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> | |
43 | ||
f7f4bccb MC |
44 | EXPORT_SYMBOL(jbd2_journal_start); |
45 | EXPORT_SYMBOL(jbd2_journal_restart); | |
46 | EXPORT_SYMBOL(jbd2_journal_extend); | |
47 | EXPORT_SYMBOL(jbd2_journal_stop); | |
48 | EXPORT_SYMBOL(jbd2_journal_lock_updates); | |
49 | EXPORT_SYMBOL(jbd2_journal_unlock_updates); | |
50 | EXPORT_SYMBOL(jbd2_journal_get_write_access); | |
51 | EXPORT_SYMBOL(jbd2_journal_get_create_access); | |
52 | EXPORT_SYMBOL(jbd2_journal_get_undo_access); | |
f7f4bccb MC |
53 | EXPORT_SYMBOL(jbd2_journal_dirty_metadata); |
54 | EXPORT_SYMBOL(jbd2_journal_release_buffer); | |
55 | EXPORT_SYMBOL(jbd2_journal_forget); | |
470decc6 DK |
56 | #if 0 |
57 | EXPORT_SYMBOL(journal_sync_buffer); | |
58 | #endif | |
f7f4bccb MC |
59 | EXPORT_SYMBOL(jbd2_journal_flush); |
60 | EXPORT_SYMBOL(jbd2_journal_revoke); | |
61 | ||
62 | EXPORT_SYMBOL(jbd2_journal_init_dev); | |
63 | EXPORT_SYMBOL(jbd2_journal_init_inode); | |
64 | EXPORT_SYMBOL(jbd2_journal_update_format); | |
65 | EXPORT_SYMBOL(jbd2_journal_check_used_features); | |
66 | EXPORT_SYMBOL(jbd2_journal_check_available_features); | |
67 | EXPORT_SYMBOL(jbd2_journal_set_features); | |
68 | EXPORT_SYMBOL(jbd2_journal_create); | |
69 | EXPORT_SYMBOL(jbd2_journal_load); | |
70 | EXPORT_SYMBOL(jbd2_journal_destroy); | |
71 | EXPORT_SYMBOL(jbd2_journal_update_superblock); | |
72 | EXPORT_SYMBOL(jbd2_journal_abort); | |
73 | EXPORT_SYMBOL(jbd2_journal_errno); | |
74 | EXPORT_SYMBOL(jbd2_journal_ack_err); | |
75 | EXPORT_SYMBOL(jbd2_journal_clear_err); | |
76 | EXPORT_SYMBOL(jbd2_log_wait_commit); | |
77 | EXPORT_SYMBOL(jbd2_journal_start_commit); | |
78 | EXPORT_SYMBOL(jbd2_journal_force_commit_nested); | |
79 | EXPORT_SYMBOL(jbd2_journal_wipe); | |
80 | EXPORT_SYMBOL(jbd2_journal_blocks_per_page); | |
81 | EXPORT_SYMBOL(jbd2_journal_invalidatepage); | |
82 | EXPORT_SYMBOL(jbd2_journal_try_to_free_buffers); | |
83 | EXPORT_SYMBOL(jbd2_journal_force_commit); | |
c851ed54 JK |
84 | EXPORT_SYMBOL(jbd2_journal_file_inode); |
85 | EXPORT_SYMBOL(jbd2_journal_init_jbd_inode); | |
86 | EXPORT_SYMBOL(jbd2_journal_release_jbd_inode); | |
87 | EXPORT_SYMBOL(jbd2_journal_begin_ordered_truncate); | |
470decc6 DK |
88 | |
89 | static int journal_convert_superblock_v1(journal_t *, journal_superblock_t *); | |
90 | static void __journal_abort_soft (journal_t *journal, int errno); | |
470decc6 DK |
91 | |
92 | /* | |
93 | * Helper function used to manage commit timeouts | |
94 | */ | |
95 | ||
96 | static void commit_timeout(unsigned long __data) | |
97 | { | |
98 | struct task_struct * p = (struct task_struct *) __data; | |
99 | ||
100 | wake_up_process(p); | |
101 | } | |
102 | ||
103 | /* | |
f7f4bccb | 104 | * kjournald2: The main thread function used to manage a logging device |
470decc6 DK |
105 | * journal. |
106 | * | |
107 | * This kernel thread is responsible for two things: | |
108 | * | |
109 | * 1) COMMIT: Every so often we need to commit the current state of the | |
110 | * filesystem to disk. The journal thread is responsible for writing | |
111 | * all of the metadata buffers to disk. | |
112 | * | |
113 | * 2) CHECKPOINT: We cannot reuse a used section of the log file until all | |
114 | * of the data in that part of the log has been rewritten elsewhere on | |
115 | * the disk. Flushing these old buffers to reclaim space in the log is | |
116 | * known as checkpointing, and this thread is responsible for that job. | |
117 | */ | |
118 | ||
f7f4bccb | 119 | static int kjournald2(void *arg) |
470decc6 DK |
120 | { |
121 | journal_t *journal = arg; | |
122 | transaction_t *transaction; | |
123 | ||
124 | /* | |
125 | * Set up an interval timer which can be used to trigger a commit wakeup | |
126 | * after the commit interval expires | |
127 | */ | |
128 | setup_timer(&journal->j_commit_timer, commit_timeout, | |
129 | (unsigned long)current); | |
130 | ||
131 | /* Record that the journal thread is running */ | |
132 | journal->j_task = current; | |
133 | wake_up(&journal->j_wait_done_commit); | |
134 | ||
f7f4bccb | 135 | printk(KERN_INFO "kjournald2 starting. Commit interval %ld seconds\n", |
470decc6 DK |
136 | journal->j_commit_interval / HZ); |
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 { | |
601 | char b[BDEVNAME_SIZE]; | |
602 | ||
603 | printk(KERN_ALERT "%s: journal block not found " | |
604 | "at offset %lu on %s\n", | |
329d291f | 605 | __func__, |
470decc6 DK |
606 | blocknr, |
607 | bdevname(journal->j_dev, b)); | |
608 | err = -EIO; | |
609 | __journal_abort_soft(journal, err); | |
610 | } | |
611 | } else { | |
612 | *retp = blocknr; /* +journal->j_blk_offset */ | |
613 | } | |
614 | return err; | |
615 | } | |
616 | ||
617 | /* | |
618 | * We play buffer_head aliasing tricks to write data/metadata blocks to | |
619 | * the journal without copying their contents, but for journal | |
620 | * descriptor blocks we do need to generate bona fide buffers. | |
621 | * | |
f7f4bccb | 622 | * After the caller of jbd2_journal_get_descriptor_buffer() has finished modifying |
470decc6 DK |
623 | * the buffer's contents they really should run flush_dcache_page(bh->b_page). |
624 | * But we don't bother doing that, so there will be coherency problems with | |
625 | * mmaps of blockdevs which hold live JBD-controlled filesystems. | |
626 | */ | |
f7f4bccb | 627 | struct journal_head *jbd2_journal_get_descriptor_buffer(journal_t *journal) |
470decc6 DK |
628 | { |
629 | struct buffer_head *bh; | |
18eba7aa | 630 | unsigned long long blocknr; |
470decc6 DK |
631 | int err; |
632 | ||
f7f4bccb | 633 | err = jbd2_journal_next_log_block(journal, &blocknr); |
470decc6 DK |
634 | |
635 | if (err) | |
636 | return NULL; | |
637 | ||
638 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
639 | lock_buffer(bh); | |
640 | memset(bh->b_data, 0, journal->j_blocksize); | |
641 | set_buffer_uptodate(bh); | |
642 | unlock_buffer(bh); | |
643 | BUFFER_TRACE(bh, "return this buffer"); | |
f7f4bccb | 644 | return jbd2_journal_add_journal_head(bh); |
470decc6 DK |
645 | } |
646 | ||
8e85fb3f JL |
647 | struct jbd2_stats_proc_session { |
648 | journal_t *journal; | |
649 | struct transaction_stats_s *stats; | |
650 | int start; | |
651 | int max; | |
652 | }; | |
653 | ||
654 | static void *jbd2_history_skip_empty(struct jbd2_stats_proc_session *s, | |
655 | struct transaction_stats_s *ts, | |
656 | int first) | |
657 | { | |
658 | if (ts == s->stats + s->max) | |
659 | ts = s->stats; | |
660 | if (!first && ts == s->stats + s->start) | |
661 | return NULL; | |
662 | while (ts->ts_type == 0) { | |
663 | ts++; | |
664 | if (ts == s->stats + s->max) | |
665 | ts = s->stats; | |
666 | if (ts == s->stats + s->start) | |
667 | return NULL; | |
668 | } | |
669 | return ts; | |
670 | ||
671 | } | |
672 | ||
673 | static void *jbd2_seq_history_start(struct seq_file *seq, loff_t *pos) | |
674 | { | |
675 | struct jbd2_stats_proc_session *s = seq->private; | |
676 | struct transaction_stats_s *ts; | |
677 | int l = *pos; | |
678 | ||
679 | if (l == 0) | |
680 | return SEQ_START_TOKEN; | |
681 | ts = jbd2_history_skip_empty(s, s->stats + s->start, 1); | |
682 | if (!ts) | |
683 | return NULL; | |
684 | l--; | |
685 | while (l) { | |
686 | ts = jbd2_history_skip_empty(s, ++ts, 0); | |
687 | if (!ts) | |
688 | break; | |
689 | l--; | |
690 | } | |
691 | return ts; | |
692 | } | |
693 | ||
694 | static void *jbd2_seq_history_next(struct seq_file *seq, void *v, loff_t *pos) | |
695 | { | |
696 | struct jbd2_stats_proc_session *s = seq->private; | |
697 | struct transaction_stats_s *ts = v; | |
698 | ||
699 | ++*pos; | |
700 | if (v == SEQ_START_TOKEN) | |
701 | return jbd2_history_skip_empty(s, s->stats + s->start, 1); | |
702 | else | |
703 | return jbd2_history_skip_empty(s, ++ts, 0); | |
704 | } | |
705 | ||
706 | static int jbd2_seq_history_show(struct seq_file *seq, void *v) | |
707 | { | |
708 | struct transaction_stats_s *ts = v; | |
709 | if (v == SEQ_START_TOKEN) { | |
710 | seq_printf(seq, "%-4s %-5s %-5s %-5s %-5s %-5s %-5s %-6s %-5s " | |
711 | "%-5s %-5s %-5s %-5s %-5s\n", "R/C", "tid", | |
712 | "wait", "run", "lock", "flush", "log", "hndls", | |
713 | "block", "inlog", "ctime", "write", "drop", | |
714 | "close"); | |
715 | return 0; | |
716 | } | |
717 | if (ts->ts_type == JBD2_STATS_RUN) | |
718 | seq_printf(seq, "%-4s %-5lu %-5u %-5u %-5u %-5u %-5u " | |
719 | "%-6lu %-5lu %-5lu\n", "R", ts->ts_tid, | |
720 | jiffies_to_msecs(ts->u.run.rs_wait), | |
721 | jiffies_to_msecs(ts->u.run.rs_running), | |
722 | jiffies_to_msecs(ts->u.run.rs_locked), | |
723 | jiffies_to_msecs(ts->u.run.rs_flushing), | |
724 | jiffies_to_msecs(ts->u.run.rs_logging), | |
725 | ts->u.run.rs_handle_count, | |
726 | ts->u.run.rs_blocks, | |
727 | ts->u.run.rs_blocks_logged); | |
728 | else if (ts->ts_type == JBD2_STATS_CHECKPOINT) | |
729 | seq_printf(seq, "%-4s %-5lu %48s %-5u %-5lu %-5lu %-5lu\n", | |
730 | "C", ts->ts_tid, " ", | |
731 | jiffies_to_msecs(ts->u.chp.cs_chp_time), | |
732 | ts->u.chp.cs_written, ts->u.chp.cs_dropped, | |
