]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/nfs/nfs4file.c
vfs: allow copy_file_range to copy across devices
[mirror_ubuntu-jammy-kernel.git] / fs / nfs / nfs4file.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * linux/fs/nfs/file.c
4 *
5 * Copyright (C) 1992 Rick Sladkey
6 */
7 #include <linux/fs.h>
8 #include <linux/file.h>
9 #include <linux/falloc.h>
10 #include <linux/nfs_fs.h>
11 #include "delegation.h"
12 #include "internal.h"
13 #include "iostat.h"
14 #include "fscache.h"
15 #include "pnfs.h"
16
17 #include "nfstrace.h"
18
19 #ifdef CONFIG_NFS_V4_2
20 #include "nfs42.h"
21 #endif
22
23 #define NFSDBG_FACILITY NFSDBG_FILE
24
25 static int
26 nfs4_file_open(struct inode *inode, struct file *filp)
27 {
28 struct nfs_open_context *ctx;
29 struct dentry *dentry = file_dentry(filp);
30 struct dentry *parent = NULL;
31 struct inode *dir;
32 unsigned openflags = filp->f_flags;
33 struct iattr attr;
34 int err;
35
36 /*
37 * If no cached dentry exists or if it's negative, NFSv4 handled the
38 * opens in ->lookup() or ->create().
39 *
40 * We only get this far for a cached positive dentry. We skipped
41 * revalidation, so handle it here by dropping the dentry and returning
42 * -EOPENSTALE. The VFS will retry the lookup/create/open.
43 */
44
45 dprintk("NFS: open file(%pd2)\n", dentry);
46
47 err = nfs_check_flags(openflags);
48 if (err)
49 return err;
50
51 if ((openflags & O_ACCMODE) == 3)
52 openflags--;
53
54 /* We can't create new files here */
55 openflags &= ~(O_CREAT|O_EXCL);
56
57 parent = dget_parent(dentry);
58 dir = d_inode(parent);
59
60 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
61 err = PTR_ERR(ctx);
62 if (IS_ERR(ctx))
63 goto out;
64
65 attr.ia_valid = ATTR_OPEN;
66 if (openflags & O_TRUNC) {
67 attr.ia_valid |= ATTR_SIZE;
68 attr.ia_size = 0;
69 filemap_write_and_wait(inode->i_mapping);
70 }
71
72 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr, NULL);
73 if (IS_ERR(inode)) {
74 err = PTR_ERR(inode);
75 switch (err) {
76 case -EPERM:
77 case -EACCES:
78 case -EDQUOT:
79 case -ENOSPC:
80 case -EROFS:
81 goto out_put_ctx;
82 default:
83 goto out_drop;
84 }
85 }
86 if (inode != d_inode(dentry))
87 goto out_drop;
88
89 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
90 nfs_file_set_open_context(filp, ctx);
91 nfs_fscache_open_file(inode, filp);
92 err = 0;
93
94 out_put_ctx:
95 put_nfs_open_context(ctx);
96 out:
97 dput(parent);
98 return err;
99
100 out_drop:
101 d_drop(dentry);
102 err = -EOPENSTALE;
103 goto out_put_ctx;
104 }
105
106 /*
107 * Flush all dirty pages, and check for write errors.
108 */
109 static int
110 nfs4_file_flush(struct file *file, fl_owner_t id)
111 {
112 struct inode *inode = file_inode(file);
113
114 dprintk("NFS: flush(%pD2)\n", file);
115
116 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
117 if ((file->f_mode & FMODE_WRITE) == 0)
118 return 0;
119
120 /*
121 * If we're holding a write delegation, then check if we're required
122 * to flush the i/o on close. If not, then just start the i/o now.
