Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1// SPDX-License-Identifier: GPL-2.0+
2/*
3 * linux/fs/jbd2/transaction.c
4 *
5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 *
7 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 *
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>
19#include <linux/jbd2.h>
20#include <linux/errno.h>
21#include <linux/slab.h>
22#include <linux/timer.h>
23#include <linux/mm.h>
24#include <linux/highmem.h>
25#include <linux/hrtimer.h>
26#include <linux/backing-dev.h>
27#include <linux/bug.h>
28#include <linux/module.h>
29#include <linux/sched/mm.h>
30
31#include <trace/events/jbd2.h>
32
33static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
34static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
35
36static struct kmem_cache *transaction_cache;
37int __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);
45 if (!transaction_cache) {
46 pr_emerg("JBD2: failed to create transaction cache\n");
47 return -ENOMEM;
48 }
49 return 0;
50}
51
52void jbd2_journal_destroy_transaction_cache(void)
53{
54 kmem_cache_destroy(transaction_cache);
55 transaction_cache = NULL;
56}
57
58void 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
65/*
66 * jbd2_get_transaction: obtain a new transaction_t object.
67 *
68 * Simply initialise a new transaction. Initialize it in
69 * RUNNING state and add it to the current journal (which should not
70 * have an existing running transaction: we only make a new transaction
71 * once we have started to commit the old one).
72 *
73 * Preconditions:
74 * The journal MUST be locked. We don't perform atomic mallocs on the
75 * new transaction and we can't block without protecting against other
76 * processes trying to touch the journal while it is in transition.
77 *
78 */
79
80static void jbd2_get_transaction(journal_t *journal,
81 transaction_t *transaction)
82{
83 transaction->t_journal = journal;
84 transaction->t_state = T_RUNNING;
85 transaction->t_start_time = ktime_get();
86 transaction->t_tid = journal->j_transaction_sequence++;
87 transaction->t_expires = jiffies + journal->j_commit_interval;
88 atomic_set(&transaction->t_updates, 0);
89 atomic_set(&transaction->t_outstanding_credits,
90 journal->j_transaction_overhead_buffers +
91 atomic_read(&journal->j_reserved_credits));
92 atomic_set(&transaction->t_outstanding_revokes, 0);
93 atomic_set(&transaction->t_handle_count, 0);
94 INIT_LIST_HEAD(&transaction->t_inode_list);
95
96 /* Set up the commit timer for the new transaction. */
97 journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
98 add_timer(&journal->j_commit_timer);
99
100 J_ASSERT(journal->j_running_transaction == NULL);
101 journal->j_running_transaction = transaction;
102 transaction->t_max_wait = 0;
103 transaction->t_start = jiffies;
104 transaction->t_requested = 0;
105}
106
107/*
108 * Handle management.
109 *
110 * A handle_t is an object which represents a single atomic update to a
111 * filesystem, and which tracks all of the modifications which form part
112 * of that one update.
113 */
114
115/*
116 * t_max_wait is carefully updated here with use of atomic compare exchange.
117 * Note that there could be multiplre threads trying to do this simultaneously
118 * hence using cmpxchg to avoid any use of locks in this case.
119 */
120static inline void update_t_max_wait(transaction_t *transaction,
121 unsigned long ts)
122{
123 unsigned long oldts, newts;
124
125 if (time_after(transaction->t_start, ts)) {
126 newts = jbd2_time_diff(ts, transaction->t_start);
127 oldts = READ_ONCE(transaction->t_max_wait);
128 while (oldts < newts)
129 oldts = cmpxchg(&transaction->t_max_wait, oldts, newts);
130 }
131}
132
133/*
134 * Wait until running transaction passes to T_FLUSH state and new transaction
135 * can thus be started. Also starts the commit if needed. The function expects
136 * running transaction to exist and releases j_state_lock.
137 */
138static void wait_transaction_locked(journal_t *journal)
139 __releases(journal->j_state_lock)
140{
141 DEFINE_WAIT(wait);
142 int need_to_start;
143 tid_t tid = journal->j_running_transaction->t_tid;
144
145 prepare_to_wait_exclusive(&journal->j_wait_transaction_locked, &wait,
146 TASK_UNINTERRUPTIBLE);
147 need_to_start = !tid_geq(journal->j_commit_request, tid);
148 read_unlock(&journal->j_state_lock);
149 if (need_to_start)
150 jbd2_log_start_commit(journal, tid);
151 jbd2_might_wait_for_commit(journal);
152 schedule();
153 finish_wait(&journal->j_wait_transaction_locked, &wait);
154}
155
156/*
157 * Wait until running transaction transitions from T_SWITCH to T_FLUSH
158 * state and new transaction can thus be started. The function releases
159 * j_state_lock.
160 */
161static void wait_transaction_switching(journal_t *journal)
162 __releases(journal->j_state_lock)
163{
164 DEFINE_WAIT(wait);
165
166 if (WARN_ON(!journal->j_running_transaction ||
167 journal->j_running_transaction->t_state != T_SWITCH)) {
168 read_unlock(&journal->j_state_lock);
169 return;
170 }
171 prepare_to_wait_exclusive(&journal->j_wait_transaction_locked, &wait,
172 TASK_UNINTERRUPTIBLE);
173 read_unlock(&journal->j_state_lock);
174 /*
175 * We don't call jbd2_might_wait_for_commit() here as there's no
176 * waiting for outstanding handles happening anymore in T_SWITCH state
177 * and handling of reserved handles actually relies on that for
178 * correctness.
179 */
180 schedule();
181 finish_wait(&journal->j_wait_transaction_locked, &wait);
182}
183
184static void sub_reserved_credits(journal_t *journal, int blocks)
185{
186 atomic_sub(blocks, &journal->j_reserved_credits);
187 wake_up(&journal->j_wait_reserved);
188}
189
190/* Maximum number of blocks for user transaction payload */
191static int jbd2_max_user_trans_buffers(journal_t *journal)
192{
193 return journal->j_max_transaction_buffers -
194 journal->j_transaction_overhead_buffers;
195}
196
197/*
198 * Wait until we can add credits for handle to the running transaction. Called
199 * with j_state_lock held for reading. Returns 0 if handle joined the running
200 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
201 * caller must retry.
202 *
203 * Note: because j_state_lock may be dropped depending on the return
204 * value, we need to fake out sparse so ti doesn't complain about a
205 * locking imbalance. Callers of add_transaction_credits will need to
206 * make a similar accomodation.
207 */
208static int add_transaction_credits(journal_t *journal, int blocks,
209 int rsv_blocks)
210__must_hold(&journal->j_state_lock)
211{
212 transaction_t *t = journal->j_running_transaction;
213 int needed;
214 int total = blocks + rsv_blocks;
215
216 /*
217 * If the current transaction is locked down for commit, wait
218 * for the lock to be released.
219 */
220 if (t->t_state != T_RUNNING) {
221 WARN_ON_ONCE(t->t_state >= T_FLUSH);
222 wait_transaction_locked(journal);
223 __acquire(&journal->j_state_lock); /* fake out sparse */
224 return 1;
225 }
226
227 /*
228 * If there is not enough space left in the log to write all
229 * potential buffers requested by this operation, we need to
230 * stall pending a log checkpoint to free some more log space.
231 */
232 needed = atomic_add_return(total, &t->t_outstanding_credits);
233 if (needed > journal->j_max_transaction_buffers) {
234 /*
235 * If the current transaction is already too large,
236 * then start to commit it: we can then go back and
237 * attach this handle to a new transaction.
238 */
239 atomic_sub(total, &t->t_outstanding_credits);
240
241 /*
242 * Is the number of reserved credits in the current transaction too
243 * big to fit this handle? Wait until reserved credits are freed.
244 */
245 if (atomic_read(&journal->j_reserved_credits) + total >
246 jbd2_max_user_trans_buffers(journal)) {
247 read_unlock(&journal->j_state_lock);
248 jbd2_might_wait_for_commit(journal);
249 wait_event(journal->j_wait_reserved,
250 atomic_read(&journal->j_reserved_credits) + total <=
251 jbd2_max_user_trans_buffers(journal));
252 __acquire(&journal->j_state_lock); /* fake out sparse */
253 return 1;
254 }
255
256 wait_transaction_locked(journal);
257 __acquire(&journal->j_state_lock); /* fake out sparse */
258 return 1;
259 }
260
261 /*
262 * The commit code assumes that it can get enough log space
263 * without forcing a checkpoint. This is *critical* for
264 * correctness: a checkpoint of a buffer which is also
265 * associated with a committing transaction creates a deadlock,
266 * so commit simply cannot force through checkpoints.
267 *
268 * We must therefore ensure the necessary space in the journal
269 * *before* starting to dirty potentially checkpointed buffers
270 * in the new transaction.
271 */
272 if (jbd2_log_space_left(journal) < journal->j_max_transaction_buffers) {
273 atomic_sub(total, &t->t_outstanding_credits);
274 read_unlock(&journal->j_state_lock);
275 jbd2_might_wait_for_commit(journal);
276 write_lock(&journal->j_state_lock);
277 if (jbd2_log_space_left(journal) <
278 journal->j_max_transaction_buffers)
279 __jbd2_log_wait_for_space(journal);
280 write_unlock(&journal->j_state_lock);
281 __acquire(&journal->j_state_lock); /* fake out sparse */
282 return 1;
283 }
284
285 /* No reservation? We are done... */
286 if (!rsv_blocks)
287 return 0;
288
289 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
290 /* We allow at most half of a transaction to be reserved */
291 if (needed > jbd2_max_user_trans_buffers(journal) / 2) {
292 sub_reserved_credits(journal, rsv_blocks);
293 atomic_sub(total, &t->t_outstanding_credits);
294 read_unlock(&journal->j_state_lock);
295 jbd2_might_wait_for_commit(journal);
296 wait_event(journal->j_wait_reserved,
297 atomic_read(&journal->j_reserved_credits) + rsv_blocks
298 <= jbd2_max_user_trans_buffers(journal) / 2);
299 __acquire(&journal->j_state_lock); /* fake out sparse */
300 return 1;
301 }
302 return 0;
303}
304
305/*
306 * start_this_handle: Given a handle, deal with any locking or stalling
307 * needed to make sure that there is enough journal space for the handle
308 * to begin. Attach the handle to a transaction and set up the
309 * transaction's buffer credits.
310 */
311
312static int start_this_handle(journal_t *journal, handle_t *handle,
313 gfp_t gfp_mask)
314{
315 transaction_t *transaction, *new_transaction = NULL;
316 int blocks = handle->h_total_credits;
317 int rsv_blocks = 0;
318 unsigned long ts = jiffies;
319
320 if (handle->h_rsv_handle)
321 rsv_blocks = handle->h_rsv_handle->h_total_credits;
322
323 /*
324 * Limit the number of reserved credits to 1/2 of maximum transaction
325 * size and limit the number of total credits to not exceed maximum
326 * transaction size per operation.
327 */
328 if (rsv_blocks > jbd2_max_user_trans_buffers(journal) / 2 ||
329 rsv_blocks + blocks > jbd2_max_user_trans_buffers(journal)) {
330 printk(KERN_ERR "JBD2: %s wants too many credits "
331 "credits:%d rsv_credits:%d max:%d\n",
332 current->comm, blocks, rsv_blocks,
333 jbd2_max_user_trans_buffers(journal));
334 WARN_ON(1);
335 return -ENOSPC;
336 }
337
338alloc_transaction:
339 /*
340 * This check is racy but it is just an optimization of allocating new
341 * transaction early if there are high chances we'll need it. If we
342 * guess wrong, we'll retry or free unused transaction.
343 */
344 if (!data_race(journal->j_running_transaction)) {
345 /*
346 * If __GFP_FS is not present, then we may be being called from
347 * inside the fs writeback layer, so we MUST NOT fail.
348 */
349 if ((gfp_mask & __GFP_FS) == 0)
350 gfp_mask |= __GFP_NOFAIL;
351 new_transaction = kmem_cache_zalloc(transaction_cache,
352 gfp_mask);
353 if (!new_transaction)
354 return -ENOMEM;
355 }
356
357 jbd2_debug(3, "New handle %p going live.\n", handle);
358
359 /*
360 * We need to hold j_state_lock until t_updates has been incremented,
361 * for proper journal barrier handling
362 */
363repeat:
364 read_lock(&journal->j_state_lock);
365 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
366 if (is_journal_aborted(journal) ||
367 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
368 read_unlock(&journal->j_state_lock);
369 jbd2_journal_free_transaction(new_transaction);
370 return -EROFS;
371 }
372
373 /*
374 * Wait on the journal's transaction barrier if necessary. Specifically
375 * we allow reserved handles to proceed because otherwise commit could
376 * deadlock on page writeback not being able to complete.
377 */
378 if (!handle->h_reserved && journal->j_barrier_count) {
379 read_unlock(&journal->j_state_lock);
380 wait_event(journal->j_wait_transaction_locked,
381 journal->j_barrier_count == 0);
382 goto repeat;
383 }
384
385 if (!journal->j_running_transaction) {
386 read_unlock(&journal->j_state_lock);
387 if (!new_transaction)
388 goto alloc_transaction;
389 write_lock(&journal->j_state_lock);
390 if (!journal->j_running_transaction &&
391 (handle->h_reserved || !journal->j_barrier_count)) {
392 jbd2_get_transaction(journal, new_transaction);
393 new_transaction = NULL;
394 }
395 write_unlock(&journal->j_state_lock);
396 goto repeat;
397 }
398
399 transaction = journal->j_running_transaction;
400
401 if (!handle->h_reserved) {
402 /* We may have dropped j_state_lock - restart in that case */
403 if (add_transaction_credits(journal, blocks, rsv_blocks)) {
404 /*
405 * add_transaction_credits releases
406 * j_state_lock on a non-zero return
407 */
408 __release(&journal->j_state_lock);
409 goto repeat;
410 }
411 } else {
412 /*
413 * We have handle reserved so we are allowed to join T_LOCKED
414 * transaction and we don't have to check for transaction size
415 * and journal space. But we still have to wait while running
416 * transaction is being switched to a committing one as it
417 * won't wait for any handles anymore.
418 */
419 if (transaction->t_state == T_SWITCH) {
420 wait_transaction_switching(journal);
421 goto repeat;
422 }
423 sub_reserved_credits(journal, blocks);
424 handle->h_reserved = 0;
425 }
426
427 /* OK, account for the buffers that this operation expects to
428 * use and add the handle to the running transaction.
429 */
430 update_t_max_wait(transaction, ts);
431 handle->h_transaction = transaction;
432 handle->h_requested_credits = blocks;
433 handle->h_revoke_credits_requested = handle->h_revoke_credits;
434 handle->h_start_jiffies = jiffies;
435 atomic_inc(&transaction->t_updates);
436 atomic_inc(&transaction->t_handle_count);
437 jbd2_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
438 handle, blocks,
439 atomic_read(&transaction->t_outstanding_credits),
440 jbd2_log_space_left(journal));
441 read_unlock(&journal->j_state_lock);
442 current->journal_info = handle;
443
444 rwsem_acquire_read(&journal->j_trans_commit_map, 0, 1, _THIS_IP_);
445 jbd2_journal_free_transaction(new_transaction);
446 /*
447 * Ensure that no allocations done while the transaction is open are
448 * going to recurse back to the fs layer.
449 */
450 handle->saved_alloc_context = memalloc_nofs_save();
451 return 0;
452}
453
454/* Allocate a new handle. This should probably be in a slab... */
455static handle_t *new_handle(int nblocks)
456{
457 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
458 if (!handle)
459 return NULL;
460 handle->h_total_credits = nblocks;
461 handle->h_ref = 1;
462
463 return handle;
464}
465
466handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
467 int revoke_records, gfp_t gfp_mask,
468 unsigned int type, unsigned int line_no)
469{
470 handle_t *handle = journal_current_handle();
471 int err;
472
473 if (!journal)
474 return ERR_PTR(-EROFS);
475
476 if (handle) {
477 if (WARN_ON_ONCE(handle->h_transaction->t_journal != journal))
478 return ERR_PTR(-EINVAL);
479 handle->h_ref++;
480 return handle;
481 }
482
483 nblocks += DIV_ROUND_UP(revoke_records,
484 journal->j_revoke_records_per_block);
485 handle = new_handle(nblocks);
486 if (!handle)
487 return ERR_PTR(-ENOMEM);
488 if (rsv_blocks) {
489 handle_t *rsv_handle;
490
491 rsv_handle = new_handle(rsv_blocks);
492 if (!rsv_handle) {
493 jbd2_free_handle(handle);
494 return ERR_PTR(-ENOMEM);
495 }
496 rsv_handle->h_reserved = 1;
497 rsv_handle->h_journal = journal;
498 handle->h_rsv_handle = rsv_handle;
499 }
500 handle->h_revoke_credits = revoke_records;
501
502 err = start_this_handle(journal, handle, gfp_mask);
503 if (err < 0) {
504 if (handle->h_rsv_handle)
505 jbd2_free_handle(handle->h_rsv_handle);
506 jbd2_free_handle(handle);
507 return ERR_PTR(err);
508 }
509 handle->h_type = type;
510 handle->h_line_no = line_no;
511 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
512 handle->h_transaction->t_tid, type,
513 line_no, nblocks);
514
515 return handle;
516}
517EXPORT_SYMBOL(jbd2__journal_start);
518
519
520/**
521 * jbd2_journal_start() - Obtain a new handle.
522 * @journal: Journal to start transaction on.
523 * @nblocks: number of block buffer we might modify
524 *
525 * We make sure that the transaction can guarantee at least nblocks of
526 * modified buffers in the log. We block until the log can guarantee
527 * that much space. Additionally, if rsv_blocks > 0, we also create another
528 * handle with rsv_blocks reserved blocks in the journal. This handle is
529 * stored in h_rsv_handle. It is not attached to any particular transaction
530 * and thus doesn't block transaction commit. If the caller uses this reserved
531 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
532 * on the parent handle will dispose the reserved one. Reserved handle has to
533 * be converted to a normal handle using jbd2_journal_start_reserved() before
534 * it can be used.
535 *
536 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
537 * on failure.
538 */
539handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
540{
541 return jbd2__journal_start(journal, nblocks, 0, 0, GFP_NOFS, 0, 0);
542}
543EXPORT_SYMBOL(jbd2_journal_start);
544
545static void __jbd2_journal_unreserve_handle(handle_t *handle, transaction_t *t)
546{
547 journal_t *journal = handle->h_journal;
548
549 WARN_ON(!handle->h_reserved);
550 sub_reserved_credits(journal, handle->h_total_credits);
551 if (t)
552 atomic_sub(handle->h_total_credits, &t->t_outstanding_credits);
553}
554
555void jbd2_journal_free_reserved(handle_t *handle)
556{
557 journal_t *journal = handle->h_journal;
558
559 /* Get j_state_lock to pin running transaction if it exists */
560 read_lock(&journal->j_state_lock);
561 __jbd2_journal_unreserve_handle(handle, journal->j_running_transaction);
562 read_unlock(&journal->j_state_lock);
563 jbd2_free_handle(handle);
564}
565EXPORT_SYMBOL(jbd2_journal_free_reserved);
566
567/**
568 * jbd2_journal_start_reserved() - start reserved handle
569 * @handle: handle to start
570 * @type: for handle statistics
571 * @line_no: for handle statistics
572 *
573 * Start handle that has been previously reserved with jbd2_journal_reserve().
574 * This attaches @handle to the running transaction (or creates one if there's
575 * not transaction running). Unlike jbd2_journal_start() this function cannot
576 * block on journal commit, checkpointing, or similar stuff. It can block on
577 * memory allocation or frozen journal though.
578 *
579 * Return 0 on success, non-zero on error - handle is freed in that case.
580 */
581int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
582 unsigned int line_no)
583{
584 journal_t *journal = handle->h_journal;
585 int ret = -EIO;
586
587 if (WARN_ON(!handle->h_reserved)) {
588 /* Someone passed in normal handle? Just stop it. */
589 jbd2_journal_stop(handle);
590 return ret;
591 }
592 /*
593 * Usefulness of mixing of reserved and unreserved handles is
594 * questionable. So far nobody seems to need it so just error out.
595 */
596 if (WARN_ON(current->journal_info)) {
597 jbd2_journal_free_reserved(handle);
598 return ret;
599 }
600
601 handle->h_journal = NULL;
602 /*
603 * GFP_NOFS is here because callers are likely from writeback or
604 * similarly constrained call sites
605 */
606 ret = start_this_handle(journal, handle, GFP_NOFS);
607 if (ret < 0) {
608 handle->h_journal = journal;
609 jbd2_journal_free_reserved(handle);
610 return ret;
611 }
612 handle->h_type = type;
613 handle->h_line_no = line_no;
614 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
615 handle->h_transaction->t_tid, type,
616 line_no, handle->h_total_credits);
617 return 0;
618}
619EXPORT_SYMBOL(jbd2_journal_start_reserved);
620
621/**
622 * jbd2_journal_extend() - extend buffer credits.
623 * @handle: handle to 'extend'
624 * @nblocks: nr blocks to try to extend by.
625 * @revoke_records: number of revoke records to try to extend by.
626 *
627 * Some transactions, such as large extends and truncates, can be done
628 * atomically all at once or in several stages. The operation requests
629 * a credit for a number of buffer modifications in advance, but can
630 * extend its credit if it needs more.
631 *
632 * jbd2_journal_extend tries to give the running handle more buffer credits.
633 * It does not guarantee that allocation - this is a best-effort only.
634 * The calling process MUST be able to deal cleanly with a failure to
635 * extend here.
636 *
637 * Return 0 on success, non-zero on failure.
638 *
639 * return code < 0 implies an error
640 * return code > 0 implies normal transaction-full status.
641 */
642int jbd2_journal_extend(handle_t *handle, int nblocks, int revoke_records)
643{
644 transaction_t *transaction = handle->h_transaction;
645 journal_t *journal;
646 int result;
647 int wanted;
648
649 if (is_handle_aborted(handle))
650 return -EROFS;
651 journal = transaction->t_journal;
652
653 result = 1;
654
655 read_lock(&journal->j_state_lock);
656
657 /* Don't extend a locked-down transaction! */
658 if (transaction->t_state != T_RUNNING) {
659 jbd2_debug(3, "denied handle %p %d blocks: "
660 "transaction not running\n", handle, nblocks);
661 goto error_out;
662 }
663
664 nblocks += DIV_ROUND_UP(
665 handle->h_revoke_credits_requested + revoke_records,
666 journal->j_revoke_records_per_block) -
667 DIV_ROUND_UP(
668 handle->h_revoke_credits_requested,
669 journal->j_revoke_records_per_block);
670 wanted = atomic_add_return(nblocks,
671 &transaction->t_outstanding_credits);
672
673 if (wanted > journal->j_max_transaction_buffers) {
674 jbd2_debug(3, "denied handle %p %d blocks: "
675 "transaction too large\n", handle, nblocks);
676 atomic_sub(nblocks, &transaction->t_outstanding_credits);
677 goto error_out;
678 }
679
680 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
681 transaction->t_tid,
682 handle->h_type, handle->h_line_no,
683 handle->h_total_credits,
684 nblocks);
685
686 handle->h_total_credits += nblocks;
687 handle->h_requested_credits += nblocks;
688 handle->h_revoke_credits += revoke_records;
689 handle->h_revoke_credits_requested += revoke_records;
690 result = 0;
691
692 jbd2_debug(3, "extended handle %p by %d\n", handle, nblocks);
693error_out:
694 read_unlock(&journal->j_state_lock);
695 return result;
696}
697
698static void stop_this_handle(handle_t *handle)
699{
700 transaction_t *transaction = handle->h_transaction;
701 journal_t *journal = transaction->t_journal;
702 int revokes;
703
704 J_ASSERT(journal_current_handle() == handle);
705 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
706 current->journal_info = NULL;
707 /*
708 * Subtract necessary revoke descriptor blocks from handle credits. We
709 * take care to account only for revoke descriptor blocks the
710 * transaction will really need as large sequences of transactions with
711 * small numbers of revokes are relatively common.
712 */
713 revokes = handle->h_revoke_credits_requested - handle->h_revoke_credits;
714 if (revokes) {
715 int t_revokes, revoke_descriptors;
716 int rr_per_blk = journal->j_revoke_records_per_block;
717
718 WARN_ON_ONCE(DIV_ROUND_UP(revokes, rr_per_blk)
719 > handle->h_total_credits);
720 t_revokes = atomic_add_return(revokes,
721 &transaction->t_outstanding_revokes);
722 revoke_descriptors =
723 DIV_ROUND_UP(t_revokes, rr_per_blk) -
724 DIV_ROUND_UP(t_revokes - revokes, rr_per_blk);
725 handle->h_total_credits -= revoke_descriptors;
726 }
727 atomic_sub(handle->h_total_credits,
728 &transaction->t_outstanding_credits);
729 if (handle->h_rsv_handle)
730 __jbd2_journal_unreserve_handle(handle->h_rsv_handle,
731 transaction);
732 if (atomic_dec_and_test(&transaction->t_updates))
733 wake_up(&journal->j_wait_updates);
734
735 rwsem_release(&journal->j_trans_commit_map, _THIS_IP_);
736 /*
737 * Scope of the GFP_NOFS context is over here and so we can restore the
738 * original alloc context.
739 */
740 memalloc_nofs_restore(handle->saved_alloc_context);
741}
742
743/**
744 * jbd2__journal_restart() - restart a handle .
745 * @handle: handle to restart
746 * @nblocks: nr credits requested
747 * @revoke_records: number of revoke record credits requested
748 * @gfp_mask: memory allocation flags (for start_this_handle)
749 *
750 * Restart a handle for a multi-transaction filesystem
751 * operation.
752 *
753 * If the jbd2_journal_extend() call above fails to grant new buffer credits
754 * to a running handle, a call to jbd2_journal_restart will commit the
755 * handle's transaction so far and reattach the handle to a new
756 * transaction capable of guaranteeing the requested number of
757 * credits. We preserve reserved handle if there's any attached to the
758 * passed in handle.
759 */
760int jbd2__journal_restart(handle_t *handle, int nblocks, int revoke_records,
761 gfp_t gfp_mask)
762{
763 transaction_t *transaction = handle->h_transaction;
764 journal_t *journal;
765 tid_t tid;
766 int need_to_start;
767 int ret;
768
769 /* If we've had an abort of any type, don't even think about
770 * actually doing the restart! */
771 if (is_handle_aborted(handle))
772 return 0;
773 journal = transaction->t_journal;
774 tid = transaction->t_tid;
775
776 /*
777 * First unlink the handle from its current transaction, and start the
778 * commit on that.
779 */
780 jbd2_debug(2, "restarting handle %p\n", handle);
781 stop_this_handle(handle);
782 handle->h_transaction = NULL;
783
784 /*
785 * TODO: If we use READ_ONCE / WRITE_ONCE for j_commit_request we can
786 * get rid of pointless j_state_lock traffic like this.
787 */
788 read_lock(&journal->j_state_lock);
789 need_to_start = !tid_geq(journal->j_commit_request, tid);
790 read_unlock(&journal->j_state_lock);
791 if (need_to_start)
792 jbd2_log_start_commit(journal, tid);
793 handle->h_total_credits = nblocks +
794 DIV_ROUND_UP(revoke_records,
795 journal->j_revoke_records_per_block);
796 handle->h_revoke_credits = revoke_records;
797 ret = start_this_handle(journal, handle, gfp_mask);
798 trace_jbd2_handle_restart(journal->j_fs_dev->bd_dev,
799 ret ? 0 : handle->h_transaction->t_tid,
800 handle->h_type, handle->h_line_no,
801 handle->h_total_credits);
802 return ret;
803}
804EXPORT_SYMBOL(jbd2__journal_restart);
805
806
807int jbd2_journal_restart(handle_t *handle, int nblocks)
808{
809 return jbd2__journal_restart(handle, nblocks, 0, GFP_NOFS);
810}
811EXPORT_SYMBOL(jbd2_journal_restart);
812
813/*
814 * Waits for any outstanding t_updates to finish.
815 * This is called with write j_state_lock held.
816 */
817void jbd2_journal_wait_updates(journal_t *journal)
818{
819 DEFINE_WAIT(wait);
820
821 while (1) {
822 /*
823 * Note that the running transaction can get freed under us if
824 * this transaction is getting committed in
825 * jbd2_journal_commit_transaction() ->
826 * jbd2_journal_free_transaction(). This can only happen when we
827 * release j_state_lock -> schedule() -> acquire j_state_lock.
828 * Hence we should everytime retrieve new j_running_transaction
829 * value (after j_state_lock release acquire cycle), else it may
830 * lead to use-after-free of old freed transaction.
831 */
832 transaction_t *transaction = journal->j_running_transaction;
833
834 if (!transaction)
835 break;
836
837 prepare_to_wait(&journal->j_wait_updates, &wait,
838 TASK_UNINTERRUPTIBLE);
839 if (!atomic_read(&transaction->t_updates)) {
840 finish_wait(&journal->j_wait_updates, &wait);
841 break;
842 }
843 write_unlock(&journal->j_state_lock);
844 schedule();
845 finish_wait(&journal->j_wait_updates, &wait);
846 write_lock(&journal->j_state_lock);
847 }
848}
849
850/**
851 * jbd2_journal_lock_updates () - establish a transaction barrier.
852 * @journal: Journal to establish a barrier on.
853 *
854 * This locks out any further updates from being started, and blocks
855 * until all existing updates have completed, returning only once the
856 * journal is in a quiescent state with no updates running.
857 *
858 * The journal lock should not be held on entry.
859 */
860void jbd2_journal_lock_updates(journal_t *journal)
861{
862 jbd2_might_wait_for_commit(journal);
863
864 write_lock(&journal->j_state_lock);
865 ++journal->j_barrier_count;
866
867 /* Wait until there are no reserved handles */
868 if (atomic_read(&journal->j_reserved_credits)) {
869 write_unlock(&journal->j_state_lock);
870 wait_event(journal->j_wait_reserved,
871 atomic_read(&journal->j_reserved_credits) == 0);
872 write_lock(&journal->j_state_lock);
873 }
874
875 /* Wait until there are no running t_updates */
876 jbd2_journal_wait_updates(journal);
877
878 write_unlock(&journal->j_state_lock);
879
880 /*
881 * We have now established a barrier against other normal updates, but
882 * we also need to barrier against other jbd2_journal_lock_updates() calls
883 * to make sure that we serialise special journal-locked operations
884 * too.
885 */
886 mutex_lock(&journal->j_barrier);
887}
888
889/**
890 * jbd2_journal_unlock_updates () - release barrier
891 * @journal: Journal to release the barrier on.
892 *
893 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
894 *
895 * Should be called without the journal lock held.
896 */
897void jbd2_journal_unlock_updates (journal_t *journal)
898{
899 J_ASSERT(journal->j_barrier_count != 0);
900
901 mutex_unlock(&journal->j_barrier);
902 write_lock(&journal->j_state_lock);
903 --journal->j_barrier_count;
904 write_unlock(&journal->j_state_lock);
905 wake_up_all(&journal->j_wait_transaction_locked);
906}
907
908static void warn_dirty_buffer(struct buffer_head *bh)
909{
910 printk(KERN_WARNING
911 "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
912 "There's a risk of filesystem corruption in case of system "
913 "crash.\n",
914 bh->b_bdev, (unsigned long long)bh->b_blocknr);
915}
916
917/* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
918static void jbd2_freeze_jh_data(struct journal_head *jh)
919{
920 char *source;
921 struct buffer_head *bh = jh2bh(jh);
922
923 J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
924 source = kmap_local_folio(bh->b_folio, bh_offset(bh));
925 /* Fire data frozen trigger just before we copy the data */
926 jbd2_buffer_frozen_trigger(jh, source, jh->b_triggers);
927 memcpy(jh->b_frozen_data, source, bh->b_size);
928 kunmap_local(source);
929
930 /*
931 * Now that the frozen data is saved off, we need to store any matching
932 * triggers.
933 */
934 jh->b_frozen_triggers = jh->b_triggers;
935}
936
937/*
938 * If the buffer is already part of the current transaction, then there
939 * is nothing we need to do. If it is already part of a prior
940 * transaction which we are still committing to disk, then we need to
941 * make sure that we do not overwrite the old copy: we do copy-out to
942 * preserve the copy going to disk. We also account the buffer against
943 * the handle's metadata buffer credits (unless the buffer is already
944 * part of the transaction, that is).
945 *
946 */
947static int
948do_get_write_access(handle_t *handle, struct journal_head *jh,
949 int force_copy)
950{
951 struct buffer_head *bh;
952 transaction_t *transaction = handle->h_transaction;
953 journal_t *journal;
954 int error;
955 char *frozen_buffer = NULL;
956 unsigned long start_lock, time_lock;
957
958 journal = transaction->t_journal;
959
960 jbd2_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
961
962 JBUFFER_TRACE(jh, "entry");
963repeat:
964 bh = jh2bh(jh);
965
966 /* @@@ Need to check for errors here at some point. */
967
968 start_lock = jiffies;
969 lock_buffer(bh);
970 spin_lock(&jh->b_state_lock);
971
972 /* If it takes too long to lock the buffer, trace it */
973 time_lock = jbd2_time_diff(start_lock, jiffies);
974 if (time_lock > HZ/10)
975 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
976 jiffies_to_msecs(time_lock));
977
978 /* We now hold the buffer lock so it is safe to query the buffer
979 * state. Is the buffer dirty?
980 *
981 * If so, there are two possibilities. The buffer may be
982 * non-journaled, and undergoing a quite legitimate writeback.
983 * Otherwise, it is journaled, and we don't expect dirty buffers
984 * in that state (the buffers should be marked JBD_Dirty
985 * instead.) So either the IO is being done under our own
986 * control and this is a bug, or it's a third party IO such as
987 * dump(8) (which may leave the buffer scheduled for read ---
988 * ie. locked but not dirty) or tune2fs (which may actually have
989 * the buffer dirtied, ugh.) */
990
991 if (buffer_dirty(bh) && jh->b_transaction) {
992 warn_dirty_buffer(bh);
993 /*
994 * We need to clean the dirty flag and we must do it under the
995 * buffer lock to be sure we don't race with running write-out.
996 */
997 JBUFFER_TRACE(jh, "Journalling dirty buffer");
998 clear_buffer_dirty(bh);
999 /*
1000 * The buffer is going to be added to BJ_Reserved list now and
1001 * nothing guarantees jbd2_journal_dirty_metadata() will be
1002 * ever called for it. So we need to set jbddirty bit here to
1003 * make sure the buffer is dirtied and written out when the
1004 * journaling machinery is done with it.
1005 */
1006 set_buffer_jbddirty(bh);
1007 }
1008
1009 error = -EROFS;
1010 if (is_handle_aborted(handle)) {
1011 spin_unlock(&jh->b_state_lock);
1012 unlock_buffer(bh);
1013 goto out;
1014 }
1015 error = 0;
1016
1017 /*
1018 * The buffer is already part of this transaction if b_transaction or
1019 * b_next_transaction points to it
1020 */
1021 if (jh->b_transaction == transaction ||
1022 jh->b_next_transaction == transaction) {
1023 unlock_buffer(bh);
1024 goto done;
1025 }
1026
1027 /*
1028 * this is the first time this transaction is touching this buffer,
1029 * reset the modified flag
1030 */
1031 jh->b_modified = 0;
1032
1033 /*
1034 * If the buffer is not journaled right now, we need to make sure it
1035 * doesn't get written to disk before the caller actually commits the
1036 * new data
1037 */
1038 if (!jh->b_transaction) {
1039 JBUFFER_TRACE(jh, "no transaction");
1040 if (WARN_ON_ONCE(jh->b_next_transaction)) {
1041 spin_unlock(&jh->b_state_lock);
1042 unlock_buffer(bh);
1043 error = -EINVAL;
1044 jbd2_journal_abort(journal, error);
1045 goto out;
1046 }
1047 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1048 /*
1049 * Make sure all stores to jh (b_modified, b_frozen_data) are
1050 * visible before attaching it to the running transaction.
1051 * Paired with barrier in jbd2_write_access_granted()
1052 */
1053 smp_wmb();
1054 spin_lock(&journal->j_list_lock);
1055 if (test_clear_buffer_dirty(bh)) {
1056 /*
1057 * Execute buffer dirty clearing and jh->b_transaction
1058 * assignment under journal->j_list_lock locked to
1059 * prevent bh being removed from checkpoint list if
1060 * the buffer is in an intermediate state (not dirty
1061 * and jh->b_transaction is NULL).
1062 */
1063 JBUFFER_TRACE(jh, "Journalling dirty buffer");
1064 set_buffer_jbddirty(bh);
1065 }
1066 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1067 spin_unlock(&journal->j_list_lock);
1068 unlock_buffer(bh);
1069 goto done;
1070 }
1071 unlock_buffer(bh);
1072
1073 /*
1074 * If there is already a copy-out version of this buffer, then we don't
1075 * need to make another one
1076 */
1077 if (jh->b_frozen_data) {
1078 JBUFFER_TRACE(jh, "has frozen data");
1079 if (WARN_ON_ONCE(jh->b_next_transaction)) {
1080 spin_unlock(&jh->b_state_lock);
1081 error = -EINVAL;
1082 jbd2_journal_abort(journal, error);
1083 goto out;
1084 }
1085 goto attach_next;
1086 }
1087
1088 JBUFFER_TRACE(jh, "owned by older transaction");
1089 if (WARN_ON_ONCE(jh->b_next_transaction ||
1090 jh->b_transaction !=
1091 journal->j_committing_transaction)) {
1092 pr_err("JBD2: %s: assertion failure: b_next_transaction=%p b_transaction=%p j_committing_transaction=%p\n",
1093 journal->j_devname, jh->b_next_transaction,
1094 jh->b_transaction, journal->j_committing_transaction);
1095 spin_unlock(&jh->b_state_lock);
1096 error = -EINVAL;
1097 jbd2_journal_abort(journal, error);
1098 goto out;
1099 }
1100
1101 /*
1102 * There is one case we have to be very careful about. If the
1103 * committing transaction is currently writing this buffer out to disk
1104 * and has NOT made a copy-out, then we cannot modify the buffer
1105 * contents at all right now. The essence of copy-out is that it is
1106 * the extra copy, not the primary copy, which gets journaled. If the
1107 * primary copy is already going to disk then we cannot do copy-out
1108 * here.
