]>
Commit | Line | Data |
---|---|---|
f5166768 | 1 | // SPDX-License-Identifier: GPL-2.0+ |
470decc6 | 2 | /* |
58862699 | 3 | * linux/fs/jbd2/transaction.c |
470decc6 DK |
4 | * |
5 | * Written by Stephen C. Tweedie <sct@redhat.com>, 1998 | |
6 | * | |
7 | * Copyright 1998 Red Hat corp --- All Rights Reserved | |
8 | * | |
470decc6 DK |
9 | * Generic filesystem transaction handling code; part of the ext2fs |
10 | * journaling system. | |
11 | * | |
12 | * This file manages transactions (compound commits managed by the | |
13 | * journaling code) and handles (individual atomic operations by the | |
14 | * filesystem). | |
15 | */ | |
16 | ||
17 | #include <linux/time.h> | |
18 | #include <linux/fs.h> | |
f7f4bccb | 19 | #include <linux/jbd2.h> |
470decc6 DK |
20 | #include <linux/errno.h> |
21 | #include <linux/slab.h> | |
22 | #include <linux/timer.h> | |
470decc6 DK |
23 | #include <linux/mm.h> |
24 | #include <linux/highmem.h> | |
e07f7183 | 25 | #include <linux/hrtimer.h> |
47def826 | 26 | #include <linux/backing-dev.h> |
44705754 | 27 | #include <linux/bug.h> |
47def826 | 28 | #include <linux/module.h> |
81378da6 | 29 | #include <linux/sched/mm.h> |
470decc6 | 30 | |
343d9c28 TT |
31 | #include <trace/events/jbd2.h> |
32 | ||
7ddae860 | 33 | static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh); |
de1b7941 | 34 | static void __jbd2_journal_unfile_buffer(struct journal_head *jh); |
7ddae860 | 35 | |
0c2022ec YY |
36 | static struct kmem_cache *transaction_cache; |
37 | int __init jbd2_journal_init_transaction_cache(void) | |
38 | { | |
39 | J_ASSERT(!transaction_cache); | |
40 | transaction_cache = kmem_cache_create("jbd2_transaction_s", | |
41 | sizeof(transaction_t), | |
42 | 0, | |
43 | SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY, | |
44 | NULL); | |
0d52154b CX |
45 | if (!transaction_cache) { |
46 | pr_emerg("JBD2: failed to create transaction cache\n"); | |
47 | return -ENOMEM; | |
48 | } | |
49 | return 0; | |
0c2022ec YY |
50 | } |
51 | ||
52 | void jbd2_journal_destroy_transaction_cache(void) | |
53 | { | |
8bdd5b60 WL |
54 | kmem_cache_destroy(transaction_cache); |
55 | transaction_cache = NULL; | |
0c2022ec YY |
56 | } |
57 | ||
58 | void jbd2_journal_free_transaction(transaction_t *transaction) | |
59 | { | |
60 | if (unlikely(ZERO_OR_NULL_PTR(transaction))) | |
61 | return; | |
62 | kmem_cache_free(transaction_cache, transaction); | |
63 | } | |
64 | ||
470decc6 | 65 | /* |
f7f4bccb | 66 | * jbd2_get_transaction: obtain a new transaction_t object. |
470decc6 | 67 | * |
0df6f469 | 68 | * Simply initialise a new transaction. Initialize it in |
470decc6 DK |
69 | * RUNNING state and add it to the current journal (which should not |
70 | * have an existing running transaction: we only make a new transaction | |
71 | * once we have started to commit the old one). | |
72 | * | |
73 | * Preconditions: | |
74 | * The journal MUST be locked. We don't perform atomic mallocs on the | |
75 | * new transaction and we can't block without protecting against other | |
76 | * processes trying to touch the journal while it is in transition. | |
77 | * | |
470decc6 DK |
78 | */ |
79 | ||
0df6f469 LS |
80 | static void jbd2_get_transaction(journal_t *journal, |
81 | transaction_t *transaction) | |
470decc6 DK |
82 | { |
83 | transaction->t_journal = journal; | |
84 | transaction->t_state = T_RUNNING; | |
e07f7183 | 85 | transaction->t_start_time = ktime_get(); |
470decc6 DK |
86 | transaction->t_tid = journal->j_transaction_sequence++; |
87 | transaction->t_expires = jiffies + journal->j_commit_interval; | |
88 | spin_lock_init(&transaction->t_handle_lock); | |
a51dca9c | 89 | atomic_set(&transaction->t_updates, 0); |
8f7d89f3 JK |
90 | atomic_set(&transaction->t_outstanding_credits, |
91 | atomic_read(&journal->j_reserved_credits)); | |
8dd42046 | 92 | atomic_set(&transaction->t_handle_count, 0); |
c851ed54 | 93 | INIT_LIST_HEAD(&transaction->t_inode_list); |
3e624fc7 | 94 | INIT_LIST_HEAD(&transaction->t_private_list); |
470decc6 DK |
95 | |
96 | /* Set up the commit timer for the new transaction. */ | |
b1f485f2 | 97 | journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires); |
470decc6 DK |
98 | add_timer(&journal->j_commit_timer); |
99 | ||
100 | J_ASSERT(journal->j_running_transaction == NULL); | |
101 | journal->j_running_transaction = transaction; | |
8e85fb3f JL |
102 | transaction->t_max_wait = 0; |
103 | transaction->t_start = jiffies; | |
9fff24aa | 104 | transaction->t_requested = 0; |
470decc6 DK |
105 | } |
106 | ||
107 | /* | |
108 | * Handle management. | |
109 | * | |
110 | * A handle_t is an object which represents a single atomic update to a | |
111 | * filesystem, and which tracks all of the modifications which form part | |
112 | * of that one update. | |
113 | */ | |
114 | ||
6d0bf005 | 115 | /* |
28e35e42 | 116 | * Update transaction's maximum wait time, if debugging is enabled. |
6d0bf005 TT |
117 | * |
118 | * In order for t_max_wait to be reliable, it must be protected by a | |
119 | * lock. But doing so will mean that start_this_handle() can not be | |
120 | * run in parallel on SMP systems, which limits our scalability. So | |
121 | * unless debugging is enabled, we no longer update t_max_wait, which | |
122 | * means that maximum wait time reported by the jbd2_run_stats | |
123 | * tracepoint will always be zero. | |
124 | */ | |
28e35e42 TM |
125 | static inline void update_t_max_wait(transaction_t *transaction, |
126 | unsigned long ts) | |
6d0bf005 TT |
127 | { |
128 | #ifdef CONFIG_JBD2_DEBUG | |
6d0bf005 TT |
129 | if (jbd2_journal_enable_debug && |
130 | time_after(transaction->t_start, ts)) { | |
131 | ts = jbd2_time_diff(ts, transaction->t_start); | |
132 | spin_lock(&transaction->t_handle_lock); | |
133 | if (ts > transaction->t_max_wait) | |
134 | transaction->t_max_wait = ts; | |
135 | spin_unlock(&transaction->t_handle_lock); | |
136 | } | |
137 | #endif | |
138 | } | |
139 | ||
8f7d89f3 | 140 | /* |
96f1e097 JK |
141 | * Wait until running transaction passes to T_FLUSH state and new transaction |
142 | * can thus be started. Also starts the commit if needed. The function expects | |
143 | * running transaction to exist and releases j_state_lock. | |
8f7d89f3 JK |
144 | */ |
145 | static void wait_transaction_locked(journal_t *journal) | |
146 | __releases(journal->j_state_lock) | |
147 | { | |
148 | DEFINE_WAIT(wait); | |
149 | int need_to_start; | |
150 | tid_t tid = journal->j_running_transaction->t_tid; | |
151 | ||
152 | prepare_to_wait(&journal->j_wait_transaction_locked, &wait, | |
153 | TASK_UNINTERRUPTIBLE); | |
154 | need_to_start = !tid_geq(journal->j_commit_request, tid); | |
155 | read_unlock(&journal->j_state_lock); | |
156 | if (need_to_start) | |
157 | jbd2_log_start_commit(journal, tid); | |
e03a9976 | 158 | jbd2_might_wait_for_commit(journal); |
8f7d89f3 JK |
159 | schedule(); |
160 | finish_wait(&journal->j_wait_transaction_locked, &wait); | |
161 | } | |
162 | ||
96f1e097 JK |
163 | /* |
164 | * Wait until running transaction transitions from T_SWITCH to T_FLUSH | |
165 | * state and new transaction can thus be started. The function releases | |
166 | * j_state_lock. | |
167 | */ | |
168 | static void wait_transaction_switching(journal_t *journal) | |
169 | __releases(journal->j_state_lock) | |
170 | { | |
171 | DEFINE_WAIT(wait); | |
172 | ||
173 | if (WARN_ON(!journal->j_running_transaction || | |
70616e51 TT |
174 | journal->j_running_transaction->t_state != T_SWITCH)) { |
175 | read_unlock(&journal->j_state_lock); | |
96f1e097 | 176 | return; |
70616e51 | 177 | } |
96f1e097 JK |
178 | prepare_to_wait(&journal->j_wait_transaction_locked, &wait, |
179 | TASK_UNINTERRUPTIBLE); | |
180 | read_unlock(&journal->j_state_lock); | |
181 | /* | |
182 | * We don't call jbd2_might_wait_for_commit() here as there's no | |
183 | * waiting for outstanding handles happening anymore in T_SWITCH state | |
184 | * and handling of reserved handles actually relies on that for | |
185 | * correctness. | |
186 | */ | |
187 | schedule(); | |
188 | finish_wait(&journal->j_wait_transaction_locked, &wait); | |
189 | } | |
190 | ||
8f7d89f3 JK |
191 | static void sub_reserved_credits(journal_t *journal, int blocks) |
192 | { | |
193 | atomic_sub(blocks, &journal->j_reserved_credits); | |
194 | wake_up(&journal->j_wait_reserved); | |
195 | } | |
196 | ||
197 | /* | |
198 | * Wait until we can add credits for handle to the running transaction. Called | |
199 | * with j_state_lock held for reading. Returns 0 if handle joined the running | |
200 | * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and | |
201 | * caller must retry. | |
202 | */ | |
203 | static int add_transaction_credits(journal_t *journal, int blocks, | |
204 | int rsv_blocks) | |
205 | { | |
206 | transaction_t *t = journal->j_running_transaction; | |
207 | int needed; | |
208 | int total = blocks + rsv_blocks; | |
209 | ||
210 | /* | |
211 | * If the current transaction is locked down for commit, wait | |
212 | * for the lock to be released. | |
213 | */ | |
96f1e097 JK |
214 | if (t->t_state != T_RUNNING) { |
215 | WARN_ON_ONCE(t->t_state >= T_FLUSH); | |
8f7d89f3 JK |
216 | wait_transaction_locked(journal); |
217 | return 1; | |
218 | } | |
219 | ||
220 | /* | |
221 | * If there is not enough space left in the log to write all | |
222 | * potential buffers requested by this operation, we need to | |
223 | * stall pending a log checkpoint to free some more log space. | |
224 | */ | |
225 | needed = atomic_add_return(total, &t->t_outstanding_credits); | |
226 | if (needed > journal->j_max_transaction_buffers) { | |
227 | /* | |
228 | * If the current transaction is already too large, | |
229 | * then start to commit it: we can then go back and | |
230 | * attach this handle to a new transaction. | |
231 | */ | |
232 | atomic_sub(total, &t->t_outstanding_credits); | |
6d3ec14d LC |
233 | |
234 | /* | |
235 | * Is the number of reserved credits in the current transaction too | |
236 | * big to fit this handle? Wait until reserved credits are freed. | |
237 | */ | |
238 | if (atomic_read(&journal->j_reserved_credits) + total > | |
239 | journal->j_max_transaction_buffers) { | |
240 | read_unlock(&journal->j_state_lock); | |
e03a9976 | 241 | jbd2_might_wait_for_commit(journal); |
6d3ec14d LC |
242 | wait_event(journal->j_wait_reserved, |
243 | atomic_read(&journal->j_reserved_credits) + total <= | |
244 | journal->j_max_transaction_buffers); | |
245 | return 1; | |
246 | } | |
247 | ||
8f7d89f3 JK |
248 | wait_transaction_locked(journal); |
249 | return 1; | |
250 | } | |
251 | ||
252 | /* | |
253 | * The commit code assumes that it can get enough log space | |
254 | * without forcing a checkpoint. This is *critical* for | |
255 | * correctness: a checkpoint of a buffer which is also | |
256 | * associated with a committing transaction creates a deadlock, | |
257 | * so commit simply cannot force through checkpoints. | |
258 | * | |
259 | * We must therefore ensure the necessary space in the journal | |
260 | * *before* starting to dirty potentially checkpointed buffers | |
261 | * in the new transaction. | |
262 | */ | |
263 | if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) { | |
264 | atomic_sub(total, &t->t_outstanding_credits); | |
265 | read_unlock(&journal->j_state_lock); | |
e03a9976 | 266 | jbd2_might_wait_for_commit(journal); |
8f7d89f3 JK |
267 | write_lock(&journal->j_state_lock); |
268 | if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) | |
269 | __jbd2_log_wait_for_space(journal); | |
270 | write_unlock(&journal->j_state_lock); | |
271 | return 1; | |
272 | } | |
273 | ||
274 | /* No reservation? We are done... */ | |
275 | if (!rsv_blocks) | |
276 | return 0; | |
277 | ||
278 | needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits); | |
279 | /* We allow at most half of a transaction to be reserved */ | |
280 | if (needed > journal->j_max_transaction_buffers / 2) { | |
281 | sub_reserved_credits(journal, rsv_blocks); | |
282 | atomic_sub(total, &t->t_outstanding_credits); | |
283 | read_unlock(&journal->j_state_lock); | |
e03a9976 | 284 | jbd2_might_wait_for_commit(journal); |
8f7d89f3 JK |
285 | wait_event(journal->j_wait_reserved, |
286 | atomic_read(&journal->j_reserved_credits) + rsv_blocks | |
287 | <= journal->j_max_transaction_buffers / 2); | |
288 | return 1; | |
289 | } | |
290 | return 0; | |
291 | } | |
292 | ||
470decc6 DK |
293 | /* |
294 | * start_this_handle: Given a handle, deal with any locking or stalling | |
295 | * needed to make sure that there is enough journal space for the handle | |
296 | * to begin. Attach the handle to a transaction and set up the | |
297 | * transaction's buffer credits. | |
298 | */ | |
299 | ||
47def826 | 300 | static int start_this_handle(journal_t *journal, handle_t *handle, |
d2159fb7 | 301 | gfp_t gfp_mask) |
470decc6 | 302 | { |
e4471831 | 303 | transaction_t *transaction, *new_transaction = NULL; |
8f7d89f3 JK |
304 | int blocks = handle->h_buffer_credits; |
305 | int rsv_blocks = 0; | |
28e35e42 | 306 | unsigned long ts = jiffies; |
470decc6 | 307 | |
6d3ec14d LC |
308 | if (handle->h_rsv_handle) |
309 | rsv_blocks = handle->h_rsv_handle->h_buffer_credits; | |
310 | ||
8f7d89f3 | 311 | /* |
6d3ec14d LC |
312 | * Limit the number of reserved credits to 1/2 of maximum transaction |
313 | * size and limit the number of total credits to not exceed maximum | |
314 | * transaction size per operation. | |
8f7d89f3 | 315 | */ |
6d3ec14d LC |
316 | if ((rsv_blocks > journal->j_max_transaction_buffers / 2) || |
317 | (rsv_blocks + blocks > journal->j_max_transaction_buffers)) { | |
318 | printk(KERN_ERR "JBD2: %s wants too many credits " | |
319 | "credits:%d rsv_credits:%d max:%d\n", | |
320 | current->comm, blocks, rsv_blocks, | |
321 | journal->j_max_transaction_buffers); | |
322 | WARN_ON(1); | |
47def826 | 323 | return -ENOSPC; |
470decc6 DK |
324 | } |
325 | ||
326 | alloc_transaction: | |
327 | if (!journal->j_running_transaction) { | |
6ccaf3e2 MH |
328 | /* |
329 | * If __GFP_FS is not present, then we may be being called from | |
330 | * inside the fs writeback layer, so we MUST NOT fail. | |
331 | */ | |
332 | if ((gfp_mask & __GFP_FS) == 0) | |
333 | gfp_mask |= __GFP_NOFAIL; | |
b2f4edb3 WG |
334 | new_transaction = kmem_cache_zalloc(transaction_cache, |
335 | gfp_mask); | |
6ccaf3e2 | 336 | if (!new_transaction) |
47def826 | 337 | return -ENOMEM; |
470decc6 DK |
338 | } |
339 | ||
340 | jbd_debug(3, "New handle %p going live.\n", handle); | |
341 | ||
470decc6 DK |
342 | /* |
343 | * We need to hold j_state_lock until t_updates has been incremented, | |
344 | * for proper journal barrier handling | |
345 | */ | |
a931da6a TT |
346 | repeat: |
347 | read_lock(&journal->j_state_lock); | |
5c2178e7 | 348 | BUG_ON(journal->j_flags & JBD2_UNMOUNT); |
470decc6 | 349 | if (is_journal_aborted(journal) || |
f7f4bccb | 350 | (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) { |
a931da6a | 351 | read_unlock(&journal->j_state_lock); |
0c2022ec | 352 | jbd2_journal_free_transaction(new_transaction); |
47def826 | 353 | return -EROFS; |
470decc6 DK |
354 | } |
355 | ||
8f7d89f3 JK |
356 | /* |
357 | * Wait on the journal's transaction barrier if necessary. Specifically | |
358 | * we allow reserved handles to proceed because otherwise commit could | |
359 | * deadlock on page writeback not being able to complete. | |
360 | */ | |
361 | if (!handle->h_reserved && journal->j_barrier_count) { | |
a931da6a | 362 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
363 | wait_event(journal->j_wait_transaction_locked, |
364 | journal->j_barrier_count == 0); | |
365 | goto repeat; | |
366 | } | |
367 | ||
368 | if (!journal->j_running_transaction) { | |
a931da6a TT |
369 | read_unlock(&journal->j_state_lock); |
370 | if (!new_transaction) | |
470decc6 | 371 | goto alloc_transaction; |
a931da6a | 372 | write_lock(&journal->j_state_lock); |
d7961c7f | 373 | if (!journal->j_running_transaction && |
8f7d89f3 | 374 | (handle->h_reserved || !journal->j_barrier_count)) { |
a931da6a TT |
375 | jbd2_get_transaction(journal, new_transaction); |
376 | new_transaction = NULL; | |
470decc6 | 377 | } |
a931da6a TT |
378 | write_unlock(&journal->j_state_lock); |
379 | goto repeat; | |
470decc6 DK |
380 | } |
381 | ||
382 | transaction = journal->j_running_transaction; | |
383 | ||
8f7d89f3 JK |
384 | if (!handle->h_reserved) { |
385 | /* We may have dropped j_state_lock - restart in that case */ | |
386 | if (add_transaction_credits(journal, blocks, rsv_blocks)) | |
387 | goto repeat; | |
388 | } else { | |
470decc6 | 389 | /* |
8f7d89f3 JK |
390 | * We have handle reserved so we are allowed to join T_LOCKED |
391 | * transaction and we don't have to check for transaction size | |
96f1e097 JK |
392 | * and journal space. But we still have to wait while running |
393 | * transaction is being switched to a committing one as it | |
394 | * won't wait for any handles anymore. | |
470decc6 | 395 | */ |
96f1e097 JK |
396 | if (transaction->t_state == T_SWITCH) { |
397 | wait_transaction_switching(journal); | |
398 | goto repeat; | |
399 | } | |
8f7d89f3 JK |
400 | sub_reserved_credits(journal, blocks); |
401 | handle->h_reserved = 0; | |
470decc6 DK |
402 | } |
403 | ||
404 | /* OK, account for the buffers that this operation expects to | |
8dd42046 | 405 | * use and add the handle to the running transaction. |
8dd42046 | 406 | */ |
28e35e42 | 407 | update_t_max_wait(transaction, ts); |
470decc6 | 408 | handle->h_transaction = transaction; |
8f7d89f3 | 409 | handle->h_requested_credits = blocks; |
343d9c28 | 410 | handle->h_start_jiffies = jiffies; |
a51dca9c | 411 | atomic_inc(&transaction->t_updates); |
8dd42046 | 412 | atomic_inc(&transaction->t_handle_count); |
8f7d89f3 JK |
413 | jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n", |
414 | handle, blocks, | |
a51dca9c | 415 | atomic_read(&transaction->t_outstanding_credits), |
76c39904 | 416 | jbd2_log_space_left(journal)); |
a931da6a | 417 | read_unlock(&journal->j_state_lock); |
41a5b913 | 418 | current->journal_info = handle; |
9599b0e5 | 419 | |
ab714aff | 420 | rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_); |
0c2022ec | 421 | jbd2_journal_free_transaction(new_transaction); |
81378da6 MH |
422 | /* |
423 | * Ensure that no allocations done while the transaction is open are | |
424 | * going to recurse back to the fs layer. | |
425 | */ | |
426 | handle->saved_alloc_context = memalloc_nofs_save(); | |
47def826 | 427 | return 0; |
470decc6 DK |
428 | } |
429 | ||
430 | /* Allocate a new handle. This should probably be in a slab... */ | |
431 | static handle_t *new_handle(int nblocks) | |
432 | { | |
af1e76d6 | 433 | handle_t *handle = jbd2_alloc_handle(GFP_NOFS); |
470decc6 DK |
434 | if (!handle) |
435 | return NULL; | |
470decc6 DK |
436 | handle->h_buffer_credits = nblocks; |
437 | handle->h_ref = 1; | |
438 | ||
439 | return handle; | |
440 | } | |
441 | ||
8f7d89f3 JK |
442 | handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks, |
443 | gfp_t gfp_mask, unsigned int type, | |
444 | unsigned int line_no) | |
470decc6 DK |
445 | { |
446 | handle_t *handle = journal_current_handle(); | |
447 | int err; | |
448 | ||
449 | if (!journal) | |
450 | return ERR_PTR(-EROFS); | |
451 | ||
452 | if (handle) { | |
453 | J_ASSERT(handle->h_transaction->t_journal == journal); | |
454 | handle->h_ref++; | |
455 | return handle; | |
456 | } | |
457 | ||
458 | handle = new_handle(nblocks); | |
459 | if (!handle) | |
460 | return ERR_PTR(-ENOMEM); | |
8f7d89f3 JK |
461 | if (rsv_blocks) { |
462 | handle_t *rsv_handle; | |
463 | ||
464 | rsv_handle = new_handle(rsv_blocks); | |
465 | if (!rsv_handle) { | |
466 | jbd2_free_handle(handle); | |
467 | return ERR_PTR(-ENOMEM); | |
468 | } | |
469 | rsv_handle->h_reserved = 1; | |
470 | rsv_handle->h_journal = journal; | |
471 | handle->h_rsv_handle = rsv_handle; | |
472 | } | |
470decc6 | 473 | |
47def826 | 474 | err = start_this_handle(journal, handle, gfp_mask); |
470decc6 | 475 | if (err < 0) { |
8f7d89f3 JK |
476 | if (handle->h_rsv_handle) |
477 | jbd2_free_handle(handle->h_rsv_handle); | |
af1e76d6 | 478 | jbd2_free_handle(handle); |
df05c1b8 | 479 | return ERR_PTR(err); |
470decc6 | 480 | } |
343d9c28 TT |
481 | handle->h_type = type; |
482 | handle->h_line_no = line_no; | |
483 | trace_jbd2_handle_start(journal->j_fs_dev->bd_dev, | |
484 | handle->h_transaction->t_tid, type, | |
485 | line_no, nblocks); | |
81378da6 | 486 | |
470decc6 DK |
487 | return handle; |
488 | } | |
47def826 TT |
489 | EXPORT_SYMBOL(jbd2__journal_start); |
490 | ||
491 | ||
91e4775d MCC |
492 | /** |
493 | * handle_t *jbd2_journal_start() - Obtain a new handle. | |
494 | * @journal: Journal to start transaction on. | |
495 | * @nblocks: number of block buffer we might modify | |
496 | * | |
497 | * We make sure that the transaction can guarantee at least nblocks of | |
498 | * modified buffers in the log. We block until the log can guarantee | |
499 | * that much space. Additionally, if rsv_blocks > 0, we also create another | |
500 | * handle with rsv_blocks reserved blocks in the journal. This handle is | |
501 | * is stored in h_rsv_handle. It is not attached to any particular transaction | |
502 | * and thus doesn't block transaction commit. If the caller uses this reserved | |
503 | * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop() | |
504 | * on the parent handle will dispose the reserved one. Reserved handle has to | |
505 | * be converted to a normal handle using jbd2_journal_start_reserved() before | |
506 | * it can be used. | |
507 | * | |
508 | * Return a pointer to a newly allocated handle, or an ERR_PTR() value | |
509 | * on failure. | |
510 | */ | |
47def826 TT |
511 | handle_t *jbd2_journal_start(journal_t *journal, int nblocks) |
512 | { | |
8f7d89f3 | 513 | return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0); |
47def826 TT |
514 | } |
515 | EXPORT_SYMBOL(jbd2_journal_start); | |
516 | ||
8f7d89f3 JK |
517 | void jbd2_journal_free_reserved(handle_t *handle) |
518 | { | |
519 | journal_t *journal = handle->h_journal; | |
520 | ||
521 | WARN_ON(!handle->h_reserved); | |
522 | sub_reserved_credits(journal, handle->h_buffer_credits); | |
523 | jbd2_free_handle(handle); | |
524 | } | |
525 | EXPORT_SYMBOL(jbd2_journal_free_reserved); | |
526 | ||
527 | /** | |
f69120ce | 528 | * int jbd2_journal_start_reserved() - start reserved handle |
8f7d89f3 | 529 | * @handle: handle to start |
f69120ce TH |
530 | * @type: for handle statistics |
531 | * @line_no: for handle statistics | |
8f7d89f3 JK |
532 | * |
533 | * Start handle that has been previously reserved with jbd2_journal_reserve(). | |
534 | * This attaches @handle to the running transaction (or creates one if there's | |
535 | * not transaction running). Unlike jbd2_journal_start() this function cannot | |
536 | * block on journal commit, checkpointing, or similar stuff. It can block on | |
537 | * memory allocation or frozen journal though. | |
538 | * | |
539 | * Return 0 on success, non-zero on error - handle is freed in that case. | |
540 | */ | |
541 | int jbd2_journal_start_reserved(handle_t *handle, unsigned int type, | |
542 | unsigned int line_no) | |
543 | { | |
544 | journal_t *journal = handle->h_journal; | |
545 | int ret = -EIO; | |
546 | ||
547 | if (WARN_ON(!handle->h_reserved)) { | |
548 | /* Someone passed in normal handle? Just stop it. */ | |
549 | jbd2_journal_stop(handle); | |
550 | return ret; | |
551 | } | |
552 | /* | |
553 | * Usefulness of mixing of reserved and unreserved handles is | |
554 | * questionable. So far nobody seems to need it so just error out. | |
555 | */ | |
556 | if (WARN_ON(current->journal_info)) { | |
557 | jbd2_journal_free_reserved(handle); | |
558 | return ret; | |
559 | } | |
560 | ||
561 | handle->h_journal = NULL; | |
8f7d89f3 JK |
562 | /* |
563 | * GFP_NOFS is here because callers are likely from writeback or | |
564 | * similarly constrained call sites | |
565 | */ | |
566 | ret = start_this_handle(journal, handle, GFP_NOFS); | |
92e3b405 | 567 | if (ret < 0) { |
b2569260 | 568 | handle->h_journal = journal; |
8f7d89f3 | 569 | jbd2_journal_free_reserved(handle); |
92e3b405 DC |
570 | return ret; |
571 | } | |
8f7d89f3 JK |
572 | handle->h_type = type; |
573 | handle->h_line_no = line_no; | |
4c273352 XW |
574 | trace_jbd2_handle_start(journal->j_fs_dev->bd_dev, |
575 | handle->h_transaction->t_tid, type, | |
576 | line_no, handle->h_buffer_credits); | |
92e3b405 | 577 | return 0; |
8f7d89f3 JK |
578 | } |
579 | EXPORT_SYMBOL(jbd2_journal_start_reserved); | |
470decc6 DK |
580 | |
581 | /** | |
f7f4bccb | 582 | * int jbd2_journal_extend() - extend buffer credits. |
470decc6 DK |
583 | * @handle: handle to 'extend' |
584 | * @nblocks: nr blocks to try to extend by. | |
585 | * | |
586 | * Some transactions, such as large extends and truncates, can be done | |
587 | * atomically all at once or in several stages. The operation requests | |
bd7ced98 | 588 | * a credit for a number of buffer modifications in advance, but can |
470decc6 DK |
589 | * extend its credit if it needs more. |
590 | * | |
f7f4bccb | 591 | * jbd2_journal_extend tries to give the running handle more buffer credits. |
470decc6 DK |
592 | * It does not guarantee that allocation - this is a best-effort only. |
593 | * The calling process MUST be able to deal cleanly with a failure to | |
594 | * extend here. | |
595 | * | |
596 | * Return 0 on success, non-zero on failure. | |
597 | * | |
598 | * return code < 0 implies an error | |
599 | * return code > 0 implies normal transaction-full status. | |
600 | */ | |
f7f4bccb | 601 | int jbd2_journal_extend(handle_t *handle, int nblocks) |
470decc6 DK |
602 | { |
603 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 604 | journal_t *journal; |
470decc6 DK |
605 | int result; |
606 | int wanted; | |
607 | ||
470decc6 | 608 | if (is_handle_aborted(handle)) |
41a5b913 TT |
609 | return -EROFS; |
610 | journal = transaction->t_journal; | |
470decc6 DK |
611 | |
612 | result = 1; | |
613 | ||
a931da6a | 614 | read_lock(&journal->j_state_lock); |
470decc6 DK |
615 | |
616 | /* Don't extend a locked-down transaction! */ | |
41a5b913 | 617 | if (transaction->t_state != T_RUNNING) { |
470decc6 DK |
618 | jbd_debug(3, "denied handle %p %d blocks: " |
619 | "transaction not running\n", handle, nblocks); | |
620 | goto error_out; | |
621 | } | |
622 | ||
623 | spin_lock(&transaction->t_handle_lock); | |
fe1e8db5 JK |
624 | wanted = atomic_add_return(nblocks, |
625 | &transaction->t_outstanding_credits); | |
470decc6 DK |
626 | |
627 | if (wanted > journal->j_max_transaction_buffers) { | |
628 | jbd_debug(3, "denied handle %p %d blocks: " | |
629 | "transaction too large\n", handle, nblocks); | |
fe1e8db5 | 630 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
470decc6 DK |
631 | goto unlock; |
632 | } | |
633 | ||
76c39904 JK |
634 | if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) > |
635 | jbd2_log_space_left(journal)) { | |
470decc6 DK |
636 | jbd_debug(3, "denied handle %p %d blocks: " |
637 | "insufficient log space\n", handle, nblocks); | |
fe1e8db5 | 638 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
470decc6 DK |
639 | goto unlock; |
640 | } | |
641 | ||
343d9c28 | 642 | trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev, |
41a5b913 | 643 | transaction->t_tid, |
343d9c28 TT |
644 | handle->h_type, handle->h_line_no, |
645 | handle->h_buffer_credits, | |
646 | nblocks); | |
647 | ||
470decc6 | 648 | handle->h_buffer_credits += nblocks; |
343d9c28 | 649 | handle->h_requested_credits += nblocks; |
470decc6 DK |
650 | result = 0; |
651 | ||
652 | jbd_debug(3, "extended handle %p by %d\n", handle, nblocks); | |
653 | unlock: | |
654 | spin_unlock(&transaction->t_handle_lock); | |
655 | error_out: | |
a931da6a | 656 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
657 | return result; |
658 | } | |
659 | ||
660 | ||
661 | /** | |
f7f4bccb | 662 | * int jbd2_journal_restart() - restart a handle . |
470decc6 DK |
663 | * @handle: handle to restart |
664 | * @nblocks: nr credits requested | |
f69120ce | 665 | * @gfp_mask: memory allocation flags (for start_this_handle) |
470decc6 DK |
666 | * |
667 | * Restart a handle for a multi-transaction filesystem | |
668 | * operation. | |
669 | * | |
f7f4bccb MC |
670 | * If the jbd2_journal_extend() call above fails to grant new buffer credits |
671 | * to a running handle, a call to jbd2_journal_restart will commit the | |
470decc6 | 672 | * handle's transaction so far and reattach the handle to a new |
bd7ced98 | 673 | * transaction capable of guaranteeing the requested number of |
8f7d89f3 JK |
674 | * credits. We preserve reserved handle if there's any attached to the |
675 | * passed in handle. | |
470decc6 | 676 | */ |
d2159fb7 | 677 | int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask) |
470decc6 DK |
678 | { |
679 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 680 | journal_t *journal; |
e4471831 TT |
681 | tid_t tid; |
682 | int need_to_start, ret; | |
470decc6 DK |
683 | |
684 | /* If we've had an abort of any type, don't even think about | |
685 | * actually doing the restart! */ | |
686 | if (is_handle_aborted(handle)) | |
687 | return 0; | |
41a5b913 | 688 | journal = transaction->t_journal; |
470decc6 DK |
689 | |
690 | /* | |
691 | * First unlink the handle from its current transaction, and start the | |
692 | * commit on that. | |
693 | */ | |
a51dca9c | 694 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 DK |
695 | J_ASSERT(journal_current_handle() == handle); |
696 | ||
a931da6a | 697 | read_lock(&journal->j_state_lock); |
470decc6 | 698 | spin_lock(&transaction->t_handle_lock); |
a51dca9c TT |
699 | atomic_sub(handle->h_buffer_credits, |
700 | &transaction->t_outstanding_credits); | |
8f7d89f3 JK |
701 | if (handle->h_rsv_handle) { |
702 | sub_reserved_credits(journal, | |
703 | handle->h_rsv_handle->h_buffer_credits); | |
704 | } | |
a51dca9c | 705 | if (atomic_dec_and_test(&transaction->t_updates)) |
470decc6 | 706 | wake_up(&journal->j_wait_updates); |
39c04153 | 707 | tid = transaction->t_tid; |
470decc6 | 708 | spin_unlock(&transaction->t_handle_lock); |
41a5b913 TT |
709 | handle->h_transaction = NULL; |
710 | current->journal_info = NULL; | |
470decc6 DK |
711 | |
712 | jbd_debug(2, "restarting handle %p\n", handle); | |
e4471831 | 713 | need_to_start = !tid_geq(journal->j_commit_request, tid); |
a931da6a | 714 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
715 | if (need_to_start) |
716 | jbd2_log_start_commit(journal, tid); | |
470decc6 | 717 | |
ab714aff | 718 | rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_); |
470decc6 | 719 | handle->h_buffer_credits = nblocks; |
b4709067 TE |
720 | /* |
721 | * Restore the original nofs context because the journal restart | |
722 | * is basically the same thing as journal stop and start. | |
723 | * start_this_handle will start a new nofs context. | |
724 | */ | |
725 | memalloc_nofs_restore(handle->saved_alloc_context); | |
47def826 | 726 | ret = start_this_handle(journal, handle, gfp_mask); |
470decc6 DK |
727 | return ret; |
728 | } | |
47def826 | 729 | EXPORT_SYMBOL(jbd2__journal_restart); |
470decc6 DK |
730 | |
731 | ||
47def826 TT |
732 | int jbd2_journal_restart(handle_t *handle, int nblocks) |
733 | { | |
734 | return jbd2__journal_restart(handle, nblocks, GFP_NOFS); | |
735 | } | |
736 | EXPORT_SYMBOL(jbd2_journal_restart); | |
737 | ||
470decc6 | 738 | /** |
f7f4bccb | 739 | * void jbd2_journal_lock_updates () - establish a transaction barrier. |
470decc6 DK |
740 | * @journal: Journal to establish a barrier on. |
741 | * | |
742 | * This locks out any further updates from being started, and blocks | |
743 | * until all existing updates have completed, returning only once the | |
744 | * journal is in a quiescent state with no updates running. | |
745 | * | |
746 | * The journal lock should not be held on entry. | |
747 | */ | |
f7f4bccb | 748 | void jbd2_journal_lock_updates(journal_t *journal) |
470decc6 DK |
749 | { |
750 | DEFINE_WAIT(wait); | |
751 | ||
1eaa566d JK |
752 | jbd2_might_wait_for_commit(journal); |
753 | ||
a931da6a | 754 | write_lock(&journal->j_state_lock); |
470decc6 DK |
755 | ++journal->j_barrier_count; |
756 | ||
8f7d89f3 JK |
757 | /* Wait until there are no reserved handles */ |
758 | if (atomic_read(&journal->j_reserved_credits)) { | |
759 | write_unlock(&journal->j_state_lock); | |
760 | wait_event(journal->j_wait_reserved, | |
761 | atomic_read(&journal->j_reserved_credits) == 0); | |
762 | write_lock(&journal->j_state_lock); | |
763 | } | |
764 | ||
470decc6 DK |
765 | /* Wait until there are no running updates */ |
766 | while (1) { | |
767 | transaction_t *transaction = journal->j_running_transaction; | |
768 | ||
769 | if (!transaction) | |
770 | break; | |
771 | ||
772 | spin_lock(&transaction->t_handle_lock); | |
9837d8e9 JK |
773 | prepare_to_wait(&journal->j_wait_updates, &wait, |
774 | TASK_UNINTERRUPTIBLE); | |
a51dca9c | 775 | if (!atomic_read(&transaction->t_updates)) { |
470decc6 | 776 | spin_unlock(&transaction->t_handle_lock); |
9837d8e9 | 777 | finish_wait(&journal->j_wait_updates, &wait); |
470decc6 DK |
778 | break; |
779 | } | |
470decc6 | 780 | spin_unlock(&transaction->t_handle_lock); |
a931da6a | 781 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
782 | schedule(); |
783 | finish_wait(&journal->j_wait_updates, &wait); | |
a931da6a | 784 | write_lock(&journal->j_state_lock); |
470decc6 | 785 | } |
a931da6a | 786 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
787 | |
788 | /* | |
789 | * We have now established a barrier against other normal updates, but | |
f7f4bccb | 790 | * we also need to barrier against other jbd2_journal_lock_updates() calls |
470decc6 DK |
791 | * to make sure that we serialise special journal-locked operations |
792 | * too. | |
793 | */ | |
794 | mutex_lock(&journal->j_barrier); | |
795 | } | |
796 | ||
797 | /** | |
f7f4bccb | 798 | * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier |
470decc6 DK |
799 | * @journal: Journal to release the barrier on. |
800 | * | |
f7f4bccb | 801 | * Release a transaction barrier obtained with jbd2_journal_lock_updates(). |
470decc6 DK |
802 | * |
803 | * Should be called without the journal lock held. | |
804 | */ | |
f7f4bccb | 805 | void jbd2_journal_unlock_updates (journal_t *journal) |
470decc6 DK |
806 | { |
807 | J_ASSERT(journal->j_barrier_count != 0); | |
808 | ||
809 | mutex_unlock(&journal->j_barrier); | |
a931da6a | 810 | write_lock(&journal->j_state_lock); |
470decc6 | 811 | --journal->j_barrier_count; |
a931da6a | 812 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
813 | wake_up(&journal->j_wait_transaction_locked); |
814 | } | |
815 | ||
f91d1d04 | 816 | static void warn_dirty_buffer(struct buffer_head *bh) |
470decc6 | 817 | { |
f91d1d04 | 818 | printk(KERN_WARNING |
a1c6f057 | 819 | "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). " |
f91d1d04 JK |
820 | "There's a risk of filesystem corruption in case of system " |
821 | "crash.\n", | |
a1c6f057 | 822 | bh->b_bdev, (unsigned long long)bh->b_blocknr); |
470decc6 DK |
823 | } |
824 | ||
ee57aba1 JK |
825 | /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */ |
826 | static void jbd2_freeze_jh_data(struct journal_head *jh) | |
827 | { | |
828 | struct page *page; | |
829 | int offset; | |
830 | char *source; | |
831 | struct buffer_head *bh = jh2bh(jh); | |
832 | ||
833 | J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n"); | |
834 | page = bh->b_page; | |
835 | offset = offset_in_page(bh->b_data); | |
836 | source = kmap_atomic(page); | |
837 | /* Fire data frozen trigger just before we copy the data */ | |
838 | jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers); | |
839 | memcpy(jh->b_frozen_data, source + offset, bh->b_size); | |
840 | kunmap_atomic(source); | |
841 | ||
842 | /* | |
843 | * Now that the frozen data is saved off, we need to store any matching | |
844 | * triggers. | |
845 | */ | |
846 | jh->b_frozen_triggers = jh->b_triggers; | |
847 | } | |
848 | ||
470decc6 DK |
849 | /* |
850 | * If the buffer is already part of the current transaction, then there | |
851 | * is nothing we need to do. If it is already part of a prior | |
852 | * transaction which we are still committing to disk, then we need to | |
853 | * make sure that we do not overwrite the old copy: we do copy-out to | |
854 | * preserve the copy going to disk. We also account the buffer against | |
855 | * the handle's metadata buffer credits (unless the buffer is already | |
856 | * part of the transaction, that is). | |
857 | * | |
858 | */ | |
859 | static int | |
860 | do_get_write_access(handle_t *handle, struct journal_head *jh, | |
861 | int force_copy) | |
862 | { | |
863 | struct buffer_head *bh; | |
41a5b913 | 864 | transaction_t *transaction = handle->h_transaction; |
470decc6 DK |
865 | journal_t *journal; |
866 | int error; | |
867 | char *frozen_buffer = NULL; | |
f783f091 | 868 | unsigned long start_lock, time_lock; |
470decc6 | 869 | |
470decc6 DK |
870 | journal = transaction->t_journal; |
871 | ||
cfef2c6a | 872 | jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy); |
470decc6 DK |
873 | |
874 | JBUFFER_TRACE(jh, "entry"); | |
875 | repeat: | |
876 | bh = jh2bh(jh); | |
877 | ||
878 | /* @@@ Need to check for errors here at some point. */ | |
879 | ||
f783f091 | 880 | start_lock = jiffies; |
470decc6 DK |
881 | lock_buffer(bh); |
882 | jbd_lock_bh_state(bh); | |
883 | ||
f783f091 TT |
884 | /* If it takes too long to lock the buffer, trace it */ |
885 | time_lock = jbd2_time_diff(start_lock, jiffies); | |
886 | if (time_lock > HZ/10) | |
887 | trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev, | |
888 | jiffies_to_msecs(time_lock)); | |
889 | ||
470decc6 DK |
890 | /* We now hold the buffer lock so it is safe to query the buffer |
891 | * state. Is the buffer dirty? | |
892 | * | |
893 | * If so, there are two possibilities. The buffer may be | |
894 | * non-journaled, and undergoing a quite legitimate writeback. | |
895 | * Otherwise, it is journaled, and we don't expect dirty buffers | |
896 | * in that state (the buffers should be marked JBD_Dirty | |
897 | * instead.) So either the IO is being done under our own | |
898 | * control and this is a bug, or it's a third party IO such as | |
899 | * dump(8) (which may leave the buffer scheduled for read --- | |
900 | * ie. locked but not dirty) or tune2fs (which may actually have | |
901 | * the buffer dirtied, ugh.) */ | |
902 | ||
903 | if (buffer_dirty(bh)) { | |
904 | /* | |
905 | * First question: is this buffer already part of the current | |
906 | * transaction or the existing committing transaction? | |
907 | */ | |
908 | if (jh->b_transaction) { | |
909 | J_ASSERT_JH(jh, | |
910 | jh->b_transaction == transaction || | |
911 | jh->b_transaction == | |
912 | journal->j_committing_transaction); | |
913 | if (jh->b_next_transaction) | |
914 | J_ASSERT_JH(jh, jh->b_next_transaction == | |
915 | transaction); | |
f91d1d04 | 916 | warn_dirty_buffer(bh); |
470decc6 DK |
917 | } |
918 | /* | |
919 | * In any case we need to clean the dirty flag and we must | |
920 | * do it under the buffer lock to be sure we don't race | |
921 | * with running write-out. | |
922 | */ | |
f91d1d04 JK |
923 | JBUFFER_TRACE(jh, "Journalling dirty buffer"); |
924 | clear_buffer_dirty(bh); | |
925 | set_buffer_jbddirty(bh); | |
470decc6 DK |
926 | } |
927 | ||
928 | unlock_buffer(bh); | |
929 | ||
930 | error = -EROFS; | |
931 | if (is_handle_aborted(handle)) { | |
932 | jbd_unlock_bh_state(bh); | |
933 | goto out; | |
934 | } | |
935 | error = 0; | |
936 | ||
937 | /* | |
938 | * The buffer is already part of this transaction if b_transaction or | |
939 | * b_next_transaction points to it | |
940 | */ | |
941 | if (jh->b_transaction == transaction || | |
942 | jh->b_next_transaction == transaction) | |
943 | goto done; | |
944 | ||
9fc7c63a JB |
945 | /* |
946 | * this is the first time this transaction is touching this buffer, | |
947 | * reset the modified flag | |
948 | */ | |
561405f0 | 949 | jh->b_modified = 0; |
9fc7c63a | 950 | |
8b00f400 JK |
951 | /* |
952 | * If the buffer is not journaled right now, we need to make sure it | |
953 | * doesn't get written to disk before the caller actually commits the | |
954 | * new data | |
955 | */ | |
956 | if (!jh->b_transaction) { | |
957 | JBUFFER_TRACE(jh, "no transaction"); | |
958 | J_ASSERT_JH(jh, !jh->b_next_transaction); | |
959 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); | |
de92c8ca JK |
960 | /* |
961 | * Make sure all stores to jh (b_modified, b_frozen_data) are | |
962 | * visible before attaching it to the running transaction. | |
963 | * Paired with barrier in jbd2_write_access_granted() | |
964 | */ | |
965 | smp_wmb(); | |
8b00f400 JK |
966 | spin_lock(&journal->j_list_lock); |
967 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); | |
968 | spin_unlock(&journal->j_list_lock); | |
969 | goto done; | |
970 | } | |
470decc6 DK |
971 | /* |
972 | * If there is already a copy-out version of this buffer, then we don't | |
973 | * need to make another one | |
974 | */ | |
975 | if (jh->b_frozen_data) { | |
976 | JBUFFER_TRACE(jh, "has frozen data"); | |
977 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
de92c8ca | 978 | goto attach_next; |
470decc6 DK |
979 | } |
980 | ||
8b00f400 JK |
981 | JBUFFER_TRACE(jh, "owned by older transaction"); |
982 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
983 | J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction); | |
470decc6 | 984 | |
8b00f400 JK |
985 | /* |
986 | * There is one case we have to be very careful about. If the | |
987 | * committing transaction is currently writing this buffer out to disk | |
988 | * and has NOT made a copy-out, then we cannot modify the buffer | |
989 | * contents at all right now. The essence of copy-out is that it is | |
990 | * the extra copy, not the primary copy, which gets journaled. If the | |
991 | * primary copy is already going to disk then we cannot do copy-out | |
992 | * here. | |
993 | */ | |
994 | if (buffer_shadow(bh)) { | |
995 | JBUFFER_TRACE(jh, "on shadow: sleep"); | |
996 | jbd_unlock_bh_state(bh); | |
997 | wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE); | |
998 | goto repeat; | |
999 | } | |
470decc6 | 1000 | |
8b00f400 JK |
1001 | /* |
1002 | * Only do the copy if the currently-owning transaction still needs it. | |
1003 | * If buffer isn't on BJ_Metadata list, the committing transaction is | |
1004 | * past that stage (here we use the fact that BH_Shadow is set under | |
1005 | * bh_state lock together with refiling to BJ_Shadow list and at this | |
1006 | * point we know the buffer doesn't have BH_Shadow set). | |
1007 | * | |
1008 | * Subtle point, though: if this is a get_undo_access, then we will be | |
1009 | * relying on the frozen_data to contain the new value of the | |
1010 | * committed_data record after the transaction, so we HAVE to force the | |
1011 | * frozen_data copy in that case. | |
1012 | */ | |
1013 | if (jh->b_jlist == BJ_Metadata || force_copy) { | |
1014 | JBUFFER_TRACE(jh, "generate frozen data"); | |
1015 | if (!frozen_buffer) { | |
1016 | JBUFFER_TRACE(jh, "allocate memory for buffer"); | |
470decc6 | 1017 | jbd_unlock_bh_state(bh); |
490c1b44 MH |
1018 | frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size, |
1019 | GFP_NOFS | __GFP_NOFAIL); | |
8b00f400 | 1020 | goto repeat; |
470decc6 | 1021 | } |
8b00f400 JK |
1022 | jh->b_frozen_data = frozen_buffer; |
1023 | frozen_buffer = NULL; | |
1024 | jbd2_freeze_jh_data(jh); | |
470decc6 | 1025 | } |
de92c8ca JK |
1026 | attach_next: |
1027 | /* | |
1028 | * Make sure all stores to jh (b_modified, b_frozen_data) are visible | |
1029 | * before attaching it to the running transaction. Paired with barrier | |
1030 | * in jbd2_write_access_granted() | |
1031 | */ | |
1032 | smp_wmb(); | |
8b00f400 | 1033 | jh->b_next_transaction = transaction; |
470decc6 DK |
1034 | |
1035 | done: | |
470decc6 DK |
1036 | jbd_unlock_bh_state(bh); |
1037 | ||
1038 | /* | |
1039 | * If we are about to journal a buffer, then any revoke pending on it is | |
1040 | * no longer valid | |
1041 | */ | |
f7f4bccb | 1042 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 DK |
1043 | |
1044 | out: | |
1045 | if (unlikely(frozen_buffer)) /* It's usually NULL */ | |
af1e76d6 | 1046 | jbd2_free(frozen_buffer, bh->b_size); |
470decc6 DK |
1047 | |
1048 | JBUFFER_TRACE(jh, "exit"); | |
1049 | return error; | |
1050 | } | |
1051 | ||
de92c8ca | 1052 | /* Fast check whether buffer is already attached to the required transaction */ |
087ffd4e JB |
1053 | static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh, |
1054 | bool undo) | |
de92c8ca JK |
1055 | { |
1056 | struct journal_head *jh; | |
1057 | bool ret = false; | |
1058 | ||
1059 | /* Dirty buffers require special handling... */ | |
1060 | if (buffer_dirty(bh)) | |
1061 | return false; | |
1062 | ||
1063 | /* | |
1064 | * RCU protects us from dereferencing freed pages. So the checks we do | |
1065 | * are guaranteed not to oops. However the jh slab object can get freed | |
1066 | * & reallocated while we work with it. So we have to be careful. When | |
1067 | * we see jh attached to the running transaction, we know it must stay | |
1068 | * so until the transaction is committed. Thus jh won't be freed and | |
1069 | * will be attached to the same bh while we run. However it can | |
1070 | * happen jh gets freed, reallocated, and attached to the transaction | |
1071 | * just after we get pointer to it from bh. So we have to be careful | |
1072 | * and recheck jh still belongs to our bh before we return success. | |
1073 | */ | |
1074 | rcu_read_lock(); | |
1075 | if (!buffer_jbd(bh)) | |
1076 | goto out; | |
1077 | /* This should be bh2jh() but that doesn't work with inline functions */ | |
1078 | jh = READ_ONCE(bh->b_private); | |
1079 | if (!jh) | |
1080 | goto out; | |
087ffd4e JB |
1081 | /* For undo access buffer must have data copied */ |
1082 | if (undo && !jh->b_committed_data) | |
1083 | goto out; | |
c33e585a QC |
1084 | if (READ_ONCE(jh->b_transaction) != handle->h_transaction && |
1085 | READ_ONCE(jh->b_next_transaction) != handle->h_transaction) | |
de92c8ca JK |
1086 | goto out; |
1087 | /* | |
1088 | * There are two reasons for the barrier here: | |
1089 | * 1) Make sure to fetch b_bh after we did previous checks so that we | |
1090 | * detect when jh went through free, realloc, attach to transaction | |
1091 | * while we were checking. Paired with implicit barrier in that path. | |
1092 | * 2) So that access to bh done after jbd2_write_access_granted() | |
1093 | * doesn't get reordered and see inconsistent state of concurrent | |
1094 | * do_get_write_access(). | |
1095 | */ | |
1096 | smp_mb(); | |
1097 | if (unlikely(jh->b_bh != bh)) | |
1098 | goto out; | |
1099 | ret = true; | |
1100 | out: | |
1101 | rcu_read_unlock(); | |
1102 | return ret; | |
1103 | } | |
1104 | ||
470decc6 | 1105 | /** |
f7f4bccb | 1106 | * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update. |
470decc6 DK |
1107 | * @handle: transaction to add buffer modifications to |
1108 | * @bh: bh to be used for metadata writes | |
470decc6 | 1109 | * |
df1b560a | 1110 | * Returns: error code or 0 on success. |
470decc6 DK |
1111 | * |
1112 | * In full data journalling mode the buffer may be of type BJ_AsyncData, | |
df1b560a | 1113 | * because we're ``write()ing`` a buffer which is also part of a shared mapping. |
470decc6 DK |
1114 | */ |
1115 | ||
f7f4bccb | 1116 | int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh) |
470decc6 | 1117 | { |
de92c8ca | 1118 | struct journal_head *jh; |
470decc6 DK |
1119 | int rc; |
1120 | ||
8a432f0e | 1121 | if (is_handle_aborted(handle)) |
1122 | return -EROFS; | |
1123 | ||
087ffd4e | 1124 | if (jbd2_write_access_granted(handle, bh, false)) |
de92c8ca JK |
1125 | return 0; |
1126 | ||
1127 | jh = jbd2_journal_add_journal_head(bh); | |
470decc6 DK |
1128 | /* We do not want to get caught playing with fields which the |
1129 | * log thread also manipulates. Make sure that the buffer | |
1130 | * completes any outstanding IO before proceeding. */ | |
1131 | rc = do_get_write_access(handle, jh, 0); | |
f7f4bccb | 1132 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1133 | return rc; |
1134 | } | |
1135 | ||
1136 | ||
1137 | /* | |
1138 | * When the user wants to journal a newly created buffer_head | |
1139 | * (ie. getblk() returned a new buffer and we are going to populate it | |
1140 | * manually rather than reading off disk), then we need to keep the | |
1141 | * buffer_head locked until it has been completely filled with new | |
1142 | * data. In this case, we should be able to make the assertion that | |
1143 | * the bh is not already part of an existing transaction. | |
1144 | * | |
1145 | * The buffer should already be locked by the caller by this point. | |
1146 | * There is no lock ranking violation: it was a newly created, | |
1147 | * unlocked buffer beforehand. */ | |
1148 | ||
1149 | /** | |
f7f4bccb | 1150 | * int jbd2_journal_get_create_access () - notify intent to use newly created bh |
470decc6 DK |
1151 | * @handle: transaction to new buffer to |
1152 | * @bh: new buffer. | |
1153 | * | |
1154 | * Call this if you create a new bh. | |
1155 | */ | |
f7f4bccb | 1156 | int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1157 | { |
1158 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 1159 | journal_t *journal; |
f7f4bccb | 1160 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
1161 | int err; |
1162 | ||
1163 | jbd_debug(5, "journal_head %p\n", jh); | |
1164 | err = -EROFS; | |
1165 | if (is_handle_aborted(handle)) | |
1166 | goto out; | |
41a5b913 | 1167 | journal = transaction->t_journal; |
470decc6 DK |
1168 | err = 0; |
1169 | ||
1170 | JBUFFER_TRACE(jh, "entry"); | |
1171 | /* | |
1172 | * The buffer may already belong to this transaction due to pre-zeroing | |
1173 | * in the filesystem's new_block code. It may also be on the previous, | |
1174 | * committing transaction's lists, but it HAS to be in Forget state in | |
1175 | * that case: the transaction must have deleted the buffer for it to be | |
1176 | * reused here. | |
1177 | */ | |
1178 | jbd_lock_bh_state(bh); | |
470decc6 DK |
1179 | J_ASSERT_JH(jh, (jh->b_transaction == transaction || |
1180 | jh->b_transaction == NULL || | |
1181 | (jh->b_transaction == journal->j_committing_transaction && | |
1182 | jh->b_jlist == BJ_Forget))); | |
1183 | ||
1184 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
1185 | J_ASSERT_JH(jh, buffer_locked(jh2bh(jh))); | |
1186 | ||
1187 | if (jh->b_transaction == NULL) { | |
f91d1d04 JK |
1188 | /* |
1189 | * Previous jbd2_journal_forget() could have left the buffer | |
1190 | * with jbddirty bit set because it was being committed. When | |
1191 | * the commit finished, we've filed the buffer for | |
1192 | * checkpointing and marked it dirty. Now we are reallocating | |
1193 | * the buffer so the transaction freeing it must have | |
1194 | * committed and so it's safe to clear the dirty bit. | |
1195 | */ | |
1196 | clear_buffer_dirty(jh2bh(jh)); | |
9fc7c63a JB |
1197 | /* first access by this transaction */ |
1198 | jh->b_modified = 0; | |
1199 | ||
470decc6 | 1200 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); |
6e4862a5 | 1201 | spin_lock(&journal->j_list_lock); |
f7f4bccb | 1202 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
559cce69 | 1203 | spin_unlock(&journal->j_list_lock); |
470decc6 | 1204 | } else if (jh->b_transaction == journal->j_committing_transaction) { |
9fc7c63a JB |
1205 | /* first access by this transaction */ |
1206 | jh->b_modified = 0; | |
1207 | ||
470decc6 | 1208 | JBUFFER_TRACE(jh, "set next transaction"); |
6e4862a5 | 1209 | spin_lock(&journal->j_list_lock); |
470decc6 | 1210 | jh->b_next_transaction = transaction; |
559cce69 | 1211 | spin_unlock(&journal->j_list_lock); |
470decc6 | 1212 | } |
470decc6 DK |
1213 | jbd_unlock_bh_state(bh); |
1214 | ||
1215 | /* | |
1216 | * akpm: I added this. ext3_alloc_branch can pick up new indirect | |
1217 | * blocks which contain freed but then revoked metadata. We need | |
1218 | * to cancel the revoke in case we end up freeing it yet again | |
1219 | * and the reallocating as data - this would cause a second revoke, | |
1220 | * which hits an assertion error. | |
1221 | */ | |
1222 | JBUFFER_TRACE(jh, "cancelling revoke"); | |
f7f4bccb | 1223 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 | 1224 | out: |
3991b400 | 1225 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1226 | return err; |
1227 | } | |
1228 | ||
1229 | /** | |
f7f4bccb | 1230 | * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with |
470decc6 DK |
1231 | * non-rewindable consequences |
1232 | * @handle: transaction | |
1233 | * @bh: buffer to undo | |
470decc6 DK |
1234 | * |
1235 | * Sometimes there is a need to distinguish between metadata which has | |
1236 | * been committed to disk and that which has not. The ext3fs code uses | |
1237 | * this for freeing and allocating space, we have to make sure that we | |
1238 | * do not reuse freed space until the deallocation has been committed, | |
1239 | * since if we overwrote that space we would make the delete | |
1240 | * un-rewindable in case of a crash. | |
1241 | * | |
f7f4bccb | 1242 | * To deal with that, jbd2_journal_get_undo_access requests write access to a |
470decc6 DK |
1243 | * buffer for parts of non-rewindable operations such as delete |
1244 | * operations on the bitmaps. The journaling code must keep a copy of | |
1245 | * the buffer's contents prior to the undo_access call until such time | |
1246 | * as we know that the buffer has definitely been committed to disk. | |
1247 | * | |
1248 | * We never need to know which transaction the committed data is part | |
1249 | * of, buffers touched here are guaranteed to be dirtied later and so | |
1250 | * will be committed to a new transaction in due course, at which point | |
1251 | * we can discard the old committed data pointer. | |
1252 | * | |
1253 | * Returns error number or 0 on success. | |
1254 | */ | |
f7f4bccb | 1255 | int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1256 | { |
1257 | int err; | |
de92c8ca | 1258 | struct journal_head *jh; |
470decc6 DK |
1259 | char *committed_data = NULL; |
1260 | ||
8a432f0e | 1261 | if (is_handle_aborted(handle)) |
1262 | return -EROFS; | |
1263 | ||
087ffd4e | 1264 | if (jbd2_write_access_granted(handle, bh, true)) |
de92c8ca | 1265 | return 0; |
470decc6 | 1266 | |
de92c8ca | 1267 | jh = jbd2_journal_add_journal_head(bh); |
01215d3e | 1268 | JBUFFER_TRACE(jh, "entry"); |
1269 | ||
470decc6 DK |
1270 | /* |
1271 | * Do this first --- it can drop the journal lock, so we want to | |
1272 | * make sure that obtaining the committed_data is done | |
1273 | * atomically wrt. completion of any outstanding commits. | |
1274 | */ | |
1275 | err = do_get_write_access(handle, jh, 1); | |
1276 | if (err) | |
1277 | goto out; | |
1278 | ||
1279 | repeat: | |
490c1b44 MH |
1280 | if (!jh->b_committed_data) |
1281 | committed_data = jbd2_alloc(jh2bh(jh)->b_size, | |
1282 | GFP_NOFS|__GFP_NOFAIL); | |
470decc6 DK |
1283 | |
1284 | jbd_lock_bh_state(bh); | |
1285 | if (!jh->b_committed_data) { | |
1286 | /* Copy out the current buffer contents into the | |
1287 | * preserved, committed copy. */ | |
1288 | JBUFFER_TRACE(jh, "generate b_committed data"); | |
1289 | if (!committed_data) { | |
1290 | jbd_unlock_bh_state(bh); | |
1291 | goto repeat; | |
1292 | } | |
1293 | ||
1294 | jh->b_committed_data = committed_data; | |
1295 | committed_data = NULL; | |
1296 | memcpy(jh->b_committed_data, bh->b_data, bh->b_size); | |
1297 | } | |
1298 | jbd_unlock_bh_state(bh); | |
1299 | out: | |
f7f4bccb | 1300 | jbd2_journal_put_journal_head(jh); |
470decc6 | 1301 | if (unlikely(committed_data)) |
af1e76d6 | 1302 | jbd2_free(committed_data, bh->b_size); |
470decc6 DK |
1303 | return err; |
1304 | } | |
1305 | ||
e06c8227 JB |
1306 | /** |
1307 | * void jbd2_journal_set_triggers() - Add triggers for commit writeout | |
1308 | * @bh: buffer to trigger on | |
1309 | * @type: struct jbd2_buffer_trigger_type containing the trigger(s). | |
1310 | * | |
1311 | * Set any triggers on this journal_head. This is always safe, because | |
1312 | * triggers for a committing buffer will be saved off, and triggers for | |
1313 | * a running transaction will match the buffer in that transaction. | |
1314 | * | |
1315 | * Call with NULL to clear the triggers. | |
1316 | */ | |
1317 | void jbd2_journal_set_triggers(struct buffer_head *bh, | |
1318 | struct jbd2_buffer_trigger_type *type) | |
1319 | { | |
ad56edad | 1320 | struct journal_head *jh = jbd2_journal_grab_journal_head(bh); |
e06c8227 | 1321 | |
ad56edad JK |
1322 | if (WARN_ON(!jh)) |
1323 | return; | |
e06c8227 | 1324 | jh->b_triggers = type; |
ad56edad | 1325 | jbd2_journal_put_journal_head(jh); |
e06c8227 JB |
1326 | } |
1327 | ||
13ceef09 | 1328 | void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data, |
e06c8227 JB |
1329 | struct jbd2_buffer_trigger_type *triggers) |
1330 | { | |
1331 | struct buffer_head *bh = jh2bh(jh); | |
1332 | ||
13ceef09 | 1333 | if (!triggers || !triggers->t_frozen) |
e06c8227 JB |
1334 | return; |
1335 | ||
13ceef09 | 1336 | triggers->t_frozen(triggers, bh, mapped_data, bh->b_size); |
e06c8227 JB |
1337 | } |
1338 | ||
1339 | void jbd2_buffer_abort_trigger(struct journal_head *jh, | |
1340 | struct jbd2_buffer_trigger_type *triggers) | |
1341 | { | |
1342 | if (!triggers || !triggers->t_abort) | |
1343 | return; | |
1344 | ||
1345 | triggers->t_abort(triggers, jh2bh(jh)); | |
1346 | } | |
1347 | ||
470decc6 | 1348 | /** |
f7f4bccb | 1349 | * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata |
470decc6 DK |
1350 | * @handle: transaction to add buffer to. |
1351 | * @bh: buffer to mark | |
1352 | * | |
1353 | * mark dirty metadata which needs to be journaled as part of the current | |
1354 | * transaction. | |
1355 | * | |
9ea7a0df TT |
1356 | * The buffer must have previously had jbd2_journal_get_write_access() |
1357 | * called so that it has a valid journal_head attached to the buffer | |
1358 | * head. | |
1359 | * | |
470decc6 DK |
1360 | * The buffer is placed on the transaction's metadata list and is marked |
1361 | * as belonging to the transaction. | |
1362 | * | |
1363 | * Returns error number or 0 on success. | |
1364 | * | |
1365 | * Special care needs to be taken if the buffer already belongs to the | |
1366 | * current committing transaction (in which case we should have frozen | |
1367 | * data present for that commit). In that case, we don't relink the | |
1368 | * buffer: that only gets done when the old transaction finally | |
1369 | * completes its commit. | |
1370 | */ | |
f7f4bccb | 1371 | int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1372 | { |
1373 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 1374 | journal_t *journal; |
ad56edad | 1375 | struct journal_head *jh; |
9ea7a0df | 1376 | int ret = 0; |
470decc6 | 1377 | |
470decc6 | 1378 | if (is_handle_aborted(handle)) |
41a5b913 | 1379 | return -EROFS; |
01215d3e | 1380 | if (!buffer_jbd(bh)) |
1381 | return -EUCLEAN; | |
1382 | ||
6e06ae88 JK |
1383 | /* |
1384 | * We don't grab jh reference here since the buffer must be part | |
1385 | * of the running transaction. | |
1386 | */ | |
1387 | jh = bh2jh(bh); | |
01215d3e | 1388 | jbd_debug(5, "journal_head %p\n", jh); |
1389 | JBUFFER_TRACE(jh, "entry"); | |
1390 | ||
6e06ae88 JK |
1391 | /* |
1392 | * This and the following assertions are unreliable since we may see jh | |
1393 | * in inconsistent state unless we grab bh_state lock. But this is | |
1394 | * crucial to catch bugs so let's do a reliable check until the | |
1395 | * lockless handling is fully proven. | |
1396 | */ | |
1397 | if (jh->b_transaction != transaction && | |
1398 | jh->b_next_transaction != transaction) { | |
1399 | jbd_lock_bh_state(bh); | |
1400 | J_ASSERT_JH(jh, jh->b_transaction == transaction || | |
1401 | jh->b_next_transaction == transaction); | |
1402 | jbd_unlock_bh_state(bh); | |
1403 | } | |
1404 | if (jh->b_modified == 1) { | |
1405 | /* If it's in our transaction it must be in BJ_Metadata list. */ | |
1406 | if (jh->b_transaction == transaction && | |
1407 | jh->b_jlist != BJ_Metadata) { | |
1408 | jbd_lock_bh_state(bh); | |
e09463f2 TT |
1409 | if (jh->b_transaction == transaction && |
1410 | jh->b_jlist != BJ_Metadata) | |
1411 | pr_err("JBD2: assertion failure: h_type=%u " | |
1412 | "h_line_no=%u block_no=%llu jlist=%u\n", | |
1413 | handle->h_type, handle->h_line_no, | |
1414 | (unsigned long long) bh->b_blocknr, | |
1415 | jh->b_jlist); | |
6e06ae88 JK |
1416 | J_ASSERT_JH(jh, jh->b_transaction != transaction || |
1417 | jh->b_jlist == BJ_Metadata); | |
1418 | jbd_unlock_bh_state(bh); | |
1419 | } | |
1420 | goto out; | |
1421 | } | |
1422 | ||
1423 | journal = transaction->t_journal; | |
470decc6 DK |
1424 | jbd_lock_bh_state(bh); |
1425 | ||
1426 | if (jh->b_modified == 0) { | |
1427 | /* | |
1428 | * This buffer's got modified and becoming part | |
1429 | * of the transaction. This needs to be done | |
1430 | * once a transaction -bzzz | |
1431 | */ | |
f6c07cad TT |
1432 | if (handle->h_buffer_credits <= 0) { |
1433 | ret = -ENOSPC; | |
1434 | goto out_unlock_bh; | |
1435 | } | |
e09463f2 | 1436 | jh->b_modified = 1; |
470decc6 DK |
1437 | handle->h_buffer_credits--; |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * fastpath, to avoid expensive locking. If this buffer is already | |
1442 | * on the running transaction's metadata list there is nothing to do. | |
1443 | * Nobody can take it off again because there is a handle open. | |
1444 | * I _think_ we're OK here with SMP barriers - a mistaken decision will | |
1445 | * result in this test being false, so we go in and take the locks. | |
1446 | */ | |
1447 | if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) { | |
1448 | JBUFFER_TRACE(jh, "fastpath"); | |
9ea7a0df TT |
1449 | if (unlikely(jh->b_transaction != |
1450 | journal->j_running_transaction)) { | |
a67c848a | 1451 | printk(KERN_ERR "JBD2: %s: " |
9ea7a0df | 1452 | "jh->b_transaction (%llu, %p, %u) != " |
66a4cb18 | 1453 | "journal->j_running_transaction (%p, %u)\n", |
9ea7a0df TT |
1454 | journal->j_devname, |
1455 | (unsigned long long) bh->b_blocknr, | |
1456 | jh->b_transaction, | |
1457 | jh->b_transaction ? jh->b_transaction->t_tid : 0, | |
1458 | journal->j_running_transaction, | |
1459 | journal->j_running_transaction ? | |
1460 | journal->j_running_transaction->t_tid : 0); | |
1461 | ret = -EINVAL; | |
1462 | } | |
470decc6 DK |
1463 | goto out_unlock_bh; |
1464 | } | |
1465 | ||
1466 | set_buffer_jbddirty(bh); | |
1467 | ||
1468 | /* | |
1469 | * Metadata already on the current transaction list doesn't | |
1470 | * need to be filed. Metadata on another transaction's list must | |
1471 | * be committing, and will be refiled once the commit completes: | |
1472 | * leave it alone for now. | |
1473 | */ | |
1474 | if (jh->b_transaction != transaction) { | |
1475 | JBUFFER_TRACE(jh, "already on other transaction"); | |
66a4cb18 TT |
1476 | if (unlikely(((jh->b_transaction != |
1477 | journal->j_committing_transaction)) || | |
1478 | (jh->b_next_transaction != transaction))) { | |
1479 | printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: " | |
1480 | "bad jh for block %llu: " | |
1481 | "transaction (%p, %u), " | |
1482 | "jh->b_transaction (%p, %u), " | |
1483 | "jh->b_next_transaction (%p, %u), jlist %u\n", | |
9ea7a0df TT |
1484 | journal->j_devname, |
1485 | (unsigned long long) bh->b_blocknr, | |
66a4cb18 | 1486 | transaction, transaction->t_tid, |
9ea7a0df | 1487 | jh->b_transaction, |
66a4cb18 TT |
1488 | jh->b_transaction ? |
1489 | jh->b_transaction->t_tid : 0, | |
9ea7a0df TT |
1490 | jh->b_next_transaction, |
1491 | jh->b_next_transaction ? | |
1492 | jh->b_next_transaction->t_tid : 0, | |
66a4cb18 TT |
1493 | jh->b_jlist); |
1494 | WARN_ON(1); | |
9ea7a0df TT |
1495 | ret = -EINVAL; |
1496 | } | |
470decc6 DK |
1497 | /* And this case is illegal: we can't reuse another |
1498 | * transaction's data buffer, ever. */ | |
1499 | goto out_unlock_bh; | |
1500 | } | |
1501 | ||
1502 | /* That test should have eliminated the following case: */ | |
4019191b | 1503 | J_ASSERT_JH(jh, jh->b_frozen_data == NULL); |
470decc6 DK |
1504 | |
1505 | JBUFFER_TRACE(jh, "file as BJ_Metadata"); | |
1506 | spin_lock(&journal->j_list_lock); | |
41a5b913 | 1507 | __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata); |
470decc6 DK |
1508 | spin_unlock(&journal->j_list_lock); |
1509 | out_unlock_bh: | |
1510 | jbd_unlock_bh_state(bh); | |
1511 | out: | |
1512 | JBUFFER_TRACE(jh, "exit"); | |
9ea7a0df | 1513 | return ret; |
470decc6 DK |
1514 | } |
1515 | ||
470decc6 | 1516 | /** |
f7f4bccb | 1517 | * void jbd2_journal_forget() - bforget() for potentially-journaled buffers. |
470decc6 DK |
1518 | * @handle: transaction handle |
1519 | * @bh: bh to 'forget' | |
1520 | * | |
1521 | * We can only do the bforget if there are no commits pending against the | |
1522 | * buffer. If the buffer is dirty in the current running transaction we | |
1523 | * can safely unlink it. | |
1524 | * | |
1525 | * bh may not be a journalled buffer at all - it may be a non-JBD | |
1526 | * buffer which came off the hashtable. Check for this. | |
1527 | * | |
1528 | * Decrements bh->b_count by one. | |
1529 | * | |
1530 | * Allow this call even if the handle has aborted --- it may be part of | |
1531 | * the caller's cleanup after an abort. | |
1532 | */ | |
f7f4bccb | 1533 | int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1534 | { |
1535 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 1536 | journal_t *journal; |
470decc6 DK |
1537 | struct journal_head *jh; |
1538 | int drop_reserve = 0; | |
1539 | int err = 0; | |
1dfc3220 | 1540 | int was_modified = 0; |
470decc6 | 1541 | |
41a5b913 TT |
1542 | if (is_handle_aborted(handle)) |
1543 | return -EROFS; | |
1544 | journal = transaction->t_journal; | |
1545 | ||
470decc6 DK |
1546 | BUFFER_TRACE(bh, "entry"); |
1547 | ||
1548 | jbd_lock_bh_state(bh); | |
470decc6 DK |
1549 | |
1550 | if (!buffer_jbd(bh)) | |
1551 | goto not_jbd; | |
1552 | jh = bh2jh(bh); | |
1553 | ||
1554 | /* Critical error: attempting to delete a bitmap buffer, maybe? | |
1555 | * Don't do any jbd operations, and return an error. */ | |
1556 | if (!J_EXPECT_JH(jh, !jh->b_committed_data, | |
1557 | "inconsistent data on disk")) { | |
1558 | err = -EIO; | |
1559 | goto not_jbd; | |
1560 | } | |
1561 | ||
48fc7f7e | 1562 | /* keep track of whether or not this transaction modified us */ |
1dfc3220 JB |
1563 | was_modified = jh->b_modified; |
1564 | ||
470decc6 DK |
1565 | /* |
1566 | * The buffer's going from the transaction, we must drop | |
1567 | * all references -bzzz | |
1568 | */ | |
1569 | jh->b_modified = 0; | |
1570 | ||
41a5b913 | 1571 | if (jh->b_transaction == transaction) { |
470decc6 DK |
1572 | J_ASSERT_JH(jh, !jh->b_frozen_data); |
1573 | ||
1574 | /* If we are forgetting a buffer which is already part | |
1575 | * of this transaction, then we can just drop it from | |
1576 | * the transaction immediately. */ | |
1577 | clear_buffer_dirty(bh); | |
1578 | clear_buffer_jbddirty(bh); | |
1579 | ||
1580 | JBUFFER_TRACE(jh, "belongs to current transaction: unfile"); | |
1581 | ||
1dfc3220 JB |
1582 | /* |
1583 | * we only want to drop a reference if this transaction | |
1584 | * modified the buffer | |
1585 | */ | |
1586 | if (was_modified) | |
1587 | drop_reserve = 1; | |
470decc6 DK |
1588 | |
1589 | /* | |
1590 | * We are no longer going to journal this buffer. | |
1591 | * However, the commit of this transaction is still | |
1592 | * important to the buffer: the delete that we are now | |
1593 | * processing might obsolete an old log entry, so by | |
1594 | * committing, we can satisfy the buffer's checkpoint. | |
1595 | * | |
1596 | * So, if we have a checkpoint on the buffer, we should | |
1597 | * now refile the buffer on our BJ_Forget list so that | |
1598 | * we know to remove the checkpoint after we commit. | |
1599 | */ | |
1600 | ||
0bfea811 | 1601 | spin_lock(&journal->j_list_lock); |
470decc6 | 1602 | if (jh->b_cp_transaction) { |
f7f4bccb MC |
1603 | __jbd2_journal_temp_unlink_buffer(jh); |
1604 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); | |
470decc6 | 1605 | } else { |
f7f4bccb | 1606 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
1607 | if (!buffer_jbd(bh)) { |
1608 | spin_unlock(&journal->j_list_lock); | |
59759926 | 1609 | goto not_jbd; |
470decc6 DK |
1610 | } |
1611 | } | |
0bfea811 | 1612 | spin_unlock(&journal->j_list_lock); |
470decc6 DK |
1613 | } else if (jh->b_transaction) { |
1614 | J_ASSERT_JH(jh, (jh->b_transaction == | |
1615 | journal->j_committing_transaction)); | |
1616 | /* However, if the buffer is still owned by a prior | |
1617 | * (committing) transaction, we can't drop it yet... */ | |
1618 | JBUFFER_TRACE(jh, "belongs to older transaction"); | |
904cdbd4 | 1619 | /* ... but we CAN drop it from the new transaction through |
1620 | * marking the buffer as freed and set j_next_transaction to | |
1621 | * the new transaction, so that not only the commit code | |
1622 | * knows it should clear dirty bits when it is done with the | |
1623 | * buffer, but also the buffer can be checkpointed only | |
1624 | * after the new transaction commits. */ | |
470decc6 | 1625 | |
904cdbd4 | 1626 | set_buffer_freed(bh); |
1627 | ||
1628 | if (!jh->b_next_transaction) { | |
0bfea811 | 1629 | spin_lock(&journal->j_list_lock); |
904cdbd4 | 1630 | jh->b_next_transaction = transaction; |
0bfea811 | 1631 | spin_unlock(&journal->j_list_lock); |
904cdbd4 | 1632 | } else { |
1633 | J_ASSERT(jh->b_next_transaction == transaction); | |
1dfc3220 JB |
1634 | |
1635 | /* | |
1636 | * only drop a reference if this transaction modified | |
1637 | * the buffer | |
1638 | */ | |
1639 | if (was_modified) | |
1640 | drop_reserve = 1; | |
470decc6 | 1641 | } |
59759926 | 1642 | } else { |
1643 | /* | |
1644 | * Finally, if the buffer is not belongs to any | |
1645 | * transaction, we can just drop it now if it has no | |
1646 | * checkpoint. | |
1647 | */ | |
1648 | spin_lock(&journal->j_list_lock); | |
1649 | if (!jh->b_cp_transaction) { | |
1650 | JBUFFER_TRACE(jh, "belongs to none transaction"); | |
1651 | spin_unlock(&journal->j_list_lock); | |
1652 | goto not_jbd; | |
1653 | } | |
1654 | ||
1655 | /* | |
1656 | * Otherwise, if the buffer has been written to disk, | |
1657 | * it is safe to remove the checkpoint and drop it. | |
1658 | */ | |
1659 | if (!buffer_dirty(bh)) { | |
1660 | __jbd2_journal_remove_checkpoint(jh); | |
1661 | spin_unlock(&journal->j_list_lock); | |
1662 | goto not_jbd; | |
1663 | } | |
1664 | ||
1665 | /* | |
1666 | * The buffer is still not written to disk, we should | |
1667 | * attach this buffer to current transaction so that the | |
1668 | * buffer can be checkpointed only after the current | |
1669 | * transaction commits. | |
1670 | */ | |
1671 | clear_buffer_dirty(bh); | |
1672 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); | |
1673 | spin_unlock(&journal->j_list_lock); | |
470decc6 DK |
1674 | } |
1675 | ||
470decc6 DK |
1676 | jbd_unlock_bh_state(bh); |
1677 | __brelse(bh); | |
1678 | drop: | |
1679 | if (drop_reserve) { | |
1680 | /* no need to reserve log space for this block -bzzz */ | |
1681 | handle->h_buffer_credits++; | |
1682 | } | |
1683 | return err; | |
59759926 | 1684 | |
1685 | not_jbd: | |
1686 | jbd_unlock_bh_state(bh); | |
1687 | __bforget(bh); | |
1688 | goto drop; | |
470decc6 DK |
1689 | } |
1690 | ||
1691 | /** | |
f7f4bccb | 1692 | * int jbd2_journal_stop() - complete a transaction |
bd7ced98 | 1693 | * @handle: transaction to complete. |
470decc6 DK |
1694 | * |
1695 | * All done for a particular handle. | |
1696 | * | |
1697 | * There is not much action needed here. We just return any remaining | |
1698 | * buffer credits to the transaction and remove the handle. The only | |
1699 | * complication is that we need to start a commit operation if the | |
1700 | * filesystem is marked for synchronous update. | |
1701 | * | |
f7f4bccb | 1702 | * jbd2_journal_stop itself will not usually return an error, but it may |
470decc6 | 1703 | * do so in unusual circumstances. In particular, expect it to |
f7f4bccb | 1704 | * return -EIO if a jbd2_journal_abort has been executed since the |
470decc6 DK |
1705 | * transaction began. |
1706 | */ | |
f7f4bccb | 1707 | int jbd2_journal_stop(handle_t *handle) |
470decc6 DK |
1708 | { |
1709 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 TT |
1710 | journal_t *journal; |
1711 | int err = 0, wait_for_commit = 0; | |
a51dca9c | 1712 | tid_t tid; |
470decc6 DK |
1713 | pid_t pid; |
1714 | ||
9d506594 LC |
1715 | if (!transaction) { |
1716 | /* | |
1717 | * Handle is already detached from the transaction so | |
1718 | * there is nothing to do other than decrease a refcount, | |
1719 | * or free the handle if refcount drops to zero | |
1720 | */ | |
1721 | if (--handle->h_ref > 0) { | |
1722 | jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1, | |
1723 | handle->h_ref); | |
1724 | return err; | |
1725 | } else { | |
1726 | if (handle->h_rsv_handle) | |
1727 | jbd2_free_handle(handle->h_rsv_handle); | |
1728 | goto free_and_exit; | |
1729 | } | |
1730 | } | |
41a5b913 TT |
1731 | journal = transaction->t_journal; |
1732 | ||
470decc6 DK |
1733 | J_ASSERT(journal_current_handle() == handle); |
1734 | ||
1735 | if (is_handle_aborted(handle)) | |
1736 | err = -EIO; | |
41a5b913 | 1737 | else |
a51dca9c | 1738 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); |
470decc6 DK |
1739 | |
1740 | if (--handle->h_ref > 0) { | |
1741 | jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1, | |
1742 | handle->h_ref); | |
1743 | return err; | |
1744 | } | |
1745 | ||
1746 | jbd_debug(4, "Handle %p going down\n", handle); | |
343d9c28 | 1747 | trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev, |
41a5b913 | 1748 | transaction->t_tid, |
343d9c28 TT |
1749 | handle->h_type, handle->h_line_no, |
1750 | jiffies - handle->h_start_jiffies, | |
1751 | handle->h_sync, handle->h_requested_credits, | |
1752 | (handle->h_requested_credits - | |
1753 | handle->h_buffer_credits)); | |
470decc6 DK |
1754 | |
1755 | /* | |
1756 | * Implement synchronous transaction batching. If the handle | |
1757 | * was synchronous, don't force a commit immediately. Let's | |
e07f7183 JB |
1758 | * yield and let another thread piggyback onto this |
1759 | * transaction. Keep doing that while new threads continue to | |
1760 | * arrive. It doesn't cost much - we're about to run a commit | |
1761 | * and sleep on IO anyway. Speeds up many-threaded, many-dir | |
1762 | * operations by 30x or more... | |
1763 | * | |
1764 | * We try and optimize the sleep time against what the | |
1765 | * underlying disk can do, instead of having a static sleep | |
1766 | * time. This is useful for the case where our storage is so | |
1767 | * fast that it is more optimal to go ahead and force a flush | |
1768 | * and wait for the transaction to be committed than it is to | |
1769 | * wait for an arbitrary amount of time for new writers to | |
1770 | * join the transaction. We achieve this by measuring how | |
1771 | * long it takes to commit a transaction, and compare it with | |
1772 | * how long this transaction has been running, and if run time | |
1773 | * < commit time then we sleep for the delta and commit. This | |
1774 | * greatly helps super fast disks that would see slowdowns as | |
1775 | * more threads started doing fsyncs. | |
470decc6 | 1776 | * |
e07f7183 JB |
1777 | * But don't do this if this process was the most recent one |
1778 | * to perform a synchronous write. We do this to detect the | |
1779 | * case where a single process is doing a stream of sync | |
1780 | * writes. No point in waiting for joiners in that case. | |
5dd21424 ES |
1781 | * |
1782 | * Setting max_batch_time to 0 disables this completely. | |
470decc6 DK |
1783 | */ |
1784 | pid = current->pid; | |
5dd21424 ES |
1785 | if (handle->h_sync && journal->j_last_sync_writer != pid && |
1786 | journal->j_max_batch_time) { | |
e07f7183 JB |
1787 | u64 commit_time, trans_time; |
1788 | ||
470decc6 | 1789 | journal->j_last_sync_writer = pid; |
e07f7183 | 1790 | |
a931da6a | 1791 | read_lock(&journal->j_state_lock); |
e07f7183 | 1792 | commit_time = journal->j_average_commit_time; |
a931da6a | 1793 | read_unlock(&journal->j_state_lock); |
e07f7183 JB |
1794 | |
1795 | trans_time = ktime_to_ns(ktime_sub(ktime_get(), | |
1796 | transaction->t_start_time)); | |
1797 | ||
30773840 TT |
1798 | commit_time = max_t(u64, commit_time, |
1799 | 1000*journal->j_min_batch_time); | |
e07f7183 | 1800 | commit_time = min_t(u64, commit_time, |
30773840 | 1801 | 1000*journal->j_max_batch_time); |
e07f7183 JB |
1802 | |
1803 | if (trans_time < commit_time) { | |
1804 | ktime_t expires = ktime_add_ns(ktime_get(), | |
1805 | commit_time); | |
1806 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1807 | schedule_hrtimeout(&expires, HRTIMER_MODE_ABS); | |
1808 | } | |
470decc6 DK |
1809 | } |
1810 | ||
7058548c TT |
1811 | if (handle->h_sync) |
1812 | transaction->t_synchronous_commit = 1; | |
470decc6 | 1813 | current->journal_info = NULL; |
a51dca9c TT |
1814 | atomic_sub(handle->h_buffer_credits, |
1815 | &transaction->t_outstanding_credits); | |
470decc6 DK |
1816 | |
1817 | /* | |
1818 | * If the handle is marked SYNC, we need to set another commit | |
1819 | * going! We also want to force a commit if the current | |
1820 | * transaction is occupying too much of the log, or if the | |
1821 | * transaction is too old now. | |
1822 | */ | |
1823 | if (handle->h_sync || | |
a51dca9c TT |
1824 | (atomic_read(&transaction->t_outstanding_credits) > |
1825 | journal->j_max_transaction_buffers) || | |
1826 | time_after_eq(jiffies, transaction->t_expires)) { | |
470decc6 DK |
1827 | /* Do this even for aborted journals: an abort still |
1828 | * completes the commit thread, it just doesn't write | |
1829 | * anything to disk. */ | |
470decc6 | 1830 | |
470decc6 DK |
1831 | jbd_debug(2, "transaction too old, requesting commit for " |
1832 | "handle %p\n", handle); | |
1833 | /* This is non-blocking */ | |
c35a56a0 | 1834 | jbd2_log_start_commit(journal, transaction->t_tid); |
470decc6 DK |
1835 | |
1836 | /* | |
f7f4bccb | 1837 | * Special case: JBD2_SYNC synchronous updates require us |
470decc6 DK |
1838 | * to wait for the commit to complete. |
1839 | */ | |
1840 | if (handle->h_sync && !(current->flags & PF_MEMALLOC)) | |
a51dca9c | 1841 | wait_for_commit = 1; |
470decc6 DK |
1842 | } |
1843 | ||
a51dca9c TT |
1844 | /* |
1845 | * Once we drop t_updates, if it goes to zero the transaction | |
25985edc | 1846 | * could start committing on us and eventually disappear. So |
a51dca9c TT |
1847 | * once we do this, we must not dereference transaction |
1848 | * pointer again. | |
1849 | */ | |
1850 | tid = transaction->t_tid; | |
1851 | if (atomic_dec_and_test(&transaction->t_updates)) { | |
1852 | wake_up(&journal->j_wait_updates); | |
1853 | if (journal->j_barrier_count) | |
1854 | wake_up(&journal->j_wait_transaction_locked); | |
1855 | } | |
1856 | ||
ab714aff | 1857 | rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_); |
7a4b188f | 1858 | |
a51dca9c TT |
1859 | if (wait_for_commit) |
1860 | err = jbd2_log_wait_commit(journal, tid); | |
1861 | ||
8f7d89f3 JK |
1862 | if (handle->h_rsv_handle) |
1863 | jbd2_journal_free_reserved(handle->h_rsv_handle); | |
41a5b913 | 1864 | free_and_exit: |
81378da6 MH |
1865 | /* |
1866 | * Scope of the GFP_NOFS context is over here and so we can restore the | |
1867 | * original alloc context. | |
1868 | */ | |
1869 | memalloc_nofs_restore(handle->saved_alloc_context); | |
af1e76d6 | 1870 | jbd2_free_handle(handle); |
470decc6 DK |
1871 | return err; |
1872 | } | |
1873 | ||
470decc6 DK |
1874 | /* |
1875 | * | |
1876 | * List management code snippets: various functions for manipulating the | |
1877 | * transaction buffer lists. | |
1878 | * | |
1879 | */ | |
1880 | ||
1881 | /* | |
1882 | * Append a buffer to a transaction list, given the transaction's list head | |
1883 | * pointer. | |
1884 | * | |
1885 | * j_list_lock is held. | |
1886 | * | |
1887 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1888 | */ | |
1889 | ||
1890 | static inline void | |
1891 | __blist_add_buffer(struct journal_head **list, struct journal_head *jh) | |
1892 | { | |
1893 | if (!*list) { | |
1894 | jh->b_tnext = jh->b_tprev = jh; | |
1895 | *list = jh; | |
1896 | } else { | |
1897 | /* Insert at the tail of the list to preserve order */ | |
1898 | struct journal_head *first = *list, *last = first->b_tprev; | |
1899 | jh->b_tprev = last; | |
1900 | jh->b_tnext = first; | |
1901 | last->b_tnext = first->b_tprev = jh; | |
1902 | } | |
1903 | } | |
1904 | ||
1905 | /* | |
1906 | * Remove a buffer from a transaction list, given the transaction's list | |
1907 | * head pointer. | |
1908 | * | |
1909 | * Called with j_list_lock held, and the journal may not be locked. | |
1910 | * | |
1911 | * jbd_lock_bh_state(jh2bh(jh)) is held. | |
1912 | */ | |
1913 | ||
1914 | static inline void | |
1915 | __blist_del_buffer(struct journal_head **list, struct journal_head *jh) | |
1916 | { | |
1917 | if (*list == jh) { | |
1918 | *list = jh->b_tnext; | |
1919 | if (*list == jh) | |
1920 | *list = NULL; | |
1921 | } | |
1922 | jh->b_tprev->b_tnext = jh->b_tnext; | |
1923 | jh->b_tnext->b_tprev = jh->b_tprev; | |
1924 | } | |
1925 | ||
1926 | /* | |
1927 | * Remove a buffer from the appropriate transaction list. | |
1928 | * | |
1929 | * Note that this function can *change* the value of | |
f5113eff JK |
1930 | * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or |
1931 | * t_reserved_list. If the caller is holding onto a copy of one of these | |
1932 | * pointers, it could go bad. Generally the caller needs to re-read the | |
1933 | * pointer from the transaction_t. | |
470decc6 | 1934 | * |
5bebccf9 | 1935 | * Called under j_list_lock. |
470decc6 | 1936 | */ |
5bebccf9 | 1937 | static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh) |
470decc6 DK |
1938 | { |
1939 | struct journal_head **list = NULL; | |
1940 | transaction_t *transaction; | |
1941 | struct buffer_head *bh = jh2bh(jh); | |
1942 | ||
1943 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
1944 | transaction = jh->b_transaction; | |
1945 | if (transaction) | |
1946 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
1947 | ||
1948 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
1949 | if (jh->b_jlist != BJ_None) | |
4019191b | 1950 | J_ASSERT_JH(jh, transaction != NULL); |
470decc6 DK |
1951 | |
1952 | switch (jh->b_jlist) { | |
1953 | case BJ_None: | |
1954 | return; | |
470decc6 DK |
1955 | case BJ_Metadata: |
1956 | transaction->t_nr_buffers--; | |
1957 | J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0); | |
1958 | list = &transaction->t_buffers; | |
1959 | break; | |
1960 | case BJ_Forget: | |
1961 | list = &transaction->t_forget; | |
1962 | break; | |
470decc6 DK |
1963 | case BJ_Shadow: |
1964 | list = &transaction->t_shadow_list; | |
1965 | break; | |
470decc6 DK |
1966 | case BJ_Reserved: |
1967 | list = &transaction->t_reserved_list; | |
1968 | break; | |
470decc6 DK |
1969 | } |
1970 | ||
1971 | __blist_del_buffer(list, jh); | |
1972 | jh->b_jlist = BJ_None; | |
e112666b TT |
1973 | if (transaction && is_journal_aborted(transaction->t_journal)) |
1974 | clear_buffer_jbddirty(bh); | |
1975 | else if (test_clear_buffer_jbddirty(bh)) | |
470decc6 DK |
1976 | mark_buffer_dirty(bh); /* Expose it to the VM */ |
1977 | } | |
1978 | ||
de1b7941 JK |
1979 | /* |
1980 | * Remove buffer from all transactions. | |
1981 | * | |
1982 | * Called with bh_state lock and j_list_lock | |
1983 | * | |
1984 | * jh and bh may be already freed when this function returns. | |
1985 | */ | |
1986 | static void __jbd2_journal_unfile_buffer(struct journal_head *jh) | |
470decc6 | 1987 | { |
412c5828 LC |
1988 | J_ASSERT_JH(jh, jh->b_transaction != NULL); |
1989 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
1990 | ||
f7f4bccb | 1991 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 | 1992 | jh->b_transaction = NULL; |
de1b7941 | 1993 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1994 | } |
1995 | ||
f7f4bccb | 1996 | void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 | 1997 | { |
de1b7941 JK |
1998 | struct buffer_head *bh = jh2bh(jh); |
1999 | ||
2000 | /* Get reference so that buffer cannot be freed before we unlock it */ | |
2001 | get_bh(bh); | |
2002 | jbd_lock_bh_state(bh); | |
470decc6 | 2003 | spin_lock(&journal->j_list_lock); |
f7f4bccb | 2004 | __jbd2_journal_unfile_buffer(jh); |
470decc6 | 2005 | spin_unlock(&journal->j_list_lock); |
de1b7941 JK |
2006 | jbd_unlock_bh_state(bh); |
2007 | __brelse(bh); | |
470decc6 DK |
2008 | } |
2009 | ||
2010 | /* | |
f7f4bccb | 2011 | * Called from jbd2_journal_try_to_free_buffers(). |
470decc6 DK |
2012 | * |
2013 | * Called under jbd_lock_bh_state(bh) | |
2014 | */ | |
2015 | static void | |
2016 | __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh) | |
2017 | { | |
2018 | struct journal_head *jh; | |
2019 | ||
2020 | jh = bh2jh(bh); | |
2021 | ||
2022 | if (buffer_locked(bh) || buffer_dirty(bh)) | |
2023 | goto out; | |
2024 | ||
d2eb0b99 | 2025 | if (jh->b_next_transaction != NULL || jh->b_transaction != NULL) |
470decc6 DK |
2026 | goto out; |
2027 | ||
2028 | spin_lock(&journal->j_list_lock); | |
d2eb0b99 | 2029 | if (jh->b_cp_transaction != NULL) { |
470decc6 | 2030 | /* written-back checkpointed metadata buffer */ |
c254c9ec JK |
2031 | JBUFFER_TRACE(jh, "remove from checkpoint list"); |
2032 | __jbd2_journal_remove_checkpoint(jh); | |
470decc6 DK |
2033 | } |
2034 | spin_unlock(&journal->j_list_lock); | |
2035 | out: | |
2036 | return; | |
2037 | } | |
2038 | ||
470decc6 | 2039 | /** |
f7f4bccb | 2040 | * int jbd2_journal_try_to_free_buffers() - try to free page buffers. |
470decc6 DK |
2041 | * @journal: journal for operation |
2042 | * @page: to try and free | |
530576bb | 2043 | * @gfp_mask: we use the mask to detect how hard should we try to release |
d0164adc MG |
2044 | * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit |
2045 | * code to release the buffers. | |
470decc6 DK |
2046 | * |
2047 | * | |
2048 | * For all the buffers on this page, | |
2049 | * if they are fully written out ordered data, move them onto BUF_CLEAN | |
2050 | * so try_to_free_buffers() can reap them. | |
2051 | * | |
2052 | * This function returns non-zero if we wish try_to_free_buffers() | |
2053 | * to be called. We do this if the page is releasable by try_to_free_buffers(). | |
2054 | * We also do it if the page has locked or dirty buffers and the caller wants | |
2055 | * us to perform sync or async writeout. | |
2056 | * | |
2057 | * This complicates JBD locking somewhat. We aren't protected by the | |
2058 | * BKL here. We wish to remove the buffer from its committing or | |
f7f4bccb | 2059 | * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer. |
470decc6 DK |
2060 | * |
2061 | * This may *change* the value of transaction_t->t_datalist, so anyone | |
2062 | * who looks at t_datalist needs to lock against this function. | |
2063 | * | |
f7f4bccb MC |
2064 | * Even worse, someone may be doing a jbd2_journal_dirty_data on this |
2065 | * buffer. So we need to lock against that. jbd2_journal_dirty_data() | |
470decc6 DK |
2066 | * will come out of the lock with the buffer dirty, which makes it |
2067 | * ineligible for release here. | |
2068 | * | |
2069 | * Who else is affected by this? hmm... Really the only contender | |
2070 | * is do_get_write_access() - it could be looking at the buffer while | |
2071 | * journal_try_to_free_buffer() is changing its state. But that | |
2072 | * cannot happen because we never reallocate freed data as metadata | |
2073 | * while the data is part of a transaction. Yes? | |
530576bb MC |
2074 | * |
2075 | * Return 0 on failure, 1 on success | |
470decc6 | 2076 | */ |
f7f4bccb | 2077 | int jbd2_journal_try_to_free_buffers(journal_t *journal, |
530576bb | 2078 | struct page *page, gfp_t gfp_mask) |
470decc6 DK |
2079 | { |
2080 | struct buffer_head *head; | |
2081 | struct buffer_head *bh; | |
0b44b921 | 2082 | bool has_write_io_error = false; |
470decc6 DK |
2083 | int ret = 0; |
2084 | ||
2085 | J_ASSERT(PageLocked(page)); | |
2086 | ||
2087 | head = page_buffers(page); | |
2088 | bh = head; | |
2089 | do { | |
2090 | struct journal_head *jh; | |
2091 | ||
2092 | /* | |
2093 | * We take our own ref against the journal_head here to avoid | |
2094 | * having to add tons of locking around each instance of | |
530576bb | 2095 | * jbd2_journal_put_journal_head(). |
470decc6 | 2096 | */ |
f7f4bccb | 2097 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
2098 | if (!jh) |
2099 | continue; | |
2100 | ||
2101 | jbd_lock_bh_state(bh); | |
2102 | __journal_try_to_free_buffer(journal, bh); | |
f7f4bccb | 2103 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2104 | jbd_unlock_bh_state(bh); |
2105 | if (buffer_jbd(bh)) | |
2106 | goto busy; | |
0b44b921 | 2107 | |
2108 | /* | |
2109 | * If we free a metadata buffer which has been failed to | |
2110 | * write out, the jbd2 checkpoint procedure will not detect | |
2111 | * this failure and may lead to filesystem inconsistency | |
2112 | * after cleanup journal tail. | |
2113 | */ | |
2114 | if (buffer_write_io_error(bh)) { | |
2115 | pr_err("JBD2: Error while async write back metadata bh %llu.", | |
2116 | (unsigned long long)bh->b_blocknr); | |
2117 | has_write_io_error = true; | |
2118 | } | |
470decc6 | 2119 | } while ((bh = bh->b_this_page) != head); |
530576bb | 2120 | |
470decc6 | 2121 | ret = try_to_free_buffers(page); |
530576bb | 2122 | |
470decc6 | 2123 | busy: |
0b44b921 | 2124 | if (has_write_io_error) |
2125 | jbd2_journal_abort(journal, -EIO); | |
2126 | ||
470decc6 DK |
2127 | return ret; |
2128 | } | |
2129 | ||
2130 | /* | |
2131 | * This buffer is no longer needed. If it is on an older transaction's | |
2132 | * checkpoint list we need to record it on this transaction's forget list | |
2133 | * to pin this buffer (and hence its checkpointing transaction) down until | |
2134 | * this transaction commits. If the buffer isn't on a checkpoint list, we | |
2135 | * release it. | |
2136 | * Returns non-zero if JBD no longer has an interest in the buffer. | |
2137 | * | |
2138 | * Called under j_list_lock. | |
2139 | * | |
2140 | * Called under jbd_lock_bh_state(bh). | |
2141 | */ | |
2142 | static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction) | |
2143 | { | |
2144 | int may_free = 1; | |
2145 | struct buffer_head *bh = jh2bh(jh); | |
2146 | ||
470decc6 DK |
2147 | if (jh->b_cp_transaction) { |
2148 | JBUFFER_TRACE(jh, "on running+cp transaction"); | |
de1b7941 | 2149 | __jbd2_journal_temp_unlink_buffer(jh); |
f91d1d04 JK |
2150 | /* |
2151 | * We don't want to write the buffer anymore, clear the | |
2152 | * bit so that we don't confuse checks in | |
2153 | * __journal_file_buffer | |
2154 | */ | |
2155 | clear_buffer_dirty(bh); | |
f7f4bccb | 2156 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); |
470decc6 DK |
2157 | may_free = 0; |
2158 | } else { | |
2159 | JBUFFER_TRACE(jh, "on running transaction"); | |
de1b7941 | 2160 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
2161 | } |
2162 | return may_free; | |
2163 | } | |
2164 | ||
2165 | /* | |
f7f4bccb | 2166 | * jbd2_journal_invalidatepage |
470decc6 DK |
2167 | * |
2168 | * This code is tricky. It has a number of cases to deal with. | |
2169 | * | |
2170 | * There are two invariants which this code relies on: | |
2171 | * | |
2172 | * i_size must be updated on disk before we start calling invalidatepage on the | |
2173 | * data. | |
2174 | * | |
2175 | * This is done in ext3 by defining an ext3_setattr method which | |
2176 | * updates i_size before truncate gets going. By maintaining this | |
2177 | * invariant, we can be sure that it is safe to throw away any buffers | |
2178 | * attached to the current transaction: once the transaction commits, | |
2179 | * we know that the data will not be needed. | |
2180 | * | |
2181 | * Note however that we can *not* throw away data belonging to the | |
2182 | * previous, committing transaction! | |
2183 | * | |
2184 | * Any disk blocks which *are* part of the previous, committing | |
2185 | * transaction (and which therefore cannot be discarded immediately) are | |
2186 | * not going to be reused in the new running transaction | |
2187 | * | |
2188 | * The bitmap committed_data images guarantee this: any block which is | |
2189 | * allocated in one transaction and removed in the next will be marked | |
2190 | * as in-use in the committed_data bitmap, so cannot be reused until | |
2191 | * the next transaction to delete the block commits. This means that | |
2192 | * leaving committing buffers dirty is quite safe: the disk blocks | |
2193 | * cannot be reallocated to a different file and so buffer aliasing is | |
2194 | * not possible. | |
2195 | * | |
2196 | * | |
2197 | * The above applies mainly to ordered data mode. In writeback mode we | |
2198 | * don't make guarantees about the order in which data hits disk --- in | |
2199 | * particular we don't guarantee that new dirty data is flushed before | |
2200 | * transaction commit --- so it is always safe just to discard data | |
2201 | * immediately in that mode. --sct | |
2202 | */ | |
2203 | ||
2204 | /* | |
2205 | * The journal_unmap_buffer helper function returns zero if the buffer | |
2206 | * concerned remains pinned as an anonymous buffer belonging to an older | |
2207 | * transaction. | |
2208 | * | |
2209 | * We're outside-transaction here. Either or both of j_running_transaction | |
2210 | * and j_committing_transaction may be NULL. | |
2211 | */ | |
b794e7a6 JK |
2212 | static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh, |
2213 | int partial_page) | |
470decc6 DK |
2214 | { |
2215 | transaction_t *transaction; | |
2216 | struct journal_head *jh; | |
2217 | int may_free = 1; | |
470decc6 DK |
2218 | |
2219 | BUFFER_TRACE(bh, "entry"); | |
2220 | ||
2221 | /* | |
2222 | * It is safe to proceed here without the j_list_lock because the | |
2223 | * buffers cannot be stolen by try_to_free_buffers as long as we are | |
2224 | * holding the page lock. --sct | |
2225 | */ | |
2226 | ||
2227 | if (!buffer_jbd(bh)) | |
2228 | goto zap_buffer_unlocked; | |
2229 | ||
87c89c23 | 2230 | /* OK, we have data buffer in journaled mode */ |
a931da6a | 2231 | write_lock(&journal->j_state_lock); |
470decc6 DK |
2232 | jbd_lock_bh_state(bh); |
2233 | spin_lock(&journal->j_list_lock); | |
2234 | ||
f7f4bccb | 2235 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
2236 | if (!jh) |
2237 | goto zap_buffer_no_jh; | |
2238 | ||
ba869023 | 2239 | /* |
2240 | * We cannot remove the buffer from checkpoint lists until the | |
2241 | * transaction adding inode to orphan list (let's call it T) | |
2242 | * is committed. Otherwise if the transaction changing the | |
2243 | * buffer would be cleaned from the journal before T is | |
2244 | * committed, a crash will cause that the correct contents of | |
2245 | * the buffer will be lost. On the other hand we have to | |
2246 | * clear the buffer dirty bit at latest at the moment when the | |
2247 | * transaction marking the buffer as freed in the filesystem | |
2248 | * structures is committed because from that moment on the | |
b794e7a6 | 2249 | * block can be reallocated and used by a different page. |
ba869023 | 2250 | * Since the block hasn't been freed yet but the inode has |
2251 | * already been added to orphan list, it is safe for us to add | |
2252 | * the buffer to BJ_Forget list of the newest transaction. | |
b794e7a6 JK |
2253 | * |
2254 | * Also we have to clear buffer_mapped flag of a truncated buffer | |
2255 | * because the buffer_head may be attached to the page straddling | |
2256 | * i_size (can happen only when blocksize < pagesize) and thus the | |
2257 | * buffer_head can be reused when the file is extended again. So we end | |
2258 | * up keeping around invalidated buffers attached to transactions' | |
2259 | * BJ_Forget list just to stop checkpointing code from cleaning up | |
2260 | * the transaction this buffer was modified in. | |
ba869023 | 2261 | */ |
470decc6 DK |
2262 | transaction = jh->b_transaction; |
2263 | if (transaction == NULL) { | |
2264 | /* First case: not on any transaction. If it | |
2265 | * has no checkpoint link, then we can zap it: | |
2266 | * it's a writeback-mode buffer so we don't care | |
2267 | * if it hits disk safely. */ | |
2268 | if (!jh->b_cp_transaction) { | |
2269 | JBUFFER_TRACE(jh, "not on any transaction: zap"); | |
2270 | goto zap_buffer; | |
2271 | } | |
2272 | ||
2273 | if (!buffer_dirty(bh)) { | |
2274 | /* bdflush has written it. We can drop it now */ | |
bc23f0c8 | 2275 | __jbd2_journal_remove_checkpoint(jh); |
470decc6 DK |
2276 | goto zap_buffer; |
2277 | } | |
2278 | ||
2279 | /* OK, it must be in the journal but still not | |
2280 | * written fully to disk: it's metadata or | |
2281 | * journaled data... */ | |
2282 | ||
2283 | if (journal->j_running_transaction) { | |
2284 | /* ... and once the current transaction has | |
2285 | * committed, the buffer won't be needed any | |
2286 | * longer. */ | |
2287 | JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget"); | |
b794e7a6 | 2288 | may_free = __dispose_buffer(jh, |
470decc6 | 2289 | journal->j_running_transaction); |
b794e7a6 | 2290 | goto zap_buffer; |
470decc6 DK |
2291 | } else { |
2292 | /* There is no currently-running transaction. So the | |
2293 | * orphan record which we wrote for this file must have | |
2294 | * passed into commit. We must attach this buffer to | |
2295 | * the committing transaction, if it exists. */ | |
2296 | if (journal->j_committing_transaction) { | |
2297 | JBUFFER_TRACE(jh, "give to committing trans"); | |
b794e7a6 | 2298 | may_free = __dispose_buffer(jh, |
470decc6 | 2299 | journal->j_committing_transaction); |
b794e7a6 | 2300 | goto zap_buffer; |
470decc6 DK |
2301 | } else { |
2302 | /* The orphan record's transaction has | |
2303 | * committed. We can cleanse this buffer */ | |
2304 | clear_buffer_jbddirty(bh); | |
bc23f0c8 | 2305 | __jbd2_journal_remove_checkpoint(jh); |
470decc6 DK |
2306 | goto zap_buffer; |
2307 | } | |
2308 | } | |
2309 | } else if (transaction == journal->j_committing_transaction) { | |
9b57988d | 2310 | JBUFFER_TRACE(jh, "on committing transaction"); |
470decc6 | 2311 | /* |
ba869023 | 2312 | * The buffer is committing, we simply cannot touch |
b794e7a6 JK |
2313 | * it. If the page is straddling i_size we have to wait |
2314 | * for commit and try again. | |
2315 | */ | |
2316 | if (partial_page) { | |
b794e7a6 JK |
2317 | jbd2_journal_put_journal_head(jh); |
2318 | spin_unlock(&journal->j_list_lock); | |
2319 | jbd_unlock_bh_state(bh); | |
2320 | write_unlock(&journal->j_state_lock); | |
53e87268 | 2321 | return -EBUSY; |
b794e7a6 JK |
2322 | } |
2323 | /* | |
334e643a | 2324 | * OK, buffer won't be reachable after truncate. We just clear |
2325 | * b_modified to not confuse transaction credit accounting, and | |
2326 | * set j_next_transaction to the running transaction (if there | |
2327 | * is one) and mark buffer as freed so that commit code knows | |
2328 | * it should clear dirty bits when it is done with the buffer. | |
ba869023 | 2329 | */ |
470decc6 | 2330 | set_buffer_freed(bh); |
ba869023 | 2331 | if (journal->j_running_transaction && buffer_jbddirty(bh)) |
2332 | jh->b_next_transaction = journal->j_running_transaction; | |
334e643a | 2333 | jh->b_modified = 0; |
f7f4bccb | 2334 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2335 | spin_unlock(&journal->j_list_lock); |
2336 | jbd_unlock_bh_state(bh); | |
a931da6a | 2337 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2338 | return 0; |
2339 | } else { | |
2340 | /* Good, the buffer belongs to the running transaction. | |
2341 | * We are writing our own transaction's data, not any | |
2342 | * previous one's, so it is safe to throw it away | |
2343 | * (remember that we expect the filesystem to have set | |
2344 | * i_size already for this truncate so recovery will not | |
2345 | * expose the disk blocks we are discarding here.) */ | |
2346 | J_ASSERT_JH(jh, transaction == journal->j_running_transaction); | |
9b57988d | 2347 | JBUFFER_TRACE(jh, "on running transaction"); |
470decc6 DK |
2348 | may_free = __dispose_buffer(jh, transaction); |
2349 | } | |
2350 | ||
2351 | zap_buffer: | |
b794e7a6 JK |
2352 | /* |
2353 | * This is tricky. Although the buffer is truncated, it may be reused | |
2354 | * if blocksize < pagesize and it is attached to the page straddling | |
2355 | * EOF. Since the buffer might have been added to BJ_Forget list of the | |
2356 | * running transaction, journal_get_write_access() won't clear | |
2357 | * b_modified and credit accounting gets confused. So clear b_modified | |
2358 | * here. | |
2359 | */ | |
2360 | jh->b_modified = 0; | |
f7f4bccb | 2361 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2362 | zap_buffer_no_jh: |
2363 | spin_unlock(&journal->j_list_lock); | |
2364 | jbd_unlock_bh_state(bh); | |
a931da6a | 2365 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
2366 | zap_buffer_unlocked: |
2367 | clear_buffer_dirty(bh); | |
2368 | J_ASSERT_BH(bh, !buffer_jbddirty(bh)); | |
2369 | clear_buffer_mapped(bh); | |
2370 | clear_buffer_req(bh); | |
2371 | clear_buffer_new(bh); | |
15291164 ES |
2372 | clear_buffer_delay(bh); |
2373 | clear_buffer_unwritten(bh); | |
470decc6 DK |
2374 | bh->b_bdev = NULL; |
2375 | return may_free; | |
2376 | } | |
2377 | ||
2378 | /** | |
f7f4bccb | 2379 | * void jbd2_journal_invalidatepage() |
470decc6 DK |
2380 | * @journal: journal to use for flush... |
2381 | * @page: page to flush | |
259709b0 LC |
2382 | * @offset: start of the range to invalidate |
2383 | * @length: length of the range to invalidate | |
470decc6 | 2384 | * |
259709b0 LC |
2385 | * Reap page buffers containing data after in the specified range in page. |
2386 | * Can return -EBUSY if buffers are part of the committing transaction and | |
2387 | * the page is straddling i_size. Caller then has to wait for current commit | |
2388 | * and try again. | |
470decc6 | 2389 | */ |
53e87268 JK |
2390 | int jbd2_journal_invalidatepage(journal_t *journal, |
2391 | struct page *page, | |
259709b0 LC |
2392 | unsigned int offset, |
2393 | unsigned int length) | |
470decc6 DK |
2394 | { |
2395 | struct buffer_head *head, *bh, *next; | |
259709b0 | 2396 | unsigned int stop = offset + length; |
470decc6 | 2397 | unsigned int curr_off = 0; |
09cbfeaf | 2398 | int partial_page = (offset || length < PAGE_SIZE); |
470decc6 | 2399 | int may_free = 1; |
53e87268 | 2400 | int ret = 0; |
470decc6 DK |
2401 | |
2402 | if (!PageLocked(page)) | |
2403 | BUG(); | |
2404 | if (!page_has_buffers(page)) | |
53e87268 | 2405 | return 0; |
470decc6 | 2406 | |
09cbfeaf | 2407 | BUG_ON(stop > PAGE_SIZE || stop < length); |
259709b0 | 2408 | |
470decc6 DK |
2409 | /* We will potentially be playing with lists other than just the |
2410 | * data lists (especially for journaled data mode), so be | |
2411 | * cautious in our locking. */ | |
2412 | ||
2413 | head = bh = page_buffers(page); | |
2414 | do { | |
2415 | unsigned int next_off = curr_off + bh->b_size; | |
2416 | next = bh->b_this_page; | |
2417 | ||
259709b0 LC |
2418 | if (next_off > stop) |
2419 | return 0; | |
2420 | ||
470decc6 DK |
2421 | if (offset <= curr_off) { |
2422 | /* This block is wholly outside the truncation point */ | |
2423 | lock_buffer(bh); | |
259709b0 | 2424 | ret = journal_unmap_buffer(journal, bh, partial_page); |
470decc6 | 2425 | unlock_buffer(bh); |
53e87268 JK |
2426 | if (ret < 0) |
2427 | return ret; | |
2428 | may_free &= ret; | |
470decc6 DK |
2429 | } |
2430 | curr_off = next_off; | |
2431 | bh = next; | |
2432 | ||
2433 | } while (bh != head); | |
2434 | ||
259709b0 | 2435 | if (!partial_page) { |
470decc6 DK |
2436 | if (may_free && try_to_free_buffers(page)) |
2437 | J_ASSERT(!page_has_buffers(page)); | |
2438 | } | |
53e87268 | 2439 | return 0; |
470decc6 DK |
2440 | } |
2441 | ||
2442 | /* | |
2443 | * File a buffer on the given transaction list. | |
2444 | */ | |
f7f4bccb | 2445 | void __jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2446 | transaction_t *transaction, int jlist) |
2447 | { | |
2448 | struct journal_head **list = NULL; | |
2449 | int was_dirty = 0; | |
2450 | struct buffer_head *bh = jh2bh(jh); | |
2451 | ||
2452 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
2453 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
2454 | ||
2455 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
2456 | J_ASSERT_JH(jh, jh->b_transaction == transaction || | |
4019191b | 2457 | jh->b_transaction == NULL); |
470decc6 DK |
2458 | |
2459 | if (jh->b_transaction && jh->b_jlist == jlist) | |
2460 | return; | |
2461 | ||
470decc6 DK |
2462 | if (jlist == BJ_Metadata || jlist == BJ_Reserved || |
2463 | jlist == BJ_Shadow || jlist == BJ_Forget) { | |
f91d1d04 JK |
2464 | /* |
2465 | * For metadata buffers, we track dirty bit in buffer_jbddirty | |
2466 | * instead of buffer_dirty. We should not see a dirty bit set | |
2467 | * here because we clear it in do_get_write_access but e.g. | |
2468 | * tune2fs can modify the sb and set the dirty bit at any time | |
2469 | * so we try to gracefully handle that. | |
2470 | */ | |
2471 | if (buffer_dirty(bh)) | |
2472 | warn_dirty_buffer(bh); | |
470decc6 DK |
2473 | if (test_clear_buffer_dirty(bh) || |
2474 | test_clear_buffer_jbddirty(bh)) | |
2475 | was_dirty = 1; | |
2476 | } | |
2477 | ||
2478 | if (jh->b_transaction) | |
f7f4bccb | 2479 | __jbd2_journal_temp_unlink_buffer(jh); |
de1b7941 JK |
2480 | else |
2481 | jbd2_journal_grab_journal_head(bh); | |
470decc6 DK |
2482 | jh->b_transaction = transaction; |
2483 | ||
2484 | switch (jlist) { | |
2485 | case BJ_None: | |
2486 | J_ASSERT_JH(jh, !jh->b_committed_data); | |
2487 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
2488 | return; | |
470decc6 DK |
2489 | case BJ_Metadata: |
2490 | transaction->t_nr_buffers++; | |
2491 | list = &transaction->t_buffers; | |
2492 | break; | |
2493 | case BJ_Forget: | |
2494 | list = &transaction->t_forget; | |
2495 | break; | |
470decc6 DK |
2496 | case BJ_Shadow: |
2497 | list = &transaction->t_shadow_list; | |
2498 | break; | |
470decc6 DK |
2499 | case BJ_Reserved: |
2500 | list = &transaction->t_reserved_list; | |
2501 | break; | |
470decc6 DK |
2502 | } |
2503 | ||
2504 | __blist_add_buffer(list, jh); | |
2505 | jh->b_jlist = jlist; | |
2506 | ||
2507 | if (was_dirty) | |
2508 | set_buffer_jbddirty(bh); | |
2509 | } | |
2510 | ||
f7f4bccb | 2511 | void jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2512 | transaction_t *transaction, int jlist) |
2513 | { | |
2514 | jbd_lock_bh_state(jh2bh(jh)); | |
2515 | spin_lock(&transaction->t_journal->j_list_lock); | |
f7f4bccb | 2516 | __jbd2_journal_file_buffer(jh, transaction, jlist); |
470decc6 DK |
2517 | spin_unlock(&transaction->t_journal->j_list_lock); |
2518 | jbd_unlock_bh_state(jh2bh(jh)); | |
2519 | } | |
2520 | ||
2521 | /* | |
2522 | * Remove a buffer from its current buffer list in preparation for | |
2523 | * dropping it from its current transaction entirely. If the buffer has | |
2524 | * already started to be used by a subsequent transaction, refile the | |
2525 | * buffer on that transaction's metadata list. | |
2526 | * | |
de1b7941 | 2527 | * Called under j_list_lock |
470decc6 | 2528 | * Called under jbd_lock_bh_state(jh2bh(jh)) |
de1b7941 JK |
2529 | * |
2530 | * jh and bh may be already free when this function returns | |
470decc6 | 2531 | */ |
f7f4bccb | 2532 | void __jbd2_journal_refile_buffer(struct journal_head *jh) |
470decc6 | 2533 | { |
ba869023 | 2534 | int was_dirty, jlist; |
470decc6 DK |
2535 | struct buffer_head *bh = jh2bh(jh); |
2536 | ||
2537 | J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh)); | |
2538 | if (jh->b_transaction) | |
2539 | assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock); | |
2540 | ||
2541 | /* If the buffer is now unused, just drop it. */ | |
2542 | if (jh->b_next_transaction == NULL) { | |
f7f4bccb | 2543 | __jbd2_journal_unfile_buffer(jh); |
470decc6 DK |
2544 | return; |
2545 | } | |
2546 | ||
2547 | /* | |
2548 | * It has been modified by a later transaction: add it to the new | |
2549 | * transaction's metadata list. | |
2550 | */ | |
2551 | ||
2552 | was_dirty = test_clear_buffer_jbddirty(bh); | |
f7f4bccb | 2553 | __jbd2_journal_temp_unlink_buffer(jh); |
412c5828 LC |
2554 | |
2555 | /* | |
2556 | * b_transaction must be set, otherwise the new b_transaction won't | |
2557 | * be holding jh reference | |
2558 | */ | |
2559 | J_ASSERT_JH(jh, jh->b_transaction != NULL); | |
2560 | ||
de1b7941 JK |
2561 | /* |
2562 | * We set b_transaction here because b_next_transaction will inherit | |
2563 | * our jh reference and thus __jbd2_journal_file_buffer() must not | |
2564 | * take a new one. | |
2565 | */ | |
c33e585a QC |
2566 | WRITE_ONCE(jh->b_transaction, jh->b_next_transaction); |
2567 | WRITE_ONCE(jh->b_next_transaction, NULL); | |
ba869023 | 2568 | if (buffer_freed(bh)) |
2569 | jlist = BJ_Forget; | |
2570 | else if (jh->b_modified) | |
2571 | jlist = BJ_Metadata; | |
2572 | else | |
2573 | jlist = BJ_Reserved; | |
2574 | __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist); | |
470decc6 DK |
2575 | J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING); |
2576 | ||
2577 | if (was_dirty) | |
2578 | set_buffer_jbddirty(bh); | |
2579 | } | |
2580 | ||
2581 | /* | |
de1b7941 JK |
2582 | * __jbd2_journal_refile_buffer() with necessary locking added. We take our |
2583 | * bh reference so that we can safely unlock bh. | |
2584 | * | |
2585 | * The jh and bh may be freed by this call. | |
470decc6 | 2586 | */ |
f7f4bccb | 2587 | void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 DK |
2588 | { |
2589 | struct buffer_head *bh = jh2bh(jh); | |
2590 | ||
de1b7941 JK |
2591 | /* Get reference so that buffer cannot be freed before we unlock it */ |
2592 | get_bh(bh); | |
470decc6 DK |
2593 | jbd_lock_bh_state(bh); |
2594 | spin_lock(&journal->j_list_lock); | |
f7f4bccb | 2595 | __jbd2_journal_refile_buffer(jh); |
470decc6 | 2596 | jbd_unlock_bh_state(bh); |
470decc6 DK |
2597 | spin_unlock(&journal->j_list_lock); |
2598 | __brelse(bh); | |
2599 | } | |
c851ed54 JK |
2600 | |
2601 | /* | |
2602 | * File inode in the inode list of the handle's transaction | |
2603 | */ | |
41617e1a | 2604 | static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode, |
6ba0e7dc | 2605 | unsigned long flags, loff_t start_byte, loff_t end_byte) |
c851ed54 JK |
2606 | { |
2607 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 2608 | journal_t *journal; |
c851ed54 JK |
2609 | |
2610 | if (is_handle_aborted(handle)) | |
41a5b913 TT |
2611 | return -EROFS; |
2612 | journal = transaction->t_journal; | |
c851ed54 JK |
2613 | |
2614 | jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino, | |
2615 | transaction->t_tid); | |
2616 | ||
c851ed54 | 2617 | spin_lock(&journal->j_list_lock); |
41617e1a | 2618 | jinode->i_flags |= flags; |
6ba0e7dc RZ |
2619 | |
2620 | if (jinode->i_dirty_end) { | |
2621 | jinode->i_dirty_start = min(jinode->i_dirty_start, start_byte); | |
2622 | jinode->i_dirty_end = max(jinode->i_dirty_end, end_byte); | |
2623 | } else { | |
2624 | jinode->i_dirty_start = start_byte; | |
2625 | jinode->i_dirty_end = end_byte; | |
2626 | } | |
2627 | ||
41617e1a | 2628 | /* Is inode already attached where we need it? */ |
c851ed54 JK |
2629 | if (jinode->i_transaction == transaction || |
2630 | jinode->i_next_transaction == transaction) | |
2631 | goto done; | |
2632 | ||
81be12c8 JK |
2633 | /* |
2634 | * We only ever set this variable to 1 so the test is safe. Since | |
2635 | * t_need_data_flush is likely to be set, we do the test to save some | |
2636 | * cacheline bouncing | |
2637 | */ | |
2638 | if (!transaction->t_need_data_flush) | |
2639 | transaction->t_need_data_flush = 1; | |
c851ed54 JK |
2640 | /* On some different transaction's list - should be |
2641 | * the committing one */ | |
2642 | if (jinode->i_transaction) { | |
2643 | J_ASSERT(jinode->i_next_transaction == NULL); | |
2644 | J_ASSERT(jinode->i_transaction == | |
2645 | journal->j_committing_transaction); | |
2646 | jinode->i_next_transaction = transaction; | |
2647 | goto done; | |
2648 | } | |
2649 | /* Not on any transaction list... */ | |
2650 | J_ASSERT(!jinode->i_next_transaction); | |
2651 | jinode->i_transaction = transaction; | |
2652 | list_add(&jinode->i_list, &transaction->t_inode_list); | |
2653 | done: | |
2654 | spin_unlock(&journal->j_list_lock); | |
2655 | ||
2656 | return 0; | |
2657 | } | |
2658 | ||
6ba0e7dc RZ |
2659 | int jbd2_journal_inode_ranged_write(handle_t *handle, |
2660 | struct jbd2_inode *jinode, loff_t start_byte, loff_t length) | |
2661 | { | |
2662 | return jbd2_journal_file_inode(handle, jinode, | |
2663 | JI_WRITE_DATA | JI_WAIT_DATA, start_byte, | |
2664 | start_byte + length - 1); | |
2665 | } | |
2666 | ||
2667 | int jbd2_journal_inode_ranged_wait(handle_t *handle, struct jbd2_inode *jinode, | |
2668 | loff_t start_byte, loff_t length) | |
2669 | { | |
2670 | return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA, | |
2671 | start_byte, start_byte + length - 1); | |
41617e1a JK |
2672 | } |
2673 | ||
c851ed54 | 2674 | /* |
7f5aa215 JK |
2675 | * File truncate and transaction commit interact with each other in a |
2676 | * non-trivial way. If a transaction writing data block A is | |
2677 | * committing, we cannot discard the data by truncate until we have | |
2678 | * written them. Otherwise if we crashed after the transaction with | |
2679 | * write has committed but before the transaction with truncate has | |
2680 | * committed, we could see stale data in block A. This function is a | |
2681 | * helper to solve this problem. It starts writeout of the truncated | |
2682 | * part in case it is in the committing transaction. | |
2683 | * | |
2684 | * Filesystem code must call this function when inode is journaled in | |
2685 | * ordered mode before truncation happens and after the inode has been | |
2686 | * placed on orphan list with the new inode size. The second condition | |
2687 | * avoids the race that someone writes new data and we start | |
2688 | * committing the transaction after this function has been called but | |
2689 | * before a transaction for truncate is started (and furthermore it | |
2690 | * allows us to optimize the case where the addition to orphan list | |
2691 | * happens in the same transaction as write --- we don't have to write | |
2692 | * any data in such case). | |
c851ed54 | 2693 | */ |
7f5aa215 JK |
2694 | int jbd2_journal_begin_ordered_truncate(journal_t *journal, |
2695 | struct jbd2_inode *jinode, | |
c851ed54 JK |
2696 | loff_t new_size) |
2697 | { | |
7f5aa215 | 2698 | transaction_t *inode_trans, *commit_trans; |
c851ed54 JK |
2699 | int ret = 0; |
2700 | ||
7f5aa215 JK |
2701 | /* This is a quick check to avoid locking if not necessary */ |
2702 | if (!jinode->i_transaction) | |
c851ed54 | 2703 | goto out; |
7f5aa215 JK |
2704 | /* Locks are here just to force reading of recent values, it is |
2705 | * enough that the transaction was not committing before we started | |
2706 | * a transaction adding the inode to orphan list */ | |
a931da6a | 2707 | read_lock(&journal->j_state_lock); |
c851ed54 | 2708 | commit_trans = journal->j_committing_transaction; |
a931da6a | 2709 | read_unlock(&journal->j_state_lock); |
7f5aa215 JK |
2710 | spin_lock(&journal->j_list_lock); |
2711 | inode_trans = jinode->i_transaction; | |
2712 | spin_unlock(&journal->j_list_lock); | |
2713 | if (inode_trans == commit_trans) { | |
2714 | ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping, | |
c851ed54 JK |
2715 | new_size, LLONG_MAX); |
2716 | if (ret) | |
2717 | jbd2_journal_abort(journal, ret); | |
2718 | } | |
2719 | out: | |
2720 | return ret; | |
2721 | } |