Commit | Line | Data |
---|---|---|
470decc6 | 1 | /* |
f7f4bccb | 2 | * linux/fs/jbd2/journal.c |
470decc6 DK |
3 | * |
4 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1998 | |
5 | * | |
6 | * Copyright 1998 Red Hat corp --- All Rights Reserved | |
7 | * | |
8 | * This file is part of the Linux kernel and is made available under | |
9 | * the terms of the GNU General Public License, version 2, or at your | |
10 | * option, any later version, incorporated herein by reference. | |
11 | * | |
12 | * Generic filesystem journal-writing code; part of the ext2fs | |
13 | * journaling system. | |
14 | * | |
15 | * This file manages journals: areas of disk reserved for logging | |
16 | * transactional updates. This includes the kernel journaling thread | |
17 | * which is responsible for scheduling updates to the log. | |
18 | * | |
19 | * We do not actually manage the physical storage of the journal in this | |
20 | * file: that is left to a per-journal policy function, which allows us | |
21 | * to store the journal within a filesystem-specified area for ext2 | |
22 | * journaling (ext2 can use a reserved inode for storing the log). | |
23 | */ | |
24 | ||
25 | #include <linux/module.h> | |
26 | #include <linux/time.h> | |
27 | #include <linux/fs.h> | |
f7f4bccb | 28 | #include <linux/jbd2.h> |
470decc6 DK |
29 | #include <linux/errno.h> |
30 | #include <linux/slab.h> | |
470decc6 DK |
31 | #include <linux/init.h> |
32 | #include <linux/mm.h> | |
7dfb7103 | 33 | #include <linux/freezer.h> |
470decc6 DK |
34 | #include <linux/pagemap.h> |
35 | #include <linux/kthread.h> | |
36 | #include <linux/poison.h> | |
37 | #include <linux/proc_fs.h> | |
0f49d5d0 | 38 | #include <linux/debugfs.h> |
8e85fb3f | 39 | #include <linux/seq_file.h> |
c225aa57 | 40 | #include <linux/math64.h> |
879c5e6b | 41 | #include <linux/hash.h> |
d2eecb03 TT |
42 | #include <linux/log2.h> |
43 | #include <linux/vmalloc.h> | |
47def826 | 44 | #include <linux/backing-dev.h> |
39e3ac25 | 45 | #include <linux/bitops.h> |
670be5a7 | 46 | #include <linux/ratelimit.h> |
879c5e6b TT |
47 | |
48 | #define CREATE_TRACE_POINTS | |
49 | #include <trace/events/jbd2.h> | |
470decc6 DK |
50 | |
51 | #include <asm/uaccess.h> | |
52 | #include <asm/page.h> | |
53 | ||
f7f4bccb MC |
54 | EXPORT_SYMBOL(jbd2_journal_extend); |
55 | EXPORT_SYMBOL(jbd2_journal_stop); | |
56 | EXPORT_SYMBOL(jbd2_journal_lock_updates); | |
57 | EXPORT_SYMBOL(jbd2_journal_unlock_updates); | |
58 | EXPORT_SYMBOL(jbd2_journal_get_write_access); | |
59 | EXPORT_SYMBOL(jbd2_journal_get_create_access); | |
60 | EXPORT_SYMBOL(jbd2_journal_get_undo_access); | |
e06c8227 | 61 | EXPORT_SYMBOL(jbd2_journal_set_triggers); |
f7f4bccb MC |
62 | EXPORT_SYMBOL(jbd2_journal_dirty_metadata); |
63 | EXPORT_SYMBOL(jbd2_journal_release_buffer); | |
64 | EXPORT_SYMBOL(jbd2_journal_forget); | |
470decc6 DK |
65 | #if 0 |
66 | EXPORT_SYMBOL(journal_sync_buffer); | |
67 | #endif | |
f7f4bccb MC |
68 | EXPORT_SYMBOL(jbd2_journal_flush); |
69 | EXPORT_SYMBOL(jbd2_journal_revoke); | |
70 | ||
71 | EXPORT_SYMBOL(jbd2_journal_init_dev); | |
72 | EXPORT_SYMBOL(jbd2_journal_init_inode); | |
f7f4bccb MC |
73 | EXPORT_SYMBOL(jbd2_journal_check_used_features); |
74 | EXPORT_SYMBOL(jbd2_journal_check_available_features); | |
75 | EXPORT_SYMBOL(jbd2_journal_set_features); | |
f7f4bccb MC |
76 | EXPORT_SYMBOL(jbd2_journal_load); |
77 | EXPORT_SYMBOL(jbd2_journal_destroy); | |
f7f4bccb MC |
78 | EXPORT_SYMBOL(jbd2_journal_abort); |
79 | EXPORT_SYMBOL(jbd2_journal_errno); | |
80 | EXPORT_SYMBOL(jbd2_journal_ack_err); | |
81 | EXPORT_SYMBOL(jbd2_journal_clear_err); | |
82 | EXPORT_SYMBOL(jbd2_log_wait_commit); | |
3b799d15 | 83 | EXPORT_SYMBOL(jbd2_log_start_commit); |
f7f4bccb MC |
84 | EXPORT_SYMBOL(jbd2_journal_start_commit); |
85 | EXPORT_SYMBOL(jbd2_journal_force_commit_nested); | |
86 | EXPORT_SYMBOL(jbd2_journal_wipe); | |
87 | EXPORT_SYMBOL(jbd2_journal_blocks_per_page); | |
88 | EXPORT_SYMBOL(jbd2_journal_invalidatepage); | |
89 | EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); | |
90 | EXPORT_SYMBOL(jbd2_journal_force_commit); | |
c851ed54 JK |
91 | EXPORT_SYMBOL(jbd2_journal_file_inode); |
92 | EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); | |
93 | EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); | |
94 | EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); | |
8aefcd55 | 95 | EXPORT_SYMBOL(jbd2_inode_cache); |
470decc6 | 96 | |
470decc6 | 97 | static void __journal_abort_soft (journal_t *journal, int errno); |
d2eecb03 | 98 | static int jbd2_journal_create_slab(size_t slab_size); |
470decc6 | 99 | |
25ed6e8a DW |
100 | /* Checksumming functions */ |
101 | int jbd2_verify_csum_type(journal_t *j, journal_superblock_t *sb) | |
102 | { | |
103 | if (!JBD2_HAS_INCOMPAT_FEATURE(j, JBD2_FEATURE_INCOMPAT_CSUM_V2)) | |
104 | return 1; | |
105 | ||
106 | return sb->s_checksum_type == JBD2_CRC32C_CHKSUM; | |
107 | } | |
108 | ||
470decc6 DK |
109 | /* |
110 | * Helper function used to manage commit timeouts | |
111 | */ | |
112 | ||
113 | static void commit_timeout(unsigned long __data) | |
114 | { | |
115 | struct task_struct * p = (struct task_struct *) __data; | |
116 | ||
117 | wake_up_process(p); | |
118 | } | |
119 | ||
120 | /* | |
f7f4bccb | 121 | * kjournald2: The main thread function used to manage a logging device |
470decc6 DK |
122 | * journal. |
123 | * | |
124 | * This kernel thread is responsible for two things: | |
125 | * | |
126 | * 1) COMMIT: Every so often we need to commit the current state of the | |
127 | * filesystem to disk. The journal thread is responsible for writing | |
128 | * all of the metadata buffers to disk. | |
129 | * | |
130 | * 2) CHECKPOINT: We cannot reuse a used section of the log file until all | |
131 | * of the data in that part of the log has been rewritten elsewhere on | |
132 | * the disk. Flushing these old buffers to reclaim space in the log is | |
133 | * known as checkpointing, and this thread is responsible for that job. | |
134 | */ | |
135 | ||
f7f4bccb | 136 | static int kjournald2(void *arg) |
470decc6 DK |
137 | { |
138 | journal_t *journal = arg; | |
139 | transaction_t *transaction; | |
140 | ||
141 | /* | |
142 | * Set up an interval timer which can be used to trigger a commit wakeup | |
143 | * after the commit interval expires | |
144 | */ | |
145 | setup_timer(&journal->j_commit_timer, commit_timeout, | |
146 | (unsigned long)current); | |
147 | ||
35c80422 NC |
148 | set_freezable(); |
149 | ||
470decc6 DK |
150 | /* Record that the journal thread is running */ |
151 | journal->j_task = current; | |
152 | wake_up(&journal->j_wait_done_commit); | |
153 | ||
470decc6 DK |
154 | /* |
155 | * And now, wait forever for commit wakeup events. | |
156 | */ | |
a931da6a | 157 | write_lock(&journal->j_state_lock); |
470decc6 DK |
158 | |
159 | loop: | |
f7f4bccb | 160 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
161 | goto end_loop; |
162 | ||
163 | jbd_debug(1, "commit_sequence=%d, commit_request=%d\n", | |
164 | journal->j_commit_sequence, journal->j_commit_request); | |
165 | ||
166 | if (journal->j_commit_sequence != journal->j_commit_request) { | |
167 | jbd_debug(1, "OK, requests differ\n"); | |
a931da6a | 168 | write_unlock(&journal->j_state_lock); |
470decc6 | 169 | del_timer_sync(&journal->j_commit_timer); |
f7f4bccb | 170 | jbd2_journal_commit_transaction(journal); |
a931da6a | 171 | write_lock(&journal->j_state_lock); |
470decc6 DK |
172 | goto loop; |
173 | } | |
174 | ||
175 | wake_up(&journal->j_wait_done_commit); | |
176 | if (freezing(current)) { | |
177 | /* | |
178 | * The simpler the better. Flushing journal isn't a | |
179 | * good idea, because that depends on threads that may | |
180 | * be already stopped. | |
181 | */ | |
f7f4bccb | 182 | jbd_debug(1, "Now suspending kjournald2\n"); |
a931da6a | 183 | write_unlock(&journal->j_state_lock); |
a0acae0e | 184 | try_to_freeze(); |
a931da6a | 185 | write_lock(&journal->j_state_lock); |
470decc6 DK |
186 | } else { |
187 | /* | |
188 | * We assume on resume that commits are already there, | |
189 | * so we don't sleep | |
190 | */ | |
191 | DEFINE_WAIT(wait); | |
192 | int should_sleep = 1; | |
193 | ||
194 | prepare_to_wait(&journal->j_wait_commit, &wait, | |
195 | TASK_INTERRUPTIBLE); | |
196 | if (journal->j_commit_sequence != journal->j_commit_request) | |
197 | should_sleep = 0; | |
198 | transaction = journal->j_running_transaction; | |
199 | if (transaction && time_after_eq(jiffies, | |
200 | transaction->t_expires)) | |
201 | should_sleep = 0; | |
f7f4bccb | 202 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
203 | should_sleep = 0; |
204 | if (should_sleep) { | |
a931da6a | 205 | write_unlock(&journal->j_state_lock); |
470decc6 | 206 | schedule(); |
a931da6a | 207 | write_lock(&journal->j_state_lock); |
470decc6 DK |
208 | } |
209 | finish_wait(&journal->j_wait_commit, &wait); | |
210 | } | |
211 | ||
f7f4bccb | 212 | jbd_debug(1, "kjournald2 wakes\n"); |
470decc6 DK |
213 | |
214 | /* | |
215 | * Were we woken up by a commit wakeup event? | |
216 | */ | |
217 | transaction = journal->j_running_transaction; | |
218 | if (transaction && time_after_eq(jiffies, transaction->t_expires)) { | |
219 | journal->j_commit_request = transaction->t_tid; | |
220 | jbd_debug(1, "woke because of timeout\n"); | |
221 | } | |
222 | goto loop; | |
223 | ||
224 | end_loop: | |
a931da6a | 225 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
226 | del_timer_sync(&journal->j_commit_timer); |
227 | journal->j_task = NULL; | |
228 | wake_up(&journal->j_wait_done_commit); | |
229 | jbd_debug(1, "Journal thread exiting.\n"); | |
230 | return 0; | |
231 | } | |
232 | ||
97f06784 | 233 | static int jbd2_journal_start_thread(journal_t *journal) |
470decc6 | 234 | { |
97f06784 PE |
235 | struct task_struct *t; |
236 | ||
90576c0b TT |
237 | t = kthread_run(kjournald2, journal, "jbd2/%s", |
238 | journal->j_devname); | |
97f06784 PE |
239 | if (IS_ERR(t)) |
240 | return PTR_ERR(t); | |
241 | ||
1076d17a | 242 | wait_event(journal->j_wait_done_commit, journal->j_task != NULL); |
97f06784 | 243 | return 0; |
470decc6 DK |
244 | } |
245 | ||
246 | static void journal_kill_thread(journal_t *journal) | |
247 | { | |
a931da6a | 248 | write_lock(&journal->j_state_lock); |
f7f4bccb | 249 | journal->j_flags |= JBD2_UNMOUNT; |
470decc6 DK |
250 | |
251 | while (journal->j_task) { | |
252 | wake_up(&journal->j_wait_commit); | |
a931da6a | 253 | write_unlock(&journal->j_state_lock); |
1076d17a | 254 | wait_event(journal->j_wait_done_commit, journal->j_task == NULL); |
a931da6a | 255 | write_lock(&journal->j_state_lock); |
470decc6 | 256 | } |
a931da6a | 257 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
258 | } |
259 | ||
260 | /* | |
f7f4bccb | 261 | * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. |
470decc6 DK |
262 | * |
263 | * Writes a metadata buffer to a given disk block. The actual IO is not | |
264 | * performed but a new buffer_head is constructed which labels the data | |
265 | * to be written with the correct destination disk block. | |
266 | * | |
267 | * Any magic-number escaping which needs to be done will cause a | |
268 | * copy-out here. If the buffer happens to start with the | |
f7f4bccb | 269 | * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the |
470decc6 DK |
270 | * magic number is only written to the log for descripter blocks. In |
271 | * this case, we copy the data and replace the first word with 0, and we | |
272 | * return a result code which indicates that this buffer needs to be | |
273 | * marked as an escaped buffer in the corresponding log descriptor | |
274 | * block. The missing word can then be restored when the block is read | |
275 | * during recovery. | |
276 | * | |
277 | * If the source buffer has already been modified by a new transaction | |
278 | * since we took the last commit snapshot, we use the frozen copy of | |
279 | * that data for IO. If we end up using the existing buffer_head's data | |
280 | * for the write, then we *have* to lock the buffer to prevent anyone | |
281 | * else from using and possibly modifying it while the IO is in | |
282 | * progress. | |
283 | * | |
284 | * The function returns a pointer to the buffer_heads to be used for IO. | |
285 | * | |
286 | * We assume that the journal has already been locked in this function. | |
287 | * | |
288 | * Return value: | |
289 | * <0: Error | |
290 | * >=0: Finished OK | |
291 | * | |
292 | * On success: | |
293 | * Bit 0 set == escape performed on the data | |
294 | * Bit 1 set == buffer copy-out performed (kfree the data after IO) | |
295 | */ | |
296 | ||
f7f4bccb | 297 | int jbd2_journal_write_metadata_buffer(transaction_t *transaction, |
470decc6 DK |
298 | struct journal_head *jh_in, |
299 | struct journal_head **jh_out, | |
18eba7aa | 300 | unsigned long long blocknr) |
