blob: 30beb11ef928f35c75d52e4d1f68857fe370172c [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
91 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
92 */
93static int __try_to_free_cp_buf(struct journal_head *jh)
94{
95 int ret = 0;
96 struct buffer_head *bh = jh2bh(jh);
97
Hidehiro Kawai44519fa2008-10-10 20:29:13 -040098 if (jh->b_jlist == BJ_None && !buffer_locked(bh) &&
99 !buffer_dirty(bh) && !buffer_write_io_error(bh)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700100 JBUFFER_TRACE(jh, "remove from checkpoint list");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700101 ret = __jbd2_journal_remove_checkpoint(jh) + 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700102 jbd_unlock_bh_state(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700103 jbd2_journal_remove_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700104 BUFFER_TRACE(bh, "release");
105 __brelse(bh);
106 } else {
107 jbd_unlock_bh_state(bh);
108 }
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;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700121 assert_spin_locked(&journal->j_state_lock);
122
123 nblocks = jbd_space_needed(journal);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700124 while (__jbd2_log_space_left(journal) < nblocks) {
125 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700126 return;
127 spin_unlock(&journal->j_state_lock);
128 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 */
141 spin_lock(&journal->j_state_lock);
Duane Griffin23f8b792008-10-08 23:28:31 -0400142 spin_lock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700143 nblocks = jbd_space_needed(journal);
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500144 space_left = __jbd2_log_space_left(journal);
145 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);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700152 spin_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 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700170 spin_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/*
179 * We were unable to perform jbd_trylock_bh_state() inside j_list_lock.
180 * The caller must restart a list walk. Wait for someone else to run
181 * jbd_unlock_bh_state().
182 */
183static void jbd_sync_bh(journal_t *journal, struct buffer_head *bh)
184 __releases(journal->j_list_lock)
185{
186 get_bh(bh);
187 spin_unlock(&journal->j_list_lock);
188 jbd_lock_bh_state(bh);
189 jbd_unlock_bh_state(bh);
190 put_bh(bh);
191}
192
193/*
194 * Clean up transaction's list of buffers submitted for io.
195 * We wait for any pending IO to complete and remove any clean
196 * buffers. Note that we take the buffers in the opposite ordering
197 * from the one in which they were submitted for IO.
198 *
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400199 * Return 0 on success, and return <0 if some buffers have failed
200 * to be written out.
201 *
Dave Kleikamp470decc2006-10-11 01:20:57 -0700202 * Called with j_list_lock held.
203 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400204static int __wait_cp_io(journal_t *journal, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700205{
206 struct journal_head *jh;
207 struct buffer_head *bh;
208 tid_t this_tid;
209 int released = 0;
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400210 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700211
212 this_tid = transaction->t_tid;
213restart:
214 /* Did somebody clean up the transaction in the meanwhile? */
215 if (journal->j_checkpoint_transactions != transaction ||
216 transaction->t_tid != this_tid)
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400217 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700218 while (!released && transaction->t_checkpoint_io_list) {
219 jh = transaction->t_checkpoint_io_list;
220 bh = jh2bh(jh);
221 if (!jbd_trylock_bh_state(bh)) {
222 jbd_sync_bh(journal, bh);
223 spin_lock(&journal->j_list_lock);
224 goto restart;
225 }
226 if (buffer_locked(bh)) {
227 atomic_inc(&bh->b_count);
228 spin_unlock(&journal->j_list_lock);
229 jbd_unlock_bh_state(bh);
230 wait_on_buffer(bh);
231 /* the journal_head may have gone by now */
232 BUFFER_TRACE(bh, "brelse");
233 __brelse(bh);
234 spin_lock(&journal->j_list_lock);
235 goto restart;
236 }
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400237 if (unlikely(buffer_write_io_error(bh)))
238 ret = -EIO;
239
Dave Kleikamp470decc2006-10-11 01:20:57 -0700240 /*
241 * Now in whatever state the buffer currently is, we know that
242 * it has been written out and so we can drop it from the list
243 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700244 released = __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700245 jbd_unlock_bh_state(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700246 jbd2_journal_remove_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700247 __brelse(bh);
248 }
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400249
250 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700251}
252
Dave Kleikamp470decc2006-10-11 01:20:57 -0700253static void
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500254__flush_batch(journal_t *journal, int *batch_count)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700255{
256 int i;
257
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500258 ll_rw_block(SWRITE, *batch_count, journal->j_chkpt_bhs);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700259 for (i = 0; i < *batch_count; i++) {
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500260 struct buffer_head *bh = journal->j_chkpt_bhs[i];
Dave Kleikamp470decc2006-10-11 01:20:57 -0700261 clear_buffer_jwrite(bh);
262 BUFFER_TRACE(bh, "brelse");
263 __brelse(bh);
264 }
265 *batch_count = 0;
266}
267
268/*
269 * Try to flush one buffer from the checkpoint list to disk.
