xref: /openbmc/linux/fs/nfs/nfs4file.c (revision aa5b395b)
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 		return nfs_open(inode, filp);
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 		default:
77 			goto out_put_ctx;
78 		case -ENOENT:
79 		case -ESTALE:
80 		case -EISDIR:
81 		case -ENOTDIR:
82 		case -ELOOP:
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 	struct nfs42_copy_notify_res *cn_resp = NULL;
137 	struct nl4_server *nss = NULL;
138 	nfs4_stateid *cnrs = NULL;
139 	ssize_t ret;
140 	bool sync = false;
141 
142 	/* Only offload copy if superblock is the same */
143 	if (file_in->f_op != &nfs4_file_operations)
144 		return -EXDEV;
145 	if (!nfs_server_capable(file_inode(file_out), NFS_CAP_COPY))
146 		return -EOPNOTSUPP;
147 	if (file_inode(file_in) == file_inode(file_out))
148 		return -EOPNOTSUPP;
149 	/* if the copy size if smaller than 2 RPC payloads, make it
150 	 * synchronous
151 	 */
152 	if (count <= 2 * NFS_SERVER(file_inode(file_in))->rsize)
153 		sync = true;
154 retry:
155 	if (!nfs42_files_from_same_server(file_in, file_out)) {
156 		/* for inter copy, if copy size if smaller than 12 RPC
157 		 * payloads, fallback to traditional copy. There are
158 		 * 14 RPCs during an NFSv4.x mount between source/dest
159 		 * servers.
160 		 */
161 		if (sync ||
162 			count <= 14 * NFS_SERVER(file_inode(file_in))->rsize)
163 			return -EOPNOTSUPP;
164 		cn_resp = kzalloc(sizeof(struct nfs42_copy_notify_res),
165 				GFP_NOFS);
166 		if (unlikely(cn_resp == NULL))
167 			return -ENOMEM;
168 
169 		ret = nfs42_proc_copy_notify(file_in, file_out, cn_resp);
170 		if (ret) {
171 			ret = -EOPNOTSUPP;
172 			goto out;
173 		}
174 		nss = &cn_resp->cnr_src;
175 		cnrs = &cn_resp->cnr_stateid;
176 	}
177 	ret = nfs42_proc_copy(file_in, pos_in, file_out, pos_out, count,
178 				nss, cnrs, sync);
179 out:
180 	if (!nfs42_files_from_same_server(file_in, file_out))
181 		kfree(cn_resp);
182 	if (ret == -EAGAIN)
183 		goto retry;
184 	return ret;
185 }
186 
187 static ssize_t nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
188 				    struct file *file_out, loff_t pos_out,
189 				    size_t count, unsigned int flags)
190 {
191 	ssize_t ret;
192 
193 	ret = __nfs4_copy_file_range(file_in, pos_in, file_out, pos_out, count,
194 				     flags);
195 	if (ret == -EOPNOTSUPP || ret == -EXDEV)
196 		ret = generic_copy_file_range(file_in, pos_in, file_out,
197 					      pos_out, count, flags);
198 	return ret;
199 }
200 
201 static loff_t nfs4_file_llseek(struct file *filep, loff_t offset, int whence)
202 {
203 	loff_t ret;
204 
205 	switch (whence) {
206 	case SEEK_HOLE:
207 	case SEEK_DATA:
208 		ret = nfs42_proc_llseek(filep, offset, whence);
209 		if (ret != -ENOTSUPP)
210 			return ret;
211 		/* Fall through */
212 	default:
213 		return nfs_file_llseek(filep, offset, whence);
214 	}
215 }
216 
217 static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t len)
218 {
219 	struct inode *inode = file_inode(filep);
220 	long ret;
221 
222 	if (!S_ISREG(inode->i_mode))
223 		return -EOPNOTSUPP;
224 
225 	if ((mode != 0) && (mode != (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)))
226 		return -EOPNOTSUPP;
227 
228 	ret = inode_newsize_ok(inode, offset + len);
229 	if (ret < 0)
230 		return ret;
231 
232 	if (mode & FALLOC_FL_PUNCH_HOLE)
233 		return nfs42_proc_deallocate(filep, offset, len);
234 	return nfs42_proc_allocate(filep, offset, len);
235 }
236 
237 static loff_t nfs42_remap_file_range(struct file *src_file, loff_t src_off,
238 		struct file *dst_file, loff_t dst_off, loff_t count,
239 		unsigned int remap_flags)
240 {
241 	struct inode *dst_inode = file_inode(dst_file);
242 	struct nfs_server *server = NFS_SERVER(dst_inode);
243 	struct inode *src_inode = file_inode(src_file);
244 	unsigned int bs = server->clone_blksize;
245 	bool same_inode = false;
246 	int ret;
247 
248 	/* NFS does not support deduplication. */
249 	if (remap_flags & REMAP_FILE_DEDUP)
250 		return -EOPNOTSUPP;
251 
252 	if (remap_flags & ~REMAP_FILE_ADVISORY)
253 		return -EINVAL;
254 
255 	/* check alignment w.r.t. clone_blksize */
256 	ret = -EINVAL;
257 	if (bs) {
258 		if (!IS_ALIGNED(src_off, bs) || !IS_ALIGNED(dst_off, bs))
259 			goto out;
260 		if (!IS_ALIGNED(count, bs) && i_size_read(src_inode) != (src_off + count))
261 			goto out;
262 	}
263 
264 	if (src_inode == dst_inode)
265 		same_inode = true;
266 
267 	/* XXX: do we lock at all? what if server needs CB_RECALL_LAYOUT? */
268 	if (same_inode) {
269 		inode_lock(src_inode);
270 	} else if (dst_inode < src_inode) {
271 		inode_lock_nested(dst_inode, I_MUTEX_PARENT);
272 		inode_lock_nested(src_inode, I_MUTEX_CHILD);
273 	} else {
274 		inode_lock_nested(src_inode, I_MUTEX_PARENT);
275 		inode_lock_nested(dst_inode, I_MUTEX_CHILD);
276 	}
277 
278 	/* flush all pending writes on both src and dst so that server
279 	 * has the latest data */
280 	ret = nfs_sync_inode(src_inode);
281 	if (ret)
282 		goto out_unlock;
283 	ret = nfs_sync_inode(dst_inode);
284 	if (ret)
285 		goto out_unlock;
286 
287 	ret = nfs42_proc_clone(src_file, dst_file, src_off, dst_off, count);
288 
289 	/* truncate inode page cache of the dst range so that future reads can fetch
290 	 * new data from server */
291 	if (!ret)
292 		truncate_inode_pages_range(&dst_inode->i_data, dst_off, dst_off + count - 1);
293 
294 out_unlock:
295 	if (same_inode) {
296 		inode_unlock(src_inode);
297 	} else if (dst_inode < src_inode) {
298 		inode_unlock(src_inode);
299 		inode_unlock(dst_inode);
300 	} else {
301 		inode_unlock(dst_inode);
302 		inode_unlock(src_inode);
303 	}
304 out:
305 	return ret < 0 ? ret : count;
306 }
307 
308 static int read_name_gen = 1;
309 #define SSC_READ_NAME_BODY "ssc_read_%d"
310 
311 struct file *
312 nfs42_ssc_open(struct vfsmount *ss_mnt, struct nfs_fh *src_fh,
313 		nfs4_stateid *stateid)
314 {
315 	struct nfs_fattr fattr;
316 	struct file *filep, *res;
317 	struct nfs_server *server;
318 	struct inode *r_ino = NULL;
319 	struct nfs_open_context *ctx;
320 	struct nfs4_state_owner *sp;
321 	char *read_name = NULL;
322 	int len, status = 0;
323 
324 	server = NFS_SERVER(ss_mnt->mnt_root->d_inode);
325 
326 	nfs_fattr_init(&fattr);
327 
328 	status = nfs4_proc_getattr(server, src_fh, &fattr, NULL, NULL);
329 	if (status < 0) {
330 		res = ERR_PTR(status);
331 		goto out;
332 	}
333 
334 	res = ERR_PTR(-ENOMEM);
335 	len = strlen(SSC_READ_NAME_BODY) + 16;
336 	read_name = kzalloc(len, GFP_NOFS);
337 	if (read_name == NULL)
338 		goto out;
339 	snprintf(read_name, len, SSC_READ_NAME_BODY, read_name_gen++);
340 
341 	r_ino = nfs_fhget(ss_mnt->mnt_root->d_inode->i_sb, src_fh, &fattr,
342 			NULL);
343 	if (IS_ERR(r_ino)) {
344 		res = ERR_CAST(r_ino);
345 		goto out_free_name;
346 	}
347 
348 	filep = alloc_file_pseudo(r_ino, ss_mnt, read_name, FMODE_READ,
349 				     r_ino->i_fop);
350 	if (IS_ERR(filep)) {
351 		res = ERR_CAST(filep);
352 		goto out_free_name;
353 	}
354 	filep->f_mode |= FMODE_READ;
355 
356 	ctx = alloc_nfs_open_context(filep->f_path.dentry, filep->f_mode,
357 					filep);
358 	if (IS_ERR(ctx)) {
359 		res = ERR_CAST(ctx);
360 		goto out_filep;
361 	}
362 
363 	res = ERR_PTR(-EINVAL);
364 	sp = nfs4_get_state_owner(server, ctx->cred, GFP_KERNEL);
365 	if (sp == NULL)
366 		goto out_ctx;
367 
368 	ctx->state = nfs4_get_open_state(r_ino, sp);
369 	if (ctx->state == NULL)
370 		goto out_stateowner;
371 
372 	set_bit(NFS_SRV_SSC_COPY_STATE, &ctx->state->flags);
373 	set_bit(NFS_OPEN_STATE, &ctx->state->flags);
374 	memcpy(&ctx->state->open_stateid.other, &stateid->other,
375 	       NFS4_STATEID_OTHER_SIZE);
376 	update_open_stateid(ctx->state, stateid, NULL, filep->f_mode);
377 
378 	nfs_file_set_open_context(filep, ctx);
379 	put_nfs_open_context(ctx);
380 
381 	file_ra_state_init(&filep->f_ra, filep->f_mapping->host->i_mapping);
382 	res = filep;
383 out_free_name:
384 	kfree(read_name);
385 out:
386 	return res;
387 out_stateowner:
388 	nfs4_put_state_owner(sp);
389 out_ctx:
390 	put_nfs_open_context(ctx);
391 out_filep:
392 	fput(filep);
393 	goto out_free_name;
394 }
395 EXPORT_SYMBOL_GPL(nfs42_ssc_open);
396 void nfs42_ssc_close(struct file *filep)
397 {
398 	struct nfs_open_context *ctx = nfs_file_open_context(filep);
399 
400 	ctx->state->flags = 0;
401 }
402 EXPORT_SYMBOL_GPL(nfs42_ssc_close);
403 #endif /* CONFIG_NFS_V4_2 */
404 
405 const struct file_operations nfs4_file_operations = {
406 	.read_iter	= nfs_file_read,
407 	.write_iter	= nfs_file_write,
408 	.mmap		= nfs_file_mmap,
409 	.open		= nfs4_file_open,
410 	.flush		= nfs4_file_flush,
411 	.release	= nfs_file_release,
412 	.fsync		= nfs_file_fsync,
413 	.lock		= nfs_lock,
414 	.flock		= nfs_flock,
415 	.splice_read	= generic_file_splice_read,
416 	.splice_write	= iter_file_splice_write,
417 	.check_flags	= nfs_check_flags,
418 	.setlease	= simple_nosetlease,
419 #ifdef CONFIG_NFS_V4_2
420 	.copy_file_range = nfs4_copy_file_range,
421 	.llseek		= nfs4_file_llseek,
422 	.fallocate	= nfs42_fallocate,
423 	.remap_file_range = nfs42_remap_file_range,
424 #else
425 	.llseek		= nfs_file_llseek,
426 #endif
427 };
428