xfs: fix dquot isolation hang
[linux-block.git] / fs / nfs / dir.c
... / ...
CommitLineData
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 = 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_open_dir_context *ctx;
73 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
74 if (ctx != NULL) {
75 ctx->duped = 0;
76 ctx->attr_gencount = NFS_I(dir)->attr_gencount;
77 ctx->dir_cookie = 0;
78 ctx->dup_cookie = 0;
79 ctx->cred = get_rpccred(cred);
80 return ctx;
81 }
82 return ERR_PTR(-ENOMEM);
83}
84
85static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
86{
87 put_rpccred(ctx->cred);
88 kfree(ctx);
89}
90
91/*
92 * Open file
93 */
94static int
95nfs_opendir(struct inode *inode, struct file *filp)
96{
97 int res = 0;
98 struct nfs_open_dir_context *ctx;
99 struct rpc_cred *cred;
100
101 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
102 filp->f_path.dentry->d_parent->d_name.name,
103 filp->f_path.dentry->d_name.name);
104
105 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
106
107 cred = rpc_lookup_cred();
108 if (IS_ERR(cred))
109 return PTR_ERR(cred);
110 ctx = alloc_nfs_open_dir_context(inode, cred);
111 if (IS_ERR(ctx)) {
112 res = PTR_ERR(ctx);
113 goto out;
114 }
115 filp->private_data = ctx;
116 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
117 /* This is a mountpoint, so d_revalidate will never
118 * have been called, so we need to refresh the
119 * inode (for close-open consistency) ourselves.
120 */
121 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
122 }
123out:
124 put_rpccred(cred);
125 return res;
126}
127
128static int
129nfs_closedir(struct inode *inode, struct file *filp)
130{
131 put_nfs_open_dir_context(filp->private_data);
132 return 0;
133}
134
135struct nfs_cache_array_entry {
136 u64 cookie;
137 u64 ino;
138 struct qstr string;
139 unsigned char d_type;
140};
141
142struct nfs_cache_array {
143 int size;
144 int eof_index;
145 u64 last_cookie;
146 struct nfs_cache_array_entry array[0];
147};
148
149typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
150typedef struct {
151 struct file *file;
152 struct page *page;
153 struct dir_context *ctx;
154 unsigned long page_index;
155 u64 *dir_cookie;
156 u64 last_cookie;
157 loff_t current_index;
158 decode_dirent_t decode;
159
160 unsigned long timestamp;
161 unsigned long gencount;
162 unsigned int cache_entry_index;
163 unsigned int plus:1;
164 unsigned int eof:1;
165} nfs_readdir_descriptor_t;
166
167/*
168 * The caller is responsible for calling nfs_readdir_release_array(page)
169 */
170static
171struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
172{
173 void *ptr;
174 if (page == NULL)
175 return ERR_PTR(-EIO);
176 ptr = kmap(page);
177 if (ptr == NULL)
178 return ERR_PTR(-ENOMEM);
179 return ptr;
180}
181
182static
183void nfs_readdir_release_array(struct page *page)
184{
185 kunmap(page);
186}
187
188/*
189 * we are freeing strings created by nfs_add_to_readdir_array()
190 */
191static
192void nfs_readdir_clear_array(struct page *page)
193{
194 struct nfs_cache_array *array;
195 int i;
196
197 array = kmap_atomic(page);
198 for (i = 0; i < array->size; i++)
199 kfree(array->array[i].string.name);
200 kunmap_atomic(array);
201}
202
203/*
204 * the caller is responsible for freeing qstr.name
205 * when called by nfs_readdir_add_to_array, the strings will be freed in
206 * nfs_clear_readdir_array()
207 */
208static
209int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
210{
211 string->len = len;
212 string->name = kmemdup(name, len, GFP_KERNEL);
213 if (string->name == NULL)
214 return -ENOMEM;
215 /*
216 * Avoid a kmemleak false positive. The pointer to the name is stored
217 * in a page cache page which kmemleak does not scan.
218 */
219 kmemleak_not_leak(string->name);
220 string->hash = full_name_hash(name, len);
221 return 0;
222}
223
224static
225int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
226{
227 struct nfs_cache_array *array = nfs_readdir_get_array(page);
228 struct nfs_cache_array_entry *cache_entry;
229 int ret;
230
231 if (IS_ERR(array))
232 return PTR_ERR(array);
233
234 cache_entry = &array->array[array->size];
235
236 /* Check that this entry lies within the page bounds */
237 ret = -ENOSPC;
238 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
239 goto out;
240
241 cache_entry->cookie = entry->prev_cookie;
242 cache_entry->ino = entry->ino;
243 cache_entry->d_type = entry->d_type;
244 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
245 if (ret)
246 goto out;
247 array->last_cookie = entry->cookie;
248 array->size++;
249 if (entry->eof != 0)
250 array->eof_index = array->size;
251out:
252 nfs_readdir_release_array(page);
253 return ret;
254}
255
256static
257int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
258{
259 loff_t diff = desc->ctx->pos - desc->current_index;
260 unsigned int index;
261
262 if (diff < 0)
263 goto out_eof;
264 if (diff >= array->size) {
265 if (array->eof_index >= 0)
266 goto out_eof;
267 return -EAGAIN;
268 }
269
270 index = (unsigned int)diff;
271 *desc->dir_cookie = array->array[index].cookie;
272 desc->cache_entry_index = index;
273 return 0;
274out_eof:
275 desc->eof = 1;
276 return -EBADCOOKIE;
277}
278
279static
280int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
281{
282 int i;
283 loff_t new_pos;
284 int status = -EAGAIN;
285
286 for (i = 0; i < array->size; i++) {
287 if (array->array[i].cookie == *desc->dir_cookie) {
288 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
289 struct nfs_open_dir_context *ctx = desc->file->private_data;
290
291 new_pos = desc->current_index + i;
292 if (ctx->attr_gencount != nfsi->attr_gencount
293 || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
294 ctx->duped = 0;
295 ctx->attr_gencount = nfsi->attr_gencount;
296 } else if (new_pos < desc->ctx->pos) {
297 if (ctx->duped > 0
298 && ctx->dup_cookie == *desc->dir_cookie) {
299 if (printk_ratelimit()) {
300 pr_notice("NFS: directory %s/%s contains a readdir loop."
301 "Please contact your server vendor. "
302 "The file: %s has duplicate cookie %llu\n",
303 desc->file->f_dentry->d_parent->d_name.name,
304 desc->file->f_dentry->d_name.name,
305 array->array[i].string.name,
306 *desc->dir_cookie);
307 }
308 status = -ELOOP;
309 goto out;
310 }
311 ctx->dup_cookie = *desc->dir_cookie;
312 ctx->duped = -1;
313 }
314 desc->ctx->pos = new_pos;
315 desc->cache_entry_index = i;
316 return 0;
317 }
318 }
319 if (array->eof_index >= 0) {
320 status = -EBADCOOKIE;
321 if (*desc->dir_cookie == array->last_cookie)
322 desc->eof = 1;
323 }
324out:
325 return status;
326}
327
328static
329int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
330{
331 struct nfs_cache_array *array;
332 int status;
333
334 array = nfs_readdir_get_array(desc->page);
335 if (IS_ERR(array)) {
336 status = PTR_ERR(array);
337 goto out;
338 }
339
340 if (*desc->dir_cookie == 0)
341 status = nfs_readdir_search_for_pos(array, desc);
342 else
343 status = nfs_readdir_search_for_cookie(array, desc);
344
345 if (status == -EAGAIN) {
346 desc->last_cookie = array->last_cookie;
347 desc->current_index += array->size;
348 desc->page_index++;
349 }
350 nfs_readdir_release_array(desc->page);
351out:
352 return status;
353}
354
355/* Fill a page with xdr information before transferring to the cache page */
356static
357int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
358 struct nfs_entry *entry, struct file *file, struct inode *inode)
359{
360 struct nfs_open_dir_context *ctx = file->private_data;
361 struct rpc_cred *cred = ctx->cred;
362 unsigned long timestamp, gencount;
363 int error;
364
365 again:
366 timestamp = jiffies;
367 gencount = nfs_inc_attr_generation_counter();
368 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
369 NFS_SERVER(inode)->dtsize, desc->plus);
370 if (error < 0) {
371 /* We requested READDIRPLUS, but the server doesn't grok it */
372 if (error == -ENOTSUPP && desc->plus) {
373 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
374 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
375 desc->plus = 0;
376 goto again;
377 }
378 goto error;
379 }
380 desc->timestamp = timestamp;
381 desc->gencount = gencount;
382error:
383 return error;
384}
385
386static int xdr_decode(nfs_readdir_descriptor_t *desc,
387 struct nfs_entry *entry, struct xdr_stream *xdr)
388{
389 int error;
390
391 error = desc->decode(xdr, entry, desc->plus);
392 if (error)
393 return error;
394 entry->fattr->time_start = desc->timestamp;
395 entry->fattr->gencount = desc->gencount;
396 return 0;
397}
398
399static
400int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
401{
402 if (dentry->d_inode == NULL)
403 goto different;
404 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
405 goto different;
406 return 1;
407different:
408 return 0;
409}
410
411static
412bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
413{
414 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
415 return false;
416 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
417 return true;
418 if (ctx->pos == 0)
419 return true;
420 return false;
421}
422
423/*
424 * This function is called by the lookup code to request the use of
425 * readdirplus to accelerate any future lookups in the same
426 * directory.
