xref: /openbmc/linux/fs/jbd2/checkpoint.c (revision 1ab142d4)
1 /*
2  * linux/fs/jbd2/checkpoint.c
3  *
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>
22 #include <linux/jbd2.h>
23 #include <linux/errno.h>
24 #include <linux/slab.h>
25 #include <linux/blkdev.h>
26 #include <trace/events/jbd2.h>
27 
28 /*
29  * Unlink a buffer from a transaction checkpoint list.
30  *
31  * Called with j_list_lock held.
32  */
33 static 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  */
51 static 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  */
68 static 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  */
93 static int __try_to_free_cp_buf(struct journal_head *jh)
94 {
95 	int ret = 0;
96 	struct buffer_head *bh = jh2bh(jh);
97 
98 	if (jh->b_jlist == BJ_None && !buffer_locked(bh) &&
99 	    !buffer_dirty(bh) && !buffer_write_io_error(bh)) {
100 		/*
101 		 * Get our reference so that bh cannot be freed before
102 		 * we unlock it
103 		 */
104 		get_bh(bh);
105 		JBUFFER_TRACE(jh, "remove from checkpoint list");
106 		ret = __jbd2_journal_remove_checkpoint(jh) + 1;
107 		jbd_unlock_bh_state(bh);
108 		BUFFER_TRACE(bh, "release");
109 		__brelse(bh);
110 	} else {
111 		jbd_unlock_bh_state(bh);
112 	}
113 	return ret;
114 }
115 
116 /*
117  * __jbd2_log_wait_for_space: wait until there is space in the journal.
118  *
119  * Called under j-state_lock *only*.  It will be unlocked if we have to wait
120  * for a checkpoint to free up some space in the log.
121  */
122 void __jbd2_log_wait_for_space(journal_t *journal)
123 {
124 	int nblocks, space_left;
125 	/* assert_spin_locked(&journal->j_state_lock); */
126 
127 	nblocks = jbd_space_needed(journal);
128 	while (__jbd2_log_space_left(journal) < nblocks) {
129 		if (journal->j_flags & JBD2_ABORT)
130 			return;
131 		write_unlock(&journal->j_state_lock);
132 		mutex_lock(&journal->j_checkpoint_mutex);
133 
134 		/*
135 		 * Test again, another process may have checkpointed while we
136 		 * were waiting for the checkpoint lock. If there are no
137 		 * transactions ready to be checkpointed, try to recover
138 		 * journal space by calling cleanup_journal_tail(), and if
139 		 * that doesn't work, by waiting for the currently committing
140 		 * transaction to complete.  If there is absolutely no way
141 		 * to make progress, this is either a BUG or corrupted
142 		 * filesystem, so abort the journal and leave a stack
143 		 * trace for forensic evidence.
144 		 */
145 		write_lock(&journal->j_state_lock);
146 		spin_lock(&journal->j_list_lock);
147 		nblocks = jbd_space_needed(journal);
148 		space_left = __jbd2_log_space_left(journal);
149 		if (space_left < nblocks) {
150 			int chkpt = journal->j_checkpoint_transactions != NULL;
151 			tid_t tid = 0;
152 
153 			if (journal->j_committing_transaction)
154 				tid = journal->j_committing_transaction->t_tid;
155 			spin_unlock(&journal->j_list_lock);
156 			write_unlock(&journal->j_state_lock);
157 			if (chkpt) {
158 				jbd2_log_do_checkpoint(journal);
159 			} else if (jbd2_cleanup_journal_tail(journal) == 0) {
160 				/* We were able to recover space; yay! */
161 				;
162 			} else if (tid) {
163 				jbd2_log_wait_commit(journal, tid);
164 			} else {
165 				printk(KERN_ERR "%s: needed %d blocks and "
166 				       "only had %d space available\n",
167 				       __func__, nblocks, space_left);
168 				printk(KERN_ERR "%s: no way to get more "
169 				       "journal space in %s\n", __func__,
170 				       journal->j_devname);
171 				WARN_ON(1);
172 				jbd2_journal_abort(journal, 0);
173 			}
174 			write_lock(&journal->j_state_lock);
175 		} else {
176 			spin_unlock(&journal->j_list_lock);
177 		}
178 		mutex_unlock(&journal->j_checkpoint_mutex);
179 	}
180 }
181 
182 /*
183  * We were unable to perform jbd_trylock_bh_state() inside j_list_lock.