270 *
271 * Return 1 if something happened which requires us to abort the current
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400272 * scan of the checkpoint list. Return <0 if the buffer has failed to
273 * be written out.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700274 *
275 * Called with j_list_lock held and drops it if 1 is returned
276 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
277 */
278static int __process_buffer(journal_t *journal, struct journal_head *jh,
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500279 int *batch_count, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700280{
281 struct buffer_head *bh = jh2bh(jh);
282 int ret = 0;
283
284 if (buffer_locked(bh)) {
285 atomic_inc(&bh->b_count);
286 spin_unlock(&journal->j_list_lock);
287 jbd_unlock_bh_state(bh);
288 wait_on_buffer(bh);
289 /* the journal_head may have gone by now */
290 BUFFER_TRACE(bh, "brelse");
291 __brelse(bh);
292 ret = 1;
293 } else if (jh->b_transaction != NULL) {
294 transaction_t *t = jh->b_transaction;
295 tid_t tid = t->t_tid;
296
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500297 transaction->t_chp_stats.cs_forced_to_close++;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700298 spin_unlock(&journal->j_list_lock);
299 jbd_unlock_bh_state(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700300 jbd2_log_start_commit(journal, tid);
301 jbd2_log_wait_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700302 ret = 1;
303 } else if (!buffer_dirty(bh)) {
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400304 ret = 1;
305 if (unlikely(buffer_write_io_error(bh)))
306 ret = -EIO;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700307 J_ASSERT_JH(jh, !buffer_jbddirty(bh));
308 BUFFER_TRACE(bh, "remove from checkpoint");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700309 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700310 spin_unlock(&journal->j_list_lock);
311 jbd_unlock_bh_state(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700312 jbd2_journal_remove_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700313 __brelse(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700314 } else {
315 /*
316 * Important: we are about to write the buffer, and
317 * possibly block, while still holding the journal lock.
318 * We cannot afford to let the transaction logic start
319 * messing around with this buffer before we write it to
320 * disk, as that would break recoverability.
321 */
322 BUFFER_TRACE(bh, "queue");
323 get_bh(bh);
324 J_ASSERT_BH(bh, !buffer_jwrite(bh));
325 set_buffer_jwrite(bh);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500326 journal->j_chkpt_bhs[*batch_count] = bh;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700327 __buffer_relink_io(jh);
328 jbd_unlock_bh_state(bh);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500329 transaction->t_chp_stats.cs_written++;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700330 (*batch_count)++;
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500331 if (*batch_count == JBD2_NR_BATCH) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700332 spin_unlock(&journal->j_list_lock);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500333 __flush_batch(journal, batch_count);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700334 ret = 1;
335 }
336 }
337 return ret;
338}
339
340/*
341 * Perform an actual checkpoint. We take the first transaction on the
342 * list of transactions to be checkpointed and send all its buffers
343 * to disk. We submit larger chunks of data at once.
344 *
345 * The journal should be locked before calling this function.
