xref: /openbmc/linux/fs/jbd2/journal.c (revision 9b9c2cd4)
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/seq_file.h>
39 #include <linux/math64.h>
40 #include <linux/hash.h>
41 #include <linux/log2.h>
42 #include <linux/vmalloc.h>
43 #include <linux/backing-dev.h>
44 #include <linux/bitops.h>
45 #include <linux/ratelimit.h>
46 
47 #define CREATE_TRACE_POINTS
48 #include <trace/events/jbd2.h>
49 
50 #include <asm/uaccess.h>
51 #include <asm/page.h>
52 
53 #ifdef CONFIG_JBD2_DEBUG
54 ushort jbd2_journal_enable_debug __read_mostly;
55 EXPORT_SYMBOL(jbd2_journal_enable_debug);
56 
57 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
58 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
59 #endif
60 
61 EXPORT_SYMBOL(jbd2_journal_extend);
62 EXPORT_SYMBOL(jbd2_journal_stop);
63 EXPORT_SYMBOL(jbd2_journal_lock_updates);
64 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
65 EXPORT_SYMBOL(jbd2_journal_get_write_access);
66 EXPORT_SYMBOL(jbd2_journal_get_create_access);
67 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
68 EXPORT_SYMBOL(jbd2_journal_set_triggers);
69 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
70 EXPORT_SYMBOL(jbd2_journal_forget);
71 #if 0
72 EXPORT_SYMBOL(journal_sync_buffer);
73 #endif
74 EXPORT_SYMBOL(jbd2_journal_flush);
75 EXPORT_SYMBOL(jbd2_journal_revoke);
76 
77 EXPORT_SYMBOL(jbd2_journal_init_dev);
78 EXPORT_SYMBOL(jbd2_journal_init_inode);
79 EXPORT_SYMBOL(jbd2_journal_check_used_features);
80 EXPORT_SYMBOL(jbd2_journal_check_available_features);
81 EXPORT_SYMBOL(jbd2_journal_set_features);
82 EXPORT_SYMBOL(jbd2_journal_load);
83 EXPORT_SYMBOL(jbd2_journal_destroy);
84 EXPORT_SYMBOL(jbd2_journal_abort);
85 EXPORT_SYMBOL(jbd2_journal_errno);
86 EXPORT_SYMBOL(jbd2_journal_ack_err);
87 EXPORT_SYMBOL(jbd2_journal_clear_err);
88 EXPORT_SYMBOL(jbd2_log_wait_commit);
89 EXPORT_SYMBOL(jbd2_log_start_commit);
90 EXPORT_SYMBOL(jbd2_journal_start_commit);
91 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
92 EXPORT_SYMBOL(jbd2_journal_wipe);
93 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
94 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
95 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
96 EXPORT_SYMBOL(jbd2_journal_force_commit);
97 EXPORT_SYMBOL(jbd2_journal_file_inode);
98 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
99 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
100 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
101 EXPORT_SYMBOL(jbd2_inode_cache);
102 
103 static void __journal_abort_soft (journal_t *journal, int errno);
104 static int jbd2_journal_create_slab(size_t slab_size);
105 
106 #ifdef CONFIG_JBD2_DEBUG
107 void __jbd2_debug(int level, const char *file, const char *func,
108 		  unsigned int line, const char *fmt, ...)
109 {
110 	struct va_format vaf;
111 	va_list args;
112 
113 	if (level > jbd2_journal_enable_debug)
114 		return;
115 	va_start(args, fmt);
116 	vaf.fmt = fmt;
117 	vaf.va = &args;
118 	printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
119 	va_end(args);
120 }
121 EXPORT_SYMBOL(__jbd2_debug);
122 #endif
123 
124 /* Checksumming functions */
125 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
126 {
127 	if (!jbd2_journal_has_csum_v2or3_feature(j))
128 		return 1;
129 
130 	return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
131 }
132 
133 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
134 {
135 	__u32 csum;
136 	__be32 old_csum;
137 
138 	old_csum = sb->s_checksum;
139 	sb->s_checksum = 0;
140 	csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
141 	sb->s_checksum = old_csum;
142 
143 	return cpu_to_be32(csum);
144 }
145 
146 static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
147 {
148 	if (!jbd2_journal_has_csum_v2or3(j))
149 		return 1;
150 
151 	return sb->s_checksum == jbd2_superblock_csum(j, sb);
152 }
153 
154 static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
155 {
156 	if (!jbd2_journal_has_csum_v2or3(j))
157 		return;
158 
159 	sb->s_checksum = jbd2_superblock_csum(j, sb);
160 }
161 
162 /*
163  * Helper function used to manage commit timeouts
164  */
165 
166 static void commit_timeout(unsigned long __data)
167 {
168 	struct task_struct * p = (struct task_struct *) __data;
169 
170 	wake_up_process(p);
171 }
172 
173 /*
174  * kjournald2: The main thread function used to manage a logging device
175  * journal.
176  *
177  * This kernel thread is responsible for two things:
178  *
179  * 1) COMMIT:  Every so often we need to commit the current state of the
180  *    filesystem to disk.  The journal thread is responsible for writing
181  *    all of the metadata buffers to disk.
182  *
183  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
184  *    of the data in that part of the log has been rewritten elsewhere on
185  *    the disk.  Flushing these old buffers to reclaim space in the log is
186  *    known as checkpointing, and this thread is responsible for that job.
187  */
188 
189 static int kjournald2(void *arg)
190 {
191 	journal_t *journal = arg;
192 	transaction_t *transaction;
193 
194 	/*
195 	 * Set up an interval timer which can be used to trigger a commit wakeup
196 	 * after the commit interval expires
197 	 */
198 	setup_timer(&journal->j_commit_timer, commit_timeout,
199 			(unsigned long)current);
200 
201 	set_freezable();
202 
203 	/* Record that the journal thread is running */
204 	journal->j_task = current;
205 	wake_up(&journal->j_wait_done_commit);
206 
207 	/*
208 	 * And now, wait forever for commit wakeup events.
209 	 */
210 	write_lock(&journal->j_state_lock);
211 
212 loop:
213 	if (journal->j_flags & JBD2_UNMOUNT)
214 		goto end_loop;
215 
216 	jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
217 		journal->j_commit_sequence, journal->j_commit_request);
218 
219 	if (journal->j_commit_sequence != journal->j_commit_request) {
220 		jbd_debug(1, "OK, requests differ\n");
221 		write_unlock(&journal->j_state_lock);
222 		del_timer_sync(&journal->j_commit_timer);
223 		jbd2_journal_commit_transaction(journal);
224 		write_lock(&journal->j_state_lock);
225 		goto loop;
226 	}
227 
228 	wake_up(&journal->j_wait_done_commit);
229 	if (freezing(current)) {
230 		/*
231 		 * The simpler the better. Flushing journal isn't a
232 		 * good idea, because that depends on threads that may
233 		 * be already stopped.
234 		 */
235 		jbd_debug(1, "Now suspending kjournald2\n");
236 		write_unlock(&journal->j_state_lock);
237 		try_to_freeze();
238 		write_lock(&journal->j_state_lock);
239 	} else {
240 		/*
241 		 * We assume on resume that commits are already there,
242 		 * so we don't sleep
243 		 */
244 		DEFINE_WAIT(wait);
245 		int should_sleep = 1;
246 
247 		prepare_to_wait(&journal->j_wait_commit, &wait,
248 				TASK_INTERRUPTIBLE);
249 		if (journal->j_commit_sequence != journal->j_commit_request)
250 			should_sleep = 0;
251 		transaction = journal->j_running_transaction;
252 		if (transaction && time_after_eq(jiffies,
253 						transaction->t_expires))
254 			should_sleep = 0;
255 		if (journal->j_flags & JBD2_UNMOUNT)
256 			should_sleep = 0;
257 		if (should_sleep) {
258 			write_unlock(&journal->j_state_lock);
259 			schedule();
260 			write_lock(&journal->j_state_lock);
261 		}
262 		finish_wait(&journal->j_wait_commit, &wait);
263 	}
264 
265 	jbd_debug(1, "kjournald2 wakes\n");
266 
267 	/*
268 	 * Were we woken up by a commit wakeup event?
269 	 */
270 	transaction = journal->j_running_transaction;
271 	if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
272 		journal->j_commit_request = transaction->t_tid;
273 		jbd_debug(1, "woke because of timeout\n");
274 	}
275 	goto loop;
276 
277 end_loop:
278 	write_unlock(&journal->j_state_lock);
279 	del_timer_sync(&journal->j_commit_timer);
280 	journal->j_task = NULL;
281 	wake_up(&journal->j_wait_done_commit);
282 	jbd_debug(1, "Journal thread exiting.\n");
283 	return 0;
284 }
285 
286 static int jbd2_journal_start_thread(journal_t *journal)
287 {
288 	struct task_struct *t;
289 
290 	t = kthread_run(kjournald2, journal, "jbd2/%s",
291 			journal->j_devname);
292 	if (IS_ERR(t))
293 		return PTR_ERR(t);
294 
295 	wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
296 	return 0;
297 }
298 
299 static void journal_kill_thread(journal_t *journal)
300 {
301 	write_lock(&journal->j_state_lock);
302 	journal->j_flags |= JBD2_UNMOUNT;
303 
304 	while (journal->j_task) {
305 		write_unlock(&journal->j_state_lock);
306 		wake_up(&journal->j_wait_commit);
307 		wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
308 		write_lock(&journal->j_state_lock);
309 	}
310 	write_unlock(&journal->j_state_lock);
311 }
312 
313 /*
314  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
315  *
316  * Writes a metadata buffer to a given disk block.  The actual IO is not
317  * performed but a new buffer_head is constructed which labels the data
318  * to be written with the correct destination disk block.
319  *
320  * Any magic-number escaping which needs to be done will cause a
321  * copy-out here.  If the buffer happens to start with the
322  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
323  * magic number is only written to the log for descripter blocks.  In
324  * this case, we copy the data and replace the first word with 0, and we
325  * return a result code which indicates that this buffer needs to be
326  * marked as an escaped buffer in the corresponding log descriptor
327  * block.  The missing word can then be restored when the block is read
328  * during recovery.
329  *
330  * If the source buffer has already been modified by a new transaction
331  * since we took the last commit snapshot, we use the frozen copy of
332  * that data for IO. If we end up using the existing buffer_head's data
333  * for the write, then we have to make sure nobody modifies it while the
334  * IO is in progress. do_get_write_access() handles this.
335  *
336  * The function returns a pointer to the buffer_head to be used for IO.
337  *
338  *
339  * Return value:
340  *  <0: Error
341  * >=0: Finished OK
342  *
343  * On success:
344  * Bit 0 set == escape performed on the data
345  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
346  */
347 
348 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
349 				  struct journal_head  *jh_in,
350 				  struct buffer_head **bh_out,
351 				  sector_t blocknr)
352 {
353 	int need_copy_out = 0;
354 	int done_copy_out = 0;
355 	int do_escape = 0;
356 	char *mapped_data;
357 	struct buffer_head *new_bh;
358 	struct page *new_page;
359 	unsigned int new_offset;
360 	struct buffer_head *bh_in = jh2bh(jh_in);
361 	journal_t *journal = transaction->t_journal;
362 
363 	/*
364 	 * The buffer really shouldn't be locked: only the current committing
365 	 * transaction is allowed to write it, so nobody else is allowed
366 	 * to do any IO.
