xref: /linux/fs/jbd2/journal.c (revision 827634added7f38b7d724cab1dccdb2b004c13c3)
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(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 retry_alloc:
375 	new_bh = alloc_buffer_head(GFP_NOFS);
376 	if (!new_bh) {
377 		/*
378 		 * Failure is not an option, but __GFP_NOFAIL is going
379 		 * away; so we retry ourselves here.
380 		 */
381 		congestion_wait(BLK_RW_ASYNC, HZ/50);
382 		goto retry_alloc;
383 	}
384 
385 	/* keep subsequent assertions sane */
386 	atomic_set(&new_bh->b_count, 1);
387 
388 	jbd_lock_bh_state(bh_in);
389 repeat:
390 	/*
391 	 * If a new transaction has already done a buffer copy-out, then
392 	 * we use that version of the data for the commit.
393 	 */
394 	if (jh_in->b_frozen_data) {
395 		done_copy_out = 1;
396 		new_page = virt_to_page(jh_in->b_frozen_data);
397 		new_offset = offset_in_page(jh_in->b_frozen_data);
398 	} else {
399 		new_page = jh2bh(jh_in)->b_page;
400 		new_offset = offset_in_page(jh2bh(jh_in)->b_data);
401 	}
402 
403 	mapped_data = kmap_atomic(new_page);
404 	/*
405 	 * Fire data frozen trigger if data already wasn't frozen.  Do this
406 	 * before checking for escaping, as the trigger may modify the magic
407 	 * offset.  If a copy-out happens afterwards, it will have the correct
408 	 * data in the buffer.
409 	 */
410 	if (!done_copy_out)
411 		jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
412 					   jh_in->b_triggers);
413 
414 	/*
415 	 * Check for escaping
416 	 */
417 	if (*((__be32 *)(mapped_data + new_offset)) ==
418 				cpu_to_be32(JBD2_MAGIC_NUMBER)) {
419 		need_copy_out = 1;
420 		do_escape = 1;
421 	}
422 	kunmap_atomic(mapped_data);
423 
424 	/*
425 	 * Do we need to do a data copy?
426 	 */
427 	if (need_copy_out && !done_copy_out) {
428 		char *tmp;
429 
430 		jbd_unlock_bh_state(bh_in);
431 		tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
432 		if (!tmp) {
433 			brelse(new_bh);
434 			return -ENOMEM;
435 		}
436 		jbd_lock_bh_state(bh_in);
437 		if (jh_in->b_frozen_data) {
438 			jbd2_free(tmp, bh_in->b_size);
439 			goto repeat;
440 		}
441 
442 		jh_in->b_frozen_data = tmp;
443 		mapped_data = kmap_atomic(new_page);
444 		memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
445 		kunmap_atomic(mapped_data);
446 
447 		new_page = virt_to_page(tmp);
448 		new_offset = offset_in_page(tmp);
449 		done_copy_out = 1;
450 
451 		/*
452 		 * This isn't strictly necessary, as we're using frozen
453 		 * data for the escaping, but it keeps consistency with
454 		 * b_frozen_data usage.
455 		 */
456 		jh_in->b_frozen_triggers = jh_in->b_triggers;
457 	}
458 
459 	/*
460 	 * Did we need to do an escaping?  Now we've done all the
461 	 * copying, we can finally do so.
462 	 */
463 	if (do_escape) {
464 		mapped_data = kmap_atomic(new_page);
465 		*((unsigned int *)(mapped_data + new_offset)) = 0;
466 		kunmap_atomic(mapped_data);
467 	}
468 
469 	set_bh_page(new_bh, new_page, new_offset);
470 	new_bh->b_size = bh_in->b_size;
471 	new_bh->b_bdev = journal->j_dev;
472 	new_bh->b_blocknr = blocknr;
473 	new_bh->b_private = bh_in;
474 	set_buffer_mapped(new_bh);
475 	set_buffer_dirty(new_bh);
476 
477 	*bh_out = new_bh;
478 
479 	/*
480 	 * The to-be-written buffer needs to get moved to the io queue,
481 	 * and the original buffer whose contents we are shadowing or
482 	 * copying is moved to the transaction's shadow queue.
483 	 */
484 	JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
485 	spin_lock(&journal->j_list_lock);
486 	__jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
487 	spin_unlock(&journal->j_list_lock);
488 	set_buffer_shadow(bh_in);
489 	jbd_unlock_bh_state(bh_in);
490 
491 	return do_escape | (done_copy_out << 1);
492 }
493 
494 /*
495  * Allocation code for the journal file.  Manage the space left in the
496  * journal, so that we can begin checkpointing when appropriate.
497  */
498 
499 /*
500  * Called with j_state_lock locked for writing.
501  * Returns true if a transaction commit was started.
502  */
503 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
504 {
505 	/* Return if the txn has already requested to be committed */
506 	if (journal->j_commit_request == target)
507 		return 0;
508 
509 	/*
510 	 * The only transaction we can possibly wait upon is the
511 	 * currently running transaction (if it exists).  Otherwise,
512 	 * the target tid must be an old one.
513 	 */
514 	if (journal->j_running_transaction &&
515 	    journal->j_running_transaction->t_tid == target) {
516 		/*
517 		 * We want a new commit: OK, mark the request and wakeup the
518 		 * commit thread.  We do _not_ do the commit ourselves.
519 		 */
520 
521 		journal->j_commit_request = target;
522 		jbd_debug(1, "JBD2: requesting commit %d/%d\n",
523 			  journal->j_commit_request,
524 			  journal->j_commit_sequence);
525 		journal->j_running_transaction->t_requested = jiffies;
526 		wake_up(&journal->j_wait_commit);
527 		return 1;
528 	} else if (!tid_geq(journal->j_commit_request, target))
529 		/* This should never happen, but if it does, preserve
530 		   the evidence before kjournald goes into a loop and
531 		   increments j_commit_sequence beyond all recognition. */
532 		WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
533 			  journal->j_commit_request,
534 			  journal->j_commit_sequence,
535 			  target, journal->j_running_transaction ?
536 			  journal->j_running_transaction->t_tid : 0);
537 	return 0;
538 }
539 
540 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
541 {
542 	int ret;
543 
544 	write_lock(&journal->j_state_lock);
545 	ret = __jbd2_log_start_commit(journal, tid);
546 	write_unlock(&journal->j_state_lock);
547 	return ret;
548 }
549 
550 /*
551  * Force and wait any uncommitted transactions.  We can only force the running
552  * transaction if we don't have an active handle, otherwise, we will deadlock.
553  * Returns: <0 in case of error,
554  *           0 if nothing to commit,
555  *           1 if transaction was successfully committed.
556  */
557 static int __jbd2_journal_force_commit(journal_t *journal)
558 {
559 	transaction_t *transaction = NULL;
560 	tid_t tid;
561 	int need_to_start = 0, ret = 0;
562 
563 	read_lock(&journal->j_state_lock);
564 	if (journal->j_running_transaction && !current->journal_info) {
565 		transaction = journal->j_running_transaction;
566 		if (!tid_geq(journal->j_commit_request, transaction->t_tid))
567 			need_to_start = 1;
568 	} else if (journal->j_committing_transaction)
569 		transaction = journal->j_committing_transaction;
570 
571 	if (!transaction) {
572 		/* Nothing to commit */
573 		read_unlock(&journal->j_state_lock);
574 		return 0;
575 	}
576 	tid = transaction->t_tid;
577 	read_unlock(&journal->j_state_lock);
578 	if (need_to_start)
579 		jbd2_log_start_commit(journal, tid);
580 	ret = jbd2_log_wait_commit(journal, tid);
581 	if (!ret)
582 		ret = 1;
583 
584 	return ret;
585 }
586 
587 /**
588  * Force and wait upon a commit if the calling process is not within
589  * transaction.  This is used for forcing out undo-protected data which contains
590  * bitmaps, when the fs is running out of space.
591  *
592  * @journal: journal to force
593  * Returns true if progress was made.
594  */
595 int jbd2_journal_force_commit_nested(journal_t *journal)
596 {
597 	int ret;
598 
599 	ret = __jbd2_journal_force_commit(journal);
600 	return ret > 0;
601 }
602 
603 /**
604  * int journal_force_commit() - force any uncommitted transactions
605  * @journal: journal to force
606  *
607  * Caller want unconditional commit. We can only force the running transaction
608  * if we don't have an active handle, otherwise, we will deadlock.
609  */
610 int jbd2_journal_force_commit(journal_t *journal)
611 {
612 	int ret;
613 
614 	J_ASSERT(!current->journal_info);
615 	ret = __jbd2_journal_force_commit(journal);
616 	if (ret > 0)
617 		ret = 0;
618 	return ret;
619 }
620 
621 /*
622  * Start a commit of the current running transaction (if any).  Returns true
623  * if a transaction is going to be committed (or is currently already
624  * committing), and fills its tid in at *ptid
625  */
626 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
627 {
628 	int ret = 0;
629 
630 	write_lock(&journal->j_state_lock);
631 	if (journal->j_running_transaction) {
632 		tid_t tid = journal->j_running_transaction->t_tid;
633 
634 		__jbd2_log_start_commit(journal, tid);
635 		/* There's a running transaction and we've just made sure
636 		 * it's commit has been scheduled. */
637 		if (ptid)
638 			*ptid = tid;
639 		ret = 1;
640 	} else if (journal->j_committing_transaction) {
641 		/*
642 		 * If commit has been started, then we have to wait for
643 		 * completion of that transaction.
644 		 */
645 		if (ptid)
646 			*ptid = journal->j_committing_transaction->t_tid;
647 		ret = 1;
648 	}
649 	write_unlock(&journal->j_state_lock);
650 	return ret;
651 }
652 
653 /*
654  * Return 1 if a given transaction has not yet sent barrier request
655  * connected with a transaction commit. If 0 is returned, transaction
656  * may or may not have sent the barrier. Used to avoid sending barrier
657  * twice in common cases.
