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 *);
59static int nfs_do_create(struct inode *dir, struct dentry *dentry,
60 umode_t mode, int open_flags);
61
62const struct file_operations nfs_dir_operations = {
63 .llseek = nfs_llseek_dir,
64 .read = generic_read_dir,
65 .iterate_shared = nfs_readdir,
66 .open = nfs_opendir,
67 .release = nfs_closedir,
68 .fsync = nfs_fsync_dir,
69};
70
71const struct address_space_operations nfs_dir_aops = {
72 .free_folio = nfs_readdir_clear_array,
73};
74
75#define NFS_INIT_DTSIZE PAGE_SIZE
76
77static struct nfs_open_dir_context *
78alloc_nfs_open_dir_context(struct inode *dir)
79{
80 struct nfs_inode *nfsi = NFS_I(dir);
81 struct nfs_open_dir_context *ctx;
82
83 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
84 if (ctx != NULL) {
85 ctx->attr_gencount = nfsi->attr_gencount;
86 ctx->dtsize = NFS_INIT_DTSIZE;
87 spin_lock(&dir->i_lock);
88 if (list_empty(&nfsi->open_files) &&
89 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
90 nfs_set_cache_invalid(dir,
91 NFS_INO_INVALID_DATA |
92 NFS_INO_REVAL_FORCED);
93 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
94 memcpy(ctx->verf, nfsi->cookieverf, sizeof(ctx->verf));
95 spin_unlock(&dir->i_lock);
96 return ctx;
97 }
98 return ERR_PTR(-ENOMEM);
99}
100
101static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
102{
103 spin_lock(&dir->i_lock);
104 list_del_rcu(&ctx->list);
105 spin_unlock(&dir->i_lock);
106 kfree_rcu(ctx, rcu_head);
107}
108
109/*
110 * Open file
111 */
112static int
113nfs_opendir(struct inode *inode, struct file *filp)
114{
115 int res = 0;
116 struct nfs_open_dir_context *ctx;
117
118 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
119
120 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
121
122 ctx = alloc_nfs_open_dir_context(inode);
123 if (IS_ERR(ctx)) {
124 res = PTR_ERR(ctx);
125 goto out;
126 }
127 filp->private_data = ctx;
128out:
129 return res;
130}
131
132static int
133nfs_closedir(struct inode *inode, struct file *filp)
134{
135 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
136 return 0;
137}
138
139struct nfs_cache_array_entry {
140 u64 cookie;
141 u64 ino;
142 const char *name;
143 unsigned int name_len;
144 unsigned char d_type;
145};
146
147struct nfs_cache_array {
148 u64 change_attr;
149 u64 last_cookie;
150 unsigned int size;
151 unsigned char folio_full : 1,
152 folio_is_eof : 1,
153 cookies_are_ordered : 1;
154 struct nfs_cache_array_entry array[] __counted_by(size);
155};
156
157struct nfs_readdir_descriptor {
158 struct file *file;
159 struct folio *folio;
160 struct dir_context *ctx;
161 pgoff_t folio_index;
162 pgoff_t folio_index_max;
163 u64 dir_cookie;
164 u64 last_cookie;
165 loff_t current_index;
166
167 __be32 verf[NFS_DIR_VERIFIER_SIZE];
168 unsigned long dir_verifier;
169 unsigned long timestamp;
170 unsigned long gencount;
171 unsigned long attr_gencount;
172 unsigned int cache_entry_index;
173 unsigned int buffer_fills;
174 unsigned int dtsize;
175 bool clear_cache;
176 bool plus;
177 bool eob;
178 bool eof;
179};
180
181static void nfs_set_dtsize(struct nfs_readdir_descriptor *desc, unsigned int sz)
182{
183 struct nfs_server *server = NFS_SERVER(file_inode(desc->file));
184 unsigned int maxsize = server->dtsize;
185
186 if (sz > maxsize)
187 sz = maxsize;
188 if (sz < NFS_MIN_FILE_IO_SIZE)
189 sz = NFS_MIN_FILE_IO_SIZE;
190 desc->dtsize = sz;
191}
192
193static void nfs_shrink_dtsize(struct nfs_readdir_descriptor *desc)
194{
195 nfs_set_dtsize(desc, desc->dtsize >> 1);
196}
197
198static void nfs_grow_dtsize(struct nfs_readdir_descriptor *desc)
199{
200 nfs_set_dtsize(desc, desc->dtsize << 1);
201}
202
203static void nfs_readdir_folio_init_array(struct folio *folio, u64 last_cookie,
204 u64 change_attr)
205{
206 struct nfs_cache_array *array;
207
208 array = kmap_local_folio(folio, 0);
209 array->change_attr = change_attr;
210 array->last_cookie = last_cookie;
211 array->size = 0;
212 array->folio_full = 0;
213 array->folio_is_eof = 0;
214 array->cookies_are_ordered = 1;
215 kunmap_local(array);
216}
217
218/*
219 * we are freeing strings created by nfs_add_to_readdir_array()
220 */
221static void nfs_readdir_clear_array(struct folio *folio)
222{
223 struct nfs_cache_array *array;
224 unsigned int i;
225
226 array = kmap_local_folio(folio, 0);
227 for (i = 0; i < array->size; i++)
228 kfree(array->array[i].name);
229 array->size = 0;
230 kunmap_local(array);
231}
232
233static void nfs_readdir_folio_reinit_array(struct folio *folio, u64 last_cookie,
234 u64 change_attr)
235{
236 nfs_readdir_clear_array(folio);
237 nfs_readdir_folio_init_array(folio, last_cookie, change_attr);
238}
239
240static struct folio *
241nfs_readdir_folio_array_alloc(u64 last_cookie, gfp_t gfp_flags)
242{
243 struct folio *folio = folio_alloc(gfp_flags, 0);
244 if (folio)
245 nfs_readdir_folio_init_array(folio, last_cookie, 0);
246 return folio;
247}
248
249static void nfs_readdir_folio_array_free(struct folio *folio)
250{
251 if (folio) {
252 nfs_readdir_clear_array(folio);
253 folio_put(folio);
254 }
255}
256
257static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array)
258{
259 return array->size == 0 ? array->last_cookie : array->array[0].cookie;
260}
261
262static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
263{
264 array->folio_is_eof = 1;
265 array->folio_full = 1;
266}
267
268static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
269{
270 return array->folio_full;
271}
272
273/*
274 * the caller is responsible for freeing qstr.name
275 * when called by nfs_readdir_add_to_array, the strings will be freed in
276 * nfs_clear_readdir_array()
277 */
278static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
279{
280 const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
281
282 /*
283 * Avoid a kmemleak false positive. The pointer to the name is stored
284 * in a page cache page which kmemleak does not scan.
285 */
286 if (ret != NULL)
287 kmemleak_not_leak(ret);
288 return ret;
289}
290
291static size_t nfs_readdir_array_maxentries(void)
292{
293 return (PAGE_SIZE - sizeof(struct nfs_cache_array)) /
294 sizeof(struct nfs_cache_array_entry);
295}
296
297/*
298 * Check that the next array entry lies entirely within the page bounds
299 */
300static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
301{
302 if (array->folio_full)
303 return -ENOSPC;
304 if (array->size == nfs_readdir_array_maxentries()) {
305 array->folio_full = 1;
306 return -ENOSPC;
307 }
308 return 0;
309}
310
311static int nfs_readdir_folio_array_append(struct folio *folio,
312 const struct nfs_entry *entry,
313 u64 *cookie)
314{
315 struct nfs_cache_array *array;
316 struct nfs_cache_array_entry *cache_entry;
317 const char *name;
318 int ret = -ENOMEM;
319
320 name = nfs_readdir_copy_name(entry->name, entry->len);
321
322 array = kmap_local_folio(folio, 0);
323 if (!name)
324 goto out;
325 ret = nfs_readdir_array_can_expand(array);
326 if (ret) {
327 kfree(name);
328 goto out;
329 }
330
331 array->size++;
332 cache_entry = &array->array[array->size - 1];
333 cache_entry->cookie = array->last_cookie;
334 cache_entry->ino = entry->ino;
335 cache_entry->d_type = entry->d_type;
336 cache_entry->name_len = entry->len;
337 cache_entry->name = name;
338 array->last_cookie = entry->cookie;
339 if (array->last_cookie <= cache_entry->cookie)
340 array->cookies_are_ordered = 0;
341 if (entry->eof != 0)
342 nfs_readdir_array_set_eof(array);
343out:
344 *cookie = array->last_cookie;
345 kunmap_local(array);
346 return ret;
347}
348
349#define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14)
350/*
351 * Hash algorithm allowing content addressible access to sequences
352 * of directory cookies. Content is addressed by the value of the
353 * cookie index of the first readdir entry in a page.
354 *
355 * We select only the first 18 bits to avoid issues with excessive
356 * memory use for the page cache XArray. 18 bits should allow the caching
357 * of 262144 pages of sequences of readdir entries. Since each page holds
358 * 127 readdir entries for a typical 64-bit system, that works out to a
359 * cache of ~ 33 million entries per directory.
360 */
361static pgoff_t nfs_readdir_folio_cookie_hash(u64 cookie)
362{
363 if (cookie == 0)
364 return 0;
365 return hash_64(cookie, 18);
366}
367
368static bool nfs_readdir_folio_validate(struct folio *folio, u64 last_cookie,
369 u64 change_attr)
370{
371 struct nfs_cache_array *array = kmap_local_folio(folio, 0);
372 int ret = true;
373
374 if (array->change_attr != change_attr)
375 ret = false;
376 if (nfs_readdir_array_index_cookie(array) != last_cookie)
377 ret = false;
378 kunmap_local(array);
379 return ret;
380}
381
382static void nfs_readdir_folio_unlock_and_put(struct folio *folio)
383{
384 folio_unlock(folio);
385 folio_put(folio);
386}
387
388static void nfs_readdir_folio_init_and_validate(struct folio *folio, u64 cookie,
389 u64 change_attr)
390{
391 if (folio_test_uptodate(folio)) {
392 if (nfs_readdir_folio_validate(folio, cookie, change_attr))
393 return;
394 nfs_readdir_clear_array(folio);
395 }
396 nfs_readdir_folio_init_array(folio, cookie, change_attr);
397 folio_mark_uptodate(folio);
398}
399
400static struct folio *nfs_readdir_folio_get_locked(struct address_space *mapping,
401 u64 cookie, u64 change_attr)
402{
403 pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
404 struct folio *folio;
405
406 folio = filemap_grab_folio(mapping, index);
407 if (IS_ERR(folio))
408 return NULL;
409 nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
410 return folio;
411}
412
413static u64 nfs_readdir_folio_last_cookie(struct folio *folio)
414{
415 struct nfs_cache_array *array;
416 u64 ret;
417
418 array = kmap_local_folio(folio, 0);
419 ret = array->last_cookie;
420 kunmap_local(array);
421 return ret;
422}
423
424static bool nfs_readdir_folio_needs_filling(struct folio *folio)
425{
426 struct nfs_cache_array *array;
427 bool ret;
428
429 array = kmap_local_folio(folio, 0);
430 ret = !nfs_readdir_array_is_full(array);
431 kunmap_local(array);
432 return ret;
433}
434
435static void nfs_readdir_folio_set_eof(struct folio *folio)
436{
437 struct nfs_cache_array *array;
438
439 array = kmap_local_folio(folio, 0);
440 nfs_readdir_array_set_eof(array);
441 kunmap_local(array);
442}
443
444static struct folio *nfs_readdir_folio_get_next(struct address_space *mapping,
445 u64 cookie, u64 change_attr)
446{
447 pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
448 struct folio *folio;
449
450 folio = __filemap_get_folio(mapping, index,
451 FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
452 mapping_gfp_mask(mapping));
453 if (IS_ERR(folio))
454 return NULL;
455 nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
456 if (nfs_readdir_folio_last_cookie(folio) != cookie)
457 nfs_readdir_folio_reinit_array(folio, cookie, change_attr);
458 return folio;
459}
460
461static inline
462int is_32bit_api(void)
463{
464#ifdef CONFIG_COMPAT
465 return in_compat_syscall();
466#else
467 return (BITS_PER_LONG == 32);
468#endif
469}
470
471static
472bool nfs_readdir_use_cookie(const struct file *filp)
473{
474 if ((filp->f_mode & FMODE_32BITHASH) ||
475 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
476 return false;
477 return true;
478}
479
480static void nfs_readdir_seek_next_array(struct nfs_cache_array *array,
481 struct nfs_readdir_descriptor *desc)
482{
483 if (array->folio_full) {
484 desc->last_cookie = array->last_cookie;
485 desc->current_index += array->size;
486 desc->cache_entry_index = 0;
487 desc->folio_index++;
488 } else
489 desc->last_cookie = nfs_readdir_array_index_cookie(array);
490}
491
492static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc)
493{
494 desc->current_index = 0;
495 desc->last_cookie = 0;
496 desc->folio_index = 0;
497}
498
499static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
500 struct nfs_readdir_descriptor *desc)
501{
502 loff_t diff = desc->ctx->pos - desc->current_index;
503 unsigned int index;
504
505 if (diff < 0)
506 goto out_eof;
507 if (diff >= array->size) {
508 if (array->folio_is_eof)
509 goto out_eof;
510 nfs_readdir_seek_next_array(array, desc);
511 return -EAGAIN;
512 }
513
514 index = (unsigned int)diff;
515 desc->dir_cookie = array->array[index].cookie;
516 desc->cache_entry_index = index;
517 return 0;
518out_eof:
519 desc->eof = true;
520 return -EBADCOOKIE;
521}
522
523static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
524 u64 cookie)
525{
526 if (!array->cookies_are_ordered)
527 return true;
528 /* Optimisation for monotonically increasing cookies */
529 if (cookie >= array->last_cookie)
530 return false;
531 if (array->size && cookie < array->array[0].cookie)
532 return false;
533 return true;
534}
535
536static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
537 struct nfs_readdir_descriptor *desc)
538{
539 unsigned int i;
540 int status = -EAGAIN;
541
542 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
543 goto check_eof;
544
545 for (i = 0; i < array->size; i++) {
546 if (array->array[i].cookie == desc->dir_cookie) {
547 if (nfs_readdir_use_cookie(desc->file))
548 desc->ctx->pos = desc->dir_cookie;
549 else
550 desc->ctx->pos = desc->current_index + i;
551 desc->cache_entry_index = i;
552 return 0;
553 }
554 }
555check_eof:
556 if (array->folio_is_eof) {
557 status = -EBADCOOKIE;
558 if (desc->dir_cookie == array->last_cookie)
559 desc->eof = true;
560 } else
561 nfs_readdir_seek_next_array(array, desc);
562 return status;
563}
564
565static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
566{
567 struct nfs_cache_array *array;
568 int status;
569
570 array = kmap_local_folio(desc->folio, 0);
571
572 if (desc->dir_cookie == 0)
573 status = nfs_readdir_search_for_pos(array, desc);
574 else
575 status = nfs_readdir_search_for_cookie(array, desc);
576
577 kunmap_local(array);
578 return status;
579}
580
581/* Fill a page with xdr information before transferring to the cache page */
582static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
583 __be32 *verf, u64 cookie,
584 struct page **pages, size_t bufsize,
585 __be32 *verf_res)
586{
587 struct inode *inode = file_inode(desc->file);
588 struct nfs_readdir_arg arg = {
589 .dentry = file_dentry(desc->file),
590 .cred = desc->file->f_cred,
591 .verf = verf,
592 .cookie = cookie,
593 .pages = pages,
594 .page_len = bufsize,
595 .plus = desc->plus,
596 };
597 struct nfs_readdir_res res = {
598 .verf = verf_res,
599 };
600 unsigned long timestamp, gencount;
601 int error;
602
603 again:
604 timestamp = jiffies;
605 gencount = nfs_inc_attr_generation_counter();
606 desc->dir_verifier = nfs_save_change_attribute(inode);
607 error = NFS_PROTO(inode)->readdir(&arg, &res);
608 if (error < 0) {
609 /* We requested READDIRPLUS, but the server doesn't grok it */
610 if (error == -ENOTSUPP && desc->plus) {
611 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
612 desc->plus = arg.plus = false;
613 goto again;
614 }
615 goto error;
616 }
617 desc->timestamp = timestamp;
618 desc->gencount = gencount;
619error:
620 return error;
621}
622
623static int xdr_decode(struct nfs_readdir_descriptor *desc,
624 struct nfs_entry *entry, struct xdr_stream *xdr)
625{
626 struct inode *inode = file_inode(desc->file);
627 int error;
628
629 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
630 if (error)
631 return error;
632 entry->fattr->time_start = desc->timestamp;
633 entry->fattr->gencount = desc->gencount;
634 return 0;
635}
636
637/* Match file and dirent using either filehandle or fileid
638 * Note: caller is responsible for checking the fsid
639 */
640static
641int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
642{
643 struct inode *inode;
644 struct nfs_inode *nfsi;
645
646 if (d_really_is_negative(dentry))
647 return 0;
648
649 inode = d_inode(dentry);
650 if (is_bad_inode(inode) || NFS_STALE(inode))
651 return 0;
652
653 nfsi = NFS_I(inode);
654 if (entry->fattr->fileid != nfsi->fileid)
655 return 0;
656 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
657 return 0;
658 return 1;
659}
660
661#define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL)
662
663static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx,
664 unsigned int cache_hits,
665 unsigned int cache_misses)
666{
667 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
668 return false;
669 if (ctx->pos == 0 ||
670 cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD)
671 return true;
672 return false;
673}
674
675/*
676 * This function is called by the getattr code to request the
677 * use of readdirplus to accelerate any future lookups in the same
678 * directory.
679 */
680void nfs_readdir_record_entry_cache_hit(struct inode *dir)
681{
682 struct nfs_inode *nfsi = NFS_I(dir);
683 struct nfs_open_dir_context *ctx;
684
685 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
686 S_ISDIR(dir->i_mode)) {
687 rcu_read_lock();
688 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
689 atomic_inc(&ctx->cache_hits);
690 rcu_read_unlock();
691 }
692}
693
694/*
695 * This function is mainly for use by nfs_getattr().
696 *
697 * If this is an 'ls -l', we want to force use of readdirplus.
698 */
699void nfs_readdir_record_entry_cache_miss(struct inode *dir)
700{
701 struct nfs_inode *nfsi = NFS_I(dir);
702 struct nfs_open_dir_context *ctx;
703
704 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
705 S_ISDIR(dir->i_mode)) {
706 rcu_read_lock();
707 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
708 atomic_inc(&ctx->cache_misses);
709 rcu_read_unlock();
710 }
711}
712
713static void nfs_lookup_advise_force_readdirplus(struct inode *dir,
714 unsigned int flags)
715{
716 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
717 return;
718 if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL))
719 return;
720 nfs_readdir_record_entry_cache_miss(dir);
721}
722
723static
724void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
725 unsigned long dir_verifier)
726{
727 struct qstr filename = QSTR_INIT(entry->name, entry->len);
728 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
729 struct dentry *dentry;
730 struct dentry *alias;
731 struct inode *inode;
732 int status;
733
734 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
735 return;
736 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
737 return;
738 if (filename.len == 0)
739 return;
740 /* Validate that the name doesn't contain any illegal '\0' */
741 if (strnlen(filename.name, filename.len) != filename.len)
742 return;
743 /* ...or '/' */
744 if (strnchr(filename.name, filename.len, '/'))
745 return;
746 if (filename.name[0] == '.') {
747 if (filename.len == 1)
748 return;
749 if (filename.len == 2 && filename.name[1] == '.')
750 return;
751 }
752 filename.hash = full_name_hash(parent, filename.name, filename.len);
753
754 dentry = d_lookup(parent, &filename);
755again:
756 if (!dentry) {
757 dentry = d_alloc_parallel(parent, &filename, &wq);
758 if (IS_ERR(dentry))
759 return;
760 }
761 if (!d_in_lookup(dentry)) {
762 /* Is there a mountpoint here? If so, just exit */
763 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
764 &entry->fattr->fsid))
765 goto out;
766 if (nfs_same_file(dentry, entry)) {
767 if (!entry->fh->size)
768 goto out;
769 nfs_set_verifier(dentry, dir_verifier);
770 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
771 if (!status)
772 nfs_setsecurity(d_inode(dentry), entry->fattr);
773 trace_nfs_readdir_lookup_revalidate(d_inode(parent),
774 dentry, 0, status);
775 goto out;
776 } else {
777 trace_nfs_readdir_lookup_revalidate_failed(
778 d_inode(parent), dentry, 0);
779 d_invalidate(dentry);
780 dput(dentry);
781 dentry = NULL;
782 goto again;
783 }
784 }
785 if (!entry->fh->size) {
786 d_lookup_done(dentry);
787 goto out;
788 }
789
790 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
791 alias = d_splice_alias(inode, dentry);
792 d_lookup_done(dentry);
793 if (alias) {
794 if (IS_ERR(alias))
795 goto out;
796 dput(dentry);
797 dentry = alias;
798 }
799 nfs_set_verifier(dentry, dir_verifier);
800 trace_nfs_readdir_lookup(d_inode(parent), dentry, 0);
801out:
802 dput(dentry);
803}
804
805static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc,
806 struct nfs_entry *entry,
807 struct xdr_stream *stream)
808{
809 int ret;
810
811 if (entry->fattr->label)
812 entry->fattr->label->len = NFS4_MAXLABELLEN;
813 ret = xdr_decode(desc, entry, stream);
814 if (ret || !desc->plus)
815 return ret;
816 nfs_prime_dcache(file_dentry(desc->file), entry, desc->dir_verifier);
817 return 0;
818}
819
820/* Perform conversion from xdr to cache array */
821static int nfs_readdir_folio_filler(struct nfs_readdir_descriptor *desc,
822 struct nfs_entry *entry,
823 struct page **xdr_pages, unsigned int buflen,
824 struct folio **arrays, size_t narrays,
825 u64 change_attr)
826{
827 struct address_space *mapping = desc->file->f_mapping;
828 struct folio *new, *folio = *arrays;
829 struct xdr_stream stream;
830 struct page *scratch;
831 struct xdr_buf buf;
832 u64 cookie;
833 int status;
834
835 scratch = alloc_page(GFP_KERNEL);
836 if (scratch == NULL)
837 return -ENOMEM;
838
839 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
840 xdr_set_scratch_page(&stream, scratch);
841
842 do {
843 status = nfs_readdir_entry_decode(desc, entry, &stream);
844 if (status != 0)
845 break;
846
847 status = nfs_readdir_folio_array_append(folio, entry, &cookie);
848 if (status != -ENOSPC)
849 continue;
850
851 if (folio->mapping != mapping) {
852 if (!--narrays)
853 break;
854 new = nfs_readdir_folio_array_alloc(cookie, GFP_KERNEL);
855 if (!new)
856 break;
857 arrays++;
858 *arrays = folio = new;
859 } else {
860 new = nfs_readdir_folio_get_next(mapping, cookie,
861 change_attr);
862 if (!new)
863 break;
864 if (folio != *arrays)
865 nfs_readdir_folio_unlock_and_put(folio);
866 folio = new;
867 }
868 desc->folio_index_max++;
869 status = nfs_readdir_folio_array_append(folio, entry, &cookie);
870 } while (!status && !entry->eof);
871
872 switch (status) {
873 case -EBADCOOKIE:
874 if (!entry->eof)
875 break;
876 nfs_readdir_folio_set_eof(folio);
877 fallthrough;
878 case -EAGAIN:
879 status = 0;
880 break;
881 case -ENOSPC:
882 status = 0;
883 if (!desc->plus)
884 break;
885 while (!nfs_readdir_entry_decode(desc, entry, &stream))
886 ;
887 }
888
889 if (folio != *arrays)
890 nfs_readdir_folio_unlock_and_put(folio);
891
892 put_page(scratch);
893 return status;
894}
895
896static void nfs_readdir_free_pages(struct page **pages, size_t npages)
897{
898 while (npages--)
899 put_page(pages[npages]);
900 kfree(pages);
901}
902
903/*
904 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
905 * to nfs_readdir_free_pages()
906 */
907static struct page **nfs_readdir_alloc_pages(size_t npages)
908{
909 struct page **pages;
910 size_t i;
911
912 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
913 if (!pages)
914 return NULL;
915 for (i = 0; i < npages; i++) {
916 struct page *page = alloc_page(GFP_KERNEL);
917 if (page == NULL)
918 goto out_freepages;
919 pages[i] = page;
920 }
921 return pages;
922
923out_freepages:
924 nfs_readdir_free_pages(pages, i);
925 return NULL;
926}
927
928static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
929 __be32 *verf_arg, __be32 *verf_res,
930 struct folio **arrays, size_t narrays)
931{
932 u64 change_attr;
933 struct page **pages;
934 struct folio *folio = *arrays;
935 struct nfs_entry *entry;
936 size_t array_size;
937 struct inode *inode = file_inode(desc->file);
938 unsigned int dtsize = desc->dtsize;
939 unsigned int pglen;
940 int status = -ENOMEM;
941
942 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
943 if (!entry)
944 return -ENOMEM;
945 entry->cookie = nfs_readdir_folio_last_cookie(folio);
946 entry->fh = nfs_alloc_fhandle();
947 entry->fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
948 entry->server = NFS_SERVER(inode);
949 if (entry->fh == NULL || entry->fattr == NULL)
950 goto out;
951
952 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
953 pages = nfs_readdir_alloc_pages(array_size);
954 if (!pages)
955 goto out;
956
957 change_attr = inode_peek_iversion_raw(inode);
958 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, pages,
959 dtsize, verf_res);
960 if (status < 0)
961 goto free_pages;
962
963 pglen = status;
964 if (pglen != 0)
965 status = nfs_readdir_folio_filler(desc, entry, pages, pglen,
966 arrays, narrays, change_attr);
967 else
968 nfs_readdir_folio_set_eof(folio);
969 desc->buffer_fills++;
970
971free_pages:
972 nfs_readdir_free_pages(pages, array_size);
973out:
974 nfs_free_fattr(entry->fattr);
975 nfs_free_fhandle(entry->fh);
976 kfree(entry);
977 return status;
978}
979
980static void nfs_readdir_folio_put(struct nfs_readdir_descriptor *desc)
981{
982 folio_put(desc->folio);
983 desc->folio = NULL;
984}
985
986static void
987nfs_readdir_folio_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
988{
989 folio_unlock(desc->folio);
990 nfs_readdir_folio_put(desc);
991}
992
993static struct folio *
994nfs_readdir_folio_get_cached(struct nfs_readdir_descriptor *desc)
995{
996 struct address_space *mapping = desc->file->f_mapping;
997 u64 change_attr = inode_peek_iversion_raw(mapping->host);
998 u64 cookie = desc->last_cookie;
999 struct folio *folio;
1000
1001 folio = nfs_readdir_folio_get_locked(mapping, cookie, change_attr);
1002 if (!folio)
1003 return NULL;
1004 if (desc->clear_cache && !nfs_readdir_folio_needs_filling(folio))
1005 nfs_readdir_folio_reinit_array(folio, cookie, change_attr);
1006 return folio;
1007}
1008
1009/*
1010 * Returns 0 if desc->dir_cookie was found on page desc->page_index
1011 * and locks the page to prevent removal from the page cache.
1012 */
1013static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
1014{
1015 struct inode *inode = file_inode(desc->file);
1016 struct nfs_inode *nfsi = NFS_I(inode);
1017 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1018 int res;
1019
1020 desc->folio = nfs_readdir_folio_get_cached(desc);
1021 if (!desc->folio)
1022 return -ENOMEM;
1023 if (nfs_readdir_folio_needs_filling(desc->folio)) {
1024 /* Grow the dtsize if we had to go back for more pages */
1025 if (desc->folio_index == desc->folio_index_max)
1026 nfs_grow_dtsize(desc);
1027 desc->folio_index_max = desc->folio_index;
1028 trace_nfs_readdir_cache_fill(desc->file, nfsi->cookieverf,
1029 desc->last_cookie,
1030 desc->folio->index, desc->dtsize);
1031 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
1032 &desc->folio, 1);
1033 if (res < 0) {
1034 nfs_readdir_folio_unlock_and_put_cached(desc);
1035 trace_nfs_readdir_cache_fill_done(inode, res);
1036 if (res == -EBADCOOKIE || res == -ENOTSYNC) {
1037 invalidate_inode_pages2(desc->file->f_mapping);
1038 nfs_readdir_rewind_search(desc);
1039 trace_nfs_readdir_invalidate_cache_range(
1040 inode, 0, MAX_LFS_FILESIZE);
1041 return -EAGAIN;
1042 }
1043 return res;
1044 }
1045 /*
1046 * Set the cookie verifier if the page cache was empty
1047 */
1048 if (desc->last_cookie == 0 &&
1049 memcmp(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf))) {
1050 memcpy(nfsi->cookieverf, verf,
1051 sizeof(nfsi->cookieverf));
1052 invalidate_inode_pages2_range(desc->file->f_mapping, 1,
1053 -1);
1054 trace_nfs_readdir_invalidate_cache_range(
1055 inode, 1, MAX_LFS_FILESIZE);
1056 }
1057 desc->clear_cache = false;
1058 }
1059 res = nfs_readdir_search_array(desc);
1060 if (res == 0)
1061 return 0;
1062 nfs_readdir_folio_unlock_and_put_cached(desc);
1063 return res;
1064}
1065
1066/* Search for desc->dir_cookie from the beginning of the page cache */
1067static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
1068{
1069 int res;
1070
1071 do {
1072 res = find_and_lock_cache_page(desc);
1073 } while (res == -EAGAIN);
1074 return res;
1075}
1076
1077#define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1078
1079/*
1080 * Once we've found the start of the dirent within a page: fill 'er up...
1081 */
1082static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
1083 const __be32 *verf)
1084{
1085 struct file *file = desc->file;
1086 struct nfs_cache_array *array;
1087 unsigned int i;
1088 bool first_emit = !desc->dir_cookie;
1089
1090 array = kmap_local_folio(desc->folio, 0);
1091 for (i = desc->cache_entry_index; i < array->size; i++) {
1092 struct nfs_cache_array_entry *ent;
1093
1094 /*
1095 * nfs_readdir_handle_cache_misses return force clear at
1096 * (cache_misses > NFS_READDIR_CACHE_MISS_THRESHOLD) for
1097 * readdir heuristic, NFS_READDIR_CACHE_MISS_THRESHOLD + 1
1098 * entries need be emitted here.
1099 */
1100 if (first_emit && i > NFS_READDIR_CACHE_MISS_THRESHOLD + 2) {
1101 desc->eob = true;
1102 break;
1103 }
1104
1105 ent = &array->array[i];
1106 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
1107 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
1108 desc->eob = true;
1109 break;
1110 }
1111 memcpy(desc->verf, verf, sizeof(desc->verf));
1112 if (i == array->size - 1) {
1113 desc->dir_cookie = array->last_cookie;
1114 nfs_readdir_seek_next_array(array, desc);
1115 } else {
1116 desc->dir_cookie = array->array[i + 1].cookie;
1117 desc->last_cookie = array->array[0].cookie;
1118 }
1119 if (nfs_readdir_use_cookie(file))
1120 desc->ctx->pos = desc->dir_cookie;
1121 else
1122 desc->ctx->pos++;
1123 }
1124 if (array->folio_is_eof)
1125 desc->eof = !desc->eob;
1126
1127 kunmap_local(array);
1128 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1129 (unsigned long long)desc->dir_cookie);
1130}
1131
1132/*
1133 * If we cannot find a cookie in our cache, we suspect that this is
1134 * because it points to a deleted file, so we ask the server to return
1135 * whatever it thinks is the next entry. We then feed this to filldir.
1136 * If all goes well, we should then be able to find our way round the
1137 * cache on the next call to readdir_search_pagecache();
1138 *
1139 * NOTE: we cannot add the anonymous page to the pagecache because
1140 * the data it contains might not be page aligned. Besides,
1141 * we should already have a complete representation of the
1142 * directory in the page cache by the time we get here.
1143 */
1144static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1145{
1146 struct folio **arrays;
1147 size_t i, sz = 512;
1148 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1149 int status = -ENOMEM;
1150
1151 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1152 (unsigned long long)desc->dir_cookie);
1153
1154 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1155 if (!arrays)
1156 goto out;
1157 arrays[0] = nfs_readdir_folio_array_alloc(desc->dir_cookie, GFP_KERNEL);
1158 if (!arrays[0])
1159 goto out;
1160
1161 desc->folio_index = 0;
1162 desc->cache_entry_index = 0;
1163 desc->last_cookie = desc->dir_cookie;
1164 desc->folio_index_max = 0;
1165
1166 trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie,
1167 -1, desc->dtsize);
1168
1169 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1170 if (status < 0) {
1171 trace_nfs_readdir_uncached_done(file_inode(desc->file), status);
1172 goto out_free;
1173 }
1174
1175 for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1176 desc->folio = arrays[i];
1177 nfs_do_filldir(desc, verf);
1178 }
1179 desc->folio = NULL;
1180
1181 /*
1182 * Grow the dtsize if we have to go back for more pages,
1183 * or shrink it if we're reading too many.
1184 */
1185 if (!desc->eof) {
1186 if (!desc->eob)
1187 nfs_grow_dtsize(desc);
1188 else if (desc->buffer_fills == 1 &&
1189 i < (desc->folio_index_max >> 1))
1190 nfs_shrink_dtsize(desc);
1191 }
1192out_free:
1193 for (i = 0; i < sz && arrays[i]; i++)
1194 nfs_readdir_folio_array_free(arrays[i]);
1195out:
1196 if (!nfs_readdir_use_cookie(desc->file))
1197 nfs_readdir_rewind_search(desc);
1198 desc->folio_index_max = -1;
1199 kfree(arrays);
1200 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1201 return status;
1202}
1203
1204static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1205 struct nfs_readdir_descriptor *desc,
1206 unsigned int cache_misses,
1207 bool force_clear)
1208{
1209 if (desc->ctx->pos == 0 || !desc->plus)
1210 return false;
1211 if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1212 return false;
1213 trace_nfs_readdir_force_readdirplus(inode);
1214 return true;
1215}
1216
1217/* The file offset position represents the dirent entry number. A
1218 last cookie cache takes care of the common case of reading the
1219 whole directory.
1220 */
1221static int nfs_readdir(struct file *file, struct dir_context *ctx)
1222{
1223 struct dentry *dentry = file_dentry(file);
1224 struct inode *inode = d_inode(dentry);
1225 struct nfs_inode *nfsi = NFS_I(inode);
1226 struct nfs_open_dir_context *dir_ctx = file->private_data;
1227 struct nfs_readdir_descriptor *desc;
1228 unsigned int cache_hits, cache_misses;
1229 bool force_clear;
1230 int res;
1231
1232 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1233 file, (long long)ctx->pos);
1234 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1235
1236 /*
1237 * ctx->pos points to the dirent entry number.
1238 * *desc->dir_cookie has the cookie for the next entry. We have
1239 * to either find the entry with the appropriate number or
1240 * revalidate the cookie.
1241 */
1242 nfs_revalidate_mapping(inode, file->f_mapping);
1243
1244 res = -ENOMEM;
1245 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1246 if (!desc)
1247 goto out;
1248 desc->file = file;
1249 desc->ctx = ctx;
1250 desc->folio_index_max = -1;
1251
1252 spin_lock(&file->f_lock);
1253 desc->dir_cookie = dir_ctx->dir_cookie;
1254 desc->folio_index = dir_ctx->page_index;
1255 desc->last_cookie = dir_ctx->last_cookie;
1256 desc->attr_gencount = dir_ctx->attr_gencount;
1257 desc->eof = dir_ctx->eof;
1258 nfs_set_dtsize(desc, dir_ctx->dtsize);
1259 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1260 cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0);
1261 cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0);
1262 force_clear = dir_ctx->force_clear;
1263 spin_unlock(&file->f_lock);
1264
1265 if (desc->eof) {
1266 res = 0;
1267 goto out_free;
1268 }
1269
1270 desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses);
1271 force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1272 force_clear);
1273 desc->clear_cache = force_clear;
1274
1275 do {
1276 res = readdir_search_pagecache(desc);
1277
1278 if (res == -EBADCOOKIE) {
1279 res = 0;
1280 /* This means either end of directory */
1281 if (desc->dir_cookie && !desc->eof) {
1282 /* Or that the server has 'lost' a cookie */
1283 res = uncached_readdir(desc);
1284 if (res == 0)
1285 continue;
1286 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1287 res = 0;
1288 }
1289 break;
1290 }
1291 if (res == -ETOOSMALL && desc->plus) {
1292 nfs_zap_caches(inode);
1293 desc->plus = false;
1294 desc->eof = false;
1295 continue;
1296 }
1297 if (res < 0)
1298 break;
1299
1300 nfs_do_filldir(desc, nfsi->cookieverf);
1301 nfs_readdir_folio_unlock_and_put_cached(desc);
1302 if (desc->folio_index == desc->folio_index_max)
1303 desc->clear_cache = force_clear;
1304 } while (!desc->eob && !desc->eof);
1305
1306 spin_lock(&file->f_lock);
1307 dir_ctx->dir_cookie = desc->dir_cookie;
1308 dir_ctx->last_cookie = desc->last_cookie;
1309 dir_ctx->attr_gencount = desc->attr_gencount;
1310 dir_ctx->page_index = desc->folio_index;
1311 dir_ctx->force_clear = force_clear;
1312 dir_ctx->eof = desc->eof;
1313 dir_ctx->dtsize = desc->dtsize;
1314 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1315 spin_unlock(&file->f_lock);
1316out_free:
1317 kfree(desc);
1318
1319out:
1320 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1321 return res;
1322}
1323
1324static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1325{
1326 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1327
1328 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1329 filp, offset, whence);
1330
1331 switch (whence) {
1332 default:
1333 return -EINVAL;
1334 case SEEK_SET:
1335 if (offset < 0)
1336 return -EINVAL;
1337 spin_lock(&filp->f_lock);
1338 break;
1339 case SEEK_CUR:
1340 if (offset == 0)
1341 return filp->f_pos;
1342 spin_lock(&filp->f_lock);
1343 offset += filp->f_pos;
1344 if (offset < 0) {
1345 spin_unlock(&filp->f_lock);
1346 return -EINVAL;
1347 }
1348 }
1349 if (offset != filp->f_pos) {
1350 filp->f_pos = offset;
1351 dir_ctx->page_index = 0;
1352 if (!nfs_readdir_use_cookie(filp)) {
1353 dir_ctx->dir_cookie = 0;
1354 dir_ctx->last_cookie = 0;
1355 } else {
1356 dir_ctx->dir_cookie = offset;
1357 dir_ctx->last_cookie = offset;
1358 }
1359 dir_ctx->eof = false;
1360 }
1361 spin_unlock(&filp->f_lock);
1362 return offset;
1363}
1364
1365/*
1366 * All directory operations under NFS are synchronous, so fsync()
1367 * is a dummy operation.
1368 */
1369static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1370 int datasync)
1371{
1372 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1373
1374 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1375 return 0;
1376}
1377
1378/**
1379 * nfs_force_lookup_revalidate - Mark the directory as having changed
1380 * @dir: pointer to directory inode
1381 *
1382 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1383 * full lookup on all child dentries of 'dir' whenever a change occurs
1384 * on the server that might have invalidated our dcache.
1385 *
1386 * Note that we reserve bit '0' as a tag to let us know when a dentry
1387 * was revalidated while holding a delegation on its inode.
1388 *
1389 * The caller should be holding dir->i_lock
1390 */
1391void nfs_force_lookup_revalidate(struct inode *dir)
1392{
1393 NFS_I(dir)->cache_change_attribute += 2;
1394}
1395EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1396
1397/**
1398 * nfs_verify_change_attribute - Detects NFS remote directory changes
1399 * @dir: pointer to parent directory inode
1400 * @verf: previously saved change attribute
1401 *
1402 * Return "false" if the verifiers doesn't match the change attribute.
1403 * This would usually indicate that the directory contents have changed on
1404 * the server, and that any dentries need revalidating.
1405 */
1406static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1407{
1408 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1409}
1410
1411static void nfs_set_verifier_delegated(unsigned long *verf)
1412{
1413 *verf |= 1UL;
1414}
1415
1416#if IS_ENABLED(CONFIG_NFS_V4)
1417static void nfs_unset_verifier_delegated(unsigned long *verf)
1418{
1419 *verf &= ~1UL;
1420}
1421#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1422
1423static bool nfs_test_verifier_delegated(unsigned long verf)
1424{
1425 return verf & 1;
1426}
1427
1428static bool nfs_verifier_is_delegated(struct dentry *dentry)
1429{
1430 return nfs_test_verifier_delegated(dentry->d_time);
1431}
1432
1433static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1434{
1435 struct inode *inode = d_inode(dentry);
1436 struct inode *dir = d_inode_rcu(dentry->d_parent);
1437
1438 if (!dir || !nfs_verify_change_attribute(dir, verf))
1439 return;
1440 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ, 0))
1441 nfs_set_verifier_delegated(&verf);
1442 dentry->d_time = verf;
1443}
1444
1445/**
1446 * nfs_set_verifier - save a parent directory verifier in the dentry
1447 * @dentry: pointer to dentry
1448 * @verf: verifier to save
1449 *
1450 * Saves the parent directory verifier in @dentry. If the inode has
1451 * a delegation, we also tag the dentry as having been revalidated
1452 * while holding a delegation so that we know we don't have to
1453 * look it up again after a directory change.
1454 */
1455void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1456{
1457
1458 spin_lock(&dentry->d_lock);
1459 nfs_set_verifier_locked(dentry, verf);
1460 spin_unlock(&dentry->d_lock);
1461}
1462EXPORT_SYMBOL_GPL(nfs_set_verifier);
1463
1464#if IS_ENABLED(CONFIG_NFS_V4)
1465/**
1466 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1467 * @inode: pointer to inode
1468 *
1469 * Iterates through the dentries in the inode alias list and clears
1470 * the tag used to indicate that the dentry has been revalidated
1471 * while holding a delegation.
1472 * This function is intended for use when the delegation is being
1473 * returned or revoked.
1474 */
1475void nfs_clear_verifier_delegated(struct inode *inode)
1476{
1477 struct dentry *alias;
1478
1479 if (!inode)
1480 return;
1481 spin_lock(&inode->i_lock);
1482 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1483 spin_lock(&alias->d_lock);
1484 nfs_unset_verifier_delegated(&alias->d_time);
1485 spin_unlock(&alias->d_lock);
1486 }
1487 spin_unlock(&inode->i_lock);
1488}
1489EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1490#endif /* IS_ENABLED(CONFIG_NFS_V4) */
1491
1492static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1493{
1494 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) &&
1495 d_really_is_negative(dentry))
1496 return dentry->d_time == inode_peek_iversion_raw(dir);
1497 return nfs_verify_change_attribute(dir, dentry->d_time);
1498}
1499
1500/*
1501 * A check for whether or not the parent directory has changed.
