net: Move prototype declaration to header file include/net/ipx.h from net/ipx/af_ipx.c
[linux/fpc-iii.git] / fs / nfs / dir.c
blobb266f734bd5394a092edcf288fcd29d54e55df13
1 /*
2 * linux/fs/nfs/dir.c
4 * Copyright (C) 1992 Rick Sladkey
6 * nfs directory handling functions
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
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>
41 #include "delegation.h"
42 #include "iostat.h"
43 #include "internal.h"
44 #include "fscache.h"
46 #include "nfstrace.h"
48 /* #define NFS_DEBUG_VERBOSE 1 */
50 static int nfs_opendir(struct inode *, struct file *);
51 static int nfs_closedir(struct inode *, struct file *);
52 static int nfs_readdir(struct file *, struct dir_context *);
53 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
54 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
55 static void nfs_readdir_clear_array(struct page*);
57 const 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,
66 const struct address_space_operations nfs_dir_aops = {
67 .freepage = nfs_readdir_clear_array,
70 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, struct rpc_cred *cred)
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;
82 return ERR_PTR(-ENOMEM);
85 static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
87 put_rpccred(ctx->cred);
88 kfree(ctx);
92 * Open file
94 static int
95 nfs_opendir(struct inode *inode, struct file *filp)
97 int res = 0;
98 struct nfs_open_dir_context *ctx;
99 struct rpc_cred *cred;
101 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
103 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
105 cred = rpc_lookup_cred();
106 if (IS_ERR(cred))
107 return PTR_ERR(cred);
108 ctx = alloc_nfs_open_dir_context(inode, cred);
109 if (IS_ERR(ctx)) {
110 res = PTR_ERR(ctx);
111 goto out;
113 filp->private_data = ctx;
114 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
115 /* This is a mountpoint, so d_revalidate will never
116 * have been called, so we need to refresh the
117 * inode (for close-open consistency) ourselves.
119 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
121 out:
122 put_rpccred(cred);
123 return res;
126 static int
127 nfs_closedir(struct inode *inode, struct file *filp)
129 put_nfs_open_dir_context(filp->private_data);
130 return 0;
133 struct nfs_cache_array_entry {
134 u64 cookie;
135 u64 ino;
136 struct qstr string;
137 unsigned char d_type;
140 struct nfs_cache_array {
141 int size;
142 int eof_index;
143 u64 last_cookie;
144 struct nfs_cache_array_entry array[0];
147 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
148 typedef struct {
149 struct file *file;
150 struct page *page;
151 struct dir_context *ctx;
152 unsigned long page_index;
153 u64 *dir_cookie;
154 u64 last_cookie;
155 loff_t current_index;
156 decode_dirent_t decode;
158 unsigned long timestamp;
159 unsigned long gencount;
160 unsigned int cache_entry_index;
161 unsigned int plus:1;
162 unsigned int eof:1;
163 } nfs_readdir_descriptor_t;
166 * The caller is responsible for calling nfs_readdir_release_array(page)
168 static
169 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
171 void *ptr;
172 if (page == NULL)
173 return ERR_PTR(-EIO);
174 ptr = kmap(page);
175 if (ptr == NULL)
176 return ERR_PTR(-ENOMEM);
177 return ptr;
180 static
181 void nfs_readdir_release_array(struct page *page)
183 kunmap(page);
187 * we are freeing strings created by nfs_add_to_readdir_array()
189 static
190 void nfs_readdir_clear_array(struct page *page)
192 struct nfs_cache_array *array;
193 int i;
195 array = kmap_atomic(page);
196 for (i = 0; i < array->size; i++)
197 kfree(array->array[i].string.name);
198 kunmap_atomic(array);
202 * the caller is responsible for freeing qstr.name
203 * when called by nfs_readdir_add_to_array, the strings will be freed in
204 * nfs_clear_readdir_array()
206 static
207 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
209 string->len = len;
210 string->name = kmemdup(name, len, GFP_KERNEL);
211 if (string->name == NULL)
212 return -ENOMEM;
214 * Avoid a kmemleak false positive. The pointer to the name is stored
215 * in a page cache page which kmemleak does not scan.
217 kmemleak_not_leak(string->name);
218 string->hash = full_name_hash(name, len);
219 return 0;
222 static
223 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
225 struct nfs_cache_array *array = nfs_readdir_get_array(page);
226 struct nfs_cache_array_entry *cache_entry;
227 int ret;
229 if (IS_ERR(array))
230 return PTR_ERR(array);
232 cache_entry = &array->array[array->size];
234 /* Check that this entry lies within the page bounds */
235 ret = -ENOSPC;
236 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
237 goto out;
239 cache_entry->cookie = entry->prev_cookie;
240 cache_entry->ino = entry->ino;
241 cache_entry->d_type = entry->d_type;
242 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
243 if (ret)
244 goto out;
245 array->last_cookie = entry->cookie;
246 array->size++;
247 if (entry->eof != 0)
248 array->eof_index = array->size;
249 out:
250 nfs_readdir_release_array(page);
251 return ret;
254 static
255 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
257 loff_t diff = desc->ctx->pos - desc->current_index;
258 unsigned int index;
260 if (diff < 0)
261 goto out_eof;
262 if (diff >= array->size) {
263 if (array->eof_index >= 0)
264 goto out_eof;
265 return -EAGAIN;
268 index = (unsigned int)diff;
269 *desc->dir_cookie = array->array[index].cookie;
270 desc->cache_entry_index = index;
271 return 0;
272 out_eof:
273 desc->eof = 1;
274 return -EBADCOOKIE;
277 static
278 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
280 int i;
281 loff_t new_pos;
282 int status = -EAGAIN;
284 for (i = 0; i < array->size; i++) {
285 if (array->array[i].cookie == *desc->dir_cookie) {
286 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
287 struct nfs_open_dir_context *ctx = desc->file->private_data;
289 new_pos = desc->current_index + i;
290 if (ctx->attr_gencount != nfsi->attr_gencount
291 || (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))) {
292 ctx->duped = 0;
293 ctx->attr_gencount = nfsi->attr_gencount;
294 } else if (new_pos < desc->ctx->pos) {
295 if (ctx->duped > 0
296 && ctx->dup_cookie == *desc->dir_cookie) {
297 if (printk_ratelimit()) {
298 pr_notice("NFS: directory %pD2 contains a readdir loop."