658  */
659 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
660 {
661 	int ret = 0;
662 	transaction_t *commit_trans;
663 
664 	if (!(journal->j_flags & JBD2_BARRIER))
665 		return 0;
666 	read_lock(&journal->j_state_lock);
667 	/* Transaction already committed? */
668 	if (tid_geq(journal->j_commit_sequence, tid))
669 		goto out;
670 	commit_trans = journal->j_committing_transaction;
671 	if (!commit_trans || commit_trans->t_tid != tid) {
672 		ret = 1;
673 		goto out;
674 	}
675 	/*
676 	 * Transaction is being committed and we already proceeded to
677 	 * submitting a flush to fs partition?
678 	 */
679 	if (journal->j_fs_dev != journal->j_dev) {
680 		if (!commit_trans->t_need_data_flush ||
681 		    commit_trans->t_state >= T_COMMIT_DFLUSH)
682 			goto out;
683 	} else {
684 		if (commit_trans->t_state >= T_COMMIT_JFLUSH)
685 			goto out;
686 	}
687 	ret = 1;
688 out:
689 	read_unlock(&journal->j_state_lock);
690 	return ret;
691 }
692 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
693 
694 /*
695  * Wait for a specified commit to complete.
696  * The caller may not hold the journal lock.
697  */
698 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
699 {
700 	int err = 0;
701 
702 	read_lock(&journal->j_state_lock);
703 #ifdef CONFIG_JBD2_DEBUG
704 	if (!tid_geq(journal->j_commit_request, tid)) {
705 		printk(KERN_ERR
706 		       "%s: error: j_commit_request=%d, tid=%d\n",
707 		       __func__, journal->j_commit_request, tid);
708 	}
709 #endif
710 	while (tid_gt(tid, journal->j_commit_sequence)) {
711 		jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
712 				  tid, journal->j_commit_sequence);
713 		read_unlock(&journal->j_state_lock);
714 		wake_up(&journal->j_wait_commit);
715 		wait_event(journal->j_wait_done_commit,
716 				!tid_gt(tid, journal->j_commit_sequence));
717 		read_lock(&journal->j_state_lock);
718 	}
719 	read_unlock(&journal->j_state_lock);
720 
721 	if (unlikely(is_journal_aborted(journal)))
722 		err = -EIO;
723 	return err;
724 }
725 
726 /*
727  * When this function returns the transaction corresponding to tid
728  * will be completed.  If the transaction has currently running, start
729  * committing that transaction before waiting for it to complete.  If
730  * the transaction id is stale, it is by definition already completed,
731  * so just return SUCCESS.
732  */
733 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
734 {
735 	int	need_to_wait = 1;
736 
737 	read_lock(&journal->j_state_lock);
738 	if (journal->j_running_transaction &&
739 	    journal->j_running_transaction->t_tid == tid) {
740 		if (journal->j_commit_request != tid) {
741 			/* transaction not yet started, so request it */
742 			read_unlock(&journal->j_state_lock);
743 			jbd2_log_start_commit(journal, tid);
744 			goto wait_commit;
745 		}
746 	} else if (!(journal->j_committing_transaction &&
747 		     journal->j_committing_transaction->t_tid == tid))
748 		need_to_wait = 0;
749 	read_unlock(&journal->j_state_lock);
750 	if (!need_to_wait)
751 		return 0;
752 wait_commit:
753 	return jbd2_log_wait_commit(journal, tid);
754 }
755 EXPORT_SYMBOL(jbd2_complete_transaction);
756 
757 /*
758  * Log buffer allocation routines:
759  */
760 
761 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
762 {
763 	unsigned long blocknr;
764 
765 	write_lock(&journal->j_state_lock);
766 	J_ASSERT(journal->j_free > 1);
767 
768 	blocknr = journal->j_head;
769 	journal->j_head++;
770 	journal->j_free--;
771 	if (journal->j_head == journal->j_last)
772 		journal->j_head = journal->j_first;
773 	write_unlock(&journal->j_state_lock);
774 	return jbd2_journal_bmap(journal, blocknr, retp);
775 }
776 
777 /*
778  * Conversion of logical to physical block numbers for the journal
779  *
780  * On external journals the journal blocks are identity-mapped, so
781  * this is a no-op.  If needed, we can use j_blk_offset - everything is
782  * ready.
783  */
784 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
785 		 unsigned long long *retp)
786 {
787 	int err = 0;
788 	unsigned long long ret;
789 
790 	if (journal->j_inode) {
791 		ret = bmap(journal->j_inode, blocknr);
792 		if (ret)
793 			*retp = ret;
794 		else {
795 			printk(KERN_ALERT "%s: journal block not found "
796 					"at offset %lu on %s\n",
797 			       __func__, blocknr, journal->j_devname);
798 			err = -EIO;
799 			__journal_abort_soft(journal, err);
800 		}
801 	} else {
802 		*retp = blocknr; /* +journal->j_blk_offset */
803 	}
804 	return err;
805 }
806 
807 /*
808  * We play buffer_head aliasing tricks to write data/metadata blocks to
809  * the journal without copying their contents, but for journal
810  * descriptor blocks we do need to generate bona fide buffers.
811  *
812  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
813  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
814  * But we don't bother doing that, so there will be coherency problems with
815  * mmaps of blockdevs which hold live JBD-controlled filesystems.
816  */
817 struct buffer_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
818 {
819 	struct buffer_head *bh;
820 	unsigned long long blocknr;
821 	int err;
822 
823 	err = jbd2_journal_next_log_block(journal, &blocknr);
824 
825 	if (err)
826 		return NULL;
827 
828 	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
829 	if (!bh)
830 		return NULL;
831 	lock_buffer(bh);
832 	memset(bh->b_data, 0, journal->j_blocksize);
833 	set_buffer_uptodate(bh);
834 	unlock_buffer(bh);
835 	BUFFER_TRACE(bh, "return this buffer");
836 	return bh;
837 }
838 
839 /*
840  * Return tid of the oldest transaction in the journal and block in the journal
841  * where the transaction starts.
842  *
843  * If the journal is now empty, return which will be the next transaction ID
844  * we will write and where will that transaction start.
845  *
846  * The return value is 0 if journal tail cannot be pushed any further, 1 if
847  * it can.
848  */
849 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
850 			      unsigned long *block)
851 {
852 	transaction_t *transaction;
853 	int ret;
854 
855 	read_lock(&journal->j_state_lock);
856 	spin_lock(&journal->j_list_lock);
857 	transaction = journal->j_checkpoint_transactions;
858 	if (transaction) {
859 		*tid = transaction->t_tid;
860 		*block = transaction->t_log_start;
861 	} else if ((transaction = journal->j_committing_transaction) != NULL) {
862 		*tid = transaction->t_tid;
863 		*block = transaction->t_log_start;
864 	} else if ((transaction = journal->j_running_transaction) != NULL) {
865 		*tid = transaction->t_tid;
866 		*block = journal->j_head;
867 	} else {
868 		*tid = journal->j_transaction_sequence;
869 		*block = journal->j_head;
870 	}
871 	ret = tid_gt(*tid, journal->j_tail_sequence);
872 	spin_unlock(&journal->j_list_lock);
873 	read_unlock(&journal->j_state_lock);
874 
875 	return ret;
876 }
877 
878 /*
879  * Update information in journal structure and in on disk journal superblock
880  * about log tail. This function does not check whether information passed in
881  * really pushes log tail further. It's responsibility of the caller to make
882  * sure provided log tail information is valid (e.g. by holding
883  * j_checkpoint_mutex all the time between computing log tail and calling this
884  * function as is the case with jbd2_cleanup_journal_tail()).
885  *
886  * Requires j_checkpoint_mutex
887  */
888 void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
889 {
890 	unsigned long freed;
891 
892 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
893 
894 	/*
895 	 * We cannot afford for write to remain in drive's caches since as
896 	 * soon as we update j_tail, next transaction can start reusing journal
897 	 * space and if we lose sb update during power failure we'd replay
898 	 * old transaction with possibly newly overwritten data.
899 	 */
900 	jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
901 	write_lock(&journal->j_state_lock);
902 	freed = block - journal->j_tail;
903 	if (block < journal->j_tail)
904 		freed += journal->j_last - journal->j_first;
905 
906 	trace_jbd2_update_log_tail(journal, tid, block, freed);
907 	jbd_debug(1,
908 		  "Cleaning journal tail from %d to %d (offset %lu), "
909 		  "freeing %lu\n",
910 		  journal->j_tail_sequence, tid, block, freed);
911 
912 	journal->j_free += freed;
913 	journal->j_tail_sequence = tid;
914 	journal->j_tail = block;
915 	write_unlock(&journal->j_state_lock);
916 }
917 
918 /*
919  * This is a variaon of __jbd2_update_log_tail which checks for validity of
920  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
921  * with other threads updating log tail.