1502 * In the case it has, we assume that the dentries are untrustworthy
1503 * and may need to be looked up again.
1504 * If rcu_walk prevents us from performing a full check, return 0.
1505 */
1506static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1507 int rcu_walk)
1508{
1509 if (IS_ROOT(dentry))
1510 return 1;
1511 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1512 return 0;
1513 if (!nfs_dentry_verify_change(dir, dentry))
1514 return 0;
1515 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1516 if (nfs_mapping_need_revalidate_inode(dir)) {
1517 if (rcu_walk)
1518 return 0;
1519 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1520 return 0;
1521 }
1522 if (!nfs_dentry_verify_change(dir, dentry))
1523 return 0;
1524 return 1;
1525}
1526
1527/*
1528 * Use intent information to check whether or not we're going to do
1529 * an O_EXCL create using this path component.
1530 */
1531static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1532{
1533 if (NFS_PROTO(dir)->version == 2)
1534 return 0;
1535 return flags & LOOKUP_EXCL;
1536}
1537
1538/*
1539 * Inode and filehandle revalidation for lookups.
1540 *
1541 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1542 * or if the intent information indicates that we're about to open this
1543 * particular file and the "nocto" mount flag is not set.
1544 *
1545 */
1546static
1547int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1548{
1549 struct nfs_server *server = NFS_SERVER(inode);
1550 int ret;
1551
1552 if (IS_AUTOMOUNT(inode))
1553 return 0;
1554
1555 if (flags & LOOKUP_OPEN) {
1556 switch (inode->i_mode & S_IFMT) {
1557 case S_IFREG:
1558 /* A NFSv4 OPEN will revalidate later */
1559 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1560 goto out;
1561 fallthrough;
1562 case S_IFDIR:
1563 if (server->flags & NFS_MOUNT_NOCTO)
1564 break;
1565 /* NFS close-to-open cache consistency validation */
1566 goto out_force;
1567 }
1568 }
1569
1570 /* VFS wants an on-the-wire revalidation */
1571 if (flags & LOOKUP_REVAL)
1572 goto out_force;
1573out:
1574 if (inode->i_nlink > 0 ||
1575 (inode->i_nlink == 0 &&
1576 test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1577 return 0;
1578 else
1579 return -ESTALE;
1580out_force:
1581 if (flags & LOOKUP_RCU)
1582 return -ECHILD;
1583 ret = __nfs_revalidate_inode(server, inode);
1584 if (ret != 0)
1585 return ret;
1586 goto out;
1587}
1588
1589static void nfs_mark_dir_for_revalidate(struct inode *inode)
1590{
1591 spin_lock(&inode->i_lock);
1592 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1593 spin_unlock(&inode->i_lock);
1594}
1595
1596/*
1597 * We judge how long we want to trust negative
1598 * dentries by looking at the parent inode mtime.
1599 *
1600 * If parent mtime has changed, we revalidate, else we wait for a
1601 * period corresponding to the parent's attribute cache timeout value.
1602 *
1603 * If LOOKUP_RCU prevents us from performing a full check, return 1
1604 * suggesting a reval is needed.
1605 *
1606 * Note that when creating a new file, or looking up a rename target,
1607 * then it shouldn't be necessary to revalidate a negative dentry.
1608 */
1609static inline
1610int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1611 unsigned int flags)
1612{
1613 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1614 return 0;
1615 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1616 return 1;
1617 /* Case insensitive server? Revalidate negative dentries */
1618 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1619 return 1;
1620 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1621}
1622
1623static int
1624nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1625 struct inode *inode, int error)
1626{
1627 switch (error) {
1628 case 1:
1629 break;
1630 case -ETIMEDOUT:
1631 if (inode && (IS_ROOT(dentry) ||
1632 NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL))
1633 error = 1;
1634 break;
1635 case -ESTALE:
1636 case -ENOENT:
1637 error = 0;
1638 fallthrough;
1639 default:
1640 /*
1641 * We can't d_drop the root of a disconnected tree:
1642 * its d_hash is on the s_anon list and d_drop() would hide
1643 * it from shrink_dcache_for_unmount(), leading to busy
1644 * inodes on unmount and further oopses.
1645 */
1646 if (inode && IS_ROOT(dentry))
1647 error = 1;
1648 break;
1649 }
1650 trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error);
1651 return error;
1652}
1653
1654static int
1655nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1656 unsigned int flags)
1657{
1658 int ret = 1;
1659 if (nfs_neg_need_reval(dir, dentry, flags)) {
1660 if (flags & LOOKUP_RCU)
1661 return -ECHILD;
1662 ret = 0;
1663 }
1664 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1665}
1666
1667static int
1668nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1669 struct inode *inode)
1670{
1671 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1672 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1673}
1674
1675static int nfs_lookup_revalidate_dentry(struct inode *dir,
1676 struct dentry *dentry,
1677 struct inode *inode, unsigned int flags)
1678{
1679 struct nfs_fh *fhandle;
1680 struct nfs_fattr *fattr;
1681 unsigned long dir_verifier;
1682 int ret;
1683
1684 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1685
1686 ret = -ENOMEM;
1687 fhandle = nfs_alloc_fhandle();
1688 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1689 if (fhandle == NULL || fattr == NULL)
1690 goto out;
1691
1692 dir_verifier = nfs_save_change_attribute(dir);
1693 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1694 if (ret < 0)
1695 goto out;
1696
1697 /* Request help from readdirplus */
1698 nfs_lookup_advise_force_readdirplus(dir, flags);
1699
1700 ret = 0;
1701 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1702 goto out;
1703 if (nfs_refresh_inode(inode, fattr) < 0)
1704 goto out;
1705
1706 nfs_setsecurity(inode, fattr);
1707 nfs_set_verifier(dentry, dir_verifier);
1708
1709 ret = 1;
1710out:
1711 nfs_free_fattr(fattr);
1712 nfs_free_fhandle(fhandle);
1713
1714 /*
1715 * If the lookup failed despite the dentry change attribute being
1716 * a match, then we should revalidate the directory cache.
1717 */
1718 if (!ret && nfs_dentry_verify_change(dir, dentry))
1719 nfs_mark_dir_for_revalidate(dir);
1720 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1721}
1722
1723/*
1724 * This is called every time the dcache has a lookup hit,
1725 * and we should check whether we can really trust that
1726 * lookup.
1727 *
1728 * NOTE! The hit can be a negative hit too, don't assume
1729 * we have an inode!
1730 *
1731 * If the parent directory is seen to have changed, we throw out the
1732 * cached dentry and do a new lookup.
1733 */
1734static int
1735nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1736 unsigned int flags)
1737{
1738 struct inode *inode;
1739 int error = 0;
1740
1741 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1742 inode = d_inode(dentry);
1743
1744 if (!inode)
1745 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1746
1747 if (is_bad_inode(inode)) {
1748 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1749 __func__, dentry);
1750 goto out_bad;
1751 }
1752
1753 if ((flags & LOOKUP_RENAME_TARGET) && d_count(dentry) < 2 &&
1754 nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1755 goto out_bad;
1756
1757 if (nfs_verifier_is_delegated(dentry))
1758 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1759
1760 /* Force a full look up iff the parent directory has changed */
1761 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1762 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1763 error = nfs_lookup_verify_inode(inode, flags);
1764 if (error) {
1765 if (error == -ESTALE)
1766 nfs_mark_dir_for_revalidate(dir);
1767 goto out_bad;
1768 }
1769 goto out_valid;
1770 }
1771
1772 if (flags & LOOKUP_RCU)
1773 return -ECHILD;
1774
1775 if (NFS_STALE(inode))
1776 goto out_bad;
1777
1778 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
1779out_valid:
1780 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1781out_bad:
1782 if (flags & LOOKUP_RCU)
1783 return -ECHILD;
1784 return nfs_lookup_revalidate_done(dir, dentry, inode, error);
1785}
1786
1787static int
1788__nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1789 int (*reval)(struct inode *, struct dentry *, unsigned int))
1790{
1791 struct dentry *parent;
1792 struct inode *dir;
1793 int ret;
1794
1795 if (flags & LOOKUP_RCU) {
1796 if (dentry->d_fsdata == NFS_FSDATA_BLOCKED)
1797 return -ECHILD;
1798 parent = READ_ONCE(dentry->d_parent);
1799 dir = d_inode_rcu(parent);
1800 if (!dir)
1801 return -ECHILD;
1802 ret = reval(dir, dentry, flags);
1803 if (parent != READ_ONCE(dentry->d_parent))
1804 return -ECHILD;
1805 } else {
1806 /* Wait for unlink to complete - see unblock_revalidate() */
1807 wait_var_event(&dentry->d_fsdata,
1808 smp_load_acquire(&dentry->d_fsdata)
1809 != NFS_FSDATA_BLOCKED);
1810 parent = dget_parent(dentry);
1811 ret = reval(d_inode(parent), dentry, flags);
1812 dput(parent);
1813 }
1814 return ret;
1815}
1816
1817static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1818{
1819 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1820}
1821
1822static void block_revalidate(struct dentry *dentry)
1823{
1824 /* old devname - just in case */
1825 kfree(dentry->d_fsdata);
1826
1827 /* Any new reference that could lead to an open
1828 * will take ->d_lock in lookup_open() -> d_lookup().
1829 * Holding this lock ensures we cannot race with
1830 * __nfs_lookup_revalidate() and removes and need
1831 * for further barriers.
1832 */
1833 lockdep_assert_held(&dentry->d_lock);
1834
1835 dentry->d_fsdata = NFS_FSDATA_BLOCKED;
1836}
1837
1838static void unblock_revalidate(struct dentry *dentry)
1839{
1840 /* store_release ensures wait_var_event() sees the update */
1841 smp_store_release(&dentry->d_fsdata, NULL);
1842 wake_up_var(&dentry->d_fsdata);
1843}
1844
1845/*
1846 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1847 * when we don't really care about the dentry name. This is called when a
1848 * pathwalk ends on a dentry that was not found via a normal lookup in the
1849 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1850 *
1851 * In this situation, we just want to verify that the inode itself is OK
1852 * since the dentry might have changed on the server.
1853 */
1854static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1855{
1856 struct inode *inode = d_inode(dentry);
1857 int error = 0;
1858
1859 /*
1860 * I believe we can only get a negative dentry here in the case of a
1861 * procfs-style symlink. Just assume it's correct for now, but we may
1862 * eventually need to do something more here.
1863 */
1864 if (!inode) {
1865 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1866 __func__, dentry);
1867 return 1;
1868 }
1869
1870 if (is_bad_inode(inode)) {
1871 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1872 __func__, dentry);
1873 return 0;
1874 }
1875
1876 error = nfs_lookup_verify_inode(inode, flags);
1877 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1878 __func__, inode->i_ino, error ? "invalid" : "valid");
1879 return !error;
1880}
1881
1882/*
1883 * This is called from dput() when d_count is going to 0.
1884 */
1885static int nfs_dentry_delete(const struct dentry *dentry)
1886{
1887 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1888 dentry, dentry->d_flags);
1889
1890 /* Unhash any dentry with a stale inode */
1891 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1892 return 1;
1893
1894 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1895 /* Unhash it, so that ->d_iput() would be called */
1896 return 1;
1897 }
1898 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1899 /* Unhash it, so that ancestors of killed async unlink
1900 * files will be cleaned up during umount */
1901 return 1;
1902 }
1903 return 0;
1904
1905}
1906
1907/* Ensure that we revalidate inode->i_nlink */
1908static void nfs_drop_nlink(struct inode *inode)
1909{
1910 spin_lock(&inode->i_lock);
1911 /* drop the inode if we're reasonably sure this is the last link */
1912 if (inode->i_nlink > 0)
1913 drop_nlink(inode);
1914 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1915 nfs_set_cache_invalid(
1916 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1917 NFS_INO_INVALID_NLINK);
1918 spin_unlock(&inode->i_lock);
1919}
1920
1921/*
1922 * Called when the dentry loses inode.
1923 * We use it to clean up silly-renamed files.
1924 */
1925static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1926{
1927 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1928 nfs_complete_unlink(dentry, inode);
1929 nfs_drop_nlink(inode);
1930 }
1931 iput(inode);
1932}
1933
1934static void nfs_d_release(struct dentry *dentry)
1935{
1936 /* free cached devname value, if it survived that far */
1937 if (unlikely(dentry->d_fsdata)) {
1938 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1939 WARN_ON(1);
1940 else
1941 kfree(dentry->d_fsdata);
1942 }
1943}
1944
1945const struct dentry_operations nfs_dentry_operations = {
1946 .d_revalidate = nfs_lookup_revalidate,
1947 .d_weak_revalidate = nfs_weak_revalidate,
1948 .d_delete = nfs_dentry_delete,
1949 .d_iput = nfs_dentry_iput,
1950 .d_automount = nfs_d_automount,
1951 .d_release = nfs_d_release,
1952};
1953EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1954
1955struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1956{
1957 struct dentry *res;
1958 struct inode *inode = NULL;
1959 struct nfs_fh *fhandle = NULL;
1960 struct nfs_fattr *fattr = NULL;
1961 unsigned long dir_verifier;
1962 int error;
1963
1964 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1965 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1966
1967 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1968 return ERR_PTR(-ENAMETOOLONG);
1969
1970 /*
1971 * If we're doing an exclusive create, optimize away the lookup
1972 * but don't hash the dentry.
1973 */
1974 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1975 return NULL;
1976
1977 res = ERR_PTR(-ENOMEM);
1978 fhandle = nfs_alloc_fhandle();
1979 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir));
1980 if (fhandle == NULL || fattr == NULL)
1981 goto out;
1982
1983 dir_verifier = nfs_save_change_attribute(dir);
1984 trace_nfs_lookup_enter(dir, dentry, flags);
1985 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1986 if (error == -ENOENT) {
1987 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1988 dir_verifier = inode_peek_iversion_raw(dir);
1989 goto no_entry;
1990 }
1991 if (error < 0) {
1992 res = ERR_PTR(error);
1993 goto out;
1994 }
1995 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1996 res = ERR_CAST(inode);
1997 if (IS_ERR(res))
1998 goto out;
1999
2000 /* Notify readdir to use READDIRPLUS */
2001 nfs_lookup_advise_force_readdirplus(dir, flags);
2002
2003no_entry:
2004 res = d_splice_alias(inode, dentry);
2005 if (res != NULL) {
2006 if (IS_ERR(res))
2007 goto out;
2008 dentry = res;
2009 }
2010 nfs_set_verifier(dentry, dir_verifier);
2011out:
2012 trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res));
2013 nfs_free_fattr(fattr);
2014 nfs_free_fhandle(fhandle);
2015 return res;
2016}
2017EXPORT_SYMBOL_GPL(nfs_lookup);
2018
2019void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
2020{
2021 /* Case insensitive server? Revalidate dentries */
2022 if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
2023 d_prune_aliases(inode);
2024}
2025EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
2026
2027#if IS_ENABLED(CONFIG_NFS_V4)
2028static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
2029
2030const struct dentry_operations nfs4_dentry_operations = {
2031 .d_revalidate = nfs4_lookup_revalidate,
2032 .d_weak_revalidate = nfs_weak_revalidate,
2033 .d_delete = nfs_dentry_delete,
2034 .d_iput = nfs_dentry_iput,
2035 .d_automount = nfs_d_automount,
2036 .d_release = nfs_d_release,
2037};
2038EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
2039
2040static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
2041{
2042 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
2043}
2044
2045static int do_open(struct inode *inode, struct file *filp)
2046{
2047 nfs_fscache_open_file(inode, filp);
2048 return 0;
2049}
2050
2051static int nfs_finish_open(struct nfs_open_context *ctx,
2052 struct dentry *dentry,
2053 struct file *file, unsigned open_flags)
2054{
2055 int err;
2056
2057 err = finish_open(file, dentry, do_open);
2058 if (err)
2059 goto out;
2060 if (S_ISREG(file_inode(file)->i_mode))
2061 nfs_file_set_open_context(file, ctx);
2062 else
2063 err = -EOPENSTALE;
2064out:
2065 return err;
2066}
2067
2068int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2069 struct file *file, unsigned open_flags,
2070 umode_t mode)
2071{
2072 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2073 struct nfs_open_context *ctx;
2074 struct dentry *res;
2075 struct iattr attr = { .ia_valid = ATTR_OPEN };
2076 struct inode *inode;
2077 unsigned int lookup_flags = 0;
2078 unsigned long dir_verifier;
2079 bool switched = false;
2080 int created = 0;
2081 int err;
2082
2083 /* Expect a negative dentry */
2084 BUG_ON(d_inode(dentry));
2085
2086 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2087 dir->i_sb->s_id, dir->i_ino, dentry);
2088
2089 err = nfs_check_flags(open_flags);
2090 if (err)
2091 return err;
2092
2093 /* NFS only supports OPEN on regular files */
2094 if ((open_flags & O_DIRECTORY)) {
2095 if (!d_in_lookup(dentry)) {
2096 /*
2097 * Hashed negative dentry with O_DIRECTORY: dentry was
2098 * revalidated and is fine, no need to perform lookup
2099 * again
2100 */
2101 return -ENOENT;
2102 }
2103 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2104 goto no_open;
2105 }
2106
2107 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
2108 return -ENAMETOOLONG;
2109
2110 if (open_flags & O_CREAT) {
2111 struct nfs_server *server = NFS_SERVER(dir);
2112
2113 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2114 mode &= ~current_umask();
2115
2116 attr.ia_valid |= ATTR_MODE;
2117 attr.ia_mode = mode;
2118 }
2119 if (open_flags & O_TRUNC) {
2120 attr.ia_valid |= ATTR_SIZE;
2121 attr.ia_size = 0;
2122 }
2123
2124 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2125 d_drop(dentry);
2126 switched = true;
2127 dentry = d_alloc_parallel(dentry->d_parent,
2128 &dentry->d_name, &wq);
2129 if (IS_ERR(dentry))
2130 return PTR_ERR(dentry);
2131 if (unlikely(!d_in_lookup(dentry)))
2132 return finish_no_open(file, dentry);
2133 }
2134
2135 ctx = create_nfs_open_context(dentry, open_flags, file);
2136 err = PTR_ERR(ctx);
2137 if (IS_ERR(ctx))
2138 goto out;
2139
2140 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
2141 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
2142 if (created)
2143 file->f_mode |= FMODE_CREATED;
2144 if (IS_ERR(inode)) {
2145 err = PTR_ERR(inode);
2146 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2147 put_nfs_open_context(ctx);
2148 d_drop(dentry);
2149 switch (err) {
2150 case -ENOENT:
2151 d_splice_alias(NULL, dentry);
2152 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
2153 dir_verifier = inode_peek_iversion_raw(dir);
2154 else
2155 dir_verifier = nfs_save_change_attribute(dir);
2156 nfs_set_verifier(dentry, dir_verifier);
2157 break;
2158 case -EISDIR:
2159 case -ENOTDIR:
2160 goto no_open;
2161 case -ELOOP:
2162 if (!(open_flags & O_NOFOLLOW))
2163 goto no_open;
2164 break;
2165 /* case -EINVAL: */
2166 default:
2167 break;
2168 }
2169 goto out;
2170 }
2171 file->f_mode |= FMODE_CAN_ODIRECT;
2172
2173 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
2174 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2175 put_nfs_open_context(ctx);
2176out:
2177 if (unlikely(switched)) {
2178 d_lookup_done(dentry);
2179 dput(dentry);
2180 }
2181 return err;
2182
2183no_open:
2184 res = nfs_lookup(dir, dentry, lookup_flags);
2185 if (!res) {
2186 inode = d_inode(dentry);
2187 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2188 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2189 res = ERR_PTR(-ENOTDIR);
2190 else if (inode && S_ISREG(inode->i_mode))
2191 res = ERR_PTR(-EOPENSTALE);
2192 } else if (!IS_ERR(res)) {
2193 inode = d_inode(res);
2194 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2195 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2196 dput(res);
2197 res = ERR_PTR(-ENOTDIR);
2198 } else if (inode && S_ISREG(inode->i_mode)) {
2199 dput(res);
2200 res = ERR_PTR(-EOPENSTALE);
2201 }
2202 }
2203 if (switched) {
2204 d_lookup_done(dentry);
2205 if (!res)
2206 res = dentry;
2207 else
2208 dput(dentry);
2209 }
2210 if (IS_ERR(res))
2211 return PTR_ERR(res);
2212 return finish_no_open(file, res);
2213}
2214EXPORT_SYMBOL_GPL(nfs_atomic_open);
2215
2216static int
2217nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2218 unsigned int flags)
2219{
2220 struct inode *inode;
2221
2222 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
2223
2224 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2225 goto full_reval;
2226 if (d_mountpoint(dentry))
2227 goto full_reval;
2228
2229 inode = d_inode(dentry);
2230
2231 /* We can't create new files in nfs_open_revalidate(), so we
2232 * optimize away revalidation of negative dentries.
2233 */
2234 if (inode == NULL)
2235 goto full_reval;
2236
2237 if (nfs_verifier_is_delegated(dentry))
2238 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2239
2240 /* NFS only supports OPEN on regular files */
2241 if (!S_ISREG(inode->i_mode))
2242 goto full_reval;
2243
2244 /* We cannot do exclusive creation on a positive dentry */
2245 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2246 goto reval_dentry;
2247
2248 /* Check if the directory changed */
2249 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2250 goto reval_dentry;
2251
2252 /* Let f_op->open() actually open (and revalidate) the file */
2253 return 1;
2254reval_dentry:
2255 if (flags & LOOKUP_RCU)
2256 return -ECHILD;
2257 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
2258
2259full_reval:
2260 return nfs_do_lookup_revalidate(dir, dentry, flags);
2261}
2262
2263static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2264{
2265 return __nfs_lookup_revalidate(dentry, flags,
2266 nfs4_do_lookup_revalidate);
2267}
2268
2269#endif /* CONFIG_NFSV4 */
2270
2271int nfs_atomic_open_v23(struct inode *dir, struct dentry *dentry,
2272 struct file *file, unsigned int open_flags,
2273 umode_t mode)
2274{
2275
2276 /* Same as look+open from lookup_open(), but with different O_TRUNC
2277 * handling.
2278 */
2279 int error = 0;
2280
2281 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
2282 return -ENAMETOOLONG;
2283
2284 if (open_flags & O_CREAT) {
2285 file->f_mode |= FMODE_CREATED;
2286 error = nfs_do_create(dir, dentry, mode, open_flags);
2287 if (error)
2288 return error;
2289 return finish_open(file, dentry, NULL);
2290 } else if (d_in_lookup(dentry)) {
2291 /* The only flags nfs_lookup considers are
2292 * LOOKUP_EXCL and LOOKUP_RENAME_TARGET, and
2293 * we want those to be zero so the lookup isn't skipped.
2294 */
2295 struct dentry *res = nfs_lookup(dir, dentry, 0);
2296
2297 d_lookup_done(dentry);
2298 if (unlikely(res)) {
2299 if (IS_ERR(res))
2300 return PTR_ERR(res);
2301 return finish_no_open(file, res);
2302 }
2303 }
2304 return finish_no_open(file, NULL);
2305
2306}
2307EXPORT_SYMBOL_GPL(nfs_atomic_open_v23);
2308
2309struct dentry *
2310nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2311 struct nfs_fattr *fattr)
2312{
2313 struct dentry *parent = dget_parent(dentry);
2314 struct inode *dir = d_inode(parent);
2315 struct inode *inode;
2316 struct dentry *d;
2317 int error;
2318
2319 d_drop(dentry);
2320
2321 if (fhandle->size == 0) {
2322 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
2323 if (error)
2324 goto out_error;
2325 }
2326 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2327 if (!(fattr->valid & NFS_ATTR_FATTR)) {
2328 struct nfs_server *server = NFS_SB(dentry->d_sb);
2329 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2330 fattr, NULL);
2331 if (error < 0)
2332 goto out_error;
2333 }
2334 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2335 d = d_splice_alias(inode, dentry);
2336out:
2337 dput(parent);
2338 return d;
2339out_error:
2340 d = ERR_PTR(error);
2341 goto out;
2342}
2343EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2344
2345/*
2346 * Code common to create, mkdir, and mknod.
2347 */
2348int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2349 struct nfs_fattr *fattr)
2350{
2351 struct dentry *d;
2352
2353 d = nfs_add_or_obtain(dentry, fhandle, fattr);
2354 if (IS_ERR(d))
2355 return PTR_ERR(d);
2356
2357 /* Callers don't care */
2358 dput(d);
2359 return 0;
2360}
2361EXPORT_SYMBOL_GPL(nfs_instantiate);
2362
2363/*
2364 * Following a failed create operation, we drop the dentry rather
2365 * than retain a negative dentry. This avoids a problem in the event
2366 * that the operation succeeded on the server, but an error in the
2367 * reply path made it appear to have failed.
2368 */
2369static int nfs_do_create(struct inode *dir, struct dentry *dentry,
2370 umode_t mode, int open_flags)
2371{
2372 struct iattr attr;
2373 int error;
2374
2375 open_flags |= O_CREAT;
2376
2377 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2378 dir->i_sb->s_id, dir->i_ino, dentry);
2379
2380 attr.ia_mode = mode;
2381 attr.ia_valid = ATTR_MODE;
2382 if (open_flags & O_TRUNC) {
2383 attr.ia_size = 0;
2384 attr.ia_valid |= ATTR_SIZE;
2385 }
2386
2387 trace_nfs_create_enter(dir, dentry, open_flags);
2388 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2389 trace_nfs_create_exit(dir, dentry, open_flags, error);
2390 if (error != 0)
2391 goto out_err;
2392 return 0;
2393out_err:
2394 d_drop(dentry);
2395 return error;
2396}
2397
2398int nfs_create(struct mnt_idmap *idmap, struct inode *dir,
2399 struct dentry *dentry, umode_t mode, bool excl)
2400{
2401 return nfs_do_create(dir, dentry, mode, excl ? O_EXCL : 0);
2402}
2403EXPORT_SYMBOL_GPL(nfs_create);
2404
2405/*
2406 * See comments for nfs_proc_create regarding failed operations.
2407 */
2408int
2409nfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
2410 struct dentry *dentry, umode_t mode, dev_t rdev)
2411{
2412 struct iattr attr;
2413 int status;
2414
2415 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2416 dir->i_sb->s_id, dir->i_ino, dentry);
2417
2418 attr.ia_mode = mode;
2419 attr.ia_valid = ATTR_MODE;
2420
2421 trace_nfs_mknod_enter(dir, dentry);
2422 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2423 trace_nfs_mknod_exit(dir, dentry, status);
2424 if (status != 0)
2425 goto out_err;
2426 return 0;
2427out_err:
2428 d_drop(dentry);
2429 return status;
2430}
2431EXPORT_SYMBOL_GPL(nfs_mknod);
2432
2433/*
2434 * See comments for nfs_proc_create regarding failed operations.
2435 */
2436int nfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2437 struct dentry *dentry, umode_t mode)
2438{
2439 struct iattr attr;
2440 int error;
2441
2442 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2443 dir->i_sb->s_id, dir->i_ino, dentry);
2444
2445 attr.ia_valid = ATTR_MODE;
2446 attr.ia_mode = mode | S_IFDIR;
2447
2448 trace_nfs_mkdir_enter(dir, dentry);
2449 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2450 trace_nfs_mkdir_exit(dir, dentry, error);
2451 if (error != 0)
2452 goto out_err;
2453 return 0;
2454out_err:
2455 d_drop(dentry);
2456 return error;
2457}
2458EXPORT_SYMBOL_GPL(nfs_mkdir);
2459
2460static void nfs_dentry_handle_enoent(struct dentry *dentry)
2461{
2462 if (simple_positive(dentry))
2463 d_delete(dentry);
2464}
2465
2466static void nfs_dentry_remove_handle_error(struct inode *dir,
2467 struct dentry *dentry, int error)
2468{
2469 switch (error) {
2470 case -ENOENT:
2471 if (d_really_is_positive(dentry))
2472 d_delete(dentry);
2473 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2474 break;
2475 case 0:
2476 nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2477 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2478 }
2479}
2480
2481int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2482{
2483 int error;
2484
2485 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2486 dir->i_sb->s_id, dir->i_ino, dentry);
2487
2488 trace_nfs_rmdir_enter(dir, dentry);
2489 if (d_really_is_positive(dentry)) {
2490 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2491 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2492 /* Ensure the VFS deletes this inode */
2493 switch (error) {
2494 case 0:
2495 clear_nlink(d_inode(dentry));
2496 break;
2497 case -ENOENT:
2498 nfs_dentry_handle_enoent(dentry);
2499 }
2500 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2501 } else
2502 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2503 nfs_dentry_remove_handle_error(dir, dentry, error);
2504 trace_nfs_rmdir_exit(dir, dentry, error);
2505
2506 return error;
2507}
2508EXPORT_SYMBOL_GPL(nfs_rmdir);
2509
2510/*
2511 * Remove a file after making sure there are no pending writes,
2512 * and after checking that the file has only one user.
2513 *
2514 * We invalidate the attribute cache and free the inode prior to the operation
2515 * to avoid possible races if the server reuses the inode.
2516 */
2517static int nfs_safe_remove(struct dentry *dentry)
2518{
2519 struct inode *dir = d_inode(dentry->d_parent);
2520 struct inode *inode = d_inode(dentry);
2521 int error = -EBUSY;
2522
2523 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2524
2525 /* If the dentry was sillyrenamed, we simply call d_delete() */
2526 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2527 error = 0;
2528 goto out;
2529 }
2530
2531 trace_nfs_remove_enter(dir, dentry);
2532 if (inode != NULL) {
2533 error = NFS_PROTO(dir)->remove(dir, dentry);
2534 if (error == 0)
2535 nfs_drop_nlink(inode);
2536 } else
2537 error = NFS_PROTO(dir)->remove(dir, dentry);
2538 if (error == -ENOENT)
2539 nfs_dentry_handle_enoent(dentry);
2540 trace_nfs_remove_exit(dir, dentry, error);
2541out:
2542 return error;
2543}
2544
2545/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2546 * belongs to an active ".nfs..." file and we return -EBUSY.
2547 *
2548 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2549 */
2550int nfs_unlink(struct inode *dir, struct dentry *dentry)
2551{
2552 int error;
2553
2554 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2555 dir->i_ino, dentry);
2556
2557 trace_nfs_unlink_enter(dir, dentry);
2558 spin_lock(&dentry->d_lock);
2559 if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2560 &NFS_I(d_inode(dentry))->flags)) {
2561 spin_unlock(&dentry->d_lock);
2562 /* Start asynchronous writeout of the inode */
2563 write_inode_now(d_inode(dentry), 0);
2564 error = nfs_sillyrename(dir, dentry);
2565 goto out;
2566 }
2567 /* We must prevent any concurrent open until the unlink
2568 * completes. ->d_revalidate will wait for ->d_fsdata
2569 * to clear. We set it here to ensure no lookup succeeds until
2570 * the unlink is complete on the server.
2571 */
2572 error = -ETXTBSY;
2573 if (WARN_ON(dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2574 WARN_ON(dentry->d_fsdata == NFS_FSDATA_BLOCKED)) {
2575 spin_unlock(&dentry->d_lock);
2576 goto out;
2577 }
2578 block_revalidate(dentry);
2579
2580 spin_unlock(&dentry->d_lock);
2581 error = nfs_safe_remove(dentry);
2582 nfs_dentry_remove_handle_error(dir, dentry, error);
2583 unblock_revalidate(dentry);
2584out:
2585 trace_nfs_unlink_exit(dir, dentry, error);
2586 return error;
2587}
2588EXPORT_SYMBOL_GPL(nfs_unlink);
2589
2590/*
2591 * To create a symbolic link, most file systems instantiate a new inode,
2592 * add a page to it containing the path, then write it out to the disk
2593 * using prepare_write/commit_write.
2594 *
2595 * Unfortunately the NFS client can't create the in-core inode first
2596 * because it needs a file handle to create an in-core inode (see
2597 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2598 * symlink request has completed on the server.
2599 *
2600 * So instead we allocate a raw page, copy the symname into it, then do
2601 * the SYMLINK request with the page as the buffer. If it succeeds, we
2602 * now have a new file handle and can instantiate an in-core NFS inode
2603 * and move the raw page into its mapping.
2604 */
2605int nfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
2606 struct dentry *dentry, const char *symname)
2607{
2608 struct folio *folio;
2609 char *kaddr;
2610 struct iattr attr;
2611 unsigned int pathlen = strlen(symname);
2612 int error;
2613
2614 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2615 dir->i_ino, dentry, symname);
2616
2617 if (pathlen > PAGE_SIZE)
2618 return -ENAMETOOLONG;
2619
2620 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2621 attr.ia_valid = ATTR_MODE;
2622
2623 folio = folio_alloc(GFP_USER, 0);
2624 if (!folio)
2625 return -ENOMEM;
2626
2627 kaddr = folio_address(folio);
2628 memcpy(kaddr, symname, pathlen);
2629 if (pathlen < PAGE_SIZE)
2630 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2631
2632 trace_nfs_symlink_enter(dir, dentry);
2633 error = NFS_PROTO(dir)->symlink(dir, dentry, folio, pathlen, &attr);
2634 trace_nfs_symlink_exit(dir, dentry, error);
2635 if (error != 0) {
2636 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2637 dir->i_sb->s_id, dir->i_ino,
2638 dentry, symname, error);
2639 d_drop(dentry);
2640 folio_put(folio);
2641 return error;
2642 }
2643
2644 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2645
2646 /*
2647 * No big deal if we can't add this page to the page cache here.
2648 * READLINK will get the missing page from the server if needed.
2649 */
2650 if (filemap_add_folio(d_inode(dentry)->i_mapping, folio, 0,
2651 GFP_KERNEL) == 0) {
2652 folio_mark_uptodate(folio);
2653 folio_unlock(folio);
2654 }
2655
2656 folio_put(folio);
2657 return 0;
2658}
2659EXPORT_SYMBOL_GPL(nfs_symlink);
2660
2661int
2662nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2663{
2664 struct inode *inode = d_inode(old_dentry);
2665 int error;
2666
2667 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2668 old_dentry, dentry);
2669
2670 trace_nfs_link_enter(inode, dir, dentry);
2671 d_drop(dentry);
2672 if (S_ISREG(inode->i_mode))
2673 nfs_sync_inode(inode);
2674 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2675 if (error == 0) {
2676 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2677 ihold(inode);
2678 d_add(dentry, inode);
2679 }
2680 trace_nfs_link_exit(inode, dir, dentry, error);
2681 return error;
2682}
2683EXPORT_SYMBOL_GPL(nfs_link);
2684
2685static void
2686nfs_unblock_rename(struct rpc_task *task, struct nfs_renamedata *data)
2687{
2688 struct dentry *new_dentry = data->new_dentry;
2689
2690 unblock_revalidate(new_dentry);
2691}
2692
2693/*
2694 * RENAME
2695 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2696 * different file handle for the same inode after a rename (e.g. when
2697 * moving to a different directory). A fail-safe method to do so would
2698 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2699 * rename the old file using the sillyrename stuff. This way, the original
2700 * file in old_dir will go away when the last process iput()s the inode.
2701 *
2702 * FIXED.
2703 *
2704 * It actually works quite well. One needs to have the possibility for
2705 * at least one ".nfs..." file in each directory the file ever gets
2706 * moved or linked to which happens automagically with the new
2707 * implementation that only depends on the dcache stuff instead of
2708 * using the inode layer
2709 *
2710 * Unfortunately, things are a little more complicated than indicated
2711 * above. For a cross-directory move, we want to make sure we can get
2712 * rid of the old inode after the operation. This means there must be
2713 * no pending writes (if it's a file), and the use count must be 1.
2714 * If these conditions are met, we can drop the dentries before doing
2715 * the rename.
2716 */
2717int nfs_rename(struct mnt_idmap *idmap, struct inode *old_dir,
2718 struct dentry *old_dentry, struct inode *new_dir,
2719 struct dentry *new_dentry, unsigned int flags)
2720{
2721 struct inode *old_inode = d_inode(old_dentry);
2722 struct inode *new_inode = d_inode(new_dentry);
2723 struct dentry *dentry = NULL;
2724 struct rpc_task *task;
2725 bool must_unblock = false;
2726 int error = -EBUSY;
2727
2728 if (flags)
2729 return -EINVAL;
2730
2731 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2732 old_dentry, new_dentry,
2733 d_count(new_dentry));
2734
2735 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2736 /*
2737 * For non-directories, check whether the target is busy and if so,
2738 * make a copy of the dentry and then do a silly-rename. If the
2739 * silly-rename succeeds, the copied dentry is hashed and becomes
2740 * the new target.
2741 */
2742 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2743 /* We must prevent any concurrent open until the unlink
2744 * completes. ->d_revalidate will wait for ->d_fsdata
2745 * to clear. We set it here to ensure no lookup succeeds until
2746 * the unlink is complete on the server.
2747 */
2748 error = -ETXTBSY;
2749 if (WARN_ON(new_dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2750 WARN_ON(new_dentry->d_fsdata == NFS_FSDATA_BLOCKED))
2751 goto out;
2752
2753 spin_lock(&new_dentry->d_lock);
2754 if (d_count(new_dentry) > 2) {
2755 int err;
2756
2757 spin_unlock(&new_dentry->d_lock);
2758
2759 /* copy the target dentry's name */
2760 dentry = d_alloc(new_dentry->d_parent,
2761 &new_dentry->d_name);
2762 if (!dentry)
2763 goto out;
2764
2765 /* silly-rename the existing target ... */
2766 err = nfs_sillyrename(new_dir, new_dentry);
2767 if (err)
2768 goto out;
2769
2770 new_dentry = dentry;
2771 new_inode = NULL;
2772 } else {
2773 block_revalidate(new_dentry);
2774 must_unblock = true;
2775 spin_unlock(&new_dentry->d_lock);
2776 }
2777
2778 }
2779
2780 if (S_ISREG(old_inode->i_mode))
2781 nfs_sync_inode(old_inode);
2782 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry,
2783 must_unblock ? nfs_unblock_rename : NULL);
2784 if (IS_ERR(task)) {
2785 if (must_unblock)
2786 unblock_revalidate(new_dentry);
2787 error = PTR_ERR(task);
2788 goto out;
2789 }
2790
2791 error = rpc_wait_for_completion_task(task);
2792 if (error != 0) {
2793 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2794 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2795 smp_wmb();
2796 } else
2797 error = task->tk_status;
2798 rpc_put_task(task);
2799 /* Ensure the inode attributes are revalidated */
2800 if (error == 0) {
2801 spin_lock(&old_inode->i_lock);
2802 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2803 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2804 NFS_INO_INVALID_CTIME |
2805 NFS_INO_REVAL_FORCED);
2806 spin_unlock(&old_inode->i_lock);
2807 }
2808out:
2809 trace_nfs_rename_exit(old_dir, old_dentry,
2810 new_dir, new_dentry, error);
2811 if (!error) {
2812 if (new_inode != NULL)
2813 nfs_drop_nlink(new_inode);
2814 /*
2815 * The d_move() should be here instead of in an async RPC completion
2816 * handler because we need the proper locks to move the dentry. If
2817 * we're interrupted by a signal, the async RPC completion handler
2818 * should mark the directories for revalidation.
2819 */
2820 d_move(old_dentry, new_dentry);
2821 nfs_set_verifier(old_dentry,
2822 nfs_save_change_attribute(new_dir));
2823 } else if (error == -ENOENT)
2824 nfs_dentry_handle_enoent(old_dentry);
2825
2826 /* new dentry created? */
2827 if (dentry)
2828 dput(dentry);
2829 return error;
2830}
2831EXPORT_SYMBOL_GPL(nfs_rename);
2832
2833static DEFINE_SPINLOCK(nfs_access_lru_lock);
2834static LIST_HEAD(nfs_access_lru_list);
2835static atomic_long_t nfs_access_nr_entries;
2836
2837static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2838module_param(nfs_access_max_cachesize, ulong, 0644);
2839MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2840
2841static void nfs_access_free_entry(struct nfs_access_entry *entry)
2842{
2843 put_group_info(entry->group_info);
2844 kfree_rcu(entry, rcu_head);
2845 smp_mb__before_atomic();
2846 atomic_long_dec(&nfs_access_nr_entries);
2847 smp_mb__after_atomic();
2848}
2849
2850static void nfs_access_free_list(struct list_head *head)
2851{
2852 struct nfs_access_entry *cache;
2853
2854 while (!list_empty(head)) {
2855 cache = list_entry(head->next, struct nfs_access_entry, lru);
2856 list_del(&cache->lru);
2857 nfs_access_free_entry(cache);
2858 }
2859}
2860
2861static unsigned long
2862nfs_do_access_cache_scan(unsigned int nr_to_scan)
2863{
2864 LIST_HEAD(head);
2865 struct nfs_inode *nfsi, *next;
2866 struct nfs_access_entry *cache;
2867 long freed = 0;
2868
2869 spin_lock(&nfs_access_lru_lock);
2870 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2871 struct inode *inode;
2872
2873 if (nr_to_scan-- == 0)
2874 break;
2875 inode = &nfsi->vfs_inode;
2876 spin_lock(&inode->i_lock);
2877 if (list_empty(&nfsi->access_cache_entry_lru))
2878 goto remove_lru_entry;
2879 cache = list_entry(nfsi->access_cache_entry_lru.next,
2880 struct nfs_access_entry, lru);
2881 list_move(&cache->lru, &head);
2882 rb_erase(&cache->rb_node, &nfsi->access_cache);
2883 freed++;
2884 if (!list_empty(&nfsi->access_cache_entry_lru))
2885 list_move_tail(&nfsi->access_cache_inode_lru,
2886 &nfs_access_lru_list);
2887 else {
2888remove_lru_entry:
2889 list_del_init(&nfsi->access_cache_inode_lru);
2890 smp_mb__before_atomic();
2891 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2892 smp_mb__after_atomic();
2893 }
2894 spin_unlock(&inode->i_lock);
2895 }
2896 spin_unlock(&nfs_access_lru_lock);
2897 nfs_access_free_list(&head);
2898 return freed;
2899}
2900
2901unsigned long
2902nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2903{
2904 int nr_to_scan = sc->nr_to_scan;
2905 gfp_t gfp_mask = sc->gfp_mask;
2906
2907 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2908 return SHRINK_STOP;
2909 return nfs_do_access_cache_scan(nr_to_scan);
2910}
2911
2912
2913unsigned long
2914nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2915{
2916 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2917}
2918
2919static void
2920nfs_access_cache_enforce_limit(void)
2921{
2922 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2923 unsigned long diff;
2924 unsigned int nr_to_scan;
2925
2926 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2927 return;
2928 nr_to_scan = 100;
2929 diff = nr_entries - nfs_access_max_cachesize;
2930 if (diff < nr_to_scan)
2931 nr_to_scan = diff;
2932 nfs_do_access_cache_scan(nr_to_scan);
2933}
2934
2935static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2936{
2937 struct rb_root *root_node = &nfsi->access_cache;
2938 struct rb_node *n;
2939 struct nfs_access_entry *entry;
2940
2941 /* Unhook entries from the cache */
2942 while ((n = rb_first(root_node)) != NULL) {
2943 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2944 rb_erase(n, root_node);
2945 list_move(&entry->lru, head);
2946 }
2947 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2948}
2949
2950void nfs_access_zap_cache(struct inode *inode)
2951{
2952 LIST_HEAD(head);
2953
2954 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2955 return;
2956 /* Remove from global LRU init */
2957 spin_lock(&nfs_access_lru_lock);
2958 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2959 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2960
2961 spin_lock(&inode->i_lock);
2962 __nfs_access_zap_cache(NFS_I(inode), &head);
2963 spin_unlock(&inode->i_lock);
2964 spin_unlock(&nfs_access_lru_lock);
2965 nfs_access_free_list(&head);
2966}
2967EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2968
2969static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2970{
2971 struct group_info *ga, *gb;
2972 int g;
2973
2974 if (uid_lt(a->fsuid, b->fsuid))
2975 return -1;
2976 if (uid_gt(a->fsuid, b->fsuid))
2977 return 1;
2978
2979 if (gid_lt(a->fsgid, b->fsgid))
2980 return -1;
2981 if (gid_gt(a->fsgid, b->fsgid))
2982 return 1;
2983
2984 ga = a->group_info;
2985 gb = b->group_info;
2986 if (ga == gb)
2987 return 0;
2988 if (ga == NULL)
2989 return -1;
2990 if (gb == NULL)
2991 return 1;
2992 if (ga->ngroups < gb->ngroups)
2993 return -1;
2994 if (ga->ngroups > gb->ngroups)
2995 return 1;
2996
2997 for (g = 0; g < ga->ngroups; g++) {
2998 if (gid_lt(ga->gid[g], gb->gid[g]))
2999 return -1;
3000 if (gid_gt(ga->gid[g], gb->gid[g]))
3001 return 1;
3002 }
3003 return 0;
3004}
3005
3006static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
3007{
3008 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
3009
3010 while (n != NULL) {
3011 struct nfs_access_entry *entry =
3012 rb_entry(n, struct nfs_access_entry, rb_node);
3013 int cmp = access_cmp(cred, entry);
3014
3015 if (cmp < 0)
3016 n = n->rb_left;
3017 else if (cmp > 0)
3018 n = n->rb_right;
3019 else
3020 return entry;
3021 }
3022 return NULL;
3023}
3024
3025static u64 nfs_access_login_time(const struct task_struct *task,
3026 const struct cred *cred)
3027{
3028 const struct task_struct *parent;
3029 const struct cred *pcred;
3030 u64 ret;
3031
3032 rcu_read_lock();
3033 for (;;) {
3034 parent = rcu_dereference(task->real_parent);
3035 pcred = __task_cred(parent);
3036 if (parent == task || cred_fscmp(pcred, cred) != 0)
3037 break;
3038 task = parent;
3039 }
3040 ret = task->start_time;
3041 rcu_read_unlock();
3042 return ret;
3043}
3044
3045static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
3046{
3047 struct nfs_inode *nfsi = NFS_I(inode);
3048 u64 login_time = nfs_access_login_time(current, cred);
3049 struct nfs_access_entry *cache;
3050 bool retry = true;
3051 int err;
3052
3053 spin_lock(&inode->i_lock);
3054 for(;;) {
3055 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3056 goto out_zap;
3057 cache = nfs_access_search_rbtree(inode, cred);
3058 err = -ENOENT;
3059 if (cache == NULL)
3060 goto out;
3061 /* Found an entry, is our attribute cache valid? */
3062 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3063 break;
3064 if (!retry)
3065 break;
3066 err = -ECHILD;
3067 if (!may_block)
3068 goto out;
3069 spin_unlock(&inode->i_lock);
3070 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
3071 if (err)
3072 return err;
3073 spin_lock(&inode->i_lock);
3074 retry = false;
3075 }
3076 err = -ENOENT;
3077 if ((s64)(login_time - cache->timestamp) > 0)
3078 goto out;
3079 *mask = cache->mask;
3080 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
3081 err = 0;
3082out:
3083 spin_unlock(&inode->i_lock);
3084 return err;
3085out_zap:
3086 spin_unlock(&inode->i_lock);
3087 nfs_access_zap_cache(inode);
3088 return -ENOENT;
3089}
3090
3091static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
3092{
3093 /* Only check the most recently returned cache entry,
3094 * but do it without locking.
3095 */
3096 struct nfs_inode *nfsi = NFS_I(inode);
3097 u64 login_time = nfs_access_login_time(current, cred);
3098 struct nfs_access_entry *cache;
3099 int err = -ECHILD;
3100 struct list_head *lh;
3101
3102 rcu_read_lock();
3103 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3104 goto out;
3105 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
3106 cache = list_entry(lh, struct nfs_access_entry, lru);
3107 if (lh == &nfsi->access_cache_entry_lru ||
3108 access_cmp(cred, cache) != 0)
3109 cache = NULL;
3110 if (cache == NULL)
3111 goto out;
3112 if ((s64)(login_time - cache->timestamp) > 0)
3113 goto out;
3114 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3115 goto out;
3116 *mask = cache->mask;
3117 err = 0;
3118out:
3119 rcu_read_unlock();
3120 return err;
3121}
3122
3123int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
3124 u32 *mask, bool may_block)
3125{
3126 int status;
3127
3128 status = nfs_access_get_cached_rcu(inode, cred, mask);
3129 if (status != 0)
3130 status = nfs_access_get_cached_locked(inode, cred, mask,
3131 may_block);
3132
3133 return status;
3134}
3135EXPORT_SYMBOL_GPL(nfs_access_get_cached);
3136
3137static void nfs_access_add_rbtree(struct inode *inode,
3138 struct nfs_access_entry *set,
3139 const struct cred *cred)
3140{
3141 struct nfs_inode *nfsi = NFS_I(inode);
3142 struct rb_root *root_node = &nfsi->access_cache;
3143 struct rb_node **p = &root_node->rb_node;
3144 struct rb_node *parent = NULL;
3145 struct nfs_access_entry *entry;
3146 int cmp;
3147
3148 spin_lock(&inode->i_lock);
3149 while (*p != NULL) {
3150 parent = *p;
3151 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3152 cmp = access_cmp(cred, entry);
3153
3154 if (cmp < 0)
3155 p = &parent->rb_left;
3156 else if (cmp > 0)
3157 p = &parent->rb_right;
3158 else
3159 goto found;
3160 }
3161 rb_link_node(&set->rb_node, parent, p);
3162 rb_insert_color(&set->rb_node, root_node);
3163 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3164 spin_unlock(&inode->i_lock);
3165 return;
3166found:
3167 rb_replace_node(parent, &set->rb_node, root_node);
3168 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3169 list_del(&entry->lru);
3170 spin_unlock(&inode->i_lock);
3171 nfs_access_free_entry(entry);
3172}
3173
3174void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3175 const struct cred *cred)
3176{
3177 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
3178 if (cache == NULL)
3179 return;
3180 RB_CLEAR_NODE(&cache->rb_node);
3181 cache->fsuid = cred->fsuid;
3182 cache->fsgid = cred->fsgid;
3183 cache->group_info = get_group_info(cred->group_info);
3184 cache->mask = set->mask;
3185 cache->timestamp = ktime_get_ns();
3186
3187 /* The above field assignments must be visible
3188 * before this item appears on the lru. We cannot easily
3189 * use rcu_assign_pointer, so just force the memory barrier.
3190 */
3191 smp_wmb();
3192 nfs_access_add_rbtree(inode, cache, cred);
3193
3194 /* Update accounting */
3195 smp_mb__before_atomic();
3196 atomic_long_inc(&nfs_access_nr_entries);
3197 smp_mb__after_atomic();
3198
3199 /* Add inode to global LRU list */
3200 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3201 spin_lock(&nfs_access_lru_lock);
3202 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
3203 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
3204 &nfs_access_lru_list);
3205 spin_unlock(&nfs_access_lru_lock);
3206 }
3207 nfs_access_cache_enforce_limit();
3208}
3209EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3210
3211#define NFS_MAY_READ (NFS_ACCESS_READ)
3212#define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3213 NFS_ACCESS_EXTEND | \
3214 NFS_ACCESS_DELETE)
3215#define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3216 NFS_ACCESS_EXTEND)
3217#define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3218#define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3219#define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3220static int
3221nfs_access_calc_mask(u32 access_result, umode_t umode)
3222{
3223 int mask = 0;
3224
3225 if (access_result & NFS_MAY_READ)
3226 mask |= MAY_READ;
3227 if (S_ISDIR(umode)) {
3228 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3229 mask |= MAY_WRITE;
3230 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3231 mask |= MAY_EXEC;
3232 } else if (S_ISREG(umode)) {
3233 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3234 mask |= MAY_WRITE;
3235 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3236 mask |= MAY_EXEC;
3237 } else if (access_result & NFS_MAY_WRITE)
3238 mask |= MAY_WRITE;
3239 return mask;
3240}
3241
3242void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3243{
3244 entry->mask = access_result;
3245}
3246EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3247
3248static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3249{
3250 struct nfs_access_entry cache;
3251 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3252 int cache_mask = -1;
3253 int status;
3254
3255 trace_nfs_access_enter(inode);
3256
3257 status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3258 if (status == 0)
3259 goto out_cached;
3260
3261 status = -ECHILD;
3262 if (!may_block)
3263 goto out;
3264
3265 /*
3266 * Determine which access bits we want to ask for...
3267 */
3268 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3269 nfs_access_xattr_mask(NFS_SERVER(inode));
3270 if (S_ISDIR(inode->i_mode))
3271 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3272 else
3273 cache.mask |= NFS_ACCESS_EXECUTE;
3274 status = NFS_PROTO(inode)->access(inode, &cache, cred);
3275 if (status != 0) {
3276 if (status == -ESTALE) {
3277 if (!S_ISDIR(inode->i_mode))
3278 nfs_set_inode_stale(inode);
3279 else
3280 nfs_zap_caches(inode);
3281 }
3282 goto out;
3283 }
3284 nfs_access_add_cache(inode, &cache, cred);
3285out_cached:
3286 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
3287 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3288 status = -EACCES;
3289out:
3290 trace_nfs_access_exit(inode, mask, cache_mask, status);
3291 return status;
3292}
3293
3294static int nfs_open_permission_mask(int openflags)
3295{
3296 int mask = 0;
3297
3298 if (openflags & __FMODE_EXEC) {
3299 /* ONLY check exec rights */
3300 mask = MAY_EXEC;
3301 } else {
3302 if ((openflags & O_ACCMODE) != O_WRONLY)
3303 mask |= MAY_READ;
3304 if ((openflags & O_ACCMODE) != O_RDONLY)
3305 mask |= MAY_WRITE;
3306 }
3307
3308 return mask;
3309}
3310
3311int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3312{
3313 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
3314}
3315EXPORT_SYMBOL_GPL(nfs_may_open);
3316
3317static int nfs_execute_ok(struct inode *inode, int mask)
3318{
3319 struct nfs_server *server = NFS_SERVER(inode);
3320 int ret = 0;
3321
3322 if (S_ISDIR(inode->i_mode))
3323 return 0;
3324 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3325 if (mask & MAY_NOT_BLOCK)
3326 return -ECHILD;
3327 ret = __nfs_revalidate_inode(server, inode);
3328 }
3329 if (ret == 0 && !execute_ok(inode))
3330 ret = -EACCES;
3331 return ret;
3332}
3333
3334int nfs_permission(struct mnt_idmap *idmap,
3335 struct inode *inode,
3336 int mask)
3337{
3338 const struct cred *cred = current_cred();
3339 int res = 0;
3340
3341 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
3342
3343 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3344 goto out;
3345 /* Is this sys_access() ? */
3346 if (mask & (MAY_ACCESS | MAY_CHDIR))
3347 goto force_lookup;
3348
3349 switch (inode->i_mode & S_IFMT) {
3350 case S_IFLNK:
3351 goto out;
3352 case S_IFREG:
3353 if ((mask & MAY_OPEN) &&
3354 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3355 return 0;
3356 break;
3357 case S_IFDIR:
3358 /*
3359 * Optimize away all write operations, since the server
3360 * will check permissions when we perform the op.
3361 */
3362 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3363 goto out;
3364 }
3365
3366force_lookup:
3367 if (!NFS_PROTO(inode)->access)
3368 goto out_notsup;
3369
3370 res = nfs_do_access(inode, cred, mask);
3371out:
3372 if (!res && (mask & MAY_EXEC))
3373 res = nfs_execute_ok(inode, mask);
3374
3375 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3376 inode->i_sb->s_id, inode->i_ino, mask, res);
3377 return res;
3378out_notsup:
3379 if (mask & MAY_NOT_BLOCK)
3380 return -ECHILD;
3381
3382 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3383 NFS_INO_INVALID_OTHER);
3384 if (res == 0)
3385 res = generic_permission(&nop_mnt_idmap, inode, mask);
3386 goto out;
3387}
3388EXPORT_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/module.h>
21#include <linux/time.h>
22#include <linux/errno.h>
23#include <linux/stat.h>
24#include <linux/fcntl.h>
25#include <linux/string.h>
26#include <linux/kernel.h>
27#include <linux/slab.h>
28#include <linux/mm.h>
29#include <linux/sunrpc/clnt.h>
30#include <linux/nfs_fs.h>
31#include <linux/nfs_mount.h>
32#include <linux/pagemap.h>
33#include <linux/pagevec.h>
34#include <linux/namei.h>
35#include <linux/mount.h>
36#include <linux/swap.h>
37#include <linux/sched.h>
38#include <linux/kmemleak.h>
39#include <linux/xattr.h>
40
41#include "delegation.h"
42#include "iostat.h"
43#include "internal.h"
44#include "fscache.h"
45
46#include "nfstrace.h"
47
48/* #define NFS_DEBUG_VERBOSE 1 */
49
50static int nfs_opendir(struct inode *, struct file *);
51static int nfs_closedir(struct inode *, struct file *);
52static int nfs_readdir(struct file *, struct dir_context *);
53static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
54static loff_t nfs_llseek_dir(struct file *, loff_t, int);
55static void nfs_readdir_clear_array(struct page*);
56
57const struct file_operations nfs_dir_operations = {
58 .llseek = nfs_llseek_dir,
59 .read = generic_read_dir,
60 .iterate_shared = nfs_readdir,
61 .open = nfs_opendir,
62 .release = nfs_closedir,
63 .fsync = nfs_fsync_dir,
64};
65
66const struct address_space_operations nfs_dir_aops = {
67 .freepage = nfs_readdir_clear_array,
68};
69
70static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
71{
72 struct nfs_inode *nfsi = NFS_I(dir);
73 struct nfs_open_dir_context *ctx;
74 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
75 if (ctx != NULL) {
76 ctx->duped = 0;
77 ctx->attr_gencount = nfsi->attr_gencount;
78 ctx->dir_cookie = 0;
79 ctx->dup_cookie = 0;
80 ctx->cred = get_rpccred(cred);
81 spin_lock(&dir->i_lock);
82 list_add(&ctx->list, &nfsi->open_files);
83 spin_unlock(&dir->i_lock);
84 return ctx;
85 }
86 return ERR_PTR(-ENOMEM);
87}
88
89static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
90{
91 spin_lock(&dir->i_lock);
92 list_del(&ctx->list);
93 spin_unlock(&dir->i_lock);
94 put_rpccred(ctx->cred);
95 kfree(ctx);
96}
97
98/*
99 * Open file
100 */
101static int
102nfs_opendir(struct inode *inode, struct file *filp)
103{
104 int res = 0;
105 struct nfs_open_dir_context *ctx;
106 struct rpc_cred *cred;
107
108 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
109
110 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
111
112 cred = rpc_lookup_cred();
113 if (IS_ERR(cred))
114 return PTR_ERR(cred);
115 ctx = alloc_nfs_open_dir_context(inode, cred);
116 if (IS_ERR(ctx)) {
117 res = PTR_ERR(ctx);
118 goto out;
119 }
120 filp->private_data = ctx;
121 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
122 /* This is a mountpoint, so d_revalidate will never
123 * have been called, so we need to refresh the
124 * inode (for close-open consistency) ourselves.
125 */
126 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
127 }
128out:
129 put_rpccred(cred);
130 return res;
131}
132
133static int
134nfs_closedir(struct inode *inode, struct file *filp)
135{
136 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
137 return 0;
138}
139
140struct nfs_cache_array_entry {
141 u64 cookie;
142 u64 ino;
143 struct qstr string;
144 unsigned char d_type;
145};
146
147struct nfs_cache_array {
148 atomic_t refcount;
149 int size;
150 int eof_index;
151 u64 last_cookie;
152 struct nfs_cache_array_entry array[0];
153};
154
155typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
156typedef struct {
157 struct file *file;
158 struct page *page;
159 struct dir_context *ctx;
160 unsigned long page_index;
161 u64 *dir_cookie;
162 u64 last_cookie;
163 loff_t current_index;
164 decode_dirent_t decode;
165
166 unsigned long timestamp;
167 unsigned long gencount;
168 unsigned int cache_entry_index;
169 unsigned int plus:1;
170 unsigned int eof:1;
171} nfs_readdir_descriptor_t;
172
173/*
174 * The caller is responsible for calling nfs_readdir_release_array(page)
175 */
176static
177struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
178{
179 void *ptr;
180 if (page == NULL)
181 return ERR_PTR(-EIO);
182 ptr = kmap(page);
183 if (ptr == NULL)
184 return ERR_PTR(-ENOMEM);
185 return ptr;
186}
187
188static
189void nfs_readdir_release_array(struct page *page)
190{
191 kunmap(page);
192}
193
194/*
195 * we are freeing strings created by nfs_add_to_readdir_array()
196 */
197static
198void nfs_readdir_clear_array(struct page *page)
199{
200 struct nfs_cache_array *array;
201 int i;
202
203 array = kmap_atomic(page);
204 if (atomic_dec_and_test(&array->refcount))
205 for (i = 0; i < array->size; i++)
206 kfree(array->array[i].string.name);
207 kunmap_atomic(array);
208}
209
210static bool grab_page(struct page *page)
211{
212 struct nfs_cache_array *array = kmap_atomic(page);
213 bool res = atomic_inc_not_zero(&array->refcount);
214 kunmap_atomic(array);
215 return res;
216}
217
218/*
219 * the caller is responsible for freeing qstr.name
220 * when called by nfs_readdir_add_to_array, the strings will be freed in
221 * nfs_clear_readdir_array()
222 */
223static
224int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
225{
226 string->len = len;
227 string->name = kmemdup(name, len, GFP_KERNEL);
228 if (string->name == NULL)
229 return -ENOMEM;
230 /*
231 * Avoid a kmemleak false positive. The pointer to the name is stored
232 * in a page cache page which kmemleak does not scan.
233 */
234 kmemleak_not_leak(string->name);
235 string->hash = full_name_hash(NULL, name, len);
236 return 0;
237}
238
239static
240int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
241{
242 struct nfs_cache_array *array = nfs_readdir_get_array(page);
243 struct nfs_cache_array_entry *cache_entry;
244 int ret;
245
246 if (IS_ERR(array))
247 return PTR_ERR(array);
248
249 cache_entry = &array->array[array->size];
250
251 /* Check that this entry lies within the page bounds */
252 ret = -ENOSPC;
253 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
254 goto out;
255
256 cache_entry->cookie = entry->prev_cookie;
257 cache_entry->ino = entry->ino;
258 cache_entry->d_type = entry->d_type;
259 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
260 if (ret)
261 goto out;
262 array->last_cookie = entry->cookie;
263 array->size++;
264 if (entry->eof != 0)
265 array->eof_index = array->size;
266out:
267 nfs_readdir_release_array(page);
268 return ret;
269}
270
271static
272int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
273{
274 loff_t diff = desc->ctx->pos - desc->current_index;
275 unsigned int index;
276
277 if (diff < 0)
278 goto out_eof;
279 if (diff >= array->size) {
280 if (array->eof_index >= 0)
281 goto out_eof;
282 return -EAGAIN;
283 }
284
285 index = (unsigned int)diff;
286 *desc->dir_cookie = array->array[index].cookie;
287 desc->cache_entry_index = index;
288 return 0;
289out_eof:
290 desc->eof = 1;
291 return -EBADCOOKIE;
292}
293
294static bool
295nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
296{
297 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
298 return false;
299 smp_rmb();
300 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
301}
302
303static
304int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
305{
306 int i;
307 loff_t new_pos;
308 int status = -EAGAIN;
309
310 for (i = 0; i < array->size; i++) {
311 if (array->array[i].cookie == *desc->dir_cookie) {
312 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
313 struct nfs_open_dir_context *ctx = desc->file->private_data;
314
315 new_pos = desc->current_index + i;
316 if (ctx->attr_gencount != nfsi->attr_gencount ||
317 !nfs_readdir_inode_mapping_valid(nfsi)) {
318 ctx->duped = 0;
319 ctx->attr_gencount = nfsi->attr_gencount;
320 } else if (new_pos < desc->ctx->pos) {
321 if (ctx->duped > 0
322 && ctx->dup_cookie == *desc->dir_cookie) {
323 if (printk_ratelimit()) {
324 pr_notice("NFS: directory %pD2 contains a readdir loop."
325 "Please contact your server vendor. "
326 "The file: %.*s has duplicate cookie %llu\n",
327 desc->file, array->array[i].string.len,
328 array->array[i].string.name, *desc->dir_cookie);
329 }
330 status = -ELOOP;
331 goto out;
332 }
333 ctx->dup_cookie = *desc->dir_cookie;
334 ctx->duped = -1;
335 }
336 desc->ctx->pos = new_pos;
337 desc->cache_entry_index = i;
338 return 0;
339 }
340 }
341 if (array->eof_index >= 0) {
342 status = -EBADCOOKIE;
343 if (*desc->dir_cookie == array->last_cookie)
344 desc->eof = 1;
345 }
346out:
347 return status;
348}
349
350static
351int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
352{
353 struct nfs_cache_array *array;
354 int status;
355
356 array = nfs_readdir_get_array(desc->page);
357 if (IS_ERR(array)) {
358 status = PTR_ERR(array);
359 goto out;
360 }
361
362 if (*desc->dir_cookie == 0)
363 status = nfs_readdir_search_for_pos(array, desc);
364 else
365 status = nfs_readdir_search_for_cookie(array, desc);
366
367 if (status == -EAGAIN) {
368 desc->last_cookie = array->last_cookie;
369 desc->current_index += array->size;
370 desc->page_index++;
371 }
372 nfs_readdir_release_array(desc->page);
373out:
374 return status;
375}
376
377/* Fill a page with xdr information before transferring to the cache page */
378static
379int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
380 struct nfs_entry *entry, struct file *file, struct inode *inode)
381{
382 struct nfs_open_dir_context *ctx = file->private_data;
383 struct rpc_cred *cred = ctx->cred;
384 unsigned long timestamp, gencount;
385 int error;
386
387 again:
388 timestamp = jiffies;
389 gencount = nfs_inc_attr_generation_counter();
390 error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages,
391 NFS_SERVER(inode)->dtsize, desc->plus);
392 if (error < 0) {
393 /* We requested READDIRPLUS, but the server doesn't grok it */
394 if (error == -ENOTSUPP && desc->plus) {
395 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
396 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
397 desc->plus = 0;
398 goto again;
399 }
400 goto error;
401 }
402 desc->timestamp = timestamp;
403 desc->gencount = gencount;
404error:
405 return error;
406}
407
408static int xdr_decode(nfs_readdir_descriptor_t *desc,
409 struct nfs_entry *entry, struct xdr_stream *xdr)
410{
411 int error;
412
413 error = desc->decode(xdr, entry, desc->plus);
414 if (error)
415 return error;
416 entry->fattr->time_start = desc->timestamp;
417 entry->fattr->gencount = desc->gencount;
418 return 0;
419}
420
421/* Match file and dirent using either filehandle or fileid
422 * Note: caller is responsible for checking the fsid
423 */
424static
425int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
426{
427 struct inode *inode;
428 struct nfs_inode *nfsi;
429
430 if (d_really_is_negative(dentry))
431 return 0;
432
433 inode = d_inode(dentry);
434 if (is_bad_inode(inode) || NFS_STALE(inode))
435 return 0;
436
437 nfsi = NFS_I(inode);
438 if (entry->fattr->fileid != nfsi->fileid)
439 return 0;
440 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
441 return 0;
442 return 1;
443}
444
445static
446bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
447{
448 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
449 return false;
450 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
451 return true;
452 if (ctx->pos == 0)
453 return true;
454 return false;
455}
456
457/*
458 * This function is called by the lookup and getattr code to request the
459 * use of readdirplus to accelerate any future lookups in the same
460 * directory.
461 */
462void nfs_advise_use_readdirplus(struct inode *dir)
463{
464 struct nfs_inode *nfsi = NFS_I(dir);
465
466 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
467 !list_empty(&nfsi->open_files))
468 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
469}
470
471/*
472 * This function is mainly for use by nfs_getattr().
473 *
474 * If this is an 'ls -l', we want to force use of readdirplus.
475 * Do this by checking if there is an active file descriptor
476 * and calling nfs_advise_use_readdirplus, then forcing a
477 * cache flush.
478 */
479void nfs_force_use_readdirplus(struct inode *dir)
480{
481 struct nfs_inode *nfsi = NFS_I(dir);
482
483 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
484 !list_empty(&nfsi->open_files)) {
485 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
486 invalidate_mapping_pages(dir->i_mapping, 0, -1);
487 }
488}
489
490static
491void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
492{
493 struct qstr filename = QSTR_INIT(entry->name, entry->len);
494 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
495 struct dentry *dentry;
496 struct dentry *alias;
497 struct inode *dir = d_inode(parent);
498 struct inode *inode;
499 int status;
500
501 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
502 return;
503 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
504 return;
505 if (filename.len == 0)
506 return;
507 /* Validate that the name doesn't contain any illegal '\0' */
508 if (strnlen(filename.name, filename.len) != filename.len)
509 return;
510 /* ...or '/' */
511 if (strnchr(filename.name, filename.len, '/'))
512 return;
513 if (filename.name[0] == '.') {
514 if (filename.len == 1)
515 return;
516 if (filename.len == 2 && filename.name[1] == '.')
517 return;
518 }
519 filename.hash = full_name_hash(parent, filename.name, filename.len);
520
521 dentry = d_lookup(parent, &filename);
522again:
523 if (!dentry) {
524 dentry = d_alloc_parallel(parent, &filename, &wq);
525 if (IS_ERR(dentry))
526 return;
527 }
528 if (!d_in_lookup(dentry)) {
529 /* Is there a mountpoint here? If so, just exit */
530 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
531 &entry->fattr->fsid))
532 goto out;
533 if (nfs_same_file(dentry, entry)) {
534 if (!entry->fh->size)
535 goto out;
536 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
537 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
538 if (!status)
539 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
540 goto out;
541 } else {
542 d_invalidate(dentry);
543 dput(dentry);
544 dentry = NULL;
545 goto again;
546 }
547 }
548 if (!entry->fh->size) {
549 d_lookup_done(dentry);
550 goto out;
551 }
552
553 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
554 alias = d_splice_alias(inode, dentry);
555 d_lookup_done(dentry);
556 if (alias) {
557 if (IS_ERR(alias))
558 goto out;
559 dput(dentry);
560 dentry = alias;
561 }
562 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
563out:
564 dput(dentry);
565}
566
567/* Perform conversion from xdr to cache array */
568static
569int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
570 struct page **xdr_pages, struct page *page, unsigned int buflen)
571{
572 struct xdr_stream stream;
573 struct xdr_buf buf;
574 struct page *scratch;
575 struct nfs_cache_array *array;
576 unsigned int count = 0;
577 int status;
578
579 scratch = alloc_page(GFP_KERNEL);
580 if (scratch == NULL)
581 return -ENOMEM;
582
583 if (buflen == 0)
584 goto out_nopages;
585
586 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
587 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
588
589 do {
590 status = xdr_decode(desc, entry, &stream);
591 if (status != 0) {
592 if (status == -EAGAIN)
593 status = 0;
594 break;
595 }
596
597 count++;
598
599 if (desc->plus != 0)
600 nfs_prime_dcache(file_dentry(desc->file), entry);
601
602 status = nfs_readdir_add_to_array(entry, page);
603 if (status != 0)
604 break;
605 } while (!entry->eof);
606
607out_nopages:
608 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
609 array = nfs_readdir_get_array(page);
610 if (!IS_ERR(array)) {
611 array->eof_index = array->size;
612 status = 0;
613 nfs_readdir_release_array(page);
614 } else
615 status = PTR_ERR(array);
616 }
617
618 put_page(scratch);
619 return status;
620}
621
622static
623void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
624{
625 unsigned int i;
626 for (i = 0; i < npages; i++)
627 put_page(pages[i]);
628}
629
630/*
631 * nfs_readdir_large_page will allocate pages that must be freed with a call
632 * to nfs_readdir_free_pagearray
633 */
634static
635int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
636{
637 unsigned int i;
638
639 for (i = 0; i < npages; i++) {
640 struct page *page = alloc_page(GFP_KERNEL);
641 if (page == NULL)
642 goto out_freepages;
643 pages[i] = page;
644 }
645 return 0;
646
647out_freepages:
648 nfs_readdir_free_pages(pages, i);
649 return -ENOMEM;
650}
651
652static
653int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
654{
655 struct page *pages[NFS_MAX_READDIR_PAGES];
656 struct nfs_entry entry;
657 struct file *file = desc->file;
658 struct nfs_cache_array *array;
659 int status = -ENOMEM;
660 unsigned int array_size = ARRAY_SIZE(pages);
661
662 entry.prev_cookie = 0;
663 entry.cookie = desc->last_cookie;
664 entry.eof = 0;
665 entry.fh = nfs_alloc_fhandle();
666 entry.fattr = nfs_alloc_fattr();
667 entry.server = NFS_SERVER(inode);
668 if (entry.fh == NULL || entry.fattr == NULL)
669 goto out;
670
671 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
672 if (IS_ERR(entry.label)) {
673 status = PTR_ERR(entry.label);
674 goto out;
675 }
676
677 array = nfs_readdir_get_array(page);
678 if (IS_ERR(array)) {
679 status = PTR_ERR(array);
680 goto out_label_free;
681 }
682 memset(array, 0, sizeof(struct nfs_cache_array));
683 atomic_set(&array->refcount, 1);
684 array->eof_index = -1;
685
686 status = nfs_readdir_alloc_pages(pages, array_size);
687 if (status < 0)
688 goto out_release_array;
689 do {
690 unsigned int pglen;
691 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
692
693 if (status < 0)
694 break;
695 pglen = status;
696 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
697 if (status < 0) {
698 if (status == -ENOSPC)
699 status = 0;
700 break;
701 }
702 } while (array->eof_index < 0);
703
704 nfs_readdir_free_pages(pages, array_size);
705out_release_array:
706 nfs_readdir_release_array(page);
707out_label_free:
708 nfs4_label_free(entry.label);
709out:
710 nfs_free_fattr(entry.fattr);
711 nfs_free_fhandle(entry.fh);
712 return status;
713}
714
715/*
716 * Now we cache directories properly, by converting xdr information
717 * to an array that can be used for lookups later. This results in
718 * fewer cache pages, since we can store more information on each page.
719 * We only need to convert from xdr once so future lookups are much simpler
720 */
721static
722int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
723{
724 struct inode *inode = file_inode(desc->file);
725 int ret;
726
727 ret = nfs_readdir_xdr_to_array(desc, page, inode);
728 if (ret < 0)
729 goto error;
730 SetPageUptodate(page);
731
732 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
733 /* Should never happen */
734 nfs_zap_mapping(inode, inode->i_mapping);
735 }
736 unlock_page(page);
737 return 0;
738 error:
739 unlock_page(page);
740 return ret;
741}
742
743static
744void cache_page_release(nfs_readdir_descriptor_t *desc)
745{
746 nfs_readdir_clear_array(desc->page);
747 put_page(desc->page);
748 desc->page = NULL;
749}
750
751static
752struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
753{
754 struct page *page;
755
756 for (;;) {
757 page = read_cache_page(desc->file->f_mapping,
758 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
759 if (IS_ERR(page) || grab_page(page))
760 break;
761 put_page(page);
762 }
763 return page;
764}
765
766/*
767 * Returns 0 if desc->dir_cookie was found on page desc->page_index
768 */
769static
770int find_cache_page(nfs_readdir_descriptor_t *desc)
771{
772 int res;
773
774 desc->page = get_cache_page(desc);
775 if (IS_ERR(desc->page))
776 return PTR_ERR(desc->page);
777
778 res = nfs_readdir_search_array(desc);
779 if (res != 0)
780 cache_page_release(desc);
781 return res;
782}
783
784/* Search for desc->dir_cookie from the beginning of the page cache */
785static inline
786int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
787{
788 int res;
789
790 if (desc->page_index == 0) {
791 desc->current_index = 0;
792 desc->last_cookie = 0;
793 }
794 do {
795 res = find_cache_page(desc);
796 } while (res == -EAGAIN);
797 return res;
798}
799
800/*
801 * Once we've found the start of the dirent within a page: fill 'er up...
802 */
803static
804int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
805{
806 struct file *file = desc->file;
807 int i = 0;
808 int res = 0;
809 struct nfs_cache_array *array = NULL;
810 struct nfs_open_dir_context *ctx = file->private_data;
811
812 array = nfs_readdir_get_array(desc->page);
813 if (IS_ERR(array)) {
814 res = PTR_ERR(array);
815 goto out;
816 }
817
818 for (i = desc->cache_entry_index; i < array->size; i++) {
819 struct nfs_cache_array_entry *ent;
820
821 ent = &array->array[i];
822 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
823 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
824 desc->eof = 1;
825 break;
826 }
827 desc->ctx->pos++;
828 if (i < (array->size-1))
829 *desc->dir_cookie = array->array[i+1].cookie;
830 else
831 *desc->dir_cookie = array->last_cookie;
832 if (ctx->duped != 0)
833 ctx->duped = 1;
834 }
835 if (array->eof_index >= 0)
836 desc->eof = 1;
837
838 nfs_readdir_release_array(desc->page);
839out:
840 cache_page_release(desc);
841 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
842 (unsigned long long)*desc->dir_cookie, res);
843 return res;
844}
845
846/*
847 * If we cannot find a cookie in our cache, we suspect that this is
848 * because it points to a deleted file, so we ask the server to return
849 * whatever it thinks is the next entry. We then feed this to filldir.
850 * If all goes well, we should then be able to find our way round the
851 * cache on the next call to readdir_search_pagecache();
852 *
853 * NOTE: we cannot add the anonymous page to the pagecache because
854 * the data it contains might not be page aligned. Besides,
855 * we should already have a complete representation of the
856 * directory in the page cache by the time we get here.
857 */
858static inline
859int uncached_readdir(nfs_readdir_descriptor_t *desc)
860{
861 struct page *page = NULL;
862 int status;
863 struct inode *inode = file_inode(desc->file);
864 struct nfs_open_dir_context *ctx = desc->file->private_data;
865
866 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
867 (unsigned long long)*desc->dir_cookie);
868
869 page = alloc_page(GFP_HIGHUSER);
870 if (!page) {
871 status = -ENOMEM;
872 goto out;
873 }
874
875 desc->page_index = 0;
876 desc->last_cookie = *desc->dir_cookie;
877 desc->page = page;
878 ctx->duped = 0;
879
880 status = nfs_readdir_xdr_to_array(desc, page, inode);
881 if (status < 0)
882 goto out_release;
883
884 status = nfs_do_filldir(desc);
885
886 out:
887 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
888 __func__, status);
889 return status;
890 out_release:
891 cache_page_release(desc);
892 goto out;
893}
894
895/* The file offset position represents the dirent entry number. A
896 last cookie cache takes care of the common case of reading the
897 whole directory.
898 */
899static int nfs_readdir(struct file *file, struct dir_context *ctx)
900{
901 struct dentry *dentry = file_dentry(file);
902 struct inode *inode = d_inode(dentry);
903 nfs_readdir_descriptor_t my_desc,
904 *desc = &my_desc;
905 struct nfs_open_dir_context *dir_ctx = file->private_data;
906 int res = 0;
907
908 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
909 file, (long long)ctx->pos);
910 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
911
912 /*
913 * ctx->pos points to the dirent entry number.
914 * *desc->dir_cookie has the cookie for the next entry. We have
915 * to either find the entry with the appropriate number or
916 * revalidate the cookie.
917 */
918 memset(desc, 0, sizeof(*desc));
919
920 desc->file = file;
921 desc->ctx = ctx;
922 desc->dir_cookie = &dir_ctx->dir_cookie;
923 desc->decode = NFS_PROTO(inode)->decode_dirent;
924 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
925
926 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
927 res = nfs_revalidate_mapping(inode, file->f_mapping);
928 if (res < 0)
929 goto out;
930
931 do {
932 res = readdir_search_pagecache(desc);
933
934 if (res == -EBADCOOKIE) {
935 res = 0;
936 /* This means either end of directory */
937 if (*desc->dir_cookie && desc->eof == 0) {
938 /* Or that the server has 'lost' a cookie */
939 res = uncached_readdir(desc);
940 if (res == 0)
941 continue;
942 }
943 break;
944 }
945 if (res == -ETOOSMALL && desc->plus) {
946 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
947 nfs_zap_caches(inode);
948 desc->page_index = 0;
949 desc->plus = 0;
950 desc->eof = 0;
951 continue;
952 }
953 if (res < 0)
954 break;
955
956 res = nfs_do_filldir(desc);
957 if (res < 0)
958 break;
959 } while (!desc->eof);
960out:
961 if (res > 0)
962 res = 0;
963 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
964 return res;
965}
966
967static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
968{
969 struct nfs_open_dir_context *dir_ctx = filp->private_data;
970
971 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
972 filp, offset, whence);
973
974 switch (whence) {
975 case 1:
976 offset += filp->f_pos;
977 case 0:
978 if (offset >= 0)
979 break;
980 default:
981 return -EINVAL;
982 }
983 if (offset != filp->f_pos) {
984 filp->f_pos = offset;
985 dir_ctx->dir_cookie = 0;
986 dir_ctx->duped = 0;
987 }
988 return offset;
989}
990
991/*
992 * All directory operations under NFS are synchronous, so fsync()
993 * is a dummy operation.
994 */
995static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
996 int datasync)
997{
998 struct inode *inode = file_inode(filp);
999
1000 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1001
1002 inode_lock(inode);
1003 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
1004 inode_unlock(inode);
1005 return 0;
1006}
1007
1008/**
1009 * nfs_force_lookup_revalidate - Mark the directory as having changed
1010 * @dir - pointer to directory inode
1011 *
1012 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1013 * full lookup on all child dentries of 'dir' whenever a change occurs
1014 * on the server that might have invalidated our dcache.
1015 *
1016 * The caller should be holding dir->i_lock
1017 */
1018void nfs_force_lookup_revalidate(struct inode *dir)
1019{
1020 NFS_I(dir)->cache_change_attribute++;
1021}
1022EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1023
1024/*
1025 * A check for whether or not the parent directory has changed.
1026 * In the case it has, we assume that the dentries are untrustworthy
1027 * and may need to be looked up again.
1028 * If rcu_walk prevents us from performing a full check, return 0.
1029 */
1030static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1031 int rcu_walk)
1032{
1033 if (IS_ROOT(dentry))
1034 return 1;
1035 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1036 return 0;
1037 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1038 return 0;
1039 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1040 if (nfs_mapping_need_revalidate_inode(dir)) {
1041 if (rcu_walk)
1042 return 0;
1043 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1044 return 0;
1045 }
1046 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1047 return 0;
1048 return 1;
1049}
1050
1051/*
1052 * Use intent information to check whether or not we're going to do
1053 * an O_EXCL create using this path component.
1054 */
1055static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1056{
1057 if (NFS_PROTO(dir)->version == 2)
1058 return 0;
1059 return flags & LOOKUP_EXCL;
1060}
1061
1062/*
1063 * Inode and filehandle revalidation for lookups.
1064 *
1065 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1066 * or if the intent information indicates that we're about to open this
1067 * particular file and the "nocto" mount flag is not set.
1068 *
1069 */
1070static
1071int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1072{
1073 struct nfs_server *server = NFS_SERVER(inode);
1074 int ret;
1075
1076 if (IS_AUTOMOUNT(inode))
1077 return 0;
1078 /* VFS wants an on-the-wire revalidation */
1079 if (flags & LOOKUP_REVAL)
1080 goto out_force;
1081 /* This is an open(2) */
1082 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1083 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1084 goto out_force;
1085out:
1086 return (inode->i_nlink == 0) ? -ENOENT : 0;
1087out_force:
1088 if (flags & LOOKUP_RCU)
1089 return -ECHILD;
1090 ret = __nfs_revalidate_inode(server, inode);
1091 if (ret != 0)
1092 return ret;
1093 goto out;
1094}
1095
1096/*
1097 * We judge how long we want to trust negative
1098 * dentries by looking at the parent inode mtime.
1099 *
1100 * If parent mtime has changed, we revalidate, else we wait for a
1101 * period corresponding to the parent's attribute cache timeout value.
1102 *
1103 * If LOOKUP_RCU prevents us from performing a full check, return 1
1104 * suggesting a reval is needed.
1105 */
1106static inline
1107int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1108 unsigned int flags)
1109{
1110 /* Don't revalidate a negative dentry if we're creating a new file */
1111 if (flags & LOOKUP_CREATE)
1112 return 0;
1113 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1114 return 1;
1115 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1116}
1117
1118/*
1119 * This is called every time the dcache has a lookup hit,
1120 * and we should check whether we can really trust that
1121 * lookup.
