blob: da3bed3e0c29d1159f2169f4b94be97ce485d966 [file] [log] [blame]
Dave Kleikampac27a0e2006-10-11 01:20:50 -07001/*
Mingming Cao617ba132006-10-11 01:20:53 -07002 * linux/fs/ext4/fsync.c
Dave Kleikampac27a0e2006-10-11 01:20:50 -07003 *
4 * Copyright (C) 1993 Stephen Tweedie (sct@redhat.com)
5 * from
6 * Copyright (C) 1992 Remy Card (card@masi.ibp.fr)
7 * Laboratoire MASI - Institut Blaise Pascal
8 * Universite Pierre et Marie Curie (Paris VI)
9 * from
10 * linux/fs/minix/truncate.c Copyright (C) 1991, 1992 Linus Torvalds
11 *
Mingming Cao617ba132006-10-11 01:20:53 -070012 * ext4fs fsync primitive
Dave Kleikampac27a0e2006-10-11 01:20:50 -070013 *
14 * Big-endian to little-endian byte-swapping/bitmaps by
15 * David S. Miller (davem@caip.rutgers.edu), 1995
16 *
17 * Removed unnecessary code duplication for little endian machines
18 * and excessive __inline__s.
19 * Andi Kleen, 1997
20 *
21 * Major simplications and cleanup - we only need to do the metadata, because
22 * we can depend on generic_block_fdatasync() to sync the data blocks.
23 */
24
25#include <linux/time.h>
26#include <linux/fs.h>
27#include <linux/sched.h>
28#include <linux/writeback.h>
Mingming Caodab291a2006-10-11 01:21:01 -070029#include <linux/jbd2.h>
Eric Sandeend755fb32008-07-11 19:27:31 -040030#include <linux/blkdev.h>
Theodore Ts'o9bffad12009-06-17 11:48:11 -040031
Christoph Hellwig3dcf5452008-04-29 18:13:32 -040032#include "ext4.h"
33#include "ext4_jbd2.h"
Dave Kleikampac27a0e2006-10-11 01:20:50 -070034
Theodore Ts'o9bffad12009-06-17 11:48:11 -040035#include <trace/events/ext4.h>
36
Theodore Ts'o4a873a42010-10-27 21:30:14 -040037static void dump_completed_IO(struct inode * inode)
38{
Tao Mae8bbe8c2011-05-09 10:25:54 -040039#ifdef EXT4FS_DEBUG
Theodore Ts'o4a873a42010-10-27 21:30:14 -040040 struct list_head *cur, *before, *after;
41 ext4_io_end_t *io, *io0, *io1;
42 unsigned long flags;
43
44 if (list_empty(&EXT4_I(inode)->i_completed_io_list)){
45 ext4_debug("inode %lu completed_io list is empty\n", inode->i_ino);
46 return;
47 }
48
49 ext4_debug("Dump inode %lu completed_io list \n", inode->i_ino);
50 spin_lock_irqsave(&EXT4_I(inode)->i_completed_io_lock, flags);
51 list_for_each_entry(io, &EXT4_I(inode)->i_completed_io_list, list){
52 cur = &io->list;
53 before = cur->prev;
54 io0 = container_of(before, ext4_io_end_t, list);
55 after = cur->next;
56 io1 = container_of(after, ext4_io_end_t, list);
57
58 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
59 io, inode->i_ino, io0, io1);
60 }
61 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
62#endif
63}
64
65/*
66 * This function is called from ext4_sync_file().
67 *
68 * When IO is completed, the work to convert unwritten extents to
69 * written is queued on workqueue but may not get immediately
70 * scheduled. When fsync is called, we need to ensure the
71 * conversion is complete before fsync returns.
72 * The inode keeps track of a list of pending/completed IO that
73 * might needs to do the conversion. This function walks through
74 * the list and convert the related unwritten extents for completed IO
75 * to written.
76 * The function return the number of pending IOs on success.
77 */
Jiaying Zhang3889fd52011-01-10 12:47:05 -050078extern int ext4_flush_completed_IO(struct inode *inode)
Theodore Ts'o4a873a42010-10-27 21:30:14 -040079{
80 ext4_io_end_t *io;
81 struct ext4_inode_info *ei = EXT4_I(inode);
82 unsigned long flags;
83 int ret = 0;
84 int ret2 = 0;
85
86 if (list_empty(&ei->i_completed_io_list))
87 return ret;
88
89 dump_completed_IO(inode);
90 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
91 while (!list_empty(&ei->i_completed_io_list)){
92 io = list_entry(ei->i_completed_io_list.next,
93 ext4_io_end_t, list);
94 /*
95 * Calling ext4_end_io_nolock() to convert completed
96 * IO to written.
