Root/
1 | /* |
2 | * linux/fs/jbd2/transaction.c |
3 | * |
4 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1998 |
5 | * |
6 | * Copyright 1998 Red Hat corp --- All Rights Reserved |
7 | * |
8 | * This file is part of the Linux kernel and is made available under |
9 | * the terms of the GNU General Public License, version 2, or at your |
10 | * option, any later version, incorporated herein by reference. |
11 | * |
12 | * Generic filesystem transaction handling code; part of the ext2fs |
13 | * journaling system. |
14 | * |
15 | * This file manages transactions (compound commits managed by the |
16 | * journaling code) and handles (individual atomic operations by the |
17 | * filesystem). |
18 | */ |
19 | |
20 | #include <linux/time.h> |
21 | #include <linux/fs.h> |
22 | #include <linux/jbd2.h> |
23 | #include <linux/errno.h> |
24 | #include <linux/slab.h> |
25 | #include <linux/timer.h> |
26 | #include <linux/mm.h> |
27 | #include <linux/highmem.h> |
28 | #include <linux/hrtimer.h> |
29 | |
30 | static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh); |
31 | |
32 | /* |
33 | * jbd2_get_transaction: obtain a new transaction_t object. |
34 | * |
35 | * Simply allocate and initialise a new transaction. Create it in |
36 | * RUNNING state and add it to the current journal (which should not |
37 | * have an existing running transaction: we only make a new transaction |
38 | * once we have started to commit the old one). |
39 | * |
40 | * Preconditions: |
41 | * The journal MUST be locked. We don't perform atomic mallocs on the |
42 | * new transaction and we can't block without protecting against other |
43 | * processes trying to touch the journal while it is in transition. |
44 | * |
45 | */ |
46 | |
47 | static transaction_t * |
48 | jbd2_get_transaction(journal_t *journal, transaction_t *transaction) |
49 | { |
50 | transaction->t_journal = journal; |
51 | transaction->t_state = T_RUNNING; |
52 | transaction->t_start_time = ktime_get(); |
53 | transaction->t_tid = journal->j_transaction_sequence++; |
54 | transaction->t_expires = jiffies + journal->j_commit_interval; |
55 | spin_lock_init(&transaction->t_handle_lock); |
56 | INIT_LIST_HEAD(&transaction->t_inode_list); |
57 | INIT_LIST_HEAD(&transaction->t_private_list); |
58 | |
59 | /* Set up the commit timer for the new transaction. */ |
60 | journal->j_commit_timer.expires = round_jiffies(transaction->t_expires); |
61 | add_timer(&journal->j_commit_timer); |
62 | |
63 | J_ASSERT(journal->j_running_transaction == NULL); |
64 | journal->j_running_transaction = transaction; |
65 | transaction->t_max_wait = 0; |
66 | transaction->t_start = jiffies; |
67 | |
68 | return transaction; |
69 | } |
70 | |
71 | /* |
72 | * Handle management. |
73 | * |
74 | * A handle_t is an object which represents a single atomic update to a |
75 | * filesystem, and which tracks all of the modifications which form part |
76 | * of that one update. |
77 | */ |
78 | |
79 | /* |
80 | * start_this_handle: Given a handle, deal with any locking or stalling |
81 | * needed to make sure that there is enough journal space for the handle |
82 | * to begin. Attach the handle to a transaction and set up the |
83 | * transaction's buffer credits. |
84 | */ |
85 | |
86 | static int start_this_handle(journal_t *journal, handle_t *handle) |
87 | { |
88 | transaction_t *transaction; |
89 | int needed; |
90 | int nblocks = handle->h_buffer_credits; |
91 | transaction_t *new_transaction = NULL; |
92 | int ret = 0; |
93 | unsigned long ts = jiffies; |
94 | |
95 | if (nblocks > journal->j_max_transaction_buffers) { |
96 | printk(KERN_ERR "JBD: %s wants too many credits (%d > %d)\n", |
97 | current->comm, nblocks, |
98 | journal->j_max_transaction_buffers); |
99 | ret = -ENOSPC; |
100 | goto out; |
101 | } |
102 | |
103 | alloc_transaction: |
104 | if (!journal->j_running_transaction) { |
105 | new_transaction = kzalloc(sizeof(*new_transaction), |
106 | GFP_NOFS|__GFP_NOFAIL); |
107 | if (!new_transaction) { |
108 | ret = -ENOMEM; |
109 | goto out; |
110 | } |
111 | } |
112 | |
113 | jbd_debug(3, "New handle %p going live.\n", handle); |
114 | |
115 | repeat: |
116 | |
117 | /* |
118 | * We need to hold j_state_lock until t_updates has been incremented, |
119 | * for proper journal barrier handling |
120 | */ |
121 | spin_lock(&journal->j_state_lock); |
122 | repeat_locked: |
123 | if (is_journal_aborted(journal) || |
124 | (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) { |
125 | spin_unlock(&journal->j_state_lock); |
126 | ret = -EROFS; |
127 | goto out; |
128 | } |
129 | |
130 | /* Wait on the journal's transaction barrier if necessary */ |
131 | if (journal->j_barrier_count) { |
132 | spin_unlock(&journal->j_state_lock); |
133 | wait_event(journal->j_wait_transaction_locked, |
134 | journal->j_barrier_count == 0); |
135 | goto repeat; |
136 | } |
137 | |
138 | if (!journal->j_running_transaction) { |
139 | if (!new_transaction) { |
140 | spin_unlock(&journal->j_state_lock); |
141 | goto alloc_transaction; |
142 | } |
143 | jbd2_get_transaction(journal, new_transaction); |
144 | new_transaction = NULL; |
145 | } |
146 | |
147 | transaction = journal->j_running_transaction; |
148 | |
149 | /* |
150 | * If the current transaction is locked down for commit, wait for the |
151 | * lock to be released. |
152 | */ |
153 | if (transaction->t_state == T_LOCKED) { |
154 | DEFINE_WAIT(wait); |
155 | |
156 | prepare_to_wait(&journal->j_wait_transaction_locked, |
157 | &wait, TASK_UNINTERRUPTIBLE); |
158 | spin_unlock(&journal->j_state_lock); |
159 | schedule(); |
160 | finish_wait(&journal->j_wait_transaction_locked, &wait); |
161 | goto repeat; |
162 | } |
163 | |
164 | /* |
165 | * If there is not enough space left in the log to write all potential |
166 | * buffers requested by this operation, we need to stall pending a log |
167 | * checkpoint to free some more log space. |
168 | */ |
169 | spin_lock(&transaction->t_handle_lock); |
170 | needed = transaction->t_outstanding_credits + nblocks; |
171 | |
172 | if (needed > journal->j_max_transaction_buffers) { |
173 | /* |
174 | * If the current transaction is already too large, then start |
175 | * to commit it: we can then go back and attach this handle to |
176 | * a new transaction. |
177 | */ |
178 | DEFINE_WAIT(wait); |
179 | |
180 | jbd_debug(2, "Handle %p starting new commit...\n", handle); |
181 | spin_unlock(&transaction->t_handle_lock); |
182 | prepare_to_wait(&journal->j_wait_transaction_locked, &wait, |
183 | TASK_UNINTERRUPTIBLE); |
184 | __jbd2_log_start_commit(journal, transaction->t_tid); |
185 | spin_unlock(&journal->j_state_lock); |
186 | schedule(); |
187 | finish_wait(&journal->j_wait_transaction_locked, &wait); |
188 | goto repeat; |
189 | } |
190 | |
191 | /* |
192 | * The commit code assumes that it can get enough log space |
193 | * without forcing a checkpoint. This is *critical* for |
194 | * correctness: a checkpoint of a buffer which is also |
195 | * associated with a committing transaction creates a deadlock, |
196 | * so commit simply cannot force through checkpoints. |
197 | * |
198 | * We must therefore ensure the necessary space in the journal |
199 | * *before* starting to dirty potentially checkpointed buffers |
200 | * in the new transaction. |
201 | * |
202 | * The worst part is, any transaction currently committing can |
203 | * reduce the free space arbitrarily. Be careful to account for |
204 | * those buffers when checkpointing. |
205 | */ |
206 | |
207 | /* |
208 | * @@@ AKPM: This seems rather over-defensive. We're giving commit |
209 | * a _lot_ of headroom: 1/4 of the journal plus the size of |
210 | * the committing transaction. Really, we only need to give it |
211 | * committing_transaction->t_outstanding_credits plus "enough" for |
212 | * the log control blocks. |
213 | * Also, this test is inconsitent with the matching one in |
214 | * jbd2_journal_extend(). |
215 | */ |
216 | if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) { |
217 | jbd_debug(2, "Handle %p waiting for checkpoint...\n", handle); |
218 | spin_unlock(&transaction->t_handle_lock); |
219 | __jbd2_log_wait_for_space(journal); |
220 | goto repeat_locked; |
221 | } |
222 | |
223 | /* OK, account for the buffers that this operation expects to |
224 | * use and add the handle to the running transaction. */ |
225 | |
226 | if (time_after(transaction->t_start, ts)) { |
227 | ts = jbd2_time_diff(ts, transaction->t_start); |
228 | if (ts > transaction->t_max_wait) |
229 | transaction->t_max_wait = ts; |
230 | } |
231 | |
232 | handle->h_transaction = transaction; |
233 | transaction->t_outstanding_credits += nblocks; |
234 | transaction->t_updates++; |
235 | transaction->t_handle_count++; |
236 | jbd_debug(4, "Handle %p given %d credits (total %d, free %d)\n", |
237 | handle, nblocks, transaction->t_outstanding_credits, |
238 | __jbd2_log_space_left(journal)); |
239 | spin_unlock(&transaction->t_handle_lock); |
240 | spin_unlock(&journal->j_state_lock); |
241 | out: |
242 | if (unlikely(new_transaction)) /* It's usually NULL */ |
243 | kfree(new_transaction); |
244 | return ret; |
245 | } |
246 | |
247 | static struct lock_class_key jbd2_handle_key; |
248 | |
249 | /* Allocate a new handle. This should probably be in a slab... */ |
250 | static handle_t *new_handle(int nblocks) |
251 | { |
252 | handle_t *handle = jbd2_alloc_handle(GFP_NOFS); |
253 | if (!handle) |
254 | return NULL; |
255 | memset(handle, 0, sizeof(*handle)); |
256 | handle->h_buffer_credits = nblocks; |
257 | handle->h_ref = 1; |
258 | |
259 | lockdep_init_map(&handle->h_lockdep_map, "jbd2_handle", |
260 | &jbd2_handle_key, 0); |
261 | |
262 | return handle; |
263 | } |
264 | |
265 | /** |
266 | * handle_t *jbd2_journal_start() - Obtain a new handle. |
267 | * @journal: Journal to start transaction on. |
268 | * @nblocks: number of block buffer we might modify |