733 | ts->u.chp.cs_forced_to_close); | |
734 | else | |
735 | J_ASSERT(0); | |
736 | return 0; | |
737 | } | |
738 | ||
739 | static void jbd2_seq_history_stop(struct seq_file *seq, void *v) | |
740 | { | |
741 | } | |
742 | ||
743 | static struct seq_operations jbd2_seq_history_ops = { | |
744 | .start = jbd2_seq_history_start, | |
745 | .next = jbd2_seq_history_next, | |
746 | .stop = jbd2_seq_history_stop, | |
747 | .show = jbd2_seq_history_show, | |
748 | }; | |
749 | ||
750 | static int jbd2_seq_history_open(struct inode *inode, struct file *file) | |
751 | { | |
752 | journal_t *journal = PDE(inode)->data; | |
753 | struct jbd2_stats_proc_session *s; | |
754 | int rc, size; | |
755 | ||
756 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
757 | if (s == NULL) | |
758 | return -ENOMEM; | |
759 | size = sizeof(struct transaction_stats_s) * journal->j_history_max; | |
760 | s->stats = kmalloc(size, GFP_KERNEL); | |
761 | if (s->stats == NULL) { | |
762 | kfree(s); | |
763 | return -ENOMEM; | |
764 | } | |
765 | spin_lock(&journal->j_history_lock); | |
766 | memcpy(s->stats, journal->j_history, size); | |
767 | s->max = journal->j_history_max; | |
768 | s->start = journal->j_history_cur % s->max; | |
769 | spin_unlock(&journal->j_history_lock); | |
770 | ||
771 | rc = seq_open(file, &jbd2_seq_history_ops); | |
772 | if (rc == 0) { | |
773 | struct seq_file *m = file->private_data; | |
774 | m->private = s; | |
775 | } else { | |
776 | kfree(s->stats); | |
777 | kfree(s); | |
778 | } | |
779 | return rc; | |
780 | ||
781 | } | |
782 | ||
783 | static int jbd2_seq_history_release(struct inode *inode, struct file *file) | |
784 | { | |
785 | struct seq_file *seq = file->private_data; | |
786 | struct jbd2_stats_proc_session *s = seq->private; | |
787 | ||
788 | kfree(s->stats); | |
789 | kfree(s); | |
790 | return seq_release(inode, file); | |
791 | } | |
792 | ||
793 | static struct file_operations jbd2_seq_history_fops = { | |
794 | .owner = THIS_MODULE, | |
795 | .open = jbd2_seq_history_open, | |
796 | .read = seq_read, | |
797 | .llseek = seq_lseek, | |
798 | .release = jbd2_seq_history_release, | |
799 | }; | |
800 | ||
801 | static void *jbd2_seq_info_start(struct seq_file *seq, loff_t *pos) | |
802 | { | |
803 | return *pos ? NULL : SEQ_START_TOKEN; | |
804 | } | |
805 | ||
806 | static void *jbd2_seq_info_next(struct seq_file *seq, void *v, loff_t *pos) | |
807 | { | |
808 | return NULL; | |
809 | } | |
810 | ||
811 | static int jbd2_seq_info_show(struct seq_file *seq, void *v) | |
812 | { | |
813 | struct jbd2_stats_proc_session *s = seq->private; | |
814 | ||
815 | if (v != SEQ_START_TOKEN) | |
816 | return 0; | |
817 | seq_printf(seq, "%lu transaction, each upto %u blocks\n", | |
818 | s->stats->ts_tid, | |
819 | s->journal->j_max_transaction_buffers); | |
820 | if (s->stats->ts_tid == 0) | |
821 | return 0; | |
822 | seq_printf(seq, "average: \n %ums waiting for transaction\n", | |
823 | jiffies_to_msecs(s->stats->u.run.rs_wait / s->stats->ts_tid)); | |
824 | seq_printf(seq, " %ums running transaction\n", | |
825 | jiffies_to_msecs(s->stats->u.run.rs_running / s->stats->ts_tid)); | |
826 | seq_printf(seq, " %ums transaction was being locked\n", | |
827 | jiffies_to_msecs(s->stats->u.run.rs_locked / s->stats->ts_tid)); | |
828 | seq_printf(seq, " %ums flushing data (in ordered mode)\n", | |
829 | jiffies_to_msecs(s->stats->u.run.rs_flushing / s->stats->ts_tid)); | |
830 | seq_printf(seq, " %ums logging transaction\n", | |
831 | jiffies_to_msecs(s->stats->u.run.rs_logging / s->stats->ts_tid)); | |
832 | seq_printf(seq, " %lu handles per transaction\n", | |
833 | s->stats->u.run.rs_handle_count / s->stats->ts_tid); | |
834 | seq_printf(seq, " %lu blocks per transaction\n", | |
835 | s->stats->u.run.rs_blocks / s->stats->ts_tid); | |
836 | seq_printf(seq, " %lu logged blocks per transaction\n", | |
837 | s->stats->u.run.rs_blocks_logged / s->stats->ts_tid); | |
838 | return 0; | |
839 | } | |
840 | ||
841 | static void jbd2_seq_info_stop(struct seq_file *seq, void *v) | |
842 | { | |
843 | } | |
844 | ||
845 | static struct seq_operations jbd2_seq_info_ops = { | |
846 | .start = jbd2_seq_info_start, | |
847 | .next = jbd2_seq_info_next, | |
848 | .stop = jbd2_seq_info_stop, | |
849 | .show = jbd2_seq_info_show, | |
850 | }; | |
851 | ||
852 | static int jbd2_seq_info_open(struct inode *inode, struct file *file) | |
853 | { | |
854 | journal_t *journal = PDE(inode)->data; | |
855 | struct jbd2_stats_proc_session *s; | |
856 | int rc, size; | |
857 | ||
858 | s = kmalloc(sizeof(*s), GFP_KERNEL); | |
859 | if (s == NULL) | |
860 | return -ENOMEM; | |
861 | size = sizeof(struct transaction_stats_s); | |
862 | s->stats = kmalloc(size, GFP_KERNEL); | |
863 | if (s->stats == NULL) { | |
864 | kfree(s); | |
865 | return -ENOMEM; | |
866 | } | |
867 | spin_lock(&journal->j_history_lock); | |
868 | memcpy(s->stats, &journal->j_stats, size); | |
869 | s->journal = journal; | |
870 | spin_unlock(&journal->j_history_lock); | |
871 | ||
872 | rc = seq_open(file, &jbd2_seq_info_ops); | |
873 | if (rc == 0) { | |
874 | struct seq_file *m = file->private_data; | |
875 | m->private = s; | |
876 | } else { | |
877 | kfree(s->stats); | |
878 | kfree(s); | |
879 | } | |
880 | return rc; | |
881 | ||
882 | } | |
883 | ||
884 | static int jbd2_seq_info_release(struct inode *inode, struct file *file) | |
885 | { | |
886 | struct seq_file *seq = file->private_data; | |
887 | struct jbd2_stats_proc_session *s = seq->private; | |
888 | kfree(s->stats); | |
889 | kfree(s); | |
890 | return seq_release(inode, file); | |
891 | } | |
892 | ||
893 | static struct file_operations jbd2_seq_info_fops = { | |
894 | .owner = THIS_MODULE, | |
895 | .open = jbd2_seq_info_open, | |
896 | .read = seq_read, | |
897 | .llseek = seq_lseek, | |
898 | .release = jbd2_seq_info_release, | |
899 | }; | |
900 | ||
901 | static struct proc_dir_entry *proc_jbd2_stats; | |
902 | ||
903 | static void jbd2_stats_proc_init(journal_t *journal) | |
904 | { | |
905 | char name[BDEVNAME_SIZE]; | |
906 | ||
f36f21ec | 907 | bdevname(journal->j_dev, name); |
8e85fb3f JL |
908 | journal->j_proc_entry = proc_mkdir(name, proc_jbd2_stats); |
909 | if (journal->j_proc_entry) { | |
79da3664 DL |
910 | proc_create_data("history", S_IRUGO, journal->j_proc_entry, |
911 | &jbd2_seq_history_fops, journal); | |
912 | proc_create_data("info", S_IRUGO, journal->j_proc_entry, | |
913 | &jbd2_seq_info_fops, journal); | |
8e85fb3f JL |
914 | } |
915 | } | |
916 | ||
917 | static void jbd2_stats_proc_exit(journal_t *journal) | |
918 | { | |
919 | char name[BDEVNAME_SIZE]; | |
920 | ||
f36f21ec | 921 | bdevname(journal->j_dev, name); |
8e85fb3f JL |
922 | remove_proc_entry("info", journal->j_proc_entry); |
923 | remove_proc_entry("history", journal->j_proc_entry); | |
924 | remove_proc_entry(name, proc_jbd2_stats); | |
925 | } | |
926 | ||
927 | static void journal_init_stats(journal_t *journal) | |
928 | { | |
929 | int size; | |
930 | ||
931 | if (!proc_jbd2_stats) | |
932 | return; | |
933 | ||
934 | journal->j_history_max = 100; | |
935 | size = sizeof(struct transaction_stats_s) * journal->j_history_max; | |
936 | journal->j_history = kzalloc(size, GFP_KERNEL); | |
937 | if (!journal->j_history) { | |
938 | journal->j_history_max = 0; | |
939 | return; | |
940 | } | |
941 | spin_lock_init(&journal->j_history_lock); | |
942 | } | |
943 | ||
470decc6 DK |
944 | /* |
945 | * Management for journal control blocks: functions to create and | |
946 | * destroy journal_t structures, and to initialise and read existing | |
947 | * journal blocks from disk. */ | |
948 | ||
949 | /* First: create and setup a journal_t object in memory. We initialise | |
950 | * very few fields yet: that has to wait until we have created the | |
951 | * journal structures from from scratch, or loaded them from disk. */ | |
952 | ||
953 | static journal_t * journal_init_common (void) | |
954 | { | |
955 | journal_t *journal; | |
956 | int err; | |
957 | ||
d802ffa8 | 958 | journal = kzalloc(sizeof(*journal), GFP_KERNEL|__GFP_NOFAIL); |
470decc6 DK |
959 | if (!journal) |
960 | goto fail; | |
470decc6 DK |
961 | |
962 | init_waitqueue_head(&journal->j_wait_transaction_locked); | |
963 | init_waitqueue_head(&journal->j_wait_logspace); | |
964 | init_waitqueue_head(&journal->j_wait_done_commit); | |
965 | init_waitqueue_head(&journal->j_wait_checkpoint); | |
966 | init_waitqueue_head(&journal->j_wait_commit); | |
967 | init_waitqueue_head(&journal->j_wait_updates); | |
968 | mutex_init(&journal->j_barrier); | |
969 | mutex_init(&journal->j_checkpoint_mutex); | |
970 | spin_lock_init(&journal->j_revoke_lock); | |
971 | spin_lock_init(&journal->j_list_lock); | |
972 | spin_lock_init(&journal->j_state_lock); | |
973 | ||
cd02ff0b | 974 | journal->j_commit_interval = (HZ * JBD2_DEFAULT_MAX_COMMIT_AGE); |
470decc6 DK |
975 | |
976 | /* The journal is marked for error until we succeed with recovery! */ | |
f7f4bccb | 977 | journal->j_flags = JBD2_ABORT; |
470decc6 DK |
978 | |
979 | /* Set up a default-sized revoke table for the new mount. */ | |
