xref: /openbmc/linux/fs/jbd2/journal.c (revision 15e47304)
1 /*
2  * linux/fs/jbd2/journal.c
3  *
4  * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5  *
6  * Copyright 1998 Red Hat corp --- 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  * Generic filesystem journal-writing code; part of the ext2fs
13  * journaling system.
14  *
15  * This file manages journals: areas of disk reserved for logging
16  * transactional updates.  This includes the kernel journaling thread
17  * which is responsible for scheduling updates to the log.
18  *
19  * We do not actually manage the physical storage of the journal in this
20  * file: that is left to a per-journal policy function, which allows us
21  * to store the journal within a filesystem-specified area for ext2
22  * journaling (ext2 can use a reserved inode for storing the log).
23  */
24 
25 #include <linux/module.h>
26 #include <linux/time.h>
27 #include <linux/fs.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
32 #include <linux/mm.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/debugfs.h>
39 #include <linux/seq_file.h>
40 #include <linux/math64.h>
41 #include <linux/hash.h>
42 #include <linux/log2.h>
43 #include <linux/vmalloc.h>
44 #include <linux/backing-dev.h>
45 #include <linux/bitops.h>
46 #include <linux/ratelimit.h>
47 
48 #define CREATE_TRACE_POINTS
49 #include <trace/events/jbd2.h>
50 
51 #include <asm/uaccess.h>
52 #include <asm/page.h>
53 
54 EXPORT_SYMBOL(jbd2_journal_extend);
55 EXPORT_SYMBOL(jbd2_journal_stop);
56 EXPORT_SYMBOL(jbd2_journal_lock_updates);
57 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
58 EXPORT_SYMBOL(jbd2_journal_get_write_access);
59 EXPORT_SYMBOL(jbd2_journal_get_create_access);
60 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
61 EXPORT_SYMBOL(jbd2_journal_set_triggers);
62 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
63 EXPORT_SYMBOL(jbd2_journal_release_buffer);
64 EXPORT_SYMBOL(jbd2_journal_forget);
65 #if 0
66 EXPORT_SYMBOL(journal_sync_buffer);
67 #endif
68 EXPORT_SYMBOL(jbd2_journal_flush);
69 EXPORT_SYMBOL(jbd2_journal_revoke);
70 
71 EXPORT_SYMBOL(jbd2_journal_init_dev);
72 EXPORT_SYMBOL(jbd2_journal_init_inode);
73 EXPORT_SYMBOL(jbd2_journal_check_used_features);
74 EXPORT_SYMBOL(jbd2_journal_check_available_features);
75 EXPORT_SYMBOL(jbd2_journal_set_features);
76 EXPORT_SYMBOL(jbd2_journal_load);
77 EXPORT_SYMBOL(jbd2_journal_destroy);
78 EXPORT_SYMBOL(jbd2_journal_abort);
79 EXPORT_SYMBOL(jbd2_journal_errno);
80 EXPORT_SYMBOL(jbd2_journal_ack_err);
81 EXPORT_SYMBOL(jbd2_journal_clear_err);
82 EXPORT_SYMBOL(jbd2_log_wait_commit);
83 EXPORT_SYMBOL(jbd2_log_start_commit);
84 EXPORT_SYMBOL(jbd2_journal_start_commit);
85 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
86 EXPORT_SYMBOL(jbd2_journal_wipe);
87 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
88 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
89 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
90 EXPORT_SYMBOL(jbd2_journal_force_commit);
91 EXPORT_SYMBOL(jbd2_journal_file_inode);
92 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
93 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
94 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
95 EXPORT_SYMBOL(jbd2_inode_cache);
96 
97 static void __journal_abort_soft (journal_t *journal, int errno);
98 static int jbd2_journal_create_slab(size_t slab_size);
99 
100 /* Checksumming functions */
101 int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
102 {
103 	if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
104 		return 1;
105 
106 	return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
107 }
108 
109 static __u32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
110 {
111 	__u32 csum, old_csum;
112 
113 	old_csum = sb->s_checksum;
114 	sb->s_checksum = 0;
115 	csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
116 	sb->s_checksum = old_csum;
117 
118 	return cpu_to_be32(csum);
119 }
120 
121 int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
122 {
123 	if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
124 		return 1;
125 
126 	return sb->s_checksum == jbd2_superblock_csum(j, sb);
127 }
128 
129 void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
130 {
131 	if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2))
132 		return;
133 
134 	sb->s_checksum = jbd2_superblock_csum(j, sb);
135 }
136 
137 /*
138  * Helper function used to manage commit timeouts
139  */
140 
141 static void commit_timeout(unsigned long __data)
142 {
143 	struct task_struct * p = (struct task_struct *) __data;
144 
145 	wake_up_process(p);
146 }
147 
148 /*
149  * kjournald2: The main thread function used to manage a logging device
150  * journal.
151  *
152  * This kernel thread is responsible for two things:
153  *
154  * 1) COMMIT:  Every so often we need to commit the current state of the
155  *    filesystem to disk.  The journal thread is responsible for writing
156  *    all of the metadata buffers to disk.
157  *
158  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
159  *    of the data in that part of the log has been rewritten elsewhere on
160  *    the disk.  Flushing these old buffers to reclaim space in the log is
161  *    known as checkpointing, and this thread is responsible for that job.
162  */
163 
164 static int kjournald2(void *arg)
165 {
166 	journal_t *journal = arg;
167 	transaction_t *transaction;
168 
169 	/*
170 	 * Set up an interval timer which can be used to trigger a commit wakeup
171 	 * after the commit interval expires
172 	 */
173 	setup_timer(&journal->j_commit_timer, commit_timeout,
174 			(unsigned long)current);
175 
176 	set_freezable();
177 
178 	/* Record that the journal thread is running */
179 	journal->j_task = current;
180 	wake_up(&journal->j_wait_done_commit);
181 
182 	/*
183 	 * And now, wait forever for commit wakeup events.
184 	 */
185 	write_lock(&journal->j_state_lock);
186 
187 loop:
188 	if (journal->j_flags & JBD2_UNMOUNT)
189 		goto end_loop;
190 
191 	jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
192 		journal->j_commit_sequence, journal->j_commit_request);
193 
194 	if (journal->j_commit_sequence != journal->j_commit_request) {
195 		jbd_debug(1, "OK, requests differ\n");
196 		write_unlock(&journal->j_state_lock);
197 		del_timer_sync(&journal->j_commit_timer);
198 		jbd2_journal_commit_transaction(journal);
199 		write_lock(&journal->j_state_lock);
200 		goto loop;
201 	}
202 
203 	wake_up(&journal->j_wait_done_commit);
204 	if (freezing(current)) {
205 		/*
206 		 * The simpler the better. Flushing journal isn't a
207 		 * good idea, because that depends on threads that may
208 		 * be already stopped.
209 		 */
210 		jbd_debug(1, "Now suspending kjournald2\n");
211 		write_unlock(&journal->j_state_lock);
212 		try_to_freeze();
213 		write_lock(&journal->j_state_lock);
214 	} else {
215 		/*
216 		 * We assume on resume that commits are already there,
217 		 * so we don't sleep
218 		 */
219 		DEFINE_WAIT(wait);
220 		int should_sleep = 1;
221 
222 		prepare_to_wait(&journal->j_wait_commit, &wait,
223 				TASK_INTERRUPTIBLE);
224 		if (journal->j_commit_sequence != journal->j_commit_request)
225 			should_sleep = 0;
226 		transaction = journal->j_running_transaction;
227 		if (transaction && time_after_eq(jiffies,
228 						transaction->t_expires))
229 			should_sleep = 0;
230 		if (journal->j_flags & JBD2_UNMOUNT)
231 			should_sleep = 0;
232 		if (should_sleep) {
233 			write_unlock(&journal->j_state_lock);
234 			schedule();
235 			write_lock(&journal->j_state_lock);
236 		}
237 		finish_wait(&journal->j_wait_commit, &wait);
238 	}
239 
240 	jbd_debug(1, "kjournald2 wakes\n");
241 
242 	/*
243 	 * Were we woken up by a commit wakeup event?
244 	 */
245 	transaction = journal->j_running_transaction;
246 	if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
247 		journal->j_commit_request = transaction->t_tid;
248 		jbd_debug(1, "woke because of timeout\n");
249 	}
250 	goto loop;
251 
252 end_loop:
253 	write_unlock(&journal->j_state_lock);
254 	del_timer_sync(&journal->j_commit_timer);
255 	journal->j_task = NULL;
256 	wake_up(&journal->j_wait_done_commit);
257 	jbd_debug(1, "Journal thread exiting.\n");
258 	return 0;
259 }
260 
261 static int jbd2_journal_start_thread(journal_t *journal)
262 {
263 	struct task_struct *t;
264 
265 	t = kthread_run(kjournald2, journal, "jbd2/%s",
266 			journal->j_devname);
267 	if (IS_ERR(t))
268 		return PTR_ERR(t);
269 
270 	wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
271 	return 0;
272 }
273 
274 static void journal_kill_thread(journal_t *journal)
275 {
276 	write_lock(&journal->j_state_lock);
277 	journal->j_flags |= JBD2_UNMOUNT;
278 
279 	while (journal->j_task) {
280 		wake_up(&journal->j_wait_commit);
281 		write_unlock(&journal->j_state_lock);
282 		wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
283 		write_lock(&journal->j_state_lock);
284 	}
285 	write_unlock(&journal->j_state_lock);
286 }
287 
288 /*
289  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
290  *
291  * Writes a metadata buffer to a given disk block.  The actual IO is not
292  * performed but a new buffer_head is constructed which labels the data
293  * to be written with the correct destination disk block.
294  *
295  * Any magic-number escaping which needs to be done will cause a
296  * copy-out here.  If the buffer happens to start with the
297  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
298  * magic number is only written to the log for descripter blocks.  In
299  * this case, we copy the data and replace the first word with 0, and we
300  * return a result code which indicates that this buffer needs to be
301  * marked as an escaped buffer in the corresponding log descriptor
302  * block.  The missing word can then be restored when the block is read
303  * during recovery.
304  *
305  * If the source buffer has already been modified by a new transaction
306  * since we took the last commit snapshot, we use the frozen copy of
307  * that data for IO.  If we end up using the existing buffer_head's data
308  * for the write, then we *have* to lock the buffer to prevent anyone
309  * else from using and possibly modifying it while the IO is in
310  * progress.
311  *
312  * The function returns a pointer to the buffer_heads to be used for IO.
313  *
314  * We assume that the journal has already been locked in this function.
315  *
316  * Return value:
317  *  <0: Error
318  * >=0: Finished OK
319  *
320  * On success:
321  * Bit 0 set == escape performed on the data
322  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
323  */
324 
325 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
326 				  struct journal_head  *jh_in,
327 				  struct journal_head **jh_out,
328 				  unsigned long long blocknr)
329 {
330 	int need_copy_out = 0;
331 	int done_copy_out = 0;
332 	int do_escape = 0;
333 	char *mapped_data;
334 	struct buffer_head *new_bh;
335 	struct journal_head *new_jh;
336 	struct page *new_page;
337 	unsigned int new_offset;
338 	struct buffer_head *bh_in = jh2bh(jh_in);
339 	journal_t *journal = transaction->t_journal;
340 
341 	/*
342 	 * The buffer really shouldn't be locked: only the current committing
343 	 * transaction is allowed to write it, so nobody else is allowed
344 	 * to do any IO.
