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/namei.h>
36 #include "delegation.h"
39 #define NFS_PARANOIA 1
40 /* #define NFS_DEBUG_VERBOSE 1 */
42 static int nfs_opendir(struct inode *, struct file *);
43 static int nfs_readdir(struct file *, void *, filldir_t);
44 static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
45 static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
46 static int nfs_mkdir(struct inode *, struct dentry *, int);
47 static int nfs_rmdir(struct inode *, struct dentry *);
48 static int nfs_unlink(struct inode *, struct dentry *);
49 static int nfs_symlink(struct inode *, struct dentry *, const char *);
50 static int nfs_link(struct dentry *, struct inode *, struct dentry *);
51 static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
52 static int nfs_rename(struct inode *, struct dentry *,
53 struct inode *, struct dentry *);
54 static int nfs_fsync_dir(struct file *, struct dentry *, int);
55 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
57 struct file_operations nfs_dir_operations = {
58 .llseek = nfs_llseek_dir,
59 .read = generic_read_dir,
60 .readdir = nfs_readdir,
62 .release = nfs_release,
63 .fsync = nfs_fsync_dir,
66 struct inode_operations nfs_dir_inode_operations = {
71 .symlink = nfs_symlink,
76 .permission = nfs_permission,
77 .getattr = nfs_getattr,
78 .setattr = nfs_setattr,
82 struct inode_operations nfs3_dir_inode_operations = {
87 .symlink = nfs_symlink,
92 .permission = nfs_permission,
93 .getattr = nfs_getattr,
94 .setattr = nfs_setattr,
95 .listxattr = nfs3_listxattr,
96 .getxattr = nfs3_getxattr,
97 .setxattr = nfs3_setxattr,
98 .removexattr = nfs3_removexattr,
100 #endif /* CONFIG_NFS_V3 */
104 static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
105 struct inode_operations nfs4_dir_inode_operations = {
106 .create = nfs_create,
107 .lookup = nfs_atomic_lookup,
109 .unlink = nfs_unlink,
110 .symlink = nfs_symlink,
114 .rename = nfs_rename,
115 .permission = nfs_permission,
116 .getattr = nfs_getattr,
117 .setattr = nfs_setattr,
118 .getxattr = nfs4_getxattr,
119 .setxattr = nfs4_setxattr,
120 .listxattr = nfs4_listxattr,
123 #endif /* CONFIG_NFS_V4 */
129 nfs_opendir(struct inode *inode, struct file *filp)
134 /* Call generic open code in order to cache credentials */
136 res = nfs_open(inode, filp);
141 typedef u32 * (*decode_dirent_t)(u32 *, struct nfs_entry *, int);
145 unsigned long page_index;
148 loff_t current_index;
149 struct nfs_entry *entry;
150 decode_dirent_t decode;
153 } nfs_readdir_descriptor_t;
155 /* Now we cache directories properly, by stuffing the dirent
156 * data directly in the page cache.
158 * Inode invalidation due to refresh etc. takes care of
159 * _everything_, no sloppy entry flushing logic, no extraneous
160 * copying, network direct to page cache, the way it was meant
163 * NOTE: Dirent information verification is done always by the
164 * page-in of the RPC reply, nowhere else, this simplies
165 * things substantially.
168 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page *page)
170 struct file *file = desc->file;
171 struct inode *inode = file->f_dentry->d_inode;
172 struct rpc_cred *cred = nfs_file_cred(file);
173 unsigned long timestamp;
176 dfprintk(VFS, "NFS: nfs_readdir_filler() reading cookie %Lu into page %lu.\n", (long long)desc->entry->cookie, page->index);
180 error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, desc->entry->cookie, page,
181 NFS_SERVER(inode)->dtsize, desc->plus);
183 /* We requested READDIRPLUS, but the server doesn't grok it */
184 if (error == -ENOTSUPP && desc->plus) {
185 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
186 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
192 SetPageUptodate(page);
193 spin_lock(&inode->i_lock);
194 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
195 spin_unlock(&inode->i_lock);
196 /* Ensure consistent page alignment of the data.
197 * Note: assumes we have exclusive access to this mapping either
198 * through inode->i_mutex or some other mechanism.
200 if (page->index == 0)
201 invalidate_inode_pages2_range(inode->i_mapping, PAGE_CACHE_SIZE, -1);
207 nfs_zap_caches(inode);
213 int dir_decode(nfs_readdir_descriptor_t *desc)
216 p = desc->decode(p, desc->entry, desc->plus);
224 void dir_page_release(nfs_readdir_descriptor_t *desc)
227 page_cache_release(desc->page);
233 * Given a pointer to a buffer that has already been filled by a call
234 * to readdir, find the next entry with cookie '*desc->dir_cookie'.
