blob: 8360674c85bcb98af88d6aaff3ee4c46fe6f31bf [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);
162 schedule();
163 finish_wait(&journal->j_wait_transaction_locked, &wait);
164}
165
166static void sub_reserved_credits(journal_t *journal, int blocks)
167{
168 atomic_sub(blocks, &journal->j_reserved_credits);
169 wake_up(&journal->j_wait_reserved);
170}
171
172/*
173 * Wait until we can add credits for handle to the running transaction. Called
174 * with j_state_lock held for reading. Returns 0 if handle joined the running
175 * transaction. Returns 1 if we had to wait, j_state_lock is dropped, and
176 * caller must retry.
177 */
178static int add_transaction_credits(journal_t *journal, int blocks,
179 int rsv_blocks)
180{
181 transaction_t *t = journal->j_running_transaction;
182 int needed;
183 int total = blocks + rsv_blocks;
184
185 /*
186 * If the current transaction is locked down for commit, wait
187 * for the lock to be released.
188 */
189 if (t->t_state == T_LOCKED) {
190 wait_transaction_locked(journal);
191 return 1;
192 }
193
194 /*
195 * If there is not enough space left in the log to write all
196 * potential buffers requested by this operation, we need to
197 * stall pending a log checkpoint to free some more log space.
198 */
199 needed = atomic_add_return(total, &t->t_outstanding_credits);
200 if (needed > journal->j_max_transaction_buffers) {
201 /*
202 * If the current transaction is already too large,
203 * then start to commit it: we can then go back and
204 * attach this handle to a new transaction.
205 */
206 atomic_sub(total, &t->t_outstanding_credits);
207 wait_transaction_locked(journal);
208 return 1;
209 }
210
211 /*
212 * The commit code assumes that it can get enough log space
213 * without forcing a checkpoint. This is *critical* for
214 * correctness: a checkpoint of a buffer which is also
215 * associated with a committing transaction creates a deadlock,
216 * so commit simply cannot force through checkpoints.
217 *
218 * We must therefore ensure the necessary space in the journal
219 * *before* starting to dirty potentially checkpointed buffers
220 * in the new transaction.
221 */
222 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal)) {
223 atomic_sub(total, &t->t_outstanding_credits);
224 read_unlock(&journal->j_state_lock);
225 write_lock(&journal->j_state_lock);
226 if (jbd2_log_space_left(journal) < jbd2_space_needed(journal))
227 __jbd2_log_wait_for_space(journal);
228 write_unlock(&journal->j_state_lock);
229 return 1;
230 }
231
232 /* No reservation? We are done... */
233 if (!rsv_blocks)
234 return 0;
235
236 needed = atomic_add_return(rsv_blocks, &journal->j_reserved_credits);
237 /* We allow at most half of a transaction to be reserved */
238 if (needed > journal->j_max_transaction_buffers / 2) {
239 sub_reserved_credits(journal, rsv_blocks);
240 atomic_sub(total, &t->t_outstanding_credits);
241 read_unlock(&journal->j_state_lock);
242 wait_event(journal->j_wait_reserved,
243 atomic_read(&journal->j_reserved_credits) + rsv_blocks
244 <= journal->j_max_transaction_buffers / 2);
245 return 1;
246 }
247 return 0;
248}
249
250/*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700251 * start_this_handle: Given a handle, deal with any locking or stalling
252 * needed to make sure that there is enough journal space for the handle
253 * to begin. Attach the handle to a transaction and set up the
254 * transaction's buffer credits.
255 */
256
Theodore Ts'o47def822010-07-27 11:56:05 -0400257static int start_this_handle(journal_t *journal, handle_t *handle,
Dan Carpenterd2159fb2011-09-04 10:20:14 -0400258 gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700259{
Theodore Ts'oe4471832011-02-12 08:18:24 -0500260 transaction_t *transaction, *new_transaction = NULL;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400261 int blocks = handle->h_buffer_credits;
262 int rsv_blocks = 0;
Tao Ma28e35e42011-05-22 21:45:26 -0400263 unsigned long ts = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700264
Jan Kara8f7d89f2013-06-04 12:35:11 -0400265 /*
266 * 1/2 of transaction can be reserved so we can practically handle
267 * only 1/2 of maximum transaction size per operation
268 */
269 if (WARN_ON(blocks > journal->j_max_transaction_buffers / 2)) {
Eryu Guanf2a44522011-11-01 19:09:18 -0400270 printk(KERN_ERR "JBD2: %s wants too many credits (%d > %d)\n",
Jan Kara8f7d89f2013-06-04 12:35:11 -0400271 current->comm, blocks,
272 journal->j_max_transaction_buffers / 2);
Theodore Ts'o47def822010-07-27 11:56:05 -0400273 return -ENOSPC;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700274 }
275
Jan Kara8f7d89f2013-06-04 12:35:11 -0400276 if (handle->h_rsv_handle)
277 rsv_blocks = handle->h_rsv_handle->h_buffer_credits;
278
Dave Kleikamp470decc2006-10-11 01:20:57 -0700279alloc_transaction:
280 if (!journal->j_running_transaction) {
Wanlong Gaob2f4edb2012-06-01 00:10:32 -0400281 new_transaction = kmem_cache_zalloc(transaction_cache,
282 gfp_mask);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700283 if (!new_transaction) {
Theodore Ts'o47def822010-07-27 11:56:05 -0400284 /*
285 * If __GFP_FS is not present, then we may be
286 * being called from inside the fs writeback
287 * layer, so we MUST NOT fail. Since
288 * __GFP_NOFAIL is going away, we will arrange
289 * to retry the allocation ourselves.
290 */
291 if ((gfp_mask & __GFP_FS) == 0) {
292 congestion_wait(BLK_RW_ASYNC, HZ/50);
293 goto alloc_transaction;
294 }
295 return -ENOMEM;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700296 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700297 }
298
299 jbd_debug(3, "New handle %p going live.\n", handle);
300
Dave Kleikamp470decc2006-10-11 01:20:57 -0700301 /*
302 * We need to hold j_state_lock until t_updates has been incremented,
303 * for proper journal barrier handling
304 */
Theodore Ts'oa931da62010-08-03 21:35:12 -0400305repeat:
306 read_lock(&journal->j_state_lock);
Theodore Ts'o5c2178e2010-10-27 21:30:04 -0400307 BUG_ON(journal->j_flags & JBD2_UNMOUNT);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700308 if (is_journal_aborted(journal) ||
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700309 (journal->j_errno != 0 && !(journal->j_flags & JBD2_ACK_ERR))) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400310 read_unlock(&journal->j_state_lock);
Yongqiang Yang0c2022e2012-02-20 17:53:02 -0500311 jbd2_journal_free_transaction(new_transaction);
Theodore Ts'o47def822010-07-27 11:56:05 -0400312 return -EROFS;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700313 }
314
Jan Kara8f7d89f2013-06-04 12:35:11 -0400315 /*
316 * Wait on the journal's transaction barrier if necessary. Specifically
317 * we allow reserved handles to proceed because otherwise commit could
318 * deadlock on page writeback not being able to complete.
319 */
320 if (!handle->h_reserved && journal->j_barrier_count) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400321 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700322 wait_event(journal->j_wait_transaction_locked,
323 journal->j_barrier_count == 0);
324 goto repeat;
325 }
326
327 if (!journal->j_running_transaction) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400328 read_unlock(&journal->j_state_lock);
329 if (!new_transaction)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700330 goto alloc_transaction;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400331 write_lock(&journal->j_state_lock);
Jan Karad7961c72012-12-21 00:15:51 -0500332 if (!journal->j_running_transaction &&
Jan Kara8f7d89f2013-06-04 12:35:11 -0400333 (handle->h_reserved || !journal->j_barrier_count)) {
Theodore Ts'oa931da62010-08-03 21:35:12 -0400334 jbd2_get_transaction(journal, new_transaction);
335 new_transaction = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700336 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400337 write_unlock(&journal->j_state_lock);
338 goto repeat;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700339 }
340
341 transaction = journal->j_running_transaction;
342
Jan Kara8f7d89f2013-06-04 12:35:11 -0400343 if (!handle->h_reserved) {
344 /* We may have dropped j_state_lock - restart in that case */
345 if (add_transaction_credits(journal, blocks, rsv_blocks))
346 goto repeat;
347 } else {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700348 /*
Jan Kara8f7d89f2013-06-04 12:35:11 -0400349 * We have handle reserved so we are allowed to join T_LOCKED
350 * transaction and we don't have to check for transaction size
351 * and journal space.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700352 */
Jan Kara8f7d89f2013-06-04 12:35:11 -0400353 sub_reserved_credits(journal, blocks);
354 handle->h_reserved = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700355 }
356
357 /* OK, account for the buffers that this operation expects to
Theodore Ts'o8dd42042010-08-03 21:38:29 -0400358 * use and add the handle to the running transaction.
Theodore Ts'o8dd42042010-08-03 21:38:29 -0400359 */
Tao Ma28e35e42011-05-22 21:45:26 -0400360 update_t_max_wait(transaction, ts);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700361 handle->h_transaction = transaction;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400362 handle->h_requested_credits = blocks;
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500363 handle->h_start_jiffies = jiffies;
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400364 atomic_inc(&transaction->t_updates);
Theodore Ts'o8dd42042010-08-03 21:38:29 -0400365 atomic_inc(&transaction->t_handle_count);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400366 jbd_debug(4, "Handle %p given %d credits (total %d, free %lu)\n",
367 handle, blocks,
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400368 atomic_read(&transaction->t_outstanding_credits),
Jan Kara76c39902013-06-04 12:12:57 -0400369 jbd2_log_space_left(journal));
Theodore Ts'oa931da62010-08-03 21:35:12 -0400370 read_unlock(&journal->j_state_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400371 current->journal_info = handle;
Jan Kara9599b0e2009-08-17 21:23:17 -0400372
373 lock_map_acquire(&handle->h_lockdep_map);
Yongqiang Yang0c2022e2012-02-20 17:53:02 -0500374 jbd2_journal_free_transaction(new_transaction);
Theodore Ts'o47def822010-07-27 11:56:05 -0400375 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700376}
377
Mingming Cao7b751062008-01-28 23:58:27 -0500378static struct lock_class_key jbd2_handle_key;
379
Dave Kleikamp470decc2006-10-11 01:20:57 -0700380/* Allocate a new handle. This should probably be in a slab... */
381static handle_t *new_handle(int nblocks)
382{
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400383 handle_t *handle = jbd2_alloc_handle(GFP_NOFS);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700384 if (!handle)
385 return NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700386 handle->h_buffer_credits = nblocks;
387 handle->h_ref = 1;
388
Mingming Cao7b751062008-01-28 23:58:27 -0500389 lockdep_init_map(&handle->h_lockdep_map, "jbd2_handle",
390 &jbd2_handle_key, 0);
391
Dave Kleikamp470decc2006-10-11 01:20:57 -0700392 return handle;
393}
394
395/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700396 * handle_t *jbd2_journal_start() - Obtain a new handle.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700397 * @journal: Journal to start transaction on.
398 * @nblocks: number of block buffer we might modify
399 *
400 * We make sure that the transaction can guarantee at least nblocks of
401 * modified buffers in the log. We block until the log can guarantee
Jan Kara8f7d89f2013-06-04 12:35:11 -0400402 * that much space. Additionally, if rsv_blocks > 0, we also create another
403 * handle with rsv_blocks reserved blocks in the journal. This handle is
404 * is stored in h_rsv_handle. It is not attached to any particular transaction
405 * and thus doesn't block transaction commit. If the caller uses this reserved
406 * handle, it has to set h_rsv_handle to NULL as otherwise jbd2_journal_stop()
407 * on the parent handle will dispose the reserved one. Reserved handle has to
408 * be converted to a normal handle using jbd2_journal_start_reserved() before
409 * it can be used.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700410 *
Eryu Guanc8675162011-05-24 17:09:58 -0400411 * Return a pointer to a newly allocated handle, or an ERR_PTR() value
412 * on failure.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700413 */
Jan Kara8f7d89f2013-06-04 12:35:11 -0400414handle_t *jbd2__journal_start(journal_t *journal, int nblocks, int rsv_blocks,
415 gfp_t gfp_mask, unsigned int type,
416 unsigned int line_no)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700417{
418 handle_t *handle = journal_current_handle();
419 int err;
420
421 if (!journal)
422 return ERR_PTR(-EROFS);
423
424 if (handle) {
425 J_ASSERT(handle->h_transaction->t_journal == journal);
426 handle->h_ref++;
427 return handle;
428 }
429
430 handle = new_handle(nblocks);
431 if (!handle)
432 return ERR_PTR(-ENOMEM);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400433 if (rsv_blocks) {
434 handle_t *rsv_handle;
435
436 rsv_handle = new_handle(rsv_blocks);
437 if (!rsv_handle) {
438 jbd2_free_handle(handle);
439 return ERR_PTR(-ENOMEM);
440 }
441 rsv_handle->h_reserved = 1;
442 rsv_handle->h_journal = journal;
443 handle->h_rsv_handle = rsv_handle;
444 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700445
Theodore Ts'o47def822010-07-27 11:56:05 -0400446 err = start_this_handle(journal, handle, gfp_mask);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700447 if (err < 0) {
Jan Kara8f7d89f2013-06-04 12:35:11 -0400448 if (handle->h_rsv_handle)
449 jbd2_free_handle(handle->h_rsv_handle);
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400450 jbd2_free_handle(handle);
Dmitry Monakhovdf05c1b82013-03-02 17:08:46 -0500451 return ERR_PTR(err);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700452 }
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500453 handle->h_type = type;
454 handle->h_line_no = line_no;
455 trace_jbd2_handle_start(journal->j_fs_dev->bd_dev,
456 handle->h_transaction->t_tid, type,
457 line_no, nblocks);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700458 return handle;
459}
Theodore Ts'o47def822010-07-27 11:56:05 -0400460EXPORT_SYMBOL(jbd2__journal_start);
461
462
463handle_t *jbd2_journal_start(journal_t *journal, int nblocks)
464{
Jan Kara8f7d89f2013-06-04 12:35:11 -0400465 return jbd2__journal_start(journal, nblocks, 0, GFP_NOFS, 0, 0);
Theodore Ts'o47def822010-07-27 11:56:05 -0400466}
467EXPORT_SYMBOL(jbd2_journal_start);
468
Jan Kara8f7d89f2013-06-04 12:35:11 -0400469void jbd2_journal_free_reserved(handle_t *handle)
470{
471 journal_t *journal = handle->h_journal;
472
473 WARN_ON(!handle->h_reserved);
474 sub_reserved_credits(journal, handle->h_buffer_credits);
475 jbd2_free_handle(handle);
476}
477EXPORT_SYMBOL(jbd2_journal_free_reserved);
478
479/**
480 * int jbd2_journal_start_reserved(handle_t *handle) - start reserved handle
481 * @handle: handle to start
482 *
483 * Start handle that has been previously reserved with jbd2_journal_reserve().
484 * This attaches @handle to the running transaction (or creates one if there's
485 * not transaction running). Unlike jbd2_journal_start() this function cannot
486 * block on journal commit, checkpointing, or similar stuff. It can block on
487 * memory allocation or frozen journal though.
488 *
489 * Return 0 on success, non-zero on error - handle is freed in that case.
490 */
491int jbd2_journal_start_reserved(handle_t *handle, unsigned int type,
492 unsigned int line_no)
493{
494 journal_t *journal = handle->h_journal;
495 int ret = -EIO;
496
497 if (WARN_ON(!handle->h_reserved)) {
498 /* Someone passed in normal handle? Just stop it. */
499 jbd2_journal_stop(handle);
500 return ret;
501 }
502 /*
503 * Usefulness of mixing of reserved and unreserved handles is
504 * questionable. So far nobody seems to need it so just error out.
505 */
506 if (WARN_ON(current->journal_info)) {
507 jbd2_journal_free_reserved(handle);
508 return ret;
509 }
510
511 handle->h_journal = NULL;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400512 /*
513 * GFP_NOFS is here because callers are likely from writeback or
514 * similarly constrained call sites
515 */
516 ret = start_this_handle(journal, handle, GFP_NOFS);
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400517 if (ret < 0)
Jan Kara8f7d89f2013-06-04 12:35:11 -0400518 jbd2_journal_free_reserved(handle);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400519 handle->h_type = type;
520 handle->h_line_no = line_no;
521 return ret;
522}
523EXPORT_SYMBOL(jbd2_journal_start_reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700524
525/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700526 * int jbd2_journal_extend() - extend buffer credits.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700527 * @handle: handle to 'extend'
528 * @nblocks: nr blocks to try to extend by.
529 *
530 * Some transactions, such as large extends and truncates, can be done
531 * atomically all at once or in several stages. The operation requests
532 * a credit for a number of buffer modications in advance, but can
533 * extend its credit if it needs more.
534 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700535 * jbd2_journal_extend tries to give the running handle more buffer credits.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700536 * It does not guarantee that allocation - this is a best-effort only.
537 * The calling process MUST be able to deal cleanly with a failure to
538 * extend here.
539 *
540 * Return 0 on success, non-zero on failure.
541 *
542 * return code < 0 implies an error
543 * return code > 0 implies normal transaction-full status.
544 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700545int jbd2_journal_extend(handle_t *handle, int nblocks)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700546{
547 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400548 journal_t *journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700549 int result;
550 int wanted;
551
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400552 WARN_ON(!transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700553 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400554 return -EROFS;
555 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700556
557 result = 1;
558
Theodore Ts'oa931da62010-08-03 21:35:12 -0400559 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700560
561 /* Don't extend a locked-down transaction! */
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400562 if (transaction->t_state != T_RUNNING) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700563 jbd_debug(3, "denied handle %p %d blocks: "
564 "transaction not running\n", handle, nblocks);
565 goto error_out;
566 }
567
568 spin_lock(&transaction->t_handle_lock);
Jan Karafe1e8db2013-06-04 12:22:15 -0400569 wanted = atomic_add_return(nblocks,
570 &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700571
572 if (wanted > journal->j_max_transaction_buffers) {
573 jbd_debug(3, "denied handle %p %d blocks: "
574 "transaction too large\n", handle, nblocks);
Jan Karafe1e8db2013-06-04 12:22:15 -0400575 atomic_sub(nblocks, &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700576 goto unlock;
577 }
578
Jan Kara76c39902013-06-04 12:12:57 -0400579 if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
580 jbd2_log_space_left(journal)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700581 jbd_debug(3, "denied handle %p %d blocks: "
582 "insufficient log space\n", handle, nblocks);
Jan Karafe1e8db2013-06-04 12:22:15 -0400583 atomic_sub(nblocks, &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700584 goto unlock;
585 }
586
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500587 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400588 transaction->t_tid,
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500589 handle->h_type, handle->h_line_no,
590 handle->h_buffer_credits,
591 nblocks);
592
Dave Kleikamp470decc2006-10-11 01:20:57 -0700593 handle->h_buffer_credits += nblocks;
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500594 handle->h_requested_credits += nblocks;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700595 result = 0;
596
597 jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
598unlock:
599 spin_unlock(&transaction->t_handle_lock);
600error_out:
Theodore Ts'oa931da62010-08-03 21:35:12 -0400601 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700602 return result;
603}
604
605
606/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700607 * int jbd2_journal_restart() - restart a handle .
Dave Kleikamp470decc2006-10-11 01:20:57 -0700608 * @handle: handle to restart
609 * @nblocks: nr credits requested
610 *
611 * Restart a handle for a multi-transaction filesystem
612 * operation.
613 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700614 * If the jbd2_journal_extend() call above fails to grant new buffer credits
615 * to a running handle, a call to jbd2_journal_restart will commit the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700616 * handle's transaction so far and reattach the handle to a new
617 * transaction capabable of guaranteeing the requested number of
Jan Kara8f7d89f2013-06-04 12:35:11 -0400618 * credits. We preserve reserved handle if there's any attached to the
619 * passed in handle.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700620 */
Dan Carpenterd2159fb2011-09-04 10:20:14 -0400621int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700622{
623 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400624 journal_t *journal;
Theodore Ts'oe4471832011-02-12 08:18:24 -0500625 tid_t tid;
626 int need_to_start, ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700627
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400628 WARN_ON(!transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700629 /* If we've had an abort of any type, don't even think about
630 * actually doing the restart! */
631 if (is_handle_aborted(handle))
632 return 0;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400633 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700634
635 /*
636 * First unlink the handle from its current transaction, and start the
637 * commit on that.
638 */
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400639 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700640 J_ASSERT(journal_current_handle() == handle);
641
Theodore Ts'oa931da62010-08-03 21:35:12 -0400642 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700643 spin_lock(&transaction->t_handle_lock);
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400644 atomic_sub(handle->h_buffer_credits,
645 &transaction->t_outstanding_credits);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400646 if (handle->h_rsv_handle) {
647 sub_reserved_credits(journal,
648 handle->h_rsv_handle->h_buffer_credits);
649 }
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400650 if (atomic_dec_and_test(&transaction->t_updates))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700651 wake_up(&journal->j_wait_updates);
Theodore Ts'o39c04152013-07-01 08:12:40 -0400652 tid = transaction->t_tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700653 spin_unlock(&transaction->t_handle_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400654 handle->h_transaction = NULL;
655 current->journal_info = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700656
657 jbd_debug(2, "restarting handle %p\n", handle);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500658 need_to_start = !tid_geq(journal->j_commit_request, tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400659 read_unlock(&journal->j_state_lock);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500660 if (need_to_start)
661 jbd2_log_start_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700662
Jan Kara9599b0e2009-08-17 21:23:17 -0400663 lock_map_release(&handle->h_lockdep_map);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700664 handle->h_buffer_credits = nblocks;
Theodore Ts'o47def822010-07-27 11:56:05 -0400665 ret = start_this_handle(journal, handle, gfp_mask);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700666 return ret;
667}
Theodore Ts'o47def822010-07-27 11:56:05 -0400668EXPORT_SYMBOL(jbd2__journal_restart);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700669
670
Theodore Ts'o47def822010-07-27 11:56:05 -0400671int jbd2_journal_restart(handle_t *handle, int nblocks)
672{
673 return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
674}
675EXPORT_SYMBOL(jbd2_journal_restart);
676
Dave Kleikamp470decc2006-10-11 01:20:57 -0700677/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700678 * void jbd2_journal_lock_updates () - establish a transaction barrier.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700679 * @journal: Journal to establish a barrier on.
680 *
681 * This locks out any further updates from being started, and blocks
682 * until all existing updates have completed, returning only once the
683 * journal is in a quiescent state with no updates running.
684 *
685 * The journal lock should not be held on entry.
686 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700687void jbd2_journal_lock_updates(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700688{
689 DEFINE_WAIT(wait);
690
Theodore Ts'oa931da62010-08-03 21:35:12 -0400691 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700692 ++journal->j_barrier_count;
693
Jan Kara8f7d89f2013-06-04 12:35:11 -0400694 /* Wait until there are no reserved handles */
695 if (atomic_read(&journal->j_reserved_credits)) {
696 write_unlock(&journal->j_state_lock);
697 wait_event(journal->j_wait_reserved,
698 atomic_read(&journal->j_reserved_credits) == 0);
699 write_lock(&journal->j_state_lock);
700 }
701
Dave Kleikamp470decc2006-10-11 01:20:57 -0700702 /* Wait until there are no running updates */
703 while (1) {
704 transaction_t *transaction = journal->j_running_transaction;
705
706 if (!transaction)
707 break;
708
709 spin_lock(&transaction->t_handle_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700710 prepare_to_wait(&journal->j_wait_updates, &wait,
711 TASK_UNINTERRUPTIBLE);
Jan Kara9837d8e2012-01-04 22:03:11 -0500712 if (!atomic_read(&transaction->t_updates)) {
713 spin_unlock(&transaction->t_handle_lock);
714 finish_wait(&journal->j_wait_updates, &wait);
715 break;
716 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700717 spin_unlock(&transaction->t_handle_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400718 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700719 schedule();
720 finish_wait(&journal->j_wait_updates, &wait);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400721 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700722 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400723 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700724
725 /*
726 * We have now established a barrier against other normal updates, but
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700727 * we also need to barrier against other jbd2_journal_lock_updates() calls
Dave Kleikamp470decc2006-10-11 01:20:57 -0700728 * to make sure that we serialise special journal-locked operations
729 * too.