345 	 *
346 	 * akpm: except if we're journalling data, and write() output is
347 	 * also part of a shared mapping, and another thread has
348 	 * decided to launch a writepage() against this buffer.
349 	 */
350 	J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
351 
352 retry_alloc:
353 	new_bh = alloc_buffer_head(GFP_NOFS);
354 	if (!new_bh) {
355 		/*
356 		 * Failure is not an option, but __GFP_NOFAIL is going
357 		 * away; so we retry ourselves here.
358 		 */
359 		congestion_wait(BLK_RW_ASYNC, HZ/50);
360 		goto retry_alloc;
361 	}
362 
363 	/* keep subsequent assertions sane */
364 	new_bh->b_state = 0;
365 	init_buffer(new_bh, NULL, NULL);
366 	atomic_set(&new_bh->b_count, 1);
367 	new_jh = jbd2_journal_add_journal_head(new_bh);	/* This sleeps */
368 
369 	/*
370 	 * If a new transaction has already done a buffer copy-out, then
371 	 * we use that version of the data for the commit.
372 	 */
373 	jbd_lock_bh_state(bh_in);
374 repeat:
375 	if (jh_in->b_frozen_data) {
376 		done_copy_out = 1;
377 		new_page = virt_to_page(jh_in->b_frozen_data);
378 		new_offset = offset_in_page(jh_in->b_frozen_data);
379 	} else {
380 		new_page = jh2bh(jh_in)->b_page;
381 		new_offset = offset_in_page(jh2bh(jh_in)->b_data);
382 	}
383 
384 	mapped_data = kmap_atomic(new_page);
385 	/*
386 	 * Fire data frozen trigger if data already wasn't frozen.  Do this
387 	 * before checking for escaping, as the trigger may modify the magic
388 	 * offset.  If a copy-out happens afterwards, it will have the correct
389 	 * data in the buffer.
390 	 */
391 	if (!done_copy_out)
392 		jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
393 					   jh_in->b_triggers);
394 
395 	/*
396 	 * Check for escaping
397 	 */
398 	if (*((__be32 *)(mapped_data + new_offset)) ==
399 				cpu_to_be32(JBD2_MAGIC_NUMBER)) {
400 		need_copy_out = 1;
401 		do_escape = 1;
402 	}
403 	kunmap_atomic(mapped_data);
404 
405 	/*
406 	 * Do we need to do a data copy?
407 	 */
408 	if (need_copy_out && !done_copy_out) {
409 		char *tmp;
410 
411 		jbd_unlock_bh_state(bh_in);
412 		tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
413 		if (!tmp) {
414 			jbd2_journal_put_journal_head(new_jh);
415 			return -ENOMEM;
416 		}
417 		jbd_lock_bh_state(bh_in);
418 		if (jh_in->b_frozen_data) {
419 			jbd2_free(tmp, bh_in->b_size);
420 			goto repeat;
421 		}
422 
423 		jh_in->b_frozen_data = tmp;
424 		mapped_data = kmap_atomic(new_page);
425 		memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
426 		kunmap_atomic(mapped_data);
427 
428 		new_page = virt_to_page(tmp);
429 		new_offset = offset_in_page(tmp);
430 		done_copy_out = 1;
431 
432 		/*
433 		 * This isn't strictly necessary, as we're using frozen
434 		 * data for the escaping, but it keeps consistency with
435 		 * b_frozen_data usage.
436 		 */
437 		jh_in->b_frozen_triggers = jh_in->b_triggers;
438 	}
439 
440 	/*
441 	 * Did we need to do an escaping?  Now we've done all the
442 	 * copying, we can finally do so.
443 	 */
444 	if (do_escape) {
445 		mapped_data = kmap_atomic(new_page);
446 		*((unsigned int *)(mapped_data + new_offset)) = 0;
447 		kunmap_atomic(mapped_data);
448 	}
449 
450 	set_bh_page(new_bh, new_page, new_offset);
451 	new_jh->b_transaction = NULL;
452 	new_bh->b_size = jh2bh(jh_in)->b_size;
453 	new_bh->b_bdev = transaction->t_journal->j_dev;
454 	new_bh->b_blocknr = blocknr;
455 	set_buffer_mapped(new_bh);
456 	set_buffer_dirty(new_bh);
457 
458 	*jh_out = new_jh;
459 
460 	/*
461 	 * The to-be-written buffer needs to get moved to the io queue,
462 	 * and the original buffer whose contents we are shadowing or
463 	 * copying is moved to the transaction's shadow queue.
464 	 */
465 	JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
466 	spin_lock(&journal->j_list_lock);
467 	__jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
468 	spin_unlock(&journal->j_list_lock);
469 	jbd_unlock_bh_state(bh_in);
470 
471 	JBUFFER_TRACE(new_jh, "file as BJ_IO");
472 	jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
473 
474 	return do_escape | (done_copy_out << 1);
475 }
476 
477 /*
478  * Allocation code for the journal file.  Manage the space left in the
479  * journal, so that we can begin checkpointing when appropriate.
480  */
481 
482 /*
483  * __jbd2_log_space_left: Return the number of free blocks left in the journal.
484  *
485  * Called with the journal already locked.
486  *
487  * Called under j_state_lock
488  */
489 
490 int __jbd2_log_space_left(journal_t *journal)
491 {
492 	int left = journal->j_free;
493 
494 	/* assert_spin_locked(&journal->j_state_lock); */
495 
496 	/*
497 	 * Be pessimistic here about the number of those free blocks which
498 	 * might be required for log descriptor control blocks.
499 	 */
500 
501 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
502 
503 	left -= MIN_LOG_RESERVED_BLOCKS;
504 
505 	if (left <= 0)
506 		return 0;
507 	left -= (left >> 3);
508 	return left;
509 }
510 
511 /*
512  * Called with j_state_lock locked for writing.
513  * Returns true if a transaction commit was started.
514  */
515 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
516 {
517 	/*
518 	 * The only transaction we can possibly wait upon is the
519 	 * currently running transaction (if it exists).  Otherwise,
520 	 * the target tid must be an old one.
521 	 */
522 	if (journal->j_running_transaction &&
523 	    journal->j_running_transaction->t_tid == target) {
524 		/*
525 		 * We want a new commit: OK, mark the request and wakeup the
526 		 * commit thread.  We do _not_ do the commit ourselves.
527 		 */
528 
529 		journal->j_commit_request = target;
530 		jbd_debug(1, "JBD2: requesting commit %d/%d\n",
531 			  journal->j_commit_request,
532 			  journal->j_commit_sequence);
533 		wake_up(&journal->j_wait_commit);
534 		return 1;
535 	} else if (!tid_geq(journal->j_commit_request, target))
536 		/* This should never happen, but if it does, preserve
537 		   the evidence before kjournald goes into a loop and
538 		   increments j_commit_sequence beyond all recognition. */
539 		WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
540 			  journal->j_commit_request,
541 			  journal->j_commit_sequence,
542 			  target, journal->j_running_transaction ?
543 			  journal->j_running_transaction->t_tid : 0);
544 	return 0;
545 }
546 
547 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
548 {
549 	int ret;
550 
551 	write_lock(&journal->j_state_lock);
552 	ret = __jbd2_log_start_commit(journal, tid);
553 	write_unlock(&journal->j_state_lock);
554 	return ret;
555 }
556 
557 /*
558  * Force and wait upon a commit if the calling process is not within
559  * transaction.  This is used for forcing out undo-protected data which contains
560  * bitmaps, when the fs is running out of space.
561  *
562  * We can only force the running transaction if we don't have an active handle;
563  * otherwise, we will deadlock.
564  *
565  * Returns true if a transaction was started.
566  */
567 int jbd2_journal_force_commit_nested(journal_t *journal)
568 {
569 	transaction_t *transaction = NULL;
570 	tid_t tid;
571 	int need_to_start = 0;
572 
573 	read_lock(&journal->j_state_lock);
574 	if (journal->j_running_transaction && !current->journal_info) {
575 		transaction = journal->j_running_transaction;
576 		if (!tid_geq(journal->j_commit_request, transaction->t_tid))
577 			need_to_start = 1;
578 	} else if (journal->j_committing_transaction)
579 		transaction = journal->j_committing_transaction;
580 
581 	if (!transaction) {
582 		read_unlock(&journal->j_state_lock);
583 		return 0;	/* Nothing to retry */
584 	}
585 
586 	tid = transaction->t_tid;
587 	read_unlock(&journal->j_state_lock);
588 	if (need_to_start)
589 		jbd2_log_start_commit(journal, tid);
590 	jbd2_log_wait_commit(journal, tid);
591 	return 1;
592 }
593 
594 /*
595  * Start a commit of the current running transaction (if any).  Returns true
596  * if a transaction is going to be committed (or is currently already
597  * committing), and fills its tid in at *ptid
598  */
599 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
600 {
601 	int ret = 0;
602 
603 	write_lock(&journal->j_state_lock);
604 	if (journal->j_running_transaction) {
605 		tid_t tid = journal->j_running_transaction->t_tid;
606 
607 		__jbd2_log_start_commit(journal, tid);
608 		/* There's a running transaction and we've just made sure
609 		 * it's commit has been scheduled. */
610 		if (ptid)
611 			*ptid = tid;
612 		ret = 1;
613 	} else if (journal->j_committing_transaction) {
614 		/*
615 		 * If commit has been started, then we have to wait for
616 		 * completion of that transaction.
617 		 */
618 		if (ptid)
619 			*ptid = journal->j_committing_transaction->t_tid;
620 		ret = 1;
621 	}
622 	write_unlock(&journal->j_state_lock);
623 	return ret;
624 }
625 
626 /*
627  * Return 1 if a given transaction has not yet sent barrier request
628  * connected with a transaction commit. If 0 is returned, transaction
629  * may or may not have sent the barrier. Used to avoid sending barrier
630  * twice in common cases.
631  */
632 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
633 {
634 	int ret = 0;
635 	transaction_t *commit_trans;
636 
637 	if (!(journal->j_flags & JBD2_BARRIER))
638 		return 0;
639 	read_lock(&journal->j_state_lock);
640 	/* Transaction already committed? */
641 	if (tid_geq(journal->j_commit_sequence, tid))
642 		goto out;
643 	commit_trans = journal->j_committing_transaction;
644 	if (!commit_trans || commit_trans->t_tid != tid) {
645 		ret = 1;
646 		goto out;
647 	}
648 	/*
649 	 * Transaction is being committed and we already proceeded to
650 	 * submitting a flush to fs partition?
651 	 */
652 	if (journal->j_fs_dev != journal->j_dev) {
653 		if (!commit_trans->t_need_data_flush ||
654 		    commit_trans->t_state >= T_COMMIT_DFLUSH)
655 			goto out;
656 	} else {
657 		if (commit_trans->t_state >= T_COMMIT_JFLUSH)
658 			goto out;
659 	}
660 	ret = 1;
661 out:
662 	read_unlock(&journal->j_state_lock);
663 	return ret;
664 }
665 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
666 
667 /*
668  * Wait for a specified commit to complete.
