xref: /openbmc/linux/fs/nfs/nfs4file.c (revision c059ee9d)
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 	fmode_t f_mode;
36 	struct iattr attr;
37 	int err;
38 
39 	/*
40 	 * If no cached dentry exists or if it's negative, NFSv4 handled the
41 	 * opens in ->lookup() or ->create().
42 	 *
43 	 * We only get this far for a cached positive dentry.  We skipped
44 	 * revalidation, so handle it here by dropping the dentry and returning
45 	 * -EOPENSTALE.  The VFS will retry the lookup/create/open.
46 	 */
47 
48 	dprintk("NFS: open file(%pd2)\n", dentry);
49 
50 	err = nfs_check_flags(openflags);
51 	if (err)
52 		return err;
53 
54 	f_mode = filp->f_mode;
55 	if ((openflags & O_ACCMODE) == 3)
56 		f_mode |= flags_to_mode(openflags);
57 
58 	/* We can't create new files here */
59 	openflags &= ~(O_CREAT|O_EXCL);
60 
61 	parent = dget_parent(dentry);
62 	dir = d_inode(parent);
63 
64 	ctx = alloc_nfs_open_context(file_dentry(filp), f_mode, filp);
65 	err = PTR_ERR(ctx);
66 	if (IS_ERR(ctx))
67 		goto out;
68 
69 	attr.ia_valid = ATTR_OPEN;
70 	if (openflags & O_TRUNC) {
71 		attr.ia_valid |= ATTR_SIZE;
72 		attr.ia_size = 0;
73 		filemap_write_and_wait(inode->i_mapping);
74 	}
75 
76 	inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, &attr, NULL);
77 	if (IS_ERR(inode)) {
78 		err = PTR_ERR(inode);
79 		switch (err) {
80 		default:
81 			goto out_put_ctx;
82 		case -ENOENT:
83 		case -ESTALE:
84 		case -EISDIR:
85 		case -ENOTDIR:
86 		case -ELOOP:
87 			goto out_drop;
88 		}
89 	}
90 	if (inode != d_inode(dentry))
91 		goto out_drop;
92 
93 	nfs_file_set_open_context(filp, ctx);
94 	nfs_fscache_open_file(inode, filp);
95 	err = 0;
96 	filp->f_mode |= FMODE_CAN_ODIRECT;
97 
98 out_put_ctx:
99 	put_nfs_open_context(ctx);
100 out:
101 	dput(parent);
102 	return err;
103 
104 out_drop:
105 	d_drop(dentry);
106 	err = -EOPENSTALE;
107 	goto out_put_ctx;
108 }
109 
110 /*
111  * Flush all dirty pages, and check for write errors.
112  */
113 static int
114 nfs4_file_flush(struct file *file, fl_owner_t id)
115 {
116 	struct inode	*inode = file_inode(file);
117 	errseq_t since;
118 
119 	dprintk("NFS: flush(%pD2)\n", file);
120 
121 	nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
122 	if ((file->f_mode & FMODE_WRITE) == 0)
123 		return 0;
124 
125 	/*
126 	 * If we're holding a write delegation, then check if we're required
127 	 * to flush the i/o on close. If not, then just start the i/o now.
128 	 */
129 	if (!nfs4_delegation_flush_on_close(inode))
130 		return filemap_fdatawrite(file->f_mapping);
131 
132 	/* Flush writes to the server and return any errors */
133 	since = filemap_sample_wb_err(file->f_mapping);
134 	nfs_wb_all(inode);
135 	return filemap_check_wb_err(file->f_mapping, since);
136 }
137 
138 #ifdef CONFIG_NFS_V4_2
139 static ssize_t __nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
140 				      struct file *file_out, loff_t pos_out,
141 				      size_t count, unsigned int flags)
142 {
143 	struct nfs42_copy_notify_res *cn_resp = NULL;
144 	struct nl4_server *nss = NULL;
145 	nfs4_stateid *cnrs = NULL;
146 	ssize_t ret;
147 	bool sync = false;
148 
149 	/* Only offload copy if superblock is the same */
150 	if (file_in->f_op != &nfs4_file_operations)
151 		return -EXDEV;
152 	if (!nfs_server_capable(file_inode(file_out), NFS_CAP_COPY) ||
153 	    !nfs_server_capable(file_inode(file_in), NFS_CAP_COPY))
154 		return -EOPNOTSUPP;
155 	if (file_inode(file_in) == file_inode(file_out))
156 		return -EOPNOTSUPP;
157 	/* if the copy size if smaller than 2 RPC payloads, make it
158 	 * synchronous
159 	 */
160 	if (count <= 2 * NFS_SERVER(file_inode(file_in))->rsize)
161 		sync = true;
162 retry:
163 	if (!nfs42_files_from_same_server(file_in, file_out)) {
164 		/*
165 		 * for inter copy, if copy size is too small
166 		 * then fallback to generic copy.
167 		 */
168 		if (sync)
169 			return -EOPNOTSUPP;
170 		cn_resp = kzalloc(sizeof(struct nfs42_copy_notify_res),
171 				  GFP_KERNEL);
172 		if (unlikely(cn_resp == NULL))
173 			return -ENOMEM;
174 
175 		ret = nfs42_proc_copy_notify(file_in, file_out, cn_resp);
176 		if (ret) {
177 			ret = -EOPNOTSUPP;
178 			goto out;
179 		}
180 		nss = &cn_resp->cnr_src;
181 		cnrs = &cn_resp->cnr_stateid;
182 	}
183 	ret = nfs42_proc_copy(file_in, pos_in, file_out, pos_out, count,
184 				nss, cnrs, sync);
185 out:
186 	kfree(cn_resp);
187 
188 	if (ret == -EAGAIN)
189 		goto retry;
190 	return ret;
191 }
192 
193 static ssize_t nfs4_copy_file_range(struct file *file_in, loff_t pos_in,
194 				    struct file *file_out, loff_t pos_out,
195 				    size_t count, unsigned int flags)
196 {
197 	ssize_t ret;
198 
199 	ret = __nfs4_copy_file_range(file_in, pos_in, file_out, pos_out, count,
200 				     flags);
201 	if (ret == -EOPNOTSUPP || ret == -EXDEV)
202 		ret = generic_copy_file_range(file_in, pos_in, file_out,
203 					      pos_out, count, flags);
204 	return ret;
205 }
206 
207 static loff_t nfs4_file_llseek(struct file *filep, loff_t offset, int whence)
208 {
209 	loff_t ret;
210 
211 	switch (whence) {
212 	case SEEK_HOLE:
213 	case SEEK_DATA:
214 		ret = nfs42_proc_llseek(filep, offset, whence);
215 		if (ret != -EOPNOTSUPP)
216 			return ret;
217 		fallthrough;
218 	default:
219 		return nfs_file_llseek(filep, offset, whence);
220 	}
221 }
222 
223 static long nfs42_fallocate(struct file *filep, int mode, loff_t offset, loff_t len)
224 {
225 	struct inode *inode = file_inode(filep);
226 	long ret;
227 
228 	if (!S_ISREG(inode->i_mode))
229 		return -EOPNOTSUPP;
230 
231 	if ((mode != 0) && (mode != (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)))
232 		return -EOPNOTSUPP;
233 
234 	ret = inode_newsize_ok(inode, offset + len);
235 	if (ret < 0)
236 		return ret;
237 
238 	if (mode & FALLOC_FL_PUNCH_HOLE)
239 		return nfs42_proc_deallocate(filep, offset, len);
240 	return nfs42_proc_allocate(filep, offset, len);
241 }
242 
243 static loff_t nfs42_remap_file_range(struct file *src_file, loff_t src_off,
244 		struct file *dst_file, loff_t dst_off, loff_t count,
245 		unsigned int remap_flags)
246 {
247 	struct inode *dst_inode = file_inode(dst_file);
248 	struct nfs_server *server = NFS_SERVER(dst_inode);
249 	struct inode *src_inode = file_inode(src_file);
250 	unsigned int bs = server->clone_blksize;
251 	bool same_inode = false;
252 	int ret;
253 
254 	/* NFS does not support deduplication. */
255 	if (remap_flags & REMAP_FILE_DEDUP)
256 		return -EOPNOTSUPP;
257 
258 	if (remap_flags & ~REMAP_FILE_ADVISORY)
259 		return -EINVAL;
260 
261 	if (IS_SWAPFILE(dst_inode) || IS_SWAPFILE(src_inode))
262 		return -ETXTBSY;
263 
264 	/* check alignment w.r.t. clone_blksize */
265 	ret = -EINVAL;
266 	if (bs) {
267 		if (!IS_ALIGNED(src_off, bs) || !IS_ALIGNED(dst_off, bs))
268 			goto out;
269 		if (!IS_ALIGNED(count, bs) && i_size_read(src_inode) != (src_off + count))
270 			goto out;
271 	}
272 
273 	if (src_inode == dst_inode)
274 		same_inode = true;
275 
276 	/* XXX: do we lock at all? what if server needs CB_RECALL_LAYOUT? */
277 	if (same_inode) {
278 		inode_lock(src_inode);
279 	} else if (dst_inode < src_inode) {
280 		inode_lock_nested(dst_inode, I_MUTEX_PARENT);
281 		inode_lock_nested(src_inode, I_MUTEX_CHILD);
282 	} else {
283 		inode_lock_nested(src_inode, I_MUTEX_PARENT);
284 		inode_lock_nested(dst_inode, I_MUTEX_CHILD);
285 	}
286 
287 	/* flush all pending writes on both src and dst so that server
288 	 * has the latest data */
289 	ret = nfs_sync_inode(src_inode);
290 	if (ret)
291 		goto out_unlock;
292 	ret = nfs_sync_inode(dst_inode);
293 	if (ret)
294 		goto out_unlock;
295 
296 	ret = nfs42_proc_clone(src_file, dst_file, src_off, dst_off, count);
297 
298 	/* truncate inode page cache of the dst range so that future reads can fetch
299 	 * new data from server */
300 	if (!ret)
301 		truncate_inode_pages_range(&dst_inode->i_data, dst_off, dst_off + count - 1);
302 
303 out_unlock:
304 	if (same_inode) {
305 		inode_unlock(src_inode);
306 	} else if (dst_inode < src_inode) {
307 		inode_unlock(src_inode);
308 		inode_unlock(dst_inode);
309 	} else {
310 		inode_unlock(dst_inode);
311 		inode_unlock(src_inode);
312 	}
313 out:
314 	return ret < 0 ? ret : count;
315 }
316 
317 static int read_name_gen = 1;
318 #define SSC_READ_NAME_BODY "ssc_read_%d"
319 
320 static struct file *__nfs42_ssc_open(struct vfsmount *ss_mnt,
321 		struct nfs_fh *src_fh, nfs4_stateid *stateid)
322 {
323 	struct nfs_fattr *fattr = nfs_alloc_fattr();
324 	struct file *filep, *res;
325 	struct nfs_server *server;
326 	struct inode *r_ino = NULL;
327 	struct nfs_open_context *ctx;
328 	struct nfs4_state_owner *sp;
329 	char *read_name = NULL;
330 	int len, status = 0;
331 
332 	server = NFS_SB(ss_mnt->mnt_sb);
333 
334 	if (!fattr)
335 		return ERR_PTR(-ENOMEM);
336 
337 	status = nfs4_proc_getattr(server, src_fh, fattr, NULL);
338 	if (status < 0) {
339 		res = ERR_PTR(status);
340 		goto out;
341 	}
342 
343 	res = ERR_PTR(-ENOMEM);
344 	len = strlen(SSC_READ_NAME_BODY) + 16;
345 	read_name = kzalloc(len, GFP_KERNEL);
346 	if (read_name == NULL)
347 		goto out;
348 	snprintf(read_name, len, SSC_READ_NAME_BODY, read_name_gen++);
349 
350 	r_ino = nfs_fhget(ss_mnt->mnt_sb, src_fh, fattr);
351 	if (IS_ERR(r_ino)) {
352 		res = ERR_CAST(r_ino);
353 		goto out_free_name;
354 	}
355 
356 	filep = alloc_file_pseudo(r_ino, ss_mnt, read_name, O_RDONLY,
357 				     r_ino->i_fop);
358 	if (IS_ERR(filep)) {
359 		res = ERR_CAST(filep);
