2 * linux/fs/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/jbd.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 = __journal_remove_checkpoint(jh) + 1;
99 jbd_unlock_bh_state(bh);
100 journal_remove_journal_head(bh);
101 BUFFER_TRACE(bh, "release");
104 jbd_unlock_bh_state(bh);
110 * __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 __log_wait_for_space(journal_t *journal)
118 assert_spin_locked(&journal->j_state_lock);
120 nblocks = jbd_space_needed(journal);
121 while (__log_space_left(journal) < nblocks) {
122 if (journal->j_flags & JFS_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 (__log_space_left(journal) < nblocks) {
134 spin_unlock(&journal->j_state_lock);
135 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)
150 spin_unlock(&journal->j_list_lock);
151 jbd_lock_bh_state(bh);
152 jbd_unlock_bh_state(bh);
157 * Clean up transaction's list of buffers submitted for io.
158 * We wait for any pending IO to complete and remove any clean
159 * buffers. Note that we take the buffers in the opposite ordering
160 * from the one in which they were submitted for IO.
162 * Called with j_list_lock held.
164 static void __wait_cp_io(journal_t *journal, transaction_t *transaction)
166 struct journal_head *jh;
167 struct buffer_head *bh;
171 this_tid = transaction->t_tid;
173 /* Did somebody clean up the transaction in the meanwhile? */
174 if (journal->j_checkpoint_transactions != transaction ||
175 transaction->t_tid != this_tid)
177 while (!released && transaction->t_checkpoint_io_list) {
178 jh = transaction->t_checkpoint_io_list;
180 if (!jbd_trylock_bh_state(bh)) {
181 jbd_sync_bh(journal, bh);
182 spin_lock(&journal->j_list_lock);
185 if (buffer_locked(bh)) {
186 atomic_inc(&bh->b_count);
187 spin_unlock(&journal->j_list_lock);
188 jbd_unlock_bh_state(bh);
190 /* the journal_head may have gone by now */
191 BUFFER_TRACE(bh, "brelse");
193 spin_lock(&journal->j_list_lock);
197 * Now in whatever state the buffer currently is, we know that
198 * it has been written out and so we can drop it from the list
200 released = __journal_remove_checkpoint(jh);
201 jbd_unlock_bh_state(bh);
202 journal_remove_journal_head(bh);
210 __flush_batch(journal_t *journal, struct buffer_head **bhs, int *batch_count)
214 ll_rw_block(SWRITE, *batch_count, bhs);
215 for (i = 0; i < *batch_count; i++) {
216 struct buffer_head *bh = bhs[i];
217 clear_buffer_jwrite(bh);
218 BUFFER_TRACE(bh, "brelse");
225 * Try to flush one buffer from the checkpoint list to disk.
227 * Return 1 if something happened which requires us to abort the current
228 * scan of the checkpoint list.
230 * Called with j_list_lock held and drops it if 1 is returned
231 * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
233 static int __process_buffer(journal_t *journal, struct journal_head *jh,
234 struct buffer_head **bhs, int *batch_count)
236 struct buffer_head *bh = jh2bh(jh);
239 if (buffer_locked(bh)) {
240 atomic_inc(&bh->b_count);
241 spin_unlock(&journal->j_list_lock);
242 jbd_unlock_bh_state(bh);
244 /* the journal_head may have gone by now */
245 BUFFER_TRACE(bh, "brelse");
248 } else if (jh->b_transaction != NULL) {
249 transaction_t *t = jh->b_transaction;
250 tid_t tid = t->t_tid;
252 spin_unlock(&journal->j_list_lock);
253 jbd_unlock_bh_state(bh);
254 log_start_commit(journal, tid);
255 log_wait_commit(journal, tid);
257 } else if (!buffer_dirty(bh)) {
258 J_ASSERT_JH(jh, !buffer_jbddirty(bh));
259 BUFFER_TRACE(bh, "remove from checkpoint");
260 __journal_remove_checkpoint(jh);
261 spin_unlock(&journal->j_list_lock);
262 jbd_unlock_bh_state(bh);
263 journal_remove_journal_head(bh);
268 * Important: we are about to write the buffer, and
269 * possibly block, while still holding the journal lock.
