2 * linux/fs/jbd2/checkpoint.c
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1999
6 * Copyright 1999 Red Hat Software --- All Rights Reserved
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.
12 * Checkpoint routines for the generic filesystem journaling code.
13 * Part of the ext2fs journaling system.
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
20 #include <linux/time.h>
22 #include <linux/jbd2.h>
23 #include <linux/errno.h>
24 #include <linux/slab.h>
27 * Unlink a buffer from a transaction checkpoint list.
29 * Called with j_list_lock held.
31 static inline void __buffer_unlink_first(struct journal_head *jh)
33 transaction_t *transaction = jh->b_cp_transaction;
35 jh->b_cpnext->b_cpprev = jh->b_cpprev;
36 jh->b_cpprev->b_cpnext = jh->b_cpnext;
37 if (transaction->t_checkpoint_list == jh) {
38 transaction->t_checkpoint_list = jh->b_cpnext;
39 if (transaction->t_checkpoint_list == jh)
40 transaction->t_checkpoint_list = NULL;
45 * Unlink a buffer from a transaction checkpoint(io) list.
47 * Called with j_list_lock held.
49 static inline void __buffer_unlink(struct journal_head *jh)
51 transaction_t *transaction = jh->b_cp_transaction;
53 __buffer_unlink_first(jh);
54 if (transaction->t_checkpoint_io_list == jh) {
55 transaction->t_checkpoint_io_list = jh->b_cpnext;
56 if (transaction->t_checkpoint_io_list == jh)
57 transaction->t_checkpoint_io_list = NULL;
62 * Move a buffer from the checkpoint list to the checkpoint io list
64 * Called with j_list_lock held
66 static inline void __buffer_relink_io(struct journal_head *jh)
68 transaction_t *transaction = jh->b_cp_transaction;
70 __buffer_unlink_first(jh);
72 if (!transaction->t_checkpoint_io_list) {
73 jh->b_cpnext = jh->b_cpprev = jh;
75 jh->b_cpnext = transaction->t_checkpoint_io_list;
76 jh->b_cpprev = transaction->t_checkpoint_io_list->b_cpprev;
77 jh->b_cpprev->b_cpnext = jh;
78 jh->b_cpnext->b_cpprev = jh;
80 transaction->t_checkpoint_io_list = jh;
84 * Try to release a checkpointed buffer from its transaction.
85 * Returns 1 if we released it and 2 if we also released the
88 * Requires j_list_lock
89 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
91 static int __try_to_free_cp_buf(struct journal_head *jh)
94 struct buffer_head *bh = jh2bh(jh);
96 if (jh->b_jlist == BJ_None && !buffer_locked(bh) && !buffer_dirty(bh)) {
97 JBUFFER_TRACE(jh, "remove from checkpoint list");
98 ret = __jbd2_journal_remove_checkpoint(jh) + 1;
99 jbd_unlock_bh_state(bh);
100 jbd2_journal_remove_journal_head(bh);
101 BUFFER_TRACE(bh, "release");
104 jbd_unlock_bh_state(bh);
110 * __jbd2_log_wait_for_space: wait until there is space in the journal.
112 * Called under j-state_lock *only*. It will be unlocked if we have to wait
113 * for a checkpoint to free up some space in the log.
115 void __jbd2_log_wait_for_space(journal_t *journal)
118 assert_spin_locked(&journal->j_state_lock);
120 nblocks = jbd_space_needed(journal);
121 while (__jbd2_log_space_left(journal) < nblocks) {
122 if (journal->j_flags & JBD2_ABORT)
124 spin_unlock(&journal->j_state_lock);
125 mutex_lock(&journal->j_checkpoint_mutex);
128 * Test again, another process may have checkpointed while we
129 * were waiting for the checkpoint lock
131 spin_lock(&journal->j_state_lock);
132 nblocks = jbd_space_needed(journal);
133 if (__jbd2_log_space_left(journal) < nblocks) {
134 spin_unlock(&journal->j_state_lock);
135 jbd2_log_do_checkpoint(journal);
136 spin_lock(&journal->j_state_lock);
138 mutex_unlock(&journal->j_checkpoint_mutex);
143 * We were unable to perform jbd_trylock_bh_state() inside j_list_lock.
