Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/journal.c | |
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> | |
28 | #include <linux/jbd.h> | |
29 | #include <linux/errno.h> | |
30 | #include <linux/slab.h> | |
31 | #include <linux/smp_lock.h> | |
32 | #include <linux/init.h> | |
33 | #include <linux/mm.h> | |
34 | #include <linux/suspend.h> | |
35 | #include <linux/pagemap.h> | |
36 | #include <asm/uaccess.h> | |
37 | #include <asm/page.h> | |
38 | #include <linux/proc_fs.h> | |
39 | ||
40 | EXPORT_SYMBOL(journal_start); | |
41 | EXPORT_SYMBOL(journal_restart); | |
42 | EXPORT_SYMBOL(journal_extend); | |
43 | EXPORT_SYMBOL(journal_stop); | |
44 | EXPORT_SYMBOL(journal_lock_updates); | |
45 | EXPORT_SYMBOL(journal_unlock_updates); | |
46 | EXPORT_SYMBOL(journal_get_write_access); | |
47 | EXPORT_SYMBOL(journal_get_create_access); | |
48 | EXPORT_SYMBOL(journal_get_undo_access); | |
49 | EXPORT_SYMBOL(journal_dirty_data); | |
50 | EXPORT_SYMBOL(journal_dirty_metadata); | |
51 | EXPORT_SYMBOL(journal_release_buffer); | |
52 | EXPORT_SYMBOL(journal_forget); | |
53 | #if 0 | |
54 | EXPORT_SYMBOL(journal_sync_buffer); | |
55 | #endif | |
56 | EXPORT_SYMBOL(journal_flush); | |
57 | EXPORT_SYMBOL(journal_revoke); | |
58 | ||
59 | EXPORT_SYMBOL(journal_init_dev); | |
60 | EXPORT_SYMBOL(journal_init_inode); | |
61 | EXPORT_SYMBOL(journal_update_format); | |
62 | EXPORT_SYMBOL(journal_check_used_features); | |
63 | EXPORT_SYMBOL(journal_check_available_features); | |
64 | EXPORT_SYMBOL(journal_set_features); | |
65 | EXPORT_SYMBOL(journal_create); | |
66 | EXPORT_SYMBOL(journal_load); | |
67 | EXPORT_SYMBOL(journal_destroy); | |
1da177e4 LT |
68 | EXPORT_SYMBOL(journal_update_superblock); |
69 | EXPORT_SYMBOL(journal_abort); | |
70 | EXPORT_SYMBOL(journal_errno); | |
71 | EXPORT_SYMBOL(journal_ack_err); | |
72 | EXPORT_SYMBOL(journal_clear_err); | |
73 | EXPORT_SYMBOL(log_wait_commit); | |
74 | EXPORT_SYMBOL(journal_start_commit); | |
75 | EXPORT_SYMBOL(journal_force_commit_nested); | |
76 | EXPORT_SYMBOL(journal_wipe); | |
77 | EXPORT_SYMBOL(journal_blocks_per_page); | |
78 | EXPORT_SYMBOL(journal_invalidatepage); | |
79 | EXPORT_SYMBOL(journal_try_to_free_buffers); | |
80 | EXPORT_SYMBOL(journal_force_commit); | |
81 | ||
82 | static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *); | |
022a4a7b | 83 | static void __journal_abort_soft (journal_t *journal, int errno); |
1da177e4 LT |
84 | |
85 | /* | |
86 | * Helper function used to manage commit timeouts | |
87 | */ | |
88 | ||
89 | static void commit_timeout(unsigned long __data) | |
90 | { | |
91 | struct task_struct * p = (struct task_struct *) __data; | |
92 | ||
93 | wake_up_process(p); | |
94 | } | |
95 | ||
1da177e4 LT |
96 | /* |
97 | * kjournald: The main thread function used to manage a logging device | |
98 | * journal. | |
99 | * | |
100 | * This kernel thread is responsible for two things: | |
101 | * | |
102 | * 1) COMMIT: Every so often we need to commit the current state of the | |
103 | * filesystem to disk. The journal thread is responsible for writing | |
104 | * all of the metadata buffers to disk. | |
105 | * | |
106 | * 2) CHECKPOINT: We cannot reuse a used section of the log file until all | |
107 | * of the data in that part of the log has been rewritten elsewhere on | |
108 | * the disk. Flushing these old buffers to reclaim space in the log is | |
109 | * known as checkpointing, and this thread is responsible for that job. | |
110 | */ | |
111 | ||
022a4a7b | 112 | static int kjournald(void *arg) |
1da177e4 LT |
113 | { |
114 | journal_t *journal = (journal_t *) arg; | |
115 | transaction_t *transaction; | |
116 | struct timer_list timer; | |
117 | ||
1da177e4 LT |
118 | daemonize("kjournald"); |
119 | ||
120 | /* Set up an interval timer which can be used to trigger a | |
121 | commit wakeup after the commit interval expires */ | |
122 | init_timer(&timer); | |
123 | timer.data = (unsigned long) current; | |
124 | timer.function = commit_timeout; | |
125 | journal->j_commit_timer = &timer; | |
126 | ||
127 | /* Record that the journal thread is running */ | |
128 | journal->j_task = current; | |
129 | wake_up(&journal->j_wait_done_commit); | |
130 | ||
131 | printk(KERN_INFO "kjournald starting. Commit interval %ld seconds\n", | |
132 | journal->j_commit_interval / HZ); | |
133 | ||
134 | /* | |
135 | * And now, wait forever for commit wakeup events. | |
136 | */ | |
137 | spin_lock(&journal->j_state_lock); | |
138 | ||
139 | loop: | |
140 | if (journal->j_flags & JFS_UNMOUNT) | |
141 | goto end_loop; | |
142 | ||
143 | jbd_debug(1, "commit_sequence=%d, commit_request=%d\n", | |
144 | journal->j_commit_sequence, journal->j_commit_request); | |
145 | ||
146 | if (journal->j_commit_sequence != journal->j_commit_request) { | |
147 | jbd_debug(1, "OK, requests differ\n"); | |
148 | spin_unlock(&journal->j_state_lock); | |
149 | del_timer_sync(journal->j_commit_timer); | |
150 | journal_commit_transaction(journal); | |
151 | spin_lock(&journal->j_state_lock); | |
152 | goto loop; | |
153 | } | |
154 | ||
155 | wake_up(&journal->j_wait_done_commit); | |
3e1d1d28 | 156 | if (freezing(current)) { |
1da177e4 LT |
157 | /* |
158 | * The simpler the better. Flushing journal isn't a | |
159 | * good idea, because that depends on threads that may | |
160 | * be already stopped. | |
161 | */ | |
162 | jbd_debug(1, "Now suspending kjournald\n"); | |
163 | spin_unlock(&journal->j_state_lock); | |
3e1d1d28 | 164 | refrigerator(); |
1da177e4 LT |
165 | spin_lock(&journal->j_state_lock); |
166 | } else { | |
167 | /* | |
168 | * We assume on resume that commits are already there, | |
169 | * so we don't sleep | |
170 | */ | |
171 | DEFINE_WAIT(wait); | |
172 | int should_sleep = 1; | |
173 | ||
174 | prepare_to_wait(&journal->j_wait_commit, &wait, | |
175 | TASK_INTERRUPTIBLE); | |
176 | if (journal->j_commit_sequence != journal->j_commit_request) | |
177 | should_sleep = 0; | |
178 | transaction = journal->j_running_transaction; | |
179 | if (transaction && time_after_eq(jiffies, | |
180 | transaction->t_expires)) | |
181 | should_sleep = 0; | |
182 | if (should_sleep) { | |
183 | spin_unlock(&journal->j_state_lock); | |
184 | schedule(); | |
185 | spin_lock(&journal->j_state_lock); | |
186 | } | |
187 | finish_wait(&journal->j_wait_commit, &wait); | |
188 | } | |
189 | ||
190 | jbd_debug(1, "kjournald wakes\n"); | |
191 | ||
192 | /* | |
193 | * Were we woken up by a commit wakeup event? | |
194 | */ | |
195 | transaction = journal->j_running_transaction; | |
196 | if (transaction && time_after_eq(jiffies, transaction->t_expires)) { | |
197 | journal->j_commit_request = transaction->t_tid; | |
198 | jbd_debug(1, "woke because of timeout\n"); | |
199 | } | |
200 | goto loop; | |
201 | ||
202 | end_loop: | |
203 | spin_unlock(&journal->j_state_lock); | |
204 | del_timer_sync(journal->j_commit_timer); | |
205 | journal->j_task = NULL; | |
206 | wake_up(&journal->j_wait_done_commit); | |
207 | jbd_debug(1, "Journal thread exiting.\n"); | |
208 | return 0; | |
209 | } | |
210 | ||
211 | static void journal_start_thread(journal_t *journal) | |
212 | { | |
213 | kernel_thread(kjournald, journal, CLONE_VM|CLONE_FS|CLONE_FILES); | |
214 | wait_event(journal->j_wait_done_commit, journal->j_task != 0); | |
215 | } | |
216 | ||
217 | static void journal_kill_thread(journal_t *journal) | |
218 | { | |
219 | spin_lock(&journal->j_state_lock); | |
220 | journal->j_flags |= JFS_UNMOUNT; | |
221 | ||
222 | while (journal->j_task) { | |
223 | wake_up(&journal->j_wait_commit); | |
224 | spin_unlock(&journal->j_state_lock); | |
225 | wait_event(journal->j_wait_done_commit, journal->j_task == 0); | |
226 | spin_lock(&journal->j_state_lock); | |
227 | } | |
228 | spin_unlock(&journal->j_state_lock); | |
229 | } | |
230 | ||
231 | /* | |
232 | * journal_write_metadata_buffer: write a metadata buffer to the journal. | |
233 | * | |
234 | * Writes a metadata buffer to a given disk block. The actual IO is not | |
235 | * performed but a new buffer_head is constructed which labels the data | |
236 | * to be written with the correct destination disk block. | |
237 | * | |
238 | * Any magic-number escaping which needs to be done will cause a | |
239 | * copy-out here. If the buffer happens to start with the | |
240 | * JFS_MAGIC_NUMBER, then we can't write it to the log directly: the | |
241 | * magic number is only written to the log for descripter blocks. In | |
242 | * this case, we copy the data and replace the first word with 0, and we | |
243 | * return a result code which indicates that this buffer needs to be | |
244 | * marked as an escaped buffer in the corresponding log descriptor | |
245 | * block. The missing word can then be restored when the block is read | |
246 | * during recovery. | |
247 | * | |
248 | * If the source buffer has already been modified by a new transaction | |
249 | * since we took the last commit snapshot, we use the frozen copy of | |
250 | * that data for IO. If we end up using the existing buffer_head's data | |
251 | * for the write, then we *have* to lock the buffer to prevent anyone | |
252 | * else from using and possibly modifying it while the IO is in | |
253 | * progress. | |
254 | * | |
255 | * The function returns a pointer to the buffer_heads to be used for IO. | |
256 | * | |
257 | * We assume that the journal has already been locked in this function. | |
258 | * | |
259 | * Return value: | |
260 | * <0: Error | |
261 | * >=0: Finished OK | |
262 | * | |
263 | * On success: | |
264 | * Bit 0 set == escape performed on the data | |
265 | * Bit 1 set == buffer copy-out performed (kfree the data after IO) | |
266 | */ | |
267 | ||
268 | int journal_write_metadata_buffer(transaction_t *transaction, | |
269 | struct journal_head *jh_in, | |
270 | struct journal_head **jh_out, | |
271 | int blocknr) | |
272 | { | |
273 | int need_copy_out = 0; | |
274 | int done_copy_out = 0; | |
275 | int do_escape = 0; | |
276 | char *mapped_data; | |
277 | struct buffer_head *new_bh; | |
278 | struct journal_head *new_jh; | |
279 | struct page *new_page; | |
280 | unsigned int new_offset; | |
281 | struct buffer_head *bh_in = jh2bh(jh_in); | |
282 | ||
283 | /* | |
284 | * The buffer really shouldn't be locked: only the current committing | |
285 | * transaction is allowed to write it, so nobody else is allowed | |
286 | * to do any IO. | |
287 | * | |
288 | * akpm: except if we're journalling data, and write() output is | |
289 | * also part of a shared mapping, and another thread has | |
290 | * decided to launch a writepage() against this buffer. | |
291 | */ | |
292 | J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); | |
293 | ||
294 | new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); | |
295 | ||
296 | /* | |
297 | * If a new transaction has already done a buffer copy-out, then | |
298 | * we use that version of the data for the commit. | |
299 | */ | |
300 | jbd_lock_bh_state(bh_in); | |
301 | repeat: | |
302 | if (jh_in->b_frozen_data) { | |
303 | done_copy_out = 1; | |
304 | new_page = virt_to_page(jh_in->b_frozen_data); | |
305 | new_offset = offset_in_page(jh_in->b_frozen_data); | |
306 | } else { | |
307 | new_page = jh2bh(jh_in)->b_page; | |
308 | new_offset = offset_in_page(jh2bh(jh_in)->b_data); | |
309 | } | |
310 | ||
311 | mapped_data = kmap_atomic(new_page, KM_USER0); | |
312 | /* | |
313 | * Check for escaping | |
314 | */ | |
315 | if (*((__be32 *)(mapped_data + new_offset)) == | |
316 | cpu_to_be32(JFS_MAGIC_NUMBER)) { | |
317 | need_copy_out = 1; | |
318 | do_escape = 1; | |
319 | } | |
320 | kunmap_atomic(mapped_data, KM_USER0); | |
321 | ||
322 | /* | |
323 | * Do we need to do a data copy? | |
324 | */ | |
325 | if (need_copy_out && !done_copy_out) { | |
326 | char *tmp; | |
327 | ||
328 | jbd_unlock_bh_state(bh_in); | |
329 | tmp = jbd_rep_kmalloc(bh_in->b_size, GFP_NOFS); | |
330 | jbd_lock_bh_state(bh_in); | |
331 | if (jh_in->b_frozen_data) { | |
332 | kfree(tmp); | |
333 | goto repeat; | |
334 | } | |
335 | ||
336 | jh_in->b_frozen_data = tmp; | |
337 | mapped_data = kmap_atomic(new_page, KM_USER0); | |
338 | memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size); | |
339 | kunmap_atomic(mapped_data, KM_USER0); | |
340 | ||
341 | new_page = virt_to_page(tmp); | |
342 | new_offset = offset_in_page(tmp); | |
343 | done_copy_out = 1; | |
344 | } | |
345 | ||
346 | /* | |
347 | * Did we need to do an escaping? Now we've done all the | |
348 | * copying, we can finally do so. | |
349 | */ | |
350 | if (do_escape) { | |
351 | mapped_data = kmap_atomic(new_page, KM_USER0); | |
352 | *((unsigned int *)(mapped_data + new_offset)) = 0; | |
353 | kunmap_atomic(mapped_data, KM_USER0); | |
354 | } | |
355 | ||
356 | /* keep subsequent assertions sane */ | |
357 | new_bh->b_state = 0; | |
358 | init_buffer(new_bh, NULL, NULL); | |
359 | atomic_set(&new_bh->b_count, 1); | |
360 | jbd_unlock_bh_state(bh_in); | |
361 | ||
362 | new_jh = journal_add_journal_head(new_bh); /* This sleeps */ | |
363 | ||
364 | set_bh_page(new_bh, new_page, new_offset); | |
365 | new_jh->b_transaction = NULL; | |
366 | new_bh->b_size = jh2bh(jh_in)->b_size; | |
367 | new_bh->b_bdev = transaction->t_journal->j_dev; | |
368 | new_bh->b_blocknr = blocknr; | |
369 | set_buffer_mapped(new_bh); | |
370 | set_buffer_dirty(new_bh); | |
371 | ||
372 | *jh_out = new_jh; | |
373 | ||
374 | /* | |
375 | * The to-be-written buffer needs to get moved to the io queue, | |
376 | * and the original buffer whose contents we are shadowing or | |
377 | * copying is moved to the transaction's shadow queue. | |
378 | */ | |
379 | JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); | |
380 | journal_file_buffer(jh_in, transaction, BJ_Shadow); | |
381 | JBUFFER_TRACE(new_jh, "file as BJ_IO"); | |
382 | journal_file_buffer(new_jh, transaction, BJ_IO); | |
383 | ||
384 | return do_escape | (done_copy_out << 1); | |
385 | } | |
386 | ||
387 | /* | |
388 | * Allocation code for the journal file. Manage the space left in the | |
389 | * journal, so that we can begin checkpointing when appropriate. | |
390 | */ | |
391 | ||
392 | /* | |
393 | * __log_space_left: Return the number of free blocks left in the journal. | |
394 | * | |
395 | * Called with the journal already locked. | |
396 | * | |
397 | * Called under j_state_lock | |
398 | */ | |
399 | ||
400 | int __log_space_left(journal_t *journal) | |
401 | { | |
402 | int left = journal->j_free; | |
403 | ||
404 | assert_spin_locked(&journal->j_state_lock); | |
405 | ||
406 | /* | |
407 | * Be pessimistic here about the number of those free blocks which | |
408 | * might be required for log descriptor control blocks. | |
409 | */ | |
410 | ||
411 | #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */ | |
412 | ||
413 | left -= MIN_LOG_RESERVED_BLOCKS; | |
414 | ||
415 | if (left <= 0) | |
416 | return 0; | |
417 | left -= (left >> 3); | |
418 | return left; | |
419 | } | |
420 | ||
421 | /* | |
422 | * Called under j_state_lock. Returns true if a transaction was started. | |
423 | */ | |
424 | int __log_start_commit(journal_t *journal, tid_t target) | |
425 | { | |
426 | /* | |
427 | * Are we already doing a recent enough commit? | |
428 | */ | |
429 | if (!tid_geq(journal->j_commit_request, target)) { | |
430 | /* | |
431 | * We want a new commit: OK, mark the request and wakup the | |
432 | * commit thread. We do _not_ do the commit ourselves. | |
433 | */ | |
434 | ||
435 | journal->j_commit_request = target; | |
436 | jbd_debug(1, "JBD: requesting commit %d/%d\n", | |
437 | journal->j_commit_request, | |
438 | journal->j_commit_sequence); | |
439 | wake_up(&journal->j_wait_commit); | |
440 | return 1; | |
441 | } | |
442 | return 0; | |
443 | } | |
444 | ||
445 | int log_start_commit(journal_t *journal, tid_t tid) | |
446 | { | |
447 | int ret; | |
448 | ||
449 | spin_lock(&journal->j_state_lock); | |
450 | ret = __log_start_commit(journal, tid); | |
451 | spin_unlock(&journal->j_state_lock); | |
452 | return ret; | |
453 | } | |
454 | ||
455 | /* | |
456 | * Force and wait upon a commit if the calling process is not within | |
457 | * transaction. This is used for forcing out undo-protected data which contains | |
458 | * bitmaps, when the fs is running out of space. | |
459 | * | |
460 | * We can only force the running transaction if we don't have an active handle; | |
461 | * otherwise, we will deadlock. | |
462 | * | |
463 | * Returns true if a transaction was started. | |
464 | */ | |
465 | int journal_force_commit_nested(journal_t *journal) | |
466 | { | |
467 | transaction_t *transaction = NULL; | |
468 | tid_t tid; | |
469 | ||
470 | spin_lock(&journal->j_state_lock); | |
471 | if (journal->j_running_transaction && !current->journal_info) { | |
472 | transaction = journal->j_running_transaction; | |
473 | __log_start_commit(journal, transaction->t_tid); | |
474 | } else if (journal->j_committing_transaction) | |
475 | transaction = journal->j_committing_transaction; | |
476 | ||
477 | if (!transaction) { | |
478 | spin_unlock(&journal->j_state_lock); | |
479 | return 0; /* Nothing to retry */ | |
480 | } | |
481 | ||
482 | tid = transaction->t_tid; | |
483 | spin_unlock(&journal->j_state_lock); | |
484 | log_wait_commit(journal, tid); | |
485 | return 1; | |
486 | } | |
487 | ||
488 | /* | |
489 | * Start a commit of the current running transaction (if any). Returns true | |
490 | * if a transaction was started, and fills its tid in at *ptid | |
491 | */ | |
492 | int journal_start_commit(journal_t *journal, tid_t *ptid) | |
493 | { | |
494 | int ret = 0; | |
495 | ||
496 | spin_lock(&journal->j_state_lock); | |
497 | if (journal->j_running_transaction) { | |
498 | tid_t tid = journal->j_running_transaction->t_tid; | |
499 | ||
500 | ret = __log_start_commit(journal, tid); | |
501 | if (ret && ptid) | |
502 | *ptid = tid; | |
503 | } else if (journal->j_committing_transaction && ptid) { | |
504 | /* | |