427 */
428static
429void nfs_advise_use_readdirplus(struct inode *dir)
430{
431 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
432}
433
434static
435void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
436{
437 struct qstr filename = QSTR_INIT(entry->name, entry->len);
438 struct dentry *dentry;
439 struct dentry *alias;
440 struct inode *dir = parent->d_inode;
441 struct inode *inode;
442 int status;
443
444 if (filename.name[0] == '.') {
445 if (filename.len == 1)
446 return;
447 if (filename.len == 2 && filename.name[1] == '.')
448 return;
449 }
450 filename.hash = full_name_hash(filename.name, filename.len);
451
452 dentry = d_lookup(parent, &filename);
453 if (dentry != NULL) {
454 if (nfs_same_file(dentry, entry)) {
455 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
456 status = nfs_refresh_inode(dentry->d_inode, entry->fattr);
457 if (!status)
458 nfs_setsecurity(dentry->d_inode, entry->fattr, entry->label);
459 goto out;
460 } else {
461 if (d_invalidate(dentry) != 0)
462 goto out;
463 dput(dentry);
464 }
465 }
466
467 dentry = d_alloc(parent, &filename);
468 if (dentry == NULL)
469 return;
470
471 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
472 if (IS_ERR(inode))
473 goto out;
474
475 alias = d_materialise_unique(dentry, inode);
476 if (IS_ERR(alias))
477 goto out;
478 else if (alias) {
479 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
480 dput(alias);
481 } else
482 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
483
484out:
485 dput(dentry);
486}
487
488/* Perform conversion from xdr to cache array */
489static
490int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
491 struct page **xdr_pages, struct page *page, unsigned int buflen)
492{
493 struct xdr_stream stream;
494 struct xdr_buf buf;
495 struct page *scratch;
496 struct nfs_cache_array *array;
497 unsigned int count = 0;
498 int status;
499
500 scratch = alloc_page(GFP_KERNEL);
501 if (scratch == NULL)
502 return -ENOMEM;
503
504 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
505 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
506
507 do {
508 status = xdr_decode(desc, entry, &stream);
509 if (status != 0) {
510 if (status == -EAGAIN)
511 status = 0;
512 break;
513 }
514
515 count++;
516
517 if (desc->plus != 0)
518 nfs_prime_dcache(desc->file->f_path.dentry, entry);
519
520 status = nfs_readdir_add_to_array(entry, page);
521 if (status != 0)
522 break;
523 } while (!entry->eof);
524
525 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
526 array = nfs_readdir_get_array(page);
527 if (!IS_ERR(array)) {
528 array->eof_index = array->size;
529 status = 0;
530 nfs_readdir_release_array(page);
531 } else
532 status = PTR_ERR(array);
533 }
534
535 put_page(scratch);
536 return status;
537}
538
539static
540void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
541{
542 unsigned int i;
543 for (i = 0; i < npages; i++)
544 put_page(pages[i]);
545}
546
547static
548void nfs_readdir_free_large_page(void *ptr, struct page **pages,
549 unsigned int npages)
550{
551 nfs_readdir_free_pagearray(pages, npages);
552}
553
554/*
555 * nfs_readdir_large_page will allocate pages that must be freed with a call
556 * to nfs_readdir_free_large_page
557 */
558static
559int nfs_readdir_large_page(struct page **pages, unsigned int npages)
560{
561 unsigned int i;
562
563 for (i = 0; i < npages; i++) {
564 struct page *page = alloc_page(GFP_KERNEL);
565 if (page == NULL)
566 goto out_freepages;
567 pages[i] = page;
568 }
569 return 0;
570
571out_freepages:
572 nfs_readdir_free_pagearray(pages, i);
573 return -ENOMEM;
574}
575
576static
577int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
578{
579 struct page *pages[NFS_MAX_READDIR_PAGES];
580 void *pages_ptr = NULL;
581 struct nfs_entry entry;
582 struct file *file = desc->file;
583 struct nfs_cache_array *array;
584 int status = -ENOMEM;
585 unsigned int array_size = ARRAY_SIZE(pages);
586
587 entry.prev_cookie = 0;
588 entry.cookie = desc->last_cookie;
589 entry.eof = 0;
590 entry.fh = nfs_alloc_fhandle();
591 entry.fattr = nfs_alloc_fattr();
592 entry.server = NFS_SERVER(inode);
593 if (entry.fh == NULL || entry.fattr == NULL)
594 goto out;
595
596 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
597 if (IS_ERR(entry.label)) {
598 status = PTR_ERR(entry.label);
599 goto out;
600 }
601
602 array = nfs_readdir_get_array(page);
603 if (IS_ERR(array)) {
604 status = PTR_ERR(array);
605 goto out_label_free;
606 }
607 memset(array, 0, sizeof(struct nfs_cache_array));
608 array->eof_index = -1;
609
610 status = nfs_readdir_large_page(pages, array_size);
611 if (status < 0)
612 goto out_release_array;
613 do {
614 unsigned int pglen;
615 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
616
617 if (status < 0)
618 break;
619 pglen = status;
620 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
621 if (status < 0) {
622 if (status == -ENOSPC)
623 status = 0;
624 break;
625 }
626 } while (array->eof_index < 0);
627
628 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
629out_release_array:
630 nfs_readdir_release_array(page);
631out_label_free:
632 nfs4_label_free(entry.label);
633out:
634 nfs_free_fattr(entry.fattr);
635 nfs_free_fhandle(entry.fh);
636 return status;
637}
638
639/*
640 * Now we cache directories properly, by converting xdr information
641 * to an array that can be used for lookups later. This results in
642 * fewer cache pages, since we can store more information on each page.
643 * We only need to convert from xdr once so future lookups are much simpler
644 */
645static
646int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
647{
648 struct inode *inode = file_inode(desc->file);
649 int ret;
650
651 ret = nfs_readdir_xdr_to_array(desc, page, inode);
652 if (ret < 0)
653 goto error;
654 SetPageUptodate(page);
655
656 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
657 /* Should never happen */
658 nfs_zap_mapping(inode, inode->i_mapping);
659 }
660 unlock_page(page);
661 return 0;
662 error:
663 unlock_page(page);
664 return ret;
665}
666
667static
668void cache_page_release(nfs_readdir_descriptor_t *desc)
669{
670 if (!desc->page->mapping)
671 nfs_readdir_clear_array(desc->page);
672 page_cache_release(desc->page);
673 desc->page = NULL;
674}
675
676static
677struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
678{
679 return read_cache_page(file_inode(desc->file)->i_mapping,
680 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
681}
682
683/*
684 * Returns 0 if desc->dir_cookie was found on page desc->page_index
685 */
686static
687int find_cache_page(nfs_readdir_descriptor_t *desc)
688{
689 int res;
690
691 desc->page = get_cache_page(desc);
692 if (IS_ERR(desc->page))
693 return PTR_ERR(desc->page);
694
695 res = nfs_readdir_search_array(desc);
696 if (res != 0)
697 cache_page_release(desc);
698 return res;
699}
700
701/* Search for desc->dir_cookie from the beginning of the page cache */
702static inline
703int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
704{
705 int res;
706
707 if (desc->page_index == 0) {
708 desc->current_index = 0;
709 desc->last_cookie = 0;
710 }
711 do {
712 res = find_cache_page(desc);
713 } while (res == -EAGAIN);
714 return res;
715}
716
717/*
718 * Once we've found the start of the dirent within a page: fill 'er up...
