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> | |
0f49d5d0 | 38 | #include <linux/debugfs.h> |
8e85fb3f | 39 | #include <linux/seq_file.h> |
c225aa57 | 40 | #include <linux/math64.h> |
879c5e6b | 41 | #include <linux/hash.h> |
d2eecb03 TT |
42 | #include <linux/log2.h> |
43 | #include <linux/vmalloc.h> | |
47def826 | 44 | #include <linux/backing-dev.h> |
39e3ac25 | 45 | #include <linux/bitops.h> |
670be5a7 | 46 | #include <linux/ratelimit.h> |
879c5e6b TT |
47 | |
48 | #define CREATE_TRACE_POINTS | |
49 | #include <trace/events/jbd2.h> | |
470decc6 DK |
50 | |
51 | #include <asm/uaccess.h> | |
52 | #include <asm/page.h> | |
53 | ||
f7f4bccb MC |
54 | EXPORT_SYMBOL(jbd2_journal_extend); |
55 | EXPORT_SYMBOL(jbd2_journal_stop); | |
56 | EXPORT_SYMBOL(jbd2_journal_lock_updates); | |
57 | EXPORT_SYMBOL(jbd2_journal_unlock_updates); | |
58 | EXPORT_SYMBOL(jbd2_journal_get_write_access); | |
59 | EXPORT_SYMBOL(jbd2_journal_get_create_access); | |
60 | EXPORT_SYMBOL(jbd2_journal_get_undo_access); | |
e06c8227 | 61 | EXPORT_SYMBOL(jbd2_journal_set_triggers); |
f7f4bccb MC |
62 | EXPORT_SYMBOL(jbd2_journal_dirty_metadata); |
63 | EXPORT_SYMBOL(jbd2_journal_release_buffer); | |
64 | EXPORT_SYMBOL(jbd2_journal_forget); | |
470decc6 DK |
65 | #if 0 |
66 | EXPORT_SYMBOL(journal_sync_buffer); | |
67 | #endif | |
f7f4bccb MC |
68 | EXPORT_SYMBOL(jbd2_journal_flush); |
69 | EXPORT_SYMBOL(jbd2_journal_revoke); | |
70 | ||
71 | EXPORT_SYMBOL(jbd2_journal_init_dev); | |
72 | EXPORT_SYMBOL(jbd2_journal_init_inode); | |
f7f4bccb MC |
73 | EXPORT_SYMBOL(jbd2_journal_check_used_features); |
74 | EXPORT_SYMBOL(jbd2_journal_check_available_features); | |
75 | EXPORT_SYMBOL(jbd2_journal_set_features); | |
f7f4bccb MC |
76 | EXPORT_SYMBOL(jbd2_journal_load); |
77 | EXPORT_SYMBOL(jbd2_journal_destroy); | |
f7f4bccb MC |
78 | EXPORT_SYMBOL(jbd2_journal_abort); |
79 | EXPORT_SYMBOL(jbd2_journal_errno); | |
80 | EXPORT_SYMBOL(jbd2_journal_ack_err); | |
81 | EXPORT_SYMBOL(jbd2_journal_clear_err); | |
82 | EXPORT_SYMBOL(jbd2_log_wait_commit); | |
3b799d15 | 83 | EXPORT_SYMBOL(jbd2_log_start_commit); |
f7f4bccb MC |
84 | EXPORT_SYMBOL(jbd2_journal_start_commit); |
85 | EXPORT_SYMBOL(jbd2_journal_force_commit_nested); | |
86 | EXPORT_SYMBOL(jbd2_journal_wipe); | |
87 | EXPORT_SYMBOL(jbd2_journal_blocks_per_page); | |
88 | EXPORT_SYMBOL(jbd2_journal_invalidatepage); | |
89 | EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); | |
90 | EXPORT_SYMBOL(jbd2_journal_force_commit); | |
c851ed54 JK |
91 | EXPORT_SYMBOL(jbd2_journal_file_inode); |
92 | EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); | |
93 | EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); | |
94 | EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); | |
8aefcd55 | 95 | EXPORT_SYMBOL(jbd2_inode_cache); |
470decc6 | 96 | |
470decc6 | 97 | static void __journal_abort_soft (journal_t *journal, int errno); |
d2eecb03 | 98 | static int jbd2_journal_create_slab(size_t slab_size); |
470decc6 | 99 | |
25ed6e8a DW |
100 | /* Checksumming functions */ |
101 | int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb) | |
102 | { | |
103 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
104 | return 1; | |
105 | ||
106 | return sb->s_checksum_type == JBD2_CRC32C_CHKSUM; | |
107 | } | |
108 | ||
4fd5ea43 DW |
109 | static __u32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb) |
110 | { | |
111 | __u32 csum, old_csum; | |
112 | ||
113 | old_csum = sb->s_checksum; | |
114 | sb->s_checksum = 0; | |
115 | csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t)); | |
116 | sb->s_checksum = old_csum; | |
117 | ||
118 | return cpu_to_be32(csum); | |
119 | } | |
120 | ||
121 | int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb) | |
122 | { | |
123 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
124 | return 1; | |
125 | ||
126 | return sb->s_checksum == jbd2_superblock_csum(j, sb); | |
127 | } | |
128 | ||
129 | void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb) | |
130 | { | |
131 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
132 | return; | |
133 | ||
134 | sb->s_checksum = jbd2_superblock_csum(j, sb); | |
135 | } | |
136 | ||
470decc6 DK |
137 | /* |
138 | * Helper function used to manage commit timeouts | |
139 | */ | |
140 | ||
141 | static void commit_timeout(unsigned long __data) | |
142 | { | |
143 | struct task_struct * p = (struct task_struct *) __data; | |
144 | ||
145 | wake_up_process(p); | |
146 | } | |
147 | ||
148 | /* | |
f7f4bccb | 149 | * kjournald2: The main thread function used to manage a logging device |
470decc6 DK |
150 | * journal. |
151 | * | |
152 | * This kernel thread is responsible for two things: | |
153 | * | |
154 | * 1) COMMIT: Every so often we need to commit the current state of the | |
155 | * filesystem to disk. The journal thread is responsible for writing | |
156 | * all of the metadata buffers to disk. | |
157 | * | |
158 | * 2) CHECKPOINT: We cannot reuse a used section of the log file until all | |
159 | * of the data in that part of the log has been rewritten elsewhere on | |
160 | * the disk. Flushing these old buffers to reclaim space in the log is | |
161 | * known as checkpointing, and this thread is responsible for that job. | |
162 | */ | |
163 | ||
f7f4bccb | 164 | static int kjournald2(void *arg) |
470decc6 DK |
165 | { |
166 | journal_t *journal = arg; | |
167 | transaction_t *transaction; | |
168 | ||
169 | /* | |
170 | * Set up an interval timer which can be used to trigger a commit wakeup | |
171 | * after the commit interval expires | |
172 | */ | |
173 | setup_timer(&journal->j_commit_timer, commit_timeout, | |
174 | (unsigned long)current); | |
175 | ||
35c80422 NC |
176 | set_freezable(); |
177 | ||
470decc6 DK |
178 | /* Record that the journal thread is running */ |
179 | journal->j_task = current; | |
180 | wake_up(&journal->j_wait_done_commit); | |
181 | ||
470decc6 DK |
182 | /* |
183 | * And now, wait forever for commit wakeup events. | |
184 | */ | |
a931da6a | 185 | write_lock(&journal->j_state_lock); |
470decc6 DK |
186 | |
187 | loop: | |
f7f4bccb | 188 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
189 | goto end_loop; |
190 | ||
191 | jbd_debug(1, "commit_sequence=%d, commit_request=%d\n", | |
192 | journal->j_commit_sequence, journal->j_commit_request); | |
193 | ||
194 | if (journal->j_commit_sequence != journal->j_commit_request) { | |
195 | jbd_debug(1, "OK, requests differ\n"); | |
a931da6a | 196 | write_unlock(&journal->j_state_lock); |
470decc6 | 197 | del_timer_sync(&journal->j_commit_timer); |
f7f4bccb | 198 | jbd2_journal_commit_transaction(journal); |
a931da6a | 199 | write_lock(&journal->j_state_lock); |
470decc6 DK |
200 | goto loop; |
201 | } | |
202 | ||
203 | wake_up(&journal->j_wait_done_commit); | |
204 | if (freezing(current)) { | |
205 | /* | |
206 | * The simpler the better. Flushing journal isn't a | |
207 | * good idea, because that depends on threads that may | |
208 | * be already stopped. | |
209 | */ | |
f7f4bccb | 210 | jbd_debug(1, "Now suspending kjournald2\n"); |
a931da6a | 211 | write_unlock(&journal->j_state_lock); |
a0acae0e | 212 | try_to_freeze(); |
a931da6a | 213 | write_lock(&journal->j_state_lock); |
470decc6 DK |
214 | } else { |
215 | /* | |
216 | * We assume on resume that commits are already there, | |
217 | * so we don't sleep | |
218 | */ | |
219 | DEFINE_WAIT(wait); | |
220 | int should_sleep = 1; | |
221 | ||
222 | prepare_to_wait(&journal->j_wait_commit, &wait, | |
223 | TASK_INTERRUPTIBLE); | |
224 | if (journal->j_commit_sequence != journal->j_commit_request) | |
225 | should_sleep = 0; | |
226 | transaction = journal->j_running_transaction; | |
227 | if (transaction && time_after_eq(jiffies, | |
228 | transaction->t_expires)) | |
229 | should_sleep = 0; | |
f7f4bccb | 230 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
231 | should_sleep = 0; |
232 | if (should_sleep) { | |
a931da6a | 233 | write_unlock(&journal->j_state_lock); |
470decc6 | 234 | schedule(); |
a931da6a | 235 | write_lock(&journal->j_state_lock); |
470decc6 DK |
236 | } |
237 | finish_wait(&journal->j_wait_commit, &wait); | |
238 | } | |
239 | ||
f7f4bccb | 240 | jbd_debug(1, "kjournald2 wakes\n"); |
470decc6 DK |
241 | |
242 | /* | |
243 | * Were we woken up by a commit wakeup event? | |
244 | */ | |
245 | transaction = journal->j_running_transaction; | |
246 | if (transaction && time_after_eq(jiffies, transaction->t_expires)) { | |
247 | journal->j_commit_request = transaction->t_tid; | |
248 | jbd_debug(1, "woke because of timeout\n"); | |
249 | } | |
250 | goto loop; | |
251 | ||
252 | end_loop: | |
a931da6a | 253 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
254 | del_timer_sync(&journal->j_commit_timer); |
255 | journal->j_task = NULL; | |
256 | wake_up(&journal->j_wait_done_commit); | |
257 | jbd_debug(1, "Journal thread exiting.\n"); | |
258 | return 0; | |
259 | } | |
260 | ||
97f06784 | 261 | static int jbd2_journal_start_thread(journal_t *journal) |
470decc6 | 262 | { |
97f06784 PE |
263 | struct task_struct *t; |
264 | ||
90576c0b TT |
265 | t = kthread_run(kjournald2, journal, "jbd2/%s", |
266 | journal->j_devname); | |
97f06784 PE |
267 | if (IS_ERR(t)) |
268 | return PTR_ERR(t); | |
269 | ||
1076d17a | 270 | wait_event(journal->j_wait_done_commit, journal->j_task != NULL); |
97f06784 | 271 | return 0; |
470decc6 DK |
272 | } |
273 | ||
274 | static void journal_kill_thread(journal_t *journal) | |
275 | { | |
a931da6a | 276 | write_lock(&journal->j_state_lock); |
f7f4bccb | 277 | journal->j_flags |= JBD2_UNMOUNT; |
470decc6 DK |
278 | |
279 | while (journal->j_task) { | |
280 | wake_up(&journal->j_wait_commit); | |
a931da6a | 281 | write_unlock(&journal->j_state_lock); |
1076d17a | 282 | wait_event(journal->j_wait_done_commit, journal->j_task == NULL); |
a931da6a | 283 | write_lock(&journal->j_state_lock); |
470decc6 | 284 | } |
a931da6a | 285 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
286 | } |
287 | ||
288 | /* | |
f7f4bccb | 289 | * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. |
470decc6 DK |
290 | * |
291 | * Writes a metadata buffer to a given disk block. The actual IO is not | |
292 | * performed but a new buffer_head is constructed which labels the data | |
293 | * to be written with the correct destination disk block. | |
294 | * | |
295 | * Any magic-number escaping which needs to be done will cause a | |
296 | * copy-out here. If the buffer happens to start with the | |
f7f4bccb | 297 | * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the |
470decc6 DK |
298 | * magic number is only written to the log for descripter blocks. In |
299 | * this case, we copy the data and replace the first word with 0, and we | |
300 | * return a result code which indicates that this buffer needs to be | |
301 | * marked as an escaped buffer in the corresponding log descriptor | |
302 | * block. The missing word can then be restored when the block is read | |
303 | * during recovery. | |
304 | * | |
305 | * If the source buffer has already been modified by a new transaction | |
306 | * since we took the last commit snapshot, we use the frozen copy of | |
307 | * that data for IO. If we end up using the existing buffer_head's data | |
308 | * for the write, then we *have* to lock the buffer to prevent anyone | |
309 | * else from using and possibly modifying it while the IO is in | |
310 | * progress. | |
311 | * | |
312 | * The function returns a pointer to the buffer_heads to be used for IO. | |
313 | * | |
314 | * We assume that the journal has already been locked in this function. | |
315 | * | |
316 | * Return value: | |
317 | * <0: Error | |
318 | * >=0: Finished OK | |
319 | * | |
320 | * On success: | |
321 | * Bit 0 set == escape performed on the data | |
322 | * Bit 1 set == buffer copy-out performed (kfree the data after IO) | |
323 | */ | |
324 | ||
f7f4bccb | 325 | int jbd2_journal_write_metadata_buffer(transaction_t *transaction, |
470decc6 DK |
326 | struct journal_head *jh_in, |
327 | struct journal_head **jh_out, | |
18eba7aa | 328 | unsigned long long blocknr) |
470decc6 DK |
329 | { |
330 | int need_copy_out = 0; | |
331 | int done_copy_out = 0; | |
332 | int do_escape = 0; | |
333 | char *mapped_data; | |
334 | struct buffer_head *new_bh; | |
335 | struct journal_head *new_jh; | |
336 | struct page *new_page; | |
337 | unsigned int new_offset; | |
338 | struct buffer_head *bh_in = jh2bh(jh_in); | |
96577c43 | 339 | journal_t *journal = transaction->t_journal; |
470decc6 DK |
340 | |
341 | /* | |
342 | * The buffer really shouldn't be locked: only the current committing | |
343 | * transaction is allowed to write it, so nobody else is allowed | |
344 | * to do any IO. | |
345 | * | |
346 | * akpm: except if we're journalling data, and write() output is | |
347 | * also part of a shared mapping, and another thread has | |
348 | * decided to launch a writepage() against this buffer. | |
349 | */ | |
350 | J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); | |
