xref: /openbmc/linux/fs/jbd2/journal.c (revision 0f613bfa)
1 // SPDX-License-Identifier: GPL-2.0+
2 /*
3  * linux/fs/jbd2/journal.c
4  *
5  * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6  *
7  * Copyright 1998 Red Hat corp --- All Rights Reserved
8  *
9  * Generic filesystem journal-writing code; part of the ext2fs
10  * journaling system.
11  *
12  * This file manages journals: areas of disk reserved for logging
13  * transactional updates.  This includes the kernel journaling thread
14  * which is responsible for scheduling updates to the log.
15  *
16  * We do not actually manage the physical storage of the journal in this
17  * file: that is left to a per-journal policy function, which allows us
18  * to store the journal within a filesystem-specified area for ext2
19  * journaling (ext2 can use a reserved inode for storing the log).
20  */
21 
22 #include <linux/module.h>
23 #include <linux/time.h>
24 #include <linux/fs.h>
25 #include <linux/jbd2.h>
26 #include <linux/errno.h>
27 #include <linux/slab.h>
28 #include <linux/init.h>
29 #include <linux/mm.h>
30 #include <linux/freezer.h>
31 #include <linux/pagemap.h>
32 #include <linux/kthread.h>
33 #include <linux/poison.h>
34 #include <linux/proc_fs.h>
35 #include <linux/seq_file.h>
36 #include <linux/math64.h>
37 #include <linux/hash.h>
38 #include <linux/log2.h>
39 #include <linux/vmalloc.h>
40 #include <linux/backing-dev.h>
41 #include <linux/bitops.h>
42 #include <linux/ratelimit.h>
43 #include <linux/sched/mm.h>
44 
45 #define CREATE_TRACE_POINTS
46 #include <trace/events/jbd2.h>
47 
48 #include <linux/uaccess.h>
49 #include <asm/page.h>
50 
51 #ifdef CONFIG_JBD2_DEBUG
52 static ushort jbd2_journal_enable_debug __read_mostly;
53 
54 module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644);
55 MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2");
56 #endif
57 
58 EXPORT_SYMBOL(jbd2_journal_extend);
59 EXPORT_SYMBOL(jbd2_journal_stop);
60 EXPORT_SYMBOL(jbd2_journal_lock_updates);
61 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
62 EXPORT_SYMBOL(jbd2_journal_get_write_access);
63 EXPORT_SYMBOL(jbd2_journal_get_create_access);
64 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
65 EXPORT_SYMBOL(jbd2_journal_set_triggers);
66 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
67 EXPORT_SYMBOL(jbd2_journal_forget);
68 EXPORT_SYMBOL(jbd2_journal_flush);
69 EXPORT_SYMBOL(jbd2_journal_revoke);
70 
71 EXPORT_SYMBOL(jbd2_journal_init_dev);
72 EXPORT_SYMBOL(jbd2_journal_init_inode);
73 EXPORT_SYMBOL(jbd2_journal_check_used_features);
74 EXPORT_SYMBOL(jbd2_journal_check_available_features);
75 EXPORT_SYMBOL(jbd2_journal_set_features);
76 EXPORT_SYMBOL(jbd2_journal_load);
77 EXPORT_SYMBOL(jbd2_journal_destroy);
78 EXPORT_SYMBOL(jbd2_journal_abort);
79 EXPORT_SYMBOL(jbd2_journal_errno);
80 EXPORT_SYMBOL(jbd2_journal_ack_err);
81 EXPORT_SYMBOL(jbd2_journal_clear_err);
82 EXPORT_SYMBOL(jbd2_log_wait_commit);
83 EXPORT_SYMBOL(jbd2_journal_start_commit);
84 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
85 EXPORT_SYMBOL(jbd2_journal_wipe);
86 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
87 EXPORT_SYMBOL(jbd2_journal_invalidate_folio);
88 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
89 EXPORT_SYMBOL(jbd2_journal_force_commit);
90 EXPORT_SYMBOL(jbd2_journal_inode_ranged_write);
91 EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait);
92 EXPORT_SYMBOL(jbd2_journal_finish_inode_data_buffers);
93 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
94 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
95 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
96 EXPORT_SYMBOL(jbd2_inode_cache);
97 
98 static int jbd2_journal_create_slab(size_t slab_size);
99 
100 #ifdef CONFIG_JBD2_DEBUG
101 void __jbd2_debug(int level, const char *file, const char *func,
102 		  unsigned int line, const char *fmt, ...)
103 {
104 	struct va_format vaf;
105 	va_list args;
106 
107 	if (level > jbd2_journal_enable_debug)
108 		return;
109 	va_start(args, fmt);
110 	vaf.fmt = fmt;
111 	vaf.va = &args;
112 	printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf);
113 	va_end(args);
114 }
115 #endif
116 
117 /* Checksumming functions */
118 static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb)
119 {
120 	if (!jbd2_journal_has_csum_v2or3_feature(j))
121 		return 1;
122 
123 	return sb->s_checksum_type == JBD2_CRC32C_CHKSUM;
124 }
125 
126 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
127 {
128 	__u32 csum;
129 	__be32 old_csum;
130 
131 	old_csum = sb->s_checksum;
132 	sb->s_checksum = 0;
133 	csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
134 	sb->s_checksum = old_csum;
135 
136 	return cpu_to_be32(csum);
137 }
138 
139 /*
140  * Helper function used to manage commit timeouts
141  */
142 
143 static void commit_timeout(struct timer_list *t)
144 {
145 	journal_t *journal = from_timer(journal, t, j_commit_timer);
146 
147 	wake_up_process(journal->j_task);
148 }
149 
150 /*
151  * kjournald2: The main thread function used to manage a logging device
152  * journal.
153  *
154  * This kernel thread is responsible for two things:
155  *
156  * 1) COMMIT:  Every so often we need to commit the current state of the
157  *    filesystem to disk.  The journal thread is responsible for writing
158  *    all of the metadata buffers to disk. If a fast commit is ongoing
159  *    journal thread waits until it's done and then continues from
160  *    there on.
161  *
162  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
163  *    of the data in that part of the log has been rewritten elsewhere on
164  *    the disk.  Flushing these old buffers to reclaim space in the log is
165  *    known as checkpointing, and this thread is responsible for that job.
166  */
167 
168 static int kjournald2(void *arg)
169 {
170 	journal_t *journal = arg;
171 	transaction_t *transaction;
172 
173 	/*
174 	 * Set up an interval timer which can be used to trigger a commit wakeup
175 	 * after the commit interval expires
176 	 */
177 	timer_setup(&journal->j_commit_timer, commit_timeout, 0);
178 
179 	set_freezable();
180 
181 	/* Record that the journal thread is running */
182 	journal->j_task = current;
183 	wake_up(&journal->j_wait_done_commit);
184 
185 	/*
186 	 * Make sure that no allocations from this kernel thread will ever
187 	 * recurse to the fs layer because we are responsible for the
188 	 * transaction commit and any fs involvement might get stuck waiting for
189 	 * the trasn. commit.
190 	 */
191 	memalloc_nofs_save();
192 
193 	/*
194 	 * And now, wait forever for commit wakeup events.
195 	 */
196 	write_lock(&journal->j_state_lock);
197 
198 loop:
199 	if (journal->j_flags & JBD2_UNMOUNT)
200 		goto end_loop;
201 
202 	jbd2_debug(1, "commit_sequence=%u, commit_request=%u\n",
203 		journal->j_commit_sequence, journal->j_commit_request);
204 
205 	if (journal->j_commit_sequence != journal->j_commit_request) {
206 		jbd2_debug(1, "OK, requests differ\n");
207 		write_unlock(&journal->j_state_lock);
208 		del_timer_sync(&journal->j_commit_timer);
209 		jbd2_journal_commit_transaction(journal);
210 		write_lock(&journal->j_state_lock);
211 		goto loop;
212 	}
213 
214 	wake_up(&journal->j_wait_done_commit);
215 	if (freezing(current)) {
216 		/*
217 		 * The simpler the better. Flushing journal isn't a
218 		 * good idea, because that depends on threads that may
219 		 * be already stopped.
220 		 */
221 		jbd2_debug(1, "Now suspending kjournald2\n");
222 		write_unlock(&journal->j_state_lock);
223 		try_to_freeze();
224 		write_lock(&journal->j_state_lock);
225 	} else {
226 		/*
227 		 * We assume on resume that commits are already there,
228 		 * so we don't sleep
229 		 */
230 		DEFINE_WAIT(wait);
231 		int should_sleep = 1;
232 
233 		prepare_to_wait(&journal->j_wait_commit, &wait,
234 				TASK_INTERRUPTIBLE);
235 		if (journal->j_commit_sequence != journal->j_commit_request)
236 			should_sleep = 0;
237 		transaction = journal->j_running_transaction;
238 		if (transaction && time_after_eq(jiffies,
239 						transaction->t_expires))
240 			should_sleep = 0;
241 		if (journal->j_flags & JBD2_UNMOUNT)
242 			should_sleep = 0;
243 		if (should_sleep) {
244 			write_unlock(&journal->j_state_lock);
245 			schedule();
246 			write_lock(&journal->j_state_lock);
247 		}
248 		finish_wait(&journal->j_wait_commit, &wait);
249 	}
250 
251 	jbd2_debug(1, "kjournald2 wakes\n");
252 
253 	/*
254 	 * Were we woken up by a commit wakeup event?
255 	 */
256 	transaction = journal->j_running_transaction;
257 	if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
258 		journal->j_commit_request = transaction->t_tid;
259 		jbd2_debug(1, "woke because of timeout\n");
260 	}
261 	goto loop;
262 
263 end_loop:
264 	del_timer_sync(&journal->j_commit_timer);
265 	journal->j_task = NULL;
266 	wake_up(&journal->j_wait_done_commit);
267 	jbd2_debug(1, "Journal thread exiting.\n");
268 	write_unlock(&journal->j_state_lock);
269 	return 0;
270 }
271 
272 static int jbd2_journal_start_thread(journal_t *journal)
273 {
274 	struct task_struct *t;
275 
276 	t = kthread_run(kjournald2, journal, "jbd2/%s",
277 			journal->j_devname);
278 	if (IS_ERR(t))
279 		return PTR_ERR(t);
280 
281 	wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
282 	return 0;
283 }
284 
285 static void journal_kill_thread(journal_t *journal)
286 {
287 	write_lock(&journal->j_state_lock);
288 	journal->j_flags |= JBD2_UNMOUNT;
289 
290 	while (journal->j_task) {
291 		write_unlock(&journal->j_state_lock);
292 		wake_up(&journal->j_wait_commit);
293 		wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
294 		write_lock(&journal->j_state_lock);
295 	}
296 	write_unlock(&journal->j_state_lock);
297 }
298 
299 /*
300  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
301  *
302  * Writes a metadata buffer to a given disk block.  The actual IO is not
303  * performed but a new buffer_head is constructed which labels the data
304  * to be written with the correct destination disk block.
305  *
306  * Any magic-number escaping which needs to be done will cause a
307  * copy-out here.  If the buffer happens to start with the
308  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
309  * magic number is only written to the log for descripter blocks.  In
310  * this case, we copy the data and replace the first word with 0, and we
311  * return a result code which indicates that this buffer needs to be
312  * marked as an escaped buffer in the corresponding log descriptor
313  * block.  The missing word can then be restored when the block is read
314  * during recovery.
315  *
316  * If the source buffer has already been modified by a new transaction
317  * since we took the last commit snapshot, we use the frozen copy of
318  * that data for IO. If we end up using the existing buffer_head's data
319  * for the write, then we have to make sure nobody modifies it while the
320  * IO is in progress. do_get_write_access() handles this.
321  *
322  * The function returns a pointer to the buffer_head to be used for IO.
323  *
324  *
325  * Return value:
326  *  <0: Error
327  * >=0: Finished OK
328  *
329  * On success:
330  * Bit 0 set == escape performed on the data
331  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
332  */
333 
334 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
335 				  struct journal_head  *jh_in,
336 				  struct buffer_head **bh_out,
337 				  sector_t blocknr)
338 {
339 	int need_copy_out = 0;
340 	int done_copy_out = 0;
341 	int do_escape = 0;
342 	char *mapped_data;
343 	struct buffer_head *new_bh;
344 	struct page *new_page;
345 	unsigned int new_offset;
346 	struct buffer_head *bh_in = jh2bh(jh_in);
347 	journal_t *journal = transaction->t_journal;
348 
349 	/*
350 	 * The buffer really shouldn't be locked: only the current committing
351 	 * transaction is allowed to write it, so nobody else is allowed
352 	 * to do any IO.
353 	 *
354 	 * akpm: except if we're journalling data, and write() output is
355 	 * also part of a shared mapping, and another thread has
356 	 * decided to launch a writepage() against this buffer.
357 	 */
358 	J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
359 
360 	new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
361 
362 	/* keep subsequent assertions sane */
363 	atomic_set(&new_bh->b_count, 1);
364 
365 	spin_lock(&jh_in->b_state_lock);
366 repeat:
367 	/*
368 	 * If a new transaction has already done a buffer copy-out, then
369 	 * we use that version of the data for the commit.
370 	 */
371 	if (jh_in->b_frozen_data) {
372 		done_copy_out = 1;
373 		new_page = virt_to_page(jh_in->b_frozen_data);
374 		new_offset = offset_in_page(jh_in->b_frozen_data);
375 	} else {
376 		new_page = jh2bh(jh_in)->b_page;
377 		new_offset = offset_in_page(jh2bh(jh_in)->b_data);
378 	}
379 
380 	mapped_data = kmap_atomic(new_page);
381 	/*
382 	 * Fire data frozen trigger if data already wasn't frozen.  Do this
383 	 * before checking for escaping, as the trigger may modify the magic
384 	 * offset.  If a copy-out happens afterwards, it will have the correct
385 	 * data in the buffer.