123 */
124 if (!nfs4_delegation_flush_on_close(inode))
125 return filemap_fdatawrite(file->f_mapping);
126
127 /* Flush writes to the server and return any errors */
128 return nfs_wb_all(inode);
129 }
130
131 #ifdef CONFIG_NFS_V4_2
132 static ssize_t __nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
133 struct file *file_out, loff_t pos_out,
134 size_t count, unsigned int flags)
135 {
136 /* Only offload copy if superblock is the same */
137 if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb)
138 return -EXDEV;
139 if (!nfs_server_capable(file_inode(file_out), NFS_CAP_COPY))
140 return -EOPNOTSUPP;
141 if (file_inode(file_in) == file_inode(file_out))
142 return -EOPNOTSUPP;
143 return nfs42_proc_copy(file_in, pos_in, file_out, pos_out, count);
144 }
145
146 static ssize_t nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
147 struct file *file_out, loff_t pos_out,
148 size_t count, unsigned int flags)
149 {
150 ssize_t ret;
151
152 ret = __nfs4_copy_file_range(file_in, pos_in, file_out, pos_out, count,
153 flags);
154 if (ret == -EOPNOTSUPP || ret == -EXDEV)
155 ret = generic_copy_file_range(file_in, pos_in, file_out,
156 pos_out, count, flags);
157 return ret;
158 }
159
160 static loff_t nfs4_file_llseek(struct file *filep, loff_t offset, int whence)
161 {
162 loff_t ret;
163
164 switch (whence) {
165 case SEEK_HOLE:
166 case SEEK_DATA:
167 ret = nfs42_proc_llseek(filep, offset, whence);
168 if (ret != -ENOTSUPP)
169 return ret;
170 /* Fall through */
171 default:
172 return nfs_file_llseek(filep, offset, whence);
173 }
174 }
175
176 static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t len)
177 {
178 struct inode *inode = file_inode(filep);
179 long ret;
180
181 if (!S_ISREG(inode->i_mode))
182 return -EOPNOTSUPP;
183
184 if ((mode != 0) && (mode != (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)))
185 return -EOPNOTSUPP;
186
187 ret = inode_newsize_ok(inode, offset + len);
188 if (ret < 0)
189 return ret;
190
191 if (mode & FALLOC_FL_PUNCH_HOLE)
192 return nfs42_proc_deallocate(filep, offset, len);
193 return nfs42_proc_allocate(filep, offset, len);
194 }
195
196 static loff_t nfs42_remap_file_range(struct file *src_file, loff_t src_off,
197 struct file *dst_file, loff_t dst_off, loff_t count,
198 unsigned int remap_flags)
199 {
200 struct inode *dst_inode = file_inode(dst_file);
201 struct nfs_server *server = NFS_SERVER(dst_inode);
202 struct inode *src_inode = file_inode(src_file);
203 unsigned int bs = server->clone_blksize;
204 bool same_inode = false;
205 int ret;
206
207 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
208 return -EINVAL;
209
210 /* check alignment w.r.t. clone_blksize */
211 ret = -EINVAL;
212 if (bs) {
213 if (!IS_ALIGNED(src_off, bs) || !IS_ALIGNED(dst_off, bs))
214 goto out;
215 if (!IS_ALIGNED(count, bs) && i_size_read(src_inode) != (src_off + count))
216 goto out;
217 }
218
219 if (src_inode == dst_inode)
220 same_inode = true;
221
222 /* XXX: do we lock at all? what if server needs CB_RECALL_LAYOUT? */
223 if (same_inode) {
224 inode_lock(src_inode);
225 } else if (dst_inode < src_inode) {
226 inode_lock_nested(dst_inode, I_MUTEX_PARENT);
227 inode_lock_nested(src_inode, I_MUTEX_CHILD);
228 } else {
229 inode_lock_nested(src_inode, I_MUTEX_PARENT);
230 inode_lock_nested(dst_inode, I_MUTEX_CHILD);
231 }
232
233 /* flush all pending writes on both src and dst so that server
234 * has the latest data */
235 ret = nfs_sync_inode(src_inode);
236 if (ret)
237 goto out_unlock;
238 ret = nfs_sync_inode(dst_inode);
239 if (ret)
240 goto out_unlock;
241
242 ret = nfs42_proc_clone(src_file, dst_file, src_off, dst_off, count);
243
244 /* truncate inode page cache of the dst range so that future reads can fetch
245 * new data from server */
246 if (!ret)
247 truncate_inode_pages_range(&dst_inode->i_data, dst_off, dst_off + count - 1);
248
249 out_unlock:
250 if (same_inode) {
251 inode_unlock(src_inode);
252 } else if (dst_inode < src_inode) {
253 inode_unlock(src_inode);
254 inode_unlock(dst_inode);
255 } else {
256 inode_unlock(dst_inode);
257 inode_unlock(src_inode);
258 }
259 out:
260 return ret < 0 ? ret : count;
261 }
262 #endif /* CONFIG_NFS_V4_2 */
263
264 const struct file_operations nfs4_file_operations = {
265 .read_iter = nfs_file_read,
266 .write_iter = nfs_file_write,
267 .mmap = nfs_file_mmap,
268 .open = nfs4_file_open,
269 .flush = nfs4_file_flush,
270 .release = nfs_file_release,
271 .fsync = nfs_file_fsync,
272 .lock = nfs_lock,
273 .flock = nfs_flock,
274 .splice_read = generic_file_splice_read,
275 .splice_write = iter_file_splice_write,
276 .check_flags = nfs_check_flags,
277 .setlease = simple_nosetlease,
278 #ifdef CONFIG_NFS_V4_2
279 .copy_file_range = nfs4_copy_file_range,
280 .llseek = nfs4_file_llseek,
281 .fallocate = nfs42_fallocate,
282 .remap_file_range = nfs42_remap_file_range,
283 #else
284 .llseek = nfs_file_llseek,
285 #endif
286 };