]>
Commit | Line | Data |
---|---|---|
470decc6 | 1 | /* |
58862699 | 2 | * linux/fs/jbd2/transaction.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 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> | |
f7f4bccb | 22 | #include <linux/jbd2.h> |
470decc6 DK |
23 | #include <linux/errno.h> |
24 | #include <linux/slab.h> | |
25 | #include <linux/timer.h> | |
470decc6 DK |
26 | #include <linux/mm.h> |
27 | #include <linux/highmem.h> | |
e07f7183 | 28 | #include <linux/hrtimer.h> |
47def826 TT |
29 | #include <linux/backing-dev.h> |
30 | #include <linux/module.h> | |
470decc6 | 31 | |
7ddae860 AB |
32 | static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh); |
33 | ||
470decc6 | 34 | /* |
f7f4bccb | 35 | * jbd2_get_transaction: obtain a new transaction_t object. |
470decc6 DK |
36 | * |
37 | * Simply allocate and initialise a new transaction. Create it in | |
38 | * RUNNING state and add it to the current journal (which should not | |
39 | * have an existing running transaction: we only make a new transaction | |
40 | * once we have started to commit the old one). | |
41 | * | |
42 | * Preconditions: | |
43 | * The journal MUST be locked. We don't perform atomic mallocs on the | |
44 | * new transaction and we can't block without protecting against other | |
45 | * processes trying to touch the journal while it is in transition. | |
46 | * | |
470decc6 DK |
47 | */ |
48 | ||
49 | static transaction_t * | |
f7f4bccb | 50 | jbd2_get_transaction(journal_t *journal, transaction_t *transaction) |
470decc6 DK |
51 | { |
52 | transaction->t_journal = journal; | |
53 | transaction->t_state = T_RUNNING; | |
e07f7183 | 54 | transaction->t_start_time = ktime_get(); |
470decc6 DK |
55 | transaction->t_tid = journal->j_transaction_sequence++; |
56 | transaction->t_expires = jiffies + journal->j_commit_interval; | |
57 | spin_lock_init(&transaction->t_handle_lock); | |
c851ed54 | 58 | INIT_LIST_HEAD(&transaction->t_inode_list); |
3e624fc7 | 59 | INIT_LIST_HEAD(&transaction->t_private_list); |
470decc6 DK |
60 | |
61 | /* Set up the commit timer for the new transaction. */ | |
b1f485f2 | 62 | journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires); |
470decc6 DK |
63 | add_timer(&journal->j_commit_timer); |
64 | ||
65 | J_ASSERT(journal->j_running_transaction == NULL); | |
66 | journal->j_running_transaction = transaction; | |
8e85fb3f JL |
67 | transaction->t_max_wait = 0; |
68 | transaction->t_start = jiffies; | |
470decc6 DK |
69 | |
70 | return transaction; | |
71 | } | |
72 | ||
73 | /* | |
74 | * Handle management. | |
75 | * | |
76 | * A handle_t is an object which represents a single atomic update to a | |
77 | * filesystem, and which tracks all of the modifications which form part | |
78 | * of that one update. | |
79 | */ | |
80 | ||
81 | /* | |
82 | * start_this_handle: Given a handle, deal with any locking or stalling | |
83 | * needed to make sure that there is enough journal space for the handle | |
84 | * to begin. Attach the handle to a transaction and set up the | |
85 | * transaction's buffer credits. | |
86 | */ | |
87 | ||
47def826 TT |
88 | static int start_this_handle(journal_t *journal, handle_t *handle, |
89 | int gfp_mask) | |
470decc6 DK |
90 | { |
91 | transaction_t *transaction; | |
92 | int needed; | |
93 | int nblocks = handle->h_buffer_credits; | |
94 | transaction_t *new_transaction = NULL; | |
8e85fb3f | 95 | unsigned long ts = jiffies; |
470decc6 DK |
96 | |
97 | if (nblocks > journal->j_max_transaction_buffers) { | |
98 | printk(KERN_ERR "JBD: %s wants too many credits (%d > %d)\n", | |
99 | current->comm, nblocks, | |
100 | journal->j_max_transaction_buffers); | |
47def826 | 101 | return -ENOSPC; |
470decc6 DK |
102 | } |
103 | ||
104 | alloc_transaction: | |
105 | if (!journal->j_running_transaction) { | |
47def826 | 106 | new_transaction = kzalloc(sizeof(*new_transaction), gfp_mask); |
470decc6 | 107 | if (!new_transaction) { |
47def826 TT |
108 | /* |
109 | * If __GFP_FS is not present, then we may be | |
110 | * being called from inside the fs writeback | |
111 | * layer, so we MUST NOT fail. Since | |
112 | * __GFP_NOFAIL is going away, we will arrange | |
113 | * to retry the allocation ourselves. | |
114 | */ | |
115 | if ((gfp_mask & __GFP_FS) == 0) { | |
116 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
117 | goto alloc_transaction; | |
118 | } | |
119 | return -ENOMEM; | |
470decc6 | 120 | } |
470decc6 DK |
121 | } |
122 | ||
123 | jbd_debug(3, "New handle %p going live.\n", handle); | |
124 | ||
125 | repeat: | |
126 | ||
127 | /* | |
128 | * We need to hold j_state_lock until t_updates has been incremented, | |
129 | * for proper journal barrier handling | |
130 | */ | |
131 | spin_lock(&journal->j_state_lock); | |
132 | repeat_locked: | |
133 | if (is_journal_aborted(journal) || | |
f7f4bccb | 134 | (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) { |
470decc6 | 135 | spin_unlock(&journal->j_state_lock); |
47def826 TT |
136 | kfree(new_transaction); |
137 | return -EROFS; | |
470decc6 DK |
138 | } |
139 | ||
140 | /* Wait on the journal's transaction barrier if necessary */ | |
141 | if (journal->j_barrier_count) { | |
142 | spin_unlock(&journal->j_state_lock); | |
143 | wait_event(journal->j_wait_transaction_locked, | |
144 | journal->j_barrier_count == 0); | |
145 | goto repeat; | |
146 | } | |
147 | ||
148 | if (!journal->j_running_transaction) { | |
149 | if (!new_transaction) { | |
150 | spin_unlock(&journal->j_state_lock); | |
151 | goto alloc_transaction; | |
152 | } | |
f7f4bccb | 153 | jbd2_get_transaction(journal, new_transaction); |
470decc6 DK |
154 | new_transaction = NULL; |
155 | } | |
156 | ||
157 | transaction = journal->j_running_transaction; | |
158 | ||
159 | /* | |
160 | * If the current transaction is locked down for commit, wait for the | |
161 | * lock to be released. | |
162 | */ | |
163 | if (transaction->t_state == T_LOCKED) { | |
164 | DEFINE_WAIT(wait); | |
165 | ||
166 | prepare_to_wait(&journal->j_wait_transaction_locked, | |
167 | &wait, TASK_UNINTERRUPTIBLE); | |
168 | spin_unlock(&journal->j_state_lock); | |
169 | schedule(); | |
170 | finish_wait(&journal->j_wait_transaction_locked, &wait); | |
171 | goto repeat; | |
172 | } | |
173 | ||
174 | /* | |
175 | * If there is not enough space left in the log to write all potential | |
176 | * buffers requested by this operation, we need to stall pending a log | |
177 | * checkpoint to free some more log space. | |
178 | */ | |
179 | spin_lock(&transaction->t_handle_lock); | |
180 | needed = transaction->t_outstanding_credits + nblocks; | |
181 | ||
182 | if (needed > journal->j_max_transaction_buffers) { | |
183 | /* | |
184 | * If the current transaction is already too large, then start | |
185 | * to commit it: we can then go back and attach this handle to | |
186 | * a new transaction. | |
187 | */ | |
188 | DEFINE_WAIT(wait); | |
189 | ||
190 | jbd_debug(2, "Handle %p starting new commit...\n", handle); | |
191 | spin_unlock(&transaction->t_handle_lock); | |
192 | prepare_to_wait(&journal->j_wait_transaction_locked, &wait, | |
193 | TASK_UNINTERRUPTIBLE); | |
f7f4bccb | 194 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
195 | spin_unlock(&journal->j_state_lock); |
196 | schedule(); | |
197 | finish_wait(&journal->j_wait_transaction_locked, &wait); | |
198 | goto repeat; | |
199 | } | |
200 | ||
201 | /* | |
202 | * The commit code assumes that it can get enough log space | |
203 | * without forcing a checkpoint. This is *critical* for | |
204 | * correctness: a checkpoint of a buffer which is also | |
205 | * associated with a committing transaction creates a deadlock, | |
206 | * so commit simply cannot force through checkpoints. | |
207 | * | |
208 | * We must therefore ensure the necessary space in the journal | |
209 | * *before* starting to dirty potentially checkpointed buffers | |
210 | * in the new transaction. | |
211 | * | |
212 | * The worst part is, any transaction currently committing can | |
213 | * reduce the free space arbitrarily. Be careful to account for | |
214 | * those buffers when checkpointing. | |
215 | */ | |
216 | ||
217 | /* | |
218 | * @@@ AKPM: This seems rather over-defensive. We're giving commit | |
219 | * a _lot_ of headroom: 1/4 of the journal plus the size of | |
220 | * the committing transaction. Really, we only need to give it | |
221 | * committing_transaction->t_outstanding_credits plus "enough" for | |
222 | * the log control blocks. | |
223 | * Also, this test is inconsitent with the matching one in | |
f7f4bccb | 224 | * jbd2_journal_extend(). |
470decc6 | 225 | */ |
f7f4bccb | 226 | if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) { |
470decc6 DK |
227 | jbd_debug(2, "Handle %p waiting for checkpoint...\n", handle); |
228 | spin_unlock(&transaction->t_handle_lock); | |
f7f4bccb | 229 | __jbd2_log_wait_for_space(journal); |
470decc6 DK |
230 | goto repeat_locked; |
231 | } | |
232 | ||
233 | /* OK, account for the buffers that this operation expects to | |
234 | * use and add the handle to the running transaction. */ | |
235 | ||
8e85fb3f JL |
236 | if (time_after(transaction->t_start, ts)) { |
237 | ts = jbd2_time_diff(ts, transaction->t_start); | |
238 | if (ts > transaction->t_max_wait) | |
239 | transaction->t_max_wait = ts; | |
240 | } | |
241 | ||
470decc6 DK |
242 | handle->h_transaction = transaction; |
243 | transaction->t_outstanding_credits += nblocks; | |
244 | transaction->t_updates++; | |
245 | transaction->t_handle_count++; | |
246 | jbd_debug(4, "Handle %p given %d credits (total %d, free %d)\n", | |
247 | handle, nblocks, transaction->t_outstanding_credits, | |
f7f4bccb | 248 | __jbd2_log_space_left(journal)); |
470decc6 DK |
249 | spin_unlock(&transaction->t_handle_lock); |
250 | spin_unlock(&journal->j_state_lock); | |
9599b0e5 JK |
251 | |
252 | lock_map_acquire(&handle->h_lockdep_map); | |
47def826 TT |
253 | kfree(new_transaction); |
254 | return 0; | |
470decc6 DK |
255 | } |
256 | ||
7b751066 MC |
257 | static struct lock_class_key jbd2_handle_key; |
258 | ||
470decc6 DK |
259 | /* Allocate a new handle. This should probably be in a slab... */ |
260 | static handle_t *new_handle(int nblocks) | |
261 | { | |
af1e76d6 | 262 | handle_t *handle = jbd2_alloc_handle(GFP_NOFS); |
470decc6 DK |
263 | if (!handle) |
264 | return NULL; | |
265 | memset(handle, 0, sizeof(*handle)); | |
266 | handle->h_buffer_credits = nblocks; | |
267 | handle->h_ref = 1; | |
268 | ||
7b751066 MC |
269 | lockdep_init_map(&handle->h_lockdep_map, "jbd2_handle", |
270 | &jbd2_handle_key, 0); | |
271 | ||
470decc6 DK |
272 | return handle; |
273 | } | |
274 | ||
275 | /** | |
f7f4bccb | 276 | * handle_t *jbd2_journal_start() - Obtain a new handle. |
470decc6 DK |
277 | * @journal: Journal to start transaction on. |
278 | * @nblocks: number of block buffer we might modify | |
279 | * | |
280 | * We make sure that the transaction can guarantee at least nblocks of | |
281 | * modified buffers in the log. We block until the log can guarantee | |
