• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  linux/fs/nfs/dir.c
4  *
5  *  Copyright (C) 1992  Rick Sladkey
6  *
7  *  nfs directory handling functions
8  *
9  * 10 Apr 1996	Added silly rename for unlink	--okir
10  * 28 Sep 1996	Improved directory cache --okir
11  * 23 Aug 1997  Claus Heine claus@momo.math.rwth-aachen.de
12  *              Re-implemented silly rename for unlink, newly implemented
13  *              silly rename for nfs_rename() following the suggestions
14  *              of Olaf Kirch (okir) found in this file.
15  *              Following Linus comments on my original hack, this version
16  *              depends only on the dcache stuff and doesn't touch the inode
17  *              layer (iput() and friends).
18  *  6 Jun 1999	Cache readdir lookups in the page cache. -DaveM
19  */
20 
21 #include <linux/module.h>
22 #include <linux/time.h>
23 #include <linux/errno.h>
24 #include <linux/stat.h>
25 #include <linux/fcntl.h>
26 #include <linux/string.h>
27 #include <linux/kernel.h>
28 #include <linux/slab.h>
29 #include <linux/mm.h>
30 #include <linux/sunrpc/clnt.h>
31 #include <linux/nfs_fs.h>
32 #include <linux/nfs_mount.h>
33 #include <linux/pagemap.h>
34 #include <linux/pagevec.h>
35 #include <linux/namei.h>
36 #include <linux/mount.h>
37 #include <linux/swap.h>
38 #include <linux/sched.h>
39 #include <linux/kmemleak.h>
40 #include <linux/xattr.h>
41 
42 #include "delegation.h"
43 #include "iostat.h"
44 #include "internal.h"
45 #include "fscache.h"
46 
47 #include "nfstrace.h"
48 
49 /* #define NFS_DEBUG_VERBOSE 1 */
50 
51 static int nfs_opendir(struct inode *, struct file *);
52 static int nfs_closedir(struct inode *, struct file *);
53 static int nfs_readdir(struct file *, struct dir_context *);
54 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
55 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
56 static void nfs_readdir_clear_array(struct page*);
57 
58 const struct file_operations nfs_dir_operations = {
59 	.llseek		= nfs_llseek_dir,
60 	.read		= generic_read_dir,
61 	.iterate_shared	= nfs_readdir,
62 	.open		= nfs_opendir,
63 	.release	= nfs_closedir,
64 	.fsync		= nfs_fsync_dir,
65 };
66 
67 const struct address_space_operations nfs_dir_aops = {
68 	.freepage = nfs_readdir_clear_array,
69 };
70 
alloc_nfs_open_dir_context(struct inode * dir,const struct cred * cred)71 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, const struct cred *cred)
72 {
73 	struct nfs_inode *nfsi = NFS_I(dir);
74 	struct nfs_open_dir_context *ctx;
75 	ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
76 	if (ctx != NULL) {
77 		ctx->duped = 0;
78 		ctx->attr_gencount = nfsi->attr_gencount;
79 		ctx->dir_cookie = 0;
80 		ctx->dup_cookie = 0;
81 		ctx->cred = get_cred(cred);
82 		spin_lock(&dir->i_lock);
83 		if (list_empty(&nfsi->open_files) &&
84 		    (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
85 			nfsi->cache_validity |= NFS_INO_INVALID_DATA |
86 				NFS_INO_REVAL_FORCED;
87 		list_add(&ctx->list, &nfsi->open_files);
88 		spin_unlock(&dir->i_lock);
89 		return ctx;
90 	}
91 	return  ERR_PTR(-ENOMEM);
92 }
93 
put_nfs_open_dir_context(struct inode * dir,struct nfs_open_dir_context * ctx)94 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
95 {
96 	spin_lock(&dir->i_lock);
97 	list_del(&ctx->list);
98 	spin_unlock(&dir->i_lock);
99 	put_cred(ctx->cred);
100 	kfree(ctx);
101 }
102 
103 /*
104  * Open file
105  */
106 static int
nfs_opendir(struct inode * inode,struct file * filp)107 nfs_opendir(struct inode *inode, struct file *filp)
108 {
109 	int res = 0;
110 	struct nfs_open_dir_context *ctx;
111 
112 	dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
113 
114 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
115 
116 	ctx = alloc_nfs_open_dir_context(inode, current_cred());
117 	if (IS_ERR(ctx)) {
118 		res = PTR_ERR(ctx);
119 		goto out;
120 	}
121 	filp->private_data = ctx;
122 out:
123 	return res;
124 }
125 
126 static int
nfs_closedir(struct inode * inode,struct file * filp)127 nfs_closedir(struct inode *inode, struct file *filp)
128 {
129 	put_nfs_open_dir_context(file_inode(filp), filp->private_data);
130 	return 0;
131 }
132 
133 struct nfs_cache_array_entry {
134 	u64 cookie;
135 	u64 ino;
136 	struct qstr string;
137 	unsigned char d_type;
138 };
139 
140 struct nfs_cache_array {
141 	int size;
142 	int eof_index;
143 	u64 last_cookie;
144 	struct nfs_cache_array_entry array[];
145 };
146 
147 typedef struct {
148 	struct file	*file;
149 	struct page	*page;
150 	struct dir_context *ctx;
151 	unsigned long	page_index;
152 	u64		*dir_cookie;
153 	u64		last_cookie;
154 	loff_t		current_index;
155 	loff_t		prev_index;
156 
157 	unsigned long	dir_verifier;
158 	unsigned long	timestamp;
159 	unsigned long	gencount;
160 	unsigned int	cache_entry_index;
161 	bool plus;
162 	bool eof;
163 } nfs_readdir_descriptor_t;
164 
165 static
nfs_readdir_init_array(struct page * page)166 void nfs_readdir_init_array(struct page *page)
167 {
168 	struct nfs_cache_array *array;
169 
170 	array = kmap_atomic(page);
171 	memset(array, 0, sizeof(struct nfs_cache_array));
172 	array->eof_index = -1;
173 	kunmap_atomic(array);
174 }
175 
176 /*
177  * we are freeing strings created by nfs_add_to_readdir_array()
178  */
179 static
nfs_readdir_clear_array(struct page * page)180 void nfs_readdir_clear_array(struct page *page)
181 {
182 	struct nfs_cache_array *array;
183 	int i;
184 
185 	array = kmap_atomic(page);
186 	for (i = 0; i < array->size; i++)
187 		kfree(array->array[i].string.name);
188 	array->size = 0;
189 	kunmap_atomic(array);
190 }
191 
192 /*
193  * the caller is responsible for freeing qstr.name
194  * when called by nfs_readdir_add_to_array, the strings will be freed in
195  * nfs_clear_readdir_array()
196  */
197 static
nfs_readdir_make_qstr(struct qstr * string,const char * name,unsigned int len)198 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
199 {
200 	string->len = len;
201 	string->name = kmemdup_nul(name, len, GFP_KERNEL);
202 	if (string->name == NULL)
203 		return -ENOMEM;
204 	/*
205 	 * Avoid a kmemleak false positive. The pointer to the name is stored
206 	 * in a page cache page which kmemleak does not scan.
207 	 */
208 	kmemleak_not_leak(string->name);
209 	string->hash = full_name_hash(NULL, name, len);
210 	return 0;
211 }
212 
213 static
nfs_readdir_add_to_array(struct nfs_entry * entry,struct page * page)214 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
215 {
216 	struct nfs_cache_array *array = kmap(page);
217 	struct nfs_cache_array_entry *cache_entry;
218 	int ret;
219 
220 	cache_entry = &array->array[array->size];
221 
222 	/* Check that this entry lies within the page bounds */
223 	ret = -ENOSPC;
224 	if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
225 		goto out;
226 
227 	cache_entry->cookie = entry->prev_cookie;
228 	cache_entry->ino = entry->ino;
229 	cache_entry->d_type = entry->d_type;
230 	ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
231 	if (ret)
232 		goto out;
233 	array->last_cookie = entry->cookie;
234 	array->size++;
235 	if (entry->eof != 0)
236 		array->eof_index = array->size;
237 out:
238 	kunmap(page);
239 	return ret;
240 }
241 
242 static inline
is_32bit_api(void)243 int is_32bit_api(void)
244 {
245 #ifdef CONFIG_COMPAT
246 	return in_compat_syscall();
247 #else
248 	return (BITS_PER_LONG == 32);
249 #endif
250 }
251 
252 static
nfs_readdir_use_cookie(const struct file * filp)253 bool nfs_readdir_use_cookie(const struct file *filp)
254 {
255 	if ((filp->f_mode & FMODE_32BITHASH) ||
256 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
257 		return false;
258 	return true;
259 }
260 
261 static
nfs_readdir_search_for_pos(struct nfs_cache_array * array,nfs_readdir_descriptor_t * desc)262 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
263 {
264 	loff_t diff = desc->ctx->pos - desc->current_index;
265 	unsigned int index;
266 
267 	if (diff < 0)
268 		goto out_eof;
269 	if (diff >= array->size) {
270 		if (array->eof_index >= 0)
271 			goto out_eof;
272 		return -EAGAIN;
273 	}
274 
275 	index = (unsigned int)diff;
276 	*desc->dir_cookie = array->array[index].cookie;
277 	desc->cache_entry_index = index;
278 	return 0;
279 out_eof:
280 	desc->eof = true;
281 	return -EBADCOOKIE;
282 }
283 
284 static bool
nfs_readdir_inode_mapping_valid(struct nfs_inode * nfsi)285 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
286 {
287 	if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
288 		return false;
289 	smp_rmb();
290 	return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
291 }
292 
293 static
nfs_readdir_search_for_cookie(struct nfs_cache_array * array,nfs_readdir_descriptor_t * desc)294 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
295 {
296 	int i;
297 	loff_t new_pos;
298 	int status = -EAGAIN;
299 
300 	for (i = 0; i < array->size; i++) {
301 		if (array->array[i].cookie == *desc->dir_cookie) {
302 			struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
303 			struct nfs_open_dir_context *ctx = desc->file->private_data;
304 
305 			new_pos = desc->current_index + i;
306 			if (ctx->attr_gencount != nfsi->attr_gencount ||
307 			    !nfs_readdir_inode_mapping_valid(nfsi)) {
308 				ctx->duped = 0;
309 				ctx->attr_gencount = nfsi->attr_gencount;
310 			} else if (new_pos < desc->prev_index) {
311 				if (ctx->duped > 0
312 				    && ctx->dup_cookie == *desc->dir_cookie) {
313 					if (printk_ratelimit()) {
314 						pr_notice("NFS: directory %pD2 contains a readdir loop."