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400346 * Called with j_checkpoint_mutex held.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700347 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700348int jbd2_log_do_checkpoint(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700349{
350 transaction_t *transaction;
351 tid_t this_tid;
352 int result;
353
354 jbd_debug(1, "Start checkpoint\n");
355
356 /*
357 * First thing: if there are any transactions in the log which
358 * don't need checkpointing, just eliminate them from the
359 * journal straight away.
360 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700361 result = jbd2_cleanup_journal_tail(journal);
Theodore Ts'o879c5e62009-06-17 11:47:48 -0400362 trace_jbd2_checkpoint(journal, result);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700363 jbd_debug(1, "cleanup_journal_tail returned %d\n", result);
364 if (result <= 0)
365 return result;
366
367 /*
368 * OK, we need to start writing disk blocks. Take one transaction
369 * and write it.
370 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400371 result = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700372 spin_lock(&journal->j_list_lock);
373 if (!journal->j_checkpoint_transactions)
374 goto out;
375 transaction = journal->j_checkpoint_transactions;
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500376 if (transaction->t_chp_stats.cs_chp_time == 0)
377 transaction->t_chp_stats.cs_chp_time = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700378 this_tid = transaction->t_tid;
379restart:
380 /*
381 * If someone cleaned up this transaction while we slept, we're
382 * done (maybe it's a new transaction, but it fell at the same
383 * address).
384 */
385 if (journal->j_checkpoint_transactions == transaction &&
386 transaction->t_tid == this_tid) {
387 int batch_count = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700388 struct journal_head *jh;
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400389 int retry = 0, err;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700390
391 while (!retry && transaction->t_checkpoint_list) {
392 struct buffer_head *bh;
393
394 jh = transaction->t_checkpoint_list;
395 bh = jh2bh(jh);
396 if (!jbd_trylock_bh_state(bh)) {
397 jbd_sync_bh(journal, bh);
398 retry = 1;
399 break;
400 }
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500401 retry = __process_buffer(journal, jh, &batch_count,
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500402 transaction);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400403 if (retry < 0 && !result)
404 result = retry;
Nick Piggin95c354f2008-01-30 13:31:20 +0100405 if (!retry && (need_resched() ||
406 spin_needbreak(&journal->j_list_lock))) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700407 spin_unlock(&journal->j_list_lock);
408 retry = 1;
409 break;
410 }
411 }
412
413 if (batch_count) {
414 if (!retry) {
415 spin_unlock(&journal->j_list_lock);
416 retry = 1;
417 }
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500418 __flush_batch(journal, &batch_count);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700419 }
420
421 if (retry) {
422 spin_lock(&journal->j_list_lock);
423 goto restart;
424 }
425 /*
426 * Now we have cleaned up the first transaction's checkpoint
427 * list. Let's clean up the second one
428 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400429 err = __wait_cp_io(journal, transaction);
430 if (!result)
431 result = err;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700432 }
433out:
434 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700435 if (result < 0)
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400436 jbd2_journal_abort(journal, result);
437 else
438 result = jbd2_cleanup_journal_tail(journal);
439
440 return (result < 0) ? result : 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700441}
442
443/*
444 * Check the list of checkpoint transactions for the journal to see if
445 * we have already got rid of any since the last update of the log tail
446 * in the journal superblock. If so, we can instantly roll the
447 * superblock forward to remove those transactions from the log.
448 *
449 * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
450 *
451 * Called with the journal lock held.
452 *
453 * This is the only part of the journaling code which really needs to be
454 * aware of transaction aborts. Checkpointing involves writing to the
455 * main filesystem area rather than to the journal, so it can proceed
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400456 * even in abort state, but we must not update the super block if
457 * checkpointing may have failed. Otherwise, we would lose some metadata
458 * buffers which should be written-back to the filesystem.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700459 */
460
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700461int jbd2_cleanup_journal_tail(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700462{
463 transaction_t * transaction;
464 tid_t first_tid;
465 unsigned long blocknr, freed;
466
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400467 if (is_journal_aborted(journal))
468 return 1;
469
Dave Kleikamp470decc2006-10-11 01:20:57 -0700470 /* OK, work out the oldest transaction remaining in the log, and
471 * the log block it starts at.