270 * We cannot afford to let the transaction logic start
271 * messing around with this buffer before we write it to
272 * disk, as that would break recoverability.
274 BUFFER_TRACE(bh, "queue");
276 J_ASSERT_BH(bh, !buffer_jwrite(bh));
277 set_buffer_jwrite(bh);
278 bhs[*batch_count] = bh;
279 __buffer_relink_io(jh);
280 jbd_unlock_bh_state(bh);
282 if (*batch_count == NR_BATCH) {
283 spin_unlock(&journal->j_list_lock);
284 __flush_batch(journal, bhs, batch_count);
292 * Perform an actual checkpoint. We take the first transaction on the
293 * list of transactions to be checkpointed and send all its buffers
294 * to disk. We submit larger chunks of data at once.
296 * The journal should be locked before calling this function.
298 int log_do_checkpoint(journal_t *journal)
300 transaction_t *transaction;
304 jbd_debug(1, "Start checkpoint\n");
307 * First thing: if there are any transactions in the log which
308 * don't need checkpointing, just eliminate them from the
309 * journal straight away.
311 result = cleanup_journal_tail(journal);
312 jbd_debug(1, "cleanup_journal_tail returned %d\n", result);
317 * OK, we need to start writing disk blocks. Take one transaction
320 spin_lock(&journal->j_list_lock);
321 if (!journal->j_checkpoint_transactions)
323 transaction = journal->j_checkpoint_transactions;
324 this_tid = transaction->t_tid;
327 * If someone cleaned up this transaction while we slept, we're
328 * done (maybe it's a new transaction, but it fell at the same
331 if (journal->j_checkpoint_transactions == transaction &&
332 transaction->t_tid == this_tid) {
334 struct buffer_head *bhs[NR_BATCH];
335 struct journal_head *jh;
338 while (!retry && transaction->t_checkpoint_list) {
339 struct buffer_head *bh;
341 jh = transaction->t_checkpoint_list;
343 if (!jbd_trylock_bh_state(bh)) {
344 jbd_sync_bh(journal, bh);
348 retry = __process_buffer(journal, jh, bhs,&batch_count);
349 if (!retry && lock_need_resched(&journal->j_list_lock)){
350 spin_unlock(&journal->j_list_lock);
358 spin_unlock(&journal->j_list_lock);
361 __flush_batch(journal, bhs, &batch_count);
365 spin_lock(&journal->j_list_lock);
369 * Now we have cleaned up the first transaction's checkpoint
370 * list. Let's clean up the second one
372 __wait_cp_io(journal, transaction);
375 spin_unlock(&journal->j_list_lock);
376 result = cleanup_journal_tail(journal);
383 * Check the list of checkpoint transactions for the journal to see if
384 * we have already got rid of any since the last update of the log tail
385 * in the journal superblock. If so, we can instantly roll the
386 * superblock forward to remove those transactions from the log.
388 * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
390 * Called with the journal lock held.
392 * This is the only part of the journaling code which really needs to be
393 * aware of transaction aborts. Checkpointing involves writing to the
394 * main filesystem area rather than to the journal, so it can proceed
395 * even in abort state, but we must not update the journal superblock if
396 * we have an abort error outstanding.
399 int cleanup_journal_tail(journal_t *journal)
401 transaction_t * transaction;
403 unsigned long blocknr, freed;
405 /* OK, work out the oldest transaction remaining in the log, and
406 * the log block it starts at.