719 */
720static
721int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
722{
723 struct file *file = desc->file;
724 int i = 0;
725 int res = 0;
726 struct nfs_cache_array *array = NULL;
727 struct nfs_open_dir_context *ctx = file->private_data;
728
729 array = nfs_readdir_get_array(desc->page);
730 if (IS_ERR(array)) {
731 res = PTR_ERR(array);
732 goto out;
733 }
734
735 for (i = desc->cache_entry_index; i < array->size; i++) {
736 struct nfs_cache_array_entry *ent;
737
738 ent = &array->array[i];
739 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
740 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
741 desc->eof = 1;
742 break;
743 }
744 desc->ctx->pos++;
745 if (i < (array->size-1))
746 *desc->dir_cookie = array->array[i+1].cookie;
747 else
748 *desc->dir_cookie = array->last_cookie;
749 if (ctx->duped != 0)
750 ctx->duped = 1;
751 }
752 if (array->eof_index >= 0)
753 desc->eof = 1;
754
755 nfs_readdir_release_array(desc->page);
756out:
757 cache_page_release(desc);
758 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
759 (unsigned long long)*desc->dir_cookie, res);
760 return res;
761}
762
763/*
764 * If we cannot find a cookie in our cache, we suspect that this is
765 * because it points to a deleted file, so we ask the server to return
766 * whatever it thinks is the next entry. We then feed this to filldir.
767 * If all goes well, we should then be able to find our way round the
768 * cache on the next call to readdir_search_pagecache();
769 *
770 * NOTE: we cannot add the anonymous page to the pagecache because
771 * the data it contains might not be page aligned. Besides,
772 * we should already have a complete representation of the
773 * directory in the page cache by the time we get here.
774 */
775static inline
776int uncached_readdir(nfs_readdir_descriptor_t *desc)
777{
778 struct page *page = NULL;
779 int status;
780 struct inode *inode = file_inode(desc->file);
781 struct nfs_open_dir_context *ctx = desc->file->private_data;
782
783 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
784 (unsigned long long)*desc->dir_cookie);
785
786 page = alloc_page(GFP_HIGHUSER);
787 if (!page) {
788 status = -ENOMEM;
789 goto out;
790 }
791
792 desc->page_index = 0;
793 desc->last_cookie = *desc->dir_cookie;
794 desc->page = page;
795 ctx->duped = 0;
796
797 status = nfs_readdir_xdr_to_array(desc, page, inode);
798 if (status < 0)
799 goto out_release;
800
801 status = nfs_do_filldir(desc);
802
803 out:
804 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
805 __func__, status);
806 return status;
807 out_release:
808 cache_page_release(desc);
809 goto out;
810}
811
812/* The file offset position represents the dirent entry number. A
813 last cookie cache takes care of the common case of reading the
814 whole directory.
815 */
816static int nfs_readdir(struct file *file, struct dir_context *ctx)
817{
818 struct dentry *dentry = file->f_path.dentry;
819 struct inode *inode = dentry->d_inode;
820 nfs_readdir_descriptor_t my_desc,
821 *desc = &my_desc;
822 struct nfs_open_dir_context *dir_ctx = file->private_data;
823 int res = 0;
824
825 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
826 dentry->d_parent->d_name.name, dentry->d_name.name,
827 (long long)ctx->pos);
828 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
829
830 /*
831 * ctx->pos points to the dirent entry number.
832 * *desc->dir_cookie has the cookie for the next entry. We have
833 * to either find the entry with the appropriate number or
834 * revalidate the cookie.
835 */
836 memset(desc, 0, sizeof(*desc));
837
838 desc->file = file;
839 desc->ctx = ctx;
840 desc->dir_cookie = &dir_ctx->dir_cookie;
841 desc->decode = NFS_PROTO(inode)->decode_dirent;
842 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
843
844 nfs_block_sillyrename(dentry);
845 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
846 res = nfs_revalidate_mapping(inode, file->f_mapping);
847 if (res < 0)
848 goto out;
849
850 do {
851 res = readdir_search_pagecache(desc);
852
853 if (res == -EBADCOOKIE) {
854 res = 0;
855 /* This means either end of directory */
856 if (*desc->dir_cookie && desc->eof == 0) {
857 /* Or that the server has 'lost' a cookie */
858 res = uncached_readdir(desc);
859 if (res == 0)
860 continue;
861 }
862 break;
863 }
864 if (res == -ETOOSMALL && desc->plus) {
865 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
866 nfs_zap_caches(inode);
867 desc->page_index = 0;
868 desc->plus = 0;
869 desc->eof = 0;
870 continue;
871 }
872 if (res < 0)
873 break;
874
875 res = nfs_do_filldir(desc);
876 if (res < 0)
877 break;
878 } while (!desc->eof);
879out:
880 nfs_unblock_sillyrename(dentry);
881 if (res > 0)
882 res = 0;
883 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
884 dentry->d_parent->d_name.name, dentry->d_name.name,
885 res);
886 return res;
887}
888
889static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
890{
891 struct dentry *dentry = filp->f_path.dentry;
892 struct inode *inode = dentry->d_inode;
893 struct nfs_open_dir_context *dir_ctx = filp->private_data;
894
895 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
896 dentry->d_parent->d_name.name,
897 dentry->d_name.name,
898 offset, whence);
899
900 mutex_lock(&inode->i_mutex);
901 switch (whence) {
902 case 1:
903 offset += filp->f_pos;
904 case 0:
905 if (offset >= 0)
906 break;
907 default:
908 offset = -EINVAL;
909 goto out;
910 }
911 if (offset != filp->f_pos) {
912 filp->f_pos = offset;
913 dir_ctx->dir_cookie = 0;
914 dir_ctx->duped = 0;
915 }
916out:
917 mutex_unlock(&inode->i_mutex);
918 return offset;
919}
920
921/*
922 * All directory operations under NFS are synchronous, so fsync()
923 * is a dummy operation.
924 */
925static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
926 int datasync)
927{
928 struct dentry *dentry = filp->f_path.dentry;
929 struct inode *inode = dentry->d_inode;
930
931 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
932 dentry->d_parent->d_name.name, dentry->d_name.name,
933 datasync);
934
935 mutex_lock(&inode->i_mutex);
936 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
937 mutex_unlock(&inode->i_mutex);
938 return 0;
939}
940
941/**
942 * nfs_force_lookup_revalidate - Mark the directory as having changed
943 * @dir - pointer to directory inode
944 *
945 * This forces the revalidation code in nfs_lookup_revalidate() to do a
946 * full lookup on all child dentries of 'dir' whenever a change occurs
947 * on the server that might have invalidated our dcache.
948 *
949 * The caller should be holding dir->i_lock
950 */
951void nfs_force_lookup_revalidate(struct inode *dir)
952{
953 NFS_I(dir)->cache_change_attribute++;
954}
955EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
956
957/*
958 * A check for whether or not the parent directory has changed.
959 * In the case it has, we assume that the dentries are untrustworthy
960 * and may need to be looked up again.
961 */
962static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
963{
964 if (IS_ROOT(dentry))
965 return 1;
966 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
967 return 0;
968 if (!nfs_verify_change_attribute(dir, dentry->d_time))
969 return 0;
970 /* Revalidate nfsi->cache_change_attribute before we declare a match */
971 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
972 return 0;
973 if (!nfs_verify_change_attribute(dir, dentry->d_time))
974 return 0;
975 return 1;
976}
977
978/*
979 * Use intent information to check whether or not we're going to do
980 * an O_EXCL create using this path component.