299 "Please contact your server vendor. "
300 "The file: %s has duplicate cookie %llu\n",
301 desc->file,
302 array->array[i].string.name,
303 *desc->dir_cookie);
305 status = -ELOOP;
306 goto out;
308 ctx->dup_cookie = *desc->dir_cookie;
309 ctx->duped = -1;
311 desc->ctx->pos = new_pos;
312 desc->cache_entry_index = i;
313 return 0;
316 if (array->eof_index >= 0) {
317 status = -EBADCOOKIE;
318 if (*desc->dir_cookie == array->last_cookie)
319 desc->eof = 1;
321 out:
322 return status;
325 static
326 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
328 struct nfs_cache_array *array;
329 int status;
331 array = nfs_readdir_get_array(desc->page);
332 if (IS_ERR(array)) {
333 status = PTR_ERR(array);
334 goto out;
337 if (*desc->dir_cookie == 0)
338 status = nfs_readdir_search_for_pos(array, desc);
339 else
340 status = nfs_readdir_search_for_cookie(array, desc);
342 if (status == -EAGAIN) {
343 desc->last_cookie = array->last_cookie;
344 desc->current_index += array->size;
345 desc->page_index++;
347 nfs_readdir_release_array(desc->page);
348 out:
349 return status;
352 /* Fill a page with xdr information before transferring to the cache page */
353 static
354 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
355 struct nfs_entry *entry, struct file *file, struct inode *inode)
357 struct nfs_open_dir_context *ctx = file->private_data;
358 struct rpc_cred *cred = ctx->cred;
359 unsigned long timestamp, gencount;
360 int error;
362 again:
363 timestamp = jiffies;
364 gencount = nfs_inc_attr_generation_counter();
365 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
366 NFS_SERVER(inode)->dtsize, desc->plus);
367 if (error < 0) {
368 /* We requested READDIRPLUS, but the server doesn't grok it */
369 if (error == -ENOTSUPP && desc->plus) {
370 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
371 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
372 desc->plus = 0;
373 goto again;
375 goto error;
377 desc->timestamp = timestamp;
378 desc->gencount = gencount;
379 error:
380 return error;
383 static int xdr_decode(nfs_readdir_descriptor_t *desc,
384 struct nfs_entry *entry, struct xdr_stream *xdr)
386 int error;
388 error = desc->decode(xdr, entry, desc->plus);
389 if (error)
390 return error;
391 entry->fattr->time_start = desc->timestamp;
392 entry->fattr->gencount = desc->gencount;
393 return 0;
396 static
397 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
399 if (dentry->d_inode == NULL)
400 goto different;
401 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
402 goto different;
403 return 1;
404 different:
405 return 0;
408 static
409 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
411 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
412 return false;
413 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
414 return true;
415 if (ctx->pos == 0)
416 return true;
417 return false;
421 * This function is called by the lookup code to request the use of
422 * readdirplus to accelerate any future lookups in the same
423 * directory.
425 static
426 void nfs_advise_use_readdirplus(struct inode *dir)
428 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
431 static
432 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
434 struct qstr filename = QSTR_INIT(entry->name, entry->len);
435 struct dentry *dentry;
436 struct dentry *alias;
437 struct inode *dir = parent->d_inode;
438 struct inode *inode;
439 int status;
441 if (filename.name[0] == '.') {
442 if (filename.len == 1)
443 return;
444 if (filename.len == 2 && filename.name[1] == '.')
445 return;
447 filename.hash = full_name_hash(filename.name, filename.len);
449 dentry = d_lookup(parent, &filename);
450 if (dentry != NULL) {
451 if (nfs_same_file(dentry, entry)) {
452 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
453 status = nfs_refresh_inode(dentry->d_inode, entry->fattr);
454 if (!status)
455 nfs_setsecurity(dentry->d_inode, entry->fattr, entry->label);
456 goto out;
457 } else {
458 if (d_invalidate(dentry) != 0)
459 goto out;
460 dput(dentry);
464 dentry = d_alloc(parent, &filename);
465 if (dentry == NULL)
466 return;
468 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
469 if (IS_ERR(inode))
470 goto out;
472 alias = d_materialise_unique(dentry, inode);
473 if (IS_ERR(alias))
474 goto out;
475 else if (alias) {
476 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
477 dput(alias);
478 } else
479 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
481 out:
482 dput(dentry);
485 /* Perform conversion from xdr to cache array */
486 static
487 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
488 struct page **xdr_pages, struct page *page, unsigned int buflen)
490 struct xdr_stream stream;
491 struct xdr_buf buf;
492 struct page *scratch;
493 struct nfs_cache_array *array;
494 unsigned int count = 0;
495 int status;
497 scratch = alloc_page(GFP_KERNEL);
498 if (scratch == NULL)
499 return -ENOMEM;
501 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
502 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
504 do {
505 status = xdr_decode(desc, entry, &stream);
506 if (status != 0) {
507 if (status == -EAGAIN)
508 status = 0;
509 break;
512 count++;
514 if (desc->plus != 0)
515 nfs_prime_dcache(desc->file->f_path.dentry, entry);
517 status = nfs_readdir_add_to_array(entry, page);
518 if (status != 0)
519 break;
520 } while (!entry->eof);
522 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
523 array = nfs_readdir_get_array(page);
524 if (!IS_ERR(array)) {
525 array->eof_index = array->size;
526 status = 0;
527 nfs_readdir_release_array(page);
528 } else
529 status = PTR_ERR(array);
532 put_page(scratch);
533 return status;
536 static
537 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
539 unsigned int i;
540 for (i = 0; i < npages; i++)
541 put_page(pages[i]);
544 static
545 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
546 unsigned int npages)
548 nfs_readdir_free_pagearray(pages, npages);
552 * nfs_readdir_large_page will allocate pages that must be freed with a call
553 * to nfs_readdir_free_large_page
555 static
556 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
558 unsigned int i;
560 for (i = 0; i < npages; i++) {
561 struct page *page = alloc_page(GFP_KERNEL);
562 if (page == NULL)
563 goto out_freepages;
564 pages[i] = page;
566 return 0;
568 out_freepages:
569 nfs_readdir_free_pagearray(pages, i);
570 return -ENOMEM;
573 static
574 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
576 struct page *pages[NFS_MAX_READDIR_PAGES];
577 void *pages_ptr = NULL;
578 struct nfs_entry entry;
579 struct file *file = desc->file;
580 struct nfs_cache_array *array;
581 int status = -ENOMEM;
582 unsigned int array_size = ARRAY_SIZE(pages);
584 entry.prev_cookie = 0;
585 entry.cookie = desc->last_cookie;
586 entry.eof = 0;
587 entry.fh = nfs_alloc_fhandle();
588 entry.fattr = nfs_alloc_fattr();
589 entry.server = NFS_SERVER(inode);
590 if (entry.fh == NULL || entry.fattr == NULL)
591 goto out;
593 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
594 if (IS_ERR(entry.label)) {
595 status = PTR_ERR(entry.label);
596 goto out;
599 array = nfs_readdir_get_array(page);
600 if (IS_ERR(array)) {
601 status = PTR_ERR(array);
602 goto out_label_free;
604 memset(array, 0, sizeof(struct nfs_cache_array));
605 array->eof_index = -1;
607 status = nfs_readdir_large_page(pages, array_size);
608 if (status < 0)
609 goto out_release_array;
610 do {
611 unsigned int pglen;
612 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
614 if (status < 0)
615 break;
616 pglen = status;
617 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
618 if (status < 0) {
619 if (status == -ENOSPC)
620 status = 0;
621 break;
623 } while (array->eof_index < 0);
625 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
626 out_release_array:
627 nfs_readdir_release_array(page);
628 out_label_free:
629 nfs4_label_free(entry.label);
630 out:
631 nfs_free_fattr(entry.fattr);
632 nfs_free_fhandle(entry.fh);
633 return status;
637 * Now we cache directories properly, by converting xdr information
638 * to an array that can be used for lookups later. This results in
639 * fewer cache pages, since we can store more information on each page.