730 */
731 mutex_lock(&journal->j_barrier);
732}
733
734/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700735 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
Dave Kleikamp470decc2006-10-11 01:20:57 -0700736 * @journal: Journal to release the barrier on.
737 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700738 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
Dave Kleikamp470decc2006-10-11 01:20:57 -0700739 *
740 * Should be called without the journal lock held.
741 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700742void jbd2_journal_unlock_updates (journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700743{
744 J_ASSERT(journal->j_barrier_count != 0);
745
746 mutex_unlock(&journal->j_barrier);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400747 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700748 --journal->j_barrier_count;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400749 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700750 wake_up(&journal->j_wait_transaction_locked);
751}
752
Jan Karaf91d1d02009-07-13 16:16:20 -0400753static void warn_dirty_buffer(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700754{
Jan Karaf91d1d02009-07-13 16:16:20 -0400755 char b[BDEVNAME_SIZE];
Dave Kleikamp470decc2006-10-11 01:20:57 -0700756
Jan Karaf91d1d02009-07-13 16:16:20 -0400757 printk(KERN_WARNING
Eryu Guanf2a44522011-11-01 19:09:18 -0400758 "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). "
Jan Karaf91d1d02009-07-13 16:16:20 -0400759 "There's a risk of filesystem corruption in case of system "
760 "crash.\n",
761 bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700762}
763
Jan Karab34090e2013-06-04 12:08:56 -0400764static int sleep_on_shadow_bh(void *word)
765{
766 io_schedule();
767 return 0;
768}
769
Dave Kleikamp470decc2006-10-11 01:20:57 -0700770/*
771 * If the buffer is already part of the current transaction, then there
772 * is nothing we need to do. If it is already part of a prior
773 * transaction which we are still committing to disk, then we need to
774 * make sure that we do not overwrite the old copy: we do copy-out to
775 * preserve the copy going to disk. We also account the buffer against
776 * the handle's metadata buffer credits (unless the buffer is already
777 * part of the transaction, that is).
778 *
779 */
780static int
781do_get_write_access(handle_t *handle, struct journal_head *jh,
782 int force_copy)
783{
784 struct buffer_head *bh;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400785 transaction_t *transaction = handle->h_transaction;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700786 journal_t *journal;
787 int error;
788 char *frozen_buffer = NULL;
789 int need_copy = 0;
Theodore Ts'of783f092013-04-21 16:47:54 -0400790 unsigned long start_lock, time_lock;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700791
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400792 WARN_ON(!transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700793 if (is_handle_aborted(handle))
794 return -EROFS;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700795 journal = transaction->t_journal;
796
Theodore Ts'ocfef2c62010-12-18 13:07:34 -0500797 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700798
799 JBUFFER_TRACE(jh, "entry");
800repeat:
801 bh = jh2bh(jh);
802
803 /* @@@ Need to check for errors here at some point. */
804
Theodore Ts'of783f092013-04-21 16:47:54 -0400805 start_lock = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700806 lock_buffer(bh);
807 jbd_lock_bh_state(bh);
808
Theodore Ts'of783f092013-04-21 16:47:54 -0400809 /* If it takes too long to lock the buffer, trace it */
810 time_lock = jbd2_time_diff(start_lock, jiffies);
811 if (time_lock > HZ/10)
812 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
813 jiffies_to_msecs(time_lock));
814
Dave Kleikamp470decc2006-10-11 01:20:57 -0700815 /* We now hold the buffer lock so it is safe to query the buffer
816 * state. Is the buffer dirty?
817 *
818 * If so, there are two possibilities. The buffer may be
819 * non-journaled, and undergoing a quite legitimate writeback.
820 * Otherwise, it is journaled, and we don't expect dirty buffers
821 * in that state (the buffers should be marked JBD_Dirty
822 * instead.) So either the IO is being done under our own
823 * control and this is a bug, or it's a third party IO such as
824 * dump(8) (which may leave the buffer scheduled for read ---
825 * ie. locked but not dirty) or tune2fs (which may actually have
826 * the buffer dirtied, ugh.) */
827
828 if (buffer_dirty(bh)) {
829 /*
830 * First question: is this buffer already part of the current
831 * transaction or the existing committing transaction?
832 */
833 if (jh->b_transaction) {
834 J_ASSERT_JH(jh,
835 jh->b_transaction == transaction ||
836 jh->b_transaction ==
837 journal->j_committing_transaction);
838 if (jh->b_next_transaction)
839 J_ASSERT_JH(jh, jh->b_next_transaction ==
840 transaction);
Jan Karaf91d1d02009-07-13 16:16:20 -0400841 warn_dirty_buffer(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700842 }
843 /*
844 * In any case we need to clean the dirty flag and we must
845 * do it under the buffer lock to be sure we don't race
846 * with running write-out.
847 */
Jan Karaf91d1d02009-07-13 16:16:20 -0400848 JBUFFER_TRACE(jh, "Journalling dirty buffer");
849 clear_buffer_dirty(bh);
850 set_buffer_jbddirty(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700851 }
852
853 unlock_buffer(bh);
854
855 error = -EROFS;
856 if (is_handle_aborted(handle)) {
857 jbd_unlock_bh_state(bh);
858 goto out;
859 }
860 error = 0;
861
862 /*
863 * The buffer is already part of this transaction if b_transaction or
864 * b_next_transaction points to it
865 */
866 if (jh->b_transaction == transaction ||
867 jh->b_next_transaction == transaction)
868 goto done;
869
870 /*
Josef Bacik9fc7c632008-04-17 10:38:59 -0400871 * this is the first time this transaction is touching this buffer,
872 * reset the modified flag
873 */
874 jh->b_modified = 0;
875
876 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700877 * If there is already a copy-out version of this buffer, then we don't
878 * need to make another one
879 */
880 if (jh->b_frozen_data) {
881 JBUFFER_TRACE(jh, "has frozen data");
882 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
883 jh->b_next_transaction = transaction;
884 goto done;
885 }
886
887 /* Is there data here we need to preserve? */
888
889 if (jh->b_transaction && jh->b_transaction != transaction) {
890 JBUFFER_TRACE(jh, "owned by older transaction");
891 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
892 J_ASSERT_JH(jh, jh->b_transaction ==
893 journal->j_committing_transaction);
894
895 /* There is one case we have to be very careful about.
896 * If the committing transaction is currently writing
897 * this buffer out to disk and has NOT made a copy-out,
898 * then we cannot modify the buffer contents at all
899 * right now. The essence of copy-out is that it is the
900 * extra copy, not the primary copy, which gets
901 * journaled. If the primary copy is already going to
902 * disk then we cannot do copy-out here. */
903
Jan Karab34090e2013-06-04 12:08:56 -0400904 if (buffer_shadow(bh)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700905 JBUFFER_TRACE(jh, "on shadow: sleep");
906 jbd_unlock_bh_state(bh);
Jan Karab34090e2013-06-04 12:08:56 -0400907 wait_on_bit(&bh->b_state, BH_Shadow,
908 sleep_on_shadow_bh, TASK_UNINTERRUPTIBLE);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700909 goto repeat;
910 }
911
Jan Karab34090e2013-06-04 12:08:56 -0400912 /*
913 * Only do the copy if the currently-owning transaction still
914 * needs it. If buffer isn't on BJ_Metadata list, the
915 * committing transaction is past that stage (here we use the
916 * fact that BH_Shadow is set under bh_state lock together with
917 * refiling to BJ_Shadow list and at this point we know the
918 * buffer doesn't have BH_Shadow set).
Dave Kleikamp470decc2006-10-11 01:20:57 -0700919 *
920 * Subtle point, though: if this is a get_undo_access,
921 * then we will be relying on the frozen_data to contain
922 * the new value of the committed_data record after the
923 * transaction, so we HAVE to force the frozen_data copy
Jan Karab34090e2013-06-04 12:08:56 -0400924 * in that case.
925 */
926 if (jh->b_jlist == BJ_Metadata || force_copy) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700927 JBUFFER_TRACE(jh, "generate frozen data");
928 if (!frozen_buffer) {
929 JBUFFER_TRACE(jh, "allocate memory for buffer");
930 jbd_unlock_bh_state(bh);
931 frozen_buffer =
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400932 jbd2_alloc(jh2bh(jh)->b_size,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700933 GFP_NOFS);
934 if (!frozen_buffer) {
Jan Kara75685072013-12-08 21:13:59 -0500935 printk(KERN_ERR
Dave Kleikamp470decc2006-10-11 01:20:57 -0700936 "%s: OOM for frozen_buffer\n",
Harvey Harrison329d2912008-04-17 10:38:59 -0400937 __func__);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700938 JBUFFER_TRACE(jh, "oom!");
939 error = -ENOMEM;
940 jbd_lock_bh_state(bh);
941 goto done;
942 }
943 goto repeat;
944 }
945 jh->b_frozen_data = frozen_buffer;
946 frozen_buffer = NULL;
947 need_copy = 1;
948 }
949 jh->b_next_transaction = transaction;
950 }
951
952
953 /*
954 * Finally, if the buffer is not journaled right now, we need to make
955 * sure it doesn't get written to disk before the caller actually
956 * commits the new data
957 */
958 if (!jh->b_transaction) {
959 JBUFFER_TRACE(jh, "no transaction");
960 J_ASSERT_JH(jh, !jh->b_next_transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700961 JBUFFER_TRACE(jh, "file as BJ_Reserved");
962 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700963 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700964 spin_unlock(&journal->j_list_lock);
965 }
966
967done:
968 if (need_copy) {
969 struct page *page;
970 int offset;
971 char *source;
972
973 J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)),
974 "Possible IO failure.\n");
975 page = jh2bh(jh)->b_page;
Theodore Ts'oa1dd5332010-12-18 13:13:40 -0500976 offset = offset_in_page(jh2bh(jh)->b_data);
Cong Wang303a8f22011-11-25 23:14:31 +0800977 source = kmap_atomic(page);
Jan Kara13ceef02010-07-14 07:56:33 +0200978 /* Fire data frozen trigger just before we copy the data */
979 jbd2_buffer_frozen_trigger(jh, source + offset,
980 jh->b_triggers);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700981 memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size);
Cong Wang303a8f22011-11-25 23:14:31 +0800982 kunmap_atomic(source);
Joel Beckere06c8222008-09-11 15:35:47 -0700983
984 /*
985 * Now that the frozen data is saved off, we need to store
986 * any matching triggers.
987 */
988 jh->b_frozen_triggers = jh->b_triggers;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700989 }
990 jbd_unlock_bh_state(bh);
991
992 /*
993 * If we are about to journal a buffer, then any revoke pending on it is
994 * no longer valid
995 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700996 jbd2_journal_cancel_revoke(handle, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700997
998out:
999 if (unlikely(frozen_buffer)) /* It's usually NULL */
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001000 jbd2_free(frozen_buffer, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001001
1002 JBUFFER_TRACE(jh, "exit");
1003 return error;
1004}
1005
1006/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001007 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001008 * @handle: transaction to add buffer modifications to
1009 * @bh: bh to be used for metadata writes
Dave Kleikamp470decc2006-10-11 01:20:57 -07001010 *
1011 * Returns an error code or 0 on success.
1012 *
1013 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1014 * because we're write()ing a buffer which is also part of a shared mapping.
1015 */
1016
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001017int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001018{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001019 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001020 int rc;
1021
1022 /* We do not want to get caught playing with fields which the
1023 * log thread also manipulates. Make sure that the buffer
1024 * completes any outstanding IO before proceeding. */
1025 rc = do_get_write_access(handle, jh, 0);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001026 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001027 return rc;
1028}
1029
1030
1031/*
1032 * When the user wants to journal a newly created buffer_head
1033 * (ie. getblk() returned a new buffer and we are going to populate it
1034 * manually rather than reading off disk), then we need to keep the
1035 * buffer_head locked until it has been completely filled with new
1036 * data. In this case, we should be able to make the assertion that
1037 * the bh is not already part of an existing transaction.
1038 *
1039 * The buffer should already be locked by the caller by this point.
1040 * There is no lock ranking violation: it was a newly created,
1041 * unlocked buffer beforehand. */
1042
1043/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001044 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
Dave Kleikamp470decc2006-10-11 01:20:57 -07001045 * @handle: transaction to new buffer to
1046 * @bh: new buffer.
1047 *
1048 * Call this if you create a new bh.
1049 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001050int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001051{
1052 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001053 journal_t *journal;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001054 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001055 int err;
1056
1057 jbd_debug(5, "journal_head %p\n", jh);
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001058 WARN_ON(!transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001059 err = -EROFS;
1060 if (is_handle_aborted(handle))
1061 goto out;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001062 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001063 err = 0;
1064
1065 JBUFFER_TRACE(jh, "entry");
1066 /*
1067 * The buffer may already belong to this transaction due to pre-zeroing
1068 * in the filesystem's new_block code. It may also be on the previous,
1069 * committing transaction's lists, but it HAS to be in Forget state in
1070 * that case: the transaction must have deleted the buffer for it to be
1071 * reused here.
1072 */
1073 jbd_lock_bh_state(bh);
1074 spin_lock(&journal->j_list_lock);
1075 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1076 jh->b_transaction == NULL ||
1077 (jh->b_transaction == journal->j_committing_transaction &&
1078 jh->b_jlist == BJ_Forget)));
1079
1080 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1081 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1082
1083 if (jh->b_transaction == NULL) {
Jan Karaf91d1d02009-07-13 16:16:20 -04001084 /*
1085 * Previous jbd2_journal_forget() could have left the buffer
1086 * with jbddirty bit set because it was being committed. When
1087 * the commit finished, we've filed the buffer for
1088 * checkpointing and marked it dirty. Now we are reallocating
1089 * the buffer so the transaction freeing it must have
1090 * committed and so it's safe to clear the dirty bit.
1091 */
1092 clear_buffer_dirty(jh2bh(jh));
Josef Bacik9fc7c632008-04-17 10:38:59 -04001093 /* first access by this transaction */
1094 jh->b_modified = 0;
1095
Dave Kleikamp470decc2006-10-11 01:20:57 -07001096 JBUFFER_TRACE(jh, "file as BJ_Reserved");
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001097 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001098 } else if (jh->b_transaction == journal->j_committing_transaction) {
Josef Bacik9fc7c632008-04-17 10:38:59 -04001099 /* first access by this transaction */
1100 jh->b_modified = 0;
1101
Dave Kleikamp470decc2006-10-11 01:20:57 -07001102 JBUFFER_TRACE(jh, "set next transaction");
1103 jh->b_next_transaction = transaction;
1104 }
1105 spin_unlock(&journal->j_list_lock);
1106 jbd_unlock_bh_state(bh);
1107
1108 /*
1109 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1110 * blocks which contain freed but then revoked metadata. We need
1111 * to cancel the revoke in case we end up freeing it yet again
1112 * and the reallocating as data - this would cause a second revoke,
1113 * which hits an assertion error.
1114 */
1115 JBUFFER_TRACE(jh, "cancelling revoke");
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001116 jbd2_journal_cancel_revoke(handle, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001117out:
Ding Dinghua3991b402011-05-25 17:43:48 -04001118 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001119 return err;
1120}
1121
1122/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001123 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
Dave Kleikamp470decc2006-10-11 01:20:57 -07001124 * non-rewindable consequences
1125 * @handle: transaction
1126 * @bh: buffer to undo
Dave Kleikamp470decc2006-10-11 01:20:57 -07001127 *
1128 * Sometimes there is a need to distinguish between metadata which has
1129 * been committed to disk and that which has not. The ext3fs code uses
1130 * this for freeing and allocating space, we have to make sure that we
1131 * do not reuse freed space until the deallocation has been committed,
1132 * since if we overwrote that space we would make the delete
1133 * un-rewindable in case of a crash.
1134 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001135 * To deal with that, jbd2_journal_get_undo_access requests write access to a
Dave Kleikamp470decc2006-10-11 01:20:57 -07001136 * buffer for parts of non-rewindable operations such as delete
1137 * operations on the bitmaps. The journaling code must keep a copy of
1138 * the buffer's contents prior to the undo_access call until such time
1139 * as we know that the buffer has definitely been committed to disk.
1140 *
1141 * We never need to know which transaction the committed data is part
1142 * of, buffers touched here are guaranteed to be dirtied later and so
1143 * will be committed to a new transaction in due course, at which point
1144 * we can discard the old committed data pointer.
1145 *
1146 * Returns error number or 0 on success.
1147 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001148int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001149{
1150 int err;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001151 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001152 char *committed_data = NULL;
1153
1154 JBUFFER_TRACE(jh, "entry");
1155
1156 /*
1157 * Do this first --- it can drop the journal lock, so we want to
1158 * make sure that obtaining the committed_data is done
1159 * atomically wrt. completion of any outstanding commits.
1160 */
1161 err = do_get_write_access(handle, jh, 1);
1162 if (err)
1163 goto out;
1164
1165repeat:
1166 if (!jh->b_committed_data) {
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001167 committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001168 if (!committed_data) {
Jan Kara75685072013-12-08 21:13:59 -05001169 printk(KERN_ERR "%s: No memory for committed data\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001170 __func__);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001171 err = -ENOMEM;
1172 goto out;
1173 }
1174 }
1175
1176 jbd_lock_bh_state(bh);
1177 if (!jh->b_committed_data) {
1178 /* Copy out the current buffer contents into the
1179 * preserved, committed copy. */
1180 JBUFFER_TRACE(jh, "generate b_committed data");
1181 if (!committed_data) {
1182 jbd_unlock_bh_state(bh);
1183 goto repeat;
1184 }
1185
1186 jh->b_committed_data = committed_data;
1187 committed_data = NULL;
1188 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1189 }
1190 jbd_unlock_bh_state(bh);
1191out:
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001192 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001193 if (unlikely(committed_data))
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001194 jbd2_free(committed_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001195 return err;
1196}
1197
1198/**
Joel Beckere06c8222008-09-11 15:35:47 -07001199 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1200 * @bh: buffer to trigger on
1201 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1202 *
1203 * Set any triggers on this journal_head. This is always safe, because
1204 * triggers for a committing buffer will be saved off, and triggers for
1205 * a running transaction will match the buffer in that transaction.
1206 *
1207 * Call with NULL to clear the triggers.
1208 */
1209void jbd2_journal_set_triggers(struct buffer_head *bh,
1210 struct jbd2_buffer_trigger_type *type)
1211{
Jan Karaad56eda2013-03-11 13:24:56 -04001212 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
Joel Beckere06c8222008-09-11 15:35:47 -07001213
Jan Karaad56eda2013-03-11 13:24:56 -04001214 if (WARN_ON(!jh))
1215 return;
Joel Beckere06c8222008-09-11 15:35:47 -07001216 jh->b_triggers = type;
Jan Karaad56eda2013-03-11 13:24:56 -04001217 jbd2_journal_put_journal_head(jh);
Joel Beckere06c8222008-09-11 15:35:47 -07001218}
1219
Jan Kara13ceef02010-07-14 07:56:33 +02001220void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
Joel Beckere06c8222008-09-11 15:35:47 -07001221 struct jbd2_buffer_trigger_type *triggers)
1222{
1223 struct buffer_head *bh = jh2bh(jh);
1224
Jan Kara13ceef02010-07-14 07:56:33 +02001225 if (!triggers || !triggers->t_frozen)
Joel Beckere06c8222008-09-11 15:35:47 -07001226 return;
1227
Jan Kara13ceef02010-07-14 07:56:33 +02001228 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
Joel Beckere06c8222008-09-11 15:35:47 -07001229}
1230
1231void jbd2_buffer_abort_trigger(struct journal_head *jh,
1232 struct jbd2_buffer_trigger_type *triggers)
1233{
1234 if (!triggers || !triggers->t_abort)
1235 return;
1236
1237 triggers->t_abort(triggers, jh2bh(jh));
1238}
1239
1240
1241
1242/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001243 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
Dave Kleikamp470decc2006-10-11 01:20:57 -07001244 * @handle: transaction to add buffer to.
1245 * @bh: buffer to mark
1246 *
1247 * mark dirty metadata which needs to be journaled as part of the current
1248 * transaction.
1249 *
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001250 * The buffer must have previously had jbd2_journal_get_write_access()
1251 * called so that it has a valid journal_head attached to the buffer
1252 * head.
1253 *
Dave Kleikamp470decc2006-10-11 01:20:57 -07001254 * The buffer is placed on the transaction's metadata list and is marked
1255 * as belonging to the transaction.
1256 *
1257 * Returns error number or 0 on success.
1258 *
1259 * Special care needs to be taken if the buffer already belongs to the
1260 * current committing transaction (in which case we should have frozen
1261 * data present for that commit). In that case, we don't relink the
1262 * buffer: that only gets done when the old transaction finally
1263 * completes its commit.
1264 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001265int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001266{
1267 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001268 journal_t *journal;
Jan Karaad56eda2013-03-11 13:24:56 -04001269 struct journal_head *jh;
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001270 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001271
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001272 WARN_ON(!transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001273 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001274 return -EROFS;
1275 journal = transaction->t_journal;
Jan Karaad56eda2013-03-11 13:24:56 -04001276 jh = jbd2_journal_grab_journal_head(bh);
1277 if (!jh) {
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001278 ret = -EUCLEAN;
1279 goto out;
1280 }
Jan Karaad56eda2013-03-11 13:24:56 -04001281 jbd_debug(5, "journal_head %p\n", jh);
1282 JBUFFER_TRACE(jh, "entry");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001283
1284 jbd_lock_bh_state(bh);
1285
1286 if (jh->b_modified == 0) {
1287 /*
1288 * This buffer's got modified and becoming part
1289 * of the transaction. This needs to be done
1290 * once a transaction -bzzz
1291 */
1292 jh->b_modified = 1;
Theodore Ts'of6c07ca2013-12-08 21:12:59 -05001293 if (handle->h_buffer_credits <= 0) {
1294 ret = -ENOSPC;
1295 goto out_unlock_bh;
1296 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001297 handle->h_buffer_credits--;
1298 }
1299
1300 /*
1301 * fastpath, to avoid expensive locking. If this buffer is already
1302 * on the running transaction's metadata list there is nothing to do.
1303 * Nobody can take it off again because there is a handle open.
1304 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1305 * result in this test being false, so we go in and take the locks.
1306 */
1307 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1308 JBUFFER_TRACE(jh, "fastpath");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001309 if (unlikely(jh->b_transaction !=
1310 journal->j_running_transaction)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001311 printk(KERN_ERR "JBD2: %s: "
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001312 "jh->b_transaction (%llu, %p, %u) != "
1313 "journal->j_running_transaction (%p, %u)",
1314 journal->j_devname,
1315 (unsigned long long) bh->b_blocknr,
1316 jh->b_transaction,
1317 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1318 journal->j_running_transaction,
1319 journal->j_running_transaction ?
1320 journal->j_running_transaction->t_tid : 0);
1321 ret = -EINVAL;
1322 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001323 goto out_unlock_bh;
1324 }
1325
1326 set_buffer_jbddirty(bh);
1327
1328 /*
1329 * Metadata already on the current transaction list doesn't
1330 * need to be filed. Metadata on another transaction's list must
1331 * be committing, and will be refiled once the commit completes:
1332 * leave it alone for now.
1333 */
1334 if (jh->b_transaction != transaction) {
1335 JBUFFER_TRACE(jh, "already on other transaction");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001336 if (unlikely(jh->b_transaction !=
1337 journal->j_committing_transaction)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001338 printk(KERN_ERR "JBD2: %s: "
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001339 "jh->b_transaction (%llu, %p, %u) != "
1340 "journal->j_committing_transaction (%p, %u)",
1341 journal->j_devname,
1342 (unsigned long long) bh->b_blocknr,
1343 jh->b_transaction,
1344 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1345 journal->j_committing_transaction,
1346 journal->j_committing_transaction ?
1347 journal->j_committing_transaction->t_tid : 0);
1348 ret = -EINVAL;
1349 }
1350 if (unlikely(jh->b_next_transaction != transaction)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001351 printk(KERN_ERR "JBD2: %s: "
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001352 "jh->b_next_transaction (%llu, %p, %u) != "
1353 "transaction (%p, %u)",
1354 journal->j_devname,
1355 (unsigned long long) bh->b_blocknr,
1356 jh->b_next_transaction,
1357 jh->b_next_transaction ?
1358 jh->b_next_transaction->t_tid : 0,
1359 transaction, transaction->t_tid);
1360 ret = -EINVAL;
1361 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001362 /* And this case is illegal: we can't reuse another
1363 * transaction's data buffer, ever. */
1364 goto out_unlock_bh;
1365 }
1366
1367 /* That test should have eliminated the following case: */
Mingming Cao40191912008-01-28 23:58:27 -05001368 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001369
1370 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1371 spin_lock(&journal->j_list_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001372 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001373 spin_unlock(&journal->j_list_lock);
1374out_unlock_bh:
1375 jbd_unlock_bh_state(bh);
Jan Karaad56eda2013-03-11 13:24:56 -04001376 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001377out:
1378 JBUFFER_TRACE(jh, "exit");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001379 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001380}
1381
Dave Kleikamp470decc2006-10-11 01:20:57 -07001382/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001383 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001384 * @handle: transaction handle
1385 * @bh: bh to 'forget'
1386 *
1387 * We can only do the bforget if there are no commits pending against the
1388 * buffer. If the buffer is dirty in the current running transaction we
1389 * can safely unlink it.
1390 *
1391 * bh may not be a journalled buffer at all - it may be a non-JBD
1392 * buffer which came off the hashtable. Check for this.
1393 *
1394 * Decrements bh->b_count by one.
1395 *
1396 * Allow this call even if the handle has aborted --- it may be part of
1397 * the caller's cleanup after an abort.
1398 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001399int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001400{
1401 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001402 journal_t *journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001403 struct journal_head *jh;
1404 int drop_reserve = 0;
1405 int err = 0;
Josef Bacik1dfc3222008-04-17 10:38:59 -04001406 int was_modified = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001407
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001408 WARN_ON(!transaction);
1409 if (is_handle_aborted(handle))
1410 return -EROFS;
1411 journal = transaction->t_journal;
1412
Dave Kleikamp470decc2006-10-11 01:20:57 -07001413 BUFFER_TRACE(bh, "entry");
1414
1415 jbd_lock_bh_state(bh);
1416 spin_lock(&journal->j_list_lock);
1417
1418 if (!buffer_jbd(bh))
1419 goto not_jbd;
1420 jh = bh2jh(bh);
1421
1422 /* Critical error: attempting to delete a bitmap buffer, maybe?
1423 * Don't do any jbd operations, and return an error. */
1424 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1425 "inconsistent data on disk")) {
1426 err = -EIO;
1427 goto not_jbd;
1428 }
1429
Adam Buchbinder48fc7f72012-09-19 21:48:00 -04001430 /* keep track of whether or not this transaction modified us */
Josef Bacik1dfc3222008-04-17 10:38:59 -04001431 was_modified = jh->b_modified;
1432
Dave Kleikamp470decc2006-10-11 01:20:57 -07001433 /*
1434 * The buffer's going from the transaction, we must drop
1435 * all references -bzzz
1436 */
1437 jh->b_modified = 0;
1438
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001439 if (jh->b_transaction == transaction) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001440 J_ASSERT_JH(jh, !jh->b_frozen_data);
1441
1442 /* If we are forgetting a buffer which is already part
1443 * of this transaction, then we can just drop it from
1444 * the transaction immediately. */
1445 clear_buffer_dirty(bh);
1446 clear_buffer_jbddirty(bh);
1447
1448 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1449
Josef Bacik1dfc3222008-04-17 10:38:59 -04001450 /*
1451 * we only want to drop a reference if this transaction
1452 * modified the buffer
1453 */
1454 if (was_modified)
1455 drop_reserve = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001456
1457 /*
1458 * We are no longer going to journal this buffer.
1459 * However, the commit of this transaction is still
1460 * important to the buffer: the delete that we are now
1461 * processing might obsolete an old log entry, so by
1462 * committing, we can satisfy the buffer's checkpoint.
1463 *
1464 * So, if we have a checkpoint on the buffer, we should
1465 * now refile the buffer on our BJ_Forget list so that
1466 * we know to remove the checkpoint after we commit.
1467 */
1468
1469 if (jh->b_cp_transaction) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001470 __jbd2_journal_temp_unlink_buffer(jh);
1471 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001472 } else {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001473 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001474 if (!buffer_jbd(bh)) {
1475 spin_unlock(&journal->j_list_lock);
1476 jbd_unlock_bh_state(bh);
1477 __bforget(bh);
1478 goto drop;
1479 }
1480 }
1481 } else if (jh->b_transaction) {
1482 J_ASSERT_JH(jh, (jh->b_transaction ==
1483 journal->j_committing_transaction));
1484 /* However, if the buffer is still owned by a prior
1485 * (committing) transaction, we can't drop it yet... */
1486 JBUFFER_TRACE(jh, "belongs to older transaction");
1487 /* ... but we CAN drop it from the new transaction if we
1488 * have also modified it since the original commit. */
1489
1490 if (jh->b_next_transaction) {
1491 J_ASSERT(jh->b_next_transaction == transaction);
1492 jh->b_next_transaction = NULL;
Josef Bacik1dfc3222008-04-17 10:38:59 -04001493
1494 /*
1495 * only drop a reference if this transaction modified
1496 * the buffer
1497 */
1498 if (was_modified)
1499 drop_reserve = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001500 }
1501 }
1502
1503not_jbd:
1504 spin_unlock(&journal->j_list_lock);
1505 jbd_unlock_bh_state(bh);
1506 __brelse(bh);
1507drop:
1508 if (drop_reserve) {
1509 /* no need to reserve log space for this block -bzzz */
1510 handle->h_buffer_credits++;
1511 }
1512 return err;
1513}
1514
1515/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001516 * int jbd2_journal_stop() - complete a transaction
Dave Kleikamp470decc2006-10-11 01:20:57 -07001517 * @handle: tranaction to complete.
1518 *
1519 * All done for a particular handle.
1520 *
1521 * There is not much action needed here. We just return any remaining
1522 * buffer credits to the transaction and remove the handle. The only
1523 * complication is that we need to start a commit operation if the
1524 * filesystem is marked for synchronous update.
1525 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001526 * jbd2_journal_stop itself will not usually return an error, but it may
Dave Kleikamp470decc2006-10-11 01:20:57 -07001527 * do so in unusual circumstances. In particular, expect it to
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001528 * return -EIO if a jbd2_journal_abort has been executed since the
Dave Kleikamp470decc2006-10-11 01:20:57 -07001529 * transaction began.
1530 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001531int jbd2_journal_stop(handle_t *handle)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001532{
1533 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001534 journal_t *journal;
1535 int err = 0, wait_for_commit = 0;
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001536 tid_t tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001537 pid_t pid;
1538
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001539 if (!transaction)
1540 goto free_and_exit;
1541 journal = transaction->t_journal;
1542
Dave Kleikamp470decc2006-10-11 01:20:57 -07001543 J_ASSERT(journal_current_handle() == handle);
1544
1545 if (is_handle_aborted(handle))
1546 err = -EIO;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001547 else
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001548 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001549
1550 if (--handle->h_ref > 0) {
1551 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1552 handle->h_ref);
1553 return err;
1554 }
1555
1556 jbd_debug(4, "Handle %p going down\n", handle);
Theodore Ts'o343d9c22013-02-08 13:00:22 -05001557 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001558 transaction->t_tid,
Theodore Ts'o343d9c22013-02-08 13:00:22 -05001559 handle->h_type, handle->h_line_no,
1560 jiffies - handle->h_start_jiffies,
1561 handle->h_sync, handle->h_requested_credits,
1562 (handle->h_requested_credits -
1563 handle->h_buffer_credits));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001564
1565 /*
1566 * Implement synchronous transaction batching. If the handle
1567 * was synchronous, don't force a commit immediately. Let's
Josef Bacike07f7182008-11-26 01:14:26 -05001568 * yield and let another thread piggyback onto this
1569 * transaction. Keep doing that while new threads continue to
1570 * arrive. It doesn't cost much - we're about to run a commit
1571 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1572 * operations by 30x or more...
Dave Kleikamp470decc2006-10-11 01:20:57 -07001573 *
Josef Bacike07f7182008-11-26 01:14:26 -05001574 * We try and optimize the sleep time against what the
1575 * underlying disk can do, instead of having a static sleep
1576 * time. This is useful for the case where our storage is so
1577 * fast that it is more optimal to go ahead and force a flush
1578 * and wait for the transaction to be committed than it is to
1579 * wait for an arbitrary amount of time for new writers to
1580 * join the transaction. We achieve this by measuring how
1581 * long it takes to commit a transaction, and compare it with
1582 * how long this transaction has been running, and if run time
1583 * < commit time then we sleep for the delta and commit. This
1584 * greatly helps super fast disks that would see slowdowns as
1585 * more threads started doing fsyncs.
1586 *
1587 * But don't do this if this process was the most recent one
1588 * to perform a synchronous write. We do this to detect the
1589 * case where a single process is doing a stream of sync
1590 * writes. No point in waiting for joiners in that case.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001591 */
1592 pid = current->pid;
1593 if (handle->h_sync && journal->j_last_sync_writer != pid) {
Josef Bacike07f7182008-11-26 01:14:26 -05001594 u64 commit_time, trans_time;
1595
Dave Kleikamp470decc2006-10-11 01:20:57 -07001596 journal->j_last_sync_writer = pid;
Josef Bacike07f7182008-11-26 01:14:26 -05001597
Theodore Ts'oa931da62010-08-03 21:35:12 -04001598 read_lock(&journal->j_state_lock);
Josef Bacike07f7182008-11-26 01:14:26 -05001599 commit_time = journal->j_average_commit_time;
Theodore Ts'oa931da62010-08-03 21:35:12 -04001600 read_unlock(&journal->j_state_lock);
Josef Bacike07f7182008-11-26 01:14:26 -05001601
1602 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1603 transaction->t_start_time));
1604
Theodore Ts'o30773842009-01-03 20:27:38 -05001605 commit_time = max_t(u64, commit_time,
1606 1000*journal->j_min_batch_time);
Josef Bacike07f7182008-11-26 01:14:26 -05001607 commit_time = min_t(u64, commit_time,
Theodore Ts'o30773842009-01-03 20:27:38 -05001608 1000*journal->j_max_batch_time);
Josef Bacike07f7182008-11-26 01:14:26 -05001609
1610 if (trans_time < commit_time) {
1611 ktime_t expires = ktime_add_ns(ktime_get(),
1612 commit_time);
1613 set_current_state(TASK_UNINTERRUPTIBLE);
1614 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1615 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001616 }
1617
Theodore Ts'o70585482009-03-25 23:35:46 -04001618 if (handle->h_sync)
1619 transaction->t_synchronous_commit = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001620 current->journal_info = NULL;
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001621 atomic_sub(handle->h_buffer_credits,
1622 &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001623
1624 /*
1625 * If the handle is marked SYNC, we need to set another commit
1626 * going! We also want to force a commit if the current
1627 * transaction is occupying too much of the log, or if the
1628 * transaction is too old now.
1629 */
1630 if (handle->h_sync ||
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001631 (atomic_read(&transaction->t_outstanding_credits) >
1632 journal->j_max_transaction_buffers) ||
1633 time_after_eq(jiffies, transaction->t_expires)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001634 /* Do this even for aborted journals: an abort still
1635 * completes the commit thread, it just doesn't write
1636 * anything to disk. */
Dave Kleikamp470decc2006-10-11 01:20:57 -07001637
Dave Kleikamp470decc2006-10-11 01:20:57 -07001638 jbd_debug(2, "transaction too old, requesting commit for "
1639 "handle %p\n", handle);
1640 /* This is non-blocking */
Theodore Ts'oc35a56a2010-05-16 05:00:00 -04001641 jbd2_log_start_commit(journal, transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001642
1643 /*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001644 * Special case: JBD2_SYNC synchronous updates require us
Dave Kleikamp470decc2006-10-11 01:20:57 -07001645 * to wait for the commit to complete.
1646 */
1647 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001648 wait_for_commit = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001649 }
1650
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001651 /*
1652 * Once we drop t_updates, if it goes to zero the transaction
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001653 * could start committing on us and eventually disappear. So
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001654 * once we do this, we must not dereference transaction
1655 * pointer again.
1656 */
1657 tid = transaction->t_tid;
1658 if (atomic_dec_and_test(&transaction->t_updates)) {
1659 wake_up(&journal->j_wait_updates);
1660 if (journal->j_barrier_count)
1661 wake_up(&journal->j_wait_transaction_locked);
1662 }
1663
1664 if (wait_for_commit)
1665 err = jbd2_log_wait_commit(journal, tid);
1666
Ingo Molnar3295f0e2008-08-11 10:30:30 +02001667 lock_map_release(&handle->h_lockdep_map);
Mingming Cao7b751062008-01-28 23:58:27 -05001668
Jan Kara8f7d89f2013-06-04 12:35:11 -04001669 if (handle->h_rsv_handle)
1670 jbd2_journal_free_reserved(handle->h_rsv_handle);
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001671free_and_exit:
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001672 jbd2_free_handle(handle);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001673 return err;
1674}
1675
Dave Kleikamp470decc2006-10-11 01:20:57 -07001676/*
1677 *
1678 * List management code snippets: various functions for manipulating the
1679 * transaction buffer lists.
1680 *
1681 */
1682
1683/*
1684 * Append a buffer to a transaction list, given the transaction's list head
1685 * pointer.
1686 *
1687 * j_list_lock is held.
1688 *
1689 * jbd_lock_bh_state(jh2bh(jh)) is held.
1690 */
1691
1692static inline void
1693__blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1694{
1695 if (!*list) {
1696 jh->b_tnext = jh->b_tprev = jh;
1697 *list = jh;
1698 } else {
1699 /* Insert at the tail of the list to preserve order */
1700 struct journal_head *first = *list, *last = first->b_tprev;
1701 jh->b_tprev = last;
1702 jh->b_tnext = first;
1703 last->b_tnext = first->b_tprev = jh;
1704 }
1705}
1706
1707/*
1708 * Remove a buffer from a transaction list, given the transaction's list
1709 * head pointer.
1710 *
1711 * Called with j_list_lock held, and the journal may not be locked.
1712 *
1713 * jbd_lock_bh_state(jh2bh(jh)) is held.
1714 */
1715
1716static inline void
1717__blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1718{
1719 if (*list == jh) {
1720 *list = jh->b_tnext;
1721 if (*list == jh)
1722 *list = NULL;
1723 }
1724 jh->b_tprev->b_tnext = jh->b_tnext;
1725 jh->b_tnext->b_tprev = jh->b_tprev;
1726}
1727
1728/*
1729 * Remove a buffer from the appropriate transaction list.
1730 *
1731 * Note that this function can *change* the value of
Jan Karaf5113ef2013-06-04 12:01:45 -04001732 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1733 * t_reserved_list. If the caller is holding onto a copy of one of these
1734 * pointers, it could go bad. Generally the caller needs to re-read the
1735 * pointer from the transaction_t.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001736 *
Jan Kara5bebccf2012-03-13 22:25:06 -04001737 * Called under j_list_lock.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001738 */
Jan Kara5bebccf2012-03-13 22:25:06 -04001739static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001740{
1741 struct journal_head **list = NULL;
1742 transaction_t *transaction;
1743 struct buffer_head *bh = jh2bh(jh);
1744
1745 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1746 transaction = jh->b_transaction;
1747 if (transaction)
1748 assert_spin_locked(&transaction->t_journal->j_list_lock);
1749
1750 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1751 if (jh->b_jlist != BJ_None)
Mingming Cao40191912008-01-28 23:58:27 -05001752 J_ASSERT_JH(jh, transaction != NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001753
1754 switch (jh->b_jlist) {
1755 case BJ_None:
1756 return;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001757 case BJ_Metadata:
1758 transaction->t_nr_buffers--;
1759 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1760 list = &transaction->t_buffers;
1761 break;
1762 case BJ_Forget:
1763 list = &transaction->t_forget;
1764 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001765 case BJ_Shadow:
1766 list = &transaction->t_shadow_list;
1767 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001768 case BJ_Reserved:
1769 list = &transaction->t_reserved_list;
1770 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001771 }
1772
1773 __blist_del_buffer(list, jh);
1774 jh->b_jlist = BJ_None;
1775 if (test_clear_buffer_jbddirty(bh))
1776 mark_buffer_dirty(bh); /* Expose it to the VM */
1777}
1778
Jan Karade1b7942011-06-13 15:38:22 -04001779/*
1780 * Remove buffer from all transactions.
1781 *
1782 * Called with bh_state lock and j_list_lock
1783 *
1784 * jh and bh may be already freed when this function returns.
1785 */
1786static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001787{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001788 __jbd2_journal_temp_unlink_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001789 jh->b_transaction = NULL;
Jan Karade1b7942011-06-13 15:38:22 -04001790 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001791}
1792
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001793void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001794{
Jan Karade1b7942011-06-13 15:38:22 -04001795 struct buffer_head *bh = jh2bh(jh);
1796
1797 /* Get reference so that buffer cannot be freed before we unlock it */
1798 get_bh(bh);
1799 jbd_lock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001800 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001801 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001802 spin_unlock(&journal->j_list_lock);
Jan Karade1b7942011-06-13 15:38:22 -04001803 jbd_unlock_bh_state(bh);
1804 __brelse(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001805}
1806
1807/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001808 * Called from jbd2_journal_try_to_free_buffers().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001809 *
1810 * Called under jbd_lock_bh_state(bh)
1811 */
1812static void
1813__journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1814{
1815 struct journal_head *jh;
1816
1817 jh = bh2jh(bh);
1818
1819 if (buffer_locked(bh) || buffer_dirty(bh))
1820 goto out;
1821
Mingming Cao40191912008-01-28 23:58:27 -05001822 if (jh->b_next_transaction != NULL)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001823 goto out;
1824
1825 spin_lock(&journal->j_list_lock);
Jan Kara87c89c22008-07-11 19:27:31 -04001826 if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001827 /* written-back checkpointed metadata buffer */
Jan Karac254c9e2012-03-13 22:27:44 -04001828 JBUFFER_TRACE(jh, "remove from checkpoint list");
1829 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001830 }
1831 spin_unlock(&journal->j_list_lock);
1832out:
1833 return;
1834}
1835
Dave Kleikamp470decc2006-10-11 01:20:57 -07001836/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001837 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001838 * @journal: journal for operation
1839 * @page: to try and free
Mingming Cao530576b2008-07-13 21:06:39 -04001840 * @gfp_mask: we use the mask to detect how hard should we try to release
1841 * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to
1842 * release the buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001843 *
1844 *
1845 * For all the buffers on this page,
1846 * if they are fully written out ordered data, move them onto BUF_CLEAN
1847 * so try_to_free_buffers() can reap them.
1848 *
1849 * This function returns non-zero if we wish try_to_free_buffers()
1850 * to be called. We do this if the page is releasable by try_to_free_buffers().
1851 * We also do it if the page has locked or dirty buffers and the caller wants
1852 * us to perform sync or async writeout.
1853 *
1854 * This complicates JBD locking somewhat. We aren't protected by the
1855 * BKL here. We wish to remove the buffer from its committing or
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001856 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001857 *
1858 * This may *change* the value of transaction_t->t_datalist, so anyone
1859 * who looks at t_datalist needs to lock against this function.
1860 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001861 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1862 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
Dave Kleikamp470decc2006-10-11 01:20:57 -07001863 * will come out of the lock with the buffer dirty, which makes it
1864 * ineligible for release here.
1865 *
1866 * Who else is affected by this? hmm... Really the only contender
1867 * is do_get_write_access() - it could be looking at the buffer while
1868 * journal_try_to_free_buffer() is changing its state. But that
1869 * cannot happen because we never reallocate freed data as metadata
1870 * while the data is part of a transaction. Yes?
Mingming Cao530576b2008-07-13 21:06:39 -04001871 *
1872 * Return 0 on failure, 1 on success
Dave Kleikamp470decc2006-10-11 01:20:57 -07001873 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001874int jbd2_journal_try_to_free_buffers(journal_t *journal,
Mingming Cao530576b2008-07-13 21:06:39 -04001875 struct page *page, gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001876{
1877 struct buffer_head *head;
1878 struct buffer_head *bh;
1879 int ret = 0;
1880
1881 J_ASSERT(PageLocked(page));
1882
1883 head = page_buffers(page);
1884 bh = head;
1885 do {
1886 struct journal_head *jh;
1887
1888 /*
1889 * We take our own ref against the journal_head here to avoid
1890 * having to add tons of locking around each instance of
Mingming Cao530576b2008-07-13 21:06:39 -04001891 * jbd2_journal_put_journal_head().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001892 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001893 jh = jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001894 if (!jh)
1895 continue;
1896
1897 jbd_lock_bh_state(bh);
1898 __journal_try_to_free_buffer(journal, bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001899 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001900 jbd_unlock_bh_state(bh);
1901 if (buffer_jbd(bh))
1902 goto busy;
1903 } while ((bh = bh->b_this_page) != head);
Mingming Cao530576b2008-07-13 21:06:39 -04001904
Dave Kleikamp470decc2006-10-11 01:20:57 -07001905 ret = try_to_free_buffers(page);
Mingming Cao530576b2008-07-13 21:06:39 -04001906
Dave Kleikamp470decc2006-10-11 01:20:57 -07001907busy:
1908 return ret;
1909}
1910
1911/*
1912 * This buffer is no longer needed. If it is on an older transaction's
1913 * checkpoint list we need to record it on this transaction's forget list
1914 * to pin this buffer (and hence its checkpointing transaction) down until
1915 * this transaction commits. If the buffer isn't on a checkpoint list, we
1916 * release it.
1917 * Returns non-zero if JBD no longer has an interest in the buffer.
1918 *
1919 * Called under j_list_lock.
1920 *
1921 * Called under jbd_lock_bh_state(bh).
1922 */
1923static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
1924{
1925 int may_free = 1;
1926 struct buffer_head *bh = jh2bh(jh);
1927
Dave Kleikamp470decc2006-10-11 01:20:57 -07001928 if (jh->b_cp_transaction) {
1929 JBUFFER_TRACE(jh, "on running+cp transaction");
Jan Karade1b7942011-06-13 15:38:22 -04001930 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karaf91d1d02009-07-13 16:16:20 -04001931 /*
1932 * We don't want to write the buffer anymore, clear the
1933 * bit so that we don't confuse checks in
1934 * __journal_file_buffer
1935 */
1936 clear_buffer_dirty(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001937 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001938 may_free = 0;
1939 } else {
1940 JBUFFER_TRACE(jh, "on running transaction");
Jan Karade1b7942011-06-13 15:38:22 -04001941 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001942 }
1943 return may_free;
1944}
1945
1946/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001947 * jbd2_journal_invalidatepage
Dave Kleikamp470decc2006-10-11 01:20:57 -07001948 *
1949 * This code is tricky. It has a number of cases to deal with.
1950 *
1951 * There are two invariants which this code relies on:
1952 *
1953 * i_size must be updated on disk before we start calling invalidatepage on the
1954 * data.
1955 *
1956 * This is done in ext3 by defining an ext3_setattr method which
1957 * updates i_size before truncate gets going. By maintaining this
1958 * invariant, we can be sure that it is safe to throw away any buffers
1959 * attached to the current transaction: once the transaction commits,
1960 * we know that the data will not be needed.
1961 *
1962 * Note however that we can *not* throw away data belonging to the
1963 * previous, committing transaction!
1964 *
1965 * Any disk blocks which *are* part of the previous, committing
1966 * transaction (and which therefore cannot be discarded immediately) are
1967 * not going to be reused in the new running transaction
1968 *
1969 * The bitmap committed_data images guarantee this: any block which is
1970 * allocated in one transaction and removed in the next will be marked
1971 * as in-use in the committed_data bitmap, so cannot be reused until
1972 * the next transaction to delete the block commits. This means that
1973 * leaving committing buffers dirty is quite safe: the disk blocks
1974 * cannot be reallocated to a different file and so buffer aliasing is
1975 * not possible.
1976 *
1977 *
1978 * The above applies mainly to ordered data mode. In writeback mode we
1979 * don't make guarantees about the order in which data hits disk --- in
1980 * particular we don't guarantee that new dirty data is flushed before
1981 * transaction commit --- so it is always safe just to discard data
1982 * immediately in that mode. --sct
1983 */
1984
1985/*
1986 * The journal_unmap_buffer helper function returns zero if the buffer
1987 * concerned remains pinned as an anonymous buffer belonging to an older
1988 * transaction.
1989 *
1990 * We're outside-transaction here. Either or both of j_running_transaction
1991 * and j_committing_transaction may be NULL.
1992 */
Jan Karab794e7a2012-09-26 23:11:13 -04001993static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
1994 int partial_page)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001995{
1996 transaction_t *transaction;
1997 struct journal_head *jh;
1998 int may_free = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001999
2000 BUFFER_TRACE(bh, "entry");
2001
2002 /*
2003 * It is safe to proceed here without the j_list_lock because the
2004 * buffers cannot be stolen by try_to_free_buffers as long as we are
2005 * holding the page lock. --sct
2006 */
2007
2008 if (!buffer_jbd(bh))
2009 goto zap_buffer_unlocked;
2010
Jan Kara87c89c22008-07-11 19:27:31 -04002011 /* OK, we have data buffer in journaled mode */
Theodore Ts'oa931da62010-08-03 21:35:12 -04002012 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002013 jbd_lock_bh_state(bh);
2014 spin_lock(&journal->j_list_lock);
2015
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002016 jh = jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002017 if (!jh)
2018 goto zap_buffer_no_jh;
2019
dingdinghuaba869022010-02-15 16:35:42 -05002020 /*
2021 * We cannot remove the buffer from checkpoint lists until the
2022 * transaction adding inode to orphan list (let's call it T)
2023 * is committed. Otherwise if the transaction changing the
2024 * buffer would be cleaned from the journal before T is
2025 * committed, a crash will cause that the correct contents of
2026 * the buffer will be lost. On the other hand we have to
2027 * clear the buffer dirty bit at latest at the moment when the
2028 * transaction marking the buffer as freed in the filesystem
2029 * structures is committed because from that moment on the
Jan Karab794e7a2012-09-26 23:11:13 -04002030 * block can be reallocated and used by a different page.
dingdinghuaba869022010-02-15 16:35:42 -05002031 * Since the block hasn't been freed yet but the inode has
2032 * already been added to orphan list, it is safe for us to add
2033 * the buffer to BJ_Forget list of the newest transaction.
Jan Karab794e7a2012-09-26 23:11:13 -04002034 *
2035 * Also we have to clear buffer_mapped flag of a truncated buffer
2036 * because the buffer_head may be attached to the page straddling
2037 * i_size (can happen only when blocksize < pagesize) and thus the
2038 * buffer_head can be reused when the file is extended again. So we end
2039 * up keeping around invalidated buffers attached to transactions'
2040 * BJ_Forget list just to stop checkpointing code from cleaning up
2041 * the transaction this buffer was modified in.
dingdinghuaba869022010-02-15 16:35:42 -05002042 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002043 transaction = jh->b_transaction;
2044 if (transaction == NULL) {
2045 /* First case: not on any transaction. If it
2046 * has no checkpoint link, then we can zap it:
2047 * it's a writeback-mode buffer so we don't care
2048 * if it hits disk safely. */
2049 if (!jh->b_cp_transaction) {
2050 JBUFFER_TRACE(jh, "not on any transaction: zap");
2051 goto zap_buffer;
2052 }
2053
2054 if (!buffer_dirty(bh)) {
2055 /* bdflush has written it. We can drop it now */
2056 goto zap_buffer;
2057 }
2058
2059 /* OK, it must be in the journal but still not
2060 * written fully to disk: it's metadata or
2061 * journaled data... */
2062
2063 if (journal->j_running_transaction) {
2064 /* ... and once the current transaction has
2065 * committed, the buffer won't be needed any
2066 * longer. */
2067 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
Jan Karab794e7a2012-09-26 23:11:13 -04002068 may_free = __dispose_buffer(jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002069 journal->j_running_transaction);
Jan Karab794e7a2012-09-26 23:11:13 -04002070 goto zap_buffer;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002071 } else {
2072 /* There is no currently-running transaction. So the
2073 * orphan record which we wrote for this file must have
2074 * passed into commit. We must attach this buffer to
2075 * the committing transaction, if it exists. */
2076 if (journal->j_committing_transaction) {
2077 JBUFFER_TRACE(jh, "give to committing trans");
Jan Karab794e7a2012-09-26 23:11:13 -04002078 may_free = __dispose_buffer(jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002079 journal->j_committing_transaction);
Jan Karab794e7a2012-09-26 23:11:13 -04002080 goto zap_buffer;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002081 } else {
2082 /* The orphan record's transaction has
2083 * committed. We can cleanse this buffer */
2084 clear_buffer_jbddirty(bh);
2085 goto zap_buffer;
2086 }
2087 }
2088 } else if (transaction == journal->j_committing_transaction) {
Eric Sandeen9b579882006-10-28 10:38:28 -07002089 JBUFFER_TRACE(jh, "on committing transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002090 /*
dingdinghuaba869022010-02-15 16:35:42 -05002091 * The buffer is committing, we simply cannot touch
Jan Karab794e7a2012-09-26 23:11:13 -04002092 * it. If the page is straddling i_size we have to wait
2093 * for commit and try again.
2094 */
2095 if (partial_page) {
Jan Karab794e7a2012-09-26 23:11:13 -04002096 jbd2_journal_put_journal_head(jh);
2097 spin_unlock(&journal->j_list_lock);
2098 jbd_unlock_bh_state(bh);
2099 write_unlock(&journal->j_state_lock);
Jan Kara53e87262012-12-25 13:29:52 -05002100 return -EBUSY;
Jan Karab794e7a2012-09-26 23:11:13 -04002101 }
2102 /*
2103 * OK, buffer won't be reachable after truncate. We just set
2104 * j_next_transaction to the running transaction (if there is
2105 * one) and mark buffer as freed so that commit code knows it
2106 * should clear dirty bits when it is done with the buffer.
dingdinghuaba869022010-02-15 16:35:42 -05002107 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002108 set_buffer_freed(bh);
dingdinghuaba869022010-02-15 16:35:42 -05002109 if (journal->j_running_transaction && buffer_jbddirty(bh))
2110 jh->b_next_transaction = journal->j_running_transaction;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002111 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002112 spin_unlock(&journal->j_list_lock);
2113 jbd_unlock_bh_state(bh);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002114 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002115 return 0;
2116 } else {
2117 /* Good, the buffer belongs to the running transaction.
2118 * We are writing our own transaction's data, not any
2119 * previous one's, so it is safe to throw it away
2120 * (remember that we expect the filesystem to have set
2121 * i_size already for this truncate so recovery will not
2122 * expose the disk blocks we are discarding here.) */
2123 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
Eric Sandeen9b579882006-10-28 10:38:28 -07002124 JBUFFER_TRACE(jh, "on running transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002125 may_free = __dispose_buffer(jh, transaction);
2126 }
2127
2128zap_buffer:
Jan Karab794e7a2012-09-26 23:11:13 -04002129 /*
2130 * This is tricky. Although the buffer is truncated, it may be reused
2131 * if blocksize < pagesize and it is attached to the page straddling
2132 * EOF. Since the buffer might have been added to BJ_Forget list of the
2133 * running transaction, journal_get_write_access() won't clear
2134 * b_modified and credit accounting gets confused. So clear b_modified
2135 * here.
2136 */
2137 jh->b_modified = 0;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002138 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002139zap_buffer_no_jh:
2140 spin_unlock(&journal->j_list_lock);
2141 jbd_unlock_bh_state(bh);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002142 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002143zap_buffer_unlocked:
2144 clear_buffer_dirty(bh);
2145 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2146 clear_buffer_mapped(bh);
2147 clear_buffer_req(bh);
2148 clear_buffer_new(bh);
Eric Sandeen15291162012-02-20 17:53:01 -05002149 clear_buffer_delay(bh);
2150 clear_buffer_unwritten(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002151 bh->b_bdev = NULL;
2152 return may_free;
2153}
2154
2155/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002156 * void jbd2_journal_invalidatepage()
Dave Kleikamp470decc2006-10-11 01:20:57 -07002157 * @journal: journal to use for flush...
2158 * @page: page to flush
Lukas Czerner259709b2013-05-21 23:20:03 -04002159 * @offset: start of the range to invalidate
2160 * @length: length of the range to invalidate
Dave Kleikamp470decc2006-10-11 01:20:57 -07002161 *
Lukas Czerner259709b2013-05-21 23:20:03 -04002162 * Reap page buffers containing data after in the specified range in page.
2163 * Can return -EBUSY if buffers are part of the committing transaction and
2164 * the page is straddling i_size. Caller then has to wait for current commit
2165 * and try again.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002166 */
Jan Kara53e87262012-12-25 13:29:52 -05002167int jbd2_journal_invalidatepage(journal_t *journal,
2168 struct page *page,
Lukas Czerner259709b2013-05-21 23:20:03 -04002169 unsigned int offset,
2170 unsigned int length)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002171{
2172 struct buffer_head *head, *bh, *next;
Lukas Czerner259709b2013-05-21 23:20:03 -04002173 unsigned int stop = offset + length;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002174 unsigned int curr_off = 0;
Lukas Czerner259709b2013-05-21 23:20:03 -04002175 int partial_page = (offset || length < PAGE_CACHE_SIZE);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002176 int may_free = 1;
Jan Kara53e87262012-12-25 13:29:52 -05002177 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002178
2179 if (!PageLocked(page))
2180 BUG();
2181 if (!page_has_buffers(page))
Jan Kara53e87262012-12-25 13:29:52 -05002182 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002183
Lukas Czerner259709b2013-05-21 23:20:03 -04002184 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
2185
Dave Kleikamp470decc2006-10-11 01:20:57 -07002186 /* We will potentially be playing with lists other than just the
2187 * data lists (especially for journaled data mode), so be
2188 * cautious in our locking. */
2189
2190 head = bh = page_buffers(page);
2191 do {
2192 unsigned int next_off = curr_off + bh->b_size;
2193 next = bh->b_this_page;
2194
Lukas Czerner259709b2013-05-21 23:20:03 -04002195 if (next_off > stop)
2196 return 0;
2197
Dave Kleikamp470decc2006-10-11 01:20:57 -07002198 if (offset <= curr_off) {
2199 /* This block is wholly outside the truncation point */
2200 lock_buffer(bh);
Lukas Czerner259709b2013-05-21 23:20:03 -04002201 ret = journal_unmap_buffer(journal, bh, partial_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002202 unlock_buffer(bh);
Jan Kara53e87262012-12-25 13:29:52 -05002203 if (ret < 0)
2204 return ret;
2205 may_free &= ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002206 }
2207 curr_off = next_off;
2208 bh = next;
2209
2210 } while (bh != head);
2211
Lukas Czerner259709b2013-05-21 23:20:03 -04002212 if (!partial_page) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002213 if (may_free && try_to_free_buffers(page))
2214 J_ASSERT(!page_has_buffers(page));
2215 }
Jan Kara53e87262012-12-25 13:29:52 -05002216 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002217}
2218
2219/*
2220 * File a buffer on the given transaction list.
2221 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002222void __jbd2_journal_file_buffer(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002223 transaction_t *transaction, int jlist)
2224{
2225 struct journal_head **list = NULL;
2226 int was_dirty = 0;
2227 struct buffer_head *bh = jh2bh(jh);
2228
2229 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2230 assert_spin_locked(&transaction->t_journal->j_list_lock);
2231
2232 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2233 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
Mingming Cao40191912008-01-28 23:58:27 -05002234 jh->b_transaction == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002235
2236 if (jh->b_transaction && jh->b_jlist == jlist)
2237 return;
2238
Dave Kleikamp470decc2006-10-11 01:20:57 -07002239 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2240 jlist == BJ_Shadow || jlist == BJ_Forget) {
Jan Karaf91d1d02009-07-13 16:16:20 -04002241 /*
2242 * For metadata buffers, we track dirty bit in buffer_jbddirty
2243 * instead of buffer_dirty. We should not see a dirty bit set
2244 * here because we clear it in do_get_write_access but e.g.
2245 * tune2fs can modify the sb and set the dirty bit at any time
2246 * so we try to gracefully handle that.
2247 */
2248 if (buffer_dirty(bh))
2249 warn_dirty_buffer(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002250 if (test_clear_buffer_dirty(bh) ||
2251 test_clear_buffer_jbddirty(bh))
2252 was_dirty = 1;
2253 }
2254
2255 if (jh->b_transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002256 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karade1b7942011-06-13 15:38:22 -04002257 else
2258 jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002259 jh->b_transaction = transaction;
2260
2261 switch (jlist) {
2262 case BJ_None:
2263 J_ASSERT_JH(jh, !jh->b_committed_data);
2264 J_ASSERT_JH(jh, !jh->b_frozen_data);
2265 return;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002266 case BJ_Metadata:
2267 transaction->t_nr_buffers++;
2268 list = &transaction->t_buffers;
2269 break;
2270 case BJ_Forget:
2271 list = &transaction->t_forget;
2272 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002273 case BJ_Shadow:
2274 list = &transaction->t_shadow_list;
2275 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002276 case BJ_Reserved:
2277 list = &transaction->t_reserved_list;
2278 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002279 }
2280
2281 __blist_add_buffer(list, jh);
2282 jh->b_jlist = jlist;
2283
2284 if (was_dirty)
2285 set_buffer_jbddirty(bh);
2286}
2287
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002288void jbd2_journal_file_buffer(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002289 transaction_t *transaction, int jlist)
2290{
2291 jbd_lock_bh_state(jh2bh(jh));
2292 spin_lock(&transaction->t_journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002293 __jbd2_journal_file_buffer(jh, transaction, jlist);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002294 spin_unlock(&transaction->t_journal->j_list_lock);
2295 jbd_unlock_bh_state(jh2bh(jh));
2296}
2297
2298/*
2299 * Remove a buffer from its current buffer list in preparation for
2300 * dropping it from its current transaction entirely. If the buffer has
2301 * already started to be used by a subsequent transaction, refile the
2302 * buffer on that transaction's metadata list.
2303 *
Jan Karade1b7942011-06-13 15:38:22 -04002304 * Called under j_list_lock
Dave Kleikamp470decc2006-10-11 01:20:57 -07002305 * Called under jbd_lock_bh_state(jh2bh(jh))
Jan Karade1b7942011-06-13 15:38:22 -04002306 *
2307 * jh and bh may be already free when this function returns
Dave Kleikamp470decc2006-10-11 01:20:57 -07002308 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002309void __jbd2_journal_refile_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002310{
dingdinghuaba869022010-02-15 16:35:42 -05002311 int was_dirty, jlist;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002312 struct buffer_head *bh = jh2bh(jh);
2313
2314 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2315 if (jh->b_transaction)
2316 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2317
2318 /* If the buffer is now unused, just drop it. */
2319 if (jh->b_next_transaction == NULL) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002320 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002321 return;
2322 }
2323
2324 /*
2325 * It has been modified by a later transaction: add it to the new
2326 * transaction's metadata list.
2327 */
2328
2329 was_dirty = test_clear_buffer_jbddirty(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002330 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karade1b7942011-06-13 15:38:22 -04002331 /*
2332 * We set b_transaction here because b_next_transaction will inherit
2333 * our jh reference and thus __jbd2_journal_file_buffer() must not
2334 * take a new one.
2335 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002336 jh->b_transaction = jh->b_next_transaction;
2337 jh->b_next_transaction = NULL;
dingdinghuaba869022010-02-15 16:35:42 -05002338 if (buffer_freed(bh))
2339 jlist = BJ_Forget;
2340 else if (jh->b_modified)
2341 jlist = BJ_Metadata;
2342 else
2343 jlist = BJ_Reserved;
2344 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002345 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2346
2347 if (was_dirty)
2348 set_buffer_jbddirty(bh);
2349}
2350
2351/*
Jan Karade1b7942011-06-13 15:38:22 -04002352 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2353 * bh reference so that we can safely unlock bh.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002354 *
Jan Karade1b7942011-06-13 15:38:22 -04002355 * The jh and bh may be freed by this call.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002356 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002357void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002358{
2359 struct buffer_head *bh = jh2bh(jh);
2360
Jan Karade1b7942011-06-13 15:38:22 -04002361 /* Get reference so that buffer cannot be freed before we unlock it */
2362 get_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002363 jbd_lock_bh_state(bh);
2364 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002365 __jbd2_journal_refile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002366 jbd_unlock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002367 spin_unlock(&journal->j_list_lock);
2368 __brelse(bh);
2369}
Jan Karac851ed52008-07-11 19:27:31 -04002370
2371/*
2372 * File inode in the inode list of the handle's transaction
2373 */
2374int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode)
2375{
2376 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002377 journal_t *journal;
Jan Karac851ed52008-07-11 19:27:31 -04002378
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002379 WARN_ON(!transaction);
Jan Karac851ed52008-07-11 19:27:31 -04002380 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002381 return -EROFS;
2382 journal = transaction->t_journal;
Jan Karac851ed52008-07-11 19:27:31 -04002383
2384 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2385 transaction->t_tid);
2386
2387 /*
2388 * First check whether inode isn't already on the transaction's
2389 * lists without taking the lock. Note that this check is safe
2390 * without the lock as we cannot race with somebody removing inode
2391 * from the transaction. The reason is that we remove inode from the
2392 * transaction only in journal_release_jbd_inode() and when we commit
2393 * the transaction. We are guarded from the first case by holding
2394 * a reference to the inode. We are safe against the second case
2395 * because if jinode->i_transaction == transaction, commit code
2396 * cannot touch the transaction because we hold reference to it,
2397 * and if jinode->i_next_transaction == transaction, commit code
2398 * will only file the inode where we want it.
2399 */
2400 if (jinode->i_transaction == transaction ||
2401 jinode->i_next_transaction == transaction)
2402 return 0;
2403
2404 spin_lock(&journal->j_list_lock);
2405
2406 if (jinode->i_transaction == transaction ||
2407 jinode->i_next_transaction == transaction)
2408 goto done;
2409
Jan Kara81be12c2011-05-24 11:52:40 -04002410 /*
2411 * We only ever set this variable to 1 so the test is safe. Since
2412 * t_need_data_flush is likely to be set, we do the test to save some
2413 * cacheline bouncing
2414 */
2415 if (!transaction->t_need_data_flush)
2416 transaction->t_need_data_flush = 1;
Jan Karac851ed52008-07-11 19:27:31 -04002417 /* On some different transaction's list - should be
2418 * the committing one */
2419 if (jinode->i_transaction) {
2420 J_ASSERT(jinode->i_next_transaction == NULL);
2421 J_ASSERT(jinode->i_transaction ==
2422 journal->j_committing_transaction);
2423 jinode->i_next_transaction = transaction;
2424 goto done;
2425 }
2426 /* Not on any transaction list... */
2427 J_ASSERT(!jinode->i_next_transaction);
2428 jinode->i_transaction = transaction;
2429 list_add(&jinode->i_list, &transaction->t_inode_list);
2430done:
2431 spin_unlock(&journal->j_list_lock);
2432
2433 return 0;
2434}
2435
2436/*
Jan Kara7f5aa212009-02-10 11:15:34 -05002437 * File truncate and transaction commit interact with each other in a
2438 * non-trivial way. If a transaction writing data block A is
2439 * committing, we cannot discard the data by truncate until we have
2440 * written them. Otherwise if we crashed after the transaction with
2441 * write has committed but before the transaction with truncate has
2442 * committed, we could see stale data in block A. This function is a
2443 * helper to solve this problem. It starts writeout of the truncated
2444 * part in case it is in the committing transaction.
2445 *
2446 * Filesystem code must call this function when inode is journaled in
2447 * ordered mode before truncation happens and after the inode has been
2448 * placed on orphan list with the new inode size. The second condition
2449 * avoids the race that someone writes new data and we start
2450 * committing the transaction after this function has been called but
2451 * before a transaction for truncate is started (and furthermore it
2452 * allows us to optimize the case where the addition to orphan list
2453 * happens in the same transaction as write --- we don't have to write
2454 * any data in such case).
Jan Karac851ed52008-07-11 19:27:31 -04002455 */
Jan Kara7f5aa212009-02-10 11:15:34 -05002456int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2457 struct jbd2_inode *jinode,
Jan Karac851ed52008-07-11 19:27:31 -04002458 loff_t new_size)
2459{
Jan Kara7f5aa212009-02-10 11:15:34 -05002460 transaction_t *inode_trans, *commit_trans;
Jan Karac851ed52008-07-11 19:27:31 -04002461 int ret = 0;
2462
Jan Kara7f5aa212009-02-10 11:15:34 -05002463 /* This is a quick check to avoid locking if not necessary */
2464 if (!jinode->i_transaction)
Jan Karac851ed52008-07-11 19:27:31 -04002465 goto out;
Jan Kara7f5aa212009-02-10 11:15:34 -05002466 /* Locks are here just to force reading of recent values, it is
2467 * enough that the transaction was not committing before we started
2468 * a transaction adding the inode to orphan list */
Theodore Ts'oa931da62010-08-03 21:35:12 -04002469 read_lock(&journal->j_state_lock);
Jan Karac851ed52008-07-11 19:27:31 -04002470 commit_trans = journal->j_committing_transaction;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002471 read_unlock(&journal->j_state_lock);
Jan Kara7f5aa212009-02-10 11:15:34 -05002472 spin_lock(&journal->j_list_lock);
2473 inode_trans = jinode->i_transaction;
2474 spin_unlock(&journal->j_list_lock);
2475 if (inode_trans == commit_trans) {
2476 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
Jan Karac851ed52008-07-11 19:27:31 -04002477 new_size, LLONG_MAX);
2478 if (ret)
2479 jbd2_journal_abort(journal, ret);
2480 }
2481out:
2482 return ret;
2483}