Commit | Line | Data |
---|---|---|
470decc6 | 1 | /* |
f7f4bccb | 2 | * linux/fs/jbd2/journal.c |
470decc6 DK |
3 | * |
4 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1998 | |
5 | * | |
6 | * Copyright 1998 Red Hat corp --- All Rights Reserved | |
7 | * | |
8 | * This file is part of the Linux kernel and is made available under | |
9 | * the terms of the GNU General Public License, version 2, or at your | |
10 | * option, any later version, incorporated herein by reference. | |
11 | * | |
12 | * Generic filesystem journal-writing code; part of the ext2fs | |
13 | * journaling system. | |
14 | * | |
15 | * This file manages journals: areas of disk reserved for logging | |
16 | * transactional updates. This includes the kernel journaling thread | |
17 | * which is responsible for scheduling updates to the log. | |
18 | * | |
19 | * We do not actually manage the physical storage of the journal in this | |
20 | * file: that is left to a per-journal policy function, which allows us | |
21 | * to store the journal within a filesystem-specified area for ext2 | |
22 | * journaling (ext2 can use a reserved inode for storing the log). | |
23 | */ | |
24 | ||
25 | #include <linux/module.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/fs.h> | |
f7f4bccb | 28 | #include <linux/jbd2.h> |
470decc6 DK |
29 | #include <linux/errno.h> |
30 | #include <linux/slab.h> | |
470decc6 DK |
31 | #include <linux/init.h> |
32 | #include <linux/mm.h> | |
7dfb7103 | 33 | #include <linux/freezer.h> |
470decc6 DK |
34 | #include <linux/pagemap.h> |
35 | #include <linux/kthread.h> | |
36 | #include <linux/poison.h> | |
37 | #include <linux/proc_fs.h> | |
8e85fb3f | 38 | #include <linux/seq_file.h> |
c225aa57 | 39 | #include <linux/math64.h> |
879c5e6b | 40 | #include <linux/hash.h> |
d2eecb03 TT |
41 | #include <linux/log2.h> |
42 | #include <linux/vmalloc.h> | |
47def826 | 43 | #include <linux/backing-dev.h> |
39e3ac25 | 44 | #include <linux/bitops.h> |
670be5a7 | 45 | #include <linux/ratelimit.h> |
879c5e6b TT |
46 | |
47 | #define CREATE_TRACE_POINTS | |
48 | #include <trace/events/jbd2.h> | |
470decc6 DK |
49 | |
50 | #include <asm/uaccess.h> | |
51 | #include <asm/page.h> | |
52 | ||
b6e96d00 TT |
53 | #ifdef CONFIG_JBD2_DEBUG |
54 | ushort jbd2_journal_enable_debug __read_mostly; | |
55 | EXPORT_SYMBOL(jbd2_journal_enable_debug); | |
56 | ||
57 | module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644); | |
58 | MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2"); | |
59 | #endif | |
60 | ||
f7f4bccb MC |
61 | EXPORT_SYMBOL(jbd2_journal_extend); |
62 | EXPORT_SYMBOL(jbd2_journal_stop); | |
63 | EXPORT_SYMBOL(jbd2_journal_lock_updates); | |
64 | EXPORT_SYMBOL(jbd2_journal_unlock_updates); | |
65 | EXPORT_SYMBOL(jbd2_journal_get_write_access); | |
66 | EXPORT_SYMBOL(jbd2_journal_get_create_access); | |
67 | EXPORT_SYMBOL(jbd2_journal_get_undo_access); | |
e06c8227 | 68 | EXPORT_SYMBOL(jbd2_journal_set_triggers); |
f7f4bccb | 69 | EXPORT_SYMBOL(jbd2_journal_dirty_metadata); |
f7f4bccb | 70 | EXPORT_SYMBOL(jbd2_journal_forget); |
470decc6 DK |
71 | #if 0 |
72 | EXPORT_SYMBOL(journal_sync_buffer); | |
73 | #endif | |
f7f4bccb MC |
74 | EXPORT_SYMBOL(jbd2_journal_flush); |
75 | EXPORT_SYMBOL(jbd2_journal_revoke); | |
76 | ||
77 | EXPORT_SYMBOL(jbd2_journal_init_dev); | |
78 | EXPORT_SYMBOL(jbd2_journal_init_inode); | |
f7f4bccb MC |
79 | EXPORT_SYMBOL(jbd2_journal_check_used_features); |
80 | EXPORT_SYMBOL(jbd2_journal_check_available_features); | |
81 | EXPORT_SYMBOL(jbd2_journal_set_features); | |
f7f4bccb MC |
82 | EXPORT_SYMBOL(jbd2_journal_load); |
83 | EXPORT_SYMBOL(jbd2_journal_destroy); | |
f7f4bccb MC |
84 | EXPORT_SYMBOL(jbd2_journal_abort); |
85 | EXPORT_SYMBOL(jbd2_journal_errno); | |
86 | EXPORT_SYMBOL(jbd2_journal_ack_err); | |
87 | EXPORT_SYMBOL(jbd2_journal_clear_err); | |
88 | EXPORT_SYMBOL(jbd2_log_wait_commit); | |
3b799d15 | 89 | EXPORT_SYMBOL(jbd2_log_start_commit); |
f7f4bccb MC |
90 | EXPORT_SYMBOL(jbd2_journal_start_commit); |
91 | EXPORT_SYMBOL(jbd2_journal_force_commit_nested); | |
92 | EXPORT_SYMBOL(jbd2_journal_wipe); | |
93 | EXPORT_SYMBOL(jbd2_journal_blocks_per_page); | |
94 | EXPORT_SYMBOL(jbd2_journal_invalidatepage); | |
95 | EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); | |
96 | EXPORT_SYMBOL(jbd2_journal_force_commit); | |
41617e1a JK |
97 | EXPORT_SYMBOL(jbd2_journal_inode_add_write); |
98 | EXPORT_SYMBOL(jbd2_journal_inode_add_wait); | |
c851ed54 JK |
99 | EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); |
100 | EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); | |
101 | EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); | |
8aefcd55 | 102 | EXPORT_SYMBOL(jbd2_inode_cache); |
470decc6 | 103 | |
470decc6 | 104 | static void __journal_abort_soft (journal_t *journal, int errno); |
d2eecb03 | 105 | static int jbd2_journal_create_slab(size_t slab_size); |
470decc6 | 106 | |
169f1a2a PG |
107 | #ifdef CONFIG_JBD2_DEBUG |
108 | void __jbd2_debug(int level, const char *file, const char *func, | |
109 | unsigned int line, const char *fmt, ...) | |
110 | { | |
111 | struct va_format vaf; | |
112 | va_list args; | |
113 | ||
114 | if (level > jbd2_journal_enable_debug) | |
115 | return; | |
116 | va_start(args, fmt); | |
117 | vaf.fmt = fmt; | |
118 | vaf.va = &args; | |
119 | printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf); | |
120 | va_end(args); | |
121 | } | |
122 | EXPORT_SYMBOL(__jbd2_debug); | |
123 | #endif | |
124 | ||
25ed6e8a | 125 | /* Checksumming functions */ |
7747e6d0 | 126 | static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb) |
25ed6e8a | 127 | { |
8595798c | 128 | if (!jbd2_journal_has_csum_v2or3_feature(j)) |
25ed6e8a DW |
129 | return 1; |
130 | ||
131 | return sb->s_checksum_type == JBD2_CRC32C_CHKSUM; | |
132 | } | |
133 | ||
18a6ea1e | 134 | static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb) |
4fd5ea43 | 135 | { |
18a6ea1e DW |
136 | __u32 csum; |
137 | __be32 old_csum; | |
4fd5ea43 DW |
138 | |
139 | old_csum = sb->s_checksum; | |
140 | sb->s_checksum = 0; | |
141 | csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t)); | |
142 | sb->s_checksum = old_csum; | |
143 | ||
144 | return cpu_to_be32(csum); | |
145 | } | |
146 | ||
7747e6d0 | 147 | static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb) |
4fd5ea43 | 148 | { |
db9ee220 | 149 | if (!jbd2_journal_has_csum_v2or3(j)) |
4fd5ea43 DW |
150 | return 1; |
151 | ||
152 | return sb->s_checksum == jbd2_superblock_csum(j, sb); | |
153 | } | |
154 | ||
7747e6d0 | 155 | static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb) |
4fd5ea43 | 156 | { |
db9ee220 | 157 | if (!jbd2_journal_has_csum_v2or3(j)) |
4fd5ea43 DW |
158 | return; |
159 | ||
160 | sb->s_checksum = jbd2_superblock_csum(j, sb); | |
161 | } | |
162 | ||
470decc6 DK |
163 | /* |
164 | * Helper function used to manage commit timeouts | |
165 | */ | |
166 | ||
167 | static void commit_timeout(unsigned long __data) | |
168 | { | |
169 | struct task_struct * p = (struct task_struct *) __data; | |
170 | ||
171 | wake_up_process(p); | |
172 | } | |
173 | ||
174 | /* | |
f7f4bccb | 175 | * kjournald2: The main thread function used to manage a logging device |
470decc6 DK |
176 | * journal. |
177 | * | |
178 | * This kernel thread is responsible for two things: | |
179 | * | |
180 | * 1) COMMIT: Every so often we need to commit the current state of the | |
181 | * filesystem to disk. The journal thread is responsible for writing | |
182 | * all of the metadata buffers to disk. | |
183 | * | |
184 | * 2) CHECKPOINT: We cannot reuse a used section of the log file until all | |
185 | * of the data in that part of the log has been rewritten elsewhere on | |
186 | * the disk. Flushing these old buffers to reclaim space in the log is | |
187 | * known as checkpointing, and this thread is responsible for that job. | |
188 | */ | |
189 | ||
f7f4bccb | 190 | static int kjournald2(void *arg) |
470decc6 DK |
191 | { |
192 | journal_t *journal = arg; | |
193 | transaction_t *transaction; | |
194 | ||
195 | /* | |
196 | * Set up an interval timer which can be used to trigger a commit wakeup | |
197 | * after the commit interval expires | |
198 | */ | |
199 | setup_timer(&journal->j_commit_timer, commit_timeout, | |
200 | (unsigned long)current); | |
201 | ||
35c80422 NC |
202 | set_freezable(); |
203 | ||
470decc6 DK |
204 | /* Record that the journal thread is running */ |
205 | journal->j_task = current; | |
206 | wake_up(&journal->j_wait_done_commit); | |
207 | ||
470decc6 DK |
208 | /* |
209 | * And now, wait forever for commit wakeup events. | |
210 | */ | |
a931da6a | 211 | write_lock(&journal->j_state_lock); |
470decc6 DK |
212 | |
213 | loop: | |
f7f4bccb | 214 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
215 | goto end_loop; |
216 | ||
217 | jbd_debug(1, "commit_sequence=%d, commit_request=%d\n", | |
218 | journal->j_commit_sequence, journal->j_commit_request); | |
219 | ||
220 | if (journal->j_commit_sequence != journal->j_commit_request) { | |
221 | jbd_debug(1, "OK, requests differ\n"); | |
a931da6a | 222 | write_unlock(&journal->j_state_lock); |
470decc6 | 223 | del_timer_sync(&journal->j_commit_timer); |
f7f4bccb | 224 | jbd2_journal_commit_transaction(journal); |
a931da6a | 225 | write_lock(&journal->j_state_lock); |
470decc6 DK |
226 | goto loop; |
227 | } | |
228 | ||
229 | wake_up(&journal->j_wait_done_commit); | |
230 | if (freezing(current)) { | |
231 | /* | |
232 | * The simpler the better. Flushing journal isn't a | |
233 | * good idea, because that depends on threads that may | |
234 | * be already stopped. | |
235 | */ | |
f7f4bccb | 236 | jbd_debug(1, "Now suspending kjournald2\n"); |
a931da6a | 237 | write_unlock(&journal->j_state_lock); |
a0acae0e | 238 | try_to_freeze(); |
a931da6a | 239 | write_lock(&journal->j_state_lock); |
470decc6 DK |
240 | } else { |
241 | /* | |
242 | * We assume on resume that commits are already there, | |
243 | * so we don't sleep | |
244 | */ | |
245 | DEFINE_WAIT(wait); | |
246 | int should_sleep = 1; | |
247 | ||
248 | prepare_to_wait(&journal->j_wait_commit, &wait, | |
249 | TASK_INTERRUPTIBLE); | |
250 | if (journal->j_commit_sequence != journal->j_commit_request) | |
251 | should_sleep = 0; | |
252 | transaction = journal->j_running_transaction; | |
253 | if (transaction && time_after_eq(jiffies, | |
254 | transaction->t_expires)) | |
255 | should_sleep = 0; | |
f7f4bccb | 256 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
257 | should_sleep = 0; |
258 | if (should_sleep) { | |
a931da6a | 259 | write_unlock(&journal->j_state_lock); |
470decc6 | 260 | schedule(); |
a931da6a | 261 | write_lock(&journal->j_state_lock); |
470decc6 DK |
262 | } |
263 | finish_wait(&journal->j_wait_commit, &wait); | |
264 | } | |
265 | ||
f7f4bccb | 266 | jbd_debug(1, "kjournald2 wakes\n"); |
470decc6 DK |
267 | |
268 | /* | |
269 | * Were we woken up by a commit wakeup event? | |
270 | */ | |
271 | transaction = journal->j_running_transaction; | |
272 | if (transaction && time_after_eq(jiffies, transaction->t_expires)) { | |
273 | journal->j_commit_request = transaction->t_tid; | |
274 | jbd_debug(1, "woke because of timeout\n"); | |
275 | } | |
276 | goto loop; | |
277 | ||
278 | end_loop: | |
a931da6a | 279 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
280 | del_timer_sync(&journal->j_commit_timer); |
281 | journal->j_task = NULL; | |
282 | wake_up(&journal->j_wait_done_commit); | |
283 | jbd_debug(1, "Journal thread exiting.\n"); | |
284 | return 0; | |
285 | } | |
286 | ||
97f06784 | 287 | static int jbd2_journal_start_thread(journal_t *journal) |
470decc6 | 288 | { |
97f06784 PE |
289 | struct task_struct *t; |
290 | ||
90576c0b TT |
291 | t = kthread_run(kjournald2, journal, "jbd2/%s", |
292 | journal->j_devname); | |
97f06784 PE |
293 | if (IS_ERR(t)) |
294 | return PTR_ERR(t); | |
295 | ||
1076d17a | 296 | wait_event(journal->j_wait_done_commit, journal->j_task != NULL); |
97f06784 | 297 | return 0; |
470decc6 DK |
298 | } |
299 | ||
300 | static void journal_kill_thread(journal_t *journal) | |
301 | { | |
a931da6a | 302 | write_lock(&journal->j_state_lock); |
f7f4bccb | 303 | journal->j_flags |= JBD2_UNMOUNT; |
470decc6 DK |
304 | |
305 | while (journal->j_task) { | |
a931da6a | 306 | write_unlock(&journal->j_state_lock); |
3469a32a | 307 | wake_up(&journal->j_wait_commit); |
1076d17a | 308 | wait_event(journal->j_wait_done_commit, journal->j_task == NULL); |
a931da6a | 309 | write_lock(&journal->j_state_lock); |
470decc6 | 310 | } |
a931da6a | 311 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
312 | } |
313 | ||
314 | /* | |
f7f4bccb | 315 | * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. |
470decc6 DK |
316 | * |
317 | * Writes a metadata buffer to a given disk block. The actual IO is not | |
318 | * performed but a new buffer_head is constructed which labels the data | |
319 | * to be written with the correct destination disk block. | |
320 | * | |
321 | * Any magic-number escaping which needs to be done will cause a | |
322 | * copy-out here. If the buffer happens to start with the | |
f7f4bccb | 323 | * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the |
470decc6 DK |
324 | * magic number is only written to the log for descripter blocks. In |
325 | * this case, we copy the data and replace the first word with 0, and we | |
326 | * return a result code which indicates that this buffer needs to be | |
327 | * marked as an escaped buffer in the corresponding log descriptor | |
328 | * block. The missing word can then be restored when the block is read | |
329 | * during recovery. | |
330 | * | |
331 | * If the source buffer has already been modified by a new transaction | |
332 | * since we took the last commit snapshot, we use the frozen copy of | |
