blob: 0fa0123151d3117c5d3678c1bc36bf7b5bb00cd4 [file] [log] [blame]
Dave Kleikamp470decc2006-10-11 01:20:57 -07001/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002 * linux/fs/jbd2/journal.c
Dave Kleikamp470decc2006-10-11 01:20:57 -07003 *
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>
Mingming Caof7f4bcc2006-10-11 01:20:59 -070028#include <linux/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070029#include <linux/errno.h>
30#include <linux/slab.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070031#include <linux/init.h>
32#include <linux/mm.h>
Nigel Cunningham7dfb7102006-12-06 20:34:23 -080033#include <linux/freezer.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070034#include <linux/pagemap.h>
35#include <linux/kthread.h>
36#include <linux/poison.h>
37#include <linux/proc_fs.h>
Jose R. Santos0f49d5d2007-07-18 08:50:18 -040038#include <linux/debugfs.h>
Johann Lombardi8e85fb32008-01-28 23:58:27 -050039#include <linux/seq_file.h>
Simon Holm Thøgersenc225aa52009-01-11 22:34:01 -050040#include <linux/math64.h>
Theodore Ts'o879c5e62009-06-17 11:47:48 -040041#include <linux/hash.h>
Theodore Ts'od2eecb02009-12-07 10:36:20 -050042#include <linux/log2.h>
43#include <linux/vmalloc.h>
Theodore Ts'o47def822010-07-27 11:56:05 -040044#include <linux/backing-dev.h>
Brian King39e3ac22010-10-27 21:25:12 -040045#include <linux/bitops.h>
Theodore Ts'o670be5a2010-12-17 10:44:16 -050046#include <linux/ratelimit.h>
Theodore Ts'o879c5e62009-06-17 11:47:48 -040047
48#define CREATE_TRACE_POINTS
49#include <trace/events/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070050
51#include <asm/uaccess.h>
52#include <asm/page.h>
Brian King39e3ac22010-10-27 21:25:12 -040053#include <asm/system.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070054
Mingming Caof7f4bcc2006-10-11 01:20:59 -070055EXPORT_SYMBOL(jbd2_journal_extend);
56EXPORT_SYMBOL(jbd2_journal_stop);
57EXPORT_SYMBOL(jbd2_journal_lock_updates);
58EXPORT_SYMBOL(jbd2_journal_unlock_updates);
59EXPORT_SYMBOL(jbd2_journal_get_write_access);
60EXPORT_SYMBOL(jbd2_journal_get_create_access);
61EXPORT_SYMBOL(jbd2_journal_get_undo_access);
Joel Beckere06c8222008-09-11 15:35:47 -070062EXPORT_SYMBOL(jbd2_journal_set_triggers);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070063EXPORT_SYMBOL(jbd2_journal_dirty_metadata);
64EXPORT_SYMBOL(jbd2_journal_release_buffer);
65EXPORT_SYMBOL(jbd2_journal_forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -070066#if 0
67EXPORT_SYMBOL(journal_sync_buffer);
68#endif
Mingming Caof7f4bcc2006-10-11 01:20:59 -070069EXPORT_SYMBOL(jbd2_journal_flush);
70EXPORT_SYMBOL(jbd2_journal_revoke);
Dave Kleikamp470decc2006-10-11 01:20:57 -070071
Mingming Caof7f4bcc2006-10-11 01:20:59 -070072EXPORT_SYMBOL(jbd2_journal_init_dev);
73EXPORT_SYMBOL(jbd2_journal_init_inode);
74EXPORT_SYMBOL(jbd2_journal_update_format);
75EXPORT_SYMBOL(jbd2_journal_check_used_features);
76EXPORT_SYMBOL(jbd2_journal_check_available_features);
77EXPORT_SYMBOL(jbd2_journal_set_features);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070078EXPORT_SYMBOL(jbd2_journal_load);
79EXPORT_SYMBOL(jbd2_journal_destroy);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070080EXPORT_SYMBOL(jbd2_journal_abort);
81EXPORT_SYMBOL(jbd2_journal_errno);
82EXPORT_SYMBOL(jbd2_journal_ack_err);
83EXPORT_SYMBOL(jbd2_journal_clear_err);
84EXPORT_SYMBOL(jbd2_log_wait_commit);
Theodore Ts'o3b799d12009-12-09 20:42:53 -050085EXPORT_SYMBOL(jbd2_log_start_commit);
Mingming Caof7f4bcc2006-10-11 01:20:59 -070086EXPORT_SYMBOL(jbd2_journal_start_commit);
87EXPORT_SYMBOL(jbd2_journal_force_commit_nested);
88EXPORT_SYMBOL(jbd2_journal_wipe);
89EXPORT_SYMBOL(jbd2_journal_blocks_per_page);
90EXPORT_SYMBOL(jbd2_journal_invalidatepage);
91EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers);
92EXPORT_SYMBOL(jbd2_journal_force_commit);
Jan Karac851ed52008-07-11 19:27:31 -040093EXPORT_SYMBOL(jbd2_journal_file_inode);
94EXPORT_SYMBOL(jbd2_journal_init_jbd_inode);
95EXPORT_SYMBOL(jbd2_journal_release_jbd_inode);
96EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -050097EXPORT_SYMBOL(jbd2_inode_cache);
Dave Kleikamp470decc2006-10-11 01:20:57 -070098
99static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *);
100static void __journal_abort_soft (journal_t *journal, int errno);
Theodore Ts'od2eecb02009-12-07 10:36:20 -0500101static int jbd2_journal_create_slab(size_t slab_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700102
103/*
104 * Helper function used to manage commit timeouts
105 */
106
107static void commit_timeout(unsigned long __data)
108{
109 struct task_struct * p = (struct task_struct *) __data;
110
111 wake_up_process(p);
112}
113
114/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700115 * kjournald2: The main thread function used to manage a logging device
Dave Kleikamp470decc2006-10-11 01:20:57 -0700116 * journal.
117 *
118 * This kernel thread is responsible for two things:
119 *
120 * 1) COMMIT: Every so often we need to commit the current state of the
121 * filesystem to disk. The journal thread is responsible for writing
122 * all of the metadata buffers to disk.
123 *
124 * 2) CHECKPOINT: We cannot reuse a used section of the log file until all
125 * of the data in that part of the log has been rewritten elsewhere on
126 * the disk. Flushing these old buffers to reclaim space in the log is
127 * known as checkpointing, and this thread is responsible for that job.
128 */
129
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700130static int kjournald2(void *arg)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700131{
132 journal_t *journal = arg;
133 transaction_t *transaction;
134
135 /*
136 * Set up an interval timer which can be used to trigger a commit wakeup
137 * after the commit interval expires
138 */
139 setup_timer(&journal->j_commit_timer, commit_timeout,
140 (unsigned long)current);
141
142 /* Record that the journal thread is running */
143 journal->j_task = current;
144 wake_up(&journal->j_wait_done_commit);
145
Dave Kleikamp470decc2006-10-11 01:20:57 -0700146 /*
147 * And now, wait forever for commit wakeup events.
148 */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400149 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700150
151loop:
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700152 if (journal->j_flags & JBD2_UNMOUNT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700153 goto end_loop;
154
155 jbd_debug(1, "commit_sequence=%d, commit_request=%d\n",
156 journal->j_commit_sequence, journal->j_commit_request);
157
158 if (journal->j_commit_sequence != journal->j_commit_request) {
159 jbd_debug(1, "OK, requests differ\n");
Theodore Ts'oa931da62010-08-03 21:35:12 -0400160 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700161 del_timer_sync(&journal->j_commit_timer);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700162 jbd2_journal_commit_transaction(journal);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400163 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700164 goto loop;
165 }
166
167 wake_up(&journal->j_wait_done_commit);
168 if (freezing(current)) {
169 /*
170 * The simpler the better. Flushing journal isn't a
171 * good idea, because that depends on threads that may
172 * be already stopped.
173 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700174 jbd_debug(1, "Now suspending kjournald2\n");
Theodore Ts'oa931da62010-08-03 21:35:12 -0400175 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700176 refrigerator();
Theodore Ts'oa931da62010-08-03 21:35:12 -0400177 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700178 } else {
179 /*
180 * We assume on resume that commits are already there,
181 * so we don't sleep
182 */
183 DEFINE_WAIT(wait);
184 int should_sleep = 1;
185
186 prepare_to_wait(&journal->j_wait_commit, &wait,
187 TASK_INTERRUPTIBLE);
188 if (journal->j_commit_sequence != journal->j_commit_request)
189 should_sleep = 0;
190 transaction = journal->j_running_transaction;
191 if (transaction && time_after_eq(jiffies,
192 transaction->t_expires))
193 should_sleep = 0;
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700194 if (journal->j_flags & JBD2_UNMOUNT)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700195 should_sleep = 0;
196 if (should_sleep) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400197 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700198 schedule();
Theodore Ts'oa931da62010-08-03 21:35:12 -0400199 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700200 }
201 finish_wait(&journal->j_wait_commit, &wait);
202 }
203
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700204 jbd_debug(1, "kjournald2 wakes\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -0700205
206 /*
207 * Were we woken up by a commit wakeup event?
208 */
209 transaction = journal->j_running_transaction;
210 if (transaction && time_after_eq(jiffies, transaction->t_expires)) {
211 journal->j_commit_request = transaction->t_tid;
212 jbd_debug(1, "woke because of timeout\n");
213 }
214 goto loop;
215
216end_loop:
Theodore Ts'oa931da62010-08-03 21:35:12 -0400217 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700218 del_timer_sync(&journal->j_commit_timer);
219 journal->j_task = NULL;
220 wake_up(&journal->j_wait_done_commit);
221 jbd_debug(1, "Journal thread exiting.\n");
222 return 0;
223}
224
Pavel Emelianov97f06782007-05-08 00:30:42 -0700225static int jbd2_journal_start_thread(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700226{
Pavel Emelianov97f06782007-05-08 00:30:42 -0700227 struct task_struct *t;
228
Theodore Ts'o90576c02009-09-29 15:51:30 -0400229 t = kthread_run(kjournald2, journal, "jbd2/%s",
230 journal->j_devname);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700231 if (IS_ERR(t))
232 return PTR_ERR(t);
233
Al Viro1076d172008-03-29 03:07:18 +0000234 wait_event(journal->j_wait_done_commit, journal->j_task != NULL);
Pavel Emelianov97f06782007-05-08 00:30:42 -0700235 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700236}
237
238static void journal_kill_thread(journal_t *journal)
239{
Theodore Ts'oa931da62010-08-03 21:35:12 -0400240 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700241 journal->j_flags |= JBD2_UNMOUNT;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700242
243 while (journal->j_task) {
244 wake_up(&journal->j_wait_commit);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400245 write_unlock(&journal->j_state_lock);
Al Viro1076d172008-03-29 03:07:18 +0000246 wait_event(journal->j_wait_done_commit, journal->j_task == NULL);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400247 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700248 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400249 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700250}
251
252/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700253 * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700254 *
255 * Writes a metadata buffer to a given disk block. The actual IO is not
256 * performed but a new buffer_head is constructed which labels the data
257 * to be written with the correct destination disk block.
258 *
259 * Any magic-number escaping which needs to be done will cause a
260 * copy-out here. If the buffer happens to start with the
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700261 * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700262 * magic number is only written to the log for descripter blocks. In
263 * this case, we copy the data and replace the first word with 0, and we
264 * return a result code which indicates that this buffer needs to be
265 * marked as an escaped buffer in the corresponding log descriptor
266 * block. The missing word can then be restored when the block is read
267 * during recovery.
268 *
269 * If the source buffer has already been modified by a new transaction
270 * since we took the last commit snapshot, we use the frozen copy of
271 * that data for IO. If we end up using the existing buffer_head's data
272 * for the write, then we *have* to lock the buffer to prevent anyone
273 * else from using and possibly modifying it while the IO is in
274 * progress.
