blob: 17159cacbd9ec95aba6f7b0a942680322858159d [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>
Theodore Ts'oede86cc2008-10-05 20:50:06 -040023#include <linux/marker.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070024#include <linux/errno.h>
25#include <linux/slab.h>
26
27/*
28 * Unlink a buffer from a transaction checkpoint list.
29 *
30 * Called with j_list_lock held.
31 */
32static inline void __buffer_unlink_first(struct journal_head *jh)
33{
34 transaction_t *transaction = jh->b_cp_transaction;
35
36 jh->b_cpnext->b_cpprev = jh->b_cpprev;
37 jh->b_cpprev->b_cpnext = jh->b_cpnext;
38 if (transaction->t_checkpoint_list == jh) {
39 transaction->t_checkpoint_list = jh->b_cpnext;
40 if (transaction->t_checkpoint_list == jh)
41 transaction->t_checkpoint_list = NULL;
42 }
43}
44
45/*
46 * Unlink a buffer from a transaction checkpoint(io) list.
47 *
48 * Called with j_list_lock held.
49 */
50static inline void __buffer_unlink(struct journal_head *jh)
51{
52 transaction_t *transaction = jh->b_cp_transaction;
53
54 __buffer_unlink_first(jh);
55 if (transaction->t_checkpoint_io_list == jh) {
56 transaction->t_checkpoint_io_list = jh->b_cpnext;
57 if (transaction->t_checkpoint_io_list == jh)
58 transaction->t_checkpoint_io_list = NULL;
59 }
60}
61
62/*
63 * Move a buffer from the checkpoint list to the checkpoint io list
64 *
65 * Called with j_list_lock held
66 */
67static inline void __buffer_relink_io(struct journal_head *jh)
68{
69 transaction_t *transaction = jh->b_cp_transaction;
70
71 __buffer_unlink_first(jh);
72
73 if (!transaction->t_checkpoint_io_list) {
74 jh->b_cpnext = jh->b_cpprev = jh;
75 } else {
76 jh->b_cpnext = transaction->t_checkpoint_io_list;
77 jh->b_cpprev = transaction->t_checkpoint_io_list->b_cpprev;
78 jh->b_cpprev->b_cpnext = jh;
79 jh->b_cpnext->b_cpprev = jh;
80 }
81 transaction->t_checkpoint_io_list = jh;
82}
83
84/*
85 * Try to release a checkpointed buffer from its transaction.
86 * Returns 1 if we released it and 2 if we also released the
87 * whole transaction.
88 *
89 * Requires j_list_lock
90 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
91 */
92static int __try_to_free_cp_buf(struct journal_head *jh)
93{
94 int ret = 0;
95 struct buffer_head *bh = jh2bh(jh);
96
Hidehiro Kawai44519fa2008-10-10 20:29:13 -040097 if (jh->b_jlist == BJ_None && !buffer_locked(bh) &&
98 !buffer_dirty(bh) && !buffer_write_io_error(bh)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -070099 JBUFFER_TRACE(jh, "remove from checkpoint list");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700100 ret = __jbd2_journal_remove_checkpoint(jh) + 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700101 jbd_unlock_bh_state(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700102 jbd2_journal_remove_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700103 BUFFER_TRACE(bh, "release");
104 __brelse(bh);
105 } else {
106 jbd_unlock_bh_state(bh);
107 }
108 return ret;
109}
110
111/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700112 * __jbd2_log_wait_for_space: wait until there is space in the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700113 *
114 * Called under j-state_lock *only*. It will be unlocked if we have to wait
115 * for a checkpoint to free up some space in the log.