669  * The caller may not hold the journal lock.
670  */
671 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
672 {
673 	int err = 0;
674 
675 	read_lock(&journal->j_state_lock);
676 #ifdef CONFIG_JBD2_DEBUG
677 	if (!tid_geq(journal->j_commit_request, tid)) {
678 		printk(KERN_EMERG
679 		       "%s: error: j_commit_request=%d, tid=%d\n",
680 		       __func__, journal->j_commit_request, tid);
681 	}
682 #endif
683 	while (tid_gt(tid, journal->j_commit_sequence)) {
684 		jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
685 				  tid, journal->j_commit_sequence);
686 		wake_up(&journal->j_wait_commit);
687 		read_unlock(&journal->j_state_lock);
688 		wait_event(journal->j_wait_done_commit,
689 				!tid_gt(tid, journal->j_commit_sequence));
690 		read_lock(&journal->j_state_lock);
691 	}
692 	read_unlock(&journal->j_state_lock);
693 
694 	if (unlikely(is_journal_aborted(journal))) {
695 		printk(KERN_EMERG "journal commit I/O error\n");
696 		err = -EIO;
697 	}
698 	return err;
699 }
700 
701 /*
702  * Log buffer allocation routines:
703  */
704 
705 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
706 {
707 	unsigned long blocknr;
708 
709 	write_lock(&journal->j_state_lock);
710 	J_ASSERT(journal->j_free > 1);
711 
712 	blocknr = journal->j_head;
713 	journal->j_head++;
714 	journal->j_free--;
715 	if (journal->j_head == journal->j_last)
716 		journal->j_head = journal->j_first;
717 	write_unlock(&journal->j_state_lock);
718 	return jbd2_journal_bmap(journal, blocknr, retp);
719 }
720 
721 /*
722  * Conversion of logical to physical block numbers for the journal
723  *
724  * On external journals the journal blocks are identity-mapped, so
725  * this is a no-op.  If needed, we can use j_blk_offset - everything is
726  * ready.
727  */
728 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
729 		 unsigned long long *retp)
730 {
731 	int err = 0;
732 	unsigned long long ret;
733 
734 	if (journal->j_inode) {
735 		ret = bmap(journal->j_inode, blocknr);
736 		if (ret)
737 			*retp = ret;
738 		else {
739 			printk(KERN_ALERT "%s: journal block not found "
740 					"at offset %lu on %s\n",
741 			       __func__, blocknr, journal->j_devname);
742 			err = -EIO;
743 			__journal_abort_soft(journal, err);
744 		}
745 	} else {
746 		*retp = blocknr; /* +journal->j_blk_offset */
747 	}
748 	return err;
749 }
750 
751 /*
752  * We play buffer_head aliasing tricks to write data/metadata blocks to
753  * the journal without copying their contents, but for journal
754  * descriptor blocks we do need to generate bona fide buffers.
755  *
756  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
757  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
758  * But we don't bother doing that, so there will be coherency problems with
759  * mmaps of blockdevs which hold live JBD-controlled filesystems.
760  */
761 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
762 {
763 	struct buffer_head *bh;
764 	unsigned long long blocknr;
765 	int err;
766 
767 	err = jbd2_journal_next_log_block(journal, &blocknr);
768 
769 	if (err)
770 		return NULL;
771 
772 	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
773 	if (!bh)
774 		return NULL;
775 	lock_buffer(bh);
776 	memset(bh->b_data, 0, journal->j_blocksize);
777 	set_buffer_uptodate(bh);
778 	unlock_buffer(bh);
779 	BUFFER_TRACE(bh, "return this buffer");
780 	return jbd2_journal_add_journal_head(bh);
781 }
782 
783 /*
784  * Return tid of the oldest transaction in the journal and block in the journal
785  * where the transaction starts.
786  *
787  * If the journal is now empty, return which will be the next transaction ID
788  * we will write and where will that transaction start.
789  *
790  * The return value is 0 if journal tail cannot be pushed any further, 1 if
791  * it can.
792  */
793 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
794 			      unsigned long *block)
795 {
796 	transaction_t *transaction;
797 	int ret;
798 
799 	read_lock(&journal->j_state_lock);
800 	spin_lock(&journal->j_list_lock);
801 	transaction = journal->j_checkpoint_transactions;
802 	if (transaction) {
803 		*tid = transaction->t_tid;
804 		*block = transaction->t_log_start;
805 	} else if ((transaction = journal->j_committing_transaction) != NULL) {
806 		*tid = transaction->t_tid;
807 		*block = transaction->t_log_start;
808 	} else if ((transaction = journal->j_running_transaction) != NULL) {
809 		*tid = transaction->t_tid;
810 		*block = journal->j_head;
811 	} else {
812 		*tid = journal->j_transaction_sequence;
813 		*block = journal->j_head;
814 	}
815 	ret = tid_gt(*tid, journal->j_tail_sequence);
816 	spin_unlock(&journal->j_list_lock);
817 	read_unlock(&journal->j_state_lock);
818 
819 	return ret;
820 }
821 
822 /*
823  * Update information in journal structure and in on disk journal superblock
824  * about log tail. This function does not check whether information passed in
825  * really pushes log tail further. It's responsibility of the caller to make
826  * sure provided log tail information is valid (e.g. by holding
827  * j_checkpoint_mutex all the time between computing log tail and calling this
828  * function as is the case with jbd2_cleanup_journal_tail()).
829  *
830  * Requires j_checkpoint_mutex
831  */
832 void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
833 {
834 	unsigned long freed;
835 
836 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
837 
838 	/*
839 	 * We cannot afford for write to remain in drive's caches since as
840 	 * soon as we update j_tail, next transaction can start reusing journal
841 	 * space and if we lose sb update during power failure we'd replay
842 	 * old transaction with possibly newly overwritten data.
843 	 */
844 	jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
845 	write_lock(&journal->j_state_lock);
846 	freed = block - journal->j_tail;
847 	if (block < journal->j_tail)
848 		freed += journal->j_last - journal->j_first;
849 
850 	trace_jbd2_update_log_tail(journal, tid, block, freed);
851 	jbd_debug(1,
852 		  "Cleaning journal tail from %d to %d (offset %lu), "
853 		  "freeing %lu\n",
854 		  journal->j_tail_sequence, tid, block, freed);
855 
856 	journal->j_free += freed;
857 	journal->j_tail_sequence = tid;
858 	journal->j_tail = block;
859 	write_unlock(&journal->j_state_lock);
860 }
861 
862 /*
863  * This is a variaon of __jbd2_update_log_tail which checks for validity of
864  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
865  * with other threads updating log tail.
866  */
867 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
868 {
869 	mutex_lock(&journal->j_checkpoint_mutex);
870 	if (tid_gt(tid, journal->j_tail_sequence))
871 		__jbd2_update_log_tail(journal, tid, block);
872 	mutex_unlock(&journal->j_checkpoint_mutex);
873 }
874 
875 struct jbd2_stats_proc_session {
876 	journal_t *journal;
877 	struct transaction_stats_s *stats;
878 	int start;
879 	int max;
880 };
881 
882 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
883 {
884 	return *pos ? NULL : SEQ_START_TOKEN;
885 }
886 
887 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
888 {
889 	return NULL;
890 }
891 
892 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
893 {
894 	struct jbd2_stats_proc_session *s = seq->private;
895 
896 	if (v != SEQ_START_TOKEN)
897 		return 0;
898 	seq_printf(seq, "%lu transaction, each up to %u blocks\n",
899 			s->stats->ts_tid,
900 			s->journal->j_max_transaction_buffers);
901 	if (s->stats->ts_tid == 0)
902 		return 0;
903 	seq_printf(seq, "average: \n  %ums waiting for transaction\n",
904 	    jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
905 	seq_printf(seq, "  %ums running transaction\n",
906 	    jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
907 	seq_printf(seq, "  %ums transaction was being locked\n",
908 	    jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
909 	seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
910 	    jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
911 	seq_printf(seq, "  %ums logging transaction\n",
912 	    jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
913 	seq_printf(seq, "  %lluus average transaction commit time\n",
914 		   div_u64(s->journal->j_average_commit_time, 1000));
915 	seq_printf(seq, "  %lu handles per transaction\n",
916 	    s->stats->run.rs_handle_count / s->stats->ts_tid);
917 	seq_printf(seq, "  %lu blocks per transaction\n",
918 	    s->stats->run.rs_blocks / s->stats->ts_tid);
919 	seq_printf(seq, "  %lu logged blocks per transaction\n",
920 	    s->stats->run.rs_blocks_logged / s->stats->ts_tid);
921 	return 0;
922 }
923 
924 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
925 {
926 }
927 
928 static const struct seq_operations jbd2_seq_info_ops = {
929 	.start  = jbd2_seq_info_start,
930 	.next   = jbd2_seq_info_next,
931 	.stop   = jbd2_seq_info_stop,
932 	.show   = jbd2_seq_info_show,
933 };
934 
935 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
936 {
937 	journal_t *journal = PDE(inode)->data;
938 	struct jbd2_stats_proc_session *s;
939 	int rc, size;
940 
941 	s = kmalloc(sizeof(*s), GFP_KERNEL);
942 	if (s == NULL)
943 		return -ENOMEM;
944 	size = sizeof(struct transaction_stats_s);
945 	s->stats = kmalloc(size, GFP_KERNEL);
946 	if (s->stats == NULL) {
947 		kfree(s);
948 		return -ENOMEM;
949 	}
950 	spin_lock(&journal->j_history_lock);
951 	memcpy(s->stats, &journal->j_stats, size);
952 	s->journal = journal;
953 	spin_unlock(&journal->j_history_lock);
954 
955 	rc = seq_open(file, &jbd2_seq_info_ops);
956 	if (rc == 0) {
957 		struct seq_file *m = file->private_data;
958 		m->private = s;
959 	} else {
960 		kfree(s->stats);
961 		kfree(s);
962 	}
963 	return rc;
964 
965 }
966 
967 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
968 {
969 	struct seq_file *seq = file->private_data;
970 	struct jbd2_stats_proc_session *s = seq->private;
971 	kfree(s->stats);
972 	kfree(s);
973 	return seq_release(inode, file);
974 }
975 
976 static const struct file_operations jbd2_seq_info_fops = {
977 	.owner		= THIS_MODULE,
978 	.open           = jbd2_seq_info_open,
979 	.read           = seq_read,
980 	.llseek         = seq_lseek,
981 	.release        = jbd2_seq_info_release,
982 };
983 
984 static struct proc_dir_entry *proc_jbd2_stats;
985 
986 static void jbd2_stats_proc_init(journal_t *journal)
987 {
988 	journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
989 	if (journal->j_proc_entry) {
990 		proc_create_data("info", S_IRUGO, journal->j_proc_entry,
991 				 &jbd2_seq_info_fops, journal);
992 	}
993 }
994 
995 static void jbd2_stats_proc_exit(journal_t *journal)
996 {
997 	remove_proc_entry("info", journal->j_proc_entry);
998 	remove_proc_entry(journal->j_devname, proc_jbd2_stats);
999 }
1000 
1001 /*
1002  * Management for journal control blocks: functions to create and
1003  * destroy journal_t structures, and to initialise and read existing
1004  * journal blocks from disk.  */
1005 
1006 /* First: create and setup a journal_t object in memory.  We initialise
1007  * very few fields yet: that has to wait until we have created the
1008  * journal structures from from scratch, or loaded them from disk. */
1009 
1010 static journal_t * journal_init_common (void)
1011 {
1012 	journal_t *journal;
1013 	int err;
1014 
1015 	journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1016 	if (!journal)
1017 		return NULL;
1018 
1019 	init_waitqueue_head(&journal->j_wait_transaction_locked);
1020 	init_waitqueue_head(&journal->j_wait_logspace);
1021 	init_waitqueue_head(&journal->j_wait_done_commit);
1022 	init_waitqueue_head(&journal->j_wait_checkpoint);
1023 	init_waitqueue_head(&journal->j_wait_commit);
1024 	init_waitqueue_head(&journal->j_wait_updates);
1025 	mutex_init(&journal->j_barrier);
1026 	mutex_init(&journal->j_checkpoint_mutex);
1027 	spin_lock_init(&journal->j_revoke_lock);
1028 	spin_lock_init(&journal->j_list_lock);
1029 	rwlock_init(&journal->j_state_lock);
1030 
1031 	journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1032 	journal->j_min_batch_time = 0;
1033 	journal->j_max_batch_time = 15000; /* 15ms */
1034 
1035 	/* The journal is marked for error until we succeed with recovery! */
1036 	journal->j_flags = JBD2_ABORT;
1037 
1038 	/* Set up a default-sized revoke table for the new mount. */
1039 	err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1040 	if (err) {
1041 		kfree(journal);
1042 		return NULL;
1043 	}
1044 
1045 	spin_lock_init(&journal->j_history_lock);
1046 
1047 	return journal;
1048 }
1049 
1050 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1051  *
1052  * Create a journal structure assigned some fixed set of disk blocks to
1053  * the journal.  We don't actually touch those disk blocks yet, but we
1054  * need to set up all of the mapping information to tell the journaling
1055  * system where the journal blocks are.