640 * We only need to convert from xdr once so future lookups are much simpler
642 static
643 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
645 struct inode *inode = file_inode(desc->file);
646 int ret;
648 ret = nfs_readdir_xdr_to_array(desc, page, inode);
649 if (ret < 0)
650 goto error;
651 SetPageUptodate(page);
653 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
654 /* Should never happen */
655 nfs_zap_mapping(inode, inode->i_mapping);
657 unlock_page(page);
658 return 0;
659 error:
660 unlock_page(page);
661 return ret;
664 static
665 void cache_page_release(nfs_readdir_descriptor_t *desc)
667 if (!desc->page->mapping)
668 nfs_readdir_clear_array(desc->page);
669 page_cache_release(desc->page);
670 desc->page = NULL;
673 static
674 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
676 return read_cache_page(file_inode(desc->file)->i_mapping,
677 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
681 * Returns 0 if desc->dir_cookie was found on page desc->page_index
683 static
684 int find_cache_page(nfs_readdir_descriptor_t *desc)
686 int res;
688 desc->page = get_cache_page(desc);
689 if (IS_ERR(desc->page))
690 return PTR_ERR(desc->page);
692 res = nfs_readdir_search_array(desc);
693 if (res != 0)
694 cache_page_release(desc);
695 return res;
698 /* Search for desc->dir_cookie from the beginning of the page cache */
699 static inline
700 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
702 int res;
704 if (desc->page_index == 0) {
705 desc->current_index = 0;
706 desc->last_cookie = 0;
708 do {
709 res = find_cache_page(desc);
710 } while (res == -EAGAIN);
711 return res;
715 * Once we've found the start of the dirent within a page: fill 'er up...
717 static
718 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
720 struct file *file = desc->file;
721 int i = 0;
722 int res = 0;
723 struct nfs_cache_array *array = NULL;
724 struct nfs_open_dir_context *ctx = file->private_data;
726 array = nfs_readdir_get_array(desc->page);
727 if (IS_ERR(array)) {
728 res = PTR_ERR(array);
729 goto out;
732 for (i = desc->cache_entry_index; i < array->size; i++) {
733 struct nfs_cache_array_entry *ent;
735 ent = &array->array[i];
736 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
737 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
738 desc->eof = 1;
739 break;
741 desc->ctx->pos++;
742 if (i < (array->size-1))
743 *desc->dir_cookie = array->array[i+1].cookie;
744 else
745 *desc->dir_cookie = array->last_cookie;
746 if (ctx->duped != 0)
747 ctx->duped = 1;
749 if (array->eof_index >= 0)
750 desc->eof = 1;
752 nfs_readdir_release_array(desc->page);
753 out:
754 cache_page_release(desc);
755 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
756 (unsigned long long)*desc->dir_cookie, res);
757 return res;
761 * If we cannot find a cookie in our cache, we suspect that this is
762 * because it points to a deleted file, so we ask the server to return
763 * whatever it thinks is the next entry. We then feed this to filldir.
764 * If all goes well, we should then be able to find our way round the
765 * cache on the next call to readdir_search_pagecache();
767 * NOTE: we cannot add the anonymous page to the pagecache because
768 * the data it contains might not be page aligned. Besides,
769 * we should already have a complete representation of the
770 * directory in the page cache by the time we get here.
772 static inline
773 int uncached_readdir(nfs_readdir_descriptor_t *desc)
775 struct page *page = NULL;
776 int status;
777 struct inode *inode = file_inode(desc->file);
778 struct nfs_open_dir_context *ctx = desc->file->private_data;
780 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
781 (unsigned long long)*desc->dir_cookie);
783 page = alloc_page(GFP_HIGHUSER);
784 if (!page) {
785 status = -ENOMEM;
786 goto out;
789 desc->page_index = 0;
790 desc->last_cookie = *desc->dir_cookie;
791 desc->page = page;
792 ctx->duped = 0;
794 status = nfs_readdir_xdr_to_array(desc, page, inode);
795 if (status < 0)
796 goto out_release;
798 status = nfs_do_filldir(desc);
800 out:
801 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
802 __func__, status);
803 return status;
804 out_release:
805 cache_page_release(desc);
806 goto out;
809 /* The file offset position represents the dirent entry number. A
810 last cookie cache takes care of the common case of reading the
811 whole directory.
813 static int nfs_readdir(struct file *file, struct dir_context *ctx)
815 struct dentry *dentry = file->f_path.dentry;
816 struct inode *inode = dentry->d_inode;
817 nfs_readdir_descriptor_t my_desc,
818 *desc = &my_desc;
819 struct nfs_open_dir_context *dir_ctx = file->private_data;
820 int res = 0;
822 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
823 file, (long long)ctx->pos);
824 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
827 * ctx->pos points to the dirent entry number.
828 * *desc->dir_cookie has the cookie for the next entry. We have
829 * to either find the entry with the appropriate number or
830 * revalidate the cookie.
832 memset(desc, 0, sizeof(*desc));
834 desc->file = file;
835 desc->ctx = ctx;
836 desc->dir_cookie = &dir_ctx->dir_cookie;
837 desc->decode = NFS_PROTO(inode)->decode_dirent;
838 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
840 nfs_block_sillyrename(dentry);
841 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
842 res = nfs_revalidate_mapping(inode, file->f_mapping);
843 if (res < 0)
844 goto out;
846 do {
847 res = readdir_search_pagecache(desc);
849 if (res == -EBADCOOKIE) {
850 res = 0;
851 /* This means either end of directory */
852 if (*desc->dir_cookie && desc->eof == 0) {
853 /* Or that the server has 'lost' a cookie */
854 res = uncached_readdir(desc);
855 if (res == 0)
856 continue;
858 break;
860 if (res == -ETOOSMALL && desc->plus) {
861 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
862 nfs_zap_caches(inode);
863 desc->page_index = 0;
864 desc->plus = 0;
865 desc->eof = 0;
866 continue;
868 if (res < 0)
869 break;
871 res = nfs_do_filldir(desc);
872 if (res < 0)
873 break;
874 } while (!desc->eof);
875 out:
876 nfs_unblock_sillyrename(dentry);
877 if (res > 0)
878 res = 0;
879 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
880 return res;
883 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
885 struct inode *inode = file_inode(filp);
886 struct nfs_open_dir_context *dir_ctx = filp->private_data;
888 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
889 filp, offset, whence);
891 mutex_lock(&inode->i_mutex);
892 switch (whence) {
893 case 1:
894 offset += filp->f_pos;
895 case 0:
896 if (offset >= 0)
897 break;
898 default:
899 offset = -EINVAL;
900 goto out;
902 if (offset != filp->f_pos) {
903 filp->f_pos = offset;
904 dir_ctx->dir_cookie = 0;
905 dir_ctx->duped = 0;
907 out:
908 mutex_unlock(&inode->i_mutex);
909 return offset;
913 * All directory operations under NFS are synchronous, so fsync()
914 * is a dummy operation.
916 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
917 int datasync)
919 struct inode *inode = file_inode(filp);
921 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
923 mutex_lock(&inode->i_mutex);
924 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
925 mutex_unlock(&inode->i_mutex);
926 return 0;
930 * nfs_force_lookup_revalidate - Mark the directory as having changed
931 * @dir - pointer to directory inode
933 * This forces the revalidation code in nfs_lookup_revalidate() to do a
934 * full lookup on all child dentries of 'dir' whenever a change occurs
935 * on the server that might have invalidated our dcache.
937 * The caller should be holding dir->i_lock
939 void nfs_force_lookup_revalidate(struct inode *dir)
941 NFS_I(dir)->cache_change_attribute++;
943 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
946 * A check for whether or not the parent directory has changed.
947 * In the case it has, we assume that the dentries are untrustworthy
948 * and may need to be looked up again.
