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
__jbd2_debug(int level,const char * file,const char * func,unsigned int line,const char * fmt,...)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 */
jbd2_superblock_csum(journal_t * j,journal_superblock_t * sb)118 static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb)
119 {
120 	__u32 csum;
121 	__be32 old_csum;
122 
123 	old_csum = sb->s_checksum;
124 	sb->s_checksum = 0;
125 	csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t));
126 	sb->s_checksum = old_csum;
127 
128 	return cpu_to_be32(csum);
129 }
130 
131 /*
132  * Helper function used to manage commit timeouts
133  */
134 
commit_timeout(struct timer_list * t)135 static void commit_timeout(struct timer_list *t)
136 {
137 	journal_t *journal = from_timer(journal, t, j_commit_timer);
138 
139 	wake_up_process(journal->j_task);
140 }
141 
142 /*
143  * kjournald2: The main thread function used to manage a logging device
144  * journal.
145  *
146  * This kernel thread is responsible for two things:
147  *
148  * 1) COMMIT:  Every so often we need to commit the current state of the
149  *    filesystem to disk.  The journal thread is responsible for writing
150  *    all of the metadata buffers to disk. If a fast commit is ongoing
151  *    journal thread waits until it's done and then continues from
152  *    there on.
153  *
154  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
155  *    of the data in that part of the log has been rewritten elsewhere on
156  *    the disk.  Flushing these old buffers to reclaim space in the log is
157  *    known as checkpointing, and this thread is responsible for that job.
158  */
159 
kjournald2(void * arg)160 static int kjournald2(void *arg)
161 {
162 	journal_t *journal = arg;
163 	transaction_t *transaction;
164 
165 	/*
166 	 * Set up an interval timer which can be used to trigger a commit wakeup
167 	 * after the commit interval expires
168 	 */
169 	timer_setup(&journal->j_commit_timer, commit_timeout, 0);
170 
171 	set_freezable();
172 
173 	/* Record that the journal thread is running */
174 	journal->j_task = current;
175 	wake_up(&journal->j_wait_done_commit);
176 
177 	/*
178 	 * Make sure that no allocations from this kernel thread will ever
179 	 * recurse to the fs layer because we are responsible for the
180 	 * transaction commit and any fs involvement might get stuck waiting for
181 	 * the trasn. commit.
182 	 */
183 	memalloc_nofs_save();
184 
185 	/*
186 	 * And now, wait forever for commit wakeup events.
187 	 */
188 	write_lock(&journal->j_state_lock);
189 
190 loop:
191 	if (journal->j_flags & JBD2_UNMOUNT)
192 		goto end_loop;
193 
194 	jbd2_debug(1, "commit_sequence=%u, commit_request=%u\n",
195 		journal->j_commit_sequence, journal->j_commit_request);
196 
197 	if (journal->j_commit_sequence != journal->j_commit_request) {
198 		jbd2_debug(1, "OK, requests differ\n");
199 		write_unlock(&journal->j_state_lock);
200 		del_timer_sync(&journal->j_commit_timer);
201 		jbd2_journal_commit_transaction(journal);
202 		write_lock(&journal->j_state_lock);
203 		goto loop;
204 	}
205 
206 	wake_up(&journal->j_wait_done_commit);
207 	if (freezing(current)) {
208 		/*
209 		 * The simpler the better. Flushing journal isn't a
210 		 * good idea, because that depends on threads that may
211 		 * be already stopped.
212 		 */
213 		jbd2_debug(1, "Now suspending kjournald2\n");
214 		write_unlock(&journal->j_state_lock);
215 		try_to_freeze();
216 		write_lock(&journal->j_state_lock);
217 	} else {
218 		/*
219 		 * We assume on resume that commits are already there,
220 		 * so we don't sleep
221 		 */
222 		DEFINE_WAIT(wait);
223 		int should_sleep = 1;
224 
225 		prepare_to_wait(&journal->j_wait_commit, &wait,
226 				TASK_INTERRUPTIBLE);
227 		if (journal->j_commit_sequence != journal->j_commit_request)
228 			should_sleep = 0;
229 		transaction = journal->j_running_transaction;
230 		if (transaction && time_after_eq(jiffies,
231 						transaction->t_expires))
232 			should_sleep = 0;
233 		if (journal->j_flags & JBD2_UNMOUNT)
234 			should_sleep = 0;
235 		if (should_sleep) {
236 			write_unlock(&journal->j_state_lock);
237 			schedule();
238 			write_lock(&journal->j_state_lock);
239 		}
240 		finish_wait(&journal->j_wait_commit, &wait);
241 	}
242 
243 	jbd2_debug(1, "kjournald2 wakes\n");
244 
245 	/*
246 	 * Were we woken up by a commit wakeup event?
247 	 */
248 	transaction = journal->j_running_transaction;
249 	if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
250 		journal->j_commit_request = transaction->t_tid;
251 		jbd2_debug(1, "woke because of timeout\n");
252 	}
253 	goto loop;
254 
255 end_loop:
256 	del_timer_sync(&journal->j_commit_timer);
257 	journal->j_task = NULL;
258 	wake_up(&journal->j_wait_done_commit);
259 	jbd2_debug(1, "Journal thread exiting.\n");
260 	write_unlock(&journal->j_state_lock);
261 	return 0;
262 }
263 
jbd2_journal_start_thread(journal_t * journal)264 static int jbd2_journal_start_thread(journal_t *journal)
265 {
266 	struct task_struct *t;
267 
268 	t = kthread_run(kjournald2, journal, "jbd2/%s",
269 			journal->j_devname);
270 	if (IS_ERR(t))
271 		return PTR_ERR(t);
272 
273 	wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
274 	return 0;
275 }
276 
journal_kill_thread(journal_t * journal)277 static void journal_kill_thread(journal_t *journal)
278 {
279 	write_lock(&journal->j_state_lock);
280 	journal->j_flags |= JBD2_UNMOUNT;
281 
282 	while (journal->j_task) {
283 		write_unlock(&journal->j_state_lock);
284 		wake_up(&journal->j_wait_commit);
285 		wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
286 		write_lock(&journal->j_state_lock);
287 	}
288 	write_unlock(&journal->j_state_lock);
289 }
290 
291 /*
292  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
293  *
294  * Writes a metadata buffer to a given disk block.  The actual IO is not
295  * performed but a new buffer_head is constructed which labels the data
296  * to be written with the correct destination disk block.
297  *
298  * Any magic-number escaping which needs to be done will cause a
299  * copy-out here.  If the buffer happens to start with the
300  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
301  * magic number is only written to the log for descripter blocks.  In
302  * this case, we copy the data and replace the first word with 0, and we
303  * return a result code which indicates that this buffer needs to be
304  * marked as an escaped buffer in the corresponding log descriptor
305  * block.  The missing word can then be restored when the block is read
306  * during recovery.
307  *
308  * If the source buffer has already been modified by a new transaction
309  * since we took the last commit snapshot, we use the frozen copy of
310  * that data for IO. If we end up using the existing buffer_head's data
311  * for the write, then we have to make sure nobody modifies it while the
312  * IO is in progress. do_get_write_access() handles this.
313  *
314  * The function returns a pointer to the buffer_head to be used for IO.
315  *
316  *
317  * Return value:
318  *  <0: Error
319  * >=0: Finished OK
320  *
321  * On success:
322  * Bit 0 set == escape performed on the data
323  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
324  */
325 
jbd2_journal_write_metadata_buffer(transaction_t * transaction,struct journal_head * jh_in,struct buffer_head ** bh_out,sector_t blocknr)326 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
327 				  struct journal_head  *jh_in,
328 				  struct buffer_head **bh_out,
329 				  sector_t blocknr)
330 {
331 	int need_copy_out = 0;
332 	int done_copy_out = 0;
333 	int do_escape = 0;
334 	char *mapped_data;
335 	struct buffer_head *new_bh;
336 	struct folio *new_folio;
337 	unsigned int new_offset;
338 	struct buffer_head *bh_in = jh2bh(jh_in);
339 	journal_t *journal = transaction->t_journal;
340 
341 	/*
342 	 * The buffer really shouldn't be locked: only the current committing
343 	 * transaction is allowed to write it, so nobody else is allowed
344 	 * to do any IO.
345 	 *
346 	 * akpm: except if we're journalling data, and write() output is
347 	 * also part of a shared mapping, and another thread has
348 	 * decided to launch a writepage() against this buffer.
349 	 */
350 	J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
351 
352 	new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
353 
354 	/* keep subsequent assertions sane */
355 	atomic_set(&new_bh->b_count, 1);
356 
357 	spin_lock(&jh_in->b_state_lock);
358 repeat:
359 	/*
360 	 * If a new transaction has already done a buffer copy-out, then
361 	 * we use that version of the data for the commit.
362 	 */
363 	if (jh_in->b_frozen_data) {
364 		done_copy_out = 1;
365 		new_folio = virt_to_folio(jh_in->b_frozen_data);
366 		new_offset = offset_in_folio(new_folio, jh_in->b_frozen_data);
367 	} else {
368 		new_folio = jh2bh(jh_in)->b_folio;
369 		new_offset = offset_in_folio(new_folio, jh2bh(jh_in)->b_data);
370 	}
371 
372 	mapped_data = kmap_local_folio(new_folio, new_offset);
373 	/*
374 	 * Fire data frozen trigger if data already wasn't frozen.  Do this
375 	 * before checking for escaping, as the trigger may modify the magic
376 	 * offset.  If a copy-out happens afterwards, it will have the correct
377 	 * data in the buffer.
378 	 */
379 	if (!done_copy_out)
380 		jbd2_buffer_frozen_trigger(jh_in, mapped_data,
381 					   jh_in->b_triggers);
382 
383 	/*
384 	 * Check for escaping
385 	 */
386 	if (*((__be32 *)mapped_data) == cpu_to_be32(JBD2_MAGIC_NUMBER)) {
387 		need_copy_out = 1;
388 		do_escape = 1;
389 	}
390 	kunmap_local(mapped_data);
391 
392 	/*
393 	 * Do we need to do a data copy?
394 	 */
395 	if (need_copy_out && !done_copy_out) {
396 		char *tmp;
397 
398 		spin_unlock(&jh_in->b_state_lock);
399 		tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
400 		if (!tmp) {
401 			brelse(new_bh);
402 			return -ENOMEM;
403 		}
404 		spin_lock(&jh_in->b_state_lock);
405 		if (jh_in->b_frozen_data) {
406 			jbd2_free(tmp, bh_in->b_size);
407 			goto repeat;
408 		}
409 
410 		jh_in->b_frozen_data = tmp;
411 		memcpy_from_folio(tmp, new_folio, new_offset, bh_in->b_size);
412 
413 		new_folio = virt_to_folio(tmp);
414 		new_offset = offset_in_folio(new_folio, tmp);
415 		done_copy_out = 1;
416 
417 		/*
418 		 * This isn't strictly necessary, as we're using frozen
419 		 * data for the escaping, but it keeps consistency with
420 		 * b_frozen_data usage.
421 		 */
422 		jh_in->b_frozen_triggers = jh_in->b_triggers;
423 	}
424 
425 	/*
426 	 * Did we need to do an escaping?  Now we've done all the
427 	 * copying, we can finally do so.
428 	 */
429 	if (do_escape) {
430 		mapped_data = kmap_local_folio(new_folio, new_offset);
431 		*((unsigned int *)mapped_data) = 0;
432 		kunmap_local(mapped_data);
433 	}
434 
435 	folio_set_bh(new_bh, new_folio, new_offset);
436 	new_bh->b_size = bh_in->b_size;
437 	new_bh->b_bdev = journal->j_dev;
438 	new_bh->b_blocknr = blocknr;
439 	new_bh->b_private = bh_in;
440 	set_buffer_mapped(new_bh);
441 	set_buffer_dirty(new_bh);
442 
443 	*bh_out = new_bh;
444 
445 	/*
446 	 * The to-be-written buffer needs to get moved to the io queue,
447 	 * and the original buffer whose contents we are shadowing or
448 	 * copying is moved to the transaction's shadow queue.
449 	 */
450 	JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
451 	spin_lock(&journal->j_list_lock);
452 	__jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
453 	spin_unlock(&journal->j_list_lock);
454 	set_buffer_shadow(bh_in);
455 	spin_unlock(&jh_in->b_state_lock);
456 
457 	return do_escape | (done_copy_out << 1);
458 }
459 
460 /*
461  * Allocation code for the journal file.  Manage the space left in the
462  * journal, so that we can begin checkpointing when appropriate.
463  */
464 
465 /*
466  * Called with j_state_lock locked for writing.
467  * Returns true if a transaction commit was started.
468  */
__jbd2_log_start_commit(journal_t * journal,tid_t target)469 static int __jbd2_log_start_commit(journal_t *journal, tid_t target)
470 {
471 	/* Return if the txn has already requested to be committed */
472 	if (journal->j_commit_request == target)
473 		return 0;
474 
475 	/*
476 	 * The only transaction we can possibly wait upon is the
477 	 * currently running transaction (if it exists).  Otherwise,
478 	 * the target tid must be an old one.
479 	 */
480 	if (journal->j_running_transaction &&
481 	    journal->j_running_transaction->t_tid == target) {
482 		/*
483 		 * We want a new commit: OK, mark the request and wakeup the
484 		 * commit thread.  We do _not_ do the commit ourselves.
485 		 */
486 
487 		journal->j_commit_request = target;
488 		jbd2_debug(1, "JBD2: requesting commit %u/%u\n",
489 			  journal->j_commit_request,
490 			  journal->j_commit_sequence);
491 		journal->j_running_transaction->t_requested = jiffies;
492 		wake_up(&journal->j_wait_commit);
493 		return 1;
494 	} else if (!tid_geq(journal->j_commit_request, target))
495 		/* This should never happen, but if it does, preserve
496 		   the evidence before kjournald goes into a loop and
497 		   increments j_commit_sequence beyond all recognition. */
498 		WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
499 			  journal->j_commit_request,
500 			  journal->j_commit_sequence,
501 			  target, journal->j_running_transaction ?