184  * The caller must restart a list walk.  Wait for someone else to run
185  * jbd_unlock_bh_state().
186  */
187 static void jbd_sync_bh(journal_t *journal, struct buffer_head *bh)
188 	__releases(journal->j_list_lock)
189 {
190 	get_bh(bh);
191 	spin_unlock(&journal->j_list_lock);
192 	jbd_lock_bh_state(bh);
193 	jbd_unlock_bh_state(bh);
194 	put_bh(bh);
195 }
196 
197 /*
198  * Clean up transaction's list of buffers submitted for io.
199  * We wait for any pending IO to complete and remove any clean
200  * buffers. Note that we take the buffers in the opposite ordering
201  * from the one in which they were submitted for IO.
202  *
203  * Return 0 on success, and return <0 if some buffers have failed
204  * to be written out.
205  *
206  * Called with j_list_lock held.
207  */
208 static int __wait_cp_io(journal_t *journal, transaction_t *transaction)
209 {
210 	struct journal_head *jh;
211 	struct buffer_head *bh;
212 	tid_t this_tid;
213 	int released = 0;
214 	int ret = 0;
215 
216 	this_tid = transaction->t_tid;
217 restart:
218 	/* Did somebody clean up the transaction in the meanwhile? */
219 	if (journal->j_checkpoint_transactions != transaction ||
220 			transaction->t_tid != this_tid)
221 		return ret;
222 	while (!released && transaction->t_checkpoint_io_list) {
223 		jh = transaction->t_checkpoint_io_list;
224 		bh = jh2bh(jh);
225 		if (!jbd_trylock_bh_state(bh)) {
226 			jbd_sync_bh(journal, bh);
227 			spin_lock(&journal->j_list_lock);
228 			goto restart;
229 		}
230 		get_bh(bh);
231 		if (buffer_locked(bh)) {
232 			spin_unlock(&journal->j_list_lock);
233 			jbd_unlock_bh_state(bh);
234 			wait_on_buffer(bh);
235 			/* the journal_head may have gone by now */
236 			BUFFER_TRACE(bh, "brelse");
237 			__brelse(bh);
238 			spin_lock(&journal->j_list_lock);
239 			goto restart;
240 		}
241 		if (unlikely(buffer_write_io_error(bh)))
242 			ret = -EIO;
243 
244 		/*
245 		 * Now in whatever state the buffer currently is, we know that
246 		 * it has been written out and so we can drop it from the list
247 		 */
248 		released = __jbd2_journal_remove_checkpoint(jh);
249 		jbd_unlock_bh_state(bh);
250 		__brelse(bh);
251 	}
252 
253 	return ret;
254 }
255 
256 static void
257 __flush_batch(journal_t *journal, int *batch_count)
258 {
259 	int i;
260 	struct blk_plug plug;
261 
262 	blk_start_plug(&plug);
263 	for (i = 0; i < *batch_count; i++)
264 		write_dirty_buffer(journal->j_chkpt_bhs[i], WRITE_SYNC);
265 	blk_finish_plug(&plug);
266 
267 	for (i = 0; i < *batch_count; i++) {
268 		struct buffer_head *bh = journal->j_chkpt_bhs[i];
269 		clear_buffer_jwrite(bh);
270 		BUFFER_TRACE(bh, "brelse");
271 		__brelse(bh);
272 	}
273 	*batch_count = 0;
274 }
275 
276 /*
277  * Try to flush one buffer from the checkpoint list to disk.
278  *
279  * Return 1 if something happened which requires us to abort the current
280  * scan of the checkpoint list.  Return <0 if the buffer has failed to
281  * be written out.