472 *
473 * If the log is now empty, we need to work out which is the
474 * next transaction ID we will write, and where it will
475 * start. */
476
477 spin_lock(&journal->j_state_lock);
478 spin_lock(&journal->j_list_lock);
479 transaction = journal->j_checkpoint_transactions;
480 if (transaction) {
481 first_tid = transaction->t_tid;
482 blocknr = transaction->t_log_start;
483 } else if ((transaction = journal->j_committing_transaction) != NULL) {
484 first_tid = transaction->t_tid;
485 blocknr = transaction->t_log_start;
486 } else if ((transaction = journal->j_running_transaction) != NULL) {
487 first_tid = transaction->t_tid;
488 blocknr = journal->j_head;
489 } else {
490 first_tid = journal->j_transaction_sequence;
491 blocknr = journal->j_head;
492 }
493 spin_unlock(&journal->j_list_lock);
494 J_ASSERT(blocknr != 0);
495
496 /* If the oldest pinned transaction is at the tail of the log
497 already then there's not much we can do right now. */
498 if (journal->j_tail_sequence == first_tid) {
499 spin_unlock(&journal->j_state_lock);
500 return 1;
501 }
502
503 /* OK, update the superblock to recover the freed space.
504 * Physical blocks come first: have we wrapped beyond the end of
505 * the log? */
506 freed = blocknr - journal->j_tail;
507 if (blocknr < journal->j_tail)
508 freed = freed + journal->j_last - journal->j_first;
509
Theodore Ts'o71f2be22009-12-23 07:45:44 -0500510 trace_jbd2_cleanup_journal_tail(journal, first_tid, blocknr, freed);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700511 jbd_debug(1,
512 "Cleaning journal tail from %d to %d (offset %lu), "
513 "freeing %lu\n",
514 journal->j_tail_sequence, first_tid, blocknr, freed);
515
516 journal->j_free += freed;
517 journal->j_tail_sequence = first_tid;
518 journal->j_tail = blocknr;
519 spin_unlock(&journal->j_state_lock);
Theodore Ts'occ3e1be2009-12-23 06:52:08 -0500520
521 /*
522 * If there is an external journal, we need to make sure that
523 * any data blocks that were recently written out --- perhaps
524 * by jbd2_log_do_checkpoint() --- are flushed out before we
525 * drop the transactions from the external journal. It's
526 * unlikely this will be necessary, especially with a
527 * appropriately sized journal, but we need this to guarantee
528 * correctness. Fortunately jbd2_cleanup_journal_tail()
529 * doesn't get called all that often.
530 */
531 if ((journal->j_fs_dev != journal->j_dev) &&
532 (journal->j_flags & JBD2_BARRIER))
533 blkdev_issue_flush(journal->j_fs_dev, NULL);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700534 if (!(journal->j_flags & JBD2_ABORT))
535 jbd2_journal_update_superblock(journal, 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700536 return 0;
537}
538
539
540/* Checkpoint list management */
541
542/*
543 * journal_clean_one_cp_list
544 *
545 * Find all the written-back checkpoint buffers in the given list and release them.
546 *
547 * Called with the journal locked.
548 * Called with j_list_lock held.
