]>
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> |
470decc6 DK |
40 | |
41 | #include <asm/uaccess.h> | |
42 | #include <asm/page.h> | |
d7cfa468 | 43 | #include <asm/div64.h> |
470decc6 | 44 | |
f7f4bccb MC |
45 | EXPORT_SYMBOL(jbd2_journal_start); |
46 | EXPORT_SYMBOL(jbd2_journal_restart); | |
47 | EXPORT_SYMBOL(jbd2_journal_extend); | |
48 | EXPORT_SYMBOL(jbd2_journal_stop); | |
49 | EXPORT_SYMBOL(jbd2_journal_lock_updates); | |
50 | EXPORT_SYMBOL(jbd2_journal_unlock_updates); | |
51 | EXPORT_SYMBOL(jbd2_journal_get_write_access); | |
52 | EXPORT_SYMBOL(jbd2_journal_get_create_access); | |
53 | EXPORT_SYMBOL(jbd2_journal_get_undo_access); | |
f7f4bccb MC |
54 | EXPORT_SYMBOL(jbd2_journal_dirty_metadata); |
55 | EXPORT_SYMBOL(jbd2_journal_release_buffer); | |
56 | EXPORT_SYMBOL(jbd2_journal_forget); | |
470decc6 DK |
57 | #if 0 |
58 | EXPORT_SYMBOL(journal_sync_buffer); | |
59 | #endif | |
f7f4bccb MC |
60 | EXPORT_SYMBOL(jbd2_journal_flush); |
61 | EXPORT_SYMBOL(jbd2_journal_revoke); | |
62 | ||
63 | EXPORT_SYMBOL(jbd2_journal_init_dev); | |
64 | EXPORT_SYMBOL(jbd2_journal_init_inode); | |
65 | EXPORT_SYMBOL(jbd2_journal_update_format); | |
66 | EXPORT_SYMBOL(jbd2_journal_check_used_features); | |
67 | EXPORT_SYMBOL(jbd2_journal_check_available_features); | |
68 | EXPORT_SYMBOL(jbd2_journal_set_features); | |
f7f4bccb MC |
69 | EXPORT_SYMBOL(jbd2_journal_load); |
70 | EXPORT_SYMBOL(jbd2_journal_destroy); | |
f7f4bccb MC |
71 | EXPORT_SYMBOL(jbd2_journal_abort); |
72 | EXPORT_SYMBOL(jbd2_journal_errno); | |
73 | EXPORT_SYMBOL(jbd2_journal_ack_err); | |
74 | EXPORT_SYMBOL(jbd2_journal_clear_err); | |
75 | EXPORT_SYMBOL(jbd2_log_wait_commit); | |
76 | EXPORT_SYMBOL(jbd2_journal_start_commit); | |
77 | EXPORT_SYMBOL(jbd2_journal_force_commit_nested); | |
78 | EXPORT_SYMBOL(jbd2_journal_wipe); | |
79 | EXPORT_SYMBOL(jbd2_journal_blocks_per_page); | |
80 | EXPORT_SYMBOL(jbd2_journal_invalidatepage); | |
81 | EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); | |
82 | EXPORT_SYMBOL(jbd2_journal_force_commit); | |
c851ed54 JK |
83 | EXPORT_SYMBOL(jbd2_journal_file_inode); |
84 | EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); | |
85 | EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); | |
86 | EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); | |
470decc6 DK |
87 | |
88 | static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *); | |
89 | static void __journal_abort_soft (journal_t *journal, int errno); | |
470decc6 DK |
90 | |
91 | /* | |
92 | * Helper function used to manage commit timeouts | |
93 | */ | |
94 | ||
95 | static void commit_timeout(unsigned long __data) | |
96 | { | |
97 | struct task_struct * p = (struct task_struct *) __data; | |
98 | ||
99 | wake_up_process(p); | |
100 | } | |
101 | ||
102 | /* | |
f7f4bccb | 103 | * kjournald2: The main thread function used to manage a logging device |
470decc6 DK |
104 | * journal. |
105 | * | |
106 | * This kernel thread is responsible for two things: | |
107 | * | |
108 | * 1) COMMIT: Every so often we need to commit the current state of the | |
109 | * filesystem to disk. The journal thread is responsible for writing | |
110 | * all of the metadata buffers to disk. | |
111 | * | |
112 | * 2) CHECKPOINT: We cannot reuse a used section of the log file until all | |
113 | * of the data in that part of the log has been rewritten elsewhere on | |
114 | * the disk. Flushing these old buffers to reclaim space in the log is | |
115 | * known as checkpointing, and this thread is responsible for that job. | |
116 | */ | |
117 | ||
f7f4bccb | 118 | static int kjournald2(void *arg) |
470decc6 DK |
119 | { |
120 | journal_t *journal = arg; | |
121 | transaction_t *transaction; | |
122 | ||
123 | /* | |
124 | * Set up an interval timer which can be used to trigger a commit wakeup | |
125 | * after the commit interval expires | |
126 | */ | |
127 | setup_timer(&journal->j_commit_timer, commit_timeout, | |
128 | (unsigned long)current); | |
129 | ||
130 | /* Record that the journal thread is running */ | |
131 | journal->j_task = current; | |
132 | wake_up(&journal->j_wait_done_commit); | |
133 | ||
4a9bf99b TT |
134 | printk(KERN_INFO "kjournald2 starting: pid %d, dev %s, " |
135 | "commit interval %ld seconds\n", current->pid, | |
136 | journal->j_devname, journal->j_commit_interval / HZ); | |
470decc6 DK |
137 | |
138 | /* | |
139 | * And now, wait forever for commit wakeup events. | |
140 | */ | |
141 | spin_lock(&journal->j_state_lock); | |
142 | ||
143 | loop: | |
f7f4bccb | 144 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
145 | goto end_loop; |
146 | ||
147 | jbd_debug(1, "commit_sequence=%d, commit_request=%d\n", | |
148 | journal->j_commit_sequence, journal->j_commit_request); | |
149 | ||
150 | if (journal->j_commit_sequence != journal->j_commit_request) { | |
151 | jbd_debug(1, "OK, requests differ\n"); | |
152 | spin_unlock(&journal->j_state_lock); | |
153 | del_timer_sync(&journal->j_commit_timer); | |
f7f4bccb | 154 | jbd2_journal_commit_transaction(journal); |
470decc6 DK |
155 | spin_lock(&journal->j_state_lock); |
156 | goto loop; | |
157 | } | |
158 | ||
159 | wake_up(&journal->j_wait_done_commit); | |
160 | if (freezing(current)) { | |
161 | /* | |
162 | * The simpler the better. Flushing journal isn't a | |
163 | * good idea, because that depends on threads that may | |
164 | * be already stopped. | |
165 | */ | |
f7f4bccb | 166 | jbd_debug(1, "Now suspending kjournald2\n"); |
470decc6 DK |
167 | spin_unlock(&journal->j_state_lock); |
168 | refrigerator(); | |
169 | spin_lock(&journal->j_state_lock); | |
170 | } else { | |
171 | /* | |
172 | * We assume on resume that commits are already there, | |
173 | * so we don't sleep | |
174 | */ | |
175 | DEFINE_WAIT(wait); | |
176 | int should_sleep = 1; | |
177 | ||
178 | prepare_to_wait(&journal->j_wait_commit, &wait, | |
179 | TASK_INTERRUPTIBLE); | |
180 | if (journal->j_commit_sequence != journal->j_commit_request) | |
181 | should_sleep = 0; | |
182 | transaction = journal->j_running_transaction; | |
183 | if (transaction && time_after_eq(jiffies, | |
184 | transaction->t_expires)) | |
185 | should_sleep = 0; | |
f7f4bccb | 186 | if (journal->j_flags & JBD2_UNMOUNT) |
470decc6 DK |
187 | should_sleep = 0; |
188 | if (should_sleep) { | |
189 | spin_unlock(&journal->j_state_lock); | |
190 | schedule(); | |
191 | spin_lock(&journal->j_state_lock); | |
192 | } | |
193 | finish_wait(&journal->j_wait_commit, &wait); | |
194 | } | |
195 | ||
f7f4bccb | 196 | jbd_debug(1, "kjournald2 wakes\n"); |
470decc6 DK |
197 | |
198 | /* | |
199 | * Were we woken up by a commit wakeup event? | |
200 | */ | |
201 | transaction = journal->j_running_transaction; | |
202 | if (transaction && time_after_eq(jiffies, transaction->t_expires)) { | |
203 | journal->j_commit_request = transaction->t_tid; | |
204 | jbd_debug(1, "woke because of timeout\n"); | |
205 | } | |
206 | goto loop; | |
207 | ||
208 | end_loop: | |
209 | spin_unlock(&journal->j_state_lock); | |
210 | del_timer_sync(&journal->j_commit_timer); | |
211 | journal->j_task = NULL; | |
212 | wake_up(&journal->j_wait_done_commit); | |
213 | jbd_debug(1, "Journal thread exiting.\n"); | |
214 | return 0; | |
215 | } | |
216 | ||
97f06784 | 217 | static int jbd2_journal_start_thread(journal_t *journal) |
470decc6 | 218 | { |
97f06784 PE |
219 | struct task_struct *t; |
220 | ||
221 | t = kthread_run(kjournald2, journal, "kjournald2"); | |
222 | if (IS_ERR(t)) | |
223 | return PTR_ERR(t); | |
224 | ||
1076d17a | 225 | wait_event(journal->j_wait_done_commit, journal->j_task != NULL); |
97f06784 | 226 | return 0; |
470decc6 DK |
227 | } |
228 | ||
229 | static void journal_kill_thread(journal_t *journal) | |
230 | { | |
231 | spin_lock(&journal->j_state_lock); | |
f7f4bccb | 232 | journal->j_flags |= JBD2_UNMOUNT; |
470decc6 DK |
233 | |
234 | while (journal->j_task) { | |
235 | wake_up(&journal->j_wait_commit); | |
236 | spin_unlock(&journal->j_state_lock); | |
1076d17a | 237 | wait_event(journal->j_wait_done_commit, journal->j_task == NULL); |
470decc6 DK |
238 | spin_lock(&journal->j_state_lock); |
239 | } | |
240 | spin_unlock(&journal->j_state_lock); | |
241 | } | |
242 | ||
243 | /* | |
f7f4bccb | 244 | * jbd2_journal_write_metadata_buffer: write a metadata buffer to the journal. |
470decc6 DK |
245 | * |
246 | * Writes a metadata buffer to a given disk block. The actual IO is not | |
247 | * performed but a new buffer_head is constructed which labels the data | |
248 | * to be written with the correct destination disk block. | |
249 | * | |
250 | * Any magic-number escaping which needs to be done will cause a | |
251 | * copy-out here. If the buffer happens to start with the | |
f7f4bccb | 252 | * JBD2_MAGIC_NUMBER, then we can't write it to the log directly: the |
470decc6 DK |
253 | * magic number is only written to the log for descripter blocks. In |
254 | * this case, we copy the data and replace the first word with 0, and we | |
255 | * return a result code which indicates that this buffer needs to be | |
256 | * marked as an escaped buffer in the corresponding log descriptor | |
257 | * block. The missing word can then be restored when the block is read | |
258 | * during recovery. | |
259 | * | |
260 | * If the source buffer has already been modified by a new transaction | |
261 | * since we took the last commit snapshot, we use the frozen copy of | |
262 | * that data for IO. If we end up using the existing buffer_head's data | |
263 | * for the write, then we *have* to lock the buffer to prevent anyone | |
264 | * else from using and possibly modifying it while the IO is in | |
265 | * progress. | |
266 | * | |
267 | * The function returns a pointer to the buffer_heads to be used for IO. | |
268 | * | |
269 | * We assume that the journal has already been locked in this function. | |
270 | * | |
271 | * Return value: | |
272 | * <0: Error | |
273 | * >=0: Finished OK | |
274 | * | |
275 | * On success: | |
276 | * Bit 0 set == escape performed on the data | |
277 | * Bit 1 set == buffer copy-out performed (kfree the data after IO) | |
278 | */ | |
279 | ||
f7f4bccb | 280 | int jbd2_journal_write_metadata_buffer(transaction_t *transaction, |
470decc6 DK |
281 | struct journal_head *jh_in, |
282 | struct journal_head **jh_out, | |
18eba7aa | 283 | unsigned long long blocknr) |
470decc6 DK |
284 | { |
285 | int need_copy_out = 0; | |
286 | int done_copy_out = 0; | |
287 | int do_escape = 0; | |
288 | char *mapped_data; | |
289 | struct buffer_head *new_bh; | |
290 | struct journal_head *new_jh; | |
291 | struct page *new_page; | |
292 | unsigned int new_offset; | |
293 | struct buffer_head *bh_in = jh2bh(jh_in); | |
294 | ||
295 | /* | |
296 | * The buffer really shouldn't be locked: only the current committing | |
297 | * transaction is allowed to write it, so nobody else is allowed | |
298 | * to do any IO. | |
299 | * | |
300 | * akpm: except if we're journalling data, and write() output is | |
301 | * also part of a shared mapping, and another thread has | |
302 | * decided to launch a writepage() against this buffer. | |
303 | */ | |