282  *
283  * Called with j_list_lock held and drops it if 1 is returned
284  * Called under jbd_lock_bh_state(jh2bh(jh)), and drops it
285  */
286 static int __process_buffer(journal_t *journal, struct journal_head *jh,
287 			    int *batch_count, transaction_t *transaction)
288 {
289 	struct buffer_head *bh = jh2bh(jh);
290 	int ret = 0;
291 
292 	if (buffer_locked(bh)) {
293 		get_bh(bh);
294 		spin_unlock(&journal->j_list_lock);
295 		jbd_unlock_bh_state(bh);
296 		wait_on_buffer(bh);
297 		/* the journal_head may have gone by now */
298 		BUFFER_TRACE(bh, "brelse");
299 		__brelse(bh);
300 		ret = 1;
301 	} else if (jh->b_transaction != NULL) {
302 		transaction_t *t = jh->b_transaction;
303 		tid_t tid = t->t_tid;
304 
305 		transaction->t_chp_stats.cs_forced_to_close++;
306 		spin_unlock(&journal->j_list_lock);
307 		jbd_unlock_bh_state(bh);
308 		if (unlikely(journal->j_flags & JBD2_UNMOUNT))
309 			/*
310 			 * The journal thread is dead; so starting and
311 			 * waiting for a commit to finish will cause
312 			 * us to wait for a _very_ long time.
313 			 */
314 			printk(KERN_ERR "JBD2: %s: "
315 			       "Waiting for Godot: block %llu\n",
316 			       journal->j_devname,
317 			       (unsigned long long) bh->b_blocknr);
318 		jbd2_log_start_commit(journal, tid);
319 		jbd2_log_wait_commit(journal, tid);
320 		ret = 1;
321 	} else if (!buffer_dirty(bh)) {
322 		ret = 1;
323 		if (unlikely(buffer_write_io_error(bh)))
324 			ret = -EIO;
325 		get_bh(bh);
326 		J_ASSERT_JH(jh, !buffer_jbddirty(bh));
327 		BUFFER_TRACE(bh, "remove from checkpoint");
328 		__jbd2_journal_remove_checkpoint(jh);
329 		spin_unlock(&journal->j_list_lock);
330 		jbd_unlock_bh_state(bh);
331 		__brelse(bh);
332 	} else {
333 		/*
334 		 * Important: we are about to write the buffer, and
335 		 * possibly block, while still holding the journal lock.
336 		 * We cannot afford to let the transaction logic start
337 		 * messing around with this buffer before we write it to
338 		 * disk, as that would break recoverability.
339 		 */
340 		BUFFER_TRACE(bh, "queue");
341 		get_bh(bh);
342 		J_ASSERT_BH(bh, !buffer_jwrite(bh));
343 		set_buffer_jwrite(bh);
344 		journal->j_chkpt_bhs[*batch_count] = bh;
345 		__buffer_relink_io(jh);
346 		jbd_unlock_bh_state(bh);
347 		transaction->t_chp_stats.cs_written++;
348 		(*batch_count)++;
349 		if (*batch_count == JBD2_NR_BATCH) {
350 			spin_unlock(&journal->j_list_lock);
351 			__flush_batch(journal, batch_count);
352 			ret = 1;
353 		}
354 	}
355 	return ret;
356 }
357 
358 /*
359  * Perform an actual checkpoint. We take the first transaction on the
360  * list of transactions to be checkpointed and send all its buffers
361  * to disk. We submit larger chunks of data at once.
362  *
363  * The journal should be locked before calling this function.
364  * Called with j_checkpoint_mutex held.
365  */
366 int jbd2_log_do_checkpoint(journal_t *journal)
367 {
368 	transaction_t *transaction;
369 	tid_t this_tid;
370 	int result;
371 
372 	jbd_debug(1, "Start checkpoint\n");
373 
374 	/*
375 	 * First thing: if there are any transactions in the log which
376 	 * don't need checkpointing, just eliminate them from the
377 	 * journal straight away.