950 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
952 if (IS_ROOT(dentry))
953 return 1;
954 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
955 return 0;
956 if (!nfs_verify_change_attribute(dir, dentry->d_time))
957 return 0;
958 /* Revalidate nfsi->cache_change_attribute before we declare a match */
959 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
960 return 0;
961 if (!nfs_verify_change_attribute(dir, dentry->d_time))
962 return 0;
963 return 1;
967 * Use intent information to check whether or not we're going to do
968 * an O_EXCL create using this path component.
970 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
972 if (NFS_PROTO(dir)->version == 2)
973 return 0;
974 return flags & LOOKUP_EXCL;
978 * Inode and filehandle revalidation for lookups.
980 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
981 * or if the intent information indicates that we're about to open this
982 * particular file and the "nocto" mount flag is not set.
985 static
986 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
988 struct nfs_server *server = NFS_SERVER(inode);
989 int ret;
991 if (IS_AUTOMOUNT(inode))
992 return 0;
993 /* VFS wants an on-the-wire revalidation */
994 if (flags & LOOKUP_REVAL)
995 goto out_force;
996 /* This is an open(2) */
997 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
998 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
999 goto out_force;
1000 out:
1001 return (inode->i_nlink == 0) ? -ENOENT : 0;
1002 out_force:
1003 ret = __nfs_revalidate_inode(server, inode);
1004 if (ret != 0)
1005 return ret;
1006 goto out;
1010 * We judge how long we want to trust negative
1011 * dentries by looking at the parent inode mtime.
1013 * If parent mtime has changed, we revalidate, else we wait for a
1014 * period corresponding to the parent's attribute cache timeout value.
1016 static inline
1017 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1018 unsigned int flags)
1020 /* Don't revalidate a negative dentry if we're creating a new file */
1021 if (flags & LOOKUP_CREATE)
1022 return 0;
1023 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1024 return 1;
1025 return !nfs_check_verifier(dir, dentry);
1029 * This is called every time the dcache has a lookup hit,
1030 * and we should check whether we can really trust that
1031 * lookup.
1033 * NOTE! The hit can be a negative hit too, don't assume
1034 * we have an inode!
1036 * If the parent directory is seen to have changed, we throw out the
1037 * cached dentry and do a new lookup.
1039 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1041 struct inode *dir;
1042 struct inode *inode;
1043 struct dentry *parent;
1044 struct nfs_fh *fhandle = NULL;
1045 struct nfs_fattr *fattr = NULL;
1046 struct nfs4_label *label = NULL;
1047 int error;
1049 if (flags & LOOKUP_RCU)
1050 return -ECHILD;
1052 parent = dget_parent(dentry);
1053 dir = parent->d_inode;
1054 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1055 inode = dentry->d_inode;
1057 if (!inode) {
1058 if (nfs_neg_need_reval(dir, dentry, flags))
1059 goto out_bad;
1060 goto out_valid_noent;
1063 if (is_bad_inode(inode)) {
1064 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1065 __func__, dentry);
1066 goto out_bad;
1069 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1070 goto out_set_verifier;
1072 /* Force a full look up iff the parent directory has changed */
1073 if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1074 if (nfs_lookup_verify_inode(inode, flags))
1075 goto out_zap_parent;
1076 goto out_valid;
1079 if (NFS_STALE(inode))
1080 goto out_bad;
1082 error = -ENOMEM;
1083 fhandle = nfs_alloc_fhandle();
1084 fattr = nfs_alloc_fattr();
1085 if (fhandle == NULL || fattr == NULL)
1086 goto out_error;
1088 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1089 if (IS_ERR(label))
1090 goto out_error;
1092 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1093 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1094 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1095 if (error)
1096 goto out_bad;
1097 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1098 goto out_bad;
1099 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1100 goto out_bad;
1102 nfs_setsecurity(inode, fattr, label);
1104 nfs_free_fattr(fattr);
1105 nfs_free_fhandle(fhandle);
1106 nfs4_label_free(label);
1108 out_set_verifier:
1109 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1110 out_valid:
1111 /* Success: notify readdir to use READDIRPLUS */
1112 nfs_advise_use_readdirplus(dir);
1113 out_valid_noent:
1114 dput(parent);
1115 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1116 __func__, dentry);
1117 return 1;
1118 out_zap_parent:
1119 nfs_zap_caches(dir);
1120 out_bad:
1121 nfs_free_fattr(fattr);
1122 nfs_free_fhandle(fhandle);
1123 nfs4_label_free(label);
1124 nfs_mark_for_revalidate(dir);
1125 if (inode && S_ISDIR(inode->i_mode)) {
1126 /* Purge readdir caches. */
1127 nfs_zap_caches(inode);
1129 * We can't d_drop the root of a disconnected tree:
1130 * its d_hash is on the s_anon list and d_drop() would hide
1131 * it from shrink_dcache_for_unmount(), leading to busy
1132 * inodes on unmount and further oopses.
1134 if (IS_ROOT(dentry))
1135 goto out_valid;
1137 /* If we have submounts, don't unhash ! */
1138 if (check_submounts_and_drop(dentry) != 0)
1139 goto out_valid;
1141 dput(parent);
1142 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1143 __func__, dentry);
1144 return 0;
1145 out_error:
1146 nfs_free_fattr(fattr);
1147 nfs_free_fhandle(fhandle);
1148 nfs4_label_free(label);
1149 dput(parent);
1150 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1151 __func__, dentry, error);
1152 return error;
1156 * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1157 * when we don't really care about the dentry name. This is called when a
1158 * pathwalk ends on a dentry that was not found via a normal lookup in the
1159 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1161 * In this situation, we just want to verify that the inode itself is OK
1162 * since the dentry might have changed on the server.
1164 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1166 int error;
1167 struct inode *inode = dentry->d_inode;
1170 * I believe we can only get a negative dentry here in the case of a
1171 * procfs-style symlink. Just assume it's correct for now, but we may
1172 * eventually need to do something more here.
1174 if (!inode) {
1175 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1176 __func__, dentry);
1177 return 1;
1180 if (is_bad_inode(inode)) {
1181 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1182 __func__, dentry);
1183 return 0;
1186 error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1187 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1188 __func__, inode->i_ino, error ? "invalid" : "valid");
1189 return !error;
1193 * This is called from dput() when d_count is going to 0.
1195 static int nfs_dentry_delete(const struct dentry *dentry)
1197 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1198 dentry, dentry->d_flags);
1200 /* Unhash any dentry with a stale inode */
1201 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1202 return 1;
1204 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1205 /* Unhash it, so that ->d_iput() would be called */
1206 return 1;
1208 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1209 /* Unhash it, so that ancestors of killed async unlink
1210 * files will be cleaned up during umount */
1211 return 1;
1213 return 0;
1217 /* Ensure that we revalidate inode->i_nlink */
1218 static void nfs_drop_nlink(struct inode *inode)
1220 spin_lock(&inode->i_lock);
1221 /* drop the inode if we're reasonably sure this is the last link */
1222 if (inode->i_nlink == 1)
1223 clear_nlink(inode);
1224 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1225 spin_unlock(&inode->i_lock);
1229 * Called when the dentry loses inode.
1230 * We use it to clean up silly-renamed files.
