blob: e1652665bd93d0cf30dece02b5a1b7692fdec811 [file] [log] [blame]
Dave Kleikamp470decc2006-10-11 01:20:57 -07001/*
Uwe Kleine-König58862692007-05-09 07:51:49 +02002 * linux/fs/jbd2/transaction.c
Dave Kleikamp470decc2006-10-11 01:20:57 -07003 *
4 * Written by Stephen C. Tweedie <sct@redhat.com>, 1998
5 *
6 * Copyright 1998 Red Hat corp --- All Rights Reserved
7 *
8 * This file is part of the Linux kernel and is made available under
9 * the terms of the GNU General Public License, version 2, or at your
10 * option, any later version, incorporated herein by reference.
11 *
12 * Generic filesystem transaction handling code; part of the ext2fs
13 * journaling system.
14 *
15 * This file manages transactions (compound commits managed by the
16 * journaling code) and handles (individual atomic operations by the
17 * filesystem).
18 */
19
20#include <linux/time.h>
21#include <linux/fs.h>
Mingming Caof7f4bcc2006-10-11 01:20:59 -070022#include <linux/jbd2.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070023#include <linux/errno.h>
24#include <linux/slab.h>
25#include <linux/timer.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070026#include <linux/mm.h>
27#include <linux/highmem.h>
Josef Bacike07f7182008-11-26 01:14:26 -050028#include <linux/hrtimer.h>
Theodore Ts'o47def822010-07-27 11:56:05 -040029#include <linux/backing-dev.h>
Randy Dunlap44705752011-10-27 04:05:13 -040030#include <linux/bug.h>
Theodore Ts'o47def822010-07-27 11:56:05 -040031#include <linux/module.h>
Dave Kleikamp470decc2006-10-11 01:20:57 -070032
Theodore Ts'o343d9c22013-02-08 13:00:22 -050033#include <trace/events/jbd2.h>
34
Adrian Bunk7ddae862006-12-06 20:38:27 -080035static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh);
Jan Karade1b7942011-06-13 15:38:22 -040036static void __jbd2_journal_unfile_buffer(struct journal_head *jh);
Adrian Bunk7ddae862006-12-06 20:38:27 -080037
Yongqiang Yang0c2022e2012-02-20 17:53:02 -050038static struct kmem_cache *transaction_cache;
39int __init jbd2_journal_init_transaction_cache(void)
40{
41 J_ASSERT(!transaction_cache);
42 transaction_cache = kmem_cache_create("jbd2_transaction_s",
43 sizeof(transaction_t),
44 0,
45 SLAB_HWCACHE_ALIGN|SLAB_TEMPORARY,
46 NULL);
47 if (transaction_cache)
48 return 0;
49 return -ENOMEM;
50}
51
52void jbd2_journal_destroy_transaction_cache(void)
53{
54 if (transaction_cache) {
55 kmem_cache_destroy(transaction_cache);
56 transaction_cache = NULL;
57 }
58}
59
60void jbd2_journal_free_transaction(transaction_t *transaction)
61{
62 if (unlikely(ZERO_OR_NULL_PTR(transaction)))
63 return;
64 kmem_cache_free(transaction_cache, transaction);
65}
66
Dave Kleikamp470decc2006-10-11 01:20:57 -070067/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -070068 * jbd2_get_transaction: obtain a new transaction_t object.
Dave Kleikamp470decc2006-10-11 01:20:57 -070069 *
70 * Simply allocate and initialise a new transaction. Create it in
71 * RUNNING state and add it to the current journal (which should not
72 * have an existing running transaction: we only make a new transaction
73 * once we have started to commit the old one).
74 *
75 * Preconditions:
76 * The journal MUST be locked. We don't perform atomic mallocs on the
77 * new transaction and we can't block without protecting against other
78 * processes trying to touch the journal while it is in transition.
79 *
Dave Kleikamp470decc2006-10-11 01:20:57 -070080 */
81
82static transaction_t *
Mingming Caof7f4bcc2006-10-11 01:20:59 -070083jbd2_get_transaction(journal_t *journal, transaction_t *transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -070084{
85 transaction->t_journal = journal;
86 transaction->t_state = T_RUNNING;
Josef Bacike07f7182008-11-26 01:14:26 -050087 transaction->t_start_time = ktime_get();
Dave Kleikamp470decc2006-10-11 01:20:57 -070088 transaction->t_tid = journal->j_transaction_sequence++;
89 transaction->t_expires = jiffies + journal->j_commit_interval;
90 spin_lock_init(&transaction->t_handle_lock);
Theodore Ts'oa51dca92010-08-02 08:43:25 -040091 atomic_set(&transaction->t_updates, 0);
Jan Kara8f7d89f2013-06-04 12:35:11 -040092 atomic_set(&transaction->t_outstanding_credits,
93 atomic_read(&journal->j_reserved_credits));
Theodore Ts'o8dd42042010-08-03 21:38:29 -040094 atomic_set(&transaction->t_handle_count, 0);
Jan Karac851ed52008-07-11 19:27:31 -040095 INIT_LIST_HEAD(&transaction->t_inode_list);
Theodore Ts'o3e624fc2008-10-16 20:00:24 -040096 INIT_LIST_HEAD(&transaction->t_private_list);
Dave Kleikamp470decc2006-10-11 01:20:57 -070097
98 /* Set up the commit timer for the new transaction. */
Andreas Dilgerb1f485f2009-08-10 22:51:53 -040099 journal->j_commit_timer.expires = round_jiffies_up(transaction->t_expires);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700100 add_timer(&journal->j_commit_timer);
101
102 J_ASSERT(journal->j_running_transaction == NULL);
103 journal->j_running_transaction = transaction;
Johann Lombardi8e85fb32008-01-28 23:58:27 -0500104 transaction->t_max_wait = 0;
105 transaction->t_start = jiffies;
Theodore Ts'o9fff24a2013-02-06 22:30:23 -0500106 transaction->t_requested = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700107
108 return transaction;
109}
110
111/*
112 * Handle management.
113 *
114 * A handle_t is an object which represents a single atomic update to a
115 * filesystem, and which tracks all of the modifications which form part
116 * of that one update.
117 */
118
119/*
Tao Ma28e35e42011-05-22 21:45:26 -0400120 * Update transaction's maximum wait time, if debugging is enabled.
Theodore Ts'o6d0bf002010-08-09 17:28:38 -0400121 *
122 * In order for t_max_wait to be reliable, it must be protected by a
123 * lock. But doing so will mean that start_this_handle() can not be
124 * run in parallel on SMP systems, which limits our scalability. So
125 * unless debugging is enabled, we no longer update t_max_wait, which
126 * means that maximum wait time reported by the jbd2_run_stats
127 * tracepoint will always be zero.
128 */
Tao Ma28e35e42011-05-22 21:45:26 -0400129static inline void update_t_max_wait(transaction_t *transaction,
130 unsigned long ts)
Theodore Ts'o6d0bf002010-08-09 17:28:38 -0400131{
132#ifdef CONFIG_JBD2_DEBUG
Theodore Ts'o6d0bf002010-08-09 17:28:38 -0400133 if (jbd2_journal_enable_debug &&
134 time_after(transaction->t_start, ts)) {
135 ts = jbd2_time_diff(ts, transaction->t_start);
136 spin_lock(&transaction->t_handle_lock);
137 if (ts > transaction->t_max_wait)
138 transaction->t_max_wait = ts;
139 spin_unlock(&transaction->t_handle_lock);
140 }
141#endif
142}
143
144/*
Jan Kara8f7d89f2013-06-04 12:35:11 -0400145 * Wait until running transaction passes T_LOCKED state. Also starts the commit
146 * if needed. The function expects running transaction to exist and releases
147 * j_state_lock.
148 */
149static void wait_transaction_locked(journal_t *journal)
150 __releases(journal->j_state_lock)
151{
152 DEFINE_WAIT(wait);
153 int need_to_start;
154 tid_t tid = journal->j_running_transaction->t_tid;
155
156 prepare_to_wait(&journal->j_wait_transaction_locked, &wait,
157 TASK_UNINTERRUPTIBLE);
158 need_to_start = !tid_geq(journal->j_commit_request, tid);
159 read_unlock(&journal->j_state_lock);
160 if (need_to_start)
161 jbd2_log_start_commit(journal, tid);
Jan Karae03a9972016-09-22 11:44:06 -0400162 jbd2_might_wait_for_commit(journal);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400163 schedule();
164 finish_wait(&journal->j_wait_transaction_locked, &wait);
165}
166
167static void sub_reserved_credits(journal_t *journal, int blocks)
168{
169 atomic_sub(blocks, &journal->j_reserved_credits);
170 wake_up(&journal->j_wait_reserved);
171}
172
173/*
174 * Wait until we can add credits for handle to the running transaction. Called
175 * with j_state_lock held for reading. Returns 0 if handle joined the running
176 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
177 * caller must retry.
178 */
179static int add_transaction_credits(journal_t *journal, int blocks,
180 int rsv_blocks)
181{
182 transaction_t *t = journal->j_running_transaction;
183 int needed;
184 int total = blocks + rsv_blocks;
185
186 /*
187 * If the current transaction is locked down for commit, wait
188 * for the lock to be released.
189 */
190 if (t->t_state == T_LOCKED) {
191 wait_transaction_locked(journal);
192 return 1;
193 }
194
195 /*
196 * If there is not enough space left in the log to write all
197 * potential buffers requested by this operation, we need to
198 * stall pending a log checkpoint to free some more log space.
199 */
200 needed = atomic_add_return(total, &t->t_outstanding_credits);
201 if (needed > journal->j_max_transaction_buffers) {
202 /*
203 * If the current transaction is already too large,
204 * then start to commit it: we can then go back and
205 * attach this handle to a new transaction.
206 */
207 atomic_sub(total, &t->t_outstanding_credits);
Lukas Czerner6d3ec142015-08-04 11:21:52 -0400208
209 /*
210 * Is the number of reserved credits in the current transaction too
211 * big to fit this handle? Wait until reserved credits are freed.
212 */
213 if (atomic_read(&journal->j_reserved_credits) + total >
214 journal->j_max_transaction_buffers) {
215 read_unlock(&journal->j_state_lock);
Jan Karae03a9972016-09-22 11:44:06 -0400216 jbd2_might_wait_for_commit(journal);
Lukas Czerner6d3ec142015-08-04 11:21:52 -0400217 wait_event(journal->j_wait_reserved,
218 atomic_read(&journal->j_reserved_credits) + total <=
219 journal->j_max_transaction_buffers);
220 return 1;
221 }
222
Jan Kara8f7d89f2013-06-04 12:35:11 -0400223 wait_transaction_locked(journal);
224 return 1;
225 }
226
227 /*
228 * The commit code assumes that it can get enough log space
229 * without forcing a checkpoint. This is *critical* for
230 * correctness: a checkpoint of a buffer which is also
231 * associated with a committing transaction creates a deadlock,
232 * so commit simply cannot force through checkpoints.
233 *
234 * We must therefore ensure the necessary space in the journal
235 * *before* starting to dirty potentially checkpointed buffers
236 * in the new transaction.
237 */
238 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) {
239 atomic_sub(total, &t->t_outstanding_credits);
240 read_unlock(&journal->j_state_lock);
Jan Karae03a9972016-09-22 11:44:06 -0400241 jbd2_might_wait_for_commit(journal);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400242 write_lock(&journal->j_state_lock);
243 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
244 __jbd2_log_wait_for_space(journal);
245 write_unlock(&journal->j_state_lock);
246 return 1;
247 }
248
249 /* No reservation? We are done... */
250 if (!rsv_blocks)
251 return 0;
252
253 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
254 /* We allow at most half of a transaction to be reserved */
255 if (needed > journal->j_max_transaction_buffers / 2) {
256 sub_reserved_credits(journal, rsv_blocks);
257 atomic_sub(total, &t->t_outstanding_credits);
258 read_unlock(&journal->j_state_lock);
Jan Karae03a9972016-09-22 11:44:06 -0400259 jbd2_might_wait_for_commit(journal);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400260 wait_event(journal->j_wait_reserved,
261 atomic_read(&journal->j_reserved_credits) + rsv_blocks
262 <= journal->j_max_transaction_buffers / 2);
263 return 1;
264 }
265 return 0;
266}
267
268/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700269 * start_this_handle: Given a handle, deal with any locking or stalling
270 * needed to make sure that there is enough journal space for the handle
271 * to begin. Attach the handle to a transaction and set up the
272 * transaction's buffer credits.
273 */
274
Theodore Ts'o47def822010-07-27 11:56:05 -0400275static int start_this_handle(journal_t *journal, handle_t *handle,
Dan Carpenterd2159fb2011-09-04 10:20:14 -0400276 gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700277{
Theodore Ts'oe4471832011-02-12 08:18:24 -0500278 transaction_t *transaction, *new_transaction = NULL;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400279 int blocks = handle->h_buffer_credits;
280 int rsv_blocks = 0;
Tao Ma28e35e42011-05-22 21:45:26 -0400281 unsigned long ts = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700282
Jan Kara8f7d89f2013-06-04 12:35:11 -0400283 if (handle->h_rsv_handle)
284 rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
285
Lukas Czerner6d3ec142015-08-04 11:21:52 -0400286 /*
287 * Limit the number of reserved credits to 1/2 of maximum transaction
288 * size and limit the number of total credits to not exceed maximum
289 * transaction size per operation.
290 */
291 if ((rsv_blocks > journal->j_max_transaction_buffers / 2) ||
292 (rsv_blocks + blocks > journal->j_max_transaction_buffers)) {
293 printk(KERN_ERR "JBD2: %s wants too many credits "
294 "credits:%d rsv_credits:%d max:%d\n",
295 current->comm, blocks, rsv_blocks,
296 journal->j_max_transaction_buffers);
297 WARN_ON(1);
298 return -ENOSPC;
299 }
300
Dave Kleikamp470decc2006-10-11 01:20:57 -0700301alloc_transaction:
302 if (!journal->j_running_transaction) {
Michal Hocko6ccaf3e2015-06-08 10:53:10 -0400303 /*
304 * If __GFP_FS is not present, then we may be being called from
305 * inside the fs writeback layer, so we MUST NOT fail.
306 */
307 if ((gfp_mask & __GFP_FS) == 0)
308 gfp_mask |= __GFP_NOFAIL;
Wanlong Gaob2f4edb2012-06-01 00:10:32 -0400309 new_transaction = kmem_cache_zalloc(transaction_cache,
310 gfp_mask);
Michal Hocko6ccaf3e2015-06-08 10:53:10 -0400311 if (!new_transaction)
Theodore Ts'o47def822010-07-27 11:56:05 -0400312 return -ENOMEM;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700313 }
314
315 jbd_debug(3, "New handle %p going live.\n", handle);
316
Dave Kleikamp470decc2006-10-11 01:20:57 -0700317 /*
318 * We need to hold j_state_lock until t_updates has been incremented,
319 * for proper journal barrier handling
320 */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400321repeat:
322 read_lock(&journal->j_state_lock);
Theodore Ts'o5c2178e2010-10-27 21:30:04 -0400323 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700324 if (is_journal_aborted(journal) ||
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700325 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400326 read_unlock(&journal->j_state_lock);
Yongqiang Yang0c2022e2012-02-20 17:53:02 -0500327 jbd2_journal_free_transaction(new_transaction);
Theodore Ts'o47def822010-07-27 11:56:05 -0400328 return -EROFS;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700329 }
330
Jan Kara8f7d89f2013-06-04 12:35:11 -0400331 /*
332 * Wait on the journal's transaction barrier if necessary. Specifically
333 * we allow reserved handles to proceed because otherwise commit could
334 * deadlock on page writeback not being able to complete.
335 */
336 if (!handle->h_reserved && journal->j_barrier_count) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400337 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700338 wait_event(journal->j_wait_transaction_locked,
339 journal->j_barrier_count == 0);
340 goto repeat;
341 }
342
343 if (!journal->j_running_transaction) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400344 read_unlock(&journal->j_state_lock);
345 if (!new_transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700346 goto alloc_transaction;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400347 write_lock(&journal->j_state_lock);
Jan Karad7961c72012-12-21 00:15:51 -0500348 if (!journal->j_running_transaction &&
Jan Kara8f7d89f2013-06-04 12:35:11 -0400349 (handle->h_reserved || !journal->j_barrier_count)) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400350 jbd2_get_transaction(journal, new_transaction);
351 new_transaction = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700352 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400353 write_unlock(&journal->j_state_lock);
354 goto repeat;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700355 }
356
357 transaction = journal->j_running_transaction;
358
Jan Kara8f7d89f2013-06-04 12:35:11 -0400359 if (!handle->h_reserved) {
360 /* We may have dropped j_state_lock - restart in that case */
361 if (add_transaction_credits(journal, blocks, rsv_blocks))
362 goto repeat;
363 } else {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700364 /*
Jan Kara8f7d89f2013-06-04 12:35:11 -0400365 * We have handle reserved so we are allowed to join T_LOCKED
366 * transaction and we don't have to check for transaction size
367 * and journal space.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700368 */
Jan Kara8f7d89f2013-06-04 12:35:11 -0400369 sub_reserved_credits(journal, blocks);
370 handle->h_reserved = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700371 }
372
373 /* OK, account for the buffers that this operation expects to
Theodore Ts'o8dd42042010-08-03 21:38:29 -0400374 * use and add the handle to the running transaction.
Theodore Ts'o8dd42042010-08-03 21:38:29 -0400375 */
Tao Ma28e35e42011-05-22 21:45:26 -0400376 update_t_max_wait(transaction, ts);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700377 handle->h_transaction = transaction;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400378 handle->h_requested_credits = blocks;
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500379 handle->h_start_jiffies = jiffies;
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400380 atomic_inc(&transaction->t_updates);
Theodore Ts'o8dd42042010-08-03 21:38:29 -0400381 atomic_inc(&transaction->t_handle_count);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400382 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
383 handle, blocks,
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400384 atomic_read(&transaction->t_outstanding_credits),
Jan Kara76c39902013-06-04 12:12:57 -0400385 jbd2_log_space_left(journal));
Theodore Ts'oa931da62010-08-03 21:35:12 -0400386 read_unlock(&journal->j_state_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400387 current->journal_info = handle;
Jan Kara9599b0e2009-08-17 21:23:17 -0400388
Jan Karaab714af2016-06-30 11:39:38 -0400389 rwsem_acquire_read(&journal->j_trans_commit_map, 0, 0, _THIS_IP_);
Yongqiang Yang0c2022e2012-02-20 17:53:02 -0500390 jbd2_journal_free_transaction(new_transaction);
Theodore Ts'o47def822010-07-27 11:56:05 -0400391 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700392}
393
394/* Allocate a new handle. This should probably be in a slab... */
395static handle_t *new_handle(int nblocks)
396{
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400397 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700398 if (!handle)
399 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700400 handle->h_buffer_credits = nblocks;
401 handle->h_ref = 1;
402
403 return handle;
404}
405
406/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700407 * handle_t *jbd2_journal_start() - Obtain a new handle.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700408 * @journal: Journal to start transaction on.
409 * @nblocks: number of block buffer we might modify
410 *
411 * We make sure that the transaction can guarantee at least nblocks of
412 * modified buffers in the log. We block until the log can guarantee
Jan Kara8f7d89f2013-06-04 12:35:11 -0400413 * that much space. Additionally, if rsv_blocks > 0, we also create another
414 * handle with rsv_blocks reserved blocks in the journal. This handle is
415 * is stored in h_rsv_handle. It is not attached to any particular transaction
416 * and thus doesn't block transaction commit. If the caller uses this reserved
417 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
418 * on the parent handle will dispose the reserved one. Reserved handle has to
419 * be converted to a normal handle using jbd2_journal_start_reserved() before
420 * it can be used.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700421 *
Eryu Guanc8675162011-05-24 17:09:58 -0400422 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
423 * on failure.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700424 */
Jan Kara8f7d89f2013-06-04 12:35:11 -0400425handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
426 gfp_t gfp_mask, unsigned int type,
427 unsigned int line_no)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700428{
429 handle_t *handle = journal_current_handle();
430 int err;
431
432 if (!journal)
433 return ERR_PTR(-EROFS);
434
435 if (handle) {
436 J_ASSERT(handle->h_transaction->t_journal == journal);
437 handle->h_ref++;
438 return handle;
439 }
440
441 handle = new_handle(nblocks);
442 if (!handle)
443 return ERR_PTR(-ENOMEM);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400444 if (rsv_blocks) {
445 handle_t *rsv_handle;
446
447 rsv_handle = new_handle(rsv_blocks);
448 if (!rsv_handle) {
449 jbd2_free_handle(handle);
450 return ERR_PTR(-ENOMEM);
451 }
452 rsv_handle->h_reserved = 1;
453 rsv_handle->h_journal = journal;
454 handle->h_rsv_handle = rsv_handle;
455 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700456
Theodore Ts'o47def822010-07-27 11:56:05 -0400457 err = start_this_handle(journal, handle, gfp_mask);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700458 if (err < 0) {
Jan Kara8f7d89f2013-06-04 12:35:11 -0400459 if (handle->h_rsv_handle)
460 jbd2_free_handle(handle->h_rsv_handle);
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400461 jbd2_free_handle(handle);
Dmitry Monakhovdf05c1b82013-03-02 17:08:46 -0500462 return ERR_PTR(err);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700463 }
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500464 handle->h_type = type;
465 handle->h_line_no = line_no;
466 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
467 handle->h_transaction->t_tid, type,
468 line_no, nblocks);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700469 return handle;
470}
Theodore Ts'o47def822010-07-27 11:56:05 -0400471EXPORT_SYMBOL(jbd2__journal_start);
472
473
474handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
475{
Jan Kara8f7d89f2013-06-04 12:35:11 -0400476 return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
Theodore Ts'o47def822010-07-27 11:56:05 -0400477}
478EXPORT_SYMBOL(jbd2_journal_start);
479
Jan Kara8f7d89f2013-06-04 12:35:11 -0400480void jbd2_journal_free_reserved(handle_t *handle)
481{
482 journal_t *journal = handle->h_journal;
483
484 WARN_ON(!handle->h_reserved);
485 sub_reserved_credits(journal, handle->h_buffer_credits);
486 jbd2_free_handle(handle);
487}
488EXPORT_SYMBOL(jbd2_journal_free_reserved);
489
490/**
491 * int jbd2_journal_start_reserved(handle_t *handle) - start reserved handle
492 * @handle: handle to start
493 *
494 * Start handle that has been previously reserved with jbd2_journal_reserve().
495 * This attaches @handle to the running transaction (or creates one if there's
496 * not transaction running). Unlike jbd2_journal_start() this function cannot
497 * block on journal commit, checkpointing, or similar stuff. It can block on
498 * memory allocation or frozen journal though.
499 *
500 * Return 0 on success, non-zero on error - handle is freed in that case.
501 */
502int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
503 unsigned int line_no)
504{
505 journal_t *journal = handle->h_journal;
506 int ret = -EIO;
507
508 if (WARN_ON(!handle->h_reserved)) {
509 /* Someone passed in normal handle? Just stop it. */
510 jbd2_journal_stop(handle);
511 return ret;
512 }
513 /*
514 * Usefulness of mixing of reserved and unreserved handles is
515 * questionable. So far nobody seems to need it so just error out.
516 */
517 if (WARN_ON(current->journal_info)) {
518 jbd2_journal_free_reserved(handle);
519 return ret;
520 }
521
522 handle->h_journal = NULL;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400523 /*
524 * GFP_NOFS is here because callers are likely from writeback or
525 * similarly constrained call sites
526 */
527 ret = start_this_handle(journal, handle, GFP_NOFS);
Dan Carpenter92e3b402014-02-17 20:33:01 -0500528 if (ret < 0) {
Jan Kara8f7d89f2013-06-04 12:35:11 -0400529 jbd2_journal_free_reserved(handle);
Dan Carpenter92e3b402014-02-17 20:33:01 -0500530 return ret;
531 }
Jan Kara8f7d89f2013-06-04 12:35:11 -0400532 handle->h_type = type;
533 handle->h_line_no = line_no;
Dan Carpenter92e3b402014-02-17 20:33:01 -0500534 return 0;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400535}
536EXPORT_SYMBOL(jbd2_journal_start_reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700537
538/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700539 * int jbd2_journal_extend() - extend buffer credits.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700540 * @handle: handle to 'extend'
541 * @nblocks: nr blocks to try to extend by.
542 *
543 * Some transactions, such as large extends and truncates, can be done
544 * atomically all at once or in several stages. The operation requests
Masanari Iidabd7ced92016-02-02 22:31:06 +0900545 * a credit for a number of buffer modifications in advance, but can
Dave Kleikamp470decc2006-10-11 01:20:57 -0700546 * extend its credit if it needs more.
547 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700548 * jbd2_journal_extend tries to give the running handle more buffer credits.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700549 * It does not guarantee that allocation - this is a best-effort only.
550 * The calling process MUST be able to deal cleanly with a failure to
551 * extend here.
552 *
553 * Return 0 on success, non-zero on failure.
554 *
555 * return code < 0 implies an error
556 * return code > 0 implies normal transaction-full status.
557 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700558int jbd2_journal_extend(handle_t *handle, int nblocks)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700559{
560 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400561 journal_t *journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700562 int result;
563 int wanted;
564
Dave Kleikamp470decc2006-10-11 01:20:57 -0700565 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400566 return -EROFS;
567 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700568
569 result = 1;
570
Theodore Ts'oa931da62010-08-03 21:35:12 -0400571 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700572
573 /* Don't extend a locked-down transaction! */
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400574 if (transaction->t_state != T_RUNNING) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700575 jbd_debug(3, "denied handle %p %d blocks: "
576 "transaction not running\n", handle, nblocks);
577 goto error_out;
578 }
579
580 spin_lock(&transaction->t_handle_lock);
Jan Karafe1e8db2013-06-04 12:22:15 -0400581 wanted = atomic_add_return(nblocks,
582 &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700583
584 if (wanted > journal->j_max_transaction_buffers) {
585 jbd_debug(3, "denied handle %p %d blocks: "
586 "transaction too large\n", handle, nblocks);
Jan Karafe1e8db2013-06-04 12:22:15 -0400587 atomic_sub(nblocks, &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700588 goto unlock;
589 }
590
Jan Kara76c39902013-06-04 12:12:57 -0400591 if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
592 jbd2_log_space_left(journal)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700593 jbd_debug(3, "denied handle %p %d blocks: "
594 "insufficient log space\n", handle, nblocks);
Jan Karafe1e8db2013-06-04 12:22:15 -0400595 atomic_sub(nblocks, &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700596 goto unlock;
597 }
598
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500599 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400600 transaction->t_tid,
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500601 handle->h_type, handle->h_line_no,
602 handle->h_buffer_credits,
603 nblocks);
604
Dave Kleikamp470decc2006-10-11 01:20:57 -0700605 handle->h_buffer_credits += nblocks;
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500606 handle->h_requested_credits += nblocks;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700607 result = 0;
608
609 jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
610unlock:
611 spin_unlock(&transaction->t_handle_lock);
612error_out:
Theodore Ts'oa931da62010-08-03 21:35:12 -0400613 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700614 return result;
615}
616
617
618/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700619 * int jbd2_journal_restart() - restart a handle .
Dave Kleikamp470decc2006-10-11 01:20:57 -0700620 * @handle: handle to restart
621 * @nblocks: nr credits requested
622 *
623 * Restart a handle for a multi-transaction filesystem
624 * operation.
625 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700626 * If the jbd2_journal_extend() call above fails to grant new buffer credits
627 * to a running handle, a call to jbd2_journal_restart will commit the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700628 * handle's transaction so far and reattach the handle to a new
Masanari Iidabd7ced92016-02-02 22:31:06 +0900629 * transaction capable of guaranteeing the requested number of
Jan Kara8f7d89f2013-06-04 12:35:11 -0400630 * credits. We preserve reserved handle if there's any attached to the
631 * passed in handle.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700632 */
Dan Carpenterd2159fb2011-09-04 10:20:14 -0400633int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700634{
635 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400636 journal_t *journal;
Theodore Ts'oe4471832011-02-12 08:18:24 -0500637 tid_t tid;
638 int need_to_start, ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700639
640 /* If we've had an abort of any type, don't even think about
641 * actually doing the restart! */
642 if (is_handle_aborted(handle))
643 return 0;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400644 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700645
646 /*
647 * First unlink the handle from its current transaction, and start the
648 * commit on that.
649 */
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400650 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700651 J_ASSERT(journal_current_handle() == handle);
652
Theodore Ts'oa931da62010-08-03 21:35:12 -0400653 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700654 spin_lock(&transaction->t_handle_lock);
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400655 atomic_sub(handle->h_buffer_credits,
656 &transaction->t_outstanding_credits);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400657 if (handle->h_rsv_handle) {
658 sub_reserved_credits(journal,
659 handle->h_rsv_handle->h_buffer_credits);
660 }
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400661 if (atomic_dec_and_test(&transaction->t_updates))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700662 wake_up(&journal->j_wait_updates);
Theodore Ts'o39c04152013-07-01 08:12:40 -0400663 tid = transaction->t_tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700664 spin_unlock(&transaction->t_handle_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400665 handle->h_transaction = NULL;
666 current->journal_info = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700667
668 jbd_debug(2, "restarting handle %p\n", handle);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500669 need_to_start = !tid_geq(journal->j_commit_request, tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400670 read_unlock(&journal->j_state_lock);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500671 if (need_to_start)
672 jbd2_log_start_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700673
Jan Karaab714af2016-06-30 11:39:38 -0400674 rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700675 handle->h_buffer_credits = nblocks;
Theodore Ts'o47def822010-07-27 11:56:05 -0400676 ret = start_this_handle(journal, handle, gfp_mask);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700677 return ret;
678}
Theodore Ts'o47def822010-07-27 11:56:05 -0400679EXPORT_SYMBOL(jbd2__journal_restart);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700680
681
Theodore Ts'o47def822010-07-27 11:56:05 -0400682int jbd2_journal_restart(handle_t *handle, int nblocks)
683{
684 return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
685}
686EXPORT_SYMBOL(jbd2_journal_restart);
687
Dave Kleikamp470decc2006-10-11 01:20:57 -0700688/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700689 * void jbd2_journal_lock_updates () - establish a transaction barrier.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700690 * @journal: Journal to establish a barrier on.
691 *
692 * This locks out any further updates from being started, and blocks
693 * until all existing updates have completed, returning only once the
694 * journal is in a quiescent state with no updates running.
695 *
696 * The journal lock should not be held on entry.
697 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700698void jbd2_journal_lock_updates(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700699{
700 DEFINE_WAIT(wait);
701
Jan Kara1eaa5662016-06-30 11:40:54 -0400702 jbd2_might_wait_for_commit(journal);
703
Theodore Ts'oa931da62010-08-03 21:35:12 -0400704 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700705 ++journal->j_barrier_count;
706
Jan Kara8f7d89f2013-06-04 12:35:11 -0400707 /* Wait until there are no reserved handles */
708 if (atomic_read(&journal->j_reserved_credits)) {
709 write_unlock(&journal->j_state_lock);
710 wait_event(journal->j_wait_reserved,
711 atomic_read(&journal->j_reserved_credits) == 0);
712 write_lock(&journal->j_state_lock);
713 }
714
Dave Kleikamp470decc2006-10-11 01:20:57 -0700715 /* Wait until there are no running updates */
716 while (1) {
717 transaction_t *transaction = journal->j_running_transaction;
718
719 if (!transaction)
720 break;
721
722 spin_lock(&transaction->t_handle_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700723 prepare_to_wait(&journal->j_wait_updates, &wait,
724 TASK_UNINTERRUPTIBLE);
Jan Kara9837d8e2012-01-04 22:03:11 -0500725 if (!atomic_read(&transaction->t_updates)) {
726 spin_unlock(&transaction->t_handle_lock);
727 finish_wait(&journal->j_wait_updates, &wait);
728 break;
729 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700730 spin_unlock(&transaction->t_handle_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400731 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700732 schedule();
733 finish_wait(&journal->j_wait_updates, &wait);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400734 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700735 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400736 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700737
738 /*
739 * We have now established a barrier against other normal updates, but
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700740 * we also need to barrier against other jbd2_journal_lock_updates() calls
Dave Kleikamp470decc2006-10-11 01:20:57 -0700741 * to make sure that we serialise special journal-locked operations
742 * too.
743 */
744 mutex_lock(&journal->j_barrier);
745}
746
747/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700748 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
Dave Kleikamp470decc2006-10-11 01:20:57 -0700749 * @journal: Journal to release the barrier on.
750 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700751 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
Dave Kleikamp470decc2006-10-11 01:20:57 -0700752 *
753 * Should be called without the journal lock held.
754 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700755void jbd2_journal_unlock_updates (journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700756{
757 J_ASSERT(journal->j_barrier_count != 0);
758
759 mutex_unlock(&journal->j_barrier);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400760 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700761 --journal->j_barrier_count;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400762 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700763 wake_up(&journal->j_wait_transaction_locked);
764}
765
Jan Karaf91d1d02009-07-13 16:16:20 -0400766static void warn_dirty_buffer(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700767{
Jan Karaf91d1d02009-07-13 16:16:20 -0400768 printk(KERN_WARNING
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400769 "JBD2: Spotted dirty metadata buffer (dev = %pg, blocknr = %llu). "
Jan Karaf91d1d02009-07-13 16:16:20 -0400770 "There's a risk of filesystem corruption in case of system "
771 "crash.\n",
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400772 bh->b_bdev, (unsigned long long)bh->b_blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700773}
774
Jan Karaee57aba2015-06-08 12:39:07 -0400775/* Call t_frozen trigger and copy buffer data into jh->b_frozen_data. */
776static void jbd2_freeze_jh_data(struct journal_head *jh)
777{
778 struct page *page;
779 int offset;
780 char *source;
781 struct buffer_head *bh = jh2bh(jh);
782
783 J_EXPECT_JH(jh, buffer_uptodate(bh), "Possible IO failure.\n");
784 page = bh->b_page;
785 offset = offset_in_page(bh->b_data);
786 source = kmap_atomic(page);
787 /* Fire data frozen trigger just before we copy the data */
788 jbd2_buffer_frozen_trigger(jh, source + offset, jh->b_triggers);
789 memcpy(jh->b_frozen_data, source + offset, bh->b_size);
790 kunmap_atomic(source);
791
792 /*
793 * Now that the frozen data is saved off, we need to store any matching
794 * triggers.
795 */
796 jh->b_frozen_triggers = jh->b_triggers;
797}
798
Dave Kleikamp470decc2006-10-11 01:20:57 -0700799/*
800 * If the buffer is already part of the current transaction, then there
801 * is nothing we need to do. If it is already part of a prior
802 * transaction which we are still committing to disk, then we need to
803 * make sure that we do not overwrite the old copy: we do copy-out to
804 * preserve the copy going to disk. We also account the buffer against
805 * the handle's metadata buffer credits (unless the buffer is already
806 * part of the transaction, that is).
807 *
808 */
809static int
810do_get_write_access(handle_t *handle, struct journal_head *jh,
811 int force_copy)
812{
813 struct buffer_head *bh;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400814 transaction_t *transaction = handle->h_transaction;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700815 journal_t *journal;
816 int error;
817 char *frozen_buffer = NULL;
Theodore Ts'of783f092013-04-21 16:47:54 -0400818 unsigned long start_lock, time_lock;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700819
820 if (is_handle_aborted(handle))
821 return -EROFS;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700822 journal = transaction->t_journal;
823
Theodore Ts'ocfef2c62010-12-18 13:07:34 -0500824 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700825
826 JBUFFER_TRACE(jh, "entry");
827repeat:
828 bh = jh2bh(jh);
829
830 /* @@@ Need to check for errors here at some point. */
831
Theodore Ts'of783f092013-04-21 16:47:54 -0400832 start_lock = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700833 lock_buffer(bh);
834 jbd_lock_bh_state(bh);
835
Theodore Ts'of783f092013-04-21 16:47:54 -0400836 /* If it takes too long to lock the buffer, trace it */
837 time_lock = jbd2_time_diff(start_lock, jiffies);
838 if (time_lock > HZ/10)
839 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
840 jiffies_to_msecs(time_lock));
841
Dave Kleikamp470decc2006-10-11 01:20:57 -0700842 /* We now hold the buffer lock so it is safe to query the buffer
843 * state. Is the buffer dirty?
844 *
845 * If so, there are two possibilities. The buffer may be
846 * non-journaled, and undergoing a quite legitimate writeback.
847 * Otherwise, it is journaled, and we don't expect dirty buffers
848 * in that state (the buffers should be marked JBD_Dirty
849 * instead.) So either the IO is being done under our own
850 * control and this is a bug, or it's a third party IO such as
851 * dump(8) (which may leave the buffer scheduled for read ---
852 * ie. locked but not dirty) or tune2fs (which may actually have
853 * the buffer dirtied, ugh.) */
854
855 if (buffer_dirty(bh)) {
856 /*
857 * First question: is this buffer already part of the current
858 * transaction or the existing committing transaction?
859 */
860 if (jh->b_transaction) {
861 J_ASSERT_JH(jh,
862 jh->b_transaction == transaction ||
863 jh->b_transaction ==
864 journal->j_committing_transaction);
865 if (jh->b_next_transaction)
866 J_ASSERT_JH(jh, jh->b_next_transaction ==
867 transaction);
Jan Karaf91d1d02009-07-13 16:16:20 -0400868 warn_dirty_buffer(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700869 }
870 /*
871 * In any case we need to clean the dirty flag and we must
872 * do it under the buffer lock to be sure we don't race
873 * with running write-out.
874 */
Jan Karaf91d1d02009-07-13 16:16:20 -0400875 JBUFFER_TRACE(jh, "Journalling dirty buffer");
876 clear_buffer_dirty(bh);
877 set_buffer_jbddirty(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700878 }
879
880 unlock_buffer(bh);
881
882 error = -EROFS;
883 if (is_handle_aborted(handle)) {
884 jbd_unlock_bh_state(bh);
885 goto out;
886 }
887 error = 0;
888
889 /*
890 * The buffer is already part of this transaction if b_transaction or
891 * b_next_transaction points to it
892 */
893 if (jh->b_transaction == transaction ||
894 jh->b_next_transaction == transaction)
895 goto done;
896
897 /*
Josef Bacik9fc7c632008-04-17 10:38:59 -0400898 * this is the first time this transaction is touching this buffer,
899 * reset the modified flag
900 */
901 jh->b_modified = 0;
902
903 /*
Jan Kara8b00f402015-06-08 12:44:21 -0400904 * If the buffer is not journaled right now, we need to make sure it
905 * doesn't get written to disk before the caller actually commits the
906 * new data
907 */
908 if (!jh->b_transaction) {
909 JBUFFER_TRACE(jh, "no transaction");
910 J_ASSERT_JH(jh, !jh->b_next_transaction);
911 JBUFFER_TRACE(jh, "file as BJ_Reserved");
Jan Karade92c8c2015-06-08 12:46:37 -0400912 /*
913 * Make sure all stores to jh (b_modified, b_frozen_data) are
914 * visible before attaching it to the running transaction.
915 * Paired with barrier in jbd2_write_access_granted()
916 */
917 smp_wmb();
Jan Kara8b00f402015-06-08 12:44:21 -0400918 spin_lock(&journal->j_list_lock);
919 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
920 spin_unlock(&journal->j_list_lock);
921 goto done;
922 }
923 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700924 * If there is already a copy-out version of this buffer, then we don't
925 * need to make another one
926 */
927 if (jh->b_frozen_data) {
928 JBUFFER_TRACE(jh, "has frozen data");
929 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
Jan Karade92c8c2015-06-08 12:46:37 -0400930 goto attach_next;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700931 }
932
Jan Kara8b00f402015-06-08 12:44:21 -0400933 JBUFFER_TRACE(jh, "owned by older transaction");
934 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
935 J_ASSERT_JH(jh, jh->b_transaction == journal->j_committing_transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700936
937 /*
Jan Kara8b00f402015-06-08 12:44:21 -0400938 * There is one case we have to be very careful about. If the
939 * committing transaction is currently writing this buffer out to disk
940 * and has NOT made a copy-out, then we cannot modify the buffer
941 * contents at all right now. The essence of copy-out is that it is
942 * the extra copy, not the primary copy, which gets journaled. If the
943 * primary copy is already going to disk then we cannot do copy-out
944 * here.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700945 */
Jan Kara8b00f402015-06-08 12:44:21 -0400946 if (buffer_shadow(bh)) {
947 JBUFFER_TRACE(jh, "on shadow: sleep");
948 jbd_unlock_bh_state(bh);
949 wait_on_bit_io(&bh->b_state, BH_Shadow, TASK_UNINTERRUPTIBLE);
950 goto repeat;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700951 }
952
Jan Kara8b00f402015-06-08 12:44:21 -0400953 /*
954 * Only do the copy if the currently-owning transaction still needs it.