378 	 */
379 	result = jbd2_cleanup_journal_tail(journal);
380 	trace_jbd2_checkpoint(journal, result);
381 	jbd_debug(1, "cleanup_journal_tail returned %d\n", result);
382 	if (result <= 0)
383 		return result;
384 
385 	/*
386 	 * OK, we need to start writing disk blocks.  Take one transaction
387 	 * and write it.
388 	 */
389 	result = 0;
390 	spin_lock(&journal->j_list_lock);
391 	if (!journal->j_checkpoint_transactions)
392 		goto out;
393 	transaction = journal->j_checkpoint_transactions;
394 	if (transaction->t_chp_stats.cs_chp_time == 0)
395 		transaction->t_chp_stats.cs_chp_time = jiffies;
396 	this_tid = transaction->t_tid;
397 restart:
398 	/*
399 	 * If someone cleaned up this transaction while we slept, we're
400 	 * done (maybe it's a new transaction, but it fell at the same
401 	 * address).
402 	 */
403 	if (journal->j_checkpoint_transactions == transaction &&
404 			transaction->t_tid == this_tid) {
405 		int batch_count = 0;
406 		struct journal_head *jh;
407 		int retry = 0, err;
408 
409 		while (!retry && transaction->t_checkpoint_list) {
410 			struct buffer_head *bh;
411 
412 			jh = transaction->t_checkpoint_list;
413 			bh = jh2bh(jh);
414 			if (!jbd_trylock_bh_state(bh)) {
415 				jbd_sync_bh(journal, bh);
416 				retry = 1;
417 				break;
418 			}
419 			retry = __process_buffer(journal, jh, &batch_count,
420 						 transaction);
421 			if (retry < 0 && !result)
422 				result = retry;
423 			if (!retry && (need_resched() ||
424 				spin_needbreak(&journal->j_list_lock))) {
425 				spin_unlock(&journal->j_list_lock);
426 				retry = 1;
427 				break;
428 			}
429 		}
430 
431 		if (batch_count) {
432 			if (!retry) {
433 				spin_unlock(&journal->j_list_lock);
434 				retry = 1;
435 			}
436 			__flush_batch(journal, &batch_count);
437 		}
438 
439 		if (retry) {
440 			spin_lock(&journal->j_list_lock);
441 			goto restart;
442 		}
443 		/*
444 		 * Now we have cleaned up the first transaction's checkpoint
445 		 * list. Let's clean up the second one
446 		 */
447 		err = __wait_cp_io(journal, transaction);
448 		if (!result)
449 			result = err;
450 	}
451 out:
452 	spin_unlock(&journal->j_list_lock);
453 	if (result < 0)
454 		jbd2_journal_abort(journal, result);
455 	else
456 		result = jbd2_cleanup_journal_tail(journal);
457 
458 	return (result < 0) ? result : 0;
459 }
460 
461 /*
462  * Check the list of checkpoint transactions for the journal to see if
463  * we have already got rid of any since the last update of the log tail
464  * in the journal superblock.  If so, we can instantly roll the
465  * superblock forward to remove those transactions from the log.
466  *
467  * Return <0 on error, 0 on success, 1 if there was nothing to clean up.
468  *
469  * Called with the journal lock held.
470  *
471  * This is the only part of the journaling code which really needs to be
472  * aware of transaction aborts.  Checkpointing involves writing to the
473  * main filesystem area rather than to the journal, so it can proceed
474  * even in abort state, but we must not update the super block if
475  * checkpointing may have failed.  Otherwise, we would lose some metadata
476  * buffers which should be written-back to the filesystem.
477  */
478 
479 int jbd2_cleanup_journal_tail(journal_t *journal)
480 {
481 	transaction_t * transaction;
482 	tid_t		first_tid;
483 	unsigned long	blocknr, freed;
484 
485 	if (is_journal_aborted(journal))
486 		return 1;
487 
488 	/* OK, work out the oldest transaction remaining in the log, and
489 	 * the log block it starts at.
