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/compat.h>
22 #include <linux/module.h>
23 #include <linux/time.h>
24 #include <linux/errno.h>
25 #include <linux/stat.h>
26 #include <linux/fcntl.h>
27 #include <linux/string.h>
28 #include <linux/kernel.h>
29 #include <linux/slab.h>
30 #include <linux/mm.h>
31 #include <linux/sunrpc/clnt.h>
32 #include <linux/nfs_fs.h>
33 #include <linux/nfs_mount.h>
34 #include <linux/pagemap.h>
35 #include <linux/pagevec.h>
36 #include <linux/namei.h>
37 #include <linux/mount.h>
38 #include <linux/swap.h>
39 #include <linux/sched.h>
40 #include <linux/kmemleak.h>
41 #include <linux/xattr.h>
42 #include <linux/hash.h>
43 
44 #include "delegation.h"
45 #include "iostat.h"
46 #include "internal.h"
47 #include "fscache.h"
48 
49 #include "nfstrace.h"
50 
51 /* #define NFS_DEBUG_VERBOSE 1 */
52 
53 static int nfs_opendir(struct inode *, struct file *);
54 static int nfs_closedir(struct inode *, struct file *);
55 static int nfs_readdir(struct file *, struct dir_context *);
56 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
57 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
58 static void nfs_readdir_free_folio(struct folio *);
59 
60 const struct file_operations nfs_dir_operations = {
61 	.llseek		= nfs_llseek_dir,
62 	.read		= generic_read_dir,
63 	.iterate_shared	= nfs_readdir,
64 	.open		= nfs_opendir,
65 	.release	= nfs_closedir,
66 	.fsync		= nfs_fsync_dir,
67 };
68 
69 const struct address_space_operations nfs_dir_aops = {
70 	.free_folio = nfs_readdir_free_folio,
71 };
72 
73 #define NFS_INIT_DTSIZE PAGE_SIZE
74 
75 static struct nfs_open_dir_context *
alloc_nfs_open_dir_context(struct inode * dir)76 alloc_nfs_open_dir_context(struct inode *dir)
77 {
78 	struct nfs_inode *nfsi = NFS_I(dir);
79 	struct nfs_open_dir_context *ctx;
80 
81 	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
82 	if (ctx != NULL) {
83 		ctx->attr_gencount = nfsi->attr_gencount;
84 		ctx->dtsize = NFS_INIT_DTSIZE;
85 		spin_lock(&dir->i_lock);
86 		if (list_empty(&nfsi->open_files) &&
87 		    (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
88 			nfs_set_cache_invalid(dir,
89 					      NFS_INO_INVALID_DATA |
90 						      NFS_INO_REVAL_FORCED);
91 		list_add_tail_rcu(&ctx->list, &nfsi->open_files);
92 		memcpy(ctx->verf, nfsi->cookieverf, sizeof(ctx->verf));
93 		spin_unlock(&dir->i_lock);
94 		return ctx;
95 	}
96 	return  ERR_PTR(-ENOMEM);
97 }
98 
put_nfs_open_dir_context(struct inode * dir,struct nfs_open_dir_context * ctx)99 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
100 {
101 	spin_lock(&dir->i_lock);
102 	list_del_rcu(&ctx->list);
103 	spin_unlock(&dir->i_lock);
104 	kfree_rcu(ctx, rcu_head);
105 }
106 
107 /*
108  * Open file
109  */
110 static int
nfs_opendir(struct inode * inode,struct file * filp)111 nfs_opendir(struct inode *inode, struct file *filp)
112 {
113 	int res = 0;
114 	struct nfs_open_dir_context *ctx;
115 
116 	dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
117 
118 	nfs_inc_stats(inode, NFSIOS_VFSOPEN);
119 
120 	ctx = alloc_nfs_open_dir_context(inode);
121 	if (IS_ERR(ctx)) {
122 		res = PTR_ERR(ctx);
123 		goto out;
124 	}
125 	filp->private_data = ctx;
126 out:
127 	return res;
128 }
129 
130 static int
nfs_closedir(struct inode * inode,struct file * filp)131 nfs_closedir(struct inode *inode, struct file *filp)
132 {
133 	put_nfs_open_dir_context(file_inode(filp), filp->private_data);
134 	return 0;
135 }
136 
137 struct nfs_cache_array_entry {
138 	u64 cookie;
139 	u64 ino;
140 	const char *name;
141 	unsigned int name_len;
142 	unsigned char d_type;
143 };
144 
145 struct nfs_cache_array {
146 	u64 change_attr;
147 	u64 last_cookie;
148 	unsigned int size;
149 	unsigned char page_full : 1,
150 		      page_is_eof : 1,
151 		      cookies_are_ordered : 1;
152 	struct nfs_cache_array_entry array[];
153 };
154 
155 struct nfs_readdir_descriptor {
156 	struct file	*file;
157 	struct page	*page;
158 	struct dir_context *ctx;
159 	pgoff_t		page_index;
160 	pgoff_t		page_index_max;
161 	u64		dir_cookie;
162 	u64		last_cookie;
163 	loff_t		current_index;
164 
165 	__be32		verf[NFS_DIR_VERIFIER_SIZE];
166 	unsigned long	dir_verifier;
167 	unsigned long	timestamp;
168 	unsigned long	gencount;
169 	unsigned long	attr_gencount;
170 	unsigned int	cache_entry_index;
171 	unsigned int	buffer_fills;
172 	unsigned int	dtsize;
173 	bool clear_cache;
174 	bool plus;
175 	bool eob;
176 	bool eof;
177 };
178 
nfs_set_dtsize(struct nfs_readdir_descriptor * desc,unsigned int sz)179 static void nfs_set_dtsize(struct nfs_readdir_descriptor *desc, unsigned int sz)
180 {
181 	struct nfs_server *server = NFS_SERVER(file_inode(desc->file));
182 	unsigned int maxsize = server->dtsize;
183 
184 	if (sz > maxsize)
185 		sz = maxsize;
186 	if (sz < NFS_MIN_FILE_IO_SIZE)
187 		sz = NFS_MIN_FILE_IO_SIZE;
188 	desc->dtsize = sz;
189 }
190 
nfs_shrink_dtsize(struct nfs_readdir_descriptor * desc)191 static void nfs_shrink_dtsize(struct nfs_readdir_descriptor *desc)
192 {
193 	nfs_set_dtsize(desc, desc->dtsize >> 1);
194 }
195 
nfs_grow_dtsize(struct nfs_readdir_descriptor * desc)196 static void nfs_grow_dtsize(struct nfs_readdir_descriptor *desc)
197 {
198 	nfs_set_dtsize(desc, desc->dtsize << 1);
199 }
200 
nfs_readdir_page_init_array(struct page * page,u64 last_cookie,u64 change_attr)201 static void nfs_readdir_page_init_array(struct page *page, u64 last_cookie,
202 					u64 change_attr)
203 {
204 	struct nfs_cache_array *array;
205 
206 	array = kmap_atomic(page);
207 	array->change_attr = change_attr;
208 	array->last_cookie = last_cookie;
209 	array->size = 0;
210 	array->page_full = 0;
211 	array->page_is_eof = 0;
212 	array->cookies_are_ordered = 1;
213 	kunmap_atomic(array);
214 }
215 
216 /*
217  * we are freeing strings created by nfs_add_to_readdir_array()
218  */
nfs_readdir_clear_array(struct page * page)219 static void nfs_readdir_clear_array(struct page *page)
220 {
221 	struct nfs_cache_array *array;
222 	unsigned int i;
223 
224 	array = kmap_atomic(page);
225 	for (i = 0; i < array->size; i++)
226 		kfree(array->array[i].name);
227 	array->size = 0;
228 	kunmap_atomic(array);
229 }
230 
nfs_readdir_free_folio(struct folio * folio)231 static void nfs_readdir_free_folio(struct folio *folio)
232 {
233 	nfs_readdir_clear_array(&folio->page);
234 }
235 
nfs_readdir_page_reinit_array(struct page * page,u64 last_cookie,u64 change_attr)236 static void nfs_readdir_page_reinit_array(struct page *page, u64 last_cookie,
237 					  u64 change_attr)
238 {
239 	nfs_readdir_clear_array(page);
240 	nfs_readdir_page_init_array(page, last_cookie, change_attr);
241 }
242 
243 static struct page *
nfs_readdir_page_array_alloc(u64 last_cookie,gfp_t gfp_flags)244 nfs_readdir_page_array_alloc(u64 last_cookie, gfp_t gfp_flags)
245 {
246 	struct page *page = alloc_page(gfp_flags);
247 	if (page)
248 		nfs_readdir_page_init_array(page, last_cookie, 0);
249 	return page;
250 }
251 
nfs_readdir_page_array_free(struct page * page)252 static void nfs_readdir_page_array_free(struct page *page)
253 {
254 	if (page) {
255 		nfs_readdir_clear_array(page);
256 		put_page(page);
257 	}
258 }
259 
nfs_readdir_array_index_cookie(struct nfs_cache_array * array)260 static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array)
261 {
262 	return array->size == 0 ? array->last_cookie : array->array[0].cookie;
263 }
264 
nfs_readdir_array_set_eof(struct nfs_cache_array * array)265 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
266 {
267 	array->page_is_eof = 1;
268 	array->page_full = 1;
269 }
270 
nfs_readdir_array_is_full(struct nfs_cache_array * array)271 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
272 {
273 	return array->page_full;
274 }
275 
276 /*
277  * the caller is responsible for freeing qstr.name
278  * when called by nfs_readdir_add_to_array, the strings will be freed in
279  * nfs_clear_readdir_array()
280  */
nfs_readdir_copy_name(const char * name,unsigned int len)281 static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
282 {
283 	const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
284 
285 	/*
286 	 * Avoid a kmemleak false positive. The pointer to the name is stored
287 	 * in a page cache page which kmemleak does not scan.
288 	 */
289 	if (ret != NULL)
290 		kmemleak_not_leak(ret);
291 	return ret;
292 }
293 
nfs_readdir_array_maxentries(void)294 static size_t nfs_readdir_array_maxentries(void)
295 {
296 	return (PAGE_SIZE - sizeof(struct nfs_cache_array)) /
297 	       sizeof(struct nfs_cache_array_entry);
298 }
299 
300 /*
301  * Check that the next array entry lies entirely within the page bounds
302  */
nfs_readdir_array_can_expand(struct nfs_cache_array * array)303 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
304 {
305 	if (array->page_full)
306 		return -ENOSPC;
307 	if (array->size == nfs_readdir_array_maxentries()) {
308 		array->page_full = 1;
309 		return -ENOSPC;
310 	}
311 	return 0;
312 }
313 
nfs_readdir_page_array_append(struct page * page,const struct nfs_entry * entry,u64 * cookie)314 static int nfs_readdir_page_array_append(struct page *page,
315 					 const struct nfs_entry *entry,
316 					 u64 *cookie)
317 {
318 	struct nfs_cache_array *array;
319 	struct nfs_cache_array_entry *cache_entry;
320 	const char *name;
321 	int ret = -ENOMEM;
322 
323 	name = nfs_readdir_copy_name(entry->name, entry->len);
324 
325 	array = kmap_atomic(page);
326 	if (!name)
327 		goto out;
328 	ret = nfs_readdir_array_can_expand(array);
329 	if (ret) {
330 		kfree(name);
331 		goto out;
332 	}
333 
334 	cache_entry = &array->array[array->size];
335 	cache_entry->cookie = array->last_cookie;
336 	cache_entry->ino = entry->ino;
337 	cache_entry->d_type = entry->d_type;
338 	cache_entry->name_len = entry->len;
339 	cache_entry->name = name;
340 	array->last_cookie = entry->cookie;
341 	if (array->last_cookie <= cache_entry->cookie)
342 		array->cookies_are_ordered = 0;
343 	array->size++;
344 	if (entry->eof != 0)
345 		nfs_readdir_array_set_eof(array);
346 out:
347 	*cookie = array->last_cookie;
348 	kunmap_atomic(array);
349 	return ret;
350 }
351 
352 #define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14)
353 /*
354  * Hash algorithm allowing content addressible access to sequences
355  * of directory cookies. Content is addressed by the value of the
356  * cookie index of the first readdir entry in a page.
357  *
358  * We select only the first 18 bits to avoid issues with excessive
359  * memory use for the page cache XArray. 18 bits should allow the caching
360  * of 262144 pages of sequences of readdir entries. Since each page holds
361  * 127 readdir entries for a typical 64-bit system, that works out to a
362  * cache of ~ 33 million entries per directory.
