cleanup_journal_tail(journal);
Theodore Ts'o879c5e62009-06-17 11:47:48 -0400354 trace_jbd2_checkpoint(journal, result);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700355 jbd_debug(1, "cleanup_journal_tail returned %d\n", result);
356 if (result <= 0)
357 return result;
358
359 /*
360 * OK, we need to start writing disk blocks. Take one transaction
361 * and write it.
362 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400363 result = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700364 spin_lock(&journal->j_list_lock);
365 if (!journal->j_checkpoint_transactions)
366 goto out;
367 transaction = journal->j_checkpoint_transactions;
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500368 if (transaction->t_chp_stats.cs_chp_time == 0)
369 transaction->t_chp_stats.cs_chp_time = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700370 this_tid = transaction->t_tid;
371restart:
372 /*
373 * If someone cleaned up this transaction while we slept, we're
374 * done (maybe it's a new transaction, but it fell at the same
375 * address).
376 */
377 if (journal->j_checkpoint_transactions == transaction &&
378 transaction->t_tid == this_tid) {
379 int batch_count = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700380 struct journal_head *jh;
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400381 int retry = 0, err;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700382
383 while (!retry && transaction->t_checkpoint_list) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700384 jh = transaction->t_checkpoint_list;
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500385 retry = __process_buffer(journal, jh, &batch_count,
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500386 transaction);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400387 if (retry < 0 && !result)
388 result = retry;
Nick Piggin95c354f2008-01-30 13:31:20 +0100389 if (!retry && (need_resched() ||
390 spin_needbreak(&journal->j_list_lock))) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700391 spin_unlock(&journal->j_list_lock);
392 retry = 1;
393 break;
394 }
395 }
396
397 if (batch_count) {
398 if (!retry) {
399 spin_unlock(&journal->j_list_lock);
400 retry = 1;
401 }
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500402 __flush_batch(journal, &batch_count);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700403 }
404
405 if (retry) {
406 spin_lock(&journal->j_list_lock);
407 goto restart;
408 }
409 /*
410 * Now we have cleaned up the first transaction's checkpoint
411 * list. Let's clean up the second one
412 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400413 err = __wait_cp_io(journal, transaction);
414 if (!result)
415 result = err;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700416 }
417out:
418 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700419 if (result < 0)
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400420 jbd2_journal_abort(journal, result);
421 else
422 result = jbd2_cleanup_journal_tail(journal);
423
424 return (result < 0) ? result : 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700425}
426
427/*
428 * Check the list of checkpoint transactions for the journal to see if
429 * we have already got rid of any since the last update of the log tail
430 * in the journal superblock. If so, we can instantly roll the
431 * superblock forward to remove those transactions from the log.
432 *
433 * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
434 *
435 * Called with the journal lock held.
436 *
437 * This is the only part of the journaling code which really needs to be
438 * aware of transaction aborts. Checkpointing involves writing to the
439 * main filesystem area rather than to the journal, so it can proceed
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400440 * even in abort state, but we must not update the super block if
441 * checkpointing may have failed. Otherwise, we would lose some metadata
442 * buffers which should be written-back to the filesystem.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700443 */
444
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700445int jbd2_cleanup_journal_tail(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700446{
Dave Kleikamp470decc2006-10-11 01:20:57 -0700447 tid_t first_tid;
Jan Kara79feb522012-03-13 22:22:54 -0400448 unsigned long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700449
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400450 if (is_journal_aborted(journal))
451 return 1;
452
Jan Kara79feb522012-03-13 22:22:54 -0400453 if (!jbd2_journal_get_log_tail(journal, &first_tid, &blocknr))
454 return 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700455 J_ASSERT(blocknr != 0);
456
Theodore Ts'occ3e1be2009-12-23 06:52:08 -0500457 /*
Jan Kara79feb522012-03-13 22:22:54 -0400458 * We need to make sure that any blocks that were recently written out
459 * --- perhaps by jbd2_log_do_checkpoint() --- are flushed out before
460 * we drop the transactions from the journal. It's unlikely this will
461 * be necessary, especially with an appropriately sized journal, but we
462 * need this to guarantee correctness. Fortunately
463 * jbd2_cleanup_journal_tail() doesn't get called all that often.