236 * If the end of the buffer has been reached, return -EAGAIN, if not,
237 * return the offset within the buffer of the next entry to be
241 int find_dirent(nfs_readdir_descriptor_t *desc)
243 struct nfs_entry *entry = desc->entry;
247 while((status = dir_decode(desc)) == 0) {
248 dfprintk(VFS, "NFS: found cookie %Lu\n", (unsigned long long)entry->cookie);
249 if (entry->prev_cookie == *desc->dir_cookie)
251 if (loop_count++ > 200) {
256 dfprintk(VFS, "NFS: find_dirent() returns %d\n", status);
261 * Given a pointer to a buffer that has already been filled by a call
262 * to readdir, find the entry at offset 'desc->file->f_pos'.
264 * If the end of the buffer has been reached, return -EAGAIN, if not,
265 * return the offset within the buffer of the next entry to be
269 int find_dirent_index(nfs_readdir_descriptor_t *desc)
271 struct nfs_entry *entry = desc->entry;
276 status = dir_decode(desc);
280 dfprintk(VFS, "NFS: found cookie %Lu at index %Ld\n", (unsigned long long)entry->cookie, desc->current_index);
282 if (desc->file->f_pos == desc->current_index) {
283 *desc->dir_cookie = entry->cookie;
286 desc->current_index++;
287 if (loop_count++ > 200) {
292 dfprintk(VFS, "NFS: find_dirent_index() returns %d\n", status);
297 * Find the given page, and call find_dirent() or find_dirent_index in
298 * order to try to return the next entry.
301 int find_dirent_page(nfs_readdir_descriptor_t *desc)
303 struct inode *inode = desc->file->f_dentry->d_inode;
307 dfprintk(VFS, "NFS: find_dirent_page() searching directory page %ld\n", desc->page_index);
309 page = read_cache_page(inode->i_mapping, desc->page_index,
310 (filler_t *)nfs_readdir_filler, desc);
312 status = PTR_ERR(page);
315 if (!PageUptodate(page))
318 /* NOTE: Someone else may have changed the READDIRPLUS flag */
320 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
321 if (*desc->dir_cookie != 0)
322 status = find_dirent(desc);
324 status = find_dirent_index(desc);
326 dir_page_release(desc);
328 dfprintk(VFS, "NFS: find_dirent_page() returns %d\n", status);
331 page_cache_release(page);
336 * Recurse through the page cache pages, and return a
337 * filled nfs_entry structure of the next directory entry if possible.
339 * The target for the search is '*desc->dir_cookie' if non-0,
340 * 'desc->file->f_pos' otherwise
343 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
348 /* Always search-by-index from the beginning of the cache */
349 if (*desc->dir_cookie == 0) {
350 dfprintk(VFS, "NFS: readdir_search_pagecache() searching for offset %Ld\n", (long long)desc->file->f_pos);
351 desc->page_index = 0;
352 desc->entry->cookie = desc->entry->prev_cookie = 0;
353 desc->entry->eof = 0;
354 desc->current_index = 0;
356 dfprintk(VFS, "NFS: readdir_search_pagecache() searching for cookie %Lu\n", (unsigned long long)*desc->dir_cookie);
359 res = find_dirent_page(desc);
362 /* Align to beginning of next page */
364 if (loop_count++ > 200) {
369 dfprintk(VFS, "NFS: readdir_search_pagecache() returned %d\n", res);
373 static inline unsigned int dt_type(struct inode *inode)
375 return (inode->i_mode >> 12) & 15;
378 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc);
381 * Once we've found the start of the dirent within a page: fill 'er up...
384 int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
387 struct file *file = desc->file;
388 struct nfs_entry *entry = desc->entry;
389 struct dentry *dentry = NULL;
390 unsigned long fileid;
394 dfprintk(VFS, "NFS: nfs_do_filldir() filling starting @ cookie %Lu\n", (long long)entry->cookie);
397 unsigned d_type = DT_UNKNOWN;
398 /* Note: entry->prev_cookie contains the cookie for
399 * retrieving the current dirent on the server */
400 fileid = nfs_fileid_to_ino_t(entry->ino);
402 /* Get a dentry if we have one */
405 dentry = nfs_readdir_lookup(desc);
407 /* Use readdirplus info */
408 if (dentry != NULL && dentry->d_inode != NULL) {
409 d_type = dt_type(dentry->d_inode);
410 fileid = dentry->d_inode->i_ino;
413 res = filldir(dirent, entry->name, entry->len,
414 file->f_pos, fileid, d_type);
418 *desc->dir_cookie = entry->cookie;
419 if (dir_decode(desc) != 0) {
423 if (loop_count++ > 200) {
428 dir_page_release(desc);
431 dfprintk(VFS, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n", (unsigned long long)*desc->dir_cookie, res);
436 * If we cannot find a cookie in our cache, we suspect that this is
437 * because it points to a deleted file, so we ask the server to return
438 * whatever it thinks is the next entry. We then feed this to filldir.