363  */
nfs_readdir_page_cookie_hash(u64 cookie)364 static pgoff_t nfs_readdir_page_cookie_hash(u64 cookie)
365 {
366 	if (cookie == 0)
367 		return 0;
368 	return hash_64(cookie, 18);
369 }
370 
nfs_readdir_page_validate(struct page * page,u64 last_cookie,u64 change_attr)371 static bool nfs_readdir_page_validate(struct page *page, u64 last_cookie,
372 				      u64 change_attr)
373 {
374 	struct nfs_cache_array *array = kmap_atomic(page);
375 	int ret = true;
376 
377 	if (array->change_attr != change_attr)
378 		ret = false;
379 	if (nfs_readdir_array_index_cookie(array) != last_cookie)
380 		ret = false;
381 	kunmap_atomic(array);
382 	return ret;
383 }
384 
nfs_readdir_page_unlock_and_put(struct page * page)385 static void nfs_readdir_page_unlock_and_put(struct page *page)
386 {
387 	unlock_page(page);
388 	put_page(page);
389 }
390 
nfs_readdir_page_init_and_validate(struct page * page,u64 cookie,u64 change_attr)391 static void nfs_readdir_page_init_and_validate(struct page *page, u64 cookie,
392 					       u64 change_attr)
393 {
394 	if (PageUptodate(page)) {
395 		if (nfs_readdir_page_validate(page, cookie, change_attr))
396 			return;
397 		nfs_readdir_clear_array(page);
398 	}
399 	nfs_readdir_page_init_array(page, cookie, change_attr);
400 	SetPageUptodate(page);
401 }
402 
nfs_readdir_page_get_locked(struct address_space * mapping,u64 cookie,u64 change_attr)403 static struct page *nfs_readdir_page_get_locked(struct address_space *mapping,
404 						u64 cookie, u64 change_attr)
405 {
406 	pgoff_t index = nfs_readdir_page_cookie_hash(cookie);
407 	struct page *page;
408 
409 	page = grab_cache_page(mapping, index);
410 	if (!page)
411 		return NULL;
412 	nfs_readdir_page_init_and_validate(page, cookie, change_attr);
413 	return page;
414 }
415 
nfs_readdir_page_last_cookie(struct page * page)416 static u64 nfs_readdir_page_last_cookie(struct page *page)
417 {
418 	struct nfs_cache_array *array;
419 	u64 ret;
420 
421 	array = kmap_atomic(page);
422 	ret = array->last_cookie;
423 	kunmap_atomic(array);
424 	return ret;
425 }
426 
nfs_readdir_page_needs_filling(struct page * page)427 static bool nfs_readdir_page_needs_filling(struct page *page)
428 {
429 	struct nfs_cache_array *array;
430 	bool ret;
431 
432 	array = kmap_atomic(page);
433 	ret = !nfs_readdir_array_is_full(array);
434 	kunmap_atomic(array);
435 	return ret;
436 }
437 
nfs_readdir_page_set_eof(struct page * page)438 static void nfs_readdir_page_set_eof(struct page *page)
439 {
440 	struct nfs_cache_array *array;
441 
442 	array = kmap_atomic(page);
443 	nfs_readdir_array_set_eof(array);
444 	kunmap_atomic(array);
445 }
446 
nfs_readdir_page_get_next(struct address_space * mapping,u64 cookie,u64 change_attr)447 static struct page *nfs_readdir_page_get_next(struct address_space *mapping,
448 					      u64 cookie, u64 change_attr)
449 {
450 	pgoff_t index = nfs_readdir_page_cookie_hash(cookie);
451 	struct page *page;
452 
453 	page = grab_cache_page_nowait(mapping, index);
454 	if (!page)
455 		return NULL;
456 	nfs_readdir_page_init_and_validate(page, cookie, change_attr);
457 	if (nfs_readdir_page_last_cookie(page) != cookie)
458 		nfs_readdir_page_reinit_array(page, cookie, change_attr);
459 	return page;
460 }
461 
462 static inline
is_32bit_api(void)463 int is_32bit_api(void)
464 {
465 #ifdef CONFIG_COMPAT
466 	return in_compat_syscall();
467 #else
468 	return (BITS_PER_LONG == 32);
469 #endif
470 }
471 
472 static
nfs_readdir_use_cookie(const struct file * filp)473 bool nfs_readdir_use_cookie(const struct file *filp)
474 {
475 	if ((filp->f_mode & FMODE_32BITHASH) ||
476 	    (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
477 		return false;
478 	return true;
479 }
480 
nfs_readdir_seek_next_array(struct nfs_cache_array * array,struct nfs_readdir_descriptor * desc)481 static void nfs_readdir_seek_next_array(struct nfs_cache_array *array,
482 					struct nfs_readdir_descriptor *desc)
483 {
484 	if (array->page_full) {
485 		desc->last_cookie = array->last_cookie;
486 		desc->current_index += array->size;
487 		desc->cache_entry_index = 0;
488 		desc->page_index++;
489 	} else
490 		desc->last_cookie = nfs_readdir_array_index_cookie(array);
491 }
492 
nfs_readdir_rewind_search(struct nfs_readdir_descriptor * desc)493 static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc)
494 {
495 	desc->current_index = 0;
496 	desc->last_cookie = 0;
497 	desc->page_index = 0;
498 }
499 
nfs_readdir_search_for_pos(struct nfs_cache_array * array,struct nfs_readdir_descriptor * desc)500 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
501 				      struct nfs_readdir_descriptor *desc)
502 {
503 	loff_t diff = desc->ctx->pos - desc->current_index;
504 	unsigned int index;
505 
506 	if (diff < 0)
507 		goto out_eof;
508 	if (diff >= array->size) {
509 		if (array->page_is_eof)
510 			goto out_eof;
511 		nfs_readdir_seek_next_array(array, desc);
512 		return -EAGAIN;
513 	}
514 
515 	index = (unsigned int)diff;
516 	desc->dir_cookie = array->array[index].cookie;
517 	desc->cache_entry_index = index;
518 	return 0;
519 out_eof:
520 	desc->eof = true;
521 	return -EBADCOOKIE;
522 }
523 
nfs_readdir_array_cookie_in_range(struct nfs_cache_array * array,u64 cookie)524 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
525 					      u64 cookie)
526 {
527 	if (!array->cookies_are_ordered)
528 		return true;
529 	/* Optimisation for monotonically increasing cookies */
530 	if (cookie >= array->last_cookie)
531 		return false;
532 	if (array->size && cookie < array->array[0].cookie)
533 		return false;
534 	return true;
535 }
536 
nfs_readdir_search_for_cookie(struct nfs_cache_array * array,struct nfs_readdir_descriptor * desc)537 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
538 					 struct nfs_readdir_descriptor *desc)
539 {
540 	unsigned int i;
541 	int status = -EAGAIN;
542 
543 	if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
544 		goto check_eof;
545 
546 	for (i = 0; i < array->size; i++) {
547 		if (array->array[i].cookie == desc->dir_cookie) {
548 			if (nfs_readdir_use_cookie(desc->file))
549 				desc->ctx->pos = desc->dir_cookie;
550 			else
551 				desc->ctx->pos = desc->current_index + i;
552 			desc->cache_entry_index = i;
553 			return 0;
554 		}
555 	}
556 check_eof:
557 	if (array->page_is_eof) {
558 		status = -EBADCOOKIE;
559 		if (desc->dir_cookie == array->last_cookie)
560 			desc->eof = true;
561 	} else
562 		nfs_readdir_seek_next_array(array, desc);
563 	return status;
564 }
565 
nfs_readdir_search_array(struct nfs_readdir_descriptor * desc)566 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
567 {
568 	struct nfs_cache_array *array;
569 	int status;
570 
571 	array = kmap_atomic(desc->page);
572 
573 	if (desc->dir_cookie == 0)
574 		status = nfs_readdir_search_for_pos(array, desc);
575 	else
576 		status = nfs_readdir_search_for_cookie(array, desc);
577 
578 	kunmap_atomic(array);
579 	return status;
580 }
581 
582 /* Fill a page with xdr information before transferring to the cache page */
nfs_readdir_xdr_filler(struct nfs_readdir_descriptor * desc,__be32 * verf,u64 cookie,struct page ** pages,size_t bufsize,__be32 * verf_res)583 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
584 				  __be32 *verf, u64 cookie,
585 				  struct page **pages, size_t bufsize,
586 				  __be32 *verf_res)
587 {
588 	struct inode *inode = file_inode(desc->file);
589 	struct nfs_readdir_arg arg = {
590 		.dentry = file_dentry(desc->file),
591 		.cred = desc->file->f_cred,
592 		.verf = verf,
593 		.cookie = cookie,
594 		.pages = pages,
595 		.page_len = bufsize,
596 		.plus = desc->plus,
597 	};
598 	struct nfs_readdir_res res = {
599 		.verf = verf_res,
600 	};
601 	unsigned long	timestamp, gencount;
602 	int		error;
603 
604  again:
605 	timestamp = jiffies;
606 	gencount = nfs_inc_attr_generation_counter();
607 	desc->dir_verifier = nfs_save_change_attribute(inode);
608 	error = NFS_PROTO(inode)->readdir(&arg, &res);
609 	if (error < 0) {
610 		/* We requested READDIRPLUS, but the server doesn't grok it */
611 		if (error == -ENOTSUPP && desc->plus) {
612 			NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
613 			desc->plus = arg.plus = false;
614 			goto again;
615 		}
616 		goto error;
617 	}
618 	desc->timestamp = timestamp;
619 	desc->gencount = gencount;
620 error:
621 	return error;
622 }
623 
xdr_decode(struct nfs_readdir_descriptor * desc,struct nfs_entry * entry,struct xdr_stream * xdr)624 static int xdr_decode(struct nfs_readdir_descriptor *desc,
625 		      struct nfs_entry *entry, struct xdr_stream *xdr)
626 {
627 	struct inode *inode = file_inode(desc->file);
628 	int error;
629 
630 	error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
631 	if (error)
632 		return error;
633 	entry->fattr->time_start = desc->timestamp;
634 	entry->fattr->gencount = desc->gencount;
635 	return 0;
636 }
637 
638 /* Match file and dirent using either filehandle or fileid
639  * Note: caller is responsible for checking the fsid
640  */
641 static
nfs_same_file(struct dentry * dentry,struct nfs_entry * entry)642 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
643 {
644 	struct inode *inode;
645 	struct nfs_inode *nfsi;
646 
647 	if (d_really_is_negative(dentry))
648 		return 0;
649 
650 	inode = d_inode(dentry);
651 	if (is_bad_inode(inode) || NFS_STALE(inode))
652 		return 0;
653 
654 	nfsi = NFS_I(inode);
655 	if (entry->fattr->fileid != nfsi->fileid)
656 		return 0;
657 	if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
658 		return 0;
659 	return 1;
660 }
661 
662 #define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL)
663 
nfs_use_readdirplus(struct inode * dir,struct dir_context * ctx,unsigned int cache_hits,unsigned int cache_misses)664 static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx,
665 				unsigned int cache_hits,
666 				unsigned int cache_misses)
667 {
668 	if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
669 		return false;
670 	if (ctx->pos == 0 ||
671 	    cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD)
672 		return true;
673 	return false;
674 }
675 
676 /*
677  * This function is called by the getattr code to request the
678  * use of readdirplus to accelerate any future lookups in the same
679  * directory.
680  */
nfs_readdir_record_entry_cache_hit(struct inode * dir)681 void nfs_readdir_record_entry_cache_hit(struct inode *dir)
682 {
683 	struct nfs_inode *nfsi = NFS_I(dir);
684 	struct nfs_open_dir_context *ctx;
685 
686 	if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
687 	    S_ISDIR(dir->i_mode)) {
688 		rcu_read_lock();
689 		list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
690 			atomic_inc(&ctx->cache_hits);
691 		rcu_read_unlock();
692 	}
693 }
694 
695 /*
696  * This function is mainly for use by nfs_getattr().
697  *
698  * If this is an 'ls -l', we want to force use of readdirplus.
699  */
nfs_readdir_record_entry_cache_miss(struct inode * dir)700 void nfs_readdir_record_entry_cache_miss(struct inode *dir)
701 {
702 	struct nfs_inode *nfsi = NFS_I(dir);
703 	struct nfs_open_dir_context *ctx;
704 
705 	if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
706 	    S_ISDIR(dir->i_mode)) {
707 		rcu_read_lock();
708 		list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
709 			atomic_inc(&ctx->cache_misses);
710 		rcu_read_unlock();
711 	}
712 }
713 
nfs_lookup_advise_force_readdirplus(struct inode * dir,unsigned int flags)714 static void nfs_lookup_advise_force_readdirplus(struct inode *dir,
715 						unsigned int flags)
716 {
717 	if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
718 		return;
719 	if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL))
720 		return;
721 	nfs_readdir_record_entry_cache_miss(dir);
722 }
723 
724 static
nfs_prime_dcache(struct dentry * parent,struct nfs_entry * entry,unsigned long dir_verifier)725 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
726 		unsigned long dir_verifier)
727 {
728 	struct qstr filename = QSTR_INIT(entry->name, entry->len);
729 	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
730 	struct dentry *dentry;
731 	struct dentry *alias;
732 	struct inode *inode;
733 	int status;
734 
735 	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
736 		return;
737 	if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
738 		return;
739 	if (filename.len == 0)
740 		return;
741 	/* Validate that the name doesn't contain any illegal '\0' */
742 	if (strnlen(filename.name, filename.len) != filename.len)
743 		return;
744 	/* ...or '/' */
745 	if (strnchr(filename.name, filename.len, '/'))
746 		return;
747 	if (filename.name[0] == '.') {
748 		if (filename.len == 1)
749 			return;
750 		if (filename.len == 2 && filename.name[1] == '.')