1056  *
1057  */
1058 
1059 /**
1060  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1061  *  @bdev: Block device on which to create the journal
1062  *  @fs_dev: Device which hold journalled filesystem for this journal.
1063  *  @start: Block nr Start of journal.
1064  *  @len:  Length of the journal in blocks.
1065  *  @blocksize: blocksize of journalling device
1066  *
1067  *  Returns: a newly created journal_t *
1068  *
1069  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1070  *  range of blocks on an arbitrary block device.
1071  *
1072  */
1073 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1074 			struct block_device *fs_dev,
1075 			unsigned long long start, int len, int blocksize)
1076 {
1077 	journal_t *journal = journal_init_common();
1078 	struct buffer_head *bh;
1079 	char *p;
1080 	int n;
1081 
1082 	if (!journal)
1083 		return NULL;
1084 
1085 	/* journal descriptor can store up to n blocks -bzzz */
1086 	journal->j_blocksize = blocksize;
1087 	journal->j_dev = bdev;
1088 	journal->j_fs_dev = fs_dev;
1089 	journal->j_blk_offset = start;
1090 	journal->j_maxlen = len;
1091 	bdevname(journal->j_dev, journal->j_devname);
1092 	p = journal->j_devname;
1093 	while ((p = strchr(p, '/')))
1094 		*p = '!';
1095 	jbd2_stats_proc_init(journal);
1096 	n = journal->j_blocksize / sizeof(journal_block_tag_t);
1097 	journal->j_wbufsize = n;
1098 	journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1099 	if (!journal->j_wbuf) {
1100 		printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1101 			__func__);
1102 		goto out_err;
1103 	}
1104 
1105 	bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1106 	if (!bh) {
1107 		printk(KERN_ERR
1108 		       "%s: Cannot get buffer for journal superblock\n",
1109 		       __func__);
1110 		goto out_err;
1111 	}
1112 	journal->j_sb_buffer = bh;
1113 	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1114 
1115 	return journal;
1116 out_err:
1117 	kfree(journal->j_wbuf);
1118 	jbd2_stats_proc_exit(journal);
1119 	kfree(journal);
1120 	return NULL;
1121 }
1122 
1123 /**
1124  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1125  *  @inode: An inode to create the journal in
1126  *
1127  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1128  * the journal.  The inode must exist already, must support bmap() and
1129  * must have all data blocks preallocated.
1130  */
1131 journal_t * jbd2_journal_init_inode (struct inode *inode)
1132 {
1133 	struct buffer_head *bh;
1134 	journal_t *journal = journal_init_common();
1135 	char *p;
1136 	int err;
1137 	int n;
1138 	unsigned long long blocknr;
1139 
1140 	if (!journal)
1141 		return NULL;
1142 
1143 	journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1144 	journal->j_inode = inode;
1145 	bdevname(journal->j_dev, journal->j_devname);
1146 	p = journal->j_devname;
1147 	while ((p = strchr(p, '/')))
1148 		*p = '!';
1149 	p = journal->j_devname + strlen(journal->j_devname);
1150 	sprintf(p, "-%lu", journal->j_inode->i_ino);
1151 	jbd_debug(1,
1152 		  "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1153 		  journal, inode->i_sb->s_id, inode->i_ino,
1154 		  (long long) inode->i_size,
1155 		  inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1156 
1157 	journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1158 	journal->j_blocksize = inode->i_sb->s_blocksize;
1159 	jbd2_stats_proc_init(journal);
1160 
1161 	/* journal descriptor can store up to n blocks -bzzz */
1162 	n = journal->j_blocksize / sizeof(journal_block_tag_t);
1163 	journal->j_wbufsize = n;
1164 	journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1165 	if (!journal->j_wbuf) {
1166 		printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1167 			__func__);
1168 		goto out_err;
1169 	}
1170 
1171 	err = jbd2_journal_bmap(journal, 0, &blocknr);
1172 	/* If that failed, give up */
1173 	if (err) {
1174 		printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1175 		       __func__);
1176 		goto out_err;
1177 	}
1178 
1179 	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1180 	if (!bh) {
1181 		printk(KERN_ERR
1182 		       "%s: Cannot get buffer for journal superblock\n",
1183 		       __func__);
1184 		goto out_err;
1185 	}
1186 	journal->j_sb_buffer = bh;
1187 	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1188 
1189 	return journal;
1190 out_err:
1191 	kfree(journal->j_wbuf);
1192 	jbd2_stats_proc_exit(journal);
1193 	kfree(journal);
1194 	return NULL;
1195 }
1196 
1197 /*
1198  * If the journal init or create aborts, we need to mark the journal
1199  * superblock as being NULL to prevent the journal destroy from writing
1200  * back a bogus superblock.
1201  */
1202 static void journal_fail_superblock (journal_t *journal)
1203 {
1204 	struct buffer_head *bh = journal->j_sb_buffer;
1205 	brelse(bh);
1206 	journal->j_sb_buffer = NULL;
1207 }
1208 
1209 /*
1210  * Given a journal_t structure, initialise the various fields for
1211  * startup of a new journaling session.  We use this both when creating
1212  * a journal, and after recovering an old journal to reset it for
1213  * subsequent use.
1214  */
1215 
1216 static int journal_reset(journal_t *journal)
1217 {
1218 	journal_superblock_t *sb = journal->j_superblock;
1219 	unsigned long long first, last;
1220 
1221 	first = be32_to_cpu(sb->s_first);
1222 	last = be32_to_cpu(sb->s_maxlen);
1223 	if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1224 		printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1225 		       first, last);
1226 		journal_fail_superblock(journal);
1227 		return -EINVAL;
1228 	}
1229 
1230 	journal->j_first = first;
1231 	journal->j_last = last;
1232 
1233 	journal->j_head = first;
1234 	journal->j_tail = first;
1235 	journal->j_free = last - first;
1236 
1237 	journal->j_tail_sequence = journal->j_transaction_sequence;
1238 	journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1239 	journal->j_commit_request = journal->j_commit_sequence;
1240 
1241 	journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1242 
1243 	/*
1244 	 * As a special case, if the on-disk copy is already marked as needing
1245 	 * no recovery (s_start == 0), then we can safely defer the superblock
1246 	 * update until the next commit by setting JBD2_FLUSHED.  This avoids
1247 	 * attempting a write to a potential-readonly device.
1248 	 */
1249 	if (sb->s_start == 0) {
1250 		jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1251 			"(start %ld, seq %d, errno %d)\n",
1252 			journal->j_tail, journal->j_tail_sequence,
1253 			journal->j_errno);
1254 		journal->j_flags |= JBD2_FLUSHED;
1255 	} else {
1256 		/* Lock here to make assertions happy... */
1257 		mutex_lock(&journal->j_checkpoint_mutex);
1258 		/*
1259 		 * Update log tail information. We use WRITE_FUA since new
1260 		 * transaction will start reusing journal space and so we
1261 		 * must make sure information about current log tail is on
1262 		 * disk before that.
1263 		 */
1264 		jbd2_journal_update_sb_log_tail(journal,
1265 						journal->j_tail_sequence,
1266 						journal->j_tail,
1267 						WRITE_FUA);
1268 		mutex_unlock(&journal->j_checkpoint_mutex);
1269 	}
1270 	return jbd2_journal_start_thread(journal);
1271 }
1272 
1273 static void jbd2_write_superblock(journal_t *journal, int write_op)
1274 {
1275 	struct buffer_head *bh = journal->j_sb_buffer;
1276 	int ret;
1277 
1278 	trace_jbd2_write_superblock(journal, write_op);
1279 	if (!(journal->j_flags & JBD2_BARRIER))
1280 		write_op &= ~(REQ_FUA | REQ_FLUSH);
1281 	lock_buffer(bh);
1282 	if (buffer_write_io_error(bh)) {
1283 		/*
1284 		 * Oh, dear.  A previous attempt to write the journal
1285 		 * superblock failed.  This could happen because the
1286 		 * USB device was yanked out.  Or it could happen to
1287 		 * be a transient write error and maybe the block will
1288 		 * be remapped.  Nothing we can do but to retry the
1289 		 * write and hope for the best.
1290 		 */
1291 		printk(KERN_ERR "JBD2: previous I/O error detected "
1292 		       "for journal superblock update for %s.\n",
1293 		       journal->j_devname);
1294 		clear_buffer_write_io_error(bh);
1295 		set_buffer_uptodate(bh);
1296 	}
1297 	get_bh(bh);
1298 	bh->b_end_io = end_buffer_write_sync;
1299 	ret = submit_bh(write_op, bh);
1300 	wait_on_buffer(bh);
1301 	if (buffer_write_io_error(bh)) {
1302 		clear_buffer_write_io_error(bh);
1303 		set_buffer_uptodate(bh);
1304 		ret = -EIO;
1305 	}
1306 	if (ret) {
1307 		printk(KERN_ERR "JBD2: Error %d detected when updating "
1308 		       "journal superblock for %s.\n", ret,
1309 		       journal->j_devname);
1310 	}
1311 }
1312 
1313 /**
1314  * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1315  * @journal: The journal to update.
