xref: /openbmc/linux/fs/jbd2/journal.c (revision e8e0929d)
1 /*
2  * linux/fs/jbd2/journal.c
3  *
4  * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5  *
6  * Copyright 1998 Red Hat corp --- All Rights Reserved
7  *
8  * This file is part of the Linux kernel and is made available under
9  * the terms of the GNU General Public License, version 2, or at your
10  * option, any later version, incorporated herein by reference.
11  *
12  * Generic filesystem journal-writing code; part of the ext2fs
13  * journaling system.
14  *
15  * This file manages journals: areas of disk reserved for logging
16  * transactional updates.  This includes the kernel journaling thread
17  * which is responsible for scheduling updates to the log.
18  *
19  * We do not actually manage the physical storage of the journal in this
20  * file: that is left to a per-journal policy function, which allows us
21  * to store the journal within a filesystem-specified area for ext2
22  * journaling (ext2 can use a reserved inode for storing the log).
23  */
24 
25 #include <linux/module.h>
26 #include <linux/time.h>
27 #include <linux/fs.h>
28 #include <linux/jbd2.h>
29 #include <linux/errno.h>
30 #include <linux/slab.h>
31 #include <linux/init.h>
32 #include <linux/mm.h>
33 #include <linux/freezer.h>
34 #include <linux/pagemap.h>
35 #include <linux/kthread.h>
36 #include <linux/poison.h>
37 #include <linux/proc_fs.h>
38 #include <linux/debugfs.h>
39 #include <linux/seq_file.h>
40 #include <linux/math64.h>
41 #include <linux/hash.h>
42 
43 #define CREATE_TRACE_POINTS
44 #include <trace/events/jbd2.h>
45 
46 #include <asm/uaccess.h>
47 #include <asm/page.h>
48 
49 EXPORT_SYMBOL(jbd2_journal_start);
50 EXPORT_SYMBOL(jbd2_journal_restart);
51 EXPORT_SYMBOL(jbd2_journal_extend);
52 EXPORT_SYMBOL(jbd2_journal_stop);
53 EXPORT_SYMBOL(jbd2_journal_lock_updates);
54 EXPORT_SYMBOL(jbd2_journal_unlock_updates);
55 EXPORT_SYMBOL(jbd2_journal_get_write_access);
56 EXPORT_SYMBOL(jbd2_journal_get_create_access);
57 EXPORT_SYMBOL(jbd2_journal_get_undo_access);
58 EXPORT_SYMBOL(jbd2_journal_set_triggers);
59 EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
60 EXPORT_SYMBOL(jbd2_journal_release_buffer);
61 EXPORT_SYMBOL(jbd2_journal_forget);
62 #if 0
63 EXPORT_SYMBOL(journal_sync_buffer);
64 #endif
65 EXPORT_SYMBOL(jbd2_journal_flush);
66 EXPORT_SYMBOL(jbd2_journal_revoke);
67 
68 EXPORT_SYMBOL(jbd2_journal_init_dev);
69 EXPORT_SYMBOL(jbd2_journal_init_inode);
70 EXPORT_SYMBOL(jbd2_journal_update_format);
71 EXPORT_SYMBOL(jbd2_journal_check_used_features);
72 EXPORT_SYMBOL(jbd2_journal_check_available_features);
73 EXPORT_SYMBOL(jbd2_journal_set_features);
74 EXPORT_SYMBOL(jbd2_journal_load);
75 EXPORT_SYMBOL(jbd2_journal_destroy);
76 EXPORT_SYMBOL(jbd2_journal_abort);
77 EXPORT_SYMBOL(jbd2_journal_errno);
78 EXPORT_SYMBOL(jbd2_journal_ack_err);
79 EXPORT_SYMBOL(jbd2_journal_clear_err);
80 EXPORT_SYMBOL(jbd2_log_wait_commit);
81 EXPORT_SYMBOL(jbd2_journal_start_commit);
82 EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
83 EXPORT_SYMBOL(jbd2_journal_wipe);
84 EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
85 EXPORT_SYMBOL(jbd2_journal_invalidatepage);
86 EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
87 EXPORT_SYMBOL(jbd2_journal_force_commit);
88 EXPORT_SYMBOL(jbd2_journal_file_inode);
89 EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
90 EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
91 EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
92 
93 static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
94 static void __journal_abort_soft (journal_t *journal, int errno);
95 
96 /*
97  * Helper function used to manage commit timeouts
98  */
99 
100 static void commit_timeout(unsigned long __data)
101 {
102 	struct task_struct * p = (struct task_struct *) __data;
103 
104 	wake_up_process(p);
105 }
106 
107 /*
108  * kjournald2: The main thread function used to manage a logging device
109  * journal.
110  *
111  * This kernel thread is responsible for two things:
112  *
113  * 1) COMMIT:  Every so often we need to commit the current state of the
114  *    filesystem to disk.  The journal thread is responsible for writing
115  *    all of the metadata buffers to disk.
116  *
117  * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
118  *    of the data in that part of the log has been rewritten elsewhere on
119  *    the disk.  Flushing these old buffers to reclaim space in the log is
120  *    known as checkpointing, and this thread is responsible for that job.
121  */
122 
123 static int kjournald2(void *arg)
124 {
125 	journal_t *journal = arg;
126 	transaction_t *transaction;
127 
128 	/*
129 	 * Set up an interval timer which can be used to trigger a commit wakeup
130 	 * after the commit interval expires
131 	 */
132 	setup_timer(&journal->j_commit_timer, commit_timeout,
133 			(unsigned long)current);
134 
135 	/* Record that the journal thread is running */
136 	journal->j_task = current;
137 	wake_up(&journal->j_wait_done_commit);
138 
139 	/*
140 	 * And now, wait forever for commit wakeup events.
141 	 */
142 	spin_lock(&journal->j_state_lock);
143 
144 loop:
145 	if (journal->j_flags & JBD2_UNMOUNT)
146 		goto end_loop;
147 
148 	jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
149 		journal->j_commit_sequence, journal->j_commit_request);
150 
151 	if (journal->j_commit_sequence != journal->j_commit_request) {
152 		jbd_debug(1, "OK, requests differ\n");
153 		spin_unlock(&journal->j_state_lock);
154 		del_timer_sync(&journal->j_commit_timer);
155 		jbd2_journal_commit_transaction(journal);
156 		spin_lock(&journal->j_state_lock);
157 		goto loop;
158 	}
159 
160 	wake_up(&journal->j_wait_done_commit);
161 	if (freezing(current)) {
162 		/*
163 		 * The simpler the better. Flushing journal isn't a
164 		 * good idea, because that depends on threads that may
165 		 * be already stopped.
166 		 */
167 		jbd_debug(1, "Now suspending kjournald2\n");
168 		spin_unlock(&journal->j_state_lock);
169 		refrigerator();
170 		spin_lock(&journal->j_state_lock);
171 	} else {
172 		/*
173 		 * We assume on resume that commits are already there,
174 		 * so we don't sleep
175 		 */
176 		DEFINE_WAIT(wait);
177 		int should_sleep = 1;
178 
179 		prepare_to_wait(&journal->j_wait_commit, &wait,
180 				TASK_INTERRUPTIBLE);
181 		if (journal->j_commit_sequence != journal->j_commit_request)
182 			should_sleep = 0;
183 		transaction = journal->j_running_transaction;
184 		if (transaction && time_after_eq(jiffies,
185 						transaction->t_expires))
186 			should_sleep = 0;
187 		if (journal->j_flags & JBD2_UNMOUNT)
188 			should_sleep = 0;
189 		if (should_sleep) {
190 			spin_unlock(&journal->j_state_lock);
191 			schedule();
192 			spin_lock(&journal->j_state_lock);
193 		}
194 		finish_wait(&journal->j_wait_commit, &wait);
195 	}
196 
197 	jbd_debug(1, "kjournald2 wakes\n");
198 
199 	/*
200 	 * Were we woken up by a commit wakeup event?
201 	 */
202 	transaction = journal->j_running_transaction;
203 	if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
204 		journal->j_commit_request = transaction->t_tid;
205 		jbd_debug(1, "woke because of timeout\n");
206 	}
207 	goto loop;
208 
209 end_loop:
210 	spin_unlock(&journal->j_state_lock);
211 	del_timer_sync(&journal->j_commit_timer);
212 	journal->j_task = NULL;
213 	wake_up(&journal->j_wait_done_commit);
214 	jbd_debug(1, "Journal thread exiting.\n");
215 	return 0;
216 }
217 
218 static int jbd2_journal_start_thread(journal_t *journal)
219 {
220 	struct task_struct *t;
221 
222 	t = kthread_run(kjournald2, journal, "jbd2/%s",
223 			journal->j_devname);
224 	if (IS_ERR(t))
225 		return PTR_ERR(t);
226 
227 	wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
228 	return 0;
229 }
230 
231 static void journal_kill_thread(journal_t *journal)
232 {
233 	spin_lock(&journal->j_state_lock);
234 	journal->j_flags |= JBD2_UNMOUNT;
235 
236 	while (journal->j_task) {
237 		wake_up(&journal->j_wait_commit);
238 		spin_unlock(&journal->j_state_lock);
239 		wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
240 		spin_lock(&journal->j_state_lock);
241 	}
242 	spin_unlock(&journal->j_state_lock);
243 }
244 
245 /*
246  * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
247  *
248  * Writes a metadata buffer to a given disk block.  The actual IO is not
249  * performed but a new buffer_head is constructed which labels the data
250  * to be written with the correct destination disk block.
251  *
252  * Any magic-number escaping which needs to be done will cause a
253  * copy-out here.  If the buffer happens to start with the
254  * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
255  * magic number is only written to the log for descripter blocks.  In
256  * this case, we copy the data and replace the first word with 0, and we
257  * return a result code which indicates that this buffer needs to be
258  * marked as an escaped buffer in the corresponding log descriptor
259  * block.  The missing word can then be restored when the block is read
260  * during recovery.
261  *
262  * If the source buffer has already been modified by a new transaction
263  * since we took the last commit snapshot, we use the frozen copy of
264  * that data for IO.  If we end up using the existing buffer_head's data
265  * for the write, then we *have* to lock the buffer to prevent anyone
266  * else from using and possibly modifying it while the IO is in
267  * progress.