116 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700117void __jbd2_log_wait_for_space(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700118{
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500119 int nblocks, space_left;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700120 assert_spin_locked(&journal->j_state_lock);
121
122 nblocks = jbd_space_needed(journal);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700123 while (__jbd2_log_space_left(journal) < nblocks) {
124 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700125 return;
126 spin_unlock(&journal->j_state_lock);
127 mutex_lock(&journal->j_checkpoint_mutex);
128
129 /*
130 * Test again, another process may have checkpointed while we
Duane Griffin23f8b792008-10-08 23:28:31 -0400131 * were waiting for the checkpoint lock. If there are no
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500132 * transactions ready to be checkpointed, try to recover
133 * journal space by calling cleanup_journal_tail(), and if
134 * that doesn't work, by waiting for the currently committing
135 * transaction to complete. If there is absolutely no way
136 * to make progress, this is either a BUG or corrupted
137 * filesystem, so abort the journal and leave a stack
138 * trace for forensic evidence.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700139 */
140 spin_lock(&journal->j_state_lock);
Duane Griffin23f8b792008-10-08 23:28:31 -0400141 spin_lock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700142 nblocks = jbd_space_needed(journal);
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500143 space_left = __jbd2_log_space_left(journal);
144 if (space_left < nblocks) {
Duane Griffin23f8b792008-10-08 23:28:31 -0400145 int chkpt = journal->j_checkpoint_transactions != NULL;
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500146 tid_t tid = 0;
Duane Griffin23f8b792008-10-08 23:28:31 -0400147
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500148 if (journal->j_committing_transaction)
149 tid = journal->j_committing_transaction->t_tid;
Duane Griffin23f8b792008-10-08 23:28:31 -0400150 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700151 spin_unlock(&journal->j_state_lock);
Duane Griffin23f8b792008-10-08 23:28:31 -0400152 if (chkpt) {
153 jbd2_log_do_checkpoint(journal);
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500154 } else if (jbd2_cleanup_journal_tail(journal) == 0) {
155 /* We were able to recover space; yay! */
156 ;
157 } else if (tid) {
158 jbd2_log_wait_commit(journal, tid);
Duane Griffin23f8b792008-10-08 23:28:31 -0400159 } else {
Theodore Ts'o8c3f25d2008-11-06 22:38:07 -0500160 printk(KERN_ERR "%s: needed %d blocks and "
161 "only had %d space available\n",
162 __func__, nblocks, space_left);
163 printk(KERN_ERR "%s: no way to get more "
164 "journal space in %s\n", __func__,
165 journal->j_devname);
166 WARN_ON(1);
Duane Griffin23f8b792008-10-08 23:28:31 -0400167 jbd2_journal_abort(journal, 0);
168 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700169 spin_lock(&journal->j_state_lock);
Duane Griffin23f8b792008-10-08 23:28:31 -0400170 } else {
171 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700172 }
173 mutex_unlock(&journal->j_checkpoint_mutex);
174 }
175}
176
177/*
178 * We were unable to perform jbd_trylock_bh_state() inside j_list_lock.
179 * The caller must restart a list walk. Wait for someone else to run
180 * jbd_unlock_bh_state().
181 */
182static void jbd_sync_bh(journal_t *journal, struct buffer_head *bh)
183 __releases(journal->j_list_lock)
184{
185 get_bh(bh);
186 spin_unlock(&journal->j_list_lock);
187 jbd_lock_bh_state(bh);
188 jbd_unlock_bh_state(bh);
189 put_bh(bh);
190}
191
192/*
193 * Clean up transaction's list of buffers submitted for io.
194 * We wait for any pending IO to complete and remove any clean
195 * buffers. Note that we take the buffers in the opposite ordering
196 * from the one in which they were submitted for IO.
197 *
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400198 * Return 0 on success, and return <0 if some buffers have failed
199 * to be written out.
200 *
Dave Kleikamp470decc2006-10-11 01:20:57 -0700201 * Called with j_list_lock held.
202 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400203static int __wait_cp_io(journal_t *journal, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700204{
205 struct journal_head *jh;
206 struct buffer_head *bh;
207 tid_t this_tid;
208 int released = 0;
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400209 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700210
211 this_tid = transaction->t_tid;
212restart:
213 /* Did somebody clean up the transaction in the meanwhile? */
214 if (journal->j_checkpoint_transactions != transaction ||
215 transaction->t_tid != this_tid)
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400216 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700217 while (!released && transaction->t_checkpoint_io_list) {
218 jh = transaction->t_checkpoint_io_list;
219 bh = jh2bh(jh);
220 if (!jbd_trylock_bh_state(bh)) {
221 jbd_sync_bh(journal, bh);
222 spin_lock(&journal->j_list_lock);
223 goto restart;
224 }
225 if (buffer_locked(bh)) {
226 atomic_inc(&bh->b_count);
227 spin_unlock(&journal->j_list_lock);
228 jbd_unlock_bh_state(bh);
229 wait_on_buffer(bh);
230 /* the journal_head may have gone by now */
231 BUFFER_TRACE(bh, "brelse");
232 __brelse(bh);
233 spin_lock(&journal->j_list_lock);
234 goto restart;
235 }
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400236 if (unlikely(buffer_write_io_error(bh)))
237 ret = -EIO;
238
Dave Kleikamp470decc2006-10-11 01:20:57 -0700239 /*
240 * Now in whatever state the buffer currently is, we know that
241 * it has been written out and so we can drop it from the list
242 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700243 released = __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700244 jbd_unlock_bh_state(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700245 jbd2_journal_remove_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700246 __brelse(bh);
247 }
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400248
249 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700250}
251
Dave Kleikamp470decc2006-10-11 01:20:57 -0700252static void
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500253__flush_batch(journal_t *journal, int *batch_count)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700254{
255 int i;
256
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500257 ll_rw_block(SWRITE, *batch_count, journal->j_chkpt_bhs);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700258 for (i = 0; i < *batch_count; i++) {
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500259 struct buffer_head *bh = journal->j_chkpt_bhs[i];
Dave Kleikamp470decc2006-10-11 01:20:57 -0700260 clear_buffer_jwrite(bh);
261 BUFFER_TRACE(bh, "brelse");
262 __brelse(bh);
263 }
264 *batch_count = 0;
265}
266
267/*
268 * Try to flush one buffer from the checkpoint list to disk.
