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 | 91 | atomic_set(&transaction->t_updates, 0); |
8f7d89f3 JK |
92 | atomic_set(&transaction->t_outstanding_credits, |
93 | atomic_read(&journal->j_reserved_credits)); | |
8dd42046 | 94 | atomic_set(&transaction->t_handle_count, 0); |
c851ed54 | 95 | INIT_LIST_HEAD(&transaction->t_inode_list); |
3e624fc7 | 96 | INIT_LIST_HEAD(&transaction->t_private_list); |
470decc6 DK |
97 | |
98 | /* Set up the commit timer for the new transaction. */ | |
b1f485f2 | 99 | journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires); |
470decc6 DK |
100 | add_timer(&journal->j_commit_timer); |
101 | ||
102 | J_ASSERT(journal->j_running_transaction == NULL); | |
103 | journal->j_running_transaction = transaction; | |
8e85fb3f JL |
104 | transaction->t_max_wait = 0; |
105 | transaction->t_start = jiffies; | |
9fff24aa | 106 | transaction->t_requested = 0; |
470decc6 DK |
107 | |
108 | return transaction; | |
109 | } | |
110 | ||
111 | /* | |
112 | * Handle management. | |
113 | * | |
114 | * A handle_t is an object which represents a single atomic update to a | |
115 | * filesystem, and which tracks all of the modifications which form part | |
116 | * of that one update. | |
117 | */ | |
118 | ||
6d0bf005 | 119 | /* |
28e35e42 | 120 | * Update transaction's maximum wait time, if debugging is enabled. |
6d0bf005 TT |
121 | * |
122 | * In order for t_max_wait to be reliable, it must be protected by a | |
123 | * lock. But doing so will mean that start_this_handle() can not be | |
124 | * run in parallel on SMP systems, which limits our scalability. So | |
125 | * unless debugging is enabled, we no longer update t_max_wait, which | |
126 | * means that maximum wait time reported by the jbd2_run_stats | |
127 | * tracepoint will always be zero. | |
128 | */ | |
28e35e42 TM |
129 | static inline void update_t_max_wait(transaction_t *transaction, |
130 | unsigned long ts) | |
6d0bf005 TT |
131 | { |
132 | #ifdef CONFIG_JBD2_DEBUG | |
6d0bf005 TT |
133 | if (jbd2_journal_enable_debug && |
134 | time_after(transaction->t_start, ts)) { | |
135 | ts = jbd2_time_diff(ts, transaction->t_start); | |
136 | spin_lock(&transaction->t_handle_lock); | |
137 | if (ts > transaction->t_max_wait) | |
138 | transaction->t_max_wait = ts; | |
139 | spin_unlock(&transaction->t_handle_lock); | |
140 | } | |
141 | #endif | |
142 | } | |
143 | ||
8f7d89f3 JK |
144 | /* |
145 | * Wait until running transaction passes T_LOCKED state. Also starts the commit | |
146 | * if needed. The function expects running transaction to exist and releases | |
147 | * j_state_lock. | |
148 | */ | |
149 | static void wait_transaction_locked(journal_t *journal) | |
150 | __releases(journal->j_state_lock) | |
151 | { | |
152 | DEFINE_WAIT(wait); | |
153 | int need_to_start; | |
154 | tid_t tid = journal->j_running_transaction->t_tid; | |
155 | ||
156 | prepare_to_wait(&journal->j_wait_transaction_locked, &wait, | |
157 | TASK_UNINTERRUPTIBLE); | |
158 | need_to_start = !tid_geq(journal->j_commit_request, tid); | |
159 | read_unlock(&journal->j_state_lock); | |
160 | if (need_to_start) | |
161 | jbd2_log_start_commit(journal, tid); | |
162 | schedule(); | |
163 | finish_wait(&journal->j_wait_transaction_locked, &wait); | |
164 | } | |
165 | ||
166 | static void sub_reserved_credits(journal_t *journal, int blocks) | |
167 | { | |
168 | atomic_sub(blocks, &journal->j_reserved_credits); | |
169 | wake_up(&journal->j_wait_reserved); | |
170 | } | |
171 | ||
172 | /* | |
173 | * Wait until we can add credits for handle to the running transaction. Called | |
174 | * with j_state_lock held for reading. Returns 0 if handle joined the running | |
175 | * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and | |
176 | * caller must retry. | |
177 | */ | |
178 | static int add_transaction_credits(journal_t *journal, int blocks, | |
179 | int rsv_blocks) | |
180 | { | |
181 | transaction_t *t = journal->j_running_transaction; | |
182 | int needed; | |
183 | int total = blocks + rsv_blocks; | |
184 | ||
1eaa566d JK |
185 | jbd2_might_wait_for_commit(journal); |
186 | ||
8f7d89f3 JK |
187 | /* |
188 | * If the current transaction is locked down for commit, wait | |
189 | * for the lock to be released. | |
190 | */ | |
191 | if (t->t_state == T_LOCKED) { | |
192 | wait_transaction_locked(journal); | |
193 | return 1; | |
194 | } | |
195 | ||
196 | /* | |
197 | * If there is not enough space left in the log to write all | |
198 | * potential buffers requested by this operation, we need to | |
199 | * stall pending a log checkpoint to free some more log space. | |
200 | */ | |
201 | needed = atomic_add_return(total, &t->t_outstanding_credits); | |
202 | if (needed > journal->j_max_transaction_buffers) { | |
203 | /* | |
204 | * If the current transaction is already too large, | |
205 | * then start to commit it: we can then go back and | |
206 | * attach this handle to a new transaction. | |
207 | */ | |
208 | atomic_sub(total, &t->t_outstanding_credits); | |
6d3ec14d LC |
209 | |
210 | /* | |
211 | * Is the number of reserved credits in the current transaction too | |
212 | * big to fit this handle? Wait until reserved credits are freed. | |
213 | */ | |
214 | if (atomic_read(&journal->j_reserved_credits) + total > | |
215 | journal->j_max_transaction_buffers) { | |
216 | read_unlock(&journal->j_state_lock); | |
217 | wait_event(journal->j_wait_reserved, | |
218 | atomic_read(&journal->j_reserved_credits) + total <= | |
219 | journal->j_max_transaction_buffers); | |
220 | return 1; | |
221 | } | |
222 | ||
8f7d89f3 JK |
223 | wait_transaction_locked(journal); |
224 | return 1; | |
225 | } | |
226 | ||
227 | /* | |
228 | * The commit code assumes that it can get enough log space | |
229 | * without forcing a checkpoint. This is *critical* for | |
230 | * correctness: a checkpoint of a buffer which is also | |
231 | * associated with a committing transaction creates a deadlock, | |
232 | * so commit simply cannot force through checkpoints. | |
233 | * | |
234 | * We must therefore ensure the necessary space in the journal | |
235 | * *before* starting to dirty potentially checkpointed buffers | |
236 | * in the new transaction. | |
237 | */ | |
238 | if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) { | |
239 | atomic_sub(total, &t->t_outstanding_credits); | |
240 | read_unlock(&journal->j_state_lock); | |
241 | write_lock(&journal->j_state_lock); | |
242 | if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) | |
243 | __jbd2_log_wait_for_space(journal); | |
244 | write_unlock(&journal->j_state_lock); | |
245 | return 1; | |
246 | } | |
247 | ||
248 | /* No reservation? We are done... */ | |
249 | if (!rsv_blocks) | |
250 | return 0; | |
251 | ||
252 | needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits); | |
253 | /* We allow at most half of a transaction to be reserved */ | |
254 | if (needed > journal->j_max_transaction_buffers / 2) { | |
255 | sub_reserved_credits(journal, rsv_blocks); | |
256 | atomic_sub(total, &t->t_outstanding_credits); | |
257 | read_unlock(&journal->j_state_lock); | |
258 | wait_event(journal->j_wait_reserved, | |
259 | atomic_read(&journal->j_reserved_credits) + rsv_blocks | |
260 | <= journal->j_max_transaction_buffers / 2); | |
261 | return 1; | |
262 | } | |
263 | return 0; | |
264 | } | |
265 | ||
470decc6 DK |
266 | /* |
267 | * start_this_handle: Given a handle, deal with any locking or stalling | |
268 | * needed to make sure that there is enough journal space for the handle | |
269 | * to begin. Attach the handle to a transaction and set up the | |
270 | * transaction's buffer credits. | |
271 | */ | |
272 | ||
47def826 | 273 | static int start_this_handle(journal_t *journal, handle_t *handle, |
d2159fb7 | 274 | gfp_t gfp_mask) |
470decc6 | 275 | { |
e4471831 | 276 | transaction_t *transaction, *new_transaction = NULL; |
8f7d89f3 JK |
277 | int blocks = handle->h_buffer_credits; |
278 | int rsv_blocks = 0; | |
28e35e42 | 279 | unsigned long ts = jiffies; |
470decc6 | 280 | |
6d3ec14d LC |
281 | if (handle->h_rsv_handle) |
282 | rsv_blocks = handle->h_rsv_handle->h_buffer_credits; | |
283 | ||
8f7d89f3 | 284 | /* |
6d3ec14d LC |
285 | * Limit the number of reserved credits to 1/2 of maximum transaction |
286 | * size and limit the number of total credits to not exceed maximum | |
287 | * transaction size per operation. | |
8f7d89f3 | 288 | */ |
6d3ec14d LC |
289 | if ((rsv_blocks > journal->j_max_transaction_buffers / 2) || |
290 | (rsv_blocks + blocks > journal->j_max_transaction_buffers)) { | |
291 | printk(KERN_ERR "JBD2: %s wants too many credits " | |
292 | "credits:%d rsv_credits:%d max:%d\n", | |
293 | current->comm, blocks, rsv_blocks, | |
294 | journal->j_max_transaction_buffers); | |
295 | WARN_ON(1); | |
47def826 | 296 | return -ENOSPC; |
470decc6 DK |
297 | } |
298 | ||
299 | alloc_transaction: | |
300 | if (!journal->j_running_transaction) { | |
6ccaf3e2 MH |
301 | /* |
302 | * If __GFP_FS is not present, then we may be being called from | |
303 | * inside the fs writeback layer, so we MUST NOT fail. | |
304 | */ | |
305 | if ((gfp_mask & __GFP_FS) == 0) | |
306 | gfp_mask |= __GFP_NOFAIL; | |
b2f4edb3 WG |
307 | new_transaction = kmem_cache_zalloc(transaction_cache, |
308 | gfp_mask); | |
6ccaf3e2 | 309 | if (!new_transaction) |
47def826 | 310 | return -ENOMEM; |
470decc6 DK |
311 | } |
312 | ||
313 | jbd_debug(3, "New handle %p going live.\n", handle); | |
314 | ||
470decc6 DK |
315 | /* |
316 | * We need to hold j_state_lock until t_updates has been incremented, | |
317 | * for proper journal barrier handling | |
318 | */ | |
a931da6a TT |
319 | repeat: |
320 | read_lock(&journal->j_state_lock); | |
5c2178e7 | 321 | BUG_ON(journal->j_flags & JBD2_UNMOUNT); |
470decc6 | 322 | if (is_journal_aborted(journal) || |
f7f4bccb | 323 | (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) { |
a931da6a | 324 | read_unlock(&journal->j_state_lock); |
0c2022ec | 325 | jbd2_journal_free_transaction(new_transaction); |
47def826 | 326 | return -EROFS; |
470decc6 DK |
327 | } |
328 | ||
8f7d89f3 JK |
329 | /* |
330 | * Wait on the journal's transaction barrier if necessary. Specifically | |
331 | * we allow reserved handles to proceed because otherwise commit could | |
332 | * deadlock on page writeback not being able to complete. | |
333 | */ | |
334 | if (!handle->h_reserved && journal->j_barrier_count) { | |
a931da6a | 335 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
336 | wait_event(journal->j_wait_transaction_locked, |
337 | journal->j_barrier_count == 0); | |
338 | goto repeat; | |
339 | } | |
340 | ||
341 | if (!journal->j_running_transaction) { | |
a931da6a TT |
342 | read_unlock(&journal->j_state_lock); |
343 | if (!new_transaction) | |
470decc6 | 344 | goto alloc_transaction; |
a931da6a | 345 | write_lock(&journal->j_state_lock); |
d7961c7f | 346 | if (!journal->j_running_transaction && |
8f7d89f3 | 347 | (handle->h_reserved || !journal->j_barrier_count)) { |
a931da6a TT |
348 | jbd2_get_transaction(journal, new_transaction); |
349 | new_transaction = NULL; | |
470decc6 | 350 | } |
a931da6a TT |
351 | write_unlock(&journal->j_state_lock); |
352 | goto repeat; | |
470decc6 DK |
353 | } |
354 | ||
355 | transaction = journal->j_running_transaction; | |
356 | ||
8f7d89f3 JK |
357 | if (!handle->h_reserved) { |
358 | /* We may have dropped j_state_lock - restart in that case */ | |
359 | if (add_transaction_credits(journal, blocks, rsv_blocks)) | |
360 | goto repeat; | |
361 | } else { | |
470decc6 | 362 | /* |
8f7d89f3 JK |
363 | * We have handle reserved so we are allowed to join T_LOCKED |
364 | * transaction and we don't have to check for transaction size | |
365 | * and journal space. | |
470decc6 | 366 | */ |
8f7d89f3 JK |
367 | sub_reserved_credits(journal, blocks); |
368 | handle->h_reserved = 0; | |
470decc6 DK |
369 | } |
370 | ||
371 | /* OK, account for the buffers that this operation expects to | |
8dd42046 | 372 | * use and add the handle to the running transaction. |
8dd42046 | 373 | */ |
28e35e42 | 374 | update_t_max_wait(transaction, ts); |
470decc6 | 375 | handle->h_transaction = transaction; |
8f7d89f3 | 376 | handle->h_requested_credits = blocks; |
343d9c28 | 377 | handle->h_start_jiffies = jiffies; |
a51dca9c | 378 | atomic_inc(&transaction->t_updates); |
8dd42046 | 379 | atomic_inc(&transaction->t_handle_count); |
8f7d89f3 JK |
380 | jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n", |
381 | handle, blocks, | |
a51dca9c | 382 | atomic_read(&transaction->t_outstanding_credits), |
76c39904 | 383 | jbd2_log_space_left(journal)); |
a931da6a | 384 | read_unlock(&journal->j_state_lock); |
41a5b913 | 385 | current->journal_info = handle; |
9599b0e5 | 386 | |
ab714aff | 387 | rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_); |
0c2022ec | 388 | jbd2_journal_free_transaction(new_transaction); |
47def826 | 389 | return 0; |
470decc6 DK |
390 | } |
391 | ||
392 | /* Allocate a new handle. This should probably be in a slab... */ | |
393 | static handle_t *new_handle(int nblocks) | |
394 | { | |
af1e76d6 | 395 | handle_t *handle = jbd2_alloc_handle(GFP_NOFS); |
470decc6 DK |
396 | if (!handle) |
397 | return NULL; | |
470decc6 DK |
398 | handle->h_buffer_credits = nblocks; |
399 | handle->h_ref = 1; | |
400 | ||
401 | return handle; | |
402 | } | |