351 | ||
47def826 TT |
352 | retry_alloc: |
353 | new_bh = alloc_buffer_head(GFP_NOFS); | |
354 | if (!new_bh) { | |
355 | /* | |
356 | * Failure is not an option, but __GFP_NOFAIL is going | |
357 | * away; so we retry ourselves here. | |
358 | */ | |
359 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
360 | goto retry_alloc; | |
361 | } | |
362 | ||
96577c43 | 363 | /* keep subsequent assertions sane */ |
364 | new_bh->b_state = 0; | |
365 | init_buffer(new_bh, NULL, NULL); | |
366 | atomic_set(&new_bh->b_count, 1); | |
367 | new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */ | |
470decc6 DK |
368 | |
369 | /* | |
370 | * If a new transaction has already done a buffer copy-out, then | |
371 | * we use that version of the data for the commit. | |
372 | */ | |
373 | jbd_lock_bh_state(bh_in); | |
374 | repeat: | |
375 | if (jh_in->b_frozen_data) { | |
376 | done_copy_out = 1; | |
377 | new_page = virt_to_page(jh_in->b_frozen_data); | |
378 | new_offset = offset_in_page(jh_in->b_frozen_data); | |
379 | } else { | |
380 | new_page = jh2bh(jh_in)->b_page; | |
381 | new_offset = offset_in_page(jh2bh(jh_in)->b_data); | |
382 | } | |
383 | ||
303a8f2a | 384 | mapped_data = kmap_atomic(new_page); |
e06c8227 | 385 | /* |
13ceef09 JK |
386 | * Fire data frozen trigger if data already wasn't frozen. Do this |
387 | * before checking for escaping, as the trigger may modify the magic | |
388 | * offset. If a copy-out happens afterwards, it will have the correct | |
389 | * data in the buffer. | |
e06c8227 | 390 | */ |
13ceef09 JK |
391 | if (!done_copy_out) |
392 | jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset, | |
393 | jh_in->b_triggers); | |
e06c8227 | 394 | |
470decc6 DK |
395 | /* |
396 | * Check for escaping | |
397 | */ | |
398 | if (*((__be32 *)(mapped_data + new_offset)) == | |
f7f4bccb | 399 | cpu_to_be32(JBD2_MAGIC_NUMBER)) { |
470decc6 DK |
400 | need_copy_out = 1; |
401 | do_escape = 1; | |
402 | } | |
303a8f2a | 403 | kunmap_atomic(mapped_data); |
470decc6 DK |
404 | |
405 | /* | |
406 | * Do we need to do a data copy? | |
407 | */ | |
408 | if (need_copy_out && !done_copy_out) { | |
409 | char *tmp; | |
410 | ||
411 | jbd_unlock_bh_state(bh_in); | |
af1e76d6 | 412 | tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); |
e6ec116b TT |
413 | if (!tmp) { |
414 | jbd2_journal_put_journal_head(new_jh); | |
415 | return -ENOMEM; | |
416 | } | |
470decc6 DK |
417 | jbd_lock_bh_state(bh_in); |
418 | if (jh_in->b_frozen_data) { | |
af1e76d6 | 419 | jbd2_free(tmp, bh_in->b_size); |
470decc6 DK |
420 | goto repeat; |
421 | } | |
422 | ||
423 | jh_in->b_frozen_data = tmp; | |
303a8f2a | 424 | mapped_data = kmap_atomic(new_page); |
470decc6 | 425 | memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size); |
303a8f2a | 426 | kunmap_atomic(mapped_data); |
470decc6 DK |
427 | |
428 | new_page = virt_to_page(tmp); | |
429 | new_offset = offset_in_page(tmp); | |
430 | done_copy_out = 1; | |
e06c8227 JB |
431 | |
432 | /* | |
433 | * This isn't strictly necessary, as we're using frozen | |
434 | * data for the escaping, but it keeps consistency with | |
435 | * b_frozen_data usage. | |
436 | */ | |
437 | jh_in->b_frozen_triggers = jh_in->b_triggers; | |
470decc6 DK |
438 | } |
439 | ||
440 | /* | |
441 | * Did we need to do an escaping? Now we've done all the | |
442 | * copying, we can finally do so. | |
443 | */ | |
444 | if (do_escape) { | |
303a8f2a | 445 | mapped_data = kmap_atomic(new_page); |
470decc6 | 446 | *((unsigned int *)(mapped_data + new_offset)) = 0; |
303a8f2a | 447 | kunmap_atomic(mapped_data); |
470decc6 DK |
448 | } |
449 | ||
470decc6 DK |
450 | set_bh_page(new_bh, new_page, new_offset); |
451 | new_jh->b_transaction = NULL; | |
452 | new_bh->b_size = jh2bh(jh_in)->b_size; | |
453 | new_bh->b_bdev = transaction->t_journal->j_dev; | |
454 | new_bh->b_blocknr = blocknr; | |
455 | set_buffer_mapped(new_bh); | |
456 | set_buffer_dirty(new_bh); | |
457 | ||
458 | *jh_out = new_jh; | |
459 | ||
460 | /* | |
461 | * The to-be-written buffer needs to get moved to the io queue, | |
462 | * and the original buffer whose contents we are shadowing or | |
463 | * copying is moved to the transaction's shadow queue. | |
464 | */ | |
465 | JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); | |
96577c43 | 466 | spin_lock(&journal->j_list_lock); |
467 | __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); | |
468 | spin_unlock(&journal->j_list_lock); | |
469 | jbd_unlock_bh_state(bh_in); | |
470 | ||
470decc6 | 471 | JBUFFER_TRACE(new_jh, "file as BJ_IO"); |
f7f4bccb | 472 | jbd2_journal_file_buffer(new_jh, transaction, BJ_IO); |
470decc6 DK |
473 | |
474 | return do_escape | (done_copy_out << 1); | |
475 | } | |
476 | ||
477 | /* | |
478 | * Allocation code for the journal file. Manage the space left in the | |
479 | * journal, so that we can begin checkpointing when appropriate. | |
480 | */ | |
481 | ||
482 | /* | |
f7f4bccb | 483 | * __jbd2_log_space_left: Return the number of free blocks left in the journal. |
470decc6 DK |
484 | * |
485 | * Called with the journal already locked. | |
486 | * | |
487 | * Called under j_state_lock | |
488 | */ | |
489 | ||
f7f4bccb | 490 | int __jbd2_log_space_left(journal_t *journal) |
470decc6 DK |
491 | { |
492 | int left = journal->j_free; | |
493 | ||
a931da6a | 494 | /* assert_spin_locked(&journal->j_state_lock); */ |
470decc6 DK |
495 | |
496 | /* | |
497 | * Be pessimistic here about the number of those free blocks which | |
498 | * might be required for log descriptor control blocks. | |
499 | */ | |
500 | ||
501 | #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */ | |
502 | ||
503 | left -= MIN_LOG_RESERVED_BLOCKS; | |
504 | ||
505 | if (left <= 0) | |
506 | return 0; | |
507 | left -= (left >> 3); | |
508 | return left; | |
509 | } | |
510 | ||
511 | /* | |
e4471831 TT |
512 | * Called with j_state_lock locked for writing. |
513 | * Returns true if a transaction commit was started. | |
470decc6 | 514 | */ |
f7f4bccb | 515 | int __jbd2_log_start_commit(journal_t *journal, tid_t target) |
470decc6 DK |
516 | { |
517 | /* | |
deeeaf13 TT |
518 | * The only transaction we can possibly wait upon is the |
519 | * currently running transaction (if it exists). Otherwise, | |
520 | * the target tid must be an old one. | |
470decc6 | 521 | */ |
deeeaf13 TT |
522 | if (journal->j_running_transaction && |
523 | journal->j_running_transaction->t_tid == target) { | |
470decc6 | 524 | /* |
bcf3d0bc | 525 | * We want a new commit: OK, mark the request and wakeup the |
470decc6 DK |
526 | * commit thread. We do _not_ do the commit ourselves. |
527 | */ | |
528 | ||
529 | journal->j_commit_request = target; | |
f2a44523 | 530 | jbd_debug(1, "JBD2: requesting commit %d/%d\n", |
470decc6 DK |
531 | journal->j_commit_request, |
532 | journal->j_commit_sequence); | |
533 | wake_up(&journal->j_wait_commit); | |
534 | return 1; | |
deeeaf13 TT |
535 | } else if (!tid_geq(journal->j_commit_request, target)) |
536 | /* This should never happen, but if it does, preserve | |
537 | the evidence before kjournald goes into a loop and | |
538 | increments j_commit_sequence beyond all recognition. */ | |
f2a44523 | 539 | WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", |
1be2add6 TT |
540 | journal->j_commit_request, |
541 | journal->j_commit_sequence, | |
542 | target, journal->j_running_transaction ? | |
543 | journal->j_running_transaction->t_tid : 0); | |
470decc6 DK |
544 | return 0; |
545 | } | |
546 | ||
f7f4bccb | 547 | int jbd2_log_start_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
548 | { |
549 | int ret; | |
550 | ||
a931da6a | 551 | write_lock(&journal->j_state_lock); |
f7f4bccb | 552 | ret = __jbd2_log_start_commit(journal, tid); |
a931da6a | 553 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
554 | return ret; |
555 | } | |
556 | ||
557 | /* | |
558 | * Force and wait upon a commit if the calling process is not within | |
559 | * transaction. This is used for forcing out undo-protected data which contains | |
560 | * bitmaps, when the fs is running out of space. | |
561 | * | |
562 | * We can only force the running transaction if we don't have an active handle; | |
563 | * otherwise, we will deadlock. | |
564 | * | |
565 | * Returns true if a transaction was started. | |
566 | */ | |
f7f4bccb | 567 | int jbd2_journal_force_commit_nested(journal_t *journal) |
470decc6 DK |
568 | { |
569 | transaction_t *transaction = NULL; | |
570 | tid_t tid; | |
e4471831 | 571 | int need_to_start = 0; |
470decc6 | 572 | |
a931da6a | 573 | read_lock(&journal->j_state_lock); |
470decc6 DK |
574 | if (journal->j_running_transaction && !current->journal_info) { |
575 | transaction = journal->j_running_transaction; | |
e4471831 TT |
576 | if (!tid_geq(journal->j_commit_request, transaction->t_tid)) |
577 | need_to_start = 1; | |
470decc6 DK |
578 | } else if (journal->j_committing_transaction) |
579 | transaction = journal->j_committing_transaction; | |
580 | ||
581 | if (!transaction) { | |
a931da6a | 582 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
583 | return 0; /* Nothing to retry */ |
584 | } | |
585 | ||
586 | tid = transaction->t_tid; | |
a931da6a | 587 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
588 | if (need_to_start) |
589 | jbd2_log_start_commit(journal, tid); | |
f7f4bccb | 590 | jbd2_log_wait_commit(journal, tid); |
470decc6 DK |
591 | return 1; |
592 | } | |
593 | ||
594 | /* | |
595 | * Start a commit of the current running transaction (if any). Returns true | |
c88ccea3 JK |
596 | * if a transaction is going to be committed (or is currently already |
597 | * committing), and fills its tid in at *ptid | |
470decc6 | 598 | */ |
f7f4bccb | 599 | int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) |
470decc6 DK |
600 | { |
601 | int ret = 0; | |
602 | ||
a931da6a | 603 | write_lock(&journal->j_state_lock); |
470decc6 DK |
604 | if (journal->j_running_transaction) { |
605 | tid_t tid = journal->j_running_transaction->t_tid; | |
606 | ||
c88ccea3 JK |
607 | __jbd2_log_start_commit(journal, tid); |
608 | /* There's a running transaction and we've just made sure | |
609 | * it's commit has been scheduled. */ | |
610 | if (ptid) | |
470decc6 | 611 | *ptid = tid; |
c88ccea3 JK |
612 | ret = 1; |
613 | } else if (journal->j_committing_transaction) { | |
470decc6 | 614 | /* |
12810ad7 AB |
615 | * If commit has been started, then we have to wait for |
616 | * completion of that transaction. | |
470decc6 | 617 | */ |
c88ccea3 JK |
618 | if (ptid) |
619 | *ptid = journal->j_committing_transaction->t_tid; | |
470decc6 DK |
620 | ret = 1; |
621 | } | |
a931da6a | 622 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
623 | return ret; |
624 | } | |
625 | ||
bbd2be36 JK |
626 | /* |
627 | * Return 1 if a given transaction has not yet sent barrier request | |
628 | * connected with a transaction commit. If 0 is returned, transaction | |
629 | * may or may not have sent the barrier. Used to avoid sending barrier | |
630 | * twice in common cases. | |
631 | */ | |
632 | int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) | |
633 | { | |
634 | int ret = 0; | |
635 | transaction_t *commit_trans; | |
636 | ||
637 | if (!(journal->j_flags & JBD2_BARRIER)) | |
638 | return 0; | |
639 | read_lock(&journal->j_state_lock); | |
640 | /* Transaction already committed? */ | |
641 | if (tid_geq(journal->j_commit_sequence, tid)) | |
642 | goto out; | |
643 | commit_trans = journal->j_committing_transaction; | |
644 | if (!commit_trans || commit_trans->t_tid != tid) { | |
645 | ret = 1; | |
646 | goto out; | |
647 | } | |
648 | /* | |
649 | * Transaction is being committed and we already proceeded to | |
650 | * submitting a flush to fs partition? | |
651 | */ | |
652 | if (journal->j_fs_dev != journal->j_dev) { | |
653 | if (!commit_trans->t_need_data_flush || | |
654 | commit_trans->t_state >= T_COMMIT_DFLUSH) | |
655 | goto out; | |
656 | } else { | |
657 | if (commit_trans->t_state >= T_COMMIT_JFLUSH) | |
658 | goto out; | |
659 | } | |
660 | ret = 1; | |
661 | out: | |
662 | read_unlock(&journal->j_state_lock); | |
663 | return ret; | |
664 | } | |
665 | EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); | |
666 | ||
470decc6 DK |
667 | /* |
668 | * Wait for a specified commit to complete. | |
669 | * The caller may not hold the journal lock. | |
670 | */ | |
f7f4bccb | 671 | int jbd2_log_wait_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
672 | { |
673 | int err = 0; | |
674 | ||
a931da6a | 675 | read_lock(&journal->j_state_lock); |
e23291b9 | 676 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
677 | if (!tid_geq(journal->j_commit_request, tid)) { |
678 | printk(KERN_EMERG | |
679 | "%s: error: j_commit_request=%d, tid=%d\n", | |
329d291f | 680 | __func__, journal->j_commit_request, tid); |
470decc6 | 681 | } |
470decc6 | 682 | #endif |
470decc6 | 683 | while (tid_gt(tid, journal->j_commit_sequence)) { |
f2a44523 | 684 | jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n", |
