Commit | Line | Data |
---|---|---|
470decc6 | 1 | /* |
f7f4bccb | 2 | * linux/fs/jbd2/journal.c |
470decc6 DK |
3 | * |
4 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1998 | |
5 | * | |
6 | * Copyright 1998 Red Hat corp --- All Rights Reserved | |
7 | * | |
8 | * This file is part of the Linux kernel and is made available under | |
9 | * the terms of the GNU General Public License, version 2, or at your | |
10 | * option, any later version, incorporated herein by reference. | |
11 | * | |
12 | * Generic filesystem journal-writing code; part of the ext2fs | |
13 | * journaling system. | |
14 | * | |
15 | * This file manages journals: areas of disk reserved for logging | |
16 | * transactional updates. This includes the kernel journaling thread | |
17 | * which is responsible for scheduling updates to the log. | |
18 | * | |
19 | * We do not actually manage the physical storage of the journal in this | |
20 | * file: that is left to a per-journal policy function, which allows us | |
21 | * to store the journal within a filesystem-specified area for ext2 | |
22 | * journaling (ext2 can use a reserved inode for storing the log). | |
23 | */ | |
24 | ||
25 | #include <linux/module.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/fs.h> | |
f7f4bccb | 28 | #include <linux/jbd2.h> |
470decc6 DK |
29 | #include <linux/errno.h> |
30 | #include <linux/slab.h> | |
470decc6 DK |
31 | #include <linux/init.h> |
32 | #include <linux/mm.h> | |
7dfb7103 | 33 | #include <linux/freezer.h> |
470decc6 DK |
34 | #include <linux/pagemap.h> |
35 | #include <linux/kthread.h> | |
36 | #include <linux/poison.h> | |
37 | #include <linux/proc_fs.h> | |
8e85fb3f | 38 | #include <linux/seq_file.h> |
c225aa57 | 39 | #include <linux/math64.h> |
879c5e6b | 40 | #include <linux/hash.h> |
d2eecb03 TT |
41 | #include <linux/log2.h> |
42 | #include <linux/vmalloc.h> | |
47def826 | 43 | #include <linux/backing-dev.h> |
39e3ac25 | 44 | #include <linux/bitops.h> |
670be5a7 | 45 | #include <linux/ratelimit.h> |
879c5e6b TT |
46 | |
47 | #define CREATE_TRACE_POINTS | |
48 | #include <trace/events/jbd2.h> | |
470decc6 DK |
49 | |
50 | #include <asm/uaccess.h> | |
51 | #include <asm/page.h> | |
52 | ||
b6e96d00 TT |
53 | #ifdef CONFIG_JBD2_DEBUG |
54 | ushort jbd2_journal_enable_debug __read_mostly; | |
55 | EXPORT_SYMBOL(jbd2_journal_enable_debug); | |
56 | ||
57 | module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644); | |
58 | MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2"); | |
59 | #endif | |
60 | ||
f7f4bccb MC |
61 | EXPORT_SYMBOL(jbd2_journal_extend); |
62 | EXPORT_SYMBOL(jbd2_journal_stop); | |
63 | EXPORT_SYMBOL(jbd2_journal_lock_updates); | |
64 | EXPORT_SYMBOL(jbd2_journal_unlock_updates); | |
65 | EXPORT_SYMBOL(jbd2_journal_get_write_access); | |
66 | EXPORT_SYMBOL(jbd2_journal_get_create_access); | |
67 | EXPORT_SYMBOL(jbd2_journal_get_undo_access); | |
e06c8227 | 68 | EXPORT_SYMBOL(jbd2_journal_set_triggers); |
f7f4bccb | 69 | EXPORT_SYMBOL(jbd2_journal_dirty_metadata); |
f7f4bccb | 70 | EXPORT_SYMBOL(jbd2_journal_forget); |
470decc6 DK |
71 | #if 0 |
72 | EXPORT_SYMBOL(journal_sync_buffer); | |
73 | #endif | |
f7f4bccb MC |
74 | EXPORT_SYMBOL(jbd2_journal_flush); |
75 | EXPORT_SYMBOL(jbd2_journal_revoke); | |
76 | ||
77 | EXPORT_SYMBOL(jbd2_journal_init_dev); | |
78 | EXPORT_SYMBOL(jbd2_journal_init_inode); | |
f7f4bccb MC |
79 | EXPORT_SYMBOL(jbd2_journal_check_used_features); |
80 | EXPORT_SYMBOL(jbd2_journal_check_available_features); | |
81 | EXPORT_SYMBOL(jbd2_journal_set_features); | |
f7f4bccb MC |
82 | EXPORT_SYMBOL(jbd2_journal_load); |
83 | EXPORT_SYMBOL(jbd2_journal_destroy); | |
f7f4bccb MC |
84 | EXPORT_SYMBOL(jbd2_journal_abort); |
85 | EXPORT_SYMBOL(jbd2_journal_errno); | |
86 | EXPORT_SYMBOL(jbd2_journal_ack_err); | |
87 | EXPORT_SYMBOL(jbd2_journal_clear_err); | |
88 | EXPORT_SYMBOL(jbd2_log_wait_commit); | |
3b799d15 | 89 | EXPORT_SYMBOL(jbd2_log_start_commit); |
f7f4bccb MC |
90 | EXPORT_SYMBOL(jbd2_journal_start_commit); |
91 | EXPORT_SYMBOL(jbd2_journal_force_commit_nested); | |
92 | EXPORT_SYMBOL(jbd2_journal_wipe); | |
93 | EXPORT_SYMBOL(jbd2_journal_blocks_per_page); | |
94 | EXPORT_SYMBOL(jbd2_journal_invalidatepage); | |
95 | EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); | |
96 | EXPORT_SYMBOL(jbd2_journal_force_commit); | |
c851ed54 JK |
97 | EXPORT_SYMBOL(jbd2_journal_file_inode); |
98 | EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); | |
99 | EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); | |
100 | EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); | |
8aefcd55 | 101 | EXPORT_SYMBOL(jbd2_inode_cache); |
470decc6 | 102 | |
470decc6 | 103 | static void __journal_abort_soft (journal_t *journal, int errno); |
d2eecb03 | 104 | static int jbd2_journal_create_slab(size_t slab_size); |
470decc6 | 105 | |
25ed6e8a DW |
106 | /* Checksumming functions */ |
107 | int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb) | |
108 | { | |
109 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
110 | return 1; | |
111 | ||
112 | return sb->s_checksum_type == JBD2_CRC32C_CHKSUM; | |
113 | } | |
114 | ||
4fd5ea43 DW |
115 | static __u32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb) |
116 | { | |
117 | __u32 csum, old_csum; | |
118 | ||
119 | old_csum = sb->s_checksum; | |
120 | sb->s_checksum = 0; | |
121 | csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t)); | |
122 | sb->s_checksum = old_csum; | |
123 | ||
124 | return cpu_to_be32(csum); | |
125 | } | |
126 | ||
127 | int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb) | |
128 | { | |
129 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
130 | return 1; | |
131 | ||
132 | return sb->s_checksum == jbd2_superblock_csum(j, sb); | |
133 | } | |
134 | ||
135 | void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb) | |
136 | { | |
137 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
138 | return; | |
139 | ||
140 | sb->s_checksum = jbd2_superblock_csum(j, sb); | |
141 | } | |
142 | ||
470decc6 DK |
143 | /* |
144 | * Helper function used to manage commit timeouts | |
145 | */ | |
146 | ||
147 | static void commit_timeout(unsigned long __data) | |
148 | { | |
149 | struct task_struct * p = (struct task_struct *) __data; | |
150 | ||
151 | wake_up_process(p); | |
152 | } | |
153 | ||
154 | /* | |
f7f4bccb | 155 | * kjournald2: The main thread function used to manage a logging device |
470decc6 DK |
156 | * journal. |
157 | * | |
158 | * This kernel thread is responsible for two things: | |
159 | * | |
160 | * 1) COMMIT: Every so often we need to commit the current state of the | |
161 | * filesystem to disk. The journal thread is responsible for writing | |
162 | * all of the metadata buffers to disk. | |
163 | * | |
164 | * 2) CHECKPOINT: We cannot reuse a used section of the log file until all | |
165 | * of the data in that part of the log has been rewritten elsewhere on | |
166 | * the disk. Flushing these old buffers to reclaim space in the log is | |
167 | * known as checkpointing, and this thread is responsible for that job. | |
168 | */ | |
169 | ||
f7f4bccb | 170 | static int kjournald2(void *arg) |
470decc6 DK |
171 | { |
172 | journal_t *journal = arg; | |
173 | transaction_t *transaction; | |
174 | ||
175 | /* | |
176 | * Set up an interval timer which can be used to trigger a commit wakeup | |
177 | * after the commit interval expires | |
178 | */ | |
179 | setup_timer(&journal->j_commit_timer, commit_timeout, | |
180 | (unsigned long)current); | |
181 | ||
35c80422 NC |
182 | set_freezable(); |
183 | ||
470decc6 DK |
184 | /* Record that the journal thread is running */ |
185 | journal->j_task = current; | |
186 | wake_up(&journal->j_wait_done_commit); | |
187 | ||
470decc6 DK |
188 | /* |
189 | * And now, wait forever for commit wakeup events. | |
190 | */ | |
a931da6a | 191 | write_lock(&journal->j_state_lock); |
470decc6 DK |
192 | |
193 | loop: | |
f7f4bccb | 194 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
195 | goto end_loop; |
196 | ||
197 | jbd_debug(1, "commit_sequence=%d, commit_request=%d\n", | |
198 | journal->j_commit_sequence, journal->j_commit_request); | |
199 | ||
200 | if (journal->j_commit_sequence != journal->j_commit_request) { | |
201 | jbd_debug(1, "OK, requests differ\n"); | |
a931da6a | 202 | write_unlock(&journal->j_state_lock); |
470decc6 | 203 | del_timer_sync(&journal->j_commit_timer); |
f7f4bccb | 204 | jbd2_journal_commit_transaction(journal); |
a931da6a | 205 | write_lock(&journal->j_state_lock); |
470decc6 DK |
206 | goto loop; |
207 | } | |
208 | ||
209 | wake_up(&journal->j_wait_done_commit); | |
210 | if (freezing(current)) { | |
211 | /* | |
212 | * The simpler the better. Flushing journal isn't a | |
213 | * good idea, because that depends on threads that may | |
214 | * be already stopped. | |
215 | */ | |
f7f4bccb | 216 | jbd_debug(1, "Now suspending kjournald2\n"); |
a931da6a | 217 | write_unlock(&journal->j_state_lock); |
a0acae0e | 218 | try_to_freeze(); |
a931da6a | 219 | write_lock(&journal->j_state_lock); |
470decc6 DK |
220 | } else { |
221 | /* | |
222 | * We assume on resume that commits are already there, | |
223 | * so we don't sleep | |
224 | */ | |
225 | DEFINE_WAIT(wait); | |
226 | int should_sleep = 1; | |
227 | ||
228 | prepare_to_wait(&journal->j_wait_commit, &wait, | |
229 | TASK_INTERRUPTIBLE); | |
230 | if (journal->j_commit_sequence != journal->j_commit_request) | |
231 | should_sleep = 0; | |
232 | transaction = journal->j_running_transaction; | |
233 | if (transaction && time_after_eq(jiffies, | |
234 | transaction->t_expires)) | |
235 | should_sleep = 0; | |
f7f4bccb | 236 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
237 | should_sleep = 0; |
238 | if (should_sleep) { | |
a931da6a | 239 | write_unlock(&journal->j_state_lock); |
470decc6 | 240 | schedule(); |
a931da6a | 241 | write_lock(&journal->j_state_lock); |
470decc6 DK |
242 | } |
243 | finish_wait(&journal->j_wait_commit, &wait); | |
244 | } | |
245 | ||
f7f4bccb | 246 | jbd_debug(1, "kjournald2 wakes\n"); |
470decc6 DK |
247 | |
248 | /* | |
249 | * Were we woken up by a commit wakeup event? | |
250 | */ | |
251 | transaction = journal->j_running_transaction; | |
252 | if (transaction && time_after_eq(jiffies, transaction->t_expires)) { | |
253 | journal->j_commit_request = transaction->t_tid; | |
254 | jbd_debug(1, "woke because of timeout\n"); | |
255 | } | |
256 | goto loop; | |
257 | ||
258 | end_loop: | |
a931da6a | 259 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
260 | del_timer_sync(&journal->j_commit_timer); |
261 | journal->j_task = NULL; | |
262 | wake_up(&journal->j_wait_done_commit); | |
263 | jbd_debug(1, "Journal thread exiting.\n"); | |
264 | return 0; | |
265 | } | |
266 | ||
97f06784 | 267 | static int jbd2_journal_start_thread(journal_t *journal) |
470decc6 | 268 | { |
97f06784 PE |
269 | struct task_struct *t; |
270 | ||
90576c0b TT |
271 | t = kthread_run(kjournald2, journal, "jbd2/%s", |
272 | journal->j_devname); | |
97f06784 PE |
273 | if (IS_ERR(t)) |
274 | return PTR_ERR(t); | |
275 | ||
1076d17a | 276 | wait_event(journal->j_wait_done_commit, journal->j_task != NULL); |
97f06784 | 277 | return 0; |
470decc6 DK |
278 | } |
279 | ||
280 | static void journal_kill_thread(journal_t *journal) | |
281 | { | |
a931da6a | 282 | write_lock(&journal->j_state_lock); |
f7f4bccb | 283 | journal->j_flags |= JBD2_UNMOUNT; |
470decc6 DK |
284 | |
285 | while (journal->j_task) { | |
286 | wake_up(&journal->j_wait_commit); | |
a931da6a | 287 | write_unlock(&journal->j_state_lock); |
1076d17a | 288 | wait_event(journal->j_wait_done_commit, journal->j_task == NULL); |
a931da6a | 289 | write_lock(&journal->j_state_lock); |
470decc6 | 290 | } |
a931da6a | 291 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
292 | } |
293 | ||
294 | /* | |
f7f4bccb | 295 | * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. |
470decc6 DK |
296 | * |
297 | * Writes a metadata buffer to a given disk block. The actual IO is not | |
298 | * performed but a new buffer_head is constructed which labels the data | |
299 | * to be written with the correct destination disk block. | |
300 | * | |
301 | * Any magic-number escaping which needs to be done will cause a | |
302 | * copy-out here. If the buffer happens to start with the | |
f7f4bccb | 303 | * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the |
470decc6 DK |
304 | * magic number is only written to the log for descripter blocks. In |
305 | * this case, we copy the data and replace the first word with 0, and we | |
306 | * return a result code which indicates that this buffer needs to be | |
307 | * marked as an escaped buffer in the corresponding log descriptor | |
308 | * block. The missing word can then be restored when the block is read | |
309 | * during recovery. | |
310 | * | |
311 | * If the source buffer has already been modified by a new transaction | |
312 | * since we took the last commit snapshot, we use the frozen copy of | |
f5113eff JK |
313 | * that data for IO. If we end up using the existing buffer_head's data |
314 | * for the write, then we have to make sure nobody modifies it while the | |
315 | * IO is in progress. do_get_write_access() handles this. | |
470decc6 | 316 | * |
f5113eff JK |