1232 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1234 if (S_ISDIR(inode->i_mode))
1235 /* drop any readdir cache as it could easily be old */
1236 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1238 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1239 nfs_complete_unlink(dentry, inode);
1240 nfs_drop_nlink(inode);
1242 iput(inode);
1245 static void nfs_d_release(struct dentry *dentry)
1247 /* free cached devname value, if it survived that far */
1248 if (unlikely(dentry->d_fsdata)) {
1249 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1250 WARN_ON(1);
1251 else
1252 kfree(dentry->d_fsdata);
1256 const struct dentry_operations nfs_dentry_operations = {
1257 .d_revalidate = nfs_lookup_revalidate,
1258 .d_weak_revalidate = nfs_weak_revalidate,
1259 .d_delete = nfs_dentry_delete,
1260 .d_iput = nfs_dentry_iput,
1261 .d_automount = nfs_d_automount,
1262 .d_release = nfs_d_release,
1264 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1266 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1268 struct dentry *res;
1269 struct dentry *parent;
1270 struct inode *inode = NULL;
1271 struct nfs_fh *fhandle = NULL;
1272 struct nfs_fattr *fattr = NULL;
1273 struct nfs4_label *label = NULL;
1274 int error;
1276 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1277 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1279 res = ERR_PTR(-ENAMETOOLONG);
1280 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1281 goto out;
1284 * If we're doing an exclusive create, optimize away the lookup
1285 * but don't hash the dentry.
1287 if (nfs_is_exclusive_create(dir, flags)) {
1288 d_instantiate(dentry, NULL);
1289 res = NULL;
1290 goto out;
1293 res = ERR_PTR(-ENOMEM);
1294 fhandle = nfs_alloc_fhandle();
1295 fattr = nfs_alloc_fattr();
1296 if (fhandle == NULL || fattr == NULL)
1297 goto out;
1299 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1300 if (IS_ERR(label))
1301 goto out;
1303 parent = dentry->d_parent;
1304 /* Protect against concurrent sillydeletes */
1305 trace_nfs_lookup_enter(dir, dentry, flags);
1306 nfs_block_sillyrename(parent);
1307 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1308 if (error == -ENOENT)
1309 goto no_entry;
1310 if (error < 0) {
1311 res = ERR_PTR(error);
1312 goto out_unblock_sillyrename;
1314 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1315 res = ERR_CAST(inode);
1316 if (IS_ERR(res))
1317 goto out_unblock_sillyrename;
1319 /* Success: notify readdir to use READDIRPLUS */
1320 nfs_advise_use_readdirplus(dir);
1322 no_entry:
1323 res = d_materialise_unique(dentry, inode);
1324 if (res != NULL) {
1325 if (IS_ERR(res))
1326 goto out_unblock_sillyrename;
1327 dentry = res;
1329 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1330 out_unblock_sillyrename:
1331 nfs_unblock_sillyrename(parent);
1332 trace_nfs_lookup_exit(dir, dentry, flags, error);
1333 nfs4_label_free(label);
1334 out:
1335 nfs_free_fattr(fattr);
1336 nfs_free_fhandle(fhandle);
1337 return res;
1339 EXPORT_SYMBOL_GPL(nfs_lookup);
1341 #if IS_ENABLED(CONFIG_NFS_V4)
1342 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1344 const struct dentry_operations nfs4_dentry_operations = {
1345 .d_revalidate = nfs4_lookup_revalidate,
1346 .d_delete = nfs_dentry_delete,
1347 .d_iput = nfs_dentry_iput,
1348 .d_automount = nfs_d_automount,
1349 .d_release = nfs_d_release,
1351 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1353 static fmode_t flags_to_mode(int flags)
1355 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1356 if ((flags & O_ACCMODE) != O_WRONLY)
1357 res |= FMODE_READ;
1358 if ((flags & O_ACCMODE) != O_RDONLY)
1359 res |= FMODE_WRITE;
1360 return res;
1363 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1365 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1368 static int do_open(struct inode *inode, struct file *filp)
1370 nfs_fscache_open_file(inode, filp);
1371 return 0;
1374 static int nfs_finish_open(struct nfs_open_context *ctx,
1375 struct dentry *dentry,
1376 struct file *file, unsigned open_flags,
1377 int *opened)
1379 int err;
1381 if ((open_flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
1382 *opened |= FILE_CREATED;
1384 err = finish_open(file, dentry, do_open, opened);
1385 if (err)
1386 goto out;
1387 nfs_file_set_open_context(file, ctx);
1389 out:
1390 return err;
1393 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1394 struct file *file, unsigned open_flags,
1395 umode_t mode, int *opened)
1397 struct nfs_open_context *ctx;
1398 struct dentry *res;
1399 struct iattr attr = { .ia_valid = ATTR_OPEN };
1400 struct inode *inode;
1401 unsigned int lookup_flags = 0;
1402 int err;
1404 /* Expect a negative dentry */
1405 BUG_ON(dentry->d_inode);
1407 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1408 dir->i_sb->s_id, dir->i_ino, dentry);
1410 err = nfs_check_flags(open_flags);
1411 if (err)
1412 return err;
1414 /* NFS only supports OPEN on regular files */
1415 if ((open_flags & O_DIRECTORY)) {
1416 if (!d_unhashed(dentry)) {
1418 * Hashed negative dentry with O_DIRECTORY: dentry was
1419 * revalidated and is fine, no need to perform lookup
1420 * again
1422 return -ENOENT;
1424 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1425 goto no_open;
1428 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1429 return -ENAMETOOLONG;
1431 if (open_flags & O_CREAT) {
1432 attr.ia_valid |= ATTR_MODE;
1433 attr.ia_mode = mode & ~current_umask();
1435 if (open_flags & O_TRUNC) {
1436 attr.ia_valid |= ATTR_SIZE;
1437 attr.ia_size = 0;
1440 ctx = create_nfs_open_context(dentry, open_flags);
1441 err = PTR_ERR(ctx);
1442 if (IS_ERR(ctx))
1443 goto out;
1445 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1446 nfs_block_sillyrename(dentry->d_parent);
1447 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
1448 nfs_unblock_sillyrename(dentry->d_parent);
1449 if (IS_ERR(inode)) {
1450 err = PTR_ERR(inode);
1451 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1452 put_nfs_open_context(ctx);
1453 switch (err) {
1454 case -ENOENT:
1455 d_drop(dentry);
1456 d_add(dentry, NULL);
1457 break;
1458 case -EISDIR:
1459 case -ENOTDIR:
1460 goto no_open;
1461 case -ELOOP:
1462 if (!(open_flags & O_NOFOLLOW))
1463 goto no_open;
1464 break;
1465 /* case -EINVAL: */
1466 default:
1467 break;
1469 goto out;
1472 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1473 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1474 put_nfs_open_context(ctx);
1475 out:
1476 return err;
1478 no_open:
1479 res = nfs_lookup(dir, dentry, lookup_flags);
1480 err = PTR_ERR(res);
1481 if (IS_ERR(res))
1482 goto out;
1484 return finish_no_open(file, res);
1486 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1488 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1490 struct dentry *parent = NULL;
1491 struct inode *inode;
1492 struct inode *dir;
1493 int ret = 0;
1495 if (flags & LOOKUP_RCU)
1496 return -ECHILD;
1498 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1499 goto no_open;
1500 if (d_mountpoint(dentry))
1501 goto no_open;
1502 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1503 goto no_open;
1505 inode = dentry->d_inode;
1506 parent = dget_parent(dentry);
1507 dir = parent->d_inode;
1509 /* We can't create new files in nfs_open_revalidate(), so we
1510 * optimize away revalidation of negative dentries.
1512 if (inode == NULL) {
1513 if (!nfs_neg_need_reval(dir, dentry, flags))
1514 ret = 1;
1515 goto out;
1518 /* NFS only supports OPEN on regular files */
1519 if (!S_ISREG(inode->i_mode))
1520 goto no_open_dput;
1521 /* We cannot do exclusive creation on a positive dentry */
1522 if (flags & LOOKUP_EXCL)
1523 goto no_open_dput;
1525 /* Let f_op->open() actually open (and revalidate) the file */
1526 ret = 1;
1528 out:
1529 dput(parent);
1530 return ret;
1532 no_open_dput:
1533 dput(parent);
1534 no_open:
1535 return nfs_lookup_revalidate(dentry, flags);
1538 #endif /* CONFIG_NFSV4 */
1541 * Code common to create, mkdir, and mknod.
1543 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1544 struct nfs_fattr *fattr,
1545 struct nfs4_label *label)
1547 struct dentry *parent = dget_parent(dentry);
1548 struct inode *dir = parent->d_inode;
1549 struct inode *inode;
1550 int error = -EACCES;
1552 d_drop(dentry);
1554 /* We may have been initialized further down */
1555 if (dentry->d_inode)
1556 goto out;
1557 if (fhandle->size == 0) {
1558 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1559 if (error)
1560 goto out_error;
1562 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1563 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1564 struct nfs_server *server = NFS_SB(dentry->d_sb);
1565 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1566 if (error < 0)
1567 goto out_error;
1569 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1570 error = PTR_ERR(inode);
1571 if (IS_ERR(inode))
1572 goto out_error;
1573 d_add(dentry, inode);
1574 out:
1575 dput(parent);
1576 return 0;
1577 out_error:
1578 nfs_mark_for_revalidate(dir);
1579 dput(parent);
1580 return error;
1582 EXPORT_SYMBOL_GPL(nfs_instantiate);
1585 * Following a failed create operation, we drop the dentry rather
1586 * than retain a negative dentry. This avoids a problem in the event
1587 * that the operation succeeded on the server, but an error in the
1588 * reply path made it appear to have failed.
1590 int nfs_create(struct inode *dir, struct dentry *dentry,
1591 umode_t mode, bool excl)
1593 struct iattr attr;
1594 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1595 int error;
1597 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1598 dir->i_sb->s_id, dir->i_ino, dentry);
1600 attr.ia_mode = mode;
1601 attr.ia_valid = ATTR_MODE;
1603 trace_nfs_create_enter(dir, dentry, open_flags);
1604 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1605 trace_nfs_create_exit(dir, dentry, open_flags, error);
1606 if (error != 0)
1607 goto out_err;
1608 return 0;
1609 out_err:
1610 d_drop(dentry);
1611 return error;
1613 EXPORT_SYMBOL_GPL(nfs_create);
1616 * See comments for nfs_proc_create regarding failed operations.
1619 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1621 struct iattr attr;
1622 int status;
1624 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1625 dir->i_sb->s_id, dir->i_ino, dentry);
1627 if (!new_valid_dev(rdev))
1628 return -EINVAL;
1630 attr.ia_mode = mode;
1631 attr.ia_valid = ATTR_MODE;
1633 trace_nfs_mknod_enter(dir, dentry);
1634 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1635 trace_nfs_mknod_exit(dir, dentry, status);
1636 if (status != 0)
1637 goto out_err;
1638 return 0;
1639 out_err:
1640 d_drop(dentry);
1641 return status;
1643 EXPORT_SYMBOL_GPL(nfs_mknod);
1646 * See comments for nfs_proc_create regarding failed operations.
1648 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1650 struct iattr attr;
1651 int error;
1653 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1654 dir->i_sb->s_id, dir->i_ino, dentry);
1656 attr.ia_valid = ATTR_MODE;
1657 attr.ia_mode = mode | S_IFDIR;
1659 trace_nfs_mkdir_enter(dir, dentry);
1660 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1661 trace_nfs_mkdir_exit(dir, dentry, error);
1662 if (error != 0)
1663 goto out_err;
1664 return 0;
1665 out_err:
1666 d_drop(dentry);
1667 return error;
1669 EXPORT_SYMBOL_GPL(nfs_mkdir);
1671 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1673 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1674 d_delete(dentry);
1677 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1679 int error;
1681 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1682 dir->i_sb->s_id, dir->i_ino, dentry);
1684 trace_nfs_rmdir_enter(dir, dentry);
1685 if (dentry->d_inode) {
1686 nfs_wait_on_sillyrename(dentry);
1687 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1688 /* Ensure the VFS deletes this inode */
1689 switch (error) {
1690 case 0:
1691 clear_nlink(dentry->d_inode);
1692 break;
1693 case -ENOENT:
1694 nfs_dentry_handle_enoent(dentry);
1696 } else
1697 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1698 trace_nfs_rmdir_exit(dir, dentry, error);
1700 return error;
1702 EXPORT_SYMBOL_GPL(nfs_rmdir);
1705 * Remove a file after making sure there are no pending writes,
1706 * and after checking that the file has only one user.
1708 * We invalidate the attribute cache and free the inode prior to the operation
1709 * to avoid possible races if the server reuses the inode.
1711 static int nfs_safe_remove(struct dentry *dentry)
1713 struct inode *dir = dentry->d_parent->d_inode;
1714 struct inode *inode = dentry->d_inode;
1715 int error = -EBUSY;
1717 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1719 /* If the dentry was sillyrenamed, we simply call d_delete() */
1720 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1721 error = 0;
1722 goto out;
1725 trace_nfs_remove_enter(dir, dentry);
1726 if (inode != NULL) {
1727 NFS_PROTO(inode)->return_delegation(inode);
1728 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1729 if (error == 0)
1730 nfs_drop_nlink(inode);
1731 } else
1732 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1733 if (error == -ENOENT)
1734 nfs_dentry_handle_enoent(dentry);
1735 trace_nfs_remove_exit(dir, dentry, error);
1736 out:
1737 return error;
1740 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1741 * belongs to an active ".nfs..." file and we return -EBUSY.
1743 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1745 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1747 int error;
1748 int need_rehash = 0;
1750 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1751 dir->i_ino, dentry);
1753 trace_nfs_unlink_enter(dir, dentry);
1754 spin_lock(&dentry->d_lock);
1755 if (d_count(dentry) > 1) {
1756 spin_unlock(&dentry->d_lock);
1757 /* Start asynchronous writeout of the inode */
1758 write_inode_now(dentry->d_inode, 0);
1759 error = nfs_sillyrename(dir, dentry);
1760 goto out;
1762 if (!d_unhashed(dentry)) {
1763 __d_drop(dentry);
1764 need_rehash = 1;
1766 spin_unlock(&dentry->d_lock);
1767 error = nfs_safe_remove(dentry);
1768 if (!error || error == -ENOENT) {
1769 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1770 } else if (need_rehash)
1771 d_rehash(dentry);
1772 out:
1773 trace_nfs_unlink_exit(dir, dentry, error);
1774 return error;
1776 EXPORT_SYMBOL_GPL(nfs_unlink);
1779 * To create a symbolic link, most file systems instantiate a new inode,
1780 * add a page to it containing the path, then write it out to the disk
1781 * using prepare_write/commit_write.
1783 * Unfortunately the NFS client can't create the in-core inode first
1784 * because it needs a file handle to create an in-core inode (see
1785 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1786 * symlink request has completed on the server.
1788 * So instead we allocate a raw page, copy the symname into it, then do
1789 * the SYMLINK request with the page as the buffer. If it succeeds, we
1790 * now have a new file handle and can instantiate an in-core NFS inode
1791 * and move the raw page into its mapping.