269 | * |
270 | * We make sure that the transaction can guarantee at least nblocks of |
271 | * modified buffers in the log. We block until the log can guarantee |
272 | * that much space. |
273 | * |
274 | * This function is visible to journal users (like ext3fs), so is not |
275 | * called with the journal already locked. |
276 | * |
277 | * Return a pointer to a newly allocated handle, or NULL on failure |
278 | */ |
279 | handle_t *jbd2_journal_start(journal_t *journal, int nblocks) |
280 | { |
281 | handle_t *handle = journal_current_handle(); |
282 | int err; |
283 | |
284 | if (!journal) |
285 | return ERR_PTR(-EROFS); |
286 | |
287 | if (handle) { |
288 | J_ASSERT(handle->h_transaction->t_journal == journal); |
289 | handle->h_ref++; |
290 | return handle; |
291 | } |
292 | |
293 | handle = new_handle(nblocks); |
294 | if (!handle) |
295 | return ERR_PTR(-ENOMEM); |
296 | |
297 | current->journal_info = handle; |
298 | |
299 | err = start_this_handle(journal, handle); |
300 | if (err < 0) { |
301 | jbd2_free_handle(handle); |
302 | current->journal_info = NULL; |
303 | handle = ERR_PTR(err); |
304 | goto out; |
305 | } |
306 | |
307 | lock_map_acquire(&handle->h_lockdep_map); |
308 | out: |
309 | return handle; |
310 | } |
311 | |
312 | /** |
313 | * int jbd2_journal_extend() - extend buffer credits. |
314 | * @handle: handle to 'extend' |
315 | * @nblocks: nr blocks to try to extend by. |
316 | * |
317 | * Some transactions, such as large extends and truncates, can be done |
318 | * atomically all at once or in several stages. The operation requests |
319 | * a credit for a number of buffer modications in advance, but can |
320 | * extend its credit if it needs more. |
321 | * |
322 | * jbd2_journal_extend tries to give the running handle more buffer credits. |
323 | * It does not guarantee that allocation - this is a best-effort only. |
324 | * The calling process MUST be able to deal cleanly with a failure to |
325 | * extend here. |
326 | * |
327 | * Return 0 on success, non-zero on failure. |
328 | * |
329 | * return code < 0 implies an error |
330 | * return code > 0 implies normal transaction-full status. |
331 | */ |
332 | int jbd2_journal_extend(handle_t *handle, int nblocks) |
333 | { |
334 | transaction_t *transaction = handle->h_transaction; |
335 | journal_t *journal = transaction->t_journal; |
336 | int result; |
337 | int wanted; |
338 | |
339 | result = -EIO; |
340 | if (is_handle_aborted(handle)) |
341 | goto out; |
342 | |
343 | result = 1; |
344 | |
345 | spin_lock(&journal->j_state_lock); |
346 | |
347 | /* Don't extend a locked-down transaction! */ |
348 | if (handle->h_transaction->t_state != T_RUNNING) { |
349 | jbd_debug(3, "denied handle %p %d blocks: " |
350 | "transaction not running\n", handle, nblocks); |
351 | goto error_out; |
352 | } |
353 | |
354 | spin_lock(&transaction->t_handle_lock); |
355 | wanted = transaction->t_outstanding_credits + nblocks; |
356 | |
357 | if (wanted > journal->j_max_transaction_buffers) { |
358 | jbd_debug(3, "denied handle %p %d blocks: " |
359 | "transaction too large\n", handle, nblocks); |
360 | goto unlock; |
361 | } |
362 | |
363 | if (wanted > __jbd2_log_space_left(journal)) { |
364 | jbd_debug(3, "denied handle %p %d blocks: " |
365 | "insufficient log space\n", handle, nblocks); |
366 | goto unlock; |
367 | } |
368 | |
369 | handle->h_buffer_credits += nblocks; |
370 | transaction->t_outstanding_credits += nblocks; |
371 | result = 0; |
372 | |
373 | jbd_debug(3, "extended handle %p by %d\n", handle, nblocks); |
374 | unlock: |
375 | spin_unlock(&transaction->t_handle_lock); |
376 | error_out: |
377 | spin_unlock(&journal->j_state_lock); |
378 | out: |
379 | return result; |
380 | } |
381 | |
382 | |
383 | /** |
384 | * int jbd2_journal_restart() - restart a handle . |
385 | * @handle: handle to restart |
386 | * @nblocks: nr credits requested |
387 | * |
388 | * Restart a handle for a multi-transaction filesystem |
389 | * operation. |
390 | * |
391 | * If the jbd2_journal_extend() call above fails to grant new buffer credits |
392 | * to a running handle, a call to jbd2_journal_restart will commit the |
393 | * handle's transaction so far and reattach the handle to a new |
394 | * transaction capabable of guaranteeing the requested number of |
395 | * credits. |
396 | */ |
397 | |
398 | int jbd2_journal_restart(handle_t *handle, int nblocks) |
399 | { |
400 | transaction_t *transaction = handle->h_transaction; |
401 | journal_t *journal = transaction->t_journal; |
402 | int ret; |
403 | |
404 | /* If we've had an abort of any type, don't even think about |
405 | * actually doing the restart! */ |
406 | if (is_handle_aborted(handle)) |
407 | return 0; |
408 | |
409 | /* |
410 | * First unlink the handle from its current transaction, and start the |
411 | * commit on that. |
412 | */ |
413 | J_ASSERT(transaction->t_updates > 0); |
414 | J_ASSERT(journal_current_handle() == handle); |
415 | |
416 | spin_lock(&journal->j_state_lock); |
417 | spin_lock(&transaction->t_handle_lock); |
418 | transaction->t_outstanding_credits -= handle->h_buffer_credits; |
419 | transaction->t_updates--; |
420 | |
421 | if (!transaction->t_updates) |
422 | wake_up(&journal->j_wait_updates); |
423 | spin_unlock(&transaction->t_handle_lock); |
424 | |
425 | jbd_debug(2, "restarting handle %p\n", handle); |
426 | __jbd2_log_start_commit(journal, transaction->t_tid); |
427 | spin_unlock(&journal->j_state_lock); |
428 | |
429 | handle->h_buffer_credits = nblocks; |
430 | ret = start_this_handle(journal, handle); |
431 | return ret; |
432 | } |
433 | |
434 | |
435 | /** |
436 | * void jbd2_journal_lock_updates () - establish a transaction barrier. |
437 | * @journal: Journal to establish a barrier on. |
438 | * |
439 | * This locks out any further updates from being started, and blocks |
440 | * until all existing updates have completed, returning only once the |
441 | * journal is in a quiescent state with no updates running. |
442 | * |
443 | * The journal lock should not be held on entry. |
444 | */ |
445 | void jbd2_journal_lock_updates(journal_t *journal) |
446 | { |
447 | DEFINE_WAIT(wait); |
448 | |
449 | spin_lock(&journal->j_state_lock); |
450 | ++journal->j_barrier_count; |
451 | |
452 | /* Wait until there are no running updates */ |
453 | while (1) { |
454 | transaction_t *transaction = journal->j_running_transaction; |
455 | |
456 | if (!transaction) |
457 | break; |
458 | |
459 | spin_lock(&transaction->t_handle_lock); |
460 | if (!transaction->t_updates) { |
461 | spin_unlock(&transaction->t_handle_lock); |
462 | break; |
463 | } |
464 | prepare_to_wait(&journal->j_wait_updates, &wait, |
465 | TASK_UNINTERRUPTIBLE); |
466 | spin_unlock(&transaction->t_handle_lock); |
467 | spin_unlock(&journal->j_state_lock); |
468 | schedule(); |
469 | finish_wait(&journal->j_wait_updates, &wait); |
470 | spin_lock(&journal->j_state_lock); |
471 | } |
472 | spin_unlock(&journal->j_state_lock); |
473 | |
474 | /* |
475 | * We have now established a barrier against other normal updates, but |
476 | * we also need to barrier against other jbd2_journal_lock_updates() calls |
477 | * to make sure that we serialise special journal-locked operations |
478 | * too. |
479 | */ |
480 | mutex_lock(&journal->j_barrier); |
481 | } |
482 | |
483 | /** |
484 | * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier |
485 | * @journal: Journal to release the barrier on. |
486 | * |
487 | * Release a transaction barrier obtained with jbd2_journal_lock_updates(). |
488 | * |
489 | * Should be called without the journal lock held. |
490 | */ |
491 | void jbd2_journal_unlock_updates (journal_t *journal) |
492 | { |
493 | J_ASSERT(journal->j_barrier_count != 0); |
494 | |
495 | mutex_unlock(&journal->j_barrier); |
496 | spin_lock(&journal->j_state_lock); |
497 | --journal->j_barrier_count; |
498 | spin_unlock(&journal->j_state_lock); |
499 | wake_up(&journal->j_wait_transaction_locked); |
500 | } |
501 | |
502 | static void warn_dirty_buffer(struct buffer_head *bh) |
503 | { |
504 | char b[BDEVNAME_SIZE]; |
505 | |
506 | printk(KERN_WARNING |
507 | "JBD: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). " |
508 | "There's a risk of filesystem corruption in case of system " |
509 | "crash.\n", |
510 | bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr); |
511 | } |
512 | |
513 | /* |
514 | * If the buffer is already part of the current transaction, then there |
515 | * is nothing we need to do. If it is already part of a prior |
516 | * transaction which we are still committing to disk, then we need to |
517 | * make sure that we do not overwrite the old copy: we do copy-out to |
518 | * preserve the copy going to disk. We also account the buffer against |
519 | * the handle's metadata buffer credits (unless the buffer is already |
520 | * part of the transaction, that is). |
521 | * |
522 | */ |
523 | static int |
524 | do_get_write_access(handle_t *handle, struct journal_head *jh, |
525 | int force_copy) |
526 | { |
527 | struct buffer_head *bh; |
528 | transaction_t *transaction; |
529 | journal_t *journal; |
530 | int error; |
531 | char *frozen_buffer = NULL; |
532 | int need_copy = 0; |
533 | |
534 | if (is_handle_aborted(handle)) |
535 | return -EROFS; |
536 | |
537 | transaction = handle->h_transaction; |
538 | journal = transaction->t_journal; |
539 | |
540 | jbd_debug(5, "buffer_head %p, force_copy %d\n", jh, force_copy); |
541 | |
542 | JBUFFER_TRACE(jh, "entry"); |
543 | repeat: |
544 | bh = jh2bh(jh); |
545 | |
546 | /* @@@ Need to check for errors here at some point. */ |
547 | |
548 | lock_buffer(bh); |
549 | jbd_lock_bh_state(bh); |
550 | |
551 | /* We now hold the buffer lock so it is safe to query the buffer |
552 | * state. Is the buffer dirty? |
553 | * |
554 | * If so, there are two possibilities. The buffer may be |