470decc6 DK |
685 | tid, journal->j_commit_sequence); |
686 | wake_up(&journal->j_wait_commit); | |
a931da6a | 687 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
688 | wait_event(journal->j_wait_done_commit, |
689 | !tid_gt(tid, journal->j_commit_sequence)); | |
a931da6a | 690 | read_lock(&journal->j_state_lock); |
470decc6 | 691 | } |
a931da6a | 692 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
693 | |
694 | if (unlikely(is_journal_aborted(journal))) { | |
695 | printk(KERN_EMERG "journal commit I/O error\n"); | |
696 | err = -EIO; | |
697 | } | |
698 | return err; | |
699 | } | |
700 | ||
701 | /* | |
702 | * Log buffer allocation routines: | |
703 | */ | |
704 | ||
18eba7aa | 705 | int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) |
470decc6 DK |
706 | { |
707 | unsigned long blocknr; | |
708 | ||
a931da6a | 709 | write_lock(&journal->j_state_lock); |
470decc6 DK |
710 | J_ASSERT(journal->j_free > 1); |
711 | ||
712 | blocknr = journal->j_head; | |
713 | journal->j_head++; | |
714 | journal->j_free--; | |
715 | if (journal->j_head == journal->j_last) | |
716 | journal->j_head = journal->j_first; | |
a931da6a | 717 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 718 | return jbd2_journal_bmap(journal, blocknr, retp); |
470decc6 DK |
719 | } |
720 | ||
721 | /* | |
722 | * Conversion of logical to physical block numbers for the journal | |
723 | * | |
724 | * On external journals the journal blocks are identity-mapped, so | |
725 | * this is a no-op. If needed, we can use j_blk_offset - everything is | |
726 | * ready. | |
727 | */ | |
f7f4bccb | 728 | int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, |
18eba7aa | 729 | unsigned long long *retp) |
470decc6 DK |
730 | { |
731 | int err = 0; | |
18eba7aa | 732 | unsigned long long ret; |
470decc6 DK |
733 | |
734 | if (journal->j_inode) { | |
735 | ret = bmap(journal->j_inode, blocknr); | |
736 | if (ret) | |
737 | *retp = ret; | |
738 | else { | |
470decc6 DK |
739 | printk(KERN_ALERT "%s: journal block not found " |
740 | "at offset %lu on %s\n", | |
05496769 | 741 | __func__, blocknr, journal->j_devname); |
470decc6 DK |
742 | err = -EIO; |
743 | __journal_abort_soft(journal, err); | |
744 | } | |
745 | } else { | |
746 | *retp = blocknr; /* +journal->j_blk_offset */ | |
747 | } | |
748 | return err; | |
749 | } | |
750 | ||
751 | /* | |
752 | * We play buffer_head aliasing tricks to write data/metadata blocks to | |
753 | * the journal without copying their contents, but for journal | |
754 | * descriptor blocks we do need to generate bona fide buffers. | |
755 | * | |
f7f4bccb | 756 | * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying |
470decc6 DK |
757 | * the buffer's contents they really should run flush_dcache_page(bh->b_page). |
758 | * But we don't bother doing that, so there will be coherency problems with | |
759 | * mmaps of blockdevs which hold live JBD-controlled filesystems. | |
760 | */ | |
f7f4bccb | 761 | struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal) |
470decc6 DK |
762 | { |
763 | struct buffer_head *bh; | |
18eba7aa | 764 | unsigned long long blocknr; |
470decc6 DK |
765 | int err; |
766 | ||
f7f4bccb | 767 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
768 | |
769 | if (err) | |
770 | return NULL; | |
771 | ||
772 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
4b905671 JK |
773 | if (!bh) |
774 | return NULL; | |
470decc6 DK |
775 | lock_buffer(bh); |
776 | memset(bh->b_data, 0, journal->j_blocksize); | |
777 | set_buffer_uptodate(bh); | |
778 | unlock_buffer(bh); | |
779 | BUFFER_TRACE(bh, "return this buffer"); | |
f7f4bccb | 780 | return jbd2_journal_add_journal_head(bh); |
470decc6 DK |
781 | } |
782 | ||
79feb521 JK |
783 | /* |
784 | * Return tid of the oldest transaction in the journal and block in the journal | |
785 | * where the transaction starts. | |
786 | * | |
787 | * If the journal is now empty, return which will be the next transaction ID | |
788 | * we will write and where will that transaction start. | |
789 | * | |
790 | * The return value is 0 if journal tail cannot be pushed any further, 1 if | |
791 | * it can. | |
792 | */ | |
793 | int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, | |
794 | unsigned long *block) | |
795 | { | |
796 | transaction_t *transaction; | |
797 | int ret; | |
798 | ||
799 | read_lock(&journal->j_state_lock); | |
800 | spin_lock(&journal->j_list_lock); | |
801 | transaction = journal->j_checkpoint_transactions; | |
802 | if (transaction) { | |
803 | *tid = transaction->t_tid; | |
804 | *block = transaction->t_log_start; | |
805 | } else if ((transaction = journal->j_committing_transaction) != NULL) { | |
806 | *tid = transaction->t_tid; | |
807 | *block = transaction->t_log_start; | |
808 | } else if ((transaction = journal->j_running_transaction) != NULL) { | |
809 | *tid = transaction->t_tid; | |
810 | *block = journal->j_head; | |
811 | } else { | |
812 | *tid = journal->j_transaction_sequence; | |
813 | *block = journal->j_head; | |
814 | } | |
815 | ret = tid_gt(*tid, journal->j_tail_sequence); | |
816 | spin_unlock(&journal->j_list_lock); | |
817 | read_unlock(&journal->j_state_lock); | |
818 | ||
819 | return ret; | |
820 | } | |
821 | ||
822 | /* | |
823 | * Update information in journal structure and in on disk journal superblock | |
824 | * about log tail. This function does not check whether information passed in | |
825 | * really pushes log tail further. It's responsibility of the caller to make | |
826 | * sure provided log tail information is valid (e.g. by holding | |
827 | * j_checkpoint_mutex all the time between computing log tail and calling this | |
828 | * function as is the case with jbd2_cleanup_journal_tail()). | |
829 | * | |
830 | * Requires j_checkpoint_mutex | |
831 | */ | |
832 | void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) | |
833 | { | |
834 | unsigned long freed; | |
835 | ||
836 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); | |
837 | ||
838 | /* | |
839 | * We cannot afford for write to remain in drive's caches since as | |
840 | * soon as we update j_tail, next transaction can start reusing journal | |
841 | * space and if we lose sb update during power failure we'd replay | |
842 | * old transaction with possibly newly overwritten data. | |
843 | */ | |
844 | jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA); | |
845 | write_lock(&journal->j_state_lock); | |
846 | freed = block - journal->j_tail; | |
847 | if (block < journal->j_tail) | |
848 | freed += journal->j_last - journal->j_first; | |
849 | ||
850 | trace_jbd2_update_log_tail(journal, tid, block, freed); | |
851 | jbd_debug(1, | |
852 | "Cleaning journal tail from %d to %d (offset %lu), " | |
853 | "freeing %lu\n", | |
854 | journal->j_tail_sequence, tid, block, freed); | |
855 | ||
856 | journal->j_free += freed; | |
857 | journal->j_tail_sequence = tid; | |
858 | journal->j_tail = block; | |
859 | write_unlock(&journal->j_state_lock); | |
860 | } | |
861 | ||
3339578f JK |
862 | /* |
863 | * This is a variaon of __jbd2_update_log_tail which checks for validity of | |
864 | * provided log tail and locks j_checkpoint_mutex. So it is safe against races | |
865 | * with other threads updating log tail. | |
866 | */ | |
867 | void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) | |
868 | { | |
869 | mutex_lock(&journal->j_checkpoint_mutex); | |
870 | if (tid_gt(tid, journal->j_tail_sequence)) | |
871 | __jbd2_update_log_tail(journal, tid, block); | |
872 | mutex_unlock(&journal->j_checkpoint_mutex); | |
873 | } | |
874 | ||
8e85fb3f JL |
875 | struct jbd2_stats_proc_session { |
876 | journal_t *journal; | |
877 | struct transaction_stats_s *stats; | |
878 | int start; | |
879 | int max; | |
880 | }; | |
881 | ||
8e85fb3f JL |
882 | static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) |
883 | { | |
884 | return *pos ? NULL : SEQ_START_TOKEN; | |
885 | } | |
886 | ||
887 | static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) | |
888 | { | |
889 | return NULL; | |
890 | } | |
891 | ||
892 | static int jbd2_seq_info_show(struct seq_file *seq, void *v) | |
893 | { | |
894 | struct jbd2_stats_proc_session *s = seq->private; | |
895 | ||
896 | if (v != SEQ_START_TOKEN) | |
897 | return 0; | |
bf699327 | 898 | seq_printf(seq, "%lu transaction, each up to %u blocks\n", |
8e85fb3f JL |
899 | s->stats->ts_tid, |
900 | s->journal->j_max_transaction_buffers); | |
901 | if (s->stats->ts_tid == 0) | |
902 | return 0; | |
903 | seq_printf(seq, "average: \n %ums waiting for transaction\n", | |
bf699327 | 904 | jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); |
8e85fb3f | 905 | seq_printf(seq, " %ums running transaction\n", |
bf699327 | 906 | jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); |
8e85fb3f | 907 | seq_printf(seq, " %ums transaction was being locked\n", |
bf699327 | 908 | jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); |
8e85fb3f | 909 | seq_printf(seq, " %ums flushing data (in ordered mode)\n", |
bf699327 | 910 | jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); |
8e85fb3f | 911 | seq_printf(seq, " %ums logging transaction\n", |
bf699327 | 912 | jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); |
c225aa57 SHT |
913 | seq_printf(seq, " %lluus average transaction commit time\n", |
914 | div_u64(s->journal->j_average_commit_time, 1000)); | |
8e85fb3f | 915 | seq_printf(seq, " %lu handles per transaction\n", |
bf699327 | 916 | s->stats->run.rs_handle_count / s->stats->ts_tid); |
8e85fb3f | 917 | seq_printf(seq, " %lu blocks per transaction\n", |
bf699327 | 918 | s->stats->run.rs_blocks / s->stats->ts_tid); |
8e85fb3f | 919 | seq_printf(seq, " %lu logged blocks per transaction\n", |
bf699327 | 920 | s->stats->run.rs_blocks_logged / s->stats->ts_tid); |
8e85fb3f JL |
921 | return 0; |
922 | } | |
923 | ||
924 | static void jbd2_seq_info_stop(struct seq_file *seq, void *v) | |
925 | { | |
926 | } | |
927 | ||
88e9d34c | 928 | static const struct seq_operations jbd2_seq_info_ops = { |
8e85fb3f JL |
929 | .start = jbd2_seq_info_start, |
930 | .next = jbd2_seq_info_next, | |
931 | .stop = jbd2_seq_info_stop, | |
932 | .show = jbd2_seq_info_show, | |
933 | }; | |
934 | ||
935 | static int jbd2_seq_info_open(struct inode *inode, struct file *file) | |
936 | { | |
937 | journal_t *journal = PDE(inode)->data; | |
938 | struct jbd2_stats_proc_session *s; | |
939 | int rc, size; | |
940 | ||
941 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
942 | if (s == NULL) | |
943 | return -ENOMEM; | |
944 | size = sizeof(struct transaction_stats_s); | |
945 | s->stats = kmalloc(size, GFP_KERNEL); | |
946 | if (s->stats == NULL) { | |
947 | kfree(s); | |
948 | return -ENOMEM; | |
949 | } | |
950 | spin_lock(&journal->j_history_lock); | |
951 | memcpy(s->stats, &journal->j_stats, size); | |
952 | s->journal = journal; | |
953 | spin_unlock(&journal->j_history_lock); | |
954 | ||
955 | rc = seq_open(file, &jbd2_seq_info_ops); | |
956 | if (rc == 0) { | |
957 | struct seq_file *m = file->private_data; | |
958 | m->private = s; | |
959 | } else { | |
960 | kfree(s->stats); | |
961 | kfree(s); | |
962 | } | |
963 | return rc; | |
964 | ||
965 | } | |
966 | ||
967 | static int jbd2_seq_info_release(struct inode *inode, struct file *file) | |
968 | { | |
969 | struct seq_file *seq = file->private_data; | |
970 | struct jbd2_stats_proc_session *s = seq->private; | |
971 | kfree(s->stats); | |
972 | kfree(s); | |
973 | return seq_release(inode, file); | |
974 | } | |
975 | ||
828c0950 | 976 | static const struct file_operations jbd2_seq_info_fops = { |
8e85fb3f JL |
977 | .owner = THIS_MODULE, |
978 | .open = jbd2_seq_info_open, | |
979 | .read = seq_read, | |
980 | .llseek = seq_lseek, | |
981 | .release = jbd2_seq_info_release, | |
982 | }; | |
983 | ||
984 | static struct proc_dir_entry *proc_jbd2_stats; | |
985 | ||
986 | static void jbd2_stats_proc_init(journal_t *journal) | |
987 | { | |
05496769 | 988 | journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); |
8e85fb3f | 989 | if (journal->j_proc_entry) { |
79da3664 DL |
990 | proc_create_data("info", S_IRUGO, journal->j_proc_entry, |
991 | &jbd2_seq_info_fops, journal); | |
8e85fb3f JL |
992 | } |
993 | } | |
994 | ||
995 | static void jbd2_stats_proc_exit(journal_t *journal) | |
996 | { | |
8e85fb3f | 997 | remove_proc_entry("info", journal->j_proc_entry); |
05496769 | 998 | remove_proc_entry(journal->j_devname, proc_jbd2_stats); |
8e85fb3f JL |
999 | } |
1000 | ||
470decc6 DK |
1001 | /* |
1002 | * Management for journal control blocks: functions to create and | |
1003 | * destroy journal_t structures, and to initialise and read existing | |
1004 | * journal blocks from disk. */ | |
1005 | ||
1006 | /* First: create and setup a journal_t object in memory. We initialise | |
1007 | * very few fields yet: that has to wait until we have created the | |
1008 | * journal structures from from scratch, or loaded them from disk. */ | |
1009 | ||