367 	 *
368 	 * akpm: except if we're journalling data, and write() output is
369 	 * also part of a shared mapping, and another thread has
370 	 * decided to launch a writepage() against this buffer.
371 	 */
372 	J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
373 
374 	new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
375 
376 	/* keep subsequent assertions sane */
377 	atomic_set(&new_bh->b_count, 1);
378 
379 	jbd_lock_bh_state(bh_in);
380 repeat:
381 	/*
382 	 * If a new transaction has already done a buffer copy-out, then
383 	 * we use that version of the data for the commit.
384 	 */
385 	if (jh_in->b_frozen_data) {
386 		done_copy_out = 1;
387 		new_page = virt_to_page(jh_in->b_frozen_data);
388 		new_offset = offset_in_page(jh_in->b_frozen_data);
389 	} else {
390 		new_page = jh2bh(jh_in)->b_page;
391 		new_offset = offset_in_page(jh2bh(jh_in)->b_data);
392 	}
393 
394 	mapped_data = kmap_atomic(new_page);
395 	/*
396 	 * Fire data frozen trigger if data already wasn't frozen.  Do this
397 	 * before checking for escaping, as the trigger may modify the magic
398 	 * offset.  If a copy-out happens afterwards, it will have the correct
399 	 * data in the buffer.
400 	 */
401 	if (!done_copy_out)
402 		jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
403 					   jh_in->b_triggers);
404 
405 	/*
406 	 * Check for escaping
407 	 */
408 	if (*((__be32 *)(mapped_data + new_offset)) ==
409 				cpu_to_be32(JBD2_MAGIC_NUMBER)) {
410 		need_copy_out = 1;
411 		do_escape = 1;
412 	}
413 	kunmap_atomic(mapped_data);
414 
415 	/*
416 	 * Do we need to do a data copy?
417 	 */
418 	if (need_copy_out && !done_copy_out) {
419 		char *tmp;
420 
421 		jbd_unlock_bh_state(bh_in);
422 		tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
423 		if (!tmp) {
424 			brelse(new_bh);
425 			return -ENOMEM;
426 		}
427 		jbd_lock_bh_state(bh_in);
428 		if (jh_in->b_frozen_data) {
429 			jbd2_free(tmp, bh_in->b_size);
430 			goto repeat;
431 		}
432 
433 		jh_in->b_frozen_data = tmp;
434 		mapped_data = kmap_atomic(new_page);
435 		memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
436 		kunmap_atomic(mapped_data);
437 
438 		new_page = virt_to_page(tmp);
439 		new_offset = offset_in_page(tmp);
440 		done_copy_out = 1;
441 
442 		/*
443 		 * This isn't strictly necessary, as we're using frozen
444 		 * data for the escaping, but it keeps consistency with
445 		 * b_frozen_data usage.
446 		 */
447 		jh_in->b_frozen_triggers = jh_in->b_triggers;
448 	}
449 
450 	/*
451 	 * Did we need to do an escaping?  Now we've done all the
452 	 * copying, we can finally do so.
453 	 */
454 	if (do_escape) {
455 		mapped_data = kmap_atomic(new_page);
456 		*((unsigned int *)(mapped_data + new_offset)) = 0;
457 		kunmap_atomic(mapped_data);
458 	}
459 
460 	set_bh_page(new_bh, new_page, new_offset);
461 	new_bh->b_size = bh_in->b_size;
462 	new_bh->b_bdev = journal->j_dev;
463 	new_bh->b_blocknr = blocknr;
464 	new_bh->b_private = bh_in;
465 	set_buffer_mapped(new_bh);
466 	set_buffer_dirty(new_bh);
467 
468 	*bh_out = new_bh;
469 
470 	/*
471 	 * The to-be-written buffer needs to get moved to the io queue,
472 	 * and the original buffer whose contents we are shadowing or
473 	 * copying is moved to the transaction's shadow queue.
474 	 */
475 	JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
476 	spin_lock(&journal->j_list_lock);
477 	__jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
478 	spin_unlock(&journal->j_list_lock);
479 	set_buffer_shadow(bh_in);
480 	jbd_unlock_bh_state(bh_in);
481 
482 	return do_escape | (done_copy_out << 1);
483 }
484 
485 /*
486  * Allocation code for the journal file.  Manage the space left in the
487  * journal, so that we can begin checkpointing when appropriate.
488  */
489 
490 /*
491  * Called with j_state_lock locked for writing.
492  * Returns true if a transaction commit was started.
493  */
494 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
495 {
496 	/* Return if the txn has already requested to be committed */
497 	if (journal->j_commit_request == target)
498 		return 0;
499 
500 	/*
501 	 * The only transaction we can possibly wait upon is the
502 	 * currently running transaction (if it exists).  Otherwise,
503 	 * the target tid must be an old one.
504 	 */
505 	if (journal->j_running_transaction &&
506 	    journal->j_running_transaction->t_tid == target) {
507 		/*
508 		 * We want a new commit: OK, mark the request and wakeup the
509 		 * commit thread.  We do _not_ do the commit ourselves.
510 		 */
511 
512 		journal->j_commit_request = target;
513 		jbd_debug(1, "JBD2: requesting commit %d/%d\n",
514 			  journal->j_commit_request,
515 			  journal->j_commit_sequence);
516 		journal->j_running_transaction->t_requested = jiffies;
517 		wake_up(&journal->j_wait_commit);
518 		return 1;
519 	} else if (!tid_geq(journal->j_commit_request, target))
520 		/* This should never happen, but if it does, preserve
521 		   the evidence before kjournald goes into a loop and
522 		   increments j_commit_sequence beyond all recognition. */
523 		WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
524 			  journal->j_commit_request,
525 			  journal->j_commit_sequence,
526 			  target, journal->j_running_transaction ?
527 			  journal->j_running_transaction->t_tid : 0);
528 	return 0;
529 }
530 
531 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
532 {
533 	int ret;
534 
535 	write_lock(&journal->j_state_lock);
536 	ret = __jbd2_log_start_commit(journal, tid);
537 	write_unlock(&journal->j_state_lock);
538 	return ret;
539 }
540 
541 /*
542  * Force and wait any uncommitted transactions.  We can only force the running
543  * transaction if we don't have an active handle, otherwise, we will deadlock.
544  * Returns: <0 in case of error,
545  *           0 if nothing to commit,
546  *           1 if transaction was successfully committed.
547  */
548 static int __jbd2_journal_force_commit(journal_t *journal)
549 {
550 	transaction_t *transaction = NULL;
551 	tid_t tid;
552 	int need_to_start = 0, ret = 0;
553 
554 	read_lock(&journal->j_state_lock);
555 	if (journal->j_running_transaction && !current->journal_info) {
556 		transaction = journal->j_running_transaction;
557 		if (!tid_geq(journal->j_commit_request, transaction->t_tid))
558 			need_to_start = 1;
559 	} else if (journal->j_committing_transaction)
560 		transaction = journal->j_committing_transaction;
561 
562 	if (!transaction) {
563 		/* Nothing to commit */
564 		read_unlock(&journal->j_state_lock);
565 		return 0;
566 	}
567 	tid = transaction->t_tid;
568 	read_unlock(&journal->j_state_lock);
569 	if (need_to_start)
570 		jbd2_log_start_commit(journal, tid);
571 	ret = jbd2_log_wait_commit(journal, tid);
572 	if (!ret)
573 		ret = 1;
574 
575 	return ret;
576 }
577 
578 /**
579  * Force and wait upon a commit if the calling process is not within
580  * transaction.  This is used for forcing out undo-protected data which contains
581  * bitmaps, when the fs is running out of space.
582  *
583  * @journal: journal to force
584  * Returns true if progress was made.
585  */
586 int jbd2_journal_force_commit_nested(journal_t *journal)
587 {
588 	int ret;
589 
590 	ret = __jbd2_journal_force_commit(journal);
591 	return ret > 0;
592 }
593 
594 /**
595  * int journal_force_commit() - force any uncommitted transactions
596  * @journal: journal to force
597  *
598  * Caller want unconditional commit. We can only force the running transaction
599  * if we don't have an active handle, otherwise, we will deadlock.
600  */
601 int jbd2_journal_force_commit(journal_t *journal)
602 {
603 	int ret;
604 
605 	J_ASSERT(!current->journal_info);
606 	ret = __jbd2_journal_force_commit(journal);
607 	if (ret > 0)
608 		ret = 0;
609 	return ret;
610 }
611 
612 /*
613  * Start a commit of the current running transaction (if any).  Returns true
614  * if a transaction is going to be committed (or is currently already
615  * committing), and fills its tid in at *ptid
616  */
617 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
618 {
619 	int ret = 0;
620 
621 	write_lock(&journal->j_state_lock);
622 	if (journal->j_running_transaction) {
623 		tid_t tid = journal->j_running_transaction->t_tid;
624 
625 		__jbd2_log_start_commit(journal, tid);
626 		/* There's a running transaction and we've just made sure
627 		 * it's commit has been scheduled. */
628 		if (ptid)
629 			*ptid = tid;
630 		ret = 1;
631 	} else if (journal->j_committing_transaction) {
632 		/*
633 		 * If commit has been started, then we have to wait for
634 		 * completion of that transaction.
635 		 */
636 		if (ptid)
637 			*ptid = journal->j_committing_transaction->t_tid;
638 		ret = 1;
639 	}
640 	write_unlock(&journal->j_state_lock);
641 	return ret;
642 }
643 
644 /*
645  * Return 1 if a given transaction has not yet sent barrier request
646  * connected with a transaction commit. If 0 is returned, transaction
647  * may or may not have sent the barrier. Used to avoid sending barrier
648  * twice in common cases.
649  */
650 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
651 {
652 	int ret = 0;
653 	transaction_t *commit_trans;
654 
655 	if (!(journal->j_flags & JBD2_BARRIER))
656 		return 0;
657 	read_lock(&journal->j_state_lock);
658 	/* Transaction already committed? */
659 	if (tid_geq(journal->j_commit_sequence, tid))
660 		goto out;
661 	commit_trans = journal->j_committing_transaction;
662 	if (!commit_trans || commit_trans->t_tid != tid) {
663 		ret = 1;
664 		goto out;
665 	}
666 	/*
667 	 * Transaction is being committed and we already proceeded to
668 	 * submitting a flush to fs partition?
669 	 */
670 	if (journal->j_fs_dev != journal->j_dev) {
671 		if (!commit_trans->t_need_data_flush ||
672 		    commit_trans->t_state >= T_COMMIT_DFLUSH)
673 			goto out;
674 	} else {
675 		if (commit_trans->t_state >= T_COMMIT_JFLUSH)
676 			goto out;
677 	}
678 	ret = 1;
679 out:
680 	read_unlock(&journal->j_state_lock);
681 	return ret;
682 }
683 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
684 
685 /*
686  * Wait for a specified commit to complete.
687  * The caller may not hold the journal lock.