275 *
276 * The function returns a pointer to the buffer_heads to be used for IO.
277 *
278 * We assume that the journal has already been locked in this function.
279 *
280 * Return value:
281 * <0: Error
282 * >=0: Finished OK
283 *
284 * On success:
285 * Bit 0 set == escape performed on the data
286 * Bit 1 set == buffer copy-out performed (kfree the data after IO)
287 */
288
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700289int jbd2_journal_write_metadata_buffer(transaction_t *transaction,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700290 struct journal_head *jh_in,
291 struct journal_head **jh_out,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700292 unsigned long long blocknr)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700293{
294 int need_copy_out = 0;
295 int done_copy_out = 0;
296 int do_escape = 0;
297 char *mapped_data;
298 struct buffer_head *new_bh;
299 struct journal_head *new_jh;
300 struct page *new_page;
301 unsigned int new_offset;
302 struct buffer_head *bh_in = jh2bh(jh_in);
dingdinghua96577c42009-07-13 17:55:35 -0400303 journal_t *journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700304
305 /*
306 * The buffer really shouldn't be locked: only the current committing
307 * transaction is allowed to write it, so nobody else is allowed
308 * to do any IO.
309 *
310 * akpm: except if we're journalling data, and write() output is
311 * also part of a shared mapping, and another thread has
312 * decided to launch a writepage() against this buffer.
313 */
314 J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in));
315
Theodore Ts'o47def822010-07-27 11:56:05 -0400316retry_alloc:
317 new_bh = alloc_buffer_head(GFP_NOFS);
318 if (!new_bh) {
319 /*
320 * Failure is not an option, but __GFP_NOFAIL is going
321 * away; so we retry ourselves here.
322 */
323 congestion_wait(BLK_RW_ASYNC, HZ/50);
324 goto retry_alloc;
325 }
326
dingdinghua96577c42009-07-13 17:55:35 -0400327 /* keep subsequent assertions sane */
328 new_bh->b_state = 0;
329 init_buffer(new_bh, NULL, NULL);
330 atomic_set(&new_bh->b_count, 1);
331 new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700332
333 /*
334 * If a new transaction has already done a buffer copy-out, then
335 * we use that version of the data for the commit.
336 */
337 jbd_lock_bh_state(bh_in);
338repeat:
339 if (jh_in->b_frozen_data) {
340 done_copy_out = 1;
341 new_page = virt_to_page(jh_in->b_frozen_data);
342 new_offset = offset_in_page(jh_in->b_frozen_data);
343 } else {
344 new_page = jh2bh(jh_in)->b_page;
345 new_offset = offset_in_page(jh2bh(jh_in)->b_data);
346 }
347
348 mapped_data = kmap_atomic(new_page, KM_USER0);
349 /*
Jan Kara13ceef02010-07-14 07:56:33 +0200350 * Fire data frozen trigger if data already wasn't frozen. Do this
351 * before checking for escaping, as the trigger may modify the magic
352 * offset. If a copy-out happens afterwards, it will have the correct
353 * data in the buffer.
Joel Beckere06c8222008-09-11 15:35:47 -0700354 */
Jan Kara13ceef02010-07-14 07:56:33 +0200355 if (!done_copy_out)
356 jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset,
357 jh_in->b_triggers);
Joel Beckere06c8222008-09-11 15:35:47 -0700358
359 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700360 * Check for escaping
361 */
362 if (*((__be32 *)(mapped_data + new_offset)) ==
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700363 cpu_to_be32(JBD2_MAGIC_NUMBER)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700364 need_copy_out = 1;
365 do_escape = 1;
366 }
367 kunmap_atomic(mapped_data, KM_USER0);
368
369 /*
370 * Do we need to do a data copy?
371 */
372 if (need_copy_out && !done_copy_out) {
373 char *tmp;
374
375 jbd_unlock_bh_state(bh_in);
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400376 tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS);
Theodore Ts'oe6ec1162009-12-01 09:04:42 -0500377 if (!tmp) {
378 jbd2_journal_put_journal_head(new_jh);
379 return -ENOMEM;
380 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700381 jbd_lock_bh_state(bh_in);
382 if (jh_in->b_frozen_data) {
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400383 jbd2_free(tmp, bh_in->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700384 goto repeat;
385 }
386
387 jh_in->b_frozen_data = tmp;
388 mapped_data = kmap_atomic(new_page, KM_USER0);
389 memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size);
390 kunmap_atomic(mapped_data, KM_USER0);
391
392 new_page = virt_to_page(tmp);
393 new_offset = offset_in_page(tmp);
394 done_copy_out = 1;
Joel Beckere06c8222008-09-11 15:35:47 -0700395
396 /*
397 * This isn't strictly necessary, as we're using frozen
398 * data for the escaping, but it keeps consistency with
399 * b_frozen_data usage.
400 */
401 jh_in->b_frozen_triggers = jh_in->b_triggers;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700402 }
403
404 /*
405 * Did we need to do an escaping? Now we've done all the
406 * copying, we can finally do so.
407 */
408 if (do_escape) {
409 mapped_data = kmap_atomic(new_page, KM_USER0);
410 *((unsigned int *)(mapped_data + new_offset)) = 0;
411 kunmap_atomic(mapped_data, KM_USER0);
412 }
413
Dave Kleikamp470decc2006-10-11 01:20:57 -0700414 set_bh_page(new_bh, new_page, new_offset);
415 new_jh->b_transaction = NULL;
416 new_bh->b_size = jh2bh(jh_in)->b_size;
417 new_bh->b_bdev = transaction->t_journal->j_dev;
418 new_bh->b_blocknr = blocknr;
419 set_buffer_mapped(new_bh);
420 set_buffer_dirty(new_bh);
421
422 *jh_out = new_jh;
423
424 /*
425 * The to-be-written buffer needs to get moved to the io queue,
426 * and the original buffer whose contents we are shadowing or
427 * copying is moved to the transaction's shadow queue.
428 */
429 JBUFFER_TRACE(jh_in, "file as BJ_Shadow");
dingdinghua96577c42009-07-13 17:55:35 -0400430 spin_lock(&journal->j_list_lock);
431 __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow);
432 spin_unlock(&journal->j_list_lock);
433 jbd_unlock_bh_state(bh_in);
434
Dave Kleikamp470decc2006-10-11 01:20:57 -0700435 JBUFFER_TRACE(new_jh, "file as BJ_IO");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700436 jbd2_journal_file_buffer(new_jh, transaction, BJ_IO);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700437
438 return do_escape | (done_copy_out << 1);
439}
440
441/*
442 * Allocation code for the journal file. Manage the space left in the
443 * journal, so that we can begin checkpointing when appropriate.
444 */
445
446/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700447 * __jbd2_log_space_left: Return the number of free blocks left in the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700448 *
449 * Called with the journal already locked.
450 *
451 * Called under j_state_lock
452 */
453
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700454int __jbd2_log_space_left(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700455{
456 int left = journal->j_free;
457
Theodore Ts'oa931da62010-08-03 21:35:12 -0400458 /* assert_spin_locked(&journal->j_state_lock); */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700459
460 /*
461 * Be pessimistic here about the number of those free blocks which
462 * might be required for log descriptor control blocks.
463 */
464
465#define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */
466
467 left -= MIN_LOG_RESERVED_BLOCKS;
468
469 if (left <= 0)
470 return 0;
471 left -= (left >> 3);
472 return left;
473}
474
475/*
Theodore Ts'oe4471832011-02-12 08:18:24 -0500476 * Called with j_state_lock locked for writing.
477 * Returns true if a transaction commit was started.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700478 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700479int __jbd2_log_start_commit(journal_t *journal, tid_t target)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700480{
481 /*
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400482 * The only transaction we can possibly wait upon is the
483 * currently running transaction (if it exists). Otherwise,
484 * the target tid must be an old one.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700485 */
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400486 if (journal->j_running_transaction &&
487 journal->j_running_transaction->t_tid == target) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700488 /*
Andrea Gelminibcf3d0b2010-10-16 15:19:14 +0200489 * We want a new commit: OK, mark the request and wakeup the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700490 * commit thread. We do _not_ do the commit ourselves.
491 */
492
493 journal->j_commit_request = target;
Eryu Guanf2a44522011-11-01 19:09:18 -0400494 jbd_debug(1, "JBD2: requesting commit %d/%d\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700495 journal->j_commit_request,
496 journal->j_commit_sequence);
497 wake_up(&journal->j_wait_commit);
498 return 1;
Theodore Ts'odeeeaf12011-05-01 18:16:26 -0400499 } else if (!tid_geq(journal->j_commit_request, target))
500 /* This should never happen, but if it does, preserve
501 the evidence before kjournald goes into a loop and
502 increments j_commit_sequence beyond all recognition. */
Eryu Guanf2a44522011-11-01 19:09:18 -0400503 WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n",
Theodore Ts'o1be2add2011-05-08 19:37:54 -0400504 journal->j_commit_request,
505 journal->j_commit_sequence,
506 target, journal->j_running_transaction ?
507 journal->j_running_transaction->t_tid : 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700508 return 0;
509}
510
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700511int jbd2_log_start_commit(journal_t *journal, tid_t tid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700512{
513 int ret;
514
Theodore Ts'oa931da62010-08-03 21:35:12 -0400515 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700516 ret = __jbd2_log_start_commit(journal, tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400517 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700518 return ret;
519}
520
521/*
522 * Force and wait upon a commit if the calling process is not within
523 * transaction. This is used for forcing out undo-protected data which contains
524 * bitmaps, when the fs is running out of space.
525 *
526 * We can only force the running transaction if we don't have an active handle;
527 * otherwise, we will deadlock.
528 *
529 * Returns true if a transaction was started.
530 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700531int jbd2_journal_force_commit_nested(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700532{
533 transaction_t *transaction = NULL;
534 tid_t tid;
Theodore Ts'oe4471832011-02-12 08:18:24 -0500535 int need_to_start = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700536
Theodore Ts'oa931da62010-08-03 21:35:12 -0400537 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700538 if (journal->j_running_transaction && !current->journal_info) {
539 transaction = journal->j_running_transaction;
Theodore Ts'oe4471832011-02-12 08:18:24 -0500540 if (!tid_geq(journal->j_commit_request, transaction->t_tid))
541 need_to_start = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700542 } else if (journal->j_committing_transaction)
543 transaction = journal->j_committing_transaction;
544
545 if (!transaction) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400546 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700547 return 0; /* Nothing to retry */
548 }
549
550 tid = transaction->t_tid;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400551 read_unlock(&journal->j_state_lock);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500552 if (need_to_start)
553 jbd2_log_start_commit(journal, tid);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700554 jbd2_log_wait_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700555 return 1;
556}
557
558/*
559 * Start a commit of the current running transaction (if any). Returns true
Jan Karac88ccea2009-02-10 11:27:46 -0500560 * if a transaction is going to be committed (or is currently already
561 * committing), and fills its tid in at *ptid
Dave Kleikamp470decc2006-10-11 01:20:57 -0700562 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700563int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700564{
565 int ret = 0;
566
Theodore Ts'oa931da62010-08-03 21:35:12 -0400567 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700568 if (journal->j_running_transaction) {
569 tid_t tid = journal->j_running_transaction->t_tid;
570
Jan Karac88ccea2009-02-10 11:27:46 -0500571 __jbd2_log_start_commit(journal, tid);
572 /* There's a running transaction and we've just made sure
573 * it's commit has been scheduled. */
574 if (ptid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700575 *ptid = tid;
Jan Karac88ccea2009-02-10 11:27:46 -0500576 ret = 1;
577 } else if (journal->j_committing_transaction) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700578 /*
579 * If ext3_write_super() recently started a commit, then we
580 * have to wait for completion of that transaction
581 */
Jan Karac88ccea2009-02-10 11:27:46 -0500582 if (ptid)
583 *ptid = journal->j_committing_transaction->t_tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700584 ret = 1;
585 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400586 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700587 return ret;
588}
589
590/*
Jan Karabbd2be32011-05-24 11:59:18 -0400591 * Return 1 if a given transaction has not yet sent barrier request
592 * connected with a transaction commit. If 0 is returned, transaction
593 * may or may not have sent the barrier. Used to avoid sending barrier
594 * twice in common cases.