386 	 */
387 	if (!done_copy_out)
388 		jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
389 					   jh_in->b_triggers);
390 
391 	/*
392 	 * Check for escaping
393 	 */
394 	if (*((__be32 *)(mapped_data + new_offset)) ==
395 				cpu_to_be32(JBD2_MAGIC_NUMBER)) {
396 		need_copy_out = 1;
397 		do_escape = 1;
398 	}
399 	kunmap_atomic(mapped_data);
400 
401 	/*
402 	 * Do we need to do a data copy?
403 	 */
404 	if (need_copy_out && !done_copy_out) {
405 		char *tmp;
406 
407 		spin_unlock(&jh_in->b_state_lock);
408 		tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
409 		if (!tmp) {
410 			brelse(new_bh);
411 			return -ENOMEM;
412 		}
413 		spin_lock(&jh_in->b_state_lock);
414 		if (jh_in->b_frozen_data) {
415 			jbd2_free(tmp, bh_in->b_size);
416 			goto repeat;
417 		}
418 
419 		jh_in->b_frozen_data = tmp;
420 		mapped_data = kmap_atomic(new_page);
421 		memcpy(tmp, mapped_data + new_offset, bh_in->b_size);
422 		kunmap_atomic(mapped_data);
423 
424 		new_page = virt_to_page(tmp);
425 		new_offset = offset_in_page(tmp);
426 		done_copy_out = 1;
427 
428 		/*
429 		 * This isn't strictly necessary, as we're using frozen
430 		 * data for the escaping, but it keeps consistency with
431 		 * b_frozen_data usage.
432 		 */
433 		jh_in->b_frozen_triggers = jh_in->b_triggers;
434 	}
435 
436 	/*
437 	 * Did we need to do an escaping?  Now we've done all the
438 	 * copying, we can finally do so.
439 	 */
440 	if (do_escape) {
441 		mapped_data = kmap_atomic(new_page);
442 		*((unsigned int *)(mapped_data + new_offset)) = 0;
443 		kunmap_atomic(mapped_data);
444 	}
445 
446 	set_bh_page(new_bh, new_page, new_offset);
447 	new_bh->b_size = bh_in->b_size;
448 	new_bh->b_bdev = journal->j_dev;
449 	new_bh->b_blocknr = blocknr;
450 	new_bh->b_private = bh_in;
451 	set_buffer_mapped(new_bh);
452 	set_buffer_dirty(new_bh);
453 
454 	*bh_out = new_bh;
455 
456 	/*
457 	 * The to-be-written buffer needs to get moved to the io queue,
458 	 * and the original buffer whose contents we are shadowing or
459 	 * copying is moved to the transaction's shadow queue.
460 	 */
461 	JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
462 	spin_lock(&journal->j_list_lock);
463 	__jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
464 	spin_unlock(&journal->j_list_lock);
465 	set_buffer_shadow(bh_in);
466 	spin_unlock(&jh_in->b_state_lock);
467 
468 	return do_escape | (done_copy_out << 1);
469 }
470 
471 /*
472  * Allocation code for the journal file.  Manage the space left in the
473  * journal, so that we can begin checkpointing when appropriate.
474  */
475 
476 /*
477  * Called with j_state_lock locked for writing.
478  * Returns true if a transaction commit was started.
479  */
480 static int __jbd2_log_start_commit(journal_t *journal, tid_t target)
481 {
482 	/* Return if the txn has already requested to be committed */
483 	if (journal->j_commit_request == target)
484 		return 0;
485 
486 	/*
487 	 * The only transaction we can possibly wait upon is the
488 	 * currently running transaction (if it exists).  Otherwise,
489 	 * the target tid must be an old one.
490 	 */
491 	if (journal->j_running_transaction &&
492 	    journal->j_running_transaction->t_tid == target) {
493 		/*
494 		 * We want a new commit: OK, mark the request and wakeup the
495 		 * commit thread.  We do _not_ do the commit ourselves.
496 		 */
497 
498 		journal->j_commit_request = target;
499 		jbd2_debug(1, "JBD2: requesting commit %u/%u\n",
500 			  journal->j_commit_request,
501 			  journal->j_commit_sequence);
502 		journal->j_running_transaction->t_requested = jiffies;
503 		wake_up(&journal->j_wait_commit);
504 		return 1;
505 	} else if (!tid_geq(journal->j_commit_request, target))
506 		/* This should never happen, but if it does, preserve
507 		   the evidence before kjournald goes into a loop and
508 		   increments j_commit_sequence beyond all recognition. */
509 		WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
510 			  journal->j_commit_request,
511 			  journal->j_commit_sequence,
512 			  target, journal->j_running_transaction ?
513 			  journal->j_running_transaction->t_tid : 0);
514 	return 0;
515 }
516 
517 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
518 {
519 	int ret;
520 
521 	write_lock(&journal->j_state_lock);
522 	ret = __jbd2_log_start_commit(journal, tid);
523 	write_unlock(&journal->j_state_lock);
524 	return ret;
525 }
526 
527 /*
528  * Force and wait any uncommitted transactions.  We can only force the running
529  * transaction if we don't have an active handle, otherwise, we will deadlock.
530  * Returns: <0 in case of error,
531  *           0 if nothing to commit,
532  *           1 if transaction was successfully committed.
533  */
534 static int __jbd2_journal_force_commit(journal_t *journal)
535 {
536 	transaction_t *transaction = NULL;
537 	tid_t tid;
538 	int need_to_start = 0, ret = 0;
539 
540 	read_lock(&journal->j_state_lock);
541 	if (journal->j_running_transaction && !current->journal_info) {
542 		transaction = journal->j_running_transaction;
543 		if (!tid_geq(journal->j_commit_request, transaction->t_tid))
544 			need_to_start = 1;
545 	} else if (journal->j_committing_transaction)
546 		transaction = journal->j_committing_transaction;
547 
548 	if (!transaction) {
549 		/* Nothing to commit */
550 		read_unlock(&journal->j_state_lock);
551 		return 0;
552 	}
553 	tid = transaction->t_tid;
554 	read_unlock(&journal->j_state_lock);
555 	if (need_to_start)
556 		jbd2_log_start_commit(journal, tid);
557 	ret = jbd2_log_wait_commit(journal, tid);
558 	if (!ret)
559 		ret = 1;
560 
561 	return ret;
562 }
563 
564 /**
565  * jbd2_journal_force_commit_nested - Force and wait upon a commit if the
566  * calling process is not within transaction.
567  *
568  * @journal: journal to force
569  * Returns true if progress was made.
570  *
571  * This is used for forcing out undo-protected data which contains
572  * bitmaps, when the fs is running out of space.
573  */
574 int jbd2_journal_force_commit_nested(journal_t *journal)
575 {
576 	int ret;
577 
578 	ret = __jbd2_journal_force_commit(journal);
579 	return ret > 0;
580 }
581 
582 /**
583  * jbd2_journal_force_commit() - force any uncommitted transactions
584  * @journal: journal to force
585  *
586  * Caller want unconditional commit. We can only force the running transaction
587  * if we don't have an active handle, otherwise, we will deadlock.
588  */
589 int jbd2_journal_force_commit(journal_t *journal)
590 {
591 	int ret;
592 
593 	J_ASSERT(!current->journal_info);
594 	ret = __jbd2_journal_force_commit(journal);
595 	if (ret > 0)
596 		ret = 0;
597 	return ret;
598 }
599 
600 /*
601  * Start a commit of the current running transaction (if any).  Returns true
602  * if a transaction is going to be committed (or is currently already
603  * committing), and fills its tid in at *ptid
604  */
605 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
606 {
607 	int ret = 0;
608 
609 	write_lock(&journal->j_state_lock);
610 	if (journal->j_running_transaction) {
611 		tid_t tid = journal->j_running_transaction->t_tid;
612 
613 		__jbd2_log_start_commit(journal, tid);
614 		/* There's a running transaction and we've just made sure
615 		 * it's commit has been scheduled. */
616 		if (ptid)
617 			*ptid = tid;
618 		ret = 1;
619 	} else if (journal->j_committing_transaction) {
620 		/*
621 		 * If commit has been started, then we have to wait for
622 		 * completion of that transaction.
623 		 */
624 		if (ptid)
625 			*ptid = journal->j_committing_transaction->t_tid;
626 		ret = 1;
627 	}
628 	write_unlock(&journal->j_state_lock);
629 	return ret;
630 }
631 
632 /*
633  * Return 1 if a given transaction has not yet sent barrier request
634  * connected with a transaction commit. If 0 is returned, transaction
635  * may or may not have sent the barrier. Used to avoid sending barrier
636  * twice in common cases.
637  */
638 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
639 {
640 	int ret = 0;
641 	transaction_t *commit_trans;
642 
643 	if (!(journal->j_flags & JBD2_BARRIER))
644 		return 0;
645 	read_lock(&journal->j_state_lock);
646 	/* Transaction already committed? */
647 	if (tid_geq(journal->j_commit_sequence, tid))
648 		goto out;
649 	commit_trans = journal->j_committing_transaction;
650 	if (!commit_trans || commit_trans->t_tid != tid) {
651 		ret = 1;
652 		goto out;
653 	}
654 	/*
655 	 * Transaction is being committed and we already proceeded to
656 	 * submitting a flush to fs partition?
657 	 */
658 	if (journal->j_fs_dev != journal->j_dev) {
659 		if (!commit_trans->t_need_data_flush ||
660 		    commit_trans->t_state >= T_COMMIT_DFLUSH)
661 			goto out;
662 	} else {
663 		if (commit_trans->t_state >= T_COMMIT_JFLUSH)
664 			goto out;
665 	}
666 	ret = 1;
667 out:
668 	read_unlock(&journal->j_state_lock);
669 	return ret;
670 }
671 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
672 
673 /*
674  * Wait for a specified commit to complete.
675  * The caller may not hold the journal lock.
676  */
677 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
678 {
679 	int err = 0;
680 
681 	read_lock(&journal->j_state_lock);
682 #ifdef CONFIG_PROVE_LOCKING
683 	/*
684 	 * Some callers make sure transaction is already committing and in that
685 	 * case we cannot block on open handles anymore. So don't warn in that
686 	 * case.
687 	 */
688 	if (tid_gt(tid, journal->j_commit_sequence) &&
689 	    (!journal->j_committing_transaction ||
690 	     journal->j_committing_transaction->t_tid != tid)) {
691 		read_unlock(&journal->j_state_lock);
692 		jbd2_might_wait_for_commit(journal);
693 		read_lock(&journal->j_state_lock);
694 	}
695 #endif
696 #ifdef CONFIG_JBD2_DEBUG
697 	if (!tid_geq(journal->j_commit_request, tid)) {
698 		printk(KERN_ERR
699 		       "%s: error: j_commit_request=%u, tid=%u\n",
700 		       __func__, journal->j_commit_request, tid);
701 	}
702 #endif
703 	while (tid_gt(tid, journal->j_commit_sequence)) {
704 		jbd2_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
705 				  tid, journal->j_commit_sequence);
706 		read_unlock(&journal->j_state_lock);
707 		wake_up(&journal->j_wait_commit);
708 		wait_event(journal->j_wait_done_commit,
709 				!tid_gt(tid, journal->j_commit_sequence));
710 		read_lock(&journal->j_state_lock);
711 	}
712 	read_unlock(&journal->j_state_lock);
713 
714 	if (unlikely(is_journal_aborted(journal)))
715 		err = -EIO;
716 	return err;
717 }
718 
719 /*
720  * Start a fast commit. If there's an ongoing fast or full commit wait for
721  * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY
722  * if a fast commit is not needed, either because there's an already a commit
723  * going on or this tid has already been committed. Returns -EINVAL if no jbd2
724  * commit has yet been performed.
725  */
726 int jbd2_fc_begin_commit(journal_t *journal, tid_t tid)
727 {
728 	if (unlikely(is_journal_aborted(journal)))
729 		return -EIO;
730 	/*
731 	 * Fast commits only allowed if at least one full commit has
732 	 * been processed.
733 	 */
734 	if (!journal->j_stats.ts_tid)
735 		return -EINVAL;
736 
737 	write_lock(&journal->j_state_lock);
738 	if (tid <= journal->j_commit_sequence) {
739 		write_unlock(&journal->j_state_lock);
740 		return -EALREADY;
741 	}
742 
743 	if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
744 	    (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) {
745 		DEFINE_WAIT(wait);
746 
747 		prepare_to_wait(&journal->j_fc_wait, &wait,
748 				TASK_UNINTERRUPTIBLE);
749 		write_unlock(&journal->j_state_lock);
750 		schedule();
751 		finish_wait(&journal->j_fc_wait, &wait);
752 		return -EALREADY;
753 	}
754 	journal->j_flags |= JBD2_FAST_COMMIT_ONGOING;
755 	write_unlock(&journal->j_state_lock);
756 	jbd2_journal_lock_updates(journal);
757 
758 	return 0;
759 }
760 EXPORT_SYMBOL(jbd2_fc_begin_commit);
761 
762 /*
763  * Stop a fast commit. If fallback is set, this function starts commit of
764  * TID tid before any other fast commit can start.
765  */
766 static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback)
767 {
768 	jbd2_journal_unlock_updates(journal);
769 	if (journal->j_fc_cleanup_callback)
770 		journal->j_fc_cleanup_callback(journal, 0, tid);
771 	write_lock(&journal->j_state_lock);
772 	journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
773 	if (fallback)
774 		journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;
775 	write_unlock(&journal->j_state_lock);
776 	wake_up(&journal->j_fc_wait);
777 	if (fallback)
778 		return jbd2_complete_transaction(journal, tid);
779 	return 0;
780 }
781 
782 int jbd2_fc_end_commit(journal_t *journal)
783 {
784 	return __jbd2_fc_end_commit(journal, 0, false);
785 }
786 EXPORT_SYMBOL(jbd2_fc_end_commit);
787 
788 int jbd2_fc_end_commit_fallback(journal_t *journal)
789 {
790 	tid_t tid;
791 
792 	read_lock(&journal->j_state_lock);
793 	tid = journal->j_running_transaction ?
794 		journal->j_running_transaction->t_tid : 0;
795 	read_unlock(&journal->j_state_lock);
796 	return __jbd2_fc_end_commit(journal, tid, true);
797 }
798 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback);
799 
800 /* Return 1 when transaction with given tid has already committed. */
801 int jbd2_transaction_committed(journal_t *journal, tid_t tid)
802 {
803 	int ret = 1;
804 
805 	read_lock(&journal->j_state_lock);
806 	if (journal->j_running_transaction &&
807 	    journal->j_running_transaction->t_tid == tid)
808 		ret = 0;
809 	if (journal->j_committing_transaction &&
810 	    journal->j_committing_transaction->t_tid == tid)
811 		ret = 0;
812 	read_unlock(&journal->j_state_lock);
813 	return ret;
814 }
815 EXPORT_SYMBOL(jbd2_transaction_committed);
816 
817 /*
818  * When this function returns the transaction corresponding to tid
819  * will be completed.  If the transaction has currently running, start
820  * committing that transaction before waiting for it to complete.  If
821  * the transaction id is stale, it is by definition already completed,
822  * so just return SUCCESS.
823  */
824 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
825 {
826 	int	need_to_wait = 1;
827 
828 	read_lock(&journal->j_state_lock);
829 	if (journal->j_running_transaction &&
830 	    journal->j_running_transaction->t_tid == tid) {
831 		if (journal->j_commit_request != tid) {
832 			/* transaction not yet started, so request it */
833 			read_unlock(&journal->j_state_lock);
834 			jbd2_log_start_commit(journal, tid);
835 			goto wait_commit;
836 		}
837 	} else if (!(journal->j_committing_transaction &&
838 		     journal->j_committing_transaction->t_tid == tid))
839 		need_to_wait = 0;
840 	read_unlock(&journal->j_state_lock);
841 	if (!need_to_wait)
842 		return 0;
843 wait_commit:
844 	return jbd2_log_wait_commit(journal, tid);
845 }
846 EXPORT_SYMBOL(jbd2_complete_transaction);
847 
848 /*
849  * Log buffer allocation routines:
850  */
851 
852 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
853 {
854 	unsigned long blocknr;
855 
856 	write_lock(&journal->j_state_lock);
857 	J_ASSERT(journal->j_free > 1);
858 
859 	blocknr = journal->j_head;
860 	journal->j_head++;
861 	journal->j_free--;
862 	if (journal->j_head == journal->j_last)
863 		journal->j_head = journal->j_first;
864 	write_unlock(&journal->j_state_lock);
865 	return jbd2_journal_bmap(journal, blocknr, retp);
866 }
867 
868 /* Map one fast commit buffer for use by the file system */
869 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out)
870 {
871 	unsigned long long pblock;
872 	unsigned long blocknr;
873 	int ret = 0;
874 	struct buffer_head *bh;
875 	int fc_off;
876 
877 	*bh_out = NULL;
878 
879 	if (journal->j_fc_off + journal->j_fc_first < journal->j_fc_last) {
880 		fc_off = journal->j_fc_off;
881 		blocknr = journal->j_fc_first + fc_off;
882 		journal->j_fc_off++;
883 	} else {
884 		ret = -EINVAL;
885 	}
886 
887 	if (ret)
888 		return ret;
889 
890 	ret = jbd2_journal_bmap(journal, blocknr, &pblock);
891 	if (ret)
892 		return ret;
893 
894 	bh = __getblk(journal->j_dev, pblock, journal->j_blocksize);
895 	if (!bh)
896 		return -ENOMEM;
897 
898 
899 	journal->j_fc_wbuf[fc_off] = bh;
900 
901 	*bh_out = bh;
902 
903 	return 0;
904 }
905 EXPORT_SYMBOL(jbd2_fc_get_buf);
906 
907 /*
908  * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
909  * for completion.
910  */
911 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks)
912 {
913 	struct buffer_head *bh;
914 	int i, j_fc_off;
915 
916 	j_fc_off = journal->j_fc_off;
917 
918 	/*
919 	 * Wait in reverse order to minimize chances of us being woken up before
920 	 * all IOs have completed
921 	 */
922 	for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) {
923 		bh = journal->j_fc_wbuf[i];
924 		wait_on_buffer(bh);
925 		/*
926 		 * Update j_fc_off so jbd2_fc_release_bufs can release remain
927 		 * buffer head.
928 		 */
929 		if (unlikely(!buffer_uptodate(bh))) {
930 			journal->j_fc_off = i + 1;
931 			return -EIO;
932 		}
933 		put_bh(bh);
934 		journal->j_fc_wbuf[i] = NULL;
935 	}
936 
937 	return 0;
938 }
939 EXPORT_SYMBOL(jbd2_fc_wait_bufs);
940 
941 int jbd2_fc_release_bufs(journal_t *journal)
942 {
943 	struct buffer_head *bh;
944 	int i, j_fc_off;
945 
946 	j_fc_off = journal->j_fc_off;
947 
948 	for (i = j_fc_off - 1; i >= 0; i--) {
949 		bh = journal->j_fc_wbuf[i];
950 		if (!bh)
951 			break;
952 		put_bh(bh);
953 		journal->j_fc_wbuf[i] = NULL;
954 	}
955 
956 	return 0;
957 }
958 EXPORT_SYMBOL(jbd2_fc_release_bufs);
959 
960 /*
961  * Conversion of logical to physical block numbers for the journal
962  *
963  * On external journals the journal blocks are identity-mapped, so
964  * this is a no-op.  If needed, we can use j_blk_offset - everything is
965  * ready.
966  */
967 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
968 		 unsigned long long *retp)
969 {
970 	int err = 0;
971 	unsigned long long ret;
972 	sector_t block = blocknr;
973 
974 	if (journal->j_bmap) {
975 		err = journal->j_bmap(journal, &block);
976 		if (err == 0)
977 			*retp = block;
978 	} else if (journal->j_inode) {
979 		ret = bmap(journal->j_inode, &block);
980 
981 		if (ret || !block) {
982 			printk(KERN_ALERT "%s: journal block not found "
983 					"at offset %lu on %s\n",
984 			       __func__, blocknr, journal->j_devname);
985 			err = -EIO;
986 			jbd2_journal_abort(journal, err);
987 		} else {
988 			*retp = block;
989 		}
990 
991 	} else {
992 		*retp = blocknr; /* +journal->j_blk_offset */
993 	}
994 	return err;
995 }
996 
997 /*
998  * We play buffer_head aliasing tricks to write data/metadata blocks to
999  * the journal without copying their contents, but for journal
1000  * descriptor blocks we do need to generate bona fide buffers.
1001  *
1002  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
1003  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
1004  * But we don't bother doing that, so there will be coherency problems with
1005  * mmaps of blockdevs which hold live JBD-controlled filesystems.
1006  */
1007 struct buffer_head *
1008 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
1009 {
1010 	journal_t *journal = transaction->t_journal;
1011 	struct buffer_head *bh;
1012 	unsigned long long blocknr;
1013 	journal_header_t *header;
1014 	int err;
1015 
1016 	err = jbd2_journal_next_log_block(journal, &blocknr);
1017 
1018 	if (err)
1019 		return NULL;
1020 
1021 	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1022 	if (!bh)
1023 		return NULL;
1024 	atomic_dec(&transaction->t_outstanding_credits);
1025 	lock_buffer(bh);
1026 	memset(bh->b_data, 0, journal->j_blocksize);
1027 	header = (journal_header_t *)bh->b_data;
1028 	header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
1029 	header->h_blocktype = cpu_to_be32(type);
1030 	header->h_sequence = cpu_to_be32(transaction->t_tid);
1031 	set_buffer_uptodate(bh);
1032 	unlock_buffer(bh);
1033 	BUFFER_TRACE(bh, "return this buffer");
1034 	return bh;
1035 }
1036 
1037 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
1038 {
1039 	struct jbd2_journal_block_tail *tail;
1040 	__u32 csum;
1041 
1042 	if (!jbd2_journal_has_csum_v2or3(j))
1043 		return;
1044 
1045 	tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
1046 			sizeof(struct jbd2_journal_block_tail));
1047 	tail->t_checksum = 0;
1048 	csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
1049 	tail->t_checksum = cpu_to_be32(csum);
1050 }
1051 
1052 /*
1053  * Return tid of the oldest transaction in the journal and block in the journal
1054  * where the transaction starts.
1055  *
1056  * If the journal is now empty, return which will be the next transaction ID
1057  * we will write and where will that transaction start.
1058  *
1059  * The return value is 0 if journal tail cannot be pushed any further, 1 if
1060  * it can.
1061  */
1062 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
1063 			      unsigned long *block)
1064 {
1065 	transaction_t *transaction;
1066 	int ret;
1067 
1068 	read_lock(&journal->j_state_lock);
1069 	spin_lock(&journal->j_list_lock);
1070 	transaction = journal->j_checkpoint_transactions;
1071 	if (transaction) {
1072 		*tid = transaction->t_tid;
1073 		*block = transaction->t_log_start;
1074 	} else if ((transaction = journal->j_committing_transaction) != NULL) {
1075 		*tid = transaction->t_tid;
1076 		*block = transaction->t_log_start;
1077 	} else if ((transaction = journal->j_running_transaction) != NULL) {
1078 		*tid = transaction->t_tid;
1079 		*block = journal->j_head;
1080 	} else {
1081 		*tid = journal->j_transaction_sequence;
1082 		*block = journal->j_head;
1083 	}
1084 	ret = tid_gt(*tid, journal->j_tail_sequence);
1085 	spin_unlock(&journal->j_list_lock);
1086 	read_unlock(&journal->j_state_lock);
1087 
1088 	return ret;
1089 }
1090 
1091 /*
1092  * Update information in journal structure and in on disk journal superblock
1093  * about log tail. This function does not check whether information passed in
1094  * really pushes log tail further. It's responsibility of the caller to make
1095  * sure provided log tail information is valid (e.g. by holding
1096  * j_checkpoint_mutex all the time between computing log tail and calling this
1097  * function as is the case with jbd2_cleanup_journal_tail()).
1098  *
1099  * Requires j_checkpoint_mutex
1100  */
1101 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1102 {
1103 	unsigned long freed;
1104 	int ret;
1105 
1106 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1107 
1108 	/*
1109 	 * We cannot afford for write to remain in drive's caches since as
1110 	 * soon as we update j_tail, next transaction can start reusing journal
1111 	 * space and if we lose sb update during power failure we'd replay
1112 	 * old transaction with possibly newly overwritten data.
1113 	 */
1114 	ret = jbd2_journal_update_sb_log_tail(journal, tid, block,
1115 					      REQ_SYNC | REQ_FUA);
1116 	if (ret)
1117 		goto out;
1118 
1119 	write_lock(&journal->j_state_lock);
1120 	freed = block - journal->j_tail;
1121 	if (block < journal->j_tail)
1122 		freed += journal->j_last - journal->j_first;
1123 
1124 	trace_jbd2_update_log_tail(journal, tid, block, freed);
1125 	jbd2_debug(1,
1126 		  "Cleaning journal tail from %u to %u (offset %lu), "
1127 		  "freeing %lu\n",
1128 		  journal->j_tail_sequence, tid, block, freed);
1129 
1130 	journal->j_free += freed;
1131 	journal->j_tail_sequence = tid;
1132 	journal->j_tail = block;
1133 	write_unlock(&journal->j_state_lock);
1134 
1135 out:
1136 	return ret;
1137 }
1138 
1139 /*
1140  * This is a variation of __jbd2_update_log_tail which checks for validity of
1141  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
1142  * with other threads updating log tail.
1143  */
1144 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1145 {
1146 	mutex_lock_io(&journal->j_checkpoint_mutex);
1147 	if (tid_gt(tid, journal->j_tail_sequence))
1148 		__jbd2_update_log_tail(journal, tid, block);
1149 	mutex_unlock(&journal->j_checkpoint_mutex);
1150 }
1151 
1152 struct jbd2_stats_proc_session {
1153 	journal_t *journal;
1154 	struct transaction_stats_s *stats;
1155 	int start;
1156 	int max;
1157 };
1158 
1159 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
1160 {
1161 	return *pos ? NULL : SEQ_START_TOKEN;
1162 }
1163 
1164 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1165 {
1166 	(*pos)++;
1167 	return NULL;
1168 }
1169 
1170 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1171 {
1172 	struct jbd2_stats_proc_session *s = seq->private;
1173 
1174 	if (v != SEQ_START_TOKEN)
1175 		return 0;
1176 	seq_printf(seq, "%lu transactions (%lu requested), "
1177 		   "each up to %u blocks\n",
1178 		   s->stats->ts_tid, s->stats->ts_requested,
1179 		   s->journal->j_max_transaction_buffers);
1180 	if (s->stats->ts_tid == 0)
1181 		return 0;
1182 	seq_printf(seq, "average: \n  %ums waiting for transaction\n",
1183 	    jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1184 	seq_printf(seq, "  %ums request delay\n",
1185 	    (s->stats->ts_requested == 0) ? 0 :
1186 	    jiffies_to_msecs(s->stats->run.rs_request_delay /
1187 			     s->stats->ts_requested));
1188 	seq_printf(seq, "  %ums running transaction\n",
1189 	    jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1190 	seq_printf(seq, "  %ums transaction was being locked\n",
1191 	    jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1192 	seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
1193 	    jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1194 	seq_printf(seq, "  %ums logging transaction\n",
1195 	    jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1196 	seq_printf(seq, "  %lluus average transaction commit time\n",
1197 		   div_u64(s->journal->j_average_commit_time, 1000));
1198 	seq_printf(seq, "  %lu handles per transaction\n",
1199 	    s->stats->run.rs_handle_count / s->stats->ts_tid);
1200 	seq_printf(seq, "  %lu blocks per transaction\n",
1201 	    s->stats->run.rs_blocks / s->stats->ts_tid);
1202 	seq_printf(seq, "  %lu logged blocks per transaction\n",
1203 	    s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1204 	return 0;
1205 }
1206 
1207 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1208 {
1209 }
1210 
1211 static const struct seq_operations jbd2_seq_info_ops = {
1212 	.start  = jbd2_seq_info_start,
1213 	.next   = jbd2_seq_info_next,
1214 	.stop   = jbd2_seq_info_stop,
1215 	.show   = jbd2_seq_info_show,
1216 };
1217 
1218 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1219 {
1220 	journal_t *journal = pde_data(inode);
1221 	struct jbd2_stats_proc_session *s;
1222 	int rc, size;
1223 
1224 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1225 	if (s == NULL)
1226 		return -ENOMEM;
1227 	size = sizeof(struct transaction_stats_s);
1228 	s->stats = kmalloc(size, GFP_KERNEL);
1229 	if (s->stats == NULL) {
1230 		kfree(s);
1231 		return -ENOMEM;
1232 	}
1233 	spin_lock(&journal->j_history_lock);
1234 	memcpy(s->stats, &journal->j_stats, size);
1235 	s->journal = journal;
1236 	spin_unlock(&journal->j_history_lock);
1237 
1238 	rc = seq_open(file, &jbd2_seq_info_ops);
1239 	if (rc == 0) {
1240 		struct seq_file *m = file->private_data;
1241 		m->private = s;
1242 	} else {
1243 		kfree(s->stats);
1244 		kfree(s);
1245 	}
1246 	return rc;
1247 
1248 }
1249 
1250 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1251 {
1252 	struct seq_file *seq = file->private_data;
1253 	struct jbd2_stats_proc_session *s = seq->private;
1254 	kfree(s->stats);
1255 	kfree(s);
1256 	return seq_release(inode, file);
1257 }
1258 
1259 static const struct proc_ops jbd2_info_proc_ops = {
1260 	.proc_open	= jbd2_seq_info_open,
1261 	.proc_read	= seq_read,
1262 	.proc_lseek	= seq_lseek,
1263 	.proc_release	= jbd2_seq_info_release,
1264 };
1265 
1266 static struct proc_dir_entry *proc_jbd2_stats;
1267 
1268 static void jbd2_stats_proc_init(journal_t *journal)
1269 {
1270 	journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1271 	if (journal->j_proc_entry) {
1272 		proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1273 				 &jbd2_info_proc_ops, journal);
1274 	}
1275 }
1276 
1277 static void jbd2_stats_proc_exit(journal_t *journal)
1278 {
1279 	remove_proc_entry("info", journal->j_proc_entry);
1280 	remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1281 }
1282 
1283 /* Minimum size of descriptor tag */
1284 static int jbd2_min_tag_size(void)
1285 {
1286 	/*
1287 	 * Tag with 32-bit block numbers does not use last four bytes of the
1288 	 * structure
1289 	 */
1290 	return sizeof(journal_block_tag_t) - 4;
1291 }
1292 
1293 /**
1294  * jbd2_journal_shrink_scan()
1295  * @shrink: shrinker to work on
1296  * @sc: reclaim request to process
1297  *
1298  * Scan the checkpointed buffer on the checkpoint list and release the
1299  * journal_head.
1300  */
1301 static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink,
1302 					      struct shrink_control *sc)
1303 {
1304 	journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1305 	unsigned long nr_to_scan = sc->nr_to_scan;
1306 	unsigned long nr_shrunk;
1307 	unsigned long count;
1308 
1309 	count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1310 	trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count);
1311 
1312 	nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan);
1313 
1314 	count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1315 	trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count);
1316 
1317 	return nr_shrunk;
1318 }
1319 
1320 /**
1321  * jbd2_journal_shrink_count()
1322  * @shrink: shrinker to work on
1323  * @sc: reclaim request to process
1324  *
1325  * Count the number of checkpoint buffers on the checkpoint list.
1326  */
1327 static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink,
1328 					       struct shrink_control *sc)
1329 {
1330 	journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1331 	unsigned long count;
1332 
1333 	count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1334 	trace_jbd2_shrink_count(journal, sc->nr_to_scan, count);
1335 
1336 	return count;
1337 }
1338 
1339 /*
1340  * Management for journal control blocks: functions to create and
1341  * destroy journal_t structures, and to initialise and read existing
1342  * journal blocks from disk.  */
1343 
1344 /* First: create and setup a journal_t object in memory.  We initialise
1345  * very few fields yet: that has to wait until we have created the
1346  * journal structures from from scratch, or loaded them from disk. */
1347 
1348 static journal_t *journal_init_common(struct block_device *bdev,
1349 			struct block_device *fs_dev,
1350 			unsigned long long start, int len, int blocksize)
1351 {
1352 	static struct lock_class_key jbd2_trans_commit_key;
1353 	journal_t *journal;
1354 	int err;
1355 	struct buffer_head *bh;
1356 	int n;
1357 
1358 	journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1359 	if (!journal)
1360 		return NULL;
1361 
1362 	init_waitqueue_head(&journal->j_wait_transaction_locked);
1363 	init_waitqueue_head(&journal->j_wait_done_commit);
1364 	init_waitqueue_head(&journal->j_wait_commit);
1365 	init_waitqueue_head(&journal->j_wait_updates);
1366 	init_waitqueue_head(&journal->j_wait_reserved);
1367 	init_waitqueue_head(&journal->j_fc_wait);
1368 	mutex_init(&journal->j_abort_mutex);
1369 	mutex_init(&journal->j_barrier);
1370 	mutex_init(&journal->j_checkpoint_mutex);
1371 	spin_lock_init(&journal->j_revoke_lock);
1372 	spin_lock_init(&journal->j_list_lock);
1373 	rwlock_init(&journal->j_state_lock);
1374 
1375 	journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1376 	journal->j_min_batch_time = 0;
1377 	journal->j_max_batch_time = 15000; /* 15ms */
1378 	atomic_set(&journal->j_reserved_credits, 0);
1379 
1380 	/* The journal is marked for error until we succeed with recovery! */
1381 	journal->j_flags = JBD2_ABORT;
1382 
1383 	/* Set up a default-sized revoke table for the new mount. */
1384 	err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1385 	if (err)
1386 		goto err_cleanup;
1387 
1388 	spin_lock_init(&journal->j_history_lock);
1389 
1390 	lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1391 			 &jbd2_trans_commit_key, 0);
1392 
1393 	/* journal descriptor can store up to n blocks -bzzz */
1394 	journal->j_blocksize = blocksize;
1395 	journal->j_dev = bdev;
1396 	journal->j_fs_dev = fs_dev;
1397 	journal->j_blk_offset = start;
1398 	journal->j_total_len = len;
1399 	/* We need enough buffers to write out full descriptor block. */
1400 	n = journal->j_blocksize / jbd2_min_tag_size();
1401 	journal->j_wbufsize = n;
1402 	journal->j_fc_wbuf = NULL;
1403 	journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1404 					GFP_KERNEL);
1405 	if (!journal->j_wbuf)
1406 		goto err_cleanup;
1407 
1408 	bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize);
1409 	if (!bh) {
1410 		pr_err("%s: Cannot get buffer for journal superblock\n",
1411 			__func__);
1412 		goto err_cleanup;
1413 	}
1414 	journal->j_sb_buffer = bh;
1415 	journal->j_superblock = (journal_superblock_t *)bh->b_data;
1416 
1417 	journal->j_shrink_transaction = NULL;
1418 	journal->j_shrinker.scan_objects = jbd2_journal_shrink_scan;
1419 	journal->j_shrinker.count_objects = jbd2_journal_shrink_count;
1420 	journal->j_shrinker.seeks = DEFAULT_SEEKS;
1421 	journal->j_shrinker.batch = journal->j_max_transaction_buffers;
1422 
1423 	if (percpu_counter_init(&journal->j_checkpoint_jh_count, 0, GFP_KERNEL))
1424 		goto err_cleanup;
1425 
1426 	if (register_shrinker(&journal->j_shrinker, "jbd2-journal:(%u:%u)",
1427 			      MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev))) {
1428 		percpu_counter_destroy(&journal->j_checkpoint_jh_count);
1429 		goto err_cleanup;
1430 	}
1431 	return journal;
1432 
1433 err_cleanup:
1434 	brelse(journal->j_sb_buffer);
1435 	kfree(journal->j_wbuf);
1436 	jbd2_journal_destroy_revoke(journal);
1437 	kfree(journal);
1438 	return NULL;
1439 }
1440 
1441 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1442  *
1443  * Create a journal structure assigned some fixed set of disk blocks to
1444  * the journal.  We don't actually touch those disk blocks yet, but we
1445  * need to set up all of the mapping information to tell the journaling
1446  * system where the journal blocks are.
1447  *
1448  */
1449 
1450 /**
1451  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1452  *  @bdev: Block device on which to create the journal
1453  *  @fs_dev: Device which hold journalled filesystem for this journal.
1454  *  @start: Block nr Start of journal.
1455  *  @len:  Length of the journal in blocks.
1456  *  @blocksize: blocksize of journalling device
1457  *
1458  *  Returns: a newly created journal_t *
1459  *
1460  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1461  *  range of blocks on an arbitrary block device.
1462  *
1463  */
1464 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1465 			struct block_device *fs_dev,
1466 			unsigned long long start, int len, int blocksize)
1467 {
1468 	journal_t *journal;
1469 
1470 	journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1471 	if (!journal)
1472 		return NULL;
1473 
1474 	snprintf(journal->j_devname, sizeof(journal->j_devname),
1475 		 "%pg", journal->j_dev);
1476 	strreplace(journal->j_devname, '/', '!');
1477 	jbd2_stats_proc_init(journal);
1478 
1479 	return journal;
1480 }
1481 
1482 /**
1483  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1484  *  @inode: An inode to create the journal in
1485  *
1486  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1487  * the journal.  The inode must exist already, must support bmap() and
1488  * must have all data blocks preallocated.
1489  */
1490 journal_t *jbd2_journal_init_inode(struct inode *inode)
1491 {
1492 	journal_t *journal;
1493 	sector_t blocknr;
1494 	char *p;
1495 	int err = 0;
1496 
1497 	blocknr = 0;
1498 	err = bmap(inode, &blocknr);
1499 
1500 	if (err || !blocknr) {
1501 		pr_err("%s: Cannot locate journal superblock\n",
1502 			__func__);
1503 		return NULL;
1504 	}
1505 
1506 	jbd2_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1507 		  inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1508 		  inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1509 
1510 	journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1511 			blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1512 			inode->i_sb->s_blocksize);
1513 	if (!journal)
1514 		return NULL;
1515 
1516 	journal->j_inode = inode;
1517 	snprintf(journal->j_devname, sizeof(journal->j_devname),
1518 		 "%pg", journal->j_dev);
1519 	p = strreplace(journal->j_devname, '/', '!');
1520 	sprintf(p, "-%lu", journal->j_inode->i_ino);
1521 	jbd2_stats_proc_init(journal);
1522 
1523 	return journal;
1524 }
1525 
1526 /*
1527  * If the journal init or create aborts, we need to mark the journal
1528  * superblock as being NULL to prevent the journal destroy from writing
1529  * back a bogus superblock.
1530  */
1531 static void journal_fail_superblock(journal_t *journal)
1532 {
1533 	struct buffer_head *bh = journal->j_sb_buffer;
1534 	brelse(bh);
1535 	journal->j_sb_buffer = NULL;
1536 }
1537 
1538 /*
1539  * Given a journal_t structure, initialise the various fields for
1540  * startup of a new journaling session.  We use this both when creating
1541  * a journal, and after recovering an old journal to reset it for
1542  * subsequent use.
1543  */
1544 
1545 static int journal_reset(journal_t *journal)
1546 {
1547 	journal_superblock_t *sb = journal->j_superblock;
1548 	unsigned long long first, last;
1549 
1550 	first = be32_to_cpu(sb->s_first);
1551 	last = be32_to_cpu(sb->s_maxlen);
1552 	if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1553 		printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1554 		       first, last);
1555 		journal_fail_superblock(journal);
1556 		return -EINVAL;
1557 	}
1558 
1559 	journal->j_first = first;
1560 	journal->j_last = last;
1561 
1562 	journal->j_head = journal->j_first;
1563 	journal->j_tail = journal->j_first;
1564 	journal->j_free = journal->j_last - journal->j_first;
1565 
1566 	journal->j_tail_sequence = journal->j_transaction_sequence;
1567 	journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1568 	journal->j_commit_request = journal->j_commit_sequence;
1569 
1570 	journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal);
1571 
1572 	/*
1573 	 * Now that journal recovery is done, turn fast commits off here. This
1574 	 * way, if fast commit was enabled before the crash but if now FS has
1575 	 * disabled it, we don't enable fast commits.
1576 	 */
1577 	jbd2_clear_feature_fast_commit(journal);
1578 
1579 	/*
1580 	 * As a special case, if the on-disk copy is already marked as needing
1581 	 * no recovery (s_start == 0), then we can safely defer the superblock
1582 	 * update until the next commit by setting JBD2_FLUSHED.  This avoids
1583 	 * attempting a write to a potential-readonly device.
1584 	 */
1585 	if (sb->s_start == 0) {
1586 		jbd2_debug(1, "JBD2: Skipping superblock update on recovered sb "
1587 			"(start %ld, seq %u, errno %d)\n",
1588 			journal->j_tail, journal->j_tail_sequence,
1589 			journal->j_errno);
1590 		journal->j_flags |= JBD2_FLUSHED;
1591 	} else {
1592 		/* Lock here to make assertions happy... */
1593 		mutex_lock_io(&journal->j_checkpoint_mutex);
1594 		/*
1595 		 * Update log tail information. We use REQ_FUA since new
1596 		 * transaction will start reusing journal space and so we
1597 		 * must make sure information about current log tail is on
1598 		 * disk before that.
1599 		 */
1600 		jbd2_journal_update_sb_log_tail(journal,
1601 						journal->j_tail_sequence,
1602 						journal->j_tail,
1603 						REQ_SYNC | REQ_FUA);
1604 		mutex_unlock(&journal->j_checkpoint_mutex);
1605 	}
1606 	return jbd2_journal_start_thread(journal);
1607 }
1608 
1609 /*
1610  * This function expects that the caller will have locked the journal
1611  * buffer head, and will return with it unlocked
1612  */
1613 static int jbd2_write_superblock(journal_t *journal, blk_opf_t write_flags)
1614 {
1615 	struct buffer_head *bh = journal->j_sb_buffer;
1616 	journal_superblock_t *sb = journal->j_superblock;
1617 	int ret = 0;
1618 
1619 	/* Buffer got discarded which means block device got invalidated */
1620 	if (!buffer_mapped(bh)) {
1621 		unlock_buffer(bh);
1622 		return -EIO;
1623 	}
1624 
1625 	trace_jbd2_write_superblock(journal, write_flags);
1626 	if (!(journal->j_flags & JBD2_BARRIER))
1627 		write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1628 	if (buffer_write_io_error(bh)) {
1629 		/*
1630 		 * Oh, dear.  A previous attempt to write the journal
1631 		 * superblock failed.  This could happen because the
1632 		 * USB device was yanked out.  Or it could happen to
1633 		 * be a transient write error and maybe the block will
1634 		 * be remapped.  Nothing we can do but to retry the
1635 		 * write and hope for the best.
1636 		 */
1637 		printk(KERN_ERR "JBD2: previous I/O error detected "
1638 		       "for journal superblock update for %s.\n",
1639 		       journal->j_devname);
1640 		clear_buffer_write_io_error(bh);
1641 		set_buffer_uptodate(bh);
1642 	}
1643 	if (jbd2_journal_has_csum_v2or3(journal))
1644 		sb->s_checksum = jbd2_superblock_csum(journal, sb);
1645 	get_bh(bh);
1646 	bh->b_end_io = end_buffer_write_sync;
1647 	submit_bh(REQ_OP_WRITE | write_flags, bh);
1648 	wait_on_buffer(bh);
1649 	if (buffer_write_io_error(bh)) {
1650 		clear_buffer_write_io_error(bh);
1651 		set_buffer_uptodate(bh);
1652 		ret = -EIO;
1653 	}
1654 	if (ret) {
1655 		printk(KERN_ERR "JBD2: I/O error when updating journal superblock for %s.\n",
1656 				journal->j_devname);
1657 		if (!is_journal_aborted(journal))
1658 			jbd2_journal_abort(journal, ret);
1659 	}
1660 
1661 	return ret;
1662 }
1663 
1664 /**
1665  * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1666  * @journal: The journal to update.
1667  * @tail_tid: TID of the new transaction at the tail of the log
1668  * @tail_block: The first block of the transaction at the tail of the log
1669  * @write_flags: Flags for the journal sb write operation
1670  *
1671  * Update a journal's superblock information about log tail and write it to
1672  * disk, waiting for the IO to complete.
1673  */
1674 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1675 				    unsigned long tail_block,
1676 				    blk_opf_t write_flags)
1677 {
1678 	journal_superblock_t *sb = journal->j_superblock;
1679 	int ret;
1680 
1681 	if (is_journal_aborted(journal))
1682 		return -EIO;
1683 	if (test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) {
1684 		jbd2_journal_abort(journal, -EIO);
1685 		return -EIO;
1686 	}
1687 
1688 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1689 	jbd2_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1690 		  tail_block, tail_tid);
1691 
1692 	lock_buffer(journal->j_sb_buffer);
1693 	sb->s_sequence = cpu_to_be32(tail_tid);
1694 	sb->s_start    = cpu_to_be32(tail_block);
1695 
1696 	ret = jbd2_write_superblock(journal, write_flags);
1697 	if (ret)
1698 		goto out;
1699 
1700 	/* Log is no longer empty */
1701 	write_lock(&journal->j_state_lock);
1702 	WARN_ON(!sb->s_sequence);
1703 	journal->j_flags &= ~JBD2_FLUSHED;
1704 	write_unlock(&journal->j_state_lock);
1705 
1706 out:
1707 	return ret;
1708 }
1709 
1710 /**
1711  * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1712  * @journal: The journal to update.
1713  * @write_flags: Flags for the journal sb write operation
1714  *
1715  * Update a journal's dynamic superblock fields to show that journal is empty.
1716  * Write updated superblock to disk waiting for IO to complete.
1717  */
1718 static void jbd2_mark_journal_empty(journal_t *journal, blk_opf_t write_flags)
1719 {
1720 	journal_superblock_t *sb = journal->j_superblock;
1721 	bool had_fast_commit = false;
1722 
1723 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1724 	lock_buffer(journal->j_sb_buffer);
1725 	if (sb->s_start == 0) {		/* Is it already empty? */
1726 		unlock_buffer(journal->j_sb_buffer);
1727 		return;
1728 	}
1729 
1730 	jbd2_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1731 		  journal->j_tail_sequence);
1732 
1733 	sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1734 	sb->s_start    = cpu_to_be32(0);
1735 	if (jbd2_has_feature_fast_commit(journal)) {
1736 		/*
1737 		 * When journal is clean, no need to commit fast commit flag and
1738 		 * make file system incompatible with older kernels.
1739 		 */
1740 		jbd2_clear_feature_fast_commit(journal);
1741 		had_fast_commit = true;
1742 	}
1743 
1744 	jbd2_write_superblock(journal, write_flags);
1745 
1746 	if (had_fast_commit)
1747 		jbd2_set_feature_fast_commit(journal);
1748 
1749 	/* Log is no longer empty */
1750 	write_lock(&journal->j_state_lock);
1751 	journal->j_flags |= JBD2_FLUSHED;
1752 	write_unlock(&journal->j_state_lock);
1753 }
1754 
1755 /**
1756  * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock)
1757  * @journal: The journal to erase.
1758  * @flags: A discard/zeroout request is sent for each physically contigous
1759  *	region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or
1760  *	JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation
1761  *	to perform.
1762  *
1763  * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes
1764  * will be explicitly written if no hardware offload is available, see
1765  * blkdev_issue_zeroout for more details.
1766  */
1767 static int __jbd2_journal_erase(journal_t *journal, unsigned int flags)
1768 {
1769 	int err = 0;
1770 	unsigned long block, log_offset; /* logical */
1771 	unsigned long long phys_block, block_start, block_stop; /* physical */
1772 	loff_t byte_start, byte_stop, byte_count;
1773 
1774 	/* flags must be set to either discard or zeroout */
1775 	if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags ||
1776 			((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1777 			(flags & JBD2_JOURNAL_FLUSH_ZEROOUT)))
1778 		return -EINVAL;
1779 
1780 	if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1781 	    !bdev_max_discard_sectors(journal->j_dev))
1782 		return -EOPNOTSUPP;
1783 
1784 	/*
1785 	 * lookup block mapping and issue discard/zeroout for each
1786 	 * contiguous region
1787 	 */
1788 	log_offset = be32_to_cpu(journal->j_superblock->s_first);
1789 	block_start =  ~0ULL;
1790 	for (block = log_offset; block < journal->j_total_len; block++) {
1791 		err = jbd2_journal_bmap(journal, block, &phys_block);
1792 		if (err) {
1793 			pr_err("JBD2: bad block at offset %lu", block);
1794 			return err;
1795 		}
1796 
1797 		if (block_start == ~0ULL) {
1798 			block_start = phys_block;
1799 			block_stop = block_start - 1;
1800 		}
1801 
1802 		/*
1803 		 * last block not contiguous with current block,
1804 		 * process last contiguous region and return to this block on
1805 		 * next loop
1806 		 */
1807 		if (phys_block != block_stop + 1) {
1808 			block--;
1809 		} else {
1810 			block_stop++;
1811 			/*
1812 			 * if this isn't the last block of journal,
1813 			 * no need to process now because next block may also
1814 			 * be part of this contiguous region
1815 			 */
1816 			if (block != journal->j_total_len - 1)
1817 				continue;
1818 		}
1819 
1820 		/*
1821 		 * end of contiguous region or this is last block of journal,
1822 		 * take care of the region
1823 		 */
1824 		byte_start = block_start * journal->j_blocksize;
1825 		byte_stop = block_stop * journal->j_blocksize;
1826 		byte_count = (block_stop - block_start + 1) *
1827 				journal->j_blocksize;
1828 
1829 		truncate_inode_pages_range(journal->j_dev->bd_inode->i_mapping,
1830 				byte_start, byte_stop);
1831 
1832 		if (flags & JBD2_JOURNAL_FLUSH_DISCARD) {
1833 			err = blkdev_issue_discard(journal->j_dev,
1834 					byte_start >> SECTOR_SHIFT,
1835 					byte_count >> SECTOR_SHIFT,
1836 					GFP_NOFS);
1837 		} else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) {
1838 			err = blkdev_issue_zeroout(journal->j_dev,
1839 					byte_start >> SECTOR_SHIFT,
1840 					byte_count >> SECTOR_SHIFT,
1841 					GFP_NOFS, 0);
1842 		}
1843 
1844 		if (unlikely(err != 0)) {
1845 			pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu",
1846 					err, block_start, block_stop);
1847 			return err;
1848 		}
1849 
1850 		/* reset start and stop after processing a region */
1851 		block_start = ~0ULL;
1852 	}
1853 
1854 	return blkdev_issue_flush(journal->j_dev);
1855 }
1856 
1857 /**
1858  * jbd2_journal_update_sb_errno() - Update error in the journal.
1859  * @journal: The journal to update.
1860  *
1861  * Update a journal's errno.  Write updated superblock to disk waiting for IO
1862  * to complete.
1863  */
1864 void jbd2_journal_update_sb_errno(journal_t *journal)
1865 {
1866 	journal_superblock_t *sb = journal->j_superblock;
1867 	int errcode;
1868 
1869 	lock_buffer(journal->j_sb_buffer);
1870 	errcode = journal->j_errno;
1871 	if (errcode == -ESHUTDOWN)
1872 		errcode = 0;
1873 	jbd2_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
1874 	sb->s_errno    = cpu_to_be32(errcode);
1875 
1876 	jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA);
1877 }
1878 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
1879 
1880 static int journal_revoke_records_per_block(journal_t *journal)
1881 {
1882 	int record_size;
1883 	int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1884 
1885 	if (jbd2_has_feature_64bit(journal))
1886 		record_size = 8;
1887 	else
1888 		record_size = 4;
1889 
1890 	if (jbd2_journal_has_csum_v2or3(journal))
1891 		space -= sizeof(struct jbd2_journal_block_tail);
1892 	return space / record_size;
1893 }
1894 
1895 /*
1896  * Read the superblock for a given journal, performing initial
1897  * validation of the format.
1898  */
1899 static int journal_get_superblock(journal_t *journal)
1900 {
1901 	struct buffer_head *bh;
1902 	journal_superblock_t *sb;
1903 	int err;
1904 
1905 	bh = journal->j_sb_buffer;
1906 
1907 	J_ASSERT(bh != NULL);
1908 	err = bh_read(bh, 0);
1909 	if (err < 0) {
1910 		printk(KERN_ERR
1911 			"JBD2: IO error reading journal superblock\n");
1912 		goto out;
1913 	}
1914 
1915 	if (buffer_verified(bh))
1916 		return 0;
1917 
1918 	sb = journal->j_superblock;
1919 
1920 	err = -EINVAL;
1921 
1922 	if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1923 	    sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1924 		printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1925 		goto out;
1926 	}
1927 
1928 	switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1929 	case JBD2_SUPERBLOCK_V1:
1930 		journal->j_format_version = 1;
1931 		break;
1932 	case JBD2_SUPERBLOCK_V2:
1933 		journal->j_format_version = 2;
1934 		break;
1935 	default:
1936 		printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1937 		goto out;
1938 	}
1939 
1940 	if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len)
1941 		journal->j_total_len = be32_to_cpu(sb->s_maxlen);
1942 	else if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) {
1943 		printk(KERN_WARNING "JBD2: journal file too short\n");
1944 		goto out;
1945 	}
1946 
1947 	if (be32_to_cpu(sb->s_first) == 0 ||
1948 	    be32_to_cpu(sb->s_first) >= journal->j_total_len) {
1949 		printk(KERN_WARNING
1950 			"JBD2: Invalid start block of journal: %u\n",
1951 			be32_to_cpu(sb->s_first));
1952 		goto out;
1953 	}
1954 
1955 	if (jbd2_has_feature_csum2(journal) &&
1956 	    jbd2_has_feature_csum3(journal)) {
1957 		/* Can't have checksum v2 and v3 at the same time! */
1958 		printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1959 		       "at the same time!\n");
1960 		goto out;
1961 	}
1962 
1963 	if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1964 	    jbd2_has_feature_checksum(journal)) {
1965 		/* Can't have checksum v1 and v2 on at the same time! */
1966 		printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1967 		       "at the same time!\n");
1968 		goto out;
1969 	}
1970 
1971 	if (!jbd2_verify_csum_type(journal, sb)) {
1972 		printk(KERN_ERR "JBD2: Unknown checksum type\n");
1973 		goto out;
1974 	}
1975 
1976 	/* Load the checksum driver */
1977 	if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1978 		journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1979 		if (IS_ERR(journal->j_chksum_driver)) {
1980 			printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1981 			err = PTR_ERR(journal->j_chksum_driver);
1982 			journal->j_chksum_driver = NULL;
1983 			goto out;
1984 		}
1985 	}
1986 
1987 	if (jbd2_journal_has_csum_v2or3(journal)) {
1988 		/* Check superblock checksum */
1989 		if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1990 			printk(KERN_ERR "JBD2: journal checksum error\n");
1991 			err = -EFSBADCRC;
1992 			goto out;
1993 		}
1994 
1995 		/* Precompute checksum seed for all metadata */
1996 		journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1997 						   sizeof(sb->s_uuid));
1998 	}
1999 
2000 	journal->j_revoke_records_per_block =
2001 				journal_revoke_records_per_block(journal);
2002 	set_buffer_verified(bh);
2003 
2004 	return 0;
2005 
2006 out:
2007 	journal_fail_superblock(journal);
2008 	return err;
2009 }
2010 
2011 /*
2012  * Load the on-disk journal superblock and read the key fields into the
2013  * journal_t.
2014  */
2015 
2016 static int load_superblock(journal_t *journal)
2017 {
2018 	int err;
2019 	journal_superblock_t *sb;
2020 	int num_fc_blocks;
2021 
2022 	err = journal_get_superblock(journal);
2023 	if (err)
2024 		return err;
2025 
2026 	sb = journal->j_superblock;
2027 
2028 	journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
2029 	journal->j_tail = be32_to_cpu(sb->s_start);
2030 	journal->j_first = be32_to_cpu(sb->s_first);
2031 	journal->j_errno = be32_to_cpu(sb->s_errno);
2032 	journal->j_last = be32_to_cpu(sb->s_maxlen);
2033 
2034 	if (jbd2_has_feature_fast_commit(journal)) {
2035 		journal->j_fc_last = be32_to_cpu(sb->s_maxlen);
2036 		num_fc_blocks = jbd2_journal_get_num_fc_blks(sb);
2037 		if (journal->j_last - num_fc_blocks >= JBD2_MIN_JOURNAL_BLOCKS)
2038 			journal->j_last = journal->j_fc_last - num_fc_blocks;
2039 		journal->j_fc_first = journal->j_last + 1;
2040 		journal->j_fc_off = 0;
2041 	}
2042 
2043 	return 0;
2044 }
2045 
2046 
2047 /**
2048  * jbd2_journal_load() - Read journal from disk.
2049  * @journal: Journal to act on.
2050  *
2051  * Given a journal_t structure which tells us which disk blocks contain
2052  * a journal, read the journal from disk to initialise the in-memory
2053  * structures.
2054  */
2055 int jbd2_journal_load(journal_t *journal)
2056 {
2057 	int err;
2058 	journal_superblock_t *sb;
2059 
2060 	err = load_superblock(journal);
2061 	if (err)
2062 		return err;
2063 
2064 	sb = journal->j_superblock;
2065 	/* If this is a V2 superblock, then we have to check the
2066 	 * features flags on it. */
2067 
2068 	if (journal->j_format_version >= 2) {
2069 		if ((sb->s_feature_ro_compat &
2070 		     ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
2071 		    (sb->s_feature_incompat &
2072 		     ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
2073 			printk(KERN_WARNING
2074 				"JBD2: Unrecognised features on journal\n");
2075 			return -EINVAL;
2076 		}
2077 	}
2078 
2079 	/*
2080 	 * Create a slab for this blocksize
2081 	 */
2082 	err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
2083 	if (err)
2084 		return err;
2085 
2086 	/* Let the recovery code check whether it needs to recover any
2087 	 * data from the journal. */
2088 	if (jbd2_journal_recover(journal))
2089 		goto recovery_error;
2090 
2091 	if (journal->j_failed_commit) {
2092 		printk(KERN_ERR "JBD2: journal transaction %u on %s "
2093 		       "is corrupt.\n", journal->j_failed_commit,
2094 		       journal->j_devname);
2095 		return -EFSCORRUPTED;
2096 	}
2097 	/*
2098 	 * clear JBD2_ABORT flag initialized in journal_init_common
2099 	 * here to update log tail information with the newest seq.
2100 	 */
2101 	journal->j_flags &= ~JBD2_ABORT;
2102 
2103 	/* OK, we've finished with the dynamic journal bits:
2104 	 * reinitialise the dynamic contents of the superblock in memory
2105 	 * and reset them on disk. */
2106 	if (journal_reset(journal))
2107 		goto recovery_error;
2108 
2109 	journal->j_flags |= JBD2_LOADED;
2110 	return 0;
2111 
2112 recovery_error:
2113 	printk(KERN_WARNING "JBD2: recovery failed\n");
2114 	return -EIO;
2115 }
2116 
2117 /**
2118  * jbd2_journal_destroy() - Release a journal_t structure.
2119  * @journal: Journal to act on.
2120  *
2121  * Release a journal_t structure once it is no longer in use by the
2122  * journaled object.
2123  * Return <0 if we couldn't clean up the journal.
2124  */
2125 int jbd2_journal_destroy(journal_t *journal)
2126 {
2127 	int err = 0;
2128 
2129 	/* Wait for the commit thread to wake up and die. */
2130 	journal_kill_thread(journal);
2131 
2132 	/* Force a final log commit */
2133 	if (journal->j_running_transaction)
2134 		jbd2_journal_commit_transaction(journal);
2135 
2136 	/* Force any old transactions to disk */
2137 
2138 	/* Totally anal locking here... */
2139 	spin_lock(&journal->j_list_lock);
2140 	while (journal->j_checkpoint_transactions != NULL) {
2141 		spin_unlock(&journal->j_list_lock);
2142 		mutex_lock_io(&journal->j_checkpoint_mutex);
2143 		err = jbd2_log_do_checkpoint(journal);
2144 		mutex_unlock(&journal->j_checkpoint_mutex);
2145 		/*
2146 		 * If checkpointing failed, just free the buffers to avoid
2147 		 * looping forever
2148 		 */
2149 		if (err) {
2150 			jbd2_journal_destroy_checkpoint(journal);
2151 			spin_lock(&journal->j_list_lock);
2152 			break;
2153 		}
2154 		spin_lock(&journal->j_list_lock);
2155 	}
2156 
2157 	J_ASSERT(journal->j_running_transaction == NULL);
2158 	J_ASSERT(journal->j_committing_transaction == NULL);
2159 	J_ASSERT(journal->j_checkpoint_transactions == NULL);
2160 	spin_unlock(&journal->j_list_lock);
2161 
2162 	/*
2163 	 * OK, all checkpoint transactions have been checked, now check the
2164 	 * write out io error flag and abort the journal if some buffer failed
2165 	 * to write back to the original location, otherwise the filesystem
2166 	 * may become inconsistent.
2167 	 */
2168 	if (!is_journal_aborted(journal) &&
2169 	    test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags))
2170 		jbd2_journal_abort(journal, -EIO);
2171 
2172 	if (journal->j_sb_buffer) {
2173 		if (!is_journal_aborted(journal)) {
2174 			mutex_lock_io(&journal->j_checkpoint_mutex);
2175 
2176 			write_lock(&journal->j_state_lock);
2177 			journal->j_tail_sequence =
2178 				++journal->j_transaction_sequence;
2179 			write_unlock(&journal->j_state_lock);
2180 
2181 			jbd2_mark_journal_empty(journal,
2182 					REQ_SYNC | REQ_PREFLUSH | REQ_FUA);
2183 			mutex_unlock(&journal->j_checkpoint_mutex);
2184 		} else
2185 			err = -EIO;
2186 		brelse(journal->j_sb_buffer);
2187 	}
2188 
2189 	if (journal->j_shrinker.flags & SHRINKER_REGISTERED) {
2190 		percpu_counter_destroy(&journal->j_checkpoint_jh_count);
2191 		unregister_shrinker(&journal->j_shrinker);
2192 	}
2193 	if (journal->j_proc_entry)
2194 		jbd2_stats_proc_exit(journal);
2195 	iput(journal->j_inode);
2196 	if (journal->j_revoke)
2197 		jbd2_journal_destroy_revoke(journal);
2198 	if (journal->j_chksum_driver)
2199 		crypto_free_shash(journal->j_chksum_driver);
2200 	kfree(journal->j_fc_wbuf);
2201 	kfree(journal->j_wbuf);
2202 	kfree(journal);
2203 
2204 	return err;
2205 }
2206 
2207 
2208 /**
2209  * jbd2_journal_check_used_features() - Check if features specified are used.
2210  * @journal: Journal to check.
2211  * @compat: bitmask of compatible features
2212  * @ro: bitmask of features that force read-only mount
2213  * @incompat: bitmask of incompatible features
2214  *
2215  * Check whether the journal uses all of a given set of
2216  * features.  Return true (non-zero) if it does.
2217  **/
2218 
2219 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
2220 				 unsigned long ro, unsigned long incompat)
2221 {
2222 	journal_superblock_t *sb;
2223 
2224 	if (!compat && !ro && !incompat)
2225 		return 1;
2226 	/* Load journal superblock if it is not loaded yet. */
2227 	if (journal->j_format_version == 0 &&
2228 	    journal_get_superblock(journal) != 0)
2229 		return 0;
2230 	if (journal->j_format_version == 1)
2231 		return 0;
2232 
2233 	sb = journal->j_superblock;
2234 
2235 	if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
2236 	    ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
2237 	    ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
2238 		return 1;
2239 
2240 	return 0;
2241 }
2242 
2243 /**
2244  * jbd2_journal_check_available_features() - Check feature set in journalling layer
2245  * @journal: Journal to check.
2246  * @compat: bitmask of compatible features
2247  * @ro: bitmask of features that force read-only mount
2248  * @incompat: bitmask of incompatible features
2249  *
2250  * Check whether the journaling code supports the use of
2251  * all of a given set of features on this journal.  Return true
2252  * (non-zero) if it can. */
2253 
2254 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
2255 				      unsigned long ro, unsigned long incompat)
2256 {
2257 	if (!compat && !ro && !incompat)
2258 		return 1;
2259 
2260 	/* We can support any known requested features iff the
2261 	 * superblock is in version 2.  Otherwise we fail to support any
2262 	 * extended sb features. */
2263 
2264 	if (journal->j_format_version != 2)
2265 		return 0;
2266 
2267 	if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
2268 	    (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
2269 	    (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
2270 		return 1;
2271 
2272 	return 0;
2273 }
2274 
2275 static int
2276 jbd2_journal_initialize_fast_commit(journal_t *journal)
2277 {
2278 	journal_superblock_t *sb = journal->j_superblock;
2279 	unsigned long long num_fc_blks;
2280 
2281 	num_fc_blks = jbd2_journal_get_num_fc_blks(sb);
2282 	if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS)
2283 		return -ENOSPC;
2284 
2285 	/* Are we called twice? */
2286 	WARN_ON(journal->j_fc_wbuf != NULL);
2287 	journal->j_fc_wbuf = kmalloc_array(num_fc_blks,
2288 				sizeof(struct buffer_head *), GFP_KERNEL);
2289 	if (!journal->j_fc_wbuf)
2290 		return -ENOMEM;
2291 
2292 	journal->j_fc_wbufsize = num_fc_blks;
2293 	journal->j_fc_last = journal->j_last;
2294 	journal->j_last = journal->j_fc_last - num_fc_blks;
2295 	journal->j_fc_first = journal->j_last + 1;
2296 	journal->j_fc_off = 0;
2297 	journal->j_free = journal->j_last - journal->j_first;
2298 	journal->j_max_transaction_buffers =
2299 		jbd2_journal_get_max_txn_bufs(journal);
2300 
2301 	return 0;
2302 }
2303 
2304 /**
2305  * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2306  * @journal: Journal to act on.
2307  * @compat: bitmask of compatible features
2308  * @ro: bitmask of features that force read-only mount
2309  * @incompat: bitmask of incompatible features
2310  *
2311  * Mark a given journal feature as present on the
2312  * superblock.  Returns true if the requested features could be set.
2313  *
2314  */
2315 
2316 int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
2317 			  unsigned long ro, unsigned long incompat)
2318 {
2319 #define INCOMPAT_FEATURE_ON(f) \
2320 		((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2321 #define COMPAT_FEATURE_ON(f) \
2322 		((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2323 	journal_superblock_t *sb;
2324 
2325 	if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
2326 		return 1;
2327 
2328 	if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
2329 		return 0;
2330 
2331 	/* If enabling v2 checksums, turn on v3 instead */
2332 	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
2333 		incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
2334 		incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
2335 	}
2336 
2337 	/* Asking for checksumming v3 and v1?  Only give them v3. */
2338 	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
2339 	    compat & JBD2_FEATURE_COMPAT_CHECKSUM)
2340 		compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
2341 
2342 	jbd2_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2343 		  compat, ro, incompat);
2344 
2345 	sb = journal->j_superblock;
2346 
2347 	if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) {
2348 		if (jbd2_journal_initialize_fast_commit(journal)) {
2349 			pr_err("JBD2: Cannot enable fast commits.\n");
2350 			return 0;
2351 		}
2352 	}
2353 
2354 	/* Load the checksum driver if necessary */
2355 	if ((journal->j_chksum_driver == NULL) &&
2356 	    INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2357 		journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
2358 		if (IS_ERR(journal->j_chksum_driver)) {
2359 			printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
2360 			journal->j_chksum_driver = NULL;
2361 			return 0;
2362 		}
2363 		/* Precompute checksum seed for all metadata */
2364 		journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2365 						   sizeof(sb->s_uuid));
2366 	}
2367 
2368 	lock_buffer(journal->j_sb_buffer);
2369 
2370 	/* If enabling v3 checksums, update superblock */
2371 	if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2372 		sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
2373 		sb->s_feature_compat &=
2374 			~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
2375 	}
2376 
2377 	/* If enabling v1 checksums, downgrade superblock */
2378 	if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
2379 		sb->s_feature_incompat &=
2380 			~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
2381 				     JBD2_FEATURE_INCOMPAT_CSUM_V3);
2382 
2383 	sb->s_feature_compat    |= cpu_to_be32(compat);
2384 	sb->s_feature_ro_compat |= cpu_to_be32(ro);
2385 	sb->s_feature_incompat  |= cpu_to_be32(incompat);
2386 	unlock_buffer(journal->j_sb_buffer);
2387 	journal->j_revoke_records_per_block =
2388 				journal_revoke_records_per_block(journal);
2389 
2390 	return 1;
2391 #undef COMPAT_FEATURE_ON
2392 #undef INCOMPAT_FEATURE_ON
2393 }
2394 
2395 /*
2396  * jbd2_journal_clear_features() - Clear a given journal feature in the
2397  * 				    superblock
2398  * @journal: Journal to act on.
2399  * @compat: bitmask of compatible features
2400  * @ro: bitmask of features that force read-only mount
2401  * @incompat: bitmask of incompatible features
2402  *
2403  * Clear a given journal feature as present on the
2404  * superblock.
2405  */
2406 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
2407 				unsigned long ro, unsigned long incompat)
2408 {
2409 	journal_superblock_t *sb;
2410 
2411 	jbd2_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2412 		  compat, ro, incompat);
2413 
2414 	sb = journal->j_superblock;
2415 
2416 	sb->s_feature_compat    &= ~cpu_to_be32(compat);
2417 	sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
2418 	sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
2419 	journal->j_revoke_records_per_block =
2420 				journal_revoke_records_per_block(journal);
2421 }
2422 EXPORT_SYMBOL(jbd2_journal_clear_features);
2423 
2424 /**
2425  * jbd2_journal_flush() - Flush journal
2426  * @journal: Journal to act on.
2427  * @flags: optional operation on the journal blocks after the flush (see below)
2428  *
2429  * Flush all data for a given journal to disk and empty the journal.
2430  * Filesystems can use this when remounting readonly to ensure that
2431  * recovery does not need to happen on remount. Optionally, a discard or zeroout
2432  * can be issued on the journal blocks after flushing.
2433  *
2434  * flags:
2435  *	JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks
2436  *	JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks
2437  */
2438 int jbd2_journal_flush(journal_t *journal, unsigned int flags)
2439 {
2440 	int err = 0;
2441 	transaction_t *transaction = NULL;
2442 
2443 	write_lock(&journal->j_state_lock);
2444 
2445 	/* Force everything buffered to the log... */
2446 	if (journal->j_running_transaction) {
2447 		transaction = journal->j_running_transaction;
2448 		__jbd2_log_start_commit(journal, transaction->t_tid);
2449 	} else if (journal->j_committing_transaction)
2450 		transaction = journal->j_committing_transaction;
2451 
2452 	/* Wait for the log commit to complete... */
2453 	if (transaction) {
2454 		tid_t tid = transaction->t_tid;
2455 
2456 		write_unlock(&journal->j_state_lock);
2457 		jbd2_log_wait_commit(journal, tid);
2458 	} else {
2459 		write_unlock(&journal->j_state_lock);
2460 	}
2461 
2462 	/* ...and flush everything in the log out to disk. */
2463 	spin_lock(&journal->j_list_lock);
2464 	while (!err && journal->j_checkpoint_transactions != NULL) {
2465 		spin_unlock(&journal->j_list_lock);
2466 		mutex_lock_io(&journal->j_checkpoint_mutex);
2467 		err = jbd2_log_do_checkpoint(journal);
2468 		mutex_unlock(&journal->j_checkpoint_mutex);
2469 		spin_lock(&journal->j_list_lock);
2470 	}
2471 	spin_unlock(&journal->j_list_lock);
2472 
2473 	if (is_journal_aborted(journal))
2474 		return -EIO;
2475 
2476 	mutex_lock_io(&journal->j_checkpoint_mutex);
2477 	if (!err) {
2478 		err = jbd2_cleanup_journal_tail(journal);
2479 		if (err < 0) {
2480 			mutex_unlock(&journal->j_checkpoint_mutex);
2481 			goto out;
2482 		}
2483 		err = 0;
2484 	}
2485 
2486 	/* Finally, mark the journal as really needing no recovery.
2487 	 * This sets s_start==0 in the underlying superblock, which is
2488 	 * the magic code for a fully-recovered superblock.  Any future
2489 	 * commits of data to the journal will restore the current
2490 	 * s_start value. */
2491 	jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2492 
2493 	if (flags)
2494 		err = __jbd2_journal_erase(journal, flags);
2495 
2496 	mutex_unlock(&journal->j_checkpoint_mutex);
2497 	write_lock(&journal->j_state_lock);
2498 	J_ASSERT(!journal->j_running_transaction);
2499 	J_ASSERT(!journal->j_committing_transaction);
2500 	J_ASSERT(!journal->j_checkpoint_transactions);
2501 	J_ASSERT(journal->j_head == journal->j_tail);
2502 	J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2503 	write_unlock(&journal->j_state_lock);
2504 out:
2505 	return err;
2506 }
2507 
2508 /**
2509  * jbd2_journal_wipe() - Wipe journal contents
2510  * @journal: Journal to act on.
2511  * @write: flag (see below)
2512  *
2513  * Wipe out all of the contents of a journal, safely.  This will produce
2514  * a warning if the journal contains any valid recovery information.
2515  * Must be called between journal_init_*() and jbd2_journal_load().
2516  *
2517  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2518  * we merely suppress recovery.
2519  */
2520 
2521 int jbd2_journal_wipe(journal_t *journal, int write)
2522 {
2523 	int err = 0;
2524 
2525 	J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2526 
2527 	err = load_superblock(journal);
2528 	if (err)
2529 		return err;
2530 
2531 	if (!journal->j_tail)
2532 		goto no_recovery;
2533 
2534 	printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2535 		write ? "Clearing" : "Ignoring");
2536 
2537 	err = jbd2_journal_skip_recovery(journal);
2538 	if (write) {
2539 		/* Lock to make assertions happy... */
2540 		mutex_lock_io(&journal->j_checkpoint_mutex);
2541 		jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA);
2542 		mutex_unlock(&journal->j_checkpoint_mutex);
2543 	}
2544 
2545  no_recovery:
2546 	return err;
2547 }
2548 
2549 /**
2550  * jbd2_journal_abort () - Shutdown the journal immediately.
2551  * @journal: the journal to shutdown.
2552  * @errno:   an error number to record in the journal indicating
2553  *           the reason for the shutdown.
2554  *
2555  * Perform a complete, immediate shutdown of the ENTIRE
2556  * journal (not of a single transaction).  This operation cannot be
2557  * undone without closing and reopening the journal.
2558  *
2559  * The jbd2_journal_abort function is intended to support higher level error
2560  * recovery mechanisms such as the ext2/ext3 remount-readonly error
2561  * mode.
2562  *
2563  * Journal abort has very specific semantics.  Any existing dirty,
2564  * unjournaled buffers in the main filesystem will still be written to
2565  * disk by bdflush, but the journaling mechanism will be suspended
2566  * immediately and no further transaction commits will be honoured.
2567  *
2568  * Any dirty, journaled buffers will be written back to disk without
2569  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2570  * filesystem, but we _do_ attempt to leave as much data as possible
2571  * behind for fsck to use for cleanup.
2572  *
2573  * Any attempt to get a new transaction handle on a journal which is in
2574  * ABORT state will just result in an -EROFS error return.  A
2575  * jbd2_journal_stop on an existing handle will return -EIO if we have
2576  * entered abort state during the update.
2577  *
2578  * Recursive transactions are not disturbed by journal abort until the
2579  * final jbd2_journal_stop, which will receive the -EIO error.
2580  *
2581  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2582  * which will be recorded (if possible) in the journal superblock.  This
2583  * allows a client to record failure conditions in the middle of a
2584  * transaction without having to complete the transaction to record the
2585  * failure to disk.  ext3_error, for example, now uses this
2586  * functionality.
2587  *
2588  */
2589 
2590 void jbd2_journal_abort(journal_t *journal, int errno)
2591 {
2592 	transaction_t *transaction;
2593 
2594 	/*
2595 	 * Lock the aborting procedure until everything is done, this avoid
2596 	 * races between filesystem's error handling flow (e.g. ext4_abort()),
2597 	 * ensure panic after the error info is written into journal's
2598 	 * superblock.
2599 	 */
2600 	mutex_lock(&journal->j_abort_mutex);
2601 	/*
2602 	 * ESHUTDOWN always takes precedence because a file system check
2603 	 * caused by any other journal abort error is not required after
2604 	 * a shutdown triggered.
2605 	 */
2606 	write_lock(&journal->j_state_lock);
2607 	if (journal->j_flags & JBD2_ABORT) {
2608 		int old_errno = journal->j_errno;
2609 
2610 		write_unlock(&journal->j_state_lock);
2611 		if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2612 			journal->j_errno = errno;
2613 			jbd2_journal_update_sb_errno(journal);
2614 		}
2615 		mutex_unlock(&journal->j_abort_mutex);
2616 		return;
2617 	}
2618 
2619 	/*
2620 	 * Mark the abort as occurred and start current running transaction
2621 	 * to release all journaled buffer.
2622 	 */
2623 	pr_err("Aborting journal on device %s.\n", journal->j_devname);
2624 
2625 	journal->j_flags |= JBD2_ABORT;
2626 	journal->j_errno = errno;
2627 	transaction = journal->j_running_transaction;
2628 	if (transaction)
2629 		__jbd2_log_start_commit(journal, transaction->t_tid);
2630 	write_unlock(&journal->j_state_lock);
2631 
2632 	/*
2633 	 * Record errno to the journal super block, so that fsck and jbd2
2634 	 * layer could realise that a filesystem check is needed.
2635 	 */
2636 	jbd2_journal_update_sb_errno(journal);
2637 	mutex_unlock(&journal->j_abort_mutex);
2638 }
2639 
2640 /**
2641  * jbd2_journal_errno() - returns the journal's error state.
2642  * @journal: journal to examine.
2643  *
2644  * This is the errno number set with jbd2_journal_abort(), the last
2645  * time the journal was mounted - if the journal was stopped
2646  * without calling abort this will be 0.
2647  *
2648  * If the journal has been aborted on this mount time -EROFS will
2649  * be returned.
2650  */
2651 int jbd2_journal_errno(journal_t *journal)
2652 {
2653 	int err;
2654 
2655 	read_lock(&journal->j_state_lock);
2656 	if (journal->j_flags & JBD2_ABORT)
2657 		err = -EROFS;
2658 	else
2659 		err = journal->j_errno;
2660 	read_unlock(&journal->j_state_lock);
2661 	return err;
2662 }
2663 
2664 /**
2665  * jbd2_journal_clear_err() - clears the journal's error state
2666  * @journal: journal to act on.
2667  *
2668  * An error must be cleared or acked to take a FS out of readonly
2669  * mode.
2670  */
2671 int jbd2_journal_clear_err(journal_t *journal)
2672 {
2673 	int err = 0;
2674 
2675 	write_lock(&journal->j_state_lock);
2676 	if (journal->j_flags & JBD2_ABORT)
2677 		err = -EROFS;
2678 	else
2679 		journal->j_errno = 0;
2680 	write_unlock(&journal->j_state_lock);
2681 	return err;
2682 }
2683 
2684 /**
2685  * jbd2_journal_ack_err() - Ack journal err.
2686  * @journal: journal to act on.
2687  *
2688  * An error must be cleared or acked to take a FS out of readonly
2689  * mode.
2690  */
2691 void jbd2_journal_ack_err(journal_t *journal)
2692 {
2693 	write_lock(&journal->j_state_lock);
2694 	if (journal->j_errno)
2695 		journal->j_flags |= JBD2_ACK_ERR;
2696 	write_unlock(&journal->j_state_lock);
2697 }
2698 
2699 int jbd2_journal_blocks_per_page(struct inode *inode)
2700 {
2701 	return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2702 }
2703 
2704 /*
2705  * helper functions to deal with 32 or 64bit block numbers.
2706  */
2707 size_t journal_tag_bytes(journal_t *journal)
2708 {
2709 	size_t sz;
2710 
2711 	if (jbd2_has_feature_csum3(journal))
2712 		return sizeof(journal_block_tag3_t);
2713 
2714 	sz = sizeof(journal_block_tag_t);
2715 
2716 	if (jbd2_has_feature_csum2(journal))
2717 		sz += sizeof(__u16);
2718 
2719 	if (jbd2_has_feature_64bit(journal))
2720 		return sz;
2721 	else
2722 		return sz - sizeof(__u32);
2723 }
2724 
2725 /*
2726  * JBD memory management
2727  *
2728  * These functions are used to allocate block-sized chunks of memory
2729  * used for making copies of buffer_head data.  Very often it will be
2730  * page-sized chunks of data, but sometimes it will be in
2731  * sub-page-size chunks.  (For example, 16k pages on Power systems
2732  * with a 4k block file system.)  For blocks smaller than a page, we
2733  * use a SLAB allocator.  There are slab caches for each block size,
2734  * which are allocated at mount time, if necessary, and we only free
2735  * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2736  * this reason we don't need to a mutex to protect access to
2737  * jbd2_slab[] allocating or releasing memory; only in
2738  * jbd2_journal_create_slab().
2739  */
2740 #define JBD2_MAX_SLABS 8
2741 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2742 
2743 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2744 	"jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2745 	"jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2746 };
2747 
2748 
2749 static void jbd2_journal_destroy_slabs(void)
2750 {
2751 	int i;
2752 
2753 	for (i = 0; i < JBD2_MAX_SLABS; i++) {
2754 		kmem_cache_destroy(jbd2_slab[i]);
2755 		jbd2_slab[i] = NULL;
2756 	}
2757 }
2758 
2759 static int jbd2_journal_create_slab(size_t size)
2760 {
2761 	static DEFINE_MUTEX(jbd2_slab_create_mutex);
2762 	int i = order_base_2(size) - 10;
2763 	size_t slab_size;
2764 
2765 	if (size == PAGE_SIZE)
2766 		return 0;
2767 
2768 	if (i >= JBD2_MAX_SLABS)
2769 		return -EINVAL;
2770 
2771 	if (unlikely(i < 0))
2772 		i = 0;
2773 	mutex_lock(&jbd2_slab_create_mutex);
2774 	if (jbd2_slab[i]) {
2775 		mutex_unlock(&jbd2_slab_create_mutex);
2776 		return 0;	/* Already created */
2777 	}
2778 
2779 	slab_size = 1 << (i+10);
2780 	jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2781 					 slab_size, 0, NULL);
2782 	mutex_unlock(&jbd2_slab_create_mutex);
2783 	if (!jbd2_slab[i]) {
2784 		printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2785 		return -ENOMEM;
2786 	}
2787 	return 0;
2788 }
2789 
2790 static struct kmem_cache *get_slab(size_t size)
2791 {
2792 	int i = order_base_2(size) - 10;
2793 
2794 	BUG_ON(i >= JBD2_MAX_SLABS);
2795 	if (unlikely(i < 0))
2796 		i = 0;
2797 	BUG_ON(jbd2_slab[i] == NULL);
2798 	return jbd2_slab[i];
2799 }
2800 
2801 void *jbd2_alloc(size_t size, gfp_t flags)
2802 {
2803 	void *ptr;
2804 
2805 	BUG_ON(size & (size-1)); /* Must be a power of 2 */
2806 
2807 	if (size < PAGE_SIZE)
2808 		ptr = kmem_cache_alloc(get_slab(size), flags);
2809 	else
2810 		ptr = (void *)__get_free_pages(flags, get_order(size));
2811 
2812 	/* Check alignment; SLUB has gotten this wrong in the past,
2813 	 * and this can lead to user data corruption! */
2814 	BUG_ON(((unsigned long) ptr) & (size-1));
2815 
2816 	return ptr;
2817 }
2818 
2819 void jbd2_free(void *ptr, size_t size)
2820 {
2821 	if (size < PAGE_SIZE)
2822 		kmem_cache_free(get_slab(size), ptr);
2823 	else
2824 		free_pages((unsigned long)ptr, get_order(size));
2825 };
2826 
2827 /*
2828  * Journal_head storage management
2829  */
2830 static struct kmem_cache *jbd2_journal_head_cache;
2831 #ifdef CONFIG_JBD2_DEBUG
2832 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2833 #endif
2834 
2835 static int __init jbd2_journal_init_journal_head_cache(void)
2836 {
2837 	J_ASSERT(!jbd2_journal_head_cache);
2838 	jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2839 				sizeof(struct journal_head),
2840 				0,		/* offset */
2841 				SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2842 				NULL);		/* ctor */
2843 	if (!jbd2_journal_head_cache) {
2844 		printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2845 		return -ENOMEM;
2846 	}
2847 	return 0;
2848 }
2849 
2850 static void jbd2_journal_destroy_journal_head_cache(void)
2851 {
2852 	kmem_cache_destroy(jbd2_journal_head_cache);
2853 	jbd2_journal_head_cache = NULL;
2854 }
2855 
2856 /*
2857  * journal_head splicing and dicing
2858  */
2859 static struct journal_head *journal_alloc_journal_head(void)
2860 {
2861 	struct journal_head *ret;
2862 
2863 #ifdef CONFIG_JBD2_DEBUG
2864 	atomic_inc(&nr_journal_heads);
2865 #endif
2866 	ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2867 	if (!ret) {
2868 		jbd2_debug(1, "out of memory for journal_head\n");
2869 		pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2870 		ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2871 				GFP_NOFS | __GFP_NOFAIL);
2872 	}
2873 	if (ret)
2874 		spin_lock_init(&ret->b_state_lock);
2875 	return ret;
2876 }
2877 
2878 static void journal_free_journal_head(struct journal_head *jh)
2879 {
2880 #ifdef CONFIG_JBD2_DEBUG
2881 	atomic_dec(&nr_journal_heads);
2882 	memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2883 #endif
2884 	kmem_cache_free(jbd2_journal_head_cache, jh);
2885 }
2886 
2887 /*
2888  * A journal_head is attached to a buffer_head whenever JBD has an
2889  * interest in the buffer.
2890  *
2891  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2892  * is set.  This bit is tested in core kernel code where we need to take
2893  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2894  * there.
2895  *
2896  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2897  *
2898  * When a buffer has its BH_JBD bit set it is immune from being released by
2899  * core kernel code, mainly via ->b_count.
2900  *
2901  * A journal_head is detached from its buffer_head when the journal_head's
2902  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2903  * transaction (b_cp_transaction) hold their references to b_jcount.
2904  *
2905  * Various places in the kernel want to attach a journal_head to a buffer_head
2906  * _before_ attaching the journal_head to a transaction.  To protect the
2907  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2908  * journal_head's b_jcount refcount by one.  The caller must call
2909  * jbd2_journal_put_journal_head() to undo this.
2910  *
2911  * So the typical usage would be:
2912  *
2913  *	(Attach a journal_head if needed.  Increments b_jcount)
2914  *	struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2915  *	...
2916  *      (Get another reference for transaction)
2917  *	jbd2_journal_grab_journal_head(bh);
2918  *	jh->b_transaction = xxx;
2919  *	(Put original reference)
2920  *	jbd2_journal_put_journal_head(jh);
2921  */
2922 
2923 /*
2924  * Give a buffer_head a journal_head.
2925  *
2926  * May sleep.
2927  */
2928 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2929 {
2930 	struct journal_head *jh;
2931 	struct journal_head *new_jh = NULL;
2932 
2933 repeat:
2934 	if (!buffer_jbd(bh))
2935 		new_jh = journal_alloc_journal_head();
2936 
2937 	jbd_lock_bh_journal_head(bh);
2938 	if (buffer_jbd(bh)) {
2939 		jh = bh2jh(bh);
2940 	} else {
2941 		J_ASSERT_BH(bh,
2942 			(atomic_read(&bh->b_count) > 0) ||
2943 			(bh->b_folio && bh->b_folio->mapping));
2944 
2945 		if (!new_jh) {
2946 			jbd_unlock_bh_journal_head(bh);
2947 			goto repeat;
2948 		}
2949 
2950 		jh = new_jh;
2951 		new_jh = NULL;		/* We consumed it */
2952 		set_buffer_jbd(bh);
2953 		bh->b_private = jh;
2954 		jh->b_bh = bh;
2955 		get_bh(bh);
2956 		BUFFER_TRACE(bh, "added journal_head");
2957 	}
2958 	jh->b_jcount++;
2959 	jbd_unlock_bh_journal_head(bh);
2960 	if (new_jh)
2961 		journal_free_journal_head(new_jh);
2962 	return bh->b_private;
2963 }
2964 
2965 /*
2966  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2967  * having a journal_head, return NULL
2968  */
2969 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2970 {
2971 	struct journal_head *jh = NULL;
2972 
2973 	jbd_lock_bh_journal_head(bh);
2974 	if (buffer_jbd(bh)) {
2975 		jh = bh2jh(bh);
2976 		jh->b_jcount++;
2977 	}
2978 	jbd_unlock_bh_journal_head(bh);
2979 	return jh;
2980 }
2981 EXPORT_SYMBOL(jbd2_journal_grab_journal_head);
2982 
2983 static void __journal_remove_journal_head(struct buffer_head *bh)
2984 {
2985 	struct journal_head *jh = bh2jh(bh);
2986 
2987 	J_ASSERT_JH(jh, jh->b_transaction == NULL);
2988 	J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2989 	J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2990 	J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2991 	J_ASSERT_BH(bh, buffer_jbd(bh));
2992 	J_ASSERT_BH(bh, jh2bh(jh) == bh);
2993 	BUFFER_TRACE(bh, "remove journal_head");
2994 
2995 	/* Unlink before dropping the lock */
2996 	bh->b_private = NULL;
2997 	jh->b_bh = NULL;	/* debug, really */
2998 	clear_buffer_jbd(bh);
2999 }
3000 
3001 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
3002 {
3003 	if (jh->b_frozen_data) {
3004 		printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
3005 		jbd2_free(jh->b_frozen_data, b_size);
3006 	}
3007 	if (jh->b_committed_data) {
3008 		printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
3009 		jbd2_free(jh->b_committed_data, b_size);
3010 	}
3011 	journal_free_journal_head(jh);
3012 }
3013 
3014 /*
3015  * Drop a reference on the passed journal_head.  If it fell to zero then
3016  * release the journal_head from the buffer_head.
3017  */
3018 void jbd2_journal_put_journal_head(struct journal_head *jh)
3019 {
3020 	struct buffer_head *bh = jh2bh(jh);
3021 
3022 	jbd_lock_bh_journal_head(bh);
3023 	J_ASSERT_JH(jh, jh->b_jcount > 0);
3024 	--jh->b_jcount;
3025 	if (!jh->b_jcount) {
3026 		__journal_remove_journal_head(bh);
3027 		jbd_unlock_bh_journal_head(bh);
3028 		journal_release_journal_head(jh, bh->b_size);
3029 		__brelse(bh);
3030 	} else {
3031 		jbd_unlock_bh_journal_head(bh);
3032 	}
3033 }
3034 EXPORT_SYMBOL(jbd2_journal_put_journal_head);
3035 
3036 /*
3037  * Initialize jbd inode head
3038  */
3039 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
3040 {
3041 	jinode->i_transaction = NULL;
3042 	jinode->i_next_transaction = NULL;
3043 	jinode->i_vfs_inode = inode;
3044 	jinode->i_flags = 0;
3045 	jinode->i_dirty_start = 0;
3046 	jinode->i_dirty_end = 0;
3047 	INIT_LIST_HEAD(&jinode->i_list);
3048 }
3049 
3050 /*
3051  * Function to be called before we start removing inode from memory (i.e.,
3052  * clear_inode() is a fine place to be called from). It removes inode from
3053  * transaction's lists.
3054  */
3055 void jbd2_journal_release_jbd_inode(journal_t *journal,
3056 				    struct jbd2_inode *jinode)
3057 {
3058 	if (!journal)
3059 		return;
3060 restart:
3061 	spin_lock(&journal->j_list_lock);
3062 	/* Is commit writing out inode - we have to wait */
3063 	if (jinode->i_flags & JI_COMMIT_RUNNING) {
3064 		wait_queue_head_t *wq;
3065 		DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
3066 		wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
3067 		prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
3068 		spin_unlock(&journal->j_list_lock);
3069 		schedule();
3070 		finish_wait(wq, &wait.wq_entry);
3071 		goto restart;
3072 	}
3073 
3074 	if (jinode->i_transaction) {
3075 		list_del(&jinode->i_list);
3076 		jinode->i_transaction = NULL;
3077 	}
3078 	spin_unlock(&journal->j_list_lock);
3079 }
3080 
3081 
3082 #ifdef CONFIG_PROC_FS
3083 
3084 #define JBD2_STATS_PROC_NAME "fs/jbd2"
3085 
3086 static void __init jbd2_create_jbd_stats_proc_entry(void)
3087 {
3088 	proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
3089 }
3090 
3091 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
3092 {
3093 	if (proc_jbd2_stats)
3094 		remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
3095 }
3096 
3097 #else
3098 
3099 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
3100 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
3101 
3102 #endif
3103 
3104 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
3105 
3106 static int __init jbd2_journal_init_inode_cache(void)
3107 {
3108 	J_ASSERT(!jbd2_inode_cache);
3109 	jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
3110 	if (!jbd2_inode_cache) {
3111 		pr_emerg("JBD2: failed to create inode cache\n");
3112 		return -ENOMEM;
3113 	}
3114 	return 0;
3115 }
3116 
3117 static int __init jbd2_journal_init_handle_cache(void)
3118 {
3119 	J_ASSERT(!jbd2_handle_cache);
3120 	jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
3121 	if (!jbd2_handle_cache) {
3122 		printk(KERN_EMERG "JBD2: failed to create handle cache\n");
3123 		return -ENOMEM;
3124 	}
3125 	return 0;
3126 }
3127 
3128 static void jbd2_journal_destroy_inode_cache(void)
3129 {
3130 	kmem_cache_destroy(jbd2_inode_cache);
3131 	jbd2_inode_cache = NULL;
3132 }
3133 
3134 static void jbd2_journal_destroy_handle_cache(void)
3135 {
3136 	kmem_cache_destroy(jbd2_handle_cache);
3137 	jbd2_handle_cache = NULL;
3138 }
3139 
3140 /*
3141  * Module startup and shutdown
3142  */
3143 
3144 static int __init journal_init_caches(void)
3145 {
3146 	int ret;
3147 
3148 	ret = jbd2_journal_init_revoke_record_cache();
3149 	if (ret == 0)
3150 		ret = jbd2_journal_init_revoke_table_cache();
3151 	if (ret == 0)
3152 		ret = jbd2_journal_init_journal_head_cache();
3153 	if (ret == 0)
3154 		ret = jbd2_journal_init_handle_cache();
3155 	if (ret == 0)
3156 		ret = jbd2_journal_init_inode_cache();
3157 	if (ret == 0)
3158 		ret = jbd2_journal_init_transaction_cache();
3159 	return ret;
3160 }
3161 
3162 static void jbd2_journal_destroy_caches(void)
3163 {
3164 	jbd2_journal_destroy_revoke_record_cache();
3165 	jbd2_journal_destroy_revoke_table_cache();
3166 	jbd2_journal_destroy_journal_head_cache();
3167 	jbd2_journal_destroy_handle_cache();
3168 	jbd2_journal_destroy_inode_cache();
3169 	jbd2_journal_destroy_transaction_cache();
3170 	jbd2_journal_destroy_slabs();
3171 }
3172 
3173 static int __init journal_init(void)
3174 {
3175 	int ret;
3176 
3177 	BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
3178 
3179 	ret = journal_init_caches();
3180 	if (ret == 0) {
3181 		jbd2_create_jbd_stats_proc_entry();
3182 	} else {
3183 		jbd2_journal_destroy_caches();
3184 	}
3185 	return ret;
3186 }
3187 
3188 static void __exit journal_exit(void)
3189 {
3190 #ifdef CONFIG_JBD2_DEBUG
3191 	int n = atomic_read(&nr_journal_heads);
3192 	if (n)
3193 		printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
3194 #endif
3195 	jbd2_remove_jbd_stats_proc_entry();
3196 	jbd2_journal_destroy_caches();
3197 }
3198 
3199 MODULE_LICENSE("GPL");
3200 module_init(journal_init);
3201 module_exit(journal_exit);
3202 
3203