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
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(file_inode(filp), 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 *, bool);
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 * we are freeing strings created by nfs_add_to_readdir_array()
176 void nfs_readdir_clear_array(struct page *page)
178 struct nfs_cache_array *array;
181 array = kmap_atomic(page);
182 for (i = 0; i < array->size; i++)
183 kfree(array->array[i].string.name);
184 kunmap_atomic(array);
188 * the caller is responsible for freeing qstr.name
189 * when called by nfs_readdir_add_to_array, the strings will be freed in
190 * nfs_clear_readdir_array()
193 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
196 string->name = kmemdup(name, len, GFP_KERNEL);
197 if (string->name == NULL)
200 * Avoid a kmemleak false positive. The pointer to the name is stored
201 * in a page cache page which kmemleak does not scan.
203 kmemleak_not_leak(string->name);
204 string->hash = full_name_hash(NULL, name, len);
209 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
211 struct nfs_cache_array *array = kmap(page);
212 struct nfs_cache_array_entry *cache_entry;
215 cache_entry = &array->array[array->size];
217 /* Check that this entry lies within the page bounds */
219 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
222 cache_entry->cookie = entry->prev_cookie;
223 cache_entry->ino = entry->ino;
224 cache_entry->d_type = entry->d_type;
225 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
228 array->last_cookie = entry->cookie;
231 array->eof_index = array->size;
238 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
240 loff_t diff = desc->ctx->pos - desc->current_index;
245 if (diff >= array->size) {
246 if (array->eof_index >= 0)
251 index = (unsigned int)diff;
252 *desc->dir_cookie = array->array[index].cookie;
253 desc->cache_entry_index = index;
261 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
263 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
266 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
270 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
274 int status = -EAGAIN;
276 for (i = 0; i < array->size; i++) {
277 if (array->array[i].cookie == *desc->dir_cookie) {
278 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
279 struct nfs_open_dir_context *ctx = desc->file->private_data;
281 new_pos = desc->current_index + i;
282 if (ctx->attr_gencount != nfsi->attr_gencount ||
283 !nfs_readdir_inode_mapping_valid(nfsi)) {
285 ctx->attr_gencount = nfsi->attr_gencount;
286 } else if (new_pos < desc->ctx->pos) {
288 && ctx->dup_cookie == *desc->dir_cookie) {
289 if (printk_ratelimit()) {
290 pr_notice("NFS: directory %pD2 contains a readdir loop."
291 "Please contact your server vendor. "
292 "The file: %.*s has duplicate cookie %llu\n",
293 desc->file, array->array[i].string.len,
294 array->array[i].string.name, *desc->dir_cookie);
299 ctx->dup_cookie = *desc->dir_cookie;
302 desc->ctx->pos = new_pos;
303 desc->cache_entry_index = i;
307 if (array->eof_index >= 0) {
308 status = -EBADCOOKIE;
309 if (*desc->dir_cookie == array->last_cookie)
317 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
319 struct nfs_cache_array *array;
322 array = kmap(desc->page);
324 if (*desc->dir_cookie == 0)
325 status = nfs_readdir_search_for_pos(array, desc);
327 status = nfs_readdir_search_for_cookie(array, desc);
329 if (status == -EAGAIN) {
330 desc->last_cookie = array->last_cookie;
331 desc->current_index += array->size;
338 /* Fill a page with xdr information before transferring to the cache page */
340 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
341 struct nfs_entry *entry, struct file *file, struct inode *inode)
343 struct nfs_open_dir_context *ctx = file->private_data;
344 struct rpc_cred *cred = ctx->cred;
345 unsigned long timestamp, gencount;
350 gencount = nfs_inc_attr_generation_counter();
351 error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages,
352 NFS_SERVER(inode)->dtsize, desc->plus);
354 /* We requested READDIRPLUS, but the server doesn't grok it */
355 if (error == -ENOTSUPP && desc->plus) {
356 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
357 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
363 desc->timestamp = timestamp;
364 desc->gencount = gencount;
369 static int xdr_decode(nfs_readdir_descriptor_t *desc,
370 struct nfs_entry *entry, struct xdr_stream *xdr)
374 error = desc->decode(xdr, entry, desc->plus);
377 entry->fattr->time_start = desc->timestamp;
378 entry->fattr->gencount = desc->gencount;
382 /* Match file and dirent using either filehandle or fileid
383 * Note: caller is responsible for checking the fsid
386 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
389 struct nfs_inode *nfsi;
391 if (d_really_is_negative(dentry))
394 inode = d_inode(dentry);
395 if (is_bad_inode(inode) || NFS_STALE(inode))
399 if (entry->fattr->fileid != nfsi->fileid)
401 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
407 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
409 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
411 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
419 * This function is called by the lookup and getattr code to request the
420 * use of readdirplus to accelerate any future lookups in the same
423 void nfs_advise_use_readdirplus(struct inode *dir)
425 struct nfs_inode *nfsi = NFS_I(dir);
427 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
428 !list_empty(&nfsi->open_files))
429 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
433 * This function is mainly for use by nfs_getattr().