490 	 *
491 	 * If the log is now empty, we need to work out which is the
492 	 * next transaction ID we will write, and where it will
493 	 * start. */
494 
495 	write_lock(&journal->j_state_lock);
496 	spin_lock(&journal->j_list_lock);
497 	transaction = journal->j_checkpoint_transactions;
498 	if (transaction) {
499 		first_tid = transaction->t_tid;
500 		blocknr = transaction->t_log_start;
501 	} else if ((transaction = journal->j_committing_transaction) != NULL) {
502 		first_tid = transaction->t_tid;
503 		blocknr = transaction->t_log_start;
504 	} else if ((transaction = journal->j_running_transaction) != NULL) {
505 		first_tid = transaction->t_tid;
506 		blocknr = journal->j_head;
507 	} else {
508 		first_tid = journal->j_transaction_sequence;
509 		blocknr = journal->j_head;
510 	}
511 	spin_unlock(&journal->j_list_lock);
512 	J_ASSERT(blocknr != 0);
513 
514 	/* If the oldest pinned transaction is at the tail of the log
515            already then there's not much we can do right now. */
516 	if (journal->j_tail_sequence == first_tid) {
517 		write_unlock(&journal->j_state_lock);
518 		return 1;
519 	}
520 
521 	/* OK, update the superblock to recover the freed space.
522 	 * Physical blocks come first: have we wrapped beyond the end of
523 	 * the log?  */
524 	freed = blocknr - journal->j_tail;
525 	if (blocknr < journal->j_tail)
526 		freed = freed + journal->j_last - journal->j_first;
527 
528 	trace_jbd2_cleanup_journal_tail(journal, first_tid, blocknr, freed);
529 	jbd_debug(1,
530 		  "Cleaning journal tail from %d to %d (offset %lu), "
531 		  "freeing %lu\n",
532 		  journal->j_tail_sequence, first_tid, blocknr, freed);
533 
534 	journal->j_free += freed;
535 	journal->j_tail_sequence = first_tid;
536 	journal->j_tail = blocknr;
537 	write_unlock(&journal->j_state_lock);
538 
539 	/*
540 	 * If there is an external journal, we need to make sure that
541 	 * any data blocks that were recently written out --- perhaps
542 	 * by jbd2_log_do_checkpoint() --- are flushed out before we
543 	 * drop the transactions from the external journal.  It's
544 	 * unlikely this will be necessary, especially with a
545 	 * appropriately sized journal, but we need this to guarantee
546 	 * correctness.  Fortunately jbd2_cleanup_journal_tail()
547 	 * doesn't get called all that often.
548 	 */
549 	if ((journal->j_fs_dev != journal->j_dev) &&
550 	    (journal->j_flags & JBD2_BARRIER))
551 		blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
552 	if (!(journal->j_flags & JBD2_ABORT))
553 		jbd2_journal_update_superblock(journal, 1);
554 	return 0;
555 }
556 
557 
558 /* Checkpoint list management */
559 
560 /*
561  * journal_clean_one_cp_list
562  *
563  * Find all the written-back checkpoint buffers in the given list and
564  * release them.
565  *
566  * Called with the journal locked.
567  * Called with j_list_lock held.
568  * Returns number of buffers reaped (for debug)
569  */
570 
571 static int journal_clean_one_cp_list(struct journal_head *jh, int *released)
572 {
573 	struct journal_head *last_jh;
574 	struct journal_head *next_jh = jh;
575 	int ret, freed = 0;
576 
577 	*released = 0;
578 	if (!jh)
579 		return 0;
580 
581 	last_jh = jh->b_cpprev;
582 	do {
583 		jh = next_jh;
584 		next_jh = jh->b_cpnext;
585 		/* Use trylock because of the ranking */
586 		if (jbd_trylock_bh_state(jh2bh(jh))) {
587 			ret = __try_to_free_cp_buf(jh);
588 			if (ret) {
589 				freed++;
590 				if (ret == 2) {
591 					*released = 1;
592 					return freed;
593 				}
594 			}
595 		}
596 		/*
597 		 * This function only frees up some memory
598 		 * if possible so we dont have an obligation
599 		 * to finish processing. Bail out if preemption
600 		 * requested:
601 		 */
602 		if (need_resched())
603 			return freed;
604 	} while (jh != last_jh);
605 
606 	return freed;
607 }
608 
609 /*
610  * journal_clean_checkpoint_list
611  *
612  * Find all the written-back checkpoint buffers in the journal and release them.