f5113eff JK |
333 | * that data for IO. If we end up using the existing buffer_head's data |
334 | * for the write, then we have to make sure nobody modifies it while the | |
335 | * IO is in progress. do_get_write_access() handles this. | |
470decc6 | 336 | * |
f5113eff JK |
337 | * The function returns a pointer to the buffer_head to be used for IO. |
338 | * | |
470decc6 DK |
339 | * |
340 | * Return value: | |
341 | * <0: Error | |
342 | * >=0: Finished OK | |
343 | * | |
344 | * On success: | |
345 | * Bit 0 set == escape performed on the data | |
346 | * Bit 1 set == buffer copy-out performed (kfree the data after IO) | |
347 | */ | |
348 | ||
f7f4bccb | 349 | int jbd2_journal_write_metadata_buffer(transaction_t *transaction, |
470decc6 | 350 | struct journal_head *jh_in, |
f5113eff JK |
351 | struct buffer_head **bh_out, |
352 | sector_t blocknr) | |
470decc6 DK |
353 | { |
354 | int need_copy_out = 0; | |
355 | int done_copy_out = 0; | |
356 | int do_escape = 0; | |
357 | char *mapped_data; | |
358 | struct buffer_head *new_bh; | |
470decc6 DK |
359 | struct page *new_page; |
360 | unsigned int new_offset; | |
361 | struct buffer_head *bh_in = jh2bh(jh_in); | |
96577c43 | 362 | journal_t *journal = transaction->t_journal; |
470decc6 DK |
363 | |
364 | /* | |
365 | * The buffer really shouldn't be locked: only the current committing | |
366 | * transaction is allowed to write it, so nobody else is allowed | |
367 | * to do any IO. | |
368 | * | |
369 | * akpm: except if we're journalling data, and write() output is | |
370 | * also part of a shared mapping, and another thread has | |
371 | * decided to launch a writepage() against this buffer. | |
372 | */ | |
373 | J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); | |
374 | ||
6ccaf3e2 | 375 | new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); |
47def826 | 376 | |
96577c43 | 377 | /* keep subsequent assertions sane */ |
96577c43 | 378 | atomic_set(&new_bh->b_count, 1); |
470decc6 | 379 | |
f5113eff JK |
380 | jbd_lock_bh_state(bh_in); |
381 | repeat: | |
470decc6 DK |
382 | /* |
383 | * If a new transaction has already done a buffer copy-out, then | |
384 | * we use that version of the data for the commit. | |
385 | */ | |
470decc6 DK |
386 | if (jh_in->b_frozen_data) { |
387 | done_copy_out = 1; | |
388 | new_page = virt_to_page(jh_in->b_frozen_data); | |
389 | new_offset = offset_in_page(jh_in->b_frozen_data); | |
390 | } else { | |
391 | new_page = jh2bh(jh_in)->b_page; | |
392 | new_offset = offset_in_page(jh2bh(jh_in)->b_data); | |
393 | } | |
394 | ||
303a8f2a | 395 | mapped_data = kmap_atomic(new_page); |
e06c8227 | 396 | /* |
13ceef09 JK |
397 | * Fire data frozen trigger if data already wasn't frozen. Do this |
398 | * before checking for escaping, as the trigger may modify the magic | |
399 | * offset. If a copy-out happens afterwards, it will have the correct | |
400 | * data in the buffer. | |
e06c8227 | 401 | */ |
13ceef09 JK |
402 | if (!done_copy_out) |
403 | jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset, | |
404 | jh_in->b_triggers); | |
e06c8227 | 405 | |
470decc6 DK |
406 | /* |
407 | * Check for escaping | |
408 | */ | |
409 | if (*((__be32 *)(mapped_data + new_offset)) == | |
f7f4bccb | 410 | cpu_to_be32(JBD2_MAGIC_NUMBER)) { |
470decc6 DK |
411 | need_copy_out = 1; |
412 | do_escape = 1; | |
413 | } | |
303a8f2a | 414 | kunmap_atomic(mapped_data); |
470decc6 DK |
415 | |
416 | /* | |
417 | * Do we need to do a data copy? | |
418 | */ | |
419 | if (need_copy_out && !done_copy_out) { | |
420 | char *tmp; | |
421 | ||
422 | jbd_unlock_bh_state(bh_in); | |
af1e76d6 | 423 | tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); |
e6ec116b | 424 | if (!tmp) { |
f5113eff | 425 | brelse(new_bh); |
e6ec116b TT |
426 | return -ENOMEM; |
427 | } | |
470decc6 DK |
428 | jbd_lock_bh_state(bh_in); |
429 | if (jh_in->b_frozen_data) { | |
af1e76d6 | 430 | jbd2_free(tmp, bh_in->b_size); |
470decc6 DK |
431 | goto repeat; |
432 | } | |
433 | ||
434 | jh_in->b_frozen_data = tmp; | |
303a8f2a | 435 | mapped_data = kmap_atomic(new_page); |
f5113eff | 436 | memcpy(tmp, mapped_data + new_offset, bh_in->b_size); |
303a8f2a | 437 | kunmap_atomic(mapped_data); |
470decc6 DK |
438 | |
439 | new_page = virt_to_page(tmp); | |
440 | new_offset = offset_in_page(tmp); | |
441 | done_copy_out = 1; | |
e06c8227 JB |
442 | |
443 | /* | |
444 | * This isn't strictly necessary, as we're using frozen | |
445 | * data for the escaping, but it keeps consistency with | |
446 | * b_frozen_data usage. | |
447 | */ | |
448 | jh_in->b_frozen_triggers = jh_in->b_triggers; | |
470decc6 DK |
449 | } |
450 | ||
451 | /* | |
452 | * Did we need to do an escaping? Now we've done all the | |
453 | * copying, we can finally do so. | |
454 | */ | |
455 | if (do_escape) { | |
303a8f2a | 456 | mapped_data = kmap_atomic(new_page); |
470decc6 | 457 | *((unsigned int *)(mapped_data + new_offset)) = 0; |
303a8f2a | 458 | kunmap_atomic(mapped_data); |
470decc6 DK |
459 | } |
460 | ||
470decc6 | 461 | set_bh_page(new_bh, new_page, new_offset); |
f5113eff JK |
462 | new_bh->b_size = bh_in->b_size; |
463 | new_bh->b_bdev = journal->j_dev; | |
470decc6 | 464 | new_bh->b_blocknr = blocknr; |
b34090e5 | 465 | new_bh->b_private = bh_in; |
470decc6 DK |
466 | set_buffer_mapped(new_bh); |
467 | set_buffer_dirty(new_bh); | |
468 | ||
f5113eff | 469 | *bh_out = new_bh; |
470decc6 DK |
470 | |
471 | /* | |
472 | * The to-be-written buffer needs to get moved to the io queue, | |
473 | * and the original buffer whose contents we are shadowing or | |
474 | * copying is moved to the transaction's shadow queue. | |
475 | */ | |
476 | JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); | |
96577c43 | 477 | spin_lock(&journal->j_list_lock); |
478 | __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); | |
479 | spin_unlock(&journal->j_list_lock); | |
b34090e5 | 480 | set_buffer_shadow(bh_in); |
96577c43 | 481 | jbd_unlock_bh_state(bh_in); |
482 | ||
470decc6 DK |
483 | return do_escape | (done_copy_out << 1); |
484 | } | |
485 | ||
486 | /* | |
487 | * Allocation code for the journal file. Manage the space left in the | |
488 | * journal, so that we can begin checkpointing when appropriate. | |
489 | */ | |
490 | ||
470decc6 | 491 | /* |
e4471831 TT |
492 | * Called with j_state_lock locked for writing. |
493 | * Returns true if a transaction commit was started. | |
470decc6 | 494 | */ |
f7f4bccb | 495 | int __jbd2_log_start_commit(journal_t *journal, tid_t target) |
470decc6 | 496 | { |
e7b04ac0 ES |
497 | /* Return if the txn has already requested to be committed */ |
498 | if (journal->j_commit_request == target) | |
499 | return 0; | |
500 | ||
470decc6 | 501 | /* |
deeeaf13 TT |
502 | * The only transaction we can possibly wait upon is the |
503 | * currently running transaction (if it exists). Otherwise, | |
504 | * the target tid must be an old one. | |
470decc6 | 505 | */ |
deeeaf13 TT |
506 | if (journal->j_running_transaction && |
507 | journal->j_running_transaction->t_tid == target) { | |
470decc6 | 508 | /* |
bcf3d0bc | 509 | * We want a new commit: OK, mark the request and wakeup the |
470decc6 DK |
510 | * commit thread. We do _not_ do the commit ourselves. |
511 | */ | |
512 | ||
513 | journal->j_commit_request = target; | |
f2a44523 | 514 | jbd_debug(1, "JBD2: requesting commit %d/%d\n", |
470decc6 DK |
515 | journal->j_commit_request, |
516 | journal->j_commit_sequence); | |
9fff24aa | 517 | journal->j_running_transaction->t_requested = jiffies; |
470decc6 DK |
518 | wake_up(&journal->j_wait_commit); |
519 | return 1; | |
deeeaf13 TT |
520 | } else if (!tid_geq(journal->j_commit_request, target)) |
521 | /* This should never happen, but if it does, preserve | |
522 | the evidence before kjournald goes into a loop and | |
523 | increments j_commit_sequence beyond all recognition. */ | |
f2a44523 | 524 | WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", |
1be2add6 TT |
525 | journal->j_commit_request, |
526 | journal->j_commit_sequence, | |
527 | target, journal->j_running_transaction ? | |
528 | journal->j_running_transaction->t_tid : 0); | |
470decc6 DK |
529 | return 0; |
530 | } | |
531 | ||
f7f4bccb | 532 | int jbd2_log_start_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
533 | { |
534 | int ret; | |
535 | ||
a931da6a | 536 | write_lock(&journal->j_state_lock); |
f7f4bccb | 537 | ret = __jbd2_log_start_commit(journal, tid); |
a931da6a | 538 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
539 | return ret; |
540 | } | |
541 | ||
542 | /* | |
9ff86446 DM |
543 | * Force and wait any uncommitted transactions. We can only force the running |
544 | * transaction if we don't have an active handle, otherwise, we will deadlock. | |
545 | * Returns: <0 in case of error, | |
546 | * 0 if nothing to commit, | |
547 | * 1 if transaction was successfully committed. | |
470decc6 | 548 | */ |
9ff86446 | 549 | static int __jbd2_journal_force_commit(journal_t *journal) |
470decc6 DK |
550 | { |
551 | transaction_t *transaction = NULL; | |
552 | tid_t tid; | |
9ff86446 | 553 | int need_to_start = 0, ret = 0; |
470decc6 | 554 | |
a931da6a | 555 | read_lock(&journal->j_state_lock); |
470decc6 DK |
556 | if (journal->j_running_transaction && !current->journal_info) { |
557 | transaction = journal->j_running_transaction; | |
e4471831 TT |
558 | if (!tid_geq(journal->j_commit_request, transaction->t_tid)) |
559 | need_to_start = 1; | |
470decc6 DK |
560 | } else if (journal->j_committing_transaction) |
561 | transaction = journal->j_committing_transaction; | |
562 | ||
563 | if (!transaction) { | |
9ff86446 | 564 | /* Nothing to commit */ |
a931da6a | 565 | read_unlock(&journal->j_state_lock); |
9ff86446 | 566 | return 0; |
470decc6 | 567 | } |
470decc6 | 568 | tid = transaction->t_tid; |
a931da6a | 569 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
570 | if (need_to_start) |
571 | jbd2_log_start_commit(journal, tid); | |
9ff86446 DM |
572 | ret = jbd2_log_wait_commit(journal, tid); |
573 | if (!ret) | |
574 | ret = 1; | |
575 | ||
576 | return ret; | |
577 | } | |
578 | ||
579 | /** | |
580 | * Force and wait upon a commit if the calling process is not within | |
581 | * transaction. This is used for forcing out undo-protected data which contains | |
582 | * bitmaps, when the fs is running out of space. | |
583 | * | |
584 | * @journal: journal to force | |
585 | * Returns true if progress was made. | |
586 | */ | |
587 | int jbd2_journal_force_commit_nested(journal_t *journal) | |
588 | { | |
589 | int ret; | |
590 | ||
591 | ret = __jbd2_journal_force_commit(journal); | |
592 | return ret > 0; | |
593 | } | |
594 | ||
595 | /** | |
596 | * int journal_force_commit() - force any uncommitted transactions | |
597 | * @journal: journal to force | |
598 | * | |
599 | * Caller want unconditional commit. We can only force the running transaction | |
600 | * if we don't have an active handle, otherwise, we will deadlock. | |
601 | */ | |
602 | int jbd2_journal_force_commit(journal_t *journal) | |
603 | { | |
604 | int ret; | |
605 | ||
606 | J_ASSERT(!current->journal_info); | |
607 | ret = __jbd2_journal_force_commit(journal); | |
608 | if (ret > 0) | |
609 | ret = 0; | |
610 | return ret; | |
470decc6 DK |
611 | } |
612 | ||
613 | /* | |
614 | * Start a commit of the current running transaction (if any). Returns true | |
c88ccea3 JK |
615 | * if a transaction is going to be committed (or is currently already |
616 | * committing), and fills its tid in at *ptid | |
470decc6 | 617 | */ |
f7f4bccb | 618 | int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) |
470decc6 DK |
619 | { |
620 | int ret = 0; | |
621 | ||
a931da6a | 622 | write_lock(&journal->j_state_lock); |
470decc6 DK |
623 | if (journal->j_running_transaction) { |
624 | tid_t tid = journal->j_running_transaction->t_tid; | |
625 | ||
c88ccea3 JK |
626 | __jbd2_log_start_commit(journal, tid); |
627 | /* There's a running transaction and we've just made sure | |
628 | * it's commit has been scheduled. */ | |
629 | if (ptid) | |
470decc6 | 630 | *ptid = tid; |
c88ccea3 JK |
631 | ret = 1; |
632 | } else if (journal->j_committing_transaction) { | |
470decc6 | 633 | /* |
12810ad7 AB |
634 | * If commit has been started, then we have to wait for |
635 | * completion of that transaction. | |
470decc6 | 636 | */ |
c88ccea3 JK |
637 | if (ptid) |
638 | *ptid = journal->j_committing_transaction->t_tid; | |
470decc6 DK |
639 | ret = 1; |
640 | } | |
a931da6a | 641 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
642 | return ret; |
643 | } | |
644 | ||
bbd2be36 JK |
645 | /* |
646 | * Return 1 if a given transaction has not yet sent barrier request | |
647 | * connected with a transaction commit. If 0 is returned, transaction | |
648 | * may or may not have sent the barrier. Used to avoid sending barrier | |
649 | * twice in common cases. | |
650 | */ | |
651 | int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) | |
652 | { | |
653 | int ret = 0; | |
654 | transaction_t *commit_trans; | |
655 | ||
656 | if (!(journal->j_flags & JBD2_BARRIER)) | |
657 | return 0; | |
658 | read_lock(&journal->j_state_lock); | |
659 | /* Transaction already committed? */ | |
660 | if (tid_geq(journal->j_commit_sequence, tid)) | |
661 | goto out; | |
662 | commit_trans = journal->j_committing_transaction; | |
663 | if (!commit_trans || commit_trans->t_tid != tid) { | |
664 | ret = 1; | |
665 | goto out; | |
666 | } | |
667 | /* | |
668 | * Transaction is being committed and we already proceeded to | |
669 | * submitting a flush to fs partition? | |
670 | */ | |
671 | if (journal->j_fs_dev != journal->j_dev) { | |
672 | if (!commit_trans->t_need_data_flush || | |
673 | commit_trans->t_state >= T_COMMIT_DFLUSH) | |
674 | goto out; | |
675 | } else { | |
676 | if (commit_trans->t_state >= T_COMMIT_JFLUSH) | |
677 | goto out; | |
678 | } | |
679 | ret = 1; | |
680 | out: | |
681 | read_unlock(&journal->j_state_lock); | |
682 | return ret; | |
683 | } | |
684 | EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); | |
685 | ||
470decc6 DK |
686 | /* |
687 | * Wait for a specified commit to complete. | |
688 | * The caller may not hold the journal lock. | |
689 | */ | |
f7f4bccb | 690 | int jbd2_log_wait_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
691 | { |
692 | int err = 0; | |
693 | ||
1eaa566d | 694 | jbd2_might_wait_for_commit(journal); |
a931da6a | 695 | read_lock(&journal->j_state_lock); |
e23291b9 | 696 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 | 697 | if (!tid_geq(journal->j_commit_request, tid)) { |
75685071 | 698 | printk(KERN_ERR |
470decc6 | 699 | "%s: error: j_commit_request=%d, tid=%d\n", |
329d291f | 700 | __func__, journal->j_commit_request, tid); |
470decc6 | 701 | } |
470decc6 | 702 | #endif |
470decc6 | 703 | while (tid_gt(tid, journal->j_commit_sequence)) { |
f2a44523 | 704 | jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n", |
470decc6 | 705 | tid, journal->j_commit_sequence); |
a931da6a | 706 | read_unlock(&journal->j_state_lock); |
3469a32a | 707 | wake_up(&journal->j_wait_commit); |
470decc6 DK |
708 | wait_event(journal->j_wait_done_commit, |
709 | !tid_gt(tid, journal->j_commit_sequence)); | |
a931da6a | 710 | read_lock(&journal->j_state_lock); |
470decc6 | 711 | } |
a931da6a | 712 | read_unlock(&journal->j_state_lock); |
470decc6 | 713 | |
75685071 | 714 | if (unlikely(is_journal_aborted(journal))) |
470decc6 | 715 | err = -EIO; |
470decc6 DK |
716 | return err; |
717 | } | |
718 | ||
d76a3a77 TT |
719 | /* |
720 | * When this function returns the transaction corresponding to tid | |
721 | * will be completed. If the transaction has currently running, start | |
722 | * committing that transaction before waiting for it to complete. If | |
723 | * the transaction id is stale, it is by definition already completed, | |
724 | * so just return SUCCESS. | |
725 | */ | |
726 | int jbd2_complete_transaction(journal_t *journal, tid_t tid) | |
727 | { | |
728 | int need_to_wait = 1; | |
729 | ||
730 | read_lock(&journal->j_state_lock); | |
731 | if (journal->j_running_transaction && | |
732 | journal->j_running_transaction->t_tid == tid) { | |
733 | if (journal->j_commit_request != tid) { | |
734 | /* transaction not yet started, so request it */ | |
735 | read_unlock(&journal->j_state_lock); | |
736 | jbd2_log_start_commit(journal, tid); | |
737 | goto wait_commit; | |
738 | } | |
739 | } else if (!(journal->j_committing_transaction && | |
740 | journal->j_committing_transaction->t_tid == tid)) | |
741 | need_to_wait = 0; | |
742 | read_unlock(&journal->j_state_lock); | |
743 | if (!need_to_wait) | |
744 | return 0; | |
745 | wait_commit: | |
746 | return jbd2_log_wait_commit(journal, tid); | |
747 | } | |
748 | EXPORT_SYMBOL(jbd2_complete_transaction); | |
749 | ||
470decc6 DK |
750 | /* |
751 | * Log buffer allocation routines: | |
752 | */ | |
753 | ||
18eba7aa | 754 | int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) |
470decc6 DK |
755 | { |
756 | unsigned long blocknr; | |
757 | ||
a931da6a | 758 | write_lock(&journal->j_state_lock); |
470decc6 DK |
759 | J_ASSERT(journal->j_free > 1); |
760 | ||
761 | blocknr = journal->j_head; | |
762 | journal->j_head++; | |
763 | journal->j_free--; | |
764 | if (journal->j_head == journal->j_last) | |
765 | journal->j_head = journal->j_first; | |
a931da6a | 766 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 767 | return jbd2_journal_bmap(journal, blocknr, retp); |
470decc6 DK |
768 | } |
769 | ||
770 | /* | |
771 | * Conversion of logical to physical block numbers for the journal | |
772 | * | |
773 | * On external journals the journal blocks are identity-mapped, so | |
774 | * this is a no-op. If needed, we can use j_blk_offset - everything is | |
775 | * ready. | |
776 | */ | |
f7f4bccb | 777 | int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, |
18eba7aa | 778 | unsigned long long *retp) |
470decc6 DK |
779 | { |
780 | int err = 0; | |
18eba7aa | 781 | unsigned long long ret; |
470decc6 DK |
782 | |
783 | if (journal->j_inode) { | |
784 | ret = bmap(journal->j_inode, blocknr); | |
785 | if (ret) | |
786 | *retp = ret; | |
787 | else { | |
470decc6 DK |
788 | printk(KERN_ALERT "%s: journal block not found " |
789 | "at offset %lu on %s\n", | |
05496769 | 790 | __func__, blocknr, journal->j_devname); |
470decc6 DK |
791 | err = -EIO; |
792 | __journal_abort_soft(journal, err); | |
793 | } | |
794 | } else { | |
795 | *retp = blocknr; /* +journal->j_blk_offset */ | |
796 | } | |
797 | return err; | |
798 | } | |
799 | ||
800 | /* | |
801 | * We play buffer_head aliasing tricks to write data/metadata blocks to | |
802 | * the journal without copying their contents, but for journal | |
803 | * descriptor blocks we do need to generate bona fide buffers. | |
804 | * | |
f7f4bccb | 805 | * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying |
470decc6 DK |
806 | * the buffer's contents they really should run flush_dcache_page(bh->b_page). |
807 | * But we don't bother doing that, so there will be coherency problems with | |
808 | * mmaps of blockdevs which hold live JBD-controlled filesystems. | |
809 | */ | |
32ab6715 JK |
810 | struct buffer_head * |
811 | jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type) | |
470decc6 | 812 | { |
32ab6715 | 813 | journal_t *journal = transaction->t_journal; |
470decc6 | 814 | struct buffer_head *bh; |
18eba7aa | 815 | unsigned long long blocknr; |
32ab6715 | 816 | journal_header_t *header; |
470decc6 DK |
817 | int err; |
818 | ||
f7f4bccb | 819 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
820 | |
821 | if (err) | |
822 | return NULL; | |
823 | ||
824 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
4b905671 JK |
825 | if (!bh) |
826 | return NULL; | |
470decc6 DK |
827 | lock_buffer(bh); |
828 | memset(bh->b_data, 0, journal->j_blocksize); | |
32ab6715 JK |
829 | header = (journal_header_t *)bh->b_data; |
830 | header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); | |
831 | header->h_blocktype = cpu_to_be32(type); | |
832 | header->h_sequence = cpu_to_be32(transaction->t_tid); | |
470decc6 DK |
833 | set_buffer_uptodate(bh); |
834 | unlock_buffer(bh); | |
835 | BUFFER_TRACE(bh, "return this buffer"); | |
e5a120ae | 836 | return bh; |
470decc6 DK |
837 | } |
838 | ||
1101cd4d JK |
839 | void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh) |
840 | { | |
841 | struct jbd2_journal_block_tail *tail; | |
842 | __u32 csum; | |
843 | ||
844 | if (!jbd2_journal_has_csum_v2or3(j)) | |
845 | return; | |
846 | ||
847 | tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize - | |
848 | sizeof(struct jbd2_journal_block_tail)); | |
849 | tail->t_checksum = 0; | |
850 | csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); | |
851 | tail->t_checksum = cpu_to_be32(csum); | |
852 | } | |
853 | ||
79feb521 JK |
854 | /* |
855 | * Return tid of the oldest transaction in the journal and block in the journal | |
856 | * where the transaction starts. | |
857 | * | |
858 | * If the journal is now empty, return which will be the next transaction ID | |
859 | * we will write and where will that transaction start. | |
860 | * | |
861 | * The return value is 0 if journal tail cannot be pushed any further, 1 if | |
862 | * it can. | |
863 | */ | |
864 | int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, | |
865 | unsigned long *block) | |
866 | { | |
867 | transaction_t *transaction; | |
868 | int ret; | |
869 | ||
870 | read_lock(&journal->j_state_lock); | |
871 | spin_lock(&journal->j_list_lock); | |
872 | transaction = journal->j_checkpoint_transactions; | |
873 | if (transaction) { | |
874 | *tid = transaction->t_tid; | |
875 | *block = transaction->t_log_start; | |
876 | } else if ((transaction = journal->j_committing_transaction) != NULL) { | |
877 | *tid = transaction->t_tid; | |
878 | *block = transaction->t_log_start; | |
879 | } else if ((transaction = journal->j_running_transaction) != NULL) { | |
880 | *tid = transaction->t_tid; | |
881 | *block = journal->j_head; | |
882 | } else { | |
883 | *tid = journal->j_transaction_sequence; | |
884 | *block = journal->j_head; | |
885 | } | |
886 | ret = tid_gt(*tid, journal->j_tail_sequence); | |
887 | spin_unlock(&journal->j_list_lock); | |
888 | read_unlock(&journal->j_state_lock); | |
889 | ||
890 | return ret; | |
891 | } | |
892 | ||
893 | /* | |
894 | * Update information in journal structure and in on disk journal superblock | |
895 | * about log tail. This function does not check whether information passed in | |
896 | * really pushes log tail further. It's responsibility of the caller to make | |
897 | * sure provided log tail information is valid (e.g. by holding | |
898 | * j_checkpoint_mutex all the time between computing log tail and calling this | |
899 | * function as is the case with jbd2_cleanup_journal_tail()). | |
900 | * | |
901 | * Requires j_checkpoint_mutex | |
902 | */ | |
6f6a6fda | 903 | int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) |
79feb521 JK |
904 | { |
905 | unsigned long freed; | |
6f6a6fda | 906 | int ret; |
79feb521 JK |
907 | |
908 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); | |
909 | ||
910 | /* | |
911 | * We cannot afford for write to remain in drive's caches since as | |
912 | * soon as we update j_tail, next transaction can start reusing journal | |
913 | * space and if we lose sb update during power failure we'd replay | |
914 | * old transaction with possibly newly overwritten data. | |
915 | */ | |
6f6a6fda JQ |
916 | ret = jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA); |
917 | if (ret) | |
918 | goto out; | |
919 | ||
79feb521 JK |
920 | write_lock(&journal->j_state_lock); |
921 | freed = block - journal->j_tail; | |
922 | if (block < journal->j_tail) | |
923 | freed += journal->j_last - journal->j_first; | |
924 | ||
925 | trace_jbd2_update_log_tail(journal, tid, block, freed); | |
926 | jbd_debug(1, | |
927 | "Cleaning journal tail from %d to %d (offset %lu), " | |
928 | "freeing %lu\n", | |
929 | journal->j_tail_sequence, tid, block, freed); | |
930 | ||
931 | journal->j_free += freed; | |
932 | journal->j_tail_sequence = tid; | |
933 | journal->j_tail = block; | |
934 | write_unlock(&journal->j_state_lock); | |
6f6a6fda JQ |
935 | |
936 | out: | |
937 | return ret; | |
79feb521 JK |
938 | } |
939 | ||
3339578f JK |
940 | /* |
941 | * This is a variaon of __jbd2_update_log_tail which checks for validity of | |
942 | * provided log tail and locks j_checkpoint_mutex. So it is safe against races | |
943 | * with other threads updating log tail. | |
944 | */ | |
945 | void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) | |
946 | { | |
947 | mutex_lock(&journal->j_checkpoint_mutex); | |
948 | if (tid_gt(tid, journal->j_tail_sequence)) | |
949 | __jbd2_update_log_tail(journal, tid, block); | |
950 | mutex_unlock(&journal->j_checkpoint_mutex); | |
951 | } | |
952 | ||
8e85fb3f JL |
953 | struct jbd2_stats_proc_session { |
954 | journal_t *journal; | |
955 | struct transaction_stats_s *stats; | |
956 | int start; | |
957 | int max; | |
958 | }; | |
959 | ||
8e85fb3f JL |
960 | static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) |
961 | { | |
962 | return *pos ? NULL : SEQ_START_TOKEN; | |
963 | } | |
964 | ||
965 | static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) | |
966 | { | |
967 | return NULL; | |
968 | } | |
969 | ||
970 | static int jbd2_seq_info_show(struct seq_file *seq, void *v) | |
971 | { | |
972 | struct jbd2_stats_proc_session *s = seq->private; | |
973 | ||
974 | if (v != SEQ_START_TOKEN) | |
975 | return 0; | |
9fff24aa TT |
976 | seq_printf(seq, "%lu transactions (%lu requested), " |
977 | "each up to %u blocks\n", | |
978 | s->stats->ts_tid, s->stats->ts_requested, | |
979 | s->journal->j_max_transaction_buffers); | |
8e85fb3f JL |
980 | if (s->stats->ts_tid == 0) |
981 | return 0; | |
982 | seq_printf(seq, "average: \n %ums waiting for transaction\n", | |
bf699327 | 983 | jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); |
9fff24aa TT |
984 | seq_printf(seq, " %ums request delay\n", |
985 | (s->stats->ts_requested == 0) ? 0 : | |
986 | jiffies_to_msecs(s->stats->run.rs_request_delay / | |
987 | s->stats->ts_requested)); | |
8e85fb3f | 988 | seq_printf(seq, " %ums running transaction\n", |
bf699327 | 989 | jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); |
8e85fb3f | 990 | seq_printf(seq, " %ums transaction was being locked\n", |
bf699327 | 991 | jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); |
8e85fb3f | 992 | seq_printf(seq, " %ums flushing data (in ordered mode)\n", |
bf699327 | 993 | jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); |
8e85fb3f | 994 | seq_printf(seq, " %ums logging transaction\n", |
bf699327 | 995 | jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); |
c225aa57 SHT |
996 | seq_printf(seq, " %lluus average transaction commit time\n", |
997 | div_u64(s->journal->j_average_commit_time, 1000)); | |
8e85fb3f | 998 | seq_printf(seq, " %lu handles per transaction\n", |
bf699327 | 999 | s->stats->run.rs_handle_count / s->stats->ts_tid); |
8e85fb3f | 1000 | seq_printf(seq, " %lu blocks per transaction\n", |
bf699327 | 1001 | s->stats->run.rs_blocks / s->stats->ts_tid); |
8e85fb3f | 1002 | seq_printf(seq, " %lu logged blocks per transaction\n", |
bf699327 | 1003 | s->stats->run.rs_blocks_logged / s->stats->ts_tid); |
8e85fb3f JL |
1004 | return 0; |
1005 | } | |
1006 | ||
1007 | static void jbd2_seq_info_stop(struct seq_file *seq, void *v) | |
1008 | { | |
1009 | } | |
1010 | ||
88e9d34c | 1011 | static const struct seq_operations jbd2_seq_info_ops = { |
8e85fb3f JL |
1012 | .start = jbd2_seq_info_start, |
1013 | .next = jbd2_seq_info_next, | |
1014 | .stop = jbd2_seq_info_stop, | |
1015 | .show = jbd2_seq_info_show, | |
1016 | }; | |
1017 | ||
1018 | static int jbd2_seq_info_open(struct inode *inode, struct file *file) | |
1019 | { | |
d9dda78b | 1020 | journal_t *journal = PDE_DATA(inode); |
8e85fb3f JL |
1021 | struct jbd2_stats_proc_session *s; |
1022 | int rc, size; | |
1023 | ||
1024 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
1025 | if (s == NULL) | |
1026 | return -ENOMEM; | |
1027 | size = sizeof(struct transaction_stats_s); | |
1028 | s->stats = kmalloc(size, GFP_KERNEL); | |
1029 | if (s->stats == NULL) { | |
1030 | kfree(s); | |
1031 | return -ENOMEM; | |
1032 | } | |
1033 | spin_lock(&journal->j_history_lock); | |
1034 | memcpy(s->stats, &journal->j_stats, size); | |
1035 | s->journal = journal; | |
1036 | spin_unlock(&journal->j_history_lock); | |
1037 | ||
1038 | rc = seq_open(file, &jbd2_seq_info_ops); | |
1039 | if (rc == 0) { | |
1040 | struct seq_file *m = file->private_data; | |
1041 | m->private = s; | |
1042 | } else { | |
1043 | kfree(s->stats); | |
1044 | kfree(s); | |
1045 | } | |
1046 | return rc; | |
1047 | ||
1048 | } | |
1049 | ||
1050 | static int jbd2_seq_info_release(struct inode *inode, struct file *file) | |
1051 | { | |
1052 | struct seq_file *seq = file->private_data; | |
1053 | struct jbd2_stats_proc_session *s = seq->private; | |
1054 | kfree(s->stats); | |
1055 | kfree(s); | |
1056 | return seq_release(inode, file); | |
1057 | } | |
1058 | ||
828c0950 | 1059 | static const struct file_operations jbd2_seq_info_fops = { |
8e85fb3f JL |
1060 | .owner = THIS_MODULE, |
1061 | .open = jbd2_seq_info_open, | |
1062 | .read = seq_read, | |
1063 | .llseek = seq_lseek, | |
1064 | .release = jbd2_seq_info_release, | |
1065 | }; | |
1066 | ||
1067 | static struct proc_dir_entry *proc_jbd2_stats; | |
1068 | ||
1069 | static void jbd2_stats_proc_init(journal_t *journal) | |
1070 | { | |
05496769 | 1071 | journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); |
8e85fb3f | 1072 | if (journal->j_proc_entry) { |
79da3664 DL |
1073 | proc_create_data("info", S_IRUGO, journal->j_proc_entry, |
1074 | &jbd2_seq_info_fops, journal); | |
8e85fb3f JL |
1075 | } |
1076 | } | |
1077 | ||
1078 | static void jbd2_stats_proc_exit(journal_t *journal) | |
1079 | { | |
8e85fb3f | 1080 | remove_proc_entry("info", journal->j_proc_entry); |
05496769 | 1081 | remove_proc_entry(journal->j_devname, proc_jbd2_stats); |
8e85fb3f JL |
1082 | } |
1083 | ||
470decc6 DK |
1084 | /* |
1085 | * Management for journal control blocks: functions to create and | |
1086 | * destroy journal_t structures, and to initialise and read existing | |
1087 | * journal blocks from disk. */ | |
1088 | ||
1089 | /* First: create and setup a journal_t object in memory. We initialise | |
1090 | * very few fields yet: that has to wait until we have created the | |
1091 | * journal structures from from scratch, or loaded them from disk. */ | |
1092 | ||
1093 | static journal_t * journal_init_common (void) | |
1094 | { | |
ab714aff | 1095 | static struct lock_class_key jbd2_trans_commit_key; |
470decc6 DK |
1096 | journal_t *journal; |
1097 | int err; | |
1098 | ||
3ebfdf88 | 1099 | journal = kzalloc(sizeof(*journal), GFP_KERNEL); |
470decc6 | 1100 | if (!journal) |
b7271b0a | 1101 | return NULL; |
470decc6 DK |
1102 | |
1103 | init_waitqueue_head(&journal->j_wait_transaction_locked); | |
470decc6 | 1104 | init_waitqueue_head(&journal->j_wait_done_commit); |
470decc6 DK |
1105 | init_waitqueue_head(&journal->j_wait_commit); |
1106 | init_waitqueue_head(&journal->j_wait_updates); | |
8f7d89f3 | 1107 | init_waitqueue_head(&journal->j_wait_reserved); |
470decc6 DK |
1108 | mutex_init(&journal->j_barrier); |
1109 | mutex_init(&journal->j_checkpoint_mutex); | |
1110 | spin_lock_init(&journal->j_revoke_lock); | |
1111 | spin_lock_init(&journal->j_list_lock); | |
a931da6a | 1112 | rwlock_init(&journal->j_state_lock); |
470decc6 | 1113 | |
cd02ff0b | 1114 | journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); |
30773840 TT |
1115 | journal->j_min_batch_time = 0; |
1116 | journal->j_max_batch_time = 15000; /* 15ms */ | |
8f7d89f3 | 1117 | atomic_set(&journal->j_reserved_credits, 0); |
470decc6 DK |
1118 | |
1119 | /* The journal is marked for error until we succeed with recovery! */ | |
f7f4bccb | 1120 | journal->j_flags = JBD2_ABORT; |
470decc6 DK |
1121 | |
1122 | /* Set up a default-sized revoke table for the new mount. */ | |
f7f4bccb | 1123 | err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); |
470decc6 DK |
1124 | if (err) { |
1125 | kfree(journal); | |
b7271b0a | 1126 | return NULL; |
470decc6 | 1127 | } |
8e85fb3f | 1128 | |
bf699327 | 1129 | spin_lock_init(&journal->j_history_lock); |
8e85fb3f | 1130 | |
ab714aff JK |
1131 | lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle", |
1132 | &jbd2_trans_commit_key, 0); | |
1133 | ||
470decc6 | 1134 | return journal; |
470decc6 DK |
1135 | } |
1136 | ||
f7f4bccb | 1137 | /* jbd2_journal_init_dev and jbd2_journal_init_inode: |
470decc6 DK |
1138 | * |
1139 | * Create a journal structure assigned some fixed set of disk blocks to | |
1140 | * the journal. We don't actually touch those disk blocks yet, but we | |
1141 | * need to set up all of the mapping information to tell the journaling | |
1142 | * system where the journal blocks are. | |
1143 | * | |
1144 | */ | |
1145 | ||
1146 | /** | |
5648ba5b | 1147 | * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure |
470decc6 DK |
1148 | * @bdev: Block device on which to create the journal |
1149 | * @fs_dev: Device which hold journalled filesystem for this journal. | |
1150 | * @start: Block nr Start of journal. | |
1151 | * @len: Length of the journal in blocks. | |
1152 | * @blocksize: blocksize of journalling device | |
5648ba5b RD |
1153 | * |
1154 | * Returns: a newly created journal_t * | |
470decc6 | 1155 | * |
f7f4bccb | 1156 | * jbd2_journal_init_dev creates a journal which maps a fixed contiguous |
470decc6 DK |
1157 | * range of blocks on an arbitrary block device. |
1158 | * | |
1159 | */ | |
f7f4bccb | 1160 | journal_t * jbd2_journal_init_dev(struct block_device *bdev, |
470decc6 | 1161 | struct block_device *fs_dev, |
18eba7aa | 1162 | unsigned long long start, int len, int blocksize) |
470decc6 DK |
1163 | { |
1164 | journal_t *journal = journal_init_common(); | |
1165 | struct buffer_head *bh; | |
1166 | int n; | |
1167 | ||
1168 | if (!journal) | |
1169 | return NULL; | |
1170 | ||
1171 | /* journal descriptor can store up to n blocks -bzzz */ | |
1172 | journal->j_blocksize = blocksize; | |
0587aa3d | 1173 | journal->j_dev = bdev; |
1174 | journal->j_fs_dev = fs_dev; | |
1175 | journal->j_blk_offset = start; | |
1176 | journal->j_maxlen = len; | |
1177 | bdevname(journal->j_dev, journal->j_devname); | |
81ae394b | 1178 | strreplace(journal->j_devname, '/', '!'); |
4b905671 | 1179 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1180 | n = journal->j_blocksize / sizeof(journal_block_tag_t); |
1181 | journal->j_wbufsize = n; | |
1182 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1183 | if (!journal->j_wbuf) { | |
25985edc | 1184 | printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", |
329d291f | 1185 | __func__); |
4b905671 | 1186 | goto out_err; |
470decc6 | 1187 | } |
470decc6 DK |
1188 | |
1189 | bh = __getblk(journal->j_dev, start, journal->j_blocksize); | |
4b905671 JK |
1190 | if (!bh) { |
1191 | printk(KERN_ERR | |
1192 | "%s: Cannot get buffer for journal superblock\n", | |
1193 | __func__); | |
1194 | goto out_err; | |
1195 | } | |
470decc6 DK |
1196 | journal->j_sb_buffer = bh; |
1197 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
4b905671 | 1198 | |
470decc6 | 1199 | return journal; |
4b905671 | 1200 | out_err: |
7b02bec0 | 1201 | kfree(journal->j_wbuf); |
4b905671 JK |
1202 | jbd2_stats_proc_exit(journal); |
1203 | kfree(journal); | |
1204 | return NULL; | |
470decc6 DK |
1205 | } |
1206 | ||
1207 | /** | |
f7f4bccb | 1208 | * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. |
470decc6 DK |
1209 | * @inode: An inode to create the journal in |
1210 | * | |
f7f4bccb | 1211 | * jbd2_journal_init_inode creates a journal which maps an on-disk inode as |
470decc6 DK |
1212 | * the journal. The inode must exist already, must support bmap() and |
1213 | * must have all data blocks preallocated. | |
1214 | */ | |
f7f4bccb | 1215 | journal_t * jbd2_journal_init_inode (struct inode *inode) |
470decc6 DK |
1216 | { |
1217 | struct buffer_head *bh; | |
1218 | journal_t *journal = journal_init_common(); | |
05496769 | 1219 | char *p; |
470decc6 DK |
1220 | int err; |
1221 | int n; | |
18eba7aa | 1222 | unsigned long long blocknr; |
470decc6 DK |
1223 | |
1224 | if (!journal) | |
1225 | return NULL; | |
1226 | ||
1227 | journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev; | |
1228 | journal->j_inode = inode; | |
05496769 | 1229 | bdevname(journal->j_dev, journal->j_devname); |
81ae394b | 1230 | p = strreplace(journal->j_devname, '/', '!'); |
90576c0b | 1231 | sprintf(p, "-%lu", journal->j_inode->i_ino); |
470decc6 DK |
1232 | jbd_debug(1, |
1233 | "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n", | |
1234 | journal, inode->i_sb->s_id, inode->i_ino, | |
1235 | (long long) inode->i_size, | |
1236 | inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); | |
1237 | ||
1238 | journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits; | |
1239 | journal->j_blocksize = inode->i_sb->s_blocksize; | |
8e85fb3f | 1240 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1241 | |
1242 | /* journal descriptor can store up to n blocks -bzzz */ | |
1243 | n = journal->j_blocksize / sizeof(journal_block_tag_t); | |
1244 | journal->j_wbufsize = n; | |
1245 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1246 | if (!journal->j_wbuf) { | |
25985edc | 1247 | printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", |
329d291f | 1248 | __func__); |
4b905671 | 1249 | goto out_err; |
470decc6 DK |
1250 | } |
1251 | ||
f7f4bccb | 1252 | err = jbd2_journal_bmap(journal, 0, &blocknr); |
470decc6 DK |
1253 | /* If that failed, give up */ |
1254 | if (err) { | |
3c26bdb4 | 1255 | printk(KERN_ERR "%s: Cannot locate journal superblock\n", |
329d291f | 1256 | __func__); |
4b905671 | 1257 | goto out_err; |
470decc6 DK |
1258 | } |
1259 | ||
a49058fa | 1260 | bh = getblk_unmovable(journal->j_dev, blocknr, journal->j_blocksize); |
4b905671 JK |
1261 | if (!bh) { |
1262 | printk(KERN_ERR | |
1263 | "%s: Cannot get buffer for journal superblock\n", | |
1264 | __func__); | |
1265 | goto out_err; | |
1266 | } | |
470decc6 DK |
1267 | journal->j_sb_buffer = bh; |
1268 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
1269 | ||
1270 | return journal; | |
4b905671 | 1271 | out_err: |
7b02bec0 | 1272 | kfree(journal->j_wbuf); |
4b905671 JK |
1273 | jbd2_stats_proc_exit(journal); |
1274 | kfree(journal); | |
1275 | return NULL; | |
470decc6 DK |
1276 | } |
1277 | ||
1278 | /* | |
1279 | * If the journal init or create aborts, we need to mark the journal | |
1280 | * superblock as being NULL to prevent the journal destroy from writing | |
1281 | * back a bogus superblock. | |
1282 | */ | |
1283 | static void journal_fail_superblock (journal_t *journal) | |
1284 | { | |
1285 | struct buffer_head *bh = journal->j_sb_buffer; | |
1286 | brelse(bh); | |
1287 | journal->j_sb_buffer = NULL; | |
1288 | } | |
1289 | ||
1290 | /* | |
1291 | * Given a journal_t structure, initialise the various fields for | |
1292 | * startup of a new journaling session. We use this both when creating | |
1293 | * a journal, and after recovering an old journal to reset it for | |
1294 | * subsequent use. | |
1295 | */ | |
1296 | ||
1297 | static int journal_reset(journal_t *journal) | |
1298 | { | |
1299 | journal_superblock_t *sb = journal->j_superblock; | |
18eba7aa | 1300 | unsigned long long first, last; |
470decc6 DK |
1301 | |
1302 | first = be32_to_cpu(sb->s_first); | |
1303 | last = be32_to_cpu(sb->s_maxlen); | |
f6f50e28 | 1304 | if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) { |
f2a44523 | 1305 | printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n", |
f6f50e28 JK |
1306 | first, last); |
1307 | journal_fail_superblock(journal); | |
1308 | return -EINVAL; | |
1309 | } | |
470decc6 DK |
1310 | |
1311 | journal->j_first = first; | |
1312 | journal->j_last = last; | |
1313 | ||
1314 | journal->j_head = first; | |
1315 | journal->j_tail = first; | |
1316 | journal->j_free = last - first; | |
1317 | ||
1318 | journal->j_tail_sequence = journal->j_transaction_sequence; | |
1319 | journal->j_commit_sequence = journal->j_transaction_sequence - 1; | |
1320 | journal->j_commit_request = journal->j_commit_sequence; | |
1321 | ||
1322 | journal->j_max_transaction_buffers = journal->j_maxlen / 4; | |
1323 | ||
470decc6 DK |
1324 | /* |
1325 | * As a special case, if the on-disk copy is already marked as needing | |
24bcc89c JK |
1326 | * no recovery (s_start == 0), then we can safely defer the superblock |
1327 | * update until the next commit by setting JBD2_FLUSHED. This avoids | |
470decc6 DK |
1328 | * attempting a write to a potential-readonly device. |
1329 | */ | |
24bcc89c | 1330 | if (sb->s_start == 0) { |
f2a44523 | 1331 | jbd_debug(1, "JBD2: Skipping superblock update on recovered sb " |
470decc6 DK |
1332 | "(start %ld, seq %d, errno %d)\n", |
1333 | journal->j_tail, journal->j_tail_sequence, | |
1334 | journal->j_errno); | |
24bcc89c JK |
1335 | journal->j_flags |= JBD2_FLUSHED; |
1336 | } else { | |
a78bb11d JK |
1337 | /* Lock here to make assertions happy... */ |
1338 | mutex_lock(&journal->j_checkpoint_mutex); | |
79feb521 JK |
1339 | /* |
1340 | * Update log tail information. We use WRITE_FUA since new | |
1341 | * transaction will start reusing journal space and so we | |
1342 | * must make sure information about current log tail is on | |
1343 | * disk before that. | |
1344 | */ | |
1345 | jbd2_journal_update_sb_log_tail(journal, | |
1346 | journal->j_tail_sequence, | |
1347 | journal->j_tail, | |
1348 | WRITE_FUA); | |
a78bb11d | 1349 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 | 1350 | } |
24bcc89c JK |
1351 | return jbd2_journal_start_thread(journal); |
1352 | } | |
470decc6 | 1353 | |
2a222ca9 | 1354 | static int jbd2_write_superblock(journal_t *journal, int write_flags) |
24bcc89c JK |
1355 | { |
1356 | struct buffer_head *bh = journal->j_sb_buffer; | |
fe52d17c | 1357 | journal_superblock_t *sb = journal->j_superblock; |
79feb521 | 1358 | int ret; |
470decc6 | 1359 | |
2a222ca9 | 1360 | trace_jbd2_write_superblock(journal, write_flags); |
79feb521 | 1361 | if (!(journal->j_flags & JBD2_BARRIER)) |
28a8f0d3 | 1362 | write_flags &= ~(REQ_FUA | REQ_PREFLUSH); |
79feb521 | 1363 | lock_buffer(bh); |
914258bf TT |
1364 | if (buffer_write_io_error(bh)) { |
1365 | /* | |
1366 | * Oh, dear. A previous attempt to write the journal | |
1367 | * superblock failed. This could happen because the | |
1368 | * USB device was yanked out. Or it could happen to | |
1369 | * be a transient write error and maybe the block will | |
1370 | * be remapped. Nothing we can do but to retry the | |
1371 | * write and hope for the best. | |
1372 | */ | |
1373 | printk(KERN_ERR "JBD2: previous I/O error detected " | |
1374 | "for journal superblock update for %s.\n", | |
1375 | journal->j_devname); | |
1376 | clear_buffer_write_io_error(bh); | |
1377 | set_buffer_uptodate(bh); | |
1378 | } | |
fe52d17c | 1379 | jbd2_superblock_csum_set(journal, sb); |
79feb521 JK |
1380 | get_bh(bh); |
1381 | bh->b_end_io = end_buffer_write_sync; | |
2a222ca9 | 1382 | ret = submit_bh(REQ_OP_WRITE, write_flags, bh); |
79feb521 | 1383 | wait_on_buffer(bh); |
24bcc89c | 1384 | if (buffer_write_io_error(bh)) { |
24bcc89c JK |
1385 | clear_buffer_write_io_error(bh); |
1386 | set_buffer_uptodate(bh); | |
79feb521 JK |
1387 | ret = -EIO; |
1388 | } | |
1389 | if (ret) { | |
1390 | printk(KERN_ERR "JBD2: Error %d detected when updating " | |
1391 | "journal superblock for %s.\n", ret, | |
1392 | journal->j_devname); | |
6f6a6fda | 1393 | jbd2_journal_abort(journal, ret); |
24bcc89c | 1394 | } |
6f6a6fda JQ |
1395 | |
1396 | return ret; | |
24bcc89c JK |
1397 | } |
1398 | ||
1399 | /** | |
1400 | * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. | |
1401 | * @journal: The journal to update. | |
79feb521 JK |
1402 | * @tail_tid: TID of the new transaction at the tail of the log |
1403 | * @tail_block: The first block of the transaction at the tail of the log | |
1404 | * @write_op: With which operation should we write the journal sb | |
24bcc89c JK |
1405 | * |
1406 | * Update a journal's superblock information about log tail and write it to | |
1407 | * disk, waiting for the IO to complete. | |
1408 | */ | |
6f6a6fda | 1409 | int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, |
79feb521 | 1410 | unsigned long tail_block, int write_op) |
24bcc89c JK |
1411 | { |
1412 | journal_superblock_t *sb = journal->j_superblock; | |
6f6a6fda | 1413 | int ret; |
24bcc89c | 1414 | |
a78bb11d | 1415 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); |
79feb521 JK |
1416 | jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", |
1417 | tail_block, tail_tid); | |
470decc6 | 1418 | |
79feb521 JK |
1419 | sb->s_sequence = cpu_to_be32(tail_tid); |
1420 | sb->s_start = cpu_to_be32(tail_block); | |
470decc6 | 1421 | |
6f6a6fda JQ |
1422 | ret = jbd2_write_superblock(journal, write_op); |
1423 | if (ret) | |
1424 | goto out; | |
470decc6 | 1425 | |
24bcc89c JK |
1426 | /* Log is no longer empty */ |
1427 | write_lock(&journal->j_state_lock); | |
1428 | WARN_ON(!sb->s_sequence); | |
1429 | journal->j_flags &= ~JBD2_FLUSHED; | |
1430 | write_unlock(&journal->j_state_lock); | |
6f6a6fda JQ |
1431 | |
1432 | out: | |
1433 | return ret; | |
24bcc89c | 1434 | } |
2201c590 | 1435 | |
24bcc89c JK |
1436 | /** |
1437 | * jbd2_mark_journal_empty() - Mark on disk journal as empty. | |
1438 | * @journal: The journal to update. | |
c0a2ad9b | 1439 | * @write_op: With which operation should we write the journal sb |
24bcc89c JK |
1440 | * |
1441 | * Update a journal's dynamic superblock fields to show that journal is empty. | |
1442 | * Write updated superblock to disk waiting for IO to complete. | |
1443 | */ | |
c0a2ad9b | 1444 | static void jbd2_mark_journal_empty(journal_t *journal, int write_op) |
24bcc89c JK |
1445 | { |
1446 | journal_superblock_t *sb = journal->j_superblock; | |
914258bf | 1447 | |
a78bb11d | 1448 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); |
a931da6a | 1449 | read_lock(&journal->j_state_lock); |
eeecef0a ES |
1450 | /* Is it already empty? */ |
1451 | if (sb->s_start == 0) { | |
1452 | read_unlock(&journal->j_state_lock); | |
1453 | return; | |
1454 | } | |
24bcc89c JK |
1455 | jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n", |
1456 | journal->j_tail_sequence); | |
470decc6 DK |
1457 | |
1458 | sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); | |
24bcc89c | 1459 | sb->s_start = cpu_to_be32(0); |
a931da6a | 1460 | read_unlock(&journal->j_state_lock); |
470decc6 | 1461 | |
c0a2ad9b | 1462 | jbd2_write_superblock(journal, write_op); |
470decc6 | 1463 | |
24bcc89c | 1464 | /* Log is no longer empty */ |
a931da6a | 1465 | write_lock(&journal->j_state_lock); |
24bcc89c | 1466 | journal->j_flags |= JBD2_FLUSHED; |
a931da6a | 1467 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1468 | } |
1469 | ||
24bcc89c JK |
1470 | |
1471 | /** | |
1472 | * jbd2_journal_update_sb_errno() - Update error in the journal. | |
1473 | * @journal: The journal to update. | |
1474 | * | |
1475 | * Update a journal's errno. Write updated superblock to disk waiting for IO | |
1476 | * to complete. | |
1477 | */ | |
d796c52e | 1478 | void jbd2_journal_update_sb_errno(journal_t *journal) |
24bcc89c JK |
1479 | { |
1480 | journal_superblock_t *sb = journal->j_superblock; | |
1481 | ||
1482 | read_lock(&journal->j_state_lock); | |
1483 | jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", | |
1484 | journal->j_errno); | |
1485 | sb->s_errno = cpu_to_be32(journal->j_errno); | |
1486 | read_unlock(&journal->j_state_lock); | |
1487 | ||
564bc402 | 1488 | jbd2_write_superblock(journal, WRITE_FUA); |
24bcc89c | 1489 | } |
d796c52e | 1490 | EXPORT_SYMBOL(jbd2_journal_update_sb_errno); |
24bcc89c | 1491 | |
470decc6 DK |
1492 | /* |
1493 | * Read the superblock for a given journal, performing initial | |
1494 | * validation of the format. | |
1495 | */ | |
470decc6 DK |
1496 | static int journal_get_superblock(journal_t *journal) |
1497 | { | |
1498 | struct buffer_head *bh; | |
1499 | journal_superblock_t *sb; | |
1500 | int err = -EIO; | |
1501 | ||
1502 | bh = journal->j_sb_buffer; | |
1503 | ||
1504 | J_ASSERT(bh != NULL); | |
1505 | if (!buffer_uptodate(bh)) { | |
dfec8a14 | 1506 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
470decc6 DK |
1507 | wait_on_buffer(bh); |
1508 | if (!buffer_uptodate(bh)) { | |
f2a44523 EG |
1509 | printk(KERN_ERR |
1510 | "JBD2: IO error reading journal superblock\n"); | |
470decc6 DK |
1511 | goto out; |
1512 | } | |
1513 | } | |
1514 | ||
25ed6e8a DW |
1515 | if (buffer_verified(bh)) |
1516 | return 0; | |
1517 | ||
470decc6 DK |
1518 | sb = journal->j_superblock; |
1519 | ||
1520 | err = -EINVAL; | |
1521 | ||
f7f4bccb | 1522 | if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || |
470decc6 | 1523 | sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { |
f2a44523 | 1524 | printk(KERN_WARNING "JBD2: no valid journal superblock found\n"); |
470decc6 DK |
1525 | goto out; |
1526 | } | |
1527 | ||
1528 | switch(be32_to_cpu(sb->s_header.h_blocktype)) { | |
f7f4bccb | 1529 | case JBD2_SUPERBLOCK_V1: |
470decc6 DK |
1530 | journal->j_format_version = 1; |
1531 | break; | |
f7f4bccb | 1532 | case JBD2_SUPERBLOCK_V2: |
470decc6 DK |
1533 | journal->j_format_version = 2; |
1534 | break; | |
1535 | default: | |
f2a44523 | 1536 | printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n"); |
470decc6 DK |
1537 | goto out; |
1538 | } | |
1539 | ||
1540 | if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen) | |
1541 | journal->j_maxlen = be32_to_cpu(sb->s_maxlen); | |
1542 | else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) { | |
f2a44523 | 1543 | printk(KERN_WARNING "JBD2: journal file too short\n"); |
470decc6 DK |
1544 | goto out; |
1545 | } | |
1546 | ||
8762202d EG |
1547 | if (be32_to_cpu(sb->s_first) == 0 || |
1548 | be32_to_cpu(sb->s_first) >= journal->j_maxlen) { | |
1549 | printk(KERN_WARNING | |
1550 | "JBD2: Invalid start block of journal: %u\n", | |
1551 | be32_to_cpu(sb->s_first)); | |
1552 | goto out; | |
1553 | } | |
1554 | ||
56316a0d DW |
1555 | if (jbd2_has_feature_csum2(journal) && |
1556 | jbd2_has_feature_csum3(journal)) { | |
db9ee220 DW |
1557 | /* Can't have checksum v2 and v3 at the same time! */ |
1558 | printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 " | |
1559 | "at the same time!\n"); | |
1560 | goto out; | |
1561 | } | |
1562 | ||
8595798c | 1563 | if (jbd2_journal_has_csum_v2or3_feature(journal) && |
56316a0d | 1564 | jbd2_has_feature_checksum(journal)) { |
feb8c6d3 DW |
1565 | /* Can't have checksum v1 and v2 on at the same time! */ |
1566 | printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 " | |
1567 | "at the same time!\n"); | |
1568 | goto out; | |
1569 | } | |
1570 | ||
25ed6e8a | 1571 | if (!jbd2_verify_csum_type(journal, sb)) { |
a67c848a | 1572 | printk(KERN_ERR "JBD2: Unknown checksum type\n"); |
25ed6e8a DW |
1573 | goto out; |
1574 | } | |
1575 | ||
01b5adce | 1576 | /* Load the checksum driver */ |
8595798c | 1577 | if (jbd2_journal_has_csum_v2or3_feature(journal)) { |
01b5adce DW |
1578 | journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); |
1579 | if (IS_ERR(journal->j_chksum_driver)) { | |
a67c848a | 1580 | printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n"); |
01b5adce DW |
1581 | err = PTR_ERR(journal->j_chksum_driver); |
1582 | journal->j_chksum_driver = NULL; | |
1583 | goto out; | |
1584 | } | |
1585 | } | |
1586 | ||
4fd5ea43 DW |
1587 | /* Check superblock checksum */ |
1588 | if (!jbd2_superblock_csum_verify(journal, sb)) { | |
a67c848a | 1589 | printk(KERN_ERR "JBD2: journal checksum error\n"); |
6a797d27 | 1590 | err = -EFSBADCRC; |
4fd5ea43 DW |
1591 | goto out; |
1592 | } | |
1593 | ||
1594 | /* Precompute checksum seed for all metadata */ | |
db9ee220 | 1595 | if (jbd2_journal_has_csum_v2or3(journal)) |
4fd5ea43 DW |
1596 | journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, |
1597 | sizeof(sb->s_uuid)); | |
1598 | ||
25ed6e8a DW |
1599 | set_buffer_verified(bh); |
1600 | ||
470decc6 DK |
1601 | return 0; |
1602 | ||
1603 | out: | |
1604 | journal_fail_superblock(journal); | |
1605 | return err; | |
1606 | } | |
1607 | ||
1608 | /* | |
1609 | * Load the on-disk journal superblock and read the key fields into the | |
1610 | * journal_t. | |
1611 | */ | |
1612 | ||
1613 | static int load_superblock(journal_t *journal) | |
1614 | { | |
1615 | int err; | |
1616 | journal_superblock_t *sb; | |
1617 | ||
1618 | err = journal_get_superblock(journal); | |
1619 | if (err) | |
1620 | return err; | |
1621 | ||
1622 | sb = journal->j_superblock; | |
1623 | ||
1624 | journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); | |
1625 | journal->j_tail = be32_to_cpu(sb->s_start); | |
1626 | journal->j_first = be32_to_cpu(sb->s_first); | |
1627 | journal->j_last = be32_to_cpu(sb->s_maxlen); | |
1628 | journal->j_errno = be32_to_cpu(sb->s_errno); | |
1629 | ||
1630 | return 0; | |
1631 | } | |
1632 | ||
1633 | ||
1634 | /** | |
f7f4bccb | 1635 | * int jbd2_journal_load() - Read journal from disk. |
470decc6 DK |
1636 | * @journal: Journal to act on. |
1637 | * | |
1638 | * Given a journal_t structure which tells us which disk blocks contain | |
1639 | * a journal, read the journal from disk to initialise the in-memory | |
1640 | * structures. | |
1641 | */ | |
f7f4bccb | 1642 | int jbd2_journal_load(journal_t *journal) |
470decc6 DK |
1643 | { |
1644 | int err; | |
1645 | journal_superblock_t *sb; | |
1646 | ||
1647 | err = load_superblock(journal); | |
1648 | if (err) | |
1649 | return err; | |
1650 | ||
1651 | sb = journal->j_superblock; | |
1652 | /* If this is a V2 superblock, then we have to check the | |
1653 | * features flags on it. */ | |
1654 | ||
1655 | if (journal->j_format_version >= 2) { | |
1656 | if ((sb->s_feature_ro_compat & | |
f7f4bccb | 1657 | ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || |
470decc6 | 1658 | (sb->s_feature_incompat & |
f7f4bccb | 1659 | ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { |
f2a44523 EG |
1660 | printk(KERN_WARNING |
1661 | "JBD2: Unrecognised features on journal\n"); | |
470decc6 DK |
1662 | return -EINVAL; |
1663 | } | |
1664 | } | |
1665 | ||
d2eecb03 TT |
1666 | /* |
1667 | * Create a slab for this blocksize | |
1668 | */ | |
1669 | err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize)); | |
1670 | if (err) | |
1671 | return err; | |
1672 | ||
470decc6 DK |
1673 | /* Let the recovery code check whether it needs to recover any |
1674 | * data from the journal. */ | |
f7f4bccb | 1675 | if (jbd2_journal_recover(journal)) |
470decc6 DK |
1676 | goto recovery_error; |
1677 | ||
e6a47428 TT |
1678 | if (journal->j_failed_commit) { |
1679 | printk(KERN_ERR "JBD2: journal transaction %u on %s " | |
1680 | "is corrupt.\n", journal->j_failed_commit, | |
1681 | journal->j_devname); | |
6a797d27 | 1682 | return -EFSCORRUPTED; |
e6a47428 TT |
1683 | } |
1684 | ||
470decc6 DK |
1685 | /* OK, we've finished with the dynamic journal bits: |
1686 | * reinitialise the dynamic contents of the superblock in memory | |
1687 | * and reset them on disk. */ | |
1688 | if (journal_reset(journal)) | |
1689 | goto recovery_error; | |
1690 | ||
f7f4bccb MC |
1691 | journal->j_flags &= ~JBD2_ABORT; |
1692 | journal->j_flags |= JBD2_LOADED; | |
470decc6 DK |
1693 | return 0; |
1694 | ||
1695 | recovery_error: | |
f2a44523 | 1696 | printk(KERN_WARNING "JBD2: recovery failed\n"); |
470decc6 DK |
1697 | return -EIO; |
1698 | } | |
1699 | ||
1700 | /** | |
f7f4bccb | 1701 | * void jbd2_journal_destroy() - Release a journal_t structure. |
470decc6 DK |
1702 | * @journal: Journal to act on. |
1703 | * | |
1704 | * Release a journal_t structure once it is no longer in use by the | |
1705 | * journaled object. | |
44519faf | 1706 | * Return <0 if we couldn't clean up the journal. |
470decc6 | 1707 | */ |
44519faf | 1708 | int jbd2_journal_destroy(journal_t *journal) |
470decc6 | 1709 | { |
44519faf HK |
1710 | int err = 0; |
1711 | ||
470decc6 DK |
1712 | /* Wait for the commit thread to wake up and die. */ |
1713 | journal_kill_thread(journal); | |
1714 | ||
1715 | /* Force a final log commit */ | |
1716 | if (journal->j_running_transaction) | |
f7f4bccb | 1717 | jbd2_journal_commit_transaction(journal); |
470decc6 DK |
1718 | |
1719 | /* Force any old transactions to disk */ | |
1720 | ||
1721 | /* Totally anal locking here... */ | |
1722 | spin_lock(&journal->j_list_lock); | |
1723 | while (journal->j_checkpoint_transactions != NULL) { | |
1724 | spin_unlock(&journal->j_list_lock); | |
1a0d3786 | 1725 | mutex_lock(&journal->j_checkpoint_mutex); |
841df7df | 1726 | err = jbd2_log_do_checkpoint(journal); |
1a0d3786 | 1727 | mutex_unlock(&journal->j_checkpoint_mutex); |
841df7df JK |
1728 | /* |
1729 | * If checkpointing failed, just free the buffers to avoid | |
1730 | * looping forever | |
1731 | */ | |
1732 | if (err) { | |
1733 | jbd2_journal_destroy_checkpoint(journal); | |
1734 | spin_lock(&journal->j_list_lock); | |
1735 | break; | |
1736 | } | |
470decc6 DK |
1737 | spin_lock(&journal->j_list_lock); |
1738 | } | |
1739 | ||
1740 | J_ASSERT(journal->j_running_transaction == NULL); | |
1741 | J_ASSERT(journal->j_committing_transaction == NULL); | |
1742 | J_ASSERT(journal->j_checkpoint_transactions == NULL); | |
1743 | spin_unlock(&journal->j_list_lock); | |
1744 | ||
470decc6 | 1745 | if (journal->j_sb_buffer) { |
44519faf | 1746 | if (!is_journal_aborted(journal)) { |
a78bb11d | 1747 | mutex_lock(&journal->j_checkpoint_mutex); |
c0a2ad9b OH |
1748 | |
1749 | write_lock(&journal->j_state_lock); | |
1750 | journal->j_tail_sequence = | |
1751 | ++journal->j_transaction_sequence; | |
1752 | write_unlock(&journal->j_state_lock); | |
1753 | ||
1754 | jbd2_mark_journal_empty(journal, WRITE_FLUSH_FUA); | |
a78bb11d JK |
1755 | mutex_unlock(&journal->j_checkpoint_mutex); |
1756 | } else | |
44519faf | 1757 | err = -EIO; |
470decc6 DK |
1758 | brelse(journal->j_sb_buffer); |
1759 | } | |
1760 | ||
8e85fb3f JL |
1761 | if (journal->j_proc_entry) |
1762 | jbd2_stats_proc_exit(journal); | |
d9f39d1e | 1763 | iput(journal->j_inode); |
470decc6 | 1764 | if (journal->j_revoke) |
f7f4bccb | 1765 | jbd2_journal_destroy_revoke(journal); |
01b5adce DW |
1766 | if (journal->j_chksum_driver) |
1767 | crypto_free_shash(journal->j_chksum_driver); | |
470decc6 DK |
1768 | kfree(journal->j_wbuf); |
1769 | kfree(journal); | |
44519faf HK |
1770 | |
1771 | return err; | |
470decc6 DK |
1772 | } |
1773 | ||
1774 | ||
1775 | /** | |
f7f4bccb | 1776 | *int jbd2_journal_check_used_features () - Check if features specified are used. |
470decc6 DK |
1777 | * @journal: Journal to check. |
1778 | * @compat: bitmask of compatible features | |
1779 | * @ro: bitmask of features that force read-only mount | |
1780 | * @incompat: bitmask of incompatible features | |
1781 | * | |
1782 | * Check whether the journal uses all of a given set of | |
1783 | * features. Return true (non-zero) if it does. | |
1784 | **/ | |
1785 | ||
f7f4bccb | 1786 | int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1787 | unsigned long ro, unsigned long incompat) |
1788 | { | |
1789 | journal_superblock_t *sb; | |
1790 | ||
1791 | if (!compat && !ro && !incompat) | |
1792 | return 1; | |
1113e1b5 PL |
1793 | /* Load journal superblock if it is not loaded yet. */ |
1794 | if (journal->j_format_version == 0 && | |
1795 | journal_get_superblock(journal) != 0) | |
1796 | return 0; | |
470decc6 DK |
1797 | if (journal->j_format_version == 1) |
1798 | return 0; | |
1799 | ||
1800 | sb = journal->j_superblock; | |
1801 | ||
1802 | if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && | |
1803 | ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && | |
1804 | ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) | |
1805 | return 1; | |
1806 | ||
1807 | return 0; | |
1808 | } | |
1809 | ||
1810 | /** | |
f7f4bccb | 1811 | * int jbd2_journal_check_available_features() - Check feature set in journalling layer |
470decc6 DK |
1812 | * @journal: Journal to check. |
1813 | * @compat: bitmask of compatible features | |
1814 | * @ro: bitmask of features that force read-only mount | |
1815 | * @incompat: bitmask of incompatible features | |
1816 | * | |
1817 | * Check whether the journaling code supports the use of | |
1818 | * all of a given set of features on this journal. Return true | |
1819 | * (non-zero) if it can. */ | |
1820 | ||
f7f4bccb | 1821 | int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1822 | unsigned long ro, unsigned long incompat) |
1823 | { | |
470decc6 DK |
1824 | if (!compat && !ro && !incompat) |
1825 | return 1; | |
1826 | ||
470decc6 DK |
1827 | /* We can support any known requested features iff the |
1828 | * superblock is in version 2. Otherwise we fail to support any | |
1829 | * extended sb features. */ | |
1830 | ||
1831 | if (journal->j_format_version != 2) | |
1832 | return 0; | |
1833 | ||
f7f4bccb MC |
1834 | if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && |
1835 | (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && | |
1836 | (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) | |
470decc6 DK |
1837 | return 1; |
1838 | ||
1839 | return 0; | |
1840 | } | |
1841 | ||
1842 | /** | |
f7f4bccb | 1843 | * int jbd2_journal_set_features () - Mark a given journal feature in the superblock |
470decc6 DK |
1844 | * @journal: Journal to act on. |
1845 | * @compat: bitmask of compatible features | |
1846 | * @ro: bitmask of features that force read-only mount | |
1847 | * @incompat: bitmask of incompatible features | |
1848 | * | |
1849 | * Mark a given journal feature as present on the | |
1850 | * superblock. Returns true if the requested features could be set. | |
1851 | * | |
1852 | */ | |
1853 | ||
f7f4bccb | 1854 | int jbd2_journal_set_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1855 | unsigned long ro, unsigned long incompat) |
1856 | { | |
25ed6e8a DW |
1857 | #define INCOMPAT_FEATURE_ON(f) \ |
1858 | ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f))) | |
1859 | #define COMPAT_FEATURE_ON(f) \ | |
1860 | ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f))) | |
470decc6 DK |
1861 | journal_superblock_t *sb; |
1862 | ||
f7f4bccb | 1863 | if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) |
470decc6 DK |
1864 | return 1; |
1865 | ||
f7f4bccb | 1866 | if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) |
470decc6 DK |
1867 | return 0; |
1868 | ||
db9ee220 DW |
1869 | /* If enabling v2 checksums, turn on v3 instead */ |
1870 | if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) { | |
1871 | incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2; | |
1872 | incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3; | |
1873 | } | |
1874 | ||
1875 | /* Asking for checksumming v3 and v1? Only give them v3. */ | |
1876 | if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 && | |
25ed6e8a DW |
1877 | compat & JBD2_FEATURE_COMPAT_CHECKSUM) |
1878 | compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM; | |
1879 | ||
470decc6 DK |
1880 | jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", |
1881 | compat, ro, incompat); | |
1882 | ||
1883 | sb = journal->j_superblock; | |
1884 | ||
db9ee220 DW |
1885 | /* If enabling v3 checksums, update superblock */ |
1886 | if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { | |
25ed6e8a DW |
1887 | sb->s_checksum_type = JBD2_CRC32C_CHKSUM; |
1888 | sb->s_feature_compat &= | |
1889 | ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM); | |
01b5adce DW |
1890 | |
1891 | /* Load the checksum driver */ | |
1892 | if (journal->j_chksum_driver == NULL) { | |
1893 | journal->j_chksum_driver = crypto_alloc_shash("crc32c", | |
1894 | 0, 0); | |
1895 | if (IS_ERR(journal->j_chksum_driver)) { | |
a67c848a | 1896 | printk(KERN_ERR "JBD2: Cannot load crc32c " |
01b5adce DW |
1897 | "driver.\n"); |
1898 | journal->j_chksum_driver = NULL; | |
1899 | return 0; | |
1900 | } | |
4fd5ea43 | 1901 | |
32f38691 | 1902 | /* Precompute checksum seed for all metadata */ |
4fd5ea43 DW |
1903 | journal->j_csum_seed = jbd2_chksum(journal, ~0, |
1904 | sb->s_uuid, | |
1905 | sizeof(sb->s_uuid)); | |
32f38691 | 1906 | } |
25ed6e8a DW |
1907 | } |
1908 | ||
1909 | /* If enabling v1 checksums, downgrade superblock */ | |
1910 | if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM)) | |
1911 | sb->s_feature_incompat &= | |
db9ee220 DW |
1912 | ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 | |
1913 | JBD2_FEATURE_INCOMPAT_CSUM_V3); | |
25ed6e8a | 1914 | |
470decc6 DK |
1915 | sb->s_feature_compat |= cpu_to_be32(compat); |
1916 | sb->s_feature_ro_compat |= cpu_to_be32(ro); | |
1917 | sb->s_feature_incompat |= cpu_to_be32(incompat); | |
1918 | ||
1919 | return 1; | |
25ed6e8a DW |
1920 | #undef COMPAT_FEATURE_ON |
1921 | #undef INCOMPAT_FEATURE_ON | |
470decc6 DK |
1922 | } |
1923 | ||
818d276c GS |
1924 | /* |
1925 | * jbd2_journal_clear_features () - Clear a given journal feature in the | |
1926 | * superblock | |
1927 | * @journal: Journal to act on. | |
1928 | * @compat: bitmask of compatible features | |
1929 | * @ro: bitmask of features that force read-only mount | |
1930 | * @incompat: bitmask of incompatible features | |
1931 | * | |
1932 | * Clear a given journal feature as present on the | |
1933 | * superblock. | |
1934 | */ | |
1935 | void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, | |
1936 | unsigned long ro, unsigned long incompat) | |
1937 | { | |
1938 | journal_superblock_t *sb; | |
1939 | ||
1940 | jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", | |
1941 | compat, ro, incompat); | |
1942 | ||
1943 | sb = journal->j_superblock; | |
1944 | ||
1945 | sb->s_feature_compat &= ~cpu_to_be32(compat); | |
1946 | sb->s_feature_ro_compat &= ~cpu_to_be32(ro); | |
1947 | sb->s_feature_incompat &= ~cpu_to_be32(incompat); | |
1948 | } | |
1949 | EXPORT_SYMBOL(jbd2_journal_clear_features); | |
470decc6 | 1950 | |
470decc6 | 1951 | /** |
f7f4bccb | 1952 | * int jbd2_journal_flush () - Flush journal |
470decc6 DK |
1953 | * @journal: Journal to act on. |
1954 | * | |
1955 | * Flush all data for a given journal to disk and empty the journal. | |
1956 | * Filesystems can use this when remounting readonly to ensure that | |
1957 | * recovery does not need to happen on remount. | |
1958 | */ | |
1959 | ||
f7f4bccb | 1960 | int jbd2_journal_flush(journal_t *journal) |
470decc6 DK |
1961 | { |
1962 | int err = 0; | |
1963 | transaction_t *transaction = NULL; | |
470decc6 | 1964 | |
a931da6a | 1965 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1966 | |
1967 | /* Force everything buffered to the log... */ | |
1968 | if (journal->j_running_transaction) { | |
1969 | transaction = journal->j_running_transaction; | |
f7f4bccb | 1970 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1971 | } else if (journal->j_committing_transaction) |
1972 | transaction = journal->j_committing_transaction; | |
1973 | ||
1974 | /* Wait for the log commit to complete... */ | |
1975 | if (transaction) { | |
1976 | tid_t tid = transaction->t_tid; | |
1977 | ||
a931da6a | 1978 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 1979 | jbd2_log_wait_commit(journal, tid); |
470decc6 | 1980 | } else { |
a931da6a | 1981 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1982 | } |
1983 | ||
1984 | /* ...and flush everything in the log out to disk. */ | |
1985 | spin_lock(&journal->j_list_lock); | |
1986 | while (!err && journal->j_checkpoint_transactions != NULL) { | |
1987 | spin_unlock(&journal->j_list_lock); | |
44519faf | 1988 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1989 | err = jbd2_log_do_checkpoint(journal); |
44519faf | 1990 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
1991 | spin_lock(&journal->j_list_lock); |
1992 | } | |
1993 | spin_unlock(&journal->j_list_lock); | |
44519faf HK |
1994 | |
1995 | if (is_journal_aborted(journal)) | |
1996 | return -EIO; | |
1997 | ||
a78bb11d | 1998 | mutex_lock(&journal->j_checkpoint_mutex); |
6f6a6fda JQ |
1999 | if (!err) { |
2000 | err = jbd2_cleanup_journal_tail(journal); | |
2001 | if (err < 0) { | |
2002 | mutex_unlock(&journal->j_checkpoint_mutex); | |
2003 | goto out; | |
2004 | } | |
2005 | err = 0; | |
2006 | } | |
470decc6 DK |
2007 | |
2008 | /* Finally, mark the journal as really needing no recovery. | |
2009 | * This sets s_start==0 in the underlying superblock, which is | |
2010 | * the magic code for a fully-recovered superblock. Any future | |
2011 | * commits of data to the journal will restore the current | |
2012 | * s_start value. */ | |
c0a2ad9b | 2013 | jbd2_mark_journal_empty(journal, WRITE_FUA); |
a78bb11d | 2014 | mutex_unlock(&journal->j_checkpoint_mutex); |
a931da6a | 2015 | write_lock(&journal->j_state_lock); |
470decc6 DK |
2016 | J_ASSERT(!journal->j_running_transaction); |
2017 | J_ASSERT(!journal->j_committing_transaction); | |
2018 | J_ASSERT(!journal->j_checkpoint_transactions); | |
2019 | J_ASSERT(journal->j_head == journal->j_tail); | |
2020 | J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); | |
a931da6a | 2021 | write_unlock(&journal->j_state_lock); |
6f6a6fda JQ |
2022 | out: |
2023 | return err; | |
470decc6 DK |
2024 | } |
2025 | ||
2026 | /** | |
f7f4bccb | 2027 | * int jbd2_journal_wipe() - Wipe journal contents |
470decc6 DK |
2028 | * @journal: Journal to act on. |
2029 | * @write: flag (see below) | |
2030 | * | |
2031 | * Wipe out all of the contents of a journal, safely. This will produce | |
2032 | * a warning if the journal contains any valid recovery information. | |
f7f4bccb | 2033 | * Must be called between journal_init_*() and jbd2_journal_load(). |
470decc6 DK |
2034 | * |
2035 | * If 'write' is non-zero, then we wipe out the journal on disk; otherwise | |
2036 | * we merely suppress recovery. | |
2037 | */ | |
2038 | ||
f7f4bccb | 2039 | int jbd2_journal_wipe(journal_t *journal, int write) |
470decc6 | 2040 | { |
470decc6 DK |
2041 | int err = 0; |
2042 | ||
f7f4bccb | 2043 | J_ASSERT (!(journal->j_flags & JBD2_LOADED)); |
470decc6 DK |
2044 | |
2045 | err = load_superblock(journal); | |
2046 | if (err) | |
2047 | return err; | |
2048 | ||
470decc6 DK |
2049 | if (!journal->j_tail) |
2050 | goto no_recovery; | |
2051 | ||
f2a44523 | 2052 | printk(KERN_WARNING "JBD2: %s recovery information on journal\n", |
470decc6 DK |
2053 | write ? "Clearing" : "Ignoring"); |
2054 | ||
f7f4bccb | 2055 | err = jbd2_journal_skip_recovery(journal); |
a78bb11d JK |
2056 | if (write) { |
2057 | /* Lock to make assertions happy... */ | |
2058 | mutex_lock(&journal->j_checkpoint_mutex); | |
c0a2ad9b | 2059 | jbd2_mark_journal_empty(journal, WRITE_FUA); |
a78bb11d JK |
2060 | mutex_unlock(&journal->j_checkpoint_mutex); |
2061 | } | |
470decc6 DK |
2062 | |
2063 | no_recovery: | |
2064 | return err; | |
2065 | } | |
2066 | ||
470decc6 DK |
2067 | /* |
2068 | * Journal abort has very specific semantics, which we describe | |
2069 | * for journal abort. | |
2070 | * | |
bfcd3555 | 2071 | * Two internal functions, which provide abort to the jbd layer |
470decc6 DK |
2072 | * itself are here. |
2073 | */ | |
2074 | ||
2075 | /* | |
2076 | * Quick version for internal journal use (doesn't lock the journal). | |
2077 | * Aborts hard --- we mark the abort as occurred, but do _nothing_ else, | |
2078 | * and don't attempt to make any other journal updates. | |
2079 | */ | |
f7f4bccb | 2080 | void __jbd2_journal_abort_hard(journal_t *journal) |
470decc6 DK |
2081 | { |
2082 | transaction_t *transaction; | |
470decc6 | 2083 | |
f7f4bccb | 2084 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2085 | return; |
2086 | ||
2087 | printk(KERN_ERR "Aborting journal on device %s.\n", | |
05496769 | 2088 | journal->j_devname); |
470decc6 | 2089 | |
a931da6a | 2090 | write_lock(&journal->j_state_lock); |
f7f4bccb | 2091 | journal->j_flags |= JBD2_ABORT; |
470decc6 DK |
2092 | transaction = journal->j_running_transaction; |
2093 | if (transaction) | |
f7f4bccb | 2094 | __jbd2_log_start_commit(journal, transaction->t_tid); |
a931da6a | 2095 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2096 | } |
2097 | ||
2098 | /* Soft abort: record the abort error status in the journal superblock, | |
2099 | * but don't do any other IO. */ | |
2100 | static void __journal_abort_soft (journal_t *journal, int errno) | |
2101 | { | |
f7f4bccb | 2102 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2103 | return; |
2104 | ||
2105 | if (!journal->j_errno) | |
2106 | journal->j_errno = errno; | |
2107 | ||
f7f4bccb | 2108 | __jbd2_journal_abort_hard(journal); |
470decc6 | 2109 | |
4327ba52 | 2110 | if (errno) { |
24bcc89c | 2111 | jbd2_journal_update_sb_errno(journal); |
4327ba52 DJ |
2112 | write_lock(&journal->j_state_lock); |
2113 | journal->j_flags |= JBD2_REC_ERR; | |
2114 | write_unlock(&journal->j_state_lock); | |
2115 | } | |
470decc6 DK |
2116 | } |
2117 | ||
2118 | /** | |
f7f4bccb | 2119 | * void jbd2_journal_abort () - Shutdown the journal immediately. |
470decc6 DK |
2120 | * @journal: the journal to shutdown. |
2121 | * @errno: an error number to record in the journal indicating | |
2122 | * the reason for the shutdown. | |
2123 | * | |
2124 | * Perform a complete, immediate shutdown of the ENTIRE | |
2125 | * journal (not of a single transaction). This operation cannot be | |
2126 | * undone without closing and reopening the journal. | |
2127 | * | |
f7f4bccb | 2128 | * The jbd2_journal_abort function is intended to support higher level error |
470decc6 DK |
2129 | * recovery mechanisms such as the ext2/ext3 remount-readonly error |
2130 | * mode. | |
2131 | * | |
2132 | * Journal abort has very specific semantics. Any existing dirty, | |
2133 | * unjournaled buffers in the main filesystem will still be written to | |
2134 | * disk by bdflush, but the journaling mechanism will be suspended | |
2135 | * immediately and no further transaction commits will be honoured. | |
2136 | * | |
2137 | * Any dirty, journaled buffers will be written back to disk without | |
2138 | * hitting the journal. Atomicity cannot be guaranteed on an aborted | |
2139 | * filesystem, but we _do_ attempt to leave as much data as possible | |
2140 | * behind for fsck to use for cleanup. | |
2141 | * | |
2142 | * Any attempt to get a new transaction handle on a journal which is in | |
2143 | * ABORT state will just result in an -EROFS error return. A | |
f7f4bccb | 2144 | * jbd2_journal_stop on an existing handle will return -EIO if we have |
470decc6 DK |
2145 | * entered abort state during the update. |
2146 | * | |
2147 | * Recursive transactions are not disturbed by journal abort until the | |
f7f4bccb | 2148 | * final jbd2_journal_stop, which will receive the -EIO error. |
470decc6 | 2149 | * |
f7f4bccb | 2150 | * Finally, the jbd2_journal_abort call allows the caller to supply an errno |
470decc6 DK |
2151 | * which will be recorded (if possible) in the journal superblock. This |
2152 | * allows a client to record failure conditions in the middle of a | |
2153 | * transaction without having to complete the transaction to record the | |
2154 | * failure to disk. ext3_error, for example, now uses this | |
2155 | * functionality. | |
2156 | * | |
2157 | * Errors which originate from within the journaling layer will NOT | |
2158 | * supply an errno; a null errno implies that absolutely no further | |
2159 | * writes are done to the journal (unless there are any already in | |
2160 | * progress). | |
2161 | * | |
2162 | */ | |
2163 | ||
f7f4bccb | 2164 | void jbd2_journal_abort(journal_t *journal, int errno) |
470decc6 DK |
2165 | { |
2166 | __journal_abort_soft(journal, errno); | |
2167 | } | |
2168 | ||
2169 | /** | |
f7f4bccb | 2170 | * int jbd2_journal_errno () - returns the journal's error state. |
470decc6 DK |
2171 | * @journal: journal to examine. |
2172 | * | |
bfcd3555 | 2173 | * This is the errno number set with jbd2_journal_abort(), the last |
470decc6 DK |
2174 | * time the journal was mounted - if the journal was stopped |
2175 | * without calling abort this will be 0. | |
2176 | * | |
2177 | * If the journal has been aborted on this mount time -EROFS will | |
2178 | * be returned. | |
2179 | */ | |
f7f4bccb | 2180 | int jbd2_journal_errno(journal_t *journal) |
470decc6 DK |
2181 | { |
2182 | int err; | |
2183 | ||
a931da6a | 2184 | read_lock(&journal->j_state_lock); |
f7f4bccb | 2185 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2186 | err = -EROFS; |
2187 | else | |
2188 | err = journal->j_errno; | |
a931da6a | 2189 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
2190 | return err; |
2191 | } | |
2192 | ||
2193 | /** | |
f7f4bccb | 2194 | * int jbd2_journal_clear_err () - clears the journal's error state |
470decc6 DK |
2195 | * @journal: journal to act on. |
2196 | * | |
bfcd3555 | 2197 | * An error must be cleared or acked to take a FS out of readonly |
470decc6 DK |
2198 | * mode. |
2199 | */ | |
f7f4bccb | 2200 | int jbd2_journal_clear_err(journal_t *journal) |
470decc6 DK |
2201 | { |
2202 | int err = 0; | |
2203 | ||
a931da6a | 2204 | write_lock(&journal->j_state_lock); |
f7f4bccb | 2205 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2206 | err = -EROFS; |
2207 | else | |
2208 | journal->j_errno = 0; | |
a931da6a | 2209 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2210 | return err; |
2211 | } | |
2212 | ||
2213 | /** | |
f7f4bccb | 2214 | * void jbd2_journal_ack_err() - Ack journal err. |
470decc6 DK |
2215 | * @journal: journal to act on. |
2216 | * | |
bfcd3555 | 2217 | * An error must be cleared or acked to take a FS out of readonly |
470decc6 DK |
2218 | * mode. |
2219 | */ | |
f7f4bccb | 2220 | void jbd2_journal_ack_err(journal_t *journal) |
470decc6 | 2221 | { |
a931da6a | 2222 | write_lock(&journal->j_state_lock); |
470decc6 | 2223 | if (journal->j_errno) |
f7f4bccb | 2224 | journal->j_flags |= JBD2_ACK_ERR; |
a931da6a | 2225 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2226 | } |
2227 | ||
f7f4bccb | 2228 | int jbd2_journal_blocks_per_page(struct inode *inode) |
470decc6 | 2229 | { |
09cbfeaf | 2230 | return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
470decc6 DK |
2231 | } |
2232 | ||
b517bea1 ZB |
2233 | /* |
2234 | * helper functions to deal with 32 or 64bit block numbers. | |
2235 | */ | |
2236 | size_t journal_tag_bytes(journal_t *journal) | |
2237 | { | |
db9ee220 DW |
2238 | size_t sz; |
2239 | ||
56316a0d | 2240 | if (jbd2_has_feature_csum3(journal)) |
db9ee220 DW |
2241 | return sizeof(journal_block_tag3_t); |
2242 | ||
2243 | sz = sizeof(journal_block_tag_t); | |
c3900875 | 2244 | |
56316a0d | 2245 | if (jbd2_has_feature_csum2(journal)) |
db9ee220 | 2246 | sz += sizeof(__u16); |
c3900875 | 2247 | |
56316a0d | 2248 | if (jbd2_has_feature_64bit(journal)) |
db9ee220 | 2249 | return sz; |
b517bea1 | 2250 | else |
db9ee220 | 2251 | return sz - sizeof(__u32); |
b517bea1 ZB |
2252 | } |
2253 | ||
d2eecb03 TT |
2254 | /* |
2255 | * JBD memory management | |
2256 | * | |
2257 | * These functions are used to allocate block-sized chunks of memory | |
2258 | * used for making copies of buffer_head data. Very often it will be | |
2259 | * page-sized chunks of data, but sometimes it will be in | |
2260 | * sub-page-size chunks. (For example, 16k pages on Power systems | |
2261 | * with a 4k block file system.) For blocks smaller than a page, we | |
2262 | * use a SLAB allocator. There are slab caches for each block size, | |
2263 | * which are allocated at mount time, if necessary, and we only free | |
2264 | * (all of) the slab caches when/if the jbd2 module is unloaded. For | |
2265 | * this reason we don't need to a mutex to protect access to | |
2266 | * jbd2_slab[] allocating or releasing memory; only in | |
2267 | * jbd2_journal_create_slab(). | |
2268 | */ | |
2269 | #define JBD2_MAX_SLABS 8 | |
2270 | static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS]; | |
d2eecb03 TT |
2271 | |
2272 | static const char *jbd2_slab_names[JBD2_MAX_SLABS] = { | |
2273 | "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k", | |
2274 | "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k" | |
2275 | }; | |
2276 | ||
2277 | ||
2278 | static void jbd2_journal_destroy_slabs(void) | |
2279 | { | |
2280 | int i; | |
2281 | ||
2282 | for (i = 0; i < JBD2_MAX_SLABS; i++) { | |
2283 | if (jbd2_slab[i]) | |
2284 | kmem_cache_destroy(jbd2_slab[i]); | |
2285 | jbd2_slab[i] = NULL; | |
2286 | } | |
2287 | } | |
2288 | ||
2289 | static int jbd2_journal_create_slab(size_t size) | |
2290 | { | |
51dfacde | 2291 | static DEFINE_MUTEX(jbd2_slab_create_mutex); |
d2eecb03 TT |
2292 | int i = order_base_2(size) - 10; |
2293 | size_t slab_size; | |
2294 | ||
2295 | if (size == PAGE_SIZE) | |
2296 | return 0; | |
2297 | ||
2298 | if (i >= JBD2_MAX_SLABS) | |
2299 | return -EINVAL; | |
2300 | ||
2301 | if (unlikely(i < 0)) | |
2302 | i = 0; | |
51dfacde | 2303 | mutex_lock(&jbd2_slab_create_mutex); |
d2eecb03 | 2304 | if (jbd2_slab[i]) { |
51dfacde | 2305 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2306 | return 0; /* Already created */ |
2307 | } | |
2308 | ||
2309 | slab_size = 1 << (i+10); | |
2310 | jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size, | |
2311 | slab_size, 0, NULL); | |
51dfacde | 2312 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2313 | if (!jbd2_slab[i]) { |
2314 | printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n"); | |
2315 | return -ENOMEM; | |
2316 | } | |
2317 | return 0; | |
2318 | } | |
2319 | ||
2320 | static struct kmem_cache *get_slab(size_t size) | |
2321 | { | |
2322 | int i = order_base_2(size) - 10; | |
2323 | ||
2324 | BUG_ON(i >= JBD2_MAX_SLABS); | |
2325 | if (unlikely(i < 0)) | |
2326 | i = 0; | |
8ac97b74 | 2327 | BUG_ON(jbd2_slab[i] == NULL); |
d2eecb03 TT |
2328 | return jbd2_slab[i]; |
2329 | } | |
2330 | ||
2331 | void *jbd2_alloc(size_t size, gfp_t flags) | |
2332 | { | |
2333 | void *ptr; | |
2334 | ||
2335 | BUG_ON(size & (size-1)); /* Must be a power of 2 */ | |
2336 | ||
f2db1971 | 2337 | if (size < PAGE_SIZE) |
d2eecb03 | 2338 | ptr = kmem_cache_alloc(get_slab(size), flags); |
f2db1971 MH |
2339 | else |
2340 | ptr = (void *)__get_free_pages(flags, get_order(size)); | |
d2eecb03 TT |
2341 | |
2342 | /* Check alignment; SLUB has gotten this wrong in the past, | |
2343 | * and this can lead to user data corruption! */ | |
2344 | BUG_ON(((unsigned long) ptr) & (size-1)); | |
2345 | ||
2346 | return ptr; | |
2347 | } | |
2348 | ||
2349 | void jbd2_free(void *ptr, size_t size) | |
2350 | { | |
f2db1971 MH |
2351 | if (size < PAGE_SIZE) |
2352 | kmem_cache_free(get_slab(size), ptr); | |
2353 | else | |
2354 | free_pages((unsigned long)ptr, get_order(size)); | |
d2eecb03 TT |
2355 | }; |
2356 | ||
470decc6 DK |
2357 | /* |
2358 | * Journal_head storage management | |
2359 | */ | |
e18b890b | 2360 | static struct kmem_cache *jbd2_journal_head_cache; |
e23291b9 | 2361 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2362 | static atomic_t nr_journal_heads = ATOMIC_INIT(0); |
2363 | #endif | |
2364 | ||
4185a2ac | 2365 | static int jbd2_journal_init_journal_head_cache(void) |
470decc6 DK |
2366 | { |
2367 | int retval; | |
2368 | ||
1076d17a | 2369 | J_ASSERT(jbd2_journal_head_cache == NULL); |
a920e941 | 2370 | jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", |
470decc6 DK |
2371 | sizeof(struct journal_head), |
2372 | 0, /* offset */ | |
de92c8ca | 2373 | SLAB_TEMPORARY | SLAB_DESTROY_BY_RCU, |
20c2df83 | 2374 | NULL); /* ctor */ |
470decc6 | 2375 | retval = 0; |
1076d17a | 2376 | if (!jbd2_journal_head_cache) { |
470decc6 | 2377 | retval = -ENOMEM; |
f2a44523 | 2378 | printk(KERN_EMERG "JBD2: no memory for journal_head cache\n"); |
470decc6 DK |
2379 | } |
2380 | return retval; | |
2381 | } | |
2382 | ||
4185a2ac | 2383 | static void jbd2_journal_destroy_journal_head_cache(void) |
470decc6 | 2384 | { |
8a9362eb DG |
2385 | if (jbd2_journal_head_cache) { |
2386 | kmem_cache_destroy(jbd2_journal_head_cache); | |
2387 | jbd2_journal_head_cache = NULL; | |
2388 | } | |
470decc6 DK |
2389 | } |
2390 | ||
2391 | /* | |
2392 | * journal_head splicing and dicing | |
2393 | */ | |
2394 | static struct journal_head *journal_alloc_journal_head(void) | |
2395 | { | |
2396 | struct journal_head *ret; | |
470decc6 | 2397 | |
e23291b9 | 2398 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2399 | atomic_inc(&nr_journal_heads); |
2400 | #endif | |
5d9cf9c6 | 2401 | ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); |
1076d17a | 2402 | if (!ret) { |
470decc6 | 2403 | jbd_debug(1, "out of memory for journal_head\n"); |
670be5a7 | 2404 | pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__); |
7b506b10 MH |
2405 | ret = kmem_cache_zalloc(jbd2_journal_head_cache, |
2406 | GFP_NOFS | __GFP_NOFAIL); | |
470decc6 DK |
2407 | } |
2408 | return ret; | |
2409 | } | |
2410 | ||
2411 | static void journal_free_journal_head(struct journal_head *jh) | |
2412 | { | |
e23291b9 | 2413 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 | 2414 | atomic_dec(&nr_journal_heads); |
cd02ff0b | 2415 | memset(jh, JBD2_POISON_FREE, sizeof(*jh)); |
470decc6 | 2416 | #endif |
f7f4bccb | 2417 | kmem_cache_free(jbd2_journal_head_cache, jh); |
470decc6 DK |
2418 | } |
2419 | ||
2420 | /* | |
2421 | * A journal_head is attached to a buffer_head whenever JBD has an | |
2422 | * interest in the buffer. | |
2423 | * | |
2424 | * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit | |
2425 | * is set. This bit is tested in core kernel code where we need to take | |
2426 | * JBD-specific actions. Testing the zeroness of ->b_private is not reliable | |
2427 | * there. | |
2428 | * | |
2429 | * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. | |
2430 | * | |
2431 | * When a buffer has its BH_JBD bit set it is immune from being released by | |
2432 | * core kernel code, mainly via ->b_count. | |
2433 | * | |
de1b7941 JK |
2434 | * A journal_head is detached from its buffer_head when the journal_head's |
2435 | * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint | |
2436 | * transaction (b_cp_transaction) hold their references to b_jcount. | |
470decc6 DK |
2437 | * |
2438 | * Various places in the kernel want to attach a journal_head to a buffer_head | |
2439 | * _before_ attaching the journal_head to a transaction. To protect the | |
f7f4bccb | 2440 | * journal_head in this situation, jbd2_journal_add_journal_head elevates the |
470decc6 | 2441 | * journal_head's b_jcount refcount by one. The caller must call |
f7f4bccb | 2442 | * jbd2_journal_put_journal_head() to undo this. |
470decc6 DK |
2443 | * |
2444 | * So the typical usage would be: | |
2445 | * | |
2446 | * (Attach a journal_head if needed. Increments b_jcount) | |
f7f4bccb | 2447 | * struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 | 2448 | * ... |
de1b7941 JK |
2449 | * (Get another reference for transaction) |
2450 | * jbd2_journal_grab_journal_head(bh); | |
470decc6 | 2451 | * jh->b_transaction = xxx; |
de1b7941 | 2452 | * (Put original reference) |
f7f4bccb | 2453 | * jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2454 | */ |
2455 | ||
2456 | /* | |
2457 | * Give a buffer_head a journal_head. | |
2458 | * | |
470decc6 DK |
2459 | * May sleep. |
2460 | */ | |
f7f4bccb | 2461 | struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) |
470decc6 DK |
2462 | { |
2463 | struct journal_head *jh; | |
2464 | struct journal_head *new_jh = NULL; | |
2465 | ||
2466 | repeat: | |
5d9cf9c6 | 2467 | if (!buffer_jbd(bh)) |
470decc6 | 2468 | new_jh = journal_alloc_journal_head(); |
470decc6 DK |
2469 | |
2470 | jbd_lock_bh_journal_head(bh); | |
2471 | if (buffer_jbd(bh)) { | |
2472 | jh = bh2jh(bh); | |
2473 | } else { | |
2474 | J_ASSERT_BH(bh, | |
2475 | (atomic_read(&bh->b_count) > 0) || | |
2476 | (bh->b_page && bh->b_page->mapping)); | |
2477 | ||
2478 | if (!new_jh) { | |
2479 | jbd_unlock_bh_journal_head(bh); | |
2480 | goto repeat; | |
2481 | } | |
2482 | ||
2483 | jh = new_jh; | |
2484 | new_jh = NULL; /* We consumed it */ | |
2485 | set_buffer_jbd(bh); | |
2486 | bh->b_private = jh; | |
2487 | jh->b_bh = bh; | |
2488 | get_bh(bh); | |
2489 | BUFFER_TRACE(bh, "added journal_head"); | |
2490 | } | |
2491 | jh->b_jcount++; | |
2492 | jbd_unlock_bh_journal_head(bh); | |
2493 | if (new_jh) | |
2494 | journal_free_journal_head(new_jh); | |
2495 | return bh->b_private; | |
2496 | } | |
2497 | ||
2498 | /* | |
2499 | * Grab a ref against this buffer_head's journal_head. If it ended up not | |
2500 | * having a journal_head, return NULL | |
2501 | */ | |
f7f4bccb | 2502 | struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) |
470decc6 DK |
2503 | { |
2504 | struct journal_head *jh = NULL; | |
2505 | ||
2506 | jbd_lock_bh_journal_head(bh); | |
2507 | if (buffer_jbd(bh)) { | |
2508 | jh = bh2jh(bh); | |
2509 | jh->b_jcount++; | |
2510 | } | |
2511 | jbd_unlock_bh_journal_head(bh); | |
2512 | return jh; | |
2513 | } | |
2514 | ||
2515 | static void __journal_remove_journal_head(struct buffer_head *bh) | |
2516 | { | |
2517 | struct journal_head *jh = bh2jh(bh); | |
2518 | ||
2519 | J_ASSERT_JH(jh, jh->b_jcount >= 0); | |
de1b7941 JK |
2520 | J_ASSERT_JH(jh, jh->b_transaction == NULL); |
2521 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
2522 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
2523 | J_ASSERT_JH(jh, jh->b_jlist == BJ_None); | |
2524 | J_ASSERT_BH(bh, buffer_jbd(bh)); | |
2525 | J_ASSERT_BH(bh, jh2bh(jh) == bh); | |
2526 | BUFFER_TRACE(bh, "remove journal_head"); | |
2527 | if (jh->b_frozen_data) { | |
2528 | printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__); | |
2529 | jbd2_free(jh->b_frozen_data, bh->b_size); | |
470decc6 | 2530 | } |
de1b7941 JK |
2531 | if (jh->b_committed_data) { |
2532 | printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__); | |
2533 | jbd2_free(jh->b_committed_data, bh->b_size); | |
2534 | } | |
2535 | bh->b_private = NULL; | |
2536 | jh->b_bh = NULL; /* debug, really */ | |
2537 | clear_buffer_jbd(bh); | |
2538 | journal_free_journal_head(jh); | |
470decc6 DK |
2539 | } |
2540 | ||
2541 | /* | |
de1b7941 | 2542 | * Drop a reference on the passed journal_head. If it fell to zero then |
470decc6 DK |
2543 | * release the journal_head from the buffer_head. |
2544 | */ | |
f7f4bccb | 2545 | void jbd2_journal_put_journal_head(struct journal_head *jh) |
470decc6 DK |
2546 | { |
2547 | struct buffer_head *bh = jh2bh(jh); | |
2548 | ||
2549 | jbd_lock_bh_journal_head(bh); | |
2550 | J_ASSERT_JH(jh, jh->b_jcount > 0); | |
2551 | --jh->b_jcount; | |
de1b7941 | 2552 | if (!jh->b_jcount) { |
470decc6 | 2553 | __journal_remove_journal_head(bh); |
de1b7941 | 2554 | jbd_unlock_bh_journal_head(bh); |
470decc6 | 2555 | __brelse(bh); |
de1b7941 JK |
2556 | } else |
2557 | jbd_unlock_bh_journal_head(bh); | |
470decc6 DK |
2558 | } |
2559 | ||
c851ed54 JK |
2560 | /* |
2561 | * Initialize jbd inode head | |
2562 | */ | |
2563 | void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) | |
2564 | { | |
2565 | jinode->i_transaction = NULL; | |
2566 | jinode->i_next_transaction = NULL; | |
2567 | jinode->i_vfs_inode = inode; | |
2568 | jinode->i_flags = 0; | |
2569 | INIT_LIST_HEAD(&jinode->i_list); | |
2570 | } | |
2571 | ||
2572 | /* | |
2573 | * Function to be called before we start removing inode from memory (i.e., | |
2574 | * clear_inode() is a fine place to be called from). It removes inode from | |
2575 | * transaction's lists. | |
2576 | */ | |
2577 | void jbd2_journal_release_jbd_inode(journal_t *journal, | |
2578 | struct jbd2_inode *jinode) | |
2579 | { | |
c851ed54 JK |
2580 | if (!journal) |
2581 | return; | |
2582 | restart: | |
2583 | spin_lock(&journal->j_list_lock); | |
2584 | /* Is commit writing out inode - we have to wait */ | |
cb0d9d47 | 2585 | if (jinode->i_flags & JI_COMMIT_RUNNING) { |
c851ed54 JK |
2586 | wait_queue_head_t *wq; |
2587 | DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); | |
2588 | wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); | |
2589 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
2590 | spin_unlock(&journal->j_list_lock); | |
2591 | schedule(); | |
2592 | finish_wait(wq, &wait.wait); | |
2593 | goto restart; | |
2594 | } | |
2595 | ||
c851ed54 JK |
2596 | if (jinode->i_transaction) { |
2597 | list_del(&jinode->i_list); | |
2598 | jinode->i_transaction = NULL; | |
2599 | } | |
2600 | spin_unlock(&journal->j_list_lock); | |
2601 | } | |
2602 | ||
470decc6 | 2603 | |
8e85fb3f JL |
2604 | #ifdef CONFIG_PROC_FS |
2605 | ||
2606 | #define JBD2_STATS_PROC_NAME "fs/jbd2" | |
2607 | ||
2608 | static void __init jbd2_create_jbd_stats_proc_entry(void) | |
2609 | { | |
2610 | proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); | |
2611 | } | |
2612 | ||
2613 | static void __exit jbd2_remove_jbd_stats_proc_entry(void) | |
2614 | { | |
2615 | if (proc_jbd2_stats) | |
2616 | remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); | |
2617 | } | |
2618 | ||
2619 | #else | |
2620 | ||
2621 | #define jbd2_create_jbd_stats_proc_entry() do {} while (0) | |
2622 | #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) | |
2623 | ||
2624 | #endif | |
2625 | ||
8aefcd55 | 2626 | struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache; |
470decc6 | 2627 | |
4185a2ac | 2628 | static int __init jbd2_journal_init_handle_cache(void) |
470decc6 | 2629 | { |
8aefcd55 | 2630 | jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY); |
f7f4bccb | 2631 | if (jbd2_handle_cache == NULL) { |
8aefcd55 TT |
2632 | printk(KERN_EMERG "JBD2: failed to create handle cache\n"); |
2633 | return -ENOMEM; | |
2634 | } | |
2635 | jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0); | |
2636 | if (jbd2_inode_cache == NULL) { | |
2637 | printk(KERN_EMERG "JBD2: failed to create inode cache\n"); | |
2638 | kmem_cache_destroy(jbd2_handle_cache); | |
470decc6 DK |
2639 | return -ENOMEM; |
2640 | } | |
2641 | return 0; | |
2642 | } | |
2643 | ||
f7f4bccb | 2644 | static void jbd2_journal_destroy_handle_cache(void) |
470decc6 | 2645 | { |
f7f4bccb MC |
2646 | if (jbd2_handle_cache) |
2647 | kmem_cache_destroy(jbd2_handle_cache); | |
8aefcd55 TT |
2648 | if (jbd2_inode_cache) |
2649 | kmem_cache_destroy(jbd2_inode_cache); | |
2650 | ||
470decc6 DK |
2651 | } |
2652 | ||
2653 | /* | |
2654 | * Module startup and shutdown | |
2655 | */ | |
2656 | ||
2657 | static int __init journal_init_caches(void) | |
2658 | { | |
2659 | int ret; | |
2660 | ||
f7f4bccb | 2661 | ret = jbd2_journal_init_revoke_caches(); |
470decc6 | 2662 | if (ret == 0) |
4185a2ac | 2663 | ret = jbd2_journal_init_journal_head_cache(); |
470decc6 | 2664 | if (ret == 0) |
4185a2ac | 2665 | ret = jbd2_journal_init_handle_cache(); |
470decc6 | 2666 | if (ret == 0) |
0c2022ec | 2667 | ret = jbd2_journal_init_transaction_cache(); |
470decc6 DK |
2668 | return ret; |
2669 | } | |
2670 | ||
f7f4bccb | 2671 | static void jbd2_journal_destroy_caches(void) |
470decc6 | 2672 | { |
f7f4bccb | 2673 | jbd2_journal_destroy_revoke_caches(); |
4185a2ac | 2674 | jbd2_journal_destroy_journal_head_cache(); |
f7f4bccb | 2675 | jbd2_journal_destroy_handle_cache(); |
0c2022ec | 2676 | jbd2_journal_destroy_transaction_cache(); |
d2eecb03 | 2677 | jbd2_journal_destroy_slabs(); |
470decc6 DK |
2678 | } |
2679 | ||
2680 | static int __init journal_init(void) | |
2681 | { | |
2682 | int ret; | |
2683 | ||
2684 | BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); | |
2685 | ||
2686 | ret = journal_init_caches(); | |
620de4e1 | 2687 | if (ret == 0) { |
620de4e1 DG |
2688 | jbd2_create_jbd_stats_proc_entry(); |
2689 | } else { | |
f7f4bccb | 2690 | jbd2_journal_destroy_caches(); |
620de4e1 | 2691 | } |
470decc6 DK |
2692 | return ret; |
2693 | } | |
2694 | ||
2695 | static void __exit journal_exit(void) | |
2696 | { | |
e23291b9 | 2697 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2698 | int n = atomic_read(&nr_journal_heads); |
2699 | if (n) | |
75685071 | 2700 | printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n); |
470decc6 | 2701 | #endif |
8e85fb3f | 2702 | jbd2_remove_jbd_stats_proc_entry(); |
f7f4bccb | 2703 | jbd2_journal_destroy_caches(); |
470decc6 DK |
2704 | } |
2705 | ||
2706 | MODULE_LICENSE("GPL"); | |
2707 | module_init(journal_init); | |
2708 | module_exit(journal_exit); | |
2709 |