]>
Commit | Line | Data |
---|---|---|
470decc6 | 1 | /* |
f7f4bccb | 2 | * linux/fs/jbd2/journal.c |
470decc6 DK |
3 | * |
4 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1998 | |
5 | * | |
6 | * Copyright 1998 Red Hat corp --- All Rights Reserved | |
7 | * | |
8 | * This file is part of the Linux kernel and is made available under | |
9 | * the terms of the GNU General Public License, version 2, or at your | |
10 | * option, any later version, incorporated herein by reference. | |
11 | * | |
12 | * Generic filesystem journal-writing code; part of the ext2fs | |
13 | * journaling system. | |
14 | * | |
15 | * This file manages journals: areas of disk reserved for logging | |
16 | * transactional updates. This includes the kernel journaling thread | |
17 | * which is responsible for scheduling updates to the log. | |
18 | * | |
19 | * We do not actually manage the physical storage of the journal in this | |
20 | * file: that is left to a per-journal policy function, which allows us | |
21 | * to store the journal within a filesystem-specified area for ext2 | |
22 | * journaling (ext2 can use a reserved inode for storing the log). | |
23 | */ | |
24 | ||
25 | #include <linux/module.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/fs.h> | |
f7f4bccb | 28 | #include <linux/jbd2.h> |
470decc6 DK |
29 | #include <linux/errno.h> |
30 | #include <linux/slab.h> | |
470decc6 DK |
31 | #include <linux/init.h> |
32 | #include <linux/mm.h> | |
7dfb7103 | 33 | #include <linux/freezer.h> |
470decc6 DK |
34 | #include <linux/pagemap.h> |
35 | #include <linux/kthread.h> | |
36 | #include <linux/poison.h> | |
37 | #include <linux/proc_fs.h> | |
8e85fb3f | 38 | #include <linux/seq_file.h> |
c225aa57 | 39 | #include <linux/math64.h> |
879c5e6b | 40 | #include <linux/hash.h> |
d2eecb03 TT |
41 | #include <linux/log2.h> |
42 | #include <linux/vmalloc.h> | |
47def826 | 43 | #include <linux/backing-dev.h> |
39e3ac25 | 44 | #include <linux/bitops.h> |
670be5a7 | 45 | #include <linux/ratelimit.h> |
eb52da3f | 46 | #include <linux/sched/mm.h> |
879c5e6b TT |
47 | |
48 | #define CREATE_TRACE_POINTS | |
49 | #include <trace/events/jbd2.h> | |
470decc6 | 50 | |
7c0f6ba6 | 51 | #include <linux/uaccess.h> |
470decc6 DK |
52 | #include <asm/page.h> |
53 | ||
b6e96d00 TT |
54 | #ifdef CONFIG_JBD2_DEBUG |
55 | ushort jbd2_journal_enable_debug __read_mostly; | |
56 | EXPORT_SYMBOL(jbd2_journal_enable_debug); | |
57 | ||
58 | module_param_named(jbd2_debug, jbd2_journal_enable_debug, ushort, 0644); | |
59 | MODULE_PARM_DESC(jbd2_debug, "Debugging level for jbd2"); | |
60 | #endif | |
61 | ||
f7f4bccb MC |
62 | EXPORT_SYMBOL(jbd2_journal_extend); |
63 | EXPORT_SYMBOL(jbd2_journal_stop); | |
64 | EXPORT_SYMBOL(jbd2_journal_lock_updates); | |
65 | EXPORT_SYMBOL(jbd2_journal_unlock_updates); | |
66 | EXPORT_SYMBOL(jbd2_journal_get_write_access); | |
67 | EXPORT_SYMBOL(jbd2_journal_get_create_access); | |
68 | EXPORT_SYMBOL(jbd2_journal_get_undo_access); | |
e06c8227 | 69 | EXPORT_SYMBOL(jbd2_journal_set_triggers); |
f7f4bccb | 70 | EXPORT_SYMBOL(jbd2_journal_dirty_metadata); |
f7f4bccb | 71 | EXPORT_SYMBOL(jbd2_journal_forget); |
470decc6 DK |
72 | #if 0 |
73 | EXPORT_SYMBOL(journal_sync_buffer); | |
74 | #endif | |
f7f4bccb MC |
75 | EXPORT_SYMBOL(jbd2_journal_flush); |
76 | EXPORT_SYMBOL(jbd2_journal_revoke); | |
77 | ||
78 | EXPORT_SYMBOL(jbd2_journal_init_dev); | |
79 | EXPORT_SYMBOL(jbd2_journal_init_inode); | |
f7f4bccb MC |
80 | EXPORT_SYMBOL(jbd2_journal_check_used_features); |
81 | EXPORT_SYMBOL(jbd2_journal_check_available_features); | |
82 | EXPORT_SYMBOL(jbd2_journal_set_features); | |
f7f4bccb MC |
83 | EXPORT_SYMBOL(jbd2_journal_load); |
84 | EXPORT_SYMBOL(jbd2_journal_destroy); | |
f7f4bccb MC |
85 | EXPORT_SYMBOL(jbd2_journal_abort); |
86 | EXPORT_SYMBOL(jbd2_journal_errno); | |
87 | EXPORT_SYMBOL(jbd2_journal_ack_err); | |
88 | EXPORT_SYMBOL(jbd2_journal_clear_err); | |
89 | EXPORT_SYMBOL(jbd2_log_wait_commit); | |
3b799d15 | 90 | EXPORT_SYMBOL(jbd2_log_start_commit); |
f7f4bccb MC |
91 | EXPORT_SYMBOL(jbd2_journal_start_commit); |
92 | EXPORT_SYMBOL(jbd2_journal_force_commit_nested); | |
93 | EXPORT_SYMBOL(jbd2_journal_wipe); | |
94 | EXPORT_SYMBOL(jbd2_journal_blocks_per_page); | |
95 | EXPORT_SYMBOL(jbd2_journal_invalidatepage); | |
96 | EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); | |
97 | EXPORT_SYMBOL(jbd2_journal_force_commit); | |
41617e1a JK |
98 | EXPORT_SYMBOL(jbd2_journal_inode_add_write); |
99 | EXPORT_SYMBOL(jbd2_journal_inode_add_wait); | |
c851ed54 JK |
100 | EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); |
101 | EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); | |
102 | EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); | |
8aefcd55 | 103 | EXPORT_SYMBOL(jbd2_inode_cache); |
470decc6 | 104 | |
470decc6 | 105 | static void __journal_abort_soft (journal_t *journal, int errno); |
d2eecb03 | 106 | static int jbd2_journal_create_slab(size_t slab_size); |
470decc6 | 107 | |
169f1a2a PG |
108 | #ifdef CONFIG_JBD2_DEBUG |
109 | void __jbd2_debug(int level, const char *file, const char *func, | |
110 | unsigned int line, const char *fmt, ...) | |
111 | { | |
112 | struct va_format vaf; | |
113 | va_list args; | |
114 | ||
115 | if (level > jbd2_journal_enable_debug) | |
116 | return; | |
117 | va_start(args, fmt); | |
118 | vaf.fmt = fmt; | |
119 | vaf.va = &args; | |
120 | printk(KERN_DEBUG "%s: (%s, %u): %pV\n", file, func, line, &vaf); | |
121 | va_end(args); | |
122 | } | |
123 | EXPORT_SYMBOL(__jbd2_debug); | |
124 | #endif | |
125 | ||
25ed6e8a | 126 | /* Checksumming functions */ |
7747e6d0 | 127 | static int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb) |
25ed6e8a | 128 | { |
8595798c | 129 | if (!jbd2_journal_has_csum_v2or3_feature(j)) |
25ed6e8a DW |
130 | return 1; |
131 | ||
132 | return sb->s_checksum_type == JBD2_CRC32C_CHKSUM; | |
133 | } | |
134 | ||
18a6ea1e | 135 | static __be32 jbd2_superblock_csum(journal_t *j, journal_superblock_t *sb) |
4fd5ea43 | 136 | { |
18a6ea1e DW |
137 | __u32 csum; |
138 | __be32 old_csum; | |
4fd5ea43 DW |
139 | |
140 | old_csum = sb->s_checksum; | |
141 | sb->s_checksum = 0; | |
142 | csum = jbd2_chksum(j, ~0, (char *)sb, sizeof(journal_superblock_t)); | |
143 | sb->s_checksum = old_csum; | |
144 | ||
145 | return cpu_to_be32(csum); | |
146 | } | |
147 | ||
7747e6d0 | 148 | static int jbd2_superblock_csum_verify(journal_t *j, journal_superblock_t *sb) |
4fd5ea43 | 149 | { |
db9ee220 | 150 | if (!jbd2_journal_has_csum_v2or3(j)) |
4fd5ea43 DW |
151 | return 1; |
152 | ||
153 | return sb->s_checksum == jbd2_superblock_csum(j, sb); | |
154 | } | |
155 | ||
7747e6d0 | 156 | static void jbd2_superblock_csum_set(journal_t *j, journal_superblock_t *sb) |
4fd5ea43 | 157 | { |
db9ee220 | 158 | if (!jbd2_journal_has_csum_v2or3(j)) |
4fd5ea43 DW |
159 | return; |
160 | ||
161 | sb->s_checksum = jbd2_superblock_csum(j, sb); | |
162 | } | |
163 | ||
470decc6 DK |
164 | /* |
165 | * Helper function used to manage commit timeouts | |
166 | */ | |
167 | ||
168 | static void commit_timeout(unsigned long __data) | |
169 | { | |
170 | struct task_struct * p = (struct task_struct *) __data; | |
171 | ||
172 | wake_up_process(p); | |
173 | } | |
174 | ||
175 | /* | |
f7f4bccb | 176 | * kjournald2: The main thread function used to manage a logging device |
470decc6 DK |
177 | * journal. |
178 | * | |
179 | * This kernel thread is responsible for two things: | |
180 | * | |
181 | * 1) COMMIT: Every so often we need to commit the current state of the | |
182 | * filesystem to disk. The journal thread is responsible for writing | |
183 | * all of the metadata buffers to disk. | |
184 | * | |
185 | * 2) CHECKPOINT: We cannot reuse a used section of the log file until all | |
186 | * of the data in that part of the log has been rewritten elsewhere on | |
187 | * the disk. Flushing these old buffers to reclaim space in the log is | |
188 | * known as checkpointing, and this thread is responsible for that job. | |
189 | */ | |
190 | ||
f7f4bccb | 191 | static int kjournald2(void *arg) |
470decc6 DK |
192 | { |
193 | journal_t *journal = arg; | |
194 | transaction_t *transaction; | |
195 | ||
196 | /* | |
197 | * Set up an interval timer which can be used to trigger a commit wakeup | |
198 | * after the commit interval expires | |
199 | */ | |
200 | setup_timer(&journal->j_commit_timer, commit_timeout, | |
201 | (unsigned long)current); | |
202 | ||
35c80422 NC |
203 | set_freezable(); |
204 | ||
470decc6 DK |
205 | /* Record that the journal thread is running */ |
206 | journal->j_task = current; | |
207 | wake_up(&journal->j_wait_done_commit); | |
208 | ||
eb52da3f MH |
209 | /* |
210 | * Make sure that no allocations from this kernel thread will ever | |
211 | * recurse to the fs layer because we are responsible for the | |
212 | * transaction commit and any fs involvement might get stuck waiting for | |
213 | * the trasn. commit. | |
214 | */ | |
215 | memalloc_nofs_save(); | |
216 | ||
470decc6 DK |
217 | /* |
218 | * And now, wait forever for commit wakeup events. | |
219 | */ | |
a931da6a | 220 | write_lock(&journal->j_state_lock); |
470decc6 DK |
221 | |
222 | loop: | |
f7f4bccb | 223 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
224 | goto end_loop; |
225 | ||
226 | jbd_debug(1, "commit_sequence=%d, commit_request=%d\n", | |
227 | journal->j_commit_sequence, journal->j_commit_request); | |
228 | ||
229 | if (journal->j_commit_sequence != journal->j_commit_request) { | |
230 | jbd_debug(1, "OK, requests differ\n"); | |
a931da6a | 231 | write_unlock(&journal->j_state_lock); |
470decc6 | 232 | del_timer_sync(&journal->j_commit_timer); |
f7f4bccb | 233 | jbd2_journal_commit_transaction(journal); |
a931da6a | 234 | write_lock(&journal->j_state_lock); |
470decc6 DK |
235 | goto loop; |
236 | } | |
237 | ||
238 | wake_up(&journal->j_wait_done_commit); | |
239 | if (freezing(current)) { | |
240 | /* | |
241 | * The simpler the better. Flushing journal isn't a | |
242 | * good idea, because that depends on threads that may | |
243 | * be already stopped. | |
244 | */ | |
f7f4bccb | 245 | jbd_debug(1, "Now suspending kjournald2\n"); |
a931da6a | 246 | write_unlock(&journal->j_state_lock); |
a0acae0e | 247 | try_to_freeze(); |
a931da6a | 248 | write_lock(&journal->j_state_lock); |
470decc6 DK |
249 | } else { |
250 | /* | |
251 | * We assume on resume that commits are already there, | |
252 | * so we don't sleep | |
253 | */ | |
254 | DEFINE_WAIT(wait); | |
255 | int should_sleep = 1; | |
256 | ||
257 | prepare_to_wait(&journal->j_wait_commit, &wait, | |
258 | TASK_INTERRUPTIBLE); | |
259 | if (journal->j_commit_sequence != journal->j_commit_request) | |
260 | should_sleep = 0; | |
261 | transaction = journal->j_running_transaction; | |
262 | if (transaction && time_after_eq(jiffies, | |
263 | transaction->t_expires)) | |
264 | should_sleep = 0; | |
f7f4bccb | 265 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
266 | should_sleep = 0; |
267 | if (should_sleep) { | |
a931da6a | 268 | write_unlock(&journal->j_state_lock); |
470decc6 | 269 | schedule(); |
a931da6a | 270 | write_lock(&journal->j_state_lock); |
470decc6 DK |
271 | } |
272 | finish_wait(&journal->j_wait_commit, &wait); | |
273 | } | |
274 | ||
f7f4bccb | 275 | jbd_debug(1, "kjournald2 wakes\n"); |
470decc6 DK |
276 | |
277 | /* | |
278 | * Were we woken up by a commit wakeup event? | |
279 | */ | |
280 | transaction = journal->j_running_transaction; | |
281 | if (transaction && time_after_eq(jiffies, transaction->t_expires)) { | |
282 | journal->j_commit_request = transaction->t_tid; | |
283 | jbd_debug(1, "woke because of timeout\n"); | |
284 | } | |
285 | goto loop; | |
286 | ||
287 | end_loop: | |
470decc6 DK |
288 | del_timer_sync(&journal->j_commit_timer); |
289 | journal->j_task = NULL; | |
290 | wake_up(&journal->j_wait_done_commit); | |
291 | jbd_debug(1, "Journal thread exiting.\n"); | |
dbfcef6b | 292 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
293 | return 0; |
294 | } | |
295 | ||
97f06784 | 296 | static int jbd2_journal_start_thread(journal_t *journal) |
470decc6 | 297 | { |
97f06784 PE |
298 | struct task_struct *t; |
299 | ||
90576c0b TT |
300 | t = kthread_run(kjournald2, journal, "jbd2/%s", |
301 | journal->j_devname); | |
97f06784 PE |
302 | if (IS_ERR(t)) |
303 | return PTR_ERR(t); | |
304 | ||
1076d17a | 305 | wait_event(journal->j_wait_done_commit, journal->j_task != NULL); |
97f06784 | 306 | return 0; |
470decc6 DK |
307 | } |
308 | ||
309 | static void journal_kill_thread(journal_t *journal) | |
310 | { | |
a931da6a | 311 | write_lock(&journal->j_state_lock); |
f7f4bccb | 312 | journal->j_flags |= JBD2_UNMOUNT; |
470decc6 DK |
313 | |
314 | while (journal->j_task) { | |
a931da6a | 315 | write_unlock(&journal->j_state_lock); |
3469a32a | 316 | wake_up(&journal->j_wait_commit); |
1076d17a | 317 | wait_event(journal->j_wait_done_commit, journal->j_task == NULL); |
a931da6a | 318 | write_lock(&journal->j_state_lock); |
470decc6 | 319 | } |
a931da6a | 320 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
321 | } |
322 | ||
323 | /* | |
f7f4bccb | 324 | * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. |
470decc6 DK |
325 | * |
326 | * Writes a metadata buffer to a given disk block. The actual IO is not | |
327 | * performed but a new buffer_head is constructed which labels the data | |
328 | * to be written with the correct destination disk block. | |
329 | * | |
330 | * Any magic-number escaping which needs to be done will cause a | |
331 | * copy-out here. If the buffer happens to start with the | |
f7f4bccb | 332 | * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the |
470decc6 DK |
333 | * magic number is only written to the log for descripter blocks. In |
334 | * this case, we copy the data and replace the first word with 0, and we | |
335 | * return a result code which indicates that this buffer needs to be | |
336 | * marked as an escaped buffer in the corresponding log descriptor | |
337 | * block. The missing word can then be restored when the block is read | |
338 | * during recovery. | |
339 | * | |
340 | * If the source buffer has already been modified by a new transaction | |
341 | * since we took the last commit snapshot, we use the frozen copy of | |
f5113eff JK |
342 | * that data for IO. If we end up using the existing buffer_head's data |
343 | * for the write, then we have to make sure nobody modifies it while the | |
344 | * IO is in progress. do_get_write_access() handles this. | |
470decc6 | 345 | * |
f5113eff JK |
346 | * The function returns a pointer to the buffer_head to be used for IO. |
347 | * | |
470decc6 DK |
348 | * |
349 | * Return value: | |
350 | * <0: Error | |
351 | * >=0: Finished OK | |
352 | * | |
353 | * On success: | |
354 | * Bit 0 set == escape performed on the data | |
355 | * Bit 1 set == buffer copy-out performed (kfree the data after IO) | |
356 | */ | |
357 | ||
f7f4bccb | 358 | int jbd2_journal_write_metadata_buffer(transaction_t *transaction, |
470decc6 | 359 | struct journal_head *jh_in, |
f5113eff JK |
360 | struct buffer_head **bh_out, |
361 | sector_t blocknr) | |
470decc6 DK |
362 | { |
363 | int need_copy_out = 0; | |
364 | int done_copy_out = 0; | |
365 | int do_escape = 0; | |
366 | char *mapped_data; | |
367 | struct buffer_head *new_bh; | |
470decc6 DK |
368 | struct page *new_page; |
369 | unsigned int new_offset; | |
370 | struct buffer_head *bh_in = jh2bh(jh_in); | |
96577c43 | 371 | journal_t *journal = transaction->t_journal; |
470decc6 DK |
372 | |
373 | /* | |
374 | * The buffer really shouldn't be locked: only the current committing | |
375 | * transaction is allowed to write it, so nobody else is allowed | |
376 | * to do any IO. | |
377 | * | |
378 | * akpm: except if we're journalling data, and write() output is | |
379 | * also part of a shared mapping, and another thread has | |
380 | * decided to launch a writepage() against this buffer. | |
381 | */ | |
382 | J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); | |
383 | ||
6ccaf3e2 | 384 | new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); |
47def826 | 385 | |
96577c43 | 386 | /* keep subsequent assertions sane */ |
96577c43 | 387 | atomic_set(&new_bh->b_count, 1); |
470decc6 | 388 | |
f5113eff JK |
389 | jbd_lock_bh_state(bh_in); |
390 | repeat: | |
470decc6 DK |
391 | /* |
392 | * If a new transaction has already done a buffer copy-out, then | |
393 | * we use that version of the data for the commit. | |
394 | */ | |
470decc6 DK |
395 | if (jh_in->b_frozen_data) { |
396 | done_copy_out = 1; | |
397 | new_page = virt_to_page(jh_in->b_frozen_data); | |
398 | new_offset = offset_in_page(jh_in->b_frozen_data); | |
399 | } else { | |
400 | new_page = jh2bh(jh_in)->b_page; | |
401 | new_offset = offset_in_page(jh2bh(jh_in)->b_data); | |
402 | } | |
403 | ||
303a8f2a | 404 | mapped_data = kmap_atomic(new_page); |
e06c8227 | 405 | /* |
13ceef09 JK |
406 | * Fire data frozen trigger if data already wasn't frozen. Do this |
407 | * before checking for escaping, as the trigger may modify the magic | |
408 | * offset. If a copy-out happens afterwards, it will have the correct | |
409 | * data in the buffer. | |
e06c8227 | 410 | */ |
13ceef09 JK |
411 | if (!done_copy_out) |
412 | jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset, | |
413 | jh_in->b_triggers); | |
e06c8227 | 414 | |
470decc6 DK |
415 | /* |
416 | * Check for escaping | |
417 | */ | |
418 | if (*((__be32 *)(mapped_data + new_offset)) == | |
f7f4bccb | 419 | cpu_to_be32(JBD2_MAGIC_NUMBER)) { |
470decc6 DK |
420 | need_copy_out = 1; |
421 | do_escape = 1; | |
422 | } | |
303a8f2a | 423 | kunmap_atomic(mapped_data); |
470decc6 DK |
424 | |
425 | /* | |
426 | * Do we need to do a data copy? | |
427 | */ | |
428 | if (need_copy_out && !done_copy_out) { | |
429 | char *tmp; | |
430 | ||
431 | jbd_unlock_bh_state(bh_in); | |
af1e76d6 | 432 | tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); |
e6ec116b | 433 | if (!tmp) { |
f5113eff | 434 | brelse(new_bh); |
e6ec116b TT |
435 | return -ENOMEM; |
436 | } | |
470decc6 DK |
437 | jbd_lock_bh_state(bh_in); |
438 | if (jh_in->b_frozen_data) { | |
af1e76d6 | 439 | jbd2_free(tmp, bh_in->b_size); |
470decc6 DK |
440 | goto repeat; |
441 | } | |
442 | ||
443 | jh_in->b_frozen_data = tmp; | |
303a8f2a | 444 | mapped_data = kmap_atomic(new_page); |
f5113eff | 445 | memcpy(tmp, mapped_data + new_offset, bh_in->b_size); |
303a8f2a | 446 | kunmap_atomic(mapped_data); |
470decc6 DK |
447 | |
448 | new_page = virt_to_page(tmp); | |
449 | new_offset = offset_in_page(tmp); | |
450 | done_copy_out = 1; | |
e06c8227 JB |
451 | |
452 | /* | |
453 | * This isn't strictly necessary, as we're using frozen | |
454 | * data for the escaping, but it keeps consistency with | |
455 | * b_frozen_data usage. | |
456 | */ | |
457 | jh_in->b_frozen_triggers = jh_in->b_triggers; | |
470decc6 DK |
458 | } |
459 | ||
460 | /* | |
461 | * Did we need to do an escaping? Now we've done all the | |
462 | * copying, we can finally do so. | |
463 | */ | |
464 | if (do_escape) { | |
303a8f2a | 465 | mapped_data = kmap_atomic(new_page); |
470decc6 | 466 | *((unsigned int *)(mapped_data + new_offset)) = 0; |
303a8f2a | 467 | kunmap_atomic(mapped_data); |
470decc6 DK |
468 | } |
469 | ||
470decc6 | 470 | set_bh_page(new_bh, new_page, new_offset); |
f5113eff JK |
471 | new_bh->b_size = bh_in->b_size; |
472 | new_bh->b_bdev = journal->j_dev; | |
470decc6 | 473 | new_bh->b_blocknr = blocknr; |
b34090e5 | 474 | new_bh->b_private = bh_in; |
470decc6 DK |
475 | set_buffer_mapped(new_bh); |
476 | set_buffer_dirty(new_bh); | |
477 | ||
f5113eff | 478 | *bh_out = new_bh; |
470decc6 DK |
479 | |
480 | /* | |
481 | * The to-be-written buffer needs to get moved to the io queue, | |
482 | * and the original buffer whose contents we are shadowing or | |
483 | * copying is moved to the transaction's shadow queue. | |
484 | */ | |
485 | JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); | |
96577c43 | 486 | spin_lock(&journal->j_list_lock); |
487 | __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); | |
488 | spin_unlock(&journal->j_list_lock); | |
b34090e5 | 489 | set_buffer_shadow(bh_in); |
96577c43 | 490 | jbd_unlock_bh_state(bh_in); |
491 | ||
470decc6 DK |
492 | return do_escape | (done_copy_out << 1); |
493 | } | |
494 | ||
495 | /* | |
496 | * Allocation code for the journal file. Manage the space left in the | |
497 | * journal, so that we can begin checkpointing when appropriate. | |
498 | */ | |
499 | ||
470decc6 | 500 | /* |
e4471831 TT |
501 | * Called with j_state_lock locked for writing. |
502 | * Returns true if a transaction commit was started. | |
470decc6 | 503 | */ |
f7f4bccb | 504 | int __jbd2_log_start_commit(journal_t *journal, tid_t target) |
470decc6 | 505 | { |
e7b04ac0 ES |
506 | /* Return if the txn has already requested to be committed */ |
507 | if (journal->j_commit_request == target) | |
508 | return 0; | |
509 | ||
470decc6 | 510 | /* |
deeeaf13 TT |
511 | * The only transaction we can possibly wait upon is the |
512 | * currently running transaction (if it exists). Otherwise, | |
513 | * the target tid must be an old one. | |
470decc6 | 514 | */ |
deeeaf13 TT |
515 | if (journal->j_running_transaction && |
516 | journal->j_running_transaction->t_tid == target) { | |
470decc6 | 517 | /* |
bcf3d0bc | 518 | * We want a new commit: OK, mark the request and wakeup the |
470decc6 DK |
519 | * commit thread. We do _not_ do the commit ourselves. |
520 | */ | |
521 | ||
522 | journal->j_commit_request = target; | |
f2a44523 | 523 | jbd_debug(1, "JBD2: requesting commit %d/%d\n", |
470decc6 DK |
524 | journal->j_commit_request, |
525 | journal->j_commit_sequence); | |
9fff24aa | 526 | journal->j_running_transaction->t_requested = jiffies; |
470decc6 DK |
527 | wake_up(&journal->j_wait_commit); |
528 | return 1; | |
deeeaf13 TT |
529 | } else if (!tid_geq(journal->j_commit_request, target)) |
530 | /* This should never happen, but if it does, preserve | |
531 | the evidence before kjournald goes into a loop and | |
532 | increments j_commit_sequence beyond all recognition. */ | |
f2a44523 | 533 | WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", |
1be2add6 TT |
534 | journal->j_commit_request, |
535 | journal->j_commit_sequence, | |
536 | target, journal->j_running_transaction ? | |
537 | journal->j_running_transaction->t_tid : 0); | |
470decc6 DK |
538 | return 0; |
539 | } | |
540 | ||
f7f4bccb | 541 | int jbd2_log_start_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
542 | { |
543 | int ret; | |
544 | ||
a931da6a | 545 | write_lock(&journal->j_state_lock); |
f7f4bccb | 546 | ret = __jbd2_log_start_commit(journal, tid); |
a931da6a | 547 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
548 | return ret; |
549 | } | |
550 | ||
551 | /* | |
9ff86446 DM |
552 | * Force and wait any uncommitted transactions. We can only force the running |
553 | * transaction if we don't have an active handle, otherwise, we will deadlock. | |
554 | * Returns: <0 in case of error, | |
555 | * 0 if nothing to commit, | |
556 | * 1 if transaction was successfully committed. | |
470decc6 | 557 | */ |
9ff86446 | 558 | static int __jbd2_journal_force_commit(journal_t *journal) |
470decc6 DK |
559 | { |
560 | transaction_t *transaction = NULL; | |
561 | tid_t tid; | |
9ff86446 | 562 | int need_to_start = 0, ret = 0; |
470decc6 | 563 | |
a931da6a | 564 | read_lock(&journal->j_state_lock); |
470decc6 DK |
565 | if (journal->j_running_transaction && !current->journal_info) { |
566 | transaction = journal->j_running_transaction; | |
e4471831 TT |
567 | if (!tid_geq(journal->j_commit_request, transaction->t_tid)) |
568 | need_to_start = 1; | |
470decc6 DK |
569 | } else if (journal->j_committing_transaction) |
570 | transaction = journal->j_committing_transaction; | |
571 | ||
572 | if (!transaction) { | |
9ff86446 | 573 | /* Nothing to commit */ |
a931da6a | 574 | read_unlock(&journal->j_state_lock); |
9ff86446 | 575 | return 0; |
470decc6 | 576 | } |
470decc6 | 577 | tid = transaction->t_tid; |
a931da6a | 578 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
579 | if (need_to_start) |
580 | jbd2_log_start_commit(journal, tid); | |
9ff86446 DM |
581 | ret = jbd2_log_wait_commit(journal, tid); |
582 | if (!ret) | |
583 | ret = 1; | |
584 | ||
585 | return ret; | |
586 | } | |
587 | ||
588 | /** | |
589 | * Force and wait upon a commit if the calling process is not within | |
590 | * transaction. This is used for forcing out undo-protected data which contains | |
591 | * bitmaps, when the fs is running out of space. | |
592 | * | |
593 | * @journal: journal to force | |
594 | * Returns true if progress was made. | |
595 | */ | |
596 | int jbd2_journal_force_commit_nested(journal_t *journal) | |
597 | { | |
598 | int ret; | |
599 | ||
600 | ret = __jbd2_journal_force_commit(journal); | |
601 | return ret > 0; | |
602 | } | |
603 | ||
604 | /** | |
605 | * int journal_force_commit() - force any uncommitted transactions | |
606 | * @journal: journal to force | |
607 | * | |
608 | * Caller want unconditional commit. We can only force the running transaction | |
609 | * if we don't have an active handle, otherwise, we will deadlock. | |
610 | */ | |
611 | int jbd2_journal_force_commit(journal_t *journal) | |
612 | { | |
613 | int ret; | |
614 | ||
615 | J_ASSERT(!current->journal_info); | |
616 | ret = __jbd2_journal_force_commit(journal); | |
617 | if (ret > 0) | |
618 | ret = 0; | |
619 | return ret; | |
470decc6 DK |
620 | } |
621 | ||
622 | /* | |
623 | * Start a commit of the current running transaction (if any). Returns true | |
c88ccea3 JK |
624 | * if a transaction is going to be committed (or is currently already |
625 | * committing), and fills its tid in at *ptid | |
470decc6 | 626 | */ |
f7f4bccb | 627 | int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) |
470decc6 DK |
628 | { |
629 | int ret = 0; | |
630 | ||
a931da6a | 631 | write_lock(&journal->j_state_lock); |
470decc6 DK |
632 | if (journal->j_running_transaction) { |
633 | tid_t tid = journal->j_running_transaction->t_tid; | |
634 | ||
c88ccea3 JK |
635 | __jbd2_log_start_commit(journal, tid); |
636 | /* There's a running transaction and we've just made sure | |
637 | * it's commit has been scheduled. */ | |
638 | if (ptid) | |
470decc6 | 639 | *ptid = tid; |
c88ccea3 JK |
640 | ret = 1; |
641 | } else if (journal->j_committing_transaction) { | |
470decc6 | 642 | /* |
12810ad7 AB |
643 | * If commit has been started, then we have to wait for |
644 | * completion of that transaction. | |
470decc6 | 645 | */ |
c88ccea3 JK |
646 | if (ptid) |
647 | *ptid = journal->j_committing_transaction->t_tid; | |
470decc6 DK |
648 | ret = 1; |
649 | } | |
a931da6a | 650 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
651 | return ret; |
652 | } | |
653 | ||
bbd2be36 JK |
654 | /* |
655 | * Return 1 if a given transaction has not yet sent barrier request | |
656 | * connected with a transaction commit. If 0 is returned, transaction | |
657 | * may or may not have sent the barrier. Used to avoid sending barrier | |
658 | * twice in common cases. | |
659 | */ | |
660 | int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) | |
661 | { | |
662 | int ret = 0; | |
663 | transaction_t *commit_trans; | |
664 | ||
665 | if (!(journal->j_flags & JBD2_BARRIER)) | |
666 | return 0; | |
667 | read_lock(&journal->j_state_lock); | |
668 | /* Transaction already committed? */ | |
669 | if (tid_geq(journal->j_commit_sequence, tid)) | |
670 | goto out; | |
671 | commit_trans = journal->j_committing_transaction; | |
672 | if (!commit_trans || commit_trans->t_tid != tid) { | |
673 | ret = 1; | |
674 | goto out; | |
675 | } | |
676 | /* | |
677 | * Transaction is being committed and we already proceeded to | |
678 | * submitting a flush to fs partition? | |
679 | */ | |
680 | if (journal->j_fs_dev != journal->j_dev) { | |
681 | if (!commit_trans->t_need_data_flush || | |
682 | commit_trans->t_state >= T_COMMIT_DFLUSH) | |
683 | goto out; | |
684 | } else { | |
685 | if (commit_trans->t_state >= T_COMMIT_JFLUSH) | |
686 | goto out; | |
687 | } | |
688 | ret = 1; | |
689 | out: | |
690 | read_unlock(&journal->j_state_lock); | |
691 | return ret; | |
692 | } | |
693 | EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); | |
694 | ||
470decc6 DK |
695 | /* |
696 | * Wait for a specified commit to complete. | |
697 | * The caller may not hold the journal lock. | |
698 | */ | |
f7f4bccb | 699 | int jbd2_log_wait_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
700 | { |
701 | int err = 0; | |
702 | ||
a931da6a | 703 | read_lock(&journal->j_state_lock); |
c52c47e4 JK |
704 | #ifdef CONFIG_PROVE_LOCKING |
705 | /* | |
706 | * Some callers make sure transaction is already committing and in that | |
707 | * case we cannot block on open handles anymore. So don't warn in that | |
708 | * case. | |
709 | */ | |
710 | if (tid_gt(tid, journal->j_commit_sequence) && | |
711 | (!journal->j_committing_transaction || | |
712 | journal->j_committing_transaction->t_tid != tid)) { | |
713 | read_unlock(&journal->j_state_lock); | |
714 | jbd2_might_wait_for_commit(journal); | |
715 | read_lock(&journal->j_state_lock); | |
716 | } | |
717 | #endif | |
e23291b9 | 718 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 | 719 | if (!tid_geq(journal->j_commit_request, tid)) { |
75685071 | 720 | printk(KERN_ERR |
470decc6 | 721 | "%s: error: j_commit_request=%d, tid=%d\n", |
329d291f | 722 | __func__, journal->j_commit_request, tid); |
470decc6 | 723 | } |
470decc6 | 724 | #endif |
470decc6 | 725 | while (tid_gt(tid, journal->j_commit_sequence)) { |
f2a44523 | 726 | jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n", |
470decc6 | 727 | tid, journal->j_commit_sequence); |
a931da6a | 728 | read_unlock(&journal->j_state_lock); |
3469a32a | 729 | wake_up(&journal->j_wait_commit); |
470decc6 DK |
730 | wait_event(journal->j_wait_done_commit, |
731 | !tid_gt(tid, journal->j_commit_sequence)); | |
a931da6a | 732 | read_lock(&journal->j_state_lock); |
470decc6 | 733 | } |
a931da6a | 734 | read_unlock(&journal->j_state_lock); |
470decc6 | 735 | |
75685071 | 736 | if (unlikely(is_journal_aborted(journal))) |
470decc6 | 737 | err = -EIO; |
470decc6 DK |
738 | return err; |
739 | } | |
740 | ||
d76a3a77 TT |
741 | /* |
742 | * When this function returns the transaction corresponding to tid | |
743 | * will be completed. If the transaction has currently running, start | |
744 | * committing that transaction before waiting for it to complete. If | |
745 | * the transaction id is stale, it is by definition already completed, | |
746 | * so just return SUCCESS. | |
747 | */ | |
748 | int jbd2_complete_transaction(journal_t *journal, tid_t tid) | |
749 | { | |
750 | int need_to_wait = 1; | |
751 | ||
752 | read_lock(&journal->j_state_lock); | |
753 | if (journal->j_running_transaction && | |
754 | journal->j_running_transaction->t_tid == tid) { | |
755 | if (journal->j_commit_request != tid) { | |
756 | /* transaction not yet started, so request it */ | |
757 | read_unlock(&journal->j_state_lock); | |
758 | jbd2_log_start_commit(journal, tid); | |
759 | goto wait_commit; | |
760 | } | |
761 | } else if (!(journal->j_committing_transaction && | |
762 | journal->j_committing_transaction->t_tid == tid)) | |
763 | need_to_wait = 0; | |
764 | read_unlock(&journal->j_state_lock); | |
765 | if (!need_to_wait) | |
766 | return 0; | |
767 | wait_commit: | |
768 | return jbd2_log_wait_commit(journal, tid); | |
769 | } | |
770 | EXPORT_SYMBOL(jbd2_complete_transaction); | |
771 | ||
470decc6 DK |
772 | /* |
773 | * Log buffer allocation routines: | |
774 | */ | |
775 | ||
18eba7aa | 776 | int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) |
470decc6 DK |
777 | { |
778 | unsigned long blocknr; | |
779 | ||
a931da6a | 780 | write_lock(&journal->j_state_lock); |
470decc6 DK |
781 | J_ASSERT(journal->j_free > 1); |
782 | ||
783 | blocknr = journal->j_head; | |
784 | journal->j_head++; | |
785 | journal->j_free--; | |
786 | if (journal->j_head == journal->j_last) | |
787 | journal->j_head = journal->j_first; | |
a931da6a | 788 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 789 | return jbd2_journal_bmap(journal, blocknr, retp); |
470decc6 DK |
790 | } |
791 | ||
792 | /* | |
793 | * Conversion of logical to physical block numbers for the journal | |
794 | * | |
795 | * On external journals the journal blocks are identity-mapped, so | |
796 | * this is a no-op. If needed, we can use j_blk_offset - everything is | |
797 | * ready. | |
798 | */ | |
f7f4bccb | 799 | int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, |
18eba7aa | 800 | unsigned long long *retp) |
470decc6 DK |
801 | { |
802 | int err = 0; | |
18eba7aa | 803 | unsigned long long ret; |
470decc6 DK |
804 | |
805 | if (journal->j_inode) { | |
806 | ret = bmap(journal->j_inode, blocknr); | |
807 | if (ret) | |
808 | *retp = ret; | |
809 | else { | |
470decc6 DK |
810 | printk(KERN_ALERT "%s: journal block not found " |
811 | "at offset %lu on %s\n", | |
05496769 | 812 | __func__, blocknr, journal->j_devname); |
470decc6 DK |
813 | err = -EIO; |
814 | __journal_abort_soft(journal, err); | |
815 | } | |
816 | } else { | |
817 | *retp = blocknr; /* +journal->j_blk_offset */ | |
818 | } | |
819 | return err; | |
820 | } | |
821 | ||
822 | /* | |
823 | * We play buffer_head aliasing tricks to write data/metadata blocks to | |
824 | * the journal without copying their contents, but for journal | |
825 | * descriptor blocks we do need to generate bona fide buffers. | |
826 | * | |
f7f4bccb | 827 | * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying |
470decc6 DK |
828 | * the buffer's contents they really should run flush_dcache_page(bh->b_page). |
829 | * But we don't bother doing that, so there will be coherency problems with | |
830 | * mmaps of blockdevs which hold live JBD-controlled filesystems. | |
831 | */ | |
32ab6715 JK |
832 | struct buffer_head * |
833 | jbd2_journal_get_descriptor_buffer(transaction_t *transaction, int type) | |
470decc6 | 834 | { |
32ab6715 | 835 | journal_t *journal = transaction->t_journal; |
470decc6 | 836 | struct buffer_head *bh; |
18eba7aa | 837 | unsigned long long blocknr; |
32ab6715 | 838 | journal_header_t *header; |
470decc6 DK |
839 | int err; |
840 | ||
f7f4bccb | 841 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
842 | |
843 | if (err) | |
844 | return NULL; | |
845 | ||
846 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
4b905671 JK |
847 | if (!bh) |
848 | return NULL; | |
470decc6 DK |
849 | lock_buffer(bh); |
850 | memset(bh->b_data, 0, journal->j_blocksize); | |
32ab6715 JK |
851 | header = (journal_header_t *)bh->b_data; |
852 | header->h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); | |
853 | header->h_blocktype = cpu_to_be32(type); | |
854 | header->h_sequence = cpu_to_be32(transaction->t_tid); | |
470decc6 DK |
855 | set_buffer_uptodate(bh); |
856 | unlock_buffer(bh); | |
857 | BUFFER_TRACE(bh, "return this buffer"); | |
e5a120ae | 858 | return bh; |
470decc6 DK |
859 | } |
860 | ||
1101cd4d JK |
861 | void jbd2_descriptor_block_csum_set(journal_t *j, struct buffer_head *bh) |
862 | { | |
863 | struct jbd2_journal_block_tail *tail; | |
864 | __u32 csum; | |
865 | ||
866 | if (!jbd2_journal_has_csum_v2or3(j)) | |
867 | return; | |
868 | ||
869 | tail = (struct jbd2_journal_block_tail *)(bh->b_data + j->j_blocksize - | |
870 | sizeof(struct jbd2_journal_block_tail)); | |
871 | tail->t_checksum = 0; | |
872 | csum = jbd2_chksum(j, j->j_csum_seed, bh->b_data, j->j_blocksize); | |
873 | tail->t_checksum = cpu_to_be32(csum); | |
874 | } | |
875 | ||
79feb521 JK |
876 | /* |
877 | * Return tid of the oldest transaction in the journal and block in the journal | |
878 | * where the transaction starts. | |
879 | * | |
880 | * If the journal is now empty, return which will be the next transaction ID | |
881 | * we will write and where will that transaction start. | |
882 | * | |
883 | * The return value is 0 if journal tail cannot be pushed any further, 1 if | |
884 | * it can. | |
885 | */ | |
886 | int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, | |
887 | unsigned long *block) | |
888 | { | |
889 | transaction_t *transaction; | |
890 | int ret; | |
891 | ||
892 | read_lock(&journal->j_state_lock); | |
893 | spin_lock(&journal->j_list_lock); | |
894 | transaction = journal->j_checkpoint_transactions; | |
895 | if (transaction) { | |
896 | *tid = transaction->t_tid; | |
897 | *block = transaction->t_log_start; | |
898 | } else if ((transaction = journal->j_committing_transaction) != NULL) { | |
899 | *tid = transaction->t_tid; | |
900 | *block = transaction->t_log_start; | |
901 | } else if ((transaction = journal->j_running_transaction) != NULL) { | |
902 | *tid = transaction->t_tid; | |
903 | *block = journal->j_head; | |
904 | } else { | |
905 | *tid = journal->j_transaction_sequence; | |
906 | *block = journal->j_head; | |
907 | } | |
908 | ret = tid_gt(*tid, journal->j_tail_sequence); | |
909 | spin_unlock(&journal->j_list_lock); | |
910 | read_unlock(&journal->j_state_lock); | |
911 | ||
912 | return ret; | |
913 | } | |
914 | ||
915 | /* | |
916 | * Update information in journal structure and in on disk journal superblock | |
917 | * about log tail. This function does not check whether information passed in | |
918 | * really pushes log tail further. It's responsibility of the caller to make | |
919 | * sure provided log tail information is valid (e.g. by holding | |
920 | * j_checkpoint_mutex all the time between computing log tail and calling this | |
921 | * function as is the case with jbd2_cleanup_journal_tail()). | |
922 | * | |
923 | * Requires j_checkpoint_mutex | |
924 | */ | |
6f6a6fda | 925 | int __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) |
79feb521 JK |
926 | { |
927 | unsigned long freed; | |
6f6a6fda | 928 | int ret; |
79feb521 JK |
929 | |
930 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); | |
931 | ||
932 | /* | |
933 | * We cannot afford for write to remain in drive's caches since as | |
934 | * soon as we update j_tail, next transaction can start reusing journal | |
935 | * space and if we lose sb update during power failure we'd replay | |
936 | * old transaction with possibly newly overwritten data. | |
937 | */ | |
17f423b5 JK |
938 | ret = jbd2_journal_update_sb_log_tail(journal, tid, block, |
939 | REQ_SYNC | REQ_FUA); | |
6f6a6fda JQ |
940 | if (ret) |
941 | goto out; | |
942 | ||
79feb521 JK |
943 | write_lock(&journal->j_state_lock); |
944 | freed = block - journal->j_tail; | |
945 | if (block < journal->j_tail) | |
946 | freed += journal->j_last - journal->j_first; | |
947 | ||
948 | trace_jbd2_update_log_tail(journal, tid, block, freed); | |
949 | jbd_debug(1, | |
950 | "Cleaning journal tail from %d to %d (offset %lu), " | |
951 | "freeing %lu\n", | |
952 | journal->j_tail_sequence, tid, block, freed); | |
953 | ||
954 | journal->j_free += freed; | |
955 | journal->j_tail_sequence = tid; | |
956 | journal->j_tail = block; | |
957 | write_unlock(&journal->j_state_lock); | |
6f6a6fda JQ |
958 | |
959 | out: | |
960 | return ret; | |
79feb521 JK |
961 | } |
962 | ||
3339578f JK |
963 | /* |
964 | * This is a variaon of __jbd2_update_log_tail which checks for validity of | |
965 | * provided log tail and locks j_checkpoint_mutex. So it is safe against races | |
966 | * with other threads updating log tail. | |
967 | */ | |
968 | void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) | |
969 | { | |
6fa7aa50 | 970 | mutex_lock_io(&journal->j_checkpoint_mutex); |
3339578f JK |
971 | if (tid_gt(tid, journal->j_tail_sequence)) |
972 | __jbd2_update_log_tail(journal, tid, block); | |
973 | mutex_unlock(&journal->j_checkpoint_mutex); | |
974 | } | |
975 | ||
8e85fb3f JL |
976 | struct jbd2_stats_proc_session { |
977 | journal_t *journal; | |
978 | struct transaction_stats_s *stats; | |
979 | int start; | |
980 | int max; | |
981 | }; | |
982 | ||
8e85fb3f JL |
983 | static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) |
984 | { | |
985 | return *pos ? NULL : SEQ_START_TOKEN; | |
986 | } | |
987 | ||
988 | static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) | |
989 | { | |
990 | return NULL; | |
991 | } | |
992 | ||
993 | static int jbd2_seq_info_show(struct seq_file *seq, void *v) | |
994 | { | |
995 | struct jbd2_stats_proc_session *s = seq->private; | |
996 | ||
997 | if (v != SEQ_START_TOKEN) | |
998 | return 0; | |
9fff24aa TT |
999 | seq_printf(seq, "%lu transactions (%lu requested), " |
1000 | "each up to %u blocks\n", | |
1001 | s->stats->ts_tid, s->stats->ts_requested, | |
1002 | s->journal->j_max_transaction_buffers); | |
8e85fb3f JL |
1003 | if (s->stats->ts_tid == 0) |
1004 | return 0; | |
1005 | seq_printf(seq, "average: \n %ums waiting for transaction\n", | |
bf699327 | 1006 | jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); |
9fff24aa TT |
1007 | seq_printf(seq, " %ums request delay\n", |
1008 | (s->stats->ts_requested == 0) ? 0 : | |
1009 | jiffies_to_msecs(s->stats->run.rs_request_delay / | |
1010 | s->stats->ts_requested)); | |
8e85fb3f | 1011 | seq_printf(seq, " %ums running transaction\n", |
bf699327 | 1012 | jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); |
8e85fb3f | 1013 | seq_printf(seq, " %ums transaction was being locked\n", |
bf699327 | 1014 | jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); |
8e85fb3f | 1015 | seq_printf(seq, " %ums flushing data (in ordered mode)\n", |
bf699327 | 1016 | jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); |
8e85fb3f | 1017 | seq_printf(seq, " %ums logging transaction\n", |
bf699327 | 1018 | jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); |
c225aa57 SHT |
1019 | seq_printf(seq, " %lluus average transaction commit time\n", |
1020 | div_u64(s->journal->j_average_commit_time, 1000)); | |
8e85fb3f | 1021 | seq_printf(seq, " %lu handles per transaction\n", |
bf699327 | 1022 | s->stats->run.rs_handle_count / s->stats->ts_tid); |
8e85fb3f | 1023 | seq_printf(seq, " %lu blocks per transaction\n", |
bf699327 | 1024 | s->stats->run.rs_blocks / s->stats->ts_tid); |
8e85fb3f | 1025 | seq_printf(seq, " %lu logged blocks per transaction\n", |
bf699327 | 1026 | s->stats->run.rs_blocks_logged / s->stats->ts_tid); |
8e85fb3f JL |
1027 | return 0; |
1028 | } | |
1029 | ||
1030 | static void jbd2_seq_info_stop(struct seq_file *seq, void *v) | |
1031 | { | |
1032 | } | |
1033 | ||
88e9d34c | 1034 | static const struct seq_operations jbd2_seq_info_ops = { |
8e85fb3f JL |
1035 | .start = jbd2_seq_info_start, |
1036 | .next = jbd2_seq_info_next, | |
1037 | .stop = jbd2_seq_info_stop, | |
1038 | .show = jbd2_seq_info_show, | |
1039 | }; | |
1040 | ||
1041 | static int jbd2_seq_info_open(struct inode *inode, struct file *file) | |
1042 | { | |
d9dda78b | 1043 | journal_t *journal = PDE_DATA(inode); |
8e85fb3f JL |
1044 | struct jbd2_stats_proc_session *s; |
1045 | int rc, size; | |
1046 | ||
1047 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
1048 | if (s == NULL) | |
1049 | return -ENOMEM; | |
1050 | size = sizeof(struct transaction_stats_s); | |
1051 | s->stats = kmalloc(size, GFP_KERNEL); | |
1052 | if (s->stats == NULL) { | |
1053 | kfree(s); | |
1054 | return -ENOMEM; | |
1055 | } | |
1056 | spin_lock(&journal->j_history_lock); | |
1057 | memcpy(s->stats, &journal->j_stats, size); | |
1058 | s->journal = journal; | |
1059 | spin_unlock(&journal->j_history_lock); | |
1060 | ||
1061 | rc = seq_open(file, &jbd2_seq_info_ops); | |
1062 | if (rc == 0) { | |
1063 | struct seq_file *m = file->private_data; | |
1064 | m->private = s; | |
1065 | } else { | |
1066 | kfree(s->stats); | |
1067 | kfree(s); | |
1068 | } | |
1069 | return rc; | |
1070 | ||
1071 | } | |
1072 | ||
1073 | static int jbd2_seq_info_release(struct inode *inode, struct file *file) | |
1074 | { | |
1075 | struct seq_file *seq = file->private_data; | |
1076 | struct jbd2_stats_proc_session *s = seq->private; | |
1077 | kfree(s->stats); | |
1078 | kfree(s); | |
1079 | return seq_release(inode, file); | |
1080 | } | |
1081 | ||
828c0950 | 1082 | static const struct file_operations jbd2_seq_info_fops = { |
8e85fb3f JL |
1083 | .owner = THIS_MODULE, |
1084 | .open = jbd2_seq_info_open, | |
1085 | .read = seq_read, | |
1086 | .llseek = seq_lseek, | |
1087 | .release = jbd2_seq_info_release, | |
1088 | }; | |
1089 | ||
1090 | static struct proc_dir_entry *proc_jbd2_stats; | |
1091 | ||
1092 | static void jbd2_stats_proc_init(journal_t *journal) | |
1093 | { | |
05496769 | 1094 | journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); |
8e85fb3f | 1095 | if (journal->j_proc_entry) { |
79da3664 DL |
1096 | proc_create_data("info", S_IRUGO, journal->j_proc_entry, |
1097 | &jbd2_seq_info_fops, journal); | |
8e85fb3f JL |
1098 | } |
1099 | } | |
1100 | ||
1101 | static void jbd2_stats_proc_exit(journal_t *journal) | |
1102 | { | |
8e85fb3f | 1103 | remove_proc_entry("info", journal->j_proc_entry); |
05496769 | 1104 | remove_proc_entry(journal->j_devname, proc_jbd2_stats); |
8e85fb3f JL |
1105 | } |
1106 | ||
470decc6 DK |
1107 | /* |
1108 | * Management for journal control blocks: functions to create and | |
1109 | * destroy journal_t structures, and to initialise and read existing | |
1110 | * journal blocks from disk. */ | |
1111 | ||
1112 | /* First: create and setup a journal_t object in memory. We initialise | |
1113 | * very few fields yet: that has to wait until we have created the | |
1114 | * journal structures from from scratch, or loaded them from disk. */ | |
1115 | ||
f0c9fd54 GT |
1116 | static journal_t *journal_init_common(struct block_device *bdev, |
1117 | struct block_device *fs_dev, | |
1118 | unsigned long long start, int len, int blocksize) | |
470decc6 | 1119 | { |
ab714aff | 1120 | static struct lock_class_key jbd2_trans_commit_key; |
470decc6 DK |
1121 | journal_t *journal; |
1122 | int err; | |
f0c9fd54 GT |
1123 | struct buffer_head *bh; |
1124 | int n; | |
470decc6 | 1125 | |
3ebfdf88 | 1126 | journal = kzalloc(sizeof(*journal), GFP_KERNEL); |
470decc6 | 1127 | if (!journal) |
b7271b0a | 1128 | return NULL; |
470decc6 DK |
1129 | |
1130 | init_waitqueue_head(&journal->j_wait_transaction_locked); | |
470decc6 | 1131 | init_waitqueue_head(&journal->j_wait_done_commit); |
470decc6 DK |
1132 | init_waitqueue_head(&journal->j_wait_commit); |
1133 | init_waitqueue_head(&journal->j_wait_updates); | |
8f7d89f3 | 1134 | init_waitqueue_head(&journal->j_wait_reserved); |
470decc6 DK |
1135 | mutex_init(&journal->j_barrier); |
1136 | mutex_init(&journal->j_checkpoint_mutex); | |
1137 | spin_lock_init(&journal->j_revoke_lock); | |
1138 | spin_lock_init(&journal->j_list_lock); | |
a931da6a | 1139 | rwlock_init(&journal->j_state_lock); |
470decc6 | 1140 | |
cd02ff0b | 1141 | journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); |
30773840 TT |
1142 | journal->j_min_batch_time = 0; |
1143 | journal->j_max_batch_time = 15000; /* 15ms */ | |
8f7d89f3 | 1144 | atomic_set(&journal->j_reserved_credits, 0); |
470decc6 DK |
1145 | |
1146 | /* The journal is marked for error until we succeed with recovery! */ | |
f7f4bccb | 1147 | journal->j_flags = JBD2_ABORT; |
470decc6 DK |
1148 | |
1149 | /* Set up a default-sized revoke table for the new mount. */ | |
f7f4bccb | 1150 | err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); |
cd9cb405 EB |
1151 | if (err) |
1152 | goto err_cleanup; | |
8e85fb3f | 1153 | |
bf699327 | 1154 | spin_lock_init(&journal->j_history_lock); |
8e85fb3f | 1155 | |
ab714aff JK |
1156 | lockdep_init_map(&journal->j_trans_commit_map, "jbd2_handle", |
1157 | &jbd2_trans_commit_key, 0); | |
1158 | ||
f0c9fd54 GT |
1159 | /* journal descriptor can store up to n blocks -bzzz */ |
1160 | journal->j_blocksize = blocksize; | |
1161 | journal->j_dev = bdev; | |
1162 | journal->j_fs_dev = fs_dev; | |
1163 | journal->j_blk_offset = start; | |
1164 | journal->j_maxlen = len; | |
1165 | n = journal->j_blocksize / sizeof(journal_block_tag_t); | |
1166 | journal->j_wbufsize = n; | |
1167 | journal->j_wbuf = kmalloc_array(n, sizeof(struct buffer_head *), | |
1168 | GFP_KERNEL); | |
cd9cb405 EB |
1169 | if (!journal->j_wbuf) |
1170 | goto err_cleanup; | |
f0c9fd54 GT |
1171 | |
1172 | bh = getblk_unmovable(journal->j_dev, start, journal->j_blocksize); | |
1173 | if (!bh) { | |
1174 | pr_err("%s: Cannot get buffer for journal superblock\n", | |
1175 | __func__); | |
cd9cb405 | 1176 | goto err_cleanup; |
f0c9fd54 GT |
1177 | } |
1178 | journal->j_sb_buffer = bh; | |
1179 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
1180 | ||
470decc6 | 1181 | return journal; |
cd9cb405 EB |
1182 | |
1183 | err_cleanup: | |
1184 | kfree(journal->j_wbuf); | |
1185 | jbd2_journal_destroy_revoke(journal); | |
1186 | kfree(journal); | |
1187 | return NULL; | |
470decc6 DK |
1188 | } |
1189 | ||
f7f4bccb | 1190 | /* jbd2_journal_init_dev and jbd2_journal_init_inode: |
470decc6 DK |
1191 | * |
1192 | * Create a journal structure assigned some fixed set of disk blocks to | |
1193 | * the journal. We don't actually touch those disk blocks yet, but we | |
1194 | * need to set up all of the mapping information to tell the journaling | |
1195 | * system where the journal blocks are. | |
1196 | * | |
1197 | */ | |
1198 | ||
1199 | /** | |
5648ba5b | 1200 | * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure |
470decc6 DK |
1201 | * @bdev: Block device on which to create the journal |
1202 | * @fs_dev: Device which hold journalled filesystem for this journal. | |
1203 | * @start: Block nr Start of journal. | |
1204 | * @len: Length of the journal in blocks. | |
1205 | * @blocksize: blocksize of journalling device | |
5648ba5b RD |
1206 | * |
1207 | * Returns: a newly created journal_t * | |
470decc6 | 1208 | * |
f7f4bccb | 1209 | * jbd2_journal_init_dev creates a journal which maps a fixed contiguous |
470decc6 DK |
1210 | * range of blocks on an arbitrary block device. |
1211 | * | |
1212 | */ | |
f0c9fd54 | 1213 | journal_t *jbd2_journal_init_dev(struct block_device *bdev, |
470decc6 | 1214 | struct block_device *fs_dev, |
18eba7aa | 1215 | unsigned long long start, int len, int blocksize) |
470decc6 | 1216 | { |
f0c9fd54 | 1217 | journal_t *journal; |
470decc6 | 1218 | |
f0c9fd54 | 1219 | journal = journal_init_common(bdev, fs_dev, start, len, blocksize); |
470decc6 DK |
1220 | if (!journal) |
1221 | return NULL; | |
1222 | ||
0587aa3d | 1223 | bdevname(journal->j_dev, journal->j_devname); |
81ae394b | 1224 | strreplace(journal->j_devname, '/', '!'); |
4b905671 | 1225 | jbd2_stats_proc_init(journal); |
4b905671 | 1226 | |
470decc6 DK |
1227 | return journal; |
1228 | } | |
1229 | ||
1230 | /** | |
f7f4bccb | 1231 | * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. |
470decc6 DK |
1232 | * @inode: An inode to create the journal in |
1233 | * | |
f7f4bccb | 1234 | * jbd2_journal_init_inode creates a journal which maps an on-disk inode as |
470decc6 DK |
1235 | * the journal. The inode must exist already, must support bmap() and |
1236 | * must have all data blocks preallocated. | |
1237 | */ | |
f0c9fd54 | 1238 | journal_t *jbd2_journal_init_inode(struct inode *inode) |
470decc6 | 1239 | { |
f0c9fd54 | 1240 | journal_t *journal; |
05496769 | 1241 | char *p; |
18eba7aa | 1242 | unsigned long long blocknr; |
470decc6 | 1243 | |
f0c9fd54 GT |
1244 | blocknr = bmap(inode, 0); |
1245 | if (!blocknr) { | |
1246 | pr_err("%s: Cannot locate journal superblock\n", | |
1247 | __func__); | |
1248 | return NULL; | |
1249 | } | |
1250 | ||
1251 | jbd_debug(1, "JBD2: inode %s/%ld, size %lld, bits %d, blksize %ld\n", | |
1252 | inode->i_sb->s_id, inode->i_ino, (long long) inode->i_size, | |
1253 | inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); | |
1254 | ||
1255 | journal = journal_init_common(inode->i_sb->s_bdev, inode->i_sb->s_bdev, | |
1256 | blocknr, inode->i_size >> inode->i_sb->s_blocksize_bits, | |
1257 | inode->i_sb->s_blocksize); | |
470decc6 DK |
1258 | if (!journal) |
1259 | return NULL; | |
1260 | ||
470decc6 | 1261 | journal->j_inode = inode; |
05496769 | 1262 | bdevname(journal->j_dev, journal->j_devname); |
81ae394b | 1263 | p = strreplace(journal->j_devname, '/', '!'); |
90576c0b | 1264 | sprintf(p, "-%lu", journal->j_inode->i_ino); |
8e85fb3f | 1265 | jbd2_stats_proc_init(journal); |
470decc6 | 1266 | |
470decc6 DK |
1267 | return journal; |
1268 | } | |
1269 | ||
1270 | /* | |
1271 | * If the journal init or create aborts, we need to mark the journal | |
1272 | * superblock as being NULL to prevent the journal destroy from writing | |
1273 | * back a bogus superblock. | |
1274 | */ | |
1275 | static void journal_fail_superblock (journal_t *journal) | |
1276 | { | |
1277 | struct buffer_head *bh = journal->j_sb_buffer; | |
1278 | brelse(bh); | |
1279 | journal->j_sb_buffer = NULL; | |
1280 | } | |
1281 | ||
1282 | /* | |
1283 | * Given a journal_t structure, initialise the various fields for | |
1284 | * startup of a new journaling session. We use this both when creating | |
1285 | * a journal, and after recovering an old journal to reset it for | |
1286 | * subsequent use. | |
1287 | */ | |
1288 | ||
1289 | static int journal_reset(journal_t *journal) | |
1290 | { | |
1291 | journal_superblock_t *sb = journal->j_superblock; | |
18eba7aa | 1292 | unsigned long long first, last; |
470decc6 DK |
1293 | |
1294 | first = be32_to_cpu(sb->s_first); | |
1295 | last = be32_to_cpu(sb->s_maxlen); | |
f6f50e28 | 1296 | if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) { |
f2a44523 | 1297 | printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n", |
f6f50e28 JK |
1298 | first, last); |
1299 | journal_fail_superblock(journal); | |
1300 | return -EINVAL; | |
1301 | } | |
470decc6 DK |
1302 | |
1303 | journal->j_first = first; | |
1304 | journal->j_last = last; | |
1305 | ||
1306 | journal->j_head = first; | |
1307 | journal->j_tail = first; | |
1308 | journal->j_free = last - first; | |
1309 | ||
1310 | journal->j_tail_sequence = journal->j_transaction_sequence; | |
1311 | journal->j_commit_sequence = journal->j_transaction_sequence - 1; | |
1312 | journal->j_commit_request = journal->j_commit_sequence; | |
1313 | ||
1314 | journal->j_max_transaction_buffers = journal->j_maxlen / 4; | |
1315 | ||
470decc6 DK |
1316 | /* |
1317 | * As a special case, if the on-disk copy is already marked as needing | |
24bcc89c JK |
1318 | * no recovery (s_start == 0), then we can safely defer the superblock |
1319 | * update until the next commit by setting JBD2_FLUSHED. This avoids | |
470decc6 DK |
1320 | * attempting a write to a potential-readonly device. |
1321 | */ | |
24bcc89c | 1322 | if (sb->s_start == 0) { |
f2a44523 | 1323 | jbd_debug(1, "JBD2: Skipping superblock update on recovered sb " |
470decc6 DK |
1324 | "(start %ld, seq %d, errno %d)\n", |
1325 | journal->j_tail, journal->j_tail_sequence, | |
1326 | journal->j_errno); | |
24bcc89c JK |
1327 | journal->j_flags |= JBD2_FLUSHED; |
1328 | } else { | |
