Btrfs: use larger system chunks
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / ext4 / fsync.c
blob00a2cb753efdeca63d6bbba4912729dc1c854182
1 /*
2 * linux/fs/ext4/fsync.c
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
12 * ext4fs fsync primitive
14 * Big-endian to little-endian byte-swapping/bitmaps by
15 * David S. Miller (davem@caip.rutgers.edu), 1995
17 * Removed unnecessary code duplication for little endian machines
18 * and excessive __inline__s.
19 * Andi Kleen, 1997
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.
25 #include <linux/time.h>
26 #include <linux/fs.h>
27 #include <linux/sched.h>
28 #include <linux/writeback.h>
29 #include <linux/jbd2.h>
30 #include <linux/blkdev.h>
32 #include "ext4.h"
33 #include "ext4_jbd2.h"
35 #include <trace/events/ext4.h>
37 static void dump_completed_IO(struct inode * inode)
39 #ifdef EXT4FS_DEBUG
40 struct list_head *cur, *before, *after;
41 ext4_io_end_t *io, *io0, *io1;
42 unsigned long flags;
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;
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);
58 ext4_debug("io 0x%p from inode %lu,prev 0x%p,next 0x%p\n",
59 io, inode->i_ino, io0, io1);
61 spin_unlock_irqrestore(&EXT4_I(inode)->i_completed_io_lock, flags);
62 #endif
66 * This function is called from ext4_sync_file().
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.
78 int ext4_flush_completed_IO(struct inode *inode)
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;
86 dump_completed_IO(inode);
87 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
88 while (!list_empty(&ei->i_completed_io_list)){
89 io = list_entry(ei->i_completed_io_list.next,
90 ext4_io_end_t, list);
91 list_del_init(&io->list);
93 * Calling ext4_end_io_nolock() to convert completed
94 * IO to written.
96 * When ext4_sync_file() is called, run_queue() may already
97 * about to flush the work corresponding to this io structure.
98 * It will be upset if it founds the io structure related
99 * to the work-to-be schedule is freed.
101 * Thus we need to keep the io structure still valid here after
102 * conversion finished. The io structure has a flag to
103 * avoid double converting from both fsync and background work
104 * queue work.
106 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
107 ret = ext4_end_io_nolock(io);
108 if (ret < 0)
109 ret2 = ret;
110 spin_lock_irqsave(&ei->i_completed_io_lock, flags);
112 spin_unlock_irqrestore(&ei->i_completed_io_lock, flags);
113 return (ret2 < 0) ? ret2 : 0;
117 * If we're not journaling and this is a just-created file, we have to
118 * sync our parent directory (if it was freshly created) since
119 * otherwise it will only be written by writeback, leaving a huge
120 * window during which a crash may lose the file. This may apply for
121 * the parent directory's parent as well, and so on recursively, if
122 * they are also freshly created.
124 static int ext4_sync_parent(struct inode *inode)
126 struct writeback_control wbc;
127 struct dentry *dentry = NULL;
128 struct inode *next;
129 int ret = 0;
131 if (!ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY))
132 return 0;
133 inode = igrab(inode);
134 while (ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
135 ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
136 dentry = NULL;
137 spin_lock(&inode->i_lock);
138 if (!list_empty(&inode->i_dentry)) {
139 dentry = list_first_entry(&inode->i_dentry,
140 struct dentry, d_alias);
141 dget(dentry);
143 spin_unlock(&inode->i_lock);
144 if (!dentry)
145 break;
146 next = igrab(dentry->d_parent->d_inode);
147 dput(dentry);
148 if (!next)
149 break;
150 iput(inode);
151 inode = next;
152 ret = sync_mapping_buffers(inode->i_mapping);
153 if (ret)
154 break;
155 memset(&wbc, 0, sizeof(wbc));
156 wbc.sync_mode = WB_SYNC_ALL;
157 wbc.nr_to_write = 0; /* only write out the inode */
158 ret = sync_inode(inode, &wbc);
159 if (ret)
160 break;
162 iput(inode);
163 return ret;
167 * __sync_file - generic_file_fsync without the locking and filemap_write
168 * @inode: inode to sync
169 * @datasync: only sync essential metadata if true
171 * This is just generic_file_fsync without the locking. This is needed for
172 * nojournal mode to make sure this inodes data/metadata makes it to disk
173 * properly. The i_mutex should be held already.
175 static int __sync_inode(struct inode *inode, int datasync)
177 int err;
178 int ret;
180 ret = sync_mapping_buffers(inode->i_mapping);
181 if (!(inode->i_state & I_DIRTY))
182 return ret;
183 if (datasync && !(inode->i_state & I_DIRTY_DATASYNC))
184 return ret;
186 err = sync_inode_metadata(inode, 1);
187 if (ret == 0)
188 ret = err;
189 return ret;
193 * akpm: A new design for ext4_sync_file().
195 * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
196 * There cannot be a transaction open by this task.
197 * Another task could have dirtied this inode. Its data can be in any
198 * state in the journalling system.
200 * What we do is just kick off a commit and wait on it. This will snapshot the
201 * inode to disk.
203 * i_mutex lock is held when entering and exiting this function
206 int ext4_sync_file(struct file *file, loff_t start, loff_t end, int datasync)
208 struct inode *inode = file->f_mapping->host;
209 struct ext4_inode_info *ei = EXT4_I(inode);
210 journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
211 int ret;
212 tid_t commit_tid;
213 bool needs_barrier = false;
215 J_ASSERT(ext4_journal_current_handle() == NULL);
217 trace_ext4_sync_file_enter(file, datasync);
219 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
220 if (ret)
221 return ret;
222 mutex_lock(&inode->i_mutex);
224 if (inode->i_sb->s_flags & MS_RDONLY)
225 goto out;
227 ret = ext4_flush_completed_IO(inode);
228 if (ret < 0)
229 goto out;
231 if (!journal) {
232 ret = __sync_inode(inode, datasync);
233 if (!ret && !list_empty(&inode->i_dentry))
234 ret = ext4_sync_parent(inode);
235 goto out;
239 * data=writeback,ordered:
240 * The caller's filemap_fdatawrite()/wait will sync the data.
241 * Metadata is in the journal, we wait for proper transaction to
242 * commit here.
244 * data=journal:
245 * filemap_fdatawrite won't do anything (the buffers are clean).
246 * ext4_force_commit will write the file data into the journal and
247 * will wait on that.
248 * filemap_fdatawait() will encounter a ton of newly-dirtied pages
249 * (they were dirtied by commit). But that's OK - the blocks are
250 * safe in-journal, which is all fsync() needs to ensure.
252 if (ext4_should_journal_data(inode)) {
253 ret = ext4_force_commit(inode->i_sb);
254 goto out;
257 commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
258 if (journal->j_flags & JBD2_BARRIER &&
259 !jbd2_trans_will_send_data_barrier(journal, commit_tid))
260 needs_barrier = true;
261 jbd2_log_start_commit(journal, commit_tid);
262 ret = jbd2_log_wait_commit(journal, commit_tid);
263 if (needs_barrier)
264 blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL);
265 out:
266 mutex_unlock(&inode->i_mutex);
267 trace_ext4_sync_file_exit(inode, ret);
268 return ret;