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>
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"
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,
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_inode *nfsi = NFS_I(dir);
73 struct nfs_open_dir_context *ctx;
74 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
77 ctx->attr_gencount = nfsi->attr_gencount;
80 ctx->cred = get_rpccred(cred);
81 spin_lock(&dir->i_lock);
82 list_add(&ctx->list, &nfsi->open_files);
83 spin_unlock(&dir->i_lock);
86 return ERR_PTR(-ENOMEM);
89 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
91 spin_lock(&dir->i_lock);
93 spin_unlock(&dir->i_lock);
94 put_rpccred(ctx->cred);
102 nfs_opendir(struct inode *inode, struct file *filp)
105 struct nfs_open_dir_context *ctx;
106 struct rpc_cred *cred;
108 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
110 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
112 cred = rpc_lookup_cred();
114 return PTR_ERR(cred);
115 ctx = alloc_nfs_open_dir_context(inode, cred);
120 filp->private_data = ctx;
121 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
122 /* This is a mountpoint, so d_revalidate will never
123 * have been called, so we need to refresh the
124 * inode (for close-open consistency) ourselves.
126 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
134 nfs_closedir(struct inode *inode, struct file *filp)
136 put_nfs_open_dir_context(filp->f_path.dentry->d_inode, filp->private_data);
140 struct nfs_cache_array_entry {
144 unsigned char d_type;
147 struct nfs_cache_array {
151 struct nfs_cache_array_entry array[0];
154 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
158 struct dir_context *ctx;
159 unsigned long page_index;
162 loff_t current_index;
163 decode_dirent_t decode;
165 unsigned long timestamp;
166 unsigned long gencount;
167 unsigned int cache_entry_index;
170 } nfs_readdir_descriptor_t;
173 * The caller is responsible for calling nfs_readdir_release_array(page)
176 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
180 return ERR_PTR(-EIO);
183 return ERR_PTR(-ENOMEM);
188 void nfs_readdir_release_array(struct page *page)
194 * we are freeing strings created by nfs_add_to_readdir_array()
197 void nfs_readdir_clear_array(struct page *page)
199 struct nfs_cache_array *array;
202 array = kmap_atomic(page);
203 for (i = 0; i < array->size; i++)
204 kfree(array->array[i].string.name);
205 kunmap_atomic(array);
209 * the caller is responsible for freeing qstr.name
210 * when called by nfs_readdir_add_to_array, the strings will be freed in
211 * nfs_clear_readdir_array()
214 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
217 string->name = kmemdup(name, len, GFP_KERNEL);
218 if (string->name == NULL)
221 * Avoid a kmemleak false positive. The pointer to the name is stored
222 * in a page cache page which kmemleak does not scan.
224 kmemleak_not_leak(string->name);
225 string->hash = full_name_hash(name, len);
230 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
232 struct nfs_cache_array *array = nfs_readdir_get_array(page);
233 struct nfs_cache_array_entry *cache_entry;
237 return PTR_ERR(array);
239 cache_entry = &array->array[array->size];
241 /* Check that this entry lies within the page bounds */
243 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
246 cache_entry->cookie = entry->prev_cookie;
247 cache_entry->ino = entry->ino;
248 cache_entry->d_type = entry->d_type;
249 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
252 array->last_cookie = entry->cookie;
255 array->eof_index = array->size;
257 nfs_readdir_release_array(page);
262 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
264 loff_t diff = desc->ctx->pos - desc->current_index;
269 if (diff >= array->size) {
270 if (array->eof_index >= 0)
275 index = (unsigned int)diff;
276 *desc->dir_cookie = array->array[index].cookie;
277 desc->cache_entry_index = index;
285 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
287 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
290 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
294 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
298 int status = -EAGAIN;
300 for (i = 0; i < array->size; i++) {
301 if (array->array[i].cookie == *desc->dir_cookie) {
302 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
303 struct nfs_open_dir_context *ctx = desc->file->private_data;
305 new_pos = desc->current_index + i;
306 if (ctx->attr_gencount != nfsi->attr_gencount ||
307 !nfs_readdir_inode_mapping_valid(nfsi)) {
309 ctx->attr_gencount = nfsi->attr_gencount;
310 } else if (new_pos < desc->ctx->pos) {
312 && ctx->dup_cookie == *desc->dir_cookie) {
313 if (printk_ratelimit()) {
314 pr_notice("NFS: directory %pD2 contains a readdir loop."
315 "Please contact your server vendor. "
316 "The file: %.*s has duplicate cookie %llu\n",
317 desc->file, array->array[i].string.len,
318 array->array[i].string.name, *desc->dir_cookie);
323 ctx->dup_cookie = *desc->dir_cookie;
326 desc->ctx->pos = new_pos;
327 desc->cache_entry_index = i;
331 if (array->eof_index >= 0) {
332 status = -EBADCOOKIE;
333 if (*desc->dir_cookie == array->last_cookie)
341 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
343 struct nfs_cache_array *array;
346 array = nfs_readdir_get_array(desc->page);
348 status = PTR_ERR(array);
352 if (*desc->dir_cookie == 0)
353 status = nfs_readdir_search_for_pos(array, desc);
355 status = nfs_readdir_search_for_cookie(array, desc);
357 if (status == -EAGAIN) {
358 desc->last_cookie = array->last_cookie;
359 desc->current_index += array->size;
362 nfs_readdir_release_array(desc->page);
367 /* Fill a page with xdr information before transferring to the cache page */
369 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
370 struct nfs_entry *entry, struct file *file, struct inode *inode)
372 struct nfs_open_dir_context *ctx = file->private_data;
373 struct rpc_cred *cred = ctx->cred;
374 unsigned long timestamp, gencount;
379 gencount = nfs_inc_attr_generation_counter();
380 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
381 NFS_SERVER(inode)->dtsize, desc->plus);
383 /* We requested READDIRPLUS, but the server doesn't grok it */
384 if (error == -ENOTSUPP && desc->plus) {
385 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
386 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
392 desc->timestamp = timestamp;
393 desc->gencount = gencount;
398 static int xdr_decode(nfs_readdir_descriptor_t *desc,
399 struct nfs_entry *entry, struct xdr_stream *xdr)
403 error = desc->decode(xdr, entry, desc->plus);
406 entry->fattr->time_start = desc->timestamp;
407 entry->fattr->gencount = desc->gencount;
412 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
414 if (dentry->d_inode == NULL)
416 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
424 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
426 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
428 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
436 * This function is called by the lookup code to request the use of
437 * readdirplus to accelerate any future lookups in the same
441 void nfs_advise_use_readdirplus(struct inode *dir)
443 set_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags);
447 * This function is mainly for use by nfs_getattr().
449 * If this is an 'ls -l', we want to force use of readdirplus.
450 * Do this by checking if there is an active file descriptor
451 * and calling nfs_advise_use_readdirplus, then forcing a
454 void nfs_force_use_readdirplus(struct inode *dir)
456 if (!list_empty(&NFS_I(dir)->open_files)) {
457 nfs_advise_use_readdirplus(dir);
458 nfs_zap_mapping(dir, dir->i_mapping);
463 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
465 struct qstr filename = QSTR_INIT(entry->name, entry->len);
466 struct dentry *dentry;
467 struct dentry *alias;
468 struct inode *dir = parent->d_inode;
472 if (filename.name[0] == '.') {
473 if (filename.len == 1)
475 if (filename.len == 2 && filename.name[1] == '.')
478 filename.hash = full_name_hash(filename.name, filename.len);
480 dentry = d_lookup(parent, &filename);
481 if (dentry != NULL) {
482 if (nfs_same_file(dentry, entry)) {
483 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
484 status = nfs_refresh_inode(dentry->d_inode, entry->fattr);
486 nfs_setsecurity(dentry->d_inode, entry->fattr, entry->label);
489 if (d_invalidate(dentry) != 0)
495 dentry = d_alloc(parent, &filename);
499 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
503 alias = d_materialise_unique(dentry, inode);
507 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
510 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
516 /* Perform conversion from xdr to cache array */
518 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
519 struct page **xdr_pages, struct page *page, unsigned int buflen)
521 struct xdr_stream stream;
523 struct page *scratch;
524 struct nfs_cache_array *array;
525 unsigned int count = 0;
528 scratch = alloc_page(GFP_KERNEL);
532 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
533 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
536 status = xdr_decode(desc, entry, &stream);
538 if (status == -EAGAIN)
546 nfs_prime_dcache(desc->file->f_path.dentry, entry);
548 status = nfs_readdir_add_to_array(entry, page);
551 } while (!entry->eof);
553 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
554 array = nfs_readdir_get_array(page);
555 if (!IS_ERR(array)) {
556 array->eof_index = array->size;
558 nfs_readdir_release_array(page);
560 status = PTR_ERR(array);
568 void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
571 for (i = 0; i < npages; i++)
576 void nfs_readdir_free_large_page(void *ptr, struct page **pages,
579 nfs_readdir_free_pagearray(pages, npages);
583 * nfs_readdir_large_page will allocate pages that must be freed with a call
584 * to nfs_readdir_free_large_page
587 int nfs_readdir_large_page(struct page **pages, unsigned int npages)
591 for (i = 0; i < npages; i++) {
592 struct page *page = alloc_page(GFP_KERNEL);
600 nfs_readdir_free_pagearray(pages, i);
605 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
607 struct page *pages[NFS_MAX_READDIR_PAGES];
608 void *pages_ptr = NULL;
609 struct nfs_entry entry;
610 struct file *file = desc->file;
611 struct nfs_cache_array *array;
612 int status = -ENOMEM;
613 unsigned int array_size = ARRAY_SIZE(pages);
615 entry.prev_cookie = 0;
616 entry.cookie = desc->last_cookie;
618 entry.fh = nfs_alloc_fhandle();
619 entry.fattr = nfs_alloc_fattr();
620 entry.server = NFS_SERVER(inode);
621 if (entry.fh == NULL || entry.fattr == NULL)
624 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
625 if (IS_ERR(entry.label)) {
626 status = PTR_ERR(entry.label);
630 array = nfs_readdir_get_array(page);
632 status = PTR_ERR(array);
635 memset(array, 0, sizeof(struct nfs_cache_array));
636 array->eof_index = -1;
638 status = nfs_readdir_large_page(pages, array_size);
640 goto out_release_array;
643 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
648 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
650 if (status == -ENOSPC)
654 } while (array->eof_index < 0);
656 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
658 nfs_readdir_release_array(page);
660 nfs4_label_free(entry.label);
662 nfs_free_fattr(entry.fattr);
663 nfs_free_fhandle(entry.fh);
668 * Now we cache directories properly, by converting xdr information
669 * to an array that can be used for lookups later. This results in
670 * fewer cache pages, since we can store more information on each page.
671 * We only need to convert from xdr once so future lookups are much simpler
674 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
676 struct inode *inode = file_inode(desc->file);
679 ret = nfs_readdir_xdr_to_array(desc, page, inode);
682 SetPageUptodate(page);
684 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
685 /* Should never happen */
686 nfs_zap_mapping(inode, inode->i_mapping);
696 void cache_page_release(nfs_readdir_descriptor_t *desc)
698 if (!desc->page->mapping)
699 nfs_readdir_clear_array(desc->page);
700 page_cache_release(desc->page);
705 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
707 return read_cache_page(file_inode(desc->file)->i_mapping,
708 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
712 * Returns 0 if desc->dir_cookie was found on page desc->page_index
715 int find_cache_page(nfs_readdir_descriptor_t *desc)
719 desc->page = get_cache_page(desc);
720 if (IS_ERR(desc->page))
721 return PTR_ERR(desc->page);
723 res = nfs_readdir_search_array(desc);
725 cache_page_release(desc);
729 /* Search for desc->dir_cookie from the beginning of the page cache */
731 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
735 if (desc->page_index == 0) {
736 desc->current_index = 0;
737 desc->last_cookie = 0;
740 res = find_cache_page(desc);
741 } while (res == -EAGAIN);
746 * Once we've found the start of the dirent within a page: fill 'er up...
