]>
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 | |
169f1a2a PG |
106 | #ifdef CONFIG_JBD2_DEBUG |
107 | void __jbd2_debug(int level, const char *file, const char *func, | |
108 | unsigned int line, const char *fmt, ...) | |
109 | { | |
110 | struct va_format vaf; | |
111 | va_list args; | |
112 | ||
113 | if (level > jbd2_journal_enable_debug) | |
114 | return; | |
115 | va_start(args, fmt); | |
116 | vaf.fmt = fmt; | |
117 | vaf.va = &args; | |
118 | printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf); | |
119 | va_end(args); | |
120 | } | |
121 | EXPORT_SYMBOL(__jbd2_debug); | |
122 | #endif | |
123 | ||
25ed6e8a | 124 | /* Checksumming functions */ |
7747e6d0 | 125 | static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb) |
25ed6e8a | 126 | { |
db9ee220 | 127 | if (!jbd2_journal_has_csum_v2or3(j)) |
25ed6e8a DW |
128 | return 1; |
129 | ||
130 | return sb->s_checksum_type == JBD2_CRC32C_CHKSUM; | |
131 | } | |
132 | ||
18a6ea1e | 133 | static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb) |
4fd5ea43 | 134 | { |
18a6ea1e DW |
135 | __u32 csum; |
136 | __be32 old_csum; | |
4fd5ea43 DW |
137 | |
138 | old_csum = sb->s_checksum; | |
139 | sb->s_checksum = 0; | |
140 | csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t)); | |
141 | sb->s_checksum = old_csum; | |
142 | ||
143 | return cpu_to_be32(csum); | |
144 | } | |
145 | ||
7747e6d0 | 146 | static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb) |
4fd5ea43 | 147 | { |
db9ee220 | 148 | if (!jbd2_journal_has_csum_v2or3(j)) |
4fd5ea43 DW |
149 | return 1; |
150 | ||
151 | return sb->s_checksum == jbd2_superblock_csum(j, sb); | |
152 | } | |
153 | ||
7747e6d0 | 154 | static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb) |
4fd5ea43 | 155 | { |
db9ee220 | 156 | if (!jbd2_journal_has_csum_v2or3(j)) |
4fd5ea43 DW |
157 | return; |
158 | ||
159 | sb->s_checksum = jbd2_superblock_csum(j, sb); | |
160 | } | |
161 | ||
470decc6 DK |
162 | /* |
163 | * Helper function used to manage commit timeouts | |
164 | */ | |
165 | ||
166 | static void commit_timeout(unsigned long __data) | |
167 | { | |
168 | struct task_struct * p = (struct task_struct *) __data; | |
169 | ||
170 | wake_up_process(p); | |
171 | } | |
172 | ||
173 | /* | |
f7f4bccb | 174 | * kjournald2: The main thread function used to manage a logging device |
470decc6 DK |
175 | * journal. |
176 | * | |
177 | * This kernel thread is responsible for two things: | |
178 | * | |
179 | * 1) COMMIT: Every so often we need to commit the current state of the | |
180 | * filesystem to disk. The journal thread is responsible for writing | |
181 | * all of the metadata buffers to disk. | |
182 | * | |
183 | * 2) CHECKPOINT: We cannot reuse a used section of the log file until all | |
184 | * of the data in that part of the log has been rewritten elsewhere on | |
185 | * the disk. Flushing these old buffers to reclaim space in the log is | |
186 | * known as checkpointing, and this thread is responsible for that job. | |
187 | */ | |
188 | ||
f7f4bccb | 189 | static int kjournald2(void *arg) |
470decc6 DK |
190 | { |
191 | journal_t *journal = arg; | |
192 | transaction_t *transaction; | |
193 | ||
194 | /* | |
195 | * Set up an interval timer which can be used to trigger a commit wakeup | |
196 | * after the commit interval expires | |
197 | */ | |
198 | setup_timer(&journal->j_commit_timer, commit_timeout, | |
199 | (unsigned long)current); | |
200 | ||
35c80422 NC |
201 | set_freezable(); |
202 | ||
470decc6 DK |
203 | /* Record that the journal thread is running */ |
204 | journal->j_task = current; | |
205 | wake_up(&journal->j_wait_done_commit); | |
206 | ||
470decc6 DK |
207 | /* |
208 | * And now, wait forever for commit wakeup events. | |
209 | */ | |
a931da6a | 210 | write_lock(&journal->j_state_lock); |
470decc6 DK |
211 | |
212 | loop: | |
f7f4bccb | 213 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
214 | goto end_loop; |
215 | ||
216 | jbd_debug(1, "commit_sequence=%d, commit_request=%d\n", | |
217 | journal->j_commit_sequence, journal->j_commit_request); | |
218 | ||
219 | if (journal->j_commit_sequence != journal->j_commit_request) { | |
220 | jbd_debug(1, "OK, requests differ\n"); | |
a931da6a | 221 | write_unlock(&journal->j_state_lock); |
470decc6 | 222 | del_timer_sync(&journal->j_commit_timer); |
f7f4bccb | 223 | jbd2_journal_commit_transaction(journal); |
a931da6a | 224 | write_lock(&journal->j_state_lock); |
470decc6 DK |
225 | goto loop; |
226 | } | |
227 | ||
228 | wake_up(&journal->j_wait_done_commit); | |
229 | if (freezing(current)) { | |
230 | /* | |
231 | * The simpler the better. Flushing journal isn't a | |
232 | * good idea, because that depends on threads that may | |
233 | * be already stopped. | |
234 | */ | |
f7f4bccb | 235 | jbd_debug(1, "Now suspending kjournald2\n"); |
a931da6a | 236 | write_unlock(&journal->j_state_lock); |
a0acae0e | 237 | try_to_freeze(); |
a931da6a | 238 | write_lock(&journal->j_state_lock); |
470decc6 DK |
239 | } else { |
240 | /* | |
241 | * We assume on resume that commits are already there, | |
242 | * so we don't sleep | |
243 | */ | |
244 | DEFINE_WAIT(wait); | |
245 | int should_sleep = 1; | |
246 | ||
247 | prepare_to_wait(&journal->j_wait_commit, &wait, | |
248 | TASK_INTERRUPTIBLE); | |
249 | if (journal->j_commit_sequence != journal->j_commit_request) | |
250 | should_sleep = 0; | |
251 | transaction = journal->j_running_transaction; | |
252 | if (transaction && time_after_eq(jiffies, | |
253 | transaction->t_expires)) | |
254 | should_sleep = 0; | |
f7f4bccb | 255 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
256 | should_sleep = 0; |
257 | if (should_sleep) { | |
a931da6a | 258 | write_unlock(&journal->j_state_lock); |
470decc6 | 259 | schedule(); |
a931da6a | 260 | write_lock(&journal->j_state_lock); |
470decc6 DK |
261 | } |
262 | finish_wait(&journal->j_wait_commit, &wait); | |
263 | } | |
264 | ||
f7f4bccb | 265 | jbd_debug(1, "kjournald2 wakes\n"); |
470decc6 DK |
266 | |
267 | /* | |
268 | * Were we woken up by a commit wakeup event? | |
269 | */ | |
270 | transaction = journal->j_running_transaction; | |
271 | if (transaction && time_after_eq(jiffies, transaction->t_expires)) { | |
272 | journal->j_commit_request = transaction->t_tid; | |
273 | jbd_debug(1, "woke because of timeout\n"); | |
274 | } | |
275 | goto loop; | |
276 | ||
277 | end_loop: | |
a931da6a | 278 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
279 | del_timer_sync(&journal->j_commit_timer); |
280 | journal->j_task = NULL; | |
281 | wake_up(&journal->j_wait_done_commit); | |
282 | jbd_debug(1, "Journal thread exiting.\n"); | |
283 | return 0; | |
284 | } | |
285 | ||
97f06784 | 286 | static int jbd2_journal_start_thread(journal_t *journal) |
470decc6 | 287 | { |
97f06784 PE |
288 | struct task_struct *t; |
289 | ||
90576c0b TT |
290 | t = kthread_run(kjournald2, journal, "jbd2/%s", |
291 | journal->j_devname); | |
97f06784 PE |
292 | if (IS_ERR(t)) |
293 | return PTR_ERR(t); | |
294 | ||
1076d17a | 295 | wait_event(journal->j_wait_done_commit, journal->j_task != NULL); |
97f06784 | 296 | return 0; |
470decc6 DK |
297 | } |
298 | ||
299 | static void journal_kill_thread(journal_t *journal) | |
300 | { | |
a931da6a | 301 | write_lock(&journal->j_state_lock); |
f7f4bccb | 302 | journal->j_flags |= JBD2_UNMOUNT; |
470decc6 DK |
303 | |
304 | while (journal->j_task) { | |
a931da6a | 305 | write_unlock(&journal->j_state_lock); |
3469a32a | 306 | wake_up(&journal->j_wait_commit); |
1076d17a | 307 | wait_event(journal->j_wait_done_commit, journal->j_task == NULL); |
a931da6a | 308 | write_lock(&journal->j_state_lock); |
470decc6 | 309 | } |
a931da6a | 310 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
311 | } |
312 | ||
313 | /* | |
f7f4bccb | 314 | * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. |
470decc6 DK |
315 | * |
316 | * Writes a metadata buffer to a given disk block. The actual IO is not | |
317 | * performed but a new buffer_head is constructed which labels the data | |
318 | * to be written with the correct destination disk block. | |
319 | * | |
320 | * Any magic-number escaping which needs to be done will cause a | |
321 | * copy-out here. If the buffer happens to start with the | |
f7f4bccb | 322 | * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the |
470decc6 DK |
323 | * magic number is only written to the log for descripter blocks. In |
324 | * this case, we copy the data and replace the first word with 0, and we | |
325 | * return a result code which indicates that this buffer needs to be | |
326 | * marked as an escaped buffer in the corresponding log descriptor | |
327 | * block. The missing word can then be restored when the block is read | |
328 | * during recovery. | |
329 | * | |
330 | * If the source buffer has already been modified by a new transaction | |
331 | * since we took the last commit snapshot, we use the frozen copy of | |
f5113eff JK |
332 | * that data for IO. If we end up using the existing buffer_head's data |
333 | * for the write, then we have to make sure nobody modifies it while the | |
334 | * IO is in progress. do_get_write_access() handles this. | |
470decc6 | 335 | * |
f5113eff JK |
336 | * The function returns a pointer to the buffer_head to be used for IO. |
337 | * | |
470decc6 DK |
338 | * |
339 | * Return value: | |
340 | * <0: Error | |
341 | * >=0: Finished OK | |
342 | * | |
343 | * On success: | |
344 | * Bit 0 set == escape performed on the data | |
345 | * Bit 1 set == buffer copy-out performed (kfree the data after IO) | |
346 | */ | |
347 | ||
f7f4bccb | 348 | int jbd2_journal_write_metadata_buffer(transaction_t *transaction, |
470decc6 | 349 | struct journal_head *jh_in, |
f5113eff JK |
350 | struct buffer_head **bh_out, |
351 | sector_t blocknr) | |
470decc6 DK |
352 | { |
353 | int need_copy_out = 0; | |
354 | int done_copy_out = 0; | |
355 | int do_escape = 0; | |
356 | char *mapped_data; | |
357 | struct buffer_head *new_bh; | |
470decc6 DK |
358 | struct page *new_page; |
359 | unsigned int new_offset; | |
360 | struct buffer_head *bh_in = jh2bh(jh_in); | |
96577c43 | 361 | journal_t *journal = transaction->t_journal; |
470decc6 DK |
362 | |
363 | /* | |
364 | * The buffer really shouldn't be locked: only the current committing | |
365 | * transaction is allowed to write it, so nobody else is allowed | |
366 | * to do any IO. | |
367 | * | |
368 | * akpm: except if we're journalling data, and write() output is | |
369 | * also part of a shared mapping, and another thread has | |
370 | * decided to launch a writepage() against this buffer. | |
371 | */ | |
372 | J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); | |
373 | ||
47def826 TT |
374 | retry_alloc: |
375 | new_bh = alloc_buffer_head(GFP_NOFS); | |
376 | if (!new_bh) { | |
377 | /* | |
378 | * Failure is not an option, but __GFP_NOFAIL is going | |
379 | * away; so we retry ourselves here. | |
380 | */ | |
381 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
382 | goto retry_alloc; | |
383 | } | |
384 | ||
96577c43 | 385 | /* keep subsequent assertions sane */ |
96577c43 | 386 | atomic_set(&new_bh->b_count, 1); |
470decc6 | 387 | |
f5113eff JK |
388 | jbd_lock_bh_state(bh_in); |
389 | repeat: | |
470decc6 DK |
390 | /* |
391 | * If a new transaction has already done a buffer copy-out, then | |
392 | * we use that version of the data for the commit. | |
393 | */ | |
470decc6 DK |
394 | if (jh_in->b_frozen_data) { |
395 | done_copy_out = 1; | |
396 | new_page = virt_to_page(jh_in->b_frozen_data); | |
397 | new_offset = offset_in_page(jh_in->b_frozen_data); | |
398 | } else { | |
399 | new_page = jh2bh(jh_in)->b_page; | |
400 | new_offset = offset_in_page(jh2bh(jh_in)->b_data); | |
401 | } | |
402 | ||
303a8f2a | 403 | mapped_data = kmap_atomic(new_page); |
e06c8227 | 404 | /* |
13ceef09 JK |
405 | * Fire data frozen trigger if data already wasn't frozen. Do this |
406 | * before checking for escaping, as the trigger may modify the magic | |
407 | * offset. If a copy-out happens afterwards, it will have the correct | |
408 | * data in the buffer. | |
e06c8227 | 409 | */ |
13ceef09 JK |
410 | if (!done_copy_out) |
411 | jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset, | |
412 | jh_in->b_triggers); | |
e06c8227 | 413 | |
470decc6 DK |
414 | /* |
415 | * Check for escaping | |
416 | */ | |
417 | if (*((__be32 *)(mapped_data + new_offset)) == | |
f7f4bccb | 418 | cpu_to_be32(JBD2_MAGIC_NUMBER)) { |
470decc6 DK |
419 | need_copy_out = 1; |
420 | do_escape = 1; | |
421 | } | |
303a8f2a | 422 | kunmap_atomic(mapped_data); |
470decc6 DK |
423 | |
424 | /* | |
425 | * Do we need to do a data copy? | |
426 | */ | |
427 | if (need_copy_out && !done_copy_out) { | |
428 | char *tmp; | |
429 | ||
430 | jbd_unlock_bh_state(bh_in); | |
af1e76d6 | 431 | tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); |
e6ec116b | 432 | if (!tmp) { |
