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_clear_array(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_clear_array,
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 folio_full : 1,
150 folio_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 folio *folio;
158 struct dir_context *ctx;
159 pgoff_t folio_index;
160 pgoff_t folio_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_folio_init_array(struct folio *folio, u64 last_cookie,
202 u64 change_attr)
203{
204 struct nfs_cache_array *array;
205
206 array = kmap_local_folio(folio, 0);
207 array->change_attr = change_attr;
208 array->last_cookie = last_cookie;
209 array->size = 0;
210 array->folio_full = 0;
211 array->folio_is_eof = 0;
212 array->cookies_are_ordered = 1;
213 kunmap_local(array);
214}
215
216/*
217 * we are freeing strings created by nfs_add_to_readdir_array()
218 */
219static void nfs_readdir_clear_array(struct folio *folio)
220{
221 struct nfs_cache_array *array;
222 unsigned int i;
223
224 array = kmap_local_folio(folio, 0);
225 for (i = 0; i < array->size; i++)
226 kfree(array->array[i].name);
227 array->size = 0;
228 kunmap_local(array);
229}
230
231static void nfs_readdir_folio_reinit_array(struct folio *folio, u64 last_cookie,
232 u64 change_attr)
233{
234 nfs_readdir_clear_array(folio);
235 nfs_readdir_folio_init_array(folio, last_cookie, change_attr);
236}
237
238static struct folio *
239nfs_readdir_folio_array_alloc(u64 last_cookie, gfp_t gfp_flags)
240{
241 struct folio *folio = folio_alloc(gfp_flags, 0);
242 if (folio)
243 nfs_readdir_folio_init_array(folio, last_cookie, 0);
244 return folio;
245}
246
247static void nfs_readdir_folio_array_free(struct folio *folio)
248{
249 if (folio) {
250 nfs_readdir_clear_array(folio);
251 folio_put(folio);
252 }
253}
254
255static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array)
256{
257 return array->size == 0 ? array->last_cookie : array->array[0].cookie;
258}
259
260static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
261{
262 array->folio_is_eof = 1;
263 array->folio_full = 1;
264}
265
266static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
267{
268 return array->folio_full;
269}
270
271/*
272 * the caller is responsible for freeing qstr.name
273 * when called by nfs_readdir_add_to_array, the strings will be freed in
274 * nfs_clear_readdir_array()
275 */
276static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
277{
278 const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
279
280 /*
281 * Avoid a kmemleak false positive. The pointer to the name is stored
282 * in a page cache page which kmemleak does not scan.
283 */
284 if (ret != NULL)
285 kmemleak_not_leak(ret);
286 return ret;
287}
288
289static size_t nfs_readdir_array_maxentries(void)
290{
291 return (PAGE_SIZE - sizeof(struct nfs_cache_array)) /
292 sizeof(struct nfs_cache_array_entry);
293}
294
295/*
296 * Check that the next array entry lies entirely within the page bounds
297 */
298static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
299{
300 if (array->folio_full)
301 return -ENOSPC;
302 if (array->size == nfs_readdir_array_maxentries()) {
303 array->folio_full = 1;
304 return -ENOSPC;
305 }
306 return 0;
307}
308
309static int nfs_readdir_folio_array_append(struct folio *folio,
310 const struct nfs_entry *entry,
311 u64 *cookie)
312{
313 struct nfs_cache_array *array;
314 struct nfs_cache_array_entry *cache_entry;
315 const char *name;
316 int ret = -ENOMEM;
317
318 name = nfs_readdir_copy_name(entry->name, entry->len);
319
320 array = kmap_local_folio(folio, 0);
321 if (!name)
322 goto out;
323 ret = nfs_readdir_array_can_expand(array);
324 if (ret) {
325 kfree(name);
326 goto out;
327 }
328
329 cache_entry = &array->array[array->size];
330 cache_entry->cookie = array->last_cookie;
331 cache_entry->ino = entry->ino;
332 cache_entry->d_type = entry->d_type;
333 cache_entry->name_len = entry->len;
334 cache_entry->name = name;
335 array->last_cookie = entry->cookie;
336 if (array->last_cookie <= cache_entry->cookie)
337 array->cookies_are_ordered = 0;
338 array->size++;
339 if (entry->eof != 0)
340 nfs_readdir_array_set_eof(array);
341out:
342 *cookie = array->last_cookie;
343 kunmap_local(array);
344 return ret;
345}
346
347#define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14)
348/*
349 * Hash algorithm allowing content addressible access to sequences
350 * of directory cookies. Content is addressed by the value of the
351 * cookie index of the first readdir entry in a page.
352 *
353 * We select only the first 18 bits to avoid issues with excessive
354 * memory use for the page cache XArray. 18 bits should allow the caching
355 * of 262144 pages of sequences of readdir entries. Since each page holds
356 * 127 readdir entries for a typical 64-bit system, that works out to a
357 * cache of ~ 33 million entries per directory.
358 */
359static pgoff_t nfs_readdir_folio_cookie_hash(u64 cookie)
360{
361 if (cookie == 0)
362 return 0;
363 return hash_64(cookie, 18);
364}
365
366static bool nfs_readdir_folio_validate(struct folio *folio, u64 last_cookie,
367 u64 change_attr)
368{
369 struct nfs_cache_array *array = kmap_local_folio(folio, 0);
370 int ret = true;
371
372 if (array->change_attr != change_attr)
373 ret = false;
374 if (nfs_readdir_array_index_cookie(array) != last_cookie)
375 ret = false;
376 kunmap_local(array);
377 return ret;
378}
379
380static void nfs_readdir_folio_unlock_and_put(struct folio *folio)
381{
382 folio_unlock(folio);
383 folio_put(folio);
384}
385
386static void nfs_readdir_folio_init_and_validate(struct folio *folio, u64 cookie,
387 u64 change_attr)
388{
389 if (folio_test_uptodate(folio)) {
390 if (nfs_readdir_folio_validate(folio, cookie, change_attr))
391 return;
392 nfs_readdir_clear_array(folio);
393 }
394 nfs_readdir_folio_init_array(folio, cookie, change_attr);
395 folio_mark_uptodate(folio);
396}
397
398static struct folio *nfs_readdir_folio_get_locked(struct address_space *mapping,
399 u64 cookie, u64 change_attr)
400{
401 pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
402 struct folio *folio;
403
404 folio = filemap_grab_folio(mapping, index);
405 if (IS_ERR(folio))
406 return NULL;
407 nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
408 return folio;
409}
410
411static u64 nfs_readdir_folio_last_cookie(struct folio *folio)
412{
413 struct nfs_cache_array *array;
414 u64 ret;
415
416 array = kmap_local_folio(folio, 0);
417 ret = array->last_cookie;
418 kunmap_local(array);
419 return ret;
420}
421
422static bool nfs_readdir_folio_needs_filling(struct folio *folio)
423{
424 struct nfs_cache_array *array;
425 bool ret;
426
427 array = kmap_local_folio(folio, 0);
428 ret = !nfs_readdir_array_is_full(array);
429 kunmap_local(array);
430 return ret;
431}
432
433static void nfs_readdir_folio_set_eof(struct folio *folio)
434{
435 struct nfs_cache_array *array;
436
437 array = kmap_local_folio(folio, 0);
438 nfs_readdir_array_set_eof(array);
439 kunmap_local(array);
440}
441
442static struct folio *nfs_readdir_folio_get_next(struct address_space *mapping,
443 u64 cookie, u64 change_attr)
444{
445 pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
446 struct folio *folio;
447
448 folio = __filemap_get_folio(mapping, index,
449 FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
450 mapping_gfp_mask(mapping));
451 if (IS_ERR(folio))
452 return NULL;
453 nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
454 if (nfs_readdir_folio_last_cookie(folio) != cookie)
455 nfs_readdir_folio_reinit_array(folio, cookie, change_attr);
456 return folio;
457}
458
459static inline
460int is_32bit_api(void)
461{
462#ifdef CONFIG_COMPAT
463 return in_compat_syscall();
464#else
465 return (BITS_PER_LONG == 32);
466#endif
467}
468
469static
470bool nfs_readdir_use_cookie(const struct file *filp)
471{
472 if ((filp->f_mode & FMODE_32BITHASH) ||
473 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
474 return false;
475 return true;
476}
477
478static void nfs_readdir_seek_next_array(struct nfs_cache_array *array,
479 struct nfs_readdir_descriptor *desc)
480{
481 if (array->folio_full) {
482 desc->last_cookie = array->last_cookie;
483 desc->current_index += array->size;
484 desc->cache_entry_index = 0;
485 desc->folio_index++;
486 } else
487 desc->last_cookie = nfs_readdir_array_index_cookie(array);
488}
489
490static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc)
491{
492 desc->current_index = 0;
493 desc->last_cookie = 0;
494 desc->folio_index = 0;
495}
496
497static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
498 struct nfs_readdir_descriptor *desc)
499{
500 loff_t diff = desc->ctx->pos - desc->current_index;
501 unsigned int index;
502
503 if (diff < 0)
504 goto out_eof;
505 if (diff >= array->size) {
506 if (array->folio_is_eof)
507 goto out_eof;
508 nfs_readdir_seek_next_array(array, desc);
509 return -EAGAIN;
510 }
511
512 index = (unsigned int)diff;
513 desc->dir_cookie = array->array[index].cookie;
514 desc->cache_entry_index = index;
515 return 0;
516out_eof:
517 desc->eof = true;
518 return -EBADCOOKIE;
519}
520
521static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
522 u64 cookie)
523{
524 if (!array->cookies_are_ordered)
525 return true;
526 /* Optimisation for monotonically increasing cookies */
527 if (cookie >= array->last_cookie)
528 return false;
529 if (array->size && cookie < array->array[0].cookie)
530 return false;
531 return true;
532}
533
534static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
535 struct nfs_readdir_descriptor *desc)
536{
537 unsigned int i;
538 int status = -EAGAIN;
539
540 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
541 goto check_eof;
542
543 for (i = 0; i < array->size; i++) {
544 if (array->array[i].cookie == desc->dir_cookie) {
545 if (nfs_readdir_use_cookie(desc->file))
546 desc->ctx->pos = desc->dir_cookie;
547 else
548 desc->ctx->pos = desc->current_index + i;
549 desc->cache_entry_index = i;
550 return 0;
551 }
552 }
553check_eof:
554 if (array->folio_is_eof) {
555 status = -EBADCOOKIE;
556 if (desc->dir_cookie == array->last_cookie)
557 desc->eof = true;
558 } else
559 nfs_readdir_seek_next_array(array, desc);
560 return status;
561}
562
563static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
564{
565 struct nfs_cache_array *array;
566 int status;
567
568 array = kmap_local_folio(desc->folio, 0);
569
570 if (desc->dir_cookie == 0)
571 status = nfs_readdir_search_for_pos(array, desc);
572 else
573 status = nfs_readdir_search_for_cookie(array, desc);
574
575 kunmap_local(array);
576 return status;
577}
578
579/* Fill a page with xdr information before transferring to the cache page */
580static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
581 __be32 *verf, u64 cookie,
582 struct page **pages, size_t bufsize,
583 __be32 *verf_res)
584{
585 struct inode *inode = file_inode(desc->file);
586 struct nfs_readdir_arg arg = {
587 .dentry = file_dentry(desc->file),
588 .cred = desc->file->f_cred,
589 .verf = verf,
590 .cookie = cookie,
591 .pages = pages,
592 .page_len = bufsize,
593 .plus = desc->plus,
594 };
595 struct nfs_readdir_res res = {
596 .verf = verf_res,
597 };
598 unsigned long timestamp, gencount;
599 int error;
600
601 again:
602 timestamp = jiffies;
603 gencount = nfs_inc_attr_generation_counter();
604 desc->dir_verifier = nfs_save_change_attribute(inode);
605 error = NFS_PROTO(inode)->readdir(&arg, &res);
606 if (error < 0) {
607 /* We requested READDIRPLUS, but the server doesn't grok it */
608 if (error == -ENOTSUPP && desc->plus) {
609 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
610 desc->plus = arg.plus = false;
611 goto again;
612 }
613 goto error;
614 }
615 desc->timestamp = timestamp;
616 desc->gencount = gencount;
617error:
618 return error;
619}
620
621static int xdr_decode(struct nfs_readdir_descriptor *desc,
622 struct nfs_entry *entry, struct xdr_stream *xdr)
623{
624 struct inode *inode = file_inode(desc->file);
625 int error;
626
627 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
628 if (error)
629 return error;
630 entry->fattr->time_start = desc->timestamp;
631 entry->fattr->gencount = desc->gencount;
632 return 0;
633}
634
635/* Match file and dirent using either filehandle or fileid
636 * Note: caller is responsible for checking the fsid
637 */
638static
639int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
640{
641 struct inode *inode;
642 struct nfs_inode *nfsi;
643
644 if (d_really_is_negative(dentry))
645 return 0;
646
647 inode = d_inode(dentry);
648 if (is_bad_inode(inode) || NFS_STALE(inode))
649 return 0;
650
651 nfsi = NFS_I(inode);
652 if (entry->fattr->fileid != nfsi->fileid)
653 return 0;
654 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
655 return 0;
656 return 1;
657}
658
659#define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL)
660
661static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx,
662 unsigned int cache_hits,
663 unsigned int cache_misses)
664{
665 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
666 return false;
667 if (ctx->pos == 0 ||
668 cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD)
669 return true;
670 return false;
671}
672
673/*
674 * This function is called by the getattr code to request the
675 * use of readdirplus to accelerate any future lookups in the same
676 * directory.
677 */
678void nfs_readdir_record_entry_cache_hit(struct inode *dir)
679{
680 struct nfs_inode *nfsi = NFS_I(dir);
681 struct nfs_open_dir_context *ctx;
682
683 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
684 S_ISDIR(dir->i_mode)) {
685 rcu_read_lock();
686 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
687 atomic_inc(&ctx->cache_hits);
688 rcu_read_unlock();
689 }
690}
691
692/*
693 * This function is mainly for use by nfs_getattr().
694 *
695 * If this is an 'ls -l', we want to force use of readdirplus.
696 */
697void nfs_readdir_record_entry_cache_miss(struct inode *dir)
698{
699 struct nfs_inode *nfsi = NFS_I(dir);
700 struct nfs_open_dir_context *ctx;
701
702 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
703 S_ISDIR(dir->i_mode)) {
704 rcu_read_lock();
705 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
706 atomic_inc(&ctx->cache_misses);
707 rcu_read_unlock();
708 }
709}
710
711static void nfs_lookup_advise_force_readdirplus(struct inode *dir,
712 unsigned int flags)
713{
714 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
715 return;
716 if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL))
717 return;
718 nfs_readdir_record_entry_cache_miss(dir);
719}
720
721static
722void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
723 unsigned long dir_verifier)
724{
725 struct qstr filename = QSTR_INIT(entry->name, entry->len);
726 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
727 struct dentry *dentry;
728 struct dentry *alias;
729 struct inode *inode;
730 int status;
731
732 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
733 return;
734 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
735 return;
736 if (filename.len == 0)
737 return;
738 /* Validate that the name doesn't contain any illegal '\0' */
739 if (strnlen(filename.name, filename.len) != filename.len)
740 return;
741 /* ...or '/' */
742 if (strnchr(filename.name, filename.len, '/'))
743 return;
744 if (filename.name[0] == '.') {
745 if (filename.len == 1)
746 return;
747 if (filename.len == 2 && filename.name[1] == '.')
748 return;
749 }
750 filename.hash = full_name_hash(parent, filename.name, filename.len);
751
752 dentry = d_lookup(parent, &filename);
753again:
754 if (!dentry) {
755 dentry = d_alloc_parallel(parent, &filename, &wq);
756 if (IS_ERR(dentry))
757 return;
758 }
759 if (!d_in_lookup(dentry)) {
760 /* Is there a mountpoint here? If so, just exit */
761 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
762 &entry->fattr->fsid))
763 goto out;
764 if (nfs_same_file(dentry, entry)) {
765 if (!entry->fh->size)
766 goto out;
767 nfs_set_verifier(dentry, dir_verifier);
768 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
769 if (!status)
770 nfs_setsecurity(d_inode(dentry), entry->fattr);
771 trace_nfs_readdir_lookup_revalidate(d_inode(parent),
772 dentry, 0, status);
773 goto out;
774 } else {
775 trace_nfs_readdir_lookup_revalidate_failed(
776 d_inode(parent), dentry, 0);
777 d_invalidate(dentry);
778 dput(dentry);
779 dentry = NULL;
780 goto again;
781 }
782 }
783 if (!entry->fh->size) {
784 d_lookup_done(dentry);
785 goto out;
786 }
787
788 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
789 alias = d_splice_alias(inode, dentry);
790 d_lookup_done(dentry);
791 if (alias) {
792 if (IS_ERR(alias))
793 goto out;
794 dput(dentry);
795 dentry = alias;
796 }
797 nfs_set_verifier(dentry, dir_verifier);
798 trace_nfs_readdir_lookup(d_inode(parent), dentry, 0);
799out:
800 dput(dentry);
801}
802
803static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc,
804 struct nfs_entry *entry,
805 struct xdr_stream *stream)
806{
807 int ret;
808
809 if (entry->fattr->label)
810 entry->fattr->label->len = NFS4_MAXLABELLEN;
811 ret = xdr_decode(desc, entry, stream);
812 if (ret || !desc->plus)
813 return ret;
814 nfs_prime_dcache(file_dentry(desc->file), entry, desc->dir_verifier);
815 return 0;
816}
817
818/* Perform conversion from xdr to cache array */
819static int nfs_readdir_folio_filler(struct nfs_readdir_descriptor *desc,
820 struct nfs_entry *entry,
821 struct page **xdr_pages, unsigned int buflen,
822 struct folio **arrays, size_t narrays,
823 u64 change_attr)
824{
825 struct address_space *mapping = desc->file->f_mapping;
826 struct folio *new, *folio = *arrays;
827 struct xdr_stream stream;
828 struct page *scratch;
829 struct xdr_buf buf;
830 u64 cookie;
831 int status;
832
833 scratch = alloc_page(GFP_KERNEL);
834 if (scratch == NULL)
835 return -ENOMEM;
836
837 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
838 xdr_set_scratch_page(&stream, scratch);
839
840 do {
841 status = nfs_readdir_entry_decode(desc, entry, &stream);
842 if (status != 0)
843 break;
844
845 status = nfs_readdir_folio_array_append(folio, entry, &cookie);
846 if (status != -ENOSPC)
847 continue;
848
849 if (folio->mapping != mapping) {
850 if (!--narrays)
851 break;
852 new = nfs_readdir_folio_array_alloc(cookie, GFP_KERNEL);
853 if (!new)
854 break;
855 arrays++;
856 *arrays = folio = new;
857 } else {
858 new = nfs_readdir_folio_get_next(mapping, cookie,
859 change_attr);
860 if (!new)
861 break;
862 if (folio != *arrays)
863 nfs_readdir_folio_unlock_and_put(folio);
864 folio = new;
865 }
866 desc->folio_index_max++;
867 status = nfs_readdir_folio_array_append(folio, entry, &cookie);
868 } while (!status && !entry->eof);
869
870 switch (status) {
871 case -EBADCOOKIE:
872 if (!entry->eof)
873 break;
874 nfs_readdir_folio_set_eof(folio);
875 fallthrough;
876 case -EAGAIN:
877 status = 0;
878 break;
879 case -ENOSPC:
880 status = 0;
881 if (!desc->plus)
882 break;
883 while (!nfs_readdir_entry_decode(desc, entry, &stream))
884 ;
885 }
886
887 if (folio != *arrays)
888 nfs_readdir_folio_unlock_and_put(folio);
889
890 put_page(scratch);
891 return status;
892}
893
894static void nfs_readdir_free_pages(struct page **pages, size_t npages)
895{
896 while (npages--)
897 put_page(pages[npages]);
898 kfree(pages);
899}
900
901/*
902 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
903 * to nfs_readdir_free_pages()
904 */
905static struct page **nfs_readdir_alloc_pages(size_t npages)
906{
907 struct page **pages;
908 size_t i;
909
910 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
911 if (!pages)
912 return NULL;
913 for (i = 0; i < npages; i++) {
914 struct page *page = alloc_page(GFP_KERNEL);
915 if (page == NULL)
916 goto out_freepages;
917 pages[i] = page;
918 }
919 return pages;
920
921out_freepages:
922 nfs_readdir_free_pages(pages, i);
923 return NULL;
924}
925
926static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
927 __be32 *verf_arg, __be32 *verf_res,
928 struct folio **arrays, size_t narrays)
929{
930 u64 change_attr;
931 struct page **pages;
932 struct folio *folio = *arrays;
933 struct nfs_entry *entry;
934 size_t array_size;
935 struct inode *inode = file_inode(desc->file);
936 unsigned int dtsize = desc->dtsize;
937 unsigned int pglen;
938 int status = -ENOMEM;
939
940 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
941 if (!entry)
942 return -ENOMEM;
943 entry->cookie = nfs_readdir_folio_last_cookie(folio);
944 entry->fh = nfs_alloc_fhandle();
945 entry->fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
946 entry->server = NFS_SERVER(inode);
947 if (entry->fh == NULL || entry->fattr == NULL)
948 goto out;
949
950 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
951 pages = nfs_readdir_alloc_pages(array_size);
952 if (!pages)
953 goto out;
954
955 change_attr = inode_peek_iversion_raw(inode);
956 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, pages,
957 dtsize, verf_res);
958 if (status < 0)
959 goto free_pages;
960
961 pglen = status;
962 if (pglen != 0)
963 status = nfs_readdir_folio_filler(desc, entry, pages, pglen,
964 arrays, narrays, change_attr);
965 else
966 nfs_readdir_folio_set_eof(folio);
967 desc->buffer_fills++;
968
969free_pages:
970 nfs_readdir_free_pages(pages, array_size);
971out:
972 nfs_free_fattr(entry->fattr);
973 nfs_free_fhandle(entry->fh);
974 kfree(entry);
975 return status;
976}
977
978static void nfs_readdir_folio_put(struct nfs_readdir_descriptor *desc)
979{
980 folio_put(desc->folio);
981 desc->folio = NULL;
982}
983
984static void
985nfs_readdir_folio_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
986{
987 folio_unlock(desc->folio);
988 nfs_readdir_folio_put(desc);
989}
990
991static struct folio *
992nfs_readdir_folio_get_cached(struct nfs_readdir_descriptor *desc)
993{
994 struct address_space *mapping = desc->file->f_mapping;
995 u64 change_attr = inode_peek_iversion_raw(mapping->host);
996 u64 cookie = desc->last_cookie;
997 struct folio *folio;
998
999 folio = nfs_readdir_folio_get_locked(mapping, cookie, change_attr);
1000 if (!folio)
1001 return NULL;
1002 if (desc->clear_cache && !nfs_readdir_folio_needs_filling(folio))
1003 nfs_readdir_folio_reinit_array(folio, cookie, change_attr);
1004 return folio;
1005}
1006
1007/*
1008 * Returns 0 if desc->dir_cookie was found on page desc->page_index
1009 * and locks the page to prevent removal from the page cache.
1010 */
1011static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
1012{
1013 struct inode *inode = file_inode(desc->file);
1014 struct nfs_inode *nfsi = NFS_I(inode);
1015 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1016 int res;
1017
1018 desc->folio = nfs_readdir_folio_get_cached(desc);
1019 if (!desc->folio)
1020 return -ENOMEM;
1021 if (nfs_readdir_folio_needs_filling(desc->folio)) {
1022 /* Grow the dtsize if we had to go back for more pages */
1023 if (desc->folio_index == desc->folio_index_max)
1024 nfs_grow_dtsize(desc);
1025 desc->folio_index_max = desc->folio_index;
1026 trace_nfs_readdir_cache_fill(desc->file, nfsi->cookieverf,
1027 desc->last_cookie,
1028 desc->folio->index, desc->dtsize);
1029 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
1030 &desc->folio, 1);
1031 if (res < 0) {
1032 nfs_readdir_folio_unlock_and_put_cached(desc);
1033 trace_nfs_readdir_cache_fill_done(inode, res);
1034 if (res == -EBADCOOKIE || res == -ENOTSYNC) {
1035 invalidate_inode_pages2(desc->file->f_mapping);
1036 nfs_readdir_rewind_search(desc);
1037 trace_nfs_readdir_invalidate_cache_range(
1038 inode, 0, MAX_LFS_FILESIZE);
1039 return -EAGAIN;
1040 }
1041 return res;
1042 }
1043 /*
1044 * Set the cookie verifier if the page cache was empty
1045 */
1046 if (desc->last_cookie == 0 &&
1047 memcmp(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf))) {
1048 memcpy(nfsi->cookieverf, verf,
1049 sizeof(nfsi->cookieverf));
1050 invalidate_inode_pages2_range(desc->file->f_mapping, 1,
1051 -1);
1052 trace_nfs_readdir_invalidate_cache_range(
1053 inode, 1, MAX_LFS_FILESIZE);
1054 }
1055 desc->clear_cache = false;
1056 }
1057 res = nfs_readdir_search_array(desc);
1058 if (res == 0)
1059 return 0;
1060 nfs_readdir_folio_unlock_and_put_cached(desc);
1061 return res;
1062}
1063
1064/* Search for desc->dir_cookie from the beginning of the page cache */
1065static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
1066{
1067 int res;
1068
1069 do {
1070 res = find_and_lock_cache_page(desc);
1071 } while (res == -EAGAIN);
1072 return res;
1073}
1074
1075#define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1076
1077/*
1078 * Once we've found the start of the dirent within a page: fill 'er up...
1079 */
1080static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
1081 const __be32 *verf)
1082{
1083 struct file *file = desc->file;
1084 struct nfs_cache_array *array;
1085 unsigned int i;
1086 bool first_emit = !desc->dir_cookie;
1087
1088 array = kmap_local_folio(desc->folio, 0);
1089 for (i = desc->cache_entry_index; i < array->size; i++) {
1090 struct nfs_cache_array_entry *ent;
1091
1092 /*
1093 * nfs_readdir_handle_cache_misses return force clear at
1094 * (cache_misses > NFS_READDIR_CACHE_MISS_THRESHOLD) for
1095 * readdir heuristic, NFS_READDIR_CACHE_MISS_THRESHOLD + 1
1096 * entries need be emitted here.
1097 */
1098 if (first_emit && i > NFS_READDIR_CACHE_MISS_THRESHOLD + 2) {
1099 desc->eob = true;
1100 break;
1101 }
1102
1103 ent = &array->array[i];
1104 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
1105 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
1106 desc->eob = true;
1107 break;
1108 }
1109 memcpy(desc->verf, verf, sizeof(desc->verf));
1110 if (i == array->size - 1) {
1111 desc->dir_cookie = array->last_cookie;
1112 nfs_readdir_seek_next_array(array, desc);
1113 } else {
1114 desc->dir_cookie = array->array[i + 1].cookie;
1115 desc->last_cookie = array->array[0].cookie;
1116 }
1117 if (nfs_readdir_use_cookie(file))
1118 desc->ctx->pos = desc->dir_cookie;
1119 else
1120 desc->ctx->pos++;
1121 }
1122 if (array->folio_is_eof)
1123 desc->eof = !desc->eob;
1124
1125 kunmap_local(array);
1126 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1127 (unsigned long long)desc->dir_cookie);
1128}
1129
1130/*
1131 * If we cannot find a cookie in our cache, we suspect that this is
1132 * because it points to a deleted file, so we ask the server to return
1133 * whatever it thinks is the next entry. We then feed this to filldir.
1134 * If all goes well, we should then be able to find our way round the
1135 * cache on the next call to readdir_search_pagecache();
1136 *
1137 * NOTE: we cannot add the anonymous page to the pagecache because
1138 * the data it contains might not be page aligned. Besides,
1139 * we should already have a complete representation of the
1140 * directory in the page cache by the time we get here.
1141 */
1142static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1143{
1144 struct folio **arrays;
1145 size_t i, sz = 512;
1146 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1147 int status = -ENOMEM;
1148
1149 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1150 (unsigned long long)desc->dir_cookie);
1151
1152 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1153 if (!arrays)
1154 goto out;
1155 arrays[0] = nfs_readdir_folio_array_alloc(desc->dir_cookie, GFP_KERNEL);
1156 if (!arrays[0])
1157 goto out;
1158
1159 desc->folio_index = 0;
1160 desc->cache_entry_index = 0;
1161 desc->last_cookie = desc->dir_cookie;
1162 desc->folio_index_max = 0;
1163
1164 trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie,
1165 -1, desc->dtsize);
1166
1167 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1168 if (status < 0) {
1169 trace_nfs_readdir_uncached_done(file_inode(desc->file), status);
1170 goto out_free;
1171 }
1172
1173 for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1174 desc->folio = arrays[i];
1175 nfs_do_filldir(desc, verf);
1176 }
1177 desc->folio = NULL;
1178
1179 /*
1180 * Grow the dtsize if we have to go back for more pages,
1181 * or shrink it if we're reading too many.
1182 */
1183 if (!desc->eof) {
1184 if (!desc->eob)
1185 nfs_grow_dtsize(desc);
1186 else if (desc->buffer_fills == 1 &&
1187 i < (desc->folio_index_max >> 1))
1188 nfs_shrink_dtsize(desc);
1189 }
1190out_free:
1191 for (i = 0; i < sz && arrays[i]; i++)
1192 nfs_readdir_folio_array_free(arrays[i]);
1193out:
1194 if (!nfs_readdir_use_cookie(desc->file))
1195 nfs_readdir_rewind_search(desc);
1196 desc->folio_index_max = -1;
1197 kfree(arrays);
1198 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1199 return status;
1200}
1201
1202static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1203 struct nfs_readdir_descriptor *desc,
1204 unsigned int cache_misses,
1205 bool force_clear)
1206{
1207 if (desc->ctx->pos == 0 || !desc->plus)
1208 return false;
1209 if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1210 return false;
1211 trace_nfs_readdir_force_readdirplus(inode);
1212 return true;
1213}
1214
1215/* The file offset position represents the dirent entry number. A
1216 last cookie cache takes care of the common case of reading the
1217 whole directory.
1218 */
1219static int nfs_readdir(struct file *file, struct dir_context *ctx)
1220{
1221 struct dentry *dentry = file_dentry(file);
1222 struct inode *inode = d_inode(dentry);
1223 struct nfs_inode *nfsi = NFS_I(inode);
1224 struct nfs_open_dir_context *dir_ctx = file->private_data;
1225 struct nfs_readdir_descriptor *desc;
1226 unsigned int cache_hits, cache_misses;
1227 bool force_clear;
1228 int res;
1229
1230 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1231 file, (long long)ctx->pos);
1232 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1233
1234 /*
1235 * ctx->pos points to the dirent entry number.
1236 * *desc->dir_cookie has the cookie for the next entry. We have
1237 * to either find the entry with the appropriate number or
1238 * revalidate the cookie.
1239 */
1240 nfs_revalidate_mapping(inode, file->f_mapping);
1241
1242 res = -ENOMEM;
1243 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1244 if (!desc)
1245 goto out;
1246 desc->file = file;
1247 desc->ctx = ctx;
1248 desc->folio_index_max = -1;
1249
1250 spin_lock(&file->f_lock);
1251 desc->dir_cookie = dir_ctx->dir_cookie;
1252 desc->folio_index = dir_ctx->page_index;
1253 desc->last_cookie = dir_ctx->last_cookie;
1254 desc->attr_gencount = dir_ctx->attr_gencount;
1255 desc->eof = dir_ctx->eof;
1256 nfs_set_dtsize(desc, dir_ctx->dtsize);
1257 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1258 cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0);
1259 cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0);
1260 force_clear = dir_ctx->force_clear;
1261 spin_unlock(&file->f_lock);
1262
1263 if (desc->eof) {
1264 res = 0;
1265 goto out_free;
1266 }
1267
1268 desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses);
1269 force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1270 force_clear);
1271 desc->clear_cache = force_clear;
1272
1273 do {
1274 res = readdir_search_pagecache(desc);
1275
1276 if (res == -EBADCOOKIE) {
1277 res = 0;
1278 /* This means either end of directory */
1279 if (desc->dir_cookie && !desc->eof) {
1280 /* Or that the server has 'lost' a cookie */
1281 res = uncached_readdir(desc);
1282 if (res == 0)
1283 continue;
1284 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1285 res = 0;
1286 }
1287 break;
1288 }
1289 if (res == -ETOOSMALL && desc->plus) {
1290 nfs_zap_caches(inode);
1291 desc->plus = false;
1292 desc->eof = false;
1293 continue;
1294 }
1295 if (res < 0)
1296 break;
1297
1298 nfs_do_filldir(desc, nfsi->cookieverf);
1299 nfs_readdir_folio_unlock_and_put_cached(desc);
1300 if (desc->folio_index == desc->folio_index_max)
1301 desc->clear_cache = force_clear;
1302 } while (!desc->eob && !desc->eof);
1303
1304 spin_lock(&file->f_lock);
1305 dir_ctx->dir_cookie = desc->dir_cookie;
1306 dir_ctx->last_cookie = desc->last_cookie;
1307 dir_ctx->attr_gencount = desc->attr_gencount;
1308 dir_ctx->page_index = desc->folio_index;
1309 dir_ctx->force_clear = force_clear;
1310 dir_ctx->eof = desc->eof;
1311 dir_ctx->dtsize = desc->dtsize;
1312 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1313 spin_unlock(&file->f_lock);
1314out_free:
1315 kfree(desc);
1316
1317out:
1318 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1319 return res;
1320}
1321
1322static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1323{
1324 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1325
1326 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1327 filp, offset, whence);
1328
1329 switch (whence) {
1330 default:
1331 return -EINVAL;
1332 case SEEK_SET:
1333 if (offset < 0)
1334 return -EINVAL;
1335 spin_lock(&filp->f_lock);
1336 break;
1337 case SEEK_CUR:
1338 if (offset == 0)
1339 return filp->f_pos;
1340 spin_lock(&filp->f_lock);
1341 offset += filp->f_pos;
1342 if (offset < 0) {
1343 spin_unlock(&filp->f_lock);
1344 return -EINVAL;
1345 }
1346 }
1347 if (offset != filp->f_pos) {
1348 filp->f_pos = offset;
1349 dir_ctx->page_index = 0;
1350 if (!nfs_readdir_use_cookie(filp)) {
1351 dir_ctx->dir_cookie = 0;
1352 dir_ctx->last_cookie = 0;
1353 } else {
1354 dir_ctx->dir_cookie = offset;
1355 dir_ctx->last_cookie = offset;
1356 }
1357 dir_ctx->eof = false;
1358 }
1359 spin_unlock(&filp->f_lock);
1360 return offset;
1361}
1362
1363/*
1364 * All directory operations under NFS are synchronous, so fsync()
1365 * is a dummy operation.
1366 */
1367static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1368 int datasync)
1369{
1370 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1371
1372 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1373 return 0;
1374}
1375
1376/**
1377 * nfs_force_lookup_revalidate - Mark the directory as having changed
1378 * @dir: pointer to directory inode
1379 *
1380 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1381 * full lookup on all child dentries of 'dir' whenever a change occurs
1382 * on the server that might have invalidated our dcache.
1383 *
1384 * Note that we reserve bit '0' as a tag to let us know when a dentry
1385 * was revalidated while holding a delegation on its inode.
1386 *
1387 * The caller should be holding dir->i_lock
1388 */
1389void nfs_force_lookup_revalidate(struct inode *dir)
1390{
1391 NFS_I(dir)->cache_change_attribute += 2;
1392}
1393EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1394
1395/**
1396 * nfs_verify_change_attribute - Detects NFS remote directory changes
1397 * @dir: pointer to parent directory inode
1398 * @verf: previously saved change attribute
1399 *
1400 * Return "false" if the verifiers doesn't match the change attribute.
1401 * This would usually indicate that the directory contents have changed on
1402 * the server, and that any dentries need revalidating.
1403 */
1404static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1405{
1406 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1407}
1408
1409static void nfs_set_verifier_delegated(unsigned long *verf)
1410{
1411 *verf |= 1UL;
1412}
1413
1414#if IS_ENABLED(CONFIG_NFS_V4)
1415static void nfs_unset_verifier_delegated(unsigned long *verf)
1416{
1417 *verf &= ~1UL;
1418}
1419#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1420
1421static bool nfs_test_verifier_delegated(unsigned long verf)
1422{
1423 return verf & 1;
1424}
1425
1426static bool nfs_verifier_is_delegated(struct dentry *dentry)
1427{
1428 return nfs_test_verifier_delegated(dentry->d_time);
1429}
1430
1431static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1432{
1433 struct inode *inode = d_inode(dentry);
1434 struct inode *dir = d_inode_rcu(dentry->d_parent);
1435
1436 if (!dir || !nfs_verify_change_attribute(dir, verf))
1437 return;
1438 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1439 nfs_set_verifier_delegated(&verf);
1440 dentry->d_time = verf;
1441}
1442
1443/**
1444 * nfs_set_verifier - save a parent directory verifier in the dentry
1445 * @dentry: pointer to dentry
1446 * @verf: verifier to save
1447 *
1448 * Saves the parent directory verifier in @dentry. If the inode has
1449 * a delegation, we also tag the dentry as having been revalidated
1450 * while holding a delegation so that we know we don't have to
1451 * look it up again after a directory change.
1452 */
1453void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1454{
1455
1456 spin_lock(&dentry->d_lock);
1457 nfs_set_verifier_locked(dentry, verf);
1458 spin_unlock(&dentry->d_lock);
1459}
1460EXPORT_SYMBOL_GPL(nfs_set_verifier);
1461
1462#if IS_ENABLED(CONFIG_NFS_V4)
1463/**
1464 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1465 * @inode: pointer to inode
1466 *
1467 * Iterates through the dentries in the inode alias list and clears
1468 * the tag used to indicate that the dentry has been revalidated
1469 * while holding a delegation.
1470 * This function is intended for use when the delegation is being
1471 * returned or revoked.
1472 */
1473void nfs_clear_verifier_delegated(struct inode *inode)
1474{
1475 struct dentry *alias;
1476
1477 if (!inode)
1478 return;
1479 spin_lock(&inode->i_lock);
1480 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1481 spin_lock(&alias->d_lock);
1482 nfs_unset_verifier_delegated(&alias->d_time);
1483 spin_unlock(&alias->d_lock);
1484 }
1485 spin_unlock(&inode->i_lock);
1486}
1487EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1488#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1489
1490static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1491{
1492 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) &&
1493 d_really_is_negative(dentry))
1494 return dentry->d_time == inode_peek_iversion_raw(dir);
1495 return nfs_verify_change_attribute(dir, dentry->d_time);
1496}
1497
1498/*
1499 * A check for whether or not the parent directory has changed.
1500 * In the case it has, we assume that the dentries are untrustworthy
1501 * and may need to be looked up again.
1502 * If rcu_walk prevents us from performing a full check, return 0.
1503 */
1504static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1505 int rcu_walk)
1506{
1507 if (IS_ROOT(dentry))
1508 return 1;
1509 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1510 return 0;
1511 if (!nfs_dentry_verify_change(dir, dentry))
1512 return 0;
1513 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1514 if (nfs_mapping_need_revalidate_inode(dir)) {
1515 if (rcu_walk)
1516 return 0;
1517 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1518 return 0;
1519 }
1520 if (!nfs_dentry_verify_change(dir, dentry))
1521 return 0;
1522 return 1;
1523}
1524
1525/*
1526 * Use intent information to check whether or not we're going to do
1527 * an O_EXCL create using this path component.
1528 */
1529static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1530{
1531 if (NFS_PROTO(dir)->version == 2)
1532 return 0;
1533 return flags & LOOKUP_EXCL;
1534}
1535
1536/*
1537 * Inode and filehandle revalidation for lookups.
1538 *
1539 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1540 * or if the intent information indicates that we're about to open this
1541 * particular file and the "nocto" mount flag is not set.
1542 *
1543 */
1544static
1545int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1546{
1547 struct nfs_server *server = NFS_SERVER(inode);
1548 int ret;
1549
1550 if (IS_AUTOMOUNT(inode))
1551 return 0;
1552
1553 if (flags & LOOKUP_OPEN) {
1554 switch (inode->i_mode & S_IFMT) {
1555 case S_IFREG:
1556 /* A NFSv4 OPEN will revalidate later */
1557 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1558 goto out;
1559 fallthrough;
1560 case S_IFDIR:
1561 if (server->flags & NFS_MOUNT_NOCTO)
1562 break;
1563 /* NFS close-to-open cache consistency validation */
1564 goto out_force;
1565 }
1566 }
1567
1568 /* VFS wants an on-the-wire revalidation */
1569 if (flags & LOOKUP_REVAL)
1570 goto out_force;
1571out:
1572 if (inode->i_nlink > 0 ||
1573 (inode->i_nlink == 0 &&
1574 test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1575 return 0;
1576 else
1577 return -ESTALE;
1578out_force:
1579 if (flags & LOOKUP_RCU)
1580 return -ECHILD;
1581 ret = __nfs_revalidate_inode(server, inode);
1582 if (ret != 0)
1583 return ret;
1584 goto out;
1585}
1586
1587static void nfs_mark_dir_for_revalidate(struct inode *inode)
1588{
1589 spin_lock(&inode->i_lock);
1590 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1591 spin_unlock(&inode->i_lock);
1592}
1593
1594/*
1595 * We judge how long we want to trust negative
1596 * dentries by looking at the parent inode mtime.
1597 *
1598 * If parent mtime has changed, we revalidate, else we wait for a
1599 * period corresponding to the parent's attribute cache timeout value.
1600 *
1601 * If LOOKUP_RCU prevents us from performing a full check, return 1
1602 * suggesting a reval is needed.
1603 *
1604 * Note that when creating a new file, or looking up a rename target,
1605 * then it shouldn't be necessary to revalidate a negative dentry.
1606 */
1607static inline
1608int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1609 unsigned int flags)
1610{
1611 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1612 return 0;
1613 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1614 return 1;
1615 /* Case insensitive server? Revalidate negative dentries */
1616 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1617 return 1;
1618 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1619}
1620
1621static int
1622nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1623 struct inode *inode, int error)
1624{
1625 switch (error) {
1626 case 1:
1627 break;
1628 case 0:
1629 /*
1630 * We can't d_drop the root of a disconnected tree:
1631 * its d_hash is on the s_anon list and d_drop() would hide
1632 * it from shrink_dcache_for_unmount(), leading to busy
1633 * inodes on unmount and further oopses.
1634 */
1635 if (inode && IS_ROOT(dentry))
1636 error = 1;
1637 break;
1638 }
1639 trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error);
1640 return error;
1641}
1642
1643static int
1644nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1645 unsigned int flags)
1646{
1647 int ret = 1;
1648 if (nfs_neg_need_reval(dir, dentry, flags)) {
1649 if (flags & LOOKUP_RCU)
1650 return -ECHILD;
1651 ret = 0;
1652 }
1653 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1654}
1655
1656static int
1657nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1658 struct inode *inode)
1659{
1660 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1661 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1662}
1663
1664static int nfs_lookup_revalidate_dentry(struct inode *dir,
1665 struct dentry *dentry,
1666 struct inode *inode, unsigned int flags)
1667{
1668 struct nfs_fh *fhandle;
1669 struct nfs_fattr *fattr;
1670 unsigned long dir_verifier;
1671 int ret;
1672
1673 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1674
1675 ret = -ENOMEM;
1676 fhandle = nfs_alloc_fhandle();
1677 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1678 if (fhandle == NULL || fattr == NULL)
1679 goto out;
1680
1681 dir_verifier = nfs_save_change_attribute(dir);
1682 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1683 if (ret < 0) {
1684 switch (ret) {
1685 case -ESTALE:
1686 case -ENOENT:
1687 ret = 0;
1688 break;
1689 case -ETIMEDOUT:
1690 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1691 ret = 1;
1692 }
1693 goto out;
1694 }
1695
1696 /* Request help from readdirplus */
1697 nfs_lookup_advise_force_readdirplus(dir, flags);
1698
1699 ret = 0;
1700 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1701 goto out;
1702 if (nfs_refresh_inode(inode, fattr) < 0)
1703 goto out;
1704
1705 nfs_setsecurity(inode, fattr);
1706 nfs_set_verifier(dentry, dir_verifier);
1707
1708 ret = 1;
1709out:
1710 nfs_free_fattr(fattr);
1711 nfs_free_fhandle(fhandle);
1712
1713 /*
1714 * If the lookup failed despite the dentry change attribute being
1715 * a match, then we should revalidate the directory cache.
1716 */
1717 if (!ret && nfs_dentry_verify_change(dir, dentry))
1718 nfs_mark_dir_for_revalidate(dir);
1719 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1720}
1721
1722/*
1723 * This is called every time the dcache has a lookup hit,
1724 * and we should check whether we can really trust that
1725 * lookup.
1726 *
1727 * NOTE! The hit can be a negative hit too, don't assume
1728 * we have an inode!
1729 *
1730 * If the parent directory is seen to have changed, we throw out the
1731 * cached dentry and do a new lookup.
1732 */
1733static int
1734nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1735 unsigned int flags)
1736{
1737 struct inode *inode;
1738 int error;
1739
1740 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1741 inode = d_inode(dentry);
1742
1743 if (!inode)
1744 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1745
1746 if (is_bad_inode(inode)) {
1747 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1748 __func__, dentry);
1749 goto out_bad;
1750 }
1751
1752 if ((flags & LOOKUP_RENAME_TARGET) && d_count(dentry) < 2 &&
1753 nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1754 goto out_bad;
1755
1756 if (nfs_verifier_is_delegated(dentry))
1757 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1758
1759 /* Force a full look up iff the parent directory has changed */
1760 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1761 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1762 error = nfs_lookup_verify_inode(inode, flags);
1763 if (error) {
1764 if (error == -ESTALE)
1765 nfs_mark_dir_for_revalidate(dir);
1766 goto out_bad;
1767 }
1768 goto out_valid;
1769 }
1770
1771 if (flags & LOOKUP_RCU)
1772 return -ECHILD;
1773
1774 if (NFS_STALE(inode))
1775 goto out_bad;
1776
1777 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
1778out_valid:
1779 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1780out_bad:
1781 if (flags & LOOKUP_RCU)
1782 return -ECHILD;
1783 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1784}
1785
1786static int
1787__nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1788 int (*reval)(struct inode *, struct dentry *, unsigned int))
1789{
1790 struct dentry *parent;
1791 struct inode *dir;
1792 int ret;
1793
1794 if (flags & LOOKUP_RCU) {
1795 if (dentry->d_fsdata == NFS_FSDATA_BLOCKED)
1796 return -ECHILD;
1797 parent = READ_ONCE(dentry->d_parent);
1798 dir = d_inode_rcu(parent);
1799 if (!dir)
1800 return -ECHILD;
1801 ret = reval(dir, dentry, flags);
1802 if (parent != READ_ONCE(dentry->d_parent))
1803 return -ECHILD;
1804 } else {
1805 /* Wait for unlink to complete */
1806 wait_var_event(&dentry->d_fsdata,
1807 dentry->d_fsdata != NFS_FSDATA_BLOCKED);
1808 parent = dget_parent(dentry);
1809 ret = reval(d_inode(parent), dentry, flags);
1810 dput(parent);
1811 }
1812 return ret;
1813}
1814
1815static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1816{
1817 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1818}
1819
1820/*
1821 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1822 * when we don't really care about the dentry name. This is called when a
1823 * pathwalk ends on a dentry that was not found via a normal lookup in the
1824 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1825 *
1826 * In this situation, we just want to verify that the inode itself is OK
1827 * since the dentry might have changed on the server.
1828 */
1829static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1830{
1831 struct inode *inode = d_inode(dentry);
1832 int error = 0;
1833
1834 /*
1835 * I believe we can only get a negative dentry here in the case of a
1836 * procfs-style symlink. Just assume it's correct for now, but we may
1837 * eventually need to do something more here.
1838 */
1839 if (!inode) {
1840 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1841 __func__, dentry);
1842 return 1;
1843 }
1844
1845 if (is_bad_inode(inode)) {
1846 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1847 __func__, dentry);
1848 return 0;
1849 }
1850
1851 error = nfs_lookup_verify_inode(inode, flags);
1852 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1853 __func__, inode->i_ino, error ? "invalid" : "valid");
1854 return !error;
1855}
1856
1857/*
1858 * This is called from dput() when d_count is going to 0.
1859 */
1860static int nfs_dentry_delete(const struct dentry *dentry)
1861{
1862 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1863 dentry, dentry->d_flags);
1864
1865 /* Unhash any dentry with a stale inode */
1866 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1867 return 1;
1868
1869 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1870 /* Unhash it, so that ->d_iput() would be called */
1871 return 1;
1872 }
1873 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1874 /* Unhash it, so that ancestors of killed async unlink
1875 * files will be cleaned up during umount */
1876 return 1;
1877 }
1878 return 0;
1879
1880}
1881
1882/* Ensure that we revalidate inode->i_nlink */
1883static void nfs_drop_nlink(struct inode *inode)
1884{
1885 spin_lock(&inode->i_lock);
1886 /* drop the inode if we're reasonably sure this is the last link */
1887 if (inode->i_nlink > 0)
1888 drop_nlink(inode);
1889 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1890 nfs_set_cache_invalid(
1891 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1892 NFS_INO_INVALID_NLINK);
1893 spin_unlock(&inode->i_lock);
1894}
1895
1896/*
1897 * Called when the dentry loses inode.
1898 * We use it to clean up silly-renamed files.
1899 */
1900static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1901{
1902 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1903 nfs_complete_unlink(dentry, inode);
1904 nfs_drop_nlink(inode);
1905 }
1906 iput(inode);
1907}
1908
1909static void nfs_d_release(struct dentry *dentry)
1910{
1911 /* free cached devname value, if it survived that far */
1912 if (unlikely(dentry->d_fsdata)) {
1913 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1914 WARN_ON(1);
1915 else
1916 kfree(dentry->d_fsdata);
1917 }
1918}
1919
1920const struct dentry_operations nfs_dentry_operations = {
1921 .d_revalidate = nfs_lookup_revalidate,
1922 .d_weak_revalidate = nfs_weak_revalidate,
1923 .d_delete = nfs_dentry_delete,
1924 .d_iput = nfs_dentry_iput,
1925 .d_automount = nfs_d_automount,
1926 .d_release = nfs_d_release,
1927};
1928EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1929
1930struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1931{
1932 struct dentry *res;
1933 struct inode *inode = NULL;
1934 struct nfs_fh *fhandle = NULL;
1935 struct nfs_fattr *fattr = NULL;
1936 unsigned long dir_verifier;
1937 int error;
1938
1939 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1940 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1941
1942 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1943 return ERR_PTR(-ENAMETOOLONG);
1944
1945 /*
1946 * If we're doing an exclusive create, optimize away the lookup
1947 * but don't hash the dentry.
1948 */
1949 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1950 return NULL;
1951
1952 res = ERR_PTR(-ENOMEM);
1953 fhandle = nfs_alloc_fhandle();
1954 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir));
1955 if (fhandle == NULL || fattr == NULL)
1956 goto out;
1957
1958 dir_verifier = nfs_save_change_attribute(dir);
1959 trace_nfs_lookup_enter(dir, dentry, flags);
1960 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1961 if (error == -ENOENT) {
1962 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1963 dir_verifier = inode_peek_iversion_raw(dir);
1964 goto no_entry;
1965 }
1966 if (error < 0) {
1967 res = ERR_PTR(error);
1968 goto out;
1969 }
1970 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1971 res = ERR_CAST(inode);
1972 if (IS_ERR(res))
1973 goto out;
1974
1975 /* Notify readdir to use READDIRPLUS */
1976 nfs_lookup_advise_force_readdirplus(dir, flags);
1977
1978no_entry:
1979 res = d_splice_alias(inode, dentry);
1980 if (res != NULL) {
1981 if (IS_ERR(res))
1982 goto out;
1983 dentry = res;
1984 }
1985 nfs_set_verifier(dentry, dir_verifier);
1986out:
1987 trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res));
1988 nfs_free_fattr(fattr);
1989 nfs_free_fhandle(fhandle);
1990 return res;
1991}
1992EXPORT_SYMBOL_GPL(nfs_lookup);
1993
1994void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
1995{
1996 /* Case insensitive server? Revalidate dentries */
1997 if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
1998 d_prune_aliases(inode);
1999}
2000EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
2001
2002#if IS_ENABLED(CONFIG_NFS_V4)
2003static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
2004
2005const struct dentry_operations nfs4_dentry_operations = {
2006 .d_revalidate = nfs4_lookup_revalidate,
2007 .d_weak_revalidate = nfs_weak_revalidate,
2008 .d_delete = nfs_dentry_delete,
2009 .d_iput = nfs_dentry_iput,
2010 .d_automount = nfs_d_automount,
2011 .d_release = nfs_d_release,
2012};
2013EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
2014
2015static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
2016{
2017 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
2018}
2019
2020static int do_open(struct inode *inode, struct file *filp)
2021{
2022 nfs_fscache_open_file(inode, filp);
2023 return 0;
2024}
2025
2026static int nfs_finish_open(struct nfs_open_context *ctx,
2027 struct dentry *dentry,
2028 struct file *file, unsigned open_flags)
2029{
2030 int err;
2031
2032 err = finish_open(file, dentry, do_open);
2033 if (err)
2034 goto out;
2035 if (S_ISREG(file_inode(file)->i_mode))
2036 nfs_file_set_open_context(file, ctx);
2037 else
2038 err = -EOPENSTALE;
2039out:
2040 return err;
2041}
2042
2043int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2044 struct file *file, unsigned open_flags,
2045 umode_t mode)
2046{
2047 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2048 struct nfs_open_context *ctx;
2049 struct dentry *res;
2050 struct iattr attr = { .ia_valid = ATTR_OPEN };
2051 struct inode *inode;
2052 unsigned int lookup_flags = 0;
2053 unsigned long dir_verifier;
2054 bool switched = false;
2055 int created = 0;
2056 int err;
2057
2058 /* Expect a negative dentry */
2059 BUG_ON(d_inode(dentry));
2060
2061 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2062 dir->i_sb->s_id, dir->i_ino, dentry);
2063
2064 err = nfs_check_flags(open_flags);
2065 if (err)
2066 return err;
2067
2068 /* NFS only supports OPEN on regular files */
2069 if ((open_flags & O_DIRECTORY)) {
2070 if (!d_in_lookup(dentry)) {
2071 /*
2072 * Hashed negative dentry with O_DIRECTORY: dentry was
2073 * revalidated and is fine, no need to perform lookup
2074 * again
2075 */
2076 return -ENOENT;
2077 }
2078 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2079 goto no_open;
2080 }
2081
2082 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
2083 return -ENAMETOOLONG;
2084
2085 if (open_flags & O_CREAT) {
2086 struct nfs_server *server = NFS_SERVER(dir);
2087
2088 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2089 mode &= ~current_umask();
2090
2091 attr.ia_valid |= ATTR_MODE;
2092 attr.ia_mode = mode;
2093 }
2094 if (open_flags & O_TRUNC) {
2095 attr.ia_valid |= ATTR_SIZE;
2096 attr.ia_size = 0;
2097 }
2098
2099 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2100 d_drop(dentry);
2101 switched = true;
2102 dentry = d_alloc_parallel(dentry->d_parent,
2103 &dentry->d_name, &wq);
2104 if (IS_ERR(dentry))
2105 return PTR_ERR(dentry);
2106 if (unlikely(!d_in_lookup(dentry)))
2107 return finish_no_open(file, dentry);
2108 }
2109
2110 ctx = create_nfs_open_context(dentry, open_flags, file);
2111 err = PTR_ERR(ctx);
2112 if (IS_ERR(ctx))
2113 goto out;
2114
2115 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
2116 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
2117 if (created)
2118 file->f_mode |= FMODE_CREATED;
2119 if (IS_ERR(inode)) {
2120 err = PTR_ERR(inode);
2121 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2122 put_nfs_open_context(ctx);
2123 d_drop(dentry);
2124 switch (err) {
2125 case -ENOENT:
2126 d_splice_alias(NULL, dentry);
2127 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
2128 dir_verifier = inode_peek_iversion_raw(dir);
2129 else
2130 dir_verifier = nfs_save_change_attribute(dir);
2131 nfs_set_verifier(dentry, dir_verifier);
2132 break;
2133 case -EISDIR:
2134 case -ENOTDIR:
2135 goto no_open;
2136 case -ELOOP:
2137 if (!(open_flags & O_NOFOLLOW))
2138 goto no_open;
2139 break;
2140 /* case -EINVAL: */
2141 default:
2142 break;
2143 }
2144 goto out;
2145 }
2146 file->f_mode |= FMODE_CAN_ODIRECT;
2147
2148 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
2149 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2150 put_nfs_open_context(ctx);
2151out:
2152 if (unlikely(switched)) {
2153 d_lookup_done(dentry);
2154 dput(dentry);
2155 }
2156 return err;
2157
2158no_open:
2159 res = nfs_lookup(dir, dentry, lookup_flags);
2160 if (!res) {
2161 inode = d_inode(dentry);
2162 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2163 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2164 res = ERR_PTR(-ENOTDIR);
2165 else if (inode && S_ISREG(inode->i_mode))
2166 res = ERR_PTR(-EOPENSTALE);
2167 } else if (!IS_ERR(res)) {
2168 inode = d_inode(res);
2169 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2170 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2171 dput(res);
2172 res = ERR_PTR(-ENOTDIR);
2173 } else if (inode && S_ISREG(inode->i_mode)) {
2174 dput(res);
2175 res = ERR_PTR(-EOPENSTALE);
2176 }
2177 }
2178 if (switched) {
2179 d_lookup_done(dentry);
2180 if (!res)
2181 res = dentry;
2182 else
2183 dput(dentry);
2184 }
2185 if (IS_ERR(res))
2186 return PTR_ERR(res);
2187 return finish_no_open(file, res);
2188}
2189EXPORT_SYMBOL_GPL(nfs_atomic_open);
2190
2191static int
2192nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2193 unsigned int flags)
2194{
2195 struct inode *inode;
2196
2197 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
2198
2199 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2200 goto full_reval;
2201 if (d_mountpoint(dentry))
2202 goto full_reval;
2203
2204 inode = d_inode(dentry);
2205
2206 /* We can't create new files in nfs_open_revalidate(), so we
2207 * optimize away revalidation of negative dentries.
2208 */
2209 if (inode == NULL)
2210 goto full_reval;
2211
2212 if (nfs_verifier_is_delegated(dentry))
2213 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2214
2215 /* NFS only supports OPEN on regular files */
2216 if (!S_ISREG(inode->i_mode))
2217 goto full_reval;
2218
2219 /* We cannot do exclusive creation on a positive dentry */
2220 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2221 goto reval_dentry;
2222
2223 /* Check if the directory changed */
2224 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2225 goto reval_dentry;
2226
2227 /* Let f_op->open() actually open (and revalidate) the file */
2228 return 1;
2229reval_dentry:
2230 if (flags & LOOKUP_RCU)
2231 return -ECHILD;
2232 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
2233
2234full_reval:
2235 return nfs_do_lookup_revalidate(dir, dentry, flags);
2236}
2237
2238static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2239{
2240 return __nfs_lookup_revalidate(dentry, flags,
2241 nfs4_do_lookup_revalidate);
2242}
2243
2244#endif /* CONFIG_NFSV4 */
2245
2246struct dentry *
2247nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2248 struct nfs_fattr *fattr)
2249{
2250 struct dentry *parent = dget_parent(dentry);
2251 struct inode *dir = d_inode(parent);
2252 struct inode *inode;
2253 struct dentry *d;
2254 int error;
2255
2256 d_drop(dentry);
2257
2258 if (fhandle->size == 0) {
2259 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
2260 if (error)
2261 goto out_error;
2262 }
2263 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2264 if (!(fattr->valid & NFS_ATTR_FATTR)) {
2265 struct nfs_server *server = NFS_SB(dentry->d_sb);
2266 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2267 fattr, NULL);
2268 if (error < 0)
2269 goto out_error;
2270 }
2271 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2272 d = d_splice_alias(inode, dentry);
2273out:
2274 dput(parent);
2275 return d;
2276out_error:
2277 d = ERR_PTR(error);
2278 goto out;
2279}
2280EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2281
2282/*
2283 * Code common to create, mkdir, and mknod.
2284 */
2285int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2286 struct nfs_fattr *fattr)
2287{
2288 struct dentry *d;
2289
2290 d = nfs_add_or_obtain(dentry, fhandle, fattr);
2291 if (IS_ERR(d))
2292 return PTR_ERR(d);
2293
2294 /* Callers don't care */
2295 dput(d);
2296 return 0;
2297}
2298EXPORT_SYMBOL_GPL(nfs_instantiate);
2299
2300/*
2301 * Following a failed create operation, we drop the dentry rather
2302 * than retain a negative dentry. This avoids a problem in the event
2303 * that the operation succeeded on the server, but an error in the
2304 * reply path made it appear to have failed.
2305 */
2306int nfs_create(struct mnt_idmap *idmap, struct inode *dir,
2307 struct dentry *dentry, umode_t mode, bool excl)
2308{
2309 struct iattr attr;
2310 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2311 int error;
2312
2313 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2314 dir->i_sb->s_id, dir->i_ino, dentry);
2315
2316 attr.ia_mode = mode;
2317 attr.ia_valid = ATTR_MODE;
2318
2319 trace_nfs_create_enter(dir, dentry, open_flags);
2320 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2321 trace_nfs_create_exit(dir, dentry, open_flags, error);
2322 if (error != 0)
2323 goto out_err;
2324 return 0;
2325out_err:
2326 d_drop(dentry);
2327 return error;
2328}
2329EXPORT_SYMBOL_GPL(nfs_create);
2330
2331/*
2332 * See comments for nfs_proc_create regarding failed operations.
2333 */
2334int
2335nfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
2336 struct dentry *dentry, umode_t mode, dev_t rdev)
2337{
2338 struct iattr attr;
2339 int status;
2340
2341 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2342 dir->i_sb->s_id, dir->i_ino, dentry);
2343
2344 attr.ia_mode = mode;
2345 attr.ia_valid = ATTR_MODE;
2346
2347 trace_nfs_mknod_enter(dir, dentry);
2348 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2349 trace_nfs_mknod_exit(dir, dentry, status);
2350 if (status != 0)
2351 goto out_err;
2352 return 0;
2353out_err:
2354 d_drop(dentry);
2355 return status;
2356}
2357EXPORT_SYMBOL_GPL(nfs_mknod);
2358
2359/*
2360 * See comments for nfs_proc_create regarding failed operations.
2361 */
2362int nfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2363 struct dentry *dentry, umode_t mode)
2364{
2365 struct iattr attr;
2366 int error;
2367
2368 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2369 dir->i_sb->s_id, dir->i_ino, dentry);
2370
2371 attr.ia_valid = ATTR_MODE;
2372 attr.ia_mode = mode | S_IFDIR;
2373
2374 trace_nfs_mkdir_enter(dir, dentry);
2375 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2376 trace_nfs_mkdir_exit(dir, dentry, error);
2377 if (error != 0)
2378 goto out_err;
2379 return 0;
2380out_err:
2381 d_drop(dentry);
2382 return error;
2383}
2384EXPORT_SYMBOL_GPL(nfs_mkdir);
2385
2386static void nfs_dentry_handle_enoent(struct dentry *dentry)
2387{
2388 if (simple_positive(dentry))
2389 d_delete(dentry);
2390}
2391
2392static void nfs_dentry_remove_handle_error(struct inode *dir,
2393 struct dentry *dentry, int error)
2394{
2395 switch (error) {
2396 case -ENOENT:
2397 if (d_really_is_positive(dentry))
2398 d_delete(dentry);
2399 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2400 break;
2401 case 0:
2402 nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2403 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2404 }
2405}
2406
2407int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2408{
2409 int error;
2410
2411 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2412 dir->i_sb->s_id, dir->i_ino, dentry);
2413
2414 trace_nfs_rmdir_enter(dir, dentry);
2415 if (d_really_is_positive(dentry)) {
2416 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2417 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2418 /* Ensure the VFS deletes this inode */
2419 switch (error) {
2420 case 0:
2421 clear_nlink(d_inode(dentry));
2422 break;
2423 case -ENOENT:
2424 nfs_dentry_handle_enoent(dentry);
2425 }
2426 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2427 } else
2428 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2429 nfs_dentry_remove_handle_error(dir, dentry, error);
2430 trace_nfs_rmdir_exit(dir, dentry, error);
2431
2432 return error;
2433}
2434EXPORT_SYMBOL_GPL(nfs_rmdir);
2435
2436/*
2437 * Remove a file after making sure there are no pending writes,
2438 * and after checking that the file has only one user.
2439 *
2440 * We invalidate the attribute cache and free the inode prior to the operation
2441 * to avoid possible races if the server reuses the inode.
2442 */
2443static int nfs_safe_remove(struct dentry *dentry)
2444{
2445 struct inode *dir = d_inode(dentry->d_parent);
2446 struct inode *inode = d_inode(dentry);
2447 int error = -EBUSY;
2448
2449 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2450
2451 /* If the dentry was sillyrenamed, we simply call d_delete() */
2452 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2453 error = 0;
2454 goto out;
2455 }
2456
2457 trace_nfs_remove_enter(dir, dentry);
2458 if (inode != NULL) {
2459 error = NFS_PROTO(dir)->remove(dir, dentry);
2460 if (error == 0)
2461 nfs_drop_nlink(inode);
2462 } else
2463 error = NFS_PROTO(dir)->remove(dir, dentry);
2464 if (error == -ENOENT)
2465 nfs_dentry_handle_enoent(dentry);
2466 trace_nfs_remove_exit(dir, dentry, error);
2467out:
2468 return error;
2469}
2470
2471/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2472 * belongs to an active ".nfs..." file and we return -EBUSY.
2473 *
2474 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2475 */
2476int nfs_unlink(struct inode *dir, struct dentry *dentry)
2477{
2478 int error;
2479
2480 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2481 dir->i_ino, dentry);
2482
2483 trace_nfs_unlink_enter(dir, dentry);
2484 spin_lock(&dentry->d_lock);
2485 if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2486 &NFS_I(d_inode(dentry))->flags)) {
2487 spin_unlock(&dentry->d_lock);
2488 /* Start asynchronous writeout of the inode */
2489 write_inode_now(d_inode(dentry), 0);
2490 error = nfs_sillyrename(dir, dentry);
2491 goto out;
2492 }
2493 /* We must prevent any concurrent open until the unlink
2494 * completes. ->d_revalidate will wait for ->d_fsdata
2495 * to clear. We set it here to ensure no lookup succeeds until
2496 * the unlink is complete on the server.
2497 */
2498 error = -ETXTBSY;
2499 if (WARN_ON(dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2500 WARN_ON(dentry->d_fsdata == NFS_FSDATA_BLOCKED)) {
2501 spin_unlock(&dentry->d_lock);
2502 goto out;
2503 }
2504 /* old devname */
2505 kfree(dentry->d_fsdata);
2506 dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2507
2508 spin_unlock(&dentry->d_lock);
2509 error = nfs_safe_remove(dentry);
2510 nfs_dentry_remove_handle_error(dir, dentry, error);
2511 dentry->d_fsdata = NULL;
2512 wake_up_var(&dentry->d_fsdata);
2513out:
2514 trace_nfs_unlink_exit(dir, dentry, error);
2515 return error;
2516}
2517EXPORT_SYMBOL_GPL(nfs_unlink);
2518
2519/*
2520 * To create a symbolic link, most file systems instantiate a new inode,
2521 * add a page to it containing the path, then write it out to the disk
2522 * using prepare_write/commit_write.
2523 *
2524 * Unfortunately the NFS client can't create the in-core inode first
2525 * because it needs a file handle to create an in-core inode (see
2526 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2527 * symlink request has completed on the server.
2528 *
2529 * So instead we allocate a raw page, copy the symname into it, then do
2530 * the SYMLINK request with the page as the buffer. If it succeeds, we
2531 * now have a new file handle and can instantiate an in-core NFS inode
2532 * and move the raw page into its mapping.
2533 */
2534int nfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
2535 struct dentry *dentry, const char *symname)
2536{
2537 struct folio *folio;
2538 char *kaddr;
2539 struct iattr attr;
2540 unsigned int pathlen = strlen(symname);
2541 int error;
2542
2543 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2544 dir->i_ino, dentry, symname);
2545
2546 if (pathlen > PAGE_SIZE)
2547 return -ENAMETOOLONG;
2548
2549 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2550 attr.ia_valid = ATTR_MODE;
2551
2552 folio = folio_alloc(GFP_USER, 0);
2553 if (!folio)
2554 return -ENOMEM;
2555
2556 kaddr = folio_address(folio);
2557 memcpy(kaddr, symname, pathlen);
2558 if (pathlen < PAGE_SIZE)
2559 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2560
2561 trace_nfs_symlink_enter(dir, dentry);
2562 error = NFS_PROTO(dir)->symlink(dir, dentry, folio, pathlen, &attr);
2563 trace_nfs_symlink_exit(dir, dentry, error);
2564 if (error != 0) {
2565 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2566 dir->i_sb->s_id, dir->i_ino,
2567 dentry, symname, error);
2568 d_drop(dentry);
2569 folio_put(folio);
2570 return error;
2571 }
2572
2573 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2574
2575 /*
2576 * No big deal if we can't add this page to the page cache here.
2577 * READLINK will get the missing page from the server if needed.
2578 */
2579 if (filemap_add_folio(d_inode(dentry)->i_mapping, folio, 0,
2580 GFP_KERNEL) == 0) {
2581 folio_mark_uptodate(folio);
2582 folio_unlock(folio);
2583 }
2584
2585 folio_put(folio);
2586 return 0;
2587}
2588EXPORT_SYMBOL_GPL(nfs_symlink);
2589
2590int
2591nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2592{
2593 struct inode *inode = d_inode(old_dentry);
2594 int error;
2595
2596 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2597 old_dentry, dentry);
2598
2599 trace_nfs_link_enter(inode, dir, dentry);
2600 d_drop(dentry);
2601 if (S_ISREG(inode->i_mode))
2602 nfs_sync_inode(inode);
2603 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2604 if (error == 0) {
2605 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2606 ihold(inode);
2607 d_add(dentry, inode);
2608 }
2609 trace_nfs_link_exit(inode, dir, dentry, error);
2610 return error;
2611}
2612EXPORT_SYMBOL_GPL(nfs_link);
2613
2614static void
2615nfs_unblock_rename(struct rpc_task *task, struct nfs_renamedata *data)
2616{
2617 struct dentry *new_dentry = data->new_dentry;
2618
2619 new_dentry->d_fsdata = NULL;
2620 wake_up_var(&new_dentry->d_fsdata);
2621}
2622
2623/*
2624 * RENAME
2625 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2626 * different file handle for the same inode after a rename (e.g. when
2627 * moving to a different directory). A fail-safe method to do so would
2628 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2629 * rename the old file using the sillyrename stuff. This way, the original
2630 * file in old_dir will go away when the last process iput()s the inode.
2631 *
2632 * FIXED.
2633 *
2634 * It actually works quite well. One needs to have the possibility for
2635 * at least one ".nfs..." file in each directory the file ever gets
2636 * moved or linked to which happens automagically with the new
2637 * implementation that only depends on the dcache stuff instead of
2638 * using the inode layer
2639 *
2640 * Unfortunately, things are a little more complicated than indicated
2641 * above. For a cross-directory move, we want to make sure we can get
2642 * rid of the old inode after the operation. This means there must be
2643 * no pending writes (if it's a file), and the use count must be 1.
2644 * If these conditions are met, we can drop the dentries before doing
2645 * the rename.
2646 */
2647int nfs_rename(struct mnt_idmap *idmap, struct inode *old_dir,
2648 struct dentry *old_dentry, struct inode *new_dir,
2649 struct dentry *new_dentry, unsigned int flags)
2650{
2651 struct inode *old_inode = d_inode(old_dentry);
2652 struct inode *new_inode = d_inode(new_dentry);
2653 struct dentry *dentry = NULL;
2654 struct rpc_task *task;
2655 bool must_unblock = false;
2656 int error = -EBUSY;
2657
2658 if (flags)
2659 return -EINVAL;
2660
2661 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2662 old_dentry, new_dentry,
2663 d_count(new_dentry));
2664
2665 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2666 /*
2667 * For non-directories, check whether the target is busy and if so,
2668 * make a copy of the dentry and then do a silly-rename. If the
2669 * silly-rename succeeds, the copied dentry is hashed and becomes
2670 * the new target.
2671 */
2672 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2673 /* We must prevent any concurrent open until the unlink
2674 * completes. ->d_revalidate will wait for ->d_fsdata
2675 * to clear. We set it here to ensure no lookup succeeds until
2676 * the unlink is complete on the server.
2677 */
2678 error = -ETXTBSY;
2679 if (WARN_ON(new_dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2680 WARN_ON(new_dentry->d_fsdata == NFS_FSDATA_BLOCKED))
2681 goto out;
2682 if (new_dentry->d_fsdata) {
2683 /* old devname */
2684 kfree(new_dentry->d_fsdata);
2685 new_dentry->d_fsdata = NULL;
2686 }
2687
2688 spin_lock(&new_dentry->d_lock);
2689 if (d_count(new_dentry) > 2) {
2690 int err;
2691
2692 spin_unlock(&new_dentry->d_lock);
2693
2694 /* copy the target dentry's name */
2695 dentry = d_alloc(new_dentry->d_parent,
2696 &new_dentry->d_name);
2697 if (!dentry)
2698 goto out;
2699
2700 /* silly-rename the existing target ... */
2701 err = nfs_sillyrename(new_dir, new_dentry);
2702 if (err)
2703 goto out;
2704
2705 new_dentry = dentry;
2706 new_inode = NULL;
2707 } else {
2708 new_dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2709 must_unblock = true;
2710 spin_unlock(&new_dentry->d_lock);
2711 }
2712
2713 }
2714
2715 if (S_ISREG(old_inode->i_mode))
2716 nfs_sync_inode(old_inode);
2717 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry,
2718 must_unblock ? nfs_unblock_rename : NULL);
2719 if (IS_ERR(task)) {
2720 error = PTR_ERR(task);
2721 goto out;
2722 }
2723
2724 error = rpc_wait_for_completion_task(task);
2725 if (error != 0) {
2726 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2727 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2728 smp_wmb();
2729 } else
2730 error = task->tk_status;
2731 rpc_put_task(task);
2732 /* Ensure the inode attributes are revalidated */
2733 if (error == 0) {
2734 spin_lock(&old_inode->i_lock);
2735 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2736 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2737 NFS_INO_INVALID_CTIME |
2738 NFS_INO_REVAL_FORCED);
2739 spin_unlock(&old_inode->i_lock);
2740 }
2741out:
2742 trace_nfs_rename_exit(old_dir, old_dentry,
2743 new_dir, new_dentry, error);
2744 if (!error) {
2745 if (new_inode != NULL)
2746 nfs_drop_nlink(new_inode);
2747 /*
2748 * The d_move() should be here instead of in an async RPC completion
2749 * handler because we need the proper locks to move the dentry. If
2750 * we're interrupted by a signal, the async RPC completion handler
2751 * should mark the directories for revalidation.
2752 */
2753 d_move(old_dentry, new_dentry);
2754 nfs_set_verifier(old_dentry,
2755 nfs_save_change_attribute(new_dir));
2756 } else if (error == -ENOENT)
2757 nfs_dentry_handle_enoent(old_dentry);
2758
2759 /* new dentry created? */
2760 if (dentry)
2761 dput(dentry);
2762 return error;
2763}
2764EXPORT_SYMBOL_GPL(nfs_rename);
2765
2766static DEFINE_SPINLOCK(nfs_access_lru_lock);
2767static LIST_HEAD(nfs_access_lru_list);
2768static atomic_long_t nfs_access_nr_entries;
2769
2770static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2771module_param(nfs_access_max_cachesize, ulong, 0644);
2772MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2773
2774static void nfs_access_free_entry(struct nfs_access_entry *entry)
2775{
2776 put_group_info(entry->group_info);
2777 kfree_rcu(entry, rcu_head);
2778 smp_mb__before_atomic();
2779 atomic_long_dec(&nfs_access_nr_entries);
2780 smp_mb__after_atomic();
2781}
2782
2783static void nfs_access_free_list(struct list_head *head)
2784{
2785 struct nfs_access_entry *cache;
2786
2787 while (!list_empty(head)) {
2788 cache = list_entry(head->next, struct nfs_access_entry, lru);
2789 list_del(&cache->lru);
2790 nfs_access_free_entry(cache);
2791 }
2792}
2793
2794static unsigned long
2795nfs_do_access_cache_scan(unsigned int nr_to_scan)
2796{
2797 LIST_HEAD(head);
2798 struct nfs_inode *nfsi, *next;
2799 struct nfs_access_entry *cache;
2800 long freed = 0;
2801
2802 spin_lock(&nfs_access_lru_lock);
2803 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2804 struct inode *inode;
2805
2806 if (nr_to_scan-- == 0)
2807 break;
2808 inode = &nfsi->vfs_inode;
2809 spin_lock(&inode->i_lock);
2810 if (list_empty(&nfsi->access_cache_entry_lru))
2811 goto remove_lru_entry;
2812 cache = list_entry(nfsi->access_cache_entry_lru.next,
2813 struct nfs_access_entry, lru);
2814 list_move(&cache->lru, &head);
2815 rb_erase(&cache->rb_node, &nfsi->access_cache);
2816 freed++;
2817 if (!list_empty(&nfsi->access_cache_entry_lru))
2818 list_move_tail(&nfsi->access_cache_inode_lru,
2819 &nfs_access_lru_list);
2820 else {
2821remove_lru_entry:
2822 list_del_init(&nfsi->access_cache_inode_lru);
2823 smp_mb__before_atomic();
2824 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2825 smp_mb__after_atomic();
2826 }
2827 spin_unlock(&inode->i_lock);
2828 }
2829 spin_unlock(&nfs_access_lru_lock);
2830 nfs_access_free_list(&head);
2831 return freed;
2832}
2833
2834unsigned long
2835nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2836{
2837 int nr_to_scan = sc->nr_to_scan;
2838 gfp_t gfp_mask = sc->gfp_mask;
2839
2840 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2841 return SHRINK_STOP;
2842 return nfs_do_access_cache_scan(nr_to_scan);
2843}
2844
2845
2846unsigned long
2847nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2848{
2849 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2850}
2851
2852static void
2853nfs_access_cache_enforce_limit(void)
2854{
2855 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2856 unsigned long diff;
2857 unsigned int nr_to_scan;
2858
2859 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2860 return;
2861 nr_to_scan = 100;
2862 diff = nr_entries - nfs_access_max_cachesize;
2863 if (diff < nr_to_scan)
2864 nr_to_scan = diff;
2865 nfs_do_access_cache_scan(nr_to_scan);
2866}
2867
2868static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2869{
2870 struct rb_root *root_node = &nfsi->access_cache;
2871 struct rb_node *n;
2872 struct nfs_access_entry *entry;
2873
2874 /* Unhook entries from the cache */
2875 while ((n = rb_first(root_node)) != NULL) {
2876 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2877 rb_erase(n, root_node);
2878 list_move(&entry->lru, head);
2879 }
2880 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2881}
2882
2883void nfs_access_zap_cache(struct inode *inode)
2884{
2885 LIST_HEAD(head);
2886
2887 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2888 return;
2889 /* Remove from global LRU init */
2890 spin_lock(&nfs_access_lru_lock);
2891 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2892 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2893
2894 spin_lock(&inode->i_lock);
2895 __nfs_access_zap_cache(NFS_I(inode), &head);
2896 spin_unlock(&inode->i_lock);
2897 spin_unlock(&nfs_access_lru_lock);
2898 nfs_access_free_list(&head);
2899}
2900EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2901
2902static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2903{
2904 struct group_info *ga, *gb;
2905 int g;
2906
2907 if (uid_lt(a->fsuid, b->fsuid))
2908 return -1;
2909 if (uid_gt(a->fsuid, b->fsuid))
2910 return 1;
2911
2912 if (gid_lt(a->fsgid, b->fsgid))
2913 return -1;
2914 if (gid_gt(a->fsgid, b->fsgid))
2915 return 1;
2916
2917 ga = a->group_info;
2918 gb = b->group_info;
2919 if (ga == gb)
2920 return 0;
2921 if (ga == NULL)
2922 return -1;
2923 if (gb == NULL)
2924 return 1;
2925 if (ga->ngroups < gb->ngroups)
2926 return -1;
2927 if (ga->ngroups > gb->ngroups)
2928 return 1;
2929
2930 for (g = 0; g < ga->ngroups; g++) {
2931 if (gid_lt(ga->gid[g], gb->gid[g]))
2932 return -1;
2933 if (gid_gt(ga->gid[g], gb->gid[g]))
2934 return 1;
2935 }
2936 return 0;
2937}
2938
2939static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2940{
2941 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2942
2943 while (n != NULL) {
2944 struct nfs_access_entry *entry =
2945 rb_entry(n, struct nfs_access_entry, rb_node);
2946 int cmp = access_cmp(cred, entry);
2947
2948 if (cmp < 0)
2949 n = n->rb_left;
2950 else if (cmp > 0)
2951 n = n->rb_right;
2952 else
2953 return entry;
2954 }
2955 return NULL;
2956}
2957
2958static u64 nfs_access_login_time(const struct task_struct *task,
2959 const struct cred *cred)
2960{
2961 const struct task_struct *parent;
2962 const struct cred *pcred;
2963 u64 ret;
2964
2965 rcu_read_lock();
2966 for (;;) {
2967 parent = rcu_dereference(task->real_parent);
2968 pcred = __task_cred(parent);
2969 if (parent == task || cred_fscmp(pcred, cred) != 0)
2970 break;
2971 task = parent;
2972 }
2973 ret = task->start_time;
2974 rcu_read_unlock();
2975 return ret;
2976}
2977
2978static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
2979{
2980 struct nfs_inode *nfsi = NFS_I(inode);
2981 u64 login_time = nfs_access_login_time(current, cred);
2982 struct nfs_access_entry *cache;
2983 bool retry = true;
2984 int err;
2985
2986 spin_lock(&inode->i_lock);
2987 for(;;) {
2988 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2989 goto out_zap;
2990 cache = nfs_access_search_rbtree(inode, cred);
2991 err = -ENOENT;
2992 if (cache == NULL)
2993 goto out;
2994 /* Found an entry, is our attribute cache valid? */
2995 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2996 break;
2997 if (!retry)
2998 break;
2999 err = -ECHILD;
3000 if (!may_block)
3001 goto out;
3002 spin_unlock(&inode->i_lock);
3003 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
3004 if (err)
3005 return err;
3006 spin_lock(&inode->i_lock);
3007 retry = false;
3008 }
3009 err = -ENOENT;
3010 if ((s64)(login_time - cache->timestamp) > 0)
3011 goto out;
3012 *mask = cache->mask;
3013 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
3014 err = 0;
3015out:
3016 spin_unlock(&inode->i_lock);
3017 return err;
3018out_zap:
3019 spin_unlock(&inode->i_lock);
3020 nfs_access_zap_cache(inode);
3021 return -ENOENT;
3022}
3023
3024static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
3025{
3026 /* Only check the most recently returned cache entry,
3027 * but do it without locking.
3028 */
3029 struct nfs_inode *nfsi = NFS_I(inode);
3030 u64 login_time = nfs_access_login_time(current, cred);
3031 struct nfs_access_entry *cache;
3032 int err = -ECHILD;
3033 struct list_head *lh;
3034
3035 rcu_read_lock();
3036 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3037 goto out;
3038 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
3039 cache = list_entry(lh, struct nfs_access_entry, lru);
3040 if (lh == &nfsi->access_cache_entry_lru ||
3041 access_cmp(cred, cache) != 0)
3042 cache = NULL;
3043 if (cache == NULL)
3044 goto out;
3045 if ((s64)(login_time - cache->timestamp) > 0)
3046 goto out;
3047 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3048 goto out;
3049 *mask = cache->mask;
3050 err = 0;
3051out:
3052 rcu_read_unlock();
3053 return err;
3054}
3055
3056int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
3057 u32 *mask, bool may_block)
3058{
3059 int status;
3060
3061 status = nfs_access_get_cached_rcu(inode, cred, mask);
3062 if (status != 0)
3063 status = nfs_access_get_cached_locked(inode, cred, mask,
3064 may_block);
3065
3066 return status;
3067}
3068EXPORT_SYMBOL_GPL(nfs_access_get_cached);
3069
3070static void nfs_access_add_rbtree(struct inode *inode,
3071 struct nfs_access_entry *set,
3072 const struct cred *cred)
3073{
3074 struct nfs_inode *nfsi = NFS_I(inode);
3075 struct rb_root *root_node = &nfsi->access_cache;
3076 struct rb_node **p = &root_node->rb_node;
3077 struct rb_node *parent = NULL;
3078 struct nfs_access_entry *entry;
3079 int cmp;
3080
3081 spin_lock(&inode->i_lock);
3082 while (*p != NULL) {
3083 parent = *p;
3084 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3085 cmp = access_cmp(cred, entry);
3086
3087 if (cmp < 0)
3088 p = &parent->rb_left;
3089 else if (cmp > 0)
3090 p = &parent->rb_right;
3091 else
3092 goto found;
3093 }
3094 rb_link_node(&set->rb_node, parent, p);
3095 rb_insert_color(&set->rb_node, root_node);
3096 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3097 spin_unlock(&inode->i_lock);
3098 return;
3099found:
3100 rb_replace_node(parent, &set->rb_node, root_node);
3101 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3102 list_del(&entry->lru);
3103 spin_unlock(&inode->i_lock);
3104 nfs_access_free_entry(entry);
3105}
3106
3107void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3108 const struct cred *cred)
3109{
3110 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
3111 if (cache == NULL)
3112 return;
3113 RB_CLEAR_NODE(&cache->rb_node);
3114 cache->fsuid = cred->fsuid;
3115 cache->fsgid = cred->fsgid;
3116 cache->group_info = get_group_info(cred->group_info);
3117 cache->mask = set->mask;
3118 cache->timestamp = ktime_get_ns();
3119
3120 /* The above field assignments must be visible
3121 * before this item appears on the lru. We cannot easily
3122 * use rcu_assign_pointer, so just force the memory barrier.
3123 */
3124 smp_wmb();
3125 nfs_access_add_rbtree(inode, cache, cred);
3126
3127 /* Update accounting */
3128 smp_mb__before_atomic();
3129 atomic_long_inc(&nfs_access_nr_entries);
3130 smp_mb__after_atomic();
3131
3132 /* Add inode to global LRU list */
3133 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3134 spin_lock(&nfs_access_lru_lock);
3135 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
3136 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
3137 &nfs_access_lru_list);
3138 spin_unlock(&nfs_access_lru_lock);
3139 }
3140 nfs_access_cache_enforce_limit();
3141}
3142EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3143
3144#define NFS_MAY_READ (NFS_ACCESS_READ)
3145#define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3146 NFS_ACCESS_EXTEND | \
3147 NFS_ACCESS_DELETE)
3148#define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3149 NFS_ACCESS_EXTEND)
3150#define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3151#define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3152#define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3153static int
3154nfs_access_calc_mask(u32 access_result, umode_t umode)
3155{
3156 int mask = 0;
3157
3158 if (access_result & NFS_MAY_READ)
3159 mask |= MAY_READ;
3160 if (S_ISDIR(umode)) {
3161 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3162 mask |= MAY_WRITE;
3163 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3164 mask |= MAY_EXEC;
3165 } else if (S_ISREG(umode)) {
3166 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3167 mask |= MAY_WRITE;
3168 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3169 mask |= MAY_EXEC;
3170 } else if (access_result & NFS_MAY_WRITE)
3171 mask |= MAY_WRITE;
3172 return mask;
3173}
3174
3175void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3176{
3177 entry->mask = access_result;
3178}
3179EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3180
3181static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3182{
3183 struct nfs_access_entry cache;
3184 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3185 int cache_mask = -1;
3186 int status;
3187
3188 trace_nfs_access_enter(inode);
3189
3190 status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3191 if (status == 0)
3192 goto out_cached;
3193
3194 status = -ECHILD;
3195 if (!may_block)
3196 goto out;
3197
3198 /*
3199 * Determine which access bits we want to ask for...
3200 */
3201 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3202 nfs_access_xattr_mask(NFS_SERVER(inode));
3203 if (S_ISDIR(inode->i_mode))
3204 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3205 else
3206 cache.mask |= NFS_ACCESS_EXECUTE;
3207 status = NFS_PROTO(inode)->access(inode, &cache, cred);
3208 if (status != 0) {
3209 if (status == -ESTALE) {
3210 if (!S_ISDIR(inode->i_mode))
3211 nfs_set_inode_stale(inode);
3212 else
3213 nfs_zap_caches(inode);
3214 }
3215 goto out;
3216 }
3217 nfs_access_add_cache(inode, &cache, cred);
3218out_cached:
3219 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
3220 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3221 status = -EACCES;
3222out:
3223 trace_nfs_access_exit(inode, mask, cache_mask, status);
3224 return status;
3225}
3226
3227static int nfs_open_permission_mask(int openflags)
3228{
3229 int mask = 0;
3230
3231 if (openflags & __FMODE_EXEC) {
3232 /* ONLY check exec rights */
3233 mask = MAY_EXEC;
3234 } else {
3235 if ((openflags & O_ACCMODE) != O_WRONLY)
3236 mask |= MAY_READ;
3237 if ((openflags & O_ACCMODE) != O_RDONLY)
3238 mask |= MAY_WRITE;
3239 }
3240
3241 return mask;
3242}
3243
3244int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3245{
3246 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
3247}
3248EXPORT_SYMBOL_GPL(nfs_may_open);
3249
3250static int nfs_execute_ok(struct inode *inode, int mask)
3251{
3252 struct nfs_server *server = NFS_SERVER(inode);
3253 int ret = 0;
3254
3255 if (S_ISDIR(inode->i_mode))
3256 return 0;
3257 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3258 if (mask & MAY_NOT_BLOCK)
3259 return -ECHILD;
3260 ret = __nfs_revalidate_inode(server, inode);
3261 }
3262 if (ret == 0 && !execute_ok(inode))
3263 ret = -EACCES;
3264 return ret;
3265}
3266
3267int nfs_permission(struct mnt_idmap *idmap,
3268 struct inode *inode,
3269 int mask)
3270{
3271 const struct cred *cred = current_cred();
3272 int res = 0;
3273
3274 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
3275
3276 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3277 goto out;
3278 /* Is this sys_access() ? */
3279 if (mask & (MAY_ACCESS | MAY_CHDIR))
3280 goto force_lookup;
3281
3282 switch (inode->i_mode & S_IFMT) {
3283 case S_IFLNK:
3284 goto out;
3285 case S_IFREG:
3286 if ((mask & MAY_OPEN) &&
3287 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3288 return 0;
3289 break;
3290 case S_IFDIR:
3291 /*
3292 * Optimize away all write operations, since the server
3293 * will check permissions when we perform the op.
3294 */
3295 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3296 goto out;
3297 }
3298
3299force_lookup:
3300 if (!NFS_PROTO(inode)->access)
3301 goto out_notsup;
3302
3303 res = nfs_do_access(inode, cred, mask);
3304out:
3305 if (!res && (mask & MAY_EXEC))
3306 res = nfs_execute_ok(inode, mask);
3307
3308 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3309 inode->i_sb->s_id, inode->i_ino, mask, res);
3310 return res;
3311out_notsup:
3312 if (mask & MAY_NOT_BLOCK)
3313 return -ECHILD;
3314
3315 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3316 NFS_INO_INVALID_OTHER);
3317 if (res == 0)
3318 res = generic_permission(&nop_mnt_idmap, inode, mask);
3319 goto out;
3320}
3321EXPORT_SYMBOL_GPL(nfs_permission);
1/*
2 * linux/fs/nfs/dir.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * nfs directory handling functions
7 *
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
18 */
19
20#include <linux/time.h>
21#include <linux/errno.h>
22#include <linux/stat.h>
23#include <linux/fcntl.h>
24#include <linux/string.h>
25#include <linux/kernel.h>
26#include <linux/slab.h>
27#include <linux/mm.h>
28#include <linux/sunrpc/clnt.h>
29#include <linux/nfs_fs.h>
30#include <linux/nfs_mount.h>
31#include <linux/pagemap.h>
32#include <linux/pagevec.h>
33#include <linux/namei.h>
34#include <linux/mount.h>
35#include <linux/sched.h>
36#include <linux/kmemleak.h>
37#include <linux/xattr.h>
38
39#include "delegation.h"
40#include "iostat.h"
41#include "internal.h"
42#include "fscache.h"
43
44/* #define NFS_DEBUG_VERBOSE 1 */
45
46static int nfs_opendir(struct inode *, struct file *);
47static int nfs_closedir(struct inode *, struct file *);
48static int nfs_readdir(struct file *, void *, filldir_t);
49static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
50static int nfs_create(struct inode *, struct dentry *, umode_t, struct nameidata *);
51static int nfs_mkdir(struct inode *, struct dentry *, umode_t);
52static int nfs_rmdir(struct inode *, struct dentry *);
53static int nfs_unlink(struct inode *, struct dentry *);
54static int nfs_symlink(struct inode *, struct dentry *, const char *);
55static int nfs_link(struct dentry *, struct inode *, struct dentry *);
56static int nfs_mknod(struct inode *, struct dentry *, umode_t, dev_t);
57static int nfs_rename(struct inode *, struct dentry *,
58 struct inode *, struct dentry *);
59static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
60static loff_t nfs_llseek_dir(struct file *, loff_t, int);
61static void nfs_readdir_clear_array(struct page*);
62
63const struct file_operations nfs_dir_operations = {
64 .llseek = nfs_llseek_dir,
65 .read = generic_read_dir,
66 .readdir = nfs_readdir,
67 .open = nfs_opendir,
68 .release = nfs_closedir,
69 .fsync = nfs_fsync_dir,
70};
71
72const struct inode_operations nfs_dir_inode_operations = {
73 .create = nfs_create,
74 .lookup = nfs_lookup,
75 .link = nfs_link,
76 .unlink = nfs_unlink,
77 .symlink = nfs_symlink,
78 .mkdir = nfs_mkdir,
79 .rmdir = nfs_rmdir,
80 .mknod = nfs_mknod,
81 .rename = nfs_rename,
82 .permission = nfs_permission,
83 .getattr = nfs_getattr,
84 .setattr = nfs_setattr,
85};
86
87const struct address_space_operations nfs_dir_aops = {
88 .freepage = nfs_readdir_clear_array,
89};
90
91#ifdef CONFIG_NFS_V3
92const struct inode_operations nfs3_dir_inode_operations = {
93 .create = nfs_create,
94 .lookup = nfs_lookup,
95 .link = nfs_link,
96 .unlink = nfs_unlink,
97 .symlink = nfs_symlink,
98 .mkdir = nfs_mkdir,
99 .rmdir = nfs_rmdir,
100 .mknod = nfs_mknod,
101 .rename = nfs_rename,
102 .permission = nfs_permission,
103 .getattr = nfs_getattr,
104 .setattr = nfs_setattr,
105 .listxattr = nfs3_listxattr,
106 .getxattr = nfs3_getxattr,
107 .setxattr = nfs3_setxattr,
108 .removexattr = nfs3_removexattr,
109};
110#endif /* CONFIG_NFS_V3 */
111
112#ifdef CONFIG_NFS_V4
113
114static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
115static int nfs_open_create(struct inode *dir, struct dentry *dentry, umode_t mode, struct nameidata *nd);
116const struct inode_operations nfs4_dir_inode_operations = {
117 .create = nfs_open_create,
118 .lookup = nfs_atomic_lookup,
119 .link = nfs_link,
120 .unlink = nfs_unlink,
121 .symlink = nfs_symlink,
122 .mkdir = nfs_mkdir,
123 .rmdir = nfs_rmdir,
124 .mknod = nfs_mknod,
125 .rename = nfs_rename,
126 .permission = nfs_permission,
127 .getattr = nfs_getattr,
128 .setattr = nfs_setattr,
129 .getxattr = generic_getxattr,
130 .setxattr = generic_setxattr,
131 .listxattr = generic_listxattr,
132 .removexattr = generic_removexattr,
133};
134
135#endif /* CONFIG_NFS_V4 */
136
137static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
138{
139 struct nfs_open_dir_context *ctx;
140 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
141 if (ctx != NULL) {
142 ctx->duped = 0;
143 ctx->attr_gencount = NFS_I(dir)->attr_gencount;
144 ctx->dir_cookie = 0;
145 ctx->dup_cookie = 0;
146 ctx->cred = get_rpccred(cred);
147 return ctx;
148 }
149 return ERR_PTR(-ENOMEM);
150}
151
152static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
153{
154 put_rpccred(ctx->cred);
155 kfree(ctx);
156}
157
158/*
159 * Open file
160 */
161static int
162nfs_opendir(struct inode *inode, struct file *filp)
163{
164 int res = 0;
165 struct nfs_open_dir_context *ctx;
166 struct rpc_cred *cred;
167
168 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
169 filp->f_path.dentry->d_parent->d_name.name,
170 filp->f_path.dentry->d_name.name);
171
172 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
173
174 cred = rpc_lookup_cred();
175 if (IS_ERR(cred))
176 return PTR_ERR(cred);
177 ctx = alloc_nfs_open_dir_context(inode, cred);
178 if (IS_ERR(ctx)) {
179 res = PTR_ERR(ctx);
180 goto out;
181 }
182 filp->private_data = ctx;
183 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
184 /* This is a mountpoint, so d_revalidate will never
185 * have been called, so we need to refresh the
186 * inode (for close-open consistency) ourselves.
187 */
188 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
189 }
190out:
191 put_rpccred(cred);
192 return res;
193}
194
195static int
196nfs_closedir(struct inode *inode, struct file *filp)
197{
198 put_nfs_open_dir_context(filp->private_data);
199 return 0;
200}
201
202struct nfs_cache_array_entry {
203 u64 cookie;
204 u64 ino;
205 struct qstr string;
206 unsigned char d_type;
207};
208
209struct nfs_cache_array {
210 int size;
211 int eof_index;
212 u64 last_cookie;
213 struct nfs_cache_array_entry array[0];
214};
215
216typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
217typedef struct {
218 struct file *file;
219 struct page *page;
220 unsigned long page_index;
221 u64 *dir_cookie;
222 u64 last_cookie;
223 loff_t current_index;
224 decode_dirent_t decode;
225
226 unsigned long timestamp;
227 unsigned long gencount;
228 unsigned int cache_entry_index;
229 unsigned int plus:1;
230 unsigned int eof:1;
231} nfs_readdir_descriptor_t;
232
233/*
234 * The caller is responsible for calling nfs_readdir_release_array(page)
235 */
236static
237struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
238{
239 void *ptr;
240 if (page == NULL)
241 return ERR_PTR(-EIO);
242 ptr = kmap(page);
243 if (ptr == NULL)
244 return ERR_PTR(-ENOMEM);
245 return ptr;
246}
247
248static
249void nfs_readdir_release_array(struct page *page)
250{
251 kunmap(page);
252}
253
254/*
255 * we are freeing strings created by nfs_add_to_readdir_array()
256 */
257static
258void nfs_readdir_clear_array(struct page *page)
259{
260 struct nfs_cache_array *array;
261 int i;
262
263 array = kmap_atomic(page);
264 for (i = 0; i < array->size; i++)
265 kfree(array->array[i].string.name);
266 kunmap_atomic(array);
267}
268
269/*
270 * the caller is responsible for freeing qstr.name
271 * when called by nfs_readdir_add_to_array, the strings will be freed in
272 * nfs_clear_readdir_array()
273 */
274static
275int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
276{
277 string->len = len;
278 string->name = kmemdup(name, len, GFP_KERNEL);
279 if (string->name == NULL)
280 return -ENOMEM;
281 /*
282 * Avoid a kmemleak false positive. The pointer to the name is stored
283 * in a page cache page which kmemleak does not scan.
284 */
285 kmemleak_not_leak(string->name);
286 string->hash = full_name_hash(name, len);
287 return 0;
288}
289
290static
291int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
292{
293 struct nfs_cache_array *array = nfs_readdir_get_array(page);
294 struct nfs_cache_array_entry *cache_entry;
295 int ret;
296
297 if (IS_ERR(array))
298 return PTR_ERR(array);
299
300 cache_entry = &array->array[array->size];
301
302 /* Check that this entry lies within the page bounds */
303 ret = -ENOSPC;
304 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
305 goto out;
306
307 cache_entry->cookie = entry->prev_cookie;
308 cache_entry->ino = entry->ino;
309 cache_entry->d_type = entry->d_type;
310 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
311 if (ret)
312 goto out;
313 array->last_cookie = entry->cookie;
314 array->size++;
315 if (entry->eof != 0)
316 array->eof_index = array->size;
317out:
318 nfs_readdir_release_array(page);
319 return ret;
320}
321
322static
323int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
324{
325 loff_t diff = desc->file->f_pos - desc->current_index;
326 unsigned int index;
327
328 if (diff < 0)
329 goto out_eof;
330 if (diff >= array->size) {
331 if (array->eof_index >= 0)
332 goto out_eof;
333 return -EAGAIN;
334 }
335
336 index = (unsigned int)diff;
337 *desc->dir_cookie = array->array[index].cookie;
338 desc->cache_entry_index = index;
339 return 0;
340out_eof:
341 desc->eof = 1;
342 return -EBADCOOKIE;
343}
344
345static
346int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
347{
348 int i;
349 loff_t new_pos;
350 int status = -EAGAIN;
351
352 for (i = 0; i < array->size; i++) {
353 if (array->array[i].cookie == *desc->dir_cookie) {
354 struct nfs_inode *nfsi = NFS_I(desc->file->f_path.dentry->d_inode);
355 struct nfs_open_dir_context *ctx = desc->file->private_data;
356
357 new_pos = desc->current_index + i;
358 if (ctx->attr_gencount != nfsi->attr_gencount
359 || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
360 ctx->duped = 0;
361 ctx->attr_gencount = nfsi->attr_gencount;
362 } else if (new_pos < desc->file->f_pos) {
363 if (ctx->duped > 0
364 && ctx->dup_cookie == *desc->dir_cookie) {
365 if (printk_ratelimit()) {
366 pr_notice("NFS: directory %s/%s contains a readdir loop."
367 "Please contact your server vendor. "
368 "The file: %s has duplicate cookie %llu\n",
369 desc->file->f_dentry->d_parent->d_name.name,
370 desc->file->f_dentry->d_name.name,
371 array->array[i].string.name,
372 *desc->dir_cookie);
373 }
374 status = -ELOOP;
375 goto out;
376 }
377 ctx->dup_cookie = *desc->dir_cookie;
378 ctx->duped = -1;
379 }
380 desc->file->f_pos = new_pos;
381 desc->cache_entry_index = i;
382 return 0;
383 }
384 }
385 if (array->eof_index >= 0) {
386 status = -EBADCOOKIE;
387 if (*desc->dir_cookie == array->last_cookie)
388 desc->eof = 1;
389 }
390out:
391 return status;
392}
393
394static
395int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
396{
397 struct nfs_cache_array *array;
398 int status;
399
400 array = nfs_readdir_get_array(desc->page);
401 if (IS_ERR(array)) {
402 status = PTR_ERR(array);
403 goto out;
404 }
405
406 if (*desc->dir_cookie == 0)
407 status = nfs_readdir_search_for_pos(array, desc);
408 else
409 status = nfs_readdir_search_for_cookie(array, desc);
410
411 if (status == -EAGAIN) {
412 desc->last_cookie = array->last_cookie;
413 desc->current_index += array->size;
414 desc->page_index++;
415 }
416 nfs_readdir_release_array(desc->page);
417out:
418 return status;
419}
420
421/* Fill a page with xdr information before transferring to the cache page */
422static
423int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
424 struct nfs_entry *entry, struct file *file, struct inode *inode)
425{
426 struct nfs_open_dir_context *ctx = file->private_data;
427 struct rpc_cred *cred = ctx->cred;
428 unsigned long timestamp, gencount;
429 int error;
430
431 again:
432 timestamp = jiffies;
433 gencount = nfs_inc_attr_generation_counter();
434 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
435 NFS_SERVER(inode)->dtsize, desc->plus);
436 if (error < 0) {
437 /* We requested READDIRPLUS, but the server doesn't grok it */
438 if (error == -ENOTSUPP && desc->plus) {
439 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
440 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
441 desc->plus = 0;
442 goto again;
443 }
444 goto error;
445 }
446 desc->timestamp = timestamp;
447 desc->gencount = gencount;
448error:
449 return error;
450}
451
452static int xdr_decode(nfs_readdir_descriptor_t *desc,
453 struct nfs_entry *entry, struct xdr_stream *xdr)
454{
455 int error;
456
457 error = desc->decode(xdr, entry, desc->plus);
458 if (error)
459 return error;
460 entry->fattr->time_start = desc->timestamp;
461 entry->fattr->gencount = desc->gencount;
462 return 0;
463}
464
465static
466int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
467{
468 if (dentry->d_inode == NULL)
469 goto different;
470 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
471 goto different;
472 return 1;
473different:
474 return 0;
475}
476
477static
478bool nfs_use_readdirplus(struct inode *dir, struct file *filp)
479{
480 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
481 return false;
482 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
483 return true;
484 if (filp->f_pos == 0)
485 return true;
486 return false;
487}
488
489/*
490 * This function is called by the lookup code to request the use of
491 * readdirplus to accelerate any future lookups in the same
492 * directory.
493 */
494static
495void nfs_advise_use_readdirplus(struct inode *dir)
496{
497 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
498}
499
500static
501void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
502{
503 struct qstr filename = QSTR_INIT(entry->name, entry->len);
504 struct dentry *dentry;
505 struct dentry *alias;
506 struct inode *dir = parent->d_inode;
507 struct inode *inode;
508
509 if (filename.name[0] == '.') {
510 if (filename.len == 1)
511 return;
512 if (filename.len == 2 && filename.name[1] == '.')
513 return;
514 }
515 filename.hash = full_name_hash(filename.name, filename.len);
516
517 dentry = d_lookup(parent, &filename);
518 if (dentry != NULL) {
519 if (nfs_same_file(dentry, entry)) {
520 nfs_refresh_inode(dentry->d_inode, entry->fattr);
521 goto out;
522 } else {
523 d_drop(dentry);
524 dput(dentry);
525 }
526 }
527
528 dentry = d_alloc(parent, &filename);
529 if (dentry == NULL)
530 return;
531
532 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
533 if (IS_ERR(inode))
534 goto out;
535
536 alias = d_materialise_unique(dentry, inode);
537 if (IS_ERR(alias))
538 goto out;
539 else if (alias) {
540 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
541 dput(alias);
542 } else
543 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
544
545out:
546 dput(dentry);
547}
548
549/* Perform conversion from xdr to cache array */
550static
551int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
552 struct page **xdr_pages, struct page *page, unsigned int buflen)
553{
554 struct xdr_stream stream;
555 struct xdr_buf buf;
556 struct page *scratch;
557 struct nfs_cache_array *array;
558 unsigned int count = 0;
559 int status;
560
561 scratch = alloc_page(GFP_KERNEL);
562 if (scratch == NULL)
563 return -ENOMEM;
564
565 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
566 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
567
568 do {
569 status = xdr_decode(desc, entry, &stream);
570 if (status != 0) {
571 if (status == -EAGAIN)
572 status = 0;
573 break;
574 }
575
576 count++;
577
578 if (desc->plus != 0)
579 nfs_prime_dcache(desc->file->f_path.dentry, entry);
580
581 status = nfs_readdir_add_to_array(entry, page);
582 if (status != 0)
583 break;
584 } while (!entry->eof);
585
586 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
587 array = nfs_readdir_get_array(page);
588 if (!IS_ERR(array)) {
589 array->eof_index = array->size;
590 status = 0;
591 nfs_readdir_release_array(page);
592 } else
593 status = PTR_ERR(array);
594 }
595
596 put_page(scratch);
597 return status;
598}
599
600static
601void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
602{
603 unsigned int i;
604 for (i = 0; i < npages; i++)
605 put_page(pages[i]);
606}
607
608static
609void nfs_readdir_free_large_page(void *ptr, struct page **pages,
610 unsigned int npages)
611{
612 nfs_readdir_free_pagearray(pages, npages);
613}
614
615/*
616 * nfs_readdir_large_page will allocate pages that must be freed with a call
617 * to nfs_readdir_free_large_page
618 */
619static
620int nfs_readdir_large_page(struct page **pages, unsigned int npages)
621{
622 unsigned int i;
623
624 for (i = 0; i < npages; i++) {
625 struct page *page = alloc_page(GFP_KERNEL);
626 if (page == NULL)
627 goto out_freepages;
628 pages[i] = page;
629 }
630 return 0;
631
632out_freepages:
633 nfs_readdir_free_pagearray(pages, i);
634 return -ENOMEM;
635}
636
637static
638int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
639{
640 struct page *pages[NFS_MAX_READDIR_PAGES];
641 void *pages_ptr = NULL;
642 struct nfs_entry entry;
643 struct file *file = desc->file;
644 struct nfs_cache_array *array;
645 int status = -ENOMEM;
646 unsigned int array_size = ARRAY_SIZE(pages);
647
648 entry.prev_cookie = 0;
649 entry.cookie = desc->last_cookie;
650 entry.eof = 0;
651 entry.fh = nfs_alloc_fhandle();
652 entry.fattr = nfs_alloc_fattr();
653 entry.server = NFS_SERVER(inode);
654 if (entry.fh == NULL || entry.fattr == NULL)
655 goto out;
656
657 array = nfs_readdir_get_array(page);
658 if (IS_ERR(array)) {
659 status = PTR_ERR(array);
660 goto out;
661 }
662 memset(array, 0, sizeof(struct nfs_cache_array));
663 array->eof_index = -1;
664
665 status = nfs_readdir_large_page(pages, array_size);
666 if (status < 0)
667 goto out_release_array;
668 do {
669 unsigned int pglen;
670 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
671
672 if (status < 0)
673 break;
674 pglen = status;
675 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
676 if (status < 0) {
677 if (status == -ENOSPC)
678 status = 0;
679 break;
680 }
681 } while (array->eof_index < 0);
682
683 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
684out_release_array:
685 nfs_readdir_release_array(page);
686out:
687 nfs_free_fattr(entry.fattr);
688 nfs_free_fhandle(entry.fh);
689 return status;
690}
691
692/*
693 * Now we cache directories properly, by converting xdr information
694 * to an array that can be used for lookups later. This results in
695 * fewer cache pages, since we can store more information on each page.
696 * We only need to convert from xdr once so future lookups are much simpler
697 */
698static
699int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
700{
701 struct inode *inode = desc->file->f_path.dentry->d_inode;
702 int ret;
703
704 ret = nfs_readdir_xdr_to_array(desc, page, inode);
705 if (ret < 0)
706 goto error;
707 SetPageUptodate(page);
708
709 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
710 /* Should never happen */
711 nfs_zap_mapping(inode, inode->i_mapping);
712 }
713 unlock_page(page);
714 return 0;
715 error:
716 unlock_page(page);
717 return ret;
718}
719
720static
721void cache_page_release(nfs_readdir_descriptor_t *desc)
722{
723 if (!desc->page->mapping)
724 nfs_readdir_clear_array(desc->page);
725 page_cache_release(desc->page);
726 desc->page = NULL;
727}
728
729static
730struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
731{
732 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
733 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
734}
735
736/*
737 * Returns 0 if desc->dir_cookie was found on page desc->page_index
738 */
739static
740int find_cache_page(nfs_readdir_descriptor_t *desc)
741{
742 int res;
743
744 desc->page = get_cache_page(desc);
745 if (IS_ERR(desc->page))
746 return PTR_ERR(desc->page);
747
748 res = nfs_readdir_search_array(desc);
749 if (res != 0)
750 cache_page_release(desc);
751 return res;
752}
753
754/* Search for desc->dir_cookie from the beginning of the page cache */
755static inline
756int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
757{
758 int res;
759
760 if (desc->page_index == 0) {
761 desc->current_index = 0;
762 desc->last_cookie = 0;
763 }
764 do {
765 res = find_cache_page(desc);
766 } while (res == -EAGAIN);
767 return res;
768}
769
770/*
771 * Once we've found the start of the dirent within a page: fill 'er up...
772 */
773static
774int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
775 filldir_t filldir)
776{
777 struct file *file = desc->file;
778 int i = 0;
779 int res = 0;
780 struct nfs_cache_array *array = NULL;
781 struct nfs_open_dir_context *ctx = file->private_data;
782
783 array = nfs_readdir_get_array(desc->page);
784 if (IS_ERR(array)) {
785 res = PTR_ERR(array);
786 goto out;
787 }
788
789 for (i = desc->cache_entry_index; i < array->size; i++) {
790 struct nfs_cache_array_entry *ent;
791
792 ent = &array->array[i];
793 if (filldir(dirent, ent->string.name, ent->string.len,
794 file->f_pos, nfs_compat_user_ino64(ent->ino),
795 ent->d_type) < 0) {
796 desc->eof = 1;
797 break;
798 }
799 file->f_pos++;
800 if (i < (array->size-1))
801 *desc->dir_cookie = array->array[i+1].cookie;
802 else
803 *desc->dir_cookie = array->last_cookie;
804 if (ctx->duped != 0)
805 ctx->duped = 1;
806 }
807 if (array->eof_index >= 0)
808 desc->eof = 1;
809
810 nfs_readdir_release_array(desc->page);
811out:
812 cache_page_release(desc);
813 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
814 (unsigned long long)*desc->dir_cookie, res);
815 return res;
816}
817
818/*
819 * If we cannot find a cookie in our cache, we suspect that this is
820 * because it points to a deleted file, so we ask the server to return
821 * whatever it thinks is the next entry. We then feed this to filldir.
822 * If all goes well, we should then be able to find our way round the
823 * cache on the next call to readdir_search_pagecache();
824 *
825 * NOTE: we cannot add the anonymous page to the pagecache because
826 * the data it contains might not be page aligned. Besides,
827 * we should already have a complete representation of the
828 * directory in the page cache by the time we get here.
829 */
830static inline
831int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
832 filldir_t filldir)
833{
834 struct page *page = NULL;
835 int status;
836 struct inode *inode = desc->file->f_path.dentry->d_inode;
837 struct nfs_open_dir_context *ctx = desc->file->private_data;
838
839 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
840 (unsigned long long)*desc->dir_cookie);
841
842 page = alloc_page(GFP_HIGHUSER);
843 if (!page) {
844 status = -ENOMEM;
845 goto out;
846 }
847
848 desc->page_index = 0;
849 desc->last_cookie = *desc->dir_cookie;
850 desc->page = page;
851 ctx->duped = 0;
852
853 status = nfs_readdir_xdr_to_array(desc, page, inode);
854 if (status < 0)
855 goto out_release;
856
857 status = nfs_do_filldir(desc, dirent, filldir);
858
859 out:
860 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
861 __func__, status);
862 return status;
863 out_release:
864 cache_page_release(desc);
865 goto out;
866}
867
868/* The file offset position represents the dirent entry number. A
869 last cookie cache takes care of the common case of reading the
870 whole directory.
871 */
872static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
873{
874 struct dentry *dentry = filp->f_path.dentry;
875 struct inode *inode = dentry->d_inode;
876 nfs_readdir_descriptor_t my_desc,
877 *desc = &my_desc;
878 struct nfs_open_dir_context *dir_ctx = filp->private_data;
879 int res;
880
881 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
882 dentry->d_parent->d_name.name, dentry->d_name.name,
883 (long long)filp->f_pos);
884 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
885
886 /*
887 * filp->f_pos points to the dirent entry number.
888 * *desc->dir_cookie has the cookie for the next entry. We have
889 * to either find the entry with the appropriate number or
890 * revalidate the cookie.
891 */
892 memset(desc, 0, sizeof(*desc));
893
894 desc->file = filp;
895 desc->dir_cookie = &dir_ctx->dir_cookie;
896 desc->decode = NFS_PROTO(inode)->decode_dirent;
897 desc->plus = nfs_use_readdirplus(inode, filp) ? 1 : 0;
898
899 nfs_block_sillyrename(dentry);
900 res = nfs_revalidate_mapping(inode, filp->f_mapping);
901 if (res < 0)
902 goto out;
903
904 do {
905 res = readdir_search_pagecache(desc);
906
907 if (res == -EBADCOOKIE) {
908 res = 0;
909 /* This means either end of directory */
910 if (*desc->dir_cookie && desc->eof == 0) {
911 /* Or that the server has 'lost' a cookie */
912 res = uncached_readdir(desc, dirent, filldir);
913 if (res == 0)
914 continue;
915 }
916 break;
917 }
918 if (res == -ETOOSMALL && desc->plus) {
919 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
920 nfs_zap_caches(inode);
921 desc->page_index = 0;
922 desc->plus = 0;
923 desc->eof = 0;
924 continue;
925 }
926 if (res < 0)
927 break;
928
929 res = nfs_do_filldir(desc, dirent, filldir);
930 if (res < 0)
931 break;
932 } while (!desc->eof);
933out:
934 nfs_unblock_sillyrename(dentry);
935 if (res > 0)
936 res = 0;
937 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
938 dentry->d_parent->d_name.name, dentry->d_name.name,
939 res);
940 return res;
941}
942
943static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
944{
945 struct dentry *dentry = filp->f_path.dentry;
946 struct inode *inode = dentry->d_inode;
947 struct nfs_open_dir_context *dir_ctx = filp->private_data;
948
949 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
950 dentry->d_parent->d_name.name,
951 dentry->d_name.name,
952 offset, origin);
953
954 mutex_lock(&inode->i_mutex);
955 switch (origin) {
956 case 1:
957 offset += filp->f_pos;
958 case 0:
959 if (offset >= 0)
960 break;
961 default:
962 offset = -EINVAL;
963 goto out;
964 }
965 if (offset != filp->f_pos) {
966 filp->f_pos = offset;
967 dir_ctx->dir_cookie = 0;
968 dir_ctx->duped = 0;
969 }
970out:
971 mutex_unlock(&inode->i_mutex);
972 return offset;
973}
974
975/*
976 * All directory operations under NFS are synchronous, so fsync()
977 * is a dummy operation.
978 */
979static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
980 int datasync)
981{
982 struct dentry *dentry = filp->f_path.dentry;
983 struct inode *inode = dentry->d_inode;
984
985 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
986 dentry->d_parent->d_name.name, dentry->d_name.name,
987 datasync);
988
989 mutex_lock(&inode->i_mutex);
990 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
991 mutex_unlock(&inode->i_mutex);
992 return 0;
993}
994
995/**
996 * nfs_force_lookup_revalidate - Mark the directory as having changed
997 * @dir - pointer to directory inode
998 *
999 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1000 * full lookup on all child dentries of 'dir' whenever a change occurs
1001 * on the server that might have invalidated our dcache.
1002 *
1003 * The caller should be holding dir->i_lock
1004 */
1005void nfs_force_lookup_revalidate(struct inode *dir)
1006{
1007 NFS_I(dir)->cache_change_attribute++;
1008}
1009
1010/*
1011 * A check for whether or not the parent directory has changed.
1012 * In the case it has, we assume that the dentries are untrustworthy
1013 * and may need to be looked up again.
1014 */
1015static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
1016{
1017 if (IS_ROOT(dentry))
1018 return 1;
1019 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1020 return 0;
1021 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1022 return 0;
1023 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1024 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1025 return 0;
1026 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1027 return 0;
1028 return 1;
1029}
1030
1031/*
1032 * Return the intent data that applies to this particular path component
1033 *
1034 * Note that the current set of intents only apply to the very last
1035 * component of the path and none of them is set before that last
1036 * component.
1037 */
1038static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd,
1039 unsigned int mask)
1040{
1041 return nd->flags & mask;
1042}
1043
1044/*
1045 * Use intent information to check whether or not we're going to do
1046 * an O_EXCL create using this path component.
1047 */
1048static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
1049{
1050 if (NFS_PROTO(dir)->version == 2)
1051 return 0;
1052 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
1053}
1054
1055/*
1056 * Inode and filehandle revalidation for lookups.
1057 *
1058 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1059 * or if the intent information indicates that we're about to open this
1060 * particular file and the "nocto" mount flag is not set.
1061 *
1062 */
1063static inline
1064int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
1065{
1066 struct nfs_server *server = NFS_SERVER(inode);
1067
1068 if (IS_AUTOMOUNT(inode))
1069 return 0;
1070 if (nd != NULL) {
1071 /* VFS wants an on-the-wire revalidation */
1072 if (nd->flags & LOOKUP_REVAL)
1073 goto out_force;
1074 /* This is an open(2) */
1075 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
1076 !(server->flags & NFS_MOUNT_NOCTO) &&
1077 (S_ISREG(inode->i_mode) ||
1078 S_ISDIR(inode->i_mode)))
1079 goto out_force;
1080 return 0;
1081 }
1082 return nfs_revalidate_inode(server, inode);
1083out_force:
1084 return __nfs_revalidate_inode(server, inode);
1085}
1086
1087/*
1088 * We judge how long we want to trust negative
1089 * dentries by looking at the parent inode mtime.
1090 *
1091 * If parent mtime has changed, we revalidate, else we wait for a
1092 * period corresponding to the parent's attribute cache timeout value.
1093 */
1094static inline
1095int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1096 struct nameidata *nd)
1097{
1098 /* Don't revalidate a negative dentry if we're creating a new file */
1099 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
1100 return 0;
1101 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1102 return 1;
1103 return !nfs_check_verifier(dir, dentry);
1104}
1105
1106/*
1107 * This is called every time the dcache has a lookup hit,
1108 * and we should check whether we can really trust that
1109 * lookup.
1110 *
1111 * NOTE! The hit can be a negative hit too, don't assume
1112 * we have an inode!
1113 *
1114 * If the parent directory is seen to have changed, we throw out the
1115 * cached dentry and do a new lookup.
1116 */
1117static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1118{
1119 struct inode *dir;
1120 struct inode *inode;
1121 struct dentry *parent;
1122 struct nfs_fh *fhandle = NULL;
1123 struct nfs_fattr *fattr = NULL;
1124 int error;
1125
1126 if (nd && (nd->flags & LOOKUP_RCU))
1127 return -ECHILD;
1128
1129 parent = dget_parent(dentry);
1130 dir = parent->d_inode;
1131 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1132 inode = dentry->d_inode;
1133
1134 if (!inode) {
1135 if (nfs_neg_need_reval(dir, dentry, nd))
1136 goto out_bad;
1137 goto out_valid_noent;
1138 }
1139
1140 if (is_bad_inode(inode)) {
1141 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1142 __func__, dentry->d_parent->d_name.name,
1143 dentry->d_name.name);
1144 goto out_bad;
1145 }
1146
1147 if (nfs_have_delegation(inode, FMODE_READ))
1148 goto out_set_verifier;
1149
1150 /* Force a full look up iff the parent directory has changed */
1151 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
1152 if (nfs_lookup_verify_inode(inode, nd))
1153 goto out_zap_parent;
1154 goto out_valid;
1155 }
1156
1157 if (NFS_STALE(inode))
1158 goto out_bad;
1159
1160 error = -ENOMEM;
1161 fhandle = nfs_alloc_fhandle();
1162 fattr = nfs_alloc_fattr();
1163 if (fhandle == NULL || fattr == NULL)
1164 goto out_error;
1165
1166 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1167 if (error)
1168 goto out_bad;
1169 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1170 goto out_bad;
1171 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1172 goto out_bad;
1173
1174 nfs_free_fattr(fattr);
1175 nfs_free_fhandle(fhandle);
1176out_set_verifier:
1177 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1178 out_valid:
1179 /* Success: notify readdir to use READDIRPLUS */
1180 nfs_advise_use_readdirplus(dir);
1181 out_valid_noent:
1182 dput(parent);
1183 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1184 __func__, dentry->d_parent->d_name.name,
1185 dentry->d_name.name);
1186 return 1;
1187out_zap_parent:
1188 nfs_zap_caches(dir);
1189 out_bad:
1190 nfs_mark_for_revalidate(dir);
1191 if (inode && S_ISDIR(inode->i_mode)) {
1192 /* Purge readdir caches. */
1193 nfs_zap_caches(inode);
1194 /* If we have submounts, don't unhash ! */
1195 if (have_submounts(dentry))
1196 goto out_valid;
1197 if (dentry->d_flags & DCACHE_DISCONNECTED)
1198 goto out_valid;
1199 shrink_dcache_parent(dentry);
1200 }
1201 d_drop(dentry);
1202 nfs_free_fattr(fattr);
1203 nfs_free_fhandle(fhandle);
1204 dput(parent);
1205 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1206 __func__, dentry->d_parent->d_name.name,
1207 dentry->d_name.name);
1208 return 0;
1209out_error:
1210 nfs_free_fattr(fattr);
1211 nfs_free_fhandle(fhandle);
1212 dput(parent);
1213 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1214 __func__, dentry->d_parent->d_name.name,
1215 dentry->d_name.name, error);
1216 return error;
1217}
1218
1219/*
1220 * This is called from dput() when d_count is going to 0.
1221 */
1222static int nfs_dentry_delete(const struct dentry *dentry)
1223{
1224 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1225 dentry->d_parent->d_name.name, dentry->d_name.name,
1226 dentry->d_flags);
1227
1228 /* Unhash any dentry with a stale inode */
1229 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1230 return 1;
1231
1232 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1233 /* Unhash it, so that ->d_iput() would be called */
1234 return 1;
1235 }
1236 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1237 /* Unhash it, so that ancestors of killed async unlink
1238 * files will be cleaned up during umount */
1239 return 1;
1240 }
1241 return 0;
1242
1243}
1244
1245static void nfs_drop_nlink(struct inode *inode)
1246{
1247 spin_lock(&inode->i_lock);
1248 if (inode->i_nlink > 0)
1249 drop_nlink(inode);
1250 spin_unlock(&inode->i_lock);
1251}
1252
1253/*
1254 * Called when the dentry loses inode.
1255 * We use it to clean up silly-renamed files.
1256 */
1257static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1258{
1259 if (S_ISDIR(inode->i_mode))
1260 /* drop any readdir cache as it could easily be old */
1261 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1262
1263 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1264 drop_nlink(inode);
1265 nfs_complete_unlink(dentry, inode);
1266 }
1267 iput(inode);
1268}
1269
1270static void nfs_d_release(struct dentry *dentry)
1271{
1272 /* free cached devname value, if it survived that far */
1273 if (unlikely(dentry->d_fsdata)) {
1274 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1275 WARN_ON(1);
1276 else
1277 kfree(dentry->d_fsdata);
1278 }
1279}
1280
1281const struct dentry_operations nfs_dentry_operations = {
1282 .d_revalidate = nfs_lookup_revalidate,
1283 .d_delete = nfs_dentry_delete,
1284 .d_iput = nfs_dentry_iput,
1285 .d_automount = nfs_d_automount,
1286 .d_release = nfs_d_release,
1287};
1288
1289static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1290{
1291 struct dentry *res;
1292 struct dentry *parent;
1293 struct inode *inode = NULL;
1294 struct nfs_fh *fhandle = NULL;
1295 struct nfs_fattr *fattr = NULL;
1296 int error;
1297
1298 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1299 dentry->d_parent->d_name.name, dentry->d_name.name);
1300 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1301
1302 res = ERR_PTR(-ENAMETOOLONG);
1303 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1304 goto out;
1305
1306 /*
1307 * If we're doing an exclusive create, optimize away the lookup
1308 * but don't hash the dentry.
1309 */
1310 if (nfs_is_exclusive_create(dir, nd)) {
1311 d_instantiate(dentry, NULL);
1312 res = NULL;
1313 goto out;
1314 }
1315
1316 res = ERR_PTR(-ENOMEM);
1317 fhandle = nfs_alloc_fhandle();
1318 fattr = nfs_alloc_fattr();
1319 if (fhandle == NULL || fattr == NULL)
1320 goto out;
1321
1322 parent = dentry->d_parent;
1323 /* Protect against concurrent sillydeletes */
1324 nfs_block_sillyrename(parent);
1325 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1326 if (error == -ENOENT)
1327 goto no_entry;
1328 if (error < 0) {
1329 res = ERR_PTR(error);
1330 goto out_unblock_sillyrename;
1331 }
1332 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1333 res = ERR_CAST(inode);
1334 if (IS_ERR(res))
1335 goto out_unblock_sillyrename;
1336
1337 /* Success: notify readdir to use READDIRPLUS */
1338 nfs_advise_use_readdirplus(dir);
1339
1340no_entry:
1341 res = d_materialise_unique(dentry, inode);
1342 if (res != NULL) {
1343 if (IS_ERR(res))
1344 goto out_unblock_sillyrename;
1345 dentry = res;
1346 }
1347 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1348out_unblock_sillyrename:
1349 nfs_unblock_sillyrename(parent);
1350out:
1351 nfs_free_fattr(fattr);
1352 nfs_free_fhandle(fhandle);
1353 return res;
1354}
1355
1356#ifdef CONFIG_NFS_V4
1357static int nfs4_lookup_revalidate(struct dentry *, struct nameidata *);
1358
1359const struct dentry_operations nfs4_dentry_operations = {
1360 .d_revalidate = nfs4_lookup_revalidate,
1361 .d_delete = nfs_dentry_delete,
1362 .d_iput = nfs_dentry_iput,
1363 .d_automount = nfs_d_automount,
1364 .d_release = nfs_d_release,
1365};
1366
1367/*
1368 * Use intent information to determine whether we need to substitute
1369 * the NFSv4-style stateful OPEN for the LOOKUP call
1370 */
1371static int is_atomic_open(struct nameidata *nd)
1372{
1373 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
1374 return 0;
1375 /* NFS does not (yet) have a stateful open for directories */
1376 if (nd->flags & LOOKUP_DIRECTORY)
1377 return 0;
1378 /* Are we trying to write to a read only partition? */
1379 if (__mnt_is_readonly(nd->path.mnt) &&
1380 (nd->intent.open.flags & (O_CREAT|O_TRUNC|O_ACCMODE)))
1381 return 0;
1382 return 1;
1383}
1384
1385static fmode_t flags_to_mode(int flags)
1386{
1387 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1388 if ((flags & O_ACCMODE) != O_WRONLY)
1389 res |= FMODE_READ;
1390 if ((flags & O_ACCMODE) != O_RDONLY)
1391 res |= FMODE_WRITE;
1392 return res;
1393}
1394
1395static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1396{
1397 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1398}
1399
1400static int do_open(struct inode *inode, struct file *filp)
1401{
1402 nfs_fscache_set_inode_cookie(inode, filp);
1403 return 0;
1404}
1405
1406static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1407{
1408 struct file *filp;
1409 int ret = 0;
1410
1411 /* If the open_intent is for execute, we have an extra check to make */
1412 if (ctx->mode & FMODE_EXEC) {
1413 ret = nfs_may_open(ctx->dentry->d_inode,
1414 ctx->cred,
1415 nd->intent.open.flags);
1416 if (ret < 0)
1417 goto out;
1418 }
1419 filp = lookup_instantiate_filp(nd, ctx->dentry, do_open);
1420 if (IS_ERR(filp))
1421 ret = PTR_ERR(filp);
1422 else
1423 nfs_file_set_open_context(filp, ctx);
1424out:
1425 put_nfs_open_context(ctx);
1426 return ret;
1427}
1428
1429static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1430{
1431 struct nfs_open_context *ctx;
1432 struct iattr attr;
1433 struct dentry *res = NULL;
1434 struct inode *inode;
1435 int open_flags;
1436 int err;
1437
1438 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1439 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1440
1441 /* Check that we are indeed trying to open this file */
1442 if (!is_atomic_open(nd))
1443 goto no_open;
1444
1445 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1446 res = ERR_PTR(-ENAMETOOLONG);
1447 goto out;
1448 }
1449
1450 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1451 * the dentry. */
1452 if (nd->flags & LOOKUP_EXCL) {
1453 d_instantiate(dentry, NULL);
1454 goto out;
1455 }
1456
1457 open_flags = nd->intent.open.flags;
1458 attr.ia_valid = ATTR_OPEN;
1459
1460 ctx = create_nfs_open_context(dentry, open_flags);
1461 res = ERR_CAST(ctx);
1462 if (IS_ERR(ctx))
1463 goto out;
1464
1465 if (nd->flags & LOOKUP_CREATE) {
1466 attr.ia_mode = nd->intent.open.create_mode;
1467 attr.ia_valid |= ATTR_MODE;
1468 attr.ia_mode &= ~current_umask();
1469 } else
1470 open_flags &= ~(O_EXCL | O_CREAT);
1471
1472 if (open_flags & O_TRUNC) {
1473 attr.ia_valid |= ATTR_SIZE;
1474 attr.ia_size = 0;
1475 }
1476
1477 /* Open the file on the server */
1478 nfs_block_sillyrename(dentry->d_parent);
1479 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1480 if (IS_ERR(inode)) {
1481 nfs_unblock_sillyrename(dentry->d_parent);
1482 put_nfs_open_context(ctx);
1483 switch (PTR_ERR(inode)) {
1484 /* Make a negative dentry */
1485 case -ENOENT:
1486 d_add(dentry, NULL);
1487 res = NULL;
1488 goto out;
1489 /* This turned out not to be a regular file */
1490 case -EISDIR:
1491 case -ENOTDIR:
1492 goto no_open;
1493 case -ELOOP:
1494 if (!(nd->intent.open.flags & O_NOFOLLOW))
1495 goto no_open;
1496 /* case -EINVAL: */
1497 default:
1498 res = ERR_CAST(inode);
1499 goto out;
1500 }
1501 }
1502 res = d_add_unique(dentry, inode);
1503 nfs_unblock_sillyrename(dentry->d_parent);
1504 if (res != NULL) {
1505 dput(ctx->dentry);
1506 ctx->dentry = dget(res);
1507 dentry = res;
1508 }
1509 err = nfs_intent_set_file(nd, ctx);
1510 if (err < 0) {
1511 if (res != NULL)
1512 dput(res);
1513 return ERR_PTR(err);
1514 }
1515out:
1516 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1517 return res;
1518no_open:
1519 return nfs_lookup(dir, dentry, nd);
1520}
1521
1522static int nfs4_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
1523{
1524 struct dentry *parent = NULL;
1525 struct inode *inode;
1526 struct inode *dir;
1527 int openflags, ret = 0;
1528
1529 if (nd && (nd->flags & LOOKUP_RCU))
1530 return -ECHILD;
1531
1532 inode = dentry->d_inode;
1533 if (!is_atomic_open(nd) || d_mountpoint(dentry))
1534 goto no_open;
1535
1536 parent = dget_parent(dentry);
1537 dir = parent->d_inode;
1538
1539 /* We can't create new files in nfs_open_revalidate(), so we
1540 * optimize away revalidation of negative dentries.
1541 */
1542 if (inode == NULL) {
1543 if (!nfs_neg_need_reval(dir, dentry, nd))
1544 ret = 1;
1545 goto out;
1546 }
1547
1548 /* NFS only supports OPEN on regular files */
1549 if (!S_ISREG(inode->i_mode))
1550 goto no_open_dput;
1551 openflags = nd->intent.open.flags;
1552 /* We cannot do exclusive creation on a positive dentry */
1553 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1554 goto no_open_dput;
1555
1556 /* Let f_op->open() actually open (and revalidate) the file */
1557 ret = 1;
1558
1559out:
1560 dput(parent);
1561 return ret;
1562
1563no_open_dput:
1564 dput(parent);
1565no_open:
1566 return nfs_lookup_revalidate(dentry, nd);
1567}
1568
1569static int nfs_open_create(struct inode *dir, struct dentry *dentry,
1570 umode_t mode, struct nameidata *nd)
1571{
1572 struct nfs_open_context *ctx = NULL;
1573 struct iattr attr;
1574 int error;
1575 int open_flags = O_CREAT|O_EXCL;
1576
1577 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1578 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1579
1580 attr.ia_mode = mode;
1581 attr.ia_valid = ATTR_MODE;
1582
1583 if (nd)
1584 open_flags = nd->intent.open.flags;
1585
1586 ctx = create_nfs_open_context(dentry, open_flags);
1587 error = PTR_ERR(ctx);
1588 if (IS_ERR(ctx))
1589 goto out_err_drop;
1590
1591 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1592 if (error != 0)
1593 goto out_put_ctx;
1594 if (nd) {
1595 error = nfs_intent_set_file(nd, ctx);
1596 if (error < 0)
1597 goto out_err;
1598 } else {
1599 put_nfs_open_context(ctx);
1600 }
1601 return 0;
1602out_put_ctx:
1603 put_nfs_open_context(ctx);
1604out_err_drop:
1605 d_drop(dentry);
1606out_err:
1607 return error;
1608}
1609
1610#endif /* CONFIG_NFSV4 */
1611
1612/*
1613 * Code common to create, mkdir, and mknod.
1614 */
1615int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1616 struct nfs_fattr *fattr)
1617{
1618 struct dentry *parent = dget_parent(dentry);
1619 struct inode *dir = parent->d_inode;
1620 struct inode *inode;
1621 int error = -EACCES;
1622
1623 d_drop(dentry);
1624
1625 /* We may have been initialized further down */
1626 if (dentry->d_inode)
1627 goto out;
1628 if (fhandle->size == 0) {
1629 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1630 if (error)
1631 goto out_error;
1632 }
1633 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1634 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1635 struct nfs_server *server = NFS_SB(dentry->d_sb);
1636 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1637 if (error < 0)
1638 goto out_error;
1639 }
1640 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1641 error = PTR_ERR(inode);
1642 if (IS_ERR(inode))
1643 goto out_error;
1644 d_add(dentry, inode);
1645out:
1646 dput(parent);
1647 return 0;
1648out_error:
1649 nfs_mark_for_revalidate(dir);
1650 dput(parent);
1651 return error;
1652}
1653
1654/*
1655 * Following a failed create operation, we drop the dentry rather
1656 * than retain a negative dentry. This avoids a problem in the event
1657 * that the operation succeeded on the server, but an error in the
1658 * reply path made it appear to have failed.
