blob: 6a3caedd22856feb650a084bfe0ef08bec838284 [file] [log] [blame]
Theodore Ts'of5166762017-12-17 22:00:59 -05001// SPDX-License-Identifier: GPL-2.0+
Dave Kleikamp470decc2006-10-11 01:20:57 -07002/*
Uwe Kleine-König58862692007-05-09 07:51:49 +02003 * linux/fs/jbd2/transaction.c
Dave Kleikamp470decc2006-10-11 01:20:57 -07004 *
5 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
6 *
7 * Copyright 1998 Red Hat corp --- All Rights Reserved
8 *
Dave Kleikamp470decc2006-10-11 01:20:57 -07009 * 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>
Mingming Caof7f4bcc2006-10-11 01:20:59 -070019#include <linux/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070020#include <linux/errno.h>
21#include <linux/slab.h>
22#include <linux/timer.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070023#include <linux/mm.h>
24#include <linux/highmem.h>
Josef Bacike07f7182008-11-26 01:14:26 -050025#include <linux/hrtimer.h>
Theodore Ts'o47def822010-07-27 11:56:05 -040026#include <linux/backing-dev.h>
Randy Dunlap44705752011-10-27 04:05:13 -040027#include <linux/bug.h>
Theodore Ts'o47def822010-07-27 11:56:05 -040028#include <linux/module.h>
Michal Hocko81378da2017-05-03 14:53:22 -070029#include <linux/sched/mm.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070030
Theodore Ts'o343d9c22013-02-08 13:00:22 -050031#include <trace/events/jbd2.h>
32
Adrian Bunk7ddae862006-12-06 20:38:27 -080033static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
Jan Karade1b7942011-06-13 15:38:22 -040034static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
Adrian Bunk7ddae862006-12-06 20:38:27 -080035
Yongqiang Yang0c2022e2012-02-20 17:53:02 -050036static 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);
Chengguang Xu0d521542019-05-10 21:15:47 -040045 if (!transaction_cache) {
46 pr_emerg("JBD2: failed to create transaction cache\n");
47 return -ENOMEM;
48 }
49 return 0;
Yongqiang Yang0c2022e2012-02-20 17:53:02 -050050}
51
52void jbd2_journal_destroy_transaction_cache(void)
53{
Wang Long8bdd5b62018-05-20 22:38:26 -040054 kmem_cache_destroy(transaction_cache);
55 transaction_cache = NULL;
Yongqiang Yang0c2022e2012-02-20 17:53:02 -050056}
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
Dave Kleikamp470decc2006-10-11 01:20:57 -070065/*
Jan Kara19014d62019-11-05 17:44:31 +010066 * Base amount of descriptor blocks we reserve for each transaction.
Jan Kara9f356e52019-11-05 17:44:24 +010067 */
Jan Kara9f356e52019-11-05 17:44:24 +010068static int jbd2_descriptor_blocks_per_trans(journal_t *journal)
69{
Jan Kara19014d62019-11-05 17:44:31 +010070 int tag_space = journal->j_blocksize - sizeof(journal_header_t);
71 int tags_per_block;
72
73 /* Subtract UUID */
74 tag_space -= 16;
75 if (jbd2_journal_has_csum_v2or3(journal))
76 tag_space -= sizeof(struct jbd2_journal_block_tail);
77 /* Commit code leaves a slack space of 16 bytes at the end of block */
78 tags_per_block = (tag_space - 16) / journal_tag_bytes(journal);
79 /*
80 * Revoke descriptors are accounted separately so we need to reserve
81 * space for commit block and normal transaction descriptor blocks.
82 */
83 return 1 + DIV_ROUND_UP(journal->j_max_transaction_buffers,
84 tags_per_block);
Jan Kara9f356e52019-11-05 17:44:24 +010085}
86
87/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -070088 * jbd2_get_transaction: obtain a new transaction_t object.
Dave Kleikamp470decc2006-10-11 01:20:57 -070089 *
Liu Song0df6f462019-03-01 00:36:57 -050090 * Simply initialise a new transaction. Initialize it in
Dave Kleikamp470decc2006-10-11 01:20:57 -070091 * RUNNING state and add it to the current journal (which should not
92 * have an existing running transaction: we only make a new transaction
93 * once we have started to commit the old one).
94 *
95 * Preconditions:
96 * The journal MUST be locked. We don't perform atomic mallocs on the
97 * new transaction and we can't block without protecting against other
98 * processes trying to touch the journal while it is in transition.
99 *
Dave Kleikamp470decc2006-10-11 01:20:57 -0700100 */
101
Liu Song0df6f462019-03-01 00:36:57 -0500102static void jbd2_get_transaction(journal_t *journal,
103 transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700104{
105 transaction->t_journal = journal;
106 transaction->t_state = T_RUNNING;
Josef Bacike07f7182008-11-26 01:14:26 -0500107 transaction->t_start_time = ktime_get();
Dave Kleikamp470decc2006-10-11 01:20:57 -0700108 transaction->t_tid = journal->j_transaction_sequence++;
109 transaction->t_expires = jiffies + journal->j_commit_interval;
110 spin_lock_init(&transaction->t_handle_lock);
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400111 atomic_set(&transaction->t_updates, 0);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400112 atomic_set(&transaction->t_outstanding_credits,
Jan Kara9f356e52019-11-05 17:44:24 +0100113 jbd2_descriptor_blocks_per_trans(journal) +
Jan Kara8f7d89f2013-06-04 12:35:11 -0400114 atomic_read(&journal->j_reserved_credits));
Jan Karafdc3ef82019-11-05 17:44:26 +0100115 atomic_set(&transaction->t_outstanding_revokes, 0);
Theodore Ts'o8dd42042010-08-03 21:38:29 -0400116 atomic_set(&transaction->t_handle_count, 0);
Jan Karac851ed52008-07-11 19:27:31 -0400117 INIT_LIST_HEAD(&transaction->t_inode_list);
Theodore Ts'o3e624fc2008-10-16 20:00:24 -0400118 INIT_LIST_HEAD(&transaction->t_private_list);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700119
120 /* Set up the commit timer for the new transaction. */
Andreas Dilgerb1f485f2009-08-10 22:51:53 -0400121 journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700122 add_timer(&journal->j_commit_timer);
123
124 J_ASSERT(journal->j_running_transaction == NULL);
125 journal->j_running_transaction = transaction;
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500126 transaction->t_max_wait = 0;
127 transaction->t_start = jiffies;
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500128 transaction->t_requested = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700129}
130
131/*
132 * Handle management.
133 *
134 * A handle_t is an object which represents a single atomic update to a
135 * filesystem, and which tracks all of the modifications which form part
136 * of that one update.
137 */
138
139/*
Tao Ma28e35e422011-05-22 21:45:26 -0400140 * Update transaction's maximum wait time, if debugging is enabled.
Theodore Ts'o6d0bf002010-08-09 17:28:38 -0400141 *
142 * In order for t_max_wait to be reliable, it must be protected by a
143 * lock. But doing so will mean that start_this_handle() can not be
144 * run in parallel on SMP systems, which limits our scalability. So
145 * unless debugging is enabled, we no longer update t_max_wait, which
146 * means that maximum wait time reported by the jbd2_run_stats
147 * tracepoint will always be zero.
148 */
Tao Ma28e35e422011-05-22 21:45:26 -0400149static inline void update_t_max_wait(transaction_t *transaction,
150 unsigned long ts)
Theodore Ts'o6d0bf002010-08-09 17:28:38 -0400151{
152#ifdef CONFIG_JBD2_DEBUG
Theodore Ts'o6d0bf002010-08-09 17:28:38 -0400153 if (jbd2_journal_enable_debug &&
154 time_after(transaction->t_start, ts)) {
155 ts = jbd2_time_diff(ts, transaction->t_start);
156 spin_lock(&transaction->t_handle_lock);
157 if (ts > transaction->t_max_wait)
158 transaction->t_max_wait = ts;
159 spin_unlock(&transaction->t_handle_lock);
160 }
161#endif
162}
163
164/*
Jan Kara96f1e092018-12-03 23:16:07 -0500165 * Wait until running transaction passes to T_FLUSH state and new transaction
166 * can thus be started. Also starts the commit if needed. The function expects
167 * running transaction to exist and releases j_state_lock.
Jan Kara8f7d89f2013-06-04 12:35:11 -0400168 */
169static void wait_transaction_locked(journal_t *journal)
170 __releases(journal->j_state_lock)
171{
172 DEFINE_WAIT(wait);
173 int need_to_start;
174 tid_t tid = journal->j_running_transaction->t_tid;
175
176 prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
177 TASK_UNINTERRUPTIBLE);
178 need_to_start = !tid_geq(journal->j_commit_request, tid);
179 read_unlock(&journal->j_state_lock);
180 if (need_to_start)
181 jbd2_log_start_commit(journal, tid);
Jan Karae03a9972016-09-22 11:44:06 -0400182 jbd2_might_wait_for_commit(journal);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400183 schedule();
184 finish_wait(&journal->j_wait_transaction_locked, &wait);
185}
186
Jan Kara96f1e092018-12-03 23:16:07 -0500187/*
188 * Wait until running transaction transitions from T_SWITCH to T_FLUSH
189 * state and new transaction can thus be started. The function releases
190 * j_state_lock.
191 */
192static void wait_transaction_switching(journal_t *journal)
193 __releases(journal->j_state_lock)
194{
195 DEFINE_WAIT(wait);
196
197 if (WARN_ON(!journal->j_running_transaction ||
Theodore Ts'o05d52332020-11-07 00:00:49 -0500198 journal->j_running_transaction->t_state != T_SWITCH)) {
199 read_unlock(&journal->j_state_lock);
Jan Kara96f1e092018-12-03 23:16:07 -0500200 return;
Theodore Ts'o05d52332020-11-07 00:00:49 -0500201 }
Jan Kara96f1e092018-12-03 23:16:07 -0500202 prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
203 TASK_UNINTERRUPTIBLE);
204 read_unlock(&journal->j_state_lock);
205 /*
206 * We don't call jbd2_might_wait_for_commit() here as there's no
207 * waiting for outstanding handles happening anymore in T_SWITCH state
208 * and handling of reserved handles actually relies on that for
209 * correctness.
210 */
211 schedule();
212 finish_wait(&journal->j_wait_transaction_locked, &wait);
213}
214
Jan Kara8f7d89f2013-06-04 12:35:11 -0400215static void sub_reserved_credits(journal_t *journal, int blocks)
216{
217 atomic_sub(blocks, &journal->j_reserved_credits);
218 wake_up(&journal->j_wait_reserved);
219}
220
221/*
222 * Wait until we can add credits for handle to the running transaction. Called
223 * with j_state_lock held for reading. Returns 0 if handle joined the running
224 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
225 * caller must retry.
Theodore Ts'ob33d9f52021-08-14 10:54:09 -0400226 *
227 * Note: because j_state_lock may be dropped depending on the return
228 * value, we need to fake out sparse so ti doesn't complain about a
229 * locking imbalance. Callers of add_transaction_credits will need to
230 * make a similar accomodation.
Jan Kara8f7d89f2013-06-04 12:35:11 -0400231 */
232static int add_transaction_credits(journal_t *journal, int blocks,
233 int rsv_blocks)
Theodore Ts'ob33d9f52021-08-14 10:54:09 -0400234__must_hold(&journal->j_state_lock)
Jan Kara8f7d89f2013-06-04 12:35:11 -0400235{
236 transaction_t *t = journal->j_running_transaction;
237 int needed;
238 int total = blocks + rsv_blocks;
239
240 /*
241 * If the current transaction is locked down for commit, wait
242 * for the lock to be released.
243 */
Jan Kara96f1e092018-12-03 23:16:07 -0500244 if (t->t_state != T_RUNNING) {
245 WARN_ON_ONCE(t->t_state >= T_FLUSH);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400246 wait_transaction_locked(journal);
Theodore Ts'ob33d9f52021-08-14 10:54:09 -0400247 __acquire(&journal->j_state_lock); /* fake out sparse */
Jan Kara8f7d89f2013-06-04 12:35:11 -0400248 return 1;
249 }
250
251 /*
252 * If there is not enough space left in the log to write all
253 * potential buffers requested by this operation, we need to
254 * stall pending a log checkpoint to free some more log space.
255 */
256 needed = atomic_add_return(total, &t->t_outstanding_credits);
257 if (needed > journal->j_max_transaction_buffers) {
258 /*
259 * If the current transaction is already too large,
260 * then start to commit it: we can then go back and
261 * attach this handle to a new transaction.
262 */
263 atomic_sub(total, &t->t_outstanding_credits);
Lukas Czerner6d3ec142015-08-04 11:21:52 -0400264
265 /*
266 * Is the number of reserved credits in the current transaction too
267 * big to fit this handle? Wait until reserved credits are freed.
268 */
269 if (atomic_read(&journal->j_reserved_credits) + total >
270 journal->j_max_transaction_buffers) {
271 read_unlock(&journal->j_state_lock);
Jan Karae03a9972016-09-22 11:44:06 -0400272 jbd2_might_wait_for_commit(journal);
Lukas Czerner6d3ec142015-08-04 11:21:52 -0400273 wait_event(journal->j_wait_reserved,
274 atomic_read(&journal->j_reserved_credits) + total <=
275 journal->j_max_transaction_buffers);
Theodore Ts'ob33d9f52021-08-14 10:54:09 -0400276 __acquire(&journal->j_state_lock); /* fake out sparse */
Lukas Czerner6d3ec142015-08-04 11:21:52 -0400277 return 1;
278 }
279
Jan Kara8f7d89f2013-06-04 12:35:11 -0400280 wait_transaction_locked(journal);
Theodore Ts'ob33d9f52021-08-14 10:54:09 -0400281 __acquire(&journal->j_state_lock); /* fake out sparse */
Jan Kara8f7d89f2013-06-04 12:35:11 -0400282 return 1;
283 }
284
285 /*
286 * The commit code assumes that it can get enough log space
287 * without forcing a checkpoint. This is *critical* for
288 * correctness: a checkpoint of a buffer which is also
289 * associated with a committing transaction creates a deadlock,
290 * so commit simply cannot force through checkpoints.
291 *
292 * We must therefore ensure the necessary space in the journal
293 * *before* starting to dirty potentially checkpointed buffers
294 * in the new transaction.
295 */
Jan Kara77444ac2019-11-05 17:44:25 +0100296 if (jbd2_log_space_left(journal) < journal->j_max_transaction_buffers) {
Jan Kara8f7d89f2013-06-04 12:35:11 -0400297 atomic_sub(total, &t->t_outstanding_credits);
298 read_unlock(&journal->j_state_lock);
Jan Karae03a9972016-09-22 11:44:06 -0400299 jbd2_might_wait_for_commit(journal);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400300 write_lock(&journal->j_state_lock);
Jan Kara77444ac2019-11-05 17:44:25 +0100301 if (jbd2_log_space_left(journal) <
302 journal->j_max_transaction_buffers)
Jan Kara8f7d89f2013-06-04 12:35:11 -0400303 __jbd2_log_wait_for_space(journal);
304 write_unlock(&journal->j_state_lock);
Theodore Ts'ob33d9f52021-08-14 10:54:09 -0400305 __acquire(&journal->j_state_lock); /* fake out sparse */
Jan Kara8f7d89f2013-06-04 12:35:11 -0400306 return 1;
307 }
308
309 /* No reservation? We are done... */
310 if (!rsv_blocks)
311 return 0;
312
313 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
314 /* We allow at most half of a transaction to be reserved */
315 if (needed > journal->j_max_transaction_buffers / 2) {
316 sub_reserved_credits(journal, rsv_blocks);
317 atomic_sub(total, &t->t_outstanding_credits);
318 read_unlock(&journal->j_state_lock);
Jan Karae03a9972016-09-22 11:44:06 -0400319 jbd2_might_wait_for_commit(journal);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400320 wait_event(journal->j_wait_reserved,
321 atomic_read(&journal->j_reserved_credits) + rsv_blocks
322 <= journal->j_max_transaction_buffers / 2);
Theodore Ts'ob33d9f52021-08-14 10:54:09 -0400323 __acquire(&journal->j_state_lock); /* fake out sparse */
Jan Kara8f7d89f2013-06-04 12:35:11 -0400324 return 1;
325 }
326 return 0;
327}
328
329/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700330 * start_this_handle: Given a handle, deal with any locking or stalling
331 * needed to make sure that there is enough journal space for the handle
332 * to begin. Attach the handle to a transaction and set up the
333 * transaction's buffer credits.
334 */
335
Theodore Ts'o47def822010-07-27 11:56:05 -0400336static int start_this_handle(journal_t *journal, handle_t *handle,
Dan Carpenterd2159fb2011-09-04 10:20:14 -0400337 gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700338{
Theodore Ts'oe4471832011-02-12 08:18:24 -0500339 transaction_t *transaction, *new_transaction = NULL;
Jan Kara933f1c12019-11-05 17:44:27 +0100340 int blocks = handle->h_total_credits;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400341 int rsv_blocks = 0;
Tao Ma28e35e422011-05-22 21:45:26 -0400342 unsigned long ts = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700343
Jan Kara8f7d89f2013-06-04 12:35:11 -0400344 if (handle->h_rsv_handle)
Jan Kara933f1c12019-11-05 17:44:27 +0100345 rsv_blocks = handle->h_rsv_handle->h_total_credits;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400346
Lukas Czerner6d3ec142015-08-04 11:21:52 -0400347 /*
348 * Limit the number of reserved credits to 1/2 of maximum transaction
349 * size and limit the number of total credits to not exceed maximum
350 * transaction size per operation.
351 */
352 if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
353 (rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
354 printk(KERN_ERR "JBD2: %s wants too many credits "
355 "credits:%d rsv_credits:%d max:%d\n",
356 current->comm, blocks, rsv_blocks,
357 journal->j_max_transaction_buffers);
358 WARN_ON(1);
359 return -ENOSPC;
360 }
361
Dave Kleikamp470decc2006-10-11 01:20:57 -0700362alloc_transaction:
Jan Kara3b1833e92021-04-06 18:17:59 +0200363 /*
364 * This check is racy but it is just an optimization of allocating new
365 * transaction early if there are high chances we'll need it. If we
366 * guess wrong, we'll retry or free unused transaction.
367 */
368 if (!data_race(journal->j_running_transaction)) {
Michal Hocko6ccaf3e2015-06-08 10:53:10 -0400369 /*
370 * If __GFP_FS is not present, then we may be being called from
371 * inside the fs writeback layer, so we MUST NOT fail.
372 */
373 if ((gfp_mask & __GFP_FS) == 0)
374 gfp_mask |= __GFP_NOFAIL;
Wanlong Gaob2f4edb2012-06-01 00:10:32 -0400375 new_transaction = kmem_cache_zalloc(transaction_cache,
376 gfp_mask);
Michal Hocko6ccaf3e2015-06-08 10:53:10 -0400377 if (!new_transaction)
Theodore Ts'o47def822010-07-27 11:56:05 -0400378 return -ENOMEM;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700379 }
380
381 jbd_debug(3, "New handle %p going live.\n", handle);
382
Dave Kleikamp470decc2006-10-11 01:20:57 -0700383 /*
384 * We need to hold j_state_lock until t_updates has been incremented,
385 * for proper journal barrier handling
386 */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400387repeat:
388 read_lock(&journal->j_state_lock);
Theodore Ts'o5c2178e2010-10-27 21:30:04 -0400389 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700390 if (is_journal_aborted(journal) ||
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700391 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400392 read_unlock(&journal->j_state_lock);
Yongqiang Yang0c2022e2012-02-20 17:53:02 -0500393 jbd2_journal_free_transaction(new_transaction);
Theodore Ts'o47def822010-07-27 11:56:05 -0400394 return -EROFS;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700395 }
396
Jan Kara8f7d89f2013-06-04 12:35:11 -0400397 /*
398 * Wait on the journal's transaction barrier if necessary. Specifically
399 * we allow reserved handles to proceed because otherwise commit could
400 * deadlock on page writeback not being able to complete.
401 */
402 if (!handle->h_reserved && journal->j_barrier_count) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400403 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700404 wait_event(journal->j_wait_transaction_locked,
405 journal->j_barrier_count == 0);
406 goto repeat;
407 }
408
409 if (!journal->j_running_transaction) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400410 read_unlock(&journal->j_state_lock);
411 if (!new_transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700412 goto alloc_transaction;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400413 write_lock(&journal->j_state_lock);
Jan Karad7961c72012-12-21 00:15:51 -0500414 if (!journal->j_running_transaction &&
Jan Kara8f7d89f2013-06-04 12:35:11 -0400415 (handle->h_reserved || !journal->j_barrier_count)) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400416 jbd2_get_transaction(journal, new_transaction);
417 new_transaction = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700418 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400419 write_unlock(&journal->j_state_lock);
420 goto repeat;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700421 }
422
423 transaction = journal->j_running_transaction;
424
Jan Kara8f7d89f2013-06-04 12:35:11 -0400425 if (!handle->h_reserved) {
426 /* We may have dropped j_state_lock - restart in that case */
Theodore Ts'ob33d9f52021-08-14 10:54:09 -0400427 if (add_transaction_credits(journal, blocks, rsv_blocks)) {
428 /*
429 * add_transaction_credits releases
430 * j_state_lock on a non-zero return
431 */
432 __release(&journal->j_state_lock);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400433 goto repeat;
Theodore Ts'ob33d9f52021-08-14 10:54:09 -0400434 }
Jan Kara8f7d89f2013-06-04 12:35:11 -0400435 } else {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700436 /*
Jan Kara8f7d89f2013-06-04 12:35:11 -0400437 * We have handle reserved so we are allowed to join T_LOCKED
438 * transaction and we don't have to check for transaction size
Jan Kara96f1e092018-12-03 23:16:07 -0500439 * and journal space. But we still have to wait while running
440 * transaction is being switched to a committing one as it
441 * won't wait for any handles anymore.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700442 */
Jan Kara96f1e092018-12-03 23:16:07 -0500443 if (transaction->t_state == T_SWITCH) {
444 wait_transaction_switching(journal);
445 goto repeat;
446 }
Jan Kara8f7d89f2013-06-04 12:35:11 -0400447 sub_reserved_credits(journal, blocks);
448 handle->h_reserved = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700449 }
450
451 /* OK, account for the buffers that this operation expects to
Theodore Ts'o8dd42042010-08-03 21:38:29 -0400452 * use and add the handle to the running transaction.
Theodore Ts'o8dd42042010-08-03 21:38:29 -0400453 */
Tao Ma28e35e422011-05-22 21:45:26 -0400454 update_t_max_wait(transaction, ts);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700455 handle->h_transaction = transaction;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400456 handle->h_requested_credits = blocks;
Jan Karafdc3ef82019-11-05 17:44:26 +0100457 handle->h_revoke_credits_requested = handle->h_revoke_credits;
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500458 handle->h_start_jiffies = jiffies;
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400459 atomic_inc(&transaction->t_updates);
Theodore Ts'o8dd42042010-08-03 21:38:29 -0400460 atomic_inc(&transaction->t_handle_count);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400461 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
462 handle, blocks,
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400463 atomic_read(&transaction->t_outstanding_credits),
Jan Kara76c39902013-06-04 12:12:57 -0400464 jbd2_log_space_left(journal));
Theodore Ts'oa931da62010-08-03 21:35:12 -0400465 read_unlock(&journal->j_state_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400466 current->journal_info = handle;
Jan Kara9599b0e2009-08-17 21:23:17 -0400467
Jan Karaab714af2016-06-30 11:39:38 -0400468 rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_);
Yongqiang Yang0c2022e2012-02-20 17:53:02 -0500469 jbd2_journal_free_transaction(new_transaction);
Michal Hocko81378da2017-05-03 14:53:22 -0700470 /*
471 * Ensure that no allocations done while the transaction is open are
472 * going to recurse back to the fs layer.
473 */
474 handle->saved_alloc_context = memalloc_nofs_save();
Theodore Ts'o47def822010-07-27 11:56:05 -0400475 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700476}
477
478/* Allocate a new handle. This should probably be in a slab... */
479static handle_t *new_handle(int nblocks)
480{
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400481 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700482 if (!handle)
483 return NULL;
Jan Kara933f1c12019-11-05 17:44:27 +0100484 handle->h_total_credits = nblocks;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700485 handle->h_ref = 1;
486
487 return handle;
488}
489
Jan Kara8f7d89f2013-06-04 12:35:11 -0400490handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
Jan Karafdc3ef82019-11-05 17:44:26 +0100491 int revoke_records, gfp_t gfp_mask,
492 unsigned int type, unsigned int line_no)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700493{
494 handle_t *handle = journal_current_handle();
495 int err;
496
497 if (!journal)
498 return ERR_PTR(-EROFS);
499
500 if (handle) {
501 J_ASSERT(handle->h_transaction->t_journal == journal);
502 handle->h_ref++;
503 return handle;
504 }
505
Jan Karafdc3ef82019-11-05 17:44:26 +0100506 nblocks += DIV_ROUND_UP(revoke_records,
507 journal->j_revoke_records_per_block);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700508 handle = new_handle(nblocks);
509 if (!handle)
510 return ERR_PTR(-ENOMEM);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400511 if (rsv_blocks) {
512 handle_t *rsv_handle;
513
514 rsv_handle = new_handle(rsv_blocks);
515 if (!rsv_handle) {
516 jbd2_free_handle(handle);
517 return ERR_PTR(-ENOMEM);
518 }
519 rsv_handle->h_reserved = 1;
520 rsv_handle->h_journal = journal;
521 handle->h_rsv_handle = rsv_handle;
522 }
Jan Karafdc3ef82019-11-05 17:44:26 +0100523 handle->h_revoke_credits = revoke_records;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700524
Theodore Ts'o47def822010-07-27 11:56:05 -0400525 err = start_this_handle(journal, handle, gfp_mask);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700526 if (err < 0) {
Jan Kara8f7d89f2013-06-04 12:35:11 -0400527 if (handle->h_rsv_handle)
528 jbd2_free_handle(handle->h_rsv_handle);
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400529 jbd2_free_handle(handle);
Dmitry Monakhovdf05c1b82013-03-02 17:08:46 -0500530 return ERR_PTR(err);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700531 }
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500532 handle->h_type = type;
533 handle->h_line_no = line_no;
534 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
535 handle->h_transaction->t_tid, type,
536 line_no, nblocks);
Michal Hocko81378da2017-05-03 14:53:22 -0700537
Dave Kleikamp470decc2006-10-11 01:20:57 -0700538 return handle;
539}
Theodore Ts'o47def822010-07-27 11:56:05 -0400540EXPORT_SYMBOL(jbd2__journal_start);
541
542
Mauro Carvalho Chehab91e47752017-05-12 07:25:21 -0300543/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +0100544 * jbd2_journal_start() - Obtain a new handle.
Mauro Carvalho Chehab91e47752017-05-12 07:25:21 -0300545 * @journal: Journal to start transaction on.
546 * @nblocks: number of block buffer we might modify
547 *
548 * We make sure that the transaction can guarantee at least nblocks of
549 * modified buffers in the log. We block until the log can guarantee
550 * that much space. Additionally, if rsv_blocks > 0, we also create another
551 * handle with rsv_blocks reserved blocks in the journal. This handle is
wangyan0c1cba62020-01-22 17:33:10 +0800552 * stored in h_rsv_handle. It is not attached to any particular transaction
Mauro Carvalho Chehab91e47752017-05-12 07:25:21 -0300553 * and thus doesn't block transaction commit. If the caller uses this reserved
554 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
555 * on the parent handle will dispose the reserved one. Reserved handle has to
556 * be converted to a normal handle using jbd2_journal_start_reserved() before
557 * it can be used.
558 *
559 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
560 * on failure.
561 */
Theodore Ts'o47def822010-07-27 11:56:05 -0400562handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
563{
Jan Karafdc3ef82019-11-05 17:44:26 +0100564 return jbd2__journal_start(journal, nblocks, 0, 0, GFP_NOFS, 0, 0);
Theodore Ts'o47def822010-07-27 11:56:05 -0400565}
566EXPORT_SYMBOL(jbd2_journal_start);
567
Jan Kara14ff6282020-05-20 15:31:19 +0200568static void __jbd2_journal_unreserve_handle(handle_t *handle, transaction_t *t)
Jan Kara8f7d89f2013-06-04 12:35:11 -0400569{
570 journal_t *journal = handle->h_journal;
571
572 WARN_ON(!handle->h_reserved);
Jan Kara933f1c12019-11-05 17:44:27 +0100573 sub_reserved_credits(journal, handle->h_total_credits);
Jan Kara14ff6282020-05-20 15:31:19 +0200574 if (t)
575 atomic_sub(handle->h_total_credits, &t->t_outstanding_credits);
Jan Karaec8b6f62019-11-05 17:44:23 +0100576}
577
578void jbd2_journal_free_reserved(handle_t *handle)
579{
Jan Kara14ff6282020-05-20 15:31:19 +0200580 journal_t *journal = handle->h_journal;
581
582 /* Get j_state_lock to pin running transaction if it exists */
583 read_lock(&journal->j_state_lock);
584 __jbd2_journal_unreserve_handle(handle, journal->j_running_transaction);
585 read_unlock(&journal->j_state_lock);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400586 jbd2_free_handle(handle);
587}
588EXPORT_SYMBOL(jbd2_journal_free_reserved);
589
590/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +0100591 * jbd2_journal_start_reserved() - start reserved handle
Jan Kara8f7d89f2013-06-04 12:35:11 -0400592 * @handle: handle to start
Tobin C. Hardingf69120c2018-01-10 00:27:29 -0500593 * @type: for handle statistics
594 * @line_no: for handle statistics
Jan Kara8f7d89f2013-06-04 12:35:11 -0400595 *
596 * Start handle that has been previously reserved with jbd2_journal_reserve().
597 * This attaches @handle to the running transaction (or creates one if there's
598 * not transaction running). Unlike jbd2_journal_start() this function cannot
599 * block on journal commit, checkpointing, or similar stuff. It can block on
600 * memory allocation or frozen journal though.
601 *
602 * Return 0 on success, non-zero on error - handle is freed in that case.
603 */
604int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
605 unsigned int line_no)
606{
607 journal_t *journal = handle->h_journal;
608 int ret = -EIO;
609
610 if (WARN_ON(!handle->h_reserved)) {
611 /* Someone passed in normal handle? Just stop it. */
612 jbd2_journal_stop(handle);
613 return ret;
614 }
615 /*
616 * Usefulness of mixing of reserved and unreserved handles is
617 * questionable. So far nobody seems to need it so just error out.
618 */
619 if (WARN_ON(current->journal_info)) {
620 jbd2_journal_free_reserved(handle);
621 return ret;
622 }
623
624 handle->h_journal = NULL;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400625 /*
626 * GFP_NOFS is here because callers are likely from writeback or
627 * similarly constrained call sites
628 */
629 ret = start_this_handle(journal, handle, GFP_NOFS);
Dan Carpenter92e3b402014-02-17 20:33:01 -0500630 if (ret < 0) {
Theodore Ts'ob2569262018-04-18 11:49:31 -0400631 handle->h_journal = journal;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400632 jbd2_journal_free_reserved(handle);
Dan Carpenter92e3b402014-02-17 20:33:01 -0500633 return ret;
634 }
Jan Kara8f7d89f2013-06-04 12:35:11 -0400635 handle->h_type = type;
636 handle->h_line_no = line_no;
Xiaoguang Wang4c273352019-08-24 23:10:17 -0400637 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
638 handle->h_transaction->t_tid, type,
Jan Kara933f1c12019-11-05 17:44:27 +0100639 line_no, handle->h_total_credits);
Dan Carpenter92e3b402014-02-17 20:33:01 -0500640 return 0;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400641}
642EXPORT_SYMBOL(jbd2_journal_start_reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700643
644/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +0100645 * jbd2_journal_extend() - extend buffer credits.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700646 * @handle: handle to 'extend'
647 * @nblocks: nr blocks to try to extend by.
Jan Karafdc3ef82019-11-05 17:44:26 +0100648 * @revoke_records: number of revoke records to try to extend by.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700649 *
650 * Some transactions, such as large extends and truncates, can be done
651 * atomically all at once or in several stages. The operation requests
Masanari Iidabd7ced92016-02-02 22:31:06 +0900652 * a credit for a number of buffer modifications in advance, but can
Dave Kleikamp470decc2006-10-11 01:20:57 -0700653 * extend its credit if it needs more.
654 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700655 * jbd2_journal_extend tries to give the running handle more buffer credits.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700656 * It does not guarantee that allocation - this is a best-effort only.
657 * The calling process MUST be able to deal cleanly with a failure to
658 * extend here.
659 *
660 * Return 0 on success, non-zero on failure.
661 *
662 * return code < 0 implies an error
663 * return code > 0 implies normal transaction-full status.
664 */
Jan Karafdc3ef82019-11-05 17:44:26 +0100665int jbd2_journal_extend(handle_t *handle, int nblocks, int revoke_records)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700666{
667 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400668 journal_t *journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700669 int result;
670 int wanted;
671
Dave Kleikamp470decc2006-10-11 01:20:57 -0700672 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400673 return -EROFS;
674 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700675
676 result = 1;
677
Theodore Ts'oa931da62010-08-03 21:35:12 -0400678 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700679
680 /* Don't extend a locked-down transaction! */
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400681 if (transaction->t_state != T_RUNNING) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700682 jbd_debug(3, "denied handle %p %d blocks: "
683 "transaction not running\n", handle, nblocks);
684 goto error_out;
685 }
686
Jan Karafdc3ef82019-11-05 17:44:26 +0100687 nblocks += DIV_ROUND_UP(
688 handle->h_revoke_credits_requested + revoke_records,
689 journal->j_revoke_records_per_block) -
690 DIV_ROUND_UP(
691 handle->h_revoke_credits_requested,
692 journal->j_revoke_records_per_block);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700693 spin_lock(&transaction->t_handle_lock);
Jan Karafe1e8db52013-06-04 12:22:15 -0400694 wanted = atomic_add_return(nblocks,
695 &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700696
697 if (wanted > journal->j_max_transaction_buffers) {
698 jbd_debug(3, "denied handle %p %d blocks: "
699 "transaction too large\n", handle, nblocks);
Jan Karafe1e8db52013-06-04 12:22:15 -0400700 atomic_sub(nblocks, &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700701 goto unlock;
702 }
703
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500704 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400705 transaction->t_tid,
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500706 handle->h_type, handle->h_line_no,
Jan Kara933f1c12019-11-05 17:44:27 +0100707 handle->h_total_credits,
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500708 nblocks);
709
Jan Kara933f1c12019-11-05 17:44:27 +0100710 handle->h_total_credits += nblocks;
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500711 handle->h_requested_credits += nblocks;
Jan Karafdc3ef82019-11-05 17:44:26 +0100712 handle->h_revoke_credits += revoke_records;
713 handle->h_revoke_credits_requested += revoke_records;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700714 result = 0;
715
716 jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
717unlock:
718 spin_unlock(&transaction->t_handle_lock);
719error_out:
Theodore Ts'oa931da62010-08-03 21:35:12 -0400720 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700721 return result;
722}
723
Jan Karaec8b6f62019-11-05 17:44:23 +0100724static void stop_this_handle(handle_t *handle)
725{
726 transaction_t *transaction = handle->h_transaction;
727 journal_t *journal = transaction->t_journal;
Jan Karafdc3ef82019-11-05 17:44:26 +0100728 int revokes;
Jan Karaec8b6f62019-11-05 17:44:23 +0100729
730 J_ASSERT(journal_current_handle() == handle);
731 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
732 current->journal_info = NULL;
Jan Karafdc3ef82019-11-05 17:44:26 +0100733 /*
734 * Subtract necessary revoke descriptor blocks from handle credits. We
735 * take care to account only for revoke descriptor blocks the
736 * transaction will really need as large sequences of transactions with
737 * small numbers of revokes are relatively common.
738 */
739 revokes = handle->h_revoke_credits_requested - handle->h_revoke_credits;
740 if (revokes) {
741 int t_revokes, revoke_descriptors;
742 int rr_per_blk = journal->j_revoke_records_per_block;
743
744 WARN_ON_ONCE(DIV_ROUND_UP(revokes, rr_per_blk)
Jan Kara933f1c12019-11-05 17:44:27 +0100745 > handle->h_total_credits);
Jan Karafdc3ef82019-11-05 17:44:26 +0100746 t_revokes = atomic_add_return(revokes,
747 &transaction->t_outstanding_revokes);
748 revoke_descriptors =
749 DIV_ROUND_UP(t_revokes, rr_per_blk) -
750 DIV_ROUND_UP(t_revokes - revokes, rr_per_blk);
Jan Kara933f1c12019-11-05 17:44:27 +0100751 handle->h_total_credits -= revoke_descriptors;
Jan Karafdc3ef82019-11-05 17:44:26 +0100752 }
Jan Kara933f1c12019-11-05 17:44:27 +0100753 atomic_sub(handle->h_total_credits,
Jan Karaec8b6f62019-11-05 17:44:23 +0100754 &transaction->t_outstanding_credits);
755 if (handle->h_rsv_handle)
Jan Kara14ff6282020-05-20 15:31:19 +0200756 __jbd2_journal_unreserve_handle(handle->h_rsv_handle,
757 transaction);
Jan Karaec8b6f62019-11-05 17:44:23 +0100758 if (atomic_dec_and_test(&transaction->t_updates))
759 wake_up(&journal->j_wait_updates);
760
Linus Torvalds50b8b3f2019-11-30 10:53:02 -0800761 rwsem_release(&journal->j_trans_commit_map, _THIS_IP_);
Jan Karaec8b6f62019-11-05 17:44:23 +0100762 /*
763 * Scope of the GFP_NOFS context is over here and so we can restore the
764 * original alloc context.
765 */
766 memalloc_nofs_restore(handle->saved_alloc_context);
767}
Dave Kleikamp470decc2006-10-11 01:20:57 -0700768
769/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +0100770 * jbd2__journal_restart() - restart a handle .
Dave Kleikamp470decc2006-10-11 01:20:57 -0700771 * @handle: handle to restart
772 * @nblocks: nr credits requested
Jan Karafdc3ef82019-11-05 17:44:26 +0100773 * @revoke_records: number of revoke record credits requested
Tobin C. Hardingf69120c2018-01-10 00:27:29 -0500774 * @gfp_mask: memory allocation flags (for start_this_handle)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700775 *
776 * Restart a handle for a multi-transaction filesystem
777 * operation.
778 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700779 * If the jbd2_journal_extend() call above fails to grant new buffer credits
780 * to a running handle, a call to jbd2_journal_restart will commit the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700781 * handle's transaction so far and reattach the handle to a new
Masanari Iidabd7ced92016-02-02 22:31:06 +0900782 * transaction capable of guaranteeing the requested number of
Jan Kara8f7d89f2013-06-04 12:35:11 -0400783 * credits. We preserve reserved handle if there's any attached to the
784 * passed in handle.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700785 */
Jan Karafdc3ef82019-11-05 17:44:26 +0100786int jbd2__journal_restart(handle_t *handle, int nblocks, int revoke_records,
787 gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700788{
789 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400790 journal_t *journal;
Theodore Ts'oe4471832011-02-12 08:18:24 -0500791 tid_t tid;
Jan Karaec8b6f62019-11-05 17:44:23 +0100792 int need_to_start;
Jan Kara0094f982019-11-05 17:44:30 +0100793 int ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700794
795 /* If we've had an abort of any type, don't even think about
796 * actually doing the restart! */
797 if (is_handle_aborted(handle))
798 return 0;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400799 journal = transaction->t_journal;
Jan Karaec8b6f62019-11-05 17:44:23 +0100800 tid = transaction->t_tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700801
802 /*
803 * First unlink the handle from its current transaction, and start the
804 * commit on that.
805 */
Dave Kleikamp470decc2006-10-11 01:20:57 -0700806 jbd_debug(2, "restarting handle %p\n", handle);
Jan Karaec8b6f62019-11-05 17:44:23 +0100807 stop_this_handle(handle);
808 handle->h_transaction = NULL;
809
810 /*
811 * TODO: If we use READ_ONCE / WRITE_ONCE for j_commit_request we can
812 * get rid of pointless j_state_lock traffic like this.
813 */
814 read_lock(&journal->j_state_lock);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500815 need_to_start = !tid_geq(journal->j_commit_request, tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400816 read_unlock(&journal->j_state_lock);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500817 if (need_to_start)
818 jbd2_log_start_commit(journal, tid);
Jan Kara933f1c12019-11-05 17:44:27 +0100819 handle->h_total_credits = nblocks +
Jan Karafdc3ef82019-11-05 17:44:26 +0100820 DIV_ROUND_UP(revoke_records,
821 journal->j_revoke_records_per_block);
822 handle->h_revoke_credits = revoke_records;
Theodore Ts'o47def822010-07-27 11:56:05 -0400823 ret = start_this_handle(journal, handle, gfp_mask);
Jan Kara0094f982019-11-05 17:44:30 +0100824 trace_jbd2_handle_restart(journal->j_fs_dev->bd_dev,
825 ret ? 0 : handle->h_transaction->t_tid,
826 handle->h_type, handle->h_line_no,
827 handle->h_total_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700828 return ret;
829}
Theodore Ts'o47def822010-07-27 11:56:05 -0400830EXPORT_SYMBOL(jbd2__journal_restart);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700831
832
Theodore Ts'o47def822010-07-27 11:56:05 -0400833int jbd2_journal_restart(handle_t *handle, int nblocks)
834{
Jan Karafdc3ef82019-11-05 17:44:26 +0100835 return jbd2__journal_restart(handle, nblocks, 0, GFP_NOFS);
Theodore Ts'o47def822010-07-27 11:56:05 -0400836}
837EXPORT_SYMBOL(jbd2_journal_restart);
838
Dave Kleikamp470decc2006-10-11 01:20:57 -0700839/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +0100840 * jbd2_journal_lock_updates () - establish a transaction barrier.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700841 * @journal: Journal to establish a barrier on.
842 *
843 * This locks out any further updates from being started, and blocks
844 * until all existing updates have completed, returning only once the
845 * journal is in a quiescent state with no updates running.
846 *
847 * The journal lock should not be held on entry.
848 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700849void jbd2_journal_lock_updates(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700850{
851 DEFINE_WAIT(wait);
852
Jan Kara1eaa5662016-06-30 11:40:54 -0400853 jbd2_might_wait_for_commit(journal);
854
Theodore Ts'oa931da62010-08-03 21:35:12 -0400855 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700856 ++journal->j_barrier_count;
857
Jan Kara8f7d89f2013-06-04 12:35:11 -0400858 /* Wait until there are no reserved handles */
859 if (atomic_read(&journal->j_reserved_credits)) {
860 write_unlock(&journal->j_state_lock);
861 wait_event(journal->j_wait_reserved,
862 atomic_read(&journal->j_reserved_credits) == 0);
863 write_lock(&journal->j_state_lock);
864 }
865
Dave Kleikamp470decc2006-10-11 01:20:57 -0700866 /* Wait until there are no running updates */
867 while (1) {
868 transaction_t *transaction = journal->j_running_transaction;
869
870 if (!transaction)
871 break;
872
873 spin_lock(&transaction->t_handle_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700874 prepare_to_wait(&journal->j_wait_updates, &wait,
875 TASK_UNINTERRUPTIBLE);
Jan Kara9837d8e2012-01-04 22:03:11 -0500876 if (!atomic_read(&transaction->t_updates)) {
877 spin_unlock(&transaction->t_handle_lock);
878 finish_wait(&journal->j_wait_updates, &wait);
879 break;
880 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700881 spin_unlock(&transaction->t_handle_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400882 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700883 schedule();
884 finish_wait(&journal->j_wait_updates, &wait);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400885 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700886 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400887 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700888
889 /*
890 * We have now established a barrier against other normal updates, but
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700891 * we also need to barrier against other jbd2_journal_lock_updates() calls
Dave Kleikamp470decc2006-10-11 01:20:57 -0700892 * to make sure that we serialise special journal-locked operations
893 * too.
894 */
895 mutex_lock(&journal->j_barrier);
896}
897
898/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +0100899 * jbd2_journal_unlock_updates () - release barrier
Dave Kleikamp470decc2006-10-11 01:20:57 -0700900 * @journal: Journal to release the barrier on.
901 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700902 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
Dave Kleikamp470decc2006-10-11 01:20:57 -0700903 *
904 * Should be called without the journal lock held.
905 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700906void jbd2_journal_unlock_updates (journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700907{
908 J_ASSERT(journal->j_barrier_count != 0);
909
910 mutex_unlock(&journal->j_barrier);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400911 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700912 --journal->j_barrier_count;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400913 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700914 wake_up(&journal->j_wait_transaction_locked);
915}
916
Jan Karaf91d1d02009-07-13 16:16:20 -0400917static void warn_dirty_buffer(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700918{
Jan Karaf91d1d02009-07-13 16:16:20 -0400919 printk(KERN_WARNING
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400920 "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
Jan Karaf91d1d02009-07-13 16:16:20 -0400921 "There's a risk of filesystem corruption in case of system "
922 "crash.\n",
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400923 bh->b_bdev, (unsigned long long)bh->b_blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700924}
925
Jan Karaee57aba2015-06-08 12:39:07 -0400926/* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
927static void jbd2_freeze_jh_data(struct journal_head *jh)
928{
929 struct page *page;
930 int offset;
931 char *source;
932 struct buffer_head *bh = jh2bh(jh);
933
934 J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
935 page = bh->b_page;
936 offset = offset_in_page(bh->b_data);
937 source = kmap_atomic(page);
938 /* Fire data frozen trigger just before we copy the data */
939 jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
940 memcpy(jh->b_frozen_data, source + offset, bh->b_size);
941 kunmap_atomic(source);
942
943 /*
944 * Now that the frozen data is saved off, we need to store any matching
945 * triggers.
946 */
947 jh->b_frozen_triggers = jh->b_triggers;
948}
949
Dave Kleikamp470decc2006-10-11 01:20:57 -0700950/*
951 * If the buffer is already part of the current transaction, then there
952 * is nothing we need to do. If it is already part of a prior
953 * transaction which we are still committing to disk, then we need to
954 * make sure that we do not overwrite the old copy: we do copy-out to
955 * preserve the copy going to disk. We also account the buffer against
956 * the handle's metadata buffer credits (unless the buffer is already
957 * part of the transaction, that is).
958 *
959 */
960static int
961do_get_write_access(handle_t *handle, struct journal_head *jh,
962 int force_copy)
963{
964 struct buffer_head *bh;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400965 transaction_t *transaction = handle->h_transaction;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700966 journal_t *journal;
967 int error;
968 char *frozen_buffer = NULL;
Theodore Ts'of783f092013-04-21 16:47:54 -0400969 unsigned long start_lock, time_lock;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700970
Dave Kleikamp470decc2006-10-11 01:20:57 -0700971 journal = transaction->t_journal;
972
Theodore Ts'ocfef2c62010-12-18 13:07:34 -0500973 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700974
975 JBUFFER_TRACE(jh, "entry");
976repeat:
977 bh = jh2bh(jh);
978
979 /* @@@ Need to check for errors here at some point. */
980
Theodore Ts'of783f092013-04-21 16:47:54 -0400981 start_lock = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700982 lock_buffer(bh);
Thomas Gleixner46417062019-08-09 14:42:32 +0200983 spin_lock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700984
Theodore Ts'of783f092013-04-21 16:47:54 -0400985 /* If it takes too long to lock the buffer, trace it */
986 time_lock = jbd2_time_diff(start_lock, jiffies);
987 if (time_lock > HZ/10)
988 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
989 jiffies_to_msecs(time_lock));
990
Dave Kleikamp470decc2006-10-11 01:20:57 -0700991 /* We now hold the buffer lock so it is safe to query the buffer
992 * state. Is the buffer dirty?
993 *
994 * If so, there are two possibilities. The buffer may be
995 * non-journaled, and undergoing a quite legitimate writeback.
996 * Otherwise, it is journaled, and we don't expect dirty buffers
997 * in that state (the buffers should be marked JBD_Dirty
998 * instead.) So either the IO is being done under our own
999 * control and this is a bug, or it's a third party IO such as
1000 * dump(8) (which may leave the buffer scheduled for read ---
1001 * ie. locked but not dirty) or tune2fs (which may actually have
1002 * the buffer dirtied, ugh.) */
1003
1004 if (buffer_dirty(bh)) {
1005 /*
1006 * First question: is this buffer already part of the current
1007 * transaction or the existing committing transaction?
1008 */
1009 if (jh->b_transaction) {
1010 J_ASSERT_JH(jh,
1011 jh->b_transaction == transaction ||
1012 jh->b_transaction ==
1013 journal->j_committing_transaction);
1014 if (jh->b_next_transaction)
1015 J_ASSERT_JH(jh, jh->b_next_transaction ==
1016 transaction);
Jan Karaf91d1d02009-07-13 16:16:20 -04001017 warn_dirty_buffer(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001018 }
1019 /*
1020 * In any case we need to clean the dirty flag and we must
1021 * do it under the buffer lock to be sure we don't race
1022 * with running write-out.
1023 */
Jan Karaf91d1d02009-07-13 16:16:20 -04001024 JBUFFER_TRACE(jh, "Journalling dirty buffer");
1025 clear_buffer_dirty(bh);
1026 set_buffer_jbddirty(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001027 }
1028
1029 unlock_buffer(bh);
1030
1031 error = -EROFS;
1032 if (is_handle_aborted(handle)) {
Thomas Gleixner46417062019-08-09 14:42:32 +02001033 spin_unlock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001034 goto out;
1035 }
1036 error = 0;
1037
1038 /*
1039 * The buffer is already part of this transaction if b_transaction or
1040 * b_next_transaction points to it
1041 */
1042 if (jh->b_transaction == transaction ||
1043 jh->b_next_transaction == transaction)
1044 goto done;
1045
1046 /*
Josef Bacik9fc7c632008-04-17 10:38:59 -04001047 * this is the first time this transaction is touching this buffer,
1048 * reset the modified flag
1049 */
Colin Ian King561405f2018-12-04 00:20:10 -05001050 jh->b_modified = 0;
Josef Bacik9fc7c632008-04-17 10:38:59 -04001051
1052 /*
Jan Kara8b00f402015-06-08 12:44:21 -04001053 * If the buffer is not journaled right now, we need to make sure it
1054 * doesn't get written to disk before the caller actually commits the
1055 * new data
1056 */
1057 if (!jh->b_transaction) {
1058 JBUFFER_TRACE(jh, "no transaction");
1059 J_ASSERT_JH(jh, !jh->b_next_transaction);
1060 JBUFFER_TRACE(jh, "file as BJ_Reserved");
Jan Karade92c8c2015-06-08 12:46:37 -04001061 /*
1062 * Make sure all stores to jh (b_modified, b_frozen_data) are
1063 * visible before attaching it to the running transaction.
1064 * Paired with barrier in jbd2_write_access_granted()
1065 */
1066 smp_wmb();
Jan Kara8b00f402015-06-08 12:44:21 -04001067 spin_lock(&journal->j_list_lock);
1068 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
1069 spin_unlock(&journal->j_list_lock);
1070 goto done;
1071 }
1072 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -07001073 * If there is already a copy-out version of this buffer, then we don't
1074 * need to make another one
1075 */
1076 if (jh->b_frozen_data) {
1077 JBUFFER_TRACE(jh, "has frozen data");
1078 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
Jan Karade92c8c2015-06-08 12:46:37 -04001079 goto attach_next;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001080 }
1081
Jan Kara8b00f402015-06-08 12:44:21 -04001082 JBUFFER_TRACE(jh, "owned by older transaction");
1083 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1084 J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001085
1086 /*
Jan Kara8b00f402015-06-08 12:44:21 -04001087 * There is one case we have to be very careful about. If the
1088 * committing transaction is currently writing this buffer out to disk
1089 * and has NOT made a copy-out, then we cannot modify the buffer
1090 * contents at all right now. The essence of copy-out is that it is
1091 * the extra copy, not the primary copy, which gets journaled. If the
1092 * primary copy is already going to disk then we cannot do copy-out
1093 * here.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001094 */
Jan Kara8b00f402015-06-08 12:44:21 -04001095 if (buffer_shadow(bh)) {
1096 JBUFFER_TRACE(jh, "on shadow: sleep");
Thomas Gleixner46417062019-08-09 14:42:32 +02001097 spin_unlock(&jh->b_state_lock);
Jan Kara8b00f402015-06-08 12:44:21 -04001098 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
1099 goto repeat;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001100 }
1101
Jan Kara8b00f402015-06-08 12:44:21 -04001102 /*
1103 * Only do the copy if the currently-owning transaction still needs it.
1104 * If buffer isn't on BJ_Metadata list, the committing transaction is
1105 * past that stage (here we use the fact that BH_Shadow is set under
1106 * bh_state lock together with refiling to BJ_Shadow list and at this
1107 * point we know the buffer doesn't have BH_Shadow set).
1108 *
1109 * Subtle point, though: if this is a get_undo_access, then we will be
1110 * relying on the frozen_data to contain the new value of the
1111 * committed_data record after the transaction, so we HAVE to force the
1112 * frozen_data copy in that case.
1113 */
1114 if (jh->b_jlist == BJ_Metadata || force_copy) {
1115 JBUFFER_TRACE(jh, "generate frozen data");
1116 if (!frozen_buffer) {
1117 JBUFFER_TRACE(jh, "allocate memory for buffer");
Thomas Gleixner46417062019-08-09 14:42:32 +02001118 spin_unlock(&jh->b_state_lock);
Michal Hocko490c1b42016-03-13 17:38:20 -04001119 frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
1120 GFP_NOFS | __GFP_NOFAIL);
Jan Kara8b00f402015-06-08 12:44:21 -04001121 goto repeat;
1122 }
1123 jh->b_frozen_data = frozen_buffer;
1124 frozen_buffer = NULL;
1125 jbd2_freeze_jh_data(jh);
1126 }
Jan Karade92c8c2015-06-08 12:46:37 -04001127attach_next:
1128 /*
1129 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
1130 * before attaching it to the running transaction. Paired with barrier
1131 * in jbd2_write_access_granted()
1132 */
1133 smp_wmb();
Jan Kara8b00f402015-06-08 12:44:21 -04001134 jh->b_next_transaction = transaction;
1135
Dave Kleikamp470decc2006-10-11 01:20:57 -07001136done:
Thomas Gleixner46417062019-08-09 14:42:32 +02001137 spin_unlock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001138
1139 /*
1140 * If we are about to journal a buffer, then any revoke pending on it is
1141 * no longer valid
1142 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001143 jbd2_journal_cancel_revoke(handle, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001144
1145out:
1146 if (unlikely(frozen_buffer)) /* It's usually NULL */
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001147 jbd2_free(frozen_buffer, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001148
1149 JBUFFER_TRACE(jh, "exit");
1150 return error;
1151}
1152
Jan Karade92c8c2015-06-08 12:46:37 -04001153/* Fast check whether buffer is already attached to the required transaction */
Junxiao Bi087ffd42015-12-04 12:29:28 -05001154static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1155 bool undo)
Jan Karade92c8c2015-06-08 12:46:37 -04001156{
1157 struct journal_head *jh;
1158 bool ret = false;
1159
1160 /* Dirty buffers require special handling... */
1161 if (buffer_dirty(bh))
1162 return false;
1163
1164 /*
1165 * RCU protects us from dereferencing freed pages. So the checks we do
1166 * are guaranteed not to oops. However the jh slab object can get freed
1167 * & reallocated while we work with it. So we have to be careful. When
1168 * we see jh attached to the running transaction, we know it must stay
1169 * so until the transaction is committed. Thus jh won't be freed and
1170 * will be attached to the same bh while we run. However it can
1171 * happen jh gets freed, reallocated, and attached to the transaction
1172 * just after we get pointer to it from bh. So we have to be careful
1173 * and recheck jh still belongs to our bh before we return success.
1174 */
1175 rcu_read_lock();
1176 if (!buffer_jbd(bh))
1177 goto out;
1178 /* This should be bh2jh() but that doesn't work with inline functions */
1179 jh = READ_ONCE(bh->b_private);
1180 if (!jh)
1181 goto out;
Junxiao Bi087ffd42015-12-04 12:29:28 -05001182 /* For undo access buffer must have data copied */
1183 if (undo && !jh->b_committed_data)
1184 goto out;
Qian Cai6c5d9112020-02-21 23:31:11 -05001185 if (READ_ONCE(jh->b_transaction) != handle->h_transaction &&
1186 READ_ONCE(jh->b_next_transaction) != handle->h_transaction)
Jan Karade92c8c2015-06-08 12:46:37 -04001187 goto out;
1188 /*
1189 * There are two reasons for the barrier here:
1190 * 1) Make sure to fetch b_bh after we did previous checks so that we
1191 * detect when jh went through free, realloc, attach to transaction
1192 * while we were checking. Paired with implicit barrier in that path.
1193 * 2) So that access to bh done after jbd2_write_access_granted()
1194 * doesn't get reordered and see inconsistent state of concurrent
1195 * do_get_write_access().
1196 */
1197 smp_mb();
1198 if (unlikely(jh->b_bh != bh))
1199 goto out;
1200 ret = true;
1201out:
1202 rcu_read_unlock();
1203 return ret;
1204}
1205
Dave Kleikamp470decc2006-10-11 01:20:57 -07001206/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +01001207 * jbd2_journal_get_write_access() - notify intent to modify a buffer
1208 * for metadata (not data) update.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001209 * @handle: transaction to add buffer modifications to
1210 * @bh: bh to be used for metadata writes
Dave Kleikamp470decc2006-10-11 01:20:57 -07001211 *
Mauro Carvalho Chehabdf1b5602017-05-12 07:58:23 -03001212 * Returns: error code or 0 on success.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001213 *
1214 * In full data journalling mode the buffer may be of type BJ_AsyncData,
Mauro Carvalho Chehabdf1b5602017-05-12 07:58:23 -03001215 * because we're ``write()ing`` a buffer which is also part of a shared mapping.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001216 */
1217
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001218int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001219{
Jan Karade92c8c2015-06-08 12:46:37 -04001220 struct journal_head *jh;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001221 int rc;
1222
wangyan8eedabf2020-02-20 21:46:14 +08001223 if (is_handle_aborted(handle))
1224 return -EROFS;
1225
Junxiao Bi087ffd42015-12-04 12:29:28 -05001226 if (jbd2_write_access_granted(handle, bh, false))
Jan Karade92c8c2015-06-08 12:46:37 -04001227 return 0;
1228
1229 jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001230 /* We do not want to get caught playing with fields which the
1231 * log thread also manipulates. Make sure that the buffer
1232 * completes any outstanding IO before proceeding. */
1233 rc = do_get_write_access(handle, jh, 0);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001234 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001235 return rc;
1236}
1237
1238
1239/*
1240 * When the user wants to journal a newly created buffer_head
1241 * (ie. getblk() returned a new buffer and we are going to populate it
1242 * manually rather than reading off disk), then we need to keep the
1243 * buffer_head locked until it has been completely filled with new
1244 * data. In this case, we should be able to make the assertion that
1245 * the bh is not already part of an existing transaction.
1246 *
1247 * The buffer should already be locked by the caller by this point.
1248 * There is no lock ranking violation: it was a newly created,
1249 * unlocked buffer beforehand. */
1250
1251/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +01001252 * jbd2_journal_get_create_access () - notify intent to use newly created bh
Dave Kleikamp470decc2006-10-11 01:20:57 -07001253 * @handle: transaction to new buffer to
1254 * @bh: new buffer.
1255 *
1256 * Call this if you create a new bh.
1257 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001258int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001259{
1260 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001261 journal_t *journal;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001262 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001263 int err;
1264
1265 jbd_debug(5, "journal_head %p\n", jh);
1266 err = -EROFS;
1267 if (is_handle_aborted(handle))
1268 goto out;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001269 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001270 err = 0;
1271
1272 JBUFFER_TRACE(jh, "entry");
1273 /*
1274 * The buffer may already belong to this transaction due to pre-zeroing
1275 * in the filesystem's new_block code. It may also be on the previous,
1276 * committing transaction's lists, but it HAS to be in Forget state in
1277 * that case: the transaction must have deleted the buffer for it to be
1278 * reused here.
1279 */
Thomas Gleixner46417062019-08-09 14:42:32 +02001280 spin_lock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001281 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1282 jh->b_transaction == NULL ||
1283 (jh->b_transaction == journal->j_committing_transaction &&
1284 jh->b_jlist == BJ_Forget)));
1285
1286 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1287 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1288
1289 if (jh->b_transaction == NULL) {
Jan Karaf91d1d02009-07-13 16:16:20 -04001290 /*
1291 * Previous jbd2_journal_forget() could have left the buffer
1292 * with jbddirty bit set because it was being committed. When
1293 * the commit finished, we've filed the buffer for
1294 * checkpointing and marked it dirty. Now we are reallocating
1295 * the buffer so the transaction freeing it must have
1296 * committed and so it's safe to clear the dirty bit.
1297 */
1298 clear_buffer_dirty(jh2bh(jh));
Josef Bacik9fc7c632008-04-17 10:38:59 -04001299 /* first access by this transaction */
1300 jh->b_modified = 0;
1301
Dave Kleikamp470decc2006-10-11 01:20:57 -07001302 JBUFFER_TRACE(jh, "file as BJ_Reserved");
Theodore Ts'o6e4862a2014-03-09 00:46:23 -05001303 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001304 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
Taesoo Kim559cce62016-10-12 23:19:18 -04001305 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001306 } else if (jh->b_transaction == journal->j_committing_transaction) {
Josef Bacik9fc7c632008-04-17 10:38:59 -04001307 /* first access by this transaction */
1308 jh->b_modified = 0;
1309
Dave Kleikamp470decc2006-10-11 01:20:57 -07001310 JBUFFER_TRACE(jh, "set next transaction");
Theodore Ts'o6e4862a2014-03-09 00:46:23 -05001311 spin_lock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001312 jh->b_next_transaction = transaction;
Taesoo Kim559cce62016-10-12 23:19:18 -04001313 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001314 }
Thomas Gleixner46417062019-08-09 14:42:32 +02001315 spin_unlock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001316
1317 /*
1318 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1319 * blocks which contain freed but then revoked metadata. We need
1320 * to cancel the revoke in case we end up freeing it yet again
1321 * and the reallocating as data - this would cause a second revoke,
1322 * which hits an assertion error.
1323 */
1324 JBUFFER_TRACE(jh, "cancelling revoke");
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001325 jbd2_journal_cancel_revoke(handle, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001326out:
Ding Dinghua3991b402011-05-25 17:43:48 -04001327 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001328 return err;
1329}
1330
1331/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +01001332 * jbd2_journal_get_undo_access() - Notify intent to modify metadata with
Dave Kleikamp470decc2006-10-11 01:20:57 -07001333 * non-rewindable consequences
1334 * @handle: transaction
1335 * @bh: buffer to undo
Dave Kleikamp470decc2006-10-11 01:20:57 -07001336 *
1337 * Sometimes there is a need to distinguish between metadata which has
1338 * been committed to disk and that which has not. The ext3fs code uses
1339 * this for freeing and allocating space, we have to make sure that we
1340 * do not reuse freed space until the deallocation has been committed,
1341 * since if we overwrote that space we would make the delete
1342 * un-rewindable in case of a crash.
1343 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001344 * To deal with that, jbd2_journal_get_undo_access requests write access to a
Dave Kleikamp470decc2006-10-11 01:20:57 -07001345 * buffer for parts of non-rewindable operations such as delete
1346 * operations on the bitmaps. The journaling code must keep a copy of
1347 * the buffer's contents prior to the undo_access call until such time
1348 * as we know that the buffer has definitely been committed to disk.
1349 *
1350 * We never need to know which transaction the committed data is part
1351 * of, buffers touched here are guaranteed to be dirtied later and so
1352 * will be committed to a new transaction in due course, at which point
1353 * we can discard the old committed data pointer.
1354 *
1355 * Returns error number or 0 on success.