749 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
751 struct file *file = desc->file;
754 struct nfs_cache_array *array = NULL;
755 struct nfs_open_dir_context *ctx = file->private_data;
757 array = nfs_readdir_get_array(desc->page);
759 res = PTR_ERR(array);
763 for (i = desc->cache_entry_index; i < array->size; i++) {
764 struct nfs_cache_array_entry *ent;
766 ent = &array->array[i];
767 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
768 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
773 if (i < (array->size-1))
774 *desc->dir_cookie = array->array[i+1].cookie;
776 *desc->dir_cookie = array->last_cookie;
780 if (array->eof_index >= 0)
783 nfs_readdir_release_array(desc->page);
785 cache_page_release(desc);
786 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
787 (unsigned long long)*desc->dir_cookie, res);
792 * If we cannot find a cookie in our cache, we suspect that this is
793 * because it points to a deleted file, so we ask the server to return
794 * whatever it thinks is the next entry. We then feed this to filldir.
795 * If all goes well, we should then be able to find our way round the
796 * cache on the next call to readdir_search_pagecache();
798 * NOTE: we cannot add the anonymous page to the pagecache because
799 * the data it contains might not be page aligned. Besides,
800 * we should already have a complete representation of the
801 * directory in the page cache by the time we get here.
804 int uncached_readdir(nfs_readdir_descriptor_t *desc)
806 struct page *page = NULL;
808 struct inode *inode = file_inode(desc->file);
809 struct nfs_open_dir_context *ctx = desc->file->private_data;
811 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
812 (unsigned long long)*desc->dir_cookie);
814 page = alloc_page(GFP_HIGHUSER);
820 desc->page_index = 0;
821 desc->last_cookie = *desc->dir_cookie;
825 status = nfs_readdir_xdr_to_array(desc, page, inode);
829 status = nfs_do_filldir(desc);
832 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
836 cache_page_release(desc);
840 static bool nfs_dir_mapping_need_revalidate(struct inode *dir)
842 struct nfs_inode *nfsi = NFS_I(dir);
844 if (nfs_attribute_cache_expired(dir))
846 if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
851 /* The file offset position represents the dirent entry number. A
852 last cookie cache takes care of the common case of reading the
855 static int nfs_readdir(struct file *file, struct dir_context *ctx)
857 struct dentry *dentry = file->f_path.dentry;
858 struct inode *inode = dentry->d_inode;
859 nfs_readdir_descriptor_t my_desc,
861 struct nfs_open_dir_context *dir_ctx = file->private_data;
864 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
865 file, (long long)ctx->pos);
866 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
869 * ctx->pos points to the dirent entry number.
870 * *desc->dir_cookie has the cookie for the next entry. We have
871 * to either find the entry with the appropriate number or
872 * revalidate the cookie.
874 memset(desc, 0, sizeof(*desc));
878 desc->dir_cookie = &dir_ctx->dir_cookie;
879 desc->decode = NFS_PROTO(inode)->decode_dirent;
880 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
882 nfs_block_sillyrename(dentry);
883 if (ctx->pos == 0 || nfs_dir_mapping_need_revalidate(inode))
884 res = nfs_revalidate_mapping(inode, file->f_mapping);
889 res = readdir_search_pagecache(desc);
891 if (res == -EBADCOOKIE) {
893 /* This means either end of directory */
894 if (*desc->dir_cookie && desc->eof == 0) {
895 /* Or that the server has 'lost' a cookie */
896 res = uncached_readdir(desc);
902 if (res == -ETOOSMALL && desc->plus) {
903 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
904 nfs_zap_caches(inode);
905 desc->page_index = 0;
913 res = nfs_do_filldir(desc);
916 } while (!desc->eof);
918 nfs_unblock_sillyrename(dentry);
921 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
925 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
927 struct inode *inode = file_inode(filp);
928 struct nfs_open_dir_context *dir_ctx = filp->private_data;
930 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
931 filp, offset, whence);
933 mutex_lock(&inode->i_mutex);
936 offset += filp->f_pos;
944 if (offset != filp->f_pos) {
945 filp->f_pos = offset;
946 dir_ctx->dir_cookie = 0;
950 mutex_unlock(&inode->i_mutex);
955 * All directory operations under NFS are synchronous, so fsync()
956 * is a dummy operation.
958 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
961 struct inode *inode = file_inode(filp);
963 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
965 mutex_lock(&inode->i_mutex);
966 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
967 mutex_unlock(&inode->i_mutex);
972 * nfs_force_lookup_revalidate - Mark the directory as having changed
973 * @dir - pointer to directory inode
975 * This forces the revalidation code in nfs_lookup_revalidate() to do a
976 * full lookup on all child dentries of 'dir' whenever a change occurs
977 * on the server that might have invalidated our dcache.
979 * The caller should be holding dir->i_lock
981 void nfs_force_lookup_revalidate(struct inode *dir)
983 NFS_I(dir)->cache_change_attribute++;
985 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
988 * A check for whether or not the parent directory has changed.
989 * In the case it has, we assume that the dentries are untrustworthy
990 * and may need to be looked up again.
992 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
996 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
998 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1000 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1001 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1003 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1009 * Use intent information to check whether or not we're going to do
1010 * an O_EXCL create using this path component.
1012 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1014 if (NFS_PROTO(dir)->version == 2)
1016 return flags & LOOKUP_EXCL;
1020 * Inode and filehandle revalidation for lookups.
1022 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1023 * or if the intent information indicates that we're about to open this
1024 * particular file and the "nocto" mount flag is not set.