144 * The caller must restart a list walk. Wait for someone else to run
145 * jbd_unlock_bh_state().
147 static void jbd_sync_bh(journal_t *journal, struct buffer_head *bh)
148 __releases(journal->j_list_lock)
151 spin_unlock(&journal->j_list_lock);
152 jbd_lock_bh_state(bh);
153 jbd_unlock_bh_state(bh);
158 * Clean up transaction's list of buffers submitted for io.
159 * We wait for any pending IO to complete and remove any clean
160 * buffers. Note that we take the buffers in the opposite ordering
161 * from the one in which they were submitted for IO.
163 * Called with j_list_lock held.
165 static void __wait_cp_io(journal_t *journal, transaction_t *transaction)
167 struct journal_head *jh;
168 struct buffer_head *bh;
172 this_tid = transaction->t_tid;
174 /* Did somebody clean up the transaction in the meanwhile? */
175 if (journal->j_checkpoint_transactions != transaction ||
176 transaction->t_tid != this_tid)
178 while (!released && transaction->t_checkpoint_io_list) {
179 jh = transaction->t_checkpoint_io_list;
181 if (!jbd_trylock_bh_state(bh)) {
182 jbd_sync_bh(journal, bh);
183 spin_lock(&journal->j_list_lock);
186 if (buffer_locked(bh)) {
187 atomic_inc(&bh->b_count);
188 spin_unlock(&journal->j_list_lock);
189 jbd_unlock_bh_state(bh);
191 /* the journal_head may have gone by now */
192 BUFFER_TRACE(bh, "brelse");
194 spin_lock(&journal->j_list_lock);
198 * Now in whatever state the buffer currently is, we know that
199 * it has been written out and so we can drop it from the list
201 released = __jbd2_journal_remove_checkpoint(jh);
202 jbd_unlock_bh_state(bh);
203 jbd2_journal_remove_journal_head(bh);
211 __flush_batch(journal_t *journal, struct buffer_head **bhs, int *batch_count)
215 ll_rw_block(SWRITE, *batch_count, bhs);
216 for (i = 0; i < *batch_count; i++) {
217 struct buffer_head *bh = bhs[i];
218 clear_buffer_jwrite(bh);
219 BUFFER_TRACE(bh, "brelse");
226 * Try to flush one buffer from the checkpoint list to disk.
228 * Return 1 if something happened which requires us to abort the current
229 * scan of the checkpoint list.
231 * Called with j_list_lock held and drops it if 1 is returned
232 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
234 static int __process_buffer(journal_t *journal, struct journal_head *jh,
235 struct buffer_head **bhs, int *batch_count,
236 transaction_t *transaction)
238 struct buffer_head *bh = jh2bh(jh);
241 if (buffer_locked(bh)) {
242 atomic_inc(&bh->b_count);
243 spin_unlock(&journal->j_list_lock);
244 jbd_unlock_bh_state(bh);
246 /* the journal_head may have gone by now */
247 BUFFER_TRACE(bh, "brelse");
250 } else if (jh->b_transaction != NULL) {
251 transaction_t *t = jh->b_transaction;
252 tid_t tid = t->t_tid;
254 transaction->t_chp_stats.cs_forced_to_close++;
255 spin_unlock(&journal->j_list_lock);
256 jbd_unlock_bh_state(bh);
257 jbd2_log_start_commit(journal, tid);
258 jbd2_log_wait_commit(journal, tid);
260 } else if (!buffer_dirty(bh)) {
261 J_ASSERT_JH(jh, !buffer_jbddirty(bh));
262 BUFFER_TRACE(bh, "remove from checkpoint");
263 __jbd2_journal_remove_checkpoint(jh);
264 spin_unlock(&journal->j_list_lock);
265 jbd_unlock_bh_state(bh);
266 jbd2_journal_remove_journal_head(bh);
271 * Important: we are about to write the buffer, and
272 * possibly block, while still holding the journal lock.
