blob: 60bb365f54a52efb8ff859ff746ac1af201bb7d2 [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);
Dan Carpenter92e3b402014-02-17 20:33:01 -0500517 if (ret < 0) {
Jan Kara8f7d89f2013-06-04 12:35:11 -0400518 jbd2_journal_free_reserved(handle);
Dan Carpenter92e3b402014-02-17 20:33:01 -0500519 return ret;
520 }
Jan Kara8f7d89f2013-06-04 12:35:11 -0400521 handle->h_type = type;
522 handle->h_line_no = line_no;
Dan Carpenter92e3b402014-02-17 20:33:01 -0500523 return 0;
Jan Kara8f7d89f2013-06-04 12:35:11 -0400524}
525EXPORT_SYMBOL(jbd2_journal_start_reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700526
527/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700528 * int jbd2_journal_extend() - extend buffer credits.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700529 * @handle: handle to 'extend'
530 * @nblocks: nr blocks to try to extend by.
531 *
532 * Some transactions, such as large extends and truncates, can be done
533 * atomically all at once or in several stages. The operation requests
534 * a credit for a number of buffer modications in advance, but can
535 * extend its credit if it needs more.
536 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700537 * jbd2_journal_extend tries to give the running handle more buffer credits.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700538 * It does not guarantee that allocation - this is a best-effort only.
539 * The calling process MUST be able to deal cleanly with a failure to
540 * extend here.
541 *
542 * Return 0 on success, non-zero on failure.
543 *
544 * return code < 0 implies an error
545 * return code > 0 implies normal transaction-full status.
546 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700547int jbd2_journal_extend(handle_t *handle, int nblocks)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700548{
549 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400550 journal_t *journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700551 int result;
552 int wanted;
553
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400554 WARN_ON(!transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700555 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400556 return -EROFS;
557 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700558
559 result = 1;
560
Theodore Ts'oa931da62010-08-03 21:35:12 -0400561 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700562
563 /* Don't extend a locked-down transaction! */
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400564 if (transaction->t_state != T_RUNNING) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700565 jbd_debug(3, "denied handle %p %d blocks: "
566 "transaction not running\n", handle, nblocks);
567 goto error_out;
568 }
569
570 spin_lock(&transaction->t_handle_lock);
Jan Karafe1e8db2013-06-04 12:22:15 -0400571 wanted = atomic_add_return(nblocks,
572 &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700573
574 if (wanted > journal->j_max_transaction_buffers) {
575 jbd_debug(3, "denied handle %p %d blocks: "
576 "transaction too large\n", handle, nblocks);
Jan Karafe1e8db2013-06-04 12:22:15 -0400577 atomic_sub(nblocks, &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700578 goto unlock;
579 }
580
Jan Kara76c39902013-06-04 12:12:57 -0400581 if (wanted + (wanted >> JBD2_CONTROL_BLOCKS_SHIFT) >
582 jbd2_log_space_left(journal)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700583 jbd_debug(3, "denied handle %p %d blocks: "
584 "insufficient log space\n", handle, nblocks);
Jan Karafe1e8db2013-06-04 12:22:15 -0400585 atomic_sub(nblocks, &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700586 goto unlock;
587 }
588
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500589 trace_jbd2_handle_extend(journal->j_fs_dev->bd_dev,
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400590 transaction->t_tid,
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500591 handle->h_type, handle->h_line_no,
592 handle->h_buffer_credits,
593 nblocks);
594
Dave Kleikamp470decc2006-10-11 01:20:57 -0700595 handle->h_buffer_credits += nblocks;
Theodore Ts'o343d9c22013-02-08 13:00:22 -0500596 handle->h_requested_credits += nblocks;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700597 result = 0;
598
599 jbd_debug(3, "extended handle %p by %d\n", handle, nblocks);
600unlock:
601 spin_unlock(&transaction->t_handle_lock);
602error_out:
Theodore Ts'oa931da62010-08-03 21:35:12 -0400603 read_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700604 return result;
605}
606
607
608/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700609 * int jbd2_journal_restart() - restart a handle .
Dave Kleikamp470decc2006-10-11 01:20:57 -0700610 * @handle: handle to restart
611 * @nblocks: nr credits requested
612 *
613 * Restart a handle for a multi-transaction filesystem
614 * operation.
615 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700616 * If the jbd2_journal_extend() call above fails to grant new buffer credits
617 * to a running handle, a call to jbd2_journal_restart will commit the
Dave Kleikamp470decc2006-10-11 01:20:57 -0700618 * handle's transaction so far and reattach the handle to a new
619 * transaction capabable of guaranteeing the requested number of
Jan Kara8f7d89f2013-06-04 12:35:11 -0400620 * credits. We preserve reserved handle if there's any attached to the
621 * passed in handle.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700622 */
Dan Carpenterd2159fb2011-09-04 10:20:14 -0400623int jbd2__journal_restart(handle_t *handle, int nblocks, gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700624{
625 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400626 journal_t *journal;
Theodore Ts'oe4471832011-02-12 08:18:24 -0500627 tid_t tid;
628 int need_to_start, ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700629
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400630 WARN_ON(!transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700631 /* If we've had an abort of any type, don't even think about
632 * actually doing the restart! */
633 if (is_handle_aborted(handle))
634 return 0;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400635 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700636
637 /*
638 * First unlink the handle from its current transaction, and start the
639 * commit on that.
640 */
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400641 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700642 J_ASSERT(journal_current_handle() == handle);
643
Theodore Ts'oa931da62010-08-03 21:35:12 -0400644 read_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700645 spin_lock(&transaction->t_handle_lock);
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400646 atomic_sub(handle->h_buffer_credits,
647 &transaction->t_outstanding_credits);
Jan Kara8f7d89f2013-06-04 12:35:11 -0400648 if (handle->h_rsv_handle) {
649 sub_reserved_credits(journal,
650 handle->h_rsv_handle->h_buffer_credits);
651 }
Theodore Ts'oa51dca92010-08-02 08:43:25 -0400652 if (atomic_dec_and_test(&transaction->t_updates))
Dave Kleikamp470decc2006-10-11 01:20:57 -0700653 wake_up(&journal->j_wait_updates);
Theodore Ts'o39c04152013-07-01 08:12:40 -0400654 tid = transaction->t_tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700655 spin_unlock(&transaction->t_handle_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400656 handle->h_transaction = NULL;
657 current->journal_info = NULL;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700658
659 jbd_debug(2, "restarting handle %p\n", handle);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500660 need_to_start = !tid_geq(journal->j_commit_request, tid);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400661 read_unlock(&journal->j_state_lock);
Theodore Ts'oe4471832011-02-12 08:18:24 -0500662 if (need_to_start)
663 jbd2_log_start_commit(journal, tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700664
Jan Kara9599b0e2009-08-17 21:23:17 -0400665 lock_map_release(&handle->h_lockdep_map);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700666 handle->h_buffer_credits = nblocks;
Theodore Ts'o47def822010-07-27 11:56:05 -0400667 ret = start_this_handle(journal, handle, gfp_mask);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700668 return ret;
669}
Theodore Ts'o47def822010-07-27 11:56:05 -0400670EXPORT_SYMBOL(jbd2__journal_restart);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700671
672
Theodore Ts'o47def822010-07-27 11:56:05 -0400673int jbd2_journal_restart(handle_t *handle, int nblocks)
674{
675 return jbd2__journal_restart(handle, nblocks, GFP_NOFS);
676}
677EXPORT_SYMBOL(jbd2_journal_restart);
678
Dave Kleikamp470decc2006-10-11 01:20:57 -0700679/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700680 * void jbd2_journal_lock_updates () - establish a transaction barrier.
Dave Kleikamp470decc2006-10-11 01:20:57 -0700681 * @journal: Journal to establish a barrier on.
682 *
683 * This locks out any further updates from being started, and blocks
684 * until all existing updates have completed, returning only once the
685 * journal is in a quiescent state with no updates running.
686 *
687 * The journal lock should not be held on entry.
688 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700689void jbd2_journal_lock_updates(journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700690{
691 DEFINE_WAIT(wait);
692
Theodore Ts'oa931da62010-08-03 21:35:12 -0400693 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700694 ++journal->j_barrier_count;
695
Jan Kara8f7d89f2013-06-04 12:35:11 -0400696 /* Wait until there are no reserved handles */
697 if (atomic_read(&journal->j_reserved_credits)) {
698 write_unlock(&journal->j_state_lock);
699 wait_event(journal->j_wait_reserved,
700 atomic_read(&journal->j_reserved_credits) == 0);
701 write_lock(&journal->j_state_lock);
702 }
703
Dave Kleikamp470decc2006-10-11 01:20:57 -0700704 /* Wait until there are no running updates */
705 while (1) {
706 transaction_t *transaction = journal->j_running_transaction;
707
708 if (!transaction)
709 break;
710
711 spin_lock(&transaction->t_handle_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700712 prepare_to_wait(&journal->j_wait_updates, &wait,
713 TASK_UNINTERRUPTIBLE);
Jan Kara9837d8e2012-01-04 22:03:11 -0500714 if (!atomic_read(&transaction->t_updates)) {
715 spin_unlock(&transaction->t_handle_lock);
716 finish_wait(&journal->j_wait_updates, &wait);
717 break;
718 }
Dave Kleikamp470decc2006-10-11 01:20:57 -0700719 spin_unlock(&transaction->t_handle_lock);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400720 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700721 schedule();
722 finish_wait(&journal->j_wait_updates, &wait);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400723 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700724 }
Theodore Ts'oa931da62010-08-03 21:35:12 -0400725 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700726
727 /*
728 * We have now established a barrier against other normal updates, but
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700729 * we also need to barrier against other jbd2_journal_lock_updates() calls
Dave Kleikamp470decc2006-10-11 01:20:57 -0700730 * to make sure that we serialise special journal-locked operations
731 * too.
732 */
733 mutex_lock(&journal->j_barrier);
734}
735
736/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700737 * void jbd2_journal_unlock_updates (journal_t* journal) - release barrier
Dave Kleikamp470decc2006-10-11 01:20:57 -0700738 * @journal: Journal to release the barrier on.
739 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700740 * Release a transaction barrier obtained with jbd2_journal_lock_updates().
Dave Kleikamp470decc2006-10-11 01:20:57 -0700741 *
742 * Should be called without the journal lock held.
743 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700744void jbd2_journal_unlock_updates (journal_t *journal)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700745{
746 J_ASSERT(journal->j_barrier_count != 0);
747
748 mutex_unlock(&journal->j_barrier);
Theodore Ts'oa931da62010-08-03 21:35:12 -0400749 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700750 --journal->j_barrier_count;
Theodore Ts'oa931da62010-08-03 21:35:12 -0400751 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700752 wake_up(&journal->j_wait_transaction_locked);
753}
754
Jan Karaf91d1d02009-07-13 16:16:20 -0400755static void warn_dirty_buffer(struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -0700756{
Jan Karaf91d1d02009-07-13 16:16:20 -0400757 char b[BDEVNAME_SIZE];
Dave Kleikamp470decc2006-10-11 01:20:57 -0700758
Jan Karaf91d1d02009-07-13 16:16:20 -0400759 printk(KERN_WARNING
Eryu Guanf2a44522011-11-01 19:09:18 -0400760 "JBD2: Spotted dirty metadata buffer (dev = %s, blocknr = %llu). "
Jan Karaf91d1d02009-07-13 16:16:20 -0400761 "There's a risk of filesystem corruption in case of system "
762 "crash.\n",
763 bdevname(bh->b_bdev, b), (unsigned long long)bh->b_blocknr);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700764}
765
Jan Karab34090e2013-06-04 12:08:56 -0400766static int sleep_on_shadow_bh(void *word)
767{
768 io_schedule();
769 return 0;
770}
771
Dave Kleikamp470decc2006-10-11 01:20:57 -0700772/*
773 * If the buffer is already part of the current transaction, then there
774 * is nothing we need to do. If it is already part of a prior
775 * transaction which we are still committing to disk, then we need to
776 * make sure that we do not overwrite the old copy: we do copy-out to
777 * preserve the copy going to disk. We also account the buffer against
778 * the handle's metadata buffer credits (unless the buffer is already
779 * part of the transaction, that is).
780 *
781 */
782static int
783do_get_write_access(handle_t *handle, struct journal_head *jh,
784 int force_copy)
785{
786 struct buffer_head *bh;
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400787 transaction_t *transaction = handle->h_transaction;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700788 journal_t *journal;
789 int error;
790 char *frozen_buffer = NULL;
791 int need_copy = 0;
Theodore Ts'of783f092013-04-21 16:47:54 -0400792 unsigned long start_lock, time_lock;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700793
Theodore Ts'o41a5b912013-07-01 08:12:41 -0400794 WARN_ON(!transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700795 if (is_handle_aborted(handle))
796 return -EROFS;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700797 journal = transaction->t_journal;
798
Theodore Ts'ocfef2c62010-12-18 13:07:34 -0500799 jbd_debug(5, "journal_head %p, force_copy %d\n", jh, force_copy);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700800
801 JBUFFER_TRACE(jh, "entry");
802repeat:
803 bh = jh2bh(jh);
804
805 /* @@@ Need to check for errors here at some point. */
806
Theodore Ts'of783f092013-04-21 16:47:54 -0400807 start_lock = jiffies;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700808 lock_buffer(bh);
809 jbd_lock_bh_state(bh);
810
Theodore Ts'of783f092013-04-21 16:47:54 -0400811 /* If it takes too long to lock the buffer, trace it */
812 time_lock = jbd2_time_diff(start_lock, jiffies);
813 if (time_lock > HZ/10)
814 trace_jbd2_lock_buffer_stall(bh->b_bdev->bd_dev,
815 jiffies_to_msecs(time_lock));
816
Dave Kleikamp470decc2006-10-11 01:20:57 -0700817 /* We now hold the buffer lock so it is safe to query the buffer
818 * state. Is the buffer dirty?
819 *
820 * If so, there are two possibilities. The buffer may be
821 * non-journaled, and undergoing a quite legitimate writeback.
822 * Otherwise, it is journaled, and we don't expect dirty buffers
823 * in that state (the buffers should be marked JBD_Dirty
824 * instead.) So either the IO is being done under our own
825 * control and this is a bug, or it's a third party IO such as
826 * dump(8) (which may leave the buffer scheduled for read ---
827 * ie. locked but not dirty) or tune2fs (which may actually have
828 * the buffer dirtied, ugh.) */
829
830 if (buffer_dirty(bh)) {
831 /*
832 * First question: is this buffer already part of the current
833 * transaction or the existing committing transaction?
834 */
835 if (jh->b_transaction) {
836 J_ASSERT_JH(jh,
837 jh->b_transaction == transaction ||
838 jh->b_transaction ==
839 journal->j_committing_transaction);
840 if (jh->b_next_transaction)
841 J_ASSERT_JH(jh, jh->b_next_transaction ==
842 transaction);
Jan Karaf91d1d02009-07-13 16:16:20 -0400843 warn_dirty_buffer(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700844 }
845 /*
846 * In any case we need to clean the dirty flag and we must
847 * do it under the buffer lock to be sure we don't race
848 * with running write-out.
849 */
Jan Karaf91d1d02009-07-13 16:16:20 -0400850 JBUFFER_TRACE(jh, "Journalling dirty buffer");
851 clear_buffer_dirty(bh);
852 set_buffer_jbddirty(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700853 }
854
855 unlock_buffer(bh);
856
857 error = -EROFS;
858 if (is_handle_aborted(handle)) {
859 jbd_unlock_bh_state(bh);
860 goto out;
861 }
862 error = 0;
863
864 /*
865 * The buffer is already part of this transaction if b_transaction or
866 * b_next_transaction points to it
867 */
868 if (jh->b_transaction == transaction ||
869 jh->b_next_transaction == transaction)
870 goto done;
871
872 /*
Josef Bacik9fc7c632008-04-17 10:38:59 -0400873 * this is the first time this transaction is touching this buffer,
874 * reset the modified flag
875 */
876 jh->b_modified = 0;
877
878 /*
Dave Kleikamp470decc2006-10-11 01:20:57 -0700879 * If there is already a copy-out version of this buffer, then we don't
880 * need to make another one
881 */
882 if (jh->b_frozen_data) {
883 JBUFFER_TRACE(jh, "has frozen data");
884 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
885 jh->b_next_transaction = transaction;
886 goto done;
887 }
888
889 /* Is there data here we need to preserve? */
890
891 if (jh->b_transaction && jh->b_transaction != transaction) {
892 JBUFFER_TRACE(jh, "owned by older transaction");
893 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
894 J_ASSERT_JH(jh, jh->b_transaction ==
895 journal->j_committing_transaction);
896
897 /* There is one case we have to be very careful about.
898 * If the committing transaction is currently writing
899 * this buffer out to disk and has NOT made a copy-out,
900 * then we cannot modify the buffer contents at all
901 * right now. The essence of copy-out is that it is the
902 * extra copy, not the primary copy, which gets
903 * journaled. If the primary copy is already going to
904 * disk then we cannot do copy-out here. */
905
Jan Karab34090e2013-06-04 12:08:56 -0400906 if (buffer_shadow(bh)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700907 JBUFFER_TRACE(jh, "on shadow: sleep");
908 jbd_unlock_bh_state(bh);
Jan Karab34090e2013-06-04 12:08:56 -0400909 wait_on_bit(&bh->b_state, BH_Shadow,
910 sleep_on_shadow_bh, TASK_UNINTERRUPTIBLE);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700911 goto repeat;
912 }
913
Jan Karab34090e2013-06-04 12:08:56 -0400914 /*
915 * Only do the copy if the currently-owning transaction still
916 * needs it. If buffer isn't on BJ_Metadata list, the
917 * committing transaction is past that stage (here we use the
918 * fact that BH_Shadow is set under bh_state lock together with
919 * refiling to BJ_Shadow list and at this point we know the
920 * buffer doesn't have BH_Shadow set).
Dave Kleikamp470decc2006-10-11 01:20:57 -0700921 *
922 * Subtle point, though: if this is a get_undo_access,
923 * then we will be relying on the frozen_data to contain
924 * the new value of the committed_data record after the
925 * transaction, so we HAVE to force the frozen_data copy
Jan Karab34090e2013-06-04 12:08:56 -0400926 * in that case.
927 */
928 if (jh->b_jlist == BJ_Metadata || force_copy) {
Dave Kleikamp470decc2006-10-11 01:20:57 -0700929 JBUFFER_TRACE(jh, "generate frozen data");
930 if (!frozen_buffer) {
931 JBUFFER_TRACE(jh, "allocate memory for buffer");
932 jbd_unlock_bh_state(bh);
933 frozen_buffer =
Mingming Caoaf1e76d2007-10-16 18:38:25 -0400934 jbd2_alloc(jh2bh(jh)->b_size,
Dave Kleikamp470decc2006-10-11 01:20:57 -0700935 GFP_NOFS);
936 if (!frozen_buffer) {
Jan Kara75685072013-12-08 21:13:59 -0500937 printk(KERN_ERR
Dave Kleikamp470decc2006-10-11 01:20:57 -0700938 "%s: OOM for frozen_buffer\n",
Harvey Harrison329d2912008-04-17 10:38:59 -0400939 __func__);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700940 JBUFFER_TRACE(jh, "oom!");
941 error = -ENOMEM;
942 jbd_lock_bh_state(bh);
943 goto done;
944 }
945 goto repeat;
946 }
947 jh->b_frozen_data = frozen_buffer;
948 frozen_buffer = NULL;
949 need_copy = 1;
950 }
951 jh->b_next_transaction = transaction;
952 }
953
954
955 /*
956 * Finally, if the buffer is not journaled right now, we need to make
957 * sure it doesn't get written to disk before the caller actually
958 * commits the new data
959 */
960 if (!jh->b_transaction) {
961 JBUFFER_TRACE(jh, "no transaction");
962 J_ASSERT_JH(jh, !jh->b_next_transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700963 JBUFFER_TRACE(jh, "file as BJ_Reserved");
964 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700965 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700966 spin_unlock(&journal->j_list_lock);
967 }
968
969done:
970 if (need_copy) {
971 struct page *page;
972 int offset;
973 char *source;
974
975 J_EXPECT_JH(jh, buffer_uptodate(jh2bh(jh)),
976 "Possible IO failure.\n");
977 page = jh2bh(jh)->b_page;
Theodore Ts'oa1dd5332010-12-18 13:13:40 -0500978 offset = offset_in_page(jh2bh(jh)->b_data);
Cong Wang303a8f22011-11-25 23:14:31 +0800979 source = kmap_atomic(page);
Jan Kara13ceef02010-07-14 07:56:33 +0200980 /* Fire data frozen trigger just before we copy the data */
981 jbd2_buffer_frozen_trigger(jh, source + offset,
982 jh->b_triggers);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700983 memcpy(jh->b_frozen_data, source+offset, jh2bh(jh)->b_size);
Cong Wang303a8f22011-11-25 23:14:31 +0800984 kunmap_atomic(source);
Joel Beckere06c8222008-09-11 15:35:47 -0700985
986 /*
987 * Now that the frozen data is saved off, we need to store
988 * any matching triggers.
989 */
990 jh->b_frozen_triggers = jh->b_triggers;
Dave Kleikamp470decc2006-10-11 01:20:57 -0700991 }
992 jbd_unlock_bh_state(bh);
993
994 /*
995 * If we are about to journal a buffer, then any revoke pending on it is
996 * no longer valid
997 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -0700998 jbd2_journal_cancel_revoke(handle, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -0700999
1000out:
1001 if (unlikely(frozen_buffer)) /* It's usually NULL */
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001002 jbd2_free(frozen_buffer, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001003
1004 JBUFFER_TRACE(jh, "exit");
1005 return error;
1006}
1007
1008/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001009 * int jbd2_journal_get_write_access() - notify intent to modify a buffer for metadata (not data) update.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001010 * @handle: transaction to add buffer modifications to
1011 * @bh: bh to be used for metadata writes
Dave Kleikamp470decc2006-10-11 01:20:57 -07001012 *
1013 * Returns an error code or 0 on success.
1014 *
1015 * In full data journalling mode the buffer may be of type BJ_AsyncData,
1016 * because we're write()ing a buffer which is also part of a shared mapping.
1017 */
1018
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001019int jbd2_journal_get_write_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001020{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001021 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001022 int rc;
1023
1024 /* We do not want to get caught playing with fields which the
1025 * log thread also manipulates. Make sure that the buffer
1026 * completes any outstanding IO before proceeding. */
1027 rc = do_get_write_access(handle, jh, 0);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001028 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001029 return rc;
1030}
1031
1032
1033/*
1034 * When the user wants to journal a newly created buffer_head
1035 * (ie. getblk() returned a new buffer and we are going to populate it
1036 * manually rather than reading off disk), then we need to keep the
1037 * buffer_head locked until it has been completely filled with new
1038 * data. In this case, we should be able to make the assertion that
1039 * the bh is not already part of an existing transaction.
1040 *
1041 * The buffer should already be locked by the caller by this point.
1042 * There is no lock ranking violation: it was a newly created,
1043 * unlocked buffer beforehand. */
1044
1045/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001046 * int jbd2_journal_get_create_access () - notify intent to use newly created bh
Dave Kleikamp470decc2006-10-11 01:20:57 -07001047 * @handle: transaction to new buffer to
1048 * @bh: new buffer.
1049 *
1050 * Call this if you create a new bh.
1051 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001052int jbd2_journal_get_create_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001053{
1054 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001055 journal_t *journal;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001056 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001057 int err;
1058
1059 jbd_debug(5, "journal_head %p\n", jh);
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001060 WARN_ON(!transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001061 err = -EROFS;
1062 if (is_handle_aborted(handle))
1063 goto out;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001064 journal = transaction->t_journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001065 err = 0;
1066
1067 JBUFFER_TRACE(jh, "entry");
1068 /*
1069 * The buffer may already belong to this transaction due to pre-zeroing
1070 * in the filesystem's new_block code. It may also be on the previous,
1071 * committing transaction's lists, but it HAS to be in Forget state in
1072 * that case: the transaction must have deleted the buffer for it to be
1073 * reused here.
1074 */
1075 jbd_lock_bh_state(bh);
1076 spin_lock(&journal->j_list_lock);
1077 J_ASSERT_JH(jh, (jh->b_transaction == transaction ||
1078 jh->b_transaction == NULL ||
1079 (jh->b_transaction == journal->j_committing_transaction &&
1080 jh->b_jlist == BJ_Forget)));
1081
1082 J_ASSERT_JH(jh, jh->b_next_transaction == NULL);
1083 J_ASSERT_JH(jh, buffer_locked(jh2bh(jh)));
1084
1085 if (jh->b_transaction == NULL) {
Jan Karaf91d1d02009-07-13 16:16:20 -04001086 /*
1087 * Previous jbd2_journal_forget() could have left the buffer
1088 * with jbddirty bit set because it was being committed. When
1089 * the commit finished, we've filed the buffer for
1090 * checkpointing and marked it dirty. Now we are reallocating
1091 * the buffer so the transaction freeing it must have
1092 * committed and so it's safe to clear the dirty bit.
1093 */
1094 clear_buffer_dirty(jh2bh(jh));
Josef Bacik9fc7c632008-04-17 10:38:59 -04001095 /* first access by this transaction */
1096 jh->b_modified = 0;
1097
Dave Kleikamp470decc2006-10-11 01:20:57 -07001098 JBUFFER_TRACE(jh, "file as BJ_Reserved");
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001099 __jbd2_journal_file_buffer(jh, transaction, BJ_Reserved);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001100 } else if (jh->b_transaction == journal->j_committing_transaction) {
Josef Bacik9fc7c632008-04-17 10:38:59 -04001101 /* first access by this transaction */
1102 jh->b_modified = 0;
1103
Dave Kleikamp470decc2006-10-11 01:20:57 -07001104 JBUFFER_TRACE(jh, "set next transaction");
1105 jh->b_next_transaction = transaction;
1106 }
1107 spin_unlock(&journal->j_list_lock);
1108 jbd_unlock_bh_state(bh);
1109
1110 /*
1111 * akpm: I added this. ext3_alloc_branch can pick up new indirect
1112 * blocks which contain freed but then revoked metadata. We need
1113 * to cancel the revoke in case we end up freeing it yet again
1114 * and the reallocating as data - this would cause a second revoke,
1115 * which hits an assertion error.
1116 */
1117 JBUFFER_TRACE(jh, "cancelling revoke");
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001118 jbd2_journal_cancel_revoke(handle, jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001119out:
Ding Dinghua3991b402011-05-25 17:43:48 -04001120 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001121 return err;
1122}
1123
1124/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001125 * int jbd2_journal_get_undo_access() - Notify intent to modify metadata with
Dave Kleikamp470decc2006-10-11 01:20:57 -07001126 * non-rewindable consequences
1127 * @handle: transaction
1128 * @bh: buffer to undo
Dave Kleikamp470decc2006-10-11 01:20:57 -07001129 *
1130 * Sometimes there is a need to distinguish between metadata which has
1131 * been committed to disk and that which has not. The ext3fs code uses
1132 * this for freeing and allocating space, we have to make sure that we
1133 * do not reuse freed space until the deallocation has been committed,
1134 * since if we overwrote that space we would make the delete
1135 * un-rewindable in case of a crash.
1136 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001137 * To deal with that, jbd2_journal_get_undo_access requests write access to a
Dave Kleikamp470decc2006-10-11 01:20:57 -07001138 * buffer for parts of non-rewindable operations such as delete
1139 * operations on the bitmaps. The journaling code must keep a copy of
1140 * the buffer's contents prior to the undo_access call until such time
1141 * as we know that the buffer has definitely been committed to disk.
1142 *
1143 * We never need to know which transaction the committed data is part
1144 * of, buffers touched here are guaranteed to be dirtied later and so
1145 * will be committed to a new transaction in due course, at which point
1146 * we can discard the old committed data pointer.
1147 *
1148 * Returns error number or 0 on success.
1149 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001150int jbd2_journal_get_undo_access(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001151{
1152 int err;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001153 struct journal_head *jh = jbd2_journal_add_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001154 char *committed_data = NULL;
1155
1156 JBUFFER_TRACE(jh, "entry");
1157
1158 /*
1159 * Do this first --- it can drop the journal lock, so we want to
1160 * make sure that obtaining the committed_data is done
1161 * atomically wrt. completion of any outstanding commits.
1162 */
1163 err = do_get_write_access(handle, jh, 1);
1164 if (err)
1165 goto out;
1166
1167repeat:
1168 if (!jh->b_committed_data) {
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001169 committed_data = jbd2_alloc(jh2bh(jh)->b_size, GFP_NOFS);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001170 if (!committed_data) {
Jan Kara75685072013-12-08 21:13:59 -05001171 printk(KERN_ERR "%s: No memory for committed data\n",
Harvey Harrison329d2912008-04-17 10:38:59 -04001172 __func__);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001173 err = -ENOMEM;
1174 goto out;
1175 }
1176 }
1177
1178 jbd_lock_bh_state(bh);
1179 if (!jh->b_committed_data) {
1180 /* Copy out the current buffer contents into the
1181 * preserved, committed copy. */
1182 JBUFFER_TRACE(jh, "generate b_committed data");
1183 if (!committed_data) {
1184 jbd_unlock_bh_state(bh);
1185 goto repeat;
1186 }
1187
1188 jh->b_committed_data = committed_data;
1189 committed_data = NULL;
1190 memcpy(jh->b_committed_data, bh->b_data, bh->b_size);
1191 }
1192 jbd_unlock_bh_state(bh);
1193out:
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001194 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001195 if (unlikely(committed_data))
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001196 jbd2_free(committed_data, bh->b_size);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001197 return err;
1198}
1199
1200/**
Joel Beckere06c8222008-09-11 15:35:47 -07001201 * void jbd2_journal_set_triggers() - Add triggers for commit writeout
1202 * @bh: buffer to trigger on
1203 * @type: struct jbd2_buffer_trigger_type containing the trigger(s).
1204 *
1205 * Set any triggers on this journal_head. This is always safe, because
1206 * triggers for a committing buffer will be saved off, and triggers for
1207 * a running transaction will match the buffer in that transaction.
1208 *
1209 * Call with NULL to clear the triggers.
1210 */
1211void jbd2_journal_set_triggers(struct buffer_head *bh,
1212 struct jbd2_buffer_trigger_type *type)
1213{
Jan Karaad56eda2013-03-11 13:24:56 -04001214 struct journal_head *jh = jbd2_journal_grab_journal_head(bh);
Joel Beckere06c8222008-09-11 15:35:47 -07001215
Jan Karaad56eda2013-03-11 13:24:56 -04001216 if (WARN_ON(!jh))
1217 return;
Joel Beckere06c8222008-09-11 15:35:47 -07001218 jh->b_triggers = type;
Jan Karaad56eda2013-03-11 13:24:56 -04001219 jbd2_journal_put_journal_head(jh);
Joel Beckere06c8222008-09-11 15:35:47 -07001220}
1221
Jan Kara13ceef02010-07-14 07:56:33 +02001222void jbd2_buffer_frozen_trigger(struct journal_head *jh, void *mapped_data,
Joel Beckere06c8222008-09-11 15:35:47 -07001223 struct jbd2_buffer_trigger_type *triggers)
1224{
1225 struct buffer_head *bh = jh2bh(jh);
1226
Jan Kara13ceef02010-07-14 07:56:33 +02001227 if (!triggers || !triggers->t_frozen)
Joel Beckere06c8222008-09-11 15:35:47 -07001228 return;
1229
Jan Kara13ceef02010-07-14 07:56:33 +02001230 triggers->t_frozen(triggers, bh, mapped_data, bh->b_size);
Joel Beckere06c8222008-09-11 15:35:47 -07001231}
1232
1233void jbd2_buffer_abort_trigger(struct journal_head *jh,
1234 struct jbd2_buffer_trigger_type *triggers)
1235{
1236 if (!triggers || !triggers->t_abort)
1237 return;
1238
1239 triggers->t_abort(triggers, jh2bh(jh));
1240}
1241
1242
1243
1244/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001245 * int jbd2_journal_dirty_metadata() - mark a buffer as containing dirty metadata
Dave Kleikamp470decc2006-10-11 01:20:57 -07001246 * @handle: transaction to add buffer to.
1247 * @bh: buffer to mark
1248 *
1249 * mark dirty metadata which needs to be journaled as part of the current
1250 * transaction.
1251 *
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001252 * The buffer must have previously had jbd2_journal_get_write_access()
1253 * called so that it has a valid journal_head attached to the buffer
1254 * head.
1255 *
Dave Kleikamp470decc2006-10-11 01:20:57 -07001256 * The buffer is placed on the transaction's metadata list and is marked
1257 * as belonging to the transaction.
1258 *
1259 * Returns error number or 0 on success.
1260 *
1261 * Special care needs to be taken if the buffer already belongs to the
1262 * current committing transaction (in which case we should have frozen
1263 * data present for that commit). In that case, we don't relink the
1264 * buffer: that only gets done when the old transaction finally
1265 * completes its commit.