1028 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1030 struct nfs_server *server = NFS_SERVER(inode);
1033 if (IS_AUTOMOUNT(inode))
1035 /* VFS wants an on-the-wire revalidation */
1036 if (flags & LOOKUP_REVAL)
1038 /* This is an open(2) */
1039 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1040 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1043 return (inode->i_nlink == 0) ? -ENOENT : 0;
1045 ret = __nfs_revalidate_inode(server, inode);
1052 * We judge how long we want to trust negative
1053 * dentries by looking at the parent inode mtime.
1055 * If parent mtime has changed, we revalidate, else we wait for a
1056 * period corresponding to the parent's attribute cache timeout value.
1059 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1062 /* Don't revalidate a negative dentry if we're creating a new file */
1063 if (flags & LOOKUP_CREATE)
1065 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1067 return !nfs_check_verifier(dir, dentry);
1071 * This is called every time the dcache has a lookup hit,
1072 * and we should check whether we can really trust that
1075 * NOTE! The hit can be a negative hit too, don't assume
1078 * If the parent directory is seen to have changed, we throw out the
1079 * cached dentry and do a new lookup.
1081 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1084 struct inode *inode;
1085 struct dentry *parent;
1086 struct nfs_fh *fhandle = NULL;
1087 struct nfs_fattr *fattr = NULL;
1088 struct nfs4_label *label = NULL;
1091 if (flags & LOOKUP_RCU)
1094 parent = dget_parent(dentry);
1095 dir = parent->d_inode;
1096 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1097 inode = dentry->d_inode;
1100 if (nfs_neg_need_reval(dir, dentry, flags))
1102 goto out_valid_noent;
1105 if (is_bad_inode(inode)) {
1106 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1111 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1112 goto out_set_verifier;
1114 /* Force a full look up iff the parent directory has changed */
1115 if (!nfs_is_exclusive_create(dir, flags) && nfs_check_verifier(dir, dentry)) {
1116 if (nfs_lookup_verify_inode(inode, flags))
1117 goto out_zap_parent;
1121 if (NFS_STALE(inode))
1125 fhandle = nfs_alloc_fhandle();
1126 fattr = nfs_alloc_fattr();
1127 if (fhandle == NULL || fattr == NULL)
1130 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1134 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1135 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1136 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1139 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1141 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1144 nfs_setsecurity(inode, fattr, label);
1146 nfs_free_fattr(fattr);
1147 nfs_free_fhandle(fhandle);
1148 nfs4_label_free(label);
1151 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1153 /* Success: notify readdir to use READDIRPLUS */
1154 nfs_advise_use_readdirplus(dir);
1157 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1161 nfs_zap_caches(dir);
1163 nfs_free_fattr(fattr);
1164 nfs_free_fhandle(fhandle);
1165 nfs4_label_free(label);
1166 nfs_mark_for_revalidate(dir);
1167 if (inode && S_ISDIR(inode->i_mode)) {
1168 /* Purge readdir caches. */
1169 nfs_zap_caches(inode);
1171 * We can't d_drop the root of a disconnected tree:
1172 * its d_hash is on the s_anon list and d_drop() would hide
1173 * it from shrink_dcache_for_unmount(), leading to busy
1174 * inodes on unmount and further oopses.
1176 if (IS_ROOT(dentry))
1179 /* If we have submounts, don't unhash ! */
1180 if (check_submounts_and_drop(dentry) != 0)
1184 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1188 nfs_free_fattr(fattr);
1189 nfs_free_fhandle(fhandle);
1190 nfs4_label_free(label);
1192 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1193 __func__, dentry, error);
1198 * A weaker form of d_revalidate for revalidating just the dentry->d_inode
1199 * when we don't really care about the dentry name. This is called when a
1200 * pathwalk ends on a dentry that was not found via a normal lookup in the
1201 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1203 * In this situation, we just want to verify that the inode itself is OK
1204 * since the dentry might have changed on the server.
1206 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1209 struct inode *inode = dentry->d_inode;
1212 * I believe we can only get a negative dentry here in the case of a
1213 * procfs-style symlink. Just assume it's correct for now, but we may
1214 * eventually need to do something more here.
1217 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1222 if (is_bad_inode(inode)) {
1223 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1228 error = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1229 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1230 __func__, inode->i_ino, error ? "invalid" : "valid");
1235 * This is called from dput() when d_count is going to 0.
1237 static int nfs_dentry_delete(const struct dentry *dentry)
1239 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1240 dentry, dentry->d_flags);
1242 /* Unhash any dentry with a stale inode */
1243 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1246 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1247 /* Unhash it, so that ->d_iput() would be called */
1250 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1251 /* Unhash it, so that ancestors of killed async unlink
1252 * files will be cleaned up during umount */
1259 /* Ensure that we revalidate inode->i_nlink */
1260 static void nfs_drop_nlink(struct inode *inode)
1262 spin_lock(&inode->i_lock);
1263 /* drop the inode if we're reasonably sure this is the last link */
1264 if (inode->i_nlink == 1)
1266 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1267 spin_unlock(&inode->i_lock);
1271 * Called when the dentry loses inode.
1272 * We use it to clean up silly-renamed files.
1274 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1276 if (S_ISDIR(inode->i_mode))
1277 /* drop any readdir cache as it could easily be old */
1278 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1280 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1281 nfs_complete_unlink(dentry, inode);
1282 nfs_drop_nlink(inode);
1287 static void nfs_d_release(struct dentry *dentry)
1289 /* free cached devname value, if it survived that far */
1290 if (unlikely(dentry->d_fsdata)) {
1291 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1294 kfree(dentry->d_fsdata);
1298 const struct dentry_operations nfs_dentry_operations = {
1299 .d_revalidate = nfs_lookup_revalidate,
1300 .d_weak_revalidate = nfs_weak_revalidate,
1301 .d_delete = nfs_dentry_delete,
1302 .d_iput = nfs_dentry_iput,
1303 .d_automount = nfs_d_automount,
1304 .d_release = nfs_d_release,
1306 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1308 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1311 struct dentry *parent;
1312 struct inode *inode = NULL;
1313 struct nfs_fh *fhandle = NULL;
1314 struct nfs_fattr *fattr = NULL;
1315 struct nfs4_label *label = NULL;
1318 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1319 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1321 res = ERR_PTR(-ENAMETOOLONG);
1322 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1326 * If we're doing an exclusive create, optimize away the lookup
1327 * but don't hash the dentry.