273 * We cannot afford to let the transaction logic start
274 * messing around with this buffer before we write it to
275 * disk, as that would break recoverability.
277 BUFFER_TRACE(bh, "queue");
279 J_ASSERT_BH(bh, !buffer_jwrite(bh));
280 set_buffer_jwrite(bh);
281 bhs[*batch_count] = bh;
282 __buffer_relink_io(jh);
283 jbd_unlock_bh_state(bh);
284 transaction->t_chp_stats.cs_written++;
286 if (*batch_count == NR_BATCH) {
287 spin_unlock(&journal->j_list_lock);
288 __flush_batch(journal, bhs, batch_count);
296 * Perform an actual checkpoint. We take the first transaction on the
297 * list of transactions to be checkpointed and send all its buffers
298 * to disk. We submit larger chunks of data at once.
300 * The journal should be locked before calling this function.
302 int jbd2_log_do_checkpoint(journal_t *journal)
304 transaction_t *transaction;
308 jbd_debug(1, "Start checkpoint\n");
311 * First thing: if there are any transactions in the log which
312 * don't need checkpointing, just eliminate them from the
313 * journal straight away.
315 result = jbd2_cleanup_journal_tail(journal);
316 jbd_debug(1, "cleanup_journal_tail returned %d\n", result);
321 * OK, we need to start writing disk blocks. Take one transaction
324 spin_lock(&journal->j_list_lock);
325 if (!journal->j_checkpoint_transactions)
327 transaction = journal->j_checkpoint_transactions;
328 if (transaction->t_chp_stats.cs_chp_time == 0)
329 transaction->t_chp_stats.cs_chp_time = jiffies;
330 this_tid = transaction->t_tid;
333 * If someone cleaned up this transaction while we slept, we're
334 * done (maybe it's a new transaction, but it fell at the same
337 if (journal->j_checkpoint_transactions == transaction &&
338 transaction->t_tid == this_tid) {
340 struct buffer_head *bhs[NR_BATCH];
341 struct journal_head *jh;
344 while (!retry && transaction->t_checkpoint_list) {
345 struct buffer_head *bh;
347 jh = transaction->t_checkpoint_list;
349 if (!jbd_trylock_bh_state(bh)) {
350 jbd_sync_bh(journal, bh);
354 retry = __process_buffer(journal, jh, bhs, &batch_count,
356 if (!retry && (need_resched() ||
357 spin_needbreak(&journal->j_list_lock))) {
358 spin_unlock(&journal->j_list_lock);
366 spin_unlock(&journal->j_list_lock);
369 __flush_batch(journal, bhs, &batch_count);
373 spin_lock(&journal->j_list_lock);
377 * Now we have cleaned up the first transaction's checkpoint
378 * list. Let's clean up the second one
380 __wait_cp_io(journal, transaction);
383 spin_unlock(&journal->j_list_lock);
384 result = jbd2_cleanup_journal_tail(journal);
391 * Check the list of checkpoint transactions for the journal to see if
392 * we have already got rid of any since the last update of the log tail
393 * in the journal superblock. If so, we can instantly roll the
394 * superblock forward to remove those transactions from the log.
396 * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
398 * Called with the journal lock held.
400 * This is the only part of the journaling code which really needs to be
401 * aware of transaction aborts. Checkpointing involves writing to the
402 * main filesystem area rather than to the journal, so it can proceed
403 * even in abort state, but we must not update the journal superblock if
404 * we have an abort error outstanding.
407 int jbd2_cleanup_journal_tail(journal_t *journal)
409 transaction_t * transaction;
411 unsigned long blocknr, freed;
413 /* OK, work out the oldest transaction remaining in the log, and
414 * the log block it starts at.
416 * If the log is now empty, we need to work out which is the
417 * next transaction ID we will write, and where it will
420 spin_lock(&journal->j_state_lock);
421 spin_lock(&journal->j_list_lock);
422 transaction = journal->j_checkpoint_transactions;
424 first_tid = transaction->t_tid;
425 blocknr = transaction->t_log_start;
426 } else if ((transaction = journal->j_committing_transaction) != NULL) {
427 first_tid = transaction->t_tid;
428 blocknr = transaction->t_log_start;
429 } else if ((transaction = journal->j_running_transaction) != NULL) {
430 first_tid = transaction->t_tid;
431 blocknr = journal->j_head;
433 first_tid = journal->j_transaction_sequence;
434 blocknr = journal->j_head;
436 spin_unlock(&journal->j_list_lock);
437 J_ASSERT(blocknr != 0);
439 /* If the oldest pinned transaction is at the tail of the log
440 already then there's not much we can do right now. */
441 if (journal->j_tail_sequence == first_tid) {
442 spin_unlock(&journal->j_state_lock);
446 /* OK, update the superblock to recover the freed space.
447 * Physical blocks come first: have we wrapped beyond the end of
449 freed = blocknr - journal->j_tail;
450 if (blocknr < journal->j_tail)
451 freed = freed + journal->j_last - journal->j_first;
454 "Cleaning journal tail from %d to %d (offset %lu), "
456 journal->j_tail_sequence, first_tid, blocknr, freed);
458 journal->j_free += freed;
459 journal->j_tail_sequence = first_tid;
460 journal->j_tail = blocknr;
461 spin_unlock(&journal->j_state_lock);
462 if (!(journal->j_flags & JBD2_ABORT))
463 jbd2_journal_update_superblock(journal, 1);
468 /* Checkpoint list management */
471 * journal_clean_one_cp_list
473 * Find all the written-back checkpoint buffers in the given list and release them.
475 * Called with the journal locked.
476 * Called with j_list_lock held.
477 * Returns number of bufers reaped (for debug)
480 static int journal_clean_one_cp_list(struct journal_head *jh, int *released)
482 struct journal_head *last_jh;
483 struct journal_head *next_jh = jh;
490 last_jh = jh->b_cpprev;
493 next_jh = jh->b_cpnext;
494 /* Use trylock because of the ranking */
495 if (jbd_trylock_bh_state(jh2bh(jh))) {
496 ret = __try_to_free_cp_buf(jh);
506 * This function only frees up some memory
507 * if possible so we dont have an obligation
508 * to finish processing. Bail out if preemption
513 } while (jh != last_jh);
519 * journal_clean_checkpoint_list
521 * Find all the written-back checkpoint buffers in the journal and release them.
523 * Called with the journal locked.
524 * Called with j_list_lock held.
525 * Returns number of buffers reaped (for debug)
528 int __jbd2_journal_clean_checkpoint_list(journal_t *journal)
530 transaction_t *transaction, *last_transaction, *next_transaction;
534 transaction = journal->j_checkpoint_transactions;
538 last_transaction = transaction->t_cpprev;
539 next_transaction = transaction;
541 transaction = next_transaction;
542 next_transaction = transaction->t_cpnext;
543 ret += journal_clean_one_cp_list(transaction->
544 t_checkpoint_list, &released);
546 * This function only frees up some memory if possible so we
547 * dont have an obligation to finish processing. Bail out if
548 * preemption requested:
555 * It is essential that we are as careful as in the case of
556 * t_checkpoint_list with removing the buffer from the list as
557 * we can possibly see not yet submitted buffers on io_list
559 ret += journal_clean_one_cp_list(transaction->
560 t_checkpoint_io_list, &released);
563 } while (transaction != last_transaction);
569 * journal_remove_checkpoint: called after a buffer has been committed
570 * to disk (either by being write-back flushed to disk, or being
571 * committed to the log).
573 * We cannot safely clean a transaction out of the log until all of the
574 * buffer updates committed in that transaction have safely been stored
575 * elsewhere on disk. To achieve this, all of the buffers in a
576 * transaction need to be maintained on the transaction's checkpoint
577 * lists until they have been rewritten, at which point this function is
578 * called to remove the buffer from the existing transaction's
581 * The function returns 1 if it frees the transaction, 0 otherwise.
583 * This function is called with the journal locked.
584 * This function is called with j_list_lock held.
585 * This function is called with jbd_lock_bh_state(jh2bh(jh))
588 int __jbd2_journal_remove_checkpoint(struct journal_head *jh)
590 transaction_t *transaction;
594 JBUFFER_TRACE(jh, "entry");
596 if ((transaction = jh->b_cp_transaction) == NULL) {
597 JBUFFER_TRACE(jh, "not on transaction");
600 journal = transaction->t_journal;
603 jh->b_cp_transaction = NULL;
605 if (transaction->t_checkpoint_list != NULL ||
606 transaction->t_checkpoint_io_list != NULL)
608 JBUFFER_TRACE(jh, "transaction has no more buffers");
611 * There is one special case to worry about: if we have just pulled the
612 * buffer off a running or committing transaction's checkpoing list,
613 * then even if the checkpoint list is empty, the transaction obviously
616 * The locking here around t_state is a bit sleazy.
617 * See the comment at the end of jbd2_journal_commit_transaction().
619 if (transaction->t_state != T_FINISHED) {
620 JBUFFER_TRACE(jh, "belongs to running/committing transaction");
624 /* OK, that was the last buffer for the transaction: we can now
625 safely remove this transaction from the log */
627 __jbd2_journal_drop_transaction(journal, transaction);
629 /* Just in case anybody was waiting for more transactions to be
631 wake_up(&journal->j_wait_logspace);
634 JBUFFER_TRACE(jh, "exit");
639 * journal_insert_checkpoint: put a committed buffer onto a checkpoint
640 * list so that we know when it is safe to clean the transaction out of
643 * Called with the journal locked.
644 * Called with j_list_lock held.
646 void __jbd2_journal_insert_checkpoint(struct journal_head *jh,
647 transaction_t *transaction)
649 JBUFFER_TRACE(jh, "entry");
650 J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh)));
651 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
653 jh->b_cp_transaction = transaction;
655 if (!transaction->t_checkpoint_list) {
656 jh->b_cpnext = jh->b_cpprev = jh;
658 jh->b_cpnext = transaction->t_checkpoint_list;
659 jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev;
660 jh->b_cpprev->b_cpnext = jh;
661 jh->b_cpnext->b_cpprev = jh;
663 transaction->t_checkpoint_list = jh;
667 * We've finished with this transaction structure: adios...
669 * The transaction must have no links except for the checkpoint by this
672 * Called with the journal locked.
673 * Called with j_list_lock held.
676 void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction)
678 assert_spin_locked(&journal->j_list_lock);
679 if (transaction->t_cpnext) {
680 transaction->t_cpnext->t_cpprev = transaction->t_cpprev;
681 transaction->t_cpprev->t_cpnext = transaction->t_cpnext;
682 if (journal->j_checkpoint_transactions == transaction)
683 journal->j_checkpoint_transactions =
684 transaction->t_cpnext;
685 if (journal->j_checkpoint_transactions == transaction)
686 journal->j_checkpoint_transactions = NULL;
689 J_ASSERT(transaction->t_state == T_FINISHED);
690 J_ASSERT(transaction->t_buffers == NULL);
691 J_ASSERT(transaction->t_sync_datalist == NULL);
692 J_ASSERT(transaction->t_forget == NULL);
693 J_ASSERT(transaction->t_iobuf_list == NULL);
694 J_ASSERT(transaction->t_shadow_list == NULL);
695 J_ASSERT(transaction->t_log_list == NULL);
696 J_ASSERT(transaction->t_checkpoint_list == NULL);
697 J_ASSERT(transaction->t_checkpoint_io_list == NULL);
698 J_ASSERT(transaction->t_updates == 0);
699 J_ASSERT(journal->j_committing_transaction != transaction);
700 J_ASSERT(journal->j_running_transaction != transaction);
702 jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid);