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/time.h>
21 #include <linux/errno.h>
22 #include <linux/stat.h>
23 #include <linux/fcntl.h>
24 #include <linux/string.h>
25 #include <linux/kernel.h>
26 #include <linux/slab.h>
28 #include <linux/sunrpc/clnt.h>
29 #include <linux/nfs_fs.h>
30 #include <linux/nfs_mount.h>
31 #include <linux/pagemap.h>
32 #include <linux/smp_lock.h>
33 #include <linux/pagevec.h>
34 #include <linux/namei.h>
35 #include <linux/mount.h>
38 #include "delegation.h"
41 #define NFS_PARANOIA 1
42 /* #define NFS_DEBUG_VERBOSE 1 */
44 static int nfs_opendir(struct inode *, struct file *);
45 static int nfs_readdir(struct file *, void *, filldir_t);
46 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
47 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
48 static int nfs_mkdir(struct inode *, struct dentry *, int);
49 static int nfs_rmdir(struct inode *, struct dentry *);
50 static int nfs_unlink(struct inode *, struct dentry *);
51 static int nfs_symlink(struct inode *, struct dentry *, const char *);
52 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
53 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
54 static int nfs_rename(struct inode *, struct dentry *,
55 struct inode *, struct dentry *);
56 static int nfs_fsync_dir(struct file *, struct dentry *, int);
57 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
59 const struct file_operations nfs_dir_operations = {
60 .llseek = nfs_llseek_dir,
61 .read = generic_read_dir,
62 .readdir = nfs_readdir,
64 .release = nfs_release,
65 .fsync = nfs_fsync_dir,
68 const struct inode_operations nfs_dir_inode_operations = {
73 .symlink = nfs_symlink,
78 .permission = nfs_permission,
79 .getattr = nfs_getattr,
80 .setattr = nfs_setattr,
84 const struct inode_operations nfs3_dir_inode_operations = {
89 .symlink = nfs_symlink,
94 .permission = nfs_permission,
95 .getattr = nfs_getattr,
96 .setattr = nfs_setattr,
97 .listxattr = nfs3_listxattr,
98 .getxattr = nfs3_getxattr,
99 .setxattr = nfs3_setxattr,
100 .removexattr = nfs3_removexattr,
102 #endif /* CONFIG_NFS_V3 */
106 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
107 const struct inode_operations nfs4_dir_inode_operations = {
108 .create = nfs_create,
109 .lookup = nfs_atomic_lookup,
111 .unlink = nfs_unlink,
112 .symlink = nfs_symlink,
116 .rename = nfs_rename,
117 .permission = nfs_permission,
118 .getattr = nfs_getattr,
119 .setattr = nfs_setattr,
120 .getxattr = nfs4_getxattr,
121 .setxattr = nfs4_setxattr,
122 .listxattr = nfs4_listxattr,
125 #endif /* CONFIG_NFS_V4 */
131 nfs_opendir(struct inode *inode, struct file *filp)
135 dfprintk(VFS, "NFS: opendir(%s/%ld)\n",
136 inode->i_sb->s_id, inode->i_ino);
139 /* Call generic open code in order to cache credentials */
140 res = nfs_open(inode, filp);
145 typedef __be32 * (*decode_dirent_t)(__be32 *, struct nfs_entry *, int);
149 unsigned long page_index;
152 loff_t current_index;
153 struct nfs_entry *entry;
154 decode_dirent_t decode;
157 unsigned long timestamp;
159 } nfs_readdir_descriptor_t;
161 /* Now we cache directories properly, by stuffing the dirent
162 * data directly in the page cache.
164 * Inode invalidation due to refresh etc. takes care of
165 * _everything_, no sloppy entry flushing logic, no extraneous
166 * copying, network direct to page cache, the way it was meant
169 * NOTE: Dirent information verification is done always by the
170 * page-in of the RPC reply, nowhere else, this simplies
171 * things substantially.
174 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
176 struct file *file = desc->file;
177 struct inode *inode = file->f_path.dentry->d_inode;
178 struct rpc_cred *cred = nfs_file_cred(file);
179 unsigned long timestamp;
182 dfprintk(DIRCACHE, "NFS: %s: reading cookie %Lu into page %lu\n",
183 __FUNCTION__, (long long)desc->entry->cookie,
188 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, desc->entry->cookie, page,
189 NFS_SERVER(inode)->dtsize, desc->plus);
191 /* We requested READDIRPLUS, but the server doesn't grok it */
192 if (error == -ENOTSUPP && desc->plus) {
193 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
194 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
200 desc->timestamp = timestamp;
201 desc->timestamp_valid = 1;
202 SetPageUptodate(page);
203 spin_lock(&inode->i_lock);
204 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
205 spin_unlock(&inode->i_lock);
206 /* Ensure consistent page alignment of the data.
207 * Note: assumes we have exclusive access to this mapping either
208 * through inode->i_mutex or some other mechanism.
210 if (page->index == 0 && invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1) < 0) {
211 /* Should never happen */
212 nfs_zap_mapping(inode, inode->i_mapping);
219 nfs_zap_caches(inode);
225 int dir_decode(nfs_readdir_descriptor_t *desc)
227 __be32 *p = desc->ptr;
228 p = desc->decode(p, desc->entry, desc->plus);
232 if (desc->timestamp_valid)
233 desc->entry->fattr->time_start = desc->timestamp;
235 desc->entry->fattr->valid &= ~NFS_ATTR_FATTR;
240 void dir_page_release(nfs_readdir_descriptor_t *desc)
243 page_cache_release(desc->page);
249 * Given a pointer to a buffer that has already been filled by a call
250 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
252 * If the end of the buffer has been reached, return -EAGAIN, if not,
253 * return the offset within the buffer of the next entry to be
257 int find_dirent(nfs_readdir_descriptor_t *desc)
259 struct nfs_entry *entry = desc->entry;
263 while((status = dir_decode(desc)) == 0) {
264 dfprintk(DIRCACHE, "NFS: %s: examining cookie %Lu\n",
265 __FUNCTION__, (unsigned long long)entry->cookie);
266 if (entry->prev_cookie == *desc->dir_cookie)
268 if (loop_count++ > 200) {
277 * Given a pointer to a buffer that has already been filled by a call
278 * to readdir, find the entry at offset 'desc->file->f_pos'.
280 * If the end of the buffer has been reached, return -EAGAIN, if not,
281 * return the offset within the buffer of the next entry to be
285 int find_dirent_index(nfs_readdir_descriptor_t *desc)
287 struct nfs_entry *entry = desc->entry;
292 status = dir_decode(desc);
296 dfprintk(DIRCACHE, "NFS: found cookie %Lu at index %Ld\n",
297 (unsigned long long)entry->cookie, desc->current_index);
299 if (desc->file->f_pos == desc->current_index) {
300 *desc->dir_cookie = entry->cookie;
303 desc->current_index++;
304 if (loop_count++ > 200) {
313 * Find the given page, and call find_dirent() or find_dirent_index in
314 * order to try to return the next entry.
317 int find_dirent_page(nfs_readdir_descriptor_t *desc)
319 struct inode *inode = desc->file->f_path.dentry->d_inode;
323 dfprintk(DIRCACHE, "NFS: %s: searching page %ld for target %Lu\n",
324 __FUNCTION__, desc->page_index,
325 (long long) *desc->dir_cookie);
327 /* If we find the page in the page_cache, we cannot be sure
328 * how fresh the data is, so we will ignore readdir_plus attributes.
330 desc->timestamp_valid = 0;
331 page = read_cache_page(inode->i_mapping, desc->page_index,
332 (filler_t *)nfs_readdir_filler, desc);
334 status = PTR_ERR(page);
337 if (!PageUptodate(page))
340 /* NOTE: Someone else may have changed the READDIRPLUS flag */
342 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
343 if (*desc->dir_cookie != 0)
344 status = find_dirent(desc);
346 status = find_dirent_index(desc);
348 dir_page_release(desc);
350 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, status);
353 page_cache_release(page);
358 * Recurse through the page cache pages, and return a
359 * filled nfs_entry structure of the next directory entry if possible.
361 * The target for the search is '*desc->dir_cookie' if non-0,
362 * 'desc->file->f_pos' otherwise
365 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
370 /* Always search-by-index from the beginning of the cache */
371 if (*desc->dir_cookie == 0) {
372 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for offset %Ld\n",
373 (long long)desc->file->f_pos);
374 desc->page_index = 0;
375 desc->entry->cookie = desc->entry->prev_cookie = 0;
376 desc->entry->eof = 0;
377 desc->current_index = 0;
379 dfprintk(DIRCACHE, "NFS: readdir_search_pagecache() searching for cookie %Lu\n",
380 (unsigned long long)*desc->dir_cookie);
383 res = find_dirent_page(desc);
386 /* Align to beginning of next page */
388 if (loop_count++ > 200) {
394 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __FUNCTION__, res);
398 static inline unsigned int dt_type(struct inode *inode)
400 return (inode->i_mode >> 12) & 15;
403 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
406 * Once we've found the start of the dirent within a page: fill 'er up...
409 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
412 struct file *file = desc->file;
413 struct nfs_entry *entry = desc->entry;
414 struct dentry *dentry = NULL;
415 unsigned long fileid;
419 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n",
420 (unsigned long long)entry->cookie);
423 unsigned d_type = DT_UNKNOWN;
424 /* Note: entry->prev_cookie contains the cookie for
425 * retrieving the current dirent on the server */
426 fileid = nfs_fileid_to_ino_t(entry->ino);
428 /* Get a dentry if we have one */
431 dentry = nfs_readdir_lookup(desc);
433 /* Use readdirplus info */
434 if (dentry != NULL && dentry->d_inode != NULL) {
435 d_type = dt_type(dentry->d_inode);
436 fileid = dentry->d_inode->i_ino;
439 res = filldir(dirent, entry->name, entry->len,
440 file->f_pos, fileid, d_type);
444 *desc->dir_cookie = entry->cookie;
445 if (dir_decode(desc) != 0) {
449 if (loop_count++ > 200) {
454 dir_page_release(desc);
457 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
458 (unsigned long long)*desc->dir_cookie, res);
463 * If we cannot find a cookie in our cache, we suspect that this is
464 * because it points to a deleted file, so we ask the server to return
465 * whatever it thinks is the next entry. We then feed this to filldir.
466 * If all goes well, we should then be able to find our way round the
467 * cache on the next call to readdir_search_pagecache();
469 * NOTE: we cannot add the anonymous page to the pagecache because
470 * the data it contains might not be page aligned. Besides,
471 * we should already have a complete representation of the
472 * directory in the page cache by the time we get here.
475 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
478 struct file *file = desc->file;
479 struct inode *inode = file->f_path.dentry->d_inode;
480 struct rpc_cred *cred = nfs_file_cred(file);
481 struct page *page = NULL;
483 unsigned long timestamp;
485 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
486 (unsigned long long)*desc->dir_cookie);
488 page = alloc_page(GFP_HIGHUSER);
494 desc->error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, *desc->dir_cookie,
496 NFS_SERVER(inode)->dtsize,
498 spin_lock(&inode->i_lock);
499 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
500 spin_unlock(&inode->i_lock);
502 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
503 if (desc->error >= 0) {
504 desc->timestamp = timestamp;
505 desc->timestamp_valid = 1;
506 if ((status = dir_decode(desc)) == 0)
507 desc->entry->prev_cookie = *desc->dir_cookie;
513 status = nfs_do_filldir(desc, dirent, filldir);
515 /* Reset read descriptor so it searches the page cache from
516 * the start upon the next call to readdir_search_pagecache() */
517 desc->page_index = 0;
518 desc->entry->cookie = desc->entry->prev_cookie = 0;
519 desc->entry->eof = 0;
521 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
522 __FUNCTION__, status);
525 dir_page_release(desc);
529 /* The file offset position represents the dirent entry number. A
530 last cookie cache takes care of the common case of reading the
533 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
535 struct dentry *dentry = filp->f_path.dentry;
536 struct inode *inode = dentry->d_inode;
537 nfs_readdir_descriptor_t my_desc,
539 struct nfs_entry my_entry;
541 struct nfs_fattr fattr;
544 dfprintk(VFS, "NFS: readdir(%s/%s) starting at cookie %Lu\n",
545 dentry->d_parent->d_name.name, dentry->d_name.name,
546 (long long)filp->f_pos);
547 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
551 res = nfs_revalidate_mapping_nolock(inode, filp->f_mapping);
558 * filp->f_pos points to the dirent entry number.
559 * *desc->dir_cookie has the cookie for the next entry. We have
560 * to either find the entry with the appropriate number or
561 * revalidate the cookie.
563 memset(desc, 0, sizeof(*desc));
566 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
567 desc->decode = NFS_PROTO(inode)->decode_dirent;
568 desc->plus = NFS_USE_READDIRPLUS(inode);
570 my_entry.cookie = my_entry.prev_cookie = 0;
573 my_entry.fattr = &fattr;
574 nfs_fattr_init(&fattr);
575 desc->entry = &my_entry;
577 while(!desc->entry->eof) {
578 res = readdir_search_pagecache(desc);
580 if (res == -EBADCOOKIE) {
581 /* This means either end of directory */
582 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
583 /* Or that the server has 'lost' a cookie */
584 res = uncached_readdir(desc, dirent, filldir);
591 if (res == -ETOOSMALL && desc->plus) {
592 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
593 nfs_zap_caches(inode);
595 desc->entry->eof = 0;
601 res = nfs_do_filldir(desc, dirent, filldir);
610 dfprintk(VFS, "NFS: readdir(%s/%s) returns %ld\n",
611 dentry->d_parent->d_name.name, dentry->d_name.name,
616 loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
618 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
621 offset += filp->f_pos;
629 if (offset != filp->f_pos) {
630 filp->f_pos = offset;
631 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
634 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
639 * All directory operations under NFS are synchronous, so fsync()
640 * is a dummy operation.
642 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
644 dfprintk(VFS, "NFS: fsync_dir(%s/%s) datasync %d\n",
645 dentry->d_parent->d_name.name, dentry->d_name.name,
652 * A check for whether or not the parent directory has changed.
653 * In the case it has, we assume that the dentries are untrustworthy
654 * and may need to be looked up again.
656 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
660 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
661 || nfs_attribute_timeout(dir))
663 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
666 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
668 dentry->d_fsdata = (void *)verf;
671 static void nfs_refresh_verifier(struct dentry * dentry, unsigned long verf)
673 if (time_after(verf, (unsigned long)dentry->d_fsdata))
674 nfs_set_verifier(dentry, verf);
678 * Whenever an NFS operation succeeds, we know that the dentry
679 * is valid, so we update the revalidation timestamp.
681 static inline void nfs_renew_times(struct dentry * dentry)
683 dentry->d_time = jiffies;
687 * Return the intent data that applies to this particular path component
689 * Note that the current set of intents only apply to the very last
690 * component of the path.
691 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
693 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
695 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
697 return nd->flags & mask;
701 * Inode and filehandle revalidation for lookups.
703 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
704 * or if the intent information indicates that we're about to open this
705 * particular file and the "nocto" mount flag is not set.
709 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
711 struct nfs_server *server = NFS_SERVER(inode);
714 /* VFS wants an on-the-wire revalidation */
715 if (nd->flags & LOOKUP_REVAL)
717 /* This is an open(2) */
718 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
719 !(server->flags & NFS_MOUNT_NOCTO) &&
720 (S_ISREG(inode->i_mode) ||
721 S_ISDIR(inode->i_mode)))
724 return nfs_revalidate_inode(server, inode);
726 return __nfs_revalidate_inode(server, inode);
730 * We judge how long we want to trust negative
731 * dentries by looking at the parent inode mtime.
733 * If parent mtime has changed, we revalidate, else we wait for a
734 * period corresponding to the parent's attribute cache timeout value.
737 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
738 struct nameidata *nd)
740 /* Don't revalidate a negative dentry if we're creating a new file */
741 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
743 return !nfs_check_verifier(dir, dentry);
747 * This is called every time the dcache has a lookup hit,
748 * and we should check whether we can really trust that
751 * NOTE! The hit can be a negative hit too, don't assume
754 * If the parent directory is seen to have changed, we throw out the
755 * cached dentry and do a new lookup.
757 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
761 struct dentry *parent;
763 struct nfs_fh fhandle;
764 struct nfs_fattr fattr;
765 unsigned long verifier;
767 parent = dget_parent(dentry);
769 dir = parent->d_inode;
770 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
771 inode = dentry->d_inode;
774 if (nfs_neg_need_reval(dir, dentry, nd))
779 if (is_bad_inode(inode)) {
780 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
781 __FUNCTION__, dentry->d_parent->d_name.name,
782 dentry->d_name.name);
786 /* Revalidate parent directory attribute cache */
787 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
790 /* Force a full look up iff the parent directory has changed */
791 if (nfs_check_verifier(dir, dentry)) {
792 if (nfs_lookup_verify_inode(inode, nd))
797 if (NFS_STALE(inode))
800 verifier = nfs_save_change_attribute(dir);
801 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
804 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
806 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
809 nfs_renew_times(dentry);
810 nfs_refresh_verifier(dentry, verifier);
814 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
815 __FUNCTION__, dentry->d_parent->d_name.name,
816 dentry->d_name.name);
822 if (inode && S_ISDIR(inode->i_mode)) {
823 /* Purge readdir caches. */
824 nfs_zap_caches(inode);
825 /* If we have submounts, don't unhash ! */
826 if (have_submounts(dentry))
828 shrink_dcache_parent(dentry);
833 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
834 __FUNCTION__, dentry->d_parent->d_name.name,
835 dentry->d_name.name);
840 * This is called from dput() when d_count is going to 0.
842 static int nfs_dentry_delete(struct dentry *dentry)
844 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
845 dentry->d_parent->d_name.name, dentry->d_name.name,
848 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
849 /* Unhash it, so that ->d_iput() would be called */
852 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
853 /* Unhash it, so that ancestors of killed async unlink
854 * files will be cleaned up during umount */
862 * Called when the dentry loses inode.
863 * We use it to clean up silly-renamed files.
865 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
867 nfs_inode_return_delegation(inode);
868 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
871 nfs_complete_unlink(dentry);
874 /* When creating a negative dentry, we want to renew d_time */
875 nfs_renew_times(dentry);
879 struct dentry_operations nfs_dentry_operations = {
880 .d_revalidate = nfs_lookup_revalidate,
881 .d_delete = nfs_dentry_delete,
882 .d_iput = nfs_dentry_iput,
886 * Use intent information to check whether or not we're going to do
887 * an O_EXCL create using this path component.
890 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
892 if (NFS_PROTO(dir)->version == 2)
894 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
896 return (nd->intent.open.flags & O_EXCL) != 0;
899 static inline int nfs_reval_fsid(struct vfsmount *mnt, struct inode *dir,
900 struct nfs_fh *fh, struct nfs_fattr *fattr)
902 struct nfs_server *server = NFS_SERVER(dir);
904 if (!nfs_fsid_equal(&server->fsid, &fattr->fsid))
905 /* Revalidate fsid on root dir */
906 return __nfs_revalidate_inode(server, mnt->mnt_root->d_inode);
910 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
913 struct inode *inode = NULL;
915 struct nfs_fh fhandle;
916 struct nfs_fattr fattr;
918 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
919 dentry->d_parent->d_name.name, dentry->d_name.name);
920 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
922 res = ERR_PTR(-ENAMETOOLONG);
923 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
926 res = ERR_PTR(-ENOMEM);
927 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
932 * If we're doing an exclusive create, optimize away the lookup
933 * but don't hash the dentry.
935 if (nfs_is_exclusive_create(dir, nd)) {
936 d_instantiate(dentry, NULL);
941 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
942 if (error == -ENOENT)
945 res = ERR_PTR(error);
948 error = nfs_reval_fsid(nd->mnt, dir, &fhandle, &fattr);
950 res = ERR_PTR(error);
953 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
954 res = (struct dentry *)inode;
959 res = d_materialise_unique(dentry, inode);
961 struct dentry *parent;
964 /* Was a directory renamed! */
965 parent = dget_parent(res);
966 if (!IS_ROOT(parent))
967 nfs_mark_for_revalidate(parent->d_inode);
971 nfs_renew_times(dentry);
972 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
980 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
982 struct dentry_operations nfs4_dentry_operations = {
983 .d_revalidate = nfs_open_revalidate,
984 .d_delete = nfs_dentry_delete,
985 .d_iput = nfs_dentry_iput,
989 * Use intent information to determine whether we need to substitute
990 * the NFSv4-style stateful OPEN for the LOOKUP call
992 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
994 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
996 /* NFS does not (yet) have a stateful open for directories */
997 if (nd->flags & LOOKUP_DIRECTORY)
999 /* Are we trying to write to a read only partition? */
1000 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
1005 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1007 struct dentry *res = NULL;
1010 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1011 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1013 /* Check that we are indeed trying to open this file */
1014 if (!is_atomic_open(dir, nd))
1017 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1018 res = ERR_PTR(-ENAMETOOLONG);
1021 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1023 /* Let vfs_create() deal with O_EXCL */
1024 if (nd->intent.open.flags & O_EXCL) {
1025 d_add(dentry, NULL);
1029 /* Open the file on the server */
1031 /* Revalidate parent directory attribute cache */
1032 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
1034 res = ERR_PTR(error);
1039 if (nd->intent.open.flags & O_CREAT) {
1040 nfs_begin_data_update(dir);
1041 res = nfs4_atomic_open(dir, dentry, nd);
1042 nfs_end_data_update(dir);
1044 res = nfs4_atomic_open(dir, dentry, nd);
1047 error = PTR_ERR(res);
1049 /* Make a negative dentry */
1053 /* This turned out not to be a regular file */
1058 if (!(nd->intent.open.flags & O_NOFOLLOW))
1064 } else if (res != NULL)
1066 nfs_renew_times(dentry);
1067 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1071 return nfs_lookup(dir, dentry, nd);
1074 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1076 struct dentry *parent = NULL;
1077 struct inode *inode = dentry->d_inode;
1079 unsigned long verifier;
1080 int openflags, ret = 0;
1082 parent = dget_parent(dentry);
1083 dir = parent->d_inode;
1084 if (!is_atomic_open(dir, nd))
1086 /* We can't create new files in nfs_open_revalidate(), so we
1087 * optimize away revalidation of negative dentries.
1091 /* NFS only supports OPEN on regular files */
1092 if (!S_ISREG(inode->i_mode))
1094 openflags = nd->intent.open.flags;
1095 /* We cannot do exclusive creation on a positive dentry */
1096 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1098 /* We can't create new files, or truncate existing ones here */
1099 openflags &= ~(O_CREAT|O_TRUNC);
1102 * Note: we're not holding inode->i_mutex and so may be racing with
1103 * operations that change the directory. We therefore save the
1104 * change attribute *before* we do the RPC call.
1107 verifier = nfs_save_change_attribute(dir);
1108 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1110 nfs_refresh_verifier(dentry, verifier);
1119 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1121 return nfs_lookup_revalidate(dentry, nd);
1123 #endif /* CONFIG_NFSV4 */
1125 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1127 struct dentry *parent = desc->file->f_path.dentry;
1128 struct inode *dir = parent->d_inode;
1129 struct nfs_entry *entry = desc->entry;
1130 struct dentry *dentry, *alias;
1131 struct qstr name = {
1132 .name = entry->name,
1135 struct inode *inode;
1139 if (name.name[0] == '.' && name.name[1] == '.')
1140 return dget_parent(parent);
1143 if (name.name[0] == '.')
1144 return dget(parent);
1146 name.hash = full_name_hash(name.name, name.len);
1147 dentry = d_lookup(parent, &name);
1148 if (dentry != NULL) {
1149 /* Is this a positive dentry that matches the readdir info? */
1150 if (dentry->d_inode != NULL &&
1151 (NFS_FILEID(dentry->d_inode) == entry->ino ||
1152 d_mountpoint(dentry))) {
1153 if (!desc->plus || entry->fh->size == 0)
1155 if (nfs_compare_fh(NFS_FH(dentry->d_inode),
1159 /* No, so d_drop to allow one to be created */
1163 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1165 /* Note: caller is already holding the dir->i_mutex! */
1166 dentry = d_alloc(parent, &name);
1169 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1170 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1171 if (IS_ERR(inode)) {
1176 alias = d_materialise_unique(dentry, inode);
1177 if (alias != NULL) {
1184 nfs_renew_times(dentry);
1185 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1188 nfs_renew_times(dentry);
1189 nfs_refresh_verifier(dentry, nfs_save_change_attribute(dir));
1194 * Code common to create, mkdir, and mknod.
1196 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1197 struct nfs_fattr *fattr)
1199 struct inode *inode;
1200 int error = -EACCES;
1202 /* We may have been initialized further down */
1203 if (dentry->d_inode)
1205 if (fhandle->size == 0) {
1206 struct inode *dir = dentry->d_parent->d_inode;
1207 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1211 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1212 struct nfs_server *server = NFS_SB(dentry->d_sb);
1213 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
1217 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1218 error = PTR_ERR(inode);
1221 d_instantiate(dentry, inode);
1222 if (d_unhashed(dentry))
1228 * Following a failed create operation, we drop the dentry rather
1229 * than retain a negative dentry. This avoids a problem in the event
1230 * that the operation succeeded on the server, but an error in the
1231 * reply path made it appear to have failed.
1233 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1234 struct nameidata *nd)
1240 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1241 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1243 attr.ia_mode = mode;
1244 attr.ia_valid = ATTR_MODE;
1246 if (nd && (nd->flags & LOOKUP_CREATE))
1247 open_flags = nd->intent.open.flags;
1250 nfs_begin_data_update(dir);
1251 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1252 nfs_end_data_update(dir);
1255 nfs_renew_times(dentry);
1256 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1266 * See comments for nfs_proc_create regarding failed operations.
1269 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1274 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1275 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1277 if (!new_valid_dev(rdev))
1280 attr.ia_mode = mode;
1281 attr.ia_valid = ATTR_MODE;
1284 nfs_begin_data_update(dir);
1285 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1286 nfs_end_data_update(dir);
1289 nfs_renew_times(dentry);
1290 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1300 * See comments for nfs_proc_create regarding failed operations.
1302 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1307 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1308 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1310 attr.ia_valid = ATTR_MODE;
1311 attr.ia_mode = mode | S_IFDIR;
1314 nfs_begin_data_update(dir);
1315 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1316 nfs_end_data_update(dir);
1319 nfs_renew_times(dentry);
1320 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1329 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1333 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1334 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1337 nfs_begin_data_update(dir);
1338 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1339 /* Ensure the VFS deletes this inode */
1340 if (error == 0 && dentry->d_inode != NULL)
1341 clear_nlink(dentry->d_inode);
1342 nfs_end_data_update(dir);
1348 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1350 static unsigned int sillycounter;
1351 const int i_inosize = sizeof(dir->i_ino)*2;
1352 const int countersize = sizeof(sillycounter)*2;
1353 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1356 struct dentry *sdentry;
1359 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1360 dentry->d_parent->d_name.name, dentry->d_name.name,
1361 atomic_read(&dentry->d_count));
1362 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1365 if (!dentry->d_inode)
1366 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1367 dentry->d_parent->d_name.name, dentry->d_name.name);
1370 * We don't allow a dentry to be silly-renamed twice.
1373 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1376 sprintf(silly, ".nfs%*.*lx",
1377 i_inosize, i_inosize, dentry->d_inode->i_ino);
1379 /* Return delegation in anticipation of the rename */
1380 nfs_inode_return_delegation(dentry->d_inode);
1384 char *suffix = silly + slen - countersize;
1388 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1390 dfprintk(VFS, "NFS: trying to rename %s to %s\n",
1391 dentry->d_name.name, silly);
1393 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1395 * N.B. Better to return EBUSY here ... it could be
1396 * dangerous to delete the file while it's in use.
1398 if (IS_ERR(sdentry))
1400 } while(sdentry->d_inode != NULL); /* need negative lookup */
1402 qsilly.name = silly;
1403 qsilly.len = strlen(silly);
1404 nfs_begin_data_update(dir);
1405 if (dentry->d_inode) {
1406 nfs_begin_data_update(dentry->d_inode);
1407 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1409 nfs_mark_for_revalidate(dentry->d_inode);
1410 nfs_end_data_update(dentry->d_inode);
1412 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1414 nfs_end_data_update(dir);
1416 nfs_renew_times(dentry);
1417 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1418 d_move(dentry, sdentry);
1419 error = nfs_async_unlink(dentry);
1420 /* If we return 0 we don't unlink */
1428 * Remove a file after making sure there are no pending writes,
1429 * and after checking that the file has only one user.
1431 * We invalidate the attribute cache and free the inode prior to the operation
1432 * to avoid possible races if the server reuses the inode.
1434 static int nfs_safe_remove(struct dentry *dentry)
1436 struct inode *dir = dentry->d_parent->d_inode;
1437 struct inode *inode = dentry->d_inode;
1440 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1441 dentry->d_parent->d_name.name, dentry->d_name.name);
1443 /* If the dentry was sillyrenamed, we simply call d_delete() */
1444 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1449 nfs_begin_data_update(dir);
1450 if (inode != NULL) {
1451 nfs_inode_return_delegation(inode);
1452 nfs_begin_data_update(inode);
1453 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1454 /* The VFS may want to delete this inode */
1457 nfs_mark_for_revalidate(inode);
1458 nfs_end_data_update(inode);
1460 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1461 nfs_end_data_update(dir);
1466 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1467 * belongs to an active ".nfs..." file and we return -EBUSY.
1469 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1471 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1474 int need_rehash = 0;
1476 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1477 dir->i_ino, dentry->d_name.name);
1480 spin_lock(&dcache_lock);
1481 spin_lock(&dentry->d_lock);
1482 if (atomic_read(&dentry->d_count) > 1) {
1483 spin_unlock(&dentry->d_lock);
1484 spin_unlock(&dcache_lock);
1485 /* Start asynchronous writeout of the inode */
1486 write_inode_now(dentry->d_inode, 0);
1487 error = nfs_sillyrename(dir, dentry);
1491 if (!d_unhashed(dentry)) {
1495 spin_unlock(&dentry->d_lock);
1496 spin_unlock(&dcache_lock);
1497 error = nfs_safe_remove(dentry);
1499 nfs_renew_times(dentry);
1500 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1501 } else if (need_rehash)
1508 * To create a symbolic link, most file systems instantiate a new inode,
1509 * add a page to it containing the path, then write it out to the disk
1510 * using prepare_write/commit_write.
1512 * Unfortunately the NFS client can't create the in-core inode first
1513 * because it needs a file handle to create an in-core inode (see
1514 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1515 * symlink request has completed on the server.
1517 * So instead we allocate a raw page, copy the symname into it, then do
1518 * the SYMLINK request with the page as the buffer. If it succeeds, we
1519 * now have a new file handle and can instantiate an in-core NFS inode
1520 * and move the raw page into its mapping.
1522 static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1524 struct pagevec lru_pvec;
1528 unsigned int pathlen = strlen(symname);
1531 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1532 dir->i_ino, dentry->d_name.name, symname);
1534 if (pathlen > PAGE_SIZE)
1535 return -ENAMETOOLONG;
1537 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1538 attr.ia_valid = ATTR_MODE;
1542 page = alloc_page(GFP_KERNEL);
1548 kaddr = kmap_atomic(page, KM_USER0);
1549 memcpy(kaddr, symname, pathlen);
1550 if (pathlen < PAGE_SIZE)
1551 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1552 kunmap_atomic(kaddr, KM_USER0);
1554 nfs_begin_data_update(dir);
1555 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1556 nfs_end_data_update(dir);
1558 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1559 dir->i_sb->s_id, dir->i_ino,
1560 dentry->d_name.name, symname, error);
1568 * No big deal if we can't add this page to the page cache here.
1569 * READLINK will get the missing page from the server if needed.