268  *
269  * The function returns a pointer to the buffer_heads to be used for IO.
270  *
271  * We assume that the journal has already been locked in this function.
272  *
273  * Return value:
274  *  <0: Error
275  * >=0: Finished OK
276  *
277  * On success:
278  * Bit 0 set == escape performed on the data
279  * Bit 1 set == buffer copy-out performed (kfree the data after IO)
280  */
281 
282 int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
283 				  struct journal_head  *jh_in,
284 				  struct journal_head **jh_out,
285 				  unsigned long long blocknr)
286 {
287 	int need_copy_out = 0;
288 	int done_copy_out = 0;
289 	int do_escape = 0;
290 	char *mapped_data;
291 	struct buffer_head *new_bh;
292 	struct journal_head *new_jh;
293 	struct page *new_page;
294 	unsigned int new_offset;
295 	struct buffer_head *bh_in = jh2bh(jh_in);
296 	struct jbd2_buffer_trigger_type *triggers;
297 	journal_t *journal = transaction->t_journal;
298 
299 	/*
300 	 * The buffer really shouldn't be locked: only the current committing
301 	 * transaction is allowed to write it, so nobody else is allowed
302 	 * to do any IO.
303 	 *
304 	 * akpm: except if we're journalling data, and write() output is
305 	 * also part of a shared mapping, and another thread has
306 	 * decided to launch a writepage() against this buffer.
307 	 */
308 	J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
309 
310 	new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL);
311 	/* keep subsequent assertions sane */
312 	new_bh->b_state = 0;
313 	init_buffer(new_bh, NULL, NULL);
314 	atomic_set(&new_bh->b_count, 1);
315 	new_jh = jbd2_journal_add_journal_head(new_bh);	/* This sleeps */
316 
317 	/*
318 	 * If a new transaction has already done a buffer copy-out, then
319 	 * we use that version of the data for the commit.
320 	 */
321 	jbd_lock_bh_state(bh_in);
322 repeat:
323 	if (jh_in->b_frozen_data) {
324 		done_copy_out = 1;
325 		new_page = virt_to_page(jh_in->b_frozen_data);
326 		new_offset = offset_in_page(jh_in->b_frozen_data);
327 		triggers = jh_in->b_frozen_triggers;
328 	} else {
329 		new_page = jh2bh(jh_in)->b_page;
330 		new_offset = offset_in_page(jh2bh(jh_in)->b_data);
331 		triggers = jh_in->b_triggers;
332 	}
333 
334 	mapped_data = kmap_atomic(new_page, KM_USER0);
335 	/*
336 	 * Fire any commit trigger.  Do this before checking for escaping,
337 	 * as the trigger may modify the magic offset.  If a copy-out
338 	 * happens afterwards, it will have the correct data in the buffer.
339 	 */
340 	jbd2_buffer_commit_trigger(jh_in, mapped_data + new_offset,
341 				   triggers);
342 
343 	/*
344 	 * Check for escaping
345 	 */
346 	if (*((__be32 *)(mapped_data + new_offset)) ==
347 				cpu_to_be32(JBD2_MAGIC_NUMBER)) {
348 		need_copy_out = 1;
349 		do_escape = 1;
350 	}
351 	kunmap_atomic(mapped_data, KM_USER0);
352 
353 	/*
354 	 * Do we need to do a data copy?
355 	 */
356 	if (need_copy_out && !done_copy_out) {
357 		char *tmp;
358 
359 		jbd_unlock_bh_state(bh_in);
360 		tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
361 		jbd_lock_bh_state(bh_in);
362 		if (jh_in->b_frozen_data) {
363 			jbd2_free(tmp, bh_in->b_size);
364 			goto repeat;
365 		}
366 
367 		jh_in->b_frozen_data = tmp;
368 		mapped_data = kmap_atomic(new_page, KM_USER0);
369 		memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
370 		kunmap_atomic(mapped_data, KM_USER0);
371 
372 		new_page = virt_to_page(tmp);
373 		new_offset = offset_in_page(tmp);
374 		done_copy_out = 1;
375 
376 		/*
377 		 * This isn't strictly necessary, as we're using frozen
378 		 * data for the escaping, but it keeps consistency with
379 		 * b_frozen_data usage.
380 		 */
381 		jh_in->b_frozen_triggers = jh_in->b_triggers;
382 	}
383 
384 	/*
385 	 * Did we need to do an escaping?  Now we've done all the
386 	 * copying, we can finally do so.
387 	 */
388 	if (do_escape) {
389 		mapped_data = kmap_atomic(new_page, KM_USER0);
390 		*((unsigned int *)(mapped_data + new_offset)) = 0;
391 		kunmap_atomic(mapped_data, KM_USER0);
392 	}
393 
394 	set_bh_page(new_bh, new_page, new_offset);
395 	new_jh->b_transaction = NULL;
396 	new_bh->b_size = jh2bh(jh_in)->b_size;
397 	new_bh->b_bdev = transaction->t_journal->j_dev;
398 	new_bh->b_blocknr = blocknr;
399 	set_buffer_mapped(new_bh);
400 	set_buffer_dirty(new_bh);
401 
402 	*jh_out = new_jh;
403 
404 	/*
405 	 * The to-be-written buffer needs to get moved to the io queue,
406 	 * and the original buffer whose contents we are shadowing or
407 	 * copying is moved to the transaction's shadow queue.
408 	 */
409 	JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
410 	spin_lock(&journal->j_list_lock);
411 	__jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
412 	spin_unlock(&journal->j_list_lock);
413 	jbd_unlock_bh_state(bh_in);
414 
415 	JBUFFER_TRACE(new_jh, "file as BJ_IO");
416 	jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
417 
418 	return do_escape | (done_copy_out << 1);
419 }
420 
421 /*
422  * Allocation code for the journal file.  Manage the space left in the
423  * journal, so that we can begin checkpointing when appropriate.
424  */
425 
426 /*
427  * __jbd2_log_space_left: Return the number of free blocks left in the journal.
428  *
429  * Called with the journal already locked.
430  *
431  * Called under j_state_lock
432  */
433 
434 int __jbd2_log_space_left(journal_t *journal)
435 {
436 	int left = journal->j_free;
437 
438 	assert_spin_locked(&journal->j_state_lock);
439 
440 	/*
441 	 * Be pessimistic here about the number of those free blocks which
442 	 * might be required for log descriptor control blocks.
443 	 */
444 
445 #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
446 
447 	left -= MIN_LOG_RESERVED_BLOCKS;
448 
449 	if (left <= 0)
450 		return 0;
451 	left -= (left >> 3);
452 	return left;
453 }
454 
455 /*
456  * Called under j_state_lock.  Returns true if a transaction commit was started.
457  */
458 int __jbd2_log_start_commit(journal_t *journal, tid_t target)
459 {
460 	/*
461 	 * Are we already doing a recent enough commit?
462 	 */
463 	if (!tid_geq(journal->j_commit_request, target)) {
464 		/*
465 		 * We want a new commit: OK, mark the request and wakup the
466 		 * commit thread.  We do _not_ do the commit ourselves.
467 		 */
468 
469 		journal->j_commit_request = target;
470 		jbd_debug(1, "JBD: requesting commit %d/%d\n",
471 			  journal->j_commit_request,
472 			  journal->j_commit_sequence);
473 		wake_up(&journal->j_wait_commit);
474 		return 1;
475 	}
476 	return 0;
477 }
478 
479 int jbd2_log_start_commit(journal_t *journal, tid_t tid)
480 {
481 	int ret;
482 
483 	spin_lock(&journal->j_state_lock);
484 	ret = __jbd2_log_start_commit(journal, tid);
485 	spin_unlock(&journal->j_state_lock);
486 	return ret;
487 }
488 
489 /*
490  * Force and wait upon a commit if the calling process is not within
491  * transaction.  This is used for forcing out undo-protected data which contains
492  * bitmaps, when the fs is running out of space.
493  *
494  * We can only force the running transaction if we don't have an active handle;
495  * otherwise, we will deadlock.
496  *
497  * Returns true if a transaction was started.
498  */
499 int jbd2_journal_force_commit_nested(journal_t *journal)
500 {
501 	transaction_t *transaction = NULL;
502 	tid_t tid;
503 
504 	spin_lock(&journal->j_state_lock);
505 	if (journal->j_running_transaction && !current->journal_info) {
506 		transaction = journal->j_running_transaction;
507 		__jbd2_log_start_commit(journal, transaction->t_tid);
508 	} else if (journal->j_committing_transaction)
509 		transaction = journal->j_committing_transaction;
510 
511 	if (!transaction) {
512 		spin_unlock(&journal->j_state_lock);
513 		return 0;	/* Nothing to retry */
514 	}
515 
516 	tid = transaction->t_tid;
517 	spin_unlock(&journal->j_state_lock);
518 	jbd2_log_wait_commit(journal, tid);
519 	return 1;
520 }
521 
522 /*
523  * Start a commit of the current running transaction (if any).  Returns true
524  * if a transaction is going to be committed (or is currently already
525  * committing), and fills its tid in at *ptid
526  */
527 int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
528 {
529 	int ret = 0;
530 
531 	spin_lock(&journal->j_state_lock);
532 	if (journal->j_running_transaction) {
533 		tid_t tid = journal->j_running_transaction->t_tid;
534 
535 		__jbd2_log_start_commit(journal, tid);
536 		/* There's a running transaction and we've just made sure
537 		 * it's commit has been scheduled. */
538 		if (ptid)
539 			*ptid = tid;
540 		ret = 1;
541 	} else if (journal->j_committing_transaction) {
542 		/*
543 		 * If ext3_write_super() recently started a commit, then we
544 		 * have to wait for completion of that transaction
545 		 */
546 		if (ptid)
547 			*ptid = journal->j_committing_transaction->t_tid;
548 		ret = 1;
549 	}
550 	spin_unlock(&journal->j_state_lock);
551 	return ret;
552 }
553 
554 /*
555  * Wait for a specified commit to complete.
556  * The caller may not hold the journal lock.
557  */
558 int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
559 {
560 	int err = 0;
561 
562 #ifdef CONFIG_JBD2_DEBUG
563 	spin_lock(&journal->j_state_lock);
564 	if (!tid_geq(journal->j_commit_request, tid)) {
565 		printk(KERN_EMERG
566 		       "%s: error: j_commit_request=%d, tid=%d\n",
567 		       __func__, journal->j_commit_request, tid);
568 	}
569 	spin_unlock(&journal->j_state_lock);
570 #endif
571 	spin_lock(&journal->j_state_lock);
572 	while (tid_gt(tid, journal->j_commit_sequence)) {
573 		jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n",
574 				  tid, journal->j_commit_sequence);
575 		wake_up(&journal->j_wait_commit);
576 		spin_unlock(&journal->j_state_lock);
577 		wait_event(journal->j_wait_done_commit,
578 				!tid_gt(tid, journal->j_commit_sequence));
579 		spin_lock(&journal->j_state_lock);
580 	}
581 	spin_unlock(&journal->j_state_lock);
582 
583 	if (unlikely(is_journal_aborted(journal))) {
584 		printk(KERN_EMERG "journal commit I/O error\n");
585 		err = -EIO;
586 	}
587 	return err;
588 }
589 
590 /*
591  * Log buffer allocation routines:
592  */
593 
594 int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
595 {
596 	unsigned long blocknr;
597 
598 	spin_lock(&journal->j_state_lock);
599 	J_ASSERT(journal->j_free > 1);
600 
601 	blocknr = journal->j_head;
602 	journal->j_head++;
603 	journal->j_free--;
604 	if (journal->j_head == journal->j_last)
605 		journal->j_head = journal->j_first;
606 	spin_unlock(&journal->j_state_lock);
607 	return jbd2_journal_bmap(journal, blocknr, retp);
608 }
609 
610 /*
611  * Conversion of logical to physical block numbers for the journal
612  *
613  * On external journals the journal blocks are identity-mapped, so
614  * this is a no-op.  If needed, we can use j_blk_offset - everything is
615  * ready.
616  */
617 int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
618 		 unsigned long long *retp)
619 {
620 	int err = 0;
621 	unsigned long long ret;
622 
623 	if (journal->j_inode) {
624 		ret = bmap(journal->j_inode, blocknr);
625 		if (ret)
626 			*retp = ret;
627 		else {
628 			printk(KERN_ALERT "%s: journal block not found "
629 					"at offset %lu on %s\n",
630 			       __func__, blocknr, journal->j_devname);
631 			err = -EIO;
632 			__journal_abort_soft(journal, err);
633 		}
634 	} else {
635 		*retp = blocknr; /* +journal->j_blk_offset */
636 	}
637 	return err;
638 }
639 
640 /*
641  * We play buffer_head aliasing tricks to write data/metadata blocks to
642  * the journal without copying their contents, but for journal
643  * descriptor blocks we do need to generate bona fide buffers.
644  *
645  * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
646  * the buffer's contents they really should run flush_dcache_page(bh->b_page).
647  * But we don't bother doing that, so there will be coherency problems with
648  * mmaps of blockdevs which hold live JBD-controlled filesystems.
649  */
650 struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
651 {
652 	struct buffer_head *bh;
653 	unsigned long long blocknr;
654 	int err;
655 
656 	err = jbd2_journal_next_log_block(journal, &blocknr);
657 
658 	if (err)
659 		return NULL;
660 
661 	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
662 	if (!bh)
663 		return NULL;
664 	lock_buffer(bh);
665 	memset(bh->b_data, 0, journal->j_blocksize);
666 	set_buffer_uptodate(bh);
667 	unlock_buffer(bh);
668 	BUFFER_TRACE(bh, "return this buffer");
669 	return jbd2_journal_add_journal_head(bh);
670 }
671 
672 struct jbd2_stats_proc_session {
673 	journal_t *journal;
674 	struct transaction_stats_s *stats;
675 	int start;
676 	int max;
677 };
678 
679 static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
680 {
681 	return *pos ? NULL : SEQ_START_TOKEN;
682 }
683 
684 static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
685 {
686 	return NULL;
687 }
688 
689 static int jbd2_seq_info_show(struct seq_file *seq, void *v)
690 {
691 	struct jbd2_stats_proc_session *s = seq->private;
692 
693 	if (v != SEQ_START_TOKEN)
694 		return 0;
695 	seq_printf(seq, "%lu transaction, each up to %u blocks\n",
696 			s->stats->ts_tid,
697 			s->journal->j_max_transaction_buffers);
698 	if (s->stats->ts_tid == 0)
699 		return 0;
700 	seq_printf(seq, "average: \n  %ums waiting for transaction\n",
701 	    jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
702 	seq_printf(seq, "  %ums running transaction\n",
703 	    jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
704 	seq_printf(seq, "  %ums transaction was being locked\n",
705 	    jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
706 	seq_printf(seq, "  %ums flushing data (in ordered mode)\n",
707 	    jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
708 	seq_printf(seq, "  %ums logging transaction\n",
709 	    jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
710 	seq_printf(seq, "  %lluus average transaction commit time\n",
711 		   div_u64(s->journal->j_average_commit_time, 1000));
712 	seq_printf(seq, "  %lu handles per transaction\n",
713 	    s->stats->run.rs_handle_count / s->stats->ts_tid);
714 	seq_printf(seq, "  %lu blocks per transaction\n",
715 	    s->stats->run.rs_blocks / s->stats->ts_tid);
716 	seq_printf(seq, "  %lu logged blocks per transaction\n",
717 	    s->stats->run.rs_blocks_logged / s->stats->ts_tid);
718 	return 0;
719 }
720 
721 static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
722 {
723 }
724 
725 static const struct seq_operations jbd2_seq_info_ops = {
726 	.start  = jbd2_seq_info_start,
727 	.next   = jbd2_seq_info_next,
728 	.stop   = jbd2_seq_info_stop,
729 	.show   = jbd2_seq_info_show,
730 };
731 
732 static int jbd2_seq_info_open(struct inode *inode, struct file *file)
733 {
734 	journal_t *journal = PDE(inode)->data;
735 	struct jbd2_stats_proc_session *s;
736 	int rc, size;
737 
738 	s = kmalloc(sizeof(*s), GFP_KERNEL);
739 	if (s == NULL)
740 		return -ENOMEM;
741 	size = sizeof(struct transaction_stats_s);
742 	s->stats = kmalloc(size, GFP_KERNEL);
743 	if (s->stats == NULL) {
744 		kfree(s);
745 		return -ENOMEM;
746 	}
747 	spin_lock(&journal->j_history_lock);
748 	memcpy(s->stats, &journal->j_stats, size);
749 	s->journal = journal;
750 	spin_unlock(&journal->j_history_lock);
751 
752 	rc = seq_open(file, &jbd2_seq_info_ops);
753 	if (rc == 0) {
754 		struct seq_file *m = file->private_data;
755 		m->private = s;
756 	} else {
757 		kfree(s->stats);
758 		kfree(s);
759 	}
760 	return rc;
761 
762 }
763 
764 static int jbd2_seq_info_release(struct inode *inode, struct file *file)
765 {
766 	struct seq_file *seq = file->private_data;
767 	struct jbd2_stats_proc_session *s = seq->private;
768 	kfree(s->stats);
769 	kfree(s);
770 	return seq_release(inode, file);
771 }
772 
773 static const struct file_operations jbd2_seq_info_fops = {
774 	.owner		= THIS_MODULE,
775 	.open           = jbd2_seq_info_open,
776 	.read           = seq_read,
777 	.llseek         = seq_lseek,
778 	.release        = jbd2_seq_info_release,
779 };
780 
781 static struct proc_dir_entry *proc_jbd2_stats;
782 
783 static void jbd2_stats_proc_init(journal_t *journal)
784 {
785 	journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
786 	if (journal->j_proc_entry) {
787 		proc_create_data("info", S_IRUGO, journal->j_proc_entry,
788 				 &jbd2_seq_info_fops, journal);
789 	}
790 }
791 
792 static void jbd2_stats_proc_exit(journal_t *journal)
793 {
794 	remove_proc_entry("info", journal->j_proc_entry);
795 	remove_proc_entry(journal->j_devname, proc_jbd2_stats);
796 }
797 
798 /*
799  * Management for journal control blocks: functions to create and
800  * destroy journal_t structures, and to initialise and read existing
801  * journal blocks from disk.  */
802 
803 /* First: create and setup a journal_t object in memory.  We initialise
804  * very few fields yet: that has to wait until we have created the
805  * journal structures from from scratch, or loaded them from disk. */
806 
807 static journal_t * journal_init_common (void)
808 {
809 	journal_t *journal;
810 	int err;
811 
812 	journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL);
813 	if (!journal)
814 		goto fail;
815 
816 	init_waitqueue_head(&journal->j_wait_transaction_locked);
817 	init_waitqueue_head(&journal->j_wait_logspace);
818 	init_waitqueue_head(&journal->j_wait_done_commit);
819 	init_waitqueue_head(&journal->j_wait_checkpoint);
820 	init_waitqueue_head(&journal->j_wait_commit);
821 	init_waitqueue_head(&journal->j_wait_updates);
822 	mutex_init(&journal->j_barrier);
823 	mutex_init(&journal->j_checkpoint_mutex);
824 	spin_lock_init(&journal->j_revoke_lock);
825 	spin_lock_init(&journal->j_list_lock);
826 	spin_lock_init(&journal->j_state_lock);
827 
828 	journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
829 	journal->j_min_batch_time = 0;
830 	journal->j_max_batch_time = 15000; /* 15ms */
831 
832 	/* The journal is marked for error until we succeed with recovery! */
833 	journal->j_flags = JBD2_ABORT;
834 
835 	/* Set up a default-sized revoke table for the new mount. */
836 	err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
837 	if (err) {
838 		kfree(journal);
839 		goto fail;
840 	}
841 
842 	spin_lock_init(&journal->j_history_lock);
843 
844 	return journal;
845 fail:
846 	return NULL;
847 }
848 
849 /* jbd2_journal_init_dev and jbd2_journal_init_inode:
850  *
851  * Create a journal structure assigned some fixed set of disk blocks to
852  * the journal.  We don't actually touch those disk blocks yet, but we
853  * need to set up all of the mapping information to tell the journaling
854  * system where the journal blocks are.
855  *
856  */
857 
858 /**
859  *  journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
860  *  @bdev: Block device on which to create the journal
861  *  @fs_dev: Device which hold journalled filesystem for this journal.
862  *  @start: Block nr Start of journal.
863  *  @len:  Length of the journal in blocks.
864  *  @blocksize: blocksize of journalling device
865  *
866  *  Returns: a newly created journal_t *
867  *
868  *  jbd2_journal_init_dev creates a journal which maps a fixed contiguous
869  *  range of blocks on an arbitrary block device.
870  *
871  */
872 journal_t * jbd2_journal_init_dev(struct block_device *bdev,
873 			struct block_device *fs_dev,
874 			unsigned long long start, int len, int blocksize)
875 {
876 	journal_t *journal = journal_init_common();
877 	struct buffer_head *bh;
878 	char *p;
879 	int n;
880 
881 	if (!journal)
882 		return NULL;
883 
884 	/* journal descriptor can store up to n blocks -bzzz */
885 	journal->j_blocksize = blocksize;
886 	jbd2_stats_proc_init(journal);
887 	n = journal->j_blocksize / sizeof(journal_block_tag_t);
888 	journal->j_wbufsize = n;
889 	journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
890 	if (!journal->j_wbuf) {
891 		printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
892 			__func__);
893 		goto out_err;
894 	}
895 	journal->j_dev = bdev;
896 	journal->j_fs_dev = fs_dev;
897 	journal->j_blk_offset = start;
898 	journal->j_maxlen = len;
899 	bdevname(journal->j_dev, journal->j_devname);
900 	p = journal->j_devname;
901 	while ((p = strchr(p, '/')))
902 		*p = '!';
903 
904 	bh = __getblk(journal->j_dev, start, journal->j_blocksize);
905 	if (!bh) {
906 		printk(KERN_ERR
907 		       "%s: Cannot get buffer for journal superblock\n",
908 		       __func__);
909 		goto out_err;
910 	}
911 	journal->j_sb_buffer = bh;
912 	journal->j_superblock = (journal_superblock_t *)bh->b_data;
913 
914 	return journal;
915 out_err:
916 	jbd2_stats_proc_exit(journal);
917 	kfree(journal);
918 	return NULL;
919 }
920 
921 /**
922  *  journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
923  *  @inode: An inode to create the journal in
924  *
925  * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
926  * the journal.  The inode must exist already, must support bmap() and
927  * must have all data blocks preallocated.
928  */
929 journal_t * jbd2_journal_init_inode (struct inode *inode)
930 {
931 	struct buffer_head *bh;
932 	journal_t *journal = journal_init_common();
933 	char *p;
934 	int err;
935 	int n;
936 	unsigned long long blocknr;
937 
938 	if (!journal)
939 		return NULL;
940 
941 	journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
942 	journal->j_inode = inode;
943 	bdevname(journal->j_dev, journal->j_devname);
944 	p = journal->j_devname;
945 	while ((p = strchr(p, '/')))
946 		*p = '!';
947 	p = journal->j_devname + strlen(journal->j_devname);
948 	sprintf(p, "-%lu", journal->j_inode->i_ino);
949 	jbd_debug(1,
950 		  "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
951 		  journal, inode->i_sb->s_id, inode->i_ino,
952 		  (long long) inode->i_size,
953 		  inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
954 
955 	journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
956 	journal->j_blocksize = inode->i_sb->s_blocksize;
957 	jbd2_stats_proc_init(journal);
958 
959 	/* journal descriptor can store up to n blocks -bzzz */
960 	n = journal->j_blocksize / sizeof(journal_block_tag_t);
961 	journal->j_wbufsize = n;
962 	journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
963 	if (!journal->j_wbuf) {
964 		printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n",
965 			__func__);
966 		goto out_err;
967 	}
968 
969 	err = jbd2_journal_bmap(journal, 0, &blocknr);
970 	/* If that failed, give up */
971 	if (err) {
972 		printk(KERN_ERR "%s: Cannnot locate journal superblock\n",
973 		       __func__);
974 		goto out_err;
975 	}
976 
977 	bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
978 	if (!bh) {
979 		printk(KERN_ERR
980 		       "%s: Cannot get buffer for journal superblock\n",
981 		       __func__);
982 		goto out_err;
983 	}
984 	journal->j_sb_buffer = bh;
985 	journal->j_superblock = (journal_superblock_t *)bh->b_data;
986 
987 	return journal;
988 out_err:
989 	jbd2_stats_proc_exit(journal);
990 	kfree(journal);
991 	return NULL;
992 }
993 
994 /*
995  * If the journal init or create aborts, we need to mark the journal
996  * superblock as being NULL to prevent the journal destroy from writing
997  * back a bogus superblock.
998  */
999 static void journal_fail_superblock (journal_t *journal)
1000 {
1001 	struct buffer_head *bh = journal->j_sb_buffer;
1002 	brelse(bh);
1003 	journal->j_sb_buffer = NULL;
1004 }
1005 
1006 /*
1007  * Given a journal_t structure, initialise the various fields for
1008  * startup of a new journaling session.  We use this both when creating
1009  * a journal, and after recovering an old journal to reset it for
1010  * subsequent use.
1011  */
1012 
1013 static int journal_reset(journal_t *journal)
1014 {
1015 	journal_superblock_t *sb = journal->j_superblock;
1016 	unsigned long long first, last;
1017 
1018 	first = be32_to_cpu(sb->s_first);
1019 	last = be32_to_cpu(sb->s_maxlen);
1020 	if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
1021 		printk(KERN_ERR "JBD: Journal too short (blocks %llu-%llu).\n",
1022 		       first, last);
1023 		journal_fail_superblock(journal);
1024 		return -EINVAL;
1025 	}
1026 
1027 	journal->j_first = first;
1028 	journal->j_last = last;
1029 
1030 	journal->j_head = first;
1031 	journal->j_tail = first;
1032 	journal->j_free = last - first;
1033 
1034 	journal->j_tail_sequence = journal->j_transaction_sequence;
1035 	journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1036 	journal->j_commit_request = journal->j_commit_sequence;
1037 
1038 	journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1039 
1040 	/* Add the dynamic fields and write it to disk. */
1041 	jbd2_journal_update_superblock(journal, 1);
1042 	return jbd2_journal_start_thread(journal);
1043 }
1044 
1045 /**
1046  * void jbd2_journal_update_superblock() - Update journal sb on disk.
1047  * @journal: The journal to update.
1048  * @wait: Set to '0' if you don't want to wait for IO completion.
1049  *
1050  * Update a journal's dynamic superblock fields and write it to disk,
1051  * optionally waiting for the IO to complete.
1052  */
1053 void jbd2_journal_update_superblock(journal_t *journal, int wait)
1054 {
1055 	journal_superblock_t *sb = journal->j_superblock;
1056 	struct buffer_head *bh = journal->j_sb_buffer;
1057 
1058 	/*
1059 	 * As a special case, if the on-disk copy is already marked as needing
1060 	 * no recovery (s_start == 0) and there are no outstanding transactions
1061 	 * in the filesystem, then we can safely defer the superblock update
1062 	 * until the next commit by setting JBD2_FLUSHED.  This avoids
1063 	 * attempting a write to a potential-readonly device.
1064 	 */
1065 	if (sb->s_start == 0 && journal->j_tail_sequence ==
1066 				journal->j_transaction_sequence) {
1067 		jbd_debug(1,"JBD: Skipping superblock update on recovered sb "
1068 			"(start %ld, seq %d, errno %d)\n",
1069 			journal->j_tail, journal->j_tail_sequence,
1070 			journal->j_errno);
1071 		goto out;
1072 	}
1073 
1074 	if (buffer_write_io_error(bh)) {
1075 		/*
1076 		 * Oh, dear.  A previous attempt to write the journal
1077 		 * superblock failed.  This could happen because the
1078 		 * USB device was yanked out.  Or it could happen to
1079 		 * be a transient write error and maybe the block will
1080 		 * be remapped.  Nothing we can do but to retry the
1081 		 * write and hope for the best.
1082 		 */
1083 		printk(KERN_ERR "JBD2: previous I/O error detected "
1084 		       "for journal superblock update for %s.\n",
1085 		       journal->j_devname);
1086 		clear_buffer_write_io_error(bh);
1087 		set_buffer_uptodate(bh);
1088 	}
1089 
1090 	spin_lock(&journal->j_state_lock);
1091 	jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n",
1092 		  journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1093 
1094 	sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1095 	sb->s_start    = cpu_to_be32(journal->j_tail);
1096 	sb->s_errno    = cpu_to_be32(journal->j_errno);
1097 	spin_unlock(&journal->j_state_lock);
1098 
1099 	BUFFER_TRACE(bh, "marking dirty");
1100 	mark_buffer_dirty(bh);
1101 	if (wait) {
1102 		sync_dirty_buffer(bh);
1103 		if (buffer_write_io_error(bh)) {
1104 			printk(KERN_ERR "JBD2: I/O error detected "
1105 			       "when updating journal superblock for %s.\n",
1106 			       journal->j_devname);
1107 			clear_buffer_write_io_error(bh);
1108 			set_buffer_uptodate(bh);
1109 		}
1110 	} else
1111 		ll_rw_block(SWRITE, 1, &bh);
1112 
1113 out:
1114 	/* If we have just flushed the log (by marking s_start==0), then
1115 	 * any future commit will have to be careful to update the
1116 	 * superblock again to re-record the true start of the log. */
1117 
1118 	spin_lock(&journal->j_state_lock);
1119 	if (sb->s_start)
1120 		journal->j_flags &= ~JBD2_FLUSHED;
1121 	else
1122 		journal->j_flags |= JBD2_FLUSHED;
1123 	spin_unlock(&journal->j_state_lock);
1124 }
1125 
1126 /*
1127  * Read the superblock for a given journal, performing initial
1128  * validation of the format.
1129  */
1130 
1131 static int journal_get_superblock(journal_t *journal)
1132 {
1133 	struct buffer_head *bh;
1134 	journal_superblock_t *sb;
1135 	int err = -EIO;
1136 
1137 	bh = journal->j_sb_buffer;
1138 
1139 	J_ASSERT(bh != NULL);
1140 	if (!buffer_uptodate(bh)) {
1141 		ll_rw_block(READ, 1, &bh);
1142 		wait_on_buffer(bh);
1143 		if (!buffer_uptodate(bh)) {
1144 			printk (KERN_ERR
1145 				"JBD: IO error reading journal superblock\n");
1146 			goto out;
1147 		}
1148 	}
1149 
1150 	sb = journal->j_superblock;
1151 
1152 	err = -EINVAL;
1153 
1154 	if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
1155 	    sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
1156 		printk(KERN_WARNING "JBD: no valid journal superblock found\n");
1157 		goto out;
1158 	}
1159 
1160 	switch(be32_to_cpu(sb->s_header.h_blocktype)) {
1161 	case JBD2_SUPERBLOCK_V1:
1162 		journal->j_format_version = 1;
1163 		break;
1164 	case JBD2_SUPERBLOCK_V2:
1165 		journal->j_format_version = 2;
1166 		break;
1167 	default:
1168 		printk(KERN_WARNING "JBD: unrecognised superblock format ID\n");
1169 		goto out;
1170 	}
1171 
1172 	if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1173 		journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1174 	else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
1175 		printk (KERN_WARNING "JBD: journal file too short\n");
1176 		goto out;
1177 	}
1178 
1179 	return 0;
1180 
1181 out:
1182 	journal_fail_superblock(journal);
1183 	return err;
1184 }
1185 
1186 /*
1187  * Load the on-disk journal superblock and read the key fields into the
1188  * journal_t.
1189  */
1190 
1191 static int load_superblock(journal_t *journal)
1192 {
1193 	int err;
1194 	journal_superblock_t *sb;
1195 
1196 	err = journal_get_superblock(journal);
1197 	if (err)
1198 		return err;
1199 
1200 	sb = journal->j_superblock;
1201 
1202 	journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1203 	journal->j_tail = be32_to_cpu(sb->s_start);
1204 	journal->j_first = be32_to_cpu(sb->s_first);
1205 	journal->j_last = be32_to_cpu(sb->s_maxlen);
1206 	journal->j_errno = be32_to_cpu(sb->s_errno);
1207 
1208 	return 0;
1209 }
1210 
1211 
1212 /**
1213  * int jbd2_journal_load() - Read journal from disk.
1214  * @journal: Journal to act on.
1215  *
1216  * Given a journal_t structure which tells us which disk blocks contain
1217  * a journal, read the journal from disk to initialise the in-memory
1218  * structures.
1219  */
1220 int jbd2_journal_load(journal_t *journal)
1221 {
1222 	int err;
1223 	journal_superblock_t *sb;
1224 
1225 	err = load_superblock(journal);
1226 	if (err)
1227 		return err;
1228 
1229 	sb = journal->j_superblock;
1230 	/* If this is a V2 superblock, then we have to check the
1231 	 * features flags on it. */
1232 
1233 	if (journal->j_format_version >= 2) {
1234 		if ((sb->s_feature_ro_compat &
1235 		     ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
1236 		    (sb->s_feature_incompat &
1237 		     ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
1238 			printk (KERN_WARNING
1239 				"JBD: Unrecognised features on journal\n");
1240 			return -EINVAL;
1241 		}
1242 	}
1243 
1244 	/* Let the recovery code check whether it needs to recover any
1245 	 * data from the journal. */
1246 	if (jbd2_journal_recover(journal))
1247 		goto recovery_error;
1248 
1249 	/* OK, we've finished with the dynamic journal bits:
1250 	 * reinitialise the dynamic contents of the superblock in memory
1251 	 * and reset them on disk. */
1252 	if (journal_reset(journal))
1253 		goto recovery_error;
1254 
1255 	journal->j_flags &= ~JBD2_ABORT;
1256 	journal->j_flags |= JBD2_LOADED;
1257 	return 0;
1258 
1259 recovery_error:
1260 	printk (KERN_WARNING "JBD: recovery failed\n");
1261 	return -EIO;
1262 }
1263 
1264 /**
1265  * void jbd2_journal_destroy() - Release a journal_t structure.
1266  * @journal: Journal to act on.
1267  *
1268  * Release a journal_t structure once it is no longer in use by the
1269  * journaled object.
1270  * Return <0 if we couldn't clean up the journal.
1271  */
1272 int jbd2_journal_destroy(journal_t *journal)
1273 {
1274 	int err = 0;
1275 
1276 	/* Wait for the commit thread to wake up and die. */
1277 	journal_kill_thread(journal);
1278 
1279 	/* Force a final log commit */
1280 	if (journal->j_running_transaction)
1281 		jbd2_journal_commit_transaction(journal);
1282 
1283 	/* Force any old transactions to disk */
1284 
1285 	/* Totally anal locking here... */
1286 	spin_lock(&journal->j_list_lock);
1287 	while (journal->j_checkpoint_transactions != NULL) {
1288 		spin_unlock(&journal->j_list_lock);
1289 		mutex_lock(&journal->j_checkpoint_mutex);
1290 		jbd2_log_do_checkpoint(journal);
1291 		mutex_unlock(&journal->j_checkpoint_mutex);
1292 		spin_lock(&journal->j_list_lock);
1293 	}
1294 
1295 	J_ASSERT(journal->j_running_transaction == NULL);
1296 	J_ASSERT(journal->j_committing_transaction == NULL);
1297 	J_ASSERT(journal->j_checkpoint_transactions == NULL);
1298 	spin_unlock(&journal->j_list_lock);
1299 
1300 	if (journal->j_sb_buffer) {
1301 		if (!is_journal_aborted(journal)) {
1302 			/* We can now mark the journal as empty. */
1303 			journal->j_tail = 0;
1304 			journal->j_tail_sequence =
1305 				++journal->j_transaction_sequence;
1306 			jbd2_journal_update_superblock(journal, 1);
1307 		} else {
1308 			err = -EIO;
1309 		}
1310 		brelse(journal->j_sb_buffer);
1311 	}
1312 
1313 	if (journal->j_proc_entry)
1314 		jbd2_stats_proc_exit(journal);
1315 	if (journal->j_inode)
1316 		iput(journal->j_inode);
1317 	if (journal->j_revoke)
1318 		jbd2_journal_destroy_revoke(journal);
1319 	kfree(journal->j_wbuf);
1320 	kfree(journal);
1321 
1322 	return err;
1323 }
1324 
1325 
1326 /**
1327  *int jbd2_journal_check_used_features () - Check if features specified are used.
1328  * @journal: Journal to check.
1329  * @compat: bitmask of compatible features
1330  * @ro: bitmask of features that force read-only mount
1331  * @incompat: bitmask of incompatible features
1332  *
1333  * Check whether the journal uses all of a given set of
1334  * features.  Return true (non-zero) if it does.
1335  **/
1336 
1337 int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
1338 				 unsigned long ro, unsigned long incompat)
1339 {
1340 	journal_superblock_t *sb;
1341 
1342 	if (!compat && !ro && !incompat)
1343 		return 1;
1344 	if (journal->j_format_version == 1)
1345 		return 0;
1346 
1347 	sb = journal->j_superblock;
1348 
1349 	if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1350 	    ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1351 	    ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1352 		return 1;
1353 
1354 	return 0;
1355 }
1356 
1357 /**
1358  * int jbd2_journal_check_available_features() - Check feature set in journalling layer
1359  * @journal: Journal to check.
1360  * @compat: bitmask of compatible features
1361  * @ro: bitmask of features that force read-only mount
1362  * @incompat: bitmask of incompatible features
1363  *
1364  * Check whether the journaling code supports the use of
1365  * all of a given set of features on this journal.  Return true
1366  * (non-zero) if it can. */
1367 
1368 int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
1369 				      unsigned long ro, unsigned long incompat)
1370 {
1371 	journal_superblock_t *sb;
1372 
1373 	if (!compat && !ro && !incompat)
1374 		return 1;
1375 
1376 	sb = journal->j_superblock;
1377 
1378 	/* We can support any known requested features iff the
1379 	 * superblock is in version 2.  Otherwise we fail to support any
1380 	 * extended sb features. */
1381 
1382 	if (journal->j_format_version != 2)
1383 		return 0;
1384 
1385 	if ((compat   & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1386 	    (ro       & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1387 	    (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
1388 		return 1;
1389 
1390 	return 0;
1391 }
1392 
1393 /**
1394  * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
1395  * @journal: Journal to act on.
1396  * @compat: bitmask of compatible features
1397  * @ro: bitmask of features that force read-only mount
1398  * @incompat: bitmask of incompatible features
1399  *
1400  * Mark a given journal feature as present on the
1401  * superblock.  Returns true if the requested features could be set.
1402  *
1403  */
1404 
1405 int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
1406 			  unsigned long ro, unsigned long incompat)
1407 {
1408 	journal_superblock_t *sb;
1409 
1410 	if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
1411 		return 1;
1412 
1413 	if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
1414 		return 0;
1415 
1416 	jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1417 		  compat, ro, incompat);
1418 
1419 	sb = journal->j_superblock;
1420 
1421 	sb->s_feature_compat    |= cpu_to_be32(compat);
1422 	sb->s_feature_ro_compat |= cpu_to_be32(ro);
1423 	sb->s_feature_incompat  |= cpu_to_be32(incompat);
1424 
1425 	return 1;
1426 }
1427 
1428 /*
1429  * jbd2_journal_clear_features () - Clear a given journal feature in the
1430  * 				    superblock
1431  * @journal: Journal to act on.
1432  * @compat: bitmask of compatible features
1433  * @ro: bitmask of features that force read-only mount
1434  * @incompat: bitmask of incompatible features
1435  *
1436  * Clear a given journal feature as present on the
1437  * superblock.
1438  */
1439 void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1440 				unsigned long ro, unsigned long incompat)
1441 {
1442 	journal_superblock_t *sb;
1443 
1444 	jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1445 		  compat, ro, incompat);
1446 
1447 	sb = journal->j_superblock;
1448 
1449 	sb->s_feature_compat    &= ~cpu_to_be32(compat);
1450 	sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1451 	sb->s_feature_incompat  &= ~cpu_to_be32(incompat);
1452 }
1453 EXPORT_SYMBOL(jbd2_journal_clear_features);
1454 
1455 /**
1456  * int jbd2_journal_update_format () - Update on-disk journal structure.
1457  * @journal: Journal to act on.
1458  *
1459  * Given an initialised but unloaded journal struct, poke about in the
1460  * on-disk structure to update it to the most recent supported version.
1461  */
1462 int jbd2_journal_update_format (journal_t *journal)
1463 {
1464 	journal_superblock_t *sb;
1465 	int err;
1466 
1467 	err = journal_get_superblock(journal);
1468 	if (err)
1469 		return err;
1470 
1471 	sb = journal->j_superblock;
1472 
1473 	switch (be32_to_cpu(sb->s_header.h_blocktype)) {
1474 	case JBD2_SUPERBLOCK_V2:
1475 		return 0;
1476 	case JBD2_SUPERBLOCK_V1:
1477 		return journal_convert_superblock_v1(journal, sb);
1478 	default:
1479 		break;
1480 	}
1481 	return -EINVAL;
1482 }
1483 
1484 static int journal_convert_superblock_v1(journal_t *journal,
1485 					 journal_superblock_t *sb)
1486 {
1487 	int offset, blocksize;
1488 	struct buffer_head *bh;
1489 
1490 	printk(KERN_WARNING
1491 		"JBD: Converting superblock from version 1 to 2.\n");
1492 
1493 	/* Pre-initialise new fields to zero */
1494 	offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1495 	blocksize = be32_to_cpu(sb->s_blocksize);
1496 	memset(&sb->s_feature_compat, 0, blocksize-offset);
1497 
1498 	sb->s_nr_users = cpu_to_be32(1);
1499 	sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
1500 	journal->j_format_version = 2;
1501 
1502 	bh = journal->j_sb_buffer;
1503 	BUFFER_TRACE(bh, "marking dirty");
1504 	mark_buffer_dirty(bh);
1505 	sync_dirty_buffer(bh);
1506 	return 0;
1507 }
1508 
1509 
1510 /**
1511  * int jbd2_journal_flush () - Flush journal
1512  * @journal: Journal to act on.
1513  *
1514  * Flush all data for a given journal to disk and empty the journal.
1515  * Filesystems can use this when remounting readonly to ensure that
1516  * recovery does not need to happen on remount.
1517  */
1518 
1519 int jbd2_journal_flush(journal_t *journal)
1520 {
1521 	int err = 0;
1522 	transaction_t *transaction = NULL;
1523 	unsigned long old_tail;
1524 
1525 	spin_lock(&journal->j_state_lock);
1526 
1527 	/* Force everything buffered to the log... */
1528 	if (journal->j_running_transaction) {
1529 		transaction = journal->j_running_transaction;
1530 		__jbd2_log_start_commit(journal, transaction->t_tid);
1531 	} else if (journal->j_committing_transaction)
1532 		transaction = journal->j_committing_transaction;
1533 
1534 	/* Wait for the log commit to complete... */
1535 	if (transaction) {
1536 		tid_t tid = transaction->t_tid;
1537 
1538 		spin_unlock(&journal->j_state_lock);
1539 		jbd2_log_wait_commit(journal, tid);
1540 	} else {
1541 		spin_unlock(&journal->j_state_lock);
1542 	}
1543 
1544 	/* ...and flush everything in the log out to disk. */
1545 	spin_lock(&journal->j_list_lock);
1546 	while (!err && journal->j_checkpoint_transactions != NULL) {
1547 		spin_unlock(&journal->j_list_lock);
1548 		mutex_lock(&journal->j_checkpoint_mutex);
1549 		err = jbd2_log_do_checkpoint(journal);
1550 		mutex_unlock(&journal->j_checkpoint_mutex);
1551 		spin_lock(&journal->j_list_lock);
1552 	}
1553 	spin_unlock(&journal->j_list_lock);
1554 
1555 	if (is_journal_aborted(journal))
1556 		return -EIO;
1557 
1558 	jbd2_cleanup_journal_tail(journal);
1559 
1560 	/* Finally, mark the journal as really needing no recovery.
1561 	 * This sets s_start==0 in the underlying superblock, which is
1562 	 * the magic code for a fully-recovered superblock.  Any future
1563 	 * commits of data to the journal will restore the current
1564 	 * s_start value. */
1565 	spin_lock(&journal->j_state_lock);
1566 	old_tail = journal->j_tail;
1567 	journal->j_tail = 0;
1568 	spin_unlock(&journal->j_state_lock);
1569 	jbd2_journal_update_superblock(journal, 1);
1570 	spin_lock(&journal->j_state_lock);
1571 	journal->j_tail = old_tail;
1572 
1573 	J_ASSERT(!journal->j_running_transaction);
1574 	J_ASSERT(!journal->j_committing_transaction);
1575 	J_ASSERT(!journal->j_checkpoint_transactions);
1576 	J_ASSERT(journal->j_head == journal->j_tail);
1577 	J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
1578 	spin_unlock(&journal->j_state_lock);
1579 	return 0;
1580 }
1581 
1582 /**
1583  * int jbd2_journal_wipe() - Wipe journal contents
1584  * @journal: Journal to act on.
1585  * @write: flag (see below)
1586  *
1587  * Wipe out all of the contents of a journal, safely.  This will produce
1588  * a warning if the journal contains any valid recovery information.
1589  * Must be called between journal_init_*() and jbd2_journal_load().
1590  *
1591  * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1592  * we merely suppress recovery.
1593  */
1594 
1595 int jbd2_journal_wipe(journal_t *journal, int write)
1596 {
1597 	journal_superblock_t *sb;
1598 	int err = 0;
1599 
1600 	J_ASSERT (!(journal->j_flags & JBD2_LOADED));
1601 
1602 	err = load_superblock(journal);
1603 	if (err)
1604 		return err;
1605 
1606 	sb = journal->j_superblock;
1607 
1608 	if (!journal->j_tail)
1609 		goto no_recovery;
1610 
1611 	printk (KERN_WARNING "JBD: %s recovery information on journal\n",
1612 		write ? "Clearing" : "Ignoring");
1613 
1614 	err = jbd2_journal_skip_recovery(journal);
1615 	if (write)
1616 		jbd2_journal_update_superblock(journal, 1);
1617 
1618  no_recovery:
1619 	return err;
1620 }
1621 
1622 /*
1623  * Journal abort has very specific semantics, which we describe
1624  * for journal abort.
1625  *
1626  * Two internal functions, which provide abort to the jbd layer
1627  * itself are here.
1628  */
1629 
1630 /*
1631  * Quick version for internal journal use (doesn't lock the journal).
1632  * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1633  * and don't attempt to make any other journal updates.
1634  */
1635 void __jbd2_journal_abort_hard(journal_t *journal)
1636 {
1637 	transaction_t *transaction;
1638 
1639 	if (journal->j_flags & JBD2_ABORT)
1640 		return;
1641 
1642 	printk(KERN_ERR "Aborting journal on device %s.\n",
1643 	       journal->j_devname);
1644 
1645 	spin_lock(&journal->j_state_lock);
1646 	journal->j_flags |= JBD2_ABORT;
1647 	transaction = journal->j_running_transaction;
1648 	if (transaction)
1649 		__jbd2_log_start_commit(journal, transaction->t_tid);
1650 	spin_unlock(&journal->j_state_lock);
1651 }
1652 
1653 /* Soft abort: record the abort error status in the journal superblock,
1654  * but don't do any other IO. */
1655 static void __journal_abort_soft (journal_t *journal, int errno)
1656 {
1657 	if (journal->j_flags & JBD2_ABORT)
1658 		return;
1659 
1660 	if (!journal->j_errno)
1661 		journal->j_errno = errno;
1662 
1663 	__jbd2_journal_abort_hard(journal);
1664 
1665 	if (errno)
1666 		jbd2_journal_update_superblock(journal, 1);
1667 }
1668 
1669 /**
1670  * void jbd2_journal_abort () - Shutdown the journal immediately.
1671  * @journal: the journal to shutdown.
1672  * @errno:   an error number to record in the journal indicating
1673  *           the reason for the shutdown.
1674  *
1675  * Perform a complete, immediate shutdown of the ENTIRE
1676  * journal (not of a single transaction).  This operation cannot be
1677  * undone without closing and reopening the journal.
1678  *
1679  * The jbd2_journal_abort function is intended to support higher level error
1680  * recovery mechanisms such as the ext2/ext3 remount-readonly error
1681  * mode.
1682  *
1683  * Journal abort has very specific semantics.  Any existing dirty,
1684  * unjournaled buffers in the main filesystem will still be written to
1685  * disk by bdflush, but the journaling mechanism will be suspended
1686  * immediately and no further transaction commits will be honoured.
1687  *
1688  * Any dirty, journaled buffers will be written back to disk without
1689  * hitting the journal.  Atomicity cannot be guaranteed on an aborted
1690  * filesystem, but we _do_ attempt to leave as much data as possible
1691  * behind for fsck to use for cleanup.
1692  *
1693  * Any attempt to get a new transaction handle on a journal which is in
1694  * ABORT state will just result in an -EROFS error return.  A
1695  * jbd2_journal_stop on an existing handle will return -EIO if we have
1696  * entered abort state during the update.
1697  *
1698  * Recursive transactions are not disturbed by journal abort until the
1699  * final jbd2_journal_stop, which will receive the -EIO error.
1700  *
1701  * Finally, the jbd2_journal_abort call allows the caller to supply an errno
1702  * which will be recorded (if possible) in the journal superblock.  This
1703  * allows a client to record failure conditions in the middle of a
1704  * transaction without having to complete the transaction to record the
1705  * failure to disk.  ext3_error, for example, now uses this
1706  * functionality.
1707  *
1708  * Errors which originate from within the journaling layer will NOT
1709  * supply an errno; a null errno implies that absolutely no further
1710  * writes are done to the journal (unless there are any already in
1711  * progress).
1712  *
1713  */
1714 
1715 void jbd2_journal_abort(journal_t *journal, int errno)
1716 {
1717 	__journal_abort_soft(journal, errno);
1718 }
1719 
1720 /**
1721  * int jbd2_journal_errno () - returns the journal's error state.
1722  * @journal: journal to examine.
1723  *
1724  * This is the errno number set with jbd2_journal_abort(), the last
1725  * time the journal was mounted - if the journal was stopped
1726  * without calling abort this will be 0.
1727  *
1728  * If the journal has been aborted on this mount time -EROFS will
1729  * be returned.
1730  */
1731 int jbd2_journal_errno(journal_t *journal)
1732 {
1733 	int err;
1734 
1735 	spin_lock(&journal->j_state_lock);
1736 	if (journal->j_flags & JBD2_ABORT)
1737 		err = -EROFS;
1738 	else
1739 		err = journal->j_errno;
1740 	spin_unlock(&journal->j_state_lock);
1741 	return err;
1742 }
1743 
1744 /**
1745  * int jbd2_journal_clear_err () - clears the journal's error state
1746  * @journal: journal to act on.
1747  *
1748  * An error must be cleared or acked to take a FS out of readonly
1749  * mode.
1750  */
1751 int jbd2_journal_clear_err(journal_t *journal)
1752 {
1753 	int err = 0;
1754 
1755 	spin_lock(&journal->j_state_lock);
1756 	if (journal->j_flags & JBD2_ABORT)
1757 		err = -EROFS;
1758 	else
1759 		journal->j_errno = 0;
1760 	spin_unlock(&journal->j_state_lock);
1761 	return err;
1762 }
1763 
1764 /**
1765  * void jbd2_journal_ack_err() - Ack journal err.
1766  * @journal: journal to act on.
1767  *
1768  * An error must be cleared or acked to take a FS out of readonly
1769  * mode.
1770  */
1771 void jbd2_journal_ack_err(journal_t *journal)
1772 {
1773 	spin_lock(&journal->j_state_lock);
1774 	if (journal->j_errno)
1775 		journal->j_flags |= JBD2_ACK_ERR;
1776 	spin_unlock(&journal->j_state_lock);
1777 }
1778 
1779 int jbd2_journal_blocks_per_page(struct inode *inode)
1780 {
1781 	return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1782 }
1783 
1784 /*
1785  * helper functions to deal with 32 or 64bit block numbers.
1786  */
1787 size_t journal_tag_bytes(journal_t *journal)
1788 {
1789 	if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
1790 		return JBD2_TAG_SIZE64;
1791 	else
1792 		return JBD2_TAG_SIZE32;
1793 }
1794 
1795 /*
1796  * Journal_head storage management
1797  */
1798 static struct kmem_cache *jbd2_journal_head_cache;
1799 #ifdef CONFIG_JBD2_DEBUG
1800 static atomic_t nr_journal_heads = ATOMIC_INIT(0);
1801 #endif
1802 
1803 static int journal_init_jbd2_journal_head_cache(void)
1804 {
1805 	int retval;
1806 
1807 	J_ASSERT(jbd2_journal_head_cache == NULL);
1808 	jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
1809 				sizeof(struct journal_head),
1810 				0,		/* offset */
1811 				SLAB_TEMPORARY,	/* flags */
1812 				NULL);		/* ctor */
1813 	retval = 0;
1814 	if (!jbd2_journal_head_cache) {
1815 		retval = -ENOMEM;
1816 		printk(KERN_EMERG "JBD: no memory for journal_head cache\n");
1817 	}
1818 	return retval;
1819 }
1820 
1821 static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
1822 {
1823 	if (jbd2_journal_head_cache) {
1824 		kmem_cache_destroy(jbd2_journal_head_cache);
1825 		jbd2_journal_head_cache = NULL;
1826 	}
1827 }
1828 
1829 /*
1830  * journal_head splicing and dicing
1831  */
1832 static struct journal_head *journal_alloc_journal_head(void)
1833 {
1834 	struct journal_head *ret;
1835 	static unsigned long last_warning;
1836 
1837 #ifdef CONFIG_JBD2_DEBUG
1838 	atomic_inc(&nr_journal_heads);
1839 #endif
1840 	ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1841 	if (!ret) {
1842 		jbd_debug(1, "out of memory for journal_head\n");
1843 		if (time_after(jiffies, last_warning + 5*HZ)) {
1844 			printk(KERN_NOTICE "ENOMEM in %s, retrying.\n",
1845 			       __func__);
1846 			last_warning = jiffies;
1847 		}
1848 		while (!ret) {
1849 			yield();
1850 			ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
1851 		}
1852 	}
1853 	return ret;
1854 }
1855 
1856 static void journal_free_journal_head(struct journal_head *jh)
1857 {
1858 #ifdef CONFIG_JBD2_DEBUG
1859 	atomic_dec(&nr_journal_heads);
1860 	memset(jh, JBD2_POISON_FREE, sizeof(*jh));
1861 #endif
1862 	kmem_cache_free(jbd2_journal_head_cache, jh);
1863 }
1864 
1865 /*
1866  * A journal_head is attached to a buffer_head whenever JBD has an
1867  * interest in the buffer.
1868  *
1869  * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
1870  * is set.  This bit is tested in core kernel code where we need to take
1871  * JBD-specific actions.  Testing the zeroness of ->b_private is not reliable
1872  * there.
1873  *
1874  * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
1875  *
1876  * When a buffer has its BH_JBD bit set it is immune from being released by
1877  * core kernel code, mainly via ->b_count.
1878  *
1879  * A journal_head may be detached from its buffer_head when the journal_head's
1880  * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL.
1881  * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the
1882  * journal_head can be dropped if needed.
1883  *
1884  * Various places in the kernel want to attach a journal_head to a buffer_head
1885  * _before_ attaching the journal_head to a transaction.  To protect the
1886  * journal_head in this situation, jbd2_journal_add_journal_head elevates the
1887  * journal_head's b_jcount refcount by one.  The caller must call
1888  * jbd2_journal_put_journal_head() to undo this.
1889  *
1890  * So the typical usage would be:
1891  *
1892  *	(Attach a journal_head if needed.  Increments b_jcount)
1893  *	struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1894  *	...
1895  *	jh->b_transaction = xxx;
1896  *	jbd2_journal_put_journal_head(jh);
1897  *
1898  * Now, the journal_head's b_jcount is zero, but it is safe from being released
1899  * because it has a non-zero b_transaction.
1900  */
1901 
1902 /*
1903  * Give a buffer_head a journal_head.
1904  *
1905  * Doesn't need the journal lock.
1906  * May sleep.
1907  */
1908 struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
1909 {
1910 	struct journal_head *jh;
1911 	struct journal_head *new_jh = NULL;
1912 
1913 repeat:
1914 	if (!buffer_jbd(bh)) {
1915 		new_jh = journal_alloc_journal_head();
1916 		memset(new_jh, 0, sizeof(*new_jh));
1917 	}
1918 
1919 	jbd_lock_bh_journal_head(bh);
1920 	if (buffer_jbd(bh)) {
1921 		jh = bh2jh(bh);
1922 	} else {
1923 		J_ASSERT_BH(bh,
1924 			(atomic_read(&bh->b_count) > 0) ||
1925 			(bh->b_page && bh->b_page->mapping));
1926 
1927 		if (!new_jh) {
1928 			jbd_unlock_bh_journal_head(bh);
1929 			goto repeat;
1930 		}
1931 
1932 		jh = new_jh;
1933 		new_jh = NULL;		/* We consumed it */
1934 		set_buffer_jbd(bh);
1935 		bh->b_private = jh;
1936 		jh->b_bh = bh;
1937 		get_bh(bh);
1938 		BUFFER_TRACE(bh, "added journal_head");
1939 	}
1940 	jh->b_jcount++;
1941 	jbd_unlock_bh_journal_head(bh);
1942 	if (new_jh)
1943 		journal_free_journal_head(new_jh);
1944 	return bh->b_private;
1945 }
1946 
1947 /*
1948  * Grab a ref against this buffer_head's journal_head.  If it ended up not
1949  * having a journal_head, return NULL
1950  */
1951 struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
1952 {
1953 	struct journal_head *jh = NULL;
1954 
1955 	jbd_lock_bh_journal_head(bh);
1956 	if (buffer_jbd(bh)) {
1957 		jh = bh2jh(bh);
1958 		jh->b_jcount++;
1959 	}
1960 	jbd_unlock_bh_journal_head(bh);
1961 	return jh;
1962 }
1963 
1964 static void __journal_remove_journal_head(struct buffer_head *bh)
1965 {
1966 	struct journal_head *jh = bh2jh(bh);
1967 
1968 	J_ASSERT_JH(jh, jh->b_jcount >= 0);
1969 
1970 	get_bh(bh);
1971 	if (jh->b_jcount == 0) {
1972 		if (jh->b_transaction == NULL &&
1973 				jh->b_next_transaction == NULL &&
1974 				jh->b_cp_transaction == NULL) {
1975 			J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
1976 			J_ASSERT_BH(bh, buffer_jbd(bh));
1977 			J_ASSERT_BH(bh, jh2bh(jh) == bh);
1978 			BUFFER_TRACE(bh, "remove journal_head");
1979 			if (jh->b_frozen_data) {
1980 				printk(KERN_WARNING "%s: freeing "
1981 						"b_frozen_data\n",
1982 						__func__);
1983 				jbd2_free(jh->b_frozen_data, bh->b_size);
1984 			}
1985 			if (jh->b_committed_data) {
1986 				printk(KERN_WARNING "%s: freeing "
1987 						"b_committed_data\n",
1988 						__func__);
1989 				jbd2_free(jh->b_committed_data, bh->b_size);
1990 			}
1991 			bh->b_private = NULL;
1992 			jh->b_bh = NULL;	/* debug, really */
1993 			clear_buffer_jbd(bh);
1994 			__brelse(bh);
1995 			journal_free_journal_head(jh);
1996 		} else {
1997 			BUFFER_TRACE(bh, "journal_head was locked");
1998 		}
1999 	}
2000 }
2001 
2002 /*
2003  * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction
2004  * and has a zero b_jcount then remove and release its journal_head.   If we did
2005  * see that the buffer is not used by any transaction we also "logically"
2006  * decrement ->b_count.
2007  *
2008  * We in fact take an additional increment on ->b_count as a convenience,
2009  * because the caller usually wants to do additional things with the bh
2010  * after calling here.
2011  * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some
2012  * time.  Once the caller has run __brelse(), the buffer is eligible for
2013  * reaping by try_to_free_buffers().
2014  */
2015 void jbd2_journal_remove_journal_head(struct buffer_head *bh)
2016 {
2017 	jbd_lock_bh_journal_head(bh);
2018 	__journal_remove_journal_head(bh);
2019 	jbd_unlock_bh_journal_head(bh);
2020 }
2021 
2022 /*
2023  * Drop a reference on the passed journal_head.  If it fell to zero then try to
2024  * release the journal_head from the buffer_head.
2025  */
2026 void jbd2_journal_put_journal_head(struct journal_head *jh)
2027 {
2028 	struct buffer_head *bh = jh2bh(jh);
2029 
2030 	jbd_lock_bh_journal_head(bh);
2031 	J_ASSERT_JH(jh, jh->b_jcount > 0);
2032 	--jh->b_jcount;
2033 	if (!jh->b_jcount && !jh->b_transaction) {
2034 		__journal_remove_journal_head(bh);
2035 		__brelse(bh);
2036 	}
2037 	jbd_unlock_bh_journal_head(bh);
2038 }
2039 
2040 /*
2041  * Initialize jbd inode head
2042  */
2043 void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2044 {
2045 	jinode->i_transaction = NULL;
2046 	jinode->i_next_transaction = NULL;
2047 	jinode->i_vfs_inode = inode;
2048 	jinode->i_flags = 0;
2049 	INIT_LIST_HEAD(&jinode->i_list);
2050 }
2051 
2052 /*
2053  * Function to be called before we start removing inode from memory (i.e.,
2054  * clear_inode() is a fine place to be called from). It removes inode from
2055  * transaction's lists.
2056  */
2057 void jbd2_journal_release_jbd_inode(journal_t *journal,
2058 				    struct jbd2_inode *jinode)
2059 {
2060 	int writeout = 0;
2061 
2062 	if (!journal)
2063 		return;
2064 restart:
2065 	spin_lock(&journal->j_list_lock);
2066 	/* Is commit writing out inode - we have to wait */
2067 	if (jinode->i_flags & JI_COMMIT_RUNNING) {
2068 		wait_queue_head_t *wq;
2069 		DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2070 		wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2071 		prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2072 		spin_unlock(&journal->j_list_lock);
2073 		schedule();
2074 		finish_wait(wq, &wait.wait);
2075 		goto restart;
2076 	}
2077 
2078 	/* Do we need to wait for data writeback? */
2079 	if (journal->j_committing_transaction == jinode->i_transaction)
2080 		writeout = 1;
2081 	if (jinode->i_transaction) {
2082 		list_del(&jinode->i_list);
2083 		jinode->i_transaction = NULL;
2084 	}
2085 	spin_unlock(&journal->j_list_lock);
2086 }
2087 
2088 /*
2089  * debugfs tunables
2090  */
2091 #ifdef CONFIG_JBD2_DEBUG
2092 u8 jbd2_journal_enable_debug __read_mostly;
2093 EXPORT_SYMBOL(jbd2_journal_enable_debug);
2094 
2095 #define JBD2_DEBUG_NAME "jbd2-debug"
2096 
2097 static struct dentry *jbd2_debugfs_dir;
2098 static struct dentry *jbd2_debug;
2099 
2100 static void __init jbd2_create_debugfs_entry(void)
2101 {
2102 	jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2103 	if (jbd2_debugfs_dir)
2104 		jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO,
2105 					       jbd2_debugfs_dir,
2106 					       &jbd2_journal_enable_debug);
2107 }
2108 
2109 static void __exit jbd2_remove_debugfs_entry(void)
2110 {
2111 	debugfs_remove(jbd2_debug);
2112 	debugfs_remove(jbd2_debugfs_dir);
2113 }
2114 
2115 #else
2116 
2117 static void __init jbd2_create_debugfs_entry(void)
2118 {
2119 }
2120 
2121 static void __exit jbd2_remove_debugfs_entry(void)
2122 {
2123 }
2124 
2125 #endif
2126 
2127 #ifdef CONFIG_PROC_FS
2128 
2129 #define JBD2_STATS_PROC_NAME "fs/jbd2"
2130 
2131 static void __init jbd2_create_jbd_stats_proc_entry(void)
2132 {
2133 	proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2134 }
2135 
2136 static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2137 {
2138 	if (proc_jbd2_stats)
2139 		remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2140 }
2141 
2142 #else
2143 
2144 #define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2145 #define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2146 
2147 #endif
2148 
2149 struct kmem_cache *jbd2_handle_cache;
2150 
2151 static int __init journal_init_handle_cache(void)
2152 {
2153 	jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle",
2154 				sizeof(handle_t),
2155 				0,		/* offset */
2156 				SLAB_TEMPORARY,	/* flags */
2157 				NULL);		/* ctor */
2158 	if (jbd2_handle_cache == NULL) {
2159 		printk(KERN_EMERG "JBD: failed to create handle cache\n");
2160 		return -ENOMEM;
2161 	}
2162 	return 0;
2163 }
2164 
2165 static void jbd2_journal_destroy_handle_cache(void)
2166 {
2167 	if (jbd2_handle_cache)
2168 		kmem_cache_destroy(jbd2_handle_cache);
2169 }
2170 
2171 /*
2172  * Module startup and shutdown
2173  */
2174 
2175 static int __init journal_init_caches(void)
2176 {
2177 	int ret;
2178 
2179 	ret = jbd2_journal_init_revoke_caches();
2180 	if (ret == 0)
2181 		ret = journal_init_jbd2_journal_head_cache();
2182 	if (ret == 0)
2183 		ret = journal_init_handle_cache();
2184 	return ret;
2185 }
2186 
2187 static void jbd2_journal_destroy_caches(void)
2188 {
2189 	jbd2_journal_destroy_revoke_caches();
2190 	jbd2_journal_destroy_jbd2_journal_head_cache();
2191 	jbd2_journal_destroy_handle_cache();
2192 }
2193 
2194 static int __init journal_init(void)
2195 {
2196 	int ret;
2197 
2198 	BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2199 
2200 	ret = journal_init_caches();
2201 	if (ret == 0) {
2202 		jbd2_create_debugfs_entry();
2203 		jbd2_create_jbd_stats_proc_entry();
2204 	} else {
2205 		jbd2_journal_destroy_caches();
2206 	}
2207 	return ret;
2208 }
2209 
2210 static void __exit journal_exit(void)
2211 {
2212 #ifdef CONFIG_JBD2_DEBUG
2213 	int n = atomic_read(&nr_journal_heads);
2214 	if (n)
2215 		printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n);
2216 #endif
2217 	jbd2_remove_debugfs_entry();
2218 	jbd2_remove_jbd_stats_proc_entry();
2219 	jbd2_journal_destroy_caches();
2220 }
2221 
2222 /*
2223  * jbd2_dev_to_name is a utility function used by the jbd2 and ext4
2224  * tracing infrastructure to map a dev_t to a device name.
2225  *
2226  * The caller should use rcu_read_lock() in order to make sure the
2227  * device name stays valid until its done with it.  We use
2228  * rcu_read_lock() as well to make sure we're safe in case the caller
2229  * gets sloppy, and because rcu_read_lock() is cheap and can be safely
2230  * nested.
2231  */
2232 struct devname_cache {
2233 	struct rcu_head	rcu;
2234 	dev_t		device;
2235 	char		devname[BDEVNAME_SIZE];
2236 };
2237 #define CACHE_SIZE_BITS 6
2238 static struct devname_cache *devcache[1 << CACHE_SIZE_BITS];
2239 static DEFINE_SPINLOCK(devname_cache_lock);
2240 
2241 static void free_devcache(struct rcu_head *rcu)
2242 {
2243 	kfree(rcu);
2244 }
2245 
2246 const char *jbd2_dev_to_name(dev_t device)
2247 {
2248 	int	i = hash_32(device, CACHE_SIZE_BITS);
2249 	char	*ret;
2250 	struct block_device *bd;
2251 	static struct devname_cache *new_dev;
2252 
2253 	rcu_read_lock();
2254 	if (devcache[i] && devcache[i]->device == device) {
2255 		ret = devcache[i]->devname;
2256 		rcu_read_unlock();
2257 		return ret;
2258 	}
2259 	rcu_read_unlock();
2260 
2261 	new_dev = kmalloc(sizeof(struct devname_cache), GFP_KERNEL);
2262 	if (!new_dev)
2263 		return "NODEV-ALLOCFAILURE"; /* Something non-NULL */
2264 	spin_lock(&devname_cache_lock);
2265 	if (devcache[i]) {
2266 		if (devcache[i]->device == device) {
2267 			kfree(new_dev);
2268 			ret = devcache[i]->devname;
2269 			spin_unlock(&devname_cache_lock);
2270 			return ret;
2271 		}
2272 		call_rcu(&devcache[i]->rcu, free_devcache);
2273 	}
2274 	devcache[i] = new_dev;
2275 	devcache[i]->device = device;
2276 	bd = bdget(device);
2277 	if (bd) {
2278 		bdevname(bd, devcache[i]->devname);
2279 		bdput(bd);
2280 	} else
2281 		__bdevname(device, devcache[i]->devname);
2282 	ret = devcache[i]->devname;
2283 	spin_unlock(&devname_cache_lock);
2284 	return ret;
2285 }
2286 EXPORT_SYMBOL(jbd2_dev_to_name);
2287 
2288 MODULE_LICENSE("GPL");
2289 module_init(journal_init);
2290 module_exit(journal_exit);
2291 
2292