613  *
614  * Called with the journal locked.
615  * Called with j_list_lock held.
616  * Returns number of buffers reaped (for debug)
617  */
618 
619 int __jbd2_journal_clean_checkpoint_list(journal_t *journal)
620 {
621 	transaction_t *transaction, *last_transaction, *next_transaction;
622 	int ret = 0;
623 	int released;
624 
625 	transaction = journal->j_checkpoint_transactions;
626 	if (!transaction)
627 		goto out;
628 
629 	last_transaction = transaction->t_cpprev;
630 	next_transaction = transaction;
631 	do {
632 		transaction = next_transaction;
633 		next_transaction = transaction->t_cpnext;
634 		ret += journal_clean_one_cp_list(transaction->
635 				t_checkpoint_list, &released);
636 		/*
637 		 * This function only frees up some memory if possible so we
638 		 * dont have an obligation to finish processing. Bail out if
639 		 * preemption requested:
640 		 */
641 		if (need_resched())
642 			goto out;
643 		if (released)
644 			continue;
645 		/*
646 		 * It is essential that we are as careful as in the case of
647 		 * t_checkpoint_list with removing the buffer from the list as
648 		 * we can possibly see not yet submitted buffers on io_list
649 		 */
650 		ret += journal_clean_one_cp_list(transaction->
651 				t_checkpoint_io_list, &released);
652 		if (need_resched())
653 			goto out;
654 	} while (transaction != last_transaction);
655 out:
656 	return ret;
657 }
658 
659 /*
660  * journal_remove_checkpoint: called after a buffer has been committed
661  * to disk (either by being write-back flushed to disk, or being
662  * committed to the log).
663  *
664  * We cannot safely clean a transaction out of the log until all of the
665  * buffer updates committed in that transaction have safely been stored
666  * elsewhere on disk.  To achieve this, all of the buffers in a
667  * transaction need to be maintained on the transaction's checkpoint
668  * lists until they have been rewritten, at which point this function is
669  * called to remove the buffer from the existing transaction's
670  * checkpoint lists.
671  *
672  * The function returns 1 if it frees the transaction, 0 otherwise.
673  * The function can free jh and bh.
674  *
675  * This function is called with j_list_lock held.
676  * This function is called with jbd_lock_bh_state(jh2bh(jh))
677  */
678 
679 int __jbd2_journal_remove_checkpoint(struct journal_head *jh)
680 {
681 	struct transaction_chp_stats_s *stats;
682 	transaction_t *transaction;
683 	journal_t *journal;
684 	int ret = 0;
685 
686 	JBUFFER_TRACE(jh, "entry");
687 
688 	if ((transaction = jh->b_cp_transaction) == NULL) {
689 		JBUFFER_TRACE(jh, "not on transaction");
690 		goto out;
691 	}
692 	journal = transaction->t_journal;
693 
694 	JBUFFER_TRACE(jh, "removing from transaction");
695 	__buffer_unlink(jh);
696 	jh->b_cp_transaction = NULL;
697 	jbd2_journal_put_journal_head(jh);
698 
699 	if (transaction->t_checkpoint_list != NULL ||
700 	    transaction->t_checkpoint_io_list != NULL)
701 		goto out;
702 
703 	/*
704 	 * There is one special case to worry about: if we have just pulled the
705 	 * buffer off a running or committing transaction's checkpoing list,
706 	 * then even if the checkpoint list is empty, the transaction obviously
707 	 * cannot be dropped!
708 	 *
709 	 * The locking here around t_state is a bit sleazy.
710 	 * See the comment at the end of jbd2_journal_commit_transaction().