f5113eff | 433 | brelse(new_bh); |
e6ec116b TT |
434 | return -ENOMEM; |
435 | } | |
470decc6 DK |
436 | jbd_lock_bh_state(bh_in); |
437 | if (jh_in->b_frozen_data) { | |
af1e76d6 | 438 | jbd2_free(tmp, bh_in->b_size); |
470decc6 DK |
439 | goto repeat; |
440 | } | |
441 | ||
442 | jh_in->b_frozen_data = tmp; | |
303a8f2a | 443 | mapped_data = kmap_atomic(new_page); |
f5113eff | 444 | memcpy(tmp, mapped_data + new_offset, bh_in->b_size); |
303a8f2a | 445 | kunmap_atomic(mapped_data); |
470decc6 DK |
446 | |
447 | new_page = virt_to_page(tmp); | |
448 | new_offset = offset_in_page(tmp); | |
449 | done_copy_out = 1; | |
e06c8227 JB |
450 | |
451 | /* | |
452 | * This isn't strictly necessary, as we're using frozen | |
453 | * data for the escaping, but it keeps consistency with | |
454 | * b_frozen_data usage. | |
455 | */ | |
456 | jh_in->b_frozen_triggers = jh_in->b_triggers; | |
470decc6 DK |
457 | } |
458 | ||
459 | /* | |
460 | * Did we need to do an escaping? Now we've done all the | |
461 | * copying, we can finally do so. | |
462 | */ | |
463 | if (do_escape) { | |
303a8f2a | 464 | mapped_data = kmap_atomic(new_page); |
470decc6 | 465 | *((unsigned int *)(mapped_data + new_offset)) = 0; |
303a8f2a | 466 | kunmap_atomic(mapped_data); |
470decc6 DK |
467 | } |
468 | ||
470decc6 | 469 | set_bh_page(new_bh, new_page, new_offset); |
f5113eff JK |
470 | new_bh->b_size = bh_in->b_size; |
471 | new_bh->b_bdev = journal->j_dev; | |
470decc6 | 472 | new_bh->b_blocknr = blocknr; |
b34090e5 | 473 | new_bh->b_private = bh_in; |
470decc6 DK |
474 | set_buffer_mapped(new_bh); |
475 | set_buffer_dirty(new_bh); | |
476 | ||
f5113eff | 477 | *bh_out = new_bh; |
470decc6 DK |
478 | |
479 | /* | |
480 | * The to-be-written buffer needs to get moved to the io queue, | |
481 | * and the original buffer whose contents we are shadowing or | |
482 | * copying is moved to the transaction's shadow queue. | |
483 | */ | |
484 | JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); | |
96577c43 | 485 | spin_lock(&journal->j_list_lock); |
486 | __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); | |
487 | spin_unlock(&journal->j_list_lock); | |
b34090e5 | 488 | set_buffer_shadow(bh_in); |
96577c43 | 489 | jbd_unlock_bh_state(bh_in); |
490 | ||
470decc6 DK |
491 | return do_escape | (done_copy_out << 1); |
492 | } | |
493 | ||
494 | /* | |
495 | * Allocation code for the journal file. Manage the space left in the | |
496 | * journal, so that we can begin checkpointing when appropriate. | |
497 | */ | |
498 | ||
470decc6 | 499 | /* |
e4471831 TT |
500 | * Called with j_state_lock locked for writing. |
501 | * Returns true if a transaction commit was started. | |
470decc6 | 502 | */ |
f7f4bccb | 503 | int __jbd2_log_start_commit(journal_t *journal, tid_t target) |
470decc6 | 504 | { |
e7b04ac0 ES |
505 | /* Return if the txn has already requested to be committed */ |
506 | if (journal->j_commit_request == target) | |
507 | return 0; | |
508 | ||
470decc6 | 509 | /* |
deeeaf13 TT |
510 | * The only transaction we can possibly wait upon is the |
511 | * currently running transaction (if it exists). Otherwise, | |
512 | * the target tid must be an old one. | |
470decc6 | 513 | */ |
deeeaf13 TT |
514 | if (journal->j_running_transaction && |
515 | journal->j_running_transaction->t_tid == target) { | |
470decc6 | 516 | /* |
bcf3d0bc | 517 | * We want a new commit: OK, mark the request and wakeup the |
470decc6 DK |
518 | * commit thread. We do _not_ do the commit ourselves. |
519 | */ | |
520 | ||
521 | journal->j_commit_request = target; | |
f2a44523 | 522 | jbd_debug(1, "JBD2: requesting commit %d/%d\n", |
470decc6 DK |
523 | journal->j_commit_request, |
524 | journal->j_commit_sequence); | |
9fff24aa | 525 | journal->j_running_transaction->t_requested = jiffies; |
470decc6 DK |
526 | wake_up(&journal->j_wait_commit); |
527 | return 1; | |
deeeaf13 TT |
528 | } else if (!tid_geq(journal->j_commit_request, target)) |
529 | /* This should never happen, but if it does, preserve | |
530 | the evidence before kjournald goes into a loop and | |
531 | increments j_commit_sequence beyond all recognition. */ | |
f2a44523 | 532 | WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", |
1be2add6 TT |
533 | journal->j_commit_request, |
534 | journal->j_commit_sequence, | |
535 | target, journal->j_running_transaction ? | |
536 | journal->j_running_transaction->t_tid : 0); | |
470decc6 DK |
537 | return 0; |
538 | } | |
539 | ||
f7f4bccb | 540 | int jbd2_log_start_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
541 | { |
542 | int ret; | |
543 | ||
a931da6a | 544 | write_lock(&journal->j_state_lock); |
f7f4bccb | 545 | ret = __jbd2_log_start_commit(journal, tid); |
a931da6a | 546 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
547 | return ret; |
548 | } | |
549 | ||
550 | /* | |
9ff86446 DM |
551 | * Force and wait any uncommitted transactions. We can only force the running |
552 | * transaction if we don't have an active handle, otherwise, we will deadlock. | |
553 | * Returns: <0 in case of error, | |
554 | * 0 if nothing to commit, | |
555 | * 1 if transaction was successfully committed. | |
470decc6 | 556 | */ |
9ff86446 | 557 | static int __jbd2_journal_force_commit(journal_t *journal) |
470decc6 DK |
558 | { |
559 | transaction_t *transaction = NULL; | |
560 | tid_t tid; | |
9ff86446 | 561 | int need_to_start = 0, ret = 0; |
470decc6 | 562 | |
a931da6a | 563 | read_lock(&journal->j_state_lock); |
470decc6 DK |
564 | if (journal->j_running_transaction && !current->journal_info) { |
565 | transaction = journal->j_running_transaction; | |
e4471831 TT |
566 | if (!tid_geq(journal->j_commit_request, transaction->t_tid)) |
567 | need_to_start = 1; | |
470decc6 DK |
568 | } else if (journal->j_committing_transaction) |
569 | transaction = journal->j_committing_transaction; | |
570 | ||
571 | if (!transaction) { | |
9ff86446 | 572 | /* Nothing to commit */ |
a931da6a | 573 | read_unlock(&journal->j_state_lock); |
9ff86446 | 574 | return 0; |
470decc6 | 575 | } |
470decc6 | 576 | tid = transaction->t_tid; |
a931da6a | 577 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
578 | if (need_to_start) |
579 | jbd2_log_start_commit(journal, tid); | |
9ff86446 DM |
580 | ret = jbd2_log_wait_commit(journal, tid); |
581 | if (!ret) | |
582 | ret = 1; | |
583 | ||
584 | return ret; | |
585 | } | |
586 | ||
587 | /** | |
588 | * Force and wait upon a commit if the calling process is not within | |
589 | * transaction. This is used for forcing out undo-protected data which contains | |
590 | * bitmaps, when the fs is running out of space. | |
591 | * | |
592 | * @journal: journal to force | |
593 | * Returns true if progress was made. | |
594 | */ | |
595 | int jbd2_journal_force_commit_nested(journal_t *journal) | |
596 | { | |
597 | int ret; | |
598 | ||
599 | ret = __jbd2_journal_force_commit(journal); | |
600 | return ret > 0; | |
601 | } | |
602 | ||
603 | /** | |
604 | * int journal_force_commit() - force any uncommitted transactions | |
605 | * @journal: journal to force | |
606 | * | |
607 | * Caller want unconditional commit. We can only force the running transaction | |
608 | * if we don't have an active handle, otherwise, we will deadlock. | |
609 | */ | |
610 | int jbd2_journal_force_commit(journal_t *journal) | |
611 | { | |
612 | int ret; | |
613 | ||
614 | J_ASSERT(!current->journal_info); | |
615 | ret = __jbd2_journal_force_commit(journal); | |
616 | if (ret > 0) | |
617 | ret = 0; | |
618 | return ret; | |
470decc6 DK |
619 | } |
620 | ||
621 | /* | |
622 | * Start a commit of the current running transaction (if any). Returns true | |
c88ccea3 JK |
623 | * if a transaction is going to be committed (or is currently already |
624 | * committing), and fills its tid in at *ptid | |
470decc6 | 625 | */ |
f7f4bccb | 626 | int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) |
470decc6 DK |
627 | { |
628 | int ret = 0; | |
629 | ||
a931da6a | 630 | write_lock(&journal->j_state_lock); |
470decc6 DK |
631 | if (journal->j_running_transaction) { |
632 | tid_t tid = journal->j_running_transaction->t_tid; | |
633 | ||
c88ccea3 JK |
634 | __jbd2_log_start_commit(journal, tid); |
635 | /* There's a running transaction and we've just made sure | |
636 | * it's commit has been scheduled. */ | |
637 | if (ptid) | |
470decc6 | 638 | *ptid = tid; |
c88ccea3 JK |
639 | ret = 1; |
640 | } else if (journal->j_committing_transaction) { | |
470decc6 | 641 | /* |
12810ad7 AB |
642 | * If commit has been started, then we have to wait for |
643 | * completion of that transaction. | |
470decc6 | 644 | */ |
c88ccea3 JK |
645 | if (ptid) |
646 | *ptid = journal->j_committing_transaction->t_tid; | |
470decc6 DK |
647 | ret = 1; |
648 | } | |
a931da6a | 649 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
650 | return ret; |
651 | } | |
652 | ||
bbd2be36 JK |
653 | /* |
654 | * Return 1 if a given transaction has not yet sent barrier request | |
655 | * connected with a transaction commit. If 0 is returned, transaction | |
656 | * may or may not have sent the barrier. Used to avoid sending barrier | |
657 | * twice in common cases. | |
658 | */ | |
659 | int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) | |
660 | { | |
661 | int ret = 0; | |
662 | transaction_t *commit_trans; | |
663 | ||
664 | if (!(journal->j_flags & JBD2_BARRIER)) | |
665 | return 0; | |
666 | read_lock(&journal->j_state_lock); | |
667 | /* Transaction already committed? */ | |
668 | if (tid_geq(journal->j_commit_sequence, tid)) | |
669 | goto out; | |
670 | commit_trans = journal->j_committing_transaction; | |
671 | if (!commit_trans || commit_trans->t_tid != tid) { | |
672 | ret = 1; | |
673 | goto out; | |
674 | } | |
675 | /* | |
676 | * Transaction is being committed and we already proceeded to | |
677 | * submitting a flush to fs partition? | |
678 | */ | |
679 | if (journal->j_fs_dev != journal->j_dev) { | |
680 | if (!commit_trans->t_need_data_flush || | |
681 | commit_trans->t_state >= T_COMMIT_DFLUSH) | |
682 | goto out; | |
683 | } else { | |
684 | if (commit_trans->t_state >= T_COMMIT_JFLUSH) | |
685 | goto out; | |
686 | } | |
687 | ret = 1; | |
688 | out: | |
689 | read_unlock(&journal->j_state_lock); | |
690 | return ret; | |
691 | } | |
692 | EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); | |
693 | ||
470decc6 DK |
694 | /* |
695 | * Wait for a specified commit to complete. | |
696 | * The caller may not hold the journal lock. | |
697 | */ | |
f7f4bccb | 698 | int jbd2_log_wait_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
699 | { |
700 | int err = 0; | |
701 | ||
a931da6a | 702 | read_lock(&journal->j_state_lock); |
e23291b9 | 703 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 | 704 | if (!tid_geq(journal->j_commit_request, tid)) { |
75685071 | 705 | printk(KERN_ERR |
470decc6 | 706 | "%s: error: j_commit_request=%d, tid=%d\n", |
329d291f | 707 | __func__, journal->j_commit_request, tid); |
470decc6 | 708 | } |
470decc6 | 709 | #endif |
470decc6 | 710 | while (tid_gt(tid, journal->j_commit_sequence)) { |
f2a44523 | 711 | jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n", |
470decc6 | 712 | tid, journal->j_commit_sequence); |
a931da6a | 713 | read_unlock(&journal->j_state_lock); |
3469a32a | 714 | wake_up(&journal->j_wait_commit); |
470decc6 DK |
715 | wait_event(journal->j_wait_done_commit, |
716 | !tid_gt(tid, journal->j_commit_sequence)); | |
a931da6a | 717 | read_lock(&journal->j_state_lock); |
470decc6 | 718 | } |
a931da6a | 719 | read_unlock(&journal->j_state_lock); |
470decc6 | 720 | |
75685071 | 721 | if (unlikely(is_journal_aborted(journal))) |
470decc6 | 722 | err = -EIO; |
470decc6 DK |
723 | return err; |
724 | } | |
725 | ||
d76a3a77 TT |
726 | /* |
727 | * When this function returns the transaction corresponding to tid | |
728 | * will be completed. If the transaction has currently running, start | |
729 | * committing that transaction before waiting for it to complete. If | |
730 | * the transaction id is stale, it is by definition already completed, | |
731 | * so just return SUCCESS. | |
732 | */ | |
733 | int jbd2_complete_transaction(journal_t *journal, tid_t tid) | |
734 | { | |
735 | int need_to_wait = 1; | |
736 | ||
737 | read_lock(&journal->j_state_lock); | |
738 | if (journal->j_running_transaction && | |
739 | journal->j_running_transaction->t_tid == tid) { | |
740 | if (journal->j_commit_request != tid) { | |
741 | /* transaction not yet started, so request it */ | |
742 | read_unlock(&journal->j_state_lock); | |
743 | jbd2_log_start_commit(journal, tid); | |
744 | goto wait_commit; | |
745 | } | |
746 | } else if (!(journal->j_committing_transaction && | |
747 | journal->j_committing_transaction->t_tid == tid)) | |
748 | need_to_wait = 0; | |
749 | read_unlock(&journal->j_state_lock); | |
750 | if (!need_to_wait) | |
751 | return 0; | |
752 | wait_commit: | |
753 | return jbd2_log_wait_commit(journal, tid); | |