1122 *
1123 * NOTE! The hit can be a negative hit too, don't assume
1124 * we have an inode!
1125 *
1126 * If the parent directory is seen to have changed, we throw out the
1127 * cached dentry and do a new lookup.
1128 */
1129static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1130{
1131 struct inode *dir;
1132 struct inode *inode;
1133 struct dentry *parent;
1134 struct nfs_fh *fhandle = NULL;
1135 struct nfs_fattr *fattr = NULL;
1136 struct nfs4_label *label = NULL;
1137 int error;
1138
1139 if (flags & LOOKUP_RCU) {
1140 parent = ACCESS_ONCE(dentry->d_parent);
1141 dir = d_inode_rcu(parent);
1142 if (!dir)
1143 return -ECHILD;
1144 } else {
1145 parent = dget_parent(dentry);
1146 dir = d_inode(parent);
1147 }
1148 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1149 inode = d_inode(dentry);
1150
1151 if (!inode) {
1152 if (nfs_neg_need_reval(dir, dentry, flags)) {
1153 if (flags & LOOKUP_RCU)
1154 return -ECHILD;
1155 goto out_bad;
1156 }
1157 goto out_valid;
1158 }
1159
1160 if (is_bad_inode(inode)) {
1161 if (flags & LOOKUP_RCU)
1162 return -ECHILD;
1163 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1164 __func__, dentry);
1165 goto out_bad;
1166 }
1167
1168 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1169 goto out_set_verifier;
1170
1171 /* Force a full look up iff the parent directory has changed */
1172 if (!nfs_is_exclusive_create(dir, flags) &&
1173 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1174
1175 if (nfs_lookup_verify_inode(inode, flags)) {
1176 if (flags & LOOKUP_RCU)
1177 return -ECHILD;
1178 goto out_zap_parent;
1179 }
1180 nfs_advise_use_readdirplus(dir);
1181 goto out_valid;
1182 }
1183
1184 if (flags & LOOKUP_RCU)
1185 return -ECHILD;
1186
1187 if (NFS_STALE(inode))
1188 goto out_bad;
1189
1190 error = -ENOMEM;
1191 fhandle = nfs_alloc_fhandle();
1192 fattr = nfs_alloc_fattr();
1193 if (fhandle == NULL || fattr == NULL)
1194 goto out_error;
1195
1196 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1197 if (IS_ERR(label))
1198 goto out_error;
1199
1200 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1201 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1202 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1203 if (error)
1204 goto out_bad;
1205 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1206 goto out_bad;
1207 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1208 goto out_bad;
1209
1210 nfs_setsecurity(inode, fattr, label);
1211
1212 nfs_free_fattr(fattr);
1213 nfs_free_fhandle(fhandle);
1214 nfs4_label_free(label);
1215
1216 /* set a readdirplus hint that we had a cache miss */
1217 nfs_force_use_readdirplus(dir);
1218
1219out_set_verifier:
1220 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1221 out_valid:
1222 if (flags & LOOKUP_RCU) {
1223 if (parent != ACCESS_ONCE(dentry->d_parent))
1224 return -ECHILD;
1225 } else
1226 dput(parent);
1227 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1228 __func__, dentry);
1229 return 1;
1230out_zap_parent:
1231 nfs_zap_caches(dir);
1232 out_bad:
1233 WARN_ON(flags & LOOKUP_RCU);
1234 nfs_free_fattr(fattr);
1235 nfs_free_fhandle(fhandle);
1236 nfs4_label_free(label);
1237 nfs_mark_for_revalidate(dir);
1238 if (inode && S_ISDIR(inode->i_mode)) {
1239 /* Purge readdir caches. */
1240 nfs_zap_caches(inode);
1241 /*
1242 * We can't d_drop the root of a disconnected tree:
1243 * its d_hash is on the s_anon list and d_drop() would hide
1244 * it from shrink_dcache_for_unmount(), leading to busy
1245 * inodes on unmount and further oopses.
1246 */
1247 if (IS_ROOT(dentry))
1248 goto out_valid;
1249 }
1250 dput(parent);
1251 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1252 __func__, dentry);
1253 return 0;
1254out_error:
1255 WARN_ON(flags & LOOKUP_RCU);
1256 nfs_free_fattr(fattr);
1257 nfs_free_fhandle(fhandle);
1258 nfs4_label_free(label);
1259 dput(parent);
1260 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1261 __func__, dentry, error);
1262 return error;
1263}
1264
1265/*
1266 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1267 * when we don't really care about the dentry name. This is called when a
1268 * pathwalk ends on a dentry that was not found via a normal lookup in the
1269 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1270 *
1271 * In this situation, we just want to verify that the inode itself is OK
1272 * since the dentry might have changed on the server.
1273 */
1274static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1275{
1276 struct inode *inode = d_inode(dentry);
1277 int error = 0;
1278
1279 /*
1280 * I believe we can only get a negative dentry here in the case of a
1281 * procfs-style symlink. Just assume it's correct for now, but we may
1282 * eventually need to do something more here.
1283 */
1284 if (!inode) {
1285 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1286 __func__, dentry);
1287 return 1;
1288 }
1289
1290 if (is_bad_inode(inode)) {
1291 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1292 __func__, dentry);
1293 return 0;
1294 }
1295
1296 if (nfs_mapping_need_revalidate_inode(inode))
1297 error = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1298 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1299 __func__, inode->i_ino, error ? "invalid" : "valid");
1300 return !error;
1301}
1302
1303/*
1304 * This is called from dput() when d_count is going to 0.
1305 */
1306static int nfs_dentry_delete(const struct dentry *dentry)
1307{
1308 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1309 dentry, dentry->d_flags);
1310
1311 /* Unhash any dentry with a stale inode */
1312 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1313 return 1;
1314
1315 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1316 /* Unhash it, so that ->d_iput() would be called */
1317 return 1;
1318 }
1319 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1320 /* Unhash it, so that ancestors of killed async unlink
1321 * files will be cleaned up during umount */
1322 return 1;
1323 }
1324 return 0;
1325
1326}
1327
1328/* Ensure that we revalidate inode->i_nlink */
1329static void nfs_drop_nlink(struct inode *inode)
1330{
1331 spin_lock(&inode->i_lock);
1332 /* drop the inode if we're reasonably sure this is the last link */
1333 if (inode->i_nlink == 1)
1334 clear_nlink(inode);
1335 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1336 spin_unlock(&inode->i_lock);
1337}
1338
1339/*
1340 * Called when the dentry loses inode.
1341 * We use it to clean up silly-renamed files.
1342 */
1343static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1344{
1345 if (S_ISDIR(inode->i_mode))
1346 /* drop any readdir cache as it could easily be old */
1347 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1348
1349 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1350 nfs_complete_unlink(dentry, inode);
1351 nfs_drop_nlink(inode);
1352 }
1353 iput(inode);
1354}
1355
1356static void nfs_d_release(struct dentry *dentry)
1357{
1358 /* free cached devname value, if it survived that far */
1359 if (unlikely(dentry->d_fsdata)) {
1360 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1361 WARN_ON(1);
1362 else
1363 kfree(dentry->d_fsdata);
1364 }
1365}
1366
1367const struct dentry_operations nfs_dentry_operations = {
1368 .d_revalidate = nfs_lookup_revalidate,
1369 .d_weak_revalidate = nfs_weak_revalidate,
1370 .d_delete = nfs_dentry_delete,
1371 .d_iput = nfs_dentry_iput,
1372 .d_automount = nfs_d_automount,
1373 .d_release = nfs_d_release,
1374};
1375EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1376
1377struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1378{
1379 struct dentry *res;
1380 struct inode *inode = NULL;
1381 struct nfs_fh *fhandle = NULL;
1382 struct nfs_fattr *fattr = NULL;
1383 struct nfs4_label *label = NULL;
1384 int error;
1385
1386 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1387 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1388
1389 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1390 return ERR_PTR(-ENAMETOOLONG);
1391
1392 /*
1393 * If we're doing an exclusive create, optimize away the lookup
1394 * but don't hash the dentry.
1395 */
1396 if (nfs_is_exclusive_create(dir, flags))
1397 return NULL;
1398
1399 res = ERR_PTR(-ENOMEM);
1400 fhandle = nfs_alloc_fhandle();
1401 fattr = nfs_alloc_fattr();
1402 if (fhandle == NULL || fattr == NULL)
1403 goto out;
1404
1405 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1406 if (IS_ERR(label))
1407 goto out;
1408
1409 trace_nfs_lookup_enter(dir, dentry, flags);
1410 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1411 if (error == -ENOENT)
1412 goto no_entry;
1413 if (error < 0) {
1414 res = ERR_PTR(error);
1415 goto out_label;
1416 }
1417 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1418 res = ERR_CAST(inode);
1419 if (IS_ERR(res))
1420 goto out_label;
1421
1422 /* Notify readdir to use READDIRPLUS */
1423 nfs_force_use_readdirplus(dir);
1424
1425no_entry:
1426 res = d_splice_alias(inode, dentry);
1427 if (res != NULL) {
1428 if (IS_ERR(res))
1429 goto out_label;
1430 dentry = res;
1431 }
1432 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1433out_label:
1434 trace_nfs_lookup_exit(dir, dentry, flags, error);
1435 nfs4_label_free(label);
1436out:
1437 nfs_free_fattr(fattr);
1438 nfs_free_fhandle(fhandle);
1439 return res;
1440}
1441EXPORT_SYMBOL_GPL(nfs_lookup);
1442
1443#if IS_ENABLED(CONFIG_NFS_V4)
1444static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1445
1446const struct dentry_operations nfs4_dentry_operations = {
1447 .d_revalidate = nfs4_lookup_revalidate,
1448 .d_delete = nfs_dentry_delete,
1449 .d_iput = nfs_dentry_iput,
1450 .d_automount = nfs_d_automount,
1451 .d_release = nfs_d_release,
1452};
1453EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1454
1455static fmode_t flags_to_mode(int flags)
1456{
1457 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1458 if ((flags & O_ACCMODE) != O_WRONLY)
1459 res |= FMODE_READ;
1460 if ((flags & O_ACCMODE) != O_RDONLY)
1461 res |= FMODE_WRITE;
1462 return res;
1463}
1464
1465static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1466{
1467 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1468}
1469
1470static int do_open(struct inode *inode, struct file *filp)
1471{
1472 nfs_fscache_open_file(inode, filp);
1473 return 0;
1474}
1475
1476static int nfs_finish_open(struct nfs_open_context *ctx,
1477 struct dentry *dentry,
1478 struct file *file, unsigned open_flags,
1479 int *opened)
1480{
1481 int err;
1482
1483 err = finish_open(file, dentry, do_open, opened);
1484 if (err)
1485 goto out;
1486 nfs_file_set_open_context(file, ctx);
1487
1488out:
1489 return err;
1490}
1491
1492int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1493 struct file *file, unsigned open_flags,
1494 umode_t mode, int *opened)
1495{
1496 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1497 struct nfs_open_context *ctx;
1498 struct dentry *res;
1499 struct iattr attr = { .ia_valid = ATTR_OPEN };
1500 struct inode *inode;
1501 unsigned int lookup_flags = 0;
1502 bool switched = false;
1503 int err;
1504
1505 /* Expect a negative dentry */
1506 BUG_ON(d_inode(dentry));
1507
1508 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1509 dir->i_sb->s_id, dir->i_ino, dentry);
1510
1511 err = nfs_check_flags(open_flags);
1512 if (err)
1513 return err;
1514
1515 /* NFS only supports OPEN on regular files */
1516 if ((open_flags & O_DIRECTORY)) {
1517 if (!d_in_lookup(dentry)) {
1518 /*
1519 * Hashed negative dentry with O_DIRECTORY: dentry was
1520 * revalidated and is fine, no need to perform lookup
1521 * again
1522 */
1523 return -ENOENT;
1524 }
1525 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1526 goto no_open;
1527 }
1528
1529 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1530 return -ENAMETOOLONG;
1531
1532 if (open_flags & O_CREAT) {
1533 struct nfs_server *server = NFS_SERVER(dir);
1534
1535 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1536 mode &= ~current_umask();
1537
1538 attr.ia_valid |= ATTR_MODE;
1539 attr.ia_mode = mode;
1540 }
1541 if (open_flags & O_TRUNC) {
1542 attr.ia_valid |= ATTR_SIZE;
1543 attr.ia_size = 0;
1544 }
1545
1546 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1547 d_drop(dentry);
1548 switched = true;
1549 dentry = d_alloc_parallel(dentry->d_parent,
1550 &dentry->d_name, &wq);
1551 if (IS_ERR(dentry))
1552 return PTR_ERR(dentry);
1553 if (unlikely(!d_in_lookup(dentry)))
1554 return finish_no_open(file, dentry);
1555 }
1556
1557 ctx = create_nfs_open_context(dentry, open_flags, file);
1558 err = PTR_ERR(ctx);
1559 if (IS_ERR(ctx))
1560 goto out;
1561
1562 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1563 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
1564 if (IS_ERR(inode)) {
1565 err = PTR_ERR(inode);
1566 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1567 put_nfs_open_context(ctx);
1568 d_drop(dentry);
1569 switch (err) {
1570 case -ENOENT:
1571 d_add(dentry, NULL);
1572 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1573 break;
1574 case -EISDIR:
1575 case -ENOTDIR:
1576 goto no_open;
1577 case -ELOOP:
1578 if (!(open_flags & O_NOFOLLOW))
1579 goto no_open;
1580 break;
1581 /* case -EINVAL: */
1582 default:
1583 break;
1584 }
1585 goto out;
1586 }
1587
1588 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1589 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1590 put_nfs_open_context(ctx);
1591out:
1592 if (unlikely(switched)) {
1593 d_lookup_done(dentry);
1594 dput(dentry);
1595 }
1596 return err;
1597
1598no_open:
1599 res = nfs_lookup(dir, dentry, lookup_flags);
1600 if (switched) {
1601 d_lookup_done(dentry);
1602 if (!res)
1603 res = dentry;
1604 else
1605 dput(dentry);
1606 }
1607 if (IS_ERR(res))
1608 return PTR_ERR(res);
1609 return finish_no_open(file, res);
1610}
1611EXPORT_SYMBOL_GPL(nfs_atomic_open);
1612
1613static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1614{
1615 struct inode *inode;
1616 int ret = 0;
1617
1618 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1619 goto no_open;
1620 if (d_mountpoint(dentry))
1621 goto no_open;
1622 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1623 goto no_open;
1624
1625 inode = d_inode(dentry);
1626
1627 /* We can't create new files in nfs_open_revalidate(), so we
1628 * optimize away revalidation of negative dentries.
1629 */
1630 if (inode == NULL) {
1631 struct dentry *parent;
1632 struct inode *dir;
1633
1634 if (flags & LOOKUP_RCU) {
1635 parent = ACCESS_ONCE(dentry->d_parent);
1636 dir = d_inode_rcu(parent);
1637 if (!dir)
1638 return -ECHILD;
1639 } else {
1640 parent = dget_parent(dentry);
1641 dir = d_inode(parent);
1642 }
1643 if (!nfs_neg_need_reval(dir, dentry, flags))
1644 ret = 1;
1645 else if (flags & LOOKUP_RCU)
1646 ret = -ECHILD;
1647 if (!(flags & LOOKUP_RCU))
1648 dput(parent);
1649 else if (parent != ACCESS_ONCE(dentry->d_parent))
1650 return -ECHILD;
1651 goto out;
1652 }
1653
1654 /* NFS only supports OPEN on regular files */
1655 if (!S_ISREG(inode->i_mode))
1656 goto no_open;
1657 /* We cannot do exclusive creation on a positive dentry */
1658 if (flags & LOOKUP_EXCL)
1659 goto no_open;
1660
1661 /* Let f_op->open() actually open (and revalidate) the file */
1662 ret = 1;
1663
1664out:
1665 return ret;
1666
1667no_open:
1668 return nfs_lookup_revalidate(dentry, flags);
1669}
1670
1671#endif /* CONFIG_NFSV4 */
1672
1673/*
1674 * Code common to create, mkdir, and mknod.
1675 */
1676int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1677 struct nfs_fattr *fattr,
1678 struct nfs4_label *label)
1679{
1680 struct dentry *parent = dget_parent(dentry);
1681 struct inode *dir = d_inode(parent);
1682 struct inode *inode;
1683 int error = -EACCES;
1684
1685 d_drop(dentry);
1686
1687 /* We may have been initialized further down */
1688 if (d_really_is_positive(dentry))
1689 goto out;
1690 if (fhandle->size == 0) {
1691 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1692 if (error)
1693 goto out_error;
1694 }
1695 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1696 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1697 struct nfs_server *server = NFS_SB(dentry->d_sb);
1698 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1699 if (error < 0)
1700 goto out_error;
1701 }
1702 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1703 error = PTR_ERR(inode);
1704 if (IS_ERR(inode))
1705 goto out_error;
1706 d_add(dentry, inode);
1707out:
1708 dput(parent);
1709 return 0;
1710out_error:
1711 nfs_mark_for_revalidate(dir);
1712 dput(parent);
1713 return error;
1714}
1715EXPORT_SYMBOL_GPL(nfs_instantiate);
1716
1717/*
1718 * Following a failed create operation, we drop the dentry rather
1719 * than retain a negative dentry. This avoids a problem in the event
1720 * that the operation succeeded on the server, but an error in the
1721 * reply path made it appear to have failed.
1722 */
1723int nfs_create(struct inode *dir, struct dentry *dentry,
1724 umode_t mode, bool excl)
1725{
1726 struct iattr attr;
1727 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1728 int error;
1729
1730 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1731 dir->i_sb->s_id, dir->i_ino, dentry);
1732
1733 attr.ia_mode = mode;
1734 attr.ia_valid = ATTR_MODE;
1735
1736 trace_nfs_create_enter(dir, dentry, open_flags);
1737 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1738 trace_nfs_create_exit(dir, dentry, open_flags, error);
1739 if (error != 0)
1740 goto out_err;
1741 return 0;
1742out_err:
1743 d_drop(dentry);
1744 return error;
1745}
1746EXPORT_SYMBOL_GPL(nfs_create);
1747
1748/*
1749 * See comments for nfs_proc_create regarding failed operations.
1750 */
1751int
1752nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1753{
1754 struct iattr attr;
1755 int status;
1756
1757 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1758 dir->i_sb->s_id, dir->i_ino, dentry);
1759
1760 attr.ia_mode = mode;
1761 attr.ia_valid = ATTR_MODE;
1762
1763 trace_nfs_mknod_enter(dir, dentry);
1764 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1765 trace_nfs_mknod_exit(dir, dentry, status);
1766 if (status != 0)
1767 goto out_err;
1768 return 0;
1769out_err:
1770 d_drop(dentry);
1771 return status;
1772}
1773EXPORT_SYMBOL_GPL(nfs_mknod);
1774
1775/*
1776 * See comments for nfs_proc_create regarding failed operations.
1777 */
1778int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1779{
1780 struct iattr attr;
1781 int error;
1782
1783 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1784 dir->i_sb->s_id, dir->i_ino, dentry);
1785
1786 attr.ia_valid = ATTR_MODE;
1787 attr.ia_mode = mode | S_IFDIR;
1788
1789 trace_nfs_mkdir_enter(dir, dentry);
1790 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1791 trace_nfs_mkdir_exit(dir, dentry, error);
1792 if (error != 0)
1793 goto out_err;
1794 return 0;
1795out_err:
1796 d_drop(dentry);
1797 return error;
1798}
1799EXPORT_SYMBOL_GPL(nfs_mkdir);
1800
1801static void nfs_dentry_handle_enoent(struct dentry *dentry)
1802{
1803 if (simple_positive(dentry))
1804 d_delete(dentry);
1805}
1806
1807int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1808{
1809 int error;
1810
1811 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1812 dir->i_sb->s_id, dir->i_ino, dentry);
1813
1814 trace_nfs_rmdir_enter(dir, dentry);
1815 if (d_really_is_positive(dentry)) {
1816 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1817 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1818 /* Ensure the VFS deletes this inode */
1819 switch (error) {
1820 case 0:
1821 clear_nlink(d_inode(dentry));
1822 break;
1823 case -ENOENT:
1824 nfs_dentry_handle_enoent(dentry);
1825 }
1826 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1827 } else
1828 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1829 trace_nfs_rmdir_exit(dir, dentry, error);
1830
1831 return error;
1832}
1833EXPORT_SYMBOL_GPL(nfs_rmdir);
1834
1835/*
1836 * Remove a file after making sure there are no pending writes,
1837 * and after checking that the file has only one user.
1838 *
1839 * We invalidate the attribute cache and free the inode prior to the operation
1840 * to avoid possible races if the server reuses the inode.
1841 */
1842static int nfs_safe_remove(struct dentry *dentry)
1843{
1844 struct inode *dir = d_inode(dentry->d_parent);
1845 struct inode *inode = d_inode(dentry);
1846 int error = -EBUSY;
1847
1848 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1849
1850 /* If the dentry was sillyrenamed, we simply call d_delete() */
1851 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1852 error = 0;
1853 goto out;
1854 }
1855
1856 trace_nfs_remove_enter(dir, dentry);
1857 if (inode != NULL) {
1858 NFS_PROTO(inode)->return_delegation(inode);
1859 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1860 if (error == 0)
1861 nfs_drop_nlink(inode);
1862 } else
1863 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1864 if (error == -ENOENT)
1865 nfs_dentry_handle_enoent(dentry);
1866 trace_nfs_remove_exit(dir, dentry, error);
1867out:
1868 return error;
1869}
1870
1871/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1872 * belongs to an active ".nfs..." file and we return -EBUSY.