555 | * non-journaled, and undergoing a quite legitimate writeback. |
556 | * Otherwise, it is journaled, and we don't expect dirty buffers |
557 | * in that state (the buffers should be marked JBD_Dirty |
558 | * instead.) So either the IO is being done under our own |
559 | * control and this is a bug, or it's a third party IO such as |
560 | * dump(8) (which may leave the buffer scheduled for read --- |
561 | * ie. locked but not dirty) or tune2fs (which may actually have |
562 | * the buffer dirtied, ugh.) */ |
563 | |
564 | if (buffer_dirty(bh)) { |
565 | /* |
566 | * First question: is this buffer already part of the current |
567 | * transaction or the existing committing transaction? |
568 | */ |
569 | if (jh->b_transaction) { |
570 | J_ASSERT_JH(jh, |
571 | jh->b_transaction == transaction || |
572 | jh->b_transaction == |
573 | journal->j_committing_transaction); |
574 | if (jh->b_next_transaction) |
575 | J_ASSERT_JH(jh, jh->b_next_transaction == |
576 | transaction); |
577 | warn_dirty_buffer(bh); |
578 | } |
579 | /* |
580 | * In any case we need to clean the dirty flag and we must |
581 | * do it under the buffer lock to be sure we don't race |
582 | * with running write-out. |
583 | */ |
584 | JBUFFER_TRACE(jh, "Journalling dirty buffer"); |
585 | clear_buffer_dirty(bh); |
586 | set_buffer_jbddirty(bh); |
587 | } |
588 | |
589 | unlock_buffer(bh); |
590 | |
591 | error = -EROFS; |
592 | if (is_handle_aborted(handle)) { |
593 | jbd_unlock_bh_state(bh); |
594 | goto out; |
595 | } |
596 | error = 0; |
597 | |
598 | /* |
599 | * The buffer is already part of this transaction if b_transaction or |
600 | * b_next_transaction points to it |
601 | */ |
602 | if (jh->b_transaction == transaction || |
603 | jh->b_next_transaction == transaction) |
604 | goto done; |
605 | |
606 | /* |
607 | * this is the first time this transaction is touching this buffer, |
608 | * reset the modified flag |
609 | */ |
610 | jh->b_modified = 0; |
611 | |
612 | /* |
613 | * If there is already a copy-out version of this buffer, then we don't |
614 | * need to make another one |
615 | */ |
616 | if (jh->b_frozen_data) { |
617 | JBUFFER_TRACE(jh, "has frozen data"); |
618 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); |
619 | jh->b_next_transaction = transaction; |
620 | goto done; |
621 | } |
622 | |
623 | /* Is there data here we need to preserve? */ |
624 | |
625 | if (jh->b_transaction && jh->b_transaction != transaction) { |
626 | JBUFFER_TRACE(jh, "owned by older transaction"); |
627 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); |
628 | J_ASSERT_JH(jh, jh->b_transaction == |
629 | journal->j_committing_transaction); |
630 | |
631 | /* There is one case we have to be very careful about. |
632 | * If the committing transaction is currently writing |
633 | * this buffer out to disk and has NOT made a copy-out, |
634 | * then we cannot modify the buffer contents at all |
635 | * right now. The essence of copy-out is that it is the |
636 | * extra copy, not the primary copy, which gets |
637 | * journaled. If the primary copy is already going to |
638 | * disk then we cannot do copy-out here. */ |
639 | |
640 | if (jh->b_jlist == BJ_Shadow) { |
641 | DEFINE_WAIT_BIT(wait, &bh->b_state, BH_Unshadow); |
642 | wait_queue_head_t *wqh; |
643 | |
644 | wqh = bit_waitqueue(&bh->b_state, BH_Unshadow); |
645 | |
646 | JBUFFER_TRACE(jh, "on shadow: sleep"); |
647 | jbd_unlock_bh_state(bh); |
648 | /* commit wakes up all shadow buffers after IO */ |
649 | for ( ; ; ) { |
650 | prepare_to_wait(wqh, &wait.wait, |
651 | TASK_UNINTERRUPTIBLE); |
652 | if (jh->b_jlist != BJ_Shadow) |
653 | break; |
654 | schedule(); |
655 | } |
656 | finish_wait(wqh, &wait.wait); |
657 | goto repeat; |
658 | } |
659 | |
660 | /* Only do the copy if the currently-owning transaction |
661 | * still needs it. If it is on the Forget list, the |
662 | * committing transaction is past that stage. The |
663 | * buffer had better remain locked during the kmalloc, |
664 | * but that should be true --- we hold the journal lock |
665 | * still and the buffer is already on the BUF_JOURNAL |
666 | * list so won't be flushed. |
667 | * |
668 | * Subtle point, though: if this is a get_undo_access, |
669 | * then we will be relying on the frozen_data to contain |
670 | * the new value of the committed_data record after the |
671 | * transaction, so we HAVE to force the frozen_data copy |
672 | * in that case. */ |
673 | |
674 | if (jh->b_jlist != BJ_Forget || force_copy) { |
675 | JBUFFER_TRACE(jh, "generate frozen data"); |
676 | if (!frozen_buffer) { |
677 | JBUFFER_TRACE(jh, "allocate memory for buffer"); |
678 | jbd_unlock_bh_state(bh); |
679 | frozen_buffer = |
680 | jbd2_alloc(jh2bh(jh)->b_size, |
681 | GFP_NOFS); |
682 | if (!frozen_buffer) { |
683 | printk(KERN_EMERG |
684 | "%s: OOM for frozen_buffer\n", |
685 | __func__); |
686 | JBUFFER_TRACE(jh, "oom!"); |
687 | error = -ENOMEM; |
688 | jbd_lock_bh_state(bh); |
689 | goto done; |
690 | } |
691 | goto repeat; |
692 | } |
693 | jh->b_frozen_data = frozen_buffer; |
694 | frozen_buffer = NULL; |
695 | need_copy = 1; |
696 | } |
697 | jh->b_next_transaction = transaction; |
698 | } |
699 | |
700 | |
701 | /* |
702 | * Finally, if the buffer is not journaled right now, we need to make |
703 | * sure it doesn't get written to disk before the caller actually |
704 | * commits the new data |
705 | */ |
706 | if (!jh->b_transaction) { |
707 | JBUFFER_TRACE(jh, "no transaction"); |
708 | J_ASSERT_JH(jh, !jh->b_next_transaction); |
709 | jh->b_transaction = transaction; |
710 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); |
711 | spin_lock(&journal->j_list_lock); |
712 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
713 | spin_unlock(&journal->j_list_lock); |
714 | } |
715 | |
716 | done: |
717 | if (need_copy) { |
718 | struct page *page; |
719 | int offset; |
720 | char *source; |
721 | |
722 | J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)), |
723 | "Possible IO failure.\n"); |
724 | page = jh2bh(jh)->b_page; |
725 | offset = ((unsigned long) jh2bh(jh)->b_data) & ~PAGE_MASK; |
726 | source = kmap_atomic(page, KM_USER0); |
727 | memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size); |
728 | kunmap_atomic(source, KM_USER0); |
729 | |
730 | /* |
731 | * Now that the frozen data is saved off, we need to store |
732 | * any matching triggers. |
733 | */ |
734 | jh->b_frozen_triggers = jh->b_triggers; |
735 | } |
736 | jbd_unlock_bh_state(bh); |
737 | |
738 | /* |
739 | * If we are about to journal a buffer, then any revoke pending on it is |
740 | * no longer valid |
741 | */ |
742 | jbd2_journal_cancel_revoke(handle, jh); |
743 | |
744 | out: |
745 | if (unlikely(frozen_buffer)) /* It's usually NULL */ |
746 | jbd2_free(frozen_buffer, bh->b_size); |
747 | |
748 | JBUFFER_TRACE(jh, "exit"); |
749 | return error; |
750 | } |
751 | |
752 | /** |
753 | * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update. |
754 | * @handle: transaction to add buffer modifications to |
755 | * @bh: bh to be used for metadata writes |
756 | * @credits: variable that will receive credits for the buffer |
757 | * |
758 | * Returns an error code or 0 on success. |
759 | * |
760 | * In full data journalling mode the buffer may be of type BJ_AsyncData, |
761 | * because we're write()ing a buffer which is also part of a shared mapping. |
762 | */ |
763 | |
764 | int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh) |
765 | { |
766 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
767 | int rc; |
768 | |
769 | /* We do not want to get caught playing with fields which the |
770 | * log thread also manipulates. Make sure that the buffer |
771 | * completes any outstanding IO before proceeding. */ |
772 | rc = do_get_write_access(handle, jh, 0); |
773 | jbd2_journal_put_journal_head(jh); |
774 | return rc; |
775 | } |
776 | |
777 | |
778 | /* |
779 | * When the user wants to journal a newly created buffer_head |
780 | * (ie. getblk() returned a new buffer and we are going to populate it |
781 | * manually rather than reading off disk), then we need to keep the |
782 | * buffer_head locked until it has been completely filled with new |
783 | * data. In this case, we should be able to make the assertion that |
784 | * the bh is not already part of an existing transaction. |
785 | * |
786 | * The buffer should already be locked by the caller by this point. |
787 | * There is no lock ranking violation: it was a newly created, |
788 | * unlocked buffer beforehand. */ |
789 | |
790 | /** |
791 | * int jbd2_journal_get_create_access () - notify intent to use newly created bh |
792 | * @handle: transaction to new buffer to |
793 | * @bh: new buffer. |
794 | * |
795 | * Call this if you create a new bh. |
796 | */ |
797 | int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh) |
798 | { |
799 | transaction_t *transaction = handle->h_transaction; |
800 | journal_t *journal = transaction->t_journal; |
801 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
802 | int err; |
803 | |
804 | jbd_debug(5, "journal_head %p\n", jh); |
805 | err = -EROFS; |
806 | if (is_handle_aborted(handle)) |
807 | goto out; |
808 | err = 0; |
809 | |
810 | JBUFFER_TRACE(jh, "entry"); |
811 | /* |
812 | * The buffer may already belong to this transaction due to pre-zeroing |
813 | * in the filesystem's new_block code. It may also be on the previous, |
814 | * committing transaction's lists, but it HAS to be in Forget state in |
815 | * that case: the transaction must have deleted the buffer for it to be |
816 | * reused here. |
817 | */ |
818 | jbd_lock_bh_state(bh); |
819 | spin_lock(&journal->j_list_lock); |
820 | J_ASSERT_JH(jh, (jh->b_transaction == transaction || |
821 | jh->b_transaction == NULL || |
822 | (jh->b_transaction == journal->j_committing_transaction && |