470decc6 DK |
301 | { |
302 | int need_copy_out = 0; | |
303 | int done_copy_out = 0; | |
304 | int do_escape = 0; | |
305 | char *mapped_data; | |
306 | struct buffer_head *new_bh; | |
307 | struct journal_head *new_jh; | |
308 | struct page *new_page; | |
309 | unsigned int new_offset; | |
310 | struct buffer_head *bh_in = jh2bh(jh_in); | |
96577c43 | 311 | journal_t *journal = transaction->t_journal; |
470decc6 DK |
312 | |
313 | /* | |
314 | * The buffer really shouldn't be locked: only the current committing | |
315 | * transaction is allowed to write it, so nobody else is allowed | |
316 | * to do any IO. | |
317 | * | |
318 | * akpm: except if we're journalling data, and write() output is | |
319 | * also part of a shared mapping, and another thread has | |
320 | * decided to launch a writepage() against this buffer. | |
321 | */ | |
322 | J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); | |
323 | ||
47def826 TT |
324 | retry_alloc: |
325 | new_bh = alloc_buffer_head(GFP_NOFS); | |
326 | if (!new_bh) { | |
327 | /* | |
328 | * Failure is not an option, but __GFP_NOFAIL is going | |
329 | * away; so we retry ourselves here. | |
330 | */ | |
331 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
332 | goto retry_alloc; | |
333 | } | |
334 | ||
96577c43 | 335 | /* keep subsequent assertions sane */ |
336 | new_bh->b_state = 0; | |
337 | init_buffer(new_bh, NULL, NULL); | |
338 | atomic_set(&new_bh->b_count, 1); | |
339 | new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */ | |
470decc6 DK |
340 | |
341 | /* | |
342 | * If a new transaction has already done a buffer copy-out, then | |
343 | * we use that version of the data for the commit. | |
344 | */ | |
345 | jbd_lock_bh_state(bh_in); | |
346 | repeat: | |
347 | if (jh_in->b_frozen_data) { | |
348 | done_copy_out = 1; | |
349 | new_page = virt_to_page(jh_in->b_frozen_data); | |
350 | new_offset = offset_in_page(jh_in->b_frozen_data); | |
351 | } else { | |
352 | new_page = jh2bh(jh_in)->b_page; | |
353 | new_offset = offset_in_page(jh2bh(jh_in)->b_data); | |
354 | } | |
355 | ||
303a8f2a | 356 | mapped_data = kmap_atomic(new_page); |
e06c8227 | 357 | /* |
13ceef09 JK |
358 | * Fire data frozen trigger if data already wasn't frozen. Do this |
359 | * before checking for escaping, as the trigger may modify the magic | |
360 | * offset. If a copy-out happens afterwards, it will have the correct | |
361 | * data in the buffer. | |
e06c8227 | 362 | */ |
13ceef09 JK |
363 | if (!done_copy_out) |
364 | jbd2_buffer_frozen_trigger(jh_in, mapped_data + new_offset, | |
365 | jh_in->b_triggers); | |
e06c8227 | 366 | |
470decc6 DK |
367 | /* |
368 | * Check for escaping | |
369 | */ | |
370 | if (*((__be32 *)(mapped_data + new_offset)) == | |
f7f4bccb | 371 | cpu_to_be32(JBD2_MAGIC_NUMBER)) { |
470decc6 DK |
372 | need_copy_out = 1; |
373 | do_escape = 1; | |
374 | } | |
303a8f2a | 375 | kunmap_atomic(mapped_data); |
470decc6 DK |
376 | |
377 | /* | |
378 | * Do we need to do a data copy? | |
379 | */ | |
380 | if (need_copy_out && !done_copy_out) { | |
381 | char *tmp; | |
382 | ||
383 | jbd_unlock_bh_state(bh_in); | |
af1e76d6 | 384 | tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); |
e6ec116b TT |
385 | if (!tmp) { |
386 | jbd2_journal_put_journal_head(new_jh); | |
387 | return -ENOMEM; | |
388 | } | |
470decc6 DK |
389 | jbd_lock_bh_state(bh_in); |
390 | if (jh_in->b_frozen_data) { | |
af1e76d6 | 391 | jbd2_free(tmp, bh_in->b_size); |
470decc6 DK |
392 | goto repeat; |
393 | } | |
394 | ||
395 | jh_in->b_frozen_data = tmp; | |
303a8f2a | 396 | mapped_data = kmap_atomic(new_page); |
470decc6 | 397 | memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size); |
303a8f2a | 398 | kunmap_atomic(mapped_data); |
470decc6 DK |
399 | |
400 | new_page = virt_to_page(tmp); | |
401 | new_offset = offset_in_page(tmp); | |
402 | done_copy_out = 1; | |
e06c8227 JB |
403 | |
404 | /* | |
405 | * This isn't strictly necessary, as we're using frozen | |
406 | * data for the escaping, but it keeps consistency with | |
407 | * b_frozen_data usage. | |
408 | */ | |
409 | jh_in->b_frozen_triggers = jh_in->b_triggers; | |
470decc6 DK |
410 | } |
411 | ||
412 | /* | |
413 | * Did we need to do an escaping? Now we've done all the | |
414 | * copying, we can finally do so. | |
415 | */ | |
416 | if (do_escape) { | |
303a8f2a | 417 | mapped_data = kmap_atomic(new_page); |
470decc6 | 418 | *((unsigned int *)(mapped_data + new_offset)) = 0; |
303a8f2a | 419 | kunmap_atomic(mapped_data); |
470decc6 DK |
420 | } |
421 | ||
470decc6 DK |
422 | set_bh_page(new_bh, new_page, new_offset); |
423 | new_jh->b_transaction = NULL; | |
424 | new_bh->b_size = jh2bh(jh_in)->b_size; | |
425 | new_bh->b_bdev = transaction->t_journal->j_dev; | |
426 | new_bh->b_blocknr = blocknr; | |
427 | set_buffer_mapped(new_bh); | |
428 | set_buffer_dirty(new_bh); | |
429 | ||
430 | *jh_out = new_jh; | |
431 | ||
432 | /* | |
433 | * The to-be-written buffer needs to get moved to the io queue, | |
434 | * and the original buffer whose contents we are shadowing or | |
435 | * copying is moved to the transaction's shadow queue. | |
436 | */ | |
437 | JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); | |
96577c43 | 438 | spin_lock(&journal->j_list_lock); |
439 | __jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); | |
440 | spin_unlock(&journal->j_list_lock); | |
441 | jbd_unlock_bh_state(bh_in); | |
442 | ||
470decc6 | 443 | JBUFFER_TRACE(new_jh, "file as BJ_IO"); |
f7f4bccb | 444 | jbd2_journal_file_buffer(new_jh, transaction, BJ_IO); |
470decc6 DK |
445 | |
446 | return do_escape | (done_copy_out << 1); | |
447 | } | |
448 | ||
449 | /* | |
450 | * Allocation code for the journal file. Manage the space left in the | |
451 | * journal, so that we can begin checkpointing when appropriate. | |
452 | */ | |
453 | ||
454 | /* | |
f7f4bccb | 455 | * __jbd2_log_space_left: Return the number of free blocks left in the journal. |
470decc6 DK |
456 | * |
457 | * Called with the journal already locked. | |
458 | * | |
459 | * Called under j_state_lock | |
460 | */ | |
461 | ||
f7f4bccb | 462 | int __jbd2_log_space_left(journal_t *journal) |
470decc6 DK |
463 | { |
464 | int left = journal->j_free; | |
465 | ||
a931da6a | 466 | /* assert_spin_locked(&journal->j_state_lock); */ |
470decc6 DK |
467 | |
468 | /* | |
469 | * Be pessimistic here about the number of those free blocks which | |
470 | * might be required for log descriptor control blocks. | |
471 | */ | |
472 | ||
473 | #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */ | |
474 | ||
475 | left -= MIN_LOG_RESERVED_BLOCKS; | |
476 | ||
477 | if (left <= 0) | |
478 | return 0; | |
479 | left -= (left >> 3); | |
480 | return left; | |
481 | } | |
482 | ||
483 | /* | |
e4471831 TT |
484 | * Called with j_state_lock locked for writing. |
485 | * Returns true if a transaction commit was started. | |
470decc6 | 486 | */ |
f7f4bccb | 487 | int __jbd2_log_start_commit(journal_t *journal, tid_t target) |
470decc6 DK |
488 | { |
489 | /* | |
deeeaf13 TT |
490 | * The only transaction we can possibly wait upon is the |
491 | * currently running transaction (if it exists). Otherwise, | |
492 | * the target tid must be an old one. | |
470decc6 | 493 | */ |
deeeaf13 TT |
494 | if (journal->j_running_transaction && |
495 | journal->j_running_transaction->t_tid == target) { | |
470decc6 | 496 | /* |
bcf3d0bc | 497 | * We want a new commit: OK, mark the request and wakeup the |
470decc6 DK |
498 | * commit thread. We do _not_ do the commit ourselves. |
499 | */ | |
500 | ||
501 | journal->j_commit_request = target; | |
f2a44523 | 502 | jbd_debug(1, "JBD2: requesting commit %d/%d\n", |
470decc6 DK |
503 | journal->j_commit_request, |
504 | journal->j_commit_sequence); | |
505 | wake_up(&journal->j_wait_commit); | |
506 | return 1; | |
deeeaf13 TT |
507 | } else if (!tid_geq(journal->j_commit_request, target)) |
508 | /* This should never happen, but if it does, preserve | |
509 | the evidence before kjournald goes into a loop and | |
510 | increments j_commit_sequence beyond all recognition. */ | |
f2a44523 | 511 | WARN_ONCE(1, "JBD2: bad log_start_commit: %u %u %u %u\n", |
1be2add6 TT |
512 | journal->j_commit_request, |
513 | journal->j_commit_sequence, | |
514 | target, journal->j_running_transaction ? | |
515 | journal->j_running_transaction->t_tid : 0); | |
470decc6 DK |
516 | return 0; |
517 | } | |
518 | ||
f7f4bccb | 519 | int jbd2_log_start_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
520 | { |
521 | int ret; | |
522 | ||
a931da6a | 523 | write_lock(&journal->j_state_lock); |
f7f4bccb | 524 | ret = __jbd2_log_start_commit(journal, tid); |
a931da6a | 525 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
526 | return ret; |
527 | } | |
528 | ||
529 | /* | |
530 | * Force and wait upon a commit if the calling process is not within | |
531 | * transaction. This is used for forcing out undo-protected data which contains | |
532 | * bitmaps, when the fs is running out of space. | |
533 | * | |
534 | * We can only force the running transaction if we don't have an active handle; | |
535 | * otherwise, we will deadlock. | |
536 | * | |
537 | * Returns true if a transaction was started. | |
538 | */ | |
f7f4bccb | 539 | int jbd2_journal_force_commit_nested(journal_t *journal) |
470decc6 DK |
540 | { |
541 | transaction_t *transaction = NULL; | |
542 | tid_t tid; | |
e4471831 | 543 | int need_to_start = 0; |
470decc6 | 544 | |
a931da6a | 545 | read_lock(&journal->j_state_lock); |
470decc6 DK |
546 | if (journal->j_running_transaction && !current->journal_info) { |
547 | transaction = journal->j_running_transaction; | |
e4471831 TT |
548 | if (!tid_geq(journal->j_commit_request, transaction->t_tid)) |
549 | need_to_start = 1; | |
470decc6 DK |
550 | } else if (journal->j_committing_transaction) |
551 | transaction = journal->j_committing_transaction; | |
552 | ||
553 | if (!transaction) { | |
a931da6a | 554 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
555 | return 0; /* Nothing to retry */ |
556 | } | |
557 | ||
558 | tid = transaction->t_tid; | |
a931da6a | 559 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
560 | if (need_to_start) |
561 | jbd2_log_start_commit(journal, tid); | |
f7f4bccb | 562 | jbd2_log_wait_commit(journal, tid); |
470decc6 DK |
563 | return 1; |
564 | } | |
565 | ||
566 | /* | |
567 | * Start a commit of the current running transaction (if any). Returns true | |
c88ccea3 JK |
568 | * if a transaction is going to be committed (or is currently already |
569 | * committing), and fills its tid in at *ptid | |
470decc6 | 570 | */ |
f7f4bccb | 571 | int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) |
470decc6 DK |
572 | { |
573 | int ret = 0; | |
574 | ||
a931da6a | 575 | write_lock(&journal->j_state_lock); |
470decc6 DK |
576 | if (journal->j_running_transaction) { |
577 | tid_t tid = journal->j_running_transaction->t_tid; | |
578 | ||
c88ccea3 JK |
579 | __jbd2_log_start_commit(journal, tid); |
580 | /* There's a running transaction and we've just made sure | |
581 | * it's commit has been scheduled. */ | |
582 | if (ptid) | |
470decc6 | 583 | *ptid = tid; |
c88ccea3 JK |
584 | ret = 1; |
585 | } else if (journal->j_committing_transaction) { | |
470decc6 DK |
586 | /* |
587 | * If ext3_write_super() recently started a commit, then we | |
588 | * have to wait for completion of that transaction | |
589 | */ | |
c88ccea3 JK |
590 | if (ptid) |
591 | *ptid = journal->j_committing_transaction->t_tid; | |
470decc6 DK |
592 | ret = 1; |
593 | } | |
a931da6a | 594 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
595 | return ret; |
596 | } | |
597 | ||
bbd2be36 JK |
598 | /* |
599 | * Return 1 if a given transaction has not yet sent barrier request | |
600 | * connected with a transaction commit. If 0 is returned, transaction | |
601 | * may or may not have sent the barrier. Used to avoid sending barrier | |
602 | * twice in common cases. | |
603 | */ | |
604 | int jbd2_trans_will_send_data_barrier(journal_t *journal, tid_t tid) | |
605 | { | |
606 | int ret = 0; | |
607 | transaction_t *commit_trans; | |
608 | ||
609 | if (!(journal->j_flags & JBD2_BARRIER)) | |
610 | return 0; | |
611 | read_lock(&journal->j_state_lock); | |
612 | /* Transaction already committed? */ | |
613 | if (tid_geq(journal->j_commit_sequence, tid)) | |
614 | goto out; | |
615 | commit_trans = journal->j_committing_transaction; | |
616 | if (!commit_trans || commit_trans->t_tid != tid) { | |
617 | ret = 1; | |
618 | goto out; | |
619 | } | |
620 | /* | |
621 | * Transaction is being committed and we already proceeded to | |
622 | * submitting a flush to fs partition? | |
623 | */ | |
624 | if (journal->j_fs_dev != journal->j_dev) { | |
625 | if (!commit_trans->t_need_data_flush || | |
626 | commit_trans->t_state >= T_COMMIT_DFLUSH) | |
627 | goto out; | |
628 | } else { | |
629 | if (commit_trans->t_state >= T_COMMIT_JFLUSH) | |
630 | goto out; | |
631 | } | |
632 | ret = 1; | |
633 | out: | |
634 | read_unlock(&journal->j_state_lock); | |
635 | return ret; | |
636 | } | |
637 | EXPORT_SYMBOL(jbd2_trans_will_send_data_barrier); | |
638 | ||
470decc6 DK |
639 | /* |
640 | * Wait for a specified commit to complete. | |
641 | * The caller may not hold the journal lock. | |
642 | */ | |
f7f4bccb | 643 | int jbd2_log_wait_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
644 | { |
645 | int err = 0; | |
646 | ||
a931da6a | 647 | read_lock(&journal->j_state_lock); |
e23291b9 | 648 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
649 | if (!tid_geq(journal->j_commit_request, tid)) { |
650 | printk(KERN_EMERG | |
651 | "%s: error: j_commit_request=%d, tid=%d\n", | |
329d291f | 652 | __func__, journal->j_commit_request, tid); |
470decc6 | 653 | } |
470decc6 | 654 | #endif |
470decc6 | 655 | while (tid_gt(tid, journal->j_commit_sequence)) { |
f2a44523 | 656 | jbd_debug(1, "JBD2: want %d, j_commit_sequence=%d\n", |
470decc6 DK |
657 | tid, journal->j_commit_sequence); |
658 | wake_up(&journal->j_wait_commit); | |
a931da6a | 659 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
660 | wait_event(journal->j_wait_done_commit, |
661 | !tid_gt(tid, journal->j_commit_sequence)); | |
a931da6a | 662 | read_lock(&journal->j_state_lock); |
470decc6 | 663 | } |
a931da6a | 664 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
665 | |
666 | if (unlikely(is_journal_aborted(journal))) { | |
667 | printk(KERN_EMERG "journal commit I/O error\n"); | |
668 | err = -EIO; | |
669 | } | |
670 | return err; | |
671 | } | |
672 | ||
673 | /* | |
674 | * Log buffer allocation routines: | |
675 | */ | |
676 | ||
18eba7aa | 677 | int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) |
470decc6 DK |
678 | { |
679 | unsigned long blocknr; | |
680 | ||
a931da6a | 681 | write_lock(&journal->j_state_lock); |
470decc6 DK |
682 | J_ASSERT(journal->j_free > 1); |
683 | ||
684 | blocknr = journal->j_head; | |
685 | journal->j_head++; | |
686 | journal->j_free--; | |
687 | if (journal->j_head == journal->j_last) | |
688 | journal->j_head = journal->j_first; | |
a931da6a | 689 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 690 | return jbd2_journal_bmap(journal, blocknr, retp); |
470decc6 DK |
691 | } |
692 | ||
693 | /* | |
694 | * Conversion of logical to physical block numbers for the journal | |
695 | * | |
696 | * On external journals the journal blocks are identity-mapped, so | |
697 | * this is a no-op. If needed, we can use j_blk_offset - everything is | |
698 | * ready. | |
699 | */ | |
f7f4bccb | 700 | int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, |
18eba7aa | 701 | unsigned long long *retp) |
470decc6 DK |
702 | { |
703 | int err = 0; | |
18eba7aa | 704 | unsigned long long ret; |
470decc6 DK |
705 | |
706 | if (journal->j_inode) { | |
707 | ret = bmap(journal->j_inode, blocknr); | |
708 | if (ret) | |
709 | *retp = ret; | |
710 | else { | |
470decc6 DK |
711 | printk(KERN_ALERT "%s: journal block not found " |
712 | "at offset %lu on %s\n", | |
05496769 | 713 | __func__, blocknr, journal->j_devname); |
470decc6 DK |
714 | err = -EIO; |
715 | __journal_abort_soft(journal, err); | |
716 | } | |
717 | } else { | |
718 | *retp = blocknr; /* +journal->j_blk_offset */ | |
719 | } | |
720 | return err; | |
721 | } | |
722 | ||
723 | /* | |
724 | * We play buffer_head aliasing tricks to write data/metadata blocks to | |
725 | * the journal without copying their contents, but for journal | |
726 | * descriptor blocks we do need to generate bona fide buffers. | |
727 | * | |
f7f4bccb | 728 | * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying |
470decc6 DK |
729 | * the buffer's contents they really should run flush_dcache_page(bh->b_page). |
730 | * But we don't bother doing that, so there will be coherency problems with | |
731 | * mmaps of blockdevs which hold live JBD-controlled filesystems. | |
732 | */ | |
f7f4bccb | 733 | struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal) |
470decc6 DK |
734 | { |
735 | struct buffer_head *bh; | |
18eba7aa | 736 | unsigned long long blocknr; |
470decc6 DK |
737 | int err; |
738 | ||
f7f4bccb | 739 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
740 | |
741 | if (err) | |
742 | return NULL; | |
743 | ||
744 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
4b905671 JK |
745 | if (!bh) |
746 | return NULL; | |
470decc6 DK |
747 | lock_buffer(bh); |
748 | memset(bh->b_data, 0, journal->j_blocksize); | |
749 | set_buffer_uptodate(bh); | |
750 | unlock_buffer(bh); | |
751 | BUFFER_TRACE(bh, "return this buffer"); | |
f7f4bccb | 752 | return jbd2_journal_add_journal_head(bh); |
470decc6 DK |
753 | } |
754 | ||
79feb521 JK |
755 | /* |
756 | * Return tid of the oldest transaction in the journal and block in the journal | |
757 | * where the transaction starts. | |
758 | * | |
759 | * If the journal is now empty, return which will be the next transaction ID | |
760 | * we will write and where will that transaction start. | |
761 | * | |
762 | * The return value is 0 if journal tail cannot be pushed any further, 1 if | |
763 | * it can. | |
764 | */ | |
765 | int jbd2_journal_get_log_tail(journal_t *journal, tid_t *tid, | |
766 | unsigned long *block) | |
767 | { | |
768 | transaction_t *transaction; | |
769 | int ret; | |
770 | ||
771 | read_lock(&journal->j_state_lock); | |
772 | spin_lock(&journal->j_list_lock); | |
773 | transaction = journal->j_checkpoint_transactions; | |
774 | if (transaction) { | |
775 | *tid = transaction->t_tid; | |
776 | *block = transaction->t_log_start; | |
777 | } else if ((transaction = journal->j_committing_transaction) != NULL) { | |
778 | *tid = transaction->t_tid; | |
779 | *block = transaction->t_log_start; | |
780 | } else if ((transaction = journal->j_running_transaction) != NULL) { | |
781 | *tid = transaction->t_tid; | |
782 | *block = journal->j_head; | |
783 | } else { | |
784 | *tid = journal->j_transaction_sequence; | |
785 | *block = journal->j_head; | |
786 | } | |
787 | ret = tid_gt(*tid, journal->j_tail_sequence); | |
788 | spin_unlock(&journal->j_list_lock); | |
789 | read_unlock(&journal->j_state_lock); | |
790 | ||
791 | return ret; | |
792 | } | |
793 | ||
794 | /* | |
795 | * Update information in journal structure and in on disk journal superblock | |
796 | * about log tail. This function does not check whether information passed in | |
797 | * really pushes log tail further. It's responsibility of the caller to make | |
798 | * sure provided log tail information is valid (e.g. by holding | |
799 | * j_checkpoint_mutex all the time between computing log tail and calling this | |
800 | * function as is the case with jbd2_cleanup_journal_tail()). | |
801 | * | |
802 | * Requires j_checkpoint_mutex | |
803 | */ | |
804 | void __jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) | |
805 | { | |
806 | unsigned long freed; | |
807 | ||
808 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); | |
809 | ||
810 | /* | |
811 | * We cannot afford for write to remain in drive's caches since as | |
812 | * soon as we update j_tail, next transaction can start reusing journal | |
813 | * space and if we lose sb update during power failure we'd replay | |
814 | * old transaction with possibly newly overwritten data. | |
815 | */ | |
816 | jbd2_journal_update_sb_log_tail(journal, tid, block, WRITE_FUA); | |
817 | write_lock(&journal->j_state_lock); | |
818 | freed = block - journal->j_tail; | |
819 | if (block < journal->j_tail) | |
820 | freed += journal->j_last - journal->j_first; | |
821 | ||
822 | trace_jbd2_update_log_tail(journal, tid, block, freed); | |
823 | jbd_debug(1, | |
824 | "Cleaning journal tail from %d to %d (offset %lu), " | |
825 | "freeing %lu\n", | |
826 | journal->j_tail_sequence, tid, block, freed); | |
827 | ||
828 | journal->j_free += freed; | |
829 | journal->j_tail_sequence = tid; | |
830 | journal->j_tail = block; | |
831 | write_unlock(&journal->j_state_lock); | |
832 | } | |
833 | ||
3339578f JK |
834 | /* |
835 | * This is a variaon of __jbd2_update_log_tail which checks for validity of | |
836 | * provided log tail and locks j_checkpoint_mutex. So it is safe against races | |
837 | * with other threads updating log tail. | |
838 | */ | |
839 | void jbd2_update_log_tail(journal_t *journal, tid_t tid, unsigned long block) | |
840 | { | |
841 | mutex_lock(&journal->j_checkpoint_mutex); | |
842 | if (tid_gt(tid, journal->j_tail_sequence)) | |
843 | __jbd2_update_log_tail(journal, tid, block); | |
844 | mutex_unlock(&journal->j_checkpoint_mutex); | |
845 | } | |
846 | ||
8e85fb3f JL |
847 | struct jbd2_stats_proc_session { |
848 | journal_t *journal; | |
849 | struct transaction_stats_s *stats; | |
850 | int start; | |
851 | int max; | |
852 | }; | |
853 | ||
8e85fb3f JL |
854 | static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) |
855 | { | |
856 | return *pos ? NULL : SEQ_START_TOKEN; | |
857 | } | |
858 | ||
859 | static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) | |
860 | { | |
861 | return NULL; | |
862 | } | |
863 | ||
864 | static int jbd2_seq_info_show(struct seq_file *seq, void *v) | |
865 | { | |
866 | struct jbd2_stats_proc_session *s = seq->private; | |
867 | ||
868 | if (v != SEQ_START_TOKEN) | |
869 | return 0; | |
bf699327 | 870 | seq_printf(seq, "%lu transaction, each up to %u blocks\n", |
8e85fb3f JL |
871 | s->stats->ts_tid, |
872 | s->journal->j_max_transaction_buffers); | |
873 | if (s->stats->ts_tid == 0) | |
874 | return 0; | |
875 | seq_printf(seq, "average: \n %ums waiting for transaction\n", | |
bf699327 | 876 | jiffies_to_msecs(s->stats->run.rs_wait / s->stats->ts_tid)); |
8e85fb3f | 877 | seq_printf(seq, " %ums running transaction\n", |
bf699327 | 878 | jiffies_to_msecs(s->stats->run.rs_running / s->stats->ts_tid)); |
8e85fb3f | 879 | seq_printf(seq, " %ums transaction was being locked\n", |
bf699327 | 880 | jiffies_to_msecs(s->stats->run.rs_locked / s->stats->ts_tid)); |
8e85fb3f | 881 | seq_printf(seq, " %ums flushing data (in ordered mode)\n", |
bf699327 | 882 | jiffies_to_msecs(s->stats->run.rs_flushing / s->stats->ts_tid)); |
8e85fb3f | 883 | seq_printf(seq, " %ums logging transaction\n", |
bf699327 | 884 | jiffies_to_msecs(s->stats->run.rs_logging / s->stats->ts_tid)); |
c225aa57 SHT |
885 | seq_printf(seq, " %lluus average transaction commit time\n", |
886 | div_u64(s->journal->j_average_commit_time, 1000)); | |
8e85fb3f | 887 | seq_printf(seq, " %lu handles per transaction\n", |
bf699327 | 888 | s->stats->run.rs_handle_count / s->stats->ts_tid); |
8e85fb3f | 889 | seq_printf(seq, " %lu blocks per transaction\n", |
bf699327 | 890 | s->stats->run.rs_blocks / s->stats->ts_tid); |
8e85fb3f | 891 | seq_printf(seq, " %lu logged blocks per transaction\n", |
bf699327 | 892 | s->stats->run.rs_blocks_logged / s->stats->ts_tid); |
8e85fb3f JL |
893 | return 0; |
894 | } | |
895 | ||
896 | static void jbd2_seq_info_stop(struct seq_file *seq, void *v) | |
897 | { | |
898 | } | |
899 | ||
88e9d34c | 900 | static const struct seq_operations jbd2_seq_info_ops = { |
8e85fb3f JL |
901 | .start = jbd2_seq_info_start, |
902 | .next = jbd2_seq_info_next, | |
903 | .stop = jbd2_seq_info_stop, | |
904 | .show = jbd2_seq_info_show, | |
905 | }; | |
906 | ||
907 | static int jbd2_seq_info_open(struct inode *inode, struct file *file) | |
908 | { | |
909 | journal_t *journal = PDE(inode)->data; | |
910 | struct jbd2_stats_proc_session *s; | |
911 | int rc, size; | |
912 | ||
913 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
914 | if (s == NULL) | |
915 | return -ENOMEM; | |
916 | size = sizeof(struct transaction_stats_s); | |
917 | s->stats = kmalloc(size, GFP_KERNEL); | |
918 | if (s->stats == NULL) { | |
919 | kfree(s); | |
920 | return -ENOMEM; | |
921 | } | |
922 | spin_lock(&journal->j_history_lock); | |
923 | memcpy(s->stats, &journal->j_stats, size); | |
924 | s->journal = journal; | |
925 | spin_unlock(&journal->j_history_lock); | |
926 | ||
927 | rc = seq_open(file, &jbd2_seq_info_ops); | |
928 | if (rc == 0) { | |
929 | struct seq_file *m = file->private_data; | |
930 | m->private = s; | |
931 | } else { | |
932 | kfree(s->stats); | |
933 | kfree(s); | |
934 | } | |
935 | return rc; | |
936 | ||
937 | } | |
938 | ||
939 | static int jbd2_seq_info_release(struct inode *inode, struct file *file) | |
940 | { | |
941 | struct seq_file *seq = file->private_data; | |
942 | struct jbd2_stats_proc_session *s = seq->private; | |
943 | kfree(s->stats); | |
944 | kfree(s); | |
945 | return seq_release(inode, file); | |
946 | } | |
947 | ||
828c0950 | 948 | static const struct file_operations jbd2_seq_info_fops = { |
8e85fb3f JL |
949 | .owner = THIS_MODULE, |
950 | .open = jbd2_seq_info_open, | |
951 | .read = seq_read, | |
952 | .llseek = seq_lseek, | |
953 | .release = jbd2_seq_info_release, | |
954 | }; | |
955 | ||
956 | static struct proc_dir_entry *proc_jbd2_stats; | |
957 | ||
958 | static void jbd2_stats_proc_init(journal_t *journal) | |
959 | { | |
05496769 | 960 | journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); |
8e85fb3f | 961 | if (journal->j_proc_entry) { |
79da3664 DL |
962 | proc_create_data("info", S_IRUGO, journal->j_proc_entry, |
963 | &jbd2_seq_info_fops, journal); | |
8e85fb3f JL |
964 | } |
965 | } | |
966 | ||
967 | static void jbd2_stats_proc_exit(journal_t *journal) | |
968 | { | |
8e85fb3f | 969 | remove_proc_entry("info", journal->j_proc_entry); |
05496769 | 970 | remove_proc_entry(journal->j_devname, proc_jbd2_stats); |
8e85fb3f JL |
971 | } |
972 | ||
470decc6 DK |
973 | /* |
974 | * Management for journal control blocks: functions to create and | |
975 | * destroy journal_t structures, and to initialise and read existing | |
976 | * journal blocks from disk. */ | |
977 | ||
978 | /* First: create and setup a journal_t object in memory. We initialise | |
979 | * very few fields yet: that has to wait until we have created the | |
980 | * journal structures from from scratch, or loaded them from disk. */ | |
981 | ||
982 | static journal_t * journal_init_common (void) | |
983 | { | |
984 | journal_t *journal; | |
985 | int err; | |
986 | ||
3ebfdf88 | 987 | journal = kzalloc(sizeof(*journal), GFP_KERNEL); |
470decc6 | 988 | if (!journal) |
b7271b0a | 989 | return NULL; |
470decc6 DK |
990 | |
991 | init_waitqueue_head(&journal->j_wait_transaction_locked); | |
992 | init_waitqueue_head(&journal->j_wait_logspace); | |
993 | init_waitqueue_head(&journal->j_wait_done_commit); | |
994 | init_waitqueue_head(&journal->j_wait_checkpoint); | |
995 | init_waitqueue_head(&journal->j_wait_commit); | |
996 | init_waitqueue_head(&journal->j_wait_updates); | |
997 | mutex_init(&journal->j_barrier); | |
998 | mutex_init(&journal->j_checkpoint_mutex); | |
999 | spin_lock_init(&journal->j_revoke_lock); | |
1000 | spin_lock_init(&journal->j_list_lock); | |
a931da6a | 1001 | rwlock_init(&journal->j_state_lock); |
470decc6 | 1002 | |
cd02ff0b | 1003 | journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); |
30773840 TT |
1004 | journal->j_min_batch_time = 0; |
1005 | journal->j_max_batch_time = 15000; /* 15ms */ | |
470decc6 DK |
1006 | |
1007 | /* The journal is marked for error until we succeed with recovery! */ | |
f7f4bccb | 1008 | journal->j_flags = JBD2_ABORT; |
470decc6 DK |
1009 | |
1010 | /* Set up a default-sized revoke table for the new mount. */ | |
f7f4bccb | 1011 | err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); |
470decc6 DK |
1012 | if (err) { |
1013 | kfree(journal); | |
b7271b0a | 1014 | return NULL; |
470decc6 | 1015 | } |
8e85fb3f | 1016 | |
bf699327 | 1017 | spin_lock_init(&journal->j_history_lock); |
8e85fb3f | 1018 | |
470decc6 | 1019 | return journal; |
470decc6 DK |
1020 | } |
1021 | ||
f7f4bccb | 1022 | /* jbd2_journal_init_dev and jbd2_journal_init_inode: |
470decc6 DK |
1023 | * |
1024 | * Create a journal structure assigned some fixed set of disk blocks to | |
1025 | * the journal. We don't actually touch those disk blocks yet, but we | |
1026 | * need to set up all of the mapping information to tell the journaling | |
1027 | * system where the journal blocks are. | |
1028 | * | |
1029 | */ | |
1030 | ||
1031 | /** | |
5648ba5b | 1032 | * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure |
470decc6 DK |
1033 | * @bdev: Block device on which to create the journal |
1034 | * @fs_dev: Device which hold journalled filesystem for this journal. | |
1035 | * @start: Block nr Start of journal. | |
1036 | * @len: Length of the journal in blocks. | |
1037 | * @blocksize: blocksize of journalling device | |
5648ba5b RD |
1038 | * |
1039 | * Returns: a newly created journal_t * | |
470decc6 | 1040 | * |
f7f4bccb | 1041 | * jbd2_journal_init_dev creates a journal which maps a fixed contiguous |
470decc6 DK |
1042 | * range of blocks on an arbitrary block device. |
1043 | * | |
1044 | */ | |
f7f4bccb | 1045 | journal_t * jbd2_journal_init_dev(struct block_device *bdev, |
470decc6 | 1046 | struct block_device *fs_dev, |
18eba7aa | 1047 | unsigned long long start, int len, int blocksize) |
470decc6 DK |
1048 | { |
1049 | journal_t *journal = journal_init_common(); | |
1050 | struct buffer_head *bh; | |
05496769 | 1051 | char *p; |
470decc6 DK |
1052 | int n; |
1053 | ||
1054 | if (!journal) | |
1055 | return NULL; | |
1056 | ||
1057 | /* journal descriptor can store up to n blocks -bzzz */ | |
1058 | journal->j_blocksize = blocksize; | |
0587aa3d | 1059 | journal->j_dev = bdev; |
1060 | journal->j_fs_dev = fs_dev; | |
1061 | journal->j_blk_offset = start; | |
1062 | journal->j_maxlen = len; | |
1063 | bdevname(journal->j_dev, journal->j_devname); | |
1064 | p = journal->j_devname; | |
1065 | while ((p = strchr(p, '/'))) | |
1066 | *p = '!'; | |
4b905671 | 1067 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1068 | n = journal->j_blocksize / sizeof(journal_block_tag_t); |
1069 | journal->j_wbufsize = n; | |
1070 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1071 | if (!journal->j_wbuf) { | |
25985edc | 1072 | printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", |
329d291f | 1073 | __func__); |
4b905671 | 1074 | goto out_err; |
470decc6 | 1075 | } |
470decc6 DK |
1076 | |
1077 | bh = __getblk(journal->j_dev, start, journal->j_blocksize); | |
4b905671 JK |
1078 | if (!bh) { |
1079 | printk(KERN_ERR | |
1080 | "%s: Cannot get buffer for journal superblock\n", | |
1081 | __func__); | |
1082 | goto out_err; | |
1083 | } | |
470decc6 DK |
1084 | journal->j_sb_buffer = bh; |
1085 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
4b905671 | 1086 | |
470decc6 | 1087 | return journal; |
4b905671 | 1088 | out_err: |
7b02bec0 | 1089 | kfree(journal->j_wbuf); |
4b905671 JK |
1090 | jbd2_stats_proc_exit(journal); |
1091 | kfree(journal); | |
1092 | return NULL; | |
470decc6 DK |
1093 | } |
1094 | ||
1095 | /** | |
f7f4bccb | 1096 | * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. |
470decc6 DK |
1097 | * @inode: An inode to create the journal in |
1098 | * | |
f7f4bccb | 1099 | * jbd2_journal_init_inode creates a journal which maps an on-disk inode as |
470decc6 DK |
1100 | * the journal. The inode must exist already, must support bmap() and |
1101 | * must have all data blocks preallocated. | |
1102 | */ | |
f7f4bccb | 1103 | journal_t * jbd2_journal_init_inode (struct inode *inode) |
470decc6 DK |
1104 | { |
1105 | struct buffer_head *bh; | |
1106 | journal_t *journal = journal_init_common(); | |
05496769 | 1107 | char *p; |
470decc6 DK |
1108 | int err; |
1109 | int n; | |
18eba7aa | 1110 | unsigned long long blocknr; |
470decc6 DK |
1111 | |
1112 | if (!journal) | |
1113 | return NULL; | |
1114 | ||
1115 | journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev; | |
1116 | journal->j_inode = inode; | |
05496769 TT |
1117 | bdevname(journal->j_dev, journal->j_devname); |
1118 | p = journal->j_devname; | |
1119 | while ((p = strchr(p, '/'))) | |
1120 | *p = '!'; | |
1121 | p = journal->j_devname + strlen(journal->j_devname); | |
90576c0b | 1122 | sprintf(p, "-%lu", journal->j_inode->i_ino); |
470decc6 DK |
1123 | jbd_debug(1, |
1124 | "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n", | |
1125 | journal, inode->i_sb->s_id, inode->i_ino, | |
1126 | (long long) inode->i_size, | |
1127 | inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); | |
1128 | ||
1129 | journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits; | |
1130 | journal->j_blocksize = inode->i_sb->s_blocksize; | |
8e85fb3f | 1131 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1132 | |
1133 | /* journal descriptor can store up to n blocks -bzzz */ | |
1134 | n = journal->j_blocksize / sizeof(journal_block_tag_t); | |
1135 | journal->j_wbufsize = n; | |
1136 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1137 | if (!journal->j_wbuf) { | |
25985edc | 1138 | printk(KERN_ERR "%s: Can't allocate bhs for commit thread\n", |
329d291f | 1139 | __func__); |
4b905671 | 1140 | goto out_err; |
470decc6 DK |
1141 | } |
1142 | ||
f7f4bccb | 1143 | err = jbd2_journal_bmap(journal, 0, &blocknr); |
470decc6 DK |
1144 | /* If that failed, give up */ |
1145 | if (err) { | |
3c26bdb4 | 1146 | printk(KERN_ERR "%s: Cannot locate journal superblock\n", |
329d291f | 1147 | __func__); |
4b905671 | 1148 | goto out_err; |
470decc6 DK |
1149 | } |
1150 | ||
1151 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
4b905671 JK |
1152 | if (!bh) { |
1153 | printk(KERN_ERR | |
1154 | "%s: Cannot get buffer for journal superblock\n", | |
1155 | __func__); | |
1156 | goto out_err; | |
1157 | } | |
470decc6 DK |
1158 | journal->j_sb_buffer = bh; |
1159 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
1160 | ||
1161 | return journal; | |
4b905671 | 1162 | out_err: |
7b02bec0 | 1163 | kfree(journal->j_wbuf); |
4b905671 JK |
1164 | jbd2_stats_proc_exit(journal); |
1165 | kfree(journal); | |
1166 | return NULL; | |
470decc6 DK |
1167 | } |
1168 | ||
1169 | /* | |
1170 | * If the journal init or create aborts, we need to mark the journal | |
1171 | * superblock as being NULL to prevent the journal destroy from writing | |
1172 | * back a bogus superblock. | |
1173 | */ | |
1174 | static void journal_fail_superblock (journal_t *journal) | |
1175 | { | |
1176 | struct buffer_head *bh = journal->j_sb_buffer; | |
1177 | brelse(bh); | |
1178 | journal->j_sb_buffer = NULL; | |
1179 | } | |
1180 | ||
1181 | /* | |
1182 | * Given a journal_t structure, initialise the various fields for | |
1183 | * startup of a new journaling session. We use this both when creating | |
1184 | * a journal, and after recovering an old journal to reset it for | |
1185 | * subsequent use. | |
1186 | */ | |
1187 | ||
1188 | static int journal_reset(journal_t *journal) | |
1189 | { | |
1190 | journal_superblock_t *sb = journal->j_superblock; | |
18eba7aa | 1191 | unsigned long long first, last; |
470decc6 DK |
1192 | |
1193 | first = be32_to_cpu(sb->s_first); | |
1194 | last = be32_to_cpu(sb->s_maxlen); | |
f6f50e28 | 1195 | if (first + JBD2_MIN_JOURNAL_BLOCKS > last + 1) { |
f2a44523 | 1196 | printk(KERN_ERR "JBD2: Journal too short (blocks %llu-%llu).\n", |
f6f50e28 JK |
1197 | first, last); |
1198 | journal_fail_superblock(journal); | |
1199 | return -EINVAL; | |
1200 | } | |
470decc6 DK |
1201 | |
1202 | journal->j_first = first; | |
1203 | journal->j_last = last; | |
1204 | ||
1205 | journal->j_head = first; | |
1206 | journal->j_tail = first; | |
1207 | journal->j_free = last - first; | |
1208 | ||
1209 | journal->j_tail_sequence = journal->j_transaction_sequence; | |
1210 | journal->j_commit_sequence = journal->j_transaction_sequence - 1; | |
1211 | journal->j_commit_request = journal->j_commit_sequence; | |
1212 | ||
1213 | journal->j_max_transaction_buffers = journal->j_maxlen / 4; | |
1214 | ||
470decc6 DK |
1215 | /* |
1216 | * As a special case, if the on-disk copy is already marked as needing | |
24bcc89c JK |
1217 | * no recovery (s_start == 0), then we can safely defer the superblock |
1218 | * update until the next commit by setting JBD2_FLUSHED. This avoids | |
470decc6 DK |
1219 | * attempting a write to a potential-readonly device. |
1220 | */ | |
24bcc89c | 1221 | if (sb->s_start == 0) { |
f2a44523 | 1222 | jbd_debug(1, "JBD2: Skipping superblock update on recovered sb " |
470decc6 DK |
1223 | "(start %ld, seq %d, errno %d)\n", |
1224 | journal->j_tail, journal->j_tail_sequence, | |
1225 | journal->j_errno); | |
24bcc89c JK |
1226 | journal->j_flags |= JBD2_FLUSHED; |
1227 | } else { | |
a78bb11d JK |
1228 | /* Lock here to make assertions happy... */ |
1229 | mutex_lock(&journal->j_checkpoint_mutex); | |
79feb521 JK |
1230 | /* |
1231 | * Update log tail information. We use WRITE_FUA since new | |
1232 | * transaction will start reusing journal space and so we | |
1233 | * must make sure information about current log tail is on | |
1234 | * disk before that. | |
1235 | */ | |
1236 | jbd2_journal_update_sb_log_tail(journal, | |
1237 | journal->j_tail_sequence, | |
1238 | journal->j_tail, | |
1239 | WRITE_FUA); | |
a78bb11d | 1240 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 | 1241 | } |
24bcc89c JK |
1242 | return jbd2_journal_start_thread(journal); |
1243 | } | |
470decc6 | 1244 | |
79feb521 | 1245 | static void jbd2_write_superblock(journal_t *journal, int write_op) |
24bcc89c JK |
1246 | { |
1247 | struct buffer_head *bh = journal->j_sb_buffer; | |
79feb521 | 1248 | int ret; |
470decc6 | 1249 | |
79feb521 JK |
1250 | trace_jbd2_write_superblock(journal, write_op); |
1251 | if (!(journal->j_flags & JBD2_BARRIER)) | |
1252 | write_op &= ~(REQ_FUA | REQ_FLUSH); | |
1253 | lock_buffer(bh); | |