1793 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1795 struct page *page;
1796 char *kaddr;
1797 struct iattr attr;
1798 unsigned int pathlen = strlen(symname);
1799 int error;
1801 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1802 dir->i_ino, dentry, symname);
1804 if (pathlen > PAGE_SIZE)
1805 return -ENAMETOOLONG;
1807 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1808 attr.ia_valid = ATTR_MODE;
1810 page = alloc_page(GFP_HIGHUSER);
1811 if (!page)
1812 return -ENOMEM;
1814 kaddr = kmap_atomic(page);
1815 memcpy(kaddr, symname, pathlen);
1816 if (pathlen < PAGE_SIZE)
1817 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1818 kunmap_atomic(kaddr);
1820 trace_nfs_symlink_enter(dir, dentry);
1821 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1822 trace_nfs_symlink_exit(dir, dentry, error);
1823 if (error != 0) {
1824 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1825 dir->i_sb->s_id, dir->i_ino,
1826 dentry, symname, error);
1827 d_drop(dentry);
1828 __free_page(page);
1829 return error;
1833 * No big deal if we can't add this page to the page cache here.
1834 * READLINK will get the missing page from the server if needed.
1836 if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0,
1837 GFP_KERNEL)) {
1838 SetPageUptodate(page);
1839 unlock_page(page);
1840 } else
1841 __free_page(page);
1843 return 0;
1845 EXPORT_SYMBOL_GPL(nfs_symlink);
1848 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1850 struct inode *inode = old_dentry->d_inode;
1851 int error;
1853 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1854 old_dentry, dentry);
1856 trace_nfs_link_enter(inode, dir, dentry);
1857 NFS_PROTO(inode)->return_delegation(inode);
1859 d_drop(dentry);
1860 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1861 if (error == 0) {
1862 ihold(inode);
1863 d_add(dentry, inode);
1865 trace_nfs_link_exit(inode, dir, dentry, error);
1866 return error;
1868 EXPORT_SYMBOL_GPL(nfs_link);
1871 * RENAME
1872 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1873 * different file handle for the same inode after a rename (e.g. when
1874 * moving to a different directory). A fail-safe method to do so would
1875 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1876 * rename the old file using the sillyrename stuff. This way, the original
1877 * file in old_dir will go away when the last process iput()s the inode.
1879 * FIXED.
1881 * It actually works quite well. One needs to have the possibility for
1882 * at least one ".nfs..." file in each directory the file ever gets
1883 * moved or linked to which happens automagically with the new
1884 * implementation that only depends on the dcache stuff instead of
1885 * using the inode layer
1887 * Unfortunately, things are a little more complicated than indicated
1888 * above. For a cross-directory move, we want to make sure we can get
1889 * rid of the old inode after the operation. This means there must be
1890 * no pending writes (if it's a file), and the use count must be 1.
1891 * If these conditions are met, we can drop the dentries before doing
1892 * the rename.
1894 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1895 struct inode *new_dir, struct dentry *new_dentry)
1897 struct inode *old_inode = old_dentry->d_inode;
1898 struct inode *new_inode = new_dentry->d_inode;
1899 struct dentry *dentry = NULL, *rehash = NULL;
1900 int error = -EBUSY;
1902 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
1903 old_dentry, new_dentry,
1904 d_count(new_dentry));
1906 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
1908 * For non-directories, check whether the target is busy and if so,
1909 * make a copy of the dentry and then do a silly-rename. If the
1910 * silly-rename succeeds, the copied dentry is hashed and becomes
1911 * the new target.
1913 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1915 * To prevent any new references to the target during the
1916 * rename, we unhash the dentry in advance.
1918 if (!d_unhashed(new_dentry)) {
1919 d_drop(new_dentry);
1920 rehash = new_dentry;
1923 if (d_count(new_dentry) > 2) {
1924 int err;
1926 /* copy the target dentry's name */
1927 dentry = d_alloc(new_dentry->d_parent,
1928 &new_dentry->d_name);
1929 if (!dentry)
1930 goto out;
1932 /* silly-rename the existing target ... */
1933 err = nfs_sillyrename(new_dir, new_dentry);
1934 if (err)
1935 goto out;
1937 new_dentry = dentry;
1938 rehash = NULL;
1939 new_inode = NULL;
1943 NFS_PROTO(old_inode)->return_delegation(old_inode);
1944 if (new_inode != NULL)
1945 NFS_PROTO(new_inode)->return_delegation(new_inode);
1947 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1948 new_dir, &new_dentry->d_name);
1949 nfs_mark_for_revalidate(old_inode);
1950 out:
1951 if (rehash)
1952 d_rehash(rehash);
1953 trace_nfs_rename_exit(old_dir, old_dentry,
1954 new_dir, new_dentry, error);
1955 if (!error) {
1956 if (new_inode != NULL)
1957 nfs_drop_nlink(new_inode);
1958 d_move(old_dentry, new_dentry);
1959 nfs_set_verifier(new_dentry,
1960 nfs_save_change_attribute(new_dir));
1961 } else if (error == -ENOENT)
1962 nfs_dentry_handle_enoent(old_dentry);
1964 /* new dentry created? */
1965 if (dentry)
1966 dput(dentry);
1967 return error;
1969 EXPORT_SYMBOL_GPL(nfs_rename);
1971 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1972 static LIST_HEAD(nfs_access_lru_list);
1973 static atomic_long_t nfs_access_nr_entries;
1975 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1977 put_rpccred(entry->cred);
1978 kfree(entry);
1979 smp_mb__before_atomic_dec();
1980 atomic_long_dec(&nfs_access_nr_entries);
1981 smp_mb__after_atomic_dec();
1984 static void nfs_access_free_list(struct list_head *head)
1986 struct nfs_access_entry *cache;
1988 while (!list_empty(head)) {
1989 cache = list_entry(head->next, struct nfs_access_entry, lru);
1990 list_del(&cache->lru);
1991 nfs_access_free_entry(cache);
1995 unsigned long
1996 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
1998 LIST_HEAD(head);
1999 struct nfs_inode *nfsi, *next;
2000 struct nfs_access_entry *cache;
2001 int nr_to_scan = sc->nr_to_scan;
2002 gfp_t gfp_mask = sc->gfp_mask;
2003 long freed = 0;
2005 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2006 return SHRINK_STOP;
2008 spin_lock(&nfs_access_lru_lock);
2009 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2010 struct inode *inode;
2012 if (nr_to_scan-- == 0)
2013 break;
2014 inode = &nfsi->vfs_inode;
2015 spin_lock(&inode->i_lock);
2016 if (list_empty(&nfsi->access_cache_entry_lru))
2017 goto remove_lru_entry;
2018 cache = list_entry(nfsi->access_cache_entry_lru.next,
2019 struct nfs_access_entry, lru);
2020 list_move(&cache->lru, &head);
2021 rb_erase(&cache->rb_node, &nfsi->access_cache);
2022 freed++;
2023 if (!list_empty(&nfsi->access_cache_entry_lru))
2024 list_move_tail(&nfsi->access_cache_inode_lru,
2025 &nfs_access_lru_list);
2026 else {
2027 remove_lru_entry:
2028 list_del_init(&nfsi->access_cache_inode_lru);
2029 smp_mb__before_clear_bit();
2030 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2031 smp_mb__after_clear_bit();
2033 spin_unlock(&inode->i_lock);
2035 spin_unlock(&nfs_access_lru_lock);
2036 nfs_access_free_list(&head);
2037 return freed;
2040 unsigned long
2041 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2043 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2046 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2048 struct rb_root *root_node = &nfsi->access_cache;