595 */
596int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid)
597{
598 int ret = 0;
599 transaction_t *commit_trans;
600
601 if (!(journal->j_flags & JBD2_BARRIER))
602 return 0;
603 read_lock(&journal->j_state_lock);
604 /* Transaction already committed? */
605 if (tid_geq(journal->j_commit_sequence, tid))
606 goto out;
607 commit_trans = journal->j_committing_transaction;
608 if (!commit_trans || commit_trans->t_tid != tid) {
609 ret = 1;
610 goto out;
611 }
612 /*
613 * Transaction is being committed and we already proceeded to
614 * submitting a flush to fs partition?
615 */
616 if (journal->j_fs_dev != journal->j_dev) {
617 if (!commit_trans->t_need_data_flush ||
618 commit_trans->t_state >= T_COMMIT_DFLUSH)
619 goto out;
620 } else {
621 if (commit_trans->t_state >= T_COMMIT_JFLUSH)
622 goto out;
623 }
624 ret = 1;
625out:
626 read_unlock(&journal->j_state_lock);
627 return ret;
628}
629EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier);
630
631/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700632 * Wait for a specified commit to complete.
633 * The caller may not hold the journal lock.
634 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700635int jbd2_log_wait_commit(journal_t *journal, tid_t tid)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700636{
637 int err = 0;
638
Theodore Ts'oa931da62010-08-03 21:35:12 -0400639 read_lock(&journal->j_state_lock);
Jose R. Santose23291b2007-07-18 08:57:06 -0400640#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -0700641 if (!tid_geq(journal->j_commit_request, tid)) {
642 printk(KERN_EMERG
643 "%s: error: j_commit_request=%d, tid=%d\n",
Harvey Harrison329d2912008-04-17 10:38:59 -0400644 __func__, journal->j_commit_request, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700645 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700646#endif
Dave Kleikamp470decc2006-10-11 01:20:57 -0700647 while (tid_gt(tid, journal->j_commit_sequence)) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400648 jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -0700649 tid, journal->j_commit_sequence);
650 wake_up(&journal->j_wait_commit);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400651 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700652 wait_event(journal->j_wait_done_commit,
653 !tid_gt(tid, journal->j_commit_sequence));
Theodore Ts'oa931da62010-08-03 21:35:12 -0400654 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700655 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400656 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700657
658 if (unlikely(is_journal_aborted(journal))) {
659 printk(KERN_EMERG "journal commit I/O error\n");
660 err = -EIO;
661 }
662 return err;
663}
664
665/*
666 * Log buffer allocation routines:
667 */
668
Mingming Cao18eba7a2006-10-11 01:21:13 -0700669int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700670{
671 unsigned long blocknr;
672
Theodore Ts'oa931da62010-08-03 21:35:12 -0400673 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700674 J_ASSERT(journal->j_free > 1);
675
676 blocknr = journal->j_head;
677 journal->j_head++;
678 journal->j_free--;
679 if (journal->j_head == journal->j_last)
680 journal->j_head = journal->j_first;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400681 write_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700682 return jbd2_journal_bmap(journal, blocknr, retp);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700683}
684
685/*
686 * Conversion of logical to physical block numbers for the journal
687 *
688 * On external journals the journal blocks are identity-mapped, so
689 * this is a no-op. If needed, we can use j_blk_offset - everything is
690 * ready.
691 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700692int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700693 unsigned long long *retp)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700694{
695 int err = 0;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700696 unsigned long long ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700697
698 if (journal->j_inode) {
699 ret = bmap(journal->j_inode, blocknr);
700 if (ret)
701 *retp = ret;
702 else {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700703 printk(KERN_ALERT "%s: journal block not found "
704 "at offset %lu on %s\n",
Theodore Ts'o05496762008-09-16 14:36:17 -0400705 __func__, blocknr, journal->j_devname);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700706 err = -EIO;
707 __journal_abort_soft(journal, err);
708 }
709 } else {
710 *retp = blocknr; /* +journal->j_blk_offset */
711 }
712 return err;
713}
714
715/*
716 * We play buffer_head aliasing tricks to write data/metadata blocks to
717 * the journal without copying their contents, but for journal
718 * descriptor blocks we do need to generate bona fide buffers.
719 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700720 * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying
Dave Kleikamp470decc2006-10-11 01:20:57 -0700721 * the buffer's contents they really should run flush_dcache_page(bh->b_page).
722 * But we don't bother doing that, so there will be coherency problems with
723 * mmaps of blockdevs which hold live JBD-controlled filesystems.
724 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700725struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700726{
727 struct buffer_head *bh;
Mingming Cao18eba7a2006-10-11 01:21:13 -0700728 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700729 int err;
730
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700731 err = jbd2_journal_next_log_block(journal, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700732
733 if (err)
734 return NULL;
735
736 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -0500737 if (!bh)
738 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700739 lock_buffer(bh);
740 memset(bh->b_data, 0, journal->j_blocksize);
741 set_buffer_uptodate(bh);
742 unlock_buffer(bh);
743 BUFFER_TRACE(bh, "return this buffer");
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700744 return jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700745}
746
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500747struct jbd2_stats_proc_session {
748 journal_t *journal;
749 struct transaction_stats_s *stats;
750 int start;
751 int max;
752};
753
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500754static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos)
755{
756 return *pos ? NULL : SEQ_START_TOKEN;
757}
758
759static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos)
760{
761 return NULL;
762}
763
764static int jbd2_seq_info_show(struct seq_file *seq, void *v)
765{
766 struct jbd2_stats_proc_session *s = seq->private;
767
768 if (v != SEQ_START_TOKEN)
769 return 0;
Theodore Ts'obf699322009-09-30 00:32:06 -0400770 seq_printf(seq, "%lu transaction, each up to %u blocks\n",
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500771 s->stats->ts_tid,
772 s->journal->j_max_transaction_buffers);
773 if (s->stats->ts_tid == 0)
774 return 0;
775 seq_printf(seq, "average: \n %ums waiting for transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400776 jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500777 seq_printf(seq, " %ums running transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400778 jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500779 seq_printf(seq, " %ums transaction was being locked\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400780 jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500781 seq_printf(seq, " %ums flushing data (in ordered mode)\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400782 jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500783 seq_printf(seq, " %ums logging transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400784 jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid));
Simon Holm Thøgersenc225aa52009-01-11 22:34:01 -0500785 seq_printf(seq, " %lluus average transaction commit time\n",
786 div_u64(s->journal->j_average_commit_time, 1000));
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500787 seq_printf(seq, " %lu handles per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400788 s->stats->run.rs_handle_count / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500789 seq_printf(seq, " %lu blocks per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400790 s->stats->run.rs_blocks / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500791 seq_printf(seq, " %lu logged blocks per transaction\n",
Theodore Ts'obf699322009-09-30 00:32:06 -0400792 s->stats->run.rs_blocks_logged / s->stats->ts_tid);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500793 return 0;
794}
795
796static void jbd2_seq_info_stop(struct seq_file *seq, void *v)
797{
798}
799
James Morris88e9d342009-09-22 16:43:43 -0700800static const struct seq_operations jbd2_seq_info_ops = {
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500801 .start = jbd2_seq_info_start,
802 .next = jbd2_seq_info_next,
803 .stop = jbd2_seq_info_stop,
804 .show = jbd2_seq_info_show,
805};
806
807static int jbd2_seq_info_open(struct inode *inode, struct file *file)
808{
809 journal_t *journal = PDE(inode)->data;
810 struct jbd2_stats_proc_session *s;
811 int rc, size;
812
813 s = kmalloc(sizeof(*s), GFP_KERNEL);
814 if (s == NULL)
815 return -ENOMEM;
816 size = sizeof(struct transaction_stats_s);
817 s->stats = kmalloc(size, GFP_KERNEL);
818 if (s->stats == NULL) {
819 kfree(s);
820 return -ENOMEM;
821 }
822 spin_lock(&journal->j_history_lock);
823 memcpy(s->stats, &journal->j_stats, size);
824 s->journal = journal;
825 spin_unlock(&journal->j_history_lock);
826
827 rc = seq_open(file, &jbd2_seq_info_ops);
828 if (rc == 0) {
829 struct seq_file *m = file->private_data;
830 m->private = s;
831 } else {
832 kfree(s->stats);
833 kfree(s);
834 }
835 return rc;
836
837}
838
839static int jbd2_seq_info_release(struct inode *inode, struct file *file)
840{
841 struct seq_file *seq = file->private_data;
842 struct jbd2_stats_proc_session *s = seq->private;
843 kfree(s->stats);
844 kfree(s);
845 return seq_release(inode, file);
846}
847
Alexey Dobriyan828c0952009-10-01 15:43:56 -0700848static const struct file_operations jbd2_seq_info_fops = {
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500849 .owner = THIS_MODULE,
850 .open = jbd2_seq_info_open,
851 .read = seq_read,
852 .llseek = seq_lseek,
853 .release = jbd2_seq_info_release,
854};
855
856static struct proc_dir_entry *proc_jbd2_stats;
857
858static void jbd2_stats_proc_init(journal_t *journal)
859{
Theodore Ts'o05496762008-09-16 14:36:17 -0400860 journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500861 if (journal->j_proc_entry) {
Denis V. Lunev79da3662008-04-29 01:02:11 -0700862 proc_create_data("info", S_IRUGO, journal->j_proc_entry,
863 &jbd2_seq_info_fops, journal);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500864 }
865}
866
867static void jbd2_stats_proc_exit(journal_t *journal)
868{
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500869 remove_proc_entry("info", journal->j_proc_entry);
Theodore Ts'o05496762008-09-16 14:36:17 -0400870 remove_proc_entry(journal->j_devname, proc_jbd2_stats);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500871}
872
Dave Kleikamp470decc2006-10-11 01:20:57 -0700873/*
874 * Management for journal control blocks: functions to create and
875 * destroy journal_t structures, and to initialise and read existing
876 * journal blocks from disk. */
877
878/* First: create and setup a journal_t object in memory. We initialise
879 * very few fields yet: that has to wait until we have created the
880 * journal structures from from scratch, or loaded them from disk. */
881
882static journal_t * journal_init_common (void)
883{
884 journal_t *journal;
885 int err;
886
Andrew Morton3ebfdf82009-12-23 08:05:15 -0500887 journal = kzalloc(sizeof(*journal), GFP_KERNEL);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700888 if (!journal)
Theodore Ts'ob7271b02010-12-18 13:39:38 -0500889 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700890
891 init_waitqueue_head(&journal->j_wait_transaction_locked);
892 init_waitqueue_head(&journal->j_wait_logspace);
893 init_waitqueue_head(&journal->j_wait_done_commit);
894 init_waitqueue_head(&journal->j_wait_checkpoint);
895 init_waitqueue_head(&journal->j_wait_commit);
896 init_waitqueue_head(&journal->j_wait_updates);
897 mutex_init(&journal->j_barrier);
898 mutex_init(&journal->j_checkpoint_mutex);
899 spin_lock_init(&journal->j_revoke_lock);
900 spin_lock_init(&journal->j_list_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400901 rwlock_init(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700902
Mingming Caocd02ff02007-10-16 18:38:25 -0400903 journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE);
Theodore Ts'o30773842009-01-03 20:27:38 -0500904 journal->j_min_batch_time = 0;
905 journal->j_max_batch_time = 15000; /* 15ms */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700906
907 /* The journal is marked for error until we succeed with recovery! */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700908 journal->j_flags = JBD2_ABORT;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700909
910 /* Set up a default-sized revoke table for the new mount. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700911 err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700912 if (err) {
913 kfree(journal);
Theodore Ts'ob7271b02010-12-18 13:39:38 -0500914 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700915 }
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500916
Theodore Ts'obf699322009-09-30 00:32:06 -0400917 spin_lock_init(&journal->j_history_lock);
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500918
Dave Kleikamp470decc2006-10-11 01:20:57 -0700919 return journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700920}
921
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700922/* jbd2_journal_init_dev and jbd2_journal_init_inode:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700923 *
924 * Create a journal structure assigned some fixed set of disk blocks to
925 * the journal. We don't actually touch those disk blocks yet, but we
926 * need to set up all of the mapping information to tell the journaling
927 * system where the journal blocks are.