269 *
270 * Return 1 if something happened which requires us to abort the current
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400271 * scan of the checkpoint list. Return <0 if the buffer has failed to
272 * be written out.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700273 *
274 * Called with j_list_lock held and drops it if 1 is returned
275 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
276 */
277static int __process_buffer(journal_t *journal, struct journal_head *jh,
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500278 int *batch_count, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700279{
280 struct buffer_head *bh = jh2bh(jh);
281 int ret = 0;
282
283 if (buffer_locked(bh)) {
284 atomic_inc(&bh->b_count);
285 spin_unlock(&journal->j_list_lock);
286 jbd_unlock_bh_state(bh);
287 wait_on_buffer(bh);
288 /* the journal_head may have gone by now */
289 BUFFER_TRACE(bh, "brelse");
290 __brelse(bh);
291 ret = 1;
292 } else if (jh->b_transaction != NULL) {
293 transaction_t *t = jh->b_transaction;
294 tid_t tid = t->t_tid;
295
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500296 transaction->t_chp_stats.cs_forced_to_close++;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700297 spin_unlock(&journal->j_list_lock);
298 jbd_unlock_bh_state(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700299 jbd2_log_start_commit(journal, tid);
300 jbd2_log_wait_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700301 ret = 1;
302 } else if (!buffer_dirty(bh)) {
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400303 ret = 1;
304 if (unlikely(buffer_write_io_error(bh)))
305 ret = -EIO;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700306 J_ASSERT_JH(jh, !buffer_jbddirty(bh));
307 BUFFER_TRACE(bh, "remove from checkpoint");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700308 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700309 spin_unlock(&journal->j_list_lock);
310 jbd_unlock_bh_state(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700311 jbd2_journal_remove_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700312 __brelse(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700313 } else {
314 /*
315 * Important: we are about to write the buffer, and
316 * possibly block, while still holding the journal lock.
317 * We cannot afford to let the transaction logic start
318 * messing around with this buffer before we write it to
319 * disk, as that would break recoverability.
320 */
321 BUFFER_TRACE(bh, "queue");
322 get_bh(bh);
323 J_ASSERT_BH(bh, !buffer_jwrite(bh));
324 set_buffer_jwrite(bh);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500325 journal->j_chkpt_bhs[*batch_count] = bh;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700326 __buffer_relink_io(jh);
327 jbd_unlock_bh_state(bh);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500328 transaction->t_chp_stats.cs_written++;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700329 (*batch_count)++;
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500330 if (*batch_count == JBD2_NR_BATCH) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700331 spin_unlock(&journal->j_list_lock);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -0500332 __flush_batch(journal, batch_count);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700333 ret = 1;
334 }
335 }
336 return ret;
337}
338
339/*
340 * Perform an actual checkpoint. We take the first transaction on the
341 * list of transactions to be checkpointed and send all its buffers
342 * to disk. We submit larger chunks of data at once.
343 *
344 * The journal should be locked before calling this function.
Hidehiro Kawai44519fa2008-10-10 20:29:13 -0400345 * Called with j_checkpoint_mutex held.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700346 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700347int jbd2_log_do_checkpoint(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700348{
349 transaction_t *transaction;
350 tid_t this_tid;
351 int result;
352
353 jbd_debug(1, "Start checkpoint\n");
354
355 /*
356 * First thing: if there are any transactions in the log which
357 * don't need checkpointing, just eliminate them from the
358 * journal straight away.