981 */
982static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
983{
984 if (NFS_PROTO(dir)->version == 2)
985 return 0;
986 return flags & LOOKUP_EXCL;
987}
988
989/*
990 * Inode and filehandle revalidation for lookups.
991 *
992 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
993 * or if the intent information indicates that we're about to open this
994 * particular file and the "nocto" mount flag is not set.
995 *
996 */
997static
998int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
999{
1000 struct nfs_server *server = NFS_SERVER(inode);
1001 int ret;
1002
1003 if (IS_AUTOMOUNT(inode))
1004 return 0;
1005 /* VFS wants an on-the-wire revalidation */
1006 if (flags & LOOKUP_REVAL)
1007 goto out_force;
1008 /* This is an open(2) */
1009 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1010 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1011 goto out_force;
1012out:
1013 return (inode->i_nlink == 0) ? -ENOENT : 0;
1014out_force:
1015 ret = __nfs_revalidate_inode(server, inode);
1016 if (ret != 0)
1017 return ret;
1018 goto out;
1019}
1020
1021/*
1022 * We judge how long we want to trust negative
1023 * dentries by looking at the parent inode mtime.
1024 *
1025 * If parent mtime has changed, we revalidate, else we wait for a
1026 * period corresponding to the parent's attribute cache timeout value.
1027 */
1028static inline
1029int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1030 unsigned int flags)
1031{
1032 /* Don't revalidate a negative dentry if we're creating a new file */
1033 if (flags & LOOKUP_CREATE)
1034 return 0;
1035 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1036 return 1;
1037 return !nfs_check_verifier(dir, dentry);
1038}
1039
1040/*
1041 * This is called every time the dcache has a lookup hit,
1042 * and we should check whether we can really trust that
1043 * lookup.
1044 *
1045 * NOTE! The hit can be a negative hit too, don't assume
1046 * we have an inode!
1047 *
1048 * If the parent directory is seen to have changed, we throw out the
1049 * cached dentry and do a new lookup.
1050 */
1051static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1052{
1053 struct inode *dir;
1054 struct inode *inode;
1055 struct dentry *parent;
1056 struct nfs_fh *fhandle = NULL;
1057 struct nfs_fattr *fattr = NULL;
1058 struct nfs4_label *label = NULL;
1059 int error;
1060
1061 if (flags & LOOKUP_RCU)
1062 return -ECHILD;
1063
1064 parent = dget_parent(dentry);
1065 dir = parent->d_inode;
1066 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1067 inode = dentry->d_inode;
1068
1069 if (!inode) {
1070 if (nfs_neg_need_reval(dir, dentry, flags))
1071 goto out_bad;
1072 goto out_valid_noent;
1073 }
1074
1075 if (is_bad_inode(inode)) {
1076 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1077 __func__, dentry->d_parent->d_name.name,
1078 dentry->d_name.name);
1079 goto out_bad;
1080 }
1081
1082 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1083 goto out_set_verifier;
1084
1085 /* Force a full look up iff the parent directory has changed */
1086 if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1087 if (nfs_lookup_verify_inode(inode, flags))
1088 goto out_zap_parent;
1089 goto out_valid;
1090 }
1091
1092 if (NFS_STALE(inode))
1093 goto out_bad;
1094
1095 error = -ENOMEM;
1096 fhandle = nfs_alloc_fhandle();
1097 fattr = nfs_alloc_fattr();
1098 if (fhandle == NULL || fattr == NULL)
1099 goto out_error;
1100
1101 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1102 if (IS_ERR(label))
1103 goto out_error;
1104
1105 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1106 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1107 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1108 if (error)
1109 goto out_bad;
1110 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1111 goto out_bad;
1112 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1113 goto out_bad;
1114
1115 nfs_setsecurity(inode, fattr, label);
1116
1117 nfs_free_fattr(fattr);
1118 nfs_free_fhandle(fhandle);
1119 nfs4_label_free(label);
1120
1121out_set_verifier:
1122 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1123 out_valid:
1124 /* Success: notify readdir to use READDIRPLUS */
1125 nfs_advise_use_readdirplus(dir);
1126 out_valid_noent:
1127 dput(parent);
1128 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
1129 __func__, dentry->d_parent->d_name.name,
1130 dentry->d_name.name);
1131 return 1;
1132out_zap_parent:
1133 nfs_zap_caches(dir);
1134 out_bad:
1135 nfs_free_fattr(fattr);
1136 nfs_free_fhandle(fhandle);
1137 nfs4_label_free(label);
1138 nfs_mark_for_revalidate(dir);
1139 if (inode && S_ISDIR(inode->i_mode)) {
1140 /* Purge readdir caches. */
1141 nfs_zap_caches(inode);
1142 if (dentry->d_flags & DCACHE_DISCONNECTED)
1143 goto out_valid;
1144 }
1145 /* If we have submounts, don't unhash ! */
1146 if (check_submounts_and_drop(dentry) != 0)
1147 goto out_valid;
1148
1149 dput(parent);
1150 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
1151 __func__, dentry->d_parent->d_name.name,
1152 dentry->d_name.name);
1153 return 0;
1154out_error:
1155 nfs_free_fattr(fattr);
1156 nfs_free_fhandle(fhandle);
1157 nfs4_label_free(label);
1158 dput(parent);
1159 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1160 __func__, dentry->d_parent->d_name.name,
1161 dentry->d_name.name, error);
1162 return error;
1163}
1164
1165/*
1166 * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1167 * when we don't really care about the dentry name. This is called when a
1168 * pathwalk ends on a dentry that was not found via a normal lookup in the
1169 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1170 *
1171 * In this situation, we just want to verify that the inode itself is OK
1172 * since the dentry might have changed on the server.
1173 */
1174static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1175{
1176 int error;
1177 struct inode *inode = dentry->d_inode;
1178
1179 /*
1180 * I believe we can only get a negative dentry here in the case of a
1181 * procfs-style symlink. Just assume it's correct for now, but we may
1182 * eventually need to do something more here.
1183 */
1184 if (!inode) {
1185 dfprintk(LOOKUPCACHE, "%s: %s/%s has negative inode\n",
1186 __func__, dentry->d_parent->d_name.name,
1187 dentry->d_name.name);
1188 return 1;
1189 }
1190
1191 if (is_bad_inode(inode)) {
1192 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
1193 __func__, dentry->d_parent->d_name.name,
1194 dentry->d_name.name);
1195 return 0;
1196 }
1197
1198 error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1199 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1200 __func__, inode->i_ino, error ? "invalid" : "valid");
1201 return !error;
1202}
1203
1204/*
1205 * This is called from dput() when d_count is going to 0.
1206 */
1207static int nfs_dentry_delete(const struct dentry *dentry)
1208{
1209 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1210 dentry->d_parent->d_name.name, dentry->d_name.name,
1211 dentry->d_flags);
1212
1213 /* Unhash any dentry with a stale inode */
1214 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1215 return 1;
1216
1217 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1218 /* Unhash it, so that ->d_iput() would be called */
1219 return 1;
1220 }
1221 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1222 /* Unhash it, so that ancestors of killed async unlink
1223 * files will be cleaned up during umount */
1224 return 1;
1225 }
1226 return 0;
1227
1228}
1229
1230/* Ensure that we revalidate inode->i_nlink */
1231static void nfs_drop_nlink(struct inode *inode)
1232{
1233 spin_lock(&inode->i_lock);
1234 /* drop the inode if we're reasonably sure this is the last link */
1235 if (inode->i_nlink == 1)
1236 clear_nlink(inode);
1237 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1238 spin_unlock(&inode->i_lock);
1239}
1240
1241/*
1242 * Called when the dentry loses inode.
1243 * We use it to clean up silly-renamed files.
1244 */
1245static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1246{
1247 if (S_ISDIR(inode->i_mode))
1248 /* drop any readdir cache as it could easily be old */
1249 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1250
1251 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1252 nfs_complete_unlink(dentry, inode);
1253 nfs_drop_nlink(inode);
1254 }
1255 iput(inode);
1256}
1257
1258static void nfs_d_release(struct dentry *dentry)
1259{
1260 /* free cached devname value, if it survived that far */
1261 if (unlikely(dentry->d_fsdata)) {
1262 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1263 WARN_ON(1);
1264 else
1265 kfree(dentry->d_fsdata);
1266 }
1267}
1268
1269const struct dentry_operations nfs_dentry_operations = {
1270 .d_revalidate = nfs_lookup_revalidate,
1271 .d_weak_revalidate = nfs_weak_revalidate,
1272 .d_delete = nfs_dentry_delete,
1273 .d_iput = nfs_dentry_iput,
1274 .d_automount = nfs_d_automount,
1275 .d_release = nfs_d_release,
1276};
1277EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1278
1279struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1280{
1281 struct dentry *res;
1282 struct dentry *parent;
1283 struct inode *inode = NULL;
1284 struct nfs_fh *fhandle = NULL;
1285 struct nfs_fattr *fattr = NULL;
1286 struct nfs4_label *label = NULL;
1287 int error;
1288
1289 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1290 dentry->d_parent->d_name.name, dentry->d_name.name);
1291 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1292
1293 res = ERR_PTR(-ENAMETOOLONG);
1294 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1295 goto out;
1296
1297 /*
1298 * If we're doing an exclusive create, optimize away the lookup
1299 * but don't hash the dentry.
1300 */
1301 if (nfs_is_exclusive_create(dir, flags)) {
1302 d_instantiate(dentry, NULL);
1303 res = NULL;
1304 goto out;
1305 }
1306
1307 res = ERR_PTR(-ENOMEM);
1308 fhandle = nfs_alloc_fhandle();
1309 fattr = nfs_alloc_fattr();
1310 if (fhandle == NULL || fattr == NULL)
1311 goto out;
1312
1313 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1314 if (IS_ERR(label))
1315 goto out;
1316
1317 parent = dentry->d_parent;
1318 /* Protect against concurrent sillydeletes */
1319 trace_nfs_lookup_enter(dir, dentry, flags);
1320 nfs_block_sillyrename(parent);
1321 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1322 if (error == -ENOENT)
1323 goto no_entry;
1324 if (error < 0) {
1325 res = ERR_PTR(error);
1326 goto out_unblock_sillyrename;
1327 }
1328 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1329 res = ERR_CAST(inode);
1330 if (IS_ERR(res))
1331 goto out_unblock_sillyrename;
1332
1333 /* Success: notify readdir to use READDIRPLUS */
1334 nfs_advise_use_readdirplus(dir);
1335
1336no_entry:
1337 res = d_materialise_unique(dentry, inode);
1338 if (res != NULL) {
1339 if (IS_ERR(res))
1340 goto out_unblock_sillyrename;
1341 dentry = res;
1342 }
1343 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1344out_unblock_sillyrename:
1345 nfs_unblock_sillyrename(parent);
1346 trace_nfs_lookup_exit(dir, dentry, flags, error);
1347 nfs4_label_free(label);
1348out:
1349 nfs_free_fattr(fattr);
1350 nfs_free_fhandle(fhandle);
1351 return res;
1352}
1353EXPORT_SYMBOL_GPL(nfs_lookup);
1354
1355#if IS_ENABLED(CONFIG_NFS_V4)
1356static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1357
1358const struct dentry_operations nfs4_dentry_operations = {
1359 .d_revalidate = nfs4_lookup_revalidate,
1360 .d_delete = nfs_dentry_delete,
1361 .d_iput = nfs_dentry_iput,
1362 .d_automount = nfs_d_automount,
1363 .d_release = nfs_d_release,
1364};
1365EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1366
1367static fmode_t flags_to_mode(int flags)
1368{
1369 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1370 if ((flags & O_ACCMODE) != O_WRONLY)
1371 res |= FMODE_READ;
1372 if ((flags & O_ACCMODE) != O_RDONLY)
1373 res |= FMODE_WRITE;
1374 return res;
1375}
1376
1377static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1378{
1379 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1380}
1381
1382static int do_open(struct inode *inode, struct file *filp)
1383{
1384 nfs_fscache_set_inode_cookie(inode, filp);
1385 return 0;
1386}
1387
1388static int nfs_finish_open(struct nfs_open_context *ctx,
1389 struct dentry *dentry,
1390 struct file *file, unsigned open_flags,
1391 int *opened)
1392{
1393 int err;
1394
1395 err = finish_open(file, dentry, do_open, opened);
1396 if (err)
1397 goto out;
1398 nfs_file_set_open_context(file, ctx);
1399
1400out:
1401 return err;
1402}
1403
1404int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1405 struct file *file, unsigned open_flags,
1406 umode_t mode, int *opened)
1407{
1408 struct nfs_open_context *ctx;
1409 struct dentry *res;
1410 struct iattr attr = { .ia_valid = ATTR_OPEN };
1411 struct inode *inode;
1412 unsigned int lookup_flags = 0;
1413 int err;
1414
1415 /* Expect a negative dentry */
1416 BUG_ON(dentry->d_inode);
1417
1418 dfprintk(VFS, "NFS: atomic_open(%s/%ld), %s\n",
1419 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1420
1421 err = nfs_check_flags(open_flags);
1422 if (err)
1423 return err;
1424
1425 /* NFS only supports OPEN on regular files */
1426 if ((open_flags & O_DIRECTORY)) {
1427 if (!d_unhashed(dentry)) {
1428 /*
1429 * Hashed negative dentry with O_DIRECTORY: dentry was
1430 * revalidated and is fine, no need to perform lookup
1431 * again
1432 */
1433 return -ENOENT;
1434 }
1435 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1436 goto no_open;
1437 }
1438
1439 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1440 return -ENAMETOOLONG;
1441
1442 if (open_flags & O_CREAT) {
1443 attr.ia_valid |= ATTR_MODE;
1444 attr.ia_mode = mode & ~current_umask();
1445 }
1446 if (open_flags & O_TRUNC) {
1447 attr.ia_valid |= ATTR_SIZE;
1448 attr.ia_size = 0;
1449 }
1450
1451 ctx = create_nfs_open_context(dentry, open_flags);
1452 err = PTR_ERR(ctx);
1453 if (IS_ERR(ctx))
1454 goto out;
1455
1456 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1457 nfs_block_sillyrename(dentry->d_parent);
1458 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
1459 nfs_unblock_sillyrename(dentry->d_parent);
1460 if (IS_ERR(inode)) {
1461 err = PTR_ERR(inode);
1462 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1463 put_nfs_open_context(ctx);
1464 switch (err) {
1465 case -ENOENT:
1466 d_drop(dentry);
1467 d_add(dentry, NULL);
1468 break;
1469 case -EISDIR:
1470 case -ENOTDIR:
1471 goto no_open;
1472 case -ELOOP:
1473 if (!(open_flags & O_NOFOLLOW))
1474 goto no_open;
1475 break;
1476 /* case -EINVAL: */
1477 default:
1478 break;
1479 }
1480 goto out;
1481 }
1482
1483 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1484 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1485 put_nfs_open_context(ctx);
1486out:
1487 return err;
1488
1489no_open:
1490 res = nfs_lookup(dir, dentry, lookup_flags);
1491 err = PTR_ERR(res);
1492 if (IS_ERR(res))
1493 goto out;
1494
1495 return finish_no_open(file, res);
1496}
1497EXPORT_SYMBOL_GPL(nfs_atomic_open);
1498
1499static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1500{
1501 struct dentry *parent = NULL;
1502 struct inode *inode;
1503 struct inode *dir;
1504 int ret = 0;
1505
1506 if (flags & LOOKUP_RCU)
1507 return -ECHILD;
1508
1509 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1510 goto no_open;
1511 if (d_mountpoint(dentry))
1512 goto no_open;
1513 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1514 goto no_open;
1515
1516 inode = dentry->d_inode;
1517 parent = dget_parent(dentry);
1518 dir = parent->d_inode;
1519
1520 /* We can't create new files in nfs_open_revalidate(), so we
1521 * optimize away revalidation of negative dentries.
1522 */
1523 if (inode == NULL) {
1524 if (!nfs_neg_need_reval(dir, dentry, flags))
1525 ret = 1;
1526 goto out;
1527 }
1528
1529 /* NFS only supports OPEN on regular files */
1530 if (!S_ISREG(inode->i_mode))
1531 goto no_open_dput;
1532 /* We cannot do exclusive creation on a positive dentry */
1533 if (flags & LOOKUP_EXCL)
1534 goto no_open_dput;
1535
1536 /* Let f_op->open() actually open (and revalidate) the file */
1537 ret = 1;
1538
1539out:
1540 dput(parent);
1541 return ret;
1542
1543no_open_dput:
1544 dput(parent);
1545no_open:
1546 return nfs_lookup_revalidate(dentry, flags);
1547}
1548
1549#endif /* CONFIG_NFSV4 */
1550
1551/*
1552 * Code common to create, mkdir, and mknod.
1553 */
1554int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1555 struct nfs_fattr *fattr,
1556 struct nfs4_label *label)
1557{
1558 struct dentry *parent = dget_parent(dentry);
1559 struct inode *dir = parent->d_inode;
1560 struct inode *inode;
1561 int error = -EACCES;
1562
1563 d_drop(dentry);
1564
1565 /* We may have been initialized further down */
1566 if (dentry->d_inode)
1567 goto out;
1568 if (fhandle->size == 0) {
1569 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1570 if (error)
1571 goto out_error;
1572 }
1573 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1574 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1575 struct nfs_server *server = NFS_SB(dentry->d_sb);
1576 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1577 if (error < 0)
1578 goto out_error;
1579 }
1580 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1581 error = PTR_ERR(inode);
1582 if (IS_ERR(inode))
1583 goto out_error;
1584 d_add(dentry, inode);
1585out:
1586 dput(parent);
1587 return 0;
1588out_error:
1589 nfs_mark_for_revalidate(dir);
1590 dput(parent);
1591 return error;
1592}
1593EXPORT_SYMBOL_GPL(nfs_instantiate);
1594
1595/*
1596 * Following a failed create operation, we drop the dentry rather
1597 * than retain a negative dentry. This avoids a problem in the event
1598 * that the operation succeeded on the server, but an error in the
1599 * reply path made it appear to have failed.
1600 */
1601int nfs_create(struct inode *dir, struct dentry *dentry,
1602 umode_t mode, bool excl)
1603{
1604 struct iattr attr;
1605 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1606 int error;
1607
1608 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1609 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1610
1611 attr.ia_mode = mode;
1612 attr.ia_valid = ATTR_MODE;
1613
1614 trace_nfs_create_enter(dir, dentry, open_flags);
1615 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1616 trace_nfs_create_exit(dir, dentry, open_flags, error);
1617 if (error != 0)
1618 goto out_err;
1619 return 0;
1620out_err:
1621 d_drop(dentry);
1622 return error;
1623}
1624EXPORT_SYMBOL_GPL(nfs_create);
1625
1626/*
1627 * See comments for nfs_proc_create regarding failed operations.
1628 */
1629int
1630nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1631{
1632 struct iattr attr;
1633 int status;
1634
1635 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1636 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1637
1638 if (!new_valid_dev(rdev))
1639 return -EINVAL;
1640
1641 attr.ia_mode = mode;
1642 attr.ia_valid = ATTR_MODE;
1643
1644 trace_nfs_mknod_enter(dir, dentry);
1645 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1646 trace_nfs_mknod_exit(dir, dentry, status);
1647 if (status != 0)
1648 goto out_err;
1649 return 0;
1650out_err:
1651 d_drop(dentry);
1652 return status;
1653}
1654EXPORT_SYMBOL_GPL(nfs_mknod);
1655
1656/*
1657 * See comments for nfs_proc_create regarding failed operations.
1658 */
1659int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1660{
1661 struct iattr attr;
1662 int error;
1663
1664 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1665 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1666
1667 attr.ia_valid = ATTR_MODE;
1668 attr.ia_mode = mode | S_IFDIR;
1669
1670 trace_nfs_mkdir_enter(dir, dentry);
1671 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1672 trace_nfs_mkdir_exit(dir, dentry, error);
1673 if (error != 0)
1674 goto out_err;
1675 return 0;
1676out_err:
1677 d_drop(dentry);
1678 return error;
1679}
1680EXPORT_SYMBOL_GPL(nfs_mkdir);
1681
1682static void nfs_dentry_handle_enoent(struct dentry *dentry)
1683{
1684 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1685 d_delete(dentry);
1686}
1687
1688int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1689{
1690 int error;
1691
1692 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1693 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1694
1695 trace_nfs_rmdir_enter(dir, dentry);
1696 if (dentry->d_inode) {
1697 nfs_wait_on_sillyrename(dentry);
1698 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1699 /* Ensure the VFS deletes this inode */
1700 switch (error) {
1701 case 0:
1702 clear_nlink(dentry->d_inode);
1703 break;
1704 case -ENOENT:
1705 nfs_dentry_handle_enoent(dentry);
1706 }
1707 } else
1708 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1709 trace_nfs_rmdir_exit(dir, dentry, error);
1710
1711 return error;
1712}
1713EXPORT_SYMBOL_GPL(nfs_rmdir);
1714
1715/*
1716 * Remove a file after making sure there are no pending writes,
1717 * and after checking that the file has only one user.
1718 *
1719 * We invalidate the attribute cache and free the inode prior to the operation
1720 * to avoid possible races if the server reuses the inode.
1721 */
1722static int nfs_safe_remove(struct dentry *dentry)
1723{
1724 struct inode *dir = dentry->d_parent->d_inode;
1725 struct inode *inode = dentry->d_inode;
1726 int error = -EBUSY;
1727
1728 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1729 dentry->d_parent->d_name.name, dentry->d_name.name);
1730
1731 /* If the dentry was sillyrenamed, we simply call d_delete() */
1732 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1733 error = 0;
1734 goto out;
1735 }
1736
1737 trace_nfs_remove_enter(dir, dentry);
1738 if (inode != NULL) {
1739 NFS_PROTO(inode)->return_delegation(inode);
1740 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1741 if (error == 0)
1742 nfs_drop_nlink(inode);
1743 } else
1744 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1745 if (error == -ENOENT)
1746 nfs_dentry_handle_enoent(dentry);
1747 trace_nfs_remove_exit(dir, dentry, error);
1748out:
1749 return error;
1750}
1751
1752/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1753 * belongs to an active ".nfs..." file and we return -EBUSY.
1754 *
1755 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1756 */
1757int nfs_unlink(struct inode *dir, struct dentry *dentry)
1758{
1759 int error;
1760 int need_rehash = 0;
1761
1762 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1763 dir->i_ino, dentry->d_name.name);
1764
1765 trace_nfs_unlink_enter(dir, dentry);
1766 spin_lock(&dentry->d_lock);
1767 if (d_count(dentry) > 1) {
1768 spin_unlock(&dentry->d_lock);
1769 /* Start asynchronous writeout of the inode */
1770 write_inode_now(dentry->d_inode, 0);
1771 error = nfs_sillyrename(dir, dentry);
1772 goto out;
1773 }
1774 if (!d_unhashed(dentry)) {
1775 __d_drop(dentry);
1776 need_rehash = 1;
1777 }
1778 spin_unlock(&dentry->d_lock);
1779 error = nfs_safe_remove(dentry);
1780 if (!error || error == -ENOENT) {
1781 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1782 } else if (need_rehash)
1783 d_rehash(dentry);
1784out:
1785 trace_nfs_unlink_exit(dir, dentry, error);
1786 return error;
1787}
1788EXPORT_SYMBOL_GPL(nfs_unlink);
1789
1790/*
1791 * To create a symbolic link, most file systems instantiate a new inode,
1792 * add a page to it containing the path, then write it out to the disk
1793 * using prepare_write/commit_write.
1794 *
1795 * Unfortunately the NFS client can't create the in-core inode first
1796 * because it needs a file handle to create an in-core inode (see
1797 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1798 * symlink request has completed on the server.
1799 *
1800 * So instead we allocate a raw page, copy the symname into it, then do
1801 * the SYMLINK request with the page as the buffer. If it succeeds, we
1802 * now have a new file handle and can instantiate an in-core NFS inode
1803 * and move the raw page into its mapping.
1804 */
1805int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1806{
1807 struct page *page;
1808 char *kaddr;
1809 struct iattr attr;
1810 unsigned int pathlen = strlen(symname);
1811 int error;
1812
1813 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1814 dir->i_ino, dentry->d_name.name, symname);
1815
1816 if (pathlen > PAGE_SIZE)
1817 return -ENAMETOOLONG;
1818
1819 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1820 attr.ia_valid = ATTR_MODE;
1821
1822 page = alloc_page(GFP_HIGHUSER);
1823 if (!page)
1824 return -ENOMEM;
1825
1826 kaddr = kmap_atomic(page);
1827 memcpy(kaddr, symname, pathlen);
1828 if (pathlen < PAGE_SIZE)
1829 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1830 kunmap_atomic(kaddr);
1831
1832 trace_nfs_symlink_enter(dir, dentry);
1833 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1834 trace_nfs_symlink_exit(dir, dentry, error);
1835 if (error != 0) {
1836 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1837 dir->i_sb->s_id, dir->i_ino,
1838 dentry->d_name.name, symname, error);
1839 d_drop(dentry);
1840 __free_page(page);
1841 return error;
1842 }
1843
1844 /*
1845 * No big deal if we can't add this page to the page cache here.
1846 * READLINK will get the missing page from the server if needed.
1847 */
1848 if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0,
1849 GFP_KERNEL)) {
1850 SetPageUptodate(page);
1851 unlock_page(page);
1852 } else
1853 __free_page(page);
1854
1855 return 0;
1856}
1857EXPORT_SYMBOL_GPL(nfs_symlink);
1858
1859int
1860nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1861{
1862 struct inode *inode = old_dentry->d_inode;
1863 int error;
1864
1865 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1866 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1867 dentry->d_parent->d_name.name, dentry->d_name.name);
1868
1869 trace_nfs_link_enter(inode, dir, dentry);
1870 NFS_PROTO(inode)->return_delegation(inode);
1871
1872 d_drop(dentry);
1873 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1874 if (error == 0) {
1875 ihold(inode);
1876 d_add(dentry, inode);
1877 }
1878 trace_nfs_link_exit(inode, dir, dentry, error);
1879 return error;
1880}
1881EXPORT_SYMBOL_GPL(nfs_link);
1882
1883/*
1884 * RENAME
1885 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1886 * different file handle for the same inode after a rename (e.g. when
1887 * moving to a different directory). A fail-safe method to do so would
1888 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1889 * rename the old file using the sillyrename stuff. This way, the original
1890 * file in old_dir will go away when the last process iput()s the inode.
1891 *
1892 * FIXED.
1893 *
1894 * It actually works quite well. One needs to have the possibility for
1895 * at least one ".nfs..." file in each directory the file ever gets
1896 * moved or linked to which happens automagically with the new
1897 * implementation that only depends on the dcache stuff instead of
1898 * using the inode layer
1899 *
1900 * Unfortunately, things are a little more complicated than indicated
1901 * above. For a cross-directory move, we want to make sure we can get
1902 * rid of the old inode after the operation. This means there must be
1903 * no pending writes (if it's a file), and the use count must be 1.
1904 * If these conditions are met, we can drop the dentries before doing
1905 * the rename.
1906 */
1907int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1908 struct inode *new_dir, struct dentry *new_dentry)
1909{
1910 struct inode *old_inode = old_dentry->d_inode;
1911 struct inode *new_inode = new_dentry->d_inode;
1912 struct dentry *dentry = NULL, *rehash = NULL;
1913 int error = -EBUSY;
1914
1915 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1916 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1917 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1918 d_count(new_dentry));
1919
1920 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
1921 /*
1922 * For non-directories, check whether the target is busy and if so,
1923 * make a copy of the dentry and then do a silly-rename. If the
1924 * silly-rename succeeds, the copied dentry is hashed and becomes
1925 * the new target.
1926 */
1927 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1928 /*
1929 * To prevent any new references to the target during the
1930 * rename, we unhash the dentry in advance.
1931 */
1932 if (!d_unhashed(new_dentry)) {
1933 d_drop(new_dentry);
1934 rehash = new_dentry;
1935 }
1936
1937 if (d_count(new_dentry) > 2) {
1938 int err;
1939
1940 /* copy the target dentry's name */
1941 dentry = d_alloc(new_dentry->d_parent,
1942 &new_dentry->d_name);
1943 if (!dentry)
1944 goto out;
1945
1946 /* silly-rename the existing target ... */
1947 err = nfs_sillyrename(new_dir, new_dentry);
1948 if (err)
1949 goto out;
1950
1951 new_dentry = dentry;
1952 rehash = NULL;
1953 new_inode = NULL;
1954 }
1955 }
1956
1957 NFS_PROTO(old_inode)->return_delegation(old_inode);
1958 if (new_inode != NULL)
1959 NFS_PROTO(new_inode)->return_delegation(new_inode);
1960
1961 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1962 new_dir, &new_dentry->d_name);
1963 nfs_mark_for_revalidate(old_inode);
1964out:
1965 if (rehash)
1966 d_rehash(rehash);
1967 trace_nfs_rename_exit(old_dir, old_dentry,
1968 new_dir, new_dentry, error);
1969 if (!error) {
1970 if (new_inode != NULL)
1971 nfs_drop_nlink(new_inode);
1972 d_move(old_dentry, new_dentry);
1973 nfs_set_verifier(new_dentry,
1974 nfs_save_change_attribute(new_dir));
1975 } else if (error == -ENOENT)
1976 nfs_dentry_handle_enoent(old_dentry);
1977
1978 /* new dentry created? */
1979 if (dentry)
1980 dput(dentry);
1981 return error;
1982}
1983EXPORT_SYMBOL_GPL(nfs_rename);
1984
1985static DEFINE_SPINLOCK(nfs_access_lru_lock);
1986static LIST_HEAD(nfs_access_lru_list);
1987static atomic_long_t nfs_access_nr_entries;
1988
1989static void nfs_access_free_entry(struct nfs_access_entry *entry)
1990{
1991 put_rpccred(entry->cred);
1992 kfree(entry);
1993 smp_mb__before_atomic_dec();
1994 atomic_long_dec(&nfs_access_nr_entries);
1995 smp_mb__after_atomic_dec();
1996}
1997
1998static void nfs_access_free_list(struct list_head *head)
1999{
2000 struct nfs_access_entry *cache;
2001
2002 while (!list_empty(head)) {
2003 cache = list_entry(head->next, struct nfs_access_entry, lru);
2004 list_del(&cache->lru);
2005 nfs_access_free_entry(cache);
2006 }
2007}
2008
2009int nfs_access_cache_shrinker(struct shrinker *shrink,
2010 struct shrink_control *sc)
2011{
2012 LIST_HEAD(head);
2013 struct nfs_inode *nfsi, *next;
2014 struct nfs_access_entry *cache;
2015 int nr_to_scan = sc->nr_to_scan;
2016 gfp_t gfp_mask = sc->gfp_mask;
2017
2018 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2019 return (nr_to_scan == 0) ? 0 : -1;
2020
2021 spin_lock(&nfs_access_lru_lock);
2022 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2023 struct inode *inode;
2024
2025 if (nr_to_scan-- == 0)
2026 break;
2027 inode = &nfsi->vfs_inode;
2028 spin_lock(&inode->i_lock);
2029 if (list_empty(&nfsi->access_cache_entry_lru))
2030 goto remove_lru_entry;
2031 cache = list_entry(nfsi->access_cache_entry_lru.next,
2032 struct nfs_access_entry, lru);
2033 list_move(&cache->lru, &head);
2034 rb_erase(&cache->rb_node, &nfsi->access_cache);
2035 if (!list_empty(&nfsi->access_cache_entry_lru))
2036 list_move_tail(&nfsi->access_cache_inode_lru,
2037 &nfs_access_lru_list);
2038 else {
2039remove_lru_entry:
2040 list_del_init(&nfsi->access_cache_inode_lru);
2041 smp_mb__before_clear_bit();
2042 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2043 smp_mb__after_clear_bit();
2044 }
2045 spin_unlock(&inode->i_lock);
2046 }
2047 spin_unlock(&nfs_access_lru_lock);
2048 nfs_access_free_list(&head);
2049 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2050}
2051
2052static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2053{