317 | * The function returns a pointer to the buffer_head to be used for IO. |
318 | * | |
470decc6 DK |
319 | * |
320 | * Return value: | |
321 | * <0: Error | |
322 | * >=0: Finished OK | |
323 | * | |
324 | * On success: | |
325 | * Bit 0 set == escape performed on the data | |
326 | * Bit 1 set == buffer copy-out performed (kfree the data after IO) | |
327 | */ | |
328 | ||
f7f4bccb | 329 | int jbd2_journal_write_metadata_buffer(transaction_t *transaction, |
470decc6 | 330 | struct journal_head *jh_in, |
f5113eff JK |
331 | struct buffer_head **bh_out, |
332 | sector_t blocknr) | |
470decc6 DK |
333 | { |
334 | int need_copy_out = 0; | |
335 | int done_copy_out = 0; | |
336 | int do_escape = 0; | |
337 | char *mapped_data; | |
338 | struct buffer_head *new_bh; | |
470decc6 DK |
339 | struct page *new_page; |
340 | unsigned int new_offset; | |
341 | struct buffer_head *bh_in = jh2bh(jh_in); | |
96577c43 | 342 | journal_t *journal = transaction->t_journal; |
470decc6 DK |
343 | |
344 | /* | |
345 | * The buffer really shouldn't be locked: only the current committing | |
346 | * transaction is allowed to write it, so nobody else is allowed | |
347 | * to do any IO. | |
348 | * | |
349 | * akpm: except if we're journalling data, and write() output is | |
350 | * also part of a shared mapping, and another thread has | |
351 | * decided to launch a writepage() against this buffer. | |
352 | */ | |
353 | J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); | |
354 | ||
47def826 TT |
355 | retry_alloc: |
356 | new_bh = alloc_buffer_head(GFP_NOFS); | |
357 | if (!new_bh) { | |
358 | /* | |
359 | * Failure is not an option, but __GFP_NOFAIL is going | |
360 | * away; so we retry ourselves here. | |
361 | */ | |
362 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
363 | goto retry_alloc; | |
364 | } | |
365 | ||
96577c43 | 366 | /* keep subsequent assertions sane */ |
96577c43 | 367 | atomic_set(&new_bh->b_count, 1); |
470decc6 | 368 | |
f5113eff JK |
369 | jbd_lock_bh_state(bh_in); |
370 | repeat: | |
470decc6 DK |
371 | /* |
372 | * If a new transaction has already done a buffer copy-out, then | |
373 | * we use that version of the data for the commit. | |
374 | */ | |
470decc6 DK |
375 | if (jh_in->b_frozen_data) { |
376 | done_copy_out = 1; | |
377 | new_page = virt_to_page(jh_in->b_frozen_data); | |
378 | new_offset = offset_in_page(jh_in->b_frozen_data); | |
379 | } else { | |
380 | new_page = jh2bh(jh_in)->b_page; | |
381 | new_offset = offset_in_page(jh2bh(jh_in)->b_data); | |
382 | } | |
383 | ||
303a8f2a | 384 | mapped_data = kmap_atomic(new_page); |
e06c8227 | 385 | /* |
13ceef09 JK |
386 | * Fire data frozen trigger if data already wasn't frozen. Do this |
387 | * before checking for escaping, as the trigger may modify the magic | |
388 | * offset. If a copy-out happens afterwards, it will have the correct | |
389 | * data in the buffer. | |
e06c8227 | 390 | */ |
13ceef09 JK |
391 | if (!done_copy_out) |
392 | jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset, | |
393 | jh_in->b_triggers); | |
e06c8227 | 394 | |
470decc6 DK |
395 | /* |
396 | * Check for escaping | |
397 | */ | |
398 | if (*((__be32 *)(mapped_data + new_offset)) == | |
f7f4bccb | 399 | cpu_to_be32(JBD2_MAGIC_NUMBER)) { |
470decc6 DK |
400 | need_copy_out = 1; |
401 | do_escape = 1; | |
402 | } | |
303a8f2a | 403 | kunmap_atomic(mapped_data); |
470decc6 DK |
404 | |
405 | /* | |
406 | * Do we need to do a data copy? | |
407 | */ | |
408 | if (need_copy_out && !done_copy_out) { | |
409 | char *tmp; | |
410 | ||
411 | jbd_unlock_bh_state(bh_in); | |
af1e76d6 | 412 | tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); |
e6ec116b | 413 | if (!tmp) { |
f5113eff | 414 | brelse(new_bh); |
e6ec116b TT |
415 | return -ENOMEM; |
416 | } | |
470decc6 DK |
417 | jbd_lock_bh_state(bh_in); |
418 | if (jh_in->b_frozen_data) { | |
af1e76d6 | 419 | jbd2_free(tmp, bh_in->b_size); |
470decc6 DK |
420 | goto repeat; |
421 | } | |
422 | ||
423 | jh_in->b_frozen_data = tmp; | |
303a8f2a | 424 | mapped_data = kmap_atomic(new_page); |
f5113eff | 425 | memcpy(tmp, mapped_data + new_offset, bh_in->b_size); |
303a8f2a | 426 | kunmap_atomic(mapped_data); |
470decc6 DK |
427 | |
428 | new_page = virt_to_page(tmp); | |
429 | new_offset = offset_in_page(tmp); | |
430 | done_copy_out = 1; | |
e06c8227 JB |
431 | |
432 | /* | |
433 | * This isn't strictly necessary, as we're using frozen | |
434 | * data for the escaping, but it keeps consistency with | |
435 | * b_frozen_data usage. | |
436 | */ | |
437 | jh_in->b_frozen_triggers = jh_in->b_triggers; | |
470decc6 DK |
438 | } |
439 | ||
440 | /* | |
441 | * Did we need to do an escaping? Now we've done all the | |
442 | * copying, we can finally do so. | |
443 | */ | |
444 | if (do_escape) { | |
303a8f2a | 445 | mapped_data = kmap_atomic(new_page); |
470decc6 | 446 | *((unsigned int *)(mapped_data + new_offset)) = 0; |
303a8f2a | 447 | kunmap_atomic(mapped_data); |
470decc6 DK |
448 | } |
449 | ||
470decc6 | 450 | set_bh_page(new_bh, new_page, new_offset); |
f5113eff JK |
451 | new_bh->b_size = bh_in->b_size; |
452 | new_bh->b_bdev = journal->j_dev; | |
470decc6 | 453 | new_bh->b_blocknr = blocknr; |
b34090e5 | 454 | new_bh->b_private = bh_in; |
470decc6 DK |
455 | set_buffer_mapped(new_bh); |
456 | set_buffer_dirty(new_bh); | |
457 | ||
f5113eff | 458 | *bh_out = new_bh; |
470decc6 DK |
459 | |
460 | /* | |
461 | * The to-be-written buffer needs to get moved to the io queue, | |
462 | * and the original buffer whose contents we are shadowing or | |
463 | * copying is moved to the transaction's shadow queue. | |
464 | */ | |
465 | JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); | |
96577c43 | 466 | spin_lock(&journal->j_list_lock); |
467 | __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); | |
468 | spin_unlock(&journal->j_list_lock); | |
b34090e5 | 469 | set_buffer_shadow(bh_in); |
96577c43 | 470 | jbd_unlock_bh_state(bh_in); |
471 | ||
470decc6 DK |
472 | return do_escape | (done_copy_out << 1); |
473 | } | |
474 | ||
475 | /* | |
476 | * Allocation code for the journal file. Manage the space left in the | |
477 | * journal, so that we can begin checkpointing when appropriate. | |
478 | */ | |
479 | ||
480 | /* | |
f7f4bccb | 481 | * __jbd2_log_space_left: Return the number of free blocks left in the journal. |
470decc6 DK |
482 | * |
483 | * Called with the journal already locked. | |
484 | * | |
485 | * Called under j_state_lock | |
486 | */ | |
487 | ||
f7f4bccb | 488 | int __jbd2_log_space_left(journal_t *journal) |
470decc6 DK |
489 | { |
490 | int left = journal->j_free; | |
491 | ||
a931da6a | 492 | /* assert_spin_locked(&journal->j_state_lock); */ |
470decc6 DK |
493 | |
494 | /* | |
495 | * Be pessimistic here about the number of those free blocks which | |
496 | * might be required for log descriptor control blocks. | |
497 | */ | |
498 | ||
499 | #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */ | |
500 | ||
501 | left -= MIN_LOG_RESERVED_BLOCKS; | |
502 | ||
503 | if (left <= 0) | |
504 | return 0; | |
505 | left -= (left >> 3); | |
506 | return left; | |
507 | } | |
508 | ||
509 | /* | |
e4471831 TT |
510 | * Called with j_state_lock locked for writing. |
511 | * Returns true if a transaction commit was started. | |
470decc6 | 512 | */ |
f7f4bccb | 513 | int __jbd2_log_start_commit(journal_t *journal, tid_t target) |
470decc6 | 514 | { |
e7b04ac0 ES |
515 | /* Return if the txn has already requested to be committed */ |
516 | if (journal->j_commit_request == target) | |
517 | return 0; | |
518 | ||
470decc6 | 519 | /* |
deeeaf13 TT |
520 | * The only transaction we can possibly wait upon is the |
521 | * currently running transaction (if it exists). Otherwise, | |
522 | * the target tid must be an old one. | |
470decc6 | 523 | */ |
deeeaf13 TT |
524 | if (journal->j_running_transaction && |
525 | journal->j_running_transaction->t_tid == target) { | |
470decc6 | 526 | /* |
bcf3d0bc | 527 | * We want a new commit: OK, mark the request and wakeup the |
470decc6 DK |
528 | * commit thread. We do _not_ do the commit ourselves. |
529 | */ | |
530 | ||
531 | journal->j_commit_request = target; | |
f2a44523 | 532 | jbd_debug(1, "JBD2: requesting commit %d/%d\n", |
470decc6 DK |
533 | journal->j_commit_request, |
534 | journal->j_commit_sequence); | |
9fff24aa | 535 | journal->j_running_transaction->t_requested = jiffies; |
470decc6 DK |
536 | wake_up(&journal->j_wait_commit); |
537 | return 1; | |
deeeaf13 TT |
538 | } else if (!tid_geq(journal->j_commit_request, target)) |
539 | /* This should never happen, but if it does, preserve | |
540 | the evidence before kjournald goes into a loop and | |
541 | increments j_commit_sequence beyond all recognition. */ | |
f2a44523 | 542 | WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", |
1be2add6 TT |
543 | journal->j_commit_request, |
544 | journal->j_commit_sequence, | |
545 | target, journal->j_running_transaction ? | |
546 | journal->j_running_transaction->t_tid : 0); | |
470decc6 DK |
547 | return 0; |
548 | } | |
549 | ||
f7f4bccb | 550 | int jbd2_log_start_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
551 | { |
552 | int ret; | |
553 | ||
a931da6a | 554 | write_lock(&journal->j_state_lock); |
f7f4bccb | 555 | ret = __jbd2_log_start_commit(journal, tid); |
a931da6a | 556 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
557 | return ret; |
558 | } | |
559 | ||
560 | /* | |
561 | * Force and wait upon a commit if the calling process is not within | |
562 | * transaction. This is used for forcing out undo-protected data which contains | |
563 | * bitmaps, when the fs is running out of space. | |
564 | * | |
565 | * We can only force the running transaction if we don't have an active handle; | |
566 | * otherwise, we will deadlock. | |
567 | * | |
568 | * Returns true if a transaction was started. | |
569 | */ | |
f7f4bccb | 570 | int jbd2_journal_force_commit_nested(journal_t *journal) |
470decc6 DK |
571 | { |
572 | transaction_t *transaction = NULL; | |
573 | tid_t tid; | |
e4471831 | 574 | int need_to_start = 0; |
470decc6 | 575 | |
a931da6a | 576 | read_lock(&journal->j_state_lock); |
470decc6 DK |
577 | if (journal->j_running_transaction && !current->journal_info) { |
578 | transaction = journal->j_running_transaction; | |
e4471831 TT |
579 | if (!tid_geq(journal->j_commit_request, transaction->t_tid)) |
580 | need_to_start = 1; | |
470decc6 DK |
581 | } else if (journal->j_committing_transaction) |
582 | transaction = journal->j_committing_transaction; | |
583 | ||
584 | if (!transaction) { | |
a931da6a | 585 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
586 | return 0; /* Nothing to retry */ |
587 | } | |
588 | ||
589 | tid = transaction->t_tid; | |
a931da6a | 590 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
591 | if (need_to_start) |
592 | jbd2_log_start_commit(journal, tid); | |
f7f4bccb | 593 | jbd2_log_wait_commit(journal, tid); |
470decc6 DK |
594 | return 1; |
595 | } | |
596 | ||
597 | /* | |
598 | * Start a commit of the current running transaction (if any). Returns true | |
c88ccea3 JK |
599 | * if a transaction is going to be committed (or is currently already |
600 | * committing), and fills its tid in at *ptid | |
470decc6 | 601 | */ |
f7f4bccb | 602 | int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) |
470decc6 DK |
603 | { |
604 | int ret = 0; | |
605 | ||
a931da6a | 606 | write_lock(&journal->j_state_lock); |
470decc6 DK |
607 | if (journal->j_running_transaction) { |
608 | tid_t tid = journal->j_running_transaction->t_tid; | |
609 | ||
c88ccea3 JK |
610 | __jbd2_log_start_commit(journal, tid); |
611 | /* There's a running transaction and we've just made sure | |
612 | * it's commit has been scheduled. */ | |
613 | if (ptid) | |
470decc6 | 614 | *ptid = tid; |
c88ccea3 JK |
615 | ret = 1; |
616 | } else if (journal->j_committing_transaction) { | |
470decc6 | 617 | /* |
12810ad7 AB |
618 | * If commit has been started, then we have to wait for |
619 | * completion of that transaction. | |
470decc6 | 620 | */ |
c88ccea3 JK |
621 | if (ptid) |
622 | *ptid = journal->j_committing_transaction->t_tid; | |
470decc6 DK |
623 | ret = 1; |
624 | } | |
a931da6a | 625 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
626 | return ret; |
627 | } | |
628 | ||
bbd2be36 JK |
629 | /* |
630 | * Return 1 if a given transaction has not yet sent barrier request | |
631 | * connected with a transaction commit. If 0 is returned, transaction | |
632 | * may or may not have sent the barrier. Used to avoid sending barrier | |
633 | * twice in common cases. | |
634 | */ | |
635 | int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) | |
636 | { | |
637 | int ret = 0; | |
638 | transaction_t *commit_trans; | |
639 | ||
640 | if (!(journal->j_flags & JBD2_BARRIER)) | |
641 | return 0; | |
642 | read_lock(&journal->j_state_lock); | |
643 | /* Transaction already committed? */ | |
644 | if (tid_geq(journal->j_commit_sequence, tid)) | |
645 | goto out; | |
646 | commit_trans = journal->j_committing_transaction; | |
647 | if (!commit_trans || commit_trans->t_tid != tid) { | |
648 | ret = 1; | |
649 | goto out; | |
650 | } | |
651 | /* | |
652 | * Transaction is being committed and we already proceeded to | |