282 | * that much space. | |
283 | * | |
284 | * This function is visible to journal users (like ext3fs), so is not | |
285 | * called with the journal already locked. | |
286 | * | |
287 | * Return a pointer to a newly allocated handle, or NULL on failure | |
288 | */ | |
47def826 | 289 | handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int gfp_mask) |
470decc6 DK |
290 | { |
291 | handle_t *handle = journal_current_handle(); | |
292 | int err; | |
293 | ||
294 | if (!journal) | |
295 | return ERR_PTR(-EROFS); | |
296 | ||
297 | if (handle) { | |
298 | J_ASSERT(handle->h_transaction->t_journal == journal); | |
299 | handle->h_ref++; | |
300 | return handle; | |
301 | } | |
302 | ||
303 | handle = new_handle(nblocks); | |
304 | if (!handle) | |
305 | return ERR_PTR(-ENOMEM); | |
306 | ||
307 | current->journal_info = handle; | |
308 | ||
47def826 | 309 | err = start_this_handle(journal, handle, gfp_mask); |
470decc6 | 310 | if (err < 0) { |
af1e76d6 | 311 | jbd2_free_handle(handle); |
470decc6 DK |
312 | current->journal_info = NULL; |
313 | handle = ERR_PTR(err); | |
7b751066 | 314 | goto out; |
470decc6 | 315 | } |
7b751066 | 316 | out: |
470decc6 DK |
317 | return handle; |
318 | } | |
47def826 TT |
319 | EXPORT_SYMBOL(jbd2__journal_start); |
320 | ||
321 | ||
322 | handle_t *jbd2_journal_start(journal_t *journal, int nblocks) | |
323 | { | |
324 | return jbd2__journal_start(journal, nblocks, GFP_NOFS); | |
325 | } | |
326 | EXPORT_SYMBOL(jbd2_journal_start); | |
327 | ||
470decc6 DK |
328 | |
329 | /** | |
f7f4bccb | 330 | * int jbd2_journal_extend() - extend buffer credits. |
470decc6 DK |
331 | * @handle: handle to 'extend' |
332 | * @nblocks: nr blocks to try to extend by. | |
333 | * | |
334 | * Some transactions, such as large extends and truncates, can be done | |
335 | * atomically all at once or in several stages. The operation requests | |
336 | * a credit for a number of buffer modications in advance, but can | |
337 | * extend its credit if it needs more. | |
338 | * | |
f7f4bccb | 339 | * jbd2_journal_extend tries to give the running handle more buffer credits. |
470decc6 DK |
340 | * It does not guarantee that allocation - this is a best-effort only. |
341 | * The calling process MUST be able to deal cleanly with a failure to | |
342 | * extend here. | |
343 | * | |
344 | * Return 0 on success, non-zero on failure. | |
345 | * | |
346 | * return code < 0 implies an error | |
347 | * return code > 0 implies normal transaction-full status. | |
348 | */ | |
f7f4bccb | 349 | int jbd2_journal_extend(handle_t *handle, int nblocks) |
470decc6 DK |
350 | { |
351 | transaction_t *transaction = handle->h_transaction; | |
352 | journal_t *journal = transaction->t_journal; | |
353 | int result; | |
354 | int wanted; | |
355 | ||
356 | result = -EIO; | |
357 | if (is_handle_aborted(handle)) | |
358 | goto out; | |
359 | ||
360 | result = 1; | |
361 | ||
362 | spin_lock(&journal->j_state_lock); | |
363 | ||
364 | /* Don't extend a locked-down transaction! */ | |
365 | if (handle->h_transaction->t_state != T_RUNNING) { | |
366 | jbd_debug(3, "denied handle %p %d blocks: " | |
367 | "transaction not running\n", handle, nblocks); | |
368 | goto error_out; | |
369 | } | |
370 | ||
371 | spin_lock(&transaction->t_handle_lock); | |
372 | wanted = transaction->t_outstanding_credits + nblocks; | |
373 | ||
374 | if (wanted > journal->j_max_transaction_buffers) { | |
375 | jbd_debug(3, "denied handle %p %d blocks: " | |
376 | "transaction too large\n", handle, nblocks); | |
377 | goto unlock; | |
378 | } | |
379 | ||
f7f4bccb | 380 | if (wanted > __jbd2_log_space_left(journal)) { |
470decc6 DK |
381 | jbd_debug(3, "denied handle %p %d blocks: " |
382 | "insufficient log space\n", handle, nblocks); | |
383 | goto unlock; | |
384 | } | |
385 | ||
386 | handle->h_buffer_credits += nblocks; | |
387 | transaction->t_outstanding_credits += nblocks; | |
388 | result = 0; | |
389 | ||
390 | jbd_debug(3, "extended handle %p by %d\n", handle, nblocks); | |
391 | unlock: | |
392 | spin_unlock(&transaction->t_handle_lock); | |
393 | error_out: | |
394 | spin_unlock(&journal->j_state_lock); | |
395 | out: | |
396 | return result; | |
397 | } | |
398 | ||
399 | ||
400 | /** | |
f7f4bccb | 401 | * int jbd2_journal_restart() - restart a handle . |
470decc6 DK |
402 | * @handle: handle to restart |
403 | * @nblocks: nr credits requested | |
404 | * | |
405 | * Restart a handle for a multi-transaction filesystem | |
406 | * operation. | |
407 | * | |
f7f4bccb MC |
408 | * If the jbd2_journal_extend() call above fails to grant new buffer credits |
409 | * to a running handle, a call to jbd2_journal_restart will commit the | |
470decc6 DK |
410 | * handle's transaction so far and reattach the handle to a new |
411 | * transaction capabable of guaranteeing the requested number of | |
412 | * credits. | |
413 | */ | |
47def826 | 414 | int jbd2__journal_restart(handle_t *handle, int nblocks, int gfp_mask) |
470decc6 DK |
415 | { |
416 | transaction_t *transaction = handle->h_transaction; | |
417 | journal_t *journal = transaction->t_journal; | |
418 | int ret; | |
419 | ||
420 | /* If we've had an abort of any type, don't even think about | |
421 | * actually doing the restart! */ | |
422 | if (is_handle_aborted(handle)) | |
423 | return 0; | |
424 | ||
425 | /* | |
426 | * First unlink the handle from its current transaction, and start the | |
427 | * commit on that. | |
428 | */ | |
429 | J_ASSERT(transaction->t_updates > 0); | |
430 | J_ASSERT(journal_current_handle() == handle); | |
431 | ||
432 | spin_lock(&journal->j_state_lock); | |
433 | spin_lock(&transaction->t_handle_lock); | |
434 | transaction->t_outstanding_credits -= handle->h_buffer_credits; | |
435 | transaction->t_updates--; | |
436 | ||
437 | if (!transaction->t_updates) | |
438 | wake_up(&journal->j_wait_updates); | |
439 | spin_unlock(&transaction->t_handle_lock); | |
440 | ||
441 | jbd_debug(2, "restarting handle %p\n", handle); | |
f7f4bccb | 442 | __jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
443 | spin_unlock(&journal->j_state_lock); |
444 | ||
9599b0e5 | 445 | lock_map_release(&handle->h_lockdep_map); |
470decc6 | 446 | handle->h_buffer_credits = nblocks; |
47def826 | 447 | ret = start_this_handle(journal, handle, gfp_mask); |
470decc6 DK |
448 | return ret; |
449 | } | |
47def826 | 450 | EXPORT_SYMBOL(jbd2__journal_restart); |
470decc6 DK |
451 | |
452 | ||
47def826 TT |
453 | int jbd2_journal_restart(handle_t *handle, int nblocks) |
454 | { | |
455 | return jbd2__journal_restart(handle, nblocks, GFP_NOFS); | |
456 | } | |
457 | EXPORT_SYMBOL(jbd2_journal_restart); | |
458 | ||
470decc6 | 459 | /** |
f7f4bccb | 460 | * void jbd2_journal_lock_updates () - establish a transaction barrier. |
470decc6 DK |
461 | * @journal: Journal to establish a barrier on. |
462 | * | |
463 | * This locks out any further updates from being started, and blocks | |
464 | * until all existing updates have completed, returning only once the | |
465 | * journal is in a quiescent state with no updates running. | |
466 | * | |
467 | * The journal lock should not be held on entry. | |
468 | */ | |
f7f4bccb | 469 | void jbd2_journal_lock_updates(journal_t *journal) |
470decc6 DK |
470 | { |
471 | DEFINE_WAIT(wait); | |
472 | ||
473 | spin_lock(&journal->j_state_lock); | |
474 | ++journal->j_barrier_count; | |
475 | ||
476 | /* Wait until there are no running updates */ | |
477 | while (1) { | |
478 | transaction_t *transaction = journal->j_running_transaction; | |
479 | ||
480 | if (!transaction) | |
481 | break; | |
482 | ||
483 | spin_lock(&transaction->t_handle_lock); | |
484 | if (!transaction->t_updates) { | |
485 | spin_unlock(&transaction->t_handle_lock); | |
486 | break; | |
487 | } | |
488 | prepare_to_wait(&journal->j_wait_updates, &wait, | |
489 | TASK_UNINTERRUPTIBLE); | |
490 | spin_unlock(&transaction->t_handle_lock); | |
491 | spin_unlock(&journal->j_state_lock); | |
492 | schedule(); | |
493 | finish_wait(&journal->j_wait_updates, &wait); | |
494 | spin_lock(&journal->j_state_lock); | |
495 | } | |
496 | spin_unlock(&journal->j_state_lock); | |
497 | ||
498 | /* | |
499 | * We have now established a barrier against other normal updates, but | |
f7f4bccb | 500 | * we also need to barrier against other jbd2_journal_lock_updates() calls |
470decc6 DK |
501 | * to make sure that we serialise special journal-locked operations |
502 | * too. | |
503 | */ | |
504 | mutex_lock(&journal->j_barrier); | |
505 | } | |
506 | ||
507 | /** | |
f7f4bccb | 508 | * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier |
470decc6 DK |
509 | * @journal: Journal to release the barrier on. |
510 | * | |
f7f4bccb | 511 | * Release a transaction barrier obtained with jbd2_journal_lock_updates(). |
470decc6 DK |
512 | * |
513 | * Should be called without the journal lock held. | |
514 | */ | |
f7f4bccb | 515 | void jbd2_journal_unlock_updates (journal_t *journal) |
470decc6 DK |
516 | { |
517 | J_ASSERT(journal->j_barrier_count != 0); | |
518 | ||
519 | mutex_unlock(&journal->j_barrier); | |
520 | spin_lock(&journal->j_state_lock); | |
521 | --journal->j_barrier_count; | |
522 | spin_unlock(&journal->j_state_lock); | |
523 | wake_up(&journal->j_wait_transaction_locked); | |
524 | } | |
525 | ||
f91d1d04 | 526 | static void warn_dirty_buffer(struct buffer_head *bh) |
470decc6 | 527 | { |
f91d1d04 | 528 | char b[BDEVNAME_SIZE]; |
470decc6 | 529 | |
f91d1d04 JK |
530 | printk(KERN_WARNING |
531 | "JBD: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). " | |
532 | "There's a risk of filesystem corruption in case of system " | |
533 | "crash.\n", | |
534 | bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr); | |
470decc6 DK |
535 | } |
536 | ||
537 | /* | |
538 | * If the buffer is already part of the current transaction, then there | |
539 | * is nothing we need to do. If it is already part of a prior | |
540 | * transaction which we are still committing to disk, then we need to | |
541 | * make sure that we do not overwrite the old copy: we do copy-out to | |
542 | * preserve the copy going to disk. We also account the buffer against | |
543 | * the handle's metadata buffer credits (unless the buffer is already | |
544 | * part of the transaction, that is). | |
545 | * | |
546 | */ | |
547 | static int | |
548 | do_get_write_access(handle_t *handle, struct journal_head *jh, | |
549 | int force_copy) | |
550 | { | |
551 | struct buffer_head *bh; | |
552 | transaction_t *transaction; | |
553 | journal_t *journal; | |
554 | int error; | |
555 | char *frozen_buffer = NULL; | |
556 | int need_copy = 0; | |
557 | ||
558 | if (is_handle_aborted(handle)) | |
559 | return -EROFS; | |
560 | ||
561 | transaction = handle->h_transaction; | |
562 | journal = transaction->t_journal; | |
563 | ||
564 | jbd_debug(5, "buffer_head %p, force_copy %d\n", jh, force_copy); | |
565 | ||
566 | JBUFFER_TRACE(jh, "entry"); | |
567 | repeat: | |
568 | bh = jh2bh(jh); | |
569 | ||
