xfs: don't serialise adjacent concurrent direct IO appending writes
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / fs / fhandle.c
blob6b088641f5bf24ce72311a3b8073ce44516b8624
1 #include <linux/syscalls.h>
2 #include <linux/slab.h>
3 #include <linux/fs.h>
4 #include <linux/file.h>
5 #include <linux/mount.h>
6 #include <linux/namei.h>
7 #include <linux/exportfs.h>
8 #include <linux/fs_struct.h>
9 #include <linux/fsnotify.h>
10 #include <linux/personality.h>
11 #include <asm/uaccess.h>
12 #include "internal.h"
14 static long do_sys_name_to_handle(struct path *path,
15 struct file_handle __user *ufh,
16 int __user *mnt_id)
18 long retval;
19 struct file_handle f_handle;
20 int handle_dwords, handle_bytes;
21 struct file_handle *handle = NULL;
24 * We need t make sure wether the file system
25 * support decoding of the file handle
27 if (!path->mnt->mnt_sb->s_export_op ||
28 !path->mnt->mnt_sb->s_export_op->fh_to_dentry)
29 return -EOPNOTSUPP;
31 if (copy_from_user(&f_handle, ufh, sizeof(struct file_handle)))
32 return -EFAULT;
34 if (f_handle.handle_bytes > MAX_HANDLE_SZ)
35 return -EINVAL;
37 handle = kmalloc(sizeof(struct file_handle) + f_handle.handle_bytes,
38 GFP_KERNEL);
39 if (!handle)
40 return -ENOMEM;
42 /* convert handle size to multiple of sizeof(u32) */
43 handle_dwords = f_handle.handle_bytes >> 2;
45 /* we ask for a non connected handle */
46 retval = exportfs_encode_fh(path->dentry,
47 (struct fid *)handle->f_handle,
48 &handle_dwords, 0);
49 handle->handle_type = retval;
50 /* convert handle size to bytes */
51 handle_bytes = handle_dwords * sizeof(u32);
52 handle->handle_bytes = handle_bytes;
53 if ((handle->handle_bytes > f_handle.handle_bytes) ||
54 (retval == 255) || (retval == -ENOSPC)) {
55 /* As per old exportfs_encode_fh documentation
56 * we could return ENOSPC to indicate overflow
57 * But file system returned 255 always. So handle
58 * both the values
61 * set the handle size to zero so we copy only
62 * non variable part of the file_handle
64 handle_bytes = 0;
65 retval = -EOVERFLOW;
66 } else
67 retval = 0;
68 /* copy the mount id */
69 if (copy_to_user(mnt_id, &path->mnt->mnt_id, sizeof(*mnt_id)) ||
70 copy_to_user(ufh, handle,
71 sizeof(struct file_handle) + handle_bytes))
72 retval = -EFAULT;
73 kfree(handle);
74 return retval;
77 /**
78 * sys_name_to_handle_at: convert name to handle
79 * @dfd: directory relative to which name is interpreted if not absolute
80 * @name: name that should be converted to handle.
81 * @handle: resulting file handle
82 * @mnt_id: mount id of the file system containing the file
83 * @flag: flag value to indicate whether to follow symlink or not
85 * @handle->handle_size indicate the space available to store the
86 * variable part of the file handle in bytes. If there is not
87 * enough space, the field is updated to return the minimum
88 * value required.
90 SYSCALL_DEFINE5(name_to_handle_at, int, dfd, const char __user *, name,
91 struct file_handle __user *, handle, int __user *, mnt_id,
92 int, flag)
94 struct path path;
95 int lookup_flags;
96 int err;
98 if ((flag & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
99 return -EINVAL;
101 lookup_flags = (flag & AT_SYMLINK_FOLLOW) ? LOOKUP_FOLLOW : 0;
102 if (flag & AT_EMPTY_PATH)
103 lookup_flags |= LOOKUP_EMPTY;
104 err = user_path_at(dfd, name, lookup_flags, &path);
105 if (!err) {
106 err = do_sys_name_to_handle(&path, handle, mnt_id);
107 path_put(&path);
109 return err;
112 static struct vfsmount *get_vfsmount_from_fd(int fd)
114 struct path path;
116 if (fd == AT_FDCWD) {
117 struct fs_struct *fs = current->fs;
118 spin_lock(&fs->lock);
119 path = fs->pwd;
120 mntget(path.mnt);
121 spin_unlock(&fs->lock);
122 } else {
123 int fput_needed;
124 struct file *file = fget_light(fd, &fput_needed);
125 if (!file)
126 return ERR_PTR(-EBADF);
127 path = file->f_path;
128 mntget(path.mnt);
129 fput_light(file, fput_needed);
131 return path.mnt;
134 static int vfs_dentry_acceptable(void *context, struct dentry *dentry)
136 return 1;
139 static int do_handle_to_path(int mountdirfd, struct file_handle *handle,
140 struct path *path)
142 int retval = 0;
143 int handle_dwords;
145 path->mnt = get_vfsmount_from_fd(mountdirfd);
146 if (IS_ERR(path->mnt)) {
147 retval = PTR_ERR(path->mnt);
148 goto out_err;
150 /* change the handle size to multiple of sizeof(u32) */
151 handle_dwords = handle->handle_bytes >> 2;
152 path->dentry = exportfs_decode_fh(path->mnt,
153 (struct fid *)handle->f_handle,
154 handle_dwords, handle->handle_type,
155 vfs_dentry_acceptable, NULL);
156 if (IS_ERR(path->dentry)) {
157 retval = PTR_ERR(path->dentry);
158 goto out_mnt;
160 return 0;
161 out_mnt:
162 mntput(path->mnt);
163 out_err:
164 return retval;
167 static int handle_to_path(int mountdirfd, struct file_handle __user *ufh,
168 struct path *path)
170 int retval = 0;
171 struct file_handle f_handle;
172 struct file_handle *handle = NULL;
175 * With handle we don't look at the execute bit on the
176 * the directory. Ideally we would like CAP_DAC_SEARCH.
177 * But we don't have that
179 if (!capable(CAP_DAC_READ_SEARCH)) {
180 retval = -EPERM;
181 goto out_err;
183 if (copy_from_user(&f_handle, ufh, sizeof(struct file_handle))) {
184 retval = -EFAULT;
185 goto out_err;
187 if ((f_handle.handle_bytes > MAX_HANDLE_SZ) ||
188 (f_handle.handle_bytes == 0)) {
189 retval = -EINVAL;
190 goto out_err;
192 handle = kmalloc(sizeof(struct file_handle) + f_handle.handle_bytes,
193 GFP_KERNEL);
194 if (!handle) {
195 retval = -ENOMEM;
196 goto out_err;
198 /* copy the full handle */
199 if (copy_from_user(handle, ufh,
200 sizeof(struct file_handle) +
201 f_handle.handle_bytes)) {
202 retval = -EFAULT;
203 goto out_handle;
206 retval = do_handle_to_path(mountdirfd, handle, path);
208 out_handle:
209 kfree(handle);
210 out_err:
211 return retval;
214 long do_handle_open(int mountdirfd,
215 struct file_handle __user *ufh, int open_flag)
217 long retval = 0;
218 struct path path;
219 struct file *file;
220 int fd;
222 retval = handle_to_path(mountdirfd, ufh, &path);
223 if (retval)
224 return retval;
226 fd = get_unused_fd_flags(open_flag);
227 if (fd < 0) {
228 path_put(&path);
229 return fd;
231 file = file_open_root(path.dentry, path.mnt, "", open_flag);
232 if (IS_ERR(file)) {
233 put_unused_fd(fd);
234 retval = PTR_ERR(file);
235 } else {
236 retval = fd;
237 fsnotify_open(file);
238 fd_install(fd, file);
240 path_put(&path);
241 return retval;
245 * sys_open_by_handle_at: Open the file handle
246 * @mountdirfd: directory file descriptor
247 * @handle: file handle to be opened
248 * @flag: open flags.
250 * @mountdirfd indicate the directory file descriptor
251 * of the mount point. file handle is decoded relative
252 * to the vfsmount pointed by the @mountdirfd. @flags
253 * value is same as the open(2) flags.
255 SYSCALL_DEFINE3(open_by_handle_at, int, mountdirfd,
256 struct file_handle __user *, handle,
257 int, flags)
259 long ret;
261 if (force_o_largefile())
262 flags |= O_LARGEFILE;
264 ret = do_handle_open(mountdirfd, handle, flags);
265 return ret;