1010 | static journal_t * journal_init_common (void) | |
1011 | { | |
1012 | journal_t *journal; | |
1013 | int err; | |
1014 | ||
3ebfdf88 | 1015 | journal = kzalloc(sizeof(*journal), GFP_KERNEL); |
470decc6 | 1016 | if (!journal) |
b7271b0a | 1017 | return NULL; |
470decc6 DK |
1018 | |
1019 | init_waitqueue_head(&journal->j_wait_transaction_locked); | |
1020 | init_waitqueue_head(&journal->j_wait_logspace); | |
1021 | init_waitqueue_head(&journal->j_wait_done_commit); | |
1022 | init_waitqueue_head(&journal->j_wait_checkpoint); | |
1023 | init_waitqueue_head(&journal->j_wait_commit); | |
1024 | init_waitqueue_head(&journal->j_wait_updates); | |
1025 | mutex_init(&journal->j_barrier); | |
1026 | mutex_init(&journal->j_checkpoint_mutex); | |
1027 | spin_lock_init(&journal->j_revoke_lock); | |
1028 | spin_lock_init(&journal->j_list_lock); | |
a931da6a | 1029 | rwlock_init(&journal->j_state_lock); |
470decc6 | 1030 | |
cd02ff0b | 1031 | journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); |
30773840 TT |
1032 | journal->j_min_batch_time = 0; |
1033 | journal->j_max_batch_time = 15000; /* 15ms */ | |
470decc6 DK |
1034 | |
1035 | /* The journal is marked for error until we succeed with recovery! */ | |
f7f4bccb | 1036 | journal->j_flags = JBD2_ABORT; |
470decc6 DK |
1037 | |
1038 | /* Set up a default-sized revoke table for the new mount. */ | |
f7f4bccb | 1039 | err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); |
470decc6 DK |
1040 | if (err) { |
1041 | kfree(journal); | |
b7271b0a | 1042 | return NULL; |
470decc6 | 1043 | } |
8e85fb3f | 1044 | |
bf699327 | 1045 | spin_lock_init(&journal->j_history_lock); |
8e85fb3f | 1046 | |
470decc6 | 1047 | return journal; |
470decc6 DK |
1048 | } |
1049 | ||
f7f4bccb | 1050 | /* jbd2_journal_init_dev and jbd2_journal_init_inode: |
470decc6 DK |
1051 | * |
1052 | * Create a journal structure assigned some fixed set of disk blocks to | |
1053 | * the journal. We don't actually touch those disk blocks yet, but we | |
1054 | * need to set up all of the mapping information to tell the journaling | |
1055 | * system where the journal blocks are. | |
1056 | * | |
1057 | */ | |
1058 | ||
1059 | /** | |
5648ba5b | 1060 | * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure |
470decc6 DK |
1061 | * @bdev: Block device on which to create the journal |
1062 | * @fs_dev: Device which hold journalled filesystem for this journal. | |
1063 | * @start: Block nr Start of journal. | |
1064 | * @len: Length of the journal in blocks. | |
1065 | * @blocksize: blocksize of journalling device | |
5648ba5b RD |
1066 | * |
1067 | * Returns: a newly created journal_t * | |
470decc6 | 1068 | * |
f7f4bccb | 1069 | * jbd2_journal_init_dev creates a journal which maps a fixed contiguous |
470decc6 DK |
1070 | * range of blocks on an arbitrary block device. |
1071 | * | |
1072 | */ | |
f7f4bccb | 1073 | journal_t * jbd2_journal_init_dev(struct block_device *bdev, |
470decc6 | 1074 | struct block_device *fs_dev, |
18eba7aa | 1075 | unsigned long long start, int len, int blocksize) |
470decc6 DK |
1076 | { |
1077 | journal_t *journal = journal_init_common(); | |
1078 | struct buffer_head *bh; | |
05496769 | 1079 | char *p; |
470decc6 DK |
1080 | int n; |
1081 | ||
1082 | if (!journal) | |
1083 | return NULL; | |
1084 | ||
1085 | /* journal descriptor can store up to n blocks -bzzz */ | |
1086 | journal->j_blocksize = blocksize; | |
0587aa3d | 1087 | journal->j_dev = bdev; |
1088 | journal->j_fs_dev = fs_dev; | |
1089 | journal->j_blk_offset = start; | |
1090 | journal->j_maxlen = len; | |
1091 | bdevname(journal->j_dev, journal->j_devname); | |
1092 | p = journal->j_devname; | |
1093 | while ((p = strchr(p, '/'))) | |
1094 | *p = '!'; | |
4b905671 | 1095 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1096 | n = journal->j_blocksize / sizeof(journal_block_tag_t); |
1097 | journal->j_wbufsize = n; | |
1098 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1099 | if (!journal->j_wbuf) { | |
25985edc | 1100 | printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", |
329d291f | 1101 | __func__); |
4b905671 | 1102 | goto out_err; |
470decc6 | 1103 | } |
470decc6 DK |
1104 | |
1105 | bh = __getblk(journal->j_dev, start, journal->j_blocksize); | |
4b905671 JK |
1106 | if (!bh) { |
1107 | printk(KERN_ERR | |
1108 | "%s: Cannot get buffer for journal superblock\n", | |
1109 | __func__); | |
1110 | goto out_err; | |
1111 | } | |
470decc6 DK |
1112 | journal->j_sb_buffer = bh; |
1113 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
4b905671 | 1114 | |
470decc6 | 1115 | return journal; |
4b905671 | 1116 | out_err: |
7b02bec0 | 1117 | kfree(journal->j_wbuf); |
4b905671 JK |
1118 | jbd2_stats_proc_exit(journal); |
1119 | kfree(journal); | |
1120 | return NULL; | |
470decc6 DK |
1121 | } |
1122 | ||
1123 | /** | |
f7f4bccb | 1124 | * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. |
470decc6 DK |
1125 | * @inode: An inode to create the journal in |
1126 | * | |
f7f4bccb | 1127 | * jbd2_journal_init_inode creates a journal which maps an on-disk inode as |
470decc6 DK |
1128 | * the journal. The inode must exist already, must support bmap() and |
1129 | * must have all data blocks preallocated. | |
1130 | */ | |
f7f4bccb | 1131 | journal_t * jbd2_journal_init_inode (struct inode *inode) |
470decc6 DK |
1132 | { |
1133 | struct buffer_head *bh; | |
1134 | journal_t *journal = journal_init_common(); | |
05496769 | 1135 | char *p; |
470decc6 DK |
1136 | int err; |
1137 | int n; | |
18eba7aa | 1138 | unsigned long long blocknr; |
470decc6 DK |
1139 | |
1140 | if (!journal) | |
1141 | return NULL; | |
1142 | ||
1143 | journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev; | |
1144 | journal->j_inode = inode; | |
05496769 TT |
1145 | bdevname(journal->j_dev, journal->j_devname); |
1146 | p = journal->j_devname; | |
1147 | while ((p = strchr(p, '/'))) | |
1148 | *p = '!'; | |
1149 | p = journal->j_devname + strlen(journal->j_devname); | |
90576c0b | 1150 | sprintf(p, "-%lu", journal->j_inode->i_ino); |
470decc6 DK |
1151 | jbd_debug(1, |
1152 | "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n", | |
1153 | journal, inode->i_sb->s_id, inode->i_ino, | |
1154 | (long long) inode->i_size, | |
1155 | inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); | |
1156 | ||
1157 | journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits; | |
1158 | journal->j_blocksize = inode->i_sb->s_blocksize; | |
8e85fb3f | 1159 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1160 | |
1161 | /* journal descriptor can store up to n blocks -bzzz */ | |
1162 | n = journal->j_blocksize / sizeof(journal_block_tag_t); | |
1163 | journal->j_wbufsize = n; | |
1164 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1165 | if (!journal->j_wbuf) { | |
25985edc | 1166 | printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", |
329d291f | 1167 | __func__); |
4b905671 | 1168 | goto out_err; |
470decc6 DK |
1169 | } |
1170 | ||
f7f4bccb | 1171 | err = jbd2_journal_bmap(journal, 0, &blocknr); |
470decc6 DK |
1172 | /* If that failed, give up */ |
1173 | if (err) { | |
3c26bdb4 | 1174 | printk(KERN_ERR "%s: Cannot locate journal superblock\n", |
329d291f | 1175 | __func__); |
4b905671 | 1176 | goto out_err; |
470decc6 DK |
1177 | } |
1178 | ||
1179 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
4b905671 JK |
1180 | if (!bh) { |
1181 | printk(KERN_ERR | |
1182 | "%s: Cannot get buffer for journal superblock\n", | |
1183 | __func__); | |
1184 | goto out_err; | |
1185 | } | |
470decc6 DK |
1186 | journal->j_sb_buffer = bh; |
1187 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
1188 | ||
1189 | return journal; | |
4b905671 | 1190 | out_err: |
7b02bec0 | 1191 | kfree(journal->j_wbuf); |
4b905671 JK |
1192 | jbd2_stats_proc_exit(journal); |
1193 | kfree(journal); | |
1194 | return NULL; | |
470decc6 DK |
1195 | } |
1196 | ||
1197 | /* | |
1198 | * If the journal init or create aborts, we need to mark the journal | |
1199 | * superblock as being NULL to prevent the journal destroy from writing | |
1200 | * back a bogus superblock. | |
1201 | */ | |
1202 | static void journal_fail_superblock (journal_t *journal) | |
1203 | { | |
1204 | struct buffer_head *bh = journal->j_sb_buffer; | |
1205 | brelse(bh); | |
1206 | journal->j_sb_buffer = NULL; | |
1207 | } | |
1208 | ||
1209 | /* | |
1210 | * Given a journal_t structure, initialise the various fields for | |
1211 | * startup of a new journaling session. We use this both when creating | |
1212 | * a journal, and after recovering an old journal to reset it for | |
1213 | * subsequent use. | |
1214 | */ | |
1215 | ||
1216 | static int journal_reset(journal_t *journal) | |
1217 | { | |
1218 | journal_superblock_t *sb = journal->j_superblock; | |
18eba7aa | 1219 | unsigned long long first, last; |
470decc6 DK |
1220 | |
1221 | first = be32_to_cpu(sb->s_first); | |
1222 | last = be32_to_cpu(sb->s_maxlen); | |
f6f50e28 | 1223 | if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) { |
f2a44523 | 1224 | printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n", |
f6f50e28 JK |
1225 | first, last); |
1226 | journal_fail_superblock(journal); | |
1227 | return -EINVAL; | |
1228 | } | |
470decc6 DK |
1229 | |
1230 | journal->j_first = first; | |
1231 | journal->j_last = last; | |
1232 | ||
1233 | journal->j_head = first; | |
1234 | journal->j_tail = first; | |
1235 | journal->j_free = last - first; | |
1236 | ||
1237 | journal->j_tail_sequence = journal->j_transaction_sequence; | |
1238 | journal->j_commit_sequence = journal->j_transaction_sequence - 1; | |
1239 | journal->j_commit_request = journal->j_commit_sequence; | |
1240 | ||
1241 | journal->j_max_transaction_buffers = journal->j_maxlen / 4; | |
1242 | ||
470decc6 DK |
1243 | /* |
1244 | * As a special case, if the on-disk copy is already marked as needing | |
24bcc89c JK |
1245 | * no recovery (s_start == 0), then we can safely defer the superblock |
1246 | * update until the next commit by setting JBD2_FLUSHED. This avoids | |
470decc6 DK |
1247 | * attempting a write to a potential-readonly device. |
1248 | */ | |
24bcc89c | 1249 | if (sb->s_start == 0) { |
f2a44523 | 1250 | jbd_debug(1, "JBD2: Skipping superblock update on recovered sb " |
470decc6 DK |
1251 | "(start %ld, seq %d, errno %d)\n", |
1252 | journal->j_tail, journal->j_tail_sequence, | |
1253 | journal->j_errno); | |
24bcc89c JK |
1254 | journal->j_flags |= JBD2_FLUSHED; |
1255 | } else { | |
a78bb11d JK |
1256 | /* Lock here to make assertions happy... */ |
1257 | mutex_lock(&journal->j_checkpoint_mutex); | |
79feb521 JK |
1258 | /* |
1259 | * Update log tail information. We use WRITE_FUA since new | |
1260 | * transaction will start reusing journal space and so we | |
1261 | * must make sure information about current log tail is on | |
1262 | * disk before that. | |
1263 | */ | |
1264 | jbd2_journal_update_sb_log_tail(journal, | |
1265 | journal->j_tail_sequence, | |
1266 | journal->j_tail, | |
1267 | WRITE_FUA); | |
a78bb11d | 1268 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 | 1269 | } |
24bcc89c JK |
1270 | return jbd2_journal_start_thread(journal); |
1271 | } | |
470decc6 | 1272 | |
79feb521 | 1273 | static void jbd2_write_superblock(journal_t *journal, int write_op) |
24bcc89c JK |
1274 | { |
1275 | struct buffer_head *bh = journal->j_sb_buffer; | |
79feb521 | 1276 | int ret; |
470decc6 | 1277 | |
79feb521 JK |
1278 | trace_jbd2_write_superblock(journal, write_op); |
1279 | if (!(journal->j_flags & JBD2_BARRIER)) | |
1280 | write_op &= ~(REQ_FUA | REQ_FLUSH); | |
1281 | lock_buffer(bh); | |
914258bf TT |
1282 | if (buffer_write_io_error(bh)) { |
1283 | /* | |
1284 | * Oh, dear. A previous attempt to write the journal | |
1285 | * superblock failed. This could happen because the | |
1286 | * USB device was yanked out. Or it could happen to | |
1287 | * be a transient write error and maybe the block will | |
1288 | * be remapped. Nothing we can do but to retry the | |
1289 | * write and hope for the best. | |
1290 | */ | |
1291 | printk(KERN_ERR "JBD2: previous I/O error detected " | |
1292 | "for journal superblock update for %s.\n", | |
1293 | journal->j_devname); | |
1294 | clear_buffer_write_io_error(bh); | |
1295 | set_buffer_uptodate(bh); | |
1296 | } | |
79feb521 JK |
1297 | get_bh(bh); |
1298 | bh->b_end_io = end_buffer_write_sync; | |
1299 | ret = submit_bh(write_op, bh); | |
1300 | wait_on_buffer(bh); | |
24bcc89c | 1301 | if (buffer_write_io_error(bh)) { |
24bcc89c JK |
1302 | clear_buffer_write_io_error(bh); |
1303 | set_buffer_uptodate(bh); | |
79feb521 JK |
1304 | ret = -EIO; |
1305 | } | |
1306 | if (ret) { | |
1307 | printk(KERN_ERR "JBD2: Error %d detected when updating " | |
1308 | "journal superblock for %s.\n", ret, | |
1309 | journal->j_devname); | |
24bcc89c JK |
1310 | } |
1311 | } | |
1312 | ||
1313 | /** | |
1314 | * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. | |
1315 | * @journal: The journal to update. | |
79feb521 JK |
1316 | * @tail_tid: TID of the new transaction at the tail of the log |