955 * If buffer isn't on BJ_Metadata list, the committing transaction is
956 * past that stage (here we use the fact that BH_Shadow is set under
957 * bh_state lock together with refiling to BJ_Shadow list and at this
958 * point we know the buffer doesn't have BH_Shadow set).
959 *
960 * Subtle point, though: if this is a get_undo_access, then we will be
961 * relying on the frozen_data to contain the new value of the
962 * committed_data record after the transaction, so we HAVE to force the
963 * frozen_data copy in that case.
964 */
965 if (jh->b_jlist == BJ_Metadata || force_copy) {
966 JBUFFER_TRACE(jh, "generate frozen data");
967 if (!frozen_buffer) {
968 JBUFFER_TRACE(jh, "allocate memory for buffer");
969 jbd_unlock_bh_state(bh);
Michal Hocko490c1b42016-03-13 17:38:20 -0400970 frozen_buffer = jbd2_alloc(jh2bh(jh)->b_size,
971 GFP_NOFS | __GFP_NOFAIL);
Jan Kara8b00f402015-06-08 12:44:21 -0400972 goto repeat;
973 }
974 jh->b_frozen_data = frozen_buffer;
975 frozen_buffer = NULL;
976 jbd2_freeze_jh_data(jh);
977 }
Jan Karade92c8c2015-06-08 12:46:37 -0400978attach_next:
979 /*
980 * Make sure all stores to jh (b_modified, b_frozen_data) are visible
981 * before attaching it to the running transaction. Paired with barrier
982 * in jbd2_write_access_granted()
983 */
984 smp_wmb();
Jan Kara8b00f402015-06-08 12:44:21 -0400985 jh->b_next_transaction = transaction;
986
Dave Kleikamp470decc2006-10-11 01:20:57 -0700987done:
Dave Kleikamp470decc2006-10-11 01:20:57 -0700988 jbd_unlock_bh_state(bh);
989
990 /*
991 * If we are about to journal a buffer, then any revoke pending on it is
992 * no longer valid
993 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700994 jbd2_journal_cancel_revoke(handle, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700995
996out:
997 if (unlikely(frozen_buffer)) /* It's usually NULL */
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400998 jbd2_free(frozen_buffer, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700999
1000 JBUFFER_TRACE(jh, "exit");
1001 return error;
1002}
1003
Jan Karade92c8c2015-06-08 12:46:37 -04001004/* Fast check whether buffer is already attached to the required transaction */
Junxiao Bi087ffd42015-12-04 12:29:28 -05001005static bool jbd2_write_access_granted(handle_t *handle, struct buffer_head *bh,
1006 bool undo)
Jan Karade92c8c2015-06-08 12:46:37 -04001007{
1008 struct journal_head *jh;
1009 bool ret = false;
1010
1011 /* Dirty buffers require special handling... */
1012 if (buffer_dirty(bh))
1013 return false;
1014
1015 /*
1016 * RCU protects us from dereferencing freed pages. So the checks we do
1017 * are guaranteed not to oops. However the jh slab object can get freed
1018 * & reallocated while we work with it. So we have to be careful. When
1019 * we see jh attached to the running transaction, we know it must stay
1020 * so until the transaction is committed. Thus jh won't be freed and
1021 * will be attached to the same bh while we run. However it can
1022 * happen jh gets freed, reallocated, and attached to the transaction
1023 * just after we get pointer to it from bh. So we have to be careful
1024 * and recheck jh still belongs to our bh before we return success.
1025 */
1026 rcu_read_lock();
1027 if (!buffer_jbd(bh))
1028 goto out;
1029 /* This should be bh2jh() but that doesn't work with inline functions */
1030 jh = READ_ONCE(bh->b_private);
1031 if (!jh)
1032 goto out;
Junxiao Bi087ffd42015-12-04 12:29:28 -05001033 /* For undo access buffer must have data copied */
1034 if (undo && !jh->b_committed_data)
1035 goto out;
Jan Karade92c8c2015-06-08 12:46:37 -04001036 if (jh->b_transaction != handle->h_transaction &&
1037 jh->b_next_transaction != handle->h_transaction)
1038 goto out;
1039 /*
1040 * There are two reasons for the barrier here:
1041 * 1) Make sure to fetch b_bh after we did previous checks so that we
1042 * detect when jh went through free, realloc, attach to transaction
1043 * while we were checking. Paired with implicit barrier in that path.
1044 * 2) So that access to bh done after jbd2_write_access_granted()
1045 * doesn't get reordered and see inconsistent state of concurrent
1046 * do_get_write_access().
1047 */
1048 smp_mb();
1049 if (unlikely(jh->b_bh != bh))
1050 goto out;
1051 ret = true;
1052out:
1053 rcu_read_unlock();
1054 return ret;
1055}
1056
Dave Kleikamp470decc2006-10-11 01:20:57 -07001057/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001058 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001059 * @handle: transaction to add buffer modifications to
1060 * @bh: bh to be used for metadata writes
Dave Kleikamp470decc2006-10-11 01:20:57 -07001061 *
1062 * Returns an error code or 0 on success.
1063 *
1064 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1065 * because we're write()ing a buffer which is also part of a shared mapping.
1066 */
1067
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001068int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001069{
Jan Karade92c8c2015-06-08 12:46:37 -04001070 struct journal_head *jh;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001071 int rc;
1072
Junxiao Bi087ffd42015-12-04 12:29:28 -05001073 if (jbd2_write_access_granted(handle, bh, false))
Jan Karade92c8c2015-06-08 12:46:37 -04001074 return 0;
1075
1076 jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001077 /* We do not want to get caught playing with fields which the
1078 * log thread also manipulates. Make sure that the buffer
1079 * completes any outstanding IO before proceeding. */
1080 rc = do_get_write_access(handle, jh, 0);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001081 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001082 return rc;
1083}
1084
1085
1086/*
1087 * When the user wants to journal a newly created buffer_head
1088 * (ie. getblk() returned a new buffer and we are going to populate it
1089 * manually rather than reading off disk), then we need to keep the
1090 * buffer_head locked until it has been completely filled with new
1091 * data. In this case, we should be able to make the assertion that
1092 * the bh is not already part of an existing transaction.
1093 *
1094 * The buffer should already be locked by the caller by this point.
1095 * There is no lock ranking violation: it was a newly created,
1096 * unlocked buffer beforehand. */
1097
1098/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001099 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
Dave Kleikamp470decc2006-10-11 01:20:57 -07001100 * @handle: transaction to new buffer to
1101 * @bh: new buffer.
1102 *
1103 * Call this if you create a new bh.
1104 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001105int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001106{
1107 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001108 journal_t *journal;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001109 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001110 int err;
1111
1112 jbd_debug(5, "journal_head %p\n", jh);
1113 err = -EROFS;
1114 if (is_handle_aborted(handle))
1115 goto out;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001116 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001117 err = 0;
1118
1119 JBUFFER_TRACE(jh, "entry");
1120 /*
1121 * The buffer may already belong to this transaction due to pre-zeroing
1122 * in the filesystem's new_block code. It may also be on the previous,
1123 * committing transaction's lists, but it HAS to be in Forget state in
1124 * that case: the transaction must have deleted the buffer for it to be
1125 * reused here.
1126 */
1127 jbd_lock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001128 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1129 jh->b_transaction == NULL ||
1130 (jh->b_transaction == journal->j_committing_transaction &&
1131 jh->b_jlist == BJ_Forget)));
1132
1133 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1134 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1135
1136 if (jh->b_transaction == NULL) {
Jan Karaf91d1d02009-07-13 16:16:20 -04001137 /*
1138 * Previous jbd2_journal_forget() could have left the buffer
1139 * with jbddirty bit set because it was being committed. When
1140 * the commit finished, we've filed the buffer for
1141 * checkpointing and marked it dirty. Now we are reallocating
1142 * the buffer so the transaction freeing it must have
1143 * committed and so it's safe to clear the dirty bit.
1144 */
1145 clear_buffer_dirty(jh2bh(jh));
Josef Bacik9fc7c632008-04-17 10:38:59 -04001146 /* first access by this transaction */
1147 jh->b_modified = 0;
1148
Dave Kleikamp470decc2006-10-11 01:20:57 -07001149 JBUFFER_TRACE(jh, "file as BJ_Reserved");
Theodore Ts'o6e4862a2014-03-09 00:46:23 -05001150 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001151 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
Taesoo Kim559cce62016-10-12 23:19:18 -04001152 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001153 } else if (jh->b_transaction == journal->j_committing_transaction) {
Josef Bacik9fc7c632008-04-17 10:38:59 -04001154 /* first access by this transaction */
1155 jh->b_modified = 0;
1156
Dave Kleikamp470decc2006-10-11 01:20:57 -07001157 JBUFFER_TRACE(jh, "set next transaction");
Theodore Ts'o6e4862a2014-03-09 00:46:23 -05001158 spin_lock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001159 jh->b_next_transaction = transaction;
Taesoo Kim559cce62016-10-12 23:19:18 -04001160 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001161 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001162 jbd_unlock_bh_state(bh);
1163
1164 /*
1165 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1166 * blocks which contain freed but then revoked metadata. We need
1167 * to cancel the revoke in case we end up freeing it yet again
1168 * and the reallocating as data - this would cause a second revoke,
1169 * which hits an assertion error.
1170 */
1171 JBUFFER_TRACE(jh, "cancelling revoke");
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001172 jbd2_journal_cancel_revoke(handle, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001173out:
Ding Dinghua3991b402011-05-25 17:43:48 -04001174 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001175 return err;
1176}
1177
1178/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001179 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
Dave Kleikamp470decc2006-10-11 01:20:57 -07001180 * non-rewindable consequences
1181 * @handle: transaction
1182 * @bh: buffer to undo
Dave Kleikamp470decc2006-10-11 01:20:57 -07001183 *
1184 * Sometimes there is a need to distinguish between metadata which has
1185 * been committed to disk and that which has not. The ext3fs code uses
1186 * this for freeing and allocating space, we have to make sure that we
1187 * do not reuse freed space until the deallocation has been committed,
1188 * since if we overwrote that space we would make the delete
1189 * un-rewindable in case of a crash.
1190 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001191 * To deal with that, jbd2_journal_get_undo_access requests write access to a
Dave Kleikamp470decc2006-10-11 01:20:57 -07001192 * buffer for parts of non-rewindable operations such as delete
1193 * operations on the bitmaps. The journaling code must keep a copy of
1194 * the buffer's contents prior to the undo_access call until such time
1195 * as we know that the buffer has definitely been committed to disk.
1196 *
1197 * We never need to know which transaction the committed data is part
1198 * of, buffers touched here are guaranteed to be dirtied later and so
1199 * will be committed to a new transaction in due course, at which point
1200 * we can discard the old committed data pointer.
1201 *
1202 * Returns error number or 0 on success.
1203 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001204int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001205{
1206 int err;
Jan Karade92c8c2015-06-08 12:46:37 -04001207 struct journal_head *jh;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001208 char *committed_data = NULL;
1209
1210 JBUFFER_TRACE(jh, "entry");
Junxiao Bi087ffd42015-12-04 12:29:28 -05001211 if (jbd2_write_access_granted(handle, bh, true))
Jan Karade92c8c2015-06-08 12:46:37 -04001212 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001213
Jan Karade92c8c2015-06-08 12:46:37 -04001214 jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001215 /*
1216 * Do this first --- it can drop the journal lock, so we want to
1217 * make sure that obtaining the committed_data is done
1218 * atomically wrt. completion of any outstanding commits.
1219 */
1220 err = do_get_write_access(handle, jh, 1);
1221 if (err)
1222 goto out;
1223
1224repeat:
Michal Hocko490c1b42016-03-13 17:38:20 -04001225 if (!jh->b_committed_data)
1226 committed_data = jbd2_alloc(jh2bh(jh)->b_size,
1227 GFP_NOFS|__GFP_NOFAIL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001228
1229 jbd_lock_bh_state(bh);
1230 if (!jh->b_committed_data) {
1231 /* Copy out the current buffer contents into the
1232 * preserved, committed copy. */
1233 JBUFFER_TRACE(jh, "generate b_committed data");
1234 if (!committed_data) {
1235 jbd_unlock_bh_state(bh);
1236 goto repeat;
1237 }
1238
1239 jh->b_committed_data = committed_data;
1240 committed_data = NULL;
1241 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1242 }
1243 jbd_unlock_bh_state(bh);
1244out:
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001245 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001246 if (unlikely(committed_data))
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001247 jbd2_free(committed_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001248 return err;
1249}
1250
1251/**
Joel Beckere06c8222008-09-11 15:35:47 -07001252 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1253 * @bh: buffer to trigger on
1254 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1255 *
1256 * Set any triggers on this journal_head. This is always safe, because
1257 * triggers for a committing buffer will be saved off, and triggers for
1258 * a running transaction will match the buffer in that transaction.
1259 *
1260 * Call with NULL to clear the triggers.
1261 */
1262void jbd2_journal_set_triggers(struct buffer_head *bh,
1263 struct jbd2_buffer_trigger_type *type)
1264{
Jan Karaad56eda2013-03-11 13:24:56 -04001265 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
Joel Beckere06c8222008-09-11 15:35:47 -07001266
Jan Karaad56eda2013-03-11 13:24:56 -04001267 if (WARN_ON(!jh))
1268 return;
Joel Beckere06c8222008-09-11 15:35:47 -07001269 jh->b_triggers = type;
Jan Karaad56eda2013-03-11 13:24:56 -04001270 jbd2_journal_put_journal_head(jh);
Joel Beckere06c8222008-09-11 15:35:47 -07001271}
1272
Jan Kara13ceef02010-07-14 07:56:33 +02001273void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
Joel Beckere06c8222008-09-11 15:35:47 -07001274 struct jbd2_buffer_trigger_type *triggers)
1275{
1276 struct buffer_head *bh = jh2bh(jh);
1277
Jan Kara13ceef02010-07-14 07:56:33 +02001278 if (!triggers || !triggers->t_frozen)
Joel Beckere06c8222008-09-11 15:35:47 -07001279 return;
1280
Jan Kara13ceef02010-07-14 07:56:33 +02001281 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
Joel Beckere06c8222008-09-11 15:35:47 -07001282}
1283
1284void jbd2_buffer_abort_trigger(struct journal_head *jh,
1285 struct jbd2_buffer_trigger_type *triggers)
1286{
1287 if (!triggers || !triggers->t_abort)
1288 return;
1289
1290 triggers->t_abort(triggers, jh2bh(jh));
1291}
1292
Joel Beckere06c8222008-09-11 15:35:47 -07001293/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001294 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
Dave Kleikamp470decc2006-10-11 01:20:57 -07001295 * @handle: transaction to add buffer to.
1296 * @bh: buffer to mark
1297 *
1298 * mark dirty metadata which needs to be journaled as part of the current
1299 * transaction.
1300 *
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001301 * The buffer must have previously had jbd2_journal_get_write_access()
1302 * called so that it has a valid journal_head attached to the buffer
1303 * head.
1304 *
Dave Kleikamp470decc2006-10-11 01:20:57 -07001305 * The buffer is placed on the transaction's metadata list and is marked
1306 * as belonging to the transaction.
1307 *
1308 * Returns error number or 0 on success.
1309 *
1310 * Special care needs to be taken if the buffer already belongs to the
1311 * current committing transaction (in which case we should have frozen
1312 * data present for that commit). In that case, we don't relink the
1313 * buffer: that only gets done when the old transaction finally
1314 * completes its commit.
1315 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001316int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001317{
1318 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001319 journal_t *journal;
Jan Karaad56eda2013-03-11 13:24:56 -04001320 struct journal_head *jh;
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001321 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001322
Dave Kleikamp470decc2006-10-11 01:20:57 -07001323 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001324 return -EROFS;
Jan Kara6e06ae82015-07-12 18:11:30 -04001325 if (!buffer_jbd(bh)) {
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001326 ret = -EUCLEAN;
1327 goto out;
1328 }
Jan Kara6e06ae82015-07-12 18:11:30 -04001329 /*
1330 * We don't grab jh reference here since the buffer must be part
1331 * of the running transaction.
1332 */
1333 jh = bh2jh(bh);
1334 /*
1335 * This and the following assertions are unreliable since we may see jh
1336 * in inconsistent state unless we grab bh_state lock. But this is
1337 * crucial to catch bugs so let's do a reliable check until the
1338 * lockless handling is fully proven.
1339 */
1340 if (jh->b_transaction != transaction &&
1341 jh->b_next_transaction != transaction) {
1342 jbd_lock_bh_state(bh);
1343 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
1344 jh->b_next_transaction == transaction);
1345 jbd_unlock_bh_state(bh);
1346 }
1347 if (jh->b_modified == 1) {
1348 /* If it's in our transaction it must be in BJ_Metadata list. */
1349 if (jh->b_transaction == transaction &&
1350 jh->b_jlist != BJ_Metadata) {
1351 jbd_lock_bh_state(bh);
1352 J_ASSERT_JH(jh, jh->b_transaction != transaction ||
1353 jh->b_jlist == BJ_Metadata);
1354 jbd_unlock_bh_state(bh);
1355 }
1356 goto out;
1357 }
1358
1359 journal = transaction->t_journal;
Jan Karaad56eda2013-03-11 13:24:56 -04001360 jbd_debug(5, "journal_head %p\n", jh);
1361 JBUFFER_TRACE(jh, "entry");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001362
1363 jbd_lock_bh_state(bh);
1364
1365 if (jh->b_modified == 0) {
1366 /*
1367 * This buffer's got modified and becoming part
1368 * of the transaction. This needs to be done
1369 * once a transaction -bzzz
1370 */
1371 jh->b_modified = 1;
Theodore Ts'of6c07ca2013-12-08 21:12:59 -05001372 if (handle->h_buffer_credits <= 0) {
1373 ret = -ENOSPC;
1374 goto out_unlock_bh;
1375 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001376 handle->h_buffer_credits--;
1377 }
1378
1379 /*
1380 * fastpath, to avoid expensive locking. If this buffer is already
1381 * on the running transaction's metadata list there is nothing to do.
1382 * Nobody can take it off again because there is a handle open.
1383 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1384 * result in this test being false, so we go in and take the locks.
1385 */
1386 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1387 JBUFFER_TRACE(jh, "fastpath");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001388 if (unlikely(jh->b_transaction !=
1389 journal->j_running_transaction)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001390 printk(KERN_ERR "JBD2: %s: "
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001391 "jh->b_transaction (%llu, %p, %u) != "
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001392 "journal->j_running_transaction (%p, %u)\n",
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001393 journal->j_devname,
1394 (unsigned long long) bh->b_blocknr,
1395 jh->b_transaction,
1396 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1397 journal->j_running_transaction,
1398 journal->j_running_transaction ?
1399 journal->j_running_transaction->t_tid : 0);
1400 ret = -EINVAL;
1401 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001402 goto out_unlock_bh;
1403 }
1404
1405 set_buffer_jbddirty(bh);
1406
1407 /*
1408 * Metadata already on the current transaction list doesn't
1409 * need to be filed. Metadata on another transaction's list must
1410 * be committing, and will be refiled once the commit completes:
1411 * leave it alone for now.
1412 */
1413 if (jh->b_transaction != transaction) {
1414 JBUFFER_TRACE(jh, "already on other transaction");
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001415 if (unlikely(((jh->b_transaction !=
1416 journal->j_committing_transaction)) ||
1417 (jh->b_next_transaction != transaction))) {
1418 printk(KERN_ERR "jbd2_journal_dirty_metadata: %s: "
1419 "bad jh for block %llu: "
1420 "transaction (%p, %u), "
1421 "jh->b_transaction (%p, %u), "
1422 "jh->b_next_transaction (%p, %u), jlist %u\n",
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001423 journal->j_devname,
1424 (unsigned long long) bh->b_blocknr,
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001425 transaction, transaction->t_tid,
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001426 jh->b_transaction,
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001427 jh->b_transaction ?
1428 jh->b_transaction->t_tid : 0,
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001429 jh->b_next_transaction,
1430 jh->b_next_transaction ?
1431 jh->b_next_transaction->t_tid : 0,
Theodore Ts'o66a4cb12014-03-12 16:38:03 -04001432 jh->b_jlist);
1433 WARN_ON(1);
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001434 ret = -EINVAL;
1435 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001436 /* And this case is illegal: we can't reuse another
1437 * transaction's data buffer, ever. */
1438 goto out_unlock_bh;
1439 }
1440
1441 /* That test should have eliminated the following case: */
Mingming Cao40191912008-01-28 23:58:27 -05001442 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001443
1444 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1445 spin_lock(&journal->j_list_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001446 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001447 spin_unlock(&journal->j_list_lock);
1448out_unlock_bh:
1449 jbd_unlock_bh_state(bh);
1450out:
1451 JBUFFER_TRACE(jh, "exit");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001452 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001453}
1454
Dave Kleikamp470decc2006-10-11 01:20:57 -07001455/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001456 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001457 * @handle: transaction handle
1458 * @bh: bh to 'forget'
1459 *
1460 * We can only do the bforget if there are no commits pending against the
1461 * buffer. If the buffer is dirty in the current running transaction we
1462 * can safely unlink it.
1463 *
1464 * bh may not be a journalled buffer at all - it may be a non-JBD
1465 * buffer which came off the hashtable. Check for this.
1466 *
1467 * Decrements bh->b_count by one.
1468 *
1469 * Allow this call even if the handle has aborted --- it may be part of
1470 * the caller's cleanup after an abort.
1471 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001472int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001473{
1474 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001475 journal_t *journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001476 struct journal_head *jh;
1477 int drop_reserve = 0;
1478 int err = 0;
Josef Bacik1dfc3222008-04-17 10:38:59 -04001479 int was_modified = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001480
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001481 if (is_handle_aborted(handle))
1482 return -EROFS;
1483 journal = transaction->t_journal;
1484
Dave Kleikamp470decc2006-10-11 01:20:57 -07001485 BUFFER_TRACE(bh, "entry");
1486
1487 jbd_lock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001488
1489 if (!buffer_jbd(bh))
1490 goto not_jbd;
1491 jh = bh2jh(bh);
1492
1493 /* Critical error: attempting to delete a bitmap buffer, maybe?
1494 * Don't do any jbd operations, and return an error. */
1495 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1496 "inconsistent data on disk")) {
1497 err = -EIO;
1498 goto not_jbd;
1499 }
1500
Adam Buchbinder48fc7f72012-09-19 21:48:00 -04001501 /* keep track of whether or not this transaction modified us */
Josef Bacik1dfc3222008-04-17 10:38:59 -04001502 was_modified = jh->b_modified;
1503
Dave Kleikamp470decc2006-10-11 01:20:57 -07001504 /*
1505 * The buffer's going from the transaction, we must drop
1506 * all references -bzzz
1507 */
1508 jh->b_modified = 0;
1509
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001510 if (jh->b_transaction == transaction) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001511 J_ASSERT_JH(jh, !jh->b_frozen_data);
1512
1513 /* If we are forgetting a buffer which is already part
1514 * of this transaction, then we can just drop it from
1515 * the transaction immediately. */
1516 clear_buffer_dirty(bh);
1517 clear_buffer_jbddirty(bh);
1518
1519 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1520
Josef Bacik1dfc3222008-04-17 10:38:59 -04001521 /*
1522 * we only want to drop a reference if this transaction
1523 * modified the buffer
1524 */
1525 if (was_modified)
1526 drop_reserve = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001527
1528 /*
1529 * We are no longer going to journal this buffer.
1530 * However, the commit of this transaction is still
1531 * important to the buffer: the delete that we are now
1532 * processing might obsolete an old log entry, so by
1533 * committing, we can satisfy the buffer's checkpoint.
1534 *
1535 * So, if we have a checkpoint on the buffer, we should
1536 * now refile the buffer on our BJ_Forget list so that
1537 * we know to remove the checkpoint after we commit.
1538 */
1539
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001540 spin_lock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001541 if (jh->b_cp_transaction) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001542 __jbd2_journal_temp_unlink_buffer(jh);
1543 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001544 } else {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001545 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001546 if (!buffer_jbd(bh)) {
1547 spin_unlock(&journal->j_list_lock);
1548 jbd_unlock_bh_state(bh);
1549 __bforget(bh);
1550 goto drop;
1551 }
1552 }
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001553 spin_unlock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001554 } else if (jh->b_transaction) {
1555 J_ASSERT_JH(jh, (jh->b_transaction ==
1556 journal->j_committing_transaction));
1557 /* However, if the buffer is still owned by a prior
1558 * (committing) transaction, we can't drop it yet... */
1559 JBUFFER_TRACE(jh, "belongs to older transaction");
1560 /* ... but we CAN drop it from the new transaction if we
1561 * have also modified it since the original commit. */
1562
1563 if (jh->b_next_transaction) {
1564 J_ASSERT(jh->b_next_transaction == transaction);
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001565 spin_lock(&journal->j_list_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001566 jh->b_next_transaction = NULL;
Theodore Ts'o0bfea812014-03-09 00:56:58 -05001567 spin_unlock(&journal->j_list_lock);
Josef Bacik1dfc3222008-04-17 10:38:59 -04001568
1569 /*
1570 * only drop a reference if this transaction modified
1571 * the buffer
1572 */
1573 if (was_modified)
1574 drop_reserve = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001575 }
1576 }
1577
1578not_jbd:
Dave Kleikamp470decc2006-10-11 01:20:57 -07001579 jbd_unlock_bh_state(bh);
1580 __brelse(bh);
1581drop:
1582 if (drop_reserve) {
1583 /* no need to reserve log space for this block -bzzz */
1584 handle->h_buffer_credits++;
1585 }
1586 return err;
1587}
1588
1589/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001590 * int jbd2_journal_stop() - complete a transaction
Masanari Iidabd7ced92016-02-02 22:31:06 +09001591 * @handle: transaction to complete.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001592 *
1593 * All done for a particular handle.
1594 *
1595 * There is not much action needed here. We just return any remaining
1596 * buffer credits to the transaction and remove the handle. The only
1597 * complication is that we need to start a commit operation if the
1598 * filesystem is marked for synchronous update.
1599 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001600 * jbd2_journal_stop itself will not usually return an error, but it may
Dave Kleikamp470decc2006-10-11 01:20:57 -07001601 * do so in unusual circumstances. In particular, expect it to
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001602 * return -EIO if a jbd2_journal_abort has been executed since the
Dave Kleikamp470decc2006-10-11 01:20:57 -07001603 * transaction began.
1604 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001605int jbd2_journal_stop(handle_t *handle)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001606{
1607 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001608 journal_t *journal;
1609 int err = 0, wait_for_commit = 0;
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001610 tid_t tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001611 pid_t pid;
1612
Lukas Czerner9d506592015-05-14 18:55:18 -04001613 if (!transaction) {
1614 /*
1615 * Handle is already detached from the transaction so
1616 * there is nothing to do other than decrease a refcount,
1617 * or free the handle if refcount drops to zero
1618 */
1619 if (--handle->h_ref > 0) {
1620 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1621 handle->h_ref);
1622 return err;
1623 } else {
1624 if (handle->h_rsv_handle)
1625 jbd2_free_handle(handle->h_rsv_handle);
1626 goto free_and_exit;
1627 }
1628 }
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001629 journal = transaction->t_journal;
1630
Dave Kleikamp470decc2006-10-11 01:20:57 -07001631 J_ASSERT(journal_current_handle() == handle);
1632
1633 if (is_handle_aborted(handle))
1634 err = -EIO;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001635 else
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001636 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001637
1638 if (--handle->h_ref > 0) {
1639 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1640 handle->h_ref);
1641 return err;
1642 }
1643
1644 jbd_debug(4, "Handle %p going down\n", handle);
Theodore Ts'o343d9c22013-02-08 13:00:22 -05001645 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001646 transaction->t_tid,
Theodore Ts'o343d9c22013-02-08 13:00:22 -05001647 handle->h_type, handle->h_line_no,
1648 jiffies - handle->h_start_jiffies,
1649 handle->h_sync, handle->h_requested_credits,
1650 (handle->h_requested_credits -
1651 handle->h_buffer_credits));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001652
1653 /*
1654 * Implement synchronous transaction batching. If the handle
1655 * was synchronous, don't force a commit immediately. Let's
Josef Bacike07f7182008-11-26 01:14:26 -05001656 * yield and let another thread piggyback onto this
1657 * transaction. Keep doing that while new threads continue to
1658 * arrive. It doesn't cost much - we're about to run a commit
1659 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1660 * operations by 30x or more...
Dave Kleikamp470decc2006-10-11 01:20:57 -07001661 *
Josef Bacike07f7182008-11-26 01:14:26 -05001662 * We try and optimize the sleep time against what the
1663 * underlying disk can do, instead of having a static sleep
1664 * time. This is useful for the case where our storage is so
1665 * fast that it is more optimal to go ahead and force a flush
1666 * and wait for the transaction to be committed than it is to
1667 * wait for an arbitrary amount of time for new writers to
1668 * join the transaction. We achieve this by measuring how
1669 * long it takes to commit a transaction, and compare it with
1670 * how long this transaction has been running, and if run time
1671 * < commit time then we sleep for the delta and commit. This
1672 * greatly helps super fast disks that would see slowdowns as
1673 * more threads started doing fsyncs.
1674 *
1675 * But don't do this if this process was the most recent one
1676 * to perform a synchronous write. We do this to detect the
1677 * case where a single process is doing a stream of sync
1678 * writes. No point in waiting for joiners in that case.
Eric Sandeen5dd21422014-07-05 19:18:22 -04001679 *
1680 * Setting max_batch_time to 0 disables this completely.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001681 */
1682 pid = current->pid;
Eric Sandeen5dd21422014-07-05 19:18:22 -04001683 if (handle->h_sync && journal->j_last_sync_writer != pid &&
1684 journal->j_max_batch_time) {
Josef Bacike07f7182008-11-26 01:14:26 -05001685 u64 commit_time, trans_time;
1686
Dave Kleikamp470decc2006-10-11 01:20:57 -07001687 journal->j_last_sync_writer = pid;
Josef Bacike07f7182008-11-26 01:14:26 -05001688
Theodore Ts'oa931da62010-08-03 21:35:12 -04001689 read_lock(&journal->j_state_lock);
Josef Bacike07f7182008-11-26 01:14:26 -05001690 commit_time = journal->j_average_commit_time;
Theodore Ts'oa931da62010-08-03 21:35:12 -04001691 read_unlock(&journal->j_state_lock);
Josef Bacike07f7182008-11-26 01:14:26 -05001692
1693 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1694 transaction->t_start_time));
1695
Theodore Ts'o30773842009-01-03 20:27:38 -05001696 commit_time = max_t(u64, commit_time,
1697 1000*journal->j_min_batch_time);
Josef Bacike07f7182008-11-26 01:14:26 -05001698 commit_time = min_t(u64, commit_time,
Theodore Ts'o30773842009-01-03 20:27:38 -05001699 1000*journal->j_max_batch_time);
Josef Bacike07f7182008-11-26 01:14:26 -05001700
1701 if (trans_time < commit_time) {
1702 ktime_t expires = ktime_add_ns(ktime_get(),
1703 commit_time);
1704 set_current_state(TASK_UNINTERRUPTIBLE);
1705 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1706 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001707 }
1708
Theodore Ts'o70585482009-03-25 23:35:46 -04001709 if (handle->h_sync)
1710 transaction->t_synchronous_commit = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001711 current->journal_info = NULL;
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001712 atomic_sub(handle->h_buffer_credits,
1713 &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001714
1715 /*
1716 * If the handle is marked SYNC, we need to set another commit
1717 * going! We also want to force a commit if the current
1718 * transaction is occupying too much of the log, or if the
1719 * transaction is too old now.
1720 */
1721 if (handle->h_sync ||
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001722 (atomic_read(&transaction->t_outstanding_credits) >
1723 journal->j_max_transaction_buffers) ||
1724 time_after_eq(jiffies, transaction->t_expires)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001725 /* Do this even for aborted journals: an abort still
1726 * completes the commit thread, it just doesn't write
1727 * anything to disk. */
Dave Kleikamp470decc2006-10-11 01:20:57 -07001728
Dave Kleikamp470decc2006-10-11 01:20:57 -07001729 jbd_debug(2, "transaction too old, requesting commit for "
1730 "handle %p\n", handle);
1731 /* This is non-blocking */
Theodore Ts'oc35a56a2010-05-16 05:00:00 -04001732 jbd2_log_start_commit(journal, transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001733
1734 /*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001735 * Special case: JBD2_SYNC synchronous updates require us
Dave Kleikamp470decc2006-10-11 01:20:57 -07001736 * to wait for the commit to complete.
1737 */
1738 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001739 wait_for_commit = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001740 }
1741
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001742 /*
1743 * Once we drop t_updates, if it goes to zero the transaction
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001744 * could start committing on us and eventually disappear. So
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001745 * once we do this, we must not dereference transaction
1746 * pointer again.
1747 */
1748 tid = transaction->t_tid;
1749 if (atomic_dec_and_test(&transaction->t_updates)) {
1750 wake_up(&journal->j_wait_updates);
1751 if (journal->j_barrier_count)
1752 wake_up(&journal->j_wait_transaction_locked);
1753 }
1754
Jan Karaab714af2016-06-30 11:39:38 -04001755 rwsem_release(&journal->j_trans_commit_map, 1, _THIS_IP_);
Jan Kara7a4b1882016-06-30 11:30:21 -04001756
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001757 if (wait_for_commit)
1758 err = jbd2_log_wait_commit(journal, tid);
1759
Jan Kara8f7d89f2013-06-04 12:35:11 -04001760 if (handle->h_rsv_handle)
1761 jbd2_journal_free_reserved(handle->h_rsv_handle);
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001762free_and_exit:
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001763 jbd2_free_handle(handle);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001764 return err;
1765}
1766
Dave Kleikamp470decc2006-10-11 01:20:57 -07001767/*
1768 *
1769 * List management code snippets: various functions for manipulating the
1770 * transaction buffer lists.
1771 *
1772 */
1773
1774/*
1775 * Append a buffer to a transaction list, given the transaction's list head
1776 * pointer.
1777 *
1778 * j_list_lock is held.
1779 *
1780 * jbd_lock_bh_state(jh2bh(jh)) is held.
1781 */
1782
1783static inline void
1784__blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1785{
1786 if (!*list) {
1787 jh->b_tnext = jh->b_tprev = jh;
1788 *list = jh;
1789 } else {
1790 /* Insert at the tail of the list to preserve order */
1791 struct journal_head *first = *list, *last = first->b_tprev;
1792 jh->b_tprev = last;
1793 jh->b_tnext = first;
1794 last->b_tnext = first->b_tprev = jh;
1795 }
1796}
1797
1798/*
1799 * Remove a buffer from a transaction list, given the transaction's list
1800 * head pointer.
1801 *
1802 * Called with j_list_lock held, and the journal may not be locked.
1803 *
1804 * jbd_lock_bh_state(jh2bh(jh)) is held.
1805 */
1806
1807static inline void
1808__blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1809{
1810 if (*list == jh) {
1811 *list = jh->b_tnext;
1812 if (*list == jh)
1813 *list = NULL;
1814 }
1815 jh->b_tprev->b_tnext = jh->b_tnext;
1816 jh->b_tnext->b_tprev = jh->b_tprev;
1817}
1818
1819/*
1820 * Remove a buffer from the appropriate transaction list.
1821 *
1822 * Note that this function can *change* the value of
Jan Karaf5113ef2013-06-04 12:01:45 -04001823 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1824 * t_reserved_list. If the caller is holding onto a copy of one of these
1825 * pointers, it could go bad. Generally the caller needs to re-read the
1826 * pointer from the transaction_t.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001827 *
Jan Kara5bebccf2012-03-13 22:25:06 -04001828 * Called under j_list_lock.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001829 */
Jan Kara5bebccf2012-03-13 22:25:06 -04001830static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001831{
1832 struct journal_head **list = NULL;
1833 transaction_t *transaction;
1834 struct buffer_head *bh = jh2bh(jh);
1835
1836 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1837 transaction = jh->b_transaction;
1838 if (transaction)
1839 assert_spin_locked(&transaction->t_journal->j_list_lock);
1840
1841 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1842 if (jh->b_jlist != BJ_None)
Mingming Cao40191912008-01-28 23:58:27 -05001843 J_ASSERT_JH(jh, transaction != NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001844
1845 switch (jh->b_jlist) {
1846 case BJ_None:
1847 return;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001848 case BJ_Metadata:
1849 transaction->t_nr_buffers--;
1850 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1851 list = &transaction->t_buffers;
1852 break;
1853 case BJ_Forget:
1854 list = &transaction->t_forget;
1855 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001856 case BJ_Shadow:
1857 list = &transaction->t_shadow_list;
1858 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001859 case BJ_Reserved:
1860 list = &transaction->t_reserved_list;
1861 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001862 }
1863
1864 __blist_del_buffer(list, jh);
1865 jh->b_jlist = BJ_None;
1866 if (test_clear_buffer_jbddirty(bh))
1867 mark_buffer_dirty(bh); /* Expose it to the VM */
1868}
1869
Jan Karade1b7942011-06-13 15:38:22 -04001870/*
1871 * Remove buffer from all transactions.
1872 *
1873 * Called with bh_state lock and j_list_lock
1874 *
1875 * jh and bh may be already freed when this function returns.
1876 */
1877static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001878{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001879 __jbd2_journal_temp_unlink_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001880 jh->b_transaction = NULL;
Jan Karade1b7942011-06-13 15:38:22 -04001881 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001882}
1883
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001884void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001885{
Jan Karade1b7942011-06-13 15:38:22 -04001886 struct buffer_head *bh = jh2bh(jh);
1887
1888 /* Get reference so that buffer cannot be freed before we unlock it */
1889 get_bh(bh);
1890 jbd_lock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001891 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001892 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001893 spin_unlock(&journal->j_list_lock);
Jan Karade1b7942011-06-13 15:38:22 -04001894 jbd_unlock_bh_state(bh);
1895 __brelse(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001896}
1897
1898/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001899 * Called from jbd2_journal_try_to_free_buffers().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001900 *
1901 * Called under jbd_lock_bh_state(bh)
1902 */
1903static void
1904__journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1905{
1906 struct journal_head *jh;
1907
1908 jh = bh2jh(bh);
1909
1910 if (buffer_locked(bh) || buffer_dirty(bh))
1911 goto out;
1912
Theodore Ts'od2eb0b92014-03-09 00:07:19 -05001913 if (jh->b_next_transaction != NULL || jh->b_transaction != NULL)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001914 goto out;
1915
1916 spin_lock(&journal->j_list_lock);
Theodore Ts'od2eb0b92014-03-09 00:07:19 -05001917 if (jh->b_cp_transaction != NULL) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001918 /* written-back checkpointed metadata buffer */
Jan Karac254c9e2012-03-13 22:27:44 -04001919 JBUFFER_TRACE(jh, "remove from checkpoint list");
1920 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001921 }
1922 spin_unlock(&journal->j_list_lock);
1923out:
1924 return;
1925}
1926
Dave Kleikamp470decc2006-10-11 01:20:57 -07001927/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001928 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001929 * @journal: journal for operation
1930 * @page: to try and free
Mingming Cao530576b2008-07-13 21:06:39 -04001931 * @gfp_mask: we use the mask to detect how hard should we try to release
Mel Gormand0164ad2015-11-06 16:28:21 -08001932 * buffers. If __GFP_DIRECT_RECLAIM and __GFP_FS is set, we wait for commit
1933 * code to release the buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001934 *
1935 *
1936 * For all the buffers on this page,
1937 * if they are fully written out ordered data, move them onto BUF_CLEAN
1938 * so try_to_free_buffers() can reap them.
1939 *
1940 * This function returns non-zero if we wish try_to_free_buffers()
1941 * to be called. We do this if the page is releasable by try_to_free_buffers().
1942 * We also do it if the page has locked or dirty buffers and the caller wants
1943 * us to perform sync or async writeout.
1944 *
1945 * This complicates JBD locking somewhat. We aren't protected by the
1946 * BKL here. We wish to remove the buffer from its committing or
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001947 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001948 *
1949 * This may *change* the value of transaction_t->t_datalist, so anyone
1950 * who looks at t_datalist needs to lock against this function.
1951 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001952 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1953 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
Dave Kleikamp470decc2006-10-11 01:20:57 -07001954 * will come out of the lock with the buffer dirty, which makes it
1955 * ineligible for release here.
1956 *
1957 * Who else is affected by this? hmm... Really the only contender
1958 * is do_get_write_access() - it could be looking at the buffer while
1959 * journal_try_to_free_buffer() is changing its state. But that
1960 * cannot happen because we never reallocate freed data as metadata
1961 * while the data is part of a transaction. Yes?
Mingming Cao530576b2008-07-13 21:06:39 -04001962 *
1963 * Return 0 on failure, 1 on success
Dave Kleikamp470decc2006-10-11 01:20:57 -07001964 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001965int jbd2_journal_try_to_free_buffers(journal_t *journal,
Mingming Cao530576b2008-07-13 21:06:39 -04001966 struct page *page, gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001967{
1968 struct buffer_head *head;
1969 struct buffer_head *bh;
1970 int ret = 0;
1971
1972 J_ASSERT(PageLocked(page));
1973
1974 head = page_buffers(page);
1975 bh = head;
1976 do {
1977 struct journal_head *jh;
1978
1979 /*
1980 * We take our own ref against the journal_head here to avoid
1981 * having to add tons of locking around each instance of
Mingming Cao530576b2008-07-13 21:06:39 -04001982 * jbd2_journal_put_journal_head().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001983 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001984 jh = jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001985 if (!jh)
1986 continue;
1987
1988 jbd_lock_bh_state(bh);
1989 __journal_try_to_free_buffer(journal, bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001990 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001991 jbd_unlock_bh_state(bh);
1992 if (buffer_jbd(bh))
1993 goto busy;
1994 } while ((bh = bh->b_this_page) != head);
Mingming Cao530576b2008-07-13 21:06:39 -04001995
Dave Kleikamp470decc2006-10-11 01:20:57 -07001996 ret = try_to_free_buffers(page);
Mingming Cao530576b2008-07-13 21:06:39 -04001997
Dave Kleikamp470decc2006-10-11 01:20:57 -07001998busy:
1999 return ret;
2000}
2001
2002/*
2003 * This buffer is no longer needed. If it is on an older transaction's
2004 * checkpoint list we need to record it on this transaction's forget list
2005 * to pin this buffer (and hence its checkpointing transaction) down until
2006 * this transaction commits. If the buffer isn't on a checkpoint list, we
2007 * release it.
2008 * Returns non-zero if JBD no longer has an interest in the buffer.
2009 *
2010 * Called under j_list_lock.
2011 *
2012 * Called under jbd_lock_bh_state(bh).
2013 */
2014static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
2015{
2016 int may_free = 1;
2017 struct buffer_head *bh = jh2bh(jh);
2018
Dave Kleikamp470decc2006-10-11 01:20:57 -07002019 if (jh->b_cp_transaction) {
2020 JBUFFER_TRACE(jh, "on running+cp transaction");
Jan Karade1b7942011-06-13 15:38:22 -04002021 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karaf91d1d02009-07-13 16:16:20 -04002022 /*
2023 * We don't want to write the buffer anymore, clear the
2024 * bit so that we don't confuse checks in
2025 * __journal_file_buffer
2026 */
2027 clear_buffer_dirty(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002028 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002029 may_free = 0;
2030 } else {
2031 JBUFFER_TRACE(jh, "on running transaction");
Jan Karade1b7942011-06-13 15:38:22 -04002032 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002033 }
2034 return may_free;
2035}
2036
2037/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002038 * jbd2_journal_invalidatepage
Dave Kleikamp470decc2006-10-11 01:20:57 -07002039 *
2040 * This code is tricky. It has a number of cases to deal with.
2041 *
2042 * There are two invariants which this code relies on:
2043 *
2044 * i_size must be updated on disk before we start calling invalidatepage on the
2045 * data.
2046 *
2047 * This is done in ext3 by defining an ext3_setattr method which
2048 * updates i_size before truncate gets going. By maintaining this
2049 * invariant, we can be sure that it is safe to throw away any buffers
2050 * attached to the current transaction: once the transaction commits,
2051 * we know that the data will not be needed.
2052 *
2053 * Note however that we can *not* throw away data belonging to the
2054 * previous, committing transaction!
2055 *
2056 * Any disk blocks which *are* part of the previous, committing
2057 * transaction (and which therefore cannot be discarded immediately) are
2058 * not going to be reused in the new running transaction
2059 *
2060 * The bitmap committed_data images guarantee this: any block which is
2061 * allocated in one transaction and removed in the next will be marked
2062 * as in-use in the committed_data bitmap, so cannot be reused until
2063 * the next transaction to delete the block commits. This means that
2064 * leaving committing buffers dirty is quite safe: the disk blocks
2065 * cannot be reallocated to a different file and so buffer aliasing is
2066 * not possible.
2067 *
2068 *
2069 * The above applies mainly to ordered data mode. In writeback mode we
2070 * don't make guarantees about the order in which data hits disk --- in
2071 * particular we don't guarantee that new dirty data is flushed before
2072 * transaction commit --- so it is always safe just to discard data
2073 * immediately in that mode. --sct
2074 */
2075
2076/*
2077 * The journal_unmap_buffer helper function returns zero if the buffer
2078 * concerned remains pinned as an anonymous buffer belonging to an older
2079 * transaction.
2080 *
2081 * We're outside-transaction here. Either or both of j_running_transaction
2082 * and j_committing_transaction may be NULL.
2083 */
Jan Karab794e7a2012-09-26 23:11:13 -04002084static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
2085 int partial_page)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002086{
2087 transaction_t *transaction;
2088 struct journal_head *jh;
2089 int may_free = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002090
2091 BUFFER_TRACE(bh, "entry");
2092
2093 /*
2094 * It is safe to proceed here without the j_list_lock because the
2095 * buffers cannot be stolen by try_to_free_buffers as long as we are
2096 * holding the page lock. --sct
2097 */
2098
2099 if (!buffer_jbd(bh))
2100 goto zap_buffer_unlocked;
2101
Jan Kara87c89c22008-07-11 19:27:31 -04002102 /* OK, we have data buffer in journaled mode */
Theodore Ts'oa931da62010-08-03 21:35:12 -04002103 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002104 jbd_lock_bh_state(bh);
2105 spin_lock(&journal->j_list_lock);
2106
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002107 jh = jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002108 if (!jh)
2109 goto zap_buffer_no_jh;
2110
dingdinghuaba869022010-02-15 16:35:42 -05002111 /*
2112 * We cannot remove the buffer from checkpoint lists until the
2113 * transaction adding inode to orphan list (let's call it T)
2114 * is committed. Otherwise if the transaction changing the
2115 * buffer would be cleaned from the journal before T is
2116 * committed, a crash will cause that the correct contents of
2117 * the buffer will be lost. On the other hand we have to
2118 * clear the buffer dirty bit at latest at the moment when the
2119 * transaction marking the buffer as freed in the filesystem
2120 * structures is committed because from that moment on the
Jan Karab794e7a2012-09-26 23:11:13 -04002121 * block can be reallocated and used by a different page.
dingdinghuaba869022010-02-15 16:35:42 -05002122 * Since the block hasn't been freed yet but the inode has
2123 * already been added to orphan list, it is safe for us to add
2124 * the buffer to BJ_Forget list of the newest transaction.
Jan Karab794e7a2012-09-26 23:11:13 -04002125 *
2126 * Also we have to clear buffer_mapped flag of a truncated buffer
2127 * because the buffer_head may be attached to the page straddling
2128 * i_size (can happen only when blocksize < pagesize) and thus the
2129 * buffer_head can be reused when the file is extended again. So we end
2130 * up keeping around invalidated buffers attached to transactions'
2131 * BJ_Forget list just to stop checkpointing code from cleaning up
2132 * the transaction this buffer was modified in.
dingdinghuaba869022010-02-15 16:35:42 -05002133 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002134 transaction = jh->b_transaction;
2135 if (transaction == NULL) {
2136 /* First case: not on any transaction. If it
2137 * has no checkpoint link, then we can zap it:
2138 * it's a writeback-mode buffer so we don't care
2139 * if it hits disk safely. */
2140 if (!jh->b_cp_transaction) {
2141 JBUFFER_TRACE(jh, "not on any transaction: zap");
2142 goto zap_buffer;
2143 }
2144
2145 if (!buffer_dirty(bh)) {
2146 /* bdflush has written it. We can drop it now */
Jan Karabc23f0c2015-11-24 15:34:35 -05002147 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002148 goto zap_buffer;
2149 }
2150
2151 /* OK, it must be in the journal but still not
2152 * written fully to disk: it's metadata or
2153 * journaled data... */
2154
2155 if (journal->j_running_transaction) {
2156 /* ... and once the current transaction has
2157 * committed, the buffer won't be needed any
2158 * longer. */
2159 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
Jan Karab794e7a2012-09-26 23:11:13 -04002160 may_free = __dispose_buffer(jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002161 journal->j_running_transaction);
Jan Karab794e7a2012-09-26 23:11:13 -04002162 goto zap_buffer;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002163 } else {
2164 /* There is no currently-running transaction. So the
2165 * orphan record which we wrote for this file must have
2166 * passed into commit. We must attach this buffer to
2167 * the committing transaction, if it exists. */
2168 if (journal->j_committing_transaction) {
2169 JBUFFER_TRACE(jh, "give to committing trans");
Jan Karab794e7a2012-09-26 23:11:13 -04002170 may_free = __dispose_buffer(jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002171 journal->j_committing_transaction);
Jan Karab794e7a2012-09-26 23:11:13 -04002172 goto zap_buffer;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002173 } else {
2174 /* The orphan record's transaction has
2175 * committed. We can cleanse this buffer */
2176 clear_buffer_jbddirty(bh);
Jan Karabc23f0c2015-11-24 15:34:35 -05002177 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002178 goto zap_buffer;
2179 }
2180 }
2181 } else if (transaction == journal->j_committing_transaction) {
Eric Sandeen9b579882006-10-28 10:38:28 -07002182 JBUFFER_TRACE(jh, "on committing transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002183 /*
dingdinghuaba869022010-02-15 16:35:42 -05002184 * The buffer is committing, we simply cannot touch
Jan Karab794e7a2012-09-26 23:11:13 -04002185 * it. If the page is straddling i_size we have to wait
2186 * for commit and try again.
2187 */
2188 if (partial_page) {
Jan Karab794e7a2012-09-26 23:11:13 -04002189 jbd2_journal_put_journal_head(jh);
2190 spin_unlock(&journal->j_list_lock);
2191 jbd_unlock_bh_state(bh);
2192 write_unlock(&journal->j_state_lock);
Jan Kara53e87262012-12-25 13:29:52 -05002193 return -EBUSY;
Jan Karab794e7a2012-09-26 23:11:13 -04002194 }
2195 /*
2196 * OK, buffer won't be reachable after truncate. We just set
2197 * j_next_transaction to the running transaction (if there is
2198 * one) and mark buffer as freed so that commit code knows it
2199 * should clear dirty bits when it is done with the buffer.
dingdinghuaba869022010-02-15 16:35:42 -05002200 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002201 set_buffer_freed(bh);
dingdinghuaba869022010-02-15 16:35:42 -05002202 if (journal->j_running_transaction && buffer_jbddirty(bh))
2203 jh->b_next_transaction = journal->j_running_transaction;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002204 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002205 spin_unlock(&journal->j_list_lock);
2206 jbd_unlock_bh_state(bh);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002207 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002208 return 0;
2209 } else {
2210 /* Good, the buffer belongs to the running transaction.
2211 * We are writing our own transaction's data, not any
2212 * previous one's, so it is safe to throw it away
2213 * (remember that we expect the filesystem to have set
2214 * i_size already for this truncate so recovery will not
2215 * expose the disk blocks we are discarding here.) */
2216 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
Eric Sandeen9b579882006-10-28 10:38:28 -07002217 JBUFFER_TRACE(jh, "on running transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002218 may_free = __dispose_buffer(jh, transaction);
2219 }
2220
2221zap_buffer:
Jan Karab794e7a2012-09-26 23:11:13 -04002222 /*
2223 * This is tricky. Although the buffer is truncated, it may be reused
2224 * if blocksize < pagesize and it is attached to the page straddling
2225 * EOF. Since the buffer might have been added to BJ_Forget list of the
2226 * running transaction, journal_get_write_access() won't clear
2227 * b_modified and credit accounting gets confused. So clear b_modified
2228 * here.
2229 */
2230 jh->b_modified = 0;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002231 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002232zap_buffer_no_jh:
2233 spin_unlock(&journal->j_list_lock);
2234 jbd_unlock_bh_state(bh);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002235 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002236zap_buffer_unlocked:
2237 clear_buffer_dirty(bh);
2238 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2239 clear_buffer_mapped(bh);
2240 clear_buffer_req(bh);
2241 clear_buffer_new(bh);
Eric Sandeen15291162012-02-20 17:53:01 -05002242 clear_buffer_delay(bh);
2243 clear_buffer_unwritten(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002244 bh->b_bdev = NULL;
2245 return may_free;
2246}
2247
2248/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002249 * void jbd2_journal_invalidatepage()
Dave Kleikamp470decc2006-10-11 01:20:57 -07002250 * @journal: journal to use for flush...
2251 * @page: page to flush
Lukas Czerner259709b2013-05-21 23:20:03 -04002252 * @offset: start of the range to invalidate
2253 * @length: length of the range to invalidate
Dave Kleikamp470decc2006-10-11 01:20:57 -07002254 *
Lukas Czerner259709b2013-05-21 23:20:03 -04002255 * Reap page buffers containing data after in the specified range in page.
2256 * Can return -EBUSY if buffers are part of the committing transaction and
2257 * the page is straddling i_size. Caller then has to wait for current commit
2258 * and try again.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002259 */
Jan Kara53e87262012-12-25 13:29:52 -05002260int jbd2_journal_invalidatepage(journal_t *journal,
2261 struct page *page,
Lukas Czerner259709b2013-05-21 23:20:03 -04002262 unsigned int offset,
2263 unsigned int length)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002264{
2265 struct buffer_head *head, *bh, *next;
Lukas Czerner259709b2013-05-21 23:20:03 -04002266 unsigned int stop = offset + length;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002267 unsigned int curr_off = 0;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002268 int partial_page = (offset || length < PAGE_SIZE);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002269 int may_free = 1;
Jan Kara53e87262012-12-25 13:29:52 -05002270 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002271
2272 if (!PageLocked(page))
2273 BUG();
2274 if (!page_has_buffers(page))
Jan Kara53e87262012-12-25 13:29:52 -05002275 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002276
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002277 BUG_ON(stop > PAGE_SIZE || stop < length);
Lukas Czerner259709b2013-05-21 23:20:03 -04002278
Dave Kleikamp470decc2006-10-11 01:20:57 -07002279 /* We will potentially be playing with lists other than just the
2280 * data lists (especially for journaled data mode), so be
2281 * cautious in our locking. */
2282
2283 head = bh = page_buffers(page);
2284 do {
2285 unsigned int next_off = curr_off + bh->b_size;
2286 next = bh->b_this_page;
2287
Lukas Czerner259709b2013-05-21 23:20:03 -04002288 if (next_off > stop)
2289 return 0;
2290
Dave Kleikamp470decc2006-10-11 01:20:57 -07002291 if (offset <= curr_off) {
2292 /* This block is wholly outside the truncation point */
2293 lock_buffer(bh);
Lukas Czerner259709b2013-05-21 23:20:03 -04002294 ret = journal_unmap_buffer(journal, bh, partial_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002295 unlock_buffer(bh);
Jan Kara53e87262012-12-25 13:29:52 -05002296 if (ret < 0)
2297 return ret;
2298 may_free &= ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002299 }
2300 curr_off = next_off;
2301 bh = next;
2302
2303 } while (bh != head);
2304
Lukas Czerner259709b2013-05-21 23:20:03 -04002305 if (!partial_page) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002306 if (may_free && try_to_free_buffers(page))
2307 J_ASSERT(!page_has_buffers(page));
2308 }
Jan Kara53e87262012-12-25 13:29:52 -05002309 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002310}
2311
2312/*
2313 * File a buffer on the given transaction list.
2314 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002315void __jbd2_journal_file_buffer(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002316 transaction_t *transaction, int jlist)
2317{
2318 struct journal_head **list = NULL;
2319 int was_dirty = 0;
2320 struct buffer_head *bh = jh2bh(jh);
2321
2322 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2323 assert_spin_locked(&transaction->t_journal->j_list_lock);
2324
2325 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2326 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
Mingming Cao40191912008-01-28 23:58:27 -05002327 jh->b_transaction == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002328
2329 if (jh->b_transaction && jh->b_jlist == jlist)
2330 return;
2331
Dave Kleikamp470decc2006-10-11 01:20:57 -07002332 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2333 jlist == BJ_Shadow || jlist == BJ_Forget) {
Jan Karaf91d1d02009-07-13 16:16:20 -04002334 /*
2335 * For metadata buffers, we track dirty bit in buffer_jbddirty
2336 * instead of buffer_dirty. We should not see a dirty bit set
2337 * here because we clear it in do_get_write_access but e.g.
2338 * tune2fs can modify the sb and set the dirty bit at any time
2339 * so we try to gracefully handle that.
2340 */
2341 if (buffer_dirty(bh))
2342 warn_dirty_buffer(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002343 if (test_clear_buffer_dirty(bh) ||
2344 test_clear_buffer_jbddirty(bh))
2345 was_dirty = 1;
2346 }
2347
2348 if (jh->b_transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002349 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karade1b7942011-06-13 15:38:22 -04002350 else
2351 jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002352 jh->b_transaction = transaction;
2353
2354 switch (jlist) {
2355 case BJ_None:
2356 J_ASSERT_JH(jh, !jh->b_committed_data);
2357 J_ASSERT_JH(jh, !jh->b_frozen_data);
2358 return;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002359 case BJ_Metadata:
2360 transaction->t_nr_buffers++;
2361 list = &transaction->t_buffers;
2362 break;
2363 case BJ_Forget:
2364 list = &transaction->t_forget;
2365 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002366 case BJ_Shadow:
2367 list = &transaction->t_shadow_list;
2368 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002369 case BJ_Reserved:
2370 list = &transaction->t_reserved_list;
2371 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002372 }
2373
2374 __blist_add_buffer(list, jh);
2375 jh->b_jlist = jlist;
2376
2377 if (was_dirty)
2378 set_buffer_jbddirty(bh);
2379}
2380
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002381void jbd2_journal_file_buffer(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002382 transaction_t *transaction, int jlist)
2383{
2384 jbd_lock_bh_state(jh2bh(jh));
2385 spin_lock(&transaction->t_journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002386 __jbd2_journal_file_buffer(jh, transaction, jlist);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002387 spin_unlock(&transaction->t_journal->j_list_lock);
2388 jbd_unlock_bh_state(jh2bh(jh));
2389}
2390
2391/*
2392 * Remove a buffer from its current buffer list in preparation for
2393 * dropping it from its current transaction entirely. If the buffer has
2394 * already started to be used by a subsequent transaction, refile the
2395 * buffer on that transaction's metadata list.
2396 *
Jan Karade1b7942011-06-13 15:38:22 -04002397 * Called under j_list_lock
Dave Kleikamp470decc2006-10-11 01:20:57 -07002398 * Called under jbd_lock_bh_state(jh2bh(jh))
Jan Karade1b7942011-06-13 15:38:22 -04002399 *
2400 * jh and bh may be already free when this function returns
Dave Kleikamp470decc2006-10-11 01:20:57 -07002401 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002402void __jbd2_journal_refile_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002403{
dingdinghuaba869022010-02-15 16:35:42 -05002404 int was_dirty, jlist;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002405 struct buffer_head *bh = jh2bh(jh);
2406
2407 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2408 if (jh->b_transaction)
2409 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2410
2411 /* If the buffer is now unused, just drop it. */
2412 if (jh->b_next_transaction == NULL) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002413 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002414 return;
2415 }
2416
2417 /*
2418 * It has been modified by a later transaction: add it to the new
2419 * transaction's metadata list.
2420 */
2421
2422 was_dirty = test_clear_buffer_jbddirty(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002423 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karade1b7942011-06-13 15:38:22 -04002424 /*
2425 * We set b_transaction here because b_next_transaction will inherit
2426 * our jh reference and thus __jbd2_journal_file_buffer() must not
2427 * take a new one.
2428 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002429 jh->b_transaction = jh->b_next_transaction;
2430 jh->b_next_transaction = NULL;
dingdinghuaba869022010-02-15 16:35:42 -05002431 if (buffer_freed(bh))
2432 jlist = BJ_Forget;
2433 else if (jh->b_modified)
2434 jlist = BJ_Metadata;
2435 else
2436 jlist = BJ_Reserved;
2437 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002438 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2439
2440 if (was_dirty)
2441 set_buffer_jbddirty(bh);
2442}
2443
2444/*
Jan Karade1b7942011-06-13 15:38:22 -04002445 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2446 * bh reference so that we can safely unlock bh.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002447 *
Jan Karade1b7942011-06-13 15:38:22 -04002448 * The jh and bh may be freed by this call.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002449 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002450void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002451{
2452 struct buffer_head *bh = jh2bh(jh);
2453
Jan Karade1b7942011-06-13 15:38:22 -04002454 /* Get reference so that buffer cannot be freed before we unlock it */
2455 get_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002456 jbd_lock_bh_state(bh);
2457 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002458 __jbd2_journal_refile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002459 jbd_unlock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002460 spin_unlock(&journal->j_list_lock);
2461 __brelse(bh);
2462}
Jan Karac851ed52008-07-11 19:27:31 -04002463
2464/*
2465 * File inode in the inode list of the handle's transaction
2466 */
Jan Kara41617e12016-04-24 00:56:07 -04002467static int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode,
2468 unsigned long flags)
Jan Karac851ed52008-07-11 19:27:31 -04002469{
2470 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002471 journal_t *journal;
Jan Karac851ed52008-07-11 19:27:31 -04002472
2473 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002474 return -EROFS;
2475 journal = transaction->t_journal;
Jan Karac851ed52008-07-11 19:27:31 -04002476
2477 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2478 transaction->t_tid);
2479
2480 /*
2481 * First check whether inode isn't already on the transaction's
2482 * lists without taking the lock. Note that this check is safe
2483 * without the lock as we cannot race with somebody removing inode
2484 * from the transaction. The reason is that we remove inode from the
2485 * transaction only in journal_release_jbd_inode() and when we commit
2486 * the transaction. We are guarded from the first case by holding
2487 * a reference to the inode. We are safe against the second case
2488 * because if jinode->i_transaction == transaction, commit code
2489 * cannot touch the transaction because we hold reference to it,
2490 * and if jinode->i_next_transaction == transaction, commit code
2491 * will only file the inode where we want it.
2492 */
Jan Kara41617e12016-04-24 00:56:07 -04002493 if ((jinode->i_transaction == transaction ||
2494 jinode->i_next_transaction == transaction) &&
2495 (jinode->i_flags & flags) == flags)
Jan Karac851ed52008-07-11 19:27:31 -04002496 return 0;
2497
2498 spin_lock(&journal->j_list_lock);
Jan Kara41617e12016-04-24 00:56:07 -04002499 jinode->i_flags |= flags;
2500 /* Is inode already attached where we need it? */
Jan Karac851ed52008-07-11 19:27:31 -04002501 if (jinode->i_transaction == transaction ||
2502 jinode->i_next_transaction == transaction)
2503 goto done;
2504
Jan Kara81be12c2011-05-24 11:52:40 -04002505 /*
2506 * We only ever set this variable to 1 so the test is safe. Since
2507 * t_need_data_flush is likely to be set, we do the test to save some
2508 * cacheline bouncing
2509 */
2510 if (!transaction->t_need_data_flush)
2511 transaction->t_need_data_flush = 1;
Jan Karac851ed52008-07-11 19:27:31 -04002512 /* On some different transaction's list - should be
2513 * the committing one */
2514 if (jinode->i_transaction) {
2515 J_ASSERT(jinode->i_next_transaction == NULL);
2516 J_ASSERT(jinode->i_transaction ==
2517 journal->j_committing_transaction);
2518 jinode->i_next_transaction = transaction;
2519 goto done;
2520 }
2521 /* Not on any transaction list... */
2522 J_ASSERT(!jinode->i_next_transaction);
2523 jinode->i_transaction = transaction;
2524 list_add(&jinode->i_list, &transaction->t_inode_list);
2525done:
2526 spin_unlock(&journal->j_list_lock);
2527
2528 return 0;
2529}
2530
Jan Kara41617e12016-04-24 00:56:07 -04002531int jbd2_journal_inode_add_write(handle_t *handle, struct jbd2_inode *jinode)
2532{
2533 return jbd2_journal_file_inode(handle, jinode,
2534 JI_WRITE_DATA | JI_WAIT_DATA);
2535}
2536
2537int jbd2_journal_inode_add_wait(handle_t *handle, struct jbd2_inode *jinode)
2538{
2539 return jbd2_journal_file_inode(handle, jinode, JI_WAIT_DATA);
2540}
2541
Jan Karac851ed52008-07-11 19:27:31 -04002542/*
Jan Kara7f5aa212009-02-10 11:15:34 -05002543 * File truncate and transaction commit interact with each other in a
2544 * non-trivial way. If a transaction writing data block A is
2545 * committing, we cannot discard the data by truncate until we have
2546 * written them. Otherwise if we crashed after the transaction with
2547 * write has committed but before the transaction with truncate has
2548 * committed, we could see stale data in block A. This function is a
2549 * helper to solve this problem. It starts writeout of the truncated
2550 * part in case it is in the committing transaction.
2551 *
2552 * Filesystem code must call this function when inode is journaled in
2553 * ordered mode before truncation happens and after the inode has been
2554 * placed on orphan list with the new inode size. The second condition
2555 * avoids the race that someone writes new data and we start
2556 * committing the transaction after this function has been called but
2557 * before a transaction for truncate is started (and furthermore it
2558 * allows us to optimize the case where the addition to orphan list
2559 * happens in the same transaction as write --- we don't have to write
2560 * any data in such case).
Jan Karac851ed52008-07-11 19:27:31 -04002561 */
Jan Kara7f5aa212009-02-10 11:15:34 -05002562int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2563 struct jbd2_inode *jinode,
Jan Karac851ed52008-07-11 19:27:31 -04002564 loff_t new_size)
2565{
Jan Kara7f5aa212009-02-10 11:15:34 -05002566 transaction_t *inode_trans, *commit_trans;
Jan Karac851ed52008-07-11 19:27:31 -04002567 int ret = 0;
2568
Jan Kara7f5aa212009-02-10 11:15:34 -05002569 /* This is a quick check to avoid locking if not necessary */
2570 if (!jinode->i_transaction)
Jan Karac851ed52008-07-11 19:27:31 -04002571 goto out;
Jan Kara7f5aa212009-02-10 11:15:34 -05002572 /* Locks are here just to force reading of recent values, it is
2573 * enough that the transaction was not committing before we started
2574 * a transaction adding the inode to orphan list */
Theodore Ts'oa931da62010-08-03 21:35:12 -04002575 read_lock(&journal->j_state_lock);
Jan Karac851ed52008-07-11 19:27:31 -04002576 commit_trans = journal->j_committing_transaction;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002577 read_unlock(&journal->j_state_lock);
Jan Kara7f5aa212009-02-10 11:15:34 -05002578 spin_lock(&journal->j_list_lock);
2579 inode_trans = jinode->i_transaction;
2580 spin_unlock(&journal->j_list_lock);
2581 if (inode_trans == commit_trans) {
2582 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
Jan Karac851ed52008-07-11 19:27:31 -04002583 new_size, LLONG_MAX);
2584 if (ret)
2585 jbd2_journal_abort(journal, ret);
2586 }
2587out:
2588 return ret;
2589}