]>
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); | |
8dd42046 | 60 | atomic_set(&transaction->t_handle_count, 0); |
c851ed54 | 61 | INIT_LIST_HEAD(&transaction->t_inode_list); |
3e624fc7 | 62 | INIT_LIST_HEAD(&transaction->t_private_list); |
470decc6 DK |
63 | |
64 | /* Set up the commit timer for the new transaction. */ | |
b1f485f2 | 65 | journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires); |
470decc6 DK |
66 | add_timer(&journal->j_commit_timer); |
67 | ||
68 | J_ASSERT(journal->j_running_transaction == NULL); | |
69 | journal->j_running_transaction = transaction; | |
8e85fb3f JL |
70 | transaction->t_max_wait = 0; |
71 | transaction->t_start = jiffies; | |
470decc6 DK |
72 | |
73 | return transaction; | |
74 | } | |
75 | ||
76 | /* | |
77 | * Handle management. | |
78 | * | |
79 | * A handle_t is an object which represents a single atomic update to a | |
80 | * filesystem, and which tracks all of the modifications which form part | |
81 | * of that one update. | |
82 | */ | |
83 | ||
6d0bf005 TT |
84 | /* |
85 | * Update transiaction's maximum wait time, if debugging is enabled. | |
86 | * | |
87 | * In order for t_max_wait to be reliable, it must be protected by a | |
88 | * lock. But doing so will mean that start_this_handle() can not be | |
89 | * run in parallel on SMP systems, which limits our scalability. So | |
90 | * unless debugging is enabled, we no longer update t_max_wait, which | |
91 | * means that maximum wait time reported by the jbd2_run_stats | |
92 | * tracepoint will always be zero. | |
93 | */ | |
94 | static inline void update_t_max_wait(transaction_t *transaction) | |
95 | { | |
96 | #ifdef CONFIG_JBD2_DEBUG | |
97 | unsigned long ts = jiffies; | |
98 | ||
99 | if (jbd2_journal_enable_debug && | |
100 | time_after(transaction->t_start, ts)) { | |
101 | ts = jbd2_time_diff(ts, transaction->t_start); | |
102 | spin_lock(&transaction->t_handle_lock); | |
103 | if (ts > transaction->t_max_wait) | |
104 | transaction->t_max_wait = ts; | |
105 | spin_unlock(&transaction->t_handle_lock); | |
106 | } | |
107 | #endif | |
108 | } | |
109 | ||
470decc6 DK |
110 | /* |
111 | * start_this_handle: Given a handle, deal with any locking or stalling | |
112 | * needed to make sure that there is enough journal space for the handle | |
113 | * to begin. Attach the handle to a transaction and set up the | |
114 | * transaction's buffer credits. | |
115 | */ | |
116 | ||
47def826 TT |
117 | static int start_this_handle(journal_t *journal, handle_t *handle, |
118 | int gfp_mask) | |
470decc6 DK |
119 | { |
120 | transaction_t *transaction; | |
121 | int needed; | |
122 | int nblocks = handle->h_buffer_credits; | |
123 | transaction_t *new_transaction = NULL; | |
470decc6 DK |
124 | |
125 | if (nblocks > journal->j_max_transaction_buffers) { | |
126 | printk(KERN_ERR "JBD: %s wants too many credits (%d > %d)\n", | |
127 | current->comm, nblocks, | |
128 | journal->j_max_transaction_buffers); | |
47def826 | 129 | return -ENOSPC; |
470decc6 DK |
130 | } |
131 | ||
132 | alloc_transaction: | |
133 | if (!journal->j_running_transaction) { | |
47def826 | 134 | new_transaction = kzalloc(sizeof(*new_transaction), gfp_mask); |
470decc6 | 135 | if (!new_transaction) { |
47def826 TT |
136 | /* |
137 | * If __GFP_FS is not present, then we may be | |
138 | * being called from inside the fs writeback | |
139 | * layer, so we MUST NOT fail. Since | |
140 | * __GFP_NOFAIL is going away, we will arrange | |
141 | * to retry the allocation ourselves. | |
142 | */ | |
143 | if ((gfp_mask & __GFP_FS) == 0) { | |
144 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
145 | goto alloc_transaction; | |
146 | } | |
147 | return -ENOMEM; | |
470decc6 | 148 | } |
470decc6 DK |
149 | } |
150 | ||
151 | jbd_debug(3, "New handle %p going live.\n", handle); | |
152 | ||
470decc6 DK |
153 | /* |
154 | * We need to hold j_state_lock until t_updates has been incremented, | |
155 | * for proper journal barrier handling | |
156 | */ | |
a931da6a TT |
157 | repeat: |
158 | read_lock(&journal->j_state_lock); | |
5c2178e7 | 159 | BUG_ON(journal->j_flags & JBD2_UNMOUNT); |
470decc6 | 160 | if (is_journal_aborted(journal) || |
f7f4bccb | 161 | (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) { |
a931da6a | 162 | read_unlock(&journal->j_state_lock); |
47def826 TT |
163 | kfree(new_transaction); |
164 | return -EROFS; | |
470decc6 DK |
165 | } |
166 | ||
167 | /* Wait on the journal's transaction barrier if necessary */ | |
168 | if (journal->j_barrier_count) { | |
a931da6a | 169 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
170 | wait_event(journal->j_wait_transaction_locked, |
171 | journal->j_barrier_count == 0); | |
172 | goto repeat; | |
173 | } | |
174 | ||
175 | if (!journal->j_running_transaction) { | |
a931da6a TT |
176 | read_unlock(&journal->j_state_lock); |
177 | if (!new_transaction) | |
470decc6 | 178 | goto alloc_transaction; |
a931da6a TT |
179 | write_lock(&journal->j_state_lock); |
180 | if (!journal->j_running_transaction) { | |
181 | jbd2_get_transaction(journal, new_transaction); | |
182 | new_transaction = NULL; | |
470decc6 | 183 | } |
a931da6a TT |
184 | write_unlock(&journal->j_state_lock); |
185 | goto repeat; | |
470decc6 DK |
186 | } |
187 | ||
188 | transaction = journal->j_running_transaction; | |
189 | ||
190 | /* | |
191 | * If the current transaction is locked down for commit, wait for the | |
192 | * lock to be released. | |
193 | */ | |
194 | if (transaction->t_state == T_LOCKED) { | |
195 | DEFINE_WAIT(wait); | |
196 | ||
197 | prepare_to_wait(&journal->j_wait_transaction_locked, | |
198 | &wait, TASK_UNINTERRUPTIBLE); | |
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 | * If there is not enough space left in the log to write all potential | |
207 | * buffers requested by this operation, we need to stall pending a log | |
208 | * checkpoint to free some more log space. | |
209 | */ | |
8dd42046 TT |
210 | needed = atomic_add_return(nblocks, |
211 | &transaction->t_outstanding_credits); | |
470decc6 DK |
212 | |
213 | if (needed > journal->j_max_transaction_buffers) { | |
214 | /* | |
215 | * If the current transaction is already too large, then start | |
216 | * to commit it: we can then go back and attach this handle to | |
217 | * a new transaction. | |
218 | */ | |
219 | DEFINE_WAIT(wait); | |
220 | ||
221 | jbd_debug(2, "Handle %p starting new commit...\n", handle); | |
8dd42046 | 222 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
470decc6 DK |
223 | prepare_to_wait(&journal->j_wait_transaction_locked, &wait, |
224 | TASK_UNINTERRUPTIBLE); | |
f7f4bccb | 225 | __jbd2_log_start_commit(journal, transaction->t_tid); |
a931da6a | 226 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
227 | schedule(); |
228 | finish_wait(&journal->j_wait_transaction_locked, &wait); | |
229 | goto repeat; | |
230 | } | |
231 | ||
232 | /* | |
233 | * The commit code assumes that it can get enough log space | |
234 | * without forcing a checkpoint. This is *critical* for | |
235 | * correctness: a checkpoint of a buffer which is also | |
236 | * associated with a committing transaction creates a deadlock, | |
237 | * so commit simply cannot force through checkpoints. | |
238 | * | |
239 | * We must therefore ensure the necessary space in the journal | |
240 | * *before* starting to dirty potentially checkpointed buffers | |
241 | * in the new transaction. | |
242 | * | |
243 | * The worst part is, any transaction currently committing can | |
244 | * reduce the free space arbitrarily. Be careful to account for | |
245 | * those buffers when checkpointing. | |
246 | */ | |
247 | ||
248 | /* | |
249 | * @@@ AKPM: This seems rather over-defensive. We're giving commit | |
250 | * a _lot_ of headroom: 1/4 of the journal plus the size of | |
251 | * the committing transaction. Really, we only need to give it | |
252 | * committing_transaction->t_outstanding_credits plus "enough" for | |
253 | * the log control blocks. | |
254 | * Also, this test is inconsitent with the matching one in | |
f7f4bccb | 255 | * jbd2_journal_extend(). |
470decc6 | 256 | */ |
f7f4bccb | 257 | if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) { |
470decc6 | 258 | jbd_debug(2, "Handle %p waiting for checkpoint...\n", handle); |
8dd42046 | 259 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
a931da6a TT |
260 | read_unlock(&journal->j_state_lock); |
261 | write_lock(&journal->j_state_lock); | |
262 | if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) | |
263 | __jbd2_log_wait_for_space(journal); | |
264 | write_unlock(&journal->j_state_lock); | |
265 | goto repeat; | |
470decc6 DK |
266 | } |
267 | ||
268 | /* OK, account for the buffers that this operation expects to | |
8dd42046 | 269 | * use and add the handle to the running transaction. |
8dd42046 | 270 | */ |
6d0bf005 | 271 | update_t_max_wait(transaction); |
470decc6 | 272 | handle->h_transaction = transaction; |
a51dca9c | 273 | atomic_inc(&transaction->t_updates); |
8dd42046 | 274 | atomic_inc(&transaction->t_handle_count); |
470decc6 | 275 | jbd_debug(4, "Handle %p given %d credits (total %d, free %d)\n", |
a51dca9c TT |
276 | handle, nblocks, |
277 | atomic_read(&transaction->t_outstanding_credits), | |
f7f4bccb | 278 | __jbd2_log_space_left(journal)); |
a931da6a | 279 | read_unlock(&journal->j_state_lock); |
9599b0e5 JK |
280 | |
281 | lock_map_acquire(&handle->h_lockdep_map); | |
47def826 TT |
282 | kfree(new_transaction); |
283 | return 0; | |
470decc6 DK |
284 | } |
285 | ||
7b751066 MC |
286 | static struct lock_class_key jbd2_handle_key; |
287 | ||
470decc6 DK |
288 | /* Allocate a new handle. This should probably be in a slab... */ |
289 | static handle_t *new_handle(int nblocks) | |
290 | { | |
af1e76d6 | 291 | handle_t *handle = jbd2_alloc_handle(GFP_NOFS); |
470decc6 DK |
292 | if (!handle) |
293 | return NULL; | |
294 | memset(handle, 0, sizeof(*handle)); | |
295 | handle->h_buffer_credits = nblocks; | |
296 | handle->h_ref = 1; | |
297 | ||
7b751066 MC |
298 | lockdep_init_map(&handle->h_lockdep_map, "jbd2_handle", |
299 | &jbd2_handle_key, 0); | |
300 | ||
470decc6 DK |
301 | return handle; |
302 | } | |
303 | ||
304 | /** | |
f7f4bccb | 305 | * handle_t *jbd2_journal_start() - Obtain a new handle. |
470decc6 DK |
306 | * @journal: Journal to start transaction on. |
307 | * @nblocks: number of block buffer we might modify | |
308 | * | |
309 | * We make sure that the transaction can guarantee at least nblocks of | |
310 | * modified buffers in the log. We block until the log can guarantee | |
311 | * that much space. | |
312 | * | |
313 | * This function is visible to journal users (like ext3fs), so is not | |
314 | * called with the journal already locked. | |
315 | * | |
316 | * Return a pointer to a newly allocated handle, or NULL on failure | |
317 | */ | |
47def826 | 318 | handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int gfp_mask) |
470decc6 DK |
319 | { |
320 | handle_t *handle = journal_current_handle(); | |
321 | int err; | |
322 | ||
323 | if (!journal) | |