914258bf TT |
1254 | if (buffer_write_io_error(bh)) { |
1255 | /* | |
1256 | * Oh, dear. A previous attempt to write the journal | |
1257 | * superblock failed. This could happen because the | |
1258 | * USB device was yanked out. Or it could happen to | |
1259 | * be a transient write error and maybe the block will | |
1260 | * be remapped. Nothing we can do but to retry the | |
1261 | * write and hope for the best. | |
1262 | */ | |
1263 | printk(KERN_ERR "JBD2: previous I/O error detected " | |
1264 | "for journal superblock update for %s.\n", | |
1265 | journal->j_devname); | |
1266 | clear_buffer_write_io_error(bh); | |
1267 | set_buffer_uptodate(bh); | |
1268 | } | |
79feb521 JK |
1269 | get_bh(bh); |
1270 | bh->b_end_io = end_buffer_write_sync; | |
1271 | ret = submit_bh(write_op, bh); | |
1272 | wait_on_buffer(bh); | |
24bcc89c | 1273 | if (buffer_write_io_error(bh)) { |
24bcc89c JK |
1274 | clear_buffer_write_io_error(bh); |
1275 | set_buffer_uptodate(bh); | |
79feb521 JK |
1276 | ret = -EIO; |
1277 | } | |
1278 | if (ret) { | |
1279 | printk(KERN_ERR "JBD2: Error %d detected when updating " | |
1280 | "journal superblock for %s.\n", ret, | |
1281 | journal->j_devname); | |
24bcc89c JK |
1282 | } |
1283 | } | |
1284 | ||
1285 | /** | |
1286 | * jbd2_journal_update_sb_log_tail() - Update log tail in journal sb on disk. | |
1287 | * @journal: The journal to update. | |
79feb521 JK |
1288 | * @tail_tid: TID of the new transaction at the tail of the log |
1289 | * @tail_block: The first block of the transaction at the tail of the log | |
1290 | * @write_op: With which operation should we write the journal sb | |
24bcc89c JK |
1291 | * |
1292 | * Update a journal's superblock information about log tail and write it to | |
1293 | * disk, waiting for the IO to complete. | |
1294 | */ | |
79feb521 JK |
1295 | void jbd2_journal_update_sb_log_tail(journal_t *journal, tid_t tail_tid, |
1296 | unsigned long tail_block, int write_op) | |
24bcc89c JK |
1297 | { |
1298 | journal_superblock_t *sb = journal->j_superblock; | |
1299 | ||
a78bb11d | 1300 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); |
79feb521 JK |
1301 | jbd_debug(1, "JBD2: updating superblock (start %lu, seq %u)\n", |
1302 | tail_block, tail_tid); | |
470decc6 | 1303 | |
79feb521 JK |
1304 | sb->s_sequence = cpu_to_be32(tail_tid); |
1305 | sb->s_start = cpu_to_be32(tail_block); | |
470decc6 | 1306 | |
79feb521 | 1307 | jbd2_write_superblock(journal, write_op); |
470decc6 | 1308 | |
24bcc89c JK |
1309 | /* Log is no longer empty */ |
1310 | write_lock(&journal->j_state_lock); | |
1311 | WARN_ON(!sb->s_sequence); | |
1312 | journal->j_flags &= ~JBD2_FLUSHED; | |
1313 | write_unlock(&journal->j_state_lock); | |
1314 | } | |
2201c590 | 1315 | |
24bcc89c JK |
1316 | /** |
1317 | * jbd2_mark_journal_empty() - Mark on disk journal as empty. | |
1318 | * @journal: The journal to update. | |
1319 | * | |
1320 | * Update a journal's dynamic superblock fields to show that journal is empty. | |
1321 | * Write updated superblock to disk waiting for IO to complete. | |
1322 | */ | |
1323 | static void jbd2_mark_journal_empty(journal_t *journal) | |
1324 | { | |
1325 | journal_superblock_t *sb = journal->j_superblock; | |
914258bf | 1326 | |
a78bb11d | 1327 | BUG_ON(!mutex_is_locked(&journal->j_checkpoint_mutex)); |
a931da6a | 1328 | read_lock(&journal->j_state_lock); |
24bcc89c JK |
1329 | jbd_debug(1, "JBD2: Marking journal as empty (seq %d)\n", |
1330 | journal->j_tail_sequence); | |
470decc6 DK |
1331 | |
1332 | sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); | |
24bcc89c | 1333 | sb->s_start = cpu_to_be32(0); |
a931da6a | 1334 | read_unlock(&journal->j_state_lock); |
470decc6 | 1335 | |
79feb521 | 1336 | jbd2_write_superblock(journal, WRITE_FUA); |
470decc6 | 1337 | |
24bcc89c | 1338 | /* Log is no longer empty */ |
a931da6a | 1339 | write_lock(&journal->j_state_lock); |
24bcc89c | 1340 | journal->j_flags |= JBD2_FLUSHED; |
a931da6a | 1341 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1342 | } |
1343 | ||
24bcc89c JK |
1344 | |
1345 | /** | |
1346 | * jbd2_journal_update_sb_errno() - Update error in the journal. | |
1347 | * @journal: The journal to update. | |
1348 | * | |
1349 | * Update a journal's errno. Write updated superblock to disk waiting for IO | |
1350 | * to complete. | |
1351 | */ | |
1352 | static void jbd2_journal_update_sb_errno(journal_t *journal) | |
1353 | { | |
1354 | journal_superblock_t *sb = journal->j_superblock; | |
1355 | ||
1356 | read_lock(&journal->j_state_lock); | |
1357 | jbd_debug(1, "JBD2: updating superblock error (errno %d)\n", | |
1358 | journal->j_errno); | |
1359 | sb->s_errno = cpu_to_be32(journal->j_errno); | |
1360 | read_unlock(&journal->j_state_lock); | |
1361 | ||
79feb521 | 1362 | jbd2_write_superblock(journal, WRITE_SYNC); |
24bcc89c JK |
1363 | } |
1364 | ||
470decc6 DK |
1365 | /* |
1366 | * Read the superblock for a given journal, performing initial | |
1367 | * validation of the format. | |
1368 | */ | |
470decc6 DK |
1369 | static int journal_get_superblock(journal_t *journal) |
1370 | { | |
1371 | struct buffer_head *bh; | |
1372 | journal_superblock_t *sb; | |
1373 | int err = -EIO; | |
1374 | ||
1375 | bh = journal->j_sb_buffer; | |
1376 | ||
1377 | J_ASSERT(bh != NULL); | |
1378 | if (!buffer_uptodate(bh)) { | |
1379 | ll_rw_block(READ, 1, &bh); | |
1380 | wait_on_buffer(bh); | |
1381 | if (!buffer_uptodate(bh)) { | |
f2a44523 EG |
1382 | printk(KERN_ERR |
1383 | "JBD2: IO error reading journal superblock\n"); | |
470decc6 DK |
1384 | goto out; |
1385 | } | |
1386 | } | |
1387 | ||
25ed6e8a DW |
1388 | if (buffer_verified(bh)) |
1389 | return 0; | |
1390 | ||
470decc6 DK |
1391 | sb = journal->j_superblock; |
1392 | ||
1393 | err = -EINVAL; | |
1394 | ||
f7f4bccb | 1395 | if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || |
470decc6 | 1396 | sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { |
f2a44523 | 1397 | printk(KERN_WARNING "JBD2: no valid journal superblock found\n"); |
470decc6 DK |
1398 | goto out; |
1399 | } | |
1400 | ||
1401 | switch(be32_to_cpu(sb->s_header.h_blocktype)) { | |
f7f4bccb | 1402 | case JBD2_SUPERBLOCK_V1: |
470decc6 DK |
1403 | journal->j_format_version = 1; |
1404 | break; | |
f7f4bccb | 1405 | case JBD2_SUPERBLOCK_V2: |
470decc6 DK |
1406 | journal->j_format_version = 2; |
1407 | break; | |
1408 | default: | |
f2a44523 | 1409 | printk(KERN_WARNING "JBD2: unrecognised superblock format ID\n"); |
470decc6 DK |
1410 | goto out; |
1411 | } | |
1412 | ||
1413 | if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen) | |
1414 | journal->j_maxlen = be32_to_cpu(sb->s_maxlen); | |
1415 | else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) { | |
f2a44523 | 1416 | printk(KERN_WARNING "JBD2: journal file too short\n"); |
470decc6 DK |
1417 | goto out; |
1418 | } | |
1419 | ||
8762202d EG |
1420 | if (be32_to_cpu(sb->s_first) == 0 || |
1421 | be32_to_cpu(sb->s_first) >= journal->j_maxlen) { | |
1422 | printk(KERN_WARNING | |
1423 | "JBD2: Invalid start block of journal: %u\n", | |
1424 | be32_to_cpu(sb->s_first)); | |
1425 | goto out; | |
1426 | } | |
1427 | ||
25ed6e8a DW |
1428 | if (JBD2_HAS_COMPAT_FEATURE(journal, JBD2_FEATURE_COMPAT_CHECKSUM) && |
1429 | JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_CSUM_V2)) { | |
1430 | /* Can't have checksum v1 and v2 on at the same time! */ | |
1431 | printk(KERN_ERR "JBD: Can't enable checksumming v1 and v2 " | |
1432 | "at the same time!\n"); | |
1433 | goto out; | |
1434 | } | |
1435 | ||
1436 | if (!jbd2_verify_csum_type(journal, sb)) { | |
1437 | printk(KERN_ERR "JBD: Unknown checksum type\n"); | |
1438 | goto out; | |
1439 | } | |
1440 | ||
1441 | set_buffer_verified(bh); | |
1442 | ||
470decc6 DK |
1443 | return 0; |
1444 | ||
1445 | out: | |
1446 | journal_fail_superblock(journal); | |
1447 | return err; | |
1448 | } | |
1449 | ||
1450 | /* | |
1451 | * Load the on-disk journal superblock and read the key fields into the | |
1452 | * journal_t. | |
1453 | */ | |
1454 | ||
1455 | static int load_superblock(journal_t *journal) | |
1456 | { | |
1457 | int err; | |
1458 | journal_superblock_t *sb; | |
1459 | ||
1460 | err = journal_get_superblock(journal); | |
1461 | if (err) | |
1462 | return err; | |
1463 | ||
1464 | sb = journal->j_superblock; | |
1465 | ||
1466 | journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); | |
1467 | journal->j_tail = be32_to_cpu(sb->s_start); | |
1468 | journal->j_first = be32_to_cpu(sb->s_first); | |
1469 | journal->j_last = be32_to_cpu(sb->s_maxlen); | |
1470 | journal->j_errno = be32_to_cpu(sb->s_errno); | |
1471 | ||
1472 | return 0; | |
1473 | } | |
1474 | ||
1475 | ||
1476 | /** | |
f7f4bccb | 1477 | * int jbd2_journal_load() - Read journal from disk. |
470decc6 DK |
1478 | * @journal: Journal to act on. |
1479 | * | |
1480 | * Given a journal_t structure which tells us which disk blocks contain | |
1481 | * a journal, read the journal from disk to initialise the in-memory | |
1482 | * structures. | |
1483 | */ | |
f7f4bccb | 1484 | int jbd2_journal_load(journal_t *journal) |
470decc6 DK |
1485 | { |
1486 | int err; | |
1487 | journal_superblock_t *sb; | |
1488 | ||
1489 | err = load_superblock(journal); | |
1490 | if (err) | |
1491 | return err; | |
1492 | ||
1493 | sb = journal->j_superblock; | |
1494 | /* If this is a V2 superblock, then we have to check the | |
1495 | * features flags on it. */ | |
1496 | ||
1497 | if (journal->j_format_version >= 2) { | |
1498 | if ((sb->s_feature_ro_compat & | |
f7f4bccb | 1499 | ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || |
470decc6 | 1500 | (sb->s_feature_incompat & |
f7f4bccb | 1501 | ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { |
f2a44523 EG |
1502 | printk(KERN_WARNING |
1503 | "JBD2: Unrecognised features on journal\n"); | |
470decc6 DK |
1504 | return -EINVAL; |
1505 | } | |
1506 | } | |
1507 | ||
d2eecb03 TT |
1508 | /* |
1509 | * Create a slab for this blocksize | |
1510 | */ | |
1511 | err = jbd2_journal_create_slab(be32_to_cpu(sb->s_blocksize)); | |
1512 | if (err) | |
1513 | return err; | |
1514 | ||
470decc6 DK |
1515 | /* Let the recovery code check whether it needs to recover any |
1516 | * data from the journal. */ | |
f7f4bccb | 1517 | if (jbd2_journal_recover(journal)) |
470decc6 DK |
1518 | goto recovery_error; |
1519 | ||
e6a47428 TT |
1520 | if (journal->j_failed_commit) { |
1521 | printk(KERN_ERR "JBD2: journal transaction %u on %s " | |
1522 | "is corrupt.\n", journal->j_failed_commit, | |
1523 | journal->j_devname); | |
1524 | return -EIO; | |
1525 | } | |
1526 | ||
470decc6 DK |
1527 | /* OK, we've finished with the dynamic journal bits: |
1528 | * reinitialise the dynamic contents of the superblock in memory | |
1529 | * and reset them on disk. */ | |
1530 | if (journal_reset(journal)) | |
1531 | goto recovery_error; | |
1532 | ||
f7f4bccb MC |
1533 | journal->j_flags &= ~JBD2_ABORT; |
1534 | journal->j_flags |= JBD2_LOADED; | |
470decc6 DK |
1535 | return 0; |
1536 | ||
1537 | recovery_error: | |
f2a44523 | 1538 | printk(KERN_WARNING "JBD2: recovery failed\n"); |
470decc6 DK |
1539 | return -EIO; |
1540 | } | |
1541 | ||
1542 | /** | |
f7f4bccb | 1543 | * void jbd2_journal_destroy() - Release a journal_t structure. |
470decc6 DK |
1544 | * @journal: Journal to act on. |
1545 | * | |
1546 | * Release a journal_t structure once it is no longer in use by the | |
1547 | * journaled object. | |
44519faf | 1548 | * Return <0 if we couldn't clean up the journal. |
470decc6 | 1549 | */ |
44519faf | 1550 | int jbd2_journal_destroy(journal_t *journal) |
470decc6 | 1551 | { |
44519faf HK |
1552 | int err = 0; |
1553 | ||
470decc6 DK |
1554 | /* Wait for the commit thread to wake up and die. */ |
1555 | journal_kill_thread(journal); | |
1556 | ||
1557 | /* Force a final log commit */ | |
1558 | if (journal->j_running_transaction) | |
f7f4bccb | 1559 | jbd2_journal_commit_transaction(journal); |
470decc6 DK |
1560 | |
1561 | /* Force any old transactions to disk */ | |
1562 | ||
1563 | /* Totally anal locking here... */ | |
1564 | spin_lock(&journal->j_list_lock); | |
1565 | while (journal->j_checkpoint_transactions != NULL) { | |
1566 | spin_unlock(&journal->j_list_lock); | |
1a0d3786 | 1567 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1568 | jbd2_log_do_checkpoint(journal); |
1a0d3786 | 1569 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
1570 | spin_lock(&journal->j_list_lock); |
1571 | } | |
1572 | ||
1573 | J_ASSERT(journal->j_running_transaction == NULL); | |
1574 | J_ASSERT(journal->j_committing_transaction == NULL); | |
1575 | J_ASSERT(journal->j_checkpoint_transactions == NULL); | |
1576 | spin_unlock(&journal->j_list_lock); | |
1577 | ||
470decc6 | 1578 | if (journal->j_sb_buffer) { |
44519faf | 1579 | if (!is_journal_aborted(journal)) { |
a78bb11d | 1580 | mutex_lock(&journal->j_checkpoint_mutex); |
24bcc89c | 1581 | jbd2_mark_journal_empty(journal); |
a78bb11d JK |
1582 | mutex_unlock(&journal->j_checkpoint_mutex); |
1583 | } else | |
44519faf | 1584 | err = -EIO; |
470decc6 DK |
1585 | brelse(journal->j_sb_buffer); |
1586 | } | |
1587 | ||
8e85fb3f JL |
1588 | if (journal->j_proc_entry) |
1589 | jbd2_stats_proc_exit(journal); | |
470decc6 DK |
1590 | if (journal->j_inode) |
1591 | iput(journal->j_inode); | |
1592 | if (journal->j_revoke) | |
f7f4bccb | 1593 | jbd2_journal_destroy_revoke(journal); |
470decc6 DK |
1594 | kfree(journal->j_wbuf); |
1595 | kfree(journal); | |
44519faf HK |
1596 | |
1597 | return err; | |
470decc6 DK |
1598 | } |
1599 | ||
1600 | ||
1601 | /** | |
f7f4bccb | 1602 | *int jbd2_journal_check_used_features () - Check if features specified are used. |
470decc6 DK |
1603 | * @journal: Journal to check. |
1604 | * @compat: bitmask of compatible features | |
1605 | * @ro: bitmask of features that force read-only mount | |
1606 | * @incompat: bitmask of incompatible features | |
1607 | * | |
1608 | * Check whether the journal uses all of a given set of | |
1609 | * features. Return true (non-zero) if it does. | |
1610 | **/ | |
1611 | ||
f7f4bccb | 1612 | int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1613 | unsigned long ro, unsigned long incompat) |
1614 | { | |
1615 | journal_superblock_t *sb; | |
1616 | ||
1617 | if (!compat && !ro && !incompat) | |
1618 | return 1; | |
1113e1b5 PL |
1619 | /* Load journal superblock if it is not loaded yet. */ |
1620 | if (journal->j_format_version == 0 && | |
1621 | journal_get_superblock(journal) != 0) | |
1622 | return 0; | |
470decc6 DK |
1623 | if (journal->j_format_version == 1) |
1624 | return 0; | |
1625 | ||
1626 | sb = journal->j_superblock; | |
1627 | ||
1628 | if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && | |
1629 | ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && | |
1630 | ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) | |
1631 | return 1; | |
1632 | ||
1633 | return 0; | |
1634 | } | |
1635 | ||
1636 | /** | |
f7f4bccb | 1637 | * int jbd2_journal_check_available_features() - Check feature set in journalling layer |
470decc6 DK |
1638 | * @journal: Journal to check. |
1639 | * @compat: bitmask of compatible features | |
1640 | * @ro: bitmask of features that force read-only mount | |
1641 | * @incompat: bitmask of incompatible features | |
1642 | * | |
1643 | * Check whether the journaling code supports the use of | |
1644 | * all of a given set of features on this journal. Return true | |
1645 | * (non-zero) if it can. */ | |
1646 | ||
f7f4bccb | 1647 | int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1648 | unsigned long ro, unsigned long incompat) |
1649 | { | |
470decc6 DK |
1650 | if (!compat && !ro && !incompat) |
1651 | return 1; | |
1652 | ||
470decc6 DK |
1653 | /* We can support any known requested features iff the |
1654 | * superblock is in version 2. Otherwise we fail to support any | |
1655 | * extended sb features. */ | |
1656 | ||
1657 | if (journal->j_format_version != 2) | |
1658 | return 0; | |
1659 | ||
f7f4bccb MC |
1660 | if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && |
1661 | (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && | |
1662 | (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) | |
470decc6 DK |
1663 | return 1; |
1664 | ||
1665 | return 0; | |
1666 | } | |
1667 | ||
1668 | /** | |
f7f4bccb | 1669 | * int jbd2_journal_set_features () - Mark a given journal feature in the superblock |
470decc6 DK |
1670 | * @journal: Journal to act on. |
1671 | * @compat: bitmask of compatible features | |
1672 | * @ro: bitmask of features that force read-only mount | |
1673 | * @incompat: bitmask of incompatible features | |
1674 | * | |
1675 | * Mark a given journal feature as present on the | |
1676 | * superblock. Returns true if the requested features could be set. | |
1677 | * | |
1678 | */ | |
1679 | ||
f7f4bccb | 1680 | int jbd2_journal_set_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1681 | unsigned long ro, unsigned long incompat) |
1682 | { | |
25ed6e8a DW |
1683 | #define INCOMPAT_FEATURE_ON(f) \ |
1684 | ((incompat & (f)) && !(sb->s_feature_incompat & cpu_to_be32(f))) | |
1685 | #define COMPAT_FEATURE_ON(f) \ | |
1686 | ((compat & (f)) && !(sb->s_feature_compat & cpu_to_be32(f))) | |
470decc6 DK |
1687 | journal_superblock_t *sb; |
1688 | ||
f7f4bccb | 1689 | if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) |
470decc6 DK |
1690 | return 1; |
1691 | ||
f7f4bccb | 1692 | if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) |
470decc6 DK |
1693 | return 0; |
1694 | ||
25ed6e8a DW |
1695 | /* Asking for checksumming v2 and v1? Only give them v2. */ |
1696 | if (incompat & JBD2_FEATURE_INCOMPAT_CSUM_V2 && | |
1697 | compat & JBD2_FEATURE_COMPAT_CHECKSUM) | |
1698 | compat &= ~JBD2_FEATURE_COMPAT_CHECKSUM; | |
1699 | ||
470decc6 DK |
1700 | jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", |
1701 | compat, ro, incompat); | |
1702 | ||
1703 | sb = journal->j_superblock; | |
1704 | ||
25ed6e8a DW |
1705 | /* If enabling v2 checksums, update superblock */ |
1706 | if (INCOMPAT_FEATURE_ON(JBD2_FEATURE_INCOMPAT_CSUM_V2)) { | |
1707 | sb->s_checksum_type = JBD2_CRC32C_CHKSUM; | |
1708 | sb->s_feature_compat &= | |
1709 | ~cpu_to_be32(JBD2_FEATURE_COMPAT_CHECKSUM); | |
1710 | } | |
1711 | ||
1712 | /* If enabling v1 checksums, downgrade superblock */ | |
1713 | if (COMPAT_FEATURE_ON(JBD2_FEATURE_COMPAT_CHECKSUM)) | |
1714 | sb->s_feature_incompat &= | |
1715 | ~cpu_to_be32(JBD2_FEATURE_INCOMPAT_CSUM_V2); | |
1716 | ||
470decc6 DK |
1717 | sb->s_feature_compat |= cpu_to_be32(compat); |
1718 | sb->s_feature_ro_compat |= cpu_to_be32(ro); | |
1719 | sb->s_feature_incompat |= cpu_to_be32(incompat); | |
1720 | ||
1721 | return 1; | |
25ed6e8a DW |
1722 | #undef COMPAT_FEATURE_ON |
1723 | #undef INCOMPAT_FEATURE_ON | |
470decc6 DK |
1724 | } |
1725 | ||
818d276c GS |
1726 | /* |
1727 | * jbd2_journal_clear_features () - Clear a given journal feature in the | |
1728 | * superblock | |
1729 | * @journal: Journal to act on. | |
1730 | * @compat: bitmask of compatible features | |
1731 | * @ro: bitmask of features that force read-only mount | |
1732 | * @incompat: bitmask of incompatible features | |
1733 | * | |
1734 | * Clear a given journal feature as present on the | |
1735 | * superblock. | |
1736 | */ | |
1737 | void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, | |
1738 | unsigned long ro, unsigned long incompat) | |
1739 | { | |
1740 | journal_superblock_t *sb; | |
1741 | ||
1742 | jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", | |
1743 | compat, ro, incompat); | |
1744 | ||
1745 | sb = journal->j_superblock; | |
1746 | ||
1747 | sb->s_feature_compat &= ~cpu_to_be32(compat); | |
1748 | sb->s_feature_ro_compat &= ~cpu_to_be32(ro); | |
1749 | sb->s_feature_incompat &= ~cpu_to_be32(incompat); | |
1750 | } | |
1751 | EXPORT_SYMBOL(jbd2_journal_clear_features); | |
470decc6 | 1752 | |
470decc6 | 1753 | /** |
f7f4bccb | 1754 | * int jbd2_journal_flush () - Flush journal |
470decc6 DK |
1755 | * @journal: Journal to act on. |
1756 | * | |
1757 | * Flush all data for a given journal to disk and empty the journal. | |
1758 | * Filesystems can use this when remounting readonly to ensure that | |
1759 | * recovery does not need to happen on remount. | |
1760 | */ | |
1761 | ||
f7f4bccb | 1762 | int jbd2_journal_flush(journal_t *journal) |
470decc6 DK |
1763 | { |
1764 | int err = 0; | |
1765 | transaction_t *transaction = NULL; | |
470decc6 | 1766 | |
a931da6a | 1767 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1768 | |
1769 | /* Force everything buffered to the log... */ | |
1770 | if (journal->j_running_transaction) { | |
1771 | transaction = journal->j_running_transaction; | |
f7f4bccb | 1772 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1773 | } else if (journal->j_committing_transaction) |
1774 | transaction = journal->j_committing_transaction; | |
1775 | ||
1776 | /* Wait for the log commit to complete... */ | |
1777 | if (transaction) { | |
1778 | tid_t tid = transaction->t_tid; | |
1779 | ||
a931da6a | 1780 | write_unlock(&journal->j_state_lock); |
f7f4bccb | 1781 | jbd2_log_wait_commit(journal, tid); |
470decc6 | 1782 | } else { |
a931da6a | 1783 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1784 | } |
1785 | ||
1786 | /* ...and flush everything in the log out to disk. */ | |
1787 | spin_lock(&journal->j_list_lock); | |
1788 | while (!err && journal->j_checkpoint_transactions != NULL) { | |
1789 | spin_unlock(&journal->j_list_lock); | |
44519faf | 1790 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1791 | err = jbd2_log_do_checkpoint(journal); |
44519faf | 1792 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
1793 | spin_lock(&journal->j_list_lock); |
1794 | } | |
1795 | spin_unlock(&journal->j_list_lock); | |
44519faf HK |
1796 | |
1797 | if (is_journal_aborted(journal)) | |
1798 | return -EIO; | |
1799 | ||
a78bb11d | 1800 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1801 | jbd2_cleanup_journal_tail(journal); |
470decc6 DK |
1802 | |
1803 | /* Finally, mark the journal as really needing no recovery. | |
1804 | * This sets s_start==0 in the underlying superblock, which is | |
1805 | * the magic code for a fully-recovered superblock. Any future | |
1806 | * commits of data to the journal will restore the current | |
1807 | * s_start value. */ | |
24bcc89c | 1808 | jbd2_mark_journal_empty(journal); |
a78bb11d | 1809 | mutex_unlock(&journal->j_checkpoint_mutex); |
a931da6a | 1810 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1811 | J_ASSERT(!journal->j_running_transaction); |
1812 | J_ASSERT(!journal->j_committing_transaction); | |
1813 | J_ASSERT(!journal->j_checkpoint_transactions); | |
1814 | J_ASSERT(journal->j_head == journal->j_tail); | |
1815 | J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); | |
a931da6a | 1816 | write_unlock(&journal->j_state_lock); |
44519faf | 1817 | return 0; |
470decc6 DK |
1818 | } |
1819 | ||
1820 | /** | |
f7f4bccb | 1821 | * int jbd2_journal_wipe() - Wipe journal contents |
470decc6 DK |
1822 | * @journal: Journal to act on. |
1823 | * @write: flag (see below) | |
1824 | * | |
1825 | * Wipe out all of the contents of a journal, safely. This will produce | |
1826 | * a warning if the journal contains any valid recovery information. | |
f7f4bccb | 1827 | * Must be called between journal_init_*() and jbd2_journal_load(). |
470decc6 DK |
1828 | * |
1829 | * If 'write' is non-zero, then we wipe out the journal on disk; otherwise | |
1830 | * we merely suppress recovery. | |
1831 | */ | |
1832 | ||
f7f4bccb | 1833 | int jbd2_journal_wipe(journal_t *journal, int write) |
470decc6 | 1834 | { |
470decc6 DK |
1835 | int err = 0; |
1836 | ||
f7f4bccb | 1837 | J_ASSERT (!(journal->j_flags & JBD2_LOADED)); |
470decc6 DK |
1838 | |
1839 | err = load_superblock(journal); | |
1840 | if (err) | |
1841 | return err; | |
1842 | ||
470decc6 DK |
1843 | if (!journal->j_tail) |
1844 | goto no_recovery; | |
1845 | ||
f2a44523 | 1846 | printk(KERN_WARNING "JBD2: %s recovery information on journal\n", |
470decc6 DK |
1847 | write ? "Clearing" : "Ignoring"); |
1848 | ||
f7f4bccb | 1849 | err = jbd2_journal_skip_recovery(journal); |
a78bb11d JK |
1850 | if (write) { |
1851 | /* Lock to make assertions happy... */ | |
1852 | mutex_lock(&journal->j_checkpoint_mutex); | |
24bcc89c | 1853 | jbd2_mark_journal_empty(journal); |
a78bb11d JK |
1854 | mutex_unlock(&journal->j_checkpoint_mutex); |
1855 | } | |
470decc6 DK |
1856 | |
1857 | no_recovery: | |
1858 | return err; | |
1859 | } | |
1860 | ||
470decc6 DK |
1861 | /* |
1862 | * Journal abort has very specific semantics, which we describe | |
1863 | * for journal abort. | |
1864 | * | |
bfcd3555 | 1865 | * Two internal functions, which provide abort to the jbd layer |
470decc6 DK |
1866 | * itself are here. |
1867 | */ | |
1868 | ||
1869 | /* | |
1870 | * Quick version for internal journal use (doesn't lock the journal). | |
1871 | * Aborts hard --- we mark the abort as occurred, but do _nothing_ else, | |
1872 | * and don't attempt to make any other journal updates. | |
1873 | */ | |
f7f4bccb | 1874 | void __jbd2_journal_abort_hard(journal_t *journal) |
470decc6 DK |
1875 | { |
1876 | transaction_t *transaction; | |
470decc6 | 1877 | |
f7f4bccb | 1878 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1879 | return; |
1880 | ||
1881 | printk(KERN_ERR "Aborting journal on device %s.\n", | |
05496769 | 1882 | journal->j_devname); |
470decc6 | 1883 | |
a931da6a | 1884 | write_lock(&journal->j_state_lock); |
f7f4bccb | 1885 | journal->j_flags |= JBD2_ABORT; |
470decc6 DK |
1886 | transaction = journal->j_running_transaction; |
1887 | if (transaction) | |