1317 | * @tail_block: The first block of the transaction at the tail of the log | |
1318 | * @write_op: With which operation should we write the journal sb | |
24bcc89c JK |
1319 | * |
1320 | * Update a journal's superblock information about log tail and write it to | |
1321 | * disk, waiting for the IO to complete. | |
1322 | */ | |
79feb521 JK |
1323 | void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, |
1324 | unsigned long tail_block, int write_op) | |
24bcc89c JK |
1325 | { |
1326 | journal_superblock_t *sb = journal->j_superblock; | |
1327 | ||
a78bb11d | 1328 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); |
79feb521 JK |
1329 | jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", |
1330 | tail_block, tail_tid); | |
470decc6 | 1331 | |
79feb521 JK |
1332 | sb->s_sequence = cpu_to_be32(tail_tid); |
1333 | sb->s_start = cpu_to_be32(tail_block); | |
470decc6 | 1334 | |
79feb521 | 1335 | jbd2_write_superblock(journal, write_op); |
470decc6 | 1336 | |
24bcc89c JK |
1337 | /* Log is no longer empty */ |
1338 | write_lock(&journal->j_state_lock); | |
1339 | WARN_ON(!sb->s_sequence); | |
1340 | journal->j_flags &= ~JBD2_FLUSHED; | |
1341 | write_unlock(&journal->j_state_lock); | |
1342 | } | |
2201c590 | 1343 | |
24bcc89c JK |
1344 | /** |
1345 | * jbd2_mark_journal_empty() - Mark on disk journal as empty. | |
1346 | * @journal: The journal to update. | |
1347 | * | |
1348 | * Update a journal's dynamic superblock fields to show that journal is empty. | |
1349 | * Write updated superblock to disk waiting for IO to complete. | |
1350 | */ | |
1351 | static void jbd2_mark_journal_empty(journal_t *journal) | |
1352 | { | |
1353 | journal_superblock_t *sb = journal->j_superblock; | |
914258bf | 1354 | |
a78bb11d | 1355 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); |
a931da6a | 1356 | read_lock(&journal->j_state_lock); |
24bcc89c JK |
1357 | jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n", |
1358 | journal->j_tail_sequence); | |
470decc6 DK |
1359 | |
1360 | sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); | |
24bcc89c | 1361 | sb->s_start = cpu_to_be32(0); |
a931da6a | 1362 | read_unlock(&journal->j_state_lock); |
470decc6 | 1363 | |
79feb521 | 1364 | jbd2_write_superblock(journal, WRITE_FUA); |
470decc6 | 1365 | |
24bcc89c | 1366 | /* Log is no longer empty */ |
a931da6a | 1367 | write_lock(&journal->j_state_lock); |
24bcc89c | 1368 | journal->j_flags |= JBD2_FLUSHED; |
a931da6a | 1369 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1370 | } |
1371 | ||
24bcc89c JK |
1372 | |
1373 | /** | |
1374 | * jbd2_journal_update_sb_errno() - Update error in the journal. | |
1375 | * @journal: The journal to update. | |
1376 | * | |
1377 | * Update a journal's errno. Write updated superblock to disk waiting for IO | |
1378 | * to complete. | |
1379 | */ | |
1380 | static void jbd2_journal_update_sb_errno(journal_t *journal) | |
1381 | { | |
1382 | journal_superblock_t *sb = journal->j_superblock; | |
1383 | ||
1384 | read_lock(&journal->j_state_lock); | |
1385 | jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", | |
1386 | journal->j_errno); | |
1387 | sb->s_errno = cpu_to_be32(journal->j_errno); | |
4fd5ea43 | 1388 | jbd2_superblock_csum_set(journal, sb); |
24bcc89c JK |
1389 | read_unlock(&journal->j_state_lock); |
1390 | ||
79feb521 | 1391 | jbd2_write_superblock(journal, WRITE_SYNC); |
24bcc89c JK |
1392 | } |
1393 | ||
470decc6 DK |
1394 | /* |
1395 | * Read the superblock for a given journal, performing initial | |
1396 | * validation of the format. | |
1397 | */ | |
470decc6 DK |
1398 | static int journal_get_superblock(journal_t *journal) |
1399 | { | |
1400 | struct buffer_head *bh; | |
1401 | journal_superblock_t *sb; | |
1402 | int err = -EIO; | |
1403 | ||
1404 | bh = journal->j_sb_buffer; | |
1405 | ||
1406 | J_ASSERT(bh != NULL); | |
1407 | if (!buffer_uptodate(bh)) { | |
1408 | ll_rw_block(READ, 1, &bh); | |
1409 | wait_on_buffer(bh); | |
1410 | if (!buffer_uptodate(bh)) { | |
f2a44523 EG |
1411 | printk(KERN_ERR |
1412 | "JBD2: IO error reading journal superblock\n"); | |
470decc6 DK |
1413 | goto out; |
1414 | } | |
1415 | } | |
1416 | ||
25ed6e8a DW |
1417 | if (buffer_verified(bh)) |
1418 | return 0; | |
1419 | ||
470decc6 DK |
1420 | sb = journal->j_superblock; |
1421 | ||
1422 | err = -EINVAL; | |
1423 | ||
f7f4bccb | 1424 | if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || |
470decc6 | 1425 | sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { |
f2a44523 | 1426 | printk(KERN_WARNING "JBD2: no valid journal superblock found\n"); |
470decc6 DK |
1427 | goto out; |
1428 | } | |
1429 | ||
1430 | switch(be32_to_cpu(sb->s_header.h_blocktype)) { | |
f7f4bccb | 1431 | case JBD2_SUPERBLOCK_V1: |
470decc6 DK |
1432 | journal->j_format_version = 1; |
1433 | break; | |
f7f4bccb | 1434 | case JBD2_SUPERBLOCK_V2: |
470decc6 DK |
1435 | journal->j_format_version = 2; |
1436 | break; | |
1437 | default: | |
f2a44523 | 1438 | printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n"); |
470decc6 DK |
1439 | goto out; |
1440 | } | |
1441 | ||
1442 | if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen) | |
1443 | journal->j_maxlen = be32_to_cpu(sb->s_maxlen); | |
1444 | else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) { | |
f2a44523 | 1445 | printk(KERN_WARNING "JBD2: journal file too short\n"); |
470decc6 DK |
1446 | goto out; |
1447 | } | |
1448 | ||
8762202d EG |
1449 | if (be32_to_cpu(sb->s_first) == 0 || |
1450 | be32_to_cpu(sb->s_first) >= journal->j_maxlen) { | |
1451 | printk(KERN_WARNING | |
1452 | "JBD2: Invalid start block of journal: %u\n", | |
1453 | be32_to_cpu(sb->s_first)); | |
1454 | goto out; | |
1455 | } | |
1456 | ||
25ed6e8a DW |
1457 | if (JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM) && |
1458 | JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) { | |
1459 | /* Can't have checksum v1 and v2 on at the same time! */ | |
1460 | printk(KERN_ERR "JBD: Can't enable checksumming v1 and v2 " | |
1461 | "at the same time!\n"); | |
1462 | goto out; | |
1463 | } | |
1464 | ||
1465 | if (!jbd2_verify_csum_type(journal, sb)) { | |
1466 | printk(KERN_ERR "JBD: Unknown checksum type\n"); | |
1467 | goto out; | |
1468 | } | |
1469 | ||
01b5adce DW |
1470 | /* Load the checksum driver */ |
1471 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) { | |
1472 | journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); | |
1473 | if (IS_ERR(journal->j_chksum_driver)) { | |
1474 | printk(KERN_ERR "JBD: Cannot load crc32c driver.\n"); | |
1475 | err = PTR_ERR(journal->j_chksum_driver); | |
1476 | journal->j_chksum_driver = NULL; | |
1477 | goto out; | |
1478 | } | |
1479 | } | |
1480 | ||
4fd5ea43 DW |
1481 | /* Check superblock checksum */ |
1482 | if (!jbd2_superblock_csum_verify(journal, sb)) { | |
1483 | printk(KERN_ERR "JBD: journal checksum error\n"); | |
1484 | goto out; | |
1485 | } | |
1486 | ||
1487 | /* Precompute checksum seed for all metadata */ | |
1488 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
1489 | journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, | |
1490 | sizeof(sb->s_uuid)); | |
1491 | ||
25ed6e8a DW |
1492 | set_buffer_verified(bh); |
1493 | ||
470decc6 DK |
1494 | return 0; |
1495 | ||
1496 | out: | |
1497 | journal_fail_superblock(journal); | |
1498 | return err; | |
1499 | } | |
1500 | ||
1501 | /* | |
1502 | * Load the on-disk journal superblock and read the key fields into the | |
1503 | * journal_t. | |
1504 | */ | |
1505 | ||
1506 | static int load_superblock(journal_t *journal) | |
1507 | { | |
1508 | int err; | |
1509 | journal_superblock_t *sb; | |
1510 | ||
1511 | err = journal_get_superblock(journal); | |
1512 | if (err) | |
1513 | return err; | |
1514 | ||
1515 | sb = journal->j_superblock; | |
1516 | ||
1517 | journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); | |
1518 | journal->j_tail = be32_to_cpu(sb->s_start); | |
1519 | journal->j_first = be32_to_cpu(sb->s_first); | |
1520 | journal->j_last = be32_to_cpu(sb->s_maxlen); | |
1521 | journal->j_errno = be32_to_cpu(sb->s_errno); | |
1522 | ||
1523 | return 0; | |
1524 | } | |
1525 | ||
1526 | ||
1527 | /** | |
f7f4bccb | 1528 | * int jbd2_journal_load() - Read journal from disk. |
470decc6 DK |
1529 | * @journal: Journal to act on. |
1530 | * | |
1531 | * Given a journal_t structure which tells us which disk blocks contain | |
1532 | * a journal, read the journal from disk to initialise the in-memory | |
1533 | * structures. | |
1534 | */ | |
f7f4bccb | 1535 | int jbd2_journal_load(journal_t *journal) |
470decc6 DK |
1536 | { |
1537 | int err; | |
1538 | journal_superblock_t *sb; | |
1539 | ||
1540 | err = load_superblock(journal); | |
1541 | if (err) | |
1542 | return err; | |
1543 | ||
1544 | sb = journal->j_superblock; | |
1545 | /* If this is a V2 superblock, then we have to check the | |
1546 | * features flags on it. */ | |
1547 | ||
1548 | if (journal->j_format_version >= 2) { | |
1549 | if ((sb->s_feature_ro_compat & | |
f7f4bccb | 1550 | ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || |
470decc6 | 1551 | (sb->s_feature_incompat & |
f7f4bccb | 1552 | ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { |
f2a44523 EG |
1553 | printk(KERN_WARNING |
1554 | "JBD2: Unrecognised features on journal\n"); | |
470decc6 DK |
1555 | return -EINVAL; |
1556 | } | |
1557 | } | |
1558 | ||
d2eecb03 TT |
1559 | /* |
1560 | * Create a slab for this blocksize | |
1561 | */ | |
1562 | err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize)); | |
1563 | if (err) | |
1564 | return err; | |
1565 | ||
470decc6 DK |
1566 | /* Let the recovery code check whether it needs to recover any |
1567 | * data from the journal. */ | |
f7f4bccb | 1568 | if (jbd2_journal_recover(journal)) |
470decc6 DK |
1569 | goto recovery_error; |
1570 | ||
e6a47428 TT |
1571 | if (journal->j_failed_commit) { |
1572 | printk(KERN_ERR "JBD2: journal transaction %u on %s " | |
1573 | "is corrupt.\n", journal->j_failed_commit, | |
1574 | journal->j_devname); | |
1575 | return -EIO; | |
1576 | } | |
1577 | ||
470decc6 DK |
1578 | /* OK, we've finished with the dynamic journal bits: |
1579 | * reinitialise the dynamic contents of the superblock in memory | |
1580 | * and reset them on disk. */ | |
1581 | if (journal_reset(journal)) | |
1582 | goto recovery_error; | |
1583 | ||
f7f4bccb MC |
1584 | journal->j_flags &= ~JBD2_ABORT; |
1585 | journal->j_flags |= JBD2_LOADED; | |
470decc6 DK |
1586 | return 0; |
1587 | ||
1588 | recovery_error: | |
f2a44523 | 1589 | printk(KERN_WARNING "JBD2: recovery failed\n"); |
470decc6 DK |
1590 | return -EIO; |
1591 | } | |
1592 | ||
1593 | /** | |
f7f4bccb | 1594 | * void jbd2_journal_destroy() - Release a journal_t structure. |
470decc6 DK |
1595 | * @journal: Journal to act on. |
1596 | * | |
1597 | * Release a journal_t structure once it is no longer in use by the | |
1598 | * journaled object. | |
44519faf | 1599 | * Return <0 if we couldn't clean up the journal. |
470decc6 | 1600 | */ |
44519faf | 1601 | int jbd2_journal_destroy(journal_t *journal) |
470decc6 | 1602 | { |
44519faf HK |
1603 | int err = 0; |
1604 | ||
470decc6 DK |
1605 | /* Wait for the commit thread to wake up and die. */ |
1606 | journal_kill_thread(journal); | |
1607 | ||
1608 | /* Force a final log commit */ | |
1609 | if (journal->j_running_transaction) | |
f7f4bccb | 1610 | jbd2_journal_commit_transaction(journal); |
470decc6 DK |
1611 | |
1612 | /* Force any old transactions to disk */ | |
1613 | ||
1614 | /* Totally anal locking here... */ | |
1615 | spin_lock(&journal->j_list_lock); | |
1616 | while (journal->j_checkpoint_transactions != NULL) { | |
1617 | spin_unlock(&journal->j_list_lock); | |
1a0d3786 | 1618 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1619 | jbd2_log_do_checkpoint(journal); |
1a0d3786 | 1620 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
1621 | spin_lock(&journal->j_list_lock); |
1622 | } | |
1623 | ||
1624 | J_ASSERT(journal->j_running_transaction == NULL); | |
1625 | J_ASSERT(journal->j_committing_transaction == NULL); | |
1626 | J_ASSERT(journal->j_checkpoint_transactions == NULL); | |
1627 | spin_unlock(&journal->j_list_lock); | |
1628 | ||
470decc6 | 1629 | if (journal->j_sb_buffer) { |
44519faf | 1630 | if (!is_journal_aborted(journal)) { |
a78bb11d | 1631 | mutex_lock(&journal->j_checkpoint_mutex); |
24bcc89c | 1632 | jbd2_mark_journal_empty(journal); |
a78bb11d JK |
1633 | mutex_unlock(&journal->j_checkpoint_mutex); |
1634 | } else | |
44519faf | 1635 | err = -EIO; |
470decc6 DK |
1636 | brelse(journal->j_sb_buffer); |
1637 | } | |
1638 | ||
8e85fb3f JL |
1639 | if (journal->j_proc_entry) |
1640 | jbd2_stats_proc_exit(journal); | |
470decc6 DK |
1641 | if (journal->j_inode) |
1642 | iput(journal->j_inode); | |
1643 | if (journal->j_revoke) | |
f7f4bccb | 1644 | jbd2_journal_destroy_revoke(journal); |
01b5adce DW |
1645 | if (journal->j_chksum_driver) |
1646 | crypto_free_shash(journal->j_chksum_driver); | |
470decc6 DK |