549 * Returns number of bufers reaped (for debug)
550 */
551
552static int journal_clean_one_cp_list(struct journal_head *jh, int *released)
553{
554 struct journal_head *last_jh;
555 struct journal_head *next_jh = jh;
556 int ret, freed = 0;
557
558 *released = 0;
559 if (!jh)
560 return 0;
561
562 last_jh = jh->b_cpprev;
563 do {
564 jh = next_jh;
565 next_jh = jh->b_cpnext;
566 /* Use trylock because of the ranking */
567 if (jbd_trylock_bh_state(jh2bh(jh))) {
568 ret = __try_to_free_cp_buf(jh);
569 if (ret) {
570 freed++;
571 if (ret == 2) {
572 *released = 1;
573 return freed;
574 }
575 }
576 }
577 /*
578 * This function only frees up some memory
579 * if possible so we dont have an obligation
580 * to finish processing. Bail out if preemption
581 * requested:
582 */
583 if (need_resched())
584 return freed;
585 } while (jh != last_jh);
586
587 return freed;
588}
589
590/*
591 * journal_clean_checkpoint_list
592 *
593 * Find all the written-back checkpoint buffers in the journal and release them.
594 *
595 * Called with the journal locked.
596 * Called with j_list_lock held.
597 * Returns number of buffers reaped (for debug)
598 */
599
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700600int __jbd2_journal_clean_checkpoint_list(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700601{
602 transaction_t *transaction, *last_transaction, *next_transaction;
603 int ret = 0;
604 int released;
605
606 transaction = journal->j_checkpoint_transactions;
607 if (!transaction)
608 goto out;
609
610 last_transaction = transaction->t_cpprev;
611 next_transaction = transaction;
612 do {
613 transaction = next_transaction;
614 next_transaction = transaction->t_cpnext;
615 ret += journal_clean_one_cp_list(transaction->
616 t_checkpoint_list, &released);
617 /*
618 * This function only frees up some memory if possible so we
619 * dont have an obligation to finish processing. Bail out if
620 * preemption requested:
621 */
622 if (need_resched())
623 goto out;
624 if (released)
625 continue;
626 /*
627 * It is essential that we are as careful as in the case of
628 * t_checkpoint_list with removing the buffer from the list as
629 * we can possibly see not yet submitted buffers on io_list
630 */
631 ret += journal_clean_one_cp_list(transaction->
632 t_checkpoint_io_list, &released);
633 if (need_resched())
634 goto out;
635 } while (transaction != last_transaction);
636out:
637 return ret;
638}
639
640/*
641 * journal_remove_checkpoint: called after a buffer has been committed
642 * to disk (either by being write-back flushed to disk, or being
643 * committed to the log).
644 *
645 * We cannot safely clean a transaction out of the log until all of the
646 * buffer updates committed in that transaction have safely been stored
647 * elsewhere on disk. To achieve this, all of the buffers in a
648 * transaction need to be maintained on the transaction's checkpoint
649 * lists until they have been rewritten, at which point this function is
650 * called to remove the buffer from the existing transaction's
651 * checkpoint lists.
652 *
653 * The function returns 1 if it frees the transaction, 0 otherwise.
654 *
655 * This function is called with the journal locked.
656 * This function is called with j_list_lock held.
657 * This function is called with jbd_lock_bh_state(jh2bh(jh))
658 */
659
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700660int __jbd2_journal_remove_checkpoint(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700661{
Theodore Ts'obf699322009-09-30 00:32:06 -0400662 struct transaction_chp_stats_s *stats;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700663 transaction_t *transaction;
664 journal_t *journal;
665 int ret = 0;
666
667 JBUFFER_TRACE(jh, "entry");
668
669 if ((transaction = jh->b_cp_transaction) == NULL) {
670 JBUFFER_TRACE(jh, "not on transaction");
671 goto out;
672 }
673 journal = transaction->t_journal;
674
675 __buffer_unlink(jh);
676 jh->b_cp_transaction = NULL;
677
678 if (transaction->t_checkpoint_list != NULL ||
679 transaction->t_checkpoint_io_list != NULL)
680 goto out;
681 JBUFFER_TRACE(jh, "transaction has no more buffers");
682
683 /*
684 * There is one special case to worry about: if we have just pulled the
Jan Karaf5a7a6b2008-01-28 23:58:27 -0500685 * buffer off a running or committing transaction's checkpoing list,
686 * then even if the checkpoint list is empty, the transaction obviously
687 * cannot be dropped!