502 			  journal->j_running_transaction->t_tid : 0);
503 	return 0;
504 }
505 
jbd2_log_start_commit(journal_t * journal,tid_t tid)506 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
507 {
508 	int ret;
509 
510 	write_lock(&journal->j_state_lock);
511 	ret = __jbd2_log_start_commit(journal, tid);
512 	write_unlock(&journal->j_state_lock);
513 	return ret;
514 }
515 
516 /*
517  * Force and wait any uncommitted transactions.  We can only force the running
518  * transaction if we don't have an active handle, otherwise, we will deadlock.
519  * Returns: <0 in case of error,
520  *           0 if nothing to commit,
521  *           1 if transaction was successfully committed.
522  */
__jbd2_journal_force_commit(journal_t * journal)523 static int __jbd2_journal_force_commit(journal_t *journal)
524 {
525 	transaction_t *transaction = NULL;
526 	tid_t tid;
527 	int need_to_start = 0, ret = 0;
528 
529 	read_lock(&journal->j_state_lock);
530 	if (journal->j_running_transaction && !current->journal_info) {
531 		transaction = journal->j_running_transaction;
532 		if (!tid_geq(journal->j_commit_request, transaction->t_tid))
533 			need_to_start = 1;
534 	} else if (journal->j_committing_transaction)
535 		transaction = journal->j_committing_transaction;
536 
537 	if (!transaction) {
538 		/* Nothing to commit */
539 		read_unlock(&journal->j_state_lock);
540 		return 0;
541 	}
542 	tid = transaction->t_tid;
543 	read_unlock(&journal->j_state_lock);
544 	if (need_to_start)
545 		jbd2_log_start_commit(journal, tid);
546 	ret = jbd2_log_wait_commit(journal, tid);
547 	if (!ret)
548 		ret = 1;
549 
550 	return ret;
551 }
552 
553 /**
554  * jbd2_journal_force_commit_nested - Force and wait upon a commit if the
555  * calling process is not within transaction.
556  *
557  * @journal: journal to force
558  * Returns true if progress was made.
559  *
560  * This is used for forcing out undo-protected data which contains
561  * bitmaps, when the fs is running out of space.
562  */
jbd2_journal_force_commit_nested(journal_t * journal)563 int jbd2_journal_force_commit_nested(journal_t *journal)
564 {
565 	int ret;
566 
567 	ret = __jbd2_journal_force_commit(journal);
568 	return ret > 0;
569 }
570 
571 /**
572  * jbd2_journal_force_commit() - force any uncommitted transactions
573  * @journal: journal to force
574  *
575  * Caller want unconditional commit. We can only force the running transaction
576  * if we don't have an active handle, otherwise, we will deadlock.
577  */
jbd2_journal_force_commit(journal_t * journal)578 int jbd2_journal_force_commit(journal_t *journal)
579 {
580 	int ret;
581 
582 	J_ASSERT(!current->journal_info);
583 	ret = __jbd2_journal_force_commit(journal);
584 	if (ret > 0)
585 		ret = 0;
586 	return ret;
587 }
588 
589 /*
590  * Start a commit of the current running transaction (if any).  Returns true
591  * if a transaction is going to be committed (or is currently already
592  * committing), and fills its tid in at *ptid
593  */
jbd2_journal_start_commit(journal_t * journal,tid_t * ptid)594 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
595 {
596 	int ret = 0;
597 
598 	write_lock(&journal->j_state_lock);
599 	if (journal->j_running_transaction) {
600 		tid_t tid = journal->j_running_transaction->t_tid;
601 
602 		__jbd2_log_start_commit(journal, tid);
603 		/* There's a running transaction and we've just made sure
604 		 * it's commit has been scheduled. */
605 		if (ptid)
606 			*ptid = tid;
607 		ret = 1;
608 	} else if (journal->j_committing_transaction) {
609 		/*
610 		 * If commit has been started, then we have to wait for
611 		 * completion of that transaction.
612 		 */
613 		if (ptid)
614 			*ptid = journal->j_committing_transaction->t_tid;
615 		ret = 1;
616 	}
617 	write_unlock(&journal->j_state_lock);
618 	return ret;
619 }
620 
621 /*
622  * Return 1 if a given transaction has not yet sent barrier request
623  * connected with a transaction commit. If 0 is returned, transaction
624  * may or may not have sent the barrier. Used to avoid sending barrier
625  * twice in common cases.
626  */
jbd2_trans_will_send_data_barrier(journal_t * journal,tid_t tid)627 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
628 {
629 	int ret = 0;
630 	transaction_t *commit_trans;
631 
632 	if (!(journal->j_flags & JBD2_BARRIER))
633 		return 0;
634 	read_lock(&journal->j_state_lock);
635 	/* Transaction already committed? */
636 	if (tid_geq(journal->j_commit_sequence, tid))
637 		goto out;
638 	commit_trans = journal->j_committing_transaction;
639 	if (!commit_trans || commit_trans->t_tid != tid) {
640 		ret = 1;
641 		goto out;
642 	}
643 	/*
644 	 * Transaction is being committed and we already proceeded to
645 	 * submitting a flush to fs partition?
646 	 */
647 	if (journal->j_fs_dev != journal->j_dev) {
648 		if (!commit_trans->t_need_data_flush ||
649 		    commit_trans->t_state >= T_COMMIT_DFLUSH)
650 			goto out;
651 	} else {
652 		if (commit_trans->t_state >= T_COMMIT_JFLUSH)
653 			goto out;
654 	}
655 	ret = 1;
656 out:
657 	read_unlock(&journal->j_state_lock);
658 	return ret;
659 }
660 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
661 
662 /*
663  * Wait for a specified commit to complete.
664  * The caller may not hold the journal lock.
665  */
jbd2_log_wait_commit(journal_t * journal,tid_t tid)666 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
667 {
668 	int err = 0;
669 
670 	read_lock(&journal->j_state_lock);
671 #ifdef CONFIG_PROVE_LOCKING
672 	/*
673 	 * Some callers make sure transaction is already committing and in that
674 	 * case we cannot block on open handles anymore. So don't warn in that
675 	 * case.
676 	 */
677 	if (tid_gt(tid, journal->j_commit_sequence) &&
678 	    (!journal->j_committing_transaction ||
679 	     journal->j_committing_transaction->t_tid != tid)) {
680 		read_unlock(&journal->j_state_lock);
681 		jbd2_might_wait_for_commit(journal);
682 		read_lock(&journal->j_state_lock);
683 	}
684 #endif
685 #ifdef CONFIG_JBD2_DEBUG
686 	if (!tid_geq(journal->j_commit_request, tid)) {
687 		printk(KERN_ERR
688 		       "%s: error: j_commit_request=%u, tid=%u\n",
689 		       __func__, journal->j_commit_request, tid);
690 	}
691 #endif
692 	while (tid_gt(tid, journal->j_commit_sequence)) {
693 		jbd2_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
694 				  tid, journal->j_commit_sequence);
695 		read_unlock(&journal->j_state_lock);
696 		wake_up(&journal->j_wait_commit);
697 		wait_event(journal->j_wait_done_commit,
698 				!tid_gt(tid, journal->j_commit_sequence));
699 		read_lock(&journal->j_state_lock);
700 	}
701 	read_unlock(&journal->j_state_lock);
702 
703 	if (unlikely(is_journal_aborted(journal)))
704 		err = -EIO;
705 	return err;
706 }
707 
708 /*
709  * Start a fast commit. If there's an ongoing fast or full commit wait for
710  * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY
711  * if a fast commit is not needed, either because there's an already a commit
712  * going on or this tid has already been committed. Returns -EINVAL if no jbd2
713  * commit has yet been performed.
714  */
jbd2_fc_begin_commit(journal_t * journal,tid_t tid)715 int jbd2_fc_begin_commit(journal_t *journal, tid_t tid)
716 {
717 	if (unlikely(is_journal_aborted(journal)))
718 		return -EIO;
719 	/*
720 	 * Fast commits only allowed if at least one full commit has
721 	 * been processed.
722 	 */
723 	if (!journal->j_stats.ts_tid)
724 		return -EINVAL;
725 
726 	write_lock(&journal->j_state_lock);
727 	if (tid <= journal->j_commit_sequence) {
728 		write_unlock(&journal->j_state_lock);
729 		return -EALREADY;
730 	}
731 
732 	if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
733 	    (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) {
734 		DEFINE_WAIT(wait);
735 
736 		prepare_to_wait(&journal->j_fc_wait, &wait,
737 				TASK_UNINTERRUPTIBLE);
738 		write_unlock(&journal->j_state_lock);
739 		schedule();
740 		finish_wait(&journal->j_fc_wait, &wait);
741 		return -EALREADY;
742 	}
743 	journal->j_flags |= JBD2_FAST_COMMIT_ONGOING;
744 	write_unlock(&journal->j_state_lock);
745 	jbd2_journal_lock_updates(journal);
746 
747 	return 0;
748 }
749 EXPORT_SYMBOL(jbd2_fc_begin_commit);
750 
751 /*
752  * Stop a fast commit. If fallback is set, this function starts commit of
753  * TID tid before any other fast commit can start.
754  */
__jbd2_fc_end_commit(journal_t * journal,tid_t tid,bool fallback)755 static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback)
756 {
757 	jbd2_journal_unlock_updates(journal);
758 	if (journal->j_fc_cleanup_callback)
759 		journal->j_fc_cleanup_callback(journal, 0, tid);
760 	write_lock(&journal->j_state_lock);
761 	journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
762 	if (fallback)
763 		journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;
764 	write_unlock(&journal->j_state_lock);
765 	wake_up(&journal->j_fc_wait);
766 	if (fallback)
767 		return jbd2_complete_transaction(journal, tid);
768 	return 0;
769 }
770 
jbd2_fc_end_commit(journal_t * journal)771 int jbd2_fc_end_commit(journal_t *journal)
772 {
773 	return __jbd2_fc_end_commit(journal, 0, false);
774 }
775 EXPORT_SYMBOL(jbd2_fc_end_commit);
776 
jbd2_fc_end_commit_fallback(journal_t * journal)777 int jbd2_fc_end_commit_fallback(journal_t *journal)
778 {
779 	tid_t tid;
780 
781 	read_lock(&journal->j_state_lock);
782 	tid = journal->j_running_transaction ?
783 		journal->j_running_transaction->t_tid : 0;
784 	read_unlock(&journal->j_state_lock);
785 	return __jbd2_fc_end_commit(journal, tid, true);
786 }
787 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback);
788 
789 /* Return 1 when transaction with given tid has already committed. */
jbd2_transaction_committed(journal_t * journal,tid_t tid)790 int jbd2_transaction_committed(journal_t *journal, tid_t tid)
791 {
792 	int ret = 1;
793 
794 	read_lock(&journal->j_state_lock);
795 	if (journal->j_running_transaction &&
796 	    journal->j_running_transaction->t_tid == tid)
797 		ret = 0;
798 	if (journal->j_committing_transaction &&
799 	    journal->j_committing_transaction->t_tid == tid)
800 		ret = 0;
801 	read_unlock(&journal->j_state_lock);
802 	return ret;
803 }
804 EXPORT_SYMBOL(jbd2_transaction_committed);
805 
806 /*
807  * When this function returns the transaction corresponding to tid
808  * will be completed.  If the transaction has currently running, start
809  * committing that transaction before waiting for it to complete.  If
810  * the transaction id is stale, it is by definition already completed,
811  * so just return SUCCESS.
812  */
jbd2_complete_transaction(journal_t * journal,tid_t tid)813 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
814 {
815 	int	need_to_wait = 1;
816 
817 	read_lock(&journal->j_state_lock);
818 	if (journal->j_running_transaction &&
819 	    journal->j_running_transaction->t_tid == tid) {
820 		if (journal->j_commit_request != tid) {
821 			/* transaction not yet started, so request it */
822 			read_unlock(&journal->j_state_lock);
823 			jbd2_log_start_commit(journal, tid);
824 			goto wait_commit;
825 		}
826 	} else if (!(journal->j_committing_transaction &&
827 		     journal->j_committing_transaction->t_tid == tid))
828 		need_to_wait = 0;
829 	read_unlock(&journal->j_state_lock);
830 	if (!need_to_wait)
831 		return 0;
832 wait_commit:
833 	return jbd2_log_wait_commit(journal, tid);
834 }
835 EXPORT_SYMBOL(jbd2_complete_transaction);
836 
837 /*
838  * Log buffer allocation routines:
839  */
840 
jbd2_journal_next_log_block(journal_t * journal,unsigned long long * retp)841 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
842 {
843 	unsigned long blocknr;
844 
845 	write_lock(&journal->j_state_lock);
846 	J_ASSERT(journal->j_free > 1);
847 
848 	blocknr = journal->j_head;
849 	journal->j_head++;
850 	journal->j_free--;
851 	if (journal->j_head == journal->j_last)
852 		journal->j_head = journal->j_first;
853 	write_unlock(&journal->j_state_lock);
854 	return jbd2_journal_bmap(journal, blocknr, retp);
855 }
856 
857 /* Map one fast commit buffer for use by the file system */
jbd2_fc_get_buf(journal_t * journal,struct buffer_head ** bh_out)858 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out)
859 {
860 	unsigned long long pblock;
861 	unsigned long blocknr;
862 	int ret = 0;
863 	struct buffer_head *bh;
864 	int fc_off;
865 
866 	*bh_out = NULL;
867 
868 	if (journal->j_fc_off + journal->j_fc_first < journal->j_fc_last) {
869 		fc_off = journal->j_fc_off;
870 		blocknr = journal->j_fc_first + fc_off;
871 		journal->j_fc_off++;
872 	} else {
873 		ret = -EINVAL;
874 	}
875 
876 	if (ret)
877 		return ret;
878 
879 	ret = jbd2_journal_bmap(journal, blocknr, &pblock);
880 	if (ret)
881 		return ret;
882 
883 	bh = __getblk(journal->j_dev, pblock, journal->j_blocksize);
884 	if (!bh)
885 		return -ENOMEM;
886 
887 
888 	journal->j_fc_wbuf[fc_off] = bh;
889 
890 	*bh_out = bh;
891 
892 	return 0;
893 }
894 EXPORT_SYMBOL(jbd2_fc_get_buf);
895 
896 /*
897  * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
898  * for completion.
899  */
jbd2_fc_wait_bufs(journal_t * journal,int num_blks)900 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks)
901 {
902 	struct buffer_head *bh;
903 	int i, j_fc_off;
904 
905 	j_fc_off = journal->j_fc_off;
906 
907 	/*
908 	 * Wait in reverse order to minimize chances of us being woken up before
909 	 * all IOs have completed
910 	 */
911 	for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) {
912 		bh = journal->j_fc_wbuf[i];
913 		wait_on_buffer(bh);
914 		/*
915 		 * Update j_fc_off so jbd2_fc_release_bufs can release remain
916 		 * buffer head.
917 		 */
918 		if (unlikely(!buffer_uptodate(bh))) {
919 			journal->j_fc_off = i + 1;
920 			return -EIO;
921 		}
922 		put_bh(bh);
923 		journal->j_fc_wbuf[i] = NULL;
924 	}
925 
926 	return 0;
927 }
928 EXPORT_SYMBOL(jbd2_fc_wait_bufs);
929 
jbd2_fc_release_bufs(journal_t * journal)930 int jbd2_fc_release_bufs(journal_t *journal)
931 {
932 	struct buffer_head *bh;
933 	int i, j_fc_off;
934 
935 	j_fc_off = journal->j_fc_off;
936 
937 	for (i = j_fc_off - 1; i >= 0; i--) {
938 		bh = journal->j_fc_wbuf[i];
939 		if (!bh)
940 			break;
941 		put_bh(bh);
942 		journal->j_fc_wbuf[i] = NULL;
943 	}
944 
945 	return 0;
946 }
947 EXPORT_SYMBOL(jbd2_fc_release_bufs);
948 
949 /*
950  * Conversion of logical to physical block numbers for the journal
951  *
952  * On external journals the journal blocks are identity-mapped, so
953  * this is a no-op.  If needed, we can use j_blk_offset - everything is
954  * ready.
955  */
jbd2_journal_bmap(journal_t * journal,unsigned long blocknr,unsigned long long * retp)956 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
957 		 unsigned long long *retp)
958 {
959 	int err = 0;
960 	unsigned long long ret;
961 	sector_t block = blocknr;
962 
963 	if (journal->j_bmap) {
964 		err = journal->j_bmap(journal, &block);
965 		if (err == 0)
966 			*retp = block;
967 	} else if (journal->j_inode) {
968 		ret = bmap(journal->j_inode, &block);
969 
970 		if (ret || !block) {
971 			printk(KERN_ALERT "%s: journal block not found "
972 					"at offset %lu on %s\n",
973 			       __func__, blocknr, journal->j_devname);
974 			err = -EIO;
975 			jbd2_journal_abort(journal, err);
976 		} else {
977 			*retp = block;
978 		}
979 
980 	} else {
981 		*retp = blocknr; /* +journal->j_blk_offset */
982 	}
983 	return err;
984 }
985 
986 /*
987  * We play buffer_head aliasing tricks to write data/metadata blocks to
988  * the journal without copying their contents, but for journal
989  * descriptor blocks we do need to generate bona fide buffers.
990  *
991  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
992  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
993  * But we don't bother doing that, so there will be coherency problems with
994  * mmaps of blockdevs which hold live JBD-controlled filesystems.
995  */
996 struct buffer_head *
jbd2_journal_get_descriptor_buffer(transaction_t * transaction,int type)997 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
998 {
999 	journal_t *journal = transaction->t_journal;
1000 	struct buffer_head *bh;
1001 	unsigned long long blocknr;
1002 	journal_header_t *header;
1003 	int err;
1004 
1005 	err = jbd2_journal_next_log_block(journal, &blocknr);
1006 
1007 	if (err)
1008 		return NULL;
1009 
1010 	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
1011 	if (!bh)
1012 		return NULL;
1013 	atomic_dec(&transaction->t_outstanding_credits);
1014 	lock_buffer(bh);
1015 	memset(bh->b_data, 0, journal->j_blocksize);
1016 	header = (journal_header_t *)bh->b_data;
1017 	header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
1018 	header->h_blocktype = cpu_to_be32(type);
1019 	header->h_sequence = cpu_to_be32(transaction->t_tid);
1020 	set_buffer_uptodate(bh);
1021 	unlock_buffer(bh);
1022 	BUFFER_TRACE(bh, "return this buffer");
1023 	return bh;
1024 }
1025 
jbd2_descriptor_block_csum_set(journal_t * j,struct buffer_head * bh)1026 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
1027 {
1028 	struct jbd2_journal_block_tail *tail;
1029 	__u32 csum;
1030 
1031 	if (!jbd2_journal_has_csum_v2or3(j))
1032 		return;
1033 
1034 	tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
1035 			sizeof(struct jbd2_journal_block_tail));
1036 	tail->t_checksum = 0;
1037 	csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
1038 	tail->t_checksum = cpu_to_be32(csum);
1039 }
1040 
1041 /*
1042  * Return tid of the oldest transaction in the journal and block in the journal
1043  * where the transaction starts.
1044  *
1045  * If the journal is now empty, return which will be the next transaction ID
1046  * we will write and where will that transaction start.
1047  *
1048  * The return value is 0 if journal tail cannot be pushed any further, 1 if
1049  * it can.
1050  */
jbd2_journal_get_log_tail(journal_t * journal,tid_t * tid,unsigned long * block)1051 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
1052 			      unsigned long *block)
1053 {
1054 	transaction_t *transaction;
1055 	int ret;
1056 
1057 	read_lock(&journal->j_state_lock);
1058 	spin_lock(&journal->j_list_lock);
1059 	transaction = journal->j_checkpoint_transactions;
1060 	if (transaction) {
1061 		*tid = transaction->t_tid;
1062 		*block = transaction->t_log_start;
1063 	} else if ((transaction = journal->j_committing_transaction) != NULL) {
1064 		*tid = transaction->t_tid;
1065 		*block = transaction->t_log_start;
1066 	} else if ((transaction = journal->j_running_transaction) != NULL) {
1067 		*tid = transaction->t_tid;
1068 		*block = journal->j_head;
1069 	} else {
1070 		*tid = journal->j_transaction_sequence;
1071 		*block = journal->j_head;
1072 	}
1073 	ret = tid_gt(*tid, journal->j_tail_sequence);
1074 	spin_unlock(&journal->j_list_lock);
1075 	read_unlock(&journal->j_state_lock);
1076 
1077 	return ret;
1078 }
1079 
1080 /*
1081  * Update information in journal structure and in on disk journal superblock
1082  * about log tail. This function does not check whether information passed in
1083  * really pushes log tail further. It's responsibility of the caller to make
1084  * sure provided log tail information is valid (e.g. by holding
1085  * j_checkpoint_mutex all the time between computing log tail and calling this
1086  * function as is the case with jbd2_cleanup_journal_tail()).
1087  *
1088  * Requires j_checkpoint_mutex
1089  */
__jbd2_update_log_tail(journal_t * journal,tid_t tid,unsigned long block)1090 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1091 {
1092 	unsigned long freed;
1093 	int ret;
1094 
1095 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1096 
1097 	/*
1098 	 * We cannot afford for write to remain in drive's caches since as
1099 	 * soon as we update j_tail, next transaction can start reusing journal
1100 	 * space and if we lose sb update during power failure we'd replay
1101 	 * old transaction with possibly newly overwritten data.
1102 	 */
1103 	ret = jbd2_journal_update_sb_log_tail(journal, tid, block, REQ_FUA);
1104 	if (ret)
1105 		goto out;
1106 
1107 	write_lock(&journal->j_state_lock);
1108 	freed = block - journal->j_tail;
1109 	if (block < journal->j_tail)
1110 		freed += journal->j_last - journal->j_first;
1111 
1112 	trace_jbd2_update_log_tail(journal, tid, block, freed);
1113 	jbd2_debug(1,
1114 		  "Cleaning journal tail from %u to %u (offset %lu), "
1115 		  "freeing %lu\n",
1116 		  journal->j_tail_sequence, tid, block, freed);
1117 
1118 	journal->j_free += freed;
1119 	journal->j_tail_sequence = tid;
1120 	journal->j_tail = block;
1121 	write_unlock(&journal->j_state_lock);
1122 
1123 out:
1124 	return ret;
1125 }
1126 
1127 /*
1128  * This is a variation of __jbd2_update_log_tail which checks for validity of
1129  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
1130  * with other threads updating log tail.
1131  */
jbd2_update_log_tail(journal_t * journal,tid_t tid,unsigned long block)1132 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1133 {
1134 	mutex_lock_io(&journal->j_checkpoint_mutex);
1135 	if (tid_gt(tid, journal->j_tail_sequence))
1136 		__jbd2_update_log_tail(journal, tid, block);
1137 	mutex_unlock(&journal->j_checkpoint_mutex);
1138 }
1139 
1140 struct jbd2_stats_proc_session {
1141 	journal_t *journal;
1142 	struct transaction_stats_s *stats;
1143 	int start;
1144 	int max;
1145 };
1146 
jbd2_seq_info_start(struct seq_file * seq,loff_t * pos)1147 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
1148 {
1149 	return *pos ? NULL : SEQ_START_TOKEN;
1150 }
1151 
jbd2_seq_info_next(struct seq_file * seq,void * v,loff_t * pos)1152 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1153 {
1154 	(*pos)++;
1155 	return NULL;
1156 }
1157 
jbd2_seq_info_show(struct seq_file * seq,void * v)1158 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1159 {
1160 	struct jbd2_stats_proc_session *s = seq->private;
1161 
1162 	if (v != SEQ_START_TOKEN)
1163 		return 0;
1164 	seq_printf(seq, "%lu transactions (%lu requested), "
1165 		   "each up to %u blocks\n",
1166 		   s->stats->ts_tid, s->stats->ts_requested,
1167 		   s->journal->j_max_transaction_buffers);
1168 	if (s->stats->ts_tid == 0)
1169 		return 0;
1170 	seq_printf(seq, "average: \n  %ums waiting for transaction\n",
1171 	    jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1172 	seq_printf(seq, "  %ums request delay\n",
1173 	    (s->stats->ts_requested == 0) ? 0 :
1174 	    jiffies_to_msecs(s->stats->run.rs_request_delay /
1175 			     s->stats->ts_requested));
1176 	seq_printf(seq, "  %ums running transaction\n",
1177 	    jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1178 	seq_printf(seq, "  %ums transaction was being locked\n",
1179 	    jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1180 	seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
1181 	    jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1182 	seq_printf(seq, "  %ums logging transaction\n",
1183 	    jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1184 	seq_printf(seq, "  %lluus average transaction commit time\n",
1185 		   div_u64(s->journal->j_average_commit_time, 1000));
1186 	seq_printf(seq, "  %lu handles per transaction\n",
1187 	    s->stats->run.rs_handle_count / s->stats->ts_tid);
1188 	seq_printf(seq, "  %lu blocks per transaction\n",
1189 	    s->stats->run.rs_blocks / s->stats->ts_tid);
1190 	seq_printf(seq, "  %lu logged blocks per transaction\n",
1191 	    s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1192 	return 0;
1193 }
1194 
jbd2_seq_info_stop(struct seq_file * seq,void * v)1195 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1196 {
1197 }
1198 
1199 static const struct seq_operations jbd2_seq_info_ops = {
1200 	.start  = jbd2_seq_info_start,
1201 	.next   = jbd2_seq_info_next,
1202 	.stop   = jbd2_seq_info_stop,
1203 	.show   = jbd2_seq_info_show,
1204 };
1205 
jbd2_seq_info_open(struct inode * inode,struct file * file)1206 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1207 {
1208 	journal_t *journal = pde_data(inode);
1209 	struct jbd2_stats_proc_session *s;
1210 	int rc, size;
1211 
1212 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1213 	if (s == NULL)
1214 		return -ENOMEM;
1215 	size = sizeof(struct transaction_stats_s);
1216 	s->stats = kmalloc(size, GFP_KERNEL);
1217 	if (s->stats == NULL) {
1218 		kfree(s);
1219 		return -ENOMEM;
1220 	}
1221 	spin_lock(&journal->j_history_lock);
1222 	memcpy(s->stats, &journal->j_stats, size);
1223 	s->journal = journal;
1224 	spin_unlock(&journal->j_history_lock);
1225 
1226 	rc = seq_open(file, &jbd2_seq_info_ops);
1227 	if (rc == 0) {
1228 		struct seq_file *m = file->private_data;
1229 		m->private = s;
1230 	} else {
1231 		kfree(s->stats);
1232 		kfree(s);
1233 	}
1234 	return rc;
1235 
1236 }
1237 
jbd2_seq_info_release(struct inode * inode,struct file * file)1238 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1239 {
1240 	struct seq_file *seq = file->private_data;
1241 	struct jbd2_stats_proc_session *s = seq->private;
1242 	kfree(s->stats);
1243 	kfree(s);
1244 	return seq_release(inode, file);
1245 }
1246 
1247 static const struct proc_ops jbd2_info_proc_ops = {
1248 	.proc_open	= jbd2_seq_info_open,
1249 	.proc_read	= seq_read,
1250 	.proc_lseek	= seq_lseek,
1251 	.proc_release	= jbd2_seq_info_release,
1252 };
1253 
1254 static struct proc_dir_entry *proc_jbd2_stats;
1255 
jbd2_stats_proc_init(journal_t * journal)1256 static void jbd2_stats_proc_init(journal_t *journal)
1257 {
1258 	journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1259 	if (journal->j_proc_entry) {
1260 		proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1261 				 &jbd2_info_proc_ops, journal);
1262 	}
1263 }
1264 
jbd2_stats_proc_exit(journal_t * journal)1265 static void jbd2_stats_proc_exit(journal_t *journal)
1266 {
1267 	remove_proc_entry("info", journal->j_proc_entry);
1268 	remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1269 }
1270 
1271 /* Minimum size of descriptor tag */
jbd2_min_tag_size(void)1272 static int jbd2_min_tag_size(void)
1273 {
1274 	/*
1275 	 * Tag with 32-bit block numbers does not use last four bytes of the
1276 	 * structure
1277 	 */
1278 	return sizeof(journal_block_tag_t) - 4;
1279 }
1280 
1281 /**
1282  * jbd2_journal_shrink_scan()
1283  * @shrink: shrinker to work on
1284  * @sc: reclaim request to process
1285  *
1286  * Scan the checkpointed buffer on the checkpoint list and release the
1287  * journal_head.
1288  */
jbd2_journal_shrink_scan(struct shrinker * shrink,struct shrink_control * sc)1289 static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink,
1290 					      struct shrink_control *sc)
1291 {
1292 	journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1293 	unsigned long nr_to_scan = sc->nr_to_scan;
1294 	unsigned long nr_shrunk;
1295 	unsigned long count;
1296 
1297 	count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1298 	trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count);
1299 
1300 	nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan);
1301 
1302 	count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1303 	trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count);
1304 
1305 	return nr_shrunk;
1306 }
1307 
1308 /**
1309  * jbd2_journal_shrink_count()
1310  * @shrink: shrinker to work on
1311  * @sc: reclaim request to process
1312  *
1313  * Count the number of checkpoint buffers on the checkpoint list.
1314  */
jbd2_journal_shrink_count(struct shrinker * shrink,struct shrink_control * sc)1315 static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink,
1316 					       struct shrink_control *sc)
1317 {
1318 	journal_t *journal = container_of(shrink, journal_t, j_shrinker);
1319 	unsigned long count;
1320 
1321 	count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1322 	trace_jbd2_shrink_count(journal, sc->nr_to_scan, count);
1323 
1324 	return count;
1325 }
1326 
1327 /*
1328  * If the journal init or create aborts, we need to mark the journal
1329  * superblock as being NULL to prevent the journal destroy from writing
1330  * back a bogus superblock.
1331  */
journal_fail_superblock(journal_t * journal)1332 static void journal_fail_superblock(journal_t *journal)
1333 {
1334 	struct buffer_head *bh = journal->j_sb_buffer;
1335 	brelse(bh);
1336 	journal->j_sb_buffer = NULL;
1337 }
1338 
1339 /*
1340  * Check the superblock for a given journal, performing initial
1341  * validation of the format.
1342  */
journal_check_superblock(journal_t * journal)1343 static int journal_check_superblock(journal_t *journal)
1344 {
1345 	journal_superblock_t *sb = journal->j_superblock;
1346 	int num_fc_blks;
1347 	int err = -EINVAL;
1348 
1349 	if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1350 	    sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1351 		printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1352 		return err;
1353 	}
1354 
1355 	if (be32_to_cpu(sb->s_header.h_blocktype) != JBD2_SUPERBLOCK_V1 &&
1356 	    be32_to_cpu(sb->s_header.h_blocktype) != JBD2_SUPERBLOCK_V2) {
1357 		printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1358 		return err;
1359 	}
1360 
1361 	if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) {
1362 		printk(KERN_WARNING "JBD2: journal file too short\n");
1363 		return err;
1364 	}
1365 
1366 	if (be32_to_cpu(sb->s_first) == 0 ||
1367 	    be32_to_cpu(sb->s_first) >= journal->j_total_len) {
1368 		printk(KERN_WARNING
1369 			"JBD2: Invalid start block of journal: %u\n",
1370 			be32_to_cpu(sb->s_first));
1371 		return err;
1372 	}
1373 
1374 	/*
1375 	 * If this is a V2 superblock, then we have to check the
1376 	 * features flags on it.
1377 	 */
1378 	if (!jbd2_format_support_feature(journal))
1379 		return 0;
1380 
1381 	if ((sb->s_feature_ro_compat &
1382 			~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1383 	    (sb->s_feature_incompat &
1384 			~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1385 		printk(KERN_WARNING "JBD2: Unrecognised features on journal\n");
1386 		return err;
1387 	}
1388 
1389 	num_fc_blks = jbd2_has_feature_fast_commit(journal) ?
1390 				jbd2_journal_get_num_fc_blks(sb) : 0;
1391 	if (be32_to_cpu(sb->s_maxlen) < JBD2_MIN_JOURNAL_BLOCKS ||
1392 	    be32_to_cpu(sb->s_maxlen) - JBD2_MIN_JOURNAL_BLOCKS < num_fc_blks) {
1393 		printk(KERN_ERR "JBD2: journal file too short %u,%d\n",
1394 		       be32_to_cpu(sb->s_maxlen), num_fc_blks);
1395 		return err;
1396 	}
1397 
1398 	if (jbd2_has_feature_csum2(journal) &&
1399 	    jbd2_has_feature_csum3(journal)) {
1400 		/* Can't have checksum v2 and v3 at the same time! */
1401 		printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1402 		       "at the same time!\n");
1403 		return err;
1404 	}
1405 
1406 	if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1407 	    jbd2_has_feature_checksum(journal)) {
1408 		/* Can't have checksum v1 and v2 on at the same time! */
1409 		printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1410 		       "at the same time!\n");
1411 		return err;
1412 	}
1413 
1414 	/* Load the checksum driver */
1415 	if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1416 		if (sb->s_checksum_type != JBD2_CRC32C_CHKSUM) {
1417 			printk(KERN_ERR "JBD2: Unknown checksum type\n");
1418 			return err;
1419 		}
1420 
1421 		journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1422 		if (IS_ERR(journal->j_chksum_driver)) {
1423 			printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1424 			err = PTR_ERR(journal->j_chksum_driver);
1425 			journal->j_chksum_driver = NULL;
1426 			return err;
1427 		}
1428 		/* Check superblock checksum */
1429 		if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1430 			printk(KERN_ERR "JBD2: journal checksum error\n");
1431 			err = -EFSBADCRC;
1432 			return err;
1433 		}
1434 	}
1435 
1436 	return 0;
1437 }
1438 
journal_revoke_records_per_block(journal_t * journal)1439 static int journal_revoke_records_per_block(journal_t *journal)
1440 {
1441 	int record_size;
1442 	int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1443 
1444 	if (jbd2_has_feature_64bit(journal))
1445 		record_size = 8;
1446 	else
1447 		record_size = 4;
1448 
1449 	if (jbd2_journal_has_csum_v2or3(journal))
1450 		space -= sizeof(struct jbd2_journal_block_tail);
1451 	return space / record_size;
1452 }
1453 
1454 /*
1455  * Load the on-disk journal superblock and read the key fields into the
1456  * journal_t.
1457  */
journal_load_superblock(journal_t * journal)1458 static int journal_load_superblock(journal_t *journal)
1459 {
1460 	int err;
1461 	struct buffer_head *bh;
1462 	journal_superblock_t *sb;
1463 
1464 	bh = getblk_unmovable(journal->j_dev, journal->j_blk_offset,
1465 			      journal->j_blocksize);
1466 	if (bh)
1467 		err = bh_read(bh, 0);
1468 	if (!bh || err < 0) {
1469 		pr_err("%s: Cannot read journal superblock\n", __func__);
1470 		brelse(bh);
1471 		return -EIO;
1472 	}
1473 
1474 	journal->j_sb_buffer = bh;
1475 	sb = (journal_superblock_t *)bh->b_data;
1476 	journal->j_superblock = sb;
1477 	err = journal_check_superblock(journal);
1478 	if (err) {
1479 		journal_fail_superblock(journal);
1480 		return err;
1481 	}
1482 
1483 	journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1484 	journal->j_tail = be32_to_cpu(sb->s_start);
1485 	journal->j_first = be32_to_cpu(sb->s_first);
1486 	journal->j_errno = be32_to_cpu(sb->s_errno);
1487 	journal->j_last = be32_to_cpu(sb->s_maxlen);
1488 
1489 	if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len)
1490 		journal->j_total_len = be32_to_cpu(sb->s_maxlen);
1491 	/* Precompute checksum seed for all metadata */
1492 	if (jbd2_journal_has_csum_v2or3(journal))
1493 		journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1494 						   sizeof(sb->s_uuid));
1495 	journal->j_revoke_records_per_block =
1496 				journal_revoke_records_per_block(journal);
1497 
1498 	if (jbd2_has_feature_fast_commit(journal)) {
1499 		journal->j_fc_last = be32_to_cpu(sb->s_maxlen);
1500 		journal->j_last = journal->j_fc_last -
1501 				  jbd2_journal_get_num_fc_blks(sb);
1502 		journal->j_fc_first = journal->j_last + 1;
1503 		journal->j_fc_off = 0;
1504 	}
1505 
1506 	return 0;
1507 }
1508 
1509 
1510 /*
1511  * Management for journal control blocks: functions to create and
1512  * destroy journal_t structures, and to initialise and read existing
1513  * journal blocks from disk.  */
1514 
1515 /* First: create and setup a journal_t object in memory.  We initialise
1516  * very few fields yet: that has to wait until we have created the
1517  * journal structures from from scratch, or loaded them from disk. */
1518 
journal_init_common(struct block_device * bdev,struct block_device * fs_dev,unsigned long long start,int len,int blocksize)1519 static journal_t *journal_init_common(struct block_device *bdev,
1520 			struct block_device *fs_dev,
1521 			unsigned long long start, int len, int blocksize)
1522 {
1523 	static struct lock_class_key jbd2_trans_commit_key;
1524 	journal_t *journal;
1525 	int err;
1526 	int n;
1527 
1528 	journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1529 	if (!journal)
1530 		return ERR_PTR(-ENOMEM);
1531 
1532 	journal->j_blocksize = blocksize;
1533 	journal->j_dev = bdev;
1534 	journal->j_fs_dev = fs_dev;
1535 	journal->j_blk_offset = start;
1536 	journal->j_total_len = len;
1537 
1538 	err = journal_load_superblock(journal);
1539 	if (err)
1540 		goto err_cleanup;
1541 
1542 	init_waitqueue_head(&journal->j_wait_transaction_locked);
1543 	init_waitqueue_head(&journal->j_wait_done_commit);
1544 	init_waitqueue_head(&journal->j_wait_commit);
1545 	init_waitqueue_head(&journal->j_wait_updates);
1546 	init_waitqueue_head(&journal->j_wait_reserved);
1547 	init_waitqueue_head(&journal->j_fc_wait);
1548 	mutex_init(&journal->j_abort_mutex);
1549 	mutex_init(&journal->j_barrier);
1550 	mutex_init(&journal->j_checkpoint_mutex);
1551 	spin_lock_init(&journal->j_revoke_lock);
1552 	spin_lock_init(&journal->j_list_lock);
1553 	spin_lock_init(&journal->j_history_lock);
1554 	rwlock_init(&journal->j_state_lock);
1555 
1556 	journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1557 	journal->j_min_batch_time = 0;
1558 	journal->j_max_batch_time = 15000; /* 15ms */
1559 	atomic_set(&journal->j_reserved_credits, 0);
1560 	lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1561 			 &jbd2_trans_commit_key, 0);
1562 
1563 	/* The journal is marked for error until we succeed with recovery! */
1564 	journal->j_flags = JBD2_ABORT;
1565 
1566 	/* Set up a default-sized revoke table for the new mount. */
1567 	err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1568 	if (err)
1569 		goto err_cleanup;
1570 
1571 	/*
1572 	 * journal descriptor can store up to n blocks, we need enough
1573 	 * buffers to write out full descriptor block.
1574 	 */
1575 	err = -ENOMEM;
1576 	n = journal->j_blocksize / jbd2_min_tag_size();
1577 	journal->j_wbufsize = n;
1578 	journal->j_fc_wbuf = NULL;
1579 	journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1580 					GFP_KERNEL);
1581 	if (!journal->j_wbuf)
1582 		goto err_cleanup;
1583 
1584 	err = percpu_counter_init(&journal->j_checkpoint_jh_count, 0,
1585 				  GFP_KERNEL);
1586 	if (err)
1587 		goto err_cleanup;
1588 
1589 	journal->j_shrink_transaction = NULL;
1590 	journal->j_shrinker.scan_objects = jbd2_journal_shrink_scan;
1591 	journal->j_shrinker.count_objects = jbd2_journal_shrink_count;
1592 	journal->j_shrinker.seeks = DEFAULT_SEEKS;
1593 	journal->j_shrinker.batch = journal->j_max_transaction_buffers;
1594 	err = register_shrinker(&journal->j_shrinker, "jbd2-journal:(%u:%u)",
1595 				MAJOR(bdev->bd_dev), MINOR(bdev->bd_dev));
1596 	if (err)
1597 		goto err_cleanup;
1598 
1599 	return journal;
1600 
1601 err_cleanup:
1602 	percpu_counter_destroy(&journal->j_checkpoint_jh_count);
1603 	if (journal->j_chksum_driver)
1604 		crypto_free_shash(journal->j_chksum_driver);
1605 	kfree(journal->j_wbuf);
1606 	jbd2_journal_destroy_revoke(journal);
1607 	journal_fail_superblock(journal);
1608 	kfree(journal);
1609 	return ERR_PTR(err);
1610 }
1611 
1612 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1613  *
1614  * Create a journal structure assigned some fixed set of disk blocks to
1615  * the journal.  We don't actually touch those disk blocks yet, but we
1616  * need to set up all of the mapping information to tell the journaling
1617  * system where the journal blocks are.
1618  *
1619  */
1620 
1621 /**
1622  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1623  *  @bdev: Block device on which to create the journal
1624  *  @fs_dev: Device which hold journalled filesystem for this journal.
1625  *  @start: Block nr Start of journal.
1626  *  @len:  Length of the journal in blocks.
1627  *  @blocksize: blocksize of journalling device
1628  *
1629  *  Returns: a newly created journal_t *
1630  *
1631  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1632  *  range of blocks on an arbitrary block device.
1633  *
1634  */
jbd2_journal_init_dev(struct block_device * bdev,struct block_device * fs_dev,unsigned long long start,int len,int blocksize)1635 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1636 			struct block_device *fs_dev,
1637 			unsigned long long start, int len, int blocksize)
1638 {
1639 	journal_t *journal;
1640 
1641 	journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1642 	if (IS_ERR(journal))
1643 		return ERR_CAST(journal);
1644 
1645 	snprintf(journal->j_devname, sizeof(journal->j_devname),
1646 		 "%pg", journal->j_dev);
1647 	strreplace(journal->j_devname, '/', '!');
1648 	jbd2_stats_proc_init(journal);
1649 
1650 	return journal;
1651 }
1652 
1653 /**
1654  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1655  *  @inode: An inode to create the journal in
1656  *
1657  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1658  * the journal.  The inode must exist already, must support bmap() and
1659  * must have all data blocks preallocated.
1660  */
jbd2_journal_init_inode(struct inode * inode)1661 journal_t *jbd2_journal_init_inode(struct inode *inode)
1662 {
1663 	journal_t *journal;
1664 	sector_t blocknr;
1665 	int err = 0;
1666 
1667 	blocknr = 0;
1668 	err = bmap(inode, &blocknr);
1669 	if (err || !blocknr) {
1670 		pr_err("%s: Cannot locate journal superblock\n", __func__);
1671 		return err ? ERR_PTR(err) : ERR_PTR(-EINVAL);
1672 	}
1673 
1674 	jbd2_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1675 		  inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1676 		  inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1677 
1678 	journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1679 			blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1680 			inode->i_sb->s_blocksize);
1681 	if (IS_ERR(journal))
1682 		return ERR_CAST(journal);
1683 
1684 	journal->j_inode = inode;
1685 	snprintf(journal->j_devname, sizeof(journal->j_devname),
1686 		 "%pg-%lu", journal->j_dev, journal->j_inode->i_ino);
1687 	strreplace(journal->j_devname, '/', '!');
1688 	jbd2_stats_proc_init(journal);
1689 
1690 	return journal;
1691 }
1692 
1693 /*
1694  * Given a journal_t structure, initialise the various fields for
1695  * startup of a new journaling session.  We use this both when creating
1696  * a journal, and after recovering an old journal to reset it for
1697  * subsequent use.
1698  */
1699 
journal_reset(journal_t * journal)1700 static int journal_reset(journal_t *journal)
1701 {
1702 	journal_superblock_t *sb = journal->j_superblock;
1703 	unsigned long long first, last;
1704 
1705 	first = be32_to_cpu(sb->s_first);
1706 	last = be32_to_cpu(sb->s_maxlen);
1707 	if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1708 		printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1709 		       first, last);
1710 		journal_fail_superblock(journal);
1711 		return -EINVAL;
1712 	}
1713 
1714 	journal->j_first = first;
1715 	journal->j_last = last;
1716 
1717 	if (journal->j_head != 0 && journal->j_flags & JBD2_CYCLE_RECORD) {
1718 		/*
1719 		 * Disable the cycled recording mode if the journal head block
1720 		 * number is not correct.
1721 		 */
1722 		if (journal->j_head < first || journal->j_head >= last) {
1723 			printk(KERN_WARNING "JBD2: Incorrect Journal head block %lu, "
1724 			       "disable journal_cycle_record\n",
1725 			       journal->j_head);
1726 			journal->j_head = journal->j_first;
1727 		}
1728 	} else {
1729 		journal->j_head = journal->j_first;
1730 	}
1731 	journal->j_tail = journal->j_head;
1732 	journal->j_free = journal->j_last - journal->j_first;
1733 
1734 	journal->j_tail_sequence = journal->j_transaction_sequence;
1735 	journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1736 	journal->j_commit_request = journal->j_commit_sequence;
1737 
1738 	journal->j_max_transaction_buffers = jbd2_journal_get_max_txn_bufs(journal);
1739 
1740 	/*
1741 	 * Now that journal recovery is done, turn fast commits off here. This
1742 	 * way, if fast commit was enabled before the crash but if now FS has
1743 	 * disabled it, we don't enable fast commits.
1744 	 */
1745 	jbd2_clear_feature_fast_commit(journal);
1746 
1747 	/*
1748 	 * As a special case, if the on-disk copy is already marked as needing
1749 	 * no recovery (s_start == 0), then we can safely defer the superblock
1750 	 * update until the next commit by setting JBD2_FLUSHED.  This avoids
1751 	 * attempting a write to a potential-readonly device.
1752 	 */
1753 	if (sb->s_start == 0) {
1754 		jbd2_debug(1, "JBD2: Skipping superblock update on recovered sb "
1755 			"(start %ld, seq %u, errno %d)\n",
1756 			journal->j_tail, journal->j_tail_sequence,
1757 			journal->j_errno);
1758 		journal->j_flags |= JBD2_FLUSHED;
1759 	} else {
1760 		/* Lock here to make assertions happy... */
1761 		mutex_lock_io(&journal->j_checkpoint_mutex);
1762 		/*
1763 		 * Update log tail information. We use REQ_FUA since new
1764 		 * transaction will start reusing journal space and so we
1765 		 * must make sure information about current log tail is on
1766 		 * disk before that.
1767 		 */
1768 		jbd2_journal_update_sb_log_tail(journal,
1769 						journal->j_tail_sequence,
1770 						journal->j_tail, REQ_FUA);
1771 		mutex_unlock(&journal->j_checkpoint_mutex);
1772 	}
1773 	return jbd2_journal_start_thread(journal);
1774 }
1775 
1776 /*
1777  * This function expects that the caller will have locked the journal
1778  * buffer head, and will return with it unlocked
1779  */
jbd2_write_superblock(journal_t * journal,blk_opf_t write_flags)1780 static int jbd2_write_superblock(journal_t *journal, blk_opf_t write_flags)
1781 {
1782 	struct buffer_head *bh = journal->j_sb_buffer;
1783 	journal_superblock_t *sb = journal->j_superblock;
1784 	int ret = 0;
1785 
1786 	/* Buffer got discarded which means block device got invalidated */
1787 	if (!buffer_mapped(bh)) {
1788 		unlock_buffer(bh);
1789 		return -EIO;
1790 	}
1791 
1792 	/*
1793 	 * Always set high priority flags to exempt from block layer's
1794 	 * QOS policies, e.g. writeback throttle.
1795 	 */
1796 	write_flags |= JBD2_JOURNAL_REQ_FLAGS;
1797 	if (!(journal->j_flags & JBD2_BARRIER))
1798 		write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1799 
1800 	trace_jbd2_write_superblock(journal, write_flags);
1801 
1802 	if (buffer_write_io_error(bh)) {
1803 		/*
1804 		 * Oh, dear.  A previous attempt to write the journal
1805 		 * superblock failed.  This could happen because the
1806 		 * USB device was yanked out.  Or it could happen to
1807 		 * be a transient write error and maybe the block will
1808 		 * be remapped.  Nothing we can do but to retry the
1809 		 * write and hope for the best.
1810 		 */
1811 		printk(KERN_ERR "JBD2: previous I/O error detected "
1812 		       "for journal superblock update for %s.\n",
1813 		       journal->j_devname);
1814 		clear_buffer_write_io_error(bh);
1815 		set_buffer_uptodate(bh);
1816 	}
1817 	if (jbd2_journal_has_csum_v2or3(journal))
1818 		sb->s_checksum = jbd2_superblock_csum(journal, sb);
1819 	get_bh(bh);
1820 	bh->b_end_io = end_buffer_write_sync;
1821 	submit_bh(REQ_OP_WRITE | write_flags, bh);
1822 	wait_on_buffer(bh);
1823 	if (buffer_write_io_error(bh)) {
1824 		clear_buffer_write_io_error(bh);
1825 		set_buffer_uptodate(bh);
1826 		ret = -EIO;
1827 	}
1828 	if (ret) {
1829 		printk(KERN_ERR "JBD2: I/O error when updating journal superblock for %s.\n",
1830 				journal->j_devname);
1831 		if (!is_journal_aborted(journal))
1832 			jbd2_journal_abort(journal, ret);
1833 	}
1834 
1835 	return ret;
1836 }
1837 
1838 /**
1839  * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1840  * @journal: The journal to update.
1841  * @tail_tid: TID of the new transaction at the tail of the log
1842  * @tail_block: The first block of the transaction at the tail of the log
1843  * @write_flags: Flags for the journal sb write operation
1844  *
1845  * Update a journal's superblock information about log tail and write it to
1846  * disk, waiting for the IO to complete.
1847  */
jbd2_journal_update_sb_log_tail(journal_t * journal,tid_t tail_tid,unsigned long tail_block,blk_opf_t write_flags)1848 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1849 				    unsigned long tail_block,
1850 				    blk_opf_t write_flags)
1851 {
1852 	journal_superblock_t *sb = journal->j_superblock;
1853 	int ret;
1854 
1855 	if (is_journal_aborted(journal))
1856 		return -EIO;
1857 	if (test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags)) {
1858 		jbd2_journal_abort(journal, -EIO);
1859 		return -EIO;
1860 	}
1861 
1862 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1863 	jbd2_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1864 		  tail_block, tail_tid);
1865 
1866 	lock_buffer(journal->j_sb_buffer);
1867 	sb->s_sequence = cpu_to_be32(tail_tid);
1868 	sb->s_start    = cpu_to_be32(tail_block);
1869 
1870 	ret = jbd2_write_superblock(journal, write_flags);
1871 	if (ret)
1872 		goto out;
1873 
1874 	/* Log is no longer empty */
1875 	write_lock(&journal->j_state_lock);
1876 	WARN_ON(!sb->s_sequence);
1877 	journal->j_flags &= ~JBD2_FLUSHED;
1878 	write_unlock(&journal->j_state_lock);
1879 
1880 out:
1881 	return ret;
1882 }
1883 
1884 /**
1885  * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1886  * @journal: The journal to update.
1887  * @write_flags: Flags for the journal sb write operation
1888  *
1889  * Update a journal's dynamic superblock fields to show that journal is empty.
1890  * Write updated superblock to disk waiting for IO to complete.
1891  */
jbd2_mark_journal_empty(journal_t * journal,blk_opf_t write_flags)1892 static void jbd2_mark_journal_empty(journal_t *journal, blk_opf_t write_flags)
1893 {
1894 	journal_superblock_t *sb = journal->j_superblock;
1895 	bool had_fast_commit = false;
1896 
1897 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1898 	lock_buffer(journal->j_sb_buffer);
1899 	if (sb->s_start == 0) {		/* Is it already empty? */
1900 		unlock_buffer(journal->j_sb_buffer);
1901 		return;
1902 	}
1903 
1904 	jbd2_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1905 		  journal->j_tail_sequence);
1906 
1907 	sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1908 	sb->s_start    = cpu_to_be32(0);
1909 	sb->s_head     = cpu_to_be32(journal->j_head);
1910 	if (jbd2_has_feature_fast_commit(journal)) {
1911 		/*
1912 		 * When journal is clean, no need to commit fast commit flag and
1913 		 * make file system incompatible with older kernels.
1914 		 */
1915 		jbd2_clear_feature_fast_commit(journal);
1916 		had_fast_commit = true;
1917 	}
1918 
1919 	jbd2_write_superblock(journal, write_flags);
1920 
1921 	if (had_fast_commit)
1922 		jbd2_set_feature_fast_commit(journal);
1923 
1924 	/* Log is no longer empty */
1925 	write_lock(&journal->j_state_lock);
1926 	journal->j_flags |= JBD2_FLUSHED;
1927 	write_unlock(&journal->j_state_lock);
1928 }
1929 
1930 /**
1931  * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock)
1932  * @journal: The journal to erase.
1933  * @flags: A discard/zeroout request is sent for each physically contigous
1934  *	region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or
1935  *	JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation
1936  *	to perform.
1937  *
1938  * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes
1939  * will be explicitly written if no hardware offload is available, see
1940  * blkdev_issue_zeroout for more details.
1941  */
__jbd2_journal_erase(journal_t * journal,unsigned int flags)1942 static int __jbd2_journal_erase(journal_t *journal, unsigned int flags)
1943 {
1944 	int err = 0;
1945 	unsigned long block, log_offset; /* logical */
1946 	unsigned long long phys_block, block_start, block_stop; /* physical */
1947 	loff_t byte_start, byte_stop, byte_count;
1948 
1949 	/* flags must be set to either discard or zeroout */
1950 	if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags ||
1951 			((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1952 			(flags & JBD2_JOURNAL_FLUSH_ZEROOUT)))
1953 		return -EINVAL;
1954 
1955 	if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1956 	    !bdev_max_discard_sectors(journal->j_dev))
1957 		return -EOPNOTSUPP;
1958 
1959 	/*
1960 	 * lookup block mapping and issue discard/zeroout for each
1961 	 * contiguous region
1962 	 */
1963 	log_offset = be32_to_cpu(journal->j_superblock->s_first);
1964 	block_start =  ~0ULL;
1965 	for (block = log_offset; block < journal->j_total_len; block++) {
1966 		err = jbd2_journal_bmap(journal, block, &phys_block);
1967 		if (err) {
1968 			pr_err("JBD2: bad block at offset %lu", block);
1969 			return err;
1970 		}
1971 
1972 		if (block_start == ~0ULL) {
1973 			block_start = phys_block;
1974 			block_stop = block_start - 1;
1975 		}
1976 
1977 		/*
1978 		 * last block not contiguous with current block,
1979 		 * process last contiguous region and return to this block on
1980 		 * next loop
1981 		 */
1982 		if (phys_block != block_stop + 1) {
1983 			block--;
1984 		} else {
1985 			block_stop++;
1986 			/*
1987 			 * if this isn't the last block of journal,
1988 			 * no need to process now because next block may also
1989 			 * be part of this contiguous region
1990 			 */
1991 			if (block != journal->j_total_len - 1)
1992 				continue;
1993 		}
1994 
1995 		/*
1996 		 * end of contiguous region or this is last block of journal,
1997 		 * take care of the region
1998 		 */
1999 		byte_start = block_start * journal->j_blocksize;
2000 		byte_stop = block_stop * journal->j_blocksize;
2001 		byte_count = (block_stop - block_start + 1) *
2002 				journal->j_blocksize;
2003 
2004 		truncate_inode_pages_range(journal->j_dev->bd_inode->i_mapping,
2005 				byte_start, byte_stop);
2006 
2007 		if (flags & JBD2_JOURNAL_FLUSH_DISCARD) {
2008 			err = blkdev_issue_discard(journal->j_dev,
2009 					byte_start >> SECTOR_SHIFT,
2010 					byte_count >> SECTOR_SHIFT,
2011 					GFP_NOFS);
2012 		} else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) {
2013 			err = blkdev_issue_zeroout(journal->j_dev,
2014 					byte_start >> SECTOR_SHIFT,
2015 					byte_count >> SECTOR_SHIFT,
2016 					GFP_NOFS, 0);
2017 		}
2018 
2019 		if (unlikely(err != 0)) {
2020 			pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu",
2021 					err, block_start, block_stop);
2022 			return err;
2023 		}
2024 
2025 		/* reset start and stop after processing a region */
2026 		block_start = ~0ULL;
2027 	}
2028 
2029 	return blkdev_issue_flush(journal->j_dev);
2030 }
2031 
2032 /**
2033  * jbd2_journal_update_sb_errno() - Update error in the journal.
2034  * @journal: The journal to update.
2035  *
2036  * Update a journal's errno.  Write updated superblock to disk waiting for IO
2037  * to complete.
2038  */
jbd2_journal_update_sb_errno(journal_t * journal)2039 void jbd2_journal_update_sb_errno(journal_t *journal)
2040 {
2041 	journal_superblock_t *sb = journal->j_superblock;
2042 	int errcode;
2043 
2044 	lock_buffer(journal->j_sb_buffer);
2045 	errcode = journal->j_errno;
2046 	if (errcode == -ESHUTDOWN)
2047 		errcode = 0;
2048 	jbd2_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
2049 	sb->s_errno    = cpu_to_be32(errcode);
2050 
2051 	jbd2_write_superblock(journal, REQ_FUA);
2052 }
2053 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
2054 
2055 /**
2056  * jbd2_journal_load() - Read journal from disk.
2057  * @journal: Journal to act on.
2058  *
2059  * Given a journal_t structure which tells us which disk blocks contain
2060  * a journal, read the journal from disk to initialise the in-memory
2061  * structures.
2062  */
jbd2_journal_load(journal_t * journal)2063 int jbd2_journal_load(journal_t *journal)
2064 {
2065 	int err;
2066 	journal_superblock_t *sb = journal->j_superblock;
2067 
2068 	/*
2069 	 * Create a slab for this blocksize
2070 	 */
2071 	err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
2072 	if (err)
2073 		return err;
2074 
2075 	/* Let the recovery code check whether it needs to recover any
2076 	 * data from the journal. */
2077 	err = jbd2_journal_recover(journal);
2078 	if (err) {
2079 		pr_warn("JBD2: journal recovery failed\n");
2080 		return err;
2081 	}
2082 
2083 	if (journal->j_failed_commit) {
2084 		printk(KERN_ERR "JBD2: journal transaction %u on %s "
2085 		       "is corrupt.\n", journal->j_failed_commit,
2086 		       journal->j_devname);
2087 		return -EFSCORRUPTED;
2088 	}
2089 	/*
2090 	 * clear JBD2_ABORT flag initialized in journal_init_common
2091 	 * here to update log tail information with the newest seq.
2092 	 */
2093 	journal->j_flags &= ~JBD2_ABORT;
2094 
2095 	/* OK, we've finished with the dynamic journal bits:
2096 	 * reinitialise the dynamic contents of the superblock in memory
2097 	 * and reset them on disk. */
2098 	err = journal_reset(journal);
2099 	if (err) {
2100 		pr_warn("JBD2: journal reset failed\n");
2101 		return err;
2102 	}
2103 
2104 	journal->j_flags |= JBD2_LOADED;
2105 	return 0;
2106 }
2107 
2108 /**
2109  * jbd2_journal_destroy() - Release a journal_t structure.
2110  * @journal: Journal to act on.
2111  *
2112  * Release a journal_t structure once it is no longer in use by the
2113  * journaled object.
2114  * Return <0 if we couldn't clean up the journal.
2115  */
jbd2_journal_destroy(journal_t * journal)2116 int jbd2_journal_destroy(journal_t *journal)
2117 {
2118 	int err = 0;
2119 
2120 	/* Wait for the commit thread to wake up and die. */
2121 	journal_kill_thread(journal);
2122 
2123 	/* Force a final log commit */
2124 	if (journal->j_running_transaction)
2125 		jbd2_journal_commit_transaction(journal);
2126 
2127 	/* Force any old transactions to disk */
2128 
2129 	/* Totally anal locking here... */
2130 	spin_lock(&journal->j_list_lock);
2131 	while (journal->j_checkpoint_transactions != NULL) {
2132 		spin_unlock(&journal->j_list_lock);
2133 		mutex_lock_io(&journal->j_checkpoint_mutex);
2134 		err = jbd2_log_do_checkpoint(journal);
2135 		mutex_unlock(&journal->j_checkpoint_mutex);
2136 		/*
2137 		 * If checkpointing failed, just free the buffers to avoid
2138 		 * looping forever
2139 		 */
2140 		if (err) {
2141 			jbd2_journal_destroy_checkpoint(journal);
2142 			spin_lock(&journal->j_list_lock);
2143 			break;
2144 		}
2145 		spin_lock(&journal->j_list_lock);
2146 	}
2147 
2148 	J_ASSERT(journal->j_running_transaction == NULL);
2149 	J_ASSERT(journal->j_committing_transaction == NULL);
2150 	J_ASSERT(journal->j_checkpoint_transactions == NULL);
2151 	spin_unlock(&journal->j_list_lock);
2152 
2153 	/*
2154 	 * OK, all checkpoint transactions have been checked, now check the
2155 	 * write out io error flag and abort the journal if some buffer failed
2156 	 * to write back to the original location, otherwise the filesystem
2157 	 * may become inconsistent.
2158 	 */
2159 	if (!is_journal_aborted(journal) &&
2160 	    test_bit(JBD2_CHECKPOINT_IO_ERROR, &journal->j_atomic_flags))
2161 		jbd2_journal_abort(journal, -EIO);
2162 
2163 	if (journal->j_sb_buffer) {
2164 		if (!is_journal_aborted(journal)) {
2165 			mutex_lock_io(&journal->j_checkpoint_mutex);
2166 
2167 			write_lock(&journal->j_state_lock);
2168 			journal->j_tail_sequence =
2169 				++journal->j_transaction_sequence;
2170 			write_unlock(&journal->j_state_lock);
2171 
2172 			jbd2_mark_journal_empty(journal, REQ_PREFLUSH | REQ_FUA);
2173 			mutex_unlock(&journal->j_checkpoint_mutex);
2174 		} else
2175 			err = -EIO;
2176 		brelse(journal->j_sb_buffer);
2177 	}
2178 
2179 	if (journal->j_shrinker.flags & SHRINKER_REGISTERED) {
2180 		percpu_counter_destroy(&journal->j_checkpoint_jh_count);
2181 		unregister_shrinker(&journal->j_shrinker);
2182 	}
2183 	if (journal->j_proc_entry)
2184 		jbd2_stats_proc_exit(journal);
2185 	iput(journal->j_inode);
2186 	if (journal->j_revoke)
2187 		jbd2_journal_destroy_revoke(journal);
2188 	if (journal->j_chksum_driver)
2189 		crypto_free_shash(journal->j_chksum_driver);
2190 	kfree(journal->j_fc_wbuf);
2191 	kfree(journal->j_wbuf);
2192 	kfree(journal);
2193 
2194 	return err;
2195 }
2196 
2197 
2198 /**
2199  * jbd2_journal_check_used_features() - Check if features specified are used.
2200  * @journal: Journal to check.
2201  * @compat: bitmask of compatible features
2202  * @ro: bitmask of features that force read-only mount
2203  * @incompat: bitmask of incompatible features
2204  *
2205  * Check whether the journal uses all of a given set of
2206  * features.  Return true (non-zero) if it does.
2207  **/
2208 
jbd2_journal_check_used_features(journal_t * journal,unsigned long compat,unsigned long ro,unsigned long incompat)2209 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
2210 				 unsigned long ro, unsigned long incompat)
2211 {
2212 	journal_superblock_t *sb;
2213 
2214 	if (!compat && !ro && !incompat)
2215 		return 1;
2216 	if (!jbd2_format_support_feature(journal))
2217 		return 0;
2218 
2219 	sb = journal->j_superblock;
2220 
2221 	if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
2222 	    ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
2223 	    ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
2224 		return 1;
2225 
2226 	return 0;
2227 }
2228 
2229 /**
2230  * jbd2_journal_check_available_features() - Check feature set in journalling layer
2231  * @journal: Journal to check.
2232  * @compat: bitmask of compatible features
2233  * @ro: bitmask of features that force read-only mount
2234  * @incompat: bitmask of incompatible features
2235  *
2236  * Check whether the journaling code supports the use of
2237  * all of a given set of features on this journal.  Return true
2238  * (non-zero) if it can. */
2239 
jbd2_journal_check_available_features(journal_t * journal,unsigned long compat,unsigned long ro,unsigned long incompat)2240 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
2241 				      unsigned long ro, unsigned long incompat)
2242 {
2243 	if (!compat && !ro && !incompat)
2244 		return 1;
2245 
2246 	if (!jbd2_format_support_feature(journal))
2247 		return 0;
2248 
2249 	if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
2250 	    (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
2251 	    (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
2252 		return 1;
2253 
2254 	return 0;
2255 }
2256 
2257 static int
jbd2_journal_initialize_fast_commit(journal_t * journal)2258 jbd2_journal_initialize_fast_commit(journal_t *journal)
2259 {
2260 	journal_superblock_t *sb = journal->j_superblock;
2261 	unsigned long long num_fc_blks;
2262 
2263 	num_fc_blks = jbd2_journal_get_num_fc_blks(sb);
2264 	if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS)
2265 		return -ENOSPC;
2266 
2267 	/* Are we called twice? */
2268 	WARN_ON(journal->j_fc_wbuf != NULL);
2269 	journal->j_fc_wbuf = kmalloc_array(num_fc_blks,
2270 				sizeof(struct buffer_head *), GFP_KERNEL);
2271 	if (!journal->j_fc_wbuf)
2272 		return -ENOMEM;
2273 
2274 	journal->j_fc_wbufsize = num_fc_blks;
2275 	journal->j_fc_last = journal->j_last;
2276 	journal->j_last = journal->j_fc_last - num_fc_blks;
2277 	journal->j_fc_first = journal->j_last + 1;
2278 	journal->j_fc_off = 0;
2279 	journal->j_free = journal->j_last - journal->j_first;
2280 	journal->j_max_transaction_buffers =
2281 		jbd2_journal_get_max_txn_bufs(journal);
2282 
2283 	return 0;
2284 }
2285 
2286 /**
2287  * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2288  * @journal: Journal to act on.
2289  * @compat: bitmask of compatible features
2290  * @ro: bitmask of features that force read-only mount
2291  * @incompat: bitmask of incompatible features
2292  *
2293  * Mark a given journal feature as present on the
2294  * superblock.  Returns true if the requested features could be set.
2295  *
2296  */
2297 
jbd2_journal_set_features(journal_t * journal,unsigned long compat,unsigned long ro,unsigned long incompat)2298 int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
2299 			  unsigned long ro, unsigned long incompat)
2300 {
2301 #define INCOMPAT_FEATURE_ON(f) \
2302 		((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2303 #define COMPAT_FEATURE_ON(f) \
2304 		((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2305 	journal_superblock_t *sb;
2306 
2307 	if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
2308 		return 1;
2309 
2310 	if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
2311 		return 0;
2312 
2313 	/* If enabling v2 checksums, turn on v3 instead */
2314 	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
2315 		incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
2316 		incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
2317 	}
2318 
2319 	/* Asking for checksumming v3 and v1?  Only give them v3. */
2320 	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
2321 	    compat & JBD2_FEATURE_COMPAT_CHECKSUM)
2322 		compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
2323 
2324 	jbd2_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2325 		  compat, ro, incompat);
2326 
2327 	sb = journal->j_superblock;
2328 
2329 	if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) {
2330 		if (jbd2_journal_initialize_fast_commit(journal)) {
2331 			pr_err("JBD2: Cannot enable fast commits.\n");
2332 			return 0;
2333 		}
2334 	}
2335 
2336 	/* Load the checksum driver if necessary */
2337 	if ((journal->j_chksum_driver == NULL) &&
2338 	    INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2339 		journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
2340 		if (IS_ERR(journal->j_chksum_driver)) {
2341 			printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
2342 			journal->j_chksum_driver = NULL;
2343 			return 0;
2344 		}
2345 		/* Precompute checksum seed for all metadata */
2346 		journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2347 						   sizeof(sb->s_uuid));
2348 	}
2349 
2350 	lock_buffer(journal->j_sb_buffer);
2351 
2352 	/* If enabling v3 checksums, update superblock */
2353 	if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2354 		sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
2355 		sb->s_feature_compat &=
2356 			~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
2357 	}
2358 
2359 	/* If enabling v1 checksums, downgrade superblock */
2360 	if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
2361 		sb->s_feature_incompat &=
2362 			~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
2363 				     JBD2_FEATURE_INCOMPAT_CSUM_V3);
2364 
2365 	sb->s_feature_compat    |= cpu_to_be32(compat);
2366 	sb->s_feature_ro_compat |= cpu_to_be32(ro);
2367 	sb->s_feature_incompat  |= cpu_to_be32(incompat);
2368 	unlock_buffer(journal->j_sb_buffer);
2369 	journal->j_revoke_records_per_block =
2370 				journal_revoke_records_per_block(journal);
2371 
2372 	return 1;
2373 #undef COMPAT_FEATURE_ON
2374 #undef INCOMPAT_FEATURE_ON
2375 }
2376 
2377 /*
2378  * jbd2_journal_clear_features() - Clear a given journal feature in the
2379  * 				    superblock
2380  * @journal: Journal to act on.
2381  * @compat: bitmask of compatible features
2382  * @ro: bitmask of features that force read-only mount
2383  * @incompat: bitmask of incompatible features
2384  *
2385  * Clear a given journal feature as present on the
2386  * superblock.
2387  */
jbd2_journal_clear_features(journal_t * journal,unsigned long compat,unsigned long ro,unsigned long incompat)2388 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
2389 				unsigned long ro, unsigned long incompat)
2390 {
2391 	journal_superblock_t *sb;
2392 
2393 	jbd2_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2394 		  compat, ro, incompat);
2395 
2396 	sb = journal->j_superblock;
2397 
2398 	sb->s_feature_compat    &= ~cpu_to_be32(compat);
2399 	sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
2400 	sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
2401 	journal->j_revoke_records_per_block =
2402 				journal_revoke_records_per_block(journal);
2403 }
2404 EXPORT_SYMBOL(jbd2_journal_clear_features);
2405 
2406 /**
2407  * jbd2_journal_flush() - Flush journal
2408  * @journal: Journal to act on.
2409  * @flags: optional operation on the journal blocks after the flush (see below)
2410  *
2411  * Flush all data for a given journal to disk and empty the journal.
2412  * Filesystems can use this when remounting readonly to ensure that
2413  * recovery does not need to happen on remount. Optionally, a discard or zeroout
2414  * can be issued on the journal blocks after flushing.
2415  *
2416  * flags:
2417  *	JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks
2418  *	JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks
2419  */
jbd2_journal_flush(journal_t * journal,unsigned int flags)2420 int jbd2_journal_flush(journal_t *journal, unsigned int flags)
2421 {
2422 	int err = 0;
2423 	transaction_t *transaction = NULL;
2424 
2425 	write_lock(&journal->j_state_lock);
2426 
2427 	/* Force everything buffered to the log... */
2428 	if (journal->j_running_transaction) {
2429 		transaction = journal->j_running_transaction;
2430 		__jbd2_log_start_commit(journal, transaction->t_tid);
2431 	} else if (journal->j_committing_transaction)
2432 		transaction = journal->j_committing_transaction;
2433 
2434 	/* Wait for the log commit to complete... */
2435 	if (transaction) {
2436 		tid_t tid = transaction->t_tid;
2437 
2438 		write_unlock(&journal->j_state_lock);
2439 		jbd2_log_wait_commit(journal, tid);
2440 	} else {
2441 		write_unlock(&journal->j_state_lock);
2442 	}
2443 
2444 	/* ...and flush everything in the log out to disk. */
2445 	spin_lock(&journal->j_list_lock);
2446 	while (!err && journal->j_checkpoint_transactions != NULL) {
2447 		spin_unlock(&journal->j_list_lock);
2448 		mutex_lock_io(&journal->j_checkpoint_mutex);
2449 		err = jbd2_log_do_checkpoint(journal);
2450 		mutex_unlock(&journal->j_checkpoint_mutex);
2451 		spin_lock(&journal->j_list_lock);
2452 	}
2453 	spin_unlock(&journal->j_list_lock);
2454 
2455 	if (is_journal_aborted(journal))
2456 		return -EIO;
2457 
2458 	mutex_lock_io(&journal->j_checkpoint_mutex);
2459 	if (!err) {
2460 		err = jbd2_cleanup_journal_tail(journal);
2461 		if (err < 0) {
2462 			mutex_unlock(&journal->j_checkpoint_mutex);
2463 			goto out;
2464 		}
2465 		err = 0;
2466 	}
2467 
2468 	/* Finally, mark the journal as really needing no recovery.
2469 	 * This sets s_start==0 in the underlying superblock, which is
2470 	 * the magic code for a fully-recovered superblock.  Any future
2471 	 * commits of data to the journal will restore the current
2472 	 * s_start value. */
2473 	jbd2_mark_journal_empty(journal, REQ_FUA);
2474 
2475 	if (flags)
2476 		err = __jbd2_journal_erase(journal, flags);
2477 
2478 	mutex_unlock(&journal->j_checkpoint_mutex);
2479 	write_lock(&journal->j_state_lock);
2480 	J_ASSERT(!journal->j_running_transaction);
2481 	J_ASSERT(!journal->j_committing_transaction);
2482 	J_ASSERT(!journal->j_checkpoint_transactions);
2483 	J_ASSERT(journal->j_head == journal->j_tail);
2484 	J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2485 	write_unlock(&journal->j_state_lock);
2486 out:
2487 	return err;
2488 }
2489 
2490 /**
2491  * jbd2_journal_wipe() - Wipe journal contents
2492  * @journal: Journal to act on.
2493  * @write: flag (see below)
2494  *
2495  * Wipe out all of the contents of a journal, safely.  This will produce
2496  * a warning if the journal contains any valid recovery information.
2497  * Must be called between journal_init_*() and jbd2_journal_load().
2498  *
2499  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2500  * we merely suppress recovery.
2501  */
2502 
jbd2_journal_wipe(journal_t * journal,int write)2503 int jbd2_journal_wipe(journal_t *journal, int write)
2504 {
2505 	int err;
2506 
2507 	J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2508 
2509 	if (!journal->j_tail)
2510 		return 0;
2511 
2512 	printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2513 		write ? "Clearing" : "Ignoring");
2514 
2515 	err = jbd2_journal_skip_recovery(journal);
2516 	if (write) {
2517 		/* Lock to make assertions happy... */
2518 		mutex_lock_io(&journal->j_checkpoint_mutex);
2519 		jbd2_mark_journal_empty(journal, REQ_FUA);
2520 		mutex_unlock(&journal->j_checkpoint_mutex);
2521 	}
2522 
2523 	return err;
2524 }
2525 
2526 /**
2527  * jbd2_journal_abort () - Shutdown the journal immediately.
2528  * @journal: the journal to shutdown.
2529  * @errno:   an error number to record in the journal indicating
2530  *           the reason for the shutdown.
2531  *
2532  * Perform a complete, immediate shutdown of the ENTIRE
2533  * journal (not of a single transaction).  This operation cannot be
2534  * undone without closing and reopening the journal.
2535  *
2536  * The jbd2_journal_abort function is intended to support higher level error
2537  * recovery mechanisms such as the ext2/ext3 remount-readonly error
2538  * mode.
2539  *
2540  * Journal abort has very specific semantics.  Any existing dirty,
2541  * unjournaled buffers in the main filesystem will still be written to
2542  * disk by bdflush, but the journaling mechanism will be suspended
2543  * immediately and no further transaction commits will be honoured.
2544  *
2545  * Any dirty, journaled buffers will be written back to disk without
2546  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2547  * filesystem, but we _do_ attempt to leave as much data as possible
2548  * behind for fsck to use for cleanup.
2549  *
2550  * Any attempt to get a new transaction handle on a journal which is in
2551  * ABORT state will just result in an -EROFS error return.  A
2552  * jbd2_journal_stop on an existing handle will return -EIO if we have
2553  * entered abort state during the update.
2554  *
2555  * Recursive transactions are not disturbed by journal abort until the
2556  * final jbd2_journal_stop, which will receive the -EIO error.
2557  *
2558  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2559  * which will be recorded (if possible) in the journal superblock.  This
2560  * allows a client to record failure conditions in the middle of a
2561  * transaction without having to complete the transaction to record the
2562  * failure to disk.  ext3_error, for example, now uses this
2563  * functionality.
2564  *
2565  */
2566 
jbd2_journal_abort(journal_t * journal,int errno)2567 void jbd2_journal_abort(journal_t *journal, int errno)
2568 {
2569 	transaction_t *transaction;
2570 
2571 	/*
2572 	 * Lock the aborting procedure until everything is done, this avoid
2573 	 * races between filesystem's error handling flow (e.g. ext4_abort()),
2574 	 * ensure panic after the error info is written into journal's
2575 	 * superblock.
2576 	 */
2577 	mutex_lock(&journal->j_abort_mutex);
2578 	/*
2579 	 * ESHUTDOWN always takes precedence because a file system check
2580 	 * caused by any other journal abort error is not required after
2581 	 * a shutdown triggered.
2582 	 */
2583 	write_lock(&journal->j_state_lock);
2584 	if (journal->j_flags & JBD2_ABORT) {
2585 		int old_errno = journal->j_errno;
2586 
2587 		write_unlock(&journal->j_state_lock);
2588 		if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2589 			journal->j_errno = errno;
2590 			jbd2_journal_update_sb_errno(journal);
2591 		}
2592 		mutex_unlock(&journal->j_abort_mutex);
2593 		return;
2594 	}
2595 
2596 	/*
2597 	 * Mark the abort as occurred and start current running transaction
2598 	 * to release all journaled buffer.
2599 	 */
2600 	pr_err("Aborting journal on device %s.\n", journal->j_devname);
2601 
2602 	journal->j_flags |= JBD2_ABORT;
2603 	journal->j_errno = errno;
2604 	transaction = journal->j_running_transaction;
2605 	if (transaction)
2606 		__jbd2_log_start_commit(journal, transaction->t_tid);
2607 	write_unlock(&journal->j_state_lock);
2608 
2609 	/*
2610 	 * Record errno to the journal super block, so that fsck and jbd2
2611 	 * layer could realise that a filesystem check is needed.
2612 	 */
2613 	jbd2_journal_update_sb_errno(journal);
2614 	mutex_unlock(&journal->j_abort_mutex);
2615 }
2616 
2617 /**
2618  * jbd2_journal_errno() - returns the journal's error state.
2619  * @journal: journal to examine.
2620  *
2621  * This is the errno number set with jbd2_journal_abort(), the last
2622  * time the journal was mounted - if the journal was stopped
2623  * without calling abort this will be 0.
2624  *
2625  * If the journal has been aborted on this mount time -EROFS will
2626  * be returned.
2627  */
jbd2_journal_errno(journal_t * journal)2628 int jbd2_journal_errno(journal_t *journal)
2629 {
2630 	int err;
2631 
2632 	read_lock(&journal->j_state_lock);
2633 	if (journal->j_flags & JBD2_ABORT)
2634 		err = -EROFS;
2635 	else
2636 		err = journal->j_errno;
2637 	read_unlock(&journal->j_state_lock);
2638 	return err;
2639 }
2640 
2641 /**
2642  * jbd2_journal_clear_err() - clears the journal's error state
2643  * @journal: journal to act on.
2644  *
2645  * An error must be cleared or acked to take a FS out of readonly
2646  * mode.
2647  */
jbd2_journal_clear_err(journal_t * journal)2648 int jbd2_journal_clear_err(journal_t *journal)
2649 {
2650 	int err = 0;
2651 
2652 	write_lock(&journal->j_state_lock);
2653 	if (journal->j_flags & JBD2_ABORT)
2654 		err = -EROFS;
2655 	else
2656 		journal->j_errno = 0;
2657 	write_unlock(&journal->j_state_lock);
2658 	return err;
2659 }
2660 
2661 /**
2662  * jbd2_journal_ack_err() - Ack journal err.
2663  * @journal: journal to act on.
2664  *
2665  * An error must be cleared or acked to take a FS out of readonly
2666  * mode.
2667  */
jbd2_journal_ack_err(journal_t * journal)2668 void jbd2_journal_ack_err(journal_t *journal)
2669 {
2670 	write_lock(&journal->j_state_lock);
2671 	if (journal->j_errno)
2672 		journal->j_flags |= JBD2_ACK_ERR;
2673 	write_unlock(&journal->j_state_lock);
2674 }
2675 
jbd2_journal_blocks_per_page(struct inode * inode)2676 int jbd2_journal_blocks_per_page(struct inode *inode)
2677 {
2678 	return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2679 }
2680 
2681 /*
2682  * helper functions to deal with 32 or 64bit block numbers.
2683  */
journal_tag_bytes(journal_t * journal)2684 size_t journal_tag_bytes(journal_t *journal)
2685 {
2686 	size_t sz;
2687 
2688 	if (jbd2_has_feature_csum3(journal))
2689 		return sizeof(journal_block_tag3_t);
2690 
2691 	sz = sizeof(journal_block_tag_t);
2692 
2693 	if (jbd2_has_feature_csum2(journal))
2694 		sz += sizeof(__u16);
2695 
2696 	if (jbd2_has_feature_64bit(journal))
2697 		return sz;
2698 	else
2699 		return sz - sizeof(__u32);
2700 }
2701 
2702 /*
2703  * JBD memory management
2704  *
2705  * These functions are used to allocate block-sized chunks of memory
2706  * used for making copies of buffer_head data.  Very often it will be
2707  * page-sized chunks of data, but sometimes it will be in
2708  * sub-page-size chunks.  (For example, 16k pages on Power systems
2709  * with a 4k block file system.)  For blocks smaller than a page, we
2710  * use a SLAB allocator.  There are slab caches for each block size,
2711  * which are allocated at mount time, if necessary, and we only free
2712  * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2713  * this reason we don't need to a mutex to protect access to
2714  * jbd2_slab[] allocating or releasing memory; only in
2715  * jbd2_journal_create_slab().
2716  */
2717 #define JBD2_MAX_SLABS 8
2718 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2719 
2720 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2721 	"jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2722 	"jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2723 };
2724 
2725 
jbd2_journal_destroy_slabs(void)2726 static void jbd2_journal_destroy_slabs(void)
2727 {
2728 	int i;
2729 
2730 	for (i = 0; i < JBD2_MAX_SLABS; i++) {
2731 		kmem_cache_destroy(jbd2_slab[i]);
2732 		jbd2_slab[i] = NULL;
2733 	}
2734 }
2735 
jbd2_journal_create_slab(size_t size)2736 static int jbd2_journal_create_slab(size_t size)
2737 {
2738 	static DEFINE_MUTEX(jbd2_slab_create_mutex);
2739 	int i = order_base_2(size) - 10;
2740 	size_t slab_size;
2741 
2742 	if (size == PAGE_SIZE)
2743 		return 0;
2744 
2745 	if (i >= JBD2_MAX_SLABS)
2746 		return -EINVAL;
2747 
2748 	if (unlikely(i < 0))
2749 		i = 0;
2750 	mutex_lock(&jbd2_slab_create_mutex);
2751 	if (jbd2_slab[i]) {
2752 		mutex_unlock(&jbd2_slab_create_mutex);
2753 		return 0;	/* Already created */
2754 	}
2755 
2756 	slab_size = 1 << (i+10);
2757 	jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2758 					 slab_size, 0, NULL);
2759 	mutex_unlock(&jbd2_slab_create_mutex);
2760 	if (!jbd2_slab[i]) {
2761 		printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2762 		return -ENOMEM;
2763 	}
2764 	return 0;
2765 }
2766 
get_slab(size_t size)2767 static struct kmem_cache *get_slab(size_t size)
2768 {
2769 	int i = order_base_2(size) - 10;
2770 
2771 	BUG_ON(i >= JBD2_MAX_SLABS);
2772 	if (unlikely(i < 0))
2773 		i = 0;
2774 	BUG_ON(jbd2_slab[i] == NULL);
2775 	return jbd2_slab[i];
2776 }
2777 
jbd2_alloc(size_t size,gfp_t flags)2778 void *jbd2_alloc(size_t size, gfp_t flags)
2779 {
2780 	void *ptr;
2781 
2782 	BUG_ON(size & (size-1)); /* Must be a power of 2 */
2783 
2784 	if (size < PAGE_SIZE)
2785 		ptr = kmem_cache_alloc(get_slab(size), flags);
2786 	else
2787 		ptr = (void *)__get_free_pages(flags, get_order(size));
2788 
2789 	/* Check alignment; SLUB has gotten this wrong in the past,
2790 	 * and this can lead to user data corruption! */
2791 	BUG_ON(((unsigned long) ptr) & (size-1));
2792 
2793 	return ptr;
2794 }
2795 
jbd2_free(void * ptr,size_t size)2796 void jbd2_free(void *ptr, size_t size)
2797 {
2798 	if (size < PAGE_SIZE)
2799 		kmem_cache_free(get_slab(size), ptr);
2800 	else
2801 		free_pages((unsigned long)ptr, get_order(size));
2802 };
2803 
2804 /*
2805  * Journal_head storage management
2806  */
2807 static struct kmem_cache *jbd2_journal_head_cache;
2808 #ifdef CONFIG_JBD2_DEBUG
2809 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2810 #endif
2811 
jbd2_journal_init_journal_head_cache(void)2812 static int __init jbd2_journal_init_journal_head_cache(void)
2813 {
2814 	J_ASSERT(!jbd2_journal_head_cache);
2815 	jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2816 				sizeof(struct journal_head),
2817 				0,		/* offset */
2818 				SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2819 				NULL);		/* ctor */
2820 	if (!jbd2_journal_head_cache) {
2821 		printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2822 		return -ENOMEM;
2823 	}
2824 	return 0;
2825 }
2826 
jbd2_journal_destroy_journal_head_cache(void)2827 static void jbd2_journal_destroy_journal_head_cache(void)
2828 {
2829 	kmem_cache_destroy(jbd2_journal_head_cache);
2830 	jbd2_journal_head_cache = NULL;
2831 }
2832 
2833 /*
2834  * journal_head splicing and dicing
2835  */
journal_alloc_journal_head(void)2836 static struct journal_head *journal_alloc_journal_head(void)
2837 {
2838 	struct journal_head *ret;
2839 
2840 #ifdef CONFIG_JBD2_DEBUG
2841 	atomic_inc(&nr_journal_heads);
2842 #endif
2843 	ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2844 	if (!ret) {
2845 		jbd2_debug(1, "out of memory for journal_head\n");
2846 		pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2847 		ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2848 				GFP_NOFS | __GFP_NOFAIL);
2849 	}
2850 	if (ret)
2851 		spin_lock_init(&ret->b_state_lock);
2852 	return ret;
2853 }
2854 
journal_free_journal_head(struct journal_head * jh)2855 static void journal_free_journal_head(struct journal_head *jh)
2856 {
2857 #ifdef CONFIG_JBD2_DEBUG
2858 	atomic_dec(&nr_journal_heads);
2859 	memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2860 #endif
2861 	kmem_cache_free(jbd2_journal_head_cache, jh);
2862 }
2863 
2864 /*
2865  * A journal_head is attached to a buffer_head whenever JBD has an
2866  * interest in the buffer.
2867  *
2868  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2869  * is set.  This bit is tested in core kernel code where we need to take
2870  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2871  * there.
2872  *
2873  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2874  *
2875  * When a buffer has its BH_JBD bit set it is immune from being released by
2876  * core kernel code, mainly via ->b_count.
2877  *
2878  * A journal_head is detached from its buffer_head when the journal_head's
2879  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2880  * transaction (b_cp_transaction) hold their references to b_jcount.
2881  *
2882  * Various places in the kernel want to attach a journal_head to a buffer_head
2883  * _before_ attaching the journal_head to a transaction.  To protect the
2884  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2885  * journal_head's b_jcount refcount by one.  The caller must call
2886  * jbd2_journal_put_journal_head() to undo this.
2887  *
2888  * So the typical usage would be:
2889  *
2890  *	(Attach a journal_head if needed.  Increments b_jcount)
2891  *	struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2892  *	...
2893  *      (Get another reference for transaction)
2894  *	jbd2_journal_grab_journal_head(bh);
2895  *	jh->b_transaction = xxx;
2896  *	(Put original reference)
2897  *	jbd2_journal_put_journal_head(jh);
2898  */
2899 
2900 /*
2901  * Give a buffer_head a journal_head.
2902  *
2903  * May sleep.
2904  */
jbd2_journal_add_journal_head(struct buffer_head * bh)2905 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2906 {
2907 	struct journal_head *jh;
2908 	struct journal_head *new_jh = NULL;
2909 
2910 repeat:
2911 	if (!buffer_jbd(bh))
2912 		new_jh = journal_alloc_journal_head();
2913 
2914 	jbd_lock_bh_journal_head(bh);
2915 	if (buffer_jbd(bh)) {
2916 		jh = bh2jh(bh);
2917 	} else {
2918 		J_ASSERT_BH(bh,
2919 			(atomic_read(&bh->b_count) > 0) ||
2920 			(bh->b_folio && bh->b_folio->mapping));
2921 
2922 		if (!new_jh) {
2923 			jbd_unlock_bh_journal_head(bh);
2924 			goto repeat;
2925 		}
2926 
2927 		jh = new_jh;
2928 		new_jh = NULL;		/* We consumed it */
2929 		set_buffer_jbd(bh);
2930 		bh->b_private = jh;
2931 		jh->b_bh = bh;
2932 		get_bh(bh);
2933 		BUFFER_TRACE(bh, "added journal_head");
2934 	}
2935 	jh->b_jcount++;
2936 	jbd_unlock_bh_journal_head(bh);
2937 	if (new_jh)
2938 		journal_free_journal_head(new_jh);
2939 	return bh->b_private;
2940 }
2941 
2942 /*
2943  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2944  * having a journal_head, return NULL
2945  */
jbd2_journal_grab_journal_head(struct buffer_head * bh)2946 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2947 {
2948 	struct journal_head *jh = NULL;
2949 
2950 	jbd_lock_bh_journal_head(bh);
2951 	if (buffer_jbd(bh)) {
2952 		jh = bh2jh(bh);
2953 		jh->b_jcount++;
2954 	}
2955 	jbd_unlock_bh_journal_head(bh);
2956 	return jh;
2957 }
2958 EXPORT_SYMBOL(jbd2_journal_grab_journal_head);
2959 
__journal_remove_journal_head(struct buffer_head * bh)2960 static void __journal_remove_journal_head(struct buffer_head *bh)
2961 {
2962 	struct journal_head *jh = bh2jh(bh);
2963 
2964 	J_ASSERT_JH(jh, jh->b_transaction == NULL);
2965 	J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2966 	J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2967 	J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2968 	J_ASSERT_BH(bh, buffer_jbd(bh));
2969 	J_ASSERT_BH(bh, jh2bh(jh) == bh);
2970 	BUFFER_TRACE(bh, "remove journal_head");
2971 
2972 	/* Unlink before dropping the lock */
2973 	bh->b_private = NULL;
2974 	jh->b_bh = NULL;	/* debug, really */
2975 	clear_buffer_jbd(bh);
2976 }
2977 
journal_release_journal_head(struct journal_head * jh,size_t b_size)2978 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2979 {
2980 	if (jh->b_frozen_data) {
2981 		printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2982 		jbd2_free(jh->b_frozen_data, b_size);
2983 	}
2984 	if (jh->b_committed_data) {
2985 		printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2986 		jbd2_free(jh->b_committed_data, b_size);
2987 	}
2988 	journal_free_journal_head(jh);
2989 }
2990 
2991 /*
2992  * Drop a reference on the passed journal_head.  If it fell to zero then
2993  * release the journal_head from the buffer_head.
2994  */
jbd2_journal_put_journal_head(struct journal_head * jh)2995 void jbd2_journal_put_journal_head(struct journal_head *jh)
2996 {
2997 	struct buffer_head *bh = jh2bh(jh);
2998 
2999 	jbd_lock_bh_journal_head(bh);
3000 	J_ASSERT_JH(jh, jh->b_jcount > 0);
3001 	--jh->b_jcount;
3002 	if (!jh->b_jcount) {
3003 		__journal_remove_journal_head(bh);
3004 		jbd_unlock_bh_journal_head(bh);
3005 		journal_release_journal_head(jh, bh->b_size);
3006 		__brelse(bh);
3007 	} else {
3008 		jbd_unlock_bh_journal_head(bh);
3009 	}
3010 }
3011 EXPORT_SYMBOL(jbd2_journal_put_journal_head);
3012 
3013 /*
3014  * Initialize jbd inode head
3015  */
jbd2_journal_init_jbd_inode(struct jbd2_inode * jinode,struct inode * inode)3016 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
3017 {
3018 	jinode->i_transaction = NULL;
3019 	jinode->i_next_transaction = NULL;
3020 	jinode->i_vfs_inode = inode;
3021 	jinode->i_flags = 0;
3022 	jinode->i_dirty_start = 0;
3023 	jinode->i_dirty_end = 0;
3024 	INIT_LIST_HEAD(&jinode->i_list);
3025 }
3026 
3027 /*
3028  * Function to be called before we start removing inode from memory (i.e.,
3029  * clear_inode() is a fine place to be called from). It removes inode from
3030  * transaction's lists.
3031  */
jbd2_journal_release_jbd_inode(journal_t * journal,struct jbd2_inode * jinode)3032 void jbd2_journal_release_jbd_inode(journal_t *journal,
3033 				    struct jbd2_inode *jinode)
3034 {
3035 	if (!journal)
3036 		return;
3037 restart:
3038 	spin_lock(&journal->j_list_lock);
3039 	/* Is commit writing out inode - we have to wait */
3040 	if (jinode->i_flags & JI_COMMIT_RUNNING) {
3041 		wait_queue_head_t *wq;
3042 		DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
3043 		wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
3044 		prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
3045 		spin_unlock(&journal->j_list_lock);
3046 		schedule();
3047 		finish_wait(wq, &wait.wq_entry);
3048 		goto restart;
3049 	}
3050 
3051 	if (jinode->i_transaction) {
3052 		list_del(&jinode->i_list);
3053 		jinode->i_transaction = NULL;
3054 	}
3055 	spin_unlock(&journal->j_list_lock);
3056 }
3057 
3058 
3059 #ifdef CONFIG_PROC_FS
3060 
3061 #define JBD2_STATS_PROC_NAME "fs/jbd2"
3062 
jbd2_create_jbd_stats_proc_entry(void)3063 static void __init jbd2_create_jbd_stats_proc_entry(void)
3064 {
3065 	proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
3066 }
3067 
jbd2_remove_jbd_stats_proc_entry(void)3068 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
3069 {
3070 	if (proc_jbd2_stats)
3071 		remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
3072 }
3073 
3074 #else
3075 
3076 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
3077 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
3078 
3079 #endif
3080 
3081 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
3082 
jbd2_journal_init_inode_cache(void)3083 static int __init jbd2_journal_init_inode_cache(void)
3084 {
3085 	J_ASSERT(!jbd2_inode_cache);
3086 	jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
3087 	if (!jbd2_inode_cache) {
3088 		pr_emerg("JBD2: failed to create inode cache\n");
3089 		return -ENOMEM;
3090 	}
3091 	return 0;
3092 }
3093 
jbd2_journal_init_handle_cache(void)3094 static int __init jbd2_journal_init_handle_cache(void)
3095 {
3096 	J_ASSERT(!jbd2_handle_cache);
3097 	jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
3098 	if (!jbd2_handle_cache) {
3099 		printk(KERN_EMERG "JBD2: failed to create handle cache\n");
3100 		return -ENOMEM;
3101 	}
3102 	return 0;
3103 }
3104 
jbd2_journal_destroy_inode_cache(void)3105 static void jbd2_journal_destroy_inode_cache(void)
3106 {
3107 	kmem_cache_destroy(jbd2_inode_cache);
3108 	jbd2_inode_cache = NULL;
3109 }
3110 
jbd2_journal_destroy_handle_cache(void)3111 static void jbd2_journal_destroy_handle_cache(void)
3112 {
3113 	kmem_cache_destroy(jbd2_handle_cache);
3114 	jbd2_handle_cache = NULL;
3115 }
3116 
3117 /*
3118  * Module startup and shutdown
3119  */
3120 
journal_init_caches(void)3121 static int __init journal_init_caches(void)
3122 {
3123 	int ret;
3124 
3125 	ret = jbd2_journal_init_revoke_record_cache();
3126 	if (ret == 0)
3127 		ret = jbd2_journal_init_revoke_table_cache();
3128 	if (ret == 0)
3129 		ret = jbd2_journal_init_journal_head_cache();
3130 	if (ret == 0)
3131 		ret = jbd2_journal_init_handle_cache();
3132 	if (ret == 0)
3133 		ret = jbd2_journal_init_inode_cache();
3134 	if (ret == 0)
3135 		ret = jbd2_journal_init_transaction_cache();
3136 	return ret;
3137 }
3138 
jbd2_journal_destroy_caches(void)3139 static void jbd2_journal_destroy_caches(void)
3140 {
3141 	jbd2_journal_destroy_revoke_record_cache();
3142 	jbd2_journal_destroy_revoke_table_cache();
3143 	jbd2_journal_destroy_journal_head_cache();
3144 	jbd2_journal_destroy_handle_cache();
3145 	jbd2_journal_destroy_inode_cache();
3146 	jbd2_journal_destroy_transaction_cache();
3147 	jbd2_journal_destroy_slabs();
3148 }
3149 
journal_init(void)3150 static int __init journal_init(void)
3151 {
3152 	int ret;
3153 
3154 	BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
3155 
3156 	ret = journal_init_caches();
3157 	if (ret == 0) {
3158 		jbd2_create_jbd_stats_proc_entry();
3159 	} else {
3160 		jbd2_journal_destroy_caches();
3161 	}
3162 	return ret;
3163 }
3164 
journal_exit(void)3165 static void __exit journal_exit(void)
3166 {
3167 #ifdef CONFIG_JBD2_DEBUG
3168 	int n = atomic_read(&nr_journal_heads);
3169 	if (n)
3170 		printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
3171 #endif
3172 	jbd2_remove_jbd_stats_proc_entry();
3173 	jbd2_journal_destroy_caches();
3174 }
3175 
3176 MODULE_LICENSE("GPL");
3177 module_init(journal_init);
3178 module_exit(journal_exit);
3179 
3180