439 * If all goes well, we should then be able to find our way round the
440 * cache on the next call to readdir_search_pagecache();
442 * NOTE: we cannot add the anonymous page to the pagecache because
443 * the data it contains might not be page aligned. Besides,
444 * we should already have a complete representation of the
445 * directory in the page cache by the time we get here.
448 int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
451 struct file *file = desc->file;
452 struct inode *inode = file->f_dentry->d_inode;
453 struct rpc_cred *cred = nfs_file_cred(file);
454 struct page *page = NULL;
457 dfprintk(VFS, "NFS: uncached_readdir() searching for cookie %Lu\n", (unsigned long long)*desc->dir_cookie);
459 page = alloc_page(GFP_HIGHUSER);
464 desc->error = NFS_PROTO(inode)->readdir(file->f_dentry, cred, *desc->dir_cookie,
466 NFS_SERVER(inode)->dtsize,
468 spin_lock(&inode->i_lock);
469 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATIME;
470 spin_unlock(&inode->i_lock);
472 desc->ptr = kmap(page); /* matching kunmap in nfs_do_filldir */
473 if (desc->error >= 0) {
474 if ((status = dir_decode(desc)) == 0)
475 desc->entry->prev_cookie = *desc->dir_cookie;
481 status = nfs_do_filldir(desc, dirent, filldir);
483 /* Reset read descriptor so it searches the page cache from
484 * the start upon the next call to readdir_search_pagecache() */
485 desc->page_index = 0;
486 desc->entry->cookie = desc->entry->prev_cookie = 0;
487 desc->entry->eof = 0;
489 dfprintk(VFS, "NFS: uncached_readdir() returns %d\n", status);
492 dir_page_release(desc);
496 /* The file offset position represents the dirent entry number. A
497 last cookie cache takes care of the common case of reading the
500 static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
502 struct dentry *dentry = filp->f_dentry;
503 struct inode *inode = dentry->d_inode;
504 nfs_readdir_descriptor_t my_desc,
506 struct nfs_entry my_entry;
508 struct nfs_fattr fattr;
511 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
515 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
522 * filp->f_pos points to the dirent entry number.
523 * *desc->dir_cookie has the cookie for the next entry. We have
524 * to either find the entry with the appropriate number or
525 * revalidate the cookie.
527 memset(desc, 0, sizeof(*desc));
530 desc->dir_cookie = &((struct nfs_open_context *)filp->private_data)->dir_cookie;
531 desc->decode = NFS_PROTO(inode)->decode_dirent;
532 desc->plus = NFS_USE_READDIRPLUS(inode);
534 my_entry.cookie = my_entry.prev_cookie = 0;
537 my_entry.fattr = &fattr;
538 nfs_fattr_init(&fattr);
539 desc->entry = &my_entry;
541 while(!desc->entry->eof) {
542 res = readdir_search_pagecache(desc);
544 if (res == -EBADCOOKIE) {
545 /* This means either end of directory */
546 if (*desc->dir_cookie && desc->entry->cookie != *desc->dir_cookie) {
547 /* Or that the server has 'lost' a cookie */
548 res = uncached_readdir(desc, dirent, filldir);
555 if (res == -ETOOSMALL && desc->plus) {
556 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_FLAGS(inode));
557 nfs_zap_caches(inode);
559 desc->entry->eof = 0;
565 res = nfs_do_filldir(desc, dirent, filldir);
577 loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
579 mutex_lock(&filp->f_dentry->d_inode->i_mutex);
582 offset += filp->f_pos;
590 if (offset != filp->f_pos) {
591 filp->f_pos = offset;
592 ((struct nfs_open_context *)filp->private_data)->dir_cookie = 0;
595 mutex_unlock(&filp->f_dentry->d_inode->i_mutex);
600 * All directory operations under NFS are synchronous, so fsync()
601 * is a dummy operation.
603 int nfs_fsync_dir(struct file *filp, struct dentry *dentry, int datasync)
609 * A check for whether or not the parent directory has changed.
610 * In the case it has, we assume that the dentries are untrustworthy
611 * and may need to be looked up again.
613 static inline int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
617 if ((NFS_I(dir)->cache_validity & NFS_INO_INVALID_ATTR) != 0
618 || nfs_attribute_timeout(dir))
620 return nfs_verify_change_attribute(dir, (unsigned long)dentry->d_fsdata);
623 static inline void nfs_set_verifier(struct dentry * dentry, unsigned long verf)
625 dentry->d_fsdata = (void *)verf;
629 * Whenever an NFS operation succeeds, we know that the dentry
630 * is valid, so we update the revalidation timestamp.
632 static inline void nfs_renew_times(struct dentry * dentry)
634 dentry->d_time = jiffies;
638 * Return the intent data that applies to this particular path component
640 * Note that the current set of intents only apply to the very last
641 * component of the path.
642 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
644 static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd, unsigned int mask)
646 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
648 return nd->flags & mask;
652 * Inode and filehandle revalidation for lookups.
654 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
655 * or if the intent information indicates that we're about to open this
656 * particular file and the "nocto" mount flag is not set.
660 int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
662 struct nfs_server *server = NFS_SERVER(inode);
665 /* VFS wants an on-the-wire revalidation */
666 if (nd->flags & LOOKUP_REVAL)
668 /* This is an open(2) */
669 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
670 !(server->flags & NFS_MOUNT_NOCTO))
673 return nfs_revalidate_inode(server, inode);
675 return __nfs_revalidate_inode(server, inode);
679 * We judge how long we want to trust negative
680 * dentries by looking at the parent inode mtime.
682 * If parent mtime has changed, we revalidate, else we wait for a
683 * period corresponding to the parent's attribute cache timeout value.
686 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
687 struct nameidata *nd)
689 /* Don't revalidate a negative dentry if we're creating a new file */
690 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
692 return !nfs_check_verifier(dir, dentry);
696 * This is called every time the dcache has a lookup hit,
697 * and we should check whether we can really trust that
700 * NOTE! The hit can be a negative hit too, don't assume
703 * If the parent directory is seen to have changed, we throw out the
704 * cached dentry and do a new lookup.
706 static int nfs_lookup_revalidate(struct dentry * dentry, struct nameidata *nd)
710 struct dentry *parent;
712 struct nfs_fh fhandle;
713 struct nfs_fattr fattr;
714 unsigned long verifier;
716 parent = dget_parent(dentry);
718 dir = parent->d_inode;
719 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
720 inode = dentry->d_inode;
723 if (nfs_neg_need_reval(dir, dentry, nd))
728 if (is_bad_inode(inode)) {
729 dfprintk(VFS, "nfs_lookup_validate: %s/%s has dud inode\n",
730 dentry->d_parent->d_name.name, dentry->d_name.name);
734 /* Revalidate parent directory attribute cache */
735 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
738 /* Force a full look up iff the parent directory has changed */
739 if (nfs_check_verifier(dir, dentry)) {
740 if (nfs_lookup_verify_inode(inode, nd))
745 if (NFS_STALE(inode))
748 verifier = nfs_save_change_attribute(dir);
749 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
752 if (nfs_compare_fh(NFS_FH(inode), &fhandle))
754 if ((error = nfs_refresh_inode(inode, &fattr)) != 0)
757 nfs_renew_times(dentry);
758 nfs_set_verifier(dentry, verifier);
767 if (inode && S_ISDIR(inode->i_mode)) {
768 /* Purge readdir caches. */
769 nfs_zap_caches(inode);
770 /* If we have submounts, don't unhash ! */
771 if (have_submounts(dentry))
773 shrink_dcache_parent(dentry);
782 * This is called from dput() when d_count is going to 0.
784 static int nfs_dentry_delete(struct dentry *dentry)
786 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
787 dentry->d_parent->d_name.name, dentry->d_name.name,
790 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
791 /* Unhash it, so that ->d_iput() would be called */
794 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
795 /* Unhash it, so that ancestors of killed async unlink
796 * files will be cleaned up during umount */
804 * Called when the dentry loses inode.
805 * We use it to clean up silly-renamed files.
807 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
809 nfs_inode_return_delegation(inode);
810 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
813 nfs_complete_unlink(dentry);
816 /* When creating a negative dentry, we want to renew d_time */
817 nfs_renew_times(dentry);
821 struct dentry_operations nfs_dentry_operations = {
822 .d_revalidate = nfs_lookup_revalidate,
823 .d_delete = nfs_dentry_delete,
824 .d_iput = nfs_dentry_iput,
828 * Use intent information to check whether or not we're going to do
829 * an O_EXCL create using this path component.
832 int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
834 if (NFS_PROTO(dir)->version == 2)
836 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_CREATE) == 0)
838 return (nd->intent.open.flags & O_EXCL) != 0;
841 static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
844 struct inode *inode = NULL;
846 struct nfs_fh fhandle;
847 struct nfs_fattr fattr;
849 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
850 dentry->d_parent->d_name.name, dentry->d_name.name);
851 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
853 res = ERR_PTR(-ENAMETOOLONG);
854 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
857 res = ERR_PTR(-ENOMEM);
858 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
862 /* If we're doing an exclusive create, optimize away the lookup */
863 if (nfs_is_exclusive_create(dir, nd))
866 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, &fhandle, &fattr);
867 if (error == -ENOENT)
870 res = ERR_PTR(error);
873 res = ERR_PTR(-EACCES);
874 inode = nfs_fhget(dentry->d_sb, &fhandle, &fattr);
878 res = d_add_unique(dentry, inode);
881 nfs_renew_times(dentry);
882 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
890 static int nfs_open_revalidate(struct dentry *, struct nameidata *);
892 struct dentry_operations nfs4_dentry_operations = {
893 .d_revalidate = nfs_open_revalidate,
894 .d_delete = nfs_dentry_delete,
895 .d_iput = nfs_dentry_iput,
899 * Use intent information to determine whether we need to substitute
900 * the NFSv4-style stateful OPEN for the LOOKUP call
902 static int is_atomic_open(struct inode *dir, struct nameidata *nd)
904 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
906 /* NFS does not (yet) have a stateful open for directories */
907 if (nd->flags & LOOKUP_DIRECTORY)
909 /* Are we trying to write to a read only partition? */
910 if (IS_RDONLY(dir) && (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
915 static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
917 struct dentry *res = NULL;
920 /* Check that we are indeed trying to open this file */
921 if (!is_atomic_open(dir, nd))
924 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
925 res = ERR_PTR(-ENAMETOOLONG);
928 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
930 /* Let vfs_create() deal with O_EXCL */
931 if (nd->intent.open.flags & O_EXCL) {
936 /* Open the file on the server */
938 /* Revalidate parent directory attribute cache */
939 error = nfs_revalidate_inode(NFS_SERVER(dir), dir);
941 res = ERR_PTR(error);
946 if (nd->intent.open.flags & O_CREAT) {
947 nfs_begin_data_update(dir);
948 res = nfs4_atomic_open(dir, dentry, nd);
949 nfs_end_data_update(dir);
951 res = nfs4_atomic_open(dir, dentry, nd);
954 error = PTR_ERR(res);
956 /* Make a negative dentry */
960 /* This turned out not to be a regular file */
965 if (!(nd->intent.open.flags & O_NOFOLLOW))
971 } else if (res != NULL)
973 nfs_renew_times(dentry);
974 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
978 return nfs_lookup(dir, dentry, nd);
981 static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
983 struct dentry *parent = NULL;
984 struct inode *inode = dentry->d_inode;
986 unsigned long verifier;
987 int openflags, ret = 0;
989 parent = dget_parent(dentry);
990 dir = parent->d_inode;
991 if (!is_atomic_open(dir, nd))
993 /* We can't create new files in nfs_open_revalidate(), so we
994 * optimize away revalidation of negative dentries.
998 /* NFS only supports OPEN on regular files */
999 if (!S_ISREG(inode->i_mode))
1001 openflags = nd->intent.open.flags;
1002 /* We cannot do exclusive creation on a positive dentry */
1003 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
1005 /* We can't create new files, or truncate existing ones here */
1006 openflags &= ~(O_CREAT|O_TRUNC);
1009 * Note: we're not holding inode->i_mutex and so may be racing with
1010 * operations that change the directory. We therefore save the
1011 * change attribute *before* we do the RPC call.
1014 verifier = nfs_save_change_attribute(dir);
1015 ret = nfs4_open_revalidate(dir, dentry, openflags, nd);
1017 nfs_set_verifier(dentry, verifier);
1026 if (inode != NULL && nfs_have_delegation(inode, FMODE_READ))
1028 return nfs_lookup_revalidate(dentry, nd);
1030 #endif /* CONFIG_NFSV4 */
1032 static struct dentry *nfs_readdir_lookup(nfs_readdir_descriptor_t *desc)
1034 struct dentry *parent = desc->file->f_dentry;
1035 struct inode *dir = parent->d_inode;
1036 struct nfs_entry *entry = desc->entry;
1037 struct dentry *dentry, *alias;
1038 struct qstr name = {
1039 .name = entry->name,
1042 struct inode *inode;
1046 if (name.name[0] == '.' && name.name[1] == '.')
1047 return dget_parent(parent);
1050 if (name.name[0] == '.')
1051 return dget(parent);
1053 name.hash = full_name_hash(name.name, name.len);
1054 dentry = d_lookup(parent, &name);
1057 if (!desc->plus || !(entry->fattr->valid & NFS_ATTR_FATTR))
1059 /* Note: caller is already holding the dir->i_mutex! */
1060 dentry = d_alloc(parent, &name);
1063 dentry->d_op = NFS_PROTO(dir)->dentry_ops;
1064 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
1069 alias = d_add_unique(dentry, inode);
1070 if (alias != NULL) {
1074 nfs_renew_times(dentry);
1075 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1080 * Code common to create, mkdir, and mknod.