1329 if (nfs_is_exclusive_create(dir, flags)) {
1330 d_instantiate(dentry, NULL);
1335 res = ERR_PTR(-ENOMEM);
1336 fhandle = nfs_alloc_fhandle();
1337 fattr = nfs_alloc_fattr();
1338 if (fhandle == NULL || fattr == NULL)
1341 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1345 parent = dentry->d_parent;
1346 /* Protect against concurrent sillydeletes */
1347 trace_nfs_lookup_enter(dir, dentry, flags);
1348 nfs_block_sillyrename(parent);
1349 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1350 if (error == -ENOENT)
1353 res = ERR_PTR(error);
1354 goto out_unblock_sillyrename;
1356 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1357 res = ERR_CAST(inode);
1359 goto out_unblock_sillyrename;
1361 /* Success: notify readdir to use READDIRPLUS */
1362 nfs_advise_use_readdirplus(dir);
1365 res = d_materialise_unique(dentry, inode);
1368 goto out_unblock_sillyrename;
1371 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1372 out_unblock_sillyrename:
1373 nfs_unblock_sillyrename(parent);
1374 trace_nfs_lookup_exit(dir, dentry, flags, error);
1375 nfs4_label_free(label);
1377 nfs_free_fattr(fattr);
1378 nfs_free_fhandle(fhandle);
1381 EXPORT_SYMBOL_GPL(nfs_lookup);
1383 #if IS_ENABLED(CONFIG_NFS_V4)
1384 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1386 const struct dentry_operations nfs4_dentry_operations = {
1387 .d_revalidate = nfs4_lookup_revalidate,
1388 .d_delete = nfs_dentry_delete,
1389 .d_iput = nfs_dentry_iput,
1390 .d_automount = nfs_d_automount,
1391 .d_release = nfs_d_release,
1393 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1395 static fmode_t flags_to_mode(int flags)
1397 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1398 if ((flags & O_ACCMODE) != O_WRONLY)
1400 if ((flags & O_ACCMODE) != O_RDONLY)
1405 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
1407 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags));
1410 static int do_open(struct inode *inode, struct file *filp)
1412 nfs_fscache_open_file(inode, filp);
1416 static int nfs_finish_open(struct nfs_open_context *ctx,
1417 struct dentry *dentry,
1418 struct file *file, unsigned open_flags,
1423 if ((open_flags & (O_CREAT | O_EXCL)) == (O_CREAT | O_EXCL))
1424 *opened |= FILE_CREATED;
1426 err = finish_open(file, dentry, do_open, opened);
1429 nfs_file_set_open_context(file, ctx);
1435 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1436 struct file *file, unsigned open_flags,
1437 umode_t mode, int *opened)
1439 struct nfs_open_context *ctx;
1441 struct iattr attr = { .ia_valid = ATTR_OPEN };
1442 struct inode *inode;
1443 unsigned int lookup_flags = 0;
1446 /* Expect a negative dentry */
1447 BUG_ON(dentry->d_inode);
1449 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1450 dir->i_sb->s_id, dir->i_ino, dentry);
1452 err = nfs_check_flags(open_flags);
1456 /* NFS only supports OPEN on regular files */
1457 if ((open_flags & O_DIRECTORY)) {
1458 if (!d_unhashed(dentry)) {
1460 * Hashed negative dentry with O_DIRECTORY: dentry was
1461 * revalidated and is fine, no need to perform lookup
1466 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1470 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1471 return -ENAMETOOLONG;
1473 if (open_flags & O_CREAT) {
1474 attr.ia_valid |= ATTR_MODE;
1475 attr.ia_mode = mode & ~current_umask();
1477 if (open_flags & O_TRUNC) {
1478 attr.ia_valid |= ATTR_SIZE;
1482 ctx = create_nfs_open_context(dentry, open_flags);
1487 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1488 nfs_block_sillyrename(dentry->d_parent);
1489 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
1490 nfs_unblock_sillyrename(dentry->d_parent);
1491 if (IS_ERR(inode)) {
1492 err = PTR_ERR(inode);
1493 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1494 put_nfs_open_context(ctx);
1498 d_add(dentry, NULL);
1504 if (!(open_flags & O_NOFOLLOW))
1514 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1515 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1516 put_nfs_open_context(ctx);
1521 res = nfs_lookup(dir, dentry, lookup_flags);
1526 return finish_no_open(file, res);
1528 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1530 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1532 struct dentry *parent = NULL;
1533 struct inode *inode;
1537 if (flags & LOOKUP_RCU)
1540 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1542 if (d_mountpoint(dentry))
1544 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1547 inode = dentry->d_inode;
1548 parent = dget_parent(dentry);
1549 dir = parent->d_inode;
1551 /* We can't create new files in nfs_open_revalidate(), so we
1552 * optimize away revalidation of negative dentries.
1554 if (inode == NULL) {
1555 if (!nfs_neg_need_reval(dir, dentry, flags))
1560 /* NFS only supports OPEN on regular files */
1561 if (!S_ISREG(inode->i_mode))
1563 /* We cannot do exclusive creation on a positive dentry */
1564 if (flags & LOOKUP_EXCL)
1567 /* Let f_op->open() actually open (and revalidate) the file */
1577 return nfs_lookup_revalidate(dentry, flags);
1580 #endif /* CONFIG_NFSV4 */
1583 * Code common to create, mkdir, and mknod.
