xref: /openbmc/linux/fs/nfs/nfs4file.c (revision 4eb5928d)
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/mount.h>
11 #include <linux/nfs_fs.h>
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 
26 static int
27 nfs4_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_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 	errseq_t since;
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 	since = filemap_sample_wb_err(file->f_mapping);
130 	nfs_wb_all(inode);
131 	return filemap_check_wb_err(file->f_mapping, since);
132 }
133 
134 #ifdef CONFIG_NFS_V4_2
135 static ssize_t __nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
136 				      struct file *file_out, loff_t pos_out,
137 				      size_t count, unsigned int flags)
138 {
139 	struct nfs42_copy_notify_res *cn_resp = NULL;
140 	struct nl4_server *nss = NULL;
141 	nfs4_stateid *cnrs = NULL;
142 	ssize_t ret;
143 	bool sync = false;
144 
145 	/* Only offload copy if superblock is the same */
146 	if (file_in->f_op != &nfs4_file_operations)
147 		return -EXDEV;
148 	if (!nfs_server_capable(file_inode(file_out), NFS_CAP_COPY))
149 		return -EOPNOTSUPP;
150 	if (file_inode(file_in) == file_inode(file_out))
151 		return -EOPNOTSUPP;
152 	/* if the copy size if smaller than 2 RPC payloads, make it
153 	 * synchronous
154 	 */
155 	if (count <= 2 * NFS_SERVER(file_inode(file_in))->rsize)
156 		sync = true;
157 retry:
158 	if (!nfs42_files_from_same_server(file_in, file_out)) {
159 		/* for inter copy, if copy size if smaller than 12 RPC
160 		 * payloads, fallback to traditional copy. There are
161 		 * 14 RPCs during an NFSv4.x mount between source/dest
162 		 * servers.
163 		 */
164 		if (sync ||
165 			count <= 14 * NFS_SERVER(file_inode(file_in))->rsize)
166 			return -EOPNOTSUPP;
167 		cn_resp = kzalloc(sizeof(struct nfs42_copy_notify_res),
168 				GFP_NOFS);
169 		if (unlikely(cn_resp == NULL))
170 			return -ENOMEM;
171 
172 		ret = nfs42_proc_copy_notify(file_in, file_out, cn_resp);
173 		if (ret) {
174 			ret = -EOPNOTSUPP;
175 			goto out;
176 		}
177 		nss = &cn_resp->cnr_src;
178 		cnrs = &cn_resp->cnr_stateid;
179 	}
180 	ret = nfs42_proc_copy(file_in, pos_in, file_out, pos_out, count,
181 				nss, cnrs, sync);
182 out:
183 	if (!nfs42_files_from_same_server(file_in, file_out))
184 		kfree(cn_resp);
185 	if (ret == -EAGAIN)
186 		goto retry;
187 	return ret;
188 }
189 
190 static ssize_t nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
191 				    struct file *file_out, loff_t pos_out,
192 				    size_t count, unsigned int flags)
193 {
194 	ssize_t ret;
195 
196 	ret = __nfs4_copy_file_range(file_in, pos_in, file_out, pos_out, count,
197 				     flags);
198 	if (ret == -EOPNOTSUPP || ret == -EXDEV)
199 		ret = generic_copy_file_range(file_in, pos_in, file_out,
200 					      pos_out, count, flags);
201 	return ret;
202 }
203 
204 static loff_t nfs4_file_llseek(struct file *filep, loff_t offset, int whence)
205 {
206 	loff_t ret;
207 
208 	switch (whence) {
209 	case SEEK_HOLE:
210 	case SEEK_DATA:
211 		ret = nfs42_proc_llseek(filep, offset, whence);
212 		if (ret != -ENOTSUPP)
213 			return ret;
214 		/* Fall through */
215 	default:
216 		return nfs_file_llseek(filep, offset, whence);
217 	}
218 }
219 
220 static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t len)
221 {
222 	struct inode *inode = file_inode(filep);
223 	long ret;
224 
225 	if (!S_ISREG(inode->i_mode))
226 		return -EOPNOTSUPP;
227 
228 	if ((mode != 0) && (mode != (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)))
229 		return -EOPNOTSUPP;
230 
231 	ret = inode_newsize_ok(inode, offset + len);
232 	if (ret < 0)
233 		return ret;
234 
235 	if (mode & FALLOC_FL_PUNCH_HOLE)
236 		return nfs42_proc_deallocate(filep, offset, len);
237 	return nfs42_proc_allocate(filep, offset, len);
238 }
239 
240 static loff_t nfs42_remap_file_range(struct file *src_file, loff_t src_off,
241 		struct file *dst_file, loff_t dst_off, loff_t count,
242 		unsigned int remap_flags)
243 {
244 	struct inode *dst_inode = file_inode(dst_file);
245 	struct nfs_server *server = NFS_SERVER(dst_inode);
246 	struct inode *src_inode = file_inode(src_file);
247 	unsigned int bs = server->clone_blksize;
248 	bool same_inode = false;
249 	int ret;
250 
251 	/* NFS does not support deduplication. */
252 	if (remap_flags & REMAP_FILE_DEDUP)
253 		return -EOPNOTSUPP;
254 
255 	if (remap_flags & ~REMAP_FILE_ADVISORY)
256 		return -EINVAL;
257 
258 	if (IS_SWAPFILE(dst_inode) || IS_SWAPFILE(src_inode))
259 		return -ETXTBSY;
260 
261 	/* check alignment w.r.t. clone_blksize */
262 	ret = -EINVAL;
263 	if (bs) {
264 		if (!IS_ALIGNED(src_off, bs) || !IS_ALIGNED(dst_off, bs))
265 			goto out;
266 		if (!IS_ALIGNED(count, bs) && i_size_read(src_inode) != (src_off + count))
267 			goto out;
268 	}
269 
270 	if (src_inode == dst_inode)
271 		same_inode = true;
272 
273 	/* XXX: do we lock at all? what if server needs CB_RECALL_LAYOUT? */
274 	if (same_inode) {
275 		inode_lock(src_inode);
276 	} else if (dst_inode < src_inode) {
277 		inode_lock_nested(dst_inode, I_MUTEX_PARENT);
278 		inode_lock_nested(src_inode, I_MUTEX_CHILD);
279 	} else {
280 		inode_lock_nested(src_inode, I_MUTEX_PARENT);
281 		inode_lock_nested(dst_inode, I_MUTEX_CHILD);
282 	}
283 
284 	/* flush all pending writes on both src and dst so that server
285 	 * has the latest data */
286 	ret = nfs_sync_inode(src_inode);
287 	if (ret)
288 		goto out_unlock;
289 	ret = nfs_sync_inode(dst_inode);
290 	if (ret)
291 		goto out_unlock;
292 
293 	ret = nfs42_proc_clone(src_file, dst_file, src_off, dst_off, count);
294 
295 	/* truncate inode page cache of the dst range so that future reads can fetch
296 	 * new data from server */
297 	if (!ret)
298 		truncate_inode_pages_range(&dst_inode->i_data, dst_off, dst_off + count - 1);
299 
300 out_unlock:
301 	if (same_inode) {
302 		inode_unlock(src_inode);
303 	} else if (dst_inode < src_inode) {
304 		inode_unlock(src_inode);
305 		inode_unlock(dst_inode);
306 	} else {
307 		inode_unlock(dst_inode);
308 		inode_unlock(src_inode);
309 	}
310 out:
311 	return ret < 0 ? ret : count;
312 }
313 
314 static int read_name_gen = 1;
315 #define SSC_READ_NAME_BODY "ssc_read_%d"
316 
317 struct file *
318 nfs42_ssc_open(struct vfsmount *ss_mnt, struct nfs_fh *src_fh,
319 		nfs4_stateid *stateid)
320 {
321 	struct nfs_fattr fattr;
322 	struct file *filep, *res;
323 	struct nfs_server *server;
324 	struct inode *r_ino = NULL;
325 	struct nfs_open_context *ctx;
326 	struct nfs4_state_owner *sp;
327 	char *read_name = NULL;
328 	int len, status = 0;
329 
330 	server = NFS_SERVER(ss_mnt->mnt_root->d_inode);
331 
332 	nfs_fattr_init(&fattr);
333 
334 	status = nfs4_proc_getattr(server, src_fh, &fattr, NULL, NULL);
335 	if (status < 0) {
336 		res = ERR_PTR(status);
337 		goto out;
338 	}
339 
340 	res = ERR_PTR(-ENOMEM);
341 	len = strlen(SSC_READ_NAME_BODY) + 16;
342 	read_name = kzalloc(len, GFP_NOFS);
343 	if (read_name == NULL)
344 		goto out;
345 	snprintf(read_name, len, SSC_READ_NAME_BODY, read_name_gen++);
346 
347 	r_ino = nfs_fhget(ss_mnt->mnt_root->d_inode->i_sb, src_fh, &fattr,
348 			NULL);
349 	if (IS_ERR(r_ino)) {
350 		res = ERR_CAST(r_ino);
351 		goto out_free_name;
352 	}
353 
354 	filep = alloc_file_pseudo(r_ino, ss_mnt, read_name, FMODE_READ,
355 				     r_ino->i_fop);
356 	if (IS_ERR(filep)) {
357 		res = ERR_CAST(filep);
358 		goto out_free_name;
359 	}
360 	filep->f_mode |= FMODE_READ;
361 
362 	ctx = alloc_nfs_open_context(filep->f_path.dentry, filep->f_mode,
363 					filep);
364 	if (IS_ERR(ctx)) {
365 		res = ERR_CAST(ctx);
366 		goto out_filep;
367 	}
368 
369 	res = ERR_PTR(-EINVAL);
370 	sp = nfs4_get_state_owner(server, ctx->cred, GFP_KERNEL);
371 	if (sp == NULL)
372 		goto out_ctx;
373 
374 	ctx->state = nfs4_get_open_state(r_ino, sp);
375 	if (ctx->state == NULL)
376 		goto out_stateowner;
377 
378 	set_bit(NFS_SRV_SSC_COPY_STATE, &ctx->state->flags);
379 	set_bit(NFS_OPEN_STATE, &ctx->state->flags);
380 	memcpy(&ctx->state->open_stateid.other, &stateid->other,
381 	       NFS4_STATEID_OTHER_SIZE);
382 	update_open_stateid(ctx->state, stateid, NULL, filep->f_mode);
383 
384 	nfs_file_set_open_context(filep, ctx);
385 	put_nfs_open_context(ctx);
386 
387 	file_ra_state_init(&filep->f_ra, filep->f_mapping->host->i_mapping);
388 	res = filep;
389 out_free_name:
390 	kfree(read_name);
391 out:
392 	return res;
393 out_stateowner:
394 	nfs4_put_state_owner(sp);
395 out_ctx:
396 	put_nfs_open_context(ctx);
397 out_filep:
398 	fput(filep);
399 	goto out_free_name;
400 }
401 EXPORT_SYMBOL_GPL(nfs42_ssc_open);
402 void nfs42_ssc_close(struct file *filep)
403 {
404 	struct nfs_open_context *ctx = nfs_file_open_context(filep);
405 
406 	ctx->state->flags = 0;
407 }
408 EXPORT_SYMBOL_GPL(nfs42_ssc_close);
409 #endif /* CONFIG_NFS_V4_2 */
410 
411 const struct file_operations nfs4_file_operations = {
412 	.read_iter	= nfs_file_read,
413 	.write_iter	= nfs_file_write,
414 	.mmap		= nfs_file_mmap,
415 	.open		= nfs4_file_open,
416 	.flush		= nfs4_file_flush,
417 	.release	= nfs_file_release,
418 	.fsync		= nfs_file_fsync,
419 	.lock		= nfs_lock,
420 	.flock		= nfs_flock,
421 	.splice_read	= generic_file_splice_read,
422 	.splice_write	= iter_file_splice_write,
423 	.check_flags	= nfs_check_flags,
424 	.setlease	= simple_nosetlease,
425 #ifdef CONFIG_NFS_V4_2
426 	.copy_file_range = nfs4_copy_file_range,
427 	.llseek		= nfs4_file_llseek,
428 	.fallocate	= nfs42_fallocate,
429 	.remap_file_range = nfs42_remap_file_range,
430 #else
431 	.llseek		= nfs_file_llseek,
432 #endif
433 };
434