653 | * submitting a flush to fs partition? | |
654 | */ | |
655 | if (journal->j_fs_dev != journal->j_dev) { | |
656 | if (!commit_trans->t_need_data_flush || | |
657 | commit_trans->t_state >= T_COMMIT_DFLUSH) | |
658 | goto out; | |
659 | } else { | |
660 | if (commit_trans->t_state >= T_COMMIT_JFLUSH) | |
661 | goto out; | |
662 | } | |
663 | ret = 1; | |
664 | out: | |
665 | read_unlock(&journal->j_state_lock); | |
666 | return ret; | |
667 | } | |
668 | EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); | |
669 | ||
470decc6 DK |
670 | /* |
671 | * Wait for a specified commit to complete. | |
672 | * The caller may not hold the journal lock. | |
673 | */ | |
f7f4bccb | 674 | int jbd2_log_wait_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
675 | { |
676 | int err = 0; | |
677 | ||
a931da6a | 678 | read_lock(&journal->j_state_lock); |
e23291b9 | 679 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
680 | if (!tid_geq(journal->j_commit_request, tid)) { |
681 | printk(KERN_EMERG | |
682 | "%s: error: j_commit_request=%d, tid=%d\n", | |
329d291f | 683 | __func__, journal->j_commit_request, tid); |
470decc6 | 684 | } |
470decc6 | 685 | #endif |
470decc6 | 686 | while (tid_gt(tid, journal->j_commit_sequence)) { |
f2a44523 | 687 | jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n", |
470decc6 DK |
688 | tid, journal->j_commit_sequence); |
689 | wake_up(&journal->j_wait_commit); | |
a931da6a | 690 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
691 | wait_event(journal->j_wait_done_commit, |
692 | !tid_gt(tid, journal->j_commit_sequence)); | |
a931da6a | 693 | read_lock(&journal->j_state_lock); |
470decc6 | 694 | } |
a931da6a | 695 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
696 | |
697 | if (unlikely(is_journal_aborted(journal))) { | |
698 | printk(KERN_EMERG "journal commit I/O error\n"); | |
699 | err = -EIO; | |
700 | } | |
701 | return err; | |
702 | } | |
703 | ||
d76a3a77 TT |
704 | /* |
705 | * When this function returns the transaction corresponding to tid | |
706 | * will be completed. If the transaction has currently running, start | |
707 | * committing that transaction before waiting for it to complete. If | |
708 | * the transaction id is stale, it is by definition already completed, | |
709 | * so just return SUCCESS. | |
710 | */ | |
711 | int jbd2_complete_transaction(journal_t *journal, tid_t tid) | |
712 | { | |
713 | int need_to_wait = 1; | |
714 | ||
715 | read_lock(&journal->j_state_lock); | |
716 | if (journal->j_running_transaction && | |
717 | journal->j_running_transaction->t_tid == tid) { | |
718 | if (journal->j_commit_request != tid) { | |
719 | /* transaction not yet started, so request it */ | |
720 | read_unlock(&journal->j_state_lock); | |
721 | jbd2_log_start_commit(journal, tid); | |
722 | goto wait_commit; | |
723 | } | |
724 | } else if (!(journal->j_committing_transaction && | |
725 | journal->j_committing_transaction->t_tid == tid)) | |
726 | need_to_wait = 0; | |
727 | read_unlock(&journal->j_state_lock); | |
728 | if (!need_to_wait) | |
729 | return 0; | |
730 | wait_commit: | |
731 | return jbd2_log_wait_commit(journal, tid); | |
732 | } | |
733 | EXPORT_SYMBOL(jbd2_complete_transaction); | |
734 | ||
470decc6 DK |
735 | /* |
736 | * Log buffer allocation routines: | |
737 | */ | |
738 | ||
18eba7aa | 739 | int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) |
470decc6 DK |
740 | { |
741 | unsigned long blocknr; | |
742 | ||
a931da6a | 743 | write_lock(&journal->j_state_lock); |
470decc6 DK |
744 | J_ASSERT(journal->j_free > 1); |
745 | ||
746 | blocknr = journal->j_head; | |
747 | journal->j_head++; | |
748 | journal->j_free--; | |
749 | if (journal->j_head == journal->j_last) | |
750 | journal->j_head = journal->j_first; | |
a931da6a | 751 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 752 | return jbd2_journal_bmap(journal, blocknr, retp); |
470decc6 DK |
753 | } |
754 | ||
755 | /* | |
756 | * Conversion of logical to physical block numbers for the journal | |
757 | * | |
758 | * On external journals the journal blocks are identity-mapped, so | |
759 | * this is a no-op. If needed, we can use j_blk_offset - everything is | |
760 | * ready. | |
761 | */ | |
f7f4bccb | 762 | int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, |
18eba7aa | 763 | unsigned long long *retp) |
470decc6 DK |
764 | { |
765 | int err = 0; | |
18eba7aa | 766 | unsigned long long ret; |
470decc6 DK |
767 | |
768 | if (journal->j_inode) { | |
769 | ret = bmap(journal->j_inode, blocknr); | |
770 | if (ret) | |
771 | *retp = ret; | |
772 | else { | |
470decc6 DK |
773 | printk(KERN_ALERT "%s: journal block not found " |
774 | "at offset %lu on %s\n", | |
05496769 | 775 | __func__, blocknr, journal->j_devname); |
470decc6 DK |
776 | err = -EIO; |
777 | __journal_abort_soft(journal, err); | |
778 | } | |
779 | } else { | |
780 | *retp = blocknr; /* +journal->j_blk_offset */ | |
781 | } | |
782 | return err; | |
783 | } | |
784 | ||
785 | /* | |
786 | * We play buffer_head aliasing tricks to write data/metadata blocks to | |
787 | * the journal without copying their contents, but for journal | |
788 | * descriptor blocks we do need to generate bona fide buffers. | |
789 | * | |
f7f4bccb | 790 | * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying |
470decc6 DK |
791 | * the buffer's contents they really should run flush_dcache_page(bh->b_page). |
792 | * But we don't bother doing that, so there will be coherency problems with | |
793 | * mmaps of blockdevs which hold live JBD-controlled filesystems. | |
794 | */ | |
e5a120ae | 795 | struct buffer_head *jbd2_journal_get_descriptor_buffer(journal_t *journal) |
470decc6 DK |
796 | { |
797 | struct buffer_head *bh; | |
18eba7aa | 798 | unsigned long long blocknr; |
470decc6 DK |
799 | int err; |
800 | ||
f7f4bccb | 801 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
802 | |
803 | if (err) | |
804 | return NULL; | |
805 | ||
806 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
4b905671 JK |
807 | if (!bh) |
808 | return NULL; | |
470decc6 DK |
809 | lock_buffer(bh); |
810 | memset(bh->b_data, 0, journal->j_blocksize); | |
811 | set_buffer_uptodate(bh); | |
812 | unlock_buffer(bh); | |
813 | BUFFER_TRACE(bh, "return this buffer"); | |
e5a120ae | 814 | return bh; |
470decc6 DK |
815 | } |
816 | ||
79feb521 JK |
817 | /* |
818 | * Return tid of the oldest transaction in the journal and block in the journal | |
819 | * where the transaction starts. | |
820 | * | |
821 | * If the journal is now empty, return which will be the next transaction ID | |
822 | * we will write and where will that transaction start. | |
823 | * | |
824 | * The return value is 0 if journal tail cannot be pushed any further, 1 if | |
825 | * it can. | |
826 | */ | |
827 | int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, | |
828 | unsigned long *block) | |
829 | { | |
830 | transaction_t *transaction; | |
831 | int ret; | |
832 | ||
833 | read_lock(&journal->j_state_lock); | |
834 | spin_lock(&journal->j_list_lock); | |
835 | transaction = journal->j_checkpoint_transactions; | |
836 | if (transaction) { | |
837 | *tid = transaction->t_tid; | |
838 | *block = transaction->t_log_start; | |
839 | } else if ((transaction = journal->j_committing_transaction) != NULL) { | |
840 | *tid = transaction->t_tid; | |
841 | *block = transaction->t_log_start; | |
842 | } else if ((transaction = journal->j_running_transaction) != NULL) { | |
843 | *tid = transaction->t_tid; | |
844 | *block = journal->j_head; | |
845 | } else { | |
846 | *tid = journal->j_transaction_sequence; | |
847 | *block = journal->j_head; | |
848 | } | |
849 | ret = tid_gt(*tid, journal->j_tail_sequence); | |
850 | spin_unlock(&journal->j_list_lock); | |
851 | read_unlock(&journal->j_state_lock); | |
852 | ||
853 | return ret; | |
854 | } | |
855 | ||
856 | /* | |
857 | * Update information in journal structure and in on disk journal superblock | |
858 | * about log tail. This function does not check whether information passed in | |
859 | * really pushes log tail further. It's responsibility of the caller to make | |
860 | * sure provided log tail information is valid (e.g. by holding | |
861 | * j_checkpoint_mutex all the time between computing log tail and calling this | |
862 | * function as is the case with jbd2_cleanup_journal_tail()). | |
863 | * | |
864 | * Requires j_checkpoint_mutex | |
865 | */ | |
866 | void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) | |
867 | { | |
868 | unsigned long freed; | |
869 | ||
870 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); | |
871 | ||
872 | /* | |
873 | * We cannot afford for write to remain in drive's caches since as | |
874 | * soon as we update j_tail, next transaction can start reusing journal | |
875 | * space and if we lose sb update during power failure we'd replay | |
876 | * old transaction with possibly newly overwritten data. | |
877 | */ | |
878 | jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA); | |
879 | write_lock(&journal->j_state_lock); | |
880 | freed = block - journal->j_tail; | |
881 | if (block < journal->j_tail) | |
882 | freed += journal->j_last - journal->j_first; | |
883 | ||
884 | trace_jbd2_update_log_tail(journal, tid, block, freed); | |
885 | jbd_debug(1, | |
886 | "Cleaning journal tail from %d to %d (offset %lu), " | |
887 | "freeing %lu\n", | |
888 | journal->j_tail_sequence, tid, block, freed); | |
889 | ||
890 | journal->j_free += freed; | |
891 | journal->j_tail_sequence = tid; | |
892 | journal->j_tail = block; | |
893 | write_unlock(&journal->j_state_lock); | |
894 | } | |
895 | ||
3339578f JK |
896 | /* |
897 | * This is a variaon of __jbd2_update_log_tail which checks for validity of | |
898 | * provided log tail and locks j_checkpoint_mutex. So it is safe against races | |
899 | * with other threads updating log tail. | |
900 | */ | |
901 | void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) | |
902 | { | |
903 | mutex_lock(&journal->j_checkpoint_mutex); | |
904 | if (tid_gt(tid, journal->j_tail_sequence)) | |
905 | __jbd2_update_log_tail(journal, tid, block); | |
906 | mutex_unlock(&journal->j_checkpoint_mutex); | |
907 | } | |
908 | ||
8e85fb3f JL |
909 | struct jbd2_stats_proc_session { |
910 | journal_t *journal; | |
911 | struct transaction_stats_s *stats; | |
912 | int start; | |
913 | int max; | |
914 | }; | |
915 | ||
8e85fb3f JL |
916 | static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) |
917 | { | |
918 | return *pos ? NULL : SEQ_START_TOKEN; | |
919 | } | |
920 | ||
921 | static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) | |
922 | { | |
923 | return NULL; | |
924 | } | |
925 | ||
926 | static int jbd2_seq_info_show(struct seq_file *seq, void *v) | |
927 | { | |
928 | struct jbd2_stats_proc_session *s = seq->private; | |
929 | ||
930 | if (v != SEQ_START_TOKEN) | |
931 | return 0; | |
9fff24aa TT |
932 | seq_printf(seq, "%lu transactions (%lu requested), " |
933 | "each up to %u blocks\n", | |
934 | s->stats->ts_tid, s->stats->ts_requested, | |
935 | s->journal->j_max_transaction_buffers); | |
8e85fb3f JL |
936 | if (s->stats->ts_tid == 0) |
937 | return 0; | |
938 | seq_printf(seq, "average: \n %ums waiting for transaction\n", | |
bf699327 | 939 | jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); |
9fff24aa TT |
940 | seq_printf(seq, " %ums request delay\n", |
941 | (s->stats->ts_requested == 0) ? 0 : | |
942 | jiffies_to_msecs(s->stats->run.rs_request_delay / | |
943 | s->stats->ts_requested)); | |
8e85fb3f | 944 | seq_printf(seq, " %ums running transaction\n", |
bf699327 | 945 | jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); |
8e85fb3f | 946 | seq_printf(seq, " %ums transaction was being locked\n", |
bf699327 | 947 | jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); |
8e85fb3f | 948 | seq_printf(seq, " %ums flushing data (in ordered mode)\n", |
bf699327 | 949 | jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); |
8e85fb3f | 950 | seq_printf(seq, " %ums logging transaction\n", |
bf699327 | 951 | jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); |
c225aa57 SHT |
952 | seq_printf(seq, " %lluus average transaction commit time\n", |
953 | div_u64(s->journal->j_average_commit_time, 1000)); | |
8e85fb3f | 954 | seq_printf(seq, " %lu handles per transaction\n", |
bf699327 | 955 | s->stats->run.rs_handle_count / s->stats->ts_tid); |
8e85fb3f | 956 | seq_printf(seq, " %lu blocks per transaction\n", |
bf699327 | 957 | s->stats->run.rs_blocks / s->stats->ts_tid); |
8e85fb3f | 958 | seq_printf(seq, " %lu logged blocks per transaction\n", |
bf699327 | 959 | s->stats->run.rs_blocks_logged / s->stats->ts_tid); |
8e85fb3f JL |
960 | return 0; |
961 | } | |
962 | ||
963 | static void jbd2_seq_info_stop(struct seq_file *seq, void *v) | |
964 | { | |
965 | } | |
966 | ||
88e9d34c | 967 | static const struct seq_operations jbd2_seq_info_ops = { |
8e85fb3f JL |
968 | .start = jbd2_seq_info_start, |
969 | .next = jbd2_seq_info_next, | |
970 | .stop = jbd2_seq_info_stop, | |
971 | .show = jbd2_seq_info_show, | |
972 | }; | |
973 | ||
974 | static int jbd2_seq_info_open(struct inode *inode, struct file *file) | |
975 | { | |
d9dda78b | 976 | journal_t *journal = PDE_DATA(inode); |
8e85fb3f JL |
977 | struct jbd2_stats_proc_session *s; |