2054 struct rb_root *root_node = &nfsi->access_cache;
2055 struct rb_node *n;
2056 struct nfs_access_entry *entry;
2057
2058 /* Unhook entries from the cache */
2059 while ((n = rb_first(root_node)) != NULL) {
2060 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2061 rb_erase(n, root_node);
2062 list_move(&entry->lru, head);
2063 }
2064 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2065}
2066
2067void nfs_access_zap_cache(struct inode *inode)
2068{
2069 LIST_HEAD(head);
2070
2071 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2072 return;
2073 /* Remove from global LRU init */
2074 spin_lock(&nfs_access_lru_lock);
2075 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2076 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2077
2078 spin_lock(&inode->i_lock);
2079 __nfs_access_zap_cache(NFS_I(inode), &head);
2080 spin_unlock(&inode->i_lock);
2081 spin_unlock(&nfs_access_lru_lock);
2082 nfs_access_free_list(&head);
2083}
2084EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2085
2086static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2087{
2088 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2089 struct nfs_access_entry *entry;
2090
2091 while (n != NULL) {
2092 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2093
2094 if (cred < entry->cred)
2095 n = n->rb_left;
2096 else if (cred > entry->cred)
2097 n = n->rb_right;
2098 else
2099 return entry;
2100 }
2101 return NULL;
2102}
2103
2104static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2105{
2106 struct nfs_inode *nfsi = NFS_I(inode);
2107 struct nfs_access_entry *cache;
2108 int err = -ENOENT;
2109
2110 spin_lock(&inode->i_lock);
2111 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2112 goto out_zap;
2113 cache = nfs_access_search_rbtree(inode, cred);
2114 if (cache == NULL)
2115 goto out;
2116 if (!nfs_have_delegated_attributes(inode) &&
2117 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2118 goto out_stale;
2119 res->jiffies = cache->jiffies;
2120 res->cred = cache->cred;
2121 res->mask = cache->mask;
2122 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2123 err = 0;
2124out:
2125 spin_unlock(&inode->i_lock);
2126 return err;
2127out_stale:
2128 rb_erase(&cache->rb_node, &nfsi->access_cache);
2129 list_del(&cache->lru);
2130 spin_unlock(&inode->i_lock);
2131 nfs_access_free_entry(cache);
2132 return -ENOENT;
2133out_zap:
2134 spin_unlock(&inode->i_lock);
2135 nfs_access_zap_cache(inode);
2136 return -ENOENT;
2137}
2138
2139static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2140{
2141 struct nfs_inode *nfsi = NFS_I(inode);
2142 struct rb_root *root_node = &nfsi->access_cache;
2143 struct rb_node **p = &root_node->rb_node;
2144 struct rb_node *parent = NULL;
2145 struct nfs_access_entry *entry;
2146
2147 spin_lock(&inode->i_lock);
2148 while (*p != NULL) {
2149 parent = *p;
2150 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2151
2152 if (set->cred < entry->cred)
2153 p = &parent->rb_left;
2154 else if (set->cred > entry->cred)
2155 p = &parent->rb_right;
2156 else
2157 goto found;
2158 }
2159 rb_link_node(&set->rb_node, parent, p);
2160 rb_insert_color(&set->rb_node, root_node);
2161 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2162 spin_unlock(&inode->i_lock);
2163 return;
2164found:
2165 rb_replace_node(parent, &set->rb_node, root_node);
2166 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2167 list_del(&entry->lru);
2168 spin_unlock(&inode->i_lock);
2169 nfs_access_free_entry(entry);
2170}
2171
2172void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2173{
2174 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2175 if (cache == NULL)
2176 return;
2177 RB_CLEAR_NODE(&cache->rb_node);
2178 cache->jiffies = set->jiffies;
2179 cache->cred = get_rpccred(set->cred);
2180 cache->mask = set->mask;
2181
2182 nfs_access_add_rbtree(inode, cache);
2183
2184 /* Update accounting */
2185 smp_mb__before_atomic_inc();
2186 atomic_long_inc(&nfs_access_nr_entries);
2187 smp_mb__after_atomic_inc();
2188
2189 /* Add inode to global LRU list */
2190 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2191 spin_lock(&nfs_access_lru_lock);
2192 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2193 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2194 &nfs_access_lru_list);
2195 spin_unlock(&nfs_access_lru_lock);
2196 }
2197}
2198EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2199
2200void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2201{
2202 entry->mask = 0;
2203 if (access_result & NFS4_ACCESS_READ)
2204 entry->mask |= MAY_READ;
2205 if (access_result &
2206 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2207 entry->mask |= MAY_WRITE;
2208 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2209 entry->mask |= MAY_EXEC;
2210}
2211EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2212
2213static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2214{
2215 struct nfs_access_entry cache;
2216 int status;
2217
2218 trace_nfs_access_enter(inode);
2219
2220 status = nfs_access_get_cached(inode, cred, &cache);
2221 if (status == 0)
2222 goto out_cached;
2223
2224 /* Be clever: ask server to check for all possible rights */
2225 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2226 cache.cred = cred;
2227 cache.jiffies = jiffies;
2228 status = NFS_PROTO(inode)->access(inode, &cache);
2229 if (status != 0) {
2230 if (status == -ESTALE) {
2231 nfs_zap_caches(inode);
2232 if (!S_ISDIR(inode->i_mode))
2233 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2234 }
2235 goto out;
2236 }
2237 nfs_access_add_cache(inode, &cache);
2238out_cached:
2239 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2240 status = -EACCES;
2241out:
2242 trace_nfs_access_exit(inode, status);
2243 return status;
2244}
2245
2246static int nfs_open_permission_mask(int openflags)
2247{
2248 int mask = 0;
2249
2250 if (openflags & __FMODE_EXEC) {
2251 /* ONLY check exec rights */
2252 mask = MAY_EXEC;
2253 } else {
2254 if ((openflags & O_ACCMODE) != O_WRONLY)
2255 mask |= MAY_READ;
2256 if ((openflags & O_ACCMODE) != O_RDONLY)
2257 mask |= MAY_WRITE;
2258 }
2259
2260 return mask;
2261}
2262
2263int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2264{
2265 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2266}
2267EXPORT_SYMBOL_GPL(nfs_may_open);
2268
2269int nfs_permission(struct inode *inode, int mask)
2270{
2271 struct rpc_cred *cred;
2272 int res = 0;
2273
2274 if (mask & MAY_NOT_BLOCK)
2275 return -ECHILD;
2276
2277 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2278
2279 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2280 goto out;
2281 /* Is this sys_access() ? */
2282 if (mask & (MAY_ACCESS | MAY_CHDIR))
2283 goto force_lookup;
2284
2285 switch (inode->i_mode & S_IFMT) {
2286 case S_IFLNK:
2287 goto out;
2288 case S_IFREG:
2289 break;
2290 case S_IFDIR:
2291 /*
2292 * Optimize away all write operations, since the server
2293 * will check permissions when we perform the op.
2294 */
2295 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2296 goto out;
2297 }
2298
2299force_lookup:
2300 if (!NFS_PROTO(inode)->access)
2301 goto out_notsup;
2302
2303 cred = rpc_lookup_cred();
2304 if (!IS_ERR(cred)) {
2305 res = nfs_do_access(inode, cred, mask);
2306 put_rpccred(cred);
2307 } else
2308 res = PTR_ERR(cred);
2309out:
2310 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2311 res = -EACCES;
2312
2313 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2314 inode->i_sb->s_id, inode->i_ino, mask, res);
2315 return res;
2316out_notsup:
2317 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2318 if (res == 0)
2319 res = generic_permission(inode, mask);
2320 goto out;
2321}
2322EXPORT_SYMBOL_GPL(nfs_permission);
2323
2324/*
2325 * Local variables:
2326 * version-control: t
2327 * kept-new-versions: 5
2328 * End:
2329 */