]>
Commit | Line | Data |
---|---|---|
f5166768 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
470decc6 | 2 | /* |
f7f4bccb | 3 | * linux/fs/jbd2/journal.c |
470decc6 DK |
4 | * |
5 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1998 | |
6 | * | |
7 | * Copyright 1998 Red Hat corp --- All Rights Reserved | |
8 | * | |
470decc6 DK |
9 | * Generic filesystem journal-writing code; part of the ext2fs |
10 | * journaling system. | |
11 | * | |
12 | * This file manages journals: areas of disk reserved for logging | |
13 | * transactional updates. This includes the kernel journaling thread | |
14 | * which is responsible for scheduling updates to the log. | |
15 | * | |
16 | * We do not actually manage the physical storage of the journal in this | |
17 | * file: that is left to a per-journal policy function, which allows us | |
18 | * to store the journal within a filesystem-specified area for ext2 | |
19 | * journaling (ext2 can use a reserved inode for storing the log). | |
20 | */ | |
21 | ||
22 | #include <linux/module.h> | |
23 | #include <linux/time.h> | |
24 | #include <linux/fs.h> | |
f7f4bccb | 25 | #include <linux/jbd2.h> |
470decc6 DK |
26 | #include <linux/errno.h> |
27 | #include <linux/slab.h> | |
470decc6 DK |
28 | #include <linux/init.h> |
29 | #include <linux/mm.h> | |
7dfb7103 | 30 | #include <linux/freezer.h> |
470decc6 DK |
31 | #include <linux/pagemap.h> |
32 | #include <linux/kthread.h> | |
33 | #include <linux/poison.h> | |
34 | #include <linux/proc_fs.h> | |
8e85fb3f | 35 | #include <linux/seq_file.h> |
c225aa57 | 36 | #include <linux/math64.h> |
879c5e6b | 37 | #include <linux/hash.h> |
d2eecb03 TT |
38 | #include <linux/log2.h> |
39 | #include <linux/vmalloc.h> | |
47def826 | 40 | #include <linux/backing-dev.h> |
39e3ac25 | 41 | #include <linux/bitops.h> |
670be5a7 | 42 | #include <linux/ratelimit.h> |
eb52da3f | 43 | #include <linux/sched/mm.h> |
879c5e6b TT |
44 | |
45 | #define CREATE_TRACE_POINTS | |
46 | #include <trace/events/jbd2.h> | |
470decc6 | 47 | |
7c0f6ba6 | 48 | #include <linux/uaccess.h> |
470decc6 DK |
49 | #include <asm/page.h> |
50 | ||
b6e96d00 TT |
51 | #ifdef CONFIG_JBD2_DEBUG |
52 | ushort jbd2_journal_enable_debug __read_mostly; | |
53 | EXPORT_SYMBOL(jbd2_journal_enable_debug); | |
54 | ||
55 | module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644); | |
56 | MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2"); | |
57 | #endif | |
58 | ||
f7f4bccb MC |
59 | EXPORT_SYMBOL(jbd2_journal_extend); |
60 | EXPORT_SYMBOL(jbd2_journal_stop); | |
61 | EXPORT_SYMBOL(jbd2_journal_lock_updates); | |
62 | EXPORT_SYMBOL(jbd2_journal_unlock_updates); | |
63 | EXPORT_SYMBOL(jbd2_journal_get_write_access); | |
64 | EXPORT_SYMBOL(jbd2_journal_get_create_access); | |
65 | EXPORT_SYMBOL(jbd2_journal_get_undo_access); | |
e06c8227 | 66 | EXPORT_SYMBOL(jbd2_journal_set_triggers); |
f7f4bccb | 67 | EXPORT_SYMBOL(jbd2_journal_dirty_metadata); |
f7f4bccb | 68 | EXPORT_SYMBOL(jbd2_journal_forget); |
f7f4bccb MC |
69 | EXPORT_SYMBOL(jbd2_journal_flush); |
70 | EXPORT_SYMBOL(jbd2_journal_revoke); | |
71 | ||
72 | EXPORT_SYMBOL(jbd2_journal_init_dev); | |
73 | EXPORT_SYMBOL(jbd2_journal_init_inode); | |
f7f4bccb MC |
74 | EXPORT_SYMBOL(jbd2_journal_check_used_features); |
75 | EXPORT_SYMBOL(jbd2_journal_check_available_features); | |
76 | EXPORT_SYMBOL(jbd2_journal_set_features); | |
f7f4bccb MC |
77 | EXPORT_SYMBOL(jbd2_journal_load); |
78 | EXPORT_SYMBOL(jbd2_journal_destroy); | |
f7f4bccb MC |
79 | EXPORT_SYMBOL(jbd2_journal_abort); |
80 | EXPORT_SYMBOL(jbd2_journal_errno); | |
81 | EXPORT_SYMBOL(jbd2_journal_ack_err); | |
82 | EXPORT_SYMBOL(jbd2_journal_clear_err); | |
83 | EXPORT_SYMBOL(jbd2_log_wait_commit); | |
3b799d15 | 84 | EXPORT_SYMBOL(jbd2_log_start_commit); |
f7f4bccb MC |
85 | EXPORT_SYMBOL(jbd2_journal_start_commit); |
86 | EXPORT_SYMBOL(jbd2_journal_force_commit_nested); | |
87 | EXPORT_SYMBOL(jbd2_journal_wipe); | |
88 | EXPORT_SYMBOL(jbd2_journal_blocks_per_page); | |
89 | EXPORT_SYMBOL(jbd2_journal_invalidatepage); | |
90 | EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); | |
91 | EXPORT_SYMBOL(jbd2_journal_force_commit); | |
41617e1a JK |
92 | EXPORT_SYMBOL(jbd2_journal_inode_add_write); |
93 | EXPORT_SYMBOL(jbd2_journal_inode_add_wait); | |
6ba0e7dc RZ |
94 | EXPORT_SYMBOL(jbd2_journal_inode_ranged_write); |
95 | EXPORT_SYMBOL(jbd2_journal_inode_ranged_wait); | |
c851ed54 JK |
96 | EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); |
97 | EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); | |
98 | EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); | |
8aefcd55 | 99 | EXPORT_SYMBOL(jbd2_inode_cache); |
470decc6 | 100 | |
470decc6 | 101 | static void __journal_abort_soft (journal_t *journal, int errno); |
d2eecb03 | 102 | static int jbd2_journal_create_slab(size_t slab_size); |
470decc6 | 103 | |
169f1a2a PG |
104 | #ifdef CONFIG_JBD2_DEBUG |
105 | void __jbd2_debug(int level, const char *file, const char *func, | |
106 | unsigned int line, const char *fmt, ...) | |
107 | { | |
108 | struct va_format vaf; | |
109 | va_list args; | |
110 | ||
111 | if (level > jbd2_journal_enable_debug) | |
112 | return; | |
113 | va_start(args, fmt); | |
114 | vaf.fmt = fmt; | |
115 | vaf.va = &args; | |
9196f571 | 116 | printk(KERN_DEBUG "%s: (%s, %u): %pV", file, func, line, &vaf); |
169f1a2a PG |
117 | va_end(args); |
118 | } | |
119 | EXPORT_SYMBOL(__jbd2_debug); | |
120 | #endif | |
121 | ||
25ed6e8a | 122 | /* Checksumming functions */ |
7747e6d0 | 123 | static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb) |
25ed6e8a | 124 | { |
8595798c | 125 | if (!jbd2_journal_has_csum_v2or3_feature(j)) |
25ed6e8a DW |
126 | return 1; |
127 | ||
128 | return sb->s_checksum_type == JBD2_CRC32C_CHKSUM; | |
129 | } | |
130 | ||
18a6ea1e | 131 | static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb) |
4fd5ea43 | 132 | { |
18a6ea1e DW |
133 | __u32 csum; |
134 | __be32 old_csum; | |
4fd5ea43 DW |
135 | |
136 | old_csum = sb->s_checksum; | |
137 | sb->s_checksum = 0; | |
138 | csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t)); | |
139 | sb->s_checksum = old_csum; | |
140 | ||
141 | return cpu_to_be32(csum); | |
142 | } | |
143 | ||
470decc6 DK |
144 | /* |
145 | * Helper function used to manage commit timeouts | |
146 | */ | |
147 | ||
e3c95788 | 148 | static void commit_timeout(struct timer_list *t) |
470decc6 | 149 | { |
e3c95788 | 150 | journal_t *journal = from_timer(journal, t, j_commit_timer); |
470decc6 | 151 | |
e3c95788 | 152 | wake_up_process(journal->j_task); |
470decc6 DK |
153 | } |
154 | ||
155 | /* | |
f7f4bccb | 156 | * kjournald2: The main thread function used to manage a logging device |
470decc6 DK |
157 | * journal. |
158 | * | |
159 | * This kernel thread is responsible for two things: | |
160 | * | |
161 | * 1) COMMIT: Every so often we need to commit the current state of the | |
162 | * filesystem to disk. The journal thread is responsible for writing | |
163 | * all of the metadata buffers to disk. | |
164 | * | |
165 | * 2) CHECKPOINT: We cannot reuse a used section of the log file until all | |
166 | * of the data in that part of the log has been rewritten elsewhere on | |
167 | * the disk. Flushing these old buffers to reclaim space in the log is | |
168 | * known as checkpointing, and this thread is responsible for that job. | |
169 | */ | |
170 | ||
f7f4bccb | 171 | static int kjournald2(void *arg) |
470decc6 DK |
172 | { |
173 | journal_t *journal = arg; | |
174 | transaction_t *transaction; | |
175 | ||
176 | /* | |
177 | * Set up an interval timer which can be used to trigger a commit wakeup | |
178 | * after the commit interval expires | |
179 | */ | |
e3c95788 | 180 | timer_setup(&journal->j_commit_timer, commit_timeout, 0); |
470decc6 | 181 | |
35c80422 NC |
182 | set_freezable(); |
183 | ||
470decc6 DK |
184 | /* Record that the journal thread is running */ |
185 | journal->j_task = current; | |
186 | wake_up(&journal->j_wait_done_commit); | |
187 | ||
eb52da3f MH |
188 | /* |
189 | * Make sure that no allocations from this kernel thread will ever | |
190 | * recurse to the fs layer because we are responsible for the | |
191 | * transaction commit and any fs involvement might get stuck waiting for | |
192 | * the trasn. commit. | |
193 | */ | |
194 | memalloc_nofs_save(); | |
195 | ||
470decc6 DK |
196 | /* |
197 | * And now, wait forever for commit wakeup events. | |
198 | */ | |
a931da6a | 199 | write_lock(&journal->j_state_lock); |
470decc6 DK |
200 | |
201 | loop: | |
f7f4bccb | 202 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
203 | goto end_loop; |
204 | ||
7821ce41 | 205 | jbd_debug(1, "commit_sequence=%u, commit_request=%u\n", |
470decc6 DK |
206 | journal->j_commit_sequence, journal->j_commit_request); |
207 | ||
208 | if (journal->j_commit_sequence != journal->j_commit_request) { | |
209 | jbd_debug(1, "OK, requests differ\n"); | |
a931da6a | 210 | write_unlock(&journal->j_state_lock); |
470decc6 | 211 | del_timer_sync(&journal->j_commit_timer); |
f7f4bccb | 212 | jbd2_journal_commit_transaction(journal); |
a931da6a | 213 | write_lock(&journal->j_state_lock); |
470decc6 DK |
214 | goto loop; |
215 | } | |
216 | ||
217 | wake_up(&journal->j_wait_done_commit); | |
218 | if (freezing(current)) { | |
219 | /* | |
220 | * The simpler the better. Flushing journal isn't a | |
221 | * good idea, because that depends on threads that may | |
222 | * be already stopped. | |
223 | */ | |
f7f4bccb | 224 | jbd_debug(1, "Now suspending kjournald2\n"); |
a931da6a | 225 | write_unlock(&journal->j_state_lock); |
a0acae0e | 226 | try_to_freeze(); |
a931da6a | 227 | write_lock(&journal->j_state_lock); |
470decc6 DK |
228 | } else { |
229 | /* | |
230 | * We assume on resume that commits are already there, | |
231 | * so we don't sleep | |
232 | */ | |
233 | DEFINE_WAIT(wait); | |
234 | int should_sleep = 1; | |
235 | ||
236 | prepare_to_wait(&journal->j_wait_commit, &wait, | |
237 | TASK_INTERRUPTIBLE); | |
238 | if (journal->j_commit_sequence != journal->j_commit_request) | |
239 | should_sleep = 0; | |
240 | transaction = journal->j_running_transaction; | |
241 | if (transaction && time_after_eq(jiffies, | |
242 | transaction->t_expires)) | |
243 | should_sleep = 0; | |
f7f4bccb | 244 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
245 | should_sleep = 0; |
246 | if (should_sleep) { | |
a931da6a | 247 | write_unlock(&journal->j_state_lock); |
470decc6 | 248 | schedule(); |
a931da6a | 249 | write_lock(&journal->j_state_lock); |
470decc6 DK |
250 | } |
251 | finish_wait(&journal->j_wait_commit, &wait); | |
252 | } | |
253 | ||
f7f4bccb | 254 | jbd_debug(1, "kjournald2 wakes\n"); |
470decc6 DK |
255 | |
256 | /* | |
257 | * Were we woken up by a commit wakeup event? | |
258 | */ | |
259 | transaction = journal->j_running_transaction; | |
260 | if (transaction && time_after_eq(jiffies, transaction->t_expires)) { | |
261 | journal->j_commit_request = transaction->t_tid; | |
262 | jbd_debug(1, "woke because of timeout\n"); | |
263 | } | |
264 | goto loop; | |
265 | ||
266 | end_loop: | |
470decc6 DK |
267 | del_timer_sync(&journal->j_commit_timer); |
268 | journal->j_task = NULL; | |
269 | wake_up(&journal->j_wait_done_commit); | |
270 | jbd_debug(1, "Journal thread exiting.\n"); | |
dbfcef6b | 271 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
272 | return 0; |
273 | } | |
274 | ||
97f06784 | 275 | static int jbd2_journal_start_thread(journal_t *journal) |
470decc6 | 276 | { |
97f06784 PE |
277 | struct task_struct *t; |
278 | ||
90576c0b TT |
279 | t = kthread_run(kjournald2, journal, "jbd2/%s", |
280 | journal->j_devname); | |
97f06784 PE |
281 | if (IS_ERR(t)) |
282 | return PTR_ERR(t); | |
283 | ||
1076d17a | 284 | wait_event(journal->j_wait_done_commit, journal->j_task != NULL); |
97f06784 | 285 | return 0; |
470decc6 DK |
286 | } |
287 | ||
288 | static void journal_kill_thread(journal_t *journal) | |
289 | { | |
a931da6a | 290 | write_lock(&journal->j_state_lock); |
f7f4bccb | 291 | journal->j_flags |= JBD2_UNMOUNT; |
470decc6 DK |
292 | |
293 | while (journal->j_task) { | |
a931da6a | 294 | write_unlock(&journal->j_state_lock); |
3469a32a | 295 | wake_up(&journal->j_wait_commit); |
1076d17a | 296 | wait_event(journal->j_wait_done_commit, journal->j_task == NULL); |
a931da6a | 297 | write_lock(&journal->j_state_lock); |
470decc6 | 298 | } |
a931da6a | 299 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
300 | } |
301 | ||
302 | /* | |
f7f4bccb | 303 | * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. |
470decc6 DK |
304 | * |
305 | * Writes a metadata buffer to a given disk block. The actual IO is not | |
306 | * performed but a new buffer_head is constructed which labels the data | |
307 | * to be written with the correct destination disk block. | |
308 | * | |
309 | * Any magic-number escaping which needs to be done will cause a | |
310 | * copy-out here. If the buffer happens to start with the | |
f7f4bccb | 311 | * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the |
470decc6 DK |
312 | * magic number is only written to the log for descripter blocks. In |
313 | * this case, we copy the data and replace the first word with 0, and we | |
314 | * return a result code which indicates that this buffer needs to be | |
315 | * marked as an escaped buffer in the corresponding log descriptor | |
316 | * block. The missing word can then be restored when the block is read | |
317 | * during recovery. | |
318 | * | |
319 | * If the source buffer has already been modified by a new transaction | |
320 | * since we took the last commit snapshot, we use the frozen copy of | |
f5113eff JK |
321 | * that data for IO. If we end up using the existing buffer_head's data |
322 | * for the write, then we have to make sure nobody modifies it while the | |
323 | * IO is in progress. do_get_write_access() handles this. | |
470decc6 | 324 | * |
f5113eff | 325 | * The function returns a pointer to the buffer_head to be used for IO. |
7821ce41 | 326 | * |
470decc6 DK |
327 | * |
328 | * Return value: | |
329 | * <0: Error | |
330 | * >=0: Finished OK | |
331 | * | |
332 | * On success: | |
333 | * Bit 0 set == escape performed on the data | |
334 | * Bit 1 set == buffer copy-out performed (kfree the data after IO) | |
335 | */ | |
336 | ||
f7f4bccb | 337 | int jbd2_journal_write_metadata_buffer(transaction_t *transaction, |
470decc6 | 338 | struct journal_head *jh_in, |
f5113eff JK |
339 | struct buffer_head **bh_out, |
340 | sector_t blocknr) | |
470decc6 DK |
341 | { |
342 | int need_copy_out = 0; | |
343 | int done_copy_out = 0; | |
344 | int do_escape = 0; | |
345 | char *mapped_data; | |
346 | struct buffer_head *new_bh; | |
470decc6 DK |
347 | struct page *new_page; |
348 | unsigned int new_offset; | |
349 | struct buffer_head *bh_in = jh2bh(jh_in); | |
96577c43 | 350 | journal_t *journal = transaction->t_journal; |
470decc6 DK |
351 | |
352 | /* | |
353 | * The buffer really shouldn't be locked: only the current committing | |
354 | * transaction is allowed to write it, so nobody else is allowed | |
355 | * to do any IO. | |
356 | * | |
357 | * akpm: except if we're journalling data, and write() output is | |
358 | * also part of a shared mapping, and another thread has | |
359 | * decided to launch a writepage() against this buffer. | |
360 | */ | |
361 | J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); | |
362 | ||
6ccaf3e2 | 363 | new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); |
47def826 | 364 | |
96577c43 | 365 | /* keep subsequent assertions sane */ |
96577c43 | 366 | atomic_set(&new_bh->b_count, 1); |
470decc6 | 367 | |
f5113eff JK |
368 | jbd_lock_bh_state(bh_in); |
369 | repeat: | |
470decc6 DK |
370 | /* |
371 | * If a new transaction has already done a buffer copy-out, then | |
372 | * we use that version of the data for the commit. | |
373 | */ | |
470decc6 DK |
374 | if (jh_in->b_frozen_data) { |
375 | done_copy_out = 1; | |
376 | new_page = virt_to_page(jh_in->b_frozen_data); | |
377 | new_offset = offset_in_page(jh_in->b_frozen_data); | |
378 | } else { | |
379 | new_page = jh2bh(jh_in)->b_page; | |
380 | new_offset = offset_in_page(jh2bh(jh_in)->b_data); | |
381 | } | |
382 | ||
303a8f2a | 383 | mapped_data = kmap_atomic(new_page); |
e06c8227 | 384 | /* |
13ceef09 JK |
385 | * Fire data frozen trigger if data already wasn't frozen. Do this |
386 | * before checking for escaping, as the trigger may modify the magic | |
387 | * offset. If a copy-out happens afterwards, it will have the correct | |
388 | * data in the buffer. | |
e06c8227 | 389 | */ |
13ceef09 JK |
390 | if (!done_copy_out) |
391 | jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset, | |
392 | jh_in->b_triggers); | |
e06c8227 | 393 | |
470decc6 DK |
394 | /* |
395 | * Check for escaping | |
396 | */ | |
397 | if (*((__be32 *)(mapped_data + new_offset)) == | |
f7f4bccb | 398 | cpu_to_be32(JBD2_MAGIC_NUMBER)) { |
470decc6 DK |
399 | need_copy_out = 1; |
400 | do_escape = 1; | |
401 | } | |
303a8f2a | 402 | kunmap_atomic(mapped_data); |
470decc6 DK |
403 | |
404 | /* | |
405 | * Do we need to do a data copy? | |
406 | */ | |
407 | if (need_copy_out && !done_copy_out) { | |
408 | char *tmp; | |
409 | ||
410 | jbd_unlock_bh_state(bh_in); | |
af1e76d6 | 411 | tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); |
e6ec116b | 412 | if (!tmp) { |
f5113eff | 413 | brelse(new_bh); |
e6ec116b TT |
414 | return -ENOMEM; |
415 | } | |
470decc6 DK |
416 | jbd_lock_bh_state(bh_in); |
417 | if (jh_in->b_frozen_data) { | |
af1e76d6 | 418 | jbd2_free(tmp, bh_in->b_size); |
470decc6 DK |
419 | goto repeat; |
420 | } | |
421 | ||
422 | jh_in->b_frozen_data = tmp; | |
303a8f2a | 423 | mapped_data = kmap_atomic(new_page); |
f5113eff | 424 | memcpy(tmp, mapped_data + new_offset, bh_in->b_size); |
303a8f2a | 425 | kunmap_atomic(mapped_data); |
470decc6 DK |
426 | |
427 | new_page = virt_to_page(tmp); | |
428 | new_offset = offset_in_page(tmp); | |
429 | done_copy_out = 1; | |
e06c8227 JB |
430 | |
431 | /* | |
432 | * This isn't strictly necessary, as we're using frozen | |
433 | * data for the escaping, but it keeps consistency with | |
434 | * b_frozen_data usage. | |
435 | */ | |
436 | jh_in->b_frozen_triggers = jh_in->b_triggers; | |
470decc6 DK |
437 | } |
438 | ||
439 | /* | |
440 | * Did we need to do an escaping? Now we've done all the | |
441 | * copying, we can finally do so. | |
442 | */ | |
443 | if (do_escape) { | |
303a8f2a | 444 | mapped_data = kmap_atomic(new_page); |
470decc6 | 445 | *((unsigned int *)(mapped_data + new_offset)) = 0; |
303a8f2a | 446 | kunmap_atomic(mapped_data); |
470decc6 DK |
447 | } |
448 | ||
470decc6 | 449 | set_bh_page(new_bh, new_page, new_offset); |
f5113eff JK |
450 | new_bh->b_size = bh_in->b_size; |
451 | new_bh->b_bdev = journal->j_dev; | |
470decc6 | 452 | new_bh->b_blocknr = blocknr; |
b34090e5 | 453 | new_bh->b_private = bh_in; |
470decc6 DK |
454 | set_buffer_mapped(new_bh); |
455 | set_buffer_dirty(new_bh); | |
456 | ||
f5113eff | 457 | *bh_out = new_bh; |
470decc6 DK |
458 | |
459 | /* | |
460 | * The to-be-written buffer needs to get moved to the io queue, | |
461 | * and the original buffer whose contents we are shadowing or | |
462 | * copying is moved to the transaction's shadow queue. | |
463 | */ | |
464 | JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); | |
96577c43 | 465 | spin_lock(&journal->j_list_lock); |
466 | __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); | |
467 | spin_unlock(&journal->j_list_lock); | |
b34090e5 | 468 | set_buffer_shadow(bh_in); |
96577c43 | 469 | jbd_unlock_bh_state(bh_in); |
470 | ||
470decc6 DK |
471 | return do_escape | (done_copy_out << 1); |
472 | } | |
473 | ||
474 | /* | |
475 | * Allocation code for the journal file. Manage the space left in the | |
476 | * journal, so that we can begin checkpointing when appropriate. | |
477 | */ | |
478 | ||
470decc6 | 479 | /* |
e4471831 TT |
480 | * Called with j_state_lock locked for writing. |
481 | * Returns true if a transaction commit was started. | |
470decc6 | 482 | */ |
f7f4bccb | 483 | int __jbd2_log_start_commit(journal_t *journal, tid_t target) |
470decc6 | 484 | { |
e7b04ac0 ES |
485 | /* Return if the txn has already requested to be committed */ |
486 | if (journal->j_commit_request == target) | |
487 | return 0; | |
488 | ||
470decc6 | 489 | /* |
deeeaf13 TT |
490 | * The only transaction we can possibly wait upon is the |
491 | * currently running transaction (if it exists). Otherwise, | |
492 | * the target tid must be an old one. | |
470decc6 | 493 | */ |
deeeaf13 TT |
494 | if (journal->j_running_transaction && |
495 | journal->j_running_transaction->t_tid == target) { | |
470decc6 | 496 | /* |
bcf3d0bc | 497 | * We want a new commit: OK, mark the request and wakeup the |
470decc6 DK |
498 | * commit thread. We do _not_ do the commit ourselves. |
499 | */ | |
500 | ||
501 | journal->j_commit_request = target; | |
7821ce41 | 502 | jbd_debug(1, "JBD2: requesting commit %u/%u\n", |
470decc6 DK |
503 | journal->j_commit_request, |
504 | journal->j_commit_sequence); | |
9fff24aa | 505 | journal->j_running_transaction->t_requested = jiffies; |
470decc6 DK |
506 | wake_up(&journal->j_wait_commit); |
507 | return 1; | |
deeeaf13 TT |
508 | } else if (!tid_geq(journal->j_commit_request, target)) |
509 | /* This should never happen, but if it does, preserve | |
510 | the evidence before kjournald goes into a loop and | |
511 | increments j_commit_sequence beyond all recognition. */ | |
f2a44523 | 512 | WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", |
1be2add6 TT |
513 | journal->j_commit_request, |
514 | journal->j_commit_sequence, | |
7821ce41 | 515 | target, journal->j_running_transaction ? |
1be2add6 | 516 | journal->j_running_transaction->t_tid : 0); |
470decc6 DK |
517 | return 0; |
518 | } | |
519 | ||
f7f4bccb | 520 | int jbd2_log_start_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
521 | { |
522 | int ret; | |
523 | ||
a931da6a | 524 | write_lock(&journal->j_state_lock); |
f7f4bccb | 525 | ret = __jbd2_log_start_commit(journal, tid); |
a931da6a | 526 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
527 | return ret; |
528 | } | |
529 | ||
530 | /* | |
9ff86446 DM |
531 | * Force and wait any uncommitted transactions. We can only force the running |
532 | * transaction if we don't have an active handle, otherwise, we will deadlock. | |
533 | * Returns: <0 in case of error, | |
534 | * 0 if nothing to commit, | |
535 | * 1 if transaction was successfully committed. | |
470decc6 | 536 | */ |
9ff86446 | 537 | static int __jbd2_journal_force_commit(journal_t *journal) |
470decc6 DK |
538 | { |
539 | transaction_t *transaction = NULL; | |
540 | tid_t tid; | |
9ff86446 | 541 | int need_to_start = 0, ret = 0; |
470decc6 | 542 | |
a931da6a | 543 | read_lock(&journal->j_state_lock); |
470decc6 DK |
544 | if (journal->j_running_transaction && !current->journal_info) { |
545 | transaction = journal->j_running_transaction; | |
e4471831 TT |
546 | if (!tid_geq(journal->j_commit_request, transaction->t_tid)) |
547 | need_to_start = 1; | |
470decc6 DK |
548 | } else if (journal->j_committing_transaction) |
549 | transaction = journal->j_committing_transaction; | |
550 | ||
551 | if (!transaction) { | |
9ff86446 | 552 | /* Nothing to commit */ |
a931da6a | 553 | read_unlock(&journal->j_state_lock); |
9ff86446 | 554 | return 0; |
470decc6 | 555 | } |
470decc6 | 556 | tid = transaction->t_tid; |
a931da6a | 557 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
558 | if (need_to_start) |
559 | jbd2_log_start_commit(journal, tid); | |
9ff86446 DM |
560 | ret = jbd2_log_wait_commit(journal, tid); |
561 | if (!ret) | |
562 | ret = 1; | |
563 | ||
564 | return ret; | |
565 | } | |
566 | ||
567 | /** | |
568 | * Force and wait upon a commit if the calling process is not within | |
569 | * transaction. This is used for forcing out undo-protected data which contains | |
570 | * bitmaps, when the fs is running out of space. | |
571 | * | |
572 | * @journal: journal to force | |
573 | * Returns true if progress was made. | |
574 | */ | |
575 | int jbd2_journal_force_commit_nested(journal_t *journal) | |
576 | { | |
577 | int ret; | |
578 | ||
579 | ret = __jbd2_journal_force_commit(journal); | |
580 | return ret > 0; | |
581 | } | |
582 | ||
583 | /** | |
584 | * int journal_force_commit() - force any uncommitted transactions | |
585 | * @journal: journal to force | |
586 | * | |
587 | * Caller want unconditional commit. We can only force the running transaction | |
588 | * if we don't have an active handle, otherwise, we will deadlock. | |
589 | */ | |
590 | int jbd2_journal_force_commit(journal_t *journal) | |
591 | { | |
592 | int ret; | |
593 | ||
594 | J_ASSERT(!current->journal_info); | |
595 | ret = __jbd2_journal_force_commit(journal); | |
596 | if (ret > 0) | |
597 | ret = 0; | |
598 | return ret; | |
470decc6 DK |
599 | } |
600 | ||
601 | /* | |
602 | * Start a commit of the current running transaction (if any). Returns true | |
c88ccea3 JK |
603 | * if a transaction is going to be committed (or is currently already |
604 | * committing), and fills its tid in at *ptid | |
470decc6 | 605 | */ |
f7f4bccb | 606 | int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) |
470decc6 DK |
607 | { |
608 | int ret = 0; | |
609 | ||
a931da6a | 610 | write_lock(&journal->j_state_lock); |
470decc6 DK |
611 | if (journal->j_running_transaction) { |
612 | tid_t tid = journal->j_running_transaction->t_tid; | |
613 | ||
c88ccea3 JK |
614 | __jbd2_log_start_commit(journal, tid); |
615 | /* There's a running transaction and we've just made sure | |
616 | * it's commit has been scheduled. */ | |
617 | if (ptid) | |
470decc6 | 618 | *ptid = tid; |
c88ccea3 JK |
619 | ret = 1; |
620 | } else if (journal->j_committing_transaction) { | |
470decc6 | 621 | /* |
12810ad7 AB |
622 | * If commit has been started, then we have to wait for |
623 | * completion of that transaction. | |
470decc6 | 624 | */ |
c88ccea3 JK |
625 | if (ptid) |
626 | *ptid = journal->j_committing_transaction->t_tid; | |
470decc6 DK |
627 | ret = 1; |
628 | } | |
a931da6a | 629 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
630 | return ret; |
631 | } | |
632 | ||
bbd2be36 JK |
633 | /* |
634 | * Return 1 if a given transaction has not yet sent barrier request | |
635 | * connected with a transaction commit. If 0 is returned, transaction | |
636 | * may or may not have sent the barrier. Used to avoid sending barrier | |
637 | * twice in common cases. | |
638 | */ | |
639 | int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) | |
640 | { | |
641 | int ret = 0; | |
642 | transaction_t *commit_trans; | |
643 | ||
644 | if (!(journal->j_flags & JBD2_BARRIER)) | |
645 | return 0; | |
646 | read_lock(&journal->j_state_lock); | |
647 | /* Transaction already committed? */ | |
648 | if (tid_geq(journal->j_commit_sequence, tid)) | |
649 | goto out; | |
650 | commit_trans = journal->j_committing_transaction; | |
651 | if (!commit_trans || commit_trans->t_tid != tid) { | |
652 | ret = 1; | |
653 | goto out; | |
654 | } | |
655 | /* | |
656 | * Transaction is being committed and we already proceeded to | |
657 | * submitting a flush to fs partition? | |
658 | */ | |
659 | if (journal->j_fs_dev != journal->j_dev) { | |
660 | if (!commit_trans->t_need_data_flush || | |
661 | commit_trans->t_state >= T_COMMIT_DFLUSH) | |
662 | goto out; | |
663 | } else { | |
664 | if (commit_trans->t_state >= T_COMMIT_JFLUSH) | |
665 | goto out; | |
666 | } | |
667 | ret = 1; | |
668 | out: | |
669 | read_unlock(&journal->j_state_lock); | |
670 | return ret; | |
671 | } | |
672 | EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); | |
673 | ||
470decc6 DK |
674 | /* |
675 | * Wait for a specified commit to complete. | |
676 | * The caller may not hold the journal lock. | |
677 | */ | |
f7f4bccb | 678 | int jbd2_log_wait_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
679 | { |
680 | int err = 0; | |
681 | ||
a931da6a | 682 | read_lock(&journal->j_state_lock); |
c52c47e4 JK |
683 | #ifdef CONFIG_PROVE_LOCKING |
684 | /* | |
685 | * Some callers make sure transaction is already committing and in that | |
686 | * case we cannot block on open handles anymore. So don't warn in that | |
687 | * case. | |
688 | */ | |
689 | if (tid_gt(tid, journal->j_commit_sequence) && | |
690 | (!journal->j_committing_transaction || | |
691 | journal->j_committing_transaction->t_tid != tid)) { | |
692 | read_unlock(&journal->j_state_lock); | |
693 | jbd2_might_wait_for_commit(journal); | |
694 | read_lock(&journal->j_state_lock); | |
695 | } | |
696 | #endif | |
e23291b9 | 697 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 | 698 | if (!tid_geq(journal->j_commit_request, tid)) { |
75685071 | 699 | printk(KERN_ERR |
7821ce41 | 700 | "%s: error: j_commit_request=%u, tid=%u\n", |
329d291f | 701 | __func__, journal->j_commit_request, tid); |
470decc6 | 702 | } |
470decc6 | 703 | #endif |
470decc6 | 704 | while (tid_gt(tid, journal->j_commit_sequence)) { |
7821ce41 | 705 | jbd_debug(1, "JBD2: want %u, j_commit_sequence=%u\n", |
470decc6 | 706 | tid, journal->j_commit_sequence); |
a931da6a | 707 | read_unlock(&journal->j_state_lock); |
3469a32a | 708 | wake_up(&journal->j_wait_commit); |
470decc6 DK |
709 | wait_event(journal->j_wait_done_commit, |
710 | !tid_gt(tid, journal->j_commit_sequence)); | |
a931da6a | 711 | read_lock(&journal->j_state_lock); |
470decc6 | 712 | } |
a931da6a | 713 | read_unlock(&journal->j_state_lock); |
470decc6 | 714 | |
75685071 | 715 | if (unlikely(is_journal_aborted(journal))) |
470decc6 | 716 | err = -EIO; |
470decc6 DK |
717 | return err; |
718 | } | |
719 | ||
b8a6176c JK |
720 | /* Return 1 when transaction with given tid has already committed. */ |
721 | int jbd2_transaction_committed(journal_t *journal, tid_t tid) | |
722 | { | |
723 | int ret = 1; | |
724 | ||
725 | read_lock(&journal->j_state_lock); | |
726 | if (journal->j_running_transaction && | |
727 | journal->j_running_transaction->t_tid == tid) | |
728 | ret = 0; | |
729 | if (journal->j_committing_transaction && | |
730 | journal->j_committing_transaction->t_tid == tid) | |
731 | ret = 0; | |
732 | read_unlock(&journal->j_state_lock); | |
733 | return ret; | |
734 | } | |
735 | EXPORT_SYMBOL(jbd2_transaction_committed); | |
736 | ||
d76a3a77 TT |
737 | /* |
738 | * When this function returns the transaction corresponding to tid | |
739 | * will be completed. If the transaction has currently running, start | |
740 | * committing that transaction before waiting for it to complete. If | |
741 | * the transaction id is stale, it is by definition already completed, | |
742 | * so just return SUCCESS. | |
743 | */ | |
744 | int jbd2_complete_transaction(journal_t *journal, tid_t tid) | |
745 | { | |
746 | int need_to_wait = 1; | |
747 | ||
748 | read_lock(&journal->j_state_lock); | |
749 | if (journal->j_running_transaction && | |
750 | journal->j_running_transaction->t_tid == tid) { | |
751 | if (journal->j_commit_request != tid) { | |
752 | /* transaction not yet started, so request it */ | |
753 | read_unlock(&journal->j_state_lock); | |
754 | jbd2_log_start_commit(journal, tid); | |
755 | goto wait_commit; | |
756 | } | |
757 | } else if (!(journal->j_committing_transaction && | |
758 | journal->j_committing_transaction->t_tid == tid)) | |
759 | need_to_wait = 0; | |
760 | read_unlock(&journal->j_state_lock); | |
761 | if (!need_to_wait) | |
762 | return 0; | |
763 | wait_commit: | |
764 | return jbd2_log_wait_commit(journal, tid); | |
765 | } | |
766 | EXPORT_SYMBOL(jbd2_complete_transaction); | |
767 | ||
470decc6 DK |
768 | /* |
769 | * Log buffer allocation routines: | |
770 | */ | |
771 | ||
18eba7aa | 772 | int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) |
470decc6 DK |
773 | { |
774 | unsigned long blocknr; | |
775 | ||
a931da6a | 776 | write_lock(&journal->j_state_lock); |
470decc6 DK |
777 | J_ASSERT(journal->j_free > 1); |
778 | ||
779 | blocknr = journal->j_head; | |
780 | journal->j_head++; | |
781 | journal->j_free--; | |
782 | if (journal->j_head == journal->j_last) | |
783 | journal->j_head = journal->j_first; | |
a931da6a | 784 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 785 | return jbd2_journal_bmap(journal, blocknr, retp); |
470decc6 DK |
786 | } |
787 | ||
788 | /* | |
789 | * Conversion of logical to physical block numbers for the journal | |
790 | * | |
791 | * On external journals the journal blocks are identity-mapped, so | |
792 | * this is a no-op. If needed, we can use j_blk_offset - everything is | |
793 | * ready. | |
794 | */ | |
f7f4bccb | 795 | int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, |
18eba7aa | 796 | unsigned long long *retp) |
470decc6 DK |
797 | { |
798 | int err = 0; | |
18eba7aa | 799 | unsigned long long ret; |
470decc6 DK |
800 | |
801 | if (journal->j_inode) { | |
802 | ret = bmap(journal->j_inode, blocknr); | |
803 | if (ret) | |
804 | *retp = ret; | |
805 | else { | |
470decc6 DK |
806 | printk(KERN_ALERT "%s: journal block not found " |
807 | "at offset %lu on %s\n", | |
05496769 | 808 | __func__, blocknr, journal->j_devname); |
470decc6 DK |
809 | err = -EIO; |
810 | __journal_abort_soft(journal, err); | |
811 | } | |
812 | } else { | |
813 | *retp = blocknr; /* +journal->j_blk_offset */ | |
814 | } | |
815 | return err; | |
816 | } | |
817 | ||
818 | /* | |
819 | * We play buffer_head aliasing tricks to write data/metadata blocks to | |
820 | * the journal without copying their contents, but for journal | |
821 | * descriptor blocks we do need to generate bona fide buffers. | |
822 | * | |
f7f4bccb | 823 | * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying |
470decc6 DK |
824 | * the buffer's contents they really should run flush_dcache_page(bh->b_page). |
825 | * But we don't bother doing that, so there will be coherency problems with | |
826 | * mmaps of blockdevs which hold live JBD-controlled filesystems. | |
827 | */ | |
32ab6715 JK |
828 | struct buffer_head * |
829 | jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type) | |
470decc6 | 830 | { |
32ab6715 | 831 | journal_t *journal = transaction->t_journal; |
470decc6 | 832 | struct buffer_head *bh; |
18eba7aa | 833 | unsigned long long blocknr; |
32ab6715 | 834 | journal_header_t *header; |
470decc6 DK |
835 | int err; |
836 | ||
f7f4bccb | 837 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
838 | |
839 | if (err) | |
840 | return NULL; | |
841 | ||
842 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
4b905671 JK |
843 | if (!bh) |
844 | return NULL; | |
470decc6 DK |
845 | lock_buffer(bh); |
846 | memset(bh->b_data, 0, journal->j_blocksize); | |
32ab6715 JK |
847 | header = (journal_header_t *)bh->b_data; |
848 | header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); | |
849 | header->h_blocktype = cpu_to_be32(type); | |
850 | header->h_sequence = cpu_to_be32(transaction->t_tid); | |
470decc6 DK |
851 | set_buffer_uptodate(bh); |
852 | unlock_buffer(bh); | |
853 | BUFFER_TRACE(bh, "return this buffer"); | |
e5a120ae | 854 | return bh; |
470decc6 DK |
855 | } |
856 | ||
1101cd4d JK |
857 | void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh) |
858 | { | |
859 | struct jbd2_journal_block_tail *tail; | |
860 | __u32 csum; | |
861 | ||
862 | if (!jbd2_journal_has_csum_v2or3(j)) | |
863 | return; | |
864 | ||
865 | tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize - | |
866 | sizeof(struct jbd2_journal_block_tail)); | |
867 | tail->t_checksum = 0; | |
868 | csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); | |
869 | tail->t_checksum = cpu_to_be32(csum); | |
870 | } | |
871 | ||
79feb521 JK |
872 | /* |
873 | * Return tid of the oldest transaction in the journal and block in the journal | |
874 | * where the transaction starts. | |
875 | * | |
876 | * If the journal is now empty, return which will be the next transaction ID | |
877 | * we will write and where will that transaction start. | |
878 | * | |
879 | * The return value is 0 if journal tail cannot be pushed any further, 1 if | |
880 | * it can. | |
881 | */ | |
882 | int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, | |
883 | unsigned long *block) | |
884 | { | |
885 | transaction_t *transaction; | |
886 | int ret; | |
887 | ||
888 | read_lock(&journal->j_state_lock); | |
889 | spin_lock(&journal->j_list_lock); | |
890 | transaction = journal->j_checkpoint_transactions; | |
891 | if (transaction) { | |
892 | *tid = transaction->t_tid; | |
893 | *block = transaction->t_log_start; | |
894 | } else if ((transaction = journal->j_committing_transaction) != NULL) { | |
895 | *tid = transaction->t_tid; | |
896 | *block = transaction->t_log_start; | |
897 | } else if ((transaction = journal->j_running_transaction) != NULL) { | |
898 | *tid = transaction->t_tid; | |
899 | *block = journal->j_head; | |
900 | } else { | |
901 | *tid = journal->j_transaction_sequence; | |
902 | *block = journal->j_head; | |
903 | } | |
904 | ret = tid_gt(*tid, journal->j_tail_sequence); | |
905 | spin_unlock(&journal->j_list_lock); | |
906 | read_unlock(&journal->j_state_lock); | |
907 | ||
908 | return ret; | |
909 | } | |
910 | ||
911 | /* | |
912 | * Update information in journal structure and in on disk journal superblock | |
913 | * about log tail. This function does not check whether information passed in | |
914 | * really pushes log tail further. It's responsibility of the caller to make | |
915 | * sure provided log tail information is valid (e.g. by holding | |
916 | * j_checkpoint_mutex all the time between computing log tail and calling this | |
917 | * function as is the case with jbd2_cleanup_journal_tail()). | |
918 | * | |
919 | * Requires j_checkpoint_mutex | |
920 | */ | |
6f6a6fda | 921 | int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) |
79feb521 JK |
922 | { |
923 | unsigned long freed; | |
6f6a6fda | 924 | int ret; |
79feb521 JK |
925 | |
926 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); | |
927 | ||
928 | /* | |
929 | * We cannot afford for write to remain in drive's caches since as | |
930 | * soon as we update j_tail, next transaction can start reusing journal | |
931 | * space and if we lose sb update during power failure we'd replay | |
932 | * old transaction with possibly newly overwritten data. | |
933 | */ | |
17f423b5 JK |
934 | ret = jbd2_journal_update_sb_log_tail(journal, tid, block, |
935 | REQ_SYNC | REQ_FUA); | |
6f6a6fda JQ |
936 | if (ret) |
937 | goto out; | |
938 | ||
79feb521 JK |
939 | write_lock(&journal->j_state_lock); |
940 | freed = block - journal->j_tail; | |
941 | if (block < journal->j_tail) | |
942 | freed += journal->j_last - journal->j_first; | |
943 | ||
944 | trace_jbd2_update_log_tail(journal, tid, block, freed); | |
945 | jbd_debug(1, | |
7821ce41 | 946 | "Cleaning journal tail from %u to %u (offset %lu), " |
79feb521 JK |
947 | "freeing %lu\n", |
948 | journal->j_tail_sequence, tid, block, freed); | |
949 | ||
950 | journal->j_free += freed; | |
951 | journal->j_tail_sequence = tid; | |
952 | journal->j_tail = block; | |
953 | write_unlock(&journal->j_state_lock); | |
6f6a6fda JQ |
954 | |
955 | out: | |
956 | return ret; | |
79feb521 JK |
957 | } |
958 | ||
3339578f | 959 | /* |
85e0c4e8 | 960 | * This is a variation of __jbd2_update_log_tail which checks for validity of |
3339578f JK |
961 | * provided log tail and locks j_checkpoint_mutex. So it is safe against races |
962 | * with other threads updating log tail. | |
963 | */ | |
964 | void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) | |
965 | { | |
6fa7aa50 | 966 | mutex_lock_io(&journal->j_checkpoint_mutex); |
3339578f JK |
967 | if (tid_gt(tid, journal->j_tail_sequence)) |
968 | __jbd2_update_log_tail(journal, tid, block); | |
969 | mutex_unlock(&journal->j_checkpoint_mutex); | |
970 | } | |
971 | ||
8e85fb3f JL |
972 | struct jbd2_stats_proc_session { |
973 | journal_t *journal; | |
974 | struct transaction_stats_s *stats; | |
975 | int start; | |
976 | int max; | |
977 | }; | |
978 | ||
8e85fb3f JL |
979 | static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) |
980 | { | |
981 | return *pos ? NULL : SEQ_START_TOKEN; | |
982 | } | |
983 | ||
984 | static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) | |
985 | { | |
986 | return NULL; | |
987 | } | |
988 | ||
989 | static int jbd2_seq_info_show(struct seq_file *seq, void *v) | |
990 | { | |
991 | struct jbd2_stats_proc_session *s = seq->private; | |
992 | ||
993 | if (v != SEQ_START_TOKEN) | |
994 | return 0; | |
9fff24aa TT |
995 | seq_printf(seq, "%lu transactions (%lu requested), " |
996 | "each up to %u blocks\n", | |
997 | s->stats->ts_tid, s->stats->ts_requested, | |
998 | s->journal->j_max_transaction_buffers); | |
8e85fb3f JL |
999 | if (s->stats->ts_tid == 0) |
1000 | return 0; | |
1001 | seq_printf(seq, "average: \n %ums waiting for transaction\n", | |
bf699327 | 1002 | jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); |
9fff24aa TT |
1003 | seq_printf(seq, " %ums request delay\n", |
1004 | (s->stats->ts_requested == 0) ? 0 : | |
1005 | jiffies_to_msecs(s->stats->run.rs_request_delay / | |
1006 | s->stats->ts_requested)); | |
8e85fb3f | 1007 | seq_printf(seq, " %ums running transaction\n", |
bf699327 | 1008 | jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); |
8e85fb3f | 1009 | seq_printf(seq, " %ums transaction was being locked\n", |
bf699327 | 1010 | jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); |
8e85fb3f | 1011 | seq_printf(seq, " %ums flushing data (in ordered mode)\n", |
bf699327 | 1012 | jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); |
8e85fb3f | 1013 | seq_printf(seq, " %ums logging transaction\n", |
bf699327 | 1014 | jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); |
c225aa57 SHT |
1015 | seq_printf(seq, " %lluus average transaction commit time\n", |
1016 | div_u64(s->journal->j_average_commit_time, 1000)); | |
8e85fb3f | 1017 | seq_printf(seq, " %lu handles per transaction\n", |
bf699327 | 1018 | s->stats->run.rs_handle_count / s->stats->ts_tid); |
8e85fb3f | 1019 | seq_printf(seq, " %lu blocks per transaction\n", |
bf699327 | 1020 | s->stats->run.rs_blocks / s->stats->ts_tid); |
8e85fb3f | 1021 | seq_printf(seq, " %lu logged blocks per transaction\n", |
bf699327 | 1022 | s->stats->run.rs_blocks_logged / s->stats->ts_tid); |
8e85fb3f JL |
1023 | return 0; |
1024 | } | |
1025 | ||
1026 | static void jbd2_seq_info_stop(struct seq_file *seq, void *v) | |
1027 | { | |
1028 | } | |
1029 | ||
88e9d34c | 1030 | static const struct seq_operations jbd2_seq_info_ops = { |
8e85fb3f JL |
1031 | .start = jbd2_seq_info_start, |
1032 | .next = jbd2_seq_info_next, | |
1033 | .stop = jbd2_seq_info_stop, | |
1034 | .show = jbd2_seq_info_show, | |
1035 | }; | |
1036 | ||
1037 | static int jbd2_seq_info_open(struct inode *inode, struct file *file) | |
1038 | { | |
d9dda78b | 1039 | journal_t *journal = PDE_DATA(inode); |
8e85fb3f JL |
1040 | struct jbd2_stats_proc_session *s; |
1041 | int rc, size; | |
1042 | ||
1043 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
1044 | if (s == NULL) | |
1045 | return -ENOMEM; | |
1046 | size = sizeof(struct transaction_stats_s); | |
1047 | s->stats = kmalloc(size, GFP_KERNEL); | |
1048 | if (s->stats == NULL) { | |
1049 | kfree(s); | |
1050 | return -ENOMEM; | |
1051 | } | |
1052 | spin_lock(&journal->j_history_lock); | |
1053 | memcpy(s->stats, &journal->j_stats, size); | |
1054 | s->journal = journal; | |
1055 | spin_unlock(&journal->j_history_lock); | |
1056 | ||
1057 | rc = seq_open(file, &jbd2_seq_info_ops); | |
1058 | if (rc == 0) { | |
1059 | struct seq_file *m = file->private_data; | |
1060 | m->private = s; | |
1061 | } else { | |
1062 | kfree(s->stats); | |
1063 | kfree(s); | |
1064 | } | |
1065 | return rc; | |
1066 | ||
1067 | } | |
1068 | ||
1069 | static int jbd2_seq_info_release(struct inode *inode, struct file *file) | |
1070 | { | |
1071 | struct seq_file *seq = file->private_data; | |
1072 | struct jbd2_stats_proc_session *s = seq->private; | |
1073 | kfree(s->stats); | |
1074 | kfree(s); | |
1075 | return seq_release(inode, file); | |
1076 | } | |
1077 | ||
828c0950 | 1078 | static const struct file_operations jbd2_seq_info_fops = { |
8e85fb3f JL |
1079 | .owner = THIS_MODULE, |
1080 | .open = jbd2_seq_info_open, | |
1081 | .read = seq_read, | |
1082 | .llseek = seq_lseek, | |
1083 | .release = jbd2_seq_info_release, | |
1084 | }; | |
1085 | ||
1086 | static struct proc_dir_entry *proc_jbd2_stats; | |
1087 | ||
1088 | static void jbd2_stats_proc_init(journal_t *journal) | |
1089 | { | |
05496769 | 1090 | journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); |
8e85fb3f | 1091 | if (journal->j_proc_entry) { |
79da3664 DL |
1092 | proc_create_data("info", S_IRUGO, journal->j_proc_entry, |
1093 | &jbd2_seq_info_fops, journal); | |
8e85fb3f JL |
1094 | } |
1095 | } | |
1096 | ||
1097 | static void jbd2_stats_proc_exit(journal_t *journal) | |
1098 | { | |
8e85fb3f | 1099 | remove_proc_entry("info", journal->j_proc_entry); |
05496769 | 1100 | remove_proc_entry(journal->j_devname, proc_jbd2_stats); |
8e85fb3f JL |
1101 | } |
1102 | ||
470decc6 DK |
1103 | /* |
1104 | * Management for journal control blocks: functions to create and | |
1105 | * destroy journal_t structures, and to initialise and read existing | |
1106 | * journal blocks from disk. */ | |
1107 | ||
1108 | /* First: create and setup a journal_t object in memory. We initialise | |
1109 | * very few fields yet: that has to wait until we have created the | |
1110 | * journal structures from from scratch, or loaded them from disk. */ | |
1111 | ||
f0c9fd54 GT |
1112 | static journal_t *journal_init_common(struct block_device *bdev, |
1113 | struct block_device *fs_dev, | |
1114 | unsigned long long start, int len, int blocksize) | |
470decc6 | 1115 | { |
ab714aff | 1116 | static struct lock_class_key jbd2_trans_commit_key; |
470decc6 DK |
1117 | journal_t *journal; |
1118 | int err; | |
f0c9fd54 GT |
1119 | struct buffer_head *bh; |
1120 | int n; | |
470decc6 | 1121 | |
3ebfdf88 | 1122 | journal = kzalloc(sizeof(*journal), GFP_KERNEL); |
470decc6 | 1123 | if (!journal) |
b7271b0a | 1124 | return NULL; |
470decc6 DK |
1125 | |
1126 | init_waitqueue_head(&journal->j_wait_transaction_locked); | |
470decc6 | 1127 | init_waitqueue_head(&journal->j_wait_done_commit); |
470decc6 DK |
1128 | init_waitqueue_head(&journal->j_wait_commit); |
1129 | init_waitqueue_head(&journal->j_wait_updates); | |
8f7d89f3 | 1130 | init_waitqueue_head(&journal->j_wait_reserved); |
470decc6 DK |
1131 | mutex_init(&journal->j_barrier); |
1132 | mutex_init(&journal->j_checkpoint_mutex); | |
1133 | spin_lock_init(&journal->j_revoke_lock); | |
1134 | spin_lock_init(&journal->j_list_lock); | |
a931da6a | 1135 | rwlock_init(&journal->j_state_lock); |
470decc6 | 1136 | |
cd02ff0b | 1137 | journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); |
30773840 TT |
1138 | journal->j_min_batch_time = 0; |
1139 | journal->j_max_batch_time = 15000; /* 15ms */ | |
8f7d89f3 | 1140 | atomic_set(&journal->j_reserved_credits, 0); |
470decc6 DK |
1141 | |
1142 | /* The journal is marked for error until we succeed with recovery! */ | |
f7f4bccb | 1143 | journal->j_flags = JBD2_ABORT; |
470decc6 DK |
1144 | |
1145 | /* Set up a default-sized revoke table for the new mount. */ | |
f7f4bccb | 1146 | err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); |
cd9cb405 EB |
1147 | if (err) |
1148 | goto err_cleanup; | |
8e85fb3f | 1149 | |
bf699327 | 1150 | spin_lock_init(&journal->j_history_lock); |
8e85fb3f | 1151 | |
ab714aff JK |
1152 | lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle", |
1153 | &jbd2_trans_commit_key, 0); | |
1154 | ||
f0c9fd54 GT |
1155 | /* journal descriptor can store up to n blocks -bzzz */ |
1156 | journal->j_blocksize = blocksize; | |
1157 | journal->j_dev = bdev; | |
1158 | journal->j_fs_dev = fs_dev; | |
1159 | journal->j_blk_offset = start; | |
1160 | journal->j_maxlen = len; | |
1161 | n = journal->j_blocksize / sizeof(journal_block_tag_t); | |
1162 | journal->j_wbufsize = n; | |
1163 | journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *), | |
1164 | GFP_KERNEL); | |
cd9cb405 EB |
1165 | if (!journal->j_wbuf) |
1166 | goto err_cleanup; | |
f0c9fd54 GT |
1167 | |
1168 | bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize); | |
1169 | if (!bh) { | |
1170 | pr_err("%s: Cannot get buffer for journal superblock\n", | |
1171 | __func__); | |
cd9cb405 | 1172 | goto err_cleanup; |
f0c9fd54 GT |
1173 | } |
1174 | journal->j_sb_buffer = bh; | |
1175 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
1176 | ||
470decc6 | 1177 | return journal; |
cd9cb405 EB |
1178 | |
1179 | err_cleanup: | |
1180 | kfree(journal->j_wbuf); | |
1181 | jbd2_journal_destroy_revoke(journal); | |
1182 | kfree(journal); | |
1183 | return NULL; | |
470decc6 DK |
1184 | } |
1185 | ||
f7f4bccb | 1186 | /* jbd2_journal_init_dev and jbd2_journal_init_inode: |
470decc6 DK |
1187 | * |
1188 | * Create a journal structure assigned some fixed set of disk blocks to | |
1189 | * the journal. We don't actually touch those disk blocks yet, but we | |
1190 | * need to set up all of the mapping information to tell the journaling | |
1191 | * system where the journal blocks are. | |
1192 | * | |
1193 | */ | |
1194 | ||
1195 | /** | |
5648ba5b | 1196 | * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure |
470decc6 DK |
1197 | * @bdev: Block device on which to create the journal |
1198 | * @fs_dev: Device which hold journalled filesystem for this journal. | |
1199 | * @start: Block nr Start of journal. | |
1200 | * @len: Length of the journal in blocks. | |
1201 | * @blocksize: blocksize of journalling device | |
5648ba5b RD |
1202 | * |
1203 | * Returns: a newly created journal_t * | |
470decc6 | 1204 | * |
f7f4bccb | 1205 | * jbd2_journal_init_dev creates a journal which maps a fixed contiguous |
470decc6 DK |
1206 | * range of blocks on an arbitrary block device. |
1207 | * | |
1208 | */ | |
f0c9fd54 | 1209 | journal_t *jbd2_journal_init_dev(struct block_device *bdev, |
470decc6 | 1210 | struct block_device *fs_dev, |
18eba7aa | 1211 | unsigned long long start, int len, int blocksize) |
470decc6 | 1212 | { |
f0c9fd54 | 1213 | journal_t *journal; |
470decc6 | 1214 | |
f0c9fd54 | 1215 | journal = journal_init_common(bdev, fs_dev, start, len, blocksize); |
470decc6 DK |
1216 | if (!journal) |
1217 | return NULL; | |
1218 | ||
0587aa3d | 1219 | bdevname(journal->j_dev, journal->j_devname); |
81ae394b | 1220 | strreplace(journal->j_devname, '/', '!'); |
4b905671 | 1221 | jbd2_stats_proc_init(journal); |
4b905671 | 1222 | |
470decc6 DK |
1223 | return journal; |
1224 | } | |
1225 | ||
1226 | /** | |
f7f4bccb | 1227 | * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. |
470decc6 DK |
1228 | * @inode: An inode to create the journal in |
1229 | * | |
f7f4bccb | 1230 | * jbd2_journal_init_inode creates a journal which maps an on-disk inode as |
470decc6 DK |
1231 | * the journal. The inode must exist already, must support bmap() and |
1232 | * must have all data blocks preallocated. | |
1233 | */ | |
f0c9fd54 | 1234 | journal_t *jbd2_journal_init_inode(struct inode *inode) |
470decc6 | 1235 | { |
f0c9fd54 | 1236 | journal_t *journal; |
05496769 | 1237 | char *p; |
18eba7aa | 1238 | unsigned long long blocknr; |
470decc6 | 1239 | |
f0c9fd54 GT |
1240 | blocknr = bmap(inode, 0); |
1241 | if (!blocknr) { | |
1242 | pr_err("%s: Cannot locate journal superblock\n", | |
1243 | __func__); | |
1244 | return NULL; | |
1245 | } | |
1246 | ||
1247 | jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n", | |
1248 | inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size, | |
1249 | inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); | |
1250 | ||
1251 | journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev, | |
1252 | blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits, | |
1253 | inode->i_sb->s_blocksize); | |
470decc6 DK |
1254 | if (!journal) |
1255 | return NULL; | |
1256 | ||
470decc6 | 1257 | journal->j_inode = inode; |
05496769 | 1258 | bdevname(journal->j_dev, journal->j_devname); |
81ae394b | 1259 | p = strreplace(journal->j_devname, '/', '!'); |
90576c0b | 1260 | sprintf(p, "-%lu", journal->j_inode->i_ino); |
8e85fb3f | 1261 | jbd2_stats_proc_init(journal); |
470decc6 | 1262 | |
470decc6 DK |
1263 | return journal; |
1264 | } | |
1265 | ||
1266 | /* | |
1267 | * If the journal init or create aborts, we need to mark the journal | |
1268 | * superblock as being NULL to prevent the journal destroy from writing | |
1269 | * back a bogus superblock. | |
1270 | */ | |
1271 | static void journal_fail_superblock (journal_t *journal) | |
1272 | { | |
1273 | struct buffer_head *bh = journal->j_sb_buffer; | |
1274 | brelse(bh); | |
1275 | journal->j_sb_buffer = NULL; | |
1276 | } | |
1277 | ||
1278 | /* | |
1279 | * Given a journal_t structure, initialise the various fields for | |
1280 | * startup of a new journaling session. We use this both when creating | |
1281 | * a journal, and after recovering an old journal to reset it for | |
1282 | * subsequent use. | |
1283 | */ | |
1284 | ||
1285 | static int journal_reset(journal_t *journal) | |
1286 | { | |
1287 | journal_superblock_t *sb = journal->j_superblock; | |
18eba7aa | 1288 | unsigned long long first, last; |
470decc6 DK |
1289 | |
1290 | first = be32_to_cpu(sb->s_first); | |
1291 | last = be32_to_cpu(sb->s_maxlen); | |
f6f50e28 | 1292 | if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) { |
f2a44523 | 1293 | printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n", |
f6f50e28 JK |
1294 | first, last); |
1295 | journal_fail_superblock(journal); | |
1296 | return -EINVAL; | |
1297 | } | |
470decc6 DK |
1298 | |
1299 | journal->j_first = first; | |
1300 | journal->j_last = last; | |
1301 | ||
1302 | journal->j_head = first; | |
1303 | journal->j_tail = first; | |
1304 | journal->j_free = last - first; | |
1305 | ||
1306 | journal->j_tail_sequence = journal->j_transaction_sequence; | |
1307 | journal->j_commit_sequence = journal->j_transaction_sequence - 1; | |
1308 | journal->j_commit_request = journal->j_commit_sequence; | |
1309 | ||
1310 | journal->j_max_transaction_buffers = journal->j_maxlen / 4; | |
1311 | ||
470decc6 DK |
1312 | /* |
1313 | * As a special case, if the on-disk copy is already marked as needing | |
24bcc89c JK |
1314 | * no recovery (s_start == 0), then we can safely defer the superblock |
1315 | * update until the next commit by setting JBD2_FLUSHED. This avoids | |
470decc6 DK |
1316 | * attempting a write to a potential-readonly device. |
1317 | */ | |
24bcc89c | 1318 | if (sb->s_start == 0) { |
f2a44523 | 1319 | jbd_debug(1, "JBD2: Skipping superblock update on recovered sb " |
7821ce41 | 1320 | "(start %ld, seq %u, errno %d)\n", |
470decc6 DK |
1321 | journal->j_tail, journal->j_tail_sequence, |
1322 | journal->j_errno); | |
24bcc89c JK |
1323 | journal->j_flags |= JBD2_FLUSHED; |
1324 | } else { | |
a78bb11d | 1325 | /* Lock here to make assertions happy... */ |
6fa7aa50 | 1326 | mutex_lock_io(&journal->j_checkpoint_mutex); |
79feb521 | 1327 | /* |
70fd7614 | 1328 | * Update log tail information. We use REQ_FUA since new |
79feb521 JK |
1329 | * transaction will start reusing journal space and so we |
1330 | * must make sure information about current log tail is on | |
1331 | * disk before that. | |
1332 | */ | |
1333 | jbd2_journal_update_sb_log_tail(journal, | |
1334 | journal->j_tail_sequence, | |
1335 | journal->j_tail, | |
17f423b5 | 1336 | REQ_SYNC | REQ_FUA); |
a78bb11d | 1337 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 | 1338 | } |
24bcc89c JK |
1339 | return jbd2_journal_start_thread(journal); |
1340 | } | |
470decc6 | 1341 | |
538bcaa6 TT |
1342 | /* |
1343 | * This function expects that the caller will have locked the journal | |
1344 | * buffer head, and will return with it unlocked | |
1345 | */ | |
2a222ca9 | 1346 | static int jbd2_write_superblock(journal_t *journal, int write_flags) |
24bcc89c JK |
1347 | { |
1348 | struct buffer_head *bh = journal->j_sb_buffer; | |
fe52d17c | 1349 | journal_superblock_t *sb = journal->j_superblock; |
79feb521 | 1350 | int ret; |
470decc6 | 1351 | |
742b06b5 JX |
1352 | /* Buffer got discarded which means block device got invalidated */ |
1353 | if (!buffer_mapped(bh)) | |
1354 | return -EIO; | |
1355 | ||
2a222ca9 | 1356 | trace_jbd2_write_superblock(journal, write_flags); |
79feb521 | 1357 | if (!(journal->j_flags & JBD2_BARRIER)) |
28a8f0d3 | 1358 | write_flags &= ~(REQ_FUA | REQ_PREFLUSH); |
914258bf TT |
1359 | if (buffer_write_io_error(bh)) { |
1360 | /* | |
1361 | * Oh, dear. A previous attempt to write the journal | |
1362 | * superblock failed. This could happen because the | |
1363 | * USB device was yanked out. Or it could happen to | |
1364 | * be a transient write error and maybe the block will | |
1365 | * be remapped. Nothing we can do but to retry the | |
1366 | * write and hope for the best. | |
1367 | */ | |
1368 | printk(KERN_ERR "JBD2: previous I/O error detected " | |
1369 | "for journal superblock update for %s.\n", | |
1370 | journal->j_devname); | |
1371 | clear_buffer_write_io_error(bh); | |
1372 | set_buffer_uptodate(bh); | |
1373 | } | |
a58ca992 TT |
1374 | if (jbd2_journal_has_csum_v2or3(journal)) |
1375 | sb->s_checksum = jbd2_superblock_csum(journal, sb); | |
79feb521 JK |
1376 | get_bh(bh); |
1377 | bh->b_end_io = end_buffer_write_sync; | |
2a222ca9 | 1378 | ret = submit_bh(REQ_OP_WRITE, write_flags, bh); |
79feb521 | 1379 | wait_on_buffer(bh); |
24bcc89c | 1380 | if (buffer_write_io_error(bh)) { |
24bcc89c JK |
1381 | clear_buffer_write_io_error(bh); |
1382 | set_buffer_uptodate(bh); | |
79feb521 JK |
1383 | ret = -EIO; |
1384 | } | |
1385 | if (ret) { | |
1386 | printk(KERN_ERR "JBD2: Error %d detected when updating " | |
1387 | "journal superblock for %s.\n", ret, | |
1388 | journal->j_devname); | |
6f6a6fda | 1389 | jbd2_journal_abort(journal, ret); |
24bcc89c | 1390 | } |
6f6a6fda JQ |
1391 | |
1392 | return ret; | |
24bcc89c JK |
1393 | } |
1394 | ||
1395 | /** | |
1396 | * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. | |
1397 | * @journal: The journal to update. | |
79feb521 JK |
1398 | * @tail_tid: TID of the new transaction at the tail of the log |
1399 | * @tail_block: The first block of the transaction at the tail of the log | |
1400 | * @write_op: With which operation should we write the journal sb | |
24bcc89c JK |
1401 | * |
1402 | * Update a journal's superblock information about log tail and write it to | |
1403 | * disk, waiting for the IO to complete. | |
1404 | */ | |
6f6a6fda | 1405 | int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, |
79feb521 | 1406 | unsigned long tail_block, int write_op) |
24bcc89c JK |
1407 | { |
1408 | journal_superblock_t *sb = journal->j_superblock; | |
6f6a6fda | 1409 | int ret; |
24bcc89c | 1410 | |
85e0c4e8 TT |
1411 | if (is_journal_aborted(journal)) |
1412 | return -EIO; | |
1413 | ||
a78bb11d | 1414 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); |
79feb521 JK |
1415 | jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", |
1416 | tail_block, tail_tid); | |
470decc6 | 1417 | |
538bcaa6 | 1418 | lock_buffer(journal->j_sb_buffer); |
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)); |
538bcaa6 TT |
1449 | lock_buffer(journal->j_sb_buffer); |
1450 | if (sb->s_start == 0) { /* Is it already empty? */ | |
1451 | unlock_buffer(journal->j_sb_buffer); | |
eeecef0a ES |
1452 | return; |
1453 | } | |
538bcaa6 | 1454 | |
7821ce41 | 1455 | jbd_debug(1, "JBD2: Marking journal as empty (seq %u)\n", |
24bcc89c | 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); |
470decc6 | 1460 | |
c0a2ad9b | 1461 | jbd2_write_superblock(journal, write_op); |
470decc6 | 1462 | |
24bcc89c | 1463 | /* Log is no longer empty */ |
a931da6a | 1464 | write_lock(&journal->j_state_lock); |
24bcc89c | 1465 | journal->j_flags |= JBD2_FLUSHED; |
a931da6a | 1466 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1467 | } |
1468 | ||
24bcc89c JK |
1469 | |
1470 | /** | |
1471 | * jbd2_journal_update_sb_errno() - Update error in the journal. | |
1472 | * @journal: The journal to update. | |
1473 | * | |
1474 | * Update a journal's errno. Write updated superblock to disk waiting for IO | |
1475 | * to complete. | |
1476 | */ | |
d796c52e | 1477 | void jbd2_journal_update_sb_errno(journal_t *journal) |
24bcc89c JK |
1478 | { |
1479 | journal_superblock_t *sb = journal->j_superblock; | |
fb7c0244 | 1480 | int errcode; |
24bcc89c | 1481 | |
538bcaa6 | 1482 | lock_buffer(journal->j_sb_buffer); |
fb7c0244 | 1483 | errcode = journal->j_errno; |
fb7c0244 TT |
1484 | if (errcode == -ESHUTDOWN) |
1485 | errcode = 0; | |
1486 | jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", errcode); | |
1487 | sb->s_errno = cpu_to_be32(errcode); | |
24bcc89c | 1488 | |
17f423b5 | 1489 | jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA); |
24bcc89c | 1490 | } |
d796c52e | 1491 | EXPORT_SYMBOL(jbd2_journal_update_sb_errno); |
24bcc89c | 1492 | |
470decc6 DK |
1493 | /* |
1494 | * Read the superblock for a given journal, performing initial | |
1495 | * validation of the format. | |
1496 | */ | |
470decc6 DK |
1497 | static int journal_get_superblock(journal_t *journal) |
1498 | { | |
1499 | struct buffer_head *bh; | |
1500 | journal_superblock_t *sb; | |
1501 | int err = -EIO; | |
1502 | ||
1503 | bh = journal->j_sb_buffer; | |
1504 | ||
1505 | J_ASSERT(bh != NULL); | |
1506 | if (!buffer_uptodate(bh)) { | |
dfec8a14 | 1507 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
470decc6 DK |
1508 | wait_on_buffer(bh); |
1509 | if (!buffer_uptodate(bh)) { | |
f2a44523 EG |
1510 | printk(KERN_ERR |
1511 | "JBD2: IO error reading journal superblock\n"); | |
470decc6 DK |
1512 | goto out; |
1513 | } | |
1514 | } | |
1515 | ||
25ed6e8a DW |
1516 | if (buffer_verified(bh)) |
1517 | return 0; | |
1518 | ||
470decc6 DK |
1519 | sb = journal->j_superblock; |
1520 | ||
1521 | err = -EINVAL; | |
1522 | ||
f7f4bccb | 1523 | if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || |
470decc6 | 1524 | sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { |
f2a44523 | 1525 | printk(KERN_WARNING "JBD2: no valid journal superblock found\n"); |
470decc6 DK |
1526 | goto out; |
1527 | } | |
1528 | ||
1529 | switch(be32_to_cpu(sb->s_header.h_blocktype)) { | |
f7f4bccb | 1530 | case JBD2_SUPERBLOCK_V1: |
470decc6 DK |
1531 | journal->j_format_version = 1; |
1532 | break; | |
f7f4bccb | 1533 | case JBD2_SUPERBLOCK_V2: |
470decc6 DK |
1534 | journal->j_format_version = 2; |
1535 | break; | |
1536 | default: | |
f2a44523 | 1537 | printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n"); |
470decc6 DK |
1538 | goto out; |
1539 | } | |
1540 | ||
1541 | if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen) | |
1542 | journal->j_maxlen = be32_to_cpu(sb->s_maxlen); | |
1543 | else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) { | |
f2a44523 | 1544 | printk(KERN_WARNING "JBD2: journal file too short\n"); |
470decc6 DK |
1545 | goto out; |
1546 | } | |
1547 | ||
8762202d EG |
1548 | if (be32_to_cpu(sb->s_first) == 0 || |
1549 | be32_to_cpu(sb->s_first) >= journal->j_maxlen) { | |
1550 | printk(KERN_WARNING | |
1551 | "JBD2: Invalid start block of journal: %u\n", | |
1552 | be32_to_cpu(sb->s_first)); | |
1553 | goto out; | |
1554 | } | |
1555 | ||
56316a0d DW |
1556 | if (jbd2_has_feature_csum2(journal) && |
1557 | jbd2_has_feature_csum3(journal)) { | |
db9ee220 DW |
1558 | /* Can't have checksum v2 and v3 at the same time! */ |
1559 | printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 " | |
1560 | "at the same time!\n"); | |
1561 | goto out; | |
1562 | } | |
1563 | ||
8595798c | 1564 | if (jbd2_journal_has_csum_v2or3_feature(journal) && |
56316a0d | 1565 | jbd2_has_feature_checksum(journal)) { |
feb8c6d3 DW |
1566 | /* Can't have checksum v1 and v2 on at the same time! */ |
1567 | printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 " | |
1568 | "at the same time!\n"); | |
1569 | goto out; | |
1570 | } | |
1571 | ||
25ed6e8a | 1572 | if (!jbd2_verify_csum_type(journal, sb)) { |
a67c848a | 1573 | printk(KERN_ERR "JBD2: Unknown checksum type\n"); |
25ed6e8a DW |
1574 | goto out; |
1575 | } | |
1576 | ||
01b5adce | 1577 | /* Load the checksum driver */ |
8595798c | 1578 | if (jbd2_journal_has_csum_v2or3_feature(journal)) { |
01b5adce DW |
1579 | journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); |
1580 | if (IS_ERR(journal->j_chksum_driver)) { | |
a67c848a | 1581 | printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n"); |
01b5adce DW |
1582 | err = PTR_ERR(journal->j_chksum_driver); |
1583 | journal->j_chksum_driver = NULL; | |
1584 | goto out; | |
1585 | } | |
1586 | } | |
1587 | ||
a58ca992 TT |
1588 | if (jbd2_journal_has_csum_v2or3(journal)) { |
1589 | /* Check superblock checksum */ | |
1590 | if (sb->s_checksum != jbd2_superblock_csum(journal, sb)) { | |
1591 | printk(KERN_ERR "JBD2: journal checksum error\n"); | |
1592 | err = -EFSBADCRC; | |
1593 | goto out; | |
1594 | } | |
4fd5ea43 | 1595 | |
a58ca992 | 1596 | /* Precompute checksum seed for all metadata */ |
4fd5ea43 DW |
1597 | journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, |
1598 | sizeof(sb->s_uuid)); | |
a58ca992 | 1599 | } |
4fd5ea43 | 1600 | |
25ed6e8a DW |
1601 | set_buffer_verified(bh); |
1602 | ||
470decc6 DK |
1603 | return 0; |
1604 | ||
1605 | out: | |
1606 | journal_fail_superblock(journal); | |
1607 | return err; | |
1608 | } | |
1609 | ||
1610 | /* | |
1611 | * Load the on-disk journal superblock and read the key fields into the | |
1612 | * journal_t. | |
1613 | */ | |
1614 | ||
1615 | static int load_superblock(journal_t *journal) | |
1616 | { | |
1617 | int err; | |
1618 | journal_superblock_t *sb; | |
1619 | ||
1620 | err = journal_get_superblock(journal); | |
1621 | if (err) | |
1622 | return err; | |
1623 | ||
1624 | sb = journal->j_superblock; | |
1625 | ||
1626 | journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); | |
1627 | journal->j_tail = be32_to_cpu(sb->s_start); | |
1628 | journal->j_first = be32_to_cpu(sb->s_first); | |
1629 | journal->j_last = be32_to_cpu(sb->s_maxlen); | |
1630 | journal->j_errno = be32_to_cpu(sb->s_errno); | |
1631 | ||
1632 | return 0; | |
1633 | } | |
1634 | ||
1635 | ||
1636 | /** | |
f7f4bccb | 1637 | * int jbd2_journal_load() - Read journal from disk. |
470decc6 DK |
1638 | * @journal: Journal to act on. |
1639 | * | |
1640 | * Given a journal_t structure which tells us which disk blocks contain | |
1641 | * a journal, read the journal from disk to initialise the in-memory | |
1642 | * structures. | |
1643 | */ | |
f7f4bccb | 1644 | int jbd2_journal_load(journal_t *journal) |
470decc6 DK |
1645 | { |
1646 | int err; | |
1647 | journal_superblock_t *sb; | |
1648 | ||
1649 | err = load_superblock(journal); | |
1650 | if (err) | |
1651 | return err; | |
1652 | ||
1653 | sb = journal->j_superblock; | |
1654 | /* If this is a V2 superblock, then we have to check the | |
1655 | * features flags on it. */ | |
1656 | ||
1657 | if (journal->j_format_version >= 2) { | |
1658 | if ((sb->s_feature_ro_compat & | |
f7f4bccb | 1659 | ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || |
470decc6 | 1660 | (sb->s_feature_incompat & |
f7f4bccb | 1661 | ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { |
f2a44523 EG |
1662 | printk(KERN_WARNING |
1663 | "JBD2: Unrecognised features on journal\n"); | |
470decc6 DK |
1664 | return -EINVAL; |
1665 | } | |
1666 | } | |
1667 | ||
d2eecb03 TT |
1668 | /* |
1669 | * Create a slab for this blocksize | |
1670 | */ | |
1671 | err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize)); | |
1672 | if (err) | |
1673 | return err; | |
1674 | ||
470decc6 DK |
1675 | /* Let the recovery code check whether it needs to recover any |
1676 | * data from the journal. */ | |
f7f4bccb | 1677 | if (jbd2_journal_recover(journal)) |
470decc6 DK |
1678 | goto recovery_error; |
1679 | ||
e6a47428 TT |
1680 | if (journal->j_failed_commit) { |
1681 | printk(KERN_ERR "JBD2: journal transaction %u on %s " | |
1682 | "is corrupt.\n", journal->j_failed_commit, | |
1683 | journal->j_devname); | |
6a797d27 | 1684 | return -EFSCORRUPTED; |
e6a47428 TT |
1685 | } |
1686 | ||
470decc6 DK |
1687 | /* OK, we've finished with the dynamic journal bits: |
1688 | * reinitialise the dynamic contents of the superblock in memory | |
1689 | * and reset them on disk. */ | |
1690 | if (journal_reset(journal)) | |
1691 | goto recovery_error; | |
1692 | ||
f7f4bccb MC |
1693 | journal->j_flags &= ~JBD2_ABORT; |
1694 | journal->j_flags |= JBD2_LOADED; | |
470decc6 DK |
1695 | return 0; |
1696 | ||
1697 | recovery_error: | |
f2a44523 | 1698 | printk(KERN_WARNING "JBD2: recovery failed\n"); |
470decc6 DK |
1699 | return -EIO; |
1700 | } | |
1701 | ||
1702 | /** | |
f7f4bccb | 1703 | * void jbd2_journal_destroy() - Release a journal_t structure. |
470decc6 DK |
1704 | * @journal: Journal to act on. |
1705 | * | |
1706 | * Release a journal_t structure once it is no longer in use by the | |
1707 | * journaled object. | |
44519faf | 1708 | * Return <0 if we couldn't clean up the journal. |
470decc6 | 1709 | */ |
44519faf | 1710 | int jbd2_journal_destroy(journal_t *journal) |
470decc6 | 1711 | { |
44519faf HK |
1712 | int err = 0; |
1713 | ||
470decc6 DK |
1714 | /* Wait for the commit thread to wake up and die. */ |
1715 | journal_kill_thread(journal); | |
1716 | ||
1717 | /* Force a final log commit */ | |
1718 | if (journal->j_running_transaction) | |
f7f4bccb | 1719 | jbd2_journal_commit_transaction(journal); |
470decc6 DK |
1720 | |
1721 | /* Force any old transactions to disk */ | |
1722 | ||
1723 | /* Totally anal locking here... */ | |
1724 | spin_lock(&journal->j_list_lock); | |
1725 | while (journal->j_checkpoint_transactions != NULL) { | |
1726 | spin_unlock(&journal->j_list_lock); | |
6fa7aa50 | 1727 | mutex_lock_io(&journal->j_checkpoint_mutex); |
841df7df | 1728 | err = jbd2_log_do_checkpoint(journal); |
1a0d3786 | 1729 | mutex_unlock(&journal->j_checkpoint_mutex); |
841df7df JK |
1730 | /* |
1731 | * If checkpointing failed, just free the buffers to avoid | |
1732 | * looping forever | |
1733 | */ | |
1734 | if (err) { | |
1735 | jbd2_journal_destroy_checkpoint(journal); | |
1736 | spin_lock(&journal->j_list_lock); | |
1737 | break; | |
1738 | } | |
470decc6 DK |
1739 | spin_lock(&journal->j_list_lock); |
1740 | } | |
1741 | ||
1742 | J_ASSERT(journal->j_running_transaction == NULL); | |
1743 | J_ASSERT(journal->j_committing_transaction == NULL); | |
1744 | J_ASSERT(journal->j_checkpoint_transactions == NULL); | |
1745 | spin_unlock(&journal->j_list_lock); | |
1746 | ||
470decc6 | 1747 | if (journal->j_sb_buffer) { |
44519faf | 1748 | if (!is_journal_aborted(journal)) { |
6fa7aa50 | 1749 | mutex_lock_io(&journal->j_checkpoint_mutex); |
c0a2ad9b OH |
1750 | |
1751 | write_lock(&journal->j_state_lock); | |
1752 | journal->j_tail_sequence = | |
1753 | ++journal->j_transaction_sequence; | |
1754 | write_unlock(&journal->j_state_lock); | |
1755 | ||
70fd7614 | 1756 | jbd2_mark_journal_empty(journal, |
17f423b5 | 1757 | REQ_SYNC | REQ_PREFLUSH | REQ_FUA); |
a78bb11d JK |
1758 | mutex_unlock(&journal->j_checkpoint_mutex); |
1759 | } else | |
44519faf | 1760 | err = -EIO; |
470decc6 DK |
1761 | brelse(journal->j_sb_buffer); |
1762 | } | |
1763 | ||
8e85fb3f JL |
1764 | if (journal->j_proc_entry) |
1765 | jbd2_stats_proc_exit(journal); | |
d9f39d1e | 1766 | iput(journal->j_inode); |
470decc6 | 1767 | if (journal->j_revoke) |
f7f4bccb | 1768 | jbd2_journal_destroy_revoke(journal); |
01b5adce DW |
1769 | if (journal->j_chksum_driver) |
1770 | crypto_free_shash(journal->j_chksum_driver); | |
470decc6 DK |
1771 | kfree(journal->j_wbuf); |
1772 | kfree(journal); | |
44519faf HK |
1773 | |
1774 | return err; | |
470decc6 DK |
1775 | } |
1776 | ||
1777 | ||
1778 | /** | |
f7f4bccb | 1779 | *int jbd2_journal_check_used_features () - Check if features specified are used. |
470decc6 DK |
1780 | * @journal: Journal to check. |
1781 | * @compat: bitmask of compatible features | |
1782 | * @ro: bitmask of features that force read-only mount | |
1783 | * @incompat: bitmask of incompatible features | |
1784 | * | |
1785 | * Check whether the journal uses all of a given set of | |
1786 | * features. Return true (non-zero) if it does. | |
1787 | **/ | |
1788 | ||
f7f4bccb | 1789 | int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1790 | unsigned long ro, unsigned long incompat) |
1791 | { | |
1792 | journal_superblock_t *sb; | |
1793 | ||
1794 | if (!compat && !ro && !incompat) | |
1795 | return 1; | |
1113e1b5 PL |
1796 | /* Load journal superblock if it is not loaded yet. */ |
1797 | if (journal->j_format_version == 0 && | |
1798 | journal_get_superblock(journal) != 0) | |
1799 | return 0; | |
470decc6 DK |
1800 | if (journal->j_format_version == 1) |
1801 | return 0; | |
1802 | ||
1803 | sb = journal->j_superblock; | |
1804 | ||
1805 | if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && | |
1806 | ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && | |
1807 | ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) | |
1808 | return 1; | |
1809 | ||
1810 | return 0; | |
1811 | } | |
1812 | ||
1813 | /** | |
f7f4bccb | 1814 | * int jbd2_journal_check_available_features() - Check feature set in journalling layer |
470decc6 DK |
1815 | * @journal: Journal to check. |
1816 | * @compat: bitmask of compatible features | |
1817 | * @ro: bitmask of features that force read-only mount | |
1818 | * @incompat: bitmask of incompatible features | |
1819 | * | |
1820 | * Check whether the journaling code supports the use of | |
1821 | * all of a given set of features on this journal. Return true | |
1822 | * (non-zero) if it can. */ | |
1823 | ||
f7f4bccb | 1824 | int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1825 | unsigned long ro, unsigned long incompat) |
1826 | { | |
470decc6 DK |
1827 | if (!compat && !ro && !incompat) |
1828 | return 1; | |
1829 | ||
470decc6 DK |
1830 | /* We can support any known requested features iff the |
1831 | * superblock is in version 2. Otherwise we fail to support any | |
1832 | * extended sb features. */ | |
1833 | ||
1834 | if (journal->j_format_version != 2) | |
1835 | return 0; | |
1836 | ||
f7f4bccb MC |
1837 | if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && |
1838 | (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && | |
1839 | (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) | |
470decc6 DK |
1840 | return 1; |
1841 | ||
1842 | return 0; | |
1843 | } | |
1844 | ||
1845 | /** | |
f7f4bccb | 1846 | * int jbd2_journal_set_features () - Mark a given journal feature in the superblock |
470decc6 DK |
1847 | * @journal: Journal to act on. |
1848 | * @compat: bitmask of compatible features | |
1849 | * @ro: bitmask of features that force read-only mount | |
1850 | * @incompat: bitmask of incompatible features | |
1851 | * | |
1852 | * Mark a given journal feature as present on the | |
1853 | * superblock. Returns true if the requested features could be set. | |
1854 | * | |
1855 | */ | |
1856 | ||
f7f4bccb | 1857 | int jbd2_journal_set_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1858 | unsigned long ro, unsigned long incompat) |
1859 | { | |
25ed6e8a DW |
1860 | #define INCOMPAT_FEATURE_ON(f) \ |
1861 | ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f))) | |
1862 | #define COMPAT_FEATURE_ON(f) \ | |
1863 | ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f))) | |
470decc6 DK |
1864 | journal_superblock_t *sb; |
1865 | ||
f7f4bccb | 1866 | if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) |
470decc6 DK |
1867 | return 1; |
1868 | ||
f7f4bccb | 1869 | if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) |
470decc6 DK |
1870 | return 0; |
1871 | ||
db9ee220 DW |
1872 | /* If enabling v2 checksums, turn on v3 instead */ |
1873 | if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) { | |
1874 | incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2; | |
1875 | incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3; | |
1876 | } | |
1877 | ||
1878 | /* Asking for checksumming v3 and v1? Only give them v3. */ | |
1879 | if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 && | |
25ed6e8a DW |
1880 | compat & JBD2_FEATURE_COMPAT_CHECKSUM) |
1881 | compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM; | |
1882 | ||
470decc6 DK |
1883 | jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", |
1884 | compat, ro, incompat); | |
1885 | ||
1886 | sb = journal->j_superblock; | |
1887 | ||
538bcaa6 TT |
1888 | /* Load the checksum driver if necessary */ |
1889 | if ((journal->j_chksum_driver == NULL) && | |
1890 | INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { | |
1891 | journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); | |
1892 | if (IS_ERR(journal->j_chksum_driver)) { | |
1893 | printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n"); | |
1894 | journal->j_chksum_driver = NULL; | |
1895 | return 0; | |
1896 | } | |
1897 | /* Precompute checksum seed for all metadata */ | |
1898 | journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, | |
1899 | sizeof(sb->s_uuid)); | |
1900 | } | |
1901 | ||
1902 | lock_buffer(journal->j_sb_buffer); | |
1903 | ||
db9ee220 DW |
1904 | /* If enabling v3 checksums, update superblock */ |
1905 | if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { | |
25ed6e8a DW |
1906 | sb->s_checksum_type = JBD2_CRC32C_CHKSUM; |
1907 | sb->s_feature_compat &= | |
1908 | ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM); | |
1909 | } | |
1910 | ||
1911 | /* If enabling v1 checksums, downgrade superblock */ | |
1912 | if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM)) | |
1913 | sb->s_feature_incompat &= | |
db9ee220 DW |
1914 | ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 | |
1915 | JBD2_FEATURE_INCOMPAT_CSUM_V3); | |
25ed6e8a | 1916 | |
470decc6 DK |
1917 | sb->s_feature_compat |= cpu_to_be32(compat); |
1918 | sb->s_feature_ro_compat |= cpu_to_be32(ro); | |
1919 | sb->s_feature_incompat |= cpu_to_be32(incompat); | |
538bcaa6 | 1920 | unlock_buffer(journal->j_sb_buffer); |
470decc6 DK |
1921 | |
1922 | return 1; | |
25ed6e8a DW |
1923 | #undef COMPAT_FEATURE_ON |
1924 | #undef INCOMPAT_FEATURE_ON | |
470decc6 DK |
1925 | } |
1926 | ||
818d276c GS |
1927 | /* |
1928 | * jbd2_journal_clear_features () - Clear a given journal feature in the | |
1929 | * superblock | |
1930 | * @journal: Journal to act on. | |
1931 | * @compat: bitmask of compatible features | |
1932 | * @ro: bitmask of features that force read-only mount | |
1933 | * @incompat: bitmask of incompatible features | |
1934 | * | |
1935 | * Clear a given journal feature as present on the | |
1936 | * superblock. | |
1937 | */ | |
1938 | void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, | |
1939 | unsigned long ro, unsigned long incompat) | |
1940 | { | |
1941 | journal_superblock_t *sb; | |
1942 | ||
1943 | jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", | |
1944 | compat, ro, incompat); | |
1945 | ||
1946 | sb = journal->j_superblock; | |
1947 | ||
1948 | sb->s_feature_compat &= ~cpu_to_be32(compat); | |
1949 | sb->s_feature_ro_compat &= ~cpu_to_be32(ro); | |
1950 | sb->s_feature_incompat &= ~cpu_to_be32(incompat); | |
1951 | } | |
1952 | EXPORT_SYMBOL(jbd2_journal_clear_features); | |
470decc6 | 1953 | |
470decc6 | 1954 | /** |
f7f4bccb | 1955 | * int jbd2_journal_flush () - Flush journal |
470decc6 DK |
1956 | * @journal: Journal to act on. |
1957 | * | |
1958 | * Flush all data for a given journal to disk and empty the journal. | |
1959 | * Filesystems can use this when remounting readonly to ensure that | |
1960 | * recovery does not need to happen on remount. | |
1961 | */ | |
1962 | ||
f7f4bccb | 1963 | int jbd2_journal_flush(journal_t *journal) |
470decc6 DK |
1964 | { |
1965 | int err = 0; | |
1966 | transaction_t *transaction = NULL; | |
470decc6 | 1967 | |
a931da6a | 1968 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1969 | |
1970 | /* Force everything buffered to the log... */ | |
1971 | if (journal->j_running_transaction) { | |
1972 | transaction = journal->j_running_transaction; | |
f7f4bccb | 1973 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1974 | } else if (journal->j_committing_transaction) |
1975 | transaction = journal->j_committing_transaction; | |
1976 | ||
1977 | /* Wait for the log commit to complete... */ | |
1978 | if (transaction) { | |
1979 | tid_t tid = transaction->t_tid; | |
1980 | ||
a931da6a | 1981 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 1982 | jbd2_log_wait_commit(journal, tid); |
470decc6 | 1983 | } else { |
a931da6a | 1984 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1985 | } |
1986 | ||
1987 | /* ...and flush everything in the log out to disk. */ | |
1988 | spin_lock(&journal->j_list_lock); | |
1989 | while (!err && journal->j_checkpoint_transactions != NULL) { | |
1990 | spin_unlock(&journal->j_list_lock); | |
6fa7aa50 | 1991 | mutex_lock_io(&journal->j_checkpoint_mutex); |
f7f4bccb | 1992 | err = jbd2_log_do_checkpoint(journal); |
44519faf | 1993 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
1994 | spin_lock(&journal->j_list_lock); |
1995 | } | |
1996 | spin_unlock(&journal->j_list_lock); | |
44519faf HK |
1997 | |
1998 | if (is_journal_aborted(journal)) | |
1999 | return -EIO; | |
2000 | ||
6fa7aa50 | 2001 | mutex_lock_io(&journal->j_checkpoint_mutex); |
6f6a6fda JQ |
2002 | if (!err) { |
2003 | err = jbd2_cleanup_journal_tail(journal); | |
2004 | if (err < 0) { | |
2005 | mutex_unlock(&journal->j_checkpoint_mutex); | |
2006 | goto out; | |
2007 | } | |
2008 | err = 0; | |
2009 | } | |
470decc6 DK |
2010 | |
2011 | /* Finally, mark the journal as really needing no recovery. | |
2012 | * This sets s_start==0 in the underlying superblock, which is | |
2013 | * the magic code for a fully-recovered superblock. Any future | |
2014 | * commits of data to the journal will restore the current | |
2015 | * s_start value. */ | |
17f423b5 | 2016 | jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA); |
a78bb11d | 2017 | mutex_unlock(&journal->j_checkpoint_mutex); |
a931da6a | 2018 | write_lock(&journal->j_state_lock); |
470decc6 DK |
2019 | J_ASSERT(!journal->j_running_transaction); |
2020 | J_ASSERT(!journal->j_committing_transaction); | |
2021 | J_ASSERT(!journal->j_checkpoint_transactions); | |
2022 | J_ASSERT(journal->j_head == journal->j_tail); | |
2023 | J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); | |
a931da6a | 2024 | write_unlock(&journal->j_state_lock); |
6f6a6fda JQ |
2025 | out: |
2026 | return err; | |
470decc6 DK |
2027 | } |
2028 | ||
2029 | /** | |
f7f4bccb | 2030 | * int jbd2_journal_wipe() - Wipe journal contents |
470decc6 DK |
2031 | * @journal: Journal to act on. |
2032 | * @write: flag (see below) | |
2033 | * | |
2034 | * Wipe out all of the contents of a journal, safely. This will produce | |
2035 | * a warning if the journal contains any valid recovery information. | |
f7f4bccb | 2036 | * Must be called between journal_init_*() and jbd2_journal_load(). |
470decc6 DK |
2037 | * |
2038 | * If 'write' is non-zero, then we wipe out the journal on disk; otherwise | |
2039 | * we merely suppress recovery. | |
2040 | */ | |
2041 | ||
f7f4bccb | 2042 | int jbd2_journal_wipe(journal_t *journal, int write) |
470decc6 | 2043 | { |
470decc6 DK |
2044 | int err = 0; |
2045 | ||
f7f4bccb | 2046 | J_ASSERT (!(journal->j_flags & JBD2_LOADED)); |
470decc6 DK |
2047 | |
2048 | err = load_superblock(journal); | |
2049 | if (err) | |
2050 | return err; | |
2051 | ||
470decc6 DK |
2052 | if (!journal->j_tail) |
2053 | goto no_recovery; | |
2054 | ||
f2a44523 | 2055 | printk(KERN_WARNING "JBD2: %s recovery information on journal\n", |
470decc6 DK |
2056 | write ? "Clearing" : "Ignoring"); |
2057 | ||
f7f4bccb | 2058 | err = jbd2_journal_skip_recovery(journal); |
a78bb11d JK |
2059 | if (write) { |
2060 | /* Lock to make assertions happy... */ | |
53cf9784 | 2061 | mutex_lock_io(&journal->j_checkpoint_mutex); |
17f423b5 | 2062 | jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA); |
a78bb11d JK |
2063 | mutex_unlock(&journal->j_checkpoint_mutex); |
2064 | } | |
470decc6 DK |
2065 | |
2066 | no_recovery: | |
2067 | return err; | |
2068 | } | |
2069 | ||
470decc6 DK |
2070 | /* |
2071 | * Journal abort has very specific semantics, which we describe | |
2072 | * for journal abort. | |
2073 | * | |
bfcd3555 | 2074 | * Two internal functions, which provide abort to the jbd layer |
470decc6 DK |
2075 | * itself are here. |
2076 | */ | |
2077 | ||
2078 | /* | |
2079 | * Quick version for internal journal use (doesn't lock the journal). | |
2080 | * Aborts hard --- we mark the abort as occurred, but do _nothing_ else, | |
2081 | * and don't attempt to make any other journal updates. | |
2082 | */ | |
f7f4bccb | 2083 | void __jbd2_journal_abort_hard(journal_t *journal) |
470decc6 DK |
2084 | { |
2085 | transaction_t *transaction; | |
470decc6 | 2086 | |
f7f4bccb | 2087 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2088 | return; |
2089 | ||
2090 | printk(KERN_ERR "Aborting journal on device %s.\n", | |
05496769 | 2091 | journal->j_devname); |
470decc6 | 2092 | |
a931da6a | 2093 | write_lock(&journal->j_state_lock); |
f7f4bccb | 2094 | journal->j_flags |= JBD2_ABORT; |
470decc6 DK |
2095 | transaction = journal->j_running_transaction; |
2096 | if (transaction) | |
f7f4bccb | 2097 | __jbd2_log_start_commit(journal, transaction->t_tid); |
a931da6a | 2098 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2099 | } |
2100 | ||
2101 | /* Soft abort: record the abort error status in the journal superblock, | |
2102 | * but don't do any other IO. */ | |
2103 | static void __journal_abort_soft (journal_t *journal, int errno) | |
2104 | { | |
fb7c0244 | 2105 | int old_errno; |
470decc6 | 2106 | |
fb7c0244 TT |
2107 | write_lock(&journal->j_state_lock); |
2108 | old_errno = journal->j_errno; | |
2109 | if (!journal->j_errno || errno == -ESHUTDOWN) | |
470decc6 DK |
2110 | journal->j_errno = errno; |
2111 | ||
fb7c0244 TT |
2112 | if (journal->j_flags & JBD2_ABORT) { |
2113 | write_unlock(&journal->j_state_lock); | |
2114 | if (!old_errno && old_errno != -ESHUTDOWN && | |
2115 | errno == -ESHUTDOWN) | |
2116 | jbd2_journal_update_sb_errno(journal); | |
2117 | return; | |
2118 | } | |
2119 | write_unlock(&journal->j_state_lock); | |
2120 | ||
f7f4bccb | 2121 | __jbd2_journal_abort_hard(journal); |
470decc6 | 2122 | |
4327ba52 | 2123 | if (errno) { |
24bcc89c | 2124 | jbd2_journal_update_sb_errno(journal); |
4327ba52 DJ |
2125 | write_lock(&journal->j_state_lock); |
2126 | journal->j_flags |= JBD2_REC_ERR; | |
2127 | write_unlock(&journal->j_state_lock); | |
2128 | } | |
470decc6 DK |
2129 | } |
2130 | ||
2131 | /** | |
f7f4bccb | 2132 | * void jbd2_journal_abort () - Shutdown the journal immediately. |
470decc6 DK |
2133 | * @journal: the journal to shutdown. |
2134 | * @errno: an error number to record in the journal indicating | |
2135 | * the reason for the shutdown. | |
2136 | * | |
2137 | * Perform a complete, immediate shutdown of the ENTIRE | |
2138 | * journal (not of a single transaction). This operation cannot be | |
2139 | * undone without closing and reopening the journal. | |
2140 | * | |
f7f4bccb | 2141 | * The jbd2_journal_abort function is intended to support higher level error |
470decc6 DK |
2142 | * recovery mechanisms such as the ext2/ext3 remount-readonly error |
2143 | * mode. | |
2144 | * | |
2145 | * Journal abort has very specific semantics. Any existing dirty, | |
2146 | * unjournaled buffers in the main filesystem will still be written to | |
2147 | * disk by bdflush, but the journaling mechanism will be suspended | |
2148 | * immediately and no further transaction commits will be honoured. | |
2149 | * | |
2150 | * Any dirty, journaled buffers will be written back to disk without | |
2151 | * hitting the journal. Atomicity cannot be guaranteed on an aborted | |
2152 | * filesystem, but we _do_ attempt to leave as much data as possible | |
2153 | * behind for fsck to use for cleanup. | |
2154 | * | |
2155 | * Any attempt to get a new transaction handle on a journal which is in | |
2156 | * ABORT state will just result in an -EROFS error return. A | |
f7f4bccb | 2157 | * jbd2_journal_stop on an existing handle will return -EIO if we have |
470decc6 DK |
2158 | * entered abort state during the update. |
2159 | * | |
2160 | * Recursive transactions are not disturbed by journal abort until the | |
f7f4bccb | 2161 | * final jbd2_journal_stop, which will receive the -EIO error. |
470decc6 | 2162 | * |
f7f4bccb | 2163 | * Finally, the jbd2_journal_abort call allows the caller to supply an errno |
470decc6 DK |
2164 | * which will be recorded (if possible) in the journal superblock. This |
2165 | * allows a client to record failure conditions in the middle of a | |
2166 | * transaction without having to complete the transaction to record the | |
2167 | * failure to disk. ext3_error, for example, now uses this | |
2168 | * functionality. | |
2169 | * | |
2170 | * Errors which originate from within the journaling layer will NOT | |
2171 | * supply an errno; a null errno implies that absolutely no further | |
2172 | * writes are done to the journal (unless there are any already in | |
2173 | * progress). | |
2174 | * | |
2175 | */ | |
2176 | ||
f7f4bccb | 2177 | void jbd2_journal_abort(journal_t *journal, int errno) |
470decc6 DK |
2178 | { |
2179 | __journal_abort_soft(journal, errno); | |
2180 | } | |
2181 | ||
2182 | /** | |
f7f4bccb | 2183 | * int jbd2_journal_errno () - returns the journal's error state. |
470decc6 DK |
2184 | * @journal: journal to examine. |
2185 | * | |
bfcd3555 | 2186 | * This is the errno number set with jbd2_journal_abort(), the last |
470decc6 DK |
2187 | * time the journal was mounted - if the journal was stopped |
2188 | * without calling abort this will be 0. | |
2189 | * | |
2190 | * If the journal has been aborted on this mount time -EROFS will | |
2191 | * be returned. | |
2192 | */ | |
f7f4bccb | 2193 | int jbd2_journal_errno(journal_t *journal) |
470decc6 DK |
2194 | { |
2195 | int err; | |
2196 | ||
a931da6a | 2197 | read_lock(&journal->j_state_lock); |
f7f4bccb | 2198 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2199 | err = -EROFS; |
2200 | else | |
2201 | err = journal->j_errno; | |
a931da6a | 2202 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
2203 | return err; |
2204 | } | |
2205 | ||
2206 | /** | |
f7f4bccb | 2207 | * int jbd2_journal_clear_err () - clears the journal's error state |
470decc6 DK |
2208 | * @journal: journal to act on. |
2209 | * | |
bfcd3555 | 2210 | * An error must be cleared or acked to take a FS out of readonly |
470decc6 DK |
2211 | * mode. |
2212 | */ | |
f7f4bccb | 2213 | int jbd2_journal_clear_err(journal_t *journal) |
470decc6 DK |
2214 | { |
2215 | int err = 0; | |
2216 | ||
a931da6a | 2217 | write_lock(&journal->j_state_lock); |
f7f4bccb | 2218 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2219 | err = -EROFS; |
2220 | else | |
2221 | journal->j_errno = 0; | |
a931da6a | 2222 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2223 | return err; |
2224 | } | |
2225 | ||
2226 | /** | |
f7f4bccb | 2227 | * void jbd2_journal_ack_err() - Ack journal err. |
470decc6 DK |
2228 | * @journal: journal to act on. |
2229 | * | |
bfcd3555 | 2230 | * An error must be cleared or acked to take a FS out of readonly |
470decc6 DK |
2231 | * mode. |
2232 | */ | |
f7f4bccb | 2233 | void jbd2_journal_ack_err(journal_t *journal) |
470decc6 | 2234 | { |
a931da6a | 2235 | write_lock(&journal->j_state_lock); |
470decc6 | 2236 | if (journal->j_errno) |
f7f4bccb | 2237 | journal->j_flags |= JBD2_ACK_ERR; |
a931da6a | 2238 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2239 | } |
2240 | ||
f7f4bccb | 2241 | int jbd2_journal_blocks_per_page(struct inode *inode) |
470decc6 | 2242 | { |
09cbfeaf | 2243 | return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
470decc6 DK |
2244 | } |
2245 | ||
b517bea1 ZB |
2246 | /* |
2247 | * helper functions to deal with 32 or 64bit block numbers. | |
2248 | */ | |
2249 | size_t journal_tag_bytes(journal_t *journal) | |
2250 | { | |
db9ee220 DW |
2251 | size_t sz; |
2252 | ||
56316a0d | 2253 | if (jbd2_has_feature_csum3(journal)) |
db9ee220 DW |
2254 | return sizeof(journal_block_tag3_t); |
2255 | ||
2256 | sz = sizeof(journal_block_tag_t); | |
c3900875 | 2257 | |
56316a0d | 2258 | if (jbd2_has_feature_csum2(journal)) |
db9ee220 | 2259 | sz += sizeof(__u16); |
c3900875 | 2260 | |
56316a0d | 2261 | if (jbd2_has_feature_64bit(journal)) |
db9ee220 | 2262 | return sz; |
b517bea1 | 2263 | else |
db9ee220 | 2264 | return sz - sizeof(__u32); |
b517bea1 ZB |
2265 | } |
2266 | ||
d2eecb03 TT |
2267 | /* |
2268 | * JBD memory management | |
2269 | * | |
2270 | * These functions are used to allocate block-sized chunks of memory | |
2271 | * used for making copies of buffer_head data. Very often it will be | |
2272 | * page-sized chunks of data, but sometimes it will be in | |
2273 | * sub-page-size chunks. (For example, 16k pages on Power systems | |
2274 | * with a 4k block file system.) For blocks smaller than a page, we | |
2275 | * use a SLAB allocator. There are slab caches for each block size, | |
2276 | * which are allocated at mount time, if necessary, and we only free | |
2277 | * (all of) the slab caches when/if the jbd2 module is unloaded. For | |
2278 | * this reason we don't need to a mutex to protect access to | |
2279 | * jbd2_slab[] allocating or releasing memory; only in | |
2280 | * jbd2_journal_create_slab(). | |
2281 | */ | |
2282 | #define JBD2_MAX_SLABS 8 | |
2283 | static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS]; | |
d2eecb03 TT |
2284 | |
2285 | static const char *jbd2_slab_names[JBD2_MAX_SLABS] = { | |
2286 | "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k", | |
2287 | "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k" | |
2288 | }; | |
2289 | ||
2290 | ||
2291 | static void jbd2_journal_destroy_slabs(void) | |
2292 | { | |
2293 | int i; | |
2294 | ||
2295 | for (i = 0; i < JBD2_MAX_SLABS; i++) { | |
8bdd5b60 | 2296 | kmem_cache_destroy(jbd2_slab[i]); |
d2eecb03 TT |
2297 | jbd2_slab[i] = NULL; |
2298 | } | |
2299 | } | |
2300 | ||
2301 | static int jbd2_journal_create_slab(size_t size) | |
2302 | { | |
51dfacde | 2303 | static DEFINE_MUTEX(jbd2_slab_create_mutex); |
d2eecb03 TT |
2304 | int i = order_base_2(size) - 10; |
2305 | size_t slab_size; | |
2306 | ||
2307 | if (size == PAGE_SIZE) | |
2308 | return 0; | |
2309 | ||
2310 | if (i >= JBD2_MAX_SLABS) | |
2311 | return -EINVAL; | |
2312 | ||
2313 | if (unlikely(i < 0)) | |
2314 | i = 0; | |
51dfacde | 2315 | mutex_lock(&jbd2_slab_create_mutex); |
d2eecb03 | 2316 | if (jbd2_slab[i]) { |
51dfacde | 2317 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2318 | return 0; /* Already created */ |
2319 | } | |
2320 | ||
2321 | slab_size = 1 << (i+10); | |
2322 | jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size, | |
2323 | slab_size, 0, NULL); | |
51dfacde | 2324 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2325 | if (!jbd2_slab[i]) { |
2326 | printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n"); | |
2327 | return -ENOMEM; | |
2328 | } | |
2329 | return 0; | |
2330 | } | |
2331 | ||
2332 | static struct kmem_cache *get_slab(size_t size) | |
2333 | { | |
2334 | int i = order_base_2(size) - 10; | |
2335 | ||
2336 | BUG_ON(i >= JBD2_MAX_SLABS); | |
2337 | if (unlikely(i < 0)) | |
2338 | i = 0; | |
8ac97b74 | 2339 | BUG_ON(jbd2_slab[i] == NULL); |
d2eecb03 TT |
2340 | return jbd2_slab[i]; |
2341 | } | |
2342 | ||
2343 | void *jbd2_alloc(size_t size, gfp_t flags) | |
2344 | { | |
2345 | void *ptr; | |
2346 | ||
2347 | BUG_ON(size & (size-1)); /* Must be a power of 2 */ | |
2348 | ||
f2db1971 | 2349 | if (size < PAGE_SIZE) |
d2eecb03 | 2350 | ptr = kmem_cache_alloc(get_slab(size), flags); |
f2db1971 MH |
2351 | else |
2352 | ptr = (void *)__get_free_pages(flags, get_order(size)); | |
d2eecb03 TT |
2353 | |
2354 | /* Check alignment; SLUB has gotten this wrong in the past, | |
2355 | * and this can lead to user data corruption! */ | |
2356 | BUG_ON(((unsigned long) ptr) & (size-1)); | |
2357 | ||
2358 | return ptr; | |
2359 | } | |
2360 | ||
2361 | void jbd2_free(void *ptr, size_t size) | |
2362 | { | |
f2db1971 MH |
2363 | if (size < PAGE_SIZE) |
2364 | kmem_cache_free(get_slab(size), ptr); | |
2365 | else | |
2366 | free_pages((unsigned long)ptr, get_order(size)); | |
d2eecb03 TT |
2367 | }; |
2368 | ||
470decc6 DK |
2369 | /* |
2370 | * Journal_head storage management | |
2371 | */ | |
e18b890b | 2372 | static struct kmem_cache *jbd2_journal_head_cache; |
e23291b9 | 2373 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2374 | static atomic_t nr_journal_heads = ATOMIC_INIT(0); |
2375 | #endif | |
2376 | ||
0d52154b | 2377 | static int __init jbd2_journal_init_journal_head_cache(void) |
470decc6 | 2378 | { |
0d52154b | 2379 | J_ASSERT(!jbd2_journal_head_cache); |
a920e941 | 2380 | jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", |
470decc6 DK |
2381 | sizeof(struct journal_head), |
2382 | 0, /* offset */ | |
5f0d5a3a | 2383 | SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU, |
20c2df83 | 2384 | NULL); /* ctor */ |
1076d17a | 2385 | if (!jbd2_journal_head_cache) { |
f2a44523 | 2386 | printk(KERN_EMERG "JBD2: no memory for journal_head cache\n"); |
0d52154b | 2387 | return -ENOMEM; |
470decc6 | 2388 | } |
0d52154b | 2389 | return 0; |
470decc6 DK |
2390 | } |
2391 | ||
4185a2ac | 2392 | static void jbd2_journal_destroy_journal_head_cache(void) |
470decc6 | 2393 | { |
8bdd5b60 WL |
2394 | kmem_cache_destroy(jbd2_journal_head_cache); |
2395 | jbd2_journal_head_cache = NULL; | |
470decc6 DK |
2396 | } |
2397 | ||
2398 | /* | |
2399 | * journal_head splicing and dicing | |
2400 | */ | |
2401 | static struct journal_head *journal_alloc_journal_head(void) | |
2402 | { | |
2403 | struct journal_head *ret; | |
470decc6 | 2404 | |
e23291b9 | 2405 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2406 | atomic_inc(&nr_journal_heads); |
2407 | #endif | |
5d9cf9c6 | 2408 | ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); |
1076d17a | 2409 | if (!ret) { |
470decc6 | 2410 | jbd_debug(1, "out of memory for journal_head\n"); |
670be5a7 | 2411 | pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__); |
7b506b10 MH |
2412 | ret = kmem_cache_zalloc(jbd2_journal_head_cache, |
2413 | GFP_NOFS | __GFP_NOFAIL); | |
470decc6 DK |
2414 | } |
2415 | return ret; | |
2416 | } | |
2417 | ||
2418 | static void journal_free_journal_head(struct journal_head *jh) | |
2419 | { | |
e23291b9 | 2420 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 | 2421 | atomic_dec(&nr_journal_heads); |
cd02ff0b | 2422 | memset(jh, JBD2_POISON_FREE, sizeof(*jh)); |
470decc6 | 2423 | #endif |
f7f4bccb | 2424 | kmem_cache_free(jbd2_journal_head_cache, jh); |
470decc6 DK |
2425 | } |
2426 | ||
2427 | /* | |
2428 | * A journal_head is attached to a buffer_head whenever JBD has an | |
2429 | * interest in the buffer. | |
2430 | * | |
2431 | * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit | |
2432 | * is set. This bit is tested in core kernel code where we need to take | |
2433 | * JBD-specific actions. Testing the zeroness of ->b_private is not reliable | |
2434 | * there. | |
2435 | * | |
2436 | * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. | |
2437 | * | |
2438 | * When a buffer has its BH_JBD bit set it is immune from being released by | |
2439 | * core kernel code, mainly via ->b_count. | |
2440 | * | |
de1b7941 JK |
2441 | * A journal_head is detached from its buffer_head when the journal_head's |
2442 | * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint | |
2443 | * transaction (b_cp_transaction) hold their references to b_jcount. | |
470decc6 DK |
2444 | * |
2445 | * Various places in the kernel want to attach a journal_head to a buffer_head | |
2446 | * _before_ attaching the journal_head to a transaction. To protect the | |
f7f4bccb | 2447 | * journal_head in this situation, jbd2_journal_add_journal_head elevates the |
470decc6 | 2448 | * journal_head's b_jcount refcount by one. The caller must call |
f7f4bccb | 2449 | * jbd2_journal_put_journal_head() to undo this. |
470decc6 DK |
2450 | * |
2451 | * So the typical usage would be: | |
2452 | * | |
2453 | * (Attach a journal_head if needed. Increments b_jcount) | |
f7f4bccb | 2454 | * struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 | 2455 | * ... |
de1b7941 JK |
2456 | * (Get another reference for transaction) |
2457 | * jbd2_journal_grab_journal_head(bh); | |
470decc6 | 2458 | * jh->b_transaction = xxx; |
de1b7941 | 2459 | * (Put original reference) |
f7f4bccb | 2460 | * jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2461 | */ |
2462 | ||
2463 | /* | |
2464 | * Give a buffer_head a journal_head. | |
2465 | * | |
470decc6 DK |
2466 | * May sleep. |
2467 | */ | |
f7f4bccb | 2468 | struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) |
470decc6 DK |
2469 | { |
2470 | struct journal_head *jh; | |
2471 | struct journal_head *new_jh = NULL; | |
2472 | ||
2473 | repeat: | |
5d9cf9c6 | 2474 | if (!buffer_jbd(bh)) |
470decc6 | 2475 | new_jh = journal_alloc_journal_head(); |
470decc6 DK |
2476 | |
2477 | jbd_lock_bh_journal_head(bh); | |
2478 | if (buffer_jbd(bh)) { | |
2479 | jh = bh2jh(bh); | |
2480 | } else { | |
2481 | J_ASSERT_BH(bh, | |
2482 | (atomic_read(&bh->b_count) > 0) || | |
2483 | (bh->b_page && bh->b_page->mapping)); | |
2484 | ||
2485 | if (!new_jh) { | |
2486 | jbd_unlock_bh_journal_head(bh); | |
2487 | goto repeat; | |
2488 | } | |
2489 | ||
2490 | jh = new_jh; | |
2491 | new_jh = NULL; /* We consumed it */ | |
2492 | set_buffer_jbd(bh); | |
2493 | bh->b_private = jh; | |
2494 | jh->b_bh = bh; | |
2495 | get_bh(bh); | |
2496 | BUFFER_TRACE(bh, "added journal_head"); | |
2497 | } | |
2498 | jh->b_jcount++; | |
2499 | jbd_unlock_bh_journal_head(bh); | |
2500 | if (new_jh) | |
2501 | journal_free_journal_head(new_jh); | |
2502 | return bh->b_private; | |
2503 | } | |
2504 | ||
2505 | /* | |
2506 | * Grab a ref against this buffer_head's journal_head. If it ended up not | |
2507 | * having a journal_head, return NULL | |
2508 | */ | |
f7f4bccb | 2509 | struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) |
470decc6 DK |
2510 | { |
2511 | struct journal_head *jh = NULL; | |
2512 | ||
2513 | jbd_lock_bh_journal_head(bh); | |
2514 | if (buffer_jbd(bh)) { | |
2515 | jh = bh2jh(bh); | |
2516 | jh->b_jcount++; | |
2517 | } | |
2518 | jbd_unlock_bh_journal_head(bh); | |
2519 | return jh; | |
2520 | } | |
2521 | ||
2522 | static void __journal_remove_journal_head(struct buffer_head *bh) | |
2523 | { | |
2524 | struct journal_head *jh = bh2jh(bh); | |
2525 | ||
2526 | J_ASSERT_JH(jh, jh->b_jcount >= 0); | |
de1b7941 JK |
2527 | J_ASSERT_JH(jh, jh->b_transaction == NULL); |
2528 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
2529 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
2530 | J_ASSERT_JH(jh, jh->b_jlist == BJ_None); | |
2531 | J_ASSERT_BH(bh, buffer_jbd(bh)); | |
2532 | J_ASSERT_BH(bh, jh2bh(jh) == bh); | |
2533 | BUFFER_TRACE(bh, "remove journal_head"); | |
2534 | if (jh->b_frozen_data) { | |
2535 | printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__); | |
2536 | jbd2_free(jh->b_frozen_data, bh->b_size); | |
470decc6 | 2537 | } |
de1b7941 JK |
2538 | if (jh->b_committed_data) { |
2539 | printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__); | |
2540 | jbd2_free(jh->b_committed_data, bh->b_size); | |
2541 | } | |
2542 | bh->b_private = NULL; | |
2543 | jh->b_bh = NULL; /* debug, really */ | |
2544 | clear_buffer_jbd(bh); | |
2545 | journal_free_journal_head(jh); | |
470decc6 DK |
2546 | } |
2547 | ||
2548 | /* | |
de1b7941 | 2549 | * Drop a reference on the passed journal_head. If it fell to zero then |
470decc6 DK |
2550 | * release the journal_head from the buffer_head. |
2551 | */ | |
f7f4bccb | 2552 | void jbd2_journal_put_journal_head(struct journal_head *jh) |
470decc6 DK |
2553 | { |
2554 | struct buffer_head *bh = jh2bh(jh); | |
2555 | ||
2556 | jbd_lock_bh_journal_head(bh); | |
2557 | J_ASSERT_JH(jh, jh->b_jcount > 0); | |
2558 | --jh->b_jcount; | |
de1b7941 | 2559 | if (!jh->b_jcount) { |
470decc6 | 2560 | __journal_remove_journal_head(bh); |
de1b7941 | 2561 | jbd_unlock_bh_journal_head(bh); |
470decc6 | 2562 | __brelse(bh); |
de1b7941 JK |
2563 | } else |
2564 | jbd_unlock_bh_journal_head(bh); | |
470decc6 DK |
2565 | } |
2566 | ||
c851ed54 JK |
2567 | /* |
2568 | * Initialize jbd inode head | |
2569 | */ | |
2570 | void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) | |
2571 | { | |
2572 | jinode->i_transaction = NULL; | |
2573 | jinode->i_next_transaction = NULL; | |
2574 | jinode->i_vfs_inode = inode; | |
2575 | jinode->i_flags = 0; | |
6ba0e7dc RZ |
2576 | jinode->i_dirty_start = 0; |
2577 | jinode->i_dirty_end = 0; | |
c851ed54 JK |
2578 | INIT_LIST_HEAD(&jinode->i_list); |
2579 | } | |
2580 | ||
2581 | /* | |
2582 | * Function to be called before we start removing inode from memory (i.e., | |
2583 | * clear_inode() is a fine place to be called from). It removes inode from | |
2584 | * transaction's lists. | |
2585 | */ | |
2586 | void jbd2_journal_release_jbd_inode(journal_t *journal, | |
2587 | struct jbd2_inode *jinode) | |
2588 | { | |
c851ed54 JK |
2589 | if (!journal) |
2590 | return; | |
2591 | restart: | |
2592 | spin_lock(&journal->j_list_lock); | |
2593 | /* Is commit writing out inode - we have to wait */ | |
cb0d9d47 | 2594 | if (jinode->i_flags & JI_COMMIT_RUNNING) { |
c851ed54 JK |
2595 | wait_queue_head_t *wq; |
2596 | DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); | |
2597 | wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); | |
21417136 | 2598 | prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
c851ed54 JK |
2599 | spin_unlock(&journal->j_list_lock); |
2600 | schedule(); | |
21417136 | 2601 | finish_wait(wq, &wait.wq_entry); |
c851ed54 JK |
2602 | goto restart; |
2603 | } | |
2604 | ||
c851ed54 JK |
2605 | if (jinode->i_transaction) { |
2606 | list_del(&jinode->i_list); | |
2607 | jinode->i_transaction = NULL; | |
2608 | } | |
2609 | spin_unlock(&journal->j_list_lock); | |
2610 | } | |
2611 | ||
470decc6 | 2612 | |
8e85fb3f JL |
2613 | #ifdef CONFIG_PROC_FS |
2614 | ||
2615 | #define JBD2_STATS_PROC_NAME "fs/jbd2" | |
2616 | ||
2617 | static void __init jbd2_create_jbd_stats_proc_entry(void) | |
2618 | { | |
2619 | proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); | |
2620 | } | |
2621 | ||
2622 | static void __exit jbd2_remove_jbd_stats_proc_entry(void) | |
2623 | { | |
2624 | if (proc_jbd2_stats) | |
2625 | remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); | |
2626 | } | |
2627 | ||
2628 | #else | |
2629 | ||
2630 | #define jbd2_create_jbd_stats_proc_entry() do {} while (0) | |
2631 | #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) | |
2632 | ||
2633 | #endif | |
2634 | ||
8aefcd55 | 2635 | struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache; |
470decc6 | 2636 | |
0d52154b CX |
2637 | static int __init jbd2_journal_init_inode_cache(void) |
2638 | { | |
2639 | J_ASSERT(!jbd2_inode_cache); | |
2640 | jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0); | |
2641 | if (!jbd2_inode_cache) { | |
2642 | pr_emerg("JBD2: failed to create inode cache\n"); | |
2643 | return -ENOMEM; | |
2644 | } | |
2645 | return 0; | |
2646 | } | |
2647 | ||
4185a2ac | 2648 | static int __init jbd2_journal_init_handle_cache(void) |
470decc6 | 2649 | { |
0d52154b | 2650 | J_ASSERT(!jbd2_handle_cache); |
8aefcd55 | 2651 | jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY); |
0d52154b | 2652 | if (!jbd2_handle_cache) { |
8aefcd55 TT |
2653 | printk(KERN_EMERG "JBD2: failed to create handle cache\n"); |
2654 | return -ENOMEM; | |
2655 | } | |
470decc6 DK |
2656 | return 0; |
2657 | } | |
2658 | ||
0d52154b CX |
2659 | static void jbd2_journal_destroy_inode_cache(void) |
2660 | { | |
2661 | kmem_cache_destroy(jbd2_inode_cache); | |
2662 | jbd2_inode_cache = NULL; | |
2663 | } | |
2664 | ||
f7f4bccb | 2665 | static void jbd2_journal_destroy_handle_cache(void) |
470decc6 | 2666 | { |
8bdd5b60 WL |
2667 | kmem_cache_destroy(jbd2_handle_cache); |
2668 | jbd2_handle_cache = NULL; | |
470decc6 DK |
2669 | } |
2670 | ||
2671 | /* | |
2672 | * Module startup and shutdown | |
2673 | */ | |
2674 | ||
2675 | static int __init journal_init_caches(void) | |
2676 | { | |
2677 | int ret; | |
2678 | ||
0d52154b CX |
2679 | ret = jbd2_journal_init_revoke_record_cache(); |
2680 | if (ret == 0) | |
2681 | ret = jbd2_journal_init_revoke_table_cache(); | |
470decc6 | 2682 | if (ret == 0) |
4185a2ac | 2683 | ret = jbd2_journal_init_journal_head_cache(); |
470decc6 | 2684 | if (ret == 0) |
4185a2ac | 2685 | ret = jbd2_journal_init_handle_cache(); |
0d52154b CX |
2686 | if (ret == 0) |
2687 | ret = jbd2_journal_init_inode_cache(); | |
470decc6 | 2688 | if (ret == 0) |
0c2022ec | 2689 | ret = jbd2_journal_init_transaction_cache(); |
470decc6 DK |
2690 | return ret; |
2691 | } | |
2692 | ||
f7f4bccb | 2693 | static void jbd2_journal_destroy_caches(void) |
470decc6 | 2694 | { |
0d52154b CX |
2695 | jbd2_journal_destroy_revoke_record_cache(); |
2696 | jbd2_journal_destroy_revoke_table_cache(); | |
4185a2ac | 2697 | jbd2_journal_destroy_journal_head_cache(); |
f7f4bccb | 2698 | jbd2_journal_destroy_handle_cache(); |
0d52154b | 2699 | jbd2_journal_destroy_inode_cache(); |
0c2022ec | 2700 | jbd2_journal_destroy_transaction_cache(); |
d2eecb03 | 2701 | jbd2_journal_destroy_slabs(); |
470decc6 DK |
2702 | } |
2703 | ||
2704 | static int __init journal_init(void) | |
2705 | { | |
2706 | int ret; | |
2707 | ||
2708 | BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); | |
2709 | ||
2710 | ret = journal_init_caches(); | |
620de4e1 | 2711 | if (ret == 0) { |
620de4e1 DG |
2712 | jbd2_create_jbd_stats_proc_entry(); |
2713 | } else { | |
f7f4bccb | 2714 | jbd2_journal_destroy_caches(); |
620de4e1 | 2715 | } |
470decc6 DK |
2716 | return ret; |
2717 | } | |
2718 | ||
2719 | static void __exit journal_exit(void) | |
2720 | { | |
e23291b9 | 2721 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2722 | int n = atomic_read(&nr_journal_heads); |
2723 | if (n) | |
75685071 | 2724 | printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n); |
470decc6 | 2725 | #endif |
8e85fb3f | 2726 | jbd2_remove_jbd_stats_proc_entry(); |
f7f4bccb | 2727 | jbd2_journal_destroy_caches(); |
470decc6 DK |
2728 | } |
2729 | ||
2730 | MODULE_LICENSE("GPL"); | |
2731 | module_init(journal_init); | |
2732 | module_exit(journal_exit); | |
2733 |