688  */
689 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
690 {
691 	int err = 0;
692 
693 	read_lock(&journal->j_state_lock);
694 #ifdef CONFIG_JBD2_DEBUG
695 	if (!tid_geq(journal->j_commit_request, tid)) {
696 		printk(KERN_ERR
697 		       "%s: error: j_commit_request=%d, tid=%d\n",
698 		       __func__, journal->j_commit_request, tid);
699 	}
700 #endif
701 	while (tid_gt(tid, journal->j_commit_sequence)) {
702 		jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
703 				  tid, journal->j_commit_sequence);
704 		read_unlock(&journal->j_state_lock);
705 		wake_up(&journal->j_wait_commit);
706 		wait_event(journal->j_wait_done_commit,
707 				!tid_gt(tid, journal->j_commit_sequence));
708 		read_lock(&journal->j_state_lock);
709 	}
710 	read_unlock(&journal->j_state_lock);
711 
712 	if (unlikely(is_journal_aborted(journal)))
713 		err = -EIO;
714 	return err;
715 }
716 
717 /*
718  * When this function returns the transaction corresponding to tid
719  * will be completed.  If the transaction has currently running, start
720  * committing that transaction before waiting for it to complete.  If
721  * the transaction id is stale, it is by definition already completed,
722  * so just return SUCCESS.
723  */
724 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
725 {
726 	int	need_to_wait = 1;
727 
728 	read_lock(&journal->j_state_lock);
729 	if (journal->j_running_transaction &&
730 	    journal->j_running_transaction->t_tid == tid) {
731 		if (journal->j_commit_request != tid) {
732 			/* transaction not yet started, so request it */
733 			read_unlock(&journal->j_state_lock);
734 			jbd2_log_start_commit(journal, tid);
735 			goto wait_commit;
736 		}
737 	} else if (!(journal->j_committing_transaction &&
738 		     journal->j_committing_transaction->t_tid == tid))
739 		need_to_wait = 0;
740 	read_unlock(&journal->j_state_lock);
741 	if (!need_to_wait)
742 		return 0;
743 wait_commit:
744 	return jbd2_log_wait_commit(journal, tid);
745 }
746 EXPORT_SYMBOL(jbd2_complete_transaction);
747 
748 /*
749  * Log buffer allocation routines:
750  */
751 
752 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
753 {
754 	unsigned long blocknr;
755 
756 	write_lock(&journal->j_state_lock);
757 	J_ASSERT(journal->j_free > 1);
758 
759 	blocknr = journal->j_head;
760 	journal->j_head++;
761 	journal->j_free--;
762 	if (journal->j_head == journal->j_last)
763 		journal->j_head = journal->j_first;
764 	write_unlock(&journal->j_state_lock);
765 	return jbd2_journal_bmap(journal, blocknr, retp);
766 }
767 
768 /*
769  * Conversion of logical to physical block numbers for the journal
770  *
771  * On external journals the journal blocks are identity-mapped, so
772  * this is a no-op.  If needed, we can use j_blk_offset - everything is
773  * ready.
774  */
775 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
776 		 unsigned long long *retp)
777 {
778 	int err = 0;
779 	unsigned long long ret;
780 
781 	if (journal->j_inode) {
782 		ret = bmap(journal->j_inode, blocknr);
783 		if (ret)
784 			*retp = ret;
785 		else {
786 			printk(KERN_ALERT "%s: journal block not found "
787 					"at offset %lu on %s\n",
788 			       __func__, blocknr, journal->j_devname);
789 			err = -EIO;
790 			__journal_abort_soft(journal, err);
791 		}
792 	} else {
793 		*retp = blocknr; /* +journal->j_blk_offset */
794 	}
795 	return err;
796 }
797 
798 /*
799  * We play buffer_head aliasing tricks to write data/metadata blocks to
800  * the journal without copying their contents, but for journal
801  * descriptor blocks we do need to generate bona fide buffers.
802  *
803  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
804  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
805  * But we don't bother doing that, so there will be coherency problems with
806  * mmaps of blockdevs which hold live JBD-controlled filesystems.
807  */
808 struct buffer_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
809 {
810 	struct buffer_head *bh;
811 	unsigned long long blocknr;
812 	int err;
813 
814 	err = jbd2_journal_next_log_block(journal, &blocknr);
815 
816 	if (err)
817 		return NULL;
818 
819 	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
820 	if (!bh)
821 		return NULL;
822 	lock_buffer(bh);
823 	memset(bh->b_data, 0, journal->j_blocksize);
824 	set_buffer_uptodate(bh);
825 	unlock_buffer(bh);
826 	BUFFER_TRACE(bh, "return this buffer");
827 	return bh;
828 }
829 
830 /*
831  * Return tid of the oldest transaction in the journal and block in the journal
832  * where the transaction starts.
833  *
834  * If the journal is now empty, return which will be the next transaction ID
835  * we will write and where will that transaction start.
836  *
837  * The return value is 0 if journal tail cannot be pushed any further, 1 if
838  * it can.
839  */
840 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
841 			      unsigned long *block)
842 {
843 	transaction_t *transaction;
844 	int ret;
845 
846 	read_lock(&journal->j_state_lock);
847 	spin_lock(&journal->j_list_lock);
848 	transaction = journal->j_checkpoint_transactions;
849 	if (transaction) {
850 		*tid = transaction->t_tid;
851 		*block = transaction->t_log_start;
852 	} else if ((transaction = journal->j_committing_transaction) != NULL) {
853 		*tid = transaction->t_tid;
854 		*block = transaction->t_log_start;
855 	} else if ((transaction = journal->j_running_transaction) != NULL) {
856 		*tid = transaction->t_tid;
857 		*block = journal->j_head;
858 	} else {
859 		*tid = journal->j_transaction_sequence;
860 		*block = journal->j_head;
861 	}
862 	ret = tid_gt(*tid, journal->j_tail_sequence);
863 	spin_unlock(&journal->j_list_lock);
864 	read_unlock(&journal->j_state_lock);
865 
866 	return ret;
867 }
868 
869 /*
870  * Update information in journal structure and in on disk journal superblock
871  * about log tail. This function does not check whether information passed in
872  * really pushes log tail further. It's responsibility of the caller to make
873  * sure provided log tail information is valid (e.g. by holding
874  * j_checkpoint_mutex all the time between computing log tail and calling this
875  * function as is the case with jbd2_cleanup_journal_tail()).
876  *
877  * Requires j_checkpoint_mutex
878  */
879 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
880 {
881 	unsigned long freed;
882 	int ret;
883 
884 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
885 
886 	/*
887 	 * We cannot afford for write to remain in drive's caches since as
888 	 * soon as we update j_tail, next transaction can start reusing journal
889 	 * space and if we lose sb update during power failure we'd replay
890 	 * old transaction with possibly newly overwritten data.
891 	 */
892 	ret = jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
893 	if (ret)
894 		goto out;
895 
896 	write_lock(&journal->j_state_lock);
897 	freed = block - journal->j_tail;
898 	if (block < journal->j_tail)
899 		freed += journal->j_last - journal->j_first;
900 
901 	trace_jbd2_update_log_tail(journal, tid, block, freed);
902 	jbd_debug(1,
903 		  "Cleaning journal tail from %d to %d (offset %lu), "
904 		  "freeing %lu\n",
905 		  journal->j_tail_sequence, tid, block, freed);
906 
907 	journal->j_free += freed;
908 	journal->j_tail_sequence = tid;
909 	journal->j_tail = block;
910 	write_unlock(&journal->j_state_lock);
911 
912 out:
913 	return ret;
914 }
915 
916 /*
917  * This is a variaon of __jbd2_update_log_tail which checks for validity of
918  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
919  * with other threads updating log tail.
920  */
921 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
922 {
923 	mutex_lock(&journal->j_checkpoint_mutex);
924 	if (tid_gt(tid, journal->j_tail_sequence))
925 		__jbd2_update_log_tail(journal, tid, block);
926 	mutex_unlock(&journal->j_checkpoint_mutex);
927 }
928 
929 struct jbd2_stats_proc_session {
930 	journal_t *journal;
931 	struct transaction_stats_s *stats;
932 	int start;
933 	int max;
934 };
935 
936 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
937 {
938 	return *pos ? NULL : SEQ_START_TOKEN;
939 }
940 
941 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
942 {
943 	return NULL;
944 }
945 
946 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
947 {
948 	struct jbd2_stats_proc_session *s = seq->private;
949 
950 	if (v != SEQ_START_TOKEN)
951 		return 0;
952 	seq_printf(seq, "%lu transactions (%lu requested), "
953 		   "each up to %u blocks\n",
954 		   s->stats->ts_tid, s->stats->ts_requested,
955 		   s->journal->j_max_transaction_buffers);
956 	if (s->stats->ts_tid == 0)
957 		return 0;
958 	seq_printf(seq, "average: \n  %ums waiting for transaction\n",
959 	    jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
960 	seq_printf(seq, "  %ums request delay\n",
961 	    (s->stats->ts_requested == 0) ? 0 :
962 	    jiffies_to_msecs(s->stats->run.rs_request_delay /
963 			     s->stats->ts_requested));
964 	seq_printf(seq, "  %ums running transaction\n",
965 	    jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
966 	seq_printf(seq, "  %ums transaction was being locked\n",
967 	    jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
968 	seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
969 	    jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
970 	seq_printf(seq, "  %ums logging transaction\n",
971 	    jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
972 	seq_printf(seq, "  %lluus average transaction commit time\n",
973 		   div_u64(s->journal->j_average_commit_time, 1000));
974 	seq_printf(seq, "  %lu handles per transaction\n",
975 	    s->stats->run.rs_handle_count / s->stats->ts_tid);
976 	seq_printf(seq, "  %lu blocks per transaction\n",
977 	    s->stats->run.rs_blocks / s->stats->ts_tid);
978 	seq_printf(seq, "  %lu logged blocks per transaction\n",
979 	    s->stats->run.rs_blocks_logged / s->stats->ts_tid);
980 	return 0;
981 }
982 
983 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
984 {
985 }
986 
987 static const struct seq_operations jbd2_seq_info_ops = {
988 	.start  = jbd2_seq_info_start,
989 	.next   = jbd2_seq_info_next,
990 	.stop   = jbd2_seq_info_stop,
991 	.show   = jbd2_seq_info_show,
992 };
993 
994 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
995 {
996 	journal_t *journal = PDE_DATA(inode);
997 	struct jbd2_stats_proc_session *s;
998 	int rc, size;
999 
1000 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1001 	if (s == NULL)
1002 		return -ENOMEM;
1003 	size = sizeof(struct transaction_stats_s);
1004 	s->stats = kmalloc(size, GFP_KERNEL);
1005 	if (s->stats == NULL) {
1006 		kfree(s);
1007 		return -ENOMEM;
1008 	}
1009 	spin_lock(&journal->j_history_lock);
1010 	memcpy(s->stats, &journal->j_stats, size);
1011 	s->journal = journal;
1012 	spin_unlock(&journal->j_history_lock);
1013 
1014 	rc = seq_open(file, &jbd2_seq_info_ops);
1015 	if (rc == 0) {
1016 		struct seq_file *m = file->private_data;
1017 		m->private = s;
1018 	} else {
1019 		kfree(s->stats);
1020 		kfree(s);
1021 	}
1022 	return rc;
1023 
1024 }
1025 
1026 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1027 {
1028 	struct seq_file *seq = file->private_data;
1029 	struct jbd2_stats_proc_session *s = seq->private;
1030 	kfree(s->stats);
1031 	kfree(s);
1032 	return seq_release(inode, file);
1033 }
1034 
1035 static const struct file_operations jbd2_seq_info_fops = {
1036 	.owner		= THIS_MODULE,
1037 	.open           = jbd2_seq_info_open,
1038 	.read           = seq_read,
1039 	.llseek         = seq_lseek,
1040 	.release        = jbd2_seq_info_release,
1041 };
1042 
1043 static struct proc_dir_entry *proc_jbd2_stats;
1044 
1045 static void jbd2_stats_proc_init(journal_t *journal)
1046 {
1047 	journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1048 	if (journal->j_proc_entry) {
1049 		proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1050 				 &jbd2_seq_info_fops, journal);
1051 	}
1052 }
1053 
1054 static void jbd2_stats_proc_exit(journal_t *journal)
1055 {
1056 	remove_proc_entry("info", journal->j_proc_entry);
1057 	remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1058 }
1059 
1060 /*
1061  * Management for journal control blocks: functions to create and
1062  * destroy journal_t structures, and to initialise and read existing
1063  * journal blocks from disk.  */
1064 
1065 /* First: create and setup a journal_t object in memory.  We initialise
1066  * very few fields yet: that has to wait until we have created the
1067  * journal structures from from scratch, or loaded them from disk. */
1068 
1069 static journal_t * journal_init_common (void)
1070 {
1071 	journal_t *journal;
1072 	int err;
1073 
1074 	journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1075 	if (!journal)
1076 		return NULL;
1077 
1078 	init_waitqueue_head(&journal->j_wait_transaction_locked);
1079 	init_waitqueue_head(&journal->j_wait_done_commit);
1080 	init_waitqueue_head(&journal->j_wait_commit);
1081 	init_waitqueue_head(&journal->j_wait_updates);
1082 	init_waitqueue_head(&journal->j_wait_reserved);
1083 	mutex_init(&journal->j_barrier);
1084 	mutex_init(&journal->j_checkpoint_mutex);
1085 	spin_lock_init(&journal->j_revoke_lock);
1086 	spin_lock_init(&journal->j_list_lock);
1087 	rwlock_init(&journal->j_state_lock);
1088 
1089 	journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1090 	journal->j_min_batch_time = 0;
1091 	journal->j_max_batch_time = 15000; /* 15ms */
1092 	atomic_set(&journal->j_reserved_credits, 0);
1093 
1094 	/* The journal is marked for error until we succeed with recovery! */
1095 	journal->j_flags = JBD2_ABORT;
1096 
1097 	/* Set up a default-sized revoke table for the new mount. */
1098 	err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1099 	if (err) {
1100 		kfree(journal);
1101 		return NULL;
1102 	}
1103 
1104 	spin_lock_init(&journal->j_history_lock);
1105 
1106 	return journal;
1107 }
1108 
1109 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1110  *
1111  * Create a journal structure assigned some fixed set of disk blocks to
1112  * the journal.  We don't actually touch those disk blocks yet, but we
1113  * need to set up all of the mapping information to tell the journaling
1114  * system where the journal blocks are.
