]>
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); | |
a51dca9c TT |
58 | atomic_set(&transaction->t_updates, 0); |
59 | atomic_set(&transaction->t_outstanding_credits, 0); | |
c851ed54 | 60 | INIT_LIST_HEAD(&transaction->t_inode_list); |
3e624fc7 | 61 | INIT_LIST_HEAD(&transaction->t_private_list); |
470decc6 DK |
62 | |
63 | /* Set up the commit timer for the new transaction. */ | |
b1f485f2 | 64 | journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires); |
470decc6 DK |
65 | add_timer(&journal->j_commit_timer); |
66 | ||
67 | J_ASSERT(journal->j_running_transaction == NULL); | |
68 | journal->j_running_transaction = transaction; | |
8e85fb3f JL |
69 | transaction->t_max_wait = 0; |
70 | transaction->t_start = jiffies; | |
470decc6 DK |
71 | |
72 | return transaction; | |
73 | } | |
74 | ||
75 | /* | |
76 | * Handle management. | |
77 | * | |
78 | * A handle_t is an object which represents a single atomic update to a | |
79 | * filesystem, and which tracks all of the modifications which form part | |
80 | * of that one update. | |
81 | */ | |
82 | ||
83 | /* | |
84 | * start_this_handle: Given a handle, deal with any locking or stalling | |
85 | * needed to make sure that there is enough journal space for the handle | |
86 | * to begin. Attach the handle to a transaction and set up the | |
87 | * transaction's buffer credits. | |
88 | */ | |
89 | ||
47def826 TT |
90 | static int start_this_handle(journal_t *journal, handle_t *handle, |
91 | int gfp_mask) | |
470decc6 DK |
92 | { |
93 | transaction_t *transaction; | |
94 | int needed; | |
95 | int nblocks = handle->h_buffer_credits; | |
96 | transaction_t *new_transaction = NULL; | |
8e85fb3f | 97 | unsigned long ts = jiffies; |
470decc6 DK |
98 | |
99 | if (nblocks > journal->j_max_transaction_buffers) { | |
100 | printk(KERN_ERR "JBD: %s wants too many credits (%d > %d)\n", | |
101 | current->comm, nblocks, | |
102 | journal->j_max_transaction_buffers); | |
47def826 | 103 | return -ENOSPC; |
470decc6 DK |
104 | } |
105 | ||
106 | alloc_transaction: | |
107 | if (!journal->j_running_transaction) { | |
47def826 | 108 | new_transaction = kzalloc(sizeof(*new_transaction), gfp_mask); |
470decc6 | 109 | if (!new_transaction) { |
47def826 TT |
110 | /* |
111 | * If __GFP_FS is not present, then we may be | |
112 | * being called from inside the fs writeback | |
113 | * layer, so we MUST NOT fail. Since | |
114 | * __GFP_NOFAIL is going away, we will arrange | |
115 | * to retry the allocation ourselves. | |
116 | */ | |
117 | if ((gfp_mask & __GFP_FS) == 0) { | |
118 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
119 | goto alloc_transaction; | |
120 | } | |
121 | return -ENOMEM; | |
470decc6 | 122 | } |
470decc6 DK |
123 | } |
124 | ||
125 | jbd_debug(3, "New handle %p going live.\n", handle); | |
126 | ||
470decc6 DK |
127 | /* |
128 | * We need to hold j_state_lock until t_updates has been incremented, | |
129 | * for proper journal barrier handling | |
130 | */ | |
a931da6a TT |
131 | repeat: |
132 | read_lock(&journal->j_state_lock); | |
470decc6 | 133 | if (is_journal_aborted(journal) || |
f7f4bccb | 134 | (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) { |
a931da6a | 135 | read_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) { | |
a931da6a | 142 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
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) { | |
a931da6a TT |
149 | read_unlock(&journal->j_state_lock); |
150 | if (!new_transaction) | |
470decc6 | 151 | goto alloc_transaction; |
a931da6a TT |
152 | write_lock(&journal->j_state_lock); |
153 | if (!journal->j_running_transaction) { | |
154 | jbd2_get_transaction(journal, new_transaction); | |
155 | new_transaction = NULL; | |
470decc6 | 156 | } |
a931da6a TT |
157 | write_unlock(&journal->j_state_lock); |
158 | goto repeat; | |
470decc6 DK |
159 | } |
160 | ||
161 | transaction = journal->j_running_transaction; | |
162 | ||
163 | /* | |
164 | * If the current transaction is locked down for commit, wait for the | |
165 | * lock to be released. | |
166 | */ | |
167 | if (transaction->t_state == T_LOCKED) { | |
168 | DEFINE_WAIT(wait); | |
169 | ||
170 | prepare_to_wait(&journal->j_wait_transaction_locked, | |
171 | &wait, TASK_UNINTERRUPTIBLE); | |
a931da6a | 172 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
173 | schedule(); |
174 | finish_wait(&journal->j_wait_transaction_locked, &wait); | |
175 | goto repeat; | |
176 | } | |
177 | ||
178 | /* | |
179 | * If there is not enough space left in the log to write all potential | |
180 | * buffers requested by this operation, we need to stall pending a log | |
181 | * checkpoint to free some more log space. | |
182 | */ | |
183 | spin_lock(&transaction->t_handle_lock); | |
a51dca9c | 184 | needed = atomic_read(&transaction->t_outstanding_credits) + nblocks; |
470decc6 DK |
185 | |
186 | if (needed > journal->j_max_transaction_buffers) { | |
187 | /* | |
188 | * If the current transaction is already too large, then start | |
189 | * to commit it: we can then go back and attach this handle to | |
190 | * a new transaction. | |
191 | */ | |
192 | DEFINE_WAIT(wait); | |
193 | ||
194 | jbd_debug(2, "Handle %p starting new commit...\n", handle); | |
195 | spin_unlock(&transaction->t_handle_lock); | |
196 | prepare_to_wait(&journal->j_wait_transaction_locked, &wait, | |
197 | TASK_UNINTERRUPTIBLE); | |
f7f4bccb | 198 | __jbd2_log_start_commit(journal, transaction->t_tid); |
a931da6a | 199 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
200 | schedule(); |
201 | finish_wait(&journal->j_wait_transaction_locked, &wait); | |
202 | goto repeat; | |
203 | } | |
204 | ||
205 | /* | |
206 | * The commit code assumes that it can get enough log space | |
207 | * without forcing a checkpoint. This is *critical* for | |
208 | * correctness: a checkpoint of a buffer which is also | |
209 | * associated with a committing transaction creates a deadlock, | |
210 | * so commit simply cannot force through checkpoints. | |
211 | * | |
212 | * We must therefore ensure the necessary space in the journal | |
213 | * *before* starting to dirty potentially checkpointed buffers | |
214 | * in the new transaction. | |
215 | * | |
216 | * The worst part is, any transaction currently committing can | |
217 | * reduce the free space arbitrarily. Be careful to account for | |
218 | * those buffers when checkpointing. | |
219 | */ | |
220 | ||
221 | /* | |
222 | * @@@ AKPM: This seems rather over-defensive. We're giving commit | |
223 | * a _lot_ of headroom: 1/4 of the journal plus the size of | |
224 | * the committing transaction. Really, we only need to give it | |
225 | * committing_transaction->t_outstanding_credits plus "enough" for | |
226 | * the log control blocks. | |
227 | * Also, this test is inconsitent with the matching one in | |
f7f4bccb | 228 | * jbd2_journal_extend(). |
470decc6 | 229 | */ |
f7f4bccb | 230 | if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) { |
470decc6 DK |
231 | jbd_debug(2, "Handle %p waiting for checkpoint...\n", handle); |
232 | spin_unlock(&transaction->t_handle_lock); | |
a931da6a TT |
233 | read_unlock(&journal->j_state_lock); |
234 | write_lock(&journal->j_state_lock); | |
235 | if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) | |
236 | __jbd2_log_wait_for_space(journal); | |
237 | write_unlock(&journal->j_state_lock); | |
238 | goto repeat; | |
470decc6 DK |
239 | } |
240 | ||
241 | /* OK, account for the buffers that this operation expects to | |
242 | * use and add the handle to the running transaction. */ | |
243 | ||
8e85fb3f JL |
244 | if (time_after(transaction->t_start, ts)) { |
245 | ts = jbd2_time_diff(ts, transaction->t_start); | |
246 | if (ts > transaction->t_max_wait) | |
247 | transaction->t_max_wait = ts; | |
248 | } | |
249 | ||
470decc6 | 250 | handle->h_transaction = transaction; |
a51dca9c TT |
251 | atomic_add(nblocks, &transaction->t_outstanding_credits); |
252 | atomic_inc(&transaction->t_updates); | |
470decc6 DK |
253 | transaction->t_handle_count++; |
254 | jbd_debug(4, "Handle %p given %d credits (total %d, free %d)\n", | |
a51dca9c TT |
255 | handle, nblocks, |
256 | atomic_read(&transaction->t_outstanding_credits), | |
f7f4bccb | 257 | __jbd2_log_space_left(journal)); |
470decc6 | 258 | spin_unlock(&transaction->t_handle_lock); |
a931da6a | 259 | read_unlock(&journal->j_state_lock); |
9599b0e5 JK |
260 | |
261 | lock_map_acquire(&handle->h_lockdep_map); | |
47def826 TT |
262 | kfree(new_transaction); |
263 | return 0; | |
470decc6 DK |
264 | } |
265 | ||
7b751066 MC |
266 | static struct lock_class_key jbd2_handle_key; |
267 | ||
470decc6 DK |
268 | /* Allocate a new handle. This should probably be in a slab... */ |
269 | static handle_t *new_handle(int nblocks) | |
270 | { | |
af1e76d6 | 271 | handle_t *handle = jbd2_alloc_handle(GFP_NOFS); |
470decc6 DK |
272 | if (!handle) |
273 | return NULL; | |
274 | memset(handle, 0, sizeof(*handle)); | |
275 | handle->h_buffer_credits = nblocks; | |
276 | handle->h_ref = 1; | |
277 | ||
7b751066 MC |
278 | lockdep_init_map(&handle->h_lockdep_map, "jbd2_handle", |
279 | &jbd2_handle_key, 0); | |
280 | ||
470decc6 DK |
281 | return handle; |
282 | } | |
283 | ||
284 | /** | |
f7f4bccb | 285 | * handle_t *jbd2_journal_start() - Obtain a new handle. |
470decc6 DK |
286 | * @journal: Journal to start transaction on. |
287 | * @nblocks: number of block buffer we might modify | |
288 | * | |
289 | * We make sure that the transaction can guarantee at least nblocks of | |
290 | * modified buffers in the log. We block until the log can guarantee | |
291 | * that much space. | |
292 | * | |
293 | * This function is visible to journal users (like ext3fs), so is not | |
294 | * called with the journal already locked. | |
295 | * | |
296 | * Return a pointer to a newly allocated handle, or NULL on failure | |
297 | */ | |
47def826 | 298 | handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int gfp_mask) |
470decc6 DK |
299 | { |
300 | handle_t *handle = journal_current_handle(); | |
301 | int err; | |
302 | ||
303 | if (!journal) | |
304 | return ERR_PTR(-EROFS); | |
305 | ||
306 | if (handle) { | |
307 | J_ASSERT(handle->h_transaction->t_journal == journal); | |
308 | handle->h_ref++; | |
309 | return handle; | |
310 | } | |
311 | ||
312 | handle = new_handle(nblocks); | |
313 | if (!handle) | |
314 | return ERR_PTR(-ENOMEM); | |
315 | ||
316 | current->journal_info = handle; | |
317 | ||
47def826 | 318 | err = start_this_handle(journal, handle, gfp_mask); |
470decc6 | 319 | if (err < 0) { |
af1e76d6 | 320 | jbd2_free_handle(handle); |
470decc6 DK |
321 | current->journal_info = NULL; |
322 | handle = ERR_PTR(err); | |
7b751066 | 323 | goto out; |
470decc6 | 324 | } |
7b751066 | 325 | out: |
470decc6 DK |
326 | return handle; |
327 | } | |
47def826 TT |
328 | EXPORT_SYMBOL(jbd2__journal_start); |
329 | ||
330 | ||
331 | handle_t *jbd2_journal_start(journal_t *journal, int nblocks) | |
332 | { | |
333 | return jbd2__journal_start(journal, nblocks, GFP_NOFS); | |
334 | } | |
335 | EXPORT_SYMBOL(jbd2_journal_start); | |
336 | ||
470decc6 DK |
337 | |
338 | /** | |
f7f4bccb | 339 | * int jbd2_journal_extend() - extend buffer credits. |
470decc6 DK |
340 | * @handle: handle to 'extend' |
341 | * @nblocks: nr blocks to try to extend by. | |
342 | * | |
343 | * Some transactions, such as large extends and truncates, can be done | |
344 | * atomically all at once or in several stages. The operation requests | |
345 | * a credit for a number of buffer modications in advance, but can | |
346 | * extend its credit if it needs more. | |
347 | * | |
f7f4bccb | 348 | * jbd2_journal_extend tries to give the running handle more buffer credits. |
470decc6 DK |
349 | * It does not guarantee that allocation - this is a best-effort only. |
350 | * The calling process MUST be able to deal cleanly with a failure to | |
351 | * extend here. | |
352 | * | |
353 | * Return 0 on success, non-zero on failure. | |
354 | * | |
355 | * return code < 0 implies an error | |
356 | * return code > 0 implies normal transaction-full status. | |
357 | */ | |
f7f4bccb | 358 | int jbd2_journal_extend(handle_t *handle, int nblocks) |
470decc6 DK |
359 | { |
360 | transaction_t *transaction = handle->h_transaction; | |
361 | journal_t *journal = transaction->t_journal; | |
362 | int result; | |
363 | int wanted; | |
364 | ||
365 | result = -EIO; | |
366 | if (is_handle_aborted(handle)) | |
367 | goto out; | |
368 | ||
369 | result = 1; | |
370 | ||
a931da6a | 371 | read_lock(&journal->j_state_lock); |
470decc6 DK |
372 | |
373 | /* Don't extend a locked-down transaction! */ | |
374 | if (handle->h_transaction->t_state != T_RUNNING) { | |
375 | jbd_debug(3, "denied handle %p %d blocks: " | |
376 | "transaction not running\n", handle, nblocks); | |
377 | goto error_out; | |
378 | } | |
379 | ||
380 | spin_lock(&transaction->t_handle_lock); | |
a51dca9c | 381 | wanted = atomic_read(&transaction->t_outstanding_credits) + nblocks; |
470decc6 DK |
382 | |
383 | if (wanted > journal->j_max_transaction_buffers) { | |
384 | jbd_debug(3, "denied handle %p %d blocks: " | |
385 | "transaction too large\n", handle, nblocks); | |
386 | goto unlock; | |
387 | } | |
388 | ||
f7f4bccb | 389 | if (wanted > __jbd2_log_space_left(journal)) { |
470decc6 DK |
390 | jbd_debug(3, "denied handle %p %d blocks: " |
391 | "insufficient log space\n", handle, nblocks); | |
392 | goto unlock; | |
393 | } | |
394 | ||
395 | handle->h_buffer_credits += nblocks; | |
a51dca9c | 396 | atomic_add(nblocks, &transaction->t_outstanding_credits); |
470decc6 DK |
397 | result = 0; |
398 | ||
399 | jbd_debug(3, "extended handle %p by %d\n", handle, nblocks); | |
400 | unlock: | |
401 | spin_unlock(&transaction->t_handle_lock); | |
402 | error_out: | |
a931da6a | 403 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
404 | out: |
405 | return result; | |
406 | } | |
407 | ||
408 | ||
409 | /** | |
f7f4bccb | 410 | * int jbd2_journal_restart() - restart a handle . |
470decc6 DK |
411 | * @handle: handle to restart |
412 | * @nblocks: nr credits requested | |
413 | * | |
414 | * Restart a handle for a multi-transaction filesystem | |
415 | * operation. | |
416 | * | |
f7f4bccb MC |
417 | * If the jbd2_journal_extend() call above fails to grant new buffer credits |
418 | * to a running handle, a call to jbd2_journal_restart will commit the | |
470decc6 DK |
419 | * handle's transaction so far and reattach the handle to a new |
420 | * transaction capabable of guaranteeing the requested number of | |
421 | * credits. | |
422 | */ | |
47def826 | 423 | int jbd2__journal_restart(handle_t *handle, int nblocks, int gfp_mask) |
470decc6 DK |
424 | { |
425 | transaction_t *transaction = handle->h_transaction; | |
426 | journal_t *journal = transaction->t_journal; | |
427 | int ret; | |
428 | ||
429 | /* If we've had an abort of any type, don't even think about | |
430 | * actually doing the restart! */ | |
431 | if (is_handle_aborted(handle)) | |
432 | return 0; | |
433 | ||
434 | /* | |
435 | * First unlink the handle from its current transaction, and start the | |
436 | * commit on that. | |
437 | */ | |
a51dca9c | 438 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 DK |
439 | J_ASSERT(journal_current_handle() == handle); |
440 | ||
a931da6a | 441 | read_lock(&journal->j_state_lock); |
470decc6 | 442 | spin_lock(&transaction->t_handle_lock); |
a51dca9c TT |
443 | atomic_sub(handle->h_buffer_credits, |
444 | &transaction->t_outstanding_credits); | |
445 | if (atomic_dec_and_test(&transaction->t_updates)) | |
470decc6 DK |
446 | wake_up(&journal->j_wait_updates); |
447 | spin_unlock(&transaction->t_handle_lock); | |
448 | ||
449 | jbd_debug(2, "restarting handle %p\n", handle); | |
f7f4bccb | 450 | __jbd2_log_start_commit(journal, transaction->t_tid); |
a931da6a | 451 | read_unlock(&journal->j_state_lock); |
470decc6 | 452 | |
9599b0e5 | 453 | lock_map_release(&handle->h_lockdep_map); |
470decc6 | 454 | handle->h_buffer_credits = nblocks; |
47def826 | 455 | ret = start_this_handle(journal, handle, gfp_mask); |
470decc6 DK |
456 | return ret; |
457 | } | |
47def826 | 458 | EXPORT_SYMBOL(jbd2__journal_restart); |
470decc6 DK |
459 | |
460 | ||
47def826 TT |
461 | int jbd2_journal_restart(handle_t *handle, int nblocks) |
462 | { | |
463 | return jbd2__journal_restart(handle, nblocks, GFP_NOFS); | |
464 | } | |
465 | EXPORT_SYMBOL(jbd2_journal_restart); | |
466 | ||
470decc6 | 467 | /** |
f7f4bccb | 468 | * void jbd2_journal_lock_updates () - establish a transaction barrier. |
470decc6 DK |
469 | * @journal: Journal to establish a barrier on. |
470 | * | |
471 | * This locks out any further updates from being started, and blocks | |
472 | * until all existing updates have completed, returning only once the | |
473 | * journal is in a quiescent state with no updates running. | |
474 | * | |
475 | * The journal lock should not be held on entry. | |
476 | */ | |
f7f4bccb | 477 | void jbd2_journal_lock_updates(journal_t *journal) |
470decc6 DK |
478 | { |
479 | DEFINE_WAIT(wait); | |
480 | ||
a931da6a | 481 | write_lock(&journal->j_state_lock); |
470decc6 DK |
482 | ++journal->j_barrier_count; |
483 | ||
484 | /* Wait until there are no running updates */ | |
485 | while (1) { | |
486 | transaction_t *transaction = journal->j_running_transaction; | |
487 | ||
488 | if (!transaction) | |
489 | break; | |
490 | ||
491 | spin_lock(&transaction->t_handle_lock); | |
a51dca9c | 492 | if (!atomic_read(&transaction->t_updates)) { |
470decc6 DK |
493 | spin_unlock(&transaction->t_handle_lock); |
494 | break; | |
495 | } | |
496 | prepare_to_wait(&journal->j_wait_updates, &wait, | |
497 | TASK_UNINTERRUPTIBLE); | |
498 | spin_unlock(&transaction->t_handle_lock); | |
a931da6a | 499 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
500 | schedule(); |
501 | finish_wait(&journal->j_wait_updates, &wait); | |
a931da6a | 502 | write_lock(&journal->j_state_lock); |
470decc6 | 503 | } |
a931da6a | 504 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
505 | |
506 | /* | |
507 | * We have now established a barrier against other normal updates, but | |
f7f4bccb | 508 | * we also need to barrier against other jbd2_journal_lock_updates() calls |
470decc6 DK |
509 | * to make sure that we serialise special journal-locked operations |
510 | * too. | |
511 | */ | |
512 | mutex_lock(&journal->j_barrier); | |
513 | } | |
514 | ||
515 | /** | |
f7f4bccb | 516 | * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier |
470decc6 DK |
517 | * @journal: Journal to release the barrier on. |
518 | * | |
f7f4bccb | 519 | * Release a transaction barrier obtained with jbd2_journal_lock_updates(). |
470decc6 DK |
520 | * |
521 | * Should be called without the journal lock held. | |
522 | */ | |
f7f4bccb | 523 | void jbd2_journal_unlock_updates (journal_t *journal) |
470decc6 DK |
524 | { |
525 | J_ASSERT(journal->j_barrier_count != 0); | |
526 | ||
527 | mutex_unlock(&journal->j_barrier); | |
a931da6a | 528 | write_lock(&journal->j_state_lock); |
470decc6 | 529 | --journal->j_barrier_count; |
a931da6a | 530 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
531 | wake_up(&journal->j_wait_transaction_locked); |
532 | } | |
533 | ||
f91d1d04 | 534 | static void warn_dirty_buffer(struct buffer_head *bh) |
470decc6 | 535 | { |
f91d1d04 | 536 | char b[BDEVNAME_SIZE]; |
470decc6 | 537 | |
f91d1d04 JK |
538 | printk(KERN_WARNING |
539 | "JBD: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). " | |
540 | "There's a risk of filesystem corruption in case of system " | |
541 | "crash.\n", | |
542 | bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr); | |
470decc6 DK |
543 | } |
544 | ||
545 | /* | |
546 | * If the buffer is already part of the current transaction, then there | |
547 | * is nothing we need to do. If it is already part of a prior | |
548 | * transaction which we are still committing to disk, then we need to | |
549 | * make sure that we do not overwrite the old copy: we do copy-out to | |
550 | * preserve the copy going to disk. We also account the buffer against | |
551 | * the handle's metadata buffer credits (unless the buffer is already | |
552 | * part of the transaction, that is). | |
553 | * | |
554 | */ | |
555 | static int | |
556 | do_get_write_access(handle_t *handle, struct journal_head *jh, | |
557 | int force_copy) | |
558 | { | |
559 | struct buffer_head *bh; | |
560 | transaction_t *transaction; | |
561 | journal_t *journal; | |
562 | int error; | |
563 | char *frozen_buffer = NULL; | |
564 | int need_copy = 0; | |
565 | ||
566 | if (is_handle_aborted(handle)) | |
567 | return -EROFS; | |
568 | ||
569 | transaction = handle->h_transaction; | |
570 | journal = transaction->t_journal; | |
571 | ||
572 | jbd_debug(5, "buffer_head %p, force_copy %d\n", jh, force_copy); | |
573 | ||
574 | JBUFFER_TRACE(jh, "entry"); | |
575 | repeat: | |
576 | bh = jh2bh(jh); | |
577 | ||
578 | /* @@@ Need to check for errors here at some point. */ | |
579 | ||
580 | lock_buffer(bh); | |
581 | jbd_lock_bh_state(bh); | |
582 | ||
583 | /* We now hold the buffer lock so it is safe to query the buffer | |
584 | * state. Is the buffer dirty? | |
585 | * | |
586 | * If so, there are two possibilities. The buffer may be | |
587 | * non-journaled, and undergoing a quite legitimate writeback. | |
588 | * Otherwise, it is journaled, and we don't expect dirty buffers | |
589 | * in that state (the buffers should be marked JBD_Dirty | |
590 | * instead.) So either the IO is being done under our own | |
591 | * control and this is a bug, or it's a third party IO such as | |
592 | * dump(8) (which may leave the buffer scheduled for read --- | |
593 | * ie. locked but not dirty) or tune2fs (which may actually have | |
594 | * the buffer dirtied, ugh.) */ | |
595 | ||
596 | if (buffer_dirty(bh)) { | |
597 | /* | |