754 | } | |
755 | EXPORT_SYMBOL(jbd2_complete_transaction); | |
756 | ||
470decc6 DK |
757 | /* |
758 | * Log buffer allocation routines: | |
759 | */ | |
760 | ||
18eba7aa | 761 | int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) |
470decc6 DK |
762 | { |
763 | unsigned long blocknr; | |
764 | ||
a931da6a | 765 | write_lock(&journal->j_state_lock); |
470decc6 DK |
766 | J_ASSERT(journal->j_free > 1); |
767 | ||
768 | blocknr = journal->j_head; | |
769 | journal->j_head++; | |
770 | journal->j_free--; | |
771 | if (journal->j_head == journal->j_last) | |
772 | journal->j_head = journal->j_first; | |
a931da6a | 773 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 774 | return jbd2_journal_bmap(journal, blocknr, retp); |
470decc6 DK |
775 | } |
776 | ||
777 | /* | |
778 | * Conversion of logical to physical block numbers for the journal | |
779 | * | |
780 | * On external journals the journal blocks are identity-mapped, so | |
781 | * this is a no-op. If needed, we can use j_blk_offset - everything is | |
782 | * ready. | |
783 | */ | |
f7f4bccb | 784 | int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, |
18eba7aa | 785 | unsigned long long *retp) |
470decc6 DK |
786 | { |
787 | int err = 0; | |
18eba7aa | 788 | unsigned long long ret; |
470decc6 DK |
789 | |
790 | if (journal->j_inode) { | |
791 | ret = bmap(journal->j_inode, blocknr); | |
792 | if (ret) | |
793 | *retp = ret; | |
794 | else { | |
470decc6 DK |
795 | printk(KERN_ALERT "%s: journal block not found " |
796 | "at offset %lu on %s\n", | |
05496769 | 797 | __func__, blocknr, journal->j_devname); |
470decc6 DK |
798 | err = -EIO; |
799 | __journal_abort_soft(journal, err); | |
800 | } | |
801 | } else { | |
802 | *retp = blocknr; /* +journal->j_blk_offset */ | |
803 | } | |
804 | return err; | |
805 | } | |
806 | ||
807 | /* | |
808 | * We play buffer_head aliasing tricks to write data/metadata blocks to | |
809 | * the journal without copying their contents, but for journal | |
810 | * descriptor blocks we do need to generate bona fide buffers. | |
811 | * | |
f7f4bccb | 812 | * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying |
470decc6 DK |
813 | * the buffer's contents they really should run flush_dcache_page(bh->b_page). |
814 | * But we don't bother doing that, so there will be coherency problems with | |
815 | * mmaps of blockdevs which hold live JBD-controlled filesystems. | |
816 | */ | |
e5a120ae | 817 | struct buffer_head *jbd2_journal_get_descriptor_buffer(journal_t *journal) |
470decc6 DK |
818 | { |
819 | struct buffer_head *bh; | |
18eba7aa | 820 | unsigned long long blocknr; |
470decc6 DK |
821 | int err; |
822 | ||
f7f4bccb | 823 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
824 | |
825 | if (err) | |
826 | return NULL; | |
827 | ||
828 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
4b905671 JK |
829 | if (!bh) |
830 | return NULL; | |
470decc6 DK |
831 | lock_buffer(bh); |
832 | memset(bh->b_data, 0, journal->j_blocksize); | |
833 | set_buffer_uptodate(bh); | |
834 | unlock_buffer(bh); | |
835 | BUFFER_TRACE(bh, "return this buffer"); | |
e5a120ae | 836 | return bh; |
470decc6 DK |
837 | } |
838 | ||
79feb521 JK |
839 | /* |
840 | * Return tid of the oldest transaction in the journal and block in the journal | |
841 | * where the transaction starts. | |
842 | * | |
843 | * If the journal is now empty, return which will be the next transaction ID | |
844 | * we will write and where will that transaction start. | |
845 | * | |
846 | * The return value is 0 if journal tail cannot be pushed any further, 1 if | |
847 | * it can. | |
848 | */ | |
849 | int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, | |
850 | unsigned long *block) | |
851 | { | |
852 | transaction_t *transaction; | |
853 | int ret; | |
854 | ||
855 | read_lock(&journal->j_state_lock); | |
856 | spin_lock(&journal->j_list_lock); | |
857 | transaction = journal->j_checkpoint_transactions; | |
858 | if (transaction) { | |
859 | *tid = transaction->t_tid; | |
860 | *block = transaction->t_log_start; | |
861 | } else if ((transaction = journal->j_committing_transaction) != NULL) { | |
862 | *tid = transaction->t_tid; | |
863 | *block = transaction->t_log_start; | |
864 | } else if ((transaction = journal->j_running_transaction) != NULL) { | |
865 | *tid = transaction->t_tid; | |
866 | *block = journal->j_head; | |
867 | } else { | |
868 | *tid = journal->j_transaction_sequence; | |
869 | *block = journal->j_head; | |
870 | } | |
871 | ret = tid_gt(*tid, journal->j_tail_sequence); | |
872 | spin_unlock(&journal->j_list_lock); | |
873 | read_unlock(&journal->j_state_lock); | |
874 | ||
875 | return ret; | |
876 | } | |
877 | ||
878 | /* | |
879 | * Update information in journal structure and in on disk journal superblock | |
880 | * about log tail. This function does not check whether information passed in | |
881 | * really pushes log tail further. It's responsibility of the caller to make | |
882 | * sure provided log tail information is valid (e.g. by holding | |
883 | * j_checkpoint_mutex all the time between computing log tail and calling this | |
884 | * function as is the case with jbd2_cleanup_journal_tail()). | |
885 | * | |
886 | * Requires j_checkpoint_mutex | |
887 | */ | |
888 | void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) | |
889 | { | |
890 | unsigned long freed; | |
891 | ||
892 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); | |
893 | ||
894 | /* | |
895 | * We cannot afford for write to remain in drive's caches since as | |
896 | * soon as we update j_tail, next transaction can start reusing journal | |
897 | * space and if we lose sb update during power failure we'd replay | |
898 | * old transaction with possibly newly overwritten data. | |
899 | */ | |
900 | jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA); | |
901 | write_lock(&journal->j_state_lock); | |
902 | freed = block - journal->j_tail; | |
903 | if (block < journal->j_tail) | |
904 | freed += journal->j_last - journal->j_first; | |
905 | ||
906 | trace_jbd2_update_log_tail(journal, tid, block, freed); | |
907 | jbd_debug(1, | |
908 | "Cleaning journal tail from %d to %d (offset %lu), " | |
909 | "freeing %lu\n", | |
910 | journal->j_tail_sequence, tid, block, freed); | |
911 | ||
912 | journal->j_free += freed; | |
913 | journal->j_tail_sequence = tid; | |
914 | journal->j_tail = block; | |
915 | write_unlock(&journal->j_state_lock); | |
916 | } | |
917 | ||
3339578f JK |
918 | /* |
919 | * This is a variaon of __jbd2_update_log_tail which checks for validity of | |
920 | * provided log tail and locks j_checkpoint_mutex. So it is safe against races | |
921 | * with other threads updating log tail. | |
922 | */ | |
923 | void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) | |
924 | { | |
925 | mutex_lock(&journal->j_checkpoint_mutex); | |
926 | if (tid_gt(tid, journal->j_tail_sequence)) | |
927 | __jbd2_update_log_tail(journal, tid, block); | |
928 | mutex_unlock(&journal->j_checkpoint_mutex); | |
929 | } | |
930 | ||
8e85fb3f JL |
931 | struct jbd2_stats_proc_session { |
932 | journal_t *journal; | |
933 | struct transaction_stats_s *stats; | |
934 | int start; | |
935 | int max; | |
936 | }; | |
937 | ||
8e85fb3f JL |
938 | static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) |
939 | { | |
940 | return *pos ? NULL : SEQ_START_TOKEN; | |
941 | } | |
942 | ||
943 | static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) | |
944 | { | |
945 | return NULL; | |
946 | } | |
947 | ||
948 | static int jbd2_seq_info_show(struct seq_file *seq, void *v) | |
949 | { | |
950 | struct jbd2_stats_proc_session *s = seq->private; | |
951 | ||
952 | if (v != SEQ_START_TOKEN) | |
953 | return 0; | |
9fff24aa TT |
954 | seq_printf(seq, "%lu transactions (%lu requested), " |
955 | "each up to %u blocks\n", | |
956 | s->stats->ts_tid, s->stats->ts_requested, | |
957 | s->journal->j_max_transaction_buffers); | |
8e85fb3f JL |
958 | if (s->stats->ts_tid == 0) |
959 | return 0; | |
960 | seq_printf(seq, "average: \n %ums waiting for transaction\n", | |
bf699327 | 961 | jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); |
9fff24aa TT |
962 | seq_printf(seq, " %ums request delay\n", |
963 | (s->stats->ts_requested == 0) ? 0 : | |
964 | jiffies_to_msecs(s->stats->run.rs_request_delay / | |
965 | s->stats->ts_requested)); | |
8e85fb3f | 966 | seq_printf(seq, " %ums running transaction\n", |
bf699327 | 967 | jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); |
8e85fb3f | 968 | seq_printf(seq, " %ums transaction was being locked\n", |
bf699327 | 969 | jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); |
8e85fb3f | 970 | seq_printf(seq, " %ums flushing data (in ordered mode)\n", |
bf699327 | 971 | jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); |
8e85fb3f | 972 | seq_printf(seq, " %ums logging transaction\n", |
bf699327 | 973 | jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); |
c225aa57 SHT |
974 | seq_printf(seq, " %lluus average transaction commit time\n", |
975 | div_u64(s->journal->j_average_commit_time, 1000)); | |
8e85fb3f | 976 | seq_printf(seq, " %lu handles per transaction\n", |
bf699327 | 977 | s->stats->run.rs_handle_count / s->stats->ts_tid); |
8e85fb3f | 978 | seq_printf(seq, " %lu blocks per transaction\n", |
bf699327 | 979 | s->stats->run.rs_blocks / s->stats->ts_tid); |
8e85fb3f | 980 | seq_printf(seq, " %lu logged blocks per transaction\n", |
bf699327 | 981 | s->stats->run.rs_blocks_logged / s->stats->ts_tid); |
8e85fb3f JL |
982 | return 0; |
983 | } | |
984 | ||
985 | static void jbd2_seq_info_stop(struct seq_file *seq, void *v) | |
986 | { | |
987 | } | |
988 | ||
88e9d34c | 989 | static const struct seq_operations jbd2_seq_info_ops = { |
8e85fb3f JL |
990 | .start = jbd2_seq_info_start, |
991 | .next = jbd2_seq_info_next, | |
992 | .stop = jbd2_seq_info_stop, | |
993 | .show = jbd2_seq_info_show, | |
994 | }; | |
995 | ||
996 | static int jbd2_seq_info_open(struct inode *inode, struct file *file) | |
997 | { | |
d9dda78b | 998 | journal_t *journal = PDE_DATA(inode); |
8e85fb3f JL |
999 | struct jbd2_stats_proc_session *s; |
1000 | int rc, size; | |
1001 | ||
1002 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
1003 | if (s == NULL) | |
1004 | return -ENOMEM; | |
1005 | size = sizeof(struct transaction_stats_s); | |
1006 | s->stats = kmalloc(size, GFP_KERNEL); | |
1007 | if (s->stats == NULL) { | |
1008 | kfree(s); | |
1009 | return -ENOMEM; | |
1010 | } | |
1011 | spin_lock(&journal->j_history_lock); | |
1012 | memcpy(s->stats, &journal->j_stats, size); | |
1013 | s->journal = journal; | |
1014 | spin_unlock(&journal->j_history_lock); | |
1015 | ||
1016 | rc = seq_open(file, &jbd2_seq_info_ops); | |
1017 | if (rc == 0) { | |
1018 | struct seq_file *m = file->private_data; | |
1019 | m->private = s; | |
1020 | } else { | |
1021 | kfree(s->stats); | |
1022 | kfree(s); | |
1023 | } | |
1024 | return rc; | |
1025 | ||
1026 | } | |
1027 | ||
1028 | static int jbd2_seq_info_release(struct inode *inode, struct file *file) | |
1029 | { | |
1030 | struct seq_file *seq = file->private_data; | |
1031 | struct jbd2_stats_proc_session *s = seq->private; | |
1032 | kfree(s->stats); | |
1033 | kfree(s); | |
1034 | return seq_release(inode, file); | |
1035 | } | |
1036 | ||
828c0950 | 1037 | static const struct file_operations jbd2_seq_info_fops = { |
8e85fb3f JL |
1038 | .owner = THIS_MODULE, |
1039 | .open = jbd2_seq_info_open, | |
1040 | .read = seq_read, | |
1041 | .llseek = seq_lseek, | |
1042 | .release = jbd2_seq_info_release, | |
1043 | }; | |
1044 | ||
1045 | static struct proc_dir_entry *proc_jbd2_stats; | |
1046 | ||
1047 | static void jbd2_stats_proc_init(journal_t *journal) | |
1048 | { | |
05496769 | 1049 | journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); |
8e85fb3f | 1050 | if (journal->j_proc_entry) { |
79da3664 DL |
1051 | proc_create_data("info", S_IRUGO, journal->j_proc_entry, |
1052 | &jbd2_seq_info_fops, journal); | |
8e85fb3f JL |
1053 | } |
1054 | } | |
1055 | ||
1056 | static void jbd2_stats_proc_exit(journal_t *journal) | |
1057 | { | |
8e85fb3f | 1058 | remove_proc_entry("info", journal->j_proc_entry); |
05496769 | 1059 | remove_proc_entry(journal->j_devname, proc_jbd2_stats); |
8e85fb3f JL |
1060 | } |
1061 | ||
470decc6 DK |
1062 | /* |
1063 | * Management for journal control blocks: functions to create and | |
1064 | * destroy journal_t structures, and to initialise and read existing | |
1065 | * journal blocks from disk. */ | |
1066 | ||
1067 | /* First: create and setup a journal_t object in memory. We initialise | |
1068 | * very few fields yet: that has to wait until we have created the | |
1069 | * journal structures from from scratch, or loaded them from disk. */ | |
1070 | ||
1071 | static journal_t * journal_init_common (void) | |
1072 | { | |
1073 | journal_t *journal; | |
1074 | int err; | |
1075 | ||