Theodore Ts'occ3e1be2009-12-23 06:52:08 -0500464 */
Jan Kara79feb522012-03-13 22:22:54 -0400465 if (journal->j_flags & JBD2_BARRIER)
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200466 blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
Jan Kara79feb522012-03-13 22:22:54 -0400467
468 __jbd2_update_log_tail(journal, first_tid, blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700469 return 0;
470}
471
472
473/* Checkpoint list management */
474
475/*
476 * journal_clean_one_cp_list
477 *
Jan Karade1b7942011-06-13 15:38:22 -0400478 * Find all the written-back checkpoint buffers in the given list and
479 * release them.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700480 *
481 * Called with the journal locked.
482 * Called with j_list_lock held.
Paul Bolle90802ed2011-12-05 13:00:34 +0100483 * Returns number of buffers reaped (for debug)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700484 */
485
486static int journal_clean_one_cp_list(struct journal_head *jh, int *released)
487{
488 struct journal_head *last_jh;
489 struct journal_head *next_jh = jh;
490 int ret, freed = 0;
491
492 *released = 0;
493 if (!jh)
494 return 0;
495
496 last_jh = jh->b_cpprev;
497 do {
498 jh = next_jh;
499 next_jh = jh->b_cpnext;
Jan Kara932bb302012-03-13 22:45:25 -0400500 ret = __try_to_free_cp_buf(jh);
501 if (ret) {
502 freed++;
503 if (ret == 2) {
504 *released = 1;
505 return freed;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700506 }
507 }
508 /*
509 * This function only frees up some memory
510 * if possible so we dont have an obligation
511 * to finish processing. Bail out if preemption
512 * requested:
513 */
514 if (need_resched())
515 return freed;
516 } while (jh != last_jh);
517
518 return freed;
519}
520
521/*
522 * journal_clean_checkpoint_list
523 *
524 * Find all the written-back checkpoint buffers in the journal and release them.
525 *
526 * Called with the journal locked.
527 * Called with j_list_lock held.
528 * Returns number of buffers reaped (for debug)
529 */
530
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700531int __jbd2_journal_clean_checkpoint_list(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700532{
533 transaction_t *transaction, *last_transaction, *next_transaction;
534 int ret = 0;
535 int released;
536
537 transaction = journal->j_checkpoint_transactions;
538 if (!transaction)
539 goto out;
540
541 last_transaction = transaction->t_cpprev;
542 next_transaction = transaction;
543 do {
544 transaction = next_transaction;
545 next_transaction = transaction->t_cpnext;
546 ret += journal_clean_one_cp_list(transaction->
547 t_checkpoint_list, &released);
548 /*
549 * This function only frees up some memory if possible so we
550 * dont have an obligation to finish processing. Bail out if
551 * preemption requested:
552 */
553 if (need_resched())
554 goto out;
555 if (released)
556 continue;
557 /*
558 * It is essential that we are as careful as in the case of
559 * t_checkpoint_list with removing the buffer from the list as
560 * we can possibly see not yet submitted buffers on io_list
561 */
562 ret += journal_clean_one_cp_list(transaction->
563 t_checkpoint_io_list, &released);
564 if (need_resched())
565 goto out;
566 } while (transaction != last_transaction);
567out:
568 return ret;
569}
570
571/*
572 * journal_remove_checkpoint: called after a buffer has been committed
573 * to disk (either by being write-back flushed to disk, or being
574 * committed to the log).
575 *
576 * We cannot safely clean a transaction out of the log until all of the
577 * buffer updates committed in that transaction have safely been stored
578 * elsewhere on disk. To achieve this, all of the buffers in a
579 * transaction need to be maintained on the transaction's checkpoint
580 * lists until they have been rewritten, at which point this function is
581 * called to remove the buffer from the existing transaction's
582 * checkpoint lists.
583 *
584 * The function returns 1 if it frees the transaction, 0 otherwise.