435 * If this is an 'ls -l', we want to force use of readdirplus.
436 * Do this by checking if there is an active file descriptor
437 * and calling nfs_advise_use_readdirplus, then forcing a
440 void nfs_force_use_readdirplus(struct inode *dir)
442 struct nfs_inode *nfsi = NFS_I(dir);
444 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
445 !list_empty(&nfsi->open_files)) {
446 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
447 invalidate_mapping_pages(dir->i_mapping, 0, -1);
452 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
454 struct qstr filename = QSTR_INIT(entry->name, entry->len);
455 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
456 struct dentry *dentry;
457 struct dentry *alias;
458 struct inode *dir = d_inode(parent);
462 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
464 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
466 if (filename.len == 0)
468 /* Validate that the name doesn't contain any illegal '\0' */
469 if (strnlen(filename.name, filename.len) != filename.len)
472 if (strnchr(filename.name, filename.len, '/'))
474 if (filename.name[0] == '.') {
475 if (filename.len == 1)
477 if (filename.len == 2 && filename.name[1] == '.')
480 filename.hash = full_name_hash(parent, filename.name, filename.len);
482 dentry = d_lookup(parent, &filename);
485 dentry = d_alloc_parallel(parent, &filename, &wq);
489 if (!d_in_lookup(dentry)) {
490 /* Is there a mountpoint here? If so, just exit */
491 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
492 &entry->fattr->fsid))
494 if (nfs_same_file(dentry, entry)) {
495 if (!entry->fh->size)
497 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
498 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
500 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
503 d_invalidate(dentry);
509 if (!entry->fh->size) {
510 d_lookup_done(dentry);
514 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
515 alias = d_splice_alias(inode, dentry);
516 d_lookup_done(dentry);
523 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
528 /* Perform conversion from xdr to cache array */
530 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
531 struct page **xdr_pages, struct page *page, unsigned int buflen)
533 struct xdr_stream stream;
535 struct page *scratch;
536 struct nfs_cache_array *array;
537 unsigned int count = 0;
540 scratch = alloc_page(GFP_KERNEL);
547 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
548 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
551 status = xdr_decode(desc, entry, &stream);
553 if (status == -EAGAIN)
561 nfs_prime_dcache(file_dentry(desc->file), entry);
563 status = nfs_readdir_add_to_array(entry, page);
566 } while (!entry->eof);
569 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
571 array->eof_index = array->size;
581 void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
584 for (i = 0; i < npages; i++)
589 * nfs_readdir_large_page will allocate pages that must be freed with a call
590 * to nfs_readdir_free_pagearray
593 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
597 for (i = 0; i < npages; i++) {
598 struct page *page = alloc_page(GFP_KERNEL);
606 nfs_readdir_free_pages(pages, i);
611 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
613 struct page *pages[NFS_MAX_READDIR_PAGES];
614 struct nfs_entry entry;
615 struct file *file = desc->file;
616 struct nfs_cache_array *array;
617 int status = -ENOMEM;
618 unsigned int array_size = ARRAY_SIZE(pages);
620 entry.prev_cookie = 0;
621 entry.cookie = desc->last_cookie;
623 entry.fh = nfs_alloc_fhandle();
624 entry.fattr = nfs_alloc_fattr();
625 entry.server = NFS_SERVER(inode);
626 if (entry.fh == NULL || entry.fattr == NULL)
629 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
630 if (IS_ERR(entry.label)) {
631 status = PTR_ERR(entry.label);
636 memset(array, 0, sizeof(struct nfs_cache_array));
637 array->eof_index = -1;
639 status = nfs_readdir_alloc_pages(pages, array_size);
641 goto out_release_array;
644 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
649 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
651 if (status == -ENOSPC)
655 } while (array->eof_index < 0);
657 nfs_readdir_free_pages(pages, array_size);
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 put_page(desc->page);
705 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
707 return read_cache_page(desc->file->f_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 = kmap(desc->page);
758 for (i = desc->cache_entry_index; i < array->size; i++) {
759 struct nfs_cache_array_entry *ent;
761 ent = &array->array[i];
762 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
763 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
768 if (i < (array->size-1))
769 *desc->dir_cookie = array->array[i+1].cookie;
771 *desc->dir_cookie = array->last_cookie;
775 if (array->eof_index >= 0)
779 cache_page_release(desc);
780 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
781 (unsigned long long)*desc->dir_cookie, res);
786 * If we cannot find a cookie in our cache, we suspect that this is
787 * because it points to a deleted file, so we ask the server to return
788 * whatever it thinks is the next entry. We then feed this to filldir.
789 * If all goes well, we should then be able to find our way round the
790 * cache on the next call to readdir_search_pagecache();
792 * NOTE: we cannot add the anonymous page to the pagecache because
793 * the data it contains might not be page aligned. Besides,
794 * we should already have a complete representation of the
795 * directory in the page cache by the time we get here.
798 int uncached_readdir(nfs_readdir_descriptor_t *desc)
800 struct page *page = NULL;
802 struct inode *inode = file_inode(desc->file);
803 struct nfs_open_dir_context *ctx = desc->file->private_data;
805 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
806 (unsigned long long)*desc->dir_cookie);
808 page = alloc_page(GFP_HIGHUSER);
814 desc->page_index = 0;
815 desc->last_cookie = *desc->dir_cookie;
819 status = nfs_readdir_xdr_to_array(desc, page, inode);
823 status = nfs_do_filldir(desc);
826 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
830 cache_page_release(desc);
834 /* The file offset position represents the dirent entry number. A
835 last cookie cache takes care of the common case of reading the
838 static int nfs_readdir(struct file *file, struct dir_context *ctx)
840 struct dentry *dentry = file_dentry(file);
841 struct inode *inode = d_inode(dentry);
842 nfs_readdir_descriptor_t my_desc,
844 struct nfs_open_dir_context *dir_ctx = file->private_data;
847 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
848 file, (long long)ctx->pos);
849 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
852 * ctx->pos points to the dirent entry number.
853 * *desc->dir_cookie has the cookie for the next entry. We have
854 * to either find the entry with the appropriate number or
855 * revalidate the cookie.
857 memset(desc, 0, sizeof(*desc));
861 desc->dir_cookie = &dir_ctx->dir_cookie;
862 desc->decode = NFS_PROTO(inode)->decode_dirent;
863 desc->plus = nfs_use_readdirplus(inode, ctx);
865 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
866 res = nfs_revalidate_mapping(inode, file->f_mapping);
871 res = readdir_search_pagecache(desc);
873 if (res == -EBADCOOKIE) {
875 /* This means either end of directory */
876 if (*desc->dir_cookie && desc->eof == 0) {
877 /* Or that the server has 'lost' a cookie */
878 res = uncached_readdir(desc);
884 if (res == -ETOOSMALL && desc->plus) {
885 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
886 nfs_zap_caches(inode);
887 desc->page_index = 0;
895 res = nfs_do_filldir(desc);
898 } while (!desc->eof);
902 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
906 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
908 struct inode *inode = file_inode(filp);
909 struct nfs_open_dir_context *dir_ctx = filp->private_data;
911 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
912 filp, offset, whence);
917 offset += filp->f_pos;
925 if (offset != filp->f_pos) {
926 filp->f_pos = offset;
927 dir_ctx->dir_cookie = 0;
936 * All directory operations under NFS are synchronous, so fsync()
937 * is a dummy operation.
939 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
942 struct inode *inode = file_inode(filp);
944 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
947 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
953 * nfs_force_lookup_revalidate - Mark the directory as having changed
954 * @dir - pointer to directory inode
956 * This forces the revalidation code in nfs_lookup_revalidate() to do a
957 * full lookup on all child dentries of 'dir' whenever a change occurs
958 * on the server that might have invalidated our dcache.
960 * The caller should be holding dir->i_lock
962 void nfs_force_lookup_revalidate(struct inode *dir)
964 NFS_I(dir)->cache_change_attribute++;
966 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
969 * A check for whether or not the parent directory has changed.
970 * In the case it has, we assume that the dentries are untrustworthy
971 * and may need to be looked up again.
972 * If rcu_walk prevents us from performing a full check, return 0.
974 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
979 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
981 if (!nfs_verify_change_attribute(dir, dentry->d_time))
983 /* Revalidate nfsi->cache_change_attribute before we declare a match */
984 if (nfs_mapping_need_revalidate_inode(dir)) {
987 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
990 if (!nfs_verify_change_attribute(dir, dentry->d_time))
996 * Use intent information to check whether or not we're going to do
997 * an O_EXCL create using this path component.
999 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1001 if (NFS_PROTO(dir)->version == 2)
1003 return flags & LOOKUP_EXCL;
1007 * Inode and filehandle revalidation for lookups.
1009 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1010 * or if the intent information indicates that we're about to open this
1011 * particular file and the "nocto" mount flag is not set.
1015 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1017 struct nfs_server *server = NFS_SERVER(inode);
1020 if (IS_AUTOMOUNT(inode))
1022 /* VFS wants an on-the-wire revalidation */
1023 if (flags & LOOKUP_REVAL)
1025 /* This is an open(2) */
1026 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1027 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1030 return (inode->i_nlink == 0) ? -ENOENT : 0;
1032 if (flags & LOOKUP_RCU)
1034 ret = __nfs_revalidate_inode(server, inode);
1041 * We judge how long we want to trust negative
1042 * dentries by looking at the parent inode mtime.
1044 * If parent mtime has changed, we revalidate, else we wait for a
1045 * period corresponding to the parent's attribute cache timeout value.
1047 * If LOOKUP_RCU prevents us from performing a full check, return 1
1048 * suggesting a reval is needed.
1051 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1054 /* Don't revalidate a negative dentry if we're creating a new file */
1055 if (flags & LOOKUP_CREATE)
1057 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1059 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1063 * This is called every time the dcache has a lookup hit,
1064 * and we should check whether we can really trust that
1067 * NOTE! The hit can be a negative hit too, don't assume
1070 * If the parent directory is seen to have changed, we throw out the
1071 * cached dentry and do a new lookup.
1073 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1076 struct inode *inode;
1077 struct dentry *parent;
1078 struct nfs_fh *fhandle = NULL;
1079 struct nfs_fattr *fattr = NULL;
1080 struct nfs4_label *label = NULL;
1083 if (flags & LOOKUP_RCU) {
1084 parent = READ_ONCE(dentry->d_parent);
1085 dir = d_inode_rcu(parent);
1089 parent = dget_parent(dentry);
1090 dir = d_inode(parent);
1092 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1093 inode = d_inode(dentry);
1096 if (nfs_neg_need_reval(dir, dentry, flags)) {
1097 if (flags & LOOKUP_RCU)
1104 if (is_bad_inode(inode)) {
1105 if (flags & LOOKUP_RCU)
1107 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1112 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1113 goto out_set_verifier;
1115 /* Force a full look up iff the parent directory has changed */
1116 if (!nfs_is_exclusive_create(dir, flags) &&
1117 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1118 error = nfs_lookup_verify_inode(inode, flags);
1120 if (flags & LOOKUP_RCU)
1122 if (error == -ESTALE)
1123 goto out_zap_parent;
1126 nfs_advise_use_readdirplus(dir);
1130 if (flags & LOOKUP_RCU)
1133 if (NFS_STALE(inode))
1137 fhandle = nfs_alloc_fhandle();
1138 fattr = nfs_alloc_fattr();
1139 if (fhandle == NULL || fattr == NULL)
1142 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1146 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1147 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1148 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1149 if (error == -ESTALE || error == -ENOENT)
1153 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1155 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1158 nfs_setsecurity(inode, fattr, label);
1160 nfs_free_fattr(fattr);
1161 nfs_free_fhandle(fhandle);
1162 nfs4_label_free(label);
1164 /* set a readdirplus hint that we had a cache miss */
1165 nfs_force_use_readdirplus(dir);
1168 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1170 if (flags & LOOKUP_RCU) {
1171 if (parent != READ_ONCE(dentry->d_parent))
1175 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1179 nfs_zap_caches(dir);
1181 WARN_ON(flags & LOOKUP_RCU);
1182 nfs_free_fattr(fattr);
1183 nfs_free_fhandle(fhandle);
1184 nfs4_label_free(label);
1185 nfs_mark_for_revalidate(dir);
1186 if (inode && S_ISDIR(inode->i_mode)) {
1187 /* Purge readdir caches. */
1188 nfs_zap_caches(inode);
1190 * We can't d_drop the root of a disconnected tree:
1191 * its d_hash is on the s_anon list and d_drop() would hide
1192 * it from shrink_dcache_for_unmount(), leading to busy
1193 * inodes on unmount and further oopses.
1195 if (IS_ROOT(dentry))
1199 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1203 WARN_ON(flags & LOOKUP_RCU);
1204 nfs_free_fattr(fattr);
1205 nfs_free_fhandle(fhandle);
1206 nfs4_label_free(label);
1208 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1209 __func__, dentry, error);
1214 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1215 * when we don't really care about the dentry name. This is called when a
1216 * pathwalk ends on a dentry that was not found via a normal lookup in the
1217 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1219 * In this situation, we just want to verify that the inode itself is OK
1220 * since the dentry might have changed on the server.
1222 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1224 struct inode *inode = d_inode(dentry);
1228 * I believe we can only get a negative dentry here in the case of a
1229 * procfs-style symlink. Just assume it's correct for now, but we may
1230 * eventually need to do something more here.
1233 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1238 if (is_bad_inode(inode)) {
1239 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1244 if (nfs_mapping_need_revalidate_inode(inode))
1245 error = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1246 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1247 __func__, inode->i_ino, error ? "invalid" : "valid");
1252 * This is called from dput() when d_count is going to 0.
1254 static int nfs_dentry_delete(const struct dentry *dentry)
1256 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1257 dentry, dentry->d_flags);
1259 /* Unhash any dentry with a stale inode */
1260 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1263 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1264 /* Unhash it, so that ->d_iput() would be called */
1267 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1268 /* Unhash it, so that ancestors of killed async unlink
1269 * files will be cleaned up during umount */
1276 /* Ensure that we revalidate inode->i_nlink */
1277 static void nfs_drop_nlink(struct inode *inode)
1279 spin_lock(&inode->i_lock);
1280 /* drop the inode if we're reasonably sure this is the last link */
1281 if (inode->i_nlink == 1)
1283 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1284 spin_unlock(&inode->i_lock);
1288 * Called when the dentry loses inode.
1289 * We use it to clean up silly-renamed files.
1291 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1293 if (S_ISDIR(inode->i_mode))
1294 /* drop any readdir cache as it could easily be old */
1295 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1297 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1298 nfs_complete_unlink(dentry, inode);
1299 nfs_drop_nlink(inode);
1304 static void nfs_d_release(struct dentry *dentry)
1306 /* free cached devname value, if it survived that far */
1307 if (unlikely(dentry->d_fsdata)) {
1308 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1311 kfree(dentry->d_fsdata);
1315 const struct dentry_operations nfs_dentry_operations = {
1316 .d_revalidate = nfs_lookup_revalidate,
1317 .d_weak_revalidate = nfs_weak_revalidate,
1318 .d_delete = nfs_dentry_delete,
1319 .d_iput = nfs_dentry_iput,
1320 .d_automount = nfs_d_automount,
1321 .d_release = nfs_d_release,
1323 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1325 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1328 struct inode *inode = NULL;
1329 struct nfs_fh *fhandle = NULL;
1330 struct nfs_fattr *fattr = NULL;
1331 struct nfs4_label *label = NULL;
1334 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1335 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1337 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1338 return ERR_PTR(-ENAMETOOLONG);
1341 * If we're doing an exclusive create, optimize away the lookup
1342 * but don't hash the dentry.
1344 if (nfs_is_exclusive_create(dir, flags))
1347 res = ERR_PTR(-ENOMEM);
1348 fhandle = nfs_alloc_fhandle();
1349 fattr = nfs_alloc_fattr();
1350 if (fhandle == NULL || fattr == NULL)
1353 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1357 trace_nfs_lookup_enter(dir, dentry, flags);
1358 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1359 if (error == -ENOENT)
1362 res = ERR_PTR(error);
1365 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1366 res = ERR_CAST(inode);
1370 /* Notify readdir to use READDIRPLUS */
1371 nfs_force_use_readdirplus(dir);
1374 res = d_splice_alias(inode, dentry);
1380 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1382 trace_nfs_lookup_exit(dir, dentry, flags, error);
1383 nfs4_label_free(label);
1385 nfs_free_fattr(fattr);
1386 nfs_free_fhandle(fhandle);
1389 EXPORT_SYMBOL_GPL(nfs_lookup);
1391 #if IS_ENABLED(CONFIG_NFS_V4)
1392 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1394 const struct dentry_operations nfs4_dentry_operations = {
1395 .d_revalidate = nfs4_lookup_revalidate,
1396 .d_delete = nfs_dentry_delete,
1397 .d_iput = nfs_dentry_iput,
1398 .d_automount = nfs_d_automount,
1399 .d_release = nfs_d_release,
1401 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1403 static fmode_t flags_to_mode(int flags)
1405 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1406 if ((flags & O_ACCMODE) != O_WRONLY)
1408 if ((flags & O_ACCMODE) != O_RDONLY)
1413 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1415 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1418 static int do_open(struct inode *inode, struct file *filp)
1420 nfs_fscache_open_file(inode, filp);
1424 static int nfs_finish_open(struct nfs_open_context *ctx,
1425 struct dentry *dentry,
1426 struct file *file, unsigned open_flags,
1431 err = finish_open(file, dentry, do_open, opened);
1434 if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1435 nfs_file_set_open_context(file, ctx);
1442 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1443 struct file *file, unsigned open_flags,
1444 umode_t mode, int *opened)
1446 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1447 struct nfs_open_context *ctx;
1449 struct iattr attr = { .ia_valid = ATTR_OPEN };
1450 struct inode *inode;
1451 unsigned int lookup_flags = 0;
1452 bool switched = false;
1455 /* Expect a negative dentry */
1456 BUG_ON(d_inode(dentry));
1458 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1459 dir->i_sb->s_id, dir->i_ino, dentry);
1461 err = nfs_check_flags(open_flags);
1465 /* NFS only supports OPEN on regular files */
1466 if ((open_flags & O_DIRECTORY)) {
1467 if (!d_in_lookup(dentry)) {
1469 * Hashed negative dentry with O_DIRECTORY: dentry was
1470 * revalidated and is fine, no need to perform lookup
1475 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1479 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1480 return -ENAMETOOLONG;
1482 if (open_flags & O_CREAT) {
1483 struct nfs_server *server = NFS_SERVER(dir);
1485 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1486 mode &= ~current_umask();
1488 attr.ia_valid |= ATTR_MODE;
1489 attr.ia_mode = mode;
1491 if (open_flags & O_TRUNC) {
1492 attr.ia_valid |= ATTR_SIZE;
1496 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1499 dentry = d_alloc_parallel(dentry->d_parent,
1500 &dentry->d_name, &wq);
1502 return PTR_ERR(dentry);
1503 if (unlikely(!d_in_lookup(dentry)))
1504 return finish_no_open(file, dentry);
1507 ctx = create_nfs_open_context(dentry, open_flags, file);
1512 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1513 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
1514 if (IS_ERR(inode)) {
1515 err = PTR_ERR(inode);
1516 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1517 put_nfs_open_context(ctx);
1521 d_splice_alias(NULL, dentry);
1522 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1528 if (!(open_flags & O_NOFOLLOW))
1538 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1539 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1540 put_nfs_open_context(ctx);
1542 if (unlikely(switched)) {
1543 d_lookup_done(dentry);
1549 res = nfs_lookup(dir, dentry, lookup_flags);
1551 d_lookup_done(dentry);
1558 return PTR_ERR(res);
1559 return finish_no_open(file, res);
1561 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1563 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1565 struct inode *inode;
1568 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1570 if (d_mountpoint(dentry))
1572 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1575 inode = d_inode(dentry);
1577 /* We can't create new files in nfs_open_revalidate(), so we
1578 * optimize away revalidation of negative dentries.
1580 if (inode == NULL) {
1581 struct dentry *parent;
1584 if (flags & LOOKUP_RCU) {
1585 parent = READ_ONCE(dentry->d_parent);
1586 dir = d_inode_rcu(parent);
1590 parent = dget_parent(dentry);
1591 dir = d_inode(parent);
1593 if (!nfs_neg_need_reval(dir, dentry, flags))
1595 else if (flags & LOOKUP_RCU)
1597 if (!(flags & LOOKUP_RCU))
1599 else if (parent != READ_ONCE(dentry->d_parent))
1604 /* NFS only supports OPEN on regular files */
1605 if (!S_ISREG(inode->i_mode))
1607 /* We cannot do exclusive creation on a positive dentry */
1608 if (flags & LOOKUP_EXCL)
1611 /* Let f_op->open() actually open (and revalidate) the file */
1618 return nfs_lookup_revalidate(dentry, flags);
1621 #endif /* CONFIG_NFSV4 */
1624 * Code common to create, mkdir, and mknod.
1626 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1627 struct nfs_fattr *fattr,
1628 struct nfs4_label *label)
1630 struct dentry *parent = dget_parent(dentry);
1631 struct inode *dir = d_inode(parent);
1632 struct inode *inode;
1633 int error = -EACCES;
1637 /* We may have been initialized further down */
1638 if (d_really_is_positive(dentry))
1640 if (fhandle->size == 0) {
1641 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1645 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1646 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1647 struct nfs_server *server = NFS_SB(dentry->d_sb);
1648 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1652 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1653 error = PTR_ERR(inode);
1656 d_add(dentry, inode);
1661 nfs_mark_for_revalidate(dir);
1665 EXPORT_SYMBOL_GPL(nfs_instantiate);
1668 * Following a failed create operation, we drop the dentry rather
1669 * than retain a negative dentry. This avoids a problem in the event
1670 * that the operation succeeded on the server, but an error in the
1671 * reply path made it appear to have failed.
1673 int nfs_create(struct inode *dir, struct dentry *dentry,
1674 umode_t mode, bool excl)
1677 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1680 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1681 dir->i_sb->s_id, dir->i_ino, dentry);
1683 attr.ia_mode = mode;
1684 attr.ia_valid = ATTR_MODE;
1686 trace_nfs_create_enter(dir, dentry, open_flags);
1687 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1688 trace_nfs_create_exit(dir, dentry, open_flags, error);
1696 EXPORT_SYMBOL_GPL(nfs_create);
1699 * See comments for nfs_proc_create regarding failed operations.
1702 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1707 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1708 dir->i_sb->s_id, dir->i_ino, dentry);
1710 attr.ia_mode = mode;
1711 attr.ia_valid = ATTR_MODE;
1713 trace_nfs_mknod_enter(dir, dentry);
1714 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1715 trace_nfs_mknod_exit(dir, dentry, status);
1723 EXPORT_SYMBOL_GPL(nfs_mknod);
1726 * See comments for nfs_proc_create regarding failed operations.
1728 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1733 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1734 dir->i_sb->s_id, dir->i_ino, dentry);
1736 attr.ia_valid = ATTR_MODE;
1737 attr.ia_mode = mode | S_IFDIR;
1739 trace_nfs_mkdir_enter(dir, dentry);
1740 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1741 trace_nfs_mkdir_exit(dir, dentry, error);
1749 EXPORT_SYMBOL_GPL(nfs_mkdir);
1751 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1753 if (simple_positive(dentry))
1757 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1761 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1762 dir->i_sb->s_id, dir->i_ino, dentry);
1764 trace_nfs_rmdir_enter(dir, dentry);
1765 if (d_really_is_positive(dentry)) {
1766 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1767 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1768 /* Ensure the VFS deletes this inode */
1771 clear_nlink(d_inode(dentry));
1774 nfs_dentry_handle_enoent(dentry);
1776 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1778 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1779 trace_nfs_rmdir_exit(dir, dentry, error);
1783 EXPORT_SYMBOL_GPL(nfs_rmdir);
1786 * Remove a file after making sure there are no pending writes,
1787 * and after checking that the file has only one user.
1789 * We invalidate the attribute cache and free the inode prior to the operation
1790 * to avoid possible races if the server reuses the inode.
1792 static int nfs_safe_remove(struct dentry *dentry)
1794 struct inode *dir = d_inode(dentry->d_parent);
1795 struct inode *inode = d_inode(dentry);
1798 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1800 /* If the dentry was sillyrenamed, we simply call d_delete() */
1801 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1806 trace_nfs_remove_enter(dir, dentry);
1807 if (inode != NULL) {
1808 NFS_PROTO(inode)->return_delegation(inode);
1809 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1811 nfs_drop_nlink(inode);
1813 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1814 if (error == -ENOENT)
1815 nfs_dentry_handle_enoent(dentry);
1816 trace_nfs_remove_exit(dir, dentry, error);
1821 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1822 * belongs to an active ".nfs..." file and we return -EBUSY.
1824 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1826 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1829 int need_rehash = 0;
1831 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1832 dir->i_ino, dentry);
1834 trace_nfs_unlink_enter(dir, dentry);
1835 spin_lock(&dentry->d_lock);
1836 if (d_count(dentry) > 1) {
1837 spin_unlock(&dentry->d_lock);
1838 /* Start asynchronous writeout of the inode */
1839 write_inode_now(d_inode(dentry), 0);
1840 error = nfs_sillyrename(dir, dentry);
1843 if (!d_unhashed(dentry)) {
1847 spin_unlock(&dentry->d_lock);
1848 error = nfs_safe_remove(dentry);
1849 if (!error || error == -ENOENT) {
1850 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1851 } else if (need_rehash)
1854 trace_nfs_unlink_exit(dir, dentry, error);
1857 EXPORT_SYMBOL_GPL(nfs_unlink);
1860 * To create a symbolic link, most file systems instantiate a new inode,
1861 * add a page to it containing the path, then write it out to the disk
1862 * using prepare_write/commit_write.
1864 * Unfortunately the NFS client can't create the in-core inode first
1865 * because it needs a file handle to create an in-core inode (see
1866 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1867 * symlink request has completed on the server.
1869 * So instead we allocate a raw page, copy the symname into it, then do
1870 * the SYMLINK request with the page as the buffer. If it succeeds, we
1871 * now have a new file handle and can instantiate an in-core NFS inode
1872 * and move the raw page into its mapping.