2049 struct rb_node *n;
2050 struct nfs_access_entry *entry;
2052 /* Unhook entries from the cache */
2053 while ((n = rb_first(root_node)) != NULL) {
2054 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2055 rb_erase(n, root_node);
2056 list_move(&entry->lru, head);
2058 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2061 void nfs_access_zap_cache(struct inode *inode)
2063 LIST_HEAD(head);
2065 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2066 return;
2067 /* Remove from global LRU init */
2068 spin_lock(&nfs_access_lru_lock);
2069 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2070 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2072 spin_lock(&inode->i_lock);
2073 __nfs_access_zap_cache(NFS_I(inode), &head);
2074 spin_unlock(&inode->i_lock);
2075 spin_unlock(&nfs_access_lru_lock);
2076 nfs_access_free_list(&head);
2078 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2080 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2082 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2083 struct nfs_access_entry *entry;
2085 while (n != NULL) {
2086 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2088 if (cred < entry->cred)
2089 n = n->rb_left;
2090 else if (cred > entry->cred)
2091 n = n->rb_right;
2092 else
2093 return entry;
2095 return NULL;
2098 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2100 struct nfs_inode *nfsi = NFS_I(inode);
2101 struct nfs_access_entry *cache;
2102 int err = -ENOENT;
2104 spin_lock(&inode->i_lock);
2105 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2106 goto out_zap;
2107 cache = nfs_access_search_rbtree(inode, cred);
2108 if (cache == NULL)
2109 goto out;
2110 if (!nfs_have_delegated_attributes(inode) &&
2111 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2112 goto out_stale;
2113 res->jiffies = cache->jiffies;
2114 res->cred = cache->cred;
2115 res->mask = cache->mask;
2116 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2117 err = 0;
2118 out:
2119 spin_unlock(&inode->i_lock);
2120 return err;
2121 out_stale:
2122 rb_erase(&cache->rb_node, &nfsi->access_cache);
2123 list_del(&cache->lru);
2124 spin_unlock(&inode->i_lock);
2125 nfs_access_free_entry(cache);
2126 return -ENOENT;
2127 out_zap:
2128 spin_unlock(&inode->i_lock);
2129 nfs_access_zap_cache(inode);
2130 return -ENOENT;
2133 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2135 struct nfs_inode *nfsi = NFS_I(inode);
2136 struct rb_root *root_node = &nfsi->access_cache;
2137 struct rb_node **p = &root_node->rb_node;
2138 struct rb_node *parent = NULL;
2139 struct nfs_access_entry *entry;
2141 spin_lock(&inode->i_lock);
2142 while (*p != NULL) {
2143 parent = *p;
2144 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2146 if (set->cred < entry->cred)
2147 p = &parent->rb_left;
2148 else if (set->cred > entry->cred)
2149 p = &parent->rb_right;
2150 else
2151 goto found;
2153 rb_link_node(&set->rb_node, parent, p);
2154 rb_insert_color(&set->rb_node, root_node);
2155 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2156 spin_unlock(&inode->i_lock);
2157 return;
2158 found:
2159 rb_replace_node(parent, &set->rb_node, root_node);
2160 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2161 list_del(&entry->lru);
2162 spin_unlock(&inode->i_lock);
2163 nfs_access_free_entry(entry);
2166 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2168 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2169 if (cache == NULL)
2170 return;
2171 RB_CLEAR_NODE(&cache->rb_node);
2172 cache->jiffies = set->jiffies;
2173 cache->cred = get_rpccred(set->cred);
2174 cache->mask = set->mask;
2176 nfs_access_add_rbtree(inode, cache);
2178 /* Update accounting */
2179 smp_mb__before_atomic_inc();
2180 atomic_long_inc(&nfs_access_nr_entries);
2181 smp_mb__after_atomic_inc();
2183 /* Add inode to global LRU list */
2184 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2185 spin_lock(&nfs_access_lru_lock);
2186 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2187 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2188 &nfs_access_lru_list);
2189 spin_unlock(&nfs_access_lru_lock);
2192 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2194 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2196 entry->mask = 0;
2197 if (access_result & NFS4_ACCESS_READ)
2198 entry->mask |= MAY_READ;
2199 if (access_result &
2200 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2201 entry->mask |= MAY_WRITE;
2202 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2203 entry->mask |= MAY_EXEC;
2205 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2207 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2209 struct nfs_access_entry cache;
2210 int status;
2212 trace_nfs_access_enter(inode);
2214 status = nfs_access_get_cached(inode, cred, &cache);
2215 if (status == 0)
2216 goto out_cached;
2218 /* Be clever: ask server to check for all possible rights */
2219 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2220 cache.cred = cred;
2221 cache.jiffies = jiffies;
2222 status = NFS_PROTO(inode)->access(inode, &cache);
2223 if (status != 0) {
2224 if (status == -ESTALE) {
2225 nfs_zap_caches(inode);
2226 if (!S_ISDIR(inode->i_mode))
2227 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2229 goto out;
2231 nfs_access_add_cache(inode, &cache);
2232 out_cached:
2233 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2234 status = -EACCES;
2235 out:
2236 trace_nfs_access_exit(inode, status);
2237 return status;
2240 static int nfs_open_permission_mask(int openflags)
2242 int mask = 0;
2244 if (openflags & __FMODE_EXEC) {
2245 /* ONLY check exec rights */
2246 mask = MAY_EXEC;
2247 } else {
2248 if ((openflags & O_ACCMODE) != O_WRONLY)
2249 mask |= MAY_READ;
2250 if ((openflags & O_ACCMODE) != O_RDONLY)
2251 mask |= MAY_WRITE;
2254 return mask;
2257 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2259 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2261 EXPORT_SYMBOL_GPL(nfs_may_open);
2263 int nfs_permission(struct inode *inode, int mask)
2265 struct rpc_cred *cred;
2266 int res = 0;
2268 if (mask & MAY_NOT_BLOCK)
2269 return -ECHILD;
2271 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2273 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2274 goto out;
2275 /* Is this sys_access() ? */
2276 if (mask & (MAY_ACCESS | MAY_CHDIR))
2277 goto force_lookup;
2279 switch (inode->i_mode & S_IFMT) {
2280 case S_IFLNK:
2281 goto out;
2282 case S_IFREG:
2283 break;
2284 case S_IFDIR:
2286 * Optimize away all write operations, since the server
2287 * will check permissions when we perform the op.
2289 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2290 goto out;
2293 force_lookup:
2294 if (!NFS_PROTO(inode)->access)
2295 goto out_notsup;
2297 cred = rpc_lookup_cred();
2298 if (!IS_ERR(cred)) {
2299 res = nfs_do_access(inode, cred, mask);
2300 put_rpccred(cred);
2301 } else
2302 res = PTR_ERR(cred);
2303 out:
2304 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2305 res = -EACCES;
2307 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2308 inode->i_sb->s_id, inode->i_ino, mask, res);
2309 return res;
2310 out_notsup:
2311 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2312 if (res == 0)
2313 res = generic_permission(inode, mask);
2314 goto out;
2316 EXPORT_SYMBOL_GPL(nfs_permission);
2319 * Local variables:
2320 * version-control: t
2321 * kept-new-versions: 5
2322 * End: