xref: /openbmc/linux/fs/nfs/nfs4file.c (revision 2572f00d)
1 /*
2  *  linux/fs/nfs/file.c
3  *
4  *  Copyright (C) 1992  Rick Sladkey
5  */
6 #include <linux/fs.h>
7 #include <linux/file.h>
8 #include <linux/falloc.h>
9 #include <linux/nfs_fs.h>
10 #include <uapi/linux/btrfs.h>	/* BTRFS_IOC_CLONE/BTRFS_IOC_CLONE_RANGE */
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 = filp->f_path.dentry;
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(filp->f_path.dentry, filp->f_mode);
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 		nfs_sync_inode(inode);
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 vfs_fsync(file, 0);
129 }
130 
131 static int
132 nfs4_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
133 {
134 	int ret;
135 	struct inode *inode = file_inode(file);
136 
137 	trace_nfs_fsync_enter(inode);
138 
139 	nfs_inode_dio_wait(inode);
140 	do {
141 		ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
142 		if (ret != 0)
143 			break;
144 		mutex_lock(&inode->i_mutex);
145 		ret = nfs_file_fsync_commit(file, start, end, datasync);
146 		if (!ret)
147 			ret = pnfs_sync_inode(inode, !!datasync);
148 		mutex_unlock(&inode->i_mutex);
149 		/*
150 		 * If nfs_file_fsync_commit detected a server reboot, then
151 		 * resend all dirty pages that might have been covered by
152 		 * the NFS_CONTEXT_RESEND_WRITES flag
153 		 */
154 		start = 0;
155 		end = LLONG_MAX;
156 	} while (ret == -EAGAIN);
157 
158 	trace_nfs_fsync_exit(inode, ret);
159 	return ret;
160 }
161 
162 #ifdef CONFIG_NFS_V4_2
163 static loff_t nfs4_file_llseek(struct file *filep, loff_t offset, int whence)
164 {
165 	loff_t ret;
166 
167 	switch (whence) {
168 	case SEEK_HOLE:
169 	case SEEK_DATA:
170 		ret = nfs42_proc_llseek(filep, offset, whence);
171 		if (ret != -ENOTSUPP)
172 			return ret;
173 	default:
174 		return nfs_file_llseek(filep, offset, whence);
175 	}
176 }
177 
178 static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t len)
179 {
180 	struct inode *inode = file_inode(filep);
181 	long ret;
182 
183 	if (!S_ISREG(inode->i_mode))
184 		return -EOPNOTSUPP;
185 
186 	if ((mode != 0) && (mode != (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)))
187 		return -EOPNOTSUPP;
188 
189 	ret = inode_newsize_ok(inode, offset + len);
190 	if (ret < 0)
191 		return ret;
192 
193 	if (mode & FALLOC_FL_PUNCH_HOLE)
194 		return nfs42_proc_deallocate(filep, offset, len);
195 	return nfs42_proc_allocate(filep, offset, len);
196 }
197 
198 static noinline long
199 nfs42_ioctl_clone(struct file *dst_file, unsigned long srcfd,
200 		  u64 src_off, u64 dst_off, u64 count)
201 {
202 	struct inode *dst_inode = file_inode(dst_file);
203 	struct nfs_server *server = NFS_SERVER(dst_inode);
204 	struct fd src_file;
205 	struct inode *src_inode;
206 	unsigned int bs = server->clone_blksize;
207 	bool same_inode = false;
208 	int ret;
209 
210 	/* dst file must be opened for writing */
211 	if (!(dst_file->f_mode & FMODE_WRITE))
212 		return -EINVAL;
213 
214 	ret = mnt_want_write_file(dst_file);
215 	if (ret)
216 		return ret;
217 
218 	src_file = fdget(srcfd);
219 	if (!src_file.file) {
220 		ret = -EBADF;
221 		goto out_drop_write;
222 	}
223 
224 	src_inode = file_inode(src_file.file);
225 
226 	if (src_inode == dst_inode)
227 		same_inode = true;
228 
229 	/* src file must be opened for reading */
230 	if (!(src_file.file->f_mode & FMODE_READ))
231 		goto out_fput;
232 
233 	/* src and dst must be regular files */
234 	ret = -EISDIR;
235 	if (!S_ISREG(src_inode->i_mode) || !S_ISREG(dst_inode->i_mode))
236 		goto out_fput;
237 
238 	ret = -EXDEV;
239 	if (src_file.file->f_path.mnt != dst_file->f_path.mnt ||
240 	    src_inode->i_sb != dst_inode->i_sb)
241 		goto out_fput;
242 
243 	/* check alignment w.r.t. clone_blksize */
244 	ret = -EINVAL;
245 	if (bs) {
246 		if (!IS_ALIGNED(src_off, bs) || !IS_ALIGNED(dst_off, bs))
247 			goto out_fput;
248 		if (!IS_ALIGNED(count, bs) && i_size_read(src_inode) != (src_off + count))
249 			goto out_fput;
250 	}
251 
252 	/* verify if ranges are overlapped within the same file */
253 	if (same_inode) {
254 		if (dst_off + count > src_off && dst_off < src_off + count)
255 			goto out_fput;
256 	}
257 
258 	/* XXX: do we lock at all? what if server needs CB_RECALL_LAYOUT? */
259 	if (same_inode) {
260 		mutex_lock(&src_inode->i_mutex);
261 	} else if (dst_inode < src_inode) {
262 		mutex_lock_nested(&dst_inode->i_mutex, I_MUTEX_PARENT);
263 		mutex_lock_nested(&src_inode->i_mutex, I_MUTEX_CHILD);
264 	} else {
265 		mutex_lock_nested(&src_inode->i_mutex, I_MUTEX_PARENT);
266 		mutex_lock_nested(&dst_inode->i_mutex, I_MUTEX_CHILD);
267 	}
268 
269 	/* flush all pending writes on both src and dst so that server
270 	 * has the latest data */
271 	ret = nfs_sync_inode(src_inode);
272 	if (ret)
273 		goto out_unlock;
274 	ret = nfs_sync_inode(dst_inode);
275 	if (ret)
276 		goto out_unlock;
277 
278 	ret = nfs42_proc_clone(src_file.file, dst_file, src_off, dst_off, count);
279 
280 	/* truncate inode page cache of the dst range so that future reads can fetch
281 	 * new data from server */
282 	if (!ret)
283 		truncate_inode_pages_range(&dst_inode->i_data, dst_off, dst_off + count - 1);
284 
285 out_unlock:
286 	if (same_inode) {
287 		mutex_unlock(&src_inode->i_mutex);
288 	} else if (dst_inode < src_inode) {
289 		mutex_unlock(&src_inode->i_mutex);
290 		mutex_unlock(&dst_inode->i_mutex);
291 	} else {
292 		mutex_unlock(&dst_inode->i_mutex);
293 		mutex_unlock(&src_inode->i_mutex);
294 	}
295 out_fput:
296 	fdput(src_file);
297 out_drop_write:
298 	mnt_drop_write_file(dst_file);
299 	return ret;
300 }
301 
302 static long nfs42_ioctl_clone_range(struct file *dst_file, void __user *argp)
303 {
304 	struct btrfs_ioctl_clone_range_args args;
305 
306 	if (copy_from_user(&args, argp, sizeof(args)))
307 		return -EFAULT;
308 
309 	return nfs42_ioctl_clone(dst_file, args.src_fd, args.src_offset,
310 				 args.dest_offset, args.src_length);
311 }
312 
313 long nfs4_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
314 {
315 	void __user *argp = (void __user *)arg;
316 
317 	switch (cmd) {
318 	case BTRFS_IOC_CLONE:
319 		return nfs42_ioctl_clone(file, arg, 0, 0, 0);
320 	case BTRFS_IOC_CLONE_RANGE:
321 		return nfs42_ioctl_clone_range(file, argp);
322 	}
323 
324 	return -ENOTTY;
325 }
326 #endif /* CONFIG_NFS_V4_2 */
327 
328 const struct file_operations nfs4_file_operations = {
329 	.read_iter	= nfs_file_read,
330 	.write_iter	= nfs_file_write,
331 	.mmap		= nfs_file_mmap,
332 	.open		= nfs4_file_open,
333 	.flush		= nfs4_file_flush,
334 	.release	= nfs_file_release,
335 	.fsync		= nfs4_file_fsync,
336 	.lock		= nfs_lock,
337 	.flock		= nfs_flock,
338 	.splice_read	= nfs_file_splice_read,
339 	.splice_write	= iter_file_splice_write,
340 	.check_flags	= nfs_check_flags,
341 	.setlease	= simple_nosetlease,
342 #ifdef CONFIG_NFS_V4_2
343 	.llseek		= nfs4_file_llseek,
344 	.fallocate	= nfs42_fallocate,
345 	.unlocked_ioctl = nfs4_ioctl,
346 	.compat_ioctl	= nfs4_ioctl,
347 #else
348 	.llseek		= nfs_file_llseek,
349 #endif
350 };
351