505 | * If ext3_write_super() recently started a commit, then we | |
506 | * have to wait for completion of that transaction | |
507 | */ | |
508 | *ptid = journal->j_committing_transaction->t_tid; | |
509 | ret = 1; | |
510 | } | |
511 | spin_unlock(&journal->j_state_lock); | |
512 | return ret; | |
513 | } | |
514 | ||
515 | /* | |
516 | * Wait for a specified commit to complete. | |
517 | * The caller may not hold the journal lock. | |
518 | */ | |
519 | int log_wait_commit(journal_t *journal, tid_t tid) | |
520 | { | |
521 | int err = 0; | |
522 | ||
523 | #ifdef CONFIG_JBD_DEBUG | |
524 | spin_lock(&journal->j_state_lock); | |
525 | if (!tid_geq(journal->j_commit_request, tid)) { | |
526 | printk(KERN_EMERG | |
527 | "%s: error: j_commit_request=%d, tid=%d\n", | |
528 | __FUNCTION__, journal->j_commit_request, tid); | |
529 | } | |
530 | spin_unlock(&journal->j_state_lock); | |
531 | #endif | |
532 | spin_lock(&journal->j_state_lock); | |
533 | while (tid_gt(tid, journal->j_commit_sequence)) { | |
534 | jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n", | |
535 | tid, journal->j_commit_sequence); | |
536 | wake_up(&journal->j_wait_commit); | |
537 | spin_unlock(&journal->j_state_lock); | |
538 | wait_event(journal->j_wait_done_commit, | |
539 | !tid_gt(tid, journal->j_commit_sequence)); | |
540 | spin_lock(&journal->j_state_lock); | |
541 | } | |
542 | spin_unlock(&journal->j_state_lock); | |
543 | ||
544 | if (unlikely(is_journal_aborted(journal))) { | |
545 | printk(KERN_EMERG "journal commit I/O error\n"); | |
546 | err = -EIO; | |
547 | } | |
548 | return err; | |
549 | } | |
550 | ||
551 | /* | |
552 | * Log buffer allocation routines: | |
553 | */ | |
554 | ||
555 | int journal_next_log_block(journal_t *journal, unsigned long *retp) | |
556 | { | |
557 | unsigned long blocknr; | |
558 | ||
559 | spin_lock(&journal->j_state_lock); | |
560 | J_ASSERT(journal->j_free > 1); | |
561 | ||
562 | blocknr = journal->j_head; | |
563 | journal->j_head++; | |
564 | journal->j_free--; | |
565 | if (journal->j_head == journal->j_last) | |
566 | journal->j_head = journal->j_first; | |
567 | spin_unlock(&journal->j_state_lock); | |
568 | return journal_bmap(journal, blocknr, retp); | |
569 | } | |
570 | ||
571 | /* | |
572 | * Conversion of logical to physical block numbers for the journal | |
573 | * | |
574 | * On external journals the journal blocks are identity-mapped, so | |
575 | * this is a no-op. If needed, we can use j_blk_offset - everything is | |
576 | * ready. | |
577 | */ | |
578 | int journal_bmap(journal_t *journal, unsigned long blocknr, | |
579 | unsigned long *retp) | |
580 | { | |
581 | int err = 0; | |
582 | unsigned long ret; | |
583 | ||
584 | if (journal->j_inode) { | |
585 | ret = bmap(journal->j_inode, blocknr); | |
586 | if (ret) | |
587 | *retp = ret; | |
588 | else { | |
589 | char b[BDEVNAME_SIZE]; | |
590 | ||
591 | printk(KERN_ALERT "%s: journal block not found " | |
592 | "at offset %lu on %s\n", | |
593 | __FUNCTION__, | |
594 | blocknr, | |
595 | bdevname(journal->j_dev, b)); | |
596 | err = -EIO; | |
597 | __journal_abort_soft(journal, err); | |
598 | } | |
599 | } else { | |
600 | *retp = blocknr; /* +journal->j_blk_offset */ | |
601 | } | |
602 | return err; | |
603 | } | |
604 | ||
605 | /* | |
606 | * We play buffer_head aliasing tricks to write data/metadata blocks to | |
607 | * the journal without copying their contents, but for journal | |
608 | * descriptor blocks we do need to generate bona fide buffers. | |
609 | * | |
610 | * After the caller of journal_get_descriptor_buffer() has finished modifying | |
611 | * the buffer's contents they really should run flush_dcache_page(bh->b_page). | |
612 | * But we don't bother doing that, so there will be coherency problems with | |
613 | * mmaps of blockdevs which hold live JBD-controlled filesystems. | |
614 | */ | |
615 | struct journal_head *journal_get_descriptor_buffer(journal_t *journal) | |
616 | { | |
617 | struct buffer_head *bh; | |
618 | unsigned long blocknr; | |
619 | int err; | |
620 | ||
621 | err = journal_next_log_block(journal, &blocknr); | |
622 | ||
623 | if (err) | |
624 | return NULL; | |
625 | ||
626 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
627 | lock_buffer(bh); | |
628 | memset(bh->b_data, 0, journal->j_blocksize); | |
629 | set_buffer_uptodate(bh); | |
630 | unlock_buffer(bh); | |
631 | BUFFER_TRACE(bh, "return this buffer"); | |
632 | return journal_add_journal_head(bh); | |
633 | } | |
634 | ||
635 | /* | |
636 | * Management for journal control blocks: functions to create and | |
637 | * destroy journal_t structures, and to initialise and read existing | |
638 | * journal blocks from disk. */ | |
639 | ||
640 | /* First: create and setup a journal_t object in memory. We initialise | |
641 | * very few fields yet: that has to wait until we have created the | |
642 | * journal structures from from scratch, or loaded them from disk. */ | |
643 | ||
644 | static journal_t * journal_init_common (void) | |
645 | { | |
646 | journal_t *journal; | |
647 | int err; | |
648 | ||
649 | journal = jbd_kmalloc(sizeof(*journal), GFP_KERNEL); | |
650 | if (!journal) | |
651 | goto fail; | |
652 | memset(journal, 0, sizeof(*journal)); | |
653 | ||
654 | init_waitqueue_head(&journal->j_wait_transaction_locked); | |
655 | init_waitqueue_head(&journal->j_wait_logspace); | |
656 | init_waitqueue_head(&journal->j_wait_done_commit); | |
657 | init_waitqueue_head(&journal->j_wait_checkpoint); | |
658 | init_waitqueue_head(&journal->j_wait_commit); | |
659 | init_waitqueue_head(&journal->j_wait_updates); | |
660 | init_MUTEX(&journal->j_barrier); | |
661 | init_MUTEX(&journal->j_checkpoint_sem); | |
662 | spin_lock_init(&journal->j_revoke_lock); | |
663 | spin_lock_init(&journal->j_list_lock); | |
664 | spin_lock_init(&journal->j_state_lock); | |
665 | ||
666 | journal->j_commit_interval = (HZ * JBD_DEFAULT_MAX_COMMIT_AGE); | |
667 | ||
668 | /* The journal is marked for error until we succeed with recovery! */ | |
669 | journal->j_flags = JFS_ABORT; | |
670 | ||
671 | /* Set up a default-sized revoke table for the new mount. */ | |
672 | err = journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); | |
673 | if (err) { | |
674 | kfree(journal); | |
675 | goto fail; | |
676 | } | |
677 | return journal; | |
678 | fail: | |
679 | return NULL; | |
680 | } | |
681 | ||
682 | /* journal_init_dev and journal_init_inode: | |
683 | * | |
684 | * Create a journal structure assigned some fixed set of disk blocks to | |
685 | * the journal. We don't actually touch those disk blocks yet, but we | |
686 | * need to set up all of the mapping information to tell the journaling | |
687 | * system where the journal blocks are. | |
688 | * | |
689 | */ | |
690 | ||
691 | /** | |
692 | * journal_t * journal_init_dev() - creates an initialises a journal structure | |
693 | * @bdev: Block device on which to create the journal | |
694 | * @fs_dev: Device which hold journalled filesystem for this journal. | |
695 | * @start: Block nr Start of journal. | |
696 | * @len: Lenght of the journal in blocks. | |
697 | * @blocksize: blocksize of journalling device | |
698 | * @returns: a newly created journal_t * | |
699 | * | |
700 | * journal_init_dev creates a journal which maps a fixed contiguous | |
701 | * range of blocks on an arbitrary block device. | |
702 | * | |
703 | */ | |
704 | journal_t * journal_init_dev(struct block_device *bdev, | |
705 | struct block_device *fs_dev, | |
706 | int start, int len, int blocksize) | |
707 | { | |
708 | journal_t *journal = journal_init_common(); | |
709 | struct buffer_head *bh; | |
710 | int n; | |
711 | ||
712 | if (!journal) | |
713 | return NULL; | |
714 | ||
715 | journal->j_dev = bdev; | |
716 | journal->j_fs_dev = fs_dev; | |
717 | journal->j_blk_offset = start; | |
718 | journal->j_maxlen = len; | |
719 | journal->j_blocksize = blocksize; | |
720 | ||
721 | bh = __getblk(journal->j_dev, start, journal->j_blocksize); | |
722 | J_ASSERT(bh != NULL); | |
723 | journal->j_sb_buffer = bh; | |
724 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
725 | ||
726 | /* journal descriptor can store up to n blocks -bzzz */ | |
727 | n = journal->j_blocksize / sizeof(journal_block_tag_t); | |
728 | journal->j_wbufsize = n; | |
729 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
730 | if (!journal->j_wbuf) { | |
731 | printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n", | |
732 | __FUNCTION__); | |
733 | kfree(journal); | |
734 | journal = NULL; | |
735 | } | |
736 | ||
737 | return journal; | |
738 | } | |
739 | ||
740 | /** | |
741 | * journal_t * journal_init_inode () - creates a journal which maps to a inode. | |
742 | * @inode: An inode to create the journal in | |
743 | * | |
744 | * journal_init_inode creates a journal which maps an on-disk inode as | |
745 | * the journal. The inode must exist already, must support bmap() and | |
746 | * must have all data blocks preallocated. | |
747 | */ | |