711 	 */
712 	if (transaction->t_state != T_FINISHED)
713 		goto out;
714 
715 	/* OK, that was the last buffer for the transaction: we can now
716 	   safely remove this transaction from the log */
717 	stats = &transaction->t_chp_stats;
718 	if (stats->cs_chp_time)
719 		stats->cs_chp_time = jbd2_time_diff(stats->cs_chp_time,
720 						    jiffies);
721 	trace_jbd2_checkpoint_stats(journal->j_fs_dev->bd_dev,
722 				    transaction->t_tid, stats);
723 
724 	__jbd2_journal_drop_transaction(journal, transaction);
725 	kfree(transaction);
726 
727 	/* Just in case anybody was waiting for more transactions to be
728            checkpointed... */
729 	wake_up(&journal->j_wait_logspace);
730 	ret = 1;
731 out:
732 	return ret;
733 }
734 
735 /*
736  * journal_insert_checkpoint: put a committed buffer onto a checkpoint
737  * list so that we know when it is safe to clean the transaction out of
738  * the log.
739  *
740  * Called with the journal locked.
741  * Called with j_list_lock held.
742  */
743 void __jbd2_journal_insert_checkpoint(struct journal_head *jh,
744 			       transaction_t *transaction)
745 {
746 	JBUFFER_TRACE(jh, "entry");
747 	J_ASSERT_JH(jh, buffer_dirty(jh2bh(jh)) || buffer_jbddirty(jh2bh(jh)));
748 	J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
749 
750 	/* Get reference for checkpointing transaction */
751 	jbd2_journal_grab_journal_head(jh2bh(jh));
752 	jh->b_cp_transaction = transaction;
753 
754 	if (!transaction->t_checkpoint_list) {
755 		jh->b_cpnext = jh->b_cpprev = jh;
756 	} else {
757 		jh->b_cpnext = transaction->t_checkpoint_list;
758 		jh->b_cpprev = transaction->t_checkpoint_list->b_cpprev;
759 		jh->b_cpprev->b_cpnext = jh;
760 		jh->b_cpnext->b_cpprev = jh;
761 	}
762 	transaction->t_checkpoint_list = jh;
763 }
764 
765 /*
766  * We've finished with this transaction structure: adios...
767  *
768  * The transaction must have no links except for the checkpoint by this
769  * point.
770  *
771  * Called with the journal locked.
772  * Called with j_list_lock held.
773  */
774 
775 void __jbd2_journal_drop_transaction(journal_t *journal, transaction_t *transaction)
776 {
777 	assert_spin_locked(&journal->j_list_lock);
778 	if (transaction->t_cpnext) {
779 		transaction->t_cpnext->t_cpprev = transaction->t_cpprev;
780 		transaction->t_cpprev->t_cpnext = transaction->t_cpnext;
781 		if (journal->j_checkpoint_transactions == transaction)
782 			journal->j_checkpoint_transactions =
783 				transaction->t_cpnext;
784 		if (journal->j_checkpoint_transactions == transaction)
785 			journal->j_checkpoint_transactions = NULL;
786 	}
787 
788 	J_ASSERT(transaction->t_state == T_FINISHED);
789 	J_ASSERT(transaction->t_buffers == NULL);
790 	J_ASSERT(transaction->t_forget == NULL);
791 	J_ASSERT(transaction->t_iobuf_list == NULL);
792 	J_ASSERT(transaction->t_shadow_list == NULL);
793 	J_ASSERT(transaction->t_log_list == NULL);
794 	J_ASSERT(transaction->t_checkpoint_list == NULL);
795 	J_ASSERT(transaction->t_checkpoint_io_list == NULL);
796 	J_ASSERT(atomic_read(&transaction->t_updates) == 0);
797 	J_ASSERT(journal->j_committing_transaction != transaction);
798 	J_ASSERT(journal->j_running_transaction != transaction);
799 
800 	jbd_debug(1, "Dropping transaction %d, all done\n", transaction->t_tid);
801 }
802