1115  *
1116  */
1117 
1118 /**
1119  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1120  *  @bdev: Block device on which to create the journal
1121  *  @fs_dev: Device which hold journalled filesystem for this journal.
1122  *  @start: Block nr Start of journal.
1123  *  @len:  Length of the journal in blocks.
1124  *  @blocksize: blocksize of journalling device
1125  *
1126  *  Returns: a newly created journal_t *
1127  *
1128  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1129  *  range of blocks on an arbitrary block device.
1130  *
1131  */
1132 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1133 			struct block_device *fs_dev,
1134 			unsigned long long start, int len, int blocksize)
1135 {
1136 	journal_t *journal = journal_init_common();
1137 	struct buffer_head *bh;
1138 	int n;
1139 
1140 	if (!journal)
1141 		return NULL;
1142 
1143 	/* journal descriptor can store up to n blocks -bzzz */
1144 	journal->j_blocksize = blocksize;
1145 	journal->j_dev = bdev;
1146 	journal->j_fs_dev = fs_dev;
1147 	journal->j_blk_offset = start;
1148 	journal->j_maxlen = len;
1149 	bdevname(journal->j_dev, journal->j_devname);
1150 	strreplace(journal->j_devname, '/', '!');
1151 	jbd2_stats_proc_init(journal);
1152 	n = journal->j_blocksize / sizeof(journal_block_tag_t);
1153 	journal->j_wbufsize = n;
1154 	journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1155 	if (!journal->j_wbuf) {
1156 		printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1157 			__func__);
1158 		goto out_err;
1159 	}
1160 
1161 	bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1162 	if (!bh) {
1163 		printk(KERN_ERR
1164 		       "%s: Cannot get buffer for journal superblock\n",
1165 		       __func__);
1166 		goto out_err;
1167 	}
1168 	journal->j_sb_buffer = bh;
1169 	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1170 
1171 	return journal;
1172 out_err:
1173 	kfree(journal->j_wbuf);
1174 	jbd2_stats_proc_exit(journal);
1175 	kfree(journal);
1176 	return NULL;
1177 }
1178 
1179 /**
1180  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1181  *  @inode: An inode to create the journal in
1182  *
1183  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1184  * the journal.  The inode must exist already, must support bmap() and
1185  * must have all data blocks preallocated.
1186  */
1187 journal_t * jbd2_journal_init_inode (struct inode *inode)
1188 {
1189 	struct buffer_head *bh;
1190 	journal_t *journal = journal_init_common();
1191 	char *p;
1192 	int err;
1193 	int n;
1194 	unsigned long long blocknr;
1195 
1196 	if (!journal)
1197 		return NULL;
1198 
1199 	journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1200 	journal->j_inode = inode;
1201 	bdevname(journal->j_dev, journal->j_devname);
1202 	p = strreplace(journal->j_devname, '/', '!');
1203 	sprintf(p, "-%lu", journal->j_inode->i_ino);
1204 	jbd_debug(1,
1205 		  "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1206 		  journal, inode->i_sb->s_id, inode->i_ino,
1207 		  (long long) inode->i_size,
1208 		  inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1209 
1210 	journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1211 	journal->j_blocksize = inode->i_sb->s_blocksize;
1212 	jbd2_stats_proc_init(journal);
1213 
1214 	/* journal descriptor can store up to n blocks -bzzz */
1215 	n = journal->j_blocksize / sizeof(journal_block_tag_t);
1216 	journal->j_wbufsize = n;
1217 	journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1218 	if (!journal->j_wbuf) {
1219 		printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1220 			__func__);
1221 		goto out_err;
1222 	}
1223 
1224 	err = jbd2_journal_bmap(journal, 0, &blocknr);
1225 	/* If that failed, give up */
1226 	if (err) {
1227 		printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1228 		       __func__);
1229 		goto out_err;
1230 	}
1231 
1232 	bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize);
1233 	if (!bh) {
1234 		printk(KERN_ERR
1235 		       "%s: Cannot get buffer for journal superblock\n",
1236 		       __func__);
1237 		goto out_err;
1238 	}
1239 	journal->j_sb_buffer = bh;
1240 	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1241 
1242 	return journal;
1243 out_err:
1244 	kfree(journal->j_wbuf);
1245 	jbd2_stats_proc_exit(journal);
1246 	kfree(journal);
1247 	return NULL;
1248 }
1249 
1250 /*
1251  * If the journal init or create aborts, we need to mark the journal
1252  * superblock as being NULL to prevent the journal destroy from writing
1253  * back a bogus superblock.
1254  */
1255 static void journal_fail_superblock (journal_t *journal)
1256 {
1257 	struct buffer_head *bh = journal->j_sb_buffer;
1258 	brelse(bh);
1259 	journal->j_sb_buffer = NULL;
1260 }
1261 
1262 /*
1263  * Given a journal_t structure, initialise the various fields for
1264  * startup of a new journaling session.  We use this both when creating
1265  * a journal, and after recovering an old journal to reset it for
1266  * subsequent use.
1267  */
1268 
1269 static int journal_reset(journal_t *journal)
1270 {
1271 	journal_superblock_t *sb = journal->j_superblock;
1272 	unsigned long long first, last;
1273 
1274 	first = be32_to_cpu(sb->s_first);
1275 	last = be32_to_cpu(sb->s_maxlen);
1276 	if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1277 		printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1278 		       first, last);
1279 		journal_fail_superblock(journal);
1280 		return -EINVAL;
1281 	}
1282 
1283 	journal->j_first = first;
1284 	journal->j_last = last;
1285 
1286 	journal->j_head = first;
1287 	journal->j_tail = first;
1288 	journal->j_free = last - first;
1289 
1290 	journal->j_tail_sequence = journal->j_transaction_sequence;
1291 	journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1292 	journal->j_commit_request = journal->j_commit_sequence;
1293 
1294 	journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1295 
1296 	/*
1297 	 * As a special case, if the on-disk copy is already marked as needing
1298 	 * no recovery (s_start == 0), then we can safely defer the superblock
1299 	 * update until the next commit by setting JBD2_FLUSHED.  This avoids
1300 	 * attempting a write to a potential-readonly device.
1301 	 */
1302 	if (sb->s_start == 0) {
1303 		jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1304 			"(start %ld, seq %d, errno %d)\n",
1305 			journal->j_tail, journal->j_tail_sequence,
1306 			journal->j_errno);
1307 		journal->j_flags |= JBD2_FLUSHED;
1308 	} else {
1309 		/* Lock here to make assertions happy... */
1310 		mutex_lock(&journal->j_checkpoint_mutex);
1311 		/*
1312 		 * Update log tail information. We use WRITE_FUA since new
1313 		 * transaction will start reusing journal space and so we
1314 		 * must make sure information about current log tail is on
1315 		 * disk before that.
1316 		 */
1317 		jbd2_journal_update_sb_log_tail(journal,
1318 						journal->j_tail_sequence,
1319 						journal->j_tail,
1320 						WRITE_FUA);
1321 		mutex_unlock(&journal->j_checkpoint_mutex);
1322 	}
1323 	return jbd2_journal_start_thread(journal);
1324 }
1325 
1326 static int jbd2_write_superblock(journal_t *journal, int write_op)
1327 {
1328 	struct buffer_head *bh = journal->j_sb_buffer;
1329 	journal_superblock_t *sb = journal->j_superblock;
1330 	int ret;
1331 
1332 	trace_jbd2_write_superblock(journal, write_op);
1333 	if (!(journal->j_flags & JBD2_BARRIER))
1334 		write_op &= ~(REQ_FUA | REQ_FLUSH);
1335 	lock_buffer(bh);
1336 	if (buffer_write_io_error(bh)) {
1337 		/*
1338 		 * Oh, dear.  A previous attempt to write the journal
1339 		 * superblock failed.  This could happen because the
1340 		 * USB device was yanked out.  Or it could happen to
1341 		 * be a transient write error and maybe the block will
1342 		 * be remapped.  Nothing we can do but to retry the
1343 		 * write and hope for the best.