748 | journal_t * journal_init_inode (struct inode *inode) | |
749 | { | |
750 | struct buffer_head *bh; | |
751 | journal_t *journal = journal_init_common(); | |
752 | int err; | |
753 | int n; | |
754 | unsigned long blocknr; | |
755 | ||
756 | if (!journal) | |
757 | return NULL; | |
758 | ||
759 | journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev; | |
760 | journal->j_inode = inode; | |
761 | jbd_debug(1, | |
762 | "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n", | |
763 | journal, inode->i_sb->s_id, inode->i_ino, | |
764 | (long long) inode->i_size, | |
765 | inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); | |
766 | ||
767 | journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits; | |
768 | journal->j_blocksize = inode->i_sb->s_blocksize; | |
769 | ||
770 | /* journal descriptor can store up to n blocks -bzzz */ | |
771 | n = journal->j_blocksize / sizeof(journal_block_tag_t); | |
772 | journal->j_wbufsize = n; | |
773 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
774 | if (!journal->j_wbuf) { | |
775 | printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n", | |
776 | __FUNCTION__); | |
777 | kfree(journal); | |
778 | return NULL; | |
779 | } | |
780 | ||
781 | err = journal_bmap(journal, 0, &blocknr); | |
782 | /* If that failed, give up */ | |
783 | if (err) { | |
784 | printk(KERN_ERR "%s: Cannnot locate journal superblock\n", | |
785 | __FUNCTION__); | |
786 | kfree(journal); | |
787 | return NULL; | |
788 | } | |
789 | ||
790 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
791 | J_ASSERT(bh != NULL); | |
792 | journal->j_sb_buffer = bh; | |
793 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
794 | ||
795 | return journal; | |
796 | } | |
797 | ||
798 | /* | |
799 | * If the journal init or create aborts, we need to mark the journal | |
800 | * superblock as being NULL to prevent the journal destroy from writing | |
801 | * back a bogus superblock. | |
802 | */ | |
803 | static void journal_fail_superblock (journal_t *journal) | |
804 | { | |
805 | struct buffer_head *bh = journal->j_sb_buffer; | |
806 | brelse(bh); | |
807 | journal->j_sb_buffer = NULL; | |
808 | } | |
809 | ||
810 | /* | |
811 | * Given a journal_t structure, initialise the various fields for | |
812 | * startup of a new journaling session. We use this both when creating | |
813 | * a journal, and after recovering an old journal to reset it for | |
814 | * subsequent use. | |
815 | */ | |
816 | ||
817 | static int journal_reset(journal_t *journal) | |
818 | { | |
819 | journal_superblock_t *sb = journal->j_superblock; | |
820 | unsigned int first, last; | |
821 | ||
822 | first = be32_to_cpu(sb->s_first); | |
823 | last = be32_to_cpu(sb->s_maxlen); | |
824 | ||
825 | journal->j_first = first; | |
826 | journal->j_last = last; | |
827 | ||
828 | journal->j_head = first; | |
829 | journal->j_tail = first; | |
830 | journal->j_free = last - first; | |
831 | ||
832 | journal->j_tail_sequence = journal->j_transaction_sequence; | |
833 | journal->j_commit_sequence = journal->j_transaction_sequence - 1; | |
834 | journal->j_commit_request = journal->j_commit_sequence; | |
835 | ||
836 | journal->j_max_transaction_buffers = journal->j_maxlen / 4; | |
837 | ||
838 | /* Add the dynamic fields and write it to disk. */ | |
839 | journal_update_superblock(journal, 1); | |
840 | journal_start_thread(journal); | |
841 | return 0; | |
842 | } | |
843 | ||
844 | /** | |
845 | * int journal_create() - Initialise the new journal file | |
846 | * @journal: Journal to create. This structure must have been initialised | |
847 | * | |
848 | * Given a journal_t structure which tells us which disk blocks we can | |
849 | * use, create a new journal superblock and initialise all of the | |
850 | * journal fields from scratch. | |
851 | **/ | |
852 | int journal_create(journal_t *journal) | |
853 | { | |
854 | unsigned long blocknr; | |
855 | struct buffer_head *bh; | |
856 | journal_superblock_t *sb; | |
857 | int i, err; | |
858 | ||
859 | if (journal->j_maxlen < JFS_MIN_JOURNAL_BLOCKS) { | |
860 | printk (KERN_ERR "Journal length (%d blocks) too short.\n", | |
861 | journal->j_maxlen); | |
862 | journal_fail_superblock(journal); | |
863 | return -EINVAL; | |
864 | } | |
865 | ||
866 | if (journal->j_inode == NULL) { | |
867 | /* | |
868 | * We don't know what block to start at! | |
869 | */ | |
870 | printk(KERN_EMERG | |
871 | "%s: creation of journal on external device!\n", | |
872 | __FUNCTION__); | |
873 | BUG(); | |
874 | } | |
875 | ||
876 | /* Zero out the entire journal on disk. We cannot afford to | |
877 | have any blocks on disk beginning with JFS_MAGIC_NUMBER. */ | |
878 | jbd_debug(1, "JBD: Zeroing out journal blocks...\n"); | |
879 | for (i = 0; i < journal->j_maxlen; i++) { | |
880 | err = journal_bmap(journal, i, &blocknr); | |
881 | if (err) | |
882 | return err; | |
883 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
884 | lock_buffer(bh); | |
885 | memset (bh->b_data, 0, journal->j_blocksize); | |
886 | BUFFER_TRACE(bh, "marking dirty"); | |
887 | mark_buffer_dirty(bh); | |
888 | BUFFER_TRACE(bh, "marking uptodate"); | |
889 | set_buffer_uptodate(bh); | |
890 | unlock_buffer(bh); | |
891 | __brelse(bh); | |
892 | } | |
893 | ||
894 | sync_blockdev(journal->j_dev); | |
895 | jbd_debug(1, "JBD: journal cleared.\n"); | |
896 | ||
897 | /* OK, fill in the initial static fields in the new superblock */ | |
898 | sb = journal->j_superblock; | |
899 | ||
900 | sb->s_header.h_magic = cpu_to_be32(JFS_MAGIC_NUMBER); | |
901 | sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2); | |
902 | ||
903 | sb->s_blocksize = cpu_to_be32(journal->j_blocksize); | |
904 | sb->s_maxlen = cpu_to_be32(journal->j_maxlen); | |
905 | sb->s_first = cpu_to_be32(1); | |
906 | ||
907 | journal->j_transaction_sequence = 1; | |
908 | ||
909 | journal->j_flags &= ~JFS_ABORT; | |
910 | journal->j_format_version = 2; | |
911 | ||
912 | return journal_reset(journal); | |
913 | } | |
914 | ||
915 | /** | |
916 | * void journal_update_superblock() - Update journal sb on disk. | |
917 | * @journal: The journal to update. | |
918 | * @wait: Set to '0' if you don't want to wait for IO completion. | |
919 | * | |
920 | * Update a journal's dynamic superblock fields and write it to disk, | |
921 | * optionally waiting for the IO to complete. | |
922 | */ | |
923 | void journal_update_superblock(journal_t *journal, int wait) | |
924 | { | |
925 | journal_superblock_t *sb = journal->j_superblock; | |
926 | struct buffer_head *bh = journal->j_sb_buffer; | |
927 | ||
928 | /* | |
929 | * As a special case, if the on-disk copy is already marked as needing | |
930 | * no recovery (s_start == 0) and there are no outstanding transactions | |
931 | * in the filesystem, then we can safely defer the superblock update | |
932 | * until the next commit by setting JFS_FLUSHED. This avoids | |
933 | * attempting a write to a potential-readonly device. | |
934 | */ | |
935 | if (sb->s_start == 0 && journal->j_tail_sequence == | |
936 | journal->j_transaction_sequence) { | |
937 | jbd_debug(1,"JBD: Skipping superblock update on recovered sb " | |
938 | "(start %ld, seq %d, errno %d)\n", | |
939 | journal->j_tail, journal->j_tail_sequence, | |
940 | journal->j_errno); | |
941 | goto out; | |
942 | } | |
943 | ||
944 | spin_lock(&journal->j_state_lock); | |
945 | jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n", | |
946 | journal->j_tail, journal->j_tail_sequence, journal->j_errno); | |
947 | ||
948 | sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); | |
949 | sb->s_start = cpu_to_be32(journal->j_tail); | |
950 | sb->s_errno = cpu_to_be32(journal->j_errno); | |
951 | spin_unlock(&journal->j_state_lock); | |
952 | ||
953 | BUFFER_TRACE(bh, "marking dirty"); | |
954 | mark_buffer_dirty(bh); | |
955 | if (wait) | |
956 | sync_dirty_buffer(bh); | |
957 | else | |
958 | ll_rw_block(WRITE, 1, &bh); | |
959 | ||
960 | out: | |
961 | /* If we have just flushed the log (by marking s_start==0), then | |
962 | * any future commit will have to be careful to update the | |
963 | * superblock again to re-record the true start of the log. */ | |
964 | ||
965 | spin_lock(&journal->j_state_lock); | |
966 | if (sb->s_start) | |
967 | journal->j_flags &= ~JFS_FLUSHED; | |
968 | else | |
969 | journal->j_flags |= JFS_FLUSHED; | |
970 | spin_unlock(&journal->j_state_lock); | |
971 | } | |
972 | ||
973 | /* | |
974 | * Read the superblock for a given journal, performing initial | |
975 | * validation of the format. | |
976 | */ | |
977 | ||
978 | static int journal_get_superblock(journal_t *journal) | |
979 | { | |
980 | struct buffer_head *bh; | |
981 | journal_superblock_t *sb; | |
982 | int err = -EIO; | |
983 | ||
984 | bh = journal->j_sb_buffer; | |
985 | ||
986 | J_ASSERT(bh != NULL); | |
987 | if (!buffer_uptodate(bh)) { | |
988 | ll_rw_block(READ, 1, &bh); | |
989 | wait_on_buffer(bh); | |
990 | if (!buffer_uptodate(bh)) { | |
991 | printk (KERN_ERR | |