Dave Kleikamp470decc2006-10-11 01:20:57 -0700688 *
Jan Karaf5a7a6b2008-01-28 23:58:27 -0500689 * The locking here around t_state is a bit sleazy.
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700690 * See the comment at the end of jbd2_journal_commit_transaction().
Dave Kleikamp470decc2006-10-11 01:20:57 -0700691 */
Jan Karaf5a7a6b2008-01-28 23:58:27 -0500692 if (transaction->t_state != T_FINISHED) {
693 JBUFFER_TRACE(jh, "belongs to running/committing transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700694 goto out;
695 }
696
697 /* OK, that was the last buffer for the transaction: we can now
698 safely remove this transaction from the log */
Theodore Ts'obf699322009-09-30 00:32:06 -0400699 stats = &transaction->t_chp_stats;
700 if (stats->cs_chp_time)
701 stats->cs_chp_time = jbd2_time_diff(stats->cs_chp_time,
702 jiffies);
703 trace_jbd2_checkpoint_stats(journal->j_fs_dev->bd_dev,
704 transaction->t_tid, stats);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700705
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700706 __jbd2_journal_drop_transaction(journal, transaction);
Aneesh Kumar K.Vfb684072008-11-06 17:50:21 -0500707 kfree(transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700708
709 /* Just in case anybody was waiting for more transactions to be
710 checkpointed... */
711 wake_up(&journal->j_wait_logspace);
712 ret = 1;
713out:
714 JBUFFER_TRACE(jh, "exit");
715 return ret;
716}
717
718/*
719 * journal_insert_checkpoint: put a committed buffer onto a checkpoint
720 * list so that we know when it is safe to clean the transaction out of
721 * the log.
722 *
723 * Called with the journal locked.
724 * Called with j_list_lock held.
725 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700726void __jbd2_journal_insert_checkpoint(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700727 transaction_t *transaction)
728{
729 JBUFFER_TRACE(jh, "entry");
730 J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh)));
731 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
732
733 jh->b_cp_transaction = transaction;
734
735 if (!transaction->t_checkpoint_list) {
736 jh->b_cpnext = jh->b_cpprev = jh;
737 } else {
738 jh->b_cpnext = transaction->t_checkpoint_list;
739 jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev;
740 jh->b_cpprev->b_cpnext = jh;
741 jh->b_cpnext->b_cpprev = jh;
742 }
743 transaction->t_checkpoint_list = jh;
744}
745
746/*
747 * We've finished with this transaction structure: adios...
748 *
749 * The transaction must have no links except for the checkpoint by this
750 * point.
751 *
752 * Called with the journal locked.
753 * Called with j_list_lock held.
754 */
755
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700756void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700757{
758 assert_spin_locked(&journal->j_list_lock);
759 if (transaction->t_cpnext) {
760 transaction->t_cpnext->t_cpprev = transaction->t_cpprev;
761 transaction->t_cpprev->t_cpnext = transaction->t_cpnext;
762 if (journal->j_checkpoint_transactions == transaction)
763 journal->j_checkpoint_transactions =
764 transaction->t_cpnext;
765 if (journal->j_checkpoint_transactions == transaction)
766 journal->j_checkpoint_transactions = NULL;
767 }
768
769 J_ASSERT(transaction->t_state == T_FINISHED);
770 J_ASSERT(transaction->t_buffers == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700771 J_ASSERT(transaction->t_forget == NULL);
772 J_ASSERT(transaction->t_iobuf_list == NULL);
773 J_ASSERT(transaction->t_shadow_list == NULL);
774 J_ASSERT(transaction->t_log_list == NULL);
775 J_ASSERT(transaction->t_checkpoint_list == NULL);
776 J_ASSERT(transaction->t_checkpoint_io_list == NULL);
777 J_ASSERT(transaction->t_updates == 0);
778 J_ASSERT(journal->j_committing_transaction != transaction);
779 J_ASSERT(journal->j_running_transaction != transaction);
780
781 jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700782}