324 | return ERR_PTR(-EROFS); | |
325 | ||
326 | if (handle) { | |
327 | J_ASSERT(handle->h_transaction->t_journal == journal); | |
328 | handle->h_ref++; | |
329 | return handle; | |
330 | } | |
331 | ||
332 | handle = new_handle(nblocks); | |
333 | if (!handle) | |
334 | return ERR_PTR(-ENOMEM); | |
335 | ||
336 | current->journal_info = handle; | |
337 | ||
47def826 | 338 | err = start_this_handle(journal, handle, gfp_mask); |
470decc6 | 339 | if (err < 0) { |
af1e76d6 | 340 | jbd2_free_handle(handle); |
470decc6 DK |
341 | current->journal_info = NULL; |
342 | handle = ERR_PTR(err); | |
7b751066 | 343 | goto out; |
470decc6 | 344 | } |
7b751066 | 345 | out: |
470decc6 DK |
346 | return handle; |
347 | } | |
47def826 TT |
348 | EXPORT_SYMBOL(jbd2__journal_start); |
349 | ||
350 | ||
351 | handle_t *jbd2_journal_start(journal_t *journal, int nblocks) | |
352 | { | |
353 | return jbd2__journal_start(journal, nblocks, GFP_NOFS); | |
354 | } | |
355 | EXPORT_SYMBOL(jbd2_journal_start); | |
356 | ||
470decc6 DK |
357 | |
358 | /** | |
f7f4bccb | 359 | * int jbd2_journal_extend() - extend buffer credits. |
470decc6 DK |
360 | * @handle: handle to 'extend' |
361 | * @nblocks: nr blocks to try to extend by. | |
362 | * | |
363 | * Some transactions, such as large extends and truncates, can be done | |
364 | * atomically all at once or in several stages. The operation requests | |
365 | * a credit for a number of buffer modications in advance, but can | |
366 | * extend its credit if it needs more. | |
367 | * | |
f7f4bccb | 368 | * jbd2_journal_extend tries to give the running handle more buffer credits. |
470decc6 DK |
369 | * It does not guarantee that allocation - this is a best-effort only. |
370 | * The calling process MUST be able to deal cleanly with a failure to | |
371 | * extend here. | |
372 | * | |
373 | * Return 0 on success, non-zero on failure. | |
374 | * | |
375 | * return code < 0 implies an error | |
376 | * return code > 0 implies normal transaction-full status. | |
377 | */ | |
f7f4bccb | 378 | int jbd2_journal_extend(handle_t *handle, int nblocks) |
470decc6 DK |
379 | { |
380 | transaction_t *transaction = handle->h_transaction; | |
381 | journal_t *journal = transaction->t_journal; | |
382 | int result; | |
383 | int wanted; | |
384 | ||
385 | result = -EIO; | |
386 | if (is_handle_aborted(handle)) | |
387 | goto out; | |
388 | ||
389 | result = 1; | |
390 | ||
a931da6a | 391 | read_lock(&journal->j_state_lock); |
470decc6 DK |
392 | |
393 | /* Don't extend a locked-down transaction! */ | |
394 | if (handle->h_transaction->t_state != T_RUNNING) { | |
395 | jbd_debug(3, "denied handle %p %d blocks: " | |
396 | "transaction not running\n", handle, nblocks); | |
397 | goto error_out; | |
398 | } | |
399 | ||
400 | spin_lock(&transaction->t_handle_lock); | |
a51dca9c | 401 | wanted = atomic_read(&transaction->t_outstanding_credits) + nblocks; |
470decc6 DK |
402 | |
403 | if (wanted > journal->j_max_transaction_buffers) { | |
404 | jbd_debug(3, "denied handle %p %d blocks: " | |
405 | "transaction too large\n", handle, nblocks); | |
406 | goto unlock; | |
407 | } | |
408 | ||
f7f4bccb | 409 | if (wanted > __jbd2_log_space_left(journal)) { |
470decc6 DK |
410 | jbd_debug(3, "denied handle %p %d blocks: " |
411 | "insufficient log space\n", handle, nblocks); | |
412 | goto unlock; | |
413 | } | |
414 | ||
415 | handle->h_buffer_credits += nblocks; | |
a51dca9c | 416 | atomic_add(nblocks, &transaction->t_outstanding_credits); |
470decc6 DK |
417 | result = 0; |
418 | ||
419 | jbd_debug(3, "extended handle %p by %d\n", handle, nblocks); | |
420 | unlock: | |
421 | spin_unlock(&transaction->t_handle_lock); | |
422 | error_out: | |
a931da6a | 423 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
424 | out: |
425 | return result; | |
426 | } | |
427 | ||
428 | ||
429 | /** | |
f7f4bccb | 430 | * int jbd2_journal_restart() - restart a handle . |
470decc6 DK |
431 | * @handle: handle to restart |
432 | * @nblocks: nr credits requested | |
433 | * | |
434 | * Restart a handle for a multi-transaction filesystem | |
435 | * operation. | |
436 | * | |
f7f4bccb MC |
437 | * If the jbd2_journal_extend() call above fails to grant new buffer credits |
438 | * to a running handle, a call to jbd2_journal_restart will commit the | |
470decc6 DK |
439 | * handle's transaction so far and reattach the handle to a new |
440 | * transaction capabable of guaranteeing the requested number of | |
441 | * credits. | |
442 | */ | |
47def826 | 443 | int jbd2__journal_restart(handle_t *handle, int nblocks, int gfp_mask) |
470decc6 DK |
444 | { |
445 | transaction_t *transaction = handle->h_transaction; | |
446 | journal_t *journal = transaction->t_journal; | |
447 | int ret; | |
448 | ||
449 | /* If we've had an abort of any type, don't even think about | |
450 | * actually doing the restart! */ | |
451 | if (is_handle_aborted(handle)) | |
452 | return 0; | |
453 | ||
454 | /* | |
455 | * First unlink the handle from its current transaction, and start the | |
456 | * commit on that. | |
457 | */ | |
a51dca9c | 458 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 DK |
459 | J_ASSERT(journal_current_handle() == handle); |
460 | ||
a931da6a | 461 | read_lock(&journal->j_state_lock); |
470decc6 | 462 | spin_lock(&transaction->t_handle_lock); |
a51dca9c TT |
463 | atomic_sub(handle->h_buffer_credits, |
464 | &transaction->t_outstanding_credits); | |
465 | if (atomic_dec_and_test(&transaction->t_updates)) | |
470decc6 DK |
466 | wake_up(&journal->j_wait_updates); |
467 | spin_unlock(&transaction->t_handle_lock); | |
468 | ||
469 | jbd_debug(2, "restarting handle %p\n", handle); | |
f7f4bccb | 470 | __jbd2_log_start_commit(journal, transaction->t_tid); |
a931da6a | 471 | read_unlock(&journal->j_state_lock); |
470decc6 | 472 | |
9599b0e5 | 473 | lock_map_release(&handle->h_lockdep_map); |
470decc6 | 474 | handle->h_buffer_credits = nblocks; |
47def826 | 475 | ret = start_this_handle(journal, handle, gfp_mask); |
470decc6 DK |
476 | return ret; |
477 | } | |
47def826 | 478 | EXPORT_SYMBOL(jbd2__journal_restart); |
470decc6 DK |
479 | |
480 | ||
47def826 TT |
481 | int jbd2_journal_restart(handle_t *handle, int nblocks) |
482 | { | |
483 | return jbd2__journal_restart(handle, nblocks, GFP_NOFS); | |
484 | } | |
485 | EXPORT_SYMBOL(jbd2_journal_restart); | |
486 | ||
470decc6 | 487 | /** |
f7f4bccb | 488 | * void jbd2_journal_lock_updates () - establish a transaction barrier. |
470decc6 DK |
489 | * @journal: Journal to establish a barrier on. |
490 | * | |
491 | * This locks out any further updates from being started, and blocks | |
492 | * until all existing updates have completed, returning only once the | |
493 | * journal is in a quiescent state with no updates running. | |
494 | * | |
495 | * The journal lock should not be held on entry. | |
496 | */ | |
f7f4bccb | 497 | void jbd2_journal_lock_updates(journal_t *journal) |
470decc6 DK |
498 | { |
499 | DEFINE_WAIT(wait); | |
500 | ||
a931da6a | 501 | write_lock(&journal->j_state_lock); |
470decc6 DK |
502 | ++journal->j_barrier_count; |
503 | ||
504 | /* Wait until there are no running updates */ | |
505 | while (1) { | |
506 | transaction_t *transaction = journal->j_running_transaction; | |
507 | ||
508 | if (!transaction) | |
509 | break; | |
510 | ||
511 | spin_lock(&transaction->t_handle_lock); | |
a51dca9c | 512 | if (!atomic_read(&transaction->t_updates)) { |
470decc6 DK |
513 | spin_unlock(&transaction->t_handle_lock); |
514 | break; | |
515 | } | |
516 | prepare_to_wait(&journal->j_wait_updates, &wait, | |
517 | TASK_UNINTERRUPTIBLE); | |
518 | spin_unlock(&transaction->t_handle_lock); | |
a931da6a | 519 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
520 | schedule(); |
521 | finish_wait(&journal->j_wait_updates, &wait); | |
a931da6a | 522 | write_lock(&journal->j_state_lock); |
470decc6 | 523 | } |
a931da6a | 524 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
525 | |
526 | /* | |
527 | * We have now established a barrier against other normal updates, but | |
f7f4bccb | 528 | * we also need to barrier against other jbd2_journal_lock_updates() calls |
470decc6 DK |
529 | * to make sure that we serialise special journal-locked operations |
530 | * too. | |
531 | */ | |
532 | mutex_lock(&journal->j_barrier); | |
533 | } | |
534 | ||
535 | /** | |
f7f4bccb | 536 | * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier |
470decc6 DK |
537 | * @journal: Journal to release the barrier on. |
538 | * | |
f7f4bccb | 539 | * Release a transaction barrier obtained with jbd2_journal_lock_updates(). |
470decc6 DK |
540 | * |
541 | * Should be called without the journal lock held. | |
542 | */ | |
f7f4bccb | 543 | void jbd2_journal_unlock_updates (journal_t *journal) |
470decc6 DK |
544 | { |
545 | J_ASSERT(journal->j_barrier_count != 0); | |
546 | ||
547 | mutex_unlock(&journal->j_barrier); | |
a931da6a | 548 | write_lock(&journal->j_state_lock); |
470decc6 | 549 | --journal->j_barrier_count; |
a931da6a | 550 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
551 | wake_up(&journal->j_wait_transaction_locked); |
552 | } | |
553 | ||
f91d1d04 | 554 | static void warn_dirty_buffer(struct buffer_head *bh) |
470decc6 | 555 | { |
f91d1d04 | 556 | char b[BDEVNAME_SIZE]; |
470decc6 | 557 | |
f91d1d04 JK |
558 | printk(KERN_WARNING |
559 | "JBD: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). " | |
560 | "There's a risk of filesystem corruption in case of system " | |
561 | "crash.\n", | |
562 | bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr); | |
470decc6 DK |
563 | } |
564 | ||
565 | /* | |
566 | * If the buffer is already part of the current transaction, then there | |
567 | * is nothing we need to do. If it is already part of a prior | |
568 | * transaction which we are still committing to disk, then we need to | |
569 | * make sure that we do not overwrite the old copy: we do copy-out to | |
570 | * preserve the copy going to disk. We also account the buffer against | |
571 | * the handle's metadata buffer credits (unless the buffer is already | |
572 | * part of the transaction, that is). | |
573 | * | |
574 | */ | |
575 | static int | |
576 | do_get_write_access(handle_t *handle, struct journal_head *jh, | |
577 | int force_copy) | |
578 | { | |
579 | struct buffer_head *bh; | |
580 | transaction_t *transaction; | |
581 | journal_t *journal; | |
582 | int error; | |
583 | char *frozen_buffer = NULL; | |
584 | int need_copy = 0; | |
585 | ||
586 | if (is_handle_aborted(handle)) | |
587 | return -EROFS; | |
588 | ||
589 | transaction = handle->h_transaction; | |
590 | journal = transaction->t_journal; | |
591 | ||
592 | jbd_debug(5, "buffer_head %p, force_copy %d\n", jh, force_copy); | |
593 | ||
594 | JBUFFER_TRACE(jh, "entry"); | |
595 | repeat: | |
596 | bh = jh2bh(jh); | |
597 | ||
598 | /* @@@ Need to check for errors here at some point. */ | |
599 | ||
600 | lock_buffer(bh); | |
601 | jbd_lock_bh_state(bh); | |
602 | ||
603 | /* We now hold the buffer lock so it is safe to query the buffer | |
604 | * state. Is the buffer dirty? | |