1316  * @tail_tid: TID of the new transaction at the tail of the log
1317  * @tail_block: The first block of the transaction at the tail of the log
1318  * @write_op: With which operation should we write the journal sb
1319  *
1320  * Update a journal's superblock information about log tail and write it to
1321  * disk, waiting for the IO to complete.
1322  */
1323 void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1324 				     unsigned long tail_block, int write_op)
1325 {
1326 	journal_superblock_t *sb = journal->j_superblock;
1327 
1328 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1329 	jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1330 		  tail_block, tail_tid);
1331 
1332 	sb->s_sequence = cpu_to_be32(tail_tid);
1333 	sb->s_start    = cpu_to_be32(tail_block);
1334 
1335 	jbd2_write_superblock(journal, write_op);
1336 
1337 	/* Log is no longer empty */
1338 	write_lock(&journal->j_state_lock);
1339 	WARN_ON(!sb->s_sequence);
1340 	journal->j_flags &= ~JBD2_FLUSHED;
1341 	write_unlock(&journal->j_state_lock);
1342 }
1343 
1344 /**
1345  * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1346  * @journal: The journal to update.
1347  *
1348  * Update a journal's dynamic superblock fields to show that journal is empty.
1349  * Write updated superblock to disk waiting for IO to complete.
1350  */
1351 static void jbd2_mark_journal_empty(journal_t *journal)
1352 {
1353 	journal_superblock_t *sb = journal->j_superblock;
1354 
1355 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1356 	read_lock(&journal->j_state_lock);
1357 	jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1358 		  journal->j_tail_sequence);
1359 
1360 	sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1361 	sb->s_start    = cpu_to_be32(0);
1362 	read_unlock(&journal->j_state_lock);
1363 
1364 	jbd2_write_superblock(journal, WRITE_FUA);
1365 
1366 	/* Log is no longer empty */
1367 	write_lock(&journal->j_state_lock);
1368 	journal->j_flags |= JBD2_FLUSHED;
1369 	write_unlock(&journal->j_state_lock);
1370 }
1371 
1372 
1373 /**
1374  * jbd2_journal_update_sb_errno() - Update error in the journal.
1375  * @journal: The journal to update.
1376  *
1377  * Update a journal's errno.  Write updated superblock to disk waiting for IO
1378  * to complete.
1379  */
1380 void jbd2_journal_update_sb_errno(journal_t *journal)
1381 {
1382 	journal_superblock_t *sb = journal->j_superblock;
1383 
1384 	read_lock(&journal->j_state_lock);
1385 	jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1386 		  journal->j_errno);
1387 	sb->s_errno    = cpu_to_be32(journal->j_errno);
1388 	jbd2_superblock_csum_set(journal, sb);
1389 	read_unlock(&journal->j_state_lock);
1390 
1391 	jbd2_write_superblock(journal, WRITE_SYNC);
1392 }
1393 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1394 
1395 /*
1396  * Read the superblock for a given journal, performing initial
1397  * validation of the format.
1398  */
1399 static int journal_get_superblock(journal_t *journal)
1400 {
1401 	struct buffer_head *bh;
1402 	journal_superblock_t *sb;
1403 	int err = -EIO;
1404 
1405 	bh = journal->j_sb_buffer;
1406 
1407 	J_ASSERT(bh != NULL);
1408 	if (!buffer_uptodate(bh)) {
1409 		ll_rw_block(READ, 1, &bh);
1410 		wait_on_buffer(bh);
1411 		if (!buffer_uptodate(bh)) {
1412 			printk(KERN_ERR
1413 				"JBD2: IO error reading journal superblock\n");
1414 			goto out;
1415 		}
1416 	}
1417 
1418 	if (buffer_verified(bh))
1419 		return 0;
1420 
1421 	sb = journal->j_superblock;
1422 
1423 	err = -EINVAL;
1424 
1425 	if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1426 	    sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1427 		printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1428 		goto out;
1429 	}
1430 
1431 	switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1432 	case JBD2_SUPERBLOCK_V1:
1433 		journal->j_format_version = 1;
1434 		break;
1435 	case JBD2_SUPERBLOCK_V2:
1436 		journal->j_format_version = 2;
1437 		break;
1438 	default:
1439 		printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1440 		goto out;
1441 	}
1442 
1443 	if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1444 		journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1445 	else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1446 		printk(KERN_WARNING "JBD2: journal file too short\n");
1447 		goto out;
1448 	}
1449 
1450 	if (be32_to_cpu(sb->s_first) == 0 ||
1451 	    be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1452 		printk(KERN_WARNING
1453 			"JBD2: Invalid start block of journal: %u\n",
1454 			be32_to_cpu(sb->s_first));
1455 		goto out;
1456 	}
1457 
1458 	if (JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM) &&
1459 	    JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1460 		/* Can't have checksum v1 and v2 on at the same time! */
1461 		printk(KERN_ERR "JBD: Can't enable checksumming v1 and v2 "
1462 		       "at the same time!\n");
1463 		goto out;
1464 	}
1465 
1466 	if (!jbd2_verify_csum_type(journal, sb)) {
1467 		printk(KERN_ERR "JBD: Unknown checksum type\n");
1468 		goto out;
1469 	}
1470 
1471 	/* Load the checksum driver */
1472 	if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1473 		journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1474 		if (IS_ERR(journal->j_chksum_driver)) {
1475 			printk(KERN_ERR "JBD: Cannot load crc32c driver.\n");
1476 			err = PTR_ERR(journal->j_chksum_driver);
1477 			journal->j_chksum_driver = NULL;
1478 			goto out;
1479 		}
1480 	}
1481 
1482 	/* Check superblock checksum */
1483 	if (!jbd2_superblock_csum_verify(journal, sb)) {
1484 		printk(KERN_ERR "JBD: journal checksum error\n");
1485 		goto out;
1486 	}
1487 
1488 	/* Precompute checksum seed for all metadata */
1489 	if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
1490 		journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1491 						   sizeof(sb->s_uuid));
1492 
1493 	set_buffer_verified(bh);
1494 
1495 	return 0;
1496 
1497 out:
1498 	journal_fail_superblock(journal);
1499 	return err;
1500 }
1501 
1502 /*
1503  * Load the on-disk journal superblock and read the key fields into the
1504  * journal_t.
1505  */
1506 
1507 static int load_superblock(journal_t *journal)
1508 {
1509 	int err;
1510 	journal_superblock_t *sb;
1511 
1512 	err = journal_get_superblock(journal);
1513 	if (err)
1514 		return err;
1515 
1516 	sb = journal->j_superblock;
1517 
1518 	journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1519 	journal->j_tail = be32_to_cpu(sb->s_start);
1520 	journal->j_first = be32_to_cpu(sb->s_first);
1521 	journal->j_last = be32_to_cpu(sb->s_maxlen);
1522 	journal->j_errno = be32_to_cpu(sb->s_errno);
1523 
1524 	return 0;
1525 }
1526 
1527 
1528 /**
1529  * int jbd2_journal_load() - Read journal from disk.
1530  * @journal: Journal to act on.
1531  *
1532  * Given a journal_t structure which tells us which disk blocks contain
1533  * a journal, read the journal from disk to initialise the in-memory
1534  * structures.
1535  */
1536 int jbd2_journal_load(journal_t *journal)
1537 {
1538 	int err;
1539 	journal_superblock_t *sb;
1540 
1541 	err = load_superblock(journal);
1542 	if (err)
1543 		return err;
1544 
1545 	sb = journal->j_superblock;
1546 	/* If this is a V2 superblock, then we have to check the
1547 	 * features flags on it. */
1548 
1549 	if (journal->j_format_version >= 2) {
1550 		if ((sb->s_feature_ro_compat &
1551 		     ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1552 		    (sb->s_feature_incompat &
1553 		     ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1554 			printk(KERN_WARNING
1555 				"JBD2: Unrecognised features on journal\n");
1556 			return -EINVAL;
1557 		}
1558 	}
1559 
1560 	/*
1561 	 * Create a slab for this blocksize
1562 	 */
1563 	err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1564 	if (err)
1565 		return err;
1566 
1567 	/* Let the recovery code check whether it needs to recover any
1568 	 * data from the journal. */
1569 	if (jbd2_journal_recover(journal))
1570 		goto recovery_error;
1571 
1572 	if (journal->j_failed_commit) {
1573 		printk(KERN_ERR "JBD2: journal transaction %u on %s "
1574 		       "is corrupt.\n", journal->j_failed_commit,
1575 		       journal->j_devname);
1576 		return -EIO;
1577 	}
1578 
1579 	/* OK, we've finished with the dynamic journal bits:
1580 	 * reinitialise the dynamic contents of the superblock in memory
1581 	 * and reset them on disk. */
1582 	if (journal_reset(journal))
1583 		goto recovery_error;
1584 
1585 	journal->j_flags &= ~JBD2_ABORT;
1586 	journal->j_flags |= JBD2_LOADED;
1587 	return 0;
1588 
1589 recovery_error:
1590 	printk(KERN_WARNING "JBD2: recovery failed\n");
1591 	return -EIO;
1592 }
1593 
1594 /**
1595  * void jbd2_journal_destroy() - Release a journal_t structure.
1596  * @journal: Journal to act on.
1597  *
1598  * Release a journal_t structure once it is no longer in use by the
1599  * journaled object.
1600  * Return <0 if we couldn't clean up the journal.
1601  */
1602 int jbd2_journal_destroy(journal_t *journal)
1603 {
1604 	int err = 0;
1605 
1606 	/* Wait for the commit thread to wake up and die. */
1607 	journal_kill_thread(journal);
1608 
1609 	/* Force a final log commit */
1610 	if (journal->j_running_transaction)
1611 		jbd2_journal_commit_transaction(journal);
1612 
1613 	/* Force any old transactions to disk */
1614 
1615 	/* Totally anal locking here... */
1616 	spin_lock(&journal->j_list_lock);
1617 	while (journal->j_checkpoint_transactions != NULL) {
1618 		spin_unlock(&journal->j_list_lock);
1619 		mutex_lock(&journal->j_checkpoint_mutex);
1620 		jbd2_log_do_checkpoint(journal);
1621 		mutex_unlock(&journal->j_checkpoint_mutex);
1622 		spin_lock(&journal->j_list_lock);
1623 	}
1624 
1625 	J_ASSERT(journal->j_running_transaction == NULL);
1626 	J_ASSERT(journal->j_committing_transaction == NULL);
1627 	J_ASSERT(journal->j_checkpoint_transactions == NULL);
1628 	spin_unlock(&journal->j_list_lock);
1629 
1630 	if (journal->j_sb_buffer) {
1631 		if (!is_journal_aborted(journal)) {
1632 			mutex_lock(&journal->j_checkpoint_mutex);
1633 			jbd2_mark_journal_empty(journal);
1634 			mutex_unlock(&journal->j_checkpoint_mutex);
1635 		} else
1636 			err = -EIO;
1637 		brelse(journal->j_sb_buffer);
1638 	}
1639 
1640 	if (journal->j_proc_entry)
1641 		jbd2_stats_proc_exit(journal);
1642 	if (journal->j_inode)
1643 		iput(journal->j_inode);
1644 	if (journal->j_revoke)
1645 		jbd2_journal_destroy_revoke(journal);
1646 	if (journal->j_chksum_driver)
1647 		crypto_free_shash(journal->j_chksum_driver);
1648 	kfree(journal->j_wbuf);
1649 	kfree(journal);
1650 
1651 	return err;
1652 }
1653 
1654 
1655 /**
1656  *int jbd2_journal_check_used_features () - Check if features specified are used.