f7f4bccb | 980 | err = jbd2_journal_init_revoke(journal, JOURNAL_REVOKE_DEFAULT_HASH); |
470decc6 DK |
981 | if (err) { |
982 | kfree(journal); | |
983 | goto fail; | |
984 | } | |
8e85fb3f JL |
985 | |
986 | journal_init_stats(journal); | |
987 | ||
470decc6 DK |
988 | return journal; |
989 | fail: | |
990 | return NULL; | |
991 | } | |
992 | ||
f7f4bccb | 993 | /* jbd2_journal_init_dev and jbd2_journal_init_inode: |
470decc6 DK |
994 | * |
995 | * Create a journal structure assigned some fixed set of disk blocks to | |
996 | * the journal. We don't actually touch those disk blocks yet, but we | |
997 | * need to set up all of the mapping information to tell the journaling | |
998 | * system where the journal blocks are. | |
999 | * | |
1000 | */ | |
1001 | ||
1002 | /** | |
5648ba5b | 1003 | * journal_t * jbd2_journal_init_dev() - creates and initialises a journal structure |
470decc6 DK |
1004 | * @bdev: Block device on which to create the journal |
1005 | * @fs_dev: Device which hold journalled filesystem for this journal. | |
1006 | * @start: Block nr Start of journal. | |
1007 | * @len: Length of the journal in blocks. | |
1008 | * @blocksize: blocksize of journalling device | |
5648ba5b RD |
1009 | * |
1010 | * Returns: a newly created journal_t * | |
470decc6 | 1011 | * |
f7f4bccb | 1012 | * jbd2_journal_init_dev creates a journal which maps a fixed contiguous |
470decc6 DK |
1013 | * range of blocks on an arbitrary block device. |
1014 | * | |
1015 | */ | |
f7f4bccb | 1016 | journal_t * jbd2_journal_init_dev(struct block_device *bdev, |
470decc6 | 1017 | struct block_device *fs_dev, |
18eba7aa | 1018 | unsigned long long start, int len, int blocksize) |
470decc6 DK |
1019 | { |
1020 | journal_t *journal = journal_init_common(); | |
1021 | struct buffer_head *bh; | |
1022 | int n; | |
1023 | ||
1024 | if (!journal) | |
1025 | return NULL; | |
1026 | ||
1027 | /* journal descriptor can store up to n blocks -bzzz */ | |
1028 | journal->j_blocksize = blocksize; | |
1029 | n = journal->j_blocksize / sizeof(journal_block_tag_t); | |
1030 | journal->j_wbufsize = n; | |
1031 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1032 | if (!journal->j_wbuf) { | |
1033 | printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n", | |
329d291f | 1034 | __func__); |
470decc6 DK |
1035 | kfree(journal); |
1036 | journal = NULL; | |
5eb30790 | 1037 | goto out; |
470decc6 DK |
1038 | } |
1039 | journal->j_dev = bdev; | |
1040 | journal->j_fs_dev = fs_dev; | |
1041 | journal->j_blk_offset = start; | |
1042 | journal->j_maxlen = len; | |
8e85fb3f | 1043 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1044 | |
1045 | bh = __getblk(journal->j_dev, start, journal->j_blocksize); | |
1046 | J_ASSERT(bh != NULL); | |
1047 | journal->j_sb_buffer = bh; | |
1048 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
5eb30790 | 1049 | out: |
470decc6 DK |
1050 | return journal; |
1051 | } | |
1052 | ||
1053 | /** | |
f7f4bccb | 1054 | * journal_t * jbd2_journal_init_inode () - creates a journal which maps to a inode. |
470decc6 DK |
1055 | * @inode: An inode to create the journal in |
1056 | * | |
f7f4bccb | 1057 | * jbd2_journal_init_inode creates a journal which maps an on-disk inode as |
470decc6 DK |
1058 | * the journal. The inode must exist already, must support bmap() and |
1059 | * must have all data blocks preallocated. | |
1060 | */ | |
f7f4bccb | 1061 | journal_t * jbd2_journal_init_inode (struct inode *inode) |
470decc6 DK |
1062 | { |
1063 | struct buffer_head *bh; | |
1064 | journal_t *journal = journal_init_common(); | |
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; | |
1074 | jbd_debug(1, | |
1075 | "journal %p: inode %s/%ld, size %Ld, bits %d, blksize %ld\n", | |
1076 | journal, inode->i_sb->s_id, inode->i_ino, | |
1077 | (long long) inode->i_size, | |
1078 | inode->i_sb->s_blocksize_bits, inode->i_sb->s_blocksize); | |
1079 | ||
1080 | journal->j_maxlen = inode->i_size >> inode->i_sb->s_blocksize_bits; | |
1081 | journal->j_blocksize = inode->i_sb->s_blocksize; | |
8e85fb3f | 1082 | jbd2_stats_proc_init(journal); |
470decc6 DK |
1083 | |
1084 | /* journal descriptor can store up to n blocks -bzzz */ | |
1085 | n = journal->j_blocksize / sizeof(journal_block_tag_t); | |
1086 | journal->j_wbufsize = n; | |
1087 | journal->j_wbuf = kmalloc(n * sizeof(struct buffer_head*), GFP_KERNEL); | |
1088 | if (!journal->j_wbuf) { | |
1089 | printk(KERN_ERR "%s: Cant allocate bhs for commit thread\n", | |
329d291f | 1090 | __func__); |
470decc6 DK |
1091 | kfree(journal); |
1092 | return NULL; | |
1093 | } | |
1094 | ||
f7f4bccb | 1095 | err = jbd2_journal_bmap(journal, 0, &blocknr); |
470decc6 DK |
1096 | /* If that failed, give up */ |
1097 | if (err) { | |
1098 | printk(KERN_ERR "%s: Cannnot locate journal superblock\n", | |
329d291f | 1099 | __func__); |
470decc6 DK |
1100 | kfree(journal); |
1101 | return NULL; | |
1102 | } | |
1103 | ||
1104 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
1105 | J_ASSERT(bh != NULL); | |
1106 | journal->j_sb_buffer = bh; | |
1107 | journal->j_superblock = (journal_superblock_t *)bh->b_data; | |
1108 | ||
1109 | return journal; | |
1110 | } | |
1111 | ||
1112 | /* | |
1113 | * If the journal init or create aborts, we need to mark the journal | |
1114 | * superblock as being NULL to prevent the journal destroy from writing | |
1115 | * back a bogus superblock. | |
1116 | */ | |
1117 | static void journal_fail_superblock (journal_t *journal) | |
1118 | { | |
1119 | struct buffer_head *bh = journal->j_sb_buffer; | |
1120 | brelse(bh); | |
1121 | journal->j_sb_buffer = NULL; | |
1122 | } | |
1123 | ||
1124 | /* | |
1125 | * Given a journal_t structure, initialise the various fields for | |
1126 | * startup of a new journaling session. We use this both when creating | |
1127 | * a journal, and after recovering an old journal to reset it for | |
1128 | * subsequent use. | |
1129 | */ | |
1130 | ||
1131 | static int journal_reset(journal_t *journal) | |
1132 | { | |
1133 | journal_superblock_t *sb = journal->j_superblock; | |
18eba7aa | 1134 | unsigned long long first, last; |
470decc6 DK |
1135 | |
1136 | first = be32_to_cpu(sb->s_first); | |
1137 | last = be32_to_cpu(sb->s_maxlen); | |
1138 | ||
1139 | journal->j_first = first; | |
1140 | journal->j_last = last; | |
1141 | ||
1142 | journal->j_head = first; | |
1143 | journal->j_tail = first; | |
1144 | journal->j_free = last - first; | |
1145 | ||
1146 | journal->j_tail_sequence = journal->j_transaction_sequence; | |
1147 | journal->j_commit_sequence = journal->j_transaction_sequence - 1; | |
1148 | journal->j_commit_request = journal->j_commit_sequence; | |
1149 | ||
1150 | journal->j_max_transaction_buffers = journal->j_maxlen / 4; | |
1151 | ||
1152 | /* Add the dynamic fields and write it to disk. */ | |
f7f4bccb | 1153 | jbd2_journal_update_superblock(journal, 1); |
97f06784 | 1154 | return jbd2_journal_start_thread(journal); |
470decc6 DK |
1155 | } |
1156 | ||
1157 | /** | |
f7f4bccb | 1158 | * int jbd2_journal_create() - Initialise the new journal file |
470decc6 DK |
1159 | * @journal: Journal to create. This structure must have been initialised |
1160 | * | |
1161 | * Given a journal_t structure which tells us which disk blocks we can | |
1162 | * use, create a new journal superblock and initialise all of the | |
1163 | * journal fields from scratch. | |
1164 | **/ | |
f7f4bccb | 1165 | int jbd2_journal_create(journal_t *journal) |
470decc6 | 1166 | { |
18eba7aa | 1167 | unsigned long long blocknr; |
470decc6 DK |
1168 | struct buffer_head *bh; |
1169 | journal_superblock_t *sb; | |
1170 | int i, err; | |
1171 | ||
f7f4bccb | 1172 | if (journal->j_maxlen < JBD2_MIN_JOURNAL_BLOCKS) { |
470decc6 DK |
1173 | printk (KERN_ERR "Journal length (%d blocks) too short.\n", |
1174 | journal->j_maxlen); | |
1175 | journal_fail_superblock(journal); | |
1176 | return -EINVAL; | |
1177 | } | |
1178 | ||
1179 | if (journal->j_inode == NULL) { | |
1180 | /* | |
1181 | * We don't know what block to start at! | |
1182 | */ | |
1183 | printk(KERN_EMERG | |
1184 | "%s: creation of journal on external device!\n", | |
329d291f | 1185 | __func__); |
470decc6 DK |
1186 | BUG(); |
1187 | } | |
1188 | ||
1189 | /* Zero out the entire journal on disk. We cannot afford to | |
f7f4bccb | 1190 | have any blocks on disk beginning with JBD2_MAGIC_NUMBER. */ |
470decc6 DK |
1191 | jbd_debug(1, "JBD: Zeroing out journal blocks...\n"); |
1192 | for (i = 0; i < journal->j_maxlen; i++) { | |
f7f4bccb | 1193 | err = jbd2_journal_bmap(journal, i, &blocknr); |
470decc6 DK |
1194 | if (err) |
1195 | return err; | |
1196 | bh = __getblk(journal->j_dev, blocknr, journal->j_blocksize); | |
1197 | lock_buffer(bh); | |
1198 | memset (bh->b_data, 0, journal->j_blocksize); | |
1199 | BUFFER_TRACE(bh, "marking dirty"); | |
1200 | mark_buffer_dirty(bh); | |
1201 | BUFFER_TRACE(bh, "marking uptodate"); | |
1202 | set_buffer_uptodate(bh); | |
1203 | unlock_buffer(bh); | |
1204 | __brelse(bh); | |
1205 | } | |
1206 | ||
1207 | sync_blockdev(journal->j_dev); | |
1208 | jbd_debug(1, "JBD: journal cleared.\n"); | |
1209 | ||
1210 | /* OK, fill in the initial static fields in the new superblock */ | |
1211 | sb = journal->j_superblock; | |
1212 | ||
f7f4bccb MC |
1213 | sb->s_header.h_magic = cpu_to_be32(JBD2_MAGIC_NUMBER); |
1214 | sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2); | |
470decc6 DK |
1215 | |
1216 | sb->s_blocksize = cpu_to_be32(journal->j_blocksize); | |
1217 | sb->s_maxlen = cpu_to_be32(journal->j_maxlen); | |
1218 | sb->s_first = cpu_to_be32(1); | |
1219 | ||
1220 | journal->j_transaction_sequence = 1; | |
1221 | ||
f7f4bccb | 1222 | journal->j_flags &= ~JBD2_ABORT; |
470decc6 DK |
1223 | journal->j_format_version = 2; |
1224 | ||
1225 | return journal_reset(journal); | |
1226 | } | |
1227 | ||
1228 | /** | |
f7f4bccb | 1229 | * void jbd2_journal_update_superblock() - Update journal sb on disk. |
470decc6 DK |
1230 | * @journal: The journal to update. |
1231 | * @wait: Set to '0' if you don't want to wait for IO completion. | |
1232 | * | |
1233 | * Update a journal's dynamic superblock fields and write it to disk, | |
1234 | * optionally waiting for the IO to complete. | |
1235 | */ | |
f7f4bccb | 1236 | void jbd2_journal_update_superblock(journal_t *journal, int wait) |
470decc6 DK |
1237 | { |
1238 | journal_superblock_t *sb = journal->j_superblock; | |
1239 | struct buffer_head *bh = journal->j_sb_buffer; | |
1240 | ||
1241 | /* | |
1242 | * As a special case, if the on-disk copy is already marked as needing | |
1243 | * no recovery (s_start == 0) and there are no outstanding transactions | |
1244 | * in the filesystem, then we can safely defer the superblock update | |
f7f4bccb | 1245 | * until the next commit by setting JBD2_FLUSHED. This avoids |
470decc6 DK |
1246 | * attempting a write to a potential-readonly device. |
1247 | */ | |
1248 | if (sb->s_start == 0 && journal->j_tail_sequence == | |
1249 | journal->j_transaction_sequence) { | |
1250 | jbd_debug(1,"JBD: Skipping superblock update on recovered sb " | |
1251 | "(start %ld, seq %d, errno %d)\n", | |
1252 | journal->j_tail, journal->j_tail_sequence, | |
1253 | journal->j_errno); | |
1254 | goto out; | |
1255 | } | |
1256 | ||
1257 | spin_lock(&journal->j_state_lock); | |
1258 | jbd_debug(1,"JBD: updating superblock (start %ld, seq %d, errno %d)\n", | |
1259 | journal->j_tail, journal->j_tail_sequence, journal->j_errno); | |
1260 | ||
1261 | sb->s_sequence = cpu_to_be32(journal->j_tail_sequence); | |
1262 | sb->s_start = cpu_to_be32(journal->j_tail); | |
1263 | sb->s_errno = cpu_to_be32(journal->j_errno); | |
1264 | spin_unlock(&journal->j_state_lock); | |
1265 | ||
1266 | BUFFER_TRACE(bh, "marking dirty"); | |
1267 | mark_buffer_dirty(bh); | |
1268 | if (wait) | |
1269 | sync_dirty_buffer(bh); | |
1270 | else | |
1271 | ll_rw_block(SWRITE, 1, &bh); | |
1272 | ||
1273 | out: | |
1274 | /* If we have just flushed the log (by marking s_start==0), then | |
1275 | * any future commit will have to be careful to update the | |
1276 | * superblock again to re-record the true start of the log. */ | |
1277 | ||
1278 | spin_lock(&journal->j_state_lock); | |
1279 | if (sb->s_start) | |
f7f4bccb | 1280 | journal->j_flags &= ~JBD2_FLUSHED; |
470decc6 | 1281 | else |
f7f4bccb | 1282 | journal->j_flags |= JBD2_FLUSHED; |
470decc6 DK |
1283 | spin_unlock(&journal->j_state_lock); |
1284 | } | |
1285 | ||
1286 | /* | |
1287 | * Read the superblock for a given journal, performing initial | |
1288 | * validation of the format. | |
1289 | */ | |
1290 | ||
1291 | static int journal_get_superblock(journal_t *journal) | |
1292 | { | |
1293 | struct buffer_head *bh; | |
1294 | journal_superblock_t *sb; | |
1295 | int err = -EIO; | |
1296 | ||
1297 | bh = journal->j_sb_buffer; | |
1298 | ||
1299 | J_ASSERT(bh != NULL); | |
1300 | if (!buffer_uptodate(bh)) { | |
1301 | ll_rw_block(READ, 1, &bh); | |
1302 | wait_on_buffer(bh); | |
1303 | if (!buffer_uptodate(bh)) { | |
1304 | printk (KERN_ERR | |
1305 | "JBD: IO error reading journal superblock\n"); | |
1306 | goto out; | |
1307 | } | |
1308 | } | |
1309 | ||
1310 | sb = journal->j_superblock; | |
1311 | ||
1312 | err = -EINVAL; | |
1313 | ||
f7f4bccb | 1314 | if (sb->s_header.h_magic != cpu_to_be32(JBD2_MAGIC_NUMBER) || |
470decc6 DK |
1315 | sb->s_blocksize != cpu_to_be32(journal->j_blocksize)) { |
1316 | printk(KERN_WARNING "JBD: no valid journal superblock found\n"); | |
1317 | goto out; | |
1318 | } | |
1319 | ||
1320 | switch(be32_to_cpu(sb->s_header.h_blocktype)) { | |
f7f4bccb | 1321 | case JBD2_SUPERBLOCK_V1: |
470decc6 DK |
1322 | journal->j_format_version = 1; |
1323 | break; | |
f7f4bccb | 1324 | case JBD2_SUPERBLOCK_V2: |
470decc6 DK |
1325 | journal->j_format_version = 2; |
1326 | break; | |
1327 | default: | |
1328 | printk(KERN_WARNING "JBD: unrecognised superblock format ID\n"); | |
1329 | goto out; | |
1330 | } | |
1331 | ||
1332 | if (be32_to_cpu(sb->s_maxlen) < journal->j_maxlen) | |
1333 | journal->j_maxlen = be32_to_cpu(sb->s_maxlen); | |
1334 | else if (be32_to_cpu(sb->s_maxlen) > journal->j_maxlen) { | |
1335 | printk (KERN_WARNING "JBD: journal file too short\n"); | |
1336 | goto out; | |
1337 | } | |
1338 | ||
1339 | return 0; | |
1340 | ||
1341 | out: | |
1342 | journal_fail_superblock(journal); | |
1343 | return err; | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * Load the on-disk journal superblock and read the key fields into the | |
1348 | * journal_t. | |
1349 | */ | |
1350 | ||
1351 | static int load_superblock(journal_t *journal) | |
1352 | { | |
1353 | int err; | |
1354 | journal_superblock_t *sb; | |
1355 | ||
1356 | err = journal_get_superblock(journal); | |
1357 | if (err) | |
1358 | return err; | |
1359 | ||
1360 | sb = journal->j_superblock; | |
1361 | ||
1362 | journal->j_tail_sequence = be32_to_cpu(sb->s_sequence); | |
1363 | journal->j_tail = be32_to_cpu(sb->s_start); | |
1364 | journal->j_first = be32_to_cpu(sb->s_first); | |
1365 | journal->j_last = be32_to_cpu(sb->s_maxlen); | |
1366 | journal->j_errno = be32_to_cpu(sb->s_errno); | |
1367 | ||
1368 | return 0; | |
1369 | } | |
1370 | ||
1371 | ||
1372 | /** | |
f7f4bccb | 1373 | * int jbd2_journal_load() - Read journal from disk. |
470decc6 DK |
1374 | * @journal: Journal to act on. |
1375 | * | |
1376 | * Given a journal_t structure which tells us which disk blocks contain | |
1377 | * a journal, read the journal from disk to initialise the in-memory | |
1378 | * structures. | |
1379 | */ | |
f7f4bccb | 1380 | int jbd2_journal_load(journal_t *journal) |
470decc6 DK |
1381 | { |
1382 | int err; | |
1383 | journal_superblock_t *sb; | |
1384 | ||
1385 | err = load_superblock(journal); | |
1386 | if (err) | |
1387 | return err; | |
1388 | ||
1389 | sb = journal->j_superblock; | |
1390 | /* If this is a V2 superblock, then we have to check the | |
1391 | * features flags on it. */ | |
1392 | ||
1393 | if (journal->j_format_version >= 2) { | |
1394 | if ((sb->s_feature_ro_compat & | |
f7f4bccb | 1395 | ~cpu_to_be32(JBD2_KNOWN_ROCOMPAT_FEATURES)) || |
470decc6 | 1396 | (sb->s_feature_incompat & |
f7f4bccb | 1397 | ~cpu_to_be32(JBD2_KNOWN_INCOMPAT_FEATURES))) { |
470decc6 DK |
1398 | printk (KERN_WARNING |
1399 | "JBD: Unrecognised features on journal\n"); | |
1400 | return -EINVAL; | |
1401 | } | |
1402 | } | |
1403 | ||
470decc6 DK |
1404 | /* Let the recovery code check whether it needs to recover any |
1405 | * data from the journal. */ | |
f7f4bccb | 1406 | if (jbd2_journal_recover(journal)) |
470decc6 DK |
1407 | goto recovery_error; |
1408 | ||
1409 | /* OK, we've finished with the dynamic journal bits: | |
1410 | * reinitialise the dynamic contents of the superblock in memory | |
1411 | * and reset them on disk. */ | |
1412 | if (journal_reset(journal)) | |
1413 | goto recovery_error; | |
1414 | ||
f7f4bccb MC |
1415 | journal->j_flags &= ~JBD2_ABORT; |
1416 | journal->j_flags |= JBD2_LOADED; | |
470decc6 DK |
1417 | return 0; |
1418 | ||
1419 | recovery_error: | |
1420 | printk (KERN_WARNING "JBD: recovery failed\n"); | |
1421 | return -EIO; | |
1422 | } | |
1423 | ||
1424 | /** | |
f7f4bccb | 1425 | * void jbd2_journal_destroy() - Release a journal_t structure. |
470decc6 DK |
1426 | * @journal: Journal to act on. |
1427 | * | |
1428 | * Release a journal_t structure once it is no longer in use by the | |
1429 | * journaled object. | |
1430 | */ | |
f7f4bccb | 1431 | void jbd2_journal_destroy(journal_t *journal) |
470decc6 DK |
1432 | { |
1433 | /* Wait for the commit thread to wake up and die. */ | |
1434 | journal_kill_thread(journal); | |
1435 | ||
1436 | /* Force a final log commit */ | |
1437 | if (journal->j_running_transaction) | |
f7f4bccb | 1438 | jbd2_journal_commit_transaction(journal); |
470decc6 DK |
1439 | |
1440 | /* Force any old transactions to disk */ | |
1441 | ||
1442 | /* Totally anal locking here... */ | |
1443 | spin_lock(&journal->j_list_lock); | |
1444 | while (journal->j_checkpoint_transactions != NULL) { | |
1445 | spin_unlock(&journal->j_list_lock); | |
f7f4bccb | 1446 | jbd2_log_do_checkpoint(journal); |
470decc6 DK |