992 | "JBD: IO error reading journal superblock\n"); | |
993 | goto out; | |
994 | } | |
995 | } | |
996 | ||
997 | sb = journal->j_superblock; | |
998 | ||
999 | err = -EINVAL; | |
1000 | ||
1001 | if (sb->s_header.h_magic != cpu_to_be32(JFS_MAGIC_NUMBER) || | |
1002 | sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { | |
1003 | printk(KERN_WARNING "JBD: no valid journal superblock found\n"); | |
1004 | goto out; | |
1005 | } | |
1006 | ||
1007 | switch(be32_to_cpu(sb->s_header.h_blocktype)) { | |
1008 | case JFS_SUPERBLOCK_V1: | |
1009 | journal->j_format_version = 1; | |
1010 | break; | |
1011 | case JFS_SUPERBLOCK_V2: | |
1012 | journal->j_format_version = 2; | |
1013 | break; | |
1014 | default: | |
1015 | printk(KERN_WARNING "JBD: unrecognised superblock format ID\n"); | |
1016 | goto out; | |
1017 | } | |
1018 | ||
1019 | if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen) | |
1020 | journal->j_maxlen = be32_to_cpu(sb->s_maxlen); | |
1021 | else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) { | |
1022 | printk (KERN_WARNING "JBD: journal file too short\n"); | |
1023 | goto out; | |
1024 | } | |
1025 | ||
1026 | return 0; | |
1027 | ||
1028 | out: | |
1029 | journal_fail_superblock(journal); | |
1030 | return err; | |
1031 | } | |
1032 | ||
1033 | /* | |
1034 | * Load the on-disk journal superblock and read the key fields into the | |
1035 | * journal_t. | |
1036 | */ | |
1037 | ||
1038 | static int load_superblock(journal_t *journal) | |
1039 | { | |
1040 | int err; | |
1041 | journal_superblock_t *sb; | |
1042 | ||
1043 | err = journal_get_superblock(journal); | |
1044 | if (err) | |
1045 | return err; | |
1046 | ||
1047 | sb = journal->j_superblock; | |
1048 | ||
1049 | journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); | |
1050 | journal->j_tail = be32_to_cpu(sb->s_start); | |
1051 | journal->j_first = be32_to_cpu(sb->s_first); | |
1052 | journal->j_last = be32_to_cpu(sb->s_maxlen); | |
1053 | journal->j_errno = be32_to_cpu(sb->s_errno); | |
1054 | ||
1055 | return 0; | |
1056 | } | |
1057 | ||
1058 | ||
1059 | /** | |
1060 | * int journal_load() - Read journal from disk. | |
1061 | * @journal: Journal to act on. | |
1062 | * | |
1063 | * Given a journal_t structure which tells us which disk blocks contain | |
1064 | * a journal, read the journal from disk to initialise the in-memory | |
1065 | * structures. | |
1066 | */ | |
1067 | int journal_load(journal_t *journal) | |
1068 | { | |
1069 | int err; | |
1070 | ||
1071 | err = load_superblock(journal); | |
1072 | if (err) | |
1073 | return err; | |
1074 | ||
1075 | /* If this is a V2 superblock, then we have to check the | |
1076 | * features flags on it. */ | |
1077 | ||
1078 | if (journal->j_format_version >= 2) { | |
1079 | journal_superblock_t *sb = journal->j_superblock; | |
1080 | ||
1081 | if ((sb->s_feature_ro_compat & | |
1082 | ~cpu_to_be32(JFS_KNOWN_ROCOMPAT_FEATURES)) || | |
1083 | (sb->s_feature_incompat & | |
1084 | ~cpu_to_be32(JFS_KNOWN_INCOMPAT_FEATURES))) { | |
1085 | printk (KERN_WARNING | |
1086 | "JBD: Unrecognised features on journal\n"); | |
1087 | return -EINVAL; | |
1088 | } | |
1089 | } | |
1090 | ||
1091 | /* Let the recovery code check whether it needs to recover any | |
1092 | * data from the journal. */ | |
1093 | if (journal_recover(journal)) | |
1094 | goto recovery_error; | |
1095 | ||
1096 | /* OK, we've finished with the dynamic journal bits: | |
1097 | * reinitialise the dynamic contents of the superblock in memory | |
1098 | * and reset them on disk. */ | |
1099 | if (journal_reset(journal)) | |
1100 | goto recovery_error; | |
1101 | ||
1102 | journal->j_flags &= ~JFS_ABORT; | |
1103 | journal->j_flags |= JFS_LOADED; | |
1104 | return 0; | |
1105 | ||
1106 | recovery_error: | |
1107 | printk (KERN_WARNING "JBD: recovery failed\n"); | |
1108 | return -EIO; | |
1109 | } | |
1110 | ||
1111 | /** | |
1112 | * void journal_destroy() - Release a journal_t structure. | |
1113 | * @journal: Journal to act on. | |
1114 | * | |
1115 | * Release a journal_t structure once it is no longer in use by the | |
1116 | * journaled object. | |
1117 | */ | |
1118 | void journal_destroy(journal_t *journal) | |
1119 | { | |
1120 | /* Wait for the commit thread to wake up and die. */ | |
1121 | journal_kill_thread(journal); | |
1122 | ||
1123 | /* Force a final log commit */ | |
1124 | if (journal->j_running_transaction) | |
1125 | journal_commit_transaction(journal); | |
1126 | ||
1127 | /* Force any old transactions to disk */ | |
1128 | ||
1129 | /* Totally anal locking here... */ | |
1130 | spin_lock(&journal->j_list_lock); | |
1131 | while (journal->j_checkpoint_transactions != NULL) { | |
1132 | spin_unlock(&journal->j_list_lock); | |
1133 | log_do_checkpoint(journal); | |
1134 | spin_lock(&journal->j_list_lock); | |
1135 | } | |
1136 | ||
1137 | J_ASSERT(journal->j_running_transaction == NULL); | |
1138 | J_ASSERT(journal->j_committing_transaction == NULL); | |
1139 | J_ASSERT(journal->j_checkpoint_transactions == NULL); | |
1140 | spin_unlock(&journal->j_list_lock); | |
1141 | ||
1142 | /* We can now mark the journal as empty. */ | |
1143 | journal->j_tail = 0; | |
1144 | journal->j_tail_sequence = ++journal->j_transaction_sequence; | |
1145 | if (journal->j_sb_buffer) { | |
1146 | journal_update_superblock(journal, 1); | |
1147 | brelse(journal->j_sb_buffer); | |
1148 | } | |
1149 | ||
1150 | if (journal->j_inode) | |
1151 | iput(journal->j_inode); | |
1152 | if (journal->j_revoke) | |
1153 | journal_destroy_revoke(journal); | |
1154 | kfree(journal->j_wbuf); | |
1155 | kfree(journal); | |
1156 | } | |
1157 | ||
1158 | ||
1159 | /** | |
1160 | *int journal_check_used_features () - Check if features specified are used. | |
1161 | * @journal: Journal to check. | |
1162 | * @compat: bitmask of compatible features | |
1163 | * @ro: bitmask of features that force read-only mount | |
1164 | * @incompat: bitmask of incompatible features | |
1165 | * | |
1166 | * Check whether the journal uses all of a given set of | |
1167 | * features. Return true (non-zero) if it does. | |
1168 | **/ | |
1169 | ||
1170 | int journal_check_used_features (journal_t *journal, unsigned long compat, | |
1171 | unsigned long ro, unsigned long incompat) | |
1172 | { | |
1173 | journal_superblock_t *sb; | |
1174 | ||
1175 | if (!compat && !ro && !incompat) | |
1176 | return 1; | |
1177 | if (journal->j_format_version == 1) | |
1178 | return 0; | |
1179 | ||
1180 | sb = journal->j_superblock; | |
1181 | ||
1182 | if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && | |
1183 | ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && | |
1184 | ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) | |
1185 | return 1; | |
1186 | ||
1187 | return 0; | |
1188 | } | |
1189 | ||
1190 | /** | |
1191 | * int journal_check_available_features() - Check feature set in journalling layer | |
1192 | * @journal: Journal to check. | |
1193 | * @compat: bitmask of compatible features | |
1194 | * @ro: bitmask of features that force read-only mount | |
1195 | * @incompat: bitmask of incompatible features | |
1196 | * | |
1197 | * Check whether the journaling code supports the use of | |
1198 | * all of a given set of features on this journal. Return true | |
1199 | * (non-zero) if it can. */ | |
1200 | ||
1201 | int journal_check_available_features (journal_t *journal, unsigned long compat, | |
1202 | unsigned long ro, unsigned long incompat) | |
1203 | { | |
1204 | journal_superblock_t *sb; | |
1205 | ||
1206 | if (!compat && !ro && !incompat) | |
1207 | return 1; | |
1208 | ||
1209 | sb = journal->j_superblock; | |
1210 | ||
1211 | /* We can support any known requested features iff the | |
1212 | * superblock is in version 2. Otherwise we fail to support any | |
1213 | * extended sb features. */ | |
1214 | ||
1215 | if (journal->j_format_version != 2) | |
1216 | return 0; | |
1217 | ||
1218 | if ((compat & JFS_KNOWN_COMPAT_FEATURES) == compat && | |
1219 | (ro & JFS_KNOWN_ROCOMPAT_FEATURES) == ro && | |
1220 | (incompat & JFS_KNOWN_INCOMPAT_FEATURES) == incompat) | |
1221 | return 1; | |
1222 | ||
1223 | return 0; | |
1224 | } | |
1225 | ||
1226 | /** | |
1227 | * int journal_set_features () - Mark a given journal feature in the superblock | |
1228 | * @journal: Journal to act on. | |
1229 | * @compat: bitmask of compatible features | |
1230 | * @ro: bitmask of features that force read-only mount | |
1231 | * @incompat: bitmask of incompatible features | |
1232 | * | |
1233 | * Mark a given journal feature as present on the | |
1234 | * superblock. Returns true if the requested features could be set. | |
1235 | * | |
1236 | */ | |
1237 | ||
1238 | int journal_set_features (journal_t *journal, unsigned long compat, | |
1239 | unsigned long ro, unsigned long incompat) | |
1240 | { | |
1241 | journal_superblock_t *sb; | |
1242 | ||
1243 | if (journal_check_used_features(journal, compat, ro, incompat)) | |
1244 | return 1; | |
1245 | ||