304 | J_ASSERT_BH(bh_in, buffer_jbddirty(bh_in)); | |
305 | ||
306 | new_bh = alloc_buffer_head(GFP_NOFS|__GFP_NOFAIL); | |
307 | ||
308 | /* | |
309 | * If a new transaction has already done a buffer copy-out, then | |
310 | * we use that version of the data for the commit. | |
311 | */ | |
312 | jbd_lock_bh_state(bh_in); | |
313 | repeat: | |
314 | if (jh_in->b_frozen_data) { | |
315 | done_copy_out = 1; | |
316 | new_page = virt_to_page(jh_in->b_frozen_data); | |
317 | new_offset = offset_in_page(jh_in->b_frozen_data); | |
318 | } else { | |
319 | new_page = jh2bh(jh_in)->b_page; | |
320 | new_offset = offset_in_page(jh2bh(jh_in)->b_data); | |
321 | } | |
322 | ||
323 | mapped_data = kmap_atomic(new_page, KM_USER0); | |
324 | /* | |
325 | * Check for escaping | |
326 | */ | |
327 | if (*((__be32 *)(mapped_data + new_offset)) == | |
f7f4bccb | 328 | cpu_to_be32(JBD2_MAGIC_NUMBER)) { |
470decc6 DK |
329 | need_copy_out = 1; |
330 | do_escape = 1; | |
331 | } | |
332 | kunmap_atomic(mapped_data, KM_USER0); | |
333 | ||
334 | /* | |
335 | * Do we need to do a data copy? | |
336 | */ | |
337 | if (need_copy_out && !done_copy_out) { | |
338 | char *tmp; | |
339 | ||
340 | jbd_unlock_bh_state(bh_in); | |
af1e76d6 | 341 | tmp = jbd2_alloc(bh_in->b_size, GFP_NOFS); |
470decc6 DK |
342 | jbd_lock_bh_state(bh_in); |
343 | if (jh_in->b_frozen_data) { | |
af1e76d6 | 344 | jbd2_free(tmp, bh_in->b_size); |
470decc6 DK |
345 | goto repeat; |
346 | } | |
347 | ||
348 | jh_in->b_frozen_data = tmp; | |
349 | mapped_data = kmap_atomic(new_page, KM_USER0); | |
350 | memcpy(tmp, mapped_data + new_offset, jh2bh(jh_in)->b_size); | |
351 | kunmap_atomic(mapped_data, KM_USER0); | |
352 | ||
353 | new_page = virt_to_page(tmp); | |
354 | new_offset = offset_in_page(tmp); | |
355 | done_copy_out = 1; | |
356 | } | |
357 | ||
358 | /* | |
359 | * Did we need to do an escaping? Now we've done all the | |
360 | * copying, we can finally do so. | |
361 | */ | |
362 | if (do_escape) { | |
363 | mapped_data = kmap_atomic(new_page, KM_USER0); | |
364 | *((unsigned int *)(mapped_data + new_offset)) = 0; | |
365 | kunmap_atomic(mapped_data, KM_USER0); | |
366 | } | |
367 | ||
368 | /* keep subsequent assertions sane */ | |
369 | new_bh->b_state = 0; | |
370 | init_buffer(new_bh, NULL, NULL); | |
371 | atomic_set(&new_bh->b_count, 1); | |
372 | jbd_unlock_bh_state(bh_in); | |
373 | ||
f7f4bccb | 374 | new_jh = jbd2_journal_add_journal_head(new_bh); /* This sleeps */ |
470decc6 DK |
375 | |
376 | set_bh_page(new_bh, new_page, new_offset); | |
377 | new_jh->b_transaction = NULL; | |
378 | new_bh->b_size = jh2bh(jh_in)->b_size; | |
379 | new_bh->b_bdev = transaction->t_journal->j_dev; | |
380 | new_bh->b_blocknr = blocknr; | |
381 | set_buffer_mapped(new_bh); | |
382 | set_buffer_dirty(new_bh); | |
383 | ||
384 | *jh_out = new_jh; | |
385 | ||
386 | /* | |
387 | * The to-be-written buffer needs to get moved to the io queue, | |
388 | * and the original buffer whose contents we are shadowing or | |
389 | * copying is moved to the transaction's shadow queue. | |
390 | */ | |
391 | JBUFFER_TRACE(jh_in, "file as BJ_Shadow"); | |
f7f4bccb | 392 | jbd2_journal_file_buffer(jh_in, transaction, BJ_Shadow); |
470decc6 | 393 | JBUFFER_TRACE(new_jh, "file as BJ_IO"); |
f7f4bccb | 394 | jbd2_journal_file_buffer(new_jh, transaction, BJ_IO); |
470decc6 DK |
395 | |
396 | return do_escape | (done_copy_out << 1); | |
397 | } | |
398 | ||
399 | /* | |
400 | * Allocation code for the journal file. Manage the space left in the | |
401 | * journal, so that we can begin checkpointing when appropriate. | |
402 | */ | |
403 | ||
404 | /* | |
f7f4bccb | 405 | * __jbd2_log_space_left: Return the number of free blocks left in the journal. |
470decc6 DK |
406 | * |
407 | * Called with the journal already locked. | |
408 | * | |
409 | * Called under j_state_lock | |
410 | */ | |
411 | ||
f7f4bccb | 412 | int __jbd2_log_space_left(journal_t *journal) |
470decc6 DK |
413 | { |
414 | int left = journal->j_free; | |
415 | ||
416 | assert_spin_locked(&journal->j_state_lock); | |
417 | ||
418 | /* | |
419 | * Be pessimistic here about the number of those free blocks which | |
420 | * might be required for log descriptor control blocks. | |
421 | */ | |
422 | ||
423 | #define MIN_LOG_RESERVED_BLOCKS 32 /* Allow for rounding errors */ | |
424 | ||
425 | left -= MIN_LOG_RESERVED_BLOCKS; | |
426 | ||
427 | if (left <= 0) | |
428 | return 0; | |
429 | left -= (left >> 3); | |
430 | return left; | |
431 | } | |
432 | ||
433 | /* | |
434 | * Called under j_state_lock. Returns true if a transaction was started. | |
435 | */ | |
f7f4bccb | 436 | int __jbd2_log_start_commit(journal_t *journal, tid_t target) |
470decc6 DK |
437 | { |
438 | /* | |
439 | * Are we already doing a recent enough commit? | |
440 | */ | |
441 | if (!tid_geq(journal->j_commit_request, target)) { | |
442 | /* | |
443 | * We want a new commit: OK, mark the request and wakup the | |
444 | * commit thread. We do _not_ do the commit ourselves. | |
445 | */ | |
446 | ||
447 | journal->j_commit_request = target; | |
448 | jbd_debug(1, "JBD: requesting commit %d/%d\n", | |
449 | journal->j_commit_request, | |
450 | journal->j_commit_sequence); | |
451 | wake_up(&journal->j_wait_commit); | |
452 | return 1; | |
453 | } | |
454 | return 0; | |
455 | } | |
456 | ||
f7f4bccb | 457 | int jbd2_log_start_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
458 | { |
459 | int ret; | |
460 | ||
461 | spin_lock(&journal->j_state_lock); | |
f7f4bccb | 462 | ret = __jbd2_log_start_commit(journal, tid); |
470decc6 DK |
463 | spin_unlock(&journal->j_state_lock); |
464 | return ret; | |
465 | } | |
466 | ||
467 | /* | |
468 | * Force and wait upon a commit if the calling process is not within | |
469 | * transaction. This is used for forcing out undo-protected data which contains | |
470 | * bitmaps, when the fs is running out of space. | |
471 | * | |
472 | * We can only force the running transaction if we don't have an active handle; | |
473 | * otherwise, we will deadlock. | |
474 | * | |
475 | * Returns true if a transaction was started. | |
476 | */ | |
f7f4bccb | 477 | int jbd2_journal_force_commit_nested(journal_t *journal) |
470decc6 DK |
478 | { |
479 | transaction_t *transaction = NULL; | |
480 | tid_t tid; | |
481 | ||
482 | spin_lock(&journal->j_state_lock); | |
483 | if (journal->j_running_transaction && !current->journal_info) { | |
484 | transaction = journal->j_running_transaction; | |
f7f4bccb | 485 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
486 | } else if (journal->j_committing_transaction) |
487 | transaction = journal->j_committing_transaction; | |
488 | ||
489 | if (!transaction) { | |
490 | spin_unlock(&journal->j_state_lock); | |
491 | return 0; /* Nothing to retry */ | |
492 | } | |
493 | ||
494 | tid = transaction->t_tid; | |
495 | spin_unlock(&journal->j_state_lock); | |
f7f4bccb | 496 | jbd2_log_wait_commit(journal, tid); |
470decc6 DK |
497 | return 1; |
498 | } | |
499 | ||
500 | /* | |
501 | * Start a commit of the current running transaction (if any). Returns true | |
502 | * if a transaction was started, and fills its tid in at *ptid | |
503 | */ | |
f7f4bccb | 504 | int jbd2_journal_start_commit(journal_t *journal, tid_t *ptid) |
470decc6 DK |
505 | { |
506 | int ret = 0; | |
507 | ||
508 | spin_lock(&journal->j_state_lock); | |
509 | if (journal->j_running_transaction) { | |
510 | tid_t tid = journal->j_running_transaction->t_tid; | |
511 | ||
f7f4bccb | 512 | ret = __jbd2_log_start_commit(journal, tid); |
470decc6 DK |
513 | if (ret && ptid) |
514 | *ptid = tid; | |
515 | } else if (journal->j_committing_transaction && ptid) { | |
516 | /* | |
517 | * If ext3_write_super() recently started a commit, then we | |
518 | * have to wait for completion of that transaction | |
519 | */ | |
520 | *ptid = journal->j_committing_transaction->t_tid; | |
521 | ret = 1; | |
522 | } | |
523 | spin_unlock(&journal->j_state_lock); | |
524 | return ret; | |
525 | } | |
526 | ||
527 | /* | |
528 | * Wait for a specified commit to complete. | |
529 | * The caller may not hold the journal lock. | |
530 | */ | |
f7f4bccb | 531 | int jbd2_log_wait_commit(journal_t *journal, tid_t tid) |
470decc6 DK |
532 | { |
533 | int err = 0; | |
534 | ||
e23291b9 | 535 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
536 | spin_lock(&journal->j_state_lock); |
537 | if (!tid_geq(journal->j_commit_request, tid)) { | |
538 | printk(KERN_EMERG | |
539 | "%s: error: j_commit_request=%d, tid=%d\n", | |
329d291f | 540 | __func__, journal->j_commit_request, tid); |
470decc6 DK |
541 | } |
542 | spin_unlock(&journal->j_state_lock); | |
543 | #endif | |
544 | spin_lock(&journal->j_state_lock); | |
545 | while (tid_gt(tid, journal->j_commit_sequence)) { | |
546 | jbd_debug(1, "JBD: want %d, j_commit_sequence=%d\n", | |
547 | tid, journal->j_commit_sequence); | |
548 | wake_up(&journal->j_wait_commit); | |
549 | spin_unlock(&journal->j_state_lock); | |
550 | wait_event(journal->j_wait_done_commit, | |
551 | !tid_gt(tid, journal->j_commit_sequence)); | |
552 | spin_lock(&journal->j_state_lock); | |
553 | } | |
554 | spin_unlock(&journal->j_state_lock); | |
555 | ||
556 | if (unlikely(is_journal_aborted(journal))) { | |
557 | printk(KERN_EMERG "journal commit I/O error\n"); | |
558 | err = -EIO; | |
559 | } | |
560 | return err; | |
561 | } | |
562 | ||
563 | /* | |
564 | * Log buffer allocation routines: | |
565 | */ | |
566 | ||
18eba7aa | 567 | int jbd2_journal_next_log_block(journal_t *journal, unsigned long long *retp) |
470decc6 DK |
568 | { |
569 | unsigned long blocknr; | |
570 | ||
571 | spin_lock(&journal->j_state_lock); | |
572 | J_ASSERT(journal->j_free > 1); | |
573 | ||
574 | blocknr = journal->j_head; | |
575 | journal->j_head++; | |
576 | journal->j_free--; | |
577 | if (journal->j_head == journal->j_last) | |
578 | journal->j_head = journal->j_first; | |
579 | spin_unlock(&journal->j_state_lock); | |
f7f4bccb | 580 | return jbd2_journal_bmap(journal, blocknr, retp); |
470decc6 DK |
581 | } |
582 | ||
583 | /* | |
584 | * Conversion of logical to physical block numbers for the journal | |
585 | * | |
586 | * On external journals the journal blocks are identity-mapped, so | |
587 | * this is a no-op. If needed, we can use j_blk_offset - everything is | |
588 | * ready. | |
589 | */ | |
f7f4bccb | 590 | int jbd2_journal_bmap(journal_t *journal, unsigned long blocknr, |
18eba7aa | 591 | unsigned long long *retp) |
470decc6 DK |
592 | { |
593 | int err = 0; | |
18eba7aa | 594 | unsigned long long ret; |
470decc6 DK |
595 | |
596 | if (journal->j_inode) { | |
597 | ret = bmap(journal->j_inode, blocknr); | |
598 | if (ret) | |
599 | *retp = ret; | |
600 | else { | |
470decc6 DK |
601 | printk(KERN_ALERT "%s: journal block not found " |
602 | "at offset %lu on %s\n", | |
05496769 | 603 | __func__, blocknr, journal->j_devname); |
470decc6 DK |
604 | err = -EIO; |
605 | __journal_abort_soft(journal, err); | |
606 | } | |
607 | } else { | |
608 | *retp = blocknr; /* +journal->j_blk_offset */ | |
609 | } | |
610 | return err; | |
611 | } | |
612 | ||
613 | /* | |