1447 | spin_lock(&journal->j_list_lock); |
1448 | } | |
1449 | ||
1450 | J_ASSERT(journal->j_running_transaction == NULL); | |
1451 | J_ASSERT(journal->j_committing_transaction == NULL); | |
1452 | J_ASSERT(journal->j_checkpoint_transactions == NULL); | |
1453 | spin_unlock(&journal->j_list_lock); | |
1454 | ||
1455 | /* We can now mark the journal as empty. */ | |
1456 | journal->j_tail = 0; | |
1457 | journal->j_tail_sequence = ++journal->j_transaction_sequence; | |
1458 | if (journal->j_sb_buffer) { | |
f7f4bccb | 1459 | jbd2_journal_update_superblock(journal, 1); |
470decc6 DK |
1460 | brelse(journal->j_sb_buffer); |
1461 | } | |
1462 | ||
8e85fb3f JL |
1463 | if (journal->j_proc_entry) |
1464 | jbd2_stats_proc_exit(journal); | |
470decc6 DK |
1465 | if (journal->j_inode) |
1466 | iput(journal->j_inode); | |
1467 | if (journal->j_revoke) | |
f7f4bccb | 1468 | jbd2_journal_destroy_revoke(journal); |
470decc6 DK |
1469 | kfree(journal->j_wbuf); |
1470 | kfree(journal); | |
1471 | } | |
1472 | ||
1473 | ||
1474 | /** | |
f7f4bccb | 1475 | *int jbd2_journal_check_used_features () - Check if features specified are used. |
470decc6 DK |
1476 | * @journal: Journal to check. |
1477 | * @compat: bitmask of compatible features | |
1478 | * @ro: bitmask of features that force read-only mount | |
1479 | * @incompat: bitmask of incompatible features | |
1480 | * | |
1481 | * Check whether the journal uses all of a given set of | |
1482 | * features. Return true (non-zero) if it does. | |
1483 | **/ | |
1484 | ||
f7f4bccb | 1485 | int jbd2_journal_check_used_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1486 | unsigned long ro, unsigned long incompat) |
1487 | { | |
1488 | journal_superblock_t *sb; | |
1489 | ||
1490 | if (!compat && !ro && !incompat) | |
1491 | return 1; | |
1492 | if (journal->j_format_version == 1) | |
1493 | return 0; | |
1494 | ||
1495 | sb = journal->j_superblock; | |
1496 | ||
1497 | if (((be32_to_cpu(sb->s_feature_compat) & compat) == compat) && | |
1498 | ((be32_to_cpu(sb->s_feature_ro_compat) & ro) == ro) && | |
1499 | ((be32_to_cpu(sb->s_feature_incompat) & incompat) == incompat)) | |
1500 | return 1; | |
1501 | ||
1502 | return 0; | |
1503 | } | |
1504 | ||
1505 | /** | |
f7f4bccb | 1506 | * int jbd2_journal_check_available_features() - Check feature set in journalling layer |
470decc6 DK |
1507 | * @journal: Journal to check. |
1508 | * @compat: bitmask of compatible features | |
1509 | * @ro: bitmask of features that force read-only mount | |
1510 | * @incompat: bitmask of incompatible features | |
1511 | * | |
1512 | * Check whether the journaling code supports the use of | |
1513 | * all of a given set of features on this journal. Return true | |
1514 | * (non-zero) if it can. */ | |
1515 | ||
f7f4bccb | 1516 | int jbd2_journal_check_available_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1517 | unsigned long ro, unsigned long incompat) |
1518 | { | |
1519 | journal_superblock_t *sb; | |
1520 | ||
1521 | if (!compat && !ro && !incompat) | |
1522 | return 1; | |
1523 | ||
1524 | sb = journal->j_superblock; | |
1525 | ||
1526 | /* We can support any known requested features iff the | |
1527 | * superblock is in version 2. Otherwise we fail to support any | |
1528 | * extended sb features. */ | |
1529 | ||
1530 | if (journal->j_format_version != 2) | |
1531 | return 0; | |
1532 | ||
f7f4bccb MC |
1533 | if ((compat & JBD2_KNOWN_COMPAT_FEATURES) == compat && |
1534 | (ro & JBD2_KNOWN_ROCOMPAT_FEATURES) == ro && | |
1535 | (incompat & JBD2_KNOWN_INCOMPAT_FEATURES) == incompat) | |
470decc6 DK |
1536 | return 1; |
1537 | ||
1538 | return 0; | |
1539 | } | |
1540 | ||
1541 | /** | |
f7f4bccb | 1542 | * int jbd2_journal_set_features () - Mark a given journal feature in the superblock |
470decc6 DK |
1543 | * @journal: Journal to act on. |
1544 | * @compat: bitmask of compatible features | |
1545 | * @ro: bitmask of features that force read-only mount | |
1546 | * @incompat: bitmask of incompatible features | |
1547 | * | |
1548 | * Mark a given journal feature as present on the | |
1549 | * superblock. Returns true if the requested features could be set. | |
1550 | * | |
1551 | */ | |
1552 | ||
f7f4bccb | 1553 | int jbd2_journal_set_features (journal_t *journal, unsigned long compat, |
470decc6 DK |
1554 | unsigned long ro, unsigned long incompat) |
1555 | { | |
1556 | journal_superblock_t *sb; | |
1557 | ||
f7f4bccb | 1558 | if (jbd2_journal_check_used_features(journal, compat, ro, incompat)) |
470decc6 DK |
1559 | return 1; |
1560 | ||
f7f4bccb | 1561 | if (!jbd2_journal_check_available_features(journal, compat, ro, incompat)) |
470decc6 DK |
1562 | return 0; |
1563 | ||
1564 | jbd_debug(1, "Setting new 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 | return 1; | |
1574 | } | |
1575 | ||
818d276c GS |
1576 | /* |
1577 | * jbd2_journal_clear_features () - Clear a given journal feature in the | |
1578 | * superblock | |
1579 | * @journal: Journal to act on. | |
1580 | * @compat: bitmask of compatible features | |
1581 | * @ro: bitmask of features that force read-only mount | |
1582 | * @incompat: bitmask of incompatible features | |
1583 | * | |
1584 | * Clear a given journal feature as present on the | |
1585 | * superblock. | |
1586 | */ | |
1587 | void jbd2_journal_clear_features(journal_t *journal, unsigned long compat, | |
1588 | unsigned long ro, unsigned long incompat) | |
1589 | { | |
1590 | journal_superblock_t *sb; | |
1591 | ||
1592 | jbd_debug(1, "Clear features 0x%lx/0x%lx/0x%lx\n", | |
1593 | compat, ro, incompat); | |
1594 | ||
1595 | sb = journal->j_superblock; | |
1596 | ||
1597 | sb->s_feature_compat &= ~cpu_to_be32(compat); | |
1598 | sb->s_feature_ro_compat &= ~cpu_to_be32(ro); | |
1599 | sb->s_feature_incompat &= ~cpu_to_be32(incompat); | |
1600 | } | |
1601 | EXPORT_SYMBOL(jbd2_journal_clear_features); | |
470decc6 DK |
1602 | |
1603 | /** | |
f7f4bccb | 1604 | * int jbd2_journal_update_format () - Update on-disk journal structure. |
470decc6 DK |
1605 | * @journal: Journal to act on. |
1606 | * | |
1607 | * Given an initialised but unloaded journal struct, poke about in the | |
1608 | * on-disk structure to update it to the most recent supported version. | |
1609 | */ | |
f7f4bccb | 1610 | int jbd2_journal_update_format (journal_t *journal) |
470decc6 DK |
1611 | { |
1612 | journal_superblock_t *sb; | |
1613 | int err; | |
1614 | ||
1615 | err = journal_get_superblock(journal); | |
1616 | if (err) | |
1617 | return err; | |
1618 | ||
1619 | sb = journal->j_superblock; | |
1620 | ||
1621 | switch (be32_to_cpu(sb->s_header.h_blocktype)) { | |
f7f4bccb | 1622 | case JBD2_SUPERBLOCK_V2: |
470decc6 | 1623 | return 0; |
f7f4bccb | 1624 | case JBD2_SUPERBLOCK_V1: |
470decc6 DK |
1625 | return journal_convert_superblock_v1(journal, sb); |
1626 | default: | |
1627 | break; | |
1628 | } | |
1629 | return -EINVAL; | |
1630 | } | |
1631 | ||
1632 | static int journal_convert_superblock_v1(journal_t *journal, | |
1633 | journal_superblock_t *sb) | |
1634 | { | |
1635 | int offset, blocksize; | |
1636 | struct buffer_head *bh; | |
1637 | ||
1638 | printk(KERN_WARNING | |
1639 | "JBD: Converting superblock from version 1 to 2.\n"); | |
1640 | ||
1641 | /* Pre-initialise new fields to zero */ | |
1642 | offset = ((char *) &(sb->s_feature_compat)) - ((char *) sb); | |
1643 | blocksize = be32_to_cpu(sb->s_blocksize); | |
1644 | memset(&sb->s_feature_compat, 0, blocksize-offset); | |
1645 | ||
1646 | sb->s_nr_users = cpu_to_be32(1); | |
f7f4bccb | 1647 | sb->s_header.h_blocktype = cpu_to_be32(JBD2_SUPERBLOCK_V2); |
470decc6 DK |
1648 | journal->j_format_version = 2; |
1649 | ||
1650 | bh = journal->j_sb_buffer; | |
1651 | BUFFER_TRACE(bh, "marking dirty"); | |
1652 | mark_buffer_dirty(bh); | |
1653 | sync_dirty_buffer(bh); | |
1654 | return 0; | |
1655 | } | |
1656 | ||
1657 | ||
1658 | /** | |
f7f4bccb | 1659 | * int jbd2_journal_flush () - Flush journal |
470decc6 DK |
1660 | * @journal: Journal to act on. |
1661 | * | |
1662 | * Flush all data for a given journal to disk and empty the journal. | |
1663 | * Filesystems can use this when remounting readonly to ensure that | |
1664 | * recovery does not need to happen on remount. | |
1665 | */ | |
1666 | ||
f7f4bccb | 1667 | int jbd2_journal_flush(journal_t *journal) |
470decc6 DK |
1668 | { |
1669 | int err = 0; | |
1670 | transaction_t *transaction = NULL; | |
1671 | unsigned long old_tail; | |
1672 | ||
1673 | spin_lock(&journal->j_state_lock); | |
1674 | ||
1675 | /* Force everything buffered to the log... */ | |
1676 | if (journal->j_running_transaction) { | |
1677 | transaction = journal->j_running_transaction; | |
f7f4bccb | 1678 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1679 | } else if (journal->j_committing_transaction) |
1680 | transaction = journal->j_committing_transaction; | |
1681 | ||
1682 | /* Wait for the log commit to complete... */ | |
1683 | if (transaction) { | |
1684 | tid_t tid = transaction->t_tid; | |
1685 | ||
1686 | spin_unlock(&journal->j_state_lock); | |
f7f4bccb | 1687 | jbd2_log_wait_commit(journal, tid); |