1647 | kfree(journal->j_wbuf); |
1648 | kfree(journal); | |
44519faf HK |
1649 | |
1650 | return err; | |
470decc6 DK |
1651 | } |
1652 | ||
1653 | ||
1654 | /** | |
f7f4bccb | 1655 | *int jbd2_journal_check_used_features () - Check if features specified are used. |
470decc6 DK |
1656 | * @journal: Journal to check. |
1657 | * @compat: bitmask of compatible features | |
1658 | * @ro: bitmask of features that force read-only mount | |
1659 | * @incompat: bitmask of incompatible features | |
1660 | * | |
1661 | * Check whether the journal uses all of a given set of | |
1662 | * features. Return true (non-zero) if it does. | |
1663 | **/ | |
1664 | ||
f7f4bccb | 1665 | int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1666 | unsigned long ro, unsigned long incompat) |
1667 | { | |
1668 | journal_superblock_t *sb; | |
1669 | ||
1670 | if (!compat && !ro && !incompat) | |
1671 | return 1; | |
1113e1b5 PL |
1672 | /* Load journal superblock if it is not loaded yet. */ |
1673 | if (journal->j_format_version == 0 && | |
1674 | journal_get_superblock(journal) != 0) | |
1675 | return 0; | |
470decc6 DK |
1676 | if (journal->j_format_version == 1) |
1677 | return 0; | |
1678 | ||
1679 | sb = journal->j_superblock; | |
1680 | ||
1681 | if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && | |
1682 | ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && | |
1683 | ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) | |
1684 | return 1; | |
1685 | ||
1686 | return 0; | |
1687 | } | |
1688 | ||
1689 | /** | |
f7f4bccb | 1690 | * int jbd2_journal_check_available_features() - Check feature set in journalling layer |
470decc6 DK |
1691 | * @journal: Journal to check. |
1692 | * @compat: bitmask of compatible features | |
1693 | * @ro: bitmask of features that force read-only mount | |
1694 | * @incompat: bitmask of incompatible features | |
1695 | * | |
1696 | * Check whether the journaling code supports the use of | |
1697 | * all of a given set of features on this journal. Return true | |
1698 | * (non-zero) if it can. */ | |
1699 | ||
f7f4bccb | 1700 | int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1701 | unsigned long ro, unsigned long incompat) |
1702 | { | |
470decc6 DK |
1703 | if (!compat && !ro && !incompat) |
1704 | return 1; | |
1705 | ||
470decc6 DK |
1706 | /* We can support any known requested features iff the |
1707 | * superblock is in version 2. Otherwise we fail to support any | |
1708 | * extended sb features. */ | |
1709 | ||
1710 | if (journal->j_format_version != 2) | |
1711 | return 0; | |
1712 | ||
f7f4bccb MC |
1713 | if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && |
1714 | (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && | |
1715 | (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) | |
470decc6 DK |
1716 | return 1; |
1717 | ||
1718 | return 0; | |
1719 | } | |
1720 | ||
1721 | /** | |
f7f4bccb | 1722 | * int jbd2_journal_set_features () - Mark a given journal feature in the superblock |
470decc6 DK |
1723 | * @journal: Journal to act on. |
1724 | * @compat: bitmask of compatible features | |
1725 | * @ro: bitmask of features that force read-only mount | |
1726 | * @incompat: bitmask of incompatible features | |
1727 | * | |
1728 | * Mark a given journal feature as present on the | |
1729 | * superblock. Returns true if the requested features could be set. | |
1730 | * | |
1731 | */ | |
1732 | ||
f7f4bccb | 1733 | int jbd2_journal_set_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1734 | unsigned long ro, unsigned long incompat) |
1735 | { | |
25ed6e8a DW |
1736 | #define INCOMPAT_FEATURE_ON(f) \ |
1737 | ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f))) | |
1738 | #define COMPAT_FEATURE_ON(f) \ | |
1739 | ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f))) | |
470decc6 DK |
1740 | journal_superblock_t *sb; |
1741 | ||
f7f4bccb | 1742 | if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) |
470decc6 DK |
1743 | return 1; |
1744 | ||
f7f4bccb | 1745 | if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) |
470decc6 DK |
1746 | return 0; |
1747 | ||
25ed6e8a DW |
1748 | /* Asking for checksumming v2 and v1? Only give them v2. */ |
1749 | if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2 && | |
1750 | compat & JBD2_FEATURE_COMPAT_CHECKSUM) | |
1751 | compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM; | |
1752 | ||
470decc6 DK |
1753 | jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", |
1754 | compat, ro, incompat); | |
1755 | ||
1756 | sb = journal->j_superblock; | |
1757 | ||
25ed6e8a DW |
1758 | /* If enabling v2 checksums, update superblock */ |
1759 | if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V2)) { | |
1760 | sb->s_checksum_type = JBD2_CRC32C_CHKSUM; | |
1761 | sb->s_feature_compat &= | |
1762 | ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM); | |
01b5adce DW |
1763 | |
1764 | /* Load the checksum driver */ | |
1765 | if (journal->j_chksum_driver == NULL) { | |
1766 | journal->j_chksum_driver = crypto_alloc_shash("crc32c", | |
1767 | 0, 0); | |
1768 | if (IS_ERR(journal->j_chksum_driver)) { | |
1769 | printk(KERN_ERR "JBD: Cannot load crc32c " | |
1770 | "driver.\n"); | |
1771 | journal->j_chksum_driver = NULL; | |
1772 | return 0; | |
1773 | } | |
1774 | } | |
4fd5ea43 DW |
1775 | |
1776 | /* Precompute checksum seed for all metadata */ | |
1777 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, | |
1778 | JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
1779 | journal->j_csum_seed = jbd2_chksum(journal, ~0, | |
1780 | sb->s_uuid, | |
1781 | sizeof(sb->s_uuid)); | |
25ed6e8a DW |
1782 | } |
1783 | ||
1784 | /* If enabling v1 checksums, downgrade superblock */ | |
1785 | if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM)) | |
1786 | sb->s_feature_incompat &= | |
1787 | ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2); | |
1788 | ||
470decc6 DK |
1789 | sb->s_feature_compat |= cpu_to_be32(compat); |
1790 | sb->s_feature_ro_compat |= cpu_to_be32(ro); | |
1791 | sb->s_feature_incompat |= cpu_to_be32(incompat); | |
1792 | ||
1793 | return 1; | |
25ed6e8a DW |
1794 | #undef COMPAT_FEATURE_ON |
1795 | #undef INCOMPAT_FEATURE_ON | |
470decc6 DK |
1796 | } |
1797 | ||
818d276c GS |
1798 | /* |
1799 | * jbd2_journal_clear_features () - Clear a given journal feature in the | |
1800 | * superblock | |
1801 | * @journal: Journal to act on. | |
1802 | * @compat: bitmask of compatible features | |
1803 | * @ro: bitmask of features that force read-only mount | |
1804 | * @incompat: bitmask of incompatible features | |
1805 | * | |
1806 | * Clear a given journal feature as present on the | |
1807 | * superblock. | |
1808 | */ | |
1809 | void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, | |
1810 | unsigned long ro, unsigned long incompat) | |
1811 | { | |
1812 | journal_superblock_t *sb; | |
1813 | ||
1814 | jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", | |
1815 | compat, ro, incompat); | |
1816 | ||
1817 | sb = journal->j_superblock; | |
1818 | ||
1819 | sb->s_feature_compat &= ~cpu_to_be32(compat); | |
1820 | sb->s_feature_ro_compat &= ~cpu_to_be32(ro); | |
1821 | sb->s_feature_incompat &= ~cpu_to_be32(incompat); | |
1822 | } | |
1823 | EXPORT_SYMBOL(jbd2_journal_clear_features); | |
470decc6 | 1824 | |
470decc6 | 1825 | /** |
f7f4bccb | 1826 | * int jbd2_journal_flush () - Flush journal |
470decc6 DK |
1827 | * @journal: Journal to act on. |
1828 | * | |
1829 | * Flush all data for a given journal to disk and empty the journal. | |
1830 | * Filesystems can use this when remounting readonly to ensure that | |
1831 | * recovery does not need to happen on remount. | |
1832 | */ | |
1833 | ||
f7f4bccb | 1834 | int jbd2_journal_flush(journal_t *journal) |
470decc6 DK |
1835 | { |
1836 | int err = 0; | |
1837 | transaction_t *transaction = NULL; | |
470decc6 | 1838 | |
a931da6a | 1839 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1840 | |
1841 | /* Force everything buffered to the log... */ | |
1842 | if (journal->j_running_transaction) { | |
1843 | transaction = journal->j_running_transaction; | |
f7f4bccb | 1844 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1845 | } else if (journal->j_committing_transaction) |
1846 | transaction = journal->j_committing_transaction; | |
1847 | ||
1848 | /* Wait for the log commit to complete... */ | |
1849 | if (transaction) { | |
1850 | tid_t tid = transaction->t_tid; | |
1851 | ||
a931da6a | 1852 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 1853 | jbd2_log_wait_commit(journal, tid); |
470decc6 | 1854 | } else { |
a931da6a | 1855 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1856 | } |
1857 | ||
1858 | /* ...and flush everything in the log out to disk. */ | |
1859 | spin_lock(&journal->j_list_lock); | |
1860 | while (!err && journal->j_checkpoint_transactions != NULL) { | |
1861 | spin_unlock(&journal->j_list_lock); | |
44519faf | 1862 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1863 | err = jbd2_log_do_checkpoint(journal); |
44519faf | 1864 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
1865 | spin_lock(&journal->j_list_lock); |
1866 | } | |
1867 | spin_unlock(&journal->j_list_lock); | |
44519faf HK |
1868 | |
1869 | if (is_journal_aborted(journal)) | |
1870 | return -EIO; | |
1871 | ||
a78bb11d | 1872 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1873 | jbd2_cleanup_journal_tail(journal); |
470decc6 DK |
1874 | |
1875 | /* Finally, mark the journal as really needing no recovery. | |
1876 | * This sets s_start==0 in the underlying superblock, which is | |
1877 | * the magic code for a fully-recovered superblock. Any future | |
1878 | * commits of data to the journal will restore the current | |
1879 | * s_start value. */ | |
24bcc89c | 1880 | jbd2_mark_journal_empty(journal); |
a78bb11d | 1881 | mutex_unlock(&journal->j_checkpoint_mutex); |
a931da6a | 1882 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1883 | J_ASSERT(!journal->j_running_transaction); |
1884 | J_ASSERT(!journal->j_committing_transaction); | |
1885 | J_ASSERT(!journal->j_checkpoint_transactions); | |
1886 | J_ASSERT(journal->j_head == journal->j_tail); | |
1887 | J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); | |
a931da6a | 1888 | write_unlock(&journal->j_state_lock); |
44519faf | 1889 | return 0; |
470decc6 DK |
1890 | } |
1891 | ||
1892 | /** | |
f7f4bccb | 1893 | * int jbd2_journal_wipe() - Wipe journal contents |
470decc6 DK |
1894 | * @journal: Journal to act on. |
1895 | * @write: flag (see below) | |
1896 | * | |
1897 | * Wipe out all of the contents of a journal, safely. This will produce | |
1898 | * a warning if the journal contains any valid recovery information. | |
f7f4bccb | 1899 | * Must be called between journal_init_*() and jbd2_journal_load(). |
470decc6 DK |
1900 | * |
1901 | * If 'write' is non-zero, then we wipe out the journal on disk; otherwise | |
1902 | * we merely suppress recovery. | |
1903 | */ | |
1904 | ||
f7f4bccb | 1905 | int jbd2_journal_wipe(journal_t *journal, int write) |
470decc6 | 1906 | { |
470decc6 DK |
1907 | int err = 0; |
1908 | ||
f7f4bccb | 1909 | J_ASSERT (!(journal->j_flags & JBD2_LOADED)); |
470decc6 DK |
1910 | |
1911 | err = load_superblock(journal); | |
1912 | if (err) | |
1913 | return err; | |
1914 | ||
470decc6 DK |
1915 | if (!journal->j_tail) |
1916 | goto no_recovery; | |
1917 | ||
f2a44523 | 1918 | printk(KERN_WARNING "JBD2: %s recovery information on journal\n", |
470decc6 DK |
1919 | write ? "Clearing" : "Ignoring"); |
1920 | ||
f7f4bccb | 1921 | err = jbd2_journal_skip_recovery(journal); |
a78bb11d JK |
1922 | if (write) { |
1923 | /* Lock to make assertions happy... */ | |
1924 | mutex_lock(&journal->j_checkpoint_mutex); | |
24bcc89c | 1925 | jbd2_mark_journal_empty(journal); |
a78bb11d JK |
1926 | mutex_unlock(&journal->j_checkpoint_mutex); |
1927 | } | |
470decc6 DK |
1928 | |
1929 | no_recovery: | |
1930 | return err; | |
1931 | } | |
1932 | ||
470decc6 DK |
1933 | /* |
1934 | * Journal abort has very specific semantics, which we describe | |
1935 | * for journal abort. | |
1936 | * | |
bfcd3555 | 1937 | * Two internal functions, which provide abort to the jbd layer |
470decc6 DK |
1938 | * itself are here. |
1939 | */ | |
1940 | ||
1941 | /* | |
1942 | * Quick version for internal journal use (doesn't lock the journal). | |
1943 | * Aborts hard --- we mark the abort as occurred, but do _nothing_ else, | |
1944 | * and don't attempt to make any other journal updates. | |
1945 | */ | |
f7f4bccb | 1946 | void __jbd2_journal_abort_hard(journal_t *journal) |
470decc6 DK |
1947 | { |
1948 | transaction_t *transaction; | |
470decc6 | 1949 | |
f7f4bccb | 1950 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1951 | return; |
1952 | ||
1953 | printk(KERN_ERR "Aborting journal on device %s.\n", | |
05496769 | 1954 | journal->j_devname); |
470decc6 | 1955 | |
a931da6a | 1956 | write_lock(&journal->j_state_lock); |
f7f4bccb | 1957 | journal->j_flags |= JBD2_ABORT; |
470decc6 DK |