1082 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1083 struct nfs_fattr *fattr)
1085 struct inode *inode;
1086 int error = -EACCES;
1088 /* We may have been initialized further down */
1089 if (dentry->d_inode)
1091 if (fhandle->size == 0) {
1092 struct inode *dir = dentry->d_parent->d_inode;
1093 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr);
1097 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1098 struct nfs_server *server = NFS_SB(dentry->d_sb);
1099 error = server->rpc_ops->getattr(server, fhandle, fattr);
1104 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1107 d_instantiate(dentry, inode);
1115 * Following a failed create operation, we drop the dentry rather
1116 * than retain a negative dentry. This avoids a problem in the event
1117 * that the operation succeeded on the server, but an error in the
1118 * reply path made it appear to have failed.
1120 static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1121 struct nameidata *nd)
1127 dfprintk(VFS, "NFS: create(%s/%ld, %s\n", dir->i_sb->s_id,
1128 dir->i_ino, dentry->d_name.name);
1130 attr.ia_mode = mode;
1131 attr.ia_valid = ATTR_MODE;
1133 if (nd && (nd->flags & LOOKUP_CREATE))
1134 open_flags = nd->intent.open.flags;
1137 nfs_begin_data_update(dir);
1138 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, nd);
1139 nfs_end_data_update(dir);
1142 nfs_renew_times(dentry);
1143 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1153 * See comments for nfs_proc_create regarding failed operations.
1156 nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1161 dfprintk(VFS, "NFS: mknod(%s/%ld, %s\n", dir->i_sb->s_id,
1162 dir->i_ino, dentry->d_name.name);
1164 if (!new_valid_dev(rdev))
1167 attr.ia_mode = mode;
1168 attr.ia_valid = ATTR_MODE;
1171 nfs_begin_data_update(dir);
1172 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1173 nfs_end_data_update(dir);
1176 nfs_renew_times(dentry);
1177 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1187 * See comments for nfs_proc_create regarding failed operations.
1189 static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1194 dfprintk(VFS, "NFS: mkdir(%s/%ld, %s\n", dir->i_sb->s_id,
1195 dir->i_ino, dentry->d_name.name);
1197 attr.ia_valid = ATTR_MODE;
1198 attr.ia_mode = mode | S_IFDIR;
1201 nfs_begin_data_update(dir);
1202 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1203 nfs_end_data_update(dir);
1206 nfs_renew_times(dentry);
1207 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1216 static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1220 dfprintk(VFS, "NFS: rmdir(%s/%ld, %s\n", dir->i_sb->s_id,
1221 dir->i_ino, dentry->d_name.name);
1224 nfs_begin_data_update(dir);
1225 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1226 /* Ensure the VFS deletes this inode */
1227 if (error == 0 && dentry->d_inode != NULL)
1228 dentry->d_inode->i_nlink = 0;
1229 nfs_end_data_update(dir);
1235 static int nfs_sillyrename(struct inode *dir, struct dentry *dentry)
1237 static unsigned int sillycounter;
1238 const int i_inosize = sizeof(dir->i_ino)*2;
1239 const int countersize = sizeof(sillycounter)*2;
1240 const int slen = sizeof(".nfs") + i_inosize + countersize - 1;
1243 struct dentry *sdentry;
1246 dfprintk(VFS, "NFS: silly-rename(%s/%s, ct=%d)\n",
1247 dentry->d_parent->d_name.name, dentry->d_name.name,
1248 atomic_read(&dentry->d_count));
1249 nfs_inc_stats(dir, NFSIOS_SILLYRENAME);
1252 if (!dentry->d_inode)
1253 printk("NFS: silly-renaming %s/%s, negative dentry??\n",
1254 dentry->d_parent->d_name.name, dentry->d_name.name);
1257 * We don't allow a dentry to be silly-renamed twice.
1260 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1263 sprintf(silly, ".nfs%*.*lx",
1264 i_inosize, i_inosize, dentry->d_inode->i_ino);
1266 /* Return delegation in anticipation of the rename */
1267 nfs_inode_return_delegation(dentry->d_inode);
1271 char *suffix = silly + slen - countersize;
1275 sprintf(suffix, "%*.*x", countersize, countersize, sillycounter);
1277 dfprintk(VFS, "trying to rename %s to %s\n",
1278 dentry->d_name.name, silly);
1280 sdentry = lookup_one_len(silly, dentry->d_parent, slen);
1282 * N.B. Better to return EBUSY here ... it could be
1283 * dangerous to delete the file while it's in use.
1285 if (IS_ERR(sdentry))
1287 } while(sdentry->d_inode != NULL); /* need negative lookup */
1289 qsilly.name = silly;
1290 qsilly.len = strlen(silly);
1291 nfs_begin_data_update(dir);
1292 if (dentry->d_inode) {
1293 nfs_begin_data_update(dentry->d_inode);
1294 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1296 nfs_mark_for_revalidate(dentry->d_inode);
1297 nfs_end_data_update(dentry->d_inode);
1299 error = NFS_PROTO(dir)->rename(dir, &dentry->d_name,
1301 nfs_end_data_update(dir);
1303 nfs_renew_times(dentry);
1304 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1305 d_move(dentry, sdentry);
1306 error = nfs_async_unlink(dentry);
1307 /* If we return 0 we don't unlink */
1315 * Remove a file after making sure there are no pending writes,
1316 * and after checking that the file has only one user.