1659 */
1660static int nfs_create(struct inode *dir, struct dentry *dentry,
1661 umode_t mode, struct nameidata *nd)
1662{
1663 struct iattr attr;
1664 int error;
1665 int open_flags = O_CREAT|O_EXCL;
1666
1667 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1668 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1669
1670 attr.ia_mode = mode;
1671 attr.ia_valid = ATTR_MODE;
1672
1673 if (nd)
1674 open_flags = nd->intent.open.flags;
1675
1676 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, NULL);
1677 if (error != 0)
1678 goto out_err;
1679 return 0;
1680out_err:
1681 d_drop(dentry);
1682 return error;
1683}
1684
1685/*
1686 * See comments for nfs_proc_create regarding failed operations.
1687 */
1688static int
1689nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1690{
1691 struct iattr attr;
1692 int status;
1693
1694 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1695 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1696
1697 if (!new_valid_dev(rdev))
1698 return -EINVAL;
1699
1700 attr.ia_mode = mode;
1701 attr.ia_valid = ATTR_MODE;
1702
1703 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1704 if (status != 0)
1705 goto out_err;
1706 return 0;
1707out_err:
1708 d_drop(dentry);
1709 return status;
1710}
1711
1712/*
1713 * See comments for nfs_proc_create regarding failed operations.
1714 */
1715static int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1716{
1717 struct iattr attr;
1718 int error;
1719
1720 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1721 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1722
1723 attr.ia_valid = ATTR_MODE;
1724 attr.ia_mode = mode | S_IFDIR;
1725
1726 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1727 if (error != 0)
1728 goto out_err;
1729 return 0;
1730out_err:
1731 d_drop(dentry);
1732 return error;
1733}
1734
1735static void nfs_dentry_handle_enoent(struct dentry *dentry)
1736{
1737 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1738 d_delete(dentry);
1739}
1740
1741static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1742{
1743 int error;
1744
1745 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1746 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1747
1748 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1749 /* Ensure the VFS deletes this inode */
1750 if (error == 0 && dentry->d_inode != NULL)
1751 clear_nlink(dentry->d_inode);
1752 else if (error == -ENOENT)
1753 nfs_dentry_handle_enoent(dentry);
1754
1755 return error;
1756}
1757
1758/*
1759 * Remove a file after making sure there are no pending writes,
1760 * and after checking that the file has only one user.
1761 *
1762 * We invalidate the attribute cache and free the inode prior to the operation
1763 * to avoid possible races if the server reuses the inode.
1764 */
1765static int nfs_safe_remove(struct dentry *dentry)
1766{
1767 struct inode *dir = dentry->d_parent->d_inode;
1768 struct inode *inode = dentry->d_inode;
1769 int error = -EBUSY;
1770
1771 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1772 dentry->d_parent->d_name.name, dentry->d_name.name);
1773
1774 /* If the dentry was sillyrenamed, we simply call d_delete() */
1775 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1776 error = 0;
1777 goto out;
1778 }
1779
1780 if (inode != NULL) {
1781 nfs_inode_return_delegation(inode);
1782 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1783 /* The VFS may want to delete this inode */
1784 if (error == 0)
1785 nfs_drop_nlink(inode);
1786 nfs_mark_for_revalidate(inode);
1787 } else
1788 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1789 if (error == -ENOENT)
1790 nfs_dentry_handle_enoent(dentry);
1791out:
1792 return error;
1793}
1794
1795/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1796 * belongs to an active ".nfs..." file and we return -EBUSY.
1797 *
1798 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1799 */
1800static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1801{
1802 int error;
1803 int need_rehash = 0;
1804
1805 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1806 dir->i_ino, dentry->d_name.name);
1807
1808 spin_lock(&dentry->d_lock);
1809 if (dentry->d_count > 1) {
1810 spin_unlock(&dentry->d_lock);
1811 /* Start asynchronous writeout of the inode */
1812 write_inode_now(dentry->d_inode, 0);
1813 error = nfs_sillyrename(dir, dentry);
1814 return error;
1815 }
1816 if (!d_unhashed(dentry)) {
1817 __d_drop(dentry);
1818 need_rehash = 1;
1819 }
1820 spin_unlock(&dentry->d_lock);
1821 error = nfs_safe_remove(dentry);
1822 if (!error || error == -ENOENT) {
1823 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1824 } else if (need_rehash)
1825 d_rehash(dentry);
1826 return error;
1827}
1828
1829/*
1830 * To create a symbolic link, most file systems instantiate a new inode,
1831 * add a page to it containing the path, then write it out to the disk
1832 * using prepare_write/commit_write.
1833 *
1834 * Unfortunately the NFS client can't create the in-core inode first
1835 * because it needs a file handle to create an in-core inode (see
1836 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1837 * symlink request has completed on the server.
1838 *
1839 * So instead we allocate a raw page, copy the symname into it, then do
1840 * the SYMLINK request with the page as the buffer. If it succeeds, we
1841 * now have a new file handle and can instantiate an in-core NFS inode
1842 * and move the raw page into its mapping.
1843 */
1844static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1845{
1846 struct pagevec lru_pvec;
1847 struct page *page;
1848 char *kaddr;
1849 struct iattr attr;
1850 unsigned int pathlen = strlen(symname);
1851 int error;
1852
1853 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1854 dir->i_ino, dentry->d_name.name, symname);
1855
1856 if (pathlen > PAGE_SIZE)
1857 return -ENAMETOOLONG;
1858
1859 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1860 attr.ia_valid = ATTR_MODE;
1861
1862 page = alloc_page(GFP_HIGHUSER);
1863 if (!page)
1864 return -ENOMEM;
1865
1866 kaddr = kmap_atomic(page);
1867 memcpy(kaddr, symname, pathlen);
1868 if (pathlen < PAGE_SIZE)
1869 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1870 kunmap_atomic(kaddr);
1871
1872 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1873 if (error != 0) {
1874 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1875 dir->i_sb->s_id, dir->i_ino,
1876 dentry->d_name.name, symname, error);
1877 d_drop(dentry);
1878 __free_page(page);
1879 return error;
1880 }
1881
1882 /*
1883 * No big deal if we can't add this page to the page cache here.
1884 * READLINK will get the missing page from the server if needed.
1885 */
1886 pagevec_init(&lru_pvec, 0);
1887 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1888 GFP_KERNEL)) {
1889 pagevec_add(&lru_pvec, page);
1890 pagevec_lru_add_file(&lru_pvec);
1891 SetPageUptodate(page);
1892 unlock_page(page);
1893 } else
1894 __free_page(page);
1895
1896 return 0;
1897}
1898
1899static int
1900nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1901{
1902 struct inode *inode = old_dentry->d_inode;
1903 int error;
1904
1905 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1906 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1907 dentry->d_parent->d_name.name, dentry->d_name.name);
1908
1909 nfs_inode_return_delegation(inode);
1910
1911 d_drop(dentry);
1912 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1913 if (error == 0) {
1914 ihold(inode);
1915 d_add(dentry, inode);
1916 }
1917 return error;
1918}
1919
1920/*
1921 * RENAME
1922 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1923 * different file handle for the same inode after a rename (e.g. when
1924 * moving to a different directory). A fail-safe method to do so would
1925 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1926 * rename the old file using the sillyrename stuff. This way, the original
1927 * file in old_dir will go away when the last process iput()s the inode.
1928 *
1929 * FIXED.
1930 *
1931 * It actually works quite well. One needs to have the possibility for
1932 * at least one ".nfs..." file in each directory the file ever gets
1933 * moved or linked to which happens automagically with the new
1934 * implementation that only depends on the dcache stuff instead of
1935 * using the inode layer
1936 *
1937 * Unfortunately, things are a little more complicated than indicated
1938 * above. For a cross-directory move, we want to make sure we can get
1939 * rid of the old inode after the operation. This means there must be
1940 * no pending writes (if it's a file), and the use count must be 1.
1941 * If these conditions are met, we can drop the dentries before doing
1942 * the rename.
1943 */
1944static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1945 struct inode *new_dir, struct dentry *new_dentry)
1946{
1947 struct inode *old_inode = old_dentry->d_inode;
1948 struct inode *new_inode = new_dentry->d_inode;
1949 struct dentry *dentry = NULL, *rehash = NULL;
1950 int error = -EBUSY;
1951
1952 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1953 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1954 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1955 new_dentry->d_count);
1956
1957 /*
1958 * For non-directories, check whether the target is busy and if so,
1959 * make a copy of the dentry and then do a silly-rename. If the
1960 * silly-rename succeeds, the copied dentry is hashed and becomes
1961 * the new target.
1962 */
1963 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1964 /*
1965 * To prevent any new references to the target during the
1966 * rename, we unhash the dentry in advance.
1967 */
1968 if (!d_unhashed(new_dentry)) {
1969 d_drop(new_dentry);
1970 rehash = new_dentry;
1971 }
1972
1973 if (new_dentry->d_count > 2) {
1974 int err;
1975
1976 /* copy the target dentry's name */
1977 dentry = d_alloc(new_dentry->d_parent,
1978 &new_dentry->d_name);
1979 if (!dentry)
1980 goto out;
1981
1982 /* silly-rename the existing target ... */
1983 err = nfs_sillyrename(new_dir, new_dentry);
1984 if (err)
1985 goto out;
1986
1987 new_dentry = dentry;
1988 rehash = NULL;
1989 new_inode = NULL;
1990 }
1991 }
1992
1993 nfs_inode_return_delegation(old_inode);
1994 if (new_inode != NULL)
1995 nfs_inode_return_delegation(new_inode);
1996
1997 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1998 new_dir, &new_dentry->d_name);
1999 nfs_mark_for_revalidate(old_inode);
2000out:
2001 if (rehash)
2002 d_rehash(rehash);
2003 if (!error) {
2004 if (new_inode != NULL)
2005 nfs_drop_nlink(new_inode);
2006 d_move(old_dentry, new_dentry);
2007 nfs_set_verifier(new_dentry,
2008 nfs_save_change_attribute(new_dir));
2009 } else if (error == -ENOENT)
2010 nfs_dentry_handle_enoent(old_dentry);
2011
2012 /* new dentry created? */
2013 if (dentry)
2014 dput(dentry);
2015 return error;
2016}
2017
2018static DEFINE_SPINLOCK(nfs_access_lru_lock);
2019static LIST_HEAD(nfs_access_lru_list);
2020static atomic_long_t nfs_access_nr_entries;
2021
2022static void nfs_access_free_entry(struct nfs_access_entry *entry)
2023{
2024 put_rpccred(entry->cred);
2025 kfree(entry);
2026 smp_mb__before_atomic_dec();
2027 atomic_long_dec(&nfs_access_nr_entries);
2028 smp_mb__after_atomic_dec();
2029}
2030
2031static void nfs_access_free_list(struct list_head *head)
2032{
2033 struct nfs_access_entry *cache;
2034
2035 while (!list_empty(head)) {
2036 cache = list_entry(head->next, struct nfs_access_entry, lru);
2037 list_del(&cache->lru);
2038 nfs_access_free_entry(cache);
2039 }
2040}
2041
2042int nfs_access_cache_shrinker(struct shrinker *shrink,
2043 struct shrink_control *sc)
2044{
2045 LIST_HEAD(head);
2046 struct nfs_inode *nfsi, *next;
2047 struct nfs_access_entry *cache;
2048 int nr_to_scan = sc->nr_to_scan;
2049 gfp_t gfp_mask = sc->gfp_mask;
2050
2051 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2052 return (nr_to_scan == 0) ? 0 : -1;
2053
2054 spin_lock(&nfs_access_lru_lock);
2055 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2056 struct inode *inode;
2057
2058 if (nr_to_scan-- == 0)
2059 break;
2060 inode = &nfsi->vfs_inode;
2061 spin_lock(&inode->i_lock);
2062 if (list_empty(&nfsi->access_cache_entry_lru))
2063 goto remove_lru_entry;
2064 cache = list_entry(nfsi->access_cache_entry_lru.next,
2065 struct nfs_access_entry, lru);
2066 list_move(&cache->lru, &head);
2067 rb_erase(&cache->rb_node, &nfsi->access_cache);
2068 if (!list_empty(&nfsi->access_cache_entry_lru))
2069 list_move_tail(&nfsi->access_cache_inode_lru,
2070 &nfs_access_lru_list);
2071 else {
2072remove_lru_entry:
2073 list_del_init(&nfsi->access_cache_inode_lru);
2074 smp_mb__before_clear_bit();
2075 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2076 smp_mb__after_clear_bit();
2077 }
2078 spin_unlock(&inode->i_lock);
2079 }
2080 spin_unlock(&nfs_access_lru_lock);
2081 nfs_access_free_list(&head);
2082 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2083}
2084
2085static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2086{
2087 struct rb_root *root_node = &nfsi->access_cache;
2088 struct rb_node *n;
2089 struct nfs_access_entry *entry;
2090
2091 /* Unhook entries from the cache */
2092 while ((n = rb_first(root_node)) != NULL) {
2093 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2094 rb_erase(n, root_node);
2095 list_move(&entry->lru, head);
2096 }
2097 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2098}
2099
2100void nfs_access_zap_cache(struct inode *inode)
2101{
2102 LIST_HEAD(head);
2103
2104 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2105 return;
2106 /* Remove from global LRU init */
2107 spin_lock(&nfs_access_lru_lock);
2108 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2109 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2110
2111 spin_lock(&inode->i_lock);
2112 __nfs_access_zap_cache(NFS_I(inode), &head);
2113 spin_unlock(&inode->i_lock);
2114 spin_unlock(&nfs_access_lru_lock);
2115 nfs_access_free_list(&head);
2116}
2117
2118static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2119{
2120 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2121 struct nfs_access_entry *entry;
2122
2123 while (n != NULL) {
2124 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2125
2126 if (cred < entry->cred)
2127 n = n->rb_left;
2128 else if (cred > entry->cred)
2129 n = n->rb_right;
2130 else
2131 return entry;
2132 }
2133 return NULL;
2134}
2135
2136static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2137{
2138 struct nfs_inode *nfsi = NFS_I(inode);
2139 struct nfs_access_entry *cache;
2140 int err = -ENOENT;
2141
2142 spin_lock(&inode->i_lock);
2143 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2144 goto out_zap;
2145 cache = nfs_access_search_rbtree(inode, cred);
2146 if (cache == NULL)
2147 goto out;
2148 if (!nfs_have_delegated_attributes(inode) &&
2149 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2150 goto out_stale;
2151 res->jiffies = cache->jiffies;
2152 res->cred = cache->cred;
2153 res->mask = cache->mask;
2154 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2155 err = 0;
2156out:
2157 spin_unlock(&inode->i_lock);
2158 return err;
2159out_stale:
2160 rb_erase(&cache->rb_node, &nfsi->access_cache);
2161 list_del(&cache->lru);
2162 spin_unlock(&inode->i_lock);
2163 nfs_access_free_entry(cache);
2164 return -ENOENT;
2165out_zap:
2166 spin_unlock(&inode->i_lock);
2167 nfs_access_zap_cache(inode);
2168 return -ENOENT;
2169}
2170
2171static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2172{
2173 struct nfs_inode *nfsi = NFS_I(inode);
2174 struct rb_root *root_node = &nfsi->access_cache;
2175 struct rb_node **p = &root_node->rb_node;
2176 struct rb_node *parent = NULL;
2177 struct nfs_access_entry *entry;
2178
2179 spin_lock(&inode->i_lock);
2180 while (*p != NULL) {
2181 parent = *p;
2182 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2183
2184 if (set->cred < entry->cred)
2185 p = &parent->rb_left;
2186 else if (set->cred > entry->cred)
2187 p = &parent->rb_right;
2188 else
2189 goto found;
2190 }
2191 rb_link_node(&set->rb_node, parent, p);
2192 rb_insert_color(&set->rb_node, root_node);
2193 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2194 spin_unlock(&inode->i_lock);
2195 return;
2196found:
2197 rb_replace_node(parent, &set->rb_node, root_node);
2198 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2199 list_del(&entry->lru);
2200 spin_unlock(&inode->i_lock);
2201 nfs_access_free_entry(entry);
2202}
2203
2204static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2205{
2206 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2207 if (cache == NULL)
2208 return;
2209 RB_CLEAR_NODE(&cache->rb_node);
2210 cache->jiffies = set->jiffies;
2211 cache->cred = get_rpccred(set->cred);
2212 cache->mask = set->mask;
2213
2214 nfs_access_add_rbtree(inode, cache);
2215
2216 /* Update accounting */
2217 smp_mb__before_atomic_inc();
2218 atomic_long_inc(&nfs_access_nr_entries);
2219 smp_mb__after_atomic_inc();
2220
2221 /* Add inode to global LRU list */
2222 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2223 spin_lock(&nfs_access_lru_lock);
2224 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2225 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2226 &nfs_access_lru_list);
2227 spin_unlock(&nfs_access_lru_lock);
2228 }
2229}
2230
2231static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2232{
2233 struct nfs_access_entry cache;
2234 int status;
2235
2236 status = nfs_access_get_cached(inode, cred, &cache);
2237 if (status == 0)
2238 goto out;
2239
2240 /* Be clever: ask server to check for all possible rights */
2241 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2242 cache.cred = cred;
2243 cache.jiffies = jiffies;
2244 status = NFS_PROTO(inode)->access(inode, &cache);
2245 if (status != 0) {
2246 if (status == -ESTALE) {
2247 nfs_zap_caches(inode);
2248 if (!S_ISDIR(inode->i_mode))
2249 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2250 }
2251 return status;
2252 }
2253 nfs_access_add_cache(inode, &cache);
2254out:
2255 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2256 return 0;
2257 return -EACCES;
2258}
2259
2260static int nfs_open_permission_mask(int openflags)
2261{
2262 int mask = 0;
2263
2264 if ((openflags & O_ACCMODE) != O_WRONLY)
2265 mask |= MAY_READ;
2266 if ((openflags & O_ACCMODE) != O_RDONLY)
2267 mask |= MAY_WRITE;
2268 if (openflags & __FMODE_EXEC)
2269 mask |= MAY_EXEC;
2270 return mask;
2271}
2272
2273int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2274{
2275 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2276}
2277
2278int nfs_permission(struct inode *inode, int mask)
2279{
2280 struct rpc_cred *cred;
2281 int res = 0;
2282
2283 if (mask & MAY_NOT_BLOCK)
2284 return -ECHILD;
2285
2286 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2287
2288 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2289 goto out;
2290 /* Is this sys_access() ? */
2291 if (mask & (MAY_ACCESS | MAY_CHDIR))
2292 goto force_lookup;
2293
2294 switch (inode->i_mode & S_IFMT) {
2295 case S_IFLNK:
2296 goto out;
2297 case S_IFREG:
2298 /* NFSv4 has atomic_open... */
2299 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
2300 && (mask & MAY_OPEN)
2301 && !(mask & MAY_EXEC))
2302 goto out;
2303 break;
2304 case S_IFDIR:
2305 /*
2306 * Optimize away all write operations, since the server
2307 * will check permissions when we perform the op.
2308 */
2309 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2310 goto out;
2311 }
2312
2313force_lookup:
2314 if (!NFS_PROTO(inode)->access)
2315 goto out_notsup;
2316
2317 cred = rpc_lookup_cred();
2318 if (!IS_ERR(cred)) {
2319 res = nfs_do_access(inode, cred, mask);
2320 put_rpccred(cred);
2321 } else
2322 res = PTR_ERR(cred);
2323out:
2324 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2325 res = -EACCES;
2326
2327 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2328 inode->i_sb->s_id, inode->i_ino, mask, res);
2329 return res;
2330out_notsup:
2331 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2332 if (res == 0)
2333 res = generic_permission(inode, mask);
2334 goto out;
2335}
2336
2337/*
2338 * Local variables:
2339 * version-control: t
2340 * kept-new-versions: 5
2341 * End:
2342 */