922  */
923 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
924 {
925 	mutex_lock(&journal->j_checkpoint_mutex);
926 	if (tid_gt(tid, journal->j_tail_sequence))
927 		__jbd2_update_log_tail(journal, tid, block);
928 	mutex_unlock(&journal->j_checkpoint_mutex);
929 }
930 
931 struct jbd2_stats_proc_session {
932 	journal_t *journal;
933 	struct transaction_stats_s *stats;
934 	int start;
935 	int max;
936 };
937 
938 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
939 {
940 	return *pos ? NULL : SEQ_START_TOKEN;
941 }
942 
943 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
944 {
945 	return NULL;
946 }
947 
948 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
949 {
950 	struct jbd2_stats_proc_session *s = seq->private;
951 
952 	if (v != SEQ_START_TOKEN)
953 		return 0;
954 	seq_printf(seq, "%lu transactions (%lu requested), "
955 		   "each up to %u blocks\n",
956 		   s->stats->ts_tid, s->stats->ts_requested,
957 		   s->journal->j_max_transaction_buffers);
958 	if (s->stats->ts_tid == 0)
959 		return 0;
960 	seq_printf(seq, "average: \n  %ums waiting for transaction\n",
961 	    jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
962 	seq_printf(seq, "  %ums request delay\n",
963 	    (s->stats->ts_requested == 0) ? 0 :
964 	    jiffies_to_msecs(s->stats->run.rs_request_delay /
965 			     s->stats->ts_requested));
966 	seq_printf(seq, "  %ums running transaction\n",
967 	    jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
968 	seq_printf(seq, "  %ums transaction was being locked\n",
969 	    jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
970 	seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
971 	    jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
972 	seq_printf(seq, "  %ums logging transaction\n",
973 	    jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
974 	seq_printf(seq, "  %lluus average transaction commit time\n",
975 		   div_u64(s->journal->j_average_commit_time, 1000));
976 	seq_printf(seq, "  %lu handles per transaction\n",
977 	    s->stats->run.rs_handle_count / s->stats->ts_tid);
978 	seq_printf(seq, "  %lu blocks per transaction\n",
979 	    s->stats->run.rs_blocks / s->stats->ts_tid);
980 	seq_printf(seq, "  %lu logged blocks per transaction\n",
981 	    s->stats->run.rs_blocks_logged / s->stats->ts_tid);
982 	return 0;
983 }
984 
985 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
986 {
987 }
988 
989 static const struct seq_operations jbd2_seq_info_ops = {
990 	.start  = jbd2_seq_info_start,
991 	.next   = jbd2_seq_info_next,
992 	.stop   = jbd2_seq_info_stop,
993 	.show   = jbd2_seq_info_show,
994 };
995 
996 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
997 {
998 	journal_t *journal = PDE_DATA(inode);
999 	struct jbd2_stats_proc_session *s;
1000 	int rc, size;
1001 
1002 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1003 	if (s == NULL)
1004 		return -ENOMEM;
1005 	size = sizeof(struct transaction_stats_s);
1006 	s->stats = kmalloc(size, GFP_KERNEL);
1007 	if (s->stats == NULL) {
1008 		kfree(s);
1009 		return -ENOMEM;
1010 	}
1011 	spin_lock(&journal->j_history_lock);
1012 	memcpy(s->stats, &journal->j_stats, size);
1013 	s->journal = journal;
1014 	spin_unlock(&journal->j_history_lock);
1015 
1016 	rc = seq_open(file, &jbd2_seq_info_ops);
1017 	if (rc == 0) {
1018 		struct seq_file *m = file->private_data;
1019 		m->private = s;
1020 	} else {
1021 		kfree(s->stats);
1022 		kfree(s);
1023 	}
1024 	return rc;
1025 
1026 }
1027 
1028 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1029 {
1030 	struct seq_file *seq = file->private_data;
1031 	struct jbd2_stats_proc_session *s = seq->private;
1032 	kfree(s->stats);
1033 	kfree(s);
1034 	return seq_release(inode, file);
1035 }
1036 
1037 static const struct file_operations jbd2_seq_info_fops = {
1038 	.owner		= THIS_MODULE,
1039 	.open           = jbd2_seq_info_open,
1040 	.read           = seq_read,
1041 	.llseek         = seq_lseek,
1042 	.release        = jbd2_seq_info_release,
1043 };
1044 
1045 static struct proc_dir_entry *proc_jbd2_stats;
1046 
1047 static void jbd2_stats_proc_init(journal_t *journal)
1048 {
1049 	journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1050 	if (journal->j_proc_entry) {
1051 		proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1052 				 &jbd2_seq_info_fops, journal);
1053 	}
1054 }
1055 
1056 static void jbd2_stats_proc_exit(journal_t *journal)
1057 {
1058 	remove_proc_entry("info", journal->j_proc_entry);
1059 	remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1060 }
1061 
1062 /*
1063  * Management for journal control blocks: functions to create and
1064  * destroy journal_t structures, and to initialise and read existing
1065  * journal blocks from disk.  */
1066 
1067 /* First: create and setup a journal_t object in memory.  We initialise
1068  * very few fields yet: that has to wait until we have created the
1069  * journal structures from from scratch, or loaded them from disk. */
1070 
1071 static journal_t * journal_init_common (void)
1072 {
1073 	journal_t *journal;
1074 	int err;
1075 
1076 	journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1077 	if (!journal)
1078 		return NULL;
1079 
1080 	init_waitqueue_head(&journal->j_wait_transaction_locked);
1081 	init_waitqueue_head(&journal->j_wait_done_commit);
1082 	init_waitqueue_head(&journal->j_wait_commit);
1083 	init_waitqueue_head(&journal->j_wait_updates);
1084 	init_waitqueue_head(&journal->j_wait_reserved);
1085 	mutex_init(&journal->j_barrier);
1086 	mutex_init(&journal->j_checkpoint_mutex);
1087 	spin_lock_init(&journal->j_revoke_lock);
1088 	spin_lock_init(&journal->j_list_lock);
1089 	rwlock_init(&journal->j_state_lock);
1090 
1091 	journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1092 	journal->j_min_batch_time = 0;
1093 	journal->j_max_batch_time = 15000; /* 15ms */
1094 	atomic_set(&journal->j_reserved_credits, 0);
1095 
1096 	/* The journal is marked for error until we succeed with recovery! */
1097 	journal->j_flags = JBD2_ABORT;
1098 
1099 	/* Set up a default-sized revoke table for the new mount. */
1100 	err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1101 	if (err) {
1102 		kfree(journal);
1103 		return NULL;
1104 	}
1105 
1106 	spin_lock_init(&journal->j_history_lock);
1107 
1108 	return journal;
1109 }
1110 
1111 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1112  *
1113  * Create a journal structure assigned some fixed set of disk blocks to
1114  * the journal.  We don't actually touch those disk blocks yet, but we
1115  * need to set up all of the mapping information to tell the journaling
1116  * system where the journal blocks are.
1117  *
1118  */
1119 
1120 /**
1121  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1122  *  @bdev: Block device on which to create the journal
1123  *  @fs_dev: Device which hold journalled filesystem for this journal.
1124  *  @start: Block nr Start of journal.
1125  *  @len:  Length of the journal in blocks.
1126  *  @blocksize: blocksize of journalling device
1127  *
1128  *  Returns: a newly created journal_t *
1129  *
1130  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1131  *  range of blocks on an arbitrary block device.
1132  *
1133  */
1134 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1135 			struct block_device *fs_dev,
1136 			unsigned long long start, int len, int blocksize)
1137 {
1138 	journal_t *journal = journal_init_common();
1139 	struct buffer_head *bh;
1140 	char *p;
1141 	int n;
1142 
1143 	if (!journal)
1144 		return NULL;
1145 
1146 	/* journal descriptor can store up to n blocks -bzzz */
1147 	journal->j_blocksize = blocksize;
1148 	journal->j_dev = bdev;
1149 	journal->j_fs_dev = fs_dev;
1150 	journal->j_blk_offset = start;
1151 	journal->j_maxlen = len;
1152 	bdevname(journal->j_dev, journal->j_devname);
1153 	p = journal->j_devname;
1154 	while ((p = strchr(p, '/')))
1155 		*p = '!';
1156 	jbd2_stats_proc_init(journal);
1157 	n = journal->j_blocksize / sizeof(journal_block_tag_t);
1158 	journal->j_wbufsize = n;
1159 	journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1160 	if (!journal->j_wbuf) {
1161 		printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1162 			__func__);
1163 		goto out_err;
1164 	}
1165 
1166 	bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1167 	if (!bh) {
1168 		printk(KERN_ERR
1169 		       "%s: Cannot get buffer for journal superblock\n",
1170 		       __func__);
1171 		goto out_err;
1172 	}
1173 	journal->j_sb_buffer = bh;
1174 	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1175 
1176 	return journal;
1177 out_err:
1178 	kfree(journal->j_wbuf);
1179 	jbd2_stats_proc_exit(journal);
1180 	kfree(journal);
1181 	return NULL;
1182 }
1183 
1184 /**
1185  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1186  *  @inode: An inode to create the journal in
1187  *
1188  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1189  * the journal.  The inode must exist already, must support bmap() and
1190  * must have all data blocks preallocated.
1191  */
1192 journal_t * jbd2_journal_init_inode (struct inode *inode)
1193 {
1194 	struct buffer_head *bh;
1195 	journal_t *journal = journal_init_common();
1196 	char *p;
1197 	int err;
1198 	int n;
1199 	unsigned long long blocknr;
1200 
1201 	if (!journal)
1202 		return NULL;
1203 
1204 	journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1205 	journal->j_inode = inode;
1206 	bdevname(journal->j_dev, journal->j_devname);
1207 	p = journal->j_devname;
1208 	while ((p = strchr(p, '/')))
1209 		*p = '!';
1210 	p = journal->j_devname + strlen(journal->j_devname);
1211 	sprintf(p, "-%lu", journal->j_inode->i_ino);
1212 	jbd_debug(1,
1213 		  "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1214 		  journal, inode->i_sb->s_id, inode->i_ino,
1215 		  (long long) inode->i_size,
1216 		  inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1217 
1218 	journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1219 	journal->j_blocksize = inode->i_sb->s_blocksize;
1220 	jbd2_stats_proc_init(journal);
1221 
1222 	/* journal descriptor can store up to n blocks -bzzz */
1223 	n = journal->j_blocksize / sizeof(journal_block_tag_t);
1224 	journal->j_wbufsize = n;
1225 	journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1226 	if (!journal->j_wbuf) {
1227 		printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1228 			__func__);
1229 		goto out_err;
1230 	}
1231 
1232 	err = jbd2_journal_bmap(journal, 0, &blocknr);
1233 	/* If that failed, give up */
1234 	if (err) {
1235 		printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1236 		       __func__);
1237 		goto out_err;
1238 	}
1239 
1240 	bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize);
1241 	if (!bh) {
1242 		printk(KERN_ERR
1243 		       "%s: Cannot get buffer for journal superblock\n",
1244 		       __func__);
1245 		goto out_err;
1246 	}
1247 	journal->j_sb_buffer = bh;
1248 	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1249 
1250 	return journal;
1251 out_err:
1252 	kfree(journal->j_wbuf);
1253 	jbd2_stats_proc_exit(journal);
1254 	kfree(journal);
1255 	return NULL;
1256 }
1257 
1258 /*
1259  * If the journal init or create aborts, we need to mark the journal
1260  * superblock as being NULL to prevent the journal destroy from writing
1261  * back a bogus superblock.
1262  */
1263 static void journal_fail_superblock (journal_t *journal)
1264 {
1265 	struct buffer_head *bh = journal->j_sb_buffer;
1266 	brelse(bh);
1267 	journal->j_sb_buffer = NULL;
1268 }
1269 
1270 /*
1271  * Given a journal_t structure, initialise the various fields for
1272  * startup of a new journaling session.  We use this both when creating
1273  * a journal, and after recovering an old journal to reset it for
1274  * subsequent use.
1275  */
1276 
1277 static int journal_reset(journal_t *journal)
1278 {
1279 	journal_superblock_t *sb = journal->j_superblock;
1280 	unsigned long long first, last;
1281 
1282 	first = be32_to_cpu(sb->s_first);
1283 	last = be32_to_cpu(sb->s_maxlen);
1284 	if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1285 		printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1286 		       first, last);
1287 		journal_fail_superblock(journal);
1288 		return -EINVAL;
1289 	}
1290 
1291 	journal->j_first = first;
1292 	journal->j_last = last;
1293 
1294 	journal->j_head = first;
1295 	journal->j_tail = first;
1296 	journal->j_free = last - first;
1297 
1298 	journal->j_tail_sequence = journal->j_transaction_sequence;
1299 	journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1300 	journal->j_commit_request = journal->j_commit_sequence;
1301 
1302 	journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1303 
1304 	/*
1305 	 * As a special case, if the on-disk copy is already marked as needing
1306 	 * no recovery (s_start == 0), then we can safely defer the superblock
1307 	 * update until the next commit by setting JBD2_FLUSHED.  This avoids
1308 	 * attempting a write to a potential-readonly device.
1309 	 */
1310 	if (sb->s_start == 0) {
1311 		jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1312 			"(start %ld, seq %d, errno %d)\n",
1313 			journal->j_tail, journal->j_tail_sequence,
1314 			journal->j_errno);
1315 		journal->j_flags |= JBD2_FLUSHED;
1316 	} else {
1317 		/* Lock here to make assertions happy... */
1318 		mutex_lock(&journal->j_checkpoint_mutex);
1319 		/*
1320 		 * Update log tail information. We use WRITE_FUA since new
1321 		 * transaction will start reusing journal space and so we
1322 		 * must make sure information about current log tail is on
1323 		 * disk before that.
1324 		 */
1325 		jbd2_journal_update_sb_log_tail(journal,
1326 						journal->j_tail_sequence,
1327 						journal->j_tail,
1328 						WRITE_FUA);
1329 		mutex_unlock(&journal->j_checkpoint_mutex);
1330 	}
1331 	return jbd2_journal_start_thread(journal);
1332 }
1333 
1334 static void jbd2_write_superblock(journal_t *journal, int write_op)
1335 {
1336 	struct buffer_head *bh = journal->j_sb_buffer;
1337 	journal_superblock_t *sb = journal->j_superblock;
1338 	int ret;
1339 
1340 	trace_jbd2_write_superblock(journal, write_op);
1341 	if (!(journal->j_flags & JBD2_BARRIER))
1342 		write_op &= ~(REQ_FUA | REQ_FLUSH);
1343 	lock_buffer(bh);
1344 	if (buffer_write_io_error(bh)) {
1345 		/*
1346 		 * Oh, dear.  A previous attempt to write the journal
1347 		 * superblock failed.  This could happen because the
1348 		 * USB device was yanked out.  Or it could happen to
1349 		 * be a transient write error and maybe the block will
1350 		 * be remapped.  Nothing we can do but to retry the
1351 		 * write and hope for the best.
1352 		 */
1353 		printk(KERN_ERR "JBD2: previous I/O error detected "
1354 		       "for journal superblock update for %s.\n",
1355 		       journal->j_devname);
1356 		clear_buffer_write_io_error(bh);
1357 		set_buffer_uptodate(bh);
1358 	}
1359 	jbd2_superblock_csum_set(journal, sb);
1360 	get_bh(bh);
1361 	bh->b_end_io = end_buffer_write_sync;
1362 	ret = submit_bh(write_op, bh);
1363 	wait_on_buffer(bh);
1364 	if (buffer_write_io_error(bh)) {
1365 		clear_buffer_write_io_error(bh);
1366 		set_buffer_uptodate(bh);
1367 		ret = -EIO;
1368 	}
1369 	if (ret) {
1370 		printk(KERN_ERR "JBD2: Error %d detected when updating "
1371 		       "journal superblock for %s.\n", ret,
1372 		       journal->j_devname);
1373 	}
1374 }
1375 
1376 /**
1377  * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1378  * @journal: The journal to update.
1379  * @tail_tid: TID of the new transaction at the tail of the log
1380  * @tail_block: The first block of the transaction at the tail of the log
1381  * @write_op: With which operation should we write the journal sb
1382  *
1383  * Update a journal's superblock information about log tail and write it to
1384  * disk, waiting for the IO to complete.
1385  */
1386 void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1387 				     unsigned long tail_block, int write_op)
1388 {
1389 	journal_superblock_t *sb = journal->j_superblock;
1390 
1391 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1392 	jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1393 		  tail_block, tail_tid);
1394 
1395 	sb->s_sequence = cpu_to_be32(tail_tid);
1396 	sb->s_start    = cpu_to_be32(tail_block);
1397 
1398 	jbd2_write_superblock(journal, write_op);
1399 
1400 	/* Log is no longer empty */
1401 	write_lock(&journal->j_state_lock);
1402 	WARN_ON(!sb->s_sequence);
1403 	journal->j_flags &= ~JBD2_FLUSHED;
1404 	write_unlock(&journal->j_state_lock);
1405 }
1406 
1407 /**
1408  * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1409  * @journal: The journal to update.
1410  *
1411  * Update a journal's dynamic superblock fields to show that journal is empty.
1412  * Write updated superblock to disk waiting for IO to complete.
1413  */
1414 static void jbd2_mark_journal_empty(journal_t *journal)
1415 {
1416 	journal_superblock_t *sb = journal->j_superblock;
1417 
1418 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1419 	read_lock(&journal->j_state_lock);
1420 	/* Is it already empty? */
1421 	if (sb->s_start == 0) {
1422 		read_unlock(&journal->j_state_lock);
1423 		return;
1424 	}
1425 	jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1426 		  journal->j_tail_sequence);
1427 
1428 	sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1429 	sb->s_start    = cpu_to_be32(0);
1430 	read_unlock(&journal->j_state_lock);
1431 
1432 	jbd2_write_superblock(journal, WRITE_FUA);
1433 
1434 	/* Log is no longer empty */
1435 	write_lock(&journal->j_state_lock);
1436 	journal->j_flags |= JBD2_FLUSHED;
1437 	write_unlock(&journal->j_state_lock);
1438 }
1439 
1440 
1441 /**
1442  * jbd2_journal_update_sb_errno() - Update error in the journal.
1443  * @journal: The journal to update.
1444  *
1445  * Update a journal's errno.  Write updated superblock to disk waiting for IO
1446  * to complete.
1447  */
1448 void jbd2_journal_update_sb_errno(journal_t *journal)
1449 {
1450 	journal_superblock_t *sb = journal->j_superblock;
1451 
1452 	read_lock(&journal->j_state_lock);
1453 	jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1454 		  journal->j_errno);
1455 	sb->s_errno    = cpu_to_be32(journal->j_errno);
1456 	read_unlock(&journal->j_state_lock);
1457 
1458 	jbd2_write_superblock(journal, WRITE_SYNC);
1459 }
1460 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1461 
1462 /*
1463  * Read the superblock for a given journal, performing initial
1464  * validation of the format.
1465  */
1466 static int journal_get_superblock(journal_t *journal)
1467 {
1468 	struct buffer_head *bh;
1469 	journal_superblock_t *sb;
1470 	int err = -EIO;
1471 
1472 	bh = journal->j_sb_buffer;
1473 
1474 	J_ASSERT(bh != NULL);
1475 	if (!buffer_uptodate(bh)) {
1476 		ll_rw_block(READ, 1, &bh);
1477 		wait_on_buffer(bh);
1478 		if (!buffer_uptodate(bh)) {
1479 			printk(KERN_ERR
1480 				"JBD2: IO error reading journal superblock\n");
1481 			goto out;
1482 		}
1483 	}
1484 
1485 	if (buffer_verified(bh))
1486 		return 0;
1487 
1488 	sb = journal->j_superblock;
1489 
1490 	err = -EINVAL;
1491 
1492 	if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1493 	    sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1494 		printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1495 		goto out;
1496 	}
1497 
1498 	switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1499 	case JBD2_SUPERBLOCK_V1:
1500 		journal->j_format_version = 1;
1501 		break;
1502 	case JBD2_SUPERBLOCK_V2:
1503 		journal->j_format_version = 2;
1504 		break;
1505 	default:
1506 		printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1507 		goto out;
1508 	}
1509 
1510 	if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1511 		journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1512 	else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1513 		printk(KERN_WARNING "JBD2: journal file too short\n");
1514 		goto out;
1515 	}
1516 
1517 	if (be32_to_cpu(sb->s_first) == 0 ||
1518 	    be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1519 		printk(KERN_WARNING
1520 			"JBD2: Invalid start block of journal: %u\n",
1521 			be32_to_cpu(sb->s_first));
1522 		goto out;
1523 	}
1524 
1525 	if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2) &&
1526 	    JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1527 		/* Can't have checksum v2 and v3 at the same time! */
1528 		printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1529 		       "at the same time!\n");
1530 		goto out;
1531 	}
1532 
1533 	if (jbd2_journal_has_csum_v2or3(journal) &&
1534 	    JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM)) {
1535 		/* Can't have checksum v1 and v2 on at the same time! */
1536 		printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1537 		       "at the same time!\n");
1538 		goto out;
1539 	}
1540 
1541 	if (!jbd2_verify_csum_type(journal, sb)) {
1542 		printk(KERN_ERR "JBD2: Unknown checksum type\n");
1543 		goto out;
1544 	}
1545 
1546 	/* Load the checksum driver */
1547 	if (jbd2_journal_has_csum_v2or3(journal)) {
1548 		journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1549 		if (IS_ERR(journal->j_chksum_driver)) {
1550 			printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1551 			err = PTR_ERR(journal->j_chksum_driver);
1552 			journal->j_chksum_driver = NULL;
1553 			goto out;
1554 		}
1555 	}
1556 
1557 	/* Check superblock checksum */
1558 	if (!jbd2_superblock_csum_verify(journal, sb)) {
1559 		printk(KERN_ERR "JBD2: journal checksum error\n");
1560 		goto out;
1561 	}
1562 
1563 	/* Precompute checksum seed for all metadata */
1564 	if (jbd2_journal_has_csum_v2or3(journal))
1565 		journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1566 						   sizeof(sb->s_uuid));
1567 
1568 	set_buffer_verified(bh);
1569 
1570 	return 0;
1571 
1572 out:
1573 	journal_fail_superblock(journal);
1574 	return err;
1575 }
1576 
1577 /*
1578  * Load the on-disk journal superblock and read the key fields into the
1579  * journal_t.
1580  */
1581 
1582 static int load_superblock(journal_t *journal)
1583 {
1584 	int err;
1585 	journal_superblock_t *sb;
1586 
1587 	err = journal_get_superblock(journal);
1588 	if (err)
1589 		return err;
1590 
1591 	sb = journal->j_superblock;
1592 
1593 	journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1594 	journal->j_tail = be32_to_cpu(sb->s_start);
1595 	journal->j_first = be32_to_cpu(sb->s_first);
1596 	journal->j_last = be32_to_cpu(sb->s_maxlen);
1597 	journal->j_errno = be32_to_cpu(sb->s_errno);
1598 
1599 	return 0;
1600 }
1601 
1602 
1603 /**
1604  * int jbd2_journal_load() - Read journal from disk.
1605  * @journal: Journal to act on.
1606  *
1607  * Given a journal_t structure which tells us which disk blocks contain
1608  * a journal, read the journal from disk to initialise the in-memory
1609  * structures.
1610  */
1611 int jbd2_journal_load(journal_t *journal)
1612 {
1613 	int err;
1614 	journal_superblock_t *sb;
1615 
1616 	err = load_superblock(journal);
1617 	if (err)
1618 		return err;
1619 
1620 	sb = journal->j_superblock;
1621 	/* If this is a V2 superblock, then we have to check the
1622 	 * features flags on it. */
1623 
1624 	if (journal->j_format_version >= 2) {
1625 		if ((sb->s_feature_ro_compat &
1626 		     ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1627 		    (sb->s_feature_incompat &
1628 		     ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1629 			printk(KERN_WARNING
1630 				"JBD2: Unrecognised features on journal\n");
1631 			return -EINVAL;
1632 		}
1633 	}
1634 
1635 	/*
1636 	 * Create a slab for this blocksize
1637 	 */
1638 	err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1639 	if (err)
1640 		return err;
1641 
1642 	/* Let the recovery code check whether it needs to recover any
1643 	 * data from the journal. */
1644 	if (jbd2_journal_recover(journal))
1645 		goto recovery_error;
1646 
1647 	if (journal->j_failed_commit) {
1648 		printk(KERN_ERR "JBD2: journal transaction %u on %s "
1649 		       "is corrupt.\n", journal->j_failed_commit,
1650 		       journal->j_devname);
1651 		return -EIO;
1652 	}
1653 
1654 	/* OK, we've finished with the dynamic journal bits:
1655 	 * reinitialise the dynamic contents of the superblock in memory
1656 	 * and reset them on disk. */
1657 	if (journal_reset(journal))
1658 		goto recovery_error;
1659 
1660 	journal->j_flags &= ~JBD2_ABORT;
1661 	journal->j_flags |= JBD2_LOADED;
1662 	return 0;
1663 
1664 recovery_error:
1665 	printk(KERN_WARNING "JBD2: recovery failed\n");
1666 	return -EIO;
1667 }
1668 
1669 /**
1670  * void jbd2_journal_destroy() - Release a journal_t structure.
1671  * @journal: Journal to act on.
1672  *
1673  * Release a journal_t structure once it is no longer in use by the
1674  * journaled object.
1675  * Return <0 if we couldn't clean up the journal.
1676  */
1677 int jbd2_journal_destroy(journal_t *journal)
1678 {
1679 	int err = 0;
1680 
1681 	/* Wait for the commit thread to wake up and die. */
1682 	journal_kill_thread(journal);
1683 
1684 	/* Force a final log commit */
1685 	if (journal->j_running_transaction)
1686 		jbd2_journal_commit_transaction(journal);
1687 
1688 	/* Force any old transactions to disk */
1689 
1690 	/* Totally anal locking here... */
1691 	spin_lock(&journal->j_list_lock);
1692 	while (journal->j_checkpoint_transactions != NULL) {
1693 		spin_unlock(&journal->j_list_lock);
1694 		mutex_lock(&journal->j_checkpoint_mutex);
1695 		jbd2_log_do_checkpoint(journal);
1696 		mutex_unlock(&journal->j_checkpoint_mutex);
1697 		spin_lock(&journal->j_list_lock);
1698 	}
1699 
1700 	J_ASSERT(journal->j_running_transaction == NULL);
1701 	J_ASSERT(journal->j_committing_transaction == NULL);
1702 	J_ASSERT(journal->j_checkpoint_transactions == NULL);
1703 	spin_unlock(&journal->j_list_lock);
1704 
1705 	if (journal->j_sb_buffer) {
1706 		if (!is_journal_aborted(journal)) {
1707 			mutex_lock(&journal->j_checkpoint_mutex);
1708 			jbd2_mark_journal_empty(journal);
1709 			mutex_unlock(&journal->j_checkpoint_mutex);
1710 		} else
1711 			err = -EIO;
1712 		brelse(journal->j_sb_buffer);
1713 	}
1714 
1715 	if (journal->j_proc_entry)
1716 		jbd2_stats_proc_exit(journal);
1717 	iput(journal->j_inode);
1718 	if (journal->j_revoke)
1719 		jbd2_journal_destroy_revoke(journal);
1720 	if (journal->j_chksum_driver)
1721 		crypto_free_shash(journal->j_chksum_driver);
1722 	kfree(journal->j_wbuf);
1723 	kfree(journal);
1724 
1725 	return err;
1726 }
1727 
1728 
1729 /**
1730  *int jbd2_journal_check_used_features () - Check if features specified are used.
1731  * @journal: Journal to check.
1732  * @compat: bitmask of compatible features
1733  * @ro: bitmask of features that force read-only mount
1734  * @incompat: bitmask of incompatible features
1735  *
1736  * Check whether the journal uses all of a given set of
1737  * features.  Return true (non-zero) if it does.
1738  **/
1739 
1740 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1741 				 unsigned long ro, unsigned long incompat)
1742 {
1743 	journal_superblock_t *sb;
1744 
1745 	if (!compat && !ro && !incompat)
1746 		return 1;
1747 	/* Load journal superblock if it is not loaded yet. */
1748 	if (journal->j_format_version == 0 &&
1749 	    journal_get_superblock(journal) != 0)
1750 		return 0;
1751 	if (journal->j_format_version == 1)
1752 		return 0;
1753 
1754 	sb = journal->j_superblock;
1755 
1756 	if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1757 	    ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1758 	    ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1759 		return 1;
1760 
1761 	return 0;
1762 }
1763 
1764 /**
1765  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1766  * @journal: Journal to check.
1767  * @compat: bitmask of compatible features
1768  * @ro: bitmask of features that force read-only mount
1769  * @incompat: bitmask of incompatible features
1770  *
1771  * Check whether the journaling code supports the use of
1772  * all of a given set of features on this journal.  Return true
1773  * (non-zero) if it can. */
1774 
1775 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1776 				      unsigned long ro, unsigned long incompat)
1777 {
1778 	if (!compat && !ro && !incompat)
1779 		return 1;
1780 
1781 	/* We can support any known requested features iff the
1782 	 * superblock is in version 2.  Otherwise we fail to support any
1783 	 * extended sb features. */
1784 
1785 	if (journal->j_format_version != 2)
1786 		return 0;
1787 
1788 	if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1789 	    (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1790 	    (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1791 		return 1;
1792 
1793 	return 0;
1794 }
1795 
1796 /**
1797  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1798  * @journal: Journal to act on.
1799  * @compat: bitmask of compatible features
1800  * @ro: bitmask of features that force read-only mount
1801  * @incompat: bitmask of incompatible features
1802  *
1803  * Mark a given journal feature as present on the
1804  * superblock.  Returns true if the requested features could be set.
1805  *
1806  */
1807 
1808 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1809 			  unsigned long ro, unsigned long incompat)
1810 {
1811 #define INCOMPAT_FEATURE_ON(f) \
1812 		((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1813 #define COMPAT_FEATURE_ON(f) \
1814 		((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1815 	journal_superblock_t *sb;
1816 
1817 	if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1818 		return 1;
1819 
1820 	if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1821 		return 0;
1822 
1823 	/* If enabling v2 checksums, turn on v3 instead */
1824 	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1825 		incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1826 		incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1827 	}
1828 
1829 	/* Asking for checksumming v3 and v1?  Only give them v3. */
1830 	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1831 	    compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1832 		compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1833 
1834 	jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1835 		  compat, ro, incompat);
1836 
1837 	sb = journal->j_superblock;
1838 
1839 	/* If enabling v3 checksums, update superblock */
1840 	if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1841 		sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1842 		sb->s_feature_compat &=
1843 			~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1844 
1845 		/* Load the checksum driver */
1846 		if (journal->j_chksum_driver == NULL) {
1847 			journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1848 								      0, 0);
1849 			if (IS_ERR(journal->j_chksum_driver)) {
1850 				printk(KERN_ERR "JBD2: Cannot load crc32c "
1851 				       "driver.\n");
1852 				journal->j_chksum_driver = NULL;
1853 				return 0;
1854 			}
1855 
1856 			/* Precompute checksum seed for all metadata */
1857 			journal->j_csum_seed = jbd2_chksum(journal, ~0,
1858 							   sb->s_uuid,
1859 							   sizeof(sb->s_uuid));
1860 		}
1861 	}
1862 
1863 	/* If enabling v1 checksums, downgrade superblock */
1864 	if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1865 		sb->s_feature_incompat &=
1866 			~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1867 				     JBD2_FEATURE_INCOMPAT_CSUM_V3);
1868 
1869 	sb->s_feature_compat    |= cpu_to_be32(compat);
1870 	sb->s_feature_ro_compat |= cpu_to_be32(ro);
1871 	sb->s_feature_incompat  |= cpu_to_be32(incompat);
1872 
1873 	return 1;
1874 #undef COMPAT_FEATURE_ON
1875 #undef INCOMPAT_FEATURE_ON
1876 }
1877 
1878 /*
1879  * jbd2_journal_clear_features () - Clear a given journal feature in the
1880  * 				    superblock
1881  * @journal: Journal to act on.
1882  * @compat: bitmask of compatible features
1883  * @ro: bitmask of features that force read-only mount
1884  * @incompat: bitmask of incompatible features
1885  *
1886  * Clear a given journal feature as present on the
1887  * superblock.
1888  */
1889 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1890 				unsigned long ro, unsigned long incompat)
1891 {
1892 	journal_superblock_t *sb;
1893 
1894 	jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1895 		  compat, ro, incompat);
1896 
1897 	sb = journal->j_superblock;
1898 
1899 	sb->s_feature_compat    &= ~cpu_to_be32(compat);
1900 	sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1901 	sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1902 }
1903 EXPORT_SYMBOL(jbd2_journal_clear_features);
1904 
1905 /**
1906  * int jbd2_journal_flush () - Flush journal
1907  * @journal: Journal to act on.
1908  *
1909  * Flush all data for a given journal to disk and empty the journal.
1910  * Filesystems can use this when remounting readonly to ensure that
1911  * recovery does not need to happen on remount.
1912  */
1913 
1914 int jbd2_journal_flush(journal_t *journal)
1915 {
1916 	int err = 0;
1917 	transaction_t *transaction = NULL;
1918 
1919 	write_lock(&journal->j_state_lock);
1920 
1921 	/* Force everything buffered to the log... */
1922 	if (journal->j_running_transaction) {
1923 		transaction = journal->j_running_transaction;
1924 		__jbd2_log_start_commit(journal, transaction->t_tid);
1925 	} else if (journal->j_committing_transaction)
1926 		transaction = journal->j_committing_transaction;
1927 
1928 	/* Wait for the log commit to complete... */
1929 	if (transaction) {
1930 		tid_t tid = transaction->t_tid;
1931 
1932 		write_unlock(&journal->j_state_lock);
1933 		jbd2_log_wait_commit(journal, tid);
1934 	} else {
1935 		write_unlock(&journal->j_state_lock);
1936 	}
1937 
1938 	/* ...and flush everything in the log out to disk. */
1939 	spin_lock(&journal->j_list_lock);
1940 	while (!err && journal->j_checkpoint_transactions != NULL) {
1941 		spin_unlock(&journal->j_list_lock);
1942 		mutex_lock(&journal->j_checkpoint_mutex);
1943 		err = jbd2_log_do_checkpoint(journal);
1944 		mutex_unlock(&journal->j_checkpoint_mutex);
1945 		spin_lock(&journal->j_list_lock);
1946 	}
1947 	spin_unlock(&journal->j_list_lock);
1948 
1949 	if (is_journal_aborted(journal))
1950 		return -EIO;
1951 
1952 	mutex_lock(&journal->j_checkpoint_mutex);
1953 	jbd2_cleanup_journal_tail(journal);
1954 
1955 	/* Finally, mark the journal as really needing no recovery.
1956 	 * This sets s_start==0 in the underlying superblock, which is
1957 	 * the magic code for a fully-recovered superblock.  Any future
1958 	 * commits of data to the journal will restore the current
1959 	 * s_start value. */
1960 	jbd2_mark_journal_empty(journal);
1961 	mutex_unlock(&journal->j_checkpoint_mutex);
1962 	write_lock(&journal->j_state_lock);
1963 	J_ASSERT(!journal->j_running_transaction);
1964 	J_ASSERT(!journal->j_committing_transaction);
1965 	J_ASSERT(!journal->j_checkpoint_transactions);
1966 	J_ASSERT(journal->j_head == journal->j_tail);
1967 	J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1968 	write_unlock(&journal->j_state_lock);
1969 	return 0;
1970 }
1971 
1972 /**
1973  * int jbd2_journal_wipe() - Wipe journal contents
1974  * @journal: Journal to act on.
1975  * @write: flag (see below)
1976  *
1977  * Wipe out all of the contents of a journal, safely.  This will produce
1978  * a warning if the journal contains any valid recovery information.
1979  * Must be called between journal_init_*() and jbd2_journal_load().
1980  *
1981  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1982  * we merely suppress recovery.
1983  */
1984 
1985 int jbd2_journal_wipe(journal_t *journal, int write)
1986 {
1987 	int err = 0;
1988 
1989 	J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1990 
1991 	err = load_superblock(journal);
1992 	if (err)
1993 		return err;
1994 
1995 	if (!journal->j_tail)
1996 		goto no_recovery;
1997 
1998 	printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
1999 		write ? "Clearing" : "Ignoring");
2000 
2001 	err = jbd2_journal_skip_recovery(journal);
2002 	if (write) {
2003 		/* Lock to make assertions happy... */
2004 		mutex_lock(&journal->j_checkpoint_mutex);
2005 		jbd2_mark_journal_empty(journal);
2006 		mutex_unlock(&journal->j_checkpoint_mutex);
2007 	}
2008 
2009  no_recovery:
2010 	return err;
2011 }
2012 
2013 /*
2014  * Journal abort has very specific semantics, which we describe
2015  * for journal abort.
2016  *
2017  * Two internal functions, which provide abort to the jbd layer
2018  * itself are here.
2019  */
2020 
2021 /*
2022  * Quick version for internal journal use (doesn't lock the journal).
2023  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2024  * and don't attempt to make any other journal updates.
2025  */
2026 void __jbd2_journal_abort_hard(journal_t *journal)
2027 {
2028 	transaction_t *transaction;
2029 
2030 	if (journal->j_flags & JBD2_ABORT)
2031 		return;
2032 
2033 	printk(KERN_ERR "Aborting journal on device %s.\n",
2034 	       journal->j_devname);
2035 
2036 	write_lock(&journal->j_state_lock);
2037 	journal->j_flags |= JBD2_ABORT;
2038 	transaction = journal->j_running_transaction;
2039 	if (transaction)
2040 		__jbd2_log_start_commit(journal, transaction->t_tid);
2041 	write_unlock(&journal->j_state_lock);
2042 }
2043 
2044 /* Soft abort: record the abort error status in the journal superblock,
2045  * but don't do any other IO. */
2046 static void __journal_abort_soft (journal_t *journal, int errno)
2047 {
2048 	if (journal->j_flags & JBD2_ABORT)
2049 		return;
2050 
2051 	if (!journal->j_errno)
2052 		journal->j_errno = errno;
2053 
2054 	__jbd2_journal_abort_hard(journal);
2055 
2056 	if (errno)
2057 		jbd2_journal_update_sb_errno(journal);
2058 }
2059 
2060 /**
2061  * void jbd2_journal_abort () - Shutdown the journal immediately.
2062  * @journal: the journal to shutdown.
2063  * @errno:   an error number to record in the journal indicating
2064  *           the reason for the shutdown.
2065  *
2066  * Perform a complete, immediate shutdown of the ENTIRE
2067  * journal (not of a single transaction).  This operation cannot be
2068  * undone without closing and reopening the journal.
2069  *
2070  * The jbd2_journal_abort function is intended to support higher level error
2071  * recovery mechanisms such as the ext2/ext3 remount-readonly error
2072  * mode.
2073  *
2074  * Journal abort has very specific semantics.  Any existing dirty,
2075  * unjournaled buffers in the main filesystem will still be written to
2076  * disk by bdflush, but the journaling mechanism will be suspended
2077  * immediately and no further transaction commits will be honoured.
2078  *
2079  * Any dirty, journaled buffers will be written back to disk without
2080  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2081  * filesystem, but we _do_ attempt to leave as much data as possible
2082  * behind for fsck to use for cleanup.
2083  *
2084  * Any attempt to get a new transaction handle on a journal which is in
2085  * ABORT state will just result in an -EROFS error return.  A
2086  * jbd2_journal_stop on an existing handle will return -EIO if we have
2087  * entered abort state during the update.
2088  *
2089  * Recursive transactions are not disturbed by journal abort until the
2090  * final jbd2_journal_stop, which will receive the -EIO error.
2091  *
2092  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2093  * which will be recorded (if possible) in the journal superblock.  This
2094  * allows a client to record failure conditions in the middle of a
2095  * transaction without having to complete the transaction to record the
2096  * failure to disk.  ext3_error, for example, now uses this
2097  * functionality.
2098  *
2099  * Errors which originate from within the journaling layer will NOT
2100  * supply an errno; a null errno implies that absolutely no further
2101  * writes are done to the journal (unless there are any already in
2102  * progress).
2103  *
2104  */
2105 
2106 void jbd2_journal_abort(journal_t *journal, int errno)
2107 {
2108 	__journal_abort_soft(journal, errno);
2109 }
2110 
2111 /**
2112  * int jbd2_journal_errno () - returns the journal's error state.
2113  * @journal: journal to examine.
2114  *
2115  * This is the errno number set with jbd2_journal_abort(), the last
2116  * time the journal was mounted - if the journal was stopped
2117  * without calling abort this will be 0.
2118  *
2119  * If the journal has been aborted on this mount time -EROFS will
2120  * be returned.
2121  */
2122 int jbd2_journal_errno(journal_t *journal)
2123 {
2124 	int err;
2125 
2126 	read_lock(&journal->j_state_lock);
2127 	if (journal->j_flags & JBD2_ABORT)
2128 		err = -EROFS;
2129 	else
2130 		err = journal->j_errno;
2131 	read_unlock(&journal->j_state_lock);
2132 	return err;
2133 }
2134 
2135 /**
2136  * int jbd2_journal_clear_err () - clears the journal's error state
2137  * @journal: journal to act on.
2138  *
2139  * An error must be cleared or acked to take a FS out of readonly
2140  * mode.
2141  */
2142 int jbd2_journal_clear_err(journal_t *journal)
2143 {
2144 	int err = 0;
2145 
2146 	write_lock(&journal->j_state_lock);
2147 	if (journal->j_flags & JBD2_ABORT)
2148 		err = -EROFS;
2149 	else
2150 		journal->j_errno = 0;
2151 	write_unlock(&journal->j_state_lock);
2152 	return err;
2153 }
2154 
2155 /**
2156  * void jbd2_journal_ack_err() - Ack journal err.
2157  * @journal: journal to act on.
2158  *
2159  * An error must be cleared or acked to take a FS out of readonly
2160  * mode.
2161  */
2162 void jbd2_journal_ack_err(journal_t *journal)
2163 {
2164 	write_lock(&journal->j_state_lock);
2165 	if (journal->j_errno)
2166 		journal->j_flags |= JBD2_ACK_ERR;
2167 	write_unlock(&journal->j_state_lock);
2168 }
2169 
2170 int jbd2_journal_blocks_per_page(struct inode *inode)
2171 {
2172 	return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
2173 }
2174 
2175 /*
2176  * helper functions to deal with 32 or 64bit block numbers.
2177  */
2178 size_t journal_tag_bytes(journal_t *journal)
2179 {
2180 	size_t sz;
2181 
2182 	if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3))
2183 		return sizeof(journal_block_tag3_t);
2184 
2185 	sz = sizeof(journal_block_tag_t);
2186 
2187 	if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2))
2188 		sz += sizeof(__u16);
2189 
2190 	if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
2191 		return sz;
2192 	else
2193 		return sz - sizeof(__u32);
2194 }
2195 
2196 /*
2197  * JBD memory management
2198  *
2199  * These functions are used to allocate block-sized chunks of memory
2200  * used for making copies of buffer_head data.  Very often it will be
2201  * page-sized chunks of data, but sometimes it will be in
2202  * sub-page-size chunks.  (For example, 16k pages on Power systems
2203  * with a 4k block file system.)  For blocks smaller than a page, we
2204  * use a SLAB allocator.  There are slab caches for each block size,
2205  * which are allocated at mount time, if necessary, and we only free
2206  * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2207  * this reason we don't need to a mutex to protect access to
2208  * jbd2_slab[] allocating or releasing memory; only in
2209  * jbd2_journal_create_slab().
2210  */
2211 #define JBD2_MAX_SLABS 8
2212 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2213 
2214 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2215 	"jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2216 	"jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2217 };
2218 
2219 
2220 static void jbd2_journal_destroy_slabs(void)
2221 {
2222 	int i;
2223 
2224 	for (i = 0; i < JBD2_MAX_SLABS; i++) {
2225 		if (jbd2_slab[i])
2226 			kmem_cache_destroy(jbd2_slab[i]);
2227 		jbd2_slab[i] = NULL;
2228 	}
2229 }
2230 
2231 static int jbd2_journal_create_slab(size_t size)
2232 {
2233 	static DEFINE_MUTEX(jbd2_slab_create_mutex);
2234 	int i = order_base_2(size) - 10;
2235 	size_t slab_size;
2236 
2237 	if (size == PAGE_SIZE)
2238 		return 0;
2239 
2240 	if (i >= JBD2_MAX_SLABS)
2241 		return -EINVAL;
2242 
2243 	if (unlikely(i < 0))
2244 		i = 0;
2245 	mutex_lock(&jbd2_slab_create_mutex);
2246 	if (jbd2_slab[i]) {
2247 		mutex_unlock(&jbd2_slab_create_mutex);
2248 		return 0;	/* Already created */
2249 	}
2250 
2251 	slab_size = 1 << (i+10);
2252 	jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2253 					 slab_size, 0, NULL);
2254 	mutex_unlock(&jbd2_slab_create_mutex);
2255 	if (!jbd2_slab[i]) {
2256 		printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2257 		return -ENOMEM;
2258 	}
2259 	return 0;
2260 }
2261 
2262 static struct kmem_cache *get_slab(size_t size)
2263 {
2264 	int i = order_base_2(size) - 10;
2265 
2266 	BUG_ON(i >= JBD2_MAX_SLABS);
2267 	if (unlikely(i < 0))
2268 		i = 0;
2269 	BUG_ON(jbd2_slab[i] == NULL);
2270 	return jbd2_slab[i];
2271 }
2272 
2273 void *jbd2_alloc(size_t size, gfp_t flags)
2274 {
2275 	void *ptr;
2276 
2277 	BUG_ON(size & (size-1)); /* Must be a power of 2 */
2278 
2279 	flags |= __GFP_REPEAT;
2280 	if (size == PAGE_SIZE)
2281 		ptr = (void *)__get_free_pages(flags, 0);
2282 	else if (size > PAGE_SIZE) {
2283 		int order = get_order(size);
2284 
2285 		if (order < 3)
2286 			ptr = (void *)__get_free_pages(flags, order);
2287 		else
2288 			ptr = vmalloc(size);
2289 	} else
2290 		ptr = kmem_cache_alloc(get_slab(size), flags);
2291 
2292 	/* Check alignment; SLUB has gotten this wrong in the past,
2293 	 * and this can lead to user data corruption! */
2294 	BUG_ON(((unsigned long) ptr) & (size-1));
2295 
2296 	return ptr;
2297 }
2298 
2299 void jbd2_free(void *ptr, size_t size)
2300 {
2301 	if (size == PAGE_SIZE) {
2302 		free_pages((unsigned long)ptr, 0);
2303 		return;
2304 	}
2305 	if (size > PAGE_SIZE) {
2306 		int order = get_order(size);
2307 
2308 		if (order < 3)
2309 			free_pages((unsigned long)ptr, order);
2310 		else
2311 			vfree(ptr);
2312 		return;
2313 	}
2314 	kmem_cache_free(get_slab(size), ptr);
2315 };
2316 
2317 /*
2318  * Journal_head storage management
2319  */
2320 static struct kmem_cache *jbd2_journal_head_cache;
2321 #ifdef CONFIG_JBD2_DEBUG
2322 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2323 #endif
2324 
2325 static int jbd2_journal_init_journal_head_cache(void)
2326 {
2327 	int retval;
2328 
2329 	J_ASSERT(jbd2_journal_head_cache == NULL);
2330 	jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2331 				sizeof(struct journal_head),
2332 				0,		/* offset */
2333 				SLAB_TEMPORARY,	/* flags */
2334 				NULL);		/* ctor */
2335 	retval = 0;
2336 	if (!jbd2_journal_head_cache) {
2337 		retval = -ENOMEM;
2338 		printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2339 	}
2340 	return retval;
2341 }
2342 
2343 static void jbd2_journal_destroy_journal_head_cache(void)
2344 {
2345 	if (jbd2_journal_head_cache) {
2346 		kmem_cache_destroy(jbd2_journal_head_cache);
2347 		jbd2_journal_head_cache = NULL;
2348 	}
2349 }
2350 
2351 /*
2352  * journal_head splicing and dicing
2353  */
2354 static struct journal_head *journal_alloc_journal_head(void)
2355 {
2356 	struct journal_head *ret;
2357 
2358 #ifdef CONFIG_JBD2_DEBUG
2359 	atomic_inc(&nr_journal_heads);
2360 #endif
2361 	ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2362 	if (!ret) {
2363 		jbd_debug(1, "out of memory for journal_head\n");
2364 		pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2365 		while (!ret) {
2366 			yield();
2367 			ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2368 		}
2369 	}
2370 	return ret;
2371 }
2372 
2373 static void journal_free_journal_head(struct journal_head *jh)
2374 {
2375 #ifdef CONFIG_JBD2_DEBUG
2376 	atomic_dec(&nr_journal_heads);
2377 	memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2378 #endif
2379 	kmem_cache_free(jbd2_journal_head_cache, jh);
2380 }
2381 
2382 /*
2383  * A journal_head is attached to a buffer_head whenever JBD has an
2384  * interest in the buffer.