823 | jh->b_jlist == BJ_Forget))); |
824 | |
825 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); |
826 | J_ASSERT_JH(jh, buffer_locked(jh2bh(jh))); |
827 | |
828 | if (jh->b_transaction == NULL) { |
829 | /* |
830 | * Previous jbd2_journal_forget() could have left the buffer |
831 | * with jbddirty bit set because it was being committed. When |
832 | * the commit finished, we've filed the buffer for |
833 | * checkpointing and marked it dirty. Now we are reallocating |
834 | * the buffer so the transaction freeing it must have |
835 | * committed and so it's safe to clear the dirty bit. |
836 | */ |
837 | clear_buffer_dirty(jh2bh(jh)); |
838 | jh->b_transaction = transaction; |
839 | |
840 | /* first access by this transaction */ |
841 | jh->b_modified = 0; |
842 | |
843 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); |
844 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
845 | } else if (jh->b_transaction == journal->j_committing_transaction) { |
846 | /* first access by this transaction */ |
847 | jh->b_modified = 0; |
848 | |
849 | JBUFFER_TRACE(jh, "set next transaction"); |
850 | jh->b_next_transaction = transaction; |
851 | } |
852 | spin_unlock(&journal->j_list_lock); |
853 | jbd_unlock_bh_state(bh); |
854 | |
855 | /* |
856 | * akpm: I added this. ext3_alloc_branch can pick up new indirect |
857 | * blocks which contain freed but then revoked metadata. We need |
858 | * to cancel the revoke in case we end up freeing it yet again |
859 | * and the reallocating as data - this would cause a second revoke, |
860 | * which hits an assertion error. |
861 | */ |
862 | JBUFFER_TRACE(jh, "cancelling revoke"); |
863 | jbd2_journal_cancel_revoke(handle, jh); |
864 | jbd2_journal_put_journal_head(jh); |
865 | out: |
866 | return err; |
867 | } |
868 | |
869 | /** |
870 | * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with |
871 | * non-rewindable consequences |
872 | * @handle: transaction |
873 | * @bh: buffer to undo |
874 | * @credits: store the number of taken credits here (if not NULL) |
875 | * |
876 | * Sometimes there is a need to distinguish between metadata which has |
877 | * been committed to disk and that which has not. The ext3fs code uses |
878 | * this for freeing and allocating space, we have to make sure that we |
879 | * do not reuse freed space until the deallocation has been committed, |
880 | * since if we overwrote that space we would make the delete |
881 | * un-rewindable in case of a crash. |
882 | * |
883 | * To deal with that, jbd2_journal_get_undo_access requests write access to a |
884 | * buffer for parts of non-rewindable operations such as delete |
885 | * operations on the bitmaps. The journaling code must keep a copy of |
886 | * the buffer's contents prior to the undo_access call until such time |
887 | * as we know that the buffer has definitely been committed to disk. |
888 | * |
889 | * We never need to know which transaction the committed data is part |
890 | * of, buffers touched here are guaranteed to be dirtied later and so |
891 | * will be committed to a new transaction in due course, at which point |
892 | * we can discard the old committed data pointer. |
893 | * |
894 | * Returns error number or 0 on success. |
895 | */ |
896 | int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh) |
897 | { |
898 | int err; |
899 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
900 | char *committed_data = NULL; |
901 | |
902 | JBUFFER_TRACE(jh, "entry"); |
903 | |
904 | /* |
905 | * Do this first --- it can drop the journal lock, so we want to |
906 | * make sure that obtaining the committed_data is done |
907 | * atomically wrt. completion of any outstanding commits. |
908 | */ |
909 | err = do_get_write_access(handle, jh, 1); |
910 | if (err) |
911 | goto out; |
912 | |
913 | repeat: |
914 | if (!jh->b_committed_data) { |
915 | committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS); |
916 | if (!committed_data) { |
917 | printk(KERN_EMERG "%s: No memory for committed data\n", |
918 | __func__); |
919 | err = -ENOMEM; |
920 | goto out; |
921 | } |
922 | } |
923 | |
924 | jbd_lock_bh_state(bh); |
925 | if (!jh->b_committed_data) { |
926 | /* Copy out the current buffer contents into the |
927 | * preserved, committed copy. */ |
928 | JBUFFER_TRACE(jh, "generate b_committed data"); |
929 | if (!committed_data) { |
930 | jbd_unlock_bh_state(bh); |
931 | goto repeat; |
932 | } |
933 | |
934 | jh->b_committed_data = committed_data; |
935 | committed_data = NULL; |
936 | memcpy(jh->b_committed_data, bh->b_data, bh->b_size); |
937 | } |
938 | jbd_unlock_bh_state(bh); |
939 | out: |
940 | jbd2_journal_put_journal_head(jh); |
941 | if (unlikely(committed_data)) |
942 | jbd2_free(committed_data, bh->b_size); |
943 | return err; |
944 | } |
945 | |
946 | /** |
947 | * void jbd2_journal_set_triggers() - Add triggers for commit writeout |
948 | * @bh: buffer to trigger on |
949 | * @type: struct jbd2_buffer_trigger_type containing the trigger(s). |
950 | * |
951 | * Set any triggers on this journal_head. This is always safe, because |
952 | * triggers for a committing buffer will be saved off, and triggers for |
953 | * a running transaction will match the buffer in that transaction. |
954 | * |
955 | * Call with NULL to clear the triggers. |
956 | */ |
957 | void jbd2_journal_set_triggers(struct buffer_head *bh, |
958 | struct jbd2_buffer_trigger_type *type) |
959 | { |
960 | struct journal_head *jh = bh2jh(bh); |
961 | |
962 | jh->b_triggers = type; |
963 | } |
964 | |
965 | void jbd2_buffer_commit_trigger(struct journal_head *jh, void *mapped_data, |
966 | struct jbd2_buffer_trigger_type *triggers) |
967 | { |
968 | struct buffer_head *bh = jh2bh(jh); |
969 | |
970 | if (!triggers || !triggers->t_commit) |
971 | return; |
972 | |
973 | triggers->t_commit(triggers, bh, mapped_data, bh->b_size); |
974 | } |
975 | |
976 | void jbd2_buffer_abort_trigger(struct journal_head *jh, |
977 | struct jbd2_buffer_trigger_type *triggers) |
978 | { |
979 | if (!triggers || !triggers->t_abort) |
980 | return; |
981 | |
982 | triggers->t_abort(triggers, jh2bh(jh)); |
983 | } |
984 | |
985 | |
986 | |
987 | /** |
988 | * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata |
989 | * @handle: transaction to add buffer to. |
990 | * @bh: buffer to mark |
991 | * |
992 | * mark dirty metadata which needs to be journaled as part of the current |
993 | * transaction. |
994 | * |
995 | * The buffer is placed on the transaction's metadata list and is marked |
996 | * as belonging to the transaction. |
997 | * |
998 | * Returns error number or 0 on success. |
999 | * |
1000 | * Special care needs to be taken if the buffer already belongs to the |
1001 | * current committing transaction (in which case we should have frozen |
1002 | * data present for that commit). In that case, we don't relink the |
1003 | * buffer: that only gets done when the old transaction finally |
1004 | * completes its commit. |
1005 | */ |
1006 | int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh) |
1007 | { |
1008 | transaction_t *transaction = handle->h_transaction; |
1009 | journal_t *journal = transaction->t_journal; |
1010 | struct journal_head *jh = bh2jh(bh); |
1011 | |
1012 | jbd_debug(5, "journal_head %p\n", jh); |
1013 | JBUFFER_TRACE(jh, "entry"); |
1014 | if (is_handle_aborted(handle)) |
1015 | goto out; |
1016 | |
1017 | jbd_lock_bh_state(bh); |
1018 | |
1019 | if (jh->b_modified == 0) { |
1020 | /* |
1021 | * This buffer's got modified and becoming part |
1022 | * of the transaction. This needs to be done |
1023 | * once a transaction -bzzz |
1024 | */ |
1025 | jh->b_modified = 1; |
1026 | J_ASSERT_JH(jh, handle->h_buffer_credits > 0); |
1027 | handle->h_buffer_credits--; |
1028 | } |
1029 | |
1030 | /* |
1031 | * fastpath, to avoid expensive locking. If this buffer is already |
1032 | * on the running transaction's metadata list there is nothing to do. |
1033 | * Nobody can take it off again because there is a handle open. |
1034 | * I _think_ we're OK here with SMP barriers - a mistaken decision will |
1035 | * result in this test being false, so we go in and take the locks. |
1036 | */ |
1037 | if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) { |
1038 | JBUFFER_TRACE(jh, "fastpath"); |
1039 | J_ASSERT_JH(jh, jh->b_transaction == |
1040 | journal->j_running_transaction); |
1041 | goto out_unlock_bh; |
1042 | } |
1043 | |
1044 | set_buffer_jbddirty(bh); |
1045 | |
1046 | /* |
1047 | * Metadata already on the current transaction list doesn't |
1048 | * need to be filed. Metadata on another transaction's list must |
1049 | * be committing, and will be refiled once the commit completes: |
1050 | * leave it alone for now. |
1051 | */ |
1052 | if (jh->b_transaction != transaction) { |
1053 | JBUFFER_TRACE(jh, "already on other transaction"); |
1054 | J_ASSERT_JH(jh, jh->b_transaction == |
1055 | journal->j_committing_transaction); |
1056 | J_ASSERT_JH(jh, jh->b_next_transaction == transaction); |
1057 | /* And this case is illegal: we can't reuse another |
1058 | * transaction's data buffer, ever. */ |
1059 | goto out_unlock_bh; |
1060 | } |
1061 | |
1062 | /* That test should have eliminated the following case: */ |
1063 | J_ASSERT_JH(jh, jh->b_frozen_data == NULL); |
1064 | |
1065 | JBUFFER_TRACE(jh, "file as BJ_Metadata"); |
1066 | spin_lock(&journal->j_list_lock); |
1067 | __jbd2_journal_file_buffer(jh, handle->h_transaction, BJ_Metadata); |
1068 | spin_unlock(&journal->j_list_lock); |
1069 | out_unlock_bh: |
1070 | jbd_unlock_bh_state(bh); |
1071 | out: |
1072 | JBUFFER_TRACE(jh, "exit"); |
1073 | return 0; |
1074 | } |
1075 | |
1076 | /* |
1077 | * jbd2_journal_release_buffer: undo a get_write_access without any buffer |
1078 | * updates, if the update decided in the end that it didn't need access. |
1079 | * |
1080 | */ |
1081 | void |