f7f4bccb | 1888 | __jbd2_log_start_commit(journal, transaction->t_tid); |
a931da6a | 1889 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1890 | } |
1891 | ||
1892 | /* Soft abort: record the abort error status in the journal superblock, | |
1893 | * but don't do any other IO. */ | |
1894 | static void __journal_abort_soft (journal_t *journal, int errno) | |
1895 | { | |
f7f4bccb | 1896 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1897 | return; |
1898 | ||
1899 | if (!journal->j_errno) | |
1900 | journal->j_errno = errno; | |
1901 | ||
f7f4bccb | 1902 | __jbd2_journal_abort_hard(journal); |
470decc6 DK |
1903 | |
1904 | if (errno) | |
24bcc89c | 1905 | jbd2_journal_update_sb_errno(journal); |
470decc6 DK |
1906 | } |
1907 | ||
1908 | /** | |
f7f4bccb | 1909 | * void jbd2_journal_abort () - Shutdown the journal immediately. |
470decc6 DK |
1910 | * @journal: the journal to shutdown. |
1911 | * @errno: an error number to record in the journal indicating | |
1912 | * the reason for the shutdown. | |
1913 | * | |
1914 | * Perform a complete, immediate shutdown of the ENTIRE | |
1915 | * journal (not of a single transaction). This operation cannot be | |
1916 | * undone without closing and reopening the journal. | |
1917 | * | |
f7f4bccb | 1918 | * The jbd2_journal_abort function is intended to support higher level error |
470decc6 DK |
1919 | * recovery mechanisms such as the ext2/ext3 remount-readonly error |
1920 | * mode. | |
1921 | * | |
1922 | * Journal abort has very specific semantics. Any existing dirty, | |
1923 | * unjournaled buffers in the main filesystem will still be written to | |
1924 | * disk by bdflush, but the journaling mechanism will be suspended | |
1925 | * immediately and no further transaction commits will be honoured. | |
1926 | * | |
1927 | * Any dirty, journaled buffers will be written back to disk without | |
1928 | * hitting the journal. Atomicity cannot be guaranteed on an aborted | |
1929 | * filesystem, but we _do_ attempt to leave as much data as possible | |
1930 | * behind for fsck to use for cleanup. | |
1931 | * | |
1932 | * Any attempt to get a new transaction handle on a journal which is in | |
1933 | * ABORT state will just result in an -EROFS error return. A | |
f7f4bccb | 1934 | * jbd2_journal_stop on an existing handle will return -EIO if we have |
470decc6 DK |
1935 | * entered abort state during the update. |
1936 | * | |
1937 | * Recursive transactions are not disturbed by journal abort until the | |
f7f4bccb | 1938 | * final jbd2_journal_stop, which will receive the -EIO error. |
470decc6 | 1939 | * |
f7f4bccb | 1940 | * Finally, the jbd2_journal_abort call allows the caller to supply an errno |
470decc6 DK |
1941 | * which will be recorded (if possible) in the journal superblock. This |
1942 | * allows a client to record failure conditions in the middle of a | |
1943 | * transaction without having to complete the transaction to record the | |
1944 | * failure to disk. ext3_error, for example, now uses this | |
1945 | * functionality. | |
1946 | * | |
1947 | * Errors which originate from within the journaling layer will NOT | |
1948 | * supply an errno; a null errno implies that absolutely no further | |
1949 | * writes are done to the journal (unless there are any already in | |
1950 | * progress). | |
1951 | * | |
1952 | */ | |
1953 | ||
f7f4bccb | 1954 | void jbd2_journal_abort(journal_t *journal, int errno) |
470decc6 DK |
1955 | { |
1956 | __journal_abort_soft(journal, errno); | |
1957 | } | |
1958 | ||
1959 | /** | |
f7f4bccb | 1960 | * int jbd2_journal_errno () - returns the journal's error state. |
470decc6 DK |
1961 | * @journal: journal to examine. |
1962 | * | |
bfcd3555 | 1963 | * This is the errno number set with jbd2_journal_abort(), the last |
470decc6 DK |
1964 | * time the journal was mounted - if the journal was stopped |
1965 | * without calling abort this will be 0. | |
1966 | * | |
1967 | * If the journal has been aborted on this mount time -EROFS will | |
1968 | * be returned. | |
1969 | */ | |
f7f4bccb | 1970 | int jbd2_journal_errno(journal_t *journal) |
470decc6 DK |
1971 | { |
1972 | int err; | |
1973 | ||
a931da6a | 1974 | read_lock(&journal->j_state_lock); |
f7f4bccb | 1975 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1976 | err = -EROFS; |
1977 | else | |
1978 | err = journal->j_errno; | |
a931da6a | 1979 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
1980 | return err; |
1981 | } | |
1982 | ||
1983 | /** | |
f7f4bccb | 1984 | * int jbd2_journal_clear_err () - clears the journal's error state |
470decc6 DK |
1985 | * @journal: journal to act on. |
1986 | * | |
bfcd3555 | 1987 | * An error must be cleared or acked to take a FS out of readonly |
470decc6 DK |
1988 | * mode. |
1989 | */ | |
f7f4bccb | 1990 | int jbd2_journal_clear_err(journal_t *journal) |
470decc6 DK |
1991 | { |
1992 | int err = 0; | |
1993 | ||
a931da6a | 1994 | write_lock(&journal->j_state_lock); |
f7f4bccb | 1995 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1996 | err = -EROFS; |
1997 | else | |
1998 | journal->j_errno = 0; | |
a931da6a | 1999 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2000 | return err; |
2001 | } | |
2002 | ||
2003 | /** | |
f7f4bccb | 2004 | * void jbd2_journal_ack_err() - Ack journal err. |
470decc6 DK |
2005 | * @journal: journal to act on. |
2006 | * | |
bfcd3555 | 2007 | * An error must be cleared or acked to take a FS out of readonly |
470decc6 DK |
2008 | * mode. |
2009 | */ | |
f7f4bccb | 2010 | void jbd2_journal_ack_err(journal_t *journal) |
470decc6 | 2011 | { |
a931da6a | 2012 | write_lock(&journal->j_state_lock); |
470decc6 | 2013 | if (journal->j_errno) |
f7f4bccb | 2014 | journal->j_flags |= JBD2_ACK_ERR; |
a931da6a | 2015 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2016 | } |
2017 | ||
f7f4bccb | 2018 | int jbd2_journal_blocks_per_page(struct inode *inode) |
470decc6 DK |
2019 | { |
2020 | return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
2021 | } | |
2022 | ||
b517bea1 ZB |
2023 | /* |
2024 | * helper functions to deal with 32 or 64bit block numbers. | |
2025 | */ | |
2026 | size_t journal_tag_bytes(journal_t *journal) | |
2027 | { | |
2028 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) | |
cd02ff0b | 2029 | return JBD2_TAG_SIZE64; |
b517bea1 | 2030 | else |
cd02ff0b | 2031 | return JBD2_TAG_SIZE32; |
b517bea1 ZB |
2032 | } |
2033 | ||
d2eecb03 TT |
2034 | /* |
2035 | * JBD memory management | |
2036 | * | |
2037 | * These functions are used to allocate block-sized chunks of memory | |
2038 | * used for making copies of buffer_head data. Very often it will be | |
2039 | * page-sized chunks of data, but sometimes it will be in | |
2040 | * sub-page-size chunks. (For example, 16k pages on Power systems | |
2041 | * with a 4k block file system.) For blocks smaller than a page, we | |
2042 | * use a SLAB allocator. There are slab caches for each block size, | |
2043 | * which are allocated at mount time, if necessary, and we only free | |
2044 | * (all of) the slab caches when/if the jbd2 module is unloaded. For | |
2045 | * this reason we don't need to a mutex to protect access to | |
2046 | * jbd2_slab[] allocating or releasing memory; only in | |
2047 | * jbd2_journal_create_slab(). | |
2048 | */ | |
2049 | #define JBD2_MAX_SLABS 8 | |
2050 | static struct kmem_cache *jbd2_slab[JBD2_MAX_SLABS]; | |
d2eecb03 TT |
2051 | |
2052 | static const char *jbd2_slab_names[JBD2_MAX_SLABS] = { | |
2053 | "jbd2_1k", "jbd2_2k", "jbd2_4k", "jbd2_8k", | |
2054 | "jbd2_16k", "jbd2_32k", "jbd2_64k", "jbd2_128k" | |
2055 | }; | |
2056 | ||
2057 | ||
2058 | static void jbd2_journal_destroy_slabs(void) | |
2059 | { | |
2060 | int i; | |
2061 | ||
2062 | for (i = 0; i < JBD2_MAX_SLABS; i++) { | |
2063 | if (jbd2_slab[i]) | |
2064 | kmem_cache_destroy(jbd2_slab[i]); | |
2065 | jbd2_slab[i] = NULL; | |
2066 | } | |
2067 | } | |
2068 | ||
2069 | static int jbd2_journal_create_slab(size_t size) | |
2070 | { | |
51dfacde | 2071 | static DEFINE_MUTEX(jbd2_slab_create_mutex); |
d2eecb03 TT |
2072 | int i = order_base_2(size) - 10; |
2073 | size_t slab_size; | |
2074 | ||
2075 | if (size == PAGE_SIZE) | |
2076 | return 0; | |
2077 | ||
2078 | if (i >= JBD2_MAX_SLABS) | |
2079 | return -EINVAL; | |
2080 | ||
2081 | if (unlikely(i < 0)) | |
2082 | i = 0; | |
51dfacde | 2083 | mutex_lock(&jbd2_slab_create_mutex); |
d2eecb03 | 2084 | if (jbd2_slab[i]) { |
51dfacde | 2085 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2086 | return 0; /* Already created */ |
2087 | } | |
2088 | ||
2089 | slab_size = 1 << (i+10); | |
2090 | jbd2_slab[i] = kmem_cache_create(jbd2_slab_names[i], slab_size, | |
2091 | slab_size, 0, NULL); | |
51dfacde | 2092 | mutex_unlock(&jbd2_slab_create_mutex); |
d2eecb03 TT |
2093 | if (!jbd2_slab[i]) { |
2094 | printk(KERN_EMERG "JBD2: no memory for jbd2_slab cache\n"); | |
2095 | return -ENOMEM; | |
2096 | } | |
2097 | return 0; | |
2098 | } | |
2099 | ||
2100 | static struct kmem_cache *get_slab(size_t size) | |
2101 | { | |
2102 | int i = order_base_2(size) - 10; | |
2103 | ||
2104 | BUG_ON(i >= JBD2_MAX_SLABS); | |
2105 | if (unlikely(i < 0)) | |
2106 | i = 0; | |
8ac97b74 | 2107 | BUG_ON(jbd2_slab[i] == NULL); |
d2eecb03 TT |
2108 | return jbd2_slab[i]; |
2109 | } | |
2110 | ||
2111 | void *jbd2_alloc(size_t size, gfp_t flags) | |
2112 | { | |
2113 | void *ptr; | |
2114 | ||
2115 | BUG_ON(size & (size-1)); /* Must be a power of 2 */ | |
2116 | ||
2117 | flags |= __GFP_REPEAT; | |
2118 | if (size == PAGE_SIZE) | |
2119 | ptr = (void *)__get_free_pages(flags, 0); | |
2120 | else if (size > PAGE_SIZE) { | |
2121 | int order = get_order(size); | |
2122 | ||
2123 | if (order < 3) | |
2124 | ptr = (void *)__get_free_pages(flags, order); | |
2125 | else | |
2126 | ptr = vmalloc(size); | |
2127 | } else | |
2128 | ptr = kmem_cache_alloc(get_slab(size), flags); | |
2129 | ||
2130 | /* Check alignment; SLUB has gotten this wrong in the past, | |
2131 | * and this can lead to user data corruption! */ | |
2132 | BUG_ON(((unsigned long) ptr) & (size-1)); | |
2133 | ||
2134 | return ptr; | |
2135 | } | |
2136 | ||
2137 | void jbd2_free(void *ptr, size_t size) | |
2138 | { | |
2139 | if (size == PAGE_SIZE) { | |
2140 | free_pages((unsigned long)ptr, 0); | |
2141 | return; | |
2142 | } | |
2143 | if (size > PAGE_SIZE) { | |
2144 | int order = get_order(size); | |
2145 | ||
2146 | if (order < 3) | |
2147 | free_pages((unsigned long)ptr, order); | |
2148 | else | |
2149 | vfree(ptr); | |
2150 | return; | |
2151 | } | |
2152 | kmem_cache_free(get_slab(size), ptr); | |
2153 | }; | |
2154 | ||
470decc6 DK |
2155 | /* |
2156 | * Journal_head storage management | |
2157 | */ | |
e18b890b | 2158 | static struct kmem_cache *jbd2_journal_head_cache; |
e23291b9 | 2159 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2160 | static atomic_t nr_journal_heads = ATOMIC_INIT(0); |
2161 | #endif | |
2162 | ||
4185a2ac | 2163 | static int jbd2_journal_init_journal_head_cache(void) |
470decc6 DK |
2164 | { |
2165 | int retval; | |
2166 | ||
1076d17a | 2167 | J_ASSERT(jbd2_journal_head_cache == NULL); |
a920e941 | 2168 | jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", |
470decc6 DK |
2169 | sizeof(struct journal_head), |
2170 | 0, /* offset */ | |
77160957 | 2171 | SLAB_TEMPORARY, /* flags */ |
20c2df83 | 2172 | NULL); /* ctor */ |
470decc6 | 2173 | retval = 0; |
1076d17a | 2174 | if (!jbd2_journal_head_cache) { |
470decc6 | 2175 | retval = -ENOMEM; |
f2a44523 | 2176 | printk(KERN_EMERG "JBD2: no memory for journal_head cache\n"); |
470decc6 DK |
2177 | } |
2178 | return retval; | |
2179 | } | |
2180 | ||
4185a2ac | 2181 | static void jbd2_journal_destroy_journal_head_cache(void) |
470decc6 | 2182 | { |
8a9362eb DG |
2183 | if (jbd2_journal_head_cache) { |
2184 | kmem_cache_destroy(jbd2_journal_head_cache); | |
2185 | jbd2_journal_head_cache = NULL; | |
2186 | } | |
470decc6 DK |
2187 | } |
2188 | ||
2189 | /* | |
2190 | * journal_head splicing and dicing | |
2191 | */ | |
2192 | static struct journal_head *journal_alloc_journal_head(void) | |
2193 | { | |
2194 | struct journal_head *ret; | |
470decc6 | 2195 | |
e23291b9 | 2196 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2197 | atomic_inc(&nr_journal_heads); |
2198 | #endif | |
f7f4bccb | 2199 | ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS); |
1076d17a | 2200 | if (!ret) { |
470decc6 | 2201 | jbd_debug(1, "out of memory for journal_head\n"); |
670be5a7 | 2202 | pr_notice_ratelimited("ENOMEM in %s, retrying.\n", __func__); |
1076d17a | 2203 | while (!ret) { |
470decc6 | 2204 | yield(); |
f7f4bccb | 2205 | ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS); |
470decc6 DK |
2206 | } |
2207 | } | |
2208 | return ret; | |
2209 | } | |
2210 | ||
2211 | static void journal_free_journal_head(struct journal_head *jh) | |
2212 | { | |