1958 | transaction = journal->j_running_transaction; |
1959 | if (transaction) | |
f7f4bccb | 1960 | __jbd2_log_start_commit(journal, transaction->t_tid); |
a931da6a | 1961 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1962 | } |
1963 | ||
1964 | /* Soft abort: record the abort error status in the journal superblock, | |
1965 | * but don't do any other IO. */ | |
1966 | static void __journal_abort_soft (journal_t *journal, int errno) | |
1967 | { | |
f7f4bccb | 1968 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1969 | return; |
1970 | ||
1971 | if (!journal->j_errno) | |
1972 | journal->j_errno = errno; | |
1973 | ||
f7f4bccb | 1974 | __jbd2_journal_abort_hard(journal); |
470decc6 DK |
1975 | |
1976 | if (errno) | |
24bcc89c | 1977 | jbd2_journal_update_sb_errno(journal); |
470decc6 DK |
1978 | } |
1979 | ||
1980 | /** | |
f7f4bccb | 1981 | * void jbd2_journal_abort () - Shutdown the journal immediately. |
470decc6 DK |
1982 | * @journal: the journal to shutdown. |
1983 | * @errno: an error number to record in the journal indicating | |
1984 | * the reason for the shutdown. | |
1985 | * | |
1986 | * Perform a complete, immediate shutdown of the ENTIRE | |
1987 | * journal (not of a single transaction). This operation cannot be | |
1988 | * undone without closing and reopening the journal. | |
1989 | * | |
f7f4bccb | 1990 | * The jbd2_journal_abort function is intended to support higher level error |
470decc6 DK |
1991 | * recovery mechanisms such as the ext2/ext3 remount-readonly error |
1992 | * mode. | |
1993 | * | |
1994 | * Journal abort has very specific semantics. Any existing dirty, | |
1995 | * unjournaled buffers in the main filesystem will still be written to | |
1996 | * disk by bdflush, but the journaling mechanism will be suspended | |
1997 | * immediately and no further transaction commits will be honoured. | |
1998 | * | |
1999 | * Any dirty, journaled buffers will be written back to disk without | |
2000 | * hitting the journal. Atomicity cannot be guaranteed on an aborted | |
2001 | * filesystem, but we _do_ attempt to leave as much data as possible | |
2002 | * behind for fsck to use for cleanup. | |
2003 | * | |
2004 | * Any attempt to get a new transaction handle on a journal which is in | |
2005 | * ABORT state will just result in an -EROFS error return. A | |
f7f4bccb | 2006 | * jbd2_journal_stop on an existing handle will return -EIO if we have |
470decc6 DK |
2007 | * entered abort state during the update. |
2008 | * | |
2009 | * Recursive transactions are not disturbed by journal abort until the | |
f7f4bccb | 2010 | * final jbd2_journal_stop, which will receive the -EIO error. |
470decc6 | 2011 | * |
f7f4bccb | 2012 | * Finally, the jbd2_journal_abort call allows the caller to supply an errno |
470decc6 DK |
2013 | * which will be recorded (if possible) in the journal superblock. This |
2014 | * allows a client to record failure conditions in the middle of a | |
2015 | * transaction without having to complete the transaction to record the | |
2016 | * failure to disk. ext3_error, for example, now uses this | |
2017 | * functionality. | |
2018 | * | |
2019 | * Errors which originate from within the journaling layer will NOT | |
2020 | * supply an errno; a null errno implies that absolutely no further | |
2021 | * writes are done to the journal (unless there are any already in | |
2022 | * progress). | |
2023 | * | |
2024 | */ | |
2025 | ||
f7f4bccb | 2026 | void jbd2_journal_abort(journal_t *journal, int errno) |
470decc6 DK |
2027 | { |
2028 | __journal_abort_soft(journal, errno); | |
2029 | } | |
2030 | ||
2031 | /** | |
f7f4bccb | 2032 | * int jbd2_journal_errno () - returns the journal's error state. |
470decc6 DK |
2033 | * @journal: journal to examine. |
2034 | * | |
bfcd3555 | 2035 | * This is the errno number set with jbd2_journal_abort(), the last |
470decc6 DK |
2036 | * time the journal was mounted - if the journal was stopped |
2037 | * without calling abort this will be 0. | |
2038 | * | |
2039 | * If the journal has been aborted on this mount time -EROFS will | |
2040 | * be returned. | |
2041 | */ | |
f7f4bccb | 2042 | int jbd2_journal_errno(journal_t *journal) |
470decc6 DK |
2043 | { |
2044 | int err; | |
2045 | ||
a931da6a | 2046 | read_lock(&journal->j_state_lock); |
f7f4bccb | 2047 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2048 | err = -EROFS; |
2049 | else | |
2050 | err = journal->j_errno; | |
a931da6a | 2051 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
2052 | return err; |
2053 | } | |
2054 | ||
2055 | /** | |
f7f4bccb | 2056 | * int jbd2_journal_clear_err () - clears the journal's error state |
470decc6 DK |
2057 | * @journal: journal to act on. |
2058 | * | |
bfcd3555 | 2059 | * An error must be cleared or acked to take a FS out of readonly |
470decc6 DK |
2060 | * mode. |
2061 | */ | |
f7f4bccb | 2062 | int jbd2_journal_clear_err(journal_t *journal) |
470decc6 DK |
2063 | { |
2064 | int err = 0; | |
2065 | ||
a931da6a | 2066 | write_lock(&journal->j_state_lock); |
f7f4bccb | 2067 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2068 | err = -EROFS; |
2069 | else | |
2070 | journal->j_errno = 0; | |
a931da6a | 2071 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2072 | return err; |
2073 | } | |
2074 | ||
2075 | /** | |
f7f4bccb | 2076 | * void jbd2_journal_ack_err() - Ack journal err. |
470decc6 DK |
2077 | * @journal: journal to act on. |
2078 | * | |
bfcd3555 | 2079 | * An error must be cleared or acked to take a FS out of readonly |
470decc6 DK |
2080 | * mode. |
2081 | */ | |
f7f4bccb | 2082 | void jbd2_journal_ack_err(journal_t *journal) |
470decc6 | 2083 | { |
a931da6a | 2084 | write_lock(&journal->j_state_lock); |
470decc6 | 2085 | if (journal->j_errno) |
f7f4bccb | 2086 | journal->j_flags |= JBD2_ACK_ERR; |
a931da6a | 2087 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2088 | } |
2089 | ||
f7f4bccb | 2090 | int jbd2_journal_blocks_per_page(struct inode *inode) |
470decc6 DK |
2091 | { |
2092 | return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
2093 | } | |
2094 | ||
b517bea1 ZB |
2095 | /* |
2096 | * helper functions to deal with 32 or 64bit block numbers. | |
2097 | */ | |
2098 | size_t journal_tag_bytes(journal_t *journal) | |
2099 | { | |
c3900875 DW |
2100 | journal_block_tag_t tag; |
2101 | size_t x = 0; | |
2102 | ||
2103 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
2104 | x += sizeof(tag.t_checksum); | |
2105 | ||
b517bea1 | 2106 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) |
c3900875 | 2107 | return x + JBD2_TAG_SIZE64; |
b517bea1 | 2108 | else |
c3900875 | 2109 | return x + JBD2_TAG_SIZE32; |
b517bea1 ZB |
2110 | } |
2111 | ||
d2eecb03 TT |
2112 | /* |
2113 | * JBD memory management | |
2114 | * | |
2115 | * These functions are used to allocate block-sized chunks of memory | |
2116 | * used for making copies of buffer_head data. Very often it will be | |
2117 | * page-sized chunks of data, but sometimes it will be in | |
2118 | * sub-page-size chunks. (For example, 16k pages on Power systems | |
2119 | * with a 4k block file system.) For blocks smaller than a page, we | |
2120 | * use a SLAB allocator. There are slab caches for each block size, | |
2121 | * which are allocated at mount time, if necessary, and we only free | |
2122 | * (all of) the slab caches when/if the jbd2 module is unloaded. For | |
2123 | * this reason we don't need to a mutex to protect access to | |
2124 | * jbd2_slab[] allocating or releasing memory; only in | |
2125 | * jbd2_journal_create_slab(). | |
2126 | */ | |
2127 | #define JBD2_MAX_SLABS 8 | |
2128 | static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS]; | |
d2eecb03 TT |
2129 | |
2130 | static const char *jbd2_slab_names[JBD2_MAX_SLABS] = { | |
2131 | "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k", | |
2132 | "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k" | |
2133 | }; | |
2134 | ||
2135 | ||
2136 | static void jbd2_journal_destroy_slabs(void) | |
2137 | { | |
2138 | int i; | |
2139 | ||
2140 | for (i = 0; i < JBD2_MAX_SLABS; i++) { | |
2141 | if (jbd2_slab[i]) | |
2142 | kmem_cache_destroy(jbd2_slab[i]); | |
2143 | jbd2_slab[i] = NULL; | |
2144 | } | |
2145 | } | |
2146 | ||
2147 | static int jbd2_journal_create_slab(size_t size) | |
2148 | { | |
51dfacde | 2149 | static DEFINE_MUTEX(jbd2_slab_create_mutex); |
d2eecb03 TT |
2150 | int i = order_base_2(size) - 10; |
2151 | size_t slab_size; | |
2152 | ||
2153 | if (size == PAGE_SIZE) | |
2154 | return 0; | |
2155 | ||
2156 | if (i >= JBD2_MAX_SLABS) | |
2157 | return -EINVAL; | |
2158 | ||
2159 | if (unlikely(i < 0)) | |
2160 | i = 0; | |
51dfacde | 2161 | mutex_lock(&jbd2_slab_create_mutex); |
d2eecb03 | 2162 | if (jbd2_slab[i]) { |
51dfacde | 2163 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2164 | return 0; /* Already created */ |
2165 | } | |
2166 | ||
2167 | slab_size = 1 << (i+10); | |
2168 | jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size, | |
2169 | slab_size, 0, NULL); | |
51dfacde | 2170 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2171 | if (!jbd2_slab[i]) { |
2172 | printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n"); | |
2173 | return -ENOMEM; | |
2174 | } | |
2175 | return 0; | |
2176 | } | |
2177 | ||
2178 | static struct kmem_cache *get_slab(size_t size) | |
2179 | { | |
2180 | int i = order_base_2(size) - 10; | |
2181 | ||
2182 | BUG_ON(i >= JBD2_MAX_SLABS); | |
2183 | if (unlikely(i < 0)) | |
2184 | i = 0; | |
8ac97b74 | 2185 | BUG_ON(jbd2_slab[i] == NULL); |
d2eecb03 TT |
2186 | return jbd2_slab[i]; |
2187 | } | |
2188 | ||
2189 | void *jbd2_alloc(size_t size, gfp_t flags) | |
2190 | { | |
2191 | void *ptr; | |
2192 | ||
2193 | BUG_ON(size & (size-1)); /* Must be a power of 2 */ | |
2194 | ||
2195 | flags |= __GFP_REPEAT; | |
2196 | if (size == PAGE_SIZE) | |
2197 | ptr = (void *)__get_free_pages(flags, 0); | |
2198 | else if (size > PAGE_SIZE) { | |
2199 | int order = get_order(size); | |
2200 | ||
2201 | if (order < 3) | |
2202 | ptr = (void *)__get_free_pages(flags, order); | |
2203 | else | |
2204 | ptr = vmalloc(size); | |
2205 | } else | |
2206 | ptr = kmem_cache_alloc(get_slab(size), flags); | |
2207 | ||
2208 | /* Check alignment; SLUB has gotten this wrong in the past, | |
2209 | * and this can lead to user data corruption! */ | |
2210 | BUG_ON(((unsigned long) ptr) & (size-1)); | |
2211 | ||
2212 | return ptr; | |
2213 | } | |
2214 | ||
2215 | void jbd2_free(void *ptr, size_t size) | |
2216 | { | |
2217 | if (size == PAGE_SIZE) { | |
2218 | free_pages((unsigned long)ptr, 0); | |
2219 | return; | |
2220 | } | |
2221 | if (size > PAGE_SIZE) { | |
2222 | int order = get_order(size); | |
2223 | ||
2224 | if (order < 3) | |
2225 | free_pages((unsigned long)ptr, order); | |
2226 | else | |
2227 | vfree(ptr); | |
2228 | return; | |
2229 | } | |
2230 | kmem_cache_free(get_slab(size), ptr); | |
2231 | }; | |
2232 | ||
470decc6 DK |
2233 | /* |
2234 | * Journal_head storage management | |
2235 | */ | |
e18b890b | 2236 | static struct kmem_cache *jbd2_journal_head_cache; |
e23291b9 | 2237 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2238 | static atomic_t nr_journal_heads = ATOMIC_INIT(0); |
2239 | #endif | |
2240 | ||
4185a2ac | 2241 | static int jbd2_journal_init_journal_head_cache(void) |
470decc6 DK |
2242 | { |
2243 | int retval; | |
2244 | ||
1076d17a | 2245 | J_ASSERT(jbd2_journal_head_cache == NULL); |
a920e941 | 2246 | jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", |
470decc6 DK |
2247 | sizeof(struct journal_head), |
2248 | 0, /* offset */ | |
77160957 | 2249 | SLAB_TEMPORARY, /* flags */ |
20c2df83 | 2250 | NULL); /* ctor */ |
470decc6 | 2251 | retval = 0; |
1076d17a | 2252 | if (!jbd2_journal_head_cache) { |
470decc6 | 2253 | retval = -ENOMEM; |
f2a44523 | 2254 | printk(KERN_EMERG "JBD2: no memory for journal_head cache\n"); |
470decc6 DK |
2255 | } |
2256 | return retval; | |
2257 | } | |
2258 | ||
4185a2ac | 2259 | static void jbd2_journal_destroy_journal_head_cache(void) |
470decc6 | 2260 | { |
8a9362eb DG |
2261 | if (jbd2_journal_head_cache) { |
2262 | kmem_cache_destroy(jbd2_journal_head_cache); | |
2263 | jbd2_journal_head_cache = NULL; | |
2264 | } | |
470decc6 DK |
2265 | } |
2266 | ||
2267 | /* | |
2268 | * journal_head splicing and dicing | |
2269 | */ | |
2270 | static struct journal_head *journal_alloc_journal_head(void) | |
2271 | { | |
2272 | struct journal_head *ret; | |
470decc6 | 2273 | |
e23291b9 | 2274 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2275 | atomic_inc(&nr_journal_heads); |
2276 | #endif | |
f7f4bccb | 2277 | ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS); |
1076d17a | 2278 | if (!ret) { |
470decc6 | 2279 | jbd_debug(1, "out of memory for journal_head\n"); |
670be5a7 | 2280 | pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__); |
1076d17a | 2281 | while (!ret) { |
470decc6 | 2282 | yield(); |
f7f4bccb | 2283 | ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS); |
470decc6 DK |
2284 | } |
2285 | } | |