1318 * We invalidate the attribute cache and free the inode prior to the operation
1319 * to avoid possible races if the server reuses the inode.
1321 static int nfs_safe_remove(struct dentry *dentry)
1323 struct inode *dir = dentry->d_parent->d_inode;
1324 struct inode *inode = dentry->d_inode;
1327 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1328 dentry->d_parent->d_name.name, dentry->d_name.name);
1330 /* If the dentry was sillyrenamed, we simply call d_delete() */
1331 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1336 nfs_begin_data_update(dir);
1337 if (inode != NULL) {
1338 nfs_inode_return_delegation(inode);
1339 nfs_begin_data_update(inode);
1340 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1341 /* The VFS may want to delete this inode */
1344 nfs_mark_for_revalidate(inode);
1345 nfs_end_data_update(inode);
1347 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1348 nfs_end_data_update(dir);
1353 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1354 * belongs to an active ".nfs..." file and we return -EBUSY.
1356 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1358 static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1361 int need_rehash = 0;
1363 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1364 dir->i_ino, dentry->d_name.name);
1367 spin_lock(&dcache_lock);
1368 spin_lock(&dentry->d_lock);
1369 if (atomic_read(&dentry->d_count) > 1) {
1370 spin_unlock(&dentry->d_lock);
1371 spin_unlock(&dcache_lock);
1372 error = nfs_sillyrename(dir, dentry);
1376 if (!d_unhashed(dentry)) {
1380 spin_unlock(&dentry->d_lock);
1381 spin_unlock(&dcache_lock);
1382 error = nfs_safe_remove(dentry);
1384 nfs_renew_times(dentry);
1385 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1386 } else if (need_rehash)
1393 nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1396 struct nfs_fattr sym_attr;
1397 struct nfs_fh sym_fh;
1398 struct qstr qsymname;
1401 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1402 dir->i_ino, dentry->d_name.name, symname);
1405 if (dentry->d_inode)
1406 printk("nfs_proc_symlink: %s/%s not negative!\n",
1407 dentry->d_parent->d_name.name, dentry->d_name.name);
1410 * Fill in the sattr for the call.
1411 * Note: SunOS 4.1.2 crashes if the mode isn't initialized!
1413 attr.ia_valid = ATTR_MODE;
1414 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1416 qsymname.name = symname;
1417 qsymname.len = strlen(symname);
1420 nfs_begin_data_update(dir);
1421 error = NFS_PROTO(dir)->symlink(dir, &dentry->d_name, &qsymname,
1422 &attr, &sym_fh, &sym_attr);
1423 nfs_end_data_update(dir);
1425 error = nfs_instantiate(dentry, &sym_fh, &sym_attr);
1427 if (error == -EEXIST)
1428 printk("nfs_proc_symlink: %s/%s already exists??\n",
1429 dentry->d_parent->d_name.name, dentry->d_name.name);
1437 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1439 struct inode *inode = old_dentry->d_inode;
1442 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1443 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1444 dentry->d_parent->d_name.name, dentry->d_name.name);
1447 nfs_begin_data_update(dir);
1448 nfs_begin_data_update(inode);
1449 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1451 atomic_inc(&inode->i_count);
1452 d_instantiate(dentry, inode);
1454 nfs_end_data_update(inode);
1455 nfs_end_data_update(dir);
1462 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1463 * different file handle for the same inode after a rename (e.g. when
1464 * moving to a different directory). A fail-safe method to do so would
1465 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1466 * rename the old file using the sillyrename stuff. This way, the original
1467 * file in old_dir will go away when the last process iput()s the inode.
1471 * It actually works quite well. One needs to have the possibility for
1472 * at least one ".nfs..." file in each directory the file ever gets
1473 * moved or linked to which happens automagically with the new
1474 * implementation that only depends on the dcache stuff instead of
1475 * using the inode layer
1477 * Unfortunately, things are a little more complicated than indicated
1478 * above. For a cross-directory move, we want to make sure we can get
1479 * rid of the old inode after the operation. This means there must be
1480 * no pending writes (if it's a file), and the use count must be 1.
1481 * If these conditions are met, we can drop the dentries before doing
1484 static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1485 struct inode *new_dir, struct dentry *new_dentry)
1487 struct inode *old_inode = old_dentry->d_inode;
1488 struct inode *new_inode = new_dentry->d_inode;
1489 struct dentry *dentry = NULL, *rehash = NULL;
1493 * To prevent any new references to the target during the rename,
1494 * we unhash the dentry and free the inode in advance.