a78bb11d | 1329 | /* Lock here to make assertions happy... */ |
6fa7aa50 | 1330 | mutex_lock_io(&journal->j_checkpoint_mutex); |
79feb521 | 1331 | /* |
70fd7614 | 1332 | * Update log tail information. We use REQ_FUA since new |
79feb521 JK |
1333 | * transaction will start reusing journal space and so we |
1334 | * must make sure information about current log tail is on | |
1335 | * disk before that. | |
1336 | */ | |
1337 | jbd2_journal_update_sb_log_tail(journal, | |
1338 | journal->j_tail_sequence, | |
1339 | journal->j_tail, | |
17f423b5 | 1340 | REQ_SYNC | REQ_FUA); |
a78bb11d | 1341 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 | 1342 | } |
24bcc89c JK |
1343 | return jbd2_journal_start_thread(journal); |
1344 | } | |
470decc6 | 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 | |
2a222ca9 | 1352 | trace_jbd2_write_superblock(journal, write_flags); |
79feb521 | 1353 | if (!(journal->j_flags & JBD2_BARRIER)) |
28a8f0d3 | 1354 | write_flags &= ~(REQ_FUA | REQ_PREFLUSH); |
79feb521 | 1355 | lock_buffer(bh); |
914258bf TT |
1356 | if (buffer_write_io_error(bh)) { |
1357 | /* | |
1358 | * Oh, dear. A previous attempt to write the journal | |
1359 | * superblock failed. This could happen because the | |
1360 | * USB device was yanked out. Or it could happen to | |
1361 | * be a transient write error and maybe the block will | |
1362 | * be remapped. Nothing we can do but to retry the | |
1363 | * write and hope for the best. | |
1364 | */ | |
1365 | printk(KERN_ERR "JBD2: previous I/O error detected " | |
1366 | "for journal superblock update for %s.\n", | |
1367 | journal->j_devname); | |
1368 | clear_buffer_write_io_error(bh); | |
1369 | set_buffer_uptodate(bh); | |
1370 | } | |
fe52d17c | 1371 | jbd2_superblock_csum_set(journal, sb); |
79feb521 JK |
1372 | get_bh(bh); |
1373 | bh->b_end_io = end_buffer_write_sync; | |
2a222ca9 | 1374 | ret = submit_bh(REQ_OP_WRITE, write_flags, bh); |
79feb521 | 1375 | wait_on_buffer(bh); |
24bcc89c | 1376 | if (buffer_write_io_error(bh)) { |
24bcc89c JK |
1377 | clear_buffer_write_io_error(bh); |
1378 | set_buffer_uptodate(bh); | |
79feb521 JK |
1379 | ret = -EIO; |
1380 | } | |
1381 | if (ret) { | |
1382 | printk(KERN_ERR "JBD2: Error %d detected when updating " | |
1383 | "journal superblock for %s.\n", ret, | |
1384 | journal->j_devname); | |
6f6a6fda | 1385 | jbd2_journal_abort(journal, ret); |
24bcc89c | 1386 | } |
6f6a6fda JQ |
1387 | |
1388 | return ret; | |
24bcc89c JK |
1389 | } |
1390 | ||
1391 | /** | |
1392 | * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. | |
1393 | * @journal: The journal to update. | |
79feb521 JK |
1394 | * @tail_tid: TID of the new transaction at the tail of the log |
1395 | * @tail_block: The first block of the transaction at the tail of the log | |
1396 | * @write_op: With which operation should we write the journal sb | |
24bcc89c JK |
1397 | * |
1398 | * Update a journal's superblock information about log tail and write it to | |
1399 | * disk, waiting for the IO to complete. | |
1400 | */ | |
6f6a6fda | 1401 | int jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, |
79feb521 | 1402 | unsigned long tail_block, int write_op) |
24bcc89c JK |
1403 | { |
1404 | journal_superblock_t *sb = journal->j_superblock; | |
6f6a6fda | 1405 | int ret; |
24bcc89c | 1406 | |
a78bb11d | 1407 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); |
79feb521 JK |
1408 | jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", |
1409 | tail_block, tail_tid); | |
470decc6 | 1410 | |
79feb521 JK |
1411 | sb->s_sequence = cpu_to_be32(tail_tid); |
1412 | sb->s_start = cpu_to_be32(tail_block); | |
470decc6 | 1413 | |
6f6a6fda JQ |
1414 | ret = jbd2_write_superblock(journal, write_op); |
1415 | if (ret) | |
1416 | goto out; | |
470decc6 | 1417 | |
24bcc89c JK |
1418 | /* Log is no longer empty */ |
1419 | write_lock(&journal->j_state_lock); | |
1420 | WARN_ON(!sb->s_sequence); | |
1421 | journal->j_flags &= ~JBD2_FLUSHED; | |
1422 | write_unlock(&journal->j_state_lock); | |
6f6a6fda JQ |
1423 | |
1424 | out: | |
1425 | return ret; | |
24bcc89c | 1426 | } |
2201c590 | 1427 | |
24bcc89c JK |
1428 | /** |
1429 | * jbd2_mark_journal_empty() - Mark on disk journal as empty. | |
1430 | * @journal: The journal to update. | |
c0a2ad9b | 1431 | * @write_op: With which operation should we write the journal sb |
24bcc89c JK |
1432 | * |
1433 | * Update a journal's dynamic superblock fields to show that journal is empty. | |
1434 | * Write updated superblock to disk waiting for IO to complete. | |
1435 | */ | |
c0a2ad9b | 1436 | static void jbd2_mark_journal_empty(journal_t *journal, int write_op) |
24bcc89c JK |
1437 | { |
1438 | journal_superblock_t *sb = journal->j_superblock; | |
914258bf | 1439 | |
a78bb11d | 1440 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); |
a931da6a | 1441 | read_lock(&journal->j_state_lock); |
eeecef0a ES |
1442 | /* Is it already empty? */ |
1443 | if (sb->s_start == 0) { | |
1444 | read_unlock(&journal->j_state_lock); | |
1445 | return; | |
1446 | } | |
24bcc89c JK |
1447 | jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n", |
1448 | journal->j_tail_sequence); | |
470decc6 DK |
1449 | |
1450 | sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); | |
24bcc89c | 1451 | sb->s_start = cpu_to_be32(0); |
a931da6a | 1452 | read_unlock(&journal->j_state_lock); |
470decc6 | 1453 | |
c0a2ad9b | 1454 | jbd2_write_superblock(journal, write_op); |
470decc6 | 1455 | |
24bcc89c | 1456 | /* Log is no longer empty */ |
a931da6a | 1457 | write_lock(&journal->j_state_lock); |
24bcc89c | 1458 | journal->j_flags |= JBD2_FLUSHED; |
a931da6a | 1459 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1460 | } |
1461 | ||
24bcc89c JK |
1462 | |
1463 | /** | |
1464 | * jbd2_journal_update_sb_errno() - Update error in the journal. | |
1465 | * @journal: The journal to update. | |
1466 | * | |
1467 | * Update a journal's errno. Write updated superblock to disk waiting for IO | |
1468 | * to complete. | |
1469 | */ | |
d796c52e | 1470 | void jbd2_journal_update_sb_errno(journal_t *journal) |
24bcc89c JK |
1471 | { |
1472 | journal_superblock_t *sb = journal->j_superblock; | |
1473 | ||
1474 | read_lock(&journal->j_state_lock); | |
1475 | jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", | |
1476 | journal->j_errno); | |
1477 | sb->s_errno = cpu_to_be32(journal->j_errno); | |
1478 | read_unlock(&journal->j_state_lock); | |
1479 | ||
17f423b5 | 1480 | jbd2_write_superblock(journal, REQ_SYNC | REQ_FUA); |
24bcc89c | 1481 | } |
d796c52e | 1482 | EXPORT_SYMBOL(jbd2_journal_update_sb_errno); |
24bcc89c | 1483 | |
470decc6 DK |
1484 | /* |
1485 | * Read the superblock for a given journal, performing initial | |
1486 | * validation of the format. | |
1487 | */ | |
470decc6 DK |
1488 | static int journal_get_superblock(journal_t *journal) |
1489 | { | |
1490 | struct buffer_head *bh; | |
1491 | journal_superblock_t *sb; | |
1492 | int err = -EIO; | |
1493 | ||
1494 | bh = journal->j_sb_buffer; | |
1495 | ||
1496 | J_ASSERT(bh != NULL); | |
1497 | if (!buffer_uptodate(bh)) { | |
dfec8a14 | 1498 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
470decc6 DK |
1499 | wait_on_buffer(bh); |
1500 | if (!buffer_uptodate(bh)) { | |
f2a44523 EG |
1501 | printk(KERN_ERR |
1502 | "JBD2: IO error reading journal superblock\n"); | |
470decc6 DK |
1503 | goto out; |
1504 | } | |
1505 | } | |
1506 | ||
25ed6e8a DW |
1507 | if (buffer_verified(bh)) |
1508 | return 0; | |
1509 | ||
470decc6 DK |
1510 | sb = journal->j_superblock; |
1511 | ||
1512 | err = -EINVAL; | |
1513 | ||
f7f4bccb | 1514 | if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || |
470decc6 | 1515 | sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { |
f2a44523 | 1516 | printk(KERN_WARNING "JBD2: no valid journal superblock found\n"); |
470decc6 DK |
1517 | goto out; |
1518 | } | |
1519 | ||
1520 | switch(be32_to_cpu(sb->s_header.h_blocktype)) { | |
f7f4bccb | 1521 | case JBD2_SUPERBLOCK_V1: |
470decc6 DK |
1522 | journal->j_format_version = 1; |
1523 | break; | |
f7f4bccb | 1524 | case JBD2_SUPERBLOCK_V2: |
470decc6 DK |
1525 | journal->j_format_version = 2; |
1526 | break; | |
1527 | default: | |
f2a44523 | 1528 | printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n"); |
470decc6 DK |
1529 | goto out; |
1530 | } | |
1531 | ||
1532 | if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen) | |
1533 | journal->j_maxlen = be32_to_cpu(sb->s_maxlen); | |
1534 | else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) { | |
f2a44523 | 1535 | printk(KERN_WARNING "JBD2: journal file too short\n"); |
470decc6 DK |
1536 | goto out; |
1537 | } | |
1538 | ||
8762202d EG |
1539 | if (be32_to_cpu(sb->s_first) == 0 || |
1540 | be32_to_cpu(sb->s_first) >= journal->j_maxlen) { | |
1541 | printk(KERN_WARNING | |
1542 | "JBD2: Invalid start block of journal: %u\n", | |
1543 | be32_to_cpu(sb->s_first)); | |
1544 | goto out; | |
1545 | } | |
1546 | ||
56316a0d DW |
1547 | if (jbd2_has_feature_csum2(journal) && |
1548 | jbd2_has_feature_csum3(journal)) { | |
db9ee220 DW |
1549 | /* Can't have checksum v2 and v3 at the same time! */ |
1550 | printk(KERN_ERR "JBD2: Can't enable checksumming v2 and v3 " | |
1551 | "at the same time!\n"); | |
1552 | goto out; | |
1553 | } | |
1554 | ||
8595798c | 1555 | if (jbd2_journal_has_csum_v2or3_feature(journal) && |
56316a0d | 1556 | jbd2_has_feature_checksum(journal)) { |
feb8c6d3 DW |
1557 | /* Can't have checksum v1 and v2 on at the same time! */ |
1558 | printk(KERN_ERR "JBD2: Can't enable checksumming v1 and v2/3 " | |
1559 | "at the same time!\n"); | |
1560 | goto out; | |
1561 | } | |
1562 | ||
25ed6e8a | 1563 | if (!jbd2_verify_csum_type(journal, sb)) { |
a67c848a | 1564 | printk(KERN_ERR "JBD2: Unknown checksum type\n"); |
25ed6e8a DW |
1565 | goto out; |
1566 | } | |
1567 | ||
01b5adce | 1568 | /* Load the checksum driver */ |
8595798c | 1569 | if (jbd2_journal_has_csum_v2or3_feature(journal)) { |
01b5adce DW |
1570 | journal->j_chksum_driver = crypto_alloc_shash("crc32c", 0, 0); |
1571 | if (IS_ERR(journal->j_chksum_driver)) { | |
a67c848a | 1572 | printk(KERN_ERR "JBD2: Cannot load crc32c driver.\n"); |
01b5adce DW |
1573 | err = PTR_ERR(journal->j_chksum_driver); |
1574 | journal->j_chksum_driver = NULL; | |
1575 | goto out; | |
1576 | } | |
1577 | } | |
1578 | ||
4fd5ea43 DW |
1579 | /* Check superblock checksum */ |
1580 | if (!jbd2_superblock_csum_verify(journal, sb)) { | |
a67c848a | 1581 | printk(KERN_ERR "JBD2: journal checksum error\n"); |
6a797d27 | 1582 | err = -EFSBADCRC; |
4fd5ea43 DW |
1583 | goto out; |
1584 | } | |
1585 | ||
1586 | /* Precompute checksum seed for all metadata */ | |
db9ee220 | 1587 | if (jbd2_journal_has_csum_v2or3(journal)) |
4fd5ea43 DW |
1588 | journal->j_csum_seed = jbd2_chksum(journal, ~0, sb->s_uuid, |
1589 | sizeof(sb->s_uuid)); | |
1590 | ||
25ed6e8a DW |
1591 | set_buffer_verified(bh); |
1592 | ||
470decc6 DK |
1593 | return 0; |
1594 | ||
1595 | out: | |
1596 | journal_fail_superblock(journal); | |
1597 | return err; | |
1598 | } | |
1599 | ||
1600 | /* | |
1601 | * Load the on-disk journal superblock and read the key fields into the | |
1602 | * journal_t. | |
1603 | */ | |
1604 | ||
1605 | static int load_superblock(journal_t *journal) | |
1606 | { | |
1607 | int err; | |
1608 | journal_superblock_t *sb; | |
1609 | ||
1610 | err = journal_get_superblock(journal); | |
1611 | if (err) | |
1612 | return err; | |
1613 | ||
1614 | sb = journal->j_superblock; | |
1615 | ||
1616 | journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); | |
1617 | journal->j_tail = be32_to_cpu(sb->s_start); | |
1618 | journal->j_first = be32_to_cpu(sb->s_first); | |
1619 | journal->j_last = be32_to_cpu(sb->s_maxlen); | |
1620 | journal->j_errno = be32_to_cpu(sb->s_errno); | |
1621 | ||
1622 | return 0; | |
1623 | } | |
1624 | ||
1625 | ||
1626 | /** | |
f7f4bccb | 1627 | * int jbd2_journal_load() - Read journal from disk. |
470decc6 DK |
1628 | * @journal: Journal to act on. |
1629 | * | |
1630 | * Given a journal_t structure which tells us which disk blocks contain | |
1631 | * a journal, read the journal from disk to initialise the in-memory | |
1632 | * structures. | |
1633 | */ | |
f7f4bccb | 1634 | int jbd2_journal_load(journal_t *journal) |
470decc6 DK |
1635 | { |
1636 | int err; | |
1637 | journal_superblock_t *sb; | |
1638 | ||
1639 | err = load_superblock(journal); | |
1640 | if (err) | |
1641 | return err; | |
1642 | ||
1643 | sb = journal->j_superblock; | |
1644 | /* If this is a V2 superblock, then we have to check the | |
1645 | * features flags on it. */ | |
1646 | ||
1647 | if (journal->j_format_version >= 2) { | |
1648 | if ((sb->s_feature_ro_compat & | |
f7f4bccb | 1649 | ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || |
470decc6 | 1650 | (sb->s_feature_incompat & |
f7f4bccb | 1651 | ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { |
f2a44523 EG |
1652 | printk(KERN_WARNING |
1653 | "JBD2: Unrecognised features on journal\n"); | |
470decc6 DK |
1654 | return -EINVAL; |
1655 | } | |
1656 | } | |
1657 | ||
d2eecb03 TT |
1658 | /* |
1659 | * Create a slab for this blocksize | |
1660 | */ | |
1661 | err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize)); | |
1662 | if (err) | |
1663 | return err; | |
1664 | ||
470decc6 DK |
1665 | /* Let the recovery code check whether it needs to recover any |
1666 | * data from the journal. */ | |
f7f4bccb | 1667 | if (jbd2_journal_recover(journal)) |
470decc6 DK |
1668 | goto recovery_error; |
1669 | ||
e6a47428 TT |
1670 | if (journal->j_failed_commit) { |
1671 | printk(KERN_ERR "JBD2: journal transaction %u on %s " | |
1672 | "is corrupt.\n", journal->j_failed_commit, | |
1673 | journal->j_devname); | |
6a797d27 | 1674 | return -EFSCORRUPTED; |
e6a47428 TT |
1675 | } |
1676 | ||
470decc6 DK |
1677 | /* OK, we've finished with the dynamic journal bits: |
1678 | * reinitialise the dynamic contents of the superblock in memory | |
1679 | * and reset them on disk. */ | |
1680 | if (journal_reset(journal)) | |
1681 | goto recovery_error; | |
1682 | ||
f7f4bccb MC |
1683 | journal->j_flags &= ~JBD2_ABORT; |
1684 | journal->j_flags |= JBD2_LOADED; | |
470decc6 DK |
1685 | return 0; |
1686 | ||
1687 | recovery_error: | |
f2a44523 | 1688 | printk(KERN_WARNING "JBD2: recovery failed\n"); |
470decc6 DK |
1689 | return -EIO; |
1690 | } | |
1691 | ||
1692 | /** | |
f7f4bccb | 1693 | * void jbd2_journal_destroy() - Release a journal_t structure. |
470decc6 DK |
1694 | * @journal: Journal to act on. |
1695 | * | |
1696 | * Release a journal_t structure once it is no longer in use by the | |
1697 | * journaled object. | |
44519faf | 1698 | * Return <0 if we couldn't clean up the journal. |
470decc6 | 1699 | */ |
44519faf | 1700 | int jbd2_journal_destroy(journal_t *journal) |
470decc6 | 1701 | { |
44519faf HK |
1702 | int err = 0; |
1703 | ||
470decc6 DK |
1704 | /* Wait for the commit thread to wake up and die. */ |
1705 | journal_kill_thread(journal); | |
1706 | ||
1707 | /* Force a final log commit */ | |
1708 | if (journal->j_running_transaction) | |
f7f4bccb | 1709 | jbd2_journal_commit_transaction(journal); |
470decc6 DK |
1710 | |
1711 | /* Force any old transactions to disk */ | |
1712 | ||
1713 | /* Totally anal locking here... */ | |
1714 | spin_lock(&journal->j_list_lock); | |
1715 | while (journal->j_checkpoint_transactions != NULL) { | |
1716 | spin_unlock(&journal->j_list_lock); | |
6fa7aa50 | 1717 | mutex_lock_io(&journal->j_checkpoint_mutex); |
841df7df | 1718 | err = jbd2_log_do_checkpoint(journal); |
1a0d3786 | 1719 | mutex_unlock(&journal->j_checkpoint_mutex); |
841df7df JK |
1720 | /* |
1721 | * If checkpointing failed, just free the buffers to avoid | |
1722 | * looping forever | |
1723 | */ | |
1724 | if (err) { | |
1725 | jbd2_journal_destroy_checkpoint(journal); | |
1726 | spin_lock(&journal->j_list_lock); | |
1727 | break; | |
1728 | } | |
470decc6 DK |
1729 | spin_lock(&journal->j_list_lock); |
1730 | } | |
1731 | ||
1732 | J_ASSERT(journal->j_running_transaction == NULL); | |
1733 | J_ASSERT(journal->j_committing_transaction == NULL); | |
1734 | J_ASSERT(journal->j_checkpoint_transactions == NULL); | |
1735 | spin_unlock(&journal->j_list_lock); | |
1736 | ||
470decc6 | 1737 | if (journal->j_sb_buffer) { |
44519faf | 1738 | if (!is_journal_aborted(journal)) { |
6fa7aa50 | 1739 | mutex_lock_io(&journal->j_checkpoint_mutex); |
c0a2ad9b OH |
1740 | |
1741 | write_lock(&journal->j_state_lock); | |
1742 | journal->j_tail_sequence = | |
1743 | ++journal->j_transaction_sequence; | |
1744 | write_unlock(&journal->j_state_lock); | |
1745 | ||
70fd7614 | 1746 | jbd2_mark_journal_empty(journal, |
17f423b5 | 1747 | REQ_SYNC | REQ_PREFLUSH | REQ_FUA); |
a78bb11d JK |
1748 | mutex_unlock(&journal->j_checkpoint_mutex); |
1749 | } else | |
44519faf | 1750 | err = -EIO; |
470decc6 DK |
1751 | brelse(journal->j_sb_buffer); |
1752 | } | |
1753 | ||
8e85fb3f JL |
1754 | if (journal->j_proc_entry) |
1755 | jbd2_stats_proc_exit(journal); | |
d9f39d1e | 1756 | iput(journal->j_inode); |
470decc6 | 1757 | if (journal->j_revoke) |
f7f4bccb | 1758 | jbd2_journal_destroy_revoke(journal); |
01b5adce DW |
1759 | if (journal->j_chksum_driver) |
1760 | crypto_free_shash(journal->j_chksum_driver); | |
470decc6 DK |
1761 | kfree(journal->j_wbuf); |
1762 | kfree(journal); | |
44519faf HK |
1763 | |
1764 | return err; | |
470decc6 DK |
1765 | } |
1766 | ||
1767 | ||
1768 | /** | |
f7f4bccb | 1769 | *int jbd2_journal_check_used_features () - Check if features specified are used. |
470decc6 DK |
1770 | * @journal: Journal to check. |
1771 | * @compat: bitmask of compatible features | |
1772 | * @ro: bitmask of features that force read-only mount | |
1773 | * @incompat: bitmask of incompatible features | |
1774 | * | |
1775 | * Check whether the journal uses all of a given set of | |
1776 | * features. Return true (non-zero) if it does. | |
1777 | **/ | |
1778 | ||
f7f4bccb | 1779 | int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1780 | unsigned long ro, unsigned long incompat) |
1781 | { | |
1782 | journal_superblock_t *sb; | |
1783 | ||
1784 | if (!compat && !ro && !incompat) | |
1785 | return 1; | |
1113e1b5 PL |
1786 | /* Load journal superblock if it is not loaded yet. */ |
1787 | if (journal->j_format_version == 0 && | |
1788 | journal_get_superblock(journal) != 0) | |
1789 | return 0; | |
470decc6 DK |
1790 | if (journal->j_format_version == 1) |
1791 | return 0; | |
1792 | ||
1793 | sb = journal->j_superblock; | |
1794 | ||
1795 | if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && | |
1796 | ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && | |
1797 | ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) | |
1798 | return 1; | |
1799 | ||
1800 | return 0; | |
1801 | } | |
1802 | ||
1803 | /** | |
f7f4bccb | 1804 | * int jbd2_journal_check_available_features() - Check feature set in journalling layer |
470decc6 DK |
1805 | * @journal: Journal to check. |
1806 | * @compat: bitmask of compatible features | |
1807 | * @ro: bitmask of features that force read-only mount | |
1808 | * @incompat: bitmask of incompatible features | |
1809 | * | |
1810 | * Check whether the journaling code supports the use of | |
1811 | * all of a given set of features on this journal. Return true | |
1812 | * (non-zero) if it can. */ | |
1813 | ||
f7f4bccb | 1814 | int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1815 | unsigned long ro, unsigned long incompat) |
1816 | { | |
470decc6 DK |
1817 | if (!compat && !ro && !incompat) |
1818 | return 1; | |
1819 | ||
470decc6 DK |
1820 | /* We can support any known requested features iff the |
1821 | * superblock is in version 2. Otherwise we fail to support any | |
1822 | * extended sb features. */ | |
1823 | ||
1824 | if (journal->j_format_version != 2) | |
1825 | return 0; | |
1826 | ||
f7f4bccb MC |
1827 | if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && |
1828 | (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && | |
1829 | (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) | |
470decc6 DK |
1830 | return 1; |
1831 | ||
1832 | return 0; | |
1833 | } | |
1834 | ||
1835 | /** | |
f7f4bccb | 1836 | * int jbd2_journal_set_features () - Mark a given journal feature in the superblock |
470decc6 DK |
1837 | * @journal: Journal to act on. |
1838 | * @compat: bitmask of compatible features | |
1839 | * @ro: bitmask of features that force read-only mount | |
1840 | * @incompat: bitmask of incompatible features | |
1841 | * | |
1842 | * Mark a given journal feature as present on the | |
1843 | * superblock. Returns true if the requested features could be set. | |
1844 | * | |
1845 | */ | |
1846 | ||
f7f4bccb | 1847 | int jbd2_journal_set_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1848 | unsigned long ro, unsigned long incompat) |
1849 | { | |
25ed6e8a DW |
1850 | #define INCOMPAT_FEATURE_ON(f) \ |
1851 | ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f))) | |
1852 | #define COMPAT_FEATURE_ON(f) \ | |
1853 | ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f))) | |
470decc6 DK |
1854 | journal_superblock_t *sb; |
1855 | ||
f7f4bccb | 1856 | if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) |
470decc6 DK |
1857 | return 1; |
1858 | ||
f7f4bccb | 1859 | if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) |
470decc6 DK |
1860 | return 0; |
1861 | ||
db9ee220 DW |
1862 | /* If enabling v2 checksums, turn on v3 instead */ |
1863 | if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2) { | |
1864 | incompat &= ~JBD2_FEATURE_INCOMPAT_CSUM_V2; | |
1865 | incompat |= JBD2_FEATURE_INCOMPAT_CSUM_V3; | |
1866 | } | |
1867 | ||
1868 | /* Asking for checksumming v3 and v1? Only give them v3. */ | |
1869 | if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V3 && | |
25ed6e8a DW |
1870 | compat & JBD2_FEATURE_COMPAT_CHECKSUM) |
1871 | compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM; | |
1872 | ||
470decc6 DK |
1873 | jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", |
1874 | compat, ro, incompat); | |
1875 | ||
1876 | sb = journal->j_superblock; | |
1877 | ||
db9ee220 DW |
1878 | /* If enabling v3 checksums, update superblock */ |
1879 | if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V3)) { | |
25ed6e8a DW |
1880 | sb->s_checksum_type = JBD2_CRC32C_CHKSUM; |
1881 | sb->s_feature_compat &= | |
1882 | ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM); | |
01b5adce DW |
1883 | |
1884 | /* Load the checksum driver */ | |
1885 | if (journal->j_chksum_driver == NULL) { | |
1886 | journal->j_chksum_driver = crypto_alloc_shash("crc32c", | |
1887 | 0, 0); | |
1888 | if (IS_ERR(journal->j_chksum_driver)) { | |
a67c848a | 1889 | printk(KERN_ERR "JBD2: Cannot load crc32c " |
01b5adce DW |
1890 | "driver.\n"); |
1891 | journal->j_chksum_driver = NULL; | |
1892 | return 0; | |
1893 | } | |
4fd5ea43 | 1894 | |
32f38691 | 1895 | /* Precompute checksum seed for all metadata */ |
4fd5ea43 DW |
1896 | journal->j_csum_seed = jbd2_chksum(journal, ~0, |
1897 | sb->s_uuid, | |
1898 | sizeof(sb->s_uuid)); | |
32f38691 | 1899 | } |
25ed6e8a DW |
1900 | } |
1901 | ||
1902 | /* If enabling v1 checksums, downgrade superblock */ | |
1903 | if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM)) | |
1904 | sb->s_feature_incompat &= | |
db9ee220 DW |
1905 | ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2 | |
1906 | JBD2_FEATURE_INCOMPAT_CSUM_V3); | |
25ed6e8a | 1907 | |
470decc6 DK |
1908 | sb->s_feature_compat |= cpu_to_be32(compat); |
1909 | sb->s_feature_ro_compat |= cpu_to_be32(ro); | |
1910 | sb->s_feature_incompat |= cpu_to_be32(incompat); | |
1911 | ||
1912 | return 1; | |
25ed6e8a DW |
1913 | #undef COMPAT_FEATURE_ON |
1914 | #undef INCOMPAT_FEATURE_ON | |
470decc6 DK |
1915 | } |
1916 | ||
818d276c GS |
1917 | /* |
1918 | * jbd2_journal_clear_features () - Clear a given journal feature in the | |
1919 | * superblock | |
1920 | * @journal: Journal to act on. | |
1921 | * @compat: bitmask of compatible features | |
1922 | * @ro: bitmask of features that force read-only mount | |
1923 | * @incompat: bitmask of incompatible features | |
1924 | * | |
1925 | * Clear a given journal feature as present on the | |
1926 | * superblock. | |
1927 | */ | |
1928 | void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, | |
1929 | unsigned long ro, unsigned long incompat) | |
1930 | { | |
1931 | journal_superblock_t *sb; | |
1932 | ||
1933 | jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", | |
1934 | compat, ro, incompat); | |
1935 | ||
1936 | sb = journal->j_superblock; | |
1937 | ||
1938 | sb->s_feature_compat &= ~cpu_to_be32(compat); | |
1939 | sb->s_feature_ro_compat &= ~cpu_to_be32(ro); | |
1940 | sb->s_feature_incompat &= ~cpu_to_be32(incompat); | |
1941 | } | |
1942 | EXPORT_SYMBOL(jbd2_journal_clear_features); | |
470decc6 | 1943 | |
470decc6 | 1944 | /** |
f7f4bccb | 1945 | * int jbd2_journal_flush () - Flush journal |
470decc6 DK |
1946 | * @journal: Journal to act on. |
1947 | * | |
1948 | * Flush all data for a given journal to disk and empty the journal. | |
1949 | * Filesystems can use this when remounting readonly to ensure that | |
1950 | * recovery does not need to happen on remount. | |
1951 | */ | |
1952 | ||
f7f4bccb | 1953 | int jbd2_journal_flush(journal_t *journal) |
470decc6 DK |
1954 | { |
1955 | int err = 0; | |
1956 | transaction_t *transaction = NULL; | |
470decc6 | 1957 | |
a931da6a | 1958 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1959 | |
1960 | /* Force everything buffered to the log... */ | |
1961 | if (journal->j_running_transaction) { | |
1962 | transaction = journal->j_running_transaction; | |
f7f4bccb | 1963 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1964 | } else if (journal->j_committing_transaction) |
1965 | transaction = journal->j_committing_transaction; | |
1966 | ||
1967 | /* Wait for the log commit to complete... */ | |
1968 | if (transaction) { | |
1969 | tid_t tid = transaction->t_tid; | |
1970 | ||
a931da6a | 1971 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 1972 | jbd2_log_wait_commit(journal, tid); |
470decc6 | 1973 | } else { |
a931da6a | 1974 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1975 | } |
1976 | ||
1977 | /* ...and flush everything in the log out to disk. */ | |
1978 | spin_lock(&journal->j_list_lock); | |
1979 | while (!err && journal->j_checkpoint_transactions != NULL) { | |
1980 | spin_unlock(&journal->j_list_lock); | |
6fa7aa50 | 1981 | mutex_lock_io(&journal->j_checkpoint_mutex); |
f7f4bccb | 1982 | err = jbd2_log_do_checkpoint(journal); |
44519faf | 1983 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
1984 | spin_lock(&journal->j_list_lock); |
1985 | } | |
1986 | spin_unlock(&journal->j_list_lock); | |
44519faf HK |
1987 | |
1988 | if (is_journal_aborted(journal)) | |
1989 | return -EIO; | |
1990 | ||
6fa7aa50 | 1991 | mutex_lock_io(&journal->j_checkpoint_mutex); |
6f6a6fda JQ |
1992 | if (!err) { |
1993 | err = jbd2_cleanup_journal_tail(journal); | |
1994 | if (err < 0) { | |
1995 | mutex_unlock(&journal->j_checkpoint_mutex); | |
1996 | goto out; | |
1997 | } | |
1998 | err = 0; | |
1999 | } | |
470decc6 DK |
2000 | |
2001 | /* Finally, mark the journal as really needing no recovery. | |
2002 | * This sets s_start==0 in the underlying superblock, which is | |
2003 | * the magic code for a fully-recovered superblock. Any future | |
2004 | * commits of data to the journal will restore the current | |
2005 | * s_start value. */ | |
17f423b5 | 2006 | jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA); |
a78bb11d | 2007 | mutex_unlock(&journal->j_checkpoint_mutex); |
a931da6a | 2008 | write_lock(&journal->j_state_lock); |
470decc6 DK |
2009 | J_ASSERT(!journal->j_running_transaction); |
2010 | J_ASSERT(!journal->j_committing_transaction); | |
2011 | J_ASSERT(!journal->j_checkpoint_transactions); | |
2012 | J_ASSERT(journal->j_head == journal->j_tail); | |
2013 | J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); | |
a931da6a | 2014 | write_unlock(&journal->j_state_lock); |
6f6a6fda JQ |
2015 | out: |
2016 | return err; | |
470decc6 DK |
2017 | } |
2018 | ||
2019 | /** | |
f7f4bccb | 2020 | * int jbd2_journal_wipe() - Wipe journal contents |
470decc6 DK |
2021 | * @journal: Journal to act on. |
2022 | * @write: flag (see below) | |
2023 | * | |
2024 | * Wipe out all of the contents of a journal, safely. This will produce | |
2025 | * a warning if the journal contains any valid recovery information. | |
f7f4bccb | 2026 | * Must be called between journal_init_*() and jbd2_journal_load(). |
470decc6 DK |
2027 | * |
2028 | * If 'write' is non-zero, then we wipe out the journal on disk; otherwise | |
2029 | * we merely suppress recovery. | |
2030 | */ | |
2031 | ||
f7f4bccb | 2032 | int jbd2_journal_wipe(journal_t *journal, int write) |
470decc6 | 2033 | { |
470decc6 DK |
2034 | int err = 0; |
2035 | ||
f7f4bccb | 2036 | J_ASSERT (!(journal->j_flags & JBD2_LOADED)); |
470decc6 DK |
2037 | |
2038 | err = load_superblock(journal); | |
2039 | if (err) | |
2040 | return err; | |
2041 | ||
470decc6 DK |
2042 | if (!journal->j_tail) |
2043 | goto no_recovery; | |
2044 | ||
f2a44523 | 2045 | printk(KERN_WARNING "JBD2: %s recovery information on journal\n", |
470decc6 DK |
2046 | write ? "Clearing" : "Ignoring"); |
2047 | ||
f7f4bccb | 2048 | err = jbd2_journal_skip_recovery(journal); |
a78bb11d JK |
2049 | if (write) { |
2050 | /* Lock to make assertions happy... */ | |
2051 | mutex_lock(&journal->j_checkpoint_mutex); | |
17f423b5 | 2052 | jbd2_mark_journal_empty(journal, REQ_SYNC | REQ_FUA); |
a78bb11d JK |
2053 | mutex_unlock(&journal->j_checkpoint_mutex); |
2054 | } | |
470decc6 DK |
2055 | |
2056 | no_recovery: | |
2057 | return err; | |
2058 | } | |
2059 | ||
470decc6 DK |
2060 | /* |
2061 | * Journal abort has very specific semantics, which we describe | |
2062 | * for journal abort. | |
2063 | * | |
bfcd3555 | 2064 | * Two internal functions, which provide abort to the jbd layer |
470decc6 DK |
2065 | * itself are here. |
2066 | */ | |
2067 | ||
2068 | /* | |
2069 | * Quick version for internal journal use (doesn't lock the journal). | |
2070 | * Aborts hard --- we mark the abort as occurred, but do _nothing_ else, | |
2071 | * and don't attempt to make any other journal updates. | |
2072 | */ | |
f7f4bccb | 2073 | void __jbd2_journal_abort_hard(journal_t *journal) |
470decc6 DK |
2074 | { |
2075 | transaction_t *transaction; | |
470decc6 | 2076 | |
f7f4bccb | 2077 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2078 | return; |
2079 | ||
2080 | printk(KERN_ERR "Aborting journal on device %s.\n", | |
05496769 | 2081 | journal->j_devname); |
470decc6 | 2082 | |
a931da6a | 2083 | write_lock(&journal->j_state_lock); |
f7f4bccb | 2084 | journal->j_flags |= JBD2_ABORT; |
470decc6 DK |
2085 | transaction = journal->j_running_transaction; |
2086 | if (transaction) | |
f7f4bccb | 2087 | __jbd2_log_start_commit(journal, transaction->t_tid); |
a931da6a | 2088 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2089 | } |
2090 | ||
2091 | /* Soft abort: record the abort error status in the journal superblock, | |
2092 | * but don't do any other IO. */ | |
2093 | static void __journal_abort_soft (journal_t *journal, int errno) | |
2094 | { | |
f7f4bccb | 2095 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2096 | return; |
2097 | ||
2098 | if (!journal->j_errno) | |
2099 | journal->j_errno = errno; | |
2100 | ||
f7f4bccb | 2101 | __jbd2_journal_abort_hard(journal); |
470decc6 | 2102 | |
4327ba52 | 2103 | if (errno) { |
24bcc89c | 2104 | jbd2_journal_update_sb_errno(journal); |
4327ba52 DJ |
2105 | write_lock(&journal->j_state_lock); |
2106 | journal->j_flags |= JBD2_REC_ERR; | |
2107 | write_unlock(&journal->j_state_lock); | |
2108 | } | |
470decc6 DK |
2109 | } |
2110 | ||
2111 | /** | |
f7f4bccb | 2112 | * void jbd2_journal_abort () - Shutdown the journal immediately. |
470decc6 DK |
2113 | * @journal: the journal to shutdown. |
2114 | * @errno: an error number to record in the journal indicating | |
2115 | * the reason for the shutdown. | |
2116 | * | |
2117 | * Perform a complete, immediate shutdown of the ENTIRE | |
2118 | * journal (not of a single transaction). This operation cannot be | |
2119 | * undone without closing and reopening the journal. | |
2120 | * | |
f7f4bccb | 2121 | * The jbd2_journal_abort function is intended to support higher level error |
470decc6 DK |
2122 | * recovery mechanisms such as the ext2/ext3 remount-readonly error |
2123 | * mode. | |
2124 | * | |
2125 | * Journal abort has very specific semantics. Any existing dirty, | |
2126 | * unjournaled buffers in the main filesystem will still be written to | |
2127 | * disk by bdflush, but the journaling mechanism will be suspended | |
2128 | * immediately and no further transaction commits will be honoured. | |
2129 | * | |
2130 | * Any dirty, journaled buffers will be written back to disk without | |
2131 | * hitting the journal. Atomicity cannot be guaranteed on an aborted | |
2132 | * filesystem, but we _do_ attempt to leave as much data as possible | |
2133 | * behind for fsck to use for cleanup. | |
2134 | * | |
2135 | * Any attempt to get a new transaction handle on a journal which is in | |
2136 | * ABORT state will just result in an -EROFS error return. A | |
f7f4bccb | 2137 | * jbd2_journal_stop on an existing handle will return -EIO if we have |
470decc6 DK |
2138 | * entered abort state during the update. |
2139 | * | |
2140 | * Recursive transactions are not disturbed by journal abort until the | |
f7f4bccb | 2141 | * final jbd2_journal_stop, which will receive the -EIO error. |
470decc6 | 2142 | * |
f7f4bccb | 2143 | * Finally, the jbd2_journal_abort call allows the caller to supply an errno |
470decc6 DK |
2144 | * which will be recorded (if possible) in the journal superblock. This |
2145 | * allows a client to record failure conditions in the middle of a | |
2146 | * transaction without having to complete the transaction to record the | |
2147 | * failure to disk. ext3_error, for example, now uses this | |
2148 | * functionality. | |
2149 | * | |
2150 | * Errors which originate from within the journaling layer will NOT | |
2151 | * supply an errno; a null errno implies that absolutely no further | |
2152 | * writes are done to the journal (unless there are any already in | |
2153 | * progress). | |
2154 | * | |
2155 | */ | |
2156 | ||
f7f4bccb | 2157 | void jbd2_journal_abort(journal_t *journal, int errno) |
470decc6 DK |
2158 | { |
2159 | __journal_abort_soft(journal, errno); | |
2160 | } | |
2161 | ||
2162 | /** | |
f7f4bccb | 2163 | * int jbd2_journal_errno () - returns the journal's error state. |
470decc6 DK |
2164 | * @journal: journal to examine. |
2165 | * | |
bfcd3555 | 2166 | * This is the errno number set with jbd2_journal_abort(), the last |
470decc6 DK |
2167 | * time the journal was mounted - if the journal was stopped |
2168 | * without calling abort this will be 0. | |
2169 | * | |
2170 | * If the journal has been aborted on this mount time -EROFS will | |
2171 | * be returned. | |
2172 | */ | |
f7f4bccb | 2173 | int jbd2_journal_errno(journal_t *journal) |
470decc6 DK |
2174 | { |
2175 | int err; | |
2176 | ||
a931da6a | 2177 | read_lock(&journal->j_state_lock); |
f7f4bccb | 2178 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2179 | err = -EROFS; |
2180 | else | |
2181 | err = journal->j_errno; | |
a931da6a | 2182 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
2183 | return err; |
2184 | } | |
2185 | ||
2186 | /** | |
f7f4bccb | 2187 | * int jbd2_journal_clear_err () - clears the journal's error state |
470decc6 DK |
2188 | * @journal: journal to act on. |
2189 | * | |
bfcd3555 | 2190 | * An error must be cleared or acked to take a FS out of readonly |
470decc6 DK |
2191 | * mode. |
2192 | */ | |
f7f4bccb | 2193 | int jbd2_journal_clear_err(journal_t *journal) |
470decc6 DK |
2194 | { |
2195 | int err = 0; | |
2196 | ||
a931da6a | 2197 | write_lock(&journal->j_state_lock); |
f7f4bccb | 2198 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
2199 | err = -EROFS; |
2200 | else | |
2201 | journal->j_errno = 0; | |
a931da6a | 2202 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2203 | return err; |
2204 | } | |
2205 | ||
2206 | /** | |
f7f4bccb | 2207 | * void jbd2_journal_ack_err() - Ack journal err. |
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 | void jbd2_journal_ack_err(journal_t *journal) |
470decc6 | 2214 | { |
a931da6a | 2215 | write_lock(&journal->j_state_lock); |
470decc6 | 2216 | if (journal->j_errno) |
f7f4bccb | 2217 | journal->j_flags |= JBD2_ACK_ERR; |
a931da6a | 2218 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2219 | } |
2220 | ||
f7f4bccb | 2221 | int jbd2_journal_blocks_per_page(struct inode *inode) |
470decc6 | 2222 | { |
09cbfeaf | 2223 | return 1 << (PAGE_SHIFT - inode->i_sb->s_blocksize_bits); |
470decc6 DK |
2224 | } |
2225 | ||
b517bea1 ZB |
2226 | /* |
2227 | * helper functions to deal with 32 or 64bit block numbers. | |
2228 | */ | |
2229 | size_t journal_tag_bytes(journal_t *journal) | |
2230 | { | |
db9ee220 DW |
2231 | size_t sz; |
2232 | ||
56316a0d | 2233 | if (jbd2_has_feature_csum3(journal)) |
db9ee220 DW |
2234 | return sizeof(journal_block_tag3_t); |
2235 | ||
2236 | sz = sizeof(journal_block_tag_t); | |
c3900875 | 2237 | |
56316a0d | 2238 | if (jbd2_has_feature_csum2(journal)) |
db9ee220 | 2239 | sz += sizeof(__u16); |
c3900875 | 2240 | |
56316a0d | 2241 | if (jbd2_has_feature_64bit(journal)) |
db9ee220 | 2242 | return sz; |
b517bea1 | 2243 | else |
db9ee220 | 2244 | return sz - sizeof(__u32); |
b517bea1 ZB |
2245 | } |
2246 | ||
d2eecb03 TT |
2247 | /* |
2248 | * JBD memory management | |
2249 | * | |
2250 | * These functions are used to allocate block-sized chunks of memory | |
2251 | * used for making copies of buffer_head data. Very often it will be | |
2252 | * page-sized chunks of data, but sometimes it will be in | |
2253 | * sub-page-size chunks. (For example, 16k pages on Power systems | |
2254 | * with a 4k block file system.) For blocks smaller than a page, we | |
2255 | * use a SLAB allocator. There are slab caches for each block size, | |
2256 | * which are allocated at mount time, if necessary, and we only free | |
2257 | * (all of) the slab caches when/if the jbd2 module is unloaded. For | |
2258 | * this reason we don't need to a mutex to protect access to | |
2259 | * jbd2_slab[] allocating or releasing memory; only in | |
2260 | * jbd2_journal_create_slab(). | |
2261 | */ | |
2262 | #define JBD2_MAX_SLABS 8 | |
2263 | static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS]; | |
d2eecb03 TT |
2264 | |
2265 | static const char *jbd2_slab_names[JBD2_MAX_SLABS] = { | |
2266 | "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k", | |
2267 | "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k" | |
2268 | }; | |
2269 | ||
2270 | ||
2271 | static void jbd2_journal_destroy_slabs(void) | |
2272 | { | |
2273 | int i; | |
2274 | ||
2275 | for (i = 0; i < JBD2_MAX_SLABS; i++) { | |
2276 | if (jbd2_slab[i]) | |
2277 | kmem_cache_destroy(jbd2_slab[i]); | |
2278 | jbd2_slab[i] = NULL; | |
2279 | } | |
2280 | } | |
2281 | ||
2282 | static int jbd2_journal_create_slab(size_t size) | |
2283 | { | |
51dfacde | 2284 | static DEFINE_MUTEX(jbd2_slab_create_mutex); |
d2eecb03 TT |
2285 | int i = order_base_2(size) - 10; |
2286 | size_t slab_size; | |
2287 | ||
2288 | if (size == PAGE_SIZE) | |
2289 | return 0; | |
2290 | ||
2291 | if (i >= JBD2_MAX_SLABS) | |
2292 | return -EINVAL; | |
2293 | ||
2294 | if (unlikely(i < 0)) | |
2295 | i = 0; | |
51dfacde | 2296 | mutex_lock(&jbd2_slab_create_mutex); |
d2eecb03 | 2297 | if (jbd2_slab[i]) { |
51dfacde | 2298 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2299 | return 0; /* Already created */ |
2300 | } | |
2301 | ||
2302 | slab_size = 1 << (i+10); | |
2303 | jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size, | |
2304 | slab_size, 0, NULL); | |
51dfacde | 2305 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2306 | if (!jbd2_slab[i]) { |
2307 | printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n"); | |
2308 | return -ENOMEM; | |
2309 | } | |
2310 | return 0; | |
2311 | } | |
2312 | ||
2313 | static struct kmem_cache *get_slab(size_t size) | |
2314 | { | |
2315 | int i = order_base_2(size) - 10; | |
2316 | ||
2317 | BUG_ON(i >= JBD2_MAX_SLABS); | |
2318 | if (unlikely(i < 0)) | |
2319 | i = 0; | |
8ac97b74 | 2320 | BUG_ON(jbd2_slab[i] == NULL); |
d2eecb03 TT |
2321 | return jbd2_slab[i]; |
2322 | } | |
2323 | ||
2324 | void *jbd2_alloc(size_t size, gfp_t flags) | |
2325 | { | |
2326 | void *ptr; | |
2327 | ||
2328 | BUG_ON(size & (size-1)); /* Must be a power of 2 */ | |
2329 | ||
f2db1971 | 2330 | if (size < PAGE_SIZE) |
d2eecb03 | 2331 | ptr = kmem_cache_alloc(get_slab(size), flags); |
f2db1971 MH |
2332 | else |
2333 | ptr = (void *)__get_free_pages(flags, get_order(size)); | |
d2eecb03 TT |
2334 | |
2335 | /* Check alignment; SLUB has gotten this wrong in the past, | |
2336 | * and this can lead to user data corruption! */ | |
2337 | BUG_ON(((unsigned long) ptr) & (size-1)); | |
2338 | ||
2339 | return ptr; | |
2340 | } | |
2341 | ||
2342 | void jbd2_free(void *ptr, size_t size) | |
2343 | { | |
f2db1971 MH |
2344 | if (size < PAGE_SIZE) |
2345 | kmem_cache_free(get_slab(size), ptr); | |
2346 | else | |
2347 | free_pages((unsigned long)ptr, get_order(size)); | |
d2eecb03 TT |
2348 | }; |
2349 | ||
470decc6 DK |
2350 | /* |
2351 | * Journal_head storage management | |
2352 | */ | |
e18b890b | 2353 | static struct kmem_cache *jbd2_journal_head_cache; |
e23291b9 | 2354 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2355 | static atomic_t nr_journal_heads = ATOMIC_INIT(0); |
2356 | #endif | |
2357 | ||
4185a2ac | 2358 | static int jbd2_journal_init_journal_head_cache(void) |
470decc6 DK |
2359 | { |
2360 | int retval; | |
2361 | ||
1076d17a | 2362 | J_ASSERT(jbd2_journal_head_cache == NULL); |
a920e941 | 2363 | jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", |
470decc6 DK |
2364 | sizeof(struct journal_head), |
2365 | 0, /* offset */ | |
5f0d5a3a | 2366 | SLAB_TEMPORARY | SLAB_TYPESAFE_BY_RCU, |
20c2df83 | 2367 | NULL); /* ctor */ |
470decc6 | 2368 | retval = 0; |
1076d17a | 2369 | if (!jbd2_journal_head_cache) { |
470decc6 | 2370 | retval = -ENOMEM; |
f2a44523 | 2371 | printk(KERN_EMERG "JBD2: no memory for journal_head cache\n"); |
470decc6 DK |
2372 | } |
2373 | return retval; | |
2374 | } | |
2375 | ||
4185a2ac | 2376 | static void jbd2_journal_destroy_journal_head_cache(void) |
470decc6 | 2377 | { |
8a9362eb DG |
2378 | if (jbd2_journal_head_cache) { |
2379 | kmem_cache_destroy(jbd2_journal_head_cache); | |
2380 | jbd2_journal_head_cache = NULL; | |
2381 | } | |
470decc6 DK |
2382 | } |
2383 | ||
2384 | /* | |
2385 | * journal_head splicing and dicing | |
2386 | */ | |
2387 | static struct journal_head *journal_alloc_journal_head(void) | |
2388 | { | |
2389 | struct journal_head *ret; | |
470decc6 | 2390 | |
e23291b9 | 2391 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2392 | atomic_inc(&nr_journal_heads); |
2393 | #endif | |
5d9cf9c6 | 2394 | ret = kmem_cache_zalloc(jbd2_journal_head_cache, GFP_NOFS); |
1076d17a | 2395 | if (!ret) { |
470decc6 | 2396 | jbd_debug(1, "out of memory for journal_head\n"); |
670be5a7 | 2397 | pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__); |
7b506b10 MH |
2398 | ret = kmem_cache_zalloc(jbd2_journal_head_cache, |
2399 | GFP_NOFS | __GFP_NOFAIL); | |
470decc6 DK |
2400 | } |
2401 | return ret; | |
2402 | } | |
2403 | ||
2404 | static void journal_free_journal_head(struct journal_head *jh) | |
2405 | { | |
e23291b9 | 2406 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 | 2407 | atomic_dec(&nr_journal_heads); |
cd02ff0b | 2408 | memset(jh, JBD2_POISON_FREE, sizeof(*jh)); |
470decc6 | 2409 | #endif |
f7f4bccb | 2410 | kmem_cache_free(jbd2_journal_head_cache, jh); |
470decc6 DK |
2411 | } |
2412 | ||
2413 | /* | |
2414 | * A journal_head is attached to a buffer_head whenever JBD has an | |
2415 | * interest in the buffer. | |
2416 | * | |
2417 | * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit | |
2418 | * is set. This bit is tested in core kernel code where we need to take | |
2419 | * JBD-specific actions. Testing the zeroness of ->b_private is not reliable | |
2420 | * there. | |
2421 | * | |
2422 | * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. | |
2423 | * | |
2424 | * When a buffer has its BH_JBD bit set it is immune from being released by | |
2425 | * core kernel code, mainly via ->b_count. | |
2426 | * | |
de1b7941 JK |
2427 | * A journal_head is detached from its buffer_head when the journal_head's |
2428 | * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint | |
2429 | * transaction (b_cp_transaction) hold their references to b_jcount. | |
470decc6 DK |
2430 | * |
2431 | * Various places in the kernel want to attach a journal_head to a buffer_head | |
2432 | * _before_ attaching the journal_head to a transaction. To protect the | |
f7f4bccb | 2433 | * journal_head in this situation, jbd2_journal_add_journal_head elevates the |
470decc6 | 2434 | * journal_head's b_jcount refcount by one. The caller must call |
f7f4bccb | 2435 | * jbd2_journal_put_journal_head() to undo this. |
470decc6 DK |
2436 | * |
2437 | * So the typical usage would be: | |
2438 | * | |
2439 | * (Attach a journal_head if needed. Increments b_jcount) | |
f7f4bccb | 2440 | * struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 | 2441 | * ... |
de1b7941 JK |
2442 | * (Get another reference for transaction) |
2443 | * jbd2_journal_grab_journal_head(bh); | |
470decc6 | 2444 | * jh->b_transaction = xxx; |
de1b7941 | 2445 | * (Put original reference) |
f7f4bccb | 2446 | * jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2447 | */ |
2448 | ||
2449 | /* | |
2450 | * Give a buffer_head a journal_head. | |
2451 | * | |
470decc6 DK |
2452 | * May sleep. |
2453 | */ | |
f7f4bccb | 2454 | struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) |
470decc6 DK |
2455 | { |
2456 | struct journal_head *jh; | |
2457 | struct journal_head *new_jh = NULL; | |
2458 | ||
2459 | repeat: | |
5d9cf9c6 | 2460 | if (!buffer_jbd(bh)) |
470decc6 | 2461 | new_jh = journal_alloc_journal_head(); |
470decc6 DK |
2462 | |
2463 | jbd_lock_bh_journal_head(bh); | |
2464 | if (buffer_jbd(bh)) { | |
2465 | jh = bh2jh(bh); | |
2466 | } else { | |
2467 | J_ASSERT_BH(bh, | |
2468 | (atomic_read(&bh->b_count) > 0) || | |
2469 | (bh->b_page && bh->b_page->mapping)); | |
2470 | ||
2471 | if (!new_jh) { | |
2472 | jbd_unlock_bh_journal_head(bh); | |
2473 | goto repeat; | |
2474 | } | |
2475 | ||
2476 | jh = new_jh; | |
2477 | new_jh = NULL; /* We consumed it */ | |
2478 | set_buffer_jbd(bh); | |
2479 | bh->b_private = jh; | |
2480 | jh->b_bh = bh; | |
2481 | get_bh(bh); | |
2482 | BUFFER_TRACE(bh, "added journal_head"); | |
2483 | } | |
2484 | jh->b_jcount++; | |
2485 | jbd_unlock_bh_journal_head(bh); | |
2486 | if (new_jh) | |
2487 | journal_free_journal_head(new_jh); | |
2488 | return bh->b_private; | |
2489 | } | |
2490 | ||
2491 | /* | |
2492 | * Grab a ref against this buffer_head's journal_head. If it ended up not | |
2493 | * having a journal_head, return NULL | |
2494 | */ | |
f7f4bccb | 2495 | struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) |
470decc6 DK |
2496 | { |
2497 | struct journal_head *jh = NULL; | |
2498 | ||
2499 | jbd_lock_bh_journal_head(bh); | |
2500 | if (buffer_jbd(bh)) { | |
2501 | jh = bh2jh(bh); | |
2502 | jh->b_jcount++; | |
2503 | } | |
2504 | jbd_unlock_bh_journal_head(bh); | |
2505 | return jh; | |
2506 | } | |
2507 | ||
2508 | static void __journal_remove_journal_head(struct buffer_head *bh) | |
2509 | { | |
2510 | struct journal_head *jh = bh2jh(bh); | |
2511 | ||
2512 | J_ASSERT_JH(jh, jh->b_jcount >= 0); | |
de1b7941 JK |
2513 | J_ASSERT_JH(jh, jh->b_transaction == NULL); |
2514 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
2515 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
2516 | J_ASSERT_JH(jh, jh->b_jlist == BJ_None); | |
2517 | J_ASSERT_BH(bh, buffer_jbd(bh)); | |
2518 | J_ASSERT_BH(bh, jh2bh(jh) == bh); | |
2519 | BUFFER_TRACE(bh, "remove journal_head"); | |
2520 | if (jh->b_frozen_data) { | |
2521 | printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__); | |
2522 | jbd2_free(jh->b_frozen_data, bh->b_size); | |
470decc6 | 2523 | } |
de1b7941 JK |
2524 | if (jh->b_committed_data) { |
2525 | printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__); | |
2526 | jbd2_free(jh->b_committed_data, bh->b_size); | |
2527 | } | |
2528 | bh->b_private = NULL; | |
2529 | jh->b_bh = NULL; /* debug, really */ | |
2530 | clear_buffer_jbd(bh); | |
2531 | journal_free_journal_head(jh); | |
470decc6 DK |
2532 | } |
2533 | ||
2534 | /* | |
de1b7941 | 2535 | * Drop a reference on the passed journal_head. If it fell to zero then |
470decc6 DK |
2536 | * release the journal_head from the buffer_head. |
2537 | */ | |
f7f4bccb | 2538 | void jbd2_journal_put_journal_head(struct journal_head *jh) |
470decc6 DK |
2539 | { |
2540 | struct buffer_head *bh = jh2bh(jh); | |
2541 | ||
2542 | jbd_lock_bh_journal_head(bh); | |
2543 | J_ASSERT_JH(jh, jh->b_jcount > 0); | |
2544 | --jh->b_jcount; | |
de1b7941 | 2545 | if (!jh->b_jcount) { |
470decc6 | 2546 | __journal_remove_journal_head(bh); |
de1b7941 | 2547 | jbd_unlock_bh_journal_head(bh); |
470decc6 | 2548 | __brelse(bh); |
de1b7941 JK |
2549 | } else |
2550 | jbd_unlock_bh_journal_head(bh); | |
470decc6 DK |
2551 | } |
2552 | ||
c851ed54 JK |
2553 | /* |
2554 | * Initialize jbd inode head | |
2555 | */ | |
2556 | void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) | |
2557 | { | |
2558 | jinode->i_transaction = NULL; | |
2559 | jinode->i_next_transaction = NULL; | |
2560 | jinode->i_vfs_inode = inode; | |
2561 | jinode->i_flags = 0; | |
2562 | INIT_LIST_HEAD(&jinode->i_list); | |
2563 | } | |
2564 | ||
2565 | /* | |
2566 | * Function to be called before we start removing inode from memory (i.e., | |
2567 | * clear_inode() is a fine place to be called from). It removes inode from | |
2568 | * transaction's lists. | |
2569 | */ | |
2570 | void jbd2_journal_release_jbd_inode(journal_t *journal, | |
2571 | struct jbd2_inode *jinode) | |
2572 | { | |
c851ed54 JK |
2573 | if (!journal) |
2574 | return; | |
2575 | restart: | |
2576 | spin_lock(&journal->j_list_lock); | |
2577 | /* Is commit writing out inode - we have to wait */ | |
cb0d9d47 | 2578 | if (jinode->i_flags & JI_COMMIT_RUNNING) { |
c851ed54 JK |
2579 | wait_queue_head_t *wq; |
2580 | DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); | |
2581 | wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); | |
21417136 | 2582 | prepare_to_wait(wq, &wait.wq_entry, TASK_UNINTERRUPTIBLE); |
c851ed54 JK |
2583 | spin_unlock(&journal->j_list_lock); |
2584 | schedule(); | |
21417136 | 2585 | finish_wait(wq, &wait.wq_entry); |
c851ed54 JK |
2586 | goto restart; |
2587 | } | |
2588 | ||
c851ed54 JK |
2589 | if (jinode->i_transaction) { |
2590 | list_del(&jinode->i_list); | |
2591 | jinode->i_transaction = NULL; | |
2592 | } | |
2593 | spin_unlock(&journal->j_list_lock); | |
2594 | } | |
2595 | ||
470decc6 | 2596 | |
8e85fb3f JL |
2597 | #ifdef CONFIG_PROC_FS |
2598 | ||
2599 | #define JBD2_STATS_PROC_NAME "fs/jbd2" | |
2600 | ||
2601 | static void __init jbd2_create_jbd_stats_proc_entry(void) | |
2602 | { | |
2603 | proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); | |
2604 | } | |
2605 | ||
2606 | static void __exit jbd2_remove_jbd_stats_proc_entry(void) | |
2607 | { | |
2608 | if (proc_jbd2_stats) | |
2609 | remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); | |
2610 | } | |
2611 | ||
2612 | #else | |
2613 | ||
2614 | #define jbd2_create_jbd_stats_proc_entry() do {} while (0) | |
2615 | #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) | |
2616 | ||
2617 | #endif | |
2618 | ||
8aefcd55 | 2619 | struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache; |
470decc6 | 2620 | |
4185a2ac | 2621 | static int __init jbd2_journal_init_handle_cache(void) |
470decc6 | 2622 | { |
8aefcd55 | 2623 | jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY); |
f7f4bccb | 2624 | if (jbd2_handle_cache == NULL) { |
8aefcd55 TT |
2625 | printk(KERN_EMERG "JBD2: failed to create handle cache\n"); |
2626 | return -ENOMEM; | |
2627 | } | |
2628 | jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0); | |
2629 | if (jbd2_inode_cache == NULL) { | |
2630 | printk(KERN_EMERG "JBD2: failed to create inode cache\n"); | |
2631 | kmem_cache_destroy(jbd2_handle_cache); | |
470decc6 DK |
2632 | return -ENOMEM; |
2633 | } | |
2634 | return 0; | |
2635 | } | |
2636 | ||
f7f4bccb | 2637 | static void jbd2_journal_destroy_handle_cache(void) |
470decc6 | 2638 | { |
f7f4bccb MC |
2639 | if (jbd2_handle_cache) |
2640 | kmem_cache_destroy(jbd2_handle_cache); | |
8aefcd55 TT |
2641 | if (jbd2_inode_cache) |
2642 | kmem_cache_destroy(jbd2_inode_cache); | |
2643 | ||
470decc6 DK |
2644 | } |
2645 | ||
2646 | /* | |
2647 | * Module startup and shutdown | |
2648 | */ | |
2649 | ||
2650 | static int __init journal_init_caches(void) | |
2651 | { | |
2652 | int ret; | |
2653 | ||
f7f4bccb | 2654 | ret = jbd2_journal_init_revoke_caches(); |
470decc6 | 2655 | if (ret == 0) |
4185a2ac | 2656 | ret = jbd2_journal_init_journal_head_cache(); |
470decc6 | 2657 | if (ret == 0) |
4185a2ac | 2658 | ret = jbd2_journal_init_handle_cache(); |
470decc6 | 2659 | if (ret == 0) |
0c2022ec | 2660 | ret = jbd2_journal_init_transaction_cache(); |
470decc6 DK |
2661 | return ret; |
2662 | } | |
2663 | ||
f7f4bccb | 2664 | static void jbd2_journal_destroy_caches(void) |
470decc6 | 2665 | { |
f7f4bccb | 2666 | jbd2_journal_destroy_revoke_caches(); |
4185a2ac | 2667 | jbd2_journal_destroy_journal_head_cache(); |
f7f4bccb | 2668 | jbd2_journal_destroy_handle_cache(); |
0c2022ec | 2669 | jbd2_journal_destroy_transaction_cache(); |
d2eecb03 | 2670 | jbd2_journal_destroy_slabs(); |
470decc6 DK |
2671 | } |
2672 | ||
2673 | static int __init journal_init(void) | |
2674 | { | |
2675 | int ret; | |
2676 | ||
2677 | BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); | |
2678 | ||
2679 | ret = journal_init_caches(); | |
620de4e1 | 2680 | if (ret == 0) { |
620de4e1 DG |
2681 | jbd2_create_jbd_stats_proc_entry(); |
2682 | } else { | |
f7f4bccb | 2683 | jbd2_journal_destroy_caches(); |
620de4e1 | 2684 | } |
470decc6 DK |
2685 | return ret; |
2686 | } | |
2687 | ||
2688 | static void __exit journal_exit(void) | |
2689 | { | |
e23291b9 | 2690 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2691 | int n = atomic_read(&nr_journal_heads); |
2692 | if (n) | |
75685071 | 2693 | printk(KERN_ERR "JBD2: leaked %d journal_heads!\n", n); |
470decc6 | 2694 | #endif |
8e85fb3f | 2695 | jbd2_remove_jbd_stats_proc_entry(); |
f7f4bccb | 2696 | jbd2_journal_destroy_caches(); |
470decc6 DK |
2697 | } |
2698 | ||
2699 | MODULE_LICENSE("GPL"); | |
2700 | module_init(journal_init); | |
2701 | module_exit(journal_exit); | |
2702 |