1657  * @journal: Journal to check.
1658  * @compat: bitmask of compatible features
1659  * @ro: bitmask of features that force read-only mount
1660  * @incompat: bitmask of incompatible features
1661  *
1662  * Check whether the journal uses all of a given set of
1663  * features.  Return true (non-zero) if it does.
1664  **/
1665 
1666 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1667 				 unsigned long ro, unsigned long incompat)
1668 {
1669 	journal_superblock_t *sb;
1670 
1671 	if (!compat && !ro && !incompat)
1672 		return 1;
1673 	/* Load journal superblock if it is not loaded yet. */
1674 	if (journal->j_format_version == 0 &&
1675 	    journal_get_superblock(journal) != 0)
1676 		return 0;
1677 	if (journal->j_format_version == 1)
1678 		return 0;
1679 
1680 	sb = journal->j_superblock;
1681 
1682 	if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1683 	    ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1684 	    ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1685 		return 1;
1686 
1687 	return 0;
1688 }
1689 
1690 /**
1691  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1692  * @journal: Journal to check.
1693  * @compat: bitmask of compatible features
1694  * @ro: bitmask of features that force read-only mount
1695  * @incompat: bitmask of incompatible features
1696  *
1697  * Check whether the journaling code supports the use of
1698  * all of a given set of features on this journal.  Return true
1699  * (non-zero) if it can. */
1700 
1701 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1702 				      unsigned long ro, unsigned long incompat)
1703 {
1704 	if (!compat && !ro && !incompat)
1705 		return 1;
1706 
1707 	/* We can support any known requested features iff the
1708 	 * superblock is in version 2.  Otherwise we fail to support any
1709 	 * extended sb features. */
1710 
1711 	if (journal->j_format_version != 2)
1712 		return 0;
1713 
1714 	if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1715 	    (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1716 	    (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1717 		return 1;
1718 
1719 	return 0;
1720 }
1721 
1722 /**
1723  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1724  * @journal: Journal to act on.
1725  * @compat: bitmask of compatible features
1726  * @ro: bitmask of features that force read-only mount
1727  * @incompat: bitmask of incompatible features
1728  *
1729  * Mark a given journal feature as present on the
1730  * superblock.  Returns true if the requested features could be set.
1731  *
1732  */
1733 
1734 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1735 			  unsigned long ro, unsigned long incompat)
1736 {
1737 #define INCOMPAT_FEATURE_ON(f) \
1738 		((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1739 #define COMPAT_FEATURE_ON(f) \
1740 		((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1741 	journal_superblock_t *sb;
1742 
1743 	if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1744 		return 1;
1745 
1746 	if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1747 		return 0;
1748 
1749 	/* Asking for checksumming v2 and v1?  Only give them v2. */
1750 	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2 &&
1751 	    compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1752 		compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1753 
1754 	jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1755 		  compat, ro, incompat);
1756 
1757 	sb = journal->j_superblock;
1758 
1759 	/* If enabling v2 checksums, update superblock */
1760 	if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V2)) {
1761 		sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1762 		sb->s_feature_compat &=
1763 			~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1764 
1765 		/* Load the checksum driver */
1766 		if (journal->j_chksum_driver == NULL) {
1767 			journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1768 								      0, 0);
1769 			if (IS_ERR(journal->j_chksum_driver)) {
1770 				printk(KERN_ERR "JBD: Cannot load crc32c "
1771 				       "driver.\n");
1772 				journal->j_chksum_driver = NULL;
1773 				return 0;
1774 			}
1775 		}
1776 
1777 		/* Precompute checksum seed for all metadata */
1778 		if (JBD2_HAS_INCOMPAT_FEATURE(journal,
1779 					      JBD2_FEATURE_INCOMPAT_CSUM_V2))
1780 			journal->j_csum_seed = jbd2_chksum(journal, ~0,
1781 							   sb->s_uuid,
1782 							   sizeof(sb->s_uuid));
1783 	}
1784 
1785 	/* If enabling v1 checksums, downgrade superblock */
1786 	if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1787 		sb->s_feature_incompat &=
1788 			~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2);
1789 
1790 	sb->s_feature_compat    |= cpu_to_be32(compat);
1791 	sb->s_feature_ro_compat |= cpu_to_be32(ro);
1792 	sb->s_feature_incompat  |= cpu_to_be32(incompat);
1793 
1794 	return 1;
1795 #undef COMPAT_FEATURE_ON
1796 #undef INCOMPAT_FEATURE_ON
1797 }
1798 
1799 /*
1800  * jbd2_journal_clear_features () - Clear a given journal feature in the
1801  * 				    superblock
1802  * @journal: Journal to act on.
1803  * @compat: bitmask of compatible features
1804  * @ro: bitmask of features that force read-only mount
1805  * @incompat: bitmask of incompatible features
1806  *
1807  * Clear a given journal feature as present on the
1808  * superblock.
1809  */
1810 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1811 				unsigned long ro, unsigned long incompat)
1812 {
1813 	journal_superblock_t *sb;
1814 
1815 	jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1816 		  compat, ro, incompat);
1817 
1818 	sb = journal->j_superblock;
1819 
1820 	sb->s_feature_compat    &= ~cpu_to_be32(compat);
1821 	sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1822 	sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1823 }
1824 EXPORT_SYMBOL(jbd2_journal_clear_features);
1825 
1826 /**
1827  * int jbd2_journal_flush () - Flush journal
1828  * @journal: Journal to act on.
1829  *
1830  * Flush all data for a given journal to disk and empty the journal.
1831  * Filesystems can use this when remounting readonly to ensure that
1832  * recovery does not need to happen on remount.
1833  */
1834 
1835 int jbd2_journal_flush(journal_t *journal)
1836 {
1837 	int err = 0;
1838 	transaction_t *transaction = NULL;
1839 
1840 	write_lock(&journal->j_state_lock);
1841 
1842 	/* Force everything buffered to the log... */
1843 	if (journal->j_running_transaction) {
1844 		transaction = journal->j_running_transaction;
1845 		__jbd2_log_start_commit(journal, transaction->t_tid);
1846 	} else if (journal->j_committing_transaction)
1847 		transaction = journal->j_committing_transaction;
1848 
1849 	/* Wait for the log commit to complete... */
1850 	if (transaction) {
1851 		tid_t tid = transaction->t_tid;
1852 
1853 		write_unlock(&journal->j_state_lock);
1854 		jbd2_log_wait_commit(journal, tid);
1855 	} else {
1856 		write_unlock(&journal->j_state_lock);
1857 	}
1858 
1859 	/* ...and flush everything in the log out to disk. */
1860 	spin_lock(&journal->j_list_lock);
1861 	while (!err && journal->j_checkpoint_transactions != NULL) {
1862 		spin_unlock(&journal->j_list_lock);
1863 		mutex_lock(&journal->j_checkpoint_mutex);
1864 		err = jbd2_log_do_checkpoint(journal);
1865 		mutex_unlock(&journal->j_checkpoint_mutex);
1866 		spin_lock(&journal->j_list_lock);
1867 	}
1868 	spin_unlock(&journal->j_list_lock);
1869 
1870 	if (is_journal_aborted(journal))
1871 		return -EIO;
1872 
1873 	mutex_lock(&journal->j_checkpoint_mutex);
1874 	jbd2_cleanup_journal_tail(journal);
1875 
1876 	/* Finally, mark the journal as really needing no recovery.
1877 	 * This sets s_start==0 in the underlying superblock, which is
1878 	 * the magic code for a fully-recovered superblock.  Any future
1879 	 * commits of data to the journal will restore the current
1880 	 * s_start value. */
1881 	jbd2_mark_journal_empty(journal);
1882 	mutex_unlock(&journal->j_checkpoint_mutex);
1883 	write_lock(&journal->j_state_lock);
1884 	J_ASSERT(!journal->j_running_transaction);
1885 	J_ASSERT(!journal->j_committing_transaction);
1886 	J_ASSERT(!journal->j_checkpoint_transactions);
1887 	J_ASSERT(journal->j_head == journal->j_tail);
1888 	J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1889 	write_unlock(&journal->j_state_lock);
1890 	return 0;
1891 }
1892 
1893 /**
1894  * int jbd2_journal_wipe() - Wipe journal contents
1895  * @journal: Journal to act on.
1896  * @write: flag (see below)
1897  *
1898  * Wipe out all of the contents of a journal, safely.  This will produce
1899  * a warning if the journal contains any valid recovery information.
1900  * Must be called between journal_init_*() and jbd2_journal_load().
1901  *
1902  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1903  * we merely suppress recovery.
1904  */
1905 
1906 int jbd2_journal_wipe(journal_t *journal, int write)
1907 {
1908 	int err = 0;
1909 
1910 	J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1911 
1912 	err = load_superblock(journal);
1913 	if (err)
1914 		return err;
1915 
1916 	if (!journal->j_tail)
1917 		goto no_recovery;
1918 
1919 	printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
1920 		write ? "Clearing" : "Ignoring");
1921 
1922 	err = jbd2_journal_skip_recovery(journal);
1923 	if (write) {
1924 		/* Lock to make assertions happy... */
1925 		mutex_lock(&journal->j_checkpoint_mutex);
1926 		jbd2_mark_journal_empty(journal);
1927 		mutex_unlock(&journal->j_checkpoint_mutex);
1928 	}
1929 
1930  no_recovery:
1931 	return err;
1932 }
1933 
1934 /*
1935  * Journal abort has very specific semantics, which we describe
1936  * for journal abort.
1937  *
1938  * Two internal functions, which provide abort to the jbd layer
1939  * itself are here.
1940  */
1941 
1942 /*
1943  * Quick version for internal journal use (doesn't lock the journal).
1944  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1945  * and don't attempt to make any other journal updates.