470decc6 DK |
1688 | } else { |
1689 | spin_unlock(&journal->j_state_lock); | |
1690 | } | |
1691 | ||
1692 | /* ...and flush everything in the log out to disk. */ | |
1693 | spin_lock(&journal->j_list_lock); | |
1694 | while (!err && journal->j_checkpoint_transactions != NULL) { | |
1695 | spin_unlock(&journal->j_list_lock); | |
f7f4bccb | 1696 | err = jbd2_log_do_checkpoint(journal); |
470decc6 DK |
1697 | spin_lock(&journal->j_list_lock); |
1698 | } | |
1699 | spin_unlock(&journal->j_list_lock); | |
f7f4bccb | 1700 | jbd2_cleanup_journal_tail(journal); |
470decc6 DK |
1701 | |
1702 | /* Finally, mark the journal as really needing no recovery. | |
1703 | * This sets s_start==0 in the underlying superblock, which is | |
1704 | * the magic code for a fully-recovered superblock. Any future | |
1705 | * commits of data to the journal will restore the current | |
1706 | * s_start value. */ | |
1707 | spin_lock(&journal->j_state_lock); | |
1708 | old_tail = journal->j_tail; | |
1709 | journal->j_tail = 0; | |
1710 | spin_unlock(&journal->j_state_lock); | |
f7f4bccb | 1711 | jbd2_journal_update_superblock(journal, 1); |
470decc6 DK |
1712 | spin_lock(&journal->j_state_lock); |
1713 | journal->j_tail = old_tail; | |
1714 | ||
1715 | J_ASSERT(!journal->j_running_transaction); | |
1716 | J_ASSERT(!journal->j_committing_transaction); | |
1717 | J_ASSERT(!journal->j_checkpoint_transactions); | |
1718 | J_ASSERT(journal->j_head == journal->j_tail); | |
1719 | J_ASSERT(journal->j_tail_sequence == journal->j_transaction_sequence); | |
1720 | spin_unlock(&journal->j_state_lock); | |
1721 | return err; | |
1722 | } | |
1723 | ||
1724 | /** | |
f7f4bccb | 1725 | * int jbd2_journal_wipe() - Wipe journal contents |
470decc6 DK |
1726 | * @journal: Journal to act on. |
1727 | * @write: flag (see below) | |
1728 | * | |
1729 | * Wipe out all of the contents of a journal, safely. This will produce | |
1730 | * a warning if the journal contains any valid recovery information. | |
f7f4bccb | 1731 | * Must be called between journal_init_*() and jbd2_journal_load(). |
470decc6 DK |
1732 | * |
1733 | * If 'write' is non-zero, then we wipe out the journal on disk; otherwise | |
1734 | * we merely suppress recovery. | |
1735 | */ | |
1736 | ||
f7f4bccb | 1737 | int jbd2_journal_wipe(journal_t *journal, int write) |
470decc6 DK |
1738 | { |
1739 | journal_superblock_t *sb; | |
1740 | int err = 0; | |
1741 | ||
f7f4bccb | 1742 | J_ASSERT (!(journal->j_flags & JBD2_LOADED)); |
470decc6 DK |
1743 | |
1744 | err = load_superblock(journal); | |
1745 | if (err) | |
1746 | return err; | |
1747 | ||
1748 | sb = journal->j_superblock; | |
1749 | ||
1750 | if (!journal->j_tail) | |
1751 | goto no_recovery; | |
1752 | ||
1753 | printk (KERN_WARNING "JBD: %s recovery information on journal\n", | |
1754 | write ? "Clearing" : "Ignoring"); | |
1755 | ||
f7f4bccb | 1756 | err = jbd2_journal_skip_recovery(journal); |
470decc6 | 1757 | if (write) |
f7f4bccb | 1758 | jbd2_journal_update_superblock(journal, 1); |
470decc6 DK |
1759 | |
1760 | no_recovery: | |
1761 | return err; | |
1762 | } | |
1763 | ||
1764 | /* | |
1765 | * journal_dev_name: format a character string to describe on what | |
1766 | * device this journal is present. | |
1767 | */ | |
1768 | ||
1769 | static const char *journal_dev_name(journal_t *journal, char *buffer) | |
1770 | { | |
1771 | struct block_device *bdev; | |
1772 | ||
1773 | if (journal->j_inode) | |
1774 | bdev = journal->j_inode->i_sb->s_bdev; | |
1775 | else | |
1776 | bdev = journal->j_dev; | |
1777 | ||
1778 | return bdevname(bdev, buffer); | |
1779 | } | |
1780 | ||
1781 | /* | |
1782 | * Journal abort has very specific semantics, which we describe | |
1783 | * for journal abort. | |
1784 | * | |
1785 | * Two internal function, which provide abort to te jbd layer | |
1786 | * itself are here. | |
1787 | */ | |
1788 | ||
1789 | /* | |
1790 | * Quick version for internal journal use (doesn't lock the journal). | |
1791 | * Aborts hard --- we mark the abort as occurred, but do _nothing_ else, | |
1792 | * and don't attempt to make any other journal updates. | |
1793 | */ | |
f7f4bccb | 1794 | void __jbd2_journal_abort_hard(journal_t *journal) |
470decc6 DK |
1795 | { |
1796 | transaction_t *transaction; | |
1797 | char b[BDEVNAME_SIZE]; | |
1798 | ||
f7f4bccb | 1799 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1800 | return; |
1801 | ||
1802 | printk(KERN_ERR "Aborting journal on device %s.\n", | |
1803 | journal_dev_name(journal, b)); | |
1804 | ||
1805 | spin_lock(&journal->j_state_lock); | |
f7f4bccb | 1806 | journal->j_flags |= JBD2_ABORT; |
470decc6 DK |
1807 | transaction = journal->j_running_transaction; |
1808 | if (transaction) | |
f7f4bccb | 1809 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1810 | spin_unlock(&journal->j_state_lock); |
1811 | } | |
1812 | ||
1813 | /* Soft abort: record the abort error status in the journal superblock, | |
1814 | * but don't do any other IO. */ | |
1815 | static void __journal_abort_soft (journal_t *journal, int errno) | |
1816 | { | |
f7f4bccb | 1817 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1818 | return; |
1819 | ||
1820 | if (!journal->j_errno) | |
1821 | journal->j_errno = errno; | |
1822 | ||
f7f4bccb | 1823 | __jbd2_journal_abort_hard(journal); |
470decc6 DK |
1824 | |
1825 | if (errno) | |
f7f4bccb | 1826 | jbd2_journal_update_superblock(journal, 1); |
470decc6 DK |
1827 | } |
1828 | ||
1829 | /** | |
f7f4bccb | 1830 | * void jbd2_journal_abort () - Shutdown the journal immediately. |
470decc6 DK |
1831 | * @journal: the journal to shutdown. |
1832 | * @errno: an error number to record in the journal indicating | |
1833 | * the reason for the shutdown. | |
1834 | * | |
1835 | * Perform a complete, immediate shutdown of the ENTIRE | |
1836 | * journal (not of a single transaction). This operation cannot be | |
1837 | * undone without closing and reopening the journal. | |
1838 | * | |
f7f4bccb | 1839 | * The jbd2_journal_abort function is intended to support higher level error |
470decc6 DK |
1840 | * recovery mechanisms such as the ext2/ext3 remount-readonly error |
1841 | * mode. | |
1842 | * | |
1843 | * Journal abort has very specific semantics. Any existing dirty, | |
1844 | * unjournaled buffers in the main filesystem will still be written to | |
1845 | * disk by bdflush, but the journaling mechanism will be suspended | |
1846 | * immediately and no further transaction commits will be honoured. | |
1847 | * | |
1848 | * Any dirty, journaled buffers will be written back to disk without | |
1849 | * hitting the journal. Atomicity cannot be guaranteed on an aborted | |
1850 | * filesystem, but we _do_ attempt to leave as much data as possible | |
1851 | * behind for fsck to use for cleanup. | |
1852 | * | |
1853 | * Any attempt to get a new transaction handle on a journal which is in | |
1854 | * ABORT state will just result in an -EROFS error return. A | |
f7f4bccb | 1855 | * jbd2_journal_stop on an existing handle will return -EIO if we have |
470decc6 DK |
1856 | * entered abort state during the update. |
1857 | * | |
1858 | * Recursive transactions are not disturbed by journal abort until the | |
f7f4bccb | 1859 | * final jbd2_journal_stop, which will receive the -EIO error. |
470decc6 | 1860 | * |
f7f4bccb | 1861 | * Finally, the jbd2_journal_abort call allows the caller to supply an errno |
470decc6 DK |
1862 | * which will be recorded (if possible) in the journal superblock. This |
1863 | * allows a client to record failure conditions in the middle of a | |
1864 | * transaction without having to complete the transaction to record the | |
1865 | * failure to disk. ext3_error, for example, now uses this | |
1866 | * functionality. | |
1867 | * | |
1868 | * Errors which originate from within the journaling layer will NOT | |
1869 | * supply an errno; a null errno implies that absolutely no further | |
1870 | * writes are done to the journal (unless there are any already in | |
1871 | * progress). | |
1872 | * | |
1873 | */ | |
1874 | ||
f7f4bccb | 1875 | void jbd2_journal_abort(journal_t *journal, int errno) |
470decc6 DK |
1876 | { |
1877 | __journal_abort_soft(journal, errno); | |
1878 | } | |
1879 | ||
1880 | /** | |
f7f4bccb | 1881 | * int jbd2_journal_errno () - returns the journal's error state. |
470decc6 DK |
1882 | * @journal: journal to examine. |
1883 | * | |
f7f4bccb | 1884 | * This is the errno numbet set with jbd2_journal_abort(), the last |
470decc6 DK |
1885 | * time the journal was mounted - if the journal was stopped |
1886 | * without calling abort this will be 0. | |
1887 | * | |
1888 | * If the journal has been aborted on this mount time -EROFS will | |
1889 | * be returned. | |
1890 | */ | |
f7f4bccb | 1891 | int jbd2_journal_errno(journal_t *journal) |
470decc6 DK |
1892 | { |
1893 | int err; | |
1894 | ||
1895 | spin_lock(&journal->j_state_lock); | |
f7f4bccb | 1896 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1897 | err = -EROFS; |
1898 | else | |
1899 | err = journal->j_errno; | |