978 | int rc, size; | |
979 | ||
980 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
981 | if (s == NULL) | |
982 | return -ENOMEM; | |
983 | size = sizeof(struct transaction_stats_s); | |
984 | s->stats = kmalloc(size, GFP_KERNEL); | |
985 | if (s->stats == NULL) { | |
986 | kfree(s); | |
987 | return -ENOMEM; | |
988 | } | |
989 | spin_lock(&journal->j_history_lock); | |
990 | memcpy(s->stats, &journal->j_stats, size); | |
991 | s->journal = journal; | |
992 | spin_unlock(&journal->j_history_lock); | |
993 | ||
994 | rc = seq_open(file, &jbd2_seq_info_ops); | |
995 | if (rc == 0) { | |
996 | struct seq_file *m = file->private_data; | |
997 | m->private = s; | |
998 | } else { | |
999 | kfree(s->stats); | |
1000 | kfree(s); | |
1001 | } | |
1002 | return rc; | |
1003 | ||
1004 | } | |
1005 | ||
1006 | static int jbd2_seq_info_release(struct inode *inode, struct file *file) | |
1007 | { | |
1008 | struct seq_file *seq = file->private_data; | |
1009 | struct jbd2_stats_proc_session *s = seq->private; | |
1010 | kfree(s->stats); | |
1011 | kfree(s); | |
1012 | return seq_release(inode, file); | |
1013 | } | |
1014 | ||
828c0950 | 1015 | static const struct file_operations jbd2_seq_info_fops = { |
8e85fb3f JL |
1016 | .owner = THIS_MODULE, |
1017 | .open = jbd2_seq_info_open, | |
1018 | .read = seq_read, | |
1019 | .llseek = seq_lseek, | |
1020 | .release = jbd2_seq_info_release, | |
1021 | }; | |
1022 | ||
1023 | static struct proc_dir_entry *proc_jbd2_stats; | |
1024 | ||
1025 | static void jbd2_stats_proc_init(journal_t *journal) | |
1026 | { | |
05496769 | 1027 | journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); |
8e85fb3f | 1028 | if (journal->j_proc_entry) { |
79da3664 DL |
1029 | proc_create_data("info", S_IRUGO, journal->j_proc_entry, |
1030 | &jbd2_seq_info_fops, journal); | |
8e85fb3f JL |
1031 | } |
1032 | } | |
1033 | ||
1034 | static void jbd2_stats_proc_exit(journal_t *journal) | |
1035 | { | |
8e85fb3f | 1036 | remove_proc_entry("info", journal->j_proc_entry); |
05496769 | 1037 | remove_proc_entry(journal->j_devname, proc_jbd2_stats); |
8e85fb3f JL |
1038 | } |
1039 | ||
470decc6 DK |
1040 | /* |
1041 | * Management for journal control blocks: functions to create and | |
1042 | * destroy journal_t structures, and to initialise and read existing | |
1043 | * journal blocks from disk. */ | |
1044 | ||
1045 | /* First: create and setup a journal_t object in memory. We initialise | |
1046 | * very few fields yet: that has to wait until we have created the | |
1047 | * journal structures from from scratch, or loaded them from disk. */ | |
1048 | ||
1049 | static journal_t * journal_init_common (void) | |
1050 | { | |
1051 | journal_t *journal; | |
1052 | int err; | |
1053 | ||
3ebfdf88 | 1054 | journal = kzalloc(sizeof(*journal), GFP_KERNEL); |
470decc6 | 1055 | if (!journal) |
b7271b0a | 1056 | return NULL; |
470decc6 DK |
1057 | |
1058 | init_waitqueue_head(&journal->j_wait_transaction_locked); | |
1059 | init_waitqueue_head(&journal->j_wait_logspace); | |
1060 | init_waitqueue_head(&journal->j_wait_done_commit); | |
1061 | init_waitqueue_head(&journal->j_wait_checkpoint); | |
1062 | init_waitqueue_head(&journal->j_wait_commit); | |
1063 | init_waitqueue_head(&journal->j_wait_updates); | |
1064 | mutex_init(&journal->j_barrier); | |
1065 | mutex_init(&journal->j_checkpoint_mutex); | |
1066 | spin_lock_init(&journal->j_revoke_lock); | |
1067 | spin_lock_init(&journal->j_list_lock); | |
a931da6a | 1068 | rwlock_init(&journal->j_state_lock); |
470decc6 | 1069 | |
cd02ff0b | 1070 | journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); |
30773840 TT |
1071 | journal->j_min_batch_time = 0; |
1072 | journal->j_max_batch_time = 15000; /* 15ms */ | |
470decc6 DK |
1073 | |
1074 | /* The journal is marked for error until we succeed with recovery! */ | |
f7f4bccb | 1075 | journal->j_flags = JBD2_ABORT; |
470decc6 DK |
1076 | |
1077 | /* Set up a default-sized revoke table for the new mount. */ | |
f7f4bccb | 1078 | err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); |
470decc6 DK |
1079 | if (err) { |
1080 | kfree(journal); | |
b7271b0a | 1081 | return NULL; |
470decc6 | 1082 | } |
8e85fb3f | 1083 | |
bf699327 | 1084 | spin_lock_init(&journal->j_history_lock); |
8e85fb3f | 1085 | |
470decc6 | 1086 | return journal; |
470decc6 DK |
1087 | } |
1088 | ||
f7f4bccb | 1089 | /* jbd2_journal_init_dev and jbd2_journal_init_inode: |
470decc6 DK |
1090 | * |
1091 | * Create a journal structure assigned some fixed set of disk blocks to | |
1092 | * the journal. We don't actually touch those disk blocks yet, but we | |
1093 | * need to set up all of the mapping information to tell the journaling | |
1094 | * system where the journal blocks are. | |
1095 | * | |
1096 | */ | |
1097 | ||
1098 | /** | |
5648ba5b | 1099 | * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure |
470decc6 DK |
1100 | * @bdev: Block device on which to create the journal |
1101 | * @fs_dev: Device which hold journalled filesystem for this journal. | |
1102 | * @start: Block nr Start of journal. | |
1103 | * @len: Length of the journal in blocks. | |
1104 | * @blocksize: blocksize of journalling device | |
5648ba5b RD |
1105 | * |
1106 | * Returns: a newly created journal_t * | |
470decc6 | 1107 | * |
f7f4bccb | 1108 | * jbd2_journal_init_dev creates a journal which maps a fixed contiguous |
470decc6 DK |
1109 | * range of blocks on an arbitrary block device. |
1110 | * | |
1111 | */ | |
f7f4bccb | 1112 | journal_t * jbd2_journal_init_dev(struct block_device *bdev, |
470decc6 | 1113 | struct block_device *fs_dev, |
18eba7aa | 1114 | unsigned long long start, int len, int blocksize) |
470decc6 DK |
1115 | { |
1116 | journal_t *journal = journal_init_common(); | |
1117 | struct buffer_head *bh; | |
05496769 | 1118 | char *p; |
470decc6 DK |
1119 | int n; |
1120 | ||
1121 | if (!journal) | |
1122 | return NULL; | |
1123 | ||
1124 | /* journal descriptor can store up to n blocks -bzzz */ | |
1125 | journal->j_blocksize = blocksize; | |
0587aa3d | 1126 | journal->j_dev = bdev; |
1127 | journal->j_fs_dev = fs_dev; | |
1128 | journal->j_blk_offset = start; | |
1129 | journal->j_maxlen = len; | |
1130 | bdevname(journal->j_dev, journal->j_devname); | |
1131 | p = journal->j_devname; | |
1132 | while ((p = strchr(p, '/'))) | |
1133 | *p = '!'; | |
4b905671 | 1134 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1135 | n = journal->j_blocksize / sizeof(journal_block_tag_t); |
1136 | journal->j_wbufsize = n; | |
1137 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1138 | if (!journal->j_wbuf) { | |
25985edc | 1139 | printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", |
329d291f | 1140 | __func__); |
4b905671 | 1141 | goto out_err; |
470decc6 | 1142 | } |
470decc6 DK |
1143 | |
1144 | bh = __getblk(journal->j_dev, start, journal->j_blocksize); | |
4b905671 JK |
1145 | if (!bh) { |
1146 | printk(KERN_ERR | |
1147 | "%s: Cannot get buffer for journal superblock\n", | |
1148 | __func__); | |
1149 | goto out_err; | |
1150 | } | |
470decc6 DK |
1151 | journal->j_sb_buffer = bh; |
1152 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
4b905671 | 1153 | |
470decc6 | 1154 | return journal; |
4b905671 | 1155 | out_err: |
7b02bec0 | 1156 | kfree(journal->j_wbuf); |
4b905671 JK |
1157 | jbd2_stats_proc_exit(journal); |
1158 | kfree(journal); | |
1159 | return NULL; | |
470decc6 DK |
1160 | } |
1161 | ||
1162 | /** | |
f7f4bccb | 1163 | * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. |
470decc6 DK |
1164 | * @inode: An inode to create the journal in |
1165 | * | |
f7f4bccb | 1166 | * jbd2_journal_init_inode creates a journal which maps an on-disk inode as |
470decc6 DK |
1167 | * the journal. The inode must exist already, must support bmap() and |
1168 | * must have all data blocks preallocated. | |
1169 | */ | |
f7f4bccb | 1170 | journal_t * jbd2_journal_init_inode (struct inode *inode) |
470decc6 DK |
1171 | { |
1172 | struct buffer_head *bh; | |
1173 | journal_t *journal = journal_init_common(); | |
05496769 | 1174 | char *p; |
470decc6 DK |
1175 | int err; |
1176 | int n; | |
18eba7aa | 1177 | unsigned long long blocknr; |
470decc6 DK |
1178 | |
1179 | if (!journal) | |
1180 | return NULL; | |
1181 | ||
1182 | journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev; | |
1183 | journal->j_inode = inode; | |
05496769 TT |
1184 | bdevname(journal->j_dev, journal->j_devname); |
1185 | p = journal->j_devname; | |
1186 | while ((p = strchr(p, '/'))) | |
1187 | *p = '!'; | |
1188 | p = journal->j_devname + strlen(journal->j_devname); | |
90576c0b | 1189 | sprintf(p, "-%lu", journal->j_inode->i_ino); |
470decc6 DK |
1190 | jbd_debug(1, |
1191 | "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n", | |
1192 | journal, inode->i_sb->s_id, inode->i_ino, | |
1193 | (long long) inode->i_size, | |
1194 | inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); | |
1195 | ||
1196 | journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits; | |
1197 | journal->j_blocksize = inode->i_sb->s_blocksize; | |
8e85fb3f | 1198 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1199 | |
1200 | /* journal descriptor can store up to n blocks -bzzz */ | |
1201 | n = journal->j_blocksize / sizeof(journal_block_tag_t); | |
1202 | journal->j_wbufsize = n; | |
1203 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1204 | if (!journal->j_wbuf) { | |
25985edc | 1205 | printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", |
329d291f | 1206 | __func__); |
4b905671 | 1207 | goto out_err; |
470decc6 DK |
1208 | } |
1209 | ||
f7f4bccb | 1210 | err = jbd2_journal_bmap(journal, 0, &blocknr); |
470decc6 DK |
1211 | /* If that failed, give up */ |
1212 | if (err) { | |
3c26bdb4 | 1213 | printk(KERN_ERR "%s: Cannot locate journal superblock\n", |
329d291f | 1214 | __func__); |
4b905671 | 1215 | goto out_err; |
470decc6 DK |
1216 | } |
1217 | ||
1218 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
4b905671 JK |
1219 | if (!bh) { |
1220 | printk(KERN_ERR | |
1221 | "%s: Cannot get buffer for journal superblock\n", | |
1222 | __func__); | |
1223 | goto out_err; | |
1224 | } | |
470decc6 DK |
1225 | journal->j_sb_buffer = bh; |
1226 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
1227 | ||
1228 | return journal; | |
4b905671 | 1229 | out_err: |
7b02bec0 | 1230 | kfree(journal->j_wbuf); |
4b905671 JK |
1231 | jbd2_stats_proc_exit(journal); |
1232 | kfree(journal); | |
1233 | return NULL; | |
470decc6 DK |
1234 | } |
1235 | ||
1236 | /* | |
1237 | * If the journal init or create aborts, we need to mark the journal | |
1238 | * superblock as being NULL to prevent the journal destroy from writing | |
1239 | * back a bogus superblock. | |
1240 | */ | |
1241 | static void journal_fail_superblock (journal_t *journal) | |
1242 | { | |
1243 | struct buffer_head *bh = journal->j_sb_buffer; | |
1244 | brelse(bh); | |
1245 | journal->j_sb_buffer = NULL; | |
1246 | } | |
1247 | ||
1248 | /* | |
1249 | * Given a journal_t structure, initialise the various fields for | |
1250 | * startup of a new journaling session. We use this both when creating | |
1251 | * a journal, and after recovering an old journal to reset it for | |
1252 | * subsequent use. | |
1253 | */ | |
1254 | ||
1255 | static int journal_reset(journal_t *journal) | |
1256 | { | |
1257 | journal_superblock_t *sb = journal->j_superblock; | |
18eba7aa | 1258 | unsigned long long first, last; |
470decc6 DK |
1259 | |
1260 | first = be32_to_cpu(sb->s_first); | |
1261 | last = be32_to_cpu(sb->s_maxlen); | |
f6f50e28 | 1262 | if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) { |
f2a44523 | 1263 | printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n", |
f6f50e28 JK |
1264 | first, last); |
1265 | journal_fail_superblock(journal); | |
1266 | return -EINVAL; | |
1267 | } | |
470decc6 DK |
1268 | |
1269 | journal->j_first = first; | |
1270 | journal->j_last = last; | |
1271 | ||
1272 | journal->j_head = first; | |
1273 | journal->j_tail = first; | |
1274 | journal->j_free = last - first; | |
1275 | ||
1276 | journal->j_tail_sequence = journal->j_transaction_sequence; | |
1277 | journal->j_commit_sequence = journal->j_transaction_sequence - 1; | |
1278 | journal->j_commit_request = journal->j_commit_sequence; | |
1279 | ||
1280 | journal->j_max_transaction_buffers = journal->j_maxlen / 4; | |
1281 | ||
470decc6 DK |
1282 | /* |
1283 | * As a special case, if the on-disk copy is already marked as needing | |
24bcc89c JK |
1284 | * no recovery (s_start == 0), then we can safely defer the superblock |
1285 | * update until the next commit by setting JBD2_FLUSHED. This avoids | |
470decc6 DK |
1286 | * attempting a write to a potential-readonly device. |
1287 | */ | |
24bcc89c | 1288 | if (sb->s_start == 0) { |
f2a44523 | 1289 | jbd_debug(1, "JBD2: Skipping superblock update on recovered sb " |
470decc6 DK |
1290 | "(start %ld, seq %d, errno %d)\n", |
1291 | journal->j_tail, journal->j_tail_sequence, | |
1292 | journal->j_errno); | |
24bcc89c JK |
1293 | journal->j_flags |= JBD2_FLUSHED; |
1294 | } else { | |
a78bb11d JK |
1295 | /* Lock here to make assertions happy... */ |
1296 | mutex_lock(&journal->j_checkpoint_mutex); | |
79feb521 JK |
1297 | /* |
1298 | * Update log tail information. We use WRITE_FUA since new | |
1299 | * transaction will start reusing journal space and so we | |
1300 | * must make sure information about current log tail is on | |
1301 | * disk before that. | |
1302 | */ | |
1303 | jbd2_journal_update_sb_log_tail(journal, | |
1304 | journal->j_tail_sequence, | |
1305 | journal->j_tail, | |
1306 | WRITE_FUA); | |
a78bb11d | 1307 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 | 1308 | } |
24bcc89c JK |
1309 | return jbd2_journal_start_thread(journal); |
1310 | } | |
470decc6 | 1311 | |
79feb521 | 1312 | static void jbd2_write_superblock(journal_t *journal, int write_op) |
24bcc89c JK |
1313 | { |
1314 | struct buffer_head *bh = journal->j_sb_buffer; | |
79feb521 | 1315 | int ret; |
470decc6 | 1316 | |
79feb521 JK |
1317 | trace_jbd2_write_superblock(journal, write_op); |
1318 | if (!(journal->j_flags & JBD2_BARRIER)) | |
1319 | write_op &= ~(REQ_FUA | REQ_FLUSH); | |
1320 | lock_buffer(bh); | |
914258bf TT |
1321 | if (buffer_write_io_error(bh)) { |
1322 | /* | |
1323 | * Oh, dear. A previous attempt to write the journal | |
1324 | * superblock failed. This could happen because the | |
1325 | * USB device was yanked out. Or it could happen to | |
1326 | * be a transient write error and maybe the block will | |
1327 | * be remapped. Nothing we can do but to retry the | |
1328 | * write and hope for the best. | |
1329 | */ | |
1330 | printk(KERN_ERR "JBD2: previous I/O error detected " | |
1331 | "for journal superblock update for %s.\n", | |
1332 | journal->j_devname); | |
1333 | clear_buffer_write_io_error(bh); | |
1334 | set_buffer_uptodate(bh); | |
1335 | } | |
79feb521 JK |
1336 | get_bh(bh); |
1337 | bh->b_end_io = end_buffer_write_sync; | |
1338 | ret = submit_bh(write_op, bh); | |
1339 | wait_on_buffer(bh); | |
24bcc89c | 1340 | if (buffer_write_io_error(bh)) { |
24bcc89c JK |
1341 | clear_buffer_write_io_error(bh); |
1342 | set_buffer_uptodate(bh); | |
79feb521 JK |
1343 | ret = -EIO; |
1344 | } | |
1345 | if (ret) { | |
1346 | printk(KERN_ERR "JBD2: Error %d detected when updating " | |
1347 | "journal superblock for %s.\n", ret, | |
1348 | journal->j_devname); | |
24bcc89c JK |
1349 | } |
1350 | } | |
1351 | ||
1352 | /** | |
1353 | * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. | |
1354 | * @journal: The journal to update. | |
79feb521 JK |
1355 | * @tail_tid: TID of the new transaction at the tail of the log |
1356 | * @tail_block: The first block of the transaction at the tail of the log | |
1357 | * @write_op: With which operation should we write the journal sb | |
24bcc89c JK |
1358 | * |
1359 | * Update a journal's superblock information about log tail and write it to | |
1360 | * disk, waiting for the IO to complete. | |
1361 | */ | |
79feb521 JK |
1362 | void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, |
1363 | unsigned long tail_block, int write_op) | |
24bcc89c JK |
1364 | { |
1365 | journal_superblock_t *sb = journal->j_superblock; | |
1366 | ||
a78bb11d | 1367 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); |
79feb521 JK |
1368 | jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", |
1369 | tail_block, tail_tid); | |
470decc6 | 1370 | |
79feb521 JK |
1371 | sb->s_sequence = cpu_to_be32(tail_tid); |
1372 | sb->s_start = cpu_to_be32(tail_block); | |
470decc6 | 1373 | |
79feb521 | 1374 | jbd2_write_superblock(journal, write_op); |
470decc6 | 1375 | |
24bcc89c JK |
1376 | /* Log is no longer empty */ |
1377 | write_lock(&journal->j_state_lock); | |
1378 | WARN_ON(!sb->s_sequence); | |
1379 | journal->j_flags &= ~JBD2_FLUSHED; | |
1380 | write_unlock(&journal->j_state_lock); | |
1381 | } | |
2201c590 | 1382 | |
24bcc89c JK |
1383 | /** |
1384 | * jbd2_mark_journal_empty() - Mark on disk journal as empty. | |
1385 | * @journal: The journal to update. | |
1386 | * | |
1387 | * Update a journal's dynamic superblock fields to show that journal is empty. | |
1388 | * Write updated superblock to disk waiting for IO to complete. | |
1389 | */ | |
1390 | static void jbd2_mark_journal_empty(journal_t *journal) | |
1391 | { | |
1392 | journal_superblock_t *sb = journal->j_superblock; | |
914258bf | 1393 | |
a78bb11d | 1394 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); |
a931da6a | 1395 | read_lock(&journal->j_state_lock); |
eeecef0a ES |
1396 | /* Is it already empty? */ |
1397 | if (sb->s_start == 0) { | |
1398 | read_unlock(&journal->j_state_lock); | |
1399 | return; | |
1400 | } | |
24bcc89c JK |
1401 | jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n", |
1402 | journal->j_tail_sequence); | |
470decc6 DK |
1403 | |
1404 | sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); | |
24bcc89c | 1405 | sb->s_start = cpu_to_be32(0); |
a931da6a | 1406 | read_unlock(&journal->j_state_lock); |
470decc6 | 1407 | |
79feb521 | 1408 | jbd2_write_superblock(journal, WRITE_FUA); |
470decc6 | 1409 | |
24bcc89c | 1410 | /* Log is no longer empty */ |
a931da6a | 1411 | write_lock(&journal->j_state_lock); |
24bcc89c | 1412 | journal->j_flags |= JBD2_FLUSHED; |
a931da6a | 1413 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1414 | } |
1415 | ||
24bcc89c JK |
1416 | |
1417 | /** | |
1418 | * jbd2_journal_update_sb_errno() - Update error in the journal. | |
1419 | * @journal: The journal to update. | |
1420 | * | |
1421 | * Update a journal's errno. Write updated superblock to disk waiting for IO | |
1422 | * to complete. | |
1423 | */ | |
d796c52e | 1424 | void jbd2_journal_update_sb_errno(journal_t *journal) |
24bcc89c JK |
1425 | { |
1426 | journal_superblock_t *sb = journal->j_superblock; | |
1427 | ||
1428 | read_lock(&journal->j_state_lock); | |
1429 | jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", | |
1430 | journal->j_errno); | |
1431 | sb->s_errno = cpu_to_be32(journal->j_errno); | |
4fd5ea43 | 1432 | jbd2_superblock_csum_set(journal, sb); |
24bcc89c JK |
1433 | read_unlock(&journal->j_state_lock); |
1434 | ||
79feb521 | 1435 | jbd2_write_superblock(journal, WRITE_SYNC); |
24bcc89c | 1436 | } |
d796c52e | 1437 | EXPORT_SYMBOL(jbd2_journal_update_sb_errno); |
24bcc89c | 1438 | |
470decc6 DK |
1439 | /* |
1440 | * Read the superblock for a given journal, performing initial | |
1441 | * validation of the format. | |
1442 | */ | |
470decc6 DK |
1443 | static int journal_get_superblock(journal_t *journal) |
1444 | { | |
1445 | struct buffer_head *bh; | |
1446 | journal_superblock_t *sb; | |
1447 | int err = -EIO; | |
1448 | ||
1449 | bh = journal->j_sb_buffer; | |
1450 | ||
1451 | J_ASSERT(bh != NULL); | |
1452 | if (!buffer_uptodate(bh)) { | |
1453 | ll_rw_block(READ, 1, &bh); | |
1454 | wait_on_buffer(bh); | |
1455 | if (!buffer_uptodate(bh)) { | |
f2a44523 EG |
1456 | printk(KERN_ERR |
1457 | "JBD2: IO error reading journal superblock\n"); | |
470decc6 DK |
1458 | goto out; |
1459 | } | |
1460 | } | |
1461 | ||
25ed6e8a DW |
1462 | if (buffer_verified(bh)) |
1463 | return 0; | |
1464 | ||
470decc6 DK |
1465 | sb = journal->j_superblock; |
1466 | ||
1467 | err = -EINVAL; | |
1468 | ||
f7f4bccb | 1469 | if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || |
470decc6 | 1470 | sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { |
f2a44523 | 1471 | printk(KERN_WARNING "JBD2: no valid journal superblock found\n"); |
470decc6 DK |
1472 | goto out; |
1473 | } | |
1474 | ||
1475 | switch(be32_to_cpu(sb->s_header.h_blocktype)) { | |
f7f4bccb | 1476 | case JBD2_SUPERBLOCK_V1: |
470decc6 DK |
1477 | journal->j_format_version = 1; |
1478 | break; | |
f7f4bccb | 1479 | case JBD2_SUPERBLOCK_V2: |
470decc6 DK |
1480 | journal->j_format_version = 2; |
1481 | break; | |
1482 | default: | |
f2a44523 | 1483 | printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n"); |
470decc6 DK |
1484 | goto out; |
1485 | } | |
1486 | ||
1487 | if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen) | |
1488 | journal->j_maxlen = be32_to_cpu(sb->s_maxlen); | |
1489 | else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) { | |
f2a44523 | 1490 | printk(KERN_WARNING "JBD2: journal file too short\n"); |
470decc6 DK |
1491 | goto out; |
1492 | } | |
1493 | ||
8762202d EG |
1494 | if (be32_to_cpu(sb->s_first) == 0 || |
1495 | be32_to_cpu(sb->s_first) >= journal->j_maxlen) { | |
1496 | printk(KERN_WARNING | |
1497 | "JBD2: Invalid start block of journal: %u\n", | |
1498 | be32_to_cpu(sb->s_first)); | |
1499 | goto out; | |
1500 | } | |
1501 | ||
25ed6e8a DW |
1502 | if (JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM) && |
1503 | JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) { | |
1504 | /* Can't have checksum v1 and v2 on at the same time! */ | |
1505 | printk(KERN_ERR "JBD: Can't enable checksumming v1 and v2 " | |
1506 | "at the same time!\n"); | |
1507 | goto out; | |
1508 | } | |
1509 | ||
1510 | if (!jbd2_verify_csum_type(journal, sb)) { | |
1511 | printk(KERN_ERR "JBD: Unknown checksum type\n"); | |
1512 | goto out; | |
1513 | } | |
1514 | ||
01b5adce DW |
1515 | /* Load the checksum driver */ |
1516 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) { | |
1517 | journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); | |
1518 | if (IS_ERR(journal->j_chksum_driver)) { | |
1519 | printk(KERN_ERR "JBD: Cannot load crc32c driver.\n"); | |
1520 | err = PTR_ERR(journal->j_chksum_driver); | |
1521 | journal->j_chksum_driver = NULL; | |
1522 | goto out; | |
1523 | } | |
1524 | } | |
1525 | ||
4fd5ea43 DW |
1526 | /* Check superblock checksum */ |
1527 | if (!jbd2_superblock_csum_verify(journal, sb)) { | |
1528 | printk(KERN_ERR "JBD: journal checksum error\n"); | |
1529 | goto out; | |
1530 | } | |
1531 | ||
1532 | /* Precompute checksum seed for all metadata */ | |
1533 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
1534 | journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, | |
1535 | sizeof(sb->s_uuid)); | |
1536 | ||
25ed6e8a DW |
1537 | set_buffer_verified(bh); |
1538 | ||
470decc6 DK |
1539 | return 0; |
1540 | ||
1541 | out: | |
1542 | journal_fail_superblock(journal); | |
1543 | return err; | |
1544 | } | |
1545 | ||
1546 | /* | |
1547 | * Load the on-disk journal superblock and read the key fields into the | |
1548 | * journal_t. | |
1549 | */ | |
1550 | ||
1551 | static int load_superblock(journal_t *journal) | |
1552 | { | |
1553 | int err; | |
1554 | journal_superblock_t *sb; | |
1555 | ||
1556 | err = journal_get_superblock(journal); | |
1557 | if (err) | |
1558 | return err; | |
1559 | ||
1560 | sb = journal->j_superblock; | |
1561 | ||
1562 | journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); | |
1563 | journal->j_tail = be32_to_cpu(sb->s_start); | |
1564 | journal->j_first = be32_to_cpu(sb->s_first); | |
1565 | journal->j_last = be32_to_cpu(sb->s_maxlen); | |
1566 | journal->j_errno = be32_to_cpu(sb->s_errno); | |
1567 | ||
1568 | return 0; | |
1569 | } | |
1570 | ||
1571 | ||
1572 | /** | |
f7f4bccb | 1573 | * int jbd2_journal_load() - Read journal from disk. |
470decc6 DK |
1574 | * @journal: Journal to act on. |
1575 | * | |
1576 | * Given a journal_t structure which tells us which disk blocks contain | |
1577 | * a journal, read the journal from disk to initialise the in-memory | |
1578 | * structures. | |
1579 | */ | |
f7f4bccb | 1580 | int jbd2_journal_load(journal_t *journal) |
470decc6 DK |
1581 | { |
1582 | int err; | |
1583 | journal_superblock_t *sb; | |
1584 | ||
1585 | err = load_superblock(journal); | |
1586 | if (err) | |
1587 | return err; | |
1588 | ||
1589 | sb = journal->j_superblock; | |
1590 | /* If this is a V2 superblock, then we have to check the | |
1591 | * features flags on it. */ | |
1592 | ||
1593 | if (journal->j_format_version >= 2) { | |
1594 | if ((sb->s_feature_ro_compat & | |
f7f4bccb | 1595 | ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || |
470decc6 | 1596 | (sb->s_feature_incompat & |
f7f4bccb | 1597 | ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { |
f2a44523 EG |
1598 | printk(KERN_WARNING |
1599 | "JBD2: Unrecognised features on journal\n"); | |
470decc6 DK |
1600 | return -EINVAL; |
1601 | } | |
1602 | } | |
1603 | ||
d2eecb03 TT |
1604 | /* |
1605 | * Create a slab for this blocksize | |
1606 | */ | |
1607 | err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize)); | |
1608 | if (err) | |
1609 | return err; | |
1610 | ||
470decc6 DK |
1611 | /* Let the recovery code check whether it needs to recover any |
1612 | * data from the journal. */ | |
f7f4bccb | 1613 | if (jbd2_journal_recover(journal)) |
470decc6 DK |
1614 | goto recovery_error; |
1615 | ||
e6a47428 TT |
1616 | if (journal->j_failed_commit) { |
1617 | printk(KERN_ERR "JBD2: journal transaction %u on %s " | |
1618 | "is corrupt.\n", journal->j_failed_commit, | |
1619 | journal->j_devname); | |
1620 | return -EIO; | |
1621 | } | |
1622 | ||
470decc6 DK |
1623 | /* OK, we've finished with the dynamic journal bits: |
1624 | * reinitialise the dynamic contents of the superblock in memory | |
1625 | * and reset them on disk. */ | |
1626 | if (journal_reset(journal)) | |
1627 | goto recovery_error; | |
1628 | ||
f7f4bccb MC |
1629 | journal->j_flags &= ~JBD2_ABORT; |
1630 | journal->j_flags |= JBD2_LOADED; | |
470decc6 DK |
1631 | return 0; |
1632 | ||
1633 | recovery_error: | |
f2a44523 | 1634 | printk(KERN_WARNING "JBD2: recovery failed\n"); |
470decc6 DK |
1635 | return -EIO; |
1636 | } | |
1637 | ||
1638 | /** | |
f7f4bccb | 1639 | * void jbd2_journal_destroy() - Release a journal_t structure. |
470decc6 DK |
1640 | * @journal: Journal to act on. |
1641 | * | |
1642 | * Release a journal_t structure once it is no longer in use by the | |
1643 | * journaled object. | |
44519faf | 1644 | * Return <0 if we couldn't clean up the journal. |
470decc6 | 1645 | */ |
44519faf | 1646 | int jbd2_journal_destroy(journal_t *journal) |
470decc6 | 1647 | { |
44519faf HK |
1648 | int err = 0; |
1649 | ||
470decc6 DK |
1650 | /* Wait for the commit thread to wake up and die. */ |
1651 | journal_kill_thread(journal); | |
1652 | ||
1653 | /* Force a final log commit */ | |
1654 | if (journal->j_running_transaction) | |
f7f4bccb | 1655 | jbd2_journal_commit_transaction(journal); |
470decc6 DK |
1656 | |
1657 | /* Force any old transactions to disk */ | |
1658 | ||
1659 | /* Totally anal locking here... */ | |
1660 | spin_lock(&journal->j_list_lock); | |
1661 | while (journal->j_checkpoint_transactions != NULL) { | |
1662 | spin_unlock(&journal->j_list_lock); | |
1a0d3786 | 1663 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1664 | jbd2_log_do_checkpoint(journal); |
1a0d3786 | 1665 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
1666 | spin_lock(&journal->j_list_lock); |
1667 | } | |
1668 | ||
1669 | J_ASSERT(journal->j_running_transaction == NULL); | |
1670 | J_ASSERT(journal->j_committing_transaction == NULL); | |
1671 | J_ASSERT(journal->j_checkpoint_transactions == NULL); | |
1672 | spin_unlock(&journal->j_list_lock); | |
1673 | ||
470decc6 | 1674 | if (journal->j_sb_buffer) { |
44519faf | 1675 | if (!is_journal_aborted(journal)) { |
a78bb11d | 1676 | mutex_lock(&journal->j_checkpoint_mutex); |
24bcc89c | 1677 | jbd2_mark_journal_empty(journal); |
a78bb11d JK |
1678 | mutex_unlock(&journal->j_checkpoint_mutex); |
1679 | } else | |
44519faf | 1680 | err = -EIO; |
470decc6 DK |
1681 | brelse(journal->j_sb_buffer); |
1682 | } | |
1683 | ||
8e85fb3f JL |
1684 | if (journal->j_proc_entry) |
1685 | jbd2_stats_proc_exit(journal); | |
470decc6 DK |
1686 | if (journal->j_inode) |
1687 | iput(journal->j_inode); | |
1688 | if (journal->j_revoke) | |
f7f4bccb | 1689 | jbd2_journal_destroy_revoke(journal); |
01b5adce DW |
1690 | if (journal->j_chksum_driver) |
1691 | crypto_free_shash(journal->j_chksum_driver); | |
470decc6 DK |
1692 | kfree(journal->j_wbuf); |
1693 | kfree(journal); | |
44519faf HK |
1694 | |
1695 | return err; | |
470decc6 DK |
1696 | } |
1697 | ||
1698 | ||
1699 | /** | |
f7f4bccb | 1700 | *int jbd2_journal_check_used_features () - Check if features specified are used. |
470decc6 DK |
1701 | * @journal: Journal to check. |
1702 | * @compat: bitmask of compatible features | |
1703 | * @ro: bitmask of features that force read-only mount | |
1704 | * @incompat: bitmask of incompatible features | |
1705 | * | |
1706 | * Check whether the journal uses all of a given set of | |
1707 | * features. Return true (non-zero) if it does. | |
1708 | **/ | |
1709 | ||
f7f4bccb | 1710 | int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1711 | unsigned long ro, unsigned long incompat) |
1712 | { | |
1713 | journal_superblock_t *sb; | |
1714 | ||
1715 | if (!compat && !ro && !incompat) | |
1716 | return 1; | |
1113e1b5 PL |
1717 | /* Load journal superblock if it is not loaded yet. */ |
1718 | if (journal->j_format_version == 0 && | |
1719 | journal_get_superblock(journal) != 0) | |
1720 | return 0; | |
470decc6 DK |
1721 | if (journal->j_format_version == 1) |
1722 | return 0; | |
1723 | ||
1724 | sb = journal->j_superblock; | |
1725 | ||
1726 | if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && | |
1727 | ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && | |
1728 | ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) | |
1729 | return 1; | |
1730 | ||
1731 | return 0; | |
1732 | } | |
1733 | ||
1734 | /** | |
f7f4bccb | 1735 | * int jbd2_journal_check_available_features() - Check feature set in journalling layer |
470decc6 DK |
1736 | * @journal: Journal to check. |
1737 | * @compat: bitmask of compatible features | |
1738 | * @ro: bitmask of features that force read-only mount | |
1739 | * @incompat: bitmask of incompatible features | |
1740 | * | |
1741 | * Check whether the journaling code supports the use of | |
1742 | * all of a given set of features on this journal. Return true | |
1743 | * (non-zero) if it can. */ | |
1744 | ||
f7f4bccb | 1745 | int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1746 | unsigned long ro, unsigned long incompat) |
1747 | { | |
470decc6 DK |
1748 | if (!compat && !ro && !incompat) |
1749 | return 1; | |
1750 | ||
470decc6 DK |
1751 | /* We can support any known requested features iff the |
1752 | * superblock is in version 2. Otherwise we fail to support any | |
1753 | * extended sb features. */ | |
1754 | ||
1755 | if (journal->j_format_version != 2) | |
1756 | return 0; | |
1757 | ||
f7f4bccb MC |
1758 | if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && |
1759 | (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && | |
1760 | (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) | |
470decc6 DK |
1761 | return 1; |
1762 | ||
1763 | return 0; | |
1764 | } | |
1765 | ||
1766 | /** | |
f7f4bccb | 1767 | * int jbd2_journal_set_features () - Mark a given journal feature in the superblock |
470decc6 DK |
1768 | * @journal: Journal to act on. |
1769 | * @compat: bitmask of compatible features | |
1770 | * @ro: bitmask of features that force read-only mount | |
1771 | * @incompat: bitmask of incompatible features | |
1772 | * | |
1773 | * Mark a given journal feature as present on the | |
1774 | * superblock. Returns true if the requested features could be set. | |
1775 | * | |
1776 | */ | |
1777 | ||
f7f4bccb | 1778 | int jbd2_journal_set_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1779 | unsigned long ro, unsigned long incompat) |
1780 | { | |
25ed6e8a DW |
1781 | #define INCOMPAT_FEATURE_ON(f) \ |
1782 | ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f))) | |
1783 | #define COMPAT_FEATURE_ON(f) \ | |
1784 | ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f))) | |
470decc6 DK |
1785 | journal_superblock_t *sb; |
1786 | ||
f7f4bccb | 1787 | if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) |
470decc6 DK |
1788 | return 1; |
1789 | ||
f7f4bccb | 1790 | if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) |
470decc6 DK |
1791 | return 0; |
1792 | ||
25ed6e8a DW |
1793 | /* Asking for checksumming v2 and v1? Only give them v2. */ |
1794 | if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2 && | |
1795 | compat & JBD2_FEATURE_COMPAT_CHECKSUM) | |
1796 | compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM; | |
1797 | ||
470decc6 DK |
1798 | jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", |
1799 | compat, ro, incompat); | |
1800 | ||
1801 | sb = journal->j_superblock; | |
1802 | ||
25ed6e8a DW |
1803 | /* If enabling v2 checksums, update superblock */ |
1804 | if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V2)) { | |
1805 | sb->s_checksum_type = JBD2_CRC32C_CHKSUM; | |
1806 | sb->s_feature_compat &= | |
1807 | ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM); | |
01b5adce DW |
1808 | |
1809 | /* Load the checksum driver */ | |
1810 | if (journal->j_chksum_driver == NULL) { | |
1811 | journal->j_chksum_driver = crypto_alloc_shash("crc32c", | |
1812 | 0, 0); | |
1813 | if (IS_ERR(journal->j_chksum_driver)) { | |
1814 | printk(KERN_ERR "JBD: Cannot load crc32c " | |
1815 | "driver.\n"); | |
1816 | journal->j_chksum_driver = NULL; | |
1817 | return 0; | |
1818 | } | |
1819 | } | |
4fd5ea43 DW |
1820 | |
1821 | /* Precompute checksum seed for all metadata */ | |
1822 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, | |
1823 | JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
1824 | journal->j_csum_seed = jbd2_chksum(journal, ~0, | |
1825 | sb->s_uuid, | |
1826 | sizeof(sb->s_uuid)); | |
25ed6e8a DW |
1827 | } |
1828 | ||
1829 | /* If enabling v1 checksums, downgrade superblock */ | |
1830 | if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM)) | |
1831 | sb->s_feature_incompat &= | |
1832 | ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2); | |
1833 | ||
470decc6 DK |
1834 | sb->s_feature_compat |= cpu_to_be32(compat); |
1835 | sb->s_feature_ro_compat |= cpu_to_be32(ro); | |
1836 | sb->s_feature_incompat |= cpu_to_be32(incompat); | |
1837 | ||
1838 | return 1; | |
25ed6e8a DW |
1839 | #undef COMPAT_FEATURE_ON |
1840 | #undef INCOMPAT_FEATURE_ON | |
470decc6 DK |
1841 | } |
1842 | ||
818d276c GS |
1843 | /* |
1844 | * jbd2_journal_clear_features () - Clear a given journal feature in the | |
1845 | * superblock | |
1846 | * @journal: Journal to act on. | |
1847 | * @compat: bitmask of compatible features | |
1848 | * @ro: bitmask of features that force read-only mount | |
1849 | * @incompat: bitmask of incompatible features | |
1850 | * | |
1851 | * Clear a given journal feature as present on the | |
1852 | * superblock. | |
1853 | */ | |
1854 | void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, | |
1855 | unsigned long ro, unsigned long incompat) | |
1856 | { | |
1857 | journal_superblock_t *sb; | |
1858 | ||
1859 | jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", | |
1860 | compat, ro, incompat); | |
1861 | ||
1862 | sb = journal->j_superblock; | |
1863 | ||
1864 | sb->s_feature_compat &= ~cpu_to_be32(compat); | |
1865 | sb->s_feature_ro_compat &= ~cpu_to_be32(ro); | |
1866 | sb->s_feature_incompat &= ~cpu_to_be32(incompat); | |
1867 | } | |
1868 | EXPORT_SYMBOL(jbd2_journal_clear_features); | |
470decc6 | 1869 | |
470decc6 | 1870 | /** |
f7f4bccb | 1871 | * int jbd2_journal_flush () - Flush journal |
470decc6 DK |
1872 | * @journal: Journal to act on. |
1873 | * | |
1874 | * Flush all data for a given journal to disk and empty the journal. | |
1875 | * Filesystems can use this when remounting readonly to ensure that | |
1876 | * recovery does not need to happen on remount. | |
1877 | */ | |
1878 | ||
f7f4bccb | 1879 | int jbd2_journal_flush(journal_t *journal) |
470decc6 DK |
1880 | { |
1881 | int err = 0; | |
1882 | transaction_t *transaction = NULL; | |
470decc6 | 1883 | |
a931da6a | 1884 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1885 | |
1886 | /* Force everything buffered to the log... */ | |
1887 | if (journal->j_running_transaction) { | |
1888 | transaction = journal->j_running_transaction; | |
f7f4bccb | 1889 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1890 | } else if (journal->j_committing_transaction) |
1891 | transaction = journal->j_committing_transaction; | |
1892 | ||
1893 | /* Wait for the log commit to complete... */ | |
1894 | if (transaction) { | |
1895 | tid_t tid = transaction->t_tid; | |
1896 | ||
a931da6a | 1897 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 1898 | jbd2_log_wait_commit(journal, tid); |
470decc6 | 1899 | } else { |
a931da6a | 1900 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1901 | } |
1902 | ||
1903 | /* ...and flush everything in the log out to disk. */ | |
1904 | spin_lock(&journal->j_list_lock); | |
1905 | while (!err && journal->j_checkpoint_transactions != NULL) { | |
1906 | spin_unlock(&journal->j_list_lock); | |
44519faf | 1907 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1908 | err = jbd2_log_do_checkpoint(journal); |
44519faf | 1909 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
1910 | spin_lock(&journal->j_list_lock); |
1911 | } | |
1912 | spin_unlock(&journal->j_list_lock); | |
44519faf HK |
1913 | |
1914 | if (is_journal_aborted(journal)) | |
1915 | return -EIO; | |
1916 | ||
a78bb11d | 1917 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1918 | jbd2_cleanup_journal_tail(journal); |
470decc6 DK |
1919 | |
1920 | /* Finally, mark the journal as really needing no recovery. | |
1921 | * This sets s_start==0 in the underlying superblock, which is | |
1922 | * the magic code for a fully-recovered superblock. Any future | |
1923 | * commits of data to the journal will restore the current | |
1924 | * s_start value. */ | |
24bcc89c | 1925 | jbd2_mark_journal_empty(journal); |
a78bb11d | 1926 | mutex_unlock(&journal->j_checkpoint_mutex); |
a931da6a | 1927 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1928 | J_ASSERT(!journal->j_running_transaction); |
1929 | J_ASSERT(!journal->j_committing_transaction); | |
1930 | J_ASSERT(!journal->j_checkpoint_transactions); | |
1931 | J_ASSERT(journal->j_head == journal->j_tail); | |
1932 | J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); | |
a931da6a | 1933 | write_unlock(&journal->j_state_lock); |
44519faf | 1934 | return 0; |
470decc6 DK |
1935 | } |
1936 | ||
1937 | /** | |
f7f4bccb | 1938 | * int jbd2_journal_wipe() - Wipe journal contents |
470decc6 DK |
1939 | * @journal: Journal to act on. |
1940 | * @write: flag (see below) | |
1941 | * | |
1942 | * Wipe out all of the contents of a journal, safely. This will produce | |
1943 | * a warning if the journal contains any valid recovery information. | |
f7f4bccb | 1944 | * Must be called between journal_init_*() and jbd2_journal_load(). |
470decc6 DK |
1945 | * |
1946 | * If 'write' is non-zero, then we wipe out the journal on disk; otherwise | |
1947 | * we merely suppress recovery. | |
1948 | */ | |
1949 | ||
f7f4bccb | 1950 | int jbd2_journal_wipe(journal_t *journal, int write) |
470decc6 | 1951 | { |
470decc6 DK |
1952 | int err = 0; |
1953 | ||
f7f4bccb | 1954 | J_ASSERT (!(journal->j_flags & JBD2_LOADED)); |
470decc6 DK |
1955 | |
1956 | err = load_superblock(journal); | |
1957 | if (err) | |
1958 | return err; | |
1959 | ||
470decc6 DK |
1960 | if (!journal->j_tail) |
1961 | goto no_recovery; | |
1962 | ||
f2a44523 | 1963 | printk(KERN_WARNING "JBD2: %s recovery information on journal\n", |
470decc6 DK |
1964 | write ? "Clearing" : "Ignoring"); |
1965 | ||
f7f4bccb | 1966 | err = jbd2_journal_skip_recovery(journal); |
a78bb11d JK |
1967 | if (write) { |
1968 | /* Lock to make assertions happy... */ | |
1969 | mutex_lock(&journal->j_checkpoint_mutex); | |
24bcc89c | 1970 | jbd2_mark_journal_empty(journal); |
a78bb11d JK |
1971 | mutex_unlock(&journal->j_checkpoint_mutex); |
1972 | } | |
470decc6 DK |
1973 | |
1974 | no_recovery: | |
1975 | return err; | |
1976 | } | |
1977 | ||
470decc6 DK |
1978 | /* |
1979 | * Journal abort has very specific semantics, which we describe | |
1980 | * for journal abort. | |
1981 | * | |
bfcd3555 | 1982 | * Two internal functions, which provide abort to the jbd layer |
470decc6 DK |
1983 | * itself are here. |
1984 | */ | |
1985 | ||
1986 | /* | |
1987 | * Quick version for internal journal use (doesn't lock the journal). | |
1988 | * Aborts hard --- we mark the abort as occurred, but do _nothing_ else, | |
1989 | * and don't attempt to make any other journal updates. | |
1990 | */ | |
f7f4bccb | 1991 | void __jbd2_journal_abort_hard(journal_t *journal) |
470decc6 DK |
1992 | { |
1993 | transaction_t *transaction; | |
470decc6 | 1994 | |
f7f4bccb | 1995 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1996 | return; |
1997 | ||
1998 | printk(KERN_ERR "Aborting journal on device %s.\n", | |
05496769 | 1999 | journal->j_devname); |
470decc6 | 2000 | |
a931da6a | 2001 | write_lock(&journal->j_state_lock); |
f7f4bccb | 2002 | journal->j_flags |= JBD2_ABORT; |
470decc6 DK |
2003 | transaction = journal->j_running_transaction; |
2004 | if (transaction) | |
f7f4bccb | 2005 | __jbd2_log_start_commit(journal, transaction->t_tid); |
a931da6a | 2006 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2007 | } |
2008 | ||
2009 | /* Soft abort: record the abort error status in the journal superblock, | |
2010 | * but don't do any other IO. */ | |
2011 | static void __journal_abort_soft (journal_t *journal, int errno) | |
2012 | { | |
f7f4bccb | 2013 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2014 | return; |
2015 | ||
2016 | if (!journal->j_errno) | |
2017 | journal->j_errno = errno; | |
2018 | ||
f7f4bccb | 2019 | __jbd2_journal_abort_hard(journal); |
470decc6 DK |
2020 | |
2021 | if (errno) | |
24bcc89c | 2022 | jbd2_journal_update_sb_errno(journal); |
470decc6 DK |
2023 | } |
2024 | ||
2025 | /** | |
f7f4bccb | 2026 | * void jbd2_journal_abort () - Shutdown the journal immediately. |
470decc6 DK |
2027 | * @journal: the journal to shutdown. |
2028 | * @errno: an error number to record in the journal indicating | |
2029 | * the reason for the shutdown. | |
2030 | * | |
2031 | * Perform a complete, immediate shutdown of the ENTIRE | |
2032 | * journal (not of a single transaction). This operation cannot be | |
2033 | * undone without closing and reopening the journal. | |
2034 | * | |
f7f4bccb | 2035 | * The jbd2_journal_abort function is intended to support higher level error |
470decc6 DK |
2036 | * recovery mechanisms such as the ext2/ext3 remount-readonly error |
2037 | * mode. | |
2038 | * | |
2039 | * Journal abort has very specific semantics. Any existing dirty, | |
2040 | * unjournaled buffers in the main filesystem will still be written to | |
2041 | * disk by bdflush, but the journaling mechanism will be suspended | |
2042 | * immediately and no further transaction commits will be honoured. | |
2043 | * | |
2044 | * Any dirty, journaled buffers will be written back to disk without | |
2045 | * hitting the journal. Atomicity cannot be guaranteed on an aborted | |
2046 | * filesystem, but we _do_ attempt to leave as much data as possible | |
2047 | * behind for fsck to use for cleanup. | |
2048 | * | |
2049 | * Any attempt to get a new transaction handle on a journal which is in | |
2050 | * ABORT state will just result in an -EROFS error return. A | |
f7f4bccb | 2051 | * jbd2_journal_stop on an existing handle will return -EIO if we have |
470decc6 DK |
2052 | * entered abort state during the update. |
2053 | * | |
2054 | * Recursive transactions are not disturbed by journal abort until the | |
f7f4bccb | 2055 | * final jbd2_journal_stop, which will receive the -EIO error. |
470decc6 | 2056 | * |
f7f4bccb | 2057 | * Finally, the jbd2_journal_abort call allows the caller to supply an errno |
470decc6 DK |
2058 | * which will be recorded (if possible) in the journal superblock. This |
2059 | * allows a client to record failure conditions in the middle of a | |
2060 | * transaction without having to complete the transaction to record the | |
2061 | * failure to disk. ext3_error, for example, now uses this | |
2062 | * functionality. | |
2063 | * | |
2064 | * Errors which originate from within the journaling layer will NOT | |
2065 | * supply an errno; a null errno implies that absolutely no further | |
2066 | * writes are done to the journal (unless there are any already in | |
2067 | * progress). | |
2068 | * | |
2069 | */ | |
2070 | ||
f7f4bccb | 2071 | void jbd2_journal_abort(journal_t *journal, int errno) |
470decc6 DK |
2072 | { |
2073 | __journal_abort_soft(journal, errno); | |
2074 | } | |
2075 | ||
2076 | /** | |
f7f4bccb | 2077 | * int jbd2_journal_errno () - returns the journal's error state. |
470decc6 DK |
2078 | * @journal: journal to examine. |
2079 | * | |
bfcd3555 | 2080 | * This is the errno number set with jbd2_journal_abort(), the last |
470decc6 DK |
2081 | * time the journal was mounted - if the journal was stopped |
2082 | * without calling abort this will be 0. | |
2083 | * | |
2084 | * If the journal has been aborted on this mount time -EROFS will | |
2085 | * be returned. | |
2086 | */ | |
f7f4bccb | 2087 | int jbd2_journal_errno(journal_t *journal) |
470decc6 DK |
2088 | { |
2089 | int err; | |
2090 | ||
a931da6a | 2091 | read_lock(&journal->j_state_lock); |
f7f4bccb | 2092 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2093 | err = -EROFS; |
2094 | else | |
2095 | err = journal->j_errno; | |
a931da6a | 2096 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
2097 | return err; |
2098 | } | |
2099 | ||
2100 | /** | |
f7f4bccb | 2101 | * int jbd2_journal_clear_err () - clears the journal's error state |
470decc6 DK |
2102 | * @journal: journal to act on. |
2103 | * | |
bfcd3555 | 2104 | * An error must be cleared or acked to take a FS out of readonly |
470decc6 DK |
2105 | * mode. |
2106 | */ | |
f7f4bccb | 2107 | int jbd2_journal_clear_err(journal_t *journal) |
470decc6 DK |
2108 | { |
2109 | int err = 0; | |
2110 | ||
a931da6a | 2111 | write_lock(&journal->j_state_lock); |
f7f4bccb | 2112 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2113 | err = -EROFS; |
2114 | else | |
2115 | journal->j_errno = 0; | |
a931da6a | 2116 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2117 | return err; |
2118 | } | |
2119 | ||
2120 | /** | |
f7f4bccb | 2121 | * void jbd2_journal_ack_err() - Ack journal err. |
470decc6 DK |
2122 | * @journal: journal to act on. |
2123 | * | |
bfcd3555 | 2124 | * An error must be cleared or acked to take a FS out of readonly |
470decc6 DK |
2125 | * mode. |
2126 | */ | |
f7f4bccb | 2127 | void jbd2_journal_ack_err(journal_t *journal) |
470decc6 | 2128 | { |
a931da6a | 2129 | write_lock(&journal->j_state_lock); |
470decc6 | 2130 | if (journal->j_errno) |
f7f4bccb | 2131 | journal->j_flags |= JBD2_ACK_ERR; |
a931da6a | 2132 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2133 | } |
2134 | ||
f7f4bccb | 2135 | int jbd2_journal_blocks_per_page(struct inode *inode) |
470decc6 DK |
2136 | { |
2137 | return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
2138 | } | |
2139 | ||
b517bea1 ZB |
2140 | /* |
2141 | * helper functions to deal with 32 or 64bit block numbers. | |
2142 | */ | |
2143 | size_t journal_tag_bytes(journal_t *journal) | |
2144 | { | |
c3900875 DW |
2145 | journal_block_tag_t tag; |
2146 | size_t x = 0; | |
2147 | ||
2148 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
2149 | x += sizeof(tag.t_checksum); | |
2150 | ||
b517bea1 | 2151 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) |
c3900875 | 2152 | return x + JBD2_TAG_SIZE64; |
b517bea1 | 2153 | else |
c3900875 | 2154 | return x + JBD2_TAG_SIZE32; |
b517bea1 ZB |
2155 | } |
2156 | ||
d2eecb03 TT |
2157 | /* |
2158 | * JBD memory management | |
2159 | * | |
2160 | * These functions are used to allocate block-sized chunks of memory | |
2161 | * used for making copies of buffer_head data. Very often it will be | |
2162 | * page-sized chunks of data, but sometimes it will be in | |
2163 | * sub-page-size chunks. (For example, 16k pages on Power systems | |
2164 | * with a 4k block file system.) For blocks smaller than a page, we | |
2165 | * use a SLAB allocator. There are slab caches for each block size, | |
2166 | * which are allocated at mount time, if necessary, and we only free | |
2167 | * (all of) the slab caches when/if the jbd2 module is unloaded. For | |
2168 | * this reason we don't need to a mutex to protect access to | |
2169 | * jbd2_slab[] allocating or releasing memory; only in | |
2170 | * jbd2_journal_create_slab(). | |
2171 | */ | |
2172 | #define JBD2_MAX_SLABS 8 | |
2173 | static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS]; | |
d2eecb03 TT |
2174 | |
2175 | static const char *jbd2_slab_names[JBD2_MAX_SLABS] = { | |
2176 | "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k", | |
2177 | "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k" | |
2178 | }; | |
2179 | ||
2180 | ||
2181 | static void jbd2_journal_destroy_slabs(void) | |
2182 | { | |
2183 | int i; | |
2184 | ||
2185 | for (i = 0; i < JBD2_MAX_SLABS; i++) { | |
2186 | if (jbd2_slab[i]) | |
2187 | kmem_cache_destroy(jbd2_slab[i]); | |
2188 | jbd2_slab[i] = NULL; | |
2189 | } | |
2190 | } | |
2191 | ||
2192 | static int jbd2_journal_create_slab(size_t size) | |
2193 | { | |
51dfacde | 2194 | static DEFINE_MUTEX(jbd2_slab_create_mutex); |
d2eecb03 TT |
2195 | int i = order_base_2(size) - 10; |
2196 | size_t slab_size; | |
2197 | ||
2198 | if (size == PAGE_SIZE) | |
2199 | return 0; | |
2200 | ||
2201 | if (i >= JBD2_MAX_SLABS) | |
2202 | return -EINVAL; | |
2203 | ||
2204 | if (unlikely(i < 0)) | |
2205 | i = 0; | |
51dfacde | 2206 | mutex_lock(&jbd2_slab_create_mutex); |
d2eecb03 | 2207 | if (jbd2_slab[i]) { |
51dfacde | 2208 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2209 | return 0; /* Already created */ |
2210 | } | |
2211 | ||
2212 | slab_size = 1 << (i+10); | |
2213 | jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size, | |
2214 | slab_size, 0, NULL); | |
51dfacde | 2215 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2216 | if (!jbd2_slab[i]) { |
2217 | printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n"); | |
2218 | return -ENOMEM; | |
2219 | } | |
2220 | return 0; | |
2221 | } | |
2222 | ||
2223 | static struct kmem_cache *get_slab(size_t size) | |
2224 | { | |
2225 | int i = order_base_2(size) - 10; | |
2226 | ||
2227 | BUG_ON(i >= JBD2_MAX_SLABS); | |
2228 | if (unlikely(i < 0)) | |
2229 | i = 0; | |
8ac97b74 | 2230 | BUG_ON(jbd2_slab[i] == NULL); |
d2eecb03 TT |
2231 | return jbd2_slab[i]; |
2232 | } | |
2233 | ||
2234 | void *jbd2_alloc(size_t size, gfp_t flags) | |
2235 | { | |
2236 | void *ptr; | |
2237 | ||
2238 | BUG_ON(size & (size-1)); /* Must be a power of 2 */ | |
2239 | ||
2240 | flags |= __GFP_REPEAT; | |
2241 | if (size == PAGE_SIZE) | |
2242 | ptr = (void *)__get_free_pages(flags, 0); | |
2243 | else if (size > PAGE_SIZE) { | |
2244 | int order = get_order(size); | |
2245 | ||
2246 | if (order < 3) | |
2247 | ptr = (void *)__get_free_pages(flags, order); | |
2248 | else | |
2249 | ptr = vmalloc(size); | |
2250 | } else | |
2251 | ptr = kmem_cache_alloc(get_slab(size), flags); | |
2252 | ||
2253 | /* Check alignment; SLUB has gotten this wrong in the past, | |
2254 | * and this can lead to user data corruption! */ | |
2255 | BUG_ON(((unsigned long) ptr) & (size-1)); | |
2256 | ||
2257 | return ptr; | |
2258 | } | |
2259 | ||
2260 | void jbd2_free(void *ptr, size_t size) | |
2261 | { | |
2262 | if (size == PAGE_SIZE) { | |
2263 | free_pages((unsigned long)ptr, 0); | |
2264 | return; | |
2265 | } | |
2266 | if (size > PAGE_SIZE) { | |
2267 | int order = get_order(size); | |
2268 | ||
2269 | if (order < 3) | |
2270 | free_pages((unsigned long)ptr, order); | |
2271 | else | |
2272 | vfree(ptr); | |
2273 | return; | |
2274 | } | |
2275 | kmem_cache_free(get_slab(size), ptr); | |
2276 | }; | |
2277 | ||
470decc6 DK |
2278 | /* |
2279 | * Journal_head storage management | |
2280 | */ | |
e18b890b | 2281 | static struct kmem_cache *jbd2_journal_head_cache; |
e23291b9 | 2282 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2283 | static atomic_t nr_journal_heads = ATOMIC_INIT(0); |
2284 | #endif | |
2285 | ||
4185a2ac | 2286 | static int jbd2_journal_init_journal_head_cache(void) |
470decc6 DK |
2287 | { |
2288 | int retval; | |
2289 | ||
1076d17a | 2290 | J_ASSERT(jbd2_journal_head_cache == NULL); |
a920e941 | 2291 | jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", |
470decc6 DK |
2292 | sizeof(struct journal_head), |
2293 | 0, /* offset */ | |
77160957 | 2294 | SLAB_TEMPORARY, /* flags */ |
20c2df83 | 2295 | NULL); /* ctor */ |
470decc6 | 2296 | retval = 0; |
1076d17a | 2297 | if (!jbd2_journal_head_cache) { |
470decc6 | 2298 | retval = -ENOMEM; |
f2a44523 | 2299 | printk(KERN_EMERG "JBD2: no memory for journal_head cache\n"); |
470decc6 DK |
2300 | } |
2301 | return retval; | |
2302 | } | |
2303 | ||
4185a2ac | 2304 | static void jbd2_journal_destroy_journal_head_cache(void) |
470decc6 | 2305 | { |
8a9362eb DG |
2306 | if (jbd2_journal_head_cache) { |
2307 | kmem_cache_destroy(jbd2_journal_head_cache); | |
2308 | jbd2_journal_head_cache = NULL; | |
2309 | } | |
470decc6 DK |
2310 | } |
2311 | ||
2312 | /* | |
2313 | * journal_head splicing and dicing | |
2314 | */ | |
2315 | static struct journal_head *journal_alloc_journal_head(void) | |
2316 | { | |
2317 | struct journal_head *ret; | |
470decc6 | 2318 | |
e23291b9 | 2319 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2320 | atomic_inc(&nr_journal_heads); |
2321 | #endif | |
5d9cf9c6 | 2322 | ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); |
1076d17a | 2323 | if (!ret) { |
470decc6 | 2324 | jbd_debug(1, "out of memory for journal_head\n"); |
670be5a7 | 2325 | pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__); |
1076d17a | 2326 | while (!ret) { |
470decc6 | 2327 | yield(); |
5d9cf9c6 | 2328 | ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); |
470decc6 DK |
2329 | } |
2330 | } | |
2331 | return ret; | |
2332 | } | |
2333 | ||
2334 | static void journal_free_journal_head(struct journal_head *jh) | |
2335 | { | |
e23291b9 | 2336 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 | 2337 | atomic_dec(&nr_journal_heads); |
cd02ff0b | 2338 | memset(jh, JBD2_POISON_FREE, sizeof(*jh)); |
470decc6 | 2339 | #endif |
f7f4bccb | 2340 | kmem_cache_free(jbd2_journal_head_cache, jh); |
470decc6 DK |
2341 | } |
2342 | ||
2343 | /* | |
2344 | * A journal_head is attached to a buffer_head whenever JBD has an | |
2345 | * interest in the buffer. | |
2346 | * | |
2347 | * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit | |
2348 | * is set. This bit is tested in core kernel code where we need to take | |
2349 | * JBD-specific actions. Testing the zeroness of ->b_private is not reliable | |
2350 | * there. | |
2351 | * | |
2352 | * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. | |
2353 | * | |
2354 | * When a buffer has its BH_JBD bit set it is immune from being released by | |
2355 | * core kernel code, mainly via ->b_count. | |
2356 | * | |
de1b7941 JK |
2357 | * A journal_head is detached from its buffer_head when the journal_head's |
2358 | * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint | |
2359 | * transaction (b_cp_transaction) hold their references to b_jcount. | |
470decc6 DK |
2360 | * |
2361 | * Various places in the kernel want to attach a journal_head to a buffer_head | |
2362 | * _before_ attaching the journal_head to a transaction. To protect the | |
f7f4bccb | 2363 | * journal_head in this situation, jbd2_journal_add_journal_head elevates the |
470decc6 | 2364 | * journal_head's b_jcount refcount by one. The caller must call |
f7f4bccb | 2365 | * jbd2_journal_put_journal_head() to undo this. |
470decc6 DK |
2366 | * |
2367 | * So the typical usage would be: | |
2368 | * | |
2369 | * (Attach a journal_head if needed. Increments b_jcount) | |
f7f4bccb | 2370 | * struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 | 2371 | * ... |
de1b7941 JK |
2372 | * (Get another reference for transaction) |
2373 | * jbd2_journal_grab_journal_head(bh); | |
470decc6 | 2374 | * jh->b_transaction = xxx; |
de1b7941 | 2375 | * (Put original reference) |
f7f4bccb | 2376 | * jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2377 | */ |
2378 | ||
2379 | /* | |
2380 | * Give a buffer_head a journal_head. | |
2381 | * | |
470decc6 DK |
2382 | * May sleep. |
2383 | */ | |
f7f4bccb | 2384 | struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) |
470decc6 DK |
2385 | { |
2386 | struct journal_head *jh; | |
2387 | struct journal_head *new_jh = NULL; | |
2388 | ||
2389 | repeat: | |
5d9cf9c6 | 2390 | if (!buffer_jbd(bh)) |
470decc6 | 2391 | new_jh = journal_alloc_journal_head(); |
470decc6 DK |
2392 | |
2393 | jbd_lock_bh_journal_head(bh); | |
2394 | if (buffer_jbd(bh)) { | |
2395 | jh = bh2jh(bh); | |
2396 | } else { | |
2397 | J_ASSERT_BH(bh, | |
2398 | (atomic_read(&bh->b_count) > 0) || | |
2399 | (bh->b_page && bh->b_page->mapping)); | |
2400 | ||
2401 | if (!new_jh) { | |
2402 | jbd_unlock_bh_journal_head(bh); | |
2403 | goto repeat; | |
2404 | } | |
2405 | ||
2406 | jh = new_jh; | |
2407 | new_jh = NULL; /* We consumed it */ | |
2408 | set_buffer_jbd(bh); | |
2409 | bh->b_private = jh; | |
2410 | jh->b_bh = bh; | |
2411 | get_bh(bh); | |
2412 | BUFFER_TRACE(bh, "added journal_head"); | |
2413 | } | |
2414 | jh->b_jcount++; | |
2415 | jbd_unlock_bh_journal_head(bh); | |
2416 | if (new_jh) | |
2417 | journal_free_journal_head(new_jh); | |
2418 | return bh->b_private; | |
2419 | } | |
2420 | ||
2421 | /* | |
2422 | * Grab a ref against this buffer_head's journal_head. If it ended up not | |
2423 | * having a journal_head, return NULL | |
2424 | */ | |
f7f4bccb | 2425 | struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) |
470decc6 DK |
2426 | { |
2427 | struct journal_head *jh = NULL; | |
2428 | ||
2429 | jbd_lock_bh_journal_head(bh); | |
2430 | if (buffer_jbd(bh)) { | |
2431 | jh = bh2jh(bh); | |
2432 | jh->b_jcount++; | |
2433 | } | |
2434 | jbd_unlock_bh_journal_head(bh); | |
2435 | return jh; | |
2436 | } | |
2437 | ||
2438 | static void __journal_remove_journal_head(struct buffer_head *bh) | |
2439 | { | |
2440 | struct journal_head *jh = bh2jh(bh); | |
2441 | ||
2442 | J_ASSERT_JH(jh, jh->b_jcount >= 0); | |
de1b7941 JK |
2443 | J_ASSERT_JH(jh, jh->b_transaction == NULL); |
2444 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
2445 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
2446 | J_ASSERT_JH(jh, jh->b_jlist == BJ_None); | |
2447 | J_ASSERT_BH(bh, buffer_jbd(bh)); | |
2448 | J_ASSERT_BH(bh, jh2bh(jh) == bh); | |
2449 | BUFFER_TRACE(bh, "remove journal_head"); | |
2450 | if (jh->b_frozen_data) { | |
2451 | printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__); | |
2452 | jbd2_free(jh->b_frozen_data, bh->b_size); | |
470decc6 | 2453 | } |
de1b7941 JK |
2454 | if (jh->b_committed_data) { |
2455 | printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__); | |
2456 | jbd2_free(jh->b_committed_data, bh->b_size); | |
2457 | } | |
2458 | bh->b_private = NULL; | |
2459 | jh->b_bh = NULL; /* debug, really */ | |
2460 | clear_buffer_jbd(bh); | |
2461 | journal_free_journal_head(jh); | |
470decc6 DK |
2462 | } |
2463 | ||
2464 | /* | |
de1b7941 | 2465 | * Drop a reference on the passed journal_head. If it fell to zero then |
470decc6 DK |
2466 | * release the journal_head from the buffer_head. |
2467 | */ | |
f7f4bccb | 2468 | void jbd2_journal_put_journal_head(struct journal_head *jh) |
470decc6 DK |
2469 | { |
2470 | struct buffer_head *bh = jh2bh(jh); | |
2471 | ||
2472 | jbd_lock_bh_journal_head(bh); | |
2473 | J_ASSERT_JH(jh, jh->b_jcount > 0); | |
2474 | --jh->b_jcount; | |
de1b7941 | 2475 | if (!jh->b_jcount) { |
470decc6 | 2476 | __journal_remove_journal_head(bh); |
de1b7941 | 2477 | jbd_unlock_bh_journal_head(bh); |
470decc6 | 2478 | __brelse(bh); |
de1b7941 JK |
2479 | } else |
2480 | jbd_unlock_bh_journal_head(bh); | |
470decc6 DK |
2481 | } |
2482 | ||
c851ed54 JK |
2483 | /* |
2484 | * Initialize jbd inode head | |
2485 | */ | |
2486 | void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) | |
2487 | { | |
2488 | jinode->i_transaction = NULL; | |
2489 | jinode->i_next_transaction = NULL; | |
2490 | jinode->i_vfs_inode = inode; | |
2491 | jinode->i_flags = 0; | |
2492 | INIT_LIST_HEAD(&jinode->i_list); | |
2493 | } | |
2494 | ||
2495 | /* | |
2496 | * Function to be called before we start removing inode from memory (i.e., | |
2497 | * clear_inode() is a fine place to be called from). It removes inode from | |
2498 | * transaction's lists. | |
2499 | */ | |
2500 | void jbd2_journal_release_jbd_inode(journal_t *journal, | |
2501 | struct jbd2_inode *jinode) | |
2502 | { | |
c851ed54 JK |
2503 | if (!journal) |
2504 | return; | |
2505 | restart: | |
2506 | spin_lock(&journal->j_list_lock); | |
2507 | /* Is commit writing out inode - we have to wait */ | |
39e3ac25 | 2508 | if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) { |
c851ed54 JK |
2509 | wait_queue_head_t *wq; |
2510 | DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); | |
2511 | wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); | |
2512 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
2513 | spin_unlock(&journal->j_list_lock); | |
2514 | schedule(); | |
2515 | finish_wait(wq, &wait.wait); | |
2516 | goto restart; | |
2517 | } | |
2518 | ||
c851ed54 JK |
2519 | if (jinode->i_transaction) { |
2520 | list_del(&jinode->i_list); | |
2521 | jinode->i_transaction = NULL; | |
2522 | } | |
2523 | spin_unlock(&journal->j_list_lock); | |
2524 | } | |
2525 | ||
470decc6 | 2526 | |
8e85fb3f JL |
2527 | #ifdef CONFIG_PROC_FS |
2528 | ||
2529 | #define JBD2_STATS_PROC_NAME "fs/jbd2" | |
2530 | ||
2531 | static void __init jbd2_create_jbd_stats_proc_entry(void) | |
2532 | { | |
2533 | proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); | |
2534 | } | |
2535 | ||
2536 | static void __exit jbd2_remove_jbd_stats_proc_entry(void) | |
2537 | { | |
2538 | if (proc_jbd2_stats) | |
2539 | remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); | |
2540 | } | |
2541 | ||
2542 | #else | |
2543 | ||
2544 | #define jbd2_create_jbd_stats_proc_entry() do {} while (0) | |
2545 | #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) | |
2546 | ||
2547 | #endif | |
2548 | ||
8aefcd55 | 2549 | struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache; |
470decc6 | 2550 | |
4185a2ac | 2551 | static int __init jbd2_journal_init_handle_cache(void) |
470decc6 | 2552 | { |
8aefcd55 | 2553 | jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY); |
f7f4bccb | 2554 | if (jbd2_handle_cache == NULL) { |
8aefcd55 TT |
2555 | printk(KERN_EMERG "JBD2: failed to create handle cache\n"); |
2556 | return -ENOMEM; | |
2557 | } | |
2558 | jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0); | |
2559 | if (jbd2_inode_cache == NULL) { | |
2560 | printk(KERN_EMERG "JBD2: failed to create inode cache\n"); | |
2561 | kmem_cache_destroy(jbd2_handle_cache); | |
470decc6 DK |
2562 | return -ENOMEM; |
2563 | } | |
2564 | return 0; | |
2565 | } | |
2566 | ||
f7f4bccb | 2567 | static void jbd2_journal_destroy_handle_cache(void) |
470decc6 | 2568 | { |
f7f4bccb MC |
2569 | if (jbd2_handle_cache) |
2570 | kmem_cache_destroy(jbd2_handle_cache); | |
8aefcd55 TT |
2571 | if (jbd2_inode_cache) |
2572 | kmem_cache_destroy(jbd2_inode_cache); | |
2573 | ||
470decc6 DK |
2574 | } |
2575 | ||
2576 | /* | |
2577 | * Module startup and shutdown | |
2578 | */ | |
2579 | ||
2580 | static int __init journal_init_caches(void) | |
2581 | { | |
2582 | int ret; | |
2583 | ||
f7f4bccb | 2584 | ret = jbd2_journal_init_revoke_caches(); |
470decc6 | 2585 | if (ret == 0) |
4185a2ac | 2586 | ret = jbd2_journal_init_journal_head_cache(); |
470decc6 | 2587 | if (ret == 0) |
4185a2ac | 2588 | ret = jbd2_journal_init_handle_cache(); |
470decc6 | 2589 | if (ret == 0) |
0c2022ec | 2590 | ret = jbd2_journal_init_transaction_cache(); |
470decc6 DK |
2591 | return ret; |
2592 | } | |
2593 | ||
f7f4bccb | 2594 | static void jbd2_journal_destroy_caches(void) |
470decc6 | 2595 | { |
f7f4bccb | 2596 | jbd2_journal_destroy_revoke_caches(); |
4185a2ac | 2597 | jbd2_journal_destroy_journal_head_cache(); |
f7f4bccb | 2598 | jbd2_journal_destroy_handle_cache(); |
0c2022ec | 2599 | jbd2_journal_destroy_transaction_cache(); |
d2eecb03 | 2600 | jbd2_journal_destroy_slabs(); |
470decc6 DK |
2601 | } |
2602 | ||
2603 | static int __init journal_init(void) | |
2604 | { | |
2605 | int ret; | |
2606 | ||
2607 | BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); | |
2608 | ||
2609 | ret = journal_init_caches(); | |
620de4e1 | 2610 | if (ret == 0) { |
620de4e1 DG |
2611 | jbd2_create_jbd_stats_proc_entry(); |
2612 | } else { | |
f7f4bccb | 2613 | jbd2_journal_destroy_caches(); |
620de4e1 | 2614 | } |
470decc6 DK |
2615 | return ret; |
2616 | } | |
2617 | ||
2618 | static void __exit journal_exit(void) | |
2619 | { | |
e23291b9 | 2620 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2621 | int n = atomic_read(&nr_journal_heads); |
2622 | if (n) | |
f2a44523 | 2623 | printk(KERN_EMERG "JBD2: leaked %d journal_heads!\n", n); |
470decc6 | 2624 | #endif |
8e85fb3f | 2625 | jbd2_remove_jbd_stats_proc_entry(); |
f7f4bccb | 2626 | jbd2_journal_destroy_caches(); |
470decc6 DK |
2627 | } |
2628 | ||
2629 | MODULE_LICENSE("GPL"); | |
2630 | module_init(journal_init); | |
2631 | module_exit(journal_exit); | |
2632 |