1946  */
1947 void __jbd2_journal_abort_hard(journal_t *journal)
1948 {
1949 	transaction_t *transaction;
1950 
1951 	if (journal->j_flags & JBD2_ABORT)
1952 		return;
1953 
1954 	printk(KERN_ERR "Aborting journal on device %s.\n",
1955 	       journal->j_devname);
1956 
1957 	write_lock(&journal->j_state_lock);
1958 	journal->j_flags |= JBD2_ABORT;
1959 	transaction = journal->j_running_transaction;
1960 	if (transaction)
1961 		__jbd2_log_start_commit(journal, transaction->t_tid);
1962 	write_unlock(&journal->j_state_lock);
1963 }
1964 
1965 /* Soft abort: record the abort error status in the journal superblock,
1966  * but don't do any other IO. */
1967 static void __journal_abort_soft (journal_t *journal, int errno)
1968 {
1969 	if (journal->j_flags & JBD2_ABORT)
1970 		return;
1971 
1972 	if (!journal->j_errno)
1973 		journal->j_errno = errno;
1974 
1975 	__jbd2_journal_abort_hard(journal);
1976 
1977 	if (errno)
1978 		jbd2_journal_update_sb_errno(journal);
1979 }
1980 
1981 /**
1982  * void jbd2_journal_abort () - Shutdown the journal immediately.
1983  * @journal: the journal to shutdown.
1984  * @errno:   an error number to record in the journal indicating
1985  *           the reason for the shutdown.
1986  *
1987  * Perform a complete, immediate shutdown of the ENTIRE
1988  * journal (not of a single transaction).  This operation cannot be
1989  * undone without closing and reopening the journal.
1990  *
1991  * The jbd2_journal_abort function is intended to support higher level error
1992  * recovery mechanisms such as the ext2/ext3 remount-readonly error
1993  * mode.
1994  *
1995  * Journal abort has very specific semantics.  Any existing dirty,
1996  * unjournaled buffers in the main filesystem will still be written to
1997  * disk by bdflush, but the journaling mechanism will be suspended
1998  * immediately and no further transaction commits will be honoured.
1999  *
2000  * Any dirty, journaled buffers will be written back to disk without
2001  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2002  * filesystem, but we _do_ attempt to leave as much data as possible
2003  * behind for fsck to use for cleanup.
2004  *
2005  * Any attempt to get a new transaction handle on a journal which is in
2006  * ABORT state will just result in an -EROFS error return.  A
2007  * jbd2_journal_stop on an existing handle will return -EIO if we have
2008  * entered abort state during the update.
2009  *
2010  * Recursive transactions are not disturbed by journal abort until the
2011  * final jbd2_journal_stop, which will receive the -EIO error.
2012  *
2013  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2014  * which will be recorded (if possible) in the journal superblock.  This
2015  * allows a client to record failure conditions in the middle of a
2016  * transaction without having to complete the transaction to record the
2017  * failure to disk.  ext3_error, for example, now uses this
2018  * functionality.
2019  *
2020  * Errors which originate from within the journaling layer will NOT
2021  * supply an errno; a null errno implies that absolutely no further
2022  * writes are done to the journal (unless there are any already in
2023  * progress).
2024  *
2025  */
2026 
2027 void jbd2_journal_abort(journal_t *journal, int errno)
2028 {
2029 	__journal_abort_soft(journal, errno);
2030 }
2031 
2032 /**
2033  * int jbd2_journal_errno () - returns the journal's error state.
2034  * @journal: journal to examine.
2035  *
2036  * This is the errno number set with jbd2_journal_abort(), the last
2037  * time the journal was mounted - if the journal was stopped
2038  * without calling abort this will be 0.
2039  *
2040  * If the journal has been aborted on this mount time -EROFS will
2041  * be returned.
2042  */
2043 int jbd2_journal_errno(journal_t *journal)
2044 {
2045 	int err;
2046 
2047 	read_lock(&journal->j_state_lock);
2048 	if (journal->j_flags & JBD2_ABORT)
2049 		err = -EROFS;
2050 	else
2051 		err = journal->j_errno;
2052 	read_unlock(&journal->j_state_lock);
2053 	return err;
2054 }
2055 
2056 /**
2057  * int jbd2_journal_clear_err () - clears the journal's error state
2058  * @journal: journal to act on.
2059  *
2060  * An error must be cleared or acked to take a FS out of readonly
2061  * mode.
2062  */
2063 int jbd2_journal_clear_err(journal_t *journal)
2064 {
2065 	int err = 0;
2066 
2067 	write_lock(&journal->j_state_lock);
2068 	if (journal->j_flags & JBD2_ABORT)
2069 		err = -EROFS;
2070 	else
2071 		journal->j_errno = 0;
2072 	write_unlock(&journal->j_state_lock);
2073 	return err;
2074 }
2075 
2076 /**
2077  * void jbd2_journal_ack_err() - Ack journal err.
2078  * @journal: journal to act on.
2079  *
2080  * An error must be cleared or acked to take a FS out of readonly
2081  * mode.
2082  */
2083 void jbd2_journal_ack_err(journal_t *journal)
2084 {
2085 	write_lock(&journal->j_state_lock);
2086 	if (journal->j_errno)
2087 		journal->j_flags |= JBD2_ACK_ERR;
2088 	write_unlock(&journal->j_state_lock);
2089 }
2090 
2091 int jbd2_journal_blocks_per_page(struct inode *inode)
2092 {
2093 	return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2094 }
2095 
2096 /*
2097  * helper functions to deal with 32 or 64bit block numbers.
2098  */
2099 size_t journal_tag_bytes(journal_t *journal)
2100 {
2101 	journal_block_tag_t tag;
2102 	size_t x = 0;
2103 
2104 	if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
2105 		x += sizeof(tag.t_checksum);
2106 
2107 	if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
2108 		return x + JBD2_TAG_SIZE64;
2109 	else
2110 		return x + JBD2_TAG_SIZE32;
2111 }
2112 
2113 /*
2114  * JBD memory management
2115  *
2116  * These functions are used to allocate block-sized chunks of memory
2117  * used for making copies of buffer_head data.  Very often it will be
2118  * page-sized chunks of data, but sometimes it will be in
2119  * sub-page-size chunks.  (For example, 16k pages on Power systems
2120  * with a 4k block file system.)  For blocks smaller than a page, we
2121  * use a SLAB allocator.  There are slab caches for each block size,
2122  * which are allocated at mount time, if necessary, and we only free
2123  * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2124  * this reason we don't need to a mutex to protect access to
2125  * jbd2_slab[] allocating or releasing memory; only in
2126  * jbd2_journal_create_slab().
2127  */
2128 #define JBD2_MAX_SLABS 8
2129 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2130 
2131 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2132 	"jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2133 	"jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2134 };
2135 
2136 
2137 static void jbd2_journal_destroy_slabs(void)
2138 {
2139 	int i;
2140 
2141 	for (i = 0; i < JBD2_MAX_SLABS; i++) {
2142 		if (jbd2_slab[i])
2143 			kmem_cache_destroy(jbd2_slab[i]);
2144 		jbd2_slab[i] = NULL;
2145 	}
2146 }
2147 
2148 static int jbd2_journal_create_slab(size_t size)
2149 {
2150 	static DEFINE_MUTEX(jbd2_slab_create_mutex);
2151 	int i = order_base_2(size) - 10;
2152 	size_t slab_size;
2153 
2154 	if (size == PAGE_SIZE)
2155 		return 0;
2156 
2157 	if (i >= JBD2_MAX_SLABS)
2158 		return -EINVAL;
2159 
2160 	if (unlikely(i < 0))
2161 		i = 0;
2162 	mutex_lock(&jbd2_slab_create_mutex);
2163 	if (jbd2_slab[i]) {
2164 		mutex_unlock(&jbd2_slab_create_mutex);
2165 		return 0;	/* Already created */
2166 	}
2167 
2168 	slab_size = 1 << (i+10);
2169 	jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2170 					 slab_size, 0, NULL);
2171 	mutex_unlock(&jbd2_slab_create_mutex);
2172 	if (!jbd2_slab[i]) {
2173 		printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2174 		return -ENOMEM;
2175 	}
2176 	return 0;
2177 }
2178 
2179 static struct kmem_cache *get_slab(size_t size)
2180 {
2181 	int i = order_base_2(size) - 10;
2182 
2183 	BUG_ON(i >= JBD2_MAX_SLABS);
2184 	if (unlikely(i < 0))
2185 		i = 0;
2186 	BUG_ON(jbd2_slab[i] == NULL);
2187 	return jbd2_slab[i];
2188 }
2189 
2190 void *jbd2_alloc(size_t size, gfp_t flags)
2191 {
2192 	void *ptr;
2193 
2194 	BUG_ON(size & (size-1)); /* Must be a power of 2 */
2195 
2196 	flags |= __GFP_REPEAT;
2197 	if (size == PAGE_SIZE)
2198 		ptr = (void *)__get_free_pages(flags, 0);
2199 	else if (size > PAGE_SIZE) {
2200 		int order = get_order(size);
2201 
2202 		if (order < 3)
2203 			ptr = (void *)__get_free_pages(flags, order);
2204 		else
2205 			ptr = vmalloc(size);
2206 	} else
2207 		ptr = kmem_cache_alloc(get_slab(size), flags);
2208 
2209 	/* Check alignment; SLUB has gotten this wrong in the past,
2210 	 * and this can lead to user data corruption! */
2211 	BUG_ON(((unsigned long) ptr) & (size-1));
2212 
2213 	return ptr;
2214 }
2215 
2216 void jbd2_free(void *ptr, size_t size)
2217 {
2218 	if (size == PAGE_SIZE) {
2219 		free_pages((unsigned long)ptr, 0);
2220 		return;
2221 	}
2222 	if (size > PAGE_SIZE) {
2223 		int order = get_order(size);
2224 
2225 		if (order < 3)
2226 			free_pages((unsigned long)ptr, order);
2227 		else
2228 			vfree(ptr);
2229 		return;
2230 	}
2231 	kmem_cache_free(get_slab(size), ptr);
2232 };
2233 
2234 /*
2235  * Journal_head storage management
2236  */
2237 static struct kmem_cache *jbd2_journal_head_cache;
2238 #ifdef CONFIG_JBD2_DEBUG
2239 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2240 #endif
2241 
2242 static int jbd2_journal_init_journal_head_cache(void)
2243 {
2244 	int retval;
2245 
2246 	J_ASSERT(jbd2_journal_head_cache == NULL);
2247 	jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2248 				sizeof(struct journal_head),
2249 				0,		/* offset */
2250 				SLAB_TEMPORARY,	/* flags */
2251 				NULL);		/* ctor */
2252 	retval = 0;
2253 	if (!jbd2_journal_head_cache) {
2254 		retval = -ENOMEM;
2255 		printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2256 	}
2257 	return retval;
2258 }
2259 
2260 static void jbd2_journal_destroy_journal_head_cache(void)
2261 {
2262 	if (jbd2_journal_head_cache) {
2263 		kmem_cache_destroy(jbd2_journal_head_cache);
2264 		jbd2_journal_head_cache = NULL;
2265 	}
2266 }
2267 
2268 /*
2269  * journal_head splicing and dicing
2270  */
2271 static struct journal_head *journal_alloc_journal_head(void)
2272 {
2273 	struct journal_head *ret;
2274 
2275 #ifdef CONFIG_JBD2_DEBUG
2276 	atomic_inc(&nr_journal_heads);
2277 #endif
2278 	ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2279 	if (!ret) {
2280 		jbd_debug(1, "out of memory for journal_head\n");
2281 		pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2282 		while (!ret) {
2283 			yield();
2284 			ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
2285 		}
2286 	}
2287 	return ret;
2288 }
2289 
2290 static void journal_free_journal_head(struct journal_head *jh)
2291 {
2292 #ifdef CONFIG_JBD2_DEBUG
2293 	atomic_dec(&nr_journal_heads);
2294 	memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2295 #endif
2296 	kmem_cache_free(jbd2_journal_head_cache, jh);
2297 }
2298 
2299 /*
2300  * A journal_head is attached to a buffer_head whenever JBD has an
2301  * interest in the buffer.
2302  *
2303  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2304  * is set.  This bit is tested in core kernel code where we need to take
2305  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2306  * there.
2307  *
2308  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2309  *
2310  * When a buffer has its BH_JBD bit set it is immune from being released by
2311  * core kernel code, mainly via ->b_count.
2312  *
2313  * A journal_head is detached from its buffer_head when the journal_head's
2314  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2315  * transaction (b_cp_transaction) hold their references to b_jcount.
2316  *
2317  * Various places in the kernel want to attach a journal_head to a buffer_head
2318  * _before_ attaching the journal_head to a transaction.  To protect the
2319  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2320  * journal_head's b_jcount refcount by one.  The caller must call
2321  * jbd2_journal_put_journal_head() to undo this.
2322  *
2323  * So the typical usage would be:
2324  *
2325  *	(Attach a journal_head if needed.  Increments b_jcount)
2326  *	struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2327  *	...
2328  *      (Get another reference for transaction)
2329  *	jbd2_journal_grab_journal_head(bh);
2330  *	jh->b_transaction = xxx;
2331  *	(Put original reference)
2332  *	jbd2_journal_put_journal_head(jh);
2333  */
2334 
2335 /*
2336  * Give a buffer_head a journal_head.
2337  *
2338  * May sleep.
2339  */
2340 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2341 {
2342 	struct journal_head *jh;
2343 	struct journal_head *new_jh = NULL;
2344 
2345 repeat:
2346 	if (!buffer_jbd(bh)) {
2347 		new_jh = journal_alloc_journal_head();
2348 		memset(new_jh, 0, sizeof(*new_jh));
2349 	}
2350 
2351 	jbd_lock_bh_journal_head(bh);
2352 	if (buffer_jbd(bh)) {
2353 		jh = bh2jh(bh);
2354 	} else {
2355 		J_ASSERT_BH(bh,
2356 			(atomic_read(&bh->b_count) > 0) ||
2357 			(bh->b_page && bh->b_page->mapping));
2358 
2359 		if (!new_jh) {
2360 			jbd_unlock_bh_journal_head(bh);
2361 			goto repeat;
2362 		}
2363 
2364 		jh = new_jh;
2365 		new_jh = NULL;		/* We consumed it */
2366 		set_buffer_jbd(bh);
2367 		bh->b_private = jh;
2368 		jh->b_bh = bh;
2369 		get_bh(bh);
2370 		BUFFER_TRACE(bh, "added journal_head");
2371 	}
2372 	jh->b_jcount++;
2373 	jbd_unlock_bh_journal_head(bh);
2374 	if (new_jh)
2375 		journal_free_journal_head(new_jh);
2376 	return bh->b_private;
2377 }
2378 
2379 /*
2380  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2381  * having a journal_head, return NULL
2382  */
2383 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2384 {
2385 	struct journal_head *jh = NULL;
2386 
2387 	jbd_lock_bh_journal_head(bh);
2388 	if (buffer_jbd(bh)) {
2389 		jh = bh2jh(bh);
2390 		jh->b_jcount++;
2391 	}
2392 	jbd_unlock_bh_journal_head(bh);
2393 	return jh;
2394 }
2395 
2396 static void __journal_remove_journal_head(struct buffer_head *bh)
2397 {
2398 	struct journal_head *jh = bh2jh(bh);
2399 
2400 	J_ASSERT_JH(jh, jh->b_jcount >= 0);
2401 	J_ASSERT_JH(jh, jh->b_transaction == NULL);
2402 	J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2403 	J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2404 	J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2405 	J_ASSERT_BH(bh, buffer_jbd(bh));
2406 	J_ASSERT_BH(bh, jh2bh(jh) == bh);
2407 	BUFFER_TRACE(bh, "remove journal_head");
2408 	if (jh->b_frozen_data) {
2409 		printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2410 		jbd2_free(jh->b_frozen_data, bh->b_size);
2411 	}
2412 	if (jh->b_committed_data) {
2413 		printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2414 		jbd2_free(jh->b_committed_data, bh->b_size);
2415 	}
2416 	bh->b_private = NULL;
2417 	jh->b_bh = NULL;	/* debug, really */
2418 	clear_buffer_jbd(bh);
2419 	journal_free_journal_head(jh);
2420 }
2421 
2422 /*
2423  * Drop a reference on the passed journal_head.  If it fell to zero then
2424  * release the journal_head from the buffer_head.
2425  */
2426 void jbd2_journal_put_journal_head(struct journal_head *jh)
2427 {
2428 	struct buffer_head *bh = jh2bh(jh);
2429 
2430 	jbd_lock_bh_journal_head(bh);
2431 	J_ASSERT_JH(jh, jh->b_jcount > 0);
2432 	--jh->b_jcount;
2433 	if (!jh->b_jcount) {
2434 		__journal_remove_journal_head(bh);
2435 		jbd_unlock_bh_journal_head(bh);
2436 		__brelse(bh);
2437 	} else
2438 		jbd_unlock_bh_journal_head(bh);
2439 }
2440 
2441 /*
2442  * Initialize jbd inode head
2443  */
2444 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2445 {
2446 	jinode->i_transaction = NULL;
2447 	jinode->i_next_transaction = NULL;
2448 	jinode->i_vfs_inode = inode;
2449 	jinode->i_flags = 0;
2450 	INIT_LIST_HEAD(&jinode->i_list);
2451 }
2452 
2453 /*
2454  * Function to be called before we start removing inode from memory (i.e.,
2455  * clear_inode() is a fine place to be called from). It removes inode from
2456  * transaction's lists.
2457  */
2458 void jbd2_journal_release_jbd_inode(journal_t *journal,
2459 				    struct jbd2_inode *jinode)
2460 {
2461 	if (!journal)
2462 		return;
2463 restart:
2464 	spin_lock(&journal->j_list_lock);
2465 	/* Is commit writing out inode - we have to wait */
2466 	if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2467 		wait_queue_head_t *wq;
2468 		DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2469 		wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2470 		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2471 		spin_unlock(&journal->j_list_lock);
2472 		schedule();
2473 		finish_wait(wq, &wait.wait);
2474 		goto restart;
2475 	}
2476 
2477 	if (jinode->i_transaction) {
2478 		list_del(&jinode->i_list);
2479 		jinode->i_transaction = NULL;
2480 	}
2481 	spin_unlock(&journal->j_list_lock);
2482 }
2483 
2484 /*
2485  * debugfs tunables
2486  */
2487 #ifdef CONFIG_JBD2_DEBUG
2488 u8 jbd2_journal_enable_debug __read_mostly;
2489 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2490 
2491 #define JBD2_DEBUG_NAME "jbd2-debug"
2492 
2493 static struct dentry *jbd2_debugfs_dir;
2494 static struct dentry *jbd2_debug;
2495 
2496 static void __init jbd2_create_debugfs_entry(void)
2497 {
2498 	jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2499 	if (jbd2_debugfs_dir)
2500 		jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME,
2501 					       S_IRUGO | S_IWUSR,
2502 					       jbd2_debugfs_dir,
2503 					       &jbd2_journal_enable_debug);
2504 }
2505 
2506 static void __exit jbd2_remove_debugfs_entry(void)
2507 {
2508 	debugfs_remove(jbd2_debug);
2509 	debugfs_remove(jbd2_debugfs_dir);
2510 }
2511 
2512 #else
2513 
2514 static void __init jbd2_create_debugfs_entry(void)
2515 {
2516 }
2517 
2518 static void __exit jbd2_remove_debugfs_entry(void)
2519 {
2520 }
2521 
2522 #endif
2523 
2524 #ifdef CONFIG_PROC_FS
2525 
2526 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2527 
2528 static void __init jbd2_create_jbd_stats_proc_entry(void)
2529 {
2530 	proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2531 }
2532 
2533 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2534 {
2535 	if (proc_jbd2_stats)
2536 		remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2537 }
2538 
2539 #else
2540 
2541 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2542 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2543 
2544 #endif
2545 
2546 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2547 
2548 static int __init jbd2_journal_init_handle_cache(void)
2549 {
2550 	jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2551 	if (jbd2_handle_cache == NULL) {
2552 		printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2553 		return -ENOMEM;
2554 	}
2555 	jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2556 	if (jbd2_inode_cache == NULL) {
2557 		printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2558 		kmem_cache_destroy(jbd2_handle_cache);
2559 		return -ENOMEM;
2560 	}
2561 	return 0;
2562 }
2563 
2564 static void jbd2_journal_destroy_handle_cache(void)
2565 {
2566 	if (jbd2_handle_cache)
2567 		kmem_cache_destroy(jbd2_handle_cache);
2568 	if (jbd2_inode_cache)
2569 		kmem_cache_destroy(jbd2_inode_cache);
2570 
2571 }
2572 
2573 /*
2574  * Module startup and shutdown
2575  */
2576 
2577 static int __init journal_init_caches(void)
2578 {
2579 	int ret;
2580 
2581 	ret = jbd2_journal_init_revoke_caches();
2582 	if (ret == 0)
2583 		ret = jbd2_journal_init_journal_head_cache();
2584 	if (ret == 0)
2585 		ret = jbd2_journal_init_handle_cache();
2586 	if (ret == 0)
2587 		ret = jbd2_journal_init_transaction_cache();
2588 	return ret;
2589 }
2590 
2591 static void jbd2_journal_destroy_caches(void)
2592 {
2593 	jbd2_journal_destroy_revoke_caches();
2594 	jbd2_journal_destroy_journal_head_cache();
2595 	jbd2_journal_destroy_handle_cache();
2596 	jbd2_journal_destroy_transaction_cache();
2597 	jbd2_journal_destroy_slabs();
2598 }
2599 
2600 static int __init journal_init(void)
2601 {
2602 	int ret;
2603 
2604 	BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2605 
2606 	ret = journal_init_caches();
2607 	if (ret == 0) {
2608 		jbd2_create_debugfs_entry();
2609 		jbd2_create_jbd_stats_proc_entry();
2610 	} else {
2611 		jbd2_journal_destroy_caches();
2612 	}
2613 	return ret;
2614 }
2615 
2616 static void __exit journal_exit(void)
2617 {
2618 #ifdef CONFIG_JBD2_DEBUG
2619 	int n = atomic_read(&nr_journal_heads);
2620 	if (n)
2621 		printk(KERN_EMERG "JBD2: leaked %d journal_heads!\n", n);
2622 #endif
2623 	jbd2_remove_debugfs_entry();
2624 	jbd2_remove_jbd_stats_proc_entry();
2625 	jbd2_journal_destroy_caches();
2626 }
2627 
2628 MODULE_LICENSE("GPL");
2629 module_init(journal_init);
2630 module_exit(journal_exit);
2631 
2632