1344 		 */
1345 		printk(KERN_ERR "JBD2: previous I/O error detected "
1346 		       "for journal superblock update for %s.\n",
1347 		       journal->j_devname);
1348 		clear_buffer_write_io_error(bh);
1349 		set_buffer_uptodate(bh);
1350 	}
1351 	jbd2_superblock_csum_set(journal, sb);
1352 	get_bh(bh);
1353 	bh->b_end_io = end_buffer_write_sync;
1354 	ret = submit_bh(write_op, bh);
1355 	wait_on_buffer(bh);
1356 	if (buffer_write_io_error(bh)) {
1357 		clear_buffer_write_io_error(bh);
1358 		set_buffer_uptodate(bh);
1359 		ret = -EIO;
1360 	}
1361 	if (ret) {
1362 		printk(KERN_ERR "JBD2: Error %d detected when updating "
1363 		       "journal superblock for %s.\n", ret,
1364 		       journal->j_devname);
1365 		jbd2_journal_abort(journal, ret);
1366 	}
1367 
1368 	return ret;
1369 }
1370 
1371 /**
1372  * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1373  * @journal: The journal to update.
1374  * @tail_tid: TID of the new transaction at the tail of the log
1375  * @tail_block: The first block of the transaction at the tail of the log
1376  * @write_op: With which operation should we write the journal sb
1377  *
1378  * Update a journal's superblock information about log tail and write it to
1379  * disk, waiting for the IO to complete.
1380  */
1381 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1382 				     unsigned long tail_block, int write_op)
1383 {
1384 	journal_superblock_t *sb = journal->j_superblock;
1385 	int ret;
1386 
1387 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1388 	jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1389 		  tail_block, tail_tid);
1390 
1391 	sb->s_sequence = cpu_to_be32(tail_tid);
1392 	sb->s_start    = cpu_to_be32(tail_block);
1393 
1394 	ret = jbd2_write_superblock(journal, write_op);
1395 	if (ret)
1396 		goto out;
1397 
1398 	/* Log is no longer empty */
1399 	write_lock(&journal->j_state_lock);
1400 	WARN_ON(!sb->s_sequence);
1401 	journal->j_flags &= ~JBD2_FLUSHED;
1402 	write_unlock(&journal->j_state_lock);
1403 
1404 out:
1405 	return ret;
1406 }
1407 
1408 /**
1409  * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1410  * @journal: The journal to update.
1411  *
1412  * Update a journal's dynamic superblock fields to show that journal is empty.
1413  * Write updated superblock to disk waiting for IO to complete.
1414  */
1415 static void jbd2_mark_journal_empty(journal_t *journal)
1416 {
1417 	journal_superblock_t *sb = journal->j_superblock;
1418 
1419 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1420 	read_lock(&journal->j_state_lock);
1421 	/* Is it already empty? */
1422 	if (sb->s_start == 0) {
1423 		read_unlock(&journal->j_state_lock);
1424 		return;
1425 	}
1426 	jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1427 		  journal->j_tail_sequence);
1428 
1429 	sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1430 	sb->s_start    = cpu_to_be32(0);
1431 	read_unlock(&journal->j_state_lock);
1432 
1433 	jbd2_write_superblock(journal, WRITE_FUA);
1434 
1435 	/* Log is no longer empty */
1436 	write_lock(&journal->j_state_lock);
1437 	journal->j_flags |= JBD2_FLUSHED;
1438 	write_unlock(&journal->j_state_lock);
1439 }
1440 
1441 
1442 /**
1443  * jbd2_journal_update_sb_errno() - Update error in the journal.
1444  * @journal: The journal to update.
1445  *
1446  * Update a journal's errno.  Write updated superblock to disk waiting for IO
1447  * to complete.
1448  */
1449 void jbd2_journal_update_sb_errno(journal_t *journal)
1450 {
1451 	journal_superblock_t *sb = journal->j_superblock;
1452 
1453 	read_lock(&journal->j_state_lock);
1454 	jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1455 		  journal->j_errno);
1456 	sb->s_errno    = cpu_to_be32(journal->j_errno);
1457 	read_unlock(&journal->j_state_lock);
1458 
1459 	jbd2_write_superblock(journal, WRITE_FUA);
1460 }
1461 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1462 
1463 /*
1464  * Read the superblock for a given journal, performing initial
1465  * validation of the format.
1466  */
1467 static int journal_get_superblock(journal_t *journal)
1468 {
1469 	struct buffer_head *bh;
1470 	journal_superblock_t *sb;
1471 	int err = -EIO;
1472 
1473 	bh = journal->j_sb_buffer;
1474 
1475 	J_ASSERT(bh != NULL);
1476 	if (!buffer_uptodate(bh)) {
1477 		ll_rw_block(READ, 1, &bh);
1478 		wait_on_buffer(bh);
1479 		if (!buffer_uptodate(bh)) {
1480 			printk(KERN_ERR
1481 				"JBD2: IO error reading journal superblock\n");
1482 			goto out;
1483 		}
1484 	}
1485 
1486 	if (buffer_verified(bh))
1487 		return 0;
1488 
1489 	sb = journal->j_superblock;
1490 
1491 	err = -EINVAL;
1492 
1493 	if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1494 	    sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1495 		printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1496 		goto out;
1497 	}
1498 
1499 	switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1500 	case JBD2_SUPERBLOCK_V1:
1501 		journal->j_format_version = 1;
1502 		break;
1503 	case JBD2_SUPERBLOCK_V2:
1504 		journal->j_format_version = 2;
1505 		break;
1506 	default:
1507 		printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1508 		goto out;
1509 	}
1510 
1511 	if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1512 		journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1513 	else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1514 		printk(KERN_WARNING "JBD2: journal file too short\n");
1515 		goto out;
1516 	}
1517 
1518 	if (be32_to_cpu(sb->s_first) == 0 ||
1519 	    be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1520 		printk(KERN_WARNING
1521 			"JBD2: Invalid start block of journal: %u\n",
1522 			be32_to_cpu(sb->s_first));
1523 		goto out;
1524 	}
1525 
1526 	if (jbd2_has_feature_csum2(journal) &&
1527 	    jbd2_has_feature_csum3(journal)) {
1528 		/* Can't have checksum v2 and v3 at the same time! */
1529 		printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1530 		       "at the same time!\n");
1531 		goto out;
1532 	}
1533 
1534 	if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1535 	    jbd2_has_feature_checksum(journal)) {
1536 		/* Can't have checksum v1 and v2 on at the same time! */
1537 		printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1538 		       "at the same time!\n");
1539 		goto out;
1540 	}
1541 
1542 	if (!jbd2_verify_csum_type(journal, sb)) {
1543 		printk(KERN_ERR "JBD2: Unknown checksum type\n");
1544 		goto out;
1545 	}
1546 
1547 	/* Load the checksum driver */
1548 	if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1549 		journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1550 		if (IS_ERR(journal->j_chksum_driver)) {
1551 			printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1552 			err = PTR_ERR(journal->j_chksum_driver);
1553 			journal->j_chksum_driver = NULL;
1554 			goto out;
1555 		}
1556 	}
1557 
1558 	/* Check superblock checksum */
1559 	if (!jbd2_superblock_csum_verify(journal, sb)) {
1560 		printk(KERN_ERR "JBD2: journal checksum error\n");
1561 		err = -EFSBADCRC;
1562 		goto out;
1563 	}
1564 
1565 	/* Precompute checksum seed for all metadata */
1566 	if (jbd2_journal_has_csum_v2or3(journal))
1567 		journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1568 						   sizeof(sb->s_uuid));
1569 
1570 	set_buffer_verified(bh);
1571 
1572 	return 0;
1573 
1574 out:
1575 	journal_fail_superblock(journal);
1576 	return err;
1577 }
1578 
1579 /*
1580  * Load the on-disk journal superblock and read the key fields into the
1581  * journal_t.
1582  */
1583 
1584 static int load_superblock(journal_t *journal)
1585 {
1586 	int err;
1587 	journal_superblock_t *sb;
1588 
1589 	err = journal_get_superblock(journal);
1590 	if (err)
1591 		return err;
1592 
1593 	sb = journal->j_superblock;
1594 
1595 	journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1596 	journal->j_tail = be32_to_cpu(sb->s_start);
1597 	journal->j_first = be32_to_cpu(sb->s_first);
1598 	journal->j_last = be32_to_cpu(sb->s_maxlen);
1599 	journal->j_errno = be32_to_cpu(sb->s_errno);
1600 
1601 	return 0;
1602 }
1603 
1604 
1605 /**
1606  * int jbd2_journal_load() - Read journal from disk.
1607  * @journal: Journal to act on.
1608  *
1609  * Given a journal_t structure which tells us which disk blocks contain
1610  * a journal, read the journal from disk to initialise the in-memory
1611  * structures.
1612  */
1613 int jbd2_journal_load(journal_t *journal)
1614 {
1615 	int err;
1616 	journal_superblock_t *sb;
1617 
1618 	err = load_superblock(journal);
1619 	if (err)
1620 		return err;
1621 
1622 	sb = journal->j_superblock;
1623 	/* If this is a V2 superblock, then we have to check the
1624 	 * features flags on it. */
1625 
1626 	if (journal->j_format_version >= 2) {
1627 		if ((sb->s_feature_ro_compat &
1628 		     ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1629 		    (sb->s_feature_incompat &
1630 		     ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1631 			printk(KERN_WARNING
1632 				"JBD2: Unrecognised features on journal\n");
1633 			return -EINVAL;
1634 		}
1635 	}
1636 
1637 	/*
1638 	 * Create a slab for this blocksize
1639 	 */
1640 	err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1641 	if (err)
1642 		return err;
1643 
1644 	/* Let the recovery code check whether it needs to recover any
1645 	 * data from the journal. */
1646 	if (jbd2_journal_recover(journal))
1647 		goto recovery_error;
1648 
1649 	if (journal->j_failed_commit) {
1650 		printk(KERN_ERR "JBD2: journal transaction %u on %s "
1651 		       "is corrupt.\n", journal->j_failed_commit,
1652 		       journal->j_devname);
1653 		return -EFSCORRUPTED;
1654 	}
1655 
1656 	/* OK, we've finished with the dynamic journal bits:
1657 	 * reinitialise the dynamic contents of the superblock in memory
1658 	 * and reset them on disk. */
1659 	if (journal_reset(journal))
1660 		goto recovery_error;
1661 
1662 	journal->j_flags &= ~JBD2_ABORT;
1663 	journal->j_flags |= JBD2_LOADED;
1664 	return 0;
1665 
1666 recovery_error:
1667 	printk(KERN_WARNING "JBD2: recovery failed\n");
1668 	return -EIO;
1669 }
1670 
1671 /**
1672  * void jbd2_journal_destroy() - Release a journal_t structure.
1673  * @journal: Journal to act on.
1674  *
1675  * Release a journal_t structure once it is no longer in use by the
1676  * journaled object.
1677  * Return <0 if we couldn't clean up the journal.