570 | /* @@@ Need to check for errors here at some point. */ | |
571 | ||
572 | lock_buffer(bh); | |
573 | jbd_lock_bh_state(bh); | |
574 | ||
575 | /* We now hold the buffer lock so it is safe to query the buffer | |
576 | * state. Is the buffer dirty? | |
577 | * | |
578 | * If so, there are two possibilities. The buffer may be | |
579 | * non-journaled, and undergoing a quite legitimate writeback. | |
580 | * Otherwise, it is journaled, and we don't expect dirty buffers | |
581 | * in that state (the buffers should be marked JBD_Dirty | |
582 | * instead.) So either the IO is being done under our own | |
583 | * control and this is a bug, or it's a third party IO such as | |
584 | * dump(8) (which may leave the buffer scheduled for read --- | |
585 | * ie. locked but not dirty) or tune2fs (which may actually have | |
586 | * the buffer dirtied, ugh.) */ | |
587 | ||
588 | if (buffer_dirty(bh)) { | |
589 | /* | |
590 | * First question: is this buffer already part of the current | |
591 | * transaction or the existing committing transaction? | |
592 | */ | |
593 | if (jh->b_transaction) { | |
594 | J_ASSERT_JH(jh, | |
595 | jh->b_transaction == transaction || | |
596 | jh->b_transaction == | |
597 | journal->j_committing_transaction); | |
598 | if (jh->b_next_transaction) | |
599 | J_ASSERT_JH(jh, jh->b_next_transaction == | |
600 | transaction); | |
f91d1d04 | 601 | warn_dirty_buffer(bh); |
470decc6 DK |
602 | } |
603 | /* | |
604 | * In any case we need to clean the dirty flag and we must | |
605 | * do it under the buffer lock to be sure we don't race | |
606 | * with running write-out. | |
607 | */ | |
f91d1d04 JK |
608 | JBUFFER_TRACE(jh, "Journalling dirty buffer"); |
609 | clear_buffer_dirty(bh); | |
610 | set_buffer_jbddirty(bh); | |
470decc6 DK |
611 | } |
612 | ||
613 | unlock_buffer(bh); | |
614 | ||
615 | error = -EROFS; | |
616 | if (is_handle_aborted(handle)) { | |
617 | jbd_unlock_bh_state(bh); | |
618 | goto out; | |
619 | } | |
620 | error = 0; | |
621 | ||
622 | /* | |
623 | * The buffer is already part of this transaction if b_transaction or | |
624 | * b_next_transaction points to it | |
625 | */ | |
626 | if (jh->b_transaction == transaction || | |
627 | jh->b_next_transaction == transaction) | |
628 | goto done; | |
629 | ||
9fc7c63a JB |
630 | /* |
631 | * this is the first time this transaction is touching this buffer, | |
632 | * reset the modified flag | |
633 | */ | |
634 | jh->b_modified = 0; | |
635 | ||
470decc6 DK |
636 | /* |
637 | * If there is already a copy-out version of this buffer, then we don't | |
638 | * need to make another one | |
639 | */ | |
640 | if (jh->b_frozen_data) { | |
641 | JBUFFER_TRACE(jh, "has frozen data"); | |
642 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
643 | jh->b_next_transaction = transaction; | |
644 | goto done; | |
645 | } | |
646 | ||
647 | /* Is there data here we need to preserve? */ | |
648 | ||
649 | if (jh->b_transaction && jh->b_transaction != transaction) { | |
650 | JBUFFER_TRACE(jh, "owned by older transaction"); | |
651 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
652 | J_ASSERT_JH(jh, jh->b_transaction == | |
653 | journal->j_committing_transaction); | |
654 | ||
655 | /* There is one case we have to be very careful about. | |
656 | * If the committing transaction is currently writing | |
657 | * this buffer out to disk and has NOT made a copy-out, | |
658 | * then we cannot modify the buffer contents at all | |
659 | * right now. The essence of copy-out is that it is the | |
660 | * extra copy, not the primary copy, which gets | |
661 | * journaled. If the primary copy is already going to | |
662 | * disk then we cannot do copy-out here. */ | |
663 | ||
664 | if (jh->b_jlist == BJ_Shadow) { | |
665 | DEFINE_WAIT_BIT(wait, &bh->b_state, BH_Unshadow); | |
666 | wait_queue_head_t *wqh; | |
667 | ||
668 | wqh = bit_waitqueue(&bh->b_state, BH_Unshadow); | |
669 | ||
670 | JBUFFER_TRACE(jh, "on shadow: sleep"); | |
671 | jbd_unlock_bh_state(bh); | |
672 | /* commit wakes up all shadow buffers after IO */ | |
673 | for ( ; ; ) { | |
674 | prepare_to_wait(wqh, &wait.wait, | |
675 | TASK_UNINTERRUPTIBLE); | |
676 | if (jh->b_jlist != BJ_Shadow) | |
677 | break; | |
678 | schedule(); | |
679 | } | |
680 | finish_wait(wqh, &wait.wait); | |
681 | goto repeat; | |
682 | } | |
683 | ||
684 | /* Only do the copy if the currently-owning transaction | |
685 | * still needs it. If it is on the Forget list, the | |
686 | * committing transaction is past that stage. The | |
687 | * buffer had better remain locked during the kmalloc, | |
688 | * but that should be true --- we hold the journal lock | |
689 | * still and the buffer is already on the BUF_JOURNAL | |
690 | * list so won't be flushed. | |
691 | * | |
692 | * Subtle point, though: if this is a get_undo_access, | |
693 | * then we will be relying on the frozen_data to contain | |
694 | * the new value of the committed_data record after the | |
695 | * transaction, so we HAVE to force the frozen_data copy | |
696 | * in that case. */ | |
697 | ||
698 | if (jh->b_jlist != BJ_Forget || force_copy) { | |
699 | JBUFFER_TRACE(jh, "generate frozen data"); | |
700 | if (!frozen_buffer) { | |
701 | JBUFFER_TRACE(jh, "allocate memory for buffer"); | |
702 | jbd_unlock_bh_state(bh); | |
703 | frozen_buffer = | |
af1e76d6 | 704 | jbd2_alloc(jh2bh(jh)->b_size, |
470decc6 DK |
705 | GFP_NOFS); |
706 | if (!frozen_buffer) { | |
707 | printk(KERN_EMERG | |
708 | "%s: OOM for frozen_buffer\n", | |
329d291f | 709 | __func__); |
470decc6 DK |
710 | JBUFFER_TRACE(jh, "oom!"); |
711 | error = -ENOMEM; | |
712 | jbd_lock_bh_state(bh); | |
713 | goto done; | |
714 | } | |
715 | goto repeat; | |
716 | } | |
717 | jh->b_frozen_data = frozen_buffer; | |
718 | frozen_buffer = NULL; | |
719 | need_copy = 1; | |
720 | } | |
721 | jh->b_next_transaction = transaction; | |
722 | } | |
723 | ||
724 | ||
725 | /* | |
726 | * Finally, if the buffer is not journaled right now, we need to make | |
727 | * sure it doesn't get written to disk before the caller actually | |
728 | * commits the new data | |
729 | */ | |
730 | if (!jh->b_transaction) { | |
731 | JBUFFER_TRACE(jh, "no transaction"); | |
732 | J_ASSERT_JH(jh, !jh->b_next_transaction); | |
733 | jh->b_transaction = transaction; | |
734 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); | |
735 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 736 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
470decc6 DK |
737 | spin_unlock(&journal->j_list_lock); |
738 | } | |
739 | ||
740 | done: | |
741 | if (need_copy) { | |
742 | struct page *page; | |
743 | int offset; | |
744 | char *source; | |
745 | ||
746 | J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)), | |
747 | "Possible IO failure.\n"); | |
748 | page = jh2bh(jh)->b_page; | |
749 | offset = ((unsigned long) jh2bh(jh)->b_data) & ~PAGE_MASK; | |
750 | source = kmap_atomic(page, KM_USER0); | |
751 | memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size); | |
752 | kunmap_atomic(source, KM_USER0); | |
e06c8227 JB |
753 | |
754 | /* | |
755 | * Now that the frozen data is saved off, we need to store | |
756 | * any matching triggers. | |
757 | */ | |
758 | jh->b_frozen_triggers = jh->b_triggers; | |
470decc6 DK |
759 | } |
760 | jbd_unlock_bh_state(bh); | |
761 | ||
762 | /* | |
763 | * If we are about to journal a buffer, then any revoke pending on it is | |
764 | * no longer valid | |
765 | */ | |
f7f4bccb | 766 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 DK |
767 | |
768 | out: | |
769 | if (unlikely(frozen_buffer)) /* It's usually NULL */ | |
af1e76d6 | 770 | jbd2_free(frozen_buffer, bh->b_size); |
470decc6 DK |
771 | |
772 | JBUFFER_TRACE(jh, "exit"); | |
773 | return error; | |
774 | } | |
775 | ||
776 | /** | |
f7f4bccb | 777 | * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update. |
470decc6 DK |
778 | * @handle: transaction to add buffer modifications to |
779 | * @bh: bh to be used for metadata writes | |
780 | * @credits: variable that will receive credits for the buffer | |
781 | * | |
782 | * Returns an error code or 0 on success. | |
783 | * | |
784 | * In full data journalling mode the buffer may be of type BJ_AsyncData, | |
785 | * because we're write()ing a buffer which is also part of a shared mapping. | |
786 | */ | |
787 | ||
f7f4bccb | 788 | int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh) |
470decc6 | 789 | { |
f7f4bccb | 790 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
791 | int rc; |
792 | ||
793 | /* We do not want to get caught playing with fields which the | |
794 | * log thread also manipulates. Make sure that the buffer | |
795 | * completes any outstanding IO before proceeding. */ | |
796 | rc = do_get_write_access(handle, jh, 0); | |
f7f4bccb | 797 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
798 | return rc; |
799 | } | |
800 | ||
801 | ||
802 | /* | |
803 | * When the user wants to journal a newly created buffer_head | |
804 | * (ie. getblk() returned a new buffer and we are going to populate it | |
805 | * manually rather than reading off disk), then we need to keep the | |
806 | * buffer_head locked until it has been completely filled with new | |
807 | * data. In this case, we should be able to make the assertion that | |
808 | * the bh is not already part of an existing transaction. | |
809 | * | |
810 | * The buffer should already be locked by the caller by this point. | |
811 | * There is no lock ranking violation: it was a newly created, | |
812 | * unlocked buffer beforehand. */ | |
813 | ||
814 | /** | |
f7f4bccb | 815 | * int jbd2_journal_get_create_access () - notify intent to use newly created bh |
470decc6 DK |
816 | * @handle: transaction to new buffer to |
817 | * @bh: new buffer. | |
818 | * | |
819 | * Call this if you create a new bh. | |
820 | */ | |
f7f4bccb | 821 | int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
822 | { |
823 | transaction_t *transaction = handle->h_transaction; | |
824 | journal_t *journal = transaction->t_journal; | |
f7f4bccb | 825 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
826 | int err; |
827 | ||
828 | jbd_debug(5, "journal_head %p\n", jh); | |
829 | err = -EROFS; | |
830 | if (is_handle_aborted(handle)) | |
831 | goto out; | |
832 | err = 0; | |
833 | ||
834 | JBUFFER_TRACE(jh, "entry"); | |
835 | /* | |
836 | * The buffer may already belong to this transaction due to pre-zeroing | |
837 | * in the filesystem's new_block code. It may also be on the previous, | |
838 | * committing transaction's lists, but it HAS to be in Forget state in | |
839 | * that case: the transaction must have deleted the buffer for it to be | |
840 | * reused here. | |
841 | */ | |
842 | jbd_lock_bh_state(bh); | |
843 | spin_lock(&journal->j_list_lock); | |
844 | J_ASSERT_JH(jh, (jh->b_transaction == transaction || | |
845 | jh->b_transaction == NULL || | |
846 | (jh->b_transaction == journal->j_committing_transaction && | |
847 | jh->b_jlist == BJ_Forget))); | |
848 | ||