315 								"Please contact your server vendor.  "
316 								"The file: %.*s has duplicate cookie %llu\n",
317 								desc->file, array->array[i].string.len,
318 								array->array[i].string.name, *desc->dir_cookie);
319 					}
320 					status = -ELOOP;
321 					goto out;
322 				}
323 				ctx->dup_cookie = *desc->dir_cookie;
324 				ctx->duped = -1;
325 			}
326 			if (nfs_readdir_use_cookie(desc->file))
327 				desc->ctx->pos = *desc->dir_cookie;
328 			else
329 				desc->ctx->pos = new_pos;
330 			desc->prev_index = new_pos;
331 			desc->cache_entry_index = i;
332 			return 0;
333 		}
334 	}
335 	if (array->eof_index >= 0) {
336 		status = -EBADCOOKIE;
337 		if (*desc->dir_cookie == array->last_cookie)
338 			desc->eof = true;
339 	}
340 out:
341 	return status;
342 }
343 
344 static
nfs_readdir_search_array(nfs_readdir_descriptor_t * desc)345 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
346 {
347 	struct nfs_cache_array *array;
348 	int status;
349 
350 	array = kmap(desc->page);
351 
352 	if (*desc->dir_cookie == 0)
353 		status = nfs_readdir_search_for_pos(array, desc);
354 	else
355 		status = nfs_readdir_search_for_cookie(array, desc);
356 
357 	if (status == -EAGAIN) {
358 		desc->last_cookie = array->last_cookie;
359 		desc->current_index += array->size;
360 		desc->page_index++;
361 	}
362 	kunmap(desc->page);
363 	return status;
364 }
365 
366 /* Fill a page with xdr information before transferring to the cache page */
367 static
nfs_readdir_xdr_filler(struct page ** pages,nfs_readdir_descriptor_t * desc,struct nfs_entry * entry,struct file * file,struct inode * inode)368 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
369 			struct nfs_entry *entry, struct file *file, struct inode *inode)
370 {
371 	struct nfs_open_dir_context *ctx = file->private_data;
372 	const struct cred *cred = ctx->cred;
373 	unsigned long	timestamp, gencount;
374 	int		error;
375 
376  again:
377 	timestamp = jiffies;
378 	gencount = nfs_inc_attr_generation_counter();
379 	desc->dir_verifier = nfs_save_change_attribute(inode);
380 	error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages,
381 					  NFS_SERVER(inode)->dtsize, desc->plus);
382 	if (error < 0) {
383 		/* We requested READDIRPLUS, but the server doesn't grok it */
384 		if (error == -ENOTSUPP && desc->plus) {
385 			NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
386 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
387 			desc->plus = false;
388 			goto again;
389 		}
390 		goto error;
391 	}
392 	desc->timestamp = timestamp;
393 	desc->gencount = gencount;
394 error:
395 	return error;
396 }
397 
xdr_decode(nfs_readdir_descriptor_t * desc,struct nfs_entry * entry,struct xdr_stream * xdr)398 static int xdr_decode(nfs_readdir_descriptor_t *desc,
399 		      struct nfs_entry *entry, struct xdr_stream *xdr)
400 {
401 	struct inode *inode = file_inode(desc->file);
402 	int error;
403 
404 	error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
405 	if (error)
406 		return error;
407 	entry->fattr->time_start = desc->timestamp;
408 	entry->fattr->gencount = desc->gencount;
409 	return 0;
410 }
411 
412 /* Match file and dirent using either filehandle or fileid
413  * Note: caller is responsible for checking the fsid
414  */
415 static
nfs_same_file(struct dentry * dentry,struct nfs_entry * entry)416 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
417 {
418 	struct inode *inode;
419 	struct nfs_inode *nfsi;
420 
421 	if (d_really_is_negative(dentry))
422 		return 0;
423 
424 	inode = d_inode(dentry);
425 	if (is_bad_inode(inode) || NFS_STALE(inode))
426 		return 0;
427 
428 	nfsi = NFS_I(inode);
429 	if (entry->fattr->fileid != nfsi->fileid)
430 		return 0;
431 	if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
432 		return 0;
433 	return 1;
434 }
435 
436 static
nfs_use_readdirplus(struct inode * dir,struct dir_context * ctx)437 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
438 {
439 	if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
440 		return false;
441 	if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
442 		return true;
443 	if (ctx->pos == 0)
444 		return true;
445 	return false;
446 }
447 
448 /*
449  * This function is called by the lookup and getattr code to request the
450  * use of readdirplus to accelerate any future lookups in the same
451  * directory.
452  */
nfs_advise_use_readdirplus(struct inode * dir)453 void nfs_advise_use_readdirplus(struct inode *dir)
454 {
455 	struct nfs_inode *nfsi = NFS_I(dir);
456 
457 	if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
458 	    !list_empty(&nfsi->open_files))
459 		set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
460 }
461 
462 /*
463  * This function is mainly for use by nfs_getattr().
464  *
465  * If this is an 'ls -l', we want to force use of readdirplus.
466  * Do this by checking if there is an active file descriptor
467  * and calling nfs_advise_use_readdirplus, then forcing a
468  * cache flush.
469  */
nfs_force_use_readdirplus(struct inode * dir)470 void nfs_force_use_readdirplus(struct inode *dir)
471 {
472 	struct nfs_inode *nfsi = NFS_I(dir);
473 
474 	if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
475 	    !list_empty(&nfsi->open_files)) {
476 		set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
477 		invalidate_mapping_pages(dir->i_mapping,
478 			nfsi->page_index + 1, -1);
479 	}
480 }
481 
482 static
nfs_prime_dcache(struct dentry * parent,struct nfs_entry * entry,unsigned long dir_verifier)483 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
484 		unsigned long dir_verifier)
485 {
486 	struct qstr filename = QSTR_INIT(entry->name, entry->len);
487 	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
488 	struct dentry *dentry;
489 	struct dentry *alias;
490 	struct inode *inode;
491 	int status;
492 
493 	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
494 		return;
495 	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
496 		return;
497 	if (filename.len == 0)
498 		return;
499 	/* Validate that the name doesn't contain any illegal '\0' */
500 	if (strnlen(filename.name, filename.len) != filename.len)
501 		return;
502 	/* ...or '/' */
503 	if (strnchr(filename.name, filename.len, '/'))
504 		return;
505 	if (filename.name[0] == '.') {
506 		if (filename.len == 1)
507 			return;
508 		if (filename.len == 2 && filename.name[1] == '.')
509 			return;
510 	}
511 	filename.hash = full_name_hash(parent, filename.name, filename.len);
512 
513 	dentry = d_lookup(parent, &filename);
514 again:
515 	if (!dentry) {
516 		dentry = d_alloc_parallel(parent, &filename, &wq);
517 		if (IS_ERR(dentry))
518 			return;
519 	}
520 	if (!d_in_lookup(dentry)) {
521 		/* Is there a mountpoint here? If so, just exit */
522 		if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
523 					&entry->fattr->fsid))
524 			goto out;
525 		if (nfs_same_file(dentry, entry)) {
526 			if (!entry->fh->size)
527 				goto out;
528 			nfs_set_verifier(dentry, dir_verifier);
529 			status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
530 			if (!status)
531 				nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
532 			goto out;
533 		} else {
534 			d_invalidate(dentry);
535 			dput(dentry);
536 			dentry = NULL;
537 			goto again;
538 		}
539 	}
540 	if (!entry->fh->size) {
541 		d_lookup_done(dentry);
542 		goto out;
543 	}
544 
545 	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
546 	alias = d_splice_alias(inode, dentry);
547 	d_lookup_done(dentry);
548 	if (alias) {
549 		if (IS_ERR(alias))
550 			goto out;
551 		dput(dentry);
552 		dentry = alias;
553 	}
554 	nfs_set_verifier(dentry, dir_verifier);
555 out:
556 	dput(dentry);
557 }
558 
559 /* Perform conversion from xdr to cache array */
560 static
nfs_readdir_page_filler(nfs_readdir_descriptor_t * desc,struct nfs_entry * entry,struct page ** xdr_pages,struct page * page,unsigned int buflen)561 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
562 				struct page **xdr_pages, struct page *page, unsigned int buflen)
563 {
564 	struct xdr_stream stream;
565 	struct xdr_buf buf;
566 	struct page *scratch;
567 	struct nfs_cache_array *array;
568 	unsigned int count = 0;
569 	int status;
570 
571 	scratch = alloc_page(GFP_KERNEL);
572 	if (scratch == NULL)
573 		return -ENOMEM;
574 
575 	if (buflen == 0)
576 		goto out_nopages;
577 
578 	xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
579 	xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
580 
581 	do {
582 		if (entry->label)
583 			entry->label->len = NFS4_MAXLABELLEN;
584 
585 		status = xdr_decode(desc, entry, &stream);
586 		if (status != 0) {
587 			if (status == -EAGAIN)
588 				status = 0;
589 			break;
590 		}
591 
592 		count++;
593 
594 		if (desc->plus)
595 			nfs_prime_dcache(file_dentry(desc->file), entry,
596 					desc->dir_verifier);
597 
598 		status = nfs_readdir_add_to_array(entry, page);
599 		if (status != 0)
600 			break;
601 	} while (!entry->eof);
602 
603 out_nopages:
604 	if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
605 		array = kmap(page);
606 		array->eof_index = array->size;
607 		status = 0;
608 		kunmap(page);
609 	}
610 
611 	put_page(scratch);
612 	return status;
613 }
614 
615 static
nfs_readdir_free_pages(struct page ** pages,unsigned int npages)616 void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
617 {
618 	unsigned int i;
619 	for (i = 0; i < npages; i++)
620 		put_page(pages[i]);
621 }
622 
623 /*
624  * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
625  * to nfs_readdir_free_pages()
626  */
627 static
nfs_readdir_alloc_pages(struct page ** pages,unsigned int npages)628 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
629 {
630 	unsigned int i;
631 
632 	for (i = 0; i < npages; i++) {
633 		struct page *page = alloc_page(GFP_KERNEL);
634 		if (page == NULL)
635 			goto out_freepages;
636 		pages[i] = page;
637 	}
638 	return 0;
639 
640 out_freepages:
641 	nfs_readdir_free_pages(pages, i);
642 	return -ENOMEM;
643 }
644 
645 static
nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t * desc,struct page * page,struct inode * inode)646 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
647 {
648 	struct page *pages[NFS_MAX_READDIR_PAGES];
649 	struct nfs_entry entry;
650 	struct file	*file = desc->file;
651 	struct nfs_cache_array *array;
652 	int status = -ENOMEM;
653 	unsigned int array_size = ARRAY_SIZE(pages);
654 
655 	nfs_readdir_init_array(page);
656 
657 	entry.prev_cookie = 0;
658 	entry.cookie = desc->last_cookie;
659 	entry.eof = 0;
660 	entry.fh = nfs_alloc_fhandle();
661 	entry.fattr = nfs_alloc_fattr();
662 	entry.server = NFS_SERVER(inode);
663 	if (entry.fh == NULL || entry.fattr == NULL)
664 		goto out;
665 
666 	entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
667 	if (IS_ERR(entry.label)) {
668 		status = PTR_ERR(entry.label);
669 		goto out;
670 	}
671 
672 	array = kmap(page);
673 
674 	status = nfs_readdir_alloc_pages(pages, array_size);
675 	if (status < 0)
676 		goto out_release_array;
677 	do {
678 		unsigned int pglen;
679 		status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
680 
681 		if (status < 0)
682 			break;
683 		pglen = status;
684 		status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
685 		if (status < 0) {
686 			if (status == -ENOSPC)
687 				status = 0;
688 			break;
689 		}
690 	} while (array->eof_index < 0);
691 
692 	nfs_readdir_free_pages(pages, array_size);
693 out_release_array:
694 	kunmap(page);
695 	nfs4_label_free(entry.label);
696 out:
697 	nfs_free_fattr(entry.fattr);
698 	nfs_free_fhandle(entry.fh);
699 	return status;
700 }
701 
702 /*
703  * Now we cache directories properly, by converting xdr information
704  * to an array that can be used for lookups later.  This results in
705  * fewer cache pages, since we can store more information on each page.
706  * We only need to convert from xdr once so future lookups are much simpler
707  */
708 static
nfs_readdir_filler(void * data,struct page * page)709 int nfs_readdir_filler(void *data, struct page* page)
710 {
711 	nfs_readdir_descriptor_t *desc = data;
712 	struct inode	*inode = file_inode(desc->file);
713 	int ret;
714 
715 	ret = nfs_readdir_xdr_to_array(desc, page, inode);
716 	if (ret < 0)
717 		goto error;
718 	SetPageUptodate(page);
719 
720 	if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
721 		/* Should never happen */
722 		nfs_zap_mapping(inode, inode->i_mapping);
723 	}
724 	unlock_page(page);
725 	return 0;
726  error:
727 	nfs_readdir_clear_array(page);
728 	unlock_page(page);
729 	return ret;
730 }
731 
732 static
cache_page_release(nfs_readdir_descriptor_t * desc)733 void cache_page_release(nfs_readdir_descriptor_t *desc)
734 {
735 	put_page(desc->page);
736 	desc->page = NULL;
737 }
738 
739 static
get_cache_page(nfs_readdir_descriptor_t * desc)740 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
741 {
742 	return read_cache_page(desc->file->f_mapping, desc->page_index,
743 			nfs_readdir_filler, desc);
744 }
745 
746 /*
747  * Returns 0 if desc->dir_cookie was found on page desc->page_index
748  * and locks the page to prevent removal from the page cache.
749  */
750 static
find_and_lock_cache_page(nfs_readdir_descriptor_t * desc)751 int find_and_lock_cache_page(nfs_readdir_descriptor_t *desc)
752 {
753 	struct inode *inode = file_inode(desc->file);
754 	struct nfs_inode *nfsi = NFS_I(inode);
755 	int res;
756 
757 	desc->page = get_cache_page(desc);
758 	if (IS_ERR(desc->page))
759 		return PTR_ERR(desc->page);
760 	res = lock_page_killable(desc->page);
761 	if (res != 0)
762 		goto error;
763 	res = -EAGAIN;
764 	if (desc->page->mapping != NULL) {
765 		res = nfs_readdir_search_array(desc);
766 		if (res == 0) {
767 			nfsi->page_index = desc->page_index;
768 			return 0;
769 		}
770 	}
771 	unlock_page(desc->page);
772 error:
773 	cache_page_release(desc);
774 	return res;
775 }
776 
777 /* Search for desc->dir_cookie from the beginning of the page cache */
778 static inline
readdir_search_pagecache(nfs_readdir_descriptor_t * desc)779 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
780 {
781 	int res;
782 
783 	if (desc->page_index == 0) {
784 		desc->current_index = 0;
785 		desc->prev_index = 0;
786 		desc->last_cookie = 0;
787 	}
788 	do {
789 		res = find_and_lock_cache_page(desc);
790 	} while (res == -EAGAIN);
791 	return res;
792 }
793 
794 /*
795  * Once we've found the start of the dirent within a page: fill 'er up...
796  */
797 static
nfs_do_filldir(nfs_readdir_descriptor_t * desc)798 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
799 {
800 	struct file	*file = desc->file;
801 	int i = 0;
802 	int res = 0;
803 	struct nfs_cache_array *array = NULL;
804 	struct nfs_open_dir_context *ctx = file->private_data;
805 
806 	array = kmap(desc->page);
807 	for (i = desc->cache_entry_index; i < array->size; i++) {
808 		struct nfs_cache_array_entry *ent;
809 
810 		ent = &array->array[i];
811 		if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
812 		    nfs_compat_user_ino64(ent->ino), ent->d_type)) {
813 			desc->eof = true;
814 			break;
815 		}
816 		if (i < (array->size-1))
817 			*desc->dir_cookie = array->array[i+1].cookie;
818 		else
819 			*desc->dir_cookie = array->last_cookie;
820 		if (nfs_readdir_use_cookie(file))
821 			desc->ctx->pos = *desc->dir_cookie;
822 		else
823 			desc->ctx->pos++;
824 		if (ctx->duped != 0)
825 			ctx->duped = 1;
826 	}
827 	if (array->eof_index >= 0)
828 		desc->eof = true;
829 
830 	kunmap(desc->page);
831 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
832 			(unsigned long long)*desc->dir_cookie, res);
833 	return res;
834 }
835 
836 /*
837  * If we cannot find a cookie in our cache, we suspect that this is
838  * because it points to a deleted file, so we ask the server to return
839  * whatever it thinks is the next entry. We then feed this to filldir.
840  * If all goes well, we should then be able to find our way round the
841  * cache on the next call to readdir_search_pagecache();
842  *
843  * NOTE: we cannot add the anonymous page to the pagecache because
844  *	 the data it contains might not be page aligned. Besides,
845  *	 we should already have a complete representation of the
846  *	 directory in the page cache by the time we get here.
847  */
848 static inline
uncached_readdir(nfs_readdir_descriptor_t * desc)849 int uncached_readdir(nfs_readdir_descriptor_t *desc)
850 {
851 	struct page	*page = NULL;
852 	int		status;
853 	struct inode *inode = file_inode(desc->file);
854 	struct nfs_open_dir_context *ctx = desc->file->private_data;
855 
856 	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
857 			(unsigned long long)*desc->dir_cookie);
858 
859 	page = alloc_page(GFP_HIGHUSER);
860 	if (!page) {
861 		status = -ENOMEM;
862 		goto out;
863 	}
864 
865 	desc->page_index = 0;
866 	desc->last_cookie = *desc->dir_cookie;
867 	desc->page = page;
868 	ctx->duped = 0;
869 
870 	status = nfs_readdir_xdr_to_array(desc, page, inode);
871 	if (status < 0)
872 		goto out_release;
873 
874 	status = nfs_do_filldir(desc);
875 
876  out_release:
877 	nfs_readdir_clear_array(desc->page);
878 	cache_page_release(desc);
879  out:
880 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
881 			__func__, status);
882 	return status;
883 }
884 
885 /* The file offset position represents the dirent entry number.  A
886    last cookie cache takes care of the common case of reading the
887    whole directory.
888  */
nfs_readdir(struct file * file,struct dir_context * ctx)889 static int nfs_readdir(struct file *file, struct dir_context *ctx)
890 {
891 	struct dentry	*dentry = file_dentry(file);
892 	struct inode	*inode = d_inode(dentry);
893 	struct nfs_open_dir_context *dir_ctx = file->private_data;
894 	nfs_readdir_descriptor_t my_desc = {
895 		.file = file,
896 		.ctx = ctx,
897 		.dir_cookie = &dir_ctx->dir_cookie,
898 		.plus = nfs_use_readdirplus(inode, ctx),
899 	},
900 			*desc = &my_desc;
901 	int res = 0;
902 
903 	dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
904 			file, (long long)ctx->pos);
905 	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
906 
907 	/*
908 	 * ctx->pos points to the dirent entry number.
909 	 * *desc->dir_cookie has the cookie for the next entry. We have
910 	 * to either find the entry with the appropriate number or
911 	 * revalidate the cookie.
912 	 */
913 	if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
914 		res = nfs_revalidate_mapping(inode, file->f_mapping);
915 	if (res < 0)
916 		goto out;
917 
918 	do {
919 		res = readdir_search_pagecache(desc);
920 
921 		if (res == -EBADCOOKIE) {
922 			res = 0;
923 			/* This means either end of directory */
924 			if (*desc->dir_cookie && !desc->eof) {
925 				/* Or that the server has 'lost' a cookie */
926 				res = uncached_readdir(desc);
927 				if (res == 0)
928 					continue;
929 			}
930 			break;
931 		}
932 		if (res == -ETOOSMALL && desc->plus) {
933 			clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
934 			nfs_zap_caches(inode);
935 			desc->page_index = 0;
936 			desc->plus = false;
937 			desc->eof = false;
938 			continue;
939 		}
940 		if (res < 0)
941 			break;
942 
943 		res = nfs_do_filldir(desc);
944 		unlock_page(desc->page);
945 		cache_page_release(desc);
946 		if (res < 0)
947 			break;
948 	} while (!desc->eof);
949 out:
950 	if (res > 0)
951 		res = 0;
952 	dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
953 	return res;
954 }
955 
nfs_llseek_dir(struct file * filp,loff_t offset,int whence)956 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
957 {
958 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
959 
960 	dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
961 			filp, offset, whence);
962 
963 	switch (whence) {
964 	default:
965 		return -EINVAL;
966 	case SEEK_SET:
967 		if (offset < 0)
968 			return -EINVAL;
969 		spin_lock(&filp->f_lock);
970 		break;
971 	case SEEK_CUR:
972 		if (offset == 0)
973 			return filp->f_pos;
974 		spin_lock(&filp->f_lock);
975 		offset += filp->f_pos;
976 		if (offset < 0) {
977 			spin_unlock(&filp->f_lock);
978 			return -EINVAL;
979 		}
980 	}
981 	if (offset != filp->f_pos) {
982 		filp->f_pos = offset;
983 		if (nfs_readdir_use_cookie(filp))
984 			dir_ctx->dir_cookie = offset;
985 		else
986 			dir_ctx->dir_cookie = 0;
987 		dir_ctx->duped = 0;
988 	}
989 	spin_unlock(&filp->f_lock);
990 	return offset;
991 }
992 
993 /*
994  * All directory operations under NFS are synchronous, so fsync()
995  * is a dummy operation.
996  */
nfs_fsync_dir(struct file * filp,loff_t start,loff_t end,int datasync)997 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
998 			 int datasync)
999 {
1000 	dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1001 
1002 	nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1003 	return 0;
1004 }
1005 
1006 /**
1007  * nfs_force_lookup_revalidate - Mark the directory as having changed
1008  * @dir: pointer to directory inode
1009  *
1010  * This forces the revalidation code in nfs_lookup_revalidate() to do a
1011  * full lookup on all child dentries of 'dir' whenever a change occurs
1012  * on the server that might have invalidated our dcache.
1013  *
1014  * Note that we reserve bit '0' as a tag to let us know when a dentry
1015  * was revalidated while holding a delegation on its inode.
1016  *
1017  * The caller should be holding dir->i_lock
1018  */
nfs_force_lookup_revalidate(struct inode * dir)1019 void nfs_force_lookup_revalidate(struct inode *dir)
1020 {
1021 	NFS_I(dir)->cache_change_attribute += 2;
1022 }
1023 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1024 
1025 /**
1026  * nfs_verify_change_attribute - Detects NFS remote directory changes
1027  * @dir: pointer to parent directory inode
1028  * @verf: previously saved change attribute
1029  *
1030  * Return "false" if the verifiers doesn't match the change attribute.
1031  * This would usually indicate that the directory contents have changed on
1032  * the server, and that any dentries need revalidating.
1033  */
nfs_verify_change_attribute(struct inode * dir,unsigned long verf)1034 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1035 {
1036 	return (verf & ~1UL) == nfs_save_change_attribute(dir);
1037 }
1038 
nfs_set_verifier_delegated(unsigned long * verf)1039 static void nfs_set_verifier_delegated(unsigned long *verf)
1040 {
1041 	*verf |= 1UL;
1042 }
1043 
1044 #if IS_ENABLED(CONFIG_NFS_V4)
nfs_unset_verifier_delegated(unsigned long * verf)1045 static void nfs_unset_verifier_delegated(unsigned long *verf)
1046 {
1047 	*verf &= ~1UL;
1048 }
1049 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1050 
nfs_test_verifier_delegated(unsigned long verf)1051 static bool nfs_test_verifier_delegated(unsigned long verf)
1052 {
1053 	return verf & 1;
1054 }
1055 
nfs_verifier_is_delegated(struct dentry * dentry)1056 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1057 {
1058 	return nfs_test_verifier_delegated(dentry->d_time);
1059 }
1060 
nfs_set_verifier_locked(struct dentry * dentry,unsigned long verf)1061 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1062 {
1063 	struct inode *inode = d_inode(dentry);
1064 	struct inode *dir = d_inode(dentry->d_parent);
1065 
1066 	if (!nfs_verify_change_attribute(dir, verf))
1067 		return;
1068 	if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1069 		nfs_set_verifier_delegated(&verf);
1070 	dentry->d_time = verf;
1071 }
1072 
1073 /**
1074  * nfs_set_verifier - save a parent directory verifier in the dentry
1075  * @dentry: pointer to dentry
1076  * @verf: verifier to save
1077  *
1078  * Saves the parent directory verifier in @dentry. If the inode has
1079  * a delegation, we also tag the dentry as having been revalidated
1080  * while holding a delegation so that we know we don't have to
1081  * look it up again after a directory change.
1082  */
nfs_set_verifier(struct dentry * dentry,unsigned long verf)1083 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1084 {
1085 
1086 	spin_lock(&dentry->d_lock);
1087 	nfs_set_verifier_locked(dentry, verf);
1088 	spin_unlock(&dentry->d_lock);
1089 }
1090 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1091 
1092 #if IS_ENABLED(CONFIG_NFS_V4)
1093 /**
1094  * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1095  * @inode: pointer to inode
1096  *
1097  * Iterates through the dentries in the inode alias list and clears
1098  * the tag used to indicate that the dentry has been revalidated
1099  * while holding a delegation.
1100  * This function is intended for use when the delegation is being
1101  * returned or revoked.
1102  */
nfs_clear_verifier_delegated(struct inode * inode)1103 void nfs_clear_verifier_delegated(struct inode *inode)
1104 {
1105 	struct dentry *alias;
1106 
1107 	if (!inode)
1108 		return;
1109 	spin_lock(&inode->i_lock);
1110 	hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1111 		spin_lock(&alias->d_lock);
1112 		nfs_unset_verifier_delegated(&alias->d_time);
1113 		spin_unlock(&alias->d_lock);
1114 	}
1115 	spin_unlock(&inode->i_lock);
1116 }
1117 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1118 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1119 
1120 /*
1121  * A check for whether or not the parent directory has changed.
1122  * In the case it has, we assume that the dentries are untrustworthy
1123  * and may need to be looked up again.
1124  * If rcu_walk prevents us from performing a full check, return 0.
1125  */
nfs_check_verifier(struct inode * dir,struct dentry * dentry,int rcu_walk)1126 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1127 			      int rcu_walk)
1128 {
1129 	if (IS_ROOT(dentry))
1130 		return 1;
1131 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1132 		return 0;
1133 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
1134 		return 0;
1135 	/* Revalidate nfsi->cache_change_attribute before we declare a match */
1136 	if (nfs_mapping_need_revalidate_inode(dir)) {
1137 		if (rcu_walk)
1138 			return 0;
1139 		if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1140 			return 0;
1141 	}
1142 	if (!nfs_verify_change_attribute(dir, dentry->d_time))
1143 		return 0;
1144 	return 1;
1145 }
1146 
1147 /*
1148  * Use intent information to check whether or not we're going to do
1149  * an O_EXCL create using this path component.
1150  */
nfs_is_exclusive_create(struct inode * dir,unsigned int flags)1151 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1152 {
1153 	if (NFS_PROTO(dir)->version == 2)
1154 		return 0;
1155 	return flags & LOOKUP_EXCL;
1156 }
1157 
1158 /*
1159  * Inode and filehandle revalidation for lookups.
1160  *
1161  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1162  * or if the intent information indicates that we're about to open this
1163  * particular file and the "nocto" mount flag is not set.
1164  *
1165  */
1166 static
nfs_lookup_verify_inode(struct inode * inode,unsigned int flags)1167 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1168 {
1169 	struct nfs_server *server = NFS_SERVER(inode);
1170 	int ret;
1171 
1172 	if (IS_AUTOMOUNT(inode))
1173 		return 0;
1174 
1175 	if (flags & LOOKUP_OPEN) {
1176 		switch (inode->i_mode & S_IFMT) {
1177 		case S_IFREG:
1178 			/* A NFSv4 OPEN will revalidate later */
1179 			if (server->caps & NFS_CAP_ATOMIC_OPEN)
1180 				goto out;
1181 			fallthrough;
1182 		case S_IFDIR:
1183 			if (server->flags & NFS_MOUNT_NOCTO)
1184 				break;
1185 			/* NFS close-to-open cache consistency validation */
1186 			goto out_force;
1187 		}
1188 	}
1189 
1190 	/* VFS wants an on-the-wire revalidation */
1191 	if (flags & LOOKUP_REVAL)
1192 		goto out_force;
1193 out:
1194 	return (inode->i_nlink == 0) ? -ESTALE : 0;
1195 out_force:
1196 	if (flags & LOOKUP_RCU)
1197 		return -ECHILD;
1198 	ret = __nfs_revalidate_inode(server, inode);
1199 	if (ret != 0)
1200 		return ret;
1201 	goto out;
1202 }
1203 
nfs_mark_dir_for_revalidate(struct inode * inode)1204 static void nfs_mark_dir_for_revalidate(struct inode *inode)
1205 {
1206 	struct nfs_inode *nfsi = NFS_I(inode);
1207 
1208 	spin_lock(&inode->i_lock);
1209 	nfsi->cache_validity |= NFS_INO_REVAL_PAGECACHE;
1210 	spin_unlock(&inode->i_lock);
1211 }
1212 
1213 /*
1214  * We judge how long we want to trust negative
1215  * dentries by looking at the parent inode mtime.
1216  *
1217  * If parent mtime has changed, we revalidate, else we wait for a
1218  * period corresponding to the parent's attribute cache timeout value.
1219  *
1220  * If LOOKUP_RCU prevents us from performing a full check, return 1
1221  * suggesting a reval is needed.
1222  *
1223  * Note that when creating a new file, or looking up a rename target,
1224  * then it shouldn't be necessary to revalidate a negative dentry.
1225  */
1226 static inline
nfs_neg_need_reval(struct inode * dir,struct dentry * dentry,unsigned int flags)1227 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1228 		       unsigned int flags)
1229 {
1230 	if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1231 		return 0;
1232 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1233 		return 1;
1234 	return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1235 }
1236 
1237 static int
nfs_lookup_revalidate_done(struct inode * dir,struct dentry * dentry,struct inode * inode,int error)1238 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1239 			   struct inode *inode, int error)
1240 {
1241 	switch (error) {
1242 	case 1:
1243 		dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1244 			__func__, dentry);
1245 		return 1;
1246 	case 0:
1247 		/*
1248 		 * We can't d_drop the root of a disconnected tree:
1249 		 * its d_hash is on the s_anon list and d_drop() would hide
1250 		 * it from shrink_dcache_for_unmount(), leading to busy
1251 		 * inodes on unmount and further oopses.
1252 		 */
1253 		if (inode && IS_ROOT(dentry))
1254 			return 1;
1255 		dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1256 				__func__, dentry);
1257 		return 0;
1258 	}
1259 	dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1260 				__func__, dentry, error);
1261 	return error;
1262 }
1263 
1264 static int
nfs_lookup_revalidate_negative(struct inode * dir,struct dentry * dentry,unsigned int flags)1265 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1266 			       unsigned int flags)
1267 {
1268 	int ret = 1;
1269 	if (nfs_neg_need_reval(dir, dentry, flags)) {
1270 		if (flags & LOOKUP_RCU)
1271 			return -ECHILD;
1272 		ret = 0;
1273 	}
1274 	return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1275 }
1276 
1277 static int
nfs_lookup_revalidate_delegated(struct inode * dir,struct dentry * dentry,struct inode * inode)1278 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1279 				struct inode *inode)
1280 {
1281 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1282 	return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1283 }
1284 
1285 static int
nfs_lookup_revalidate_dentry(struct inode * dir,struct dentry * dentry,struct inode * inode)1286 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry,
1287 			     struct inode *inode)
1288 {
1289 	struct nfs_fh *fhandle;
1290 	struct nfs_fattr *fattr;
1291 	struct nfs4_label *label;
1292 	unsigned long dir_verifier;
1293 	int ret;
1294 
1295 	ret = -ENOMEM;
1296 	fhandle = nfs_alloc_fhandle();
1297 	fattr = nfs_alloc_fattr();
1298 	label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1299 	if (fhandle == NULL || fattr == NULL || IS_ERR(label))
1300 		goto out;
1301 
1302 	dir_verifier = nfs_save_change_attribute(dir);
1303 	ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1304 	if (ret < 0) {
1305 		switch (ret) {
1306 		case -ESTALE:
1307 		case -ENOENT:
1308 			ret = 0;
1309 			break;
1310 		case -ETIMEDOUT:
1311 			if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1312 				ret = 1;
1313 		}
1314 		goto out;
1315 	}
1316 	ret = 0;
1317 	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1318 		goto out;
1319 	if (nfs_refresh_inode(inode, fattr) < 0)
1320 		goto out;
1321 
1322 	nfs_setsecurity(inode, fattr, label);
1323 	nfs_set_verifier(dentry, dir_verifier);
1324 
1325 	/* set a readdirplus hint that we had a cache miss */
1326 	nfs_force_use_readdirplus(dir);
1327 	ret = 1;
1328 out:
1329 	nfs_free_fattr(fattr);
1330 	nfs_free_fhandle(fhandle);
1331 	nfs4_label_free(label);
1332 
1333 	/*
1334 	 * If the lookup failed despite the dentry change attribute being
1335 	 * a match, then we should revalidate the directory cache.
1336 	 */
1337 	if (!ret && nfs_verify_change_attribute(dir, dentry->d_time))
1338 		nfs_mark_dir_for_revalidate(dir);
1339 	return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1340 }
1341 
1342 /*
1343  * This is called every time the dcache has a lookup hit,
1344  * and we should check whether we can really trust that
1345  * lookup.
1346  *
1347  * NOTE! The hit can be a negative hit too, don't assume
1348  * we have an inode!
1349  *
1350  * If the parent directory is seen to have changed, we throw out the
1351  * cached dentry and do a new lookup.
1352  */
1353 static int
nfs_do_lookup_revalidate(struct inode * dir,struct dentry * dentry,unsigned int flags)1354 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1355 			 unsigned int flags)
1356 {
1357 	struct inode *inode;
1358 	int error;
1359 
1360 	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1361 	inode = d_inode(dentry);
1362 
1363 	if (!inode)
1364 		return nfs_lookup_revalidate_negative(dir, dentry, flags);
1365 
1366 	if (is_bad_inode(inode)) {
1367 		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1368 				__func__, dentry);
1369 		goto out_bad;
1370 	}
1371 
1372 	if (nfs_verifier_is_delegated(dentry))
1373 		return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1374 
1375 	/* Force a full look up iff the parent directory has changed */
1376 	if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1377 	    nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1378 		error = nfs_lookup_verify_inode(inode, flags);
1379 		if (error) {
1380 			if (error == -ESTALE)
1381 				nfs_mark_dir_for_revalidate(dir);
1382 			goto out_bad;
1383 		}
1384 		nfs_advise_use_readdirplus(dir);
1385 		goto out_valid;
1386 	}
1387 
1388 	if (flags & LOOKUP_RCU)
1389 		return -ECHILD;
1390 
1391 	if (NFS_STALE(inode))
1392 		goto out_bad;
1393 
1394 	trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1395 	error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1396 	trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1397 	return error;
1398 out_valid:
1399 	return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1400 out_bad:
1401 	if (flags & LOOKUP_RCU)
1402 		return -ECHILD;
1403 	return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1404 }
1405 
1406 static int
__nfs_lookup_revalidate(struct dentry * dentry,unsigned int flags,int (* reval)(struct inode *,struct dentry *,unsigned int))1407 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1408 			int (*reval)(struct inode *, struct dentry *, unsigned int))
1409 {
1410 	struct dentry *parent;
1411 	struct inode *dir;
1412 	int ret;
1413 
1414 	if (flags & LOOKUP_RCU) {
1415 		parent = READ_ONCE(dentry->d_parent);
1416 		dir = d_inode_rcu(parent);
1417 		if (!dir)
1418 			return -ECHILD;
1419 		ret = reval(dir, dentry, flags);
1420 		if (parent != READ_ONCE(dentry->d_parent))
1421 			return -ECHILD;
1422 	} else {
1423 		parent = dget_parent(dentry);
1424 		ret = reval(d_inode(parent), dentry, flags);
1425 		dput(parent);
1426 	}
1427 	return ret;
1428 }
1429 
nfs_lookup_revalidate(struct dentry * dentry,unsigned int flags)1430 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1431 {
1432 	return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1433 }
1434 
1435 /*
1436  * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1437  * when we don't really care about the dentry name. This is called when a
1438  * pathwalk ends on a dentry that was not found via a normal lookup in the
1439  * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1440  *
1441  * In this situation, we just want to verify that the inode itself is OK
1442  * since the dentry might have changed on the server.
1443  */
nfs_weak_revalidate(struct dentry * dentry,unsigned int flags)1444 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1445 {
1446 	struct inode *inode = d_inode(dentry);
1447 	int error = 0;
1448 
1449 	/*
1450 	 * I believe we can only get a negative dentry here in the case of a
1451 	 * procfs-style symlink. Just assume it's correct for now, but we may
1452 	 * eventually need to do something more here.
1453 	 */
1454 	if (!inode) {
1455 		dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1456 				__func__, dentry);
1457 		return 1;
1458 	}
1459 
1460 	if (is_bad_inode(inode)) {
1461 		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1462 				__func__, dentry);
1463 		return 0;
1464 	}
1465 
1466 	error = nfs_lookup_verify_inode(inode, flags);
1467 	dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1468 			__func__, inode->i_ino, error ? "invalid" : "valid");
1469 	return !error;
1470 }
1471 
1472 /*
1473  * This is called from dput() when d_count is going to 0.
1474  */
nfs_dentry_delete(const struct dentry * dentry)1475 static int nfs_dentry_delete(const struct dentry *dentry)
1476 {
1477 	dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1478 		dentry, dentry->d_flags);
1479 
1480 	/* Unhash any dentry with a stale inode */
1481 	if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1482 		return 1;
1483 
1484 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1485 		/* Unhash it, so that ->d_iput() would be called */
1486 		return 1;
1487 	}
1488 	if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1489 		/* Unhash it, so that ancestors of killed async unlink
1490 		 * files will be cleaned up during umount */
1491 		return 1;
1492 	}
1493 	return 0;
1494 
1495 }
1496 
1497 /* Ensure that we revalidate inode->i_nlink */
nfs_drop_nlink(struct inode * inode)1498 static void nfs_drop_nlink(struct inode *inode)
1499 {
1500 	spin_lock(&inode->i_lock);
1501 	/* drop the inode if we're reasonably sure this is the last link */
1502 	if (inode->i_nlink > 0)
1503 		drop_nlink(inode);
1504 	NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1505 	NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE
1506 		| NFS_INO_INVALID_CTIME
1507 		| NFS_INO_INVALID_OTHER
1508 		| NFS_INO_REVAL_FORCED;
1509 	spin_unlock(&inode->i_lock);
1510 }
1511 
1512 /*
1513  * Called when the dentry loses inode.
1514  * We use it to clean up silly-renamed files.
1515  */
nfs_dentry_iput(struct dentry * dentry,struct inode * inode)1516 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1517 {
1518 	if (S_ISDIR(inode->i_mode))
1519 		/* drop any readdir cache as it could easily be old */
1520 		NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1521 
1522 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1523 		nfs_complete_unlink(dentry, inode);
1524 		nfs_drop_nlink(inode);
1525 	}
1526 	iput(inode);
1527 }
1528 
nfs_d_release(struct dentry * dentry)1529 static void nfs_d_release(struct dentry *dentry)
1530 {
1531 	/* free cached devname value, if it survived that far */
1532 	if (unlikely(dentry->d_fsdata)) {
1533 		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1534 			WARN_ON(1);
1535 		else
1536 			kfree(dentry->d_fsdata);
1537 	}
1538 }
1539 
1540 const struct dentry_operations nfs_dentry_operations = {
1541 	.d_revalidate	= nfs_lookup_revalidate,
1542 	.d_weak_revalidate	= nfs_weak_revalidate,
1543 	.d_delete	= nfs_dentry_delete,
1544 	.d_iput		= nfs_dentry_iput,
1545 	.d_automount	= nfs_d_automount,
1546 	.d_release	= nfs_d_release,
1547 };
1548 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1549 
nfs_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)1550 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1551 {
1552 	struct dentry *res;
1553 	struct inode *inode = NULL;
1554 	struct nfs_fh *fhandle = NULL;
1555 	struct nfs_fattr *fattr = NULL;
1556 	struct nfs4_label *label = NULL;
1557 	unsigned long dir_verifier;
1558 	int error;
1559 
1560 	dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1561 	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1562 
1563 	if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1564 		return ERR_PTR(-ENAMETOOLONG);
1565 
1566 	/*
1567 	 * If we're doing an exclusive create, optimize away the lookup
1568 	 * but don't hash the dentry.
1569 	 */
1570 	if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1571 		return NULL;
1572 
1573 	res = ERR_PTR(-ENOMEM);
1574 	fhandle = nfs_alloc_fhandle();
1575 	fattr = nfs_alloc_fattr();
1576 	if (fhandle == NULL || fattr == NULL)
1577 		goto out;
1578 
1579 	label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1580 	if (IS_ERR(label))
1581 		goto out;
1582 
1583 	dir_verifier = nfs_save_change_attribute(dir);
1584 	trace_nfs_lookup_enter(dir, dentry, flags);
1585 	error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1586 	if (error == -ENOENT)
1587 		goto no_entry;
1588 	if (error < 0) {
1589 		res = ERR_PTR(error);
1590 		goto out_label;
1591 	}
1592 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1593 	res = ERR_CAST(inode);
1594 	if (IS_ERR(res))
1595 		goto out_label;
1596 
1597 	/* Notify readdir to use READDIRPLUS */
1598 	nfs_force_use_readdirplus(dir);
1599 
1600 no_entry:
1601 	res = d_splice_alias(inode, dentry);
1602 	if (res != NULL) {
1603 		if (IS_ERR(res))
1604 			goto out_label;
1605 		dentry = res;
1606 	}
1607 	nfs_set_verifier(dentry, dir_verifier);
1608 out_label:
1609 	trace_nfs_lookup_exit(dir, dentry, flags, error);
1610 	nfs4_label_free(label);
1611 out:
1612 	nfs_free_fattr(fattr);
1613 	nfs_free_fhandle(fhandle);
1614 	return res;
1615 }
1616 EXPORT_SYMBOL_GPL(nfs_lookup);
1617 
1618 #if IS_ENABLED(CONFIG_NFS_V4)
1619 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1620 
1621 const struct dentry_operations nfs4_dentry_operations = {
1622 	.d_revalidate	= nfs4_lookup_revalidate,
1623 	.d_weak_revalidate	= nfs_weak_revalidate,
1624 	.d_delete	= nfs_dentry_delete,
1625 	.d_iput		= nfs_dentry_iput,
1626 	.d_automount	= nfs_d_automount,
1627 	.d_release	= nfs_d_release,
1628 };
1629 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1630 
create_nfs_open_context(struct dentry * dentry,int open_flags,struct file * filp)1631 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1632 {
1633 	return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1634 }
1635 
do_open(struct inode * inode,struct file * filp)1636 static int do_open(struct inode *inode, struct file *filp)
1637 {
1638 	nfs_fscache_open_file(inode, filp);
1639 	return 0;
1640 }
1641 
nfs_finish_open(struct nfs_open_context * ctx,struct dentry * dentry,struct file * file,unsigned open_flags)1642 static int nfs_finish_open(struct nfs_open_context *ctx,
1643 			   struct dentry *dentry,
1644 			   struct file *file, unsigned open_flags)
1645 {
1646 	int err;
1647 
1648 	err = finish_open(file, dentry, do_open);
1649 	if (err)
1650 		goto out;
1651 	if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1652 		nfs_file_set_open_context(file, ctx);
1653 	else
1654 		err = -EOPENSTALE;
1655 out:
1656 	return err;
1657 }
1658 
nfs_atomic_open(struct inode * dir,struct dentry * dentry,struct file * file,unsigned open_flags,umode_t mode)1659 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1660 		    struct file *file, unsigned open_flags,
1661 		    umode_t mode)
1662 {
1663 	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1664 	struct nfs_open_context *ctx;
1665 	struct dentry *res;
1666 	struct iattr attr = { .ia_valid = ATTR_OPEN };
1667 	struct inode *inode;
1668 	unsigned int lookup_flags = 0;
1669 	bool switched = false;
1670 	int created = 0;
1671 	int err;
1672 
1673 	/* Expect a negative dentry */
1674 	BUG_ON(d_inode(dentry));
1675 
1676 	dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1677 			dir->i_sb->s_id, dir->i_ino, dentry);
1678 
1679 	err = nfs_check_flags(open_flags);
1680 	if (err)
1681 		return err;
1682 
1683 	/* NFS only supports OPEN on regular files */
1684 	if ((open_flags & O_DIRECTORY)) {
1685 		if (!d_in_lookup(dentry)) {
1686 			/*
1687 			 * Hashed negative dentry with O_DIRECTORY: dentry was
1688 			 * revalidated and is fine, no need to perform lookup
1689 			 * again
1690 			 */
1691 			return -ENOENT;
1692 		}
1693 		lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1694 		goto no_open;
1695 	}
1696 
1697 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1698 		return -ENAMETOOLONG;
1699 
1700 	if (open_flags & O_CREAT) {
1701 		struct nfs_server *server = NFS_SERVER(dir);
1702 
1703 		if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1704 			mode &= ~current_umask();
1705 
1706 		attr.ia_valid |= ATTR_MODE;
1707 		attr.ia_mode = mode;
1708 	}
1709 	if (open_flags & O_TRUNC) {
1710 		attr.ia_valid |= ATTR_SIZE;
1711 		attr.ia_size = 0;
1712 	}
1713 
1714 	if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1715 		d_drop(dentry);
1716 		switched = true;
1717 		dentry = d_alloc_parallel(dentry->d_parent,
1718 					  &dentry->d_name, &wq);
1719 		if (IS_ERR(dentry))
1720 			return PTR_ERR(dentry);
1721 		if (unlikely(!d_in_lookup(dentry)))
1722 			return finish_no_open(file, dentry);
1723 	}
1724 
1725 	ctx = create_nfs_open_context(dentry, open_flags, file);
1726 	err = PTR_ERR(ctx);
1727 	if (IS_ERR(ctx))
1728 		goto out;
1729 
1730 	trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1731 	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1732 	if (created)
1733 		file->f_mode |= FMODE_CREATED;
1734 	if (IS_ERR(inode)) {
1735 		err = PTR_ERR(inode);
1736 		trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1737 		put_nfs_open_context(ctx);
1738 		d_drop(dentry);
1739 		switch (err) {
1740 		case -ENOENT:
1741 			d_splice_alias(NULL, dentry);
1742 			nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1743 			break;
1744 		case -EISDIR:
1745 		case -ENOTDIR:
1746 			goto no_open;
1747 		case -ELOOP:
1748 			if (!(open_flags & O_NOFOLLOW))
1749 				goto no_open;
1750 			break;
1751 			/* case -EINVAL: */
1752 		default:
1753 			break;
1754 		}
1755 		goto out;
1756 	}
1757 
1758 	err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1759 	trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1760 	put_nfs_open_context(ctx);
1761 out:
1762 	if (unlikely(switched)) {
1763 		d_lookup_done(dentry);
1764 		dput(dentry);
1765 	}
1766 	return err;
1767 
1768 no_open:
1769 	res = nfs_lookup(dir, dentry, lookup_flags);
1770 	if (!res) {
1771 		inode = d_inode(dentry);
1772 		if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
1773 		    !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
1774 			res = ERR_PTR(-ENOTDIR);
1775 		else if (inode && S_ISREG(inode->i_mode))
1776 			res = ERR_PTR(-EOPENSTALE);
1777 	} else if (!IS_ERR(res)) {
1778 		inode = d_inode(res);
1779 		if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
1780 		    !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
1781 			dput(res);
1782 			res = ERR_PTR(-ENOTDIR);
1783 		} else if (inode && S_ISREG(inode->i_mode)) {
1784 			dput(res);
1785 			res = ERR_PTR(-EOPENSTALE);
1786 		}
1787 	}
1788 	if (switched) {
1789 		d_lookup_done(dentry);
1790 		if (!res)
1791 			res = dentry;
1792 		else
1793 			dput(dentry);
1794 	}
1795 	if (IS_ERR(res))
1796 		return PTR_ERR(res);
1797 	return finish_no_open(file, res);
1798 }
1799 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1800 
1801 static int
nfs4_do_lookup_revalidate(struct inode * dir,struct dentry * dentry,unsigned int flags)1802 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1803 			  unsigned int flags)
1804 {
1805 	struct inode *inode;
1806 
1807 	if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1808 		goto full_reval;
1809 	if (d_mountpoint(dentry))
1810 		goto full_reval;
1811 
1812 	inode = d_inode(dentry);
1813 
1814 	/* We can't create new files in nfs_open_revalidate(), so we
1815 	 * optimize away revalidation of negative dentries.
1816 	 */
1817 	if (inode == NULL)
1818 		goto full_reval;
1819 
1820 	if (nfs_verifier_is_delegated(dentry))
1821 		return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1822 
1823 	/* NFS only supports OPEN on regular files */
1824 	if (!S_ISREG(inode->i_mode))
1825 		goto full_reval;
1826 
1827 	/* We cannot do exclusive creation on a positive dentry */
1828 	if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
1829 		goto reval_dentry;
1830 
1831 	/* Check if the directory changed */
1832 	if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
1833 		goto reval_dentry;
1834 
1835 	/* Let f_op->open() actually open (and revalidate) the file */
1836 	return 1;
1837 reval_dentry:
1838 	if (flags & LOOKUP_RCU)
1839 		return -ECHILD;
1840 	return nfs_lookup_revalidate_dentry(dir, dentry, inode);
1841 
1842 full_reval:
1843 	return nfs_do_lookup_revalidate(dir, dentry, flags);
1844 }
1845 
nfs4_lookup_revalidate(struct dentry * dentry,unsigned int flags)1846 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1847 {
1848 	return __nfs_lookup_revalidate(dentry, flags,
1849 			nfs4_do_lookup_revalidate);
1850 }
1851 
1852 #endif /* CONFIG_NFSV4 */
1853 
1854 struct dentry *
nfs_add_or_obtain(struct dentry * dentry,struct nfs_fh * fhandle,struct nfs_fattr * fattr,struct nfs4_label * label)1855 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
1856 				struct nfs_fattr *fattr,
1857 				struct nfs4_label *label)
1858 {
1859 	struct dentry *parent = dget_parent(dentry);
1860 	struct inode *dir = d_inode(parent);
1861 	struct inode *inode;
1862 	struct dentry *d;
1863 	int error;
1864 
1865 	d_drop(dentry);
1866 
1867 	if (fhandle->size == 0) {
1868 		error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL);
1869 		if (error)
1870 			goto out_error;
1871 	}
1872 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1873 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
1874 		struct nfs_server *server = NFS_SB(dentry->d_sb);
1875 		error = server->nfs_client->rpc_ops->getattr(server, fhandle,
1876 				fattr, NULL, NULL);
1877 		if (error < 0)
1878 			goto out_error;
1879 	}
1880 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1881 	d = d_splice_alias(inode, dentry);
1882 out:
1883 	dput(parent);
1884 	return d;
1885 out_error:
1886 	d = ERR_PTR(error);
1887 	goto out;
1888 }
1889 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
1890 
1891 /*
1892  * Code common to create, mkdir, and mknod.
1893  */
nfs_instantiate(struct dentry * dentry,struct nfs_fh * fhandle,struct nfs_fattr * fattr,struct nfs4_label * label)1894 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1895 				struct nfs_fattr *fattr,
1896 				struct nfs4_label *label)
1897 {
1898 	struct dentry *d;
1899 
1900 	d = nfs_add_or_obtain(dentry, fhandle, fattr, label);
1901 	if (IS_ERR(d))
1902 		return PTR_ERR(d);
1903 
1904 	/* Callers don't care */
1905 	dput(d);
1906 	return 0;
1907 }
1908 EXPORT_SYMBOL_GPL(nfs_instantiate);
1909 
1910 /*
1911  * Following a failed create operation, we drop the dentry rather
1912  * than retain a negative dentry. This avoids a problem in the event
1913  * that the operation succeeded on the server, but an error in the
1914  * reply path made it appear to have failed.
1915  */
nfs_create(struct inode * dir,struct dentry * dentry,umode_t mode,bool excl)1916 int nfs_create(struct inode *dir, struct dentry *dentry,
1917 		umode_t mode, bool excl)
1918 {
1919 	struct iattr attr;
1920 	int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1921 	int error;
1922 
1923 	dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1924 			dir->i_sb->s_id, dir->i_ino, dentry);
1925 
1926 	attr.ia_mode = mode;
1927 	attr.ia_valid = ATTR_MODE;
1928 
1929 	trace_nfs_create_enter(dir, dentry, open_flags);
1930 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1931 	trace_nfs_create_exit(dir, dentry, open_flags, error);
1932 	if (error != 0)
1933 		goto out_err;
1934 	return 0;
1935 out_err:
1936 	d_drop(dentry);
1937 	return error;
1938 }
1939 EXPORT_SYMBOL_GPL(nfs_create);
1940 
1941 /*
1942  * See comments for nfs_proc_create regarding failed operations.
1943  */
1944 int
nfs_mknod(struct inode * dir,struct dentry * dentry,umode_t mode,dev_t rdev)1945 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1946 {
1947 	struct iattr attr;
1948 	int status;
1949 
1950 	dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1951 			dir->i_sb->s_id, dir->i_ino, dentry);
1952 
1953 	attr.ia_mode = mode;
1954 	attr.ia_valid = ATTR_MODE;
1955 
1956 	trace_nfs_mknod_enter(dir, dentry);
1957 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1958 	trace_nfs_mknod_exit(dir, dentry, status);
1959 	if (status != 0)
1960 		goto out_err;
1961 	return 0;
1962 out_err:
1963 	d_drop(dentry);
1964 	return status;
1965 }
1966 EXPORT_SYMBOL_GPL(nfs_mknod);
1967 
1968 /*
1969  * See comments for nfs_proc_create regarding failed operations.
1970  */
nfs_mkdir(struct inode * dir,struct dentry * dentry,umode_t mode)1971 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1972 {
1973 	struct iattr attr;
1974 	int error;
1975 
1976 	dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1977 			dir->i_sb->s_id, dir->i_ino, dentry);
1978 
1979 	attr.ia_valid = ATTR_MODE;
1980 	attr.ia_mode = mode | S_IFDIR;
1981 
1982 	trace_nfs_mkdir_enter(dir, dentry);
1983 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1984 	trace_nfs_mkdir_exit(dir, dentry, error);
1985 	if (error != 0)
1986 		goto out_err;
1987 	return 0;
1988 out_err:
1989 	d_drop(dentry);
1990 	return error;
1991 }
1992 EXPORT_SYMBOL_GPL(nfs_mkdir);
1993 
nfs_dentry_handle_enoent(struct dentry * dentry)1994 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1995 {
1996 	if (simple_positive(dentry))
1997 		d_delete(dentry);
1998 }
1999 
nfs_rmdir(struct inode * dir,struct dentry * dentry)2000 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2001 {
2002 	int error;
2003 
2004 	dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2005 			dir->i_sb->s_id, dir->i_ino, dentry);
2006 
2007 	trace_nfs_rmdir_enter(dir, dentry);
2008 	if (d_really_is_positive(dentry)) {
2009 		down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2010 		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2011 		/* Ensure the VFS deletes this inode */
2012 		switch (error) {
2013 		case 0:
2014 			clear_nlink(d_inode(dentry));
2015 			break;
2016 		case -ENOENT:
2017 			nfs_dentry_handle_enoent(dentry);
2018 		}
2019 		up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2020 	} else
2021 		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2022 	trace_nfs_rmdir_exit(dir, dentry, error);
2023 
2024 	return error;
2025 }
2026 EXPORT_SYMBOL_GPL(nfs_rmdir);
2027 
2028 /*
2029  * Remove a file after making sure there are no pending writes,
2030  * and after checking that the file has only one user.
2031  *
2032  * We invalidate the attribute cache and free the inode prior to the operation
2033  * to avoid possible races if the server reuses the inode.
2034  */
nfs_safe_remove(struct dentry * dentry)2035 static int nfs_safe_remove(struct dentry *dentry)
2036 {
2037 	struct inode *dir = d_inode(dentry->d_parent);
2038 	struct inode *inode = d_inode(dentry);
2039 	int error = -EBUSY;
2040 
2041 	dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2042 
2043 	/* If the dentry was sillyrenamed, we simply call d_delete() */
2044 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2045 		error = 0;
2046 		goto out;
2047 	}
2048 
2049 	trace_nfs_remove_enter(dir, dentry);
2050 	if (inode != NULL) {
2051 		error = NFS_PROTO(dir)->remove(dir, dentry);
2052 		if (error == 0)
2053 			nfs_drop_nlink(inode);
2054 	} else
2055 		error = NFS_PROTO(dir)->remove(dir, dentry);
2056 	if (error == -ENOENT)
2057 		nfs_dentry_handle_enoent(dentry);
2058 	trace_nfs_remove_exit(dir, dentry, error);
2059 out:
2060 	return error;
2061 }
2062 
2063 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
2064  *  belongs to an active ".nfs..." file and we return -EBUSY.
2065  *
2066  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
2067  */
nfs_unlink(struct inode * dir,struct dentry * dentry)2068 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2069 {
2070 	int error;
2071 	int need_rehash = 0;
2072 
2073 	dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2074 		dir->i_ino, dentry);
2075 
2076 	trace_nfs_unlink_enter(dir, dentry);
2077 	spin_lock(&dentry->d_lock);
2078 	if (d_count(dentry) > 1) {
2079 		spin_unlock(&dentry->d_lock);
2080 		/* Start asynchronous writeout of the inode */
2081 		write_inode_now(d_inode(dentry), 0);
2082 		error = nfs_sillyrename(dir, dentry);
2083 		goto out;
2084 	}
2085 	if (!d_unhashed(dentry)) {
2086 		__d_drop(dentry);
2087 		need_rehash = 1;
2088 	}
2089 	spin_unlock(&dentry->d_lock);
2090 	error = nfs_safe_remove(dentry);
2091 	if (!error || error == -ENOENT) {
2092 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2093 	} else if (need_rehash)
2094 		d_rehash(dentry);
2095 out:
2096 	trace_nfs_unlink_exit(dir, dentry, error);
2097 	return error;
2098 }
2099 EXPORT_SYMBOL_GPL(nfs_unlink);
2100 
2101 /*
2102  * To create a symbolic link, most file systems instantiate a new inode,
2103  * add a page to it containing the path, then write it out to the disk
2104  * using prepare_write/commit_write.
2105  *
2106  * Unfortunately the NFS client can't create the in-core inode first
2107  * because it needs a file handle to create an in-core inode (see
2108  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
2109  * symlink request has completed on the server.
2110  *
2111  * So instead we allocate a raw page, copy the symname into it, then do
2112  * the SYMLINK request with the page as the buffer.  If it succeeds, we
2113  * now have a new file handle and can instantiate an in-core NFS inode
2114  * and move the raw page into its mapping.
2115  */
nfs_symlink(struct inode * dir,struct dentry * dentry,const char * symname)2116 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2117 {
2118 	struct page *page;
2119 	char *kaddr;
2120 	struct iattr attr;
2121 	unsigned int pathlen = strlen(symname);
2122 	int error;
2123 
2124 	dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2125 		dir->i_ino, dentry, symname);
2126 
2127 	if (pathlen > PAGE_SIZE)
2128 		return -ENAMETOOLONG;
2129 
2130 	attr.ia_mode = S_IFLNK | S_IRWXUGO;
2131 	attr.ia_valid = ATTR_MODE;
2132 
2133 	page = alloc_page(GFP_USER);
2134 	if (!page)
2135 		return -ENOMEM;
2136 
2137 	kaddr = page_address(page);
2138 	memcpy(kaddr, symname, pathlen);
2139 	if (pathlen < PAGE_SIZE)
2140 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2141 
2142 	trace_nfs_symlink_enter(dir, dentry);
2143 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2144 	trace_nfs_symlink_exit(dir, dentry, error);
2145 	if (error != 0) {
2146 		dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2147 			dir->i_sb->s_id, dir->i_ino,
2148 			dentry, symname, error);
2149 		d_drop(dentry);
2150 		__free_page(page);
2151 		return error;
2152 	}
2153 
2154 	/*
2155 	 * No big deal if we can't add this page to the page cache here.
2156 	 * READLINK will get the missing page from the server if needed.
2157 	 */
2158 	if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2159 							GFP_KERNEL)) {
2160 		SetPageUptodate(page);
2161 		unlock_page(page);
2162 		/*
2163 		 * add_to_page_cache_lru() grabs an extra page refcount.
2164 		 * Drop it here to avoid leaking this page later.
2165 		 */
2166 		put_page(page);
2167 	} else
2168 		__free_page(page);
2169 
2170 	return 0;
2171 }
2172 EXPORT_SYMBOL_GPL(nfs_symlink);
2173 
2174 int
nfs_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)2175 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2176 {
2177 	struct inode *inode = d_inode(old_dentry);
2178 	int error;
2179 
2180 	dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2181 		old_dentry, dentry);
2182 
2183 	trace_nfs_link_enter(inode, dir, dentry);
2184 	d_drop(dentry);
2185 	if (S_ISREG(inode->i_mode))
2186 		nfs_sync_inode(inode);
2187 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2188 	if (error == 0) {
2189 		ihold(inode);
2190 		d_add(dentry, inode);
2191 	}
2192 	trace_nfs_link_exit(inode, dir, dentry, error);
2193 	return error;
2194 }
2195 EXPORT_SYMBOL_GPL(nfs_link);
2196 
2197 /*
2198  * RENAME
2199  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2200  * different file handle for the same inode after a rename (e.g. when
2201  * moving to a different directory). A fail-safe method to do so would
2202  * be to look up old_dir/old_name, create a link to new_dir/new_name and
2203  * rename the old file using the sillyrename stuff. This way, the original
2204  * file in old_dir will go away when the last process iput()s the inode.
2205  *
2206  * FIXED.
2207  *
2208  * It actually works quite well. One needs to have the possibility for
2209  * at least one ".nfs..." file in each directory the file ever gets
2210  * moved or linked to which happens automagically with the new
2211  * implementation that only depends on the dcache stuff instead of
2212  * using the inode layer
2213  *
2214  * Unfortunately, things are a little more complicated than indicated
2215  * above. For a cross-directory move, we want to make sure we can get
2216  * rid of the old inode after the operation.  This means there must be
2217  * no pending writes (if it's a file), and the use count must be 1.
2218  * If these conditions are met, we can drop the dentries before doing
2219  * the rename.
2220  */
nfs_rename(struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry,unsigned int flags)2221 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2222 	       struct inode *new_dir, struct dentry *new_dentry,
2223 	       unsigned int flags)
2224 {
2225 	struct inode *old_inode = d_inode(old_dentry);
2226 	struct inode *new_inode = d_inode(new_dentry);
2227 	struct dentry *dentry = NULL, *rehash = NULL;
2228 	struct rpc_task *task;
2229 	int error = -EBUSY;
2230 
2231 	if (flags)
2232 		return -EINVAL;
2233 
2234 	dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2235 		 old_dentry, new_dentry,
2236 		 d_count(new_dentry));
2237 
2238 	trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2239 	/*
2240 	 * For non-directories, check whether the target is busy and if so,
2241 	 * make a copy of the dentry and then do a silly-rename. If the
2242 	 * silly-rename succeeds, the copied dentry is hashed and becomes
2243 	 * the new target.
2244 	 */
2245 	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2246 		/*
2247 		 * To prevent any new references to the target during the
2248 		 * rename, we unhash the dentry in advance.
2249 		 */
2250 		if (!d_unhashed(new_dentry)) {
2251 			d_drop(new_dentry);
2252 			rehash = new_dentry;
2253 		}
2254 
2255 		if (d_count(new_dentry) > 2) {
2256 			int err;
2257 
2258 			/* copy the target dentry's name */
2259 			dentry = d_alloc(new_dentry->d_parent,
2260 					 &new_dentry->d_name);
2261 			if (!dentry)
2262 				goto out;
2263 
2264 			/* silly-rename the existing target ... */
2265 			err = nfs_sillyrename(new_dir, new_dentry);
2266 			if (err)
2267 				goto out;
2268 
2269 			new_dentry = dentry;
2270 			rehash = NULL;
2271 			new_inode = NULL;
2272 		}
2273 	}
2274 
2275 	if (S_ISREG(old_inode->i_mode))
2276 		nfs_sync_inode(old_inode);
2277 	task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2278 	if (IS_ERR(task)) {
2279 		error = PTR_ERR(task);
2280 		goto out;
2281 	}
2282 
2283 	error = rpc_wait_for_completion_task(task);
2284 	if (error != 0) {
2285 		((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2286 		/* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2287 		smp_wmb();
2288 	} else
2289 		error = task->tk_status;
2290 	rpc_put_task(task);
2291 	/* Ensure the inode attributes are revalidated */
2292 	if (error == 0) {
2293 		spin_lock(&old_inode->i_lock);
2294 		NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2295 		NFS_I(old_inode)->cache_validity |= NFS_INO_INVALID_CHANGE
2296 			| NFS_INO_INVALID_CTIME
2297 			| NFS_INO_REVAL_FORCED;
2298 		spin_unlock(&old_inode->i_lock);
2299 	}
2300 out:
2301 	if (rehash)
2302 		d_rehash(rehash);
2303 	trace_nfs_rename_exit(old_dir, old_dentry,
2304 			new_dir, new_dentry, error);
2305 	if (!error) {
2306 		if (new_inode != NULL)
2307 			nfs_drop_nlink(new_inode);
2308 		/*
2309 		 * The d_move() should be here instead of in an async RPC completion
2310 		 * handler because we need the proper locks to move the dentry.  If
2311 		 * we're interrupted by a signal, the async RPC completion handler
2312 		 * should mark the directories for revalidation.
2313 		 */
2314 		d_move(old_dentry, new_dentry);
2315 		nfs_set_verifier(old_dentry,
2316 					nfs_save_change_attribute(new_dir));
2317 	} else if (error == -ENOENT)
2318 		nfs_dentry_handle_enoent(old_dentry);
2319 
2320 	/* new dentry created? */
2321 	if (dentry)
2322 		dput(dentry);
2323 	return error;
2324 }
2325 EXPORT_SYMBOL_GPL(nfs_rename);
2326 
2327 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2328 static LIST_HEAD(nfs_access_lru_list);
2329 static atomic_long_t nfs_access_nr_entries;
2330 
2331 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2332 module_param(nfs_access_max_cachesize, ulong, 0644);
2333 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2334 
nfs_access_free_entry(struct nfs_access_entry * entry)2335 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2336 {
2337 	put_cred(entry->cred);
2338 	kfree_rcu(entry, rcu_head);
2339 	smp_mb__before_atomic();
2340 	atomic_long_dec(&nfs_access_nr_entries);
2341 	smp_mb__after_atomic();
2342 }
2343 
nfs_access_free_list(struct list_head * head)2344 static void nfs_access_free_list(struct list_head *head)
2345 {
2346 	struct nfs_access_entry *cache;
2347 
2348 	while (!list_empty(head)) {
2349 		cache = list_entry(head->next, struct nfs_access_entry, lru);
2350 		list_del(&cache->lru);
2351 		nfs_access_free_entry(cache);
2352 	}
2353 }
2354 
2355 static unsigned long
nfs_do_access_cache_scan(unsigned int nr_to_scan)2356 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2357 {
2358 	LIST_HEAD(head);
2359 	struct nfs_inode *nfsi, *next;
2360 	struct nfs_access_entry *cache;
2361 	long freed = 0;
2362 
2363 	spin_lock(&nfs_access_lru_lock);
2364 	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2365 		struct inode *inode;
2366 
2367 		if (nr_to_scan-- == 0)
2368 			break;
2369 		inode = &nfsi->vfs_inode;
2370 		spin_lock(&inode->i_lock);
2371 		if (list_empty(&nfsi->access_cache_entry_lru))
2372 			goto remove_lru_entry;
2373 		cache = list_entry(nfsi->access_cache_entry_lru.next,
2374 				struct nfs_access_entry, lru);
2375 		list_move(&cache->lru, &head);
2376 		rb_erase(&cache->rb_node, &nfsi->access_cache);
2377 		freed++;
2378 		if (!list_empty(&nfsi->access_cache_entry_lru))
2379 			list_move_tail(&nfsi->access_cache_inode_lru,
2380 					&nfs_access_lru_list);
2381 		else {
2382 remove_lru_entry:
2383 			list_del_init(&nfsi->access_cache_inode_lru);
2384 			smp_mb__before_atomic();
2385 			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2386 			smp_mb__after_atomic();
2387 		}
2388 		spin_unlock(&inode->i_lock);
2389 	}
2390 	spin_unlock(&nfs_access_lru_lock);
2391 	nfs_access_free_list(&head);
2392 	return freed;
2393 }
2394 
2395 unsigned long
nfs_access_cache_scan(struct shrinker * shrink,struct shrink_control * sc)2396 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2397 {
2398 	int nr_to_scan = sc->nr_to_scan;
2399 	gfp_t gfp_mask = sc->gfp_mask;
2400 
2401 	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2402 		return SHRINK_STOP;
2403 	return nfs_do_access_cache_scan(nr_to_scan);
2404 }
2405 
2406 
2407 unsigned long
nfs_access_cache_count(struct shrinker * shrink,struct shrink_control * sc)2408 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2409 {
2410 	return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2411 }
2412 
2413 static void
nfs_access_cache_enforce_limit(void)2414 nfs_access_cache_enforce_limit(void)
2415 {
2416 	long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2417 	unsigned long diff;
2418 	unsigned int nr_to_scan;
2419 
2420 	if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2421 		return;
2422 	nr_to_scan = 100;
2423 	diff = nr_entries - nfs_access_max_cachesize;
2424 	if (diff < nr_to_scan)
2425 		nr_to_scan = diff;
2426 	nfs_do_access_cache_scan(nr_to_scan);
2427 }
2428 
__nfs_access_zap_cache(struct nfs_inode * nfsi,struct list_head * head)2429 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2430 {
2431 	struct rb_root *root_node = &nfsi->access_cache;
2432 	struct rb_node *n;
2433 	struct nfs_access_entry *entry;
2434 
2435 	/* Unhook entries from the cache */
2436 	while ((n = rb_first(root_node)) != NULL) {
2437 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2438 		rb_erase(n, root_node);
2439 		list_move(&entry->lru, head);
2440 	}
2441 	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2442 }
2443 
nfs_access_zap_cache(struct inode * inode)2444 void nfs_access_zap_cache(struct inode *inode)
2445 {
2446 	LIST_HEAD(head);
2447 
2448 	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2449 		return;
2450 	/* Remove from global LRU init */
2451 	spin_lock(&nfs_access_lru_lock);
2452 	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2453 		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2454 
2455 	spin_lock(&inode->i_lock);
2456 	__nfs_access_zap_cache(NFS_I(inode), &head);
2457 	spin_unlock(&inode->i_lock);
2458 	spin_unlock(&nfs_access_lru_lock);
2459 	nfs_access_free_list(&head);
2460 }
2461 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2462 
nfs_access_search_rbtree(struct inode * inode,const struct cred * cred)2463 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2464 {
2465 	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2466 
2467 	while (n != NULL) {
2468 		struct nfs_access_entry *entry =
2469 			rb_entry(n, struct nfs_access_entry, rb_node);
2470 		int cmp = cred_fscmp(cred, entry->cred);
2471 
2472 		if (cmp < 0)
2473 			n = n->rb_left;
2474 		else if (cmp > 0)
2475 			n = n->rb_right;
2476 		else
2477 			return entry;
2478 	}
2479 	return NULL;
2480 }
2481 
nfs_access_get_cached_locked(struct inode * inode,const struct cred * cred,u32 * mask,bool may_block)2482 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
2483 {
2484 	struct nfs_inode *nfsi = NFS_I(inode);
2485 	struct nfs_access_entry *cache;
2486 	bool retry = true;
2487 	int err;
2488 
2489 	spin_lock(&inode->i_lock);
2490 	for(;;) {
2491 		if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2492 			goto out_zap;
2493 		cache = nfs_access_search_rbtree(inode, cred);
2494 		err = -ENOENT;
2495 		if (cache == NULL)
2496 			goto out;
2497 		/* Found an entry, is our attribute cache valid? */
2498 		if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2499 			break;
2500 		if (!retry)
2501 			break;
2502 		err = -ECHILD;
2503 		if (!may_block)
2504 			goto out;
2505 		spin_unlock(&inode->i_lock);
2506 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2507 		if (err)
2508 			return err;
2509 		spin_lock(&inode->i_lock);
2510 		retry = false;
2511 	}
2512 	*mask = cache->mask;
2513 	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2514 	err = 0;
2515 out:
2516 	spin_unlock(&inode->i_lock);
2517 	return err;
2518 out_zap:
2519 	spin_unlock(&inode->i_lock);
2520 	nfs_access_zap_cache(inode);
2521 	return -ENOENT;
2522 }
2523 
nfs_access_get_cached_rcu(struct inode * inode,const struct cred * cred,u32 * mask)2524 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
2525 {
2526 	/* Only check the most recently returned cache entry,
2527 	 * but do it without locking.