598 | * First question: is this buffer already part of the current | |
599 | * transaction or the existing committing transaction? | |
600 | */ | |
601 | if (jh->b_transaction) { | |
602 | J_ASSERT_JH(jh, | |
603 | jh->b_transaction == transaction || | |
604 | jh->b_transaction == | |
605 | journal->j_committing_transaction); | |
606 | if (jh->b_next_transaction) | |
607 | J_ASSERT_JH(jh, jh->b_next_transaction == | |
608 | transaction); | |
f91d1d04 | 609 | warn_dirty_buffer(bh); |
470decc6 DK |
610 | } |
611 | /* | |
612 | * In any case we need to clean the dirty flag and we must | |
613 | * do it under the buffer lock to be sure we don't race | |
614 | * with running write-out. | |
615 | */ | |
f91d1d04 JK |
616 | JBUFFER_TRACE(jh, "Journalling dirty buffer"); |
617 | clear_buffer_dirty(bh); | |
618 | set_buffer_jbddirty(bh); | |
470decc6 DK |
619 | } |
620 | ||
621 | unlock_buffer(bh); | |
622 | ||
623 | error = -EROFS; | |
624 | if (is_handle_aborted(handle)) { | |
625 | jbd_unlock_bh_state(bh); | |
626 | goto out; | |
627 | } | |
628 | error = 0; | |
629 | ||
630 | /* | |
631 | * The buffer is already part of this transaction if b_transaction or | |
632 | * b_next_transaction points to it | |
633 | */ | |
634 | if (jh->b_transaction == transaction || | |
635 | jh->b_next_transaction == transaction) | |
636 | goto done; | |
637 | ||
9fc7c63a JB |
638 | /* |
639 | * this is the first time this transaction is touching this buffer, | |
640 | * reset the modified flag | |
641 | */ | |
642 | jh->b_modified = 0; | |
643 | ||
470decc6 DK |
644 | /* |
645 | * If there is already a copy-out version of this buffer, then we don't | |
646 | * need to make another one | |
647 | */ | |
648 | if (jh->b_frozen_data) { | |
649 | JBUFFER_TRACE(jh, "has frozen data"); | |
650 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
651 | jh->b_next_transaction = transaction; | |
652 | goto done; | |
653 | } | |
654 | ||
655 | /* Is there data here we need to preserve? */ | |
656 | ||
657 | if (jh->b_transaction && jh->b_transaction != transaction) { | |
658 | JBUFFER_TRACE(jh, "owned by older transaction"); | |
659 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
660 | J_ASSERT_JH(jh, jh->b_transaction == | |
661 | journal->j_committing_transaction); | |
662 | ||
663 | /* There is one case we have to be very careful about. | |
664 | * If the committing transaction is currently writing | |
665 | * this buffer out to disk and has NOT made a copy-out, | |
666 | * then we cannot modify the buffer contents at all | |
667 | * right now. The essence of copy-out is that it is the | |
668 | * extra copy, not the primary copy, which gets | |
669 | * journaled. If the primary copy is already going to | |
670 | * disk then we cannot do copy-out here. */ | |
671 | ||
672 | if (jh->b_jlist == BJ_Shadow) { | |
673 | DEFINE_WAIT_BIT(wait, &bh->b_state, BH_Unshadow); | |
674 | wait_queue_head_t *wqh; | |
675 | ||
676 | wqh = bit_waitqueue(&bh->b_state, BH_Unshadow); | |
677 | ||
678 | JBUFFER_TRACE(jh, "on shadow: sleep"); | |
679 | jbd_unlock_bh_state(bh); | |
680 | /* commit wakes up all shadow buffers after IO */ | |
681 | for ( ; ; ) { | |
682 | prepare_to_wait(wqh, &wait.wait, | |
683 | TASK_UNINTERRUPTIBLE); | |
684 | if (jh->b_jlist != BJ_Shadow) | |
685 | break; | |
686 | schedule(); | |
687 | } | |
688 | finish_wait(wqh, &wait.wait); | |
689 | goto repeat; | |
690 | } | |
691 | ||
692 | /* Only do the copy if the currently-owning transaction | |
693 | * still needs it. If it is on the Forget list, the | |
694 | * committing transaction is past that stage. The | |
695 | * buffer had better remain locked during the kmalloc, | |
696 | * but that should be true --- we hold the journal lock | |
697 | * still and the buffer is already on the BUF_JOURNAL | |
698 | * list so won't be flushed. | |
699 | * | |
700 | * Subtle point, though: if this is a get_undo_access, | |
701 | * then we will be relying on the frozen_data to contain | |
702 | * the new value of the committed_data record after the | |
703 | * transaction, so we HAVE to force the frozen_data copy | |
704 | * in that case. */ | |
705 | ||
706 | if (jh->b_jlist != BJ_Forget || force_copy) { | |
707 | JBUFFER_TRACE(jh, "generate frozen data"); | |
708 | if (!frozen_buffer) { | |
709 | JBUFFER_TRACE(jh, "allocate memory for buffer"); | |
710 | jbd_unlock_bh_state(bh); | |
711 | frozen_buffer = | |
af1e76d6 | 712 | jbd2_alloc(jh2bh(jh)->b_size, |
470decc6 DK |
713 | GFP_NOFS); |
714 | if (!frozen_buffer) { | |
715 | printk(KERN_EMERG | |
716 | "%s: OOM for frozen_buffer\n", | |
329d291f | 717 | __func__); |
470decc6 DK |
718 | JBUFFER_TRACE(jh, "oom!"); |
719 | error = -ENOMEM; | |
720 | jbd_lock_bh_state(bh); | |
721 | goto done; | |
722 | } | |
723 | goto repeat; | |
724 | } | |
725 | jh->b_frozen_data = frozen_buffer; | |
726 | frozen_buffer = NULL; | |
727 | need_copy = 1; | |
728 | } | |
729 | jh->b_next_transaction = transaction; | |
730 | } | |
731 | ||
732 | ||
733 | /* | |
734 | * Finally, if the buffer is not journaled right now, we need to make | |
735 | * sure it doesn't get written to disk before the caller actually | |
736 | * commits the new data | |
737 | */ | |
738 | if (!jh->b_transaction) { | |
739 | JBUFFER_TRACE(jh, "no transaction"); | |
740 | J_ASSERT_JH(jh, !jh->b_next_transaction); | |
741 | jh->b_transaction = transaction; | |
742 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); | |
743 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 744 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
470decc6 DK |
745 | spin_unlock(&journal->j_list_lock); |
746 | } | |
747 | ||
748 | done: | |
749 | if (need_copy) { | |
750 | struct page *page; | |
751 | int offset; | |
752 | char *source; | |
753 | ||
754 | J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)), | |
755 | "Possible IO failure.\n"); | |
756 | page = jh2bh(jh)->b_page; | |
757 | offset = ((unsigned long) jh2bh(jh)->b_data) & ~PAGE_MASK; | |
758 | source = kmap_atomic(page, KM_USER0); | |
759 | memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size); | |
760 | kunmap_atomic(source, KM_USER0); | |
e06c8227 JB |
761 | |
762 | /* | |
763 | * Now that the frozen data is saved off, we need to store | |
764 | * any matching triggers. | |
765 | */ | |
766 | jh->b_frozen_triggers = jh->b_triggers; | |
470decc6 DK |
767 | } |
768 | jbd_unlock_bh_state(bh); | |
769 | ||
770 | /* | |
771 | * If we are about to journal a buffer, then any revoke pending on it is | |
772 | * no longer valid | |
773 | */ | |
f7f4bccb | 774 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 DK |
775 | |
776 | out: | |
777 | if (unlikely(frozen_buffer)) /* It's usually NULL */ | |
af1e76d6 | 778 | jbd2_free(frozen_buffer, bh->b_size); |
470decc6 DK |
779 | |
780 | JBUFFER_TRACE(jh, "exit"); | |
781 | return error; | |
782 | } | |
783 | ||
784 | /** | |
f7f4bccb | 785 | * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update. |
470decc6 DK |
786 | * @handle: transaction to add buffer modifications to |
787 | * @bh: bh to be used for metadata writes | |
788 | * @credits: variable that will receive credits for the buffer | |
789 | * | |
790 | * Returns an error code or 0 on success. | |
791 | * | |
792 | * In full data journalling mode the buffer may be of type BJ_AsyncData, | |
793 | * because we're write()ing a buffer which is also part of a shared mapping. | |
794 | */ | |
795 | ||
f7f4bccb | 796 | int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh) |
470decc6 | 797 | { |
f7f4bccb | 798 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
799 | int rc; |
800 | ||
801 | /* We do not want to get caught playing with fields which the | |
802 | * log thread also manipulates. Make sure that the buffer | |
803 | * completes any outstanding IO before proceeding. */ | |
804 | rc = do_get_write_access(handle, jh, 0); | |
f7f4bccb | 805 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
806 | return rc; |
807 | } | |
808 | ||
809 | ||
810 | /* | |
811 | * When the user wants to journal a newly created buffer_head | |
812 | * (ie. getblk() returned a new buffer and we are going to populate it | |
813 | * manually rather than reading off disk), then we need to keep the | |
814 | * buffer_head locked until it has been completely filled with new | |
815 | * data. In this case, we should be able to make the assertion that | |
816 | * the bh is not already part of an existing transaction. | |
817 | * | |
818 | * The buffer should already be locked by the caller by this point. | |
819 | * There is no lock ranking violation: it was a newly created, | |
820 | * unlocked buffer beforehand. */ | |
821 | ||
822 | /** | |
f7f4bccb | 823 | * int jbd2_journal_get_create_access () - notify intent to use newly created bh |
470decc6 DK |
824 | * @handle: transaction to new buffer to |
825 | * @bh: new buffer. | |
826 | * | |
827 | * Call this if you create a new bh. | |
828 | */ | |
f7f4bccb | 829 | int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
830 | { |
831 | transaction_t *transaction = handle->h_transaction; | |
832 | journal_t *journal = transaction->t_journal; | |
f7f4bccb | 833 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
834 | int err; |
835 | ||
836 | jbd_debug(5, "journal_head %p\n", jh); | |
837 | err = -EROFS; | |
838 | if (is_handle_aborted(handle)) | |
839 | goto out; | |
840 | err = 0; | |
841 | ||
842 | JBUFFER_TRACE(jh, "entry"); | |
843 | /* | |
844 | * The buffer may already belong to this transaction due to pre-zeroing | |
845 | * in the filesystem's new_block code. It may also be on the previous, | |
846 | * committing transaction's lists, but it HAS to be in Forget state in | |
847 | * that case: the transaction must have deleted the buffer for it to be | |
848 | * reused here. | |
849 | */ | |
850 | jbd_lock_bh_state(bh); | |
851 | spin_lock(&journal->j_list_lock); | |
852 | J_ASSERT_JH(jh, (jh->b_transaction == transaction || | |
853 | jh->b_transaction == NULL || | |
854 | (jh->b_transaction == journal->j_committing_transaction && | |
855 | jh->b_jlist == BJ_Forget))); | |
856 | ||
857 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
858 | J_ASSERT_JH(jh, buffer_locked(jh2bh(jh))); | |
859 | ||
860 | if (jh->b_transaction == NULL) { | |
f91d1d04 JK |
861 | /* |
862 | * Previous jbd2_journal_forget() could have left the buffer | |
863 | * with jbddirty bit set because it was being committed. When | |
864 | * the commit finished, we've filed the buffer for | |
865 | * checkpointing and marked it dirty. Now we are reallocating | |
866 | * the buffer so the transaction freeing it must have | |
867 | * committed and so it's safe to clear the dirty bit. | |
868 | */ | |
869 | clear_buffer_dirty(jh2bh(jh)); | |
470decc6 | 870 | jh->b_transaction = transaction; |
9fc7c63a JB |
871 | |
872 | /* first access by this transaction */ | |
873 | jh->b_modified = 0; | |
874 | ||
470decc6 | 875 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); |