1109 */
1110 if (buffer_shadow(bh)) {
1111 JBUFFER_TRACE(jh, "on shadow: sleep");
1112 spin_unlock(&jh->b_state_lock);
1113 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
1114 goto repeat;
1115 }
1116
1117 /*
1118 * Only do the copy if the currently-owning transaction still needs it.
1119 * If buffer isn't on BJ_Metadata list, the committing transaction is
1120 * past that stage (here we use the fact that BH_Shadow is set under
1121 * bh_state lock together with refiling to BJ_Shadow list and at this
1122 * point we know the buffer doesn't have BH_Shadow set).
1123 *
1124 * Subtle point, though: if this is a get_undo_access, then we will be
1125 * relying on the frozen_data to contain the new value of the
1126 * committed_data record after the transaction, so we HAVE to force the
1127 * frozen_data copy in that case.
1128 */
1129 if (jh->b_jlist == BJ_Metadata || force_copy) {
1130 JBUFFER_TRACE(jh, "generate frozen data");
1131 if (!frozen_buffer) {
1132 JBUFFER_TRACE(jh, "allocate memory for buffer");
1133 spin_unlock(&jh->b_state_lock);
1134 frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
1135 GFP_NOFS | __GFP_NOFAIL);
1136 goto repeat;
1137 }
1138 jh->b_frozen_data = frozen_buffer;
1139 frozen_buffer = NULL;
1140 jbd2_freeze_jh_data(jh);
1141 }
1142attach_next:
1143 /*
1144 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
1145 * before attaching it to the running transaction. Paired with barrier
1146 * in jbd2_write_access_granted()
1147 */
1148 smp_wmb();
1149 jh->b_next_transaction = transaction;
1150
1151done:
1152 spin_unlock(&jh->b_state_lock);
1153
1154 /*
1155 * If we are about to journal a buffer, then any revoke pending on it is
1156 * no longer valid
1157 */
1158 jbd2_journal_cancel_revoke(handle, jh);
1159
1160out:
1161 if (unlikely(frozen_buffer)) /* It's usually NULL */
1162 jbd2_free(frozen_buffer, bh->b_size);
1163
1164 JBUFFER_TRACE(jh, "exit");
1165 return error;
1166}
1167
1168/* Fast check whether buffer is already attached to the required transaction */
1169static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1170 bool undo)
1171{
1172 struct journal_head *jh;
1173 bool ret = false;
1174
1175 /* Dirty buffers require special handling... */
1176 if (buffer_dirty(bh))
1177 return false;
1178
1179 /*
1180 * RCU protects us from dereferencing freed pages. So the checks we do
1181 * are guaranteed not to oops. However the jh slab object can get freed
1182 * & reallocated while we work with it. So we have to be careful. When
1183 * we see jh attached to the running transaction, we know it must stay
1184 * so until the transaction is committed. Thus jh won't be freed and
1185 * will be attached to the same bh while we run. However it can
1186 * happen jh gets freed, reallocated, and attached to the transaction
1187 * just after we get pointer to it from bh. So we have to be careful
1188 * and recheck jh still belongs to our bh before we return success.
1189 */
1190 rcu_read_lock();
1191 if (!buffer_jbd(bh))
1192 goto out;
1193 /* This should be bh2jh() but that doesn't work with inline functions */
1194 jh = READ_ONCE(bh->b_private);
1195 if (!jh)
1196 goto out;
1197 /* For undo access buffer must have data copied */
1198 if (undo && !jh->b_committed_data)
1199 goto out;
1200 if (READ_ONCE(jh->b_transaction) != handle->h_transaction &&
1201 READ_ONCE(jh->b_next_transaction) != handle->h_transaction)
1202 goto out;
1203 /*
1204 * There are two reasons for the barrier here:
1205 * 1) Make sure to fetch b_bh after we did previous checks so that we
1206 * detect when jh went through free, realloc, attach to transaction
1207 * while we were checking. Paired with implicit barrier in that path.
1208 * 2) So that access to bh done after jbd2_write_access_granted()
1209 * doesn't get reordered and see inconsistent state of concurrent
1210 * do_get_write_access().
1211 */
1212 smp_mb();
1213 if (unlikely(jh->b_bh != bh))
1214 goto out;
1215 ret = true;
1216out:
1217 rcu_read_unlock();
1218 return ret;
1219}
1220
1221/**
1222 * jbd2_journal_get_write_access() - notify intent to modify a buffer
1223 * for metadata (not data) update.
1224 * @handle: transaction to add buffer modifications to
1225 * @bh: bh to be used for metadata writes
1226 *
1227 * Returns: error code or 0 on success.
1228 *
1229 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1230 * because we're ``write()ing`` a buffer which is also part of a shared mapping.
1231 */
1232
1233int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
1234{
1235 struct journal_head *jh;
1236 journal_t *journal;
1237 int rc;
1238
1239 if (is_handle_aborted(handle))
1240 return -EROFS;
1241
1242 journal = handle->h_transaction->t_journal;
1243 rc = jbd2_check_fs_dev_write_error(journal);
1244 if (rc) {
1245 /*
1246 * If the fs dev has writeback errors, it may have failed
1247 * to async write out metadata buffers in the background.
1248 * In this case, we could read old data from disk and write
1249 * it out again, which may lead to on-disk filesystem
1250 * inconsistency. Aborting journal can avoid it happen.
1251 */
1252 jbd2_journal_abort(journal, rc);
1253 return -EIO;
1254 }
1255
1256 if (jbd2_write_access_granted(handle, bh, false))
1257 return 0;
1258
1259 jh = jbd2_journal_add_journal_head(bh);
1260 /* We do not want to get caught playing with fields which the
1261 * log thread also manipulates. Make sure that the buffer
1262 * completes any outstanding IO before proceeding. */
1263 rc = do_get_write_access(handle, jh, 0);
1264 jbd2_journal_put_journal_head(jh);
1265 return rc;
1266}
1267
1268
1269/*
1270 * When the user wants to journal a newly created buffer_head
1271 * (ie. getblk() returned a new buffer and we are going to populate it
1272 * manually rather than reading off disk), then we need to keep the
1273 * buffer_head locked until it has been completely filled with new
1274 * data. In this case, we should be able to make the assertion that
1275 * the bh is not already part of an existing transaction.
1276 *
1277 * The buffer should already be locked by the caller by this point.
1278 * There is no lock ranking violation: it was a newly created,
1279 * unlocked buffer beforehand. */
1280
1281/**
1282 * jbd2_journal_get_create_access () - notify intent to use newly created bh
1283 * @handle: transaction to new buffer to
1284 * @bh: new buffer.
1285 *
1286 * Call this if you create a new bh.
1287 */
1288int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
1289{
1290 transaction_t *transaction = handle->h_transaction;
1291 journal_t *journal;
1292 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
1293 int err;
1294
1295 jbd2_debug(5, "journal_head %p\n", jh);
1296 err = -EROFS;
1297 if (is_handle_aborted(handle))
1298 goto out;
1299 journal = transaction->t_journal;
1300 err = 0;
1301
1302 JBUFFER_TRACE(jh, "entry");
1303 /*
1304 * The buffer may already belong to this transaction due to pre-zeroing
1305 * in the filesystem's new_block code. It may also be on the previous,
1306 * committing transaction's lists, but it HAS to be in Forget state in
1307 * that case: the transaction must have deleted the buffer for it to be
1308 * reused here.
1309 * In the case of file system data inconsistency, for example, if the
1310 * block bitmap of a referenced block is not set, it can lead to the
1311 * situation where a block being committed is allocated and used again.
1312 * As a result, the following condition will not be satisfied, so here
1313 * we directly trigger a JBD abort instead of immediately invoking
1314 * bugon.
1315 */
1316 spin_lock(&jh->b_state_lock);
1317 if (!(jh->b_transaction == transaction || jh->b_transaction == NULL ||
1318 (jh->b_transaction == journal->j_committing_transaction &&
1319 jh->b_jlist == BJ_Forget)) || jh->b_next_transaction != NULL) {
1320 err = -EROFS;
1321 spin_unlock(&jh->b_state_lock);
1322 jbd2_journal_abort(journal, err);
1323 goto out;
1324 }
1325
1326 if (WARN_ON_ONCE(!buffer_locked(jh2bh(jh)))) {
1327 err = -EINVAL;
1328 spin_unlock(&jh->b_state_lock);
1329 jbd2_journal_abort(journal, err);
1330 goto out;
1331 }
1332
1333 if (jh->b_transaction == NULL) {
1334 /*
1335 * Previous jbd2_journal_forget() could have left the buffer
1336 * with jbddirty bit set because it was being committed. When
1337 * the commit finished, we've filed the buffer for
1338 * checkpointing and marked it dirty. Now we are reallocating
1339 * the buffer so the transaction freeing it must have
1340 * committed and so it's safe to clear the dirty bit.
1341 */
1342 clear_buffer_dirty(jh2bh(jh));
1343 /* first access by this transaction */
1344 jh->b_modified = 0;
1345
1346 JBUFFER_TRACE(jh, "file as BJ_Reserved");
1347 spin_lock(&journal->j_list_lock);
1348 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1349 spin_unlock(&journal->j_list_lock);
1350 } else if (jh->b_transaction == journal->j_committing_transaction) {
1351 /* first access by this transaction */
1352 jh->b_modified = 0;
1353
1354 JBUFFER_TRACE(jh, "set next transaction");
1355 spin_lock(&journal->j_list_lock);
1356 jh->b_next_transaction = transaction;
1357 spin_unlock(&journal->j_list_lock);
1358 }
1359 spin_unlock(&jh->b_state_lock);
1360
1361 /*
1362 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1363 * blocks which contain freed but then revoked metadata. We need
1364 * to cancel the revoke in case we end up freeing it yet again
1365 * and the reallocating as data - this would cause a second revoke,
1366 * which hits an assertion error.
1367 */
1368 JBUFFER_TRACE(jh, "cancelling revoke");
1369 jbd2_journal_cancel_revoke(handle, jh);
1370out:
1371 jbd2_journal_put_journal_head(jh);
1372 return err;
1373}
1374
1375/**
1376 * jbd2_journal_get_undo_access() - Notify intent to modify metadata with
1377 * non-rewindable consequences
1378 * @handle: transaction
1379 * @bh: buffer to undo
1380 *
1381 * Sometimes there is a need to distinguish between metadata which has
1382 * been committed to disk and that which has not. The ext3fs code uses
1383 * this for freeing and allocating space, we have to make sure that we
1384 * do not reuse freed space until the deallocation has been committed,
1385 * since if we overwrote that space we would make the delete
1386 * un-rewindable in case of a crash.
1387 *
1388 * To deal with that, jbd2_journal_get_undo_access requests write access to a
1389 * buffer for parts of non-rewindable operations such as delete
1390 * operations on the bitmaps. The journaling code must keep a copy of
1391 * the buffer's contents prior to the undo_access call until such time
1392 * as we know that the buffer has definitely been committed to disk.
1393 *
1394 * We never need to know which transaction the committed data is part
1395 * of, buffers touched here are guaranteed to be dirtied later and so
1396 * will be committed to a new transaction in due course, at which point
1397 * we can discard the old committed data pointer.
1398 *
1399 * Returns error number or 0 on success.
1400 */
1401int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
1402{
1403 int err;
1404 struct journal_head *jh;
1405 char *committed_data = NULL;
1406
1407 if (is_handle_aborted(handle))
1408 return -EROFS;
1409
1410 if (jbd2_write_access_granted(handle, bh, true))
1411 return 0;
1412
1413 jh = jbd2_journal_add_journal_head(bh);
1414 JBUFFER_TRACE(jh, "entry");
1415
1416 /*
1417 * Do this first --- it can drop the journal lock, so we want to
1418 * make sure that obtaining the committed_data is done
1419 * atomically wrt. completion of any outstanding commits.
1420 */
1421 err = do_get_write_access(handle, jh, 1);
1422 if (err)
1423 goto out;
1424
1425repeat:
1426 if (!jh->b_committed_data)
1427 committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1428 GFP_NOFS|__GFP_NOFAIL);
1429
1430 spin_lock(&jh->b_state_lock);
1431 if (!jh->b_committed_data) {
1432 /* Copy out the current buffer contents into the
1433 * preserved, committed copy. */
1434 JBUFFER_TRACE(jh, "generate b_committed data");
1435 if (!committed_data) {
1436 spin_unlock(&jh->b_state_lock);
1437 goto repeat;
1438 }
1439
1440 jh->b_committed_data = committed_data;
1441 committed_data = NULL;
1442 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1443 }
1444 spin_unlock(&jh->b_state_lock);
1445out:
1446 jbd2_journal_put_journal_head(jh);
1447 if (unlikely(committed_data))
1448 jbd2_free(committed_data, bh->b_size);
1449 return err;
1450}
1451
1452/**
1453 * jbd2_journal_set_triggers() - Add triggers for commit writeout
1454 * @bh: buffer to trigger on
1455 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1456 *
1457 * Set any triggers on this journal_head. This is always safe, because
1458 * triggers for a committing buffer will be saved off, and triggers for
1459 * a running transaction will match the buffer in that transaction.
1460 *
1461 * Call with NULL to clear the triggers.
1462 */
1463void jbd2_journal_set_triggers(struct buffer_head *bh,
1464 struct jbd2_buffer_trigger_type *type)
1465{
1466 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
1467
1468 if (WARN_ON_ONCE(!jh))
1469 return;
1470 jh->b_triggers = type;
1471 jbd2_journal_put_journal_head(jh);
1472}
1473
1474void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
1475 struct jbd2_buffer_trigger_type *triggers)
1476{
1477 struct buffer_head *bh = jh2bh(jh);
1478
1479 if (!triggers || !triggers->t_frozen)
1480 return;
1481
1482 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
1483}
1484
1485void jbd2_buffer_abort_trigger(struct journal_head *jh,
1486 struct jbd2_buffer_trigger_type *triggers)
1487{
1488 if (!triggers || !triggers->t_abort)
1489 return;
1490
1491 triggers->t_abort(triggers, jh2bh(jh));
1492}
1493
1494/**
1495 * jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
1496 * @handle: transaction to add buffer to.
1497 * @bh: buffer to mark
1498 *
1499 * mark dirty metadata which needs to be journaled as part of the current
1500 * transaction.
1501 *
1502 * The buffer must have previously had jbd2_journal_get_write_access()
1503 * called so that it has a valid journal_head attached to the buffer
1504 * head.
1505 *
1506 * The buffer is placed on the transaction's metadata list and is marked
1507 * as belonging to the transaction.
1508 *
1509 * Returns error number or 0 on success.
1510 *
1511 * Special care needs to be taken if the buffer already belongs to the
1512 * current committing transaction (in which case we should have frozen
1513 * data present for that commit). In that case, we don't relink the
1514 * buffer: that only gets done when the old transaction finally
1515 * completes its commit.
1516 */
1517int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
1518{
1519 transaction_t *transaction = handle->h_transaction;
1520 journal_t *journal = transaction->t_journal;
1521 struct journal_head *jh;
1522 int ret = 0;
1523
1524 if (!buffer_jbd(bh))
1525 return -EUCLEAN;
1526
1527 /*
1528 * We don't grab jh reference here since the buffer must be part
1529 * of the running transaction.
1530 */
1531 jh = bh2jh(bh);
1532 jbd2_debug(5, "journal_head %p\n", jh);
1533 JBUFFER_TRACE(jh, "entry");
1534
1535 /*
1536 * This and the following assertions are unreliable since we may see jh
1537 * in inconsistent state unless we grab bh_state lock. But this is
1538 * crucial to catch bugs so let's do a reliable check until the
1539 * lockless handling is fully proven.
1540 */
1541 if (data_race(jh->b_transaction != transaction &&
1542 jh->b_next_transaction != transaction)) {
1543 spin_lock(&jh->b_state_lock);
1544 if (WARN_ON_ONCE(jh->b_transaction != transaction &&
1545 jh->b_next_transaction != transaction)) {
1546 pr_err("JBD2: %s: assertion failure: b_transaction=%p transaction=%p b_next_transaction=%p\n",
1547 journal->j_devname, jh->b_transaction,
1548 transaction, jh->b_next_transaction);
1549 ret = -EINVAL;
1550 goto out_unlock_bh;
1551 }
1552 spin_unlock(&jh->b_state_lock);
1553 }
1554 if (data_race(jh->b_modified == 1)) {
1555 /* If it's in our transaction it must be in BJ_Metadata list. */
1556 if (data_race(jh->b_transaction == transaction &&
1557 jh->b_jlist != BJ_Metadata)) {
1558 spin_lock(&jh->b_state_lock);
1559 if (WARN_ON_ONCE(jh->b_transaction == transaction &&
1560 jh->b_jlist != BJ_Metadata)) {
1561 pr_err("JBD2: assertion failure: h_type=%u h_line_no=%u block_no=%llu jlist=%u\n",
1562 handle->h_type, handle->h_line_no,
1563 (unsigned long long) bh->b_blocknr,
1564 jh->b_jlist);
1565 ret = -EINVAL;
1566 goto out_unlock_bh;
1567 }
1568 spin_unlock(&jh->b_state_lock);
1569 }
1570 goto out;
1571 }
1572
1573 spin_lock(&jh->b_state_lock);
1574
1575 if (is_handle_aborted(handle)) {
1576 /*
1577 * Check journal aborting with @jh->b_state_lock locked,
1578 * since 'jh->b_transaction' could be replaced with
1579 * 'jh->b_next_transaction' during old transaction
1580 * committing if journal aborted, which may fail
1581 * assertion on 'jh->b_frozen_data == NULL'.
1582 */
1583 ret = -EROFS;
1584 goto out_unlock_bh;
1585 }
1586
1587 if (jh->b_modified == 0) {
1588 /*
1589 * This buffer's got modified and becoming part
1590 * of the transaction. This needs to be done
1591 * once a transaction -bzzz
1592 */
1593 if (WARN_ON_ONCE(jbd2_handle_buffer_credits(handle) <= 0)) {
1594 ret = -ENOSPC;
1595 goto out_unlock_bh;
1596 }
1597 jh->b_modified = 1;
1598 handle->h_total_credits--;
1599 }
1600
1601 /*
1602 * fastpath, to avoid expensive locking. If this buffer is already
1603 * on the running transaction's metadata list there is nothing to do.
1604 * Nobody can take it off again because there is a handle open.
1605 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1606 * result in this test being false, so we go in and take the locks.
1607 */
1608 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1609 JBUFFER_TRACE(jh, "fastpath");
1610 if (unlikely(jh->b_transaction !=
1611 journal->j_running_transaction)) {
1612 printk(KERN_ERR "JBD2: %s: "
1613 "jh->b_transaction (%llu, %p, %u) != "
1614 "journal->j_running_transaction (%p, %u)\n",
1615 journal->j_devname,
1616 (unsigned long long) bh->b_blocknr,
1617 jh->b_transaction,
1618 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1619 journal->j_running_transaction,
1620 journal->j_running_transaction ?
1621 journal->j_running_transaction->t_tid : 0);
1622 ret = -EINVAL;
1623 }
1624 goto out_unlock_bh;
1625 }
1626
1627 set_buffer_jbddirty(bh);
1628
1629 /*
1630 * Metadata already on the current transaction list doesn't
1631 * need to be filed. Metadata on another transaction's list must
1632 * be committing, and will be refiled once the commit completes:
1633 * leave it alone for now.
1634 */
1635 if (jh->b_transaction != transaction) {
1636 JBUFFER_TRACE(jh, "already on other transaction");
1637 if (unlikely(((jh->b_transaction !=
1638 journal->j_committing_transaction)) ||
1639 (jh->b_next_transaction != transaction))) {
1640 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1641 "bad jh for block %llu: "
1642 "transaction (%p, %u), "
1643 "jh->b_transaction (%p, %u), "
1644 "jh->b_next_transaction (%p, %u), jlist %u\n",
1645 journal->j_devname,
1646 (unsigned long long) bh->b_blocknr,
1647 transaction, transaction->t_tid,
1648 jh->b_transaction,
1649 jh->b_transaction ?
1650 jh->b_transaction->t_tid : 0,
1651 jh->b_next_transaction,
1652 jh->b_next_transaction ?
1653 jh->b_next_transaction->t_tid : 0,
1654 jh->b_jlist);
1655 WARN_ON(1);
1656 ret = -EINVAL;
1657 }
1658 /* And this case is illegal: we can't reuse another
1659 * transaction's data buffer, ever. */
1660 goto out_unlock_bh;
1661 }
1662
1663 /* That test should have eliminated the following case: */
1664 if (WARN_ON_ONCE(jh->b_frozen_data)) {
1665 ret = -EINVAL;
1666 goto out_unlock_bh;
1667 }
1668
1669 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1670 spin_lock(&journal->j_list_lock);
1671 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
1672 spin_unlock(&journal->j_list_lock);
1673out_unlock_bh:
1674 spin_unlock(&jh->b_state_lock);
1675out:
1676 JBUFFER_TRACE(jh, "exit");
1677 return ret;
1678}
1679
1680/**
1681 * jbd2_journal_forget() - bforget() for potentially-journaled buffers.
1682 * @handle: transaction handle
1683 * @bh: bh to 'forget'
1684 *
1685 * We can only do the bforget if there are no commits pending against the
1686 * buffer. If the buffer is dirty in the current running transaction we
1687 * can safely unlink it.
1688 *
1689 * bh may not be a journalled buffer at all - it may be a non-JBD
1690 * buffer which came off the hashtable. Check for this.
1691 *
1692 * Decrements bh->b_count by one.
1693 *
1694 * Allow this call even if the handle has aborted --- it may be part of
1695 * the caller's cleanup after an abort.
1696 */
1697int jbd2_journal_forget(handle_t *handle, struct buffer_head *bh)
1698{
1699 transaction_t *transaction = handle->h_transaction;
1700 journal_t *journal;
1701 struct journal_head *jh;
1702 int drop_reserve = 0;
1703 int err = 0;
1704 int was_modified = 0;
1705 int wait_for_writeback = 0;
1706 int abort_journal = 0;
1707
1708 if (is_handle_aborted(handle))
1709 return -EROFS;
1710 journal = transaction->t_journal;
1711
1712 BUFFER_TRACE(bh, "entry");
1713
1714 jh = jbd2_journal_grab_journal_head(bh);
1715 if (!jh) {
1716 __bforget(bh);
1717 return 0;
1718 }
1719
1720 spin_lock(&jh->b_state_lock);
1721
1722 /* Critical error: attempting to delete a bitmap buffer, maybe?
1723 * Don't do any jbd operations, and return an error. */
1724 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1725 "inconsistent data on disk")) {
1726 err = -EIO;
1727 goto drop;
1728 }
1729
1730 /* keep track of whether or not this transaction modified us */
1731 was_modified = jh->b_modified;
1732
1733 /*
1734 * The buffer's going from the transaction, we must drop
1735 * all references -bzzz
1736 */
1737 jh->b_modified = 0;
1738
1739 if (jh->b_transaction == transaction) {
1740 if (WARN_ON_ONCE(jh->b_frozen_data)) {
1741 err = -EINVAL;
1742 abort_journal = 1;
1743 goto drop;
1744 }
1745
1746 /* If we are forgetting a buffer which is already part
1747 * of this transaction, then we can just drop it from
1748 * the transaction immediately. */
1749 clear_buffer_dirty(bh);
1750 clear_buffer_jbddirty(bh);
1751
1752 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1753
1754 /*
1755 * we only want to drop a reference if this transaction
1756 * modified the buffer
1757 */
1758 if (was_modified)
1759 drop_reserve = 1;
1760
1761 /*
1762 * We are no longer going to journal this buffer.
1763 * However, the commit of this transaction is still
1764 * important to the buffer: the delete that we are now
1765 * processing might obsolete an old log entry, so by
1766 * committing, we can satisfy the buffer's checkpoint.
1767 *
1768 * So, if we have a checkpoint on the buffer, we should
1769 * now refile the buffer on our BJ_Forget list so that
1770 * we know to remove the checkpoint after we commit.
1771 */
1772
1773 spin_lock(&journal->j_list_lock);
1774 if (jh->b_cp_transaction) {
1775 __jbd2_journal_temp_unlink_buffer(jh);
1776 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1777 } else {
1778 __jbd2_journal_unfile_buffer(jh);
1779 jbd2_journal_put_journal_head(jh);
1780 }
1781 spin_unlock(&journal->j_list_lock);
1782 } else if (jh->b_transaction) {
1783 if (WARN_ON_ONCE(jh->b_transaction != journal->j_committing_transaction)) {
1784 err = -EINVAL;
1785 abort_journal = 1;
1786 goto drop;
1787 }
1788 /* However, if the buffer is still owned by a prior
1789 * (committing) transaction, we can't drop it yet... */
1790 JBUFFER_TRACE(jh, "belongs to older transaction");
1791 /* ... but we CAN drop it from the new transaction through
1792 * marking the buffer as freed and set j_next_transaction to
1793 * the new transaction, so that not only the commit code
1794 * knows it should clear dirty bits when it is done with the
1795 * buffer, but also the buffer can be checkpointed only
1796 * after the new transaction commits. */
1797
1798 set_buffer_freed(bh);
1799
1800 if (!jh->b_next_transaction) {
1801 spin_lock(&journal->j_list_lock);
1802 jh->b_next_transaction = transaction;
1803 spin_unlock(&journal->j_list_lock);
1804 } else {
1805 if (WARN_ON_ONCE(jh->b_next_transaction != transaction)) {
1806 err = -EINVAL;
1807 abort_journal = 1;
1808 goto drop;
1809 }
1810
1811 /*
1812 * only drop a reference if this transaction modified
1813 * the buffer
1814 */
1815 if (was_modified)
1816 drop_reserve = 1;
1817 }
1818 } else {
1819 /*
1820 * Finally, if the buffer is not belongs to any
1821 * transaction, we can just drop it now if it has no
1822 * checkpoint.
1823 */
1824 spin_lock(&journal->j_list_lock);
1825 if (!jh->b_cp_transaction) {
1826 JBUFFER_TRACE(jh, "belongs to none transaction");
1827 spin_unlock(&journal->j_list_lock);
1828 goto drop;
1829 }
1830
1831 /*
1832 * Otherwise, if the buffer has been written to disk,
1833 * it is safe to remove the checkpoint and drop it.
1834 */
1835 if (jbd2_journal_try_remove_checkpoint(jh) >= 0) {
1836 spin_unlock(&journal->j_list_lock);
1837 goto drop;
1838 }
1839
1840 /*
1841 * The buffer has not yet been written to disk. We should
1842 * either clear the buffer or ensure that the ongoing I/O
1843 * is completed, and attach this buffer to current
1844 * transaction so that the buffer can be checkpointed only
1845 * after the current transaction commits.
1846 */
1847 clear_buffer_dirty(bh);
1848 wait_for_writeback = 1;
1849 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1850 spin_unlock(&journal->j_list_lock);
1851 }
1852drop:
1853 __brelse(bh);
1854 spin_unlock(&jh->b_state_lock);
1855 if (abort_journal)
1856 jbd2_journal_abort(journal, err);
1857 if (wait_for_writeback)
1858 wait_on_buffer(bh);
1859 jbd2_journal_put_journal_head(jh);
1860 if (drop_reserve) {
1861 /* no need to reserve log space for this block -bzzz */
1862 handle->h_total_credits++;
1863 }
1864 return err;
1865}
1866
1867/**
1868 * jbd2_journal_stop() - complete a transaction
1869 * @handle: transaction to complete.
1870 *
1871 * All done for a particular handle.
1872 *
1873 * There is not much action needed here. We just return any remaining
1874 * buffer credits to the transaction and remove the handle. The only
1875 * complication is that we need to start a commit operation if the
1876 * filesystem is marked for synchronous update.
1877 *
1878 * jbd2_journal_stop itself will not usually return an error, but it may
1879 * do so in unusual circumstances. In particular, expect it to
1880 * return -EIO if a jbd2_journal_abort has been executed since the
1881 * transaction began.
1882 */
1883int jbd2_journal_stop(handle_t *handle)
1884{
1885 transaction_t *transaction = handle->h_transaction;
1886 journal_t *journal;
1887 int err = 0, wait_for_commit = 0;
1888 tid_t tid;
1889 pid_t pid;
1890
1891 if (--handle->h_ref > 0) {
1892 jbd2_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1893 handle->h_ref);
1894 if (is_handle_aborted(handle))
1895 return -EIO;
1896 return 0;
1897 }
1898 if (!transaction) {
1899 /*
1900 * Handle is already detached from the transaction so there is
1901 * nothing to do other than free the handle.
1902 */
1903 memalloc_nofs_restore(handle->saved_alloc_context);
1904 goto free_and_exit;
1905 }
1906 journal = transaction->t_journal;
1907 tid = transaction->t_tid;
1908
1909 if (is_handle_aborted(handle))
1910 err = -EIO;
1911
1912 jbd2_debug(4, "Handle %p going down\n", handle);
1913 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
1914 tid, handle->h_type, handle->h_line_no,
1915 jiffies - handle->h_start_jiffies,
1916 handle->h_sync, handle->h_requested_credits,
1917 (handle->h_requested_credits -
1918 handle->h_total_credits));
1919
1920 /*
1921 * Implement synchronous transaction batching. If the handle
1922 * was synchronous, don't force a commit immediately. Let's
1923 * yield and let another thread piggyback onto this
1924 * transaction. Keep doing that while new threads continue to
1925 * arrive. It doesn't cost much - we're about to run a commit
1926 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1927 * operations by 30x or more...
1928 *
1929 * We try and optimize the sleep time against what the
1930 * underlying disk can do, instead of having a static sleep
1931 * time. This is useful for the case where our storage is so
1932 * fast that it is more optimal to go ahead and force a flush
1933 * and wait for the transaction to be committed than it is to
1934 * wait for an arbitrary amount of time for new writers to
1935 * join the transaction. We achieve this by measuring how
1936 * long it takes to commit a transaction, and compare it with
1937 * how long this transaction has been running, and if run time
1938 * < commit time then we sleep for the delta and commit. This
1939 * greatly helps super fast disks that would see slowdowns as
1940 * more threads started doing fsyncs.
1941 *
1942 * But don't do this if this process was the most recent one
1943 * to perform a synchronous write. We do this to detect the
1944 * case where a single process is doing a stream of sync
1945 * writes. No point in waiting for joiners in that case.
1946 *
1947 * Setting max_batch_time to 0 disables this completely.
1948 */
1949 pid = current->pid;
1950 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1951 journal->j_max_batch_time) {
1952 u64 commit_time, trans_time;
1953
1954 journal->j_last_sync_writer = pid;
1955
1956 read_lock(&journal->j_state_lock);
1957 commit_time = journal->j_average_commit_time;
1958 read_unlock(&journal->j_state_lock);
1959
1960 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1961 transaction->t_start_time));
1962
1963 commit_time = max_t(u64, commit_time,
1964 1000*journal->j_min_batch_time);
1965 commit_time = min_t(u64, commit_time,
1966 1000*journal->j_max_batch_time);
1967
1968 if (trans_time < commit_time) {
1969 ktime_t expires = ktime_add_ns(ktime_get(),
1970 commit_time);
1971 set_current_state(TASK_UNINTERRUPTIBLE);
1972 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1973 }
1974 }
1975
1976 if (handle->h_sync)
1977 transaction->t_synchronous_commit = 1;
1978
1979 /*
1980 * If the handle is marked SYNC, we need to set another commit
1981 * going! We also want to force a commit if the transaction is too
1982 * old now.
1983 */
1984 if (handle->h_sync ||
1985 time_after_eq(jiffies, transaction->t_expires)) {
1986 /* Do this even for aborted journals: an abort still
1987 * completes the commit thread, it just doesn't write
1988 * anything to disk. */
1989
1990 jbd2_debug(2, "transaction too old, requesting commit for "
1991 "handle %p\n", handle);
1992 /* This is non-blocking */
1993 jbd2_log_start_commit(journal, tid);
1994
1995 /*
1996 * Special case: JBD2_SYNC synchronous updates require us
1997 * to wait for the commit to complete.
1998 */
1999 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
2000 wait_for_commit = 1;
2001 }
2002
2003 /*
2004 * Once stop_this_handle() drops t_updates, the transaction could start
2005 * committing on us and eventually disappear. So we must not
2006 * dereference transaction pointer again after calling
2007 * stop_this_handle().
2008 */
2009 stop_this_handle(handle);
2010
2011 if (wait_for_commit)
2012 err = jbd2_log_wait_commit(journal, tid);
2013
2014free_and_exit:
2015 if (handle->h_rsv_handle)
2016 jbd2_free_handle(handle->h_rsv_handle);
2017 jbd2_free_handle(handle);
2018 return err;
2019}
2020
2021/*
2022 *
2023 * List management code snippets: various functions for manipulating the
2024 * transaction buffer lists.
2025 *
2026 */
2027
2028/*
2029 * Append a buffer to a transaction list, given the transaction's list head
2030 * pointer.
2031 *
2032 * j_list_lock is held.
2033 *
2034 * jh->b_state_lock is held.
2035 */
2036
2037static inline void
2038__blist_add_buffer(struct journal_head **list, struct journal_head *jh)
2039{
2040 if (!*list) {
2041 jh->b_tnext = jh->b_tprev = jh;
2042 *list = jh;
2043 } else {
2044 /* Insert at the tail of the list to preserve order */
2045 struct journal_head *first = *list, *last = first->b_tprev;
2046 jh->b_tprev = last;
2047 jh->b_tnext = first;
2048 last->b_tnext = first->b_tprev = jh;
2049 }
2050}
2051
2052/*
2053 * Remove a buffer from a transaction list, given the transaction's list
2054 * head pointer.
2055 *
2056 * Called with j_list_lock held, and the journal may not be locked.
2057 *
2058 * jh->b_state_lock is held.
2059 */
2060
2061static inline void
2062__blist_del_buffer(struct journal_head **list, struct journal_head *jh)
2063{
2064 if (*list == jh) {
2065 *list = jh->b_tnext;
2066 if (*list == jh)
2067 *list = NULL;
2068 }
2069 jh->b_tprev->b_tnext = jh->b_tnext;
2070 jh->b_tnext->b_tprev = jh->b_tprev;
2071}
2072
2073/*
2074 * Remove a buffer from the appropriate transaction list.
2075 *
2076 * Note that this function can *change* the value of
2077 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
2078 * t_reserved_list. If the caller is holding onto a copy of one of these
2079 * pointers, it could go bad. Generally the caller needs to re-read the
2080 * pointer from the transaction_t.
2081 *
2082 * Called under j_list_lock.
2083 */
2084static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
2085{
2086 struct journal_head **list = NULL;
2087 transaction_t *transaction;
2088 struct buffer_head *bh = jh2bh(jh);
2089
2090 lockdep_assert_held(&jh->b_state_lock);
2091 transaction = jh->b_transaction;
2092 if (transaction)
2093 assert_spin_locked(&transaction->t_journal->j_list_lock);
2094
2095 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2096 if (jh->b_jlist != BJ_None)
2097 J_ASSERT_JH(jh, transaction != NULL);
2098
2099 switch (jh->b_jlist) {
2100 case BJ_None:
2101 return;
2102 case BJ_Metadata:
2103 transaction->t_nr_buffers--;
2104 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
2105 list = &transaction->t_buffers;
2106 break;
2107 case BJ_Forget:
2108 list = &transaction->t_forget;
2109 break;
2110 case BJ_Shadow:
2111 list = &transaction->t_shadow_list;
2112 break;
2113 case BJ_Reserved:
2114 list = &transaction->t_reserved_list;
2115 break;
2116 }
2117
2118 __blist_del_buffer(list, jh);
2119 jh->b_jlist = BJ_None;
2120 if (transaction && is_journal_aborted(transaction->t_journal))
2121 clear_buffer_jbddirty(bh);
2122 else if (test_clear_buffer_jbddirty(bh))
2123 mark_buffer_dirty(bh); /* Expose it to the VM */
2124}
2125
2126/*
2127 * Remove buffer from all transactions. The caller is responsible for dropping
2128 * the jh reference that belonged to the transaction.
2129 *
2130 * Called with bh_state lock and j_list_lock
2131 */
2132static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
2133{
2134 J_ASSERT_JH(jh, jh->b_transaction != NULL);
2135 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2136
2137 __jbd2_journal_temp_unlink_buffer(jh);
2138 jh->b_transaction = NULL;
2139}
2140
2141/**
2142 * jbd2_journal_try_to_free_buffers() - try to free page buffers.
2143 * @journal: journal for operation
2144 * @folio: Folio to detach data from.
2145 *
2146 * For all the buffers on this page,
2147 * if they are fully written out ordered data, move them onto BUF_CLEAN
2148 * so try_to_free_buffers() can reap them.
2149 *
2150 * This function returns non-zero if we wish try_to_free_buffers()
2151 * to be called. We do this if the page is releasable by try_to_free_buffers().
2152 * We also do it if the page has locked or dirty buffers and the caller wants
2153 * us to perform sync or async writeout.
2154 *
2155 * This complicates JBD locking somewhat. We aren't protected by the
2156 * BKL here. We wish to remove the buffer from its committing or
2157 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
2158 *
2159 * This may *change* the value of transaction_t->t_datalist, so anyone
2160 * who looks at t_datalist needs to lock against this function.
2161 *
2162 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
2163 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
2164 * will come out of the lock with the buffer dirty, which makes it
2165 * ineligible for release here.
2166 *
2167 * Who else is affected by this? hmm... Really the only contender
2168 * is do_get_write_access() - it could be looking at the buffer while
2169 * journal_try_to_free_buffer() is changing its state. But that
2170 * cannot happen because we never reallocate freed data as metadata
2171 * while the data is part of a transaction. Yes?
2172 *
2173 * Return false on failure, true on success
2174 */
2175bool jbd2_journal_try_to_free_buffers(journal_t *journal, struct folio *folio)
2176{
2177 struct buffer_head *head;
2178 struct buffer_head *bh;
2179 bool ret = false;
2180
2181 if (WARN_ON_ONCE(!folio_test_locked(folio)))
2182 return false;
2183
2184 head = folio_buffers(folio);
2185 bh = head;
2186 do {
2187 struct journal_head *jh;
2188
2189 /*
2190 * We take our own ref against the journal_head here to avoid
2191 * having to add tons of locking around each instance of
2192 * jbd2_journal_put_journal_head().
2193 */
2194 jh = jbd2_journal_grab_journal_head(bh);
2195 if (!jh)
2196 continue;
2197
2198 spin_lock(&jh->b_state_lock);
2199 if (!jh->b_transaction && !jh->b_next_transaction) {
2200 spin_lock(&journal->j_list_lock);
2201 /* Remove written-back checkpointed metadata buffer */
2202 if (jh->b_cp_transaction != NULL)
2203 jbd2_journal_try_remove_checkpoint(jh);
2204 spin_unlock(&journal->j_list_lock);
2205 }
2206 spin_unlock(&jh->b_state_lock);
2207 jbd2_journal_put_journal_head(jh);
2208 if (buffer_jbd(bh))
2209 goto busy;
2210 } while ((bh = bh->b_this_page) != head);
2211
2212 ret = try_to_free_buffers(folio);
2213busy:
2214 return ret;
2215}
2216
2217/*
2218 * This buffer is no longer needed. If it is on an older transaction's
2219 * checkpoint list we need to record it on this transaction's forget list
2220 * to pin this buffer (and hence its checkpointing transaction) down until
2221 * this transaction commits. If the buffer isn't on a checkpoint list, we
2222 * release it.
2223 * Returns non-zero if JBD no longer has an interest in the buffer.
2224 *
2225 * Called under j_list_lock.
2226 *
2227 * Called under jh->b_state_lock.
2228 */
2229static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2230{
2231 int may_free = 1;
2232 struct buffer_head *bh = jh2bh(jh);
2233
2234 if (jh->b_cp_transaction) {
2235 JBUFFER_TRACE(jh, "on running+cp transaction");
2236 __jbd2_journal_temp_unlink_buffer(jh);
2237 /*
2238 * We don't want to write the buffer anymore, clear the
2239 * bit so that we don't confuse checks in
2240 * __jbd2_journal_file_buffer
2241 */
2242 clear_buffer_dirty(bh);
2243 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
2244 may_free = 0;
2245 } else {
2246 JBUFFER_TRACE(jh, "on running transaction");
2247 __jbd2_journal_unfile_buffer(jh);
2248 jbd2_journal_put_journal_head(jh);
2249 }
2250 return may_free;
2251}
2252
2253/*
2254 * jbd2_journal_invalidate_folio
2255 *
2256 * This code is tricky. It has a number of cases to deal with.
2257 *
2258 * There are two invariants which this code relies on:
2259 *
2260 * i_size must be updated on disk before we start calling invalidate_folio
2261 * on the data.
2262 *
2263 * This is done in ext3 by defining an ext3_setattr method which
2264 * updates i_size before truncate gets going. By maintaining this
2265 * invariant, we can be sure that it is safe to throw away any buffers
2266 * attached to the current transaction: once the transaction commits,
2267 * we know that the data will not be needed.
2268 *
2269 * Note however that we can *not* throw away data belonging to the
2270 * previous, committing transaction!
2271 *
2272 * Any disk blocks which *are* part of the previous, committing
2273 * transaction (and which therefore cannot be discarded immediately) are
2274 * not going to be reused in the new running transaction
2275 *
2276 * The bitmap committed_data images guarantee this: any block which is
2277 * allocated in one transaction and removed in the next will be marked
2278 * as in-use in the committed_data bitmap, so cannot be reused until
2279 * the next transaction to delete the block commits. This means that
2280 * leaving committing buffers dirty is quite safe: the disk blocks
2281 * cannot be reallocated to a different file and so buffer aliasing is
2282 * not possible.
2283 *
2284 *
2285 * The above applies mainly to ordered data mode. In writeback mode we
2286 * don't make guarantees about the order in which data hits disk --- in
2287 * particular we don't guarantee that new dirty data is flushed before
2288 * transaction commit --- so it is always safe just to discard data
2289 * immediately in that mode. --sct
2290 */
2291
2292/*
2293 * The journal_unmap_buffer helper function returns zero if the buffer
2294 * concerned remains pinned as an anonymous buffer belonging to an older
2295 * transaction.
2296 *
2297 * We're outside-transaction here. Either or both of j_running_transaction
2298 * and j_committing_transaction may be NULL.
2299 */
2300static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2301 int partial_page)
2302{
2303 transaction_t *transaction;
2304 struct journal_head *jh;
2305 int may_free = 1;
2306
2307 BUFFER_TRACE(bh, "entry");
2308
2309 /*
2310 * It is safe to proceed here without the j_list_lock because the
2311 * buffers cannot be stolen by try_to_free_buffers as long as we are
2312 * holding the page lock. --sct
2313 */
2314
2315 jh = jbd2_journal_grab_journal_head(bh);
2316 if (!jh)
2317 goto zap_buffer_unlocked;
2318
2319 /* OK, we have data buffer in journaled mode */
2320 write_lock(&journal->j_state_lock);
2321 spin_lock(&jh->b_state_lock);
2322 spin_lock(&journal->j_list_lock);
2323
2324 /*
2325 * We cannot remove the buffer from checkpoint lists until the
2326 * transaction adding inode to orphan list (let's call it T)
2327 * is committed. Otherwise if the transaction changing the
2328 * buffer would be cleaned from the journal before T is
2329 * committed, a crash will cause that the correct contents of
2330 * the buffer will be lost. On the other hand we have to
2331 * clear the buffer dirty bit at latest at the moment when the
2332 * transaction marking the buffer as freed in the filesystem
2333 * structures is committed because from that moment on the
2334 * block can be reallocated and used by a different page.
2335 * Since the block hasn't been freed yet but the inode has
2336 * already been added to orphan list, it is safe for us to add
2337 * the buffer to BJ_Forget list of the newest transaction.
2338 *
2339 * Also we have to clear buffer_mapped flag of a truncated buffer
2340 * because the buffer_head may be attached to the page straddling
2341 * i_size (can happen only when blocksize < pagesize) and thus the
2342 * buffer_head can be reused when the file is extended again. So we end
2343 * up keeping around invalidated buffers attached to transactions'
2344 * BJ_Forget list just to stop checkpointing code from cleaning up
2345 * the transaction this buffer was modified in.
2346 */
2347 transaction = jh->b_transaction;
2348 if (transaction == NULL) {
2349 /* First case: not on any transaction. If it
2350 * has no checkpoint link, then we can zap it:
2351 * it's a writeback-mode buffer so we don't care
2352 * if it hits disk safely. */
2353 if (!jh->b_cp_transaction) {
2354 JBUFFER_TRACE(jh, "not on any transaction: zap");
2355 goto zap_buffer;
2356 }
2357
2358 if (!buffer_dirty(bh)) {
2359 /* bdflush has written it. We can drop it now */
2360 __jbd2_journal_remove_checkpoint(jh);
2361 goto zap_buffer;
2362 }
2363
2364 /* OK, it must be in the journal but still not
2365 * written fully to disk: it's metadata or
2366 * journaled data... */
2367
2368 if (journal->j_running_transaction) {
2369 /* ... and once the current transaction has
2370 * committed, the buffer won't be needed any
2371 * longer. */
2372 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
2373 may_free = __dispose_buffer(jh,
2374 journal->j_running_transaction);
2375 goto zap_buffer;
2376 } else {
2377 /* There is no currently-running transaction. So the
2378 * orphan record which we wrote for this file must have
2379 * passed into commit. We must attach this buffer to
2380 * the committing transaction, if it exists. */
2381 if (journal->j_committing_transaction) {
2382 JBUFFER_TRACE(jh, "give to committing trans");
2383 may_free = __dispose_buffer(jh,
2384 journal->j_committing_transaction);
2385 goto zap_buffer;
2386 } else {
2387 /* The orphan record's transaction has
2388 * committed. We can cleanse this buffer */
2389 clear_buffer_jbddirty(bh);
2390 __jbd2_journal_remove_checkpoint(jh);
2391 goto zap_buffer;
2392 }
2393 }
2394 } else if (transaction == journal->j_committing_transaction) {
2395 JBUFFER_TRACE(jh, "on committing transaction");
2396 /*
2397 * The buffer is committing, we simply cannot touch
2398 * it. If the page is straddling i_size we have to wait
2399 * for commit and try again.
2400 */
2401 if (partial_page) {
2402 spin_unlock(&journal->j_list_lock);
2403 spin_unlock(&jh->b_state_lock);
2404 write_unlock(&journal->j_state_lock);
2405 jbd2_journal_put_journal_head(jh);
2406 /* Already zapped buffer? Nothing to do... */
2407 if (!bh->b_bdev)
2408 return 0;
2409 return -EBUSY;
2410 }
2411 /*
2412 * OK, buffer won't be reachable after truncate. We just clear
2413 * b_modified to not confuse transaction credit accounting, and
2414 * set j_next_transaction to the running transaction (if there
2415 * is one) and mark buffer as freed so that commit code knows
2416 * it should clear dirty bits when it is done with the buffer.
2417 */
2418 set_buffer_freed(bh);
2419 if (journal->j_running_transaction && buffer_jbddirty(bh))
2420 jh->b_next_transaction = journal->j_running_transaction;
2421 jh->b_modified = 0;
2422 spin_unlock(&journal->j_list_lock);
2423 spin_unlock(&jh->b_state_lock);
2424 write_unlock(&journal->j_state_lock);
2425 jbd2_journal_put_journal_head(jh);
2426 return 0;
2427 } else {
2428 /* Good, the buffer belongs to the running transaction.
2429 * We are writing our own transaction's data, not any
2430 * previous one's, so it is safe to throw it away
2431 * (remember that we expect the filesystem to have set
2432 * i_size already for this truncate so recovery will not
2433 * expose the disk blocks we are discarding here.) */
2434 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
2435 JBUFFER_TRACE(jh, "on running transaction");
2436 may_free = __dispose_buffer(jh, transaction);
2437 }
2438
2439zap_buffer:
2440 /*
2441 * This is tricky. Although the buffer is truncated, it may be reused
2442 * if blocksize < pagesize and it is attached to the page straddling
2443 * EOF. Since the buffer might have been added to BJ_Forget list of the
2444 * running transaction, journal_get_write_access() won't clear
2445 * b_modified and credit accounting gets confused. So clear b_modified
2446 * here.
2447 */
2448 jh->b_modified = 0;
2449 spin_unlock(&journal->j_list_lock);
2450 spin_unlock(&jh->b_state_lock);
2451 write_unlock(&journal->j_state_lock);
2452 jbd2_journal_put_journal_head(jh);
2453zap_buffer_unlocked:
2454 clear_buffer_dirty(bh);
2455 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2456 clear_buffer_mapped(bh);
2457 clear_buffer_req(bh);
2458 clear_buffer_new(bh);
2459 clear_buffer_delay(bh);
2460 clear_buffer_unwritten(bh);
2461 bh->b_bdev = NULL;
2462 return may_free;
2463}
2464
2465/**
2466 * jbd2_journal_invalidate_folio()
2467 * @journal: journal to use for flush...
2468 * @folio: folio to flush
2469 * @offset: start of the range to invalidate
2470 * @length: length of the range to invalidate
2471 *
2472 * Reap page buffers containing data after in the specified range in page.
2473 * Can return -EBUSY if buffers are part of the committing transaction and
2474 * the page is straddling i_size. Caller then has to wait for current commit
2475 * and try again.
2476 */
2477int jbd2_journal_invalidate_folio(journal_t *journal, struct folio *folio,
2478 size_t offset, size_t length)
2479{
2480 struct buffer_head *head, *bh, *next;
2481 unsigned int stop = offset + length;
2482 unsigned int curr_off = 0;
2483 int partial_page = (offset || length < folio_size(folio));
2484 int may_free = 1;
2485 int ret = 0;
2486
2487 if (!folio_test_locked(folio))
2488 BUG();
2489 head = folio_buffers(folio);
2490 if (!head)
2491 return 0;
2492
2493 BUG_ON(stop > folio_size(folio) || stop < length);
2494
2495 /* We will potentially be playing with lists other than just the
2496 * data lists (especially for journaled data mode), so be
2497 * cautious in our locking. */
2498
2499 bh = head;
2500 do {
2501 unsigned int next_off = curr_off + bh->b_size;
2502 next = bh->b_this_page;
2503
2504 if (next_off > stop)
2505 return 0;
2506
2507 if (offset <= curr_off) {
2508 /* This block is wholly outside the truncation point */
2509 lock_buffer(bh);
2510 ret = journal_unmap_buffer(journal, bh, partial_page);
2511 unlock_buffer(bh);
2512 if (ret < 0)
2513 return ret;
2514 may_free &= ret;
2515 }
2516 curr_off = next_off;
2517 bh = next;
2518
2519 } while (bh != head);
2520
2521 if (!partial_page) {
2522 if (may_free && try_to_free_buffers(folio))
2523 J_ASSERT(!folio_buffers(folio));
2524 }
2525 return 0;
2526}
2527
2528/*
2529 * File a buffer on the given transaction list.
2530 */
2531void __jbd2_journal_file_buffer(struct journal_head *jh,
2532 transaction_t *transaction, int jlist)
2533{
2534 struct journal_head **list = NULL;
2535 int was_dirty = 0;
2536 struct buffer_head *bh = jh2bh(jh);
2537
2538 lockdep_assert_held(&jh->b_state_lock);
2539 assert_spin_locked(&transaction->t_journal->j_list_lock);
2540
2541 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2542 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
2543 jh->b_transaction == NULL);
2544
2545 if (jh->b_transaction && jh->b_jlist == jlist)
2546 return;
2547
2548 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2549 jlist == BJ_Shadow || jlist == BJ_Forget) {
2550 /*
2551 * For metadata buffers, we track dirty bit in buffer_jbddirty
2552 * instead of buffer_dirty. We should not see a dirty bit set
2553 * here because we clear it in do_get_write_access but e.g.
2554 * tune2fs can modify the sb and set the dirty bit at any time
2555 * so we try to gracefully handle that.
2556 */
2557 if (buffer_dirty(bh))
2558 warn_dirty_buffer(bh);
2559 if (test_clear_buffer_dirty(bh) ||
2560 test_clear_buffer_jbddirty(bh))
2561 was_dirty = 1;
2562 }
2563
2564 if (jh->b_transaction)
2565 __jbd2_journal_temp_unlink_buffer(jh);
2566 else
2567 jbd2_journal_grab_journal_head(bh);
2568 jh->b_transaction = transaction;
2569
2570 switch (jlist) {
2571 case BJ_None:
2572 J_ASSERT_JH(jh, !jh->b_committed_data);
2573 J_ASSERT_JH(jh, !jh->b_frozen_data);
2574 return;
2575 case BJ_Metadata:
2576 transaction->t_nr_buffers++;
2577 list = &transaction->t_buffers;
2578 break;
2579 case BJ_Forget:
2580 list = &transaction->t_forget;
2581 break;
2582 case BJ_Shadow:
2583 list = &transaction->t_shadow_list;
2584 break;
2585 case BJ_Reserved:
2586 list = &transaction->t_reserved_list;
2587 break;
2588 }
2589
2590 __blist_add_buffer(list, jh);
2591 jh->b_jlist = jlist;
2592
2593 if (was_dirty)
2594 set_buffer_jbddirty(bh);
2595}
2596
2597void jbd2_journal_file_buffer(struct journal_head *jh,
2598 transaction_t *transaction, int jlist)
2599{
2600 spin_lock(&jh->b_state_lock);
2601 spin_lock(&transaction->t_journal->j_list_lock);
2602 __jbd2_journal_file_buffer(jh, transaction, jlist);
2603 spin_unlock(&transaction->t_journal->j_list_lock);
2604 spin_unlock(&jh->b_state_lock);
2605}
2606
2607/*
2608 * Remove a buffer from its current buffer list in preparation for
2609 * dropping it from its current transaction entirely. If the buffer has
2610 * already started to be used by a subsequent transaction, refile the
2611 * buffer on that transaction's metadata list.
2612 *
2613 * Called under j_list_lock
2614 * Called under jh->b_state_lock
2615 *
2616 * When this function returns true, there's no next transaction to refile to
2617 * and the caller has to drop jh reference through
2618 * jbd2_journal_put_journal_head().
2619 */
2620bool __jbd2_journal_refile_buffer(struct journal_head *jh)
2621{
2622 int was_dirty, jlist;
2623 struct buffer_head *bh = jh2bh(jh);
2624
2625 lockdep_assert_held(&jh->b_state_lock);
2626 if (jh->b_transaction)
2627 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2628
2629 /* If the buffer is now unused, just drop it. */
2630 if (jh->b_next_transaction == NULL) {
2631 __jbd2_journal_unfile_buffer(jh);
2632 return true;
2633 }
2634
2635 /*
2636 * It has been modified by a later transaction: add it to the new
2637 * transaction's metadata list.
2638 */
2639
2640 was_dirty = test_clear_buffer_jbddirty(bh);
2641 __jbd2_journal_temp_unlink_buffer(jh);
2642
2643 /*
2644 * b_transaction must be set, otherwise the new b_transaction won't
2645 * be holding jh reference
2646 */
2647 J_ASSERT_JH(jh, jh->b_transaction != NULL);
2648
2649 /*
2650 * We set b_transaction here because b_next_transaction will inherit
2651 * our jh reference and thus __jbd2_journal_file_buffer() must not
2652 * take a new one.
2653 */
2654 WRITE_ONCE(jh->b_transaction, jh->b_next_transaction);
2655 WRITE_ONCE(jh->b_next_transaction, NULL);
2656 if (buffer_freed(bh))
2657 jlist = BJ_Forget;
2658 else if (jh->b_modified)
2659 jlist = BJ_Metadata;
2660 else
2661 jlist = BJ_Reserved;
2662 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
2663 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2664
2665 if (was_dirty)
2666 set_buffer_jbddirty(bh);
2667 return false;
2668}
2669
2670/*
2671 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2672 * bh reference so that we can safely unlock bh.
2673 *
2674 * The jh and bh may be freed by this call.
2675 */
2676void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
2677{
2678 bool drop;
2679
2680 spin_lock(&jh->b_state_lock);
2681 spin_lock(&journal->j_list_lock);
2682 drop = __jbd2_journal_refile_buffer(jh);
2683 spin_unlock(&jh->b_state_lock);
2684 spin_unlock(&journal->j_list_lock);
2685 if (drop)
2686 jbd2_journal_put_journal_head(jh);
2687}
2688
2689/*
2690 * File inode in the inode list of the handle's transaction
2691 */
2692static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
2693 unsigned long flags, loff_t start_byte, loff_t end_byte)
2694{
2695 transaction_t *transaction = handle->h_transaction;
2696 journal_t *journal;
2697 pgoff_t start_page, end_page;
2698 int err = 0;
2699 int abort_transaction = 0;
2700
2701 if (is_handle_aborted(handle))
2702 return -EROFS;
2703 journal = transaction->t_journal;
2704
2705 jbd2_debug(4, "Adding inode %llu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2706 transaction->t_tid);
2707
2708 start_page = (pgoff_t)(start_byte >> PAGE_SHIFT);
2709 end_page = (pgoff_t)(end_byte >> PAGE_SHIFT) + 1;
2710
2711 spin_lock(&journal->j_list_lock);
2712 WRITE_ONCE(jinode->i_flags, jinode->i_flags | flags);
2713
2714 if (jinode->i_dirty_start_page != jinode->i_dirty_end_page) {
2715 WRITE_ONCE(jinode->i_dirty_start_page,
2716 min(jinode->i_dirty_start_page, start_page));
2717 WRITE_ONCE(jinode->i_dirty_end_page,
2718 max(jinode->i_dirty_end_page, end_page));
2719 } else {
2720 /* Publish a new non-empty range by making end visible first. */
2721 WRITE_ONCE(jinode->i_dirty_end_page, end_page);
2722 WRITE_ONCE(jinode->i_dirty_start_page, start_page);
2723 }
2724
2725 /* Is inode already attached where we need it? */
2726 if (jinode->i_transaction == transaction ||
2727 jinode->i_next_transaction == transaction)
2728 goto done;
2729
2730 /*
2731 * We only ever set this variable to 1 so the test is safe. Since
2732 * t_need_data_flush is likely to be set, we do the test to save some
2733 * cacheline bouncing
2734 */
2735 if (!transaction->t_need_data_flush)
2736 transaction->t_need_data_flush = 1;
2737 /* On some different transaction's list - should be
2738 * the committing one */
2739 if (jinode->i_transaction) {
2740 if (WARN_ON_ONCE(jinode->i_next_transaction ||
2741 jinode->i_transaction !=
2742 journal->j_committing_transaction)) {
2743 pr_err("JBD2: %s: assertion failure: i_next_transaction=%p i_transaction=%p j_committing_transaction=%p\n",
2744 journal->j_devname, jinode->i_next_transaction,
2745 jinode->i_transaction,
2746 journal->j_committing_transaction);
2747 err = -EINVAL;
2748 abort_transaction = 1;
2749 goto done;
2750 }
2751 jinode->i_next_transaction = transaction;
2752 goto done;
2753 }
2754 /* Not on any transaction list... */
2755 if (WARN_ON_ONCE(jinode->i_next_transaction)) {
2756 err = -EINVAL;
2757 abort_transaction = 1;
2758 goto done;
2759 }
2760 jinode->i_transaction = transaction;
2761 list_add(&jinode->i_list, &transaction->t_inode_list);
2762done:
2763 spin_unlock(&journal->j_list_lock);
2764 if (abort_transaction)
2765 jbd2_journal_abort(journal, err);
2766 return err;
2767}
2768
2769int jbd2_journal_inode_ranged_write(handle_t *handle,
2770 struct jbd2_inode *jinode, loff_t start_byte, loff_t length)
2771{
2772 return jbd2_journal_file_inode(handle, jinode,
2773 JI_WRITE_DATA | JI_WAIT_DATA, start_byte,
2774 start_byte + length - 1);
2775}
2776
2777int jbd2_journal_inode_ranged_wait(handle_t *handle, struct jbd2_inode *jinode,
2778 loff_t start_byte, loff_t length)
2779{
2780 return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA,
2781 start_byte, start_byte + length - 1);
2782}
2783
2784/*
2785 * File truncate and transaction commit interact with each other in a
2786 * non-trivial way. If a transaction writing data block A is
2787 * committing, we cannot discard the data by truncate until we have
2788 * written them. Otherwise if we crashed after the transaction with
2789 * write has committed but before the transaction with truncate has
2790 * committed, we could see stale data in block A. This function is a
2791 * helper to solve this problem. It starts writeout of the truncated
2792 * part in case it is in the committing transaction.
2793 *
2794 * Filesystem code must call this function when inode is journaled in
2795 * ordered mode before truncation happens and after the inode has been
2796 * placed on orphan list with the new inode size. The second condition
2797 * avoids the race that someone writes new data and we start
2798 * committing the transaction after this function has been called but
2799 * before a transaction for truncate is started (and furthermore it
2800 * allows us to optimize the case where the addition to orphan list
2801 * happens in the same transaction as write --- we don't have to write
2802 * any data in such case).
2803 */
2804int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2805 struct jbd2_inode *jinode,
2806 loff_t new_size)
2807{
2808 transaction_t *inode_trans, *commit_trans;
2809 int ret = 0;
2810
2811 /* This is a quick check to avoid locking if not necessary */
2812 if (!READ_ONCE(jinode->i_transaction))
2813 goto out;
2814 /* Locks are here just to force reading of recent values, it is
2815 * enough that the transaction was not committing before we started
2816 * a transaction adding the inode to orphan list */
2817 read_lock(&journal->j_state_lock);
2818 commit_trans = journal->j_committing_transaction;
2819 read_unlock(&journal->j_state_lock);
2820 spin_lock(&journal->j_list_lock);
2821 inode_trans = jinode->i_transaction;
2822 spin_unlock(&journal->j_list_lock);
2823 if (inode_trans == commit_trans) {
2824 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
2825 new_size, LLONG_MAX);
2826 if (ret)
2827 jbd2_journal_abort(journal, ret);
2828 }
2829out:
2830 return ret;
2831}