ext4: let ext4_group_add_blocks() return an error code
[linux-2.6.git] / fs / ext4 / file.c
1 /*
2  *  linux/fs/ext4/file.c
3  *
4  * Copyright (C) 1992, 1993, 1994, 1995
5  * Remy Card (card@masi.ibp.fr)
6  * Laboratoire MASI - Institut Blaise Pascal
7  * Universite Pierre et Marie Curie (Paris VI)
8  *
9  *  from
10  *
11  *  linux/fs/minix/file.c
12  *
13  *  Copyright (C) 1991, 1992  Linus Torvalds
14  *
15  *  ext4 fs regular file handling primitives
16  *
17  *  64-bit file support on 64-bit platforms by Jakub Jelinek
18  *      (jj@sunsite.ms.mff.cuni.cz)
19  */
20
21 #include <linux/time.h>
22 #include <linux/fs.h>
23 #include <linux/jbd2.h>
24 #include <linux/mount.h>
25 #include <linux/path.h>
26 #include <linux/quotaops.h>
27 #include "ext4.h"
28 #include "ext4_jbd2.h"
29 #include "xattr.h"
30 #include "acl.h"
31
32 /*
33  * Called when an inode is released. Note that this is different
34  * from ext4_file_open: open gets called at every open, but release
35  * gets called only when /all/ the files are closed.
36  */
37 static int ext4_release_file(struct inode *inode, struct file *filp)
38 {
39         if (ext4_test_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE)) {
40                 ext4_alloc_da_blocks(inode);
41                 ext4_clear_inode_state(inode, EXT4_STATE_DA_ALLOC_CLOSE);
42         }
43         /* if we are the last writer on the inode, drop the block reservation */
44         if ((filp->f_mode & FMODE_WRITE) &&
45                         (atomic_read(&inode->i_writecount) == 1) &&
46                         !EXT4_I(inode)->i_reserved_data_blocks)
47         {
48                 down_write(&EXT4_I(inode)->i_data_sem);
49                 ext4_discard_preallocations(inode);
50                 up_write(&EXT4_I(inode)->i_data_sem);
51         }
52         if (is_dx(inode) && filp->private_data)
53                 ext4_htree_free_dir_info(filp->private_data);
54
55         return 0;
56 }
57
58 static void ext4_aiodio_wait(struct inode *inode)
59 {
60         wait_queue_head_t *wq = ext4_ioend_wq(inode);
61
62         wait_event(*wq, (atomic_read(&EXT4_I(inode)->i_aiodio_unwritten) == 0));
63 }
64
65 /*
66  * This tests whether the IO in question is block-aligned or not.
67  * Ext4 utilizes unwritten extents when hole-filling during direct IO, and they
68  * are converted to written only after the IO is complete.  Until they are
69  * mapped, these blocks appear as holes, so dio_zero_block() will assume that
70  * it needs to zero out portions of the start and/or end block.  If 2 AIO
71  * threads are at work on the same unwritten block, they must be synchronized
72  * or one thread will zero the other's data, causing corruption.
73  */
74 static int
75 ext4_unaligned_aio(struct inode *inode, const struct iovec *iov,
76                    unsigned long nr_segs, loff_t pos)
77 {
78         struct super_block *sb = inode->i_sb;
79         int blockmask = sb->s_blocksize - 1;
80         size_t count = iov_length(iov, nr_segs);
81         loff_t final_size = pos + count;
82
83         if (pos >= inode->i_size)
84                 return 0;
85
86         if ((pos & blockmask) || (final_size & blockmask))
87                 return 1;
88
89         return 0;
90 }
91
92 static ssize_t
93 ext4_file_write(struct kiocb *iocb, const struct iovec *iov,
94                 unsigned long nr_segs, loff_t pos)
95 {
96         struct inode *inode = iocb->ki_filp->f_path.dentry->d_inode;
97         int unaligned_aio = 0;
98         int ret;
99
100         /*
101          * If we have encountered a bitmap-format file, the size limit
102          * is smaller than s_maxbytes, which is for extent-mapped files.
103          */
104
105         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS))) {
106                 struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
107                 size_t length = iov_length(iov, nr_segs);
108
109                 if ((pos > sbi->s_bitmap_maxbytes ||
110                     (pos == sbi->s_bitmap_maxbytes && length > 0)))
111                         return -EFBIG;
112
113                 if (pos + length > sbi->s_bitmap_maxbytes) {
114                         nr_segs = iov_shorten((struct iovec *)iov, nr_segs,
115                                               sbi->s_bitmap_maxbytes - pos);
116                 }
117         } else if (unlikely((iocb->ki_filp->f_flags & O_DIRECT) &&
118                    !is_sync_kiocb(iocb))) {
119                 unaligned_aio = ext4_unaligned_aio(inode, iov, nr_segs, pos);
120         }
121
122         /* Unaligned direct AIO must be serialized; see comment above */
123         if (unaligned_aio) {
124                 static unsigned long unaligned_warn_time;
125
126                 /* Warn about this once per day */
127                 if (printk_timed_ratelimit(&unaligned_warn_time, 60*60*24*HZ))
128                         ext4_msg(inode->i_sb, KERN_WARNING,
129                                  "Unaligned AIO/DIO on inode %ld by %s; "
130                                  "performance will be poor.",
131                                  inode->i_ino, current->comm);
132                 mutex_lock(ext4_aio_mutex(inode));
133                 ext4_aiodio_wait(inode);
134         }
135
136         ret = generic_file_aio_write(iocb, iov, nr_segs, pos);
137
138         if (unaligned_aio)
139                 mutex_unlock(ext4_aio_mutex(inode));
140
141         return ret;
142 }
143
144 static const struct vm_operations_struct ext4_file_vm_ops = {
145         .fault          = filemap_fault,
146         .page_mkwrite   = ext4_page_mkwrite,
147 };
148
149 static int ext4_file_mmap(struct file *file, struct vm_area_struct *vma)
150 {
151         struct address_space *mapping = file->f_mapping;
152
153         if (!mapping->a_ops->readpage)
154                 return -ENOEXEC;
155         file_accessed(file);
156         vma->vm_ops = &ext4_file_vm_ops;
157         vma->vm_flags |= VM_CAN_NONLINEAR;
158         return 0;
159 }
160
161 static int ext4_file_open(struct inode * inode, struct file * filp)
162 {
163         struct super_block *sb = inode->i_sb;
164         struct ext4_sb_info *sbi = EXT4_SB(inode->i_sb);
165         struct ext4_inode_info *ei = EXT4_I(inode);
166         struct vfsmount *mnt = filp->f_path.mnt;
167         struct path path;
168         char buf[64], *cp;
169
170         if (unlikely(!(sbi->s_mount_flags & EXT4_MF_MNTDIR_SAMPLED) &&
171                      !(sb->s_flags & MS_RDONLY))) {
172                 sbi->s_mount_flags |= EXT4_MF_MNTDIR_SAMPLED;
173                 /*
174                  * Sample where the filesystem has been mounted and
175                  * store it in the superblock for sysadmin convenience
176                  * when trying to sort through large numbers of block
177                  * devices or filesystem images.
178                  */
179                 memset(buf, 0, sizeof(buf));
180                 path.mnt = mnt;
181                 path.dentry = mnt->mnt_root;
182                 cp = d_path(&path, buf, sizeof(buf));
183                 if (!IS_ERR(cp)) {
184                         memcpy(sbi->s_es->s_last_mounted, cp,
185                                sizeof(sbi->s_es->s_last_mounted));
186                         ext4_mark_super_dirty(sb);
187                 }
188         }
189         /*
190          * Set up the jbd2_inode if we are opening the inode for
191          * writing and the journal is present
192          */
193         if (sbi->s_journal && !ei->jinode && (filp->f_mode & FMODE_WRITE)) {
194                 struct jbd2_inode *jinode = jbd2_alloc_inode(GFP_KERNEL);
195
196                 spin_lock(&inode->i_lock);
197                 if (!ei->jinode) {
198                         if (!jinode) {
199                                 spin_unlock(&inode->i_lock);
200                                 return -ENOMEM;
201                         }
202                         ei->jinode = jinode;
203                         jbd2_journal_init_jbd_inode(ei->jinode, inode);
204                         jinode = NULL;
205                 }
206                 spin_unlock(&inode->i_lock);
207                 if (unlikely(jinode != NULL))
208                         jbd2_free_inode(jinode);
209         }
210         return dquot_file_open(inode, filp);
211 }
212
213 /*
214  * ext4_llseek() copied from generic_file_llseek() to handle both
215  * block-mapped and extent-mapped maxbytes values. This should
216  * otherwise be identical with generic_file_llseek().
217  */
218 loff_t ext4_llseek(struct file *file, loff_t offset, int origin)
219 {
220         struct inode *inode = file->f_mapping->host;
221         loff_t maxbytes;
222
223         if (!(ext4_test_inode_flag(inode, EXT4_INODE_EXTENTS)))
224                 maxbytes = EXT4_SB(inode->i_sb)->s_bitmap_maxbytes;
225         else
226                 maxbytes = inode->i_sb->s_maxbytes;
227         mutex_lock(&inode->i_mutex);
228         switch (origin) {
229         case SEEK_END:
230                 offset += inode->i_size;
231                 break;
232         case SEEK_CUR:
233                 if (offset == 0) {
234                         mutex_unlock(&inode->i_mutex);
235                         return file->f_pos;
236                 }
237                 offset += file->f_pos;
238                 break;
239         }
240
241         if (offset < 0 || offset > maxbytes) {
242                 mutex_unlock(&inode->i_mutex);
243                 return -EINVAL;
244         }
245
246         if (offset != file->f_pos) {
247                 file->f_pos = offset;
248                 file->f_version = 0;
249         }
250         mutex_unlock(&inode->i_mutex);
251
252         return offset;
253 }
254
255 const struct file_operations ext4_file_operations = {
256         .llseek         = ext4_llseek,
257         .read           = do_sync_read,
258         .write          = do_sync_write,
259         .aio_read       = generic_file_aio_read,
260         .aio_write      = ext4_file_write,
261         .unlocked_ioctl = ext4_ioctl,
262 #ifdef CONFIG_COMPAT
263         .compat_ioctl   = ext4_compat_ioctl,
264 #endif
265         .mmap           = ext4_file_mmap,
266         .open           = ext4_file_open,
267         .release        = ext4_release_file,
268         .fsync          = ext4_sync_file,
269         .splice_read    = generic_file_splice_read,
270         .splice_write   = generic_file_splice_write,
271         .fallocate      = ext4_fallocate,
272 };
273
274 const struct inode_operations ext4_file_inode_operations = {
275         .setattr        = ext4_setattr,
276         .getattr        = ext4_getattr,
277 #ifdef CONFIG_EXT4_FS_XATTR
278         .setxattr       = generic_setxattr,
279         .getxattr       = generic_getxattr,
280         .listxattr      = ext4_listxattr,
281         .removexattr    = generic_removexattr,
282 #endif
283         .check_acl      = ext4_check_acl,
284         .fiemap         = ext4_fiemap,
285 };
286