1873 *
1874 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1875 */
1876int nfs_unlink(struct inode *dir, struct dentry *dentry)
1877{
1878 int error;
1879 int need_rehash = 0;
1880
1881 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1882 dir->i_ino, dentry);
1883
1884 trace_nfs_unlink_enter(dir, dentry);
1885 spin_lock(&dentry->d_lock);
1886 if (d_count(dentry) > 1) {
1887 spin_unlock(&dentry->d_lock);
1888 /* Start asynchronous writeout of the inode */
1889 write_inode_now(d_inode(dentry), 0);
1890 error = nfs_sillyrename(dir, dentry);
1891 goto out;
1892 }
1893 if (!d_unhashed(dentry)) {
1894 __d_drop(dentry);
1895 need_rehash = 1;
1896 }
1897 spin_unlock(&dentry->d_lock);
1898 error = nfs_safe_remove(dentry);
1899 if (!error || error == -ENOENT) {
1900 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1901 } else if (need_rehash)
1902 d_rehash(dentry);
1903out:
1904 trace_nfs_unlink_exit(dir, dentry, error);
1905 return error;
1906}
1907EXPORT_SYMBOL_GPL(nfs_unlink);
1908
1909/*
1910 * To create a symbolic link, most file systems instantiate a new inode,
1911 * add a page to it containing the path, then write it out to the disk
1912 * using prepare_write/commit_write.
1913 *
1914 * Unfortunately the NFS client can't create the in-core inode first
1915 * because it needs a file handle to create an in-core inode (see
1916 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1917 * symlink request has completed on the server.
1918 *
1919 * So instead we allocate a raw page, copy the symname into it, then do
1920 * the SYMLINK request with the page as the buffer. If it succeeds, we
1921 * now have a new file handle and can instantiate an in-core NFS inode
1922 * and move the raw page into its mapping.
1923 */
1924int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1925{
1926 struct page *page;
1927 char *kaddr;
1928 struct iattr attr;
1929 unsigned int pathlen = strlen(symname);
1930 int error;
1931
1932 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1933 dir->i_ino, dentry, symname);
1934
1935 if (pathlen > PAGE_SIZE)
1936 return -ENAMETOOLONG;
1937
1938 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1939 attr.ia_valid = ATTR_MODE;
1940
1941 page = alloc_page(GFP_USER);
1942 if (!page)
1943 return -ENOMEM;
1944
1945 kaddr = page_address(page);
1946 memcpy(kaddr, symname, pathlen);
1947 if (pathlen < PAGE_SIZE)
1948 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1949
1950 trace_nfs_symlink_enter(dir, dentry);
1951 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1952 trace_nfs_symlink_exit(dir, dentry, error);
1953 if (error != 0) {
1954 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1955 dir->i_sb->s_id, dir->i_ino,
1956 dentry, symname, error);
1957 d_drop(dentry);
1958 __free_page(page);
1959 return error;
1960 }
1961
1962 /*
1963 * No big deal if we can't add this page to the page cache here.
1964 * READLINK will get the missing page from the server if needed.
1965 */
1966 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
1967 GFP_KERNEL)) {
1968 SetPageUptodate(page);
1969 unlock_page(page);
1970 /*
1971 * add_to_page_cache_lru() grabs an extra page refcount.
1972 * Drop it here to avoid leaking this page later.
1973 */
1974 put_page(page);
1975 } else
1976 __free_page(page);
1977
1978 return 0;
1979}
1980EXPORT_SYMBOL_GPL(nfs_symlink);
1981
1982int
1983nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1984{
1985 struct inode *inode = d_inode(old_dentry);
1986 int error;
1987
1988 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1989 old_dentry, dentry);
1990
1991 trace_nfs_link_enter(inode, dir, dentry);
1992 NFS_PROTO(inode)->return_delegation(inode);
1993
1994 d_drop(dentry);
1995 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1996 if (error == 0) {
1997 ihold(inode);
1998 d_add(dentry, inode);
1999 }
2000 trace_nfs_link_exit(inode, dir, dentry, error);
2001 return error;
2002}
2003EXPORT_SYMBOL_GPL(nfs_link);
2004
2005/*
2006 * RENAME
2007 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2008 * different file handle for the same inode after a rename (e.g. when
2009 * moving to a different directory). A fail-safe method to do so would
2010 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2011 * rename the old file using the sillyrename stuff. This way, the original
2012 * file in old_dir will go away when the last process iput()s the inode.
2013 *
2014 * FIXED.
2015 *
2016 * It actually works quite well. One needs to have the possibility for
2017 * at least one ".nfs..." file in each directory the file ever gets
2018 * moved or linked to which happens automagically with the new
2019 * implementation that only depends on the dcache stuff instead of
2020 * using the inode layer
2021 *
2022 * Unfortunately, things are a little more complicated than indicated
2023 * above. For a cross-directory move, we want to make sure we can get
2024 * rid of the old inode after the operation. This means there must be
2025 * no pending writes (if it's a file), and the use count must be 1.
2026 * If these conditions are met, we can drop the dentries before doing
2027 * the rename.
2028 */
2029int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2030 struct inode *new_dir, struct dentry *new_dentry,
2031 unsigned int flags)
2032{
2033 struct inode *old_inode = d_inode(old_dentry);
2034 struct inode *new_inode = d_inode(new_dentry);
2035 struct dentry *dentry = NULL, *rehash = NULL;
2036 struct rpc_task *task;
2037 int error = -EBUSY;
2038
2039 if (flags)
2040 return -EINVAL;
2041
2042 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2043 old_dentry, new_dentry,
2044 d_count(new_dentry));
2045
2046 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2047 /*
2048 * For non-directories, check whether the target is busy and if so,
2049 * make a copy of the dentry and then do a silly-rename. If the
2050 * silly-rename succeeds, the copied dentry is hashed and becomes
2051 * the new target.
2052 */
2053 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2054 /*
2055 * To prevent any new references to the target during the
2056 * rename, we unhash the dentry in advance.
2057 */
2058 if (!d_unhashed(new_dentry)) {
2059 d_drop(new_dentry);
2060 rehash = new_dentry;
2061 }
2062
2063 if (d_count(new_dentry) > 2) {
2064 int err;
2065
2066 /* copy the target dentry's name */
2067 dentry = d_alloc(new_dentry->d_parent,
2068 &new_dentry->d_name);
2069 if (!dentry)
2070 goto out;
2071
2072 /* silly-rename the existing target ... */
2073 err = nfs_sillyrename(new_dir, new_dentry);
2074 if (err)
2075 goto out;
2076
2077 new_dentry = dentry;
2078 rehash = NULL;
2079 new_inode = NULL;
2080 }
2081 }
2082
2083 NFS_PROTO(old_inode)->return_delegation(old_inode);
2084 if (new_inode != NULL)
2085 NFS_PROTO(new_inode)->return_delegation(new_inode);
2086
2087 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2088 if (IS_ERR(task)) {
2089 error = PTR_ERR(task);
2090 goto out;
2091 }
2092
2093 error = rpc_wait_for_completion_task(task);
2094 if (error == 0)
2095 error = task->tk_status;
2096 rpc_put_task(task);
2097 nfs_mark_for_revalidate(old_inode);
2098out:
2099 if (rehash)
2100 d_rehash(rehash);
2101 trace_nfs_rename_exit(old_dir, old_dentry,
2102 new_dir, new_dentry, error);
2103 if (!error) {
2104 if (new_inode != NULL)
2105 nfs_drop_nlink(new_inode);
2106 d_move(old_dentry, new_dentry);
2107 nfs_set_verifier(new_dentry,
2108 nfs_save_change_attribute(new_dir));
2109 } else if (error == -ENOENT)
2110 nfs_dentry_handle_enoent(old_dentry);
2111
2112 /* new dentry created? */
2113 if (dentry)
2114 dput(dentry);
2115 return error;
2116}
2117EXPORT_SYMBOL_GPL(nfs_rename);
2118
2119static DEFINE_SPINLOCK(nfs_access_lru_lock);
2120static LIST_HEAD(nfs_access_lru_list);
2121static atomic_long_t nfs_access_nr_entries;
2122
2123static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2124module_param(nfs_access_max_cachesize, ulong, 0644);
2125MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2126
2127static void nfs_access_free_entry(struct nfs_access_entry *entry)
2128{
2129 put_rpccred(entry->cred);
2130 kfree_rcu(entry, rcu_head);
2131 smp_mb__before_atomic();
2132 atomic_long_dec(&nfs_access_nr_entries);
2133 smp_mb__after_atomic();
2134}
2135
2136static void nfs_access_free_list(struct list_head *head)
2137{
2138 struct nfs_access_entry *cache;
2139
2140 while (!list_empty(head)) {
2141 cache = list_entry(head->next, struct nfs_access_entry, lru);
2142 list_del(&cache->lru);
2143 nfs_access_free_entry(cache);
2144 }
2145}
2146
2147static unsigned long
2148nfs_do_access_cache_scan(unsigned int nr_to_scan)
2149{
2150 LIST_HEAD(head);
2151 struct nfs_inode *nfsi, *next;
2152 struct nfs_access_entry *cache;
2153 long freed = 0;
2154
2155 spin_lock(&nfs_access_lru_lock);
2156 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2157 struct inode *inode;
2158
2159 if (nr_to_scan-- == 0)
2160 break;
2161 inode = &nfsi->vfs_inode;
2162 spin_lock(&inode->i_lock);
2163 if (list_empty(&nfsi->access_cache_entry_lru))
2164 goto remove_lru_entry;
2165 cache = list_entry(nfsi->access_cache_entry_lru.next,
2166 struct nfs_access_entry, lru);
2167 list_move(&cache->lru, &head);
2168 rb_erase(&cache->rb_node, &nfsi->access_cache);
2169 freed++;
2170 if (!list_empty(&nfsi->access_cache_entry_lru))
2171 list_move_tail(&nfsi->access_cache_inode_lru,
2172 &nfs_access_lru_list);
2173 else {
2174remove_lru_entry:
2175 list_del_init(&nfsi->access_cache_inode_lru);
2176 smp_mb__before_atomic();
2177 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2178 smp_mb__after_atomic();
2179 }
2180 spin_unlock(&inode->i_lock);
2181 }
2182 spin_unlock(&nfs_access_lru_lock);
2183 nfs_access_free_list(&head);
2184 return freed;
2185}
2186
2187unsigned long
2188nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2189{
2190 int nr_to_scan = sc->nr_to_scan;
2191 gfp_t gfp_mask = sc->gfp_mask;
2192
2193 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2194 return SHRINK_STOP;
2195 return nfs_do_access_cache_scan(nr_to_scan);
2196}
2197
2198
2199unsigned long
2200nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2201{
2202 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2203}
2204
2205static void
2206nfs_access_cache_enforce_limit(void)
2207{
2208 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2209 unsigned long diff;
2210 unsigned int nr_to_scan;
2211
2212 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2213 return;
2214 nr_to_scan = 100;
2215 diff = nr_entries - nfs_access_max_cachesize;
2216 if (diff < nr_to_scan)
2217 nr_to_scan = diff;
2218 nfs_do_access_cache_scan(nr_to_scan);
2219}
2220
2221static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2222{
2223 struct rb_root *root_node = &nfsi->access_cache;
2224 struct rb_node *n;
2225 struct nfs_access_entry *entry;
2226
2227 /* Unhook entries from the cache */
2228 while ((n = rb_first(root_node)) != NULL) {
2229 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2230 rb_erase(n, root_node);
2231 list_move(&entry->lru, head);
2232 }
2233 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2234}
2235
2236void nfs_access_zap_cache(struct inode *inode)
2237{
2238 LIST_HEAD(head);
2239
2240 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2241 return;
2242 /* Remove from global LRU init */
2243 spin_lock(&nfs_access_lru_lock);
2244 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2245 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2246
2247 spin_lock(&inode->i_lock);
2248 __nfs_access_zap_cache(NFS_I(inode), &head);
2249 spin_unlock(&inode->i_lock);
2250 spin_unlock(&nfs_access_lru_lock);
2251 nfs_access_free_list(&head);
2252}
2253EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2254
2255static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2256{
2257 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2258 struct nfs_access_entry *entry;
2259
2260 while (n != NULL) {
2261 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2262
2263 if (cred < entry->cred)
2264 n = n->rb_left;
2265 else if (cred > entry->cred)
2266 n = n->rb_right;
2267 else
2268 return entry;
2269 }
2270 return NULL;
2271}
2272
2273static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res, bool may_block)
2274{
2275 struct nfs_inode *nfsi = NFS_I(inode);
2276 struct nfs_access_entry *cache;
2277 bool retry = true;
2278 int err;
2279
2280 spin_lock(&inode->i_lock);
2281 for(;;) {
2282 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2283 goto out_zap;
2284 cache = nfs_access_search_rbtree(inode, cred);
2285 err = -ENOENT;
2286 if (cache == NULL)
2287 goto out;
2288 /* Found an entry, is our attribute cache valid? */
2289 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2290 break;
2291 err = -ECHILD;
2292 if (!may_block)
2293 goto out;
2294 if (!retry)
2295 goto out_zap;
2296 spin_unlock(&inode->i_lock);
2297 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2298 if (err)
2299 return err;
2300 spin_lock(&inode->i_lock);
2301 retry = false;
2302 }
2303 res->jiffies = cache->jiffies;
2304 res->cred = cache->cred;
2305 res->mask = cache->mask;
2306 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2307 err = 0;
2308out:
2309 spin_unlock(&inode->i_lock);
2310 return err;
2311out_zap:
2312 spin_unlock(&inode->i_lock);
2313 nfs_access_zap_cache(inode);
2314 return -ENOENT;
2315}
2316
2317static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2318{
2319 /* Only check the most recently returned cache entry,
2320 * but do it without locking.
2321 */
2322 struct nfs_inode *nfsi = NFS_I(inode);
2323 struct nfs_access_entry *cache;
2324 int err = -ECHILD;
2325 struct list_head *lh;
2326
2327 rcu_read_lock();
2328 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2329 goto out;
2330 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2331 cache = list_entry(lh, struct nfs_access_entry, lru);
2332 if (lh == &nfsi->access_cache_entry_lru ||
2333 cred != cache->cred)
2334 cache = NULL;
2335 if (cache == NULL)
2336 goto out;
2337 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2338 goto out;
2339 res->jiffies = cache->jiffies;
2340 res->cred = cache->cred;
2341 res->mask = cache->mask;
2342 err = 0;
2343out:
2344 rcu_read_unlock();
2345 return err;
2346}
2347
2348static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2349{
2350 struct nfs_inode *nfsi = NFS_I(inode);
2351 struct rb_root *root_node = &nfsi->access_cache;
2352 struct rb_node **p = &root_node->rb_node;
2353 struct rb_node *parent = NULL;
2354 struct nfs_access_entry *entry;
2355
2356 spin_lock(&inode->i_lock);
2357 while (*p != NULL) {
2358 parent = *p;
2359 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2360
2361 if (set->cred < entry->cred)
2362 p = &parent->rb_left;
2363 else if (set->cred > entry->cred)
2364 p = &parent->rb_right;
2365 else
2366 goto found;
2367 }
2368 rb_link_node(&set->rb_node, parent, p);
2369 rb_insert_color(&set->rb_node, root_node);
2370 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2371 spin_unlock(&inode->i_lock);
2372 return;
2373found:
2374 rb_replace_node(parent, &set->rb_node, root_node);
2375 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2376 list_del(&entry->lru);
2377 spin_unlock(&inode->i_lock);
2378 nfs_access_free_entry(entry);
2379}
2380
2381void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2382{
2383 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2384 if (cache == NULL)
2385 return;
2386 RB_CLEAR_NODE(&cache->rb_node);
2387 cache->jiffies = set->jiffies;
2388 cache->cred = get_rpccred(set->cred);
2389 cache->mask = set->mask;
2390
2391 /* The above field assignments must be visible
2392 * before this item appears on the lru. We cannot easily
2393 * use rcu_assign_pointer, so just force the memory barrier.
2394 */
2395 smp_wmb();
2396 nfs_access_add_rbtree(inode, cache);
2397
2398 /* Update accounting */
2399 smp_mb__before_atomic();
2400 atomic_long_inc(&nfs_access_nr_entries);
2401 smp_mb__after_atomic();
2402
2403 /* Add inode to global LRU list */
2404 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2405 spin_lock(&nfs_access_lru_lock);
2406 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2407 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2408 &nfs_access_lru_list);
2409 spin_unlock(&nfs_access_lru_lock);
2410 }
2411 nfs_access_cache_enforce_limit();
2412}
2413EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2414
2415void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2416{
2417 entry->mask = 0;
2418 if (access_result & NFS4_ACCESS_READ)
2419 entry->mask |= MAY_READ;
2420 if (access_result &
2421 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2422 entry->mask |= MAY_WRITE;
2423 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2424 entry->mask |= MAY_EXEC;
2425}
2426EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2427
2428static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2429{
2430 struct nfs_access_entry cache;
2431 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2432 int status;
2433
2434 trace_nfs_access_enter(inode);
2435
2436 status = nfs_access_get_cached_rcu(inode, cred, &cache);
2437 if (status != 0)
2438 status = nfs_access_get_cached(inode, cred, &cache, may_block);
2439 if (status == 0)
2440 goto out_cached;
2441
2442 status = -ECHILD;
2443 if (!may_block)
2444 goto out;
2445
2446 /* Be clever: ask server to check for all possible rights */
2447 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2448 cache.cred = cred;
2449 cache.jiffies = jiffies;
2450 status = NFS_PROTO(inode)->access(inode, &cache);
2451 if (status != 0) {
2452 if (status == -ESTALE) {
2453 nfs_zap_caches(inode);
2454 if (!S_ISDIR(inode->i_mode))
2455 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2456 }
2457 goto out;
2458 }
2459 nfs_access_add_cache(inode, &cache);
2460out_cached:
2461 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2462 status = -EACCES;
2463out:
2464 trace_nfs_access_exit(inode, status);
2465 return status;
2466}
2467
2468static int nfs_open_permission_mask(int openflags)
2469{
2470 int mask = 0;
2471
2472 if (openflags & __FMODE_EXEC) {
2473 /* ONLY check exec rights */
2474 mask = MAY_EXEC;
2475 } else {
2476 if ((openflags & O_ACCMODE) != O_WRONLY)
2477 mask |= MAY_READ;
2478 if ((openflags & O_ACCMODE) != O_RDONLY)
2479 mask |= MAY_WRITE;
2480 }
2481
2482 return mask;
2483}
2484
2485int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2486{
2487 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2488}
2489EXPORT_SYMBOL_GPL(nfs_may_open);
2490
2491static int nfs_execute_ok(struct inode *inode, int mask)
2492{
2493 struct nfs_server *server = NFS_SERVER(inode);
2494 int ret = 0;
2495
2496 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) {
2497 if (mask & MAY_NOT_BLOCK)
2498 return -ECHILD;
2499 ret = __nfs_revalidate_inode(server, inode);
2500 }
2501 if (ret == 0 && !execute_ok(inode))
2502 ret = -EACCES;
2503 return ret;
2504}
2505
2506int nfs_permission(struct inode *inode, int mask)
2507{
2508 struct rpc_cred *cred;
2509 int res = 0;
2510
2511 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2512
2513 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2514 goto out;
2515 /* Is this sys_access() ? */
2516 if (mask & (MAY_ACCESS | MAY_CHDIR))
2517 goto force_lookup;
2518
2519 switch (inode->i_mode & S_IFMT) {
2520 case S_IFLNK:
2521 goto out;
2522 case S_IFREG:
2523 if ((mask & MAY_OPEN) &&
2524 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2525 return 0;
2526 break;
2527 case S_IFDIR:
2528 /*
2529 * Optimize away all write operations, since the server
2530 * will check permissions when we perform the op.
2531 */
2532 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2533 goto out;
2534 }
2535
2536force_lookup:
2537 if (!NFS_PROTO(inode)->access)
2538 goto out_notsup;
2539
2540 /* Always try fast lookups first */
2541 rcu_read_lock();
2542 cred = rpc_lookup_cred_nonblock();
2543 if (!IS_ERR(cred))
2544 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2545 else
2546 res = PTR_ERR(cred);
2547 rcu_read_unlock();
2548 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2549 /* Fast lookup failed, try the slow way */
2550 cred = rpc_lookup_cred();
2551 if (!IS_ERR(cred)) {
2552 res = nfs_do_access(inode, cred, mask);
2553 put_rpccred(cred);
2554 } else
2555 res = PTR_ERR(cred);
2556 }
2557out:
2558 if (!res && (mask & MAY_EXEC))
2559 res = nfs_execute_ok(inode, mask);
2560
2561 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2562 inode->i_sb->s_id, inode->i_ino, mask, res);
2563 return res;
2564out_notsup:
2565 if (mask & MAY_NOT_BLOCK)
2566 return -ECHILD;
2567
2568 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2569 if (res == 0)
2570 res = generic_permission(inode, mask);
2571 goto out;
2572}
2573EXPORT_SYMBOL_GPL(nfs_permission);
2574
2575/*
2576 * Local variables:
2577 * version-control: t
2578 * kept-new-versions: 5
2579 * End:
2580 */