xref: /openbmc/linux/fs/nfs/nfs4file.c (revision dc023681)
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 <linux/nfs_ssc.h>
13 #include "delegation.h"
14 #include "internal.h"
15 #include "iostat.h"
16 #include "fscache.h"
17 #include "pnfs.h"
18 
19 #include "nfstrace.h"
20 
21 #ifdef CONFIG_NFS_V4_2
22 #include "nfs42.h"
23 #endif
24 
25 #define NFSDBG_FACILITY		NFSDBG_FILE
26 
27 static int
28 nfs4_file_open(struct inode *inode, struct file *filp)
29 {
30 	struct nfs_open_context *ctx;
31 	struct dentry *dentry = file_dentry(filp);
32 	struct dentry *parent = NULL;
33 	struct inode *dir;
34 	unsigned openflags = filp->f_flags;
35 	struct iattr attr;
36 	int err;
37 
38 	/*
39 	 * If no cached dentry exists or if it's negative, NFSv4 handled the
40 	 * opens in ->lookup() or ->create().
41 	 *
42 	 * We only get this far for a cached positive dentry.  We skipped
43 	 * revalidation, so handle it here by dropping the dentry and returning
44 	 * -EOPENSTALE.  The VFS will retry the lookup/create/open.
45 	 */
46 
47 	dprintk("NFS: open file(%pd2)\n", dentry);
48 
49 	err = nfs_check_flags(openflags);
50 	if (err)
51 		return err;
52 
53 	if ((openflags & O_ACCMODE) == 3)
54 		return nfs_open(inode, filp);
55 
56 	/* We can't create new files here */
57 	openflags &= ~(O_CREAT|O_EXCL);
58 
59 	parent = dget_parent(dentry);
60 	dir = d_inode(parent);
61 
62 	ctx = alloc_nfs_open_context(file_dentry(filp), filp->f_mode, filp);
63 	err = PTR_ERR(ctx);
64 	if (IS_ERR(ctx))
65 		goto out;
66 
67 	attr.ia_valid = ATTR_OPEN;
68 	if (openflags & O_TRUNC) {
69 		attr.ia_valid |= ATTR_SIZE;
70 		attr.ia_size = 0;
71 		filemap_write_and_wait(inode->i_mapping);
72 	}
73 
74 	inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr, NULL);
75 	if (IS_ERR(inode)) {
76 		err = PTR_ERR(inode);
77 		switch (err) {
78 		default:
79 			goto out_put_ctx;
80 		case -ENOENT:
81 		case -ESTALE:
82 		case -EISDIR:
83 		case -ENOTDIR:
84 		case -ELOOP:
85 			goto out_drop;
86 		}
87 	}
88 	if (inode != d_inode(dentry))
89 		goto out_drop;
90 
91 	nfs_file_set_open_context(filp, ctx);
92 	nfs_fscache_open_file(inode, filp);
93 	err = 0;
94 
95 out_put_ctx:
96 	put_nfs_open_context(ctx);
97 out:
98 	dput(parent);
99 	return err;
100 
101 out_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  */
110 static int
111 nfs4_file_flush(struct file *file, fl_owner_t id)
112 {
113 	struct inode	*inode = file_inode(file);
114 	errseq_t since;
115 
116 	dprintk("NFS: flush(%pD2)\n", file);
117 
118 	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
119 	if ((file->f_mode & FMODE_WRITE) == 0)
120 		return 0;
121 
122 	/*
123 	 * If we're holding a write delegation, then check if we're required
124 	 * to flush the i/o on close. If not, then just start the i/o now.
125 	 */
126 	if (!nfs4_delegation_flush_on_close(inode))
127 		return filemap_fdatawrite(file->f_mapping);
128 
129 	/* Flush writes to the server and return any errors */
130 	since = filemap_sample_wb_err(file->f_mapping);
131 	nfs_wb_all(inode);
132 	return filemap_check_wb_err(file->f_mapping, since);
133 }
134 
135 #ifdef CONFIG_NFS_V4_2
136 static ssize_t __nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
137 				      struct file *file_out, loff_t pos_out,
138 				      size_t count, unsigned int flags)
139 {
140 	struct nfs42_copy_notify_res *cn_resp = NULL;
141 	struct nl4_server *nss = NULL;
142 	nfs4_stateid *cnrs = NULL;
143 	ssize_t ret;
144 	bool sync = false;
145 
146 	/* Only offload copy if superblock is the same */
147 	if (file_in->f_op != &nfs4_file_operations)
148 		return -EXDEV;
149 	if (!nfs_server_capable(file_inode(file_out), NFS_CAP_COPY) ||
150 	    !nfs_server_capable(file_inode(file_in), NFS_CAP_COPY))
151 		return -EOPNOTSUPP;
152 	if (file_inode(file_in) == file_inode(file_out))
153 		return -EOPNOTSUPP;
154 	/* if the copy size if smaller than 2 RPC payloads, make it
155 	 * synchronous
156 	 */
157 	if (count <= 2 * NFS_SERVER(file_inode(file_in))->rsize)
158 		sync = true;
159 retry:
160 	if (!nfs42_files_from_same_server(file_in, file_out)) {
161 		/*
162 		 * for inter copy, if copy size is too small
163 		 * then fallback to generic copy.
164 		 */
165 		if (sync)
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 != -EOPNOTSUPP)
213 			return ret;
214 		fallthrough;
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 static struct file *__nfs42_ssc_open(struct vfsmount *ss_mnt,
318 		struct nfs_fh *src_fh, nfs4_stateid *stateid)
319 {
320 	struct nfs_fattr fattr;
321 	struct file *filep, *res;
322 	struct nfs_server *server;
323 	struct inode *r_ino = NULL;
324 	struct nfs_open_context *ctx;
325 	struct nfs4_state_owner *sp;
326 	char *read_name = NULL;
327 	int len, status = 0;
328 
329 	server = NFS_SERVER(ss_mnt->mnt_root->d_inode);
330 
331 	nfs_fattr_init(&fattr);
332 
333 	status = nfs4_proc_getattr(server, src_fh, &fattr, NULL, NULL);
334 	if (status < 0) {
335 		res = ERR_PTR(status);
336 		goto out;
337 	}
338 
339 	res = ERR_PTR(-ENOMEM);
340 	len = strlen(SSC_READ_NAME_BODY) + 16;
341 	read_name = kzalloc(len, GFP_NOFS);
342 	if (read_name == NULL)
343 		goto out;
344 	snprintf(read_name, len, SSC_READ_NAME_BODY, read_name_gen++);
345 
346 	r_ino = nfs_fhget(ss_mnt->mnt_root->d_inode->i_sb, src_fh, &fattr,
347 			NULL);
348 	if (IS_ERR(r_ino)) {
349 		res = ERR_CAST(r_ino);
350 		goto out_free_name;
351 	}
352 
353 	filep = alloc_file_pseudo(r_ino, ss_mnt, read_name, FMODE_READ,
354 				     r_ino->i_fop);
355 	if (IS_ERR(filep)) {
356 		res = ERR_CAST(filep);
357 		goto out_free_name;
358 	}
359 	filep->f_mode |= FMODE_READ;
360 
361 	ctx = alloc_nfs_open_context(filep->f_path.dentry, filep->f_mode,
362 					filep);
363 	if (IS_ERR(ctx)) {
364 		res = ERR_CAST(ctx);
365 		goto out_filep;
366 	}
367 
368 	res = ERR_PTR(-EINVAL);
369 	sp = nfs4_get_state_owner(server, ctx->cred, GFP_KERNEL);
370 	if (sp == NULL)
371 		goto out_ctx;
372 
373 	ctx->state = nfs4_get_open_state(r_ino, sp);
374 	if (ctx->state == NULL)
375 		goto out_stateowner;
376 
377 	set_bit(NFS_SRV_SSC_COPY_STATE, &ctx->state->flags);
378 	memcpy(&ctx->state->open_stateid.other, &stateid->other,
379 	       NFS4_STATEID_OTHER_SIZE);
380 	update_open_stateid(ctx->state, stateid, NULL, filep->f_mode);
381 	set_bit(NFS_OPEN_STATE, &ctx->state->flags);
382 
383 	nfs_file_set_open_context(filep, ctx);
384 	put_nfs_open_context(ctx);
385 
386 	file_ra_state_init(&filep->f_ra, filep->f_mapping->host->i_mapping);
387 	res = filep;
388 out_free_name:
389 	kfree(read_name);
390 out:
391 	return res;
392 out_stateowner:
393 	nfs4_put_state_owner(sp);
394 out_ctx:
395 	put_nfs_open_context(ctx);
396 out_filep:
397 	fput(filep);
398 	goto out_free_name;
399 }
400 
401 static void __nfs42_ssc_close(struct file *filep)
402 {
403 	struct nfs_open_context *ctx = nfs_file_open_context(filep);
404 
405 	ctx->state->flags = 0;
406 }
407 
408 static const struct nfs4_ssc_client_ops nfs4_ssc_clnt_ops_tbl = {
409 	.sco_open = __nfs42_ssc_open,
410 	.sco_close = __nfs42_ssc_close,
411 };
412 
413 /**
414  * nfs42_ssc_register_ops - Wrapper to register NFS_V4 ops in nfs_common
415  *
416  * Return values:
417  *   None
418  */
419 void nfs42_ssc_register_ops(void)
420 {
421 	nfs42_ssc_register(&nfs4_ssc_clnt_ops_tbl);
422 }
423 
424 /**
425  * nfs42_ssc_unregister_ops - wrapper to un-register NFS_V4 ops in nfs_common
426  *
427  * Return values:
428  *   None.
429  */
430 void nfs42_ssc_unregister_ops(void)
431 {
432 	nfs42_ssc_unregister(&nfs4_ssc_clnt_ops_tbl);
433 }
434 #endif /* CONFIG_NFS_V4_2 */
435 
436 static int nfs4_setlease(struct file *file, long arg, struct file_lock **lease,
437 			 void **priv)
438 {
439 	return nfs4_proc_setlease(file, arg, lease, priv);
440 }
441 
442 const struct file_operations nfs4_file_operations = {
443 	.read_iter	= nfs_file_read,
444 	.write_iter	= nfs_file_write,
445 	.mmap		= nfs_file_mmap,
446 	.open		= nfs4_file_open,
447 	.flush		= nfs4_file_flush,
448 	.release	= nfs_file_release,
449 	.fsync		= nfs_file_fsync,
450 	.lock		= nfs_lock,
451 	.flock		= nfs_flock,
452 	.splice_read	= generic_file_splice_read,
453 	.splice_write	= iter_file_splice_write,
454 	.check_flags	= nfs_check_flags,
455 	.setlease	= nfs4_setlease,
456 #ifdef CONFIG_NFS_V4_2
457 	.copy_file_range = nfs4_copy_file_range,
458 	.llseek		= nfs4_file_llseek,
459 	.fallocate	= nfs42_fallocate,
460 	.remap_file_range = nfs42_remap_file_range,
461 #else
462 	.llseek		= nfs_file_llseek,
463 #endif
464 };
465