1678  */
1679 int jbd2_journal_destroy(journal_t *journal)
1680 {
1681 	int err = 0;
1682 
1683 	/* Wait for the commit thread to wake up and die. */
1684 	journal_kill_thread(journal);
1685 
1686 	/* Force a final log commit */
1687 	if (journal->j_running_transaction)
1688 		jbd2_journal_commit_transaction(journal);
1689 
1690 	/* Force any old transactions to disk */
1691 
1692 	/* Totally anal locking here... */
1693 	spin_lock(&journal->j_list_lock);
1694 	while (journal->j_checkpoint_transactions != NULL) {
1695 		spin_unlock(&journal->j_list_lock);
1696 		mutex_lock(&journal->j_checkpoint_mutex);
1697 		err = jbd2_log_do_checkpoint(journal);
1698 		mutex_unlock(&journal->j_checkpoint_mutex);
1699 		/*
1700 		 * If checkpointing failed, just free the buffers to avoid
1701 		 * looping forever
1702 		 */
1703 		if (err) {
1704 			jbd2_journal_destroy_checkpoint(journal);
1705 			spin_lock(&journal->j_list_lock);
1706 			break;
1707 		}
1708 		spin_lock(&journal->j_list_lock);
1709 	}
1710 
1711 	J_ASSERT(journal->j_running_transaction == NULL);
1712 	J_ASSERT(journal->j_committing_transaction == NULL);
1713 	J_ASSERT(journal->j_checkpoint_transactions == NULL);
1714 	spin_unlock(&journal->j_list_lock);
1715 
1716 	if (journal->j_sb_buffer) {
1717 		if (!is_journal_aborted(journal)) {
1718 			mutex_lock(&journal->j_checkpoint_mutex);
1719 			jbd2_mark_journal_empty(journal);
1720 			mutex_unlock(&journal->j_checkpoint_mutex);
1721 		} else
1722 			err = -EIO;
1723 		brelse(journal->j_sb_buffer);
1724 	}
1725 
1726 	if (journal->j_proc_entry)
1727 		jbd2_stats_proc_exit(journal);
1728 	iput(journal->j_inode);
1729 	if (journal->j_revoke)
1730 		jbd2_journal_destroy_revoke(journal);
1731 	if (journal->j_chksum_driver)
1732 		crypto_free_shash(journal->j_chksum_driver);
1733 	kfree(journal->j_wbuf);
1734 	kfree(journal);
1735 
1736 	return err;
1737 }
1738 
1739 
1740 /**
1741  *int jbd2_journal_check_used_features () - Check if features specified are used.
1742  * @journal: Journal to check.
1743  * @compat: bitmask of compatible features
1744  * @ro: bitmask of features that force read-only mount
1745  * @incompat: bitmask of incompatible features
1746  *
1747  * Check whether the journal uses all of a given set of
1748  * features.  Return true (non-zero) if it does.
1749  **/
1750 
1751 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1752 				 unsigned long ro, unsigned long incompat)
1753 {
1754 	journal_superblock_t *sb;
1755 
1756 	if (!compat && !ro && !incompat)
1757 		return 1;
1758 	/* Load journal superblock if it is not loaded yet. */
1759 	if (journal->j_format_version == 0 &&
1760 	    journal_get_superblock(journal) != 0)
1761 		return 0;
1762 	if (journal->j_format_version == 1)
1763 		return 0;
1764 
1765 	sb = journal->j_superblock;
1766 
1767 	if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1768 	    ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1769 	    ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1770 		return 1;
1771 
1772 	return 0;
1773 }
1774 
1775 /**
1776  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1777  * @journal: Journal to check.
1778  * @compat: bitmask of compatible features
1779  * @ro: bitmask of features that force read-only mount
1780  * @incompat: bitmask of incompatible features
1781  *
1782  * Check whether the journaling code supports the use of
1783  * all of a given set of features on this journal.  Return true
1784  * (non-zero) if it can. */
1785 
1786 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1787 				      unsigned long ro, unsigned long incompat)
1788 {
1789 	if (!compat && !ro && !incompat)
1790 		return 1;
1791 
1792 	/* We can support any known requested features iff the
1793 	 * superblock is in version 2.  Otherwise we fail to support any
1794 	 * extended sb features. */
1795 
1796 	if (journal->j_format_version != 2)
1797 		return 0;
1798 
1799 	if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1800 	    (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1801 	    (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1802 		return 1;
1803 
1804 	return 0;
1805 }
1806 
1807 /**
1808  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1809  * @journal: Journal to act on.
1810  * @compat: bitmask of compatible features
1811  * @ro: bitmask of features that force read-only mount
1812  * @incompat: bitmask of incompatible features
1813  *
1814  * Mark a given journal feature as present on the
1815  * superblock.  Returns true if the requested features could be set.
1816  *
1817  */
1818 
1819 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1820 			  unsigned long ro, unsigned long incompat)
1821 {
1822 #define INCOMPAT_FEATURE_ON(f) \
1823 		((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1824 #define COMPAT_FEATURE_ON(f) \
1825 		((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1826 	journal_superblock_t *sb;
1827 
1828 	if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1829 		return 1;
1830 
1831 	if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1832 		return 0;
1833 
1834 	/* If enabling v2 checksums, turn on v3 instead */
1835 	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1836 		incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1837 		incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1838 	}
1839 
1840 	/* Asking for checksumming v3 and v1?  Only give them v3. */
1841 	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1842 	    compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1843 		compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1844 
1845 	jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1846 		  compat, ro, incompat);
1847 
1848 	sb = journal->j_superblock;
1849 
1850 	/* If enabling v3 checksums, update superblock */
1851 	if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1852 		sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1853 		sb->s_feature_compat &=
1854 			~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1855 
1856 		/* Load the checksum driver */
1857 		if (journal->j_chksum_driver == NULL) {
1858 			journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1859 								      0, 0);
1860 			if (IS_ERR(journal->j_chksum_driver)) {
1861 				printk(KERN_ERR "JBD2: Cannot load crc32c "
1862 				       "driver.\n");
1863 				journal->j_chksum_driver = NULL;
1864 				return 0;
1865 			}
1866 
1867 			/* Precompute checksum seed for all metadata */
1868 			journal->j_csum_seed = jbd2_chksum(journal, ~0,
1869 							   sb->s_uuid,
1870 							   sizeof(sb->s_uuid));
1871 		}
1872 	}
1873 
1874 	/* If enabling v1 checksums, downgrade superblock */
1875 	if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1876 		sb->s_feature_incompat &=
1877 			~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1878 				     JBD2_FEATURE_INCOMPAT_CSUM_V3);
1879 
1880 	sb->s_feature_compat    |= cpu_to_be32(compat);
1881 	sb->s_feature_ro_compat |= cpu_to_be32(ro);
1882 	sb->s_feature_incompat  |= cpu_to_be32(incompat);
1883 
1884 	return 1;
1885 #undef COMPAT_FEATURE_ON
1886 #undef INCOMPAT_FEATURE_ON
1887 }
1888 
1889 /*
1890  * jbd2_journal_clear_features () - Clear a given journal feature in the
1891  * 				    superblock
1892  * @journal: Journal to act on.
1893  * @compat: bitmask of compatible features
1894  * @ro: bitmask of features that force read-only mount
1895  * @incompat: bitmask of incompatible features
1896  *
1897  * Clear a given journal feature as present on the
1898  * superblock.
1899  */
1900 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1901 				unsigned long ro, unsigned long incompat)
1902 {
1903 	journal_superblock_t *sb;
1904 
1905 	jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1906 		  compat, ro, incompat);
1907 
1908 	sb = journal->j_superblock;
1909 
1910 	sb->s_feature_compat    &= ~cpu_to_be32(compat);
1911 	sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1912 	sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1913 }
1914 EXPORT_SYMBOL(jbd2_journal_clear_features);
1915 
1916 /**
1917  * int jbd2_journal_flush () - Flush journal
1918  * @journal: Journal to act on.
1919  *
1920  * Flush all data for a given journal to disk and empty the journal.
1921  * Filesystems can use this when remounting readonly to ensure that
1922  * recovery does not need to happen on remount.
1923  */
1924 
1925 int jbd2_journal_flush(journal_t *journal)
1926 {
1927 	int err = 0;
1928 	transaction_t *transaction = NULL;
1929 
1930 	write_lock(&journal->j_state_lock);
1931 
1932 	/* Force everything buffered to the log... */
1933 	if (journal->j_running_transaction) {
1934 		transaction = journal->j_running_transaction;
1935 		__jbd2_log_start_commit(journal, transaction->t_tid);
1936 	} else if (journal->j_committing_transaction)
1937 		transaction = journal->j_committing_transaction;
1938 
1939 	/* Wait for the log commit to complete... */
1940 	if (transaction) {
1941 		tid_t tid = transaction->t_tid;
1942 
1943 		write_unlock(&journal->j_state_lock);
1944 		jbd2_log_wait_commit(journal, tid);
1945 	} else {
1946 		write_unlock(&journal->j_state_lock);
1947 	}
1948 
1949 	/* ...and flush everything in the log out to disk. */
1950 	spin_lock(&journal->j_list_lock);
1951 	while (!err && journal->j_checkpoint_transactions != NULL) {
1952 		spin_unlock(&journal->j_list_lock);
1953 		mutex_lock(&journal->j_checkpoint_mutex);
1954 		err = jbd2_log_do_checkpoint(journal);
1955 		mutex_unlock(&journal->j_checkpoint_mutex);
1956 		spin_lock(&journal->j_list_lock);
1957 	}
1958 	spin_unlock(&journal->j_list_lock);
1959 
1960 	if (is_journal_aborted(journal))
1961 		return -EIO;
1962 
1963 	mutex_lock(&journal->j_checkpoint_mutex);
1964 	if (!err) {
1965 		err = jbd2_cleanup_journal_tail(journal);
1966 		if (err < 0) {
1967 			mutex_unlock(&journal->j_checkpoint_mutex);
1968 			goto out;
1969 		}
1970 		err = 0;
1971 	}
1972 
1973 	/* Finally, mark the journal as really needing no recovery.
1974 	 * This sets s_start==0 in the underlying superblock, which is
1975 	 * the magic code for a fully-recovered superblock.  Any future
1976 	 * commits of data to the journal will restore the current
1977 	 * s_start value. */
1978 	jbd2_mark_journal_empty(journal);
1979 	mutex_unlock(&journal->j_checkpoint_mutex);
1980 	write_lock(&journal->j_state_lock);
1981 	J_ASSERT(!journal->j_running_transaction);
1982 	J_ASSERT(!journal->j_committing_transaction);
1983 	J_ASSERT(!journal->j_checkpoint_transactions);
1984 	J_ASSERT(journal->j_head == journal->j_tail);
1985 	J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1986 	write_unlock(&journal->j_state_lock);
1987 out:
1988 	return err;
1989 }
1990 
1991 /**
1992  * int jbd2_journal_wipe() - Wipe journal contents
1993  * @journal: Journal to act on.
1994  * @write: flag (see below)
1995  *
1996  * Wipe out all of the contents of a journal, safely.  This will produce
1997  * a warning if the journal contains any valid recovery information.
1998  * Must be called between journal_init_*() and jbd2_journal_load().
1999  *
2000  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2001  * we merely suppress recovery.
2002  */
2003 
2004 int jbd2_journal_wipe(journal_t *journal, int write)
2005 {
2006 	int err = 0;
2007 
2008 	J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2009 
2010 	err = load_superblock(journal);
2011 	if (err)
2012 		return err;
2013 
2014 	if (!journal->j_tail)
2015 		goto no_recovery;
2016 
2017 	printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2018 		write ? "Clearing" : "Ignoring");
2019 
2020 	err = jbd2_journal_skip_recovery(journal);
2021 	if (write) {
2022 		/* Lock to make assertions happy... */
2023 		mutex_lock(&journal->j_checkpoint_mutex);
2024 		jbd2_mark_journal_empty(journal);
2025 		mutex_unlock(&journal->j_checkpoint_mutex);
2026 	}
2027 
2028  no_recovery:
2029 	return err;
2030 }
2031 
2032 /*
2033  * Journal abort has very specific semantics, which we describe
2034  * for journal abort.