359 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700360 result = jbd2_cleanup_journal_tail(journal);
Theodore Ts'oede86cc2008-10-05 20:50:06 -0400361 trace_mark(jbd2_checkpoint, "dev %s need_checkpoint %d",
362 journal->j_devname, 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
510 jbd_debug(1,
511 "Cleaning journal tail from %d to %d (offset %lu), "
512 "freeing %lu\n",
513 journal->j_tail_sequence, first_tid, blocknr, freed);
514
515 journal->j_free += freed;
516 journal->j_tail_sequence = first_tid;
517 journal->j_tail = blocknr;
518 spin_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700519 if (!(journal->j_flags & JBD2_ABORT))
520 jbd2_journal_update_superblock(journal, 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700521 return 0;
522}
523
524
525/* Checkpoint list management */
526
527/*
528 * journal_clean_one_cp_list
529 *
530 * Find all the written-back checkpoint buffers in the given list and release them.
531 *
532 * Called with the journal locked.
533 * Called with j_list_lock held.
534 * Returns number of bufers reaped (for debug)
535 */
536
537static int journal_clean_one_cp_list(struct journal_head *jh, int *released)
538{
539 struct journal_head *last_jh;
540 struct journal_head *next_jh = jh;
541 int ret, freed = 0;
542
543 *released = 0;
544 if (!jh)
545 return 0;
546
547 last_jh = jh->b_cpprev;
548 do {
549 jh = next_jh;
550 next_jh = jh->b_cpnext;
551 /* Use trylock because of the ranking */
552 if (jbd_trylock_bh_state(jh2bh(jh))) {
553 ret = __try_to_free_cp_buf(jh);
554 if (ret) {
555 freed++;
556 if (ret == 2) {
557 *released = 1;
558 return freed;
559 }
560 }
561 }
562 /*
563 * This function only frees up some memory
564 * if possible so we dont have an obligation
565 * to finish processing. Bail out if preemption
566 * requested:
567 */
568 if (need_resched())
569 return freed;
570 } while (jh != last_jh);
571
572 return freed;
573}
574
575/*
576 * journal_clean_checkpoint_list
577 *
578 * Find all the written-back checkpoint buffers in the journal and release them.
579 *
580 * Called with the journal locked.
581 * Called with j_list_lock held.
582 * Returns number of buffers reaped (for debug)
583 */
584
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700585int __jbd2_journal_clean_checkpoint_list(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700586{
587 transaction_t *transaction, *last_transaction, *next_transaction;
588 int ret = 0;
589 int released;
590
591 transaction = journal->j_checkpoint_transactions;
592 if (!transaction)
593 goto out;
594
595 last_transaction = transaction->t_cpprev;
596 next_transaction = transaction;
597 do {
598 transaction = next_transaction;
599 next_transaction = transaction->t_cpnext;
600 ret += journal_clean_one_cp_list(transaction->
601 t_checkpoint_list, &released);
602 /*
603 * This function only frees up some memory if possible so we
604 * dont have an obligation to finish processing. Bail out if
605 * preemption requested:
606 */
607 if (need_resched())
608 goto out;
609 if (released)
610 continue;
611 /*
612 * It is essential that we are as careful as in the case of
613 * t_checkpoint_list with removing the buffer from the list as
614 * we can possibly see not yet submitted buffers on io_list
615 */
616 ret += journal_clean_one_cp_list(transaction->
617 t_checkpoint_io_list, &released);
618 if (need_resched())
619 goto out;
620 } while (transaction != last_transaction);
621out:
622 return ret;
623}
624
625/*
626 * journal_remove_checkpoint: called after a buffer has been committed
627 * to disk (either by being write-back flushed to disk, or being
628 * committed to the log).
629 *
630 * We cannot safely clean a transaction out of the log until all of the
631 * buffer updates committed in that transaction have safely been stored
632 * elsewhere on disk. To achieve this, all of the buffers in a
633 * transaction need to be maintained on the transaction's checkpoint
634 * lists until they have been rewritten, at which point this function is
635 * called to remove the buffer from the existing transaction's
636 * checkpoint lists.
637 *
638 * The function returns 1 if it frees the transaction, 0 otherwise.
639 *
640 * This function is called with the journal locked.
641 * This function is called with j_list_lock held.