849 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
850 | J_ASSERT_JH(jh, buffer_locked(jh2bh(jh))); | |
851 | ||
852 | if (jh->b_transaction == NULL) { | |
f91d1d04 JK |
853 | /* |
854 | * Previous jbd2_journal_forget() could have left the buffer | |
855 | * with jbddirty bit set because it was being committed. When | |
856 | * the commit finished, we've filed the buffer for | |
857 | * checkpointing and marked it dirty. Now we are reallocating | |
858 | * the buffer so the transaction freeing it must have | |
859 | * committed and so it's safe to clear the dirty bit. | |
860 | */ | |
861 | clear_buffer_dirty(jh2bh(jh)); | |
470decc6 | 862 | jh->b_transaction = transaction; |
9fc7c63a JB |
863 | |
864 | /* first access by this transaction */ | |
865 | jh->b_modified = 0; | |
866 | ||
470decc6 | 867 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); |
f7f4bccb | 868 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
470decc6 | 869 | } else if (jh->b_transaction == journal->j_committing_transaction) { |
9fc7c63a JB |
870 | /* first access by this transaction */ |
871 | jh->b_modified = 0; | |
872 | ||
470decc6 DK |
873 | JBUFFER_TRACE(jh, "set next transaction"); |
874 | jh->b_next_transaction = transaction; | |
875 | } | |
876 | spin_unlock(&journal->j_list_lock); | |
877 | jbd_unlock_bh_state(bh); | |
878 | ||
879 | /* | |
880 | * akpm: I added this. ext3_alloc_branch can pick up new indirect | |
881 | * blocks which contain freed but then revoked metadata. We need | |
882 | * to cancel the revoke in case we end up freeing it yet again | |
883 | * and the reallocating as data - this would cause a second revoke, | |
884 | * which hits an assertion error. | |
885 | */ | |
886 | JBUFFER_TRACE(jh, "cancelling revoke"); | |
f7f4bccb MC |
887 | jbd2_journal_cancel_revoke(handle, jh); |
888 | jbd2_journal_put_journal_head(jh); | |
470decc6 DK |
889 | out: |
890 | return err; | |
891 | } | |
892 | ||
893 | /** | |
f7f4bccb | 894 | * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with |
470decc6 DK |
895 | * non-rewindable consequences |
896 | * @handle: transaction | |
897 | * @bh: buffer to undo | |
898 | * @credits: store the number of taken credits here (if not NULL) | |
899 | * | |
900 | * Sometimes there is a need to distinguish between metadata which has | |
901 | * been committed to disk and that which has not. The ext3fs code uses | |
902 | * this for freeing and allocating space, we have to make sure that we | |
903 | * do not reuse freed space until the deallocation has been committed, | |
904 | * since if we overwrote that space we would make the delete | |
905 | * un-rewindable in case of a crash. | |
906 | * | |
f7f4bccb | 907 | * To deal with that, jbd2_journal_get_undo_access requests write access to a |
470decc6 DK |
908 | * buffer for parts of non-rewindable operations such as delete |
909 | * operations on the bitmaps. The journaling code must keep a copy of | |
910 | * the buffer's contents prior to the undo_access call until such time | |
911 | * as we know that the buffer has definitely been committed to disk. | |
912 | * | |
913 | * We never need to know which transaction the committed data is part | |
914 | * of, buffers touched here are guaranteed to be dirtied later and so | |
915 | * will be committed to a new transaction in due course, at which point | |
916 | * we can discard the old committed data pointer. | |
917 | * | |
918 | * Returns error number or 0 on success. | |
919 | */ | |
f7f4bccb | 920 | int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
921 | { |
922 | int err; | |
f7f4bccb | 923 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
924 | char *committed_data = NULL; |
925 | ||
926 | JBUFFER_TRACE(jh, "entry"); | |
927 | ||
928 | /* | |
929 | * Do this first --- it can drop the journal lock, so we want to | |
930 | * make sure that obtaining the committed_data is done | |
931 | * atomically wrt. completion of any outstanding commits. | |
932 | */ | |
933 | err = do_get_write_access(handle, jh, 1); | |
934 | if (err) | |
935 | goto out; | |
936 | ||
937 | repeat: | |
938 | if (!jh->b_committed_data) { | |
af1e76d6 | 939 | committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS); |
470decc6 DK |
940 | if (!committed_data) { |
941 | printk(KERN_EMERG "%s: No memory for committed data\n", | |
329d291f | 942 | __func__); |
470decc6 DK |
943 | err = -ENOMEM; |
944 | goto out; | |
945 | } | |
946 | } | |
947 | ||
948 | jbd_lock_bh_state(bh); | |
949 | if (!jh->b_committed_data) { | |
950 | /* Copy out the current buffer contents into the | |
951 | * preserved, committed copy. */ | |
952 | JBUFFER_TRACE(jh, "generate b_committed data"); | |
953 | if (!committed_data) { | |
954 | jbd_unlock_bh_state(bh); | |
955 | goto repeat; | |
956 | } | |
957 | ||
958 | jh->b_committed_data = committed_data; | |
959 | committed_data = NULL; | |
960 | memcpy(jh->b_committed_data, bh->b_data, bh->b_size); | |
961 | } | |
962 | jbd_unlock_bh_state(bh); | |
963 | out: | |
f7f4bccb | 964 | jbd2_journal_put_journal_head(jh); |
470decc6 | 965 | if (unlikely(committed_data)) |
af1e76d6 | 966 | jbd2_free(committed_data, bh->b_size); |
470decc6 DK |
967 | return err; |
968 | } | |
969 | ||
e06c8227 JB |
970 | /** |
971 | * void jbd2_journal_set_triggers() - Add triggers for commit writeout | |
972 | * @bh: buffer to trigger on | |
973 | * @type: struct jbd2_buffer_trigger_type containing the trigger(s). | |
974 | * | |
975 | * Set any triggers on this journal_head. This is always safe, because | |
976 | * triggers for a committing buffer will be saved off, and triggers for | |
977 | * a running transaction will match the buffer in that transaction. | |
978 | * | |
979 | * Call with NULL to clear the triggers. | |
980 | */ | |
981 | void jbd2_journal_set_triggers(struct buffer_head *bh, | |
982 | struct jbd2_buffer_trigger_type *type) | |
983 | { | |
984 | struct journal_head *jh = bh2jh(bh); | |
985 | ||
986 | jh->b_triggers = type; | |
987 | } | |
988 | ||
989 | void jbd2_buffer_commit_trigger(struct journal_head *jh, void *mapped_data, | |
990 | struct jbd2_buffer_trigger_type *triggers) | |
991 | { | |
992 | struct buffer_head *bh = jh2bh(jh); | |
993 | ||
994 | if (!triggers || !triggers->t_commit) | |
995 | return; | |
996 | ||
997 | triggers->t_commit(triggers, bh, mapped_data, bh->b_size); | |
998 | } | |
999 | ||
1000 | void jbd2_buffer_abort_trigger(struct journal_head *jh, | |
1001 | struct jbd2_buffer_trigger_type *triggers) | |
1002 | { | |
1003 | if (!triggers || !triggers->t_abort) | |
1004 | return; | |
1005 | ||
1006 | triggers->t_abort(triggers, jh2bh(jh)); | |
1007 | } | |
1008 | ||
1009 | ||
1010 | ||
470decc6 | 1011 | /** |
f7f4bccb | 1012 | * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata |
470decc6 DK |
1013 | * @handle: transaction to add buffer to. |
1014 | * @bh: buffer to mark | |
1015 | * | |
1016 | * mark dirty metadata which needs to be journaled as part of the current | |
1017 | * transaction. | |
1018 | * | |
1019 | * The buffer is placed on the transaction's metadata list and is marked | |
1020 | * as belonging to the transaction. | |
1021 | * | |
1022 | * Returns error number or 0 on success. | |
1023 | * | |
1024 | * Special care needs to be taken if the buffer already belongs to the | |
1025 | * current committing transaction (in which case we should have frozen | |
1026 | * data present for that commit). In that case, we don't relink the | |
1027 | * buffer: that only gets done when the old transaction finally | |
1028 | * completes its commit. | |
1029 | */ | |
f7f4bccb | 1030 | int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1031 | { |
1032 | transaction_t *transaction = handle->h_transaction; | |
1033 | journal_t *journal = transaction->t_journal; | |
1034 | struct journal_head *jh = bh2jh(bh); | |
1035 | ||
1036 | jbd_debug(5, "journal_head %p\n", jh); | |
1037 | JBUFFER_TRACE(jh, "entry"); | |
1038 | if (is_handle_aborted(handle)) | |
1039 | goto out; | |
1040 | ||
1041 | jbd_lock_bh_state(bh); | |
1042 | ||
1043 | if (jh->b_modified == 0) { | |
1044 | /* | |
1045 | * This buffer's got modified and becoming part | |
1046 | * of the transaction. This needs to be done | |
1047 | * once a transaction -bzzz | |
1048 | */ | |
1049 | jh->b_modified = 1; | |
1050 | J_ASSERT_JH(jh, handle->h_buffer_credits > 0); | |
1051 | handle->h_buffer_credits--; | |
1052 | } | |
1053 | ||
1054 | /* | |
1055 | * fastpath, to avoid expensive locking. If this buffer is already | |
1056 | * on the running transaction's metadata list there is nothing to do. | |
1057 | * Nobody can take it off again because there is a handle open. | |
1058 | * I _think_ we're OK here with SMP barriers - a mistaken decision will | |
1059 | * result in this test being false, so we go in and take the locks. | |
1060 | */ | |
1061 | if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) { | |
1062 | JBUFFER_TRACE(jh, "fastpath"); | |
1063 | J_ASSERT_JH(jh, jh->b_transaction == | |
1064 | journal->j_running_transaction); | |
1065 | goto out_unlock_bh; | |
1066 | } | |
1067 | ||
1068 | set_buffer_jbddirty(bh); | |
1069 | ||
1070 | /* | |
1071 | * Metadata already on the current transaction list doesn't | |
1072 | * need to be filed. Metadata on another transaction's list must | |
1073 | * be committing, and will be refiled once the commit completes: | |
1074 | * leave it alone for now. | |
1075 | */ | |
1076 | if (jh->b_transaction != transaction) { | |
1077 | JBUFFER_TRACE(jh, "already on other transaction"); | |
1078 | J_ASSERT_JH(jh, jh->b_transaction == | |
1079 | journal->j_committing_transaction); | |
1080 | J_ASSERT_JH(jh, jh->b_next_transaction == transaction); | |
1081 | /* And this case is illegal: we can't reuse another | |
1082 | * transaction's data buffer, ever. */ | |
1083 | goto out_unlock_bh; | |
1084 | } | |
1085 | ||
1086 | /* That test should have eliminated the following case: */ | |
4019191b | 1087 | J_ASSERT_JH(jh, jh->b_frozen_data == NULL); |
470decc6 DK |
1088 | |
1089 | JBUFFER_TRACE(jh, "file as BJ_Metadata"); | |
1090 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 1091 | __jbd2_journal_file_buffer(jh, handle->h_transaction, BJ_Metadata); |
470decc6 DK |
1092 | spin_unlock(&journal->j_list_lock); |
1093 | out_unlock_bh: | |
1094 | jbd_unlock_bh_state(bh); | |
1095 | out: | |
1096 | JBUFFER_TRACE(jh, "exit"); | |
1097 | return 0; | |
1098 | } | |
1099 | ||
1100 | /* | |
f7f4bccb | 1101 | * jbd2_journal_release_buffer: undo a get_write_access without any buffer |
470decc6 DK |
1102 | * updates, if the update decided in the end that it didn't need access. |
1103 | * | |
1104 | */ | |
1105 | void | |
f7f4bccb | 1106 | jbd2_journal_release_buffer(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1107 | { |
1108 | BUFFER_TRACE(bh, "entry"); | |
1109 | } | |
1110 | ||
1111 | /** | |
f7f4bccb | 1112 | * void jbd2_journal_forget() - bforget() for potentially-journaled buffers. |
470decc6 DK |
1113 | * @handle: transaction handle |
1114 | * @bh: bh to 'forget' | |