1082 | jbd2_journal_release_buffer(handle_t *handle, struct buffer_head *bh) |
1083 | { |
1084 | BUFFER_TRACE(bh, "entry"); |
1085 | } |
1086 | |
1087 | /** |
1088 | * void jbd2_journal_forget() - bforget() for potentially-journaled buffers. |
1089 | * @handle: transaction handle |
1090 | * @bh: bh to 'forget' |
1091 | * |
1092 | * We can only do the bforget if there are no commits pending against the |
1093 | * buffer. If the buffer is dirty in the current running transaction we |
1094 | * can safely unlink it. |
1095 | * |
1096 | * bh may not be a journalled buffer at all - it may be a non-JBD |
1097 | * buffer which came off the hashtable. Check for this. |
1098 | * |
1099 | * Decrements bh->b_count by one. |
1100 | * |
1101 | * Allow this call even if the handle has aborted --- it may be part of |
1102 | * the caller's cleanup after an abort. |
1103 | */ |
1104 | int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh) |
1105 | { |
1106 | transaction_t *transaction = handle->h_transaction; |
1107 | journal_t *journal = transaction->t_journal; |
1108 | struct journal_head *jh; |
1109 | int drop_reserve = 0; |
1110 | int err = 0; |
1111 | int was_modified = 0; |
1112 | |
1113 | BUFFER_TRACE(bh, "entry"); |
1114 | |
1115 | jbd_lock_bh_state(bh); |
1116 | spin_lock(&journal->j_list_lock); |
1117 | |
1118 | if (!buffer_jbd(bh)) |
1119 | goto not_jbd; |
1120 | jh = bh2jh(bh); |
1121 | |
1122 | /* Critical error: attempting to delete a bitmap buffer, maybe? |
1123 | * Don't do any jbd operations, and return an error. */ |
1124 | if (!J_EXPECT_JH(jh, !jh->b_committed_data, |
1125 | "inconsistent data on disk")) { |
1126 | err = -EIO; |
1127 | goto not_jbd; |
1128 | } |
1129 | |
1130 | /* keep track of wether or not this transaction modified us */ |
1131 | was_modified = jh->b_modified; |
1132 | |
1133 | /* |
1134 | * The buffer's going from the transaction, we must drop |
1135 | * all references -bzzz |
1136 | */ |
1137 | jh->b_modified = 0; |
1138 | |
1139 | if (jh->b_transaction == handle->h_transaction) { |
1140 | J_ASSERT_JH(jh, !jh->b_frozen_data); |
1141 | |
1142 | /* If we are forgetting a buffer which is already part |
1143 | * of this transaction, then we can just drop it from |
1144 | * the transaction immediately. */ |
1145 | clear_buffer_dirty(bh); |
1146 | clear_buffer_jbddirty(bh); |
1147 | |
1148 | JBUFFER_TRACE(jh, "belongs to current transaction: unfile"); |
1149 | |
1150 | /* |
1151 | * we only want to drop a reference if this transaction |
1152 | * modified the buffer |
1153 | */ |
1154 | if (was_modified) |
1155 | drop_reserve = 1; |
1156 | |
1157 | /* |
1158 | * We are no longer going to journal this buffer. |
1159 | * However, the commit of this transaction is still |
1160 | * important to the buffer: the delete that we are now |
1161 | * processing might obsolete an old log entry, so by |
1162 | * committing, we can satisfy the buffer's checkpoint. |
1163 | * |
1164 | * So, if we have a checkpoint on the buffer, we should |
1165 | * now refile the buffer on our BJ_Forget list so that |
1166 | * we know to remove the checkpoint after we commit. |
1167 | */ |
1168 | |
1169 | if (jh->b_cp_transaction) { |
1170 | __jbd2_journal_temp_unlink_buffer(jh); |
1171 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); |
1172 | } else { |
1173 | __jbd2_journal_unfile_buffer(jh); |
1174 | jbd2_journal_remove_journal_head(bh); |
1175 | __brelse(bh); |
1176 | if (!buffer_jbd(bh)) { |
1177 | spin_unlock(&journal->j_list_lock); |
1178 | jbd_unlock_bh_state(bh); |
1179 | __bforget(bh); |
1180 | goto drop; |
1181 | } |
1182 | } |
1183 | } else if (jh->b_transaction) { |
1184 | J_ASSERT_JH(jh, (jh->b_transaction == |
1185 | journal->j_committing_transaction)); |
1186 | /* However, if the buffer is still owned by a prior |
1187 | * (committing) transaction, we can't drop it yet... */ |
1188 | JBUFFER_TRACE(jh, "belongs to older transaction"); |
1189 | /* ... but we CAN drop it from the new transaction if we |
1190 | * have also modified it since the original commit. */ |
1191 | |
1192 | if (jh->b_next_transaction) { |
1193 | J_ASSERT(jh->b_next_transaction == transaction); |
1194 | jh->b_next_transaction = NULL; |
1195 | |
1196 | /* |
1197 | * only drop a reference if this transaction modified |
1198 | * the buffer |
1199 | */ |
1200 | if (was_modified) |
1201 | drop_reserve = 1; |
1202 | } |
1203 | } |
1204 | |
1205 | not_jbd: |
1206 | spin_unlock(&journal->j_list_lock); |
1207 | jbd_unlock_bh_state(bh); |
1208 | __brelse(bh); |
1209 | drop: |
1210 | if (drop_reserve) { |
1211 | /* no need to reserve log space for this block -bzzz */ |
1212 | handle->h_buffer_credits++; |
1213 | } |
1214 | return err; |
1215 | } |
1216 | |
1217 | /** |
1218 | * int jbd2_journal_stop() - complete a transaction |
1219 | * @handle: tranaction to complete. |
1220 | * |
1221 | * All done for a particular handle. |
1222 | * |
1223 | * There is not much action needed here. We just return any remaining |
1224 | * buffer credits to the transaction and remove the handle. The only |
1225 | * complication is that we need to start a commit operation if the |
1226 | * filesystem is marked for synchronous update. |
1227 | * |
1228 | * jbd2_journal_stop itself will not usually return an error, but it may |
1229 | * do so in unusual circumstances. In particular, expect it to |
1230 | * return -EIO if a jbd2_journal_abort has been executed since the |
1231 | * transaction began. |
1232 | */ |
1233 | int jbd2_journal_stop(handle_t *handle) |
1234 | { |
1235 | transaction_t *transaction = handle->h_transaction; |
1236 | journal_t *journal = transaction->t_journal; |
1237 | int err; |
1238 | pid_t pid; |
1239 | |
1240 | J_ASSERT(journal_current_handle() == handle); |
1241 | |
1242 | if (is_handle_aborted(handle)) |
1243 | err = -EIO; |
1244 | else { |
1245 | J_ASSERT(transaction->t_updates > 0); |
1246 | err = 0; |
1247 | } |
1248 | |
1249 | if (--handle->h_ref > 0) { |
1250 | jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1, |
1251 | handle->h_ref); |
1252 | return err; |
1253 | } |
1254 | |
1255 | jbd_debug(4, "Handle %p going down\n", handle); |
1256 | |
1257 | /* |
1258 | * Implement synchronous transaction batching. If the handle |
1259 | * was synchronous, don't force a commit immediately. Let's |
1260 | * yield and let another thread piggyback onto this |
1261 | * transaction. Keep doing that while new threads continue to |
1262 | * arrive. It doesn't cost much - we're about to run a commit |
1263 | * and sleep on IO anyway. Speeds up many-threaded, many-dir |
1264 | * operations by 30x or more... |
1265 | * |
1266 | * We try and optimize the sleep time against what the |
1267 | * underlying disk can do, instead of having a static sleep |
1268 | * time. This is useful for the case where our storage is so |
1269 | * fast that it is more optimal to go ahead and force a flush |
1270 | * and wait for the transaction to be committed than it is to |
1271 | * wait for an arbitrary amount of time for new writers to |
1272 | * join the transaction. We achieve this by measuring how |
1273 | * long it takes to commit a transaction, and compare it with |
1274 | * how long this transaction has been running, and if run time |
1275 | * < commit time then we sleep for the delta and commit. This |
1276 | * greatly helps super fast disks that would see slowdowns as |
1277 | * more threads started doing fsyncs. |
1278 | * |
1279 | * But don't do this if this process was the most recent one |
1280 | * to perform a synchronous write. We do this to detect the |
1281 | * case where a single process is doing a stream of sync |
1282 | * writes. No point in waiting for joiners in that case. |
1283 | */ |
1284 | pid = current->pid; |
1285 | if (handle->h_sync && journal->j_last_sync_writer != pid) { |
1286 | u64 commit_time, trans_time; |
1287 | |
1288 | journal->j_last_sync_writer = pid; |
1289 | |
1290 | spin_lock(&journal->j_state_lock); |
1291 | commit_time = journal->j_average_commit_time; |
1292 | spin_unlock(&journal->j_state_lock); |
1293 | |
1294 | trans_time = ktime_to_ns(ktime_sub(ktime_get(), |
1295 | transaction->t_start_time)); |
1296 | |
1297 | commit_time = max_t(u64, commit_time, |
1298 | 1000*journal->j_min_batch_time); |
1299 | commit_time = min_t(u64, commit_time, |
1300 | 1000*journal->j_max_batch_time); |
1301 | |
1302 | if (trans_time < commit_time) { |
1303 | ktime_t expires = ktime_add_ns(ktime_get(), |
1304 | commit_time); |
1305 | set_current_state(TASK_UNINTERRUPTIBLE); |
1306 | schedule_hrtimeout(&expires, HRTIMER_MODE_ABS); |
1307 | } |
1308 | } |
1309 | |
1310 | if (handle->h_sync) |
1311 | transaction->t_synchronous_commit = 1; |
1312 | current->journal_info = NULL; |
1313 | spin_lock(&journal->j_state_lock); |
1314 | spin_lock(&transaction->t_handle_lock); |
1315 | transaction->t_outstanding_credits -= handle->h_buffer_credits; |
1316 | transaction->t_updates--; |
1317 | if (!transaction->t_updates) { |
1318 | wake_up(&journal->j_wait_updates); |
1319 | if (journal->j_barrier_count) |
1320 | wake_up(&journal->j_wait_transaction_locked); |
1321 | } |
1322 | |
1323 | /* |
1324 | * If the handle is marked SYNC, we need to set another commit |
1325 | * going! We also want to force a commit if the current |
1326 | * transaction is occupying too much of the log, or if the |
1327 | * transaction is too old now. |
1328 | */ |
1329 | if (handle->h_sync || |
1330 | transaction->t_outstanding_credits > |
1331 | journal->j_max_transaction_buffers || |
1332 | time_after_eq(jiffies, transaction->t_expires)) { |
1333 | /* Do this even for aborted journals: an abort still |
1334 | * completes the commit thread, it just doesn't write |
1335 | * anything to disk. */ |
1336 | tid_t tid = transaction->t_tid; |
1337 | |
1338 | spin_unlock(&transaction->t_handle_lock); |
1339 | jbd_debug(2, "transaction too old, requesting commit for " |
1340 | "handle %p\n", handle); |
1341 | /* This is non-blocking */ |
1342 | __jbd2_log_start_commit(journal, transaction->t_tid); |
1343 | spin_unlock(&journal->j_state_lock); |
1344 | |
1345 | /* |
1346 | * Special case: JBD2_SYNC synchronous updates require us |
1347 | * to wait for the commit to complete. |
1348 | */ |
1349 | if (handle->h_sync && !(current->flags & PF_MEMALLOC)) |
1350 | err = jbd2_log_wait_commit(journal, tid); |
1351 | } else { |
1352 | spin_unlock(&transaction->t_handle_lock); |
1353 | spin_unlock(&journal->j_state_lock); |
1354 | } |
1355 | |
1356 | lock_map_release(&handle->h_lockdep_map); |
1357 | |
1358 | jbd2_free_handle(handle); |
1359 | return err; |
1360 | } |
1361 | |
1362 | /** |
1363 | * int jbd2_journal_force_commit() - force any uncommitted transactions |
1364 | * @journal: journal to force |
1365 | * |
1366 | * For synchronous operations: force any uncommitted transactions |
1367 | * to disk. May seem kludgy, but it reuses all the handle batching |
1368 | * code in a very simple manner. |
1369 | */ |
1370 | int jbd2_journal_force_commit(journal_t *journal) |
1371 | { |
1372 | handle_t *handle; |
1373 | int ret; |
1374 | |
1375 | handle = jbd2_journal_start(journal, 1); |
1376 | if (IS_ERR(handle)) { |
1377 | ret = PTR_ERR(handle); |
1378 | } else { |
1379 | handle->h_sync = 1; |
1380 | ret = jbd2_journal_stop(handle); |
1381 | } |
1382 | return ret; |
1383 | } |
1384 | |
1385 | /* |
1386 | * |
1387 | * List management code snippets: various functions for manipulating the |
1388 | * transaction buffer lists. |
1389 | * |
1390 | */ |
1391 | |
1392 | /* |
1393 | * Append a buffer to a transaction list, given the transaction's list head |
1394 | * pointer. |
1395 | * |
1396 | * j_list_lock is held. |
1397 | * |
1398 | * jbd_lock_bh_state(jh2bh(jh)) is held. |
1399 | */ |
1400 | |
1401 | static inline void |
1402 | __blist_add_buffer(struct journal_head **list, struct journal_head *jh) |
1403 | { |
1404 | if (!*list) { |
1405 | jh->b_tnext = jh->b_tprev = jh; |
1406 | *list = jh; |
1407 | } else { |
1408 | /* Insert at the tail of the list to preserve order */ |
1409 | struct journal_head *first = *list, *last = first->b_tprev; |
1410 | jh->b_tprev = last; |
1411 | jh->b_tnext = first; |
1412 | last->b_tnext = first->b_tprev = jh; |
1413 | } |
1414 | } |
1415 | |
1416 | /* |
1417 | * Remove a buffer from a transaction list, given the transaction's list |
1418 | * head pointer. |
1419 | * |
1420 | * Called with j_list_lock held, and the journal may not be locked. |
1421 | * |
1422 | * jbd_lock_bh_state(jh2bh(jh)) is held. |
1423 | */ |
1424 | |
1425 | static inline void |
1426 | __blist_del_buffer(struct journal_head **list, struct journal_head *jh) |
1427 | { |
1428 | if (*list == jh) { |
1429 | *list = jh->b_tnext; |
1430 | if (*list == jh) |
1431 | *list = NULL; |
1432 | } |
1433 | jh->b_tprev->b_tnext = jh->b_tnext; |
1434 | jh->b_tnext->b_tprev = jh->b_tprev; |
1435 | } |
1436 | |
1437 | /* |
1438 | * Remove a buffer from the appropriate transaction list. |
1439 | * |
1440 | * Note that this function can *change* the value of |
1441 | * bh->b_transaction->t_buffers, t_forget, t_iobuf_list, t_shadow_list, |
1442 | * t_log_list or t_reserved_list. If the caller is holding onto a copy of one |
1443 | * of these pointers, it could go bad. Generally the caller needs to re-read |
1444 | * the pointer from the transaction_t. |
1445 | * |
1446 | * Called under j_list_lock. The journal may not be locked. |
1447 | */ |
1448 | void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh) |
1449 | { |
1450 | struct journal_head **list = NULL; |
1451 | transaction_t *transaction; |
1452 | struct buffer_head *bh = jh2bh(jh); |
1453 | |
1454 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); |
1455 | transaction = jh->b_transaction; |
1456 | if (transaction) |
1457 | assert_spin_locked(&transaction->t_journal->j_list_lock); |
1458 | |
1459 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); |
1460 | if (jh->b_jlist != BJ_None) |
1461 | J_ASSERT_JH(jh, transaction != NULL); |
1462 | |
1463 | switch (jh->b_jlist) { |
1464 | case BJ_None: |
1465 | return; |
1466 | case BJ_Metadata: |
1467 | transaction->t_nr_buffers--; |
1468 | J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0); |
1469 | list = &transaction->t_buffers; |
1470 | break; |
1471 | case BJ_Forget: |
1472 | list = &transaction->t_forget; |
1473 | break; |
1474 | case BJ_IO: |
1475 | list = &transaction->t_iobuf_list; |
1476 | break; |
1477 | case BJ_Shadow: |
1478 | list = &transaction->t_shadow_list; |
1479 | break; |
1480 | case BJ_LogCtl: |
1481 | list = &transaction->t_log_list; |
1482 | break; |
1483 | case BJ_Reserved: |
1484 | list = &transaction->t_reserved_list; |
1485 | break; |
1486 | } |
1487 | |
1488 | __blist_del_buffer(list, jh); |
1489 | jh->b_jlist = BJ_None; |
1490 | if (test_clear_buffer_jbddirty(bh)) |
1491 | mark_buffer_dirty(bh); /* Expose it to the VM */ |
1492 | } |
1493 | |
1494 | void __jbd2_journal_unfile_buffer(struct journal_head *jh) |
1495 | { |
1496 | __jbd2_journal_temp_unlink_buffer(jh); |
1497 | jh->b_transaction = NULL; |
1498 | } |
1499 | |
1500 | void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh) |
1501 | { |
1502 | jbd_lock_bh_state(jh2bh(jh)); |
1503 | spin_lock(&journal->j_list_lock); |
1504 | __jbd2_journal_unfile_buffer(jh); |
1505 | spin_unlock(&journal->j_list_lock); |
1506 | jbd_unlock_bh_state(jh2bh(jh)); |
1507 | } |
1508 | |
1509 | /* |
1510 | * Called from jbd2_journal_try_to_free_buffers(). |
1511 | * |
1512 | * Called under jbd_lock_bh_state(bh) |
1513 | */ |
1514 | static void |
1515 | __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh) |
1516 | { |
1517 | struct journal_head *jh; |
1518 | |
1519 | jh = bh2jh(bh); |
1520 | |
1521 | if (buffer_locked(bh) || buffer_dirty(bh)) |
1522 | goto out; |
1523 | |
1524 | if (jh->b_next_transaction != NULL) |
1525 | goto out; |
1526 | |
1527 | spin_lock(&journal->j_list_lock); |
1528 | if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) { |
1529 | /* written-back checkpointed metadata buffer */ |
1530 | if (jh->b_jlist == BJ_None) { |
1531 | JBUFFER_TRACE(jh, "remove from checkpoint list"); |
1532 | __jbd2_journal_remove_checkpoint(jh); |
1533 | jbd2_journal_remove_journal_head(bh); |
1534 | __brelse(bh); |
1535 | } |
1536 | } |
1537 | spin_unlock(&journal->j_list_lock); |
1538 | out: |
1539 | return; |
1540 | } |
1541 | |
1542 | /** |
1543 | * int jbd2_journal_try_to_free_buffers() - try to free page buffers. |
1544 | * @journal: journal for operation |
1545 | * @page: to try and free |
1546 | * @gfp_mask: we use the mask to detect how hard should we try to release |
1547 | * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to |
1548 | * release the buffers. |
1549 | * |
1550 | * |
1551 | * For all the buffers on this page, |
1552 | * if they are fully written out ordered data, move them onto BUF_CLEAN |
1553 | * so try_to_free_buffers() can reap them. |
1554 | * |
1555 | * This function returns non-zero if we wish try_to_free_buffers() |
1556 | * to be called. We do this if the page is releasable by try_to_free_buffers(). |
1557 | * We also do it if the page has locked or dirty buffers and the caller wants |
1558 | * us to perform sync or async writeout. |
1559 | * |
1560 | * This complicates JBD locking somewhat. We aren't protected by the |
1561 | * BKL here. We wish to remove the buffer from its committing or |
1562 | * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer. |
1563 | * |
1564 | * This may *change* the value of transaction_t->t_datalist, so anyone |
1565 | * who looks at t_datalist needs to lock against this function. |
1566 | * |
1567 | * Even worse, someone may be doing a jbd2_journal_dirty_data on this |
1568 | * buffer. So we need to lock against that. jbd2_journal_dirty_data() |
1569 | * will come out of the lock with the buffer dirty, which makes it |
1570 | * ineligible for release here. |
1571 | * |
1572 | * Who else is affected by this? hmm... Really the only contender |
1573 | * is do_get_write_access() - it could be looking at the buffer while |
1574 | * journal_try_to_free_buffer() is changing its state. But that |
1575 | * cannot happen because we never reallocate freed data as metadata |
1576 | * while the data is part of a transaction. Yes? |
1577 | * |
1578 | * Return 0 on failure, 1 on success |
1579 | */ |
1580 | int jbd2_journal_try_to_free_buffers(journal_t *journal, |
1581 | struct page *page, gfp_t gfp_mask) |
1582 | { |
1583 | struct buffer_head *head; |
1584 | struct buffer_head *bh; |
1585 | int ret = 0; |
1586 | |
1587 | J_ASSERT(PageLocked(page)); |
1588 | |
1589 | head = page_buffers(page); |
1590 | bh = head; |
1591 | do { |
1592 | struct journal_head *jh; |
1593 | |
1594 | /* |
1595 | * We take our own ref against the journal_head here to avoid |
1596 | * having to add tons of locking around each instance of |
1597 | * jbd2_journal_remove_journal_head() and |
1598 | * jbd2_journal_put_journal_head(). |
1599 | */ |
1600 | jh = jbd2_journal_grab_journal_head(bh); |
1601 | if (!jh) |
1602 | continue; |
1603 | |
1604 | jbd_lock_bh_state(bh); |
1605 | __journal_try_to_free_buffer(journal, bh); |
1606 | jbd2_journal_put_journal_head(jh); |
1607 | jbd_unlock_bh_state(bh); |
1608 | if (buffer_jbd(bh)) |
1609 | goto busy; |
1610 | } while ((bh = bh->b_this_page) != head); |
1611 | |
1612 | ret = try_to_free_buffers(page); |
1613 | |
1614 | busy: |
1615 | return ret; |
1616 | } |
1617 | |
1618 | /* |
1619 | * This buffer is no longer needed. If it is on an older transaction's |
1620 | * checkpoint list we need to record it on this transaction's forget list |
1621 | * to pin this buffer (and hence its checkpointing transaction) down until |
1622 | * this transaction commits. If the buffer isn't on a checkpoint list, we |
1623 | * release it. |
1624 | * Returns non-zero if JBD no longer has an interest in the buffer. |
1625 | * |
1626 | * Called under j_list_lock. |
1627 | * |
1628 | * Called under jbd_lock_bh_state(bh). |
1629 | */ |
1630 | static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction) |
1631 | { |
1632 | int may_free = 1; |
1633 | struct buffer_head *bh = jh2bh(jh); |
1634 | |
1635 | __jbd2_journal_unfile_buffer(jh); |
1636 | |
1637 | if (jh->b_cp_transaction) { |
1638 | JBUFFER_TRACE(jh, "on running+cp transaction"); |
1639 | /* |
1640 | * We don't want to write the buffer anymore, clear the |
1641 | * bit so that we don't confuse checks in |
1642 | * __journal_file_buffer |
1643 | */ |
1644 | clear_buffer_dirty(bh); |
1645 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); |
1646 | may_free = 0; |
1647 | } else { |
1648 | JBUFFER_TRACE(jh, "on running transaction"); |
1649 | jbd2_journal_remove_journal_head(bh); |
1650 | __brelse(bh); |
1651 | } |
1652 | return may_free; |
1653 | } |
1654 | |
1655 | /* |
1656 | * jbd2_journal_invalidatepage |
1657 | * |
1658 | * This code is tricky. It has a number of cases to deal with. |
1659 | * |
1660 | * There are two invariants which this code relies on: |
1661 | * |
1662 | * i_size must be updated on disk before we start calling invalidatepage on the |
1663 | * data. |
1664 | * |
1665 | * This is done in ext3 by defining an ext3_setattr method which |
1666 | * updates i_size before truncate gets going. By maintaining this |
1667 | * invariant, we can be sure that it is safe to throw away any buffers |
1668 | * attached to the current transaction: once the transaction commits, |
1669 | * we know that the data will not be needed. |
1670 | * |
1671 | * Note however that we can *not* throw away data belonging to the |
1672 | * previous, committing transaction! |
1673 | * |
1674 | * Any disk blocks which *are* part of the previous, committing |
1675 | * transaction (and which therefore cannot be discarded immediately) are |
1676 | * not going to be reused in the new running transaction |
1677 | * |
1678 | * The bitmap committed_data images guarantee this: any block which is |
1679 | * allocated in one transaction and removed in the next will be marked |
1680 | * as in-use in the committed_data bitmap, so cannot be reused until |
1681 | * the next transaction to delete the block commits. This means that |
1682 | * leaving committing buffers dirty is quite safe: the disk blocks |
1683 | * cannot be reallocated to a different file and so buffer aliasing is |
1684 | * not possible. |
1685 | * |
1686 | * |
1687 | * The above applies mainly to ordered data mode. In writeback mode we |
1688 | * don't make guarantees about the order in which data hits disk --- in |
1689 | * particular we don't guarantee that new dirty data is flushed before |
1690 | * transaction commit --- so it is always safe just to discard data |
1691 | * immediately in that mode. --sct |
1692 | */ |
1693 | |
1694 | /* |
1695 | * The journal_unmap_buffer helper function returns zero if the buffer |
1696 | * concerned remains pinned as an anonymous buffer belonging to an older |
1697 | * transaction. |
1698 | * |
1699 | * We're outside-transaction here. Either or both of j_running_transaction |
1700 | * and j_committing_transaction may be NULL. |
1701 | */ |
1702 | static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh) |
1703 | { |
1704 | transaction_t *transaction; |
1705 | struct journal_head *jh; |
1706 | int may_free = 1; |
1707 | int ret; |
1708 | |
1709 | BUFFER_TRACE(bh, "entry"); |
1710 | |
1711 | /* |
1712 | * It is safe to proceed here without the j_list_lock because the |
1713 | * buffers cannot be stolen by try_to_free_buffers as long as we are |
1714 | * holding the page lock. --sct |
1715 | */ |
1716 | |
1717 | if (!buffer_jbd(bh)) |
1718 | goto zap_buffer_unlocked; |
1719 | |
1720 | /* OK, we have data buffer in journaled mode */ |
1721 | spin_lock(&journal->j_state_lock); |
1722 | jbd_lock_bh_state(bh); |
1723 | spin_lock(&journal->j_list_lock); |
1724 | |
1725 | jh = jbd2_journal_grab_journal_head(bh); |
1726 | if (!jh) |
1727 | goto zap_buffer_no_jh; |
1728 | |
1729 | transaction = jh->b_transaction; |
1730 | if (transaction == NULL) { |
1731 | /* First case: not on any transaction. If it |
1732 | * has no checkpoint link, then we can zap it: |
1733 | * it's a writeback-mode buffer so we don't care |
1734 | * if it hits disk safely. */ |
1735 | if (!jh->b_cp_transaction) { |
1736 | JBUFFER_TRACE(jh, "not on any transaction: zap"); |
1737 | goto zap_buffer; |
1738 | } |
1739 | |
1740 | if (!buffer_dirty(bh)) { |
1741 | /* bdflush has written it. We can drop it now */ |
1742 | goto zap_buffer; |
1743 | } |
1744 | |
1745 | /* OK, it must be in the journal but still not |
1746 | * written fully to disk: it's metadata or |
1747 | * journaled data... */ |
1748 | |
1749 | if (journal->j_running_transaction) { |
1750 | /* ... and once the current transaction has |
1751 | * committed, the buffer won't be needed any |
1752 | * longer. */ |
1753 | JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget"); |
1754 | ret = __dispose_buffer(jh, |
1755 | journal->j_running_transaction); |
1756 | jbd2_journal_put_journal_head(jh); |
1757 | spin_unlock(&journal->j_list_lock); |
1758 | jbd_unlock_bh_state(bh); |
1759 | spin_unlock(&journal->j_state_lock); |
1760 | return ret; |
1761 | } else { |
1762 | /* There is no currently-running transaction. So the |
1763 | * orphan record which we wrote for this file must have |
1764 | * passed into commit. We must attach this buffer to |
1765 | * the committing transaction, if it exists. */ |
1766 | if (journal->j_committing_transaction) { |
1767 | JBUFFER_TRACE(jh, "give to committing trans"); |
1768 | ret = __dispose_buffer(jh, |
1769 | journal->j_committing_transaction); |
1770 | jbd2_journal_put_journal_head(jh); |
1771 | spin_unlock(&journal->j_list_lock); |
1772 | jbd_unlock_bh_state(bh); |
1773 | spin_unlock(&journal->j_state_lock); |
1774 | return ret; |
1775 | } else { |
1776 | /* The orphan record's transaction has |
1777 | * committed. We can cleanse this buffer */ |
1778 | clear_buffer_jbddirty(bh); |
1779 | goto zap_buffer; |
1780 | } |
1781 | } |
1782 | } else if (transaction == journal->j_committing_transaction) { |
1783 | JBUFFER_TRACE(jh, "on committing transaction"); |
1784 | /* |
1785 | * If it is committing, we simply cannot touch it. We |
1786 | * can remove it's next_transaction pointer from the |
1787 | * running transaction if that is set, but nothing |
1788 | * else. */ |
1789 | set_buffer_freed(bh); |
1790 | if (jh->b_next_transaction) { |
1791 | J_ASSERT(jh->b_next_transaction == |
1792 | journal->j_running_transaction); |
1793 | jh->b_next_transaction = NULL; |
1794 | } |
1795 | jbd2_journal_put_journal_head(jh); |
1796 | spin_unlock(&journal->j_list_lock); |
1797 | jbd_unlock_bh_state(bh); |
1798 | spin_unlock(&journal->j_state_lock); |
1799 | return 0; |
1800 | } else { |
1801 | /* Good, the buffer belongs to the running transaction. |
1802 | * We are writing our own transaction's data, not any |
1803 | * previous one's, so it is safe to throw it away |
1804 | * (remember that we expect the filesystem to have set |
1805 | * i_size already for this truncate so recovery will not |
1806 | * expose the disk blocks we are discarding here.) */ |
1807 | J_ASSERT_JH(jh, transaction == journal->j_running_transaction); |
1808 | JBUFFER_TRACE(jh, "on running transaction"); |
1809 | may_free = __dispose_buffer(jh, transaction); |
1810 | } |
1811 | |
1812 | zap_buffer: |
1813 | jbd2_journal_put_journal_head(jh); |
1814 | zap_buffer_no_jh: |
1815 | spin_unlock(&journal->j_list_lock); |
1816 | jbd_unlock_bh_state(bh); |
1817 | spin_unlock(&journal->j_state_lock); |
1818 | zap_buffer_unlocked: |
1819 | clear_buffer_dirty(bh); |
1820 | J_ASSERT_BH(bh, !buffer_jbddirty(bh)); |
1821 | clear_buffer_mapped(bh); |
1822 | clear_buffer_req(bh); |
1823 | clear_buffer_new(bh); |
1824 | bh->b_bdev = NULL; |
1825 | return may_free; |
1826 | } |
1827 | |
1828 | /** |
1829 | * void jbd2_journal_invalidatepage() |
1830 | * @journal: journal to use for flush... |
1831 | * @page: page to flush |
1832 | * @offset: length of page to invalidate. |
1833 | * |
1834 | * Reap page buffers containing data after offset in page. |
1835 | * |
1836 | */ |
1837 | void jbd2_journal_invalidatepage(journal_t *journal, |
1838 | struct page *page, |
1839 | unsigned long offset) |
1840 | { |
1841 | struct buffer_head *head, *bh, *next; |
1842 | unsigned int curr_off = 0; |
1843 | int may_free = 1; |
1844 | |
1845 | if (!PageLocked(page)) |
1846 | BUG(); |
1847 | if (!page_has_buffers(page)) |
1848 | return; |
1849 | |
1850 | /* We will potentially be playing with lists other than just the |
1851 | * data lists (especially for journaled data mode), so be |
1852 | * cautious in our locking. */ |
1853 | |
1854 | head = bh = page_buffers(page); |
1855 | do { |
1856 | unsigned int next_off = curr_off + bh->b_size; |
1857 | next = bh->b_this_page; |
1858 | |
1859 | if (offset <= curr_off) { |
1860 | /* This block is wholly outside the truncation point */ |
1861 | lock_buffer(bh); |
1862 | may_free &= journal_unmap_buffer(journal, bh); |
1863 | unlock_buffer(bh); |
1864 | } |
1865 | curr_off = next_off; |
1866 | bh = next; |
1867 | |
1868 | } while (bh != head); |
1869 | |
1870 | if (!offset) { |
1871 | if (may_free && try_to_free_buffers(page)) |
1872 | J_ASSERT(!page_has_buffers(page)); |
1873 | } |
1874 | } |
1875 | |
1876 | /* |
1877 | * File a buffer on the given transaction list. |
1878 | */ |
1879 | void __jbd2_journal_file_buffer(struct journal_head *jh, |
1880 | transaction_t *transaction, int jlist) |
1881 | { |
1882 | struct journal_head **list = NULL; |
1883 | int was_dirty = 0; |
1884 | struct buffer_head *bh = jh2bh(jh); |
1885 | |
1886 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); |
1887 | assert_spin_locked(&transaction->t_journal->j_list_lock); |
1888 | |
1889 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); |
1890 | J_ASSERT_JH(jh, jh->b_transaction == transaction || |
1891 | jh->b_transaction == NULL); |
1892 | |
1893 | if (jh->b_transaction && jh->b_jlist == jlist) |
1894 | return; |
1895 | |
1896 | if (jlist == BJ_Metadata || jlist == BJ_Reserved || |
1897 | jlist == BJ_Shadow || jlist == BJ_Forget) { |
1898 | /* |
1899 | * For metadata buffers, we track dirty bit in buffer_jbddirty |
1900 | * instead of buffer_dirty. We should not see a dirty bit set |
1901 | * here because we clear it in do_get_write_access but e.g. |
1902 | * tune2fs can modify the sb and set the dirty bit at any time |
1903 | * so we try to gracefully handle that. |
1904 | */ |
1905 | if (buffer_dirty(bh)) |
1906 | warn_dirty_buffer(bh); |
1907 | if (test_clear_buffer_dirty(bh) || |
1908 | test_clear_buffer_jbddirty(bh)) |
1909 | was_dirty = 1; |
1910 | } |
1911 | |
1912 | if (jh->b_transaction) |
1913 | __jbd2_journal_temp_unlink_buffer(jh); |
1914 | jh->b_transaction = transaction; |
1915 | |
1916 | switch (jlist) { |
1917 | case BJ_None: |
1918 | J_ASSERT_JH(jh, !jh->b_committed_data); |
1919 | J_ASSERT_JH(jh, !jh->b_frozen_data); |
1920 | return; |
1921 | case BJ_Metadata: |
1922 | transaction->t_nr_buffers++; |
1923 | list = &transaction->t_buffers; |
1924 | break; |
1925 | case BJ_Forget: |
1926 | list = &transaction->t_forget; |
1927 | break; |
1928 | case BJ_IO: |
1929 | list = &transaction->t_iobuf_list; |
1930 | break; |
1931 | case BJ_Shadow: |
1932 | list = &transaction->t_shadow_list; |
1933 | break; |
1934 | case BJ_LogCtl: |
1935 | list = &transaction->t_log_list; |
1936 | break; |
1937 | case BJ_Reserved: |
1938 | list = &transaction->t_reserved_list; |
1939 | break; |
1940 | } |
1941 | |
1942 | __blist_add_buffer(list, jh); |
1943 | jh->b_jlist = jlist; |
1944 | |
1945 | if (was_dirty) |
1946 | set_buffer_jbddirty(bh); |
1947 | } |
1948 | |
1949 | void jbd2_journal_file_buffer(struct journal_head *jh, |
1950 | transaction_t *transaction, int jlist) |
1951 | { |
1952 | jbd_lock_bh_state(jh2bh(jh)); |
1953 | spin_lock(&transaction->t_journal->j_list_lock); |
1954 | __jbd2_journal_file_buffer(jh, transaction, jlist); |
1955 | spin_unlock(&transaction->t_journal->j_list_lock); |
1956 | jbd_unlock_bh_state(jh2bh(jh)); |
1957 | } |
1958 | |
1959 | /* |
1960 | * Remove a buffer from its current buffer list in preparation for |
1961 | * dropping it from its current transaction entirely. If the buffer has |
1962 | * already started to be used by a subsequent transaction, refile the |
1963 | * buffer on that transaction's metadata list. |
1964 | * |
1965 | * Called under journal->j_list_lock |
1966 | * |
1967 | * Called under jbd_lock_bh_state(jh2bh(jh)) |
1968 | */ |
1969 | void __jbd2_journal_refile_buffer(struct journal_head *jh) |
1970 | { |
1971 | int was_dirty; |
1972 | struct buffer_head *bh = jh2bh(jh); |
1973 | |
1974 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); |
1975 | if (jh->b_transaction) |
1976 | assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock); |
1977 | |
1978 | /* If the buffer is now unused, just drop it. */ |
1979 | if (jh->b_next_transaction == NULL) { |
1980 | __jbd2_journal_unfile_buffer(jh); |
1981 | return; |
1982 | } |
1983 | |
1984 | /* |
1985 | * It has been modified by a later transaction: add it to the new |
1986 | * transaction's metadata list. |
1987 | */ |
1988 | |
1989 | was_dirty = test_clear_buffer_jbddirty(bh); |
1990 | __jbd2_journal_temp_unlink_buffer(jh); |
1991 | jh->b_transaction = jh->b_next_transaction; |
1992 | jh->b_next_transaction = NULL; |
1993 | __jbd2_journal_file_buffer(jh, jh->b_transaction, |
1994 | jh->b_modified ? BJ_Metadata : BJ_Reserved); |
1995 | J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING); |
1996 | |
1997 | if (was_dirty) |
1998 | set_buffer_jbddirty(bh); |
1999 | } |
2000 | |
2001 | /* |
2002 | * For the unlocked version of this call, also make sure that any |
2003 | * hanging journal_head is cleaned up if necessary. |
2004 | * |
2005 | * __jbd2_journal_refile_buffer is usually called as part of a single locked |
2006 | * operation on a buffer_head, in which the caller is probably going to |
2007 | * be hooking the journal_head onto other lists. In that case it is up |
2008 | * to the caller to remove the journal_head if necessary. For the |
2009 | * unlocked jbd2_journal_refile_buffer call, the caller isn't going to be |
2010 | * doing anything else to the buffer so we need to do the cleanup |
2011 | * ourselves to avoid a jh leak. |
2012 | * |
2013 | * *** The journal_head may be freed by this call! *** |
2014 | */ |
2015 | void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh) |
2016 | { |
2017 | struct buffer_head *bh = jh2bh(jh); |
2018 | |
2019 | jbd_lock_bh_state(bh); |
2020 | spin_lock(&journal->j_list_lock); |
2021 | |
2022 | __jbd2_journal_refile_buffer(jh); |
2023 | jbd_unlock_bh_state(bh); |
2024 | jbd2_journal_remove_journal_head(bh); |
2025 | |
2026 | spin_unlock(&journal->j_list_lock); |
2027 | __brelse(bh); |
2028 | } |
2029 | |
2030 | /* |
2031 | * File inode in the inode list of the handle's transaction |
2032 | */ |
2033 | int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode) |
2034 | { |
2035 | transaction_t *transaction = handle->h_transaction; |
2036 | journal_t *journal = transaction->t_journal; |
2037 | |
2038 | if (is_handle_aborted(handle)) |
2039 | return -EIO; |
2040 | |
2041 | jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino, |
2042 | transaction->t_tid); |
2043 | |
2044 | /* |
2045 | * First check whether inode isn't already on the transaction's |
2046 | * lists without taking the lock. Note that this check is safe |
2047 | * without the lock as we cannot race with somebody removing inode |
2048 | * from the transaction. The reason is that we remove inode from the |
2049 | * transaction only in journal_release_jbd_inode() and when we commit |
2050 | * the transaction. We are guarded from the first case by holding |
2051 | * a reference to the inode. We are safe against the second case |
2052 | * because if jinode->i_transaction == transaction, commit code |
2053 | * cannot touch the transaction because we hold reference to it, |
2054 | * and if jinode->i_next_transaction == transaction, commit code |
2055 | * will only file the inode where we want it. |
2056 | */ |
2057 | if (jinode->i_transaction == transaction || |
2058 | jinode->i_next_transaction == transaction) |
2059 | return 0; |
2060 | |
2061 | spin_lock(&journal->j_list_lock); |
2062 | |
2063 | if (jinode->i_transaction == transaction || |
2064 | jinode->i_next_transaction == transaction) |
2065 | goto done; |
2066 | |
2067 | /* On some different transaction's list - should be |
2068 | * the committing one */ |
2069 | if (jinode->i_transaction) { |
2070 | J_ASSERT(jinode->i_next_transaction == NULL); |
2071 | J_ASSERT(jinode->i_transaction == |
2072 | journal->j_committing_transaction); |
2073 | jinode->i_next_transaction = transaction; |
2074 | goto done; |
2075 | } |
2076 | /* Not on any transaction list... */ |
2077 | J_ASSERT(!jinode->i_next_transaction); |
2078 | jinode->i_transaction = transaction; |
2079 | list_add(&jinode->i_list, &transaction->t_inode_list); |
2080 | done: |
2081 | spin_unlock(&journal->j_list_lock); |
2082 | |
2083 | return 0; |
2084 | } |
2085 | |
2086 | /* |
2087 | * File truncate and transaction commit interact with each other in a |
2088 | * non-trivial way. If a transaction writing data block A is |
2089 | * committing, we cannot discard the data by truncate until we have |
2090 | * written them. Otherwise if we crashed after the transaction with |
2091 | * write has committed but before the transaction with truncate has |
2092 | * committed, we could see stale data in block A. This function is a |
2093 | * helper to solve this problem. It starts writeout of the truncated |
2094 | * part in case it is in the committing transaction. |
2095 | * |
2096 | * Filesystem code must call this function when inode is journaled in |
2097 | * ordered mode before truncation happens and after the inode has been |
2098 | * placed on orphan list with the new inode size. The second condition |
2099 | * avoids the race that someone writes new data and we start |
2100 | * committing the transaction after this function has been called but |
2101 | * before a transaction for truncate is started (and furthermore it |
2102 | * allows us to optimize the case where the addition to orphan list |
2103 | * happens in the same transaction as write --- we don't have to write |
2104 | * any data in such case). |
2105 | */ |
2106 | int jbd2_journal_begin_ordered_truncate(journal_t *journal, |
2107 | struct jbd2_inode *jinode, |
2108 | loff_t new_size) |
2109 | { |
2110 | transaction_t *inode_trans, *commit_trans; |
2111 | int ret = 0; |
2112 | |
2113 | /* This is a quick check to avoid locking if not necessary */ |
2114 | if (!jinode->i_transaction) |
2115 | goto out; |
2116 | /* Locks are here just to force reading of recent values, it is |
2117 | * enough that the transaction was not committing before we started |
2118 | * a transaction adding the inode to orphan list */ |
2119 | spin_lock(&journal->j_state_lock); |
2120 | commit_trans = journal->j_committing_transaction; |
2121 | spin_unlock(&journal->j_state_lock); |
2122 | spin_lock(&journal->j_list_lock); |
2123 | inode_trans = jinode->i_transaction; |
2124 | spin_unlock(&journal->j_list_lock); |
2125 | if (inode_trans == commit_trans) { |
2126 | ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping, |
2127 | new_size, LLONG_MAX); |
2128 | if (ret) |
2129 | jbd2_journal_abort(journal, ret); |
2130 | } |
2131 | out: |
2132 | return ret; |
2133 | } |
2134 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9