Jan Karade1b7942011-06-13 15:38:22 -0400585 * The function can free jh and bh.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700586 *
Dave Kleikamp470decc2006-10-11 01:20:57 -0700587 * This function is called with j_list_lock held.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700588 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700589int __jbd2_journal_remove_checkpoint(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700590{
Theodore Ts'obf699322009-09-30 00:32:06 -0400591 struct transaction_chp_stats_s *stats;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700592 transaction_t *transaction;
593 journal_t *journal;
594 int ret = 0;
595
596 JBUFFER_TRACE(jh, "entry");
597
598 if ((transaction = jh->b_cp_transaction) == NULL) {
599 JBUFFER_TRACE(jh, "not on transaction");
600 goto out;
601 }
602 journal = transaction->t_journal;
603
Jan Karade1b7942011-06-13 15:38:22 -0400604 JBUFFER_TRACE(jh, "removing from transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700605 __buffer_unlink(jh);
606 jh->b_cp_transaction = NULL;
Jan Karade1b7942011-06-13 15:38:22 -0400607 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700608
609 if (transaction->t_checkpoint_list != NULL ||
610 transaction->t_checkpoint_io_list != NULL)
611 goto out;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700612
613 /*
614 * There is one special case to worry about: if we have just pulled the
Jan Karaf5a7a6b2008-01-28 23:58:27 -0500615 * buffer off a running or committing transaction's checkpoing list,
616 * then even if the checkpoint list is empty, the transaction obviously
617 * cannot be dropped!
Dave Kleikamp470decc2006-10-11 01:20:57 -0700618 *
Jan Karaf5a7a6b2008-01-28 23:58:27 -0500619 * The locking here around t_state is a bit sleazy.
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700620 * See the comment at the end of jbd2_journal_commit_transaction().
Dave Kleikamp470decc2006-10-11 01:20:57 -0700621 */
Jan Karade1b7942011-06-13 15:38:22 -0400622 if (transaction->t_state != T_FINISHED)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700623 goto out;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700624
625 /* OK, that was the last buffer for the transaction: we can now
626 safely remove this transaction from the log */
Theodore Ts'obf699322009-09-30 00:32:06 -0400627 stats = &transaction->t_chp_stats;
628 if (stats->cs_chp_time)
629 stats->cs_chp_time = jbd2_time_diff(stats->cs_chp_time,
630 jiffies);
631 trace_jbd2_checkpoint_stats(journal->j_fs_dev->bd_dev,
632 transaction->t_tid, stats);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700633
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700634 __jbd2_journal_drop_transaction(journal, transaction);
Yongqiang Yang0c2022e2012-02-20 17:53:02 -0500635 jbd2_journal_free_transaction(transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700636 ret = 1;
637out:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700638 return ret;
639}
640
641/*
642 * journal_insert_checkpoint: put a committed buffer onto a checkpoint
643 * list so that we know when it is safe to clean the transaction out of
644 * the log.
645 *
646 * Called with the journal locked.
647 * Called with j_list_lock held.
648 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700649void __jbd2_journal_insert_checkpoint(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700650 transaction_t *transaction)
651{
652 JBUFFER_TRACE(jh, "entry");
653 J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh)));
654 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
655
Jan Karade1b7942011-06-13 15:38:22 -0400656 /* Get reference for checkpointing transaction */
657 jbd2_journal_grab_journal_head(jh2bh(jh));
Dave Kleikamp470decc2006-10-11 01:20:57 -0700658 jh->b_cp_transaction = transaction;
659
660 if (!transaction->t_checkpoint_list) {
661 jh->b_cpnext = jh->b_cpprev = jh;
662 } else {
663 jh->b_cpnext = transaction->t_checkpoint_list;
664 jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev;
665 jh->b_cpprev->b_cpnext = jh;
666 jh->b_cpnext->b_cpprev = jh;
667 }
668 transaction->t_checkpoint_list = jh;
669}
670
671/*
672 * We've finished with this transaction structure: adios...
673 *
674 * The transaction must have no links except for the checkpoint by this
675 * point.
676 *
677 * Called with the journal locked.
678 * Called with j_list_lock held.
679 */
680
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700681void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700682{
683 assert_spin_locked(&journal->j_list_lock);
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);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700696 J_ASSERT(transaction->t_forget == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700697 J_ASSERT(transaction->t_shadow_list == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700698 J_ASSERT(transaction->t_checkpoint_list == NULL);
699 J_ASSERT(transaction->t_checkpoint_io_list == NULL);
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400700 J_ASSERT(atomic_read(&transaction->t_updates) == 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700701 J_ASSERT(journal->j_committing_transaction != transaction);
702 J_ASSERT(journal->j_running_transaction != transaction);
703
Seiji Aguchi2201c592012-02-20 17:53:01 -0500704 trace_jbd2_drop_transaction(journal, transaction);
705
Dave Kleikamp470decc2006-10-11 01:20:57 -0700706 jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700707}