403 | ||
404 | /** | |
f7f4bccb | 405 | * handle_t *jbd2_journal_start() - Obtain a new handle. |
470decc6 DK |
406 | * @journal: Journal to start transaction on. |
407 | * @nblocks: number of block buffer we might modify | |
408 | * | |
409 | * We make sure that the transaction can guarantee at least nblocks of | |
410 | * modified buffers in the log. We block until the log can guarantee | |
8f7d89f3 JK |
411 | * that much space. Additionally, if rsv_blocks > 0, we also create another |
412 | * handle with rsv_blocks reserved blocks in the journal. This handle is | |
413 | * is stored in h_rsv_handle. It is not attached to any particular transaction | |
414 | * and thus doesn't block transaction commit. If the caller uses this reserved | |
415 | * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop() | |
416 | * on the parent handle will dispose the reserved one. Reserved handle has to | |
417 | * be converted to a normal handle using jbd2_journal_start_reserved() before | |
418 | * it can be used. | |
470decc6 | 419 | * |
c867516d EG |
420 | * Return a pointer to a newly allocated handle, or an ERR_PTR() value |
421 | * on failure. | |
470decc6 | 422 | */ |
8f7d89f3 JK |
423 | handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks, |
424 | gfp_t gfp_mask, unsigned int type, | |
425 | unsigned int line_no) | |
470decc6 DK |
426 | { |
427 | handle_t *handle = journal_current_handle(); | |
428 | int err; | |
429 | ||
430 | if (!journal) | |
431 | return ERR_PTR(-EROFS); | |
432 | ||
433 | if (handle) { | |
434 | J_ASSERT(handle->h_transaction->t_journal == journal); | |
435 | handle->h_ref++; | |
436 | return handle; | |
437 | } | |
438 | ||
439 | handle = new_handle(nblocks); | |
440 | if (!handle) | |
441 | return ERR_PTR(-ENOMEM); | |
8f7d89f3 JK |
442 | if (rsv_blocks) { |
443 | handle_t *rsv_handle; | |
444 | ||
445 | rsv_handle = new_handle(rsv_blocks); | |
446 | if (!rsv_handle) { | |
447 | jbd2_free_handle(handle); | |
448 | return ERR_PTR(-ENOMEM); | |
449 | } | |
450 | rsv_handle->h_reserved = 1; | |
451 | rsv_handle->h_journal = journal; | |
452 | handle->h_rsv_handle = rsv_handle; | |
453 | } | |
470decc6 | 454 | |
47def826 | 455 | err = start_this_handle(journal, handle, gfp_mask); |
470decc6 | 456 | if (err < 0) { |
8f7d89f3 JK |
457 | if (handle->h_rsv_handle) |
458 | jbd2_free_handle(handle->h_rsv_handle); | |
af1e76d6 | 459 | jbd2_free_handle(handle); |
df05c1b8 | 460 | return ERR_PTR(err); |
470decc6 | 461 | } |
343d9c28 TT |
462 | handle->h_type = type; |
463 | handle->h_line_no = line_no; | |
464 | trace_jbd2_handle_start(journal->j_fs_dev->bd_dev, | |
465 | handle->h_transaction->t_tid, type, | |
466 | line_no, nblocks); | |
470decc6 DK |
467 | return handle; |
468 | } | |
47def826 TT |
469 | EXPORT_SYMBOL(jbd2__journal_start); |
470 | ||
471 | ||
472 | handle_t *jbd2_journal_start(journal_t *journal, int nblocks) | |
473 | { | |
8f7d89f3 | 474 | return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0); |
47def826 TT |
475 | } |
476 | EXPORT_SYMBOL(jbd2_journal_start); | |
477 | ||
8f7d89f3 JK |
478 | void jbd2_journal_free_reserved(handle_t *handle) |
479 | { | |
480 | journal_t *journal = handle->h_journal; | |
481 | ||
482 | WARN_ON(!handle->h_reserved); | |
483 | sub_reserved_credits(journal, handle->h_buffer_credits); | |
484 | jbd2_free_handle(handle); | |
485 | } | |
486 | EXPORT_SYMBOL(jbd2_journal_free_reserved); | |
487 | ||
488 | /** | |
489 | * int jbd2_journal_start_reserved(handle_t *handle) - start reserved handle | |
490 | * @handle: handle to start | |
491 | * | |
492 | * Start handle that has been previously reserved with jbd2_journal_reserve(). | |
493 | * This attaches @handle to the running transaction (or creates one if there's | |
494 | * not transaction running). Unlike jbd2_journal_start() this function cannot | |
495 | * block on journal commit, checkpointing, or similar stuff. It can block on | |
496 | * memory allocation or frozen journal though. | |
497 | * | |
498 | * Return 0 on success, non-zero on error - handle is freed in that case. | |
499 | */ | |
500 | int jbd2_journal_start_reserved(handle_t *handle, unsigned int type, | |
501 | unsigned int line_no) | |
502 | { | |
503 | journal_t *journal = handle->h_journal; | |
504 | int ret = -EIO; | |
505 | ||
506 | if (WARN_ON(!handle->h_reserved)) { | |
507 | /* Someone passed in normal handle? Just stop it. */ | |
508 | jbd2_journal_stop(handle); | |
509 | return ret; | |
510 | } | |
511 | /* | |
512 | * Usefulness of mixing of reserved and unreserved handles is | |
513 | * questionable. So far nobody seems to need it so just error out. | |
514 | */ | |
515 | if (WARN_ON(current->journal_info)) { | |
516 | jbd2_journal_free_reserved(handle); | |
517 | return ret; | |
518 | } | |
519 | ||
520 | handle->h_journal = NULL; | |
8f7d89f3 JK |
521 | /* |
522 | * GFP_NOFS is here because callers are likely from writeback or | |
523 | * similarly constrained call sites | |
524 | */ | |
525 | ret = start_this_handle(journal, handle, GFP_NOFS); | |
92e3b405 | 526 | if (ret < 0) { |
8f7d89f3 | 527 | jbd2_journal_free_reserved(handle); |
92e3b405 DC |
528 | return ret; |
529 | } | |
8f7d89f3 JK |
530 | handle->h_type = type; |
531 | handle->h_line_no = line_no; | |
92e3b405 | 532 | return 0; |
8f7d89f3 JK |
533 | } |
534 | EXPORT_SYMBOL(jbd2_journal_start_reserved); | |
470decc6 DK |
535 | |
536 | /** | |
f7f4bccb | 537 | * int jbd2_journal_extend() - extend buffer credits. |
470decc6 DK |
538 | * @handle: handle to 'extend' |
539 | * @nblocks: nr blocks to try to extend by. | |
540 | * | |
541 | * Some transactions, such as large extends and truncates, can be done | |
542 | * atomically all at once or in several stages. The operation requests | |
bd7ced98 | 543 | * a credit for a number of buffer modifications in advance, but can |
470decc6 DK |
544 | * extend its credit if it needs more. |
545 | * | |
f7f4bccb | 546 | * jbd2_journal_extend tries to give the running handle more buffer credits. |
470decc6 DK |
547 | * It does not guarantee that allocation - this is a best-effort only. |
548 | * The calling process MUST be able to deal cleanly with a failure to | |
549 | * extend here. | |
550 | * | |
551 | * Return 0 on success, non-zero on failure. | |
552 | * | |
553 | * return code < 0 implies an error | |
554 | * return code > 0 implies normal transaction-full status. | |
555 | */ | |
f7f4bccb | 556 | int jbd2_journal_extend(handle_t *handle, int nblocks) |
470decc6 DK |
557 | { |
558 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 559 | journal_t *journal; |
470decc6 DK |
560 | int result; |
561 | int wanted; | |
562 | ||
470decc6 | 563 | if (is_handle_aborted(handle)) |
41a5b913 TT |
564 | return -EROFS; |
565 | journal = transaction->t_journal; | |
470decc6 DK |
566 | |
567 | result = 1; | |
568 | ||
a931da6a | 569 | read_lock(&journal->j_state_lock); |
470decc6 DK |
570 | |
571 | /* Don't extend a locked-down transaction! */ | |
41a5b913 | 572 | if (transaction->t_state != T_RUNNING) { |
470decc6 DK |
573 | jbd_debug(3, "denied handle %p %d blocks: " |
574 | "transaction not running\n", handle, nblocks); | |
575 | goto error_out; | |
576 | } | |
577 | ||
578 | spin_lock(&transaction->t_handle_lock); | |
fe1e8db5 JK |
579 | wanted = atomic_add_return(nblocks, |
580 | &transaction->t_outstanding_credits); | |
470decc6 DK |
581 | |
582 | if (wanted > journal->j_max_transaction_buffers) { | |
583 | jbd_debug(3, "denied handle %p %d blocks: " | |
584 | "transaction too large\n", handle, nblocks); | |
fe1e8db5 | 585 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
470decc6 DK |
586 | goto unlock; |
587 | } | |
588 | ||
76c39904 JK |
589 | if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) > |
590 | jbd2_log_space_left(journal)) { | |
470decc6 DK |
591 | jbd_debug(3, "denied handle %p %d blocks: " |
592 | "insufficient log space\n", handle, nblocks); | |
fe1e8db5 | 593 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
470decc6 DK |
594 | goto unlock; |
595 | } | |
596 | ||
343d9c28 | 597 | trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev, |
41a5b913 | 598 | transaction->t_tid, |
343d9c28 TT |
599 | handle->h_type, handle->h_line_no, |
600 | handle->h_buffer_credits, | |
601 | nblocks); | |
602 | ||
470decc6 | 603 | handle->h_buffer_credits += nblocks; |
343d9c28 | 604 | handle->h_requested_credits += nblocks; |
470decc6 DK |
605 | result = 0; |
606 | ||
607 | jbd_debug(3, "extended handle %p by %d\n", handle, nblocks); | |
608 | unlock: | |
609 | spin_unlock(&transaction->t_handle_lock); | |
610 | error_out: | |
a931da6a | 611 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
612 | return result; |
613 | } | |
614 | ||
615 | ||
616 | /** | |
f7f4bccb | 617 | * int jbd2_journal_restart() - restart a handle . |
470decc6 DK |
618 | * @handle: handle to restart |
619 | * @nblocks: nr credits requested | |
620 | * | |
621 | * Restart a handle for a multi-transaction filesystem | |
622 | * operation. | |
623 | * | |
f7f4bccb MC |
624 | * If the jbd2_journal_extend() call above fails to grant new buffer credits |
625 | * to a running handle, a call to jbd2_journal_restart will commit the | |
470decc6 | 626 | * handle's transaction so far and reattach the handle to a new |
bd7ced98 | 627 | * transaction capable of guaranteeing the requested number of |
8f7d89f3 JK |
628 | * credits. We preserve reserved handle if there's any attached to the |
629 | * passed in handle. | |
470decc6 | 630 | */ |
d2159fb7 | 631 | int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask) |
470decc6 DK |
632 | { |
633 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 634 | journal_t *journal; |
e4471831 TT |
635 | tid_t tid; |
636 | int need_to_start, ret; | |
470decc6 DK |
637 | |
638 | /* If we've had an abort of any type, don't even think about | |
639 | * actually doing the restart! */ | |
640 | if (is_handle_aborted(handle)) | |
641 | return 0; | |
41a5b913 | 642 | journal = transaction->t_journal; |
470decc6 DK |
643 | |
644 | /* | |
645 | * First unlink the handle from its current transaction, and start the | |
646 | * commit on that. | |
647 | */ | |
a51dca9c | 648 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 DK |
649 | J_ASSERT(journal_current_handle() == handle); |
650 | ||
a931da6a | 651 | read_lock(&journal->j_state_lock); |
470decc6 | 652 | spin_lock(&transaction->t_handle_lock); |
a51dca9c TT |
653 | atomic_sub(handle->h_buffer_credits, |
654 | &transaction->t_outstanding_credits); | |
8f7d89f3 JK |
655 | if (handle->h_rsv_handle) { |
656 | sub_reserved_credits(journal, | |
657 | handle->h_rsv_handle->h_buffer_credits); | |
658 | } | |
a51dca9c | 659 | if (atomic_dec_and_test(&transaction->t_updates)) |
470decc6 | 660 | wake_up(&journal->j_wait_updates); |
39c04153 | 661 | tid = transaction->t_tid; |
470decc6 | 662 | spin_unlock(&transaction->t_handle_lock); |
41a5b913 TT |
663 | handle->h_transaction = NULL; |
664 | current->journal_info = NULL; | |
470decc6 DK |
665 | |
666 | jbd_debug(2, "restarting handle %p\n", handle); | |
e4471831 | 667 | need_to_start = !tid_geq(journal->j_commit_request, tid); |
a931da6a | 668 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
669 | if (need_to_start) |
670 | jbd2_log_start_commit(journal, tid); | |
470decc6 | 671 | |
ab714aff | 672 | rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_); |
470decc6 | 673 | handle->h_buffer_credits = nblocks; |
47def826 | 674 | ret = start_this_handle(journal, handle, gfp_mask); |
470decc6 DK |
675 | return ret; |
676 | } | |
47def826 | 677 | EXPORT_SYMBOL(jbd2__journal_restart); |
470decc6 DK |
678 | |
679 | ||
47def826 TT |
680 | int jbd2_journal_restart(handle_t *handle, int nblocks) |
681 | { | |
682 | return jbd2__journal_restart(handle, nblocks, GFP_NOFS); | |
683 | } | |
684 | EXPORT_SYMBOL(jbd2_journal_restart); | |
685 | ||
470decc6 | 686 | /** |
f7f4bccb | 687 | * void jbd2_journal_lock_updates () - establish a transaction barrier. |
470decc6 DK |
688 | * @journal: Journal to establish a barrier on. |
689 | * | |
690 | * This locks out any further updates from being started, and blocks | |
691 | * until all existing updates have completed, returning only once the | |
692 | * journal is in a quiescent state with no updates running. | |
693 | * | |
694 | * The journal lock should not be held on entry. | |
695 | */ | |
f7f4bccb | 696 | void jbd2_journal_lock_updates(journal_t *journal) |
470decc6 DK |
697 | { |
698 | DEFINE_WAIT(wait); | |
699 | ||
1eaa566d JK |
700 | jbd2_might_wait_for_commit(journal); |
701 | ||
a931da6a | 702 | write_lock(&journal->j_state_lock); |
470decc6 DK |
703 | ++journal->j_barrier_count; |
704 | ||
8f7d89f3 JK |
705 | /* Wait until there are no reserved handles */ |
706 | if (atomic_read(&journal->j_reserved_credits)) { | |
707 | write_unlock(&journal->j_state_lock); | |
708 | wait_event(journal->j_wait_reserved, | |
709 | atomic_read(&journal->j_reserved_credits) == 0); | |
710 | write_lock(&journal->j_state_lock); | |
711 | } | |
712 | ||
470decc6 DK |
713 | /* Wait until there are no running updates */ |
714 | while (1) { | |
715 | transaction_t *transaction = journal->j_running_transaction; | |
716 | ||
717 | if (!transaction) | |
718 | break; | |
719 | ||
720 | spin_lock(&transaction->t_handle_lock); | |
9837d8e9 JK |
721 | prepare_to_wait(&journal->j_wait_updates, &wait, |
722 | TASK_UNINTERRUPTIBLE); | |
a51dca9c | 723 | if (!atomic_read(&transaction->t_updates)) { |
470decc6 | 724 | spin_unlock(&transaction->t_handle_lock); |
9837d8e9 | 725 | finish_wait(&journal->j_wait_updates, &wait); |
470decc6 DK |
726 | break; |
727 | } | |
470decc6 | 728 | spin_unlock(&transaction->t_handle_lock); |
a931da6a | 729 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
730 | schedule(); |
731 | finish_wait(&journal->j_wait_updates, &wait); | |
a931da6a | 732 | write_lock(&journal->j_state_lock); |
470decc6 | 733 | } |
a931da6a | 734 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
735 | |
736 | /* | |
737 | * We have now established a barrier against other normal updates, but | |
f7f4bccb | 738 | * we also need to barrier against other jbd2_journal_lock_updates() calls |
470decc6 DK |
739 | * to make sure that we serialise special journal-locked operations |
740 | * too. | |
741 | */ | |
742 | mutex_lock(&journal->j_barrier); | |
743 | } | |
744 | ||
745 | /** | |
f7f4bccb | 746 | * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier |
470decc6 DK |
747 | * @journal: Journal to release the barrier on. |
748 | * | |
f7f4bccb | 749 | * Release a transaction barrier obtained with jbd2_journal_lock_updates(). |
470decc6 DK |
750 | * |
751 | * Should be called without the journal lock held. | |
752 | */ | |
f7f4bccb | 753 | void jbd2_journal_unlock_updates (journal_t *journal) |
470decc6 DK |
754 | { |
755 | J_ASSERT(journal->j_barrier_count != 0); | |
756 | ||
757 | mutex_unlock(&journal->j_barrier); | |
a931da6a | 758 | write_lock(&journal->j_state_lock); |
470decc6 | 759 | --journal->j_barrier_count; |
a931da6a | 760 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
761 | wake_up(&journal->j_wait_transaction_locked); |
762 | } | |
763 | ||
f91d1d04 | 764 | static void warn_dirty_buffer(struct buffer_head *bh) |
470decc6 | 765 | { |
f91d1d04 | 766 | printk(KERN_WARNING |
a1c6f057 | 767 | "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). " |
f91d1d04 JK |
768 | "There's a risk of filesystem corruption in case of system " |
769 | "crash.\n", | |
a1c6f057 | 770 | bh->b_bdev, (unsigned long long)bh->b_blocknr); |
470decc6 DK |
771 | } |
772 | ||
ee57aba1 JK |
773 | /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */ |
774 | static void jbd2_freeze_jh_data(struct journal_head *jh) | |
775 | { | |
776 | struct page *page; | |
777 | int offset; | |
778 | char *source; | |
779 | struct buffer_head *bh = jh2bh(jh); | |
780 | ||
781 | J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n"); | |
782 | page = bh->b_page; | |
783 | offset = offset_in_page(bh->b_data); | |
784 | source = kmap_atomic(page); | |
785 | /* Fire data frozen trigger just before we copy the data */ | |
786 | jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers); | |
787 | memcpy(jh->b_frozen_data, source + offset, bh->b_size); | |
788 | kunmap_atomic(source); | |
789 | ||
790 | /* | |
791 | * Now that the frozen data is saved off, we need to store any matching | |
792 | * triggers. | |
793 | */ | |
794 | jh->b_frozen_triggers = jh->b_triggers; | |
795 | } | |
796 | ||
470decc6 DK |
797 | /* |
798 | * If the buffer is already part of the current transaction, then there | |
799 | * is nothing we need to do. If it is already part of a prior | |
800 | * transaction which we are still committing to disk, then we need to | |
801 | * make sure that we do not overwrite the old copy: we do copy-out to | |
802 | * preserve the copy going to disk. We also account the buffer against | |
803 | * the handle's metadata buffer credits (unless the buffer is already | |
804 | * part of the transaction, that is). | |
805 | * | |
806 | */ | |
807 | static int | |
808 | do_get_write_access(handle_t *handle, struct journal_head *jh, | |
809 | int force_copy) | |
810 | { | |
811 | struct buffer_head *bh; | |
41a5b913 | 812 | transaction_t *transaction = handle->h_transaction; |
470decc6 DK |
813 | journal_t *journal; |
814 | int error; | |
815 | char *frozen_buffer = NULL; | |
f783f091 | 816 | unsigned long start_lock, time_lock; |
470decc6 DK |
817 | |
818 | if (is_handle_aborted(handle)) | |
819 | return -EROFS; | |
470decc6 DK |
820 | journal = transaction->t_journal; |
821 | ||
cfef2c6a | 822 | jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy); |
470decc6 DK |
823 | |
824 | JBUFFER_TRACE(jh, "entry"); | |
825 | repeat: | |
826 | bh = jh2bh(jh); | |
827 | ||
828 | /* @@@ Need to check for errors here at some point. */ | |
829 | ||
f783f091 | 830 | start_lock = jiffies; |
470decc6 DK |
831 | lock_buffer(bh); |
832 | jbd_lock_bh_state(bh); | |
833 | ||
f783f091 TT |
834 | /* If it takes too long to lock the buffer, trace it */ |
835 | time_lock = jbd2_time_diff(start_lock, jiffies); | |
836 | if (time_lock > HZ/10) | |
837 | trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev, | |
838 | jiffies_to_msecs(time_lock)); | |
839 | ||
470decc6 DK |
840 | /* We now hold the buffer lock so it is safe to query the buffer |
841 | * state. Is the buffer dirty? | |
842 | * | |
843 | * If so, there are two possibilities. The buffer may be | |
844 | * non-journaled, and undergoing a quite legitimate writeback. | |
845 | * Otherwise, it is journaled, and we don't expect dirty buffers | |
846 | * in that state (the buffers should be marked JBD_Dirty | |
847 | * instead.) So either the IO is being done under our own | |
848 | * control and this is a bug, or it's a third party IO such as | |
849 | * dump(8) (which may leave the buffer scheduled for read --- | |
850 | * ie. locked but not dirty) or tune2fs (which may actually have | |
851 | * the buffer dirtied, ugh.) */ | |
852 | ||
853 | if (buffer_dirty(bh)) { | |
854 | /* | |
855 | * First question: is this buffer already part of the current | |
856 | * transaction or the existing committing transaction? | |
857 | */ | |
858 | if (jh->b_transaction) { | |
859 | J_ASSERT_JH(jh, | |
860 | jh->b_transaction == transaction || | |
861 | jh->b_transaction == | |
862 | journal->j_committing_transaction); | |
863 | if (jh->b_next_transaction) | |
864 | J_ASSERT_JH(jh, jh->b_next_transaction == | |
865 | transaction); | |
f91d1d04 | 866 | warn_dirty_buffer(bh); |
470decc6 DK |
867 | } |
868 | /* | |
869 | * In any case we need to clean the dirty flag and we must | |
870 | * do it under the buffer lock to be sure we don't race | |
871 | * with running write-out. | |
872 | */ | |
f91d1d04 JK |
873 | JBUFFER_TRACE(jh, "Journalling dirty buffer"); |
874 | clear_buffer_dirty(bh); | |
875 | set_buffer_jbddirty(bh); | |
470decc6 DK |
876 | } |
877 | ||
878 | unlock_buffer(bh); | |
879 | ||
880 | error = -EROFS; | |
881 | if (is_handle_aborted(handle)) { | |
882 | jbd_unlock_bh_state(bh); | |
883 | goto out; | |
884 | } | |
885 | error = 0; | |
886 | ||
887 | /* | |
888 | * The buffer is already part of this transaction if b_transaction or | |
889 | * b_next_transaction points to it | |
890 | */ | |
891 | if (jh->b_transaction == transaction || | |
892 | jh->b_next_transaction == transaction) | |
893 | goto done; | |
894 | ||
9fc7c63a JB |
895 | /* |
896 | * this is the first time this transaction is touching this buffer, | |
897 | * reset the modified flag | |
898 | */ | |
899 | jh->b_modified = 0; | |
900 | ||
8b00f400 JK |
901 | /* |
902 | * If the buffer is not journaled right now, we need to make sure it | |
903 | * doesn't get written to disk before the caller actually commits the | |
904 | * new data | |
905 | */ | |
906 | if (!jh->b_transaction) { | |
907 | JBUFFER_TRACE(jh, "no transaction"); | |
908 | J_ASSERT_JH(jh, !jh->b_next_transaction); | |
909 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); | |
de92c8ca JK |
910 | /* |
911 | * Make sure all stores to jh (b_modified, b_frozen_data) are | |
912 | * visible before attaching it to the running transaction. | |
913 | * Paired with barrier in jbd2_write_access_granted() | |
914 | */ | |
915 | smp_wmb(); | |
8b00f400 JK |
916 | spin_lock(&journal->j_list_lock); |
917 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); | |
918 | spin_unlock(&journal->j_list_lock); | |
919 | goto done; | |
920 | } | |
470decc6 DK |
921 | /* |
922 | * If there is already a copy-out version of this buffer, then we don't | |
923 | * need to make another one | |
924 | */ | |
925 | if (jh->b_frozen_data) { | |
926 | JBUFFER_TRACE(jh, "has frozen data"); | |
927 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
de92c8ca | 928 | goto attach_next; |
470decc6 DK |
929 | } |
930 | ||
8b00f400 JK |
931 | JBUFFER_TRACE(jh, "owned by older transaction"); |
932 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
933 | J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction); | |
470decc6 | 934 | |
8b00f400 JK |
935 | /* |
936 | * There is one case we have to be very careful about. If the | |
937 | * committing transaction is currently writing this buffer out to disk | |
938 | * and has NOT made a copy-out, then we cannot modify the buffer | |
939 | * contents at all right now. The essence of copy-out is that it is | |
940 | * the extra copy, not the primary copy, which gets journaled. If the | |
941 | * primary copy is already going to disk then we cannot do copy-out | |
942 | * here. | |
943 | */ | |
944 | if (buffer_shadow(bh)) { | |
945 | JBUFFER_TRACE(jh, "on shadow: sleep"); | |
946 | jbd_unlock_bh_state(bh); | |
947 | wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE); | |
948 | goto repeat; | |
949 | } | |
470decc6 | 950 | |
8b00f400 JK |
951 | /* |
952 | * Only do the copy if the currently-owning transaction still needs it. | |
953 | * If buffer isn't on BJ_Metadata list, the committing transaction is | |
954 | * past that stage (here we use the fact that BH_Shadow is set under | |
955 | * bh_state lock together with refiling to BJ_Shadow list and at this | |
956 | * point we know the buffer doesn't have BH_Shadow set). | |
957 | * | |
958 | * Subtle point, though: if this is a get_undo_access, then we will be | |
959 | * relying on the frozen_data to contain the new value of the | |
960 | * committed_data record after the transaction, so we HAVE to force the | |
961 | * frozen_data copy in that case. | |
962 | */ | |
963 | if (jh->b_jlist == BJ_Metadata || force_copy) { | |
964 | JBUFFER_TRACE(jh, "generate frozen data"); | |
965 | if (!frozen_buffer) { | |
966 | JBUFFER_TRACE(jh, "allocate memory for buffer"); | |
470decc6 | 967 | jbd_unlock_bh_state(bh); |
490c1b44 MH |
968 | frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size, |
969 | GFP_NOFS | __GFP_NOFAIL); | |
8b00f400 | 970 | goto repeat; |
470decc6 | 971 | } |
8b00f400 JK |
972 | jh->b_frozen_data = frozen_buffer; |
973 | frozen_buffer = NULL; | |
974 | jbd2_freeze_jh_data(jh); | |
470decc6 | 975 | } |
de92c8ca JK |
976 | attach_next: |
977 | /* | |
978 | * Make sure all stores to jh (b_modified, b_frozen_data) are visible | |
979 | * before attaching it to the running transaction. Paired with barrier | |
980 | * in jbd2_write_access_granted() | |
981 | */ | |
982 | smp_wmb(); | |
8b00f400 | 983 | jh->b_next_transaction = transaction; |
470decc6 DK |
984 | |
985 | done: | |
470decc6 DK |
986 | jbd_unlock_bh_state(bh); |
987 | ||
988 | /* | |
989 | * If we are about to journal a buffer, then any revoke pending on it is | |
990 | * no longer valid | |
991 | */ | |
f7f4bccb | 992 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 DK |
993 | |
994 | out: | |
995 | if (unlikely(frozen_buffer)) /* It's usually NULL */ | |
af1e76d6 | 996 | jbd2_free(frozen_buffer, bh->b_size); |
470decc6 DK |
997 | |
998 | JBUFFER_TRACE(jh, "exit"); | |
999 | return error; | |
1000 | } | |
1001 | ||
de92c8ca | 1002 | /* Fast check whether buffer is already attached to the required transaction */ |
087ffd4e JB |
1003 | static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh, |
1004 | bool undo) | |
de92c8ca JK |
1005 | { |
1006 | struct journal_head *jh; | |
1007 | bool ret = false; | |
1008 | ||
1009 | /* Dirty buffers require special handling... */ | |
1010 | if (buffer_dirty(bh)) | |
1011 | return false; | |
1012 | ||
1013 | /* | |
1014 | * RCU protects us from dereferencing freed pages. So the checks we do | |
1015 | * are guaranteed not to oops. However the jh slab object can get freed | |
1016 | * & reallocated while we work with it. So we have to be careful. When | |
1017 | * we see jh attached to the running transaction, we know it must stay | |
1018 | * so until the transaction is committed. Thus jh won't be freed and | |
1019 | * will be attached to the same bh while we run. However it can | |
1020 | * happen jh gets freed, reallocated, and attached to the transaction | |
1021 | * just after we get pointer to it from bh. So we have to be careful | |
1022 | * and recheck jh still belongs to our bh before we return success. | |
1023 | */ | |
1024 | rcu_read_lock(); | |
1025 | if (!buffer_jbd(bh)) | |
1026 | goto out; | |
1027 | /* This should be bh2jh() but that doesn't work with inline functions */ | |
1028 | jh = READ_ONCE(bh->b_private); | |
1029 | if (!jh) | |
1030 | goto out; | |
087ffd4e JB |
1031 | /* For undo access buffer must have data copied */ |
1032 | if (undo && !jh->b_committed_data) | |
1033 | goto out; | |
de92c8ca JK |
1034 | if (jh->b_transaction != handle->h_transaction && |
1035 | jh->b_next_transaction != handle->h_transaction) | |
1036 | goto out; | |
1037 | /* | |
1038 | * There are two reasons for the barrier here: | |
1039 | * 1) Make sure to fetch b_bh after we did previous checks so that we | |
1040 | * detect when jh went through free, realloc, attach to transaction | |
1041 | * while we were checking. Paired with implicit barrier in that path. | |
1042 | * 2) So that access to bh done after jbd2_write_access_granted() | |
1043 | * doesn't get reordered and see inconsistent state of concurrent | |
1044 | * do_get_write_access(). | |
1045 | */ | |
1046 | smp_mb(); | |
1047 | if (unlikely(jh->b_bh != bh)) | |
1048 | goto out; | |
1049 | ret = true; | |
1050 | out: | |
1051 | rcu_read_unlock(); | |
1052 | return ret; | |
1053 | } | |
1054 | ||
470decc6 | 1055 | /** |
f7f4bccb | 1056 | * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update. |
470decc6 DK |
1057 | * @handle: transaction to add buffer modifications to |
1058 | * @bh: bh to be used for metadata writes | |
470decc6 DK |
1059 | * |
1060 | * Returns an error code or 0 on success. | |
1061 | * | |
1062 | * In full data journalling mode the buffer may be of type BJ_AsyncData, | |
1063 | * because we're write()ing a buffer which is also part of a shared mapping. | |
1064 | */ | |
1065 | ||
f7f4bccb | 1066 | int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh) |
470decc6 | 1067 | { |
de92c8ca | 1068 | struct journal_head *jh; |
470decc6 DK |
1069 | int rc; |
1070 | ||
087ffd4e | 1071 | if (jbd2_write_access_granted(handle, bh, false)) |
de92c8ca JK |
1072 | return 0; |
1073 | ||
1074 | jh = jbd2_journal_add_journal_head(bh); | |
470decc6 DK |
1075 | /* We do not want to get caught playing with fields which the |
1076 | * log thread also manipulates. Make sure that the buffer | |
1077 | * completes any outstanding IO before proceeding. */ | |
1078 | rc = do_get_write_access(handle, jh, 0); | |
f7f4bccb | 1079 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1080 | return rc; |
1081 | } | |
1082 | ||
1083 | ||
1084 | /* | |
1085 | * When the user wants to journal a newly created buffer_head | |
1086 | * (ie. getblk() returned a new buffer and we are going to populate it | |
1087 | * manually rather than reading off disk), then we need to keep the | |
1088 | * buffer_head locked until it has been completely filled with new | |
1089 | * data. In this case, we should be able to make the assertion that | |
1090 | * the bh is not already part of an existing transaction. | |
1091 | * | |
1092 | * The buffer should already be locked by the caller by this point. | |
1093 | * There is no lock ranking violation: it was a newly created, | |
1094 | * unlocked buffer beforehand. */ | |
1095 | ||
1096 | /** | |
f7f4bccb | 1097 | * int jbd2_journal_get_create_access () - notify intent to use newly created bh |
470decc6 DK |
1098 | * @handle: transaction to new buffer to |
1099 | * @bh: new buffer. | |
1100 | * | |
1101 | * Call this if you create a new bh. | |
1102 | */ | |
f7f4bccb | 1103 | int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1104 | { |
1105 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 1106 | journal_t *journal; |
f7f4bccb | 1107 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
1108 | int err; |
1109 | ||
1110 | jbd_debug(5, "journal_head %p\n", jh); | |
1111 | err = -EROFS; | |
1112 | if (is_handle_aborted(handle)) | |
1113 | goto out; | |
41a5b913 | 1114 | journal = transaction->t_journal; |
470decc6 DK |
1115 | err = 0; |
1116 | ||
1117 | JBUFFER_TRACE(jh, "entry"); | |
1118 | /* | |
1119 | * The buffer may already belong to this transaction due to pre-zeroing | |
1120 | * in the filesystem's new_block code. It may also be on the previous, | |
1121 | * committing transaction's lists, but it HAS to be in Forget state in | |
1122 | * that case: the transaction must have deleted the buffer for it to be | |
1123 | * reused here. | |
1124 | */ | |
1125 | jbd_lock_bh_state(bh); | |
470decc6 DK |
1126 | J_ASSERT_JH(jh, (jh->b_transaction == transaction || |
1127 | jh->b_transaction == NULL || | |
1128 | (jh->b_transaction == journal->j_committing_transaction && | |
1129 | jh->b_jlist == BJ_Forget))); | |
1130 | ||
1131 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
1132 | J_ASSERT_JH(jh, buffer_locked(jh2bh(jh))); | |
1133 | ||
1134 | if (jh->b_transaction == NULL) { | |
f91d1d04 JK |
1135 | /* |
1136 | * Previous jbd2_journal_forget() could have left the buffer | |
1137 | * with jbddirty bit set because it was being committed. When | |
1138 | * the commit finished, we've filed the buffer for | |
1139 | * checkpointing and marked it dirty. Now we are reallocating | |
1140 | * the buffer so the transaction freeing it must have | |
1141 | * committed and so it's safe to clear the dirty bit. | |
1142 | */ | |
1143 | clear_buffer_dirty(jh2bh(jh)); | |
9fc7c63a JB |
1144 | /* first access by this transaction */ |
1145 | jh->b_modified = 0; | |
1146 | ||
470decc6 | 1147 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); |
6e4862a5 | 1148 | spin_lock(&journal->j_list_lock); |
f7f4bccb | 1149 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
470decc6 | 1150 | } else if (jh->b_transaction == journal->j_committing_transaction) { |
9fc7c63a JB |
1151 | /* first access by this transaction */ |
1152 | jh->b_modified = 0; | |
1153 | ||
470decc6 | 1154 | JBUFFER_TRACE(jh, "set next transaction"); |
6e4862a5 | 1155 | spin_lock(&journal->j_list_lock); |
470decc6 DK |
1156 | jh->b_next_transaction = transaction; |
1157 | } | |
1158 | spin_unlock(&journal->j_list_lock); | |
1159 | jbd_unlock_bh_state(bh); | |
1160 | ||
1161 | /* | |
1162 | * akpm: I added this. ext3_alloc_branch can pick up new indirect | |
1163 | * blocks which contain freed but then revoked metadata. We need | |
1164 | * to cancel the revoke in case we end up freeing it yet again | |
1165 | * and the reallocating as data - this would cause a second revoke, | |
1166 | * which hits an assertion error. | |
1167 | */ | |
1168 | JBUFFER_TRACE(jh, "cancelling revoke"); | |
f7f4bccb | 1169 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 | 1170 | out: |
3991b400 | 1171 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1172 | return err; |
1173 | } | |
1174 | ||
1175 | /** | |
f7f4bccb | 1176 | * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with |
470decc6 DK |
1177 | * non-rewindable consequences |
1178 | * @handle: transaction | |
1179 | * @bh: buffer to undo | |
470decc6 DK |
1180 | * |
1181 | * Sometimes there is a need to distinguish between metadata which has | |
1182 | * been committed to disk and that which has not. The ext3fs code uses | |
1183 | * this for freeing and allocating space, we have to make sure that we | |
1184 | * do not reuse freed space until the deallocation has been committed, | |
1185 | * since if we overwrote that space we would make the delete | |
1186 | * un-rewindable in case of a crash. | |
1187 | * | |
f7f4bccb | 1188 | * To deal with that, jbd2_journal_get_undo_access requests write access to a |
470decc6 DK |
1189 | * buffer for parts of non-rewindable operations such as delete |
1190 | * operations on the bitmaps. The journaling code must keep a copy of | |
1191 | * the buffer's contents prior to the undo_access call until such time | |
1192 | * as we know that the buffer has definitely been committed to disk. | |
1193 | * | |
1194 | * We never need to know which transaction the committed data is part | |
1195 | * of, buffers touched here are guaranteed to be dirtied later and so | |
1196 | * will be committed to a new transaction in due course, at which point | |
1197 | * we can discard the old committed data pointer. | |
1198 | * | |
1199 | * Returns error number or 0 on success. | |
1200 | */ | |
f7f4bccb | 1201 | int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1202 | { |
1203 | int err; | |
de92c8ca | 1204 | struct journal_head *jh; |
470decc6 DK |
1205 | char *committed_data = NULL; |
1206 | ||
1207 | JBUFFER_TRACE(jh, "entry"); | |
087ffd4e | 1208 | if (jbd2_write_access_granted(handle, bh, true)) |
de92c8ca | 1209 | return 0; |
470decc6 | 1210 | |
de92c8ca | 1211 | jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
1212 | /* |
1213 | * Do this first --- it can drop the journal lock, so we want to | |
1214 | * make sure that obtaining the committed_data is done | |
1215 | * atomically wrt. completion of any outstanding commits. | |
1216 | */ | |
1217 | err = do_get_write_access(handle, jh, 1); | |
1218 | if (err) | |
1219 | goto out; | |
1220 | ||
1221 | repeat: | |
490c1b44 MH |
1222 | if (!jh->b_committed_data) |
1223 | committed_data = jbd2_alloc(jh2bh(jh)->b_size, | |
1224 | GFP_NOFS|__GFP_NOFAIL); | |
470decc6 DK |
1225 | |
1226 | jbd_lock_bh_state(bh); | |
1227 | if (!jh->b_committed_data) { | |
1228 | /* Copy out the current buffer contents into the | |
1229 | * preserved, committed copy. */ | |
1230 | JBUFFER_TRACE(jh, "generate b_committed data"); | |
1231 | if (!committed_data) { | |
1232 | jbd_unlock_bh_state(bh); | |
1233 | goto repeat; | |
1234 | } | |
1235 | ||
1236 | jh->b_committed_data = committed_data; | |
1237 | committed_data = NULL; | |
1238 | memcpy(jh->b_committed_data, bh->b_data, bh->b_size); | |
1239 | } | |
1240 | jbd_unlock_bh_state(bh); | |
1241 | out: | |
f7f4bccb | 1242 | jbd2_journal_put_journal_head(jh); |
470decc6 | 1243 | if (unlikely(committed_data)) |
af1e76d6 | 1244 | jbd2_free(committed_data, bh->b_size); |
470decc6 DK |
1245 | return err; |
1246 | } | |
1247 | ||
e06c8227 JB |
1248 | /** |
1249 | * void jbd2_journal_set_triggers() - Add triggers for commit writeout | |
1250 | * @bh: buffer to trigger on | |
1251 | * @type: struct jbd2_buffer_trigger_type containing the trigger(s). | |
1252 | * | |
1253 | * Set any triggers on this journal_head. This is always safe, because | |
1254 | * triggers for a committing buffer will be saved off, and triggers for | |
1255 | * a running transaction will match the buffer in that transaction. | |
1256 | * | |
1257 | * Call with NULL to clear the triggers. | |
1258 | */ | |
1259 | void jbd2_journal_set_triggers(struct buffer_head *bh, | |
1260 | struct jbd2_buffer_trigger_type *type) | |
1261 | { | |
ad56edad | 1262 | struct journal_head *jh = jbd2_journal_grab_journal_head(bh); |
e06c8227 | 1263 | |
ad56edad JK |
1264 | if (WARN_ON(!jh)) |
1265 | return; | |
e06c8227 | 1266 | jh->b_triggers = type; |
ad56edad | 1267 | jbd2_journal_put_journal_head(jh); |
e06c8227 JB |
1268 | } |
1269 | ||
13ceef09 | 1270 | void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data, |
e06c8227 JB |
1271 | struct jbd2_buffer_trigger_type *triggers) |
1272 | { | |
1273 | struct buffer_head *bh = jh2bh(jh); | |
1274 | ||
13ceef09 | 1275 | if (!triggers || !triggers->t_frozen) |
e06c8227 JB |
1276 | return; |
1277 | ||
13ceef09 | 1278 | triggers->t_frozen(triggers, bh, mapped_data, bh->b_size); |
e06c8227 JB |
1279 | } |
1280 | ||
1281 | void jbd2_buffer_abort_trigger(struct journal_head *jh, | |
1282 | struct jbd2_buffer_trigger_type *triggers) | |
1283 | { | |
1284 | if (!triggers || !triggers->t_abort) | |
1285 | return; | |
1286 | ||
1287 | triggers->t_abort(triggers, jh2bh(jh)); | |
1288 | } | |
1289 | ||
470decc6 | 1290 | /** |
f7f4bccb | 1291 | * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata |
470decc6 DK |
1292 | * @handle: transaction to add buffer to. |
1293 | * @bh: buffer to mark | |
1294 | * | |
1295 | * mark dirty metadata which needs to be journaled as part of the current | |
1296 | * transaction. | |
1297 | * | |
9ea7a0df TT |
1298 | * The buffer must have previously had jbd2_journal_get_write_access() |
1299 | * called so that it has a valid journal_head attached to the buffer | |
1300 | * head. | |
1301 | * | |
470decc6 DK |
1302 | * The buffer is placed on the transaction's metadata list and is marked |
1303 | * as belonging to the transaction. | |
1304 | * | |
1305 | * Returns error number or 0 on success. | |
1306 | * | |
1307 | * Special care needs to be taken if the buffer already belongs to the | |
1308 | * current committing transaction (in which case we should have frozen | |
1309 | * data present for that commit). In that case, we don't relink the | |
1310 | * buffer: that only gets done when the old transaction finally | |
1311 | * completes its commit. | |
1312 | */ | |
f7f4bccb | 1313 | int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1314 | { |
1315 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 1316 | journal_t *journal; |
ad56edad | 1317 | struct journal_head *jh; |
9ea7a0df | 1318 | int ret = 0; |
470decc6 | 1319 | |
470decc6 | 1320 | if (is_handle_aborted(handle)) |
41a5b913 | 1321 | return -EROFS; |
6e06ae88 | 1322 | if (!buffer_jbd(bh)) { |
9ea7a0df TT |
1323 | ret = -EUCLEAN; |
1324 | goto out; | |
1325 | } | |
6e06ae88 JK |
1326 | /* |
1327 | * We don't grab jh reference here since the buffer must be part | |
1328 | * of the running transaction. | |
1329 | */ | |
1330 | jh = bh2jh(bh); | |
1331 | /* | |
1332 | * This and the following assertions are unreliable since we may see jh | |
1333 | * in inconsistent state unless we grab bh_state lock. But this is | |
1334 | * crucial to catch bugs so let's do a reliable check until the | |
1335 | * lockless handling is fully proven. | |
1336 | */ | |
1337 | if (jh->b_transaction != transaction && | |
1338 | jh->b_next_transaction != transaction) { | |
1339 | jbd_lock_bh_state(bh); | |
1340 | J_ASSERT_JH(jh, jh->b_transaction == transaction || | |
1341 | jh->b_next_transaction == transaction); | |
1342 | jbd_unlock_bh_state(bh); | |
1343 | } | |
1344 | if (jh->b_modified == 1) { | |
1345 | /* If it's in our transaction it must be in BJ_Metadata list. */ | |
1346 | if (jh->b_transaction == transaction && | |
1347 | jh->b_jlist != BJ_Metadata) { | |
1348 | jbd_lock_bh_state(bh); | |
1349 | J_ASSERT_JH(jh, jh->b_transaction != transaction || | |
1350 | jh->b_jlist == BJ_Metadata); | |
1351 | jbd_unlock_bh_state(bh); | |
1352 | } | |
1353 | goto out; | |
1354 | } | |
1355 | ||
1356 | journal = transaction->t_journal; | |
ad56edad JK |
1357 | jbd_debug(5, "journal_head %p\n", jh); |
1358 | JBUFFER_TRACE(jh, "entry"); | |
470decc6 DK |
1359 | |
1360 | jbd_lock_bh_state(bh); | |
1361 | ||
1362 | if (jh->b_modified == 0) { | |
1363 | /* | |
1364 | * This buffer's got modified and becoming part | |
1365 | * of the transaction. This needs to be done | |
1366 | * once a transaction -bzzz | |
1367 | */ | |
1368 | jh->b_modified = 1; | |
f6c07cad TT |
1369 | if (handle->h_buffer_credits <= 0) { |
1370 | ret = -ENOSPC; | |
1371 | goto out_unlock_bh; | |
1372 | } | |
470decc6 DK |
1373 | handle->h_buffer_credits--; |
1374 | } | |
1375 | ||
1376 | /* | |
1377 | * fastpath, to avoid expensive locking. If this buffer is already | |
1378 | * on the running transaction's metadata list there is nothing to do. | |
1379 | * Nobody can take it off again because there is a handle open. | |
1380 | * I _think_ we're OK here with SMP barriers - a mistaken decision will | |
1381 | * result in this test being false, so we go in and take the locks. | |
1382 | */ | |
1383 | if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) { | |
1384 | JBUFFER_TRACE(jh, "fastpath"); | |
9ea7a0df TT |
1385 | if (unlikely(jh->b_transaction != |
1386 | journal->j_running_transaction)) { | |
a67c848a | 1387 | printk(KERN_ERR "JBD2: %s: " |
9ea7a0df | 1388 | "jh->b_transaction (%llu, %p, %u) != " |
66a4cb18 | 1389 | "journal->j_running_transaction (%p, %u)\n", |
9ea7a0df TT |
1390 | journal->j_devname, |
1391 | (unsigned long long) bh->b_blocknr, | |
1392 | jh->b_transaction, | |
1393 | jh->b_transaction ? jh->b_transaction->t_tid : 0, | |
1394 | journal->j_running_transaction, | |
1395 | journal->j_running_transaction ? | |
1396 | journal->j_running_transaction->t_tid : 0); | |
1397 | ret = -EINVAL; | |
1398 | } | |
470decc6 DK |
1399 | goto out_unlock_bh; |
1400 | } | |
1401 | ||
1402 | set_buffer_jbddirty(bh); | |
1403 | ||
1404 | /* | |
1405 | * Metadata already on the current transaction list doesn't | |
1406 | * need to be filed. Metadata on another transaction's list must | |
1407 | * be committing, and will be refiled once the commit completes: | |
1408 | * leave it alone for now. | |
1409 | */ | |
1410 | if (jh->b_transaction != transaction) { | |
1411 | JBUFFER_TRACE(jh, "already on other transaction"); | |
66a4cb18 TT |
1412 | if (unlikely(((jh->b_transaction != |
1413 | journal->j_committing_transaction)) || | |
1414 | (jh->b_next_transaction != transaction))) { | |
1415 | printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: " | |
1416 | "bad jh for block %llu: " | |
1417 | "transaction (%p, %u), " | |
1418 | "jh->b_transaction (%p, %u), " | |
1419 | "jh->b_next_transaction (%p, %u), jlist %u\n", | |
9ea7a0df TT |
1420 | journal->j_devname, |
1421 | (unsigned long long) bh->b_blocknr, | |
66a4cb18 | 1422 | transaction, transaction->t_tid, |
9ea7a0df | 1423 | jh->b_transaction, |
66a4cb18 TT |
1424 | jh->b_transaction ? |
1425 | jh->b_transaction->t_tid : 0, | |
9ea7a0df TT |
1426 | jh->b_next_transaction, |
1427 | jh->b_next_transaction ? | |
1428 | jh->b_next_transaction->t_tid : 0, | |
66a4cb18 TT |
1429 | jh->b_jlist); |
1430 | WARN_ON(1); | |
9ea7a0df TT |
1431 | ret = -EINVAL; |
1432 | } | |
470decc6 DK |
1433 | /* And this case is illegal: we can't reuse another |
1434 | * transaction's data buffer, ever. */ | |
1435 | goto out_unlock_bh; | |
1436 | } | |
1437 | ||
1438 | /* That test should have eliminated the following case: */ | |
4019191b | 1439 | J_ASSERT_JH(jh, jh->b_frozen_data == NULL); |
470decc6 DK |
1440 | |
1441 | JBUFFER_TRACE(jh, "file as BJ_Metadata"); | |
1442 | spin_lock(&journal->j_list_lock); | |
41a5b913 | 1443 | __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata); |
470decc6 DK |
1444 | spin_unlock(&journal->j_list_lock); |
1445 | out_unlock_bh: | |
1446 | jbd_unlock_bh_state(bh); | |
1447 | out: | |
1448 | JBUFFER_TRACE(jh, "exit"); | |
9ea7a0df | 1449 | return ret; |
470decc6 DK |
1450 | } |
1451 | ||
470decc6 | 1452 | /** |
f7f4bccb | 1453 | * void jbd2_journal_forget() - bforget() for potentially-journaled buffers. |
470decc6 DK |
1454 | * @handle: transaction handle |
1455 | * @bh: bh to 'forget' | |
1456 | * | |
1457 | * We can only do the bforget if there are no commits pending against the | |
1458 | * buffer. If the buffer is dirty in the current running transaction we | |
1459 | * can safely unlink it. | |
1460 | * | |
1461 | * bh may not be a journalled buffer at all - it may be a non-JBD | |
1462 | * buffer which came off the hashtable. Check for this. | |
1463 | * | |
1464 | * Decrements bh->b_count by one. | |
1465 | * | |
1466 | * Allow this call even if the handle has aborted --- it may be part of | |
1467 | * the caller's cleanup after an abort. | |
1468 | */ | |
f7f4bccb | 1469 | int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1470 | { |
1471 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 1472 | journal_t *journal; |
470decc6 DK |
1473 | struct journal_head *jh; |
1474 | int drop_reserve = 0; | |
1475 | int err = 0; | |
1dfc3220 | 1476 | int was_modified = 0; |
470decc6 | 1477 | |
41a5b913 TT |
1478 | if (is_handle_aborted(handle)) |
1479 | return -EROFS; | |
1480 | journal = transaction->t_journal; | |
1481 | ||
470decc6 DK |
1482 | BUFFER_TRACE(bh, "entry"); |
1483 | ||
1484 | jbd_lock_bh_state(bh); | |
470decc6 DK |
1485 | |
1486 | if (!buffer_jbd(bh)) | |
1487 | goto not_jbd; | |
1488 | jh = bh2jh(bh); | |
1489 | ||
1490 | /* Critical error: attempting to delete a bitmap buffer, maybe? | |
1491 | * Don't do any jbd operations, and return an error. */ | |
1492 | if (!J_EXPECT_JH(jh, !jh->b_committed_data, | |
1493 | "inconsistent data on disk")) { | |
1494 | err = -EIO; | |
1495 | goto not_jbd; | |
1496 | } | |
1497 | ||
48fc7f7e | 1498 | /* keep track of whether or not this transaction modified us */ |
1dfc3220 JB |
1499 | was_modified = jh->b_modified; |
1500 | ||
470decc6 DK |
1501 | /* |
1502 | * The buffer's going from the transaction, we must drop | |
1503 | * all references -bzzz | |
1504 | */ | |
1505 | jh->b_modified = 0; | |
1506 | ||
41a5b913 | 1507 | if (jh->b_transaction == transaction) { |
470decc6 DK |
1508 | J_ASSERT_JH(jh, !jh->b_frozen_data); |
1509 | ||
1510 | /* If we are forgetting a buffer which is already part | |
1511 | * of this transaction, then we can just drop it from | |
1512 | * the transaction immediately. */ | |
1513 | clear_buffer_dirty(bh); | |
1514 | clear_buffer_jbddirty(bh); | |
1515 | ||
1516 | JBUFFER_TRACE(jh, "belongs to current transaction: unfile"); | |
1517 | ||
1dfc3220 JB |
1518 | /* |
1519 | * we only want to drop a reference if this transaction | |
1520 | * modified the buffer | |
1521 | */ | |
1522 | if (was_modified) | |
1523 | drop_reserve = 1; | |
470decc6 DK |
1524 | |
1525 | /* | |
1526 | * We are no longer going to journal this buffer. | |
1527 | * However, the commit of this transaction is still | |
1528 | * important to the buffer: the delete that we are now | |
1529 | * processing might obsolete an old log entry, so by | |
1530 | * committing, we can satisfy the buffer's checkpoint. | |
1531 | * | |
1532 | * So, if we have a checkpoint on the buffer, we should | |
1533 | * now refile the buffer on our BJ_Forget list so that | |
1534 | * we know to remove the checkpoint after we commit. | |
1535 | */ | |
1536 | ||
0bfea811 | 1537 | spin_lock(&journal->j_list_lock); |
470decc6 | 1538 | if (jh->b_cp_transaction) { |
f7f4bccb MC |
1539 | __jbd2_journal_temp_unlink_buffer(jh); |
1540 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); | |
470decc6 | 1541 | } else { |
f7f4bccb | 1542 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
1543 | if (!buffer_jbd(bh)) { |
1544 | spin_unlock(&journal->j_list_lock); | |
1545 | jbd_unlock_bh_state(bh); | |
1546 | __bforget(bh); | |
1547 | goto drop; | |
1548 | } | |
1549 | } | |
0bfea811 | 1550 | spin_unlock(&journal->j_list_lock); |
470decc6 DK |
1551 | } else if (jh->b_transaction) { |
1552 | J_ASSERT_JH(jh, (jh->b_transaction == | |
1553 | journal->j_committing_transaction)); | |
1554 | /* However, if the buffer is still owned by a prior | |
1555 | * (committing) transaction, we can't drop it yet... */ | |
1556 | JBUFFER_TRACE(jh, "belongs to older transaction"); | |
1557 | /* ... but we CAN drop it from the new transaction if we | |
1558 | * have also modified it since the original commit. */ | |
1559 | ||
1560 | if (jh->b_next_transaction) { | |
1561 | J_ASSERT(jh->b_next_transaction == transaction); | |
0bfea811 | 1562 | spin_lock(&journal->j_list_lock); |
470decc6 | 1563 | jh->b_next_transaction = NULL; |
0bfea811 | 1564 | spin_unlock(&journal->j_list_lock); |
1dfc3220 JB |
1565 | |
1566 | /* | |
1567 | * only drop a reference if this transaction modified | |
1568 | * the buffer | |
1569 | */ | |
1570 | if (was_modified) | |
1571 | drop_reserve = 1; | |
470decc6 DK |
1572 | } |
1573 | } | |
1574 | ||
1575 | not_jbd: | |
470decc6 DK |
1576 | jbd_unlock_bh_state(bh); |
1577 | __brelse(bh); | |
1578 | drop: | |
1579 | if (drop_reserve) { | |
1580 | /* no need to reserve log space for this block -bzzz */ | |
1581 | handle->h_buffer_credits++; | |
1582 | } | |
1583 | return err; | |
1584 | } | |
1585 | ||
1586 | /** | |
f7f4bccb | 1587 | * int jbd2_journal_stop() - complete a transaction |
bd7ced98 | 1588 | * @handle: transaction to complete. |
470decc6 DK |
1589 | * |
1590 | * All done for a particular handle. | |
1591 | * | |
1592 | * There is not much action needed here. We just return any remaining | |
1593 | * buffer credits to the transaction and remove the handle. The only | |
1594 | * complication is that we need to start a commit operation if the | |
1595 | * filesystem is marked for synchronous update. | |
1596 | * | |
f7f4bccb | 1597 | * jbd2_journal_stop itself will not usually return an error, but it may |
470decc6 | 1598 | * do so in unusual circumstances. In particular, expect it to |
f7f4bccb | 1599 | * return -EIO if a jbd2_journal_abort has been executed since the |
470decc6 DK |
1600 | * transaction began. |
1601 | */ | |
f7f4bccb | 1602 | int jbd2_journal_stop(handle_t *handle) |
470decc6 DK |
1603 | { |
1604 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 TT |
1605 | journal_t *journal; |
1606 | int err = 0, wait_for_commit = 0; | |
a51dca9c | 1607 | tid_t tid; |
470decc6 DK |
1608 | pid_t pid; |
1609 | ||
9d506594 LC |
1610 | if (!transaction) { |
1611 | /* | |
1612 | * Handle is already detached from the transaction so | |
1613 | * there is nothing to do other than decrease a refcount, | |
1614 | * or free the handle if refcount drops to zero | |
1615 | */ | |
1616 | if (--handle->h_ref > 0) { | |
1617 | jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1, | |
1618 | handle->h_ref); | |
1619 | return err; | |
1620 | } else { | |
1621 | if (handle->h_rsv_handle) | |
1622 | jbd2_free_handle(handle->h_rsv_handle); | |
1623 | goto free_and_exit; | |
1624 | } | |
1625 | } | |
41a5b913 TT |
1626 | journal = transaction->t_journal; |
1627 | ||
470decc6 DK |
1628 | J_ASSERT(journal_current_handle() == handle); |
1629 | ||
1630 | if (is_handle_aborted(handle)) | |
1631 | err = -EIO; | |
41a5b913 | 1632 | else |
a51dca9c | 1633 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 DK |
1634 | |
1635 | if (--handle->h_ref > 0) { | |
1636 | jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1, | |
1637 | handle->h_ref); | |
1638 | return err; | |
1639 | } | |
1640 | ||
1641 | jbd_debug(4, "Handle %p going down\n", handle); | |
343d9c28 | 1642 | trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev, |
41a5b913 | 1643 | transaction->t_tid, |
343d9c28 TT |
1644 | handle->h_type, handle->h_line_no, |
1645 | jiffies - handle->h_start_jiffies, | |
1646 | handle->h_sync, handle->h_requested_credits, | |
1647 | (handle->h_requested_credits - | |
1648 | handle->h_buffer_credits)); | |
470decc6 DK |
1649 | |
1650 | /* | |
1651 | * Implement synchronous transaction batching. If the handle | |
1652 | * was synchronous, don't force a commit immediately. Let's | |
e07f7183 JB |
1653 | * yield and let another thread piggyback onto this |
1654 | * transaction. Keep doing that while new threads continue to | |
1655 | * arrive. It doesn't cost much - we're about to run a commit | |
1656 | * and sleep on IO anyway. Speeds up many-threaded, many-dir | |
1657 | * operations by 30x or more... | |
1658 | * | |
1659 | * We try and optimize the sleep time against what the | |
1660 | * underlying disk can do, instead of having a static sleep | |
1661 | * time. This is useful for the case where our storage is so | |
1662 | * fast that it is more optimal to go ahead and force a flush | |
1663 | * and wait for the transaction to be committed than it is to | |
1664 | * wait for an arbitrary amount of time for new writers to | |
1665 | * join the transaction. We achieve this by measuring how | |
1666 | * long it takes to commit a transaction, and compare it with | |
1667 | * how long this transaction has been running, and if run time | |
1668 | * < commit time then we sleep for the delta and commit. This | |
1669 | * greatly helps super fast disks that would see slowdowns as | |
1670 | * more threads started doing fsyncs. | |
470decc6 | 1671 | * |
e07f7183 JB |
1672 | * But don't do this if this process was the most recent one |
1673 | * to perform a synchronous write. We do this to detect the | |
1674 | * case where a single process is doing a stream of sync | |
1675 | * writes. No point in waiting for joiners in that case. | |
5dd21424 ES |
1676 | * |
1677 | * Setting max_batch_time to 0 disables this completely. | |
470decc6 DK |
1678 | */ |
1679 | pid = current->pid; | |
5dd21424 ES |
1680 | if (handle->h_sync && journal->j_last_sync_writer != pid && |
1681 | journal->j_max_batch_time) { | |
e07f7183 JB |
1682 | u64 commit_time, trans_time; |
1683 | ||
470decc6 | 1684 | journal->j_last_sync_writer = pid; |
e07f7183 | 1685 | |
a931da6a | 1686 | read_lock(&journal->j_state_lock); |
e07f7183 | 1687 | commit_time = journal->j_average_commit_time; |
a931da6a | 1688 | read_unlock(&journal->j_state_lock); |
e07f7183 JB |
1689 | |
1690 | trans_time = ktime_to_ns(ktime_sub(ktime_get(), | |
1691 | transaction->t_start_time)); | |
1692 | ||
30773840 TT |
1693 | commit_time = max_t(u64, commit_time, |
1694 | 1000*journal->j_min_batch_time); | |
e07f7183 | 1695 | commit_time = min_t(u64, commit_time, |
30773840 | 1696 | 1000*journal->j_max_batch_time); |
e07f7183 JB |
1697 | |
1698 | if (trans_time < commit_time) { | |
1699 | ktime_t expires = ktime_add_ns(ktime_get(), | |
1700 | commit_time); | |
1701 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1702 | schedule_hrtimeout(&expires, HRTIMER_MODE_ABS); | |
1703 | } | |
470decc6 DK |
1704 | } |
1705 | ||
7058548c TT |
1706 | if (handle->h_sync) |
1707 | transaction->t_synchronous_commit = 1; | |
470decc6 | 1708 | current->journal_info = NULL; |
a51dca9c TT |
1709 | atomic_sub(handle->h_buffer_credits, |
1710 | &transaction->t_outstanding_credits); | |
470decc6 DK |
1711 | |
1712 | /* | |
1713 | * If the handle is marked SYNC, we need to set another commit | |
1714 | * going! We also want to force a commit if the current | |
1715 | * transaction is occupying too much of the log, or if the | |
1716 | * transaction is too old now. | |
1717 | */ | |
1718 | if (handle->h_sync || | |
a51dca9c TT |
1719 | (atomic_read(&transaction->t_outstanding_credits) > |
1720 | journal->j_max_transaction_buffers) || | |
1721 | time_after_eq(jiffies, transaction->t_expires)) { | |
470decc6 DK |
1722 | /* Do this even for aborted journals: an abort still |
1723 | * completes the commit thread, it just doesn't write | |
1724 | * anything to disk. */ | |
470decc6 | 1725 | |
470decc6 DK |
1726 | jbd_debug(2, "transaction too old, requesting commit for " |
1727 | "handle %p\n", handle); | |
1728 | /* This is non-blocking */ | |
c35a56a0 | 1729 | jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1730 | |
1731 | /* | |
f7f4bccb | 1732 | * Special case: JBD2_SYNC synchronous updates require us |
470decc6 DK |
1733 | * to wait for the commit to complete. |
1734 | */ | |
1735 | if (handle->h_sync && !(current->flags & PF_MEMALLOC)) | |
a51dca9c | 1736 | wait_for_commit = 1; |
470decc6 DK |
1737 | } |
1738 | ||
a51dca9c TT |
1739 | /* |
1740 | * Once we drop t_updates, if it goes to zero the transaction | |
25985edc | 1741 | * could start committing on us and eventually disappear. So |
a51dca9c TT |
1742 | * once we do this, we must not dereference transaction |
1743 | * pointer again. | |
1744 | */ | |
1745 | tid = transaction->t_tid; | |
1746 | if (atomic_dec_and_test(&transaction->t_updates)) { | |
1747 | wake_up(&journal->j_wait_updates); | |
1748 | if (journal->j_barrier_count) | |
1749 | wake_up(&journal->j_wait_transaction_locked); | |
1750 | } | |
1751 | ||
ab714aff | 1752 | rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_); |
7a4b188f | 1753 | |
a51dca9c TT |
1754 | if (wait_for_commit) |
1755 | err = jbd2_log_wait_commit(journal, tid); | |
1756 | ||
8f7d89f3 JK |
1757 | if (handle->h_rsv_handle) |
1758 | jbd2_journal_free_reserved(handle->h_rsv_handle); | |
41a5b913 | 1759 | free_and_exit: |
af1e76d6 | 1760 | jbd2_free_handle(handle); |
470decc6 DK |
1761 | return err; |
1762 | } | |
1763 | ||
470decc6 DK |
1764 | /* |
1765 | * | |
1766 | * List management code snippets: various functions for manipulating the | |
1767 | * transaction buffer lists. | |
1768 | * | |
1769 | */ | |
1770 | ||
1771 | /* | |
1772 | * Append a buffer to a transaction list, given the transaction's list head | |
1773 | * pointer. | |
1774 | * | |
1775 | * j_list_lock is held. | |
1776 | * | |
1777 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1778 | */ | |
1779 | ||
1780 | static inline void | |
1781 | __blist_add_buffer(struct journal_head **list, struct journal_head *jh) | |
1782 | { | |
1783 | if (!*list) { | |
1784 | jh->b_tnext = jh->b_tprev = jh; | |
1785 | *list = jh; | |
1786 | } else { | |
1787 | /* Insert at the tail of the list to preserve order */ | |
1788 | struct journal_head *first = *list, *last = first->b_tprev; | |
1789 | jh->b_tprev = last; | |
1790 | jh->b_tnext = first; | |
1791 | last->b_tnext = first->b_tprev = jh; | |
1792 | } | |
1793 | } | |
1794 | ||
1795 | /* | |
1796 | * Remove a buffer from a transaction list, given the transaction's list | |
1797 | * head pointer. | |
1798 | * | |
1799 | * Called with j_list_lock held, and the journal may not be locked. | |
1800 | * | |
1801 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1802 | */ | |
1803 | ||
1804 | static inline void | |
1805 | __blist_del_buffer(struct journal_head **list, struct journal_head *jh) | |
1806 | { | |
1807 | if (*list == jh) { | |
1808 | *list = jh->b_tnext; | |
1809 | if (*list == jh) | |
1810 | *list = NULL; | |
1811 | } | |
1812 | jh->b_tprev->b_tnext = jh->b_tnext; | |
1813 | jh->b_tnext->b_tprev = jh->b_tprev; | |
1814 | } | |
1815 | ||
1816 | /* | |
1817 | * Remove a buffer from the appropriate transaction list. | |
1818 | * | |
1819 | * Note that this function can *change* the value of | |
f5113eff JK |
1820 | * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or |
1821 | * t_reserved_list. If the caller is holding onto a copy of one of these | |
1822 | * pointers, it could go bad. Generally the caller needs to re-read the | |
1823 | * pointer from the transaction_t. | |
470decc6 | 1824 | * |
5bebccf9 | 1825 | * Called under j_list_lock. |
470decc6 | 1826 | */ |
5bebccf9 | 1827 | static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh) |
470decc6 DK |
1828 | { |
1829 | struct journal_head **list = NULL; | |
1830 | transaction_t *transaction; | |
1831 | struct buffer_head *bh = jh2bh(jh); | |
1832 | ||
1833 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
1834 | transaction = jh->b_transaction; | |
1835 | if (transaction) | |
1836 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
1837 | ||
1838 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
1839 | if (jh->b_jlist != BJ_None) | |
4019191b | 1840 | J_ASSERT_JH(jh, transaction != NULL); |
470decc6 DK |
1841 | |
1842 | switch (jh->b_jlist) { | |
1843 | case BJ_None: | |
1844 | return; | |
470decc6 DK |
1845 | case BJ_Metadata: |
1846 | transaction->t_nr_buffers--; | |
1847 | J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0); | |
1848 | list = &transaction->t_buffers; | |
1849 | break; | |
1850 | case BJ_Forget: | |
1851 | list = &transaction->t_forget; | |
1852 | break; | |
470decc6 DK |
1853 | case BJ_Shadow: |
1854 | list = &transaction->t_shadow_list; | |
1855 | break; | |
470decc6 DK |
1856 | case BJ_Reserved: |
1857 | list = &transaction->t_reserved_list; | |
1858 | break; | |
470decc6 DK |
1859 | } |
1860 | ||
1861 | __blist_del_buffer(list, jh); | |
1862 | jh->b_jlist = BJ_None; | |
1863 | if (test_clear_buffer_jbddirty(bh)) | |
1864 | mark_buffer_dirty(bh); /* Expose it to the VM */ | |
1865 | } | |
1866 | ||
de1b7941 JK |
1867 | /* |
1868 | * Remove buffer from all transactions. | |
1869 | * | |
1870 | * Called with bh_state lock and j_list_lock | |
1871 | * | |
1872 | * jh and bh may be already freed when this function returns. | |
1873 | */ | |
1874 | static void __jbd2_journal_unfile_buffer(struct journal_head *jh) | |
470decc6 | 1875 | { |
f7f4bccb | 1876 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 | 1877 | jh->b_transaction = NULL; |
de1b7941 | 1878 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1879 | } |
1880 | ||
f7f4bccb | 1881 | void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 | 1882 | { |
de1b7941 JK |
1883 | struct buffer_head *bh = jh2bh(jh); |
1884 | ||
1885 | /* Get reference so that buffer cannot be freed before we unlock it */ | |
1886 | get_bh(bh); | |
1887 | jbd_lock_bh_state(bh); | |
470decc6 | 1888 | spin_lock(&journal->j_list_lock); |
f7f4bccb | 1889 | __jbd2_journal_unfile_buffer(jh); |
470decc6 | 1890 | spin_unlock(&journal->j_list_lock); |
de1b7941 JK |
1891 | jbd_unlock_bh_state(bh); |
1892 | __brelse(bh); | |
470decc6 DK |
1893 | } |
1894 | ||
1895 | /* | |
f7f4bccb | 1896 | * Called from jbd2_journal_try_to_free_buffers(). |
470decc6 DK |
1897 | * |
1898 | * Called under jbd_lock_bh_state(bh) | |
1899 | */ | |
1900 | static void | |
1901 | __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh) | |
1902 | { | |
1903 | struct journal_head *jh; | |
1904 | ||
1905 | jh = bh2jh(bh); | |
1906 | ||
1907 | if (buffer_locked(bh) || buffer_dirty(bh)) | |
1908 | goto out; | |
1909 | ||
d2eb0b99 | 1910 | if (jh->b_next_transaction != NULL || jh->b_transaction != NULL) |
470decc6 DK |
1911 | goto out; |
1912 | ||
1913 | spin_lock(&journal->j_list_lock); | |
d2eb0b99 | 1914 | if (jh->b_cp_transaction != NULL) { |
470decc6 | 1915 | /* written-back checkpointed metadata buffer */ |
c254c9ec JK |
1916 | JBUFFER_TRACE(jh, "remove from checkpoint list"); |
1917 | __jbd2_journal_remove_checkpoint(jh); | |
470decc6 DK |
1918 | } |
1919 | spin_unlock(&journal->j_list_lock); | |
1920 | out: | |
1921 | return; | |
1922 | } | |
1923 | ||
470decc6 | 1924 | /** |
f7f4bccb | 1925 | * int jbd2_journal_try_to_free_buffers() - try to free page buffers. |
470decc6 DK |
1926 | * @journal: journal for operation |
1927 | * @page: to try and free | |
530576bb | 1928 | * @gfp_mask: we use the mask to detect how hard should we try to release |
d0164adc MG |
1929 | * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit |
1930 | * code to release the buffers. | |
470decc6 DK |
1931 | * |
1932 | * | |
1933 | * For all the buffers on this page, | |
1934 | * if they are fully written out ordered data, move them onto BUF_CLEAN | |
1935 | * so try_to_free_buffers() can reap them. | |
1936 | * | |
1937 | * This function returns non-zero if we wish try_to_free_buffers() | |
1938 | * to be called. We do this if the page is releasable by try_to_free_buffers(). | |
1939 | * We also do it if the page has locked or dirty buffers and the caller wants | |
1940 | * us to perform sync or async writeout. | |
1941 | * | |
1942 | * This complicates JBD locking somewhat. We aren't protected by the | |
1943 | * BKL here. We wish to remove the buffer from its committing or | |
f7f4bccb | 1944 | * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer. |
470decc6 DK |
1945 | * |
1946 | * This may *change* the value of transaction_t->t_datalist, so anyone | |
1947 | * who looks at t_datalist needs to lock against this function. | |
1948 | * | |
f7f4bccb MC |
1949 | * Even worse, someone may be doing a jbd2_journal_dirty_data on this |
1950 | * buffer. So we need to lock against that. jbd2_journal_dirty_data() | |
470decc6 DK |
1951 | * will come out of the lock with the buffer dirty, which makes it |
1952 | * ineligible for release here. | |
1953 | * | |
1954 | * Who else is affected by this? hmm... Really the only contender | |
1955 | * is do_get_write_access() - it could be looking at the buffer while | |
1956 | * journal_try_to_free_buffer() is changing its state. But that | |
1957 | * cannot happen because we never reallocate freed data as metadata | |
1958 | * while the data is part of a transaction. Yes? | |
530576bb MC |
1959 | * |
1960 | * Return 0 on failure, 1 on success | |
470decc6 | 1961 | */ |
f7f4bccb | 1962 | int jbd2_journal_try_to_free_buffers(journal_t *journal, |
530576bb | 1963 | struct page *page, gfp_t gfp_mask) |
470decc6 DK |
1964 | { |
1965 | struct buffer_head *head; | |
1966 | struct buffer_head *bh; | |
1967 | int ret = 0; | |
1968 | ||
1969 | J_ASSERT(PageLocked(page)); | |
1970 | ||
1971 | head = page_buffers(page); | |
1972 | bh = head; | |
1973 | do { | |
1974 | struct journal_head *jh; | |
1975 | ||
1976 | /* | |
1977 | * We take our own ref against the journal_head here to avoid | |
1978 | * having to add tons of locking around each instance of | |
530576bb | 1979 | * jbd2_journal_put_journal_head(). |
470decc6 | 1980 | */ |
f7f4bccb | 1981 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
1982 | if (!jh) |
1983 | continue; | |
1984 | ||
1985 | jbd_lock_bh_state(bh); | |
1986 | __journal_try_to_free_buffer(journal, bh); | |
f7f4bccb | 1987 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1988 | jbd_unlock_bh_state(bh); |
1989 | if (buffer_jbd(bh)) | |
1990 | goto busy; | |
1991 | } while ((bh = bh->b_this_page) != head); | |
530576bb | 1992 | |
470decc6 | 1993 | ret = try_to_free_buffers(page); |
530576bb | 1994 | |
470decc6 DK |
1995 | busy: |
1996 | return ret; | |
1997 | } | |
1998 | ||
1999 | /* | |
2000 | * This buffer is no longer needed. If it is on an older transaction's | |
2001 | * checkpoint list we need to record it on this transaction's forget list | |
2002 | * to pin this buffer (and hence its checkpointing transaction) down until | |
2003 | * this transaction commits. If the buffer isn't on a checkpoint list, we | |
2004 | * release it. | |
2005 | * Returns non-zero if JBD no longer has an interest in the buffer. | |
2006 | * | |
2007 | * Called under j_list_lock. | |
2008 | * | |
2009 | * Called under jbd_lock_bh_state(bh). | |
2010 | */ | |
2011 | static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction) | |
2012 | { | |
2013 | int may_free = 1; | |
2014 | struct buffer_head *bh = jh2bh(jh); | |
2015 | ||
470decc6 DK |
2016 | if (jh->b_cp_transaction) { |
2017 | JBUFFER_TRACE(jh, "on running+cp transaction"); | |
de1b7941 | 2018 | __jbd2_journal_temp_unlink_buffer(jh); |
f91d1d04 JK |
2019 | /* |
2020 | * We don't want to write the buffer anymore, clear the | |
2021 | * bit so that we don't confuse checks in | |
2022 | * __journal_file_buffer | |
2023 | */ | |
2024 | clear_buffer_dirty(bh); | |
f7f4bccb | 2025 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); |
470decc6 DK |
2026 | may_free = 0; |
2027 | } else { | |
2028 | JBUFFER_TRACE(jh, "on running transaction"); | |
de1b7941 | 2029 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
2030 | } |
2031 | return may_free; | |
2032 | } | |
2033 | ||
2034 | /* | |
f7f4bccb | 2035 | * jbd2_journal_invalidatepage |
470decc6 DK |
2036 | * |
2037 | * This code is tricky. It has a number of cases to deal with. | |
2038 | * | |
2039 | * There are two invariants which this code relies on: | |
2040 | * | |
2041 | * i_size must be updated on disk before we start calling invalidatepage on the | |
2042 | * data. | |
2043 | * | |
2044 | * This is done in ext3 by defining an ext3_setattr method which | |
2045 | * updates i_size before truncate gets going. By maintaining this | |
2046 | * invariant, we can be sure that it is safe to throw away any buffers | |
2047 | * attached to the current transaction: once the transaction commits, | |
2048 | * we know that the data will not be needed. | |
2049 | * | |
2050 | * Note however that we can *not* throw away data belonging to the | |
2051 | * previous, committing transaction! | |
2052 | * | |
2053 | * Any disk blocks which *are* part of the previous, committing | |
2054 | * transaction (and which therefore cannot be discarded immediately) are | |
2055 | * not going to be reused in the new running transaction | |
2056 | * | |
2057 | * The bitmap committed_data images guarantee this: any block which is | |
2058 | * allocated in one transaction and removed in the next will be marked | |
2059 | * as in-use in the committed_data bitmap, so cannot be reused until | |
2060 | * the next transaction to delete the block commits. This means that | |
2061 | * leaving committing buffers dirty is quite safe: the disk blocks | |
2062 | * cannot be reallocated to a different file and so buffer aliasing is | |
2063 | * not possible. | |
2064 | * | |
2065 | * | |
2066 | * The above applies mainly to ordered data mode. In writeback mode we | |
2067 | * don't make guarantees about the order in which data hits disk --- in | |
2068 | * particular we don't guarantee that new dirty data is flushed before | |
2069 | * transaction commit --- so it is always safe just to discard data | |
2070 | * immediately in that mode. --sct | |
2071 | */ | |
2072 | ||
2073 | /* | |
2074 | * The journal_unmap_buffer helper function returns zero if the buffer | |
2075 | * concerned remains pinned as an anonymous buffer belonging to an older | |
2076 | * transaction. | |
2077 | * | |
2078 | * We're outside-transaction here. Either or both of j_running_transaction | |
2079 | * and j_committing_transaction may be NULL. | |
2080 | */ | |
b794e7a6 JK |
2081 | static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh, |
2082 | int partial_page) | |
470decc6 DK |
2083 | { |
2084 | transaction_t *transaction; | |
2085 | struct journal_head *jh; | |
2086 | int may_free = 1; | |
470decc6 DK |
2087 | |
2088 | BUFFER_TRACE(bh, "entry"); | |
2089 | ||
2090 | /* | |
2091 | * It is safe to proceed here without the j_list_lock because the | |
2092 | * buffers cannot be stolen by try_to_free_buffers as long as we are | |
2093 | * holding the page lock. --sct | |
2094 | */ | |
2095 | ||
2096 | if (!buffer_jbd(bh)) | |
2097 | goto zap_buffer_unlocked; | |
2098 | ||
87c89c23 | 2099 | /* OK, we have data buffer in journaled mode */ |
a931da6a | 2100 | write_lock(&journal->j_state_lock); |
470decc6 DK |
2101 | jbd_lock_bh_state(bh); |
2102 | spin_lock(&journal->j_list_lock); | |
2103 | ||
f7f4bccb | 2104 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
2105 | if (!jh) |
2106 | goto zap_buffer_no_jh; | |
2107 | ||
ba869023 | 2108 | /* |
2109 | * We cannot remove the buffer from checkpoint lists until the | |
2110 | * transaction adding inode to orphan list (let's call it T) | |
2111 | * is committed. Otherwise if the transaction changing the | |
2112 | * buffer would be cleaned from the journal before T is | |
2113 | * committed, a crash will cause that the correct contents of | |
2114 | * the buffer will be lost. On the other hand we have to | |
2115 | * clear the buffer dirty bit at latest at the moment when the | |
2116 | * transaction marking the buffer as freed in the filesystem | |
2117 | * structures is committed because from that moment on the | |
b794e7a6 | 2118 | * block can be reallocated and used by a different page. |
ba869023 | 2119 | * Since the block hasn't been freed yet but the inode has |
2120 | * already been added to orphan list, it is safe for us to add | |
2121 | * the buffer to BJ_Forget list of the newest transaction. | |
b794e7a6 JK |
2122 | * |
2123 | * Also we have to clear buffer_mapped flag of a truncated buffer | |
2124 | * because the buffer_head may be attached to the page straddling | |
2125 | * i_size (can happen only when blocksize < pagesize) and thus the | |
2126 | * buffer_head can be reused when the file is extended again. So we end | |
2127 | * up keeping around invalidated buffers attached to transactions' | |
2128 | * BJ_Forget list just to stop checkpointing code from cleaning up | |
2129 | * the transaction this buffer was modified in. | |
ba869023 | 2130 | */ |
470decc6 DK |
2131 | transaction = jh->b_transaction; |
2132 | if (transaction == NULL) { | |
2133 | /* First case: not on any transaction. If it | |
2134 | * has no checkpoint link, then we can zap it: | |
2135 | * it's a writeback-mode buffer so we don't care | |
2136 | * if it hits disk safely. */ | |
2137 | if (!jh->b_cp_transaction) { | |
2138 | JBUFFER_TRACE(jh, "not on any transaction: zap"); | |
2139 | goto zap_buffer; | |
2140 | } | |
2141 | ||
2142 | if (!buffer_dirty(bh)) { | |
2143 | /* bdflush has written it. We can drop it now */ | |
bc23f0c8 | 2144 | __jbd2_journal_remove_checkpoint(jh); |
470decc6 DK |
2145 | goto zap_buffer; |
2146 | } | |
2147 | ||
2148 | /* OK, it must be in the journal but still not | |
2149 | * written fully to disk: it's metadata or | |
2150 | * journaled data... */ | |
2151 | ||
2152 | if (journal->j_running_transaction) { | |
2153 | /* ... and once the current transaction has | |
2154 | * committed, the buffer won't be needed any | |
2155 | * longer. */ | |
2156 | JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget"); | |
b794e7a6 | 2157 | may_free = __dispose_buffer(jh, |
470decc6 | 2158 | journal->j_running_transaction); |
b794e7a6 | 2159 | goto zap_buffer; |
470decc6 DK |
2160 | } else { |
2161 | /* There is no currently-running transaction. So the | |
2162 | * orphan record which we wrote for this file must have | |
2163 | * passed into commit. We must attach this buffer to | |
2164 | * the committing transaction, if it exists. */ | |
2165 | if (journal->j_committing_transaction) { | |
2166 | JBUFFER_TRACE(jh, "give to committing trans"); | |
b794e7a6 | 2167 | may_free = __dispose_buffer(jh, |
470decc6 | 2168 | journal->j_committing_transaction); |
b794e7a6 | 2169 | goto zap_buffer; |
470decc6 DK |
2170 | } else { |
2171 | /* The orphan record's transaction has | |
2172 | * committed. We can cleanse this buffer */ | |
2173 | clear_buffer_jbddirty(bh); | |
bc23f0c8 | 2174 | __jbd2_journal_remove_checkpoint(jh); |
470decc6 DK |
2175 | goto zap_buffer; |
2176 | } | |
2177 | } | |
2178 | } else if (transaction == journal->j_committing_transaction) { | |
9b57988d | 2179 | JBUFFER_TRACE(jh, "on committing transaction"); |
470decc6 | 2180 | /* |
ba869023 | 2181 | * The buffer is committing, we simply cannot touch |
b794e7a6 JK |
2182 | * it. If the page is straddling i_size we have to wait |
2183 | * for commit and try again. | |
2184 | */ | |
2185 | if (partial_page) { | |
b794e7a6 JK |
2186 | jbd2_journal_put_journal_head(jh); |
2187 | spin_unlock(&journal->j_list_lock); | |
2188 | jbd_unlock_bh_state(bh); | |
2189 | write_unlock(&journal->j_state_lock); | |
53e87268 | 2190 | return -EBUSY; |
b794e7a6 JK |
2191 | } |
2192 | /* | |
2193 | * OK, buffer won't be reachable after truncate. We just set | |
2194 | * j_next_transaction to the running transaction (if there is | |
2195 | * one) and mark buffer as freed so that commit code knows it | |
2196 | * should clear dirty bits when it is done with the buffer. | |
ba869023 | 2197 | */ |
470decc6 | 2198 | set_buffer_freed(bh); |
ba869023 | 2199 | if (journal->j_running_transaction && buffer_jbddirty(bh)) |
2200 | jh->b_next_transaction = journal->j_running_transaction; | |
f7f4bccb | 2201 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2202 | spin_unlock(&journal->j_list_lock); |
2203 | jbd_unlock_bh_state(bh); | |
a931da6a | 2204 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2205 | return 0; |
2206 | } else { | |
2207 | /* Good, the buffer belongs to the running transaction. | |
2208 | * We are writing our own transaction's data, not any | |
2209 | * previous one's, so it is safe to throw it away | |
2210 | * (remember that we expect the filesystem to have set | |
2211 | * i_size already for this truncate so recovery will not | |
2212 | * expose the disk blocks we are discarding here.) */ | |
2213 | J_ASSERT_JH(jh, transaction == journal->j_running_transaction); | |
9b57988d | 2214 | JBUFFER_TRACE(jh, "on running transaction"); |
470decc6 DK |
2215 | may_free = __dispose_buffer(jh, transaction); |
2216 | } | |
2217 | ||
2218 | zap_buffer: | |
b794e7a6 JK |
2219 | /* |
2220 | * This is tricky. Although the buffer is truncated, it may be reused | |
2221 | * if blocksize < pagesize and it is attached to the page straddling | |
2222 | * EOF. Since the buffer might have been added to BJ_Forget list of the | |
2223 | * running transaction, journal_get_write_access() won't clear | |
2224 | * b_modified and credit accounting gets confused. So clear b_modified | |
2225 | * here. | |
2226 | */ | |
2227 | jh->b_modified = 0; | |
f7f4bccb | 2228 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2229 | zap_buffer_no_jh: |
2230 | spin_unlock(&journal->j_list_lock); | |
2231 | jbd_unlock_bh_state(bh); | |
a931da6a | 2232 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2233 | zap_buffer_unlocked: |
2234 | clear_buffer_dirty(bh); | |
2235 | J_ASSERT_BH(bh, !buffer_jbddirty(bh)); | |
2236 | clear_buffer_mapped(bh); | |
2237 | clear_buffer_req(bh); | |
2238 | clear_buffer_new(bh); | |
15291164 ES |
2239 | clear_buffer_delay(bh); |
2240 | clear_buffer_unwritten(bh); | |
470decc6 DK |
2241 | bh->b_bdev = NULL; |
2242 | return may_free; | |
2243 | } | |
2244 | ||
2245 | /** | |
f7f4bccb | 2246 | * void jbd2_journal_invalidatepage() |
470decc6 DK |
2247 | * @journal: journal to use for flush... |
2248 | * @page: page to flush | |
259709b0 LC |
2249 | * @offset: start of the range to invalidate |
2250 | * @length: length of the range to invalidate | |
470decc6 | 2251 | * |
259709b0 LC |
2252 | * Reap page buffers containing data after in the specified range in page. |
2253 | * Can return -EBUSY if buffers are part of the committing transaction and | |
2254 | * the page is straddling i_size. Caller then has to wait for current commit | |
2255 | * and try again. | |
470decc6 | 2256 | */ |
53e87268 JK |
2257 | int jbd2_journal_invalidatepage(journal_t *journal, |
2258 | struct page *page, | |
259709b0 LC |
2259 | unsigned int offset, |
2260 | unsigned int length) | |
470decc6 DK |
2261 | { |
2262 | struct buffer_head *head, *bh, *next; | |
259709b0 | 2263 | unsigned int stop = offset + length; |
470decc6 | 2264 | unsigned int curr_off = 0; |
09cbfeaf | 2265 | int partial_page = (offset || length < PAGE_SIZE); |
470decc6 | 2266 | int may_free = 1; |
53e87268 | 2267 | int ret = 0; |
470decc6 DK |
2268 | |
2269 | if (!PageLocked(page)) | |
2270 | BUG(); | |
2271 | if (!page_has_buffers(page)) | |
53e87268 | 2272 | return 0; |
470decc6 | 2273 | |
09cbfeaf | 2274 | BUG_ON(stop > PAGE_SIZE || stop < length); |
259709b0 | 2275 | |
470decc6 DK |
2276 | /* We will potentially be playing with lists other than just the |
2277 | * data lists (especially for journaled data mode), so be | |
2278 | * cautious in our locking. */ | |
2279 | ||
2280 | head = bh = page_buffers(page); | |
2281 | do { | |
2282 | unsigned int next_off = curr_off + bh->b_size; | |
2283 | next = bh->b_this_page; | |
2284 | ||
259709b0 LC |
2285 | if (next_off > stop) |
2286 | return 0; | |
2287 | ||
470decc6 DK |
2288 | if (offset <= curr_off) { |
2289 | /* This block is wholly outside the truncation point */ | |
2290 | lock_buffer(bh); | |
259709b0 | 2291 | ret = journal_unmap_buffer(journal, bh, partial_page); |
470decc6 | 2292 | unlock_buffer(bh); |
53e87268 JK |
2293 | if (ret < 0) |
2294 | return ret; | |
2295 | may_free &= ret; | |
470decc6 DK |
2296 | } |
2297 | curr_off = next_off; | |
2298 | bh = next; | |
2299 | ||
2300 | } while (bh != head); | |
2301 | ||
259709b0 | 2302 | if (!partial_page) { |
470decc6 DK |
2303 | if (may_free && try_to_free_buffers(page)) |
2304 | J_ASSERT(!page_has_buffers(page)); | |
2305 | } | |
53e87268 | 2306 | return 0; |
470decc6 DK |
2307 | } |
2308 | ||
2309 | /* | |
2310 | * File a buffer on the given transaction list. | |
2311 | */ | |
f7f4bccb | 2312 | void __jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2313 | transaction_t *transaction, int jlist) |
2314 | { | |
2315 | struct journal_head **list = NULL; | |
2316 | int was_dirty = 0; | |
2317 | struct buffer_head *bh = jh2bh(jh); | |
2318 | ||
2319 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
2320 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
2321 | ||
2322 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
2323 | J_ASSERT_JH(jh, jh->b_transaction == transaction || | |
4019191b | 2324 | jh->b_transaction == NULL); |
470decc6 DK |
2325 | |
2326 | if (jh->b_transaction && jh->b_jlist == jlist) | |
2327 | return; | |
2328 | ||
470decc6 DK |
2329 | if (jlist == BJ_Metadata || jlist == BJ_Reserved || |
2330 | jlist == BJ_Shadow || jlist == BJ_Forget) { | |
f91d1d04 JK |
2331 | /* |
2332 | * For metadata buffers, we track dirty bit in buffer_jbddirty | |
2333 | * instead of buffer_dirty. We should not see a dirty bit set | |
2334 | * here because we clear it in do_get_write_access but e.g. | |
2335 | * tune2fs can modify the sb and set the dirty bit at any time | |
2336 | * so we try to gracefully handle that. | |
2337 | */ | |
2338 | if (buffer_dirty(bh)) | |
2339 | warn_dirty_buffer(bh); | |
470decc6 DK |
2340 | if (test_clear_buffer_dirty(bh) || |
2341 | test_clear_buffer_jbddirty(bh)) | |
2342 | was_dirty = 1; | |
2343 | } | |
2344 | ||
2345 | if (jh->b_transaction) | |
f7f4bccb | 2346 | __jbd2_journal_temp_unlink_buffer(jh); |
de1b7941 JK |
2347 | else |
2348 | jbd2_journal_grab_journal_head(bh); | |
470decc6 DK |
2349 | jh->b_transaction = transaction; |
2350 | ||
2351 | switch (jlist) { | |
2352 | case BJ_None: | |
2353 | J_ASSERT_JH(jh, !jh->b_committed_data); | |
2354 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
2355 | return; | |
470decc6 DK |
2356 | case BJ_Metadata: |
2357 | transaction->t_nr_buffers++; | |
2358 | list = &transaction->t_buffers; | |
2359 | break; | |
2360 | case BJ_Forget: | |
2361 | list = &transaction->t_forget; | |
2362 | break; | |
470decc6 DK |
2363 | case BJ_Shadow: |
2364 | list = &transaction->t_shadow_list; | |
2365 | break; | |
470decc6 DK |
2366 | case BJ_Reserved: |
2367 | list = &transaction->t_reserved_list; | |
2368 | break; | |
470decc6 DK |
2369 | } |
2370 | ||
2371 | __blist_add_buffer(list, jh); | |
2372 | jh->b_jlist = jlist; | |
2373 | ||
2374 | if (was_dirty) | |
2375 | set_buffer_jbddirty(bh); | |
2376 | } | |
2377 | ||
f7f4bccb | 2378 | void jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2379 | transaction_t *transaction, int jlist) |
2380 | { | |
2381 | jbd_lock_bh_state(jh2bh(jh)); | |
2382 | spin_lock(&transaction->t_journal->j_list_lock); | |
f7f4bccb | 2383 | __jbd2_journal_file_buffer(jh, transaction, jlist); |
470decc6 DK |
2384 | spin_unlock(&transaction->t_journal->j_list_lock); |
2385 | jbd_unlock_bh_state(jh2bh(jh)); | |
2386 | } | |
2387 | ||
2388 | /* | |
2389 | * Remove a buffer from its current buffer list in preparation for | |
2390 | * dropping it from its current transaction entirely. If the buffer has | |
2391 | * already started to be used by a subsequent transaction, refile the | |
2392 | * buffer on that transaction's metadata list. | |
2393 | * | |
de1b7941 | 2394 | * Called under j_list_lock |
470decc6 | 2395 | * Called under jbd_lock_bh_state(jh2bh(jh)) |
de1b7941 JK |
2396 | * |
2397 | * jh and bh may be already free when this function returns | |
470decc6 | 2398 | */ |
f7f4bccb | 2399 | void __jbd2_journal_refile_buffer(struct journal_head *jh) |
470decc6 | 2400 | { |
ba869023 | 2401 | int was_dirty, jlist; |
470decc6 DK |
2402 | struct buffer_head *bh = jh2bh(jh); |
2403 | ||
2404 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
2405 | if (jh->b_transaction) | |
2406 | assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock); | |
2407 | ||
2408 | /* If the buffer is now unused, just drop it. */ | |
2409 | if (jh->b_next_transaction == NULL) { | |
f7f4bccb | 2410 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
2411 | return; |
2412 | } | |
2413 | ||
2414 | /* | |
2415 | * It has been modified by a later transaction: add it to the new | |
2416 | * transaction's metadata list. | |
2417 | */ | |
2418 | ||
2419 | was_dirty = test_clear_buffer_jbddirty(bh); | |
f7f4bccb | 2420 | __jbd2_journal_temp_unlink_buffer(jh); |
de1b7941 JK |
2421 | /* |
2422 | * We set b_transaction here because b_next_transaction will inherit | |
2423 | * our jh reference and thus __jbd2_journal_file_buffer() must not | |
2424 | * take a new one. | |
2425 | */ | |
470decc6 DK |
2426 | jh->b_transaction = jh->b_next_transaction; |
2427 | jh->b_next_transaction = NULL; | |
ba869023 | 2428 | if (buffer_freed(bh)) |
2429 | jlist = BJ_Forget; | |
2430 | else if (jh->b_modified) | |
2431 | jlist = BJ_Metadata; | |
2432 | else | |
2433 | jlist = BJ_Reserved; | |
2434 | __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist); | |
470decc6 DK |
2435 | J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING); |
2436 | ||
2437 | if (was_dirty) | |
2438 | set_buffer_jbddirty(bh); | |
2439 | } | |
2440 | ||
2441 | /* | |
de1b7941 JK |
2442 | * __jbd2_journal_refile_buffer() with necessary locking added. We take our |
2443 | * bh reference so that we can safely unlock bh. | |
2444 | * | |
2445 | * The jh and bh may be freed by this call. | |
470decc6 | 2446 | */ |
f7f4bccb | 2447 | void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 DK |
2448 | { |
2449 | struct buffer_head *bh = jh2bh(jh); | |
2450 | ||
de1b7941 JK |
2451 | /* Get reference so that buffer cannot be freed before we unlock it */ |
2452 | get_bh(bh); | |
470decc6 DK |
2453 | jbd_lock_bh_state(bh); |
2454 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 2455 | __jbd2_journal_refile_buffer(jh); |
470decc6 | 2456 | jbd_unlock_bh_state(bh); |
470decc6 DK |
2457 | spin_unlock(&journal->j_list_lock); |
2458 | __brelse(bh); | |
2459 | } | |
c851ed54 JK |
2460 | |
2461 | /* | |
2462 | * File inode in the inode list of the handle's transaction | |
2463 | */ | |
41617e1a JK |
2464 | static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode, |
2465 | unsigned long flags) | |
c851ed54 JK |
2466 | { |
2467 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 2468 | journal_t *journal; |
c851ed54 JK |
2469 | |
2470 | if (is_handle_aborted(handle)) | |
41a5b913 TT |
2471 | return -EROFS; |
2472 | journal = transaction->t_journal; | |
c851ed54 JK |
2473 | |
2474 | jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino, | |
2475 | transaction->t_tid); | |
2476 | ||
2477 | /* | |
2478 | * First check whether inode isn't already on the transaction's | |
2479 | * lists without taking the lock. Note that this check is safe | |
2480 | * without the lock as we cannot race with somebody removing inode | |
2481 | * from the transaction. The reason is that we remove inode from the | |
2482 | * transaction only in journal_release_jbd_inode() and when we commit | |
2483 | * the transaction. We are guarded from the first case by holding | |
2484 | * a reference to the inode. We are safe against the second case | |
2485 | * because if jinode->i_transaction == transaction, commit code | |
2486 | * cannot touch the transaction because we hold reference to it, | |
2487 | * and if jinode->i_next_transaction == transaction, commit code | |
2488 | * will only file the inode where we want it. | |
2489 | */ | |
41617e1a JK |
2490 | if ((jinode->i_transaction == transaction || |
2491 | jinode->i_next_transaction == transaction) && | |
2492 | (jinode->i_flags & flags) == flags) | |
c851ed54 JK |
2493 | return 0; |
2494 | ||
2495 | spin_lock(&journal->j_list_lock); | |
41617e1a JK |
2496 | jinode->i_flags |= flags; |
2497 | /* Is inode already attached where we need it? */ | |
c851ed54 JK |
2498 | if (jinode->i_transaction == transaction || |
2499 | jinode->i_next_transaction == transaction) | |
2500 | goto done; | |
2501 | ||
81be12c8 JK |
2502 | /* |
2503 | * We only ever set this variable to 1 so the test is safe. Since | |
2504 | * t_need_data_flush is likely to be set, we do the test to save some | |
2505 | * cacheline bouncing | |
2506 | */ | |
2507 | if (!transaction->t_need_data_flush) | |
2508 | transaction->t_need_data_flush = 1; | |
c851ed54 JK |
2509 | /* On some different transaction's list - should be |
2510 | * the committing one */ | |
2511 | if (jinode->i_transaction) { | |
2512 | J_ASSERT(jinode->i_next_transaction == NULL); | |
2513 | J_ASSERT(jinode->i_transaction == | |
2514 | journal->j_committing_transaction); | |
2515 | jinode->i_next_transaction = transaction; | |
2516 | goto done; | |
2517 | } | |
2518 | /* Not on any transaction list... */ | |
2519 | J_ASSERT(!jinode->i_next_transaction); | |
2520 | jinode->i_transaction = transaction; | |
2521 | list_add(&jinode->i_list, &transaction->t_inode_list); | |
2522 | done: | |
2523 | spin_unlock(&journal->j_list_lock); | |
2524 | ||
2525 | return 0; | |
2526 | } | |
2527 | ||
41617e1a JK |
2528 | int jbd2_journal_inode_add_write(handle_t *handle, struct jbd2_inode *jinode) |
2529 | { | |
2530 | return jbd2_journal_file_inode(handle, jinode, | |
2531 | JI_WRITE_DATA | JI_WAIT_DATA); | |
2532 | } | |
2533 | ||
2534 | int jbd2_journal_inode_add_wait(handle_t *handle, struct jbd2_inode *jinode) | |
2535 | { | |
2536 | return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA); | |
2537 | } | |
2538 | ||
c851ed54 | 2539 | /* |
7f5aa215 JK |
2540 | * File truncate and transaction commit interact with each other in a |
2541 | * non-trivial way. If a transaction writing data block A is | |
2542 | * committing, we cannot discard the data by truncate until we have | |
2543 | * written them. Otherwise if we crashed after the transaction with | |
2544 | * write has committed but before the transaction with truncate has | |
2545 | * committed, we could see stale data in block A. This function is a | |
2546 | * helper to solve this problem. It starts writeout of the truncated | |
2547 | * part in case it is in the committing transaction. | |
2548 | * | |
2549 | * Filesystem code must call this function when inode is journaled in | |
2550 | * ordered mode before truncation happens and after the inode has been | |
2551 | * placed on orphan list with the new inode size. The second condition | |
2552 | * avoids the race that someone writes new data and we start | |
2553 | * committing the transaction after this function has been called but | |
2554 | * before a transaction for truncate is started (and furthermore it | |
2555 | * allows us to optimize the case where the addition to orphan list | |
2556 | * happens in the same transaction as write --- we don't have to write | |
2557 | * any data in such case). | |
c851ed54 | 2558 | */ |
7f5aa215 JK |
2559 | int jbd2_journal_begin_ordered_truncate(journal_t *journal, |
2560 | struct jbd2_inode *jinode, | |
c851ed54 JK |
2561 | loff_t new_size) |
2562 | { | |
7f5aa215 | 2563 | transaction_t *inode_trans, *commit_trans; |
c851ed54 JK |
2564 | int ret = 0; |
2565 | ||
7f5aa215 JK |
2566 | /* This is a quick check to avoid locking if not necessary */ |
2567 | if (!jinode->i_transaction) | |
c851ed54 | 2568 | goto out; |
7f5aa215 JK |
2569 | /* Locks are here just to force reading of recent values, it is |
2570 | * enough that the transaction was not committing before we started | |
2571 | * a transaction adding the inode to orphan list */ | |
a931da6a | 2572 | read_lock(&journal->j_state_lock); |
c851ed54 | 2573 | commit_trans = journal->j_committing_transaction; |
a931da6a | 2574 | read_unlock(&journal->j_state_lock); |
7f5aa215 JK |
2575 | spin_lock(&journal->j_list_lock); |
2576 | inode_trans = jinode->i_transaction; | |
2577 | spin_unlock(&journal->j_list_lock); | |
2578 | if (inode_trans == commit_trans) { | |
2579 | ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping, | |
c851ed54 JK |
2580 | new_size, LLONG_MAX); |
2581 | if (ret) | |
2582 | jbd2_journal_abort(journal, ret); | |
2583 | } | |
2584 | out: | |
2585 | return ret; | |
2586 | } |