1497 if (!d_unhashed(new_dentry)) {
1499 rehash = new_dentry;
1502 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1503 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1504 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
1505 atomic_read(&new_dentry->d_count));
1508 * First check whether the target is busy ... we can't
1509 * safely do _any_ rename if the target is in use.
1511 * For files, make a copy of the dentry and then do a
1512 * silly-rename. If the silly-rename succeeds, the
1513 * copied dentry is hashed and becomes the new target.
1517 if (S_ISDIR(new_inode->i_mode)) {
1519 if (!S_ISDIR(old_inode->i_mode))
1521 } else if (atomic_read(&new_dentry->d_count) > 2) {
1523 /* copy the target dentry's name */
1524 dentry = d_alloc(new_dentry->d_parent,
1525 &new_dentry->d_name);
1529 /* silly-rename the existing target ... */
1530 err = nfs_sillyrename(new_dir, new_dentry);
1532 new_dentry = rehash = dentry;
1534 /* instantiate the replacement target */
1535 d_instantiate(new_dentry, NULL);
1536 } else if (atomic_read(&new_dentry->d_count) > 1) {
1537 /* dentry still busy? */
1539 printk("nfs_rename: target %s/%s busy, d_count=%d\n",
1540 new_dentry->d_parent->d_name.name,
1541 new_dentry->d_name.name,
1542 atomic_read(&new_dentry->d_count));
1547 new_inode->i_nlink--;
1551 * ... prune child dentries and writebacks if needed.
1553 if (atomic_read(&old_dentry->d_count) > 1) {
1554 nfs_wb_all(old_inode);
1555 shrink_dcache_parent(old_dentry);
1557 nfs_inode_return_delegation(old_inode);
1559 if (new_inode != NULL) {
1560 nfs_inode_return_delegation(new_inode);
1561 d_delete(new_dentry);
1564 nfs_begin_data_update(old_dir);
1565 nfs_begin_data_update(new_dir);
1566 nfs_begin_data_update(old_inode);
1567 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1568 new_dir, &new_dentry->d_name);
1569 nfs_mark_for_revalidate(old_inode);
1570 nfs_end_data_update(old_inode);
1571 nfs_end_data_update(new_dir);
1572 nfs_end_data_update(old_dir);
1577 if (!S_ISDIR(old_inode->i_mode))
1578 d_move(old_dentry, new_dentry);
1579 nfs_renew_times(new_dentry);
1580 nfs_set_verifier(new_dentry, nfs_save_change_attribute(new_dir));
1583 /* new dentry created? */
1590 int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
1592 struct nfs_inode *nfsi = NFS_I(inode);
1593 struct nfs_access_entry *cache = &nfsi->cache_access;
1595 if (cache->cred != cred
1596 || time_after(jiffies, cache->jiffies + NFS_ATTRTIMEO(inode))
1597 || (nfsi->cache_validity & NFS_INO_INVALID_ACCESS))
1599 memcpy(res, cache, sizeof(*res));
1603 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
1605 struct nfs_inode *nfsi = NFS_I(inode);
1606 struct nfs_access_entry *cache = &nfsi->cache_access;
1608 if (cache->cred != set->cred) {
1610 put_rpccred(cache->cred);
1611 cache->cred = get_rpccred(set->cred);
1613 /* FIXME: replace current access_cache BKL reliance with inode->i_lock */
1614 spin_lock(&inode->i_lock);
1615 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
1616 spin_unlock(&inode->i_lock);
1617 cache->jiffies = set->jiffies;
1618 cache->mask = set->mask;
1621 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
1623 struct nfs_access_entry cache;
1626 status = nfs_access_get_cached(inode, cred, &cache);
1630 /* Be clever: ask server to check for all possible rights */
1631 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
1633 cache.jiffies = jiffies;
1634 status = NFS_PROTO(inode)->access(inode, &cache);
1637 nfs_access_add_cache(inode, &cache);
1639 if ((cache.mask & mask) == mask)
1644 int nfs_permission(struct inode *inode, int mask, struct nameidata *nd)
1646 struct rpc_cred *cred;
1649 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
1653 /* Is this sys_access() ? */
1654 if (nd != NULL && (nd->flags & LOOKUP_ACCESS))
1657 switch (inode->i_mode & S_IFMT) {
1661 /* NFSv4 has atomic_open... */
1662 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
1664 && (nd->flags & LOOKUP_OPEN))
1669 * Optimize away all write operations, since the server
1670 * will check permissions when we perform the op.
1672 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
1679 if (!NFS_PROTO(inode)->access)
1682 cred = rpcauth_lookupcred(NFS_CLIENT(inode)->cl_auth, 0);
1683 if (!IS_ERR(cred)) {
1684 res = nfs_do_access(inode, cred, mask);
1687 res = PTR_ERR(cred);
1692 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
1694 res = generic_permission(inode, mask, NULL);
1701 * version-control: t
1702 * kept-new-versions: 5