1585 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1586 struct nfs_fattr *fattr,
1587 struct nfs4_label *label)
1589 struct dentry *parent = dget_parent(dentry);
1590 struct inode *dir = parent->d_inode;
1591 struct inode *inode;
1592 int error = -EACCES;
1596 /* We may have been initialized further down */
1597 if (dentry->d_inode)
1599 if (fhandle->size == 0) {
1600 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1604 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1605 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1606 struct nfs_server *server = NFS_SB(dentry->d_sb);
1607 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1611 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1612 error = PTR_ERR(inode);
1615 d_add(dentry, inode);
1620 nfs_mark_for_revalidate(dir);
1624 EXPORT_SYMBOL_GPL(nfs_instantiate);
1627 * Following a failed create operation, we drop the dentry rather
1628 * than retain a negative dentry. This avoids a problem in the event
1629 * that the operation succeeded on the server, but an error in the
1630 * reply path made it appear to have failed.
1632 int nfs_create(struct inode *dir, struct dentry *dentry,
1633 umode_t mode, bool excl)
1636 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1639 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1640 dir->i_sb->s_id, dir->i_ino, dentry);
1642 attr.ia_mode = mode;
1643 attr.ia_valid = ATTR_MODE;
1645 trace_nfs_create_enter(dir, dentry, open_flags);
1646 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1647 trace_nfs_create_exit(dir, dentry, open_flags, error);
1655 EXPORT_SYMBOL_GPL(nfs_create);
1658 * See comments for nfs_proc_create regarding failed operations.
1661 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1666 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1667 dir->i_sb->s_id, dir->i_ino, dentry);
1669 if (!new_valid_dev(rdev))
1672 attr.ia_mode = mode;
1673 attr.ia_valid = ATTR_MODE;
1675 trace_nfs_mknod_enter(dir, dentry);
1676 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1677 trace_nfs_mknod_exit(dir, dentry, status);
1685 EXPORT_SYMBOL_GPL(nfs_mknod);
1688 * See comments for nfs_proc_create regarding failed operations.
1690 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1695 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1696 dir->i_sb->s_id, dir->i_ino, dentry);
1698 attr.ia_valid = ATTR_MODE;
1699 attr.ia_mode = mode | S_IFDIR;
1701 trace_nfs_mkdir_enter(dir, dentry);
1702 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1703 trace_nfs_mkdir_exit(dir, dentry, error);
1711 EXPORT_SYMBOL_GPL(nfs_mkdir);
1713 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1715 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1719 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1723 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1724 dir->i_sb->s_id, dir->i_ino, dentry);
1726 trace_nfs_rmdir_enter(dir, dentry);
1727 if (dentry->d_inode) {
1728 nfs_wait_on_sillyrename(dentry);
1729 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1730 /* Ensure the VFS deletes this inode */
1733 clear_nlink(dentry->d_inode);
1736 nfs_dentry_handle_enoent(dentry);
1739 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1740 trace_nfs_rmdir_exit(dir, dentry, error);
1744 EXPORT_SYMBOL_GPL(nfs_rmdir);
1747 * Remove a file after making sure there are no pending writes,
1748 * and after checking that the file has only one user.
1750 * We invalidate the attribute cache and free the inode prior to the operation
1751 * to avoid possible races if the server reuses the inode.
1753 static int nfs_safe_remove(struct dentry *dentry)
1755 struct inode *dir = dentry->d_parent->d_inode;
1756 struct inode *inode = dentry->d_inode;
1759 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1761 /* If the dentry was sillyrenamed, we simply call d_delete() */
1762 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1767 trace_nfs_remove_enter(dir, dentry);
1768 if (inode != NULL) {
1769 NFS_PROTO(inode)->return_delegation(inode);
1770 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1772 nfs_drop_nlink(inode);
1774 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1775 if (error == -ENOENT)
1776 nfs_dentry_handle_enoent(dentry);
1777 trace_nfs_remove_exit(dir, dentry, error);
1782 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1783 * belongs to an active ".nfs..." file and we return -EBUSY.
1785 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1787 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1790 int need_rehash = 0;
1792 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1793 dir->i_ino, dentry);
1795 trace_nfs_unlink_enter(dir, dentry);
1796 spin_lock(&dentry->d_lock);
1797 if (d_count(dentry) > 1) {
1798 spin_unlock(&dentry->d_lock);
1799 /* Start asynchronous writeout of the inode */
1800 write_inode_now(dentry->d_inode, 0);
1801 error = nfs_sillyrename(dir, dentry);
1804 if (!d_unhashed(dentry)) {
1808 spin_unlock(&dentry->d_lock);
1809 error = nfs_safe_remove(dentry);
1810 if (!error || error == -ENOENT) {
1811 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1812 } else if (need_rehash)
1815 trace_nfs_unlink_exit(dir, dentry, error);
1818 EXPORT_SYMBOL_GPL(nfs_unlink);
1821 * To create a symbolic link, most file systems instantiate a new inode,
1822 * add a page to it containing the path, then write it out to the disk
1823 * using prepare_write/commit_write.
1825 * Unfortunately the NFS client can't create the in-core inode first
1826 * because it needs a file handle to create an in-core inode (see
1827 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1828 * symlink request has completed on the server.
1830 * So instead we allocate a raw page, copy the symname into it, then do
1831 * the SYMLINK request with the page as the buffer. If it succeeds, we
1832 * now have a new file handle and can instantiate an in-core NFS inode
1833 * and move the raw page into its mapping.
1835 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1840 unsigned int pathlen = strlen(symname);
1843 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1844 dir->i_ino, dentry, symname);
1846 if (pathlen > PAGE_SIZE)
1847 return -ENAMETOOLONG;
1849 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1850 attr.ia_valid = ATTR_MODE;
1852 page = alloc_page(GFP_HIGHUSER);
1856 kaddr = kmap_atomic(page);
1857 memcpy(kaddr, symname, pathlen);
1858 if (pathlen < PAGE_SIZE)
1859 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1860 kunmap_atomic(kaddr);
1862 trace_nfs_symlink_enter(dir, dentry);
1863 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1864 trace_nfs_symlink_exit(dir, dentry, error);
1866 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1867 dir->i_sb->s_id, dir->i_ino,
1868 dentry, symname, error);
1875 * No big deal if we can't add this page to the page cache here.
1876 * READLINK will get the missing page from the server if needed.
1878 if (!add_to_page_cache_lru(page, dentry->d_inode->i_mapping, 0,
1880 SetPageUptodate(page);
1883 * add_to_page_cache_lru() grabs an extra page refcount.
1884 * Drop it here to avoid leaking this page later.
1886 page_cache_release(page);
1892 EXPORT_SYMBOL_GPL(nfs_symlink);
1895 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1897 struct inode *inode = old_dentry->d_inode;
1900 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1901 old_dentry, dentry);
1903 trace_nfs_link_enter(inode, dir, dentry);
1904 NFS_PROTO(inode)->return_delegation(inode);
1907 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1910 d_add(dentry, inode);
1912 trace_nfs_link_exit(inode, dir, dentry, error);
1915 EXPORT_SYMBOL_GPL(nfs_link);
1919 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1920 * different file handle for the same inode after a rename (e.g. when
1921 * moving to a different directory). A fail-safe method to do so would
1922 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1923 * rename the old file using the sillyrename stuff. This way, the original
1924 * file in old_dir will go away when the last process iput()s the inode.
1928 * It actually works quite well. One needs to have the possibility for
1929 * at least one ".nfs..." file in each directory the file ever gets
1930 * moved or linked to which happens automagically with the new
1931 * implementation that only depends on the dcache stuff instead of
1932 * using the inode layer
1934 * Unfortunately, things are a little more complicated than indicated
1935 * above. For a cross-directory move, we want to make sure we can get
1936 * rid of the old inode after the operation. This means there must be
1937 * no pending writes (if it's a file), and the use count must be 1.
1938 * If these conditions are met, we can drop the dentries before doing
1941 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1942 struct inode *new_dir, struct dentry *new_dentry)
1944 struct inode *old_inode = old_dentry->d_inode;
1945 struct inode *new_inode = new_dentry->d_inode;
1946 struct dentry *dentry = NULL, *rehash = NULL;
1947 struct rpc_task *task;
1950 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
1951 old_dentry, new_dentry,
1952 d_count(new_dentry));
1954 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
1956 * For non-directories, check whether the target is busy and if so,
1957 * make a copy of the dentry and then do a silly-rename. If the
1958 * silly-rename succeeds, the copied dentry is hashed and becomes
1961 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1963 * To prevent any new references to the target during the
1964 * rename, we unhash the dentry in advance.
1966 if (!d_unhashed(new_dentry)) {
1968 rehash = new_dentry;
1971 if (d_count(new_dentry) > 2) {
1974 /* copy the target dentry's name */
1975 dentry = d_alloc(new_dentry->d_parent,
1976 &new_dentry->d_name);
1980 /* silly-rename the existing target ... */
1981 err = nfs_sillyrename(new_dir, new_dentry);
1985 new_dentry = dentry;
1991 NFS_PROTO(old_inode)->return_delegation(old_inode);
1992 if (new_inode != NULL)
1993 NFS_PROTO(new_inode)->return_delegation(new_inode);
1995 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
1997 error = PTR_ERR(task);
2001 error = rpc_wait_for_completion_task(task);
2003 error = task->tk_status;
2005 nfs_mark_for_revalidate(old_inode);
2009 trace_nfs_rename_exit(old_dir, old_dentry,
2010 new_dir, new_dentry, error);
2012 if (new_inode != NULL)
2013 nfs_drop_nlink(new_inode);
2014 d_move(old_dentry, new_dentry);
2015 nfs_set_verifier(new_dentry,
2016 nfs_save_change_attribute(new_dir));
2017 } else if (error == -ENOENT)
2018 nfs_dentry_handle_enoent(old_dentry);
2020 /* new dentry created? */
2025 EXPORT_SYMBOL_GPL(nfs_rename);
2027 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2028 static LIST_HEAD(nfs_access_lru_list);
2029 static atomic_long_t nfs_access_nr_entries;
2031 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2033 put_rpccred(entry->cred);
2035 smp_mb__before_atomic();
2036 atomic_long_dec(&nfs_access_nr_entries);
2037 smp_mb__after_atomic();
2040 static void nfs_access_free_list(struct list_head *head)
2042 struct nfs_access_entry *cache;
2044 while (!list_empty(head)) {
2045 cache = list_entry(head->next, struct nfs_access_entry, lru);
2046 list_del(&cache->lru);
2047 nfs_access_free_entry(cache);
2052 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2055 struct nfs_inode *nfsi, *next;
2056 struct nfs_access_entry *cache;
2057 int nr_to_scan = sc->nr_to_scan;
2058 gfp_t gfp_mask = sc->gfp_mask;
2061 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2064 spin_lock(&nfs_access_lru_lock);
2065 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2066 struct inode *inode;
2068 if (nr_to_scan-- == 0)
2070 inode = &nfsi->vfs_inode;
2071 spin_lock(&inode->i_lock);
2072 if (list_empty(&nfsi->access_cache_entry_lru))
2073 goto remove_lru_entry;
2074 cache = list_entry(nfsi->access_cache_entry_lru.next,
2075 struct nfs_access_entry, lru);
2076 list_move(&cache->lru, &head);
2077 rb_erase(&cache->rb_node, &nfsi->access_cache);
2079 if (!list_empty(&nfsi->access_cache_entry_lru))
2080 list_move_tail(&nfsi->access_cache_inode_lru,
2081 &nfs_access_lru_list);
2084 list_del_init(&nfsi->access_cache_inode_lru);
2085 smp_mb__before_atomic();
2086 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2087 smp_mb__after_atomic();
2089 spin_unlock(&inode->i_lock);