2528 	 */
2529 	struct nfs_inode *nfsi = NFS_I(inode);
2530 	struct nfs_access_entry *cache;
2531 	int err = -ECHILD;
2532 	struct list_head *lh;
2533 
2534 	rcu_read_lock();
2535 	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2536 		goto out;
2537 	lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
2538 	cache = list_entry(lh, struct nfs_access_entry, lru);
2539 	if (lh == &nfsi->access_cache_entry_lru ||
2540 	    cred_fscmp(cred, cache->cred) != 0)
2541 		cache = NULL;
2542 	if (cache == NULL)
2543 		goto out;
2544 	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2545 		goto out;
2546 	*mask = cache->mask;
2547 	err = 0;
2548 out:
2549 	rcu_read_unlock();
2550 	return err;
2551 }
2552 
nfs_access_get_cached(struct inode * inode,const struct cred * cred,u32 * mask,bool may_block)2553 int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
2554 			  u32 *mask, bool may_block)
2555 {
2556 	int status;
2557 
2558 	status = nfs_access_get_cached_rcu(inode, cred, mask);
2559 	if (status != 0)
2560 		status = nfs_access_get_cached_locked(inode, cred, mask,
2561 		    may_block);
2562 
2563 	return status;
2564 }
2565 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
2566 
nfs_access_add_rbtree(struct inode * inode,struct nfs_access_entry * set)2567 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2568 {
2569 	struct nfs_inode *nfsi = NFS_I(inode);
2570 	struct rb_root *root_node = &nfsi->access_cache;
2571 	struct rb_node **p = &root_node->rb_node;
2572 	struct rb_node *parent = NULL;
2573 	struct nfs_access_entry *entry;
2574 	int cmp;
2575 
2576 	spin_lock(&inode->i_lock);
2577 	while (*p != NULL) {
2578 		parent = *p;
2579 		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2580 		cmp = cred_fscmp(set->cred, entry->cred);
2581 
2582 		if (cmp < 0)
2583 			p = &parent->rb_left;
2584 		else if (cmp > 0)
2585 			p = &parent->rb_right;
2586 		else
2587 			goto found;
2588 	}
2589 	rb_link_node(&set->rb_node, parent, p);
2590 	rb_insert_color(&set->rb_node, root_node);
2591 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2592 	spin_unlock(&inode->i_lock);
2593 	return;
2594 found:
2595 	rb_replace_node(parent, &set->rb_node, root_node);
2596 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2597 	list_del(&entry->lru);
2598 	spin_unlock(&inode->i_lock);
2599 	nfs_access_free_entry(entry);
2600 }
2601 
nfs_access_add_cache(struct inode * inode,struct nfs_access_entry * set)2602 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2603 {
2604 	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2605 	if (cache == NULL)
2606 		return;
2607 	RB_CLEAR_NODE(&cache->rb_node);
2608 	cache->cred = get_cred(set->cred);
2609 	cache->mask = set->mask;
2610 
2611 	/* The above field assignments must be visible
2612 	 * before this item appears on the lru.  We cannot easily
2613 	 * use rcu_assign_pointer, so just force the memory barrier.
2614 	 */
2615 	smp_wmb();
2616 	nfs_access_add_rbtree(inode, cache);
2617 
2618 	/* Update accounting */
2619 	smp_mb__before_atomic();
2620 	atomic_long_inc(&nfs_access_nr_entries);
2621 	smp_mb__after_atomic();
2622 
2623 	/* Add inode to global LRU list */
2624 	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2625 		spin_lock(&nfs_access_lru_lock);
2626 		if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2627 			list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2628 					&nfs_access_lru_list);
2629 		spin_unlock(&nfs_access_lru_lock);
2630 	}
2631 	nfs_access_cache_enforce_limit();
2632 }
2633 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2634 
2635 #define NFS_MAY_READ (NFS_ACCESS_READ)
2636 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2637 		NFS_ACCESS_EXTEND | \
2638 		NFS_ACCESS_DELETE)
2639 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2640 		NFS_ACCESS_EXTEND)
2641 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2642 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2643 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2644 static int
nfs_access_calc_mask(u32 access_result,umode_t umode)2645 nfs_access_calc_mask(u32 access_result, umode_t umode)
2646 {
2647 	int mask = 0;
2648 
2649 	if (access_result & NFS_MAY_READ)
2650 		mask |= MAY_READ;
2651 	if (S_ISDIR(umode)) {
2652 		if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2653 			mask |= MAY_WRITE;
2654 		if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2655 			mask |= MAY_EXEC;
2656 	} else if (S_ISREG(umode)) {
2657 		if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2658 			mask |= MAY_WRITE;
2659 		if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2660 			mask |= MAY_EXEC;
2661 	} else if (access_result & NFS_MAY_WRITE)
2662 			mask |= MAY_WRITE;
2663 	return mask;
2664 }
2665 
nfs_access_set_mask(struct nfs_access_entry * entry,u32 access_result)2666 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2667 {
2668 	entry->mask = access_result;
2669 }
2670 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2671 
nfs_do_access(struct inode * inode,const struct cred * cred,int mask)2672 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
2673 {
2674 	struct nfs_access_entry cache;
2675 	bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2676 	int cache_mask = -1;
2677 	int status;
2678 
2679 	trace_nfs_access_enter(inode);
2680 
2681 	status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
2682 	if (status == 0)
2683 		goto out_cached;
2684 
2685 	status = -ECHILD;
2686 	if (!may_block)
2687 		goto out;
2688 
2689 	/*
2690 	 * Determine which access bits we want to ask for...
2691 	 */
2692 	cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2693 	if (nfs_server_capable(inode, NFS_CAP_XATTR)) {
2694 		cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE |
2695 		    NFS_ACCESS_XALIST;
2696 	}
2697 	if (S_ISDIR(inode->i_mode))
2698 		cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2699 	else
2700 		cache.mask |= NFS_ACCESS_EXECUTE;
2701 	cache.cred = cred;
2702 	status = NFS_PROTO(inode)->access(inode, &cache);
2703 	if (status != 0) {
2704 		if (status == -ESTALE) {
2705 			if (!S_ISDIR(inode->i_mode))
2706 				nfs_set_inode_stale(inode);
2707 			else
2708 				nfs_zap_caches(inode);
2709 		}
2710 		goto out;
2711 	}
2712 	nfs_access_add_cache(inode, &cache);
2713 out_cached:
2714 	cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2715 	if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2716 		status = -EACCES;
2717 out:
2718 	trace_nfs_access_exit(inode, mask, cache_mask, status);
2719 	return status;
2720 }
2721 
nfs_open_permission_mask(int openflags)2722 static int nfs_open_permission_mask(int openflags)
2723 {
2724 	int mask = 0;
2725 
2726 	if (openflags & __FMODE_EXEC) {
2727 		/* ONLY check exec rights */
2728 		mask = MAY_EXEC;
2729 	} else {
2730 		if ((openflags & O_ACCMODE) != O_WRONLY)
2731 			mask |= MAY_READ;
2732 		if ((openflags & O_ACCMODE) != O_RDONLY)
2733 			mask |= MAY_WRITE;
2734 	}
2735 
2736 	return mask;
2737 }
2738 
nfs_may_open(struct inode * inode,const struct cred * cred,int openflags)2739 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
2740 {
2741 	return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2742 }
2743 EXPORT_SYMBOL_GPL(nfs_may_open);
2744 
nfs_execute_ok(struct inode * inode,int mask)2745 static int nfs_execute_ok(struct inode *inode, int mask)
2746 {
2747 	struct nfs_server *server = NFS_SERVER(inode);
2748 	int ret = 0;
2749 
2750 	if (S_ISDIR(inode->i_mode))
2751 		return 0;
2752 	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) {
2753 		if (mask & MAY_NOT_BLOCK)
2754 			return -ECHILD;
2755 		ret = __nfs_revalidate_inode(server, inode);
2756 	}
2757 	if (ret == 0 && !execute_ok(inode))
2758 		ret = -EACCES;
2759 	return ret;
2760 }
2761 
nfs_permission(struct inode * inode,int mask)2762 int nfs_permission(struct inode *inode, int mask)
2763 {
2764 	const struct cred *cred = current_cred();
2765 	int res = 0;
2766 
2767 	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2768 
2769 	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2770 		goto out;
2771 	/* Is this sys_access() ? */
2772 	if (mask & (MAY_ACCESS | MAY_CHDIR))
2773 		goto force_lookup;
2774 
2775 	switch (inode->i_mode & S_IFMT) {
2776 		case S_IFLNK:
2777 			goto out;
2778 		case S_IFREG:
2779 			if ((mask & MAY_OPEN) &&
2780 			   nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2781 				return 0;
2782 			break;
2783 		case S_IFDIR:
2784 			/*
2785 			 * Optimize away all write operations, since the server
2786 			 * will check permissions when we perform the op.
2787 			 */
2788 			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2789 				goto out;
2790 	}
2791 
2792 force_lookup:
2793 	if (!NFS_PROTO(inode)->access)
2794 		goto out_notsup;
2795 
2796 	res = nfs_do_access(inode, cred, mask);
2797 out:
2798 	if (!res && (mask & MAY_EXEC))
2799 		res = nfs_execute_ok(inode, mask);
2800 
2801 	dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2802 		inode->i_sb->s_id, inode->i_ino, mask, res);
2803 	return res;
2804 out_notsup:
2805 	if (mask & MAY_NOT_BLOCK)
2806 		return -ECHILD;
2807 
2808 	res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2809 	if (res == 0)
2810 		res = generic_permission(inode, mask);
2811 	goto out;
2812 }
2813 EXPORT_SYMBOL_GPL(nfs_permission);
2814 
2815 /*
2816  * Local variables:
2817  *  version-control: t
2818  *  kept-new-versions: 5
2819  * End:
2820  */
2821