Loading...
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
53static int nfs_opendir(struct inode *, struct file *);
54static int nfs_closedir(struct inode *, struct file *);
55static int nfs_readdir(struct file *, struct dir_context *);
56static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
57static loff_t nfs_llseek_dir(struct file *, loff_t, int);
58static void nfs_readdir_free_folio(struct folio *);
59
60const 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
69const struct address_space_operations nfs_dir_aops = {
70 .free_folio = nfs_readdir_free_folio,
71};
72
73#define NFS_INIT_DTSIZE PAGE_SIZE
74
75static struct nfs_open_dir_context *
76alloc_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
99static 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 */
110static int
111nfs_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;
126out:
127 return res;
128}
129
130static int
131nfs_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
137struct 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
145struct 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
155struct 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
179static 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
191static void nfs_shrink_dtsize(struct nfs_readdir_descriptor *desc)
192{
193 nfs_set_dtsize(desc, desc->dtsize >> 1);
194}
195
196static void nfs_grow_dtsize(struct nfs_readdir_descriptor *desc)
197{
198 nfs_set_dtsize(desc, desc->dtsize << 1);
199}
200
201static 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 */
219static 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
231static void nfs_readdir_free_folio(struct folio *folio)
232{
233 nfs_readdir_clear_array(&folio->page);
234}
235
236static 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
243static struct page *
244nfs_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
252static 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
260static 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
265static 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
271static 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 */
281static 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
294static 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 */
303static 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
314static 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);
346out:
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 */
364static 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
371static 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
385static void nfs_readdir_page_unlock_and_put(struct page *page)
386{
387 unlock_page(page);
388 put_page(page);
389}
390
391static 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
403static 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
416static 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
427static 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
438static 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
447static 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
462static inline
463int 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
472static
473bool 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
481static 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
493static 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
500static 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;
519out_eof:
520 desc->eof = true;
521 return -EBADCOOKIE;
522}
523
524static 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
537static 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 }
556check_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
566static 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 */
583static 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;
620error:
621 return error;
622}
623
624static 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 */
641static
642int 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
664static 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 */
681void 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 */
700void 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
714static 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
724static
725void 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);
756again:
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);
802out:
803 dput(dentry);
804}
805
806static 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 */
822static 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
896static 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 */
907static 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
923out_freepages:
924 nfs_readdir_free_pages(pages, i);
925 return NULL;
926}
927
928static 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
971free_pages:
972 nfs_readdir_free_pages(pages, array_size);
973out:
974 nfs_free_fattr(entry->fattr);
975 nfs_free_fhandle(entry->fh);
976 kfree(entry);
977 return status;
978}
979
980static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc)
981{
982 put_page(desc->page);
983 desc->page = NULL;
984}
985
986static void
987nfs_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
993static struct page *
994nfs_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 */
1013static 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 */
1067static 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#define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1078
1079/*
1080 * Once we've found the start of the dirent within a page: fill 'er up...
1081 */
1082static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
1083 const __be32 *verf)
1084{
1085 struct file *file = desc->file;
1086 struct nfs_cache_array *array;
1087 unsigned int i;
1088 bool first_emit = !desc->dir_cookie;
1089
1090 array = kmap_local_page(desc->page);
1091 for (i = desc->cache_entry_index; i < array->size; i++) {
1092 struct nfs_cache_array_entry *ent;
1093
1094 ent = &array->array[i];
1095 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
1096 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
1097 desc->eob = true;
1098 break;
1099 }
1100 memcpy(desc->verf, verf, sizeof(desc->verf));
1101 if (i == array->size - 1) {
1102 desc->dir_cookie = array->last_cookie;
1103 nfs_readdir_seek_next_array(array, desc);
1104 } else {
1105 desc->dir_cookie = array->array[i + 1].cookie;
1106 desc->last_cookie = array->array[0].cookie;
1107 }
1108 if (nfs_readdir_use_cookie(file))
1109 desc->ctx->pos = desc->dir_cookie;
1110 else
1111 desc->ctx->pos++;
1112 if (first_emit && i > NFS_READDIR_CACHE_MISS_THRESHOLD + 1) {
1113 desc->eob = true;
1114 break;
1115 }
1116 }
1117 if (array->page_is_eof)
1118 desc->eof = !desc->eob;
1119
1120 kunmap_local(array);
1121 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1122 (unsigned long long)desc->dir_cookie);
1123}
1124
1125/*
1126 * If we cannot find a cookie in our cache, we suspect that this is
1127 * because it points to a deleted file, so we ask the server to return
1128 * whatever it thinks is the next entry. We then feed this to filldir.
1129 * If all goes well, we should then be able to find our way round the
1130 * cache on the next call to readdir_search_pagecache();
1131 *
1132 * NOTE: we cannot add the anonymous page to the pagecache because
1133 * the data it contains might not be page aligned. Besides,
1134 * we should already have a complete representation of the
1135 * directory in the page cache by the time we get here.
1136 */
1137static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1138{
1139 struct page **arrays;
1140 size_t i, sz = 512;
1141 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1142 int status = -ENOMEM;
1143
1144 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1145 (unsigned long long)desc->dir_cookie);
1146
1147 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1148 if (!arrays)
1149 goto out;
1150 arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL);
1151 if (!arrays[0])
1152 goto out;
1153
1154 desc->page_index = 0;
1155 desc->cache_entry_index = 0;
1156 desc->last_cookie = desc->dir_cookie;
1157 desc->page_index_max = 0;
1158
1159 trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie,
1160 -1, desc->dtsize);
1161
1162 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1163 if (status < 0) {
1164 trace_nfs_readdir_uncached_done(file_inode(desc->file), status);
1165 goto out_free;
1166 }
1167
1168 for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1169 desc->page = arrays[i];
1170 nfs_do_filldir(desc, verf);
1171 }
1172 desc->page = NULL;
1173
1174 /*
1175 * Grow the dtsize if we have to go back for more pages,
1176 * or shrink it if we're reading too many.
1177 */
1178 if (!desc->eof) {
1179 if (!desc->eob)
1180 nfs_grow_dtsize(desc);
1181 else if (desc->buffer_fills == 1 &&
1182 i < (desc->page_index_max >> 1))
1183 nfs_shrink_dtsize(desc);
1184 }
1185out_free:
1186 for (i = 0; i < sz && arrays[i]; i++)
1187 nfs_readdir_page_array_free(arrays[i]);
1188out:
1189 if (!nfs_readdir_use_cookie(desc->file))
1190 nfs_readdir_rewind_search(desc);
1191 desc->page_index_max = -1;
1192 kfree(arrays);
1193 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1194 return status;
1195}
1196
1197static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1198 struct nfs_readdir_descriptor *desc,
1199 unsigned int cache_misses,
1200 bool force_clear)
1201{
1202 if (desc->ctx->pos == 0 || !desc->plus)
1203 return false;
1204 if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1205 return false;
1206 trace_nfs_readdir_force_readdirplus(inode);
1207 return true;
1208}
1209
1210/* The file offset position represents the dirent entry number. A
1211 last cookie cache takes care of the common case of reading the
1212 whole directory.
1213 */
1214static int nfs_readdir(struct file *file, struct dir_context *ctx)
1215{
1216 struct dentry *dentry = file_dentry(file);
1217 struct inode *inode = d_inode(dentry);
1218 struct nfs_inode *nfsi = NFS_I(inode);
1219 struct nfs_open_dir_context *dir_ctx = file->private_data;
1220 struct nfs_readdir_descriptor *desc;
1221 unsigned int cache_hits, cache_misses;
1222 bool force_clear;
1223 int res;
1224
1225 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1226 file, (long long)ctx->pos);
1227 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1228
1229 /*
1230 * ctx->pos points to the dirent entry number.
1231 * *desc->dir_cookie has the cookie for the next entry. We have
1232 * to either find the entry with the appropriate number or
1233 * revalidate the cookie.
1234 */
1235 nfs_revalidate_mapping(inode, file->f_mapping);
1236
1237 res = -ENOMEM;
1238 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1239 if (!desc)
1240 goto out;
1241 desc->file = file;
1242 desc->ctx = ctx;
1243 desc->page_index_max = -1;
1244
1245 spin_lock(&file->f_lock);
1246 desc->dir_cookie = dir_ctx->dir_cookie;
1247 desc->page_index = dir_ctx->page_index;
1248 desc->last_cookie = dir_ctx->last_cookie;
1249 desc->attr_gencount = dir_ctx->attr_gencount;
1250 desc->eof = dir_ctx->eof;
1251 nfs_set_dtsize(desc, dir_ctx->dtsize);
1252 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1253 cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0);
1254 cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0);
1255 force_clear = dir_ctx->force_clear;
1256 spin_unlock(&file->f_lock);
1257
1258 if (desc->eof) {
1259 res = 0;
1260 goto out_free;
1261 }
1262
1263 desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses);
1264 force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1265 force_clear);
1266 desc->clear_cache = force_clear;
1267
1268 do {
1269 res = readdir_search_pagecache(desc);
1270
1271 if (res == -EBADCOOKIE) {
1272 res = 0;
1273 /* This means either end of directory */
1274 if (desc->dir_cookie && !desc->eof) {
1275 /* Or that the server has 'lost' a cookie */
1276 res = uncached_readdir(desc);
1277 if (res == 0)
1278 continue;
1279 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1280 res = 0;
1281 }
1282 break;
1283 }
1284 if (res == -ETOOSMALL && desc->plus) {
1285 nfs_zap_caches(inode);
1286 desc->plus = false;
1287 desc->eof = false;
1288 continue;
1289 }
1290 if (res < 0)
1291 break;
1292
1293 nfs_do_filldir(desc, nfsi->cookieverf);
1294 nfs_readdir_page_unlock_and_put_cached(desc);
1295 if (desc->page_index == desc->page_index_max)
1296 desc->clear_cache = force_clear;
1297 } while (!desc->eob && !desc->eof);
1298
1299 spin_lock(&file->f_lock);
1300 dir_ctx->dir_cookie = desc->dir_cookie;
1301 dir_ctx->last_cookie = desc->last_cookie;
1302 dir_ctx->attr_gencount = desc->attr_gencount;
1303 dir_ctx->page_index = desc->page_index;
1304 dir_ctx->force_clear = force_clear;
1305 dir_ctx->eof = desc->eof;
1306 dir_ctx->dtsize = desc->dtsize;
1307 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1308 spin_unlock(&file->f_lock);
1309out_free:
1310 kfree(desc);
1311
1312out:
1313 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1314 return res;
1315}
1316
1317static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1318{
1319 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1320
1321 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1322 filp, offset, whence);
1323
1324 switch (whence) {
1325 default:
1326 return -EINVAL;
1327 case SEEK_SET:
1328 if (offset < 0)
1329 return -EINVAL;
1330 spin_lock(&filp->f_lock);
1331 break;
1332 case SEEK_CUR:
1333 if (offset == 0)
1334 return filp->f_pos;
1335 spin_lock(&filp->f_lock);
1336 offset += filp->f_pos;
1337 if (offset < 0) {
1338 spin_unlock(&filp->f_lock);
1339 return -EINVAL;
1340 }
1341 }
1342 if (offset != filp->f_pos) {
1343 filp->f_pos = offset;
1344 dir_ctx->page_index = 0;
1345 if (!nfs_readdir_use_cookie(filp)) {
1346 dir_ctx->dir_cookie = 0;
1347 dir_ctx->last_cookie = 0;
1348 } else {
1349 dir_ctx->dir_cookie = offset;
1350 dir_ctx->last_cookie = offset;
1351 }
1352 dir_ctx->eof = false;
1353 }
1354 spin_unlock(&filp->f_lock);
1355 return offset;
1356}
1357
1358/*
1359 * All directory operations under NFS are synchronous, so fsync()
1360 * is a dummy operation.
1361 */
1362static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1363 int datasync)
1364{
1365 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1366
1367 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1368 return 0;
1369}
1370
1371/**
1372 * nfs_force_lookup_revalidate - Mark the directory as having changed
1373 * @dir: pointer to directory inode
1374 *
1375 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1376 * full lookup on all child dentries of 'dir' whenever a change occurs
1377 * on the server that might have invalidated our dcache.
1378 *
1379 * Note that we reserve bit '0' as a tag to let us know when a dentry
1380 * was revalidated while holding a delegation on its inode.
1381 *
1382 * The caller should be holding dir->i_lock
1383 */
1384void nfs_force_lookup_revalidate(struct inode *dir)
1385{
1386 NFS_I(dir)->cache_change_attribute += 2;
1387}
1388EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1389
1390/**
1391 * nfs_verify_change_attribute - Detects NFS remote directory changes
1392 * @dir: pointer to parent directory inode
1393 * @verf: previously saved change attribute
1394 *
1395 * Return "false" if the verifiers doesn't match the change attribute.
1396 * This would usually indicate that the directory contents have changed on
1397 * the server, and that any dentries need revalidating.
1398 */
1399static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1400{
1401 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1402}
1403
1404static void nfs_set_verifier_delegated(unsigned long *verf)
1405{
1406 *verf |= 1UL;
1407}
1408
1409#if IS_ENABLED(CONFIG_NFS_V4)
1410static void nfs_unset_verifier_delegated(unsigned long *verf)
1411{
1412 *verf &= ~1UL;
1413}
1414#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1415
1416static bool nfs_test_verifier_delegated(unsigned long verf)
1417{
1418 return verf & 1;
1419}
1420
1421static bool nfs_verifier_is_delegated(struct dentry *dentry)
1422{
1423 return nfs_test_verifier_delegated(dentry->d_time);
1424}
1425
1426static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1427{
1428 struct inode *inode = d_inode(dentry);
1429 struct inode *dir = d_inode(dentry->d_parent);
1430
1431 if (!nfs_verify_change_attribute(dir, verf))
1432 return;
1433 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1434 nfs_set_verifier_delegated(&verf);
1435 dentry->d_time = verf;
1436}
1437
1438/**
1439 * nfs_set_verifier - save a parent directory verifier in the dentry
1440 * @dentry: pointer to dentry
1441 * @verf: verifier to save
1442 *
1443 * Saves the parent directory verifier in @dentry. If the inode has
1444 * a delegation, we also tag the dentry as having been revalidated
1445 * while holding a delegation so that we know we don't have to
1446 * look it up again after a directory change.
1447 */
1448void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1449{
1450
1451 spin_lock(&dentry->d_lock);
1452 nfs_set_verifier_locked(dentry, verf);
1453 spin_unlock(&dentry->d_lock);
1454}
1455EXPORT_SYMBOL_GPL(nfs_set_verifier);
1456
1457#if IS_ENABLED(CONFIG_NFS_V4)
1458/**
1459 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1460 * @inode: pointer to inode
1461 *
1462 * Iterates through the dentries in the inode alias list and clears
1463 * the tag used to indicate that the dentry has been revalidated
1464 * while holding a delegation.
1465 * This function is intended for use when the delegation is being
1466 * returned or revoked.
1467 */
1468void nfs_clear_verifier_delegated(struct inode *inode)
1469{
1470 struct dentry *alias;
1471
1472 if (!inode)
1473 return;
1474 spin_lock(&inode->i_lock);
1475 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1476 spin_lock(&alias->d_lock);
1477 nfs_unset_verifier_delegated(&alias->d_time);
1478 spin_unlock(&alias->d_lock);
1479 }
1480 spin_unlock(&inode->i_lock);
1481}
1482EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1483#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1484
1485static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1486{
1487 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) &&
1488 d_really_is_negative(dentry))
1489 return dentry->d_time == inode_peek_iversion_raw(dir);
1490 return nfs_verify_change_attribute(dir, dentry->d_time);
1491}
1492
1493/*
1494 * A check for whether or not the parent directory has changed.
1495 * In the case it has, we assume that the dentries are untrustworthy
1496 * and may need to be looked up again.
1497 * If rcu_walk prevents us from performing a full check, return 0.
1498 */
1499static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1500 int rcu_walk)
1501{
1502 if (IS_ROOT(dentry))
1503 return 1;
1504 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1505 return 0;
1506 if (!nfs_dentry_verify_change(dir, dentry))
1507 return 0;
1508 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1509 if (nfs_mapping_need_revalidate_inode(dir)) {
1510 if (rcu_walk)
1511 return 0;
1512 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1513 return 0;
1514 }
1515 if (!nfs_dentry_verify_change(dir, dentry))
1516 return 0;
1517 return 1;
1518}
1519
1520/*
1521 * Use intent information to check whether or not we're going to do
1522 * an O_EXCL create using this path component.
1523 */
1524static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1525{
1526 if (NFS_PROTO(dir)->version == 2)
1527 return 0;
1528 return flags & LOOKUP_EXCL;
1529}
1530
1531/*
1532 * Inode and filehandle revalidation for lookups.
1533 *
1534 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1535 * or if the intent information indicates that we're about to open this
1536 * particular file and the "nocto" mount flag is not set.
1537 *
1538 */
1539static
1540int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1541{
1542 struct nfs_server *server = NFS_SERVER(inode);
1543 int ret;
1544
1545 if (IS_AUTOMOUNT(inode))
1546 return 0;
1547
1548 if (flags & LOOKUP_OPEN) {
1549 switch (inode->i_mode & S_IFMT) {
1550 case S_IFREG:
1551 /* A NFSv4 OPEN will revalidate later */
1552 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1553 goto out;
1554 fallthrough;
1555 case S_IFDIR:
1556 if (server->flags & NFS_MOUNT_NOCTO)
1557 break;
1558 /* NFS close-to-open cache consistency validation */
1559 goto out_force;
1560 }
1561 }
1562
1563 /* VFS wants an on-the-wire revalidation */
1564 if (flags & LOOKUP_REVAL)
1565 goto out_force;
1566out:
1567 if (inode->i_nlink > 0 ||
1568 (inode->i_nlink == 0 &&
1569 test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1570 return 0;
1571 else
1572 return -ESTALE;
1573out_force:
1574 if (flags & LOOKUP_RCU)
1575 return -ECHILD;
1576 ret = __nfs_revalidate_inode(server, inode);
1577 if (ret != 0)
1578 return ret;
1579 goto out;
1580}
1581
1582static void nfs_mark_dir_for_revalidate(struct inode *inode)
1583{
1584 spin_lock(&inode->i_lock);
1585 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1586 spin_unlock(&inode->i_lock);
1587}
1588
1589/*
1590 * We judge how long we want to trust negative
1591 * dentries by looking at the parent inode mtime.
1592 *
1593 * If parent mtime has changed, we revalidate, else we wait for a
1594 * period corresponding to the parent's attribute cache timeout value.
1595 *
1596 * If LOOKUP_RCU prevents us from performing a full check, return 1
1597 * suggesting a reval is needed.
1598 *
1599 * Note that when creating a new file, or looking up a rename target,
1600 * then it shouldn't be necessary to revalidate a negative dentry.
1601 */
1602static inline
1603int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1604 unsigned int flags)
1605{
1606 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1607 return 0;
1608 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1609 return 1;
1610 /* Case insensitive server? Revalidate negative dentries */
1611 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1612 return 1;
1613 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1614}
1615
1616static int
1617nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1618 struct inode *inode, int error)
1619{
1620 switch (error) {
1621 case 1:
1622 break;
1623 case 0:
1624 /*
1625 * We can't d_drop the root of a disconnected tree:
1626 * its d_hash is on the s_anon list and d_drop() would hide
1627 * it from shrink_dcache_for_unmount(), leading to busy
1628 * inodes on unmount and further oopses.
1629 */
1630 if (inode && IS_ROOT(dentry))
1631 error = 1;
1632 break;
1633 }
1634 trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error);
1635 return error;
1636}
1637
1638static int
1639nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1640 unsigned int flags)
1641{
1642 int ret = 1;
1643 if (nfs_neg_need_reval(dir, dentry, flags)) {
1644 if (flags & LOOKUP_RCU)
1645 return -ECHILD;
1646 ret = 0;
1647 }
1648 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1649}
1650
1651static int
1652nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1653 struct inode *inode)
1654{
1655 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1656 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1657}
1658
1659static int nfs_lookup_revalidate_dentry(struct inode *dir,
1660 struct dentry *dentry,
1661 struct inode *inode, unsigned int flags)
1662{
1663 struct nfs_fh *fhandle;
1664 struct nfs_fattr *fattr;
1665 unsigned long dir_verifier;
1666 int ret;
1667
1668 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1669
1670 ret = -ENOMEM;
1671 fhandle = nfs_alloc_fhandle();
1672 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1673 if (fhandle == NULL || fattr == NULL)
1674 goto out;
1675
1676 dir_verifier = nfs_save_change_attribute(dir);
1677 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1678 if (ret < 0) {
1679 switch (ret) {
1680 case -ESTALE:
1681 case -ENOENT:
1682 ret = 0;
1683 break;
1684 case -ETIMEDOUT:
1685 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1686 ret = 1;
1687 }
1688 goto out;
1689 }
1690
1691 /* Request help from readdirplus */
1692 nfs_lookup_advise_force_readdirplus(dir, flags);
1693
1694 ret = 0;
1695 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1696 goto out;
1697 if (nfs_refresh_inode(inode, fattr) < 0)
1698 goto out;
1699
1700 nfs_setsecurity(inode, fattr);
1701 nfs_set_verifier(dentry, dir_verifier);
1702
1703 ret = 1;
1704out:
1705 nfs_free_fattr(fattr);
1706 nfs_free_fhandle(fhandle);
1707
1708 /*
1709 * If the lookup failed despite the dentry change attribute being
1710 * a match, then we should revalidate the directory cache.
1711 */
1712 if (!ret && nfs_dentry_verify_change(dir, dentry))
1713 nfs_mark_dir_for_revalidate(dir);
1714 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1715}
1716
1717/*
1718 * This is called every time the dcache has a lookup hit,
1719 * and we should check whether we can really trust that
1720 * lookup.
1721 *
1722 * NOTE! The hit can be a negative hit too, don't assume
1723 * we have an inode!
1724 *
1725 * If the parent directory is seen to have changed, we throw out the
1726 * cached dentry and do a new lookup.
1727 */
1728static int
1729nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1730 unsigned int flags)
1731{
1732 struct inode *inode;
1733 int error;
1734
1735 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1736 inode = d_inode(dentry);
1737
1738 if (!inode)
1739 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1740
1741 if (is_bad_inode(inode)) {
1742 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1743 __func__, dentry);
1744 goto out_bad;
1745 }
1746
1747 if ((flags & LOOKUP_RENAME_TARGET) && d_count(dentry) < 2 &&
1748 nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1749 goto out_bad;
1750
1751 if (nfs_verifier_is_delegated(dentry))
1752 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1753
1754 /* Force a full look up iff the parent directory has changed */
1755 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1756 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1757 error = nfs_lookup_verify_inode(inode, flags);
1758 if (error) {
1759 if (error == -ESTALE)
1760 nfs_mark_dir_for_revalidate(dir);
1761 goto out_bad;
1762 }
1763 goto out_valid;
1764 }
1765
1766 if (flags & LOOKUP_RCU)
1767 return -ECHILD;
1768
1769 if (NFS_STALE(inode))
1770 goto out_bad;
1771
1772 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
1773out_valid:
1774 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1775out_bad:
1776 if (flags & LOOKUP_RCU)
1777 return -ECHILD;
1778 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1779}
1780
1781static int
1782__nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1783 int (*reval)(struct inode *, struct dentry *, unsigned int))
1784{
1785 struct dentry *parent;
1786 struct inode *dir;
1787 int ret;
1788
1789 if (flags & LOOKUP_RCU) {
1790 if (dentry->d_fsdata == NFS_FSDATA_BLOCKED)
1791 return -ECHILD;
1792 parent = READ_ONCE(dentry->d_parent);
1793 dir = d_inode_rcu(parent);
1794 if (!dir)
1795 return -ECHILD;
1796 ret = reval(dir, dentry, flags);
1797 if (parent != READ_ONCE(dentry->d_parent))
1798 return -ECHILD;
1799 } else {
1800 /* Wait for unlink to complete */
1801 wait_var_event(&dentry->d_fsdata,
1802 dentry->d_fsdata != NFS_FSDATA_BLOCKED);
1803 parent = dget_parent(dentry);
1804 ret = reval(d_inode(parent), dentry, flags);
1805 dput(parent);
1806 }
1807 return ret;
1808}
1809
1810static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1811{
1812 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1813}
1814
1815/*
1816 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1817 * when we don't really care about the dentry name. This is called when a
1818 * pathwalk ends on a dentry that was not found via a normal lookup in the
1819 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1820 *
1821 * In this situation, we just want to verify that the inode itself is OK
1822 * since the dentry might have changed on the server.
1823 */
1824static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1825{
1826 struct inode *inode = d_inode(dentry);
1827 int error = 0;
1828
1829 /*
1830 * I believe we can only get a negative dentry here in the case of a
1831 * procfs-style symlink. Just assume it's correct for now, but we may
1832 * eventually need to do something more here.
1833 */
1834 if (!inode) {
1835 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1836 __func__, dentry);
1837 return 1;
1838 }
1839
1840 if (is_bad_inode(inode)) {
1841 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1842 __func__, dentry);
1843 return 0;
1844 }
1845
1846 error = nfs_lookup_verify_inode(inode, flags);
1847 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1848 __func__, inode->i_ino, error ? "invalid" : "valid");
1849 return !error;
1850}
1851
1852/*
1853 * This is called from dput() when d_count is going to 0.
1854 */
1855static int nfs_dentry_delete(const struct dentry *dentry)
1856{
1857 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1858 dentry, dentry->d_flags);
1859
1860 /* Unhash any dentry with a stale inode */
1861 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1862 return 1;
1863
1864 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1865 /* Unhash it, so that ->d_iput() would be called */
1866 return 1;
1867 }
1868 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1869 /* Unhash it, so that ancestors of killed async unlink
1870 * files will be cleaned up during umount */
1871 return 1;
1872 }
1873 return 0;
1874
1875}
1876
1877/* Ensure that we revalidate inode->i_nlink */
1878static void nfs_drop_nlink(struct inode *inode)
1879{
1880 spin_lock(&inode->i_lock);
1881 /* drop the inode if we're reasonably sure this is the last link */
1882 if (inode->i_nlink > 0)
1883 drop_nlink(inode);
1884 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1885 nfs_set_cache_invalid(
1886 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1887 NFS_INO_INVALID_NLINK);
1888 spin_unlock(&inode->i_lock);
1889}
1890
1891/*
1892 * Called when the dentry loses inode.
1893 * We use it to clean up silly-renamed files.
1894 */
1895static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1896{
1897 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1898 nfs_complete_unlink(dentry, inode);
1899 nfs_drop_nlink(inode);
1900 }
1901 iput(inode);
1902}
1903
1904static void nfs_d_release(struct dentry *dentry)
1905{
1906 /* free cached devname value, if it survived that far */
1907 if (unlikely(dentry->d_fsdata)) {
1908 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1909 WARN_ON(1);
1910 else
1911 kfree(dentry->d_fsdata);
1912 }
1913}
1914
1915const struct dentry_operations nfs_dentry_operations = {
1916 .d_revalidate = nfs_lookup_revalidate,
1917 .d_weak_revalidate = nfs_weak_revalidate,
1918 .d_delete = nfs_dentry_delete,
1919 .d_iput = nfs_dentry_iput,
1920 .d_automount = nfs_d_automount,
1921 .d_release = nfs_d_release,
1922};
1923EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1924
1925struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1926{
1927 struct dentry *res;
1928 struct inode *inode = NULL;
1929 struct nfs_fh *fhandle = NULL;
1930 struct nfs_fattr *fattr = NULL;
1931 unsigned long dir_verifier;
1932 int error;
1933
1934 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1935 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1936
1937 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1938 return ERR_PTR(-ENAMETOOLONG);
1939
1940 /*
1941 * If we're doing an exclusive create, optimize away the lookup
1942 * but don't hash the dentry.
1943 */
1944 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1945 return NULL;
1946
1947 res = ERR_PTR(-ENOMEM);
1948 fhandle = nfs_alloc_fhandle();
1949 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir));
1950 if (fhandle == NULL || fattr == NULL)
1951 goto out;
1952
1953 dir_verifier = nfs_save_change_attribute(dir);
1954 trace_nfs_lookup_enter(dir, dentry, flags);
1955 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1956 if (error == -ENOENT) {
1957 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1958 dir_verifier = inode_peek_iversion_raw(dir);
1959 goto no_entry;
1960 }
1961 if (error < 0) {
1962 res = ERR_PTR(error);
1963 goto out;
1964 }
1965 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1966 res = ERR_CAST(inode);
1967 if (IS_ERR(res))
1968 goto out;
1969
1970 /* Notify readdir to use READDIRPLUS */
1971 nfs_lookup_advise_force_readdirplus(dir, flags);
1972
1973no_entry:
1974 res = d_splice_alias(inode, dentry);
1975 if (res != NULL) {
1976 if (IS_ERR(res))
1977 goto out;
1978 dentry = res;
1979 }
1980 nfs_set_verifier(dentry, dir_verifier);
1981out:
1982 trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res));
1983 nfs_free_fattr(fattr);
1984 nfs_free_fhandle(fhandle);
1985 return res;
1986}
1987EXPORT_SYMBOL_GPL(nfs_lookup);
1988
1989void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
1990{
1991 /* Case insensitive server? Revalidate dentries */
1992 if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
1993 d_prune_aliases(inode);
1994}
1995EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
1996
1997#if IS_ENABLED(CONFIG_NFS_V4)
1998static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1999
2000const struct dentry_operations nfs4_dentry_operations = {
2001 .d_revalidate = nfs4_lookup_revalidate,
2002 .d_weak_revalidate = nfs_weak_revalidate,
2003 .d_delete = nfs_dentry_delete,
2004 .d_iput = nfs_dentry_iput,
2005 .d_automount = nfs_d_automount,
2006 .d_release = nfs_d_release,
2007};
2008EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
2009
2010static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
2011{
2012 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
2013}
2014
2015static int do_open(struct inode *inode, struct file *filp)
2016{
2017 nfs_fscache_open_file(inode, filp);
2018 return 0;
2019}
2020
2021static int nfs_finish_open(struct nfs_open_context *ctx,
2022 struct dentry *dentry,
2023 struct file *file, unsigned open_flags)
2024{
2025 int err;
2026
2027 err = finish_open(file, dentry, do_open);
2028 if (err)
2029 goto out;
2030 if (S_ISREG(file_inode(file)->i_mode))
2031 nfs_file_set_open_context(file, ctx);
2032 else
2033 err = -EOPENSTALE;
2034out:
2035 return err;
2036}
2037
2038int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2039 struct file *file, unsigned open_flags,
2040 umode_t mode)
2041{
2042 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2043 struct nfs_open_context *ctx;
2044 struct dentry *res;
2045 struct iattr attr = { .ia_valid = ATTR_OPEN };
2046 struct inode *inode;
2047 unsigned int lookup_flags = 0;
2048 unsigned long dir_verifier;
2049 bool switched = false;
2050 int created = 0;
2051 int err;
2052
2053 /* Expect a negative dentry */
2054 BUG_ON(d_inode(dentry));
2055
2056 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2057 dir->i_sb->s_id, dir->i_ino, dentry);
2058
2059 err = nfs_check_flags(open_flags);
2060 if (err)
2061 return err;
2062
2063 /* NFS only supports OPEN on regular files */
2064 if ((open_flags & O_DIRECTORY)) {
2065 if (!d_in_lookup(dentry)) {
2066 /*
2067 * Hashed negative dentry with O_DIRECTORY: dentry was
2068 * revalidated and is fine, no need to perform lookup
2069 * again
2070 */
2071 return -ENOENT;
2072 }
2073 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2074 goto no_open;
2075 }
2076
2077 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
2078 return -ENAMETOOLONG;
2079
2080 if (open_flags & O_CREAT) {
2081 struct nfs_server *server = NFS_SERVER(dir);
2082
2083 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2084 mode &= ~current_umask();
2085
2086 attr.ia_valid |= ATTR_MODE;
2087 attr.ia_mode = mode;
2088 }
2089 if (open_flags & O_TRUNC) {
2090 attr.ia_valid |= ATTR_SIZE;
2091 attr.ia_size = 0;
2092 }
2093
2094 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2095 d_drop(dentry);
2096 switched = true;
2097 dentry = d_alloc_parallel(dentry->d_parent,
2098 &dentry->d_name, &wq);
2099 if (IS_ERR(dentry))
2100 return PTR_ERR(dentry);
2101 if (unlikely(!d_in_lookup(dentry)))
2102 return finish_no_open(file, dentry);
2103 }
2104
2105 ctx = create_nfs_open_context(dentry, open_flags, file);
2106 err = PTR_ERR(ctx);
2107 if (IS_ERR(ctx))
2108 goto out;
2109
2110 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
2111 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
2112 if (created)
2113 file->f_mode |= FMODE_CREATED;
2114 if (IS_ERR(inode)) {
2115 err = PTR_ERR(inode);
2116 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2117 put_nfs_open_context(ctx);
2118 d_drop(dentry);
2119 switch (err) {
2120 case -ENOENT:
2121 d_splice_alias(NULL, dentry);
2122 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
2123 dir_verifier = inode_peek_iversion_raw(dir);
2124 else
2125 dir_verifier = nfs_save_change_attribute(dir);
2126 nfs_set_verifier(dentry, dir_verifier);
2127 break;
2128 case -EISDIR:
2129 case -ENOTDIR:
2130 goto no_open;
2131 case -ELOOP:
2132 if (!(open_flags & O_NOFOLLOW))
2133 goto no_open;
2134 break;
2135 /* case -EINVAL: */
2136 default:
2137 break;
2138 }
2139 goto out;
2140 }
2141 file->f_mode |= FMODE_CAN_ODIRECT;
2142
2143 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
2144 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2145 put_nfs_open_context(ctx);
2146out:
2147 if (unlikely(switched)) {
2148 d_lookup_done(dentry);
2149 dput(dentry);
2150 }
2151 return err;
2152
2153no_open:
2154 res = nfs_lookup(dir, dentry, lookup_flags);
2155 if (!res) {
2156 inode = d_inode(dentry);
2157 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2158 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2159 res = ERR_PTR(-ENOTDIR);
2160 else if (inode && S_ISREG(inode->i_mode))
2161 res = ERR_PTR(-EOPENSTALE);
2162 } else if (!IS_ERR(res)) {
2163 inode = d_inode(res);
2164 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2165 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2166 dput(res);
2167 res = ERR_PTR(-ENOTDIR);
2168 } else if (inode && S_ISREG(inode->i_mode)) {
2169 dput(res);
2170 res = ERR_PTR(-EOPENSTALE);
2171 }
2172 }
2173 if (switched) {
2174 d_lookup_done(dentry);
2175 if (!res)
2176 res = dentry;
2177 else
2178 dput(dentry);
2179 }
2180 if (IS_ERR(res))
2181 return PTR_ERR(res);
2182 return finish_no_open(file, res);
2183}
2184EXPORT_SYMBOL_GPL(nfs_atomic_open);
2185
2186static int
2187nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2188 unsigned int flags)
2189{
2190 struct inode *inode;
2191
2192 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2193 goto full_reval;
2194 if (d_mountpoint(dentry))
2195 goto full_reval;
2196
2197 inode = d_inode(dentry);
2198
2199 /* We can't create new files in nfs_open_revalidate(), so we
2200 * optimize away revalidation of negative dentries.
2201 */
2202 if (inode == NULL)
2203 goto full_reval;
2204
2205 if (nfs_verifier_is_delegated(dentry))
2206 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2207
2208 /* NFS only supports OPEN on regular files */
2209 if (!S_ISREG(inode->i_mode))
2210 goto full_reval;
2211
2212 /* We cannot do exclusive creation on a positive dentry */
2213 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2214 goto reval_dentry;
2215
2216 /* Check if the directory changed */
2217 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2218 goto reval_dentry;
2219
2220 /* Let f_op->open() actually open (and revalidate) the file */
2221 return 1;
2222reval_dentry:
2223 if (flags & LOOKUP_RCU)
2224 return -ECHILD;
2225 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
2226
2227full_reval:
2228 return nfs_do_lookup_revalidate(dir, dentry, flags);
2229}
2230
2231static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2232{
2233 return __nfs_lookup_revalidate(dentry, flags,
2234 nfs4_do_lookup_revalidate);
2235}
2236
2237#endif /* CONFIG_NFSV4 */
2238
2239struct dentry *
2240nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2241 struct nfs_fattr *fattr)
2242{
2243 struct dentry *parent = dget_parent(dentry);
2244 struct inode *dir = d_inode(parent);
2245 struct inode *inode;
2246 struct dentry *d;
2247 int error;
2248
2249 d_drop(dentry);
2250
2251 if (fhandle->size == 0) {
2252 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
2253 if (error)
2254 goto out_error;
2255 }
2256 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2257 if (!(fattr->valid & NFS_ATTR_FATTR)) {
2258 struct nfs_server *server = NFS_SB(dentry->d_sb);
2259 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2260 fattr, NULL);
2261 if (error < 0)
2262 goto out_error;
2263 }
2264 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2265 d = d_splice_alias(inode, dentry);
2266out:
2267 dput(parent);
2268 return d;
2269out_error:
2270 d = ERR_PTR(error);
2271 goto out;
2272}
2273EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2274
2275/*
2276 * Code common to create, mkdir, and mknod.
2277 */
2278int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2279 struct nfs_fattr *fattr)
2280{
2281 struct dentry *d;
2282
2283 d = nfs_add_or_obtain(dentry, fhandle, fattr);
2284 if (IS_ERR(d))
2285 return PTR_ERR(d);
2286
2287 /* Callers don't care */
2288 dput(d);
2289 return 0;
2290}
2291EXPORT_SYMBOL_GPL(nfs_instantiate);
2292
2293/*
2294 * Following a failed create operation, we drop the dentry rather
2295 * than retain a negative dentry. This avoids a problem in the event
2296 * that the operation succeeded on the server, but an error in the
2297 * reply path made it appear to have failed.
2298 */
2299int nfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2300 struct dentry *dentry, umode_t mode, bool excl)
2301{
2302 struct iattr attr;
2303 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2304 int error;
2305
2306 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2307 dir->i_sb->s_id, dir->i_ino, dentry);
2308
2309 attr.ia_mode = mode;
2310 attr.ia_valid = ATTR_MODE;
2311
2312 trace_nfs_create_enter(dir, dentry, open_flags);
2313 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2314 trace_nfs_create_exit(dir, dentry, open_flags, error);
2315 if (error != 0)
2316 goto out_err;
2317 return 0;
2318out_err:
2319 d_drop(dentry);
2320 return error;
2321}
2322EXPORT_SYMBOL_GPL(nfs_create);
2323
2324/*
2325 * See comments for nfs_proc_create regarding failed operations.
2326 */
2327int
2328nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2329 struct dentry *dentry, umode_t mode, dev_t rdev)
2330{
2331 struct iattr attr;
2332 int status;
2333
2334 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2335 dir->i_sb->s_id, dir->i_ino, dentry);
2336
2337 attr.ia_mode = mode;
2338 attr.ia_valid = ATTR_MODE;
2339
2340 trace_nfs_mknod_enter(dir, dentry);
2341 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2342 trace_nfs_mknod_exit(dir, dentry, status);
2343 if (status != 0)
2344 goto out_err;
2345 return 0;
2346out_err:
2347 d_drop(dentry);
2348 return status;
2349}
2350EXPORT_SYMBOL_GPL(nfs_mknod);
2351
2352/*
2353 * See comments for nfs_proc_create regarding failed operations.
2354 */
2355int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2356 struct dentry *dentry, umode_t mode)
2357{
2358 struct iattr attr;
2359 int error;
2360
2361 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2362 dir->i_sb->s_id, dir->i_ino, dentry);
2363
2364 attr.ia_valid = ATTR_MODE;
2365 attr.ia_mode = mode | S_IFDIR;
2366
2367 trace_nfs_mkdir_enter(dir, dentry);
2368 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2369 trace_nfs_mkdir_exit(dir, dentry, error);
2370 if (error != 0)
2371 goto out_err;
2372 return 0;
2373out_err:
2374 d_drop(dentry);
2375 return error;
2376}
2377EXPORT_SYMBOL_GPL(nfs_mkdir);
2378
2379static void nfs_dentry_handle_enoent(struct dentry *dentry)
2380{
2381 if (simple_positive(dentry))
2382 d_delete(dentry);
2383}
2384
2385static void nfs_dentry_remove_handle_error(struct inode *dir,
2386 struct dentry *dentry, int error)
2387{
2388 switch (error) {
2389 case -ENOENT:
2390 if (d_really_is_positive(dentry))
2391 d_delete(dentry);
2392 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2393 break;
2394 case 0:
2395 nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2396 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2397 }
2398}
2399
2400int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2401{
2402 int error;
2403
2404 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2405 dir->i_sb->s_id, dir->i_ino, dentry);
2406
2407 trace_nfs_rmdir_enter(dir, dentry);
2408 if (d_really_is_positive(dentry)) {
2409 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2410 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2411 /* Ensure the VFS deletes this inode */
2412 switch (error) {
2413 case 0:
2414 clear_nlink(d_inode(dentry));
2415 break;
2416 case -ENOENT:
2417 nfs_dentry_handle_enoent(dentry);
2418 }
2419 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2420 } else
2421 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2422 nfs_dentry_remove_handle_error(dir, dentry, error);
2423 trace_nfs_rmdir_exit(dir, dentry, error);
2424
2425 return error;
2426}
2427EXPORT_SYMBOL_GPL(nfs_rmdir);
2428
2429/*
2430 * Remove a file after making sure there are no pending writes,
2431 * and after checking that the file has only one user.
2432 *
2433 * We invalidate the attribute cache and free the inode prior to the operation
2434 * to avoid possible races if the server reuses the inode.
2435 */
2436static int nfs_safe_remove(struct dentry *dentry)
2437{
2438 struct inode *dir = d_inode(dentry->d_parent);
2439 struct inode *inode = d_inode(dentry);
2440 int error = -EBUSY;
2441
2442 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2443
2444 /* If the dentry was sillyrenamed, we simply call d_delete() */
2445 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2446 error = 0;
2447 goto out;
2448 }
2449
2450 trace_nfs_remove_enter(dir, dentry);
2451 if (inode != NULL) {
2452 error = NFS_PROTO(dir)->remove(dir, dentry);
2453 if (error == 0)
2454 nfs_drop_nlink(inode);
2455 } else
2456 error = NFS_PROTO(dir)->remove(dir, dentry);
2457 if (error == -ENOENT)
2458 nfs_dentry_handle_enoent(dentry);
2459 trace_nfs_remove_exit(dir, dentry, error);
2460out:
2461 return error;
2462}
2463
2464/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2465 * belongs to an active ".nfs..." file and we return -EBUSY.
2466 *
2467 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2468 */
2469int nfs_unlink(struct inode *dir, struct dentry *dentry)
2470{
2471 int error;
2472
2473 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2474 dir->i_ino, dentry);
2475
2476 trace_nfs_unlink_enter(dir, dentry);
2477 spin_lock(&dentry->d_lock);
2478 if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2479 &NFS_I(d_inode(dentry))->flags)) {
2480 spin_unlock(&dentry->d_lock);
2481 /* Start asynchronous writeout of the inode */
2482 write_inode_now(d_inode(dentry), 0);
2483 error = nfs_sillyrename(dir, dentry);
2484 goto out;
2485 }
2486 /* We must prevent any concurrent open until the unlink
2487 * completes. ->d_revalidate will wait for ->d_fsdata
2488 * to clear. We set it here to ensure no lookup succeeds until
2489 * the unlink is complete on the server.
2490 */
2491 error = -ETXTBSY;
2492 if (WARN_ON(dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2493 WARN_ON(dentry->d_fsdata == NFS_FSDATA_BLOCKED)) {
2494 spin_unlock(&dentry->d_lock);
2495 goto out;
2496 }
2497 /* old devname */
2498 kfree(dentry->d_fsdata);
2499 dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2500
2501 spin_unlock(&dentry->d_lock);
2502 error = nfs_safe_remove(dentry);
2503 nfs_dentry_remove_handle_error(dir, dentry, error);
2504 dentry->d_fsdata = NULL;
2505 wake_up_var(&dentry->d_fsdata);
2506out:
2507 trace_nfs_unlink_exit(dir, dentry, error);
2508 return error;
2509}
2510EXPORT_SYMBOL_GPL(nfs_unlink);
2511
2512/*
2513 * To create a symbolic link, most file systems instantiate a new inode,
2514 * add a page to it containing the path, then write it out to the disk
2515 * using prepare_write/commit_write.
2516 *
2517 * Unfortunately the NFS client can't create the in-core inode first
2518 * because it needs a file handle to create an in-core inode (see
2519 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2520 * symlink request has completed on the server.
2521 *
2522 * So instead we allocate a raw page, copy the symname into it, then do
2523 * the SYMLINK request with the page as the buffer. If it succeeds, we
2524 * now have a new file handle and can instantiate an in-core NFS inode
2525 * and move the raw page into its mapping.
2526 */
2527int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
2528 struct dentry *dentry, const char *symname)
2529{
2530 struct page *page;
2531 char *kaddr;
2532 struct iattr attr;
2533 unsigned int pathlen = strlen(symname);
2534 int error;
2535
2536 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2537 dir->i_ino, dentry, symname);
2538
2539 if (pathlen > PAGE_SIZE)
2540 return -ENAMETOOLONG;
2541
2542 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2543 attr.ia_valid = ATTR_MODE;
2544
2545 page = alloc_page(GFP_USER);
2546 if (!page)
2547 return -ENOMEM;
2548
2549 kaddr = page_address(page);
2550 memcpy(kaddr, symname, pathlen);
2551 if (pathlen < PAGE_SIZE)
2552 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2553
2554 trace_nfs_symlink_enter(dir, dentry);
2555 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2556 trace_nfs_symlink_exit(dir, dentry, error);
2557 if (error != 0) {
2558 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2559 dir->i_sb->s_id, dir->i_ino,
2560 dentry, symname, error);
2561 d_drop(dentry);
2562 __free_page(page);
2563 return error;
2564 }
2565
2566 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2567
2568 /*
2569 * No big deal if we can't add this page to the page cache here.
2570 * READLINK will get the missing page from the server if needed.
2571 */
2572 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2573 GFP_KERNEL)) {
2574 SetPageUptodate(page);
2575 unlock_page(page);
2576 /*
2577 * add_to_page_cache_lru() grabs an extra page refcount.
2578 * Drop it here to avoid leaking this page later.
2579 */
2580 put_page(page);
2581 } else
2582 __free_page(page);
2583
2584 return 0;
2585}
2586EXPORT_SYMBOL_GPL(nfs_symlink);
2587
2588int
2589nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2590{
2591 struct inode *inode = d_inode(old_dentry);
2592 int error;
2593
2594 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2595 old_dentry, dentry);
2596
2597 trace_nfs_link_enter(inode, dir, dentry);
2598 d_drop(dentry);
2599 if (S_ISREG(inode->i_mode))
2600 nfs_sync_inode(inode);
2601 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2602 if (error == 0) {
2603 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2604 ihold(inode);
2605 d_add(dentry, inode);
2606 }
2607 trace_nfs_link_exit(inode, dir, dentry, error);
2608 return error;
2609}
2610EXPORT_SYMBOL_GPL(nfs_link);
2611
2612static void
2613nfs_unblock_rename(struct rpc_task *task, struct nfs_renamedata *data)
2614{
2615 struct dentry *new_dentry = data->new_dentry;
2616
2617 new_dentry->d_fsdata = NULL;
2618 wake_up_var(&new_dentry->d_fsdata);
2619}
2620
2621/*
2622 * RENAME
2623 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2624 * different file handle for the same inode after a rename (e.g. when
2625 * moving to a different directory). A fail-safe method to do so would
2626 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2627 * rename the old file using the sillyrename stuff. This way, the original
2628 * file in old_dir will go away when the last process iput()s the inode.
2629 *
2630 * FIXED.
2631 *
2632 * It actually works quite well. One needs to have the possibility for
2633 * at least one ".nfs..." file in each directory the file ever gets
2634 * moved or linked to which happens automagically with the new
2635 * implementation that only depends on the dcache stuff instead of
2636 * using the inode layer
2637 *
2638 * Unfortunately, things are a little more complicated than indicated
2639 * above. For a cross-directory move, we want to make sure we can get
2640 * rid of the old inode after the operation. This means there must be
2641 * no pending writes (if it's a file), and the use count must be 1.
2642 * If these conditions are met, we can drop the dentries before doing
2643 * the rename.
2644 */
2645int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
2646 struct dentry *old_dentry, struct inode *new_dir,
2647 struct dentry *new_dentry, unsigned int flags)
2648{
2649 struct inode *old_inode = d_inode(old_dentry);
2650 struct inode *new_inode = d_inode(new_dentry);
2651 struct dentry *dentry = NULL;
2652 struct rpc_task *task;
2653 bool must_unblock = false;
2654 int error = -EBUSY;
2655
2656 if (flags)
2657 return -EINVAL;
2658
2659 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2660 old_dentry, new_dentry,
2661 d_count(new_dentry));
2662
2663 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2664 /*
2665 * For non-directories, check whether the target is busy and if so,
2666 * make a copy of the dentry and then do a silly-rename. If the
2667 * silly-rename succeeds, the copied dentry is hashed and becomes
2668 * the new target.
2669 */
2670 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2671 /* We must prevent any concurrent open until the unlink
2672 * completes. ->d_revalidate will wait for ->d_fsdata
2673 * to clear. We set it here to ensure no lookup succeeds until
2674 * the unlink is complete on the server.
2675 */
2676 error = -ETXTBSY;
2677 if (WARN_ON(new_dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2678 WARN_ON(new_dentry->d_fsdata == NFS_FSDATA_BLOCKED))
2679 goto out;
2680 if (new_dentry->d_fsdata) {
2681 /* old devname */
2682 kfree(new_dentry->d_fsdata);
2683 new_dentry->d_fsdata = NULL;
2684 }
2685
2686 spin_lock(&new_dentry->d_lock);
2687 if (d_count(new_dentry) > 2) {
2688 int err;
2689
2690 spin_unlock(&new_dentry->d_lock);
2691
2692 /* copy the target dentry's name */
2693 dentry = d_alloc(new_dentry->d_parent,
2694 &new_dentry->d_name);
2695 if (!dentry)
2696 goto out;
2697
2698 /* silly-rename the existing target ... */
2699 err = nfs_sillyrename(new_dir, new_dentry);
2700 if (err)
2701 goto out;
2702
2703 new_dentry = dentry;
2704 new_inode = NULL;
2705 } else {
2706 new_dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2707 must_unblock = true;
2708 spin_unlock(&new_dentry->d_lock);
2709 }
2710
2711 }
2712
2713 if (S_ISREG(old_inode->i_mode))
2714 nfs_sync_inode(old_inode);
2715 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry,
2716 must_unblock ? nfs_unblock_rename : NULL);
2717 if (IS_ERR(task)) {
2718 error = PTR_ERR(task);
2719 goto out;
2720 }
2721
2722 error = rpc_wait_for_completion_task(task);
2723 if (error != 0) {
2724 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2725 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2726 smp_wmb();
2727 } else
2728 error = task->tk_status;
2729 rpc_put_task(task);
2730 /* Ensure the inode attributes are revalidated */
2731 if (error == 0) {
2732 spin_lock(&old_inode->i_lock);
2733 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2734 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2735 NFS_INO_INVALID_CTIME |
2736 NFS_INO_REVAL_FORCED);
2737 spin_unlock(&old_inode->i_lock);
2738 }
2739out:
2740 trace_nfs_rename_exit(old_dir, old_dentry,
2741 new_dir, new_dentry, error);
2742 if (!error) {
2743 if (new_inode != NULL)
2744 nfs_drop_nlink(new_inode);
2745 /*
2746 * The d_move() should be here instead of in an async RPC completion
2747 * handler because we need the proper locks to move the dentry. If
2748 * we're interrupted by a signal, the async RPC completion handler
2749 * should mark the directories for revalidation.
2750 */
2751 d_move(old_dentry, new_dentry);
2752 nfs_set_verifier(old_dentry,
2753 nfs_save_change_attribute(new_dir));
2754 } else if (error == -ENOENT)
2755 nfs_dentry_handle_enoent(old_dentry);
2756
2757 /* new dentry created? */
2758 if (dentry)
2759 dput(dentry);
2760 return error;
2761}
2762EXPORT_SYMBOL_GPL(nfs_rename);
2763
2764static DEFINE_SPINLOCK(nfs_access_lru_lock);
2765static LIST_HEAD(nfs_access_lru_list);
2766static atomic_long_t nfs_access_nr_entries;
2767
2768static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2769module_param(nfs_access_max_cachesize, ulong, 0644);
2770MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2771
2772static void nfs_access_free_entry(struct nfs_access_entry *entry)
2773{
2774 put_group_info(entry->group_info);
2775 kfree_rcu(entry, rcu_head);
2776 smp_mb__before_atomic();
2777 atomic_long_dec(&nfs_access_nr_entries);
2778 smp_mb__after_atomic();
2779}
2780
2781static void nfs_access_free_list(struct list_head *head)
2782{
2783 struct nfs_access_entry *cache;
2784
2785 while (!list_empty(head)) {
2786 cache = list_entry(head->next, struct nfs_access_entry, lru);
2787 list_del(&cache->lru);
2788 nfs_access_free_entry(cache);
2789 }
2790}
2791
2792static unsigned long
2793nfs_do_access_cache_scan(unsigned int nr_to_scan)
2794{
2795 LIST_HEAD(head);
2796 struct nfs_inode *nfsi, *next;
2797 struct nfs_access_entry *cache;
2798 long freed = 0;
2799
2800 spin_lock(&nfs_access_lru_lock);
2801 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2802 struct inode *inode;
2803
2804 if (nr_to_scan-- == 0)
2805 break;
2806 inode = &nfsi->vfs_inode;
2807 spin_lock(&inode->i_lock);
2808 if (list_empty(&nfsi->access_cache_entry_lru))
2809 goto remove_lru_entry;
2810 cache = list_entry(nfsi->access_cache_entry_lru.next,
2811 struct nfs_access_entry, lru);
2812 list_move(&cache->lru, &head);
2813 rb_erase(&cache->rb_node, &nfsi->access_cache);
2814 freed++;
2815 if (!list_empty(&nfsi->access_cache_entry_lru))
2816 list_move_tail(&nfsi->access_cache_inode_lru,
2817 &nfs_access_lru_list);
2818 else {
2819remove_lru_entry:
2820 list_del_init(&nfsi->access_cache_inode_lru);
2821 smp_mb__before_atomic();
2822 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2823 smp_mb__after_atomic();
2824 }
2825 spin_unlock(&inode->i_lock);
2826 }
2827 spin_unlock(&nfs_access_lru_lock);
2828 nfs_access_free_list(&head);
2829 return freed;
2830}
2831
2832unsigned long
2833nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2834{
2835 int nr_to_scan = sc->nr_to_scan;
2836 gfp_t gfp_mask = sc->gfp_mask;
2837
2838 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2839 return SHRINK_STOP;
2840 return nfs_do_access_cache_scan(nr_to_scan);
2841}
2842
2843
2844unsigned long
2845nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2846{
2847 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2848}
2849
2850static void
2851nfs_access_cache_enforce_limit(void)
2852{
2853 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2854 unsigned long diff;
2855 unsigned int nr_to_scan;
2856
2857 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2858 return;
2859 nr_to_scan = 100;
2860 diff = nr_entries - nfs_access_max_cachesize;
2861 if (diff < nr_to_scan)
2862 nr_to_scan = diff;
2863 nfs_do_access_cache_scan(nr_to_scan);
2864}
2865
2866static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2867{
2868 struct rb_root *root_node = &nfsi->access_cache;
2869 struct rb_node *n;
2870 struct nfs_access_entry *entry;
2871
2872 /* Unhook entries from the cache */
2873 while ((n = rb_first(root_node)) != NULL) {
2874 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2875 rb_erase(n, root_node);
2876 list_move(&entry->lru, head);
2877 }
2878 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2879}
2880
2881void nfs_access_zap_cache(struct inode *inode)
2882{
2883 LIST_HEAD(head);
2884
2885 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2886 return;
2887 /* Remove from global LRU init */
2888 spin_lock(&nfs_access_lru_lock);
2889 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2890 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2891
2892 spin_lock(&inode->i_lock);
2893 __nfs_access_zap_cache(NFS_I(inode), &head);
2894 spin_unlock(&inode->i_lock);
2895 spin_unlock(&nfs_access_lru_lock);
2896 nfs_access_free_list(&head);
2897}
2898EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2899
2900static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2901{
2902 struct group_info *ga, *gb;
2903 int g;
2904
2905 if (uid_lt(a->fsuid, b->fsuid))
2906 return -1;
2907 if (uid_gt(a->fsuid, b->fsuid))
2908 return 1;
2909
2910 if (gid_lt(a->fsgid, b->fsgid))
2911 return -1;
2912 if (gid_gt(a->fsgid, b->fsgid))
2913 return 1;
2914
2915 ga = a->group_info;
2916 gb = b->group_info;
2917 if (ga == gb)
2918 return 0;
2919 if (ga == NULL)
2920 return -1;
2921 if (gb == NULL)
2922 return 1;
2923 if (ga->ngroups < gb->ngroups)
2924 return -1;
2925 if (ga->ngroups > gb->ngroups)
2926 return 1;
2927
2928 for (g = 0; g < ga->ngroups; g++) {
2929 if (gid_lt(ga->gid[g], gb->gid[g]))
2930 return -1;
2931 if (gid_gt(ga->gid[g], gb->gid[g]))
2932 return 1;
2933 }
2934 return 0;
2935}
2936
2937static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2938{
2939 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2940
2941 while (n != NULL) {
2942 struct nfs_access_entry *entry =
2943 rb_entry(n, struct nfs_access_entry, rb_node);
2944 int cmp = access_cmp(cred, entry);
2945
2946 if (cmp < 0)
2947 n = n->rb_left;
2948 else if (cmp > 0)
2949 n = n->rb_right;
2950 else
2951 return entry;
2952 }
2953 return NULL;
2954}
2955
2956static u64 nfs_access_login_time(const struct task_struct *task,
2957 const struct cred *cred)
2958{
2959 const struct task_struct *parent;
2960 const struct cred *pcred;
2961 u64 ret;
2962
2963 rcu_read_lock();
2964 for (;;) {
2965 parent = rcu_dereference(task->real_parent);
2966 pcred = rcu_dereference(parent->cred);
2967 if (parent == task || cred_fscmp(pcred, cred) != 0)
2968 break;
2969 task = parent;
2970 }
2971 ret = task->start_time;
2972 rcu_read_unlock();
2973 return ret;
2974}
2975
2976static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
2977{
2978 struct nfs_inode *nfsi = NFS_I(inode);
2979 u64 login_time = nfs_access_login_time(current, cred);
2980 struct nfs_access_entry *cache;
2981 bool retry = true;
2982 int err;
2983
2984 spin_lock(&inode->i_lock);
2985 for(;;) {
2986 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2987 goto out_zap;
2988 cache = nfs_access_search_rbtree(inode, cred);
2989 err = -ENOENT;
2990 if (cache == NULL)
2991 goto out;
2992 /* Found an entry, is our attribute cache valid? */
2993 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2994 break;
2995 if (!retry)
2996 break;
2997 err = -ECHILD;
2998 if (!may_block)
2999 goto out;
3000 spin_unlock(&inode->i_lock);
3001 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
3002 if (err)
3003 return err;
3004 spin_lock(&inode->i_lock);
3005 retry = false;
3006 }
3007 err = -ENOENT;
3008 if ((s64)(login_time - cache->timestamp) > 0)
3009 goto out;
3010 *mask = cache->mask;
3011 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
3012 err = 0;
3013out:
3014 spin_unlock(&inode->i_lock);
3015 return err;
3016out_zap:
3017 spin_unlock(&inode->i_lock);
3018 nfs_access_zap_cache(inode);
3019 return -ENOENT;
3020}
3021
3022static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
3023{
3024 /* Only check the most recently returned cache entry,
3025 * but do it without locking.
3026 */
3027 struct nfs_inode *nfsi = NFS_I(inode);
3028 u64 login_time = nfs_access_login_time(current, cred);
3029 struct nfs_access_entry *cache;
3030 int err = -ECHILD;
3031 struct list_head *lh;
3032
3033 rcu_read_lock();
3034 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3035 goto out;
3036 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
3037 cache = list_entry(lh, struct nfs_access_entry, lru);
3038 if (lh == &nfsi->access_cache_entry_lru ||
3039 access_cmp(cred, cache) != 0)
3040 cache = NULL;
3041 if (cache == NULL)
3042 goto out;
3043 if ((s64)(login_time - cache->timestamp) > 0)
3044 goto out;
3045 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3046 goto out;
3047 *mask = cache->mask;
3048 err = 0;
3049out:
3050 rcu_read_unlock();
3051 return err;
3052}
3053
3054int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
3055 u32 *mask, bool may_block)
3056{
3057 int status;
3058
3059 status = nfs_access_get_cached_rcu(inode, cred, mask);
3060 if (status != 0)
3061 status = nfs_access_get_cached_locked(inode, cred, mask,
3062 may_block);
3063
3064 return status;
3065}
3066EXPORT_SYMBOL_GPL(nfs_access_get_cached);
3067
3068static void nfs_access_add_rbtree(struct inode *inode,
3069 struct nfs_access_entry *set,
3070 const struct cred *cred)
3071{
3072 struct nfs_inode *nfsi = NFS_I(inode);
3073 struct rb_root *root_node = &nfsi->access_cache;
3074 struct rb_node **p = &root_node->rb_node;
3075 struct rb_node *parent = NULL;
3076 struct nfs_access_entry *entry;
3077 int cmp;
3078
3079 spin_lock(&inode->i_lock);
3080 while (*p != NULL) {
3081 parent = *p;
3082 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3083 cmp = access_cmp(cred, entry);
3084
3085 if (cmp < 0)
3086 p = &parent->rb_left;
3087 else if (cmp > 0)
3088 p = &parent->rb_right;
3089 else
3090 goto found;
3091 }
3092 set->timestamp = ktime_get_ns();
3093 rb_link_node(&set->rb_node, parent, p);
3094 rb_insert_color(&set->rb_node, root_node);
3095 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3096 spin_unlock(&inode->i_lock);
3097 return;
3098found:
3099 rb_replace_node(parent, &set->rb_node, root_node);
3100 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3101 list_del(&entry->lru);
3102 spin_unlock(&inode->i_lock);
3103 nfs_access_free_entry(entry);
3104}
3105
3106void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3107 const struct cred *cred)
3108{
3109 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
3110 if (cache == NULL)
3111 return;
3112 RB_CLEAR_NODE(&cache->rb_node);
3113 cache->fsuid = cred->fsuid;
3114 cache->fsgid = cred->fsgid;
3115 cache->group_info = get_group_info(cred->group_info);
3116 cache->mask = set->mask;
3117
3118 /* The above field assignments must be visible
3119 * before this item appears on the lru. We cannot easily
3120 * use rcu_assign_pointer, so just force the memory barrier.
3121 */
3122 smp_wmb();
3123 nfs_access_add_rbtree(inode, cache, cred);
3124
3125 /* Update accounting */
3126 smp_mb__before_atomic();
3127 atomic_long_inc(&nfs_access_nr_entries);
3128 smp_mb__after_atomic();
3129
3130 /* Add inode to global LRU list */
3131 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3132 spin_lock(&nfs_access_lru_lock);
3133 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
3134 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
3135 &nfs_access_lru_list);
3136 spin_unlock(&nfs_access_lru_lock);
3137 }
3138 nfs_access_cache_enforce_limit();
3139}
3140EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3141
3142#define NFS_MAY_READ (NFS_ACCESS_READ)
3143#define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3144 NFS_ACCESS_EXTEND | \
3145 NFS_ACCESS_DELETE)
3146#define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3147 NFS_ACCESS_EXTEND)
3148#define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3149#define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3150#define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3151static int
3152nfs_access_calc_mask(u32 access_result, umode_t umode)
3153{
3154 int mask = 0;
3155
3156 if (access_result & NFS_MAY_READ)
3157 mask |= MAY_READ;
3158 if (S_ISDIR(umode)) {
3159 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3160 mask |= MAY_WRITE;
3161 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3162 mask |= MAY_EXEC;
3163 } else if (S_ISREG(umode)) {
3164 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3165 mask |= MAY_WRITE;
3166 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3167 mask |= MAY_EXEC;
3168 } else if (access_result & NFS_MAY_WRITE)
3169 mask |= MAY_WRITE;
3170 return mask;
3171}
3172
3173void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3174{
3175 entry->mask = access_result;
3176}
3177EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3178
3179static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3180{
3181 struct nfs_access_entry cache;
3182 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3183 int cache_mask = -1;
3184 int status;
3185
3186 trace_nfs_access_enter(inode);
3187
3188 status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3189 if (status == 0)
3190 goto out_cached;
3191
3192 status = -ECHILD;
3193 if (!may_block)
3194 goto out;
3195
3196 /*
3197 * Determine which access bits we want to ask for...
3198 */
3199 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3200 nfs_access_xattr_mask(NFS_SERVER(inode));
3201 if (S_ISDIR(inode->i_mode))
3202 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3203 else
3204 cache.mask |= NFS_ACCESS_EXECUTE;
3205 status = NFS_PROTO(inode)->access(inode, &cache, cred);
3206 if (status != 0) {
3207 if (status == -ESTALE) {
3208 if (!S_ISDIR(inode->i_mode))
3209 nfs_set_inode_stale(inode);
3210 else
3211 nfs_zap_caches(inode);
3212 }
3213 goto out;
3214 }
3215 nfs_access_add_cache(inode, &cache, cred);
3216out_cached:
3217 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
3218 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3219 status = -EACCES;
3220out:
3221 trace_nfs_access_exit(inode, mask, cache_mask, status);
3222 return status;
3223}
3224
3225static int nfs_open_permission_mask(int openflags)
3226{
3227 int mask = 0;
3228
3229 if (openflags & __FMODE_EXEC) {
3230 /* ONLY check exec rights */
3231 mask = MAY_EXEC;
3232 } else {
3233 if ((openflags & O_ACCMODE) != O_WRONLY)
3234 mask |= MAY_READ;
3235 if ((openflags & O_ACCMODE) != O_RDONLY)
3236 mask |= MAY_WRITE;
3237 }
3238
3239 return mask;
3240}
3241
3242int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3243{
3244 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
3245}
3246EXPORT_SYMBOL_GPL(nfs_may_open);
3247
3248static int nfs_execute_ok(struct inode *inode, int mask)
3249{
3250 struct nfs_server *server = NFS_SERVER(inode);
3251 int ret = 0;
3252
3253 if (S_ISDIR(inode->i_mode))
3254 return 0;
3255 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3256 if (mask & MAY_NOT_BLOCK)
3257 return -ECHILD;
3258 ret = __nfs_revalidate_inode(server, inode);
3259 }
3260 if (ret == 0 && !execute_ok(inode))
3261 ret = -EACCES;
3262 return ret;
3263}
3264
3265int nfs_permission(struct user_namespace *mnt_userns,
3266 struct inode *inode,
3267 int mask)
3268{
3269 const struct cred *cred = current_cred();
3270 int res = 0;
3271
3272 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
3273
3274 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3275 goto out;
3276 /* Is this sys_access() ? */
3277 if (mask & (MAY_ACCESS | MAY_CHDIR))
3278 goto force_lookup;
3279
3280 switch (inode->i_mode & S_IFMT) {
3281 case S_IFLNK:
3282 goto out;
3283 case S_IFREG:
3284 if ((mask & MAY_OPEN) &&
3285 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3286 return 0;
3287 break;
3288 case S_IFDIR:
3289 /*
3290 * Optimize away all write operations, since the server
3291 * will check permissions when we perform the op.
3292 */
3293 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3294 goto out;
3295 }
3296
3297force_lookup:
3298 if (!NFS_PROTO(inode)->access)
3299 goto out_notsup;
3300
3301 res = nfs_do_access(inode, cred, mask);
3302out:
3303 if (!res && (mask & MAY_EXEC))
3304 res = nfs_execute_ok(inode, mask);
3305
3306 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3307 inode->i_sb->s_id, inode->i_ino, mask, res);
3308 return res;
3309out_notsup:
3310 if (mask & MAY_NOT_BLOCK)
3311 return -ECHILD;
3312
3313 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3314 NFS_INO_INVALID_OTHER);
3315 if (res == 0)
3316 res = generic_permission(&init_user_ns, inode, mask);
3317 goto out;
3318}
3319EXPORT_SYMBOL_GPL(nfs_permission);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * linux/fs/nfs/dir.c
4 *
5 * Copyright (C) 1992 Rick Sladkey
6 *
7 * nfs directory handling functions
8 *
9 * 10 Apr 1996 Added silly rename for unlink --okir
10 * 28 Sep 1996 Improved directory cache --okir
11 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
12 * Re-implemented silly rename for unlink, newly implemented
13 * silly rename for nfs_rename() following the suggestions
14 * of Olaf Kirch (okir) found in this file.
15 * Following Linus comments on my original hack, this version
16 * depends only on the dcache stuff and doesn't touch the inode
17 * layer (iput() and friends).
18 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
19 */
20
21#include <linux/module.h>
22#include <linux/time.h>
23#include <linux/errno.h>
24#include <linux/stat.h>
25#include <linux/fcntl.h>
26#include <linux/string.h>
27#include <linux/kernel.h>
28#include <linux/slab.h>
29#include <linux/mm.h>
30#include <linux/sunrpc/clnt.h>
31#include <linux/nfs_fs.h>
32#include <linux/nfs_mount.h>
33#include <linux/pagemap.h>
34#include <linux/pagevec.h>
35#include <linux/namei.h>
36#include <linux/mount.h>
37#include <linux/swap.h>
38#include <linux/sched.h>
39#include <linux/kmemleak.h>
40#include <linux/xattr.h>
41
42#include "delegation.h"
43#include "iostat.h"
44#include "internal.h"
45#include "fscache.h"
46
47#include "nfstrace.h"
48
49/* #define NFS_DEBUG_VERBOSE 1 */
50
51static int nfs_opendir(struct inode *, struct file *);
52static int nfs_closedir(struct inode *, struct file *);
53static int nfs_readdir(struct file *, struct dir_context *);
54static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
55static loff_t nfs_llseek_dir(struct file *, loff_t, int);
56static void nfs_readdir_clear_array(struct page*);
57
58const struct file_operations nfs_dir_operations = {
59 .llseek = nfs_llseek_dir,
60 .read = generic_read_dir,
61 .iterate_shared = nfs_readdir,
62 .open = nfs_opendir,
63 .release = nfs_closedir,
64 .fsync = nfs_fsync_dir,
65};
66
67const struct address_space_operations nfs_dir_aops = {
68 .freepage = nfs_readdir_clear_array,
69};
70
71static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir)
72{
73 struct nfs_inode *nfsi = NFS_I(dir);
74 struct nfs_open_dir_context *ctx;
75 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
76 if (ctx != NULL) {
77 ctx->duped = 0;
78 ctx->attr_gencount = nfsi->attr_gencount;
79 ctx->dir_cookie = 0;
80 ctx->dup_cookie = 0;
81 spin_lock(&dir->i_lock);
82 if (list_empty(&nfsi->open_files) &&
83 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
84 nfs_set_cache_invalid(dir,
85 NFS_INO_INVALID_DATA |
86 NFS_INO_REVAL_FORCED);
87 list_add(&ctx->list, &nfsi->open_files);
88 spin_unlock(&dir->i_lock);
89 return ctx;
90 }
91 return ERR_PTR(-ENOMEM);
92}
93
94static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
95{
96 spin_lock(&dir->i_lock);
97 list_del(&ctx->list);
98 spin_unlock(&dir->i_lock);
99 kfree(ctx);
100}
101
102/*
103 * Open file
104 */
105static int
106nfs_opendir(struct inode *inode, struct file *filp)
107{
108 int res = 0;
109 struct nfs_open_dir_context *ctx;
110
111 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
112
113 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
114
115 ctx = alloc_nfs_open_dir_context(inode);
116 if (IS_ERR(ctx)) {
117 res = PTR_ERR(ctx);
118 goto out;
119 }
120 filp->private_data = ctx;
121out:
122 return res;
123}
124
125static int
126nfs_closedir(struct inode *inode, struct file *filp)
127{
128 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
129 return 0;
130}
131
132struct nfs_cache_array_entry {
133 u64 cookie;
134 u64 ino;
135 const char *name;
136 unsigned int name_len;
137 unsigned char d_type;
138};
139
140struct nfs_cache_array {
141 u64 last_cookie;
142 unsigned int size;
143 unsigned char page_full : 1,
144 page_is_eof : 1,
145 cookies_are_ordered : 1;
146 struct nfs_cache_array_entry array[];
147};
148
149struct nfs_readdir_descriptor {
150 struct file *file;
151 struct page *page;
152 struct dir_context *ctx;
153 pgoff_t page_index;
154 u64 dir_cookie;
155 u64 last_cookie;
156 u64 dup_cookie;
157 loff_t current_index;
158 loff_t prev_index;
159
160 __be32 verf[NFS_DIR_VERIFIER_SIZE];
161 unsigned long dir_verifier;
162 unsigned long timestamp;
163 unsigned long gencount;
164 unsigned long attr_gencount;
165 unsigned int cache_entry_index;
166 signed char duped;
167 bool plus;
168 bool eof;
169};
170
171static void nfs_readdir_array_init(struct nfs_cache_array *array)
172{
173 memset(array, 0, sizeof(struct nfs_cache_array));
174}
175
176static void nfs_readdir_page_init_array(struct page *page, u64 last_cookie)
177{
178 struct nfs_cache_array *array;
179
180 array = kmap_atomic(page);
181 nfs_readdir_array_init(array);
182 array->last_cookie = last_cookie;
183 array->cookies_are_ordered = 1;
184 kunmap_atomic(array);
185}
186
187/*
188 * we are freeing strings created by nfs_add_to_readdir_array()
189 */
190static
191void nfs_readdir_clear_array(struct page *page)
192{
193 struct nfs_cache_array *array;
194 int i;
195
196 array = kmap_atomic(page);
197 for (i = 0; i < array->size; i++)
198 kfree(array->array[i].name);
199 nfs_readdir_array_init(array);
200 kunmap_atomic(array);
201}
202
203static struct page *
204nfs_readdir_page_array_alloc(u64 last_cookie, gfp_t gfp_flags)
205{
206 struct page *page = alloc_page(gfp_flags);
207 if (page)
208 nfs_readdir_page_init_array(page, last_cookie);
209 return page;
210}
211
212static void nfs_readdir_page_array_free(struct page *page)
213{
214 if (page) {
215 nfs_readdir_clear_array(page);
216 put_page(page);
217 }
218}
219
220static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
221{
222 array->page_is_eof = 1;
223 array->page_full = 1;
224}
225
226static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
227{
228 return array->page_full;
229}
230
231/*
232 * the caller is responsible for freeing qstr.name
233 * when called by nfs_readdir_add_to_array, the strings will be freed in
234 * nfs_clear_readdir_array()
235 */
236static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
237{
238 const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
239
240 /*
241 * Avoid a kmemleak false positive. The pointer to the name is stored
242 * in a page cache page which kmemleak does not scan.
243 */
244 if (ret != NULL)
245 kmemleak_not_leak(ret);
246 return ret;
247}
248
249/*
250 * Check that the next array entry lies entirely within the page bounds
251 */
252static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
253{
254 struct nfs_cache_array_entry *cache_entry;
255
256 if (array->page_full)
257 return -ENOSPC;
258 cache_entry = &array->array[array->size + 1];
259 if ((char *)cache_entry - (char *)array > PAGE_SIZE) {
260 array->page_full = 1;
261 return -ENOSPC;
262 }
263 return 0;
264}
265
266static
267int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
268{
269 struct nfs_cache_array *array;
270 struct nfs_cache_array_entry *cache_entry;
271 const char *name;
272 int ret;
273
274 name = nfs_readdir_copy_name(entry->name, entry->len);
275 if (!name)
276 return -ENOMEM;
277
278 array = kmap_atomic(page);
279 ret = nfs_readdir_array_can_expand(array);
280 if (ret) {
281 kfree(name);
282 goto out;
283 }
284
285 cache_entry = &array->array[array->size];
286 cache_entry->cookie = entry->prev_cookie;
287 cache_entry->ino = entry->ino;
288 cache_entry->d_type = entry->d_type;
289 cache_entry->name_len = entry->len;
290 cache_entry->name = name;
291 array->last_cookie = entry->cookie;
292 if (array->last_cookie <= cache_entry->cookie)
293 array->cookies_are_ordered = 0;
294 array->size++;
295 if (entry->eof != 0)
296 nfs_readdir_array_set_eof(array);
297out:
298 kunmap_atomic(array);
299 return ret;
300}
301
302static struct page *nfs_readdir_page_get_locked(struct address_space *mapping,
303 pgoff_t index, u64 last_cookie)
304{
305 struct page *page;
306
307 page = grab_cache_page(mapping, index);
308 if (page && !PageUptodate(page)) {
309 nfs_readdir_page_init_array(page, last_cookie);
310 if (invalidate_inode_pages2_range(mapping, index + 1, -1) < 0)
311 nfs_zap_mapping(mapping->host, mapping);
312 SetPageUptodate(page);
313 }
314
315 return page;
316}
317
318static u64 nfs_readdir_page_last_cookie(struct page *page)
319{
320 struct nfs_cache_array *array;
321 u64 ret;
322
323 array = kmap_atomic(page);
324 ret = array->last_cookie;
325 kunmap_atomic(array);
326 return ret;
327}
328
329static bool nfs_readdir_page_needs_filling(struct page *page)
330{
331 struct nfs_cache_array *array;
332 bool ret;
333
334 array = kmap_atomic(page);
335 ret = !nfs_readdir_array_is_full(array);
336 kunmap_atomic(array);
337 return ret;
338}
339
340static void nfs_readdir_page_set_eof(struct page *page)
341{
342 struct nfs_cache_array *array;
343
344 array = kmap_atomic(page);
345 nfs_readdir_array_set_eof(array);
346 kunmap_atomic(array);
347}
348
349static void nfs_readdir_page_unlock_and_put(struct page *page)
350{
351 unlock_page(page);
352 put_page(page);
353}
354
355static struct page *nfs_readdir_page_get_next(struct address_space *mapping,
356 pgoff_t index, u64 cookie)
357{
358 struct page *page;
359
360 page = nfs_readdir_page_get_locked(mapping, index, cookie);
361 if (page) {
362 if (nfs_readdir_page_last_cookie(page) == cookie)
363 return page;
364 nfs_readdir_page_unlock_and_put(page);
365 }
366 return NULL;
367}
368
369static inline
370int is_32bit_api(void)
371{
372#ifdef CONFIG_COMPAT
373 return in_compat_syscall();
374#else
375 return (BITS_PER_LONG == 32);
376#endif
377}
378
379static
380bool nfs_readdir_use_cookie(const struct file *filp)
381{
382 if ((filp->f_mode & FMODE_32BITHASH) ||
383 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
384 return false;
385 return true;
386}
387
388static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
389 struct nfs_readdir_descriptor *desc)
390{
391 loff_t diff = desc->ctx->pos - desc->current_index;
392 unsigned int index;
393
394 if (diff < 0)
395 goto out_eof;
396 if (diff >= array->size) {
397 if (array->page_is_eof)
398 goto out_eof;
399 return -EAGAIN;
400 }
401
402 index = (unsigned int)diff;
403 desc->dir_cookie = array->array[index].cookie;
404 desc->cache_entry_index = index;
405 return 0;
406out_eof:
407 desc->eof = true;
408 return -EBADCOOKIE;
409}
410
411static bool
412nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
413{
414 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
415 return false;
416 smp_rmb();
417 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
418}
419
420static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
421 u64 cookie)
422{
423 if (!array->cookies_are_ordered)
424 return true;
425 /* Optimisation for monotonically increasing cookies */
426 if (cookie >= array->last_cookie)
427 return false;
428 if (array->size && cookie < array->array[0].cookie)
429 return false;
430 return true;
431}
432
433static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
434 struct nfs_readdir_descriptor *desc)
435{
436 int i;
437 loff_t new_pos;
438 int status = -EAGAIN;
439
440 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
441 goto check_eof;
442
443 for (i = 0; i < array->size; i++) {
444 if (array->array[i].cookie == desc->dir_cookie) {
445 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
446
447 new_pos = desc->current_index + i;
448 if (desc->attr_gencount != nfsi->attr_gencount ||
449 !nfs_readdir_inode_mapping_valid(nfsi)) {
450 desc->duped = 0;
451 desc->attr_gencount = nfsi->attr_gencount;
452 } else if (new_pos < desc->prev_index) {
453 if (desc->duped > 0
454 && desc->dup_cookie == desc->dir_cookie) {
455 if (printk_ratelimit()) {
456 pr_notice("NFS: directory %pD2 contains a readdir loop."
457 "Please contact your server vendor. "
458 "The file: %s has duplicate cookie %llu\n",
459 desc->file, array->array[i].name, desc->dir_cookie);
460 }
461 status = -ELOOP;
462 goto out;
463 }
464 desc->dup_cookie = desc->dir_cookie;
465 desc->duped = -1;
466 }
467 if (nfs_readdir_use_cookie(desc->file))
468 desc->ctx->pos = desc->dir_cookie;
469 else
470 desc->ctx->pos = new_pos;
471 desc->prev_index = new_pos;
472 desc->cache_entry_index = i;
473 return 0;
474 }
475 }
476check_eof:
477 if (array->page_is_eof) {
478 status = -EBADCOOKIE;
479 if (desc->dir_cookie == array->last_cookie)
480 desc->eof = true;
481 }
482out:
483 return status;
484}
485
486static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
487{
488 struct nfs_cache_array *array;
489 int status;
490
491 array = kmap_atomic(desc->page);
492
493 if (desc->dir_cookie == 0)
494 status = nfs_readdir_search_for_pos(array, desc);
495 else
496 status = nfs_readdir_search_for_cookie(array, desc);
497
498 if (status == -EAGAIN) {
499 desc->last_cookie = array->last_cookie;
500 desc->current_index += array->size;
501 desc->page_index++;
502 }
503 kunmap_atomic(array);
504 return status;
505}
506
507/* Fill a page with xdr information before transferring to the cache page */
508static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
509 __be32 *verf, u64 cookie,
510 struct page **pages, size_t bufsize,
511 __be32 *verf_res)
512{
513 struct inode *inode = file_inode(desc->file);
514 struct nfs_readdir_arg arg = {
515 .dentry = file_dentry(desc->file),
516 .cred = desc->file->f_cred,
517 .verf = verf,
518 .cookie = cookie,
519 .pages = pages,
520 .page_len = bufsize,
521 .plus = desc->plus,
522 };
523 struct nfs_readdir_res res = {
524 .verf = verf_res,
525 };
526 unsigned long timestamp, gencount;
527 int error;
528
529 again:
530 timestamp = jiffies;
531 gencount = nfs_inc_attr_generation_counter();
532 desc->dir_verifier = nfs_save_change_attribute(inode);
533 error = NFS_PROTO(inode)->readdir(&arg, &res);
534 if (error < 0) {
535 /* We requested READDIRPLUS, but the server doesn't grok it */
536 if (error == -ENOTSUPP && desc->plus) {
537 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
538 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
539 desc->plus = arg.plus = false;
540 goto again;
541 }
542 goto error;
543 }
544 desc->timestamp = timestamp;
545 desc->gencount = gencount;
546error:
547 return error;
548}
549
550static int xdr_decode(struct nfs_readdir_descriptor *desc,
551 struct nfs_entry *entry, struct xdr_stream *xdr)
552{
553 struct inode *inode = file_inode(desc->file);
554 int error;
555
556 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
557 if (error)
558 return error;
559 entry->fattr->time_start = desc->timestamp;
560 entry->fattr->gencount = desc->gencount;
561 return 0;
562}
563
564/* Match file and dirent using either filehandle or fileid
565 * Note: caller is responsible for checking the fsid
566 */
567static
568int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
569{
570 struct inode *inode;
571 struct nfs_inode *nfsi;
572
573 if (d_really_is_negative(dentry))
574 return 0;
575
576 inode = d_inode(dentry);
577 if (is_bad_inode(inode) || NFS_STALE(inode))
578 return 0;
579
580 nfsi = NFS_I(inode);
581 if (entry->fattr->fileid != nfsi->fileid)
582 return 0;
583 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
584 return 0;
585 return 1;
586}
587
588static
589bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
590{
591 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
592 return false;
593 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
594 return true;
595 if (ctx->pos == 0)
596 return true;
597 return false;
598}
599
600/*
601 * This function is called by the lookup and getattr code to request the
602 * use of readdirplus to accelerate any future lookups in the same
603 * directory.
604 */
605void nfs_advise_use_readdirplus(struct inode *dir)
606{
607 struct nfs_inode *nfsi = NFS_I(dir);
608
609 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
610 !list_empty(&nfsi->open_files))
611 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
612}
613
614/*
615 * This function is mainly for use by nfs_getattr().
616 *
617 * If this is an 'ls -l', we want to force use of readdirplus.
618 * Do this by checking if there is an active file descriptor
619 * and calling nfs_advise_use_readdirplus, then forcing a
620 * cache flush.
621 */
622void nfs_force_use_readdirplus(struct inode *dir)
623{
624 struct nfs_inode *nfsi = NFS_I(dir);
625
626 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
627 !list_empty(&nfsi->open_files)) {
628 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
629 invalidate_mapping_pages(dir->i_mapping,
630 nfsi->page_index + 1, -1);
631 }
632}
633
634static
635void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
636 unsigned long dir_verifier)
637{
638 struct qstr filename = QSTR_INIT(entry->name, entry->len);
639 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
640 struct dentry *dentry;
641 struct dentry *alias;
642 struct inode *inode;
643 int status;
644
645 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
646 return;
647 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
648 return;
649 if (filename.len == 0)
650 return;
651 /* Validate that the name doesn't contain any illegal '\0' */
652 if (strnlen(filename.name, filename.len) != filename.len)
653 return;
654 /* ...or '/' */
655 if (strnchr(filename.name, filename.len, '/'))
656 return;
657 if (filename.name[0] == '.') {
658 if (filename.len == 1)
659 return;
660 if (filename.len == 2 && filename.name[1] == '.')
661 return;
662 }
663 filename.hash = full_name_hash(parent, filename.name, filename.len);
664
665 dentry = d_lookup(parent, &filename);
666again:
667 if (!dentry) {
668 dentry = d_alloc_parallel(parent, &filename, &wq);
669 if (IS_ERR(dentry))
670 return;
671 }
672 if (!d_in_lookup(dentry)) {
673 /* Is there a mountpoint here? If so, just exit */
674 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
675 &entry->fattr->fsid))
676 goto out;
677 if (nfs_same_file(dentry, entry)) {
678 if (!entry->fh->size)
679 goto out;
680 nfs_set_verifier(dentry, dir_verifier);
681 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
682 if (!status)
683 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
684 goto out;
685 } else {
686 d_invalidate(dentry);
687 dput(dentry);
688 dentry = NULL;
689 goto again;
690 }
691 }
692 if (!entry->fh->size) {
693 d_lookup_done(dentry);
694 goto out;
695 }
696
697 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
698 alias = d_splice_alias(inode, dentry);
699 d_lookup_done(dentry);
700 if (alias) {
701 if (IS_ERR(alias))
702 goto out;
703 dput(dentry);
704 dentry = alias;
705 }
706 nfs_set_verifier(dentry, dir_verifier);
707out:
708 dput(dentry);
709}
710
711/* Perform conversion from xdr to cache array */
712static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc,
713 struct nfs_entry *entry,
714 struct page **xdr_pages,
715 unsigned int buflen,
716 struct page **arrays,
717 size_t narrays)
718{
719 struct address_space *mapping = desc->file->f_mapping;
720 struct xdr_stream stream;
721 struct xdr_buf buf;
722 struct page *scratch, *new, *page = *arrays;
723 int status;
724
725 scratch = alloc_page(GFP_KERNEL);
726 if (scratch == NULL)
727 return -ENOMEM;
728
729 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
730 xdr_set_scratch_page(&stream, scratch);
731
732 do {
733 if (entry->label)
734 entry->label->len = NFS4_MAXLABELLEN;
735
736 status = xdr_decode(desc, entry, &stream);
737 if (status != 0)
738 break;
739
740 if (desc->plus)
741 nfs_prime_dcache(file_dentry(desc->file), entry,
742 desc->dir_verifier);
743
744 status = nfs_readdir_add_to_array(entry, page);
745 if (status != -ENOSPC)
746 continue;
747
748 if (page->mapping != mapping) {
749 if (!--narrays)
750 break;
751 new = nfs_readdir_page_array_alloc(entry->prev_cookie,
752 GFP_KERNEL);
753 if (!new)
754 break;
755 arrays++;
756 *arrays = page = new;
757 } else {
758 new = nfs_readdir_page_get_next(mapping,
759 page->index + 1,
760 entry->prev_cookie);
761 if (!new)
762 break;
763 if (page != *arrays)
764 nfs_readdir_page_unlock_and_put(page);
765 page = new;
766 }
767 status = nfs_readdir_add_to_array(entry, page);
768 } while (!status && !entry->eof);
769
770 switch (status) {
771 case -EBADCOOKIE:
772 if (entry->eof) {
773 nfs_readdir_page_set_eof(page);
774 status = 0;
775 }
776 break;
777 case -ENOSPC:
778 case -EAGAIN:
779 status = 0;
780 break;
781 }
782
783 if (page != *arrays)
784 nfs_readdir_page_unlock_and_put(page);
785
786 put_page(scratch);
787 return status;
788}
789
790static void nfs_readdir_free_pages(struct page **pages, size_t npages)
791{
792 while (npages--)
793 put_page(pages[npages]);
794 kfree(pages);
795}
796
797/*
798 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
799 * to nfs_readdir_free_pages()
800 */
801static struct page **nfs_readdir_alloc_pages(size_t npages)
802{
803 struct page **pages;
804 size_t i;
805
806 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
807 if (!pages)
808 return NULL;
809 for (i = 0; i < npages; i++) {
810 struct page *page = alloc_page(GFP_KERNEL);
811 if (page == NULL)
812 goto out_freepages;
813 pages[i] = page;
814 }
815 return pages;
816
817out_freepages:
818 nfs_readdir_free_pages(pages, i);
819 return NULL;
820}
821
822static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
823 __be32 *verf_arg, __be32 *verf_res,
824 struct page **arrays, size_t narrays)
825{
826 struct page **pages;
827 struct page *page = *arrays;
828 struct nfs_entry *entry;
829 size_t array_size;
830 struct inode *inode = file_inode(desc->file);
831 size_t dtsize = NFS_SERVER(inode)->dtsize;
832 int status = -ENOMEM;
833
834 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
835 if (!entry)
836 return -ENOMEM;
837 entry->cookie = nfs_readdir_page_last_cookie(page);
838 entry->fh = nfs_alloc_fhandle();
839 entry->fattr = nfs_alloc_fattr();
840 entry->server = NFS_SERVER(inode);
841 if (entry->fh == NULL || entry->fattr == NULL)
842 goto out;
843
844 entry->label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
845 if (IS_ERR(entry->label)) {
846 status = PTR_ERR(entry->label);
847 goto out;
848 }
849
850 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
851 pages = nfs_readdir_alloc_pages(array_size);
852 if (!pages)
853 goto out_release_label;
854
855 do {
856 unsigned int pglen;
857 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie,
858 pages, dtsize,
859 verf_res);
860 if (status < 0)
861 break;
862
863 pglen = status;
864 if (pglen == 0) {
865 nfs_readdir_page_set_eof(page);
866 break;
867 }
868
869 verf_arg = verf_res;
870
871 status = nfs_readdir_page_filler(desc, entry, pages, pglen,
872 arrays, narrays);
873 } while (!status && nfs_readdir_page_needs_filling(page));
874
875 nfs_readdir_free_pages(pages, array_size);
876out_release_label:
877 nfs4_label_free(entry->label);
878out:
879 nfs_free_fattr(entry->fattr);
880 nfs_free_fhandle(entry->fh);
881 kfree(entry);
882 return status;
883}
884
885static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc)
886{
887 put_page(desc->page);
888 desc->page = NULL;
889}
890
891static void
892nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
893{
894 unlock_page(desc->page);
895 nfs_readdir_page_put(desc);
896}
897
898static struct page *
899nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc)
900{
901 return nfs_readdir_page_get_locked(desc->file->f_mapping,
902 desc->page_index,
903 desc->last_cookie);
904}
905
906/*
907 * Returns 0 if desc->dir_cookie was found on page desc->page_index
908 * and locks the page to prevent removal from the page cache.
909 */
910static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
911{
912 struct inode *inode = file_inode(desc->file);
913 struct nfs_inode *nfsi = NFS_I(inode);
914 __be32 verf[NFS_DIR_VERIFIER_SIZE];
915 int res;
916
917 desc->page = nfs_readdir_page_get_cached(desc);
918 if (!desc->page)
919 return -ENOMEM;
920 if (nfs_readdir_page_needs_filling(desc->page)) {
921 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
922 &desc->page, 1);
923 if (res < 0) {
924 nfs_readdir_page_unlock_and_put_cached(desc);
925 if (res == -EBADCOOKIE || res == -ENOTSYNC) {
926 invalidate_inode_pages2(desc->file->f_mapping);
927 desc->page_index = 0;
928 return -EAGAIN;
929 }
930 return res;
931 }
932 /*
933 * Set the cookie verifier if the page cache was empty
934 */
935 if (desc->page_index == 0)
936 memcpy(nfsi->cookieverf, verf,
937 sizeof(nfsi->cookieverf));
938 }
939 res = nfs_readdir_search_array(desc);
940 if (res == 0) {
941 nfsi->page_index = desc->page_index;
942 return 0;
943 }
944 nfs_readdir_page_unlock_and_put_cached(desc);
945 return res;
946}
947
948static bool nfs_readdir_dont_search_cache(struct nfs_readdir_descriptor *desc)
949{
950 struct address_space *mapping = desc->file->f_mapping;
951 struct inode *dir = file_inode(desc->file);
952 unsigned int dtsize = NFS_SERVER(dir)->dtsize;
953 loff_t size = i_size_read(dir);
954
955 /*
956 * Default to uncached readdir if the page cache is empty, and
957 * we're looking for a non-zero cookie in a large directory.
958 */
959 return desc->dir_cookie != 0 && mapping->nrpages == 0 && size > dtsize;
960}
961
962/* Search for desc->dir_cookie from the beginning of the page cache */
963static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
964{
965 int res;
966
967 if (nfs_readdir_dont_search_cache(desc))
968 return -EBADCOOKIE;
969
970 do {
971 if (desc->page_index == 0) {
972 desc->current_index = 0;
973 desc->prev_index = 0;
974 desc->last_cookie = 0;
975 }
976 res = find_and_lock_cache_page(desc);
977 } while (res == -EAGAIN);
978 return res;
979}
980
981/*
982 * Once we've found the start of the dirent within a page: fill 'er up...
983 */
984static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
985 const __be32 *verf)
986{
987 struct file *file = desc->file;
988 struct nfs_cache_array *array;
989 unsigned int i = 0;
990
991 array = kmap(desc->page);
992 for (i = desc->cache_entry_index; i < array->size; i++) {
993 struct nfs_cache_array_entry *ent;
994
995 ent = &array->array[i];
996 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
997 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
998 desc->eof = true;
999 break;
1000 }
1001 memcpy(desc->verf, verf, sizeof(desc->verf));
1002 if (i < (array->size-1))
1003 desc->dir_cookie = array->array[i+1].cookie;
1004 else
1005 desc->dir_cookie = array->last_cookie;
1006 if (nfs_readdir_use_cookie(file))
1007 desc->ctx->pos = desc->dir_cookie;
1008 else
1009 desc->ctx->pos++;
1010 if (desc->duped != 0)
1011 desc->duped = 1;
1012 }
1013 if (array->page_is_eof)
1014 desc->eof = true;
1015
1016 kunmap(desc->page);
1017 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1018 (unsigned long long)desc->dir_cookie);
1019}
1020
1021/*
1022 * If we cannot find a cookie in our cache, we suspect that this is
1023 * because it points to a deleted file, so we ask the server to return
1024 * whatever it thinks is the next entry. We then feed this to filldir.
1025 * If all goes well, we should then be able to find our way round the
1026 * cache on the next call to readdir_search_pagecache();
1027 *
1028 * NOTE: we cannot add the anonymous page to the pagecache because
1029 * the data it contains might not be page aligned. Besides,
1030 * we should already have a complete representation of the
1031 * directory in the page cache by the time we get here.
1032 */
1033static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1034{
1035 struct page **arrays;
1036 size_t i, sz = 512;
1037 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1038 int status = -ENOMEM;
1039
1040 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1041 (unsigned long long)desc->dir_cookie);
1042
1043 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1044 if (!arrays)
1045 goto out;
1046 arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL);
1047 if (!arrays[0])
1048 goto out;
1049
1050 desc->page_index = 0;
1051 desc->last_cookie = desc->dir_cookie;
1052 desc->duped = 0;
1053
1054 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1055
1056 for (i = 0; !desc->eof && i < sz && arrays[i]; i++) {
1057 desc->page = arrays[i];
1058 nfs_do_filldir(desc, verf);
1059 }
1060 desc->page = NULL;
1061
1062
1063 for (i = 0; i < sz && arrays[i]; i++)
1064 nfs_readdir_page_array_free(arrays[i]);
1065out:
1066 kfree(arrays);
1067 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1068 return status;
1069}
1070
1071/* The file offset position represents the dirent entry number. A
1072 last cookie cache takes care of the common case of reading the
1073 whole directory.
1074 */
1075static int nfs_readdir(struct file *file, struct dir_context *ctx)
1076{
1077 struct dentry *dentry = file_dentry(file);
1078 struct inode *inode = d_inode(dentry);
1079 struct nfs_inode *nfsi = NFS_I(inode);
1080 struct nfs_open_dir_context *dir_ctx = file->private_data;
1081 struct nfs_readdir_descriptor *desc;
1082 int res;
1083
1084 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1085 file, (long long)ctx->pos);
1086 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1087
1088 /*
1089 * ctx->pos points to the dirent entry number.
1090 * *desc->dir_cookie has the cookie for the next entry. We have
1091 * to either find the entry with the appropriate number or
1092 * revalidate the cookie.
1093 */
1094 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) {
1095 res = nfs_revalidate_mapping(inode, file->f_mapping);
1096 if (res < 0)
1097 goto out;
1098 }
1099
1100 res = -ENOMEM;
1101 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1102 if (!desc)
1103 goto out;
1104 desc->file = file;
1105 desc->ctx = ctx;
1106 desc->plus = nfs_use_readdirplus(inode, ctx);
1107
1108 spin_lock(&file->f_lock);
1109 desc->dir_cookie = dir_ctx->dir_cookie;
1110 desc->dup_cookie = dir_ctx->dup_cookie;
1111 desc->duped = dir_ctx->duped;
1112 desc->attr_gencount = dir_ctx->attr_gencount;
1113 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1114 spin_unlock(&file->f_lock);
1115
1116 do {
1117 res = readdir_search_pagecache(desc);
1118
1119 if (res == -EBADCOOKIE) {
1120 res = 0;
1121 /* This means either end of directory */
1122 if (desc->dir_cookie && !desc->eof) {
1123 /* Or that the server has 'lost' a cookie */
1124 res = uncached_readdir(desc);
1125 if (res == 0)
1126 continue;
1127 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1128 res = 0;
1129 }
1130 break;
1131 }
1132 if (res == -ETOOSMALL && desc->plus) {
1133 clear_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
1134 nfs_zap_caches(inode);
1135 desc->page_index = 0;
1136 desc->plus = false;
1137 desc->eof = false;
1138 continue;
1139 }
1140 if (res < 0)
1141 break;
1142
1143 nfs_do_filldir(desc, nfsi->cookieverf);
1144 nfs_readdir_page_unlock_and_put_cached(desc);
1145 } while (!desc->eof);
1146
1147 spin_lock(&file->f_lock);
1148 dir_ctx->dir_cookie = desc->dir_cookie;
1149 dir_ctx->dup_cookie = desc->dup_cookie;
1150 dir_ctx->duped = desc->duped;
1151 dir_ctx->attr_gencount = desc->attr_gencount;
1152 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1153 spin_unlock(&file->f_lock);
1154
1155 kfree(desc);
1156
1157out:
1158 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1159 return res;
1160}
1161
1162static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1163{
1164 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1165
1166 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1167 filp, offset, whence);
1168
1169 switch (whence) {
1170 default:
1171 return -EINVAL;
1172 case SEEK_SET:
1173 if (offset < 0)
1174 return -EINVAL;
1175 spin_lock(&filp->f_lock);
1176 break;
1177 case SEEK_CUR:
1178 if (offset == 0)
1179 return filp->f_pos;
1180 spin_lock(&filp->f_lock);
1181 offset += filp->f_pos;
1182 if (offset < 0) {
1183 spin_unlock(&filp->f_lock);
1184 return -EINVAL;
1185 }
1186 }
1187 if (offset != filp->f_pos) {
1188 filp->f_pos = offset;
1189 if (nfs_readdir_use_cookie(filp))
1190 dir_ctx->dir_cookie = offset;
1191 else
1192 dir_ctx->dir_cookie = 0;
1193 if (offset == 0)
1194 memset(dir_ctx->verf, 0, sizeof(dir_ctx->verf));
1195 dir_ctx->duped = 0;
1196 }
1197 spin_unlock(&filp->f_lock);
1198 return offset;
1199}
1200
1201/*
1202 * All directory operations under NFS are synchronous, so fsync()
1203 * is a dummy operation.
1204 */
1205static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1206 int datasync)
1207{
1208 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1209
1210 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1211 return 0;
1212}
1213
1214/**
1215 * nfs_force_lookup_revalidate - Mark the directory as having changed
1216 * @dir: pointer to directory inode
1217 *
1218 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1219 * full lookup on all child dentries of 'dir' whenever a change occurs
1220 * on the server that might have invalidated our dcache.
1221 *
1222 * Note that we reserve bit '0' as a tag to let us know when a dentry
1223 * was revalidated while holding a delegation on its inode.
1224 *
1225 * The caller should be holding dir->i_lock
1226 */
1227void nfs_force_lookup_revalidate(struct inode *dir)
1228{
1229 NFS_I(dir)->cache_change_attribute += 2;
1230}
1231EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1232
1233/**
1234 * nfs_verify_change_attribute - Detects NFS remote directory changes
1235 * @dir: pointer to parent directory inode
1236 * @verf: previously saved change attribute
1237 *
1238 * Return "false" if the verifiers doesn't match the change attribute.
1239 * This would usually indicate that the directory contents have changed on
1240 * the server, and that any dentries need revalidating.
1241 */
1242static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1243{
1244 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1245}
1246
1247static void nfs_set_verifier_delegated(unsigned long *verf)
1248{
1249 *verf |= 1UL;
1250}
1251
1252#if IS_ENABLED(CONFIG_NFS_V4)
1253static void nfs_unset_verifier_delegated(unsigned long *verf)
1254{
1255 *verf &= ~1UL;
1256}
1257#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1258
1259static bool nfs_test_verifier_delegated(unsigned long verf)
1260{
1261 return verf & 1;
1262}
1263
1264static bool nfs_verifier_is_delegated(struct dentry *dentry)
1265{
1266 return nfs_test_verifier_delegated(dentry->d_time);
1267}
1268
1269static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1270{
1271 struct inode *inode = d_inode(dentry);
1272
1273 if (!nfs_verifier_is_delegated(dentry) &&
1274 !nfs_verify_change_attribute(d_inode(dentry->d_parent), verf))
1275 goto out;
1276 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1277 nfs_set_verifier_delegated(&verf);
1278out:
1279 dentry->d_time = verf;
1280}
1281
1282/**
1283 * nfs_set_verifier - save a parent directory verifier in the dentry
1284 * @dentry: pointer to dentry
1285 * @verf: verifier to save
1286 *
1287 * Saves the parent directory verifier in @dentry. If the inode has
1288 * a delegation, we also tag the dentry as having been revalidated
1289 * while holding a delegation so that we know we don't have to
1290 * look it up again after a directory change.
1291 */
1292void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1293{
1294
1295 spin_lock(&dentry->d_lock);
1296 nfs_set_verifier_locked(dentry, verf);
1297 spin_unlock(&dentry->d_lock);
1298}
1299EXPORT_SYMBOL_GPL(nfs_set_verifier);
1300
1301#if IS_ENABLED(CONFIG_NFS_V4)
1302/**
1303 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1304 * @inode: pointer to inode
1305 *
1306 * Iterates through the dentries in the inode alias list and clears
1307 * the tag used to indicate that the dentry has been revalidated
1308 * while holding a delegation.
1309 * This function is intended for use when the delegation is being
1310 * returned or revoked.
1311 */
1312void nfs_clear_verifier_delegated(struct inode *inode)
1313{
1314 struct dentry *alias;
1315
1316 if (!inode)
1317 return;
1318 spin_lock(&inode->i_lock);
1319 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1320 spin_lock(&alias->d_lock);
1321 nfs_unset_verifier_delegated(&alias->d_time);
1322 spin_unlock(&alias->d_lock);
1323 }
1324 spin_unlock(&inode->i_lock);
1325}
1326EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1327#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1328
1329/*
1330 * A check for whether or not the parent directory has changed.
1331 * In the case it has, we assume that the dentries are untrustworthy
1332 * and may need to be looked up again.
1333 * If rcu_walk prevents us from performing a full check, return 0.
1334 */
1335static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1336 int rcu_walk)
1337{
1338 if (IS_ROOT(dentry))
1339 return 1;
1340 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1341 return 0;
1342 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1343 return 0;
1344 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1345 if (nfs_mapping_need_revalidate_inode(dir)) {
1346 if (rcu_walk)
1347 return 0;
1348 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1349 return 0;
1350 }
1351 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1352 return 0;
1353 return 1;
1354}
1355
1356/*
1357 * Use intent information to check whether or not we're going to do
1358 * an O_EXCL create using this path component.
1359 */
1360static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1361{
1362 if (NFS_PROTO(dir)->version == 2)
1363 return 0;
1364 return flags & LOOKUP_EXCL;
1365}
1366
1367/*
1368 * Inode and filehandle revalidation for lookups.
1369 *
1370 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1371 * or if the intent information indicates that we're about to open this
1372 * particular file and the "nocto" mount flag is not set.
1373 *
1374 */
1375static
1376int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1377{
1378 struct nfs_server *server = NFS_SERVER(inode);
1379 int ret;
1380
1381 if (IS_AUTOMOUNT(inode))
1382 return 0;
1383
1384 if (flags & LOOKUP_OPEN) {
1385 switch (inode->i_mode & S_IFMT) {
1386 case S_IFREG:
1387 /* A NFSv4 OPEN will revalidate later */
1388 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1389 goto out;
1390 fallthrough;
1391 case S_IFDIR:
1392 if (server->flags & NFS_MOUNT_NOCTO)
1393 break;
1394 /* NFS close-to-open cache consistency validation */
1395 goto out_force;
1396 }
1397 }
1398
1399 /* VFS wants an on-the-wire revalidation */
1400 if (flags & LOOKUP_REVAL)
1401 goto out_force;
1402out:
1403 return (inode->i_nlink == 0) ? -ESTALE : 0;
1404out_force:
1405 if (flags & LOOKUP_RCU)
1406 return -ECHILD;
1407 ret = __nfs_revalidate_inode(server, inode);
1408 if (ret != 0)
1409 return ret;
1410 goto out;
1411}
1412
1413static void nfs_mark_dir_for_revalidate(struct inode *inode)
1414{
1415 spin_lock(&inode->i_lock);
1416 nfs_set_cache_invalid(inode, NFS_INO_REVAL_PAGECACHE);
1417 spin_unlock(&inode->i_lock);
1418}
1419
1420/*
1421 * We judge how long we want to trust negative
1422 * dentries by looking at the parent inode mtime.
1423 *
1424 * If parent mtime has changed, we revalidate, else we wait for a
1425 * period corresponding to the parent's attribute cache timeout value.
1426 *
1427 * If LOOKUP_RCU prevents us from performing a full check, return 1
1428 * suggesting a reval is needed.
1429 *
1430 * Note that when creating a new file, or looking up a rename target,
1431 * then it shouldn't be necessary to revalidate a negative dentry.
1432 */
1433static inline
1434int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1435 unsigned int flags)
1436{
1437 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1438 return 0;
1439 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1440 return 1;
1441 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1442}
1443
1444static int
1445nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1446 struct inode *inode, int error)
1447{
1448 switch (error) {
1449 case 1:
1450 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1451 __func__, dentry);
1452 return 1;
1453 case 0:
1454 /*
1455 * We can't d_drop the root of a disconnected tree:
1456 * its d_hash is on the s_anon list and d_drop() would hide
1457 * it from shrink_dcache_for_unmount(), leading to busy
1458 * inodes on unmount and further oopses.
1459 */
1460 if (inode && IS_ROOT(dentry))
1461 return 1;
1462 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1463 __func__, dentry);
1464 return 0;
1465 }
1466 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1467 __func__, dentry, error);
1468 return error;
1469}
1470
1471static int
1472nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1473 unsigned int flags)
1474{
1475 int ret = 1;
1476 if (nfs_neg_need_reval(dir, dentry, flags)) {
1477 if (flags & LOOKUP_RCU)
1478 return -ECHILD;
1479 ret = 0;
1480 }
1481 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1482}
1483
1484static int
1485nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1486 struct inode *inode)
1487{
1488 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1489 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1490}
1491
1492static int
1493nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry,
1494 struct inode *inode)
1495{
1496 struct nfs_fh *fhandle;
1497 struct nfs_fattr *fattr;
1498 struct nfs4_label *label;
1499 unsigned long dir_verifier;
1500 int ret;
1501
1502 ret = -ENOMEM;
1503 fhandle = nfs_alloc_fhandle();
1504 fattr = nfs_alloc_fattr();
1505 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1506 if (fhandle == NULL || fattr == NULL || IS_ERR(label))
1507 goto out;
1508
1509 dir_verifier = nfs_save_change_attribute(dir);
1510 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1511 if (ret < 0) {
1512 switch (ret) {
1513 case -ESTALE:
1514 case -ENOENT:
1515 ret = 0;
1516 break;
1517 case -ETIMEDOUT:
1518 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1519 ret = 1;
1520 }
1521 goto out;
1522 }
1523 ret = 0;
1524 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1525 goto out;
1526 if (nfs_refresh_inode(inode, fattr) < 0)
1527 goto out;
1528
1529 nfs_setsecurity(inode, fattr, label);
1530 nfs_set_verifier(dentry, dir_verifier);
1531
1532 /* set a readdirplus hint that we had a cache miss */
1533 nfs_force_use_readdirplus(dir);
1534 ret = 1;
1535out:
1536 nfs_free_fattr(fattr);
1537 nfs_free_fhandle(fhandle);
1538 nfs4_label_free(label);
1539
1540 /*
1541 * If the lookup failed despite the dentry change attribute being
1542 * a match, then we should revalidate the directory cache.
1543 */
1544 if (!ret && nfs_verify_change_attribute(dir, dentry->d_time))
1545 nfs_mark_dir_for_revalidate(dir);
1546 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1547}
1548
1549/*
1550 * This is called every time the dcache has a lookup hit,
1551 * and we should check whether we can really trust that
1552 * lookup.
1553 *
1554 * NOTE! The hit can be a negative hit too, don't assume
1555 * we have an inode!
1556 *
1557 * If the parent directory is seen to have changed, we throw out the
1558 * cached dentry and do a new lookup.
1559 */
1560static int
1561nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1562 unsigned int flags)
1563{
1564 struct inode *inode;
1565 int error;
1566
1567 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1568 inode = d_inode(dentry);
1569
1570 if (!inode)
1571 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1572
1573 if (is_bad_inode(inode)) {
1574 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1575 __func__, dentry);
1576 goto out_bad;
1577 }
1578
1579 if (nfs_verifier_is_delegated(dentry))
1580 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1581
1582 /* Force a full look up iff the parent directory has changed */
1583 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1584 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1585 error = nfs_lookup_verify_inode(inode, flags);
1586 if (error) {
1587 if (error == -ESTALE)
1588 nfs_mark_dir_for_revalidate(dir);
1589 goto out_bad;
1590 }
1591 nfs_advise_use_readdirplus(dir);
1592 goto out_valid;
1593 }
1594
1595 if (flags & LOOKUP_RCU)
1596 return -ECHILD;
1597
1598 if (NFS_STALE(inode))
1599 goto out_bad;
1600
1601 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1602 error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1603 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1604 return error;
1605out_valid:
1606 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1607out_bad:
1608 if (flags & LOOKUP_RCU)
1609 return -ECHILD;
1610 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1611}
1612
1613static int
1614__nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1615 int (*reval)(struct inode *, struct dentry *, unsigned int))
1616{
1617 struct dentry *parent;
1618 struct inode *dir;
1619 int ret;
1620
1621 if (flags & LOOKUP_RCU) {
1622 parent = READ_ONCE(dentry->d_parent);
1623 dir = d_inode_rcu(parent);
1624 if (!dir)
1625 return -ECHILD;
1626 ret = reval(dir, dentry, flags);
1627 if (parent != READ_ONCE(dentry->d_parent))
1628 return -ECHILD;
1629 } else {
1630 parent = dget_parent(dentry);
1631 ret = reval(d_inode(parent), dentry, flags);
1632 dput(parent);
1633 }
1634 return ret;
1635}
1636
1637static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1638{
1639 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1640}
1641
1642/*
1643 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1644 * when we don't really care about the dentry name. This is called when a
1645 * pathwalk ends on a dentry that was not found via a normal lookup in the
1646 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1647 *
1648 * In this situation, we just want to verify that the inode itself is OK
1649 * since the dentry might have changed on the server.
1650 */
1651static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1652{
1653 struct inode *inode = d_inode(dentry);
1654 int error = 0;
1655
1656 /*
1657 * I believe we can only get a negative dentry here in the case of a
1658 * procfs-style symlink. Just assume it's correct for now, but we may
1659 * eventually need to do something more here.
1660 */
1661 if (!inode) {
1662 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1663 __func__, dentry);
1664 return 1;
1665 }
1666
1667 if (is_bad_inode(inode)) {
1668 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1669 __func__, dentry);
1670 return 0;
1671 }
1672
1673 error = nfs_lookup_verify_inode(inode, flags);
1674 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1675 __func__, inode->i_ino, error ? "invalid" : "valid");
1676 return !error;
1677}
1678
1679/*
1680 * This is called from dput() when d_count is going to 0.
1681 */
1682static int nfs_dentry_delete(const struct dentry *dentry)
1683{
1684 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1685 dentry, dentry->d_flags);
1686
1687 /* Unhash any dentry with a stale inode */
1688 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1689 return 1;
1690
1691 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1692 /* Unhash it, so that ->d_iput() would be called */
1693 return 1;
1694 }
1695 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1696 /* Unhash it, so that ancestors of killed async unlink
1697 * files will be cleaned up during umount */
1698 return 1;
1699 }
1700 return 0;
1701
1702}
1703
1704/* Ensure that we revalidate inode->i_nlink */
1705static void nfs_drop_nlink(struct inode *inode)
1706{
1707 spin_lock(&inode->i_lock);
1708 /* drop the inode if we're reasonably sure this is the last link */
1709 if (inode->i_nlink > 0)
1710 drop_nlink(inode);
1711 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1712 nfs_set_cache_invalid(
1713 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1714 NFS_INO_INVALID_NLINK);
1715 spin_unlock(&inode->i_lock);
1716}
1717
1718/*
1719 * Called when the dentry loses inode.
1720 * We use it to clean up silly-renamed files.
1721 */
1722static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1723{
1724 if (S_ISDIR(inode->i_mode))
1725 /* drop any readdir cache as it could easily be old */
1726 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1727
1728 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1729 nfs_complete_unlink(dentry, inode);
1730 nfs_drop_nlink(inode);
1731 }
1732 iput(inode);
1733}
1734
1735static void nfs_d_release(struct dentry *dentry)
1736{
1737 /* free cached devname value, if it survived that far */
1738 if (unlikely(dentry->d_fsdata)) {
1739 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1740 WARN_ON(1);
1741 else
1742 kfree(dentry->d_fsdata);
1743 }
1744}
1745
1746const struct dentry_operations nfs_dentry_operations = {
1747 .d_revalidate = nfs_lookup_revalidate,
1748 .d_weak_revalidate = nfs_weak_revalidate,
1749 .d_delete = nfs_dentry_delete,
1750 .d_iput = nfs_dentry_iput,
1751 .d_automount = nfs_d_automount,
1752 .d_release = nfs_d_release,
1753};
1754EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1755
1756struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1757{
1758 struct dentry *res;
1759 struct inode *inode = NULL;
1760 struct nfs_fh *fhandle = NULL;
1761 struct nfs_fattr *fattr = NULL;
1762 struct nfs4_label *label = NULL;
1763 unsigned long dir_verifier;
1764 int error;
1765
1766 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1767 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1768
1769 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1770 return ERR_PTR(-ENAMETOOLONG);
1771
1772 /*
1773 * If we're doing an exclusive create, optimize away the lookup
1774 * but don't hash the dentry.
1775 */
1776 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1777 return NULL;
1778
1779 res = ERR_PTR(-ENOMEM);
1780 fhandle = nfs_alloc_fhandle();
1781 fattr = nfs_alloc_fattr();
1782 if (fhandle == NULL || fattr == NULL)
1783 goto out;
1784
1785 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1786 if (IS_ERR(label))
1787 goto out;
1788
1789 dir_verifier = nfs_save_change_attribute(dir);
1790 trace_nfs_lookup_enter(dir, dentry, flags);
1791 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1792 if (error == -ENOENT)
1793 goto no_entry;
1794 if (error < 0) {
1795 res = ERR_PTR(error);
1796 goto out_label;
1797 }
1798 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1799 res = ERR_CAST(inode);
1800 if (IS_ERR(res))
1801 goto out_label;
1802
1803 /* Notify readdir to use READDIRPLUS */
1804 nfs_force_use_readdirplus(dir);
1805
1806no_entry:
1807 res = d_splice_alias(inode, dentry);
1808 if (res != NULL) {
1809 if (IS_ERR(res))
1810 goto out_label;
1811 dentry = res;
1812 }
1813 nfs_set_verifier(dentry, dir_verifier);
1814out_label:
1815 trace_nfs_lookup_exit(dir, dentry, flags, error);
1816 nfs4_label_free(label);
1817out:
1818 nfs_free_fattr(fattr);
1819 nfs_free_fhandle(fhandle);
1820 return res;
1821}
1822EXPORT_SYMBOL_GPL(nfs_lookup);
1823
1824#if IS_ENABLED(CONFIG_NFS_V4)
1825static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1826
1827const struct dentry_operations nfs4_dentry_operations = {
1828 .d_revalidate = nfs4_lookup_revalidate,
1829 .d_weak_revalidate = nfs_weak_revalidate,
1830 .d_delete = nfs_dentry_delete,
1831 .d_iput = nfs_dentry_iput,
1832 .d_automount = nfs_d_automount,
1833 .d_release = nfs_d_release,
1834};
1835EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1836
1837static fmode_t flags_to_mode(int flags)
1838{
1839 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1840 if ((flags & O_ACCMODE) != O_WRONLY)
1841 res |= FMODE_READ;
1842 if ((flags & O_ACCMODE) != O_RDONLY)
1843 res |= FMODE_WRITE;
1844 return res;
1845}
1846
1847static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1848{
1849 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1850}
1851
1852static int do_open(struct inode *inode, struct file *filp)
1853{
1854 nfs_fscache_open_file(inode, filp);
1855 return 0;
1856}
1857
1858static int nfs_finish_open(struct nfs_open_context *ctx,
1859 struct dentry *dentry,
1860 struct file *file, unsigned open_flags)
1861{
1862 int err;
1863
1864 err = finish_open(file, dentry, do_open);
1865 if (err)
1866 goto out;
1867 if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1868 nfs_file_set_open_context(file, ctx);
1869 else
1870 err = -EOPENSTALE;
1871out:
1872 return err;
1873}
1874
1875int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1876 struct file *file, unsigned open_flags,
1877 umode_t mode)
1878{
1879 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1880 struct nfs_open_context *ctx;
1881 struct dentry *res;
1882 struct iattr attr = { .ia_valid = ATTR_OPEN };
1883 struct inode *inode;
1884 unsigned int lookup_flags = 0;
1885 bool switched = false;
1886 int created = 0;
1887 int err;
1888
1889 /* Expect a negative dentry */
1890 BUG_ON(d_inode(dentry));
1891
1892 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1893 dir->i_sb->s_id, dir->i_ino, dentry);
1894
1895 err = nfs_check_flags(open_flags);
1896 if (err)
1897 return err;
1898
1899 /* NFS only supports OPEN on regular files */
1900 if ((open_flags & O_DIRECTORY)) {
1901 if (!d_in_lookup(dentry)) {
1902 /*
1903 * Hashed negative dentry with O_DIRECTORY: dentry was
1904 * revalidated and is fine, no need to perform lookup
1905 * again
1906 */
1907 return -ENOENT;
1908 }
1909 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1910 goto no_open;
1911 }
1912
1913 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1914 return -ENAMETOOLONG;
1915
1916 if (open_flags & O_CREAT) {
1917 struct nfs_server *server = NFS_SERVER(dir);
1918
1919 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1920 mode &= ~current_umask();
1921
1922 attr.ia_valid |= ATTR_MODE;
1923 attr.ia_mode = mode;
1924 }
1925 if (open_flags & O_TRUNC) {
1926 attr.ia_valid |= ATTR_SIZE;
1927 attr.ia_size = 0;
1928 }
1929
1930 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1931 d_drop(dentry);
1932 switched = true;
1933 dentry = d_alloc_parallel(dentry->d_parent,
1934 &dentry->d_name, &wq);
1935 if (IS_ERR(dentry))
1936 return PTR_ERR(dentry);
1937 if (unlikely(!d_in_lookup(dentry)))
1938 return finish_no_open(file, dentry);
1939 }
1940
1941 ctx = create_nfs_open_context(dentry, open_flags, file);
1942 err = PTR_ERR(ctx);
1943 if (IS_ERR(ctx))
1944 goto out;
1945
1946 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1947 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1948 if (created)
1949 file->f_mode |= FMODE_CREATED;
1950 if (IS_ERR(inode)) {
1951 err = PTR_ERR(inode);
1952 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1953 put_nfs_open_context(ctx);
1954 d_drop(dentry);
1955 switch (err) {
1956 case -ENOENT:
1957 d_splice_alias(NULL, dentry);
1958 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1959 break;
1960 case -EISDIR:
1961 case -ENOTDIR:
1962 goto no_open;
1963 case -ELOOP:
1964 if (!(open_flags & O_NOFOLLOW))
1965 goto no_open;
1966 break;
1967 /* case -EINVAL: */
1968 default:
1969 break;
1970 }
1971 goto out;
1972 }
1973
1974 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1975 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1976 put_nfs_open_context(ctx);
1977out:
1978 if (unlikely(switched)) {
1979 d_lookup_done(dentry);
1980 dput(dentry);
1981 }
1982 return err;
1983
1984no_open:
1985 res = nfs_lookup(dir, dentry, lookup_flags);
1986 if (switched) {
1987 d_lookup_done(dentry);
1988 if (!res)
1989 res = dentry;
1990 else
1991 dput(dentry);
1992 }
1993 if (IS_ERR(res))
1994 return PTR_ERR(res);
1995 return finish_no_open(file, res);
1996}
1997EXPORT_SYMBOL_GPL(nfs_atomic_open);
1998
1999static int
2000nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2001 unsigned int flags)
2002{
2003 struct inode *inode;
2004
2005 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2006 goto full_reval;
2007 if (d_mountpoint(dentry))
2008 goto full_reval;
2009
2010 inode = d_inode(dentry);
2011
2012 /* We can't create new files in nfs_open_revalidate(), so we
2013 * optimize away revalidation of negative dentries.
2014 */
2015 if (inode == NULL)
2016 goto full_reval;
2017
2018 if (nfs_verifier_is_delegated(dentry))
2019 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2020
2021 /* NFS only supports OPEN on regular files */
2022 if (!S_ISREG(inode->i_mode))
2023 goto full_reval;
2024
2025 /* We cannot do exclusive creation on a positive dentry */
2026 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2027 goto reval_dentry;
2028
2029 /* Check if the directory changed */
2030 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2031 goto reval_dentry;
2032
2033 /* Let f_op->open() actually open (and revalidate) the file */
2034 return 1;
2035reval_dentry:
2036 if (flags & LOOKUP_RCU)
2037 return -ECHILD;
2038 return nfs_lookup_revalidate_dentry(dir, dentry, inode);
2039
2040full_reval:
2041 return nfs_do_lookup_revalidate(dir, dentry, flags);
2042}
2043
2044static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2045{
2046 return __nfs_lookup_revalidate(dentry, flags,
2047 nfs4_do_lookup_revalidate);
2048}
2049
2050#endif /* CONFIG_NFSV4 */
2051
2052struct dentry *
2053nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2054 struct nfs_fattr *fattr,
2055 struct nfs4_label *label)
2056{
2057 struct dentry *parent = dget_parent(dentry);
2058 struct inode *dir = d_inode(parent);
2059 struct inode *inode;
2060 struct dentry *d;
2061 int error;
2062
2063 d_drop(dentry);
2064
2065 if (fhandle->size == 0) {
2066 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL);
2067 if (error)
2068 goto out_error;
2069 }
2070 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2071 if (!(fattr->valid & NFS_ATTR_FATTR)) {
2072 struct nfs_server *server = NFS_SB(dentry->d_sb);
2073 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2074 fattr, NULL, NULL);
2075 if (error < 0)
2076 goto out_error;
2077 }
2078 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
2079 d = d_splice_alias(inode, dentry);
2080out:
2081 dput(parent);
2082 return d;
2083out_error:
2084 d = ERR_PTR(error);
2085 goto out;
2086}
2087EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2088
2089/*
2090 * Code common to create, mkdir, and mknod.
2091 */
2092int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2093 struct nfs_fattr *fattr,
2094 struct nfs4_label *label)
2095{
2096 struct dentry *d;
2097
2098 d = nfs_add_or_obtain(dentry, fhandle, fattr, label);
2099 if (IS_ERR(d))
2100 return PTR_ERR(d);
2101
2102 /* Callers don't care */
2103 dput(d);
2104 return 0;
2105}
2106EXPORT_SYMBOL_GPL(nfs_instantiate);
2107
2108/*
2109 * Following a failed create operation, we drop the dentry rather
2110 * than retain a negative dentry. This avoids a problem in the event
2111 * that the operation succeeded on the server, but an error in the
2112 * reply path made it appear to have failed.
2113 */
2114int nfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2115 struct dentry *dentry, umode_t mode, bool excl)
2116{
2117 struct iattr attr;
2118 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2119 int error;
2120
2121 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2122 dir->i_sb->s_id, dir->i_ino, dentry);
2123
2124 attr.ia_mode = mode;
2125 attr.ia_valid = ATTR_MODE;
2126
2127 trace_nfs_create_enter(dir, dentry, open_flags);
2128 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2129 trace_nfs_create_exit(dir, dentry, open_flags, error);
2130 if (error != 0)
2131 goto out_err;
2132 return 0;
2133out_err:
2134 d_drop(dentry);
2135 return error;
2136}
2137EXPORT_SYMBOL_GPL(nfs_create);
2138
2139/*
2140 * See comments for nfs_proc_create regarding failed operations.
2141 */
2142int
2143nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2144 struct dentry *dentry, umode_t mode, dev_t rdev)
2145{
2146 struct iattr attr;
2147 int status;
2148
2149 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2150 dir->i_sb->s_id, dir->i_ino, dentry);
2151
2152 attr.ia_mode = mode;
2153 attr.ia_valid = ATTR_MODE;
2154
2155 trace_nfs_mknod_enter(dir, dentry);
2156 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2157 trace_nfs_mknod_exit(dir, dentry, status);
2158 if (status != 0)
2159 goto out_err;
2160 return 0;
2161out_err:
2162 d_drop(dentry);
2163 return status;
2164}
2165EXPORT_SYMBOL_GPL(nfs_mknod);
2166
2167/*
2168 * See comments for nfs_proc_create regarding failed operations.
2169 */
2170int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2171 struct dentry *dentry, umode_t mode)
2172{
2173 struct iattr attr;
2174 int error;
2175
2176 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2177 dir->i_sb->s_id, dir->i_ino, dentry);
2178
2179 attr.ia_valid = ATTR_MODE;
2180 attr.ia_mode = mode | S_IFDIR;
2181
2182 trace_nfs_mkdir_enter(dir, dentry);
2183 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2184 trace_nfs_mkdir_exit(dir, dentry, error);
2185 if (error != 0)
2186 goto out_err;
2187 return 0;
2188out_err:
2189 d_drop(dentry);
2190 return error;
2191}
2192EXPORT_SYMBOL_GPL(nfs_mkdir);
2193
2194static void nfs_dentry_handle_enoent(struct dentry *dentry)
2195{
2196 if (simple_positive(dentry))
2197 d_delete(dentry);
2198}
2199
2200int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2201{
2202 int error;
2203
2204 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2205 dir->i_sb->s_id, dir->i_ino, dentry);
2206
2207 trace_nfs_rmdir_enter(dir, dentry);
2208 if (d_really_is_positive(dentry)) {
2209 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2210 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2211 /* Ensure the VFS deletes this inode */
2212 switch (error) {
2213 case 0:
2214 clear_nlink(d_inode(dentry));
2215 break;
2216 case -ENOENT:
2217 nfs_dentry_handle_enoent(dentry);
2218 }
2219 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2220 } else
2221 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2222 trace_nfs_rmdir_exit(dir, dentry, error);
2223
2224 return error;
2225}
2226EXPORT_SYMBOL_GPL(nfs_rmdir);
2227
2228/*
2229 * Remove a file after making sure there are no pending writes,
2230 * and after checking that the file has only one user.
2231 *
2232 * We invalidate the attribute cache and free the inode prior to the operation
2233 * to avoid possible races if the server reuses the inode.
2234 */
2235static int nfs_safe_remove(struct dentry *dentry)
2236{
2237 struct inode *dir = d_inode(dentry->d_parent);
2238 struct inode *inode = d_inode(dentry);
2239 int error = -EBUSY;
2240
2241 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2242
2243 /* If the dentry was sillyrenamed, we simply call d_delete() */
2244 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2245 error = 0;
2246 goto out;
2247 }
2248
2249 trace_nfs_remove_enter(dir, dentry);
2250 if (inode != NULL) {
2251 error = NFS_PROTO(dir)->remove(dir, dentry);
2252 if (error == 0)
2253 nfs_drop_nlink(inode);
2254 } else
2255 error = NFS_PROTO(dir)->remove(dir, dentry);
2256 if (error == -ENOENT)
2257 nfs_dentry_handle_enoent(dentry);
2258 trace_nfs_remove_exit(dir, dentry, error);
2259out:
2260 return error;
2261}
2262
2263/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2264 * belongs to an active ".nfs..." file and we return -EBUSY.
2265 *
2266 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2267 */
2268int nfs_unlink(struct inode *dir, struct dentry *dentry)
2269{
2270 int error;
2271 int need_rehash = 0;
2272
2273 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2274 dir->i_ino, dentry);
2275
2276 trace_nfs_unlink_enter(dir, dentry);
2277 spin_lock(&dentry->d_lock);
2278 if (d_count(dentry) > 1) {
2279 spin_unlock(&dentry->d_lock);
2280 /* Start asynchronous writeout of the inode */
2281 write_inode_now(d_inode(dentry), 0);
2282 error = nfs_sillyrename(dir, dentry);
2283 goto out;
2284 }
2285 if (!d_unhashed(dentry)) {
2286 __d_drop(dentry);
2287 need_rehash = 1;
2288 }
2289 spin_unlock(&dentry->d_lock);
2290 error = nfs_safe_remove(dentry);
2291 if (!error || error == -ENOENT) {
2292 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2293 } else if (need_rehash)
2294 d_rehash(dentry);
2295out:
2296 trace_nfs_unlink_exit(dir, dentry, error);
2297 return error;
2298}
2299EXPORT_SYMBOL_GPL(nfs_unlink);
2300
2301/*
2302 * To create a symbolic link, most file systems instantiate a new inode,
2303 * add a page to it containing the path, then write it out to the disk
2304 * using prepare_write/commit_write.
2305 *
2306 * Unfortunately the NFS client can't create the in-core inode first
2307 * because it needs a file handle to create an in-core inode (see
2308 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2309 * symlink request has completed on the server.
2310 *
2311 * So instead we allocate a raw page, copy the symname into it, then do
2312 * the SYMLINK request with the page as the buffer. If it succeeds, we
2313 * now have a new file handle and can instantiate an in-core NFS inode
2314 * and move the raw page into its mapping.
2315 */
2316int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
2317 struct dentry *dentry, const char *symname)
2318{
2319 struct page *page;
2320 char *kaddr;
2321 struct iattr attr;
2322 unsigned int pathlen = strlen(symname);
2323 int error;
2324
2325 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2326 dir->i_ino, dentry, symname);
2327
2328 if (pathlen > PAGE_SIZE)
2329 return -ENAMETOOLONG;
2330
2331 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2332 attr.ia_valid = ATTR_MODE;
2333
2334 page = alloc_page(GFP_USER);
2335 if (!page)
2336 return -ENOMEM;
2337
2338 kaddr = page_address(page);
2339 memcpy(kaddr, symname, pathlen);
2340 if (pathlen < PAGE_SIZE)
2341 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2342
2343 trace_nfs_symlink_enter(dir, dentry);
2344 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2345 trace_nfs_symlink_exit(dir, dentry, error);
2346 if (error != 0) {
2347 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2348 dir->i_sb->s_id, dir->i_ino,
2349 dentry, symname, error);
2350 d_drop(dentry);
2351 __free_page(page);
2352 return error;
2353 }
2354
2355 /*
2356 * No big deal if we can't add this page to the page cache here.
2357 * READLINK will get the missing page from the server if needed.
2358 */
2359 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2360 GFP_KERNEL)) {
2361 SetPageUptodate(page);
2362 unlock_page(page);
2363 /*
2364 * add_to_page_cache_lru() grabs an extra page refcount.