2091 spin_unlock(&nfs_access_lru_lock);
2092 nfs_access_free_list(&head);
2097 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2099 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2102 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2104 struct rb_root *root_node = &nfsi->access_cache;
2106 struct nfs_access_entry *entry;
2108 /* Unhook entries from the cache */
2109 while ((n = rb_first(root_node)) != NULL) {
2110 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2111 rb_erase(n, root_node);
2112 list_move(&entry->lru, head);
2114 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2117 void nfs_access_zap_cache(struct inode *inode)
2121 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2123 /* Remove from global LRU init */
2124 spin_lock(&nfs_access_lru_lock);
2125 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2126 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2128 spin_lock(&inode->i_lock);
2129 __nfs_access_zap_cache(NFS_I(inode), &head);
2130 spin_unlock(&inode->i_lock);
2131 spin_unlock(&nfs_access_lru_lock);
2132 nfs_access_free_list(&head);
2134 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2136 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2138 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2139 struct nfs_access_entry *entry;
2142 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2144 if (cred < entry->cred)
2146 else if (cred > entry->cred)
2154 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2156 struct nfs_inode *nfsi = NFS_I(inode);
2157 struct nfs_access_entry *cache;
2160 spin_lock(&inode->i_lock);
2161 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2163 cache = nfs_access_search_rbtree(inode, cred);
2166 if (!nfs_have_delegated_attributes(inode) &&
2167 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
2169 res->jiffies = cache->jiffies;
2170 res->cred = cache->cred;
2171 res->mask = cache->mask;
2172 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2175 spin_unlock(&inode->i_lock);
2178 rb_erase(&cache->rb_node, &nfsi->access_cache);
2179 list_del(&cache->lru);
2180 spin_unlock(&inode->i_lock);
2181 nfs_access_free_entry(cache);
2184 spin_unlock(&inode->i_lock);
2185 nfs_access_zap_cache(inode);
2189 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2191 struct nfs_inode *nfsi = NFS_I(inode);
2192 struct rb_root *root_node = &nfsi->access_cache;
2193 struct rb_node **p = &root_node->rb_node;
2194 struct rb_node *parent = NULL;
2195 struct nfs_access_entry *entry;
2197 spin_lock(&inode->i_lock);
2198 while (*p != NULL) {
2200 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2202 if (set->cred < entry->cred)
2203 p = &parent->rb_left;
2204 else if (set->cred > entry->cred)
2205 p = &parent->rb_right;
2209 rb_link_node(&set->rb_node, parent, p);
2210 rb_insert_color(&set->rb_node, root_node);
2211 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2212 spin_unlock(&inode->i_lock);
2215 rb_replace_node(parent, &set->rb_node, root_node);
2216 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2217 list_del(&entry->lru);
2218 spin_unlock(&inode->i_lock);
2219 nfs_access_free_entry(entry);
2222 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2224 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2227 RB_CLEAR_NODE(&cache->rb_node);
2228 cache->jiffies = set->jiffies;
2229 cache->cred = get_rpccred(set->cred);
2230 cache->mask = set->mask;
2232 nfs_access_add_rbtree(inode, cache);
2234 /* Update accounting */
2235 smp_mb__before_atomic();
2236 atomic_long_inc(&nfs_access_nr_entries);
2237 smp_mb__after_atomic();
2239 /* Add inode to global LRU list */
2240 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2241 spin_lock(&nfs_access_lru_lock);
2242 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2243 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2244 &nfs_access_lru_list);
2245 spin_unlock(&nfs_access_lru_lock);
2248 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2250 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2253 if (access_result & NFS4_ACCESS_READ)
2254 entry->mask |= MAY_READ;
2256 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2257 entry->mask |= MAY_WRITE;
2258 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2259 entry->mask |= MAY_EXEC;
2261 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2263 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2265 struct nfs_access_entry cache;
2268 trace_nfs_access_enter(inode);
2270 status = nfs_access_get_cached(inode, cred, &cache);
2274 /* Be clever: ask server to check for all possible rights */
2275 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2277 cache.jiffies = jiffies;
2278 status = NFS_PROTO(inode)->access(inode, &cache);
2280 if (status == -ESTALE) {
2281 nfs_zap_caches(inode);
2282 if (!S_ISDIR(inode->i_mode))
2283 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2287 nfs_access_add_cache(inode, &cache);
2289 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2292 trace_nfs_access_exit(inode, status);
2296 static int nfs_open_permission_mask(int openflags)
2300 if (openflags & __FMODE_EXEC) {
2301 /* ONLY check exec rights */
2304 if ((openflags & O_ACCMODE) != O_WRONLY)
2306 if ((openflags & O_ACCMODE) != O_RDONLY)
2313 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2315 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2317 EXPORT_SYMBOL_GPL(nfs_may_open);
2319 int nfs_permission(struct inode *inode, int mask)
2321 struct rpc_cred *cred;
2324 if (mask & MAY_NOT_BLOCK)
2327 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2329 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2331 /* Is this sys_access() ? */
2332 if (mask & (MAY_ACCESS | MAY_CHDIR))
2335 switch (inode->i_mode & S_IFMT) {
2342 * Optimize away all write operations, since the server
2343 * will check permissions when we perform the op.
2345 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2350 if (!NFS_PROTO(inode)->access)
2353 cred = rpc_lookup_cred();
2354 if (!IS_ERR(cred)) {
2355 res = nfs_do_access(inode, cred, mask);
2358 res = PTR_ERR(cred);
2360 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2363 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2364 inode->i_sb->s_id, inode->i_ino, mask, res);
2367 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2369 res = generic_permission(inode, mask);
2372 EXPORT_SYMBOL_GPL(nfs_permission);
2376 * version-control: t
2377 * kept-new-versions: 5