408 * If the log is now empty, we need to work out which is the
409 * next transaction ID we will write, and where it will
412 spin_lock(&journal->j_state_lock);
413 spin_lock(&journal->j_list_lock);
414 transaction = journal->j_checkpoint_transactions;
416 first_tid = transaction->t_tid;
417 blocknr = transaction->t_log_start;
418 } else if ((transaction = journal->j_committing_transaction) != NULL) {
419 first_tid = transaction->t_tid;
420 blocknr = transaction->t_log_start;
421 } else if ((transaction = journal->j_running_transaction) != NULL) {
422 first_tid = transaction->t_tid;
423 blocknr = journal->j_head;
425 first_tid = journal->j_transaction_sequence;
426 blocknr = journal->j_head;
428 spin_unlock(&journal->j_list_lock);
429 J_ASSERT(blocknr != 0);
431 /* If the oldest pinned transaction is at the tail of the log
432 already then there's not much we can do right now. */
433 if (journal->j_tail_sequence == first_tid) {
434 spin_unlock(&journal->j_state_lock);
438 /* OK, update the superblock to recover the freed space.
439 * Physical blocks come first: have we wrapped beyond the end of
441 freed = blocknr - journal->j_tail;
442 if (blocknr < journal->j_tail)
443 freed = freed + journal->j_last - journal->j_first;
446 "Cleaning journal tail from %d to %d (offset %lu), "
448 journal->j_tail_sequence, first_tid, blocknr, freed);
450 journal->j_free += freed;
451 journal->j_tail_sequence = first_tid;
452 journal->j_tail = blocknr;
453 spin_unlock(&journal->j_state_lock);
454 if (!(journal->j_flags & JFS_ABORT))
455 journal_update_superblock(journal, 1);
460 /* Checkpoint list management */
463 * journal_clean_one_cp_list
465 * Find all the written-back checkpoint buffers in the given list and release them.
467 * Called with the journal locked.
468 * Called with j_list_lock held.
469 * Returns number of bufers reaped (for debug)
472 static int journal_clean_one_cp_list(struct journal_head *jh, int *released)
474 struct journal_head *last_jh;
475 struct journal_head *next_jh = jh;
482 last_jh = jh->b_cpprev;
485 next_jh = jh->b_cpnext;
486 /* Use trylock because of the ranking */
487 if (jbd_trylock_bh_state(jh2bh(jh))) {
488 ret = __try_to_free_cp_buf(jh);
498 * This function only frees up some memory
499 * if possible so we dont have an obligation
500 * to finish processing. Bail out if preemption
505 } while (jh != last_jh);
511 * journal_clean_checkpoint_list
513 * Find all the written-back checkpoint buffers in the journal and release them.
515 * Called with the journal locked.
516 * Called with j_list_lock held.
517 * Returns number of buffers reaped (for debug)
520 int __journal_clean_checkpoint_list(journal_t *journal)
522 transaction_t *transaction, *last_transaction, *next_transaction;
526 transaction = journal->j_checkpoint_transactions;
530 last_transaction = transaction->t_cpprev;
531 next_transaction = transaction;
533 transaction = next_transaction;
534 next_transaction = transaction->t_cpnext;
535 ret += journal_clean_one_cp_list(transaction->
536 t_checkpoint_list, &released);
538 * This function only frees up some memory if possible so we
539 * dont have an obligation to finish processing. Bail out if
540 * preemption requested:
547 * It is essential that we are as careful as in the case of
548 * t_checkpoint_list with removing the buffer from the list as
549 * we can possibly see not yet submitted buffers on io_list
551 ret += journal_clean_one_cp_list(transaction->
552 t_checkpoint_io_list, &released);
555 } while (transaction != last_transaction);
561 * journal_remove_checkpoint: called after a buffer has been committed
562 * to disk (either by being write-back flushed to disk, or being
563 * committed to the log).
565 * We cannot safely clean a transaction out of the log until all of the
566 * buffer updates committed in that transaction have safely been stored
567 * elsewhere on disk. To achieve this, all of the buffers in a
568 * transaction need to be maintained on the transaction's checkpoint
569 * lists until they have been rewritten, at which point this function is
570 * called to remove the buffer from the existing transaction's
573 * The function returns 1 if it frees the transaction, 0 otherwise.
575 * This function is called with the journal locked.
576 * This function is called with j_list_lock held.
577 * This function is called with jbd_lock_bh_state(jh2bh(jh))
580 int __journal_remove_checkpoint(struct journal_head *jh)
582 transaction_t *transaction;
586 JBUFFER_TRACE(jh, "entry");
588 if ((transaction = jh->b_cp_transaction) == NULL) {
589 JBUFFER_TRACE(jh, "not on transaction");
592 journal = transaction->t_journal;
595 jh->b_cp_transaction = NULL;
597 if (transaction->t_checkpoint_list != NULL ||
598 transaction->t_checkpoint_io_list != NULL)
600 JBUFFER_TRACE(jh, "transaction has no more buffers");
603 * There is one special case to worry about: if we have just pulled the
604 * buffer off a committing transaction's forget list, then even if the
605 * checkpoint list is empty, the transaction obviously cannot be
608 * The locking here around j_committing_transaction is a bit sleazy.
609 * See the comment at the end of journal_commit_transaction().
611 if (transaction == journal->j_committing_transaction) {
612 JBUFFER_TRACE(jh, "belongs to committing transaction");
616 /* OK, that was the last buffer for the transaction: we can now
617 safely remove this transaction from the log */
619 __journal_drop_transaction(journal, transaction);
621 /* Just in case anybody was waiting for more transactions to be
623 wake_up(&journal->j_wait_logspace);
626 JBUFFER_TRACE(jh, "exit");
631 * journal_insert_checkpoint: put a committed buffer onto a checkpoint
632 * list so that we know when it is safe to clean the transaction out of
635 * Called with the journal locked.
636 * Called with j_list_lock held.
638 void __journal_insert_checkpoint(struct journal_head *jh,
639 transaction_t *transaction)
641 JBUFFER_TRACE(jh, "entry");
642 J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh)));
643 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
645 jh->b_cp_transaction = transaction;
647 if (!transaction->t_checkpoint_list) {
648 jh->b_cpnext = jh->b_cpprev = jh;
650 jh->b_cpnext = transaction->t_checkpoint_list;
651 jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev;
652 jh->b_cpprev->b_cpnext = jh;
653 jh->b_cpnext->b_cpprev = jh;
655 transaction->t_checkpoint_list = jh;
659 * We've finished with this transaction structure: adios...
661 * The transaction must have no links except for the checkpoint by this
664 * Called with the journal locked.
665 * Called with j_list_lock held.
668 void __journal_drop_transaction(journal_t *journal, transaction_t *transaction)
670 assert_spin_locked(&journal->j_list_lock);
671 if (transaction->t_cpnext) {
672 transaction->t_cpnext->t_cpprev = transaction->t_cpprev;
673 transaction->t_cpprev->t_cpnext = transaction->t_cpnext;
674 if (journal->j_checkpoint_transactions == transaction)
675 journal->j_checkpoint_transactions =
676 transaction->t_cpnext;
677 if (journal->j_checkpoint_transactions == transaction)
678 journal->j_checkpoint_transactions = NULL;
681 J_ASSERT(transaction->t_state == T_FINISHED);
682 J_ASSERT(transaction->t_buffers == NULL);
683 J_ASSERT(transaction->t_sync_datalist == NULL);
684 J_ASSERT(transaction->t_forget == NULL);
685 J_ASSERT(transaction->t_iobuf_list == NULL);
686 J_ASSERT(transaction->t_shadow_list == NULL);
687 J_ASSERT(transaction->t_log_list == NULL);
688 J_ASSERT(transaction->t_checkpoint_list == NULL);
689 J_ASSERT(transaction->t_checkpoint_io_list == NULL);
690 J_ASSERT(transaction->t_updates == 0);
691 J_ASSERT(journal->j_committing_transaction != transaction);
692 J_ASSERT(journal->j_running_transaction != transaction);
694 jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid);