xref: /linux/fs/jbd2/journal.c (revision 6140ceb9)
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 
224 		prepare_to_wait(&journal->j_wait_commit, &wait,
225 				TASK_INTERRUPTIBLE);
226 		transaction = journal->j_running_transaction;
227 		if (transaction == NULL ||
228 		    time_before(jiffies, transaction->t_expires)) {
229 			write_unlock(&journal->j_state_lock);
230 			schedule();
231 			write_lock(&journal->j_state_lock);
232 		}
233 		finish_wait(&journal->j_wait_commit, &wait);
234 	}
235 
236 	jbd2_debug(1, "kjournald2 wakes\n");
237 
238 	/*
239 	 * Were we woken up by a commit wakeup event?
240 	 */
241 	transaction = journal->j_running_transaction;
242 	if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
243 		journal->j_commit_request = transaction->t_tid;
244 		jbd2_debug(1, "woke because of timeout\n");
245 	}
246 	goto loop;
247 
248 end_loop:
249 	del_timer_sync(&journal->j_commit_timer);
250 	journal->j_task = NULL;
251 	wake_up(&journal->j_wait_done_commit);
252 	jbd2_debug(1, "Journal thread exiting.\n");
253 	write_unlock(&journal->j_state_lock);
254 	return 0;
255 }
256 
jbd2_journal_start_thread(journal_t * journal)257 static int jbd2_journal_start_thread(journal_t *journal)
258 {
259 	struct task_struct *t;
260 
261 	t = kthread_run(kjournald2, journal, "jbd2/%s",
262 			journal->j_devname);
263 	if (IS_ERR(t))
264 		return PTR_ERR(t);
265 
266 	wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
267 	return 0;
268 }
269 
journal_kill_thread(journal_t * journal)270 static void journal_kill_thread(journal_t *journal)
271 {
272 	write_lock(&journal->j_state_lock);
273 	journal->j_flags |= JBD2_UNMOUNT;
274 
275 	while (journal->j_task) {
276 		write_unlock(&journal->j_state_lock);
277 		wake_up(&journal->j_wait_commit);
278 		wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
279 		write_lock(&journal->j_state_lock);
280 	}
281 	write_unlock(&journal->j_state_lock);
282 }
283 
jbd2_data_needs_escaping(char * data)284 static inline bool jbd2_data_needs_escaping(char *data)
285 {
286 	return *((__be32 *)data) == cpu_to_be32(JBD2_MAGIC_NUMBER);
287 }
288 
jbd2_data_do_escape(char * data)289 static inline void jbd2_data_do_escape(char *data)
290 {
291 	*((unsigned int *)data) = 0;
292 }
293 
294 /*
295  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
296  *
297  * Writes a metadata buffer to a given disk block.  The actual IO is not
298  * performed but a new buffer_head is constructed which labels the data
299  * to be written with the correct destination disk block.
300  *
301  * Any magic-number escaping which needs to be done will cause a
302  * copy-out here.  If the buffer happens to start with the
303  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
304  * magic number is only written to the log for descripter blocks.  In
305  * this case, we copy the data and replace the first word with 0, and we
306  * return a result code which indicates that this buffer needs to be
307  * marked as an escaped buffer in the corresponding log descriptor
308  * block.  The missing word can then be restored when the block is read
309  * during recovery.
310  *
311  * If the source buffer has already been modified by a new transaction
312  * since we took the last commit snapshot, we use the frozen copy of
313  * that data for IO. If we end up using the existing buffer_head's data
314  * for the write, then we have to make sure nobody modifies it while the
315  * IO is in progress. do_get_write_access() handles this.
316  *
317  * The function returns a pointer to the buffer_head to be used for IO.
318  *
319  *
320  * Return value:
321  *  <0: Error
322  *  =0: Finished OK without escape
323  *  =1: Finished OK with escape
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 do_escape = 0;
332 	struct buffer_head *new_bh;
333 	struct folio *new_folio;
334 	unsigned int new_offset;
335 	struct buffer_head *bh_in = jh2bh(jh_in);
336 	journal_t *journal = transaction->t_journal;
337 
338 	/*
339 	 * The buffer really shouldn't be locked: only the current committing
340 	 * transaction is allowed to write it, so nobody else is allowed
341 	 * to do any IO.
342 	 *
343 	 * akpm: except if we're journalling data, and write() output is
344 	 * also part of a shared mapping, and another thread has
345 	 * decided to launch a writepage() against this buffer.
346 	 */
347 	J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
348 
349 	new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
350 
351 	/* keep subsequent assertions sane */
352 	atomic_set(&new_bh->b_count, 1);
353 
354 	spin_lock(&jh_in->b_state_lock);
355 	/*
356 	 * If a new transaction has already done a buffer copy-out, then
357 	 * we use that version of the data for the commit.
358 	 */
359 	if (jh_in->b_frozen_data) {
360 		new_folio = virt_to_folio(jh_in->b_frozen_data);
361 		new_offset = offset_in_folio(new_folio, jh_in->b_frozen_data);
362 		do_escape = jbd2_data_needs_escaping(jh_in->b_frozen_data);
363 		if (do_escape)
364 			jbd2_data_do_escape(jh_in->b_frozen_data);
365 	} else {
366 		char *tmp;
367 		char *mapped_data;
368 
369 		new_folio = bh_in->b_folio;
370 		new_offset = offset_in_folio(new_folio, bh_in->b_data);
371 		mapped_data = kmap_local_folio(new_folio, new_offset);
372 		/*
373 		 * Fire data frozen trigger if data already wasn't frozen. Do
374 		 * this before checking for escaping, as the trigger may modify
375 		 * the magic offset.  If a copy-out happens afterwards, it will
376 		 * have the correct data in the buffer.
377 		 */
378 		jbd2_buffer_frozen_trigger(jh_in, mapped_data,
379 					   jh_in->b_triggers);
380 		do_escape = jbd2_data_needs_escaping(mapped_data);
381 		kunmap_local(mapped_data);
382 		/*
383 		 * Do we need to do a data copy?
384 		 */
385 		if (!do_escape)
386 			goto escape_done;
387 
388 		spin_unlock(&jh_in->b_state_lock);
389 		tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
390 		if (!tmp) {
391 			brelse(new_bh);
392 			free_buffer_head(new_bh);
393 			return -ENOMEM;
394 		}
395 		spin_lock(&jh_in->b_state_lock);
396 		if (jh_in->b_frozen_data) {
397 			jbd2_free(tmp, bh_in->b_size);
398 			goto copy_done;
399 		}
400 
401 		jh_in->b_frozen_data = tmp;
402 		memcpy_from_folio(tmp, new_folio, new_offset, bh_in->b_size);
403 		/*
404 		 * This isn't strictly necessary, as we're using frozen
405 		 * data for the escaping, but it keeps consistency with
406 		 * b_frozen_data usage.
407 		 */
408 		jh_in->b_frozen_triggers = jh_in->b_triggers;
409 
410 copy_done:
411 		new_folio = virt_to_folio(jh_in->b_frozen_data);
412 		new_offset = offset_in_folio(new_folio, jh_in->b_frozen_data);
413 		jbd2_data_do_escape(jh_in->b_frozen_data);
414 	}
415 
416 escape_done:
417 	folio_set_bh(new_bh, new_folio, new_offset);
418 	new_bh->b_size = bh_in->b_size;
419 	new_bh->b_bdev = journal->j_dev;
420 	new_bh->b_blocknr = blocknr;
421 	new_bh->b_private = bh_in;
422 	set_buffer_mapped(new_bh);
423 	set_buffer_dirty(new_bh);
424 
425 	*bh_out = new_bh;
426 
427 	/*
428 	 * The to-be-written buffer needs to get moved to the io queue,
429 	 * and the original buffer whose contents we are shadowing or
430 	 * copying is moved to the transaction's shadow queue.
431 	 */
432 	JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
433 	spin_lock(&journal->j_list_lock);
434 	__jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
435 	spin_unlock(&journal->j_list_lock);
436 	set_buffer_shadow(bh_in);
437 	spin_unlock(&jh_in->b_state_lock);
438 
439 	return do_escape;
440 }
441 
442 /*
443  * Allocation code for the journal file.  Manage the space left in the
444  * journal, so that we can begin checkpointing when appropriate.
445  */
446 
447 /*
448  * Called with j_state_lock locked for writing.
449  * Returns true if a transaction commit was started.
450  */
__jbd2_log_start_commit(journal_t * journal,tid_t target)451 static int __jbd2_log_start_commit(journal_t *journal, tid_t target)
452 {
453 	/* Return if the txn has already requested to be committed */
454 	if (journal->j_commit_request == target)
455 		return 0;
456 
457 	/*
458 	 * The only transaction we can possibly wait upon is the
459 	 * currently running transaction (if it exists).  Otherwise,
460 	 * the target tid must be an old one.
461 	 */
462 	if (journal->j_running_transaction &&
463 	    journal->j_running_transaction->t_tid == target) {
464 		/*
465 		 * We want a new commit: OK, mark the request and wakeup the
466 		 * commit thread.  We do _not_ do the commit ourselves.
467 		 */
468 
469 		journal->j_commit_request = target;
470 		jbd2_debug(1, "JBD2: requesting commit %u/%u\n",
471 			  journal->j_commit_request,
472 			  journal->j_commit_sequence);
473 		journal->j_running_transaction->t_requested = jiffies;
474 		wake_up(&journal->j_wait_commit);
475 		return 1;
476 	} else if (!tid_geq(journal->j_commit_request, target))
477 		/* This should never happen, but if it does, preserve
478 		   the evidence before kjournald goes into a loop and
479 		   increments j_commit_sequence beyond all recognition. */
480 		WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
481 			  journal->j_commit_request,
482 			  journal->j_commit_sequence,
483 			  target, journal->j_running_transaction ?
484 			  journal->j_running_transaction->t_tid : 0);
485 	return 0;
486 }
487 
jbd2_log_start_commit(journal_t * journal,tid_t tid)488 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
489 {
490 	int ret;
491 
492 	write_lock(&journal->j_state_lock);
493 	ret = __jbd2_log_start_commit(journal, tid);
494 	write_unlock(&journal->j_state_lock);
495 	return ret;
496 }
497 
498 /*
499  * Force and wait any uncommitted transactions.  We can only force the running
500  * transaction if we don't have an active handle, otherwise, we will deadlock.
501  * Returns: <0 in case of error,
502  *           0 if nothing to commit,
503  *           1 if transaction was successfully committed.
504  */
__jbd2_journal_force_commit(journal_t * journal)505 static int __jbd2_journal_force_commit(journal_t *journal)
506 {
507 	transaction_t *transaction = NULL;
508 	tid_t tid;
509 	int need_to_start = 0, ret = 0;
510 
511 	read_lock(&journal->j_state_lock);
512 	if (journal->j_running_transaction && !current->journal_info) {
513 		transaction = journal->j_running_transaction;
514 		if (!tid_geq(journal->j_commit_request, transaction->t_tid))
515 			need_to_start = 1;
516 	} else if (journal->j_committing_transaction)
517 		transaction = journal->j_committing_transaction;
518 
519 	if (!transaction) {
520 		/* Nothing to commit */
521 		read_unlock(&journal->j_state_lock);
522 		return 0;
523 	}
524 	tid = transaction->t_tid;
525 	read_unlock(&journal->j_state_lock);
526 	if (need_to_start)
527 		jbd2_log_start_commit(journal, tid);
528 	ret = jbd2_log_wait_commit(journal, tid);
529 	if (!ret)
530 		ret = 1;
531 
532 	return ret;
533 }
534 
535 /**
536  * jbd2_journal_force_commit_nested - Force and wait upon a commit if the
537  * calling process is not within transaction.
538  *
539  * @journal: journal to force
540  * Returns true if progress was made.
541  *
542  * This is used for forcing out undo-protected data which contains
543  * bitmaps, when the fs is running out of space.
544  */
jbd2_journal_force_commit_nested(journal_t * journal)545 int jbd2_journal_force_commit_nested(journal_t *journal)
546 {
547 	int ret;
548 
549 	ret = __jbd2_journal_force_commit(journal);
550 	return ret > 0;
551 }
552 
553 /**
554  * jbd2_journal_force_commit() - force any uncommitted transactions
555  * @journal: journal to force
556  *
557  * Caller want unconditional commit. We can only force the running transaction
558  * if we don't have an active handle, otherwise, we will deadlock.
559  */
jbd2_journal_force_commit(journal_t * journal)560 int jbd2_journal_force_commit(journal_t *journal)
561 {
562 	int ret;
563 
564 	J_ASSERT(!current->journal_info);
565 	ret = __jbd2_journal_force_commit(journal);
566 	if (ret > 0)
567 		ret = 0;
568 	return ret;
569 }
570 
571 /*
572  * Start a commit of the current running transaction (if any).  Returns true
573  * if a transaction is going to be committed (or is currently already
574  * committing), and fills its tid in at *ptid
575  */
jbd2_journal_start_commit(journal_t * journal,tid_t * ptid)576 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
577 {
578 	int ret = 0;
579 
580 	write_lock(&journal->j_state_lock);
581 	if (journal->j_running_transaction) {
582 		tid_t tid = journal->j_running_transaction->t_tid;
583 
584 		__jbd2_log_start_commit(journal, tid);
585 		/* There's a running transaction and we've just made sure
586 		 * it's commit has been scheduled. */
587 		if (ptid)
588 			*ptid = tid;
589 		ret = 1;
590 	} else if (journal->j_committing_transaction) {
591 		/*
592 		 * If commit has been started, then we have to wait for
593 		 * completion of that transaction.
594 		 */
595 		if (ptid)
596 			*ptid = journal->j_committing_transaction->t_tid;
597 		ret = 1;
598 	}
599 	write_unlock(&journal->j_state_lock);
600 	return ret;
601 }
602 
603 /*
604  * Return 1 if a given transaction has not yet sent barrier request
605  * connected with a transaction commit. If 0 is returned, transaction
606  * may or may not have sent the barrier. Used to avoid sending barrier
607  * twice in common cases.
608  */
jbd2_trans_will_send_data_barrier(journal_t * journal,tid_t tid)609 int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
610 {
611 	int ret = 0;
612 	transaction_t *commit_trans;
613 
614 	if (!(journal->j_flags & JBD2_BARRIER))
615 		return 0;
616 	read_lock(&journal->j_state_lock);
617 	/* Transaction already committed? */
618 	if (tid_geq(journal->j_commit_sequence, tid))
619 		goto out;
620 	commit_trans = journal->j_committing_transaction;
621 	if (!commit_trans || commit_trans->t_tid != tid) {
622 		ret = 1;
623 		goto out;
624 	}
625 	/*
626 	 * Transaction is being committed and we already proceeded to
627 	 * submitting a flush to fs partition?
628 	 */
629 	if (journal->j_fs_dev != journal->j_dev) {
630 		if (!commit_trans->t_need_data_flush ||
631 		    commit_trans->t_state >= T_COMMIT_DFLUSH)
632 			goto out;
633 	} else {
634 		if (commit_trans->t_state >= T_COMMIT_JFLUSH)
635 			goto out;
636 	}
637 	ret = 1;
638 out:
639 	read_unlock(&journal->j_state_lock);
640 	return ret;
641 }
642 EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
643 
644 /*
645  * Wait for a specified commit to complete.
646  * The caller may not hold the journal lock.
647  */
jbd2_log_wait_commit(journal_t * journal,tid_t tid)648 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
649 {
650 	int err = 0;
651 
652 	read_lock(&journal->j_state_lock);
653 #ifdef CONFIG_PROVE_LOCKING
654 	/*
655 	 * Some callers make sure transaction is already committing and in that
656 	 * case we cannot block on open handles anymore. So don't warn in that
657 	 * case.
658 	 */
659 	if (tid_gt(tid, journal->j_commit_sequence) &&
660 	    (!journal->j_committing_transaction ||
661 	     journal->j_committing_transaction->t_tid != tid)) {
662 		read_unlock(&journal->j_state_lock);
663 		jbd2_might_wait_for_commit(journal);
664 		read_lock(&journal->j_state_lock);
665 	}
666 #endif
667 #ifdef CONFIG_JBD2_DEBUG
668 	if (!tid_geq(journal->j_commit_request, tid)) {
669 		printk(KERN_ERR
670 		       "%s: error: j_commit_request=%u, tid=%u\n",
671 		       __func__, journal->j_commit_request, tid);
672 	}
673 #endif
674 	while (tid_gt(tid, journal->j_commit_sequence)) {
675 		jbd2_debug(1, "JBD2: want %u, j_commit_sequence=%u\n",
676 				  tid, journal->j_commit_sequence);
677 		read_unlock(&journal->j_state_lock);
678 		wake_up(&journal->j_wait_commit);
679 		wait_event(journal->j_wait_done_commit,
680 				!tid_gt(tid, journal->j_commit_sequence));
681 		read_lock(&journal->j_state_lock);
682 	}
683 	read_unlock(&journal->j_state_lock);
684 
685 	if (unlikely(is_journal_aborted(journal)))
686 		err = -EIO;
687 	return err;
688 }
689 
690 /*
691  * Start a fast commit. If there's an ongoing fast or full commit wait for
692  * it to complete. Returns 0 if a new fast commit was started. Returns -EALREADY
693  * if a fast commit is not needed, either because there's an already a commit
694  * going on or this tid has already been committed. Returns -EINVAL if no jbd2
695  * commit has yet been performed.
696  */
jbd2_fc_begin_commit(journal_t * journal,tid_t tid)697 int jbd2_fc_begin_commit(journal_t *journal, tid_t tid)
698 {
699 	if (unlikely(is_journal_aborted(journal)))
700 		return -EIO;
701 	/*
702 	 * Fast commits only allowed if at least one full commit has
703 	 * been processed.
704 	 */
705 	if (!journal->j_stats.ts_tid)
706 		return -EINVAL;
707 
708 	write_lock(&journal->j_state_lock);
709 	if (tid_geq(journal->j_commit_sequence, tid)) {
710 		write_unlock(&journal->j_state_lock);
711 		return -EALREADY;
712 	}
713 
714 	if (journal->j_flags & JBD2_FULL_COMMIT_ONGOING ||
715 	    (journal->j_flags & JBD2_FAST_COMMIT_ONGOING)) {
716 		DEFINE_WAIT(wait);
717 
718 		prepare_to_wait(&journal->j_fc_wait, &wait,
719 				TASK_UNINTERRUPTIBLE);
720 		write_unlock(&journal->j_state_lock);
721 		schedule();
722 		finish_wait(&journal->j_fc_wait, &wait);
723 		return -EALREADY;
724 	}
725 	journal->j_flags |= JBD2_FAST_COMMIT_ONGOING;
726 	write_unlock(&journal->j_state_lock);
727 	jbd2_journal_lock_updates(journal);
728 
729 	return 0;
730 }
731 EXPORT_SYMBOL(jbd2_fc_begin_commit);
732 
733 /*
734  * Stop a fast commit. If fallback is set, this function starts commit of
735  * TID tid before any other fast commit can start.
736  */
__jbd2_fc_end_commit(journal_t * journal,tid_t tid,bool fallback)737 static int __jbd2_fc_end_commit(journal_t *journal, tid_t tid, bool fallback)
738 {
739 	if (journal->j_fc_cleanup_callback)
740 		journal->j_fc_cleanup_callback(journal, 0, tid);
741 	jbd2_journal_unlock_updates(journal);
742 	write_lock(&journal->j_state_lock);
743 	journal->j_flags &= ~JBD2_FAST_COMMIT_ONGOING;
744 	if (fallback)
745 		journal->j_flags |= JBD2_FULL_COMMIT_ONGOING;
746 	write_unlock(&journal->j_state_lock);
747 	wake_up(&journal->j_fc_wait);
748 	if (fallback)
749 		return jbd2_complete_transaction(journal, tid);
750 	return 0;
751 }
752 
jbd2_fc_end_commit(journal_t * journal)753 int jbd2_fc_end_commit(journal_t *journal)
754 {
755 	return __jbd2_fc_end_commit(journal, 0, false);
756 }
757 EXPORT_SYMBOL(jbd2_fc_end_commit);
758 
jbd2_fc_end_commit_fallback(journal_t * journal)759 int jbd2_fc_end_commit_fallback(journal_t *journal)
760 {
761 	tid_t tid;
762 
763 	read_lock(&journal->j_state_lock);
764 	tid = journal->j_running_transaction ?
765 		journal->j_running_transaction->t_tid : 0;
766 	read_unlock(&journal->j_state_lock);
767 	return __jbd2_fc_end_commit(journal, tid, true);
768 }
769 EXPORT_SYMBOL(jbd2_fc_end_commit_fallback);
770 
771 /* Return 1 when transaction with given tid has already committed. */
jbd2_transaction_committed(journal_t * journal,tid_t tid)772 int jbd2_transaction_committed(journal_t *journal, tid_t tid)
773 {
774 	return tid_geq(READ_ONCE(journal->j_commit_sequence), tid);
775 }
776 EXPORT_SYMBOL(jbd2_transaction_committed);
777 
778 /*
779  * When this function returns the transaction corresponding to tid
780  * will be completed.  If the transaction has currently running, start
781  * committing that transaction before waiting for it to complete.  If
782  * the transaction id is stale, it is by definition already completed,
783  * so just return SUCCESS.
784  */
jbd2_complete_transaction(journal_t * journal,tid_t tid)785 int jbd2_complete_transaction(journal_t *journal, tid_t tid)
786 {
787 	int	need_to_wait = 1;
788 
789 	read_lock(&journal->j_state_lock);
790 	if (journal->j_running_transaction &&
791 	    journal->j_running_transaction->t_tid == tid) {
792 		if (journal->j_commit_request != tid) {
793 			/* transaction not yet started, so request it */
794 			read_unlock(&journal->j_state_lock);
795 			jbd2_log_start_commit(journal, tid);
796 			goto wait_commit;
797 		}
798 	} else if (!(journal->j_committing_transaction &&
799 		     journal->j_committing_transaction->t_tid == tid))
800 		need_to_wait = 0;
801 	read_unlock(&journal->j_state_lock);
802 	if (!need_to_wait)
803 		return 0;
804 wait_commit:
805 	return jbd2_log_wait_commit(journal, tid);
806 }
807 EXPORT_SYMBOL(jbd2_complete_transaction);
808 
809 /*
810  * Log buffer allocation routines:
811  */
812 
jbd2_journal_next_log_block(journal_t * journal,unsigned long long * retp)813 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
814 {
815 	unsigned long blocknr;
816 
817 	write_lock(&journal->j_state_lock);
818 	J_ASSERT(journal->j_free > 1);
819 
820 	blocknr = journal->j_head;
821 	journal->j_head++;
822 	journal->j_free--;
823 	if (journal->j_head == journal->j_last)
824 		journal->j_head = journal->j_first;
825 	write_unlock(&journal->j_state_lock);
826 	return jbd2_journal_bmap(journal, blocknr, retp);
827 }
828 
829 /* Map one fast commit buffer for use by the file system */
jbd2_fc_get_buf(journal_t * journal,struct buffer_head ** bh_out)830 int jbd2_fc_get_buf(journal_t *journal, struct buffer_head **bh_out)
831 {
832 	unsigned long long pblock;
833 	unsigned long blocknr;
834 	int ret = 0;
835 	struct buffer_head *bh;
836 	int fc_off;
837 
838 	*bh_out = NULL;
839 
840 	if (journal->j_fc_off + journal->j_fc_first >= journal->j_fc_last)
841 		return -EINVAL;
842 
843 	fc_off = journal->j_fc_off;
844 	blocknr = journal->j_fc_first + fc_off;
845 	journal->j_fc_off++;
846 	ret = jbd2_journal_bmap(journal, blocknr, &pblock);
847 	if (ret)
848 		return ret;
849 
850 	bh = __getblk(journal->j_dev, pblock, journal->j_blocksize);
851 	if (!bh)
852 		return -ENOMEM;
853 
854 	journal->j_fc_wbuf[fc_off] = bh;
855 
856 	*bh_out = bh;
857 
858 	return 0;
859 }
860 EXPORT_SYMBOL(jbd2_fc_get_buf);
861 
862 /*
863  * Wait on fast commit buffers that were allocated by jbd2_fc_get_buf
864  * for completion.
865  */
jbd2_fc_wait_bufs(journal_t * journal,int num_blks)866 int jbd2_fc_wait_bufs(journal_t *journal, int num_blks)
867 {
868 	struct buffer_head *bh;
869 	int i, j_fc_off;
870 
871 	j_fc_off = journal->j_fc_off;
872 
873 	/*
874 	 * Wait in reverse order to minimize chances of us being woken up before
875 	 * all IOs have completed
876 	 */
877 	for (i = j_fc_off - 1; i >= j_fc_off - num_blks; i--) {
878 		bh = journal->j_fc_wbuf[i];
879 		wait_on_buffer(bh);
880 		/*
881 		 * Update j_fc_off so jbd2_fc_release_bufs can release remain
882 		 * buffer head.
883 		 */
884 		if (unlikely(!buffer_uptodate(bh))) {
885 			journal->j_fc_off = i + 1;
886 			return -EIO;
887 		}
888 		put_bh(bh);
889 		journal->j_fc_wbuf[i] = NULL;
890 	}
891 
892 	return 0;
893 }
894 EXPORT_SYMBOL(jbd2_fc_wait_bufs);
895 
jbd2_fc_release_bufs(journal_t * journal)896 void jbd2_fc_release_bufs(journal_t *journal)
897 {
898 	struct buffer_head *bh;
899 	int i, j_fc_off;
900 
901 	j_fc_off = journal->j_fc_off;
902 
903 	for (i = j_fc_off - 1; i >= 0; i--) {
904 		bh = journal->j_fc_wbuf[i];
905 		if (!bh)
906 			break;
907 		put_bh(bh);
908 		journal->j_fc_wbuf[i] = NULL;
909 	}
910 }
911 EXPORT_SYMBOL(jbd2_fc_release_bufs);
912 
913 /*
914  * Conversion of logical to physical block numbers for the journal
915  *
916  * On external journals the journal blocks are identity-mapped, so
917  * this is a no-op.  If needed, we can use j_blk_offset - everything is
918  * ready.
919  */
jbd2_journal_bmap(journal_t * journal,unsigned long blocknr,unsigned long long * retp)920 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
921 		 unsigned long long *retp)
922 {
923 	int err = 0;
924 	unsigned long long ret;
925 	sector_t block = blocknr;
926 
927 	if (journal->j_bmap) {
928 		err = journal->j_bmap(journal, &block);
929 		if (err == 0)
930 			*retp = block;
931 	} else if (journal->j_inode) {
932 		ret = bmap(journal->j_inode, &block);
933 
934 		if (ret || !block) {
935 			printk(KERN_ALERT "%s: journal block not found "
936 					"at offset %lu on %s\n",
937 			       __func__, blocknr, journal->j_devname);
938 			err = -EIO;
939 			jbd2_journal_abort(journal, err);
940 		} else {
941 			*retp = block;
942 		}
943 
944 	} else {
945 		*retp = blocknr; /* +journal->j_blk_offset */
946 	}
947 	return err;
948 }
949 
950 /*
951  * We play buffer_head aliasing tricks to write data/metadata blocks to
952  * the journal without copying their contents, but for journal
953  * descriptor blocks we do need to generate bona fide buffers.
954  *
955  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
956  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
957  * But we don't bother doing that, so there will be coherency problems with
958  * mmaps of blockdevs which hold live JBD-controlled filesystems.
959  */
960 struct buffer_head *
jbd2_journal_get_descriptor_buffer(transaction_t * transaction,int type)961 jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type)
962 {
963 	journal_t *journal = transaction->t_journal;
964 	struct buffer_head *bh;
965 	unsigned long long blocknr;
966 	journal_header_t *header;
967 	int err;
968 
969 	err = jbd2_journal_next_log_block(journal, &blocknr);
970 
971 	if (err)
972 		return NULL;
973 
974 	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
975 	if (!bh)
976 		return NULL;
977 	atomic_dec(&transaction->t_outstanding_credits);
978 	lock_buffer(bh);
979 	memset(bh->b_data, 0, journal->j_blocksize);
980 	header = (journal_header_t *)bh->b_data;
981 	header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER);
982 	header->h_blocktype = cpu_to_be32(type);
983 	header->h_sequence = cpu_to_be32(transaction->t_tid);
984 	set_buffer_uptodate(bh);
985 	unlock_buffer(bh);
986 	BUFFER_TRACE(bh, "return this buffer");
987 	return bh;
988 }
989 
jbd2_descriptor_block_csum_set(journal_t * j,struct buffer_head * bh)990 void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh)
991 {
992 	struct jbd2_journal_block_tail *tail;
993 	__u32 csum;
994 
995 	if (!jbd2_journal_has_csum_v2or3(j))
996 		return;
997 
998 	tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize -
999 			sizeof(struct jbd2_journal_block_tail));
1000 	tail->t_checksum = 0;
1001 	csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize);
1002 	tail->t_checksum = cpu_to_be32(csum);
1003 }
1004 
1005 /*
1006  * Return tid of the oldest transaction in the journal and block in the journal
1007  * where the transaction starts.
1008  *
1009  * If the journal is now empty, return which will be the next transaction ID
1010  * we will write and where will that transaction start.
1011  *
1012  * The return value is 0 if journal tail cannot be pushed any further, 1 if
1013  * it can.
1014  */
jbd2_journal_get_log_tail(journal_t * journal,tid_t * tid,unsigned long * block)1015 int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid,
1016 			      unsigned long *block)
1017 {
1018 	transaction_t *transaction;
1019 	int ret;
1020 
1021 	read_lock(&journal->j_state_lock);
1022 	spin_lock(&journal->j_list_lock);
1023 	transaction = journal->j_checkpoint_transactions;
1024 	if (transaction) {
1025 		*tid = transaction->t_tid;
1026 		*block = transaction->t_log_start;
1027 	} else if ((transaction = journal->j_committing_transaction) != NULL) {
1028 		*tid = transaction->t_tid;
1029 		*block = transaction->t_log_start;
1030 	} else if ((transaction = journal->j_running_transaction) != NULL) {
1031 		*tid = transaction->t_tid;
1032 		*block = journal->j_head;
1033 	} else {
1034 		*tid = journal->j_transaction_sequence;
1035 		*block = journal->j_head;
1036 	}
1037 	ret = tid_gt(*tid, journal->j_tail_sequence);
1038 	spin_unlock(&journal->j_list_lock);
1039 	read_unlock(&journal->j_state_lock);
1040 
1041 	return ret;
1042 }
1043 
1044 /*
1045  * Update information in journal structure and in on disk journal superblock
1046  * about log tail. This function does not check whether information passed in
1047  * really pushes log tail further. It's responsibility of the caller to make
1048  * sure provided log tail information is valid (e.g. by holding
1049  * j_checkpoint_mutex all the time between computing log tail and calling this
1050  * function as is the case with jbd2_cleanup_journal_tail()).
1051  *
1052  * Requires j_checkpoint_mutex
1053  */
__jbd2_update_log_tail(journal_t * journal,tid_t tid,unsigned long block)1054 int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1055 {
1056 	unsigned long freed;
1057 	int ret;
1058 
1059 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1060 
1061 	/*
1062 	 * We cannot afford for write to remain in drive's caches since as
1063 	 * soon as we update j_tail, next transaction can start reusing journal
1064 	 * space and if we lose sb update during power failure we'd replay
1065 	 * old transaction with possibly newly overwritten data.
1066 	 */
1067 	ret = jbd2_journal_update_sb_log_tail(journal, tid, block, REQ_FUA);
1068 	if (ret)
1069 		goto out;
1070 
1071 	write_lock(&journal->j_state_lock);
1072 	freed = block - journal->j_tail;
1073 	if (block < journal->j_tail)
1074 		freed += journal->j_last - journal->j_first;
1075 
1076 	trace_jbd2_update_log_tail(journal, tid, block, freed);
1077 	jbd2_debug(1,
1078 		  "Cleaning journal tail from %u to %u (offset %lu), "
1079 		  "freeing %lu\n",
1080 		  journal->j_tail_sequence, tid, block, freed);
1081 
1082 	journal->j_free += freed;
1083 	journal->j_tail_sequence = tid;
1084 	journal->j_tail = block;
1085 	write_unlock(&journal->j_state_lock);
1086 
1087 out:
1088 	return ret;
1089 }
1090 
1091 /*
1092  * This is a variation of __jbd2_update_log_tail which checks for validity of
1093  * provided log tail and locks j_checkpoint_mutex. So it is safe against races
1094  * with other threads updating log tail.
1095  */
jbd2_update_log_tail(journal_t * journal,tid_t tid,unsigned long block)1096 void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block)
1097 {
1098 	mutex_lock_io(&journal->j_checkpoint_mutex);
1099 	if (tid_gt(tid, journal->j_tail_sequence))
1100 		__jbd2_update_log_tail(journal, tid, block);
1101 	mutex_unlock(&journal->j_checkpoint_mutex);
1102 }
1103 
1104 struct jbd2_stats_proc_session {
1105 	journal_t *journal;
1106 	struct transaction_stats_s *stats;
1107 	int start;
1108 	int max;
1109 };
1110 
jbd2_seq_info_start(struct seq_file * seq,loff_t * pos)1111 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
1112 {
1113 	return *pos ? NULL : SEQ_START_TOKEN;
1114 }
1115 
jbd2_seq_info_next(struct seq_file * seq,void * v,loff_t * pos)1116 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
1117 {
1118 	(*pos)++;
1119 	return NULL;
1120 }
1121 
jbd2_seq_info_show(struct seq_file * seq,void * v)1122 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
1123 {
1124 	struct jbd2_stats_proc_session *s = seq->private;
1125 
1126 	if (v != SEQ_START_TOKEN)
1127 		return 0;
1128 	seq_printf(seq, "%lu transactions (%lu requested), "
1129 		   "each up to %u blocks\n",
1130 		   s->stats->ts_tid, s->stats->ts_requested,
1131 		   s->journal->j_max_transaction_buffers);
1132 	if (s->stats->ts_tid == 0)
1133 		return 0;
1134 	seq_printf(seq, "average: \n  %ums waiting for transaction\n",
1135 	    jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
1136 	seq_printf(seq, "  %ums request delay\n",
1137 	    (s->stats->ts_requested == 0) ? 0 :
1138 	    jiffies_to_msecs(s->stats->run.rs_request_delay /
1139 			     s->stats->ts_requested));
1140 	seq_printf(seq, "  %ums running transaction\n",
1141 	    jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
1142 	seq_printf(seq, "  %ums transaction was being locked\n",
1143 	    jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
1144 	seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
1145 	    jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
1146 	seq_printf(seq, "  %ums logging transaction\n",
1147 	    jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
1148 	seq_printf(seq, "  %lluus average transaction commit time\n",
1149 		   div_u64(s->journal->j_average_commit_time, 1000));
1150 	seq_printf(seq, "  %lu handles per transaction\n",
1151 	    s->stats->run.rs_handle_count / s->stats->ts_tid);
1152 	seq_printf(seq, "  %lu blocks per transaction\n",
1153 	    s->stats->run.rs_blocks / s->stats->ts_tid);
1154 	seq_printf(seq, "  %lu logged blocks per transaction\n",
1155 	    s->stats->run.rs_blocks_logged / s->stats->ts_tid);
1156 	return 0;
1157 }
1158 
jbd2_seq_info_stop(struct seq_file * seq,void * v)1159 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
1160 {
1161 }
1162 
1163 static const struct seq_operations jbd2_seq_info_ops = {
1164 	.start  = jbd2_seq_info_start,
1165 	.next   = jbd2_seq_info_next,
1166 	.stop   = jbd2_seq_info_stop,
1167 	.show   = jbd2_seq_info_show,
1168 };
1169 
jbd2_seq_info_open(struct inode * inode,struct file * file)1170 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
1171 {
1172 	journal_t *journal = pde_data(inode);
1173 	struct jbd2_stats_proc_session *s;
1174 	int rc, size;
1175 
1176 	s = kmalloc(sizeof(*s), GFP_KERNEL);
1177 	if (s == NULL)
1178 		return -ENOMEM;
1179 	size = sizeof(struct transaction_stats_s);
1180 	s->stats = kmalloc(size, GFP_KERNEL);
1181 	if (s->stats == NULL) {
1182 		kfree(s);
1183 		return -ENOMEM;
1184 	}
1185 	spin_lock(&journal->j_history_lock);
1186 	memcpy(s->stats, &journal->j_stats, size);
1187 	s->journal = journal;
1188 	spin_unlock(&journal->j_history_lock);
1189 
1190 	rc = seq_open(file, &jbd2_seq_info_ops);
1191 	if (rc == 0) {
1192 		struct seq_file *m = file->private_data;
1193 		m->private = s;
1194 	} else {
1195 		kfree(s->stats);
1196 		kfree(s);
1197 	}
1198 	return rc;
1199 
1200 }
1201 
jbd2_seq_info_release(struct inode * inode,struct file * file)1202 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
1203 {
1204 	struct seq_file *seq = file->private_data;
1205 	struct jbd2_stats_proc_session *s = seq->private;
1206 	kfree(s->stats);
1207 	kfree(s);
1208 	return seq_release(inode, file);
1209 }
1210 
1211 static const struct proc_ops jbd2_info_proc_ops = {
1212 	.proc_open	= jbd2_seq_info_open,
1213 	.proc_read	= seq_read,
1214 	.proc_lseek	= seq_lseek,
1215 	.proc_release	= jbd2_seq_info_release,
1216 };
1217 
1218 static struct proc_dir_entry *proc_jbd2_stats;
1219 
jbd2_stats_proc_init(journal_t * journal)1220 static void jbd2_stats_proc_init(journal_t *journal)
1221 {
1222 	journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
1223 	if (journal->j_proc_entry) {
1224 		proc_create_data("info", S_IRUGO, journal->j_proc_entry,
1225 				 &jbd2_info_proc_ops, journal);
1226 	}
1227 }
1228 
jbd2_stats_proc_exit(journal_t * journal)1229 static void jbd2_stats_proc_exit(journal_t *journal)
1230 {
1231 	remove_proc_entry("info", journal->j_proc_entry);
1232 	remove_proc_entry(journal->j_devname, proc_jbd2_stats);
1233 }
1234 
1235 /* Minimum size of descriptor tag */
jbd2_min_tag_size(void)1236 static int jbd2_min_tag_size(void)
1237 {
1238 	/*
1239 	 * Tag with 32-bit block numbers does not use last four bytes of the
1240 	 * structure
1241 	 */
1242 	return sizeof(journal_block_tag_t) - 4;
1243 }
1244 
1245 /**
1246  * jbd2_journal_shrink_scan()
1247  * @shrink: shrinker to work on
1248  * @sc: reclaim request to process
1249  *
1250  * Scan the checkpointed buffer on the checkpoint list and release the
1251  * journal_head.
1252  */
jbd2_journal_shrink_scan(struct shrinker * shrink,struct shrink_control * sc)1253 static unsigned long jbd2_journal_shrink_scan(struct shrinker *shrink,
1254 					      struct shrink_control *sc)
1255 {
1256 	journal_t *journal = shrink->private_data;
1257 	unsigned long nr_to_scan = sc->nr_to_scan;
1258 	unsigned long nr_shrunk;
1259 	unsigned long count;
1260 
1261 	count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1262 	trace_jbd2_shrink_scan_enter(journal, sc->nr_to_scan, count);
1263 
1264 	nr_shrunk = jbd2_journal_shrink_checkpoint_list(journal, &nr_to_scan);
1265 
1266 	count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1267 	trace_jbd2_shrink_scan_exit(journal, nr_to_scan, nr_shrunk, count);
1268 
1269 	return nr_shrunk;
1270 }
1271 
1272 /**
1273  * jbd2_journal_shrink_count()
1274  * @shrink: shrinker to work on
1275  * @sc: reclaim request to process
1276  *
1277  * Count the number of checkpoint buffers on the checkpoint list.
1278  */
jbd2_journal_shrink_count(struct shrinker * shrink,struct shrink_control * sc)1279 static unsigned long jbd2_journal_shrink_count(struct shrinker *shrink,
1280 					       struct shrink_control *sc)
1281 {
1282 	journal_t *journal = shrink->private_data;
1283 	unsigned long count;
1284 
1285 	count = percpu_counter_read_positive(&journal->j_checkpoint_jh_count);
1286 	trace_jbd2_shrink_count(journal, sc->nr_to_scan, count);
1287 
1288 	return count;
1289 }
1290 
1291 /*
1292  * If the journal init or create aborts, we need to mark the journal
1293  * superblock as being NULL to prevent the journal destroy from writing
1294  * back a bogus superblock.
1295  */
journal_fail_superblock(journal_t * journal)1296 static void journal_fail_superblock(journal_t *journal)
1297 {
1298 	struct buffer_head *bh = journal->j_sb_buffer;
1299 	brelse(bh);
1300 	journal->j_sb_buffer = NULL;
1301 }
1302 
1303 /*
1304  * Check the superblock for a given journal, performing initial
1305  * validation of the format.
1306  */
journal_check_superblock(journal_t * journal)1307 static int journal_check_superblock(journal_t *journal)
1308 {
1309 	journal_superblock_t *sb = journal->j_superblock;
1310 	int num_fc_blks;
1311 	int err = -EINVAL;
1312 
1313 	if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1314 	    sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1315 		printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
1316 		return err;
1317 	}
1318 
1319 	if (be32_to_cpu(sb->s_header.h_blocktype) != JBD2_SUPERBLOCK_V1 &&
1320 	    be32_to_cpu(sb->s_header.h_blocktype) != JBD2_SUPERBLOCK_V2) {
1321 		printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
1322 		return err;
1323 	}
1324 
1325 	if (be32_to_cpu(sb->s_maxlen) > journal->j_total_len) {
1326 		printk(KERN_WARNING "JBD2: journal file too short\n");
1327 		return err;
1328 	}
1329 
1330 	if (be32_to_cpu(sb->s_first) == 0 ||
1331 	    be32_to_cpu(sb->s_first) >= journal->j_total_len) {
1332 		printk(KERN_WARNING
1333 			"JBD2: Invalid start block of journal: %u\n",
1334 			be32_to_cpu(sb->s_first));
1335 		return err;
1336 	}
1337 
1338 	/*
1339 	 * If this is a V2 superblock, then we have to check the
1340 	 * features flags on it.
1341 	 */
1342 	if (!jbd2_format_support_feature(journal))
1343 		return 0;
1344 
1345 	if ((sb->s_feature_ro_compat &
1346 			~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1347 	    (sb->s_feature_incompat &
1348 			~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1349 		printk(KERN_WARNING "JBD2: Unrecognised features on journal\n");
1350 		return err;
1351 	}
1352 
1353 	num_fc_blks = jbd2_has_feature_fast_commit(journal) ?
1354 				jbd2_journal_get_num_fc_blks(sb) : 0;
1355 	if (be32_to_cpu(sb->s_maxlen) < JBD2_MIN_JOURNAL_BLOCKS ||
1356 	    be32_to_cpu(sb->s_maxlen) - JBD2_MIN_JOURNAL_BLOCKS < num_fc_blks) {
1357 		printk(KERN_ERR "JBD2: journal file too short %u,%d\n",
1358 		       be32_to_cpu(sb->s_maxlen), num_fc_blks);
1359 		return err;
1360 	}
1361 
1362 	if (jbd2_has_feature_csum2(journal) &&
1363 	    jbd2_has_feature_csum3(journal)) {
1364 		/* Can't have checksum v2 and v3 at the same time! */
1365 		printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 "
1366 		       "at the same time!\n");
1367 		return err;
1368 	}
1369 
1370 	if (jbd2_journal_has_csum_v2or3_feature(journal) &&
1371 	    jbd2_has_feature_checksum(journal)) {
1372 		/* Can't have checksum v1 and v2 on at the same time! */
1373 		printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 "
1374 		       "at the same time!\n");
1375 		return err;
1376 	}
1377 
1378 	/* Load the checksum driver */
1379 	if (jbd2_journal_has_csum_v2or3_feature(journal)) {
1380 		if (sb->s_checksum_type != JBD2_CRC32C_CHKSUM) {
1381 			printk(KERN_ERR "JBD2: Unknown checksum type\n");
1382 			return err;
1383 		}
1384 
1385 		journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
1386 		if (IS_ERR(journal->j_chksum_driver)) {
1387 			printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
1388 			err = PTR_ERR(journal->j_chksum_driver);
1389 			journal->j_chksum_driver = NULL;
1390 			return err;
1391 		}
1392 		/* Check superblock checksum */
1393 		if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) {
1394 			printk(KERN_ERR "JBD2: journal checksum error\n");
1395 			err = -EFSBADCRC;
1396 			return err;
1397 		}
1398 	}
1399 
1400 	return 0;
1401 }
1402 
journal_revoke_records_per_block(journal_t * journal)1403 static int journal_revoke_records_per_block(journal_t *journal)
1404 {
1405 	int record_size;
1406 	int space = journal->j_blocksize - sizeof(jbd2_journal_revoke_header_t);
1407 
1408 	if (jbd2_has_feature_64bit(journal))
1409 		record_size = 8;
1410 	else
1411 		record_size = 4;
1412 
1413 	if (jbd2_journal_has_csum_v2or3(journal))
1414 		space -= sizeof(struct jbd2_journal_block_tail);
1415 	return space / record_size;
1416 }
1417 
jbd2_journal_get_max_txn_bufs(journal_t * journal)1418 static int jbd2_journal_get_max_txn_bufs(journal_t *journal)
1419 {
1420 	return (journal->j_total_len - journal->j_fc_wbufsize) / 3;
1421 }
1422 
1423 /*
1424  * Base amount of descriptor blocks we reserve for each transaction.
1425  */
jbd2_descriptor_blocks_per_trans(journal_t * journal)1426 static int jbd2_descriptor_blocks_per_trans(journal_t *journal)
1427 {
1428 	int tag_space = journal->j_blocksize - sizeof(journal_header_t);
1429 	int tags_per_block;
1430 
1431 	/* Subtract UUID */
1432 	tag_space -= 16;
1433 	if (jbd2_journal_has_csum_v2or3(journal))
1434 		tag_space -= sizeof(struct jbd2_journal_block_tail);
1435 	/* Commit code leaves a slack space of 16 bytes at the end of block */
1436 	tags_per_block = (tag_space - 16) / journal_tag_bytes(journal);
1437 	/*
1438 	 * Revoke descriptors are accounted separately so we need to reserve
1439 	 * space for commit block and normal transaction descriptor blocks.
1440 	 */
1441 	return 1 + DIV_ROUND_UP(jbd2_journal_get_max_txn_bufs(journal),
1442 				tags_per_block);
1443 }
1444 
1445 /*
1446  * Initialize number of blocks each transaction reserves for its bookkeeping
1447  * and maximum number of blocks a transaction can use. This needs to be called
1448  * after the journal size and the fastcommit area size are initialized.
1449  */
jbd2_journal_init_transaction_limits(journal_t * journal)1450 static void jbd2_journal_init_transaction_limits(journal_t *journal)
1451 {
1452 	journal->j_revoke_records_per_block =
1453 				journal_revoke_records_per_block(journal);
1454 	journal->j_transaction_overhead_buffers =
1455 				jbd2_descriptor_blocks_per_trans(journal);
1456 	journal->j_max_transaction_buffers =
1457 				jbd2_journal_get_max_txn_bufs(journal);
1458 }
1459 
1460 /*
1461  * Load the on-disk journal superblock and read the key fields into the
1462  * journal_t.
1463  */
journal_load_superblock(journal_t * journal)1464 static int journal_load_superblock(journal_t *journal)
1465 {
1466 	int err;
1467 	struct buffer_head *bh;
1468 	journal_superblock_t *sb;
1469 
1470 	bh = getblk_unmovable(journal->j_dev, journal->j_blk_offset,
1471 			      journal->j_blocksize);
1472 	if (bh)
1473 		err = bh_read(bh, 0);
1474 	if (!bh || err < 0) {
1475 		pr_err("%s: Cannot read journal superblock\n", __func__);
1476 		brelse(bh);
1477 		return -EIO;
1478 	}
1479 
1480 	journal->j_sb_buffer = bh;
1481 	sb = (journal_superblock_t *)bh->b_data;
1482 	journal->j_superblock = sb;
1483 	err = journal_check_superblock(journal);
1484 	if (err) {
1485 		journal_fail_superblock(journal);
1486 		return err;
1487 	}
1488 
1489 	journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1490 	journal->j_tail = be32_to_cpu(sb->s_start);
1491 	journal->j_first = be32_to_cpu(sb->s_first);
1492 	journal->j_errno = be32_to_cpu(sb->s_errno);
1493 	journal->j_last = be32_to_cpu(sb->s_maxlen);
1494 
1495 	if (be32_to_cpu(sb->s_maxlen) < journal->j_total_len)
1496 		journal->j_total_len = be32_to_cpu(sb->s_maxlen);
1497 	/* Precompute checksum seed for all metadata */
1498 	if (jbd2_journal_has_csum_v2or3(journal))
1499 		journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
1500 						   sizeof(sb->s_uuid));
1501 	/* After journal features are set, we can compute transaction limits */
1502 	jbd2_journal_init_transaction_limits(journal);
1503 
1504 	if (jbd2_has_feature_fast_commit(journal)) {
1505 		journal->j_fc_last = be32_to_cpu(sb->s_maxlen);
1506 		journal->j_last = journal->j_fc_last -
1507 				  jbd2_journal_get_num_fc_blks(sb);
1508 		journal->j_fc_first = journal->j_last + 1;
1509 		journal->j_fc_off = 0;
1510 	}
1511 
1512 	return 0;
1513 }
1514 
1515 
1516 /*
1517  * Management for journal control blocks: functions to create and
1518  * destroy journal_t structures, and to initialise and read existing
1519  * journal blocks from disk.  */
1520 
1521 /* First: create and setup a journal_t object in memory.  We initialise
1522  * very few fields yet: that has to wait until we have created the
1523  * journal structures from from scratch, or loaded them from disk. */
1524 
journal_init_common(struct block_device * bdev,struct block_device * fs_dev,unsigned long long start,int len,int blocksize)1525 static journal_t *journal_init_common(struct block_device *bdev,
1526 			struct block_device *fs_dev,
1527 			unsigned long long start, int len, int blocksize)
1528 {
1529 	static struct lock_class_key jbd2_trans_commit_key;
1530 	journal_t *journal;
1531 	int err;
1532 	int n;
1533 
1534 	journal = kzalloc(sizeof(*journal), GFP_KERNEL);
1535 	if (!journal)
1536 		return ERR_PTR(-ENOMEM);
1537 
1538 	journal->j_blocksize = blocksize;
1539 	journal->j_dev = bdev;
1540 	journal->j_fs_dev = fs_dev;
1541 	journal->j_blk_offset = start;
1542 	journal->j_total_len = len;
1543 	jbd2_init_fs_dev_write_error(journal);
1544 
1545 	err = journal_load_superblock(journal);
1546 	if (err)
1547 		goto err_cleanup;
1548 
1549 	init_waitqueue_head(&journal->j_wait_transaction_locked);
1550 	init_waitqueue_head(&journal->j_wait_done_commit);
1551 	init_waitqueue_head(&journal->j_wait_commit);
1552 	init_waitqueue_head(&journal->j_wait_updates);
1553 	init_waitqueue_head(&journal->j_wait_reserved);
1554 	init_waitqueue_head(&journal->j_fc_wait);
1555 	mutex_init(&journal->j_abort_mutex);
1556 	mutex_init(&journal->j_barrier);
1557 	mutex_init(&journal->j_checkpoint_mutex);
1558 	spin_lock_init(&journal->j_revoke_lock);
1559 	spin_lock_init(&journal->j_list_lock);
1560 	spin_lock_init(&journal->j_history_lock);
1561 	rwlock_init(&journal->j_state_lock);
1562 
1563 	journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
1564 	journal->j_min_batch_time = 0;
1565 	journal->j_max_batch_time = 15000; /* 15ms */
1566 	atomic_set(&journal->j_reserved_credits, 0);
1567 	lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle",
1568 			 &jbd2_trans_commit_key, 0);
1569 
1570 	/* The journal is marked for error until we succeed with recovery! */
1571 	journal->j_flags = JBD2_ABORT;
1572 
1573 	/* Set up a default-sized revoke table for the new mount. */
1574 	err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
1575 	if (err)
1576 		goto err_cleanup;
1577 
1578 	/*
1579 	 * journal descriptor can store up to n blocks, we need enough
1580 	 * buffers to write out full descriptor block.
1581 	 */
1582 	err = -ENOMEM;
1583 	n = journal->j_blocksize / jbd2_min_tag_size();
1584 	journal->j_wbufsize = n;
1585 	journal->j_fc_wbuf = NULL;
1586 	journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *),
1587 					GFP_KERNEL);
1588 	if (!journal->j_wbuf)
1589 		goto err_cleanup;
1590 
1591 	err = percpu_counter_init(&journal->j_checkpoint_jh_count, 0,
1592 				  GFP_KERNEL);
1593 	if (err)
1594 		goto err_cleanup;
1595 
1596 	journal->j_shrink_transaction = NULL;
1597 
1598 	journal->j_shrinker = shrinker_alloc(0, "jbd2-journal:(%u:%u)",
1599 					     MAJOR(bdev->bd_dev),
1600 					     MINOR(bdev->bd_dev));
1601 	if (!journal->j_shrinker) {
1602 		err = -ENOMEM;
1603 		goto err_cleanup;
1604 	}
1605 
1606 	journal->j_shrinker->scan_objects = jbd2_journal_shrink_scan;
1607 	journal->j_shrinker->count_objects = jbd2_journal_shrink_count;
1608 	journal->j_shrinker->private_data = journal;
1609 
1610 	shrinker_register(journal->j_shrinker);
1611 
1612 	return journal;
1613 
1614 err_cleanup:
1615 	percpu_counter_destroy(&journal->j_checkpoint_jh_count);
1616 	if (journal->j_chksum_driver)
1617 		crypto_free_shash(journal->j_chksum_driver);
1618 	kfree(journal->j_wbuf);
1619 	jbd2_journal_destroy_revoke(journal);
1620 	journal_fail_superblock(journal);
1621 	kfree(journal);
1622 	return ERR_PTR(err);
1623 }
1624 
1625 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
1626  *
1627  * Create a journal structure assigned some fixed set of disk blocks to
1628  * the journal.  We don't actually touch those disk blocks yet, but we
1629  * need to set up all of the mapping information to tell the journaling
1630  * system where the journal blocks are.
1631  *
1632  */
1633 
1634 /**
1635  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
1636  *  @bdev: Block device on which to create the journal
1637  *  @fs_dev: Device which hold journalled filesystem for this journal.
1638  *  @start: Block nr Start of journal.
1639  *  @len:  Length of the journal in blocks.
1640  *  @blocksize: blocksize of journalling device
1641  *
1642  *  Returns: a newly created journal_t *
1643  *
1644  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
1645  *  range of blocks on an arbitrary block device.
1646  *
1647  */
jbd2_journal_init_dev(struct block_device * bdev,struct block_device * fs_dev,unsigned long long start,int len,int blocksize)1648 journal_t *jbd2_journal_init_dev(struct block_device *bdev,
1649 			struct block_device *fs_dev,
1650 			unsigned long long start, int len, int blocksize)
1651 {
1652 	journal_t *journal;
1653 
1654 	journal = journal_init_common(bdev, fs_dev, start, len, blocksize);
1655 	if (IS_ERR(journal))
1656 		return ERR_CAST(journal);
1657 
1658 	snprintf(journal->j_devname, sizeof(journal->j_devname),
1659 		 "%pg", journal->j_dev);
1660 	strreplace(journal->j_devname, '/', '!');
1661 	jbd2_stats_proc_init(journal);
1662 
1663 	return journal;
1664 }
1665 
1666 /**
1667  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
1668  *  @inode: An inode to create the journal in
1669  *
1670  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
1671  * the journal.  The inode must exist already, must support bmap() and
1672  * must have all data blocks preallocated.
1673  */
jbd2_journal_init_inode(struct inode * inode)1674 journal_t *jbd2_journal_init_inode(struct inode *inode)
1675 {
1676 	journal_t *journal;
1677 	sector_t blocknr;
1678 	int err = 0;
1679 
1680 	blocknr = 0;
1681 	err = bmap(inode, &blocknr);
1682 	if (err || !blocknr) {
1683 		pr_err("%s: Cannot locate journal superblock\n", __func__);
1684 		return err ? ERR_PTR(err) : ERR_PTR(-EINVAL);
1685 	}
1686 
1687 	jbd2_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n",
1688 		  inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size,
1689 		  inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1690 
1691 	journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev,
1692 			blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits,
1693 			inode->i_sb->s_blocksize);
1694 	if (IS_ERR(journal))
1695 		return ERR_CAST(journal);
1696 
1697 	journal->j_inode = inode;
1698 	snprintf(journal->j_devname, sizeof(journal->j_devname),
1699 		 "%pg-%lu", journal->j_dev, journal->j_inode->i_ino);
1700 	strreplace(journal->j_devname, '/', '!');
1701 	jbd2_stats_proc_init(journal);
1702 
1703 	return journal;
1704 }
1705 
1706 /*
1707  * Given a journal_t structure, initialise the various fields for
1708  * startup of a new journaling session.  We use this both when creating
1709  * a journal, and after recovering an old journal to reset it for
1710  * subsequent use.
1711  */
1712 
journal_reset(journal_t * journal)1713 static int journal_reset(journal_t *journal)
1714 {
1715 	journal_superblock_t *sb = journal->j_superblock;
1716 	unsigned long long first, last;
1717 
1718 	first = be32_to_cpu(sb->s_first);
1719 	last = be32_to_cpu(sb->s_maxlen);
1720 	if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1721 		printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
1722 		       first, last);
1723 		journal_fail_superblock(journal);
1724 		return -EINVAL;
1725 	}
1726 
1727 	journal->j_first = first;
1728 	journal->j_last = last;
1729 
1730 	if (journal->j_head != 0 && journal->j_flags & JBD2_CYCLE_RECORD) {
1731 		/*
1732 		 * Disable the cycled recording mode if the journal head block
1733 		 * number is not correct.
1734 		 */
1735 		if (journal->j_head < first || journal->j_head >= last) {
1736 			printk(KERN_WARNING "JBD2: Incorrect Journal head block %lu, "
1737 			       "disable journal_cycle_record\n",
1738 			       journal->j_head);
1739 			journal->j_head = journal->j_first;
1740 		}
1741 	} else {
1742 		journal->j_head = journal->j_first;
1743 	}
1744 	journal->j_tail = journal->j_head;
1745 	journal->j_free = journal->j_last - journal->j_first;
1746 
1747 	journal->j_tail_sequence = journal->j_transaction_sequence;
1748 	journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1749 	journal->j_commit_request = journal->j_commit_sequence;
1750 
1751 	/*
1752 	 * Now that journal recovery is done, turn fast commits off here. This
1753 	 * way, if fast commit was enabled before the crash but if now FS has
1754 	 * disabled it, we don't enable fast commits.
1755 	 */
1756 	jbd2_clear_feature_fast_commit(journal);
1757 
1758 	/*
1759 	 * As a special case, if the on-disk copy is already marked as needing
1760 	 * no recovery (s_start == 0), then we can safely defer the superblock
1761 	 * update until the next commit by setting JBD2_FLUSHED.  This avoids
1762 	 * attempting a write to a potential-readonly device.
1763 	 */
1764 	if (sb->s_start == 0) {
1765 		jbd2_debug(1, "JBD2: Skipping superblock update on recovered sb "
1766 			"(start %ld, seq %u, errno %d)\n",
1767 			journal->j_tail, journal->j_tail_sequence,
1768 			journal->j_errno);
1769 		journal->j_flags |= JBD2_FLUSHED;
1770 	} else {
1771 		/* Lock here to make assertions happy... */
1772 		mutex_lock_io(&journal->j_checkpoint_mutex);
1773 		/*
1774 		 * Update log tail information. We use REQ_FUA since new
1775 		 * transaction will start reusing journal space and so we
1776 		 * must make sure information about current log tail is on
1777 		 * disk before that.
1778 		 */
1779 		jbd2_journal_update_sb_log_tail(journal,
1780 						journal->j_tail_sequence,
1781 						journal->j_tail, REQ_FUA);
1782 		mutex_unlock(&journal->j_checkpoint_mutex);
1783 	}
1784 	return jbd2_journal_start_thread(journal);
1785 }
1786 
1787 /*
1788  * This function expects that the caller will have locked the journal
1789  * buffer head, and will return with it unlocked
1790  */
jbd2_write_superblock(journal_t * journal,blk_opf_t write_flags)1791 static int jbd2_write_superblock(journal_t *journal, blk_opf_t write_flags)
1792 {
1793 	struct buffer_head *bh = journal->j_sb_buffer;
1794 	journal_superblock_t *sb = journal->j_superblock;
1795 	int ret = 0;
1796 
1797 	/* Buffer got discarded which means block device got invalidated */
1798 	if (!buffer_mapped(bh)) {
1799 		unlock_buffer(bh);
1800 		return -EIO;
1801 	}
1802 
1803 	/*
1804 	 * Always set high priority flags to exempt from block layer's
1805 	 * QOS policies, e.g. writeback throttle.
1806 	 */
1807 	write_flags |= JBD2_JOURNAL_REQ_FLAGS;
1808 	if (!(journal->j_flags & JBD2_BARRIER))
1809 		write_flags &= ~(REQ_FUA | REQ_PREFLUSH);
1810 
1811 	trace_jbd2_write_superblock(journal, write_flags);
1812 
1813 	if (buffer_write_io_error(bh)) {
1814 		/*
1815 		 * Oh, dear.  A previous attempt to write the journal
1816 		 * superblock failed.  This could happen because the
1817 		 * USB device was yanked out.  Or it could happen to
1818 		 * be a transient write error and maybe the block will
1819 		 * be remapped.  Nothing we can do but to retry the
1820 		 * write and hope for the best.
1821 		 */
1822 		printk(KERN_ERR "JBD2: previous I/O error detected "
1823 		       "for journal superblock update for %s.\n",
1824 		       journal->j_devname);
1825 		clear_buffer_write_io_error(bh);
1826 		set_buffer_uptodate(bh);
1827 	}
1828 	if (jbd2_journal_has_csum_v2or3(journal))
1829 		sb->s_checksum = jbd2_superblock_csum(journal, sb);
1830 	get_bh(bh);
1831 	bh->b_end_io = end_buffer_write_sync;
1832 	submit_bh(REQ_OP_WRITE | write_flags, bh);
1833 	wait_on_buffer(bh);
1834 	if (buffer_write_io_error(bh)) {
1835 		clear_buffer_write_io_error(bh);
1836 		set_buffer_uptodate(bh);
1837 		ret = -EIO;
1838 	}
1839 	if (ret) {
1840 		printk(KERN_ERR "JBD2: I/O error when updating journal superblock for %s.\n",
1841 				journal->j_devname);
1842 		if (!is_journal_aborted(journal))
1843 			jbd2_journal_abort(journal, ret);
1844 	}
1845 
1846 	return ret;
1847 }
1848 
1849 /**
1850  * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk.
1851  * @journal: The journal to update.
1852  * @tail_tid: TID of the new transaction at the tail of the log
1853  * @tail_block: The first block of the transaction at the tail of the log
1854  * @write_flags: Flags for the journal sb write operation
1855  *
1856  * Update a journal's superblock information about log tail and write it to
1857  * disk, waiting for the IO to complete.
1858  */
jbd2_journal_update_sb_log_tail(journal_t * journal,tid_t tail_tid,unsigned long tail_block,blk_opf_t write_flags)1859 int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid,
1860 				    unsigned long tail_block,
1861 				    blk_opf_t write_flags)
1862 {
1863 	journal_superblock_t *sb = journal->j_superblock;
1864 	int ret;
1865 
1866 	if (is_journal_aborted(journal))
1867 		return -EIO;
1868 	if (jbd2_check_fs_dev_write_error(journal)) {
1869 		jbd2_journal_abort(journal, -EIO);
1870 		return -EIO;
1871 	}
1872 
1873 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1874 	jbd2_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n",
1875 		  tail_block, tail_tid);
1876 
1877 	lock_buffer(journal->j_sb_buffer);
1878 	sb->s_sequence = cpu_to_be32(tail_tid);
1879 	sb->s_start    = cpu_to_be32(tail_block);
1880 
1881 	ret = jbd2_write_superblock(journal, write_flags);
1882 	if (ret)
1883 		goto out;
1884 
1885 	/* Log is no longer empty */
1886 	write_lock(&journal->j_state_lock);
1887 	WARN_ON(!sb->s_sequence);
1888 	journal->j_flags &= ~JBD2_FLUSHED;
1889 	write_unlock(&journal->j_state_lock);
1890 
1891 out:
1892 	return ret;
1893 }
1894 
1895 /**
1896  * jbd2_mark_journal_empty() - Mark on disk journal as empty.
1897  * @journal: The journal to update.
1898  * @write_flags: Flags for the journal sb write operation
1899  *
1900  * Update a journal's dynamic superblock fields to show that journal is empty.
1901  * Write updated superblock to disk waiting for IO to complete.
1902  */
jbd2_mark_journal_empty(journal_t * journal,blk_opf_t write_flags)1903 static void jbd2_mark_journal_empty(journal_t *journal, blk_opf_t write_flags)
1904 {
1905 	journal_superblock_t *sb = journal->j_superblock;
1906 	bool had_fast_commit = false;
1907 
1908 	BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex));
1909 	lock_buffer(journal->j_sb_buffer);
1910 	if (sb->s_start == 0) {		/* Is it already empty? */
1911 		unlock_buffer(journal->j_sb_buffer);
1912 		return;
1913 	}
1914 
1915 	jbd2_debug(1, "JBD2: Marking journal as empty (seq %u)\n",
1916 		  journal->j_tail_sequence);
1917 
1918 	sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1919 	sb->s_start    = cpu_to_be32(0);
1920 	sb->s_head     = cpu_to_be32(journal->j_head);
1921 	if (jbd2_has_feature_fast_commit(journal)) {
1922 		/*
1923 		 * When journal is clean, no need to commit fast commit flag and
1924 		 * make file system incompatible with older kernels.
1925 		 */
1926 		jbd2_clear_feature_fast_commit(journal);
1927 		had_fast_commit = true;
1928 	}
1929 
1930 	jbd2_write_superblock(journal, write_flags);
1931 
1932 	if (had_fast_commit)
1933 		jbd2_set_feature_fast_commit(journal);
1934 
1935 	/* Log is empty */
1936 	write_lock(&journal->j_state_lock);
1937 	journal->j_flags |= JBD2_FLUSHED;
1938 	write_unlock(&journal->j_state_lock);
1939 }
1940 
1941 /**
1942  * __jbd2_journal_erase() - Discard or zeroout journal blocks (excluding superblock)
1943  * @journal: The journal to erase.
1944  * @flags: A discard/zeroout request is sent for each physically contigous
1945  *	region of the journal. Either JBD2_JOURNAL_FLUSH_DISCARD or
1946  *	JBD2_JOURNAL_FLUSH_ZEROOUT must be set to determine which operation
1947  *	to perform.
1948  *
1949  * Note: JBD2_JOURNAL_FLUSH_ZEROOUT attempts to use hardware offload. Zeroes
1950  * will be explicitly written if no hardware offload is available, see
1951  * blkdev_issue_zeroout for more details.
1952  */
__jbd2_journal_erase(journal_t * journal,unsigned int flags)1953 static int __jbd2_journal_erase(journal_t *journal, unsigned int flags)
1954 {
1955 	int err = 0;
1956 	unsigned long block, log_offset; /* logical */
1957 	unsigned long long phys_block, block_start, block_stop; /* physical */
1958 	loff_t byte_start, byte_stop, byte_count;
1959 
1960 	/* flags must be set to either discard or zeroout */
1961 	if ((flags & ~JBD2_JOURNAL_FLUSH_VALID) || !flags ||
1962 			((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1963 			(flags & JBD2_JOURNAL_FLUSH_ZEROOUT)))
1964 		return -EINVAL;
1965 
1966 	if ((flags & JBD2_JOURNAL_FLUSH_DISCARD) &&
1967 	    !bdev_max_discard_sectors(journal->j_dev))
1968 		return -EOPNOTSUPP;
1969 
1970 	/*
1971 	 * lookup block mapping and issue discard/zeroout for each
1972 	 * contiguous region
1973 	 */
1974 	log_offset = be32_to_cpu(journal->j_superblock->s_first);
1975 	block_start =  ~0ULL;
1976 	for (block = log_offset; block < journal->j_total_len; block++) {
1977 		err = jbd2_journal_bmap(journal, block, &phys_block);
1978 		if (err) {
1979 			pr_err("JBD2: bad block at offset %lu", block);
1980 			return err;
1981 		}
1982 
1983 		if (block_start == ~0ULL) {
1984 			block_start = phys_block;
1985 			block_stop = block_start - 1;
1986 		}
1987 
1988 		/*
1989 		 * last block not contiguous with current block,
1990 		 * process last contiguous region and return to this block on
1991 		 * next loop
1992 		 */
1993 		if (phys_block != block_stop + 1) {
1994 			block--;
1995 		} else {
1996 			block_stop++;
1997 			/*
1998 			 * if this isn't the last block of journal,
1999 			 * no need to process now because next block may also
2000 			 * be part of this contiguous region
2001 			 */
2002 			if (block != journal->j_total_len - 1)
2003 				continue;
2004 		}
2005 
2006 		/*
2007 		 * end of contiguous region or this is last block of journal,
2008 		 * take care of the region
2009 		 */
2010 		byte_start = block_start * journal->j_blocksize;
2011 		byte_stop = block_stop * journal->j_blocksize;
2012 		byte_count = (block_stop - block_start + 1) *
2013 				journal->j_blocksize;
2014 
2015 		truncate_inode_pages_range(journal->j_dev->bd_mapping,
2016 				byte_start, byte_stop);
2017 
2018 		if (flags & JBD2_JOURNAL_FLUSH_DISCARD) {
2019 			err = blkdev_issue_discard(journal->j_dev,
2020 					byte_start >> SECTOR_SHIFT,
2021 					byte_count >> SECTOR_SHIFT,
2022 					GFP_NOFS);
2023 		} else if (flags & JBD2_JOURNAL_FLUSH_ZEROOUT) {
2024 			err = blkdev_issue_zeroout(journal->j_dev,
2025 					byte_start >> SECTOR_SHIFT,
2026 					byte_count >> SECTOR_SHIFT,
2027 					GFP_NOFS, 0);
2028 		}
2029 
2030 		if (unlikely(err != 0)) {
2031 			pr_err("JBD2: (error %d) unable to wipe journal at physical blocks %llu - %llu",
2032 					err, block_start, block_stop);
2033 			return err;
2034 		}
2035 
2036 		/* reset start and stop after processing a region */
2037 		block_start = ~0ULL;
2038 	}
2039 
2040 	return blkdev_issue_flush(journal->j_dev);
2041 }
2042 
2043 /**
2044  * jbd2_journal_update_sb_errno() - Update error in the journal.
2045  * @journal: The journal to update.
2046  *
2047  * Update a journal's errno.  Write updated superblock to disk waiting for IO
2048  * to complete.
2049  */
jbd2_journal_update_sb_errno(journal_t * journal)2050 void jbd2_journal_update_sb_errno(journal_t *journal)
2051 {
2052 	journal_superblock_t *sb = journal->j_superblock;
2053 	int errcode;
2054 
2055 	lock_buffer(journal->j_sb_buffer);
2056 	errcode = journal->j_errno;
2057 	if (errcode == -ESHUTDOWN)
2058 		errcode = 0;
2059 	jbd2_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode);
2060 	sb->s_errno    = cpu_to_be32(errcode);
2061 
2062 	jbd2_write_superblock(journal, REQ_FUA);
2063 }
2064 EXPORT_SYMBOL(jbd2_journal_update_sb_errno);
2065 
2066 /**
2067  * jbd2_journal_load() - Read journal from disk.
2068  * @journal: Journal to act on.
2069  *
2070  * Given a journal_t structure which tells us which disk blocks contain
2071  * a journal, read the journal from disk to initialise the in-memory
2072  * structures.
2073  */
jbd2_journal_load(journal_t * journal)2074 int jbd2_journal_load(journal_t *journal)
2075 {
2076 	int err;
2077 	journal_superblock_t *sb = journal->j_superblock;
2078 
2079 	/*
2080 	 * Create a slab for this blocksize
2081 	 */
2082 	err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
2083 	if (err)
2084 		return err;
2085 
2086 	/* Let the recovery code check whether it needs to recover any
2087 	 * data from the journal. */
2088 	err = jbd2_journal_recover(journal);
2089 	if (err) {
2090 		pr_warn("JBD2: journal recovery failed\n");
2091 		return err;
2092 	}
2093 
2094 	if (journal->j_failed_commit) {
2095 		printk(KERN_ERR "JBD2: journal transaction %u on %s "
2096 		       "is corrupt.\n", journal->j_failed_commit,
2097 		       journal->j_devname);
2098 		return -EFSCORRUPTED;
2099 	}
2100 	/*
2101 	 * clear JBD2_ABORT flag initialized in journal_init_common
2102 	 * here to update log tail information with the newest seq.
2103 	 */
2104 	journal->j_flags &= ~JBD2_ABORT;
2105 
2106 	/* OK, we've finished with the dynamic journal bits:
2107 	 * reinitialise the dynamic contents of the superblock in memory
2108 	 * and reset them on disk. */
2109 	err = journal_reset(journal);
2110 	if (err) {
2111 		pr_warn("JBD2: journal reset failed\n");
2112 		return err;
2113 	}
2114 
2115 	journal->j_flags |= JBD2_LOADED;
2116 	return 0;
2117 }
2118 
2119 /**
2120  * jbd2_journal_destroy() - Release a journal_t structure.
2121  * @journal: Journal to act on.
2122  *
2123  * Release a journal_t structure once it is no longer in use by the
2124  * journaled object.
2125  * Return <0 if we couldn't clean up the journal.
2126  */
jbd2_journal_destroy(journal_t * journal)2127 int jbd2_journal_destroy(journal_t *journal)
2128 {
2129 	int err = 0;
2130 
2131 	/* Wait for the commit thread to wake up and die. */
2132 	journal_kill_thread(journal);
2133 
2134 	/* Force a final log commit */
2135 	if (journal->j_running_transaction)
2136 		jbd2_journal_commit_transaction(journal);
2137 
2138 	/* Force any old transactions to disk */
2139 
2140 	/* Totally anal locking here... */
2141 	spin_lock(&journal->j_list_lock);
2142 	while (journal->j_checkpoint_transactions != NULL) {
2143 		spin_unlock(&journal->j_list_lock);
2144 		mutex_lock_io(&journal->j_checkpoint_mutex);
2145 		err = jbd2_log_do_checkpoint(journal);
2146 		mutex_unlock(&journal->j_checkpoint_mutex);
2147 		/*
2148 		 * If checkpointing failed, just free the buffers to avoid
2149 		 * looping forever
2150 		 */
2151 		if (err) {
2152 			jbd2_journal_destroy_checkpoint(journal);
2153 			spin_lock(&journal->j_list_lock);
2154 			break;
2155 		}
2156 		spin_lock(&journal->j_list_lock);
2157 	}
2158 
2159 	J_ASSERT(journal->j_running_transaction == NULL);
2160 	J_ASSERT(journal->j_committing_transaction == NULL);
2161 	J_ASSERT(journal->j_checkpoint_transactions == NULL);
2162 	spin_unlock(&journal->j_list_lock);
2163 
2164 	/*
2165 	 * OK, all checkpoint transactions have been checked, now check the
2166 	 * writeback errseq of fs dev and abort the journal if some buffer
2167 	 * failed to write back to the original location, otherwise the
2168 	 * filesystem may become inconsistent.
2169 	 */
2170 	if (!is_journal_aborted(journal) &&
2171 	    jbd2_check_fs_dev_write_error(journal))
2172 		jbd2_journal_abort(journal, -EIO);
2173 
2174 	if (journal->j_sb_buffer) {
2175 		if (!is_journal_aborted(journal)) {
2176 			mutex_lock_io(&journal->j_checkpoint_mutex);
2177 
2178 			write_lock(&journal->j_state_lock);
2179 			journal->j_tail_sequence =
2180 				++journal->j_transaction_sequence;
2181 			write_unlock(&journal->j_state_lock);
2182 
2183 			jbd2_mark_journal_empty(journal, REQ_PREFLUSH | REQ_FUA);
2184 			mutex_unlock(&journal->j_checkpoint_mutex);
2185 		} else
2186 			err = -EIO;
2187 		brelse(journal->j_sb_buffer);
2188 	}
2189 
2190 	if (journal->j_shrinker) {
2191 		percpu_counter_destroy(&journal->j_checkpoint_jh_count);
2192 		shrinker_free(journal->j_shrinker);
2193 	}
2194 	if (journal->j_proc_entry)
2195 		jbd2_stats_proc_exit(journal);
2196 	iput(journal->j_inode);
2197 	if (journal->j_revoke)
2198 		jbd2_journal_destroy_revoke(journal);
2199 	if (journal->j_chksum_driver)
2200 		crypto_free_shash(journal->j_chksum_driver);
2201 	kfree(journal->j_fc_wbuf);
2202 	kfree(journal->j_wbuf);
2203 	kfree(journal);
2204 
2205 	return err;
2206 }
2207 
2208 
2209 /**
2210  * jbd2_journal_check_used_features() - Check if features specified are used.
2211  * @journal: Journal to check.
2212  * @compat: bitmask of compatible features
2213  * @ro: bitmask of features that force read-only mount
2214  * @incompat: bitmask of incompatible features
2215  *
2216  * Check whether the journal uses all of a given set of
2217  * features.  Return true (non-zero) if it does.
2218  **/
2219 
jbd2_journal_check_used_features(journal_t * journal,unsigned long compat,unsigned long ro,unsigned long incompat)2220 int jbd2_journal_check_used_features(journal_t *journal, unsigned long compat,
2221 				 unsigned long ro, unsigned long incompat)
2222 {
2223 	journal_superblock_t *sb;
2224 
2225 	if (!compat && !ro && !incompat)
2226 		return 1;
2227 	if (!jbd2_format_support_feature(journal))
2228 		return 0;
2229 
2230 	sb = journal->j_superblock;
2231 
2232 	if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
2233 	    ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
2234 	    ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
2235 		return 1;
2236 
2237 	return 0;
2238 }
2239 
2240 /**
2241  * jbd2_journal_check_available_features() - Check feature set in journalling layer
2242  * @journal: Journal to check.
2243  * @compat: bitmask of compatible features
2244  * @ro: bitmask of features that force read-only mount
2245  * @incompat: bitmask of incompatible features
2246  *
2247  * Check whether the journaling code supports the use of
2248  * all of a given set of features on this journal.  Return true
2249  * (non-zero) if it can. */
2250 
jbd2_journal_check_available_features(journal_t * journal,unsigned long compat,unsigned long ro,unsigned long incompat)2251 int jbd2_journal_check_available_features(journal_t *journal, unsigned long compat,
2252 				      unsigned long ro, unsigned long incompat)
2253 {
2254 	if (!compat && !ro && !incompat)
2255 		return 1;
2256 
2257 	if (!jbd2_format_support_feature(journal))
2258 		return 0;
2259 
2260 	if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
2261 	    (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
2262 	    (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
2263 		return 1;
2264 
2265 	return 0;
2266 }
2267 
2268 static int
jbd2_journal_initialize_fast_commit(journal_t * journal)2269 jbd2_journal_initialize_fast_commit(journal_t *journal)
2270 {
2271 	journal_superblock_t *sb = journal->j_superblock;
2272 	unsigned long long num_fc_blks;
2273 
2274 	num_fc_blks = jbd2_journal_get_num_fc_blks(sb);
2275 	if (journal->j_last - num_fc_blks < JBD2_MIN_JOURNAL_BLOCKS)
2276 		return -ENOSPC;
2277 
2278 	/* Are we called twice? */
2279 	WARN_ON(journal->j_fc_wbuf != NULL);
2280 	journal->j_fc_wbuf = kmalloc_array(num_fc_blks,
2281 				sizeof(struct buffer_head *), GFP_KERNEL);
2282 	if (!journal->j_fc_wbuf)
2283 		return -ENOMEM;
2284 
2285 	journal->j_fc_wbufsize = num_fc_blks;
2286 	journal->j_fc_last = journal->j_last;
2287 	journal->j_last = journal->j_fc_last - num_fc_blks;
2288 	journal->j_fc_first = journal->j_last + 1;
2289 	journal->j_fc_off = 0;
2290 	journal->j_free = journal->j_last - journal->j_first;
2291 
2292 	return 0;
2293 }
2294 
2295 /**
2296  * jbd2_journal_set_features() - Mark a given journal feature in the superblock
2297  * @journal: Journal to act on.
2298  * @compat: bitmask of compatible features
2299  * @ro: bitmask of features that force read-only mount
2300  * @incompat: bitmask of incompatible features
2301  *
2302  * Mark a given journal feature as present on the
2303  * superblock.  Returns true if the requested features could be set.
2304  *
2305  */
2306 
jbd2_journal_set_features(journal_t * journal,unsigned long compat,unsigned long ro,unsigned long incompat)2307 int jbd2_journal_set_features(journal_t *journal, unsigned long compat,
2308 			  unsigned long ro, unsigned long incompat)
2309 {
2310 #define INCOMPAT_FEATURE_ON(f) \
2311 		((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f)))
2312 #define COMPAT_FEATURE_ON(f) \
2313 		((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f)))
2314 	journal_superblock_t *sb;
2315 
2316 	if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
2317 		return 1;
2318 
2319 	if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
2320 		return 0;
2321 
2322 	/* If enabling v2 checksums, turn on v3 instead */
2323 	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) {
2324 		incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2;
2325 		incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3;
2326 	}
2327 
2328 	/* Asking for checksumming v3 and v1?  Only give them v3. */
2329 	if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 &&
2330 	    compat & JBD2_FEATURE_COMPAT_CHECKSUM)
2331 		compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM;
2332 
2333 	jbd2_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
2334 		  compat, ro, incompat);
2335 
2336 	sb = journal->j_superblock;
2337 
2338 	if (incompat & JBD2_FEATURE_INCOMPAT_FAST_COMMIT) {
2339 		if (jbd2_journal_initialize_fast_commit(journal)) {
2340 			pr_err("JBD2: Cannot enable fast commits.\n");
2341 			return 0;
2342 		}
2343 	}
2344 
2345 	/* Load the checksum driver if necessary */
2346 	if ((journal->j_chksum_driver == NULL) &&
2347 	    INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2348 		journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
2349 		if (IS_ERR(journal->j_chksum_driver)) {
2350 			printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n");
2351 			journal->j_chksum_driver = NULL;
2352 			return 0;
2353 		}
2354 		/* Precompute checksum seed for all metadata */
2355 		journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid,
2356 						   sizeof(sb->s_uuid));
2357 	}
2358 
2359 	lock_buffer(journal->j_sb_buffer);
2360 
2361 	/* If enabling v3 checksums, update superblock */
2362 	if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) {
2363 		sb->s_checksum_type = JBD2_CRC32C_CHKSUM;
2364 		sb->s_feature_compat &=
2365 			~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM);
2366 	}
2367 
2368 	/* If enabling v1 checksums, downgrade superblock */
2369 	if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM))
2370 		sb->s_feature_incompat &=
2371 			~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 |
2372 				     JBD2_FEATURE_INCOMPAT_CSUM_V3);
2373 
2374 	sb->s_feature_compat    |= cpu_to_be32(compat);
2375 	sb->s_feature_ro_compat |= cpu_to_be32(ro);
2376 	sb->s_feature_incompat  |= cpu_to_be32(incompat);
2377 	unlock_buffer(journal->j_sb_buffer);
2378 	jbd2_journal_init_transaction_limits(journal);
2379 
2380 	return 1;
2381 #undef COMPAT_FEATURE_ON
2382 #undef INCOMPAT_FEATURE_ON
2383 }
2384 
2385 /*
2386  * jbd2_journal_clear_features() - Clear a given journal feature in the
2387  * 				    superblock
2388  * @journal: Journal to act on.
2389  * @compat: bitmask of compatible features
2390  * @ro: bitmask of features that force read-only mount
2391  * @incompat: bitmask of incompatible features
2392  *
2393  * Clear a given journal feature as present on the
2394  * superblock.
2395  */
jbd2_journal_clear_features(journal_t * journal,unsigned long compat,unsigned long ro,unsigned long incompat)2396 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
2397 				unsigned long ro, unsigned long incompat)
2398 {
2399 	journal_superblock_t *sb;
2400 
2401 	jbd2_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
2402 		  compat, ro, incompat);
2403 
2404 	sb = journal->j_superblock;
2405 
2406 	sb->s_feature_compat    &= ~cpu_to_be32(compat);
2407 	sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
2408 	sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
2409 	jbd2_journal_init_transaction_limits(journal);
2410 }
2411 EXPORT_SYMBOL(jbd2_journal_clear_features);
2412 
2413 /**
2414  * jbd2_journal_flush() - Flush journal
2415  * @journal: Journal to act on.
2416  * @flags: optional operation on the journal blocks after the flush (see below)
2417  *
2418  * Flush all data for a given journal to disk and empty the journal.
2419  * Filesystems can use this when remounting readonly to ensure that
2420  * recovery does not need to happen on remount. Optionally, a discard or zeroout
2421  * can be issued on the journal blocks after flushing.
2422  *
2423  * flags:
2424  *	JBD2_JOURNAL_FLUSH_DISCARD: issues discards for the journal blocks
2425  *	JBD2_JOURNAL_FLUSH_ZEROOUT: issues zeroouts for the journal blocks
2426  */
jbd2_journal_flush(journal_t * journal,unsigned int flags)2427 int jbd2_journal_flush(journal_t *journal, unsigned int flags)
2428 {
2429 	int err = 0;
2430 	transaction_t *transaction = NULL;
2431 
2432 	write_lock(&journal->j_state_lock);
2433 
2434 	/* Force everything buffered to the log... */
2435 	if (journal->j_running_transaction) {
2436 		transaction = journal->j_running_transaction;
2437 		__jbd2_log_start_commit(journal, transaction->t_tid);
2438 	} else if (journal->j_committing_transaction)
2439 		transaction = journal->j_committing_transaction;
2440 
2441 	/* Wait for the log commit to complete... */
2442 	if (transaction) {
2443 		tid_t tid = transaction->t_tid;
2444 
2445 		write_unlock(&journal->j_state_lock);
2446 		jbd2_log_wait_commit(journal, tid);
2447 	} else {
2448 		write_unlock(&journal->j_state_lock);
2449 	}
2450 
2451 	/* ...and flush everything in the log out to disk. */
2452 	spin_lock(&journal->j_list_lock);
2453 	while (!err && journal->j_checkpoint_transactions != NULL) {
2454 		spin_unlock(&journal->j_list_lock);
2455 		mutex_lock_io(&journal->j_checkpoint_mutex);
2456 		err = jbd2_log_do_checkpoint(journal);
2457 		mutex_unlock(&journal->j_checkpoint_mutex);
2458 		spin_lock(&journal->j_list_lock);
2459 	}
2460 	spin_unlock(&journal->j_list_lock);
2461 
2462 	if (is_journal_aborted(journal))
2463 		return -EIO;
2464 
2465 	mutex_lock_io(&journal->j_checkpoint_mutex);
2466 	if (!err) {
2467 		err = jbd2_cleanup_journal_tail(journal);
2468 		if (err < 0) {
2469 			mutex_unlock(&journal->j_checkpoint_mutex);
2470 			goto out;
2471 		}
2472 		err = 0;
2473 	}
2474 
2475 	/* Finally, mark the journal as really needing no recovery.
2476 	 * This sets s_start==0 in the underlying superblock, which is
2477 	 * the magic code for a fully-recovered superblock.  Any future
2478 	 * commits of data to the journal will restore the current
2479 	 * s_start value. */
2480 	jbd2_mark_journal_empty(journal, REQ_FUA);
2481 
2482 	if (flags)
2483 		err = __jbd2_journal_erase(journal, flags);
2484 
2485 	mutex_unlock(&journal->j_checkpoint_mutex);
2486 	write_lock(&journal->j_state_lock);
2487 	J_ASSERT(!journal->j_running_transaction);
2488 	J_ASSERT(!journal->j_committing_transaction);
2489 	J_ASSERT(!journal->j_checkpoint_transactions);
2490 	J_ASSERT(journal->j_head == journal->j_tail);
2491 	J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
2492 	write_unlock(&journal->j_state_lock);
2493 out:
2494 	return err;
2495 }
2496 
2497 /**
2498  * jbd2_journal_wipe() - Wipe journal contents
2499  * @journal: Journal to act on.
2500  * @write: flag (see below)
2501  *
2502  * Wipe out all of the contents of a journal, safely.  This will produce
2503  * a warning if the journal contains any valid recovery information.
2504  * Must be called between journal_init_*() and jbd2_journal_load().
2505  *
2506  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
2507  * we merely suppress recovery.
2508  */
2509 
jbd2_journal_wipe(journal_t * journal,int write)2510 int jbd2_journal_wipe(journal_t *journal, int write)
2511 {
2512 	int err;
2513 
2514 	J_ASSERT (!(journal->j_flags & JBD2_LOADED));
2515 
2516 	if (!journal->j_tail)
2517 		return 0;
2518 
2519 	printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
2520 		write ? "Clearing" : "Ignoring");
2521 
2522 	err = jbd2_journal_skip_recovery(journal);
2523 	if (write) {
2524 		/* Lock to make assertions happy... */
2525 		mutex_lock_io(&journal->j_checkpoint_mutex);
2526 		jbd2_mark_journal_empty(journal, REQ_FUA);
2527 		mutex_unlock(&journal->j_checkpoint_mutex);
2528 	}
2529 
2530 	return err;
2531 }
2532 
2533 /**
2534  * jbd2_journal_abort () - Shutdown the journal immediately.
2535  * @journal: the journal to shutdown.
2536  * @errno:   an error number to record in the journal indicating
2537  *           the reason for the shutdown.
2538  *
2539  * Perform a complete, immediate shutdown of the ENTIRE
2540  * journal (not of a single transaction).  This operation cannot be
2541  * undone without closing and reopening the journal.
2542  *
2543  * The jbd2_journal_abort function is intended to support higher level error
2544  * recovery mechanisms such as the ext2/ext3 remount-readonly error
2545  * mode.
2546  *
2547  * Journal abort has very specific semantics.  Any existing dirty,
2548  * unjournaled buffers in the main filesystem will still be written to
2549  * disk by bdflush, but the journaling mechanism will be suspended
2550  * immediately and no further transaction commits will be honoured.
2551  *
2552  * Any dirty, journaled buffers will be written back to disk without
2553  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
2554  * filesystem, but we _do_ attempt to leave as much data as possible
2555  * behind for fsck to use for cleanup.
2556  *
2557  * Any attempt to get a new transaction handle on a journal which is in
2558  * ABORT state will just result in an -EROFS error return.  A
2559  * jbd2_journal_stop on an existing handle will return -EIO if we have
2560  * entered abort state during the update.
2561  *
2562  * Recursive transactions are not disturbed by journal abort until the
2563  * final jbd2_journal_stop, which will receive the -EIO error.
2564  *
2565  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
2566  * which will be recorded (if possible) in the journal superblock.  This
2567  * allows a client to record failure conditions in the middle of a
2568  * transaction without having to complete the transaction to record the
2569  * failure to disk.  ext3_error, for example, now uses this
2570  * functionality.
2571  *
2572  */
2573 
jbd2_journal_abort(journal_t * journal,int errno)2574 void jbd2_journal_abort(journal_t *journal, int errno)
2575 {
2576 	transaction_t *transaction;
2577 
2578 	/*
2579 	 * Lock the aborting procedure until everything is done, this avoid
2580 	 * races between filesystem's error handling flow (e.g. ext4_abort()),
2581 	 * ensure panic after the error info is written into journal's
2582 	 * superblock.
2583 	 */
2584 	mutex_lock(&journal->j_abort_mutex);
2585 	/*
2586 	 * ESHUTDOWN always takes precedence because a file system check
2587 	 * caused by any other journal abort error is not required after
2588 	 * a shutdown triggered.
2589 	 */
2590 	write_lock(&journal->j_state_lock);
2591 	if (journal->j_flags & JBD2_ABORT) {
2592 		int old_errno = journal->j_errno;
2593 
2594 		write_unlock(&journal->j_state_lock);
2595 		if (old_errno != -ESHUTDOWN && errno == -ESHUTDOWN) {
2596 			journal->j_errno = errno;
2597 			jbd2_journal_update_sb_errno(journal);
2598 		}
2599 		mutex_unlock(&journal->j_abort_mutex);
2600 		return;
2601 	}
2602 
2603 	/*
2604 	 * Mark the abort as occurred and start current running transaction
2605 	 * to release all journaled buffer.
2606 	 */
2607 	pr_err("Aborting journal on device %s.\n", journal->j_devname);
2608 
2609 	journal->j_flags |= JBD2_ABORT;
2610 	journal->j_errno = errno;
2611 	transaction = journal->j_running_transaction;
2612 	if (transaction)
2613 		__jbd2_log_start_commit(journal, transaction->t_tid);
2614 	write_unlock(&journal->j_state_lock);
2615 
2616 	/*
2617 	 * Record errno to the journal super block, so that fsck and jbd2
2618 	 * layer could realise that a filesystem check is needed.
2619 	 */
2620 	jbd2_journal_update_sb_errno(journal);
2621 	mutex_unlock(&journal->j_abort_mutex);
2622 }
2623 
2624 /**
2625  * jbd2_journal_errno() - returns the journal's error state.
2626  * @journal: journal to examine.
2627  *
2628  * This is the errno number set with jbd2_journal_abort(), the last
2629  * time the journal was mounted - if the journal was stopped
2630  * without calling abort this will be 0.
2631  *
2632  * If the journal has been aborted on this mount time -EROFS will
2633  * be returned.
2634  */
jbd2_journal_errno(journal_t * journal)2635 int jbd2_journal_errno(journal_t *journal)
2636 {
2637 	int err;
2638 
2639 	read_lock(&journal->j_state_lock);
2640 	if (journal->j_flags & JBD2_ABORT)
2641 		err = -EROFS;
2642 	else
2643 		err = journal->j_errno;
2644 	read_unlock(&journal->j_state_lock);
2645 	return err;
2646 }
2647 
2648 /**
2649  * jbd2_journal_clear_err() - clears the journal's error state
2650  * @journal: journal to act on.
2651  *
2652  * An error must be cleared or acked to take a FS out of readonly
2653  * mode.
2654  */
jbd2_journal_clear_err(journal_t * journal)2655 int jbd2_journal_clear_err(journal_t *journal)
2656 {
2657 	int err = 0;
2658 
2659 	write_lock(&journal->j_state_lock);
2660 	if (journal->j_flags & JBD2_ABORT)
2661 		err = -EROFS;
2662 	else
2663 		journal->j_errno = 0;
2664 	write_unlock(&journal->j_state_lock);
2665 	return err;
2666 }
2667 
2668 /**
2669  * jbd2_journal_ack_err() - Ack journal err.
2670  * @journal: journal to act on.
2671  *
2672  * An error must be cleared or acked to take a FS out of readonly
2673  * mode.
2674  */
jbd2_journal_ack_err(journal_t * journal)2675 void jbd2_journal_ack_err(journal_t *journal)
2676 {
2677 	write_lock(&journal->j_state_lock);
2678 	if (journal->j_errno)
2679 		journal->j_flags |= JBD2_ACK_ERR;
2680 	write_unlock(&journal->j_state_lock);
2681 }
2682 
jbd2_journal_blocks_per_page(struct inode * inode)2683 int jbd2_journal_blocks_per_page(struct inode *inode)
2684 {
2685 	return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits);
2686 }
2687 
2688 /*
2689  * helper functions to deal with 32 or 64bit block numbers.
2690  */
journal_tag_bytes(journal_t * journal)2691 size_t journal_tag_bytes(journal_t *journal)
2692 {
2693 	size_t sz;
2694 
2695 	if (jbd2_has_feature_csum3(journal))
2696 		return sizeof(journal_block_tag3_t);
2697 
2698 	sz = sizeof(journal_block_tag_t);
2699 
2700 	if (jbd2_has_feature_csum2(journal))
2701 		sz += sizeof(__u16);
2702 
2703 	if (jbd2_has_feature_64bit(journal))
2704 		return sz;
2705 	else
2706 		return sz - sizeof(__u32);
2707 }
2708 
2709 /*
2710  * JBD memory management
2711  *
2712  * These functions are used to allocate block-sized chunks of memory
2713  * used for making copies of buffer_head data.  Very often it will be
2714  * page-sized chunks of data, but sometimes it will be in
2715  * sub-page-size chunks.  (For example, 16k pages on Power systems
2716  * with a 4k block file system.)  For blocks smaller than a page, we
2717  * use a SLAB allocator.  There are slab caches for each block size,
2718  * which are allocated at mount time, if necessary, and we only free
2719  * (all of) the slab caches when/if the jbd2 module is unloaded.  For
2720  * this reason we don't need to a mutex to protect access to
2721  * jbd2_slab[] allocating or releasing memory; only in
2722  * jbd2_journal_create_slab().
2723  */
2724 #define JBD2_MAX_SLABS 8
2725 static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
2726 
2727 static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
2728 	"jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
2729 	"jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
2730 };
2731 
2732 
jbd2_journal_destroy_slabs(void)2733 static void jbd2_journal_destroy_slabs(void)
2734 {
2735 	int i;
2736 
2737 	for (i = 0; i < JBD2_MAX_SLABS; i++) {
2738 		kmem_cache_destroy(jbd2_slab[i]);
2739 		jbd2_slab[i] = NULL;
2740 	}
2741 }
2742 
jbd2_journal_create_slab(size_t size)2743 static int jbd2_journal_create_slab(size_t size)
2744 {
2745 	static DEFINE_MUTEX(jbd2_slab_create_mutex);
2746 	int i = order_base_2(size) - 10;
2747 	size_t slab_size;
2748 
2749 	if (size == PAGE_SIZE)
2750 		return 0;
2751 
2752 	if (i >= JBD2_MAX_SLABS)
2753 		return -EINVAL;
2754 
2755 	if (unlikely(i < 0))
2756 		i = 0;
2757 	mutex_lock(&jbd2_slab_create_mutex);
2758 	if (jbd2_slab[i]) {
2759 		mutex_unlock(&jbd2_slab_create_mutex);
2760 		return 0;	/* Already created */
2761 	}
2762 
2763 	slab_size = 1 << (i+10);
2764 	jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
2765 					 slab_size, 0, NULL);
2766 	mutex_unlock(&jbd2_slab_create_mutex);
2767 	if (!jbd2_slab[i]) {
2768 		printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
2769 		return -ENOMEM;
2770 	}
2771 	return 0;
2772 }
2773 
get_slab(size_t size)2774 static struct kmem_cache *get_slab(size_t size)
2775 {
2776 	int i = order_base_2(size) - 10;
2777 
2778 	BUG_ON(i >= JBD2_MAX_SLABS);
2779 	if (unlikely(i < 0))
2780 		i = 0;
2781 	BUG_ON(jbd2_slab[i] == NULL);
2782 	return jbd2_slab[i];
2783 }
2784 
jbd2_alloc(size_t size,gfp_t flags)2785 void *jbd2_alloc(size_t size, gfp_t flags)
2786 {
2787 	void *ptr;
2788 
2789 	BUG_ON(size & (size-1)); /* Must be a power of 2 */
2790 
2791 	if (size < PAGE_SIZE)
2792 		ptr = kmem_cache_alloc(get_slab(size), flags);
2793 	else
2794 		ptr = (void *)__get_free_pages(flags, get_order(size));
2795 
2796 	/* Check alignment; SLUB has gotten this wrong in the past,
2797 	 * and this can lead to user data corruption! */
2798 	BUG_ON(((unsigned long) ptr) & (size-1));
2799 
2800 	return ptr;
2801 }
2802 
jbd2_free(void * ptr,size_t size)2803 void jbd2_free(void *ptr, size_t size)
2804 {
2805 	if (size < PAGE_SIZE)
2806 		kmem_cache_free(get_slab(size), ptr);
2807 	else
2808 		free_pages((unsigned long)ptr, get_order(size));
2809 };
2810 
2811 /*
2812  * Journal_head storage management
2813  */
2814 static struct kmem_cache *jbd2_journal_head_cache;
2815 #ifdef CONFIG_JBD2_DEBUG
2816 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2817 #endif
2818 
jbd2_journal_init_journal_head_cache(void)2819 static int __init jbd2_journal_init_journal_head_cache(void)
2820 {
2821 	J_ASSERT(!jbd2_journal_head_cache);
2822 	jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
2823 				sizeof(struct journal_head),
2824 				0,		/* offset */
2825 				SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU,
2826 				NULL);		/* ctor */
2827 	if (!jbd2_journal_head_cache) {
2828 		printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
2829 		return -ENOMEM;
2830 	}
2831 	return 0;
2832 }
2833 
jbd2_journal_destroy_journal_head_cache(void)2834 static void jbd2_journal_destroy_journal_head_cache(void)
2835 {
2836 	kmem_cache_destroy(jbd2_journal_head_cache);
2837 	jbd2_journal_head_cache = NULL;
2838 }
2839 
2840 /*
2841  * journal_head splicing and dicing
2842  */
journal_alloc_journal_head(void)2843 static struct journal_head *journal_alloc_journal_head(void)
2844 {
2845 	struct journal_head *ret;
2846 
2847 #ifdef CONFIG_JBD2_DEBUG
2848 	atomic_inc(&nr_journal_heads);
2849 #endif
2850 	ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS);
2851 	if (!ret) {
2852 		jbd2_debug(1, "out of memory for journal_head\n");
2853 		pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
2854 		ret = kmem_cache_zalloc(jbd2_journal_head_cache,
2855 				GFP_NOFS | __GFP_NOFAIL);
2856 	}
2857 	spin_lock_init(&ret->b_state_lock);
2858 	return ret;
2859 }
2860 
journal_free_journal_head(struct journal_head * jh)2861 static void journal_free_journal_head(struct journal_head *jh)
2862 {
2863 #ifdef CONFIG_JBD2_DEBUG
2864 	atomic_dec(&nr_journal_heads);
2865 	memset(jh, JBD2_POISON_FREE, sizeof(*jh));
2866 #endif
2867 	kmem_cache_free(jbd2_journal_head_cache, jh);
2868 }
2869 
2870 /*
2871  * A journal_head is attached to a buffer_head whenever JBD has an
2872  * interest in the buffer.
2873  *
2874  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2875  * is set.  This bit is tested in core kernel code where we need to take
2876  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
2877  * there.
2878  *
2879  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2880  *
2881  * When a buffer has its BH_JBD bit set it is immune from being released by
2882  * core kernel code, mainly via ->b_count.
2883  *
2884  * A journal_head is detached from its buffer_head when the journal_head's
2885  * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2886  * transaction (b_cp_transaction) hold their references to b_jcount.
2887  *
2888  * Various places in the kernel want to attach a journal_head to a buffer_head
2889  * _before_ attaching the journal_head to a transaction.  To protect the
2890  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
2891  * journal_head's b_jcount refcount by one.  The caller must call
2892  * jbd2_journal_put_journal_head() to undo this.
2893  *
2894  * So the typical usage would be:
2895  *
2896  *	(Attach a journal_head if needed.  Increments b_jcount)
2897  *	struct journal_head *jh = jbd2_journal_add_journal_head(bh);
2898  *	...
2899  *      (Get another reference for transaction)
2900  *	jbd2_journal_grab_journal_head(bh);
2901  *	jh->b_transaction = xxx;
2902  *	(Put original reference)
2903  *	jbd2_journal_put_journal_head(jh);
2904  */
2905 
2906 /*
2907  * Give a buffer_head a journal_head.
2908  *
2909  * May sleep.
2910  */
jbd2_journal_add_journal_head(struct buffer_head * bh)2911 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
2912 {
2913 	struct journal_head *jh;
2914 	struct journal_head *new_jh = NULL;
2915 
2916 repeat:
2917 	if (!buffer_jbd(bh))
2918 		new_jh = journal_alloc_journal_head();
2919 
2920 	jbd_lock_bh_journal_head(bh);
2921 	if (buffer_jbd(bh)) {
2922 		jh = bh2jh(bh);
2923 	} else {
2924 		J_ASSERT_BH(bh,
2925 			(atomic_read(&bh->b_count) > 0) ||
2926 			(bh->b_folio && bh->b_folio->mapping));
2927 
2928 		if (!new_jh) {
2929 			jbd_unlock_bh_journal_head(bh);
2930 			goto repeat;
2931 		}
2932 
2933 		jh = new_jh;
2934 		new_jh = NULL;		/* We consumed it */
2935 		set_buffer_jbd(bh);
2936 		bh->b_private = jh;
2937 		jh->b_bh = bh;
2938 		get_bh(bh);
2939 		BUFFER_TRACE(bh, "added journal_head");
2940 	}
2941 	jh->b_jcount++;
2942 	jbd_unlock_bh_journal_head(bh);
2943 	if (new_jh)
2944 		journal_free_journal_head(new_jh);
2945 	return bh->b_private;
2946 }
2947 
2948 /*
2949  * Grab a ref against this buffer_head's journal_head.  If it ended up not
2950  * having a journal_head, return NULL
2951  */
jbd2_journal_grab_journal_head(struct buffer_head * bh)2952 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
2953 {
2954 	struct journal_head *jh = NULL;
2955 
2956 	jbd_lock_bh_journal_head(bh);
2957 	if (buffer_jbd(bh)) {
2958 		jh = bh2jh(bh);
2959 		jh->b_jcount++;
2960 	}
2961 	jbd_unlock_bh_journal_head(bh);
2962 	return jh;
2963 }
2964 EXPORT_SYMBOL(jbd2_journal_grab_journal_head);
2965 
__journal_remove_journal_head(struct buffer_head * bh)2966 static void __journal_remove_journal_head(struct buffer_head *bh)
2967 {
2968 	struct journal_head *jh = bh2jh(bh);
2969 
2970 	J_ASSERT_JH(jh, jh->b_transaction == NULL);
2971 	J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2972 	J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2973 	J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2974 	J_ASSERT_BH(bh, buffer_jbd(bh));
2975 	J_ASSERT_BH(bh, jh2bh(jh) == bh);
2976 	BUFFER_TRACE(bh, "remove journal_head");
2977 
2978 	/* Unlink before dropping the lock */
2979 	bh->b_private = NULL;
2980 	jh->b_bh = NULL;	/* debug, really */
2981 	clear_buffer_jbd(bh);
2982 }
2983 
journal_release_journal_head(struct journal_head * jh,size_t b_size)2984 static void journal_release_journal_head(struct journal_head *jh, size_t b_size)
2985 {
2986 	if (jh->b_frozen_data) {
2987 		printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2988 		jbd2_free(jh->b_frozen_data, b_size);
2989 	}
2990 	if (jh->b_committed_data) {
2991 		printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2992 		jbd2_free(jh->b_committed_data, b_size);
2993 	}
2994 	journal_free_journal_head(jh);
2995 }
2996 
2997 /*
2998  * Drop a reference on the passed journal_head.  If it fell to zero then
2999  * release the journal_head from the buffer_head.
3000  */
jbd2_journal_put_journal_head(struct journal_head * jh)3001 void jbd2_journal_put_journal_head(struct journal_head *jh)
3002 {
3003 	struct buffer_head *bh = jh2bh(jh);
3004 
3005 	jbd_lock_bh_journal_head(bh);
3006 	J_ASSERT_JH(jh, jh->b_jcount > 0);
3007 	--jh->b_jcount;
3008 	if (!jh->b_jcount) {
3009 		__journal_remove_journal_head(bh);
3010 		jbd_unlock_bh_journal_head(bh);
3011 		journal_release_journal_head(jh, bh->b_size);
3012 		__brelse(bh);
3013 	} else {
3014 		jbd_unlock_bh_journal_head(bh);
3015 	}
3016 }
3017 EXPORT_SYMBOL(jbd2_journal_put_journal_head);
3018 
3019 /*
3020  * Initialize jbd inode head
3021  */
jbd2_journal_init_jbd_inode(struct jbd2_inode * jinode,struct inode * inode)3022 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
3023 {
3024 	jinode->i_transaction = NULL;
3025 	jinode->i_next_transaction = NULL;
3026 	jinode->i_vfs_inode = inode;
3027 	jinode->i_flags = 0;
3028 	jinode->i_dirty_start = 0;
3029 	jinode->i_dirty_end = 0;
3030 	INIT_LIST_HEAD(&jinode->i_list);
3031 }
3032 
3033 /*
3034  * Function to be called before we start removing inode from memory (i.e.,
3035  * clear_inode() is a fine place to be called from). It removes inode from
3036  * transaction's lists.
3037  */
jbd2_journal_release_jbd_inode(journal_t * journal,struct jbd2_inode * jinode)3038 void jbd2_journal_release_jbd_inode(journal_t *journal,
3039 				    struct jbd2_inode *jinode)
3040 {
3041 	if (!journal)
3042 		return;
3043 restart:
3044 	spin_lock(&journal->j_list_lock);
3045 	/* Is commit writing out inode - we have to wait */
3046 	if (jinode->i_flags & JI_COMMIT_RUNNING) {
3047 		wait_queue_head_t *wq;
3048 		DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
3049 		wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
3050 		prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE);
3051 		spin_unlock(&journal->j_list_lock);
3052 		schedule();
3053 		finish_wait(wq, &wait.wq_entry);
3054 		goto restart;
3055 	}
3056 
3057 	if (jinode->i_transaction) {
3058 		list_del(&jinode->i_list);
3059 		jinode->i_transaction = NULL;
3060 	}
3061 	spin_unlock(&journal->j_list_lock);
3062 }
3063 
3064 
3065 #ifdef CONFIG_PROC_FS
3066 
3067 #define JBD2_STATS_PROC_NAME "fs/jbd2"
3068 
jbd2_create_jbd_stats_proc_entry(void)3069 static void __init jbd2_create_jbd_stats_proc_entry(void)
3070 {
3071 	proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
3072 }
3073 
jbd2_remove_jbd_stats_proc_entry(void)3074 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
3075 {
3076 	if (proc_jbd2_stats)
3077 		remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
3078 }
3079 
3080 #else
3081 
3082 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
3083 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
3084 
3085 #endif
3086 
3087 struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
3088 
jbd2_journal_init_inode_cache(void)3089 static int __init jbd2_journal_init_inode_cache(void)
3090 {
3091 	J_ASSERT(!jbd2_inode_cache);
3092 	jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
3093 	if (!jbd2_inode_cache) {
3094 		pr_emerg("JBD2: failed to create inode cache\n");
3095 		return -ENOMEM;
3096 	}
3097 	return 0;
3098 }
3099 
jbd2_journal_init_handle_cache(void)3100 static int __init jbd2_journal_init_handle_cache(void)
3101 {
3102 	J_ASSERT(!jbd2_handle_cache);
3103 	jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
3104 	if (!jbd2_handle_cache) {
3105 		printk(KERN_EMERG "JBD2: failed to create handle cache\n");
3106 		return -ENOMEM;
3107 	}
3108 	return 0;
3109 }
3110 
jbd2_journal_destroy_inode_cache(void)3111 static void jbd2_journal_destroy_inode_cache(void)
3112 {
3113 	kmem_cache_destroy(jbd2_inode_cache);
3114 	jbd2_inode_cache = NULL;
3115 }
3116 
jbd2_journal_destroy_handle_cache(void)3117 static void jbd2_journal_destroy_handle_cache(void)
3118 {
3119 	kmem_cache_destroy(jbd2_handle_cache);
3120 	jbd2_handle_cache = NULL;
3121 }
3122 
3123 /*
3124  * Module startup and shutdown
3125  */
3126 
journal_init_caches(void)3127 static int __init journal_init_caches(void)
3128 {
3129 	int ret;
3130 
3131 	ret = jbd2_journal_init_revoke_record_cache();
3132 	if (ret == 0)
3133 		ret = jbd2_journal_init_revoke_table_cache();
3134 	if (ret == 0)
3135 		ret = jbd2_journal_init_journal_head_cache();
3136 	if (ret == 0)
3137 		ret = jbd2_journal_init_handle_cache();
3138 	if (ret == 0)
3139 		ret = jbd2_journal_init_inode_cache();
3140 	if (ret == 0)
3141 		ret = jbd2_journal_init_transaction_cache();
3142 	return ret;
3143 }
3144 
jbd2_journal_destroy_caches(void)3145 static void jbd2_journal_destroy_caches(void)
3146 {
3147 	jbd2_journal_destroy_revoke_record_cache();
3148 	jbd2_journal_destroy_revoke_table_cache();
3149 	jbd2_journal_destroy_journal_head_cache();
3150 	jbd2_journal_destroy_handle_cache();
3151 	jbd2_journal_destroy_inode_cache();
3152 	jbd2_journal_destroy_transaction_cache();
3153 	jbd2_journal_destroy_slabs();
3154 }
3155 
journal_init(void)3156 static int __init journal_init(void)
3157 {
3158 	int ret;
3159 
3160 	BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
3161 
3162 	ret = journal_init_caches();
3163 	if (ret == 0) {
3164 		jbd2_create_jbd_stats_proc_entry();
3165 	} else {
3166 		jbd2_journal_destroy_caches();
3167 	}
3168 	return ret;
3169 }
3170 
journal_exit(void)3171 static void __exit journal_exit(void)
3172 {
3173 #ifdef CONFIG_JBD2_DEBUG
3174 	int n = atomic_read(&nr_journal_heads);
3175 	if (n)
3176 		printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n);
3177 #endif
3178 	jbd2_remove_jbd_stats_proc_entry();
3179 	jbd2_journal_destroy_caches();
3180 }
3181 
3182 MODULE_DESCRIPTION("Generic filesystem journal-writing module");
3183 MODULE_LICENSE("GPL");
3184 module_init(journal_init);
3185 module_exit(journal_exit);
3186 
3187