1900 | spin_unlock(&journal->j_state_lock); | |
1901 | return err; | |
1902 | } | |
1903 | ||
1904 | /** | |
f7f4bccb | 1905 | * int jbd2_journal_clear_err () - clears the journal's error state |
470decc6 DK |
1906 | * @journal: journal to act on. |
1907 | * | |
1908 | * An error must be cleared or Acked to take a FS out of readonly | |
1909 | * mode. | |
1910 | */ | |
f7f4bccb | 1911 | int jbd2_journal_clear_err(journal_t *journal) |
470decc6 DK |
1912 | { |
1913 | int err = 0; | |
1914 | ||
1915 | spin_lock(&journal->j_state_lock); | |
f7f4bccb | 1916 | if (journal->j_flags & JBD2_ABORT) |
470decc6 DK |
1917 | err = -EROFS; |
1918 | else | |
1919 | journal->j_errno = 0; | |
1920 | spin_unlock(&journal->j_state_lock); | |
1921 | return err; | |
1922 | } | |
1923 | ||
1924 | /** | |
f7f4bccb | 1925 | * void jbd2_journal_ack_err() - Ack journal err. |
470decc6 DK |
1926 | * @journal: journal to act on. |
1927 | * | |
1928 | * An error must be cleared or Acked to take a FS out of readonly | |
1929 | * mode. | |
1930 | */ | |
f7f4bccb | 1931 | void jbd2_journal_ack_err(journal_t *journal) |
470decc6 DK |
1932 | { |
1933 | spin_lock(&journal->j_state_lock); | |
1934 | if (journal->j_errno) | |
f7f4bccb | 1935 | journal->j_flags |= JBD2_ACK_ERR; |
470decc6 DK |
1936 | spin_unlock(&journal->j_state_lock); |
1937 | } | |
1938 | ||
f7f4bccb | 1939 | int jbd2_journal_blocks_per_page(struct inode *inode) |
470decc6 DK |
1940 | { |
1941 | return 1 << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits); | |
1942 | } | |
1943 | ||
b517bea1 ZB |
1944 | /* |
1945 | * helper functions to deal with 32 or 64bit block numbers. | |
1946 | */ | |
1947 | size_t journal_tag_bytes(journal_t *journal) | |
1948 | { | |
1949 | if (JBD2_HAS_INCOMPAT_FEATURE(journal, JBD2_FEATURE_INCOMPAT_64BIT)) | |
cd02ff0b | 1950 | return JBD2_TAG_SIZE64; |
b517bea1 | 1951 | else |
cd02ff0b | 1952 | return JBD2_TAG_SIZE32; |
b517bea1 ZB |
1953 | } |
1954 | ||
470decc6 DK |
1955 | /* |
1956 | * Journal_head storage management | |
1957 | */ | |
e18b890b | 1958 | static struct kmem_cache *jbd2_journal_head_cache; |
e23291b9 | 1959 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
1960 | static atomic_t nr_journal_heads = ATOMIC_INIT(0); |
1961 | #endif | |
1962 | ||
f7f4bccb | 1963 | static int journal_init_jbd2_journal_head_cache(void) |
470decc6 DK |
1964 | { |
1965 | int retval; | |
1966 | ||
1076d17a | 1967 | J_ASSERT(jbd2_journal_head_cache == NULL); |
a920e941 | 1968 | jbd2_journal_head_cache = kmem_cache_create("jbd2_journal_head", |
470decc6 DK |
1969 | sizeof(struct journal_head), |
1970 | 0, /* offset */ | |
77160957 | 1971 | SLAB_TEMPORARY, /* flags */ |
20c2df83 | 1972 | NULL); /* ctor */ |
470decc6 | 1973 | retval = 0; |
1076d17a | 1974 | if (!jbd2_journal_head_cache) { |
470decc6 DK |
1975 | retval = -ENOMEM; |
1976 | printk(KERN_EMERG "JBD: no memory for journal_head cache\n"); | |
1977 | } | |
1978 | return retval; | |
1979 | } | |
1980 | ||
f7f4bccb | 1981 | static void jbd2_journal_destroy_jbd2_journal_head_cache(void) |
470decc6 | 1982 | { |
8a9362eb DG |
1983 | if (jbd2_journal_head_cache) { |
1984 | kmem_cache_destroy(jbd2_journal_head_cache); | |
1985 | jbd2_journal_head_cache = NULL; | |
1986 | } | |
470decc6 DK |
1987 | } |
1988 | ||
1989 | /* | |
1990 | * journal_head splicing and dicing | |
1991 | */ | |
1992 | static struct journal_head *journal_alloc_journal_head(void) | |
1993 | { | |
1994 | struct journal_head *ret; | |
1995 | static unsigned long last_warning; | |
1996 | ||
e23291b9 | 1997 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
1998 | atomic_inc(&nr_journal_heads); |
1999 | #endif | |
f7f4bccb | 2000 | ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS); |
1076d17a | 2001 | if (!ret) { |
470decc6 DK |
2002 | jbd_debug(1, "out of memory for journal_head\n"); |
2003 | if (time_after(jiffies, last_warning + 5*HZ)) { | |
2004 | printk(KERN_NOTICE "ENOMEM in %s, retrying.\n", | |
329d291f | 2005 | __func__); |
470decc6 DK |
2006 | last_warning = jiffies; |
2007 | } | |
1076d17a | 2008 | while (!ret) { |
470decc6 | 2009 | yield(); |
f7f4bccb | 2010 | ret = kmem_cache_alloc(jbd2_journal_head_cache, GFP_NOFS); |
470decc6 DK |
2011 | } |
2012 | } | |
2013 | return ret; | |
2014 | } | |
2015 | ||
2016 | static void journal_free_journal_head(struct journal_head *jh) | |
2017 | { | |
e23291b9 | 2018 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 | 2019 | atomic_dec(&nr_journal_heads); |
cd02ff0b | 2020 | memset(jh, JBD2_POISON_FREE, sizeof(*jh)); |
470decc6 | 2021 | #endif |
f7f4bccb | 2022 | kmem_cache_free(jbd2_journal_head_cache, jh); |
470decc6 DK |
2023 | } |
2024 | ||
2025 | /* | |
2026 | * A journal_head is attached to a buffer_head whenever JBD has an | |
2027 | * interest in the buffer. | |
2028 | * | |
2029 | * Whenever a buffer has an attached journal_head, its ->b_state:BH_JBD bit | |
2030 | * is set. This bit is tested in core kernel code where we need to take | |
2031 | * JBD-specific actions. Testing the zeroness of ->b_private is not reliable | |
2032 | * there. | |
2033 | * | |
2034 | * When a buffer has its BH_JBD bit set, its ->b_count is elevated by one. | |
2035 | * | |
2036 | * When a buffer has its BH_JBD bit set it is immune from being released by | |
2037 | * core kernel code, mainly via ->b_count. | |
2038 | * | |
2039 | * A journal_head may be detached from its buffer_head when the journal_head's | |
2040 | * b_transaction, b_cp_transaction and b_next_transaction pointers are NULL. | |
f7f4bccb | 2041 | * Various places in JBD call jbd2_journal_remove_journal_head() to indicate that the |
470decc6 DK |
2042 | * journal_head can be dropped if needed. |
2043 | * | |
2044 | * Various places in the kernel want to attach a journal_head to a buffer_head | |
2045 | * _before_ attaching the journal_head to a transaction. To protect the | |
f7f4bccb | 2046 | * journal_head in this situation, jbd2_journal_add_journal_head elevates the |
470decc6 | 2047 | * journal_head's b_jcount refcount by one. The caller must call |
f7f4bccb | 2048 | * jbd2_journal_put_journal_head() to undo this. |
470decc6 DK |
2049 | * |
2050 | * So the typical usage would be: | |
2051 | * | |
2052 | * (Attach a journal_head if needed. Increments b_jcount) | |
f7f4bccb | 2053 | * struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
2054 | * ... |
2055 | * jh->b_transaction = xxx; | |
f7f4bccb | 2056 | * jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2057 | * |
2058 | * Now, the journal_head's b_jcount is zero, but it is safe from being released | |
2059 | * because it has a non-zero b_transaction. | |
2060 | */ | |
2061 | ||
2062 | /* | |
2063 | * Give a buffer_head a journal_head. | |
2064 | * | |
2065 | * Doesn't need the journal lock. | |
2066 | * May sleep. | |
2067 | */ | |
f7f4bccb | 2068 | struct journal_head *jbd2_journal_add_journal_head(struct buffer_head *bh) |
470decc6 DK |
2069 | { |
2070 | struct journal_head *jh; | |
2071 | struct journal_head *new_jh = NULL; | |
2072 | ||
2073 | repeat: | |
2074 | if (!buffer_jbd(bh)) { | |
2075 | new_jh = journal_alloc_journal_head(); | |
2076 | memset(new_jh, 0, sizeof(*new_jh)); | |
2077 | } | |
2078 | ||
2079 | jbd_lock_bh_journal_head(bh); | |
2080 | if (buffer_jbd(bh)) { | |
2081 | jh = bh2jh(bh); | |
2082 | } else { | |
2083 | J_ASSERT_BH(bh, | |
2084 | (atomic_read(&bh->b_count) > 0) || | |
2085 | (bh->b_page && bh->b_page->mapping)); | |
2086 | ||
2087 | if (!new_jh) { | |
2088 | jbd_unlock_bh_journal_head(bh); | |
2089 | goto repeat; | |
2090 | } | |
2091 | ||
2092 | jh = new_jh; | |
2093 | new_jh = NULL; /* We consumed it */ | |
2094 | set_buffer_jbd(bh); | |
2095 | bh->b_private = jh; | |
2096 | jh->b_bh = bh; | |
2097 | get_bh(bh); | |
2098 | BUFFER_TRACE(bh, "added journal_head"); | |
2099 | } | |
2100 | jh->b_jcount++; | |
2101 | jbd_unlock_bh_journal_head(bh); | |
2102 | if (new_jh) | |
2103 | journal_free_journal_head(new_jh); | |
2104 | return bh->b_private; | |
2105 | } | |
2106 | ||
2107 | /* | |
2108 | * Grab a ref against this buffer_head's journal_head. If it ended up not | |
2109 | * having a journal_head, return NULL | |
2110 | */ | |
f7f4bccb | 2111 | struct journal_head *jbd2_journal_grab_journal_head(struct buffer_head *bh) |
470decc6 DK |
2112 | { |
2113 | struct journal_head *jh = NULL; | |
2114 | ||
2115 | jbd_lock_bh_journal_head(bh); | |
2116 | if (buffer_jbd(bh)) { | |
2117 | jh = bh2jh(bh); | |
2118 | jh->b_jcount++; | |
2119 | } | |
2120 | jbd_unlock_bh_journal_head(bh); | |
2121 | return jh; | |
2122 | } | |
2123 | ||
2124 | static void __journal_remove_journal_head(struct buffer_head *bh) | |
2125 | { | |
2126 | struct journal_head *jh = bh2jh(bh); | |
2127 | ||
2128 | J_ASSERT_JH(jh, jh->b_jcount >= 0); | |
2129 | ||
2130 | get_bh(bh); | |
2131 | if (jh->b_jcount == 0) { | |
2132 | if (jh->b_transaction == NULL && | |
2133 | jh->b_next_transaction == NULL && | |
2134 | jh->b_cp_transaction == NULL) { | |
2135 | J_ASSERT_JH(jh, jh->b_jlist == BJ_None); | |
2136 | J_ASSERT_BH(bh, buffer_jbd(bh)); | |
2137 | J_ASSERT_BH(bh, jh2bh(jh) == bh); | |