2035  *
2036  * Two internal functions, which provide abort to the jbd layer
2037  * itself are here.
2038  */
2039 
2040 /*
2041  * Quick version for internal journal use (doesn't lock the journal).
2042  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2043  * and don't attempt to make any other journal updates.
2044  */
2045 void __jbd2_journal_abort_hard(journal_t *journal)
2046 {
2047 	transaction_t *transaction;
2048 
2049 	if (journal->j_flags & JBD2_ABORT)
2050 		return;
2051 
2052 	printk(KERN_ERR "Aborting journal on device %s.\n",
2053 	       journal->j_devname);
2054 
2055 	write_lock(&journal->j_state_lock);
2056 	journal->j_flags |= JBD2_ABORT;
2057 	transaction = journal->j_running_transaction;
2058 	if (transaction)
2059 		__jbd2_log_start_commit(journal, transaction->t_tid);
2060 	write_unlock(&journal->j_state_lock);
2061 }
2062 
2063 /* Soft abort: record the abort error status in the journal superblock,
2064  * but don't do any other IO. */
2065 static void __journal_abort_soft (journal_t *journal, int errno)
2066 {
2067 	if (journal->j_flags & JBD2_ABORT)
2068 		return;
2069 
2070 	if (!journal->j_errno)
2071 		journal->j_errno = errno;
2072 
2073 	__jbd2_journal_abort_hard(journal);
2074 
2075 	if (errno) {
2076 		jbd2_journal_update_sb_errno(journal);
2077 		write_lock(&journal->j_state_lock);
2078 		journal->j_flags |= JBD2_REC_ERR;
2079 		write_unlock(&journal->j_state_lock);
2080 	}
2081 }
2082 
2083 /**
2084  * void jbd2_journal_abort () - Shutdown the journal immediately.
2085  * @journal: the journal to shutdown.
2086  * @errno:   an error number to record in the journal indicating
2087  *           the reason for the shutdown.
2088  *
2089  * Perform a complete, immediate shutdown of the ENTIRE
2090  * journal (not of a single transaction).  This operation cannot be
2091  * undone without closing and reopening the journal.
2092  *
2093  * The jbd2_journal_abort function is intended to support higher level error
2094  * recovery mechanisms such as the ext2/ext3 remount-readonly error
2095  * mode.
2096  *
2097  * Journal abort has very specific semantics.  Any existing dirty,
2098  * unjournaled buffers in the main filesystem will still be written to
2099  * disk by bdflush, but the journaling mechanism will be suspended
2100  * immediately and no further transaction commits will be honoured.
2101  *
2102  * Any dirty, journaled buffers will be written back to disk without
2103  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2104  * filesystem, but we _do_ attempt to leave as much data as possible
2105  * behind for fsck to use for cleanup.
2106  *
2107  * Any attempt to get a new transaction handle on a journal which is in
2108  * ABORT state will just result in an -EROFS error return.  A
2109  * jbd2_journal_stop on an existing handle will return -EIO if we have
2110  * entered abort state during the update.
2111  *
2112  * Recursive transactions are not disturbed by journal abort until the
2113  * final jbd2_journal_stop, which will receive the -EIO error.
2114  *
2115  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2116  * which will be recorded (if possible) in the journal superblock.  This
2117  * allows a client to record failure conditions in the middle of a
2118  * transaction without having to complete the transaction to record the
2119  * failure to disk.  ext3_error, for example, now uses this
2120  * functionality.
2121  *
2122  * Errors which originate from within the journaling layer will NOT
2123  * supply an errno; a null errno implies that absolutely no further
2124  * writes are done to the journal (unless there are any already in
2125  * progress).
2126  *
2127  */
2128 
2129 void jbd2_journal_abort(journal_t *journal, int errno)
2130 {
2131 	__journal_abort_soft(journal, errno);
2132 }
2133 
2134 /**
2135  * int jbd2_journal_errno () - returns the journal's error state.
2136  * @journal: journal to examine.
2137  *
2138  * This is the errno number set with jbd2_journal_abort(), the last
2139  * time the journal was mounted - if the journal was stopped
2140  * without calling abort this will be 0.
2141  *
2142  * If the journal has been aborted on this mount time -EROFS will
2143  * be returned.
2144  */
2145 int jbd2_journal_errno(journal_t *journal)
2146 {
2147 	int err;
2148 
2149 	read_lock(&journal->j_state_lock);
2150 	if (journal->j_flags & JBD2_ABORT)
2151 		err = -EROFS;
2152 	else
2153 		err = journal->j_errno;
2154 	read_unlock(&journal->j_state_lock);
2155 	return err;
2156 }
2157 
2158 /**
2159  * int jbd2_journal_clear_err () - clears the journal's error state
2160  * @journal: journal to act on.
2161  *
2162  * An error must be cleared or acked to take a FS out of readonly
2163  * mode.
2164  */
2165 int jbd2_journal_clear_err(journal_t *journal)
2166 {
2167 	int err = 0;
2168 
2169 	write_lock(&journal->j_state_lock);
2170 	if (journal->j_flags & JBD2_ABORT)
2171 		err = -EROFS;
2172 	else
2173 		journal->j_errno = 0;
2174 	write_unlock(&journal->j_state_lock);
2175 	return err;
2176 }
2177 
2178 /**
2179  * void jbd2_journal_ack_err() - Ack journal err.
2180  * @journal: journal to act on.
2181  *
2182  * An error must be cleared or acked to take a FS out of readonly
2183  * mode.
2184  */
2185 void jbd2_journal_ack_err(journal_t *journal)
2186 {
2187 	write_lock(&journal->j_state_lock);
2188 	if (journal->j_errno)
2189 		journal->j_flags |= JBD2_ACK_ERR;
2190 	write_unlock(&journal->j_state_lock);
2191 }
2192 
2193 int jbd2_journal_blocks_per_page(struct inode *inode)
2194 {
2195 	return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2196 }
2197 
2198 /*
2199  * helper functions to deal with 32 or 64bit block numbers.
2200  */
2201 size_t journal_tag_bytes(journal_t *journal)
2202 {
2203 	size_t sz;
2204 
2205 	if (jbd2_has_feature_csum3(journal))
2206 		return sizeof(journal_block_tag3_t);
2207 
2208 	sz = sizeof(journal_block_tag_t);
2209 
2210 	if (jbd2_has_feature_csum2(journal))
2211 		sz += sizeof(__u16);
2212 
2213 	if (jbd2_has_feature_64bit(journal))
2214 		return sz;
2215 	else
2216 		return sz - sizeof(__u32);
2217 }
2218 
2219 /*
2220  * JBD memory management
2221  *
2222  * These functions are used to allocate block-sized chunks of memory
2223  * used for making copies of buffer_head data.  Very often it will be
2224  * page-sized chunks of data, but sometimes it will be in
2225  * sub-page-size chunks.  (For example, 16k pages on Power systems
2226  * with a 4k block file system.)  For blocks smaller than a page, we
2227  * use a SLAB allocator.  There are slab caches for each block size,
2228  * which are allocated at mount time, if necessary, and we only free
2229  * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2230  * this reason we don't need to a mutex to protect access to
2231  * jbd2_slab[] allocating or releasing memory; only in
2232  * jbd2_journal_create_slab().
2233  */
2234 #define JBD2_MAX_SLABS 8
2235 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2236 
2237 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2238 	"jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2239 	"jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2240 };
2241 
2242 
2243 static void jbd2_journal_destroy_slabs(void)
2244 {
2245 	int i;
2246 
2247 	for (i = 0; i < JBD2_MAX_SLABS; i++) {
2248 		if (jbd2_slab[i])
2249 			kmem_cache_destroy(jbd2_slab[i]);
2250 		jbd2_slab[i] = NULL;
2251 	}
2252 }
2253 
2254 static int jbd2_journal_create_slab(size_t size)
2255 {
2256 	static DEFINE_MUTEX(jbd2_slab_create_mutex);
2257 	int i = order_base_2(size) - 10;
2258 	size_t slab_size;
2259 
2260 	if (size == PAGE_SIZE)
2261 		return 0;
2262 
2263 	if (i >= JBD2_MAX_SLABS)
2264 		return -EINVAL;
2265 
2266 	if (unlikely(i < 0))
2267 		i = 0;
2268 	mutex_lock(&jbd2_slab_create_mutex);
2269 	if (jbd2_slab[i]) {
2270 		mutex_unlock(&jbd2_slab_create_mutex);
2271 		return 0;	/* Already created */
2272 	}
2273 
2274 	slab_size = 1 << (i+10);
2275 	jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2276 					 slab_size, 0, NULL);
2277 	mutex_unlock(&jbd2_slab_create_mutex);
2278 	if (!jbd2_slab[i]) {
2279 		printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2280 		return -ENOMEM;
2281 	}
2282 	return 0;
2283 }
2284 
2285 static struct kmem_cache *get_slab(size_t size)
2286 {
2287 	int i = order_base_2(size) - 10;
2288 
2289 	BUG_ON(i >= JBD2_MAX_SLABS);
2290 	if (unlikely(i < 0))
2291 		i = 0;
2292 	BUG_ON(jbd2_slab[i] == NULL);
2293 	return jbd2_slab[i];
2294 }
2295 
2296 void *jbd2_alloc(size_t size, gfp_t flags)
2297 {
2298 	void *ptr;
2299 
2300 	BUG_ON(size & (size-1)); /* Must be a power of 2 */
2301 
2302 	flags |= __GFP_REPEAT;
2303 	if (size == PAGE_SIZE)
2304 		ptr = (void *)__get_free_pages(flags, 0);
2305 	else if (size > PAGE_SIZE) {
2306 		int order = get_order(size);
2307 
2308 		if (order < 3)
2309 			ptr = (void *)__get_free_pages(flags, order);
2310 		else
2311 			ptr = vmalloc(size);
2312 	} else
2313 		ptr = kmem_cache_alloc(get_slab(size), flags);
2314 
2315 	/* Check alignment; SLUB has gotten this wrong in the past,
2316 	 * and this can lead to user data corruption! */
2317 	BUG_ON(((unsigned long) ptr) & (size-1));
2318 
2319 	return ptr;
2320 }
2321 
2322 void jbd2_free(void *ptr, size_t size)
2323 {
2324 	if (size == PAGE_SIZE) {
2325 		free_pages((unsigned long)ptr, 0);
2326 		return;
2327 	}
2328 	if (size > PAGE_SIZE) {
2329 		int order = get_order(size);
2330 
2331 		if (order < 3)
2332 			free_pages((unsigned long)ptr, order);
2333 		else
2334 			vfree(ptr);
2335 		return;
2336 	}
2337 	kmem_cache_free(get_slab(size), ptr);
2338 };
2339 
2340 /*
2341  * Journal_head storage management
2342  */
2343 static struct kmem_cache *jbd2_journal_head_cache;
2344 #ifdef CONFIG_JBD2_DEBUG
2345 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2346 #endif
2347 
2348 static int jbd2_journal_init_journal_head_cache(void)
2349 {
2350 	int retval;
2351 
2352 	J_ASSERT(jbd2_journal_head_cache == NULL);
2353 	jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2354 				sizeof(struct journal_head),
2355 				0,		/* offset */
2356 				SLAB_TEMPORARY | SLAB_DESTROY_BY_RCU,
2357 				NULL);		/* ctor */
2358 	retval = 0;
2359 	if (!jbd2_journal_head_cache) {
2360 		retval = -ENOMEM;
2361 		printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2362 	}
2363 	return retval;
2364 }
2365 
2366 static void jbd2_journal_destroy_journal_head_cache(void)
2367 {
2368 	if (jbd2_journal_head_cache) {
2369 		kmem_cache_destroy(jbd2_journal_head_cache);
2370 		jbd2_journal_head_cache = NULL;
2371 	}
2372 }
2373 
2374 /*
2375  * journal_head splicing and dicing
2376  */
2377 static struct journal_head *journal_alloc_journal_head(void)
2378 {
2379 	struct journal_head *ret;
2380 
2381 #ifdef CONFIG_JBD2_DEBUG
2382 	atomic_inc(&nr_journal_heads);
2383 #endif
2384 	ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2385 	if (!ret) {
2386 		jbd_debug(1, "out of memory for journal_head\n");
2387 		pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2388 		ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2389 				GFP_NOFS | __GFP_NOFAIL);
2390 	}
2391 	return ret;
2392 }
2393 
2394 static void journal_free_journal_head(struct journal_head *jh)
2395 {
2396 #ifdef CONFIG_JBD2_DEBUG
2397 	atomic_dec(&nr_journal_heads);
2398 	memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2399 #endif
2400 	kmem_cache_free(jbd2_journal_head_cache, jh);
2401 }
2402 
2403 /*
2404  * A journal_head is attached to a buffer_head whenever JBD has an
2405  * interest in the buffer.
