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 | ||
9f356e5a | 65 | /* |
19014d69 | 66 | * Base amount of descriptor blocks we reserve for each transaction. |
9f356e5a | 67 | */ |
9f356e5a JK |
68 | static int jbd2_descriptor_blocks_per_trans(journal_t *journal) |
69 | { | |
19014d69 JK |
70 | int tag_space = journal->j_blocksize - sizeof(journal_header_t); |
71 | int tags_per_block; | |
72 | ||
73 | /* Subtract UUID */ | |
74 | tag_space -= 16; | |
75 | if (jbd2_journal_has_csum_v2or3(journal)) | |
76 | tag_space -= sizeof(struct jbd2_journal_block_tail); | |
77 | /* Commit code leaves a slack space of 16 bytes at the end of block */ | |
78 | tags_per_block = (tag_space - 16) / journal_tag_bytes(journal); | |
79 | /* | |
80 | * Revoke descriptors are accounted separately so we need to reserve | |
81 | * space for commit block and normal transaction descriptor blocks. | |
82 | */ | |
83 | return 1 + DIV_ROUND_UP(journal->j_max_transaction_buffers, | |
84 | tags_per_block); | |
9f356e5a JK |
85 | } |
86 | ||
470decc6 | 87 | /* |
f7f4bccb | 88 | * jbd2_get_transaction: obtain a new transaction_t object. |
470decc6 | 89 | * |
0df6f469 | 90 | * Simply initialise a new transaction. Initialize it in |
470decc6 DK |
91 | * RUNNING state and add it to the current journal (which should not |
92 | * have an existing running transaction: we only make a new transaction | |
93 | * once we have started to commit the old one). | |
94 | * | |
95 | * Preconditions: | |
96 | * The journal MUST be locked. We don't perform atomic mallocs on the | |
97 | * new transaction and we can't block without protecting against other | |
98 | * processes trying to touch the journal while it is in transition. | |
99 | * | |
470decc6 DK |
100 | */ |
101 | ||
0df6f469 LS |
102 | static void jbd2_get_transaction(journal_t *journal, |
103 | transaction_t *transaction) | |
470decc6 DK |
104 | { |
105 | transaction->t_journal = journal; | |
106 | transaction->t_state = T_RUNNING; | |
e07f7183 | 107 | transaction->t_start_time = ktime_get(); |
470decc6 DK |
108 | transaction->t_tid = journal->j_transaction_sequence++; |
109 | transaction->t_expires = jiffies + journal->j_commit_interval; | |
a51dca9c | 110 | atomic_set(&transaction->t_updates, 0); |
8f7d89f3 | 111 | atomic_set(&transaction->t_outstanding_credits, |
9f356e5a | 112 | jbd2_descriptor_blocks_per_trans(journal) + |
8f7d89f3 | 113 | atomic_read(&journal->j_reserved_credits)); |
fdc3ef88 | 114 | atomic_set(&transaction->t_outstanding_revokes, 0); |
8dd42046 | 115 | atomic_set(&transaction->t_handle_count, 0); |
c851ed54 | 116 | INIT_LIST_HEAD(&transaction->t_inode_list); |
3e624fc7 | 117 | INIT_LIST_HEAD(&transaction->t_private_list); |
470decc6 DK |
118 | |
119 | /* Set up the commit timer for the new transaction. */ | |
b1f485f2 | 120 | journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires); |
470decc6 DK |
121 | add_timer(&journal->j_commit_timer); |
122 | ||
123 | J_ASSERT(journal->j_running_transaction == NULL); | |
124 | journal->j_running_transaction = transaction; | |
8e85fb3f JL |
125 | transaction->t_max_wait = 0; |
126 | transaction->t_start = jiffies; | |
9fff24aa | 127 | transaction->t_requested = 0; |
470decc6 DK |
128 | } |
129 | ||
130 | /* | |
131 | * Handle management. | |
132 | * | |
133 | * A handle_t is an object which represents a single atomic update to a | |
134 | * filesystem, and which tracks all of the modifications which form part | |
135 | * of that one update. | |
136 | */ | |
137 | ||
6d0bf005 | 138 | /* |
28e35e42 | 139 | * Update transaction's maximum wait time, if debugging is enabled. |
6d0bf005 | 140 | * |
f7f497cb RH |
141 | * t_max_wait is carefully updated here with use of atomic compare exchange. |
142 | * Note that there could be multiplre threads trying to do this simultaneously | |
143 | * hence using cmpxchg to avoid any use of locks in this case. | |
2d442920 | 144 | * With this t_max_wait can be updated w/o enabling jbd2_journal_enable_debug. |
6d0bf005 | 145 | */ |
28e35e42 TM |
146 | static inline void update_t_max_wait(transaction_t *transaction, |
147 | unsigned long ts) | |
6d0bf005 | 148 | { |
f7f497cb | 149 | unsigned long oldts, newts; |
2d442920 RH |
150 | |
151 | if (time_after(transaction->t_start, ts)) { | |
f7f497cb RH |
152 | newts = jbd2_time_diff(ts, transaction->t_start); |
153 | oldts = READ_ONCE(transaction->t_max_wait); | |
154 | while (oldts < newts) | |
155 | oldts = cmpxchg(&transaction->t_max_wait, oldts, newts); | |
6d0bf005 | 156 | } |
6d0bf005 TT |
157 | } |
158 | ||
8f7d89f3 | 159 | /* |
96f1e097 JK |
160 | * Wait until running transaction passes to T_FLUSH state and new transaction |
161 | * can thus be started. Also starts the commit if needed. The function expects | |
162 | * running transaction to exist and releases j_state_lock. | |
8f7d89f3 JK |
163 | */ |
164 | static void wait_transaction_locked(journal_t *journal) | |
165 | __releases(journal->j_state_lock) | |
166 | { | |
167 | DEFINE_WAIT(wait); | |
168 | int need_to_start; | |
169 | tid_t tid = journal->j_running_transaction->t_tid; | |
170 | ||
34fc8768 | 171 | prepare_to_wait_exclusive(&journal->j_wait_transaction_locked, &wait, |
8f7d89f3 JK |
172 | TASK_UNINTERRUPTIBLE); |
173 | need_to_start = !tid_geq(journal->j_commit_request, tid); | |
174 | read_unlock(&journal->j_state_lock); | |
175 | if (need_to_start) | |
176 | jbd2_log_start_commit(journal, tid); | |
e03a9976 | 177 | jbd2_might_wait_for_commit(journal); |
8f7d89f3 JK |
178 | schedule(); |
179 | finish_wait(&journal->j_wait_transaction_locked, &wait); | |
180 | } | |
181 | ||
96f1e097 JK |
182 | /* |
183 | * Wait until running transaction transitions from T_SWITCH to T_FLUSH | |
184 | * state and new transaction can thus be started. The function releases | |
185 | * j_state_lock. | |
186 | */ | |
187 | static void wait_transaction_switching(journal_t *journal) | |
188 | __releases(journal->j_state_lock) | |
189 | { | |
190 | DEFINE_WAIT(wait); | |
191 | ||
192 | if (WARN_ON(!journal->j_running_transaction || | |
05d5233d TT |
193 | journal->j_running_transaction->t_state != T_SWITCH)) { |
194 | read_unlock(&journal->j_state_lock); | |
96f1e097 | 195 | return; |
05d5233d | 196 | } |
34fc8768 | 197 | prepare_to_wait_exclusive(&journal->j_wait_transaction_locked, &wait, |
96f1e097 JK |
198 | TASK_UNINTERRUPTIBLE); |
199 | read_unlock(&journal->j_state_lock); | |
200 | /* | |
201 | * We don't call jbd2_might_wait_for_commit() here as there's no | |
202 | * waiting for outstanding handles happening anymore in T_SWITCH state | |
203 | * and handling of reserved handles actually relies on that for | |
204 | * correctness. | |
205 | */ | |
206 | schedule(); | |
207 | finish_wait(&journal->j_wait_transaction_locked, &wait); | |
208 | } | |
209 | ||
8f7d89f3 JK |
210 | static void sub_reserved_credits(journal_t *journal, int blocks) |
211 | { | |
212 | atomic_sub(blocks, &journal->j_reserved_credits); | |
213 | wake_up(&journal->j_wait_reserved); | |
214 | } | |
215 | ||
216 | /* | |
217 | * Wait until we can add credits for handle to the running transaction. Called | |
218 | * with j_state_lock held for reading. Returns 0 if handle joined the running | |
219 | * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and | |
220 | * caller must retry. | |
b33d9f59 TT |
221 | * |
222 | * Note: because j_state_lock may be dropped depending on the return | |
223 | * value, we need to fake out sparse so ti doesn't complain about a | |
224 | * locking imbalance. Callers of add_transaction_credits will need to | |
225 | * make a similar accomodation. | |
8f7d89f3 JK |
226 | */ |
227 | static int add_transaction_credits(journal_t *journal, int blocks, | |
228 | int rsv_blocks) | |
b33d9f59 | 229 | __must_hold(&journal->j_state_lock) |
8f7d89f3 JK |
230 | { |
231 | transaction_t *t = journal->j_running_transaction; | |
232 | int needed; | |
233 | int total = blocks + rsv_blocks; | |
234 | ||
235 | /* | |
236 | * If the current transaction is locked down for commit, wait | |
237 | * for the lock to be released. | |
238 | */ | |
96f1e097 JK |
239 | if (t->t_state != T_RUNNING) { |
240 | WARN_ON_ONCE(t->t_state >= T_FLUSH); | |
8f7d89f3 | 241 | wait_transaction_locked(journal); |
b33d9f59 | 242 | __acquire(&journal->j_state_lock); /* fake out sparse */ |
8f7d89f3 JK |
243 | return 1; |
244 | } | |
245 | ||
246 | /* | |
247 | * If there is not enough space left in the log to write all | |
248 | * potential buffers requested by this operation, we need to | |
249 | * stall pending a log checkpoint to free some more log space. | |
250 | */ | |
251 | needed = atomic_add_return(total, &t->t_outstanding_credits); | |
252 | if (needed > journal->j_max_transaction_buffers) { | |
253 | /* | |
254 | * If the current transaction is already too large, | |
255 | * then start to commit it: we can then go back and | |
256 | * attach this handle to a new transaction. | |
257 | */ | |
258 | atomic_sub(total, &t->t_outstanding_credits); | |
6d3ec14d LC |
259 | |
260 | /* | |
261 | * Is the number of reserved credits in the current transaction too | |
262 | * big to fit this handle? Wait until reserved credits are freed. | |
263 | */ | |
264 | if (atomic_read(&journal->j_reserved_credits) + total > | |
265 | journal->j_max_transaction_buffers) { | |
266 | read_unlock(&journal->j_state_lock); | |
e03a9976 | 267 | jbd2_might_wait_for_commit(journal); |
6d3ec14d LC |
268 | wait_event(journal->j_wait_reserved, |
269 | atomic_read(&journal->j_reserved_credits) + total <= | |
270 | journal->j_max_transaction_buffers); | |
b33d9f59 | 271 | __acquire(&journal->j_state_lock); /* fake out sparse */ |
6d3ec14d LC |
272 | return 1; |
273 | } | |
274 | ||
8f7d89f3 | 275 | wait_transaction_locked(journal); |
b33d9f59 | 276 | __acquire(&journal->j_state_lock); /* fake out sparse */ |
8f7d89f3 JK |
277 | return 1; |
278 | } | |
279 | ||
280 | /* | |
281 | * The commit code assumes that it can get enough log space | |
282 | * without forcing a checkpoint. This is *critical* for | |
283 | * correctness: a checkpoint of a buffer which is also | |
284 | * associated with a committing transaction creates a deadlock, | |
285 | * so commit simply cannot force through checkpoints. | |
286 | * | |
287 | * We must therefore ensure the necessary space in the journal | |
288 | * *before* starting to dirty potentially checkpointed buffers | |
289 | * in the new transaction. | |
290 | */ | |
77444ac4 | 291 | if (jbd2_log_space_left(journal) < journal->j_max_transaction_buffers) { |
8f7d89f3 JK |
292 | atomic_sub(total, &t->t_outstanding_credits); |
293 | read_unlock(&journal->j_state_lock); | |
e03a9976 | 294 | jbd2_might_wait_for_commit(journal); |
8f7d89f3 | 295 | write_lock(&journal->j_state_lock); |
77444ac4 JK |
296 | if (jbd2_log_space_left(journal) < |
297 | journal->j_max_transaction_buffers) | |
8f7d89f3 JK |
298 | __jbd2_log_wait_for_space(journal); |
299 | write_unlock(&journal->j_state_lock); | |
b33d9f59 | 300 | __acquire(&journal->j_state_lock); /* fake out sparse */ |
8f7d89f3 JK |
301 | return 1; |
302 | } | |
303 | ||
304 | /* No reservation? We are done... */ | |
305 | if (!rsv_blocks) | |
306 | return 0; | |
307 | ||
308 | needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits); | |
309 | /* We allow at most half of a transaction to be reserved */ | |
310 | if (needed > journal->j_max_transaction_buffers / 2) { | |
311 | sub_reserved_credits(journal, rsv_blocks); | |
312 | atomic_sub(total, &t->t_outstanding_credits); | |
313 | read_unlock(&journal->j_state_lock); | |
e03a9976 | 314 | jbd2_might_wait_for_commit(journal); |
8f7d89f3 JK |
315 | wait_event(journal->j_wait_reserved, |
316 | atomic_read(&journal->j_reserved_credits) + rsv_blocks | |
317 | <= journal->j_max_transaction_buffers / 2); | |
b33d9f59 | 318 | __acquire(&journal->j_state_lock); /* fake out sparse */ |
8f7d89f3 JK |
319 | return 1; |
320 | } | |
321 | return 0; | |
322 | } | |
323 | ||
470decc6 DK |
324 | /* |
325 | * start_this_handle: Given a handle, deal with any locking or stalling | |
326 | * needed to make sure that there is enough journal space for the handle | |
327 | * to begin. Attach the handle to a transaction and set up the | |
328 | * transaction's buffer credits. | |
329 | */ | |
330 | ||
47def826 | 331 | static int start_this_handle(journal_t *journal, handle_t *handle, |
d2159fb7 | 332 | gfp_t gfp_mask) |
470decc6 | 333 | { |
e4471831 | 334 | transaction_t *transaction, *new_transaction = NULL; |
933f1c1e | 335 | int blocks = handle->h_total_credits; |
8f7d89f3 | 336 | int rsv_blocks = 0; |
28e35e42 | 337 | unsigned long ts = jiffies; |
470decc6 | 338 | |
6d3ec14d | 339 | if (handle->h_rsv_handle) |
933f1c1e | 340 | rsv_blocks = handle->h_rsv_handle->h_total_credits; |
6d3ec14d | 341 | |
8f7d89f3 | 342 | /* |
6d3ec14d LC |
343 | * Limit the number of reserved credits to 1/2 of maximum transaction |
344 | * size and limit the number of total credits to not exceed maximum | |
345 | * transaction size per operation. | |
8f7d89f3 | 346 | */ |
6d3ec14d LC |
347 | if ((rsv_blocks > journal->j_max_transaction_buffers / 2) || |
348 | (rsv_blocks + blocks > journal->j_max_transaction_buffers)) { | |
349 | printk(KERN_ERR "JBD2: %s wants too many credits " | |
350 | "credits:%d rsv_credits:%d max:%d\n", | |
351 | current->comm, blocks, rsv_blocks, | |
352 | journal->j_max_transaction_buffers); | |
353 | WARN_ON(1); | |
47def826 | 354 | return -ENOSPC; |
470decc6 DK |
355 | } |
356 | ||
357 | alloc_transaction: | |
3b1833e9 JK |
358 | /* |
359 | * This check is racy but it is just an optimization of allocating new | |
360 | * transaction early if there are high chances we'll need it. If we | |
361 | * guess wrong, we'll retry or free unused transaction. | |
362 | */ | |
363 | if (!data_race(journal->j_running_transaction)) { | |
6ccaf3e2 MH |
364 | /* |
365 | * If __GFP_FS is not present, then we may be being called from | |
366 | * inside the fs writeback layer, so we MUST NOT fail. | |
367 | */ | |
368 | if ((gfp_mask & __GFP_FS) == 0) | |
369 | gfp_mask |= __GFP_NOFAIL; | |
b2f4edb3 WG |
370 | new_transaction = kmem_cache_zalloc(transaction_cache, |
371 | gfp_mask); | |
6ccaf3e2 | 372 | if (!new_transaction) |
47def826 | 373 | return -ENOMEM; |
470decc6 DK |
374 | } |
375 | ||
cb3b3bf2 | 376 | jbd2_debug(3, "New handle %p going live.\n", handle); |
470decc6 | 377 | |
470decc6 DK |
378 | /* |
379 | * We need to hold j_state_lock until t_updates has been incremented, | |
380 | * for proper journal barrier handling | |
381 | */ | |
a931da6a TT |
382 | repeat: |
383 | read_lock(&journal->j_state_lock); | |
5c2178e7 | 384 | BUG_ON(journal->j_flags & JBD2_UNMOUNT); |
470decc6 | 385 | if (is_journal_aborted(journal) || |
f7f4bccb | 386 | (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) { |
a931da6a | 387 | read_unlock(&journal->j_state_lock); |
0c2022ec | 388 | jbd2_journal_free_transaction(new_transaction); |
47def826 | 389 | return -EROFS; |
470decc6 DK |
390 | } |
391 | ||
8f7d89f3 JK |
392 | /* |
393 | * Wait on the journal's transaction barrier if necessary. Specifically | |
394 | * we allow reserved handles to proceed because otherwise commit could | |
395 | * deadlock on page writeback not being able to complete. | |
396 | */ | |
397 | if (!handle->h_reserved && journal->j_barrier_count) { | |
a931da6a | 398 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
399 | wait_event(journal->j_wait_transaction_locked, |
400 | journal->j_barrier_count == 0); | |
401 | goto repeat; | |
402 | } | |
403 | ||
404 | if (!journal->j_running_transaction) { | |
a931da6a TT |
405 | read_unlock(&journal->j_state_lock); |
406 | if (!new_transaction) | |
470decc6 | 407 | goto alloc_transaction; |
a931da6a | 408 | write_lock(&journal->j_state_lock); |
d7961c7f | 409 | if (!journal->j_running_transaction && |
8f7d89f3 | 410 | (handle->h_reserved || !journal->j_barrier_count)) { |
a931da6a TT |
411 | jbd2_get_transaction(journal, new_transaction); |
412 | new_transaction = NULL; | |
470decc6 | 413 | } |
a931da6a TT |
414 | write_unlock(&journal->j_state_lock); |
415 | goto repeat; | |
470decc6 DK |
416 | } |
417 | ||
418 | transaction = journal->j_running_transaction; | |
419 | ||
8f7d89f3 JK |
420 | if (!handle->h_reserved) { |
421 | /* We may have dropped j_state_lock - restart in that case */ | |
b33d9f59 TT |
422 | if (add_transaction_credits(journal, blocks, rsv_blocks)) { |
423 | /* | |
424 | * add_transaction_credits releases | |
425 | * j_state_lock on a non-zero return | |
426 | */ | |
427 | __release(&journal->j_state_lock); | |
8f7d89f3 | 428 | goto repeat; |
b33d9f59 | 429 | } |
8f7d89f3 | 430 | } else { |
470decc6 | 431 | /* |
8f7d89f3 JK |
432 | * We have handle reserved so we are allowed to join T_LOCKED |
433 | * transaction and we don't have to check for transaction size | |
96f1e097 JK |
434 | * and journal space. But we still have to wait while running |
435 | * transaction is being switched to a committing one as it | |
436 | * won't wait for any handles anymore. | |
470decc6 | 437 | */ |
96f1e097 JK |
438 | if (transaction->t_state == T_SWITCH) { |
439 | wait_transaction_switching(journal); | |
440 | goto repeat; | |
441 | } | |
8f7d89f3 JK |
442 | sub_reserved_credits(journal, blocks); |
443 | handle->h_reserved = 0; | |
470decc6 DK |
444 | } |
445 | ||
446 | /* OK, account for the buffers that this operation expects to | |
4f981868 | 447 | * use and add the handle to the running transaction. |
8dd42046 | 448 | */ |
28e35e42 | 449 | update_t_max_wait(transaction, ts); |
470decc6 | 450 | handle->h_transaction = transaction; |
8f7d89f3 | 451 | handle->h_requested_credits = blocks; |
fdc3ef88 | 452 | handle->h_revoke_credits_requested = handle->h_revoke_credits; |
343d9c28 | 453 | handle->h_start_jiffies = jiffies; |
a51dca9c | 454 | atomic_inc(&transaction->t_updates); |
8dd42046 | 455 | atomic_inc(&transaction->t_handle_count); |
cb3b3bf2 | 456 | jbd2_debug(4, "Handle %p given %d credits (total %d, free %lu)\n", |
8f7d89f3 | 457 | handle, blocks, |
a51dca9c | 458 | atomic_read(&transaction->t_outstanding_credits), |
76c39904 | 459 | jbd2_log_space_left(journal)); |
a931da6a | 460 | read_unlock(&journal->j_state_lock); |
41a5b913 | 461 | current->journal_info = handle; |
9599b0e5 | 462 | |
ab714aff | 463 | rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_); |
0c2022ec | 464 | jbd2_journal_free_transaction(new_transaction); |
81378da6 MH |
465 | /* |
466 | * Ensure that no allocations done while the transaction is open are | |
467 | * going to recurse back to the fs layer. | |
468 | */ | |
469 | handle->saved_alloc_context = memalloc_nofs_save(); | |
47def826 | 470 | return 0; |
470decc6 DK |
471 | } |
472 | ||
473 | /* Allocate a new handle. This should probably be in a slab... */ | |
474 | static handle_t *new_handle(int nblocks) | |
475 | { | |
af1e76d6 | 476 | handle_t *handle = jbd2_alloc_handle(GFP_NOFS); |
470decc6 DK |
477 | if (!handle) |
478 | return NULL; | |
933f1c1e | 479 | handle->h_total_credits = nblocks; |
470decc6 DK |
480 | handle->h_ref = 1; |
481 | ||
482 | return handle; | |
483 | } | |
484 | ||
8f7d89f3 | 485 | handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks, |
fdc3ef88 JK |
486 | int revoke_records, gfp_t gfp_mask, |
487 | unsigned int type, unsigned int line_no) | |
470decc6 DK |
488 | { |
489 | handle_t *handle = journal_current_handle(); | |
490 | int err; | |
491 | ||
492 | if (!journal) | |
493 | return ERR_PTR(-EROFS); | |
494 | ||
495 | if (handle) { | |
496 | J_ASSERT(handle->h_transaction->t_journal == journal); | |
497 | handle->h_ref++; | |
498 | return handle; | |
499 | } | |
500 | ||
fdc3ef88 JK |
501 | nblocks += DIV_ROUND_UP(revoke_records, |
502 | journal->j_revoke_records_per_block); | |
470decc6 DK |
503 | handle = new_handle(nblocks); |
504 | if (!handle) | |
505 | return ERR_PTR(-ENOMEM); | |
8f7d89f3 JK |
506 | if (rsv_blocks) { |
507 | handle_t *rsv_handle; | |
508 | ||
509 | rsv_handle = new_handle(rsv_blocks); | |
510 | if (!rsv_handle) { | |
511 | jbd2_free_handle(handle); | |
512 | return ERR_PTR(-ENOMEM); | |
513 | } | |
514 | rsv_handle->h_reserved = 1; | |
515 | rsv_handle->h_journal = journal; | |
516 | handle->h_rsv_handle = rsv_handle; | |
517 | } | |
fdc3ef88 | 518 | handle->h_revoke_credits = revoke_records; |
470decc6 | 519 | |
47def826 | 520 | err = start_this_handle(journal, handle, gfp_mask); |
470decc6 | 521 | if (err < 0) { |
8f7d89f3 JK |
522 | if (handle->h_rsv_handle) |
523 | jbd2_free_handle(handle->h_rsv_handle); | |
af1e76d6 | 524 | jbd2_free_handle(handle); |
df05c1b8 | 525 | return ERR_PTR(err); |
470decc6 | 526 | } |
343d9c28 TT |
527 | handle->h_type = type; |
528 | handle->h_line_no = line_no; | |
529 | trace_jbd2_handle_start(journal->j_fs_dev->bd_dev, | |
530 | handle->h_transaction->t_tid, type, | |
531 | line_no, nblocks); | |
81378da6 | 532 | |
470decc6 DK |
533 | return handle; |
534 | } | |
47def826 TT |
535 | EXPORT_SYMBOL(jbd2__journal_start); |
536 | ||
537 | ||
91e4775d | 538 | /** |
2bf31d94 | 539 | * jbd2_journal_start() - Obtain a new handle. |
91e4775d MCC |
540 | * @journal: Journal to start transaction on. |
541 | * @nblocks: number of block buffer we might modify | |
542 | * | |
543 | * We make sure that the transaction can guarantee at least nblocks of | |
544 | * modified buffers in the log. We block until the log can guarantee | |
545 | * that much space. Additionally, if rsv_blocks > 0, we also create another | |
546 | * handle with rsv_blocks reserved blocks in the journal. This handle is | |
0c1cba6c | 547 | * stored in h_rsv_handle. It is not attached to any particular transaction |
91e4775d MCC |
548 | * and thus doesn't block transaction commit. If the caller uses this reserved |
549 | * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop() | |
550 | * on the parent handle will dispose the reserved one. Reserved handle has to | |
551 | * be converted to a normal handle using jbd2_journal_start_reserved() before | |
552 | * it can be used. | |
553 | * | |
554 | * Return a pointer to a newly allocated handle, or an ERR_PTR() value | |
555 | * on failure. | |
556 | */ | |
47def826 TT |
557 | handle_t *jbd2_journal_start(journal_t *journal, int nblocks) |
558 | { | |
fdc3ef88 | 559 | return jbd2__journal_start(journal, nblocks, 0, 0, GFP_NOFS, 0, 0); |
47def826 TT |
560 | } |
561 | EXPORT_SYMBOL(jbd2_journal_start); | |
562 | ||
14ff6286 | 563 | static void __jbd2_journal_unreserve_handle(handle_t *handle, transaction_t *t) |
8f7d89f3 JK |
564 | { |
565 | journal_t *journal = handle->h_journal; | |
566 | ||
567 | WARN_ON(!handle->h_reserved); | |
933f1c1e | 568 | sub_reserved_credits(journal, handle->h_total_credits); |
14ff6286 JK |
569 | if (t) |
570 | atomic_sub(handle->h_total_credits, &t->t_outstanding_credits); | |
ec8b6f60 JK |
571 | } |
572 | ||
573 | void jbd2_journal_free_reserved(handle_t *handle) | |
574 | { | |
14ff6286 JK |
575 | journal_t *journal = handle->h_journal; |
576 | ||
577 | /* Get j_state_lock to pin running transaction if it exists */ | |
578 | read_lock(&journal->j_state_lock); | |
579 | __jbd2_journal_unreserve_handle(handle, journal->j_running_transaction); | |
580 | read_unlock(&journal->j_state_lock); | |
8f7d89f3 JK |
581 | jbd2_free_handle(handle); |
582 | } | |
583 | EXPORT_SYMBOL(jbd2_journal_free_reserved); | |
584 | ||
585 | /** | |
2bf31d94 | 586 | * jbd2_journal_start_reserved() - start reserved handle |
8f7d89f3 | 587 | * @handle: handle to start |
f69120ce TH |
588 | * @type: for handle statistics |
589 | * @line_no: for handle statistics | |
8f7d89f3 JK |
590 | * |
591 | * Start handle that has been previously reserved with jbd2_journal_reserve(). | |
592 | * This attaches @handle to the running transaction (or creates one if there's | |
593 | * not transaction running). Unlike jbd2_journal_start() this function cannot | |
594 | * block on journal commit, checkpointing, or similar stuff. It can block on | |
595 | * memory allocation or frozen journal though. | |
596 | * | |
597 | * Return 0 on success, non-zero on error - handle is freed in that case. | |
598 | */ | |
599 | int jbd2_journal_start_reserved(handle_t *handle, unsigned int type, | |
600 | unsigned int line_no) | |
601 | { | |
602 | journal_t *journal = handle->h_journal; | |
603 | int ret = -EIO; | |
604 | ||
605 | if (WARN_ON(!handle->h_reserved)) { | |
606 | /* Someone passed in normal handle? Just stop it. */ | |
607 | jbd2_journal_stop(handle); | |
608 | return ret; | |
609 | } | |
610 | /* | |
611 | * Usefulness of mixing of reserved and unreserved handles is | |
612 | * questionable. So far nobody seems to need it so just error out. | |
613 | */ | |
614 | if (WARN_ON(current->journal_info)) { | |
615 | jbd2_journal_free_reserved(handle); | |
616 | return ret; | |
617 | } | |
618 | ||
619 | handle->h_journal = NULL; | |
8f7d89f3 JK |
620 | /* |
621 | * GFP_NOFS is here because callers are likely from writeback or | |
622 | * similarly constrained call sites | |
623 | */ | |
624 | ret = start_this_handle(journal, handle, GFP_NOFS); | |
92e3b405 | 625 | if (ret < 0) { |
b2569260 | 626 | handle->h_journal = journal; |
8f7d89f3 | 627 | jbd2_journal_free_reserved(handle); |
92e3b405 DC |
628 | return ret; |
629 | } | |
8f7d89f3 JK |
630 | handle->h_type = type; |
631 | handle->h_line_no = line_no; | |
4c273352 XW |
632 | trace_jbd2_handle_start(journal->j_fs_dev->bd_dev, |
633 | handle->h_transaction->t_tid, type, | |
933f1c1e | 634 | line_no, handle->h_total_credits); |
92e3b405 | 635 | return 0; |
8f7d89f3 JK |
636 | } |
637 | EXPORT_SYMBOL(jbd2_journal_start_reserved); | |
470decc6 DK |
638 | |
639 | /** | |
2bf31d94 | 640 | * jbd2_journal_extend() - extend buffer credits. |
470decc6 DK |
641 | * @handle: handle to 'extend' |
642 | * @nblocks: nr blocks to try to extend by. | |
fdc3ef88 | 643 | * @revoke_records: number of revoke records to try to extend by. |
470decc6 DK |
644 | * |
645 | * Some transactions, such as large extends and truncates, can be done | |
646 | * atomically all at once or in several stages. The operation requests | |
bd7ced98 | 647 | * a credit for a number of buffer modifications in advance, but can |
470decc6 DK |
648 | * extend its credit if it needs more. |
649 | * | |
f7f4bccb | 650 | * jbd2_journal_extend tries to give the running handle more buffer credits. |
470decc6 DK |
651 | * It does not guarantee that allocation - this is a best-effort only. |
652 | * The calling process MUST be able to deal cleanly with a failure to | |
653 | * extend here. | |
654 | * | |
655 | * Return 0 on success, non-zero on failure. | |
656 | * | |
657 | * return code < 0 implies an error | |
658 | * return code > 0 implies normal transaction-full status. | |
659 | */ | |
fdc3ef88 | 660 | int jbd2_journal_extend(handle_t *handle, int nblocks, int revoke_records) |
470decc6 DK |
661 | { |
662 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 663 | journal_t *journal; |
470decc6 DK |
664 | int result; |
665 | int wanted; | |
666 | ||
470decc6 | 667 | if (is_handle_aborted(handle)) |
41a5b913 TT |
668 | return -EROFS; |
669 | journal = transaction->t_journal; | |
470decc6 DK |
670 | |
671 | result = 1; | |
672 | ||
a931da6a | 673 | read_lock(&journal->j_state_lock); |
470decc6 DK |
674 | |
675 | /* Don't extend a locked-down transaction! */ | |
41a5b913 | 676 | if (transaction->t_state != T_RUNNING) { |
cb3b3bf2 | 677 | jbd2_debug(3, "denied handle %p %d blocks: " |
470decc6 DK |
678 | "transaction not running\n", handle, nblocks); |
679 | goto error_out; | |
680 | } | |
681 | ||
fdc3ef88 JK |
682 | nblocks += DIV_ROUND_UP( |
683 | handle->h_revoke_credits_requested + revoke_records, | |
684 | journal->j_revoke_records_per_block) - | |
685 | DIV_ROUND_UP( | |
686 | handle->h_revoke_credits_requested, | |
687 | journal->j_revoke_records_per_block); | |
fe1e8db5 JK |
688 | wanted = atomic_add_return(nblocks, |
689 | &transaction->t_outstanding_credits); | |
470decc6 DK |
690 | |
691 | if (wanted > journal->j_max_transaction_buffers) { | |
cb3b3bf2 | 692 | jbd2_debug(3, "denied handle %p %d blocks: " |
470decc6 | 693 | "transaction too large\n", handle, nblocks); |
fe1e8db5 | 694 | atomic_sub(nblocks, &transaction->t_outstanding_credits); |
f7f497cb | 695 | goto error_out; |
470decc6 DK |
696 | } |
697 | ||
343d9c28 | 698 | trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev, |
41a5b913 | 699 | transaction->t_tid, |
343d9c28 | 700 | handle->h_type, handle->h_line_no, |
933f1c1e | 701 | handle->h_total_credits, |
343d9c28 TT |
702 | nblocks); |
703 | ||
933f1c1e | 704 | handle->h_total_credits += nblocks; |
343d9c28 | 705 | handle->h_requested_credits += nblocks; |
fdc3ef88 JK |
706 | handle->h_revoke_credits += revoke_records; |
707 | handle->h_revoke_credits_requested += revoke_records; | |
470decc6 DK |
708 | result = 0; |
709 | ||
cb3b3bf2 | 710 | jbd2_debug(3, "extended handle %p by %d\n", handle, nblocks); |
470decc6 | 711 | error_out: |
a931da6a | 712 | read_unlock(&journal->j_state_lock); |
470decc6 DK |
713 | return result; |
714 | } | |
715 | ||
ec8b6f60 JK |
716 | static void stop_this_handle(handle_t *handle) |
717 | { | |
718 | transaction_t *transaction = handle->h_transaction; | |
719 | journal_t *journal = transaction->t_journal; | |
fdc3ef88 | 720 | int revokes; |
ec8b6f60 JK |
721 | |
722 | J_ASSERT(journal_current_handle() == handle); | |
723 | J_ASSERT(atomic_read(&transaction->t_updates) > 0); | |
724 | current->journal_info = NULL; | |
fdc3ef88 JK |
725 | /* |
726 | * Subtract necessary revoke descriptor blocks from handle credits. We | |
727 | * take care to account only for revoke descriptor blocks the | |
728 | * transaction will really need as large sequences of transactions with | |
729 | * small numbers of revokes are relatively common. | |
730 | */ | |
731 | revokes = handle->h_revoke_credits_requested - handle->h_revoke_credits; | |
732 | if (revokes) { | |
733 | int t_revokes, revoke_descriptors; | |
734 | int rr_per_blk = journal->j_revoke_records_per_block; | |
735 | ||
736 | WARN_ON_ONCE(DIV_ROUND_UP(revokes, rr_per_blk) | |
933f1c1e | 737 | > handle->h_total_credits); |
fdc3ef88 JK |
738 | t_revokes = atomic_add_return(revokes, |
739 | &transaction->t_outstanding_revokes); | |
740 | revoke_descriptors = | |
741 | DIV_ROUND_UP(t_revokes, rr_per_blk) - | |
742 | DIV_ROUND_UP(t_revokes - revokes, rr_per_blk); | |
933f1c1e | 743 | handle->h_total_credits -= revoke_descriptors; |
fdc3ef88 | 744 | } |
933f1c1e | 745 | atomic_sub(handle->h_total_credits, |
ec8b6f60 JK |
746 | &transaction->t_outstanding_credits); |
747 | if (handle->h_rsv_handle) | |
14ff6286 JK |
748 | __jbd2_journal_unreserve_handle(handle->h_rsv_handle, |
749 | transaction); | |
ec8b6f60 JK |
750 | if (atomic_dec_and_test(&transaction->t_updates)) |
751 | wake_up(&journal->j_wait_updates); | |
752 | ||
50b8b3f8 | 753 | rwsem_release(&journal->j_trans_commit_map, _THIS_IP_); |
ec8b6f60 JK |
754 | /* |
755 | * Scope of the GFP_NOFS context is over here and so we can restore the | |
756 | * original alloc context. | |
757 | */ | |
758 | memalloc_nofs_restore(handle->saved_alloc_context); | |
759 | } | |
470decc6 DK |
760 | |
761 | /** | |
2bf31d94 | 762 | * jbd2__journal_restart() - restart a handle . |
470decc6 DK |
763 | * @handle: handle to restart |
764 | * @nblocks: nr credits requested | |
fdc3ef88 | 765 | * @revoke_records: number of revoke record credits requested |
f69120ce | 766 | * @gfp_mask: memory allocation flags (for start_this_handle) |
470decc6 DK |
767 | * |
768 | * Restart a handle for a multi-transaction filesystem | |
769 | * operation. | |
770 | * | |
f7f4bccb MC |
771 | * If the jbd2_journal_extend() call above fails to grant new buffer credits |
772 | * to a running handle, a call to jbd2_journal_restart will commit the | |
470decc6 | 773 | * handle's transaction so far and reattach the handle to a new |
bd7ced98 | 774 | * transaction capable of guaranteeing the requested number of |
8f7d89f3 JK |
775 | * credits. We preserve reserved handle if there's any attached to the |
776 | * passed in handle. | |
470decc6 | 777 | */ |
fdc3ef88 JK |
778 | int jbd2__journal_restart(handle_t *handle, int nblocks, int revoke_records, |
779 | gfp_t gfp_mask) | |
470decc6 DK |
780 | { |
781 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 782 | journal_t *journal; |
e4471831 | 783 | tid_t tid; |
ec8b6f60 | 784 | int need_to_start; |
0094f981 | 785 | int ret; |
470decc6 DK |
786 | |
787 | /* If we've had an abort of any type, don't even think about | |
788 | * actually doing the restart! */ | |
789 | if (is_handle_aborted(handle)) | |
790 | return 0; | |
41a5b913 | 791 | journal = transaction->t_journal; |
ec8b6f60 | 792 | tid = transaction->t_tid; |
470decc6 DK |
793 | |
794 | /* | |
795 | * First unlink the handle from its current transaction, and start the | |
796 | * commit on that. | |
797 | */ | |
cb3b3bf2 | 798 | jbd2_debug(2, "restarting handle %p\n", handle); |
ec8b6f60 | 799 | stop_this_handle(handle); |
41a5b913 | 800 | handle->h_transaction = NULL; |
470decc6 | 801 | |
ec8b6f60 JK |
802 | /* |
803 | * TODO: If we use READ_ONCE / WRITE_ONCE for j_commit_request we can | |
804 | * get rid of pointless j_state_lock traffic like this. | |
805 | */ | |
806 | read_lock(&journal->j_state_lock); | |
e4471831 | 807 | need_to_start = !tid_geq(journal->j_commit_request, tid); |
a931da6a | 808 | read_unlock(&journal->j_state_lock); |
e4471831 TT |
809 | if (need_to_start) |
810 | jbd2_log_start_commit(journal, tid); | |
933f1c1e | 811 | handle->h_total_credits = nblocks + |
fdc3ef88 JK |
812 | DIV_ROUND_UP(revoke_records, |
813 | journal->j_revoke_records_per_block); | |
814 | handle->h_revoke_credits = revoke_records; | |
47def826 | 815 | ret = start_this_handle(journal, handle, gfp_mask); |
0094f981 JK |
816 | trace_jbd2_handle_restart(journal->j_fs_dev->bd_dev, |
817 | ret ? 0 : handle->h_transaction->t_tid, | |
818 | handle->h_type, handle->h_line_no, | |
819 | handle->h_total_credits); | |
470decc6 DK |
820 | return ret; |
821 | } | |
47def826 | 822 | EXPORT_SYMBOL(jbd2__journal_restart); |
470decc6 DK |
823 | |
824 | ||
47def826 TT |
825 | int jbd2_journal_restart(handle_t *handle, int nblocks) |
826 | { | |
fdc3ef88 | 827 | return jbd2__journal_restart(handle, nblocks, 0, GFP_NOFS); |
47def826 TT |
828 | } |
829 | EXPORT_SYMBOL(jbd2_journal_restart); | |
830 | ||
4f981868 RH |
831 | /* |
832 | * Waits for any outstanding t_updates to finish. | |
833 | * This is called with write j_state_lock held. | |
834 | */ | |
835 | void jbd2_journal_wait_updates(journal_t *journal) | |
836 | { | |
cc16eeca | 837 | DEFINE_WAIT(wait); |
4f981868 | 838 | |
cc16eeca RH |
839 | while (1) { |
840 | /* | |
841 | * Note that the running transaction can get freed under us if | |
842 | * this transaction is getting committed in | |
843 | * jbd2_journal_commit_transaction() -> | |
844 | * jbd2_journal_free_transaction(). This can only happen when we | |
845 | * release j_state_lock -> schedule() -> acquire j_state_lock. | |
846 | * Hence we should everytime retrieve new j_running_transaction | |
847 | * value (after j_state_lock release acquire cycle), else it may | |
848 | * lead to use-after-free of old freed transaction. | |
849 | */ | |
850 | transaction_t *transaction = journal->j_running_transaction; | |
4f981868 | 851 | |
cc16eeca RH |
852 | if (!transaction) |
853 | break; | |
4f981868 RH |
854 | |
855 | prepare_to_wait(&journal->j_wait_updates, &wait, | |
cc16eeca RH |
856 | TASK_UNINTERRUPTIBLE); |
857 | if (!atomic_read(&transaction->t_updates)) { | |
cc16eeca RH |
858 | finish_wait(&journal->j_wait_updates, &wait); |
859 | break; | |
4f981868 | 860 | } |
cc16eeca RH |
861 | write_unlock(&journal->j_state_lock); |
862 | schedule(); | |
4f981868 | 863 | finish_wait(&journal->j_wait_updates, &wait); |
cc16eeca | 864 | write_lock(&journal->j_state_lock); |
4f981868 | 865 | } |
4f981868 RH |
866 | } |
867 | ||
470decc6 | 868 | /** |
2bf31d94 | 869 | * jbd2_journal_lock_updates () - establish a transaction barrier. |
470decc6 DK |
870 | * @journal: Journal to establish a barrier on. |
871 | * | |
872 | * This locks out any further updates from being started, and blocks | |
873 | * until all existing updates have completed, returning only once the | |
874 | * journal is in a quiescent state with no updates running. | |
875 | * | |
876 | * The journal lock should not be held on entry. | |
877 | */ | |
f7f4bccb | 878 | void jbd2_journal_lock_updates(journal_t *journal) |
470decc6 | 879 | { |
1eaa566d JK |
880 | jbd2_might_wait_for_commit(journal); |
881 | ||
a931da6a | 882 | write_lock(&journal->j_state_lock); |
470decc6 DK |
883 | ++journal->j_barrier_count; |
884 | ||
8f7d89f3 JK |
885 | /* Wait until there are no reserved handles */ |
886 | if (atomic_read(&journal->j_reserved_credits)) { | |
887 | write_unlock(&journal->j_state_lock); | |
888 | wait_event(journal->j_wait_reserved, | |
889 | atomic_read(&journal->j_reserved_credits) == 0); | |
890 | write_lock(&journal->j_state_lock); | |
891 | } | |
892 | ||
4f981868 RH |
893 | /* Wait until there are no running t_updates */ |
894 | jbd2_journal_wait_updates(journal); | |
470decc6 | 895 | |
a931da6a | 896 | write_unlock(&journal->j_state_lock); |
470decc6 DK |
897 | |
898 | /* | |
899 | * We have now established a barrier against other normal updates, but | |
f7f4bccb | 900 | * we also need to barrier against other jbd2_journal_lock_updates() calls |
470decc6 DK |
901 | * to make sure that we serialise special journal-locked operations |
902 | * too. | |
903 | */ | |
904 | mutex_lock(&journal->j_barrier); | |
905 | } | |
906 | ||
907 | /** | |
2bf31d94 | 908 | * jbd2_journal_unlock_updates () - release barrier |
470decc6 DK |
909 | * @journal: Journal to release the barrier on. |
910 | * | |
f7f4bccb | 911 | * Release a transaction barrier obtained with jbd2_journal_lock_updates(). |
470decc6 DK |
912 | * |
913 | * Should be called without the journal lock held. | |
914 | */ | |
f7f4bccb | 915 | void jbd2_journal_unlock_updates (journal_t *journal) |
470decc6 DK |
916 | { |
917 | J_ASSERT(journal->j_barrier_count != 0); | |
918 | ||
919 | mutex_unlock(&journal->j_barrier); | |
a931da6a | 920 | write_lock(&journal->j_state_lock); |
470decc6 | 921 | --journal->j_barrier_count; |
a931da6a | 922 | write_unlock(&journal->j_state_lock); |
34fc8768 | 923 | wake_up_all(&journal->j_wait_transaction_locked); |
470decc6 DK |
924 | } |
925 | ||
f91d1d04 | 926 | static void warn_dirty_buffer(struct buffer_head *bh) |
470decc6 | 927 | { |
f91d1d04 | 928 | printk(KERN_WARNING |
a1c6f057 | 929 | "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). " |
f91d1d04 JK |
930 | "There's a risk of filesystem corruption in case of system " |
931 | "crash.\n", | |
a1c6f057 | 932 | bh->b_bdev, (unsigned long long)bh->b_blocknr); |
470decc6 DK |
933 | } |
934 | ||
ee57aba1 JK |
935 | /* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */ |
936 | static void jbd2_freeze_jh_data(struct journal_head *jh) | |
937 | { | |
938 | struct page *page; | |
939 | int offset; | |
940 | char *source; | |
941 | struct buffer_head *bh = jh2bh(jh); | |
942 | ||
943 | J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n"); | |
944 | page = bh->b_page; | |
945 | offset = offset_in_page(bh->b_data); | |
946 | source = kmap_atomic(page); | |
947 | /* Fire data frozen trigger just before we copy the data */ | |
948 | jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers); | |
949 | memcpy(jh->b_frozen_data, source + offset, bh->b_size); | |
950 | kunmap_atomic(source); | |
951 | ||
952 | /* | |
953 | * Now that the frozen data is saved off, we need to store any matching | |
954 | * triggers. | |
955 | */ | |
956 | jh->b_frozen_triggers = jh->b_triggers; | |
957 | } | |
958 | ||
470decc6 DK |
959 | /* |
960 | * If the buffer is already part of the current transaction, then there | |
961 | * is nothing we need to do. If it is already part of a prior | |
962 | * transaction which we are still committing to disk, then we need to | |
963 | * make sure that we do not overwrite the old copy: we do copy-out to | |
964 | * preserve the copy going to disk. We also account the buffer against | |
965 | * the handle's metadata buffer credits (unless the buffer is already | |
966 | * part of the transaction, that is). | |
967 | * | |
968 | */ | |
969 | static int | |
970 | do_get_write_access(handle_t *handle, struct journal_head *jh, | |
971 | int force_copy) | |
972 | { | |
973 | struct buffer_head *bh; | |
41a5b913 | 974 | transaction_t *transaction = handle->h_transaction; |
470decc6 DK |
975 | journal_t *journal; |
976 | int error; | |
977 | char *frozen_buffer = NULL; | |
f783f091 | 978 | unsigned long start_lock, time_lock; |
470decc6 | 979 | |
470decc6 DK |
980 | journal = transaction->t_journal; |
981 | ||
cb3b3bf2 | 982 | jbd2_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy); |
470decc6 DK |
983 | |
984 | JBUFFER_TRACE(jh, "entry"); | |
985 | repeat: | |
986 | bh = jh2bh(jh); | |
987 | ||
988 | /* @@@ Need to check for errors here at some point. */ | |
989 | ||
f783f091 | 990 | start_lock = jiffies; |
470decc6 | 991 | lock_buffer(bh); |
46417064 | 992 | spin_lock(&jh->b_state_lock); |
470decc6 | 993 | |
f783f091 TT |
994 | /* If it takes too long to lock the buffer, trace it */ |
995 | time_lock = jbd2_time_diff(start_lock, jiffies); | |
996 | if (time_lock > HZ/10) | |
997 | trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev, | |
998 | jiffies_to_msecs(time_lock)); | |
999 | ||
470decc6 DK |
1000 | /* We now hold the buffer lock so it is safe to query the buffer |
1001 | * state. Is the buffer dirty? | |
1002 | * | |
1003 | * If so, there are two possibilities. The buffer may be | |
1004 | * non-journaled, and undergoing a quite legitimate writeback. | |
1005 | * Otherwise, it is journaled, and we don't expect dirty buffers | |
1006 | * in that state (the buffers should be marked JBD_Dirty | |
1007 | * instead.) So either the IO is being done under our own | |
1008 | * control and this is a bug, or it's a third party IO such as | |
1009 | * dump(8) (which may leave the buffer scheduled for read --- | |
1010 | * ie. locked but not dirty) or tune2fs (which may actually have | |
1011 | * the buffer dirtied, ugh.) */ | |
1012 | ||
e6b9bd72 ZC |
1013 | if (buffer_dirty(bh) && jh->b_transaction) { |
1014 | warn_dirty_buffer(bh); | |
470decc6 | 1015 | /* |
e6b9bd72 ZC |
1016 | * We need to clean the dirty flag and we must do it under the |
1017 | * buffer lock to be sure we don't race with running write-out. | |
470decc6 | 1018 | */ |
f91d1d04 JK |
1019 | JBUFFER_TRACE(jh, "Journalling dirty buffer"); |
1020 | clear_buffer_dirty(bh); | |
e6b9bd72 ZC |
1021 | /* |
1022 | * The buffer is going to be added to BJ_Reserved list now and | |
1023 | * nothing guarantees jbd2_journal_dirty_metadata() will be | |
1024 | * ever called for it. So we need to set jbddirty bit here to | |
1025 | * make sure the buffer is dirtied and written out when the | |
1026 | * journaling machinery is done with it. | |
1027 | */ | |
f91d1d04 | 1028 | set_buffer_jbddirty(bh); |
470decc6 DK |
1029 | } |
1030 | ||
470decc6 DK |
1031 | error = -EROFS; |
1032 | if (is_handle_aborted(handle)) { | |
46417064 | 1033 | spin_unlock(&jh->b_state_lock); |
e6b9bd72 | 1034 | unlock_buffer(bh); |
470decc6 DK |
1035 | goto out; |
1036 | } | |
1037 | error = 0; | |
1038 | ||
1039 | /* | |
1040 | * The buffer is already part of this transaction if b_transaction or | |
1041 | * b_next_transaction points to it | |
1042 | */ | |
1043 | if (jh->b_transaction == transaction || | |
e6b9bd72 ZC |
1044 | jh->b_next_transaction == transaction) { |
1045 | unlock_buffer(bh); | |
470decc6 | 1046 | goto done; |
e6b9bd72 | 1047 | } |
470decc6 | 1048 | |
9fc7c63a JB |
1049 | /* |
1050 | * this is the first time this transaction is touching this buffer, | |
1051 | * reset the modified flag | |
1052 | */ | |
561405f0 | 1053 | jh->b_modified = 0; |
9fc7c63a | 1054 | |
8b00f400 JK |
1055 | /* |
1056 | * If the buffer is not journaled right now, we need to make sure it | |
1057 | * doesn't get written to disk before the caller actually commits the | |
1058 | * new data | |
1059 | */ | |
1060 | if (!jh->b_transaction) { | |
1061 | JBUFFER_TRACE(jh, "no transaction"); | |
1062 | J_ASSERT_JH(jh, !jh->b_next_transaction); | |
1063 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); | |
de92c8ca JK |
1064 | /* |
1065 | * Make sure all stores to jh (b_modified, b_frozen_data) are | |
1066 | * visible before attaching it to the running transaction. | |
1067 | * Paired with barrier in jbd2_write_access_granted() | |
1068 | */ | |
1069 | smp_wmb(); | |
8b00f400 | 1070 | spin_lock(&journal->j_list_lock); |
e6b9bd72 ZC |
1071 | if (test_clear_buffer_dirty(bh)) { |
1072 | /* | |
1073 | * Execute buffer dirty clearing and jh->b_transaction | |
1074 | * assignment under journal->j_list_lock locked to | |
1075 | * prevent bh being removed from checkpoint list if | |
1076 | * the buffer is in an intermediate state (not dirty | |
1077 | * and jh->b_transaction is NULL). | |
1078 | */ | |
1079 | JBUFFER_TRACE(jh, "Journalling dirty buffer"); | |
1080 | set_buffer_jbddirty(bh); | |
1081 | } | |
8b00f400 JK |
1082 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
1083 | spin_unlock(&journal->j_list_lock); | |
e6b9bd72 | 1084 | unlock_buffer(bh); |
8b00f400 JK |
1085 | goto done; |
1086 | } | |
e6b9bd72 ZC |
1087 | unlock_buffer(bh); |
1088 | ||
470decc6 DK |
1089 | /* |
1090 | * If there is already a copy-out version of this buffer, then we don't | |
1091 | * need to make another one | |
1092 | */ | |
1093 | if (jh->b_frozen_data) { | |
1094 | JBUFFER_TRACE(jh, "has frozen data"); | |
1095 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
de92c8ca | 1096 | goto attach_next; |
470decc6 DK |
1097 | } |
1098 | ||
8b00f400 JK |
1099 | JBUFFER_TRACE(jh, "owned by older transaction"); |
1100 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
1101 | J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction); | |
470decc6 | 1102 | |
8b00f400 JK |
1103 | /* |
1104 | * There is one case we have to be very careful about. If the | |
1105 | * committing transaction is currently writing this buffer out to disk | |
1106 | * and has NOT made a copy-out, then we cannot modify the buffer | |
1107 | * contents at all right now. The essence of copy-out is that it is | |
1108 | * the extra copy, not the primary copy, which gets journaled. If the | |
1109 | * primary copy is already going to disk then we cannot do copy-out | |
1110 | * here. | |
1111 | */ | |
1112 | if (buffer_shadow(bh)) { | |
1113 | JBUFFER_TRACE(jh, "on shadow: sleep"); | |
46417064 | 1114 | spin_unlock(&jh->b_state_lock); |
8b00f400 JK |
1115 | wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE); |
1116 | goto repeat; | |
1117 | } | |
470decc6 | 1118 | |
8b00f400 JK |
1119 | /* |
1120 | * Only do the copy if the currently-owning transaction still needs it. | |
1121 | * If buffer isn't on BJ_Metadata list, the committing transaction is | |
1122 | * past that stage (here we use the fact that BH_Shadow is set under | |
1123 | * bh_state lock together with refiling to BJ_Shadow list and at this | |
1124 | * point we know the buffer doesn't have BH_Shadow set). | |
1125 | * | |
1126 | * Subtle point, though: if this is a get_undo_access, then we will be | |
1127 | * relying on the frozen_data to contain the new value of the | |
1128 | * committed_data record after the transaction, so we HAVE to force the | |
1129 | * frozen_data copy in that case. | |
1130 | */ | |
1131 | if (jh->b_jlist == BJ_Metadata || force_copy) { | |
1132 | JBUFFER_TRACE(jh, "generate frozen data"); | |
1133 | if (!frozen_buffer) { | |
1134 | JBUFFER_TRACE(jh, "allocate memory for buffer"); | |
46417064 | 1135 | spin_unlock(&jh->b_state_lock); |
490c1b44 MH |
1136 | frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size, |
1137 | GFP_NOFS | __GFP_NOFAIL); | |
8b00f400 | 1138 | goto repeat; |
470decc6 | 1139 | } |
8b00f400 JK |
1140 | jh->b_frozen_data = frozen_buffer; |
1141 | frozen_buffer = NULL; | |
1142 | jbd2_freeze_jh_data(jh); | |
470decc6 | 1143 | } |
de92c8ca JK |
1144 | attach_next: |
1145 | /* | |
1146 | * Make sure all stores to jh (b_modified, b_frozen_data) are visible | |
1147 | * before attaching it to the running transaction. Paired with barrier | |
1148 | * in jbd2_write_access_granted() | |
1149 | */ | |
1150 | smp_wmb(); | |
8b00f400 | 1151 | jh->b_next_transaction = transaction; |
470decc6 DK |
1152 | |
1153 | done: | |
46417064 | 1154 | spin_unlock(&jh->b_state_lock); |
470decc6 DK |
1155 | |
1156 | /* | |
1157 | * If we are about to journal a buffer, then any revoke pending on it is | |
1158 | * no longer valid | |
1159 | */ | |
f7f4bccb | 1160 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 DK |
1161 | |
1162 | out: | |
1163 | if (unlikely(frozen_buffer)) /* It's usually NULL */ | |
af1e76d6 | 1164 | jbd2_free(frozen_buffer, bh->b_size); |
470decc6 DK |
1165 | |
1166 | JBUFFER_TRACE(jh, "exit"); | |
1167 | return error; | |
1168 | } | |
1169 | ||
de92c8ca | 1170 | /* Fast check whether buffer is already attached to the required transaction */ |
087ffd4e JB |
1171 | static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh, |
1172 | bool undo) | |
de92c8ca JK |
1173 | { |
1174 | struct journal_head *jh; | |
1175 | bool ret = false; | |
1176 | ||
1177 | /* Dirty buffers require special handling... */ | |
1178 | if (buffer_dirty(bh)) | |
1179 | return false; | |
1180 | ||
1181 | /* | |
1182 | * RCU protects us from dereferencing freed pages. So the checks we do | |
1183 | * are guaranteed not to oops. However the jh slab object can get freed | |
1184 | * & reallocated while we work with it. So we have to be careful. When | |
1185 | * we see jh attached to the running transaction, we know it must stay | |
1186 | * so until the transaction is committed. Thus jh won't be freed and | |
1187 | * will be attached to the same bh while we run. However it can | |
1188 | * happen jh gets freed, reallocated, and attached to the transaction | |
1189 | * just after we get pointer to it from bh. So we have to be careful | |
1190 | * and recheck jh still belongs to our bh before we return success. | |
1191 | */ | |
1192 | rcu_read_lock(); | |
1193 | if (!buffer_jbd(bh)) | |
1194 | goto out; | |
1195 | /* This should be bh2jh() but that doesn't work with inline functions */ | |
1196 | jh = READ_ONCE(bh->b_private); | |
1197 | if (!jh) | |
1198 | goto out; | |
087ffd4e JB |
1199 | /* For undo access buffer must have data copied */ |
1200 | if (undo && !jh->b_committed_data) | |
1201 | goto out; | |
6c5d9112 QC |
1202 | if (READ_ONCE(jh->b_transaction) != handle->h_transaction && |
1203 | READ_ONCE(jh->b_next_transaction) != handle->h_transaction) | |
de92c8ca JK |
1204 | goto out; |
1205 | /* | |
1206 | * There are two reasons for the barrier here: | |
1207 | * 1) Make sure to fetch b_bh after we did previous checks so that we | |
1208 | * detect when jh went through free, realloc, attach to transaction | |
1209 | * while we were checking. Paired with implicit barrier in that path. | |
1210 | * 2) So that access to bh done after jbd2_write_access_granted() | |
1211 | * doesn't get reordered and see inconsistent state of concurrent | |
1212 | * do_get_write_access(). | |
1213 | */ | |
1214 | smp_mb(); | |
1215 | if (unlikely(jh->b_bh != bh)) | |
1216 | goto out; | |
1217 | ret = true; | |
1218 | out: | |
1219 | rcu_read_unlock(); | |
1220 | return ret; | |
1221 | } | |
1222 | ||
470decc6 | 1223 | /** |
2bf31d94 MCC |
1224 | * jbd2_journal_get_write_access() - notify intent to modify a buffer |
1225 | * for metadata (not data) update. | |
470decc6 DK |
1226 | * @handle: transaction to add buffer modifications to |
1227 | * @bh: bh to be used for metadata writes | |
470decc6 | 1228 | * |
df1b560a | 1229 | * Returns: error code or 0 on success. |
470decc6 DK |
1230 | * |
1231 | * In full data journalling mode the buffer may be of type BJ_AsyncData, | |
df1b560a | 1232 | * because we're ``write()ing`` a buffer which is also part of a shared mapping. |
470decc6 DK |
1233 | */ |
1234 | ||
f7f4bccb | 1235 | int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh) |
470decc6 | 1236 | { |
de92c8ca | 1237 | struct journal_head *jh; |
470decc6 DK |
1238 | int rc; |
1239 | ||
8eedabfd | 1240 | if (is_handle_aborted(handle)) |
1241 | return -EROFS; | |
1242 | ||
087ffd4e | 1243 | if (jbd2_write_access_granted(handle, bh, false)) |
de92c8ca JK |
1244 | return 0; |
1245 | ||
1246 | jh = jbd2_journal_add_journal_head(bh); | |
470decc6 DK |
1247 | /* We do not want to get caught playing with fields which the |
1248 | * log thread also manipulates. Make sure that the buffer | |
1249 | * completes any outstanding IO before proceeding. */ | |
1250 | rc = do_get_write_access(handle, jh, 0); | |
f7f4bccb | 1251 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1252 | return rc; |
1253 | } | |
1254 | ||
1255 | ||
1256 | /* | |
1257 | * When the user wants to journal a newly created buffer_head | |
1258 | * (ie. getblk() returned a new buffer and we are going to populate it | |
1259 | * manually rather than reading off disk), then we need to keep the | |
1260 | * buffer_head locked until it has been completely filled with new | |
1261 | * data. In this case, we should be able to make the assertion that | |
1262 | * the bh is not already part of an existing transaction. | |
1263 | * | |
1264 | * The buffer should already be locked by the caller by this point. | |
1265 | * There is no lock ranking violation: it was a newly created, | |
1266 | * unlocked buffer beforehand. */ | |
1267 | ||
1268 | /** | |
2bf31d94 | 1269 | * jbd2_journal_get_create_access () - notify intent to use newly created bh |
470decc6 DK |
1270 | * @handle: transaction to new buffer to |
1271 | * @bh: new buffer. | |
1272 | * | |
1273 | * Call this if you create a new bh. | |
1274 | */ | |
f7f4bccb | 1275 | int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1276 | { |
1277 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 1278 | journal_t *journal; |
f7f4bccb | 1279 | struct journal_head *jh = jbd2_journal_add_journal_head(bh); |
470decc6 DK |
1280 | int err; |
1281 | ||
cb3b3bf2 | 1282 | jbd2_debug(5, "journal_head %p\n", jh); |
470decc6 DK |
1283 | err = -EROFS; |
1284 | if (is_handle_aborted(handle)) | |
1285 | goto out; | |
41a5b913 | 1286 | journal = transaction->t_journal; |
470decc6 DK |
1287 | err = 0; |
1288 | ||
1289 | JBUFFER_TRACE(jh, "entry"); | |
1290 | /* | |
1291 | * The buffer may already belong to this transaction due to pre-zeroing | |
1292 | * in the filesystem's new_block code. It may also be on the previous, | |
1293 | * committing transaction's lists, but it HAS to be in Forget state in | |
1294 | * that case: the transaction must have deleted the buffer for it to be | |
1295 | * reused here. | |
1296 | */ | |
46417064 | 1297 | spin_lock(&jh->b_state_lock); |
470decc6 DK |
1298 | J_ASSERT_JH(jh, (jh->b_transaction == transaction || |
1299 | jh->b_transaction == NULL || | |
1300 | (jh->b_transaction == journal->j_committing_transaction && | |
1301 | jh->b_jlist == BJ_Forget))); | |
1302 | ||
1303 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
1304 | J_ASSERT_JH(jh, buffer_locked(jh2bh(jh))); | |
1305 | ||
1306 | if (jh->b_transaction == NULL) { | |
f91d1d04 JK |
1307 | /* |
1308 | * Previous jbd2_journal_forget() could have left the buffer | |
1309 | * with jbddirty bit set because it was being committed. When | |
1310 | * the commit finished, we've filed the buffer for | |
1311 | * checkpointing and marked it dirty. Now we are reallocating | |
1312 | * the buffer so the transaction freeing it must have | |
1313 | * committed and so it's safe to clear the dirty bit. | |
1314 | */ | |
1315 | clear_buffer_dirty(jh2bh(jh)); | |
9fc7c63a JB |
1316 | /* first access by this transaction */ |
1317 | jh->b_modified = 0; | |
1318 | ||
470decc6 | 1319 | JBUFFER_TRACE(jh, "file as BJ_Reserved"); |
6e4862a5 | 1320 | spin_lock(&journal->j_list_lock); |
f7f4bccb | 1321 | __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved); |
559cce69 | 1322 | spin_unlock(&journal->j_list_lock); |
470decc6 | 1323 | } else if (jh->b_transaction == journal->j_committing_transaction) { |
9fc7c63a JB |
1324 | /* first access by this transaction */ |
1325 | jh->b_modified = 0; | |
1326 | ||
470decc6 | 1327 | JBUFFER_TRACE(jh, "set next transaction"); |
6e4862a5 | 1328 | spin_lock(&journal->j_list_lock); |
470decc6 | 1329 | jh->b_next_transaction = transaction; |
559cce69 | 1330 | spin_unlock(&journal->j_list_lock); |
470decc6 | 1331 | } |
46417064 | 1332 | spin_unlock(&jh->b_state_lock); |
470decc6 DK |
1333 | |
1334 | /* | |
1335 | * akpm: I added this. ext3_alloc_branch can pick up new indirect | |
1336 | * blocks which contain freed but then revoked metadata. We need | |
1337 | * to cancel the revoke in case we end up freeing it yet again | |
1338 | * and the reallocating as data - this would cause a second revoke, | |
1339 | * which hits an assertion error. | |
1340 | */ | |
1341 | JBUFFER_TRACE(jh, "cancelling revoke"); | |
f7f4bccb | 1342 | jbd2_journal_cancel_revoke(handle, jh); |
470decc6 | 1343 | out: |
3991b400 | 1344 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
1345 | return err; |
1346 | } | |
1347 | ||
1348 | /** | |
2bf31d94 | 1349 | * jbd2_journal_get_undo_access() - Notify intent to modify metadata with |
470decc6 DK |
1350 | * non-rewindable consequences |
1351 | * @handle: transaction | |
1352 | * @bh: buffer to undo | |
470decc6 DK |
1353 | * |
1354 | * Sometimes there is a need to distinguish between metadata which has | |
1355 | * been committed to disk and that which has not. The ext3fs code uses | |
1356 | * this for freeing and allocating space, we have to make sure that we | |
1357 | * do not reuse freed space until the deallocation has been committed, | |
1358 | * since if we overwrote that space we would make the delete | |
1359 | * un-rewindable in case of a crash. | |
1360 | * | |
f7f4bccb | 1361 | * To deal with that, jbd2_journal_get_undo_access requests write access to a |
470decc6 DK |
1362 | * buffer for parts of non-rewindable operations such as delete |
1363 | * operations on the bitmaps. The journaling code must keep a copy of | |
1364 | * the buffer's contents prior to the undo_access call until such time | |
1365 | * as we know that the buffer has definitely been committed to disk. | |
1366 | * | |
1367 | * We never need to know which transaction the committed data is part | |
1368 | * of, buffers touched here are guaranteed to be dirtied later and so | |
1369 | * will be committed to a new transaction in due course, at which point | |
1370 | * we can discard the old committed data pointer. | |
1371 | * | |
1372 | * Returns error number or 0 on success. | |
1373 | */ | |
f7f4bccb | 1374 | int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1375 | { |
1376 | int err; | |
de92c8ca | 1377 | struct journal_head *jh; |
470decc6 DK |
1378 | char *committed_data = NULL; |
1379 | ||
8eedabfd | 1380 | if (is_handle_aborted(handle)) |
1381 | return -EROFS; | |
1382 | ||
087ffd4e | 1383 | if (jbd2_write_access_granted(handle, bh, true)) |
de92c8ca | 1384 | return 0; |
470decc6 | 1385 | |
de92c8ca | 1386 | jh = jbd2_journal_add_journal_head(bh); |
01215d3e | 1387 | JBUFFER_TRACE(jh, "entry"); |
1388 | ||
470decc6 DK |
1389 | /* |
1390 | * Do this first --- it can drop the journal lock, so we want to | |
1391 | * make sure that obtaining the committed_data is done | |
1392 | * atomically wrt. completion of any outstanding commits. | |
1393 | */ | |
1394 | err = do_get_write_access(handle, jh, 1); | |
1395 | if (err) | |
1396 | goto out; | |
1397 | ||
1398 | repeat: | |
490c1b44 MH |
1399 | if (!jh->b_committed_data) |
1400 | committed_data = jbd2_alloc(jh2bh(jh)->b_size, | |
1401 | GFP_NOFS|__GFP_NOFAIL); | |
470decc6 | 1402 | |
46417064 | 1403 | spin_lock(&jh->b_state_lock); |
470decc6 DK |
1404 | if (!jh->b_committed_data) { |
1405 | /* Copy out the current buffer contents into the | |
1406 | * preserved, committed copy. */ | |
1407 | JBUFFER_TRACE(jh, "generate b_committed data"); | |
1408 | if (!committed_data) { | |
46417064 | 1409 | spin_unlock(&jh->b_state_lock); |
470decc6 DK |
1410 | goto repeat; |
1411 | } | |
1412 | ||
1413 | jh->b_committed_data = committed_data; | |
1414 | committed_data = NULL; | |
1415 | memcpy(jh->b_committed_data, bh->b_data, bh->b_size); | |
1416 | } | |
46417064 | 1417 | spin_unlock(&jh->b_state_lock); |
470decc6 | 1418 | out: |
f7f4bccb | 1419 | jbd2_journal_put_journal_head(jh); |
470decc6 | 1420 | if (unlikely(committed_data)) |
af1e76d6 | 1421 | jbd2_free(committed_data, bh->b_size); |
470decc6 DK |
1422 | return err; |
1423 | } | |
1424 | ||
e06c8227 | 1425 | /** |
2bf31d94 | 1426 | * jbd2_journal_set_triggers() - Add triggers for commit writeout |
e06c8227 JB |
1427 | * @bh: buffer to trigger on |
1428 | * @type: struct jbd2_buffer_trigger_type containing the trigger(s). | |
1429 | * | |
1430 | * Set any triggers on this journal_head. This is always safe, because | |
1431 | * triggers for a committing buffer will be saved off, and triggers for | |
1432 | * a running transaction will match the buffer in that transaction. | |
1433 | * | |
1434 | * Call with NULL to clear the triggers. | |
1435 | */ | |
1436 | void jbd2_journal_set_triggers(struct buffer_head *bh, | |
1437 | struct jbd2_buffer_trigger_type *type) | |
1438 | { | |
ad56edad | 1439 | struct journal_head *jh = jbd2_journal_grab_journal_head(bh); |
e06c8227 | 1440 | |
188c299e | 1441 | if (WARN_ON_ONCE(!jh)) |
ad56edad | 1442 | return; |
e06c8227 | 1443 | jh->b_triggers = type; |
ad56edad | 1444 | jbd2_journal_put_journal_head(jh); |
e06c8227 JB |
1445 | } |
1446 | ||
13ceef09 | 1447 | void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data, |
e06c8227 JB |
1448 | struct jbd2_buffer_trigger_type *triggers) |
1449 | { | |
1450 | struct buffer_head *bh = jh2bh(jh); | |
1451 | ||
13ceef09 | 1452 | if (!triggers || !triggers->t_frozen) |
e06c8227 JB |
1453 | return; |
1454 | ||
13ceef09 | 1455 | triggers->t_frozen(triggers, bh, mapped_data, bh->b_size); |
e06c8227 JB |
1456 | } |
1457 | ||
1458 | void jbd2_buffer_abort_trigger(struct journal_head *jh, | |
1459 | struct jbd2_buffer_trigger_type *triggers) | |
1460 | { | |
1461 | if (!triggers || !triggers->t_abort) | |
1462 | return; | |
1463 | ||
1464 | triggers->t_abort(triggers, jh2bh(jh)); | |
1465 | } | |
1466 | ||
470decc6 | 1467 | /** |
2bf31d94 | 1468 | * jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata |
470decc6 DK |
1469 | * @handle: transaction to add buffer to. |
1470 | * @bh: buffer to mark | |
1471 | * | |
1472 | * mark dirty metadata which needs to be journaled as part of the current | |
1473 | * transaction. | |
1474 | * | |
9ea7a0df TT |
1475 | * The buffer must have previously had jbd2_journal_get_write_access() |
1476 | * called so that it has a valid journal_head attached to the buffer | |
1477 | * head. | |
1478 | * | |
470decc6 DK |
1479 | * The buffer is placed on the transaction's metadata list and is marked |
1480 | * as belonging to the transaction. | |
1481 | * | |
1482 | * Returns error number or 0 on success. | |
1483 | * | |
1484 | * Special care needs to be taken if the buffer already belongs to the | |
1485 | * current committing transaction (in which case we should have frozen | |
1486 | * data present for that commit). In that case, we don't relink the | |
1487 | * buffer: that only gets done when the old transaction finally | |
1488 | * completes its commit. | |
1489 | */ | |
f7f4bccb | 1490 | int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1491 | { |
1492 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 1493 | journal_t *journal; |
ad56edad | 1494 | struct journal_head *jh; |
9ea7a0df | 1495 | int ret = 0; |
470decc6 | 1496 | |
01215d3e | 1497 | if (!buffer_jbd(bh)) |
1498 | return -EUCLEAN; | |
1499 | ||
6e06ae88 JK |
1500 | /* |
1501 | * We don't grab jh reference here since the buffer must be part | |
1502 | * of the running transaction. | |
1503 | */ | |
1504 | jh = bh2jh(bh); | |
cb3b3bf2 | 1505 | jbd2_debug(5, "journal_head %p\n", jh); |
01215d3e | 1506 | JBUFFER_TRACE(jh, "entry"); |
1507 | ||
6e06ae88 JK |
1508 | /* |
1509 | * This and the following assertions are unreliable since we may see jh | |
1510 | * in inconsistent state unless we grab bh_state lock. But this is | |
1511 | * crucial to catch bugs so let's do a reliable check until the | |
1512 | * lockless handling is fully proven. | |
1513 | */ | |
83fe6b18 JK |
1514 | if (data_race(jh->b_transaction != transaction && |
1515 | jh->b_next_transaction != transaction)) { | |
46417064 | 1516 | spin_lock(&jh->b_state_lock); |
6e06ae88 JK |
1517 | J_ASSERT_JH(jh, jh->b_transaction == transaction || |
1518 | jh->b_next_transaction == transaction); | |
46417064 | 1519 | spin_unlock(&jh->b_state_lock); |
6e06ae88 JK |
1520 | } |
1521 | if (jh->b_modified == 1) { | |
1522 | /* If it's in our transaction it must be in BJ_Metadata list. */ | |
83fe6b18 JK |
1523 | if (data_race(jh->b_transaction == transaction && |
1524 | jh->b_jlist != BJ_Metadata)) { | |
46417064 | 1525 | spin_lock(&jh->b_state_lock); |
e09463f2 TT |
1526 | if (jh->b_transaction == transaction && |
1527 | jh->b_jlist != BJ_Metadata) | |
1528 | pr_err("JBD2: assertion failure: h_type=%u " | |
1529 | "h_line_no=%u block_no=%llu jlist=%u\n", | |
1530 | handle->h_type, handle->h_line_no, | |
1531 | (unsigned long long) bh->b_blocknr, | |
1532 | jh->b_jlist); | |
6e06ae88 JK |
1533 | J_ASSERT_JH(jh, jh->b_transaction != transaction || |
1534 | jh->b_jlist == BJ_Metadata); | |
46417064 | 1535 | spin_unlock(&jh->b_state_lock); |
6e06ae88 JK |
1536 | } |
1537 | goto out; | |
1538 | } | |
1539 | ||
1540 | journal = transaction->t_journal; | |
46417064 | 1541 | spin_lock(&jh->b_state_lock); |
470decc6 | 1542 | |
4a734f08 ZC |
1543 | if (is_handle_aborted(handle)) { |
1544 | /* | |
1545 | * Check journal aborting with @jh->b_state_lock locked, | |
1546 | * since 'jh->b_transaction' could be replaced with | |
1547 | * 'jh->b_next_transaction' during old transaction | |
1548 | * committing if journal aborted, which may fail | |
1549 | * assertion on 'jh->b_frozen_data == NULL'. | |
1550 | */ | |
1551 | ret = -EROFS; | |
1552 | goto out_unlock_bh; | |
1553 | } | |
1554 | ||
470decc6 DK |
1555 | if (jh->b_modified == 0) { |
1556 | /* | |
1557 | * This buffer's got modified and becoming part | |
1558 | * of the transaction. This needs to be done | |
1559 | * once a transaction -bzzz | |
1560 | */ | |
d090707e | 1561 | if (WARN_ON_ONCE(jbd2_handle_buffer_credits(handle) <= 0)) { |
f6c07cad TT |
1562 | ret = -ENOSPC; |
1563 | goto out_unlock_bh; | |
1564 | } | |
e09463f2 | 1565 | jh->b_modified = 1; |
933f1c1e | 1566 | handle->h_total_credits--; |
470decc6 DK |
1567 | } |
1568 | ||
1569 | /* | |
1570 | * fastpath, to avoid expensive locking. If this buffer is already | |
1571 | * on the running transaction's metadata list there is nothing to do. | |
1572 | * Nobody can take it off again because there is a handle open. | |
1573 | * I _think_ we're OK here with SMP barriers - a mistaken decision will | |
1574 | * result in this test being false, so we go in and take the locks. | |
1575 | */ | |
1576 | if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) { | |
1577 | JBUFFER_TRACE(jh, "fastpath"); | |
9ea7a0df TT |
1578 | if (unlikely(jh->b_transaction != |
1579 | journal->j_running_transaction)) { | |
a67c848a | 1580 | printk(KERN_ERR "JBD2: %s: " |
9ea7a0df | 1581 | "jh->b_transaction (%llu, %p, %u) != " |
66a4cb18 | 1582 | "journal->j_running_transaction (%p, %u)\n", |
9ea7a0df TT |
1583 | journal->j_devname, |
1584 | (unsigned long long) bh->b_blocknr, | |
1585 | jh->b_transaction, | |
1586 | jh->b_transaction ? jh->b_transaction->t_tid : 0, | |
1587 | journal->j_running_transaction, | |
1588 | journal->j_running_transaction ? | |
1589 | journal->j_running_transaction->t_tid : 0); | |
1590 | ret = -EINVAL; | |
1591 | } | |
470decc6 DK |
1592 | goto out_unlock_bh; |
1593 | } | |
1594 | ||
1595 | set_buffer_jbddirty(bh); | |
1596 | ||
1597 | /* | |
1598 | * Metadata already on the current transaction list doesn't | |
1599 | * need to be filed. Metadata on another transaction's list must | |
1600 | * be committing, and will be refiled once the commit completes: | |
1601 | * leave it alone for now. | |
1602 | */ | |
1603 | if (jh->b_transaction != transaction) { | |
1604 | JBUFFER_TRACE(jh, "already on other transaction"); | |
66a4cb18 TT |
1605 | if (unlikely(((jh->b_transaction != |
1606 | journal->j_committing_transaction)) || | |
1607 | (jh->b_next_transaction != transaction))) { | |
1608 | printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: " | |
1609 | "bad jh for block %llu: " | |
1610 | "transaction (%p, %u), " | |
1611 | "jh->b_transaction (%p, %u), " | |
1612 | "jh->b_next_transaction (%p, %u), jlist %u\n", | |
9ea7a0df TT |
1613 | journal->j_devname, |
1614 | (unsigned long long) bh->b_blocknr, | |
66a4cb18 | 1615 | transaction, transaction->t_tid, |
9ea7a0df | 1616 | jh->b_transaction, |
66a4cb18 TT |
1617 | jh->b_transaction ? |
1618 | jh->b_transaction->t_tid : 0, | |
9ea7a0df TT |
1619 | jh->b_next_transaction, |
1620 | jh->b_next_transaction ? | |
1621 | jh->b_next_transaction->t_tid : 0, | |
66a4cb18 TT |
1622 | jh->b_jlist); |
1623 | WARN_ON(1); | |
9ea7a0df TT |
1624 | ret = -EINVAL; |
1625 | } | |
470decc6 DK |
1626 | /* And this case is illegal: we can't reuse another |
1627 | * transaction's data buffer, ever. */ | |
1628 | goto out_unlock_bh; | |
1629 | } | |
1630 | ||
1631 | /* That test should have eliminated the following case: */ | |
4019191b | 1632 | J_ASSERT_JH(jh, jh->b_frozen_data == NULL); |
470decc6 DK |
1633 | |
1634 | JBUFFER_TRACE(jh, "file as BJ_Metadata"); | |
1635 | spin_lock(&journal->j_list_lock); | |
41a5b913 | 1636 | __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata); |
470decc6 DK |
1637 | spin_unlock(&journal->j_list_lock); |
1638 | out_unlock_bh: | |
46417064 | 1639 | spin_unlock(&jh->b_state_lock); |
470decc6 DK |
1640 | out: |
1641 | JBUFFER_TRACE(jh, "exit"); | |
9ea7a0df | 1642 | return ret; |
470decc6 DK |
1643 | } |
1644 | ||
470decc6 | 1645 | /** |
2bf31d94 | 1646 | * jbd2_journal_forget() - bforget() for potentially-journaled buffers. |
470decc6 DK |
1647 | * @handle: transaction handle |
1648 | * @bh: bh to 'forget' | |
1649 | * | |
1650 | * We can only do the bforget if there are no commits pending against the | |
1651 | * buffer. If the buffer is dirty in the current running transaction we | |
1652 | * can safely unlink it. | |
1653 | * | |
1654 | * bh may not be a journalled buffer at all - it may be a non-JBD | |
1655 | * buffer which came off the hashtable. Check for this. | |
1656 | * | |
1657 | * Decrements bh->b_count by one. | |
1658 | * | |
1659 | * Allow this call even if the handle has aborted --- it may be part of | |
1660 | * the caller's cleanup after an abort. | |
1661 | */ | |
8d6ce136 | 1662 | int jbd2_journal_forget(handle_t *handle, struct buffer_head *bh) |
470decc6 DK |
1663 | { |
1664 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 1665 | journal_t *journal; |
470decc6 DK |
1666 | struct journal_head *jh; |
1667 | int drop_reserve = 0; | |
1668 | int err = 0; | |
1dfc3220 | 1669 | int was_modified = 0; |
470decc6 | 1670 | |
41a5b913 TT |
1671 | if (is_handle_aborted(handle)) |
1672 | return -EROFS; | |
1673 | journal = transaction->t_journal; | |
1674 | ||
470decc6 DK |
1675 | BUFFER_TRACE(bh, "entry"); |
1676 | ||
46417064 TG |
1677 | jh = jbd2_journal_grab_journal_head(bh); |
1678 | if (!jh) { | |
1679 | __bforget(bh); | |
1680 | return 0; | |
1681 | } | |
470decc6 | 1682 | |
46417064 | 1683 | spin_lock(&jh->b_state_lock); |
470decc6 DK |
1684 | |
1685 | /* Critical error: attempting to delete a bitmap buffer, maybe? | |
1686 | * Don't do any jbd operations, and return an error. */ | |
1687 | if (!J_EXPECT_JH(jh, !jh->b_committed_data, | |
1688 | "inconsistent data on disk")) { | |
1689 | err = -EIO; | |
2e710ff0 | 1690 | goto drop; |
470decc6 DK |
1691 | } |
1692 | ||
48fc7f7e | 1693 | /* keep track of whether or not this transaction modified us */ |
1dfc3220 JB |
1694 | was_modified = jh->b_modified; |
1695 | ||
470decc6 DK |
1696 | /* |
1697 | * The buffer's going from the transaction, we must drop | |
1698 | * all references -bzzz | |
1699 | */ | |
1700 | jh->b_modified = 0; | |
1701 | ||
41a5b913 | 1702 | if (jh->b_transaction == transaction) { |
470decc6 DK |
1703 | J_ASSERT_JH(jh, !jh->b_frozen_data); |
1704 | ||
1705 | /* If we are forgetting a buffer which is already part | |
1706 | * of this transaction, then we can just drop it from | |
1707 | * the transaction immediately. */ | |
1708 | clear_buffer_dirty(bh); | |
1709 | clear_buffer_jbddirty(bh); | |
1710 | ||
1711 | JBUFFER_TRACE(jh, "belongs to current transaction: unfile"); | |
1712 | ||
1dfc3220 JB |
1713 | /* |
1714 | * we only want to drop a reference if this transaction | |
1715 | * modified the buffer | |
1716 | */ | |
1717 | if (was_modified) | |
1718 | drop_reserve = 1; | |
470decc6 DK |
1719 | |
1720 | /* | |
1721 | * We are no longer going to journal this buffer. | |
1722 | * However, the commit of this transaction is still | |
1723 | * important to the buffer: the delete that we are now | |
1724 | * processing might obsolete an old log entry, so by | |
1725 | * committing, we can satisfy the buffer's checkpoint. | |
1726 | * | |
1727 | * So, if we have a checkpoint on the buffer, we should | |
1728 | * now refile the buffer on our BJ_Forget list so that | |
1729 | * we know to remove the checkpoint after we commit. | |
1730 | */ | |
1731 | ||
0bfea811 | 1732 | spin_lock(&journal->j_list_lock); |
470decc6 | 1733 | if (jh->b_cp_transaction) { |
f7f4bccb MC |
1734 | __jbd2_journal_temp_unlink_buffer(jh); |
1735 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); | |
470decc6 | 1736 | } else { |
f7f4bccb | 1737 | __jbd2_journal_unfile_buffer(jh); |
93108ebb | 1738 | jbd2_journal_put_journal_head(jh); |
470decc6 | 1739 | } |
0bfea811 | 1740 | spin_unlock(&journal->j_list_lock); |
470decc6 DK |
1741 | } else if (jh->b_transaction) { |
1742 | J_ASSERT_JH(jh, (jh->b_transaction == | |
1743 | journal->j_committing_transaction)); | |
1744 | /* However, if the buffer is still owned by a prior | |
1745 | * (committing) transaction, we can't drop it yet... */ | |
1746 | JBUFFER_TRACE(jh, "belongs to older transaction"); | |
904cdbd4 | 1747 | /* ... but we CAN drop it from the new transaction through |
1748 | * marking the buffer as freed and set j_next_transaction to | |
1749 | * the new transaction, so that not only the commit code | |
1750 | * knows it should clear dirty bits when it is done with the | |
1751 | * buffer, but also the buffer can be checkpointed only | |
1752 | * after the new transaction commits. */ | |
470decc6 | 1753 | |
904cdbd4 | 1754 | set_buffer_freed(bh); |
1755 | ||
1756 | if (!jh->b_next_transaction) { | |
0bfea811 | 1757 | spin_lock(&journal->j_list_lock); |
904cdbd4 | 1758 | jh->b_next_transaction = transaction; |
0bfea811 | 1759 | spin_unlock(&journal->j_list_lock); |
904cdbd4 | 1760 | } else { |
1761 | J_ASSERT(jh->b_next_transaction == transaction); | |
1dfc3220 JB |
1762 | |
1763 | /* | |
1764 | * only drop a reference if this transaction modified | |
1765 | * the buffer | |
1766 | */ | |
1767 | if (was_modified) | |
1768 | drop_reserve = 1; | |
470decc6 | 1769 | } |
59759926 | 1770 | } else { |
1771 | /* | |
1772 | * Finally, if the buffer is not belongs to any | |
1773 | * transaction, we can just drop it now if it has no | |
1774 | * checkpoint. | |
1775 | */ | |
1776 | spin_lock(&journal->j_list_lock); | |
1777 | if (!jh->b_cp_transaction) { | |
1778 | JBUFFER_TRACE(jh, "belongs to none transaction"); | |
1779 | spin_unlock(&journal->j_list_lock); | |
2e710ff0 | 1780 | goto drop; |
59759926 | 1781 | } |
1782 | ||
1783 | /* | |
1784 | * Otherwise, if the buffer has been written to disk, | |
1785 | * it is safe to remove the checkpoint and drop it. | |
1786 | */ | |
1787 | if (!buffer_dirty(bh)) { | |
1788 | __jbd2_journal_remove_checkpoint(jh); | |
1789 | spin_unlock(&journal->j_list_lock); | |
2e710ff0 | 1790 | goto drop; |
59759926 | 1791 | } |
1792 | ||
1793 | /* | |
1794 | * The buffer is still not written to disk, we should | |
1795 | * attach this buffer to current transaction so that the | |
1796 | * buffer can be checkpointed only after the current | |
1797 | * transaction commits. | |
1798 | */ | |
1799 | clear_buffer_dirty(bh); | |
1800 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); | |
1801 | spin_unlock(&journal->j_list_lock); | |
470decc6 | 1802 | } |
470decc6 | 1803 | drop: |
470decc6 | 1804 | __brelse(bh); |
46417064 TG |
1805 | spin_unlock(&jh->b_state_lock); |
1806 | jbd2_journal_put_journal_head(jh); | |
470decc6 DK |
1807 | if (drop_reserve) { |
1808 | /* no need to reserve log space for this block -bzzz */ | |
933f1c1e | 1809 | handle->h_total_credits++; |
470decc6 DK |
1810 | } |
1811 | return err; | |
1812 | } | |
1813 | ||
1814 | /** | |
2bf31d94 | 1815 | * jbd2_journal_stop() - complete a transaction |
bd7ced98 | 1816 | * @handle: transaction to complete. |
470decc6 DK |
1817 | * |
1818 | * All done for a particular handle. | |
1819 | * | |
1820 | * There is not much action needed here. We just return any remaining | |
1821 | * buffer credits to the transaction and remove the handle. The only | |
1822 | * complication is that we need to start a commit operation if the | |
1823 | * filesystem is marked for synchronous update. | |
1824 | * | |
f7f4bccb | 1825 | * jbd2_journal_stop itself will not usually return an error, but it may |
470decc6 | 1826 | * do so in unusual circumstances. In particular, expect it to |
f7f4bccb | 1827 | * return -EIO if a jbd2_journal_abort has been executed since the |
470decc6 DK |
1828 | * transaction began. |
1829 | */ | |
f7f4bccb | 1830 | int jbd2_journal_stop(handle_t *handle) |
470decc6 DK |
1831 | { |
1832 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 TT |
1833 | journal_t *journal; |
1834 | int err = 0, wait_for_commit = 0; | |
a51dca9c | 1835 | tid_t tid; |
470decc6 DK |
1836 | pid_t pid; |
1837 | ||
dfaf5ffd | 1838 | if (--handle->h_ref > 0) { |
cb3b3bf2 | 1839 | jbd2_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1, |
dfaf5ffd JK |
1840 | handle->h_ref); |
1841 | if (is_handle_aborted(handle)) | |
1842 | return -EIO; | |
1843 | return 0; | |
1844 | } | |
9d506594 LC |
1845 | if (!transaction) { |
1846 | /* | |
dfaf5ffd JK |
1847 | * Handle is already detached from the transaction so there is |
1848 | * nothing to do other than free the handle. | |
9d506594 | 1849 | */ |
ec8b6f60 | 1850 | memalloc_nofs_restore(handle->saved_alloc_context); |
dfaf5ffd | 1851 | goto free_and_exit; |
9d506594 | 1852 | } |
41a5b913 | 1853 | journal = transaction->t_journal; |
dfaf5ffd | 1854 | tid = transaction->t_tid; |
470decc6 DK |
1855 | |
1856 | if (is_handle_aborted(handle)) | |
1857 | err = -EIO; | |
470decc6 | 1858 | |
cb3b3bf2 | 1859 | jbd2_debug(4, "Handle %p going down\n", handle); |
343d9c28 | 1860 | trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev, |
dfaf5ffd | 1861 | tid, handle->h_type, handle->h_line_no, |
343d9c28 TT |
1862 | jiffies - handle->h_start_jiffies, |
1863 | handle->h_sync, handle->h_requested_credits, | |
1864 | (handle->h_requested_credits - | |
933f1c1e | 1865 | handle->h_total_credits)); |
470decc6 DK |
1866 | |
1867 | /* | |
1868 | * Implement synchronous transaction batching. If the handle | |
1869 | * was synchronous, don't force a commit immediately. Let's | |
e07f7183 JB |
1870 | * yield and let another thread piggyback onto this |
1871 | * transaction. Keep doing that while new threads continue to | |
1872 | * arrive. It doesn't cost much - we're about to run a commit | |
1873 | * and sleep on IO anyway. Speeds up many-threaded, many-dir | |
1874 | * operations by 30x or more... | |
1875 | * | |
1876 | * We try and optimize the sleep time against what the | |
1877 | * underlying disk can do, instead of having a static sleep | |
1878 | * time. This is useful for the case where our storage is so | |
1879 | * fast that it is more optimal to go ahead and force a flush | |
1880 | * and wait for the transaction to be committed than it is to | |
1881 | * wait for an arbitrary amount of time for new writers to | |
1882 | * join the transaction. We achieve this by measuring how | |
1883 | * long it takes to commit a transaction, and compare it with | |
1884 | * how long this transaction has been running, and if run time | |
1885 | * < commit time then we sleep for the delta and commit. This | |
1886 | * greatly helps super fast disks that would see slowdowns as | |
1887 | * more threads started doing fsyncs. | |
470decc6 | 1888 | * |
e07f7183 JB |
1889 | * But don't do this if this process was the most recent one |
1890 | * to perform a synchronous write. We do this to detect the | |
1891 | * case where a single process is doing a stream of sync | |
1892 | * writes. No point in waiting for joiners in that case. | |
5dd21424 ES |
1893 | * |
1894 | * Setting max_batch_time to 0 disables this completely. | |
470decc6 DK |
1895 | */ |
1896 | pid = current->pid; | |
5dd21424 ES |
1897 | if (handle->h_sync && journal->j_last_sync_writer != pid && |
1898 | journal->j_max_batch_time) { | |
e07f7183 JB |
1899 | u64 commit_time, trans_time; |
1900 | ||
470decc6 | 1901 | journal->j_last_sync_writer = pid; |
e07f7183 | 1902 | |
a931da6a | 1903 | read_lock(&journal->j_state_lock); |
e07f7183 | 1904 | commit_time = journal->j_average_commit_time; |
a931da6a | 1905 | read_unlock(&journal->j_state_lock); |
e07f7183 JB |
1906 | |
1907 | trans_time = ktime_to_ns(ktime_sub(ktime_get(), | |
1908 | transaction->t_start_time)); | |
1909 | ||
30773840 TT |
1910 | commit_time = max_t(u64, commit_time, |
1911 | 1000*journal->j_min_batch_time); | |
e07f7183 | 1912 | commit_time = min_t(u64, commit_time, |
30773840 | 1913 | 1000*journal->j_max_batch_time); |
e07f7183 JB |
1914 | |
1915 | if (trans_time < commit_time) { | |
1916 | ktime_t expires = ktime_add_ns(ktime_get(), | |
1917 | commit_time); | |
1918 | set_current_state(TASK_UNINTERRUPTIBLE); | |
1919 | schedule_hrtimeout(&expires, HRTIMER_MODE_ABS); | |
1920 | } | |
470decc6 DK |
1921 | } |
1922 | ||
7058548c TT |
1923 | if (handle->h_sync) |
1924 | transaction->t_synchronous_commit = 1; | |
470decc6 DK |
1925 | |
1926 | /* | |
1927 | * If the handle is marked SYNC, we need to set another commit | |
150549ed JK |
1928 | * going! We also want to force a commit if the transaction is too |
1929 | * old now. | |
470decc6 DK |
1930 | */ |
1931 | if (handle->h_sync || | |
a51dca9c | 1932 | time_after_eq(jiffies, transaction->t_expires)) { |
470decc6 DK |
1933 | /* Do this even for aborted journals: an abort still |
1934 | * completes the commit thread, it just doesn't write | |
1935 | * anything to disk. */ | |
470decc6 | 1936 | |
cb3b3bf2 | 1937 | jbd2_debug(2, "transaction too old, requesting commit for " |
470decc6 DK |
1938 | "handle %p\n", handle); |
1939 | /* This is non-blocking */ | |
dfaf5ffd | 1940 | jbd2_log_start_commit(journal, tid); |
470decc6 DK |
1941 | |
1942 | /* | |
f7f4bccb | 1943 | * Special case: JBD2_SYNC synchronous updates require us |
470decc6 DK |
1944 | * to wait for the commit to complete. |
1945 | */ | |
1946 | if (handle->h_sync && !(current->flags & PF_MEMALLOC)) | |
a51dca9c | 1947 | wait_for_commit = 1; |
470decc6 DK |
1948 | } |
1949 | ||
a51dca9c | 1950 | /* |
ec8b6f60 JK |
1951 | * Once stop_this_handle() drops t_updates, the transaction could start |
1952 | * committing on us and eventually disappear. So we must not | |
1953 | * dereference transaction pointer again after calling | |
1954 | * stop_this_handle(). | |
a51dca9c | 1955 | */ |
ec8b6f60 | 1956 | stop_this_handle(handle); |
7a4b188f | 1957 | |
a51dca9c TT |
1958 | if (wait_for_commit) |
1959 | err = jbd2_log_wait_commit(journal, tid); | |
1960 | ||
41a5b913 | 1961 | free_and_exit: |
ec8b6f60 JK |
1962 | if (handle->h_rsv_handle) |
1963 | jbd2_free_handle(handle->h_rsv_handle); | |
af1e76d6 | 1964 | jbd2_free_handle(handle); |
470decc6 DK |
1965 | return err; |
1966 | } | |
1967 | ||
470decc6 DK |
1968 | /* |
1969 | * | |
1970 | * List management code snippets: various functions for manipulating the | |
1971 | * transaction buffer lists. | |
1972 | * | |
1973 | */ | |
1974 | ||
1975 | /* | |
1976 | * Append a buffer to a transaction list, given the transaction's list head | |
1977 | * pointer. | |
1978 | * | |
1979 | * j_list_lock is held. | |
1980 | * | |
46417064 | 1981 | * jh->b_state_lock is held. |
470decc6 DK |
1982 | */ |
1983 | ||
1984 | static inline void | |
1985 | __blist_add_buffer(struct journal_head **list, struct journal_head *jh) | |
1986 | { | |
1987 | if (!*list) { | |
1988 | jh->b_tnext = jh->b_tprev = jh; | |
1989 | *list = jh; | |
1990 | } else { | |
1991 | /* Insert at the tail of the list to preserve order */ | |
1992 | struct journal_head *first = *list, *last = first->b_tprev; | |
1993 | jh->b_tprev = last; | |
1994 | jh->b_tnext = first; | |
1995 | last->b_tnext = first->b_tprev = jh; | |
1996 | } | |
1997 | } | |
1998 | ||
1999 | /* | |
2000 | * Remove a buffer from a transaction list, given the transaction's list | |
2001 | * head pointer. | |
2002 | * | |
2003 | * Called with j_list_lock held, and the journal may not be locked. | |
2004 | * | |
46417064 | 2005 | * jh->b_state_lock is held. |
470decc6 DK |
2006 | */ |
2007 | ||
2008 | static inline void | |
2009 | __blist_del_buffer(struct journal_head **list, struct journal_head *jh) | |
2010 | { | |
2011 | if (*list == jh) { | |
2012 | *list = jh->b_tnext; | |
2013 | if (*list == jh) | |
2014 | *list = NULL; | |
2015 | } | |
2016 | jh->b_tprev->b_tnext = jh->b_tnext; | |
2017 | jh->b_tnext->b_tprev = jh->b_tprev; | |
2018 | } | |
2019 | ||
2020 | /* | |
2021 | * Remove a buffer from the appropriate transaction list. | |
2022 | * | |
2023 | * Note that this function can *change* the value of | |
f5113eff JK |
2024 | * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or |
2025 | * t_reserved_list. If the caller is holding onto a copy of one of these | |
2026 | * pointers, it could go bad. Generally the caller needs to re-read the | |
2027 | * pointer from the transaction_t. | |
470decc6 | 2028 | * |
5bebccf9 | 2029 | * Called under j_list_lock. |
470decc6 | 2030 | */ |
5bebccf9 | 2031 | static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh) |
470decc6 DK |
2032 | { |
2033 | struct journal_head **list = NULL; | |
2034 | transaction_t *transaction; | |
2035 | struct buffer_head *bh = jh2bh(jh); | |
2036 | ||
46417064 | 2037 | lockdep_assert_held(&jh->b_state_lock); |
470decc6 DK |
2038 | transaction = jh->b_transaction; |
2039 | if (transaction) | |
2040 | assert_spin_locked(&transaction->t_journal->j_list_lock); | |
2041 | ||
2042 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
2043 | if (jh->b_jlist != BJ_None) | |
4019191b | 2044 | J_ASSERT_JH(jh, transaction != NULL); |
470decc6 DK |
2045 | |
2046 | switch (jh->b_jlist) { | |
2047 | case BJ_None: | |
2048 | return; | |
470decc6 DK |
2049 | case BJ_Metadata: |
2050 | transaction->t_nr_buffers--; | |
2051 | J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0); | |
2052 | list = &transaction->t_buffers; | |
2053 | break; | |
2054 | case BJ_Forget: | |
2055 | list = &transaction->t_forget; | |
2056 | break; | |
470decc6 DK |
2057 | case BJ_Shadow: |
2058 | list = &transaction->t_shadow_list; | |
2059 | break; | |
470decc6 DK |
2060 | case BJ_Reserved: |
2061 | list = &transaction->t_reserved_list; | |
2062 | break; | |
470decc6 DK |
2063 | } |
2064 | ||
2065 | __blist_del_buffer(list, jh); | |
2066 | jh->b_jlist = BJ_None; | |
e112666b TT |
2067 | if (transaction && is_journal_aborted(transaction->t_journal)) |
2068 | clear_buffer_jbddirty(bh); | |
2069 | else if (test_clear_buffer_jbddirty(bh)) | |
470decc6 DK |
2070 | mark_buffer_dirty(bh); /* Expose it to the VM */ |
2071 | } | |
2072 | ||
de1b7941 | 2073 | /* |
93108ebb JK |
2074 | * Remove buffer from all transactions. The caller is responsible for dropping |
2075 | * the jh reference that belonged to the transaction. | |
de1b7941 JK |
2076 | * |
2077 | * Called with bh_state lock and j_list_lock | |
de1b7941 JK |
2078 | */ |
2079 | static void __jbd2_journal_unfile_buffer(struct journal_head *jh) | |
470decc6 | 2080 | { |
24dc9864 LC |
2081 | J_ASSERT_JH(jh, jh->b_transaction != NULL); |
2082 | J_ASSERT_JH(jh, jh->b_next_transaction == NULL); | |
2083 | ||
f7f4bccb | 2084 | __jbd2_journal_temp_unlink_buffer(jh); |
470decc6 DK |
2085 | jh->b_transaction = NULL; |
2086 | } | |
2087 | ||
f7f4bccb | 2088 | void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 | 2089 | { |
de1b7941 JK |
2090 | struct buffer_head *bh = jh2bh(jh); |
2091 | ||
2092 | /* Get reference so that buffer cannot be freed before we unlock it */ | |
2093 | get_bh(bh); | |
46417064 | 2094 | spin_lock(&jh->b_state_lock); |
470decc6 | 2095 | spin_lock(&journal->j_list_lock); |
f7f4bccb | 2096 | __jbd2_journal_unfile_buffer(jh); |
470decc6 | 2097 | spin_unlock(&journal->j_list_lock); |
46417064 | 2098 | spin_unlock(&jh->b_state_lock); |
93108ebb | 2099 | jbd2_journal_put_journal_head(jh); |
de1b7941 | 2100 | __brelse(bh); |
470decc6 DK |
2101 | } |
2102 | ||
2103 | /* | |
f7f4bccb | 2104 | * Called from jbd2_journal_try_to_free_buffers(). |
470decc6 | 2105 | * |
46417064 | 2106 | * Called under jh->b_state_lock |
470decc6 DK |
2107 | */ |
2108 | static void | |
2109 | __journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh) | |
2110 | { | |
2111 | struct journal_head *jh; | |
2112 | ||
2113 | jh = bh2jh(bh); | |
2114 | ||
2115 | if (buffer_locked(bh) || buffer_dirty(bh)) | |
2116 | goto out; | |
2117 | ||
d2eb0b99 | 2118 | if (jh->b_next_transaction != NULL || jh->b_transaction != NULL) |
470decc6 DK |
2119 | goto out; |
2120 | ||
2121 | spin_lock(&journal->j_list_lock); | |
d2eb0b99 | 2122 | if (jh->b_cp_transaction != NULL) { |
470decc6 | 2123 | /* written-back checkpointed metadata buffer */ |
c254c9ec JK |
2124 | JBUFFER_TRACE(jh, "remove from checkpoint list"); |
2125 | __jbd2_journal_remove_checkpoint(jh); | |
470decc6 DK |
2126 | } |
2127 | spin_unlock(&journal->j_list_lock); | |
2128 | out: | |
2129 | return; | |
2130 | } | |
2131 | ||
470decc6 | 2132 | /** |
2bf31d94 | 2133 | * jbd2_journal_try_to_free_buffers() - try to free page buffers. |
470decc6 | 2134 | * @journal: journal for operation |
4f5bf127 | 2135 | * @folio: Folio to detach data from. |
470decc6 DK |
2136 | * |
2137 | * For all the buffers on this page, | |
2138 | * if they are fully written out ordered data, move them onto BUF_CLEAN | |
2139 | * so try_to_free_buffers() can reap them. | |
2140 | * | |
2141 | * This function returns non-zero if we wish try_to_free_buffers() | |
2142 | * to be called. We do this if the page is releasable by try_to_free_buffers(). | |
2143 | * We also do it if the page has locked or dirty buffers and the caller wants | |
2144 | * us to perform sync or async writeout. | |
2145 | * | |
2146 | * This complicates JBD locking somewhat. We aren't protected by the | |
2147 | * BKL here. We wish to remove the buffer from its committing or | |
f7f4bccb | 2148 | * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer. |
470decc6 DK |
2149 | * |
2150 | * This may *change* the value of transaction_t->t_datalist, so anyone | |
2151 | * who looks at t_datalist needs to lock against this function. | |
2152 | * | |
f7f4bccb MC |
2153 | * Even worse, someone may be doing a jbd2_journal_dirty_data on this |
2154 | * buffer. So we need to lock against that. jbd2_journal_dirty_data() | |
470decc6 DK |
2155 | * will come out of the lock with the buffer dirty, which makes it |
2156 | * ineligible for release here. | |
2157 | * | |
2158 | * Who else is affected by this? hmm... Really the only contender | |
2159 | * is do_get_write_access() - it could be looking at the buffer while | |
2160 | * journal_try_to_free_buffer() is changing its state. But that | |
2161 | * cannot happen because we never reallocate freed data as metadata | |
2162 | * while the data is part of a transaction. Yes? | |
530576bb | 2163 | * |
c56a6eb0 | 2164 | * Return false on failure, true on success |
470decc6 | 2165 | */ |
c56a6eb0 | 2166 | bool jbd2_journal_try_to_free_buffers(journal_t *journal, struct folio *folio) |
470decc6 DK |
2167 | { |
2168 | struct buffer_head *head; | |
2169 | struct buffer_head *bh; | |
c56a6eb0 | 2170 | bool ret = false; |
470decc6 | 2171 | |
c56a6eb0 | 2172 | J_ASSERT(folio_test_locked(folio)); |
470decc6 | 2173 | |
c56a6eb0 | 2174 | head = folio_buffers(folio); |
470decc6 DK |
2175 | bh = head; |
2176 | do { | |
2177 | struct journal_head *jh; | |
2178 | ||
2179 | /* | |
2180 | * We take our own ref against the journal_head here to avoid | |
2181 | * having to add tons of locking around each instance of | |
530576bb | 2182 | * jbd2_journal_put_journal_head(). |
470decc6 | 2183 | */ |
f7f4bccb | 2184 | jh = jbd2_journal_grab_journal_head(bh); |
470decc6 DK |
2185 | if (!jh) |
2186 | continue; | |
2187 | ||
46417064 | 2188 | spin_lock(&jh->b_state_lock); |
470decc6 | 2189 | __journal_try_to_free_buffer(journal, bh); |
46417064 | 2190 | spin_unlock(&jh->b_state_lock); |
f7f4bccb | 2191 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2192 | if (buffer_jbd(bh)) |
2193 | goto busy; | |
2194 | } while ((bh = bh->b_this_page) != head); | |
530576bb | 2195 | |
68189fef | 2196 | ret = try_to_free_buffers(folio); |
470decc6 DK |
2197 | busy: |
2198 | return ret; | |
2199 | } | |
2200 | ||
2201 | /* | |
2202 | * This buffer is no longer needed. If it is on an older transaction's | |
2203 | * checkpoint list we need to record it on this transaction's forget list | |
2204 | * to pin this buffer (and hence its checkpointing transaction) down until | |
2205 | * this transaction commits. If the buffer isn't on a checkpoint list, we | |
2206 | * release it. | |
2207 | * Returns non-zero if JBD no longer has an interest in the buffer. | |
2208 | * | |
2209 | * Called under j_list_lock. | |
2210 | * | |
46417064 | 2211 | * Called under jh->b_state_lock. |
470decc6 DK |
2212 | */ |
2213 | static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction) | |
2214 | { | |
2215 | int may_free = 1; | |
2216 | struct buffer_head *bh = jh2bh(jh); | |
2217 | ||
470decc6 DK |
2218 | if (jh->b_cp_transaction) { |
2219 | JBUFFER_TRACE(jh, "on running+cp transaction"); | |
de1b7941 | 2220 | __jbd2_journal_temp_unlink_buffer(jh); |
f91d1d04 JK |
2221 | /* |
2222 | * We don't want to write the buffer anymore, clear the | |
2223 | * bit so that we don't confuse checks in | |
2224 | * __journal_file_buffer | |
2225 | */ | |
2226 | clear_buffer_dirty(bh); | |
f7f4bccb | 2227 | __jbd2_journal_file_buffer(jh, transaction, BJ_Forget); |
470decc6 DK |
2228 | may_free = 0; |
2229 | } else { | |
2230 | JBUFFER_TRACE(jh, "on running transaction"); | |
de1b7941 | 2231 | __jbd2_journal_unfile_buffer(jh); |
93108ebb | 2232 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2233 | } |
2234 | return may_free; | |
2235 | } | |
2236 | ||
2237 | /* | |
ccd16945 | 2238 | * jbd2_journal_invalidate_folio |
470decc6 DK |
2239 | * |
2240 | * This code is tricky. It has a number of cases to deal with. | |
2241 | * | |
2242 | * There are two invariants which this code relies on: | |
2243 | * | |
ccd16945 MWO |
2244 | * i_size must be updated on disk before we start calling invalidate_folio |
2245 | * on the data. | |
470decc6 DK |
2246 | * |
2247 | * This is done in ext3 by defining an ext3_setattr method which | |
2248 | * updates i_size before truncate gets going. By maintaining this | |
2249 | * invariant, we can be sure that it is safe to throw away any buffers | |
2250 | * attached to the current transaction: once the transaction commits, | |
2251 | * we know that the data will not be needed. | |
2252 | * | |
2253 | * Note however that we can *not* throw away data belonging to the | |
2254 | * previous, committing transaction! | |
2255 | * | |
2256 | * Any disk blocks which *are* part of the previous, committing | |
2257 | * transaction (and which therefore cannot be discarded immediately) are | |
2258 | * not going to be reused in the new running transaction | |
2259 | * | |
2260 | * The bitmap committed_data images guarantee this: any block which is | |
2261 | * allocated in one transaction and removed in the next will be marked | |
2262 | * as in-use in the committed_data bitmap, so cannot be reused until | |
2263 | * the next transaction to delete the block commits. This means that | |
2264 | * leaving committing buffers dirty is quite safe: the disk blocks | |
2265 | * cannot be reallocated to a different file and so buffer aliasing is | |
2266 | * not possible. | |
2267 | * | |
2268 | * | |
2269 | * The above applies mainly to ordered data mode. In writeback mode we | |
2270 | * don't make guarantees about the order in which data hits disk --- in | |
2271 | * particular we don't guarantee that new dirty data is flushed before | |
2272 | * transaction commit --- so it is always safe just to discard data | |
2273 | * immediately in that mode. --sct | |
2274 | */ | |
2275 | ||
2276 | /* | |
2277 | * The journal_unmap_buffer helper function returns zero if the buffer | |
2278 | * concerned remains pinned as an anonymous buffer belonging to an older | |
2279 | * transaction. | |
2280 | * | |
2281 | * We're outside-transaction here. Either or both of j_running_transaction | |
2282 | * and j_committing_transaction may be NULL. | |
2283 | */ | |
b794e7a6 JK |
2284 | static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh, |
2285 | int partial_page) | |
470decc6 DK |
2286 | { |
2287 | transaction_t *transaction; | |
2288 | struct journal_head *jh; | |
2289 | int may_free = 1; | |
470decc6 DK |
2290 | |
2291 | BUFFER_TRACE(bh, "entry"); | |
2292 | ||
2293 | /* | |
2294 | * It is safe to proceed here without the j_list_lock because the | |
2295 | * buffers cannot be stolen by try_to_free_buffers as long as we are | |
2296 | * holding the page lock. --sct | |
2297 | */ | |
2298 | ||
d84560f7 TG |
2299 | jh = jbd2_journal_grab_journal_head(bh); |
2300 | if (!jh) | |
470decc6 DK |
2301 | goto zap_buffer_unlocked; |
2302 | ||
87c89c23 | 2303 | /* OK, we have data buffer in journaled mode */ |
a931da6a | 2304 | write_lock(&journal->j_state_lock); |
46417064 | 2305 | spin_lock(&jh->b_state_lock); |
470decc6 DK |
2306 | spin_lock(&journal->j_list_lock); |
2307 | ||
ba869023 | 2308 | /* |
2309 | * We cannot remove the buffer from checkpoint lists until the | |
2310 | * transaction adding inode to orphan list (let's call it T) | |
2311 | * is committed. Otherwise if the transaction changing the | |
2312 | * buffer would be cleaned from the journal before T is | |
2313 | * committed, a crash will cause that the correct contents of | |
2314 | * the buffer will be lost. On the other hand we have to | |
2315 | * clear the buffer dirty bit at latest at the moment when the | |
2316 | * transaction marking the buffer as freed in the filesystem | |
2317 | * structures is committed because from that moment on the | |
b794e7a6 | 2318 | * block can be reallocated and used by a different page. |
ba869023 | 2319 | * Since the block hasn't been freed yet but the inode has |
2320 | * already been added to orphan list, it is safe for us to add | |
2321 | * the buffer to BJ_Forget list of the newest transaction. | |
b794e7a6 JK |
2322 | * |
2323 | * Also we have to clear buffer_mapped flag of a truncated buffer | |
2324 | * because the buffer_head may be attached to the page straddling | |
2325 | * i_size (can happen only when blocksize < pagesize) and thus the | |
2326 | * buffer_head can be reused when the file is extended again. So we end | |
2327 | * up keeping around invalidated buffers attached to transactions' | |
2328 | * BJ_Forget list just to stop checkpointing code from cleaning up | |
2329 | * the transaction this buffer was modified in. | |
ba869023 | 2330 | */ |
470decc6 DK |
2331 | transaction = jh->b_transaction; |
2332 | if (transaction == NULL) { | |
2333 | /* First case: not on any transaction. If it | |
2334 | * has no checkpoint link, then we can zap it: | |
2335 | * it's a writeback-mode buffer so we don't care | |
2336 | * if it hits disk safely. */ | |
2337 | if (!jh->b_cp_transaction) { | |
2338 | JBUFFER_TRACE(jh, "not on any transaction: zap"); | |
2339 | goto zap_buffer; | |
2340 | } | |
2341 | ||
2342 | if (!buffer_dirty(bh)) { | |
2343 | /* bdflush has written it. We can drop it now */ | |
bc23f0c8 | 2344 | __jbd2_journal_remove_checkpoint(jh); |
470decc6 DK |
2345 | goto zap_buffer; |
2346 | } | |
2347 | ||
2348 | /* OK, it must be in the journal but still not | |
2349 | * written fully to disk: it's metadata or | |
2350 | * journaled data... */ | |
2351 | ||
2352 | if (journal->j_running_transaction) { | |
2353 | /* ... and once the current transaction has | |
2354 | * committed, the buffer won't be needed any | |
2355 | * longer. */ | |
2356 | JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget"); | |
b794e7a6 | 2357 | may_free = __dispose_buffer(jh, |
470decc6 | 2358 | journal->j_running_transaction); |
b794e7a6 | 2359 | goto zap_buffer; |
470decc6 DK |
2360 | } else { |
2361 | /* There is no currently-running transaction. So the | |
2362 | * orphan record which we wrote for this file must have | |
2363 | * passed into commit. We must attach this buffer to | |
2364 | * the committing transaction, if it exists. */ | |
2365 | if (journal->j_committing_transaction) { | |
2366 | JBUFFER_TRACE(jh, "give to committing trans"); | |
b794e7a6 | 2367 | may_free = __dispose_buffer(jh, |
470decc6 | 2368 | journal->j_committing_transaction); |
b794e7a6 | 2369 | goto zap_buffer; |
470decc6 DK |
2370 | } else { |
2371 | /* The orphan record's transaction has | |
2372 | * committed. We can cleanse this buffer */ | |
2373 | clear_buffer_jbddirty(bh); | |
bc23f0c8 | 2374 | __jbd2_journal_remove_checkpoint(jh); |
470decc6 DK |
2375 | goto zap_buffer; |
2376 | } | |
2377 | } | |
2378 | } else if (transaction == journal->j_committing_transaction) { | |
9b57988d | 2379 | JBUFFER_TRACE(jh, "on committing transaction"); |
470decc6 | 2380 | /* |
ba869023 | 2381 | * The buffer is committing, we simply cannot touch |
b794e7a6 JK |
2382 | * it. If the page is straddling i_size we have to wait |
2383 | * for commit and try again. | |
2384 | */ | |
2385 | if (partial_page) { | |
b794e7a6 | 2386 | spin_unlock(&journal->j_list_lock); |
46417064 | 2387 | spin_unlock(&jh->b_state_lock); |
b794e7a6 | 2388 | write_unlock(&journal->j_state_lock); |
46417064 | 2389 | jbd2_journal_put_journal_head(jh); |
bd159398 JK |
2390 | /* Already zapped buffer? Nothing to do... */ |
2391 | if (!bh->b_bdev) | |
2392 | return 0; | |
53e87268 | 2393 | return -EBUSY; |
b794e7a6 JK |
2394 | } |
2395 | /* | |
6a66a7de | 2396 | * OK, buffer won't be reachable after truncate. We just clear |
2397 | * b_modified to not confuse transaction credit accounting, and | |
2398 | * set j_next_transaction to the running transaction (if there | |
2399 | * is one) and mark buffer as freed so that commit code knows | |
2400 | * it should clear dirty bits when it is done with the buffer. | |
ba869023 | 2401 | */ |
470decc6 | 2402 | set_buffer_freed(bh); |
ba869023 | 2403 | if (journal->j_running_transaction && buffer_jbddirty(bh)) |
2404 | jh->b_next_transaction = journal->j_running_transaction; | |
6a66a7de | 2405 | jh->b_modified = 0; |
470decc6 | 2406 | spin_unlock(&journal->j_list_lock); |
46417064 | 2407 | spin_unlock(&jh->b_state_lock); |
a931da6a | 2408 | write_unlock(&journal->j_state_lock); |
46417064 | 2409 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2410 | return 0; |
2411 | } else { | |
2412 | /* Good, the buffer belongs to the running transaction. | |
2413 | * We are writing our own transaction's data, not any | |
2414 | * previous one's, so it is safe to throw it away | |
2415 | * (remember that we expect the filesystem to have set | |
2416 | * i_size already for this truncate so recovery will not | |
2417 | * expose the disk blocks we are discarding here.) */ | |
2418 | J_ASSERT_JH(jh, transaction == journal->j_running_transaction); | |
9b57988d | 2419 | JBUFFER_TRACE(jh, "on running transaction"); |
470decc6 DK |
2420 | may_free = __dispose_buffer(jh, transaction); |
2421 | } | |
2422 | ||
2423 | zap_buffer: | |
b794e7a6 JK |
2424 | /* |
2425 | * This is tricky. Although the buffer is truncated, it may be reused | |
2426 | * if blocksize < pagesize and it is attached to the page straddling | |
2427 | * EOF. Since the buffer might have been added to BJ_Forget list of the | |
2428 | * running transaction, journal_get_write_access() won't clear | |
2429 | * b_modified and credit accounting gets confused. So clear b_modified | |
2430 | * here. | |
2431 | */ | |
2432 | jh->b_modified = 0; | |
470decc6 | 2433 | spin_unlock(&journal->j_list_lock); |
46417064 | 2434 | spin_unlock(&jh->b_state_lock); |
a931da6a | 2435 | write_unlock(&journal->j_state_lock); |
46417064 | 2436 | jbd2_journal_put_journal_head(jh); |
470decc6 DK |
2437 | zap_buffer_unlocked: |
2438 | clear_buffer_dirty(bh); | |
2439 | J_ASSERT_BH(bh, !buffer_jbddirty(bh)); | |
2440 | clear_buffer_mapped(bh); | |
2441 | clear_buffer_req(bh); | |
2442 | clear_buffer_new(bh); | |
15291164 ES |
2443 | clear_buffer_delay(bh); |
2444 | clear_buffer_unwritten(bh); | |
470decc6 DK |
2445 | bh->b_bdev = NULL; |
2446 | return may_free; | |
2447 | } | |
2448 | ||
2449 | /** | |
ccd16945 | 2450 | * jbd2_journal_invalidate_folio() |
470decc6 | 2451 | * @journal: journal to use for flush... |
ccd16945 | 2452 | * @folio: folio to flush |
259709b0 LC |
2453 | * @offset: start of the range to invalidate |
2454 | * @length: length of the range to invalidate | |
470decc6 | 2455 | * |
259709b0 LC |
2456 | * Reap page buffers containing data after in the specified range in page. |
2457 | * Can return -EBUSY if buffers are part of the committing transaction and | |
2458 | * the page is straddling i_size. Caller then has to wait for current commit | |
2459 | * and try again. | |
470decc6 | 2460 | */ |
ccd16945 MWO |
2461 | int jbd2_journal_invalidate_folio(journal_t *journal, struct folio *folio, |
2462 | size_t offset, size_t length) | |
470decc6 DK |
2463 | { |
2464 | struct buffer_head *head, *bh, *next; | |
259709b0 | 2465 | unsigned int stop = offset + length; |
470decc6 | 2466 | unsigned int curr_off = 0; |
ccd16945 | 2467 | int partial_page = (offset || length < folio_size(folio)); |
470decc6 | 2468 | int may_free = 1; |
53e87268 | 2469 | int ret = 0; |
470decc6 | 2470 | |
ccd16945 | 2471 | if (!folio_test_locked(folio)) |
470decc6 | 2472 | BUG(); |
ccd16945 MWO |
2473 | head = folio_buffers(folio); |
2474 | if (!head) | |
53e87268 | 2475 | return 0; |
470decc6 | 2476 | |
ccd16945 | 2477 | BUG_ON(stop > folio_size(folio) || stop < length); |
259709b0 | 2478 | |
470decc6 DK |
2479 | /* We will potentially be playing with lists other than just the |
2480 | * data lists (especially for journaled data mode), so be | |
2481 | * cautious in our locking. */ | |
2482 | ||
ccd16945 | 2483 | bh = head; |
470decc6 DK |
2484 | do { |
2485 | unsigned int next_off = curr_off + bh->b_size; | |
2486 | next = bh->b_this_page; | |
2487 | ||
259709b0 LC |
2488 | if (next_off > stop) |
2489 | return 0; | |
2490 | ||
470decc6 DK |
2491 | if (offset <= curr_off) { |
2492 | /* This block is wholly outside the truncation point */ | |
2493 | lock_buffer(bh); | |
259709b0 | 2494 | ret = journal_unmap_buffer(journal, bh, partial_page); |
470decc6 | 2495 | unlock_buffer(bh); |
53e87268 JK |
2496 | if (ret < 0) |
2497 | return ret; | |
2498 | may_free &= ret; | |
470decc6 DK |
2499 | } |
2500 | curr_off = next_off; | |
2501 | bh = next; | |
2502 | ||
2503 | } while (bh != head); | |
2504 | ||
259709b0 | 2505 | if (!partial_page) { |
68189fef | 2506 | if (may_free && try_to_free_buffers(folio)) |
ccd16945 | 2507 | J_ASSERT(!folio_buffers(folio)); |
470decc6 | 2508 | } |
53e87268 | 2509 | return 0; |
470decc6 DK |
2510 | } |
2511 | ||
2512 | /* | |
2513 | * File a buffer on the given transaction list. | |
2514 | */ | |
f7f4bccb | 2515 | void __jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2516 | transaction_t *transaction, int jlist) |
2517 | { | |
2518 | struct journal_head **list = NULL; | |
2519 | int was_dirty = 0; | |
2520 | struct buffer_head *bh = jh2bh(jh); | |
2521 | ||
46417064 | 2522 | lockdep_assert_held(&jh->b_state_lock); |
470decc6 DK |
2523 | assert_spin_locked(&transaction->t_journal->j_list_lock); |
2524 | ||
2525 | J_ASSERT_JH(jh, jh->b_jlist < BJ_Types); | |
2526 | J_ASSERT_JH(jh, jh->b_transaction == transaction || | |
4019191b | 2527 | jh->b_transaction == NULL); |
470decc6 DK |
2528 | |
2529 | if (jh->b_transaction && jh->b_jlist == jlist) | |
2530 | return; | |
2531 | ||
470decc6 DK |
2532 | if (jlist == BJ_Metadata || jlist == BJ_Reserved || |
2533 | jlist == BJ_Shadow || jlist == BJ_Forget) { | |
f91d1d04 JK |
2534 | /* |
2535 | * For metadata buffers, we track dirty bit in buffer_jbddirty | |
2536 | * instead of buffer_dirty. We should not see a dirty bit set | |
2537 | * here because we clear it in do_get_write_access but e.g. | |
2538 | * tune2fs can modify the sb and set the dirty bit at any time | |
2539 | * so we try to gracefully handle that. | |
2540 | */ | |
2541 | if (buffer_dirty(bh)) | |
2542 | warn_dirty_buffer(bh); | |
470decc6 DK |
2543 | if (test_clear_buffer_dirty(bh) || |
2544 | test_clear_buffer_jbddirty(bh)) | |
2545 | was_dirty = 1; | |
2546 | } | |
2547 | ||
2548 | if (jh->b_transaction) | |
f7f4bccb | 2549 | __jbd2_journal_temp_unlink_buffer(jh); |
de1b7941 JK |
2550 | else |
2551 | jbd2_journal_grab_journal_head(bh); | |
470decc6 DK |
2552 | jh->b_transaction = transaction; |
2553 | ||
2554 | switch (jlist) { | |
2555 | case BJ_None: | |
2556 | J_ASSERT_JH(jh, !jh->b_committed_data); | |
2557 | J_ASSERT_JH(jh, !jh->b_frozen_data); | |
2558 | return; | |
470decc6 DK |
2559 | case BJ_Metadata: |
2560 | transaction->t_nr_buffers++; | |
2561 | list = &transaction->t_buffers; | |
2562 | break; | |
2563 | case BJ_Forget: | |
2564 | list = &transaction->t_forget; | |
2565 | break; | |
470decc6 DK |
2566 | case BJ_Shadow: |
2567 | list = &transaction->t_shadow_list; | |
2568 | break; | |
470decc6 DK |
2569 | case BJ_Reserved: |
2570 | list = &transaction->t_reserved_list; | |
2571 | break; | |
470decc6 DK |
2572 | } |
2573 | ||
2574 | __blist_add_buffer(list, jh); | |
2575 | jh->b_jlist = jlist; | |
2576 | ||
2577 | if (was_dirty) | |
2578 | set_buffer_jbddirty(bh); | |
2579 | } | |
2580 | ||
f7f4bccb | 2581 | void jbd2_journal_file_buffer(struct journal_head *jh, |
470decc6 DK |
2582 | transaction_t *transaction, int jlist) |
2583 | { | |
46417064 | 2584 | spin_lock(&jh->b_state_lock); |
470decc6 | 2585 | spin_lock(&transaction->t_journal->j_list_lock); |
f7f4bccb | 2586 | __jbd2_journal_file_buffer(jh, transaction, jlist); |
470decc6 | 2587 | spin_unlock(&transaction->t_journal->j_list_lock); |
46417064 | 2588 | spin_unlock(&jh->b_state_lock); |
470decc6 DK |
2589 | } |
2590 | ||
2591 | /* | |
2592 | * Remove a buffer from its current buffer list in preparation for | |
2593 | * dropping it from its current transaction entirely. If the buffer has | |
2594 | * already started to be used by a subsequent transaction, refile the | |
2595 | * buffer on that transaction's metadata list. | |
2596 | * | |
de1b7941 | 2597 | * Called under j_list_lock |
46417064 | 2598 | * Called under jh->b_state_lock |
de1b7941 | 2599 | * |
93108ebb JK |
2600 | * When this function returns true, there's no next transaction to refile to |
2601 | * and the caller has to drop jh reference through | |
2602 | * jbd2_journal_put_journal_head(). | |
470decc6 | 2603 | */ |
93108ebb | 2604 | bool __jbd2_journal_refile_buffer(struct journal_head *jh) |
470decc6 | 2605 | { |
ba869023 | 2606 | int was_dirty, jlist; |
470decc6 DK |
2607 | struct buffer_head *bh = jh2bh(jh); |
2608 | ||
46417064 | 2609 | lockdep_assert_held(&jh->b_state_lock); |
470decc6 DK |
2610 | if (jh->b_transaction) |
2611 | assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock); | |
2612 | ||
2613 | /* If the buffer is now unused, just drop it. */ | |
2614 | if (jh->b_next_transaction == NULL) { | |
f7f4bccb | 2615 | __jbd2_journal_unfile_buffer(jh); |
93108ebb | 2616 | return true; |
470decc6 DK |
2617 | } |
2618 | ||
2619 | /* | |
2620 | * It has been modified by a later transaction: add it to the new | |
2621 | * transaction's metadata list. | |
2622 | */ | |
2623 | ||
2624 | was_dirty = test_clear_buffer_jbddirty(bh); | |
f7f4bccb | 2625 | __jbd2_journal_temp_unlink_buffer(jh); |
24dc9864 LC |
2626 | |
2627 | /* | |
2628 | * b_transaction must be set, otherwise the new b_transaction won't | |
2629 | * be holding jh reference | |
2630 | */ | |
2631 | J_ASSERT_JH(jh, jh->b_transaction != NULL); | |
2632 | ||
de1b7941 JK |
2633 | /* |
2634 | * We set b_transaction here because b_next_transaction will inherit | |
2635 | * our jh reference and thus __jbd2_journal_file_buffer() must not | |
2636 | * take a new one. | |
2637 | */ | |
6c5d9112 QC |
2638 | WRITE_ONCE(jh->b_transaction, jh->b_next_transaction); |
2639 | WRITE_ONCE(jh->b_next_transaction, NULL); | |
ba869023 | 2640 | if (buffer_freed(bh)) |
2641 | jlist = BJ_Forget; | |
2642 | else if (jh->b_modified) | |
2643 | jlist = BJ_Metadata; | |
2644 | else | |
2645 | jlist = BJ_Reserved; | |
2646 | __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist); | |
470decc6 DK |
2647 | J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING); |
2648 | ||
2649 | if (was_dirty) | |
2650 | set_buffer_jbddirty(bh); | |
93108ebb | 2651 | return false; |
470decc6 DK |
2652 | } |
2653 | ||
2654 | /* | |
de1b7941 JK |
2655 | * __jbd2_journal_refile_buffer() with necessary locking added. We take our |
2656 | * bh reference so that we can safely unlock bh. | |
2657 | * | |
2658 | * The jh and bh may be freed by this call. | |
470decc6 | 2659 | */ |
f7f4bccb | 2660 | void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh) |
470decc6 | 2661 | { |
93108ebb | 2662 | bool drop; |
470decc6 | 2663 | |
46417064 | 2664 | spin_lock(&jh->b_state_lock); |
470decc6 | 2665 | spin_lock(&journal->j_list_lock); |
93108ebb | 2666 | drop = __jbd2_journal_refile_buffer(jh); |
46417064 | 2667 | spin_unlock(&jh->b_state_lock); |
470decc6 | 2668 | spin_unlock(&journal->j_list_lock); |
93108ebb JK |
2669 | if (drop) |
2670 | jbd2_journal_put_journal_head(jh); | |
470decc6 | 2671 | } |
c851ed54 JK |
2672 | |
2673 | /* | |
2674 | * File inode in the inode list of the handle's transaction | |
2675 | */ | |
41617e1a | 2676 | static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode, |
6ba0e7dc | 2677 | unsigned long flags, loff_t start_byte, loff_t end_byte) |
c851ed54 JK |
2678 | { |
2679 | transaction_t *transaction = handle->h_transaction; | |
41a5b913 | 2680 | journal_t *journal; |
c851ed54 JK |
2681 | |
2682 | if (is_handle_aborted(handle)) | |
41a5b913 TT |
2683 | return -EROFS; |
2684 | journal = transaction->t_journal; | |
c851ed54 | 2685 | |
cb3b3bf2 | 2686 | jbd2_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino, |
c851ed54 JK |
2687 | transaction->t_tid); |
2688 | ||
c851ed54 | 2689 | spin_lock(&journal->j_list_lock); |
41617e1a | 2690 | jinode->i_flags |= flags; |
6ba0e7dc RZ |
2691 | |
2692 | if (jinode->i_dirty_end) { | |
2693 | jinode->i_dirty_start = min(jinode->i_dirty_start, start_byte); | |
2694 | jinode->i_dirty_end = max(jinode->i_dirty_end, end_byte); | |
2695 | } else { | |
2696 | jinode->i_dirty_start = start_byte; | |
2697 | jinode->i_dirty_end = end_byte; | |
2698 | } | |
2699 | ||
41617e1a | 2700 | /* Is inode already attached where we need it? */ |
c851ed54 JK |
2701 | if (jinode->i_transaction == transaction || |
2702 | jinode->i_next_transaction == transaction) | |
2703 | goto done; | |
2704 | ||
81be12c8 JK |
2705 | /* |
2706 | * We only ever set this variable to 1 so the test is safe. Since | |
2707 | * t_need_data_flush is likely to be set, we do the test to save some | |
2708 | * cacheline bouncing | |
2709 | */ | |
2710 | if (!transaction->t_need_data_flush) | |
2711 | transaction->t_need_data_flush = 1; | |
c851ed54 JK |
2712 | /* On some different transaction's list - should be |
2713 | * the committing one */ | |
2714 | if (jinode->i_transaction) { | |
2715 | J_ASSERT(jinode->i_next_transaction == NULL); | |
2716 | J_ASSERT(jinode->i_transaction == | |
2717 | journal->j_committing_transaction); | |
2718 | jinode->i_next_transaction = transaction; | |
2719 | goto done; | |
2720 | } | |
2721 | /* Not on any transaction list... */ | |
2722 | J_ASSERT(!jinode->i_next_transaction); | |
2723 | jinode->i_transaction = transaction; | |
2724 | list_add(&jinode->i_list, &transaction->t_inode_list); | |
2725 | done: | |
2726 | spin_unlock(&journal->j_list_lock); | |
2727 | ||
2728 | return 0; | |
2729 | } | |
2730 | ||
6ba0e7dc RZ |
2731 | int jbd2_journal_inode_ranged_write(handle_t *handle, |
2732 | struct jbd2_inode *jinode, loff_t start_byte, loff_t length) | |
2733 | { | |
2734 | return jbd2_journal_file_inode(handle, jinode, | |
2735 | JI_WRITE_DATA | JI_WAIT_DATA, start_byte, | |
2736 | start_byte + length - 1); | |
2737 | } | |
2738 | ||
2739 | int jbd2_journal_inode_ranged_wait(handle_t *handle, struct jbd2_inode *jinode, | |
2740 | loff_t start_byte, loff_t length) | |
2741 | { | |
2742 | return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA, | |
2743 | start_byte, start_byte + length - 1); | |
41617e1a JK |
2744 | } |
2745 | ||
c851ed54 | 2746 | /* |
7f5aa215 JK |
2747 | * File truncate and transaction commit interact with each other in a |
2748 | * non-trivial way. If a transaction writing data block A is | |
2749 | * committing, we cannot discard the data by truncate until we have | |
2750 | * written them. Otherwise if we crashed after the transaction with | |
2751 | * write has committed but before the transaction with truncate has | |
2752 | * committed, we could see stale data in block A. This function is a | |
2753 | * helper to solve this problem. It starts writeout of the truncated | |
2754 | * part in case it is in the committing transaction. | |
2755 | * | |
2756 | * Filesystem code must call this function when inode is journaled in | |
2757 | * ordered mode before truncation happens and after the inode has been | |
2758 | * placed on orphan list with the new inode size. The second condition | |
2759 | * avoids the race that someone writes new data and we start | |
2760 | * committing the transaction after this function has been called but | |
2761 | * before a transaction for truncate is started (and furthermore it | |
2762 | * allows us to optimize the case where the addition to orphan list | |
2763 | * happens in the same transaction as write --- we don't have to write | |
2764 | * any data in such case). | |
c851ed54 | 2765 | */ |
7f5aa215 JK |
2766 | int jbd2_journal_begin_ordered_truncate(journal_t *journal, |
2767 | struct jbd2_inode *jinode, | |
c851ed54 JK |
2768 | loff_t new_size) |
2769 | { | |
7f5aa215 | 2770 | transaction_t *inode_trans, *commit_trans; |
c851ed54 JK |
2771 | int ret = 0; |
2772 | ||
7f5aa215 JK |
2773 | /* This is a quick check to avoid locking if not necessary */ |
2774 | if (!jinode->i_transaction) | |
c851ed54 | 2775 | goto out; |
7f5aa215 JK |
2776 | /* Locks are here just to force reading of recent values, it is |
2777 | * enough that the transaction was not committing before we started | |
2778 | * a transaction adding the inode to orphan list */ | |
a931da6a | 2779 | read_lock(&journal->j_state_lock); |
c851ed54 | 2780 | commit_trans = journal->j_committing_transaction; |
a931da6a | 2781 | read_unlock(&journal->j_state_lock); |
7f5aa215 JK |
2782 | spin_lock(&journal->j_list_lock); |
2783 | inode_trans = jinode->i_transaction; | |
2784 | spin_unlock(&journal->j_list_lock); | |
2785 | if (inode_trans == commit_trans) { | |
2786 | ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping, | |
c851ed54 JK |
2787 | new_size, LLONG_MAX); |
2788 | if (ret) | |
2789 | jbd2_journal_abort(journal, ret); | |
2790 | } | |
2791 | out: | |
2792 | return ret; | |
2793 | } |