blob: b8d316a338bc96b6f37e33303625ee613b2e111d [file] [log] [blame]
rjw1f884582022-01-06 17:20:42 +08001// 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 <uapi/linux/btrfs.h> /* BTRFS_IOC_CLONE/BTRFS_IOC_CLONE_RANGE */
12#include "delegation.h"
13#include "internal.h"
14#include "iostat.h"
15#include "fscache.h"
16#include "pnfs.h"
17
18#include "nfstrace.h"
19
20#ifdef CONFIG_NFS_V4_2
21#include "nfs42.h"
22#endif
23
24#define NFSDBG_FACILITY NFSDBG_FILE
25
26static int
27nfs4_file_open(struct inode *inode, struct file *filp)
28{
29 struct nfs_open_context *ctx;
30 struct dentry *dentry = file_dentry(filp);
31 struct dentry *parent = NULL;
32 struct inode *dir;
33 unsigned openflags = filp->f_flags;
34 struct iattr attr;
35 int err;
36
37 /*
38 * If no cached dentry exists or if it's negative, NFSv4 handled the
39 * opens in ->lookup() or ->create().
40 *
41 * We only get this far for a cached positive dentry. We skipped
42 * revalidation, so handle it here by dropping the dentry and returning
43 * -EOPENSTALE. The VFS will retry the lookup/create/open.
44 */
45
46 dprintk("NFS: open file(%pd2)\n", dentry);
47
48 err = nfs_check_flags(openflags);
49 if (err)
50 return err;
51
52 if ((openflags & O_ACCMODE) == 3)
53 return nfs_open(inode, filp);
54
55 /* We can't create new files here */
56 openflags &= ~(O_CREAT|O_EXCL);
57
58 parent = dget_parent(dentry);
59 dir = d_inode(parent);
60
61 ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
62 err = PTR_ERR(ctx);
63 if (IS_ERR(ctx))
64 goto out;
65
66 attr.ia_valid = ATTR_OPEN;
67 if (openflags & O_TRUNC) {
68 attr.ia_valid |= ATTR_SIZE;
69 attr.ia_size = 0;
70 filemap_write_and_wait(inode->i_mapping);
71 }
72
73 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr, NULL);
74 if (IS_ERR(inode)) {
75 err = PTR_ERR(inode);
76 switch (err) {
77 default:
78 goto out_put_ctx;
79 case -ENOENT:
80 case -ESTALE:
81 case -EISDIR:
82 case -ENOTDIR:
83 case -ELOOP:
84 goto out_drop;
85 }
86 }
87 if (inode != d_inode(dentry))
88 goto out_drop;
89
90 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
91 nfs_file_set_open_context(filp, ctx);
92 nfs_fscache_open_file(inode, filp);
93 err = 0;
94
95out_put_ctx:
96 put_nfs_open_context(ctx);
97out:
98 dput(parent);
99 return err;
100
101out_drop:
102 d_drop(dentry);
103 err = -EOPENSTALE;
104 goto out_put_ctx;
105}
106
107/*
108 * Flush all dirty pages, and check for write errors.
109 */
110static int
111nfs4_file_flush(struct file *file, fl_owner_t id)
112{
113 struct inode *inode = file_inode(file);
114
115 dprintk("NFS: flush(%pD2)\n", file);
116
117 nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
118 if ((file->f_mode & FMODE_WRITE) == 0)
119 return 0;
120
121 /*
122 * If we're holding a write delegation, then check if we're required
123 * to flush the i/o on close. If not, then just start the i/o now.
124 */
125 if (!nfs4_delegation_flush_on_close(inode))
126 return filemap_fdatawrite(file->f_mapping);
127
128 /* Flush writes to the server and return any errors */
129 return vfs_fsync(file, 0);
130}
131
132#ifdef CONFIG_NFS_V4_2
133static ssize_t nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
134 struct file *file_out, loff_t pos_out,
135 size_t count, unsigned int flags)
136{
137 if (file_inode(file_in) == file_inode(file_out))
138 return -EINVAL;
139
140 return nfs42_proc_copy(file_in, pos_in, file_out, pos_out, count);
141}
142
143static loff_t nfs4_file_llseek(struct file *filep, loff_t offset, int whence)
144{
145 loff_t ret;
146
147 switch (whence) {
148 case SEEK_HOLE:
149 case SEEK_DATA:
150 ret = nfs42_proc_llseek(filep, offset, whence);
151 if (ret != -ENOTSUPP)
152 return ret;
153 default:
154 return nfs_file_llseek(filep, offset, whence);
155 }
156}
157
158static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t len)
159{
160 struct inode *inode = file_inode(filep);
161 long ret;
162
163 if (!S_ISREG(inode->i_mode))
164 return -EOPNOTSUPP;
165
166 if ((mode != 0) && (mode != (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)))
167 return -EOPNOTSUPP;
168
169 ret = inode_newsize_ok(inode, offset + len);
170 if (ret < 0)
171 return ret;
172
173 if (mode & FALLOC_FL_PUNCH_HOLE)
174 return nfs42_proc_deallocate(filep, offset, len);
175 return nfs42_proc_allocate(filep, offset, len);
176}
177
178static int nfs42_clone_file_range(struct file *src_file, loff_t src_off,
179 struct file *dst_file, loff_t dst_off, u64 count)
180{
181 struct inode *dst_inode = file_inode(dst_file);
182 struct nfs_server *server = NFS_SERVER(dst_inode);
183 struct inode *src_inode = file_inode(src_file);
184 unsigned int bs = server->clone_blksize;
185 bool same_inode = false;
186 int ret;
187
188 /* check alignment w.r.t. clone_blksize */
189 ret = -EINVAL;
190 if (bs) {
191 if (!IS_ALIGNED(src_off, bs) || !IS_ALIGNED(dst_off, bs))
192 goto out;
193 if (!IS_ALIGNED(count, bs) && i_size_read(src_inode) != (src_off + count))
194 goto out;
195 }
196
197 if (src_inode == dst_inode)
198 same_inode = true;
199
200 /* XXX: do we lock at all? what if server needs CB_RECALL_LAYOUT? */
201 if (same_inode) {
202 inode_lock(src_inode);
203 } else if (dst_inode < src_inode) {
204 inode_lock_nested(dst_inode, I_MUTEX_PARENT);
205 inode_lock_nested(src_inode, I_MUTEX_CHILD);
206 } else {
207 inode_lock_nested(src_inode, I_MUTEX_PARENT);
208 inode_lock_nested(dst_inode, I_MUTEX_CHILD);
209 }
210
211 /* flush all pending writes on both src and dst so that server
212 * has the latest data */
213 ret = nfs_sync_inode(src_inode);
214 if (ret)
215 goto out_unlock;
216 ret = nfs_sync_inode(dst_inode);
217 if (ret)
218 goto out_unlock;
219
220 ret = nfs42_proc_clone(src_file, dst_file, src_off, dst_off, count);
221
222 /* truncate inode page cache of the dst range so that future reads can fetch
223 * new data from server */
224 if (!ret)
225 truncate_inode_pages_range(&dst_inode->i_data, dst_off, dst_off + count - 1);
226
227out_unlock:
228 if (same_inode) {
229 inode_unlock(src_inode);
230 } else if (dst_inode < src_inode) {
231 inode_unlock(src_inode);
232 inode_unlock(dst_inode);
233 } else {
234 inode_unlock(dst_inode);
235 inode_unlock(src_inode);
236 }
237out:
238 return ret;
239}
240#endif /* CONFIG_NFS_V4_2 */
241
242const struct file_operations nfs4_file_operations = {
243 .read_iter = nfs_file_read,
244 .write_iter = nfs_file_write,
245 .mmap = nfs_file_mmap,
246 .open = nfs4_file_open,
247 .flush = nfs4_file_flush,
248 .release = nfs_file_release,
249 .fsync = nfs_file_fsync,
250 .lock = nfs_lock,
251 .flock = nfs_flock,
252 .splice_read = generic_file_splice_read,
253 .splice_write = iter_file_splice_write,
254 .check_flags = nfs_check_flags,
255 .setlease = simple_nosetlease,
256#ifdef CONFIG_NFS_V4_2
257 .copy_file_range = nfs4_copy_file_range,
258 .llseek = nfs4_file_llseek,
259 .fallocate = nfs42_fallocate,
260 .clone_file_range = nfs42_clone_file_range,
261#else
262 .llseek = nfs_file_llseek,
263#endif
264};