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 | */ | |
286 | ||
287 | /* | |
288 | * @@@ AKPM: This seems rather over-defensive. We're giving commit | |
289 | * a _lot_ of headroom: 1/4 of the journal plus the size of | |
290 | * the committing transaction. Really, we only need to give it | |
291 | * committing_transaction->t_outstanding_credits plus "enough" for | |
292 | * the log control blocks. | |
a34f0b31 | 293 | * Also, this test is inconsistent with the matching one in |
f7f4bccb | 294 | * jbd2_journal_extend(). |
470decc6 | 295 | */ |
f7f4bccb | 296 | if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) { |
470decc6 | 297 | jbd_debug(2, "Handle %p waiting for checkpoint...\n", handle); |
8dd42046 | 298 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
a931da6a TT |
299 | read_unlock(&journal->j_state_lock); |
300 | write_lock(&journal->j_state_lock); | |
301 | if (__jbd2_log_space_left(journal) < jbd_space_needed(journal)) | |
302 | __jbd2_log_wait_for_space(journal); | |
303 | write_unlock(&journal->j_state_lock); | |
304 | goto repeat; | |
470decc6 DK |
305 | } |
306 | ||
307 | /* OK, account for the buffers that this operation expects to | |
8dd42046 | 308 | * use and add the handle to the running transaction. |
8dd42046 | 309 | */ |
28e35e42 | 310 | update_t_max_wait(transaction, ts); |
470decc6 | 311 | handle->h_transaction = transaction; |
343d9c28 TT |
312 | handle->h_requested_credits = nblocks; |
313 | handle->h_start_jiffies = jiffies; | |
a51dca9c | 314 | atomic_inc(&transaction->t_updates); |
8dd42046 | 315 | atomic_inc(&transaction->t_handle_count); |
470decc6 | 316 | jbd_debug(4, "Handle %p given %d credits (total %d, free %d)\n", |
a51dca9c TT |
317 | handle, nblocks, |
318 | atomic_read(&transaction->t_outstanding_credits), | |
f7f4bccb | 319 | __jbd2_log_space_left(journal)); |
a931da6a | 320 | read_unlock(&journal->j_state_lock); |
9599b0e5 JK |
321 | |
322 | lock_map_acquire(&handle->h_lockdep_map); | |
0c2022ec | 323 | jbd2_journal_free_transaction(new_transaction); |
47def826 | 324 | return 0; |
470decc6 DK |
325 | } |
326 | ||
7b751066 MC |
327 | static struct lock_class_key jbd2_handle_key; |
328 | ||
470decc6 DK |
329 | /* Allocate a new handle. This should probably be in a slab... */ |
330 | static handle_t *new_handle(int nblocks) | |
331 | { | |
af1e76d6 | 332 | handle_t *handle = jbd2_alloc_handle(GFP_NOFS); |
470decc6 DK |
333 | if (!handle) |
334 | return NULL; | |
470decc6 DK |
335 | handle->h_buffer_credits = nblocks; |
336 | handle->h_ref = 1; | |
337 | ||
7b751066 MC |
338 | lockdep_init_map(&handle->h_lockdep_map, "jbd2_handle", |
339 | &jbd2_handle_key, 0); | |
340 | ||
470decc6 DK |
341 | return handle; |
342 | } | |
343 | ||
344 | /** | |
f7f4bccb | 345 | * handle_t *jbd2_journal_start() - Obtain a new handle. |
470decc6 DK |
346 | * @journal: Journal to start transaction on. |
347 | * @nblocks: number of block buffer we might modify | |
348 | * | |
349 | * We make sure that the transaction can guarantee at least nblocks of | |
350 | * modified buffers in the log. We block until the log can guarantee | |
351 | * that much space. | |
352 | * | |
353 | * This function is visible to journal users (like ext3fs), so is not | |
354 | * called with the journal already locked. | |
355 | * | |
c867516d EG |
356 | * Return a pointer to a newly allocated handle, or an ERR_PTR() value |
357 | * on failure. | |
470decc6 | 358 | */ |
343d9c28 TT |
359 | handle_t *jbd2__journal_start(journal_t *journal, int nblocks, gfp_t gfp_mask, |
360 | unsigned int type, unsigned int line_no) | |
470decc6 DK |
361 | { |
362 | handle_t *handle = journal_current_handle(); | |
363 | int err; | |
364 | ||
365 | if (!journal) | |
366 | return ERR_PTR(-EROFS); | |
367 | ||
368 | if (handle) { | |
369 | J_ASSERT(handle->h_transaction->t_journal == journal); | |
370 | handle->h_ref++; | |
371 | return handle; | |
372 | } | |
373 | ||
374 | handle = new_handle(nblocks); | |
375 | if (!handle) | |
376 | return ERR_PTR(-ENOMEM); | |
377 | ||
378 | current->journal_info = handle; | |
379 | ||
47def826 | 380 | err = start_this_handle(journal, handle, gfp_mask); |
470decc6 | 381 | if (err < 0) { |
af1e76d6 | 382 | jbd2_free_handle(handle); |
470decc6 | 383 | current->journal_info = NULL; |
df05c1b8 | 384 | return ERR_PTR(err); |
470decc6 | 385 | } |
343d9c28 TT |
386 | handle->h_type = type; |
387 | handle->h_line_no = line_no; | |
388 | trace_jbd2_handle_start(journal->j_fs_dev->bd_dev, | |
389 | handle->h_transaction->t_tid, type, | |
390 | line_no, nblocks); | |
470decc6 DK |
391 | return handle; |
392 | } | |
47def826 TT |
393 | EXPORT_SYMBOL(jbd2__journal_start); |
394 | ||
395 | ||
396 | handle_t *jbd2_journal_start(journal_t *journal, int nblocks) | |
397 | { | |
343d9c28 | 398 | return jbd2__journal_start(journal, nblocks, GFP_NOFS, 0, 0); |
47def826 TT |
399 | } |
400 | EXPORT_SYMBOL(jbd2_journal_start); | |
401 | ||
470decc6 DK |
402 | |
403 | /** | |
f7f4bccb | 404 | * int jbd2_journal_extend() - extend buffer credits. |
470decc6 DK |
405 | * @handle: handle to 'extend' |
406 | * @nblocks: nr blocks to try to extend by. | |
407 | * | |
408 | * Some transactions, such as large extends and truncates, can be done | |
409 | * atomically all at once or in several stages. The operation requests | |
410 | * a credit for a number of buffer modications in advance, but can | |
411 | * extend its credit if it needs more. | |
412 | * | |
f7f4bccb | 413 | * jbd2_journal_extend tries to give the running handle more buffer credits. |
470decc6 DK |
414 | * It does not guarantee that allocation - this is a best-effort only. |
415 | * The calling process MUST be able to deal cleanly with a failure to | |
416 | * extend here. | |
417 | * | |
418 | * Return 0 on success, non-zero on failure. | |
419 | * | |
420 | * return code < 0 implies an error | |
421 | * return code > 0 implies normal transaction-full status. | |
422 | */ | |
f7f4bccb | 423 | int jbd2_journal_extend(handle_t *handle, int nblocks) |
470decc6 DK |
424 | { |
425 | transaction_t *transaction = handle->h_transaction; | |
426 | journal_t *journal = transaction->t_journal; | |
427 | int result; | |
428 | int wanted; | |
429 | ||
430 | result = -EIO; | |
431 | if (is_handle_aborted(handle)) | |
432 | goto out; | |
433 | ||
434 | result = 1; | |
435 | ||
a931da6a | 436 | read_lock(&journal->j_state_lock); |
470decc6 DK |
437 | |
438 | /* Don't extend a locked-down transaction! */ | |
439 | if (handle->h_transaction->t_state != T_RUNNING) { | |
440 | jbd_debug(3, "denied handle %p %d blocks: " | |
441 | "transaction not running\n", handle, nblocks); | |
442 | goto error_out; | |
443 | } | |
444 | ||
445 | spin_lock(&transaction->t_handle_lock); | |
a51dca9c | 446 | wanted = atomic_read(&transaction->t_outstanding_credits) + nblocks; |
470decc6 DK |
447 | |
448 | if (wanted > journal->j_max_transaction_buffers) { | |
449 | jbd_debug(3, "denied handle %p %d blocks: " | |
450 | "transaction too large\n", handle, nblocks); | |
451 | goto unlock; | |
452 | } | |
453 | ||
f7f4bccb | 454 | if (wanted > __jbd2_log_space_left(journal)) { |
470decc6 DK |
455 | jbd_debug(3, "denied handle %p %d blocks: " |
456 | "insufficient log space\n", handle, nblocks); | |
457 | goto unlock; | |
458 | } | |
459 | ||
343d9c28 TT |
460 | trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev, |
461 | handle->h_transaction->t_tid, | |
462 | handle->h_type, handle->h_line_no, | |
463 | handle->h_buffer_credits, | |
464 | nblocks); | |
465 | ||
470decc6 | 466 | handle->h_buffer_credits += nblocks; |
343d9c28 | 467 | handle->h_requested_credits += nblocks; |
a51dca9c | 468 | atomic_add(nblocks, &transaction->t_outstanding_credits); |
470decc6 DK |
469 | result = 0; |
470 | ||
471 | jbd_debug(3, "extended handle %p by %d\n", handle, nblocks); | |
472 | unlock: | |
473 | spin_unlock(&transaction->t_handle_lock); | |
474 | error_out: | |
a931da6a | 475 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
476 | out: |
477 | return result; | |
478 | } | |
479 | ||
480 | ||
481 | /** | |
f7f4bccb | 482 | * int jbd2_journal_restart() - restart a handle . |
470decc6 DK |
483 | * @handle: handle to restart |
484 | * @nblocks: nr credits requested | |
485 | * | |
486 | * Restart a handle for a multi-transaction filesystem | |
487 | * operation. | |
488 | * | |
f7f4bccb MC |
489 | * If the jbd2_journal_extend() call above fails to grant new buffer credits |
490 | * to a running handle, a call to jbd2_journal_restart will commit the | |
470decc6 DK |
491 | * handle's transaction so far and reattach the handle to a new |
492 | * transaction capabable of guaranteeing the requested number of | |
493 | * credits. | |
494 | */ | |
d2159fb7 | 495 | int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask) |
470decc6 DK |
496 | { |
497 | transaction_t *transaction = handle->h_transaction; | |
498 | journal_t *journal = transaction->t_journal; | |
e4471831 TT |
499 | tid_t tid; |
500 | int need_to_start, ret; | |
470decc6 DK |
501 | |
502 | /* If we've had an abort of any type, don't even think about | |
503 | * actually doing the restart! */ | |
504 | if (is_handle_aborted(handle)) | |
505 | return 0; | |
506 | ||
507 | /* | |
508 | * First unlink the handle from its current transaction, and start the | |
509 | * commit on that. | |
510 | */ | |
a51dca9c | 511 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 DK |
512 | J_ASSERT(journal_current_handle() == handle); |
513 | ||
a931da6a | 514 | read_lock(&journal->j_state_lock); |
470decc6 | 515 | spin_lock(&transaction->t_handle_lock); |
a51dca9c TT |
516 | atomic_sub(handle->h_buffer_credits, |
517 | &transaction->t_outstanding_credits); | |
518 | if (atomic_dec_and_test(&transaction->t_updates)) | |
470decc6 DK |
519 | wake_up(&journal->j_wait_updates); |
520 | spin_unlock(&transaction->t_handle_lock); | |
521 | ||
522 | jbd_debug(2, "restarting handle %p\n", handle); | |
e4471831 TT |
523 | tid = transaction->t_tid; |
524 | need_to_start = !tid_geq(journal->j_commit_request, tid); | |
a931da6a | 525 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
526 | if (need_to_start) |
527 | jbd2_log_start_commit(journal, tid); | |
470decc6 | 528 | |
9599b0e5 | 529 | lock_map_release(&handle->h_lockdep_map); |
470decc6 | 530 | handle->h_buffer_credits = nblocks; |
47def826 | 531 | ret = start_this_handle(journal, handle, gfp_mask); |
470decc6 DK |
532 | return ret; |
533 | } | |
47def826 | 534 | EXPORT_SYMBOL(jbd2__journal_restart); |
470decc6 DK |
535 | |
536 | ||
47def826 TT |
537 | int jbd2_journal_restart(handle_t *handle, int nblocks) |
538 | { | |
539 | return jbd2__journal_restart(handle, nblocks, GFP_NOFS); | |
540 | } | |
541 | EXPORT_SYMBOL(jbd2_journal_restart); | |
542 | ||
470decc6 | 543 | /** |
f7f4bccb | 544 | * void jbd2_journal_lock_updates () - establish a transaction barrier. |
470decc6 DK |
545 | * @journal: Journal to establish a barrier on. |
546 | * | |
547 | * This locks out any further updates from being started, and blocks | |
548 | * until all existing updates have completed, returning only once the | |
549 | * journal is in a quiescent state with no updates running. | |
550 | * | |
551 | * The journal lock should not be held on entry. | |
552 | */ | |
f7f4bccb | 553 | void jbd2_journal_lock_updates(journal_t *journal) |
470decc6 DK |
554 | { |
555 | DEFINE_WAIT(wait); | |
556 | ||
a931da6a | 557 | write_lock(&journal->j_state_lock); |
470decc6 DK |
558 | ++journal->j_barrier_count; |
559 | ||
560 | /* Wait until there are no running updates */ | |
561 | while (1) { | |
562 | transaction_t *transaction = journal->j_running_transaction; | |
563 | ||
564 | if (!transaction) | |
565 | break; | |
566 | ||
567 | spin_lock(&transaction->t_handle_lock); | |
9837d8e9 JK |
568 | prepare_to_wait(&journal->j_wait_updates, &wait, |
569 | TASK_UNINTERRUPTIBLE); | |
a51dca9c | 570 | if (!atomic_read(&transaction->t_updates)) { |
470decc6 | 571 | spin_unlock(&transaction->t_handle_lock); |
9837d8e9 | 572 | finish_wait(&journal->j_wait_updates, &wait); |
470decc6 DK |
573 | break; |
574 | } | |
470decc6 | 575 | spin_unlock(&transaction->t_handle_lock); |
a931da6a | 576 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
577 | schedule(); |
578 | finish_wait(&journal->j_wait_updates, &wait); | |
a931da6a | 579 | write_lock(&journal->j_state_lock); |
470decc6 | 580 | } |
a931da6a | 581 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
582 | |
583 | /* | |
584 | * We have now established a barrier against other normal updates, but | |
f7f4bccb | 585 | * we also need to barrier against other jbd2_journal_lock_updates() calls |
470decc6 DK |
586 | * to make sure that we serialise special journal-locked operations |
587 | * too. | |
588 | */ | |
589 | mutex_lock(&journal->j_barrier); | |
590 | } | |
591 | ||
592 | /** | |
f7f4bccb | 593 | * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier |
470decc6 DK |
594 | * @journal: Journal to release the barrier on. |
595 | * | |
f7f4bccb | 596 | * Release a transaction barrier obtained with jbd2_journal_lock_updates(). |
470decc6 DK |
597 | * |
598 | * Should be called without the journal lock held. | |
599 | */ | |
f7f4bccb | 600 | void jbd2_journal_unlock_updates (journal_t *journal) |
470decc6 DK |
601 | { |
602 | J_ASSERT(journal->j_barrier_count != 0); | |
603 | ||
604 | mutex_unlock(&journal->j_barrier); | |
a931da6a | 605 | write_lock(&journal->j_state_lock); |
470decc6 | 606 | --journal->j_barrier_count; |
a931da6a | 607 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
608 | wake_up(&journal->j_wait_transaction_locked); |
609 | } | |
610 | ||
f91d1d04 | 611 | static void warn_dirty_buffer(struct buffer_head *bh) |
470decc6 | 612 | { |
f91d1d04 | 613 | char b[BDEVNAME_SIZE]; |
470decc6 | 614 | |
f91d1d04 | 615 | printk(KERN_WARNING |
f2a44523 | 616 | "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). " |
f91d1d04 JK |
617 | "There's a risk of filesystem corruption in case of system " |
618 | "crash.\n", | |
619 | bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr); | |
470decc6 DK |
620 | } |
621 | ||
622 | /* | |
623 | * If the buffer is already part of the current transaction, then there | |
624 | * is nothing we need to do. If it is already part of a prior | |
625 | * transaction which we are still committing to disk, then we need to | |
626 | * make sure that we do not overwrite the old copy: we do copy-out to | |
627 | * preserve the copy going to disk. We also account the buffer against | |
628 | * the handle's metadata buffer credits (unless the buffer is already | |
629 | * part of the transaction, that is). | |
630 | * | |
631 | */ | |
632 | static int | |
633 | do_get_write_access(handle_t *handle, struct journal_head *jh, | |
634 | int force_copy) | |
635 | { | |
636 | struct buffer_head *bh; | |
637 | transaction_t *transaction; | |
638 | journal_t *journal; | |
639 | int error; | |
640 | char *frozen_buffer = NULL; | |
641 | int need_copy = 0; | |
f783f091 | 642 | unsigned long start_lock, time_lock; |
470decc6 DK |
643 | |
644 | if (is_handle_aborted(handle)) | |
645 | return -EROFS; | |
646 | ||
647 | transaction = handle->h_transaction; | |
648 | journal = transaction->t_journal; | |
649 | ||
cfef2c6a | 650 | jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy); |
470decc6 DK |
651 | |
652 | JBUFFER_TRACE(jh, "entry"); | |
653 | repeat: | |
654 | bh = jh2bh(jh); | |
655 | ||
656 | /* @@@ Need to check for errors here at some point. */ | |
657 | ||
f783f091 | 658 | start_lock = jiffies; |
470decc6 DK |
659 | lock_buffer(bh); |
660 | jbd_lock_bh_state(bh); | |
661 | ||
f783f091 TT |
662 | /* If it takes too long to lock the buffer, trace it */ |
663 | time_lock = jbd2_time_diff(start_lock, jiffies); | |
664 | if (time_lock > HZ/10) | |
665 | trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev, | |
666 | jiffies_to_msecs(time_lock)); | |
667 | ||
470decc6 DK |
668 | /* We now hold the buffer lock so it is safe to query the buffer |
669 | * state. Is the buffer dirty? | |
670 | * | |
671 | * If so, there are two possibilities. The buffer may be | |
672 | * non-journaled, and undergoing a quite legitimate writeback. | |
673 | * Otherwise, it is journaled, and we don't expect dirty buffers | |
674 | * in that state (the buffers should be marked JBD_Dirty | |
675 | * instead.) So either the IO is being done under our own | |
676 | * control and this is a bug, or it's a third party IO such as | |
677 | * dump(8) (which may leave the buffer scheduled for read --- | |
678 | * ie. locked but not dirty) or tune2fs (which may actually have | |
679 | * the buffer dirtied, ugh.) */ | |
680 | ||
681 | if (buffer_dirty(bh)) { | |
682 | /* | |
683 | * First question: is this buffer already part of the current | |
684 | * transaction or the existing committing transaction? | |
685 | */ | |
686 | if (jh->b_transaction) { | |
687 | J_ASSERT_JH(jh, | |
688 | jh->b_transaction == transaction || | |
689 | jh->b_transaction == | |
690 | journal->j_committing_transaction); | |
691 | if (jh->b_next_transaction) | |
692 | J_ASSERT_JH(jh, jh->b_next_transaction == | |
693 | transaction); | |
f91d1d04 | 694 | warn_dirty_buffer(bh); |
470decc6 DK |
695 | } |
696 | /* | |
697 | * In any case we need to clean the dirty flag and we must | |
698 | * do it under the buffer lock to be sure we don't race | |
699 | * with running write-out. | |
700 | */ | |
f91d1d04 JK |
701 | JBUFFER_TRACE(jh, "Journalling dirty buffer"); |
702 | clear_buffer_dirty(bh); | |
703 | set_buffer_jbddirty(bh); | |
470decc6 DK |
704 | } |
705 | ||
706 | unlock_buffer(bh); | |
707 | ||
708 | error = -EROFS; | |
709 | if (is_handle_aborted(handle)) { | |
710 | jbd_unlock_bh_state(bh); | |
711 | goto out; | |
712 | } | |
713 | error = 0; | |
714 | ||
715 | /* | |
716 | * The buffer is already part of this transaction if b_transaction or | |
717 | * b_next_transaction points to it | |
718 | */ | |
719 | if (jh->b_transaction == transaction || | |
720 | jh->b_next_transaction == transaction) | |
721 | goto done; | |
722 | ||
9fc7c63a JB |
723 | /* |
724 | * this is the first time this transaction is touching this buffer, | |
725 | * reset the modified flag | |
726 | */ | |
727 | jh->b_modified = 0; | |
728 | ||
470decc6 DK |
729 | /* |
730 | * If there is already a copy-out version of this buffer, then we don't | |
731 | * need to make another one | |
732 | */ | |
733 | if (jh->b_frozen_data) { | |
734 | JBUFFER_TRACE(jh, "has frozen data"); | |
735 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
736 | jh->b_next_transaction = transaction; | |
737 | goto done; | |
738 | } | |
739 | ||
740 | /* Is there data here we need to preserve? */ | |
741 | ||
742 | if (jh->b_transaction && jh->b_transaction != transaction) { | |
743 | JBUFFER_TRACE(jh, "owned by older transaction"); | |
744 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
745 | J_ASSERT_JH(jh, jh->b_transaction == | |
746 | journal->j_committing_transaction); | |
747 | ||
748 | /* There is one case we have to be very careful about. | |
749 | * If the committing transaction is currently writing | |
750 | * this buffer out to disk and has NOT made a copy-out, | |
751 | * then we cannot modify the buffer contents at all | |
752 | * right now. The essence of copy-out is that it is the | |
753 | * extra copy, not the primary copy, which gets | |
754 | * journaled. If the primary copy is already going to | |
755 | * disk then we cannot do copy-out here. */ | |
756 | ||
757 | if (jh->b_jlist == BJ_Shadow) { | |
758 | DEFINE_WAIT_BIT(wait, &bh->b_state, BH_Unshadow); | |
759 | wait_queue_head_t *wqh; | |
760 | ||
761 | wqh = bit_waitqueue(&bh->b_state, BH_Unshadow); | |
762 | ||
763 | JBUFFER_TRACE(jh, "on shadow: sleep"); | |
764 | jbd_unlock_bh_state(bh); | |
765 | /* commit wakes up all shadow buffers after IO */ | |
766 | for ( ; ; ) { | |
767 | prepare_to_wait(wqh, &wait.wait, | |
768 | TASK_UNINTERRUPTIBLE); | |
769 | if (jh->b_jlist != BJ_Shadow) | |
770 | break; | |
771 | schedule(); | |
772 | } | |
773 | finish_wait(wqh, &wait.wait); | |
774 | goto repeat; | |
775 | } | |
776 | ||
777 | /* Only do the copy if the currently-owning transaction | |
778 | * still needs it. If it is on the Forget list, the | |
779 | * committing transaction is past that stage. The | |
780 | * buffer had better remain locked during the kmalloc, | |
781 | * but that should be true --- we hold the journal lock | |
782 | * still and the buffer is already on the BUF_JOURNAL | |
783 | * list so won't be flushed. | |
784 | * | |
785 | * Subtle point, though: if this is a get_undo_access, | |
786 | * then we will be relying on the frozen_data to contain | |
787 | * the new value of the committed_data record after the | |
788 | * transaction, so we HAVE to force the frozen_data copy | |
789 | * in that case. */ | |
790 | ||
791 | if (jh->b_jlist != BJ_Forget || force_copy) { | |
792 | JBUFFER_TRACE(jh, "generate frozen data"); | |
793 | if (!frozen_buffer) { | |
794 | JBUFFER_TRACE(jh, "allocate memory for buffer"); | |
795 | jbd_unlock_bh_state(bh); | |
796 | frozen_buffer = | |
af1e76d6 | 797 | jbd2_alloc(jh2bh(jh)->b_size, |
470decc6 DK |
798 | GFP_NOFS); |
799 | if (!frozen_buffer) { | |
800 | printk(KERN_EMERG | |
801 | "%s: OOM for frozen_buffer\n", | |
329d291f | 802 | __func__); |
470decc6 DK |
803 | JBUFFER_TRACE(jh, "oom!"); |
804 | error = -ENOMEM; | |
805 | jbd_lock_bh_state(bh); | |
806 | goto done; | |
807 | } | |
808 | goto repeat; | |
809 | } | |
810 | jh->b_frozen_data = frozen_buffer; | |
811 | frozen_buffer = NULL; | |
812 | need_copy = 1; | |
813 | } | |
814 | jh->b_next_transaction = transaction; | |
815 | } | |
816 | ||
817 | ||
818 | /* | |
819 | * Finally, if the buffer is not journaled right now, we need to make | |
820 | * sure it doesn't get written to disk before the caller actually | |
821 | * commits the new data | |
822 | */ | |
823 | if (!jh->b_transaction) { | |
824 | JBUFFER_TRACE(jh, "no transaction"); | |
825 | J_ASSERT_JH(jh, !jh->b_next_transaction); | |
470decc6 DK |
826 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); |
827 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 828 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
470decc6 DK |
829 | spin_unlock(&journal->j_list_lock); |
830 | } | |
831 | ||
832 | done: | |
833 | if (need_copy) { | |
834 | struct page *page; | |
835 | int offset; | |
836 | char *source; | |
837 | ||
838 | J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)), | |
839 | "Possible IO failure.\n"); | |
840 | page = jh2bh(jh)->b_page; | |
a1dd5331 | 841 | offset = offset_in_page(jh2bh(jh)->b_data); |
303a8f2a | 842 | source = kmap_atomic(page); |
13ceef09 JK |
843 | /* Fire data frozen trigger just before we copy the data */ |
844 | jbd2_buffer_frozen_trigger(jh, source + offset, | |
845 | jh->b_triggers); | |
470decc6 | 846 | memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size); |
303a8f2a | 847 | kunmap_atomic(source); |
e06c8227 JB |
848 | |
849 | /* | |
850 | * Now that the frozen data is saved off, we need to store | |
851 | * any matching triggers. | |
852 | */ | |
853 | jh->b_frozen_triggers = jh->b_triggers; | |
470decc6 DK |
854 | } |
855 | jbd_unlock_bh_state(bh); | |
856 | ||
857 | /* | |
858 | * If we are about to journal a buffer, then any revoke pending on it is | |
859 | * no longer valid | |
860 | */ | |
f7f4bccb | 861 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 DK |
862 | |
863 | out: | |
864 | if (unlikely(frozen_buffer)) /* It's usually NULL */ | |
af1e76d6 | 865 | jbd2_free(frozen_buffer, bh->b_size); |
470decc6 DK |
866 | |
867 | JBUFFER_TRACE(jh, "exit"); | |
868 | return error; | |
869 | } | |
870 | ||
871 | /** | |
f7f4bccb | 872 | * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update. |
470decc6 DK |
873 | * @handle: transaction to add buffer modifications to |
874 | * @bh: bh to be used for metadata writes | |
470decc6 DK |
875 | * |
876 | * Returns an error code or 0 on success. | |
877 | * | |
878 | * In full data journalling mode the buffer may be of type BJ_AsyncData, | |
879 | * because we're write()ing a buffer which is also part of a shared mapping. | |
880 | */ | |
881 | ||
f7f4bccb | 882 | int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh) |
470decc6 | 883 | { |
f7f4bccb | 884 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
885 | int rc; |
886 | ||
887 | /* We do not want to get caught playing with fields which the | |
888 | * log thread also manipulates. Make sure that the buffer | |
889 | * completes any outstanding IO before proceeding. */ | |
890 | rc = do_get_write_access(handle, jh, 0); | |
f7f4bccb | 891 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
892 | return rc; |
893 | } | |
894 | ||
895 | ||
896 | /* | |
897 | * When the user wants to journal a newly created buffer_head | |
898 | * (ie. getblk() returned a new buffer and we are going to populate it | |
899 | * manually rather than reading off disk), then we need to keep the | |
900 | * buffer_head locked until it has been completely filled with new | |
901 | * data. In this case, we should be able to make the assertion that | |
902 | * the bh is not already part of an existing transaction. | |
903 | * | |
904 | * The buffer should already be locked by the caller by this point. | |
905 | * There is no lock ranking violation: it was a newly created, | |
906 | * unlocked buffer beforehand. */ | |
907 | ||
908 | /** | |
f7f4bccb | 909 | * int jbd2_journal_get_create_access () - notify intent to use newly created bh |
470decc6 DK |
910 | * @handle: transaction to new buffer to |
911 | * @bh: new buffer. | |
912 | * | |
913 | * Call this if you create a new bh. | |
914 | */ | |
f7f4bccb | 915 | int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
916 | { |
917 | transaction_t *transaction = handle->h_transaction; | |
918 | journal_t *journal = transaction->t_journal; | |
f7f4bccb | 919 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
920 | int err; |
921 | ||
922 | jbd_debug(5, "journal_head %p\n", jh); | |
923 | err = -EROFS; | |
924 | if (is_handle_aborted(handle)) | |
925 | goto out; | |
926 | err = 0; | |
927 | ||
928 | JBUFFER_TRACE(jh, "entry"); | |
929 | /* | |
930 | * The buffer may already belong to this transaction due to pre-zeroing | |
931 | * in the filesystem's new_block code. It may also be on the previous, | |
932 | * committing transaction's lists, but it HAS to be in Forget state in | |
933 | * that case: the transaction must have deleted the buffer for it to be | |
934 | * reused here. | |
935 | */ | |
936 | jbd_lock_bh_state(bh); | |
937 | spin_lock(&journal->j_list_lock); | |
938 | J_ASSERT_JH(jh, (jh->b_transaction == transaction || | |
939 | jh->b_transaction == NULL || | |
940 | (jh->b_transaction == journal->j_committing_transaction && | |
941 | jh->b_jlist == BJ_Forget))); | |
942 | ||
943 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
944 | J_ASSERT_JH(jh, buffer_locked(jh2bh(jh))); | |
945 | ||
946 | if (jh->b_transaction == NULL) { | |
f91d1d04 JK |
947 | /* |
948 | * Previous jbd2_journal_forget() could have left the buffer | |
949 | * with jbddirty bit set because it was being committed. When | |
950 | * the commit finished, we've filed the buffer for | |
951 | * checkpointing and marked it dirty. Now we are reallocating | |
952 | * the buffer so the transaction freeing it must have | |
953 | * committed and so it's safe to clear the dirty bit. | |
954 | */ | |
955 | clear_buffer_dirty(jh2bh(jh)); | |
9fc7c63a JB |
956 | /* first access by this transaction */ |
957 | jh->b_modified = 0; | |
958 | ||
470decc6 | 959 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); |
f7f4bccb | 960 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
470decc6 | 961 | } else if (jh->b_transaction == journal->j_committing_transaction) { |
9fc7c63a JB |
962 | /* first access by this transaction */ |
963 | jh->b_modified = 0; | |
964 | ||
470decc6 DK |
965 | JBUFFER_TRACE(jh, "set next transaction"); |
966 | jh->b_next_transaction = transaction; | |
967 | } | |
968 | spin_unlock(&journal->j_list_lock); | |
969 | jbd_unlock_bh_state(bh); | |
970 | ||
971 | /* | |
972 | * akpm: I added this. ext3_alloc_branch can pick up new indirect | |
973 | * blocks which contain freed but then revoked metadata. We need | |
974 | * to cancel the revoke in case we end up freeing it yet again | |
975 | * and the reallocating as data - this would cause a second revoke, | |
976 | * which hits an assertion error. | |
977 | */ | |
978 | JBUFFER_TRACE(jh, "cancelling revoke"); | |
f7f4bccb | 979 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 | 980 | out: |
3991b400 | 981 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
982 | return err; |
983 | } | |
984 | ||
985 | /** | |
f7f4bccb | 986 | * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with |
470decc6 DK |
987 | * non-rewindable consequences |
988 | * @handle: transaction | |
989 | * @bh: buffer to undo | |
470decc6 DK |
990 | * |
991 | * Sometimes there is a need to distinguish between metadata which has | |
992 | * been committed to disk and that which has not. The ext3fs code uses | |
993 | * this for freeing and allocating space, we have to make sure that we | |
994 | * do not reuse freed space until the deallocation has been committed, | |
995 | * since if we overwrote that space we would make the delete | |
996 | * un-rewindable in case of a crash. | |
997 | * | |
f7f4bccb | 998 | * To deal with that, jbd2_journal_get_undo_access requests write access to a |
470decc6 DK |
999 | * buffer for parts of non-rewindable operations such as delete |
1000 | * operations on the bitmaps. The journaling code must keep a copy of | |
1001 | * the buffer's contents prior to the undo_access call until such time | |
1002 | * as we know that the buffer has definitely been committed to disk. | |
1003 | * | |
1004 | * We never need to know which transaction the committed data is part | |
1005 | * of, buffers touched here are guaranteed to be dirtied later and so | |
1006 | * will be committed to a new transaction in due course, at which point | |
1007 | * we can discard the old committed data pointer. | |
1008 | * | |
1009 | * Returns error number or 0 on success. | |
1010 | */ | |
f7f4bccb | 1011 | int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1012 | { |
1013 | int err; | |
f7f4bccb | 1014 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
1015 | char *committed_data = NULL; |
1016 | ||
1017 | JBUFFER_TRACE(jh, "entry"); | |
1018 | ||
1019 | /* | |
1020 | * Do this first --- it can drop the journal lock, so we want to | |
1021 | * make sure that obtaining the committed_data is done | |
1022 | * atomically wrt. completion of any outstanding commits. | |
1023 | */ | |
1024 | err = do_get_write_access(handle, jh, 1); | |
1025 | if (err) | |
1026 | goto out; | |
1027 | ||
1028 | repeat: | |
1029 | if (!jh->b_committed_data) { | |
af1e76d6 | 1030 | committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS); |
470decc6 DK |
1031 | if (!committed_data) { |
1032 | printk(KERN_EMERG "%s: No memory for committed data\n", | |
329d291f | 1033 | __func__); |
470decc6 DK |
1034 | err = -ENOMEM; |
1035 | goto out; | |
1036 | } | |
1037 | } | |
1038 | ||
1039 | jbd_lock_bh_state(bh); | |
1040 | if (!jh->b_committed_data) { | |
1041 | /* Copy out the current buffer contents into the | |
1042 | * preserved, committed copy. */ | |
1043 | JBUFFER_TRACE(jh, "generate b_committed data"); | |
1044 | if (!committed_data) { | |
1045 | jbd_unlock_bh_state(bh); | |
1046 | goto repeat; | |
1047 | } | |
1048 | ||
1049 | jh->b_committed_data = committed_data; | |
1050 | committed_data = NULL; | |
1051 | memcpy(jh->b_committed_data, bh->b_data, bh->b_size); | |
1052 | } | |
1053 | jbd_unlock_bh_state(bh); | |
1054 | out: | |
f7f4bccb | 1055 | jbd2_journal_put_journal_head(jh); |
470decc6 | 1056 | if (unlikely(committed_data)) |
af1e76d6 | 1057 | jbd2_free(committed_data, bh->b_size); |
470decc6 DK |
1058 | return err; |
1059 | } | |
1060 | ||
e06c8227 JB |
1061 | /** |
1062 | * void jbd2_journal_set_triggers() - Add triggers for commit writeout | |
1063 | * @bh: buffer to trigger on | |
1064 | * @type: struct jbd2_buffer_trigger_type containing the trigger(s). | |
1065 | * | |
1066 | * Set any triggers on this journal_head. This is always safe, because | |
1067 | * triggers for a committing buffer will be saved off, and triggers for | |
1068 | * a running transaction will match the buffer in that transaction. | |
1069 | * | |
1070 | * Call with NULL to clear the triggers. | |
1071 | */ | |
1072 | void jbd2_journal_set_triggers(struct buffer_head *bh, | |
1073 | struct jbd2_buffer_trigger_type *type) | |
1074 | { | |
ad56edad | 1075 | struct journal_head *jh = jbd2_journal_grab_journal_head(bh); |
e06c8227 | 1076 | |
ad56edad JK |
1077 | if (WARN_ON(!jh)) |
1078 | return; | |
e06c8227 | 1079 | jh->b_triggers = type; |
ad56edad | 1080 | jbd2_journal_put_journal_head(jh); |
e06c8227 JB |
1081 | } |
1082 | ||
13ceef09 | 1083 | void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data, |
e06c8227 JB |
1084 | struct jbd2_buffer_trigger_type *triggers) |
1085 | { | |
1086 | struct buffer_head *bh = jh2bh(jh); | |
1087 | ||
13ceef09 | 1088 | if (!triggers || !triggers->t_frozen) |
e06c8227 JB |
1089 | return; |
1090 | ||
13ceef09 | 1091 | triggers->t_frozen(triggers, bh, mapped_data, bh->b_size); |
e06c8227 JB |
1092 | } |
1093 | ||
1094 | void jbd2_buffer_abort_trigger(struct journal_head *jh, | |
1095 | struct jbd2_buffer_trigger_type *triggers) | |
1096 | { | |
1097 | if (!triggers || !triggers->t_abort) | |
1098 | return; | |
1099 | ||
1100 | triggers->t_abort(triggers, jh2bh(jh)); | |
1101 | } | |
1102 | ||
1103 | ||
1104 | ||
470decc6 | 1105 | /** |
f7f4bccb | 1106 | * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata |
470decc6 DK |
1107 | * @handle: transaction to add buffer to. |
1108 | * @bh: buffer to mark | |
1109 | * | |
1110 | * mark dirty metadata which needs to be journaled as part of the current | |
1111 | * transaction. | |
1112 | * | |
9ea7a0df TT |
1113 | * The buffer must have previously had jbd2_journal_get_write_access() |
1114 | * called so that it has a valid journal_head attached to the buffer | |
1115 | * head. | |
1116 | * | |
470decc6 DK |
1117 | * The buffer is placed on the transaction's metadata list and is marked |
1118 | * as belonging to the transaction. | |
1119 | * | |
1120 | * Returns error number or 0 on success. | |
1121 | * | |
1122 | * Special care needs to be taken if the buffer already belongs to the | |
1123 | * current committing transaction (in which case we should have frozen | |
1124 | * data present for that commit). In that case, we don't relink the | |
1125 | * buffer: that only gets done when the old transaction finally | |
1126 | * completes its commit. | |
1127 | */ | |
f7f4bccb | 1128 | int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1129 | { |
1130 | transaction_t *transaction = handle->h_transaction; | |
1131 | journal_t *journal = transaction->t_journal; | |
ad56edad | 1132 | struct journal_head *jh; |
9ea7a0df | 1133 | int ret = 0; |
470decc6 | 1134 | |
470decc6 DK |
1135 | if (is_handle_aborted(handle)) |
1136 | goto out; | |
ad56edad JK |
1137 | jh = jbd2_journal_grab_journal_head(bh); |
1138 | if (!jh) { | |
9ea7a0df TT |
1139 | ret = -EUCLEAN; |
1140 | goto out; | |
1141 | } | |
ad56edad JK |
1142 | jbd_debug(5, "journal_head %p\n", jh); |
1143 | JBUFFER_TRACE(jh, "entry"); | |
470decc6 DK |
1144 | |
1145 | jbd_lock_bh_state(bh); | |
1146 | ||
1147 | if (jh->b_modified == 0) { | |
1148 | /* | |
1149 | * This buffer's got modified and becoming part | |
1150 | * of the transaction. This needs to be done | |
1151 | * once a transaction -bzzz | |
1152 | */ | |
1153 | jh->b_modified = 1; | |
1154 | J_ASSERT_JH(jh, handle->h_buffer_credits > 0); | |
1155 | handle->h_buffer_credits--; | |
1156 | } | |
1157 | ||
1158 | /* | |
1159 | * fastpath, to avoid expensive locking. If this buffer is already | |
1160 | * on the running transaction's metadata list there is nothing to do. | |
1161 | * Nobody can take it off again because there is a handle open. | |
1162 | * I _think_ we're OK here with SMP barriers - a mistaken decision will | |
1163 | * result in this test being false, so we go in and take the locks. | |
1164 | */ | |
1165 | if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) { | |
1166 | JBUFFER_TRACE(jh, "fastpath"); | |
9ea7a0df TT |
1167 | if (unlikely(jh->b_transaction != |
1168 | journal->j_running_transaction)) { | |
1169 | printk(KERN_EMERG "JBD: %s: " | |
1170 | "jh->b_transaction (%llu, %p, %u) != " | |
1171 | "journal->j_running_transaction (%p, %u)", | |
1172 | journal->j_devname, | |
1173 | (unsigned long long) bh->b_blocknr, | |
1174 | jh->b_transaction, | |
1175 | jh->b_transaction ? jh->b_transaction->t_tid : 0, | |
1176 | journal->j_running_transaction, | |
1177 | journal->j_running_transaction ? | |
1178 | journal->j_running_transaction->t_tid : 0); | |
1179 | ret = -EINVAL; | |
1180 | } | |
470decc6 DK |
1181 | goto out_unlock_bh; |
1182 | } | |
1183 | ||
1184 | set_buffer_jbddirty(bh); | |
1185 | ||
1186 | /* | |
1187 | * Metadata already on the current transaction list doesn't | |
1188 | * need to be filed. Metadata on another transaction's list must | |
1189 | * be committing, and will be refiled once the commit completes: | |
1190 | * leave it alone for now. | |
1191 | */ | |
1192 | if (jh->b_transaction != transaction) { | |
1193 | JBUFFER_TRACE(jh, "already on other transaction"); | |
9ea7a0df TT |
1194 | if (unlikely(jh->b_transaction != |
1195 | journal->j_committing_transaction)) { | |
1196 | printk(KERN_EMERG "JBD: %s: " | |
1197 | "jh->b_transaction (%llu, %p, %u) != " | |
1198 | "journal->j_committing_transaction (%p, %u)", | |
1199 | journal->j_devname, | |
1200 | (unsigned long long) bh->b_blocknr, | |
1201 | jh->b_transaction, | |
1202 | jh->b_transaction ? jh->b_transaction->t_tid : 0, | |
1203 | journal->j_committing_transaction, | |
1204 | journal->j_committing_transaction ? | |
1205 | journal->j_committing_transaction->t_tid : 0); | |
1206 | ret = -EINVAL; | |
1207 | } | |
1208 | if (unlikely(jh->b_next_transaction != transaction)) { | |
1209 | printk(KERN_EMERG "JBD: %s: " | |
1210 | "jh->b_next_transaction (%llu, %p, %u) != " | |
1211 | "transaction (%p, %u)", | |
1212 | journal->j_devname, | |
1213 | (unsigned long long) bh->b_blocknr, | |
1214 | jh->b_next_transaction, | |
1215 | jh->b_next_transaction ? | |
1216 | jh->b_next_transaction->t_tid : 0, | |
1217 | transaction, transaction->t_tid); | |
1218 | ret = -EINVAL; | |
1219 | } | |
470decc6 DK |
1220 | /* And this case is illegal: we can't reuse another |
1221 | * transaction's data buffer, ever. */ | |
1222 | goto out_unlock_bh; | |
1223 | } | |
1224 | ||
1225 | /* That test should have eliminated the following case: */ | |
4019191b | 1226 | J_ASSERT_JH(jh, jh->b_frozen_data == NULL); |
470decc6 DK |
1227 | |
1228 | JBUFFER_TRACE(jh, "file as BJ_Metadata"); | |
1229 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 1230 | __jbd2_journal_file_buffer(jh, handle->h_transaction, BJ_Metadata); |
470decc6 DK |
1231 | spin_unlock(&journal->j_list_lock); |
1232 | out_unlock_bh: | |
1233 | jbd_unlock_bh_state(bh); | |
ad56edad | 1234 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1235 | out: |
1236 | JBUFFER_TRACE(jh, "exit"); | |
44705754 | 1237 | WARN_ON(ret); /* All errors are bugs, so dump the stack */ |
9ea7a0df | 1238 | return ret; |
470decc6 DK |
1239 | } |
1240 | ||
470decc6 | 1241 | /** |
f7f4bccb | 1242 | * void jbd2_journal_forget() - bforget() for potentially-journaled buffers. |
470decc6 DK |
1243 | * @handle: transaction handle |
1244 | * @bh: bh to 'forget' | |
1245 | * | |
1246 | * We can only do the bforget if there are no commits pending against the | |
1247 | * buffer. If the buffer is dirty in the current running transaction we | |
1248 | * can safely unlink it. | |
1249 | * | |
1250 | * bh may not be a journalled buffer at all - it may be a non-JBD | |
1251 | * buffer which came off the hashtable. Check for this. | |
1252 | * | |
1253 | * Decrements bh->b_count by one. | |
1254 | * | |
1255 | * Allow this call even if the handle has aborted --- it may be part of | |
1256 | * the caller's cleanup after an abort. | |
1257 | */ | |
f7f4bccb | 1258 | int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1259 | { |
1260 | transaction_t *transaction = handle->h_transaction; | |
1261 | journal_t *journal = transaction->t_journal; | |
1262 | struct journal_head *jh; | |
1263 | int drop_reserve = 0; | |
1264 | int err = 0; | |
1dfc3220 | 1265 | int was_modified = 0; |
470decc6 DK |
1266 | |
1267 | BUFFER_TRACE(bh, "entry"); | |
1268 | ||
1269 | jbd_lock_bh_state(bh); | |
1270 | spin_lock(&journal->j_list_lock); | |
1271 | ||
1272 | if (!buffer_jbd(bh)) | |
1273 | goto not_jbd; | |
1274 | jh = bh2jh(bh); | |
1275 | ||
1276 | /* Critical error: attempting to delete a bitmap buffer, maybe? | |
1277 | * Don't do any jbd operations, and return an error. */ | |
1278 | if (!J_EXPECT_JH(jh, !jh->b_committed_data, | |
1279 | "inconsistent data on disk")) { | |
1280 | err = -EIO; | |
1281 | goto not_jbd; | |
1282 | } | |
1283 | ||
48fc7f7e | 1284 | /* keep track of whether or not this transaction modified us */ |
1dfc3220 JB |
1285 | was_modified = jh->b_modified; |
1286 | ||
470decc6 DK |
1287 | /* |
1288 | * The buffer's going from the transaction, we must drop | |
1289 | * all references -bzzz | |
1290 | */ | |
1291 | jh->b_modified = 0; | |
1292 | ||
1293 | if (jh->b_transaction == handle->h_transaction) { | |
1294 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
1295 | ||
1296 | /* If we are forgetting a buffer which is already part | |
1297 | * of this transaction, then we can just drop it from | |
1298 | * the transaction immediately. */ | |
1299 | clear_buffer_dirty(bh); | |
1300 | clear_buffer_jbddirty(bh); | |
1301 | ||
1302 | JBUFFER_TRACE(jh, "belongs to current transaction: unfile"); | |
1303 | ||
1dfc3220 JB |
1304 | /* |
1305 | * we only want to drop a reference if this transaction | |
1306 | * modified the buffer | |
1307 | */ | |
1308 | if (was_modified) | |
1309 | drop_reserve = 1; | |
470decc6 DK |
1310 | |
1311 | /* | |
1312 | * We are no longer going to journal this buffer. | |
1313 | * However, the commit of this transaction is still | |
1314 | * important to the buffer: the delete that we are now | |
1315 | * processing might obsolete an old log entry, so by | |
1316 | * committing, we can satisfy the buffer's checkpoint. | |
1317 | * | |
1318 | * So, if we have a checkpoint on the buffer, we should | |
1319 | * now refile the buffer on our BJ_Forget list so that | |
1320 | * we know to remove the checkpoint after we commit. | |
1321 | */ | |
1322 | ||
1323 | if (jh->b_cp_transaction) { | |
f7f4bccb MC |
1324 | __jbd2_journal_temp_unlink_buffer(jh); |
1325 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); | |
470decc6 | 1326 | } else { |
f7f4bccb | 1327 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
1328 | if (!buffer_jbd(bh)) { |
1329 | spin_unlock(&journal->j_list_lock); | |
1330 | jbd_unlock_bh_state(bh); | |
1331 | __bforget(bh); | |
1332 | goto drop; | |
1333 | } | |
1334 | } | |
1335 | } else if (jh->b_transaction) { | |
1336 | J_ASSERT_JH(jh, (jh->b_transaction == | |
1337 | journal->j_committing_transaction)); | |
1338 | /* However, if the buffer is still owned by a prior | |
1339 | * (committing) transaction, we can't drop it yet... */ | |
1340 | JBUFFER_TRACE(jh, "belongs to older transaction"); | |
1341 | /* ... but we CAN drop it from the new transaction if we | |
1342 | * have also modified it since the original commit. */ | |
1343 | ||
1344 | if (jh->b_next_transaction) { | |
1345 | J_ASSERT(jh->b_next_transaction == transaction); | |
1346 | jh->b_next_transaction = NULL; | |
1dfc3220 JB |
1347 | |
1348 | /* | |
1349 | * only drop a reference if this transaction modified | |
1350 | * the buffer | |
1351 | */ | |
1352 | if (was_modified) | |
1353 | drop_reserve = 1; | |
470decc6 DK |
1354 | } |
1355 | } | |
1356 | ||
1357 | not_jbd: | |
1358 | spin_unlock(&journal->j_list_lock); | |
1359 | jbd_unlock_bh_state(bh); | |
1360 | __brelse(bh); | |
1361 | drop: | |
1362 | if (drop_reserve) { | |
1363 | /* no need to reserve log space for this block -bzzz */ | |
1364 | handle->h_buffer_credits++; | |
1365 | } | |
1366 | return err; | |
1367 | } | |
1368 | ||
1369 | /** | |
f7f4bccb | 1370 | * int jbd2_journal_stop() - complete a transaction |
470decc6 DK |
1371 | * @handle: tranaction to complete. |
1372 | * | |
1373 | * All done for a particular handle. | |
1374 | * | |
1375 | * There is not much action needed here. We just return any remaining | |
1376 | * buffer credits to the transaction and remove the handle. The only | |
1377 | * complication is that we need to start a commit operation if the | |
1378 | * filesystem is marked for synchronous update. | |
1379 | * | |
f7f4bccb | 1380 | * jbd2_journal_stop itself will not usually return an error, but it may |
470decc6 | 1381 | * do so in unusual circumstances. In particular, expect it to |
f7f4bccb | 1382 | * return -EIO if a jbd2_journal_abort has been executed since the |
470decc6 DK |
1383 | * transaction began. |
1384 | */ | |
f7f4bccb | 1385 | int jbd2_journal_stop(handle_t *handle) |
470decc6 DK |
1386 | { |
1387 | transaction_t *transaction = handle->h_transaction; | |
1388 | journal_t *journal = transaction->t_journal; | |
a51dca9c TT |
1389 | int err, wait_for_commit = 0; |
1390 | tid_t tid; | |
470decc6 DK |
1391 | pid_t pid; |
1392 | ||
470decc6 DK |
1393 | J_ASSERT(journal_current_handle() == handle); |
1394 | ||
1395 | if (is_handle_aborted(handle)) | |
1396 | err = -EIO; | |
3e2a532b | 1397 | else { |
a51dca9c | 1398 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 | 1399 | err = 0; |
3e2a532b | 1400 | } |
470decc6 DK |
1401 | |
1402 | if (--handle->h_ref > 0) { | |
1403 | jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1, | |
1404 | handle->h_ref); | |
1405 | return err; | |
1406 | } | |
1407 | ||
1408 | jbd_debug(4, "Handle %p going down\n", handle); | |
343d9c28 TT |
1409 | trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev, |
1410 | handle->h_transaction->t_tid, | |
1411 | handle->h_type, handle->h_line_no, | |
1412 | jiffies - handle->h_start_jiffies, | |
1413 | handle->h_sync, handle->h_requested_credits, | |
1414 | (handle->h_requested_credits - | |
1415 | handle->h_buffer_credits)); | |
470decc6 DK |
1416 | |
1417 | /* | |
1418 | * Implement synchronous transaction batching. If the handle | |
1419 | * was synchronous, don't force a commit immediately. Let's | |
e07f7183 JB |
1420 | * yield and let another thread piggyback onto this |
1421 | * transaction. Keep doing that while new threads continue to | |
1422 | * arrive. It doesn't cost much - we're about to run a commit | |
1423 | * and sleep on IO anyway. Speeds up many-threaded, many-dir | |
1424 | * operations by 30x or more... | |
1425 | * | |
1426 | * We try and optimize the sleep time against what the | |
1427 | * underlying disk can do, instead of having a static sleep | |
1428 | * time. This is useful for the case where our storage is so | |
1429 | * fast that it is more optimal to go ahead and force a flush | |
1430 | * and wait for the transaction to be committed than it is to | |
1431 | * wait for an arbitrary amount of time for new writers to | |
1432 | * join the transaction. We achieve this by measuring how | |
1433 | * long it takes to commit a transaction, and compare it with | |
1434 | * how long this transaction has been running, and if run time | |
1435 | * < commit time then we sleep for the delta and commit. This | |
1436 | * greatly helps super fast disks that would see slowdowns as | |
1437 | * more threads started doing fsyncs. | |
470decc6 | 1438 | * |
e07f7183 JB |
1439 | * But don't do this if this process was the most recent one |
1440 | * to perform a synchronous write. We do this to detect the | |
1441 | * case where a single process is doing a stream of sync | |
1442 | * writes. No point in waiting for joiners in that case. | |
470decc6 DK |
1443 | */ |
1444 | pid = current->pid; | |
1445 | if (handle->h_sync && journal->j_last_sync_writer != pid) { | |
e07f7183 JB |
1446 | u64 commit_time, trans_time; |
1447 | ||
470decc6 | 1448 | journal->j_last_sync_writer = pid; |
e07f7183 | 1449 | |
a931da6a | 1450 | read_lock(&journal->j_state_lock); |
e07f7183 | 1451 | commit_time = journal->j_average_commit_time; |
a931da6a | 1452 | read_unlock(&journal->j_state_lock); |
e07f7183 JB |
1453 | |
1454 | trans_time = ktime_to_ns(ktime_sub(ktime_get(), | |
1455 | transaction->t_start_time)); | |
1456 | ||
30773840 TT |
1457 | commit_time = max_t(u64, commit_time, |
1458 | 1000*journal->j_min_batch_time); | |
e07f7183 | 1459 | commit_time = min_t(u64, commit_time, |
30773840 | 1460 | 1000*journal->j_max_batch_time); |
e07f7183 JB |
1461 | |
1462 | if (trans_time < commit_time) { | |
1463 | ktime_t expires = ktime_add_ns(ktime_get(), | |
1464 | commit_time); | |
1465 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1466 | schedule_hrtimeout(&expires, HRTIMER_MODE_ABS); | |
1467 | } | |
470decc6 DK |
1468 | } |
1469 | ||
7058548c TT |
1470 | if (handle->h_sync) |
1471 | transaction->t_synchronous_commit = 1; | |
470decc6 | 1472 | current->journal_info = NULL; |
a51dca9c TT |
1473 | atomic_sub(handle->h_buffer_credits, |
1474 | &transaction->t_outstanding_credits); | |
470decc6 DK |
1475 | |
1476 | /* | |
1477 | * If the handle is marked SYNC, we need to set another commit | |
1478 | * going! We also want to force a commit if the current | |
1479 | * transaction is occupying too much of the log, or if the | |
1480 | * transaction is too old now. | |
1481 | */ | |
1482 | if (handle->h_sync || | |
a51dca9c TT |
1483 | (atomic_read(&transaction->t_outstanding_credits) > |
1484 | journal->j_max_transaction_buffers) || | |
1485 | time_after_eq(jiffies, transaction->t_expires)) { | |
470decc6 DK |
1486 | /* Do this even for aborted journals: an abort still |
1487 | * completes the commit thread, it just doesn't write | |
1488 | * anything to disk. */ | |
470decc6 | 1489 | |
470decc6 DK |
1490 | jbd_debug(2, "transaction too old, requesting commit for " |
1491 | "handle %p\n", handle); | |
1492 | /* This is non-blocking */ | |
c35a56a0 | 1493 | jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1494 | |
1495 | /* | |
f7f4bccb | 1496 | * Special case: JBD2_SYNC synchronous updates require us |
470decc6 DK |
1497 | * to wait for the commit to complete. |
1498 | */ | |
1499 | if (handle->h_sync && !(current->flags & PF_MEMALLOC)) | |
a51dca9c | 1500 | wait_for_commit = 1; |
470decc6 DK |
1501 | } |
1502 | ||
a51dca9c TT |
1503 | /* |
1504 | * Once we drop t_updates, if it goes to zero the transaction | |
25985edc | 1505 | * could start committing on us and eventually disappear. So |
a51dca9c TT |
1506 | * once we do this, we must not dereference transaction |
1507 | * pointer again. | |
1508 | */ | |
1509 | tid = transaction->t_tid; | |
1510 | if (atomic_dec_and_test(&transaction->t_updates)) { | |
1511 | wake_up(&journal->j_wait_updates); | |
1512 | if (journal->j_barrier_count) | |
1513 | wake_up(&journal->j_wait_transaction_locked); | |
1514 | } | |
1515 | ||
1516 | if (wait_for_commit) | |
1517 | err = jbd2_log_wait_commit(journal, tid); | |
1518 | ||
3295f0ef | 1519 | lock_map_release(&handle->h_lockdep_map); |
7b751066 | 1520 | |
af1e76d6 | 1521 | jbd2_free_handle(handle); |
470decc6 DK |
1522 | return err; |
1523 | } | |
1524 | ||
5648ba5b RD |
1525 | /** |
1526 | * int jbd2_journal_force_commit() - force any uncommitted transactions | |
470decc6 DK |
1527 | * @journal: journal to force |
1528 | * | |
1529 | * For synchronous operations: force any uncommitted transactions | |
1530 | * to disk. May seem kludgy, but it reuses all the handle batching | |
1531 | * code in a very simple manner. | |
1532 | */ | |
f7f4bccb | 1533 | int jbd2_journal_force_commit(journal_t *journal) |
470decc6 DK |
1534 | { |
1535 | handle_t *handle; | |
1536 | int ret; | |
1537 | ||
f7f4bccb | 1538 | handle = jbd2_journal_start(journal, 1); |
470decc6 DK |
1539 | if (IS_ERR(handle)) { |
1540 | ret = PTR_ERR(handle); | |
1541 | } else { | |
1542 | handle->h_sync = 1; | |
f7f4bccb | 1543 | ret = jbd2_journal_stop(handle); |
470decc6 DK |
1544 | } |
1545 | return ret; | |
1546 | } | |
1547 | ||
1548 | /* | |
1549 | * | |
1550 | * List management code snippets: various functions for manipulating the | |
1551 | * transaction buffer lists. | |
1552 | * | |
1553 | */ | |
1554 | ||
1555 | /* | |
1556 | * Append a buffer to a transaction list, given the transaction's list head | |
1557 | * pointer. | |
1558 | * | |
1559 | * j_list_lock is held. | |
1560 | * | |
1561 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1562 | */ | |
1563 | ||
1564 | static inline void | |
1565 | __blist_add_buffer(struct journal_head **list, struct journal_head *jh) | |
1566 | { | |
1567 | if (!*list) { | |
1568 | jh->b_tnext = jh->b_tprev = jh; | |
1569 | *list = jh; | |
1570 | } else { | |
1571 | /* Insert at the tail of the list to preserve order */ | |
1572 | struct journal_head *first = *list, *last = first->b_tprev; | |
1573 | jh->b_tprev = last; | |
1574 | jh->b_tnext = first; | |
1575 | last->b_tnext = first->b_tprev = jh; | |
1576 | } | |
1577 | } | |
1578 | ||
1579 | /* | |
1580 | * Remove a buffer from a transaction list, given the transaction's list | |
1581 | * head pointer. | |
1582 | * | |
1583 | * Called with j_list_lock held, and the journal may not be locked. | |
1584 | * | |
1585 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1586 | */ | |
1587 | ||
1588 | static inline void | |
1589 | __blist_del_buffer(struct journal_head **list, struct journal_head *jh) | |
1590 | { | |
1591 | if (*list == jh) { | |
1592 | *list = jh->b_tnext; | |
1593 | if (*list == jh) | |
1594 | *list = NULL; | |
1595 | } | |
1596 | jh->b_tprev->b_tnext = jh->b_tnext; | |
1597 | jh->b_tnext->b_tprev = jh->b_tprev; | |
1598 | } | |
1599 | ||
1600 | /* | |
1601 | * Remove a buffer from the appropriate transaction list. | |
1602 | * | |
1603 | * Note that this function can *change* the value of | |
87c89c23 JK |
1604 | * bh->b_transaction->t_buffers, t_forget, t_iobuf_list, t_shadow_list, |
1605 | * t_log_list or t_reserved_list. If the caller is holding onto a copy of one | |
1606 | * of these pointers, it could go bad. Generally the caller needs to re-read | |
1607 | * the pointer from the transaction_t. | |
470decc6 | 1608 | * |
5bebccf9 | 1609 | * Called under j_list_lock. |
470decc6 | 1610 | */ |
5bebccf9 | 1611 | static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh) |
470decc6 DK |
1612 | { |
1613 | struct journal_head **list = NULL; | |
1614 | transaction_t *transaction; | |
1615 | struct buffer_head *bh = jh2bh(jh); | |
1616 | ||
1617 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
1618 | transaction = jh->b_transaction; | |
1619 | if (transaction) | |
1620 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
1621 | ||
1622 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
1623 | if (jh->b_jlist != BJ_None) | |
4019191b | 1624 | J_ASSERT_JH(jh, transaction != NULL); |
470decc6 DK |
1625 | |
1626 | switch (jh->b_jlist) { | |
1627 | case BJ_None: | |
1628 | return; | |
470decc6 DK |
1629 | case BJ_Metadata: |
1630 | transaction->t_nr_buffers--; | |
1631 | J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0); | |
1632 | list = &transaction->t_buffers; | |
1633 | break; | |
1634 | case BJ_Forget: | |
1635 | list = &transaction->t_forget; | |
1636 | break; | |
1637 | case BJ_IO: | |
1638 | list = &transaction->t_iobuf_list; | |
1639 | break; | |
1640 | case BJ_Shadow: | |
1641 | list = &transaction->t_shadow_list; | |
1642 | break; | |
1643 | case BJ_LogCtl: | |
1644 | list = &transaction->t_log_list; | |
1645 | break; | |
1646 | case BJ_Reserved: | |
1647 | list = &transaction->t_reserved_list; | |
1648 | break; | |
470decc6 DK |
1649 | } |
1650 | ||
1651 | __blist_del_buffer(list, jh); | |
1652 | jh->b_jlist = BJ_None; | |
1653 | if (test_clear_buffer_jbddirty(bh)) | |
1654 | mark_buffer_dirty(bh); /* Expose it to the VM */ | |
1655 | } | |
1656 | ||
de1b7941 JK |
1657 | /* |
1658 | * Remove buffer from all transactions. | |
1659 | * | |
1660 | * Called with bh_state lock and j_list_lock | |
1661 | * | |
1662 | * jh and bh may be already freed when this function returns. | |
1663 | */ | |
1664 | static void __jbd2_journal_unfile_buffer(struct journal_head *jh) | |
470decc6 | 1665 | { |
f7f4bccb | 1666 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 | 1667 | jh->b_transaction = NULL; |
de1b7941 | 1668 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1669 | } |
1670 | ||
f7f4bccb | 1671 | void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 | 1672 | { |
de1b7941 JK |
1673 | struct buffer_head *bh = jh2bh(jh); |
1674 | ||
1675 | /* Get reference so that buffer cannot be freed before we unlock it */ | |
1676 | get_bh(bh); | |
1677 | jbd_lock_bh_state(bh); | |
470decc6 | 1678 | spin_lock(&journal->j_list_lock); |
f7f4bccb | 1679 | __jbd2_journal_unfile_buffer(jh); |
470decc6 | 1680 | spin_unlock(&journal->j_list_lock); |
de1b7941 JK |
1681 | jbd_unlock_bh_state(bh); |
1682 | __brelse(bh); | |
470decc6 DK |
1683 | } |
1684 | ||
1685 | /* | |
f7f4bccb | 1686 | * Called from jbd2_journal_try_to_free_buffers(). |
470decc6 DK |
1687 | * |
1688 | * Called under jbd_lock_bh_state(bh) | |
1689 | */ | |
1690 | static void | |
1691 | __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh) | |
1692 | { | |
1693 | struct journal_head *jh; | |
1694 | ||
1695 | jh = bh2jh(bh); | |
1696 | ||
1697 | if (buffer_locked(bh) || buffer_dirty(bh)) | |
1698 | goto out; | |
1699 | ||
4019191b | 1700 | if (jh->b_next_transaction != NULL) |
470decc6 DK |
1701 | goto out; |
1702 | ||
1703 | spin_lock(&journal->j_list_lock); | |
87c89c23 | 1704 | if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) { |
470decc6 | 1705 | /* written-back checkpointed metadata buffer */ |
c254c9ec JK |
1706 | JBUFFER_TRACE(jh, "remove from checkpoint list"); |
1707 | __jbd2_journal_remove_checkpoint(jh); | |
470decc6 DK |
1708 | } |
1709 | spin_unlock(&journal->j_list_lock); | |
1710 | out: | |
1711 | return; | |
1712 | } | |
1713 | ||
470decc6 | 1714 | /** |
f7f4bccb | 1715 | * int jbd2_journal_try_to_free_buffers() - try to free page buffers. |
470decc6 DK |
1716 | * @journal: journal for operation |
1717 | * @page: to try and free | |
530576bb MC |
1718 | * @gfp_mask: we use the mask to detect how hard should we try to release |
1719 | * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to | |
1720 | * release the buffers. | |
470decc6 DK |
1721 | * |
1722 | * | |
1723 | * For all the buffers on this page, | |
1724 | * if they are fully written out ordered data, move them onto BUF_CLEAN | |
1725 | * so try_to_free_buffers() can reap them. | |
1726 | * | |
1727 | * This function returns non-zero if we wish try_to_free_buffers() | |
1728 | * to be called. We do this if the page is releasable by try_to_free_buffers(). | |
1729 | * We also do it if the page has locked or dirty buffers and the caller wants | |
1730 | * us to perform sync or async writeout. | |
1731 | * | |
1732 | * This complicates JBD locking somewhat. We aren't protected by the | |
1733 | * BKL here. We wish to remove the buffer from its committing or | |
f7f4bccb | 1734 | * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer. |
470decc6 DK |
1735 | * |
1736 | * This may *change* the value of transaction_t->t_datalist, so anyone | |
1737 | * who looks at t_datalist needs to lock against this function. | |
1738 | * | |
f7f4bccb MC |
1739 | * Even worse, someone may be doing a jbd2_journal_dirty_data on this |
1740 | * buffer. So we need to lock against that. jbd2_journal_dirty_data() | |
470decc6 DK |
1741 | * will come out of the lock with the buffer dirty, which makes it |
1742 | * ineligible for release here. | |
1743 | * | |
1744 | * Who else is affected by this? hmm... Really the only contender | |
1745 | * is do_get_write_access() - it could be looking at the buffer while | |
1746 | * journal_try_to_free_buffer() is changing its state. But that | |
1747 | * cannot happen because we never reallocate freed data as metadata | |
1748 | * while the data is part of a transaction. Yes? | |
530576bb MC |
1749 | * |
1750 | * Return 0 on failure, 1 on success | |
470decc6 | 1751 | */ |
f7f4bccb | 1752 | int jbd2_journal_try_to_free_buffers(journal_t *journal, |
530576bb | 1753 | struct page *page, gfp_t gfp_mask) |
470decc6 DK |
1754 | { |
1755 | struct buffer_head *head; | |
1756 | struct buffer_head *bh; | |
1757 | int ret = 0; | |
1758 | ||
1759 | J_ASSERT(PageLocked(page)); | |
1760 | ||
1761 | head = page_buffers(page); | |
1762 | bh = head; | |
1763 | do { | |
1764 | struct journal_head *jh; | |
1765 | ||
1766 | /* | |
1767 | * We take our own ref against the journal_head here to avoid | |
1768 | * having to add tons of locking around each instance of | |
530576bb | 1769 | * jbd2_journal_put_journal_head(). |
470decc6 | 1770 | */ |
f7f4bccb | 1771 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
1772 | if (!jh) |
1773 | continue; | |
1774 | ||
1775 | jbd_lock_bh_state(bh); | |
1776 | __journal_try_to_free_buffer(journal, bh); | |
f7f4bccb | 1777 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1778 | jbd_unlock_bh_state(bh); |
1779 | if (buffer_jbd(bh)) | |
1780 | goto busy; | |
1781 | } while ((bh = bh->b_this_page) != head); | |
530576bb | 1782 | |
470decc6 | 1783 | ret = try_to_free_buffers(page); |
530576bb | 1784 | |
470decc6 DK |
1785 | busy: |
1786 | return ret; | |
1787 | } | |
1788 | ||
1789 | /* | |
1790 | * This buffer is no longer needed. If it is on an older transaction's | |
1791 | * checkpoint list we need to record it on this transaction's forget list | |
1792 | * to pin this buffer (and hence its checkpointing transaction) down until | |
1793 | * this transaction commits. If the buffer isn't on a checkpoint list, we | |
1794 | * release it. | |
1795 | * Returns non-zero if JBD no longer has an interest in the buffer. | |
1796 | * | |
1797 | * Called under j_list_lock. | |
1798 | * | |
1799 | * Called under jbd_lock_bh_state(bh). | |
1800 | */ | |
1801 | static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction) | |
1802 | { | |
1803 | int may_free = 1; | |
1804 | struct buffer_head *bh = jh2bh(jh); | |
1805 | ||
470decc6 DK |
1806 | if (jh->b_cp_transaction) { |
1807 | JBUFFER_TRACE(jh, "on running+cp transaction"); | |
de1b7941 | 1808 | __jbd2_journal_temp_unlink_buffer(jh); |
f91d1d04 JK |
1809 | /* |
1810 | * We don't want to write the buffer anymore, clear the | |
1811 | * bit so that we don't confuse checks in | |
1812 | * __journal_file_buffer | |
1813 | */ | |
1814 | clear_buffer_dirty(bh); | |
f7f4bccb | 1815 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); |
470decc6 DK |
1816 | may_free = 0; |
1817 | } else { | |
1818 | JBUFFER_TRACE(jh, "on running transaction"); | |
de1b7941 | 1819 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
1820 | } |
1821 | return may_free; | |
1822 | } | |
1823 | ||
1824 | /* | |
f7f4bccb | 1825 | * jbd2_journal_invalidatepage |
470decc6 DK |
1826 | * |
1827 | * This code is tricky. It has a number of cases to deal with. | |
1828 | * | |
1829 | * There are two invariants which this code relies on: | |
1830 | * | |
1831 | * i_size must be updated on disk before we start calling invalidatepage on the | |
1832 | * data. | |
1833 | * | |
1834 | * This is done in ext3 by defining an ext3_setattr method which | |
1835 | * updates i_size before truncate gets going. By maintaining this | |
1836 | * invariant, we can be sure that it is safe to throw away any buffers | |
1837 | * attached to the current transaction: once the transaction commits, | |
1838 | * we know that the data will not be needed. | |
1839 | * | |
1840 | * Note however that we can *not* throw away data belonging to the | |
1841 | * previous, committing transaction! | |
1842 | * | |
1843 | * Any disk blocks which *are* part of the previous, committing | |
1844 | * transaction (and which therefore cannot be discarded immediately) are | |
1845 | * not going to be reused in the new running transaction | |
1846 | * | |
1847 | * The bitmap committed_data images guarantee this: any block which is | |
1848 | * allocated in one transaction and removed in the next will be marked | |
1849 | * as in-use in the committed_data bitmap, so cannot be reused until | |
1850 | * the next transaction to delete the block commits. This means that | |
1851 | * leaving committing buffers dirty is quite safe: the disk blocks | |
1852 | * cannot be reallocated to a different file and so buffer aliasing is | |
1853 | * not possible. | |
1854 | * | |
1855 | * | |
1856 | * The above applies mainly to ordered data mode. In writeback mode we | |
1857 | * don't make guarantees about the order in which data hits disk --- in | |
1858 | * particular we don't guarantee that new dirty data is flushed before | |
1859 | * transaction commit --- so it is always safe just to discard data | |
1860 | * immediately in that mode. --sct | |
1861 | */ | |
1862 | ||
1863 | /* | |
1864 | * The journal_unmap_buffer helper function returns zero if the buffer | |
1865 | * concerned remains pinned as an anonymous buffer belonging to an older | |
1866 | * transaction. | |
1867 | * | |
1868 | * We're outside-transaction here. Either or both of j_running_transaction | |
1869 | * and j_committing_transaction may be NULL. | |
1870 | */ | |
b794e7a6 JK |
1871 | static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh, |
1872 | int partial_page) | |
470decc6 DK |
1873 | { |
1874 | transaction_t *transaction; | |
1875 | struct journal_head *jh; | |
1876 | int may_free = 1; | |
470decc6 DK |
1877 | |
1878 | BUFFER_TRACE(bh, "entry"); | |
1879 | ||
1880 | /* | |
1881 | * It is safe to proceed here without the j_list_lock because the | |
1882 | * buffers cannot be stolen by try_to_free_buffers as long as we are | |
1883 | * holding the page lock. --sct | |
1884 | */ | |
1885 | ||
1886 | if (!buffer_jbd(bh)) | |
1887 | goto zap_buffer_unlocked; | |
1888 | ||
87c89c23 | 1889 | /* OK, we have data buffer in journaled mode */ |
a931da6a | 1890 | write_lock(&journal->j_state_lock); |
470decc6 DK |
1891 | jbd_lock_bh_state(bh); |
1892 | spin_lock(&journal->j_list_lock); | |
1893 | ||
f7f4bccb | 1894 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
1895 | if (!jh) |
1896 | goto zap_buffer_no_jh; | |
1897 | ||
ba869023 | 1898 | /* |
1899 | * We cannot remove the buffer from checkpoint lists until the | |
1900 | * transaction adding inode to orphan list (let's call it T) | |
1901 | * is committed. Otherwise if the transaction changing the | |
1902 | * buffer would be cleaned from the journal before T is | |
1903 | * committed, a crash will cause that the correct contents of | |
1904 | * the buffer will be lost. On the other hand we have to | |
1905 | * clear the buffer dirty bit at latest at the moment when the | |
1906 | * transaction marking the buffer as freed in the filesystem | |
1907 | * structures is committed because from that moment on the | |
b794e7a6 | 1908 | * block can be reallocated and used by a different page. |
ba869023 | 1909 | * Since the block hasn't been freed yet but the inode has |
1910 | * already been added to orphan list, it is safe for us to add | |
1911 | * the buffer to BJ_Forget list of the newest transaction. | |
b794e7a6 JK |
1912 | * |
1913 | * Also we have to clear buffer_mapped flag of a truncated buffer | |
1914 | * because the buffer_head may be attached to the page straddling | |
1915 | * i_size (can happen only when blocksize < pagesize) and thus the | |
1916 | * buffer_head can be reused when the file is extended again. So we end | |
1917 | * up keeping around invalidated buffers attached to transactions' | |
1918 | * BJ_Forget list just to stop checkpointing code from cleaning up | |
1919 | * the transaction this buffer was modified in. | |
ba869023 | 1920 | */ |
470decc6 DK |
1921 | transaction = jh->b_transaction; |
1922 | if (transaction == NULL) { | |
1923 | /* First case: not on any transaction. If it | |
1924 | * has no checkpoint link, then we can zap it: | |
1925 | * it's a writeback-mode buffer so we don't care | |
1926 | * if it hits disk safely. */ | |
1927 | if (!jh->b_cp_transaction) { | |
1928 | JBUFFER_TRACE(jh, "not on any transaction: zap"); | |
1929 | goto zap_buffer; | |
1930 | } | |
1931 | ||
1932 | if (!buffer_dirty(bh)) { | |
1933 | /* bdflush has written it. We can drop it now */ | |
1934 | goto zap_buffer; | |
1935 | } | |
1936 | ||
1937 | /* OK, it must be in the journal but still not | |
1938 | * written fully to disk: it's metadata or | |
1939 | * journaled data... */ | |
1940 | ||
1941 | if (journal->j_running_transaction) { | |
1942 | /* ... and once the current transaction has | |
1943 | * committed, the buffer won't be needed any | |
1944 | * longer. */ | |
1945 | JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget"); | |
b794e7a6 | 1946 | may_free = __dispose_buffer(jh, |
470decc6 | 1947 | journal->j_running_transaction); |
b794e7a6 | 1948 | goto zap_buffer; |
470decc6 DK |
1949 | } else { |
1950 | /* There is no currently-running transaction. So the | |
1951 | * orphan record which we wrote for this file must have | |
1952 | * passed into commit. We must attach this buffer to | |
1953 | * the committing transaction, if it exists. */ | |
1954 | if (journal->j_committing_transaction) { | |
1955 | JBUFFER_TRACE(jh, "give to committing trans"); | |
b794e7a6 | 1956 | may_free = __dispose_buffer(jh, |
470decc6 | 1957 | journal->j_committing_transaction); |
b794e7a6 | 1958 | goto zap_buffer; |
470decc6 DK |
1959 | } else { |
1960 | /* The orphan record's transaction has | |
1961 | * committed. We can cleanse this buffer */ | |
1962 | clear_buffer_jbddirty(bh); | |
1963 | goto zap_buffer; | |
1964 | } | |
1965 | } | |
1966 | } else if (transaction == journal->j_committing_transaction) { | |
9b57988d | 1967 | JBUFFER_TRACE(jh, "on committing transaction"); |
470decc6 | 1968 | /* |
ba869023 | 1969 | * The buffer is committing, we simply cannot touch |
b794e7a6 JK |
1970 | * it. If the page is straddling i_size we have to wait |
1971 | * for commit and try again. | |
1972 | */ | |
1973 | if (partial_page) { | |
b794e7a6 JK |
1974 | jbd2_journal_put_journal_head(jh); |
1975 | spin_unlock(&journal->j_list_lock); | |
1976 | jbd_unlock_bh_state(bh); | |
1977 | write_unlock(&journal->j_state_lock); | |
53e87268 | 1978 | return -EBUSY; |
b794e7a6 JK |
1979 | } |
1980 | /* | |
1981 | * OK, buffer won't be reachable after truncate. We just set | |
1982 | * j_next_transaction to the running transaction (if there is | |
1983 | * one) and mark buffer as freed so that commit code knows it | |
1984 | * should clear dirty bits when it is done with the buffer. | |
ba869023 | 1985 | */ |
470decc6 | 1986 | set_buffer_freed(bh); |
ba869023 | 1987 | if (journal->j_running_transaction && buffer_jbddirty(bh)) |
1988 | jh->b_next_transaction = journal->j_running_transaction; | |
f7f4bccb | 1989 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1990 | spin_unlock(&journal->j_list_lock); |
1991 | jbd_unlock_bh_state(bh); | |
a931da6a | 1992 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
1993 | return 0; |
1994 | } else { | |
1995 | /* Good, the buffer belongs to the running transaction. | |
1996 | * We are writing our own transaction's data, not any | |
1997 | * previous one's, so it is safe to throw it away | |
1998 | * (remember that we expect the filesystem to have set | |
1999 | * i_size already for this truncate so recovery will not | |
2000 | * expose the disk blocks we are discarding here.) */ | |
2001 | J_ASSERT_JH(jh, transaction == journal->j_running_transaction); | |
9b57988d | 2002 | JBUFFER_TRACE(jh, "on running transaction"); |
470decc6 DK |
2003 | may_free = __dispose_buffer(jh, transaction); |
2004 | } | |
2005 | ||
2006 | zap_buffer: | |
b794e7a6 JK |
2007 | /* |
2008 | * This is tricky. Although the buffer is truncated, it may be reused | |
2009 | * if blocksize < pagesize and it is attached to the page straddling | |
2010 | * EOF. Since the buffer might have been added to BJ_Forget list of the | |
2011 | * running transaction, journal_get_write_access() won't clear | |
2012 | * b_modified and credit accounting gets confused. So clear b_modified | |
2013 | * here. | |
2014 | */ | |
2015 | jh->b_modified = 0; | |
f7f4bccb | 2016 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2017 | zap_buffer_no_jh: |
2018 | spin_unlock(&journal->j_list_lock); | |
2019 | jbd_unlock_bh_state(bh); | |
a931da6a | 2020 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2021 | zap_buffer_unlocked: |
2022 | clear_buffer_dirty(bh); | |
2023 | J_ASSERT_BH(bh, !buffer_jbddirty(bh)); | |
2024 | clear_buffer_mapped(bh); | |
2025 | clear_buffer_req(bh); | |
2026 | clear_buffer_new(bh); | |
15291164 ES |
2027 | clear_buffer_delay(bh); |
2028 | clear_buffer_unwritten(bh); | |
470decc6 DK |
2029 | bh->b_bdev = NULL; |
2030 | return may_free; | |
2031 | } | |
2032 | ||
2033 | /** | |
f7f4bccb | 2034 | * void jbd2_journal_invalidatepage() |
470decc6 DK |
2035 | * @journal: journal to use for flush... |
2036 | * @page: page to flush | |
2037 | * @offset: length of page to invalidate. | |
2038 | * | |
53e87268 JK |
2039 | * Reap page buffers containing data after offset in page. Can return -EBUSY |
2040 | * if buffers are part of the committing transaction and the page is straddling | |
2041 | * i_size. Caller then has to wait for current commit and try again. | |
470decc6 | 2042 | */ |
53e87268 JK |
2043 | int jbd2_journal_invalidatepage(journal_t *journal, |
2044 | struct page *page, | |
2045 | unsigned long offset) | |
470decc6 DK |
2046 | { |
2047 | struct buffer_head *head, *bh, *next; | |
2048 | unsigned int curr_off = 0; | |
2049 | int may_free = 1; | |
53e87268 | 2050 | int ret = 0; |
470decc6 DK |
2051 | |
2052 | if (!PageLocked(page)) | |
2053 | BUG(); | |
2054 | if (!page_has_buffers(page)) | |
53e87268 | 2055 | return 0; |
470decc6 DK |
2056 | |
2057 | /* We will potentially be playing with lists other than just the | |
2058 | * data lists (especially for journaled data mode), so be | |
2059 | * cautious in our locking. */ | |
2060 | ||
2061 | head = bh = page_buffers(page); | |
2062 | do { | |
2063 | unsigned int next_off = curr_off + bh->b_size; | |
2064 | next = bh->b_this_page; | |
2065 | ||
2066 | if (offset <= curr_off) { | |
2067 | /* This block is wholly outside the truncation point */ | |
2068 | lock_buffer(bh); | |
53e87268 | 2069 | ret = journal_unmap_buffer(journal, bh, offset > 0); |
470decc6 | 2070 | unlock_buffer(bh); |
53e87268 JK |
2071 | if (ret < 0) |
2072 | return ret; | |
2073 | may_free &= ret; | |
470decc6 DK |
2074 | } |
2075 | curr_off = next_off; | |
2076 | bh = next; | |
2077 | ||
2078 | } while (bh != head); | |
2079 | ||
2080 | if (!offset) { | |
2081 | if (may_free && try_to_free_buffers(page)) | |
2082 | J_ASSERT(!page_has_buffers(page)); | |
2083 | } | |
53e87268 | 2084 | return 0; |
470decc6 DK |
2085 | } |
2086 | ||
2087 | /* | |
2088 | * File a buffer on the given transaction list. | |
2089 | */ | |
f7f4bccb | 2090 | void __jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2091 | transaction_t *transaction, int jlist) |
2092 | { | |
2093 | struct journal_head **list = NULL; | |
2094 | int was_dirty = 0; | |
2095 | struct buffer_head *bh = jh2bh(jh); | |
2096 | ||
2097 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
2098 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
2099 | ||
2100 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
2101 | J_ASSERT_JH(jh, jh->b_transaction == transaction || | |
4019191b | 2102 | jh->b_transaction == NULL); |
470decc6 DK |
2103 | |
2104 | if (jh->b_transaction && jh->b_jlist == jlist) | |
2105 | return; | |
2106 | ||
470decc6 DK |
2107 | if (jlist == BJ_Metadata || jlist == BJ_Reserved || |
2108 | jlist == BJ_Shadow || jlist == BJ_Forget) { | |
f91d1d04 JK |
2109 | /* |
2110 | * For metadata buffers, we track dirty bit in buffer_jbddirty | |
2111 | * instead of buffer_dirty. We should not see a dirty bit set | |
2112 | * here because we clear it in do_get_write_access but e.g. | |
2113 | * tune2fs can modify the sb and set the dirty bit at any time | |
2114 | * so we try to gracefully handle that. | |
2115 | */ | |
2116 | if (buffer_dirty(bh)) | |
2117 | warn_dirty_buffer(bh); | |
470decc6 DK |
2118 | if (test_clear_buffer_dirty(bh) || |
2119 | test_clear_buffer_jbddirty(bh)) | |
2120 | was_dirty = 1; | |
2121 | } | |
2122 | ||
2123 | if (jh->b_transaction) | |
f7f4bccb | 2124 | __jbd2_journal_temp_unlink_buffer(jh); |
de1b7941 JK |
2125 | else |
2126 | jbd2_journal_grab_journal_head(bh); | |
470decc6 DK |
2127 | jh->b_transaction = transaction; |
2128 | ||
2129 | switch (jlist) { | |
2130 | case BJ_None: | |
2131 | J_ASSERT_JH(jh, !jh->b_committed_data); | |
2132 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
2133 | return; | |
470decc6 DK |
2134 | case BJ_Metadata: |
2135 | transaction->t_nr_buffers++; | |
2136 | list = &transaction->t_buffers; | |
2137 | break; | |
2138 | case BJ_Forget: | |
2139 | list = &transaction->t_forget; | |
2140 | break; | |
2141 | case BJ_IO: | |
2142 | list = &transaction->t_iobuf_list; | |
2143 | break; | |
2144 | case BJ_Shadow: | |
2145 | list = &transaction->t_shadow_list; | |
2146 | break; | |
2147 | case BJ_LogCtl: | |
2148 | list = &transaction->t_log_list; | |
2149 | break; | |
2150 | case BJ_Reserved: | |
2151 | list = &transaction->t_reserved_list; | |
2152 | break; | |
470decc6 DK |
2153 | } |
2154 | ||
2155 | __blist_add_buffer(list, jh); | |
2156 | jh->b_jlist = jlist; | |
2157 | ||
2158 | if (was_dirty) | |
2159 | set_buffer_jbddirty(bh); | |
2160 | } | |
2161 | ||
f7f4bccb | 2162 | void jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2163 | transaction_t *transaction, int jlist) |
2164 | { | |
2165 | jbd_lock_bh_state(jh2bh(jh)); | |
2166 | spin_lock(&transaction->t_journal->j_list_lock); | |
f7f4bccb | 2167 | __jbd2_journal_file_buffer(jh, transaction, jlist); |
470decc6 DK |
2168 | spin_unlock(&transaction->t_journal->j_list_lock); |
2169 | jbd_unlock_bh_state(jh2bh(jh)); | |
2170 | } | |
2171 | ||
2172 | /* | |
2173 | * Remove a buffer from its current buffer list in preparation for | |
2174 | * dropping it from its current transaction entirely. If the buffer has | |
2175 | * already started to be used by a subsequent transaction, refile the | |
2176 | * buffer on that transaction's metadata list. | |
2177 | * | |
de1b7941 | 2178 | * Called under j_list_lock |
470decc6 | 2179 | * Called under jbd_lock_bh_state(jh2bh(jh)) |
de1b7941 JK |
2180 | * |
2181 | * jh and bh may be already free when this function returns | |
470decc6 | 2182 | */ |
f7f4bccb | 2183 | void __jbd2_journal_refile_buffer(struct journal_head *jh) |
470decc6 | 2184 | { |
ba869023 | 2185 | int was_dirty, jlist; |
470decc6 DK |
2186 | struct buffer_head *bh = jh2bh(jh); |
2187 | ||
2188 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
2189 | if (jh->b_transaction) | |
2190 | assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock); | |
2191 | ||
2192 | /* If the buffer is now unused, just drop it. */ | |
2193 | if (jh->b_next_transaction == NULL) { | |
f7f4bccb | 2194 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
2195 | return; |
2196 | } | |
2197 | ||
2198 | /* | |
2199 | * It has been modified by a later transaction: add it to the new | |
2200 | * transaction's metadata list. | |
2201 | */ | |
2202 | ||
2203 | was_dirty = test_clear_buffer_jbddirty(bh); | |
f7f4bccb | 2204 | __jbd2_journal_temp_unlink_buffer(jh); |
de1b7941 JK |
2205 | /* |
2206 | * We set b_transaction here because b_next_transaction will inherit | |
2207 | * our jh reference and thus __jbd2_journal_file_buffer() must not | |
2208 | * take a new one. | |
2209 | */ | |
470decc6 DK |
2210 | jh->b_transaction = jh->b_next_transaction; |
2211 | jh->b_next_transaction = NULL; | |
ba869023 | 2212 | if (buffer_freed(bh)) |
2213 | jlist = BJ_Forget; | |
2214 | else if (jh->b_modified) | |
2215 | jlist = BJ_Metadata; | |
2216 | else | |
2217 | jlist = BJ_Reserved; | |
2218 | __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist); | |
470decc6 DK |
2219 | J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING); |
2220 | ||
2221 | if (was_dirty) | |
2222 | set_buffer_jbddirty(bh); | |
2223 | } | |
2224 | ||
2225 | /* | |
de1b7941 JK |
2226 | * __jbd2_journal_refile_buffer() with necessary locking added. We take our |
2227 | * bh reference so that we can safely unlock bh. | |
2228 | * | |
2229 | * The jh and bh may be freed by this call. | |
470decc6 | 2230 | */ |
f7f4bccb | 2231 | void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 DK |
2232 | { |
2233 | struct buffer_head *bh = jh2bh(jh); | |
2234 | ||
de1b7941 JK |
2235 | /* Get reference so that buffer cannot be freed before we unlock it */ |
2236 | get_bh(bh); | |
470decc6 DK |
2237 | jbd_lock_bh_state(bh); |
2238 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 2239 | __jbd2_journal_refile_buffer(jh); |
470decc6 | 2240 | jbd_unlock_bh_state(bh); |
470decc6 DK |
2241 | spin_unlock(&journal->j_list_lock); |
2242 | __brelse(bh); | |
2243 | } | |
c851ed54 JK |
2244 | |
2245 | /* | |
2246 | * File inode in the inode list of the handle's transaction | |
2247 | */ | |
2248 | int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode) | |
2249 | { | |
2250 | transaction_t *transaction = handle->h_transaction; | |
2251 | journal_t *journal = transaction->t_journal; | |
2252 | ||
2253 | if (is_handle_aborted(handle)) | |
2254 | return -EIO; | |
2255 | ||
2256 | jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino, | |
2257 | transaction->t_tid); | |
2258 | ||
2259 | /* | |
2260 | * First check whether inode isn't already on the transaction's | |
2261 | * lists without taking the lock. Note that this check is safe | |
2262 | * without the lock as we cannot race with somebody removing inode | |
2263 | * from the transaction. The reason is that we remove inode from the | |
2264 | * transaction only in journal_release_jbd_inode() and when we commit | |
2265 | * the transaction. We are guarded from the first case by holding | |
2266 | * a reference to the inode. We are safe against the second case | |
2267 | * because if jinode->i_transaction == transaction, commit code | |
2268 | * cannot touch the transaction because we hold reference to it, | |
2269 | * and if jinode->i_next_transaction == transaction, commit code | |
2270 | * will only file the inode where we want it. | |
2271 | */ | |
2272 | if (jinode->i_transaction == transaction || | |
2273 | jinode->i_next_transaction == transaction) | |
2274 | return 0; | |
2275 | ||
2276 | spin_lock(&journal->j_list_lock); | |
2277 | ||
2278 | if (jinode->i_transaction == transaction || | |
2279 | jinode->i_next_transaction == transaction) | |
2280 | goto done; | |
2281 | ||
81be12c8 JK |
2282 | /* |
2283 | * We only ever set this variable to 1 so the test is safe. Since | |
2284 | * t_need_data_flush is likely to be set, we do the test to save some | |
2285 | * cacheline bouncing | |
2286 | */ | |
2287 | if (!transaction->t_need_data_flush) | |
2288 | transaction->t_need_data_flush = 1; | |
c851ed54 JK |
2289 | /* On some different transaction's list - should be |
2290 | * the committing one */ | |
2291 | if (jinode->i_transaction) { | |
2292 | J_ASSERT(jinode->i_next_transaction == NULL); | |
2293 | J_ASSERT(jinode->i_transaction == | |
2294 | journal->j_committing_transaction); | |
2295 | jinode->i_next_transaction = transaction; | |
2296 | goto done; | |
2297 | } | |
2298 | /* Not on any transaction list... */ | |
2299 | J_ASSERT(!jinode->i_next_transaction); | |
2300 | jinode->i_transaction = transaction; | |
2301 | list_add(&jinode->i_list, &transaction->t_inode_list); | |
2302 | done: | |
2303 | spin_unlock(&journal->j_list_lock); | |
2304 | ||
2305 | return 0; | |
2306 | } | |
2307 | ||
2308 | /* | |
7f5aa215 JK |
2309 | * File truncate and transaction commit interact with each other in a |
2310 | * non-trivial way. If a transaction writing data block A is | |
2311 | * committing, we cannot discard the data by truncate until we have | |
2312 | * written them. Otherwise if we crashed after the transaction with | |
2313 | * write has committed but before the transaction with truncate has | |
2314 | * committed, we could see stale data in block A. This function is a | |
2315 | * helper to solve this problem. It starts writeout of the truncated | |
2316 | * part in case it is in the committing transaction. | |
2317 | * | |
2318 | * Filesystem code must call this function when inode is journaled in | |
2319 | * ordered mode before truncation happens and after the inode has been | |
2320 | * placed on orphan list with the new inode size. The second condition | |
2321 | * avoids the race that someone writes new data and we start | |
2322 | * committing the transaction after this function has been called but | |
2323 | * before a transaction for truncate is started (and furthermore it | |
2324 | * allows us to optimize the case where the addition to orphan list | |
2325 | * happens in the same transaction as write --- we don't have to write | |
2326 | * any data in such case). | |
c851ed54 | 2327 | */ |
7f5aa215 JK |
2328 | int jbd2_journal_begin_ordered_truncate(journal_t *journal, |
2329 | struct jbd2_inode *jinode, | |
c851ed54 JK |
2330 | loff_t new_size) |
2331 | { | |
7f5aa215 | 2332 | transaction_t *inode_trans, *commit_trans; |
c851ed54 JK |
2333 | int ret = 0; |
2334 | ||
7f5aa215 JK |
2335 | /* This is a quick check to avoid locking if not necessary */ |
2336 | if (!jinode->i_transaction) | |
c851ed54 | 2337 | goto out; |
7f5aa215 JK |
2338 | /* Locks are here just to force reading of recent values, it is |
2339 | * enough that the transaction was not committing before we started | |
2340 | * a transaction adding the inode to orphan list */ | |
a931da6a | 2341 | read_lock(&journal->j_state_lock); |
c851ed54 | 2342 | commit_trans = journal->j_committing_transaction; |
a931da6a | 2343 | read_unlock(&journal->j_state_lock); |
7f5aa215 JK |
2344 | spin_lock(&journal->j_list_lock); |
2345 | inode_trans = jinode->i_transaction; | |
2346 | spin_unlock(&journal->j_list_lock); | |
2347 | if (inode_trans == commit_trans) { | |
2348 | ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping, | |
c851ed54 JK |
2349 | new_size, LLONG_MAX); |
2350 | if (ret) | |
2351 | jbd2_journal_abort(journal, ret); | |
2352 | } | |
2353 | out: | |
2354 | return ret; | |
2355 | } |