605 | * | |
606 | * If so, there are two possibilities. The buffer may be | |
607 | * non-journaled, and undergoing a quite legitimate writeback. | |
608 | * Otherwise, it is journaled, and we don't expect dirty buffers | |
609 | * in that state (the buffers should be marked JBD_Dirty | |
610 | * instead.) So either the IO is being done under our own | |
611 | * control and this is a bug, or it's a third party IO such as | |
612 | * dump(8) (which may leave the buffer scheduled for read --- | |
613 | * ie. locked but not dirty) or tune2fs (which may actually have | |
614 | * the buffer dirtied, ugh.) */ | |
615 | ||
616 | if (buffer_dirty(bh)) { | |
617 | /* | |
618 | * First question: is this buffer already part of the current | |
619 | * transaction or the existing committing transaction? | |
620 | */ | |
621 | if (jh->b_transaction) { | |
622 | J_ASSERT_JH(jh, | |
623 | jh->b_transaction == transaction || | |
624 | jh->b_transaction == | |
625 | journal->j_committing_transaction); | |
626 | if (jh->b_next_transaction) | |
627 | J_ASSERT_JH(jh, jh->b_next_transaction == | |
628 | transaction); | |
f91d1d04 | 629 | warn_dirty_buffer(bh); |
470decc6 DK |
630 | } |
631 | /* | |
632 | * In any case we need to clean the dirty flag and we must | |
633 | * do it under the buffer lock to be sure we don't race | |
634 | * with running write-out. | |
635 | */ | |
f91d1d04 JK |
636 | JBUFFER_TRACE(jh, "Journalling dirty buffer"); |
637 | clear_buffer_dirty(bh); | |
638 | set_buffer_jbddirty(bh); | |
470decc6 DK |
639 | } |
640 | ||
641 | unlock_buffer(bh); | |
642 | ||
643 | error = -EROFS; | |
644 | if (is_handle_aborted(handle)) { | |
645 | jbd_unlock_bh_state(bh); | |
646 | goto out; | |
647 | } | |
648 | error = 0; | |
649 | ||
650 | /* | |
651 | * The buffer is already part of this transaction if b_transaction or | |
652 | * b_next_transaction points to it | |
653 | */ | |
654 | if (jh->b_transaction == transaction || | |
655 | jh->b_next_transaction == transaction) | |
656 | goto done; | |
657 | ||
9fc7c63a JB |
658 | /* |
659 | * this is the first time this transaction is touching this buffer, | |
660 | * reset the modified flag | |
661 | */ | |
662 | jh->b_modified = 0; | |
663 | ||
470decc6 DK |
664 | /* |
665 | * If there is already a copy-out version of this buffer, then we don't | |
666 | * need to make another one | |
667 | */ | |
668 | if (jh->b_frozen_data) { | |
669 | JBUFFER_TRACE(jh, "has frozen data"); | |
670 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
671 | jh->b_next_transaction = transaction; | |
672 | goto done; | |
673 | } | |
674 | ||
675 | /* Is there data here we need to preserve? */ | |
676 | ||
677 | if (jh->b_transaction && jh->b_transaction != transaction) { | |
678 | JBUFFER_TRACE(jh, "owned by older transaction"); | |
679 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
680 | J_ASSERT_JH(jh, jh->b_transaction == | |
681 | journal->j_committing_transaction); | |
682 | ||
683 | /* There is one case we have to be very careful about. | |
684 | * If the committing transaction is currently writing | |
685 | * this buffer out to disk and has NOT made a copy-out, | |
686 | * then we cannot modify the buffer contents at all | |
687 | * right now. The essence of copy-out is that it is the | |
688 | * extra copy, not the primary copy, which gets | |
689 | * journaled. If the primary copy is already going to | |
690 | * disk then we cannot do copy-out here. */ | |
691 | ||
692 | if (jh->b_jlist == BJ_Shadow) { | |
693 | DEFINE_WAIT_BIT(wait, &bh->b_state, BH_Unshadow); | |
694 | wait_queue_head_t *wqh; | |
695 | ||
696 | wqh = bit_waitqueue(&bh->b_state, BH_Unshadow); | |
697 | ||
698 | JBUFFER_TRACE(jh, "on shadow: sleep"); | |
699 | jbd_unlock_bh_state(bh); | |
700 | /* commit wakes up all shadow buffers after IO */ | |
701 | for ( ; ; ) { | |
702 | prepare_to_wait(wqh, &wait.wait, | |
703 | TASK_UNINTERRUPTIBLE); | |
704 | if (jh->b_jlist != BJ_Shadow) | |
705 | break; | |
706 | schedule(); | |
707 | } | |
708 | finish_wait(wqh, &wait.wait); | |
709 | goto repeat; | |
710 | } | |
711 | ||
712 | /* Only do the copy if the currently-owning transaction | |
713 | * still needs it. If it is on the Forget list, the | |
714 | * committing transaction is past that stage. The | |
715 | * buffer had better remain locked during the kmalloc, | |
716 | * but that should be true --- we hold the journal lock | |
717 | * still and the buffer is already on the BUF_JOURNAL | |
718 | * list so won't be flushed. | |
719 | * | |
720 | * Subtle point, though: if this is a get_undo_access, | |
721 | * then we will be relying on the frozen_data to contain | |
722 | * the new value of the committed_data record after the | |
723 | * transaction, so we HAVE to force the frozen_data copy | |
724 | * in that case. */ | |
725 | ||
726 | if (jh->b_jlist != BJ_Forget || force_copy) { | |
727 | JBUFFER_TRACE(jh, "generate frozen data"); | |
728 | if (!frozen_buffer) { | |
729 | JBUFFER_TRACE(jh, "allocate memory for buffer"); | |
730 | jbd_unlock_bh_state(bh); | |
731 | frozen_buffer = | |
af1e76d6 | 732 | jbd2_alloc(jh2bh(jh)->b_size, |
470decc6 DK |
733 | GFP_NOFS); |
734 | if (!frozen_buffer) { | |
735 | printk(KERN_EMERG | |
736 | "%s: OOM for frozen_buffer\n", | |
329d291f | 737 | __func__); |
470decc6 DK |
738 | JBUFFER_TRACE(jh, "oom!"); |
739 | error = -ENOMEM; | |
740 | jbd_lock_bh_state(bh); | |
741 | goto done; | |
742 | } | |
743 | goto repeat; | |
744 | } | |
745 | jh->b_frozen_data = frozen_buffer; | |
746 | frozen_buffer = NULL; | |
747 | need_copy = 1; | |
748 | } | |
749 | jh->b_next_transaction = transaction; | |
750 | } | |
751 | ||
752 | ||
753 | /* | |
754 | * Finally, if the buffer is not journaled right now, we need to make | |
755 | * sure it doesn't get written to disk before the caller actually | |
756 | * commits the new data | |
757 | */ | |
758 | if (!jh->b_transaction) { | |
759 | JBUFFER_TRACE(jh, "no transaction"); | |
760 | J_ASSERT_JH(jh, !jh->b_next_transaction); | |
761 | jh->b_transaction = transaction; | |
762 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); | |
763 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 764 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
470decc6 DK |
765 | spin_unlock(&journal->j_list_lock); |
766 | } | |
767 | ||
768 | done: | |
769 | if (need_copy) { | |
770 | struct page *page; | |
771 | int offset; | |
772 | char *source; | |
773 | ||
774 | J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)), | |
775 | "Possible IO failure.\n"); | |
776 | page = jh2bh(jh)->b_page; | |
777 | offset = ((unsigned long) jh2bh(jh)->b_data) & ~PAGE_MASK; | |
778 | source = kmap_atomic(page, KM_USER0); | |
13ceef09 JK |
779 | /* Fire data frozen trigger just before we copy the data */ |
780 | jbd2_buffer_frozen_trigger(jh, source + offset, | |
781 | jh->b_triggers); | |
470decc6 DK |
782 | memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size); |
783 | kunmap_atomic(source, KM_USER0); | |
e06c8227 JB |
784 | |
785 | /* | |
786 | * Now that the frozen data is saved off, we need to store | |
787 | * any matching triggers. | |
788 | */ | |
789 | jh->b_frozen_triggers = jh->b_triggers; | |
470decc6 DK |
790 | } |
791 | jbd_unlock_bh_state(bh); | |
792 | ||
793 | /* | |
794 | * If we are about to journal a buffer, then any revoke pending on it is | |
795 | * no longer valid | |
796 | */ | |
f7f4bccb | 797 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 DK |
798 | |
799 | out: | |
800 | if (unlikely(frozen_buffer)) /* It's usually NULL */ | |
af1e76d6 | 801 | jbd2_free(frozen_buffer, bh->b_size); |
470decc6 DK |
802 | |
803 | JBUFFER_TRACE(jh, "exit"); | |
804 | return error; | |
805 | } | |
806 | ||
807 | /** | |
f7f4bccb | 808 | * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update. |
470decc6 DK |
809 | * @handle: transaction to add buffer modifications to |
810 | * @bh: bh to be used for metadata writes | |
811 | * @credits: variable that will receive credits for the buffer | |
812 | * | |
813 | * Returns an error code or 0 on success. | |
814 | * | |
815 | * In full data journalling mode the buffer may be of type BJ_AsyncData, | |
816 | * because we're write()ing a buffer which is also part of a shared mapping. | |
817 | */ | |
818 | ||
f7f4bccb | 819 | int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh) |
470decc6 | 820 | { |
f7f4bccb | 821 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
822 | int rc; |
823 | ||
824 | /* We do not want to get caught playing with fields which the | |
825 | * log thread also manipulates. Make sure that the buffer | |
826 | * completes any outstanding IO before proceeding. */ | |
827 | rc = do_get_write_access(handle, jh, 0); | |
f7f4bccb | 828 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
829 | return rc; |
830 | } | |
831 | ||
832 | ||
833 | /* | |
834 | * When the user wants to journal a newly created buffer_head | |
835 | * (ie. getblk() returned a new buffer and we are going to populate it | |
836 | * manually rather than reading off disk), then we need to keep the | |
837 | * buffer_head locked until it has been completely filled with new | |
838 | * data. In this case, we should be able to make the assertion that | |
839 | * the bh is not already part of an existing transaction. | |
840 | * | |
841 | * The buffer should already be locked by the caller by this point. | |
842 | * There is no lock ranking violation: it was a newly created, | |
843 | * unlocked buffer beforehand. */ | |
844 | ||
845 | /** | |
f7f4bccb | 846 | * int jbd2_journal_get_create_access () - notify intent to use newly created bh |
470decc6 DK |
847 | * @handle: transaction to new buffer to |
848 | * @bh: new buffer. | |
849 | * | |
850 | * Call this if you create a new bh. | |
851 | */ | |
f7f4bccb | 852 | int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
853 | { |
854 | transaction_t *transaction = handle->h_transaction; | |
855 | journal_t *journal = transaction->t_journal; | |
f7f4bccb | 856 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
857 | int err; |
858 | ||
859 | jbd_debug(5, "journal_head %p\n", jh); | |
860 | err = -EROFS; | |
861 | if (is_handle_aborted(handle)) | |
862 | goto out; | |
863 | err = 0; | |
864 | ||
865 | JBUFFER_TRACE(jh, "entry"); | |
866 | /* | |
867 | * The buffer may already belong to this transaction due to pre-zeroing | |
868 | * in the filesystem's new_block code. It may also be on the previous, | |
869 | * committing transaction's lists, but it HAS to be in Forget state in | |
870 | * that case: the transaction must have deleted the buffer for it to be | |
871 | * reused here. | |
872 | */ | |
873 | jbd_lock_bh_state(bh); | |
874 | spin_lock(&journal->j_list_lock); | |
875 | J_ASSERT_JH(jh, (jh->b_transaction == transaction || | |
876 | jh->b_transaction == NULL || | |
877 | (jh->b_transaction == journal->j_committing_transaction && | |
878 | jh->b_jlist == BJ_Forget))); | |
879 | ||
880 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
881 | J_ASSERT_JH(jh, buffer_locked(jh2bh(jh))); | |
882 | ||
883 | if (jh->b_transaction == NULL) { | |
f91d1d04 JK |
884 | /* |
885 | * Previous jbd2_journal_forget() could have left the buffer | |
886 | * with jbddirty bit set because it was being committed. When | |
887 | * the commit finished, we've filed the buffer for | |
888 | * checkpointing and marked it dirty. Now we are reallocating | |
889 | * the buffer so the transaction freeing it must have | |
890 | * committed and so it's safe to clear the dirty bit. | |
891 | */ | |
892 | clear_buffer_dirty(jh2bh(jh)); | |
470decc6 | 893 | jh->b_transaction = transaction; |
9fc7c63a JB |
894 | |
895 | /* first access by this transaction */ | |
896 | jh->b_modified = 0; | |
897 | ||
470decc6 | 898 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); |
f7f4bccb | 899 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
470decc6 | 900 | } else if (jh->b_transaction == journal->j_committing_transaction) { |
9fc7c63a JB |
901 | /* first access by this transaction */ |
902 | jh->b_modified = 0; | |
903 | ||
470decc6 DK |
904 | JBUFFER_TRACE(jh, "set next transaction"); |
905 | jh->b_next_transaction = transaction; | |
906 | } | |
907 | spin_unlock(&journal->j_list_lock); | |
908 | jbd_unlock_bh_state(bh); | |
909 | ||
910 | /* | |
911 | * akpm: I added this. ext3_alloc_branch can pick up new indirect | |
912 | * blocks which contain freed but then revoked metadata. We need | |
913 | * to cancel the revoke in case we end up freeing it yet again | |
914 | * and the reallocating as data - this would cause a second revoke, | |
915 | * which hits an assertion error. | |
916 | */ | |
917 | JBUFFER_TRACE(jh, "cancelling revoke"); | |
f7f4bccb MC |
918 | jbd2_journal_cancel_revoke(handle, jh); |
919 | jbd2_journal_put_journal_head(jh); | |
470decc6 DK |
920 | out: |
921 | return err; | |
922 | } | |
923 | ||
924 | /** | |
f7f4bccb | 925 | * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with |
470decc6 DK |
926 | * non-rewindable consequences |
927 | * @handle: transaction | |
928 | * @bh: buffer to undo | |
929 | * @credits: store the number of taken credits here (if not NULL) | |
930 | * | |
931 | * Sometimes there is a need to distinguish between metadata which has | |
932 | * been committed to disk and that which has not. The ext3fs code uses | |
933 | * this for freeing and allocating space, we have to make sure that we | |
934 | * do not reuse freed space until the deallocation has been committed, | |
935 | * since if we overwrote that space we would make the delete | |
936 | * un-rewindable in case of a crash. | |
937 | * | |
f7f4bccb | 938 | * To deal with that, jbd2_journal_get_undo_access requests write access to a |
470decc6 DK |
939 | * buffer for parts of non-rewindable operations such as delete |
940 | * operations on the bitmaps. The journaling code must keep a copy of | |
941 | * the buffer's contents prior to the undo_access call until such time | |
942 | * as we know that the buffer has definitely been committed to disk. | |
943 | * | |
944 | * We never need to know which transaction the committed data is part | |
945 | * of, buffers touched here are guaranteed to be dirtied later and so | |
946 | * will be committed to a new transaction in due course, at which point | |
947 | * we can discard the old committed data pointer. | |
948 | * | |
949 | * Returns error number or 0 on success. | |
950 | */ | |
f7f4bccb | 951 | int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
952 | { |
953 | int err; | |
f7f4bccb | 954 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
955 | char *committed_data = NULL; |
956 | ||
957 | JBUFFER_TRACE(jh, "entry"); | |
958 | ||
959 | /* | |
960 | * Do this first --- it can drop the journal lock, so we want to | |
961 | * make sure that obtaining the committed_data is done | |
962 | * atomically wrt. completion of any outstanding commits. | |
963 | */ | |
964 | err = do_get_write_access(handle, jh, 1); | |
965 | if (err) | |
966 | goto out; | |
967 | ||
968 | repeat: | |
969 | if (!jh->b_committed_data) { | |
af1e76d6 | 970 | committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS); |
470decc6 DK |
971 | if (!committed_data) { |
972 | printk(KERN_EMERG "%s: No memory for committed data\n", | |
329d291f | 973 | __func__); |
470decc6 DK |
974 | err = -ENOMEM; |
975 | goto out; | |
976 | } | |
977 | } | |
978 | ||
979 | jbd_lock_bh_state(bh); | |
980 | if (!jh->b_committed_data) { | |
981 | /* Copy out the current buffer contents into the | |
982 | * preserved, committed copy. */ | |
983 | JBUFFER_TRACE(jh, "generate b_committed data"); | |
984 | if (!committed_data) { | |
985 | jbd_unlock_bh_state(bh); | |
986 | goto repeat; | |
987 | } | |
988 | ||
989 | jh->b_committed_data = committed_data; | |
990 | committed_data = NULL; | |
991 | memcpy(jh->b_committed_data, bh->b_data, bh->b_size); | |
992 | } | |
993 | jbd_unlock_bh_state(bh); | |
994 | out: | |
f7f4bccb | 995 | jbd2_journal_put_journal_head(jh); |
470decc6 | 996 | if (unlikely(committed_data)) |
af1e76d6 | 997 | jbd2_free(committed_data, bh->b_size); |
470decc6 DK |
998 | return err; |
999 | } | |
1000 | ||
e06c8227 JB |
1001 | /** |
1002 | * void jbd2_journal_set_triggers() - Add triggers for commit writeout | |
1003 | * @bh: buffer to trigger on | |
1004 | * @type: struct jbd2_buffer_trigger_type containing the trigger(s). | |
1005 | * | |
1006 | * Set any triggers on this journal_head. This is always safe, because | |
1007 | * triggers for a committing buffer will be saved off, and triggers for | |
1008 | * a running transaction will match the buffer in that transaction. | |
1009 | * | |
1010 | * Call with NULL to clear the triggers. | |
1011 | */ | |
1012 | void jbd2_journal_set_triggers(struct buffer_head *bh, | |
1013 | struct jbd2_buffer_trigger_type *type) | |
1014 | { | |
1015 | struct journal_head *jh = bh2jh(bh); | |
1016 | ||
1017 | jh->b_triggers = type; | |
1018 | } | |
1019 | ||
13ceef09 | 1020 | void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data, |
e06c8227 JB |
1021 | struct jbd2_buffer_trigger_type *triggers) |
1022 | { | |
1023 | struct buffer_head *bh = jh2bh(jh); | |
1024 | ||
13ceef09 | 1025 | if (!triggers || !triggers->t_frozen) |
e06c8227 JB |
1026 | return; |
1027 | ||
13ceef09 | 1028 | triggers->t_frozen(triggers, bh, mapped_data, bh->b_size); |
e06c8227 JB |
1029 | } |
1030 | ||
1031 | void jbd2_buffer_abort_trigger(struct journal_head *jh, | |
1032 | struct jbd2_buffer_trigger_type *triggers) | |
1033 | { | |
1034 | if (!triggers || !triggers->t_abort) | |
1035 | return; | |
1036 | ||
1037 | triggers->t_abort(triggers, jh2bh(jh)); | |
1038 | } | |
1039 | ||
1040 | ||
1041 | ||
470decc6 | 1042 | /** |
f7f4bccb | 1043 | * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata |
470decc6 DK |
1044 | * @handle: transaction to add buffer to. |
1045 | * @bh: buffer to mark | |
1046 | * | |
1047 | * mark dirty metadata which needs to be journaled as part of the current | |
1048 | * transaction. | |
1049 | * | |
1050 | * The buffer is placed on the transaction's metadata list and is marked | |
1051 | * as belonging to the transaction. | |
1052 | * | |
1053 | * Returns error number or 0 on success. | |
1054 | * | |
1055 | * Special care needs to be taken if the buffer already belongs to the | |
1056 | * current committing transaction (in which case we should have frozen | |
1057 | * data present for that commit). In that case, we don't relink the | |
1058 | * buffer: that only gets done when the old transaction finally | |
1059 | * completes its commit. | |
1060 | */ | |
f7f4bccb | 1061 | int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1062 | { |
1063 | transaction_t *transaction = handle->h_transaction; | |
1064 | journal_t *journal = transaction->t_journal; | |
1065 | struct journal_head *jh = bh2jh(bh); | |
1066 | ||
1067 | jbd_debug(5, "journal_head %p\n", jh); | |
1068 | JBUFFER_TRACE(jh, "entry"); | |
1069 | if (is_handle_aborted(handle)) | |
1070 | goto out; | |
1071 | ||
1072 | jbd_lock_bh_state(bh); | |
1073 | ||
1074 | if (jh->b_modified == 0) { | |
1075 | /* | |
1076 | * This buffer's got modified and becoming part | |
1077 | * of the transaction. This needs to be done | |
1078 | * once a transaction -bzzz | |
1079 | */ | |
1080 | jh->b_modified = 1; | |
1081 | J_ASSERT_JH(jh, handle->h_buffer_credits > 0); | |
1082 | handle->h_buffer_credits--; | |
1083 | } | |
1084 | ||
1085 | /* | |
1086 | * fastpath, to avoid expensive locking. If this buffer is already | |
1087 | * on the running transaction's metadata list there is nothing to do. | |
1088 | * Nobody can take it off again because there is a handle open. | |
1089 | * I _think_ we're OK here with SMP barriers - a mistaken decision will | |
1090 | * result in this test being false, so we go in and take the locks. | |
1091 | */ | |
1092 | if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) { | |
1093 | JBUFFER_TRACE(jh, "fastpath"); | |
1094 | J_ASSERT_JH(jh, jh->b_transaction == | |
1095 | journal->j_running_transaction); | |
1096 | goto out_unlock_bh; | |
1097 | } | |
1098 | ||
1099 | set_buffer_jbddirty(bh); | |
1100 | ||
1101 | /* | |
1102 | * Metadata already on the current transaction list doesn't | |
1103 | * need to be filed. Metadata on another transaction's list must | |
1104 | * be committing, and will be refiled once the commit completes: | |
1105 | * leave it alone for now. | |
1106 | */ | |
1107 | if (jh->b_transaction != transaction) { | |
1108 | JBUFFER_TRACE(jh, "already on other transaction"); | |
1109 | J_ASSERT_JH(jh, jh->b_transaction == | |
1110 | journal->j_committing_transaction); | |
1111 | J_ASSERT_JH(jh, jh->b_next_transaction == transaction); | |
1112 | /* And this case is illegal: we can't reuse another | |
1113 | * transaction's data buffer, ever. */ | |
1114 | goto out_unlock_bh; | |
1115 | } | |
1116 | ||
1117 | /* That test should have eliminated the following case: */ | |
4019191b | 1118 | J_ASSERT_JH(jh, jh->b_frozen_data == NULL); |
470decc6 DK |
1119 | |
1120 | JBUFFER_TRACE(jh, "file as BJ_Metadata"); | |
1121 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 1122 | __jbd2_journal_file_buffer(jh, handle->h_transaction, BJ_Metadata); |
470decc6 DK |
1123 | spin_unlock(&journal->j_list_lock); |
1124 | out_unlock_bh: | |
1125 | jbd_unlock_bh_state(bh); | |
1126 | out: | |
1127 | JBUFFER_TRACE(jh, "exit"); | |
1128 | return 0; | |
1129 | } | |
1130 | ||
1131 | /* | |
f7f4bccb | 1132 | * jbd2_journal_release_buffer: undo a get_write_access without any buffer |
470decc6 DK |
1133 | * updates, if the update decided in the end that it didn't need access. |
1134 | * | |
1135 | */ | |
1136 | void | |
f7f4bccb | 1137 | jbd2_journal_release_buffer(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1138 | { |
1139 | BUFFER_TRACE(bh, "entry"); | |
1140 | } | |
1141 | ||
1142 | /** | |
f7f4bccb | 1143 | * void jbd2_journal_forget() - bforget() for potentially-journaled buffers. |
470decc6 DK |
1144 | * @handle: transaction handle |
1145 | * @bh: bh to 'forget' | |
1146 | * | |
1147 | * We can only do the bforget if there are no commits pending against the | |
1148 | * buffer. If the buffer is dirty in the current running transaction we | |
1149 | * can safely unlink it. | |
1150 | * | |
1151 | * bh may not be a journalled buffer at all - it may be a non-JBD | |
1152 | * buffer which came off the hashtable. Check for this. | |
1153 | * | |
1154 | * Decrements bh->b_count by one. | |
1155 | * | |
1156 | * Allow this call even if the handle has aborted --- it may be part of | |
1157 | * the caller's cleanup after an abort. | |
1158 | */ | |
f7f4bccb | 1159 | int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1160 | { |
1161 | transaction_t *transaction = handle->h_transaction; | |
1162 | journal_t *journal = transaction->t_journal; | |
1163 | struct journal_head *jh; | |
1164 | int drop_reserve = 0; | |
1165 | int err = 0; | |
1dfc3220 | 1166 | int was_modified = 0; |
470decc6 DK |
1167 | |
1168 | BUFFER_TRACE(bh, "entry"); | |
1169 | ||
1170 | jbd_lock_bh_state(bh); | |
1171 | spin_lock(&journal->j_list_lock); | |
1172 | ||
1173 | if (!buffer_jbd(bh)) | |
1174 | goto not_jbd; | |
1175 | jh = bh2jh(bh); | |
1176 | ||
1177 | /* Critical error: attempting to delete a bitmap buffer, maybe? | |
1178 | * Don't do any jbd operations, and return an error. */ | |
1179 | if (!J_EXPECT_JH(jh, !jh->b_committed_data, | |
1180 | "inconsistent data on disk")) { | |
1181 | err = -EIO; | |
1182 | goto not_jbd; | |
1183 | } | |
1184 | ||
1dfc3220 JB |
1185 | /* keep track of wether or not this transaction modified us */ |
1186 | was_modified = jh->b_modified; | |
1187 | ||
470decc6 DK |
1188 | /* |
1189 | * The buffer's going from the transaction, we must drop | |
1190 | * all references -bzzz | |
1191 | */ | |
1192 | jh->b_modified = 0; | |
1193 | ||
1194 | if (jh->b_transaction == handle->h_transaction) { | |
1195 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
1196 | ||
1197 | /* If we are forgetting a buffer which is already part | |
1198 | * of this transaction, then we can just drop it from | |
1199 | * the transaction immediately. */ | |
1200 | clear_buffer_dirty(bh); | |
1201 | clear_buffer_jbddirty(bh); | |
1202 | ||
1203 | JBUFFER_TRACE(jh, "belongs to current transaction: unfile"); | |
1204 | ||
1dfc3220 JB |
1205 | /* |
1206 | * we only want to drop a reference if this transaction | |
1207 | * modified the buffer | |
1208 | */ | |
1209 | if (was_modified) | |
1210 | drop_reserve = 1; | |
470decc6 DK |
1211 | |
1212 | /* | |
1213 | * We are no longer going to journal this buffer. | |
1214 | * However, the commit of this transaction is still | |
1215 | * important to the buffer: the delete that we are now | |
1216 | * processing might obsolete an old log entry, so by | |
1217 | * committing, we can satisfy the buffer's checkpoint. | |
1218 | * | |
1219 | * So, if we have a checkpoint on the buffer, we should | |
1220 | * now refile the buffer on our BJ_Forget list so that | |
1221 | * we know to remove the checkpoint after we commit. | |
1222 | */ | |
1223 | ||
1224 | if (jh->b_cp_transaction) { | |
f7f4bccb MC |
1225 | __jbd2_journal_temp_unlink_buffer(jh); |
1226 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); | |
470decc6 | 1227 | } else { |
f7f4bccb MC |
1228 | __jbd2_journal_unfile_buffer(jh); |
1229 | jbd2_journal_remove_journal_head(bh); | |
470decc6 DK |
1230 | __brelse(bh); |
1231 | if (!buffer_jbd(bh)) { | |
1232 | spin_unlock(&journal->j_list_lock); | |
1233 | jbd_unlock_bh_state(bh); | |
1234 | __bforget(bh); | |
1235 | goto drop; | |
1236 | } | |
1237 | } | |
1238 | } else if (jh->b_transaction) { | |
1239 | J_ASSERT_JH(jh, (jh->b_transaction == | |
1240 | journal->j_committing_transaction)); | |
1241 | /* However, if the buffer is still owned by a prior | |
1242 | * (committing) transaction, we can't drop it yet... */ | |
1243 | JBUFFER_TRACE(jh, "belongs to older transaction"); | |
1244 | /* ... but we CAN drop it from the new transaction if we | |
1245 | * have also modified it since the original commit. */ | |
1246 | ||
1247 | if (jh->b_next_transaction) { | |
1248 | J_ASSERT(jh->b_next_transaction == transaction); | |
1249 | jh->b_next_transaction = NULL; | |
1dfc3220 JB |
1250 | |
1251 | /* | |
1252 | * only drop a reference if this transaction modified | |
1253 | * the buffer | |
1254 | */ | |
1255 | if (was_modified) | |
1256 | drop_reserve = 1; | |
470decc6 DK |
1257 | } |
1258 | } | |
1259 | ||
1260 | not_jbd: | |
1261 | spin_unlock(&journal->j_list_lock); | |
1262 | jbd_unlock_bh_state(bh); | |
1263 | __brelse(bh); | |
1264 | drop: | |
1265 | if (drop_reserve) { | |
1266 | /* no need to reserve log space for this block -bzzz */ | |
1267 | handle->h_buffer_credits++; | |
1268 | } | |
1269 | return err; | |
1270 | } | |
1271 | ||
1272 | /** | |
f7f4bccb | 1273 | * int jbd2_journal_stop() - complete a transaction |
470decc6 DK |
1274 | * @handle: tranaction to complete. |
1275 | * | |
1276 | * All done for a particular handle. | |
1277 | * | |
1278 | * There is not much action needed here. We just return any remaining | |
1279 | * buffer credits to the transaction and remove the handle. The only | |
1280 | * complication is that we need to start a commit operation if the | |
1281 | * filesystem is marked for synchronous update. | |
1282 | * | |
f7f4bccb | 1283 | * jbd2_journal_stop itself will not usually return an error, but it may |
470decc6 | 1284 | * do so in unusual circumstances. In particular, expect it to |
f7f4bccb | 1285 | * return -EIO if a jbd2_journal_abort has been executed since the |
470decc6 DK |
1286 | * transaction began. |
1287 | */ | |
f7f4bccb | 1288 | int jbd2_journal_stop(handle_t *handle) |
470decc6 DK |
1289 | { |
1290 | transaction_t *transaction = handle->h_transaction; | |
1291 | journal_t *journal = transaction->t_journal; | |
a51dca9c TT |
1292 | int err, wait_for_commit = 0; |
1293 | tid_t tid; | |
470decc6 DK |
1294 | pid_t pid; |
1295 | ||
470decc6 DK |
1296 | J_ASSERT(journal_current_handle() == handle); |
1297 | ||
1298 | if (is_handle_aborted(handle)) | |
1299 | err = -EIO; | |
3e2a532b | 1300 | else { |
a51dca9c | 1301 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 | 1302 | err = 0; |
3e2a532b | 1303 | } |
470decc6 DK |
1304 | |
1305 | if (--handle->h_ref > 0) { | |
1306 | jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1, | |
1307 | handle->h_ref); | |
1308 | return err; | |
1309 | } | |
1310 | ||
1311 | jbd_debug(4, "Handle %p going down\n", handle); | |
1312 | ||
1313 | /* | |
1314 | * Implement synchronous transaction batching. If the handle | |
1315 | * was synchronous, don't force a commit immediately. Let's | |
e07f7183 JB |
1316 | * yield and let another thread piggyback onto this |
1317 | * transaction. Keep doing that while new threads continue to | |
1318 | * arrive. It doesn't cost much - we're about to run a commit | |
1319 | * and sleep on IO anyway. Speeds up many-threaded, many-dir | |
1320 | * operations by 30x or more... | |
1321 | * | |
1322 | * We try and optimize the sleep time against what the | |
1323 | * underlying disk can do, instead of having a static sleep | |
1324 | * time. This is useful for the case where our storage is so | |
1325 | * fast that it is more optimal to go ahead and force a flush | |
1326 | * and wait for the transaction to be committed than it is to | |
1327 | * wait for an arbitrary amount of time for new writers to | |
1328 | * join the transaction. We achieve this by measuring how | |
1329 | * long it takes to commit a transaction, and compare it with | |
1330 | * how long this transaction has been running, and if run time | |
1331 | * < commit time then we sleep for the delta and commit. This | |
1332 | * greatly helps super fast disks that would see slowdowns as | |
1333 | * more threads started doing fsyncs. | |
470decc6 | 1334 | * |
e07f7183 JB |
1335 | * But don't do this if this process was the most recent one |
1336 | * to perform a synchronous write. We do this to detect the | |
1337 | * case where a single process is doing a stream of sync | |
1338 | * writes. No point in waiting for joiners in that case. | |
470decc6 DK |
1339 | */ |
1340 | pid = current->pid; | |
1341 | if (handle->h_sync && journal->j_last_sync_writer != pid) { | |
e07f7183 JB |
1342 | u64 commit_time, trans_time; |
1343 | ||
470decc6 | 1344 | journal->j_last_sync_writer = pid; |
e07f7183 | 1345 | |
a931da6a | 1346 | read_lock(&journal->j_state_lock); |
e07f7183 | 1347 | commit_time = journal->j_average_commit_time; |
a931da6a | 1348 | read_unlock(&journal->j_state_lock); |
e07f7183 JB |
1349 | |
1350 | trans_time = ktime_to_ns(ktime_sub(ktime_get(), | |
1351 | transaction->t_start_time)); | |
1352 | ||
30773840 TT |
1353 | commit_time = max_t(u64, commit_time, |
1354 | 1000*journal->j_min_batch_time); | |
e07f7183 | 1355 | commit_time = min_t(u64, commit_time, |
30773840 | 1356 | 1000*journal->j_max_batch_time); |
e07f7183 JB |
1357 | |
1358 | if (trans_time < commit_time) { | |
1359 | ktime_t expires = ktime_add_ns(ktime_get(), | |
1360 | commit_time); | |
1361 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1362 | schedule_hrtimeout(&expires, HRTIMER_MODE_ABS); | |
1363 | } | |
470decc6 DK |
1364 | } |
1365 | ||
7058548c TT |
1366 | if (handle->h_sync) |
1367 | transaction->t_synchronous_commit = 1; | |
470decc6 | 1368 | current->journal_info = NULL; |
a51dca9c TT |
1369 | atomic_sub(handle->h_buffer_credits, |
1370 | &transaction->t_outstanding_credits); | |
470decc6 DK |
1371 | |
1372 | /* | |
1373 | * If the handle is marked SYNC, we need to set another commit | |
1374 | * going! We also want to force a commit if the current | |
1375 | * transaction is occupying too much of the log, or if the | |
1376 | * transaction is too old now. | |
1377 | */ | |
1378 | if (handle->h_sync || | |
a51dca9c TT |
1379 | (atomic_read(&transaction->t_outstanding_credits) > |
1380 | journal->j_max_transaction_buffers) || | |
1381 | time_after_eq(jiffies, transaction->t_expires)) { | |
470decc6 DK |
1382 | /* Do this even for aborted journals: an abort still |
1383 | * completes the commit thread, it just doesn't write | |
1384 | * anything to disk. */ | |
470decc6 | 1385 | |
470decc6 DK |
1386 | jbd_debug(2, "transaction too old, requesting commit for " |
1387 | "handle %p\n", handle); | |
1388 | /* This is non-blocking */ | |
c35a56a0 | 1389 | jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1390 | |
1391 | /* | |
f7f4bccb | 1392 | * Special case: JBD2_SYNC synchronous updates require us |
470decc6 DK |
1393 | * to wait for the commit to complete. |
1394 | */ | |
1395 | if (handle->h_sync && !(current->flags & PF_MEMALLOC)) | |
a51dca9c | 1396 | wait_for_commit = 1; |
470decc6 DK |
1397 | } |
1398 | ||
a51dca9c TT |
1399 | /* |
1400 | * Once we drop t_updates, if it goes to zero the transaction | |
1401 | * could start commiting on us and eventually disappear. So | |
1402 | * once we do this, we must not dereference transaction | |
1403 | * pointer again. | |
1404 | */ | |
1405 | tid = transaction->t_tid; | |
1406 | if (atomic_dec_and_test(&transaction->t_updates)) { | |
1407 | wake_up(&journal->j_wait_updates); | |
1408 | if (journal->j_barrier_count) | |
1409 | wake_up(&journal->j_wait_transaction_locked); | |
1410 | } | |
1411 | ||
1412 | if (wait_for_commit) | |
1413 | err = jbd2_log_wait_commit(journal, tid); | |
1414 | ||
3295f0ef | 1415 | lock_map_release(&handle->h_lockdep_map); |
7b751066 | 1416 | |
af1e76d6 | 1417 | jbd2_free_handle(handle); |
470decc6 DK |
1418 | return err; |
1419 | } | |
1420 | ||
5648ba5b RD |
1421 | /** |
1422 | * int jbd2_journal_force_commit() - force any uncommitted transactions | |
470decc6 DK |
1423 | * @journal: journal to force |
1424 | * | |
1425 | * For synchronous operations: force any uncommitted transactions | |
1426 | * to disk. May seem kludgy, but it reuses all the handle batching | |
1427 | * code in a very simple manner. | |
1428 | */ | |
f7f4bccb | 1429 | int jbd2_journal_force_commit(journal_t *journal) |
470decc6 DK |
1430 | { |
1431 | handle_t *handle; | |
1432 | int ret; | |
1433 | ||
f7f4bccb | 1434 | handle = jbd2_journal_start(journal, 1); |
470decc6 DK |
1435 | if (IS_ERR(handle)) { |
1436 | ret = PTR_ERR(handle); | |
1437 | } else { | |
1438 | handle->h_sync = 1; | |
f7f4bccb | 1439 | ret = jbd2_journal_stop(handle); |
470decc6 DK |
1440 | } |
1441 | return ret; | |
1442 | } | |
1443 | ||
1444 | /* | |
1445 | * | |
1446 | * List management code snippets: various functions for manipulating the | |
1447 | * transaction buffer lists. | |
1448 | * | |
1449 | */ | |
1450 | ||
1451 | /* | |
1452 | * Append a buffer to a transaction list, given the transaction's list head | |
1453 | * pointer. | |
1454 | * | |
1455 | * j_list_lock is held. | |
1456 | * | |
1457 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1458 | */ | |
1459 | ||
1460 | static inline void | |
1461 | __blist_add_buffer(struct journal_head **list, struct journal_head *jh) | |
1462 | { | |
1463 | if (!*list) { | |
1464 | jh->b_tnext = jh->b_tprev = jh; | |
1465 | *list = jh; | |
1466 | } else { | |
1467 | /* Insert at the tail of the list to preserve order */ | |
1468 | struct journal_head *first = *list, *last = first->b_tprev; | |
1469 | jh->b_tprev = last; | |
1470 | jh->b_tnext = first; | |
1471 | last->b_tnext = first->b_tprev = jh; | |
1472 | } | |
1473 | } | |
1474 | ||
1475 | /* | |
1476 | * Remove a buffer from a transaction list, given the transaction's list | |
1477 | * head pointer. | |
1478 | * | |
1479 | * Called with j_list_lock held, and the journal may not be locked. | |
1480 | * | |
1481 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1482 | */ | |
1483 | ||
1484 | static inline void | |
1485 | __blist_del_buffer(struct journal_head **list, struct journal_head *jh) | |
1486 | { | |
1487 | if (*list == jh) { | |
1488 | *list = jh->b_tnext; | |
1489 | if (*list == jh) | |
1490 | *list = NULL; | |
1491 | } | |
1492 | jh->b_tprev->b_tnext = jh->b_tnext; | |
1493 | jh->b_tnext->b_tprev = jh->b_tprev; | |
1494 | } | |
1495 | ||
1496 | /* | |
1497 | * Remove a buffer from the appropriate transaction list. | |
1498 | * | |
1499 | * Note that this function can *change* the value of | |
87c89c23 JK |
1500 | * bh->b_transaction->t_buffers, t_forget, t_iobuf_list, t_shadow_list, |
1501 | * t_log_list or t_reserved_list. If the caller is holding onto a copy of one | |
1502 | * of these pointers, it could go bad. Generally the caller needs to re-read | |
1503 | * the pointer from the transaction_t. | |
470decc6 DK |
1504 | * |
1505 | * Called under j_list_lock. The journal may not be locked. | |
1506 | */ | |
f7f4bccb | 1507 | void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh) |
470decc6 DK |
1508 | { |
1509 | struct journal_head **list = NULL; | |
1510 | transaction_t *transaction; | |
1511 | struct buffer_head *bh = jh2bh(jh); | |
1512 | ||
1513 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
1514 | transaction = jh->b_transaction; | |
1515 | if (transaction) | |
1516 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
1517 | ||
1518 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
1519 | if (jh->b_jlist != BJ_None) | |
4019191b | 1520 | J_ASSERT_JH(jh, transaction != NULL); |
470decc6 DK |
1521 | |
1522 | switch (jh->b_jlist) { | |
1523 | case BJ_None: | |
1524 | return; | |
470decc6 DK |
1525 | case BJ_Metadata: |
1526 | transaction->t_nr_buffers--; | |
1527 | J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0); | |
1528 | list = &transaction->t_buffers; | |
1529 | break; | |
1530 | case BJ_Forget: | |
1531 | list = &transaction->t_forget; | |
1532 | break; | |
1533 | case BJ_IO: | |
1534 | list = &transaction->t_iobuf_list; | |
1535 | break; | |
1536 | case BJ_Shadow: | |
1537 | list = &transaction->t_shadow_list; | |
1538 | break; | |
1539 | case BJ_LogCtl: | |
1540 | list = &transaction->t_log_list; | |
1541 | break; | |
1542 | case BJ_Reserved: | |
1543 | list = &transaction->t_reserved_list; | |
1544 | break; | |
470decc6 DK |
1545 | } |
1546 | ||
1547 | __blist_del_buffer(list, jh); | |
1548 | jh->b_jlist = BJ_None; | |
1549 | if (test_clear_buffer_jbddirty(bh)) | |
1550 | mark_buffer_dirty(bh); /* Expose it to the VM */ | |
1551 | } | |
1552 | ||
f7f4bccb | 1553 | void __jbd2_journal_unfile_buffer(struct journal_head *jh) |
470decc6 | 1554 | { |
f7f4bccb | 1555 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 DK |
1556 | jh->b_transaction = NULL; |
1557 | } | |
1558 | ||
f7f4bccb | 1559 | void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 DK |
1560 | { |
1561 | jbd_lock_bh_state(jh2bh(jh)); | |
1562 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 1563 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
1564 | spin_unlock(&journal->j_list_lock); |
1565 | jbd_unlock_bh_state(jh2bh(jh)); | |
1566 | } | |
1567 | ||
1568 | /* | |
f7f4bccb | 1569 | * Called from jbd2_journal_try_to_free_buffers(). |
470decc6 DK |
1570 | * |
1571 | * Called under jbd_lock_bh_state(bh) | |
1572 | */ | |
1573 | static void | |
1574 | __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh) | |
1575 | { | |
1576 | struct journal_head *jh; | |
1577 | ||
1578 | jh = bh2jh(bh); | |
1579 | ||
1580 | if (buffer_locked(bh) || buffer_dirty(bh)) | |
1581 | goto out; | |
1582 | ||
4019191b | 1583 | if (jh->b_next_transaction != NULL) |
470decc6 DK |
1584 | goto out; |
1585 | ||
1586 | spin_lock(&journal->j_list_lock); | |
87c89c23 | 1587 | if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) { |
470decc6 DK |
1588 | /* written-back checkpointed metadata buffer */ |
1589 | if (jh->b_jlist == BJ_None) { | |
1590 | JBUFFER_TRACE(jh, "remove from checkpoint list"); | |
f7f4bccb MC |
1591 | __jbd2_journal_remove_checkpoint(jh); |
1592 | jbd2_journal_remove_journal_head(bh); | |
470decc6 DK |
1593 | __brelse(bh); |
1594 | } | |
1595 | } | |
1596 | spin_unlock(&journal->j_list_lock); | |
1597 | out: | |
1598 | return; | |
1599 | } | |
1600 | ||
470decc6 | 1601 | /** |
f7f4bccb | 1602 | * int jbd2_journal_try_to_free_buffers() - try to free page buffers. |
470decc6 DK |
1603 | * @journal: journal for operation |
1604 | * @page: to try and free | |
530576bb MC |
1605 | * @gfp_mask: we use the mask to detect how hard should we try to release |
1606 | * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to | |
1607 | * release the buffers. | |
470decc6 DK |
1608 | * |
1609 | * | |
1610 | * For all the buffers on this page, | |
1611 | * if they are fully written out ordered data, move them onto BUF_CLEAN | |
1612 | * so try_to_free_buffers() can reap them. | |
1613 | * | |
1614 | * This function returns non-zero if we wish try_to_free_buffers() | |
1615 | * to be called. We do this if the page is releasable by try_to_free_buffers(). | |
1616 | * We also do it if the page has locked or dirty buffers and the caller wants | |
1617 | * us to perform sync or async writeout. | |
1618 | * | |
1619 | * This complicates JBD locking somewhat. We aren't protected by the | |
1620 | * BKL here. We wish to remove the buffer from its committing or | |
f7f4bccb | 1621 | * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer. |
470decc6 DK |
1622 | * |
1623 | * This may *change* the value of transaction_t->t_datalist, so anyone | |
1624 | * who looks at t_datalist needs to lock against this function. | |
1625 | * | |
f7f4bccb MC |
1626 | * Even worse, someone may be doing a jbd2_journal_dirty_data on this |
1627 | * buffer. So we need to lock against that. jbd2_journal_dirty_data() | |
470decc6 DK |
1628 | * will come out of the lock with the buffer dirty, which makes it |
1629 | * ineligible for release here. | |
1630 | * | |
1631 | * Who else is affected by this? hmm... Really the only contender | |
1632 | * is do_get_write_access() - it could be looking at the buffer while | |
1633 | * journal_try_to_free_buffer() is changing its state. But that | |
1634 | * cannot happen because we never reallocate freed data as metadata | |
1635 | * while the data is part of a transaction. Yes? | |
530576bb MC |
1636 | * |
1637 | * Return 0 on failure, 1 on success | |
470decc6 | 1638 | */ |
f7f4bccb | 1639 | int jbd2_journal_try_to_free_buffers(journal_t *journal, |
530576bb | 1640 | struct page *page, gfp_t gfp_mask) |
470decc6 DK |
1641 | { |
1642 | struct buffer_head *head; | |
1643 | struct buffer_head *bh; | |
1644 | int ret = 0; | |
1645 | ||
1646 | J_ASSERT(PageLocked(page)); | |
1647 | ||
1648 | head = page_buffers(page); | |
1649 | bh = head; | |
1650 | do { | |
1651 | struct journal_head *jh; | |
1652 | ||
1653 | /* | |
1654 | * We take our own ref against the journal_head here to avoid | |
1655 | * having to add tons of locking around each instance of | |
530576bb MC |
1656 | * jbd2_journal_remove_journal_head() and |
1657 | * jbd2_journal_put_journal_head(). | |
470decc6 | 1658 | */ |
f7f4bccb | 1659 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
1660 | if (!jh) |
1661 | continue; | |
1662 | ||
1663 | jbd_lock_bh_state(bh); | |
1664 | __journal_try_to_free_buffer(journal, bh); | |
f7f4bccb | 1665 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1666 | jbd_unlock_bh_state(bh); |
1667 | if (buffer_jbd(bh)) | |
1668 | goto busy; | |
1669 | } while ((bh = bh->b_this_page) != head); | |
530576bb | 1670 | |
470decc6 | 1671 | ret = try_to_free_buffers(page); |
530576bb | 1672 | |
470decc6 DK |
1673 | busy: |
1674 | return ret; | |
1675 | } | |
1676 | ||
1677 | /* | |
1678 | * This buffer is no longer needed. If it is on an older transaction's | |
1679 | * checkpoint list we need to record it on this transaction's forget list | |
1680 | * to pin this buffer (and hence its checkpointing transaction) down until | |
1681 | * this transaction commits. If the buffer isn't on a checkpoint list, we | |
1682 | * release it. | |
1683 | * Returns non-zero if JBD no longer has an interest in the buffer. | |
1684 | * | |
1685 | * Called under j_list_lock. | |
1686 | * | |
1687 | * Called under jbd_lock_bh_state(bh). | |
1688 | */ | |
1689 | static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction) | |
1690 | { | |
1691 | int may_free = 1; | |
1692 | struct buffer_head *bh = jh2bh(jh); | |
1693 | ||
f7f4bccb | 1694 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
1695 | |
1696 | if (jh->b_cp_transaction) { | |
1697 | JBUFFER_TRACE(jh, "on running+cp transaction"); | |
f91d1d04 JK |
1698 | /* |
1699 | * We don't want to write the buffer anymore, clear the | |
1700 | * bit so that we don't confuse checks in | |
1701 | * __journal_file_buffer | |
1702 | */ | |
1703 | clear_buffer_dirty(bh); | |
f7f4bccb | 1704 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); |
470decc6 DK |
1705 | may_free = 0; |
1706 | } else { | |
1707 | JBUFFER_TRACE(jh, "on running transaction"); | |
f7f4bccb | 1708 | jbd2_journal_remove_journal_head(bh); |
470decc6 DK |
1709 | __brelse(bh); |
1710 | } | |
1711 | return may_free; | |
1712 | } | |
1713 | ||
1714 | /* | |
f7f4bccb | 1715 | * jbd2_journal_invalidatepage |
470decc6 DK |
1716 | * |
1717 | * This code is tricky. It has a number of cases to deal with. | |
1718 | * | |
1719 | * There are two invariants which this code relies on: | |
1720 | * | |
1721 | * i_size must be updated on disk before we start calling invalidatepage on the | |
1722 | * data. | |
1723 | * | |
1724 | * This is done in ext3 by defining an ext3_setattr method which | |
1725 | * updates i_size before truncate gets going. By maintaining this | |
1726 | * invariant, we can be sure that it is safe to throw away any buffers | |
1727 | * attached to the current transaction: once the transaction commits, | |
1728 | * we know that the data will not be needed. | |
1729 | * | |
1730 | * Note however that we can *not* throw away data belonging to the | |
1731 | * previous, committing transaction! | |
1732 | * | |
1733 | * Any disk blocks which *are* part of the previous, committing | |
1734 | * transaction (and which therefore cannot be discarded immediately) are | |
1735 | * not going to be reused in the new running transaction | |
1736 | * | |
1737 | * The bitmap committed_data images guarantee this: any block which is | |
1738 | * allocated in one transaction and removed in the next will be marked | |
1739 | * as in-use in the committed_data bitmap, so cannot be reused until | |
1740 | * the next transaction to delete the block commits. This means that | |
1741 | * leaving committing buffers dirty is quite safe: the disk blocks | |
1742 | * cannot be reallocated to a different file and so buffer aliasing is | |
1743 | * not possible. | |
1744 | * | |
1745 | * | |
1746 | * The above applies mainly to ordered data mode. In writeback mode we | |
1747 | * don't make guarantees about the order in which data hits disk --- in | |
1748 | * particular we don't guarantee that new dirty data is flushed before | |
1749 | * transaction commit --- so it is always safe just to discard data | |
1750 | * immediately in that mode. --sct | |
1751 | */ | |
1752 | ||
1753 | /* | |
1754 | * The journal_unmap_buffer helper function returns zero if the buffer | |
1755 | * concerned remains pinned as an anonymous buffer belonging to an older | |
1756 | * transaction. | |
1757 | * | |
1758 | * We're outside-transaction here. Either or both of j_running_transaction | |
1759 | * and j_committing_transaction may be NULL. | |
1760 | */ | |
1761 | static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh) | |
1762 | { | |
1763 | transaction_t *transaction; | |
1764 | struct journal_head *jh; | |
1765 | int may_free = 1; | |
1766 | int ret; | |
1767 | ||
1768 | BUFFER_TRACE(bh, "entry"); | |
1769 | ||
1770 | /* | |
1771 | * It is safe to proceed here without the j_list_lock because the | |
1772 | * buffers cannot be stolen by try_to_free_buffers as long as we are | |
1773 | * holding the page lock. --sct | |
1774 | */ | |
1775 | ||
1776 | if (!buffer_jbd(bh)) | |
1777 | goto zap_buffer_unlocked; | |
1778 | ||
87c89c23 | 1779 | /* OK, we have data buffer in journaled mode */ |
a931da6a | 1780 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1781 | jbd_lock_bh_state(bh); |
1782 | spin_lock(&journal->j_list_lock); | |
1783 | ||
f7f4bccb | 1784 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
1785 | if (!jh) |
1786 | goto zap_buffer_no_jh; | |
1787 | ||
ba869023 | 1788 | /* |
1789 | * We cannot remove the buffer from checkpoint lists until the | |
1790 | * transaction adding inode to orphan list (let's call it T) | |
1791 | * is committed. Otherwise if the transaction changing the | |
1792 | * buffer would be cleaned from the journal before T is | |
1793 | * committed, a crash will cause that the correct contents of | |
1794 | * the buffer will be lost. On the other hand we have to | |
1795 | * clear the buffer dirty bit at latest at the moment when the | |
1796 | * transaction marking the buffer as freed in the filesystem | |
1797 | * structures is committed because from that moment on the | |
1798 | * buffer can be reallocated and used by a different page. | |
1799 | * Since the block hasn't been freed yet but the inode has | |
1800 | * already been added to orphan list, it is safe for us to add | |
1801 | * the buffer to BJ_Forget list of the newest transaction. | |
1802 | */ | |
470decc6 DK |
1803 | transaction = jh->b_transaction; |
1804 | if (transaction == NULL) { | |
1805 | /* First case: not on any transaction. If it | |
1806 | * has no checkpoint link, then we can zap it: | |
1807 | * it's a writeback-mode buffer so we don't care | |
1808 | * if it hits disk safely. */ | |
1809 | if (!jh->b_cp_transaction) { | |
1810 | JBUFFER_TRACE(jh, "not on any transaction: zap"); | |
1811 | goto zap_buffer; | |
1812 | } | |
1813 | ||
1814 | if (!buffer_dirty(bh)) { | |
1815 | /* bdflush has written it. We can drop it now */ | |
1816 | goto zap_buffer; | |
1817 | } | |
1818 | ||
1819 | /* OK, it must be in the journal but still not | |
1820 | * written fully to disk: it's metadata or | |
1821 | * journaled data... */ | |
1822 | ||
1823 | if (journal->j_running_transaction) { | |
1824 | /* ... and once the current transaction has | |
1825 | * committed, the buffer won't be needed any | |
1826 | * longer. */ | |
1827 | JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget"); | |
1828 | ret = __dispose_buffer(jh, | |
1829 | 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); | |
a931da6a | 1833 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1834 | return ret; |
1835 | } else { | |
1836 | /* There is no currently-running transaction. So the | |
1837 | * orphan record which we wrote for this file must have | |
1838 | * passed into commit. We must attach this buffer to | |
1839 | * the committing transaction, if it exists. */ | |
1840 | if (journal->j_committing_transaction) { | |
1841 | JBUFFER_TRACE(jh, "give to committing trans"); | |
1842 | ret = __dispose_buffer(jh, | |
1843 | journal->j_committing_transaction); | |
f7f4bccb | 1844 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1845 | spin_unlock(&journal->j_list_lock); |
1846 | jbd_unlock_bh_state(bh); | |
a931da6a | 1847 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1848 | return ret; |
1849 | } else { | |
1850 | /* The orphan record's transaction has | |
1851 | * committed. We can cleanse this buffer */ | |
1852 | clear_buffer_jbddirty(bh); | |
1853 | goto zap_buffer; | |
1854 | } | |
1855 | } | |
1856 | } else if (transaction == journal->j_committing_transaction) { | |
9b57988d | 1857 | JBUFFER_TRACE(jh, "on committing transaction"); |
470decc6 | 1858 | /* |
ba869023 | 1859 | * The buffer is committing, we simply cannot touch |
1860 | * it. So we just set j_next_transaction to the | |
1861 | * running transaction (if there is one) and mark | |
1862 | * buffer as freed so that commit code knows it should | |
1863 | * clear dirty bits when it is done with the buffer. | |
1864 | */ | |
470decc6 | 1865 | set_buffer_freed(bh); |
ba869023 | 1866 | if (journal->j_running_transaction && buffer_jbddirty(bh)) |
1867 | jh->b_next_transaction = journal->j_running_transaction; | |
f7f4bccb | 1868 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1869 | spin_unlock(&journal->j_list_lock); |
1870 | jbd_unlock_bh_state(bh); | |
a931da6a | 1871 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1872 | return 0; |
1873 | } else { | |
1874 | /* Good, the buffer belongs to the running transaction. | |
1875 | * We are writing our own transaction's data, not any | |
1876 | * previous one's, so it is safe to throw it away | |
1877 | * (remember that we expect the filesystem to have set | |
1878 | * i_size already for this truncate so recovery will not | |
1879 | * expose the disk blocks we are discarding here.) */ | |
1880 | J_ASSERT_JH(jh, transaction == journal->j_running_transaction); | |
9b57988d | 1881 | JBUFFER_TRACE(jh, "on running transaction"); |
470decc6 DK |
1882 | may_free = __dispose_buffer(jh, transaction); |
1883 | } | |
1884 | ||
1885 | zap_buffer: | |
f7f4bccb | 1886 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1887 | zap_buffer_no_jh: |
1888 | spin_unlock(&journal->j_list_lock); | |
1889 | jbd_unlock_bh_state(bh); | |
a931da6a | 1890 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1891 | zap_buffer_unlocked: |
1892 | clear_buffer_dirty(bh); | |
1893 | J_ASSERT_BH(bh, !buffer_jbddirty(bh)); | |
1894 | clear_buffer_mapped(bh); | |
1895 | clear_buffer_req(bh); | |
1896 | clear_buffer_new(bh); | |
1897 | bh->b_bdev = NULL; | |
1898 | return may_free; | |
1899 | } | |
1900 | ||
1901 | /** | |
f7f4bccb | 1902 | * void jbd2_journal_invalidatepage() |
470decc6 DK |
1903 | * @journal: journal to use for flush... |
1904 | * @page: page to flush | |
1905 | * @offset: length of page to invalidate. | |
1906 | * | |
1907 | * Reap page buffers containing data after offset in page. | |
1908 | * | |
1909 | */ | |
f7f4bccb | 1910 | void jbd2_journal_invalidatepage(journal_t *journal, |
470decc6 DK |
1911 | struct page *page, |
1912 | unsigned long offset) | |
1913 | { | |
1914 | struct buffer_head *head, *bh, *next; | |
1915 | unsigned int curr_off = 0; | |
1916 | int may_free = 1; | |
1917 | ||
1918 | if (!PageLocked(page)) | |
1919 | BUG(); | |
1920 | if (!page_has_buffers(page)) | |
1921 | return; | |
1922 | ||
1923 | /* We will potentially be playing with lists other than just the | |
1924 | * data lists (especially for journaled data mode), so be | |
1925 | * cautious in our locking. */ | |
1926 | ||
1927 | head = bh = page_buffers(page); | |
1928 | do { | |
1929 | unsigned int next_off = curr_off + bh->b_size; | |
1930 | next = bh->b_this_page; | |
1931 | ||
1932 | if (offset <= curr_off) { | |
1933 | /* This block is wholly outside the truncation point */ | |
1934 | lock_buffer(bh); | |
1935 | may_free &= journal_unmap_buffer(journal, bh); | |
1936 | unlock_buffer(bh); | |
1937 | } | |
1938 | curr_off = next_off; | |
1939 | bh = next; | |
1940 | ||
1941 | } while (bh != head); | |
1942 | ||
1943 | if (!offset) { | |
1944 | if (may_free && try_to_free_buffers(page)) | |
1945 | J_ASSERT(!page_has_buffers(page)); | |
1946 | } | |
1947 | } | |
1948 | ||
1949 | /* | |
1950 | * File a buffer on the given transaction list. | |
1951 | */ | |
f7f4bccb | 1952 | void __jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
1953 | transaction_t *transaction, int jlist) |
1954 | { | |
1955 | struct journal_head **list = NULL; | |
1956 | int was_dirty = 0; | |
1957 | struct buffer_head *bh = jh2bh(jh); | |
1958 | ||
1959 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
1960 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
1961 | ||
1962 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
1963 | J_ASSERT_JH(jh, jh->b_transaction == transaction || | |
4019191b | 1964 | jh->b_transaction == NULL); |
470decc6 DK |
1965 | |
1966 | if (jh->b_transaction && jh->b_jlist == jlist) | |
1967 | return; | |
1968 | ||
470decc6 DK |
1969 | if (jlist == BJ_Metadata || jlist == BJ_Reserved || |
1970 | jlist == BJ_Shadow || jlist == BJ_Forget) { | |
f91d1d04 JK |
1971 | /* |
1972 | * For metadata buffers, we track dirty bit in buffer_jbddirty | |
1973 | * instead of buffer_dirty. We should not see a dirty bit set | |
1974 | * here because we clear it in do_get_write_access but e.g. | |
1975 | * tune2fs can modify the sb and set the dirty bit at any time | |
1976 | * so we try to gracefully handle that. | |
1977 | */ | |
1978 | if (buffer_dirty(bh)) | |
1979 | warn_dirty_buffer(bh); | |
470decc6 DK |
1980 | if (test_clear_buffer_dirty(bh) || |
1981 | test_clear_buffer_jbddirty(bh)) | |
1982 | was_dirty = 1; | |
1983 | } | |
1984 | ||
1985 | if (jh->b_transaction) | |
f7f4bccb | 1986 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 DK |
1987 | jh->b_transaction = transaction; |
1988 | ||
1989 | switch (jlist) { | |
1990 | case BJ_None: | |
1991 | J_ASSERT_JH(jh, !jh->b_committed_data); | |
1992 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
1993 | return; | |
470decc6 DK |
1994 | case BJ_Metadata: |
1995 | transaction->t_nr_buffers++; | |
1996 | list = &transaction->t_buffers; | |
1997 | break; | |
1998 | case BJ_Forget: | |
1999 | list = &transaction->t_forget; | |
2000 | break; | |
2001 | case BJ_IO: | |
2002 | list = &transaction->t_iobuf_list; | |
2003 | break; | |
2004 | case BJ_Shadow: | |
2005 | list = &transaction->t_shadow_list; | |
2006 | break; | |
2007 | case BJ_LogCtl: | |
2008 | list = &transaction->t_log_list; | |
2009 | break; | |
2010 | case BJ_Reserved: | |
2011 | list = &transaction->t_reserved_list; | |
2012 | break; | |
470decc6 DK |
2013 | } |
2014 | ||
2015 | __blist_add_buffer(list, jh); | |
2016 | jh->b_jlist = jlist; | |
2017 | ||
2018 | if (was_dirty) | |
2019 | set_buffer_jbddirty(bh); | |
2020 | } | |
2021 | ||
f7f4bccb | 2022 | void jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2023 | transaction_t *transaction, int jlist) |
2024 | { | |
2025 | jbd_lock_bh_state(jh2bh(jh)); | |
2026 | spin_lock(&transaction->t_journal->j_list_lock); | |
f7f4bccb | 2027 | __jbd2_journal_file_buffer(jh, transaction, jlist); |
470decc6 DK |
2028 | spin_unlock(&transaction->t_journal->j_list_lock); |
2029 | jbd_unlock_bh_state(jh2bh(jh)); | |
2030 | } | |
2031 | ||
2032 | /* | |
2033 | * Remove a buffer from its current buffer list in preparation for | |
2034 | * dropping it from its current transaction entirely. If the buffer has | |
2035 | * already started to be used by a subsequent transaction, refile the | |
2036 | * buffer on that transaction's metadata list. | |
2037 | * | |
2038 | * Called under journal->j_list_lock | |
2039 | * | |
2040 | * Called under jbd_lock_bh_state(jh2bh(jh)) | |
2041 | */ | |
f7f4bccb | 2042 | void __jbd2_journal_refile_buffer(struct journal_head *jh) |
470decc6 | 2043 | { |
ba869023 | 2044 | int was_dirty, jlist; |
470decc6 DK |
2045 | struct buffer_head *bh = jh2bh(jh); |
2046 | ||
2047 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
2048 | if (jh->b_transaction) | |
2049 | assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock); | |
2050 | ||
2051 | /* If the buffer is now unused, just drop it. */ | |
2052 | if (jh->b_next_transaction == NULL) { | |
f7f4bccb | 2053 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
2054 | return; |
2055 | } | |
2056 | ||
2057 | /* | |
2058 | * It has been modified by a later transaction: add it to the new | |
2059 | * transaction's metadata list. | |
2060 | */ | |
2061 | ||
2062 | was_dirty = test_clear_buffer_jbddirty(bh); | |
f7f4bccb | 2063 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 DK |
2064 | jh->b_transaction = jh->b_next_transaction; |
2065 | jh->b_next_transaction = NULL; | |
ba869023 | 2066 | if (buffer_freed(bh)) |
2067 | jlist = BJ_Forget; | |
2068 | else if (jh->b_modified) | |
2069 | jlist = BJ_Metadata; | |
2070 | else | |
2071 | jlist = BJ_Reserved; | |
2072 | __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist); | |
470decc6 DK |
2073 | J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING); |
2074 | ||
2075 | if (was_dirty) | |
2076 | set_buffer_jbddirty(bh); | |
2077 | } | |
2078 | ||
2079 | /* | |
2080 | * For the unlocked version of this call, also make sure that any | |
2081 | * hanging journal_head is cleaned up if necessary. | |
2082 | * | |
f7f4bccb | 2083 | * __jbd2_journal_refile_buffer is usually called as part of a single locked |
470decc6 DK |
2084 | * operation on a buffer_head, in which the caller is probably going to |
2085 | * be hooking the journal_head onto other lists. In that case it is up | |
2086 | * to the caller to remove the journal_head if necessary. For the | |
f7f4bccb | 2087 | * unlocked jbd2_journal_refile_buffer call, the caller isn't going to be |
470decc6 DK |
2088 | * doing anything else to the buffer so we need to do the cleanup |
2089 | * ourselves to avoid a jh leak. | |
2090 | * | |
2091 | * *** The journal_head may be freed by this call! *** | |
2092 | */ | |
f7f4bccb | 2093 | void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 DK |
2094 | { |
2095 | struct buffer_head *bh = jh2bh(jh); | |
2096 | ||
2097 | jbd_lock_bh_state(bh); | |
2098 | spin_lock(&journal->j_list_lock); | |
2099 | ||
f7f4bccb | 2100 | __jbd2_journal_refile_buffer(jh); |
470decc6 | 2101 | jbd_unlock_bh_state(bh); |
f7f4bccb | 2102 | jbd2_journal_remove_journal_head(bh); |
470decc6 DK |
2103 | |
2104 | spin_unlock(&journal->j_list_lock); | |
2105 | __brelse(bh); | |
2106 | } | |
c851ed54 JK |
2107 | |
2108 | /* | |
2109 | * File inode in the inode list of the handle's transaction | |
2110 | */ | |
2111 | int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode) | |
2112 | { | |
2113 | transaction_t *transaction = handle->h_transaction; | |
2114 | journal_t *journal = transaction->t_journal; | |
2115 | ||
2116 | if (is_handle_aborted(handle)) | |
2117 | return -EIO; | |
2118 | ||
2119 | jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino, | |
2120 | transaction->t_tid); | |
2121 | ||
2122 | /* | |
2123 | * First check whether inode isn't already on the transaction's | |
2124 | * lists without taking the lock. Note that this check is safe | |
2125 | * without the lock as we cannot race with somebody removing inode | |
2126 | * from the transaction. The reason is that we remove inode from the | |
2127 | * transaction only in journal_release_jbd_inode() and when we commit | |
2128 | * the transaction. We are guarded from the first case by holding | |
2129 | * a reference to the inode. We are safe against the second case | |
2130 | * because if jinode->i_transaction == transaction, commit code | |
2131 | * cannot touch the transaction because we hold reference to it, | |
2132 | * and if jinode->i_next_transaction == transaction, commit code | |
2133 | * will only file the inode where we want it. | |
2134 | */ | |
2135 | if (jinode->i_transaction == transaction || | |
2136 | jinode->i_next_transaction == transaction) | |
2137 | return 0; | |
2138 | ||
2139 | spin_lock(&journal->j_list_lock); | |
2140 | ||
2141 | if (jinode->i_transaction == transaction || | |
2142 | jinode->i_next_transaction == transaction) | |
2143 | goto done; | |
2144 | ||
2145 | /* On some different transaction's list - should be | |
2146 | * the committing one */ | |
2147 | if (jinode->i_transaction) { | |
2148 | J_ASSERT(jinode->i_next_transaction == NULL); | |
2149 | J_ASSERT(jinode->i_transaction == | |
2150 | journal->j_committing_transaction); | |
2151 | jinode->i_next_transaction = transaction; | |
2152 | goto done; | |
2153 | } | |
2154 | /* Not on any transaction list... */ | |
2155 | J_ASSERT(!jinode->i_next_transaction); | |
2156 | jinode->i_transaction = transaction; | |
2157 | list_add(&jinode->i_list, &transaction->t_inode_list); | |
2158 | done: | |
2159 | spin_unlock(&journal->j_list_lock); | |
2160 | ||
2161 | return 0; | |
2162 | } | |
2163 | ||
2164 | /* | |
7f5aa215 JK |
2165 | * File truncate and transaction commit interact with each other in a |
2166 | * non-trivial way. If a transaction writing data block A is | |
2167 | * committing, we cannot discard the data by truncate until we have | |
2168 | * written them. Otherwise if we crashed after the transaction with | |
2169 | * write has committed but before the transaction with truncate has | |
2170 | * committed, we could see stale data in block A. This function is a | |
2171 | * helper to solve this problem. It starts writeout of the truncated | |
2172 | * part in case it is in the committing transaction. | |
2173 | * | |
2174 | * Filesystem code must call this function when inode is journaled in | |
2175 | * ordered mode before truncation happens and after the inode has been | |
2176 | * placed on orphan list with the new inode size. The second condition | |
2177 | * avoids the race that someone writes new data and we start | |
2178 | * committing the transaction after this function has been called but | |
2179 | * before a transaction for truncate is started (and furthermore it | |
2180 | * allows us to optimize the case where the addition to orphan list | |
2181 | * happens in the same transaction as write --- we don't have to write | |
2182 | * any data in such case). | |
c851ed54 | 2183 | */ |
7f5aa215 JK |
2184 | int jbd2_journal_begin_ordered_truncate(journal_t *journal, |
2185 | struct jbd2_inode *jinode, | |
c851ed54 JK |
2186 | loff_t new_size) |
2187 | { | |
7f5aa215 | 2188 | transaction_t *inode_trans, *commit_trans; |
c851ed54 JK |
2189 | int ret = 0; |
2190 | ||
7f5aa215 JK |
2191 | /* This is a quick check to avoid locking if not necessary */ |
2192 | if (!jinode->i_transaction) | |
c851ed54 | 2193 | goto out; |
7f5aa215 JK |
2194 | /* Locks are here just to force reading of recent values, it is |
2195 | * enough that the transaction was not committing before we started | |
2196 | * a transaction adding the inode to orphan list */ | |
a931da6a | 2197 | read_lock(&journal->j_state_lock); |
c851ed54 | 2198 | commit_trans = journal->j_committing_transaction; |
a931da6a | 2199 | read_unlock(&journal->j_state_lock); |
7f5aa215 JK |
2200 | spin_lock(&journal->j_list_lock); |
2201 | inode_trans = jinode->i_transaction; | |
2202 | spin_unlock(&journal->j_list_lock); | |
2203 | if (inode_trans == commit_trans) { | |
2204 | ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping, | |
c851ed54 JK |
2205 | new_size, LLONG_MAX); |
2206 | if (ret) | |
2207 | jbd2_journal_abort(journal, ret); | |
2208 | } | |
2209 | out: | |
2210 | return ret; | |
2211 | } |