360 		goto out_free_name;
361 	}
362 
363 	ctx = alloc_nfs_open_context(filep->f_path.dentry, filep->f_mode,
364 					filep);
365 	if (IS_ERR(ctx)) {
366 		res = ERR_CAST(ctx);
367 		goto out_filep;
368 	}
369 
370 	res = ERR_PTR(-EINVAL);
371 	sp = nfs4_get_state_owner(server, ctx->cred, GFP_KERNEL);
372 	if (sp == NULL)
373 		goto out_ctx;
374 
375 	ctx->state = nfs4_get_open_state(r_ino, sp);
376 	if (ctx->state == NULL)
377 		goto out_stateowner;
378 
379 	set_bit(NFS_SRV_SSC_COPY_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 	set_bit(NFS_OPEN_STATE, &ctx->state->flags);
384 
385 	nfs_file_set_open_context(filep, ctx);
386 	put_nfs_open_context(ctx);
387 
388 	file_ra_state_init(&filep->f_ra, filep->f_mapping->host->i_mapping);
389 	res = filep;
390 out_free_name:
391 	kfree(read_name);
392 out:
393 	nfs_free_fattr(fattr);
394 	return res;
395 out_stateowner:
396 	nfs4_put_state_owner(sp);
397 out_ctx:
398 	put_nfs_open_context(ctx);
399 out_filep:
400 	fput(filep);
401 	goto out_free_name;
402 }
403 
404 static void __nfs42_ssc_close(struct file *filep)
405 {
406 	struct nfs_open_context *ctx = nfs_file_open_context(filep);
407 
408 	ctx->state->flags = 0;
409 }
410 
411 static const struct nfs4_ssc_client_ops nfs4_ssc_clnt_ops_tbl = {
412 	.sco_open = __nfs42_ssc_open,
413 	.sco_close = __nfs42_ssc_close,
414 };
415 
416 /**
417  * nfs42_ssc_register_ops - Wrapper to register NFS_V4 ops in nfs_common
418  *
419  * Return values:
420  *   None
421  */
422 void nfs42_ssc_register_ops(void)
423 {
424 	nfs42_ssc_register(&nfs4_ssc_clnt_ops_tbl);
425 }
426 
427 /**
428  * nfs42_ssc_unregister_ops - wrapper to un-register NFS_V4 ops in nfs_common
429  *
430  * Return values:
431  *   None.
432  */
433 void nfs42_ssc_unregister_ops(void)
434 {
435 	nfs42_ssc_unregister(&nfs4_ssc_clnt_ops_tbl);
436 }
437 #endif /* CONFIG_NFS_V4_2 */
438 
439 static int nfs4_setlease(struct file *file, long arg, struct file_lock **lease,
440 			 void **priv)
441 {
442 	return nfs4_proc_setlease(file, arg, lease, priv);
443 }
444 
445 const struct file_operations nfs4_file_operations = {
446 	.read_iter	= nfs_file_read,
447 	.write_iter	= nfs_file_write,
448 	.mmap		= nfs_file_mmap,
449 	.open		= nfs4_file_open,
450 	.flush		= nfs4_file_flush,
451 	.release	= nfs_file_release,
452 	.fsync		= nfs_file_fsync,
453 	.lock		= nfs_lock,
454 	.flock		= nfs_flock,
455 	.splice_read	= generic_file_splice_read,
456 	.splice_write	= iter_file_splice_write,
457 	.check_flags	= nfs_check_flags,
458 	.setlease	= nfs4_setlease,
459 #ifdef CONFIG_NFS_V4_2
460 	.copy_file_range = nfs4_copy_file_range,
461 	.llseek		= nfs4_file_llseek,
462 	.fallocate	= nfs42_fallocate,
463 	.remap_file_range = nfs42_remap_file_range,
464 #else
465 	.llseek		= nfs_file_llseek,
466 #endif
467 };
468