2406  *
2407  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2408  * is set.  This bit is tested in core kernel code where we need to take
2409  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2410  * there.
2411  *
2412  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2413  *
2414  * When a buffer has its BH_JBD bit set it is immune from being released by
2415  * core kernel code, mainly via ->b_count.
2416  *
2417  * A journal_head is detached from its buffer_head when the journal_head's
2418  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2419  * transaction (b_cp_transaction) hold their references to b_jcount.
2420  *
2421  * Various places in the kernel want to attach a journal_head to a buffer_head
2422  * _before_ attaching the journal_head to a transaction.  To protect the
2423  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2424  * journal_head's b_jcount refcount by one.  The caller must call
2425  * jbd2_journal_put_journal_head() to undo this.
2426  *
2427  * So the typical usage would be:
2428  *
2429  *	(Attach a journal_head if needed.  Increments b_jcount)
2430  *	struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2431  *	...
2432  *      (Get another reference for transaction)
2433  *	jbd2_journal_grab_journal_head(bh);
2434  *	jh->b_transaction = xxx;
2435  *	(Put original reference)
2436  *	jbd2_journal_put_journal_head(jh);
2437  */
2438 
2439 /*
2440  * Give a buffer_head a journal_head.
2441  *
2442  * May sleep.
2443  */
2444 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2445 {
2446 	struct journal_head *jh;
2447 	struct journal_head *new_jh = NULL;
2448 
2449 repeat:
2450 	if (!buffer_jbd(bh))
2451 		new_jh = journal_alloc_journal_head();
2452 
2453 	jbd_lock_bh_journal_head(bh);
2454 	if (buffer_jbd(bh)) {
2455 		jh = bh2jh(bh);
2456 	} else {
2457 		J_ASSERT_BH(bh,
2458 			(atomic_read(&bh->b_count) > 0) ||
2459 			(bh->b_page && bh->b_page->mapping));
2460 
2461 		if (!new_jh) {
2462 			jbd_unlock_bh_journal_head(bh);
2463 			goto repeat;
2464 		}
2465 
2466 		jh = new_jh;
2467 		new_jh = NULL;		/* We consumed it */
2468 		set_buffer_jbd(bh);
2469 		bh->b_private = jh;
2470 		jh->b_bh = bh;
2471 		get_bh(bh);
2472 		BUFFER_TRACE(bh, "added journal_head");
2473 	}
2474 	jh->b_jcount++;
2475 	jbd_unlock_bh_journal_head(bh);
2476 	if (new_jh)
2477 		journal_free_journal_head(new_jh);
2478 	return bh->b_private;
2479 }
2480 
2481 /*
2482  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2483  * having a journal_head, return NULL
2484  */
2485 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2486 {
2487 	struct journal_head *jh = NULL;
2488 
2489 	jbd_lock_bh_journal_head(bh);
2490 	if (buffer_jbd(bh)) {
2491 		jh = bh2jh(bh);
2492 		jh->b_jcount++;
2493 	}
2494 	jbd_unlock_bh_journal_head(bh);
2495 	return jh;
2496 }
2497 
2498 static void __journal_remove_journal_head(struct buffer_head *bh)
2499 {
2500 	struct journal_head *jh = bh2jh(bh);
2501 
2502 	J_ASSERT_JH(jh, jh->b_jcount >= 0);
2503 	J_ASSERT_JH(jh, jh->b_transaction == NULL);
2504 	J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2505 	J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2506 	J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2507 	J_ASSERT_BH(bh, buffer_jbd(bh));
2508 	J_ASSERT_BH(bh, jh2bh(jh) == bh);
2509 	BUFFER_TRACE(bh, "remove journal_head");
2510 	if (jh->b_frozen_data) {
2511 		printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2512 		jbd2_free(jh->b_frozen_data, bh->b_size);
2513 	}
2514 	if (jh->b_committed_data) {
2515 		printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2516 		jbd2_free(jh->b_committed_data, bh->b_size);
2517 	}
2518 	bh->b_private = NULL;
2519 	jh->b_bh = NULL;	/* debug, really */
2520 	clear_buffer_jbd(bh);
2521 	journal_free_journal_head(jh);
2522 }
2523 
2524 /*
2525  * Drop a reference on the passed journal_head.  If it fell to zero then
2526  * release the journal_head from the buffer_head.
2527  */
2528 void jbd2_journal_put_journal_head(struct journal_head *jh)
2529 {
2530 	struct buffer_head *bh = jh2bh(jh);
2531 
2532 	jbd_lock_bh_journal_head(bh);
2533 	J_ASSERT_JH(jh, jh->b_jcount > 0);
2534 	--jh->b_jcount;
2535 	if (!jh->b_jcount) {
2536 		__journal_remove_journal_head(bh);
2537 		jbd_unlock_bh_journal_head(bh);
2538 		__brelse(bh);
2539 	} else
2540 		jbd_unlock_bh_journal_head(bh);
2541 }
2542 
2543 /*
2544  * Initialize jbd inode head
2545  */
2546 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2547 {
2548 	jinode->i_transaction = NULL;
2549 	jinode->i_next_transaction = NULL;
2550 	jinode->i_vfs_inode = inode;
2551 	jinode->i_flags = 0;
2552 	INIT_LIST_HEAD(&jinode->i_list);
2553 }
2554 
2555 /*
2556  * Function to be called before we start removing inode from memory (i.e.,
2557  * clear_inode() is a fine place to be called from). It removes inode from
2558  * transaction's lists.
2559  */
2560 void jbd2_journal_release_jbd_inode(journal_t *journal,
2561 				    struct jbd2_inode *jinode)
2562 {
2563 	if (!journal)
2564 		return;
2565 restart:
2566 	spin_lock(&journal->j_list_lock);
2567 	/* Is commit writing out inode - we have to wait */
2568 	if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2569 		wait_queue_head_t *wq;
2570 		DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2571 		wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2572 		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2573 		spin_unlock(&journal->j_list_lock);
2574 		schedule();
2575 		finish_wait(wq, &wait.wait);
2576 		goto restart;
2577 	}
2578 
2579 	if (jinode->i_transaction) {
2580 		list_del(&jinode->i_list);
2581 		jinode->i_transaction = NULL;
2582 	}
2583 	spin_unlock(&journal->j_list_lock);
2584 }
2585 
2586 
2587 #ifdef CONFIG_PROC_FS
2588 
2589 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2590 
2591 static void __init jbd2_create_jbd_stats_proc_entry(void)
2592 {
2593 	proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2594 }
2595 
2596 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2597 {
2598 	if (proc_jbd2_stats)
2599 		remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2600 }
2601 
2602 #else
2603 
2604 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2605 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2606 
2607 #endif
2608 
2609 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2610 
2611 static int __init jbd2_journal_init_handle_cache(void)
2612 {
2613 	jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2614 	if (jbd2_handle_cache == NULL) {
2615 		printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2616 		return -ENOMEM;
2617 	}
2618 	jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2619 	if (jbd2_inode_cache == NULL) {
2620 		printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2621 		kmem_cache_destroy(jbd2_handle_cache);
2622 		return -ENOMEM;
2623 	}
2624 	return 0;
2625 }
2626 
2627 static void jbd2_journal_destroy_handle_cache(void)
2628 {
2629 	if (jbd2_handle_cache)
2630 		kmem_cache_destroy(jbd2_handle_cache);
2631 	if (jbd2_inode_cache)
2632 		kmem_cache_destroy(jbd2_inode_cache);
2633 
2634 }
2635 
2636 /*
2637  * Module startup and shutdown
2638  */
2639 
2640 static int __init journal_init_caches(void)
2641 {
2642 	int ret;
2643 
2644 	ret = jbd2_journal_init_revoke_caches();
2645 	if (ret == 0)
2646 		ret = jbd2_journal_init_journal_head_cache();
2647 	if (ret == 0)
2648 		ret = jbd2_journal_init_handle_cache();
2649 	if (ret == 0)
2650 		ret = jbd2_journal_init_transaction_cache();
2651 	return ret;
2652 }
2653 
2654 static void jbd2_journal_destroy_caches(void)
2655 {
2656 	jbd2_journal_destroy_revoke_caches();
2657 	jbd2_journal_destroy_journal_head_cache();
2658 	jbd2_journal_destroy_handle_cache();
2659 	jbd2_journal_destroy_transaction_cache();
2660 	jbd2_journal_destroy_slabs();
2661 }
2662 
2663 static int __init journal_init(void)
2664 {
2665 	int ret;
2666 
2667 	BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2668 
2669 	ret = journal_init_caches();
2670 	if (ret == 0) {
2671 		jbd2_create_jbd_stats_proc_entry();
2672 	} else {
2673 		jbd2_journal_destroy_caches();
2674 	}
2675 	return ret;
2676 }
2677 
2678 static void __exit journal_exit(void)
2679 {
2680 #ifdef CONFIG_JBD2_DEBUG
2681 	int n = atomic_read(&nr_journal_heads);
2682 	if (n)
2683 		printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2684 #endif
2685 	jbd2_remove_jbd_stats_proc_entry();
2686 	jbd2_journal_destroy_caches();
2687 }
2688 
2689 MODULE_LICENSE("GPL");
2690 module_init(journal_init);
2691 module_exit(journal_exit);
2692 
2693