e23291b9 | 2213 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 | 2214 | atomic_dec(&nr_journal_heads); |
cd02ff0b | 2215 | memset(jh, JBD2_POISON_FREE, sizeof(*jh)); |
470decc6 | 2216 | #endif |
f7f4bccb | 2217 | kmem_cache_free(jbd2_journal_head_cache, jh); |
470decc6 DK |
2218 | } |
2219 | ||
2220 | /* | |
2221 | * A journal_head is attached to a buffer_head whenever JBD has an | |
2222 | * interest in the buffer. | |
2223 | * | |
2224 | * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit | |
2225 | * is set. This bit is tested in core kernel code where we need to take | |
2226 | * JBD-specific actions. Testing the zeroness of ->b_private is not reliable | |
2227 | * there. | |
2228 | * | |
2229 | * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. | |
2230 | * | |
2231 | * When a buffer has its BH_JBD bit set it is immune from being released by | |
2232 | * core kernel code, mainly via ->b_count. | |
2233 | * | |
de1b7941 JK |
2234 | * A journal_head is detached from its buffer_head when the journal_head's |
2235 | * b_jcount reaches zero. Running transaction (b_transaction) and checkpoint | |
2236 | * transaction (b_cp_transaction) hold their references to b_jcount. | |
470decc6 DK |
2237 | * |
2238 | * Various places in the kernel want to attach a journal_head to a buffer_head | |
2239 | * _before_ attaching the journal_head to a transaction. To protect the | |
f7f4bccb | 2240 | * journal_head in this situation, jbd2_journal_add_journal_head elevates the |
470decc6 | 2241 | * journal_head's b_jcount refcount by one. The caller must call |
f7f4bccb | 2242 | * jbd2_journal_put_journal_head() to undo this. |
470decc6 DK |
2243 | * |
2244 | * So the typical usage would be: | |
2245 | * | |
2246 | * (Attach a journal_head if needed. Increments b_jcount) | |
f7f4bccb | 2247 | * struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 | 2248 | * ... |
de1b7941 JK |
2249 | * (Get another reference for transaction) |
2250 | * jbd2_journal_grab_journal_head(bh); | |
470decc6 | 2251 | * jh->b_transaction = xxx; |
de1b7941 | 2252 | * (Put original reference) |
f7f4bccb | 2253 | * jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2254 | */ |
2255 | ||
2256 | /* | |
2257 | * Give a buffer_head a journal_head. | |
2258 | * | |
470decc6 DK |
2259 | * May sleep. |
2260 | */ | |
f7f4bccb | 2261 | struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) |
470decc6 DK |
2262 | { |
2263 | struct journal_head *jh; | |
2264 | struct journal_head *new_jh = NULL; | |
2265 | ||
2266 | repeat: | |
2267 | if (!buffer_jbd(bh)) { | |
2268 | new_jh = journal_alloc_journal_head(); | |
2269 | memset(new_jh, 0, sizeof(*new_jh)); | |
2270 | } | |
2271 | ||
2272 | jbd_lock_bh_journal_head(bh); | |
2273 | if (buffer_jbd(bh)) { | |
2274 | jh = bh2jh(bh); | |
2275 | } else { | |
2276 | J_ASSERT_BH(bh, | |
2277 | (atomic_read(&bh->b_count) > 0) || | |
2278 | (bh->b_page && bh->b_page->mapping)); | |
2279 | ||
2280 | if (!new_jh) { | |
2281 | jbd_unlock_bh_journal_head(bh); | |
2282 | goto repeat; | |
2283 | } | |
2284 | ||
2285 | jh = new_jh; | |
2286 | new_jh = NULL; /* We consumed it */ | |
2287 | set_buffer_jbd(bh); | |
2288 | bh->b_private = jh; | |
2289 | jh->b_bh = bh; | |
2290 | get_bh(bh); | |
2291 | BUFFER_TRACE(bh, "added journal_head"); | |
2292 | } | |
2293 | jh->b_jcount++; | |
2294 | jbd_unlock_bh_journal_head(bh); | |
2295 | if (new_jh) | |
2296 | journal_free_journal_head(new_jh); | |
2297 | return bh->b_private; | |
2298 | } | |
2299 | ||
2300 | /* | |
2301 | * Grab a ref against this buffer_head's journal_head. If it ended up not | |
2302 | * having a journal_head, return NULL | |
2303 | */ | |
f7f4bccb | 2304 | struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) |
470decc6 DK |
2305 | { |
2306 | struct journal_head *jh = NULL; | |
2307 | ||
2308 | jbd_lock_bh_journal_head(bh); | |
2309 | if (buffer_jbd(bh)) { | |
2310 | jh = bh2jh(bh); | |
2311 | jh->b_jcount++; | |
2312 | } | |
2313 | jbd_unlock_bh_journal_head(bh); | |
2314 | return jh; | |
2315 | } | |
2316 | ||
2317 | static void __journal_remove_journal_head(struct buffer_head *bh) | |
2318 | { | |
2319 | struct journal_head *jh = bh2jh(bh); | |
2320 | ||
2321 | J_ASSERT_JH(jh, jh->b_jcount >= 0); | |
de1b7941 JK |
2322 | J_ASSERT_JH(jh, jh->b_transaction == NULL); |
2323 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
2324 | J_ASSERT_JH(jh, jh->b_cp_transaction == NULL); | |
2325 | J_ASSERT_JH(jh, jh->b_jlist == BJ_None); | |
2326 | J_ASSERT_BH(bh, buffer_jbd(bh)); | |
2327 | J_ASSERT_BH(bh, jh2bh(jh) == bh); | |
2328 | BUFFER_TRACE(bh, "remove journal_head"); | |
2329 | if (jh->b_frozen_data) { | |
2330 | printk(KERN_WARNING "%s: freeing b_frozen_data\n", __func__); | |
2331 | jbd2_free(jh->b_frozen_data, bh->b_size); | |
470decc6 | 2332 | } |
de1b7941 JK |
2333 | if (jh->b_committed_data) { |
2334 | printk(KERN_WARNING "%s: freeing b_committed_data\n", __func__); | |
2335 | jbd2_free(jh->b_committed_data, bh->b_size); | |
2336 | } | |
2337 | bh->b_private = NULL; | |
2338 | jh->b_bh = NULL; /* debug, really */ | |
2339 | clear_buffer_jbd(bh); | |
2340 | journal_free_journal_head(jh); | |
470decc6 DK |
2341 | } |
2342 | ||
2343 | /* | |
de1b7941 | 2344 | * Drop a reference on the passed journal_head. If it fell to zero then |
470decc6 DK |
2345 | * release the journal_head from the buffer_head. |
2346 | */ | |
f7f4bccb | 2347 | void jbd2_journal_put_journal_head(struct journal_head *jh) |
470decc6 DK |
2348 | { |
2349 | struct buffer_head *bh = jh2bh(jh); | |
2350 | ||
2351 | jbd_lock_bh_journal_head(bh); | |
2352 | J_ASSERT_JH(jh, jh->b_jcount > 0); | |
2353 | --jh->b_jcount; | |
de1b7941 | 2354 | if (!jh->b_jcount) { |
470decc6 | 2355 | __journal_remove_journal_head(bh); |
de1b7941 | 2356 | jbd_unlock_bh_journal_head(bh); |
470decc6 | 2357 | __brelse(bh); |
de1b7941 JK |
2358 | } else |
2359 | jbd_unlock_bh_journal_head(bh); | |
470decc6 DK |
2360 | } |
2361 | ||
c851ed54 JK |
2362 | /* |
2363 | * Initialize jbd inode head | |
2364 | */ | |
2365 | void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) | |
2366 | { | |
2367 | jinode->i_transaction = NULL; | |
2368 | jinode->i_next_transaction = NULL; | |
2369 | jinode->i_vfs_inode = inode; | |
2370 | jinode->i_flags = 0; | |
2371 | INIT_LIST_HEAD(&jinode->i_list); | |
2372 | } | |
2373 | ||
2374 | /* | |
2375 | * Function to be called before we start removing inode from memory (i.e., | |
2376 | * clear_inode() is a fine place to be called from). It removes inode from | |
2377 | * transaction's lists. | |
2378 | */ | |
2379 | void jbd2_journal_release_jbd_inode(journal_t *journal, | |
2380 | struct jbd2_inode *jinode) | |
2381 | { | |
c851ed54 JK |
2382 | if (!journal) |
2383 | return; | |
2384 | restart: | |
2385 | spin_lock(&journal->j_list_lock); | |
2386 | /* Is commit writing out inode - we have to wait */ | |
39e3ac25 | 2387 | if (test_bit(__JI_COMMIT_RUNNING, &jinode->i_flags)) { |
c851ed54 JK |
2388 | wait_queue_head_t *wq; |
2389 | DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); | |
2390 | wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); | |
2391 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
2392 | spin_unlock(&journal->j_list_lock); | |
2393 | schedule(); | |
2394 | finish_wait(wq, &wait.wait); | |
2395 | goto restart; | |
2396 | } | |
2397 | ||
c851ed54 JK |
2398 | if (jinode->i_transaction) { |
2399 | list_del(&jinode->i_list); | |
2400 | jinode->i_transaction = NULL; | |
2401 | } | |
2402 | spin_unlock(&journal->j_list_lock); | |
2403 | } | |
2404 | ||
470decc6 | 2405 | /* |
0f49d5d0 | 2406 | * debugfs tunables |
470decc6 | 2407 | */ |
6f38c74f JS |
2408 | #ifdef CONFIG_JBD2_DEBUG |
2409 | u8 jbd2_journal_enable_debug __read_mostly; | |
f7f4bccb | 2410 | EXPORT_SYMBOL(jbd2_journal_enable_debug); |
470decc6 | 2411 | |
0f49d5d0 | 2412 | #define JBD2_DEBUG_NAME "jbd2-debug" |
470decc6 | 2413 | |
6f38c74f JS |
2414 | static struct dentry *jbd2_debugfs_dir; |
2415 | static struct dentry *jbd2_debug; | |
470decc6 | 2416 | |
0f49d5d0 JS |
2417 | static void __init jbd2_create_debugfs_entry(void) |
2418 | { | |
2419 | jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL); | |
2420 | if (jbd2_debugfs_dir) | |
765f8361 YK |
2421 | jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, |
2422 | S_IRUGO | S_IWUSR, | |
0f49d5d0 JS |
2423 | jbd2_debugfs_dir, |
2424 | &jbd2_journal_enable_debug); | |
470decc6 DK |
2425 | } |
2426 | ||
0f49d5d0 | 2427 | static void __exit jbd2_remove_debugfs_entry(void) |
470decc6 | 2428 | { |
6f38c74f JS |
2429 | debugfs_remove(jbd2_debug); |
2430 | debugfs_remove(jbd2_debugfs_dir); | |
470decc6 DK |
2431 | } |
2432 | ||
0f49d5d0 | 2433 | #else |
470decc6 | 2434 | |
0f49d5d0 | 2435 | static void __init jbd2_create_debugfs_entry(void) |
470decc6 | 2436 | { |
470decc6 DK |
2437 | } |
2438 | ||
0f49d5d0 | 2439 | static void __exit jbd2_remove_debugfs_entry(void) |
470decc6 | 2440 | { |
470decc6 DK |
2441 | } |
2442 | ||
470decc6 DK |
2443 | #endif |
2444 | ||
8e85fb3f JL |
2445 | #ifdef CONFIG_PROC_FS |
2446 | ||
2447 | #define JBD2_STATS_PROC_NAME "fs/jbd2" | |
2448 | ||
2449 | static void __init jbd2_create_jbd_stats_proc_entry(void) | |
2450 | { | |
2451 | proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); | |
2452 | } | |
2453 | ||
2454 | static void __exit jbd2_remove_jbd_stats_proc_entry(void) | |
2455 | { | |
2456 | if (proc_jbd2_stats) | |
2457 | remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); | |
2458 | } | |
2459 | ||
2460 | #else | |
2461 | ||
2462 | #define jbd2_create_jbd_stats_proc_entry() do {} while (0) | |
2463 | #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) | |
2464 | ||
2465 | #endif | |
2466 | ||
8aefcd55 | 2467 | struct kmem_cache *jbd2_handle_cache, *jbd2_inode_cache; |
470decc6 | 2468 | |
4185a2ac | 2469 | static int __init jbd2_journal_init_handle_cache(void) |
470decc6 | 2470 | { |
8aefcd55 | 2471 | jbd2_handle_cache = KMEM_CACHE(jbd2_journal_handle, SLAB_TEMPORARY); |
f7f4bccb | 2472 | if (jbd2_handle_cache == NULL) { |
8aefcd55 TT |
2473 | printk(KERN_EMERG "JBD2: failed to create handle cache\n"); |
2474 | return -ENOMEM; | |
2475 | } | |
2476 | jbd2_inode_cache = KMEM_CACHE(jbd2_inode, 0); | |
2477 | if (jbd2_inode_cache == NULL) { | |
2478 | printk(KERN_EMERG "JBD2: failed to create inode cache\n"); | |
2479 | kmem_cache_destroy(jbd2_handle_cache); | |
470decc6 DK |
2480 | return -ENOMEM; |
2481 | } | |
2482 | return 0; | |
2483 | } | |
2484 | ||
f7f4bccb | 2485 | static void jbd2_journal_destroy_handle_cache(void) |
470decc6 | 2486 | { |
f7f4bccb MC |
2487 | if (jbd2_handle_cache) |
2488 | kmem_cache_destroy(jbd2_handle_cache); | |
8aefcd55 TT |
2489 | if (jbd2_inode_cache) |
2490 | kmem_cache_destroy(jbd2_inode_cache); | |
2491 | ||
470decc6 DK |
2492 | } |
2493 | ||
2494 | /* | |
2495 | * Module startup and shutdown | |
2496 | */ | |
2497 | ||
2498 | static int __init journal_init_caches(void) | |
2499 | { | |
2500 | int ret; | |
2501 | ||
f7f4bccb | 2502 | ret = jbd2_journal_init_revoke_caches(); |
470decc6 | 2503 | if (ret == 0) |
4185a2ac | 2504 | ret = jbd2_journal_init_journal_head_cache(); |
470decc6 | 2505 | if (ret == 0) |
4185a2ac | 2506 | ret = jbd2_journal_init_handle_cache(); |
470decc6 | 2507 | if (ret == 0) |
0c2022ec | 2508 | ret = jbd2_journal_init_transaction_cache(); |
470decc6 DK |
2509 | return ret; |
2510 | } | |
2511 | ||
f7f4bccb | 2512 | static void jbd2_journal_destroy_caches(void) |
470decc6 | 2513 | { |
f7f4bccb | 2514 | jbd2_journal_destroy_revoke_caches(); |
4185a2ac | 2515 | jbd2_journal_destroy_journal_head_cache(); |
f7f4bccb | 2516 | jbd2_journal_destroy_handle_cache(); |
0c2022ec | 2517 | jbd2_journal_destroy_transaction_cache(); |
d2eecb03 | 2518 | jbd2_journal_destroy_slabs(); |
470decc6 DK |
2519 | } |
2520 | ||
2521 | static int __init journal_init(void) | |
2522 | { | |
2523 | int ret; | |
2524 | ||
2525 | BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); | |
2526 | ||
2527 | ret = journal_init_caches(); | |
620de4e1 DG |
2528 | if (ret == 0) { |
2529 | jbd2_create_debugfs_entry(); | |
2530 | jbd2_create_jbd_stats_proc_entry(); | |
2531 | } else { | |
f7f4bccb | 2532 | jbd2_journal_destroy_caches(); |
620de4e1 | 2533 | } |
470decc6 DK |
2534 | return ret; |
2535 | } | |
2536 | ||
2537 | static void __exit journal_exit(void) | |
2538 | { | |
e23291b9 | 2539 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2540 | int n = atomic_read(&nr_journal_heads); |
2541 | if (n) | |
f2a44523 | 2542 | printk(KERN_EMERG "JBD2: leaked %d journal_heads!\n", n); |
470decc6 | 2543 | #endif |
0f49d5d0 | 2544 | jbd2_remove_debugfs_entry(); |
8e85fb3f | 2545 | jbd2_remove_jbd_stats_proc_entry(); |
f7f4bccb | 2546 | jbd2_journal_destroy_caches(); |
470decc6 DK |
2547 | } |
2548 | ||
2549 | MODULE_LICENSE("GPL"); | |
2550 | module_init(journal_init); | |
2551 | module_exit(journal_exit); | |
2552 |