928 *
929 */
930
931/**
Randy Dunlap5648ba52008-04-17 10:38:59 -0400932 * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure
Dave Kleikamp470decc2006-10-11 01:20:57 -0700933 * @bdev: Block device on which to create the journal
934 * @fs_dev: Device which hold journalled filesystem for this journal.
935 * @start: Block nr Start of journal.
936 * @len: Length of the journal in blocks.
937 * @blocksize: blocksize of journalling device
Randy Dunlap5648ba52008-04-17 10:38:59 -0400938 *
939 * Returns: a newly created journal_t *
Dave Kleikamp470decc2006-10-11 01:20:57 -0700940 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700941 * jbd2_journal_init_dev creates a journal which maps a fixed contiguous
Dave Kleikamp470decc2006-10-11 01:20:57 -0700942 * range of blocks on an arbitrary block device.
943 *
944 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700945journal_t * jbd2_journal_init_dev(struct block_device *bdev,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700946 struct block_device *fs_dev,
Mingming Cao18eba7a2006-10-11 01:21:13 -0700947 unsigned long long start, int len, int blocksize)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700948{
949 journal_t *journal = journal_init_common();
950 struct buffer_head *bh;
Theodore Ts'o05496762008-09-16 14:36:17 -0400951 char *p;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700952 int n;
953
954 if (!journal)
955 return NULL;
956
957 /* journal descriptor can store up to n blocks -bzzz */
958 journal->j_blocksize = blocksize;
yangsheng0587aa32010-11-17 21:46:26 -0500959 journal->j_dev = bdev;
960 journal->j_fs_dev = fs_dev;
961 journal->j_blk_offset = start;
962 journal->j_maxlen = len;
963 bdevname(journal->j_dev, journal->j_devname);
964 p = journal->j_devname;
965 while ((p = strchr(p, '/')))
966 *p = '!';
Jan Kara4b905672009-01-06 14:53:35 -0500967 jbd2_stats_proc_init(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700968 n = journal->j_blocksize / sizeof(journal_block_tag_t);
969 journal->j_wbufsize = n;
970 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
971 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300972 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison329d2912008-04-17 10:38:59 -0400973 __func__);
Jan Kara4b905672009-01-06 14:53:35 -0500974 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700975 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700976
977 bh = __getblk(journal->j_dev, start, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -0500978 if (!bh) {
979 printk(KERN_ERR
980 "%s: Cannot get buffer for journal superblock\n",
981 __func__);
982 goto out_err;
983 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700984 journal->j_sb_buffer = bh;
985 journal->j_superblock = (journal_superblock_t *)bh->b_data;
Jan Kara4b905672009-01-06 14:53:35 -0500986
Dave Kleikamp470decc2006-10-11 01:20:57 -0700987 return journal;
Jan Kara4b905672009-01-06 14:53:35 -0500988out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +0800989 kfree(journal->j_wbuf);
Jan Kara4b905672009-01-06 14:53:35 -0500990 jbd2_stats_proc_exit(journal);
991 kfree(journal);
992 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700993}
994
995/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700996 * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700997 * @inode: An inode to create the journal in
998 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700999 * jbd2_journal_init_inode creates a journal which maps an on-disk inode as
Dave Kleikamp470decc2006-10-11 01:20:57 -07001000 * the journal. The inode must exist already, must support bmap() and
1001 * must have all data blocks preallocated.
1002 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001003journal_t * jbd2_journal_init_inode (struct inode *inode)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001004{
1005 struct buffer_head *bh;
1006 journal_t *journal = journal_init_common();
Theodore Ts'o05496762008-09-16 14:36:17 -04001007 char *p;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001008 int err;
1009 int n;
Mingming Cao18eba7a2006-10-11 01:21:13 -07001010 unsigned long long blocknr;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001011
1012 if (!journal)
1013 return NULL;
1014
1015 journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev;
1016 journal->j_inode = inode;
Theodore Ts'o05496762008-09-16 14:36:17 -04001017 bdevname(journal->j_dev, journal->j_devname);
1018 p = journal->j_devname;
1019 while ((p = strchr(p, '/')))
1020 *p = '!';
1021 p = journal->j_devname + strlen(journal->j_devname);
Theodore Ts'o90576c02009-09-29 15:51:30 -04001022 sprintf(p, "-%lu", journal->j_inode->i_ino);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001023 jbd_debug(1,
1024 "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n",
1025 journal, inode->i_sb->s_id, inode->i_ino,
1026 (long long) inode->i_size,
1027 inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize);
1028
1029 journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits;
1030 journal->j_blocksize = inode->i_sb->s_blocksize;
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001031 jbd2_stats_proc_init(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001032
1033 /* journal descriptor can store up to n blocks -bzzz */
1034 n = journal->j_blocksize / sizeof(journal_block_tag_t);
1035 journal->j_wbufsize = n;
1036 journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL);
1037 if (!journal->j_wbuf) {
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001038 printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001039 __func__);
Jan Kara4b905672009-01-06 14:53:35 -05001040 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001041 }
1042
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001043 err = jbd2_journal_bmap(journal, 0, &blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001044 /* If that failed, give up */
1045 if (err) {
Justin P. Mattock3c26bdb2011-02-26 20:34:05 -08001046 printk(KERN_ERR "%s: Cannot locate journal superblock\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001047 __func__);
Jan Kara4b905672009-01-06 14:53:35 -05001048 goto out_err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001049 }
1050
1051 bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize);
Jan Kara4b905672009-01-06 14:53:35 -05001052 if (!bh) {
1053 printk(KERN_ERR
1054 "%s: Cannot get buffer for journal superblock\n",
1055 __func__);
1056 goto out_err;
1057 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001058 journal->j_sb_buffer = bh;
1059 journal->j_superblock = (journal_superblock_t *)bh->b_data;
1060
1061 return journal;
Jan Kara4b905672009-01-06 14:53:35 -05001062out_err:
Tao Ma7b02bec2009-11-10 17:13:22 +08001063 kfree(journal->j_wbuf);
Jan Kara4b905672009-01-06 14:53:35 -05001064 jbd2_stats_proc_exit(journal);
1065 kfree(journal);
1066 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001067}
1068
1069/*
1070 * If the journal init or create aborts, we need to mark the journal
1071 * superblock as being NULL to prevent the journal destroy from writing
1072 * back a bogus superblock.
1073 */
1074static void journal_fail_superblock (journal_t *journal)
1075{
1076 struct buffer_head *bh = journal->j_sb_buffer;
1077 brelse(bh);
1078 journal->j_sb_buffer = NULL;
1079}
1080
1081/*
1082 * Given a journal_t structure, initialise the various fields for
1083 * startup of a new journaling session. We use this both when creating
1084 * a journal, and after recovering an old journal to reset it for
1085 * subsequent use.
1086 */
1087
1088static int journal_reset(journal_t *journal)
1089{
1090 journal_superblock_t *sb = journal->j_superblock;
Mingming Cao18eba7a2006-10-11 01:21:13 -07001091 unsigned long long first, last;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001092
1093 first = be32_to_cpu(sb->s_first);
1094 last = be32_to_cpu(sb->s_maxlen);
Jan Karaf6f50e22009-07-17 10:40:01 -04001095 if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001096 printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n",
Jan Karaf6f50e22009-07-17 10:40:01 -04001097 first, last);
1098 journal_fail_superblock(journal);
1099 return -EINVAL;
1100 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001101
1102 journal->j_first = first;
1103 journal->j_last = last;
1104
1105 journal->j_head = first;
1106 journal->j_tail = first;
1107 journal->j_free = last - first;
1108
1109 journal->j_tail_sequence = journal->j_transaction_sequence;
1110 journal->j_commit_sequence = journal->j_transaction_sequence - 1;
1111 journal->j_commit_request = journal->j_commit_sequence;
1112
1113 journal->j_max_transaction_buffers = journal->j_maxlen / 4;
1114
1115 /* Add the dynamic fields and write it to disk. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001116 jbd2_journal_update_superblock(journal, 1);
Pavel Emelianov97f06782007-05-08 00:30:42 -07001117 return jbd2_journal_start_thread(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001118}
1119
1120/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001121 * void jbd2_journal_update_superblock() - Update journal sb on disk.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001122 * @journal: The journal to update.
1123 * @wait: Set to '0' if you don't want to wait for IO completion.
1124 *
1125 * Update a journal's dynamic superblock fields and write it to disk,
1126 * optionally waiting for the IO to complete.
1127 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001128void jbd2_journal_update_superblock(journal_t *journal, int wait)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001129{
1130 journal_superblock_t *sb = journal->j_superblock;
1131 struct buffer_head *bh = journal->j_sb_buffer;
1132
1133 /*
1134 * As a special case, if the on-disk copy is already marked as needing
1135 * no recovery (s_start == 0) and there are no outstanding transactions
1136 * in the filesystem, then we can safely defer the superblock update
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001137 * until the next commit by setting JBD2_FLUSHED. This avoids
Dave Kleikamp470decc2006-10-11 01:20:57 -07001138 * attempting a write to a potential-readonly device.
1139 */
1140 if (sb->s_start == 0 && journal->j_tail_sequence ==
1141 journal->j_transaction_sequence) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001142 jbd_debug(1, "JBD2: Skipping superblock update on recovered sb "
Dave Kleikamp470decc2006-10-11 01:20:57 -07001143 "(start %ld, seq %d, errno %d)\n",
1144 journal->j_tail, journal->j_tail_sequence,
1145 journal->j_errno);
1146 goto out;
1147 }
1148
Theodore Ts'o914258b2008-10-06 21:35:40 -04001149 if (buffer_write_io_error(bh)) {
1150 /*
1151 * Oh, dear. A previous attempt to write the journal
1152 * superblock failed. This could happen because the
1153 * USB device was yanked out. Or it could happen to
1154 * be a transient write error and maybe the block will
1155 * be remapped. Nothing we can do but to retry the
1156 * write and hope for the best.