751 			return;
752 	}
753 	filename.hash = full_name_hash(parent, filename.name, filename.len);
754 
755 	dentry = d_lookup(parent, &filename);
756 again:
757 	if (!dentry) {
758 		dentry = d_alloc_parallel(parent, &filename, &wq);
759 		if (IS_ERR(dentry))
760 			return;
761 	}
762 	if (!d_in_lookup(dentry)) {
763 		/* Is there a mountpoint here? If so, just exit */
764 		if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
765 					&entry->fattr->fsid))
766 			goto out;
767 		if (nfs_same_file(dentry, entry)) {
768 			if (!entry->fh->size)
769 				goto out;
770 			nfs_set_verifier(dentry, dir_verifier);
771 			status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
772 			if (!status)
773 				nfs_setsecurity(d_inode(dentry), entry->fattr);
774 			trace_nfs_readdir_lookup_revalidate(d_inode(parent),
775 							    dentry, 0, status);
776 			goto out;
777 		} else {
778 			trace_nfs_readdir_lookup_revalidate_failed(
779 				d_inode(parent), dentry, 0);
780 			d_invalidate(dentry);
781 			dput(dentry);
782 			dentry = NULL;
783 			goto again;
784 		}
785 	}
786 	if (!entry->fh->size) {
787 		d_lookup_done(dentry);
788 		goto out;
789 	}
790 
791 	inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
792 	alias = d_splice_alias(inode, dentry);
793 	d_lookup_done(dentry);
794 	if (alias) {
795 		if (IS_ERR(alias))
796 			goto out;
797 		dput(dentry);
798 		dentry = alias;
799 	}
800 	nfs_set_verifier(dentry, dir_verifier);
801 	trace_nfs_readdir_lookup(d_inode(parent), dentry, 0);
802 out:
803 	dput(dentry);
804 }
805 
nfs_readdir_entry_decode(struct nfs_readdir_descriptor * desc,struct nfs_entry * entry,struct xdr_stream * stream)806 static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc,
807 				    struct nfs_entry *entry,
808 				    struct xdr_stream *stream)
809 {
810 	int ret;
811 
812 	if (entry->fattr->label)
813 		entry->fattr->label->len = NFS4_MAXLABELLEN;
814 	ret = xdr_decode(desc, entry, stream);
815 	if (ret || !desc->plus)
816 		return ret;
817 	nfs_prime_dcache(file_dentry(desc->file), entry, desc->dir_verifier);
818 	return 0;
819 }
820 
821 /* Perform conversion from xdr to cache array */
nfs_readdir_page_filler(struct nfs_readdir_descriptor * desc,struct nfs_entry * entry,struct page ** xdr_pages,unsigned int buflen,struct page ** arrays,size_t narrays,u64 change_attr)822 static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc,
823 				   struct nfs_entry *entry,
824 				   struct page **xdr_pages, unsigned int buflen,
825 				   struct page **arrays, size_t narrays,
826 				   u64 change_attr)
827 {
828 	struct address_space *mapping = desc->file->f_mapping;
829 	struct xdr_stream stream;
830 	struct xdr_buf buf;
831 	struct page *scratch, *new, *page = *arrays;
832 	u64 cookie;
833 	int status;
834 
835 	scratch = alloc_page(GFP_KERNEL);
836 	if (scratch == NULL)
837 		return -ENOMEM;
838 
839 	xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
840 	xdr_set_scratch_page(&stream, scratch);
841 
842 	do {
843 		status = nfs_readdir_entry_decode(desc, entry, &stream);
844 		if (status != 0)
845 			break;
846 
847 		status = nfs_readdir_page_array_append(page, entry, &cookie);
848 		if (status != -ENOSPC)
849 			continue;
850 
851 		if (page->mapping != mapping) {
852 			if (!--narrays)
853 				break;
854 			new = nfs_readdir_page_array_alloc(cookie, GFP_KERNEL);
855 			if (!new)
856 				break;
857 			arrays++;
858 			*arrays = page = new;
859 		} else {
860 			new = nfs_readdir_page_get_next(mapping, cookie,
861 							change_attr);
862 			if (!new)
863 				break;
864 			if (page != *arrays)
865 				nfs_readdir_page_unlock_and_put(page);
866 			page = new;
867 		}
868 		desc->page_index_max++;
869 		status = nfs_readdir_page_array_append(page, entry, &cookie);
870 	} while (!status && !entry->eof);
871 
872 	switch (status) {
873 	case -EBADCOOKIE:
874 		if (!entry->eof)
875 			break;
876 		nfs_readdir_page_set_eof(page);
877 		fallthrough;
878 	case -EAGAIN:
879 		status = 0;
880 		break;
881 	case -ENOSPC:
882 		status = 0;
883 		if (!desc->plus)
884 			break;
885 		while (!nfs_readdir_entry_decode(desc, entry, &stream))
886 			;
887 	}
888 
889 	if (page != *arrays)
890 		nfs_readdir_page_unlock_and_put(page);
891 
892 	put_page(scratch);
893 	return status;
894 }
895 
nfs_readdir_free_pages(struct page ** pages,size_t npages)896 static void nfs_readdir_free_pages(struct page **pages, size_t npages)
897 {
898 	while (npages--)
899 		put_page(pages[npages]);
900 	kfree(pages);
901 }
902 
903 /*
904  * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
905  * to nfs_readdir_free_pages()
906  */
nfs_readdir_alloc_pages(size_t npages)907 static struct page **nfs_readdir_alloc_pages(size_t npages)
908 {
909 	struct page **pages;
910 	size_t i;
911 
912 	pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
913 	if (!pages)
914 		return NULL;
915 	for (i = 0; i < npages; i++) {
916 		struct page *page = alloc_page(GFP_KERNEL);
917 		if (page == NULL)
918 			goto out_freepages;
919 		pages[i] = page;
920 	}
921 	return pages;
922 
923 out_freepages:
924 	nfs_readdir_free_pages(pages, i);
925 	return NULL;
926 }
927 
nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor * desc,__be32 * verf_arg,__be32 * verf_res,struct page ** arrays,size_t narrays)928 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
929 				    __be32 *verf_arg, __be32 *verf_res,
930 				    struct page **arrays, size_t narrays)
931 {
932 	u64 change_attr;
933 	struct page **pages;
934 	struct page *page = *arrays;
935 	struct nfs_entry *entry;
936 	size_t array_size;
937 	struct inode *inode = file_inode(desc->file);
938 	unsigned int dtsize = desc->dtsize;
939 	unsigned int pglen;
940 	int status = -ENOMEM;
941 
942 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
943 	if (!entry)
944 		return -ENOMEM;
945 	entry->cookie = nfs_readdir_page_last_cookie(page);
946 	entry->fh = nfs_alloc_fhandle();
947 	entry->fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
948 	entry->server = NFS_SERVER(inode);
949 	if (entry->fh == NULL || entry->fattr == NULL)
950 		goto out;
951 
952 	array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
953 	pages = nfs_readdir_alloc_pages(array_size);
954 	if (!pages)
955 		goto out;
956 
957 	change_attr = inode_peek_iversion_raw(inode);
958 	status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, pages,
959 					dtsize, verf_res);
960 	if (status < 0)
961 		goto free_pages;
962 
963 	pglen = status;
964 	if (pglen != 0)
965 		status = nfs_readdir_page_filler(desc, entry, pages, pglen,
966 						 arrays, narrays, change_attr);
967 	else
968 		nfs_readdir_page_set_eof(page);
969 	desc->buffer_fills++;
970 
971 free_pages:
972 	nfs_readdir_free_pages(pages, array_size);
973 out:
974 	nfs_free_fattr(entry->fattr);
975 	nfs_free_fhandle(entry->fh);
976 	kfree(entry);
977 	return status;
978 }
979 
nfs_readdir_page_put(struct nfs_readdir_descriptor * desc)980 static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc)
981 {
982 	put_page(desc->page);
983 	desc->page = NULL;
984 }
985 
986 static void
nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor * desc)987 nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
988 {
989 	unlock_page(desc->page);
990 	nfs_readdir_page_put(desc);
991 }
992 
993 static struct page *
nfs_readdir_page_get_cached(struct nfs_readdir_descriptor * desc)994 nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc)
995 {
996 	struct address_space *mapping = desc->file->f_mapping;
997 	u64 change_attr = inode_peek_iversion_raw(mapping->host);
998 	u64 cookie = desc->last_cookie;
999 	struct page *page;
1000 
1001 	page = nfs_readdir_page_get_locked(mapping, cookie, change_attr);
1002 	if (!page)
1003 		return NULL;
1004 	if (desc->clear_cache && !nfs_readdir_page_needs_filling(page))
1005 		nfs_readdir_page_reinit_array(page, cookie, change_attr);
1006 	return page;
1007 }
1008 
1009 /*
1010  * Returns 0 if desc->dir_cookie was found on page desc->page_index
1011  * and locks the page to prevent removal from the page cache.
1012  */
find_and_lock_cache_page(struct nfs_readdir_descriptor * desc)1013 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
1014 {
1015 	struct inode *inode = file_inode(desc->file);
1016 	struct nfs_inode *nfsi = NFS_I(inode);
1017 	__be32 verf[NFS_DIR_VERIFIER_SIZE];
1018 	int res;
1019 
1020 	desc->page = nfs_readdir_page_get_cached(desc);
1021 	if (!desc->page)
1022 		return -ENOMEM;
1023 	if (nfs_readdir_page_needs_filling(desc->page)) {
1024 		/* Grow the dtsize if we had to go back for more pages */
1025 		if (desc->page_index == desc->page_index_max)
1026 			nfs_grow_dtsize(desc);
1027 		desc->page_index_max = desc->page_index;
1028 		trace_nfs_readdir_cache_fill(desc->file, nfsi->cookieverf,
1029 					     desc->last_cookie,
1030 					     desc->page->index, desc->dtsize);
1031 		res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
1032 					       &desc->page, 1);
1033 		if (res < 0) {
1034 			nfs_readdir_page_unlock_and_put_cached(desc);
1035 			trace_nfs_readdir_cache_fill_done(inode, res);
1036 			if (res == -EBADCOOKIE || res == -ENOTSYNC) {
1037 				invalidate_inode_pages2(desc->file->f_mapping);
1038 				nfs_readdir_rewind_search(desc);
1039 				trace_nfs_readdir_invalidate_cache_range(
1040 					inode, 0, MAX_LFS_FILESIZE);
1041 				return -EAGAIN;
1042 			}
1043 			return res;
1044 		}
1045 		/*
1046 		 * Set the cookie verifier if the page cache was empty
1047 		 */
1048 		if (desc->last_cookie == 0 &&
1049 		    memcmp(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf))) {
1050 			memcpy(nfsi->cookieverf, verf,
1051 			       sizeof(nfsi->cookieverf));
1052 			invalidate_inode_pages2_range(desc->file->f_mapping, 1,
1053 						      -1);
1054 			trace_nfs_readdir_invalidate_cache_range(
1055 				inode, 1, MAX_LFS_FILESIZE);
1056 		}
1057 		desc->clear_cache = false;
1058 	}
1059 	res = nfs_readdir_search_array(desc);
1060 	if (res == 0)
1061 		return 0;
1062 	nfs_readdir_page_unlock_and_put_cached(desc);
1063 	return res;
1064 }
1065 
1066 /* Search for desc->dir_cookie from the beginning of the page cache */
readdir_search_pagecache(struct nfs_readdir_descriptor * desc)1067 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
1068 {
1069 	int res;
1070 
1071 	do {
1072 		res = find_and_lock_cache_page(desc);
1073 	} while (res == -EAGAIN);
1074 	return res;
1075 }
1076 
1077 /*
1078  * Once we've found the start of the dirent within a page: fill 'er up...
1079  */
nfs_do_filldir(struct nfs_readdir_descriptor * desc,const __be32 * verf)1080 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
1081 			   const __be32 *verf)
1082 {
1083 	struct file	*file = desc->file;
1084 	struct nfs_cache_array *array;
1085 	unsigned int i;
1086 
1087 	array = kmap_local_page(desc->page);
1088 	for (i = desc->cache_entry_index; i < array->size; i++) {
1089 		struct nfs_cache_array_entry *ent;
1090 
1091 		ent = &array->array[i];
1092 		if (!dir_emit(desc->ctx, ent->name, ent->name_len,
1093 		    nfs_compat_user_ino64(ent->ino), ent->d_type)) {
1094 			desc->eob = true;
1095 			break;
1096 		}
1097 		memcpy(desc->verf, verf, sizeof(desc->verf));
1098 		if (i == array->size - 1) {
1099 			desc->dir_cookie = array->last_cookie;
1100 			nfs_readdir_seek_next_array(array, desc);
1101 		} else {
1102 			desc->dir_cookie = array->array[i + 1].cookie;
1103 			desc->last_cookie = array->array[0].cookie;
1104 		}
1105 		if (nfs_readdir_use_cookie(file))
1106 			desc->ctx->pos = desc->dir_cookie;
1107 		else
1108 			desc->ctx->pos++;
1109 	}
1110 	if (array->page_is_eof)
1111 		desc->eof = !desc->eob;
1112 
1113 	kunmap_local(array);
1114 	dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1115 			(unsigned long long)desc->dir_cookie);
1116 }
1117 
1118 /*
1119  * If we cannot find a cookie in our cache, we suspect that this is
1120  * because it points to a deleted file, so we ask the server to return
1121  * whatever it thinks is the next entry. We then feed this to filldir.
1122  * If all goes well, we should then be able to find our way round the
1123  * cache on the next call to readdir_search_pagecache();
1124  *
1125  * NOTE: we cannot add the anonymous page to the pagecache because
1126  *	 the data it contains might not be page aligned. Besides,
1127  *	 we should already have a complete representation of the
1128  *	 directory in the page cache by the time we get here.
1129  */
uncached_readdir(struct nfs_readdir_descriptor * desc)1130 static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1131 {
1132 	struct page	**arrays;
1133 	size_t		i, sz = 512;
1134 	__be32		verf[NFS_DIR_VERIFIER_SIZE];
1135 	int		status = -ENOMEM;
1136 
1137 	dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1138 			(unsigned long long)desc->dir_cookie);
1139 
1140 	arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1141 	if (!arrays)
1142 		goto out;
1143 	arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL);
1144 	if (!arrays[0])
1145 		goto out;
1146 
1147 	desc->page_index = 0;
1148 	desc->cache_entry_index = 0;
1149 	desc->last_cookie = desc->dir_cookie;
1150 	desc->page_index_max = 0;
1151 
1152 	trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie,
1153 				   -1, desc->dtsize);
1154 
1155 	status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1156 	if (status < 0) {
1157 		trace_nfs_readdir_uncached_done(file_inode(desc->file), status);
1158 		goto out_free;
1159 	}
1160 
1161 	for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1162 		desc->page = arrays[i];
1163 		nfs_do_filldir(desc, verf);
1164 	}
1165 	desc->page = NULL;
1166 
1167 	/*
1168 	 * Grow the dtsize if we have to go back for more pages,
1169 	 * or shrink it if we're reading too many.
1170 	 */
1171 	if (!desc->eof) {
1172 		if (!desc->eob)
1173 			nfs_grow_dtsize(desc);
1174 		else if (desc->buffer_fills == 1 &&
1175 			 i < (desc->page_index_max >> 1))
1176 			nfs_shrink_dtsize(desc);
1177 	}
1178 out_free:
1179 	for (i = 0; i < sz && arrays[i]; i++)
1180 		nfs_readdir_page_array_free(arrays[i]);
1181 out:
1182 	if (!nfs_readdir_use_cookie(desc->file))
1183 		nfs_readdir_rewind_search(desc);
1184 	desc->page_index_max = -1;
1185 	kfree(arrays);
1186 	dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1187 	return status;
1188 }
1189 
1190 #define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1191 
nfs_readdir_handle_cache_misses(struct inode * inode,struct nfs_readdir_descriptor * desc,unsigned int cache_misses,bool force_clear)1192 static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1193 					    struct nfs_readdir_descriptor *desc,
1194 					    unsigned int cache_misses,
1195 					    bool force_clear)
1196 {
1197 	if (desc->ctx->pos == 0 || !desc->plus)
1198 		return false;
1199 	if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1200 		return false;
1201 	trace_nfs_readdir_force_readdirplus(inode);
1202 	return true;
1203 }
1204 
1205 /* The file offset position represents the dirent entry number.  A
1206    last cookie cache takes care of the common case of reading the
1207    whole directory.
1208  */
nfs_readdir(struct file * file,struct dir_context * ctx)1209 static int nfs_readdir(struct file *file, struct dir_context *ctx)
1210 {
1211 	struct dentry	*dentry = file_dentry(file);
1212 	struct inode	*inode = d_inode(dentry);
1213 	struct nfs_inode *nfsi = NFS_I(inode);
1214 	struct nfs_open_dir_context *dir_ctx = file->private_data;
1215 	struct nfs_readdir_descriptor *desc;
1216 	unsigned int cache_hits, cache_misses;
1217 	bool force_clear;
1218 	int res;
1219 
1220 	dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1221 			file, (long long)ctx->pos);
1222 	nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1223 
1224 	/*
1225 	 * ctx->pos points to the dirent entry number.
1226 	 * *desc->dir_cookie has the cookie for the next entry. We have
1227 	 * to either find the entry with the appropriate number or
1228 	 * revalidate the cookie.
1229 	 */
1230 	nfs_revalidate_mapping(inode, file->f_mapping);
1231 
1232 	res = -ENOMEM;
1233 	desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1234 	if (!desc)
1235 		goto out;
1236 	desc->file = file;
1237 	desc->ctx = ctx;
1238 	desc->page_index_max = -1;
1239 
1240 	spin_lock(&file->f_lock);
1241 	desc->dir_cookie = dir_ctx->dir_cookie;
1242 	desc->page_index = dir_ctx->page_index;
1243 	desc->last_cookie = dir_ctx->last_cookie;
1244 	desc->attr_gencount = dir_ctx->attr_gencount;
1245 	desc->eof = dir_ctx->eof;
1246 	nfs_set_dtsize(desc, dir_ctx->dtsize);
1247 	memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1248 	cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0);
1249 	cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0);
1250 	force_clear = dir_ctx->force_clear;
1251 	spin_unlock(&file->f_lock);
1252 
1253 	if (desc->eof) {
1254 		res = 0;
1255 		goto out_free;
1256 	}
1257 
1258 	desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses);
1259 	force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1260 						      force_clear);
1261 	desc->clear_cache = force_clear;
1262 
1263 	do {
1264 		res = readdir_search_pagecache(desc);
1265 
1266 		if (res == -EBADCOOKIE) {
1267 			res = 0;
1268 			/* This means either end of directory */
1269 			if (desc->dir_cookie && !desc->eof) {
1270 				/* Or that the server has 'lost' a cookie */
1271 				res = uncached_readdir(desc);
1272 				if (res == 0)
1273 					continue;
1274 				if (res == -EBADCOOKIE || res == -ENOTSYNC)
1275 					res = 0;
1276 			}
1277 			break;
1278 		}
1279 		if (res == -ETOOSMALL && desc->plus) {
1280 			nfs_zap_caches(inode);
1281 			desc->plus = false;
1282 			desc->eof = false;
1283 			continue;
1284 		}
1285 		if (res < 0)
1286 			break;
1287 
1288 		nfs_do_filldir(desc, nfsi->cookieverf);
1289 		nfs_readdir_page_unlock_and_put_cached(desc);
1290 		if (desc->page_index == desc->page_index_max)
1291 			desc->clear_cache = force_clear;
1292 	} while (!desc->eob && !desc->eof);
1293 
1294 	spin_lock(&file->f_lock);
1295 	dir_ctx->dir_cookie = desc->dir_cookie;
1296 	dir_ctx->last_cookie = desc->last_cookie;
1297 	dir_ctx->attr_gencount = desc->attr_gencount;
1298 	dir_ctx->page_index = desc->page_index;
1299 	dir_ctx->force_clear = force_clear;
1300 	dir_ctx->eof = desc->eof;
1301 	dir_ctx->dtsize = desc->dtsize;
1302 	memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1303 	spin_unlock(&file->f_lock);
1304 out_free:
1305 	kfree(desc);
1306 
1307 out:
1308 	dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1309 	return res;
1310 }
1311 
nfs_llseek_dir(struct file * filp,loff_t offset,int whence)1312 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1313 {
1314 	struct nfs_open_dir_context *dir_ctx = filp->private_data;
1315 
1316 	dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1317 			filp, offset, whence);
1318 
1319 	switch (whence) {
1320 	default:
1321 		return -EINVAL;
1322 	case SEEK_SET:
1323 		if (offset < 0)
1324 			return -EINVAL;
1325 		spin_lock(&filp->f_lock);
1326 		break;
1327 	case SEEK_CUR:
1328 		if (offset == 0)
1329 			return filp->f_pos;
1330 		spin_lock(&filp->f_lock);
1331 		offset += filp->f_pos;
1332 		if (offset < 0) {
1333 			spin_unlock(&filp->f_lock);
1334 			return -EINVAL;
1335 		}
1336 	}
1337 	if (offset != filp->f_pos) {
1338 		filp->f_pos = offset;
1339 		dir_ctx->page_index = 0;
1340 		if (!nfs_readdir_use_cookie(filp)) {
1341 			dir_ctx->dir_cookie = 0;
1342 			dir_ctx->last_cookie = 0;
1343 		} else {
1344 			dir_ctx->dir_cookie = offset;
1345 			dir_ctx->last_cookie = offset;
1346 		}
1347 		dir_ctx->eof = false;
1348 	}
1349 	spin_unlock(&filp->f_lock);
1350 	return offset;
1351 }
1352 
1353 /*
1354  * All directory operations under NFS are synchronous, so fsync()
1355  * is a dummy operation.
1356  */
nfs_fsync_dir(struct file * filp,loff_t start,loff_t end,int datasync)1357 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1358 			 int datasync)
1359 {
1360 	dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1361 
1362 	nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1363 	return 0;
1364 }
1365 
1366 /**
1367  * nfs_force_lookup_revalidate - Mark the directory as having changed
1368  * @dir: pointer to directory inode
1369  *
1370  * This forces the revalidation code in nfs_lookup_revalidate() to do a
1371  * full lookup on all child dentries of 'dir' whenever a change occurs
1372  * on the server that might have invalidated our dcache.
1373  *
1374  * Note that we reserve bit '0' as a tag to let us know when a dentry
1375  * was revalidated while holding a delegation on its inode.
1376  *
1377  * The caller should be holding dir->i_lock
1378  */
nfs_force_lookup_revalidate(struct inode * dir)1379 void nfs_force_lookup_revalidate(struct inode *dir)
1380 {
1381 	NFS_I(dir)->cache_change_attribute += 2;
1382 }
1383 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1384 
1385 /**
1386  * nfs_verify_change_attribute - Detects NFS remote directory changes
1387  * @dir: pointer to parent directory inode
1388  * @verf: previously saved change attribute
1389  *
1390  * Return "false" if the verifiers doesn't match the change attribute.
1391  * This would usually indicate that the directory contents have changed on
1392  * the server, and that any dentries need revalidating.
1393  */
nfs_verify_change_attribute(struct inode * dir,unsigned long verf)1394 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1395 {
1396 	return (verf & ~1UL) == nfs_save_change_attribute(dir);
1397 }
1398 
nfs_set_verifier_delegated(unsigned long * verf)1399 static void nfs_set_verifier_delegated(unsigned long *verf)
1400 {
1401 	*verf |= 1UL;
1402 }
1403 
1404 #if IS_ENABLED(CONFIG_NFS_V4)
nfs_unset_verifier_delegated(unsigned long * verf)1405 static void nfs_unset_verifier_delegated(unsigned long *verf)
1406 {
1407 	*verf &= ~1UL;
1408 }
1409 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1410 
nfs_test_verifier_delegated(unsigned long verf)1411 static bool nfs_test_verifier_delegated(unsigned long verf)
1412 {
1413 	return verf & 1;
1414 }
1415 
nfs_verifier_is_delegated(struct dentry * dentry)1416 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1417 {
1418 	return nfs_test_verifier_delegated(dentry->d_time);
1419 }
1420 
nfs_set_verifier_locked(struct dentry * dentry,unsigned long verf)1421 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1422 {
1423 	struct inode *inode = d_inode(dentry);
1424 	struct inode *dir = d_inode(dentry->d_parent);
1425 
1426 	if (!nfs_verify_change_attribute(dir, verf))
1427 		return;
1428 	if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1429 		nfs_set_verifier_delegated(&verf);
1430 	dentry->d_time = verf;
1431 }
1432 
1433 /**
1434  * nfs_set_verifier - save a parent directory verifier in the dentry
1435  * @dentry: pointer to dentry
1436  * @verf: verifier to save
1437  *
1438  * Saves the parent directory verifier in @dentry. If the inode has
1439  * a delegation, we also tag the dentry as having been revalidated
1440  * while holding a delegation so that we know we don't have to
1441  * look it up again after a directory change.
1442  */
nfs_set_verifier(struct dentry * dentry,unsigned long verf)1443 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1444 {
1445 
1446 	spin_lock(&dentry->d_lock);
1447 	nfs_set_verifier_locked(dentry, verf);
1448 	spin_unlock(&dentry->d_lock);
1449 }
1450 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1451 
1452 #if IS_ENABLED(CONFIG_NFS_V4)
1453 /**
1454  * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1455  * @inode: pointer to inode
1456  *
1457  * Iterates through the dentries in the inode alias list and clears
1458  * the tag used to indicate that the dentry has been revalidated
1459  * while holding a delegation.
1460  * This function is intended for use when the delegation is being
1461  * returned or revoked.
1462  */
nfs_clear_verifier_delegated(struct inode * inode)1463 void nfs_clear_verifier_delegated(struct inode *inode)
1464 {
1465 	struct dentry *alias;
1466 
1467 	if (!inode)
1468 		return;
1469 	spin_lock(&inode->i_lock);
1470 	hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1471 		spin_lock(&alias->d_lock);
1472 		nfs_unset_verifier_delegated(&alias->d_time);
1473 		spin_unlock(&alias->d_lock);
1474 	}
1475 	spin_unlock(&inode->i_lock);
1476 }
1477 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1478 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1479 
nfs_dentry_verify_change(struct inode * dir,struct dentry * dentry)1480 static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1481 {
1482 	if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) &&
1483 	    d_really_is_negative(dentry))
1484 		return dentry->d_time == inode_peek_iversion_raw(dir);
1485 	return nfs_verify_change_attribute(dir, dentry->d_time);
1486 }
1487 
1488 /*
1489  * A check for whether or not the parent directory has changed.
1490  * In the case it has, we assume that the dentries are untrustworthy
1491  * and may need to be looked up again.
1492  * If rcu_walk prevents us from performing a full check, return 0.
1493  */
nfs_check_verifier(struct inode * dir,struct dentry * dentry,int rcu_walk)1494 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1495 			      int rcu_walk)
1496 {
1497 	if (IS_ROOT(dentry))
1498 		return 1;
1499 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1500 		return 0;
1501 	if (!nfs_dentry_verify_change(dir, dentry))
1502 		return 0;
1503 	/* Revalidate nfsi->cache_change_attribute before we declare a match */
1504 	if (nfs_mapping_need_revalidate_inode(dir)) {
1505 		if (rcu_walk)
1506 			return 0;
1507 		if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1508 			return 0;
1509 	}
1510 	if (!nfs_dentry_verify_change(dir, dentry))
1511 		return 0;
1512 	return 1;
1513 }
1514 
1515 /*
1516  * Use intent information to check whether or not we're going to do
1517  * an O_EXCL create using this path component.
1518  */
nfs_is_exclusive_create(struct inode * dir,unsigned int flags)1519 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1520 {
1521 	if (NFS_PROTO(dir)->version == 2)
1522 		return 0;
1523 	return flags & LOOKUP_EXCL;
1524 }
1525 
1526 /*
1527  * Inode and filehandle revalidation for lookups.
1528  *
1529  * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1530  * or if the intent information indicates that we're about to open this
1531  * particular file and the "nocto" mount flag is not set.
1532  *
1533  */
1534 static
nfs_lookup_verify_inode(struct inode * inode,unsigned int flags)1535 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1536 {
1537 	struct nfs_server *server = NFS_SERVER(inode);
1538 	int ret;
1539 
1540 	if (IS_AUTOMOUNT(inode))
1541 		return 0;
1542 
1543 	if (flags & LOOKUP_OPEN) {
1544 		switch (inode->i_mode & S_IFMT) {
1545 		case S_IFREG:
1546 			/* A NFSv4 OPEN will revalidate later */
1547 			if (server->caps & NFS_CAP_ATOMIC_OPEN)
1548 				goto out;
1549 			fallthrough;
1550 		case S_IFDIR:
1551 			if (server->flags & NFS_MOUNT_NOCTO)
1552 				break;
1553 			/* NFS close-to-open cache consistency validation */
1554 			goto out_force;
1555 		}
1556 	}
1557 
1558 	/* VFS wants an on-the-wire revalidation */
1559 	if (flags & LOOKUP_REVAL)
1560 		goto out_force;
1561 out:
1562 	if (inode->i_nlink > 0 ||
1563 	    (inode->i_nlink == 0 &&
1564 	     test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1565 		return 0;
1566 	else
1567 		return -ESTALE;
1568 out_force:
1569 	if (flags & LOOKUP_RCU)
1570 		return -ECHILD;
1571 	ret = __nfs_revalidate_inode(server, inode);
1572 	if (ret != 0)
1573 		return ret;
1574 	goto out;
1575 }
1576 
nfs_mark_dir_for_revalidate(struct inode * inode)1577 static void nfs_mark_dir_for_revalidate(struct inode *inode)
1578 {
1579 	spin_lock(&inode->i_lock);
1580 	nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1581 	spin_unlock(&inode->i_lock);
1582 }
1583 
1584 /*
1585  * We judge how long we want to trust negative
1586  * dentries by looking at the parent inode mtime.
1587  *
1588  * If parent mtime has changed, we revalidate, else we wait for a
1589  * period corresponding to the parent's attribute cache timeout value.
1590  *
1591  * If LOOKUP_RCU prevents us from performing a full check, return 1
1592  * suggesting a reval is needed.
1593  *
1594  * Note that when creating a new file, or looking up a rename target,
1595  * then it shouldn't be necessary to revalidate a negative dentry.
1596  */
1597 static inline
nfs_neg_need_reval(struct inode * dir,struct dentry * dentry,unsigned int flags)1598 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1599 		       unsigned int flags)
1600 {
1601 	if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1602 		return 0;
1603 	if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1604 		return 1;
1605 	/* Case insensitive server? Revalidate negative dentries */
1606 	if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1607 		return 1;
1608 	return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1609 }
1610 
1611 static int
nfs_lookup_revalidate_done(struct inode * dir,struct dentry * dentry,struct inode * inode,int error)1612 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1613 			   struct inode *inode, int error)
1614 {
1615 	switch (error) {
1616 	case 1:
1617 		break;
1618 	case 0:
1619 		/*
1620 		 * We can't d_drop the root of a disconnected tree:
1621 		 * its d_hash is on the s_anon list and d_drop() would hide
1622 		 * it from shrink_dcache_for_unmount(), leading to busy
1623 		 * inodes on unmount and further oopses.
1624 		 */
1625 		if (inode && IS_ROOT(dentry))
1626 			error = 1;
1627 		break;
1628 	}
1629 	trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error);
1630 	return error;
1631 }
1632 
1633 static int
nfs_lookup_revalidate_negative(struct inode * dir,struct dentry * dentry,unsigned int flags)1634 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1635 			       unsigned int flags)
1636 {
1637 	int ret = 1;
1638 	if (nfs_neg_need_reval(dir, dentry, flags)) {
1639 		if (flags & LOOKUP_RCU)
1640 			return -ECHILD;
1641 		ret = 0;
1642 	}
1643 	return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1644 }
1645 
1646 static int
nfs_lookup_revalidate_delegated(struct inode * dir,struct dentry * dentry,struct inode * inode)1647 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1648 				struct inode *inode)
1649 {
1650 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1651 	return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1652 }
1653 
nfs_lookup_revalidate_dentry(struct inode * dir,struct dentry * dentry,struct inode * inode,unsigned int flags)1654 static int nfs_lookup_revalidate_dentry(struct inode *dir,
1655 					struct dentry *dentry,
1656 					struct inode *inode, unsigned int flags)
1657 {
1658 	struct nfs_fh *fhandle;
1659 	struct nfs_fattr *fattr;
1660 	unsigned long dir_verifier;
1661 	int ret;
1662 
1663 	trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1664 
1665 	ret = -ENOMEM;
1666 	fhandle = nfs_alloc_fhandle();
1667 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1668 	if (fhandle == NULL || fattr == NULL)
1669 		goto out;
1670 
1671 	dir_verifier = nfs_save_change_attribute(dir);
1672 	ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1673 	if (ret < 0) {
1674 		switch (ret) {
1675 		case -ESTALE:
1676 		case -ENOENT:
1677 			ret = 0;
1678 			break;
1679 		case -ETIMEDOUT:
1680 			if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1681 				ret = 1;
1682 		}
1683 		goto out;
1684 	}
1685 
1686 	/* Request help from readdirplus */
1687 	nfs_lookup_advise_force_readdirplus(dir, flags);
1688 
1689 	ret = 0;
1690 	if (nfs_compare_fh(NFS_FH(inode), fhandle))
1691 		goto out;
1692 	if (nfs_refresh_inode(inode, fattr) < 0)
1693 		goto out;
1694 
1695 	nfs_setsecurity(inode, fattr);
1696 	nfs_set_verifier(dentry, dir_verifier);
1697 
1698 	ret = 1;
1699 out:
1700 	nfs_free_fattr(fattr);
1701 	nfs_free_fhandle(fhandle);
1702 
1703 	/*
1704 	 * If the lookup failed despite the dentry change attribute being
1705 	 * a match, then we should revalidate the directory cache.
1706 	 */
1707 	if (!ret && nfs_dentry_verify_change(dir, dentry))
1708 		nfs_mark_dir_for_revalidate(dir);
1709 	return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1710 }
1711 
1712 /*
1713  * This is called every time the dcache has a lookup hit,
1714  * and we should check whether we can really trust that
1715  * lookup.
1716  *
1717  * NOTE! The hit can be a negative hit too, don't assume
1718  * we have an inode!
1719  *
1720  * If the parent directory is seen to have changed, we throw out the
1721  * cached dentry and do a new lookup.
1722  */
1723 static int
nfs_do_lookup_revalidate(struct inode * dir,struct dentry * dentry,unsigned int flags)1724 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1725 			 unsigned int flags)
1726 {
1727 	struct inode *inode;
1728 	int error;
1729 
1730 	nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1731 	inode = d_inode(dentry);
1732 
1733 	if (!inode)
1734 		return nfs_lookup_revalidate_negative(dir, dentry, flags);
1735 
1736 	if (is_bad_inode(inode)) {
1737 		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1738 				__func__, dentry);
1739 		goto out_bad;
1740 	}
1741 
1742 	if ((flags & LOOKUP_RENAME_TARGET) && d_count(dentry) < 2 &&
1743 	    nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1744 		goto out_bad;
1745 
1746 	if (nfs_verifier_is_delegated(dentry))
1747 		return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1748 
1749 	/* Force a full look up iff the parent directory has changed */
1750 	if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1751 	    nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1752 		error = nfs_lookup_verify_inode(inode, flags);
1753 		if (error) {
1754 			if (error == -ESTALE)
1755 				nfs_mark_dir_for_revalidate(dir);
1756 			goto out_bad;
1757 		}
1758 		goto out_valid;
1759 	}
1760 
1761 	if (flags & LOOKUP_RCU)
1762 		return -ECHILD;
1763 
1764 	if (NFS_STALE(inode))
1765 		goto out_bad;
1766 
1767 	return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
1768 out_valid:
1769 	return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1770 out_bad:
1771 	if (flags & LOOKUP_RCU)
1772 		return -ECHILD;
1773 	return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1774 }
1775 
1776 static int
__nfs_lookup_revalidate(struct dentry * dentry,unsigned int flags,int (* reval)(struct inode *,struct dentry *,unsigned int))1777 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1778 			int (*reval)(struct inode *, struct dentry *, unsigned int))
1779 {
1780 	struct dentry *parent;
1781 	struct inode *dir;
1782 	int ret;
1783 
1784 	if (flags & LOOKUP_RCU) {
1785 		if (dentry->d_fsdata == NFS_FSDATA_BLOCKED)
1786 			return -ECHILD;
1787 		parent = READ_ONCE(dentry->d_parent);
1788 		dir = d_inode_rcu(parent);
1789 		if (!dir)
1790 			return -ECHILD;
1791 		ret = reval(dir, dentry, flags);
1792 		if (parent != READ_ONCE(dentry->d_parent))
1793 			return -ECHILD;
1794 	} else {
1795 		/* Wait for unlink to complete */
1796 		wait_var_event(&dentry->d_fsdata,
1797 			       dentry->d_fsdata != NFS_FSDATA_BLOCKED);
1798 		parent = dget_parent(dentry);
1799 		ret = reval(d_inode(parent), dentry, flags);
1800 		dput(parent);
1801 	}
1802 	return ret;
1803 }
1804 
nfs_lookup_revalidate(struct dentry * dentry,unsigned int flags)1805 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1806 {
1807 	return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1808 }
1809 
1810 /*
1811  * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1812  * when we don't really care about the dentry name. This is called when a
1813  * pathwalk ends on a dentry that was not found via a normal lookup in the
1814  * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1815  *
1816  * In this situation, we just want to verify that the inode itself is OK
1817  * since the dentry might have changed on the server.
1818  */
nfs_weak_revalidate(struct dentry * dentry,unsigned int flags)1819 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1820 {
1821 	struct inode *inode = d_inode(dentry);
1822 	int error = 0;
1823 
1824 	/*
1825 	 * I believe we can only get a negative dentry here in the case of a
1826 	 * procfs-style symlink. Just assume it's correct for now, but we may
1827 	 * eventually need to do something more here.
1828 	 */
1829 	if (!inode) {
1830 		dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1831 				__func__, dentry);
1832 		return 1;
1833 	}
1834 
1835 	if (is_bad_inode(inode)) {
1836 		dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1837 				__func__, dentry);
1838 		return 0;
1839 	}
1840 
1841 	error = nfs_lookup_verify_inode(inode, flags);
1842 	dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1843 			__func__, inode->i_ino, error ? "invalid" : "valid");
1844 	return !error;
1845 }
1846 
1847 /*
1848  * This is called from dput() when d_count is going to 0.
1849  */
nfs_dentry_delete(const struct dentry * dentry)1850 static int nfs_dentry_delete(const struct dentry *dentry)
1851 {
1852 	dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1853 		dentry, dentry->d_flags);
1854 
1855 	/* Unhash any dentry with a stale inode */
1856 	if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1857 		return 1;
1858 
1859 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1860 		/* Unhash it, so that ->d_iput() would be called */
1861 		return 1;
1862 	}
1863 	if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1864 		/* Unhash it, so that ancestors of killed async unlink
1865 		 * files will be cleaned up during umount */
1866 		return 1;
1867 	}
1868 	return 0;
1869 
1870 }
1871 
1872 /* Ensure that we revalidate inode->i_nlink */
nfs_drop_nlink(struct inode * inode)1873 static void nfs_drop_nlink(struct inode *inode)
1874 {
1875 	spin_lock(&inode->i_lock);
1876 	/* drop the inode if we're reasonably sure this is the last link */
1877 	if (inode->i_nlink > 0)
1878 		drop_nlink(inode);
1879 	NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1880 	nfs_set_cache_invalid(
1881 		inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1882 			       NFS_INO_INVALID_NLINK);
1883 	spin_unlock(&inode->i_lock);
1884 }
1885 
1886 /*
1887  * Called when the dentry loses inode.
1888  * We use it to clean up silly-renamed files.
1889  */
nfs_dentry_iput(struct dentry * dentry,struct inode * inode)1890 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1891 {
1892 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1893 		nfs_complete_unlink(dentry, inode);
1894 		nfs_drop_nlink(inode);
1895 	}
1896 	iput(inode);
1897 }
1898 
nfs_d_release(struct dentry * dentry)1899 static void nfs_d_release(struct dentry *dentry)
1900 {
1901 	/* free cached devname value, if it survived that far */
1902 	if (unlikely(dentry->d_fsdata)) {
1903 		if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1904 			WARN_ON(1);
1905 		else
1906 			kfree(dentry->d_fsdata);
1907 	}
1908 }
1909 
1910 const struct dentry_operations nfs_dentry_operations = {
1911 	.d_revalidate	= nfs_lookup_revalidate,
1912 	.d_weak_revalidate	= nfs_weak_revalidate,
1913 	.d_delete	= nfs_dentry_delete,
1914 	.d_iput		= nfs_dentry_iput,
1915 	.d_automount	= nfs_d_automount,
1916 	.d_release	= nfs_d_release,
1917 };
1918 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1919 
nfs_lookup(struct inode * dir,struct dentry * dentry,unsigned int flags)1920 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1921 {
1922 	struct dentry *res;
1923 	struct inode *inode = NULL;
1924 	struct nfs_fh *fhandle = NULL;
1925 	struct nfs_fattr *fattr = NULL;
1926 	unsigned long dir_verifier;
1927 	int error;
1928 
1929 	dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1930 	nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1931 
1932 	if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1933 		return ERR_PTR(-ENAMETOOLONG);
1934 
1935 	/*
1936 	 * If we're doing an exclusive create, optimize away the lookup
1937 	 * but don't hash the dentry.
1938 	 */
1939 	if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1940 		return NULL;
1941 
1942 	res = ERR_PTR(-ENOMEM);
1943 	fhandle = nfs_alloc_fhandle();
1944 	fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir));
1945 	if (fhandle == NULL || fattr == NULL)
1946 		goto out;
1947 
1948 	dir_verifier = nfs_save_change_attribute(dir);
1949 	trace_nfs_lookup_enter(dir, dentry, flags);
1950 	error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1951 	if (error == -ENOENT) {
1952 		if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1953 			dir_verifier = inode_peek_iversion_raw(dir);
1954 		goto no_entry;
1955 	}
1956 	if (error < 0) {
1957 		res = ERR_PTR(error);
1958 		goto out;
1959 	}
1960 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1961 	res = ERR_CAST(inode);
1962 	if (IS_ERR(res))
1963 		goto out;
1964 
1965 	/* Notify readdir to use READDIRPLUS */
1966 	nfs_lookup_advise_force_readdirplus(dir, flags);
1967 
1968 no_entry:
1969 	res = d_splice_alias(inode, dentry);
1970 	if (res != NULL) {
1971 		if (IS_ERR(res))
1972 			goto out;
1973 		dentry = res;
1974 	}
1975 	nfs_set_verifier(dentry, dir_verifier);
1976 out:
1977 	trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res));
1978 	nfs_free_fattr(fattr);
1979 	nfs_free_fhandle(fhandle);
1980 	return res;
1981 }
1982 EXPORT_SYMBOL_GPL(nfs_lookup);
1983 
nfs_d_prune_case_insensitive_aliases(struct inode * inode)1984 void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
1985 {
1986 	/* Case insensitive server? Revalidate dentries */
1987 	if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
1988 		d_prune_aliases(inode);
1989 }
1990 EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
1991 
1992 #if IS_ENABLED(CONFIG_NFS_V4)
1993 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1994 
1995 const struct dentry_operations nfs4_dentry_operations = {
1996 	.d_revalidate	= nfs4_lookup_revalidate,
1997 	.d_weak_revalidate	= nfs_weak_revalidate,
1998 	.d_delete	= nfs_dentry_delete,
1999 	.d_iput		= nfs_dentry_iput,
2000 	.d_automount	= nfs_d_automount,
2001 	.d_release	= nfs_d_release,
2002 };
2003 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
2004 
create_nfs_open_context(struct dentry * dentry,int open_flags,struct file * filp)2005 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
2006 {
2007 	return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
2008 }
2009 
do_open(struct inode * inode,struct file * filp)2010 static int do_open(struct inode *inode, struct file *filp)
2011 {
2012 	nfs_fscache_open_file(inode, filp);
2013 	return 0;
2014 }
2015 
nfs_finish_open(struct nfs_open_context * ctx,struct dentry * dentry,struct file * file,unsigned open_flags)2016 static int nfs_finish_open(struct nfs_open_context *ctx,
2017 			   struct dentry *dentry,
2018 			   struct file *file, unsigned open_flags)
2019 {
2020 	int err;
2021 
2022 	err = finish_open(file, dentry, do_open);
2023 	if (err)
2024 		goto out;
2025 	if (S_ISREG(file_inode(file)->i_mode))
2026 		nfs_file_set_open_context(file, ctx);
2027 	else
2028 		err = -EOPENSTALE;
2029 out:
2030 	return err;
2031 }
2032 
nfs_atomic_open(struct inode * dir,struct dentry * dentry,struct file * file,unsigned open_flags,umode_t mode)2033 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2034 		    struct file *file, unsigned open_flags,
2035 		    umode_t mode)
2036 {
2037 	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2038 	struct nfs_open_context *ctx;
2039 	struct dentry *res;
2040 	struct iattr attr = { .ia_valid = ATTR_OPEN };
2041 	struct inode *inode;
2042 	unsigned int lookup_flags = 0;
2043 	unsigned long dir_verifier;
2044 	bool switched = false;
2045 	int created = 0;
2046 	int err;
2047 
2048 	/* Expect a negative dentry */
2049 	BUG_ON(d_inode(dentry));
2050 
2051 	dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2052 			dir->i_sb->s_id, dir->i_ino, dentry);
2053 
2054 	err = nfs_check_flags(open_flags);
2055 	if (err)
2056 		return err;
2057 
2058 	/* NFS only supports OPEN on regular files */
2059 	if ((open_flags & O_DIRECTORY)) {
2060 		if (!d_in_lookup(dentry)) {
2061 			/*
2062 			 * Hashed negative dentry with O_DIRECTORY: dentry was
2063 			 * revalidated and is fine, no need to perform lookup
2064 			 * again
2065 			 */
2066 			return -ENOENT;
2067 		}
2068 		lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2069 		goto no_open;
2070 	}
2071 
2072 	if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
2073 		return -ENAMETOOLONG;
2074 
2075 	if (open_flags & O_CREAT) {
2076 		struct nfs_server *server = NFS_SERVER(dir);
2077 
2078 		if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2079 			mode &= ~current_umask();
2080 
2081 		attr.ia_valid |= ATTR_MODE;
2082 		attr.ia_mode = mode;
2083 	}
2084 	if (open_flags & O_TRUNC) {
2085 		attr.ia_valid |= ATTR_SIZE;
2086 		attr.ia_size = 0;
2087 	}
2088 
2089 	if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2090 		d_drop(dentry);
2091 		switched = true;
2092 		dentry = d_alloc_parallel(dentry->d_parent,
2093 					  &dentry->d_name, &wq);
2094 		if (IS_ERR(dentry))
2095 			return PTR_ERR(dentry);
2096 		if (unlikely(!d_in_lookup(dentry)))
2097 			return finish_no_open(file, dentry);
2098 	}
2099 
2100 	ctx = create_nfs_open_context(dentry, open_flags, file);
2101 	err = PTR_ERR(ctx);
2102 	if (IS_ERR(ctx))
2103 		goto out;
2104 
2105 	trace_nfs_atomic_open_enter(dir, ctx, open_flags);
2106 	inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
2107 	if (created)
2108 		file->f_mode |= FMODE_CREATED;
2109 	if (IS_ERR(inode)) {
2110 		err = PTR_ERR(inode);
2111 		trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2112 		put_nfs_open_context(ctx);
2113 		d_drop(dentry);
2114 		switch (err) {
2115 		case -ENOENT:
2116 			d_splice_alias(NULL, dentry);
2117 			if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
2118 				dir_verifier = inode_peek_iversion_raw(dir);
2119 			else
2120 				dir_verifier = nfs_save_change_attribute(dir);
2121 			nfs_set_verifier(dentry, dir_verifier);
2122 			break;
2123 		case -EISDIR:
2124 		case -ENOTDIR:
2125 			goto no_open;
2126 		case -ELOOP:
2127 			if (!(open_flags & O_NOFOLLOW))
2128 				goto no_open;
2129 			break;
2130 			/* case -EINVAL: */
2131 		default:
2132 			break;
2133 		}
2134 		goto out;
2135 	}
2136 	file->f_mode |= FMODE_CAN_ODIRECT;
2137 
2138 	err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
2139 	trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2140 	put_nfs_open_context(ctx);
2141 out:
2142 	if (unlikely(switched)) {
2143 		d_lookup_done(dentry);
2144 		dput(dentry);
2145 	}
2146 	return err;
2147 
2148 no_open:
2149 	res = nfs_lookup(dir, dentry, lookup_flags);
2150 	if (!res) {
2151 		inode = d_inode(dentry);
2152 		if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2153 		    !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2154 			res = ERR_PTR(-ENOTDIR);
2155 		else if (inode && S_ISREG(inode->i_mode))
2156 			res = ERR_PTR(-EOPENSTALE);
2157 	} else if (!IS_ERR(res)) {
2158 		inode = d_inode(res);
2159 		if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2160 		    !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2161 			dput(res);
2162 			res = ERR_PTR(-ENOTDIR);
2163 		} else if (inode && S_ISREG(inode->i_mode)) {
2164 			dput(res);
2165 			res = ERR_PTR(-EOPENSTALE);
2166 		}
2167 	}
2168 	if (switched) {
2169 		d_lookup_done(dentry);
2170 		if (!res)
2171 			res = dentry;
2172 		else
2173 			dput(dentry);
2174 	}
2175 	if (IS_ERR(res))
2176 		return PTR_ERR(res);
2177 	return finish_no_open(file, res);
2178 }
2179 EXPORT_SYMBOL_GPL(nfs_atomic_open);
2180 
2181 static int
nfs4_do_lookup_revalidate(struct inode * dir,struct dentry * dentry,unsigned int flags)2182 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2183 			  unsigned int flags)
2184 {
2185 	struct inode *inode;
2186 
2187 	if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2188 		goto full_reval;
2189 	if (d_mountpoint(dentry))
2190 		goto full_reval;
2191 
2192 	inode = d_inode(dentry);
2193 
2194 	/* We can't create new files in nfs_open_revalidate(), so we
2195 	 * optimize away revalidation of negative dentries.
2196 	 */
2197 	if (inode == NULL)
2198 		goto full_reval;
2199 
2200 	if (nfs_verifier_is_delegated(dentry))
2201 		return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2202 
2203 	/* NFS only supports OPEN on regular files */
2204 	if (!S_ISREG(inode->i_mode))
2205 		goto full_reval;
2206 
2207 	/* We cannot do exclusive creation on a positive dentry */
2208 	if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2209 		goto reval_dentry;
2210 
2211 	/* Check if the directory changed */
2212 	if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2213 		goto reval_dentry;
2214 
2215 	/* Let f_op->open() actually open (and revalidate) the file */
2216 	return 1;
2217 reval_dentry:
2218 	if (flags & LOOKUP_RCU)
2219 		return -ECHILD;
2220 	return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
2221 
2222 full_reval:
2223 	return nfs_do_lookup_revalidate(dir, dentry, flags);
2224 }
2225 
nfs4_lookup_revalidate(struct dentry * dentry,unsigned int flags)2226 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2227 {
2228 	return __nfs_lookup_revalidate(dentry, flags,
2229 			nfs4_do_lookup_revalidate);
2230 }
2231 
2232 #endif /* CONFIG_NFSV4 */
2233 
2234 struct dentry *
nfs_add_or_obtain(struct dentry * dentry,struct nfs_fh * fhandle,struct nfs_fattr * fattr)2235 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2236 				struct nfs_fattr *fattr)
2237 {
2238 	struct dentry *parent = dget_parent(dentry);
2239 	struct inode *dir = d_inode(parent);
2240 	struct inode *inode;
2241 	struct dentry *d;
2242 	int error;
2243 
2244 	d_drop(dentry);
2245 
2246 	if (fhandle->size == 0) {
2247 		error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
2248 		if (error)
2249 			goto out_error;
2250 	}
2251 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2252 	if (!(fattr->valid & NFS_ATTR_FATTR)) {
2253 		struct nfs_server *server = NFS_SB(dentry->d_sb);
2254 		error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2255 				fattr, NULL);
2256 		if (error < 0)
2257 			goto out_error;
2258 	}
2259 	inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2260 	d = d_splice_alias(inode, dentry);
2261 out:
2262 	dput(parent);
2263 	return d;
2264 out_error:
2265 	d = ERR_PTR(error);
2266 	goto out;
2267 }
2268 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2269 
2270 /*
2271  * Code common to create, mkdir, and mknod.
2272  */
nfs_instantiate(struct dentry * dentry,struct nfs_fh * fhandle,struct nfs_fattr * fattr)2273 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2274 				struct nfs_fattr *fattr)
2275 {
2276 	struct dentry *d;
2277 
2278 	d = nfs_add_or_obtain(dentry, fhandle, fattr);
2279 	if (IS_ERR(d))
2280 		return PTR_ERR(d);
2281 
2282 	/* Callers don't care */
2283 	dput(d);
2284 	return 0;
2285 }
2286 EXPORT_SYMBOL_GPL(nfs_instantiate);
2287 
2288 /*
2289  * Following a failed create operation, we drop the dentry rather
2290  * than retain a negative dentry. This avoids a problem in the event
2291  * that the operation succeeded on the server, but an error in the
2292  * reply path made it appear to have failed.
2293  */
nfs_create(struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry,umode_t mode,bool excl)2294 int nfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2295 	       struct dentry *dentry, umode_t mode, bool excl)
2296 {
2297 	struct iattr attr;
2298 	int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2299 	int error;
2300 
2301 	dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2302 			dir->i_sb->s_id, dir->i_ino, dentry);
2303 
2304 	attr.ia_mode = mode;
2305 	attr.ia_valid = ATTR_MODE;
2306 
2307 	trace_nfs_create_enter(dir, dentry, open_flags);
2308 	error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2309 	trace_nfs_create_exit(dir, dentry, open_flags, error);
2310 	if (error != 0)
2311 		goto out_err;
2312 	return 0;
2313 out_err:
2314 	d_drop(dentry);
2315 	return error;
2316 }
2317 EXPORT_SYMBOL_GPL(nfs_create);
2318 
2319 /*
2320  * See comments for nfs_proc_create regarding failed operations.
2321  */
2322 int
nfs_mknod(struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry,umode_t mode,dev_t rdev)2323 nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2324 	  struct dentry *dentry, umode_t mode, dev_t rdev)
2325 {
2326 	struct iattr attr;
2327 	int status;
2328 
2329 	dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2330 			dir->i_sb->s_id, dir->i_ino, dentry);
2331 
2332 	attr.ia_mode = mode;
2333 	attr.ia_valid = ATTR_MODE;
2334 
2335 	trace_nfs_mknod_enter(dir, dentry);
2336 	status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2337 	trace_nfs_mknod_exit(dir, dentry, status);
2338 	if (status != 0)
2339 		goto out_err;
2340 	return 0;
2341 out_err:
2342 	d_drop(dentry);
2343 	return status;
2344 }
2345 EXPORT_SYMBOL_GPL(nfs_mknod);
2346 
2347 /*
2348  * See comments for nfs_proc_create regarding failed operations.
2349  */
nfs_mkdir(struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry,umode_t mode)2350 int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2351 	      struct dentry *dentry, umode_t mode)
2352 {
2353 	struct iattr attr;
2354 	int error;
2355 
2356 	dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2357 			dir->i_sb->s_id, dir->i_ino, dentry);
2358 
2359 	attr.ia_valid = ATTR_MODE;
2360 	attr.ia_mode = mode | S_IFDIR;
2361 
2362 	trace_nfs_mkdir_enter(dir, dentry);
2363 	error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2364 	trace_nfs_mkdir_exit(dir, dentry, error);
2365 	if (error != 0)
2366 		goto out_err;
2367 	return 0;
2368 out_err:
2369 	d_drop(dentry);
2370 	return error;
2371 }
2372 EXPORT_SYMBOL_GPL(nfs_mkdir);
2373 
nfs_dentry_handle_enoent(struct dentry * dentry)2374 static void nfs_dentry_handle_enoent(struct dentry *dentry)
2375 {
2376 	if (simple_positive(dentry))
2377 		d_delete(dentry);
2378 }
2379 
nfs_dentry_remove_handle_error(struct inode * dir,struct dentry * dentry,int error)2380 static void nfs_dentry_remove_handle_error(struct inode *dir,
2381 					   struct dentry *dentry, int error)
2382 {
2383 	switch (error) {
2384 	case -ENOENT:
2385 		if (d_really_is_positive(dentry))
2386 			d_delete(dentry);
2387 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2388 		break;
2389 	case 0:
2390 		nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2391 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2392 	}
2393 }
2394 
nfs_rmdir(struct inode * dir,struct dentry * dentry)2395 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2396 {
2397 	int error;
2398 
2399 	dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2400 			dir->i_sb->s_id, dir->i_ino, dentry);
2401 
2402 	trace_nfs_rmdir_enter(dir, dentry);
2403 	if (d_really_is_positive(dentry)) {
2404 		down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2405 		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2406 		/* Ensure the VFS deletes this inode */
2407 		switch (error) {
2408 		case 0:
2409 			clear_nlink(d_inode(dentry));
2410 			break;
2411 		case -ENOENT:
2412 			nfs_dentry_handle_enoent(dentry);
2413 		}
2414 		up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2415 	} else
2416 		error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2417 	nfs_dentry_remove_handle_error(dir, dentry, error);
2418 	trace_nfs_rmdir_exit(dir, dentry, error);
2419 
2420 	return error;
2421 }
2422 EXPORT_SYMBOL_GPL(nfs_rmdir);
2423 
2424 /*
2425  * Remove a file after making sure there are no pending writes,
2426  * and after checking that the file has only one user.
2427  *
2428  * We invalidate the attribute cache and free the inode prior to the operation
2429  * to avoid possible races if the server reuses the inode.
2430  */
nfs_safe_remove(struct dentry * dentry)2431 static int nfs_safe_remove(struct dentry *dentry)
2432 {
2433 	struct inode *dir = d_inode(dentry->d_parent);
2434 	struct inode *inode = d_inode(dentry);
2435 	int error = -EBUSY;
2436 
2437 	dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2438 
2439 	/* If the dentry was sillyrenamed, we simply call d_delete() */
2440 	if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2441 		error = 0;
2442 		goto out;
2443 	}
2444 
2445 	trace_nfs_remove_enter(dir, dentry);
2446 	if (inode != NULL) {
2447 		error = NFS_PROTO(dir)->remove(dir, dentry);
2448 		if (error == 0)
2449 			nfs_drop_nlink(inode);
2450 	} else
2451 		error = NFS_PROTO(dir)->remove(dir, dentry);
2452 	if (error == -ENOENT)
2453 		nfs_dentry_handle_enoent(dentry);
2454 	trace_nfs_remove_exit(dir, dentry, error);
2455 out:
2456 	return error;
2457 }
2458 
2459 /*  We do silly rename. In case sillyrename() returns -EBUSY, the inode
2460  *  belongs to an active ".nfs..." file and we return -EBUSY.
2461  *
2462  *  If sillyrename() returns 0, we do nothing, otherwise we unlink.
2463  */
nfs_unlink(struct inode * dir,struct dentry * dentry)2464 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2465 {
2466 	int error;
2467 
2468 	dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2469 		dir->i_ino, dentry);
2470 
2471 	trace_nfs_unlink_enter(dir, dentry);
2472 	spin_lock(&dentry->d_lock);
2473 	if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2474 					     &NFS_I(d_inode(dentry))->flags)) {
2475 		spin_unlock(&dentry->d_lock);
2476 		/* Start asynchronous writeout of the inode */
2477 		write_inode_now(d_inode(dentry), 0);
2478 		error = nfs_sillyrename(dir, dentry);
2479 		goto out;
2480 	}
2481 	/* We must prevent any concurrent open until the unlink
2482 	 * completes.  ->d_revalidate will wait for ->d_fsdata
2483 	 * to clear.  We set it here to ensure no lookup succeeds until
2484 	 * the unlink is complete on the server.
2485 	 */
2486 	error = -ETXTBSY;
2487 	if (WARN_ON(dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2488 	    WARN_ON(dentry->d_fsdata == NFS_FSDATA_BLOCKED)) {
2489 		spin_unlock(&dentry->d_lock);
2490 		goto out;
2491 	}
2492 	/* old devname */
2493 	kfree(dentry->d_fsdata);
2494 	dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2495 
2496 	spin_unlock(&dentry->d_lock);
2497 	error = nfs_safe_remove(dentry);
2498 	nfs_dentry_remove_handle_error(dir, dentry, error);
2499 	dentry->d_fsdata = NULL;
2500 	wake_up_var(&dentry->d_fsdata);
2501 out:
2502 	trace_nfs_unlink_exit(dir, dentry, error);
2503 	return error;
2504 }
2505 EXPORT_SYMBOL_GPL(nfs_unlink);
2506 
2507 /*
2508  * To create a symbolic link, most file systems instantiate a new inode,
2509  * add a page to it containing the path, then write it out to the disk
2510  * using prepare_write/commit_write.
2511  *
2512  * Unfortunately the NFS client can't create the in-core inode first
2513  * because it needs a file handle to create an in-core inode (see
2514  * fs/nfs/inode.c:nfs_fhget).  We only have a file handle *after* the
2515  * symlink request has completed on the server.
2516  *
2517  * So instead we allocate a raw page, copy the symname into it, then do
2518  * the SYMLINK request with the page as the buffer.  If it succeeds, we
2519  * now have a new file handle and can instantiate an in-core NFS inode
2520  * and move the raw page into its mapping.
2521  */
nfs_symlink(struct user_namespace * mnt_userns,struct inode * dir,struct dentry * dentry,const char * symname)2522 int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
2523 		struct dentry *dentry, const char *symname)
2524 {
2525 	struct page *page;
2526 	char *kaddr;
2527 	struct iattr attr;
2528 	unsigned int pathlen = strlen(symname);
2529 	int error;
2530 
2531 	dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2532 		dir->i_ino, dentry, symname);
2533 
2534 	if (pathlen > PAGE_SIZE)
2535 		return -ENAMETOOLONG;
2536 
2537 	attr.ia_mode = S_IFLNK | S_IRWXUGO;
2538 	attr.ia_valid = ATTR_MODE;
2539 
2540 	page = alloc_page(GFP_USER);
2541 	if (!page)
2542 		return -ENOMEM;
2543 
2544 	kaddr = page_address(page);
2545 	memcpy(kaddr, symname, pathlen);
2546 	if (pathlen < PAGE_SIZE)
2547 		memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2548 
2549 	trace_nfs_symlink_enter(dir, dentry);
2550 	error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2551 	trace_nfs_symlink_exit(dir, dentry, error);
2552 	if (error != 0) {
2553 		dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2554 			dir->i_sb->s_id, dir->i_ino,
2555 			dentry, symname, error);
2556 		d_drop(dentry);
2557 		__free_page(page);
2558 		return error;
2559 	}
2560 
2561 	nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2562 
2563 	/*
2564 	 * No big deal if we can't add this page to the page cache here.
2565 	 * READLINK will get the missing page from the server if needed.
2566 	 */
2567 	if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2568 							GFP_KERNEL)) {
2569 		SetPageUptodate(page);
2570 		unlock_page(page);
2571 		/*
2572 		 * add_to_page_cache_lru() grabs an extra page refcount.
2573 		 * Drop it here to avoid leaking this page later.
2574 		 */
2575 		put_page(page);
2576 	} else
2577 		__free_page(page);
2578 
2579 	return 0;
2580 }
2581 EXPORT_SYMBOL_GPL(nfs_symlink);
2582 
2583 int
nfs_link(struct dentry * old_dentry,struct inode * dir,struct dentry * dentry)2584 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2585 {
2586 	struct inode *inode = d_inode(old_dentry);
2587 	int error;
2588 
2589 	dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2590 		old_dentry, dentry);
2591 
2592 	trace_nfs_link_enter(inode, dir, dentry);
2593 	d_drop(dentry);
2594 	if (S_ISREG(inode->i_mode))
2595 		nfs_sync_inode(inode);
2596 	error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2597 	if (error == 0) {
2598 		nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2599 		ihold(inode);
2600 		d_add(dentry, inode);
2601 	}
2602 	trace_nfs_link_exit(inode, dir, dentry, error);
2603 	return error;
2604 }
2605 EXPORT_SYMBOL_GPL(nfs_link);
2606 
2607 static void
nfs_unblock_rename(struct rpc_task * task,struct nfs_renamedata * data)2608 nfs_unblock_rename(struct rpc_task *task, struct nfs_renamedata *data)
2609 {
2610 	struct dentry *new_dentry = data->new_dentry;
2611 
2612 	new_dentry->d_fsdata = NULL;
2613 	wake_up_var(&new_dentry->d_fsdata);
2614 }
2615 
2616 /*
2617  * RENAME
2618  * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2619  * different file handle for the same inode after a rename (e.g. when
2620  * moving to a different directory). A fail-safe method to do so would
2621  * be to look up old_dir/old_name, create a link to new_dir/new_name and
2622  * rename the old file using the sillyrename stuff. This way, the original
2623  * file in old_dir will go away when the last process iput()s the inode.
2624  *
2625  * FIXED.
2626  *
2627  * It actually works quite well. One needs to have the possibility for
2628  * at least one ".nfs..." file in each directory the file ever gets
2629  * moved or linked to which happens automagically with the new
2630  * implementation that only depends on the dcache stuff instead of
2631  * using the inode layer
2632  *
2633  * Unfortunately, things are a little more complicated than indicated
2634  * above. For a cross-directory move, we want to make sure we can get
2635  * rid of the old inode after the operation.  This means there must be
2636  * no pending writes (if it's a file), and the use count must be 1.
2637  * If these conditions are met, we can drop the dentries before doing
2638  * the rename.
2639  */
nfs_rename(struct user_namespace * mnt_userns,struct inode * old_dir,struct dentry * old_dentry,struct inode * new_dir,struct dentry * new_dentry,unsigned int flags)2640 int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
2641 	       struct dentry *old_dentry, struct inode *new_dir,
2642 	       struct dentry *new_dentry, unsigned int flags)
2643 {
2644 	struct inode *old_inode = d_inode(old_dentry);
2645 	struct inode *new_inode = d_inode(new_dentry);
2646 	struct dentry *dentry = NULL;
2647 	struct rpc_task *task;
2648 	bool must_unblock = false;
2649 	int error = -EBUSY;
2650 
2651 	if (flags)
2652 		return -EINVAL;
2653 
2654 	dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2655 		 old_dentry, new_dentry,
2656 		 d_count(new_dentry));
2657 
2658 	trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2659 	/*
2660 	 * For non-directories, check whether the target is busy and if so,
2661 	 * make a copy of the dentry and then do a silly-rename. If the
2662 	 * silly-rename succeeds, the copied dentry is hashed and becomes
2663 	 * the new target.
2664 	 */
2665 	if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2666 		/* We must prevent any concurrent open until the unlink
2667 		 * completes.  ->d_revalidate will wait for ->d_fsdata
2668 		 * to clear.  We set it here to ensure no lookup succeeds until
2669 		 * the unlink is complete on the server.
2670 		 */
2671 		error = -ETXTBSY;
2672 		if (WARN_ON(new_dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2673 		    WARN_ON(new_dentry->d_fsdata == NFS_FSDATA_BLOCKED))
2674 			goto out;
2675 		if (new_dentry->d_fsdata) {
2676 			/* old devname */
2677 			kfree(new_dentry->d_fsdata);
2678 			new_dentry->d_fsdata = NULL;
2679 		}
2680 
2681 		spin_lock(&new_dentry->d_lock);
2682 		if (d_count(new_dentry) > 2) {
2683 			int err;
2684 
2685 			spin_unlock(&new_dentry->d_lock);
2686 
2687 			/* copy the target dentry's name */
2688 			dentry = d_alloc(new_dentry->d_parent,
2689 					 &new_dentry->d_name);
2690 			if (!dentry)
2691 				goto out;
2692 
2693 			/* silly-rename the existing target ... */
2694 			err = nfs_sillyrename(new_dir, new_dentry);
2695 			if (err)
2696 				goto out;
2697 
2698 			new_dentry = dentry;
2699 			new_inode = NULL;
2700 		} else {
2701 			new_dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2702 			must_unblock = true;
2703 			spin_unlock(&new_dentry->d_lock);
2704 		}
2705 
2706 	}
2707 
2708 	if (S_ISREG(old_inode->i_mode))
2709 		nfs_sync_inode(old_inode);
2710 	task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry,
2711 				must_unblock ? nfs_unblock_rename : NULL);
2712 	if (IS_ERR(task)) {
2713 		error = PTR_ERR(task);
2714 		goto out;
2715 	}
2716 
2717 	error = rpc_wait_for_completion_task(task);
2718 	if (error != 0) {
2719 		((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2720 		/* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2721 		smp_wmb();
2722 	} else
2723 		error = task->tk_status;
2724 	rpc_put_task(task);
2725 	/* Ensure the inode attributes are revalidated */
2726 	if (error == 0) {
2727 		spin_lock(&old_inode->i_lock);
2728 		NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2729 		nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2730 							 NFS_INO_INVALID_CTIME |
2731 							 NFS_INO_REVAL_FORCED);
2732 		spin_unlock(&old_inode->i_lock);
2733 	}
2734 out:
2735 	trace_nfs_rename_exit(old_dir, old_dentry,
2736 			new_dir, new_dentry, error);
2737 	if (!error) {
2738 		if (new_inode != NULL)
2739 			nfs_drop_nlink(new_inode);
2740 		/*
2741 		 * The d_move() should be here instead of in an async RPC completion
2742 		 * handler because we need the proper locks to move the dentry.  If
2743 		 * we're interrupted by a signal, the async RPC completion handler
2744 		 * should mark the directories for revalidation.
2745 		 */
2746 		d_move(old_dentry, new_dentry);
2747 		nfs_set_verifier(old_dentry,
2748 					nfs_save_change_attribute(new_dir));
2749 	} else if (error == -ENOENT)
2750 		nfs_dentry_handle_enoent(old_dentry);
2751 
2752 	/* new dentry created? */
2753 	if (dentry)
2754 		dput(dentry);
2755 	return error;
2756 }
2757 EXPORT_SYMBOL_GPL(nfs_rename);
2758 
2759 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2760 static LIST_HEAD(nfs_access_lru_list);
2761 static atomic_long_t nfs_access_nr_entries;
2762 
2763 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2764 module_param(nfs_access_max_cachesize, ulong, 0644);
2765 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2766 
nfs_access_free_entry(struct nfs_access_entry * entry)2767 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2768 {
2769 	put_group_info(entry->group_info);
2770 	kfree_rcu(entry, rcu_head);
2771 	smp_mb__before_atomic();
2772 	atomic_long_dec(&nfs_access_nr_entries);
2773 	smp_mb__after_atomic();
2774 }
2775 
nfs_access_free_list(struct list_head * head)2776 static void nfs_access_free_list(struct list_head *head)
2777 {
2778 	struct nfs_access_entry *cache;
2779 
2780 	while (!list_empty(head)) {
2781 		cache = list_entry(head->next, struct nfs_access_entry, lru);
2782 		list_del(&cache->lru);
2783 		nfs_access_free_entry(cache);
2784 	}
2785 }
2786 
2787 static unsigned long
nfs_do_access_cache_scan(unsigned int nr_to_scan)2788 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2789 {
2790 	LIST_HEAD(head);
2791 	struct nfs_inode *nfsi, *next;
2792 	struct nfs_access_entry *cache;
2793 	long freed = 0;
2794 
2795 	spin_lock(&nfs_access_lru_lock);
2796 	list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2797 		struct inode *inode;
2798 
2799 		if (nr_to_scan-- == 0)
2800 			break;
2801 		inode = &nfsi->vfs_inode;
2802 		spin_lock(&inode->i_lock);
2803 		if (list_empty(&nfsi->access_cache_entry_lru))
2804 			goto remove_lru_entry;
2805 		cache = list_entry(nfsi->access_cache_entry_lru.next,
2806 				struct nfs_access_entry, lru);
2807 		list_move(&cache->lru, &head);
2808 		rb_erase(&cache->rb_node, &nfsi->access_cache);
2809 		freed++;
2810 		if (!list_empty(&nfsi->access_cache_entry_lru))
2811 			list_move_tail(&nfsi->access_cache_inode_lru,
2812 					&nfs_access_lru_list);
2813 		else {
2814 remove_lru_entry:
2815 			list_del_init(&nfsi->access_cache_inode_lru);
2816 			smp_mb__before_atomic();
2817 			clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2818 			smp_mb__after_atomic();
2819 		}
2820 		spin_unlock(&inode->i_lock);
2821 	}
2822 	spin_unlock(&nfs_access_lru_lock);
2823 	nfs_access_free_list(&head);
2824 	return freed;
2825 }
2826 
2827 unsigned long
nfs_access_cache_scan(struct shrinker * shrink,struct shrink_control * sc)2828 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2829 {
2830 	int nr_to_scan = sc->nr_to_scan;
2831 	gfp_t gfp_mask = sc->gfp_mask;
2832 
2833 	if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2834 		return SHRINK_STOP;
2835 	return nfs_do_access_cache_scan(nr_to_scan);
2836 }
2837 
2838 
2839 unsigned long
nfs_access_cache_count(struct shrinker * shrink,struct shrink_control * sc)2840 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2841 {
2842 	return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2843 }
2844 
2845 static void
nfs_access_cache_enforce_limit(void)2846 nfs_access_cache_enforce_limit(void)
2847 {
2848 	long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2849 	unsigned long diff;
2850 	unsigned int nr_to_scan;
2851 
2852 	if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2853 		return;
2854 	nr_to_scan = 100;
2855 	diff = nr_entries - nfs_access_max_cachesize;
2856 	if (diff < nr_to_scan)
2857 		nr_to_scan = diff;
2858 	nfs_do_access_cache_scan(nr_to_scan);
2859 }
2860 
__nfs_access_zap_cache(struct nfs_inode * nfsi,struct list_head * head)2861 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2862 {
2863 	struct rb_root *root_node = &nfsi->access_cache;
2864 	struct rb_node *n;