f7f4bccb | 876 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
470decc6 | 877 | } else if (jh->b_transaction == journal->j_committing_transaction) { |
9fc7c63a JB |
878 | /* first access by this transaction */ |
879 | jh->b_modified = 0; | |
880 | ||
470decc6 DK |
881 | JBUFFER_TRACE(jh, "set next transaction"); |
882 | jh->b_next_transaction = transaction; | |
883 | } | |
884 | spin_unlock(&journal->j_list_lock); | |
885 | jbd_unlock_bh_state(bh); | |
886 | ||
887 | /* | |
888 | * akpm: I added this. ext3_alloc_branch can pick up new indirect | |
889 | * blocks which contain freed but then revoked metadata. We need | |
890 | * to cancel the revoke in case we end up freeing it yet again | |
891 | * and the reallocating as data - this would cause a second revoke, | |
892 | * which hits an assertion error. | |
893 | */ | |
894 | JBUFFER_TRACE(jh, "cancelling revoke"); | |
f7f4bccb MC |
895 | jbd2_journal_cancel_revoke(handle, jh); |
896 | jbd2_journal_put_journal_head(jh); | |
470decc6 DK |
897 | out: |
898 | return err; | |
899 | } | |
900 | ||
901 | /** | |
f7f4bccb | 902 | * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with |
470decc6 DK |
903 | * non-rewindable consequences |
904 | * @handle: transaction | |
905 | * @bh: buffer to undo | |
906 | * @credits: store the number of taken credits here (if not NULL) | |
907 | * | |
908 | * Sometimes there is a need to distinguish between metadata which has | |
909 | * been committed to disk and that which has not. The ext3fs code uses | |
910 | * this for freeing and allocating space, we have to make sure that we | |
911 | * do not reuse freed space until the deallocation has been committed, | |
912 | * since if we overwrote that space we would make the delete | |
913 | * un-rewindable in case of a crash. | |
914 | * | |
f7f4bccb | 915 | * To deal with that, jbd2_journal_get_undo_access requests write access to a |
470decc6 DK |
916 | * buffer for parts of non-rewindable operations such as delete |
917 | * operations on the bitmaps. The journaling code must keep a copy of | |
918 | * the buffer's contents prior to the undo_access call until such time | |
919 | * as we know that the buffer has definitely been committed to disk. | |
920 | * | |
921 | * We never need to know which transaction the committed data is part | |
922 | * of, buffers touched here are guaranteed to be dirtied later and so | |
923 | * will be committed to a new transaction in due course, at which point | |
924 | * we can discard the old committed data pointer. | |
925 | * | |
926 | * Returns error number or 0 on success. | |
927 | */ | |
f7f4bccb | 928 | int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
929 | { |
930 | int err; | |
f7f4bccb | 931 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
932 | char *committed_data = NULL; |
933 | ||
934 | JBUFFER_TRACE(jh, "entry"); | |
935 | ||
936 | /* | |
937 | * Do this first --- it can drop the journal lock, so we want to | |
938 | * make sure that obtaining the committed_data is done | |
939 | * atomically wrt. completion of any outstanding commits. | |
940 | */ | |
941 | err = do_get_write_access(handle, jh, 1); | |
942 | if (err) | |
943 | goto out; | |
944 | ||
945 | repeat: | |
946 | if (!jh->b_committed_data) { | |
af1e76d6 | 947 | committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS); |
470decc6 DK |
948 | if (!committed_data) { |
949 | printk(KERN_EMERG "%s: No memory for committed data\n", | |
329d291f | 950 | __func__); |
470decc6 DK |
951 | err = -ENOMEM; |
952 | goto out; | |
953 | } | |
954 | } | |
955 | ||
956 | jbd_lock_bh_state(bh); | |
957 | if (!jh->b_committed_data) { | |
958 | /* Copy out the current buffer contents into the | |
959 | * preserved, committed copy. */ | |
960 | JBUFFER_TRACE(jh, "generate b_committed data"); | |
961 | if (!committed_data) { | |
962 | jbd_unlock_bh_state(bh); | |
963 | goto repeat; | |
964 | } | |
965 | ||
966 | jh->b_committed_data = committed_data; | |
967 | committed_data = NULL; | |
968 | memcpy(jh->b_committed_data, bh->b_data, bh->b_size); | |
969 | } | |
970 | jbd_unlock_bh_state(bh); | |
971 | out: | |
f7f4bccb | 972 | jbd2_journal_put_journal_head(jh); |
470decc6 | 973 | if (unlikely(committed_data)) |
af1e76d6 | 974 | jbd2_free(committed_data, bh->b_size); |
470decc6 DK |
975 | return err; |
976 | } | |
977 | ||
e06c8227 JB |
978 | /** |
979 | * void jbd2_journal_set_triggers() - Add triggers for commit writeout | |
980 | * @bh: buffer to trigger on | |
981 | * @type: struct jbd2_buffer_trigger_type containing the trigger(s). | |
982 | * | |
983 | * Set any triggers on this journal_head. This is always safe, because | |
984 | * triggers for a committing buffer will be saved off, and triggers for | |
985 | * a running transaction will match the buffer in that transaction. | |
986 | * | |
987 | * Call with NULL to clear the triggers. | |
988 | */ | |
989 | void jbd2_journal_set_triggers(struct buffer_head *bh, | |
990 | struct jbd2_buffer_trigger_type *type) | |
991 | { | |
992 | struct journal_head *jh = bh2jh(bh); | |
993 | ||
994 | jh->b_triggers = type; | |
995 | } | |
996 | ||
997 | void jbd2_buffer_commit_trigger(struct journal_head *jh, void *mapped_data, | |
998 | struct jbd2_buffer_trigger_type *triggers) | |
999 | { | |
1000 | struct buffer_head *bh = jh2bh(jh); | |
1001 | ||
1002 | if (!triggers || !triggers->t_commit) | |
1003 | return; | |
1004 | ||
1005 | triggers->t_commit(triggers, bh, mapped_data, bh->b_size); | |
1006 | } | |
1007 | ||
1008 | void jbd2_buffer_abort_trigger(struct journal_head *jh, | |
1009 | struct jbd2_buffer_trigger_type *triggers) | |
1010 | { | |
1011 | if (!triggers || !triggers->t_abort) | |
1012 | return; | |
1013 | ||
1014 | triggers->t_abort(triggers, jh2bh(jh)); | |
1015 | } | |
1016 | ||
1017 | ||
1018 | ||
470decc6 | 1019 | /** |
f7f4bccb | 1020 | * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata |
470decc6 DK |
1021 | * @handle: transaction to add buffer to. |
1022 | * @bh: buffer to mark | |
1023 | * | |
1024 | * mark dirty metadata which needs to be journaled as part of the current | |
1025 | * transaction. | |
1026 | * | |
1027 | * The buffer is placed on the transaction's metadata list and is marked | |
1028 | * as belonging to the transaction. | |
1029 | * | |
1030 | * Returns error number or 0 on success. | |
1031 | * | |
1032 | * Special care needs to be taken if the buffer already belongs to the | |
1033 | * current committing transaction (in which case we should have frozen | |
1034 | * data present for that commit). In that case, we don't relink the | |
1035 | * buffer: that only gets done when the old transaction finally | |
1036 | * completes its commit. | |
1037 | */ | |
f7f4bccb | 1038 | int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1039 | { |
1040 | transaction_t *transaction = handle->h_transaction; | |
1041 | journal_t *journal = transaction->t_journal; | |
1042 | struct journal_head *jh = bh2jh(bh); | |
1043 | ||
1044 | jbd_debug(5, "journal_head %p\n", jh); | |
1045 | JBUFFER_TRACE(jh, "entry"); | |
1046 | if (is_handle_aborted(handle)) | |
1047 | goto out; | |
1048 | ||
1049 | jbd_lock_bh_state(bh); | |
1050 | ||
1051 | if (jh->b_modified == 0) { | |
1052 | /* | |
1053 | * This buffer's got modified and becoming part | |
1054 | * of the transaction. This needs to be done | |
1055 | * once a transaction -bzzz | |
1056 | */ | |
1057 | jh->b_modified = 1; | |
1058 | J_ASSERT_JH(jh, handle->h_buffer_credits > 0); | |
1059 | handle->h_buffer_credits--; | |
1060 | } | |
1061 | ||
1062 | /* | |
1063 | * fastpath, to avoid expensive locking. If this buffer is already | |
1064 | * on the running transaction's metadata list there is nothing to do. | |
1065 | * Nobody can take it off again because there is a handle open. | |
1066 | * I _think_ we're OK here with SMP barriers - a mistaken decision will | |
1067 | * result in this test being false, so we go in and take the locks. | |
1068 | */ | |
1069 | if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) { | |
1070 | JBUFFER_TRACE(jh, "fastpath"); | |
1071 | J_ASSERT_JH(jh, jh->b_transaction == | |
1072 | journal->j_running_transaction); | |
1073 | goto out_unlock_bh; | |
1074 | } | |
1075 | ||
1076 | set_buffer_jbddirty(bh); | |
1077 | ||
1078 | /* | |
1079 | * Metadata already on the current transaction list doesn't | |
1080 | * need to be filed. Metadata on another transaction's list must | |
1081 | * be committing, and will be refiled once the commit completes: | |
1082 | * leave it alone for now. | |
1083 | */ | |
1084 | if (jh->b_transaction != transaction) { | |
1085 | JBUFFER_TRACE(jh, "already on other transaction"); | |
1086 | J_ASSERT_JH(jh, jh->b_transaction == | |
1087 | journal->j_committing_transaction); | |
1088 | J_ASSERT_JH(jh, jh->b_next_transaction == transaction); | |
1089 | /* And this case is illegal: we can't reuse another | |
1090 | * transaction's data buffer, ever. */ | |
1091 | goto out_unlock_bh; | |
1092 | } | |
1093 | ||
1094 | /* That test should have eliminated the following case: */ | |
4019191b | 1095 | J_ASSERT_JH(jh, jh->b_frozen_data == NULL); |
470decc6 DK |
1096 | |
1097 | JBUFFER_TRACE(jh, "file as BJ_Metadata"); | |
1098 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 1099 | __jbd2_journal_file_buffer(jh, handle->h_transaction, BJ_Metadata); |
470decc6 DK |
1100 | spin_unlock(&journal->j_list_lock); |
1101 | out_unlock_bh: | |
1102 | jbd_unlock_bh_state(bh); | |
1103 | out: | |
1104 | JBUFFER_TRACE(jh, "exit"); | |
1105 | return 0; | |
1106 | } | |
1107 | ||
1108 | /* | |
f7f4bccb | 1109 | * jbd2_journal_release_buffer: undo a get_write_access without any buffer |
470decc6 DK |
1110 | * updates, if the update decided in the end that it didn't need access. |
1111 | * | |
1112 | */ | |
1113 | void | |
f7f4bccb | 1114 | jbd2_journal_release_buffer(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1115 | { |
1116 | BUFFER_TRACE(bh, "entry"); | |
1117 | } | |
1118 | ||
1119 | /** | |
f7f4bccb | 1120 | * void jbd2_journal_forget() - bforget() for potentially-journaled buffers. |
470decc6 DK |
1121 | * @handle: transaction handle |
1122 | * @bh: bh to 'forget' | |
1123 | * | |
1124 | * We can only do the bforget if there are no commits pending against the | |
1125 | * buffer. If the buffer is dirty in the current running transaction we | |
1126 | * can safely unlink it. | |
1127 | * | |
1128 | * bh may not be a journalled buffer at all - it may be a non-JBD | |
1129 | * buffer which came off the hashtable. Check for this. | |
1130 | * | |
1131 | * Decrements bh->b_count by one. | |
1132 | * | |
1133 | * Allow this call even if the handle has aborted --- it may be part of | |
1134 | * the caller's cleanup after an abort. | |
1135 | */ | |
f7f4bccb | 1136 | int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1137 | { |
1138 | transaction_t *transaction = handle->h_transaction; | |
1139 | journal_t *journal = transaction->t_journal; | |
1140 | struct journal_head *jh; | |
1141 | int drop_reserve = 0; | |
1142 | int err = 0; | |
1dfc3220 | 1143 | int was_modified = 0; |
470decc6 DK |
1144 | |
1145 | BUFFER_TRACE(bh, "entry"); | |
1146 | ||
1147 | jbd_lock_bh_state(bh); | |
1148 | spin_lock(&journal->j_list_lock); | |
1149 | ||
1150 | if (!buffer_jbd(bh)) | |
1151 | goto not_jbd; | |
1152 | jh = bh2jh(bh); | |
1153 | ||
1154 | /* Critical error: attempting to delete a bitmap buffer, maybe? | |
1155 | * Don't do any jbd operations, and return an error. */ | |
1156 | if (!J_EXPECT_JH(jh, !jh->b_committed_data, | |
1157 | "inconsistent data on disk")) { | |
1158 | err = -EIO; | |
1159 | goto not_jbd; | |
1160 | } | |
1161 | ||
1dfc3220 JB |
1162 | /* keep track of wether or not this transaction modified us */ |
1163 | was_modified = jh->b_modified; | |
1164 | ||
470decc6 DK |
1165 | /* |
1166 | * The buffer's going from the transaction, we must drop | |
1167 | * all references -bzzz | |
1168 | */ | |
1169 | jh->b_modified = 0; | |
1170 | ||
1171 | if (jh->b_transaction == handle->h_transaction) { | |
1172 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
1173 | ||
1174 | /* If we are forgetting a buffer which is already part | |
1175 | * of this transaction, then we can just drop it from | |
1176 | * the transaction immediately. */ | |
1177 | clear_buffer_dirty(bh); | |
1178 | clear_buffer_jbddirty(bh); | |
1179 | ||
1180 | JBUFFER_TRACE(jh, "belongs to current transaction: unfile"); | |
1181 | ||
1dfc3220 JB |
1182 | /* |
1183 | * we only want to drop a reference if this transaction | |
1184 | * modified the buffer | |
1185 | */ | |
1186 | if (was_modified) | |
1187 | drop_reserve = 1; | |
470decc6 DK |
1188 | |
1189 | /* | |
1190 | * We are no longer going to journal this buffer. | |
1191 | * However, the commit of this transaction is still | |
1192 | * important to the buffer: the delete that we are now | |
1193 | * processing might obsolete an old log entry, so by | |
1194 | * committing, we can satisfy the buffer's checkpoint. | |
1195 | * | |
1196 | * So, if we have a checkpoint on the buffer, we should | |
1197 | * now refile the buffer on our BJ_Forget list so that | |
1198 | * we know to remove the checkpoint after we commit. | |
1199 | */ | |
1200 | ||
1201 | if (jh->b_cp_transaction) { | |
f7f4bccb MC |
1202 | __jbd2_journal_temp_unlink_buffer(jh); |
1203 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); | |
470decc6 | 1204 | } else { |
f7f4bccb MC |
1205 | __jbd2_journal_unfile_buffer(jh); |
1206 | jbd2_journal_remove_journal_head(bh); | |
470decc6 DK |
1207 | __brelse(bh); |
1208 | if (!buffer_jbd(bh)) { | |
1209 | spin_unlock(&journal->j_list_lock); | |
1210 | jbd_unlock_bh_state(bh); | |
1211 | __bforget(bh); | |
1212 | goto drop; | |
1213 | } | |
1214 | } | |
1215 | } else if (jh->b_transaction) { | |
1216 | J_ASSERT_JH(jh, (jh->b_transaction == | |
1217 | journal->j_committing_transaction)); | |
1218 | /* However, if the buffer is still owned by a prior | |
1219 | * (committing) transaction, we can't drop it yet... */ | |
1220 | JBUFFER_TRACE(jh, "belongs to older transaction"); | |
1221 | /* ... but we CAN drop it from the new transaction if we | |
1222 | * have also modified it since the original commit. */ | |
1223 | ||
1224 | if (jh->b_next_transaction) { | |
1225 | J_ASSERT(jh->b_next_transaction == transaction); | |
1226 | jh->b_next_transaction = NULL; | |
1dfc3220 JB |
1227 | |
1228 | /* | |
1229 | * only drop a reference if this transaction modified | |
1230 | * the buffer | |
1231 | */ | |
1232 | if (was_modified) | |
1233 | drop_reserve = 1; | |
470decc6 DK |
1234 | } |
1235 | } | |
1236 | ||
1237 | not_jbd: | |
1238 | spin_unlock(&journal->j_list_lock); | |
1239 | jbd_unlock_bh_state(bh); | |
1240 | __brelse(bh); | |
1241 | drop: | |
1242 | if (drop_reserve) { | |
1243 | /* no need to reserve log space for this block -bzzz */ | |
1244 | handle->h_buffer_credits++; | |
1245 | } | |
1246 | return err; | |
1247 | } | |
1248 | ||
1249 | /** | |
f7f4bccb | 1250 | * int jbd2_journal_stop() - complete a transaction |
470decc6 DK |
1251 | * @handle: tranaction to complete. |
1252 | * | |
1253 | * All done for a particular handle. | |
1254 | * | |
1255 | * There is not much action needed here. We just return any remaining | |
1256 | * buffer credits to the transaction and remove the handle. The only | |
1257 | * complication is that we need to start a commit operation if the | |
1258 | * filesystem is marked for synchronous update. | |
1259 | * | |
f7f4bccb | 1260 | * jbd2_journal_stop itself will not usually return an error, but it may |
470decc6 | 1261 | * do so in unusual circumstances. In particular, expect it to |
f7f4bccb | 1262 | * return -EIO if a jbd2_journal_abort has been executed since the |
470decc6 DK |
1263 | * transaction began. |
1264 | */ | |
f7f4bccb | 1265 | int jbd2_journal_stop(handle_t *handle) |
470decc6 DK |
1266 | { |
1267 | transaction_t *transaction = handle->h_transaction; | |
1268 | journal_t *journal = transaction->t_journal; | |
a51dca9c TT |
1269 | int err, wait_for_commit = 0; |
1270 | tid_t tid; | |
470decc6 DK |
1271 | pid_t pid; |
1272 | ||
470decc6 DK |
1273 | J_ASSERT(journal_current_handle() == handle); |
1274 | ||
1275 | if (is_handle_aborted(handle)) | |
1276 | err = -EIO; | |
3e2a532b | 1277 | else { |
a51dca9c | 1278 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 | 1279 | err = 0; |
3e2a532b | 1280 | } |
470decc6 DK |
1281 | |
1282 | if (--handle->h_ref > 0) { | |
1283 | jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1, | |
1284 | handle->h_ref); | |
1285 | return err; | |
1286 | } | |
1287 | ||
1288 | jbd_debug(4, "Handle %p going down\n", handle); | |
1289 | ||
1290 | /* | |
1291 | * Implement synchronous transaction batching. If the handle | |
1292 | * was synchronous, don't force a commit immediately. Let's | |
e07f7183 JB |
1293 | * yield and let another thread piggyback onto this |
1294 | * transaction. Keep doing that while new threads continue to | |
1295 | * arrive. It doesn't cost much - we're about to run a commit | |
1296 | * and sleep on IO anyway. Speeds up many-threaded, many-dir | |
1297 | * operations by 30x or more... | |
1298 | * | |
1299 | * We try and optimize the sleep time against what the | |
1300 | * underlying disk can do, instead of having a static sleep | |
1301 | * time. This is useful for the case where our storage is so | |
1302 | * fast that it is more optimal to go ahead and force a flush | |
1303 | * and wait for the transaction to be committed than it is to | |
1304 | * wait for an arbitrary amount of time for new writers to | |
1305 | * join the transaction. We achieve this by measuring how | |
1306 | * long it takes to commit a transaction, and compare it with | |
1307 | * how long this transaction has been running, and if run time | |
1308 | * < commit time then we sleep for the delta and commit. This | |
1309 | * greatly helps super fast disks that would see slowdowns as | |
1310 | * more threads started doing fsyncs. | |
470decc6 | 1311 | * |
e07f7183 JB |
1312 | * But don't do this if this process was the most recent one |
1313 | * to perform a synchronous write. We do this to detect the | |
1314 | * case where a single process is doing a stream of sync | |
1315 | * writes. No point in waiting for joiners in that case. | |
470decc6 DK |
1316 | */ |
1317 | pid = current->pid; | |
1318 | if (handle->h_sync && journal->j_last_sync_writer != pid) { | |
e07f7183 JB |
1319 | u64 commit_time, trans_time; |
1320 | ||
470decc6 | 1321 | journal->j_last_sync_writer = pid; |
e07f7183 | 1322 | |
a931da6a | 1323 | read_lock(&journal->j_state_lock); |
e07f7183 | 1324 | commit_time = journal->j_average_commit_time; |
a931da6a | 1325 | read_unlock(&journal->j_state_lock); |
e07f7183 JB |
1326 | |
1327 | trans_time = ktime_to_ns(ktime_sub(ktime_get(), | |
1328 | transaction->t_start_time)); | |
1329 | ||
30773840 TT |
1330 | commit_time = max_t(u64, commit_time, |
1331 | 1000*journal->j_min_batch_time); | |
e07f7183 | 1332 | commit_time = min_t(u64, commit_time, |
30773840 | 1333 | 1000*journal->j_max_batch_time); |
e07f7183 JB |
1334 | |
1335 | if (trans_time < commit_time) { | |
1336 | ktime_t expires = ktime_add_ns(ktime_get(), | |
1337 | commit_time); | |
1338 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1339 | schedule_hrtimeout(&expires, HRTIMER_MODE_ABS); | |
1340 | } | |
470decc6 DK |
1341 | } |
1342 | ||
7058548c TT |
1343 | if (handle->h_sync) |
1344 | transaction->t_synchronous_commit = 1; | |
470decc6 | 1345 | current->journal_info = NULL; |
a51dca9c TT |
1346 | atomic_sub(handle->h_buffer_credits, |
1347 | &transaction->t_outstanding_credits); | |
470decc6 DK |
1348 | |
1349 | /* | |
1350 | * If the handle is marked SYNC, we need to set another commit | |
1351 | * going! We also want to force a commit if the current | |
1352 | * transaction is occupying too much of the log, or if the | |
1353 | * transaction is too old now. | |
1354 | */ | |
1355 | if (handle->h_sync || | |
a51dca9c TT |
1356 | (atomic_read(&transaction->t_outstanding_credits) > |
1357 | journal->j_max_transaction_buffers) || | |
1358 | time_after_eq(jiffies, transaction->t_expires)) { | |
470decc6 DK |
1359 | /* Do this even for aborted journals: an abort still |
1360 | * completes the commit thread, it just doesn't write | |
1361 | * anything to disk. */ | |
470decc6 | 1362 | |
470decc6 DK |
1363 | jbd_debug(2, "transaction too old, requesting commit for " |
1364 | "handle %p\n", handle); | |
1365 | /* This is non-blocking */ | |
c35a56a0 | 1366 | jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1367 | |
1368 | /* | |
f7f4bccb | 1369 | * Special case: JBD2_SYNC synchronous updates require us |
470decc6 DK |
1370 | * to wait for the commit to complete. |
1371 | */ | |
1372 | if (handle->h_sync && !(current->flags & PF_MEMALLOC)) | |
a51dca9c | 1373 | wait_for_commit = 1; |
470decc6 DK |
1374 | } |
1375 | ||
a51dca9c TT |
1376 | /* |
1377 | * Once we drop t_updates, if it goes to zero the transaction | |
1378 | * could start commiting on us and eventually disappear. So | |
1379 | * once we do this, we must not dereference transaction | |
1380 | * pointer again. | |
1381 | */ | |
1382 | tid = transaction->t_tid; | |
1383 | if (atomic_dec_and_test(&transaction->t_updates)) { | |
1384 | wake_up(&journal->j_wait_updates); | |
1385 | if (journal->j_barrier_count) | |
1386 | wake_up(&journal->j_wait_transaction_locked); | |
1387 | } | |
1388 | ||
1389 | if (wait_for_commit) | |
1390 | err = jbd2_log_wait_commit(journal, tid); | |
1391 | ||
3295f0ef | 1392 | lock_map_release(&handle->h_lockdep_map); |
7b751066 | 1393 | |
af1e76d6 | 1394 | jbd2_free_handle(handle); |
470decc6 DK |
1395 | return err; |
1396 | } | |
1397 | ||
5648ba5b RD |
1398 | /** |
1399 | * int jbd2_journal_force_commit() - force any uncommitted transactions | |
470decc6 DK |
1400 | * @journal: journal to force |
1401 | * | |
1402 | * For synchronous operations: force any uncommitted transactions | |
1403 | * to disk. May seem kludgy, but it reuses all the handle batching | |
1404 | * code in a very simple manner. | |
1405 | */ | |
f7f4bccb | 1406 | int jbd2_journal_force_commit(journal_t *journal) |
470decc6 DK |
1407 | { |
1408 | handle_t *handle; | |
1409 | int ret; | |
1410 | ||
f7f4bccb | 1411 | handle = jbd2_journal_start(journal, 1); |
470decc6 DK |
1412 | if (IS_ERR(handle)) { |
1413 | ret = PTR_ERR(handle); | |
1414 | } else { | |
1415 | handle->h_sync = 1; | |
f7f4bccb | 1416 | ret = jbd2_journal_stop(handle); |
470decc6 DK |
1417 | } |
1418 | return ret; | |
1419 | } | |
1420 | ||
1421 | /* | |
1422 | * | |
1423 | * List management code snippets: various functions for manipulating the | |
1424 | * transaction buffer lists. | |
1425 | * | |
1426 | */ | |
1427 | ||
1428 | /* | |
1429 | * Append a buffer to a transaction list, given the transaction's list head | |
1430 | * pointer. | |
1431 | * | |
1432 | * j_list_lock is held. | |
1433 | * | |
1434 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1435 | */ | |
1436 | ||
1437 | static inline void | |
1438 | __blist_add_buffer(struct journal_head **list, struct journal_head *jh) | |
1439 | { | |
1440 | if (!*list) { | |
1441 | jh->b_tnext = jh->b_tprev = jh; | |
1442 | *list = jh; | |
1443 | } else { | |
1444 | /* Insert at the tail of the list to preserve order */ | |
1445 | struct journal_head *first = *list, *last = first->b_tprev; | |
1446 | jh->b_tprev = last; | |
1447 | jh->b_tnext = first; | |
1448 | last->b_tnext = first->b_tprev = jh; | |
1449 | } | |
1450 | } | |
1451 | ||
1452 | /* | |
1453 | * Remove a buffer from a transaction list, given the transaction's list | |
1454 | * head pointer. | |
1455 | * | |
1456 | * Called with j_list_lock held, and the journal may not be locked. | |
1457 | * | |
1458 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1459 | */ | |
1460 | ||
1461 | static inline void | |
1462 | __blist_del_buffer(struct journal_head **list, struct journal_head *jh) | |
1463 | { | |
1464 | if (*list == jh) { | |
1465 | *list = jh->b_tnext; | |
1466 | if (*list == jh) | |
1467 | *list = NULL; | |
1468 | } | |
1469 | jh->b_tprev->b_tnext = jh->b_tnext; | |
1470 | jh->b_tnext->b_tprev = jh->b_tprev; | |
1471 | } | |
1472 | ||
1473 | /* | |
1474 | * Remove a buffer from the appropriate transaction list. | |
1475 | * | |
1476 | * Note that this function can *change* the value of | |
87c89c23 JK |
1477 | * bh->b_transaction->t_buffers, t_forget, t_iobuf_list, t_shadow_list, |
1478 | * t_log_list or t_reserved_list. If the caller is holding onto a copy of one | |
1479 | * of these pointers, it could go bad. Generally the caller needs to re-read | |
1480 | * the pointer from the transaction_t. | |
470decc6 DK |
1481 | * |
1482 | * Called under j_list_lock. The journal may not be locked. | |
1483 | */ | |
f7f4bccb | 1484 | void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh) |
470decc6 DK |
1485 | { |
1486 | struct journal_head **list = NULL; | |
1487 | transaction_t *transaction; | |
1488 | struct buffer_head *bh = jh2bh(jh); | |
1489 | ||
1490 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
1491 | transaction = jh->b_transaction; | |
1492 | if (transaction) | |
1493 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
1494 | ||
1495 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
1496 | if (jh->b_jlist != BJ_None) | |
4019191b | 1497 | J_ASSERT_JH(jh, transaction != NULL); |
470decc6 DK |
1498 | |
1499 | switch (jh->b_jlist) { | |
1500 | case BJ_None: | |
1501 | return; | |
470decc6 DK |
1502 | case BJ_Metadata: |
1503 | transaction->t_nr_buffers--; | |
1504 | J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0); | |
1505 | list = &transaction->t_buffers; | |
1506 | break; | |
1507 | case BJ_Forget: | |
1508 | list = &transaction->t_forget; | |
1509 | break; | |
1510 | case BJ_IO: | |
1511 | list = &transaction->t_iobuf_list; | |
1512 | break; | |
1513 | case BJ_Shadow: | |
1514 | list = &transaction->t_shadow_list; | |
1515 | break; | |
1516 | case BJ_LogCtl: | |
1517 | list = &transaction->t_log_list; | |
1518 | break; | |
1519 | case BJ_Reserved: | |
1520 | list = &transaction->t_reserved_list; | |
1521 | break; | |
470decc6 DK |
1522 | } |
1523 | ||
1524 | __blist_del_buffer(list, jh); | |
1525 | jh->b_jlist = BJ_None; | |
1526 | if (test_clear_buffer_jbddirty(bh)) | |
1527 | mark_buffer_dirty(bh); /* Expose it to the VM */ | |
1528 | } | |
1529 | ||
f7f4bccb | 1530 | void __jbd2_journal_unfile_buffer(struct journal_head *jh) |
470decc6 | 1531 | { |
f7f4bccb | 1532 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 DK |
1533 | jh->b_transaction = NULL; |
1534 | } | |
1535 | ||
f7f4bccb | 1536 | void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 DK |
1537 | { |
1538 | jbd_lock_bh_state(jh2bh(jh)); | |
1539 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 1540 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
1541 | spin_unlock(&journal->j_list_lock); |
1542 | jbd_unlock_bh_state(jh2bh(jh)); | |
1543 | } | |
1544 | ||
1545 | /* | |
f7f4bccb | 1546 | * Called from jbd2_journal_try_to_free_buffers(). |
470decc6 DK |
1547 | * |
1548 | * Called under jbd_lock_bh_state(bh) | |
1549 | */ | |
1550 | static void | |
1551 | __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh) | |
1552 | { | |
1553 | struct journal_head *jh; | |
1554 | ||
1555 | jh = bh2jh(bh); | |
1556 | ||
1557 | if (buffer_locked(bh) || buffer_dirty(bh)) | |
1558 | goto out; | |
1559 | ||
4019191b | 1560 | if (jh->b_next_transaction != NULL) |
470decc6 DK |
1561 | goto out; |
1562 | ||
1563 | spin_lock(&journal->j_list_lock); | |
87c89c23 | 1564 | if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) { |
470decc6 DK |
1565 | /* written-back checkpointed metadata buffer */ |
1566 | if (jh->b_jlist == BJ_None) { | |
1567 | JBUFFER_TRACE(jh, "remove from checkpoint list"); | |
f7f4bccb MC |
1568 | __jbd2_journal_remove_checkpoint(jh); |
1569 | jbd2_journal_remove_journal_head(bh); | |
470decc6 DK |
1570 | __brelse(bh); |
1571 | } | |
1572 | } | |
1573 | spin_unlock(&journal->j_list_lock); | |
1574 | out: | |
1575 | return; | |
1576 | } | |
1577 | ||
470decc6 | 1578 | /** |
f7f4bccb | 1579 | * int jbd2_journal_try_to_free_buffers() - try to free page buffers. |
470decc6 DK |
1580 | * @journal: journal for operation |
1581 | * @page: to try and free | |
530576bb MC |
1582 | * @gfp_mask: we use the mask to detect how hard should we try to release |
1583 | * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to | |
1584 | * release the buffers. | |
470decc6 DK |
1585 | * |
1586 | * | |
1587 | * For all the buffers on this page, | |
1588 | * if they are fully written out ordered data, move them onto BUF_CLEAN | |
1589 | * so try_to_free_buffers() can reap them. | |
1590 | * | |
1591 | * This function returns non-zero if we wish try_to_free_buffers() | |
1592 | * to be called. We do this if the page is releasable by try_to_free_buffers(). | |
1593 | * We also do it if the page has locked or dirty buffers and the caller wants | |
1594 | * us to perform sync or async writeout. | |
1595 | * | |
1596 | * This complicates JBD locking somewhat. We aren't protected by the | |
1597 | * BKL here. We wish to remove the buffer from its committing or | |
f7f4bccb | 1598 | * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer. |
470decc6 DK |
1599 | * |
1600 | * This may *change* the value of transaction_t->t_datalist, so anyone | |
1601 | * who looks at t_datalist needs to lock against this function. | |
1602 | * | |
f7f4bccb MC |
1603 | * Even worse, someone may be doing a jbd2_journal_dirty_data on this |
1604 | * buffer. So we need to lock against that. jbd2_journal_dirty_data() | |
470decc6 DK |
1605 | * will come out of the lock with the buffer dirty, which makes it |
1606 | * ineligible for release here. | |
1607 | * | |
1608 | * Who else is affected by this? hmm... Really the only contender | |
1609 | * is do_get_write_access() - it could be looking at the buffer while | |
1610 | * journal_try_to_free_buffer() is changing its state. But that | |
1611 | * cannot happen because we never reallocate freed data as metadata | |
1612 | * while the data is part of a transaction. Yes? | |
530576bb MC |
1613 | * |
1614 | * Return 0 on failure, 1 on success | |
470decc6 | 1615 | */ |
f7f4bccb | 1616 | int jbd2_journal_try_to_free_buffers(journal_t *journal, |
530576bb | 1617 | struct page *page, gfp_t gfp_mask) |
470decc6 DK |
1618 | { |
1619 | struct buffer_head *head; | |
1620 | struct buffer_head *bh; | |
1621 | int ret = 0; | |
1622 | ||
1623 | J_ASSERT(PageLocked(page)); | |
1624 | ||
1625 | head = page_buffers(page); | |
1626 | bh = head; | |
1627 | do { | |
1628 | struct journal_head *jh; | |
1629 | ||
1630 | /* | |
1631 | * We take our own ref against the journal_head here to avoid | |
1632 | * having to add tons of locking around each instance of | |
530576bb MC |
1633 | * jbd2_journal_remove_journal_head() and |
1634 | * jbd2_journal_put_journal_head(). | |
470decc6 | 1635 | */ |
f7f4bccb | 1636 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
1637 | if (!jh) |
1638 | continue; | |
1639 | ||
1640 | jbd_lock_bh_state(bh); | |
1641 | __journal_try_to_free_buffer(journal, bh); | |
f7f4bccb | 1642 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1643 | jbd_unlock_bh_state(bh); |
1644 | if (buffer_jbd(bh)) | |
1645 | goto busy; | |
1646 | } while ((bh = bh->b_this_page) != head); | |
530576bb | 1647 | |
470decc6 | 1648 | ret = try_to_free_buffers(page); |
530576bb | 1649 | |
470decc6 DK |
1650 | busy: |
1651 | return ret; | |
1652 | } | |
1653 | ||
1654 | /* | |
1655 | * This buffer is no longer needed. If it is on an older transaction's | |
1656 | * checkpoint list we need to record it on this transaction's forget list | |
1657 | * to pin this buffer (and hence its checkpointing transaction) down until | |
1658 | * this transaction commits. If the buffer isn't on a checkpoint list, we | |
1659 | * release it. | |
1660 | * Returns non-zero if JBD no longer has an interest in the buffer. | |
1661 | * | |
1662 | * Called under j_list_lock. | |
1663 | * | |
1664 | * Called under jbd_lock_bh_state(bh). | |
1665 | */ | |
1666 | static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction) | |
1667 | { | |
1668 | int may_free = 1; | |
1669 | struct buffer_head *bh = jh2bh(jh); | |
1670 | ||
f7f4bccb | 1671 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
1672 | |
1673 | if (jh->b_cp_transaction) { | |
1674 | JBUFFER_TRACE(jh, "on running+cp transaction"); | |
f91d1d04 JK |
1675 | /* |
1676 | * We don't want to write the buffer anymore, clear the | |
1677 | * bit so that we don't confuse checks in | |
1678 | * __journal_file_buffer | |
1679 | */ | |
1680 | clear_buffer_dirty(bh); | |
f7f4bccb | 1681 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); |
470decc6 DK |
1682 | may_free = 0; |
1683 | } else { | |
1684 | JBUFFER_TRACE(jh, "on running transaction"); | |
f7f4bccb | 1685 | jbd2_journal_remove_journal_head(bh); |
470decc6 DK |
1686 | __brelse(bh); |
1687 | } | |
1688 | return may_free; | |
1689 | } | |
1690 | ||
1691 | /* | |
f7f4bccb | 1692 | * jbd2_journal_invalidatepage |
470decc6 DK |
1693 | * |
1694 | * This code is tricky. It has a number of cases to deal with. | |
1695 | * | |
1696 | * There are two invariants which this code relies on: | |
1697 | * | |
1698 | * i_size must be updated on disk before we start calling invalidatepage on the | |
1699 | * data. | |
1700 | * | |
1701 | * This is done in ext3 by defining an ext3_setattr method which | |
1702 | * updates i_size before truncate gets going. By maintaining this | |
1703 | * invariant, we can be sure that it is safe to throw away any buffers | |
1704 | * attached to the current transaction: once the transaction commits, | |
1705 | * we know that the data will not be needed. | |
1706 | * | |
1707 | * Note however that we can *not* throw away data belonging to the | |
1708 | * previous, committing transaction! | |
1709 | * | |
1710 | * Any disk blocks which *are* part of the previous, committing | |
1711 | * transaction (and which therefore cannot be discarded immediately) are | |
1712 | * not going to be reused in the new running transaction | |
1713 | * | |
1714 | * The bitmap committed_data images guarantee this: any block which is | |
1715 | * allocated in one transaction and removed in the next will be marked | |
1716 | * as in-use in the committed_data bitmap, so cannot be reused until | |
1717 | * the next transaction to delete the block commits. This means that | |
1718 | * leaving committing buffers dirty is quite safe: the disk blocks | |
1719 | * cannot be reallocated to a different file and so buffer aliasing is | |
1720 | * not possible. | |
1721 | * | |
1722 | * | |
1723 | * The above applies mainly to ordered data mode. In writeback mode we | |
1724 | * don't make guarantees about the order in which data hits disk --- in | |
1725 | * particular we don't guarantee that new dirty data is flushed before | |
1726 | * transaction commit --- so it is always safe just to discard data | |
1727 | * immediately in that mode. --sct | |
1728 | */ | |
1729 | ||
1730 | /* | |
1731 | * The journal_unmap_buffer helper function returns zero if the buffer | |
1732 | * concerned remains pinned as an anonymous buffer belonging to an older | |
1733 | * transaction. | |
1734 | * | |
1735 | * We're outside-transaction here. Either or both of j_running_transaction | |
1736 | * and j_committing_transaction may be NULL. | |
1737 | */ | |
1738 | static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh) | |
1739 | { | |
1740 | transaction_t *transaction; | |
1741 | struct journal_head *jh; | |
1742 | int may_free = 1; | |
1743 | int ret; | |
1744 | ||
1745 | BUFFER_TRACE(bh, "entry"); | |
1746 | ||
1747 | /* | |
1748 | * It is safe to proceed here without the j_list_lock because the | |
1749 | * buffers cannot be stolen by try_to_free_buffers as long as we are | |
1750 | * holding the page lock. --sct | |
1751 | */ | |
1752 | ||
1753 | if (!buffer_jbd(bh)) | |
1754 | goto zap_buffer_unlocked; | |
1755 | ||
87c89c23 | 1756 | /* OK, we have data buffer in journaled mode */ |
a931da6a | 1757 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1758 | jbd_lock_bh_state(bh); |
1759 | spin_lock(&journal->j_list_lock); | |
1760 | ||
f7f4bccb | 1761 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
1762 | if (!jh) |
1763 | goto zap_buffer_no_jh; | |
1764 | ||
ba869023 | 1765 | /* |
1766 | * We cannot remove the buffer from checkpoint lists until the | |
1767 | * transaction adding inode to orphan list (let's call it T) | |
1768 | * is committed. Otherwise if the transaction changing the | |
1769 | * buffer would be cleaned from the journal before T is | |
1770 | * committed, a crash will cause that the correct contents of | |
1771 | * the buffer will be lost. On the other hand we have to | |
1772 | * clear the buffer dirty bit at latest at the moment when the | |
1773 | * transaction marking the buffer as freed in the filesystem | |
1774 | * structures is committed because from that moment on the | |
1775 | * buffer can be reallocated and used by a different page. | |
1776 | * Since the block hasn't been freed yet but the inode has | |
1777 | * already been added to orphan list, it is safe for us to add | |
1778 | * the buffer to BJ_Forget list of the newest transaction. | |
1779 | */ | |
470decc6 DK |
1780 | transaction = jh->b_transaction; |
1781 | if (transaction == NULL) { | |
1782 | /* First case: not on any transaction. If it | |
1783 | * has no checkpoint link, then we can zap it: | |
1784 | * it's a writeback-mode buffer so we don't care | |
1785 | * if it hits disk safely. */ | |
1786 | if (!jh->b_cp_transaction) { | |
1787 | JBUFFER_TRACE(jh, "not on any transaction: zap"); | |
1788 | goto zap_buffer; | |
1789 | } | |
1790 | ||
1791 | if (!buffer_dirty(bh)) { | |
1792 | /* bdflush has written it. We can drop it now */ | |
1793 | goto zap_buffer; | |
1794 | } | |
1795 | ||
1796 | /* OK, it must be in the journal but still not | |
1797 | * written fully to disk: it's metadata or | |
1798 | * journaled data... */ | |
1799 | ||
1800 | if (journal->j_running_transaction) { | |
1801 | /* ... and once the current transaction has | |
1802 | * committed, the buffer won't be needed any | |
1803 | * longer. */ | |
1804 | JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget"); | |
1805 | ret = __dispose_buffer(jh, | |
1806 | journal->j_running_transaction); | |
f7f4bccb | 1807 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1808 | spin_unlock(&journal->j_list_lock); |
1809 | jbd_unlock_bh_state(bh); | |
a931da6a | 1810 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1811 | return ret; |
1812 | } else { | |
1813 | /* There is no currently-running transaction. So the | |
1814 | * orphan record which we wrote for this file must have | |
1815 | * passed into commit. We must attach this buffer to | |
1816 | * the committing transaction, if it exists. */ | |
1817 | if (journal->j_committing_transaction) { | |
1818 | JBUFFER_TRACE(jh, "give to committing trans"); | |
1819 | ret = __dispose_buffer(jh, | |
1820 | journal->j_committing_transaction); | |
f7f4bccb | 1821 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1822 | spin_unlock(&journal->j_list_lock); |
1823 | jbd_unlock_bh_state(bh); | |
a931da6a | 1824 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1825 | return ret; |
1826 | } else { | |
1827 | /* The orphan record's transaction has | |
1828 | * committed. We can cleanse this buffer */ | |
1829 | clear_buffer_jbddirty(bh); | |
1830 | goto zap_buffer; | |
1831 | } | |
1832 | } | |
1833 | } else if (transaction == journal->j_committing_transaction) { | |
9b57988d | 1834 | JBUFFER_TRACE(jh, "on committing transaction"); |
470decc6 | 1835 | /* |
ba869023 | 1836 | * The buffer is committing, we simply cannot touch |
1837 | * it. So we just set j_next_transaction to the | |
1838 | * running transaction (if there is one) and mark | |
1839 | * buffer as freed so that commit code knows it should | |
1840 | * clear dirty bits when it is done with the buffer. | |
1841 | */ | |
470decc6 | 1842 | set_buffer_freed(bh); |
ba869023 | 1843 | if (journal->j_running_transaction && buffer_jbddirty(bh)) |
1844 | jh->b_next_transaction = journal->j_running_transaction; | |
f7f4bccb | 1845 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1846 | spin_unlock(&journal->j_list_lock); |
1847 | jbd_unlock_bh_state(bh); | |
a931da6a | 1848 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1849 | return 0; |
1850 | } else { | |
1851 | /* Good, the buffer belongs to the running transaction. | |
1852 | * We are writing our own transaction's data, not any | |
1853 | * previous one's, so it is safe to throw it away | |
1854 | * (remember that we expect the filesystem to have set | |
1855 | * i_size already for this truncate so recovery will not | |
1856 | * expose the disk blocks we are discarding here.) */ | |
1857 | J_ASSERT_JH(jh, transaction == journal->j_running_transaction); | |
9b57988d | 1858 | JBUFFER_TRACE(jh, "on running transaction"); |
470decc6 DK |
1859 | may_free = __dispose_buffer(jh, transaction); |
1860 | } | |
1861 | ||
1862 | zap_buffer: | |
f7f4bccb | 1863 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1864 | zap_buffer_no_jh: |
1865 | spin_unlock(&journal->j_list_lock); | |
1866 | jbd_unlock_bh_state(bh); | |
a931da6a | 1867 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1868 | zap_buffer_unlocked: |
1869 | clear_buffer_dirty(bh); | |
1870 | J_ASSERT_BH(bh, !buffer_jbddirty(bh)); | |
1871 | clear_buffer_mapped(bh); | |
1872 | clear_buffer_req(bh); | |
1873 | clear_buffer_new(bh); | |
1874 | bh->b_bdev = NULL; | |
1875 | return may_free; | |
1876 | } | |
1877 | ||
1878 | /** | |
f7f4bccb | 1879 | * void jbd2_journal_invalidatepage() |
470decc6 DK |
1880 | * @journal: journal to use for flush... |
1881 | * @page: page to flush | |
1882 | * @offset: length of page to invalidate. | |
1883 | * | |
1884 | * Reap page buffers containing data after offset in page. | |
1885 | * | |
1886 | */ | |
f7f4bccb | 1887 | void jbd2_journal_invalidatepage(journal_t *journal, |
470decc6 DK |
1888 | struct page *page, |
1889 | unsigned long offset) | |
1890 | { | |
1891 | struct buffer_head *head, *bh, *next; | |
1892 | unsigned int curr_off = 0; | |
1893 | int may_free = 1; | |
1894 | ||
1895 | if (!PageLocked(page)) | |
1896 | BUG(); | |
1897 | if (!page_has_buffers(page)) | |
1898 | return; | |
1899 | ||
1900 | /* We will potentially be playing with lists other than just the | |
1901 | * data lists (especially for journaled data mode), so be | |
1902 | * cautious in our locking. */ | |
1903 | ||
1904 | head = bh = page_buffers(page); | |
1905 | do { | |
1906 | unsigned int next_off = curr_off + bh->b_size; | |
1907 | next = bh->b_this_page; | |
1908 | ||
1909 | if (offset <= curr_off) { | |
1910 | /* This block is wholly outside the truncation point */ | |
1911 | lock_buffer(bh); | |
1912 | may_free &= journal_unmap_buffer(journal, bh); | |
1913 | unlock_buffer(bh); | |
1914 | } | |
1915 | curr_off = next_off; | |
1916 | bh = next; | |
1917 | ||
1918 | } while (bh != head); | |
1919 | ||
1920 | if (!offset) { | |
1921 | if (may_free && try_to_free_buffers(page)) | |
1922 | J_ASSERT(!page_has_buffers(page)); | |
1923 | } | |
1924 | } | |
1925 | ||
1926 | /* | |
1927 | * File a buffer on the given transaction list. | |
1928 | */ | |
f7f4bccb | 1929 | void __jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
1930 | transaction_t *transaction, int jlist) |
1931 | { | |
1932 | struct journal_head **list = NULL; | |
1933 | int was_dirty = 0; | |
1934 | struct buffer_head *bh = jh2bh(jh); | |
1935 | ||
1936 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
1937 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
1938 | ||
1939 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
1940 | J_ASSERT_JH(jh, jh->b_transaction == transaction || | |
4019191b | 1941 | jh->b_transaction == NULL); |
470decc6 DK |
1942 | |
1943 | if (jh->b_transaction && jh->b_jlist == jlist) | |
1944 | return; | |
1945 | ||
470decc6 DK |
1946 | if (jlist == BJ_Metadata || jlist == BJ_Reserved || |
1947 | jlist == BJ_Shadow || jlist == BJ_Forget) { | |
f91d1d04 JK |
1948 | /* |
1949 | * For metadata buffers, we track dirty bit in buffer_jbddirty | |
1950 | * instead of buffer_dirty. We should not see a dirty bit set | |
1951 | * here because we clear it in do_get_write_access but e.g. | |
1952 | * tune2fs can modify the sb and set the dirty bit at any time | |
1953 | * so we try to gracefully handle that. | |
1954 | */ | |
1955 | if (buffer_dirty(bh)) | |
1956 | warn_dirty_buffer(bh); | |
470decc6 DK |
1957 | if (test_clear_buffer_dirty(bh) || |
1958 | test_clear_buffer_jbddirty(bh)) | |
1959 | was_dirty = 1; | |
1960 | } | |
1961 | ||
1962 | if (jh->b_transaction) | |
f7f4bccb | 1963 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 DK |
1964 | jh->b_transaction = transaction; |
1965 | ||
1966 | switch (jlist) { | |
1967 | case BJ_None: | |
1968 | J_ASSERT_JH(jh, !jh->b_committed_data); | |
1969 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
1970 | return; | |
470decc6 DK |
1971 | case BJ_Metadata: |
1972 | transaction->t_nr_buffers++; | |
1973 | list = &transaction->t_buffers; | |
1974 | break; | |
1975 | case BJ_Forget: | |
1976 | list = &transaction->t_forget; | |
1977 | break; | |
1978 | case BJ_IO: | |
1979 | list = &transaction->t_iobuf_list; | |
1980 | break; | |
1981 | case BJ_Shadow: | |
1982 | list = &transaction->t_shadow_list; | |
1983 | break; | |
1984 | case BJ_LogCtl: | |
1985 | list = &transaction->t_log_list; | |
1986 | break; | |
1987 | case BJ_Reserved: | |
1988 | list = &transaction->t_reserved_list; | |
1989 | break; | |
470decc6 DK |
1990 | } |
1991 | ||
1992 | __blist_add_buffer(list, jh); | |
1993 | jh->b_jlist = jlist; | |
1994 | ||
1995 | if (was_dirty) | |
1996 | set_buffer_jbddirty(bh); | |
1997 | } | |
1998 | ||
f7f4bccb | 1999 | void jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2000 | transaction_t *transaction, int jlist) |
2001 | { | |
2002 | jbd_lock_bh_state(jh2bh(jh)); | |
2003 | spin_lock(&transaction->t_journal->j_list_lock); | |
f7f4bccb | 2004 | __jbd2_journal_file_buffer(jh, transaction, jlist); |
470decc6 DK |
2005 | spin_unlock(&transaction->t_journal->j_list_lock); |
2006 | jbd_unlock_bh_state(jh2bh(jh)); | |
2007 | } | |
2008 | ||
2009 | /* | |
2010 | * Remove a buffer from its current buffer list in preparation for | |
2011 | * dropping it from its current transaction entirely. If the buffer has | |
2012 | * already started to be used by a subsequent transaction, refile the | |
2013 | * buffer on that transaction's metadata list. | |
2014 | * | |
2015 | * Called under journal->j_list_lock | |
2016 | * | |
2017 | * Called under jbd_lock_bh_state(jh2bh(jh)) | |
2018 | */ | |
f7f4bccb | 2019 | void __jbd2_journal_refile_buffer(struct journal_head *jh) |
470decc6 | 2020 | { |
ba869023 | 2021 | int was_dirty, jlist; |
470decc6 DK |
2022 | struct buffer_head *bh = jh2bh(jh); |
2023 | ||
2024 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
2025 | if (jh->b_transaction) | |
2026 | assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock); | |
2027 | ||
2028 | /* If the buffer is now unused, just drop it. */ | |
2029 | if (jh->b_next_transaction == NULL) { | |
f7f4bccb | 2030 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
2031 | return; |
2032 | } | |
2033 | ||
2034 | /* | |
2035 | * It has been modified by a later transaction: add it to the new | |
2036 | * transaction's metadata list. | |
2037 | */ | |
2038 | ||
2039 | was_dirty = test_clear_buffer_jbddirty(bh); | |
f7f4bccb | 2040 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 DK |
2041 | jh->b_transaction = jh->b_next_transaction; |
2042 | jh->b_next_transaction = NULL; | |
ba869023 | 2043 | if (buffer_freed(bh)) |
2044 | jlist = BJ_Forget; | |
2045 | else if (jh->b_modified) | |
2046 | jlist = BJ_Metadata; | |
2047 | else | |
2048 | jlist = BJ_Reserved; | |
2049 | __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist); | |
470decc6 DK |
2050 | J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING); |
2051 | ||
2052 | if (was_dirty) | |
2053 | set_buffer_jbddirty(bh); | |
2054 | } | |
2055 | ||
2056 | /* | |
2057 | * For the unlocked version of this call, also make sure that any | |
2058 | * hanging journal_head is cleaned up if necessary. | |
2059 | * | |
f7f4bccb | 2060 | * __jbd2_journal_refile_buffer is usually called as part of a single locked |
470decc6 DK |
2061 | * operation on a buffer_head, in which the caller is probably going to |
2062 | * be hooking the journal_head onto other lists. In that case it is up | |
2063 | * to the caller to remove the journal_head if necessary. For the | |
f7f4bccb | 2064 | * unlocked jbd2_journal_refile_buffer call, the caller isn't going to be |
470decc6 DK |
2065 | * doing anything else to the buffer so we need to do the cleanup |
2066 | * ourselves to avoid a jh leak. | |
2067 | * | |
2068 | * *** The journal_head may be freed by this call! *** | |
2069 | */ | |
f7f4bccb | 2070 | void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 DK |
2071 | { |
2072 | struct buffer_head *bh = jh2bh(jh); | |
2073 | ||
2074 | jbd_lock_bh_state(bh); | |
2075 | spin_lock(&journal->j_list_lock); | |
2076 | ||
f7f4bccb | 2077 | __jbd2_journal_refile_buffer(jh); |
470decc6 | 2078 | jbd_unlock_bh_state(bh); |
f7f4bccb | 2079 | jbd2_journal_remove_journal_head(bh); |
470decc6 DK |
2080 | |
2081 | spin_unlock(&journal->j_list_lock); | |
2082 | __brelse(bh); | |
2083 | } | |
c851ed54 JK |
2084 | |
2085 | /* | |
2086 | * File inode in the inode list of the handle's transaction | |
2087 | */ | |
2088 | int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode) | |
2089 | { | |
2090 | transaction_t *transaction = handle->h_transaction; | |
2091 | journal_t *journal = transaction->t_journal; | |
2092 | ||
2093 | if (is_handle_aborted(handle)) | |
2094 | return -EIO; | |
2095 | ||
2096 | jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino, | |
2097 | transaction->t_tid); | |
2098 | ||
2099 | /* | |
2100 | * First check whether inode isn't already on the transaction's | |
2101 | * lists without taking the lock. Note that this check is safe | |
2102 | * without the lock as we cannot race with somebody removing inode | |
2103 | * from the transaction. The reason is that we remove inode from the | |
2104 | * transaction only in journal_release_jbd_inode() and when we commit | |
2105 | * the transaction. We are guarded from the first case by holding | |
2106 | * a reference to the inode. We are safe against the second case | |
2107 | * because if jinode->i_transaction == transaction, commit code | |
2108 | * cannot touch the transaction because we hold reference to it, | |
2109 | * and if jinode->i_next_transaction == transaction, commit code | |
2110 | * will only file the inode where we want it. | |
2111 | */ | |
2112 | if (jinode->i_transaction == transaction || | |
2113 | jinode->i_next_transaction == transaction) | |
2114 | return 0; | |
2115 | ||
2116 | spin_lock(&journal->j_list_lock); | |
2117 | ||
2118 | if (jinode->i_transaction == transaction || | |
2119 | jinode->i_next_transaction == transaction) | |
2120 | goto done; | |
2121 | ||
2122 | /* On some different transaction's list - should be | |
2123 | * the committing one */ | |
2124 | if (jinode->i_transaction) { | |
2125 | J_ASSERT(jinode->i_next_transaction == NULL); | |
2126 | J_ASSERT(jinode->i_transaction == | |
2127 | journal->j_committing_transaction); | |
2128 | jinode->i_next_transaction = transaction; | |
2129 | goto done; | |
2130 | } | |
2131 | /* Not on any transaction list... */ | |
2132 | J_ASSERT(!jinode->i_next_transaction); | |
2133 | jinode->i_transaction = transaction; | |
2134 | list_add(&jinode->i_list, &transaction->t_inode_list); | |
2135 | done: | |
2136 | spin_unlock(&journal->j_list_lock); | |
2137 | ||
2138 | return 0; | |
2139 | } | |
2140 | ||
2141 | /* | |
7f5aa215 JK |
2142 | * File truncate and transaction commit interact with each other in a |
2143 | * non-trivial way. If a transaction writing data block A is | |
2144 | * committing, we cannot discard the data by truncate until we have | |
2145 | * written them. Otherwise if we crashed after the transaction with | |
2146 | * write has committed but before the transaction with truncate has | |
2147 | * committed, we could see stale data in block A. This function is a | |
2148 | * helper to solve this problem. It starts writeout of the truncated | |
2149 | * part in case it is in the committing transaction. | |
2150 | * | |
2151 | * Filesystem code must call this function when inode is journaled in | |
2152 | * ordered mode before truncation happens and after the inode has been | |
2153 | * placed on orphan list with the new inode size. The second condition | |
2154 | * avoids the race that someone writes new data and we start | |
2155 | * committing the transaction after this function has been called but | |
2156 | * before a transaction for truncate is started (and furthermore it | |
2157 | * allows us to optimize the case where the addition to orphan list | |
2158 | * happens in the same transaction as write --- we don't have to write | |
2159 | * any data in such case). | |
c851ed54 | 2160 | */ |
7f5aa215 JK |
2161 | int jbd2_journal_begin_ordered_truncate(journal_t *journal, |
2162 | struct jbd2_inode *jinode, | |
c851ed54 JK |
2163 | loff_t new_size) |
2164 | { | |
7f5aa215 | 2165 | transaction_t *inode_trans, *commit_trans; |
c851ed54 JK |
2166 | int ret = 0; |
2167 | ||
7f5aa215 JK |
2168 | /* This is a quick check to avoid locking if not necessary */ |
2169 | if (!jinode->i_transaction) | |
c851ed54 | 2170 | goto out; |
7f5aa215 JK |
2171 | /* Locks are here just to force reading of recent values, it is |
2172 | * enough that the transaction was not committing before we started | |
2173 | * a transaction adding the inode to orphan list */ | |
a931da6a | 2174 | read_lock(&journal->j_state_lock); |
c851ed54 | 2175 | commit_trans = journal->j_committing_transaction; |
a931da6a | 2176 | read_unlock(&journal->j_state_lock); |
7f5aa215 JK |
2177 | spin_lock(&journal->j_list_lock); |
2178 | inode_trans = jinode->i_transaction; | |
2179 | spin_unlock(&journal->j_list_lock); | |
2180 | if (inode_trans == commit_trans) { | |
2181 | ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping, | |
c851ed54 JK |
2182 | new_size, LLONG_MAX); |
2183 | if (ret) | |
2184 | jbd2_journal_abort(journal, ret); | |
2185 | } | |
2186 | out: | |
2187 | return ret; | |
2188 | } |