1157 */
1158 printk(KERN_ERR "JBD2: previous I/O error detected "
1159 "for journal superblock update for %s.\n",
1160 journal->j_devname);
1161 clear_buffer_write_io_error(bh);
1162 set_buffer_uptodate(bh);
1163 }
1164
Theodore Ts'oa931da62010-08-03 21:35:12 -04001165 read_lock(&journal->j_state_lock);
Eryu Guanf2a44522011-11-01 19:09:18 -04001166 jbd_debug(1, "JBD2: updating superblock (start %ld, seq %d, errno %d)\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -07001167 journal->j_tail, journal->j_tail_sequence, journal->j_errno);
1168
1169 sb->s_sequence = cpu_to_be32(journal->j_tail_sequence);
1170 sb->s_start = cpu_to_be32(journal->j_tail);
1171 sb->s_errno = cpu_to_be32(journal->j_errno);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001172 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001173
1174 BUFFER_TRACE(bh, "marking dirty");
1175 mark_buffer_dirty(bh);
Theodore Ts'o914258b2008-10-06 21:35:40 -04001176 if (wait) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001177 sync_dirty_buffer(bh);
Theodore Ts'o914258b2008-10-06 21:35:40 -04001178 if (buffer_write_io_error(bh)) {
1179 printk(KERN_ERR "JBD2: I/O error detected "
1180 "when updating journal superblock for %s.\n",
1181 journal->j_devname);
1182 clear_buffer_write_io_error(bh);
1183 set_buffer_uptodate(bh);
1184 }
1185 } else
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02001186 write_dirty_buffer(bh, WRITE);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001187
1188out:
1189 /* If we have just flushed the log (by marking s_start==0), then
1190 * any future commit will have to be careful to update the
1191 * superblock again to re-record the true start of the log. */
1192
Theodore Ts'oa931da62010-08-03 21:35:12 -04001193 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001194 if (sb->s_start)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001195 journal->j_flags &= ~JBD2_FLUSHED;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001196 else
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001197 journal->j_flags |= JBD2_FLUSHED;
Theodore Ts'oa931da62010-08-03 21:35:12 -04001198 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001199}
1200
1201/*
1202 * Read the superblock for a given journal, performing initial
1203 * validation of the format.
1204 */
1205
1206static int journal_get_superblock(journal_t *journal)
1207{
1208 struct buffer_head *bh;
1209 journal_superblock_t *sb;
1210 int err = -EIO;
1211
1212 bh = journal->j_sb_buffer;
1213
1214 J_ASSERT(bh != NULL);
1215 if (!buffer_uptodate(bh)) {
1216 ll_rw_block(READ, 1, &bh);
1217 wait_on_buffer(bh);
1218 if (!buffer_uptodate(bh)) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001219 printk(KERN_ERR
1220 "JBD2: IO error reading journal superblock\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001221 goto out;
1222 }
1223 }
1224
1225 sb = journal->j_superblock;
1226
1227 err = -EINVAL;
1228
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001229 if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) ||
Dave Kleikamp470decc2006-10-11 01:20:57 -07001230 sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001231 printk(KERN_WARNING "JBD2: no valid journal superblock found\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001232 goto out;
1233 }
1234
1235 switch(be32_to_cpu(sb->s_header.h_blocktype)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001236 case JBD2_SUPERBLOCK_V1:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001237 journal->j_format_version = 1;
1238 break;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001239 case JBD2_SUPERBLOCK_V2:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001240 journal->j_format_version = 2;
1241 break;
1242 default:
Eryu Guanf2a44522011-11-01 19:09:18 -04001243 printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001244 goto out;
1245 }
1246
1247 if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen)
1248 journal->j_maxlen = be32_to_cpu(sb->s_maxlen);
1249 else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001250 printk(KERN_WARNING "JBD2: journal file too short\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001251 goto out;
1252 }
1253
Eryu Guan87622022011-11-01 19:04:59 -04001254 if (be32_to_cpu(sb->s_first) == 0 ||
1255 be32_to_cpu(sb->s_first) >= journal->j_maxlen) {
1256 printk(KERN_WARNING
1257 "JBD2: Invalid start block of journal: %u\n",
1258 be32_to_cpu(sb->s_first));
1259 goto out;
1260 }
1261
Dave Kleikamp470decc2006-10-11 01:20:57 -07001262 return 0;
1263
1264out:
1265 journal_fail_superblock(journal);
1266 return err;
1267}
1268
1269/*
1270 * Load the on-disk journal superblock and read the key fields into the
1271 * journal_t.
1272 */
1273
1274static int load_superblock(journal_t *journal)
1275{
1276 int err;
1277 journal_superblock_t *sb;
1278
1279 err = journal_get_superblock(journal);
1280 if (err)
1281 return err;
1282
1283 sb = journal->j_superblock;
1284
1285 journal->j_tail_sequence = be32_to_cpu(sb->s_sequence);
1286 journal->j_tail = be32_to_cpu(sb->s_start);
1287 journal->j_first = be32_to_cpu(sb->s_first);
1288 journal->j_last = be32_to_cpu(sb->s_maxlen);
1289 journal->j_errno = be32_to_cpu(sb->s_errno);
1290
1291 return 0;
1292}
1293
1294
1295/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001296 * int jbd2_journal_load() - Read journal from disk.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001297 * @journal: Journal to act on.
1298 *
1299 * Given a journal_t structure which tells us which disk blocks contain
1300 * a journal, read the journal from disk to initialise the in-memory
1301 * structures.
1302 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001303int jbd2_journal_load(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001304{
1305 int err;
1306 journal_superblock_t *sb;
1307
1308 err = load_superblock(journal);
1309 if (err)
1310 return err;
1311
1312 sb = journal->j_superblock;
1313 /* If this is a V2 superblock, then we have to check the
1314 * features flags on it. */
1315
1316 if (journal->j_format_version >= 2) {
1317 if ((sb->s_feature_ro_compat &
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001318 ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) ||
Dave Kleikamp470decc2006-10-11 01:20:57 -07001319 (sb->s_feature_incompat &
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001320 ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) {
Eryu Guanf2a44522011-11-01 19:09:18 -04001321 printk(KERN_WARNING
1322 "JBD2: Unrecognised features on journal\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001323 return -EINVAL;
1324 }
1325 }
1326
Theodore Ts'od2eecb02009-12-07 10:36:20 -05001327 /*
1328 * Create a slab for this blocksize
1329 */
1330 err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize));
1331 if (err)
1332 return err;
1333
Dave Kleikamp470decc2006-10-11 01:20:57 -07001334 /* Let the recovery code check whether it needs to recover any
1335 * data from the journal. */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001336 if (jbd2_journal_recover(journal))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001337 goto recovery_error;
1338
Theodore Ts'oe6a47422009-11-15 15:31:37 -05001339 if (journal->j_failed_commit) {
1340 printk(KERN_ERR "JBD2: journal transaction %u on %s "
1341 "is corrupt.\n", journal->j_failed_commit,
1342 journal->j_devname);
1343 return -EIO;
1344 }
1345
Dave Kleikamp470decc2006-10-11 01:20:57 -07001346 /* OK, we've finished with the dynamic journal bits:
1347 * reinitialise the dynamic contents of the superblock in memory
1348 * and reset them on disk. */
1349 if (journal_reset(journal))
1350 goto recovery_error;
1351
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001352 journal->j_flags &= ~JBD2_ABORT;
1353 journal->j_flags |= JBD2_LOADED;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001354 return 0;
1355
1356recovery_error:
Eryu Guanf2a44522011-11-01 19:09:18 -04001357 printk(KERN_WARNING "JBD2: recovery failed\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001358 return -EIO;
1359}
1360
1361/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001362 * void jbd2_journal_destroy() - Release a journal_t structure.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001363 * @journal: Journal to act on.
1364 *
1365 * Release a journal_t structure once it is no longer in use by the
1366 * journaled object.
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001367 * Return <0 if we couldn't clean up the journal.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001368 */
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001369int jbd2_journal_destroy(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001370{
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001371 int err = 0;
1372
Dave Kleikamp470decc2006-10-11 01:20:57 -07001373 /* Wait for the commit thread to wake up and die. */
1374 journal_kill_thread(journal);
1375
1376 /* Force a final log commit */
1377 if (journal->j_running_transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001378 jbd2_journal_commit_transaction(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001379
1380 /* Force any old transactions to disk */
1381
1382 /* Totally anal locking here... */
1383 spin_lock(&journal->j_list_lock);
1384 while (journal->j_checkpoint_transactions != NULL) {
1385 spin_unlock(&journal->j_list_lock);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -05001386 mutex_lock(&journal->j_checkpoint_mutex);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001387 jbd2_log_do_checkpoint(journal);
Theodore Ts'o1a0d3782008-11-05 00:09:22 -05001388 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001389 spin_lock(&journal->j_list_lock);
1390 }
1391
1392 J_ASSERT(journal->j_running_transaction == NULL);
1393 J_ASSERT(journal->j_committing_transaction == NULL);
1394 J_ASSERT(journal->j_checkpoint_transactions == NULL);
1395 spin_unlock(&journal->j_list_lock);
1396
Dave Kleikamp470decc2006-10-11 01:20:57 -07001397 if (journal->j_sb_buffer) {
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001398 if (!is_journal_aborted(journal)) {
1399 /* We can now mark the journal as empty. */
1400 journal->j_tail = 0;
1401 journal->j_tail_sequence =
1402 ++journal->j_transaction_sequence;
1403 jbd2_journal_update_superblock(journal, 1);
1404 } else {
1405 err = -EIO;
1406 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001407 brelse(journal->j_sb_buffer);
1408 }
1409
Johann Lombardi8e85fb32008-01-28 23:58:27 -05001410 if (journal->j_proc_entry)
1411 jbd2_stats_proc_exit(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001412 if (journal->j_inode)
1413 iput(journal->j_inode);
1414 if (journal->j_revoke)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001415 jbd2_journal_destroy_revoke(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001416 kfree(journal->j_wbuf);
1417 kfree(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001418
1419 return err;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001420}
1421
1422
1423/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001424 *int jbd2_journal_check_used_features () - Check if features specified are used.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001425 * @journal: Journal to check.
1426 * @compat: bitmask of compatible features
1427 * @ro: bitmask of features that force read-only mount
1428 * @incompat: bitmask of incompatible features
1429 *
1430 * Check whether the journal uses all of a given set of
1431 * features. Return true (non-zero) if it does.
1432 **/
1433
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001434int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001435 unsigned long ro, unsigned long incompat)
1436{
1437 journal_superblock_t *sb;
1438
1439 if (!compat && !ro && !incompat)
1440 return 1;
Patrick J. LoPresti1113e1b2010-07-22 15:04:16 -07001441 /* Load journal superblock if it is not loaded yet. */
1442 if (journal->j_format_version == 0 &&
1443 journal_get_superblock(journal) != 0)
1444 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001445 if (journal->j_format_version == 1)
1446 return 0;
1447
1448 sb = journal->j_superblock;
1449
1450 if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) &&
1451 ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) &&
1452 ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat))
1453 return 1;
1454
1455 return 0;
1456}
1457
1458/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001459 * int jbd2_journal_check_available_features() - Check feature set in journalling layer
Dave Kleikamp470decc2006-10-11 01:20:57 -07001460 * @journal: Journal to check.
1461 * @compat: bitmask of compatible features
1462 * @ro: bitmask of features that force read-only mount
1463 * @incompat: bitmask of incompatible features
1464 *
1465 * Check whether the journaling code supports the use of
1466 * all of a given set of features on this journal. Return true
1467 * (non-zero) if it can. */
1468
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001469int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001470 unsigned long ro, unsigned long incompat)
1471{
Dave Kleikamp470decc2006-10-11 01:20:57 -07001472 if (!compat && !ro && !incompat)
1473 return 1;
1474
Dave Kleikamp470decc2006-10-11 01:20:57 -07001475 /* We can support any known requested features iff the
1476 * superblock is in version 2. Otherwise we fail to support any
1477 * extended sb features. */
1478
1479 if (journal->j_format_version != 2)
1480 return 0;
1481
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001482 if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat &&
1483 (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro &&
1484 (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001485 return 1;
1486
1487 return 0;
1488}
1489
1490/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001491 * int jbd2_journal_set_features () - Mark a given journal feature in the superblock
Dave Kleikamp470decc2006-10-11 01:20:57 -07001492 * @journal: Journal to act on.
1493 * @compat: bitmask of compatible features
1494 * @ro: bitmask of features that force read-only mount
1495 * @incompat: bitmask of incompatible features
1496 *
1497 * Mark a given journal feature as present on the
1498 * superblock. Returns true if the requested features could be set.