642 * This function is called with jbd_lock_bh_state(jh2bh(jh))
643 */
644
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700645int __jbd2_journal_remove_checkpoint(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700646{
647 transaction_t *transaction;
648 journal_t *journal;
649 int ret = 0;
650
651 JBUFFER_TRACE(jh, "entry");
652
653 if ((transaction = jh->b_cp_transaction) == NULL) {
654 JBUFFER_TRACE(jh, "not on transaction");
655 goto out;
656 }
657 journal = transaction->t_journal;
658
659 __buffer_unlink(jh);
660 jh->b_cp_transaction = NULL;
661
662 if (transaction->t_checkpoint_list != NULL ||
663 transaction->t_checkpoint_io_list != NULL)
664 goto out;
665 JBUFFER_TRACE(jh, "transaction has no more buffers");
666
667 /*
668 * There is one special case to worry about: if we have just pulled the
Jan Karaf5a7a6b2008-01-28 23:58:27 -0500669 * buffer off a running or committing transaction's checkpoing list,
670 * then even if the checkpoint list is empty, the transaction obviously
671 * cannot be dropped!
Dave Kleikamp470decc2006-10-11 01:20:57 -0700672 *
Jan Karaf5a7a6b2008-01-28 23:58:27 -0500673 * The locking here around t_state is a bit sleazy.
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700674 * See the comment at the end of jbd2_journal_commit_transaction().
Dave Kleikamp470decc2006-10-11 01:20:57 -0700675 */
Jan Karaf5a7a6b2008-01-28 23:58:27 -0500676 if (transaction->t_state != T_FINISHED) {
677 JBUFFER_TRACE(jh, "belongs to running/committing transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700678 goto out;
679 }
680
681 /* OK, that was the last buffer for the transaction: we can now
682 safely remove this transaction from the log */
683
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700684 __jbd2_journal_drop_transaction(journal, transaction);
Aneesh Kumar K.Vfb684072008-11-06 17:50:21 -0500685 kfree(transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700686
687 /* Just in case anybody was waiting for more transactions to be
688 checkpointed... */
689 wake_up(&journal->j_wait_logspace);
690 ret = 1;
691out:
692 JBUFFER_TRACE(jh, "exit");
693 return ret;
694}
695
696/*
697 * journal_insert_checkpoint: put a committed buffer onto a checkpoint
698 * list so that we know when it is safe to clean the transaction out of
699 * the log.
700 *
701 * Called with the journal locked.
702 * Called with j_list_lock held.
703 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700704void __jbd2_journal_insert_checkpoint(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700705 transaction_t *transaction)
706{
707 JBUFFER_TRACE(jh, "entry");
708 J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh)));
709 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
710
711 jh->b_cp_transaction = transaction;
712
713 if (!transaction->t_checkpoint_list) {
714 jh->b_cpnext = jh->b_cpprev = jh;
715 } else {
716 jh->b_cpnext = transaction->t_checkpoint_list;
717 jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev;
718 jh->b_cpprev->b_cpnext = jh;
719 jh->b_cpnext->b_cpprev = jh;
720 }
721 transaction->t_checkpoint_list = jh;
722}
723
724/*
725 * We've finished with this transaction structure: adios...
726 *
727 * The transaction must have no links except for the checkpoint by this
728 * point.
729 *
730 * Called with the journal locked.
731 * Called with j_list_lock held.
732 */
733
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700734void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700735{
736 assert_spin_locked(&journal->j_list_lock);
737 if (transaction->t_cpnext) {
738 transaction->t_cpnext->t_cpprev = transaction->t_cpprev;
739 transaction->t_cpprev->t_cpnext = transaction->t_cpnext;
740 if (journal->j_checkpoint_transactions == transaction)
741 journal->j_checkpoint_transactions =
742 transaction->t_cpnext;
743 if (journal->j_checkpoint_transactions == transaction)
744 journal->j_checkpoint_transactions = NULL;
745 }
746
747 J_ASSERT(transaction->t_state == T_FINISHED);
748 J_ASSERT(transaction->t_buffers == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700749 J_ASSERT(transaction->t_forget == NULL);
750 J_ASSERT(transaction->t_iobuf_list == NULL);
751 J_ASSERT(transaction->t_shadow_list == NULL);
752 J_ASSERT(transaction->t_log_list == NULL);
753 J_ASSERT(transaction->t_checkpoint_list == NULL);
754 J_ASSERT(transaction->t_checkpoint_io_list == NULL);
755 J_ASSERT(transaction->t_updates == 0);
756 J_ASSERT(journal->j_committing_transaction != transaction);
757 J_ASSERT(journal->j_running_transaction != transaction);
758
759 jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700760}