1266 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001267int jbd2_journal_dirty_metadata(handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001268{
1269 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001270 journal_t *journal;
Jan Karaad56eda2013-03-11 13:24:56 -04001271 struct journal_head *jh;
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001272 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001273
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001274 WARN_ON(!transaction);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001275 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001276 return -EROFS;
1277 journal = transaction->t_journal;
Jan Karaad56eda2013-03-11 13:24:56 -04001278 jh = jbd2_journal_grab_journal_head(bh);
1279 if (!jh) {
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001280 ret = -EUCLEAN;
1281 goto out;
1282 }
Jan Karaad56eda2013-03-11 13:24:56 -04001283 jbd_debug(5, "journal_head %p\n", jh);
1284 JBUFFER_TRACE(jh, "entry");
Dave Kleikamp470decc2006-10-11 01:20:57 -07001285
1286 jbd_lock_bh_state(bh);
1287
1288 if (jh->b_modified == 0) {
1289 /*
1290 * This buffer's got modified and becoming part
1291 * of the transaction. This needs to be done
1292 * once a transaction -bzzz
1293 */
1294 jh->b_modified = 1;
Theodore Ts'of6c07ca2013-12-08 21:12:59 -05001295 if (handle->h_buffer_credits <= 0) {
1296 ret = -ENOSPC;
1297 goto out_unlock_bh;
1298 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001299 handle->h_buffer_credits--;
1300 }
1301
1302 /*
1303 * fastpath, to avoid expensive locking. If this buffer is already
1304 * on the running transaction's metadata list there is nothing to do.
1305 * Nobody can take it off again because there is a handle open.
1306 * I _think_ we're OK here with SMP barriers - a mistaken decision will
1307 * result in this test being false, so we go in and take the locks.
1308 */
1309 if (jh->b_transaction == transaction && jh->b_jlist == BJ_Metadata) {
1310 JBUFFER_TRACE(jh, "fastpath");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001311 if (unlikely(jh->b_transaction !=
1312 journal->j_running_transaction)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001313 printk(KERN_ERR "JBD2: %s: "
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001314 "jh->b_transaction (%llu, %p, %u) != "
1315 "journal->j_running_transaction (%p, %u)",
1316 journal->j_devname,
1317 (unsigned long long) bh->b_blocknr,
1318 jh->b_transaction,
1319 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1320 journal->j_running_transaction,
1321 journal->j_running_transaction ?
1322 journal->j_running_transaction->t_tid : 0);
1323 ret = -EINVAL;
1324 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001325 goto out_unlock_bh;
1326 }
1327
1328 set_buffer_jbddirty(bh);
1329
1330 /*
1331 * Metadata already on the current transaction list doesn't
1332 * need to be filed. Metadata on another transaction's list must
1333 * be committing, and will be refiled once the commit completes:
1334 * leave it alone for now.
1335 */
1336 if (jh->b_transaction != transaction) {
1337 JBUFFER_TRACE(jh, "already on other transaction");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001338 if (unlikely(jh->b_transaction !=
1339 journal->j_committing_transaction)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001340 printk(KERN_ERR "JBD2: %s: "
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001341 "jh->b_transaction (%llu, %p, %u) != "
1342 "journal->j_committing_transaction (%p, %u)",
1343 journal->j_devname,
1344 (unsigned long long) bh->b_blocknr,
1345 jh->b_transaction,
1346 jh->b_transaction ? jh->b_transaction->t_tid : 0,
1347 journal->j_committing_transaction,
1348 journal->j_committing_transaction ?
1349 journal->j_committing_transaction->t_tid : 0);
1350 ret = -EINVAL;
1351 }
1352 if (unlikely(jh->b_next_transaction != transaction)) {
Dmitry Monakhova67c8482013-12-08 21:14:59 -05001353 printk(KERN_ERR "JBD2: %s: "
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001354 "jh->b_next_transaction (%llu, %p, %u) != "
1355 "transaction (%p, %u)",
1356 journal->j_devname,
1357 (unsigned long long) bh->b_blocknr,
1358 jh->b_next_transaction,
1359 jh->b_next_transaction ?
1360 jh->b_next_transaction->t_tid : 0,
1361 transaction, transaction->t_tid);
1362 ret = -EINVAL;
1363 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001364 /* And this case is illegal: we can't reuse another
1365 * transaction's data buffer, ever. */
1366 goto out_unlock_bh;
1367 }
1368
1369 /* That test should have eliminated the following case: */
Mingming Cao40191912008-01-28 23:58:27 -05001370 J_ASSERT_JH(jh, jh->b_frozen_data == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001371
1372 JBUFFER_TRACE(jh, "file as BJ_Metadata");
1373 spin_lock(&journal->j_list_lock);
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001374 __jbd2_journal_file_buffer(jh, transaction, BJ_Metadata);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001375 spin_unlock(&journal->j_list_lock);
1376out_unlock_bh:
1377 jbd_unlock_bh_state(bh);
Jan Karaad56eda2013-03-11 13:24:56 -04001378 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001379out:
1380 JBUFFER_TRACE(jh, "exit");
Theodore Ts'o9ea7a0d2011-09-04 10:18:14 -04001381 return ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001382}
1383
Dave Kleikamp470decc2006-10-11 01:20:57 -07001384/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001385 * void jbd2_journal_forget() - bforget() for potentially-journaled buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001386 * @handle: transaction handle
1387 * @bh: bh to 'forget'
1388 *
1389 * We can only do the bforget if there are no commits pending against the
1390 * buffer. If the buffer is dirty in the current running transaction we
1391 * can safely unlink it.
1392 *
1393 * bh may not be a journalled buffer at all - it may be a non-JBD
1394 * buffer which came off the hashtable. Check for this.
1395 *
1396 * Decrements bh->b_count by one.
1397 *
1398 * Allow this call even if the handle has aborted --- it may be part of
1399 * the caller's cleanup after an abort.
1400 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001401int jbd2_journal_forget (handle_t *handle, struct buffer_head *bh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001402{
1403 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001404 journal_t *journal;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001405 struct journal_head *jh;
1406 int drop_reserve = 0;
1407 int err = 0;
Josef Bacik1dfc3222008-04-17 10:38:59 -04001408 int was_modified = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001409
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001410 WARN_ON(!transaction);
1411 if (is_handle_aborted(handle))
1412 return -EROFS;
1413 journal = transaction->t_journal;
1414
Dave Kleikamp470decc2006-10-11 01:20:57 -07001415 BUFFER_TRACE(bh, "entry");
1416
1417 jbd_lock_bh_state(bh);
1418 spin_lock(&journal->j_list_lock);
1419
1420 if (!buffer_jbd(bh))
1421 goto not_jbd;
1422 jh = bh2jh(bh);
1423
1424 /* Critical error: attempting to delete a bitmap buffer, maybe?
1425 * Don't do any jbd operations, and return an error. */
1426 if (!J_EXPECT_JH(jh, !jh->b_committed_data,
1427 "inconsistent data on disk")) {
1428 err = -EIO;
1429 goto not_jbd;
1430 }
1431
Adam Buchbinder48fc7f72012-09-19 21:48:00 -04001432 /* keep track of whether or not this transaction modified us */
Josef Bacik1dfc3222008-04-17 10:38:59 -04001433 was_modified = jh->b_modified;
1434
Dave Kleikamp470decc2006-10-11 01:20:57 -07001435 /*
1436 * The buffer's going from the transaction, we must drop
1437 * all references -bzzz
1438 */
1439 jh->b_modified = 0;
1440
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001441 if (jh->b_transaction == transaction) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001442 J_ASSERT_JH(jh, !jh->b_frozen_data);
1443
1444 /* If we are forgetting a buffer which is already part
1445 * of this transaction, then we can just drop it from
1446 * the transaction immediately. */
1447 clear_buffer_dirty(bh);
1448 clear_buffer_jbddirty(bh);
1449
1450 JBUFFER_TRACE(jh, "belongs to current transaction: unfile");
1451
Josef Bacik1dfc3222008-04-17 10:38:59 -04001452 /*
1453 * we only want to drop a reference if this transaction
1454 * modified the buffer
1455 */
1456 if (was_modified)
1457 drop_reserve = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001458
1459 /*
1460 * We are no longer going to journal this buffer.
1461 * However, the commit of this transaction is still
1462 * important to the buffer: the delete that we are now
1463 * processing might obsolete an old log entry, so by
1464 * committing, we can satisfy the buffer's checkpoint.
1465 *
1466 * So, if we have a checkpoint on the buffer, we should
1467 * now refile the buffer on our BJ_Forget list so that
1468 * we know to remove the checkpoint after we commit.
1469 */
1470
1471 if (jh->b_cp_transaction) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001472 __jbd2_journal_temp_unlink_buffer(jh);
1473 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001474 } else {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001475 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001476 if (!buffer_jbd(bh)) {
1477 spin_unlock(&journal->j_list_lock);
1478 jbd_unlock_bh_state(bh);
1479 __bforget(bh);
1480 goto drop;
1481 }
1482 }
1483 } else if (jh->b_transaction) {
1484 J_ASSERT_JH(jh, (jh->b_transaction ==
1485 journal->j_committing_transaction));
1486 /* However, if the buffer is still owned by a prior
1487 * (committing) transaction, we can't drop it yet... */
1488 JBUFFER_TRACE(jh, "belongs to older transaction");
1489 /* ... but we CAN drop it from the new transaction if we
1490 * have also modified it since the original commit. */
1491
1492 if (jh->b_next_transaction) {
1493 J_ASSERT(jh->b_next_transaction == transaction);
1494 jh->b_next_transaction = NULL;
Josef Bacik1dfc3222008-04-17 10:38:59 -04001495
1496 /*
1497 * only drop a reference if this transaction modified
1498 * the buffer
1499 */
1500 if (was_modified)
1501 drop_reserve = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001502 }
1503 }
1504
1505not_jbd:
1506 spin_unlock(&journal->j_list_lock);
1507 jbd_unlock_bh_state(bh);
1508 __brelse(bh);
1509drop:
1510 if (drop_reserve) {
1511 /* no need to reserve log space for this block -bzzz */
1512 handle->h_buffer_credits++;
1513 }
1514 return err;
1515}
1516
1517/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001518 * int jbd2_journal_stop() - complete a transaction
Dave Kleikamp470decc2006-10-11 01:20:57 -07001519 * @handle: tranaction to complete.
1520 *
1521 * All done for a particular handle.
1522 *
1523 * There is not much action needed here. We just return any remaining
1524 * buffer credits to the transaction and remove the handle. The only
1525 * complication is that we need to start a commit operation if the
1526 * filesystem is marked for synchronous update.
1527 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001528 * jbd2_journal_stop itself will not usually return an error, but it may
Dave Kleikamp470decc2006-10-11 01:20:57 -07001529 * do so in unusual circumstances. In particular, expect it to
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001530 * return -EIO if a jbd2_journal_abort has been executed since the
Dave Kleikamp470decc2006-10-11 01:20:57 -07001531 * transaction began.
1532 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001533int jbd2_journal_stop(handle_t *handle)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001534{
1535 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001536 journal_t *journal;
1537 int err = 0, wait_for_commit = 0;
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001538 tid_t tid;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001539 pid_t pid;
1540
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001541 if (!transaction)
1542 goto free_and_exit;
1543 journal = transaction->t_journal;
1544
Dave Kleikamp470decc2006-10-11 01:20:57 -07001545 J_ASSERT(journal_current_handle() == handle);
1546
1547 if (is_handle_aborted(handle))
1548 err = -EIO;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001549 else
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001550 J_ASSERT(atomic_read(&transaction->t_updates) > 0);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001551
1552 if (--handle->h_ref > 0) {
1553 jbd_debug(4, "h_ref %d -> %d\n", handle->h_ref + 1,
1554 handle->h_ref);
1555 return err;
1556 }
1557
1558 jbd_debug(4, "Handle %p going down\n", handle);
Theodore Ts'o343d9c22013-02-08 13:00:22 -05001559 trace_jbd2_handle_stats(journal->j_fs_dev->bd_dev,
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001560 transaction->t_tid,
Theodore Ts'o343d9c22013-02-08 13:00:22 -05001561 handle->h_type, handle->h_line_no,
1562 jiffies - handle->h_start_jiffies,
1563 handle->h_sync, handle->h_requested_credits,
1564 (handle->h_requested_credits -
1565 handle->h_buffer_credits));
Dave Kleikamp470decc2006-10-11 01:20:57 -07001566
1567 /*
1568 * Implement synchronous transaction batching. If the handle
1569 * was synchronous, don't force a commit immediately. Let's
Josef Bacike07f7182008-11-26 01:14:26 -05001570 * yield and let another thread piggyback onto this
1571 * transaction. Keep doing that while new threads continue to
1572 * arrive. It doesn't cost much - we're about to run a commit
1573 * and sleep on IO anyway. Speeds up many-threaded, many-dir
1574 * operations by 30x or more...
Dave Kleikamp470decc2006-10-11 01:20:57 -07001575 *
Josef Bacike07f7182008-11-26 01:14:26 -05001576 * We try and optimize the sleep time against what the
1577 * underlying disk can do, instead of having a static sleep
1578 * time. This is useful for the case where our storage is so
1579 * fast that it is more optimal to go ahead and force a flush
1580 * and wait for the transaction to be committed than it is to
1581 * wait for an arbitrary amount of time for new writers to
1582 * join the transaction. We achieve this by measuring how
1583 * long it takes to commit a transaction, and compare it with
1584 * how long this transaction has been running, and if run time
1585 * < commit time then we sleep for the delta and commit. This
1586 * greatly helps super fast disks that would see slowdowns as
1587 * more threads started doing fsyncs.
1588 *
1589 * But don't do this if this process was the most recent one
1590 * to perform a synchronous write. We do this to detect the
1591 * case where a single process is doing a stream of sync
1592 * writes. No point in waiting for joiners in that case.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001593 */
1594 pid = current->pid;
1595 if (handle->h_sync && journal->j_last_sync_writer != pid) {
Josef Bacike07f7182008-11-26 01:14:26 -05001596 u64 commit_time, trans_time;
1597
Dave Kleikamp470decc2006-10-11 01:20:57 -07001598 journal->j_last_sync_writer = pid;
Josef Bacike07f7182008-11-26 01:14:26 -05001599
Theodore Ts'oa931da62010-08-03 21:35:12 -04001600 read_lock(&journal->j_state_lock);
Josef Bacike07f7182008-11-26 01:14:26 -05001601 commit_time = journal->j_average_commit_time;
Theodore Ts'oa931da62010-08-03 21:35:12 -04001602 read_unlock(&journal->j_state_lock);
Josef Bacike07f7182008-11-26 01:14:26 -05001603
1604 trans_time = ktime_to_ns(ktime_sub(ktime_get(),
1605 transaction->t_start_time));
1606
Theodore Ts'o30773842009-01-03 20:27:38 -05001607 commit_time = max_t(u64, commit_time,
1608 1000*journal->j_min_batch_time);
Josef Bacike07f7182008-11-26 01:14:26 -05001609 commit_time = min_t(u64, commit_time,
Theodore Ts'o30773842009-01-03 20:27:38 -05001610 1000*journal->j_max_batch_time);
Josef Bacike07f7182008-11-26 01:14:26 -05001611
1612 if (trans_time < commit_time) {
1613 ktime_t expires = ktime_add_ns(ktime_get(),
1614 commit_time);
1615 set_current_state(TASK_UNINTERRUPTIBLE);
1616 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1617 }
Dave Kleikamp470decc2006-10-11 01:20:57 -07001618 }
1619
Theodore Ts'o70585482009-03-25 23:35:46 -04001620 if (handle->h_sync)
1621 transaction->t_synchronous_commit = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001622 current->journal_info = NULL;
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001623 atomic_sub(handle->h_buffer_credits,
1624 &transaction->t_outstanding_credits);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001625
1626 /*
1627 * If the handle is marked SYNC, we need to set another commit
1628 * going! We also want to force a commit if the current
1629 * transaction is occupying too much of the log, or if the
1630 * transaction is too old now.
1631 */
1632 if (handle->h_sync ||
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001633 (atomic_read(&transaction->t_outstanding_credits) >
1634 journal->j_max_transaction_buffers) ||
1635 time_after_eq(jiffies, transaction->t_expires)) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001636 /* Do this even for aborted journals: an abort still
1637 * completes the commit thread, it just doesn't write
1638 * anything to disk. */
Dave Kleikamp470decc2006-10-11 01:20:57 -07001639
Dave Kleikamp470decc2006-10-11 01:20:57 -07001640 jbd_debug(2, "transaction too old, requesting commit for "
1641 "handle %p\n", handle);
1642 /* This is non-blocking */
Theodore Ts'oc35a56a2010-05-16 05:00:00 -04001643 jbd2_log_start_commit(journal, transaction->t_tid);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001644
1645 /*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001646 * Special case: JBD2_SYNC synchronous updates require us
Dave Kleikamp470decc2006-10-11 01:20:57 -07001647 * to wait for the commit to complete.
1648 */
1649 if (handle->h_sync && !(current->flags & PF_MEMALLOC))
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001650 wait_for_commit = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001651 }
1652
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001653 /*
1654 * Once we drop t_updates, if it goes to zero the transaction
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001655 * could start committing on us and eventually disappear. So
Theodore Ts'oa51dca92010-08-02 08:43:25 -04001656 * once we do this, we must not dereference transaction
1657 * pointer again.
1658 */
1659 tid = transaction->t_tid;
1660 if (atomic_dec_and_test(&transaction->t_updates)) {
1661 wake_up(&journal->j_wait_updates);
1662 if (journal->j_barrier_count)
1663 wake_up(&journal->j_wait_transaction_locked);
1664 }
1665
1666 if (wait_for_commit)
1667 err = jbd2_log_wait_commit(journal, tid);
1668
Ingo Molnar3295f0e2008-08-11 10:30:30 +02001669 lock_map_release(&handle->h_lockdep_map);
Mingming Cao7b751062008-01-28 23:58:27 -05001670
Jan Kara8f7d89f2013-06-04 12:35:11 -04001671 if (handle->h_rsv_handle)
1672 jbd2_journal_free_reserved(handle->h_rsv_handle);
Theodore Ts'o41a5b912013-07-01 08:12:41 -04001673free_and_exit:
Mingming Caoaf1e76d2007-10-16 18:38:25 -04001674 jbd2_free_handle(handle);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001675 return err;
1676}
1677
Dave Kleikamp470decc2006-10-11 01:20:57 -07001678/*
1679 *
1680 * List management code snippets: various functions for manipulating the
1681 * transaction buffer lists.
1682 *
1683 */
1684
1685/*
1686 * Append a buffer to a transaction list, given the transaction's list head
1687 * pointer.
1688 *
1689 * j_list_lock is held.
1690 *
1691 * jbd_lock_bh_state(jh2bh(jh)) is held.
1692 */
1693
1694static inline void
1695__blist_add_buffer(struct journal_head **list, struct journal_head *jh)
1696{
1697 if (!*list) {
1698 jh->b_tnext = jh->b_tprev = jh;
1699 *list = jh;
1700 } else {
1701 /* Insert at the tail of the list to preserve order */
1702 struct journal_head *first = *list, *last = first->b_tprev;
1703 jh->b_tprev = last;
1704 jh->b_tnext = first;
1705 last->b_tnext = first->b_tprev = jh;
1706 }
1707}
1708
1709/*
1710 * Remove a buffer from a transaction list, given the transaction's list
1711 * head pointer.
1712 *
1713 * Called with j_list_lock held, and the journal may not be locked.
1714 *
1715 * jbd_lock_bh_state(jh2bh(jh)) is held.
1716 */
1717
1718static inline void
1719__blist_del_buffer(struct journal_head **list, struct journal_head *jh)
1720{
1721 if (*list == jh) {
1722 *list = jh->b_tnext;
1723 if (*list == jh)
1724 *list = NULL;
1725 }
1726 jh->b_tprev->b_tnext = jh->b_tnext;
1727 jh->b_tnext->b_tprev = jh->b_tprev;
1728}
1729
1730/*
1731 * Remove a buffer from the appropriate transaction list.
1732 *
1733 * Note that this function can *change* the value of
Jan Karaf5113ef2013-06-04 12:01:45 -04001734 * bh->b_transaction->t_buffers, t_forget, t_shadow_list, t_log_list or
1735 * t_reserved_list. If the caller is holding onto a copy of one of these
1736 * pointers, it could go bad. Generally the caller needs to re-read the
1737 * pointer from the transaction_t.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001738 *
Jan Kara5bebccf2012-03-13 22:25:06 -04001739 * Called under j_list_lock.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001740 */
Jan Kara5bebccf2012-03-13 22:25:06 -04001741static void __jbd2_journal_temp_unlink_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001742{
1743 struct journal_head **list = NULL;
1744 transaction_t *transaction;
1745 struct buffer_head *bh = jh2bh(jh);
1746
1747 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
1748 transaction = jh->b_transaction;
1749 if (transaction)
1750 assert_spin_locked(&transaction->t_journal->j_list_lock);
1751
1752 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
1753 if (jh->b_jlist != BJ_None)
Mingming Cao40191912008-01-28 23:58:27 -05001754 J_ASSERT_JH(jh, transaction != NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001755
1756 switch (jh->b_jlist) {
1757 case BJ_None:
1758 return;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001759 case BJ_Metadata:
1760 transaction->t_nr_buffers--;
1761 J_ASSERT_JH(jh, transaction->t_nr_buffers >= 0);
1762 list = &transaction->t_buffers;
1763 break;
1764 case BJ_Forget:
1765 list = &transaction->t_forget;
1766 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001767 case BJ_Shadow:
1768 list = &transaction->t_shadow_list;
1769 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001770 case BJ_Reserved:
1771 list = &transaction->t_reserved_list;
1772 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07001773 }
1774
1775 __blist_del_buffer(list, jh);
1776 jh->b_jlist = BJ_None;
1777 if (test_clear_buffer_jbddirty(bh))
1778 mark_buffer_dirty(bh); /* Expose it to the VM */
1779}
1780
Jan Karade1b7942011-06-13 15:38:22 -04001781/*
1782 * Remove buffer from all transactions.
1783 *
1784 * Called with bh_state lock and j_list_lock
1785 *
1786 * jh and bh may be already freed when this function returns.
1787 */
1788static void __jbd2_journal_unfile_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001789{
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001790 __jbd2_journal_temp_unlink_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001791 jh->b_transaction = NULL;
Jan Karade1b7942011-06-13 15:38:22 -04001792 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001793}
1794
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001795void jbd2_journal_unfile_buffer(journal_t *journal, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001796{
Jan Karade1b7942011-06-13 15:38:22 -04001797 struct buffer_head *bh = jh2bh(jh);
1798
1799 /* Get reference so that buffer cannot be freed before we unlock it */
1800 get_bh(bh);
1801 jbd_lock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001802 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001803 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001804 spin_unlock(&journal->j_list_lock);
Jan Karade1b7942011-06-13 15:38:22 -04001805 jbd_unlock_bh_state(bh);
1806 __brelse(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001807}
1808
1809/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001810 * Called from jbd2_journal_try_to_free_buffers().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001811 *
1812 * Called under jbd_lock_bh_state(bh)
1813 */
1814static void
1815__journal_try_to_free_buffer(journal_t *journal, struct buffer_head *bh)
1816{
1817 struct journal_head *jh;
1818
1819 jh = bh2jh(bh);
1820
1821 if (buffer_locked(bh) || buffer_dirty(bh))
1822 goto out;
1823
Mingming Cao40191912008-01-28 23:58:27 -05001824 if (jh->b_next_transaction != NULL)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001825 goto out;
1826
1827 spin_lock(&journal->j_list_lock);
Jan Kara87c89c22008-07-11 19:27:31 -04001828 if (jh->b_cp_transaction != NULL && jh->b_transaction == NULL) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07001829 /* written-back checkpointed metadata buffer */
Jan Karac254c9e2012-03-13 22:27:44 -04001830 JBUFFER_TRACE(jh, "remove from checkpoint list");
1831 __jbd2_journal_remove_checkpoint(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001832 }
1833 spin_unlock(&journal->j_list_lock);
1834out:
1835 return;
1836}
1837
Dave Kleikamp470decc2006-10-11 01:20:57 -07001838/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001839 * int jbd2_journal_try_to_free_buffers() - try to free page buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001840 * @journal: journal for operation
1841 * @page: to try and free
Mingming Cao530576b2008-07-13 21:06:39 -04001842 * @gfp_mask: we use the mask to detect how hard should we try to release
1843 * buffers. If __GFP_WAIT and __GFP_FS is set, we wait for commit code to
1844 * release the buffers.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001845 *
1846 *
1847 * For all the buffers on this page,
1848 * if they are fully written out ordered data, move them onto BUF_CLEAN
1849 * so try_to_free_buffers() can reap them.
1850 *
1851 * This function returns non-zero if we wish try_to_free_buffers()
1852 * to be called. We do this if the page is releasable by try_to_free_buffers().
1853 * We also do it if the page has locked or dirty buffers and the caller wants
1854 * us to perform sync or async writeout.
1855 *
1856 * This complicates JBD locking somewhat. We aren't protected by the
1857 * BKL here. We wish to remove the buffer from its committing or
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001858 * running transaction's ->t_datalist via __jbd2_journal_unfile_buffer.
Dave Kleikamp470decc2006-10-11 01:20:57 -07001859 *
1860 * This may *change* the value of transaction_t->t_datalist, so anyone
1861 * who looks at t_datalist needs to lock against this function.
1862 *
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001863 * Even worse, someone may be doing a jbd2_journal_dirty_data on this
1864 * buffer. So we need to lock against that. jbd2_journal_dirty_data()
Dave Kleikamp470decc2006-10-11 01:20:57 -07001865 * will come out of the lock with the buffer dirty, which makes it
1866 * ineligible for release here.
1867 *
1868 * Who else is affected by this? hmm... Really the only contender
1869 * is do_get_write_access() - it could be looking at the buffer while
1870 * journal_try_to_free_buffer() is changing its state. But that
1871 * cannot happen because we never reallocate freed data as metadata
1872 * while the data is part of a transaction. Yes?
Mingming Cao530576b2008-07-13 21:06:39 -04001873 *
1874 * Return 0 on failure, 1 on success
Dave Kleikamp470decc2006-10-11 01:20:57 -07001875 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001876int jbd2_journal_try_to_free_buffers(journal_t *journal,
Mingming Cao530576b2008-07-13 21:06:39 -04001877 struct page *page, gfp_t gfp_mask)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001878{
1879 struct buffer_head *head;
1880 struct buffer_head *bh;
1881 int ret = 0;
1882
1883 J_ASSERT(PageLocked(page));
1884
1885 head = page_buffers(page);
1886 bh = head;
1887 do {
1888 struct journal_head *jh;
1889
1890 /*
1891 * We take our own ref against the journal_head here to avoid
1892 * having to add tons of locking around each instance of
Mingming Cao530576b2008-07-13 21:06:39 -04001893 * jbd2_journal_put_journal_head().
Dave Kleikamp470decc2006-10-11 01:20:57 -07001894 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001895 jh = jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001896 if (!jh)
1897 continue;
1898
1899 jbd_lock_bh_state(bh);
1900 __journal_try_to_free_buffer(journal, bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001901 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001902 jbd_unlock_bh_state(bh);
1903 if (buffer_jbd(bh))
1904 goto busy;
1905 } while ((bh = bh->b_this_page) != head);
Mingming Cao530576b2008-07-13 21:06:39 -04001906
Dave Kleikamp470decc2006-10-11 01:20:57 -07001907 ret = try_to_free_buffers(page);
Mingming Cao530576b2008-07-13 21:06:39 -04001908
Dave Kleikamp470decc2006-10-11 01:20:57 -07001909busy:
1910 return ret;
1911}
1912
1913/*
1914 * This buffer is no longer needed. If it is on an older transaction's
1915 * checkpoint list we need to record it on this transaction's forget list
1916 * to pin this buffer (and hence its checkpointing transaction) down until
1917 * this transaction commits. If the buffer isn't on a checkpoint list, we
1918 * release it.
1919 * Returns non-zero if JBD no longer has an interest in the buffer.
1920 *
1921 * Called under j_list_lock.
1922 *
1923 * Called under jbd_lock_bh_state(bh).
1924 */
1925static int __dispose_buffer(struct journal_head *jh, transaction_t *transaction)
1926{
1927 int may_free = 1;
1928 struct buffer_head *bh = jh2bh(jh);
1929
Dave Kleikamp470decc2006-10-11 01:20:57 -07001930 if (jh->b_cp_transaction) {
1931 JBUFFER_TRACE(jh, "on running+cp transaction");
Jan Karade1b7942011-06-13 15:38:22 -04001932 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karaf91d1d02009-07-13 16:16:20 -04001933 /*
1934 * We don't want to write the buffer anymore, clear the
1935 * bit so that we don't confuse checks in
1936 * __journal_file_buffer
1937 */
1938 clear_buffer_dirty(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001939 __jbd2_journal_file_buffer(jh, transaction, BJ_Forget);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001940 may_free = 0;
1941 } else {
1942 JBUFFER_TRACE(jh, "on running transaction");
Jan Karade1b7942011-06-13 15:38:22 -04001943 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07001944 }
1945 return may_free;
1946}
1947
1948/*
Mingming Caof7f4bcc2006-10-11 01:20:59 -07001949 * jbd2_journal_invalidatepage
Dave Kleikamp470decc2006-10-11 01:20:57 -07001950 *
1951 * This code is tricky. It has a number of cases to deal with.
1952 *
1953 * There are two invariants which this code relies on:
1954 *
1955 * i_size must be updated on disk before we start calling invalidatepage on the
1956 * data.
1957 *
1958 * This is done in ext3 by defining an ext3_setattr method which
1959 * updates i_size before truncate gets going. By maintaining this
1960 * invariant, we can be sure that it is safe to throw away any buffers
1961 * attached to the current transaction: once the transaction commits,
1962 * we know that the data will not be needed.
1963 *
1964 * Note however that we can *not* throw away data belonging to the
1965 * previous, committing transaction!
1966 *
1967 * Any disk blocks which *are* part of the previous, committing
1968 * transaction (and which therefore cannot be discarded immediately) are
1969 * not going to be reused in the new running transaction
1970 *
1971 * The bitmap committed_data images guarantee this: any block which is
1972 * allocated in one transaction and removed in the next will be marked
1973 * as in-use in the committed_data bitmap, so cannot be reused until
1974 * the next transaction to delete the block commits. This means that
1975 * leaving committing buffers dirty is quite safe: the disk blocks
1976 * cannot be reallocated to a different file and so buffer aliasing is
1977 * not possible.
1978 *
1979 *
1980 * The above applies mainly to ordered data mode. In writeback mode we
1981 * don't make guarantees about the order in which data hits disk --- in
1982 * particular we don't guarantee that new dirty data is flushed before
1983 * transaction commit --- so it is always safe just to discard data
1984 * immediately in that mode. --sct
1985 */
1986
1987/*
1988 * The journal_unmap_buffer helper function returns zero if the buffer
1989 * concerned remains pinned as an anonymous buffer belonging to an older
1990 * transaction.
1991 *
1992 * We're outside-transaction here. Either or both of j_running_transaction
1993 * and j_committing_transaction may be NULL.
1994 */
Jan Karab794e7a2012-09-26 23:11:13 -04001995static int journal_unmap_buffer(journal_t *journal, struct buffer_head *bh,
1996 int partial_page)
Dave Kleikamp470decc2006-10-11 01:20:57 -07001997{
1998 transaction_t *transaction;
1999 struct journal_head *jh;
2000 int may_free = 1;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002001
2002 BUFFER_TRACE(bh, "entry");
2003
2004 /*
2005 * It is safe to proceed here without the j_list_lock because the
2006 * buffers cannot be stolen by try_to_free_buffers as long as we are
2007 * holding the page lock. --sct
2008 */
2009
2010 if (!buffer_jbd(bh))
2011 goto zap_buffer_unlocked;
2012
Jan Kara87c89c22008-07-11 19:27:31 -04002013 /* OK, we have data buffer in journaled mode */
Theodore Ts'oa931da62010-08-03 21:35:12 -04002014 write_lock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002015 jbd_lock_bh_state(bh);
2016 spin_lock(&journal->j_list_lock);
2017
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002018 jh = jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002019 if (!jh)
2020 goto zap_buffer_no_jh;
2021
dingdinghuaba869022010-02-15 16:35:42 -05002022 /*
2023 * We cannot remove the buffer from checkpoint lists until the
2024 * transaction adding inode to orphan list (let's call it T)
2025 * is committed. Otherwise if the transaction changing the
2026 * buffer would be cleaned from the journal before T is
2027 * committed, a crash will cause that the correct contents of
2028 * the buffer will be lost. On the other hand we have to
2029 * clear the buffer dirty bit at latest at the moment when the
2030 * transaction marking the buffer as freed in the filesystem
2031 * structures is committed because from that moment on the
Jan Karab794e7a2012-09-26 23:11:13 -04002032 * block can be reallocated and used by a different page.
dingdinghuaba869022010-02-15 16:35:42 -05002033 * Since the block hasn't been freed yet but the inode has
2034 * already been added to orphan list, it is safe for us to add
2035 * the buffer to BJ_Forget list of the newest transaction.
Jan Karab794e7a2012-09-26 23:11:13 -04002036 *
2037 * Also we have to clear buffer_mapped flag of a truncated buffer
2038 * because the buffer_head may be attached to the page straddling
2039 * i_size (can happen only when blocksize < pagesize) and thus the
2040 * buffer_head can be reused when the file is extended again. So we end
2041 * up keeping around invalidated buffers attached to transactions'
2042 * BJ_Forget list just to stop checkpointing code from cleaning up
2043 * the transaction this buffer was modified in.
dingdinghuaba869022010-02-15 16:35:42 -05002044 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002045 transaction = jh->b_transaction;
2046 if (transaction == NULL) {
2047 /* First case: not on any transaction. If it
2048 * has no checkpoint link, then we can zap it:
2049 * it's a writeback-mode buffer so we don't care
2050 * if it hits disk safely. */
2051 if (!jh->b_cp_transaction) {
2052 JBUFFER_TRACE(jh, "not on any transaction: zap");
2053 goto zap_buffer;
2054 }
2055
2056 if (!buffer_dirty(bh)) {
2057 /* bdflush has written it. We can drop it now */
2058 goto zap_buffer;
2059 }
2060
2061 /* OK, it must be in the journal but still not
2062 * written fully to disk: it's metadata or
2063 * journaled data... */
2064
2065 if (journal->j_running_transaction) {
2066 /* ... and once the current transaction has
2067 * committed, the buffer won't be needed any
2068 * longer. */
2069 JBUFFER_TRACE(jh, "checkpointed: add to BJ_Forget");
Jan Karab794e7a2012-09-26 23:11:13 -04002070 may_free = __dispose_buffer(jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002071 journal->j_running_transaction);
Jan Karab794e7a2012-09-26 23:11:13 -04002072 goto zap_buffer;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002073 } else {
2074 /* There is no currently-running transaction. So the
2075 * orphan record which we wrote for this file must have
2076 * passed into commit. We must attach this buffer to
2077 * the committing transaction, if it exists. */
2078 if (journal->j_committing_transaction) {
2079 JBUFFER_TRACE(jh, "give to committing trans");
Jan Karab794e7a2012-09-26 23:11:13 -04002080 may_free = __dispose_buffer(jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002081 journal->j_committing_transaction);
Jan Karab794e7a2012-09-26 23:11:13 -04002082 goto zap_buffer;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002083 } else {
2084 /* The orphan record's transaction has
2085 * committed. We can cleanse this buffer */
2086 clear_buffer_jbddirty(bh);
2087 goto zap_buffer;
2088 }
2089 }
2090 } else if (transaction == journal->j_committing_transaction) {
Eric Sandeen9b579882006-10-28 10:38:28 -07002091 JBUFFER_TRACE(jh, "on committing transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002092 /*
dingdinghuaba869022010-02-15 16:35:42 -05002093 * The buffer is committing, we simply cannot touch
Jan Karab794e7a2012-09-26 23:11:13 -04002094 * it. If the page is straddling i_size we have to wait
2095 * for commit and try again.
2096 */
2097 if (partial_page) {
Jan Karab794e7a2012-09-26 23:11:13 -04002098 jbd2_journal_put_journal_head(jh);
2099 spin_unlock(&journal->j_list_lock);
2100 jbd_unlock_bh_state(bh);
2101 write_unlock(&journal->j_state_lock);
Jan Kara53e87262012-12-25 13:29:52 -05002102 return -EBUSY;
Jan Karab794e7a2012-09-26 23:11:13 -04002103 }
2104 /*
2105 * OK, buffer won't be reachable after truncate. We just set
2106 * j_next_transaction to the running transaction (if there is
2107 * one) and mark buffer as freed so that commit code knows it
2108 * should clear dirty bits when it is done with the buffer.
dingdinghuaba869022010-02-15 16:35:42 -05002109 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002110 set_buffer_freed(bh);
dingdinghuaba869022010-02-15 16:35:42 -05002111 if (journal->j_running_transaction && buffer_jbddirty(bh))
2112 jh->b_next_transaction = journal->j_running_transaction;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002113 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002114 spin_unlock(&journal->j_list_lock);
2115 jbd_unlock_bh_state(bh);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002116 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002117 return 0;
2118 } else {
2119 /* Good, the buffer belongs to the running transaction.
2120 * We are writing our own transaction's data, not any
2121 * previous one's, so it is safe to throw it away
2122 * (remember that we expect the filesystem to have set
2123 * i_size already for this truncate so recovery will not
2124 * expose the disk blocks we are discarding here.) */
2125 J_ASSERT_JH(jh, transaction == journal->j_running_transaction);
Eric Sandeen9b579882006-10-28 10:38:28 -07002126 JBUFFER_TRACE(jh, "on running transaction");
Dave Kleikamp470decc2006-10-11 01:20:57 -07002127 may_free = __dispose_buffer(jh, transaction);
2128 }
2129
2130zap_buffer:
Jan Karab794e7a2012-09-26 23:11:13 -04002131 /*
2132 * This is tricky. Although the buffer is truncated, it may be reused
2133 * if blocksize < pagesize and it is attached to the page straddling
2134 * EOF. Since the buffer might have been added to BJ_Forget list of the
2135 * running transaction, journal_get_write_access() won't clear
2136 * b_modified and credit accounting gets confused. So clear b_modified
2137 * here.
2138 */
2139 jh->b_modified = 0;
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002140 jbd2_journal_put_journal_head(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002141zap_buffer_no_jh:
2142 spin_unlock(&journal->j_list_lock);
2143 jbd_unlock_bh_state(bh);
Theodore Ts'oa931da62010-08-03 21:35:12 -04002144 write_unlock(&journal->j_state_lock);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002145zap_buffer_unlocked:
2146 clear_buffer_dirty(bh);
2147 J_ASSERT_BH(bh, !buffer_jbddirty(bh));
2148 clear_buffer_mapped(bh);
2149 clear_buffer_req(bh);
2150 clear_buffer_new(bh);
Eric Sandeen15291162012-02-20 17:53:01 -05002151 clear_buffer_delay(bh);
2152 clear_buffer_unwritten(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002153 bh->b_bdev = NULL;
2154 return may_free;
2155}
2156
2157/**
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002158 * void jbd2_journal_invalidatepage()
Dave Kleikamp470decc2006-10-11 01:20:57 -07002159 * @journal: journal to use for flush...
2160 * @page: page to flush
Lukas Czerner259709b2013-05-21 23:20:03 -04002161 * @offset: start of the range to invalidate
2162 * @length: length of the range to invalidate
Dave Kleikamp470decc2006-10-11 01:20:57 -07002163 *
Lukas Czerner259709b2013-05-21 23:20:03 -04002164 * Reap page buffers containing data after in the specified range in page.
2165 * Can return -EBUSY if buffers are part of the committing transaction and
2166 * the page is straddling i_size. Caller then has to wait for current commit
2167 * and try again.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002168 */
Jan Kara53e87262012-12-25 13:29:52 -05002169int jbd2_journal_invalidatepage(journal_t *journal,
2170 struct page *page,
Lukas Czerner259709b2013-05-21 23:20:03 -04002171 unsigned int offset,
2172 unsigned int length)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002173{
2174 struct buffer_head *head, *bh, *next;
Lukas Czerner259709b2013-05-21 23:20:03 -04002175 unsigned int stop = offset + length;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002176 unsigned int curr_off = 0;
Lukas Czerner259709b2013-05-21 23:20:03 -04002177 int partial_page = (offset || length < PAGE_CACHE_SIZE);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002178 int may_free = 1;
Jan Kara53e87262012-12-25 13:29:52 -05002179 int ret = 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002180
2181 if (!PageLocked(page))
2182 BUG();
2183 if (!page_has_buffers(page))
Jan Kara53e87262012-12-25 13:29:52 -05002184 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002185
Lukas Czerner259709b2013-05-21 23:20:03 -04002186 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
2187
Dave Kleikamp470decc2006-10-11 01:20:57 -07002188 /* We will potentially be playing with lists other than just the
2189 * data lists (especially for journaled data mode), so be
2190 * cautious in our locking. */
2191
2192 head = bh = page_buffers(page);
2193 do {
2194 unsigned int next_off = curr_off + bh->b_size;
2195 next = bh->b_this_page;
2196
Lukas Czerner259709b2013-05-21 23:20:03 -04002197 if (next_off > stop)
2198 return 0;
2199
Dave Kleikamp470decc2006-10-11 01:20:57 -07002200 if (offset <= curr_off) {
2201 /* This block is wholly outside the truncation point */
2202 lock_buffer(bh);
Lukas Czerner259709b2013-05-21 23:20:03 -04002203 ret = journal_unmap_buffer(journal, bh, partial_page);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002204 unlock_buffer(bh);
Jan Kara53e87262012-12-25 13:29:52 -05002205 if (ret < 0)
2206 return ret;
2207 may_free &= ret;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002208 }
2209 curr_off = next_off;
2210 bh = next;
2211
2212 } while (bh != head);
2213
Lukas Czerner259709b2013-05-21 23:20:03 -04002214 if (!partial_page) {
Dave Kleikamp470decc2006-10-11 01:20:57 -07002215 if (may_free && try_to_free_buffers(page))
2216 J_ASSERT(!page_has_buffers(page));
2217 }
Jan Kara53e87262012-12-25 13:29:52 -05002218 return 0;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002219}
2220
2221/*
2222 * File a buffer on the given transaction list.
2223 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002224void __jbd2_journal_file_buffer(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002225 transaction_t *transaction, int jlist)
2226{
2227 struct journal_head **list = NULL;
2228 int was_dirty = 0;
2229 struct buffer_head *bh = jh2bh(jh);
2230
2231 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2232 assert_spin_locked(&transaction->t_journal->j_list_lock);
2233
2234 J_ASSERT_JH(jh, jh->b_jlist < BJ_Types);
2235 J_ASSERT_JH(jh, jh->b_transaction == transaction ||
Mingming Cao40191912008-01-28 23:58:27 -05002236 jh->b_transaction == NULL);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002237
2238 if (jh->b_transaction && jh->b_jlist == jlist)
2239 return;
2240
Dave Kleikamp470decc2006-10-11 01:20:57 -07002241 if (jlist == BJ_Metadata || jlist == BJ_Reserved ||
2242 jlist == BJ_Shadow || jlist == BJ_Forget) {
Jan Karaf91d1d02009-07-13 16:16:20 -04002243 /*
2244 * For metadata buffers, we track dirty bit in buffer_jbddirty
2245 * instead of buffer_dirty. We should not see a dirty bit set
2246 * here because we clear it in do_get_write_access but e.g.
2247 * tune2fs can modify the sb and set the dirty bit at any time
2248 * so we try to gracefully handle that.
2249 */
2250 if (buffer_dirty(bh))
2251 warn_dirty_buffer(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002252 if (test_clear_buffer_dirty(bh) ||
2253 test_clear_buffer_jbddirty(bh))
2254 was_dirty = 1;
2255 }
2256
2257 if (jh->b_transaction)
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002258 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karade1b7942011-06-13 15:38:22 -04002259 else
2260 jbd2_journal_grab_journal_head(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002261 jh->b_transaction = transaction;
2262
2263 switch (jlist) {
2264 case BJ_None:
2265 J_ASSERT_JH(jh, !jh->b_committed_data);
2266 J_ASSERT_JH(jh, !jh->b_frozen_data);
2267 return;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002268 case BJ_Metadata:
2269 transaction->t_nr_buffers++;
2270 list = &transaction->t_buffers;
2271 break;
2272 case BJ_Forget:
2273 list = &transaction->t_forget;
2274 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002275 case BJ_Shadow:
2276 list = &transaction->t_shadow_list;
2277 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002278 case BJ_Reserved:
2279 list = &transaction->t_reserved_list;
2280 break;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002281 }
2282
2283 __blist_add_buffer(list, jh);
2284 jh->b_jlist = jlist;
2285
2286 if (was_dirty)
2287 set_buffer_jbddirty(bh);
2288}
2289
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002290void jbd2_journal_file_buffer(struct journal_head *jh,
Dave Kleikamp470decc2006-10-11 01:20:57 -07002291 transaction_t *transaction, int jlist)
2292{
2293 jbd_lock_bh_state(jh2bh(jh));
2294 spin_lock(&transaction->t_journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002295 __jbd2_journal_file_buffer(jh, transaction, jlist);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002296 spin_unlock(&transaction->t_journal->j_list_lock);
2297 jbd_unlock_bh_state(jh2bh(jh));
2298}
2299
2300/*
2301 * Remove a buffer from its current buffer list in preparation for
2302 * dropping it from its current transaction entirely. If the buffer has
2303 * already started to be used by a subsequent transaction, refile the
2304 * buffer on that transaction's metadata list.
2305 *
Jan Karade1b7942011-06-13 15:38:22 -04002306 * Called under j_list_lock
Dave Kleikamp470decc2006-10-11 01:20:57 -07002307 * Called under jbd_lock_bh_state(jh2bh(jh))
Jan Karade1b7942011-06-13 15:38:22 -04002308 *
2309 * jh and bh may be already free when this function returns
Dave Kleikamp470decc2006-10-11 01:20:57 -07002310 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002311void __jbd2_journal_refile_buffer(struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002312{
dingdinghuaba869022010-02-15 16:35:42 -05002313 int was_dirty, jlist;
Dave Kleikamp470decc2006-10-11 01:20:57 -07002314 struct buffer_head *bh = jh2bh(jh);
2315
2316 J_ASSERT_JH(jh, jbd_is_locked_bh_state(bh));
2317 if (jh->b_transaction)
2318 assert_spin_locked(&jh->b_transaction->t_journal->j_list_lock);
2319
2320 /* If the buffer is now unused, just drop it. */
2321 if (jh->b_next_transaction == NULL) {
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002322 __jbd2_journal_unfile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002323 return;
2324 }
2325
2326 /*
2327 * It has been modified by a later transaction: add it to the new
2328 * transaction's metadata list.
2329 */
2330
2331 was_dirty = test_clear_buffer_jbddirty(bh);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002332 __jbd2_journal_temp_unlink_buffer(jh);
Jan Karade1b7942011-06-13 15:38:22 -04002333 /*
2334 * We set b_transaction here because b_next_transaction will inherit
2335 * our jh reference and thus __jbd2_journal_file_buffer() must not
2336 * take a new one.
2337 */
Dave Kleikamp470decc2006-10-11 01:20:57 -07002338 jh->b_transaction = jh->b_next_transaction;
2339 jh->b_next_transaction = NULL;
dingdinghuaba869022010-02-15 16:35:42 -05002340 if (buffer_freed(bh))
2341 jlist = BJ_Forget;
2342 else if (jh->b_modified)
2343 jlist = BJ_Metadata;
2344 else
2345 jlist = BJ_Reserved;
2346 __jbd2_journal_file_buffer(jh, jh->b_transaction, jlist);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002347 J_ASSERT_JH(jh, jh->b_transaction->t_state == T_RUNNING);
2348
2349 if (was_dirty)
2350 set_buffer_jbddirty(bh);
2351}
2352
2353/*
Jan Karade1b7942011-06-13 15:38:22 -04002354 * __jbd2_journal_refile_buffer() with necessary locking added. We take our
2355 * bh reference so that we can safely unlock bh.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002356 *
Jan Karade1b7942011-06-13 15:38:22 -04002357 * The jh and bh may be freed by this call.
Dave Kleikamp470decc2006-10-11 01:20:57 -07002358 */
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002359void jbd2_journal_refile_buffer(journal_t *journal, struct journal_head *jh)
Dave Kleikamp470decc2006-10-11 01:20:57 -07002360{
2361 struct buffer_head *bh = jh2bh(jh);
2362
Jan Karade1b7942011-06-13 15:38:22 -04002363 /* Get reference so that buffer cannot be freed before we unlock it */
2364 get_bh(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002365 jbd_lock_bh_state(bh);
2366 spin_lock(&journal->j_list_lock);
Mingming Caof7f4bcc2006-10-11 01:20:59 -07002367 __jbd2_journal_refile_buffer(jh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002368 jbd_unlock_bh_state(bh);
Dave Kleikamp470decc2006-10-11 01:20:57 -07002369 spin_unlock(&journal->j_list_lock);
2370 __brelse(bh);
2371}
Jan Karac851ed52008-07-11 19:27:31 -04002372
2373/*
2374 * File inode in the inode list of the handle's transaction
2375 */
2376int jbd2_journal_file_inode(handle_t *handle, struct jbd2_inode *jinode)
2377{
2378 transaction_t *transaction = handle->h_transaction;
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002379 journal_t *journal;
Jan Karac851ed52008-07-11 19:27:31 -04002380
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002381 WARN_ON(!transaction);
Jan Karac851ed52008-07-11 19:27:31 -04002382 if (is_handle_aborted(handle))
Theodore Ts'o41a5b912013-07-01 08:12:41 -04002383 return -EROFS;
2384 journal = transaction->t_journal;
Jan Karac851ed52008-07-11 19:27:31 -04002385
2386 jbd_debug(4, "Adding inode %lu, tid:%d\n", jinode->i_vfs_inode->i_ino,
2387 transaction->t_tid);
2388
2389 /*
2390 * First check whether inode isn't already on the transaction's
2391 * lists without taking the lock. Note that this check is safe
2392 * without the lock as we cannot race with somebody removing inode
2393 * from the transaction. The reason is that we remove inode from the
2394 * transaction only in journal_release_jbd_inode() and when we commit
2395 * the transaction. We are guarded from the first case by holding
2396 * a reference to the inode. We are safe against the second case
2397 * because if jinode->i_transaction == transaction, commit code
2398 * cannot touch the transaction because we hold reference to it,
2399 * and if jinode->i_next_transaction == transaction, commit code
2400 * will only file the inode where we want it.
2401 */
2402 if (jinode->i_transaction == transaction ||
2403 jinode->i_next_transaction == transaction)
2404 return 0;
2405
2406 spin_lock(&journal->j_list_lock);
2407
2408 if (jinode->i_transaction == transaction ||
2409 jinode->i_next_transaction == transaction)
2410 goto done;
2411
Jan Kara81be12c2011-05-24 11:52:40 -04002412 /*
2413 * We only ever set this variable to 1 so the test is safe. Since
2414 * t_need_data_flush is likely to be set, we do the test to save some
2415 * cacheline bouncing
2416 */
2417 if (!transaction->t_need_data_flush)
2418 transaction->t_need_data_flush = 1;
Jan Karac851ed52008-07-11 19:27:31 -04002419 /* On some different transaction's list - should be
2420 * the committing one */
2421 if (jinode->i_transaction) {
2422 J_ASSERT(jinode->i_next_transaction == NULL);
2423 J_ASSERT(jinode->i_transaction ==
2424 journal->j_committing_transaction);
2425 jinode->i_next_transaction = transaction;
2426 goto done;
2427 }
2428 /* Not on any transaction list... */
2429 J_ASSERT(!jinode->i_next_transaction);
2430 jinode->i_transaction = transaction;
2431 list_add(&jinode->i_list, &transaction->t_inode_list);
2432done:
2433 spin_unlock(&journal->j_list_lock);
2434
2435 return 0;
2436}
2437
2438/*
Jan Kara7f5aa212009-02-10 11:15:34 -05002439 * File truncate and transaction commit interact with each other in a
2440 * non-trivial way. If a transaction writing data block A is
2441 * committing, we cannot discard the data by truncate until we have
2442 * written them. Otherwise if we crashed after the transaction with
2443 * write has committed but before the transaction with truncate has
2444 * committed, we could see stale data in block A. This function is a
2445 * helper to solve this problem. It starts writeout of the truncated
2446 * part in case it is in the committing transaction.
2447 *
2448 * Filesystem code must call this function when inode is journaled in
2449 * ordered mode before truncation happens and after the inode has been
2450 * placed on orphan list with the new inode size. The second condition
2451 * avoids the race that someone writes new data and we start
2452 * committing the transaction after this function has been called but
2453 * before a transaction for truncate is started (and furthermore it
2454 * allows us to optimize the case where the addition to orphan list
2455 * happens in the same transaction as write --- we don't have to write
2456 * any data in such case).
Jan Karac851ed52008-07-11 19:27:31 -04002457 */
Jan Kara7f5aa212009-02-10 11:15:34 -05002458int jbd2_journal_begin_ordered_truncate(journal_t *journal,
2459 struct jbd2_inode *jinode,
Jan Karac851ed52008-07-11 19:27:31 -04002460 loff_t new_size)
2461{
Jan Kara7f5aa212009-02-10 11:15:34 -05002462 transaction_t *inode_trans, *commit_trans;
Jan Karac851ed52008-07-11 19:27:31 -04002463 int ret = 0;
2464
Jan Kara7f5aa212009-02-10 11:15:34 -05002465 /* This is a quick check to avoid locking if not necessary */
2466 if (!jinode->i_transaction)
Jan Karac851ed52008-07-11 19:27:31 -04002467 goto out;
Jan Kara7f5aa212009-02-10 11:15:34 -05002468 /* Locks are here just to force reading of recent values, it is
2469 * enough that the transaction was not committing before we started
2470 * a transaction adding the inode to orphan list */
Theodore Ts'oa931da62010-08-03 21:35:12 -04002471 read_lock(&journal->j_state_lock);
Jan Karac851ed52008-07-11 19:27:31 -04002472 commit_trans = journal->j_committing_transaction;
Theodore Ts'oa931da62010-08-03 21:35:12 -04002473 read_unlock(&journal->j_state_lock);
Jan Kara7f5aa212009-02-10 11:15:34 -05002474 spin_lock(&journal->j_list_lock);
2475 inode_trans = jinode->i_transaction;
2476 spin_unlock(&journal->j_list_lock);
2477 if (inode_trans == commit_trans) {
2478 ret = filemap_fdatawrite_range(jinode->i_vfs_inode->i_mapping,
Jan Karac851ed52008-07-11 19:27:31 -04002479 new_size, LLONG_MAX);
2480 if (ret)
2481 jbd2_journal_abort(journal, ret);
2482 }
2483out:
2484 return ret;
2485}