3ebfdf88 | 1076 | journal = kzalloc(sizeof(*journal), GFP_KERNEL); |
470decc6 | 1077 | if (!journal) |
b7271b0a | 1078 | return NULL; |
470decc6 DK |
1079 | |
1080 | init_waitqueue_head(&journal->j_wait_transaction_locked); | |
470decc6 | 1081 | init_waitqueue_head(&journal->j_wait_done_commit); |
470decc6 DK |
1082 | init_waitqueue_head(&journal->j_wait_commit); |
1083 | init_waitqueue_head(&journal->j_wait_updates); | |
8f7d89f3 | 1084 | init_waitqueue_head(&journal->j_wait_reserved); |
470decc6 DK |
1085 | mutex_init(&journal->j_barrier); |
1086 | mutex_init(&journal->j_checkpoint_mutex); | |
1087 | spin_lock_init(&journal->j_revoke_lock); | |
1088 | spin_lock_init(&journal->j_list_lock); | |
a931da6a | 1089 | rwlock_init(&journal->j_state_lock); |
470decc6 | 1090 | |
cd02ff0b | 1091 | journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); |
30773840 TT |
1092 | journal->j_min_batch_time = 0; |
1093 | journal->j_max_batch_time = 15000; /* 15ms */ | |
8f7d89f3 | 1094 | atomic_set(&journal->j_reserved_credits, 0); |
470decc6 DK |
1095 | |
1096 | /* The journal is marked for error until we succeed with recovery! */ | |
f7f4bccb | 1097 | journal->j_flags = JBD2_ABORT; |
470decc6 DK |
1098 | |
1099 | /* Set up a default-sized revoke table for the new mount. */ | |
f7f4bccb | 1100 | err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); |
470decc6 DK |
1101 | if (err) { |
1102 | kfree(journal); | |
b7271b0a | 1103 | return NULL; |
470decc6 | 1104 | } |
8e85fb3f | 1105 | |
bf699327 | 1106 | spin_lock_init(&journal->j_history_lock); |
8e85fb3f | 1107 | |
470decc6 | 1108 | return journal; |
470decc6 DK |
1109 | } |
1110 | ||
f7f4bccb | 1111 | /* jbd2_journal_init_dev and jbd2_journal_init_inode: |
470decc6 DK |
1112 | * |
1113 | * Create a journal structure assigned some fixed set of disk blocks to | |
1114 | * the journal. We don't actually touch those disk blocks yet, but we | |
1115 | * need to set up all of the mapping information to tell the journaling | |
1116 | * system where the journal blocks are. | |
1117 | * | |
1118 | */ | |
1119 | ||
1120 | /** | |
5648ba5b | 1121 | * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure |
470decc6 DK |
1122 | * @bdev: Block device on which to create the journal |
1123 | * @fs_dev: Device which hold journalled filesystem for this journal. | |
1124 | * @start: Block nr Start of journal. | |
1125 | * @len: Length of the journal in blocks. | |
1126 | * @blocksize: blocksize of journalling device | |
5648ba5b RD |
1127 | * |
1128 | * Returns: a newly created journal_t * | |
470decc6 | 1129 | * |
f7f4bccb | 1130 | * jbd2_journal_init_dev creates a journal which maps a fixed contiguous |
470decc6 DK |
1131 | * range of blocks on an arbitrary block device. |
1132 | * | |
1133 | */ | |
f7f4bccb | 1134 | journal_t * jbd2_journal_init_dev(struct block_device *bdev, |
470decc6 | 1135 | struct block_device *fs_dev, |
18eba7aa | 1136 | unsigned long long start, int len, int blocksize) |
470decc6 DK |
1137 | { |
1138 | journal_t *journal = journal_init_common(); | |
1139 | struct buffer_head *bh; | |
05496769 | 1140 | char *p; |
470decc6 DK |
1141 | int n; |
1142 | ||
1143 | if (!journal) | |
1144 | return NULL; | |
1145 | ||
1146 | /* journal descriptor can store up to n blocks -bzzz */ | |
1147 | journal->j_blocksize = blocksize; | |
0587aa3d | 1148 | journal->j_dev = bdev; |
1149 | journal->j_fs_dev = fs_dev; | |
1150 | journal->j_blk_offset = start; | |
1151 | journal->j_maxlen = len; | |
1152 | bdevname(journal->j_dev, journal->j_devname); | |
1153 | p = journal->j_devname; | |
1154 | while ((p = strchr(p, '/'))) | |
1155 | *p = '!'; | |
4b905671 | 1156 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1157 | n = journal->j_blocksize / sizeof(journal_block_tag_t); |
1158 | journal->j_wbufsize = n; | |
1159 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1160 | if (!journal->j_wbuf) { | |
25985edc | 1161 | printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", |
329d291f | 1162 | __func__); |
4b905671 | 1163 | goto out_err; |
470decc6 | 1164 | } |
470decc6 DK |
1165 | |
1166 | bh = __getblk(journal->j_dev, start, journal->j_blocksize); | |
4b905671 JK |
1167 | if (!bh) { |
1168 | printk(KERN_ERR | |
1169 | "%s: Cannot get buffer for journal superblock\n", | |
1170 | __func__); | |
1171 | goto out_err; | |
1172 | } | |
470decc6 DK |
1173 | journal->j_sb_buffer = bh; |
1174 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
4b905671 | 1175 | |
470decc6 | 1176 | return journal; |
4b905671 | 1177 | out_err: |
7b02bec0 | 1178 | kfree(journal->j_wbuf); |
4b905671 JK |
1179 | jbd2_stats_proc_exit(journal); |
1180 | kfree(journal); | |
1181 | return NULL; | |
470decc6 DK |
1182 | } |
1183 | ||
1184 | /** | |
f7f4bccb | 1185 | * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. |
470decc6 DK |
1186 | * @inode: An inode to create the journal in |
1187 | * | |
f7f4bccb | 1188 | * jbd2_journal_init_inode creates a journal which maps an on-disk inode as |
470decc6 DK |
1189 | * the journal. The inode must exist already, must support bmap() and |
1190 | * must have all data blocks preallocated. | |
1191 | */ | |
f7f4bccb | 1192 | journal_t * jbd2_journal_init_inode (struct inode *inode) |
470decc6 DK |
1193 | { |
1194 | struct buffer_head *bh; | |
1195 | journal_t *journal = journal_init_common(); | |
05496769 | 1196 | char *p; |
470decc6 DK |
1197 | int err; |
1198 | int n; | |
18eba7aa | 1199 | unsigned long long blocknr; |
470decc6 DK |
1200 | |
1201 | if (!journal) | |
1202 | return NULL; | |
1203 | ||
1204 | journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev; | |
1205 | journal->j_inode = inode; | |
05496769 TT |
1206 | bdevname(journal->j_dev, journal->j_devname); |
1207 | p = journal->j_devname; | |
1208 | while ((p = strchr(p, '/'))) | |
1209 | *p = '!'; | |
1210 | p = journal->j_devname + strlen(journal->j_devname); | |
90576c0b | 1211 | sprintf(p, "-%lu", journal->j_inode->i_ino); |
470decc6 DK |
1212 | jbd_debug(1, |
1213 | "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n", | |
1214 | journal, inode->i_sb->s_id, inode->i_ino, | |
1215 | (long long) inode->i_size, | |
1216 | inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); | |
1217 | ||
1218 | journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits; | |
1219 | journal->j_blocksize = inode->i_sb->s_blocksize; | |
8e85fb3f | 1220 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1221 | |
1222 | /* journal descriptor can store up to n blocks -bzzz */ | |
1223 | n = journal->j_blocksize / sizeof(journal_block_tag_t); | |
1224 | journal->j_wbufsize = n; | |
1225 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1226 | if (!journal->j_wbuf) { | |
25985edc | 1227 | printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", |
329d291f | 1228 | __func__); |
4b905671 | 1229 | goto out_err; |
470decc6 DK |
1230 | } |
1231 | ||
f7f4bccb | 1232 | err = jbd2_journal_bmap(journal, 0, &blocknr); |
470decc6 DK |
1233 | /* If that failed, give up */ |
1234 | if (err) { | |
3c26bdb4 | 1235 | printk(KERN_ERR "%s: Cannot locate journal superblock\n", |
329d291f | 1236 | __func__); |
4b905671 | 1237 | goto out_err; |
470decc6 DK |
1238 | } |
1239 | ||
a49058fa | 1240 | bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize); |
4b905671 JK |
1241 | if (!bh) { |
1242 | printk(KERN_ERR | |
1243 | "%s: Cannot get buffer for journal superblock\n", | |
1244 | __func__); | |
1245 | goto out_err; | |
1246 | } | |
470decc6 DK |
1247 | journal->j_sb_buffer = bh; |
1248 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
1249 | ||
1250 | return journal; | |
4b905671 | 1251 | out_err: |
7b02bec0 | 1252 | kfree(journal->j_wbuf); |
4b905671 JK |
1253 | jbd2_stats_proc_exit(journal); |
1254 | kfree(journal); | |
1255 | return NULL; | |
470decc6 DK |
1256 | } |
1257 | ||
1258 | /* | |
1259 | * If the journal init or create aborts, we need to mark the journal | |
1260 | * superblock as being NULL to prevent the journal destroy from writing | |
1261 | * back a bogus superblock. | |
1262 | */ | |
1263 | static void journal_fail_superblock (journal_t *journal) | |
1264 | { | |
1265 | struct buffer_head *bh = journal->j_sb_buffer; | |
1266 | brelse(bh); | |
1267 | journal->j_sb_buffer = NULL; | |
1268 | } | |
1269 | ||
1270 | /* | |
1271 | * Given a journal_t structure, initialise the various fields for | |
1272 | * startup of a new journaling session. We use this both when creating | |
1273 | * a journal, and after recovering an old journal to reset it for | |
1274 | * subsequent use. | |
1275 | */ | |
1276 | ||
1277 | static int journal_reset(journal_t *journal) | |
1278 | { | |
1279 | journal_superblock_t *sb = journal->j_superblock; | |
18eba7aa | 1280 | unsigned long long first, last; |
470decc6 DK |
1281 | |
1282 | first = be32_to_cpu(sb->s_first); | |
1283 | last = be32_to_cpu(sb->s_maxlen); | |
f6f50e28 | 1284 | if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) { |
f2a44523 | 1285 | printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n", |
f6f50e28 JK |
1286 | first, last); |
1287 | journal_fail_superblock(journal); | |
1288 | return -EINVAL; | |
1289 | } | |
470decc6 DK |
1290 | |
1291 | journal->j_first = first; | |
1292 | journal->j_last = last; | |
1293 | ||
1294 | journal->j_head = first; | |
1295 | journal->j_tail = first; | |
1296 | journal->j_free = last - first; | |
1297 | ||
1298 | journal->j_tail_sequence = journal->j_transaction_sequence; | |
1299 | journal->j_commit_sequence = journal->j_transaction_sequence - 1; | |
1300 | journal->j_commit_request = journal->j_commit_sequence; | |
1301 | ||
1302 | journal->j_max_transaction_buffers = journal->j_maxlen / 4; | |
1303 | ||
470decc6 DK |
1304 | /* |
1305 | * As a special case, if the on-disk copy is already marked as needing | |
24bcc89c JK |
1306 | * no recovery (s_start == 0), then we can safely defer the superblock |
1307 | * update until the next commit by setting JBD2_FLUSHED. This avoids | |
470decc6 DK |
1308 | * attempting a write to a potential-readonly device. |
1309 | */ | |
24bcc89c | 1310 | if (sb->s_start == 0) { |
f2a44523 | 1311 | jbd_debug(1, "JBD2: Skipping superblock update on recovered sb " |
470decc6 DK |
1312 | "(start %ld, seq %d, errno %d)\n", |
1313 | journal->j_tail, journal->j_tail_sequence, | |
1314 | journal->j_errno); | |
24bcc89c JK |
1315 | journal->j_flags |= JBD2_FLUSHED; |
1316 | } else { | |
a78bb11d JK |
1317 | /* Lock here to make assertions happy... */ |
1318 | mutex_lock(&journal->j_checkpoint_mutex); | |
79feb521 JK |
1319 | /* |
1320 | * Update log tail information. We use WRITE_FUA since new | |
1321 | * transaction will start reusing journal space and so we | |
1322 | * must make sure information about current log tail is on | |
1323 | * disk before that. | |
1324 | */ | |
1325 | jbd2_journal_update_sb_log_tail(journal, | |
1326 | journal->j_tail_sequence, | |
1327 | journal->j_tail, | |
1328 | WRITE_FUA); | |
a78bb11d | 1329 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 | 1330 | } |
24bcc89c JK |
1331 | return jbd2_journal_start_thread(journal); |
1332 | } | |
470decc6 | 1333 | |
79feb521 | 1334 | static void jbd2_write_superblock(journal_t *journal, int write_op) |
24bcc89c JK |
1335 | { |
1336 | struct buffer_head *bh = journal->j_sb_buffer; | |
fe52d17c | 1337 | journal_superblock_t *sb = journal->j_superblock; |
79feb521 | 1338 | int ret; |
470decc6 | 1339 | |
79feb521 JK |
1340 | trace_jbd2_write_superblock(journal, write_op); |
1341 | if (!(journal->j_flags & JBD2_BARRIER)) | |
1342 | write_op &= ~(REQ_FUA | REQ_FLUSH); | |
1343 | lock_buffer(bh); | |
914258bf TT |
1344 | if (buffer_write_io_error(bh)) { |
1345 | /* | |
1346 | * Oh, dear. A previous attempt to write the journal | |
1347 | * superblock failed. This could happen because the | |
1348 | * USB device was yanked out. Or it could happen to | |
1349 | * be a transient write error and maybe the block will | |
1350 | * be remapped. Nothing we can do but to retry the | |
1351 | * write and hope for the best. | |
1352 | */ | |
1353 | printk(KERN_ERR "JBD2: previous I/O error detected " | |
1354 | "for journal superblock update for %s.\n", | |
1355 | journal->j_devname); | |
1356 | clear_buffer_write_io_error(bh); | |
1357 | set_buffer_uptodate(bh); | |
1358 | } | |
fe52d17c | 1359 | jbd2_superblock_csum_set(journal, sb); |
79feb521 JK |
1360 | get_bh(bh); |
1361 | bh->b_end_io = end_buffer_write_sync; | |
1362 | ret = submit_bh(write_op, bh); | |
1363 | wait_on_buffer(bh); | |
24bcc89c | 1364 | if (buffer_write_io_error(bh)) { |
24bcc89c JK |
1365 | clear_buffer_write_io_error(bh); |
1366 | set_buffer_uptodate(bh); | |
79feb521 JK |
1367 | ret = -EIO; |
1368 | } | |
1369 | if (ret) { | |
1370 | printk(KERN_ERR "JBD2: Error %d detected when updating " | |
1371 | "journal superblock for %s.\n", ret, | |
1372 | journal->j_devname); | |
24bcc89c JK |
1373 | } |
1374 | } | |
1375 | ||
1376 | /** | |