2385  *
2386  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2387  * is set.  This bit is tested in core kernel code where we need to take
2388  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2389  * there.
2390  *
2391  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2392  *
2393  * When a buffer has its BH_JBD bit set it is immune from being released by
2394  * core kernel code, mainly via ->b_count.
2395  *
2396  * A journal_head is detached from its buffer_head when the journal_head's
2397  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2398  * transaction (b_cp_transaction) hold their references to b_jcount.
2399  *
2400  * Various places in the kernel want to attach a journal_head to a buffer_head
2401  * _before_ attaching the journal_head to a transaction.  To protect the
2402  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2403  * journal_head's b_jcount refcount by one.  The caller must call
2404  * jbd2_journal_put_journal_head() to undo this.
2405  *
2406  * So the typical usage would be:
2407  *
2408  *	(Attach a journal_head if needed.  Increments b_jcount)
2409  *	struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2410  *	...
2411  *      (Get another reference for transaction)
2412  *	jbd2_journal_grab_journal_head(bh);
2413  *	jh->b_transaction = xxx;
2414  *	(Put original reference)
2415  *	jbd2_journal_put_journal_head(jh);
2416  */
2417 
2418 /*
2419  * Give a buffer_head a journal_head.
2420  *
2421  * May sleep.
2422  */
2423 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2424 {
2425 	struct journal_head *jh;
2426 	struct journal_head *new_jh = NULL;
2427 
2428 repeat:
2429 	if (!buffer_jbd(bh))
2430 		new_jh = journal_alloc_journal_head();
2431 
2432 	jbd_lock_bh_journal_head(bh);
2433 	if (buffer_jbd(bh)) {
2434 		jh = bh2jh(bh);
2435 	} else {
2436 		J_ASSERT_BH(bh,
2437 			(atomic_read(&bh->b_count) > 0) ||
2438 			(bh->b_page && bh->b_page->mapping));
2439 
2440 		if (!new_jh) {
2441 			jbd_unlock_bh_journal_head(bh);
2442 			goto repeat;
2443 		}
2444 
2445 		jh = new_jh;
2446 		new_jh = NULL;		/* We consumed it */
2447 		set_buffer_jbd(bh);
2448 		bh->b_private = jh;
2449 		jh->b_bh = bh;
2450 		get_bh(bh);
2451 		BUFFER_TRACE(bh, "added journal_head");
2452 	}
2453 	jh->b_jcount++;
2454 	jbd_unlock_bh_journal_head(bh);
2455 	if (new_jh)
2456 		journal_free_journal_head(new_jh);
2457 	return bh->b_private;
2458 }
2459 
2460 /*
2461  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2462  * having a journal_head, return NULL
2463  */
2464 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2465 {
2466 	struct journal_head *jh = NULL;
2467 
2468 	jbd_lock_bh_journal_head(bh);
2469 	if (buffer_jbd(bh)) {
2470 		jh = bh2jh(bh);
2471 		jh->b_jcount++;
2472 	}
2473 	jbd_unlock_bh_journal_head(bh);
2474 	return jh;
2475 }
2476 
2477 static void __journal_remove_journal_head(struct buffer_head *bh)
2478 {
2479 	struct journal_head *jh = bh2jh(bh);
2480 
2481 	J_ASSERT_JH(jh, jh->b_jcount >= 0);
2482 	J_ASSERT_JH(jh, jh->b_transaction == NULL);
2483 	J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2484 	J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2485 	J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2486 	J_ASSERT_BH(bh, buffer_jbd(bh));
2487 	J_ASSERT_BH(bh, jh2bh(jh) == bh);
2488 	BUFFER_TRACE(bh, "remove journal_head");
2489 	if (jh->b_frozen_data) {
2490 		printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2491 		jbd2_free(jh->b_frozen_data, bh->b_size);
2492 	}
2493 	if (jh->b_committed_data) {
2494 		printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2495 		jbd2_free(jh->b_committed_data, bh->b_size);
2496 	}
2497 	bh->b_private = NULL;
2498 	jh->b_bh = NULL;	/* debug, really */
2499 	clear_buffer_jbd(bh);
2500 	journal_free_journal_head(jh);
2501 }
2502 
2503 /*
2504  * Drop a reference on the passed journal_head.  If it fell to zero then
2505  * release the journal_head from the buffer_head.
2506  */
2507 void jbd2_journal_put_journal_head(struct journal_head *jh)
2508 {
2509 	struct buffer_head *bh = jh2bh(jh);
2510 
2511 	jbd_lock_bh_journal_head(bh);
2512 	J_ASSERT_JH(jh, jh->b_jcount > 0);
2513 	--jh->b_jcount;
2514 	if (!jh->b_jcount) {
2515 		__journal_remove_journal_head(bh);
2516 		jbd_unlock_bh_journal_head(bh);
2517 		__brelse(bh);
2518 	} else
2519 		jbd_unlock_bh_journal_head(bh);
2520 }
2521 
2522 /*
2523  * Initialize jbd inode head
2524  */
2525 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2526 {
2527 	jinode->i_transaction = NULL;
2528 	jinode->i_next_transaction = NULL;
2529 	jinode->i_vfs_inode = inode;
2530 	jinode->i_flags = 0;
2531 	INIT_LIST_HEAD(&jinode->i_list);
2532 }
2533 
2534 /*
2535  * Function to be called before we start removing inode from memory (i.e.,
2536  * clear_inode() is a fine place to be called from). It removes inode from
2537  * transaction's lists.
2538  */
2539 void jbd2_journal_release_jbd_inode(journal_t *journal,
2540 				    struct jbd2_inode *jinode)
2541 {
2542 	if (!journal)
2543 		return;
2544 restart:
2545 	spin_lock(&journal->j_list_lock);
2546 	/* Is commit writing out inode - we have to wait */
2547 	if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
2548 		wait_queue_head_t *wq;
2549 		DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2550 		wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2551 		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2552 		spin_unlock(&journal->j_list_lock);
2553 		schedule();
2554 		finish_wait(wq, &wait.wait);
2555 		goto restart;
2556 	}
2557 
2558 	if (jinode->i_transaction) {
2559 		list_del(&jinode->i_list);
2560 		jinode->i_transaction = NULL;
2561 	}
2562 	spin_unlock(&journal->j_list_lock);
2563 }
2564 
2565 
2566 #ifdef CONFIG_PROC_FS
2567 
2568 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2569 
2570 static void __init jbd2_create_jbd_stats_proc_entry(void)
2571 {
2572 	proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2573 }
2574 
2575 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2576 {
2577 	if (proc_jbd2_stats)
2578 		remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2579 }
2580 
2581 #else
2582 
2583 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2584 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2585 
2586 #endif
2587 
2588 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2589 
2590 static int __init jbd2_journal_init_handle_cache(void)
2591 {
2592 	jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2593 	if (jbd2_handle_cache == NULL) {
2594 		printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2595 		return -ENOMEM;
2596 	}
2597 	jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2598 	if (jbd2_inode_cache == NULL) {
2599 		printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2600 		kmem_cache_destroy(jbd2_handle_cache);
2601 		return -ENOMEM;
2602 	}
2603 	return 0;
2604 }
2605 
2606 static void jbd2_journal_destroy_handle_cache(void)
2607 {
2608 	if (jbd2_handle_cache)
2609 		kmem_cache_destroy(jbd2_handle_cache);
2610 	if (jbd2_inode_cache)
2611 		kmem_cache_destroy(jbd2_inode_cache);
2612 
2613 }
2614 
2615 /*
2616  * Module startup and shutdown
2617  */
2618 
2619 static int __init journal_init_caches(void)
2620 {
2621 	int ret;
2622 
2623 	ret = jbd2_journal_init_revoke_caches();
2624 	if (ret == 0)
2625 		ret = jbd2_journal_init_journal_head_cache();
2626 	if (ret == 0)
2627 		ret = jbd2_journal_init_handle_cache();
2628 	if (ret == 0)
2629 		ret = jbd2_journal_init_transaction_cache();
2630 	return ret;
2631 }
2632 
2633 static void jbd2_journal_destroy_caches(void)
2634 {
2635 	jbd2_journal_destroy_revoke_caches();
2636 	jbd2_journal_destroy_journal_head_cache();
2637 	jbd2_journal_destroy_handle_cache();
2638 	jbd2_journal_destroy_transaction_cache();
2639 	jbd2_journal_destroy_slabs();
2640 }
2641 
2642 static int __init journal_init(void)
2643 {
2644 	int ret;
2645 
2646 	BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2647 
2648 	ret = journal_init_caches();
2649 	if (ret == 0) {
2650 		jbd2_create_jbd_stats_proc_entry();
2651 	} else {
2652 		jbd2_journal_destroy_caches();
2653 	}
2654 	return ret;
2655 }
2656 
2657 static void __exit journal_exit(void)
2658 {
2659 #ifdef CONFIG_JBD2_DEBUG
2660 	int n = atomic_read(&nr_journal_heads);
2661 	if (n)
2662 		printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2663 #endif
2664 	jbd2_remove_jbd_stats_proc_entry();
2665 	jbd2_journal_destroy_caches();
2666 }
2667 
2668 MODULE_LICENSE("GPL");
2669 module_init(journal_init);
2670 module_exit(journal_exit);
2671 
2672