1115 | * | |
1116 | * We can only do the bforget if there are no commits pending against the | |
1117 | * buffer. If the buffer is dirty in the current running transaction we | |
1118 | * can safely unlink it. | |
1119 | * | |
1120 | * bh may not be a journalled buffer at all - it may be a non-JBD | |
1121 | * buffer which came off the hashtable. Check for this. | |
1122 | * | |
1123 | * Decrements bh->b_count by one. | |
1124 | * | |
1125 | * Allow this call even if the handle has aborted --- it may be part of | |
1126 | * the caller's cleanup after an abort. | |
1127 | */ | |
f7f4bccb | 1128 | int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1129 | { |
1130 | transaction_t *transaction = handle->h_transaction; | |
1131 | journal_t *journal = transaction->t_journal; | |
1132 | struct journal_head *jh; | |
1133 | int drop_reserve = 0; | |
1134 | int err = 0; | |
1dfc3220 | 1135 | int was_modified = 0; |
470decc6 DK |
1136 | |
1137 | BUFFER_TRACE(bh, "entry"); | |
1138 | ||
1139 | jbd_lock_bh_state(bh); | |
1140 | spin_lock(&journal->j_list_lock); | |
1141 | ||
1142 | if (!buffer_jbd(bh)) | |
1143 | goto not_jbd; | |
1144 | jh = bh2jh(bh); | |
1145 | ||
1146 | /* Critical error: attempting to delete a bitmap buffer, maybe? | |
1147 | * Don't do any jbd operations, and return an error. */ | |
1148 | if (!J_EXPECT_JH(jh, !jh->b_committed_data, | |
1149 | "inconsistent data on disk")) { | |
1150 | err = -EIO; | |
1151 | goto not_jbd; | |
1152 | } | |
1153 | ||
1dfc3220 JB |
1154 | /* keep track of wether or not this transaction modified us */ |
1155 | was_modified = jh->b_modified; | |
1156 | ||
470decc6 DK |
1157 | /* |
1158 | * The buffer's going from the transaction, we must drop | |
1159 | * all references -bzzz | |
1160 | */ | |
1161 | jh->b_modified = 0; | |
1162 | ||
1163 | if (jh->b_transaction == handle->h_transaction) { | |
1164 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
1165 | ||
1166 | /* If we are forgetting a buffer which is already part | |
1167 | * of this transaction, then we can just drop it from | |
1168 | * the transaction immediately. */ | |
1169 | clear_buffer_dirty(bh); | |
1170 | clear_buffer_jbddirty(bh); | |
1171 | ||
1172 | JBUFFER_TRACE(jh, "belongs to current transaction: unfile"); | |
1173 | ||
1dfc3220 JB |
1174 | /* |
1175 | * we only want to drop a reference if this transaction | |
1176 | * modified the buffer | |
1177 | */ | |
1178 | if (was_modified) | |
1179 | drop_reserve = 1; | |
470decc6 DK |
1180 | |
1181 | /* | |
1182 | * We are no longer going to journal this buffer. | |
1183 | * However, the commit of this transaction is still | |
1184 | * important to the buffer: the delete that we are now | |
1185 | * processing might obsolete an old log entry, so by | |
1186 | * committing, we can satisfy the buffer's checkpoint. | |
1187 | * | |
1188 | * So, if we have a checkpoint on the buffer, we should | |
1189 | * now refile the buffer on our BJ_Forget list so that | |
1190 | * we know to remove the checkpoint after we commit. | |
1191 | */ | |
1192 | ||
1193 | if (jh->b_cp_transaction) { | |
f7f4bccb MC |
1194 | __jbd2_journal_temp_unlink_buffer(jh); |
1195 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); | |
470decc6 | 1196 | } else { |
f7f4bccb MC |
1197 | __jbd2_journal_unfile_buffer(jh); |
1198 | jbd2_journal_remove_journal_head(bh); | |
470decc6 DK |
1199 | __brelse(bh); |
1200 | if (!buffer_jbd(bh)) { | |
1201 | spin_unlock(&journal->j_list_lock); | |
1202 | jbd_unlock_bh_state(bh); | |
1203 | __bforget(bh); | |
1204 | goto drop; | |
1205 | } | |
1206 | } | |
1207 | } else if (jh->b_transaction) { | |
1208 | J_ASSERT_JH(jh, (jh->b_transaction == | |
1209 | journal->j_committing_transaction)); | |
1210 | /* However, if the buffer is still owned by a prior | |
1211 | * (committing) transaction, we can't drop it yet... */ | |
1212 | JBUFFER_TRACE(jh, "belongs to older transaction"); | |
1213 | /* ... but we CAN drop it from the new transaction if we | |
1214 | * have also modified it since the original commit. */ | |
1215 | ||
1216 | if (jh->b_next_transaction) { | |
1217 | J_ASSERT(jh->b_next_transaction == transaction); | |
1218 | jh->b_next_transaction = NULL; | |
1dfc3220 JB |
1219 | |
1220 | /* | |
1221 | * only drop a reference if this transaction modified | |
1222 | * the buffer | |
1223 | */ | |
1224 | if (was_modified) | |
1225 | drop_reserve = 1; | |
470decc6 DK |
1226 | } |
1227 | } | |
1228 | ||
1229 | not_jbd: | |
1230 | spin_unlock(&journal->j_list_lock); | |
1231 | jbd_unlock_bh_state(bh); | |
1232 | __brelse(bh); | |
1233 | drop: | |
1234 | if (drop_reserve) { | |
1235 | /* no need to reserve log space for this block -bzzz */ | |
1236 | handle->h_buffer_credits++; | |
1237 | } | |
1238 | return err; | |
1239 | } | |
1240 | ||
1241 | /** | |
f7f4bccb | 1242 | * int jbd2_journal_stop() - complete a transaction |
470decc6 DK |
1243 | * @handle: tranaction to complete. |
1244 | * | |
1245 | * All done for a particular handle. | |
1246 | * | |
1247 | * There is not much action needed here. We just return any remaining | |
1248 | * buffer credits to the transaction and remove the handle. The only | |
1249 | * complication is that we need to start a commit operation if the | |
1250 | * filesystem is marked for synchronous update. | |
1251 | * | |
f7f4bccb | 1252 | * jbd2_journal_stop itself will not usually return an error, but it may |
470decc6 | 1253 | * do so in unusual circumstances. In particular, expect it to |
f7f4bccb | 1254 | * return -EIO if a jbd2_journal_abort has been executed since the |
470decc6 DK |
1255 | * transaction began. |
1256 | */ | |
f7f4bccb | 1257 | int jbd2_journal_stop(handle_t *handle) |
470decc6 DK |
1258 | { |
1259 | transaction_t *transaction = handle->h_transaction; | |
1260 | journal_t *journal = transaction->t_journal; | |
e07f7183 | 1261 | int err; |
470decc6 DK |
1262 | pid_t pid; |
1263 | ||
470decc6 DK |
1264 | J_ASSERT(journal_current_handle() == handle); |
1265 | ||
1266 | if (is_handle_aborted(handle)) | |
1267 | err = -EIO; | |
3e2a532b OH |
1268 | else { |
1269 | J_ASSERT(transaction->t_updates > 0); | |
470decc6 | 1270 | err = 0; |
3e2a532b | 1271 | } |
470decc6 DK |
1272 | |
1273 | if (--handle->h_ref > 0) { | |
1274 | jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1, | |
1275 | handle->h_ref); | |
1276 | return err; | |
1277 | } | |
1278 | ||
1279 | jbd_debug(4, "Handle %p going down\n", handle); | |
1280 | ||
1281 | /* | |
1282 | * Implement synchronous transaction batching. If the handle | |
1283 | * was synchronous, don't force a commit immediately. Let's | |
e07f7183 JB |
1284 | * yield and let another thread piggyback onto this |
1285 | * transaction. Keep doing that while new threads continue to | |
1286 | * arrive. It doesn't cost much - we're about to run a commit | |
1287 | * and sleep on IO anyway. Speeds up many-threaded, many-dir | |
1288 | * operations by 30x or more... | |
1289 | * | |
1290 | * We try and optimize the sleep time against what the | |
1291 | * underlying disk can do, instead of having a static sleep | |
1292 | * time. This is useful for the case where our storage is so | |
1293 | * fast that it is more optimal to go ahead and force a flush | |
1294 | * and wait for the transaction to be committed than it is to | |
1295 | * wait for an arbitrary amount of time for new writers to | |
1296 | * join the transaction. We achieve this by measuring how | |
1297 | * long it takes to commit a transaction, and compare it with | |
1298 | * how long this transaction has been running, and if run time | |
1299 | * < commit time then we sleep for the delta and commit. This | |
1300 | * greatly helps super fast disks that would see slowdowns as | |
1301 | * more threads started doing fsyncs. | |
470decc6 | 1302 | * |
e07f7183 JB |
1303 | * But don't do this if this process was the most recent one |
1304 | * to perform a synchronous write. We do this to detect the | |
1305 | * case where a single process is doing a stream of sync | |
1306 | * writes. No point in waiting for joiners in that case. | |
470decc6 DK |
1307 | */ |
1308 | pid = current->pid; | |
1309 | if (handle->h_sync && journal->j_last_sync_writer != pid) { | |
e07f7183 JB |
1310 | u64 commit_time, trans_time; |
1311 | ||
470decc6 | 1312 | journal->j_last_sync_writer = pid; |
e07f7183 JB |
1313 | |
1314 | spin_lock(&journal->j_state_lock); | |
1315 | commit_time = journal->j_average_commit_time; | |
1316 | spin_unlock(&journal->j_state_lock); | |
1317 | ||
1318 | trans_time = ktime_to_ns(ktime_sub(ktime_get(), | |
1319 | transaction->t_start_time)); | |
1320 | ||
30773840 TT |
1321 | commit_time = max_t(u64, commit_time, |
1322 | 1000*journal->j_min_batch_time); | |
e07f7183 | 1323 | commit_time = min_t(u64, commit_time, |
30773840 | 1324 | 1000*journal->j_max_batch_time); |
e07f7183 JB |
1325 | |
1326 | if (trans_time < commit_time) { | |
1327 | ktime_t expires = ktime_add_ns(ktime_get(), | |
1328 | commit_time); | |
1329 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1330 | schedule_hrtimeout(&expires, HRTIMER_MODE_ABS); | |
1331 | } | |
470decc6 DK |
1332 | } |
1333 | ||
7058548c TT |
1334 | if (handle->h_sync) |
1335 | transaction->t_synchronous_commit = 1; | |
470decc6 | 1336 | current->journal_info = NULL; |
470decc6 DK |
1337 | spin_lock(&transaction->t_handle_lock); |
1338 | transaction->t_outstanding_credits -= handle->h_buffer_credits; | |
1339 | transaction->t_updates--; | |
1340 | if (!transaction->t_updates) { | |
1341 | wake_up(&journal->j_wait_updates); | |
1342 | if (journal->j_barrier_count) | |
1343 | wake_up(&journal->j_wait_transaction_locked); | |
1344 | } | |
1345 | ||
1346 | /* | |
1347 | * If the handle is marked SYNC, we need to set another commit | |
1348 | * going! We also want to force a commit if the current | |
1349 | * transaction is occupying too much of the log, or if the | |
1350 | * transaction is too old now. | |
1351 | */ | |
1352 | if (handle->h_sync || | |
1353 | transaction->t_outstanding_credits > | |
1354 | journal->j_max_transaction_buffers || | |
1355 | time_after_eq(jiffies, transaction->t_expires)) { | |
1356 | /* Do this even for aborted journals: an abort still | |
1357 | * completes the commit thread, it just doesn't write | |
1358 | * anything to disk. */ | |
1359 | tid_t tid = transaction->t_tid; | |
1360 | ||
1361 | spin_unlock(&transaction->t_handle_lock); | |
1362 | jbd_debug(2, "transaction too old, requesting commit for " | |
1363 | "handle %p\n", handle); | |
1364 | /* This is non-blocking */ | |
c35a56a0 | 1365 | jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1366 | |
1367 | /* | |
f7f4bccb | 1368 | * Special case: JBD2_SYNC synchronous updates require us |
470decc6 DK |
1369 | * to wait for the commit to complete. |
1370 | */ | |
1371 | if (handle->h_sync && !(current->flags & PF_MEMALLOC)) | |
f7f4bccb | 1372 | err = jbd2_log_wait_commit(journal, tid); |
470decc6 DK |
1373 | } else { |
1374 | spin_unlock(&transaction->t_handle_lock); | |
470decc6 DK |