1377 | * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. | |
1378 | * @journal: The journal to update. | |
79feb521 JK |
1379 | * @tail_tid: TID of the new transaction at the tail of the log |
1380 | * @tail_block: The first block of the transaction at the tail of the log | |
1381 | * @write_op: With which operation should we write the journal sb | |
24bcc89c JK |
1382 | * |
1383 | * Update a journal's superblock information about log tail and write it to | |
1384 | * disk, waiting for the IO to complete. | |
1385 | */ | |
79feb521 JK |
1386 | void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, |
1387 | unsigned long tail_block, int write_op) | |
24bcc89c JK |
1388 | { |
1389 | journal_superblock_t *sb = journal->j_superblock; | |
1390 | ||
a78bb11d | 1391 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); |
79feb521 JK |
1392 | jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", |
1393 | tail_block, tail_tid); | |
470decc6 | 1394 | |
79feb521 JK |
1395 | sb->s_sequence = cpu_to_be32(tail_tid); |
1396 | sb->s_start = cpu_to_be32(tail_block); | |
470decc6 | 1397 | |
79feb521 | 1398 | jbd2_write_superblock(journal, write_op); |
470decc6 | 1399 | |
24bcc89c JK |
1400 | /* Log is no longer empty */ |
1401 | write_lock(&journal->j_state_lock); | |
1402 | WARN_ON(!sb->s_sequence); | |
1403 | journal->j_flags &= ~JBD2_FLUSHED; | |
1404 | write_unlock(&journal->j_state_lock); | |
1405 | } | |
2201c590 | 1406 | |
24bcc89c JK |
1407 | /** |
1408 | * jbd2_mark_journal_empty() - Mark on disk journal as empty. | |
1409 | * @journal: The journal to update. | |
1410 | * | |
1411 | * Update a journal's dynamic superblock fields to show that journal is empty. | |
1412 | * Write updated superblock to disk waiting for IO to complete. | |
1413 | */ | |
1414 | static void jbd2_mark_journal_empty(journal_t *journal) | |
1415 | { | |
1416 | journal_superblock_t *sb = journal->j_superblock; | |
914258bf | 1417 | |
a78bb11d | 1418 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); |
a931da6a | 1419 | read_lock(&journal->j_state_lock); |
eeecef0a ES |
1420 | /* Is it already empty? */ |
1421 | if (sb->s_start == 0) { | |
1422 | read_unlock(&journal->j_state_lock); | |
1423 | return; | |
1424 | } | |
24bcc89c JK |
1425 | jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n", |
1426 | journal->j_tail_sequence); | |
470decc6 DK |
1427 | |
1428 | sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); | |
24bcc89c | 1429 | sb->s_start = cpu_to_be32(0); |
a931da6a | 1430 | read_unlock(&journal->j_state_lock); |
470decc6 | 1431 | |
79feb521 | 1432 | jbd2_write_superblock(journal, WRITE_FUA); |
470decc6 | 1433 | |
24bcc89c | 1434 | /* Log is no longer empty */ |
a931da6a | 1435 | write_lock(&journal->j_state_lock); |
24bcc89c | 1436 | journal->j_flags |= JBD2_FLUSHED; |
a931da6a | 1437 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1438 | } |
1439 | ||
24bcc89c JK |
1440 | |
1441 | /** | |
1442 | * jbd2_journal_update_sb_errno() - Update error in the journal. | |
1443 | * @journal: The journal to update. | |
1444 | * | |
1445 | * Update a journal's errno. Write updated superblock to disk waiting for IO | |
1446 | * to complete. | |
1447 | */ | |
d796c52e | 1448 | void jbd2_journal_update_sb_errno(journal_t *journal) |
24bcc89c JK |
1449 | { |
1450 | journal_superblock_t *sb = journal->j_superblock; | |
1451 | ||
1452 | read_lock(&journal->j_state_lock); | |
1453 | jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", | |
1454 | journal->j_errno); | |
1455 | sb->s_errno = cpu_to_be32(journal->j_errno); | |
1456 | read_unlock(&journal->j_state_lock); | |
1457 | ||
79feb521 | 1458 | jbd2_write_superblock(journal, WRITE_SYNC); |
24bcc89c | 1459 | } |
d796c52e | 1460 | EXPORT_SYMBOL(jbd2_journal_update_sb_errno); |
24bcc89c | 1461 | |
470decc6 DK |
1462 | /* |
1463 | * Read the superblock for a given journal, performing initial | |
1464 | * validation of the format. | |
1465 | */ | |
470decc6 DK |
1466 | static int journal_get_superblock(journal_t *journal) |
1467 | { | |
1468 | struct buffer_head *bh; | |
1469 | journal_superblock_t *sb; | |
1470 | int err = -EIO; | |
1471 | ||
1472 | bh = journal->j_sb_buffer; | |
1473 | ||
1474 | J_ASSERT(bh != NULL); | |
1475 | if (!buffer_uptodate(bh)) { | |
1476 | ll_rw_block(READ, 1, &bh); | |
1477 | wait_on_buffer(bh); | |
1478 | if (!buffer_uptodate(bh)) { | |
f2a44523 EG |
1479 | printk(KERN_ERR |
1480 | "JBD2: IO error reading journal superblock\n"); | |
470decc6 DK |
1481 | goto out; |
1482 | } | |
1483 | } | |
1484 | ||
25ed6e8a DW |
1485 | if (buffer_verified(bh)) |
1486 | return 0; | |
1487 | ||
470decc6 DK |
1488 | sb = journal->j_superblock; |
1489 | ||
1490 | err = -EINVAL; | |
1491 | ||
f7f4bccb | 1492 | if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || |
470decc6 | 1493 | sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { |
f2a44523 | 1494 | printk(KERN_WARNING "JBD2: no valid journal superblock found\n"); |
470decc6 DK |
1495 | goto out; |
1496 | } | |
1497 | ||
1498 | switch(be32_to_cpu(sb->s_header.h_blocktype)) { | |
f7f4bccb | 1499 | case JBD2_SUPERBLOCK_V1: |
470decc6 DK |
1500 | journal->j_format_version = 1; |
1501 | break; | |
f7f4bccb | 1502 | case JBD2_SUPERBLOCK_V2: |
470decc6 DK |
1503 | journal->j_format_version = 2; |
1504 | break; | |
1505 | default: | |
f2a44523 | 1506 | printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n"); |
470decc6 DK |
1507 | goto out; |
1508 | } | |
1509 | ||
1510 | if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen) | |
1511 | journal->j_maxlen = be32_to_cpu(sb->s_maxlen); | |
1512 | else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) { | |
f2a44523 | 1513 | printk(KERN_WARNING "JBD2: journal file too short\n"); |
470decc6 DK |
1514 | goto out; |
1515 | } | |
1516 | ||
8762202d EG |
1517 | if (be32_to_cpu(sb->s_first) == 0 || |
1518 | be32_to_cpu(sb->s_first) >= journal->j_maxlen) { | |
1519 | printk(KERN_WARNING | |
1520 | "JBD2: Invalid start block of journal: %u\n", | |
1521 | be32_to_cpu(sb->s_first)); | |
1522 | goto out; | |
1523 | } | |
1524 | ||
db9ee220 DW |
1525 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2) && |
1526 | JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3)) { | |
1527 | /* Can't have checksum v2 and v3 at the same time! */ | |
1528 | printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 " | |
1529 | "at the same time!\n"); | |
1530 | goto out; | |
1531 | } | |
1532 | ||
feb8c6d3 DW |
1533 | if (jbd2_journal_has_csum_v2or3(journal) && |
1534 | JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM)) { | |
1535 | /* Can't have checksum v1 and v2 on at the same time! */ | |
1536 | printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 " | |
1537 | "at the same time!\n"); | |
1538 | goto out; | |
1539 | } | |
1540 | ||
25ed6e8a | 1541 | if (!jbd2_verify_csum_type(journal, sb)) { |
a67c848a | 1542 | printk(KERN_ERR "JBD2: Unknown checksum type\n"); |
25ed6e8a DW |
1543 | goto out; |
1544 | } | |
1545 | ||
01b5adce | 1546 | /* Load the checksum driver */ |
db9ee220 | 1547 | if (jbd2_journal_has_csum_v2or3(journal)) { |
01b5adce DW |
1548 | journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); |
1549 | if (IS_ERR(journal->j_chksum_driver)) { | |
a67c848a | 1550 | printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n"); |
01b5adce DW |
1551 | err = PTR_ERR(journal->j_chksum_driver); |
1552 | journal->j_chksum_driver = NULL; | |
1553 | goto out; | |
1554 | } | |
1555 | } | |
1556 | ||
4fd5ea43 DW |
1557 | /* Check superblock checksum */ |
1558 | if (!jbd2_superblock_csum_verify(journal, sb)) { | |
a67c848a | 1559 | printk(KERN_ERR "JBD2: journal checksum error\n"); |
4fd5ea43 DW |
1560 | goto out; |
1561 | } | |
1562 | ||
1563 | /* Precompute checksum seed for all metadata */ | |
db9ee220 | 1564 | if (jbd2_journal_has_csum_v2or3(journal)) |
4fd5ea43 DW |
1565 | journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, |
1566 | sizeof(sb->s_uuid)); | |
1567 | ||
25ed6e8a DW |
1568 | set_buffer_verified(bh); |
1569 | ||
470decc6 DK |
1570 | return 0; |
1571 | ||
1572 | out: | |
1573 | journal_fail_superblock(journal); | |
1574 | return err; | |
1575 | } | |
1576 | ||
1577 | /* | |
1578 | * Load the on-disk journal superblock and read the key fields into the | |
1579 | * journal_t. | |
1580 | */ | |
1581 | ||
1582 | static int load_superblock(journal_t *journal) | |
1583 | { | |
1584 | int err; | |
1585 | journal_superblock_t *sb; | |
1586 | ||
1587 | err = journal_get_superblock(journal); | |
1588 | if (err) | |
1589 | return err; | |
1590 | ||
1591 | sb = journal->j_superblock; | |
1592 | ||
1593 | journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); | |
1594 | journal->j_tail = be32_to_cpu(sb->s_start); | |
1595 | journal->j_first = be32_to_cpu(sb->s_first); | |
1596 | journal->j_last = be32_to_cpu(sb->s_maxlen); | |
1597 | journal->j_errno = be32_to_cpu(sb->s_errno); | |
1598 | ||
1599 | return 0; | |
1600 | } | |
1601 | ||
1602 | ||
1603 | /** | |
f7f4bccb | 1604 | * int jbd2_journal_load() - Read journal from disk. |
470decc6 DK |
1605 | * @journal: Journal to act on. |
1606 | * | |
1607 | * Given a journal_t structure which tells us which disk blocks contain | |
1608 | * a journal, read the journal from disk to initialise the in-memory | |
1609 | * structures. | |
1610 | */ | |
f7f4bccb | 1611 | int jbd2_journal_load(journal_t *journal) |
470decc6 DK |
1612 | { |
1613 | int err; | |
1614 | journal_superblock_t *sb; | |
1615 | ||
1616 | err = load_superblock(journal); | |
1617 | if (err) | |
1618 | return err; | |
1619 | ||
1620 | sb = journal->j_superblock; | |
1621 | /* If this is a V2 superblock, then we have to check the | |
1622 | * features flags on it. */ | |
1623 | ||
1624 | if (journal->j_format_version >= 2) { | |
1625 | if ((sb->s_feature_ro_compat & | |
f7f4bccb | 1626 | ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || |
470decc6 | 1627 | (sb->s_feature_incompat & |
f7f4bccb | 1628 | ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { |
f2a44523 EG |
1629 | printk(KERN_WARNING |
1630 | "JBD2: Unrecognised features on journal\n"); | |
470decc6 DK |
1631 | return -EINVAL; |
1632 | } | |
1633 | } | |
1634 | ||
d2eecb03 TT |
1635 | /* |
1636 | * Create a slab for this blocksize | |
1637 | */ | |
1638 | err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize)); | |
1639 | if (err) | |
1640 | return err; | |
1641 | ||
470decc6 DK |
1642 | /* Let the recovery code check whether it needs to recover any |
1643 | * data from the journal. */ | |
f7f4bccb | 1644 | if (jbd2_journal_recover(journal)) |
470decc6 DK |
1645 | goto recovery_error; |
1646 | ||
e6a47428 TT |
1647 | if (journal->j_failed_commit) { |
1648 | printk(KERN_ERR "JBD2: journal transaction %u on %s " | |
1649 | "is corrupt.\n", journal->j_failed_commit, | |
1650 | journal->j_devname); | |
1651 | return -EIO; | |
1652 | } | |
1653 | ||
470decc6 DK |
1654 | /* OK, we've finished with the dynamic journal bits: |
1655 | * reinitialise the dynamic contents of the superblock in memory | |
1656 | * and reset them on disk. */ | |
1657 | if (journal_reset(journal)) | |
1658 | goto recovery_error; | |
1659 | ||
f7f4bccb MC |
1660 | journal->j_flags &= ~JBD2_ABORT; |
1661 | journal->j_flags |= JBD2_LOADED; | |
470decc6 DK |
1662 | return 0; |
1663 | ||
1664 | recovery_error: | |
f2a44523 | 1665 | printk(KERN_WARNING "JBD2: recovery failed\n"); |
470decc6 DK |
1666 | return -EIO; |
1667 | } | |
1668 | ||
1669 | /** | |
f7f4bccb | 1670 | * void jbd2_journal_destroy() - Release a journal_t structure. |
470decc6 DK |
1671 | * @journal: Journal to act on. |
1672 | * | |
1673 | * Release a journal_t structure once it is no longer in use by the | |
1674 | * journaled object. | |
44519faf | 1675 | * Return <0 if we couldn't clean up the journal. |
470decc6 | 1676 | */ |
44519faf | 1677 | int jbd2_journal_destroy(journal_t *journal) |
470decc6 | 1678 | { |
44519faf HK |
1679 | int err = 0; |
1680 | ||
470decc6 DK |
1681 | /* Wait for the commit thread to wake up and die. */ |
1682 | journal_kill_thread(journal); | |
1683 | ||
1684 | /* Force a final log commit */ | |
1685 | if (journal->j_running_transaction) | |
f7f4bccb | 1686 | jbd2_journal_commit_transaction(journal); |
470decc6 DK |
1687 | |
1688 | /* Force any old transactions to disk */ | |
1689 | ||
1690 | /* Totally anal locking here... */ | |
1691 | spin_lock(&journal->j_list_lock); | |
1692 | while (journal->j_checkpoint_transactions != NULL) { | |
1693 | spin_unlock(&journal->j_list_lock); | |
1a0d3786 | 1694 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1695 | jbd2_log_do_checkpoint(journal); |
1a0d3786 | 1696 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
1697 | spin_lock(&journal->j_list_lock); |
1698 | } | |
1699 | ||
1700 | J_ASSERT(journal->j_running_transaction == NULL); | |
1701 | J_ASSERT(journal->j_committing_transaction == NULL); | |
1702 | J_ASSERT(journal->j_checkpoint_transactions == NULL); | |
1703 | spin_unlock(&journal->j_list_lock); | |
1704 | ||
470decc6 | 1705 | if (journal->j_sb_buffer) { |
44519faf | 1706 | if (!is_journal_aborted(journal)) { |
a78bb11d | 1707 | mutex_lock(&journal->j_checkpoint_mutex); |
24bcc89c | 1708 | jbd2_mark_journal_empty(journal); |
a78bb11d JK |
1709 | mutex_unlock(&journal->j_checkpoint_mutex); |
1710 | } else | |
44519faf | 1711 | err = -EIO; |
470decc6 DK |
1712 | brelse(journal->j_sb_buffer); |
1713 | } | |
1714 | ||
8e85fb3f JL |
1715 | if (journal->j_proc_entry) |
1716 | jbd2_stats_proc_exit(journal); | |
d9f39d1e | 1717 | iput(journal->j_inode); |
470decc6 | 1718 | if (journal->j_revoke) |
f7f4bccb | 1719 | jbd2_journal_destroy_revoke(journal); |
01b5adce DW |
1720 | if (journal->j_chksum_driver) |
1721 | crypto_free_shash(journal->j_chksum_driver); | |
470decc6 DK |
1722 | kfree(journal->j_wbuf); |
1723 | kfree(journal); | |
44519faf HK |
1724 | |
1725 | return err; | |
470decc6 DK |
1726 | } |
1727 | ||
1728 | ||
1729 | /** | |
f7f4bccb | 1730 | *int jbd2_journal_check_used_features () - Check if features specified are used. |
470decc6 DK |
1731 | * @journal: Journal to check. |
1732 | * @compat: bitmask of compatible features | |
1733 | * @ro: bitmask of features that force read-only mount | |
1734 | * @incompat: bitmask of incompatible features | |
1735 | * | |
1736 | * Check whether the journal uses all of a given set of | |
1737 | * features. Return true (non-zero) if it does. | |
1738 | **/ | |
1739 | ||
f7f4bccb | 1740 | int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1741 | unsigned long ro, unsigned long incompat) |
1742 | { | |
1743 | journal_superblock_t *sb; | |
1744 | ||
1745 | if (!compat && !ro && !incompat) | |
1746 | return 1; | |
1113e1b5 PL |
1747 | /* Load journal superblock if it is not loaded yet. */ |
1748 | if (journal->j_format_version == 0 && | |
1749 | journal_get_superblock(journal) != 0) | |
1750 | return 0; | |
470decc6 DK |
1751 | if (journal->j_format_version == 1) |
1752 | return 0; | |
1753 | ||
1754 | sb = journal->j_superblock; | |
1755 | ||
1756 | if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && | |
1757 | ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && | |
1758 | ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) | |
1759 | return 1; | |
1760 | ||
1761 | return 0; | |
1762 | } | |
1763 | ||
1764 | /** | |
f7f4bccb | 1765 | * int jbd2_journal_check_available_features() - Check feature set in journalling layer |
470decc6 DK |
1766 | * @journal: Journal to check. |
1767 | * @compat: bitmask of compatible features | |
1768 | * @ro: bitmask of features that force read-only mount | |
1769 | * @incompat: bitmask of incompatible features | |
1770 | * | |
1771 | * Check whether the journaling code supports the use of | |
1772 | * all of a given set of features on this journal. Return true | |
1773 | * (non-zero) if it can. */ | |
1774 | ||
f7f4bccb | 1775 | int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1776 | unsigned long ro, unsigned long incompat) |
1777 | { | |
470decc6 DK |
1778 | if (!compat && !ro && !incompat) |
1779 | return 1; | |
1780 | ||
470decc6 DK |
1781 | /* We can support any known requested features iff the |
1782 | * superblock is in version 2. Otherwise we fail to support any | |
1783 | * extended sb features. */ | |
1784 | ||
1785 | if (journal->j_format_version != 2) | |
1786 | return 0; | |
1787 | ||
f7f4bccb MC |
1788 | if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && |
1789 | (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && | |
1790 | (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) | |
470decc6 DK |
1791 | return 1; |
1792 | ||
1793 | return 0; | |
1794 | } | |
1795 | ||
1796 | /** | |
f7f4bccb | 1797 | * int jbd2_journal_set_features () - Mark a given journal feature in the superblock |
470decc6 DK |
1798 | * @journal: Journal to act on. |
1799 | * @compat: bitmask of compatible features | |
1800 | * @ro: bitmask of features that force read-only mount | |
1801 | * @incompat: bitmask of incompatible features | |
1802 | * | |
1803 | * Mark a given journal feature as present on the | |
1804 | * superblock. Returns true if the requested features could be set. | |
1805 | * | |
1806 | */ | |
1807 | ||
f7f4bccb | 1808 | int jbd2_journal_set_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1809 | unsigned long ro, unsigned long incompat) |
1810 | { | |
25ed6e8a DW |
1811 | #define INCOMPAT_FEATURE_ON(f) \ |
1812 | ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f))) | |
1813 | #define COMPAT_FEATURE_ON(f) \ | |
1814 | ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f))) | |
470decc6 DK |
1815 | journal_superblock_t *sb; |
1816 | ||
f7f4bccb | 1817 | if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) |
470decc6 DK |
1818 | return 1; |
1819 | ||
f7f4bccb | 1820 | if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) |
470decc6 DK |
1821 | return 0; |
1822 | ||
db9ee220 DW |
1823 | /* If enabling v2 checksums, turn on v3 instead */ |
1824 | if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) { | |
1825 | incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2; | |
1826 | incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3; | |
1827 | } | |
1828 | ||
1829 | /* Asking for checksumming v3 and v1? Only give them v3. */ | |
1830 | if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 && | |
25ed6e8a DW |
1831 | compat & JBD2_FEATURE_COMPAT_CHECKSUM) |
1832 | compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM; | |
1833 | ||
470decc6 DK |
1834 | jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", |
1835 | compat, ro, incompat); | |
1836 | ||
1837 | sb = journal->j_superblock; | |
1838 | ||
db9ee220 DW |
1839 | /* If enabling v3 checksums, update superblock */ |
1840 | if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { | |
25ed6e8a DW |
1841 | sb->s_checksum_type = JBD2_CRC32C_CHKSUM; |
1842 | sb->s_feature_compat &= | |
1843 | ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM); | |
01b5adce DW |
1844 | |
1845 | /* Load the checksum driver */ | |
1846 | if (journal->j_chksum_driver == NULL) { | |
1847 | journal->j_chksum_driver = crypto_alloc_shash("crc32c", | |
1848 | 0, 0); | |
1849 | if (IS_ERR(journal->j_chksum_driver)) { | |
a67c848a | 1850 | printk(KERN_ERR "JBD2: Cannot load crc32c " |
01b5adce DW |
1851 | "driver.\n"); |
1852 | journal->j_chksum_driver = NULL; | |
1853 | return 0; | |
1854 | } | |
4fd5ea43 | 1855 | |
32f38691 | 1856 | /* Precompute checksum seed for all metadata */ |
4fd5ea43 DW |
1857 | journal->j_csum_seed = jbd2_chksum(journal, ~0, |
1858 | sb->s_uuid, | |
1859 | sizeof(sb->s_uuid)); | |
32f38691 | 1860 | } |
25ed6e8a DW |
1861 | } |
1862 | ||
1863 | /* If enabling v1 checksums, downgrade superblock */ | |
1864 | if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM)) | |
1865 | sb->s_feature_incompat &= | |
db9ee220 DW |
1866 | ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 | |
1867 | JBD2_FEATURE_INCOMPAT_CSUM_V3); | |
25ed6e8a | 1868 | |
470decc6 DK |
1869 | sb->s_feature_compat |= cpu_to_be32(compat); |
1870 | sb->s_feature_ro_compat |= cpu_to_be32(ro); | |
1871 | sb->s_feature_incompat |= cpu_to_be32(incompat); | |
1872 | ||
1873 | return 1; | |
25ed6e8a DW |
1874 | #undef COMPAT_FEATURE_ON |
1875 | #undef INCOMPAT_FEATURE_ON | |
470decc6 DK |
1876 | } |
1877 | ||
818d276c GS |
1878 | /* |
1879 | * jbd2_journal_clear_features () - Clear a given journal feature in the | |
1880 | * superblock | |
1881 | * @journal: Journal to act on. | |
1882 | * @compat: bitmask of compatible features | |
1883 | * @ro: bitmask of features that force read-only mount | |
1884 | * @incompat: bitmask of incompatible features | |
1885 | * | |
1886 | * Clear a given journal feature as present on the | |
1887 | * superblock. | |
1888 | */ | |
1889 | void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, | |
1890 | unsigned long ro, unsigned long incompat) | |
1891 | { | |
1892 | journal_superblock_t *sb; | |
1893 | ||
1894 | jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", | |
1895 | compat, ro, incompat); | |
1896 | ||
1897 | sb = journal->j_superblock; | |
1898 | ||
1899 | sb->s_feature_compat &= ~cpu_to_be32(compat); | |
1900 | sb->s_feature_ro_compat &= ~cpu_to_be32(ro); | |
1901 | sb->s_feature_incompat &= ~cpu_to_be32(incompat); | |
1902 | } | |
1903 | EXPORT_SYMBOL(jbd2_journal_clear_features); | |
470decc6 | 1904 | |
470decc6 | 1905 | /** |
f7f4bccb | 1906 | * int jbd2_journal_flush () - Flush journal |
470decc6 DK |
1907 | * @journal: Journal to act on. |
1908 | * | |
1909 | * Flush all data for a given journal to disk and empty the journal. | |
1910 | * Filesystems can use this when remounting readonly to ensure that | |
1911 | * recovery does not need to happen on remount. | |
1912 | */ | |
1913 | ||
f7f4bccb | 1914 | int jbd2_journal_flush(journal_t *journal) |
470decc6 DK |
1915 | { |
1916 | int err = 0; | |
1917 | transaction_t *transaction = NULL; | |
470decc6 | 1918 | |
a931da6a | 1919 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1920 | |
1921 | /* Force everything buffered to the log... */ | |
1922 | if (journal->j_running_transaction) { | |
1923 | transaction = journal->j_running_transaction; | |
f7f4bccb | 1924 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1925 | } else if (journal->j_committing_transaction) |
1926 | transaction = journal->j_committing_transaction; | |
1927 | ||
1928 | /* Wait for the log commit to complete... */ | |
1929 | if (transaction) { | |
1930 | tid_t tid = transaction->t_tid; | |
1931 | ||
a931da6a | 1932 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 1933 | jbd2_log_wait_commit(journal, tid); |
470decc6 | 1934 | } else { |
a931da6a | 1935 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1936 | } |
1937 | ||
1938 | /* ...and flush everything in the log out to disk. */ | |
1939 | spin_lock(&journal->j_list_lock); | |
1940 | while (!err && journal->j_checkpoint_transactions != NULL) { | |
1941 | spin_unlock(&journal->j_list_lock); | |
44519faf | 1942 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1943 | err = jbd2_log_do_checkpoint(journal); |
44519faf | 1944 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
1945 | spin_lock(&journal->j_list_lock); |
1946 | } | |
1947 | spin_unlock(&journal->j_list_lock); | |
44519faf HK |
1948 | |
1949 | if (is_journal_aborted(journal)) | |
1950 | return -EIO; | |
1951 | ||
a78bb11d | 1952 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1953 | jbd2_cleanup_journal_tail(journal); |
470decc6 DK |
1954 | |
1955 | /* Finally, mark the journal as really needing no recovery. | |
1956 | * This sets s_start==0 in the underlying superblock, which is | |
1957 | * the magic code for a fully-recovered superblock. Any future | |
1958 | * commits of data to the journal will restore the current | |
1959 | * s_start value. */ | |
24bcc89c | 1960 | jbd2_mark_journal_empty(journal); |
a78bb11d | 1961 | mutex_unlock(&journal->j_checkpoint_mutex); |
a931da6a | 1962 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1963 | J_ASSERT(!journal->j_running_transaction); |
1964 | J_ASSERT(!journal->j_committing_transaction); | |
1965 | J_ASSERT(!journal->j_checkpoint_transactions); | |
1966 | J_ASSERT(journal->j_head == journal->j_tail); | |
1967 | J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); | |
a931da6a | 1968 | write_unlock(&journal->j_state_lock); |
44519faf | 1969 | return 0; |
470decc6 DK |
1970 | } |
1971 | ||
1972 | /** | |
f7f4bccb | 1973 | * int jbd2_journal_wipe() - Wipe journal contents |
470decc6 DK |
1974 | * @journal: Journal to act on. |
1975 | * @write: flag (see below) | |
1976 | * | |
1977 | * Wipe out all of the contents of a journal, safely. This will produce | |
1978 | * a warning if the journal contains any valid recovery information. | |
f7f4bccb | 1979 | * Must be called between journal_init_*() and jbd2_journal_load(). |
470decc6 DK |
1980 | * |
1981 | * If 'write' is non-zero, then we wipe out the journal on disk; otherwise | |
1982 | * we merely suppress recovery. | |
1983 | */ | |
1984 | ||
f7f4bccb | 1985 | int jbd2_journal_wipe(journal_t *journal, int write) |
470decc6 | 1986 | { |
470decc6 DK |
1987 | int err = 0; |
1988 | ||
f7f4bccb | 1989 | J_ASSERT (!(journal->j_flags & JBD2_LOADED)); |
470decc6 DK |
1990 | |
1991 | err = load_superblock(journal); | |
1992 | if (err) | |
1993 | return err; | |
1994 | ||
470decc6 DK |
1995 | if (!journal->j_tail) |
1996 | goto no_recovery; | |
1997 | ||
f2a44523 | 1998 | printk(KERN_WARNING "JBD2: %s recovery information on journal\n", |
470decc6 DK |
1999 | write ? "Clearing" : "Ignoring"); |
2000 | ||
f7f4bccb | 2001 | err = jbd2_journal_skip_recovery(journal); |
a78bb11d JK |
2002 | if (write) { |
2003 | /* Lock to make assertions happy... */ | |
2004 | mutex_lock(&journal->j_checkpoint_mutex); | |
24bcc89c | 2005 | jbd2_mark_journal_empty(journal); |
a78bb11d JK |
2006 | mutex_unlock(&journal->j_checkpoint_mutex); |
2007 | } | |
470decc6 DK |
2008 | |
2009 | no_recovery: | |
2010 | return err; | |
2011 | } | |
2012 | ||
470decc6 DK |
2013 | /* |
2014 | * Journal abort has very specific semantics, which we describe | |
2015 | * for journal abort. | |
2016 | * | |
bfcd3555 | 2017 | * Two internal functions, which provide abort to the jbd layer |
470decc6 DK |
2018 | * itself are here. |
2019 | */ | |
2020 | ||
2021 | /* | |
2022 | * Quick version for internal journal use (doesn't lock the journal). | |
2023 | * Aborts hard --- we mark the abort as occurred, but do _nothing_ else, | |
2024 | * and don't attempt to make any other journal updates. | |
2025 | */ | |
f7f4bccb | 2026 | void __jbd2_journal_abort_hard(journal_t *journal) |
470decc6 DK |
2027 | { |
2028 | transaction_t *transaction; | |
470decc6 | 2029 | |
f7f4bccb | 2030 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2031 | return; |
2032 | ||
2033 | printk(KERN_ERR "Aborting journal on device %s.\n", | |
05496769 | 2034 | journal->j_devname); |
470decc6 | 2035 | |
a931da6a | 2036 | write_lock(&journal->j_state_lock); |
f7f4bccb | 2037 | journal->j_flags |= JBD2_ABORT; |
470decc6 DK |
2038 | transaction = journal->j_running_transaction; |
2039 | if (transaction) | |
f7f4bccb | 2040 | __jbd2_log_start_commit(journal, transaction->t_tid); |
a931da6a | 2041 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2042 | } |
2043 | ||
2044 | /* Soft abort: record the abort error status in the journal superblock, | |
2045 | * but don't do any other IO. */ | |
2046 | static void __journal_abort_soft (journal_t *journal, int errno) | |
2047 | { | |
f7f4bccb | 2048 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2049 | return; |
2050 | ||
2051 | if (!journal->j_errno) | |
2052 | journal->j_errno = errno; | |
2053 | ||
f7f4bccb | 2054 | __jbd2_journal_abort_hard(journal); |
470decc6 DK |
2055 | |
2056 | if (errno) | |
24bcc89c | 2057 | jbd2_journal_update_sb_errno(journal); |
470decc6 DK |
2058 | } |
2059 | ||
2060 | /** | |
f7f4bccb | 2061 | * void jbd2_journal_abort () - Shutdown the journal immediately. |
470decc6 DK |
2062 | * @journal: the journal to shutdown. |
2063 | * @errno: an error number to record in the journal indicating | |
2064 | * the reason for the shutdown. | |
2065 | * | |
2066 | * Perform a complete, immediate shutdown of the ENTIRE | |
2067 | * journal (not of a single transaction). This operation cannot be | |
2068 | * undone without closing and reopening the journal. | |
2069 | * | |
f7f4bccb | 2070 | * The jbd2_journal_abort function is intended to support higher level error |
470decc6 DK |
2071 | * recovery mechanisms such as the ext2/ext3 remount-readonly error |
2072 | * mode. | |
2073 | * | |
2074 | * Journal abort has very specific semantics. Any existing dirty, | |
2075 | * unjournaled buffers in the main filesystem will still be written to | |
2076 | * disk by bdflush, but the journaling mechanism will be suspended | |
2077 | * immediately and no further transaction commits will be honoured. | |
2078 | * | |
2079 | * Any dirty, journaled buffers will be written back to disk without | |
2080 | * hitting the journal. Atomicity cannot be guaranteed on an aborted | |
2081 | * filesystem, but we _do_ attempt to leave as much data as possible | |
2082 | * behind for fsck to use for cleanup. | |
2083 | * | |
2084 | * Any attempt to get a new transaction handle on a journal which is in | |
2085 | * ABORT state will just result in an -EROFS error return. A | |
f7f4bccb | 2086 | * jbd2_journal_stop on an existing handle will return -EIO if we have |
470decc6 DK |
2087 | * entered abort state during the update. |
2088 | * | |
2089 | * Recursive transactions are not disturbed by journal abort until the | |
f7f4bccb | 2090 | * final jbd2_journal_stop, which will receive the -EIO error. |
470decc6 | 2091 | * |
f7f4bccb | 2092 | * Finally, the jbd2_journal_abort call allows the caller to supply an errno |
470decc6 DK |
2093 | * which will be recorded (if possible) in the journal superblock. This |
2094 | * allows a client to record failure conditions in the middle of a | |
2095 | * transaction without having to complete the transaction to record the | |
2096 | * failure to disk. ext3_error, for example, now uses this | |
2097 | * functionality. | |
2098 | * | |
2099 | * Errors which originate from within the journaling layer will NOT | |
2100 | * supply an errno; a null errno implies that absolutely no further | |
2101 | * writes are done to the journal (unless there are any already in | |
2102 | * progress). | |
2103 | * | |
2104 | */ | |
2105 | ||
f7f4bccb | 2106 | void jbd2_journal_abort(journal_t *journal, int errno) |
470decc6 DK |
2107 | { |
2108 | __journal_abort_soft(journal, errno); | |
2109 | } | |
2110 | ||
2111 | /** | |
f7f4bccb | 2112 | * int jbd2_journal_errno () - returns the journal's error state. |
470decc6 DK |
2113 | * @journal: journal to examine. |
2114 | * | |
bfcd3555 | 2115 | * This is the errno number set with jbd2_journal_abort(), the last |
470decc6 DK |
2116 | * time the journal was mounted - if the journal was stopped |
2117 | * without calling abort this will be 0. | |
2118 | * | |
2119 | * If the journal has been aborted on this mount time -EROFS will | |
2120 | * be returned. | |
2121 | */ | |
f7f4bccb | 2122 | int jbd2_journal_errno(journal_t *journal) |
470decc6 DK |
2123 | { |
2124 | int err; | |
2125 | ||
a931da6a | 2126 | read_lock(&journal->j_state_lock); |
f7f4bccb | 2127 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2128 | err = -EROFS; |
2129 | else | |
2130 | err = journal->j_errno; | |
a931da6a | 2131 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
2132 | return err; |
2133 | } | |
2134 | ||
2135 | /** | |
f7f4bccb | 2136 | * int jbd2_journal_clear_err () - clears the journal's error state |
470decc6 DK |
2137 | * @journal: journal to act on. |
2138 | * | |
bfcd3555 | 2139 | * An error must be cleared or acked to take a FS out of readonly |
470decc6 DK |
2140 | * mode. |
2141 | */ | |
f7f4bccb | 2142 | int jbd2_journal_clear_err(journal_t *journal) |
470decc6 DK |
2143 | { |
2144 | int err = 0; | |
2145 | ||
a931da6a | 2146 | write_lock(&journal->j_state_lock); |
f7f4bccb | 2147 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2148 | err = -EROFS; |
2149 | else | |
2150 | journal->j_errno = 0; | |
a931da6a | 2151 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2152 | return err; |
2153 | } | |
2154 | ||
2155 | /** | |
f7f4bccb | 2156 | * void jbd2_journal_ack_err() - Ack journal err. |
470decc6 DK |
2157 | * @journal: journal to act on. |
2158 | * | |
bfcd3555 | 2159 | * An error must be cleared or acked to take a FS out of readonly |
470decc6 DK |
2160 | * mode. |
2161 | */ | |
f7f4bccb | 2162 | void jbd2_journal_ack_err(journal_t *journal) |
470decc6 | 2163 | { |
a931da6a | 2164 | write_lock(&journal->j_state_lock); |
470decc6 | 2165 | if (journal->j_errno) |
f7f4bccb | 2166 | journal->j_flags |= JBD2_ACK_ERR; |
a931da6a | 2167 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2168 | } |
2169 | ||
f7f4bccb | 2170 | int jbd2_journal_blocks_per_page(struct inode *inode) |
470decc6 DK |
2171 | { |
2172 | return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
2173 | } | |
2174 | ||
b517bea1 ZB |
2175 | /* |
2176 | * helper functions to deal with 32 or 64bit block numbers. | |
2177 | */ | |
2178 | size_t journal_tag_bytes(journal_t *journal) | |
2179 | { | |
db9ee220 DW |
2180 | size_t sz; |
2181 | ||
2182 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V3)) | |
2183 | return sizeof(journal_block_tag3_t); | |
2184 | ||
2185 | sz = sizeof(journal_block_tag_t); | |
c3900875 DW |
2186 | |
2187 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
db9ee220 | 2188 | sz += sizeof(__u16); |
c3900875 | 2189 | |
b517bea1 | 2190 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) |
db9ee220 | 2191 | return sz; |
b517bea1 | 2192 | else |
db9ee220 | 2193 | return sz - sizeof(__u32); |
b517bea1 ZB |
2194 | } |
2195 | ||
d2eecb03 TT |
2196 | /* |
2197 | * JBD memory management | |
2198 | * | |
2199 | * These functions are used to allocate block-sized chunks of memory | |
2200 | * used for making copies of buffer_head data. Very often it will be | |
2201 | * page-sized chunks of data, but sometimes it will be in | |
2202 | * sub-page-size chunks. (For example, 16k pages on Power systems | |
2203 | * with a 4k block file system.) For blocks smaller than a page, we | |
2204 | * use a SLAB allocator. There are slab caches for each block size, | |
2205 | * which are allocated at mount time, if necessary, and we only free | |
2206 | * (all of) the slab caches when/if the jbd2 module is unloaded. For | |
2207 | * this reason we don't need to a mutex to protect access to | |
2208 | * jbd2_slab[] allocating or releasing memory; only in | |
2209 | * jbd2_journal_create_slab(). | |
2210 | */ | |
2211 | #define JBD2_MAX_SLABS 8 | |
2212 | static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS]; | |
d2eecb03 TT |
2213 | |
2214 | static const char *jbd2_slab_names[JBD2_MAX_SLABS] = { | |
2215 | "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k", | |
2216 | "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k" | |
2217 | }; | |
2218 | ||
2219 | ||
2220 | static void jbd2_journal_destroy_slabs(void) | |
2221 | { | |
2222 | int i; | |
2223 | ||
2224 | for (i = 0; i < JBD2_MAX_SLABS; i++) { | |
2225 | if (jbd2_slab[i]) | |
2226 | kmem_cache_destroy(jbd2_slab[i]); | |
2227 | jbd2_slab[i] = NULL; | |
2228 | } | |
2229 | } | |
2230 | ||
2231 | static int jbd2_journal_create_slab(size_t size) | |
2232 | { | |
51dfacde | 2233 | static DEFINE_MUTEX(jbd2_slab_create_mutex); |
d2eecb03 TT |
2234 | int i = order_base_2(size) - 10; |
2235 | size_t slab_size; | |
2236 | ||
2237 | if (size == PAGE_SIZE) | |
2238 | return 0; | |
2239 | ||
2240 | if (i >= JBD2_MAX_SLABS) | |
2241 | return -EINVAL; | |
2242 | ||
2243 | if (unlikely(i < 0)) | |
2244 | i = 0; | |
51dfacde | 2245 | mutex_lock(&jbd2_slab_create_mutex); |
d2eecb03 | 2246 | if (jbd2_slab[i]) { |
51dfacde | 2247 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2248 | return 0; /* Already created */ |
2249 | } | |
2250 | ||
2251 | slab_size = 1 << (i+10); | |
2252 | jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size, | |
2253 | slab_size, 0, NULL); | |
51dfacde | 2254 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2255 | if (!jbd2_slab[i]) { |
2256 | printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n"); | |
2257 | return -ENOMEM; | |
2258 | } | |
2259 | return 0; | |
2260 | } | |
2261 | ||
2262 | static struct kmem_cache *get_slab(size_t size) | |
2263 | { | |
2264 | int i = order_base_2(size) - 10; | |
2265 | ||
2266 | BUG_ON(i >= JBD2_MAX_SLABS); | |
2267 | if (unlikely(i < 0)) | |
2268 | i = 0; | |
8ac97b74 | 2269 | BUG_ON(jbd2_slab[i] == NULL); |
d2eecb03 TT |
2270 | return jbd2_slab[i]; |
2271 | } | |
2272 | ||
2273 | void *jbd2_alloc(size_t size, gfp_t flags) | |
2274 | { | |
2275 | void *ptr; | |
2276 | ||
2277 | BUG_ON(size & (size-1)); /* Must be a power of 2 */ | |
2278 | ||
2279 | flags |= __GFP_REPEAT; | |
2280 | if (size == PAGE_SIZE) | |
2281 | ptr = (void *)__get_free_pages(flags, 0); | |
2282 | else if (size > PAGE_SIZE) { | |
2283 | int order = get_order(size); | |
2284 | ||
2285 | if (order < 3) | |
2286 | ptr = (void *)__get_free_pages(flags, order); | |
2287 | else | |
2288 | ptr = vmalloc(size); | |
2289 | } else | |
2290 | ptr = kmem_cache_alloc(get_slab(size), flags); | |
2291 | ||
2292 | /* Check alignment; SLUB has gotten this wrong in the past, | |
2293 | * and this can lead to user data corruption! */ | |
2294 | BUG_ON(((unsigned long) ptr) & (size-1)); | |
2295 | ||
2296 | return ptr; | |
2297 | } | |
2298 | ||
2299 | void jbd2_free(void *ptr, size_t size) | |
2300 | { | |
2301 | if (size == PAGE_SIZE) { | |
2302 | free_pages((unsigned long)ptr, 0); | |
2303 | return; | |
2304 | } | |
2305 | if (size > PAGE_SIZE) { | |
2306 | int order = get_order(size); | |
2307 | ||
2308 | if (order < 3) | |
2309 | free_pages((unsigned long)ptr, order); | |
2310 | else | |
2311 | vfree(ptr); | |
2312 | return; | |
2313 | } | |
2314 | kmem_cache_free(get_slab(size), ptr); | |
2315 | }; | |
2316 | ||
470decc6 DK |
2317 | /* |
2318 | * Journal_head storage management | |
2319 | */ | |
e18b890b | 2320 | static struct kmem_cache *jbd2_journal_head_cache; |
e23291b9 | 2321 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2322 | static atomic_t nr_journal_heads = ATOMIC_INIT(0); |
2323 | #endif | |
2324 | ||
4185a2ac | 2325 | static int jbd2_journal_init_journal_head_cache(void) |
470decc6 DK |
2326 | { |
2327 | int retval; | |
2328 | ||
1076d17a | 2329 | J_ASSERT(jbd2_journal_head_cache == NULL); |
a920e941 | 2330 | jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", |
470decc6 DK |
2331 | sizeof(struct journal_head), |
2332 | 0, /* offset */ | |
77160957 | 2333 | SLAB_TEMPORARY, /* flags */ |
20c2df83 | 2334 | NULL); /* ctor */ |
470decc6 | 2335 | retval = 0; |
1076d17a | 2336 | if (!jbd2_journal_head_cache) { |
470decc6 | 2337 | retval = -ENOMEM; |
f2a44523 | 2338 | printk(KERN_EMERG "JBD2: no memory for journal_head cache\n"); |
470decc6 DK |
2339 | } |
2340 | return retval; | |
2341 | } | |
2342 | ||
4185a2ac | 2343 | static void jbd2_journal_destroy_journal_head_cache(void) |
470decc6 | 2344 | { |
8a9362eb DG |
2345 | if (jbd2_journal_head_cache) { |
2346 | kmem_cache_destroy(jbd2_journal_head_cache); | |
2347 | jbd2_journal_head_cache = NULL; | |
2348 | } | |
470decc6 DK |
2349 | } |
2350 | ||
2351 | /* | |
2352 | * journal_head splicing and dicing | |
2353 | */ | |
2354 | static struct journal_head *journal_alloc_journal_head(void) | |
2355 | { | |
2356 | struct journal_head *ret; | |
470decc6 | 2357 | |
e23291b9 | 2358 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2359 | atomic_inc(&nr_journal_heads); |
2360 | #endif | |
5d9cf9c6 | 2361 | ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); |
1076d17a | 2362 | if (!ret) { |
470decc6 | 2363 | jbd_debug(1, "out of memory for journal_head\n"); |
670be5a7 | 2364 | pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__); |
1076d17a | 2365 | while (!ret) { |
470decc6 | 2366 | yield(); |
5d9cf9c6 | 2367 | ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); |
470decc6 DK |
2368 | } |
2369 | } | |
2370 | return ret; | |
2371 | } | |
2372 | ||
2373 | static void journal_free_journal_head(struct journal_head *jh) | |
2374 | { | |
e23291b9 | 2375 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 | 2376 | atomic_dec(&nr_journal_heads); |
cd02ff0b | 2377 | memset(jh, JBD2_POISON_FREE, sizeof(*jh)); |
470decc6 | 2378 | #endif |
f7f4bccb | 2379 | kmem_cache_free(jbd2_journal_head_cache, jh); |
470decc6 DK |
2380 | } |
2381 | ||
2382 | /* | |
2383 | * A journal_head is attached to a buffer_head whenever JBD has an | |
2384 | * interest in the buffer. | |
2385 | * | |
2386 | * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit | |
2387 | * is set. This bit is tested in core kernel code where we need to take | |
2388 | * JBD-specific actions. Testing the zeroness of ->b_private is not reliable | |
2389 | * there. | |
2390 | * | |
2391 | * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. | |
2392 | * | |
2393 | * When a buffer has its BH_JBD bit set it is immune from being released by | |
2394 | * core kernel code, mainly via ->b_count. | |
2395 | * | |
de1b7941 JK |
2396 | * A journal_head is detached from its buffer_head when the journal_head's |
2397 | * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint | |
2398 | * transaction (b_cp_transaction) hold their references to b_jcount. | |
470decc6 DK |
2399 | * |
2400 | * Various places in the kernel want to attach a journal_head to a buffer_head | |
2401 | * _before_ attaching the journal_head to a transaction. To protect the | |
f7f4bccb | 2402 | * journal_head in this situation, jbd2_journal_add_journal_head elevates the |
470decc6 | 2403 | * journal_head's b_jcount refcount by one. The caller must call |
f7f4bccb | 2404 | * jbd2_journal_put_journal_head() to undo this. |
470decc6 DK |
2405 | * |
2406 | * So the typical usage would be: | |
2407 | * | |
2408 | * (Attach a journal_head if needed. Increments b_jcount) | |
f7f4bccb | 2409 | * struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 | 2410 | * ... |
de1b7941 JK |
2411 | * (Get another reference for transaction) |
2412 | * jbd2_journal_grab_journal_head(bh); | |
470decc6 | 2413 | * jh->b_transaction = xxx; |
de1b7941 | 2414 | * (Put original reference) |
f7f4bccb | 2415 | * jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2416 | */ |
2417 | ||
2418 | /* | |
2419 | * Give a buffer_head a journal_head. | |
2420 | * | |
470decc6 DK |
2421 | * May sleep. |
2422 | */ | |
f7f4bccb | 2423 | struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) |
470decc6 DK |
2424 | { |
2425 | struct journal_head *jh; | |
2426 | struct journal_head *new_jh = NULL; | |
2427 | ||
2428 | repeat: | |
5d9cf9c6 | 2429 | if (!buffer_jbd(bh)) |
470decc6 | 2430 | new_jh = journal_alloc_journal_head(); |
470decc6 DK |
2431 | |
2432 | jbd_lock_bh_journal_head(bh); | |
2433 | if (buffer_jbd(bh)) { | |
2434 | jh = bh2jh(bh); | |
2435 | } else { | |
2436 | J_ASSERT_BH(bh, | |
2437 | (atomic_read(&bh->b_count) > 0) || | |
2438 | (bh->b_page && bh->b_page->mapping)); | |
2439 | ||
2440 | if (!new_jh) { | |
2441 | jbd_unlock_bh_journal_head(bh); | |
2442 | goto repeat; | |
2443 | } | |
2444 | ||
2445 | jh = new_jh; | |
2446 | new_jh = NULL; /* We consumed it */ | |
2447 | set_buffer_jbd(bh); | |
2448 | bh->b_private = jh; | |
2449 | jh->b_bh = bh; | |
2450 | get_bh(bh); | |
2451 | BUFFER_TRACE(bh, "added journal_head"); | |
2452 | } | |
2453 | jh->b_jcount++; | |
2454 | jbd_unlock_bh_journal_head(bh); | |
2455 | if (new_jh) | |
2456 | journal_free_journal_head(new_jh); | |
2457 | return bh->b_private; | |
2458 | } | |
2459 | ||
2460 | /* | |
2461 | * Grab a ref against this buffer_head's journal_head. If it ended up not | |
2462 | * having a journal_head, return NULL | |
2463 | */ | |
f7f4bccb | 2464 | struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) |
470decc6 DK |
2465 | { |
2466 | struct journal_head *jh = NULL; | |
2467 | ||
2468 | jbd_lock_bh_journal_head(bh); | |
2469 | if (buffer_jbd(bh)) { | |
2470 | jh = bh2jh(bh); | |
2471 | jh->b_jcount++; | |
2472 | } | |
2473 | jbd_unlock_bh_journal_head(bh); | |
2474 | return jh; | |
2475 | } | |
2476 | ||
2477 | static void __journal_remove_journal_head(struct buffer_head *bh) | |
2478 | { | |
2479 | struct journal_head *jh = bh2jh(bh); | |
2480 | ||
2481 | J_ASSERT_JH(jh, jh->b_jcount >= 0); | |
de1b7941 JK |
2482 | J_ASSERT_JH(jh, jh->b_transaction == NULL); |
2483 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
2484 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
2485 | J_ASSERT_JH(jh, jh->b_jlist == BJ_None); | |
2486 | J_ASSERT_BH(bh, buffer_jbd(bh)); | |
2487 | J_ASSERT_BH(bh, jh2bh(jh) == bh); | |
2488 | BUFFER_TRACE(bh, "remove journal_head"); | |
2489 | if (jh->b_frozen_data) { | |
2490 | printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__); | |
2491 | jbd2_free(jh->b_frozen_data, bh->b_size); | |
470decc6 | 2492 | } |
de1b7941 JK |
2493 | if (jh->b_committed_data) { |
2494 | printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__); | |
2495 | jbd2_free(jh->b_committed_data, bh->b_size); | |
2496 | } | |
2497 | bh->b_private = NULL; | |
2498 | jh->b_bh = NULL; /* debug, really */ | |
2499 | clear_buffer_jbd(bh); | |
2500 | journal_free_journal_head(jh); | |
470decc6 DK |
2501 | } |
2502 | ||
2503 | /* | |
de1b7941 | 2504 | * Drop a reference on the passed journal_head. If it fell to zero then |
470decc6 DK |
2505 | * release the journal_head from the buffer_head. |
2506 | */ | |
f7f4bccb | 2507 | void jbd2_journal_put_journal_head(struct journal_head *jh) |
470decc6 DK |
2508 | { |
2509 | struct buffer_head *bh = jh2bh(jh); | |
2510 | ||
2511 | jbd_lock_bh_journal_head(bh); | |
2512 | J_ASSERT_JH(jh, jh->b_jcount > 0); | |
2513 | --jh->b_jcount; | |
de1b7941 | 2514 | if (!jh->b_jcount) { |
470decc6 | 2515 | __journal_remove_journal_head(bh); |
de1b7941 | 2516 | jbd_unlock_bh_journal_head(bh); |
470decc6 | 2517 | __brelse(bh); |
de1b7941 JK |
2518 | } else |
2519 | jbd_unlock_bh_journal_head(bh); | |
470decc6 DK |
2520 | } |
2521 | ||
c851ed54 JK |
2522 | /* |
2523 | * Initialize jbd inode head | |
2524 | */ | |
2525 | void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) | |
2526 | { | |
2527 | jinode->i_transaction = NULL; | |
2528 | jinode->i_next_transaction = NULL; | |
2529 | jinode->i_vfs_inode = inode; | |
2530 | jinode->i_flags = 0; | |
2531 | INIT_LIST_HEAD(&jinode->i_list); | |
2532 | } | |
2533 | ||
2534 | /* | |
2535 | * Function to be called before we start removing inode from memory (i.e., | |
2536 | * clear_inode() is a fine place to be called from). It removes inode from | |
2537 | * transaction's lists. | |
2538 | */ | |
2539 | void jbd2_journal_release_jbd_inode(journal_t *journal, | |
2540 | struct jbd2_inode *jinode) | |
2541 | { | |
c851ed54 JK |
2542 | if (!journal) |
2543 | return; | |
2544 | restart: | |
2545 | spin_lock(&journal->j_list_lock); | |
2546 | /* Is commit writing out inode - we have to wait */ | |
39e3ac25 | 2547 | if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) { |
c851ed54 JK |
2548 | wait_queue_head_t *wq; |
2549 | DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); | |
2550 | wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); | |
2551 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
2552 | spin_unlock(&journal->j_list_lock); | |
2553 | schedule(); | |
2554 | finish_wait(wq, &wait.wait); | |
2555 | goto restart; | |
2556 | } | |
2557 | ||
c851ed54 JK |
2558 | if (jinode->i_transaction) { |
2559 | list_del(&jinode->i_list); | |
2560 | jinode->i_transaction = NULL; | |
2561 | } | |
2562 | spin_unlock(&journal->j_list_lock); | |
2563 | } | |
2564 | ||
470decc6 | 2565 | |
8e85fb3f JL |
2566 | #ifdef CONFIG_PROC_FS |
2567 | ||
2568 | #define JBD2_STATS_PROC_NAME "fs/jbd2" | |
2569 | ||
2570 | static void __init jbd2_create_jbd_stats_proc_entry(void) | |
2571 | { | |
2572 | proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); | |
2573 | } | |
2574 | ||
2575 | static void __exit jbd2_remove_jbd_stats_proc_entry(void) | |
2576 | { | |
2577 | if (proc_jbd2_stats) | |
2578 | remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); | |
2579 | } | |
2580 | ||
2581 | #else | |
2582 | ||
2583 | #define jbd2_create_jbd_stats_proc_entry() do {} while (0) | |
2584 | #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) | |
2585 | ||
2586 | #endif | |
2587 | ||
8aefcd55 | 2588 | struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache; |
470decc6 | 2589 | |
4185a2ac | 2590 | static int __init jbd2_journal_init_handle_cache(void) |
470decc6 | 2591 | { |
8aefcd55 | 2592 | jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY); |
f7f4bccb | 2593 | if (jbd2_handle_cache == NULL) { |
8aefcd55 TT |
2594 | printk(KERN_EMERG "JBD2: failed to create handle cache\n"); |
2595 | return -ENOMEM; | |
2596 | } | |
2597 | jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0); | |
2598 | if (jbd2_inode_cache == NULL) { | |
2599 | printk(KERN_EMERG "JBD2: failed to create inode cache\n"); | |
2600 | kmem_cache_destroy(jbd2_handle_cache); | |
470decc6 DK |
2601 | return -ENOMEM; |
2602 | } | |
2603 | return 0; | |
2604 | } | |
2605 | ||
f7f4bccb | 2606 | static void jbd2_journal_destroy_handle_cache(void) |
470decc6 | 2607 | { |
f7f4bccb MC |
2608 | if (jbd2_handle_cache) |
2609 | kmem_cache_destroy(jbd2_handle_cache); | |
8aefcd55 TT |
2610 | if (jbd2_inode_cache) |
2611 | kmem_cache_destroy(jbd2_inode_cache); | |
2612 | ||
470decc6 DK |
2613 | } |
2614 | ||
2615 | /* | |
2616 | * Module startup and shutdown | |
2617 | */ | |
2618 | ||
2619 | static int __init journal_init_caches(void) | |
2620 | { | |
2621 | int ret; | |
2622 | ||
f7f4bccb | 2623 | ret = jbd2_journal_init_revoke_caches(); |
470decc6 | 2624 | if (ret == 0) |
4185a2ac | 2625 | ret = jbd2_journal_init_journal_head_cache(); |
470decc6 | 2626 | if (ret == 0) |
4185a2ac | 2627 | ret = jbd2_journal_init_handle_cache(); |
470decc6 | 2628 | if (ret == 0) |
0c2022ec | 2629 | ret = jbd2_journal_init_transaction_cache(); |
470decc6 DK |
2630 | return ret; |
2631 | } | |
2632 | ||
f7f4bccb | 2633 | static void jbd2_journal_destroy_caches(void) |
470decc6 | 2634 | { |
f7f4bccb | 2635 | jbd2_journal_destroy_revoke_caches(); |
4185a2ac | 2636 | jbd2_journal_destroy_journal_head_cache(); |
f7f4bccb | 2637 | jbd2_journal_destroy_handle_cache(); |
0c2022ec | 2638 | jbd2_journal_destroy_transaction_cache(); |
d2eecb03 | 2639 | jbd2_journal_destroy_slabs(); |
470decc6 DK |
2640 | } |
2641 | ||
2642 | static int __init journal_init(void) | |
2643 | { | |
2644 | int ret; | |
2645 | ||
2646 | BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); | |
2647 | ||
2648 | ret = journal_init_caches(); | |
620de4e1 | 2649 | if (ret == 0) { |
620de4e1 DG |
2650 | jbd2_create_jbd_stats_proc_entry(); |
2651 | } else { | |
f7f4bccb | 2652 | jbd2_journal_destroy_caches(); |
620de4e1 | 2653 | } |
470decc6 DK |
2654 | return ret; |
2655 | } | |
2656 | ||
2657 | static void __exit journal_exit(void) | |
2658 | { | |
e23291b9 | 2659 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2660 | int n = atomic_read(&nr_journal_heads); |
2661 | if (n) | |
75685071 | 2662 | printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n); |
470decc6 | 2663 | #endif |
8e85fb3f | 2664 | jbd2_remove_jbd_stats_proc_entry(); |
f7f4bccb | 2665 | jbd2_journal_destroy_caches(); |
470decc6 DK |
2666 | } |
2667 | ||
2668 | MODULE_LICENSE("GPL"); | |
2669 | module_init(journal_init); | |
2670 | module_exit(journal_exit); | |
2671 |