xref: /openbmc/linux/fs/jbd2/journal.c (revision 4f205687)
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_inode_add_write);
98 EXPORT_SYMBOL(jbd2_journal_inode_add_wait);
99 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
100 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
101 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
102 EXPORT_SYMBOL(jbd2_inode_cache);
103 
104 static void __journal_abort_soft (journal_t *journal, int errno);
105 static int jbd2_journal_create_slab(size_t slab_size);
106 
107 #ifdef CONFIG_JBD2_DEBUG
108 void __jbd2_debug(int level, const char *file, const char *func,
109 		  unsigned int line, const char *fmt, ...)
110 {
111 	struct va_format vaf;
112 	va_list args;
113 
114 	if (level > jbd2_journal_enable_debug)
115 		return;
116 	va_start(args, fmt);
117 	vaf.fmt = fmt;
118 	vaf.va = &args;
119 	printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf);
120 	va_end(args);
121 }
122 EXPORT_SYMBOL(__jbd2_debug);
123 #endif
124 
125 /* Checksumming functions */
126 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
127 {
128 	if (!jbd2_journal_has_csum_v2or3_feature(j))
129 		return 1;
130 
131 	return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
132 }
133 
134 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
135 {
136 	__u32 csum;
137 	__be32 old_csum;
138 
139 	old_csum = sb->s_checksum;
140 	sb->s_checksum = 0;
141 	csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
142 	sb->s_checksum = old_csum;
143 
144 	return cpu_to_be32(csum);
145 }
146 
147 static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb)
148 {
149 	if (!jbd2_journal_has_csum_v2or3(j))
150 		return 1;
151 
152 	return sb->s_checksum == jbd2_superblock_csum(j, sb);
153 }
154 
155 static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb)
156 {
157 	if (!jbd2_journal_has_csum_v2or3(j))
158 		return;
159 
160 	sb->s_checksum = jbd2_superblock_csum(j, sb);
161 }
162 
163 /*
164  * Helper function used to manage commit timeouts
165  */
166 
167 static void commit_timeout(unsigned long __data)
168 {
169 	struct task_struct * p = (struct task_struct *) __data;
170 
171 	wake_up_process(p);
172 }
173 
174 /*
175  * kjournald2: The main thread function used to manage a logging device
176  * journal.
177  *
178  * This kernel thread is responsible for two things:
179  *
180  * 1) COMMIT:  Every so often we need to commit the current state of the
181  *    filesystem to disk.  The journal thread is responsible for writing
182  *    all of the metadata buffers to disk.
183  *
184  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
185  *    of the data in that part of the log has been rewritten elsewhere on
186  *    the disk.  Flushing these old buffers to reclaim space in the log is
187  *    known as checkpointing, and this thread is responsible for that job.
188  */
189 
190 static int kjournald2(void *arg)
191 {
192 	journal_t *journal = arg;
193 	transaction_t *transaction;
194 
195 	/*
196 	 * Set up an interval timer which can be used to trigger a commit wakeup
197 	 * after the commit interval expires
198 	 */
199 	setup_timer(&journal->j_commit_timer, commit_timeout,
200 			(unsigned long)current);
201 
202 	set_freezable();
203 
204 	/* Record that the journal thread is running */
205 	journal->j_task = current;
206 	wake_up(&journal->j_wait_done_commit);
207 
208 	/*
209 	 * And now, wait forever for commit wakeup events.
210 	 */
211 	write_lock(&journal->j_state_lock);
212 
213 loop:
214 	if (journal->j_flags & JBD2_UNMOUNT)
215 		goto end_loop;
216 
217 	jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
218 		journal->j_commit_sequence, journal->j_commit_request);
219 
220 	if (journal->j_commit_sequence != journal->j_commit_request) {
221 		jbd_debug(1, "OK, requests differ\n");
222 		write_unlock(&journal->j_state_lock);
223 		del_timer_sync(&journal->j_commit_timer);
224 		jbd2_journal_commit_transaction(journal);
225 		write_lock(&journal->j_state_lock);
226 		goto loop;
227 	}
228 
229 	wake_up(&journal->j_wait_done_commit);
230 	if (freezing(current)) {
231 		/*
232 		 * The simpler the better. Flushing journal isn't a
233 		 * good idea, because that depends on threads that may
234 		 * be already stopped.
235 		 */
236 		jbd_debug(1, "Now suspending kjournald2\n");
237 		write_unlock(&journal->j_state_lock);
238 		try_to_freeze();
239 		write_lock(&journal->j_state_lock);
240 	} else {
241 		/*
242 		 * We assume on resume that commits are already there,
243 		 * so we don't sleep
244 		 */
245 		DEFINE_WAIT(wait);
246 		int should_sleep = 1;
247 
248 		prepare_to_wait(&journal->j_wait_commit, &wait,
249 				TASK_INTERRUPTIBLE);
250 		if (journal->j_commit_sequence != journal->j_commit_request)
251 			should_sleep = 0;
252 		transaction = journal->j_running_transaction;
253 		if (transaction && time_after_eq(jiffies,
254 						transaction->t_expires))
255 			should_sleep = 0;
256 		if (journal->j_flags & JBD2_UNMOUNT)
257 			should_sleep = 0;
258 		if (should_sleep) {
259 			write_unlock(&journal->j_state_lock);
260 			schedule();
261 			write_lock(&journal->j_state_lock);
262 		}
263 		finish_wait(&journal->j_wait_commit, &wait);
264 	}
265 
266 	jbd_debug(1, "kjournald2 wakes\n");
267 
268 	/*
269 	 * Were we woken up by a commit wakeup event?
270 	 */
271 	transaction = journal->j_running_transaction;
272 	if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
273 		journal->j_commit_request = transaction->t_tid;
274 		jbd_debug(1, "woke because of timeout\n");
275 	}
276 	goto loop;
277 
278 end_loop:
279 	write_unlock(&journal->j_state_lock);
280 	del_timer_sync(&journal->j_commit_timer);
281 	journal->j_task = NULL;
282 	wake_up(&journal->j_wait_done_commit);
283 	jbd_debug(1, "Journal thread exiting.\n");
284 	return 0;
285 }
286 
287 static int jbd2_journal_start_thread(journal_t *journal)
288 {
289 	struct task_struct *t;
290 
291 	t = kthread_run(kjournald2, journal, "jbd2/%s",
292 			journal->j_devname);
293 	if (IS_ERR(t))
294 		return PTR_ERR(t);
295 
296 	wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
297 	return 0;
298 }
299 
300 static void journal_kill_thread(journal_t *journal)
301 {
302 	write_lock(&journal->j_state_lock);
303 	journal->j_flags |= JBD2_UNMOUNT;
304 
305 	while (journal->j_task) {
306 		write_unlock(&journal->j_state_lock);
307 		wake_up(&journal->j_wait_commit);
308 		wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
309 		write_lock(&journal->j_state_lock);
310 	}
311 	write_unlock(&journal->j_state_lock);
312 }
313 
314 /*
315  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
316  *
317  * Writes a metadata buffer to a given disk block.  The actual IO is not
318  * performed but a new buffer_head is constructed which labels the data
319  * to be written with the correct destination disk block.
320  *
321  * Any magic-number escaping which needs to be done will cause a
322  * copy-out here.  If the buffer happens to start with the
323  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
324  * magic number is only written to the log for descripter blocks.  In
325  * this case, we copy the data and replace the first word with 0, and we
326  * return a result code which indicates that this buffer needs to be
327  * marked as an escaped buffer in the corresponding log descriptor
328  * block.  The missing word can then be restored when the block is read
329  * during recovery.
330  *
331  * If the source buffer has already been modified by a new transaction
332  * since we took the last commit snapshot, we use the frozen copy of
333  * that data for IO. If we end up using the existing buffer_head's data
334  * for the write, then we have to make sure nobody modifies it while the
335  * IO is in progress. do_get_write_access() handles this.
336  *
337  * The function returns a pointer to the buffer_head to be used for IO.
338  *
339  *
340  * Return value:
341  *  <0: Error
342  * >=0: Finished OK
343  *
344  * On success:
345  * Bit 0 set == escape performed on the data
346  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
347  */
348 
349 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
350 				  struct journal_head  *jh_in,
351 				  struct buffer_head **bh_out,
352 				  sector_t blocknr)
353 {
354 	int need_copy_out = 0;
355 	int done_copy_out = 0;
356 	int do_escape = 0;
357 	char *mapped_data;
358 	struct buffer_head *new_bh;
359 	struct page *new_page;
360 	unsigned int new_offset;
361 	struct buffer_head *bh_in = jh2bh(jh_in);
362 	journal_t *journal = transaction->t_journal;
363 
364 	/*
365 	 * The buffer really shouldn't be locked: only the current committing
366 	 * transaction is allowed to write it, so nobody else is allowed
367 	 * to do any IO.
368 	 *
369 	 * akpm: except if we're journalling data, and write() output is
370 	 * also part of a shared mapping, and another thread has
371 	 * decided to launch a writepage() against this buffer.
372 	 */
373 	J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
374 
375 	new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
376 
377 	/* keep subsequent assertions sane */
378 	atomic_set(&new_bh->b_count, 1);
379 
380 	jbd_lock_bh_state(bh_in);
381 repeat:
382 	/*
383 	 * If a new transaction has already done a buffer copy-out, then
384 	 * we use that version of the data for the commit.
385 	 */
386 	if (jh_in->b_frozen_data) {
387 		done_copy_out = 1;
388 		new_page = virt_to_page(jh_in->b_frozen_data);
389 		new_offset = offset_in_page(jh_in->b_frozen_data);
390 	} else {
391 		new_page = jh2bh(jh_in)->b_page;
392 		new_offset = offset_in_page(jh2bh(jh_in)->b_data);
393 	}
394 
395 	mapped_data = kmap_atomic(new_page);
396 	/*
397 	 * Fire data frozen trigger if data already wasn't frozen.  Do this
398 	 * before checking for escaping, as the trigger may modify the magic
399 	 * offset.  If a copy-out happens afterwards, it will have the correct
400 	 * data in the buffer.
401 	 */
402 	if (!done_copy_out)
403 		jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
404 					   jh_in->b_triggers);
405 
406 	/*
407 	 * Check for escaping
408 	 */
409 	if (*((__be32 *)(mapped_data + new_offset)) ==
410 				cpu_to_be32(JBD2_MAGIC_NUMBER)) {
411 		need_copy_out = 1;
412 		do_escape = 1;
413 	}
414 	kunmap_atomic(mapped_data);
415 
416 	/*
417 	 * Do we need to do a data copy?
418 	 */
419 	if (need_copy_out && !done_copy_out) {
420 		char *tmp;
421 
422 		jbd_unlock_bh_state(bh_in);
423 		tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
424 		if (!tmp) {
425 			brelse(new_bh);
426 			return -ENOMEM;
427 		}
428 		jbd_lock_bh_state(bh_in);
429 		if (jh_in->b_frozen_data) {
430 			jbd2_free(tmp, bh_in->b_size);
431 			goto repeat;
432 		}
433 
434 		jh_in->b_frozen_data = tmp;
435 		mapped_data = kmap_atomic(new_page);
436 		memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
437 		kunmap_atomic(mapped_data);
438 
439 		new_page = virt_to_page(tmp);
440 		new_offset = offset_in_page(tmp);
441 		done_copy_out = 1;
442 
443 		/*
444 		 * This isn't strictly necessary, as we're using frozen
445 		 * data for the escaping, but it keeps consistency with
446 		 * b_frozen_data usage.
447 		 */
448 		jh_in->b_frozen_triggers = jh_in->b_triggers;
449 	}
450 
451 	/*
452 	 * Did we need to do an escaping?  Now we've done all the
453 	 * copying, we can finally do so.
454 	 */
455 	if (do_escape) {
456 		mapped_data = kmap_atomic(new_page);
457 		*((unsigned int *)(mapped_data + new_offset)) = 0;
458 		kunmap_atomic(mapped_data);
459 	}
460 
461 	set_bh_page(new_bh, new_page, new_offset);
462 	new_bh->b_size = bh_in->b_size;
463 	new_bh->b_bdev = journal->j_dev;
464 	new_bh->b_blocknr = blocknr;
465 	new_bh->b_private = bh_in;
466 	set_buffer_mapped(new_bh);
467 	set_buffer_dirty(new_bh);
468 
469 	*bh_out = new_bh;
470 
471 	/*
472 	 * The to-be-written buffer needs to get moved to the io queue,
473 	 * and the original buffer whose contents we are shadowing or
474 	 * copying is moved to the transaction's shadow queue.
475 	 */
476 	JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
477 	spin_lock(&journal->j_list_lock);
478 	__jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
479 	spin_unlock(&journal->j_list_lock);
480 	set_buffer_shadow(bh_in);
481 	jbd_unlock_bh_state(bh_in);
482 
483 	return do_escape | (done_copy_out << 1);
484 }
485 
486 /*
487  * Allocation code for the journal file.  Manage the space left in the
488  * journal, so that we can begin checkpointing when appropriate.
489  */
490 
491 /*
492  * Called with j_state_lock locked for writing.
493  * Returns true if a transaction commit was started.
494  */
495 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
496 {
497 	/* Return if the txn has already requested to be committed */
498 	if (journal->j_commit_request == target)
499 		return 0;
500 
501 	/*
502 	 * The only transaction we can possibly wait upon is the
503 	 * currently running transaction (if it exists).  Otherwise,
504 	 * the target tid must be an old one.
505 	 */
506 	if (journal->j_running_transaction &&
507 	    journal->j_running_transaction->t_tid == target) {
508 		/*
509 		 * We want a new commit: OK, mark the request and wakeup the
510 		 * commit thread.  We do _not_ do the commit ourselves.
511 		 */
512 
513 		journal->j_commit_request = target;
514 		jbd_debug(1, "JBD2: requesting commit %d/%d\n",
515 			  journal->j_commit_request,
516 			  journal->j_commit_sequence);
517 		journal->j_running_transaction->t_requested = jiffies;
518 		wake_up(&journal->j_wait_commit);
519 		return 1;
520 	} else if (!tid_geq(journal->j_commit_request, target))
521 		/* This should never happen, but if it does, preserve
522 		   the evidence before kjournald goes into a loop and
523 		   increments j_commit_sequence beyond all recognition. */
524 		WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
525 			  journal->j_commit_request,
526 			  journal->j_commit_sequence,
527 			  target, journal->j_running_transaction ?
528 			  journal->j_running_transaction->t_tid : 0);
529 	return 0;
530 }
531 
532 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
533 {
534 	int ret;
535 
536 	write_lock(&journal->j_state_lock);
537 	ret = __jbd2_log_start_commit(journal, tid);
538 	write_unlock(&journal->j_state_lock);
539 	return ret;
540 }
541 
542 /*
543  * Force and wait any uncommitted transactions.  We can only force the running
544  * transaction if we don't have an active handle, otherwise, we will deadlock.
545  * Returns: <0 in case of error,
546  *           0 if nothing to commit,
547  *           1 if transaction was successfully committed.
548  */
549 static int __jbd2_journal_force_commit(journal_t *journal)
550 {
551 	transaction_t *transaction = NULL;
552 	tid_t tid;
553 	int need_to_start = 0, ret = 0;
554 
555 	read_lock(&journal->j_state_lock);
556 	if (journal->j_running_transaction && !current->journal_info) {
557 		transaction = journal->j_running_transaction;
558 		if (!tid_geq(journal->j_commit_request, transaction->t_tid))
559 			need_to_start = 1;
560 	} else if (journal->j_committing_transaction)
561 		transaction = journal->j_committing_transaction;
562 
563 	if (!transaction) {
564 		/* Nothing to commit */
565 		read_unlock(&journal->j_state_lock);
566 		return 0;
567 	}
568 	tid = transaction->t_tid;
569 	read_unlock(&journal->j_state_lock);
570 	if (need_to_start)
571 		jbd2_log_start_commit(journal, tid);
572 	ret = jbd2_log_wait_commit(journal, tid);
573 	if (!ret)
574 		ret = 1;
575 
576 	return ret;
577 }
578 
579 /**
580  * Force and wait upon a commit if the calling process is not within
581  * transaction.  This is used for forcing out undo-protected data which contains
582  * bitmaps, when the fs is running out of space.
583  *
584  * @journal: journal to force
585  * Returns true if progress was made.
586  */
587 int jbd2_journal_force_commit_nested(journal_t *journal)
588 {
589 	int ret;
590 
591 	ret = __jbd2_journal_force_commit(journal);
592 	return ret > 0;
593 }
594 
595 /**
596  * int journal_force_commit() - force any uncommitted transactions
597  * @journal: journal to force
598  *
599  * Caller want unconditional commit. We can only force the running transaction
600  * if we don't have an active handle, otherwise, we will deadlock.
601  */
602 int jbd2_journal_force_commit(journal_t *journal)
603 {
604 	int ret;
605 
606 	J_ASSERT(!current->journal_info);
607 	ret = __jbd2_journal_force_commit(journal);
608 	if (ret > 0)
609 		ret = 0;
610 	return ret;
611 }
612 
613 /*
614  * Start a commit of the current running transaction (if any).  Returns true
615  * if a transaction is going to be committed (or is currently already
616  * committing), and fills its tid in at *ptid
617  */
618 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
619 {
620 	int ret = 0;
621 
622 	write_lock(&journal->j_state_lock);
623 	if (journal->j_running_transaction) {
624 		tid_t tid = journal->j_running_transaction->t_tid;
625 
626 		__jbd2_log_start_commit(journal, tid);
627 		/* There's a running transaction and we've just made sure
628 		 * it's commit has been scheduled. */
629 		if (ptid)
630 			*ptid = tid;
631 		ret = 1;
632 	} else if (journal->j_committing_transaction) {
633 		/*
634 		 * If commit has been started, then we have to wait for
635 		 * completion of that transaction.
636 		 */
637 		if (ptid)
638 			*ptid = journal->j_committing_transaction->t_tid;
639 		ret = 1;
640 	}
641 	write_unlock(&journal->j_state_lock);
642 	return ret;
643 }
644 
645 /*
646  * Return 1 if a given transaction has not yet sent barrier request
647  * connected with a transaction commit. If 0 is returned, transaction
648  * may or may not have sent the barrier. Used to avoid sending barrier
649  * twice in common cases.
650  */
651 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
652 {
653 	int ret = 0;
654 	transaction_t *commit_trans;
655 
656 	if (!(journal->j_flags & JBD2_BARRIER))
657 		return 0;
658 	read_lock(&journal->j_state_lock);
659 	/* Transaction already committed? */
660 	if (tid_geq(journal->j_commit_sequence, tid))
661 		goto out;
662 	commit_trans = journal->j_committing_transaction;
663 	if (!commit_trans || commit_trans->t_tid != tid) {
664 		ret = 1;
665 		goto out;
666 	}
667 	/*
668 	 * Transaction is being committed and we already proceeded to
669 	 * submitting a flush to fs partition?
670 	 */
671 	if (journal->j_fs_dev != journal->j_dev) {
672 		if (!commit_trans->t_need_data_flush ||
673 		    commit_trans->t_state >= T_COMMIT_DFLUSH)
674 			goto out;
675 	} else {
676 		if (commit_trans->t_state >= T_COMMIT_JFLUSH)
677 			goto out;
678 	}
679 	ret = 1;
680 out:
681 	read_unlock(&journal->j_state_lock);
682 	return ret;
683 }
684 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
685 
686 /*
687  * Wait for a specified commit to complete.
688  * The caller may not hold the journal lock.
689  */
690 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
691 {
692 	int err = 0;
693 
694 	read_lock(&journal->j_state_lock);
695 #ifdef CONFIG_JBD2_DEBUG
696 	if (!tid_geq(journal->j_commit_request, tid)) {
697 		printk(KERN_ERR
698 		       "%s: error: j_commit_request=%d, tid=%d\n",
699 		       __func__, journal->j_commit_request, tid);
700 	}
701 #endif
702 	while (tid_gt(tid, journal->j_commit_sequence)) {
703 		jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
704 				  tid, journal->j_commit_sequence);
705 		read_unlock(&journal->j_state_lock);
706 		wake_up(&journal->j_wait_commit);
707 		wait_event(journal->j_wait_done_commit,
708 				!tid_gt(tid, journal->j_commit_sequence));
709 		read_lock(&journal->j_state_lock);
710 	}
711 	read_unlock(&journal->j_state_lock);
712 
713 	if (unlikely(is_journal_aborted(journal)))
714 		err = -EIO;
715 	return err;
716 }
717 
718 /*
719  * When this function returns the transaction corresponding to tid
720  * will be completed.  If the transaction has currently running, start
721  * committing that transaction before waiting for it to complete.  If
722  * the transaction id is stale, it is by definition already completed,
723  * so just return SUCCESS.
724  */
725 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
726 {
727 	int	need_to_wait = 1;
728 
729 	read_lock(&journal->j_state_lock);
730 	if (journal->j_running_transaction &&
731 	    journal->j_running_transaction->t_tid == tid) {
732 		if (journal->j_commit_request != tid) {
733 			/* transaction not yet started, so request it */
734 			read_unlock(&journal->j_state_lock);
735 			jbd2_log_start_commit(journal, tid);
736 			goto wait_commit;
737 		}
738 	} else if (!(journal->j_committing_transaction &&
739 		     journal->j_committing_transaction->t_tid == tid))
740 		need_to_wait = 0;
741 	read_unlock(&journal->j_state_lock);
742 	if (!need_to_wait)
743 		return 0;
744 wait_commit:
745 	return jbd2_log_wait_commit(journal, tid);
746 }
747 EXPORT_SYMBOL(jbd2_complete_transaction);
748 
749 /*
750  * Log buffer allocation routines:
751  */
752 
753 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
754 {
755 	unsigned long blocknr;
756 
757 	write_lock(&journal->j_state_lock);
758 	J_ASSERT(journal->j_free > 1);
759 
760 	blocknr = journal->j_head;
761 	journal->j_head++;
762 	journal->j_free--;
763 	if (journal->j_head == journal->j_last)
764 		journal->j_head = journal->j_first;
765 	write_unlock(&journal->j_state_lock);
766 	return jbd2_journal_bmap(journal, blocknr, retp);
767 }
768 
769 /*
770  * Conversion of logical to physical block numbers for the journal
771  *
772  * On external journals the journal blocks are identity-mapped, so
773  * this is a no-op.  If needed, we can use j_blk_offset - everything is
774  * ready.
775  */
776 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
777 		 unsigned long long *retp)
778 {
779 	int err = 0;
780 	unsigned long long ret;
781 
782 	if (journal->j_inode) {
783 		ret = bmap(journal->j_inode, blocknr);
784 		if (ret)
785 			*retp = ret;
786 		else {
787 			printk(KERN_ALERT "%s: journal block not found "
788 					"at offset %lu on %s\n",
789 			       __func__, blocknr, journal->j_devname);
790 			err = -EIO;
791 			__journal_abort_soft(journal, err);
792 		}
793 	} else {
794 		*retp = blocknr; /* +journal->j_blk_offset */
795 	}
796 	return err;
797 }
798 
799 /*
800  * We play buffer_head aliasing tricks to write data/metadata blocks to
801  * the journal without copying their contents, but for journal
802  * descriptor blocks we do need to generate bona fide buffers.
803  *
804  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
805  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
806  * But we don't bother doing that, so there will be coherency problems with
807  * mmaps of blockdevs which hold live JBD-controlled filesystems.
808  */
809 struct buffer_head *
810 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
811 {
812 	journal_t *journal = transaction->t_journal;
813 	struct buffer_head *bh;
814 	unsigned long long blocknr;
815 	journal_header_t *header;
816 	int err;
817 
818 	err = jbd2_journal_next_log_block(journal, &blocknr);
819 
820 	if (err)
821 		return NULL;
822 
823 	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
824 	if (!bh)
825 		return NULL;
826 	lock_buffer(bh);
827 	memset(bh->b_data, 0, journal->j_blocksize);
828 	header = (journal_header_t *)bh->b_data;
829 	header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
830 	header->h_blocktype = cpu_to_be32(type);
831 	header->h_sequence = cpu_to_be32(transaction->t_tid);
832 	set_buffer_uptodate(bh);
833 	unlock_buffer(bh);
834 	BUFFER_TRACE(bh, "return this buffer");
835 	return bh;
836 }
837 
838 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
839 {
840 	struct jbd2_journal_block_tail *tail;
841 	__u32 csum;
842 
843 	if (!jbd2_journal_has_csum_v2or3(j))
844 		return;
845 
846 	tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
847 			sizeof(struct jbd2_journal_block_tail));
848 	tail->t_checksum = 0;
849 	csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
850 	tail->t_checksum = cpu_to_be32(csum);
851 }
852 
853 /*
854  * Return tid of the oldest transaction in the journal and block in the journal
855  * where the transaction starts.
856  *
857  * If the journal is now empty, return which will be the next transaction ID
858  * we will write and where will that transaction start.
859  *
860  * The return value is 0 if journal tail cannot be pushed any further, 1 if
861  * it can.
862  */
863 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
864 			      unsigned long *block)
865 {
866 	transaction_t *transaction;
867 	int ret;
868 
869 	read_lock(&journal->j_state_lock);
870 	spin_lock(&journal->j_list_lock);
871 	transaction = journal->j_checkpoint_transactions;
872 	if (transaction) {
873 		*tid = transaction->t_tid;
874 		*block = transaction->t_log_start;
875 	} else if ((transaction = journal->j_committing_transaction) != NULL) {
876 		*tid = transaction->t_tid;
877 		*block = transaction->t_log_start;
878 	} else if ((transaction = journal->j_running_transaction) != NULL) {
879 		*tid = transaction->t_tid;
880 		*block = journal->j_head;
881 	} else {
882 		*tid = journal->j_transaction_sequence;
883 		*block = journal->j_head;
884 	}
885 	ret = tid_gt(*tid, journal->j_tail_sequence);
886 	spin_unlock(&journal->j_list_lock);
887 	read_unlock(&journal->j_state_lock);
888 
889 	return ret;
890 }
891 
892 /*
893  * Update information in journal structure and in on disk journal superblock
894  * about log tail. This function does not check whether information passed in
895  * really pushes log tail further. It's responsibility of the caller to make
896  * sure provided log tail information is valid (e.g. by holding
897  * j_checkpoint_mutex all the time between computing log tail and calling this
898  * function as is the case with jbd2_cleanup_journal_tail()).
899  *
900  * Requires j_checkpoint_mutex
901  */
902 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
903 {
904 	unsigned long freed;
905 	int ret;
906 
907 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
908 
909 	/*
910 	 * We cannot afford for write to remain in drive's caches since as
911 	 * soon as we update j_tail, next transaction can start reusing journal
912 	 * space and if we lose sb update during power failure we'd replay
913 	 * old transaction with possibly newly overwritten data.
914 	 */
915 	ret = jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA);
916 	if (ret)
917 		goto out;
918 
919 	write_lock(&journal->j_state_lock);
920 	freed = block - journal->j_tail;
921 	if (block < journal->j_tail)
922 		freed += journal->j_last - journal->j_first;
923 
924 	trace_jbd2_update_log_tail(journal, tid, block, freed);
925 	jbd_debug(1,
926 		  "Cleaning journal tail from %d to %d (offset %lu), "
927 		  "freeing %lu\n",
928 		  journal->j_tail_sequence, tid, block, freed);
929 
930 	journal->j_free += freed;
931 	journal->j_tail_sequence = tid;
932 	journal->j_tail = block;
933 	write_unlock(&journal->j_state_lock);
934 
935 out:
936 	return ret;
937 }
938 
939 /*
940  * This is a variaon of __jbd2_update_log_tail which checks for validity of
941  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
942  * with other threads updating log tail.
943  */
944 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
945 {
946 	mutex_lock(&journal->j_checkpoint_mutex);
947 	if (tid_gt(tid, journal->j_tail_sequence))
948 		__jbd2_update_log_tail(journal, tid, block);
949 	mutex_unlock(&journal->j_checkpoint_mutex);
950 }
951 
952 struct jbd2_stats_proc_session {
953 	journal_t *journal;
954 	struct transaction_stats_s *stats;
955 	int start;
956 	int max;
957 };
958 
959 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
960 {
961 	return *pos ? NULL : SEQ_START_TOKEN;
962 }
963 
964 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
965 {
966 	return NULL;
967 }
968 
969 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
970 {
971 	struct jbd2_stats_proc_session *s = seq->private;
972 
973 	if (v != SEQ_START_TOKEN)
974 		return 0;
975 	seq_printf(seq, "%lu transactions (%lu requested), "
976 		   "each up to %u blocks\n",
977 		   s->stats->ts_tid, s->stats->ts_requested,
978 		   s->journal->j_max_transaction_buffers);
979 	if (s->stats->ts_tid == 0)
980 		return 0;
981 	seq_printf(seq, "average: \n  %ums waiting for transaction\n",
982 	    jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
983 	seq_printf(seq, "  %ums request delay\n",
984 	    (s->stats->ts_requested == 0) ? 0 :
985 	    jiffies_to_msecs(s->stats->run.rs_request_delay /
986 			     s->stats->ts_requested));
987 	seq_printf(seq, "  %ums running transaction\n",
988 	    jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
989 	seq_printf(seq, "  %ums transaction was being locked\n",
990 	    jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
991 	seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
992 	    jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
993 	seq_printf(seq, "  %ums logging transaction\n",
994 	    jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
995 	seq_printf(seq, "  %lluus average transaction commit time\n",
996 		   div_u64(s->journal->j_average_commit_time, 1000));
997 	seq_printf(seq, "  %lu handles per transaction\n",
998 	    s->stats->run.rs_handle_count / s->stats->ts_tid);
999 	seq_printf(seq, "  %lu blocks per transaction\n",
1000 	    s->stats->run.rs_blocks / s->stats->ts_tid);
1001 	seq_printf(seq, "  %lu logged blocks per transaction\n",
1002 	    s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1003 	return 0;
1004 }
1005 
1006 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1007 {
1008 }
1009 
1010 static const struct seq_operations jbd2_seq_info_ops = {
1011 	.start  = jbd2_seq_info_start,
1012 	.next   = jbd2_seq_info_next,
1013 	.stop   = jbd2_seq_info_stop,
1014 	.show   = jbd2_seq_info_show,
1015 };
1016 
1017 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1018 {
1019 	journal_t *journal = PDE_DATA(inode);
1020 	struct jbd2_stats_proc_session *s;
1021 	int rc, size;
1022 
1023 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1024 	if (s == NULL)
1025 		return -ENOMEM;
1026 	size = sizeof(struct transaction_stats_s);
1027 	s->stats = kmalloc(size, GFP_KERNEL);
1028 	if (s->stats == NULL) {
1029 		kfree(s);
1030 		return -ENOMEM;
1031 	}
1032 	spin_lock(&journal->j_history_lock);
1033 	memcpy(s->stats, &journal->j_stats, size);
1034 	s->journal = journal;
1035 	spin_unlock(&journal->j_history_lock);
1036 
1037 	rc = seq_open(file, &jbd2_seq_info_ops);
1038 	if (rc == 0) {
1039 		struct seq_file *m = file->private_data;
1040 		m->private = s;
1041 	} else {
1042 		kfree(s->stats);
1043 		kfree(s);
1044 	}
1045 	return rc;
1046 
1047 }
1048 
1049 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1050 {
1051 	struct seq_file *seq = file->private_data;
1052 	struct jbd2_stats_proc_session *s = seq->private;
1053 	kfree(s->stats);
1054 	kfree(s);
1055 	return seq_release(inode, file);
1056 }
1057 
1058 static const struct file_operations jbd2_seq_info_fops = {
1059 	.owner		= THIS_MODULE,
1060 	.open           = jbd2_seq_info_open,
1061 	.read           = seq_read,
1062 	.llseek         = seq_lseek,
1063 	.release        = jbd2_seq_info_release,
1064 };
1065 
1066 static struct proc_dir_entry *proc_jbd2_stats;
1067 
1068 static void jbd2_stats_proc_init(journal_t *journal)
1069 {
1070 	journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1071 	if (journal->j_proc_entry) {
1072 		proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1073 				 &jbd2_seq_info_fops, journal);
1074 	}
1075 }
1076 
1077 static void jbd2_stats_proc_exit(journal_t *journal)
1078 {
1079 	remove_proc_entry("info", journal->j_proc_entry);
1080 	remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1081 }
1082 
1083 /*
1084  * Management for journal control blocks: functions to create and
1085  * destroy journal_t structures, and to initialise and read existing
1086  * journal blocks from disk.  */
1087 
1088 /* First: create and setup a journal_t object in memory.  We initialise
1089  * very few fields yet: that has to wait until we have created the
1090  * journal structures from from scratch, or loaded them from disk. */
1091 
1092 static journal_t * journal_init_common (void)
1093 {
1094 	journal_t *journal;
1095 	int err;
1096 
1097 	journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1098 	if (!journal)
1099 		return NULL;
1100 
1101 	init_waitqueue_head(&journal->j_wait_transaction_locked);
1102 	init_waitqueue_head(&journal->j_wait_done_commit);
1103 	init_waitqueue_head(&journal->j_wait_commit);
1104 	init_waitqueue_head(&journal->j_wait_updates);
1105 	init_waitqueue_head(&journal->j_wait_reserved);
1106 	mutex_init(&journal->j_barrier);
1107 	mutex_init(&journal->j_checkpoint_mutex);
1108 	spin_lock_init(&journal->j_revoke_lock);
1109 	spin_lock_init(&journal->j_list_lock);
1110 	rwlock_init(&journal->j_state_lock);
1111 
1112 	journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1113 	journal->j_min_batch_time = 0;
1114 	journal->j_max_batch_time = 15000; /* 15ms */
1115 	atomic_set(&journal->j_reserved_credits, 0);
1116 
1117 	/* The journal is marked for error until we succeed with recovery! */
1118 	journal->j_flags = JBD2_ABORT;
1119 
1120 	/* Set up a default-sized revoke table for the new mount. */
1121 	err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1122 	if (err) {
1123 		kfree(journal);
1124 		return NULL;
1125 	}
1126 
1127 	spin_lock_init(&journal->j_history_lock);
1128 
1129 	return journal;
1130 }
1131 
1132 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1133  *
1134  * Create a journal structure assigned some fixed set of disk blocks to
1135  * the journal.  We don't actually touch those disk blocks yet, but we
1136  * need to set up all of the mapping information to tell the journaling
1137  * system where the journal blocks are.
1138  *
1139  */
1140 
1141 /**
1142  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1143  *  @bdev: Block device on which to create the journal
1144  *  @fs_dev: Device which hold journalled filesystem for this journal.
1145  *  @start: Block nr Start of journal.
1146  *  @len:  Length of the journal in blocks.
1147  *  @blocksize: blocksize of journalling device
1148  *
1149  *  Returns: a newly created journal_t *
1150  *
1151  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1152  *  range of blocks on an arbitrary block device.
1153  *
1154  */
1155 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
1156 			struct block_device *fs_dev,
1157 			unsigned long long start, int len, int blocksize)
1158 {
1159 	journal_t *journal = journal_init_common();
1160 	struct buffer_head *bh;
1161 	int n;
1162 
1163 	if (!journal)
1164 		return NULL;
1165 
1166 	/* journal descriptor can store up to n blocks -bzzz */
1167 	journal->j_blocksize = blocksize;
1168 	journal->j_dev = bdev;
1169 	journal->j_fs_dev = fs_dev;
1170 	journal->j_blk_offset = start;
1171 	journal->j_maxlen = len;
1172 	bdevname(journal->j_dev, journal->j_devname);
1173 	strreplace(journal->j_devname, '/', '!');
1174 	jbd2_stats_proc_init(journal);
1175 	n = journal->j_blocksize / sizeof(journal_block_tag_t);
1176 	journal->j_wbufsize = n;
1177 	journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1178 	if (!journal->j_wbuf) {
1179 		printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1180 			__func__);
1181 		goto out_err;
1182 	}
1183 
1184 	bh = __getblk(journal->j_dev, start, journal->j_blocksize);
1185 	if (!bh) {
1186 		printk(KERN_ERR
1187 		       "%s: Cannot get buffer for journal superblock\n",
1188 		       __func__);
1189 		goto out_err;
1190 	}
1191 	journal->j_sb_buffer = bh;
1192 	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1193 
1194 	return journal;
1195 out_err:
1196 	kfree(journal->j_wbuf);
1197 	jbd2_stats_proc_exit(journal);
1198 	kfree(journal);
1199 	return NULL;
1200 }
1201 
1202 /**
1203  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1204  *  @inode: An inode to create the journal in
1205  *
1206  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1207  * the journal.  The inode must exist already, must support bmap() and
1208  * must have all data blocks preallocated.
1209  */
1210 journal_t * jbd2_journal_init_inode (struct inode *inode)
1211 {
1212 	struct buffer_head *bh;
1213 	journal_t *journal = journal_init_common();
1214 	char *p;
1215 	int err;
1216 	int n;
1217 	unsigned long long blocknr;
1218 
1219 	if (!journal)
1220 		return NULL;
1221 
1222 	journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1223 	journal->j_inode = inode;
1224 	bdevname(journal->j_dev, journal->j_devname);
1225 	p = strreplace(journal->j_devname, '/', '!');
1226 	sprintf(p, "-%lu", journal->j_inode->i_ino);
1227 	jbd_debug(1,
1228 		  "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1229 		  journal, inode->i_sb->s_id, inode->i_ino,
1230 		  (long long) inode->i_size,
1231 		  inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1232 
1233 	journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1234 	journal->j_blocksize = inode->i_sb->s_blocksize;
1235 	jbd2_stats_proc_init(journal);
1236 
1237 	/* journal descriptor can store up to n blocks -bzzz */
1238 	n = journal->j_blocksize / sizeof(journal_block_tag_t);
1239 	journal->j_wbufsize = n;
1240 	journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1241 	if (!journal->j_wbuf) {
1242 		printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
1243 			__func__);
1244 		goto out_err;
1245 	}
1246 
1247 	err = jbd2_journal_bmap(journal, 0, &blocknr);
1248 	/* If that failed, give up */
1249 	if (err) {
1250 		printk(KERN_ERR "%s: Cannot locate journal superblock\n",
1251 		       __func__);
1252 		goto out_err;
1253 	}
1254 
1255 	bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize);
1256 	if (!bh) {
1257 		printk(KERN_ERR
1258 		       "%s: Cannot get buffer for journal superblock\n",
1259 		       __func__);
1260 		goto out_err;
1261 	}
1262 	journal->j_sb_buffer = bh;
1263 	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1264 
1265 	return journal;
1266 out_err:
1267 	kfree(journal->j_wbuf);
1268 	jbd2_stats_proc_exit(journal);
1269 	kfree(journal);
1270 	return NULL;
1271 }
1272 
1273 /*
1274  * If the journal init or create aborts, we need to mark the journal
1275  * superblock as being NULL to prevent the journal destroy from writing
1276  * back a bogus superblock.
1277  */
1278 static void journal_fail_superblock (journal_t *journal)
1279 {
1280 	struct buffer_head *bh = journal->j_sb_buffer;
1281 	brelse(bh);
1282 	journal->j_sb_buffer = NULL;
1283 }
1284 
1285 /*
1286  * Given a journal_t structure, initialise the various fields for
1287  * startup of a new journaling session.  We use this both when creating
1288  * a journal, and after recovering an old journal to reset it for
1289  * subsequent use.
1290  */
1291 
1292 static int journal_reset(journal_t *journal)
1293 {
1294 	journal_superblock_t *sb = journal->j_superblock;
1295 	unsigned long long first, last;
1296 
1297 	first = be32_to_cpu(sb->s_first);
1298 	last = be32_to_cpu(sb->s_maxlen);
1299 	if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1300 		printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1301 		       first, last);
1302 		journal_fail_superblock(journal);
1303 		return -EINVAL;
1304 	}
1305 
1306 	journal->j_first = first;
1307 	journal->j_last = last;
1308 
1309 	journal->j_head = first;
1310 	journal->j_tail = first;
1311 	journal->j_free = last - first;
1312 
1313 	journal->j_tail_sequence = journal->j_transaction_sequence;
1314 	journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1315 	journal->j_commit_request = journal->j_commit_sequence;
1316 
1317 	journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1318 
1319 	/*
1320 	 * As a special case, if the on-disk copy is already marked as needing
1321 	 * no recovery (s_start == 0), then we can safely defer the superblock
1322 	 * update until the next commit by setting JBD2_FLUSHED.  This avoids
1323 	 * attempting a write to a potential-readonly device.
1324 	 */
1325 	if (sb->s_start == 0) {
1326 		jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
1327 			"(start %ld, seq %d, errno %d)\n",
1328 			journal->j_tail, journal->j_tail_sequence,
1329 			journal->j_errno);
1330 		journal->j_flags |= JBD2_FLUSHED;
1331 	} else {
1332 		/* Lock here to make assertions happy... */
1333 		mutex_lock(&journal->j_checkpoint_mutex);
1334 		/*
1335 		 * Update log tail information. We use WRITE_FUA since new
1336 		 * transaction will start reusing journal space and so we
1337 		 * must make sure information about current log tail is on
1338 		 * disk before that.
1339 		 */
1340 		jbd2_journal_update_sb_log_tail(journal,
1341 						journal->j_tail_sequence,
1342 						journal->j_tail,
1343 						WRITE_FUA);
1344 		mutex_unlock(&journal->j_checkpoint_mutex);
1345 	}
1346 	return jbd2_journal_start_thread(journal);
1347 }
1348 
1349 static int jbd2_write_superblock(journal_t *journal, int write_op)
1350 {
1351 	struct buffer_head *bh = journal->j_sb_buffer;
1352 	journal_superblock_t *sb = journal->j_superblock;
1353 	int ret;
1354 
1355 	trace_jbd2_write_superblock(journal, write_op);
1356 	if (!(journal->j_flags & JBD2_BARRIER))
1357 		write_op &= ~(REQ_FUA | REQ_FLUSH);
1358 	lock_buffer(bh);
1359 	if (buffer_write_io_error(bh)) {
1360 		/*
1361 		 * Oh, dear.  A previous attempt to write the journal
1362 		 * superblock failed.  This could happen because the
1363 		 * USB device was yanked out.  Or it could happen to
1364 		 * be a transient write error and maybe the block will
1365 		 * be remapped.  Nothing we can do but to retry the
1366 		 * write and hope for the best.
1367 		 */
1368 		printk(KERN_ERR "JBD2: previous I/O error detected "
1369 		       "for journal superblock update for %s.\n",
1370 		       journal->j_devname);
1371 		clear_buffer_write_io_error(bh);
1372 		set_buffer_uptodate(bh);
1373 	}
1374 	jbd2_superblock_csum_set(journal, sb);
1375 	get_bh(bh);
1376 	bh->b_end_io = end_buffer_write_sync;
1377 	ret = submit_bh(write_op, bh);
1378 	wait_on_buffer(bh);
1379 	if (buffer_write_io_error(bh)) {
1380 		clear_buffer_write_io_error(bh);
1381 		set_buffer_uptodate(bh);
1382 		ret = -EIO;
1383 	}
1384 	if (ret) {
1385 		printk(KERN_ERR "JBD2: Error %d detected when updating "
1386 		       "journal superblock for %s.\n", ret,
1387 		       journal->j_devname);
1388 		jbd2_journal_abort(journal, ret);
1389 	}
1390 
1391 	return ret;
1392 }
1393 
1394 /**
1395  * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1396  * @journal: The journal to update.
1397  * @tail_tid: TID of the new transaction at the tail of the log
1398  * @tail_block: The first block of the transaction at the tail of the log
1399  * @write_op: With which operation should we write the journal sb
1400  *
1401  * Update a journal's superblock information about log tail and write it to
1402  * disk, waiting for the IO to complete.
1403  */
1404 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1405 				     unsigned long tail_block, int write_op)
1406 {
1407 	journal_superblock_t *sb = journal->j_superblock;
1408 	int ret;
1409 
1410 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1411 	jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1412 		  tail_block, tail_tid);
1413 
1414 	sb->s_sequence = cpu_to_be32(tail_tid);
1415 	sb->s_start    = cpu_to_be32(tail_block);
1416 
1417 	ret = jbd2_write_superblock(journal, write_op);
1418 	if (ret)
1419 		goto out;
1420 
1421 	/* Log is no longer empty */
1422 	write_lock(&journal->j_state_lock);
1423 	WARN_ON(!sb->s_sequence);
1424 	journal->j_flags &= ~JBD2_FLUSHED;
1425 	write_unlock(&journal->j_state_lock);
1426 
1427 out:
1428 	return ret;
1429 }
1430 
1431 /**
1432  * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1433  * @journal: The journal to update.
1434  * @write_op: With which operation should we write the journal sb
1435  *
1436  * Update a journal's dynamic superblock fields to show that journal is empty.
1437  * Write updated superblock to disk waiting for IO to complete.
1438  */
1439 static void jbd2_mark_journal_empty(journal_t *journal, int write_op)
1440 {
1441 	journal_superblock_t *sb = journal->j_superblock;
1442 
1443 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1444 	read_lock(&journal->j_state_lock);
1445 	/* Is it already empty? */
1446 	if (sb->s_start == 0) {
1447 		read_unlock(&journal->j_state_lock);
1448 		return;
1449 	}
1450 	jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n",
1451 		  journal->j_tail_sequence);
1452 
1453 	sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1454 	sb->s_start    = cpu_to_be32(0);
1455 	read_unlock(&journal->j_state_lock);
1456 
1457 	jbd2_write_superblock(journal, write_op);
1458 
1459 	/* Log is no longer empty */
1460 	write_lock(&journal->j_state_lock);
1461 	journal->j_flags |= JBD2_FLUSHED;
1462 	write_unlock(&journal->j_state_lock);
1463 }
1464 
1465 
1466 /**
1467  * jbd2_journal_update_sb_errno() - Update error in the journal.
1468  * @journal: The journal to update.
1469  *
1470  * Update a journal's errno.  Write updated superblock to disk waiting for IO
1471  * to complete.
1472  */
1473 void jbd2_journal_update_sb_errno(journal_t *journal)
1474 {
1475 	journal_superblock_t *sb = journal->j_superblock;
1476 
1477 	read_lock(&journal->j_state_lock);
1478 	jbd_debug(1, "JBD2: updating superblock error (errno %d)\n",
1479 		  journal->j_errno);
1480 	sb->s_errno    = cpu_to_be32(journal->j_errno);
1481 	read_unlock(&journal->j_state_lock);
1482 
1483 	jbd2_write_superblock(journal, WRITE_FUA);
1484 }
1485 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1486 
1487 /*
1488  * Read the superblock for a given journal, performing initial
1489  * validation of the format.
1490  */
1491 static int journal_get_superblock(journal_t *journal)
1492 {
1493 	struct buffer_head *bh;
1494 	journal_superblock_t *sb;
1495 	int err = -EIO;
1496 
1497 	bh = journal->j_sb_buffer;
1498 
1499 	J_ASSERT(bh != NULL);
1500 	if (!buffer_uptodate(bh)) {
1501 		ll_rw_block(READ, 1, &bh);
1502 		wait_on_buffer(bh);
1503 		if (!buffer_uptodate(bh)) {
1504 			printk(KERN_ERR
1505 				"JBD2: IO error reading journal superblock\n");
1506 			goto out;
1507 		}
1508 	}
1509 
1510 	if (buffer_verified(bh))
1511 		return 0;
1512 
1513 	sb = journal->j_superblock;
1514 
1515 	err = -EINVAL;
1516 
1517 	if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1518 	    sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1519 		printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1520 		goto out;
1521 	}
1522 
1523 	switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1524 	case JBD2_SUPERBLOCK_V1:
1525 		journal->j_format_version = 1;
1526 		break;
1527 	case JBD2_SUPERBLOCK_V2:
1528 		journal->j_format_version = 2;
1529 		break;
1530 	default:
1531 		printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1532 		goto out;
1533 	}
1534 
1535 	if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1536 		journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1537 	else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1538 		printk(KERN_WARNING "JBD2: journal file too short\n");
1539 		goto out;
1540 	}
1541 
1542 	if (be32_to_cpu(sb->s_first) == 0 ||
1543 	    be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1544 		printk(KERN_WARNING
1545 			"JBD2: Invalid start block of journal: %u\n",
1546 			be32_to_cpu(sb->s_first));
1547 		goto out;
1548 	}
1549 
1550 	if (jbd2_has_feature_csum2(journal) &&
1551 	    jbd2_has_feature_csum3(journal)) {
1552 		/* Can't have checksum v2 and v3 at the same time! */
1553 		printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1554 		       "at the same time!\n");
1555 		goto out;
1556 	}
1557 
1558 	if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1559 	    jbd2_has_feature_checksum(journal)) {
1560 		/* Can't have checksum v1 and v2 on at the same time! */
1561 		printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1562 		       "at the same time!\n");
1563 		goto out;
1564 	}
1565 
1566 	if (!jbd2_verify_csum_type(journal, sb)) {
1567 		printk(KERN_ERR "JBD2: Unknown checksum type\n");
1568 		goto out;
1569 	}
1570 
1571 	/* Load the checksum driver */
1572 	if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1573 		journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1574 		if (IS_ERR(journal->j_chksum_driver)) {
1575 			printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1576 			err = PTR_ERR(journal->j_chksum_driver);
1577 			journal->j_chksum_driver = NULL;
1578 			goto out;
1579 		}
1580 	}
1581 
1582 	/* Check superblock checksum */
1583 	if (!jbd2_superblock_csum_verify(journal, sb)) {
1584 		printk(KERN_ERR "JBD2: journal checksum error\n");
1585 		err = -EFSBADCRC;
1586 		goto out;
1587 	}
1588 
1589 	/* Precompute checksum seed for all metadata */
1590 	if (jbd2_journal_has_csum_v2or3(journal))
1591 		journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1592 						   sizeof(sb->s_uuid));
1593 
1594 	set_buffer_verified(bh);
1595 
1596 	return 0;
1597 
1598 out:
1599 	journal_fail_superblock(journal);
1600 	return err;
1601 }
1602 
1603 /*
1604  * Load the on-disk journal superblock and read the key fields into the
1605  * journal_t.
1606  */
1607 
1608 static int load_superblock(journal_t *journal)
1609 {
1610 	int err;
1611 	journal_superblock_t *sb;
1612 
1613 	err = journal_get_superblock(journal);
1614 	if (err)
1615 		return err;
1616 
1617 	sb = journal->j_superblock;
1618 
1619 	journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1620 	journal->j_tail = be32_to_cpu(sb->s_start);
1621 	journal->j_first = be32_to_cpu(sb->s_first);
1622 	journal->j_last = be32_to_cpu(sb->s_maxlen);
1623 	journal->j_errno = be32_to_cpu(sb->s_errno);
1624 
1625 	return 0;
1626 }
1627 
1628 
1629 /**
1630  * int jbd2_journal_load() - Read journal from disk.
1631  * @journal: Journal to act on.
1632  *
1633  * Given a journal_t structure which tells us which disk blocks contain
1634  * a journal, read the journal from disk to initialise the in-memory
1635  * structures.
1636  */
1637 int jbd2_journal_load(journal_t *journal)
1638 {
1639 	int err;
1640 	journal_superblock_t *sb;
1641 
1642 	err = load_superblock(journal);
1643 	if (err)
1644 		return err;
1645 
1646 	sb = journal->j_superblock;
1647 	/* If this is a V2 superblock, then we have to check the
1648 	 * features flags on it. */
1649 
1650 	if (journal->j_format_version >= 2) {
1651 		if ((sb->s_feature_ro_compat &
1652 		     ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1653 		    (sb->s_feature_incompat &
1654 		     ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1655 			printk(KERN_WARNING
1656 				"JBD2: Unrecognised features on journal\n");
1657 			return -EINVAL;
1658 		}
1659 	}
1660 
1661 	/*
1662 	 * Create a slab for this blocksize
1663 	 */
1664 	err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1665 	if (err)
1666 		return err;
1667 
1668 	/* Let the recovery code check whether it needs to recover any
1669 	 * data from the journal. */
1670 	if (jbd2_journal_recover(journal))
1671 		goto recovery_error;
1672 
1673 	if (journal->j_failed_commit) {
1674 		printk(KERN_ERR "JBD2: journal transaction %u on %s "
1675 		       "is corrupt.\n", journal->j_failed_commit,
1676 		       journal->j_devname);
1677 		return -EFSCORRUPTED;
1678 	}
1679 
1680 	/* OK, we've finished with the dynamic journal bits:
1681 	 * reinitialise the dynamic contents of the superblock in memory
1682 	 * and reset them on disk. */
1683 	if (journal_reset(journal))
1684 		goto recovery_error;
1685 
1686 	journal->j_flags &= ~JBD2_ABORT;
1687 	journal->j_flags |= JBD2_LOADED;
1688 	return 0;
1689 
1690 recovery_error:
1691 	printk(KERN_WARNING "JBD2: recovery failed\n");
1692 	return -EIO;
1693 }
1694 
1695 /**
1696  * void jbd2_journal_destroy() - Release a journal_t structure.
1697  * @journal: Journal to act on.
1698  *
1699  * Release a journal_t structure once it is no longer in use by the
1700  * journaled object.
1701  * Return <0 if we couldn't clean up the journal.
1702  */
1703 int jbd2_journal_destroy(journal_t *journal)
1704 {
1705 	int err = 0;
1706 
1707 	/* Wait for the commit thread to wake up and die. */
1708 	journal_kill_thread(journal);
1709 
1710 	/* Force a final log commit */
1711 	if (journal->j_running_transaction)
1712 		jbd2_journal_commit_transaction(journal);
1713 
1714 	/* Force any old transactions to disk */
1715 
1716 	/* Totally anal locking here... */
1717 	spin_lock(&journal->j_list_lock);
1718 	while (journal->j_checkpoint_transactions != NULL) {
1719 		spin_unlock(&journal->j_list_lock);
1720 		mutex_lock(&journal->j_checkpoint_mutex);
1721 		err = jbd2_log_do_checkpoint(journal);
1722 		mutex_unlock(&journal->j_checkpoint_mutex);
1723 		/*
1724 		 * If checkpointing failed, just free the buffers to avoid
1725 		 * looping forever
1726 		 */
1727 		if (err) {
1728 			jbd2_journal_destroy_checkpoint(journal);
1729 			spin_lock(&journal->j_list_lock);
1730 			break;
1731 		}
1732 		spin_lock(&journal->j_list_lock);
1733 	}
1734 
1735 	J_ASSERT(journal->j_running_transaction == NULL);
1736 	J_ASSERT(journal->j_committing_transaction == NULL);
1737 	J_ASSERT(journal->j_checkpoint_transactions == NULL);
1738 	spin_unlock(&journal->j_list_lock);
1739 
1740 	if (journal->j_sb_buffer) {
1741 		if (!is_journal_aborted(journal)) {
1742 			mutex_lock(&journal->j_checkpoint_mutex);
1743 
1744 			write_lock(&journal->j_state_lock);
1745 			journal->j_tail_sequence =
1746 				++journal->j_transaction_sequence;
1747 			write_unlock(&journal->j_state_lock);
1748 
1749 			jbd2_mark_journal_empty(journal, WRITE_FLUSH_FUA);
1750 			mutex_unlock(&journal->j_checkpoint_mutex);
1751 		} else
1752 			err = -EIO;
1753 		brelse(journal->j_sb_buffer);
1754 	}
1755 
1756 	if (journal->j_proc_entry)
1757 		jbd2_stats_proc_exit(journal);
1758 	iput(journal->j_inode);
1759 	if (journal->j_revoke)
1760 		jbd2_journal_destroy_revoke(journal);
1761 	if (journal->j_chksum_driver)
1762 		crypto_free_shash(journal->j_chksum_driver);
1763 	kfree(journal->j_wbuf);
1764 	kfree(journal);
1765 
1766 	return err;
1767 }
1768 
1769 
1770 /**
1771  *int jbd2_journal_check_used_features () - Check if features specified are used.
1772  * @journal: Journal to check.
1773  * @compat: bitmask of compatible features
1774  * @ro: bitmask of features that force read-only mount
1775  * @incompat: bitmask of incompatible features
1776  *
1777  * Check whether the journal uses all of a given set of
1778  * features.  Return true (non-zero) if it does.
1779  **/
1780 
1781 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1782 				 unsigned long ro, unsigned long incompat)
1783 {
1784 	journal_superblock_t *sb;
1785 
1786 	if (!compat && !ro && !incompat)
1787 		return 1;
1788 	/* Load journal superblock if it is not loaded yet. */
1789 	if (journal->j_format_version == 0 &&
1790 	    journal_get_superblock(journal) != 0)
1791 		return 0;
1792 	if (journal->j_format_version == 1)
1793 		return 0;
1794 
1795 	sb = journal->j_superblock;
1796 
1797 	if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1798 	    ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1799 	    ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1800 		return 1;
1801 
1802 	return 0;
1803 }
1804 
1805 /**
1806  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1807  * @journal: Journal to check.
1808  * @compat: bitmask of compatible features
1809  * @ro: bitmask of features that force read-only mount
1810  * @incompat: bitmask of incompatible features
1811  *
1812  * Check whether the journaling code supports the use of
1813  * all of a given set of features on this journal.  Return true
1814  * (non-zero) if it can. */
1815 
1816 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1817 				      unsigned long ro, unsigned long incompat)
1818 {
1819 	if (!compat && !ro && !incompat)
1820 		return 1;
1821 
1822 	/* We can support any known requested features iff the
1823 	 * superblock is in version 2.  Otherwise we fail to support any
1824 	 * extended sb features. */
1825 
1826 	if (journal->j_format_version != 2)
1827 		return 0;
1828 
1829 	if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1830 	    (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1831 	    (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1832 		return 1;
1833 
1834 	return 0;
1835 }
1836 
1837 /**
1838  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1839  * @journal: Journal to act on.
1840  * @compat: bitmask of compatible features
1841  * @ro: bitmask of features that force read-only mount
1842  * @incompat: bitmask of incompatible features
1843  *
1844  * Mark a given journal feature as present on the
1845  * superblock.  Returns true if the requested features could be set.
1846  *
1847  */
1848 
1849 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1850 			  unsigned long ro, unsigned long incompat)
1851 {
1852 #define INCOMPAT_FEATURE_ON(f) \
1853 		((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
1854 #define COMPAT_FEATURE_ON(f) \
1855 		((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
1856 	journal_superblock_t *sb;
1857 
1858 	if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1859 		return 1;
1860 
1861 	if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1862 		return 0;
1863 
1864 	/* If enabling v2 checksums, turn on v3 instead */
1865 	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
1866 		incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
1867 		incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
1868 	}
1869 
1870 	/* Asking for checksumming v3 and v1?  Only give them v3. */
1871 	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
1872 	    compat & JBD2_FEATURE_COMPAT_CHECKSUM)
1873 		compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
1874 
1875 	jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1876 		  compat, ro, incompat);
1877 
1878 	sb = journal->j_superblock;
1879 
1880 	/* If enabling v3 checksums, update superblock */
1881 	if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
1882 		sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
1883 		sb->s_feature_compat &=
1884 			~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
1885 
1886 		/* Load the checksum driver */
1887 		if (journal->j_chksum_driver == NULL) {
1888 			journal->j_chksum_driver = crypto_alloc_shash("crc32c",
1889 								      0, 0);
1890 			if (IS_ERR(journal->j_chksum_driver)) {
1891 				printk(KERN_ERR "JBD2: Cannot load crc32c "
1892 				       "driver.\n");
1893 				journal->j_chksum_driver = NULL;
1894 				return 0;
1895 			}
1896 
1897 			/* Precompute checksum seed for all metadata */
1898 			journal->j_csum_seed = jbd2_chksum(journal, ~0,
1899 							   sb->s_uuid,
1900 							   sizeof(sb->s_uuid));
1901 		}
1902 	}
1903 
1904 	/* If enabling v1 checksums, downgrade superblock */
1905 	if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
1906 		sb->s_feature_incompat &=
1907 			~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
1908 				     JBD2_FEATURE_INCOMPAT_CSUM_V3);
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 	return 1;
1915 #undef COMPAT_FEATURE_ON
1916 #undef INCOMPAT_FEATURE_ON
1917 }
1918 
1919 /*
1920  * jbd2_journal_clear_features () - Clear a given journal feature in the
1921  * 				    superblock
1922  * @journal: Journal to act on.
1923  * @compat: bitmask of compatible features
1924  * @ro: bitmask of features that force read-only mount
1925  * @incompat: bitmask of incompatible features
1926  *
1927  * Clear a given journal feature as present on the
1928  * superblock.
1929  */
1930 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1931 				unsigned long ro, unsigned long incompat)
1932 {
1933 	journal_superblock_t *sb;
1934 
1935 	jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1936 		  compat, ro, incompat);
1937 
1938 	sb = journal->j_superblock;
1939 
1940 	sb->s_feature_compat    &= ~cpu_to_be32(compat);
1941 	sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1942 	sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1943 }
1944 EXPORT_SYMBOL(jbd2_journal_clear_features);
1945 
1946 /**
1947  * int jbd2_journal_flush () - Flush journal
1948  * @journal: Journal to act on.
1949  *
1950  * Flush all data for a given journal to disk and empty the journal.
1951  * Filesystems can use this when remounting readonly to ensure that
1952  * recovery does not need to happen on remount.
1953  */
1954 
1955 int jbd2_journal_flush(journal_t *journal)
1956 {
1957 	int err = 0;
1958 	transaction_t *transaction = NULL;
1959 
1960 	write_lock(&journal->j_state_lock);
1961 
1962 	/* Force everything buffered to the log... */
1963 	if (journal->j_running_transaction) {
1964 		transaction = journal->j_running_transaction;
1965 		__jbd2_log_start_commit(journal, transaction->t_tid);
1966 	} else if (journal->j_committing_transaction)
1967 		transaction = journal->j_committing_transaction;
1968 
1969 	/* Wait for the log commit to complete... */
1970 	if (transaction) {
1971 		tid_t tid = transaction->t_tid;
1972 
1973 		write_unlock(&journal->j_state_lock);
1974 		jbd2_log_wait_commit(journal, tid);
1975 	} else {
1976 		write_unlock(&journal->j_state_lock);
1977 	}
1978 
1979 	/* ...and flush everything in the log out to disk. */
1980 	spin_lock(&journal->j_list_lock);
1981 	while (!err && journal->j_checkpoint_transactions != NULL) {
1982 		spin_unlock(&journal->j_list_lock);
1983 		mutex_lock(&journal->j_checkpoint_mutex);
1984 		err = jbd2_log_do_checkpoint(journal);
1985 		mutex_unlock(&journal->j_checkpoint_mutex);
1986 		spin_lock(&journal->j_list_lock);
1987 	}
1988 	spin_unlock(&journal->j_list_lock);
1989 
1990 	if (is_journal_aborted(journal))
1991 		return -EIO;
1992 
1993 	mutex_lock(&journal->j_checkpoint_mutex);
1994 	if (!err) {
1995 		err = jbd2_cleanup_journal_tail(journal);
1996 		if (err < 0) {
1997 			mutex_unlock(&journal->j_checkpoint_mutex);
1998 			goto out;
1999 		}
2000 		err = 0;
2001 	}
2002 
2003 	/* Finally, mark the journal as really needing no recovery.
2004 	 * This sets s_start==0 in the underlying superblock, which is
2005 	 * the magic code for a fully-recovered superblock.  Any future
2006 	 * commits of data to the journal will restore the current
2007 	 * s_start value. */
2008 	jbd2_mark_journal_empty(journal, WRITE_FUA);
2009 	mutex_unlock(&journal->j_checkpoint_mutex);
2010 	write_lock(&journal->j_state_lock);
2011 	J_ASSERT(!journal->j_running_transaction);
2012 	J_ASSERT(!journal->j_committing_transaction);
2013 	J_ASSERT(!journal->j_checkpoint_transactions);
2014 	J_ASSERT(journal->j_head == journal->j_tail);
2015 	J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2016 	write_unlock(&journal->j_state_lock);
2017 out:
2018 	return err;
2019 }
2020 
2021 /**
2022  * int jbd2_journal_wipe() - Wipe journal contents
2023  * @journal: Journal to act on.
2024  * @write: flag (see below)
2025  *
2026  * Wipe out all of the contents of a journal, safely.  This will produce
2027  * a warning if the journal contains any valid recovery information.
2028  * Must be called between journal_init_*() and jbd2_journal_load().
2029  *
2030  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2031  * we merely suppress recovery.
2032  */
2033 
2034 int jbd2_journal_wipe(journal_t *journal, int write)
2035 {
2036 	int err = 0;
2037 
2038 	J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2039 
2040 	err = load_superblock(journal);
2041 	if (err)
2042 		return err;
2043 
2044 	if (!journal->j_tail)
2045 		goto no_recovery;
2046 
2047 	printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2048 		write ? "Clearing" : "Ignoring");
2049 
2050 	err = jbd2_journal_skip_recovery(journal);
2051 	if (write) {
2052 		/* Lock to make assertions happy... */
2053 		mutex_lock(&journal->j_checkpoint_mutex);
2054 		jbd2_mark_journal_empty(journal, WRITE_FUA);
2055 		mutex_unlock(&journal->j_checkpoint_mutex);
2056 	}
2057 
2058  no_recovery:
2059 	return err;
2060 }
2061 
2062 /*
2063  * Journal abort has very specific semantics, which we describe
2064  * for journal abort.
2065  *
2066  * Two internal functions, which provide abort to the jbd layer
2067  * itself are here.
2068  */
2069 
2070 /*
2071  * Quick version for internal journal use (doesn't lock the journal).
2072  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
2073  * and don't attempt to make any other journal updates.
2074  */
2075 void __jbd2_journal_abort_hard(journal_t *journal)
2076 {
2077 	transaction_t *transaction;
2078 
2079 	if (journal->j_flags & JBD2_ABORT)
2080 		return;
2081 
2082 	printk(KERN_ERR "Aborting journal on device %s.\n",
2083 	       journal->j_devname);
2084 
2085 	write_lock(&journal->j_state_lock);
2086 	journal->j_flags |= JBD2_ABORT;
2087 	transaction = journal->j_running_transaction;
2088 	if (transaction)
2089 		__jbd2_log_start_commit(journal, transaction->t_tid);
2090 	write_unlock(&journal->j_state_lock);
2091 }
2092 
2093 /* Soft abort: record the abort error status in the journal superblock,
2094  * but don't do any other IO. */
2095 static void __journal_abort_soft (journal_t *journal, int errno)
2096 {
2097 	if (journal->j_flags & JBD2_ABORT)
2098 		return;
2099 
2100 	if (!journal->j_errno)
2101 		journal->j_errno = errno;
2102 
2103 	__jbd2_journal_abort_hard(journal);
2104 
2105 	if (errno) {
2106 		jbd2_journal_update_sb_errno(journal);
2107 		write_lock(&journal->j_state_lock);
2108 		journal->j_flags |= JBD2_REC_ERR;
2109 		write_unlock(&journal->j_state_lock);
2110 	}
2111 }
2112 
2113 /**
2114  * void jbd2_journal_abort () - Shutdown the journal immediately.
2115  * @journal: the journal to shutdown.
2116  * @errno:   an error number to record in the journal indicating
2117  *           the reason for the shutdown.
2118  *
2119  * Perform a complete, immediate shutdown of the ENTIRE
2120  * journal (not of a single transaction).  This operation cannot be
2121  * undone without closing and reopening the journal.
2122  *
2123  * The jbd2_journal_abort function is intended to support higher level error
2124  * recovery mechanisms such as the ext2/ext3 remount-readonly error
2125  * mode.
2126  *
2127  * Journal abort has very specific semantics.  Any existing dirty,
2128  * unjournaled buffers in the main filesystem will still be written to
2129  * disk by bdflush, but the journaling mechanism will be suspended
2130  * immediately and no further transaction commits will be honoured.
2131  *
2132  * Any dirty, journaled buffers will be written back to disk without
2133  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2134  * filesystem, but we _do_ attempt to leave as much data as possible
2135  * behind for fsck to use for cleanup.
2136  *
2137  * Any attempt to get a new transaction handle on a journal which is in
2138  * ABORT state will just result in an -EROFS error return.  A
2139  * jbd2_journal_stop on an existing handle will return -EIO if we have
2140  * entered abort state during the update.
2141  *
2142  * Recursive transactions are not disturbed by journal abort until the
2143  * final jbd2_journal_stop, which will receive the -EIO error.
2144  *
2145  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2146  * which will be recorded (if possible) in the journal superblock.  This
2147  * allows a client to record failure conditions in the middle of a
2148  * transaction without having to complete the transaction to record the
2149  * failure to disk.  ext3_error, for example, now uses this
2150  * functionality.
2151  *
2152  * Errors which originate from within the journaling layer will NOT
2153  * supply an errno; a null errno implies that absolutely no further
2154  * writes are done to the journal (unless there are any already in
2155  * progress).
2156  *
2157  */
2158 
2159 void jbd2_journal_abort(journal_t *journal, int errno)
2160 {
2161 	__journal_abort_soft(journal, errno);
2162 }
2163 
2164 /**
2165  * int jbd2_journal_errno () - returns the journal's error state.
2166  * @journal: journal to examine.
2167  *
2168  * This is the errno number set with jbd2_journal_abort(), the last
2169  * time the journal was mounted - if the journal was stopped
2170  * without calling abort this will be 0.
2171  *
2172  * If the journal has been aborted on this mount time -EROFS will
2173  * be returned.
2174  */
2175 int jbd2_journal_errno(journal_t *journal)
2176 {
2177 	int err;
2178 
2179 	read_lock(&journal->j_state_lock);
2180 	if (journal->j_flags & JBD2_ABORT)
2181 		err = -EROFS;
2182 	else
2183 		err = journal->j_errno;
2184 	read_unlock(&journal->j_state_lock);
2185 	return err;
2186 }
2187 
2188 /**
2189  * int jbd2_journal_clear_err () - clears the journal's error state
2190  * @journal: journal to act on.
2191  *
2192  * An error must be cleared or acked to take a FS out of readonly
2193  * mode.
2194  */
2195 int jbd2_journal_clear_err(journal_t *journal)
2196 {
2197 	int err = 0;
2198 
2199 	write_lock(&journal->j_state_lock);
2200 	if (journal->j_flags & JBD2_ABORT)
2201 		err = -EROFS;
2202 	else
2203 		journal->j_errno = 0;
2204 	write_unlock(&journal->j_state_lock);
2205 	return err;
2206 }
2207 
2208 /**
2209  * void jbd2_journal_ack_err() - Ack journal err.
2210  * @journal: journal to act on.
2211  *
2212  * An error must be cleared or acked to take a FS out of readonly
2213  * mode.
2214  */
2215 void jbd2_journal_ack_err(journal_t *journal)
2216 {
2217 	write_lock(&journal->j_state_lock);
2218 	if (journal->j_errno)
2219 		journal->j_flags |= JBD2_ACK_ERR;
2220 	write_unlock(&journal->j_state_lock);
2221 }
2222 
2223 int jbd2_journal_blocks_per_page(struct inode *inode)
2224 {
2225 	return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2226 }
2227 
2228 /*
2229  * helper functions to deal with 32 or 64bit block numbers.
2230  */
2231 size_t journal_tag_bytes(journal_t *journal)
2232 {
2233 	size_t sz;
2234 
2235 	if (jbd2_has_feature_csum3(journal))
2236 		return sizeof(journal_block_tag3_t);
2237 
2238 	sz = sizeof(journal_block_tag_t);
2239 
2240 	if (jbd2_has_feature_csum2(journal))
2241 		sz += sizeof(__u16);
2242 
2243 	if (jbd2_has_feature_64bit(journal))
2244 		return sz;
2245 	else
2246 		return sz - sizeof(__u32);
2247 }
2248 
2249 /*
2250  * JBD memory management
2251  *
2252  * These functions are used to allocate block-sized chunks of memory
2253  * used for making copies of buffer_head data.  Very often it will be
2254  * page-sized chunks of data, but sometimes it will be in
2255  * sub-page-size chunks.  (For example, 16k pages on Power systems
2256  * with a 4k block file system.)  For blocks smaller than a page, we
2257  * use a SLAB allocator.  There are slab caches for each block size,
2258  * which are allocated at mount time, if necessary, and we only free
2259  * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2260  * this reason we don't need to a mutex to protect access to
2261  * jbd2_slab[] allocating or releasing memory; only in
2262  * jbd2_journal_create_slab().
2263  */
2264 #define JBD2_MAX_SLABS 8
2265 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2266 
2267 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2268 	"jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2269 	"jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2270 };
2271 
2272 
2273 static void jbd2_journal_destroy_slabs(void)
2274 {
2275 	int i;
2276 
2277 	for (i = 0; i < JBD2_MAX_SLABS; i++) {
2278 		if (jbd2_slab[i])
2279 			kmem_cache_destroy(jbd2_slab[i]);
2280 		jbd2_slab[i] = NULL;
2281 	}
2282 }
2283 
2284 static int jbd2_journal_create_slab(size_t size)
2285 {
2286 	static DEFINE_MUTEX(jbd2_slab_create_mutex);
2287 	int i = order_base_2(size) - 10;
2288 	size_t slab_size;
2289 
2290 	if (size == PAGE_SIZE)
2291 		return 0;
2292 
2293 	if (i >= JBD2_MAX_SLABS)
2294 		return -EINVAL;
2295 
2296 	if (unlikely(i < 0))
2297 		i = 0;
2298 	mutex_lock(&jbd2_slab_create_mutex);
2299 	if (jbd2_slab[i]) {
2300 		mutex_unlock(&jbd2_slab_create_mutex);
2301 		return 0;	/* Already created */
2302 	}
2303 
2304 	slab_size = 1 << (i+10);
2305 	jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2306 					 slab_size, 0, NULL);
2307 	mutex_unlock(&jbd2_slab_create_mutex);
2308 	if (!jbd2_slab[i]) {
2309 		printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2310 		return -ENOMEM;
2311 	}
2312 	return 0;
2313 }
2314 
2315 static struct kmem_cache *get_slab(size_t size)
2316 {
2317 	int i = order_base_2(size) - 10;
2318 
2319 	BUG_ON(i >= JBD2_MAX_SLABS);
2320 	if (unlikely(i < 0))
2321 		i = 0;
2322 	BUG_ON(jbd2_slab[i] == NULL);
2323 	return jbd2_slab[i];
2324 }
2325 
2326 void *jbd2_alloc(size_t size, gfp_t flags)
2327 {
2328 	void *ptr;
2329 
2330 	BUG_ON(size & (size-1)); /* Must be a power of 2 */
2331 
2332 	if (size < PAGE_SIZE)
2333 		ptr = kmem_cache_alloc(get_slab(size), flags);
2334 	else
2335 		ptr = (void *)__get_free_pages(flags, get_order(size));
2336 
2337 	/* Check alignment; SLUB has gotten this wrong in the past,
2338 	 * and this can lead to user data corruption! */
2339 	BUG_ON(((unsigned long) ptr) & (size-1));
2340 
2341 	return ptr;
2342 }
2343 
2344 void jbd2_free(void *ptr, size_t size)
2345 {
2346 	if (size < PAGE_SIZE)
2347 		kmem_cache_free(get_slab(size), ptr);
2348 	else
2349 		free_pages((unsigned long)ptr, get_order(size));
2350 };
2351 
2352 /*
2353  * Journal_head storage management
2354  */
2355 static struct kmem_cache *jbd2_journal_head_cache;
2356 #ifdef CONFIG_JBD2_DEBUG
2357 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2358 #endif
2359 
2360 static int jbd2_journal_init_journal_head_cache(void)
2361 {
2362 	int retval;
2363 
2364 	J_ASSERT(jbd2_journal_head_cache == NULL);
2365 	jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2366 				sizeof(struct journal_head),
2367 				0,		/* offset */
2368 				SLAB_TEMPORARY | SLAB_DESTROY_BY_RCU,
2369 				NULL);		/* ctor */
2370 	retval = 0;
2371 	if (!jbd2_journal_head_cache) {
2372 		retval = -ENOMEM;
2373 		printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2374 	}
2375 	return retval;
2376 }
2377 
2378 static void jbd2_journal_destroy_journal_head_cache(void)
2379 {
2380 	if (jbd2_journal_head_cache) {
2381 		kmem_cache_destroy(jbd2_journal_head_cache);
2382 		jbd2_journal_head_cache = NULL;
2383 	}
2384 }
2385 
2386 /*
2387  * journal_head splicing and dicing
2388  */
2389 static struct journal_head *journal_alloc_journal_head(void)
2390 {
2391 	struct journal_head *ret;
2392 
2393 #ifdef CONFIG_JBD2_DEBUG
2394 	atomic_inc(&nr_journal_heads);
2395 #endif
2396 	ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2397 	if (!ret) {
2398 		jbd_debug(1, "out of memory for journal_head\n");
2399 		pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2400 		ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2401 				GFP_NOFS | __GFP_NOFAIL);
2402 	}
2403 	return ret;
2404 }
2405 
2406 static void journal_free_journal_head(struct journal_head *jh)
2407 {
2408 #ifdef CONFIG_JBD2_DEBUG
2409 	atomic_dec(&nr_journal_heads);
2410 	memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2411 #endif
2412 	kmem_cache_free(jbd2_journal_head_cache, jh);
2413 }
2414 
2415 /*
2416  * A journal_head is attached to a buffer_head whenever JBD has an
2417  * interest in the buffer.
2418  *
2419  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2420  * is set.  This bit is tested in core kernel code where we need to take
2421  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2422  * there.
2423  *
2424  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2425  *
2426  * When a buffer has its BH_JBD bit set it is immune from being released by
2427  * core kernel code, mainly via ->b_count.
2428  *
2429  * A journal_head is detached from its buffer_head when the journal_head's
2430  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2431  * transaction (b_cp_transaction) hold their references to b_jcount.
2432  *
2433  * Various places in the kernel want to attach a journal_head to a buffer_head
2434  * _before_ attaching the journal_head to a transaction.  To protect the
2435  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2436  * journal_head's b_jcount refcount by one.  The caller must call
2437  * jbd2_journal_put_journal_head() to undo this.
2438  *
2439  * So the typical usage would be:
2440  *
2441  *	(Attach a journal_head if needed.  Increments b_jcount)
2442  *	struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2443  *	...
2444  *      (Get another reference for transaction)
2445  *	jbd2_journal_grab_journal_head(bh);
2446  *	jh->b_transaction = xxx;
2447  *	(Put original reference)
2448  *	jbd2_journal_put_journal_head(jh);
2449  */
2450 
2451 /*
2452  * Give a buffer_head a journal_head.
2453  *
2454  * May sleep.
2455  */
2456 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2457 {
2458 	struct journal_head *jh;
2459 	struct journal_head *new_jh = NULL;
2460 
2461 repeat:
2462 	if (!buffer_jbd(bh))
2463 		new_jh = journal_alloc_journal_head();
2464 
2465 	jbd_lock_bh_journal_head(bh);
2466 	if (buffer_jbd(bh)) {
2467 		jh = bh2jh(bh);
2468 	} else {
2469 		J_ASSERT_BH(bh,
2470 			(atomic_read(&bh->b_count) > 0) ||
2471 			(bh->b_page && bh->b_page->mapping));
2472 
2473 		if (!new_jh) {
2474 			jbd_unlock_bh_journal_head(bh);
2475 			goto repeat;
2476 		}
2477 
2478 		jh = new_jh;
2479 		new_jh = NULL;		/* We consumed it */
2480 		set_buffer_jbd(bh);
2481 		bh->b_private = jh;
2482 		jh->b_bh = bh;
2483 		get_bh(bh);
2484 		BUFFER_TRACE(bh, "added journal_head");
2485 	}
2486 	jh->b_jcount++;
2487 	jbd_unlock_bh_journal_head(bh);
2488 	if (new_jh)
2489 		journal_free_journal_head(new_jh);
2490 	return bh->b_private;
2491 }
2492 
2493 /*
2494  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2495  * having a journal_head, return NULL
2496  */
2497 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2498 {
2499 	struct journal_head *jh = NULL;
2500 
2501 	jbd_lock_bh_journal_head(bh);
2502 	if (buffer_jbd(bh)) {
2503 		jh = bh2jh(bh);
2504 		jh->b_jcount++;
2505 	}
2506 	jbd_unlock_bh_journal_head(bh);
2507 	return jh;
2508 }
2509 
2510 static void __journal_remove_journal_head(struct buffer_head *bh)
2511 {
2512 	struct journal_head *jh = bh2jh(bh);
2513 
2514 	J_ASSERT_JH(jh, jh->b_jcount >= 0);
2515 	J_ASSERT_JH(jh, jh->b_transaction == NULL);
2516 	J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2517 	J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2518 	J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2519 	J_ASSERT_BH(bh, buffer_jbd(bh));
2520 	J_ASSERT_BH(bh, jh2bh(jh) == bh);
2521 	BUFFER_TRACE(bh, "remove journal_head");
2522 	if (jh->b_frozen_data) {
2523 		printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2524 		jbd2_free(jh->b_frozen_data, bh->b_size);
2525 	}
2526 	if (jh->b_committed_data) {
2527 		printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2528 		jbd2_free(jh->b_committed_data, bh->b_size);
2529 	}
2530 	bh->b_private = NULL;
2531 	jh->b_bh = NULL;	/* debug, really */
2532 	clear_buffer_jbd(bh);
2533 	journal_free_journal_head(jh);
2534 }
2535 
2536 /*
2537  * Drop a reference on the passed journal_head.  If it fell to zero then
2538  * release the journal_head from the buffer_head.
2539  */
2540 void jbd2_journal_put_journal_head(struct journal_head *jh)
2541 {
2542 	struct buffer_head *bh = jh2bh(jh);
2543 
2544 	jbd_lock_bh_journal_head(bh);
2545 	J_ASSERT_JH(jh, jh->b_jcount > 0);
2546 	--jh->b_jcount;
2547 	if (!jh->b_jcount) {
2548 		__journal_remove_journal_head(bh);
2549 		jbd_unlock_bh_journal_head(bh);
2550 		__brelse(bh);
2551 	} else
2552 		jbd_unlock_bh_journal_head(bh);
2553 }
2554 
2555 /*
2556  * Initialize jbd inode head
2557  */
2558 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2559 {
2560 	jinode->i_transaction = NULL;
2561 	jinode->i_next_transaction = NULL;
2562 	jinode->i_vfs_inode = inode;
2563 	jinode->i_flags = 0;
2564 	INIT_LIST_HEAD(&jinode->i_list);
2565 }
2566 
2567 /*
2568  * Function to be called before we start removing inode from memory (i.e.,
2569  * clear_inode() is a fine place to be called from). It removes inode from
2570  * transaction's lists.
2571  */
2572 void jbd2_journal_release_jbd_inode(journal_t *journal,
2573 				    struct jbd2_inode *jinode)
2574 {
2575 	if (!journal)
2576 		return;
2577 restart:
2578 	spin_lock(&journal->j_list_lock);
2579 	/* Is commit writing out inode - we have to wait */
2580 	if (jinode->i_flags & JI_COMMIT_RUNNING) {
2581 		wait_queue_head_t *wq;
2582 		DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2583 		wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2584 		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2585 		spin_unlock(&journal->j_list_lock);
2586 		schedule();
2587 		finish_wait(wq, &wait.wait);
2588 		goto restart;
2589 	}
2590 
2591 	if (jinode->i_transaction) {
2592 		list_del(&jinode->i_list);
2593 		jinode->i_transaction = NULL;
2594 	}
2595 	spin_unlock(&journal->j_list_lock);
2596 }
2597 
2598 
2599 #ifdef CONFIG_PROC_FS
2600 
2601 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2602 
2603 static void __init jbd2_create_jbd_stats_proc_entry(void)
2604 {
2605 	proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2606 }
2607 
2608 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2609 {
2610 	if (proc_jbd2_stats)
2611 		remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2612 }
2613 
2614 #else
2615 
2616 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2617 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2618 
2619 #endif
2620 
2621 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
2622 
2623 static int __init jbd2_journal_init_handle_cache(void)
2624 {
2625 	jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
2626 	if (jbd2_handle_cache == NULL) {
2627 		printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2628 		return -ENOMEM;
2629 	}
2630 	jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2631 	if (jbd2_inode_cache == NULL) {
2632 		printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2633 		kmem_cache_destroy(jbd2_handle_cache);
2634 		return -ENOMEM;
2635 	}
2636 	return 0;
2637 }
2638 
2639 static void jbd2_journal_destroy_handle_cache(void)
2640 {
2641 	if (jbd2_handle_cache)
2642 		kmem_cache_destroy(jbd2_handle_cache);
2643 	if (jbd2_inode_cache)
2644 		kmem_cache_destroy(jbd2_inode_cache);
2645 
2646 }
2647 
2648 /*
2649  * Module startup and shutdown
2650  */
2651 
2652 static int __init journal_init_caches(void)
2653 {
2654 	int ret;
2655 
2656 	ret = jbd2_journal_init_revoke_caches();
2657 	if (ret == 0)
2658 		ret = jbd2_journal_init_journal_head_cache();
2659 	if (ret == 0)
2660 		ret = jbd2_journal_init_handle_cache();
2661 	if (ret == 0)
2662 		ret = jbd2_journal_init_transaction_cache();
2663 	return ret;
2664 }
2665 
2666 static void jbd2_journal_destroy_caches(void)
2667 {
2668 	jbd2_journal_destroy_revoke_caches();
2669 	jbd2_journal_destroy_journal_head_cache();
2670 	jbd2_journal_destroy_handle_cache();
2671 	jbd2_journal_destroy_transaction_cache();
2672 	jbd2_journal_destroy_slabs();
2673 }
2674 
2675 static int __init journal_init(void)
2676 {
2677 	int ret;
2678 
2679 	BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2680 
2681 	ret = journal_init_caches();
2682 	if (ret == 0) {
2683 		jbd2_create_jbd_stats_proc_entry();
2684 	} else {
2685 		jbd2_journal_destroy_caches();
2686 	}
2687 	return ret;
2688 }
2689 
2690 static void __exit journal_exit(void)
2691 {
2692 #ifdef CONFIG_JBD2_DEBUG
2693 	int n = atomic_read(&nr_journal_heads);
2694 	if (n)
2695 		printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
2696 #endif
2697 	jbd2_remove_jbd_stats_proc_entry();
2698 	jbd2_journal_destroy_caches();
2699 }
2700 
2701 MODULE_LICENSE("GPL");
2702 module_init(journal_init);
2703 module_exit(journal_exit);
2704 
2705