2365 * Drop it here to avoid leaking this page later.
2366 */
2367 put_page(page);
2368 } else
2369 __free_page(page);
2370
2371 return 0;
2372}
2373EXPORT_SYMBOL_GPL(nfs_symlink);
2374
2375int
2376nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2377{
2378 struct inode *inode = d_inode(old_dentry);
2379 int error;
2380
2381 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2382 old_dentry, dentry);
2383
2384 trace_nfs_link_enter(inode, dir, dentry);
2385 d_drop(dentry);
2386 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2387 if (error == 0) {
2388 ihold(inode);
2389 d_add(dentry, inode);
2390 }
2391 trace_nfs_link_exit(inode, dir, dentry, error);
2392 return error;
2393}
2394EXPORT_SYMBOL_GPL(nfs_link);
2395
2396/*
2397 * RENAME
2398 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2399 * different file handle for the same inode after a rename (e.g. when
2400 * moving to a different directory). A fail-safe method to do so would
2401 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2402 * rename the old file using the sillyrename stuff. This way, the original
2403 * file in old_dir will go away when the last process iput()s the inode.
2404 *
2405 * FIXED.
2406 *
2407 * It actually works quite well. One needs to have the possibility for
2408 * at least one ".nfs..." file in each directory the file ever gets
2409 * moved or linked to which happens automagically with the new
2410 * implementation that only depends on the dcache stuff instead of
2411 * using the inode layer
2412 *
2413 * Unfortunately, things are a little more complicated than indicated
2414 * above. For a cross-directory move, we want to make sure we can get
2415 * rid of the old inode after the operation. This means there must be
2416 * no pending writes (if it's a file), and the use count must be 1.
2417 * If these conditions are met, we can drop the dentries before doing
2418 * the rename.
2419 */
2420int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
2421 struct dentry *old_dentry, struct inode *new_dir,
2422 struct dentry *new_dentry, unsigned int flags)
2423{
2424 struct inode *old_inode = d_inode(old_dentry);
2425 struct inode *new_inode = d_inode(new_dentry);
2426 struct dentry *dentry = NULL, *rehash = NULL;
2427 struct rpc_task *task;
2428 int error = -EBUSY;
2429
2430 if (flags)
2431 return -EINVAL;
2432
2433 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2434 old_dentry, new_dentry,
2435 d_count(new_dentry));
2436
2437 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2438 /*
2439 * For non-directories, check whether the target is busy and if so,
2440 * make a copy of the dentry and then do a silly-rename. If the
2441 * silly-rename succeeds, the copied dentry is hashed and becomes
2442 * the new target.
2443 */
2444 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2445 /*
2446 * To prevent any new references to the target during the
2447 * rename, we unhash the dentry in advance.
2448 */
2449 if (!d_unhashed(new_dentry)) {
2450 d_drop(new_dentry);
2451 rehash = new_dentry;
2452 }
2453
2454 if (d_count(new_dentry) > 2) {
2455 int err;
2456
2457 /* copy the target dentry's name */
2458 dentry = d_alloc(new_dentry->d_parent,
2459 &new_dentry->d_name);
2460 if (!dentry)
2461 goto out;
2462
2463 /* silly-rename the existing target ... */
2464 err = nfs_sillyrename(new_dir, new_dentry);
2465 if (err)
2466 goto out;
2467
2468 new_dentry = dentry;
2469 rehash = NULL;
2470 new_inode = NULL;
2471 }
2472 }
2473
2474 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2475 if (IS_ERR(task)) {
2476 error = PTR_ERR(task);
2477 goto out;
2478 }
2479
2480 error = rpc_wait_for_completion_task(task);
2481 if (error != 0) {
2482 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2483 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2484 smp_wmb();
2485 } else
2486 error = task->tk_status;
2487 rpc_put_task(task);
2488 /* Ensure the inode attributes are revalidated */
2489 if (error == 0) {
2490 spin_lock(&old_inode->i_lock);
2491 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2492 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2493 NFS_INO_INVALID_CTIME |
2494 NFS_INO_REVAL_FORCED);
2495 spin_unlock(&old_inode->i_lock);
2496 }
2497out:
2498 if (rehash)
2499 d_rehash(rehash);
2500 trace_nfs_rename_exit(old_dir, old_dentry,
2501 new_dir, new_dentry, error);
2502 if (!error) {
2503 if (new_inode != NULL)
2504 nfs_drop_nlink(new_inode);
2505 /*
2506 * The d_move() should be here instead of in an async RPC completion
2507 * handler because we need the proper locks to move the dentry. If
2508 * we're interrupted by a signal, the async RPC completion handler
2509 * should mark the directories for revalidation.
2510 */
2511 d_move(old_dentry, new_dentry);
2512 nfs_set_verifier(old_dentry,
2513 nfs_save_change_attribute(new_dir));
2514 } else if (error == -ENOENT)
2515 nfs_dentry_handle_enoent(old_dentry);
2516
2517 /* new dentry created? */
2518 if (dentry)
2519 dput(dentry);
2520 return error;
2521}
2522EXPORT_SYMBOL_GPL(nfs_rename);
2523
2524static DEFINE_SPINLOCK(nfs_access_lru_lock);
2525static LIST_HEAD(nfs_access_lru_list);
2526static atomic_long_t nfs_access_nr_entries;
2527
2528static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2529module_param(nfs_access_max_cachesize, ulong, 0644);
2530MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2531
2532static void nfs_access_free_entry(struct nfs_access_entry *entry)
2533{
2534 put_cred(entry->cred);
2535 kfree_rcu(entry, rcu_head);
2536 smp_mb__before_atomic();
2537 atomic_long_dec(&nfs_access_nr_entries);
2538 smp_mb__after_atomic();
2539}
2540
2541static void nfs_access_free_list(struct list_head *head)
2542{
2543 struct nfs_access_entry *cache;
2544
2545 while (!list_empty(head)) {
2546 cache = list_entry(head->next, struct nfs_access_entry, lru);
2547 list_del(&cache->lru);
2548 nfs_access_free_entry(cache);
2549 }
2550}
2551
2552static unsigned long
2553nfs_do_access_cache_scan(unsigned int nr_to_scan)
2554{
2555 LIST_HEAD(head);
2556 struct nfs_inode *nfsi, *next;
2557 struct nfs_access_entry *cache;
2558 long freed = 0;
2559
2560 spin_lock(&nfs_access_lru_lock);
2561 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2562 struct inode *inode;
2563
2564 if (nr_to_scan-- == 0)
2565 break;
2566 inode = &nfsi->vfs_inode;
2567 spin_lock(&inode->i_lock);
2568 if (list_empty(&nfsi->access_cache_entry_lru))
2569 goto remove_lru_entry;
2570 cache = list_entry(nfsi->access_cache_entry_lru.next,
2571 struct nfs_access_entry, lru);
2572 list_move(&cache->lru, &head);
2573 rb_erase(&cache->rb_node, &nfsi->access_cache);
2574 freed++;
2575 if (!list_empty(&nfsi->access_cache_entry_lru))
2576 list_move_tail(&nfsi->access_cache_inode_lru,
2577 &nfs_access_lru_list);
2578 else {
2579remove_lru_entry:
2580 list_del_init(&nfsi->access_cache_inode_lru);
2581 smp_mb__before_atomic();
2582 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2583 smp_mb__after_atomic();
2584 }
2585 spin_unlock(&inode->i_lock);
2586 }
2587 spin_unlock(&nfs_access_lru_lock);
2588 nfs_access_free_list(&head);
2589 return freed;
2590}
2591
2592unsigned long
2593nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2594{
2595 int nr_to_scan = sc->nr_to_scan;
2596 gfp_t gfp_mask = sc->gfp_mask;
2597
2598 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2599 return SHRINK_STOP;
2600 return nfs_do_access_cache_scan(nr_to_scan);
2601}
2602
2603
2604unsigned long
2605nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2606{
2607 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2608}
2609
2610static void
2611nfs_access_cache_enforce_limit(void)
2612{
2613 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2614 unsigned long diff;
2615 unsigned int nr_to_scan;
2616
2617 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2618 return;
2619 nr_to_scan = 100;
2620 diff = nr_entries - nfs_access_max_cachesize;
2621 if (diff < nr_to_scan)
2622 nr_to_scan = diff;
2623 nfs_do_access_cache_scan(nr_to_scan);
2624}
2625
2626static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2627{
2628 struct rb_root *root_node = &nfsi->access_cache;
2629 struct rb_node *n;
2630 struct nfs_access_entry *entry;
2631
2632 /* Unhook entries from the cache */
2633 while ((n = rb_first(root_node)) != NULL) {
2634 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2635 rb_erase(n, root_node);
2636 list_move(&entry->lru, head);
2637 }
2638 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2639}
2640
2641void nfs_access_zap_cache(struct inode *inode)
2642{
2643 LIST_HEAD(head);
2644
2645 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2646 return;
2647 /* Remove from global LRU init */
2648 spin_lock(&nfs_access_lru_lock);
2649 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2650 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2651
2652 spin_lock(&inode->i_lock);
2653 __nfs_access_zap_cache(NFS_I(inode), &head);
2654 spin_unlock(&inode->i_lock);
2655 spin_unlock(&nfs_access_lru_lock);
2656 nfs_access_free_list(&head);
2657}
2658EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2659
2660static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2661{
2662 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2663
2664 while (n != NULL) {
2665 struct nfs_access_entry *entry =
2666 rb_entry(n, struct nfs_access_entry, rb_node);
2667 int cmp = cred_fscmp(cred, entry->cred);
2668
2669 if (cmp < 0)
2670 n = n->rb_left;
2671 else if (cmp > 0)
2672 n = n->rb_right;
2673 else
2674 return entry;
2675 }
2676 return NULL;
2677}
2678
2679static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block)
2680{
2681 struct nfs_inode *nfsi = NFS_I(inode);
2682 struct nfs_access_entry *cache;
2683 bool retry = true;
2684 int err;
2685
2686 spin_lock(&inode->i_lock);
2687 for(;;) {
2688 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2689 goto out_zap;
2690 cache = nfs_access_search_rbtree(inode, cred);
2691 err = -ENOENT;
2692 if (cache == NULL)
2693 goto out;
2694 /* Found an entry, is our attribute cache valid? */
2695 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2696 break;
2697 if (!retry)
2698 break;
2699 err = -ECHILD;
2700 if (!may_block)
2701 goto out;
2702 spin_unlock(&inode->i_lock);
2703 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2704 if (err)
2705 return err;
2706 spin_lock(&inode->i_lock);
2707 retry = false;
2708 }
2709 res->cred = cache->cred;
2710 res->mask = cache->mask;
2711 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2712 err = 0;
2713out:
2714 spin_unlock(&inode->i_lock);
2715 return err;
2716out_zap:
2717 spin_unlock(&inode->i_lock);
2718 nfs_access_zap_cache(inode);
2719 return -ENOENT;
2720}
2721
2722static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res)
2723{
2724 /* Only check the most recently returned cache entry,
2725 * but do it without locking.
2726 */
2727 struct nfs_inode *nfsi = NFS_I(inode);
2728 struct nfs_access_entry *cache;
2729 int err = -ECHILD;
2730 struct list_head *lh;
2731
2732 rcu_read_lock();
2733 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2734 goto out;
2735 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
2736 cache = list_entry(lh, struct nfs_access_entry, lru);
2737 if (lh == &nfsi->access_cache_entry_lru ||
2738 cred_fscmp(cred, cache->cred) != 0)
2739 cache = NULL;
2740 if (cache == NULL)
2741 goto out;
2742 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2743 goto out;
2744 res->cred = cache->cred;
2745 res->mask = cache->mask;
2746 err = 0;
2747out:
2748 rcu_read_unlock();
2749 return err;
2750}
2751
2752int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct
2753nfs_access_entry *res, bool may_block)
2754{
2755 int status;
2756
2757 status = nfs_access_get_cached_rcu(inode, cred, res);
2758 if (status != 0)
2759 status = nfs_access_get_cached_locked(inode, cred, res,
2760 may_block);
2761
2762 return status;
2763}
2764EXPORT_SYMBOL_GPL(nfs_access_get_cached);
2765
2766static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2767{
2768 struct nfs_inode *nfsi = NFS_I(inode);
2769 struct rb_root *root_node = &nfsi->access_cache;
2770 struct rb_node **p = &root_node->rb_node;
2771 struct rb_node *parent = NULL;
2772 struct nfs_access_entry *entry;
2773 int cmp;
2774
2775 spin_lock(&inode->i_lock);
2776 while (*p != NULL) {
2777 parent = *p;
2778 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2779 cmp = cred_fscmp(set->cred, entry->cred);
2780
2781 if (cmp < 0)
2782 p = &parent->rb_left;
2783 else if (cmp > 0)
2784 p = &parent->rb_right;
2785 else
2786 goto found;
2787 }
2788 rb_link_node(&set->rb_node, parent, p);
2789 rb_insert_color(&set->rb_node, root_node);
2790 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2791 spin_unlock(&inode->i_lock);
2792 return;
2793found:
2794 rb_replace_node(parent, &set->rb_node, root_node);
2795 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2796 list_del(&entry->lru);
2797 spin_unlock(&inode->i_lock);
2798 nfs_access_free_entry(entry);
2799}
2800
2801void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2802{
2803 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2804 if (cache == NULL)
2805 return;
2806 RB_CLEAR_NODE(&cache->rb_node);
2807 cache->cred = get_cred(set->cred);
2808 cache->mask = set->mask;
2809
2810 /* The above field assignments must be visible
2811 * before this item appears on the lru. We cannot easily
2812 * use rcu_assign_pointer, so just force the memory barrier.
2813 */
2814 smp_wmb();
2815 nfs_access_add_rbtree(inode, cache);
2816
2817 /* Update accounting */
2818 smp_mb__before_atomic();
2819 atomic_long_inc(&nfs_access_nr_entries);
2820 smp_mb__after_atomic();
2821
2822 /* Add inode to global LRU list */
2823 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2824 spin_lock(&nfs_access_lru_lock);
2825 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2826 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2827 &nfs_access_lru_list);
2828 spin_unlock(&nfs_access_lru_lock);
2829 }
2830 nfs_access_cache_enforce_limit();
2831}
2832EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2833
2834#define NFS_MAY_READ (NFS_ACCESS_READ)
2835#define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2836 NFS_ACCESS_EXTEND | \
2837 NFS_ACCESS_DELETE)
2838#define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2839 NFS_ACCESS_EXTEND)
2840#define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2841#define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2842#define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2843static int
2844nfs_access_calc_mask(u32 access_result, umode_t umode)
2845{
2846 int mask = 0;
2847
2848 if (access_result & NFS_MAY_READ)
2849 mask |= MAY_READ;
2850 if (S_ISDIR(umode)) {
2851 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2852 mask |= MAY_WRITE;
2853 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2854 mask |= MAY_EXEC;
2855 } else if (S_ISREG(umode)) {
2856 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2857 mask |= MAY_WRITE;
2858 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2859 mask |= MAY_EXEC;
2860 } else if (access_result & NFS_MAY_WRITE)
2861 mask |= MAY_WRITE;
2862 return mask;
2863}
2864
2865void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2866{
2867 entry->mask = access_result;
2868}
2869EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2870
2871static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
2872{
2873 struct nfs_access_entry cache;
2874 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2875 int cache_mask = -1;
2876 int status;
2877
2878 trace_nfs_access_enter(inode);
2879
2880 status = nfs_access_get_cached(inode, cred, &cache, may_block);
2881 if (status == 0)
2882 goto out_cached;
2883
2884 status = -ECHILD;
2885 if (!may_block)
2886 goto out;
2887
2888 /*
2889 * Determine which access bits we want to ask for...
2890 */
2891 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2892 if (nfs_server_capable(inode, NFS_CAP_XATTR)) {
2893 cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE |
2894 NFS_ACCESS_XALIST;
2895 }
2896 if (S_ISDIR(inode->i_mode))
2897 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2898 else
2899 cache.mask |= NFS_ACCESS_EXECUTE;
2900 cache.cred = cred;
2901 status = NFS_PROTO(inode)->access(inode, &cache);
2902 if (status != 0) {
2903 if (status == -ESTALE) {
2904 if (!S_ISDIR(inode->i_mode))
2905 nfs_set_inode_stale(inode);
2906 else
2907 nfs_zap_caches(inode);
2908 }
2909 goto out;
2910 }
2911 nfs_access_add_cache(inode, &cache);
2912out_cached:
2913 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2914 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2915 status = -EACCES;
2916out:
2917 trace_nfs_access_exit(inode, mask, cache_mask, status);
2918 return status;
2919}
2920
2921static int nfs_open_permission_mask(int openflags)
2922{
2923 int mask = 0;
2924
2925 if (openflags & __FMODE_EXEC) {
2926 /* ONLY check exec rights */
2927 mask = MAY_EXEC;
2928 } else {
2929 if ((openflags & O_ACCMODE) != O_WRONLY)
2930 mask |= MAY_READ;
2931 if ((openflags & O_ACCMODE) != O_RDONLY)
2932 mask |= MAY_WRITE;
2933 }
2934
2935 return mask;
2936}
2937
2938int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
2939{
2940 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2941}
2942EXPORT_SYMBOL_GPL(nfs_may_open);
2943
2944static int nfs_execute_ok(struct inode *inode, int mask)
2945{
2946 struct nfs_server *server = NFS_SERVER(inode);
2947 int ret = 0;
2948
2949 if (S_ISDIR(inode->i_mode))
2950 return 0;
2951 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
2952 if (mask & MAY_NOT_BLOCK)
2953 return -ECHILD;
2954 ret = __nfs_revalidate_inode(server, inode);
2955 }
2956 if (ret == 0 && !execute_ok(inode))
2957 ret = -EACCES;
2958 return ret;
2959}
2960
2961int nfs_permission(struct user_namespace *mnt_userns,
2962 struct inode *inode,
2963 int mask)
2964{
2965 const struct cred *cred = current_cred();
2966 int res = 0;
2967
2968 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2969
2970 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2971 goto out;
2972 /* Is this sys_access() ? */
2973 if (mask & (MAY_ACCESS | MAY_CHDIR))
2974 goto force_lookup;
2975
2976 switch (inode->i_mode & S_IFMT) {
2977 case S_IFLNK:
2978 goto out;
2979 case S_IFREG:
2980 if ((mask & MAY_OPEN) &&
2981 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2982 return 0;
2983 break;
2984 case S_IFDIR:
2985 /*
2986 * Optimize away all write operations, since the server
2987 * will check permissions when we perform the op.
2988 */
2989 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2990 goto out;
2991 }
2992
2993force_lookup:
2994 if (!NFS_PROTO(inode)->access)
2995 goto out_notsup;
2996
2997 res = nfs_do_access(inode, cred, mask);
2998out:
2999 if (!res && (mask & MAY_EXEC))
3000 res = nfs_execute_ok(inode, mask);
3001
3002 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3003 inode->i_sb->s_id, inode->i_ino, mask, res);
3004 return res;
3005out_notsup:
3006 if (mask & MAY_NOT_BLOCK)
3007 return -ECHILD;
3008
3009 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3010 NFS_INO_INVALID_OTHER);
3011 if (res == 0)
3012 res = generic_permission(&init_user_ns, inode, mask);
3013 goto out;
3014}
3015EXPORT_SYMBOL_GPL(nfs_permission);