1356 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001357int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001358{
1359 int err;
Jan Karade92c8c2015-06-08 12:46:37 -04001360 struct journal_head *jh;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001361 char *committed_data = NULL;
1362
wangyan8eedabf2020-02-20 21:46:14 +08001363 if (is_handle_aborted(handle))
1364 return -EROFS;
1365
Junxiao Bi087ffd42015-12-04 12:29:28 -05001366 if (jbd2_write_access_granted(handle, bh, true))
Jan Karade92c8c2015-06-08 12:46:37 -04001367 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001368
Jan Karade92c8c2015-06-08 12:46:37 -04001369 jh = jbd2_journal_add_journal_head(bh);
zhangyi (F)01215d32019-02-21 11:24:09 -05001370 JBUFFER_TRACE(jh, "entry");
1371
Dave Kleikamp470decc2006-10-11 01:20:57 -07001372 /*
1373 * Do this first --- it can drop the journal lock, so we want to
1374 * make sure that obtaining the committed_data is done
1375 * atomically wrt. completion of any outstanding commits.
1376 */
1377 err = do_get_write_access(handle, jh, 1);
1378 if (err)
1379 goto out;
1380
1381repeat:
Michal Hocko490c1b42016-03-13 17:38:20 -04001382 if (!jh->b_committed_data)
1383 committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1384 GFP_NOFS|__GFP_NOFAIL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001385
Thomas Gleixner46417062019-08-09 14:42:32 +02001386 spin_lock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001387 if (!jh->b_committed_data) {
1388 /* Copy out the current buffer contents into the
1389 * preserved, committed copy. */
1390 JBUFFER_TRACE(jh, "generate b_committed data");
1391 if (!committed_data) {
Thomas Gleixner46417062019-08-09 14:42:32 +02001392 spin_unlock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001393 goto repeat;
1394 }
1395
1396 jh->b_committed_data = committed_data;
1397 committed_data = NULL;
1398 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1399 }
Thomas Gleixner46417062019-08-09 14:42:32 +02001400 spin_unlock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001401out:
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001402 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001403 if (unlikely(committed_data))
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001404 jbd2_free(committed_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001405 return err;
1406}
1407
1408/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +01001409 * jbd2_journal_set_triggers() - Add triggers for commit writeout
Joel Beckere06c8222008-09-11 15:35:47 -07001410 * @bh: buffer to trigger on
1411 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1412 *
1413 * Set any triggers on this journal_head. This is always safe, because
1414 * triggers for a committing buffer will be saved off, and triggers for
1415 * a running transaction will match the buffer in that transaction.
1416 *
1417 * Call with NULL to clear the triggers.
1418 */
1419void jbd2_journal_set_triggers(struct buffer_head *bh,
1420 struct jbd2_buffer_trigger_type *type)
1421{
Jan Karaad56eda2013-03-11 13:24:56 -04001422 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
Joel Beckere06c8222008-09-11 15:35:47 -07001423
Jan Kara188c2992021-08-16 11:57:04 +02001424 if (WARN_ON_ONCE(!jh))
Jan Karaad56eda2013-03-11 13:24:56 -04001425 return;
Joel Beckere06c8222008-09-11 15:35:47 -07001426 jh->b_triggers = type;
Jan Karaad56eda2013-03-11 13:24:56 -04001427 jbd2_journal_put_journal_head(jh);
Joel Beckere06c8222008-09-11 15:35:47 -07001428}
1429
Jan Kara13ceef02010-07-14 07:56:33 +02001430void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
Joel Beckere06c8222008-09-11 15:35:47 -07001431 struct jbd2_buffer_trigger_type *triggers)
1432{
1433 struct buffer_head *bh = jh2bh(jh);
1434
Jan Kara13ceef02010-07-14 07:56:33 +02001435 if (!triggers || !triggers->t_frozen)
Joel Beckere06c8222008-09-11 15:35:47 -07001436 return;
1437
Jan Kara13ceef02010-07-14 07:56:33 +02001438 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
Joel Beckere06c8222008-09-11 15:35:47 -07001439}
1440
1441void jbd2_buffer_abort_trigger(struct journal_head *jh,
1442 struct jbd2_buffer_trigger_type *triggers)
1443{
1444 if (!triggers || !triggers->t_abort)
1445 return;
1446
1447 triggers->t_abort(triggers, jh2bh(jh));
1448}
1449
Joel Beckere06c8222008-09-11 15:35:47 -07001450/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +01001451 * jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
Dave Kleikamp470decc2006-10-11 01:20:57 -07001452 * @handle: transaction to add buffer to.
1453 * @bh: buffer to mark
1454 *
1455 * mark dirty metadata which needs to be journaled as part of the current
1456 * transaction.
1457 *
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001458 * The buffer must have previously had jbd2_journal_get_write_access()
1459 * called so that it has a valid journal_head attached to the buffer
1460 * head.
1461 *
Dave Kleikamp470decc2006-10-11 01:20:57 -07001462 * The buffer is placed on the transaction's metadata list and is marked
1463 * as belonging to the transaction.
1464 *
1465 * Returns error number or 0 on success.
1466 *
1467 * Special care needs to be taken if the buffer already belongs to the
1468 * current committing transaction (in which case we should have frozen
1469 * data present for that commit). In that case, we don't relink the
1470 * buffer: that only gets done when the old transaction finally
1471 * completes its commit.
1472 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001473int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001474{
1475 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001476 journal_t *journal;
Jan Karaad56eda2013-03-11 13:24:56 -04001477 struct journal_head *jh;
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001478 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001479
Dave Kleikamp470decc2006-10-11 01:20:57 -07001480 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001481 return -EROFS;
zhangyi (F)01215d32019-02-21 11:24:09 -05001482 if (!buffer_jbd(bh))
1483 return -EUCLEAN;
1484
Jan Kara6e06ae82015-07-12 18:11:30 -04001485 /*
1486 * We don't grab jh reference here since the buffer must be part
1487 * of the running transaction.
1488 */
1489 jh = bh2jh(bh);
zhangyi (F)01215d32019-02-21 11:24:09 -05001490 jbd_debug(5, "journal_head %p\n", jh);
1491 JBUFFER_TRACE(jh, "entry");
1492
Jan Kara6e06ae82015-07-12 18:11:30 -04001493 /*
1494 * This and the following assertions are unreliable since we may see jh
1495 * in inconsistent state unless we grab bh_state lock. But this is
1496 * crucial to catch bugs so let's do a reliable check until the
1497 * lockless handling is fully proven.
1498 */
Jan Kara83fe6b12021-04-06 18:18:00 +02001499 if (data_race(jh->b_transaction != transaction &&
1500 jh->b_next_transaction != transaction)) {
Thomas Gleixner46417062019-08-09 14:42:32 +02001501 spin_lock(&jh->b_state_lock);
Jan Kara6e06ae82015-07-12 18:11:30 -04001502 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1503 jh->b_next_transaction == transaction);
Thomas Gleixner46417062019-08-09 14:42:32 +02001504 spin_unlock(&jh->b_state_lock);
Jan Kara6e06ae82015-07-12 18:11:30 -04001505 }
1506 if (jh->b_modified == 1) {
1507 /* If it's in our transaction it must be in BJ_Metadata list. */
Jan Kara83fe6b12021-04-06 18:18:00 +02001508 if (data_race(jh->b_transaction == transaction &&
1509 jh->b_jlist != BJ_Metadata)) {
Thomas Gleixner46417062019-08-09 14:42:32 +02001510 spin_lock(&jh->b_state_lock);
Theodore Ts'oe09463f2018-06-16 20:21:45 -04001511 if (jh->b_transaction == transaction &&
1512 jh->b_jlist != BJ_Metadata)
1513 pr_err("JBD2: assertion failure: h_type=%u "
1514 "h_line_no=%u block_no=%llu jlist=%u\n",
1515 handle->h_type, handle->h_line_no,
1516 (unsigned long long) bh->b_blocknr,
1517 jh->b_jlist);
Jan Kara6e06ae82015-07-12 18:11:30 -04001518 J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1519 jh->b_jlist == BJ_Metadata);
Thomas Gleixner46417062019-08-09 14:42:32 +02001520 spin_unlock(&jh->b_state_lock);
Jan Kara6e06ae82015-07-12 18:11:30 -04001521 }
1522 goto out;
1523 }
1524
1525 journal = transaction->t_journal;
Thomas Gleixner46417062019-08-09 14:42:32 +02001526 spin_lock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001527
1528 if (jh->b_modified == 0) {
1529 /*
1530 * This buffer's got modified and becoming part
1531 * of the transaction. This needs to be done
1532 * once a transaction -bzzz
1533 */
Jan Karad0907072019-11-05 17:44:28 +01001534 if (WARN_ON_ONCE(jbd2_handle_buffer_credits(handle) <= 0)) {
Theodore Ts'of6c07ca2013-12-08 21:12:59 -05001535 ret = -ENOSPC;
1536 goto out_unlock_bh;
1537 }
Theodore Ts'oe09463f2018-06-16 20:21:45 -04001538 jh->b_modified = 1;
Jan Kara933f1c12019-11-05 17:44:27 +01001539 handle->h_total_credits--;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001540 }
1541
1542 /*
1543 * fastpath, to avoid expensive locking. If this buffer is already
1544 * on the running transaction's metadata list there is nothing to do.
1545 * Nobody can take it off again because there is a handle open.
1546 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1547 * result in this test being false, so we go in and take the locks.
1548 */
1549 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1550 JBUFFER_TRACE(jh, "fastpath");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001551 if (unlikely(jh->b_transaction !=
1552 journal->j_running_transaction)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001553 printk(KERN_ERR "JBD2: %s: "
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001554 "jh->b_transaction (%llu, %p, %u) != "
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001555 "journal->j_running_transaction (%p, %u)\n",
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001556 journal->j_devname,
1557 (unsigned long long) bh->b_blocknr,
1558 jh->b_transaction,
1559 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1560 journal->j_running_transaction,
1561 journal->j_running_transaction ?
1562 journal->j_running_transaction->t_tid : 0);
1563 ret = -EINVAL;
1564 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001565 goto out_unlock_bh;
1566 }
1567
1568 set_buffer_jbddirty(bh);
1569
1570 /*
1571 * Metadata already on the current transaction list doesn't
1572 * need to be filed. Metadata on another transaction's list must
1573 * be committing, and will be refiled once the commit completes:
1574 * leave it alone for now.
1575 */
1576 if (jh->b_transaction != transaction) {
1577 JBUFFER_TRACE(jh, "already on other transaction");
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001578 if (unlikely(((jh->b_transaction !=
1579 journal->j_committing_transaction)) ||
1580 (jh->b_next_transaction != transaction))) {
1581 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1582 "bad jh for block %llu: "
1583 "transaction (%p, %u), "
1584 "jh->b_transaction (%p, %u), "
1585 "jh->b_next_transaction (%p, %u), jlist %u\n",
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001586 journal->j_devname,
1587 (unsigned long long) bh->b_blocknr,
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001588 transaction, transaction->t_tid,
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001589 jh->b_transaction,
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001590 jh->b_transaction ?
1591 jh->b_transaction->t_tid : 0,
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001592 jh->b_next_transaction,
1593 jh->b_next_transaction ?
1594 jh->b_next_transaction->t_tid : 0,
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001595 jh->b_jlist);
1596 WARN_ON(1);
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001597 ret = -EINVAL;
1598 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001599 /* And this case is illegal: we can't reuse another
1600 * transaction's data buffer, ever. */
1601 goto out_unlock_bh;
1602 }
1603
1604 /* That test should have eliminated the following case: */
Mingming Cao40191912008-01-28 23:58:27 -05001605 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001606
1607 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1608 spin_lock(&journal->j_list_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001609 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001610 spin_unlock(&journal->j_list_lock);
1611out_unlock_bh:
Thomas Gleixner46417062019-08-09 14:42:32 +02001612 spin_unlock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001613out:
1614 JBUFFER_TRACE(jh, "exit");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001615 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001616}
1617
Dave Kleikamp470decc2006-10-11 01:20:57 -07001618/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +01001619 * jbd2_journal_forget() - bforget() for potentially-journaled buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001620 * @handle: transaction handle
1621 * @bh: bh to 'forget'
1622 *
1623 * We can only do the bforget if there are no commits pending against the
1624 * buffer. If the buffer is dirty in the current running transaction we
1625 * can safely unlink it.
1626 *
1627 * bh may not be a journalled buffer at all - it may be a non-JBD
1628 * buffer which came off the hashtable. Check for this.
1629 *
1630 * Decrements bh->b_count by one.
1631 *
1632 * Allow this call even if the handle has aborted --- it may be part of
1633 * the caller's cleanup after an abort.
1634 */
Shijie Luo8d6ce132020-01-23 01:43:25 -05001635int jbd2_journal_forget(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001636{
1637 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001638 journal_t *journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001639 struct journal_head *jh;
1640 int drop_reserve = 0;
1641 int err = 0;
Josef Bacik1dfc3222008-04-17 10:38:59 -04001642 int was_modified = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001643
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001644 if (is_handle_aborted(handle))
1645 return -EROFS;
1646 journal = transaction->t_journal;
1647
Dave Kleikamp470decc2006-10-11 01:20:57 -07001648 BUFFER_TRACE(bh, "entry");
1649
Thomas Gleixner46417062019-08-09 14:42:32 +02001650 jh = jbd2_journal_grab_journal_head(bh);
1651 if (!jh) {
1652 __bforget(bh);
1653 return 0;
1654 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001655
Thomas Gleixner46417062019-08-09 14:42:32 +02001656 spin_lock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001657
1658 /* Critical error: attempting to delete a bitmap buffer, maybe?
1659 * Don't do any jbd operations, and return an error. */
1660 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1661 "inconsistent data on disk")) {
1662 err = -EIO;
Jan Kara2e710ff2019-08-09 14:42:31 +02001663 goto drop;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001664 }
1665
Adam Buchbinder48fc7f72012-09-19 21:48:00 -04001666 /* keep track of whether or not this transaction modified us */
Josef Bacik1dfc3222008-04-17 10:38:59 -04001667 was_modified = jh->b_modified;
1668
Dave Kleikamp470decc2006-10-11 01:20:57 -07001669 /*
1670 * The buffer's going from the transaction, we must drop
1671 * all references -bzzz
1672 */
1673 jh->b_modified = 0;
1674
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001675 if (jh->b_transaction == transaction) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001676 J_ASSERT_JH(jh, !jh->b_frozen_data);
1677
1678 /* If we are forgetting a buffer which is already part
1679 * of this transaction, then we can just drop it from
1680 * the transaction immediately. */
1681 clear_buffer_dirty(bh);
1682 clear_buffer_jbddirty(bh);
1683
1684 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1685
Josef Bacik1dfc3222008-04-17 10:38:59 -04001686 /*
1687 * we only want to drop a reference if this transaction
1688 * modified the buffer
1689 */
1690 if (was_modified)
1691 drop_reserve = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001692
1693 /*
1694 * We are no longer going to journal this buffer.
1695 * However, the commit of this transaction is still
1696 * important to the buffer: the delete that we are now
1697 * processing might obsolete an old log entry, so by
1698 * committing, we can satisfy the buffer's checkpoint.
1699 *
1700 * So, if we have a checkpoint on the buffer, we should
1701 * now refile the buffer on our BJ_Forget list so that
1702 * we know to remove the checkpoint after we commit.
1703 */
1704
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001705 spin_lock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001706 if (jh->b_cp_transaction) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001707 __jbd2_journal_temp_unlink_buffer(jh);
1708 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001709 } else {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001710 __jbd2_journal_unfile_buffer(jh);
Jan Kara93108eb2019-08-09 14:42:29 +02001711 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001712 }
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001713 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001714 } else if (jh->b_transaction) {
1715 J_ASSERT_JH(jh, (jh->b_transaction ==
1716 journal->j_committing_transaction));
1717 /* However, if the buffer is still owned by a prior
1718 * (committing) transaction, we can't drop it yet... */
1719 JBUFFER_TRACE(jh, "belongs to older transaction");
zhangyi (F)904cdbd2019-02-10 23:23:04 -05001720 /* ... but we CAN drop it from the new transaction through
1721 * marking the buffer as freed and set j_next_transaction to
1722 * the new transaction, so that not only the commit code
1723 * knows it should clear dirty bits when it is done with the
1724 * buffer, but also the buffer can be checkpointed only
1725 * after the new transaction commits. */
Dave Kleikamp470decc2006-10-11 01:20:57 -07001726
zhangyi (F)904cdbd2019-02-10 23:23:04 -05001727 set_buffer_freed(bh);
1728
1729 if (!jh->b_next_transaction) {
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001730 spin_lock(&journal->j_list_lock);
zhangyi (F)904cdbd2019-02-10 23:23:04 -05001731 jh->b_next_transaction = transaction;
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001732 spin_unlock(&journal->j_list_lock);
zhangyi (F)904cdbd2019-02-10 23:23:04 -05001733 } else {
1734 J_ASSERT(jh->b_next_transaction == transaction);
Josef Bacik1dfc3222008-04-17 10:38:59 -04001735
1736 /*
1737 * only drop a reference if this transaction modified
1738 * the buffer
1739 */
1740 if (was_modified)
1741 drop_reserve = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001742 }
zhangyi (F)59759922019-02-10 23:26:06 -05001743 } else {
1744 /*
1745 * Finally, if the buffer is not belongs to any
1746 * transaction, we can just drop it now if it has no
1747 * checkpoint.
1748 */
1749 spin_lock(&journal->j_list_lock);
1750 if (!jh->b_cp_transaction) {
1751 JBUFFER_TRACE(jh, "belongs to none transaction");
1752 spin_unlock(&journal->j_list_lock);
Jan Kara2e710ff2019-08-09 14:42:31 +02001753 goto drop;
zhangyi (F)59759922019-02-10 23:26:06 -05001754 }
1755
1756 /*
1757 * Otherwise, if the buffer has been written to disk,
1758 * it is safe to remove the checkpoint and drop it.
1759 */
1760 if (!buffer_dirty(bh)) {
1761 __jbd2_journal_remove_checkpoint(jh);
1762 spin_unlock(&journal->j_list_lock);
Jan Kara2e710ff2019-08-09 14:42:31 +02001763 goto drop;
zhangyi (F)59759922019-02-10 23:26:06 -05001764 }
1765
1766 /*
1767 * The buffer is still not written to disk, we should
1768 * attach this buffer to current transaction so that the
1769 * buffer can be checkpointed only after the current
1770 * transaction commits.
1771 */
1772 clear_buffer_dirty(bh);
1773 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
1774 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001775 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001776drop:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001777 __brelse(bh);
Thomas Gleixner46417062019-08-09 14:42:32 +02001778 spin_unlock(&jh->b_state_lock);
1779 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001780 if (drop_reserve) {
1781 /* no need to reserve log space for this block -bzzz */
Jan Kara933f1c12019-11-05 17:44:27 +01001782 handle->h_total_credits++;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001783 }
1784 return err;
1785}
1786
1787/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +01001788 * jbd2_journal_stop() - complete a transaction
Masanari Iidabd7ced92016-02-02 22:31:06 +09001789 * @handle: transaction to complete.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001790 *
1791 * All done for a particular handle.
1792 *
1793 * There is not much action needed here. We just return any remaining
1794 * buffer credits to the transaction and remove the handle. The only
1795 * complication is that we need to start a commit operation if the
1796 * filesystem is marked for synchronous update.
1797 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001798 * jbd2_journal_stop itself will not usually return an error, but it may
Dave Kleikamp470decc2006-10-11 01:20:57 -07001799 * do so in unusual circumstances. In particular, expect it to
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001800 * return -EIO if a jbd2_journal_abort has been executed since the
Dave Kleikamp470decc2006-10-11 01:20:57 -07001801 * transaction began.
1802 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001803int jbd2_journal_stop(handle_t *handle)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001804{
1805 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001806 journal_t *journal;
1807 int err = 0, wait_for_commit = 0;
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001808 tid_t tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001809 pid_t pid;
1810
Jan Karadfaf5ff2019-11-05 17:44:20 +01001811 if (--handle->h_ref > 0) {
1812 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1813 handle->h_ref);
1814 if (is_handle_aborted(handle))
1815 return -EIO;
1816 return 0;
1817 }
Lukas Czerner9d506592015-05-14 18:55:18 -04001818 if (!transaction) {
1819 /*
Jan Karadfaf5ff2019-11-05 17:44:20 +01001820 * Handle is already detached from the transaction so there is
1821 * nothing to do other than free the handle.
Lukas Czerner9d506592015-05-14 18:55:18 -04001822 */
Jan Karaec8b6f62019-11-05 17:44:23 +01001823 memalloc_nofs_restore(handle->saved_alloc_context);
Jan Karadfaf5ff2019-11-05 17:44:20 +01001824 goto free_and_exit;
Lukas Czerner9d506592015-05-14 18:55:18 -04001825 }
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001826 journal = transaction->t_journal;
Jan Karadfaf5ff2019-11-05 17:44:20 +01001827 tid = transaction->t_tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001828
1829 if (is_handle_aborted(handle))
1830 err = -EIO;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001831
1832 jbd_debug(4, "Handle %p going down\n", handle);
Theodore Ts'o343d9c22013-02-08 13:00:22 -05001833 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
Jan Karadfaf5ff2019-11-05 17:44:20 +01001834 tid, handle->h_type, handle->h_line_no,
Theodore Ts'o343d9c22013-02-08 13:00:22 -05001835 jiffies - handle->h_start_jiffies,
1836 handle->h_sync, handle->h_requested_credits,
1837 (handle->h_requested_credits -
Jan Kara933f1c12019-11-05 17:44:27 +01001838 handle->h_total_credits));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001839
1840 /*
1841 * Implement synchronous transaction batching. If the handle
1842 * was synchronous, don't force a commit immediately. Let's
Josef Bacike07f7182008-11-26 01:14:26 -05001843 * yield and let another thread piggyback onto this
1844 * transaction. Keep doing that while new threads continue to
1845 * arrive. It doesn't cost much - we're about to run a commit
1846 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1847 * operations by 30x or more...
Dave Kleikamp470decc2006-10-11 01:20:57 -07001848 *
Josef Bacike07f7182008-11-26 01:14:26 -05001849 * We try and optimize the sleep time against what the
1850 * underlying disk can do, instead of having a static sleep
1851 * time. This is useful for the case where our storage is so
1852 * fast that it is more optimal to go ahead and force a flush
1853 * and wait for the transaction to be committed than it is to
1854 * wait for an arbitrary amount of time for new writers to
1855 * join the transaction. We achieve this by measuring how
1856 * long it takes to commit a transaction, and compare it with
1857 * how long this transaction has been running, and if run time
1858 * < commit time then we sleep for the delta and commit. This
1859 * greatly helps super fast disks that would see slowdowns as
1860 * more threads started doing fsyncs.
1861 *
1862 * But don't do this if this process was the most recent one
1863 * to perform a synchronous write. We do this to detect the
1864 * case where a single process is doing a stream of sync
1865 * writes. No point in waiting for joiners in that case.
Eric Sandeen5dd21422014-07-05 19:18:22 -04001866 *
1867 * Setting max_batch_time to 0 disables this completely.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001868 */
1869 pid = current->pid;
Eric Sandeen5dd21422014-07-05 19:18:22 -04001870 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1871 journal->j_max_batch_time) {
Josef Bacike07f7182008-11-26 01:14:26 -05001872 u64 commit_time, trans_time;
1873
Dave Kleikamp470decc2006-10-11 01:20:57 -07001874 journal->j_last_sync_writer = pid;
Josef Bacike07f7182008-11-26 01:14:26 -05001875
Theodore Ts'oa931da62010-08-03 21:35:12 -04001876 read_lock(&journal->j_state_lock);
Josef Bacike07f7182008-11-26 01:14:26 -05001877 commit_time = journal->j_average_commit_time;
Theodore Ts'oa931da62010-08-03 21:35:12 -04001878 read_unlock(&journal->j_state_lock);
Josef Bacike07f7182008-11-26 01:14:26 -05001879
1880 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1881 transaction->t_start_time));
1882
Theodore Ts'o30773842009-01-03 20:27:38 -05001883 commit_time = max_t(u64, commit_time,
1884 1000*journal->j_min_batch_time);
Josef Bacike07f7182008-11-26 01:14:26 -05001885 commit_time = min_t(u64, commit_time,
Theodore Ts'o30773842009-01-03 20:27:38 -05001886 1000*journal->j_max_batch_time);
Josef Bacike07f7182008-11-26 01:14:26 -05001887
1888 if (trans_time < commit_time) {
1889 ktime_t expires = ktime_add_ns(ktime_get(),
1890 commit_time);
1891 set_current_state(TASK_UNINTERRUPTIBLE);
1892 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1893 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001894 }
1895
Theodore Ts'o70585482009-03-25 23:35:46 -04001896 if (handle->h_sync)
1897 transaction->t_synchronous_commit = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001898
1899 /*
1900 * If the handle is marked SYNC, we need to set another commit
Jan Kara150549e2019-11-05 17:44:21 +01001901 * going! We also want to force a commit if the transaction is too
1902 * old now.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001903 */
1904 if (handle->h_sync ||
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001905 time_after_eq(jiffies, transaction->t_expires)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001906 /* Do this even for aborted journals: an abort still
1907 * completes the commit thread, it just doesn't write
1908 * anything to disk. */
Dave Kleikamp470decc2006-10-11 01:20:57 -07001909
Dave Kleikamp470decc2006-10-11 01:20:57 -07001910 jbd_debug(2, "transaction too old, requesting commit for "
1911 "handle %p\n", handle);
1912 /* This is non-blocking */
Jan Karadfaf5ff2019-11-05 17:44:20 +01001913 jbd2_log_start_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001914
1915 /*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001916 * Special case: JBD2_SYNC synchronous updates require us
Dave Kleikamp470decc2006-10-11 01:20:57 -07001917 * to wait for the commit to complete.
1918 */
1919 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001920 wait_for_commit = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001921 }
1922
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001923 /*
Jan Karaec8b6f62019-11-05 17:44:23 +01001924 * Once stop_this_handle() drops t_updates, the transaction could start
1925 * committing on us and eventually disappear. So we must not
1926 * dereference transaction pointer again after calling
1927 * stop_this_handle().
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001928 */
Jan Karaec8b6f62019-11-05 17:44:23 +01001929 stop_this_handle(handle);
Jan Kara7a4b1882016-06-30 11:30:21 -04001930
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001931 if (wait_for_commit)
1932 err = jbd2_log_wait_commit(journal, tid);
1933
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001934free_and_exit:
Jan Karaec8b6f62019-11-05 17:44:23 +01001935 if (handle->h_rsv_handle)
1936 jbd2_free_handle(handle->h_rsv_handle);
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001937 jbd2_free_handle(handle);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001938 return err;
1939}
1940
Dave Kleikamp470decc2006-10-11 01:20:57 -07001941/*
1942 *
1943 * List management code snippets: various functions for manipulating the
1944 * transaction buffer lists.
1945 *
1946 */
1947
1948/*
1949 * Append a buffer to a transaction list, given the transaction's list head
1950 * pointer.
1951 *
1952 * j_list_lock is held.
1953 *
Thomas Gleixner46417062019-08-09 14:42:32 +02001954 * jh->b_state_lock is held.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001955 */
1956
1957static inline void
1958__blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1959{
1960 if (!*list) {
1961 jh->b_tnext = jh->b_tprev = jh;
1962 *list = jh;
1963 } else {
1964 /* Insert at the tail of the list to preserve order */
1965 struct journal_head *first = *list, *last = first->b_tprev;
1966 jh->b_tprev = last;
1967 jh->b_tnext = first;
1968 last->b_tnext = first->b_tprev = jh;
1969 }
1970}
1971
1972/*
1973 * Remove a buffer from a transaction list, given the transaction's list
1974 * head pointer.
1975 *
1976 * Called with j_list_lock held, and the journal may not be locked.
1977 *
Thomas Gleixner46417062019-08-09 14:42:32 +02001978 * jh->b_state_lock is held.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001979 */
1980
1981static inline void
1982__blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1983{
1984 if (*list == jh) {
1985 *list = jh->b_tnext;
1986 if (*list == jh)
1987 *list = NULL;
1988 }
1989 jh->b_tprev->b_tnext = jh->b_tnext;
1990 jh->b_tnext->b_tprev = jh->b_tprev;
1991}
1992
1993/*
1994 * Remove a buffer from the appropriate transaction list.
1995 *
1996 * Note that this function can *change* the value of
Jan Karaf5113ef2013-06-04 12:01:45 -04001997 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1998 * t_reserved_list. If the caller is holding onto a copy of one of these
1999 * pointers, it could go bad. Generally the caller needs to re-read the
2000 * pointer from the transaction_t.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002001 *
Jan Kara5bebccf2012-03-13 22:25:06 -04002002 * Called under j_list_lock.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002003 */
Jan Kara5bebccf2012-03-13 22:25:06 -04002004static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002005{
2006 struct journal_head **list = NULL;
2007 transaction_t *transaction;
2008 struct buffer_head *bh = jh2bh(jh);
2009
Thomas Gleixner46417062019-08-09 14:42:32 +02002010 lockdep_assert_held(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002011 transaction = jh->b_transaction;
2012 if (transaction)
2013 assert_spin_locked(&transaction->t_journal->j_list_lock);
2014
2015 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2016 if (jh->b_jlist != BJ_None)
Mingming Cao40191912008-01-28 23:58:27 -05002017 J_ASSERT_JH(jh, transaction != NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002018
2019 switch (jh->b_jlist) {
2020 case BJ_None:
2021 return;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002022 case BJ_Metadata:
2023 transaction->t_nr_buffers--;
2024 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
2025 list = &transaction->t_buffers;
2026 break;
2027 case BJ_Forget:
2028 list = &transaction->t_forget;
2029 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002030 case BJ_Shadow:
2031 list = &transaction->t_shadow_list;
2032 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002033 case BJ_Reserved:
2034 list = &transaction->t_reserved_list;
2035 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002036 }
2037
2038 __blist_del_buffer(list, jh);
2039 jh->b_jlist = BJ_None;
Theodore Ts'oe1126662017-02-04 23:14:19 -05002040 if (transaction && is_journal_aborted(transaction->t_journal))
2041 clear_buffer_jbddirty(bh);
2042 else if (test_clear_buffer_jbddirty(bh))
Dave Kleikamp470decc2006-10-11 01:20:57 -07002043 mark_buffer_dirty(bh); /* Expose it to the VM */
2044}
2045
Jan Karade1b7942011-06-13 15:38:22 -04002046/*
Jan Kara93108eb2019-08-09 14:42:29 +02002047 * Remove buffer from all transactions. The caller is responsible for dropping
2048 * the jh reference that belonged to the transaction.
Jan Karade1b7942011-06-13 15:38:22 -04002049 *
2050 * Called with bh_state lock and j_list_lock
Jan Karade1b7942011-06-13 15:38:22 -04002051 */
2052static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002053{
Lukas Czerner24dc9862020-06-17 11:25:49 +02002054 J_ASSERT_JH(jh, jh->b_transaction != NULL);
2055 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
2056
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002057 __jbd2_journal_temp_unlink_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002058 jh->b_transaction = NULL;
2059}
2060
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002061void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002062{
Jan Karade1b7942011-06-13 15:38:22 -04002063 struct buffer_head *bh = jh2bh(jh);
2064
2065 /* Get reference so that buffer cannot be freed before we unlock it */
2066 get_bh(bh);
Thomas Gleixner46417062019-08-09 14:42:32 +02002067 spin_lock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002068 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002069 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002070 spin_unlock(&journal->j_list_lock);
Thomas Gleixner46417062019-08-09 14:42:32 +02002071 spin_unlock(&jh->b_state_lock);
Jan Kara93108eb2019-08-09 14:42:29 +02002072 jbd2_journal_put_journal_head(jh);
Jan Karade1b7942011-06-13 15:38:22 -04002073 __brelse(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002074}
2075
2076/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002077 * Called from jbd2_journal_try_to_free_buffers().
Dave Kleikamp470decc2006-10-11 01:20:57 -07002078 *
Thomas Gleixner46417062019-08-09 14:42:32 +02002079 * Called under jh->b_state_lock
Dave Kleikamp470decc2006-10-11 01:20:57 -07002080 */
2081static void
2082__journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
2083{
2084 struct journal_head *jh;
2085
2086 jh = bh2jh(bh);
2087
2088 if (buffer_locked(bh) || buffer_dirty(bh))
2089 goto out;
2090
Theodore Ts'od2eb0b92014-03-09 00:07:19 -05002091 if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002092 goto out;
2093
2094 spin_lock(&journal->j_list_lock);
Theodore Ts'od2eb0b92014-03-09 00:07:19 -05002095 if (jh->b_cp_transaction != NULL) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002096 /* written-back checkpointed metadata buffer */
Jan Karac254c9e2012-03-13 22:27:44 -04002097 JBUFFER_TRACE(jh, "remove from checkpoint list");
2098 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002099 }
2100 spin_unlock(&journal->j_list_lock);
2101out:
2102 return;
2103}
2104
Dave Kleikamp470decc2006-10-11 01:20:57 -07002105/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +01002106 * jbd2_journal_try_to_free_buffers() - try to free page buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002107 * @journal: journal for operation
2108 * @page: to try and free
Dave Kleikamp470decc2006-10-11 01:20:57 -07002109 *
2110 * For all the buffers on this page,
2111 * if they are fully written out ordered data, move them onto BUF_CLEAN
2112 * so try_to_free_buffers() can reap them.
2113 *
2114 * This function returns non-zero if we wish try_to_free_buffers()
2115 * to be called. We do this if the page is releasable by try_to_free_buffers().
2116 * We also do it if the page has locked or dirty buffers and the caller wants
2117 * us to perform sync or async writeout.
2118 *
2119 * This complicates JBD locking somewhat. We aren't protected by the
2120 * BKL here. We wish to remove the buffer from its committing or
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002121 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002122 *
2123 * This may *change* the value of transaction_t->t_datalist, so anyone
2124 * who looks at t_datalist needs to lock against this function.
2125 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002126 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
2127 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
Dave Kleikamp470decc2006-10-11 01:20:57 -07002128 * will come out of the lock with the buffer dirty, which makes it
2129 * ineligible for release here.
2130 *
2131 * Who else is affected by this? hmm... Really the only contender
2132 * is do_get_write_access() - it could be looking at the buffer while
2133 * journal_try_to_free_buffer() is changing its state. But that
2134 * cannot happen because we never reallocate freed data as metadata
2135 * while the data is part of a transaction. Yes?
Mingming Cao530576b2008-07-13 21:06:39 -04002136 *
2137 * Return 0 on failure, 1 on success
Dave Kleikamp470decc2006-10-11 01:20:57 -07002138 */
zhangyi (F)529a7812020-06-20 10:54:27 +08002139int jbd2_journal_try_to_free_buffers(journal_t *journal, struct page *page)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002140{
2141 struct buffer_head *head;
2142 struct buffer_head *bh;
2143 int ret = 0;
2144
2145 J_ASSERT(PageLocked(page));
2146
2147 head = page_buffers(page);
2148 bh = head;
2149 do {
2150 struct journal_head *jh;
2151
2152 /*
2153 * We take our own ref against the journal_head here to avoid
2154 * having to add tons of locking around each instance of
Mingming Cao530576b2008-07-13 21:06:39 -04002155 * jbd2_journal_put_journal_head().
Dave Kleikamp470decc2006-10-11 01:20:57 -07002156 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002157 jh = jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002158 if (!jh)
2159 continue;
2160
Thomas Gleixner46417062019-08-09 14:42:32 +02002161 spin_lock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002162 __journal_try_to_free_buffer(journal, bh);
Thomas Gleixner46417062019-08-09 14:42:32 +02002163 spin_unlock(&jh->b_state_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002164 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002165 if (buffer_jbd(bh))
2166 goto busy;
2167 } while ((bh = bh->b_this_page) != head);
Mingming Cao530576b2008-07-13 21:06:39 -04002168
Dave Kleikamp470decc2006-10-11 01:20:57 -07002169 ret = try_to_free_buffers(page);
2170busy:
2171 return ret;
2172}
2173
2174/*
2175 * This buffer is no longer needed. If it is on an older transaction's
2176 * checkpoint list we need to record it on this transaction's forget list
2177 * to pin this buffer (and hence its checkpointing transaction) down until
2178 * this transaction commits. If the buffer isn't on a checkpoint list, we
2179 * release it.
2180 * Returns non-zero if JBD no longer has an interest in the buffer.
2181 *
2182 * Called under j_list_lock.
2183 *
Thomas Gleixner46417062019-08-09 14:42:32 +02002184 * Called under jh->b_state_lock.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002185 */
2186static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2187{
2188 int may_free = 1;
2189 struct buffer_head *bh = jh2bh(jh);
2190
Dave Kleikamp470decc2006-10-11 01:20:57 -07002191 if (jh->b_cp_transaction) {
2192 JBUFFER_TRACE(jh, "on running+cp transaction");
Jan Karade1b7942011-06-13 15:38:22 -04002193 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karaf91d1d02009-07-13 16:16:20 -04002194 /*
2195 * We don't want to write the buffer anymore, clear the
2196 * bit so that we don't confuse checks in
2197 * __journal_file_buffer
2198 */
2199 clear_buffer_dirty(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002200 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002201 may_free = 0;
2202 } else {
2203 JBUFFER_TRACE(jh, "on running transaction");
Jan Karade1b7942011-06-13 15:38:22 -04002204 __jbd2_journal_unfile_buffer(jh);
Jan Kara93108eb2019-08-09 14:42:29 +02002205 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002206 }
2207 return may_free;
2208}
2209
2210/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002211 * jbd2_journal_invalidatepage
Dave Kleikamp470decc2006-10-11 01:20:57 -07002212 *
2213 * This code is tricky. It has a number of cases to deal with.
2214 *
2215 * There are two invariants which this code relies on:
2216 *
2217 * i_size must be updated on disk before we start calling invalidatepage on the
2218 * data.
2219 *
2220 * This is done in ext3 by defining an ext3_setattr method which
2221 * updates i_size before truncate gets going. By maintaining this
2222 * invariant, we can be sure that it is safe to throw away any buffers
2223 * attached to the current transaction: once the transaction commits,
2224 * we know that the data will not be needed.
2225 *
2226 * Note however that we can *not* throw away data belonging to the
2227 * previous, committing transaction!
2228 *
2229 * Any disk blocks which *are* part of the previous, committing
2230 * transaction (and which therefore cannot be discarded immediately) are
2231 * not going to be reused in the new running transaction
2232 *
2233 * The bitmap committed_data images guarantee this: any block which is
2234 * allocated in one transaction and removed in the next will be marked
2235 * as in-use in the committed_data bitmap, so cannot be reused until
2236 * the next transaction to delete the block commits. This means that
2237 * leaving committing buffers dirty is quite safe: the disk blocks
2238 * cannot be reallocated to a different file and so buffer aliasing is
2239 * not possible.
2240 *
2241 *
2242 * The above applies mainly to ordered data mode. In writeback mode we
2243 * don't make guarantees about the order in which data hits disk --- in
2244 * particular we don't guarantee that new dirty data is flushed before
2245 * transaction commit --- so it is always safe just to discard data
2246 * immediately in that mode. --sct
2247 */
2248
2249/*
2250 * The journal_unmap_buffer helper function returns zero if the buffer
2251 * concerned remains pinned as an anonymous buffer belonging to an older
2252 * transaction.
2253 *
2254 * We're outside-transaction here. Either or both of j_running_transaction
2255 * and j_committing_transaction may be NULL.
2256 */
Jan Karab794e7a2012-09-26 23:11:13 -04002257static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2258 int partial_page)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002259{
2260 transaction_t *transaction;
2261 struct journal_head *jh;
2262 int may_free = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002263
2264 BUFFER_TRACE(bh, "entry");
2265
2266 /*
2267 * It is safe to proceed here without the j_list_lock because the
2268 * buffers cannot be stolen by try_to_free_buffers as long as we are
2269 * holding the page lock. --sct
2270 */
2271
Thomas Gleixnerd84560f2019-08-09 14:42:27 +02002272 jh = jbd2_journal_grab_journal_head(bh);
2273 if (!jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002274 goto zap_buffer_unlocked;
2275
Jan Kara87c89c22008-07-11 19:27:31 -04002276 /* OK, we have data buffer in journaled mode */
Theodore Ts'oa931da62010-08-03 21:35:12 -04002277 write_lock(&journal->j_state_lock);
Thomas Gleixner46417062019-08-09 14:42:32 +02002278 spin_lock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002279 spin_lock(&journal->j_list_lock);
2280
dingdinghuaba869022010-02-15 16:35:42 -05002281 /*
2282 * We cannot remove the buffer from checkpoint lists until the
2283 * transaction adding inode to orphan list (let's call it T)
2284 * is committed. Otherwise if the transaction changing the
2285 * buffer would be cleaned from the journal before T is
2286 * committed, a crash will cause that the correct contents of
2287 * the buffer will be lost. On the other hand we have to
2288 * clear the buffer dirty bit at latest at the moment when the
2289 * transaction marking the buffer as freed in the filesystem
2290 * structures is committed because from that moment on the
Jan Karab794e7a2012-09-26 23:11:13 -04002291 * block can be reallocated and used by a different page.
dingdinghuaba869022010-02-15 16:35:42 -05002292 * Since the block hasn't been freed yet but the inode has
2293 * already been added to orphan list, it is safe for us to add
2294 * the buffer to BJ_Forget list of the newest transaction.
Jan Karab794e7a2012-09-26 23:11:13 -04002295 *
2296 * Also we have to clear buffer_mapped flag of a truncated buffer
2297 * because the buffer_head may be attached to the page straddling
2298 * i_size (can happen only when blocksize < pagesize) and thus the
2299 * buffer_head can be reused when the file is extended again. So we end
2300 * up keeping around invalidated buffers attached to transactions'
2301 * BJ_Forget list just to stop checkpointing code from cleaning up
2302 * the transaction this buffer was modified in.
dingdinghuaba869022010-02-15 16:35:42 -05002303 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002304 transaction = jh->b_transaction;
2305 if (transaction == NULL) {
2306 /* First case: not on any transaction. If it
2307 * has no checkpoint link, then we can zap it:
2308 * it's a writeback-mode buffer so we don't care
2309 * if it hits disk safely. */
2310 if (!jh->b_cp_transaction) {
2311 JBUFFER_TRACE(jh, "not on any transaction: zap");
2312 goto zap_buffer;
2313 }
2314
2315 if (!buffer_dirty(bh)) {
2316 /* bdflush has written it. We can drop it now */
Jan Karabc23f0c2015-11-24 15:34:35 -05002317 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002318 goto zap_buffer;
2319 }
2320
2321 /* OK, it must be in the journal but still not
2322 * written fully to disk: it's metadata or
2323 * journaled data... */
2324
2325 if (journal->j_running_transaction) {
2326 /* ... and once the current transaction has
2327 * committed, the buffer won't be needed any
2328 * longer. */
2329 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
Jan Karab794e7a2012-09-26 23:11:13 -04002330 may_free = __dispose_buffer(jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002331 journal->j_running_transaction);
Jan Karab794e7a2012-09-26 23:11:13 -04002332 goto zap_buffer;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002333 } else {
2334 /* There is no currently-running transaction. So the
2335 * orphan record which we wrote for this file must have
2336 * passed into commit. We must attach this buffer to
2337 * the committing transaction, if it exists. */
2338 if (journal->j_committing_transaction) {
2339 JBUFFER_TRACE(jh, "give to committing trans");
Jan Karab794e7a2012-09-26 23:11:13 -04002340 may_free = __dispose_buffer(jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002341 journal->j_committing_transaction);
Jan Karab794e7a2012-09-26 23:11:13 -04002342 goto zap_buffer;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002343 } else {
2344 /* The orphan record's transaction has
2345 * committed. We can cleanse this buffer */
2346 clear_buffer_jbddirty(bh);
Jan Karabc23f0c2015-11-24 15:34:35 -05002347 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002348 goto zap_buffer;
2349 }
2350 }
2351 } else if (transaction == journal->j_committing_transaction) {
Eric Sandeen9b579882006-10-28 10:38:28 -07002352 JBUFFER_TRACE(jh, "on committing transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002353 /*
dingdinghuaba869022010-02-15 16:35:42 -05002354 * The buffer is committing, we simply cannot touch
Jan Karab794e7a2012-09-26 23:11:13 -04002355 * it. If the page is straddling i_size we have to wait
2356 * for commit and try again.
2357 */
2358 if (partial_page) {
Jan Karab794e7a2012-09-26 23:11:13 -04002359 spin_unlock(&journal->j_list_lock);
Thomas Gleixner46417062019-08-09 14:42:32 +02002360 spin_unlock(&jh->b_state_lock);
Jan Karab794e7a2012-09-26 23:11:13 -04002361 write_unlock(&journal->j_state_lock);
Thomas Gleixner46417062019-08-09 14:42:32 +02002362 jbd2_journal_put_journal_head(jh);
Jan Kara53e87262012-12-25 13:29:52 -05002363 return -EBUSY;
Jan Karab794e7a2012-09-26 23:11:13 -04002364 }
2365 /*
zhangyi (F)6a66a7d2020-02-13 14:38:20 +08002366 * OK, buffer won't be reachable after truncate. We just clear
2367 * b_modified to not confuse transaction credit accounting, and
2368 * set j_next_transaction to the running transaction (if there
2369 * is one) and mark buffer as freed so that commit code knows
2370 * it should clear dirty bits when it is done with the buffer.
dingdinghuaba869022010-02-15 16:35:42 -05002371 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002372 set_buffer_freed(bh);
dingdinghuaba869022010-02-15 16:35:42 -05002373 if (journal->j_running_transaction && buffer_jbddirty(bh))
2374 jh->b_next_transaction = journal->j_running_transaction;
zhangyi (F)6a66a7d2020-02-13 14:38:20 +08002375 jh->b_modified = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002376 spin_unlock(&journal->j_list_lock);
Thomas Gleixner46417062019-08-09 14:42:32 +02002377 spin_unlock(&jh->b_state_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002378 write_unlock(&journal->j_state_lock);
Thomas Gleixner46417062019-08-09 14:42:32 +02002379 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002380 return 0;
2381 } else {
2382 /* Good, the buffer belongs to the running transaction.
2383 * We are writing our own transaction's data, not any
2384 * previous one's, so it is safe to throw it away
2385 * (remember that we expect the filesystem to have set
2386 * i_size already for this truncate so recovery will not
2387 * expose the disk blocks we are discarding here.) */
2388 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
Eric Sandeen9b579882006-10-28 10:38:28 -07002389 JBUFFER_TRACE(jh, "on running transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002390 may_free = __dispose_buffer(jh, transaction);
2391 }
2392
2393zap_buffer:
Jan Karab794e7a2012-09-26 23:11:13 -04002394 /*
2395 * This is tricky. Although the buffer is truncated, it may be reused
2396 * if blocksize < pagesize and it is attached to the page straddling
2397 * EOF. Since the buffer might have been added to BJ_Forget list of the
2398 * running transaction, journal_get_write_access() won't clear
2399 * b_modified and credit accounting gets confused. So clear b_modified
2400 * here.
2401 */
2402 jh->b_modified = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002403 spin_unlock(&journal->j_list_lock);
Thomas Gleixner46417062019-08-09 14:42:32 +02002404 spin_unlock(&jh->b_state_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002405 write_unlock(&journal->j_state_lock);
Thomas Gleixner46417062019-08-09 14:42:32 +02002406 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002407zap_buffer_unlocked:
2408 clear_buffer_dirty(bh);
2409 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2410 clear_buffer_mapped(bh);
2411 clear_buffer_req(bh);
2412 clear_buffer_new(bh);
Eric Sandeen15291162012-02-20 17:53:01 -05002413 clear_buffer_delay(bh);
2414 clear_buffer_unwritten(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002415 bh->b_bdev = NULL;
2416 return may_free;
2417}
2418
2419/**
Mauro Carvalho Chehab2bf31d92020-11-16 11:18:08 +01002420 * jbd2_journal_invalidatepage()
Dave Kleikamp470decc2006-10-11 01:20:57 -07002421 * @journal: journal to use for flush...
2422 * @page: page to flush
Lukas Czerner259709b2013-05-21 23:20:03 -04002423 * @offset: start of the range to invalidate
2424 * @length: length of the range to invalidate
Dave Kleikamp470decc2006-10-11 01:20:57 -07002425 *
Lukas Czerner259709b2013-05-21 23:20:03 -04002426 * Reap page buffers containing data after in the specified range in page.
2427 * Can return -EBUSY if buffers are part of the committing transaction and
2428 * the page is straddling i_size. Caller then has to wait for current commit
2429 * and try again.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002430 */
Jan Kara53e87262012-12-25 13:29:52 -05002431int jbd2_journal_invalidatepage(journal_t *journal,
2432 struct page *page,
Lukas Czerner259709b2013-05-21 23:20:03 -04002433 unsigned int offset,
2434 unsigned int length)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002435{
2436 struct buffer_head *head, *bh, *next;
Lukas Czerner259709b2013-05-21 23:20:03 -04002437 unsigned int stop = offset + length;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002438 unsigned int curr_off = 0;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002439 int partial_page = (offset || length < PAGE_SIZE);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002440 int may_free = 1;
Jan Kara53e87262012-12-25 13:29:52 -05002441 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002442
2443 if (!PageLocked(page))
2444 BUG();
2445 if (!page_has_buffers(page))
Jan Kara53e87262012-12-25 13:29:52 -05002446 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002447
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002448 BUG_ON(stop > PAGE_SIZE || stop < length);
Lukas Czerner259709b2013-05-21 23:20:03 -04002449
Dave Kleikamp470decc2006-10-11 01:20:57 -07002450 /* We will potentially be playing with lists other than just the
2451 * data lists (especially for journaled data mode), so be
2452 * cautious in our locking. */
2453
2454 head = bh = page_buffers(page);
2455 do {
2456 unsigned int next_off = curr_off + bh->b_size;
2457 next = bh->b_this_page;
2458
Lukas Czerner259709b2013-05-21 23:20:03 -04002459 if (next_off > stop)
2460 return 0;
2461
Dave Kleikamp470decc2006-10-11 01:20:57 -07002462 if (offset <= curr_off) {
2463 /* This block is wholly outside the truncation point */
2464 lock_buffer(bh);
Lukas Czerner259709b2013-05-21 23:20:03 -04002465 ret = journal_unmap_buffer(journal, bh, partial_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002466 unlock_buffer(bh);
Jan Kara53e87262012-12-25 13:29:52 -05002467 if (ret < 0)
2468 return ret;
2469 may_free &= ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002470 }
2471 curr_off = next_off;
2472 bh = next;
2473
2474 } while (bh != head);
2475
Lukas Czerner259709b2013-05-21 23:20:03 -04002476 if (!partial_page) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002477 if (may_free && try_to_free_buffers(page))
2478 J_ASSERT(!page_has_buffers(page));
2479 }
Jan Kara53e87262012-12-25 13:29:52 -05002480 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002481}
2482
2483/*
2484 * File a buffer on the given transaction list.
2485 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002486void __jbd2_journal_file_buffer(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002487 transaction_t *transaction, int jlist)
2488{
2489 struct journal_head **list = NULL;
2490 int was_dirty = 0;
2491 struct buffer_head *bh = jh2bh(jh);
2492
Thomas Gleixner46417062019-08-09 14:42:32 +02002493 lockdep_assert_held(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002494 assert_spin_locked(&transaction->t_journal->j_list_lock);
2495
2496 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2497 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
Mingming Cao40191912008-01-28 23:58:27 -05002498 jh->b_transaction == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002499
2500 if (jh->b_transaction && jh->b_jlist == jlist)
2501 return;
2502
Dave Kleikamp470decc2006-10-11 01:20:57 -07002503 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2504 jlist == BJ_Shadow || jlist == BJ_Forget) {
Jan Karaf91d1d02009-07-13 16:16:20 -04002505 /*
2506 * For metadata buffers, we track dirty bit in buffer_jbddirty
2507 * instead of buffer_dirty. We should not see a dirty bit set
2508 * here because we clear it in do_get_write_access but e.g.
2509 * tune2fs can modify the sb and set the dirty bit at any time
2510 * so we try to gracefully handle that.
2511 */
2512 if (buffer_dirty(bh))
2513 warn_dirty_buffer(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002514 if (test_clear_buffer_dirty(bh) ||
2515 test_clear_buffer_jbddirty(bh))
2516 was_dirty = 1;
2517 }
2518
2519 if (jh->b_transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002520 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karade1b7942011-06-13 15:38:22 -04002521 else
2522 jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002523 jh->b_transaction = transaction;
2524
2525 switch (jlist) {
2526 case BJ_None:
2527 J_ASSERT_JH(jh, !jh->b_committed_data);
2528 J_ASSERT_JH(jh, !jh->b_frozen_data);
2529 return;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002530 case BJ_Metadata:
2531 transaction->t_nr_buffers++;
2532 list = &transaction->t_buffers;
2533 break;
2534 case BJ_Forget:
2535 list = &transaction->t_forget;
2536 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002537 case BJ_Shadow:
2538 list = &transaction->t_shadow_list;
2539 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002540 case BJ_Reserved:
2541 list = &transaction->t_reserved_list;
2542 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002543 }
2544
2545 __blist_add_buffer(list, jh);
2546 jh->b_jlist = jlist;
2547
2548 if (was_dirty)
2549 set_buffer_jbddirty(bh);
2550}
2551
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002552void jbd2_journal_file_buffer(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002553 transaction_t *transaction, int jlist)
2554{
Thomas Gleixner46417062019-08-09 14:42:32 +02002555 spin_lock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002556 spin_lock(&transaction->t_journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002557 __jbd2_journal_file_buffer(jh, transaction, jlist);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002558 spin_unlock(&transaction->t_journal->j_list_lock);
Thomas Gleixner46417062019-08-09 14:42:32 +02002559 spin_unlock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002560}
2561
2562/*
2563 * Remove a buffer from its current buffer list in preparation for
2564 * dropping it from its current transaction entirely. If the buffer has
2565 * already started to be used by a subsequent transaction, refile the
2566 * buffer on that transaction's metadata list.
2567 *
Jan Karade1b7942011-06-13 15:38:22 -04002568 * Called under j_list_lock
Thomas Gleixner46417062019-08-09 14:42:32 +02002569 * Called under jh->b_state_lock
Jan Karade1b7942011-06-13 15:38:22 -04002570 *
Jan Kara93108eb2019-08-09 14:42:29 +02002571 * When this function returns true, there's no next transaction to refile to
2572 * and the caller has to drop jh reference through
2573 * jbd2_journal_put_journal_head().
Dave Kleikamp470decc2006-10-11 01:20:57 -07002574 */
Jan Kara93108eb2019-08-09 14:42:29 +02002575bool __jbd2_journal_refile_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002576{
dingdinghuaba869022010-02-15 16:35:42 -05002577 int was_dirty, jlist;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002578 struct buffer_head *bh = jh2bh(jh);
2579
Thomas Gleixner46417062019-08-09 14:42:32 +02002580 lockdep_assert_held(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002581 if (jh->b_transaction)
2582 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2583
2584 /* If the buffer is now unused, just drop it. */
2585 if (jh->b_next_transaction == NULL) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002586 __jbd2_journal_unfile_buffer(jh);
Jan Kara93108eb2019-08-09 14:42:29 +02002587 return true;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002588 }
2589
2590 /*
2591 * It has been modified by a later transaction: add it to the new
2592 * transaction's metadata list.
2593 */
2594
2595 was_dirty = test_clear_buffer_jbddirty(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002596 __jbd2_journal_temp_unlink_buffer(jh);
Lukas Czerner24dc9862020-06-17 11:25:49 +02002597
2598 /*
2599 * b_transaction must be set, otherwise the new b_transaction won't
2600 * be holding jh reference
2601 */
2602 J_ASSERT_JH(jh, jh->b_transaction != NULL);
2603
Jan Karade1b7942011-06-13 15:38:22 -04002604 /*
2605 * We set b_transaction here because b_next_transaction will inherit
2606 * our jh reference and thus __jbd2_journal_file_buffer() must not
2607 * take a new one.
2608 */
Qian Cai6c5d9112020-02-21 23:31:11 -05002609 WRITE_ONCE(jh->b_transaction, jh->b_next_transaction);
2610 WRITE_ONCE(jh->b_next_transaction, NULL);
dingdinghuaba869022010-02-15 16:35:42 -05002611 if (buffer_freed(bh))
2612 jlist = BJ_Forget;
2613 else if (jh->b_modified)
2614 jlist = BJ_Metadata;
2615 else
2616 jlist = BJ_Reserved;
2617 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002618 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2619
2620 if (was_dirty)
2621 set_buffer_jbddirty(bh);
Jan Kara93108eb2019-08-09 14:42:29 +02002622 return false;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002623}
2624
2625/*
Jan Karade1b7942011-06-13 15:38:22 -04002626 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2627 * bh reference so that we can safely unlock bh.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002628 *
Jan Karade1b7942011-06-13 15:38:22 -04002629 * The jh and bh may be freed by this call.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002630 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002631void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002632{
Jan Kara93108eb2019-08-09 14:42:29 +02002633 bool drop;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002634
Thomas Gleixner46417062019-08-09 14:42:32 +02002635 spin_lock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002636 spin_lock(&journal->j_list_lock);
Jan Kara93108eb2019-08-09 14:42:29 +02002637 drop = __jbd2_journal_refile_buffer(jh);
Thomas Gleixner46417062019-08-09 14:42:32 +02002638 spin_unlock(&jh->b_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002639 spin_unlock(&journal->j_list_lock);
Jan Kara93108eb2019-08-09 14:42:29 +02002640 if (drop)
2641 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002642}
Jan Karac851ed52008-07-11 19:27:31 -04002643
2644/*
2645 * File inode in the inode list of the handle's transaction
2646 */
Jan Kara41617e12016-04-24 00:56:07 -04002647static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
Ross Zwisler6ba0e7d2019-06-20 17:24:56 -04002648 unsigned long flags, loff_t start_byte, loff_t end_byte)
Jan Karac851ed52008-07-11 19:27:31 -04002649{
2650 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002651 journal_t *journal;
Jan Karac851ed52008-07-11 19:27:31 -04002652
2653 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002654 return -EROFS;
2655 journal = transaction->t_journal;
Jan Karac851ed52008-07-11 19:27:31 -04002656
2657 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2658 transaction->t_tid);
2659
Jan Karac851ed52008-07-11 19:27:31 -04002660 spin_lock(&journal->j_list_lock);
Jan Kara41617e12016-04-24 00:56:07 -04002661 jinode->i_flags |= flags;
Ross Zwisler6ba0e7d2019-06-20 17:24:56 -04002662
2663 if (jinode->i_dirty_end) {
2664 jinode->i_dirty_start = min(jinode->i_dirty_start, start_byte);
2665 jinode->i_dirty_end = max(jinode->i_dirty_end, end_byte);
2666 } else {
2667 jinode->i_dirty_start = start_byte;
2668 jinode->i_dirty_end = end_byte;
2669 }
2670
Jan Kara41617e12016-04-24 00:56:07 -04002671 /* Is inode already attached where we need it? */
Jan Karac851ed52008-07-11 19:27:31 -04002672 if (jinode->i_transaction == transaction ||
2673 jinode->i_next_transaction == transaction)
2674 goto done;
2675
Jan Kara81be12c2011-05-24 11:52:40 -04002676 /*
2677 * We only ever set this variable to 1 so the test is safe. Since
2678 * t_need_data_flush is likely to be set, we do the test to save some
2679 * cacheline bouncing
2680 */
2681 if (!transaction->t_need_data_flush)
2682 transaction->t_need_data_flush = 1;
Jan Karac851ed52008-07-11 19:27:31 -04002683 /* On some different transaction's list - should be
2684 * the committing one */
2685 if (jinode->i_transaction) {
2686 J_ASSERT(jinode->i_next_transaction == NULL);
2687 J_ASSERT(jinode->i_transaction ==
2688 journal->j_committing_transaction);
2689 jinode->i_next_transaction = transaction;
2690 goto done;
2691 }
2692 /* Not on any transaction list... */
2693 J_ASSERT(!jinode->i_next_transaction);
2694 jinode->i_transaction = transaction;
2695 list_add(&jinode->i_list, &transaction->t_inode_list);
2696done:
2697 spin_unlock(&journal->j_list_lock);
2698
2699 return 0;
2700}
2701
Ross Zwisler6ba0e7d2019-06-20 17:24:56 -04002702int jbd2_journal_inode_ranged_write(handle_t *handle,
2703 struct jbd2_inode *jinode, loff_t start_byte, loff_t length)
2704{
2705 return jbd2_journal_file_inode(handle, jinode,
2706 JI_WRITE_DATA | JI_WAIT_DATA, start_byte,
2707 start_byte + length - 1);
2708}
2709
2710int jbd2_journal_inode_ranged_wait(handle_t *handle, struct jbd2_inode *jinode,
2711 loff_t start_byte, loff_t length)
2712{
2713 return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA,
2714 start_byte, start_byte + length - 1);
Jan Kara41617e12016-04-24 00:56:07 -04002715}
2716
Jan Karac851ed52008-07-11 19:27:31 -04002717/*
Jan Kara7f5aa212009-02-10 11:15:34 -05002718 * File truncate and transaction commit interact with each other in a
2719 * non-trivial way. If a transaction writing data block A is
2720 * committing, we cannot discard the data by truncate until we have
2721 * written them. Otherwise if we crashed after the transaction with
2722 * write has committed but before the transaction with truncate has
2723 * committed, we could see stale data in block A. This function is a
2724 * helper to solve this problem. It starts writeout of the truncated
2725 * part in case it is in the committing transaction.
2726 *
2727 * Filesystem code must call this function when inode is journaled in
2728 * ordered mode before truncation happens and after the inode has been
2729 * placed on orphan list with the new inode size. The second condition
2730 * avoids the race that someone writes new data and we start
2731 * committing the transaction after this function has been called but
2732 * before a transaction for truncate is started (and furthermore it
2733 * allows us to optimize the case where the addition to orphan list
2734 * happens in the same transaction as write --- we don't have to write
2735 * any data in such case).
Jan Karac851ed52008-07-11 19:27:31 -04002736 */
Jan Kara7f5aa212009-02-10 11:15:34 -05002737int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2738 struct jbd2_inode *jinode,
Jan Karac851ed52008-07-11 19:27:31 -04002739 loff_t new_size)
2740{
Jan Kara7f5aa212009-02-10 11:15:34 -05002741 transaction_t *inode_trans, *commit_trans;
Jan Karac851ed52008-07-11 19:27:31 -04002742 int ret = 0;
2743
Jan Kara7f5aa212009-02-10 11:15:34 -05002744 /* This is a quick check to avoid locking if not necessary */
2745 if (!jinode->i_transaction)
Jan Karac851ed52008-07-11 19:27:31 -04002746 goto out;
Jan Kara7f5aa212009-02-10 11:15:34 -05002747 /* Locks are here just to force reading of recent values, it is
2748 * enough that the transaction was not committing before we started
2749 * a transaction adding the inode to orphan list */
Theodore Ts'oa931da62010-08-03 21:35:12 -04002750 read_lock(&journal->j_state_lock);
Jan Karac851ed52008-07-11 19:27:31 -04002751 commit_trans = journal->j_committing_transaction;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002752 read_unlock(&journal->j_state_lock);
Jan Kara7f5aa212009-02-10 11:15:34 -05002753 spin_lock(&journal->j_list_lock);
2754 inode_trans = jinode->i_transaction;
2755 spin_unlock(&journal->j_list_lock);
2756 if (inode_trans == commit_trans) {
2757 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
Jan Karac851ed52008-07-11 19:27:31 -04002758 new_size, LLONG_MAX);
2759 if (ret)
2760 jbd2_journal_abort(journal, ret);
2761 }
2762out:
2763 return ret;
2764}