1246 | if (!journal_check_available_features(journal, compat, ro, incompat)) | |
1247 | return 0; | |
1248 | ||
1249 | jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", | |
1250 | compat, ro, incompat); | |
1251 | ||
1252 | sb = journal->j_superblock; | |
1253 | ||
1254 | sb->s_feature_compat |= cpu_to_be32(compat); | |
1255 | sb->s_feature_ro_compat |= cpu_to_be32(ro); | |
1256 | sb->s_feature_incompat |= cpu_to_be32(incompat); | |
1257 | ||
1258 | return 1; | |
1259 | } | |
1260 | ||
1261 | ||
1262 | /** | |
1263 | * int journal_update_format () - Update on-disk journal structure. | |
1264 | * @journal: Journal to act on. | |
1265 | * | |
1266 | * Given an initialised but unloaded journal struct, poke about in the | |
1267 | * on-disk structure to update it to the most recent supported version. | |
1268 | */ | |
1269 | int journal_update_format (journal_t *journal) | |
1270 | { | |
1271 | journal_superblock_t *sb; | |
1272 | int err; | |
1273 | ||
1274 | err = journal_get_superblock(journal); | |
1275 | if (err) | |
1276 | return err; | |
1277 | ||
1278 | sb = journal->j_superblock; | |
1279 | ||
1280 | switch (be32_to_cpu(sb->s_header.h_blocktype)) { | |
1281 | case JFS_SUPERBLOCK_V2: | |
1282 | return 0; | |
1283 | case JFS_SUPERBLOCK_V1: | |
1284 | return journal_convert_superblock_v1(journal, sb); | |
1285 | default: | |
1286 | break; | |
1287 | } | |
1288 | return -EINVAL; | |
1289 | } | |
1290 | ||
1291 | static int journal_convert_superblock_v1(journal_t *journal, | |
1292 | journal_superblock_t *sb) | |
1293 | { | |
1294 | int offset, blocksize; | |
1295 | struct buffer_head *bh; | |
1296 | ||
1297 | printk(KERN_WARNING | |
1298 | "JBD: Converting superblock from version 1 to 2.\n"); | |
1299 | ||
1300 | /* Pre-initialise new fields to zero */ | |
1301 | offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb); | |
1302 | blocksize = be32_to_cpu(sb->s_blocksize); | |
1303 | memset(&sb->s_feature_compat, 0, blocksize-offset); | |
1304 | ||
1305 | sb->s_nr_users = cpu_to_be32(1); | |
1306 | sb->s_header.h_blocktype = cpu_to_be32(JFS_SUPERBLOCK_V2); | |
1307 | journal->j_format_version = 2; | |
1308 | ||
1309 | bh = journal->j_sb_buffer; | |
1310 | BUFFER_TRACE(bh, "marking dirty"); | |
1311 | mark_buffer_dirty(bh); | |
1312 | sync_dirty_buffer(bh); | |
1313 | return 0; | |
1314 | } | |
1315 | ||
1316 | ||
1317 | /** | |
1318 | * int journal_flush () - Flush journal | |
1319 | * @journal: Journal to act on. | |
1320 | * | |
1321 | * Flush all data for a given journal to disk and empty the journal. | |
1322 | * Filesystems can use this when remounting readonly to ensure that | |
1323 | * recovery does not need to happen on remount. | |
1324 | */ | |
1325 | ||
1326 | int journal_flush(journal_t *journal) | |
1327 | { | |
1328 | int err = 0; | |
1329 | transaction_t *transaction = NULL; | |
1330 | unsigned long old_tail; | |
1331 | ||
1332 | spin_lock(&journal->j_state_lock); | |
1333 | ||
1334 | /* Force everything buffered to the log... */ | |
1335 | if (journal->j_running_transaction) { | |
1336 | transaction = journal->j_running_transaction; | |
1337 | __log_start_commit(journal, transaction->t_tid); | |
1338 | } else if (journal->j_committing_transaction) | |
1339 | transaction = journal->j_committing_transaction; | |
1340 | ||
1341 | /* Wait for the log commit to complete... */ | |
1342 | if (transaction) { | |
1343 | tid_t tid = transaction->t_tid; | |
1344 | ||
1345 | spin_unlock(&journal->j_state_lock); | |
1346 | log_wait_commit(journal, tid); | |
1347 | } else { | |
1348 | spin_unlock(&journal->j_state_lock); | |
1349 | } | |
1350 | ||
1351 | /* ...and flush everything in the log out to disk. */ | |
1352 | spin_lock(&journal->j_list_lock); | |
1353 | while (!err && journal->j_checkpoint_transactions != NULL) { | |
1354 | spin_unlock(&journal->j_list_lock); | |
1355 | err = log_do_checkpoint(journal); | |
1356 | spin_lock(&journal->j_list_lock); | |
1357 | } | |
1358 | spin_unlock(&journal->j_list_lock); | |
1359 | cleanup_journal_tail(journal); | |
1360 | ||
1361 | /* Finally, mark the journal as really needing no recovery. | |
1362 | * This sets s_start==0 in the underlying superblock, which is | |
1363 | * the magic code for a fully-recovered superblock. Any future | |
1364 | * commits of data to the journal will restore the current | |
1365 | * s_start value. */ | |
1366 | spin_lock(&journal->j_state_lock); | |
1367 | old_tail = journal->j_tail; | |
1368 | journal->j_tail = 0; | |
1369 | spin_unlock(&journal->j_state_lock); | |
1370 | journal_update_superblock(journal, 1); | |
1371 | spin_lock(&journal->j_state_lock); | |
1372 | journal->j_tail = old_tail; | |
1373 | ||
1374 | J_ASSERT(!journal->j_running_transaction); | |
1375 | J_ASSERT(!journal->j_committing_transaction); | |
1376 | J_ASSERT(!journal->j_checkpoint_transactions); | |
1377 | J_ASSERT(journal->j_head == journal->j_tail); | |
1378 | J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); | |
1379 | spin_unlock(&journal->j_state_lock); | |
1380 | return err; | |
1381 | } | |
1382 | ||
1383 | /** | |
1384 | * int journal_wipe() - Wipe journal contents | |
1385 | * @journal: Journal to act on. | |
1386 | * @write: flag (see below) | |
1387 | * | |
1388 | * Wipe out all of the contents of a journal, safely. This will produce | |
1389 | * a warning if the journal contains any valid recovery information. | |
1390 | * Must be called between journal_init_*() and journal_load(). | |
1391 | * | |
1392 | * If 'write' is non-zero, then we wipe out the journal on disk; otherwise | |
1393 | * we merely suppress recovery. | |
1394 | */ | |
1395 | ||
1396 | int journal_wipe(journal_t *journal, int write) | |
1397 | { | |
1398 | journal_superblock_t *sb; | |
1399 | int err = 0; | |
1400 | ||
1401 | J_ASSERT (!(journal->j_flags & JFS_LOADED)); | |
1402 | ||
1403 | err = load_superblock(journal); | |
1404 | if (err) | |
1405 | return err; | |
1406 | ||
1407 | sb = journal->j_superblock; | |
1408 | ||
1409 | if (!journal->j_tail) | |
1410 | goto no_recovery; | |
1411 | ||
1412 | printk (KERN_WARNING "JBD: %s recovery information on journal\n", | |
1413 | write ? "Clearing" : "Ignoring"); | |
1414 | ||
1415 | err = journal_skip_recovery(journal); | |
1416 | if (write) | |
1417 | journal_update_superblock(journal, 1); | |
1418 | ||
1419 | no_recovery: | |
1420 | return err; | |
1421 | } | |
1422 | ||
1423 | /* | |
1424 | * journal_dev_name: format a character string to describe on what | |
1425 | * device this journal is present. | |
1426 | */ | |
1427 | ||
022a4a7b | 1428 | static const char *journal_dev_name(journal_t *journal, char *buffer) |
1da177e4 LT |
1429 | { |
1430 | struct block_device *bdev; | |
1431 | ||
1432 | if (journal->j_inode) | |
1433 | bdev = journal->j_inode->i_sb->s_bdev; | |
1434 | else | |
1435 | bdev = journal->j_dev; | |
1436 | ||
1437 | return bdevname(bdev, buffer); | |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * Journal abort has very specific semantics, which we describe | |
1442 | * for journal abort. | |
1443 | * | |
1444 | * Two internal function, which provide abort to te jbd layer | |
1445 | * itself are here. | |
1446 | */ | |
1447 | ||
1448 | /* | |
1449 | * Quick version for internal journal use (doesn't lock the journal). | |
1450 | * Aborts hard --- we mark the abort as occurred, but do _nothing_ else, | |
1451 | * and don't attempt to make any other journal updates. | |
1452 | */ | |
1453 | void __journal_abort_hard(journal_t *journal) | |
1454 | { | |
1455 | transaction_t *transaction; | |
1456 | char b[BDEVNAME_SIZE]; | |
1457 | ||
1458 | if (journal->j_flags & JFS_ABORT) | |
1459 | return; | |
1460 | ||
1461 | printk(KERN_ERR "Aborting journal on device %s.\n", | |
1462 | journal_dev_name(journal, b)); | |
1463 | ||
1464 | spin_lock(&journal->j_state_lock); | |
1465 | journal->j_flags |= JFS_ABORT; | |
1466 | transaction = journal->j_running_transaction; | |
1467 | if (transaction) | |
1468 | __log_start_commit(journal, transaction->t_tid); | |
1469 | spin_unlock(&journal->j_state_lock); | |
1470 | } | |
1471 | ||
1472 | /* Soft abort: record the abort error status in the journal superblock, | |
1473 | * but don't do any other IO. */ | |
022a4a7b | 1474 | static void __journal_abort_soft (journal_t *journal, int errno) |
1da177e4 LT |
1475 | { |
1476 | if (journal->j_flags & JFS_ABORT) | |
1477 | return; | |
1478 | ||
1479 | if (!journal->j_errno) | |
1480 | journal->j_errno = errno; | |
1481 | ||
1482 | __journal_abort_hard(journal); | |
1483 | ||
1484 | if (errno) | |
1485 | journal_update_superblock(journal, 1); | |
1486 | } | |
1487 | ||
1488 | /** | |
1489 | * void journal_abort () - Shutdown the journal immediately. | |
1490 | * @journal: the journal to shutdown. | |
1491 | * @errno: an error number to record in the journal indicating | |
1492 | * the reason for the shutdown. | |
1493 | * | |
1494 | * Perform a complete, immediate shutdown of the ENTIRE | |
1495 | * journal (not of a single transaction). This operation cannot be | |