1499 *
1500 */
1501
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001502int jbd2_journal_set_features (journal_t *journal, unsigned long compat,
Dave Kleikamp470decc2006-10-11 01:20:57 -07001503 unsigned long ro, unsigned long incompat)
1504{
1505 journal_superblock_t *sb;
1506
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001507 if (jbd2_journal_check_used_features(journal, compat, ro, incompat))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001508 return 1;
1509
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001510 if (!jbd2_journal_check_available_features(journal, compat, ro, incompat))
Dave Kleikamp470decc2006-10-11 01:20:57 -07001511 return 0;
1512
1513 jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n",
1514 compat, ro, incompat);
1515
1516 sb = journal->j_superblock;
1517
1518 sb->s_feature_compat |= cpu_to_be32(compat);
1519 sb->s_feature_ro_compat |= cpu_to_be32(ro);
1520 sb->s_feature_incompat |= cpu_to_be32(incompat);
1521
1522 return 1;
1523}
1524
Girish Shilamkar818d2762008-01-28 23:58:27 -05001525/*
1526 * jbd2_journal_clear_features () - Clear a given journal feature in the
1527 * superblock
1528 * @journal: Journal to act on.
1529 * @compat: bitmask of compatible features
1530 * @ro: bitmask of features that force read-only mount
1531 * @incompat: bitmask of incompatible features
1532 *
1533 * Clear a given journal feature as present on the
1534 * superblock.
1535 */
1536void jbd2_journal_clear_features(journal_t *journal, unsigned long compat,
1537 unsigned long ro, unsigned long incompat)
1538{
1539 journal_superblock_t *sb;
1540
1541 jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n",
1542 compat, ro, incompat);
1543
1544 sb = journal->j_superblock;
1545
1546 sb->s_feature_compat &= ~cpu_to_be32(compat);
1547 sb->s_feature_ro_compat &= ~cpu_to_be32(ro);
1548 sb->s_feature_incompat &= ~cpu_to_be32(incompat);
1549}
1550EXPORT_SYMBOL(jbd2_journal_clear_features);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001551
1552/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001553 * int jbd2_journal_update_format () - Update on-disk journal structure.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001554 * @journal: Journal to act on.
1555 *
1556 * Given an initialised but unloaded journal struct, poke about in the
1557 * on-disk structure to update it to the most recent supported version.
1558 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001559int jbd2_journal_update_format (journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001560{
1561 journal_superblock_t *sb;
1562 int err;
1563
1564 err = journal_get_superblock(journal);
1565 if (err)
1566 return err;
1567
1568 sb = journal->j_superblock;
1569
1570 switch (be32_to_cpu(sb->s_header.h_blocktype)) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001571 case JBD2_SUPERBLOCK_V2:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001572 return 0;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001573 case JBD2_SUPERBLOCK_V1:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001574 return journal_convert_superblock_v1(journal, sb);
1575 default:
1576 break;
1577 }
1578 return -EINVAL;
1579}
1580
1581static int journal_convert_superblock_v1(journal_t *journal,
1582 journal_superblock_t *sb)
1583{
1584 int offset, blocksize;
1585 struct buffer_head *bh;
1586
1587 printk(KERN_WARNING
Eryu Guanf2a44522011-11-01 19:09:18 -04001588 "JBD2: Converting superblock from version 1 to 2.\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001589
1590 /* Pre-initialise new fields to zero */
1591 offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb);
1592 blocksize = be32_to_cpu(sb->s_blocksize);
1593 memset(&sb->s_feature_compat, 0, blocksize-offset);
1594
1595 sb->s_nr_users = cpu_to_be32(1);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001596 sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001597 journal->j_format_version = 2;
1598
1599 bh = journal->j_sb_buffer;
1600 BUFFER_TRACE(bh, "marking dirty");
1601 mark_buffer_dirty(bh);
1602 sync_dirty_buffer(bh);
1603 return 0;
1604}
1605
1606
1607/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001608 * int jbd2_journal_flush () - Flush journal
Dave Kleikamp470decc2006-10-11 01:20:57 -07001609 * @journal: Journal to act on.
1610 *
1611 * Flush all data for a given journal to disk and empty the journal.
1612 * Filesystems can use this when remounting readonly to ensure that
1613 * recovery does not need to happen on remount.
1614 */
1615
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001616int jbd2_journal_flush(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001617{
1618 int err = 0;
1619 transaction_t *transaction = NULL;
1620 unsigned long old_tail;
1621
Theodore Ts'oa931da62010-08-03 21:35:12 -04001622 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001623
1624 /* Force everything buffered to the log... */
1625 if (journal->j_running_transaction) {
1626 transaction = journal->j_running_transaction;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001627 __jbd2_log_start_commit(journal, transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001628 } else if (journal->j_committing_transaction)
1629 transaction = journal->j_committing_transaction;
1630
1631 /* Wait for the log commit to complete... */
1632 if (transaction) {
1633 tid_t tid = transaction->t_tid;
1634
Theodore Ts'oa931da62010-08-03 21:35:12 -04001635 write_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001636 jbd2_log_wait_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001637 } else {
Theodore Ts'oa931da62010-08-03 21:35:12 -04001638 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001639 }
1640
1641 /* ...and flush everything in the log out to disk. */
1642 spin_lock(&journal->j_list_lock);
1643 while (!err && journal->j_checkpoint_transactions != NULL) {
1644 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001645 mutex_lock(&journal->j_checkpoint_mutex);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001646 err = jbd2_log_do_checkpoint(journal);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001647 mutex_unlock(&journal->j_checkpoint_mutex);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001648 spin_lock(&journal->j_list_lock);
1649 }
1650 spin_unlock(&journal->j_list_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001651
1652 if (is_journal_aborted(journal))
1653 return -EIO;
1654
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001655 jbd2_cleanup_journal_tail(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001656
1657 /* Finally, mark the journal as really needing no recovery.
1658 * This sets s_start==0 in the underlying superblock, which is
1659 * the magic code for a fully-recovered superblock. Any future
1660 * commits of data to the journal will restore the current
1661 * s_start value. */
Theodore Ts'oa931da62010-08-03 21:35:12 -04001662 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001663 old_tail = journal->j_tail;
1664 journal->j_tail = 0;
Theodore Ts'oa931da62010-08-03 21:35:12 -04001665 write_unlock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001666 jbd2_journal_update_superblock(journal, 1);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001667 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001668 journal->j_tail = old_tail;
1669
1670 J_ASSERT(!journal->j_running_transaction);
1671 J_ASSERT(!journal->j_committing_transaction);
1672 J_ASSERT(!journal->j_checkpoint_transactions);
1673 J_ASSERT(journal->j_head == journal->j_tail);
1674 J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001675 write_unlock(&journal->j_state_lock);
Hidehiro Kawai44519fa2008-10-10 20:29:13 -04001676 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001677}
1678
1679/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001680 * int jbd2_journal_wipe() - Wipe journal contents
Dave Kleikamp470decc2006-10-11 01:20:57 -07001681 * @journal: Journal to act on.
1682 * @write: flag (see below)
1683 *
1684 * Wipe out all of the contents of a journal, safely. This will produce
1685 * a warning if the journal contains any valid recovery information.
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001686 * Must be called between journal_init_*() and jbd2_journal_load().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001687 *
1688 * If 'write' is non-zero, then we wipe out the journal on disk; otherwise
1689 * we merely suppress recovery.
1690 */
1691
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001692int jbd2_journal_wipe(journal_t *journal, int write)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001693{
Dave Kleikamp470decc2006-10-11 01:20:57 -07001694 int err = 0;
1695
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001696 J_ASSERT (!(journal->j_flags & JBD2_LOADED));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001697
1698 err = load_superblock(journal);
1699 if (err)
1700 return err;
1701
Dave Kleikamp470decc2006-10-11 01:20:57 -07001702 if (!journal->j_tail)
1703 goto no_recovery;
1704
Eryu Guanf2a44522011-11-01 19:09:18 -04001705 printk(KERN_WARNING "JBD2: %s recovery information on journal\n",
Dave Kleikamp470decc2006-10-11 01:20:57 -07001706 write ? "Clearing" : "Ignoring");
1707
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001708 err = jbd2_journal_skip_recovery(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001709 if (write)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001710 jbd2_journal_update_superblock(journal, 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001711
1712 no_recovery:
1713 return err;
1714}
1715
1716/*
Dave Kleikamp470decc2006-10-11 01:20:57 -07001717 * Journal abort has very specific semantics, which we describe
1718 * for journal abort.
1719 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04001720 * Two internal functions, which provide abort to the jbd layer
Dave Kleikamp470decc2006-10-11 01:20:57 -07001721 * itself are here.
1722 */
1723
1724/*
1725 * Quick version for internal journal use (doesn't lock the journal).
1726 * Aborts hard --- we mark the abort as occurred, but do _nothing_ else,
1727 * and don't attempt to make any other journal updates.
1728 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001729void __jbd2_journal_abort_hard(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001730{
1731 transaction_t *transaction;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001732
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001733 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001734 return;
1735
1736 printk(KERN_ERR "Aborting journal on device %s.\n",
Theodore Ts'o05496762008-09-16 14:36:17 -04001737 journal->j_devname);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001738
Theodore Ts'oa931da62010-08-03 21:35:12 -04001739 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001740 journal->j_flags |= JBD2_ABORT;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001741 transaction = journal->j_running_transaction;
1742 if (transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001743 __jbd2_log_start_commit(journal, transaction->t_tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -04001744 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001745}
1746
1747/* Soft abort: record the abort error status in the journal superblock,
1748 * but don't do any other IO. */
1749static void __journal_abort_soft (journal_t *journal, int errno)
1750{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001751 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001752 return;
1753
1754 if (!journal->j_errno)
1755 journal->j_errno = errno;
1756
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001757 __jbd2_journal_abort_hard(journal);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001758
1759 if (errno)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001760 jbd2_journal_update_superblock(journal, 1);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001761}
1762
1763/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001764 * void jbd2_journal_abort () - Shutdown the journal immediately.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001765 * @journal: the journal to shutdown.
1766 * @errno: an error number to record in the journal indicating
1767 * the reason for the shutdown.
1768 *
1769 * Perform a complete, immediate shutdown of the ENTIRE
1770 * journal (not of a single transaction). This operation cannot be
1771 * undone without closing and reopening the journal.
1772 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001773 * The jbd2_journal_abort function is intended to support higher level error
Dave Kleikamp470decc2006-10-11 01:20:57 -07001774 * recovery mechanisms such as the ext2/ext3 remount-readonly error
1775 * mode.
1776 *
1777 * Journal abort has very specific semantics. Any existing dirty,
1778 * unjournaled buffers in the main filesystem will still be written to
1779 * disk by bdflush, but the journaling mechanism will be suspended
1780 * immediately and no further transaction commits will be honoured.
1781 *
1782 * Any dirty, journaled buffers will be written back to disk without
1783 * hitting the journal. Atomicity cannot be guaranteed on an aborted
1784 * filesystem, but we _do_ attempt to leave as much data as possible
1785 * behind for fsck to use for cleanup.
1786 *
1787 * Any attempt to get a new transaction handle on a journal which is in
1788 * ABORT state will just result in an -EROFS error return. A
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001789 * jbd2_journal_stop on an existing handle will return -EIO if we have
Dave Kleikamp470decc2006-10-11 01:20:57 -07001790 * entered abort state during the update.
1791 *
1792 * Recursive transactions are not disturbed by journal abort until the
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001793 * final jbd2_journal_stop, which will receive the -EIO error.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001794 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001795 * Finally, the jbd2_journal_abort call allows the caller to supply an errno
Dave Kleikamp470decc2006-10-11 01:20:57 -07001796 * which will be recorded (if possible) in the journal superblock. This
1797 * allows a client to record failure conditions in the middle of a
1798 * transaction without having to complete the transaction to record the
1799 * failure to disk. ext3_error, for example, now uses this
1800 * functionality.
1801 *
1802 * Errors which originate from within the journaling layer will NOT
1803 * supply an errno; a null errno implies that absolutely no further
1804 * writes are done to the journal (unless there are any already in
1805 * progress).
1806 *
1807 */
1808
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001809void jbd2_journal_abort(journal_t *journal, int errno)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001810{
1811 __journal_abort_soft(journal, errno);
1812}
1813
1814/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001815 * int jbd2_journal_errno () - returns the journal's error state.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001816 * @journal: journal to examine.
1817 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04001818 * This is the errno number set with jbd2_journal_abort(), the last
Dave Kleikamp470decc2006-10-11 01:20:57 -07001819 * time the journal was mounted - if the journal was stopped
1820 * without calling abort this will be 0.
1821 *
1822 * If the journal has been aborted on this mount time -EROFS will
1823 * be returned.
1824 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001825int jbd2_journal_errno(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001826{
1827 int err;
1828
Theodore Ts'oa931da62010-08-03 21:35:12 -04001829 read_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001830 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001831 err = -EROFS;
1832 else
1833 err = journal->j_errno;
Theodore Ts'oa931da62010-08-03 21:35:12 -04001834 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001835 return err;
1836}
1837
1838/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001839 * int jbd2_journal_clear_err () - clears the journal's error state
Dave Kleikamp470decc2006-10-11 01:20:57 -07001840 * @journal: journal to act on.
1841 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04001842 * An error must be cleared or acked to take a FS out of readonly
Dave Kleikamp470decc2006-10-11 01:20:57 -07001843 * mode.
1844 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001845int jbd2_journal_clear_err(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001846{
1847 int err = 0;
1848
Theodore Ts'oa931da62010-08-03 21:35:12 -04001849 write_lock(&journal->j_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001850 if (journal->j_flags & JBD2_ABORT)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001851 err = -EROFS;
1852 else
1853 journal->j_errno = 0;
Theodore Ts'oa931da62010-08-03 21:35:12 -04001854 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001855 return err;
1856}
1857
1858/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001859 * void jbd2_journal_ack_err() - Ack journal err.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001860 * @journal: journal to act on.
1861 *
Alberto Bertoglibfcd3552009-06-09 00:06:20 -04001862 * An error must be cleared or acked to take a FS out of readonly
Dave Kleikamp470decc2006-10-11 01:20:57 -07001863 * mode.
1864 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001865void jbd2_journal_ack_err(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001866{
Theodore Ts'oa931da62010-08-03 21:35:12 -04001867 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001868 if (journal->j_errno)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001869 journal->j_flags |= JBD2_ACK_ERR;
Theodore Ts'oa931da62010-08-03 21:35:12 -04001870 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001871}
1872
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001873int jbd2_journal_blocks_per_page(struct inode *inode)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001874{
1875 return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1876}
1877
1878/*
Zach Brownb517bea2006-10-11 01:21:08 -07001879 * helper functions to deal with 32 or 64bit block numbers.
1880 */
1881size_t journal_tag_bytes(journal_t *journal)
1882{
1883 if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT))
Mingming Caocd02ff02007-10-16 18:38:25 -04001884 return JBD2_TAG_SIZE64;
Zach Brownb517bea2006-10-11 01:21:08 -07001885 else
Mingming Caocd02ff02007-10-16 18:38:25 -04001886 return JBD2_TAG_SIZE32;
Zach Brownb517bea2006-10-11 01:21:08 -07001887}
1888
1889/*
Theodore Ts'od2eecb02009-12-07 10:36:20 -05001890 * JBD memory management
1891 *
1892 * These functions are used to allocate block-sized chunks of memory
1893 * used for making copies of buffer_head data. Very often it will be
1894 * page-sized chunks of data, but sometimes it will be in
1895 * sub-page-size chunks. (For example, 16k pages on Power systems
1896 * with a 4k block file system.) For blocks smaller than a page, we
1897 * use a SLAB allocator. There are slab caches for each block size,
1898 * which are allocated at mount time, if necessary, and we only free
1899 * (all of) the slab caches when/if the jbd2 module is unloaded. For
1900 * this reason we don't need to a mutex to protect access to
1901 * jbd2_slab[] allocating or releasing memory; only in
1902 * jbd2_journal_create_slab().
1903 */
1904#define JBD2_MAX_SLABS 8
1905static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS];
Theodore Ts'od2eecb02009-12-07 10:36:20 -05001906
1907static const char *jbd2_slab_names[JBD2_MAX_SLABS] = {
1908 "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k",
1909 "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k"
1910};
1911
1912
1913static void jbd2_journal_destroy_slabs(void)
1914{
1915 int i;
1916
1917 for (i = 0; i < JBD2_MAX_SLABS; i++) {
1918 if (jbd2_slab[i])
1919 kmem_cache_destroy(jbd2_slab[i]);
1920 jbd2_slab[i] = NULL;
1921 }
1922}
1923
1924static int jbd2_journal_create_slab(size_t size)
1925{
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02001926 static DEFINE_MUTEX(jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05001927 int i = order_base_2(size) - 10;
1928 size_t slab_size;
1929
1930 if (size == PAGE_SIZE)
1931 return 0;
1932
1933 if (i >= JBD2_MAX_SLABS)
1934 return -EINVAL;
1935
1936 if (unlikely(i < 0))
1937 i = 0;
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02001938 mutex_lock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05001939 if (jbd2_slab[i]) {
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02001940 mutex_unlock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05001941 return 0; /* Already created */
1942 }
1943
1944 slab_size = 1 << (i+10);
1945 jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size,
1946 slab_size, 0, NULL);
Thomas Gleixner51dfacde2010-10-16 22:34:39 +02001947 mutex_unlock(&jbd2_slab_create_mutex);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05001948 if (!jbd2_slab[i]) {
1949 printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n");
1950 return -ENOMEM;
1951 }
1952 return 0;
1953}
1954
1955static struct kmem_cache *get_slab(size_t size)
1956{
1957 int i = order_base_2(size) - 10;
1958
1959 BUG_ON(i >= JBD2_MAX_SLABS);
1960 if (unlikely(i < 0))
1961 i = 0;
Bill Pemberton8ac97b72010-04-30 09:34:31 -04001962 BUG_ON(jbd2_slab[i] == NULL);
Theodore Ts'od2eecb02009-12-07 10:36:20 -05001963 return jbd2_slab[i];
1964}
1965
1966void *jbd2_alloc(size_t size, gfp_t flags)
1967{
1968 void *ptr;
1969
1970 BUG_ON(size & (size-1)); /* Must be a power of 2 */
1971
1972 flags |= __GFP_REPEAT;
1973 if (size == PAGE_SIZE)
1974 ptr = (void *)__get_free_pages(flags, 0);
1975 else if (size > PAGE_SIZE) {
1976 int order = get_order(size);
1977
1978 if (order < 3)
1979 ptr = (void *)__get_free_pages(flags, order);
1980 else
1981 ptr = vmalloc(size);
1982 } else
1983 ptr = kmem_cache_alloc(get_slab(size), flags);
1984
1985 /* Check alignment; SLUB has gotten this wrong in the past,
1986 * and this can lead to user data corruption! */
1987 BUG_ON(((unsigned long) ptr) & (size-1));
1988
1989 return ptr;
1990}
1991
1992void jbd2_free(void *ptr, size_t size)
1993{
1994 if (size == PAGE_SIZE) {
1995 free_pages((unsigned long)ptr, 0);
1996 return;
1997 }
1998 if (size > PAGE_SIZE) {
1999 int order = get_order(size);
2000
2001 if (order < 3)
2002 free_pages((unsigned long)ptr, order);
2003 else
2004 vfree(ptr);
2005 return;
2006 }
2007 kmem_cache_free(get_slab(size), ptr);
2008};
2009
2010/*
Dave Kleikamp470decc2006-10-11 01:20:57 -07002011 * Journal_head storage management
2012 */
Christoph Lametere18b8902006-12-06 20:33:20 -08002013static struct kmem_cache *jbd2_journal_head_cache;
Jose R. Santose23291b2007-07-18 08:57:06 -04002014#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002015static atomic_t nr_journal_heads = ATOMIC_INIT(0);
2016#endif
2017
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002018static int journal_init_jbd2_journal_head_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002019{
2020 int retval;
2021
Al Viro1076d172008-03-29 03:07:18 +00002022 J_ASSERT(jbd2_journal_head_cache == NULL);
Johann Lombardia920e942006-10-11 01:21:00 -07002023 jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head",
Dave Kleikamp470decc2006-10-11 01:20:57 -07002024 sizeof(struct journal_head),
2025 0, /* offset */
Mingming Cao77160952008-01-28 23:58:27 -05002026 SLAB_TEMPORARY, /* flags */
Paul Mundt20c2df82007-07-20 10:11:58 +09002027 NULL); /* ctor */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002028 retval = 0;
Al Viro1076d172008-03-29 03:07:18 +00002029 if (!jbd2_journal_head_cache) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002030 retval = -ENOMEM;
Eryu Guanf2a44522011-11-01 19:09:18 -04002031 printk(KERN_EMERG "JBD2: no memory for journal_head cache\n");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002032 }
2033 return retval;
2034}
2035
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002036static void jbd2_journal_destroy_jbd2_journal_head_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002037{
Duane Griffin8a9362e2008-04-17 10:38:59 -04002038 if (jbd2_journal_head_cache) {
2039 kmem_cache_destroy(jbd2_journal_head_cache);
2040 jbd2_journal_head_cache = NULL;
2041 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002042}
2043
2044/*
2045 * journal_head splicing and dicing
2046 */
2047static struct journal_head *journal_alloc_journal_head(void)
2048{
2049 struct journal_head *ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002050
Jose R. Santose23291b2007-07-18 08:57:06 -04002051#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002052 atomic_inc(&nr_journal_heads);
2053#endif
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002054 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
Al Viro1076d172008-03-29 03:07:18 +00002055 if (!ret) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002056 jbd_debug(1, "out of memory for journal_head\n");
Theodore Ts'o670be5a2010-12-17 10:44:16 -05002057 pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__);
Al Viro1076d172008-03-29 03:07:18 +00002058 while (!ret) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002059 yield();
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002060 ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002061 }
2062 }
2063 return ret;
2064}
2065
2066static void journal_free_journal_head(struct journal_head *jh)
2067{
Jose R. Santose23291b2007-07-18 08:57:06 -04002068#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002069 atomic_dec(&nr_journal_heads);
Mingming Caocd02ff02007-10-16 18:38:25 -04002070 memset(jh, JBD2_POISON_FREE, sizeof(*jh));
Dave Kleikamp470decc2006-10-11 01:20:57 -07002071#endif
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002072 kmem_cache_free(jbd2_journal_head_cache, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002073}
2074
2075/*
2076 * A journal_head is attached to a buffer_head whenever JBD has an
2077 * interest in the buffer.
2078 *
2079 * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit
2080 * is set. This bit is tested in core kernel code where we need to take
2081 * JBD-specific actions. Testing the zeroness of ->b_private is not reliable
2082 * there.
2083 *
2084 * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one.
2085 *
2086 * When a buffer has its BH_JBD bit set it is immune from being released by
2087 * core kernel code, mainly via ->b_count.
2088 *
Jan Karade1b7942011-06-13 15:38:22 -04002089 * A journal_head is detached from its buffer_head when the journal_head's
2090 * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint
2091 * transaction (b_cp_transaction) hold their references to b_jcount.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002092 *
2093 * Various places in the kernel want to attach a journal_head to a buffer_head
2094 * _before_ attaching the journal_head to a transaction. To protect the
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002095 * journal_head in this situation, jbd2_journal_add_journal_head elevates the
Dave Kleikamp470decc2006-10-11 01:20:57 -07002096 * journal_head's b_jcount refcount by one. The caller must call
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002097 * jbd2_journal_put_journal_head() to undo this.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002098 *
2099 * So the typical usage would be:
2100 *
2101 * (Attach a journal_head if needed. Increments b_jcount)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002102 * struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002103 * ...
Jan Karade1b7942011-06-13 15:38:22 -04002104 * (Get another reference for transaction)
2105 * jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002106 * jh->b_transaction = xxx;
Jan Karade1b7942011-06-13 15:38:22 -04002107 * (Put original reference)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002108 * jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002109 */
2110
2111/*
2112 * Give a buffer_head a journal_head.
2113 *
Dave Kleikamp470decc2006-10-11 01:20:57 -07002114 * May sleep.
2115 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002116struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002117{
2118 struct journal_head *jh;
2119 struct journal_head *new_jh = NULL;
2120
2121repeat:
2122 if (!buffer_jbd(bh)) {
2123 new_jh = journal_alloc_journal_head();
2124 memset(new_jh, 0, sizeof(*new_jh));
2125 }
2126
2127 jbd_lock_bh_journal_head(bh);
2128 if (buffer_jbd(bh)) {
2129 jh = bh2jh(bh);
2130 } else {
2131 J_ASSERT_BH(bh,
2132 (atomic_read(&bh->b_count) > 0) ||
2133 (bh->b_page && bh->b_page->mapping));
2134
2135 if (!new_jh) {
2136 jbd_unlock_bh_journal_head(bh);
2137 goto repeat;
2138 }
2139
2140 jh = new_jh;
2141 new_jh = NULL; /* We consumed it */
2142 set_buffer_jbd(bh);
2143 bh->b_private = jh;
2144 jh->b_bh = bh;
2145 get_bh(bh);
2146 BUFFER_TRACE(bh, "added journal_head");
2147 }
2148 jh->b_jcount++;
2149 jbd_unlock_bh_journal_head(bh);
2150 if (new_jh)
2151 journal_free_journal_head(new_jh);
2152 return bh->b_private;
2153}
2154
2155/*
2156 * Grab a ref against this buffer_head's journal_head. If it ended up not
2157 * having a journal_head, return NULL
2158 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002159struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002160{
2161 struct journal_head *jh = NULL;
2162
2163 jbd_lock_bh_journal_head(bh);
2164 if (buffer_jbd(bh)) {
2165 jh = bh2jh(bh);
2166 jh->b_jcount++;
2167 }
2168 jbd_unlock_bh_journal_head(bh);
2169 return jh;
2170}
2171
2172static void __journal_remove_journal_head(struct buffer_head *bh)
2173{
2174 struct journal_head *jh = bh2jh(bh);
2175
2176 J_ASSERT_JH(jh, jh->b_jcount >= 0);
Jan Karade1b7942011-06-13 15:38:22 -04002177 J_ASSERT_JH(jh, jh->b_transaction == NULL);
2178 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2179 J_ASSERT_JH(jh, jh->b_cp_transaction == NULL);
2180 J_ASSERT_JH(jh, jh->b_jlist == BJ_None);
2181 J_ASSERT_BH(bh, buffer_jbd(bh));
2182 J_ASSERT_BH(bh, jh2bh(jh) == bh);
2183 BUFFER_TRACE(bh, "remove journal_head");
2184 if (jh->b_frozen_data) {
2185 printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__);
2186 jbd2_free(jh->b_frozen_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002187 }
Jan Karade1b7942011-06-13 15:38:22 -04002188 if (jh->b_committed_data) {
2189 printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__);
2190 jbd2_free(jh->b_committed_data, bh->b_size);
2191 }
2192 bh->b_private = NULL;
2193 jh->b_bh = NULL; /* debug, really */
2194 clear_buffer_jbd(bh);
2195 journal_free_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002196}
2197
2198/*
Jan Karade1b7942011-06-13 15:38:22 -04002199 * Drop a reference on the passed journal_head. If it fell to zero then
Dave Kleikamp470decc2006-10-11 01:20:57 -07002200 * release the journal_head from the buffer_head.
2201 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002202void jbd2_journal_put_journal_head(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002203{
2204 struct buffer_head *bh = jh2bh(jh);
2205
2206 jbd_lock_bh_journal_head(bh);
2207 J_ASSERT_JH(jh, jh->b_jcount > 0);
2208 --jh->b_jcount;
Jan Karade1b7942011-06-13 15:38:22 -04002209 if (!jh->b_jcount) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002210 __journal_remove_journal_head(bh);
Jan Karade1b7942011-06-13 15:38:22 -04002211 jbd_unlock_bh_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002212 __brelse(bh);
Jan Karade1b7942011-06-13 15:38:22 -04002213 } else
2214 jbd_unlock_bh_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002215}
2216
2217/*
Jan Karac851ed52008-07-11 19:27:31 -04002218 * Initialize jbd inode head
2219 */
2220void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode)
2221{
2222 jinode->i_transaction = NULL;
2223 jinode->i_next_transaction = NULL;
2224 jinode->i_vfs_inode = inode;
2225 jinode->i_flags = 0;
2226 INIT_LIST_HEAD(&jinode->i_list);
2227}
2228
2229/*
2230 * Function to be called before we start removing inode from memory (i.e.,
2231 * clear_inode() is a fine place to be called from). It removes inode from
2232 * transaction's lists.
2233 */
2234void jbd2_journal_release_jbd_inode(journal_t *journal,
2235 struct jbd2_inode *jinode)
2236{
Jan Karac851ed52008-07-11 19:27:31 -04002237 if (!journal)
2238 return;
2239restart:
2240 spin_lock(&journal->j_list_lock);
2241 /* Is commit writing out inode - we have to wait */
Brian King39e3ac22010-10-27 21:25:12 -04002242 if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) {
Jan Karac851ed52008-07-11 19:27:31 -04002243 wait_queue_head_t *wq;
2244 DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING);
2245 wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING);
2246 prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE);
2247 spin_unlock(&journal->j_list_lock);
2248 schedule();
2249 finish_wait(wq, &wait.wait);
2250 goto restart;
2251 }
2252
Jan Karac851ed52008-07-11 19:27:31 -04002253 if (jinode->i_transaction) {
2254 list_del(&jinode->i_list);
2255 jinode->i_transaction = NULL;
2256 }
2257 spin_unlock(&journal->j_list_lock);
2258}
2259
2260/*
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002261 * debugfs tunables
Dave Kleikamp470decc2006-10-11 01:20:57 -07002262 */
Jose R. Santos6f38c742007-10-16 18:38:25 -04002263#ifdef CONFIG_JBD2_DEBUG
2264u8 jbd2_journal_enable_debug __read_mostly;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002265EXPORT_SYMBOL(jbd2_journal_enable_debug);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002266
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002267#define JBD2_DEBUG_NAME "jbd2-debug"
Dave Kleikamp470decc2006-10-11 01:20:57 -07002268
Jose R. Santos6f38c742007-10-16 18:38:25 -04002269static struct dentry *jbd2_debugfs_dir;
2270static struct dentry *jbd2_debug;
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002271
2272static void __init jbd2_create_debugfs_entry(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002273{
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002274 jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL);
2275 if (jbd2_debugfs_dir)
Yin Kangkai765f8362009-12-15 14:48:25 -08002276 jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME,
2277 S_IRUGO | S_IWUSR,
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002278 jbd2_debugfs_dir,
2279 &jbd2_journal_enable_debug);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002280}
2281
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002282static void __exit jbd2_remove_debugfs_entry(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002283{
Jose R. Santos6f38c742007-10-16 18:38:25 -04002284 debugfs_remove(jbd2_debug);
2285 debugfs_remove(jbd2_debugfs_dir);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002286}
2287
2288#else
2289
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002290static void __init jbd2_create_debugfs_entry(void)
2291{
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002292}
2293
2294static void __exit jbd2_remove_debugfs_entry(void)
2295{
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002296}
Dave Kleikamp470decc2006-10-11 01:20:57 -07002297
2298#endif
2299
Johann Lombardi8e85fb32008-01-28 23:58:27 -05002300#ifdef CONFIG_PROC_FS
2301
2302#define JBD2_STATS_PROC_NAME "fs/jbd2"
2303
2304static void __init jbd2_create_jbd_stats_proc_entry(void)
2305{
2306 proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL);
2307}
2308
2309static void __exit jbd2_remove_jbd_stats_proc_entry(void)
2310{
2311 if (proc_jbd2_stats)
2312 remove_proc_entry(JBD2_STATS_PROC_NAME, NULL);
2313}
2314
2315#else
2316
2317#define jbd2_create_jbd_stats_proc_entry() do {} while (0)
2318#define jbd2_remove_jbd_stats_proc_entry() do {} while (0)
2319
2320#endif
2321
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002322struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002323
2324static int __init journal_init_handle_cache(void)
2325{
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002326 jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002327 if (jbd2_handle_cache == NULL) {
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002328 printk(KERN_EMERG "JBD2: failed to create handle cache\n");
2329 return -ENOMEM;
2330 }
2331 jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0);
2332 if (jbd2_inode_cache == NULL) {
2333 printk(KERN_EMERG "JBD2: failed to create inode cache\n");
2334 kmem_cache_destroy(jbd2_handle_cache);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002335 return -ENOMEM;
2336 }
2337 return 0;
2338}
2339
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002340static void jbd2_journal_destroy_handle_cache(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002341{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002342 if (jbd2_handle_cache)
2343 kmem_cache_destroy(jbd2_handle_cache);
Theodore Ts'o8aefcd52011-01-10 12:29:43 -05002344 if (jbd2_inode_cache)
2345 kmem_cache_destroy(jbd2_inode_cache);
2346
Dave Kleikamp470decc2006-10-11 01:20:57 -07002347}
2348
2349/*
2350 * Module startup and shutdown
2351 */
2352
2353static int __init journal_init_caches(void)
2354{
2355 int ret;
2356
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002357 ret = jbd2_journal_init_revoke_caches();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002358 if (ret == 0)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002359 ret = journal_init_jbd2_journal_head_cache();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002360 if (ret == 0)
2361 ret = journal_init_handle_cache();
2362 return ret;
2363}
2364
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002365static void jbd2_journal_destroy_caches(void)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002366{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002367 jbd2_journal_destroy_revoke_caches();
2368 jbd2_journal_destroy_jbd2_journal_head_cache();
2369 jbd2_journal_destroy_handle_cache();
Theodore Ts'od2eecb02009-12-07 10:36:20 -05002370 jbd2_journal_destroy_slabs();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002371}
2372
2373static int __init journal_init(void)
2374{
2375 int ret;
2376
2377 BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024);
2378
2379 ret = journal_init_caches();
Duane Griffin620de4e2008-04-29 22:02:47 -04002380 if (ret == 0) {
2381 jbd2_create_debugfs_entry();
2382 jbd2_create_jbd_stats_proc_entry();
2383 } else {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002384 jbd2_journal_destroy_caches();
Duane Griffin620de4e2008-04-29 22:02:47 -04002385 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07002386 return ret;
2387}
2388
2389static void __exit journal_exit(void)
2390{
Jose R. Santose23291b2007-07-18 08:57:06 -04002391#ifdef CONFIG_JBD2_DEBUG
Dave Kleikamp470decc2006-10-11 01:20:57 -07002392 int n = atomic_read(&nr_journal_heads);
2393 if (n)
Eryu Guanf2a44522011-11-01 19:09:18 -04002394 printk(KERN_EMERG "JBD2: leaked %d journal_heads!\n", n);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002395#endif
Jose R. Santos0f49d5d2007-07-18 08:50:18 -04002396 jbd2_remove_debugfs_entry();
Johann Lombardi8e85fb32008-01-28 23:58:27 -05002397 jbd2_remove_jbd_stats_proc_entry();
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002398 jbd2_journal_destroy_caches();
Dave Kleikamp470decc2006-10-11 01:20:57 -07002399}
2400
Dave Kleikamp470decc2006-10-11 01:20:57 -07002401MODULE_LICENSE("GPL");
2402module_init(journal_init);
2403module_exit(journal_exit);
2404