2138 | BUFFER_TRACE(bh, "remove journal_head"); | |
2139 | if (jh->b_frozen_data) { | |
2140 | printk(KERN_WARNING "%s: freeing " | |
2141 | "b_frozen_data\n", | |
329d291f | 2142 | __func__); |
af1e76d6 | 2143 | jbd2_free(jh->b_frozen_data, bh->b_size); |
470decc6 DK |
2144 | } |
2145 | if (jh->b_committed_data) { | |
2146 | printk(KERN_WARNING "%s: freeing " | |
2147 | "b_committed_data\n", | |
329d291f | 2148 | __func__); |
af1e76d6 | 2149 | jbd2_free(jh->b_committed_data, bh->b_size); |
470decc6 DK |
2150 | } |
2151 | bh->b_private = NULL; | |
2152 | jh->b_bh = NULL; /* debug, really */ | |
2153 | clear_buffer_jbd(bh); | |
2154 | __brelse(bh); | |
2155 | journal_free_journal_head(jh); | |
2156 | } else { | |
2157 | BUFFER_TRACE(bh, "journal_head was locked"); | |
2158 | } | |
2159 | } | |
2160 | } | |
2161 | ||
2162 | /* | |
f7f4bccb | 2163 | * jbd2_journal_remove_journal_head(): if the buffer isn't attached to a transaction |
470decc6 DK |
2164 | * and has a zero b_jcount then remove and release its journal_head. If we did |
2165 | * see that the buffer is not used by any transaction we also "logically" | |
2166 | * decrement ->b_count. | |
2167 | * | |
2168 | * We in fact take an additional increment on ->b_count as a convenience, | |
2169 | * because the caller usually wants to do additional things with the bh | |
2170 | * after calling here. | |
f7f4bccb | 2171 | * The caller of jbd2_journal_remove_journal_head() *must* run __brelse(bh) at some |
470decc6 DK |
2172 | * time. Once the caller has run __brelse(), the buffer is eligible for |
2173 | * reaping by try_to_free_buffers(). | |
2174 | */ | |
f7f4bccb | 2175 | void jbd2_journal_remove_journal_head(struct buffer_head *bh) |
470decc6 DK |
2176 | { |
2177 | jbd_lock_bh_journal_head(bh); | |
2178 | __journal_remove_journal_head(bh); | |
2179 | jbd_unlock_bh_journal_head(bh); | |
2180 | } | |
2181 | ||
2182 | /* | |
2183 | * Drop a reference on the passed journal_head. If it fell to zero then try to | |
2184 | * release the journal_head from the buffer_head. | |
2185 | */ | |
f7f4bccb | 2186 | void jbd2_journal_put_journal_head(struct journal_head *jh) |
470decc6 DK |
2187 | { |
2188 | struct buffer_head *bh = jh2bh(jh); | |
2189 | ||
2190 | jbd_lock_bh_journal_head(bh); | |
2191 | J_ASSERT_JH(jh, jh->b_jcount > 0); | |
2192 | --jh->b_jcount; | |
2193 | if (!jh->b_jcount && !jh->b_transaction) { | |
2194 | __journal_remove_journal_head(bh); | |
2195 | __brelse(bh); | |
2196 | } | |
2197 | jbd_unlock_bh_journal_head(bh); | |
2198 | } | |
2199 | ||
c851ed54 JK |
2200 | /* |
2201 | * Initialize jbd inode head | |
2202 | */ | |
2203 | void jbd2_journal_init_jbd_inode(struct jbd2_inode *jinode, struct inode *inode) | |
2204 | { | |
2205 | jinode->i_transaction = NULL; | |
2206 | jinode->i_next_transaction = NULL; | |
2207 | jinode->i_vfs_inode = inode; | |
2208 | jinode->i_flags = 0; | |
2209 | INIT_LIST_HEAD(&jinode->i_list); | |
2210 | } | |
2211 | ||
2212 | /* | |
2213 | * Function to be called before we start removing inode from memory (i.e., | |
2214 | * clear_inode() is a fine place to be called from). It removes inode from | |
2215 | * transaction's lists. | |
2216 | */ | |
2217 | void jbd2_journal_release_jbd_inode(journal_t *journal, | |
2218 | struct jbd2_inode *jinode) | |
2219 | { | |
2220 | int writeout = 0; | |
2221 | ||
2222 | if (!journal) | |
2223 | return; | |
2224 | restart: | |
2225 | spin_lock(&journal->j_list_lock); | |
2226 | /* Is commit writing out inode - we have to wait */ | |
2227 | if (jinode->i_flags & JI_COMMIT_RUNNING) { | |
2228 | wait_queue_head_t *wq; | |
2229 | DEFINE_WAIT_BIT(wait, &jinode->i_flags, __JI_COMMIT_RUNNING); | |
2230 | wq = bit_waitqueue(&jinode->i_flags, __JI_COMMIT_RUNNING); | |
2231 | prepare_to_wait(wq, &wait.wait, TASK_UNINTERRUPTIBLE); | |
2232 | spin_unlock(&journal->j_list_lock); | |
2233 | schedule(); | |
2234 | finish_wait(wq, &wait.wait); | |
2235 | goto restart; | |
2236 | } | |
2237 | ||
2238 | /* Do we need to wait for data writeback? */ | |
2239 | if (journal->j_committing_transaction == jinode->i_transaction) | |
2240 | writeout = 1; | |
2241 | if (jinode->i_transaction) { | |
2242 | list_del(&jinode->i_list); | |
2243 | jinode->i_transaction = NULL; | |
2244 | } | |
2245 | spin_unlock(&journal->j_list_lock); | |
2246 | } | |
2247 | ||
470decc6 | 2248 | /* |
0f49d5d0 | 2249 | * debugfs tunables |
470decc6 | 2250 | */ |
6f38c74f JS |
2251 | #ifdef CONFIG_JBD2_DEBUG |
2252 | u8 jbd2_journal_enable_debug __read_mostly; | |
f7f4bccb | 2253 | EXPORT_SYMBOL(jbd2_journal_enable_debug); |
470decc6 | 2254 | |
0f49d5d0 | 2255 | #define JBD2_DEBUG_NAME "jbd2-debug" |
470decc6 | 2256 | |
6f38c74f JS |
2257 | static struct dentry *jbd2_debugfs_dir; |
2258 | static struct dentry *jbd2_debug; | |
470decc6 | 2259 | |
0f49d5d0 JS |
2260 | static void __init jbd2_create_debugfs_entry(void) |
2261 | { | |
2262 | jbd2_debugfs_dir = debugfs_create_dir("jbd2", NULL); | |
2263 | if (jbd2_debugfs_dir) | |
2264 | jbd2_debug = debugfs_create_u8(JBD2_DEBUG_NAME, S_IRUGO, | |
2265 | jbd2_debugfs_dir, | |
2266 | &jbd2_journal_enable_debug); | |
470decc6 DK |
2267 | } |
2268 | ||
0f49d5d0 | 2269 | static void __exit jbd2_remove_debugfs_entry(void) |
470decc6 | 2270 | { |
6f38c74f JS |
2271 | debugfs_remove(jbd2_debug); |
2272 | debugfs_remove(jbd2_debugfs_dir); | |
470decc6 DK |
2273 | } |
2274 | ||
0f49d5d0 | 2275 | #else |
470decc6 | 2276 | |
0f49d5d0 | 2277 | static void __init jbd2_create_debugfs_entry(void) |
470decc6 | 2278 | { |
470decc6 DK |
2279 | } |
2280 | ||
0f49d5d0 | 2281 | static void __exit jbd2_remove_debugfs_entry(void) |
470decc6 | 2282 | { |
470decc6 DK |
2283 | } |
2284 | ||
470decc6 DK |
2285 | #endif |
2286 | ||
8e85fb3f JL |
2287 | #ifdef CONFIG_PROC_FS |
2288 | ||
2289 | #define JBD2_STATS_PROC_NAME "fs/jbd2" | |
2290 | ||
2291 | static void __init jbd2_create_jbd_stats_proc_entry(void) | |
2292 | { | |
2293 | proc_jbd2_stats = proc_mkdir(JBD2_STATS_PROC_NAME, NULL); | |
2294 | } | |
2295 | ||
2296 | static void __exit jbd2_remove_jbd_stats_proc_entry(void) | |
2297 | { | |
2298 | if (proc_jbd2_stats) | |
2299 | remove_proc_entry(JBD2_STATS_PROC_NAME, NULL); | |
2300 | } | |
2301 | ||
2302 | #else | |
2303 | ||
2304 | #define jbd2_create_jbd_stats_proc_entry() do {} while (0) | |
2305 | #define jbd2_remove_jbd_stats_proc_entry() do {} while (0) | |
2306 | ||
2307 | #endif | |
2308 | ||
e18b890b | 2309 | struct kmem_cache *jbd2_handle_cache; |
470decc6 DK |
2310 | |
2311 | static int __init journal_init_handle_cache(void) | |
2312 | { | |
a920e941 | 2313 | jbd2_handle_cache = kmem_cache_create("jbd2_journal_handle", |
470decc6 DK |
2314 | sizeof(handle_t), |
2315 | 0, /* offset */ | |
77160957 | 2316 | SLAB_TEMPORARY, /* flags */ |
20c2df83 | 2317 | NULL); /* ctor */ |
f7f4bccb | 2318 | if (jbd2_handle_cache == NULL) { |
470decc6 DK |
2319 | printk(KERN_EMERG "JBD: failed to create handle cache\n"); |
2320 | return -ENOMEM; | |
2321 | } | |
2322 | return 0; | |
2323 | } | |
2324 | ||
f7f4bccb | 2325 | static void jbd2_journal_destroy_handle_cache(void) |
470decc6 | 2326 | { |
f7f4bccb MC |
2327 | if (jbd2_handle_cache) |
2328 | kmem_cache_destroy(jbd2_handle_cache); | |
470decc6 DK |
2329 | } |
2330 | ||
2331 | /* | |
2332 | * Module startup and shutdown | |
2333 | */ | |
2334 | ||
2335 | static int __init journal_init_caches(void) | |
2336 | { | |
2337 | int ret; | |
2338 | ||
f7f4bccb | 2339 | ret = jbd2_journal_init_revoke_caches(); |
470decc6 | 2340 | if (ret == 0) |
f7f4bccb | 2341 | ret = journal_init_jbd2_journal_head_cache(); |
470decc6 DK |
2342 | if (ret == 0) |
2343 | ret = journal_init_handle_cache(); | |
2344 | return ret; | |
2345 | } | |
2346 | ||
f7f4bccb | 2347 | static void jbd2_journal_destroy_caches(void) |
470decc6 | 2348 | { |
f7f4bccb MC |
2349 | jbd2_journal_destroy_revoke_caches(); |
2350 | jbd2_journal_destroy_jbd2_journal_head_cache(); | |
2351 | jbd2_journal_destroy_handle_cache(); | |
470decc6 DK |
2352 | } |
2353 | ||
2354 | static int __init journal_init(void) | |
2355 | { | |
2356 | int ret; | |
2357 | ||
2358 | BUILD_BUG_ON(sizeof(struct journal_superblock_s) != 1024); | |
2359 | ||
2360 | ret = journal_init_caches(); | |
620de4e1 DG |
2361 | if (ret == 0) { |
2362 | jbd2_create_debugfs_entry(); | |
2363 | jbd2_create_jbd_stats_proc_entry(); | |
2364 | } else { | |
f7f4bccb | 2365 | jbd2_journal_destroy_caches(); |
620de4e1 | 2366 | } |
470decc6 DK |
2367 | return ret; |
2368 | } | |
2369 | ||
2370 | static void __exit journal_exit(void) | |
2371 | { | |
e23291b9 | 2372 | #ifdef CONFIG_JBD2_DEBUG |
470decc6 DK |
2373 | int n = atomic_read(&nr_journal_heads); |
2374 | if (n) | |
2375 | printk(KERN_EMERG "JBD: leaked %d journal_heads!\n", n); | |
2376 | #endif | |
0f49d5d0 | 2377 | jbd2_remove_debugfs_entry(); |
8e85fb3f | 2378 | jbd2_remove_jbd_stats_proc_entry(); |
f7f4bccb | 2379 | jbd2_journal_destroy_caches(); |
470decc6 DK |
2380 | } |
2381 | ||
2382 | MODULE_LICENSE("GPL"); | |
2383 | module_init(journal_init); | |
2384 | module_exit(journal_exit); | |
2385 |