1375 | } |
1376 | ||
3295f0ef | 1377 | lock_map_release(&handle->h_lockdep_map); |
7b751066 | 1378 | |
af1e76d6 | 1379 | jbd2_free_handle(handle); |
470decc6 DK |
1380 | return err; |
1381 | } | |
1382 | ||
5648ba5b RD |
1383 | /** |
1384 | * int jbd2_journal_force_commit() - force any uncommitted transactions | |
470decc6 DK |
1385 | * @journal: journal to force |
1386 | * | |
1387 | * For synchronous operations: force any uncommitted transactions | |
1388 | * to disk. May seem kludgy, but it reuses all the handle batching | |
1389 | * code in a very simple manner. | |
1390 | */ | |
f7f4bccb | 1391 | int jbd2_journal_force_commit(journal_t *journal) |
470decc6 DK |
1392 | { |
1393 | handle_t *handle; | |
1394 | int ret; | |
1395 | ||
f7f4bccb | 1396 | handle = jbd2_journal_start(journal, 1); |
470decc6 DK |
1397 | if (IS_ERR(handle)) { |
1398 | ret = PTR_ERR(handle); | |
1399 | } else { | |
1400 | handle->h_sync = 1; | |
f7f4bccb | 1401 | ret = jbd2_journal_stop(handle); |
470decc6 DK |
1402 | } |
1403 | return ret; | |
1404 | } | |
1405 | ||
1406 | /* | |
1407 | * | |
1408 | * List management code snippets: various functions for manipulating the | |
1409 | * transaction buffer lists. | |
1410 | * | |
1411 | */ | |
1412 | ||
1413 | /* | |
1414 | * Append a buffer to a transaction list, given the transaction's list head | |
1415 | * pointer. | |
1416 | * | |
1417 | * j_list_lock is held. | |
1418 | * | |
1419 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1420 | */ | |
1421 | ||
1422 | static inline void | |
1423 | __blist_add_buffer(struct journal_head **list, struct journal_head *jh) | |
1424 | { | |
1425 | if (!*list) { | |
1426 | jh->b_tnext = jh->b_tprev = jh; | |
1427 | *list = jh; | |
1428 | } else { | |
1429 | /* Insert at the tail of the list to preserve order */ | |
1430 | struct journal_head *first = *list, *last = first->b_tprev; | |
1431 | jh->b_tprev = last; | |
1432 | jh->b_tnext = first; | |
1433 | last->b_tnext = first->b_tprev = jh; | |
1434 | } | |
1435 | } | |
1436 | ||
1437 | /* | |
1438 | * Remove a buffer from a transaction list, given the transaction's list | |
1439 | * head pointer. | |
1440 | * | |
1441 | * Called with j_list_lock held, and the journal may not be locked. | |
1442 | * | |
1443 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1444 | */ | |
1445 | ||
1446 | static inline void | |
1447 | __blist_del_buffer(struct journal_head **list, struct journal_head *jh) | |
1448 | { | |
1449 | if (*list == jh) { | |
1450 | *list = jh->b_tnext; | |
1451 | if (*list == jh) | |
1452 | *list = NULL; | |
1453 | } | |
1454 | jh->b_tprev->b_tnext = jh->b_tnext; | |
1455 | jh->b_tnext->b_tprev = jh->b_tprev; | |
1456 | } | |
1457 | ||
1458 | /* | |
1459 | * Remove a buffer from the appropriate transaction list. | |
1460 | * | |
1461 | * Note that this function can *change* the value of | |
87c89c23 JK |
1462 | * bh->b_transaction->t_buffers, t_forget, t_iobuf_list, t_shadow_list, |
1463 | * t_log_list or t_reserved_list. If the caller is holding onto a copy of one | |
1464 | * of these pointers, it could go bad. Generally the caller needs to re-read | |
1465 | * the pointer from the transaction_t. | |
470decc6 DK |
1466 | * |
1467 | * Called under j_list_lock. The journal may not be locked. | |
1468 | */ | |
f7f4bccb | 1469 | void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh) |
470decc6 DK |
1470 | { |
1471 | struct journal_head **list = NULL; | |
1472 | transaction_t *transaction; | |
1473 | struct buffer_head *bh = jh2bh(jh); | |
1474 | ||
1475 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
1476 | transaction = jh->b_transaction; | |
1477 | if (transaction) | |
1478 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
1479 | ||
1480 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
1481 | if (jh->b_jlist != BJ_None) | |
4019191b | 1482 | J_ASSERT_JH(jh, transaction != NULL); |
470decc6 DK |
1483 | |
1484 | switch (jh->b_jlist) { | |
1485 | case BJ_None: | |
1486 | return; | |
470decc6 DK |
1487 | case BJ_Metadata: |
1488 | transaction->t_nr_buffers--; | |
1489 | J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0); | |
1490 | list = &transaction->t_buffers; | |
1491 | break; | |
1492 | case BJ_Forget: | |
1493 | list = &transaction->t_forget; | |
1494 | break; | |
1495 | case BJ_IO: | |
1496 | list = &transaction->t_iobuf_list; | |
1497 | break; | |
1498 | case BJ_Shadow: | |
1499 | list = &transaction->t_shadow_list; | |
1500 | break; | |
1501 | case BJ_LogCtl: | |
1502 | list = &transaction->t_log_list; | |
1503 | break; | |
1504 | case BJ_Reserved: | |
1505 | list = &transaction->t_reserved_list; | |
1506 | break; | |
470decc6 DK |
1507 | } |
1508 | ||
1509 | __blist_del_buffer(list, jh); | |
1510 | jh->b_jlist = BJ_None; | |
1511 | if (test_clear_buffer_jbddirty(bh)) | |
1512 | mark_buffer_dirty(bh); /* Expose it to the VM */ | |
1513 | } | |
1514 | ||
f7f4bccb | 1515 | void __jbd2_journal_unfile_buffer(struct journal_head *jh) |
470decc6 | 1516 | { |
f7f4bccb | 1517 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 DK |
1518 | jh->b_transaction = NULL; |
1519 | } | |
1520 | ||
f7f4bccb | 1521 | void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 DK |
1522 | { |
1523 | jbd_lock_bh_state(jh2bh(jh)); | |
1524 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 1525 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
1526 | spin_unlock(&journal->j_list_lock); |
1527 | jbd_unlock_bh_state(jh2bh(jh)); | |
1528 | } | |
1529 | ||
1530 | /* | |
f7f4bccb | 1531 | * Called from jbd2_journal_try_to_free_buffers(). |
470decc6 DK |
1532 | * |
1533 | * Called under jbd_lock_bh_state(bh) | |
1534 | */ | |
1535 | static void | |
1536 | __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh) | |
1537 | { | |
1538 | struct journal_head *jh; | |
1539 | ||
1540 | jh = bh2jh(bh); | |
1541 | ||
1542 | if (buffer_locked(bh) || buffer_dirty(bh)) | |
1543 | goto out; | |
1544 | ||
4019191b | 1545 | if (jh->b_next_transaction != NULL) |
470decc6 DK |
1546 | goto out; |
1547 | ||
1548 | spin_lock(&journal->j_list_lock); | |
87c89c23 | 1549 | if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) { |
470decc6 DK |
1550 | /* written-back checkpointed metadata buffer */ |
1551 | if (jh->b_jlist == BJ_None) { | |
1552 | JBUFFER_TRACE(jh, "remove from checkpoint list"); | |
f7f4bccb MC |
1553 | __jbd2_journal_remove_checkpoint(jh); |
1554 | jbd2_journal_remove_journal_head(bh); | |
470decc6 DK |
1555 | __brelse(bh); |
1556 | } | |
1557 | } | |
1558 | spin_unlock(&journal->j_list_lock); | |
1559 | out: | |
1560 | return; | |
1561 | } | |
1562 | ||
470decc6 | 1563 | /** |
f7f4bccb | 1564 | * int jbd2_journal_try_to_free_buffers() - try to free page buffers. |
470decc6 DK |
1565 | * @journal: journal for operation |
1566 | * @page: to try and free | |
530576bb MC |
1567 | * @gfp_mask: we use the mask to detect how hard should we try to release |
1568 | * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to | |
1569 | * release the buffers. | |
470decc6 DK |
1570 | * |
1571 | * | |
1572 | * For all the buffers on this page, | |
1573 | * if they are fully written out ordered data, move them onto BUF_CLEAN | |
1574 | * so try_to_free_buffers() can reap them. | |
1575 | * | |
1576 | * This function returns non-zero if we wish try_to_free_buffers() | |
1577 | * to be called. We do this if the page is releasable by try_to_free_buffers(). | |
1578 | * We also do it if the page has locked or dirty buffers and the caller wants | |
1579 | * us to perform sync or async writeout. | |
1580 | * | |
1581 | * This complicates JBD locking somewhat. We aren't protected by the | |
1582 | * BKL here. We wish to remove the buffer from its committing or | |
f7f4bccb | 1583 | * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer. |
470decc6 DK |
1584 | * |
1585 | * This may *change* the value of transaction_t->t_datalist, so anyone | |
1586 | * who looks at t_datalist needs to lock against this function. | |
1587 | * | |
f7f4bccb MC |
1588 | * Even worse, someone may be doing a jbd2_journal_dirty_data on this |
1589 | * buffer. So we need to lock against that. jbd2_journal_dirty_data() | |
470decc6 DK |
1590 | * will come out of the lock with the buffer dirty, which makes it |
1591 | * ineligible for release here. | |
1592 | * | |
1593 | * Who else is affected by this? hmm... Really the only contender | |
1594 | * is do_get_write_access() - it could be looking at the buffer while | |
1595 | * journal_try_to_free_buffer() is changing its state. But that | |
1596 | * cannot happen because we never reallocate freed data as metadata | |
1597 | * while the data is part of a transaction. Yes? | |
530576bb MC |
1598 | * |
1599 | * Return 0 on failure, 1 on success | |
470decc6 | 1600 | */ |
f7f4bccb | 1601 | int jbd2_journal_try_to_free_buffers(journal_t *journal, |
530576bb | 1602 | struct page *page, gfp_t gfp_mask) |
470decc6 DK |
1603 | { |
1604 | struct buffer_head *head; | |
1605 | struct buffer_head *bh; | |
1606 | int ret = 0; | |
1607 | ||
1608 | J_ASSERT(PageLocked(page)); | |
1609 | ||
1610 | head = page_buffers(page); | |
1611 | bh = head; | |
1612 | do { | |
1613 | struct journal_head *jh; | |
1614 | ||
1615 | /* | |
1616 | * We take our own ref against the journal_head here to avoid | |
1617 | * having to add tons of locking around each instance of | |
530576bb MC |
1618 | * jbd2_journal_remove_journal_head() and |
1619 | * jbd2_journal_put_journal_head(). | |
470decc6 | 1620 | */ |
f7f4bccb | 1621 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
1622 | if (!jh) |
1623 | continue; | |
1624 | ||
1625 | jbd_lock_bh_state(bh); | |
1626 | __journal_try_to_free_buffer(journal, bh); | |
f7f4bccb | 1627 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1628 | jbd_unlock_bh_state(bh); |
1629 | if (buffer_jbd(bh)) | |
1630 | goto busy; | |
1631 | } while ((bh = bh->b_this_page) != head); | |
530576bb | 1632 | |
470decc6 | 1633 | ret = try_to_free_buffers(page); |
530576bb | 1634 | |
470decc6 DK |
1635 | busy: |
1636 | return ret; | |
1637 | } | |
1638 | ||
1639 | /* | |
1640 | * This buffer is no longer needed. If it is on an older transaction's | |
1641 | * checkpoint list we need to record it on this transaction's forget list | |
1642 | * to pin this buffer (and hence its checkpointing transaction) down until | |
1643 | * this transaction commits. If the buffer isn't on a checkpoint list, we | |
1644 | * release it. | |
1645 | * Returns non-zero if JBD no longer has an interest in the buffer. | |
1646 | * | |
1647 | * Called under j_list_lock. | |
1648 | * | |
1649 | * Called under jbd_lock_bh_state(bh). | |
1650 | */ | |
1651 | static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction) | |
1652 | { | |
1653 | int may_free = 1; | |
1654 | struct buffer_head *bh = jh2bh(jh); | |
1655 | ||
f7f4bccb | 1656 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
1657 | |
1658 | if (jh->b_cp_transaction) { | |
1659 | JBUFFER_TRACE(jh, "on running+cp transaction"); | |
f91d1d04 JK |
1660 | /* |
1661 | * We don't want to write the buffer anymore, clear the | |
1662 | * bit so that we don't confuse checks in | |
1663 | * __journal_file_buffer | |
1664 | */ | |
1665 | clear_buffer_dirty(bh); | |
f7f4bccb | 1666 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); |
470decc6 DK |
1667 | may_free = 0; |
1668 | } else { | |
1669 | JBUFFER_TRACE(jh, "on running transaction"); | |
f7f4bccb | 1670 | jbd2_journal_remove_journal_head(bh); |
470decc6 DK |
1671 | __brelse(bh); |
1672 | } | |
1673 | return may_free; | |
1674 | } | |
1675 | ||
1676 | /* | |
f7f4bccb | 1677 | * jbd2_journal_invalidatepage |
470decc6 DK |
1678 | * |
1679 | * This code is tricky. It has a number of cases to deal with. | |
1680 | * | |
1681 | * There are two invariants which this code relies on: | |
1682 | * | |
1683 | * i_size must be updated on disk before we start calling invalidatepage on the | |
1684 | * data. | |
1685 | * | |
1686 | * This is done in ext3 by defining an ext3_setattr method which | |
1687 | * updates i_size before truncate gets going. By maintaining this | |
1688 | * invariant, we can be sure that it is safe to throw away any buffers | |
1689 | * attached to the current transaction: once the transaction commits, | |
1690 | * we know that the data will not be needed. | |
1691 | * | |
1692 | * Note however that we can *not* throw away data belonging to the | |
1693 | * previous, committing transaction! | |
1694 | * | |
1695 | * Any disk blocks which *are* part of the previous, committing | |
1696 | * transaction (and which therefore cannot be discarded immediately) are | |
1697 | * not going to be reused in the new running transaction | |
1698 | * | |
1699 | * The bitmap committed_data images guarantee this: any block which is | |
1700 | * allocated in one transaction and removed in the next will be marked | |
1701 | * as in-use in the committed_data bitmap, so cannot be reused until | |
1702 | * the next transaction to delete the block commits. This means that | |
1703 | * leaving committing buffers dirty is quite safe: the disk blocks | |
1704 | * cannot be reallocated to a different file and so buffer aliasing is | |
1705 | * not possible. | |
1706 | * | |
1707 | * | |
1708 | * The above applies mainly to ordered data mode. In writeback mode we | |
1709 | * don't make guarantees about the order in which data hits disk --- in | |
1710 | * particular we don't guarantee that new dirty data is flushed before | |
1711 | * transaction commit --- so it is always safe just to discard data | |
1712 | * immediately in that mode. --sct | |
1713 | */ | |
1714 | ||
1715 | /* | |
1716 | * The journal_unmap_buffer helper function returns zero if the buffer | |
1717 | * concerned remains pinned as an anonymous buffer belonging to an older | |
1718 | * transaction. | |
1719 | * | |
1720 | * We're outside-transaction here. Either or both of j_running_transaction | |
1721 | * and j_committing_transaction may be NULL. | |
1722 | */ | |
1723 | static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh) | |
1724 | { | |
1725 | transaction_t *transaction; | |
1726 | struct journal_head *jh; | |
1727 | int may_free = 1; | |
1728 | int ret; | |
1729 | ||
1730 | BUFFER_TRACE(bh, "entry"); | |
1731 | ||
1732 | /* | |
1733 | * It is safe to proceed here without the j_list_lock because the | |
1734 | * buffers cannot be stolen by try_to_free_buffers as long as we are | |
1735 | * holding the page lock. --sct | |
1736 | */ | |
1737 | ||
1738 | if (!buffer_jbd(bh)) | |
1739 | goto zap_buffer_unlocked; | |
1740 | ||
87c89c23 | 1741 | /* OK, we have data buffer in journaled mode */ |
470decc6 DK |
1742 | spin_lock(&journal->j_state_lock); |
1743 | jbd_lock_bh_state(bh); | |
1744 | spin_lock(&journal->j_list_lock); | |
1745 | ||
f7f4bccb | 1746 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
1747 | if (!jh) |
1748 | goto zap_buffer_no_jh; | |
1749 | ||
ba869023 | 1750 | /* |
1751 | * We cannot remove the buffer from checkpoint lists until the | |
1752 | * transaction adding inode to orphan list (let's call it T) | |
1753 | * is committed. Otherwise if the transaction changing the | |
1754 | * buffer would be cleaned from the journal before T is | |
1755 | * committed, a crash will cause that the correct contents of | |
1756 | * the buffer will be lost. On the other hand we have to | |
1757 | * clear the buffer dirty bit at latest at the moment when the | |
1758 | * transaction marking the buffer as freed in the filesystem | |
1759 | * structures is committed because from that moment on the | |
1760 | * buffer can be reallocated and used by a different page. | |
1761 | * Since the block hasn't been freed yet but the inode has | |
1762 | * already been added to orphan list, it is safe for us to add | |
1763 | * the buffer to BJ_Forget list of the newest transaction. | |
1764 | */ | |
470decc6 DK |
1765 | transaction = jh->b_transaction; |
1766 | if (transaction == NULL) { | |
1767 | /* First case: not on any transaction. If it | |
1768 | * has no checkpoint link, then we can zap it: | |
1769 | * it's a writeback-mode buffer so we don't care | |
1770 | * if it hits disk safely. */ | |
1771 | if (!jh->b_cp_transaction) { | |
1772 | JBUFFER_TRACE(jh, "not on any transaction: zap"); | |
1773 | goto zap_buffer; | |
1774 | } | |
1775 | ||
1776 | if (!buffer_dirty(bh)) { | |
1777 | /* bdflush has written it. We can drop it now */ | |
1778 | goto zap_buffer; | |
1779 | } | |
1780 | ||
1781 | /* OK, it must be in the journal but still not | |
1782 | * written fully to disk: it's metadata or | |
1783 | * journaled data... */ | |
1784 | ||
1785 | if (journal->j_running_transaction) { | |
1786 | /* ... and once the current transaction has | |
1787 | * committed, the buffer won't be needed any | |
1788 | * longer. */ | |
1789 | JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget"); | |
1790 | ret = __dispose_buffer(jh, | |
1791 | journal->j_running_transaction); | |
f7f4bccb | 1792 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1793 | spin_unlock(&journal->j_list_lock); |
1794 | jbd_unlock_bh_state(bh); | |
1795 | spin_unlock(&journal->j_state_lock); | |
1796 | return ret; | |
1797 | } else { | |
1798 | /* There is no currently-running transaction. So the | |
1799 | * orphan record which we wrote for this file must have | |
1800 | * passed into commit. We must attach this buffer to | |
1801 | * the committing transaction, if it exists. */ | |
1802 | if (journal->j_committing_transaction) { | |
1803 | JBUFFER_TRACE(jh, "give to committing trans"); | |
1804 | ret = __dispose_buffer(jh, | |
1805 | journal->j_committing_transaction); | |
f7f4bccb | 1806 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1807 | spin_unlock(&journal->j_list_lock); |
1808 | jbd_unlock_bh_state(bh); | |
1809 | spin_unlock(&journal->j_state_lock); | |
1810 | return ret; | |
1811 | } else { | |
1812 | /* The orphan record's transaction has | |
1813 | * committed. We can cleanse this buffer */ | |
1814 | clear_buffer_jbddirty(bh); | |
1815 | goto zap_buffer; | |
1816 | } | |
1817 | } | |
1818 | } else if (transaction == journal->j_committing_transaction) { | |
9b57988d | 1819 | JBUFFER_TRACE(jh, "on committing transaction"); |
470decc6 | 1820 | /* |
ba869023 | 1821 | * The buffer is committing, we simply cannot touch |
1822 | * it. So we just set j_next_transaction to the | |
1823 | * running transaction (if there is one) and mark | |
1824 | * buffer as freed so that commit code knows it should | |
1825 | * clear dirty bits when it is done with the buffer. | |
1826 | */ | |
470decc6 | 1827 | set_buffer_freed(bh); |
ba869023 | 1828 | if (journal->j_running_transaction && buffer_jbddirty(bh)) |
1829 | jh->b_next_transaction = journal->j_running_transaction; | |
f7f4bccb | 1830 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1831 | spin_unlock(&journal->j_list_lock); |
1832 | jbd_unlock_bh_state(bh); | |
1833 | spin_unlock(&journal->j_state_lock); | |
1834 | return 0; | |
1835 | } else { | |
1836 | /* Good, the buffer belongs to the running transaction. | |
1837 | * We are writing our own transaction's data, not any | |
1838 | * previous one's, so it is safe to throw it away | |
1839 | * (remember that we expect the filesystem to have set | |
1840 | * i_size already for this truncate so recovery will not | |
1841 | * expose the disk blocks we are discarding here.) */ | |
1842 | J_ASSERT_JH(jh, transaction == journal->j_running_transaction); | |
9b57988d | 1843 | JBUFFER_TRACE(jh, "on running transaction"); |
470decc6 DK |
1844 | may_free = __dispose_buffer(jh, transaction); |
1845 | } | |
1846 | ||
1847 | zap_buffer: | |
f7f4bccb | 1848 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1849 | zap_buffer_no_jh: |
1850 | spin_unlock(&journal->j_list_lock); | |
1851 | jbd_unlock_bh_state(bh); | |
1852 | spin_unlock(&journal->j_state_lock); | |
1853 | zap_buffer_unlocked: | |
1854 | clear_buffer_dirty(bh); | |
1855 | J_ASSERT_BH(bh, !buffer_jbddirty(bh)); | |
1856 | clear_buffer_mapped(bh); | |
1857 | clear_buffer_req(bh); | |
1858 | clear_buffer_new(bh); | |
1859 | bh->b_bdev = NULL; | |
1860 | return may_free; | |
1861 | } | |
1862 | ||
1863 | /** | |
f7f4bccb | 1864 | * void jbd2_journal_invalidatepage() |
470decc6 DK |
1865 | * @journal: journal to use for flush... |
1866 | * @page: page to flush | |
1867 | * @offset: length of page to invalidate. | |
1868 | * | |
1869 | * Reap page buffers containing data after offset in page. | |
1870 | * | |
1871 | */ | |
f7f4bccb | 1872 | void jbd2_journal_invalidatepage(journal_t *journal, |
470decc6 DK |
1873 | struct page *page, |
1874 | unsigned long offset) | |
1875 | { | |
1876 | struct buffer_head *head, *bh, *next; | |
1877 | unsigned int curr_off = 0; | |
1878 | int may_free = 1; | |
1879 | ||
1880 | if (!PageLocked(page)) | |
1881 | BUG(); | |
1882 | if (!page_has_buffers(page)) | |
1883 | return; | |
1884 | ||
1885 | /* We will potentially be playing with lists other than just the | |
1886 | * data lists (especially for journaled data mode), so be | |
1887 | * cautious in our locking. */ | |
1888 | ||
1889 | head = bh = page_buffers(page); | |
1890 | do { | |
1891 | unsigned int next_off = curr_off + bh->b_size; | |
1892 | next = bh->b_this_page; | |
1893 | ||
1894 | if (offset <= curr_off) { | |
1895 | /* This block is wholly outside the truncation point */ | |
1896 | lock_buffer(bh); | |
1897 | may_free &= journal_unmap_buffer(journal, bh); | |
1898 | unlock_buffer(bh); | |
1899 | } | |
1900 | curr_off = next_off; | |
1901 | bh = next; | |
1902 | ||
1903 | } while (bh != head); | |
1904 | ||
1905 | if (!offset) { | |
1906 | if (may_free && try_to_free_buffers(page)) | |
1907 | J_ASSERT(!page_has_buffers(page)); | |
1908 | } | |
1909 | } | |
1910 | ||
1911 | /* | |
1912 | * File a buffer on the given transaction list. | |
1913 | */ | |
f7f4bccb | 1914 | void __jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
1915 | transaction_t *transaction, int jlist) |
1916 | { | |
1917 | struct journal_head **list = NULL; | |
1918 | int was_dirty = 0; | |
1919 | struct buffer_head *bh = jh2bh(jh); | |
1920 | ||
1921 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
1922 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
1923 | ||
1924 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
1925 | J_ASSERT_JH(jh, jh->b_transaction == transaction || | |
4019191b | 1926 | jh->b_transaction == NULL); |
470decc6 DK |
1927 | |
1928 | if (jh->b_transaction && jh->b_jlist == jlist) | |
1929 | return; | |
1930 | ||
470decc6 DK |
1931 | if (jlist == BJ_Metadata || jlist == BJ_Reserved || |
1932 | jlist == BJ_Shadow || jlist == BJ_Forget) { | |
f91d1d04 JK |
1933 | /* |
1934 | * For metadata buffers, we track dirty bit in buffer_jbddirty | |
1935 | * instead of buffer_dirty. We should not see a dirty bit set | |
1936 | * here because we clear it in do_get_write_access but e.g. | |
1937 | * tune2fs can modify the sb and set the dirty bit at any time | |
1938 | * so we try to gracefully handle that. | |
1939 | */ | |
1940 | if (buffer_dirty(bh)) | |
1941 | warn_dirty_buffer(bh); | |
470decc6 DK |
1942 | if (test_clear_buffer_dirty(bh) || |
1943 | test_clear_buffer_jbddirty(bh)) | |
1944 | was_dirty = 1; | |
1945 | } | |
1946 | ||
1947 | if (jh->b_transaction) | |
f7f4bccb | 1948 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 DK |
1949 | jh->b_transaction = transaction; |
1950 | ||
1951 | switch (jlist) { | |
1952 | case BJ_None: | |
1953 | J_ASSERT_JH(jh, !jh->b_committed_data); | |
1954 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
1955 | return; | |
470decc6 DK |
1956 | case BJ_Metadata: |
1957 | transaction->t_nr_buffers++; | |
1958 | list = &transaction->t_buffers; | |
1959 | break; | |
1960 | case BJ_Forget: | |
1961 | list = &transaction->t_forget; | |
1962 | break; | |
1963 | case BJ_IO: | |
1964 | list = &transaction->t_iobuf_list; | |
1965 | break; | |
1966 | case BJ_Shadow: | |
1967 | list = &transaction->t_shadow_list; | |
1968 | break; | |
1969 | case BJ_LogCtl: | |
1970 | list = &transaction->t_log_list; | |
1971 | break; | |
1972 | case BJ_Reserved: | |
1973 | list = &transaction->t_reserved_list; | |
1974 | break; | |
470decc6 DK |
1975 | } |
1976 | ||
1977 | __blist_add_buffer(list, jh); | |
1978 | jh->b_jlist = jlist; | |
1979 | ||
1980 | if (was_dirty) | |
1981 | set_buffer_jbddirty(bh); | |
1982 | } | |
1983 | ||
f7f4bccb | 1984 | void jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
1985 | transaction_t *transaction, int jlist) |
1986 | { | |
1987 | jbd_lock_bh_state(jh2bh(jh)); | |
1988 | spin_lock(&transaction->t_journal->j_list_lock); | |
f7f4bccb | 1989 | __jbd2_journal_file_buffer(jh, transaction, jlist); |
470decc6 DK |
1990 | spin_unlock(&transaction->t_journal->j_list_lock); |
1991 | jbd_unlock_bh_state(jh2bh(jh)); | |
1992 | } | |
1993 | ||
1994 | /* | |
1995 | * Remove a buffer from its current buffer list in preparation for | |
1996 | * dropping it from its current transaction entirely. If the buffer has | |
1997 | * already started to be used by a subsequent transaction, refile the | |
1998 | * buffer on that transaction's metadata list. | |
1999 | * | |
2000 | * Called under journal->j_list_lock | |
2001 | * | |
2002 | * Called under jbd_lock_bh_state(jh2bh(jh)) | |
2003 | */ | |
f7f4bccb | 2004 | void __jbd2_journal_refile_buffer(struct journal_head *jh) |
470decc6 | 2005 | { |
ba869023 | 2006 | int was_dirty, jlist; |
470decc6 DK |
2007 | struct buffer_head *bh = jh2bh(jh); |
2008 | ||
2009 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
2010 | if (jh->b_transaction) | |
2011 | assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock); | |
2012 | ||
2013 | /* If the buffer is now unused, just drop it. */ | |
2014 | if (jh->b_next_transaction == NULL) { | |
f7f4bccb | 2015 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
2016 | return; |
2017 | } | |
2018 | ||
2019 | /* | |
2020 | * It has been modified by a later transaction: add it to the new | |
2021 | * transaction's metadata list. | |
2022 | */ | |
2023 | ||
2024 | was_dirty = test_clear_buffer_jbddirty(bh); | |
f7f4bccb | 2025 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 DK |
2026 | jh->b_transaction = jh->b_next_transaction; |
2027 | jh->b_next_transaction = NULL; | |
ba869023 | 2028 | if (buffer_freed(bh)) |
2029 | jlist = BJ_Forget; | |
2030 | else if (jh->b_modified) | |
2031 | jlist = BJ_Metadata; | |
2032 | else | |
2033 | jlist = BJ_Reserved; | |
2034 | __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist); | |
470decc6 DK |
2035 | J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING); |
2036 | ||
2037 | if (was_dirty) | |
2038 | set_buffer_jbddirty(bh); | |
2039 | } | |
2040 | ||
2041 | /* | |
2042 | * For the unlocked version of this call, also make sure that any | |
2043 | * hanging journal_head is cleaned up if necessary. | |
2044 | * | |
f7f4bccb | 2045 | * __jbd2_journal_refile_buffer is usually called as part of a single locked |
470decc6 DK |
2046 | * operation on a buffer_head, in which the caller is probably going to |
2047 | * be hooking the journal_head onto other lists. In that case it is up | |
2048 | * to the caller to remove the journal_head if necessary. For the | |
f7f4bccb | 2049 | * unlocked jbd2_journal_refile_buffer call, the caller isn't going to be |
470decc6 DK |
2050 | * doing anything else to the buffer so we need to do the cleanup |
2051 | * ourselves to avoid a jh leak. | |
2052 | * | |
2053 | * *** The journal_head may be freed by this call! *** | |
2054 | */ | |
f7f4bccb | 2055 | void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 DK |
2056 | { |
2057 | struct buffer_head *bh = jh2bh(jh); | |
2058 | ||
2059 | jbd_lock_bh_state(bh); | |
2060 | spin_lock(&journal->j_list_lock); | |
2061 | ||
f7f4bccb | 2062 | __jbd2_journal_refile_buffer(jh); |
470decc6 | 2063 | jbd_unlock_bh_state(bh); |
f7f4bccb | 2064 | jbd2_journal_remove_journal_head(bh); |
470decc6 DK |
2065 | |
2066 | spin_unlock(&journal->j_list_lock); | |
2067 | __brelse(bh); | |
2068 | } | |
c851ed54 JK |
2069 | |
2070 | /* | |
2071 | * File inode in the inode list of the handle's transaction | |
2072 | */ | |
2073 | int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode) | |
2074 | { | |
2075 | transaction_t *transaction = handle->h_transaction; | |
2076 | journal_t *journal = transaction->t_journal; | |
2077 | ||
2078 | if (is_handle_aborted(handle)) | |
2079 | return -EIO; | |
2080 | ||
2081 | jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino, | |
2082 | transaction->t_tid); | |
2083 | ||
2084 | /* | |
2085 | * First check whether inode isn't already on the transaction's | |
2086 | * lists without taking the lock. Note that this check is safe | |
2087 | * without the lock as we cannot race with somebody removing inode | |
2088 | * from the transaction. The reason is that we remove inode from the | |
2089 | * transaction only in journal_release_jbd_inode() and when we commit | |
2090 | * the transaction. We are guarded from the first case by holding | |
2091 | * a reference to the inode. We are safe against the second case | |
2092 | * because if jinode->i_transaction == transaction, commit code | |
2093 | * cannot touch the transaction because we hold reference to it, | |
2094 | * and if jinode->i_next_transaction == transaction, commit code | |
2095 | * will only file the inode where we want it. | |
2096 | */ | |
2097 | if (jinode->i_transaction == transaction || | |
2098 | jinode->i_next_transaction == transaction) | |
2099 | return 0; | |
2100 | ||
2101 | spin_lock(&journal->j_list_lock); | |
2102 | ||
2103 | if (jinode->i_transaction == transaction || | |
2104 | jinode->i_next_transaction == transaction) | |
2105 | goto done; | |
2106 | ||
2107 | /* On some different transaction's list - should be | |
2108 | * the committing one */ | |
2109 | if (jinode->i_transaction) { | |
2110 | J_ASSERT(jinode->i_next_transaction == NULL); | |
2111 | J_ASSERT(jinode->i_transaction == | |
2112 | journal->j_committing_transaction); | |
2113 | jinode->i_next_transaction = transaction; | |
2114 | goto done; | |
2115 | } | |
2116 | /* Not on any transaction list... */ | |
2117 | J_ASSERT(!jinode->i_next_transaction); | |
2118 | jinode->i_transaction = transaction; | |
2119 | list_add(&jinode->i_list, &transaction->t_inode_list); | |
2120 | done: | |
2121 | spin_unlock(&journal->j_list_lock); | |
2122 | ||
2123 | return 0; | |
2124 | } | |
2125 | ||
2126 | /* | |
7f5aa215 JK |
2127 | * File truncate and transaction commit interact with each other in a |
2128 | * non-trivial way. If a transaction writing data block A is | |
2129 | * committing, we cannot discard the data by truncate until we have | |
2130 | * written them. Otherwise if we crashed after the transaction with | |
2131 | * write has committed but before the transaction with truncate has | |
2132 | * committed, we could see stale data in block A. This function is a | |
2133 | * helper to solve this problem. It starts writeout of the truncated | |
2134 | * part in case it is in the committing transaction. | |
2135 | * | |
2136 | * Filesystem code must call this function when inode is journaled in | |
2137 | * ordered mode before truncation happens and after the inode has been | |
2138 | * placed on orphan list with the new inode size. The second condition | |
2139 | * avoids the race that someone writes new data and we start | |
2140 | * committing the transaction after this function has been called but | |
2141 | * before a transaction for truncate is started (and furthermore it | |
2142 | * allows us to optimize the case where the addition to orphan list | |
2143 | * happens in the same transaction as write --- we don't have to write | |
2144 | * any data in such case). | |
c851ed54 | 2145 | */ |
7f5aa215 JK |
2146 | int jbd2_journal_begin_ordered_truncate(journal_t *journal, |
2147 | struct jbd2_inode *jinode, | |
c851ed54 JK |
2148 | loff_t new_size) |
2149 | { | |
7f5aa215 | 2150 | transaction_t *inode_trans, *commit_trans; |
c851ed54 JK |
2151 | int ret = 0; |
2152 | ||
7f5aa215 JK |
2153 | /* This is a quick check to avoid locking if not necessary */ |
2154 | if (!jinode->i_transaction) | |
c851ed54 | 2155 | goto out; |
7f5aa215 JK |
2156 | /* Locks are here just to force reading of recent values, it is |
2157 | * enough that the transaction was not committing before we started | |
2158 | * a transaction adding the inode to orphan list */ | |
c851ed54 JK |
2159 | spin_lock(&journal->j_state_lock); |
2160 | commit_trans = journal->j_committing_transaction; | |
2161 | spin_unlock(&journal->j_state_lock); | |
7f5aa215 JK |
2162 | spin_lock(&journal->j_list_lock); |
2163 | inode_trans = jinode->i_transaction; | |
2164 | spin_unlock(&journal->j_list_lock); | |
2165 | if (inode_trans == commit_trans) { | |
2166 | ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping, | |
c851ed54 JK |
2167 | new_size, LLONG_MAX); |
2168 | if (ret) | |
2169 | jbd2_journal_abort(journal, ret); | |
2170 | } | |
2171 | out: | |
2172 | return ret; | |
2173 | } |