]>
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 | 29 | #include <linux/backing-dev.h> |
44705754 | 30 | #include <linux/bug.h> |
47def826 | 31 | #include <linux/module.h> |
470decc6 | 32 | |
343d9c28 TT |
33 | #include <trace/events/jbd2.h> |
34 | ||
7ddae860 | 35 | static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh); |
de1b7941 | 36 | static void __jbd2_journal_unfile_buffer(struct journal_head *jh); |
7ddae860 | 37 | |
0c2022ec YY |
38 | static struct kmem_cache *transaction_cache; |
39 | int __init jbd2_journal_init_transaction_cache(void) | |
40 | { | |
41 | J_ASSERT(!transaction_cache); | |
42 | transaction_cache = kmem_cache_create("jbd2_transaction_s", | |
43 | sizeof(transaction_t), | |
44 | 0, | |
45 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY, | |
46 | NULL); | |
47 | if (transaction_cache) | |
48 | return 0; | |
49 | return -ENOMEM; | |
50 | } | |
51 | ||
52 | void jbd2_journal_destroy_transaction_cache(void) | |
53 | { | |
54 | if (transaction_cache) { | |
55 | kmem_cache_destroy(transaction_cache); | |
56 | transaction_cache = NULL; | |
57 | } | |
58 | } | |
59 | ||
60 | void jbd2_journal_free_transaction(transaction_t *transaction) | |
61 | { | |
62 | if (unlikely(ZERO_OR_NULL_PTR(transaction))) | |
63 | return; | |
64 | kmem_cache_free(transaction_cache, transaction); | |
65 | } | |
66 | ||
470decc6 | 67 | /* |
f7f4bccb | 68 | * jbd2_get_transaction: obtain a new transaction_t object. |
470decc6 DK |
69 | * |
70 | * Simply allocate and initialise a new transaction. Create it in | |
71 | * RUNNING state and add it to the current journal (which should not | |
72 | * have an existing running transaction: we only make a new transaction | |
73 | * once we have started to commit the old one). | |
74 | * | |
75 | * Preconditions: | |
76 | * The journal MUST be locked. We don't perform atomic mallocs on the | |
77 | * new transaction and we can't block without protecting against other | |
78 | * processes trying to touch the journal while it is in transition. | |
79 | * | |
470decc6 DK |
80 | */ |
81 | ||
82 | static transaction_t * | |
f7f4bccb | 83 | jbd2_get_transaction(journal_t *journal, transaction_t *transaction) |
470decc6 DK |
84 | { |
85 | transaction->t_journal = journal; | |
86 | transaction->t_state = T_RUNNING; | |
e07f7183 | 87 | transaction->t_start_time = ktime_get(); |
470decc6 DK |
88 | transaction->t_tid = journal->j_transaction_sequence++; |
89 | transaction->t_expires = jiffies + journal->j_commit_interval; | |
90 | spin_lock_init(&transaction->t_handle_lock); | |
a51dca9c TT |
91 | atomic_set(&transaction->t_updates, 0); |
92 | atomic_set(&transaction->t_outstanding_credits, 0); | |
8dd42046 | 93 | atomic_set(&transaction->t_handle_count, 0); |
c851ed54 | 94 | INIT_LIST_HEAD(&transaction->t_inode_list); |
3e624fc7 | 95 | INIT_LIST_HEAD(&transaction->t_private_list); |
470decc6 DK |
96 | |
97 | /* Set up the commit timer for the new transaction. */ | |
b1f485f2 | 98 | journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires); |
470decc6 DK |
99 | add_timer(&journal->j_commit_timer); |
100 | ||
101 | J_ASSERT(journal->j_running_transaction == NULL); | |
102 | journal->j_running_transaction = transaction; | |
8e85fb3f JL |
103 | transaction->t_max_wait = 0; |
104 | transaction->t_start = jiffies; | |
9fff24aa | 105 | transaction->t_requested = 0; |
470decc6 DK |
106 | |
107 | return transaction; | |
108 | } | |
109 | ||
110 | /* | |
111 | * Handle management. | |
112 | * | |
113 | * A handle_t is an object which represents a single atomic update to a | |
114 | * filesystem, and which tracks all of the modifications which form part | |
115 | * of that one update. | |
116 | */ | |
117 | ||
6d0bf005 | 118 | /* |
28e35e42 | 119 | * Update transaction's maximum wait time, if debugging is enabled. |
6d0bf005 TT |
120 | * |
121 | * In order for t_max_wait to be reliable, it must be protected by a | |
122 | * lock. But doing so will mean that start_this_handle() can not be | |
123 | * run in parallel on SMP systems, which limits our scalability. So | |
124 | * unless debugging is enabled, we no longer update t_max_wait, which | |
125 | * means that maximum wait time reported by the jbd2_run_stats | |
126 | * tracepoint will always be zero. | |
127 | */ | |
28e35e42 TM |
128 | static inline void update_t_max_wait(transaction_t *transaction, |
129 | unsigned long ts) | |
6d0bf005 TT |
130 | { |
131 | #ifdef CONFIG_JBD2_DEBUG | |
6d0bf005 TT |
132 | if (jbd2_journal_enable_debug && |
133 | time_after(transaction->t_start, ts)) { | |
134 | ts = jbd2_time_diff(ts, transaction->t_start); | |
135 | spin_lock(&transaction->t_handle_lock); | |
136 | if (ts > transaction->t_max_wait) | |
137 | transaction->t_max_wait = ts; | |
138 | spin_unlock(&transaction->t_handle_lock); | |
139 | } | |
140 | #endif | |
141 | } | |
142 | ||
470decc6 DK |
143 | /* |
144 | * start_this_handle: Given a handle, deal with any locking or stalling | |
145 | * needed to make sure that there is enough journal space for the handle | |
146 | * to begin. Attach the handle to a transaction and set up the | |
147 | * transaction's buffer credits. | |
148 | */ | |
149 | ||
47def826 | 150 | static int start_this_handle(journal_t *journal, handle_t *handle, |
d2159fb7 | 151 | gfp_t gfp_mask) |
470decc6 | 152 | { |
e4471831 TT |
153 | transaction_t *transaction, *new_transaction = NULL; |
154 | tid_t tid; | |
155 | int needed, need_to_start; | |
156 | int nblocks = handle->h_buffer_credits; | |
28e35e42 | 157 | unsigned long ts = jiffies; |
470decc6 DK |
158 | |
159 | if (nblocks > journal->j_max_transaction_buffers) { | |
f2a44523 | 160 | printk(KERN_ERR "JBD2: %s wants too many credits (%d > %d)\n", |
470decc6 DK |
161 | current->comm, nblocks, |
162 | journal->j_max_transaction_buffers); | |
47def826 | 163 | return -ENOSPC; |
470decc6 DK |
164 | } |
165 | ||
166 | alloc_transaction: | |
167 | if (!journal->j_running_transaction) { | |
b2f4edb3 WG |
168 | new_transaction = kmem_cache_zalloc(transaction_cache, |
169 | gfp_mask); | |
470decc6 | 170 | if (!new_transaction) { |
47def826 TT |
171 | /* |
172 | * If __GFP_FS is not present, then we may be | |
173 | * being called from inside the fs writeback | |
174 | * layer, so we MUST NOT fail. Since | |
175 | * __GFP_NOFAIL is going away, we will arrange | |
176 | * to retry the allocation ourselves. | |
177 | */ | |
178 | if ((gfp_mask & __GFP_FS) == 0) { | |
179 | congestion_wait(BLK_RW_ASYNC, HZ/50); | |
180 | goto alloc_transaction; | |
181 | } | |
182 | return -ENOMEM; | |
470decc6 | 183 | } |
470decc6 DK |
184 | } |
185 | ||
186 | jbd_debug(3, "New handle %p going live.\n", handle); | |
187 | ||
470decc6 DK |
188 | /* |
189 | * We need to hold j_state_lock until t_updates has been incremented, | |
190 | * for proper journal barrier handling | |
191 | */ | |
a931da6a TT |
192 | repeat: |
193 | read_lock(&journal->j_state_lock); | |
5c2178e7 | 194 | BUG_ON(journal->j_flags & JBD2_UNMOUNT); |
470decc6 | 195 | if (is_journal_aborted(journal) || |
f7f4bccb | 196 | (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) { |
a931da6a | 197 | read_unlock(&journal->j_state_lock); |
0c2022ec | 198 | jbd2_journal_free_transaction(new_transaction); |
47def826 | 199 | return -EROFS; |
470decc6 DK |
200 | } |
201 | ||
202 | /* Wait on the journal's transaction barrier if necessary */ | |
203 | if (journal->j_barrier_count) { | |
a931da6a | 204 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
205 | wait_event(journal->j_wait_transaction_locked, |
206 | journal->j_barrier_count == 0); | |
207 | goto repeat; | |
208 | } | |
209 | ||
210 | if (!journal->j_running_transaction) { | |
a931da6a TT |
211 | read_unlock(&journal->j_state_lock); |
212 | if (!new_transaction) | |
470decc6 | 213 | goto alloc_transaction; |
a931da6a | 214 | write_lock(&journal->j_state_lock); |
d7961c7f JK |
215 | if (!journal->j_running_transaction && |
216 | !journal->j_barrier_count) { | |
a931da6a TT |
217 | jbd2_get_transaction(journal, new_transaction); |
218 | new_transaction = NULL; | |
470decc6 | 219 | } |
a931da6a TT |
220 | write_unlock(&journal->j_state_lock); |
221 | goto repeat; | |
470decc6 DK |
222 | } |
223 | ||
224 | transaction = journal->j_running_transaction; | |
225 | ||
226 | /* | |
227 | * If the current transaction is locked down for commit, wait for the | |
228 | * lock to be released. | |
229 | */ | |
230 | if (transaction->t_state == T_LOCKED) { | |
231 | DEFINE_WAIT(wait); | |
232 | ||
233 | prepare_to_wait(&journal->j_wait_transaction_locked, | |
234 | &wait, TASK_UNINTERRUPTIBLE); | |
a931da6a | 235 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
236 | schedule(); |
237 | finish_wait(&journal->j_wait_transaction_locked, &wait); | |
238 | goto repeat; | |
239 | } | |
240 | ||
241 | /* | |
242 | * If there is not enough space left in the log to write all potential | |
243 | * buffers requested by this operation, we need to stall pending a log | |
244 | * checkpoint to free some more log space. | |
245 | */ | |
8dd42046 TT |
246 | needed = atomic_add_return(nblocks, |
247 | &transaction->t_outstanding_credits); | |
470decc6 DK |
248 | |
249 | if (needed > journal->j_max_transaction_buffers) { | |
250 | /* | |
251 | * If the current transaction is already too large, then start | |
252 | * to commit it: we can then go back and attach this handle to | |
253 | * a new transaction. | |
254 | */ | |
255 | DEFINE_WAIT(wait); | |
256 | ||
257 | jbd_debug(2, "Handle %p starting new commit...\n", handle); | |
8dd42046 | 258 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
470decc6 DK |
259 | prepare_to_wait(&journal->j_wait_transaction_locked, &wait, |
260 | TASK_UNINTERRUPTIBLE); | |
e4471831 TT |
261 | tid = transaction->t_tid; |
262 | need_to_start = !tid_geq(journal->j_commit_request, tid); | |
a931da6a | 263 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
264 | if (need_to_start) |
265 | jbd2_log_start_commit(journal, tid); | |
470decc6 DK |
266 | schedule(); |
267 | finish_wait(&journal->j_wait_transaction_locked, &wait); | |
268 | goto repeat; | |
269 | } | |
270 | ||
271 | /* | |
272 | * The commit code assumes that it can get enough log space | |
273 | * without forcing a checkpoint. This is *critical* for | |
274 | * correctness: a checkpoint of a buffer which is also | |
275 | * associated with a committing transaction creates a deadlock, | |
276 | * so commit simply cannot force through checkpoints. | |
277 | * | |
278 | * We must therefore ensure the necessary space in the journal | |
279 | * *before* starting to dirty potentially checkpointed buffers | |
280 | * in the new transaction. | |
281 | * | |
282 | * The worst part is, any transaction currently committing can | |
283 | * reduce the free space arbitrarily. Be careful to account for | |
284 | * those buffers when checkpointing. | |
285 | */ | |
f7f4bccb | 286 | if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) { |
470decc6 | 287 | jbd_debug(2, "Handle %p waiting for checkpoint...\n", handle); |
8dd42046 | 288 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
a931da6a TT |
289 | read_unlock(&journal->j_state_lock); |
290 | write_lock(&journal->j_state_lock); | |
291 | if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) | |
292 | __jbd2_log_wait_for_space(journal); | |
293 | write_unlock(&journal->j_state_lock); | |
294 | goto repeat; | |
470decc6 DK |
295 | } |
296 | ||
297 | /* OK, account for the buffers that this operation expects to | |
8dd42046 | 298 | * use and add the handle to the running transaction. |
8dd42046 | 299 | */ |
28e35e42 | 300 | update_t_max_wait(transaction, ts); |
470decc6 | 301 | handle->h_transaction = transaction; |
343d9c28 TT |
302 | handle->h_requested_credits = nblocks; |
303 | handle->h_start_jiffies = jiffies; | |
a51dca9c | 304 | atomic_inc(&transaction->t_updates); |
8dd42046 | 305 | atomic_inc(&transaction->t_handle_count); |
470decc6 | 306 | jbd_debug(4, "Handle %p given %d credits (total %d, free %d)\n", |
a51dca9c TT |
307 | handle, nblocks, |
308 | atomic_read(&transaction->t_outstanding_credits), | |
f7f4bccb | 309 | __jbd2_log_space_left(journal)); |
a931da6a | 310 | read_unlock(&journal->j_state_lock); |
9599b0e5 JK |
311 | |
312 | lock_map_acquire(&handle->h_lockdep_map); | |
0c2022ec | 313 | jbd2_journal_free_transaction(new_transaction); |
47def826 | 314 | return 0; |
470decc6 DK |
315 | } |
316 | ||
7b751066 MC |
317 | static struct lock_class_key jbd2_handle_key; |
318 | ||
470decc6 DK |
319 | /* Allocate a new handle. This should probably be in a slab... */ |
320 | static handle_t *new_handle(int nblocks) | |
321 | { | |
af1e76d6 | 322 | handle_t *handle = jbd2_alloc_handle(GFP_NOFS); |
470decc6 DK |
323 | if (!handle) |
324 | return NULL; | |
470decc6 DK |
325 | handle->h_buffer_credits = nblocks; |
326 | handle->h_ref = 1; | |
327 | ||
7b751066 MC |
328 | lockdep_init_map(&handle->h_lockdep_map, "jbd2_handle", |
329 | &jbd2_handle_key, 0); | |
330 | ||
470decc6 DK |
331 | return handle; |
332 | } | |
333 | ||
334 | /** | |
f7f4bccb | 335 | * handle_t *jbd2_journal_start() - Obtain a new handle. |
470decc6 DK |
336 | * @journal: Journal to start transaction on. |
337 | * @nblocks: number of block buffer we might modify | |
338 | * | |
339 | * We make sure that the transaction can guarantee at least nblocks of | |
340 | * modified buffers in the log. We block until the log can guarantee | |
341 | * that much space. | |
342 | * | |
343 | * This function is visible to journal users (like ext3fs), so is not | |
344 | * called with the journal already locked. | |
345 | * | |
c867516d EG |
346 | * Return a pointer to a newly allocated handle, or an ERR_PTR() value |
347 | * on failure. | |
470decc6 | 348 | */ |
343d9c28 TT |
349 | handle_t *jbd2__journal_start(journal_t *journal, int nblocks, gfp_t gfp_mask, |
350 | unsigned int type, unsigned int line_no) | |
470decc6 DK |
351 | { |
352 | handle_t *handle = journal_current_handle(); | |
353 | int err; | |
354 | ||
355 | if (!journal) | |
356 | return ERR_PTR(-EROFS); | |
357 | ||
358 | if (handle) { | |
359 | J_ASSERT(handle->h_transaction->t_journal == journal); | |
360 | handle->h_ref++; | |
361 | return handle; | |
362 | } | |
363 | ||
364 | handle = new_handle(nblocks); | |
365 | if (!handle) | |
366 | return ERR_PTR(-ENOMEM); | |
367 | ||
368 | current->journal_info = handle; | |
369 | ||
47def826 | 370 | err = start_this_handle(journal, handle, gfp_mask); |
470decc6 | 371 | if (err < 0) { |
af1e76d6 | 372 | jbd2_free_handle(handle); |
470decc6 | 373 | current->journal_info = NULL; |
df05c1b8 | 374 | return ERR_PTR(err); |
470decc6 | 375 | } |
343d9c28 TT |
376 | handle->h_type = type; |
377 | handle->h_line_no = line_no; | |
378 | trace_jbd2_handle_start(journal->j_fs_dev->bd_dev, | |
379 | handle->h_transaction->t_tid, type, | |
380 | line_no, nblocks); | |
470decc6 DK |
381 | return handle; |
382 | } | |
47def826 TT |
383 | EXPORT_SYMBOL(jbd2__journal_start); |
384 | ||
385 | ||
386 | handle_t *jbd2_journal_start(journal_t *journal, int nblocks) | |
387 | { | |
343d9c28 | 388 | return jbd2__journal_start(journal, nblocks, GFP_NOFS, 0, 0); |
47def826 TT |
389 | } |
390 | EXPORT_SYMBOL(jbd2_journal_start); | |
391 | ||
470decc6 DK |
392 | |
393 | /** | |
f7f4bccb | 394 | * int jbd2_journal_extend() - extend buffer credits. |
470decc6 DK |
395 | * @handle: handle to 'extend' |
396 | * @nblocks: nr blocks to try to extend by. | |
397 | * | |
398 | * Some transactions, such as large extends and truncates, can be done | |
399 | * atomically all at once or in several stages. The operation requests | |
400 | * a credit for a number of buffer modications in advance, but can | |
401 | * extend its credit if it needs more. | |
402 | * | |
f7f4bccb | 403 | * jbd2_journal_extend tries to give the running handle more buffer credits. |
470decc6 DK |
404 | * It does not guarantee that allocation - this is a best-effort only. |
405 | * The calling process MUST be able to deal cleanly with a failure to | |
406 | * extend here. | |
407 | * | |
408 | * Return 0 on success, non-zero on failure. | |
409 | * | |
410 | * return code < 0 implies an error | |
411 | * return code > 0 implies normal transaction-full status. | |
412 | */ | |
f7f4bccb | 413 | int jbd2_journal_extend(handle_t *handle, int nblocks) |
470decc6 DK |
414 | { |
415 | transaction_t *transaction = handle->h_transaction; | |
416 | journal_t *journal = transaction->t_journal; | |
417 | int result; | |
418 | int wanted; | |
419 | ||
420 | result = -EIO; | |
421 | if (is_handle_aborted(handle)) | |
422 | goto out; | |
423 | ||
424 | result = 1; | |
425 | ||
a931da6a | 426 | read_lock(&journal->j_state_lock); |
470decc6 DK |
427 | |
428 | /* Don't extend a locked-down transaction! */ | |
429 | if (handle->h_transaction->t_state != T_RUNNING) { | |
430 | jbd_debug(3, "denied handle %p %d blocks: " | |
431 | "transaction not running\n", handle, nblocks); | |
432 | goto error_out; | |
433 | } | |
434 | ||
435 | spin_lock(&transaction->t_handle_lock); | |
a51dca9c | 436 | wanted = atomic_read(&transaction->t_outstanding_credits) + nblocks; |
470decc6 DK |
437 | |
438 | if (wanted > journal->j_max_transaction_buffers) { | |
439 | jbd_debug(3, "denied handle %p %d blocks: " | |
440 | "transaction too large\n", handle, nblocks); | |
441 | goto unlock; | |
442 | } | |
443 | ||
f7f4bccb | 444 | if (wanted > __jbd2_log_space_left(journal)) { |
470decc6 DK |
445 | jbd_debug(3, "denied handle %p %d blocks: " |
446 | "insufficient log space\n", handle, nblocks); | |
447 | goto unlock; | |
448 | } | |
449 | ||
343d9c28 TT |
450 | trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev, |
451 | handle->h_transaction->t_tid, | |
452 | handle->h_type, handle->h_line_no, | |
453 | handle->h_buffer_credits, | |
454 | nblocks); | |
455 | ||
470decc6 | 456 | handle->h_buffer_credits += nblocks; |
343d9c28 | 457 | handle->h_requested_credits += nblocks; |
a51dca9c | 458 | atomic_add(nblocks, &transaction->t_outstanding_credits); |
470decc6 DK |
459 | result = 0; |
460 | ||
461 | jbd_debug(3, "extended handle %p by %d\n", handle, nblocks); | |
462 | unlock: | |
463 | spin_unlock(&transaction->t_handle_lock); | |
464 | error_out: | |
a931da6a | 465 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
466 | out: |
467 | return result; | |
468 | } | |
469 | ||
470 | ||
471 | /** | |
f7f4bccb | 472 | * int jbd2_journal_restart() - restart a handle . |
470decc6 DK |
473 | * @handle: handle to restart |
474 | * @nblocks: nr credits requested | |
475 | * | |
476 | * Restart a handle for a multi-transaction filesystem | |
477 | * operation. | |
478 | * | |
f7f4bccb MC |
479 | * If the jbd2_journal_extend() call above fails to grant new buffer credits |
480 | * to a running handle, a call to jbd2_journal_restart will commit the | |
470decc6 DK |
481 | * handle's transaction so far and reattach the handle to a new |
482 | * transaction capabable of guaranteeing the requested number of | |
483 | * credits. | |
484 | */ | |
d2159fb7 | 485 | int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask) |
470decc6 DK |
486 | { |
487 | transaction_t *transaction = handle->h_transaction; | |
488 | journal_t *journal = transaction->t_journal; | |
e4471831 TT |
489 | tid_t tid; |
490 | int need_to_start, ret; | |
470decc6 DK |
491 | |
492 | /* If we've had an abort of any type, don't even think about | |
493 | * actually doing the restart! */ | |
494 | if (is_handle_aborted(handle)) | |
495 | return 0; | |
496 | ||
497 | /* | |
498 | * First unlink the handle from its current transaction, and start the | |
499 | * commit on that. | |
500 | */ | |
a51dca9c | 501 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 DK |
502 | J_ASSERT(journal_current_handle() == handle); |
503 | ||
a931da6a | 504 | read_lock(&journal->j_state_lock); |
470decc6 | 505 | spin_lock(&transaction->t_handle_lock); |
a51dca9c TT |
506 | atomic_sub(handle->h_buffer_credits, |
507 | &transaction->t_outstanding_credits); | |
508 | if (atomic_dec_and_test(&transaction->t_updates)) | |
470decc6 DK |
509 | wake_up(&journal->j_wait_updates); |
510 | spin_unlock(&transaction->t_handle_lock); | |
511 | ||
512 | jbd_debug(2, "restarting handle %p\n", handle); | |
e4471831 TT |
513 | tid = transaction->t_tid; |
514 | need_to_start = !tid_geq(journal->j_commit_request, tid); | |
a931da6a | 515 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
516 | if (need_to_start) |
517 | jbd2_log_start_commit(journal, tid); | |
470decc6 | 518 | |
9599b0e5 | 519 | lock_map_release(&handle->h_lockdep_map); |
470decc6 | 520 | handle->h_buffer_credits = nblocks; |
47def826 | 521 | ret = start_this_handle(journal, handle, gfp_mask); |
470decc6 DK |
522 | return ret; |
523 | } | |
47def826 | 524 | EXPORT_SYMBOL(jbd2__journal_restart); |
470decc6 DK |
525 | |
526 | ||
47def826 TT |
527 | int jbd2_journal_restart(handle_t *handle, int nblocks) |
528 | { | |
529 | return jbd2__journal_restart(handle, nblocks, GFP_NOFS); | |
530 | } | |
531 | EXPORT_SYMBOL(jbd2_journal_restart); | |
532 | ||
470decc6 | 533 | /** |
f7f4bccb | 534 | * void jbd2_journal_lock_updates () - establish a transaction barrier. |
470decc6 DK |
535 | * @journal: Journal to establish a barrier on. |
536 | * | |
537 | * This locks out any further updates from being started, and blocks | |
538 | * until all existing updates have completed, returning only once the | |
539 | * journal is in a quiescent state with no updates running. | |
540 | * | |
541 | * The journal lock should not be held on entry. | |
542 | */ | |
f7f4bccb | 543 | void jbd2_journal_lock_updates(journal_t *journal) |
470decc6 DK |
544 | { |
545 | DEFINE_WAIT(wait); | |
546 | ||
a931da6a | 547 | write_lock(&journal->j_state_lock); |
470decc6 DK |
548 | ++journal->j_barrier_count; |
549 | ||
550 | /* Wait until there are no running updates */ | |
551 | while (1) { | |
552 | transaction_t *transaction = journal->j_running_transaction; | |
553 | ||
554 | if (!transaction) | |
555 | break; | |
556 | ||
557 | spin_lock(&transaction->t_handle_lock); | |
9837d8e9 JK |
558 | prepare_to_wait(&journal->j_wait_updates, &wait, |
559 | TASK_UNINTERRUPTIBLE); | |
a51dca9c | 560 | if (!atomic_read(&transaction->t_updates)) { |
470decc6 | 561 | spin_unlock(&transaction->t_handle_lock); |
9837d8e9 | 562 | finish_wait(&journal->j_wait_updates, &wait); |
470decc6 DK |
563 | break; |
564 | } | |
470decc6 | 565 | spin_unlock(&transaction->t_handle_lock); |
a931da6a | 566 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
567 | schedule(); |
568 | finish_wait(&journal->j_wait_updates, &wait); | |
a931da6a | 569 | write_lock(&journal->j_state_lock); |
470decc6 | 570 | } |
a931da6a | 571 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
572 | |
573 | /* | |
574 | * We have now established a barrier against other normal updates, but | |
f7f4bccb | 575 | * we also need to barrier against other jbd2_journal_lock_updates() calls |
470decc6 DK |
576 | * to make sure that we serialise special journal-locked operations |
577 | * too. | |
578 | */ | |
579 | mutex_lock(&journal->j_barrier); | |
580 | } | |
581 | ||
582 | /** | |
f7f4bccb | 583 | * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier |
470decc6 DK |
584 | * @journal: Journal to release the barrier on. |
585 | * | |
f7f4bccb | 586 | * Release a transaction barrier obtained with jbd2_journal_lock_updates(). |
470decc6 DK |
587 | * |
588 | * Should be called without the journal lock held. | |
589 | */ | |
f7f4bccb | 590 | void jbd2_journal_unlock_updates (journal_t *journal) |
470decc6 DK |
591 | { |
592 | J_ASSERT(journal->j_barrier_count != 0); | |
593 | ||
594 | mutex_unlock(&journal->j_barrier); | |
a931da6a | 595 | write_lock(&journal->j_state_lock); |
470decc6 | 596 | --journal->j_barrier_count; |
a931da6a | 597 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
598 | wake_up(&journal->j_wait_transaction_locked); |
599 | } | |
600 | ||
f91d1d04 | 601 | static void warn_dirty_buffer(struct buffer_head *bh) |
470decc6 | 602 | { |
f91d1d04 | 603 | char b[BDEVNAME_SIZE]; |
470decc6 | 604 | |
f91d1d04 | 605 | printk(KERN_WARNING |
f2a44523 | 606 | "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). " |
f91d1d04 JK |
607 | "There's a risk of filesystem corruption in case of system " |
608 | "crash.\n", | |
609 | bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr); | |
470decc6 DK |
610 | } |
611 | ||
b34090e5 JK |
612 | static int sleep_on_shadow_bh(void *word) |
613 | { | |
614 | io_schedule(); | |
615 | return 0; | |
616 | } | |
617 | ||
470decc6 DK |
618 | /* |
619 | * If the buffer is already part of the current transaction, then there | |
620 | * is nothing we need to do. If it is already part of a prior | |
621 | * transaction which we are still committing to disk, then we need to | |
622 | * make sure that we do not overwrite the old copy: we do copy-out to | |
623 | * preserve the copy going to disk. We also account the buffer against | |
624 | * the handle's metadata buffer credits (unless the buffer is already | |
625 | * part of the transaction, that is). | |
626 | * | |
627 | */ | |
628 | static int | |
629 | do_get_write_access(handle_t *handle, struct journal_head *jh, | |
630 | int force_copy) | |
631 | { | |
632 | struct buffer_head *bh; | |
633 | transaction_t *transaction; | |
634 | journal_t *journal; | |
635 | int error; | |
636 | char *frozen_buffer = NULL; | |
637 | int need_copy = 0; | |
f783f091 | 638 | unsigned long start_lock, time_lock; |
470decc6 DK |
639 | |
640 | if (is_handle_aborted(handle)) | |
641 | return -EROFS; | |
642 | ||
643 | transaction = handle->h_transaction; | |
644 | journal = transaction->t_journal; | |
645 | ||
cfef2c6a | 646 | jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy); |
470decc6 DK |
647 | |
648 | JBUFFER_TRACE(jh, "entry"); | |
649 | repeat: | |
650 | bh = jh2bh(jh); | |
651 | ||
652 | /* @@@ Need to check for errors here at some point. */ | |
653 | ||
f783f091 | 654 | start_lock = jiffies; |
470decc6 DK |
655 | lock_buffer(bh); |
656 | jbd_lock_bh_state(bh); | |
657 | ||
f783f091 TT |
658 | /* If it takes too long to lock the buffer, trace it */ |
659 | time_lock = jbd2_time_diff(start_lock, jiffies); | |
660 | if (time_lock > HZ/10) | |
661 | trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev, | |
662 | jiffies_to_msecs(time_lock)); | |
663 | ||
470decc6 DK |
664 | /* We now hold the buffer lock so it is safe to query the buffer |
665 | * state. Is the buffer dirty? | |
666 | * | |
667 | * If so, there are two possibilities. The buffer may be | |
668 | * non-journaled, and undergoing a quite legitimate writeback. | |
669 | * Otherwise, it is journaled, and we don't expect dirty buffers | |
670 | * in that state (the buffers should be marked JBD_Dirty | |
671 | * instead.) So either the IO is being done under our own | |
672 | * control and this is a bug, or it's a third party IO such as | |
673 | * dump(8) (which may leave the buffer scheduled for read --- | |
674 | * ie. locked but not dirty) or tune2fs (which may actually have | |
675 | * the buffer dirtied, ugh.) */ | |
676 | ||
677 | if (buffer_dirty(bh)) { | |
678 | /* | |
679 | * First question: is this buffer already part of the current | |
680 | * transaction or the existing committing transaction? | |
681 | */ | |
682 | if (jh->b_transaction) { | |
683 | J_ASSERT_JH(jh, | |
684 | jh->b_transaction == transaction || | |
685 | jh->b_transaction == | |
686 | journal->j_committing_transaction); | |
687 | if (jh->b_next_transaction) | |
688 | J_ASSERT_JH(jh, jh->b_next_transaction == | |
689 | transaction); | |
f91d1d04 | 690 | warn_dirty_buffer(bh); |
470decc6 DK |
691 | } |
692 | /* | |
693 | * In any case we need to clean the dirty flag and we must | |
694 | * do it under the buffer lock to be sure we don't race | |
695 | * with running write-out. | |
696 | */ | |
f91d1d04 JK |
697 | JBUFFER_TRACE(jh, "Journalling dirty buffer"); |
698 | clear_buffer_dirty(bh); | |
699 | set_buffer_jbddirty(bh); | |
470decc6 DK |
700 | } |
701 | ||
702 | unlock_buffer(bh); | |
703 | ||
704 | error = -EROFS; | |
705 | if (is_handle_aborted(handle)) { | |
706 | jbd_unlock_bh_state(bh); | |
707 | goto out; | |
708 | } | |
709 | error = 0; | |
710 | ||
711 | /* | |
712 | * The buffer is already part of this transaction if b_transaction or | |
713 | * b_next_transaction points to it | |
714 | */ | |
715 | if (jh->b_transaction == transaction || | |
716 | jh->b_next_transaction == transaction) | |
717 | goto done; | |
718 | ||
9fc7c63a JB |
719 | /* |
720 | * this is the first time this transaction is touching this buffer, | |
721 | * reset the modified flag | |
722 | */ | |
723 | jh->b_modified = 0; | |
724 | ||
470decc6 DK |
725 | /* |
726 | * If there is already a copy-out version of this buffer, then we don't | |
727 | * need to make another one | |
728 | */ | |
729 | if (jh->b_frozen_data) { | |
730 | JBUFFER_TRACE(jh, "has frozen data"); | |
731 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
732 | jh->b_next_transaction = transaction; | |
733 | goto done; | |
734 | } | |
735 | ||
736 | /* Is there data here we need to preserve? */ | |
737 | ||
738 | if (jh->b_transaction && jh->b_transaction != transaction) { | |
739 | JBUFFER_TRACE(jh, "owned by older transaction"); | |
740 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
741 | J_ASSERT_JH(jh, jh->b_transaction == | |
742 | journal->j_committing_transaction); | |
743 | ||
744 | /* There is one case we have to be very careful about. | |
745 | * If the committing transaction is currently writing | |
746 | * this buffer out to disk and has NOT made a copy-out, | |
747 | * then we cannot modify the buffer contents at all | |
748 | * right now. The essence of copy-out is that it is the | |
749 | * extra copy, not the primary copy, which gets | |
750 | * journaled. If the primary copy is already going to | |
751 | * disk then we cannot do copy-out here. */ | |
752 | ||
b34090e5 | 753 | if (buffer_shadow(bh)) { |
470decc6 DK |
754 | JBUFFER_TRACE(jh, "on shadow: sleep"); |
755 | jbd_unlock_bh_state(bh); | |
b34090e5 JK |
756 | wait_on_bit(&bh->b_state, BH_Shadow, |
757 | sleep_on_shadow_bh, TASK_UNINTERRUPTIBLE); | |
470decc6 DK |
758 | goto repeat; |
759 | } | |
760 | ||
b34090e5 JK |
761 | /* |
762 | * Only do the copy if the currently-owning transaction still | |
763 | * needs it. If buffer isn't on BJ_Metadata list, the | |
764 | * committing transaction is past that stage (here we use the | |
765 | * fact that BH_Shadow is set under bh_state lock together with | |
766 | * refiling to BJ_Shadow list and at this point we know the | |
767 | * buffer doesn't have BH_Shadow set). | |
470decc6 DK |
768 | * |
769 | * Subtle point, though: if this is a get_undo_access, | |
770 | * then we will be relying on the frozen_data to contain | |
771 | * the new value of the committed_data record after the | |
772 | * transaction, so we HAVE to force the frozen_data copy | |
b34090e5 JK |
773 | * in that case. |
774 | */ | |
775 | if (jh->b_jlist == BJ_Metadata || force_copy) { | |
470decc6 DK |
776 | JBUFFER_TRACE(jh, "generate frozen data"); |
777 | if (!frozen_buffer) { | |
778 | JBUFFER_TRACE(jh, "allocate memory for buffer"); | |
779 | jbd_unlock_bh_state(bh); | |
780 | frozen_buffer = | |
af1e76d6 | 781 | jbd2_alloc(jh2bh(jh)->b_size, |
470decc6 DK |
782 | GFP_NOFS); |
783 | if (!frozen_buffer) { | |
784 | printk(KERN_EMERG | |
785 | "%s: OOM for frozen_buffer\n", | |
329d291f | 786 | __func__); |
470decc6 DK |
787 | JBUFFER_TRACE(jh, "oom!"); |
788 | error = -ENOMEM; | |
789 | jbd_lock_bh_state(bh); | |
790 | goto done; | |
791 | } | |
792 | goto repeat; | |
793 | } | |
794 | jh->b_frozen_data = frozen_buffer; | |
795 | frozen_buffer = NULL; | |
796 | need_copy = 1; | |
797 | } | |
798 | jh->b_next_transaction = transaction; | |
799 | } | |
800 | ||
801 | ||
802 | /* | |
803 | * Finally, if the buffer is not journaled right now, we need to make | |
804 | * sure it doesn't get written to disk before the caller actually | |
805 | * commits the new data | |
806 | */ | |
807 | if (!jh->b_transaction) { | |
808 | JBUFFER_TRACE(jh, "no transaction"); | |
809 | J_ASSERT_JH(jh, !jh->b_next_transaction); | |
470decc6 DK |
810 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); |
811 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 812 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
470decc6 DK |
813 | spin_unlock(&journal->j_list_lock); |
814 | } | |
815 | ||
816 | done: | |
817 | if (need_copy) { | |
818 | struct page *page; | |
819 | int offset; | |
820 | char *source; | |
821 | ||
822 | J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)), | |
823 | "Possible IO failure.\n"); | |
824 | page = jh2bh(jh)->b_page; | |
a1dd5331 | 825 | offset = offset_in_page(jh2bh(jh)->b_data); |
303a8f2a | 826 | source = kmap_atomic(page); |
13ceef09 JK |
827 | /* Fire data frozen trigger just before we copy the data */ |
828 | jbd2_buffer_frozen_trigger(jh, source + offset, | |
829 | jh->b_triggers); | |
470decc6 | 830 | memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size); |
303a8f2a | 831 | kunmap_atomic(source); |
e06c8227 JB |
832 | |
833 | /* | |
834 | * Now that the frozen data is saved off, we need to store | |
835 | * any matching triggers. | |
836 | */ | |
837 | jh->b_frozen_triggers = jh->b_triggers; | |
470decc6 DK |
838 | } |
839 | jbd_unlock_bh_state(bh); | |
840 | ||
841 | /* | |
842 | * If we are about to journal a buffer, then any revoke pending on it is | |
843 | * no longer valid | |
844 | */ | |
f7f4bccb | 845 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 DK |
846 | |
847 | out: | |
848 | if (unlikely(frozen_buffer)) /* It's usually NULL */ | |
af1e76d6 | 849 | jbd2_free(frozen_buffer, bh->b_size); |
470decc6 DK |
850 | |
851 | JBUFFER_TRACE(jh, "exit"); | |
852 | return error; | |
853 | } | |
854 | ||
855 | /** | |
f7f4bccb | 856 | * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update. |
470decc6 DK |
857 | * @handle: transaction to add buffer modifications to |
858 | * @bh: bh to be used for metadata writes | |
470decc6 DK |
859 | * |
860 | * Returns an error code or 0 on success. | |
861 | * | |
862 | * In full data journalling mode the buffer may be of type BJ_AsyncData, | |
863 | * because we're write()ing a buffer which is also part of a shared mapping. | |
864 | */ | |
865 | ||
f7f4bccb | 866 | int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh) |
470decc6 | 867 | { |
f7f4bccb | 868 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
869 | int rc; |
870 | ||
871 | /* We do not want to get caught playing with fields which the | |
872 | * log thread also manipulates. Make sure that the buffer | |
873 | * completes any outstanding IO before proceeding. */ | |
874 | rc = do_get_write_access(handle, jh, 0); | |
f7f4bccb | 875 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
876 | return rc; |
877 | } | |
878 | ||
879 | ||
880 | /* | |
881 | * When the user wants to journal a newly created buffer_head | |
882 | * (ie. getblk() returned a new buffer and we are going to populate it | |
883 | * manually rather than reading off disk), then we need to keep the | |
884 | * buffer_head locked until it has been completely filled with new | |
885 | * data. In this case, we should be able to make the assertion that | |
886 | * the bh is not already part of an existing transaction. | |
887 | * | |
888 | * The buffer should already be locked by the caller by this point. | |
889 | * There is no lock ranking violation: it was a newly created, | |
890 | * unlocked buffer beforehand. */ | |
891 | ||
892 | /** | |
f7f4bccb | 893 | * int jbd2_journal_get_create_access () - notify intent to use newly created bh |
470decc6 DK |
894 | * @handle: transaction to new buffer to |
895 | * @bh: new buffer. | |
896 | * | |
897 | * Call this if you create a new bh. | |
898 | */ | |
f7f4bccb | 899 | int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
900 | { |
901 | transaction_t *transaction = handle->h_transaction; | |
902 | journal_t *journal = transaction->t_journal; | |
f7f4bccb | 903 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
904 | int err; |
905 | ||
906 | jbd_debug(5, "journal_head %p\n", jh); | |
907 | err = -EROFS; | |
908 | if (is_handle_aborted(handle)) | |
909 | goto out; | |
910 | err = 0; | |
911 | ||
912 | JBUFFER_TRACE(jh, "entry"); | |
913 | /* | |
914 | * The buffer may already belong to this transaction due to pre-zeroing | |
915 | * in the filesystem's new_block code. It may also be on the previous, | |
916 | * committing transaction's lists, but it HAS to be in Forget state in | |
917 | * that case: the transaction must have deleted the buffer for it to be | |
918 | * reused here. | |
919 | */ | |
920 | jbd_lock_bh_state(bh); | |
921 | spin_lock(&journal->j_list_lock); | |
922 | J_ASSERT_JH(jh, (jh->b_transaction == transaction || | |
923 | jh->b_transaction == NULL || | |
924 | (jh->b_transaction == journal->j_committing_transaction && | |
925 | jh->b_jlist == BJ_Forget))); | |
926 | ||
927 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
928 | J_ASSERT_JH(jh, buffer_locked(jh2bh(jh))); | |
929 | ||
930 | if (jh->b_transaction == NULL) { | |
f91d1d04 JK |
931 | /* |
932 | * Previous jbd2_journal_forget() could have left the buffer | |
933 | * with jbddirty bit set because it was being committed. When | |
934 | * the commit finished, we've filed the buffer for | |
935 | * checkpointing and marked it dirty. Now we are reallocating | |
936 | * the buffer so the transaction freeing it must have | |
937 | * committed and so it's safe to clear the dirty bit. | |
938 | */ | |
939 | clear_buffer_dirty(jh2bh(jh)); | |
9fc7c63a JB |
940 | /* first access by this transaction */ |
941 | jh->b_modified = 0; | |
942 | ||
470decc6 | 943 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); |
f7f4bccb | 944 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
470decc6 | 945 | } else if (jh->b_transaction == journal->j_committing_transaction) { |
9fc7c63a JB |
946 | /* first access by this transaction */ |
947 | jh->b_modified = 0; | |
948 | ||
470decc6 DK |
949 | JBUFFER_TRACE(jh, "set next transaction"); |
950 | jh->b_next_transaction = transaction; | |
951 | } | |
952 | spin_unlock(&journal->j_list_lock); | |
953 | jbd_unlock_bh_state(bh); | |
954 | ||
955 | /* | |
956 | * akpm: I added this. ext3_alloc_branch can pick up new indirect | |
957 | * blocks which contain freed but then revoked metadata. We need | |
958 | * to cancel the revoke in case we end up freeing it yet again | |
959 | * and the reallocating as data - this would cause a second revoke, | |
960 | * which hits an assertion error. | |
961 | */ | |
962 | JBUFFER_TRACE(jh, "cancelling revoke"); | |
f7f4bccb | 963 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 | 964 | out: |
3991b400 | 965 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
966 | return err; |
967 | } | |
968 | ||
969 | /** | |
f7f4bccb | 970 | * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with |
470decc6 DK |
971 | * non-rewindable consequences |
972 | * @handle: transaction | |
973 | * @bh: buffer to undo | |
470decc6 DK |
974 | * |
975 | * Sometimes there is a need to distinguish between metadata which has | |
976 | * been committed to disk and that which has not. The ext3fs code uses | |
977 | * this for freeing and allocating space, we have to make sure that we | |
978 | * do not reuse freed space until the deallocation has been committed, | |
979 | * since if we overwrote that space we would make the delete | |
980 | * un-rewindable in case of a crash. | |
981 | * | |
f7f4bccb | 982 | * To deal with that, jbd2_journal_get_undo_access requests write access to a |
470decc6 DK |
983 | * buffer for parts of non-rewindable operations such as delete |
984 | * operations on the bitmaps. The journaling code must keep a copy of | |
985 | * the buffer's contents prior to the undo_access call until such time | |
986 | * as we know that the buffer has definitely been committed to disk. | |
987 | * | |
988 | * We never need to know which transaction the committed data is part | |
989 | * of, buffers touched here are guaranteed to be dirtied later and so | |
990 | * will be committed to a new transaction in due course, at which point | |
991 | * we can discard the old committed data pointer. | |
992 | * | |
993 | * Returns error number or 0 on success. | |
994 | */ | |
f7f4bccb | 995 | int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
996 | { |
997 | int err; | |
f7f4bccb | 998 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
999 | char *committed_data = NULL; |
1000 | ||
1001 | JBUFFER_TRACE(jh, "entry"); | |
1002 | ||
1003 | /* | |
1004 | * Do this first --- it can drop the journal lock, so we want to | |
1005 | * make sure that obtaining the committed_data is done | |
1006 | * atomically wrt. completion of any outstanding commits. | |
1007 | */ | |
1008 | err = do_get_write_access(handle, jh, 1); | |
1009 | if (err) | |
1010 | goto out; | |
1011 | ||
1012 | repeat: | |
1013 | if (!jh->b_committed_data) { | |
af1e76d6 | 1014 | committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS); |
470decc6 DK |
1015 | if (!committed_data) { |
1016 | printk(KERN_EMERG "%s: No memory for committed data\n", | |
329d291f | 1017 | __func__); |
470decc6 DK |
1018 | err = -ENOMEM; |
1019 | goto out; | |
1020 | } | |
1021 | } | |
1022 | ||
1023 | jbd_lock_bh_state(bh); | |
1024 | if (!jh->b_committed_data) { | |
1025 | /* Copy out the current buffer contents into the | |
1026 | * preserved, committed copy. */ | |
1027 | JBUFFER_TRACE(jh, "generate b_committed data"); | |
1028 | if (!committed_data) { | |
1029 | jbd_unlock_bh_state(bh); | |
1030 | goto repeat; | |
1031 | } | |
1032 | ||
1033 | jh->b_committed_data = committed_data; | |
1034 | committed_data = NULL; | |
1035 | memcpy(jh->b_committed_data, bh->b_data, bh->b_size); | |
1036 | } | |
1037 | jbd_unlock_bh_state(bh); | |
1038 | out: | |
f7f4bccb | 1039 | jbd2_journal_put_journal_head(jh); |
470decc6 | 1040 | if (unlikely(committed_data)) |
af1e76d6 | 1041 | jbd2_free(committed_data, bh->b_size); |
470decc6 DK |
1042 | return err; |
1043 | } | |
1044 | ||
e06c8227 JB |
1045 | /** |
1046 | * void jbd2_journal_set_triggers() - Add triggers for commit writeout | |
1047 | * @bh: buffer to trigger on | |
1048 | * @type: struct jbd2_buffer_trigger_type containing the trigger(s). | |
1049 | * | |
1050 | * Set any triggers on this journal_head. This is always safe, because | |
1051 | * triggers for a committing buffer will be saved off, and triggers for | |
1052 | * a running transaction will match the buffer in that transaction. | |
1053 | * | |
1054 | * Call with NULL to clear the triggers. | |
1055 | */ | |
1056 | void jbd2_journal_set_triggers(struct buffer_head *bh, | |
1057 | struct jbd2_buffer_trigger_type *type) | |
1058 | { | |
ad56edad | 1059 | struct journal_head *jh = jbd2_journal_grab_journal_head(bh); |
e06c8227 | 1060 | |
ad56edad JK |
1061 | if (WARN_ON(!jh)) |
1062 | return; | |
e06c8227 | 1063 | jh->b_triggers = type; |
ad56edad | 1064 | jbd2_journal_put_journal_head(jh); |
e06c8227 JB |
1065 | } |
1066 | ||
13ceef09 | 1067 | void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data, |
e06c8227 JB |
1068 | struct jbd2_buffer_trigger_type *triggers) |
1069 | { | |
1070 | struct buffer_head *bh = jh2bh(jh); | |
1071 | ||
13ceef09 | 1072 | if (!triggers || !triggers->t_frozen) |
e06c8227 JB |
1073 | return; |
1074 | ||
13ceef09 | 1075 | triggers->t_frozen(triggers, bh, mapped_data, bh->b_size); |
e06c8227 JB |
1076 | } |
1077 | ||
1078 | void jbd2_buffer_abort_trigger(struct journal_head *jh, | |
1079 | struct jbd2_buffer_trigger_type *triggers) | |
1080 | { | |
1081 | if (!triggers || !triggers->t_abort) | |
1082 | return; | |
1083 | ||
1084 | triggers->t_abort(triggers, jh2bh(jh)); | |
1085 | } | |
1086 | ||
1087 | ||
1088 | ||
470decc6 | 1089 | /** |
f7f4bccb | 1090 | * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata |
470decc6 DK |
1091 | * @handle: transaction to add buffer to. |
1092 | * @bh: buffer to mark | |
1093 | * | |
1094 | * mark dirty metadata which needs to be journaled as part of the current | |
1095 | * transaction. | |
1096 | * | |
9ea7a0df TT |
1097 | * The buffer must have previously had jbd2_journal_get_write_access() |
1098 | * called so that it has a valid journal_head attached to the buffer | |
1099 | * head. | |
1100 | * | |
470decc6 DK |
1101 | * The buffer is placed on the transaction's metadata list and is marked |
1102 | * as belonging to the transaction. | |
1103 | * | |
1104 | * Returns error number or 0 on success. | |
1105 | * | |
1106 | * Special care needs to be taken if the buffer already belongs to the | |
1107 | * current committing transaction (in which case we should have frozen | |
1108 | * data present for that commit). In that case, we don't relink the | |
1109 | * buffer: that only gets done when the old transaction finally | |
1110 | * completes its commit. | |
1111 | */ | |
f7f4bccb | 1112 | int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1113 | { |
1114 | transaction_t *transaction = handle->h_transaction; | |
1115 | journal_t *journal = transaction->t_journal; | |
ad56edad | 1116 | struct journal_head *jh; |
9ea7a0df | 1117 | int ret = 0; |
470decc6 | 1118 | |
470decc6 DK |
1119 | if (is_handle_aborted(handle)) |
1120 | goto out; | |
ad56edad JK |
1121 | jh = jbd2_journal_grab_journal_head(bh); |
1122 | if (!jh) { | |
9ea7a0df TT |
1123 | ret = -EUCLEAN; |
1124 | goto out; | |
1125 | } | |
ad56edad JK |
1126 | jbd_debug(5, "journal_head %p\n", jh); |
1127 | JBUFFER_TRACE(jh, "entry"); | |
470decc6 DK |
1128 | |
1129 | jbd_lock_bh_state(bh); | |
1130 | ||
1131 | if (jh->b_modified == 0) { | |
1132 | /* | |
1133 | * This buffer's got modified and becoming part | |
1134 | * of the transaction. This needs to be done | |
1135 | * once a transaction -bzzz | |
1136 | */ | |
1137 | jh->b_modified = 1; | |
1138 | J_ASSERT_JH(jh, handle->h_buffer_credits > 0); | |
1139 | handle->h_buffer_credits--; | |
1140 | } | |
1141 | ||
1142 | /* | |
1143 | * fastpath, to avoid expensive locking. If this buffer is already | |
1144 | * on the running transaction's metadata list there is nothing to do. | |
1145 | * Nobody can take it off again because there is a handle open. | |
1146 | * I _think_ we're OK here with SMP barriers - a mistaken decision will | |
1147 | * result in this test being false, so we go in and take the locks. | |
1148 | */ | |
1149 | if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) { | |
1150 | JBUFFER_TRACE(jh, "fastpath"); | |
9ea7a0df TT |
1151 | if (unlikely(jh->b_transaction != |
1152 | journal->j_running_transaction)) { | |
1153 | printk(KERN_EMERG "JBD: %s: " | |
1154 | "jh->b_transaction (%llu, %p, %u) != " | |
1155 | "journal->j_running_transaction (%p, %u)", | |
1156 | journal->j_devname, | |
1157 | (unsigned long long) bh->b_blocknr, | |
1158 | jh->b_transaction, | |
1159 | jh->b_transaction ? jh->b_transaction->t_tid : 0, | |
1160 | journal->j_running_transaction, | |
1161 | journal->j_running_transaction ? | |
1162 | journal->j_running_transaction->t_tid : 0); | |
1163 | ret = -EINVAL; | |
1164 | } | |
470decc6 DK |
1165 | goto out_unlock_bh; |
1166 | } | |
1167 | ||
1168 | set_buffer_jbddirty(bh); | |
1169 | ||
1170 | /* | |
1171 | * Metadata already on the current transaction list doesn't | |
1172 | * need to be filed. Metadata on another transaction's list must | |
1173 | * be committing, and will be refiled once the commit completes: | |
1174 | * leave it alone for now. | |
1175 | */ | |
1176 | if (jh->b_transaction != transaction) { | |
1177 | JBUFFER_TRACE(jh, "already on other transaction"); | |
9ea7a0df TT |
1178 | if (unlikely(jh->b_transaction != |
1179 | journal->j_committing_transaction)) { | |
1180 | printk(KERN_EMERG "JBD: %s: " | |
1181 | "jh->b_transaction (%llu, %p, %u) != " | |
1182 | "journal->j_committing_transaction (%p, %u)", | |
1183 | journal->j_devname, | |
1184 | (unsigned long long) bh->b_blocknr, | |
1185 | jh->b_transaction, | |
1186 | jh->b_transaction ? jh->b_transaction->t_tid : 0, | |
1187 | journal->j_committing_transaction, | |
1188 | journal->j_committing_transaction ? | |
1189 | journal->j_committing_transaction->t_tid : 0); | |
1190 | ret = -EINVAL; | |
1191 | } | |
1192 | if (unlikely(jh->b_next_transaction != transaction)) { | |
1193 | printk(KERN_EMERG "JBD: %s: " | |
1194 | "jh->b_next_transaction (%llu, %p, %u) != " | |
1195 | "transaction (%p, %u)", | |
1196 | journal->j_devname, | |
1197 | (unsigned long long) bh->b_blocknr, | |
1198 | jh->b_next_transaction, | |
1199 | jh->b_next_transaction ? | |
1200 | jh->b_next_transaction->t_tid : 0, | |
1201 | transaction, transaction->t_tid); | |
1202 | ret = -EINVAL; | |
1203 | } | |
470decc6 DK |
1204 | /* And this case is illegal: we can't reuse another |
1205 | * transaction's data buffer, ever. */ | |
1206 | goto out_unlock_bh; | |
1207 | } | |
1208 | ||
1209 | /* That test should have eliminated the following case: */ | |
4019191b | 1210 | J_ASSERT_JH(jh, jh->b_frozen_data == NULL); |
470decc6 DK |
1211 | |
1212 | JBUFFER_TRACE(jh, "file as BJ_Metadata"); | |
1213 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 1214 | __jbd2_journal_file_buffer(jh, handle->h_transaction, BJ_Metadata); |
470decc6 DK |
1215 | spin_unlock(&journal->j_list_lock); |
1216 | out_unlock_bh: | |
1217 | jbd_unlock_bh_state(bh); | |
ad56edad | 1218 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1219 | out: |
1220 | JBUFFER_TRACE(jh, "exit"); | |
44705754 | 1221 | WARN_ON(ret); /* All errors are bugs, so dump the stack */ |
9ea7a0df | 1222 | return ret; |
470decc6 DK |
1223 | } |
1224 | ||
470decc6 | 1225 | /** |
f7f4bccb | 1226 | * void jbd2_journal_forget() - bforget() for potentially-journaled buffers. |
470decc6 DK |
1227 | * @handle: transaction handle |
1228 | * @bh: bh to 'forget' | |
1229 | * | |
1230 | * We can only do the bforget if there are no commits pending against the | |
1231 | * buffer. If the buffer is dirty in the current running transaction we | |
1232 | * can safely unlink it. | |
1233 | * | |
1234 | * bh may not be a journalled buffer at all - it may be a non-JBD | |
1235 | * buffer which came off the hashtable. Check for this. | |
1236 | * | |
1237 | * Decrements bh->b_count by one. | |
1238 | * | |
1239 | * Allow this call even if the handle has aborted --- it may be part of | |
1240 | * the caller's cleanup after an abort. | |
1241 | */ | |
f7f4bccb | 1242 | int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1243 | { |
1244 | transaction_t *transaction = handle->h_transaction; | |
1245 | journal_t *journal = transaction->t_journal; | |
1246 | struct journal_head *jh; | |
1247 | int drop_reserve = 0; | |
1248 | int err = 0; | |
1dfc3220 | 1249 | int was_modified = 0; |
470decc6 DK |
1250 | |
1251 | BUFFER_TRACE(bh, "entry"); | |
1252 | ||
1253 | jbd_lock_bh_state(bh); | |
1254 | spin_lock(&journal->j_list_lock); | |
1255 | ||
1256 | if (!buffer_jbd(bh)) | |
1257 | goto not_jbd; | |
1258 | jh = bh2jh(bh); | |
1259 | ||
1260 | /* Critical error: attempting to delete a bitmap buffer, maybe? | |
1261 | * Don't do any jbd operations, and return an error. */ | |
1262 | if (!J_EXPECT_JH(jh, !jh->b_committed_data, | |
1263 | "inconsistent data on disk")) { | |
1264 | err = -EIO; | |
1265 | goto not_jbd; | |
1266 | } | |
1267 | ||
48fc7f7e | 1268 | /* keep track of whether or not this transaction modified us */ |
1dfc3220 JB |
1269 | was_modified = jh->b_modified; |
1270 | ||
470decc6 DK |
1271 | /* |
1272 | * The buffer's going from the transaction, we must drop | |
1273 | * all references -bzzz | |
1274 | */ | |
1275 | jh->b_modified = 0; | |
1276 | ||
1277 | if (jh->b_transaction == handle->h_transaction) { | |
1278 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
1279 | ||
1280 | /* If we are forgetting a buffer which is already part | |
1281 | * of this transaction, then we can just drop it from | |
1282 | * the transaction immediately. */ | |
1283 | clear_buffer_dirty(bh); | |
1284 | clear_buffer_jbddirty(bh); | |
1285 | ||
1286 | JBUFFER_TRACE(jh, "belongs to current transaction: unfile"); | |
1287 | ||
1dfc3220 JB |
1288 | /* |
1289 | * we only want to drop a reference if this transaction | |
1290 | * modified the buffer | |
1291 | */ | |
1292 | if (was_modified) | |
1293 | drop_reserve = 1; | |
470decc6 DK |
1294 | |
1295 | /* | |
1296 | * We are no longer going to journal this buffer. | |
1297 | * However, the commit of this transaction is still | |
1298 | * important to the buffer: the delete that we are now | |
1299 | * processing might obsolete an old log entry, so by | |
1300 | * committing, we can satisfy the buffer's checkpoint. | |
1301 | * | |
1302 | * So, if we have a checkpoint on the buffer, we should | |
1303 | * now refile the buffer on our BJ_Forget list so that | |
1304 | * we know to remove the checkpoint after we commit. | |
1305 | */ | |
1306 | ||
1307 | if (jh->b_cp_transaction) { | |
f7f4bccb MC |
1308 | __jbd2_journal_temp_unlink_buffer(jh); |
1309 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); | |
470decc6 | 1310 | } else { |
f7f4bccb | 1311 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
1312 | if (!buffer_jbd(bh)) { |
1313 | spin_unlock(&journal->j_list_lock); | |
1314 | jbd_unlock_bh_state(bh); | |
1315 | __bforget(bh); | |
1316 | goto drop; | |
1317 | } | |
1318 | } | |
1319 | } else if (jh->b_transaction) { | |
1320 | J_ASSERT_JH(jh, (jh->b_transaction == | |
1321 | journal->j_committing_transaction)); | |
1322 | /* However, if the buffer is still owned by a prior | |
1323 | * (committing) transaction, we can't drop it yet... */ | |
1324 | JBUFFER_TRACE(jh, "belongs to older transaction"); | |
1325 | /* ... but we CAN drop it from the new transaction if we | |
1326 | * have also modified it since the original commit. */ | |
1327 | ||
1328 | if (jh->b_next_transaction) { | |
1329 | J_ASSERT(jh->b_next_transaction == transaction); | |
1330 | jh->b_next_transaction = NULL; | |
1dfc3220 JB |
1331 | |
1332 | /* | |
1333 | * only drop a reference if this transaction modified | |
1334 | * the buffer | |
1335 | */ | |
1336 | if (was_modified) | |
1337 | drop_reserve = 1; | |
470decc6 DK |
1338 | } |
1339 | } | |
1340 | ||
1341 | not_jbd: | |
1342 | spin_unlock(&journal->j_list_lock); | |
1343 | jbd_unlock_bh_state(bh); | |
1344 | __brelse(bh); | |
1345 | drop: | |
1346 | if (drop_reserve) { | |
1347 | /* no need to reserve log space for this block -bzzz */ | |
1348 | handle->h_buffer_credits++; | |
1349 | } | |
1350 | return err; | |
1351 | } | |
1352 | ||
1353 | /** | |
f7f4bccb | 1354 | * int jbd2_journal_stop() - complete a transaction |
470decc6 DK |
1355 | * @handle: tranaction to complete. |
1356 | * | |
1357 | * All done for a particular handle. | |
1358 | * | |
1359 | * There is not much action needed here. We just return any remaining | |
1360 | * buffer credits to the transaction and remove the handle. The only | |
1361 | * complication is that we need to start a commit operation if the | |
1362 | * filesystem is marked for synchronous update. | |
1363 | * | |
f7f4bccb | 1364 | * jbd2_journal_stop itself will not usually return an error, but it may |
470decc6 | 1365 | * do so in unusual circumstances. In particular, expect it to |
f7f4bccb | 1366 | * return -EIO if a jbd2_journal_abort has been executed since the |
470decc6 DK |
1367 | * transaction began. |
1368 | */ | |
f7f4bccb | 1369 | int jbd2_journal_stop(handle_t *handle) |
470decc6 DK |
1370 | { |
1371 | transaction_t *transaction = handle->h_transaction; | |
1372 | journal_t *journal = transaction->t_journal; | |
a51dca9c TT |
1373 | int err, wait_for_commit = 0; |
1374 | tid_t tid; | |
470decc6 DK |
1375 | pid_t pid; |
1376 | ||
470decc6 DK |
1377 | J_ASSERT(journal_current_handle() == handle); |
1378 | ||
1379 | if (is_handle_aborted(handle)) | |
1380 | err = -EIO; | |
3e2a532b | 1381 | else { |
a51dca9c | 1382 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 | 1383 | err = 0; |
3e2a532b | 1384 | } |
470decc6 DK |
1385 | |
1386 | if (--handle->h_ref > 0) { | |
1387 | jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1, | |
1388 | handle->h_ref); | |
1389 | return err; | |
1390 | } | |
1391 | ||
1392 | jbd_debug(4, "Handle %p going down\n", handle); | |
343d9c28 TT |
1393 | trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev, |
1394 | handle->h_transaction->t_tid, | |
1395 | handle->h_type, handle->h_line_no, | |
1396 | jiffies - handle->h_start_jiffies, | |
1397 | handle->h_sync, handle->h_requested_credits, | |
1398 | (handle->h_requested_credits - | |
1399 | handle->h_buffer_credits)); | |
470decc6 DK |
1400 | |
1401 | /* | |
1402 | * Implement synchronous transaction batching. If the handle | |
1403 | * was synchronous, don't force a commit immediately. Let's | |
e07f7183 JB |
1404 | * yield and let another thread piggyback onto this |
1405 | * transaction. Keep doing that while new threads continue to | |
1406 | * arrive. It doesn't cost much - we're about to run a commit | |
1407 | * and sleep on IO anyway. Speeds up many-threaded, many-dir | |
1408 | * operations by 30x or more... | |
1409 | * | |
1410 | * We try and optimize the sleep time against what the | |
1411 | * underlying disk can do, instead of having a static sleep | |
1412 | * time. This is useful for the case where our storage is so | |
1413 | * fast that it is more optimal to go ahead and force a flush | |
1414 | * and wait for the transaction to be committed than it is to | |
1415 | * wait for an arbitrary amount of time for new writers to | |
1416 | * join the transaction. We achieve this by measuring how | |
1417 | * long it takes to commit a transaction, and compare it with | |
1418 | * how long this transaction has been running, and if run time | |
1419 | * < commit time then we sleep for the delta and commit. This | |
1420 | * greatly helps super fast disks that would see slowdowns as | |
1421 | * more threads started doing fsyncs. | |
470decc6 | 1422 | * |
e07f7183 JB |
1423 | * But don't do this if this process was the most recent one |
1424 | * to perform a synchronous write. We do this to detect the | |
1425 | * case where a single process is doing a stream of sync | |
1426 | * writes. No point in waiting for joiners in that case. | |
470decc6 DK |
1427 | */ |
1428 | pid = current->pid; | |
1429 | if (handle->h_sync && journal->j_last_sync_writer != pid) { | |
e07f7183 JB |
1430 | u64 commit_time, trans_time; |
1431 | ||
470decc6 | 1432 | journal->j_last_sync_writer = pid; |
e07f7183 | 1433 | |
a931da6a | 1434 | read_lock(&journal->j_state_lock); |
e07f7183 | 1435 | commit_time = journal->j_average_commit_time; |
a931da6a | 1436 | read_unlock(&journal->j_state_lock); |
e07f7183 JB |
1437 | |
1438 | trans_time = ktime_to_ns(ktime_sub(ktime_get(), | |
1439 | transaction->t_start_time)); | |
1440 | ||
30773840 TT |
1441 | commit_time = max_t(u64, commit_time, |
1442 | 1000*journal->j_min_batch_time); | |
e07f7183 | 1443 | commit_time = min_t(u64, commit_time, |
30773840 | 1444 | 1000*journal->j_max_batch_time); |
e07f7183 JB |
1445 | |
1446 | if (trans_time < commit_time) { | |
1447 | ktime_t expires = ktime_add_ns(ktime_get(), | |
1448 | commit_time); | |
1449 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1450 | schedule_hrtimeout(&expires, HRTIMER_MODE_ABS); | |
1451 | } | |
470decc6 DK |
1452 | } |
1453 | ||
7058548c TT |
1454 | if (handle->h_sync) |
1455 | transaction->t_synchronous_commit = 1; | |
470decc6 | 1456 | current->journal_info = NULL; |
a51dca9c TT |
1457 | atomic_sub(handle->h_buffer_credits, |
1458 | &transaction->t_outstanding_credits); | |
470decc6 DK |
1459 | |
1460 | /* | |
1461 | * If the handle is marked SYNC, we need to set another commit | |
1462 | * going! We also want to force a commit if the current | |
1463 | * transaction is occupying too much of the log, or if the | |
1464 | * transaction is too old now. | |
1465 | */ | |
1466 | if (handle->h_sync || | |
a51dca9c TT |
1467 | (atomic_read(&transaction->t_outstanding_credits) > |
1468 | journal->j_max_transaction_buffers) || | |
1469 | time_after_eq(jiffies, transaction->t_expires)) { | |
470decc6 DK |
1470 | /* Do this even for aborted journals: an abort still |
1471 | * completes the commit thread, it just doesn't write | |
1472 | * anything to disk. */ | |
470decc6 | 1473 | |
470decc6 DK |
1474 | jbd_debug(2, "transaction too old, requesting commit for " |
1475 | "handle %p\n", handle); | |
1476 | /* This is non-blocking */ | |
c35a56a0 | 1477 | jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1478 | |
1479 | /* | |
f7f4bccb | 1480 | * Special case: JBD2_SYNC synchronous updates require us |
470decc6 DK |
1481 | * to wait for the commit to complete. |
1482 | */ | |
1483 | if (handle->h_sync && !(current->flags & PF_MEMALLOC)) | |
a51dca9c | 1484 | wait_for_commit = 1; |
470decc6 DK |
1485 | } |
1486 | ||
a51dca9c TT |
1487 | /* |
1488 | * Once we drop t_updates, if it goes to zero the transaction | |
25985edc | 1489 | * could start committing on us and eventually disappear. So |
a51dca9c TT |
1490 | * once we do this, we must not dereference transaction |
1491 | * pointer again. | |
1492 | */ | |
1493 | tid = transaction->t_tid; | |
1494 | if (atomic_dec_and_test(&transaction->t_updates)) { | |
1495 | wake_up(&journal->j_wait_updates); | |
1496 | if (journal->j_barrier_count) | |
1497 | wake_up(&journal->j_wait_transaction_locked); | |
1498 | } | |
1499 | ||
1500 | if (wait_for_commit) | |
1501 | err = jbd2_log_wait_commit(journal, tid); | |
1502 | ||
3295f0ef | 1503 | lock_map_release(&handle->h_lockdep_map); |
7b751066 | 1504 | |
af1e76d6 | 1505 | jbd2_free_handle(handle); |
470decc6 DK |
1506 | return err; |
1507 | } | |
1508 | ||
5648ba5b RD |
1509 | /** |
1510 | * int jbd2_journal_force_commit() - force any uncommitted transactions | |
470decc6 DK |
1511 | * @journal: journal to force |
1512 | * | |
1513 | * For synchronous operations: force any uncommitted transactions | |
1514 | * to disk. May seem kludgy, but it reuses all the handle batching | |
1515 | * code in a very simple manner. | |
1516 | */ | |
f7f4bccb | 1517 | int jbd2_journal_force_commit(journal_t *journal) |
470decc6 DK |
1518 | { |
1519 | handle_t *handle; | |
1520 | int ret; | |
1521 | ||
f7f4bccb | 1522 | handle = jbd2_journal_start(journal, 1); |
470decc6 DK |
1523 | if (IS_ERR(handle)) { |
1524 | ret = PTR_ERR(handle); | |
1525 | } else { | |
1526 | handle->h_sync = 1; | |
f7f4bccb | 1527 | ret = jbd2_journal_stop(handle); |
470decc6 DK |
1528 | } |
1529 | return ret; | |
1530 | } | |
1531 | ||
1532 | /* | |
1533 | * | |
1534 | * List management code snippets: various functions for manipulating the | |
1535 | * transaction buffer lists. | |
1536 | * | |
1537 | */ | |
1538 | ||
1539 | /* | |
1540 | * Append a buffer to a transaction list, given the transaction's list head | |
1541 | * pointer. | |
1542 | * | |
1543 | * j_list_lock is held. | |
1544 | * | |
1545 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1546 | */ | |
1547 | ||
1548 | static inline void | |
1549 | __blist_add_buffer(struct journal_head **list, struct journal_head *jh) | |
1550 | { | |
1551 | if (!*list) { | |
1552 | jh->b_tnext = jh->b_tprev = jh; | |
1553 | *list = jh; | |
1554 | } else { | |
1555 | /* Insert at the tail of the list to preserve order */ | |
1556 | struct journal_head *first = *list, *last = first->b_tprev; | |
1557 | jh->b_tprev = last; | |
1558 | jh->b_tnext = first; | |
1559 | last->b_tnext = first->b_tprev = jh; | |
1560 | } | |
1561 | } | |
1562 | ||
1563 | /* | |
1564 | * Remove a buffer from a transaction list, given the transaction's list | |
1565 | * head pointer. | |
1566 | * | |
1567 | * Called with j_list_lock held, and the journal may not be locked. | |
1568 | * | |
1569 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1570 | */ | |
1571 | ||
1572 | static inline void | |
1573 | __blist_del_buffer(struct journal_head **list, struct journal_head *jh) | |
1574 | { | |
1575 | if (*list == jh) { | |
1576 | *list = jh->b_tnext; | |
1577 | if (*list == jh) | |
1578 | *list = NULL; | |
1579 | } | |
1580 | jh->b_tprev->b_tnext = jh->b_tnext; | |
1581 | jh->b_tnext->b_tprev = jh->b_tprev; | |
1582 | } | |
1583 | ||
1584 | /* | |
1585 | * Remove a buffer from the appropriate transaction list. | |
1586 | * | |
1587 | * Note that this function can *change* the value of | |
f5113eff JK |
1588 | * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or |
1589 | * t_reserved_list. If the caller is holding onto a copy of one of these | |
1590 | * pointers, it could go bad. Generally the caller needs to re-read the | |
1591 | * pointer from the transaction_t. | |
470decc6 | 1592 | * |
5bebccf9 | 1593 | * Called under j_list_lock. |
470decc6 | 1594 | */ |
5bebccf9 | 1595 | static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh) |
470decc6 DK |
1596 | { |
1597 | struct journal_head **list = NULL; | |
1598 | transaction_t *transaction; | |
1599 | struct buffer_head *bh = jh2bh(jh); | |
1600 | ||
1601 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
1602 | transaction = jh->b_transaction; | |
1603 | if (transaction) | |
1604 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
1605 | ||
1606 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
1607 | if (jh->b_jlist != BJ_None) | |
4019191b | 1608 | J_ASSERT_JH(jh, transaction != NULL); |
470decc6 DK |
1609 | |
1610 | switch (jh->b_jlist) { | |
1611 | case BJ_None: | |
1612 | return; | |
470decc6 DK |
1613 | case BJ_Metadata: |
1614 | transaction->t_nr_buffers--; | |
1615 | J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0); | |
1616 | list = &transaction->t_buffers; | |
1617 | break; | |
1618 | case BJ_Forget: | |
1619 | list = &transaction->t_forget; | |
1620 | break; | |
470decc6 DK |
1621 | case BJ_Shadow: |
1622 | list = &transaction->t_shadow_list; | |
1623 | break; | |
470decc6 DK |
1624 | case BJ_Reserved: |
1625 | list = &transaction->t_reserved_list; | |
1626 | break; | |
470decc6 DK |
1627 | } |
1628 | ||
1629 | __blist_del_buffer(list, jh); | |
1630 | jh->b_jlist = BJ_None; | |
1631 | if (test_clear_buffer_jbddirty(bh)) | |
1632 | mark_buffer_dirty(bh); /* Expose it to the VM */ | |
1633 | } | |
1634 | ||
de1b7941 JK |
1635 | /* |
1636 | * Remove buffer from all transactions. | |
1637 | * | |
1638 | * Called with bh_state lock and j_list_lock | |
1639 | * | |
1640 | * jh and bh may be already freed when this function returns. | |
1641 | */ | |
1642 | static void __jbd2_journal_unfile_buffer(struct journal_head *jh) | |
470decc6 | 1643 | { |
f7f4bccb | 1644 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 | 1645 | jh->b_transaction = NULL; |
de1b7941 | 1646 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1647 | } |
1648 | ||
f7f4bccb | 1649 | void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 | 1650 | { |
de1b7941 JK |
1651 | struct buffer_head *bh = jh2bh(jh); |
1652 | ||
1653 | /* Get reference so that buffer cannot be freed before we unlock it */ | |
1654 | get_bh(bh); | |
1655 | jbd_lock_bh_state(bh); | |
470decc6 | 1656 | spin_lock(&journal->j_list_lock); |
f7f4bccb | 1657 | __jbd2_journal_unfile_buffer(jh); |
470decc6 | 1658 | spin_unlock(&journal->j_list_lock); |
de1b7941 JK |
1659 | jbd_unlock_bh_state(bh); |
1660 | __brelse(bh); | |
470decc6 DK |
1661 | } |
1662 | ||
1663 | /* | |
f7f4bccb | 1664 | * Called from jbd2_journal_try_to_free_buffers(). |
470decc6 DK |
1665 | * |
1666 | * Called under jbd_lock_bh_state(bh) | |
1667 | */ | |
1668 | static void | |
1669 | __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh) | |
1670 | { | |
1671 | struct journal_head *jh; | |
1672 | ||
1673 | jh = bh2jh(bh); | |
1674 | ||
1675 | if (buffer_locked(bh) || buffer_dirty(bh)) | |
1676 | goto out; | |
1677 | ||
4019191b | 1678 | if (jh->b_next_transaction != NULL) |
470decc6 DK |
1679 | goto out; |
1680 | ||
1681 | spin_lock(&journal->j_list_lock); | |
87c89c23 | 1682 | if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) { |
470decc6 | 1683 | /* written-back checkpointed metadata buffer */ |
c254c9ec JK |
1684 | JBUFFER_TRACE(jh, "remove from checkpoint list"); |
1685 | __jbd2_journal_remove_checkpoint(jh); | |
470decc6 DK |
1686 | } |
1687 | spin_unlock(&journal->j_list_lock); | |
1688 | out: | |
1689 | return; | |
1690 | } | |
1691 | ||
470decc6 | 1692 | /** |
f7f4bccb | 1693 | * int jbd2_journal_try_to_free_buffers() - try to free page buffers. |
470decc6 DK |
1694 | * @journal: journal for operation |
1695 | * @page: to try and free | |
530576bb MC |
1696 | * @gfp_mask: we use the mask to detect how hard should we try to release |
1697 | * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to | |
1698 | * release the buffers. | |
470decc6 DK |
1699 | * |
1700 | * | |
1701 | * For all the buffers on this page, | |
1702 | * if they are fully written out ordered data, move them onto BUF_CLEAN | |
1703 | * so try_to_free_buffers() can reap them. | |
1704 | * | |
1705 | * This function returns non-zero if we wish try_to_free_buffers() | |
1706 | * to be called. We do this if the page is releasable by try_to_free_buffers(). | |
1707 | * We also do it if the page has locked or dirty buffers and the caller wants | |
1708 | * us to perform sync or async writeout. | |
1709 | * | |
1710 | * This complicates JBD locking somewhat. We aren't protected by the | |
1711 | * BKL here. We wish to remove the buffer from its committing or | |
f7f4bccb | 1712 | * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer. |
470decc6 DK |
1713 | * |
1714 | * This may *change* the value of transaction_t->t_datalist, so anyone | |
1715 | * who looks at t_datalist needs to lock against this function. | |
1716 | * | |
f7f4bccb MC |
1717 | * Even worse, someone may be doing a jbd2_journal_dirty_data on this |
1718 | * buffer. So we need to lock against that. jbd2_journal_dirty_data() | |
470decc6 DK |
1719 | * will come out of the lock with the buffer dirty, which makes it |
1720 | * ineligible for release here. | |
1721 | * | |
1722 | * Who else is affected by this? hmm... Really the only contender | |
1723 | * is do_get_write_access() - it could be looking at the buffer while | |
1724 | * journal_try_to_free_buffer() is changing its state. But that | |
1725 | * cannot happen because we never reallocate freed data as metadata | |
1726 | * while the data is part of a transaction. Yes? | |
530576bb MC |
1727 | * |
1728 | * Return 0 on failure, 1 on success | |
470decc6 | 1729 | */ |
f7f4bccb | 1730 | int jbd2_journal_try_to_free_buffers(journal_t *journal, |
530576bb | 1731 | struct page *page, gfp_t gfp_mask) |
470decc6 DK |
1732 | { |
1733 | struct buffer_head *head; | |
1734 | struct buffer_head *bh; | |
1735 | int ret = 0; | |
1736 | ||
1737 | J_ASSERT(PageLocked(page)); | |
1738 | ||
1739 | head = page_buffers(page); | |
1740 | bh = head; | |
1741 | do { | |
1742 | struct journal_head *jh; | |
1743 | ||
1744 | /* | |
1745 | * We take our own ref against the journal_head here to avoid | |
1746 | * having to add tons of locking around each instance of | |
530576bb | 1747 | * jbd2_journal_put_journal_head(). |
470decc6 | 1748 | */ |
f7f4bccb | 1749 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
1750 | if (!jh) |
1751 | continue; | |
1752 | ||
1753 | jbd_lock_bh_state(bh); | |
1754 | __journal_try_to_free_buffer(journal, bh); | |
f7f4bccb | 1755 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1756 | jbd_unlock_bh_state(bh); |
1757 | if (buffer_jbd(bh)) | |
1758 | goto busy; | |
1759 | } while ((bh = bh->b_this_page) != head); | |
530576bb | 1760 | |
470decc6 | 1761 | ret = try_to_free_buffers(page); |
530576bb | 1762 | |
470decc6 DK |
1763 | busy: |
1764 | return ret; | |
1765 | } | |
1766 | ||
1767 | /* | |
1768 | * This buffer is no longer needed. If it is on an older transaction's | |
1769 | * checkpoint list we need to record it on this transaction's forget list | |
1770 | * to pin this buffer (and hence its checkpointing transaction) down until | |
1771 | * this transaction commits. If the buffer isn't on a checkpoint list, we | |
1772 | * release it. | |
1773 | * Returns non-zero if JBD no longer has an interest in the buffer. | |
1774 | * | |
1775 | * Called under j_list_lock. | |
1776 | * | |
1777 | * Called under jbd_lock_bh_state(bh). | |
1778 | */ | |
1779 | static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction) | |
1780 | { | |
1781 | int may_free = 1; | |
1782 | struct buffer_head *bh = jh2bh(jh); | |
1783 | ||
470decc6 DK |
1784 | if (jh->b_cp_transaction) { |
1785 | JBUFFER_TRACE(jh, "on running+cp transaction"); | |
de1b7941 | 1786 | __jbd2_journal_temp_unlink_buffer(jh); |
f91d1d04 JK |
1787 | /* |
1788 | * We don't want to write the buffer anymore, clear the | |
1789 | * bit so that we don't confuse checks in | |
1790 | * __journal_file_buffer | |
1791 | */ | |
1792 | clear_buffer_dirty(bh); | |
f7f4bccb | 1793 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); |
470decc6 DK |
1794 | may_free = 0; |
1795 | } else { | |
1796 | JBUFFER_TRACE(jh, "on running transaction"); | |
de1b7941 | 1797 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
1798 | } |
1799 | return may_free; | |
1800 | } | |
1801 | ||
1802 | /* | |
f7f4bccb | 1803 | * jbd2_journal_invalidatepage |
470decc6 DK |
1804 | * |
1805 | * This code is tricky. It has a number of cases to deal with. | |
1806 | * | |
1807 | * There are two invariants which this code relies on: | |
1808 | * | |
1809 | * i_size must be updated on disk before we start calling invalidatepage on the | |
1810 | * data. | |
1811 | * | |
1812 | * This is done in ext3 by defining an ext3_setattr method which | |
1813 | * updates i_size before truncate gets going. By maintaining this | |
1814 | * invariant, we can be sure that it is safe to throw away any buffers | |
1815 | * attached to the current transaction: once the transaction commits, | |
1816 | * we know that the data will not be needed. | |
1817 | * | |
1818 | * Note however that we can *not* throw away data belonging to the | |
1819 | * previous, committing transaction! | |
1820 | * | |
1821 | * Any disk blocks which *are* part of the previous, committing | |
1822 | * transaction (and which therefore cannot be discarded immediately) are | |
1823 | * not going to be reused in the new running transaction | |
1824 | * | |
1825 | * The bitmap committed_data images guarantee this: any block which is | |
1826 | * allocated in one transaction and removed in the next will be marked | |
1827 | * as in-use in the committed_data bitmap, so cannot be reused until | |
1828 | * the next transaction to delete the block commits. This means that | |
1829 | * leaving committing buffers dirty is quite safe: the disk blocks | |
1830 | * cannot be reallocated to a different file and so buffer aliasing is | |
1831 | * not possible. | |
1832 | * | |
1833 | * | |
1834 | * The above applies mainly to ordered data mode. In writeback mode we | |
1835 | * don't make guarantees about the order in which data hits disk --- in | |
1836 | * particular we don't guarantee that new dirty data is flushed before | |
1837 | * transaction commit --- so it is always safe just to discard data | |
1838 | * immediately in that mode. --sct | |
1839 | */ | |
1840 | ||
1841 | /* | |
1842 | * The journal_unmap_buffer helper function returns zero if the buffer | |
1843 | * concerned remains pinned as an anonymous buffer belonging to an older | |
1844 | * transaction. | |
1845 | * | |
1846 | * We're outside-transaction here. Either or both of j_running_transaction | |
1847 | * and j_committing_transaction may be NULL. | |
1848 | */ | |
b794e7a6 JK |
1849 | static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh, |
1850 | int partial_page) | |
470decc6 DK |
1851 | { |
1852 | transaction_t *transaction; | |
1853 | struct journal_head *jh; | |
1854 | int may_free = 1; | |
470decc6 DK |
1855 | |
1856 | BUFFER_TRACE(bh, "entry"); | |
1857 | ||
1858 | /* | |
1859 | * It is safe to proceed here without the j_list_lock because the | |
1860 | * buffers cannot be stolen by try_to_free_buffers as long as we are | |
1861 | * holding the page lock. --sct | |
1862 | */ | |
1863 | ||
1864 | if (!buffer_jbd(bh)) | |
1865 | goto zap_buffer_unlocked; | |
1866 | ||
87c89c23 | 1867 | /* OK, we have data buffer in journaled mode */ |
a931da6a | 1868 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1869 | jbd_lock_bh_state(bh); |
1870 | spin_lock(&journal->j_list_lock); | |
1871 | ||
f7f4bccb | 1872 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
1873 | if (!jh) |
1874 | goto zap_buffer_no_jh; | |
1875 | ||
ba869023 | 1876 | /* |
1877 | * We cannot remove the buffer from checkpoint lists until the | |
1878 | * transaction adding inode to orphan list (let's call it T) | |
1879 | * is committed. Otherwise if the transaction changing the | |
1880 | * buffer would be cleaned from the journal before T is | |
1881 | * committed, a crash will cause that the correct contents of | |
1882 | * the buffer will be lost. On the other hand we have to | |
1883 | * clear the buffer dirty bit at latest at the moment when the | |
1884 | * transaction marking the buffer as freed in the filesystem | |
1885 | * structures is committed because from that moment on the | |
b794e7a6 | 1886 | * block can be reallocated and used by a different page. |
ba869023 | 1887 | * Since the block hasn't been freed yet but the inode has |
1888 | * already been added to orphan list, it is safe for us to add | |
1889 | * the buffer to BJ_Forget list of the newest transaction. | |
b794e7a6 JK |
1890 | * |
1891 | * Also we have to clear buffer_mapped flag of a truncated buffer | |
1892 | * because the buffer_head may be attached to the page straddling | |
1893 | * i_size (can happen only when blocksize < pagesize) and thus the | |
1894 | * buffer_head can be reused when the file is extended again. So we end | |
1895 | * up keeping around invalidated buffers attached to transactions' | |
1896 | * BJ_Forget list just to stop checkpointing code from cleaning up | |
1897 | * the transaction this buffer was modified in. | |
ba869023 | 1898 | */ |
470decc6 DK |
1899 | transaction = jh->b_transaction; |
1900 | if (transaction == NULL) { | |
1901 | /* First case: not on any transaction. If it | |
1902 | * has no checkpoint link, then we can zap it: | |
1903 | * it's a writeback-mode buffer so we don't care | |
1904 | * if it hits disk safely. */ | |
1905 | if (!jh->b_cp_transaction) { | |
1906 | JBUFFER_TRACE(jh, "not on any transaction: zap"); | |
1907 | goto zap_buffer; | |
1908 | } | |
1909 | ||
1910 | if (!buffer_dirty(bh)) { | |
1911 | /* bdflush has written it. We can drop it now */ | |
1912 | goto zap_buffer; | |
1913 | } | |
1914 | ||
1915 | /* OK, it must be in the journal but still not | |
1916 | * written fully to disk: it's metadata or | |
1917 | * journaled data... */ | |
1918 | ||
1919 | if (journal->j_running_transaction) { | |
1920 | /* ... and once the current transaction has | |
1921 | * committed, the buffer won't be needed any | |
1922 | * longer. */ | |
1923 | JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget"); | |
b794e7a6 | 1924 | may_free = __dispose_buffer(jh, |
470decc6 | 1925 | journal->j_running_transaction); |
b794e7a6 | 1926 | goto zap_buffer; |
470decc6 DK |
1927 | } else { |
1928 | /* There is no currently-running transaction. So the | |
1929 | * orphan record which we wrote for this file must have | |
1930 | * passed into commit. We must attach this buffer to | |
1931 | * the committing transaction, if it exists. */ | |
1932 | if (journal->j_committing_transaction) { | |
1933 | JBUFFER_TRACE(jh, "give to committing trans"); | |
b794e7a6 | 1934 | may_free = __dispose_buffer(jh, |
470decc6 | 1935 | journal->j_committing_transaction); |
b794e7a6 | 1936 | goto zap_buffer; |
470decc6 DK |
1937 | } else { |
1938 | /* The orphan record's transaction has | |
1939 | * committed. We can cleanse this buffer */ | |
1940 | clear_buffer_jbddirty(bh); | |
1941 | goto zap_buffer; | |
1942 | } | |
1943 | } | |
1944 | } else if (transaction == journal->j_committing_transaction) { | |
9b57988d | 1945 | JBUFFER_TRACE(jh, "on committing transaction"); |
470decc6 | 1946 | /* |
ba869023 | 1947 | * The buffer is committing, we simply cannot touch |
b794e7a6 JK |
1948 | * it. If the page is straddling i_size we have to wait |
1949 | * for commit and try again. | |
1950 | */ | |
1951 | if (partial_page) { | |
b794e7a6 JK |
1952 | jbd2_journal_put_journal_head(jh); |
1953 | spin_unlock(&journal->j_list_lock); | |
1954 | jbd_unlock_bh_state(bh); | |
1955 | write_unlock(&journal->j_state_lock); | |
53e87268 | 1956 | return -EBUSY; |
b794e7a6 JK |
1957 | } |
1958 | /* | |
1959 | * OK, buffer won't be reachable after truncate. We just set | |
1960 | * j_next_transaction to the running transaction (if there is | |
1961 | * one) and mark buffer as freed so that commit code knows it | |
1962 | * should clear dirty bits when it is done with the buffer. | |
ba869023 | 1963 | */ |
470decc6 | 1964 | set_buffer_freed(bh); |
ba869023 | 1965 | if (journal->j_running_transaction && buffer_jbddirty(bh)) |
1966 | jh->b_next_transaction = journal->j_running_transaction; | |
f7f4bccb | 1967 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1968 | spin_unlock(&journal->j_list_lock); |
1969 | jbd_unlock_bh_state(bh); | |
a931da6a | 1970 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1971 | return 0; |
1972 | } else { | |
1973 | /* Good, the buffer belongs to the running transaction. | |
1974 | * We are writing our own transaction's data, not any | |
1975 | * previous one's, so it is safe to throw it away | |
1976 | * (remember that we expect the filesystem to have set | |
1977 | * i_size already for this truncate so recovery will not | |
1978 | * expose the disk blocks we are discarding here.) */ | |
1979 | J_ASSERT_JH(jh, transaction == journal->j_running_transaction); | |
9b57988d | 1980 | JBUFFER_TRACE(jh, "on running transaction"); |
470decc6 DK |
1981 | may_free = __dispose_buffer(jh, transaction); |
1982 | } | |
1983 | ||
1984 | zap_buffer: | |
b794e7a6 JK |
1985 | /* |
1986 | * This is tricky. Although the buffer is truncated, it may be reused | |
1987 | * if blocksize < pagesize and it is attached to the page straddling | |
1988 | * EOF. Since the buffer might have been added to BJ_Forget list of the | |
1989 | * running transaction, journal_get_write_access() won't clear | |
1990 | * b_modified and credit accounting gets confused. So clear b_modified | |
1991 | * here. | |
1992 | */ | |
1993 | jh->b_modified = 0; | |
f7f4bccb | 1994 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1995 | zap_buffer_no_jh: |
1996 | spin_unlock(&journal->j_list_lock); | |
1997 | jbd_unlock_bh_state(bh); | |
a931da6a | 1998 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1999 | zap_buffer_unlocked: |
2000 | clear_buffer_dirty(bh); | |
2001 | J_ASSERT_BH(bh, !buffer_jbddirty(bh)); | |
2002 | clear_buffer_mapped(bh); | |
2003 | clear_buffer_req(bh); | |
2004 | clear_buffer_new(bh); | |
15291164 ES |
2005 | clear_buffer_delay(bh); |
2006 | clear_buffer_unwritten(bh); | |
470decc6 DK |
2007 | bh->b_bdev = NULL; |
2008 | return may_free; | |
2009 | } | |
2010 | ||
2011 | /** | |
f7f4bccb | 2012 | * void jbd2_journal_invalidatepage() |
470decc6 DK |
2013 | * @journal: journal to use for flush... |
2014 | * @page: page to flush | |
259709b0 LC |
2015 | * @offset: start of the range to invalidate |
2016 | * @length: length of the range to invalidate | |
470decc6 | 2017 | * |
259709b0 LC |
2018 | * Reap page buffers containing data after in the specified range in page. |
2019 | * Can return -EBUSY if buffers are part of the committing transaction and | |
2020 | * the page is straddling i_size. Caller then has to wait for current commit | |
2021 | * and try again. | |
470decc6 | 2022 | */ |
53e87268 JK |
2023 | int jbd2_journal_invalidatepage(journal_t *journal, |
2024 | struct page *page, | |
259709b0 LC |
2025 | unsigned int offset, |
2026 | unsigned int length) | |
470decc6 DK |
2027 | { |
2028 | struct buffer_head *head, *bh, *next; | |
259709b0 | 2029 | unsigned int stop = offset + length; |
470decc6 | 2030 | unsigned int curr_off = 0; |
259709b0 | 2031 | int partial_page = (offset || length < PAGE_CACHE_SIZE); |
470decc6 | 2032 | int may_free = 1; |
53e87268 | 2033 | int ret = 0; |
470decc6 DK |
2034 | |
2035 | if (!PageLocked(page)) | |
2036 | BUG(); | |
2037 | if (!page_has_buffers(page)) | |
53e87268 | 2038 | return 0; |
470decc6 | 2039 | |
259709b0 LC |
2040 | BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); |
2041 | ||
470decc6 DK |
2042 | /* We will potentially be playing with lists other than just the |
2043 | * data lists (especially for journaled data mode), so be | |
2044 | * cautious in our locking. */ | |
2045 | ||
2046 | head = bh = page_buffers(page); | |
2047 | do { | |
2048 | unsigned int next_off = curr_off + bh->b_size; | |
2049 | next = bh->b_this_page; | |
2050 | ||
259709b0 LC |
2051 | if (next_off > stop) |
2052 | return 0; | |
2053 | ||
470decc6 DK |
2054 | if (offset <= curr_off) { |
2055 | /* This block is wholly outside the truncation point */ | |
2056 | lock_buffer(bh); | |
259709b0 | 2057 | ret = journal_unmap_buffer(journal, bh, partial_page); |
470decc6 | 2058 | unlock_buffer(bh); |
53e87268 JK |
2059 | if (ret < 0) |
2060 | return ret; | |
2061 | may_free &= ret; | |
470decc6 DK |
2062 | } |
2063 | curr_off = next_off; | |
2064 | bh = next; | |
2065 | ||
2066 | } while (bh != head); | |
2067 | ||
259709b0 | 2068 | if (!partial_page) { |
470decc6 DK |
2069 | if (may_free && try_to_free_buffers(page)) |
2070 | J_ASSERT(!page_has_buffers(page)); | |
2071 | } | |
53e87268 | 2072 | return 0; |
470decc6 DK |
2073 | } |
2074 | ||
2075 | /* | |
2076 | * File a buffer on the given transaction list. | |
2077 | */ | |
f7f4bccb | 2078 | void __jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2079 | transaction_t *transaction, int jlist) |
2080 | { | |
2081 | struct journal_head **list = NULL; | |
2082 | int was_dirty = 0; | |
2083 | struct buffer_head *bh = jh2bh(jh); | |
2084 | ||
2085 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
2086 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
2087 | ||
2088 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
2089 | J_ASSERT_JH(jh, jh->b_transaction == transaction || | |
4019191b | 2090 | jh->b_transaction == NULL); |
470decc6 DK |
2091 | |
2092 | if (jh->b_transaction && jh->b_jlist == jlist) | |
2093 | return; | |
2094 | ||
470decc6 DK |
2095 | if (jlist == BJ_Metadata || jlist == BJ_Reserved || |
2096 | jlist == BJ_Shadow || jlist == BJ_Forget) { | |
f91d1d04 JK |
2097 | /* |
2098 | * For metadata buffers, we track dirty bit in buffer_jbddirty | |
2099 | * instead of buffer_dirty. We should not see a dirty bit set | |
2100 | * here because we clear it in do_get_write_access but e.g. | |
2101 | * tune2fs can modify the sb and set the dirty bit at any time | |
2102 | * so we try to gracefully handle that. | |
2103 | */ | |
2104 | if (buffer_dirty(bh)) | |
2105 | warn_dirty_buffer(bh); | |
470decc6 DK |
2106 | if (test_clear_buffer_dirty(bh) || |
2107 | test_clear_buffer_jbddirty(bh)) | |
2108 | was_dirty = 1; | |
2109 | } | |
2110 | ||
2111 | if (jh->b_transaction) | |
f7f4bccb | 2112 | __jbd2_journal_temp_unlink_buffer(jh); |
de1b7941 JK |
2113 | else |
2114 | jbd2_journal_grab_journal_head(bh); | |
470decc6 DK |
2115 | jh->b_transaction = transaction; |
2116 | ||
2117 | switch (jlist) { | |
2118 | case BJ_None: | |
2119 | J_ASSERT_JH(jh, !jh->b_committed_data); | |
2120 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
2121 | return; | |
470decc6 DK |
2122 | case BJ_Metadata: |
2123 | transaction->t_nr_buffers++; | |
2124 | list = &transaction->t_buffers; | |
2125 | break; | |
2126 | case BJ_Forget: | |
2127 | list = &transaction->t_forget; | |
2128 | break; | |
470decc6 DK |
2129 | case BJ_Shadow: |
2130 | list = &transaction->t_shadow_list; | |
2131 | break; | |
470decc6 DK |
2132 | case BJ_Reserved: |
2133 | list = &transaction->t_reserved_list; | |
2134 | break; | |
470decc6 DK |
2135 | } |
2136 | ||
2137 | __blist_add_buffer(list, jh); | |
2138 | jh->b_jlist = jlist; | |
2139 | ||
2140 | if (was_dirty) | |
2141 | set_buffer_jbddirty(bh); | |
2142 | } | |
2143 | ||
f7f4bccb | 2144 | void jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2145 | transaction_t *transaction, int jlist) |
2146 | { | |
2147 | jbd_lock_bh_state(jh2bh(jh)); | |
2148 | spin_lock(&transaction->t_journal->j_list_lock); | |
f7f4bccb | 2149 | __jbd2_journal_file_buffer(jh, transaction, jlist); |
470decc6 DK |
2150 | spin_unlock(&transaction->t_journal->j_list_lock); |
2151 | jbd_unlock_bh_state(jh2bh(jh)); | |
2152 | } | |
2153 | ||
2154 | /* | |
2155 | * Remove a buffer from its current buffer list in preparation for | |
2156 | * dropping it from its current transaction entirely. If the buffer has | |
2157 | * already started to be used by a subsequent transaction, refile the | |
2158 | * buffer on that transaction's metadata list. | |
2159 | * | |
de1b7941 | 2160 | * Called under j_list_lock |
470decc6 | 2161 | * Called under jbd_lock_bh_state(jh2bh(jh)) |
de1b7941 JK |
2162 | * |
2163 | * jh and bh may be already free when this function returns | |
470decc6 | 2164 | */ |
f7f4bccb | 2165 | void __jbd2_journal_refile_buffer(struct journal_head *jh) |
470decc6 | 2166 | { |
ba869023 | 2167 | int was_dirty, jlist; |
470decc6 DK |
2168 | struct buffer_head *bh = jh2bh(jh); |
2169 | ||
2170 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
2171 | if (jh->b_transaction) | |
2172 | assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock); | |
2173 | ||
2174 | /* If the buffer is now unused, just drop it. */ | |
2175 | if (jh->b_next_transaction == NULL) { | |
f7f4bccb | 2176 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
2177 | return; |
2178 | } | |
2179 | ||
2180 | /* | |
2181 | * It has been modified by a later transaction: add it to the new | |
2182 | * transaction's metadata list. | |
2183 | */ | |
2184 | ||
2185 | was_dirty = test_clear_buffer_jbddirty(bh); | |
f7f4bccb | 2186 | __jbd2_journal_temp_unlink_buffer(jh); |
de1b7941 JK |
2187 | /* |
2188 | * We set b_transaction here because b_next_transaction will inherit | |
2189 | * our jh reference and thus __jbd2_journal_file_buffer() must not | |
2190 | * take a new one. | |
2191 | */ | |
470decc6 DK |
2192 | jh->b_transaction = jh->b_next_transaction; |
2193 | jh->b_next_transaction = NULL; | |
ba869023 | 2194 | if (buffer_freed(bh)) |
2195 | jlist = BJ_Forget; | |
2196 | else if (jh->b_modified) | |
2197 | jlist = BJ_Metadata; | |
2198 | else | |
2199 | jlist = BJ_Reserved; | |
2200 | __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist); | |
470decc6 DK |
2201 | J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING); |
2202 | ||
2203 | if (was_dirty) | |
2204 | set_buffer_jbddirty(bh); | |
2205 | } | |
2206 | ||
2207 | /* | |
de1b7941 JK |
2208 | * __jbd2_journal_refile_buffer() with necessary locking added. We take our |
2209 | * bh reference so that we can safely unlock bh. | |
2210 | * | |
2211 | * The jh and bh may be freed by this call. | |
470decc6 | 2212 | */ |
f7f4bccb | 2213 | void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 DK |
2214 | { |
2215 | struct buffer_head *bh = jh2bh(jh); | |
2216 | ||
de1b7941 JK |
2217 | /* Get reference so that buffer cannot be freed before we unlock it */ |
2218 | get_bh(bh); | |
470decc6 DK |
2219 | jbd_lock_bh_state(bh); |
2220 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 2221 | __jbd2_journal_refile_buffer(jh); |
470decc6 | 2222 | jbd_unlock_bh_state(bh); |
470decc6 DK |
2223 | spin_unlock(&journal->j_list_lock); |
2224 | __brelse(bh); | |
2225 | } | |
c851ed54 JK |
2226 | |
2227 | /* | |
2228 | * File inode in the inode list of the handle's transaction | |
2229 | */ | |
2230 | int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode) | |
2231 | { | |
2232 | transaction_t *transaction = handle->h_transaction; | |
2233 | journal_t *journal = transaction->t_journal; | |
2234 | ||
2235 | if (is_handle_aborted(handle)) | |
2236 | return -EIO; | |
2237 | ||
2238 | jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino, | |
2239 | transaction->t_tid); | |
2240 | ||
2241 | /* | |
2242 | * First check whether inode isn't already on the transaction's | |
2243 | * lists without taking the lock. Note that this check is safe | |
2244 | * without the lock as we cannot race with somebody removing inode | |
2245 | * from the transaction. The reason is that we remove inode from the | |
2246 | * transaction only in journal_release_jbd_inode() and when we commit | |
2247 | * the transaction. We are guarded from the first case by holding | |
2248 | * a reference to the inode. We are safe against the second case | |
2249 | * because if jinode->i_transaction == transaction, commit code | |
2250 | * cannot touch the transaction because we hold reference to it, | |
2251 | * and if jinode->i_next_transaction == transaction, commit code | |
2252 | * will only file the inode where we want it. | |
2253 | */ | |
2254 | if (jinode->i_transaction == transaction || | |
2255 | jinode->i_next_transaction == transaction) | |
2256 | return 0; | |
2257 | ||
2258 | spin_lock(&journal->j_list_lock); | |
2259 | ||
2260 | if (jinode->i_transaction == transaction || | |
2261 | jinode->i_next_transaction == transaction) | |
2262 | goto done; | |
2263 | ||
81be12c8 JK |
2264 | /* |
2265 | * We only ever set this variable to 1 so the test is safe. Since | |
2266 | * t_need_data_flush is likely to be set, we do the test to save some | |
2267 | * cacheline bouncing | |
2268 | */ | |
2269 | if (!transaction->t_need_data_flush) | |
2270 | transaction->t_need_data_flush = 1; | |
c851ed54 JK |
2271 | /* On some different transaction's list - should be |
2272 | * the committing one */ | |
2273 | if (jinode->i_transaction) { | |
2274 | J_ASSERT(jinode->i_next_transaction == NULL); | |
2275 | J_ASSERT(jinode->i_transaction == | |
2276 | journal->j_committing_transaction); | |
2277 | jinode->i_next_transaction = transaction; | |
2278 | goto done; | |
2279 | } | |
2280 | /* Not on any transaction list... */ | |
2281 | J_ASSERT(!jinode->i_next_transaction); | |
2282 | jinode->i_transaction = transaction; | |
2283 | list_add(&jinode->i_list, &transaction->t_inode_list); | |
2284 | done: | |
2285 | spin_unlock(&journal->j_list_lock); | |
2286 | ||
2287 | return 0; | |
2288 | } | |
2289 | ||
2290 | /* | |
7f5aa215 JK |
2291 | * File truncate and transaction commit interact with each other in a |
2292 | * non-trivial way. If a transaction writing data block A is | |
2293 | * committing, we cannot discard the data by truncate until we have | |
2294 | * written them. Otherwise if we crashed after the transaction with | |
2295 | * write has committed but before the transaction with truncate has | |
2296 | * committed, we could see stale data in block A. This function is a | |
2297 | * helper to solve this problem. It starts writeout of the truncated | |
2298 | * part in case it is in the committing transaction. | |
2299 | * | |
2300 | * Filesystem code must call this function when inode is journaled in | |
2301 | * ordered mode before truncation happens and after the inode has been | |
2302 | * placed on orphan list with the new inode size. The second condition | |
2303 | * avoids the race that someone writes new data and we start | |
2304 | * committing the transaction after this function has been called but | |
2305 | * before a transaction for truncate is started (and furthermore it | |
2306 | * allows us to optimize the case where the addition to orphan list | |
2307 | * happens in the same transaction as write --- we don't have to write | |
2308 | * any data in such case). | |
c851ed54 | 2309 | */ |
7f5aa215 JK |
2310 | int jbd2_journal_begin_ordered_truncate(journal_t *journal, |
2311 | struct jbd2_inode *jinode, | |
c851ed54 JK |
2312 | loff_t new_size) |
2313 | { | |
7f5aa215 | 2314 | transaction_t *inode_trans, *commit_trans; |
c851ed54 JK |
2315 | int ret = 0; |
2316 | ||
7f5aa215 JK |
2317 | /* This is a quick check to avoid locking if not necessary */ |
2318 | if (!jinode->i_transaction) | |
c851ed54 | 2319 | goto out; |
7f5aa215 JK |
2320 | /* Locks are here just to force reading of recent values, it is |
2321 | * enough that the transaction was not committing before we started | |
2322 | * a transaction adding the inode to orphan list */ | |
a931da6a | 2323 | read_lock(&journal->j_state_lock); |
c851ed54 | 2324 | commit_trans = journal->j_committing_transaction; |
a931da6a | 2325 | read_unlock(&journal->j_state_lock); |
7f5aa215 JK |
2326 | spin_lock(&journal->j_list_lock); |
2327 | inode_trans = jinode->i_transaction; | |
2328 | spin_unlock(&journal->j_list_lock); | |
2329 | if (inode_trans == commit_trans) { | |
2330 | ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping, | |
c851ed54 JK |
2331 | new_size, LLONG_MAX); |
2332 | if (ret) | |
2333 | jbd2_journal_abort(journal, ret); | |
2334 | } | |
2335 | out: | |
2336 | return ret; | |
2337 | } |