97 *
98 * When ext4_sync_file() is called, run_queue() may already
99 * about to flush the work corresponding to this io structure.
100 * It will be upset if it founds the io structure related
101 * to the work-to-be schedule is freed.
102 *
103 * Thus we need to keep the io structure still valid here after
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300104 * conversion finished. The io structure has a flag to
Theodore Ts'o4a873a42010-10-27 21:30:14 -0400105 * avoid double converting from both fsync and background work
106 * queue work.
107 */
108 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
109 ret = ext4_end_io_nolock(io);
110 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
111 if (ret < 0)
112 ret2 = ret;
113 else
114 list_del_init(&io->list);
115 }
116 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
117 return (ret2 < 0) ? ret2 : 0;
118}
119
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700120/*
Frank Mayhar14ece102010-05-17 08:00:00 -0400121 * If we're not journaling and this is a just-created file, we have to
122 * sync our parent directory (if it was freshly created) since
123 * otherwise it will only be written by writeback, leaving a huge
124 * window during which a crash may lose the file. This may apply for
125 * the parent directory's parent as well, and so on recursively, if
126 * they are also freshly created.
127 */
Curt Wohlgemuth0893ed42011-04-10 22:05:31 -0400128static int ext4_sync_parent(struct inode *inode)
Frank Mayhar14ece102010-05-17 08:00:00 -0400129{
Curt Wohlgemuth0893ed42011-04-10 22:05:31 -0400130 struct writeback_control wbc;
Frank Mayhar14ece102010-05-17 08:00:00 -0400131 struct dentry *dentry = NULL;
Curt Wohlgemuth0893ed42011-04-10 22:05:31 -0400132 int ret = 0;
Frank Mayhar14ece102010-05-17 08:00:00 -0400133
134 while (inode && ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
135 ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
136 dentry = list_entry(inode->i_dentry.next,
137 struct dentry, d_alias);
138 if (!dentry || !dentry->d_parent || !dentry->d_parent->d_inode)
139 break;
140 inode = dentry->d_parent->d_inode;
Curt Wohlgemuth0893ed42011-04-10 22:05:31 -0400141 ret = sync_mapping_buffers(inode->i_mapping);
142 if (ret)
143 break;
144 memset(&wbc, 0, sizeof(wbc));
145 wbc.sync_mode = WB_SYNC_ALL;
146 wbc.nr_to_write = 0; /* only write out the inode */
147 ret = sync_inode(inode, &wbc);
148 if (ret)
149 break;
Frank Mayhar14ece102010-05-17 08:00:00 -0400150 }
Curt Wohlgemuth0893ed42011-04-10 22:05:31 -0400151 return ret;
Frank Mayhar14ece102010-05-17 08:00:00 -0400152}
153
Josef Bacik02c24a82011-07-16 20:44:56 -0400154/**
155 * __sync_file - generic_file_fsync without the locking and filemap_write
156 * @inode: inode to sync
157 * @datasync: only sync essential metadata if true
158 *
159 * This is just generic_file_fsync without the locking. This is needed for
160 * nojournal mode to make sure this inodes data/metadata makes it to disk
161 * properly. The i_mutex should be held already.
162 */
163static int __sync_inode(struct inode *inode, int datasync)
164{
165 int err;
166 int ret;
167
168 ret = sync_mapping_buffers(inode->i_mapping);
169 if (!(inode->i_state & I_DIRTY))
170 return ret;
171 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
172 return ret;
173
174 err = sync_inode_metadata(inode, 1);
175 if (ret == 0)
176 ret = err;
177 return ret;
178}
179
Frank Mayhar14ece102010-05-17 08:00:00 -0400180/*
Mingming Cao617ba132006-10-11 01:20:53 -0700181 * akpm: A new design for ext4_sync_file().
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700182 *
183 * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
184 * There cannot be a transaction open by this task.
185 * Another task could have dirtied this inode. Its data can be in any
186 * state in the journalling system.
187 *
188 * What we do is just kick off a commit and wait on it. This will snapshot the
189 * inode to disk.
Mingming Cao8d5d02e2009-09-28 15:48:29 -0400190 *
191 * i_mutex lock is held when entering and exiting this function
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700192 */
193
Josef Bacik02c24a82011-07-16 20:44:56 -0400194int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700195{
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200196 struct inode *inode = file->f_mapping->host;
Jan Karab436b9b2009-12-08 23:51:10 -0500197 struct ext4_inode_info *ei = EXT4_I(inode);
Eric Sandeend755fb32008-07-11 19:27:31 -0400198 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
Jan Karab436b9b2009-12-08 23:51:10 -0500199 int ret;
200 tid_t commit_tid;
Jan Kara93628ff2011-05-24 12:00:54 -0400201 bool needs_barrier = false;
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700202
Aneesh Kumar K.Vac398492007-10-16 18:38:25 -0400203 J_ASSERT(ext4_journal_current_handle() == NULL);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700204
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400205 trace_ext4_sync_file_enter(file, datasync);
Theodore Ts'oede86cc2008-10-05 20:50:06 -0400206
Josef Bacik02c24a82011-07-16 20:44:56 -0400207 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
208 if (ret)
209 return ret;
210 mutex_lock(&inode->i_mutex);
211
Jan Karab436b9b2009-12-08 23:51:10 -0500212 if (inode->i_sb->s_flags & MS_RDONLY)
Josef Bacik02c24a82011-07-16 20:44:56 -0400213 goto out;
Jan Karab436b9b2009-12-08 23:51:10 -0500214
Jiaying Zhang3889fd52011-01-10 12:47:05 -0500215 ret = ext4_flush_completed_IO(inode);
Mingming Cao8d5d02e2009-09-28 15:48:29 -0400216 if (ret < 0)
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400217 goto out;
Theodore Ts'o60e66792010-05-17 07:00:00 -0400218
Frank Mayhar14ece102010-05-17 08:00:00 -0400219 if (!journal) {
Josef Bacik02c24a82011-07-16 20:44:56 -0400220 ret = __sync_inode(inode, datasync);
Frank Mayhar14ece102010-05-17 08:00:00 -0400221 if (!ret && !list_empty(&inode->i_dentry))
Curt Wohlgemuth0893ed42011-04-10 22:05:31 -0400222 ret = ext4_sync_parent(inode);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400223 goto out;
Frank Mayhar14ece102010-05-17 08:00:00 -0400224 }
Jan Karab436b9b2009-12-08 23:51:10 -0500225
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700226 /*
Jan Karab436b9b2009-12-08 23:51:10 -0500227 * data=writeback,ordered:
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700228 * The caller's filemap_fdatawrite()/wait will sync the data.
Jan Karab436b9b2009-12-08 23:51:10 -0500229 * Metadata is in the journal, we wait for proper transaction to
230 * commit here.
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700231 *
232 * data=journal:
233 * filemap_fdatawrite won't do anything (the buffers are clean).
Mingming Cao617ba132006-10-11 01:20:53 -0700234 * ext4_force_commit will write the file data into the journal and
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700235 * will wait on that.
236 * filemap_fdatawait() will encounter a ton of newly-dirtied pages
237 * (they were dirtied by commit). But that's OK - the blocks are
238 * safe in-journal, which is all fsync() needs to ensure.
239 */
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400240 if (ext4_should_journal_data(inode)) {
241 ret = ext4_force_commit(inode->i_sb);
242 goto out;
243 }
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700244
Jan Karab436b9b2009-12-08 23:51:10 -0500245 commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
Jan Kara93628ff2011-05-24 12:00:54 -0400246 if (journal->j_flags & JBD2_BARRIER &&
247 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
248 needs_barrier = true;
249 jbd2_log_start_commit(journal, commit_tid);
250 ret = jbd2_log_wait_commit(journal, commit_tid);
251 if (needs_barrier)
Christoph Hellwigdd3932e2010-09-16 20:51:46 +0200252 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400253 out:
Josef Bacik02c24a82011-07-16 20:44:56 -0400254 mutex_unlock(&inode->i_mutex);
Jiaying Zhang0562e0b2011-03-21 21:38:05 -0400255 trace_ext4_sync_file_exit(inode, ret);
Dave Kleikampac27a0e2006-10-11 01:20:50 -0700256 return ret;
257}