2865 	struct nfs_access_entry *entry;
2866 
2867 	/* Unhook entries from the cache */
2868 	while ((n = rb_first(root_node)) != NULL) {
2869 		entry = rb_entry(n, struct nfs_access_entry, rb_node);
2870 		rb_erase(n, root_node);
2871 		list_move(&entry->lru, head);
2872 	}
2873 	nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2874 }
2875 
nfs_access_zap_cache(struct inode * inode)2876 void nfs_access_zap_cache(struct inode *inode)
2877 {
2878 	LIST_HEAD(head);
2879 
2880 	if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2881 		return;
2882 	/* Remove from global LRU init */
2883 	spin_lock(&nfs_access_lru_lock);
2884 	if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2885 		list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2886 
2887 	spin_lock(&inode->i_lock);
2888 	__nfs_access_zap_cache(NFS_I(inode), &head);
2889 	spin_unlock(&inode->i_lock);
2890 	spin_unlock(&nfs_access_lru_lock);
2891 	nfs_access_free_list(&head);
2892 }
2893 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2894 
access_cmp(const struct cred * a,const struct nfs_access_entry * b)2895 static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2896 {
2897 	struct group_info *ga, *gb;
2898 	int g;
2899 
2900 	if (uid_lt(a->fsuid, b->fsuid))
2901 		return -1;
2902 	if (uid_gt(a->fsuid, b->fsuid))
2903 		return 1;
2904 
2905 	if (gid_lt(a->fsgid, b->fsgid))
2906 		return -1;
2907 	if (gid_gt(a->fsgid, b->fsgid))
2908 		return 1;
2909 
2910 	ga = a->group_info;
2911 	gb = b->group_info;
2912 	if (ga == gb)
2913 		return 0;
2914 	if (ga == NULL)
2915 		return -1;
2916 	if (gb == NULL)
2917 		return 1;
2918 	if (ga->ngroups < gb->ngroups)
2919 		return -1;
2920 	if (ga->ngroups > gb->ngroups)
2921 		return 1;
2922 
2923 	for (g = 0; g < ga->ngroups; g++) {
2924 		if (gid_lt(ga->gid[g], gb->gid[g]))
2925 			return -1;
2926 		if (gid_gt(ga->gid[g], gb->gid[g]))
2927 			return 1;
2928 	}
2929 	return 0;
2930 }
2931 
nfs_access_search_rbtree(struct inode * inode,const struct cred * cred)2932 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2933 {
2934 	struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2935 
2936 	while (n != NULL) {
2937 		struct nfs_access_entry *entry =
2938 			rb_entry(n, struct nfs_access_entry, rb_node);
2939 		int cmp = access_cmp(cred, entry);
2940 
2941 		if (cmp < 0)
2942 			n = n->rb_left;
2943 		else if (cmp > 0)
2944 			n = n->rb_right;
2945 		else
2946 			return entry;
2947 	}
2948 	return NULL;
2949 }
2950 
nfs_access_get_cached_locked(struct inode * inode,const struct cred * cred,u32 * mask,bool may_block)2951 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
2952 {
2953 	struct nfs_inode *nfsi = NFS_I(inode);
2954 	struct nfs_access_entry *cache;
2955 	bool retry = true;
2956 	int err;
2957 
2958 	spin_lock(&inode->i_lock);
2959 	for(;;) {
2960 		if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2961 			goto out_zap;
2962 		cache = nfs_access_search_rbtree(inode, cred);
2963 		err = -ENOENT;
2964 		if (cache == NULL)
2965 			goto out;
2966 		/* Found an entry, is our attribute cache valid? */
2967 		if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2968 			break;
2969 		if (!retry)
2970 			break;
2971 		err = -ECHILD;
2972 		if (!may_block)
2973 			goto out;
2974 		spin_unlock(&inode->i_lock);
2975 		err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2976 		if (err)
2977 			return err;
2978 		spin_lock(&inode->i_lock);
2979 		retry = false;
2980 	}
2981 	*mask = cache->mask;
2982 	list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2983 	err = 0;
2984 out:
2985 	spin_unlock(&inode->i_lock);
2986 	return err;
2987 out_zap:
2988 	spin_unlock(&inode->i_lock);
2989 	nfs_access_zap_cache(inode);
2990 	return -ENOENT;
2991 }
2992 
nfs_access_get_cached_rcu(struct inode * inode,const struct cred * cred,u32 * mask)2993 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
2994 {
2995 	/* Only check the most recently returned cache entry,
2996 	 * but do it without locking.
2997 	 */
2998 	struct nfs_inode *nfsi = NFS_I(inode);
2999 	struct nfs_access_entry *cache;
3000 	int err = -ECHILD;
3001 	struct list_head *lh;
3002 
3003 	rcu_read_lock();
3004 	if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3005 		goto out;
3006 	lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
3007 	cache = list_entry(lh, struct nfs_access_entry, lru);
3008 	if (lh == &nfsi->access_cache_entry_lru ||
3009 	    access_cmp(cred, cache) != 0)
3010 		cache = NULL;
3011 	if (cache == NULL)
3012 		goto out;
3013 	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3014 		goto out;
3015 	*mask = cache->mask;
3016 	err = 0;
3017 out:
3018 	rcu_read_unlock();
3019 	return err;
3020 }
3021 
nfs_access_get_cached(struct inode * inode,const struct cred * cred,u32 * mask,bool may_block)3022 int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
3023 			  u32 *mask, bool may_block)
3024 {
3025 	int status;
3026 
3027 	status = nfs_access_get_cached_rcu(inode, cred, mask);
3028 	if (status != 0)
3029 		status = nfs_access_get_cached_locked(inode, cred, mask,
3030 		    may_block);
3031 
3032 	return status;
3033 }
3034 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
3035 
nfs_access_add_rbtree(struct inode * inode,struct nfs_access_entry * set,const struct cred * cred)3036 static void nfs_access_add_rbtree(struct inode *inode,
3037 				  struct nfs_access_entry *set,
3038 				  const struct cred *cred)
3039 {
3040 	struct nfs_inode *nfsi = NFS_I(inode);
3041 	struct rb_root *root_node = &nfsi->access_cache;
3042 	struct rb_node **p = &root_node->rb_node;
3043 	struct rb_node *parent = NULL;
3044 	struct nfs_access_entry *entry;
3045 	int cmp;
3046 
3047 	spin_lock(&inode->i_lock);
3048 	while (*p != NULL) {
3049 		parent = *p;
3050 		entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3051 		cmp = access_cmp(cred, entry);
3052 
3053 		if (cmp < 0)
3054 			p = &parent->rb_left;
3055 		else if (cmp > 0)
3056 			p = &parent->rb_right;
3057 		else
3058 			goto found;
3059 	}
3060 	rb_link_node(&set->rb_node, parent, p);
3061 	rb_insert_color(&set->rb_node, root_node);
3062 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3063 	spin_unlock(&inode->i_lock);
3064 	return;
3065 found:
3066 	rb_replace_node(parent, &set->rb_node, root_node);
3067 	list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3068 	list_del(&entry->lru);
3069 	spin_unlock(&inode->i_lock);
3070 	nfs_access_free_entry(entry);
3071 }
3072 
nfs_access_add_cache(struct inode * inode,struct nfs_access_entry * set,const struct cred * cred)3073 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3074 			  const struct cred *cred)
3075 {
3076 	struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
3077 	if (cache == NULL)
3078 		return;
3079 	RB_CLEAR_NODE(&cache->rb_node);
3080 	cache->fsuid = cred->fsuid;
3081 	cache->fsgid = cred->fsgid;
3082 	cache->group_info = get_group_info(cred->group_info);
3083 	cache->mask = set->mask;
3084 
3085 	/* The above field assignments must be visible
3086 	 * before this item appears on the lru.  We cannot easily
3087 	 * use rcu_assign_pointer, so just force the memory barrier.
3088 	 */
3089 	smp_wmb();
3090 	nfs_access_add_rbtree(inode, cache, cred);
3091 
3092 	/* Update accounting */
3093 	smp_mb__before_atomic();
3094 	atomic_long_inc(&nfs_access_nr_entries);
3095 	smp_mb__after_atomic();
3096 
3097 	/* Add inode to global LRU list */
3098 	if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3099 		spin_lock(&nfs_access_lru_lock);
3100 		if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
3101 			list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
3102 					&nfs_access_lru_list);
3103 		spin_unlock(&nfs_access_lru_lock);
3104 	}
3105 	nfs_access_cache_enforce_limit();
3106 }
3107 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3108 
3109 #define NFS_MAY_READ (NFS_ACCESS_READ)
3110 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3111 		NFS_ACCESS_EXTEND | \
3112 		NFS_ACCESS_DELETE)
3113 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3114 		NFS_ACCESS_EXTEND)
3115 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3116 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3117 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3118 static int
nfs_access_calc_mask(u32 access_result,umode_t umode)3119 nfs_access_calc_mask(u32 access_result, umode_t umode)
3120 {
3121 	int mask = 0;
3122 
3123 	if (access_result & NFS_MAY_READ)
3124 		mask |= MAY_READ;
3125 	if (S_ISDIR(umode)) {
3126 		if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3127 			mask |= MAY_WRITE;
3128 		if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3129 			mask |= MAY_EXEC;
3130 	} else if (S_ISREG(umode)) {
3131 		if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3132 			mask |= MAY_WRITE;
3133 		if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3134 			mask |= MAY_EXEC;
3135 	} else if (access_result & NFS_MAY_WRITE)
3136 			mask |= MAY_WRITE;
3137 	return mask;
3138 }
3139 
nfs_access_set_mask(struct nfs_access_entry * entry,u32 access_result)3140 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3141 {
3142 	entry->mask = access_result;
3143 }
3144 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3145 
nfs_do_access(struct inode * inode,const struct cred * cred,int mask)3146 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3147 {
3148 	struct nfs_access_entry cache;
3149 	bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3150 	int cache_mask = -1;
3151 	int status;
3152 
3153 	trace_nfs_access_enter(inode);
3154 
3155 	status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3156 	if (status == 0)
3157 		goto out_cached;
3158 
3159 	status = -ECHILD;
3160 	if (!may_block)
3161 		goto out;
3162 
3163 	/*
3164 	 * Determine which access bits we want to ask for...
3165 	 */
3166 	cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3167 		     nfs_access_xattr_mask(NFS_SERVER(inode));
3168 	if (S_ISDIR(inode->i_mode))
3169 		cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3170 	else
3171 		cache.mask |= NFS_ACCESS_EXECUTE;
3172 	status = NFS_PROTO(inode)->access(inode, &cache, cred);
3173 	if (status != 0) {
3174 		if (status == -ESTALE) {
3175 			if (!S_ISDIR(inode->i_mode))
3176 				nfs_set_inode_stale(inode);
3177 			else
3178 				nfs_zap_caches(inode);
3179 		}
3180 		goto out;
3181 	}
3182 	nfs_access_add_cache(inode, &cache, cred);
3183 out_cached:
3184 	cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
3185 	if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3186 		status = -EACCES;
3187 out:
3188 	trace_nfs_access_exit(inode, mask, cache_mask, status);
3189 	return status;
3190 }
3191 
nfs_open_permission_mask(int openflags)3192 static int nfs_open_permission_mask(int openflags)
3193 {
3194 	int mask = 0;
3195 
3196 	if (openflags & __FMODE_EXEC) {
3197 		/* ONLY check exec rights */
3198 		mask = MAY_EXEC;
3199 	} else {
3200 		if ((openflags & O_ACCMODE) != O_WRONLY)
3201 			mask |= MAY_READ;
3202 		if ((openflags & O_ACCMODE) != O_RDONLY)
3203 			mask |= MAY_WRITE;
3204 	}
3205 
3206 	return mask;
3207 }
3208 
nfs_may_open(struct inode * inode,const struct cred * cred,int openflags)3209 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3210 {
3211 	return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
3212 }
3213 EXPORT_SYMBOL_GPL(nfs_may_open);
3214 
nfs_execute_ok(struct inode * inode,int mask)3215 static int nfs_execute_ok(struct inode *inode, int mask)
3216 {
3217 	struct nfs_server *server = NFS_SERVER(inode);
3218 	int ret = 0;
3219 
3220 	if (S_ISDIR(inode->i_mode))
3221 		return 0;
3222 	if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3223 		if (mask & MAY_NOT_BLOCK)
3224 			return -ECHILD;
3225 		ret = __nfs_revalidate_inode(server, inode);
3226 	}
3227 	if (ret == 0 && !execute_ok(inode))
3228 		ret = -EACCES;
3229 	return ret;
3230 }
3231 
nfs_permission(struct user_namespace * mnt_userns,struct inode * inode,int mask)3232 int nfs_permission(struct user_namespace *mnt_userns,
3233 		   struct inode *inode,
3234 		   int mask)
3235 {
3236 	const struct cred *cred = current_cred();
3237 	int res = 0;
3238 
3239 	nfs_inc_stats(inode, NFSIOS_VFSACCESS);
3240 
3241 	if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3242 		goto out;
3243 	/* Is this sys_access() ? */
3244 	if (mask & (MAY_ACCESS | MAY_CHDIR))
3245 		goto force_lookup;
3246 
3247 	switch (inode->i_mode & S_IFMT) {
3248 		case S_IFLNK:
3249 			goto out;
3250 		case S_IFREG:
3251 			if ((mask & MAY_OPEN) &&
3252 			   nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3253 				return 0;
3254 			break;
3255 		case S_IFDIR:
3256 			/*
3257 			 * Optimize away all write operations, since the server
3258 			 * will check permissions when we perform the op.
3259 			 */
3260 			if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3261 				goto out;
3262 	}
3263 
3264 force_lookup:
3265 	if (!NFS_PROTO(inode)->access)
3266 		goto out_notsup;
3267 
3268 	res = nfs_do_access(inode, cred, mask);
3269 out:
3270 	if (!res && (mask & MAY_EXEC))
3271 		res = nfs_execute_ok(inode, mask);
3272 
3273 	dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3274 		inode->i_sb->s_id, inode->i_ino, mask, res);
3275 	return res;
3276 out_notsup:
3277 	if (mask & MAY_NOT_BLOCK)
3278 		return -ECHILD;
3279 
3280 	res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3281 						  NFS_INO_INVALID_OTHER);
3282 	if (res == 0)
3283 		res = generic_permission(&init_user_ns, inode, mask);
3284 	goto out;
3285 }
3286 EXPORT_SYMBOL_GPL(nfs_permission);
3287