1571 pagevec_init(&lru_pvec, 0);
1572 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1574 pagevec_add(&lru_pvec, page);
1575 pagevec_lru_add(&lru_pvec);
1576 SetPageUptodate(page);
1586 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1588 struct inode *inode = old_dentry->d_inode;
1591 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1592 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1593 dentry->d_parent->d_name.name, dentry->d_name.name);
1596 nfs_begin_data_update(dir);
1597 nfs_begin_data_update(inode);
1598 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1600 atomic_inc(&inode->i_count);
1601 d_instantiate(dentry, inode);
1603 nfs_end_data_update(inode);
1604 nfs_end_data_update(dir);
1611 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1612 * different file handle for the same inode after a rename (e.g. when
1613 * moving to a different directory). A fail-safe method to do so would
1614 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1615 * rename the old file using the sillyrename stuff. This way, the original
1616 * file in old_dir will go away when the last process iput()s the inode.
1620 * It actually works quite well. One needs to have the possibility for
1621 * at least one ".nfs..." file in each directory the file ever gets
1622 * moved or linked to which happens automagically with the new
1623 * implementation that only depends on the dcache stuff instead of
1624 * using the inode layer
1626 * Unfortunately, things are a little more complicated than indicated
1627 * above. For a cross-directory move, we want to make sure we can get
1628 * rid of the old inode after the operation. This means there must be
1629 * no pending writes (if it's a file), and the use count must be 1.
1630 * If these conditions are met, we can drop the dentries before doing
1633 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1634 struct inode *new_dir, struct dentry *new_dentry)
1636 struct inode *old_inode = old_dentry->d_inode;
1637 struct inode *new_inode = new_dentry->d_inode;
1638 struct dentry *dentry = NULL, *rehash = NULL;
1642 * To prevent any new references to the target during the rename,
1643 * we unhash the dentry and free the inode in advance.
1646 if (!d_unhashed(new_dentry)) {
1648 rehash = new_dentry;
1651 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1652 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1653 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1654 atomic_read(&new_dentry->d_count));
1657 * First check whether the target is busy ... we can't
1658 * safely do _any_ rename if the target is in use.
1660 * For files, make a copy of the dentry and then do a
1661 * silly-rename. If the silly-rename succeeds, the
1662 * copied dentry is hashed and becomes the new target.
1666 if (S_ISDIR(new_inode->i_mode)) {
1668 if (!S_ISDIR(old_inode->i_mode))
1670 } else if (atomic_read(&new_dentry->d_count) > 2) {
1672 /* copy the target dentry's name */
1673 dentry = d_alloc(new_dentry->d_parent,
1674 &new_dentry->d_name);
1678 /* silly-rename the existing target ... */
1679 err = nfs_sillyrename(new_dir, new_dentry);
1681 new_dentry = rehash = dentry;
1683 /* instantiate the replacement target */
1684 d_instantiate(new_dentry, NULL);
1685 } else if (atomic_read(&new_dentry->d_count) > 1) {
1686 /* dentry still busy? */
1688 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1689 new_dentry->d_parent->d_name.name,
1690 new_dentry->d_name.name,
1691 atomic_read(&new_dentry->d_count));
1696 drop_nlink(new_inode);
1700 * ... prune child dentries and writebacks if needed.
1702 if (atomic_read(&old_dentry->d_count) > 1) {
1703 if (S_ISREG(old_inode->i_mode))
1704 nfs_wb_all(old_inode);
1705 shrink_dcache_parent(old_dentry);
1707 nfs_inode_return_delegation(old_inode);
1709 if (new_inode != NULL) {
1710 nfs_inode_return_delegation(new_inode);
1711 d_delete(new_dentry);
1714 nfs_begin_data_update(old_dir);
1715 nfs_begin_data_update(new_dir);
1716 nfs_begin_data_update(old_inode);
1717 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1718 new_dir, &new_dentry->d_name);
1719 nfs_mark_for_revalidate(old_inode);
1720 nfs_end_data_update(old_inode);
1721 nfs_end_data_update(new_dir);
1722 nfs_end_data_update(old_dir);
1727 d_move(old_dentry, new_dentry);
1728 nfs_renew_times(new_dentry);
1729 nfs_refresh_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1732 /* new dentry created? */
1739 static DEFINE_SPINLOCK(nfs_access_lru_lock);
1740 static LIST_HEAD(nfs_access_lru_list);
1741 static atomic_long_t nfs_access_nr_entries;
1743 static void nfs_access_free_entry(struct nfs_access_entry *entry)
1745 put_rpccred(entry->cred);
1747 smp_mb__before_atomic_dec();
1748 atomic_long_dec(&nfs_access_nr_entries);
1749 smp_mb__after_atomic_dec();
1752 int nfs_access_cache_shrinker(int nr_to_scan, gfp_t gfp_mask)
1755 struct nfs_inode *nfsi;
1756 struct nfs_access_entry *cache;
1758 spin_lock(&nfs_access_lru_lock);
1760 list_for_each_entry(nfsi, &nfs_access_lru_list, access_cache_inode_lru) {
1761 struct inode *inode;
1763 if (nr_to_scan-- == 0)
1765 inode = igrab(&nfsi->vfs_inode);
1768 spin_lock(&inode->i_lock);
1769 if (list_empty(&nfsi->access_cache_entry_lru))
1770 goto remove_lru_entry;
1771 cache = list_entry(nfsi->access_cache_entry_lru.next,
1772 struct nfs_access_entry, lru);
1773 list_move(&cache->lru, &head);
1774 rb_erase(&cache->rb_node, &nfsi->access_cache);
1775 if (!list_empty(&nfsi->access_cache_entry_lru))
1776 list_move_tail(&nfsi->access_cache_inode_lru,
1777 &nfs_access_lru_list);
1780 list_del_init(&nfsi->access_cache_inode_lru);
1781 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
1783 spin_unlock(&inode->i_lock);
1787 spin_unlock(&nfs_access_lru_lock);
1788 while (!list_empty(&head)) {
1789 cache = list_entry(head.next, struct nfs_access_entry, lru);
1790 list_del(&cache->lru);
1791 nfs_access_free_entry(cache);
1793 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
1796 static void __nfs_access_zap_cache(struct inode *inode)
1798 struct nfs_inode *nfsi = NFS_I(inode);
1799 struct rb_root *root_node = &nfsi->access_cache;
1800 struct rb_node *n, *dispose = NULL;
1801 struct nfs_access_entry *entry;
1803 /* Unhook entries from the cache */
1804 while ((n = rb_first(root_node)) != NULL) {
1805 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1806 rb_erase(n, root_node);
1807 list_del(&entry->lru);
1808 n->rb_left = dispose;
1811 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1812 spin_unlock(&inode->i_lock);
1814 /* Now kill them all! */
1815 while (dispose != NULL) {
1817 dispose = n->rb_left;
1818 nfs_access_free_entry(rb_entry(n, struct nfs_access_entry, rb_node));
1822 void nfs_access_zap_cache(struct inode *inode)
1824 /* Remove from global LRU init */
1825 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1826 spin_lock(&nfs_access_lru_lock);
1827 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
1828 spin_unlock(&nfs_access_lru_lock);
1831 spin_lock(&inode->i_lock);
1832 /* This will release the spinlock */
1833 __nfs_access_zap_cache(inode);
1836 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
1838 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
1839 struct nfs_access_entry *entry;
1842 entry = rb_entry(n, struct nfs_access_entry, rb_node);
1844 if (cred < entry->cred)
1846 else if (cred > entry->cred)
1854 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1856 struct nfs_inode *nfsi = NFS_I(inode);
1857 struct nfs_access_entry *cache;
1860 spin_lock(&inode->i_lock);
1861 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
1863 cache = nfs_access_search_rbtree(inode, cred);
1866 if (time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode)))
1868 res->jiffies = cache->jiffies;
1869 res->cred = cache->cred;
1870 res->mask = cache->mask;
1871 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
1874 spin_unlock(&inode->i_lock);
1877 rb_erase(&cache->rb_node, &nfsi->access_cache);
1878 list_del(&cache->lru);
1879 spin_unlock(&inode->i_lock);
1880 nfs_access_free_entry(cache);
1883 /* This will release the spinlock */
1884 __nfs_access_zap_cache(inode);
1888 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
1890 struct nfs_inode *nfsi = NFS_I(inode);
1891 struct rb_root *root_node = &nfsi->access_cache;
1892 struct rb_node **p = &root_node->rb_node;
1893 struct rb_node *parent = NULL;
1894 struct nfs_access_entry *entry;
1896 spin_lock(&inode->i_lock);
1897 while (*p != NULL) {
1899 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
1901 if (set->cred < entry->cred)
1902 p = &parent->rb_left;
1903 else if (set->cred > entry->cred)
1904 p = &parent->rb_right;
1908 rb_link_node(&set->rb_node, parent, p);
1909 rb_insert_color(&set->rb_node, root_node);
1910 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1911 spin_unlock(&inode->i_lock);
1914 rb_replace_node(parent, &set->rb_node, root_node);
1915 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
1916 list_del(&entry->lru);
1917 spin_unlock(&inode->i_lock);
1918 nfs_access_free_entry(entry);
1921 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1923 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
1926 RB_CLEAR_NODE(&cache->rb_node);
1927 cache->jiffies = set->jiffies;
1928 cache->cred = get_rpccred(set->cred);
1929 cache->mask = set->mask;
1931 nfs_access_add_rbtree(inode, cache);
1933 /* Update accounting */
1934 smp_mb__before_atomic_inc();
1935 atomic_long_inc(&nfs_access_nr_entries);
1936 smp_mb__after_atomic_inc();
1938 /* Add inode to global LRU list */
1939 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_FLAGS(inode))) {
1940 spin_lock(&nfs_access_lru_lock);
1941 list_add_tail(&NFS_I(inode)->access_cache_inode_lru, &nfs_access_lru_list);
1942 spin_unlock(&nfs_access_lru_lock);
1946 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1948 struct nfs_access_entry cache;
1951 status = nfs_access_get_cached(inode, cred, &cache);
1955 /* Be clever: ask server to check for all possible rights */
1956 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1958 cache.jiffies = jiffies;
1959 status = NFS_PROTO(inode)->access(inode, &cache);
1962 nfs_access_add_cache(inode, &cache);
1964 if ((cache.mask & mask) == mask)
1969 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1971 struct rpc_cred *cred;
1974 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1978 /* Is this sys_access() ? */
1979 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1982 switch (inode->i_mode & S_IFMT) {
1986 /* NFSv4 has atomic_open... */
1987 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1989 && (nd->flags & LOOKUP_OPEN))
1994 * Optimize away all write operations, since the server
1995 * will check permissions when we perform the op.
1997 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2004 if (!NFS_PROTO(inode)->access)
2007 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
2008 if (!IS_ERR(cred)) {
2009 res = nfs_do_access(inode, cred, mask);
2012 res = PTR_ERR(cred);
2015 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2016 inode->i_sb->s_id, inode->i_ino, mask, res);
2019 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2021 res = generic_permission(inode, mask, NULL);
2028 * version-control: t
2029 * kept-new-versions: 5