2286 | return ret; | |
2287 | } | |
2288 | ||
2289 | static void journal_free_journal_head(struct journal_head *jh) | |
2290 | { | |
e23291b9 | 2291 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 | 2292 | atomic_dec(&nr_journal_heads); |
cd02ff0b | 2293 | memset(jh, JBD2_POISON_FREE, sizeof(*jh)); |
470decc6 | 2294 | #endif |
f7f4bccb | 2295 | kmem_cache_free(jbd2_journal_head_cache, jh); |
470decc6 DK |
2296 | } |
2297 | ||
2298 | /* | |
2299 | * A journal_head is attached to a buffer_head whenever JBD has an | |
2300 | * interest in the buffer. | |
2301 | * | |
2302 | * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit | |
2303 | * is set. This bit is tested in core kernel code where we need to take | |
2304 | * JBD-specific actions. Testing the zeroness of ->b_private is not reliable | |
2305 | * there. | |
2306 | * | |
2307 | * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. | |
2308 | * | |
2309 | * When a buffer has its BH_JBD bit set it is immune from being released by | |
2310 | * core kernel code, mainly via ->b_count. | |
2311 | * | |
de1b7941 JK |
2312 | * A journal_head is detached from its buffer_head when the journal_head's |
2313 | * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint | |
2314 | * transaction (b_cp_transaction) hold their references to b_jcount. | |
470decc6 DK |
2315 | * |
2316 | * Various places in the kernel want to attach a journal_head to a buffer_head | |
2317 | * _before_ attaching the journal_head to a transaction. To protect the | |
f7f4bccb | 2318 | * journal_head in this situation, jbd2_journal_add_journal_head elevates the |
470decc6 | 2319 | * journal_head's b_jcount refcount by one. The caller must call |
f7f4bccb | 2320 | * jbd2_journal_put_journal_head() to undo this. |
470decc6 DK |
2321 | * |
2322 | * So the typical usage would be: | |
2323 | * | |
2324 | * (Attach a journal_head if needed. Increments b_jcount) | |
f7f4bccb | 2325 | * struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 | 2326 | * ... |
de1b7941 JK |
2327 | * (Get another reference for transaction) |
2328 | * jbd2_journal_grab_journal_head(bh); | |
470decc6 | 2329 | * jh->b_transaction = xxx; |
de1b7941 | 2330 | * (Put original reference) |
f7f4bccb | 2331 | * jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2332 | */ |
2333 | ||
2334 | /* | |
2335 | * Give a buffer_head a journal_head. | |
2336 | * | |
470decc6 DK |
2337 | * May sleep. |
2338 | */ | |
f7f4bccb | 2339 | struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) |
470decc6 DK |
2340 | { |
2341 | struct journal_head *jh; | |
2342 | struct journal_head *new_jh = NULL; | |
2343 | ||
2344 | repeat: | |
2345 | if (!buffer_jbd(bh)) { | |
2346 | new_jh = journal_alloc_journal_head(); | |
2347 | memset(new_jh, 0, sizeof(*new_jh)); | |
2348 | } | |
2349 | ||
2350 | jbd_lock_bh_journal_head(bh); | |
2351 | if (buffer_jbd(bh)) { | |
2352 | jh = bh2jh(bh); | |
2353 | } else { | |
2354 | J_ASSERT_BH(bh, | |
2355 | (atomic_read(&bh->b_count) > 0) || | |
2356 | (bh->b_page && bh->b_page->mapping)); | |
2357 | ||
2358 | if (!new_jh) { | |
2359 | jbd_unlock_bh_journal_head(bh); | |
2360 | goto repeat; | |
2361 | } | |
2362 | ||
2363 | jh = new_jh; | |
2364 | new_jh = NULL; /* We consumed it */ | |
2365 | set_buffer_jbd(bh); | |
2366 | bh->b_private = jh; | |
2367 | jh->b_bh = bh; | |
2368 | get_bh(bh); | |
2369 | BUFFER_TRACE(bh, "added journal_head"); | |
2370 | } | |
2371 | jh->b_jcount++; | |
2372 | jbd_unlock_bh_journal_head(bh); | |
2373 | if (new_jh) | |
2374 | journal_free_journal_head(new_jh); | |
2375 | return bh->b_private; | |
2376 | } | |
2377 | ||
2378 | /* | |
2379 | * Grab a ref against this buffer_head's journal_head. If it ended up not | |
2380 | * having a journal_head, return NULL | |
2381 | */ | |
f7f4bccb | 2382 | struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) |
470decc6 DK |
2383 | { |
2384 | struct journal_head *jh = NULL; | |
2385 | ||
2386 | jbd_lock_bh_journal_head(bh); | |
2387 | if (buffer_jbd(bh)) { | |
2388 | jh = bh2jh(bh); | |
2389 | jh->b_jcount++; | |
2390 | } | |
2391 | jbd_unlock_bh_journal_head(bh); | |
2392 | return jh; | |
2393 | } | |
2394 | ||
2395 | static void __journal_remove_journal_head(struct buffer_head *bh) | |
2396 | { | |
2397 | struct journal_head *jh = bh2jh(bh); | |
2398 | ||
2399 | J_ASSERT_JH(jh, jh->b_jcount >= 0); | |
de1b7941 JK |
2400 | J_ASSERT_JH(jh, jh->b_transaction == NULL); |
2401 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
2402 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
2403 | J_ASSERT_JH(jh, jh->b_jlist == BJ_None); | |
2404 | J_ASSERT_BH(bh, buffer_jbd(bh)); | |
2405 | J_ASSERT_BH(bh, jh2bh(jh) == bh); | |
2406 | BUFFER_TRACE(bh, "remove journal_head"); | |
2407 | if (jh->b_frozen_data) { | |
2408 | printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__); | |
2409 | jbd2_free(jh->b_frozen_data, bh->b_size); | |
470decc6 | 2410 | } |
de1b7941 JK |
2411 | if (jh->b_committed_data) { |
2412 | printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__); | |
2413 | jbd2_free(jh->b_committed_data, bh->b_size); | |
2414 | } | |
2415 | bh->b_private = NULL; | |
2416 | jh->b_bh = NULL; /* debug, really */ | |
2417 | clear_buffer_jbd(bh); | |
2418 | journal_free_journal_head(jh); | |
470decc6 DK |
2419 | } |
2420 | ||
2421 | /* | |
de1b7941 | 2422 | * Drop a reference on the passed journal_head. If it fell to zero then |
470decc6 DK |
2423 | * release the journal_head from the buffer_head. |
2424 | */ | |
f7f4bccb | 2425 | void jbd2_journal_put_journal_head(struct journal_head *jh) |
470decc6 DK |
2426 | { |
2427 | struct buffer_head *bh = jh2bh(jh); | |
2428 | ||
2429 | jbd_lock_bh_journal_head(bh); | |
2430 | J_ASSERT_JH(jh, jh->b_jcount > 0); | |
2431 | --jh->b_jcount; | |
de1b7941 | 2432 | if (!jh->b_jcount) { |
470decc6 | 2433 | __journal_remove_journal_head(bh); |
de1b7941 | 2434 | jbd_unlock_bh_journal_head(bh); |
470decc6 | 2435 | __brelse(bh); |
de1b7941 JK |
2436 | } else |
2437 | jbd_unlock_bh_journal_head(bh); | |
470decc6 DK |
2438 | } |
2439 | ||
c851ed54 JK |
2440 | /* |
2441 | * Initialize jbd inode head | |
2442 | */ | |
2443 | void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) | |
2444 | { | |
2445 | jinode->i_transaction = NULL; | |
2446 | jinode->i_next_transaction = NULL; | |
2447 | jinode->i_vfs_inode = inode; | |
2448 | jinode->i_flags = 0; | |
2449 | INIT_LIST_HEAD(&jinode->i_list); | |
2450 | } | |
2451 | ||
2452 | /* | |
2453 | * Function to be called before we start removing inode from memory (i.e., | |
2454 | * clear_inode() is a fine place to be called from). It removes inode from | |
2455 | * transaction's lists. | |
2456 | */ | |
2457 | void jbd2_journal_release_jbd_inode(journal_t *journal, | |
2458 | struct jbd2_inode *jinode) | |
2459 | { | |
c851ed54 JK |
2460 | if (!journal) |
2461 | return; | |
2462 | restart: | |
2463 | spin_lock(&journal->j_list_lock); | |
2464 | /* Is commit writing out inode - we have to wait */ | |
39e3ac25 | 2465 | if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) { |
c851ed54 JK |
2466 | wait_queue_head_t *wq; |
2467 | DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); | |
2468 | wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); | |
2469 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
2470 | spin_unlock(&journal->j_list_lock); | |
2471 | schedule(); | |
2472 | finish_wait(wq, &wait.wait); | |
2473 | goto restart; | |
2474 | } | |
2475 | ||
c851ed54 JK |
2476 | if (jinode->i_transaction) { |
2477 | list_del(&jinode->i_list); | |
2478 | jinode->i_transaction = NULL; | |
2479 | } | |
2480 | spin_unlock(&journal->j_list_lock); | |
2481 | } | |
2482 | ||
470decc6 | 2483 | /* |
0f49d5d0 | 2484 | * debugfs tunables |
470decc6 | 2485 | */ |
6f38c74f JS |
2486 | #ifdef CONFIG_JBD2_DEBUG |
2487 | u8 jbd2_journal_enable_debug __read_mostly; | |
f7f4bccb | 2488 | EXPORT_SYMBOL(jbd2_journal_enable_debug); |
470decc6 | 2489 | |
0f49d5d0 | 2490 | #define JBD2_DEBUG_NAME "jbd2-debug" |
470decc6 | 2491 | |
6f38c74f JS |
2492 | static struct dentry *jbd2_debugfs_dir; |
2493 | static struct dentry *jbd2_debug; | |
470decc6 | 2494 | |
0f49d5d0 JS |
2495 | static void __init jbd2_create_debugfs_entry(void) |
2496 | { | |
2497 | jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL); | |
2498 | if (jbd2_debugfs_dir) | |
765f8361 YK |
2499 | jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, |
2500 | S_IRUGO | S_IWUSR, | |
0f49d5d0 JS |
2501 | jbd2_debugfs_dir, |
2502 | &jbd2_journal_enable_debug); | |
470decc6 DK |
2503 | } |
2504 | ||
0f49d5d0 | 2505 | static void __exit jbd2_remove_debugfs_entry(void) |
470decc6 | 2506 | { |
6f38c74f JS |
2507 | debugfs_remove(jbd2_debug); |
2508 | debugfs_remove(jbd2_debugfs_dir); | |
470decc6 DK |
2509 | } |
2510 | ||
0f49d5d0 | 2511 | #else |
470decc6 | 2512 | |
0f49d5d0 | 2513 | static void __init jbd2_create_debugfs_entry(void) |
470decc6 | 2514 | { |
470decc6 DK |
2515 | } |
2516 | ||
0f49d5d0 | 2517 | static void __exit jbd2_remove_debugfs_entry(void) |
470decc6 | 2518 | { |
470decc6 DK |
2519 | } |
2520 | ||
470decc6 DK |
2521 | #endif |
2522 | ||
8e85fb3f JL |
2523 | #ifdef CONFIG_PROC_FS |
2524 | ||
2525 | #define JBD2_STATS_PROC_NAME "fs/jbd2" | |
2526 | ||
2527 | static void __init jbd2_create_jbd_stats_proc_entry(void) | |
2528 | { | |
2529 | proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); | |
2530 | } | |
2531 | ||
2532 | static void __exit jbd2_remove_jbd_stats_proc_entry(void) | |
2533 | { | |
2534 | if (proc_jbd2_stats) | |
2535 | remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); | |
2536 | } | |
2537 | ||
2538 | #else | |
2539 | ||
2540 | #define jbd2_create_jbd_stats_proc_entry() do {} while (0) | |
2541 | #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) | |
2542 | ||
2543 | #endif | |
2544 | ||
8aefcd55 | 2545 | struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache; |
470decc6 | 2546 | |
4185a2ac | 2547 | static int __init jbd2_journal_init_handle_cache(void) |
470decc6 | 2548 | { |
8aefcd55 | 2549 | jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY); |
f7f4bccb | 2550 | if (jbd2_handle_cache == NULL) { |
8aefcd55 TT |
2551 | printk(KERN_EMERG "JBD2: failed to create handle cache\n"); |
2552 | return -ENOMEM; | |
2553 | } | |
2554 | jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0); | |
2555 | if (jbd2_inode_cache == NULL) { | |
2556 | printk(KERN_EMERG "JBD2: failed to create inode cache\n"); | |
2557 | kmem_cache_destroy(jbd2_handle_cache); | |
470decc6 DK |
2558 | return -ENOMEM; |
2559 | } | |
2560 | return 0; | |
2561 | } | |
2562 | ||
f7f4bccb | 2563 | static void jbd2_journal_destroy_handle_cache(void) |
470decc6 | 2564 | { |
f7f4bccb MC |
2565 | if (jbd2_handle_cache) |
2566 | kmem_cache_destroy(jbd2_handle_cache); | |
8aefcd55 TT |
2567 | if (jbd2_inode_cache) |
2568 | kmem_cache_destroy(jbd2_inode_cache); | |
2569 | ||
470decc6 DK |
2570 | } |
2571 | ||
2572 | /* | |
2573 | * Module startup and shutdown | |
2574 | */ | |
2575 | ||
2576 | static int __init journal_init_caches(void) | |
2577 | { | |
2578 | int ret; | |
2579 | ||
f7f4bccb | 2580 | ret = jbd2_journal_init_revoke_caches(); |
470decc6 | 2581 | if (ret == 0) |
4185a2ac | 2582 | ret = jbd2_journal_init_journal_head_cache(); |
470decc6 | 2583 | if (ret == 0) |
4185a2ac | 2584 | ret = jbd2_journal_init_handle_cache(); |
470decc6 | 2585 | if (ret == 0) |
0c2022ec | 2586 | ret = jbd2_journal_init_transaction_cache(); |
470decc6 DK |
2587 | return ret; |
2588 | } | |
2589 | ||
f7f4bccb | 2590 | static void jbd2_journal_destroy_caches(void) |
470decc6 | 2591 | { |
f7f4bccb | 2592 | jbd2_journal_destroy_revoke_caches(); |
4185a2ac | 2593 | jbd2_journal_destroy_journal_head_cache(); |
f7f4bccb | 2594 | jbd2_journal_destroy_handle_cache(); |
0c2022ec | 2595 | jbd2_journal_destroy_transaction_cache(); |
d2eecb03 | 2596 | jbd2_journal_destroy_slabs(); |
470decc6 DK |
2597 | } |
2598 | ||
2599 | static int __init journal_init(void) | |
2600 | { | |
2601 | int ret; | |
2602 | ||
2603 | BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); | |
2604 | ||
2605 | ret = journal_init_caches(); | |
620de4e1 DG |
2606 | if (ret == 0) { |
2607 | jbd2_create_debugfs_entry(); | |
2608 | jbd2_create_jbd_stats_proc_entry(); | |
2609 | } else { | |
f7f4bccb | 2610 | jbd2_journal_destroy_caches(); |
620de4e1 | 2611 | } |
470decc6 DK |
2612 | return ret; |
2613 | } | |
2614 | ||
2615 | static void __exit journal_exit(void) | |
2616 | { | |
e23291b9 | 2617 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2618 | int n = atomic_read(&nr_journal_heads); |
2619 | if (n) | |
f2a44523 | 2620 | printk(KERN_EMERG "JBD2: leaked %d journal_heads!\n", n); |
470decc6 | 2621 | #endif |
0f49d5d0 | 2622 | jbd2_remove_debugfs_entry(); |
8e85fb3f | 2623 | jbd2_remove_jbd_stats_proc_entry(); |
f7f4bccb | 2624 | jbd2_journal_destroy_caches(); |
470decc6 DK |
2625 | } |
2626 | ||
2627 | MODULE_LICENSE("GPL"); | |
2628 | module_init(journal_init); | |
2629 | module_exit(journal_exit); | |
2630 |