1496 | * undone without closing and reopening the journal. | |
1497 | * | |
1498 | * The journal_abort function is intended to support higher level error | |
1499 | * recovery mechanisms such as the ext2/ext3 remount-readonly error | |
1500 | * mode. | |
1501 | * | |
1502 | * Journal abort has very specific semantics. Any existing dirty, | |
1503 | * unjournaled buffers in the main filesystem will still be written to | |
1504 | * disk by bdflush, but the journaling mechanism will be suspended | |
1505 | * immediately and no further transaction commits will be honoured. | |
1506 | * | |
1507 | * Any dirty, journaled buffers will be written back to disk without | |
1508 | * hitting the journal. Atomicity cannot be guaranteed on an aborted | |
1509 | * filesystem, but we _do_ attempt to leave as much data as possible | |
1510 | * behind for fsck to use for cleanup. | |
1511 | * | |
1512 | * Any attempt to get a new transaction handle on a journal which is in | |
1513 | * ABORT state will just result in an -EROFS error return. A | |
1514 | * journal_stop on an existing handle will return -EIO if we have | |
1515 | * entered abort state during the update. | |
1516 | * | |
1517 | * Recursive transactions are not disturbed by journal abort until the | |
1518 | * final journal_stop, which will receive the -EIO error. | |
1519 | * | |
1520 | * Finally, the journal_abort call allows the caller to supply an errno | |
1521 | * which will be recorded (if possible) in the journal superblock. This | |
1522 | * allows a client to record failure conditions in the middle of a | |
1523 | * transaction without having to complete the transaction to record the | |
1524 | * failure to disk. ext3_error, for example, now uses this | |
1525 | * functionality. | |
1526 | * | |
1527 | * Errors which originate from within the journaling layer will NOT | |
1528 | * supply an errno; a null errno implies that absolutely no further | |
1529 | * writes are done to the journal (unless there are any already in | |
1530 | * progress). | |
1531 | * | |
1532 | */ | |
1533 | ||
1534 | void journal_abort(journal_t *journal, int errno) | |
1535 | { | |
1536 | __journal_abort_soft(journal, errno); | |
1537 | } | |
1538 | ||
1539 | /** | |
1540 | * int journal_errno () - returns the journal's error state. | |
1541 | * @journal: journal to examine. | |
1542 | * | |
1543 | * This is the errno numbet set with journal_abort(), the last | |
1544 | * time the journal was mounted - if the journal was stopped | |
1545 | * without calling abort this will be 0. | |
1546 | * | |
1547 | * If the journal has been aborted on this mount time -EROFS will | |
1548 | * be returned. | |
1549 | */ | |
1550 | int journal_errno(journal_t *journal) | |
1551 | { | |
1552 | int err; | |
1553 | ||
1554 | spin_lock(&journal->j_state_lock); | |
1555 | if (journal->j_flags & JFS_ABORT) | |
1556 | err = -EROFS; | |
1557 | else | |
1558 | err = journal->j_errno; | |
1559 | spin_unlock(&journal->j_state_lock); | |
1560 | return err; | |
1561 | } | |
1562 | ||
1563 | /** | |
1564 | * int journal_clear_err () - clears the journal's error state | |
1565 | * @journal: journal to act on. | |
1566 | * | |
1567 | * An error must be cleared or Acked to take a FS out of readonly | |
1568 | * mode. | |
1569 | */ | |
1570 | int journal_clear_err(journal_t *journal) | |
1571 | { | |
1572 | int err = 0; | |
1573 | ||
1574 | spin_lock(&journal->j_state_lock); | |
1575 | if (journal->j_flags & JFS_ABORT) | |
1576 | err = -EROFS; | |
1577 | else | |
1578 | journal->j_errno = 0; | |
1579 | spin_unlock(&journal->j_state_lock); | |
1580 | return err; | |
1581 | } | |
1582 | ||
1583 | /** | |
1584 | * void journal_ack_err() - Ack journal err. | |
1585 | * @journal: journal to act on. | |
1586 | * | |
1587 | * An error must be cleared or Acked to take a FS out of readonly | |
1588 | * mode. | |
1589 | */ | |
1590 | void journal_ack_err(journal_t *journal) | |
1591 | { | |
1592 | spin_lock(&journal->j_state_lock); | |
1593 | if (journal->j_errno) | |
1594 | journal->j_flags |= JFS_ACK_ERR; | |
1595 | spin_unlock(&journal->j_state_lock); | |
1596 | } | |
1597 | ||
1598 | int journal_blocks_per_page(struct inode *inode) | |
1599 | { | |
1600 | return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
1601 | } | |
1602 | ||
1603 | /* | |
1604 | * Simple support for retrying memory allocations. Introduced to help to | |
1605 | * debug different VM deadlock avoidance strategies. | |
1606 | */ | |
1607 | void * __jbd_kmalloc (const char *where, size_t size, int flags, int retry) | |
1608 | { | |
1609 | return kmalloc(size, flags | (retry ? __GFP_NOFAIL : 0)); | |
1610 | } | |
1611 | ||
1612 | /* | |
1613 | * Journal_head storage management | |
1614 | */ | |
1615 | static kmem_cache_t *journal_head_cache; | |
1616 | #ifdef CONFIG_JBD_DEBUG | |
1617 | static atomic_t nr_journal_heads = ATOMIC_INIT(0); | |
1618 | #endif | |
1619 | ||
1620 | static int journal_init_journal_head_cache(void) | |
1621 | { | |
1622 | int retval; | |
1623 | ||
1624 | J_ASSERT(journal_head_cache == 0); | |
1625 | journal_head_cache = kmem_cache_create("journal_head", | |
1626 | sizeof(struct journal_head), | |
1627 | 0, /* offset */ | |
1628 | 0, /* flags */ | |
1629 | NULL, /* ctor */ | |
1630 | NULL); /* dtor */ | |
1631 | retval = 0; | |
1632 | if (journal_head_cache == 0) { | |
1633 | retval = -ENOMEM; | |
1634 | printk(KERN_EMERG "JBD: no memory for journal_head cache\n"); | |
1635 | } | |
1636 | return retval; | |
1637 | } | |
1638 | ||
1639 | static void journal_destroy_journal_head_cache(void) | |
1640 | { | |
1641 | J_ASSERT(journal_head_cache != NULL); | |
1642 | kmem_cache_destroy(journal_head_cache); | |
1643 | journal_head_cache = NULL; | |
1644 | } | |
1645 | ||
1646 | /* | |
1647 | * journal_head splicing and dicing | |
1648 | */ | |
1649 | static struct journal_head *journal_alloc_journal_head(void) | |
1650 | { | |
1651 | struct journal_head *ret; | |
1652 | static unsigned long last_warning; | |
1653 | ||
1654 | #ifdef CONFIG_JBD_DEBUG | |
1655 | atomic_inc(&nr_journal_heads); | |
1656 | #endif | |
1657 | ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS); | |
1658 | if (ret == 0) { | |
1659 | jbd_debug(1, "out of memory for journal_head\n"); | |
1660 | if (time_after(jiffies, last_warning + 5*HZ)) { | |
1661 | printk(KERN_NOTICE "ENOMEM in %s, retrying.\n", | |
1662 | __FUNCTION__); | |
1663 | last_warning = jiffies; | |
1664 | } | |
1665 | while (ret == 0) { | |
1666 | yield(); | |
1667 | ret = kmem_cache_alloc(journal_head_cache, GFP_NOFS); | |
1668 | } | |
1669 | } | |
1670 | return ret; | |
1671 | } | |
1672 | ||
1673 | static void journal_free_journal_head(struct journal_head *jh) | |
1674 | { | |
1675 | #ifdef CONFIG_JBD_DEBUG | |
1676 | atomic_dec(&nr_journal_heads); | |
1677 | memset(jh, 0x5b, sizeof(*jh)); | |
1678 | #endif | |
1679 | kmem_cache_free(journal_head_cache, jh); | |
1680 | } | |
1681 | ||
1682 | /* | |
1683 | * A journal_head is attached to a buffer_head whenever JBD has an | |
1684 | * interest in the buffer. | |
1685 | * | |
1686 | * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit | |
1687 | * is set. This bit is tested in core kernel code where we need to take | |
1688 | * JBD-specific actions. Testing the zeroness of ->b_private is not reliable | |
1689 | * there. | |
1690 | * | |
1691 | * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. | |
1692 | * | |
1693 | * When a buffer has its BH_JBD bit set it is immune from being released by | |
1694 | * core kernel code, mainly via ->b_count. | |
1695 | * | |
1696 | * A journal_head may be detached from its buffer_head when the journal_head's | |
1697 | * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL. | |
1698 | * Various places in JBD call journal_remove_journal_head() to indicate that the | |
1699 | * journal_head can be dropped if needed. | |
1700 | * | |
1701 | * Various places in the kernel want to attach a journal_head to a buffer_head | |
1702 | * _before_ attaching the journal_head to a transaction. To protect the | |
1703 | * journal_head in this situation, journal_add_journal_head elevates the | |
1704 | * journal_head's b_jcount refcount by one. The caller must call | |
1705 | * journal_put_journal_head() to undo this. | |
1706 | * | |
1707 | * So the typical usage would be: | |
1708 | * | |
1709 | * (Attach a journal_head if needed. Increments b_jcount) | |
1710 | * struct journal_head *jh = journal_add_journal_head(bh); | |
1711 | * ... | |
1712 | * jh->b_transaction = xxx; | |
1713 | * journal_put_journal_head(jh); | |
1714 | * | |
1715 | * Now, the journal_head's b_jcount is zero, but it is safe from being released | |
1716 | * because it has a non-zero b_transaction. | |
1717 | */ | |
1718 | ||
1719 | /* | |
1720 | * Give a buffer_head a journal_head. | |
1721 | * | |
1722 | * Doesn't need the journal lock. | |
1723 | * May sleep. | |
1724 | */ | |
1725 | struct journal_head *journal_add_journal_head(struct buffer_head *bh) | |
1726 | { | |
1727 | struct journal_head *jh; | |
1728 | struct journal_head *new_jh = NULL; | |
1729 | ||
1730 | repeat: | |
1731 | if (!buffer_jbd(bh)) { | |
1732 | new_jh = journal_alloc_journal_head(); | |
1733 | memset(new_jh, 0, sizeof(*new_jh)); | |
1734 | } | |
1735 | ||
1736 | jbd_lock_bh_journal_head(bh); | |
1737 | if (buffer_jbd(bh)) { | |
1738 | jh = bh2jh(bh); | |
1739 | } else { | |
1740 | J_ASSERT_BH(bh, | |
1741 | (atomic_read(&bh->b_count) > 0) || | |
1742 | (bh->b_page && bh->b_page->mapping)); | |
1743 | ||
1744 | if (!new_jh) { | |
1745 | jbd_unlock_bh_journal_head(bh); | |
1746 | goto repeat; | |
1747 | } | |
1748 | ||
1749 | jh = new_jh; | |
1750 | new_jh = NULL; /* We consumed it */ | |
1751 | set_buffer_jbd(bh); | |
1752 | bh->b_private = jh; | |
1753 | jh->b_bh = bh; | |
1754 | get_bh(bh); | |
1755 | BUFFER_TRACE(bh, "added journal_head"); | |
1756 | } | |
1757 | jh->b_jcount++; | |
1758 | jbd_unlock_bh_journal_head(bh); | |
1759 | if (new_jh) | |
1760 | journal_free_journal_head(new_jh); | |
1761 | return bh->b_private; | |
1762 | } | |
1763 | ||
1764 | /* | |
1765 | * Grab a ref against this buffer_head's journal_head. If it ended up not | |
1766 | * having a journal_head, return NULL | |
1767 | */ | |
1768 | struct journal_head *journal_grab_journal_head(struct buffer_head *bh) | |
1769 | { | |
1770 | struct journal_head *jh = NULL; | |
1771 | ||
1772 | jbd_lock_bh_journal_head(bh); | |
1773 | if (buffer_jbd(bh)) { | |
1774 | jh = bh2jh(bh); | |
1775 | jh->b_jcount++; | |
1776 | } | |
1777 | jbd_unlock_bh_journal_head(bh); | |
1778 | return jh; | |
1779 | } | |
1780 | ||
1781 | static void __journal_remove_journal_head(struct buffer_head *bh) | |
1782 | { | |
1783 | struct journal_head *jh = bh2jh(bh); | |
1784 | ||
1785 | J_ASSERT_JH(jh, jh->b_jcount >= 0); | |
1786 | ||
1787 | get_bh(bh); | |
1788 | if (jh->b_jcount == 0) { | |
1789 | if (jh->b_transaction == NULL && | |
1790 | jh->b_next_transaction == NULL && | |
1791 | jh->b_cp_transaction == NULL) { | |
1792 | J_ASSERT_JH(jh, jh->b_jlist == BJ_None); | |
1793 | J_ASSERT_BH(bh, buffer_jbd(bh)); | |
1794 | J_ASSERT_BH(bh, jh2bh(jh) == bh); | |
1795 | BUFFER_TRACE(bh, "remove journal_head"); | |
1796 | if (jh->b_frozen_data) { | |
1797 | printk(KERN_WARNING "%s: freeing " | |
1798 | "b_frozen_data\n", | |
1799 | __FUNCTION__); | |
1800 | kfree(jh->b_frozen_data); | |
1801 | } | |
1802 | if (jh->b_committed_data) { | |
1803 | printk(KERN_WARNING "%s: freeing " | |
1804 | "b_committed_data\n", | |
1805 | __FUNCTION__); | |
1806 | kfree(jh->b_committed_data); | |
1807 | } | |
1808 | bh->b_private = NULL; | |
1809 | jh->b_bh = NULL; /* debug, really */ | |
1810 | clear_buffer_jbd(bh); | |
1811 | __brelse(bh); | |
1812 | journal_free_journal_head(jh); | |
1813 | } else { | |
1814 | BUFFER_TRACE(bh, "journal_head was locked"); | |
1815 | } | |
1816 | } | |
1817 | } | |
1818 | ||
1819 | /* | |
1820 | * journal_remove_journal_head(): if the buffer isn't attached to a transaction | |
1821 | * and has a zero b_jcount then remove and release its journal_head. If we did | |
1822 | * see that the buffer is not used by any transaction we also "logically" | |
1823 | * decrement ->b_count. | |
1824 | * | |
1825 | * We in fact take an additional increment on ->b_count as a convenience, | |
1826 | * because the caller usually wants to do additional things with the bh | |
1827 | * after calling here. | |
1828 | * The caller of journal_remove_journal_head() *must* run __brelse(bh) at some | |
1829 | * time. Once the caller has run __brelse(), the buffer is eligible for | |
1830 | * reaping by try_to_free_buffers(). | |
1831 | */ | |
1832 | void journal_remove_journal_head(struct buffer_head *bh) | |
1833 | { | |
1834 | jbd_lock_bh_journal_head(bh); | |
1835 | __journal_remove_journal_head(bh); | |
1836 | jbd_unlock_bh_journal_head(bh); | |
1837 | } | |
1838 | ||
1839 | /* | |
1840 | * Drop a reference on the passed journal_head. If it fell to zero then try to | |
1841 | * release the journal_head from the buffer_head. | |
1842 | */ | |
1843 | void journal_put_journal_head(struct journal_head *jh) | |
1844 | { | |
1845 | struct buffer_head *bh = jh2bh(jh); | |
1846 | ||
1847 | jbd_lock_bh_journal_head(bh); | |
1848 | J_ASSERT_JH(jh, jh->b_jcount > 0); | |
1849 | --jh->b_jcount; | |
1850 | if (!jh->b_jcount && !jh->b_transaction) { | |
1851 | __journal_remove_journal_head(bh); | |
1852 | __brelse(bh); | |
1853 | } | |
1854 | jbd_unlock_bh_journal_head(bh); | |
1855 | } | |
1856 | ||
1857 | /* | |
1858 | * /proc tunables | |
1859 | */ | |
1860 | #if defined(CONFIG_JBD_DEBUG) | |
1861 | int journal_enable_debug; | |
1862 | EXPORT_SYMBOL(journal_enable_debug); | |
1863 | #endif | |
1864 | ||
1865 | #if defined(CONFIG_JBD_DEBUG) && defined(CONFIG_PROC_FS) | |
1866 | ||
1867 | static struct proc_dir_entry *proc_jbd_debug; | |
1868 | ||
022a4a7b | 1869 | static int read_jbd_debug(char *page, char **start, off_t off, |
1da177e4 LT |
1870 | int count, int *eof, void *data) |
1871 | { | |
1872 | int ret; | |
1873 | ||
1874 | ret = sprintf(page + off, "%d\n", journal_enable_debug); | |
1875 | *eof = 1; | |
1876 | return ret; | |
1877 | } | |
1878 | ||
022a4a7b | 1879 | static int write_jbd_debug(struct file *file, const char __user *buffer, |
1da177e4 LT |
1880 | unsigned long count, void *data) |
1881 | { | |
1882 | char buf[32]; | |
1883 | ||
1884 | if (count > ARRAY_SIZE(buf) - 1) | |
1885 | count = ARRAY_SIZE(buf) - 1; | |
1886 | if (copy_from_user(buf, buffer, count)) | |
1887 | return -EFAULT; | |
1888 | buf[ARRAY_SIZE(buf) - 1] = '\0'; | |
1889 | journal_enable_debug = simple_strtoul(buf, NULL, 10); | |
1890 | return count; | |
1891 | } | |
1892 | ||
1893 | #define JBD_PROC_NAME "sys/fs/jbd-debug" | |
1894 | ||
1895 | static void __init create_jbd_proc_entry(void) | |
1896 | { | |
1897 | proc_jbd_debug = create_proc_entry(JBD_PROC_NAME, 0644, NULL); | |
1898 | if (proc_jbd_debug) { | |
1899 | /* Why is this so hard? */ | |
1900 | proc_jbd_debug->read_proc = read_jbd_debug; | |
1901 | proc_jbd_debug->write_proc = write_jbd_debug; | |
1902 | } | |
1903 | } | |
1904 | ||
1905 | static void __exit remove_jbd_proc_entry(void) | |
1906 | { | |
1907 | if (proc_jbd_debug) | |
1908 | remove_proc_entry(JBD_PROC_NAME, NULL); | |
1909 | } | |
1910 | ||
1911 | #else | |
1912 | ||
1913 | #define create_jbd_proc_entry() do {} while (0) | |
1914 | #define remove_jbd_proc_entry() do {} while (0) | |
1915 | ||
1916 | #endif | |
1917 | ||
1918 | kmem_cache_t *jbd_handle_cache; | |
1919 | ||
1920 | static int __init journal_init_handle_cache(void) | |
1921 | { | |
1922 | jbd_handle_cache = kmem_cache_create("journal_handle", | |
1923 | sizeof(handle_t), | |
1924 | 0, /* offset */ | |
1925 | 0, /* flags */ | |
1926 | NULL, /* ctor */ | |
1927 | NULL); /* dtor */ | |
1928 | if (jbd_handle_cache == NULL) { | |
1929 | printk(KERN_EMERG "JBD: failed to create handle cache\n"); | |
1930 | return -ENOMEM; | |
1931 | } | |
1932 | return 0; | |
1933 | } | |
1934 | ||
1935 | static void journal_destroy_handle_cache(void) | |
1936 | { | |
1937 | if (jbd_handle_cache) | |
1938 | kmem_cache_destroy(jbd_handle_cache); | |
1939 | } | |
1940 | ||
1941 | /* | |
1942 | * Module startup and shutdown | |
1943 | */ | |
1944 | ||
1945 | static int __init journal_init_caches(void) | |
1946 | { | |
1947 | int ret; | |
1948 | ||
1949 | ret = journal_init_revoke_caches(); | |
1950 | if (ret == 0) | |
1951 | ret = journal_init_journal_head_cache(); | |
1952 | if (ret == 0) | |
1953 | ret = journal_init_handle_cache(); | |
1954 | return ret; | |
1955 | } | |
1956 | ||
1957 | static void journal_destroy_caches(void) | |
1958 | { | |
1959 | journal_destroy_revoke_caches(); | |
1960 | journal_destroy_journal_head_cache(); | |
1961 | journal_destroy_handle_cache(); | |
1962 | } | |
1963 | ||
1964 | static int __init journal_init(void) | |
1965 | { | |
1966 | int ret; | |
1967 | ||
022a4a7b AB |
1968 | /* Static check for data structure consistency. There's no code |
1969 | * invoked --- we'll just get a linker failure if things aren't right. | |
1970 | */ | |
1971 | extern void journal_bad_superblock_size(void); | |
1972 | if (sizeof(struct journal_superblock_s) != 1024) | |
1973 | journal_bad_superblock_size(); | |
1974 | ||
1975 | ||
1da177e4 LT |
1976 | ret = journal_init_caches(); |
1977 | if (ret != 0) | |
1978 | journal_destroy_caches(); | |
1979 | create_jbd_proc_entry(); | |
1980 | return ret; | |
1981 | } | |
1982 | ||
1983 | static void __exit journal_exit(void) | |
1984 | { | |
1985 | #ifdef CONFIG_JBD_DEBUG | |
1986 | int n = atomic_read(&nr_journal_heads); | |
1987 | if (n) | |
1988 | printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n); | |
1989 | #endif | |
1990 | remove_jbd_proc_entry(); | |
1991 | journal_destroy_caches(); | |
1992 | } | |
1993 | ||
1994 | MODULE_LICENSE("GPL"); | |
1995 | module_init(journal_init); | |
1996 | module_exit(journal_exit); | |
1997 |