614 | * We play buffer_head aliasing tricks to write data/metadata blocks to | |
615 | * the journal without copying their contents, but for journal | |
616 | * descriptor blocks we do need to generate bona fide buffers. | |
617 | * | |
f7f4bccb | 618 | * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying |
470decc6 DK |
619 | * the buffer's contents they really should run flush_dcache_page(bh->b_page). |
620 | * But we don't bother doing that, so there will be coherency problems with | |
621 | * mmaps of blockdevs which hold live JBD-controlled filesystems. | |
622 | */ | |
f7f4bccb | 623 | struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal) |
470decc6 DK |
624 | { |
625 | struct buffer_head *bh; | |
18eba7aa | 626 | unsigned long long blocknr; |
470decc6 DK |
627 | int err; |
628 | ||
f7f4bccb | 629 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
630 | |
631 | if (err) | |
632 | return NULL; | |
633 | ||
634 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
635 | lock_buffer(bh); | |
636 | memset(bh->b_data, 0, journal->j_blocksize); | |
637 | set_buffer_uptodate(bh); | |
638 | unlock_buffer(bh); | |
639 | BUFFER_TRACE(bh, "return this buffer"); | |
f7f4bccb | 640 | return jbd2_journal_add_journal_head(bh); |
470decc6 DK |
641 | } |
642 | ||
8e85fb3f JL |
643 | struct jbd2_stats_proc_session { |
644 | journal_t *journal; | |
645 | struct transaction_stats_s *stats; | |
646 | int start; | |
647 | int max; | |
648 | }; | |
649 | ||
650 | static void *jbd2_history_skip_empty(struct jbd2_stats_proc_session *s, | |
651 | struct transaction_stats_s *ts, | |
652 | int first) | |
653 | { | |
654 | if (ts == s->stats + s->max) | |
655 | ts = s->stats; | |
656 | if (!first && ts == s->stats + s->start) | |
657 | return NULL; | |
658 | while (ts->ts_type == 0) { | |
659 | ts++; | |
660 | if (ts == s->stats + s->max) | |
661 | ts = s->stats; | |
662 | if (ts == s->stats + s->start) | |
663 | return NULL; | |
664 | } | |
665 | return ts; | |
666 | ||
667 | } | |
668 | ||
669 | static void *jbd2_seq_history_start(struct seq_file *seq, loff_t *pos) | |
670 | { | |
671 | struct jbd2_stats_proc_session *s = seq->private; | |
672 | struct transaction_stats_s *ts; | |
673 | int l = *pos; | |
674 | ||
675 | if (l == 0) | |
676 | return SEQ_START_TOKEN; | |
677 | ts = jbd2_history_skip_empty(s, s->stats + s->start, 1); | |
678 | if (!ts) | |
679 | return NULL; | |
680 | l--; | |
681 | while (l) { | |
682 | ts = jbd2_history_skip_empty(s, ++ts, 0); | |
683 | if (!ts) | |
684 | break; | |
685 | l--; | |
686 | } | |
687 | return ts; | |
688 | } | |
689 | ||
690 | static void *jbd2_seq_history_next(struct seq_file *seq, void *v, loff_t *pos) | |
691 | { | |
692 | struct jbd2_stats_proc_session *s = seq->private; | |
693 | struct transaction_stats_s *ts = v; | |
694 | ||
695 | ++*pos; | |
696 | if (v == SEQ_START_TOKEN) | |
697 | return jbd2_history_skip_empty(s, s->stats + s->start, 1); | |
698 | else | |
699 | return jbd2_history_skip_empty(s, ++ts, 0); | |
700 | } | |
701 | ||
702 | static int jbd2_seq_history_show(struct seq_file *seq, void *v) | |
703 | { | |
704 | struct transaction_stats_s *ts = v; | |
705 | if (v == SEQ_START_TOKEN) { | |
706 | seq_printf(seq, "%-4s %-5s %-5s %-5s %-5s %-5s %-5s %-6s %-5s " | |
707 | "%-5s %-5s %-5s %-5s %-5s\n", "R/C", "tid", | |
708 | "wait", "run", "lock", "flush", "log", "hndls", | |
709 | "block", "inlog", "ctime", "write", "drop", | |
710 | "close"); | |
711 | return 0; | |
712 | } | |
713 | if (ts->ts_type == JBD2_STATS_RUN) | |
714 | seq_printf(seq, "%-4s %-5lu %-5u %-5u %-5u %-5u %-5u " | |
715 | "%-6lu %-5lu %-5lu\n", "R", ts->ts_tid, | |
716 | jiffies_to_msecs(ts->u.run.rs_wait), | |
717 | jiffies_to_msecs(ts->u.run.rs_running), | |
718 | jiffies_to_msecs(ts->u.run.rs_locked), | |
719 | jiffies_to_msecs(ts->u.run.rs_flushing), | |
720 | jiffies_to_msecs(ts->u.run.rs_logging), | |
721 | ts->u.run.rs_handle_count, | |
722 | ts->u.run.rs_blocks, | |
723 | ts->u.run.rs_blocks_logged); | |
724 | else if (ts->ts_type == JBD2_STATS_CHECKPOINT) | |
725 | seq_printf(seq, "%-4s %-5lu %48s %-5u %-5lu %-5lu %-5lu\n", | |
726 | "C", ts->ts_tid, " ", | |
727 | jiffies_to_msecs(ts->u.chp.cs_chp_time), | |
728 | ts->u.chp.cs_written, ts->u.chp.cs_dropped, | |
729 | ts->u.chp.cs_forced_to_close); | |
730 | else | |
731 | J_ASSERT(0); | |
732 | return 0; | |
733 | } | |
734 | ||
735 | static void jbd2_seq_history_stop(struct seq_file *seq, void *v) | |
736 | { | |
737 | } | |
738 | ||
739 | static struct seq_operations jbd2_seq_history_ops = { | |
740 | .start = jbd2_seq_history_start, | |
741 | .next = jbd2_seq_history_next, | |
742 | .stop = jbd2_seq_history_stop, | |
743 | .show = jbd2_seq_history_show, | |
744 | }; | |
745 | ||
746 | static int jbd2_seq_history_open(struct inode *inode, struct file *file) | |
747 | { | |
748 | journal_t *journal = PDE(inode)->data; | |
749 | struct jbd2_stats_proc_session *s; | |
750 | int rc, size; | |
751 | ||
752 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
753 | if (s == NULL) | |
754 | return -ENOMEM; | |
755 | size = sizeof(struct transaction_stats_s) * journal->j_history_max; | |
756 | s->stats = kmalloc(size, GFP_KERNEL); | |
757 | if (s->stats == NULL) { | |
758 | kfree(s); | |
759 | return -ENOMEM; | |
760 | } | |
761 | spin_lock(&journal->j_history_lock); | |
762 | memcpy(s->stats, journal->j_history, size); | |
763 | s->max = journal->j_history_max; | |
764 | s->start = journal->j_history_cur % s->max; | |
765 | spin_unlock(&journal->j_history_lock); | |
766 | ||
767 | rc = seq_open(file, &jbd2_seq_history_ops); | |
768 | if (rc == 0) { | |
769 | struct seq_file *m = file->private_data; | |
770 | m->private = s; | |
771 | } else { | |
772 | kfree(s->stats); | |
773 | kfree(s); | |
774 | } | |
775 | return rc; | |
776 | ||
777 | } | |
778 | ||
779 | static int jbd2_seq_history_release(struct inode *inode, struct file *file) | |
780 | { | |
781 | struct seq_file *seq = file->private_data; | |
782 | struct jbd2_stats_proc_session *s = seq->private; | |
783 | ||
784 | kfree(s->stats); | |
785 | kfree(s); | |
786 | return seq_release(inode, file); | |
787 | } | |
788 | ||
789 | static struct file_operations jbd2_seq_history_fops = { | |
790 | .owner = THIS_MODULE, | |
791 | .open = jbd2_seq_history_open, | |
792 | .read = seq_read, | |
793 | .llseek = seq_lseek, | |
794 | .release = jbd2_seq_history_release, | |
795 | }; | |
796 | ||
797 | static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) | |
798 | { | |
799 | return *pos ? NULL : SEQ_START_TOKEN; | |
800 | } | |
801 | ||
802 | static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) | |
803 | { | |
804 | return NULL; | |
805 | } | |
806 | ||
807 | static int jbd2_seq_info_show(struct seq_file *seq, void *v) | |
808 | { | |
809 | struct jbd2_stats_proc_session *s = seq->private; | |
810 | ||
811 | if (v != SEQ_START_TOKEN) | |
812 | return 0; | |
813 | seq_printf(seq, "%lu transaction, each upto %u blocks\n", | |
814 | s->stats->ts_tid, | |
815 | s->journal->j_max_transaction_buffers); | |
816 | if (s->stats->ts_tid == 0) | |
817 | return 0; | |
818 | seq_printf(seq, "average: \n %ums waiting for transaction\n", | |
819 | jiffies_to_msecs(s->stats->u.run.rs_wait / s->stats->ts_tid)); | |
820 | seq_printf(seq, " %ums running transaction\n", | |
821 | jiffies_to_msecs(s->stats->u.run.rs_running / s->stats->ts_tid)); | |
822 | seq_printf(seq, " %ums transaction was being locked\n", | |
823 | jiffies_to_msecs(s->stats->u.run.rs_locked / s->stats->ts_tid)); | |
824 | seq_printf(seq, " %ums flushing data (in ordered mode)\n", | |
825 | jiffies_to_msecs(s->stats->u.run.rs_flushing / s->stats->ts_tid)); | |
826 | seq_printf(seq, " %ums logging transaction\n", | |
827 | jiffies_to_msecs(s->stats->u.run.rs_logging / s->stats->ts_tid)); | |
d7cfa468 TT |
828 | seq_printf(seq, " %luus average transaction commit time\n", |
829 | do_div(s->journal->j_average_commit_time, 1000)); | |
8e85fb3f JL |
830 | seq_printf(seq, " %lu handles per transaction\n", |
831 | s->stats->u.run.rs_handle_count / s->stats->ts_tid); | |
832 | seq_printf(seq, " %lu blocks per transaction\n", | |
833 | s->stats->u.run.rs_blocks / s->stats->ts_tid); | |
834 | seq_printf(seq, " %lu logged blocks per transaction\n", | |
835 | s->stats->u.run.rs_blocks_logged / s->stats->ts_tid); | |
836 | return 0; | |
837 | } | |
838 | ||
839 | static void jbd2_seq_info_stop(struct seq_file *seq, void *v) | |
840 | { | |
841 | } | |
842 | ||
843 | static struct seq_operations jbd2_seq_info_ops = { | |
844 | .start = jbd2_seq_info_start, | |
845 | .next = jbd2_seq_info_next, | |
846 | .stop = jbd2_seq_info_stop, | |
847 | .show = jbd2_seq_info_show, | |
848 | }; | |
849 | ||
850 | static int jbd2_seq_info_open(struct inode *inode, struct file *file) | |
851 | { | |
852 | journal_t *journal = PDE(inode)->data; | |
853 | struct jbd2_stats_proc_session *s; | |
854 | int rc, size; | |
855 | ||
856 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
857 | if (s == NULL) | |
858 | return -ENOMEM; | |
859 | size = sizeof(struct transaction_stats_s); | |
860 | s->stats = kmalloc(size, GFP_KERNEL); | |
861 | if (s->stats == NULL) { | |
862 | kfree(s); | |
863 | return -ENOMEM; | |
864 | } | |
865 | spin_lock(&journal->j_history_lock); | |
866 | memcpy(s->stats, &journal->j_stats, size); | |
867 | s->journal = journal; | |
868 | spin_unlock(&journal->j_history_lock); | |
869 | ||
870 | rc = seq_open(file, &jbd2_seq_info_ops); | |
871 | if (rc == 0) { | |
872 | struct seq_file *m = file->private_data; | |
873 | m->private = s; | |
874 | } else { | |
875 | kfree(s->stats); | |
876 | kfree(s); | |
877 | } | |
878 | return rc; | |
879 | ||
880 | } | |
881 | ||
882 | static int jbd2_seq_info_release(struct inode *inode, struct file *file) | |
883 | { | |
884 | struct seq_file *seq = file->private_data; | |
885 | struct jbd2_stats_proc_session *s = seq->private; | |
886 | kfree(s->stats); | |
887 | kfree(s); | |
888 | return seq_release(inode, file); | |
889 | } | |
890 | ||
891 | static struct file_operations jbd2_seq_info_fops = { | |
892 | .owner = THIS_MODULE, | |
893 | .open = jbd2_seq_info_open, | |
894 | .read = seq_read, | |
895 | .llseek = seq_lseek, | |
896 | .release = jbd2_seq_info_release, | |
897 | }; | |
898 | ||
899 | static struct proc_dir_entry *proc_jbd2_stats; | |
900 | ||
901 | static void jbd2_stats_proc_init(journal_t *journal) | |
902 | { | |
05496769 | 903 | journal->j_proc_entry = proc_mkdir(journal->j_devname, proc_jbd2_stats); |
8e85fb3f | 904 | if (journal->j_proc_entry) { |
79da3664 DL |
905 | proc_create_data("history", S_IRUGO, journal->j_proc_entry, |
906 | &jbd2_seq_history_fops, journal); | |
907 | proc_create_data("info", S_IRUGO, journal->j_proc_entry, | |
908 | &jbd2_seq_info_fops, journal); | |
8e85fb3f JL |
909 | } |
910 | } | |
911 | ||
912 | static void jbd2_stats_proc_exit(journal_t *journal) | |
913 | { | |
8e85fb3f JL |
914 | remove_proc_entry("info", journal->j_proc_entry); |
915 | remove_proc_entry("history", journal->j_proc_entry); | |
05496769 | 916 | remove_proc_entry(journal->j_devname, proc_jbd2_stats); |
8e85fb3f JL |
917 | } |
918 | ||
919 | static void journal_init_stats(journal_t *journal) | |
920 | { | |
921 | int size; | |
922 | ||
923 | if (!proc_jbd2_stats) | |
924 | return; | |
925 | ||
926 | journal->j_history_max = 100; | |
927 | size = sizeof(struct transaction_stats_s) * journal->j_history_max; | |
928 | journal->j_history = kzalloc(size, GFP_KERNEL); | |
929 | if (!journal->j_history) { | |
930 | journal->j_history_max = 0; | |
931 | return; | |
932 | } | |
933 | spin_lock_init(&journal->j_history_lock); | |
934 | } | |
935 | ||
470decc6 DK |
936 | /* |
937 | * Management for journal control blocks: functions to create and | |
938 | * destroy journal_t structures, and to initialise and read existing | |
939 | * journal blocks from disk. */ | |
940 | ||
941 | /* First: create and setup a journal_t object in memory. We initialise | |
942 | * very few fields yet: that has to wait until we have created the | |
943 | * journal structures from from scratch, or loaded them from disk. */ | |
944 | ||
945 | static journal_t * journal_init_common (void) | |
946 | { | |
947 | journal_t *journal; | |
948 | int err; | |
949 | ||
d802ffa8 | 950 | journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL); |
470decc6 DK |
951 | if (!journal) |
952 | goto fail; | |
470decc6 DK |
953 | |
954 | init_waitqueue_head(&journal->j_wait_transaction_locked); | |
955 | init_waitqueue_head(&journal->j_wait_logspace); | |
956 | init_waitqueue_head(&journal->j_wait_done_commit); | |
957 | init_waitqueue_head(&journal->j_wait_checkpoint); | |
958 | init_waitqueue_head(&journal->j_wait_commit); | |
959 | init_waitqueue_head(&journal->j_wait_updates); | |
960 | mutex_init(&journal->j_barrier); | |
961 | mutex_init(&journal->j_checkpoint_mutex); | |
962 | spin_lock_init(&journal->j_revoke_lock); | |
963 | spin_lock_init(&journal->j_list_lock); | |
964 | spin_lock_init(&journal->j_state_lock); | |
965 | ||
cd02ff0b | 966 | journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); |
30773840 TT |
967 | journal->j_min_batch_time = 0; |
968 | journal->j_max_batch_time = 15000; /* 15ms */ | |
470decc6 DK |
969 | |
970 | /* The journal is marked for error until we succeed with recovery! */ | |
f7f4bccb | 971 | journal->j_flags = JBD2_ABORT; |
470decc6 DK |
972 | |
973 | /* Set up a default-sized revoke table for the new mount. */ | |
f7f4bccb | 974 | err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); |
470decc6 DK |
975 | if (err) { |
976 | kfree(journal); | |
977 | goto fail; | |
978 | } | |
8e85fb3f JL |
979 | |
980 | journal_init_stats(journal); | |
981 | ||
470decc6 DK |
982 | return journal; |
983 | fail: | |
984 | return NULL; | |
985 | } | |
986 | ||
f7f4bccb | 987 | /* jbd2_journal_init_dev and jbd2_journal_init_inode: |
470decc6 DK |
988 | * |
989 | * Create a journal structure assigned some fixed set of disk blocks to | |
990 | * the journal. We don't actually touch those disk blocks yet, but we | |
991 | * need to set up all of the mapping information to tell the journaling | |
992 | * system where the journal blocks are. | |
993 | * | |
994 | */ | |
995 | ||
996 | /** | |
5648ba5b | 997 | * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure |
470decc6 DK |
998 | * @bdev: Block device on which to create the journal |
999 | * @fs_dev: Device which hold journalled filesystem for this journal. | |
1000 | * @start: Block nr Start of journal. | |
1001 | * @len: Length of the journal in blocks. | |
1002 | * @blocksize: blocksize of journalling device | |
5648ba5b RD |
1003 | * |
1004 | * Returns: a newly created journal_t * | |
470decc6 | 1005 | * |
f7f4bccb | 1006 | * jbd2_journal_init_dev creates a journal which maps a fixed contiguous |
470decc6 DK |
1007 | * range of blocks on an arbitrary block device. |
1008 | * | |
1009 | */ | |
f7f4bccb | 1010 | journal_t * jbd2_journal_init_dev(struct block_device *bdev, |
470decc6 | 1011 | struct block_device *fs_dev, |
18eba7aa | 1012 | unsigned long long start, int len, int blocksize) |
470decc6 DK |
1013 | { |
1014 | journal_t *journal = journal_init_common(); | |
1015 | struct buffer_head *bh; | |
05496769 | 1016 | char *p; |
470decc6 DK |
1017 | int n; |
1018 | ||
1019 | if (!journal) | |
1020 | return NULL; | |
1021 | ||
1022 | /* journal descriptor can store up to n blocks -bzzz */ | |
1023 | journal->j_blocksize = blocksize; | |
1024 | n = journal->j_blocksize / sizeof(journal_block_tag_t); | |
1025 | journal->j_wbufsize = n; | |
1026 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1027 | if (!journal->j_wbuf) { | |
1028 | printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n", | |
329d291f | 1029 | __func__); |
470decc6 DK |
1030 | kfree(journal); |
1031 | journal = NULL; | |
5eb30790 | 1032 | goto out; |
470decc6 DK |
1033 | } |
1034 | journal->j_dev = bdev; | |
1035 | journal->j_fs_dev = fs_dev; | |
1036 | journal->j_blk_offset = start; | |
1037 | journal->j_maxlen = len; | |
05496769 TT |
1038 | bdevname(journal->j_dev, journal->j_devname); |
1039 | p = journal->j_devname; | |
1040 | while ((p = strchr(p, '/'))) | |
1041 | *p = '!'; | |
8e85fb3f | 1042 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1043 | |
1044 | bh = __getblk(journal->j_dev, start, journal->j_blocksize); | |
1045 | J_ASSERT(bh != NULL); | |
1046 | journal->j_sb_buffer = bh; | |
1047 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
5eb30790 | 1048 | out: |
470decc6 DK |
1049 | return journal; |
1050 | } | |
1051 | ||
1052 | /** | |
f7f4bccb | 1053 | * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. |
470decc6 DK |
1054 | * @inode: An inode to create the journal in |
1055 | * | |
f7f4bccb | 1056 | * jbd2_journal_init_inode creates a journal which maps an on-disk inode as |
470decc6 DK |
1057 | * the journal. The inode must exist already, must support bmap() and |
1058 | * must have all data blocks preallocated. | |
1059 | */ | |
f7f4bccb | 1060 | journal_t * jbd2_journal_init_inode (struct inode *inode) |
470decc6 DK |
1061 | { |
1062 | struct buffer_head *bh; | |
1063 | journal_t *journal = journal_init_common(); | |
05496769 | 1064 | char *p; |
470decc6 DK |
1065 | int err; |
1066 | int n; | |
18eba7aa | 1067 | unsigned long long blocknr; |
470decc6 DK |
1068 | |
1069 | if (!journal) | |
1070 | return NULL; | |
1071 | ||
1072 | journal->j_dev = journal->j_fs_dev = inode->i_sb->s_bdev; | |
1073 | journal->j_inode = inode; | |
05496769 TT |
1074 | bdevname(journal->j_dev, journal->j_devname); |
1075 | p = journal->j_devname; | |
1076 | while ((p = strchr(p, '/'))) | |
1077 | *p = '!'; | |
1078 | p = journal->j_devname + strlen(journal->j_devname); | |
1079 | sprintf(p, ":%lu", journal->j_inode->i_ino); | |
470decc6 DK |
1080 | jbd_debug(1, |
1081 | "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n", | |
1082 | journal, inode->i_sb->s_id, inode->i_ino, | |
1083 | (long long) inode->i_size, | |
1084 | inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); | |
1085 | ||
1086 | journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits; | |
1087 | journal->j_blocksize = inode->i_sb->s_blocksize; | |
8e85fb3f | 1088 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1089 | |
1090 | /* journal descriptor can store up to n blocks -bzzz */ | |
1091 | n = journal->j_blocksize / sizeof(journal_block_tag_t); | |
1092 | journal->j_wbufsize = n; | |
1093 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1094 | if (!journal->j_wbuf) { | |
1095 | printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n", | |
329d291f | 1096 | __func__); |
2423840d | 1097 | jbd2_stats_proc_exit(journal); |
470decc6 DK |
1098 | kfree(journal); |
1099 | return NULL; | |
1100 | } | |
1101 | ||
f7f4bccb | 1102 | err = jbd2_journal_bmap(journal, 0, &blocknr); |
470decc6 DK |
1103 | /* If that failed, give up */ |
1104 | if (err) { | |
1105 | printk(KERN_ERR "%s: Cannnot locate journal superblock\n", | |
329d291f | 1106 | __func__); |
2423840d | 1107 | jbd2_stats_proc_exit(journal); |
470decc6 DK |
1108 | kfree(journal); |
1109 | return NULL; | |
1110 | } | |
1111 | ||
1112 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
1113 | J_ASSERT(bh != NULL); | |
1114 | journal->j_sb_buffer = bh; | |
1115 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
1116 | ||
1117 | return journal; | |
1118 | } | |
1119 | ||
1120 | /* | |
1121 | * If the journal init or create aborts, we need to mark the journal | |
1122 | * superblock as being NULL to prevent the journal destroy from writing | |
1123 | * back a bogus superblock. | |
1124 | */ | |
1125 | static void journal_fail_superblock (journal_t *journal) | |
1126 | { | |
1127 | struct buffer_head *bh = journal->j_sb_buffer; | |
1128 | brelse(bh); | |
1129 | journal->j_sb_buffer = NULL; | |
1130 | } | |
1131 | ||
1132 | /* | |
1133 | * Given a journal_t structure, initialise the various fields for | |
1134 | * startup of a new journaling session. We use this both when creating | |
1135 | * a journal, and after recovering an old journal to reset it for | |
1136 | * subsequent use. | |
1137 | */ | |
1138 | ||
1139 | static int journal_reset(journal_t *journal) | |
1140 | { | |
1141 | journal_superblock_t *sb = journal->j_superblock; | |
18eba7aa | 1142 | unsigned long long first, last; |
470decc6 DK |
1143 | |
1144 | first = be32_to_cpu(sb->s_first); | |
1145 | last = be32_to_cpu(sb->s_maxlen); | |
1146 | ||
1147 | journal->j_first = first; | |
1148 | journal->j_last = last; | |
1149 | ||
1150 | journal->j_head = first; | |
1151 | journal->j_tail = first; | |
1152 | journal->j_free = last - first; | |
1153 | ||
1154 | journal->j_tail_sequence = journal->j_transaction_sequence; | |
1155 | journal->j_commit_sequence = journal->j_transaction_sequence - 1; | |
1156 | journal->j_commit_request = journal->j_commit_sequence; | |
1157 | ||
1158 | journal->j_max_transaction_buffers = journal->j_maxlen / 4; | |
1159 | ||
1160 | /* Add the dynamic fields and write it to disk. */ | |
f7f4bccb | 1161 | jbd2_journal_update_superblock(journal, 1); |
97f06784 | 1162 | return jbd2_journal_start_thread(journal); |
470decc6 DK |
1163 | } |
1164 | ||
470decc6 | 1165 | /** |
f7f4bccb | 1166 | * void jbd2_journal_update_superblock() - Update journal sb on disk. |
470decc6 DK |
1167 | * @journal: The journal to update. |
1168 | * @wait: Set to '0' if you don't want to wait for IO completion. | |
1169 | * | |
1170 | * Update a journal's dynamic superblock fields and write it to disk, | |
1171 | * optionally waiting for the IO to complete. | |
1172 | */ | |
f7f4bccb | 1173 | void jbd2_journal_update_superblock(journal_t *journal, int wait) |
470decc6 DK |
1174 | { |
1175 | journal_superblock_t *sb = journal->j_superblock; | |
1176 | struct buffer_head *bh = journal->j_sb_buffer; | |
1177 | ||
1178 | /* | |
1179 | * As a special case, if the on-disk copy is already marked as needing | |
1180 | * no recovery (s_start == 0) and there are no outstanding transactions | |
1181 | * in the filesystem, then we can safely defer the superblock update | |
f7f4bccb | 1182 | * until the next commit by setting JBD2_FLUSHED. This avoids |
470decc6 DK |
1183 | * attempting a write to a potential-readonly device. |
1184 | */ | |
1185 | if (sb->s_start == 0 && journal->j_tail_sequence == | |
1186 | journal->j_transaction_sequence) { | |
1187 | jbd_debug(1,"JBD: Skipping superblock update on recovered sb " | |
1188 | "(start %ld, seq %d, errno %d)\n", | |
1189 | journal->j_tail, journal->j_tail_sequence, | |
1190 | journal->j_errno); | |
1191 | goto out; | |
1192 | } | |
1193 | ||
914258bf TT |
1194 | if (buffer_write_io_error(bh)) { |
1195 | /* | |
1196 | * Oh, dear. A previous attempt to write the journal | |
1197 | * superblock failed. This could happen because the | |
1198 | * USB device was yanked out. Or it could happen to | |
1199 | * be a transient write error and maybe the block will | |
1200 | * be remapped. Nothing we can do but to retry the | |
1201 | * write and hope for the best. | |
1202 | */ | |
1203 | printk(KERN_ERR "JBD2: previous I/O error detected " | |
1204 | "for journal superblock update for %s.\n", | |
1205 | journal->j_devname); | |
1206 | clear_buffer_write_io_error(bh); | |
1207 | set_buffer_uptodate(bh); | |
1208 | } | |
1209 | ||
470decc6 DK |
1210 | spin_lock(&journal->j_state_lock); |
1211 | jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n", | |
1212 | journal->j_tail, journal->j_tail_sequence, journal->j_errno); | |
1213 | ||
1214 | sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); | |
1215 | sb->s_start = cpu_to_be32(journal->j_tail); | |
1216 | sb->s_errno = cpu_to_be32(journal->j_errno); | |
1217 | spin_unlock(&journal->j_state_lock); | |
1218 | ||
1219 | BUFFER_TRACE(bh, "marking dirty"); | |
1220 | mark_buffer_dirty(bh); | |
914258bf | 1221 | if (wait) { |
470decc6 | 1222 | sync_dirty_buffer(bh); |
914258bf TT |
1223 | if (buffer_write_io_error(bh)) { |
1224 | printk(KERN_ERR "JBD2: I/O error detected " | |
1225 | "when updating journal superblock for %s.\n", | |
1226 | journal->j_devname); | |
1227 | clear_buffer_write_io_error(bh); | |
1228 | set_buffer_uptodate(bh); | |
1229 | } | |
1230 | } else | |
470decc6 DK |
1231 | ll_rw_block(SWRITE, 1, &bh); |
1232 | ||
1233 | out: | |
1234 | /* If we have just flushed the log (by marking s_start==0), then | |
1235 | * any future commit will have to be careful to update the | |
1236 | * superblock again to re-record the true start of the log. */ | |
1237 | ||
1238 | spin_lock(&journal->j_state_lock); | |
1239 | if (sb->s_start) | |
f7f4bccb | 1240 | journal->j_flags &= ~JBD2_FLUSHED; |
470decc6 | 1241 | else |
f7f4bccb | 1242 | journal->j_flags |= JBD2_FLUSHED; |
470decc6 DK |
1243 | spin_unlock(&journal->j_state_lock); |
1244 | } | |
1245 | ||
1246 | /* | |
1247 | * Read the superblock for a given journal, performing initial | |
1248 | * validation of the format. | |
1249 | */ | |
1250 | ||
1251 | static int journal_get_superblock(journal_t *journal) | |
1252 | { | |
1253 | struct buffer_head *bh; | |
1254 | journal_superblock_t *sb; | |
1255 | int err = -EIO; | |
1256 | ||
1257 | bh = journal->j_sb_buffer; | |
1258 | ||
1259 | J_ASSERT(bh != NULL); | |
1260 | if (!buffer_uptodate(bh)) { | |
1261 | ll_rw_block(READ, 1, &bh); | |
1262 | wait_on_buffer(bh); | |
1263 | if (!buffer_uptodate(bh)) { | |
1264 | printk (KERN_ERR | |
1265 | "JBD: IO error reading journal superblock\n"); | |
1266 | goto out; | |
1267 | } | |
1268 | } | |
1269 | ||
1270 | sb = journal->j_superblock; | |
1271 | ||
1272 | err = -EINVAL; | |
1273 | ||
f7f4bccb | 1274 | if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || |
470decc6 DK |
1275 | sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { |
1276 | printk(KERN_WARNING "JBD: no valid journal superblock found\n"); | |
1277 | goto out; | |
1278 | } | |
1279 | ||
1280 | switch(be32_to_cpu(sb->s_header.h_blocktype)) { | |
f7f4bccb | 1281 | case JBD2_SUPERBLOCK_V1: |
470decc6 DK |
1282 | journal->j_format_version = 1; |
1283 | break; | |
f7f4bccb | 1284 | case JBD2_SUPERBLOCK_V2: |
470decc6 DK |
1285 | journal->j_format_version = 2; |
1286 | break; | |
1287 | default: | |
1288 | printk(KERN_WARNING "JBD: unrecognised superblock format ID\n"); | |
1289 | goto out; | |
1290 | } | |
1291 | ||
1292 | if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen) | |
1293 | journal->j_maxlen = be32_to_cpu(sb->s_maxlen); | |
1294 | else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) { | |
1295 | printk (KERN_WARNING "JBD: journal file too short\n"); | |
1296 | goto out; | |
1297 | } | |
1298 | ||
1299 | return 0; | |
1300 | ||
1301 | out: | |
1302 | journal_fail_superblock(journal); | |
1303 | return err; | |
1304 | } | |
1305 | ||
1306 | /* | |
1307 | * Load the on-disk journal superblock and read the key fields into the | |
1308 | * journal_t. | |
1309 | */ | |
1310 | ||
1311 | static int load_superblock(journal_t *journal) | |
1312 | { | |
1313 | int err; | |
1314 | journal_superblock_t *sb; | |
1315 | ||
1316 | err = journal_get_superblock(journal); | |
1317 | if (err) | |
1318 | return err; | |
1319 | ||
1320 | sb = journal->j_superblock; | |
1321 | ||
1322 | journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); | |
1323 | journal->j_tail = be32_to_cpu(sb->s_start); | |
1324 | journal->j_first = be32_to_cpu(sb->s_first); | |
1325 | journal->j_last = be32_to_cpu(sb->s_maxlen); | |
1326 | journal->j_errno = be32_to_cpu(sb->s_errno); | |
1327 | ||
1328 | return 0; | |
1329 | } | |
1330 | ||
1331 | ||
1332 | /** | |
f7f4bccb | 1333 | * int jbd2_journal_load() - Read journal from disk. |
470decc6 DK |
1334 | * @journal: Journal to act on. |
1335 | * | |
1336 | * Given a journal_t structure which tells us which disk blocks contain | |
1337 | * a journal, read the journal from disk to initialise the in-memory | |
1338 | * structures. | |
1339 | */ | |
f7f4bccb | 1340 | int jbd2_journal_load(journal_t *journal) |
470decc6 DK |
1341 | { |
1342 | int err; | |
1343 | journal_superblock_t *sb; | |
1344 | ||
1345 | err = load_superblock(journal); | |
1346 | if (err) | |
1347 | return err; | |
1348 | ||
1349 | sb = journal->j_superblock; | |
1350 | /* If this is a V2 superblock, then we have to check the | |
1351 | * features flags on it. */ | |
1352 | ||
1353 | if (journal->j_format_version >= 2) { | |
1354 | if ((sb->s_feature_ro_compat & | |
f7f4bccb | 1355 | ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || |
470decc6 | 1356 | (sb->s_feature_incompat & |
f7f4bccb | 1357 | ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { |
470decc6 DK |
1358 | printk (KERN_WARNING |
1359 | "JBD: Unrecognised features on journal\n"); | |
1360 | return -EINVAL; | |
1361 | } | |
1362 | } | |
1363 | ||
470decc6 DK |
1364 | /* Let the recovery code check whether it needs to recover any |
1365 | * data from the journal. */ | |
f7f4bccb | 1366 | if (jbd2_journal_recover(journal)) |
470decc6 DK |
1367 | goto recovery_error; |
1368 | ||
1369 | /* OK, we've finished with the dynamic journal bits: | |
1370 | * reinitialise the dynamic contents of the superblock in memory | |
1371 | * and reset them on disk. */ | |
1372 | if (journal_reset(journal)) | |
1373 | goto recovery_error; | |
1374 | ||
f7f4bccb MC |
1375 | journal->j_flags &= ~JBD2_ABORT; |
1376 | journal->j_flags |= JBD2_LOADED; | |
470decc6 DK |
1377 | return 0; |
1378 | ||
1379 | recovery_error: | |
1380 | printk (KERN_WARNING "JBD: recovery failed\n"); | |
1381 | return -EIO; | |
1382 | } | |
1383 | ||
1384 | /** | |
f7f4bccb | 1385 | * void jbd2_journal_destroy() - Release a journal_t structure. |
470decc6 DK |
1386 | * @journal: Journal to act on. |
1387 | * | |
1388 | * Release a journal_t structure once it is no longer in use by the | |
1389 | * journaled object. | |
44519faf | 1390 | * Return <0 if we couldn't clean up the journal. |
470decc6 | 1391 | */ |
44519faf | 1392 | int jbd2_journal_destroy(journal_t *journal) |
470decc6 | 1393 | { |
44519faf HK |
1394 | int err = 0; |
1395 | ||
470decc6 DK |
1396 | /* Wait for the commit thread to wake up and die. */ |
1397 | journal_kill_thread(journal); | |
1398 | ||
1399 | /* Force a final log commit */ | |
1400 | if (journal->j_running_transaction) | |
f7f4bccb | 1401 | jbd2_journal_commit_transaction(journal); |
470decc6 DK |
1402 | |
1403 | /* Force any old transactions to disk */ | |
1404 | ||
1405 | /* Totally anal locking here... */ | |
1406 | spin_lock(&journal->j_list_lock); | |
1407 | while (journal->j_checkpoint_transactions != NULL) { | |
1408 | spin_unlock(&journal->j_list_lock); | |
1a0d3786 | 1409 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1410 | jbd2_log_do_checkpoint(journal); |
1a0d3786 | 1411 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
1412 | spin_lock(&journal->j_list_lock); |
1413 | } | |
1414 | ||
1415 | J_ASSERT(journal->j_running_transaction == NULL); | |
1416 | J_ASSERT(journal->j_committing_transaction == NULL); | |
1417 | J_ASSERT(journal->j_checkpoint_transactions == NULL); | |
1418 | spin_unlock(&journal->j_list_lock); | |
1419 | ||
470decc6 | 1420 | if (journal->j_sb_buffer) { |
44519faf HK |
1421 | if (!is_journal_aborted(journal)) { |
1422 | /* We can now mark the journal as empty. */ | |
1423 | journal->j_tail = 0; | |
1424 | journal->j_tail_sequence = | |
1425 | ++journal->j_transaction_sequence; | |
1426 | jbd2_journal_update_superblock(journal, 1); | |
1427 | } else { | |
1428 | err = -EIO; | |
1429 | } | |
470decc6 DK |
1430 | brelse(journal->j_sb_buffer); |
1431 | } | |
1432 | ||
8e85fb3f JL |
1433 | if (journal->j_proc_entry) |
1434 | jbd2_stats_proc_exit(journal); | |
470decc6 DK |
1435 | if (journal->j_inode) |
1436 | iput(journal->j_inode); | |
1437 | if (journal->j_revoke) | |
f7f4bccb | 1438 | jbd2_journal_destroy_revoke(journal); |
470decc6 DK |
1439 | kfree(journal->j_wbuf); |
1440 | kfree(journal); | |
44519faf HK |
1441 | |
1442 | return err; | |
470decc6 DK |
1443 | } |
1444 | ||
1445 | ||
1446 | /** | |
f7f4bccb | 1447 | *int jbd2_journal_check_used_features () - Check if features specified are used. |
470decc6 DK |
1448 | * @journal: Journal to check. |
1449 | * @compat: bitmask of compatible features | |
1450 | * @ro: bitmask of features that force read-only mount | |
1451 | * @incompat: bitmask of incompatible features | |
1452 | * | |
1453 | * Check whether the journal uses all of a given set of | |
1454 | * features. Return true (non-zero) if it does. | |
1455 | **/ | |
1456 | ||
f7f4bccb | 1457 | int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1458 | unsigned long ro, unsigned long incompat) |
1459 | { | |
1460 | journal_superblock_t *sb; | |
1461 | ||
1462 | if (!compat && !ro && !incompat) | |
1463 | return 1; | |
1464 | if (journal->j_format_version == 1) | |
1465 | return 0; | |
1466 | ||
1467 | sb = journal->j_superblock; | |
1468 | ||
1469 | if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && | |
1470 | ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && | |
1471 | ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) | |
1472 | return 1; | |
1473 | ||
1474 | return 0; | |
1475 | } | |
1476 | ||
1477 | /** | |
f7f4bccb | 1478 | * int jbd2_journal_check_available_features() - Check feature set in journalling layer |
470decc6 DK |
1479 | * @journal: Journal to check. |
1480 | * @compat: bitmask of compatible features | |
1481 | * @ro: bitmask of features that force read-only mount | |
1482 | * @incompat: bitmask of incompatible features | |
1483 | * | |
1484 | * Check whether the journaling code supports the use of | |
1485 | * all of a given set of features on this journal. Return true | |
1486 | * (non-zero) if it can. */ | |
1487 | ||
f7f4bccb | 1488 | int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1489 | unsigned long ro, unsigned long incompat) |
1490 | { | |
1491 | journal_superblock_t *sb; | |
1492 | ||
1493 | if (!compat && !ro && !incompat) | |
1494 | return 1; | |
1495 | ||
1496 | sb = journal->j_superblock; | |
1497 | ||
1498 | /* We can support any known requested features iff the | |
1499 | * superblock is in version 2. Otherwise we fail to support any | |
1500 | * extended sb features. */ | |
1501 | ||
1502 | if (journal->j_format_version != 2) | |
1503 | return 0; | |
1504 | ||
f7f4bccb MC |
1505 | if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && |
1506 | (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && | |
1507 | (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) | |
470decc6 DK |
1508 | return 1; |
1509 | ||
1510 | return 0; | |
1511 | } | |
1512 | ||
1513 | /** | |
f7f4bccb | 1514 | * int jbd2_journal_set_features () - Mark a given journal feature in the superblock |
470decc6 DK |
1515 | * @journal: Journal to act on. |
1516 | * @compat: bitmask of compatible features | |
1517 | * @ro: bitmask of features that force read-only mount | |
1518 | * @incompat: bitmask of incompatible features | |
1519 | * | |
1520 | * Mark a given journal feature as present on the | |
1521 | * superblock. Returns true if the requested features could be set. | |
1522 | * | |
1523 | */ | |
1524 | ||
f7f4bccb | 1525 | int jbd2_journal_set_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1526 | unsigned long ro, unsigned long incompat) |
1527 | { | |
1528 | journal_superblock_t *sb; | |
1529 | ||
f7f4bccb | 1530 | if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) |
470decc6 DK |
1531 | return 1; |
1532 | ||
f7f4bccb | 1533 | if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) |
470decc6 DK |
1534 | return 0; |
1535 | ||
1536 | jbd_debug(1, "Setting new features 0x%lx/0x%lx/0x%lx\n", | |
1537 | compat, ro, incompat); | |
1538 | ||
1539 | sb = journal->j_superblock; | |
1540 | ||
1541 | sb->s_feature_compat |= cpu_to_be32(compat); | |
1542 | sb->s_feature_ro_compat |= cpu_to_be32(ro); | |
1543 | sb->s_feature_incompat |= cpu_to_be32(incompat); | |
1544 | ||
1545 | return 1; | |
1546 | } | |
1547 | ||
818d276c GS |
1548 | /* |
1549 | * jbd2_journal_clear_features () - Clear a given journal feature in the | |
1550 | * superblock | |
1551 | * @journal: Journal to act on. | |
1552 | * @compat: bitmask of compatible features | |
1553 | * @ro: bitmask of features that force read-only mount | |
1554 | * @incompat: bitmask of incompatible features | |
1555 | * | |
1556 | * Clear a given journal feature as present on the | |
1557 | * superblock. | |
1558 | */ | |
1559 | void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, | |
1560 | unsigned long ro, unsigned long incompat) | |
1561 | { | |
1562 | journal_superblock_t *sb; | |
1563 | ||
1564 | jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", | |
1565 | compat, ro, incompat); | |
1566 | ||
1567 | sb = journal->j_superblock; | |
1568 | ||
1569 | sb->s_feature_compat &= ~cpu_to_be32(compat); | |
1570 | sb->s_feature_ro_compat &= ~cpu_to_be32(ro); | |
1571 | sb->s_feature_incompat &= ~cpu_to_be32(incompat); | |
1572 | } | |
1573 | EXPORT_SYMBOL(jbd2_journal_clear_features); | |
470decc6 DK |
1574 | |
1575 | /** | |
f7f4bccb | 1576 | * int jbd2_journal_update_format () - Update on-disk journal structure. |
470decc6 DK |
1577 | * @journal: Journal to act on. |
1578 | * | |
1579 | * Given an initialised but unloaded journal struct, poke about in the | |
1580 | * on-disk structure to update it to the most recent supported version. | |
1581 | */ | |
f7f4bccb | 1582 | int jbd2_journal_update_format (journal_t *journal) |
470decc6 DK |
1583 | { |
1584 | journal_superblock_t *sb; | |
1585 | int err; | |
1586 | ||
1587 | err = journal_get_superblock(journal); | |
1588 | if (err) | |
1589 | return err; | |
1590 | ||
1591 | sb = journal->j_superblock; | |
1592 | ||
1593 | switch (be32_to_cpu(sb->s_header.h_blocktype)) { | |
f7f4bccb | 1594 | case JBD2_SUPERBLOCK_V2: |
470decc6 | 1595 | return 0; |
f7f4bccb | 1596 | case JBD2_SUPERBLOCK_V1: |
470decc6 DK |
1597 | return journal_convert_superblock_v1(journal, sb); |
1598 | default: | |
1599 | break; | |
1600 | } | |
1601 | return -EINVAL; | |
1602 | } | |
1603 | ||
1604 | static int journal_convert_superblock_v1(journal_t *journal, | |
1605 | journal_superblock_t *sb) | |
1606 | { | |
1607 | int offset, blocksize; | |
1608 | struct buffer_head *bh; | |
1609 | ||
1610 | printk(KERN_WARNING | |
1611 | "JBD: Converting superblock from version 1 to 2.\n"); | |
1612 | ||
1613 | /* Pre-initialise new fields to zero */ | |
1614 | offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb); | |
1615 | blocksize = be32_to_cpu(sb->s_blocksize); | |
1616 | memset(&sb->s_feature_compat, 0, blocksize-offset); | |
1617 | ||
1618 | sb->s_nr_users = cpu_to_be32(1); | |
f7f4bccb | 1619 | sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2); |
470decc6 DK |
1620 | journal->j_format_version = 2; |
1621 | ||
1622 | bh = journal->j_sb_buffer; | |
1623 | BUFFER_TRACE(bh, "marking dirty"); | |
1624 | mark_buffer_dirty(bh); | |
1625 | sync_dirty_buffer(bh); | |
1626 | return 0; | |
1627 | } | |
1628 | ||
1629 | ||
1630 | /** | |
f7f4bccb | 1631 | * int jbd2_journal_flush () - Flush journal |
470decc6 DK |
1632 | * @journal: Journal to act on. |
1633 | * | |
1634 | * Flush all data for a given journal to disk and empty the journal. | |
1635 | * Filesystems can use this when remounting readonly to ensure that | |
1636 | * recovery does not need to happen on remount. | |
1637 | */ | |
1638 | ||
f7f4bccb | 1639 | int jbd2_journal_flush(journal_t *journal) |
470decc6 DK |
1640 | { |
1641 | int err = 0; | |
1642 | transaction_t *transaction = NULL; | |
1643 | unsigned long old_tail; | |
1644 | ||
1645 | spin_lock(&journal->j_state_lock); | |
1646 | ||
1647 | /* Force everything buffered to the log... */ | |
1648 | if (journal->j_running_transaction) { | |
1649 | transaction = journal->j_running_transaction; | |
f7f4bccb | 1650 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1651 | } else if (journal->j_committing_transaction) |
1652 | transaction = journal->j_committing_transaction; | |
1653 | ||
1654 | /* Wait for the log commit to complete... */ | |
1655 | if (transaction) { | |
1656 | tid_t tid = transaction->t_tid; | |
1657 | ||
1658 | spin_unlock(&journal->j_state_lock); | |
f7f4bccb | 1659 | jbd2_log_wait_commit(journal, tid); |
470decc6 DK |
1660 | } else { |
1661 | spin_unlock(&journal->j_state_lock); | |
1662 | } | |
1663 | ||
1664 | /* ...and flush everything in the log out to disk. */ | |
1665 | spin_lock(&journal->j_list_lock); | |
1666 | while (!err && journal->j_checkpoint_transactions != NULL) { | |
1667 | spin_unlock(&journal->j_list_lock); | |
44519faf | 1668 | mutex_lock(&journal->j_checkpoint_mutex); |
f7f4bccb | 1669 | err = jbd2_log_do_checkpoint(journal); |
44519faf | 1670 | mutex_unlock(&journal->j_checkpoint_mutex); |
470decc6 DK |
1671 | spin_lock(&journal->j_list_lock); |
1672 | } | |
1673 | spin_unlock(&journal->j_list_lock); | |
44519faf HK |
1674 | |
1675 | if (is_journal_aborted(journal)) | |
1676 | return -EIO; | |
1677 | ||
f7f4bccb | 1678 | jbd2_cleanup_journal_tail(journal); |
470decc6 DK |
1679 | |
1680 | /* Finally, mark the journal as really needing no recovery. | |
1681 | * This sets s_start==0 in the underlying superblock, which is | |
1682 | * the magic code for a fully-recovered superblock. Any future | |
1683 | * commits of data to the journal will restore the current | |
1684 | * s_start value. */ | |
1685 | spin_lock(&journal->j_state_lock); | |
1686 | old_tail = journal->j_tail; | |
1687 | journal->j_tail = 0; | |
1688 | spin_unlock(&journal->j_state_lock); | |
f7f4bccb | 1689 | jbd2_journal_update_superblock(journal, 1); |
470decc6 DK |
1690 | spin_lock(&journal->j_state_lock); |
1691 | journal->j_tail = old_tail; | |
1692 | ||
1693 | J_ASSERT(!journal->j_running_transaction); | |
1694 | J_ASSERT(!journal->j_committing_transaction); | |
1695 | J_ASSERT(!journal->j_checkpoint_transactions); | |
1696 | J_ASSERT(journal->j_head == journal->j_tail); | |
1697 | J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); | |
1698 | spin_unlock(&journal->j_state_lock); | |
44519faf | 1699 | return 0; |
470decc6 DK |
1700 | } |
1701 | ||
1702 | /** | |
f7f4bccb | 1703 | * int jbd2_journal_wipe() - Wipe journal contents |
470decc6 DK |
1704 | * @journal: Journal to act on. |
1705 | * @write: flag (see below) | |
1706 | * | |
1707 | * Wipe out all of the contents of a journal, safely. This will produce | |
1708 | * a warning if the journal contains any valid recovery information. | |
f7f4bccb | 1709 | * Must be called between journal_init_*() and jbd2_journal_load(). |
470decc6 DK |
1710 | * |
1711 | * If 'write' is non-zero, then we wipe out the journal on disk; otherwise | |
1712 | * we merely suppress recovery. | |
1713 | */ | |
1714 | ||
f7f4bccb | 1715 | int jbd2_journal_wipe(journal_t *journal, int write) |
470decc6 DK |
1716 | { |
1717 | journal_superblock_t *sb; | |
1718 | int err = 0; | |
1719 | ||
f7f4bccb | 1720 | J_ASSERT (!(journal->j_flags & JBD2_LOADED)); |
470decc6 DK |
1721 | |
1722 | err = load_superblock(journal); | |
1723 | if (err) | |
1724 | return err; | |
1725 | ||
1726 | sb = journal->j_superblock; | |
1727 | ||
1728 | if (!journal->j_tail) | |
1729 | goto no_recovery; | |
1730 | ||
1731 | printk (KERN_WARNING "JBD: %s recovery information on journal\n", | |
1732 | write ? "Clearing" : "Ignoring"); | |
1733 | ||
f7f4bccb | 1734 | err = jbd2_journal_skip_recovery(journal); |
470decc6 | 1735 | if (write) |
f7f4bccb | 1736 | jbd2_journal_update_superblock(journal, 1); |
470decc6 DK |
1737 | |
1738 | no_recovery: | |
1739 | return err; | |
1740 | } | |
1741 | ||
470decc6 DK |
1742 | /* |
1743 | * Journal abort has very specific semantics, which we describe | |
1744 | * for journal abort. | |
1745 | * | |
1746 | * Two internal function, which provide abort to te jbd layer | |
1747 | * itself are here. | |
1748 | */ | |
1749 | ||
1750 | /* | |
1751 | * Quick version for internal journal use (doesn't lock the journal). | |
1752 | * Aborts hard --- we mark the abort as occurred, but do _nothing_ else, | |
1753 | * and don't attempt to make any other journal updates. | |
1754 | */ | |
f7f4bccb | 1755 | void __jbd2_journal_abort_hard(journal_t *journal) |
470decc6 DK |
1756 | { |
1757 | transaction_t *transaction; | |
470decc6 | 1758 | |
f7f4bccb | 1759 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1760 | return; |
1761 | ||
1762 | printk(KERN_ERR "Aborting journal on device %s.\n", | |
05496769 | 1763 | journal->j_devname); |
470decc6 DK |
1764 | |
1765 | spin_lock(&journal->j_state_lock); | |
f7f4bccb | 1766 | journal->j_flags |= JBD2_ABORT; |
470decc6 DK |
1767 | transaction = journal->j_running_transaction; |
1768 | if (transaction) | |
f7f4bccb | 1769 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1770 | spin_unlock(&journal->j_state_lock); |
1771 | } | |
1772 | ||
1773 | /* Soft abort: record the abort error status in the journal superblock, | |
1774 | * but don't do any other IO. */ | |
1775 | static void __journal_abort_soft (journal_t *journal, int errno) | |
1776 | { | |
f7f4bccb | 1777 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1778 | return; |
1779 | ||
1780 | if (!journal->j_errno) | |
1781 | journal->j_errno = errno; | |
1782 | ||
f7f4bccb | 1783 | __jbd2_journal_abort_hard(journal); |
470decc6 DK |
1784 | |
1785 | if (errno) | |
f7f4bccb | 1786 | jbd2_journal_update_superblock(journal, 1); |
470decc6 DK |
1787 | } |
1788 | ||
1789 | /** | |
f7f4bccb | 1790 | * void jbd2_journal_abort () - Shutdown the journal immediately. |
470decc6 DK |
1791 | * @journal: the journal to shutdown. |
1792 | * @errno: an error number to record in the journal indicating | |
1793 | * the reason for the shutdown. | |
1794 | * | |
1795 | * Perform a complete, immediate shutdown of the ENTIRE | |
1796 | * journal (not of a single transaction). This operation cannot be | |
1797 | * undone without closing and reopening the journal. | |
1798 | * | |
f7f4bccb | 1799 | * The jbd2_journal_abort function is intended to support higher level error |
470decc6 DK |
1800 | * recovery mechanisms such as the ext2/ext3 remount-readonly error |
1801 | * mode. | |
1802 | * | |
1803 | * Journal abort has very specific semantics. Any existing dirty, | |
1804 | * unjournaled buffers in the main filesystem will still be written to | |
1805 | * disk by bdflush, but the journaling mechanism will be suspended | |
1806 | * immediately and no further transaction commits will be honoured. | |
1807 | * | |
1808 | * Any dirty, journaled buffers will be written back to disk without | |
1809 | * hitting the journal. Atomicity cannot be guaranteed on an aborted | |
1810 | * filesystem, but we _do_ attempt to leave as much data as possible | |
1811 | * behind for fsck to use for cleanup. | |
1812 | * | |
1813 | * Any attempt to get a new transaction handle on a journal which is in | |
1814 | * ABORT state will just result in an -EROFS error return. A | |
f7f4bccb | 1815 | * jbd2_journal_stop on an existing handle will return -EIO if we have |
470decc6 DK |
1816 | * entered abort state during the update. |
1817 | * | |
1818 | * Recursive transactions are not disturbed by journal abort until the | |
f7f4bccb | 1819 | * final jbd2_journal_stop, which will receive the -EIO error. |
470decc6 | 1820 | * |
f7f4bccb | 1821 | * Finally, the jbd2_journal_abort call allows the caller to supply an errno |
470decc6 DK |
1822 | * which will be recorded (if possible) in the journal superblock. This |
1823 | * allows a client to record failure conditions in the middle of a | |
1824 | * transaction without having to complete the transaction to record the | |
1825 | * failure to disk. ext3_error, for example, now uses this | |
1826 | * functionality. | |
1827 | * | |
1828 | * Errors which originate from within the journaling layer will NOT | |
1829 | * supply an errno; a null errno implies that absolutely no further | |
1830 | * writes are done to the journal (unless there are any already in | |
1831 | * progress). | |
1832 | * | |
1833 | */ | |
1834 | ||
f7f4bccb | 1835 | void jbd2_journal_abort(journal_t *journal, int errno) |
470decc6 DK |
1836 | { |
1837 | __journal_abort_soft(journal, errno); | |
1838 | } | |
1839 | ||
1840 | /** | |
f7f4bccb | 1841 | * int jbd2_journal_errno () - returns the journal's error state. |
470decc6 DK |
1842 | * @journal: journal to examine. |
1843 | * | |
f7f4bccb | 1844 | * This is the errno numbet set with jbd2_journal_abort(), the last |
470decc6 DK |
1845 | * time the journal was mounted - if the journal was stopped |
1846 | * without calling abort this will be 0. | |
1847 | * | |
1848 | * If the journal has been aborted on this mount time -EROFS will | |
1849 | * be returned. | |
1850 | */ | |
f7f4bccb | 1851 | int jbd2_journal_errno(journal_t *journal) |
470decc6 DK |
1852 | { |
1853 | int err; | |
1854 | ||
1855 | spin_lock(&journal->j_state_lock); | |
f7f4bccb | 1856 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1857 | err = -EROFS; |
1858 | else | |
1859 | err = journal->j_errno; | |
1860 | spin_unlock(&journal->j_state_lock); | |
1861 | return err; | |
1862 | } | |
1863 | ||
1864 | /** | |
f7f4bccb | 1865 | * int jbd2_journal_clear_err () - clears the journal's error state |
470decc6 DK |
1866 | * @journal: journal to act on. |
1867 | * | |
1868 | * An error must be cleared or Acked to take a FS out of readonly | |
1869 | * mode. | |
1870 | */ | |
f7f4bccb | 1871 | int jbd2_journal_clear_err(journal_t *journal) |
470decc6 DK |
1872 | { |
1873 | int err = 0; | |
1874 | ||
1875 | spin_lock(&journal->j_state_lock); | |
f7f4bccb | 1876 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1877 | err = -EROFS; |
1878 | else | |
1879 | journal->j_errno = 0; | |
1880 | spin_unlock(&journal->j_state_lock); | |
1881 | return err; | |
1882 | } | |
1883 | ||
1884 | /** | |
f7f4bccb | 1885 | * void jbd2_journal_ack_err() - Ack journal err. |
470decc6 DK |
1886 | * @journal: journal to act on. |
1887 | * | |
1888 | * An error must be cleared or Acked to take a FS out of readonly | |
1889 | * mode. | |
1890 | */ | |
f7f4bccb | 1891 | void jbd2_journal_ack_err(journal_t *journal) |
470decc6 DK |
1892 | { |
1893 | spin_lock(&journal->j_state_lock); | |
1894 | if (journal->j_errno) | |
f7f4bccb | 1895 | journal->j_flags |= JBD2_ACK_ERR; |
470decc6 DK |
1896 | spin_unlock(&journal->j_state_lock); |
1897 | } | |
1898 | ||
f7f4bccb | 1899 | int jbd2_journal_blocks_per_page(struct inode *inode) |
470decc6 DK |
1900 | { |
1901 | return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
1902 | } | |
1903 | ||
b517bea1 ZB |
1904 | /* |
1905 | * helper functions to deal with 32 or 64bit block numbers. | |
1906 | */ | |
1907 | size_t journal_tag_bytes(journal_t *journal) | |
1908 | { | |
1909 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) | |
cd02ff0b | 1910 | return JBD2_TAG_SIZE64; |
b517bea1 | 1911 | else |
cd02ff0b | 1912 | return JBD2_TAG_SIZE32; |
b517bea1 ZB |
1913 | } |
1914 | ||
470decc6 DK |
1915 | /* |
1916 | * Journal_head storage management | |
1917 | */ | |
e18b890b | 1918 | static struct kmem_cache *jbd2_journal_head_cache; |
e23291b9 | 1919 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
1920 | static atomic_t nr_journal_heads = ATOMIC_INIT(0); |
1921 | #endif | |
1922 | ||
f7f4bccb | 1923 | static int journal_init_jbd2_journal_head_cache(void) |
470decc6 DK |
1924 | { |
1925 | int retval; | |
1926 | ||
1076d17a | 1927 | J_ASSERT(jbd2_journal_head_cache == NULL); |
a920e941 | 1928 | jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", |
470decc6 DK |
1929 | sizeof(struct journal_head), |
1930 | 0, /* offset */ | |
77160957 | 1931 | SLAB_TEMPORARY, /* flags */ |
20c2df83 | 1932 | NULL); /* ctor */ |
470decc6 | 1933 | retval = 0; |
1076d17a | 1934 | if (!jbd2_journal_head_cache) { |
470decc6 DK |
1935 | retval = -ENOMEM; |
1936 | printk(KERN_EMERG "JBD: no memory for journal_head cache\n"); | |
1937 | } | |
1938 | return retval; | |
1939 | } | |
1940 | ||
f7f4bccb | 1941 | static void jbd2_journal_destroy_jbd2_journal_head_cache(void) |
470decc6 | 1942 | { |
8a9362eb DG |
1943 | if (jbd2_journal_head_cache) { |
1944 | kmem_cache_destroy(jbd2_journal_head_cache); | |
1945 | jbd2_journal_head_cache = NULL; | |
1946 | } | |
470decc6 DK |
1947 | } |
1948 | ||
1949 | /* | |
1950 | * journal_head splicing and dicing | |
1951 | */ | |
1952 | static struct journal_head *journal_alloc_journal_head(void) | |
1953 | { | |
1954 | struct journal_head *ret; | |
1955 | static unsigned long last_warning; | |
1956 | ||
e23291b9 | 1957 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
1958 | atomic_inc(&nr_journal_heads); |
1959 | #endif | |
f7f4bccb | 1960 | ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS); |
1076d17a | 1961 | if (!ret) { |
470decc6 DK |
1962 | jbd_debug(1, "out of memory for journal_head\n"); |
1963 | if (time_after(jiffies, last_warning + 5*HZ)) { | |
1964 | printk(KERN_NOTICE "ENOMEM in %s, retrying.\n", | |
329d291f | 1965 | __func__); |
470decc6 DK |
1966 | last_warning = jiffies; |
1967 | } | |
1076d17a | 1968 | while (!ret) { |
470decc6 | 1969 | yield(); |
f7f4bccb | 1970 | ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS); |
470decc6 DK |
1971 | } |
1972 | } | |
1973 | return ret; | |
1974 | } | |
1975 | ||
1976 | static void journal_free_journal_head(struct journal_head *jh) | |
1977 | { | |
e23291b9 | 1978 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 | 1979 | atomic_dec(&nr_journal_heads); |
cd02ff0b | 1980 | memset(jh, JBD2_POISON_FREE, sizeof(*jh)); |
470decc6 | 1981 | #endif |
f7f4bccb | 1982 | kmem_cache_free(jbd2_journal_head_cache, jh); |
470decc6 DK |
1983 | } |
1984 | ||
1985 | /* | |
1986 | * A journal_head is attached to a buffer_head whenever JBD has an | |
1987 | * interest in the buffer. | |
1988 | * | |
1989 | * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit | |
1990 | * is set. This bit is tested in core kernel code where we need to take | |
1991 | * JBD-specific actions. Testing the zeroness of ->b_private is not reliable | |
1992 | * there. | |
1993 | * | |
1994 | * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. | |
1995 | * | |
1996 | * When a buffer has its BH_JBD bit set it is immune from being released by | |
1997 | * core kernel code, mainly via ->b_count. | |
1998 | * | |
1999 | * A journal_head may be detached from its buffer_head when the journal_head's | |
2000 | * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL. | |
f7f4bccb | 2001 | * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the |
470decc6 DK |
2002 | * journal_head can be dropped if needed. |
2003 | * | |
2004 | * Various places in the kernel want to attach a journal_head to a buffer_head | |
2005 | * _before_ attaching the journal_head to a transaction. To protect the | |
f7f4bccb | 2006 | * journal_head in this situation, jbd2_journal_add_journal_head elevates the |
470decc6 | 2007 | * journal_head's b_jcount refcount by one. The caller must call |
f7f4bccb | 2008 | * jbd2_journal_put_journal_head() to undo this. |
470decc6 DK |
2009 | * |
2010 | * So the typical usage would be: | |
2011 | * | |
2012 | * (Attach a journal_head if needed. Increments b_jcount) | |
f7f4bccb | 2013 | * struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
2014 | * ... |
2015 | * jh->b_transaction = xxx; | |
f7f4bccb | 2016 | * jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2017 | * |
2018 | * Now, the journal_head's b_jcount is zero, but it is safe from being released | |
2019 | * because it has a non-zero b_transaction. | |
2020 | */ | |
2021 | ||
2022 | /* | |
2023 | * Give a buffer_head a journal_head. | |
2024 | * | |
2025 | * Doesn't need the journal lock. | |
2026 | * May sleep. | |
2027 | */ | |
f7f4bccb | 2028 | struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) |
470decc6 DK |
2029 | { |
2030 | struct journal_head *jh; | |
2031 | struct journal_head *new_jh = NULL; | |
2032 | ||
2033 | repeat: | |
2034 | if (!buffer_jbd(bh)) { | |
2035 | new_jh = journal_alloc_journal_head(); | |
2036 | memset(new_jh, 0, sizeof(*new_jh)); | |
2037 | } | |
2038 | ||
2039 | jbd_lock_bh_journal_head(bh); | |
2040 | if (buffer_jbd(bh)) { | |
2041 | jh = bh2jh(bh); | |
2042 | } else { | |
2043 | J_ASSERT_BH(bh, | |
2044 | (atomic_read(&bh->b_count) > 0) || | |
2045 | (bh->b_page && bh->b_page->mapping)); | |
2046 | ||
2047 | if (!new_jh) { | |
2048 | jbd_unlock_bh_journal_head(bh); | |
2049 | goto repeat; | |
2050 | } | |
2051 | ||
2052 | jh = new_jh; | |
2053 | new_jh = NULL; /* We consumed it */ | |
2054 | set_buffer_jbd(bh); | |
2055 | bh->b_private = jh; | |
2056 | jh->b_bh = bh; | |
2057 | get_bh(bh); | |
2058 | BUFFER_TRACE(bh, "added journal_head"); | |
2059 | } | |
2060 | jh->b_jcount++; | |
2061 | jbd_unlock_bh_journal_head(bh); | |
2062 | if (new_jh) | |
2063 | journal_free_journal_head(new_jh); | |
2064 | return bh->b_private; | |
2065 | } | |
2066 | ||
2067 | /* | |
2068 | * Grab a ref against this buffer_head's journal_head. If it ended up not | |
2069 | * having a journal_head, return NULL | |
2070 | */ | |
f7f4bccb | 2071 | struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) |
470decc6 DK |
2072 | { |
2073 | struct journal_head *jh = NULL; | |
2074 | ||
2075 | jbd_lock_bh_journal_head(bh); | |
2076 | if (buffer_jbd(bh)) { | |
2077 | jh = bh2jh(bh); | |
2078 | jh->b_jcount++; | |
2079 | } | |
2080 | jbd_unlock_bh_journal_head(bh); | |
2081 | return jh; | |
2082 | } | |
2083 | ||
2084 | static void __journal_remove_journal_head(struct buffer_head *bh) | |
2085 | { | |
2086 | struct journal_head *jh = bh2jh(bh); | |
2087 | ||
2088 | J_ASSERT_JH(jh, jh->b_jcount >= 0); | |
2089 | ||
2090 | get_bh(bh); | |
2091 | if (jh->b_jcount == 0) { | |
2092 | if (jh->b_transaction == NULL && | |
2093 | jh->b_next_transaction == NULL && | |
2094 | jh->b_cp_transaction == NULL) { | |
2095 | J_ASSERT_JH(jh, jh->b_jlist == BJ_None); | |
2096 | J_ASSERT_BH(bh, buffer_jbd(bh)); | |
2097 | J_ASSERT_BH(bh, jh2bh(jh) == bh); | |
2098 | BUFFER_TRACE(bh, "remove journal_head"); | |
2099 | if (jh->b_frozen_data) { | |
2100 | printk(KERN_WARNING "%s: freeing " | |
2101 | "b_frozen_data\n", | |
329d291f | 2102 | __func__); |
af1e76d6 | 2103 | jbd2_free(jh->b_frozen_data, bh->b_size); |
470decc6 DK |
2104 | } |
2105 | if (jh->b_committed_data) { | |
2106 | printk(KERN_WARNING "%s: freeing " | |
2107 | "b_committed_data\n", | |
329d291f | 2108 | __func__); |
af1e76d6 | 2109 | jbd2_free(jh->b_committed_data, bh->b_size); |
470decc6 DK |
2110 | } |
2111 | bh->b_private = NULL; | |
2112 | jh->b_bh = NULL; /* debug, really */ | |
2113 | clear_buffer_jbd(bh); | |
2114 | __brelse(bh); | |
2115 | journal_free_journal_head(jh); | |
2116 | } else { | |
2117 | BUFFER_TRACE(bh, "journal_head was locked"); | |
2118 | } | |
2119 | } | |
2120 | } | |
2121 | ||
2122 | /* | |
f7f4bccb | 2123 | * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction |
470decc6 DK |
2124 | * and has a zero b_jcount then remove and release its journal_head. If we did |
2125 | * see that the buffer is not used by any transaction we also "logically" | |
2126 | * decrement ->b_count. | |
2127 | * | |
2128 | * We in fact take an additional increment on ->b_count as a convenience, | |
2129 | * because the caller usually wants to do additional things with the bh | |
2130 | * after calling here. | |
f7f4bccb | 2131 | * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some |
470decc6 DK |
2132 | * time. Once the caller has run __brelse(), the buffer is eligible for |
2133 | * reaping by try_to_free_buffers(). | |
2134 | */ | |
f7f4bccb | 2135 | void jbd2_journal_remove_journal_head(struct buffer_head *bh) |
470decc6 DK |
2136 | { |
2137 | jbd_lock_bh_journal_head(bh); | |
2138 | __journal_remove_journal_head(bh); | |
2139 | jbd_unlock_bh_journal_head(bh); | |
2140 | } | |
2141 | ||
2142 | /* | |
2143 | * Drop a reference on the passed journal_head. If it fell to zero then try to | |
2144 | * release the journal_head from the buffer_head. | |
2145 | */ | |
f7f4bccb | 2146 | void jbd2_journal_put_journal_head(struct journal_head *jh) |
470decc6 DK |
2147 | { |
2148 | struct buffer_head *bh = jh2bh(jh); | |
2149 | ||
2150 | jbd_lock_bh_journal_head(bh); | |
2151 | J_ASSERT_JH(jh, jh->b_jcount > 0); | |
2152 | --jh->b_jcount; | |
2153 | if (!jh->b_jcount && !jh->b_transaction) { | |
2154 | __journal_remove_journal_head(bh); | |
2155 | __brelse(bh); | |
2156 | } | |
2157 | jbd_unlock_bh_journal_head(bh); | |
2158 | } | |
2159 | ||
c851ed54 JK |
2160 | /* |
2161 | * Initialize jbd inode head | |
2162 | */ | |
2163 | void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) | |
2164 | { | |
2165 | jinode->i_transaction = NULL; | |
2166 | jinode->i_next_transaction = NULL; | |
2167 | jinode->i_vfs_inode = inode; | |
2168 | jinode->i_flags = 0; | |
2169 | INIT_LIST_HEAD(&jinode->i_list); | |
2170 | } | |
2171 | ||
2172 | /* | |
2173 | * Function to be called before we start removing inode from memory (i.e., | |
2174 | * clear_inode() is a fine place to be called from). It removes inode from | |
2175 | * transaction's lists. | |
2176 | */ | |
2177 | void jbd2_journal_release_jbd_inode(journal_t *journal, | |
2178 | struct jbd2_inode *jinode) | |
2179 | { | |
2180 | int writeout = 0; | |
2181 | ||
2182 | if (!journal) | |
2183 | return; | |
2184 | restart: | |
2185 | spin_lock(&journal->j_list_lock); | |
2186 | /* Is commit writing out inode - we have to wait */ | |
2187 | if (jinode->i_flags & JI_COMMIT_RUNNING) { | |
2188 | wait_queue_head_t *wq; | |
2189 | DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); | |
2190 | wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); | |
2191 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
2192 | spin_unlock(&journal->j_list_lock); | |
2193 | schedule(); | |
2194 | finish_wait(wq, &wait.wait); | |
2195 | goto restart; | |
2196 | } | |
2197 | ||
2198 | /* Do we need to wait for data writeback? */ | |
2199 | if (journal->j_committing_transaction == jinode->i_transaction) | |
2200 | writeout = 1; | |
2201 | if (jinode->i_transaction) { | |
2202 | list_del(&jinode->i_list); | |
2203 | jinode->i_transaction = NULL; | |
2204 | } | |
2205 | spin_unlock(&journal->j_list_lock); | |
2206 | } | |
2207 | ||
470decc6 | 2208 | /* |
0f49d5d0 | 2209 | * debugfs tunables |
470decc6 | 2210 | */ |
6f38c74f JS |
2211 | #ifdef CONFIG_JBD2_DEBUG |
2212 | u8 jbd2_journal_enable_debug __read_mostly; | |
f7f4bccb | 2213 | EXPORT_SYMBOL(jbd2_journal_enable_debug); |
470decc6 | 2214 | |
0f49d5d0 | 2215 | #define JBD2_DEBUG_NAME "jbd2-debug" |
470decc6 | 2216 | |
6f38c74f JS |
2217 | static struct dentry *jbd2_debugfs_dir; |
2218 | static struct dentry *jbd2_debug; | |
470decc6 | 2219 | |
0f49d5d0 JS |
2220 | static void __init jbd2_create_debugfs_entry(void) |
2221 | { | |
2222 | jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL); | |
2223 | if (jbd2_debugfs_dir) | |
2224 | jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO, | |
2225 | jbd2_debugfs_dir, | |
2226 | &jbd2_journal_enable_debug); | |
470decc6 DK |
2227 | } |
2228 | ||
0f49d5d0 | 2229 | static void __exit jbd2_remove_debugfs_entry(void) |
470decc6 | 2230 | { |
6f38c74f JS |
2231 | debugfs_remove(jbd2_debug); |
2232 | debugfs_remove(jbd2_debugfs_dir); | |
470decc6 DK |
2233 | } |
2234 | ||
0f49d5d0 | 2235 | #else |
470decc6 | 2236 | |
0f49d5d0 | 2237 | static void __init jbd2_create_debugfs_entry(void) |
470decc6 | 2238 | { |
470decc6 DK |
2239 | } |
2240 | ||
0f49d5d0 | 2241 | static void __exit jbd2_remove_debugfs_entry(void) |
470decc6 | 2242 | { |
470decc6 DK |
2243 | } |
2244 | ||
470decc6 DK |
2245 | #endif |
2246 | ||
8e85fb3f JL |
2247 | #ifdef CONFIG_PROC_FS |
2248 | ||
2249 | #define JBD2_STATS_PROC_NAME "fs/jbd2" | |
2250 | ||
2251 | static void __init jbd2_create_jbd_stats_proc_entry(void) | |
2252 | { | |
2253 | proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); | |
2254 | } | |
2255 | ||
2256 | static void __exit jbd2_remove_jbd_stats_proc_entry(void) | |
2257 | { | |
2258 | if (proc_jbd2_stats) | |
2259 | remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); | |
2260 | } | |
2261 | ||
2262 | #else | |
2263 | ||
2264 | #define jbd2_create_jbd_stats_proc_entry() do {} while (0) | |
2265 | #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) | |
2266 | ||
2267 | #endif | |
2268 | ||
e18b890b | 2269 | struct kmem_cache *jbd2_handle_cache; |
470decc6 DK |
2270 | |
2271 | static int __init journal_init_handle_cache(void) | |
2272 | { | |
a920e941 | 2273 | jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle", |
470decc6 DK |
2274 | sizeof(handle_t), |
2275 | 0, /* offset */ | |
77160957 | 2276 | SLAB_TEMPORARY, /* flags */ |
20c2df83 | 2277 | NULL); /* ctor */ |
f7f4bccb | 2278 | if (jbd2_handle_cache == NULL) { |
470decc6 DK |
2279 | printk(KERN_EMERG "JBD: failed to create handle cache\n"); |
2280 | return -ENOMEM; | |
2281 | } | |
2282 | return 0; | |
2283 | } | |
2284 | ||
f7f4bccb | 2285 | static void jbd2_journal_destroy_handle_cache(void) |
470decc6 | 2286 | { |
f7f4bccb MC |
2287 | if (jbd2_handle_cache) |
2288 | kmem_cache_destroy(jbd2_handle_cache); | |
470decc6 DK |
2289 | } |
2290 | ||
2291 | /* | |
2292 | * Module startup and shutdown | |
2293 | */ | |
2294 | ||
2295 | static int __init journal_init_caches(void) | |
2296 | { | |
2297 | int ret; | |
2298 | ||
f7f4bccb | 2299 | ret = jbd2_journal_init_revoke_caches(); |
470decc6 | 2300 | if (ret == 0) |
f7f4bccb | 2301 | ret = journal_init_jbd2_journal_head_cache(); |
470decc6 DK |
2302 | if (ret == 0) |
2303 | ret = journal_init_handle_cache(); | |
2304 | return ret; | |
2305 | } | |
2306 | ||
f7f4bccb | 2307 | static void jbd2_journal_destroy_caches(void) |
470decc6 | 2308 | { |
f7f4bccb MC |
2309 | jbd2_journal_destroy_revoke_caches(); |
2310 | jbd2_journal_destroy_jbd2_journal_head_cache(); | |
2311 | jbd2_journal_destroy_handle_cache(); | |
470decc6 DK |
2312 | } |
2313 | ||
2314 | static int __init journal_init(void) | |
2315 | { | |
2316 | int ret; | |
2317 | ||
2318 | BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); | |
2319 | ||
2320 | ret = journal_init_caches(); | |
620de4e1 DG |
2321 | if (ret == 0) { |
2322 | jbd2_create_debugfs_entry(); | |
2323 | jbd2_create_jbd_stats_proc_entry(); | |
2324 | } else { | |
f7f4bccb | 2325 | jbd2_journal_destroy_caches(); |
620de4e1 | 2326 | } |
470decc6 DK |
2327 | return ret; |
2328 | } | |
2329 | ||
2330 | static void __exit journal_exit(void) | |
2331 | { | |
e23291b9 | 2332 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2333 | int n = atomic_read(&nr_journal_heads); |
2334 | if (n) | |
2335 | printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n); | |
2336 | #endif | |
0f49d5d0 | 2337 | jbd2_remove_debugfs_entry(); |
8e85fb3f | 2338 | jbd2_remove_jbd_stats_proc_entry(); |
f7f4bccb | 2339 | jbd2_journal_destroy_caches(); |
470decc6 DK |
2340 | } |
2341 | ||
2342 | MODULE_LICENSE("GPL"); | |
2343 | module_init(journal_init); | |
2344 | module_exit(journal_exit); | |
2345 |