1874 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1879 unsigned int pathlen = strlen(symname);
1882 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1883 dir->i_ino, dentry, symname);
1885 if (pathlen > PAGE_SIZE)
1886 return -ENAMETOOLONG;
1888 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1889 attr.ia_valid = ATTR_MODE;
1891 page = alloc_page(GFP_USER);
1895 kaddr = page_address(page);
1896 memcpy(kaddr, symname, pathlen);
1897 if (pathlen < PAGE_SIZE)
1898 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1900 trace_nfs_symlink_enter(dir, dentry);
1901 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1902 trace_nfs_symlink_exit(dir, dentry, error);
1904 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1905 dir->i_sb->s_id, dir->i_ino,
1906 dentry, symname, error);
1913 * No big deal if we can't add this page to the page cache here.
1914 * READLINK will get the missing page from the server if needed.
1916 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
1918 SetPageUptodate(page);
1921 * add_to_page_cache_lru() grabs an extra page refcount.
1922 * Drop it here to avoid leaking this page later.
1930 EXPORT_SYMBOL_GPL(nfs_symlink);
1933 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1935 struct inode *inode = d_inode(old_dentry);
1938 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1939 old_dentry, dentry);
1941 trace_nfs_link_enter(inode, dir, dentry);
1942 NFS_PROTO(inode)->return_delegation(inode);
1945 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1948 d_add(dentry, inode);
1950 trace_nfs_link_exit(inode, dir, dentry, error);
1953 EXPORT_SYMBOL_GPL(nfs_link);
1957 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1958 * different file handle for the same inode after a rename (e.g. when
1959 * moving to a different directory). A fail-safe method to do so would
1960 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1961 * rename the old file using the sillyrename stuff. This way, the original
1962 * file in old_dir will go away when the last process iput()s the inode.
1966 * It actually works quite well. One needs to have the possibility for
1967 * at least one ".nfs..." file in each directory the file ever gets
1968 * moved or linked to which happens automagically with the new
1969 * implementation that only depends on the dcache stuff instead of
1970 * using the inode layer
1972 * Unfortunately, things are a little more complicated than indicated
1973 * above. For a cross-directory move, we want to make sure we can get
1974 * rid of the old inode after the operation. This means there must be
1975 * no pending writes (if it's a file), and the use count must be 1.
1976 * If these conditions are met, we can drop the dentries before doing
1979 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1980 struct inode *new_dir, struct dentry *new_dentry,
1983 struct inode *old_inode = d_inode(old_dentry);
1984 struct inode *new_inode = d_inode(new_dentry);
1985 struct dentry *dentry = NULL, *rehash = NULL;
1986 struct rpc_task *task;
1992 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
1993 old_dentry, new_dentry,
1994 d_count(new_dentry));
1996 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
1998 * For non-directories, check whether the target is busy and if so,
1999 * make a copy of the dentry and then do a silly-rename. If the
2000 * silly-rename succeeds, the copied dentry is hashed and becomes
2003 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2005 * To prevent any new references to the target during the
2006 * rename, we unhash the dentry in advance.
2008 if (!d_unhashed(new_dentry)) {
2010 rehash = new_dentry;
2013 if (d_count(new_dentry) > 2) {
2016 /* copy the target dentry's name */
2017 dentry = d_alloc(new_dentry->d_parent,
2018 &new_dentry->d_name);
2022 /* silly-rename the existing target ... */
2023 err = nfs_sillyrename(new_dir, new_dentry);
2027 new_dentry = dentry;
2033 NFS_PROTO(old_inode)->return_delegation(old_inode);
2034 if (new_inode != NULL)
2035 NFS_PROTO(new_inode)->return_delegation(new_inode);
2037 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2039 error = PTR_ERR(task);
2043 error = rpc_wait_for_completion_task(task);
2045 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2046 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2049 error = task->tk_status;
2051 nfs_mark_for_revalidate(old_inode);
2055 trace_nfs_rename_exit(old_dir, old_dentry,
2056 new_dir, new_dentry, error);
2058 if (new_inode != NULL)
2059 nfs_drop_nlink(new_inode);
2061 * The d_move() should be here instead of in an async RPC completion
2062 * handler because we need the proper locks to move the dentry. If
2063 * we're interrupted by a signal, the async RPC completion handler
2064 * should mark the directories for revalidation.
2066 d_move(old_dentry, new_dentry);
2067 nfs_set_verifier(new_dentry,
2068 nfs_save_change_attribute(new_dir));
2069 } else if (error == -ENOENT)
2070 nfs_dentry_handle_enoent(old_dentry);
2072 /* new dentry created? */
2077 EXPORT_SYMBOL_GPL(nfs_rename);
2079 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2080 static LIST_HEAD(nfs_access_lru_list);
2081 static atomic_long_t nfs_access_nr_entries;
2083 static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2084 module_param(nfs_access_max_cachesize, ulong, 0644);
2085 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2087 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2089 put_rpccred(entry->cred);
2090 kfree_rcu(entry, rcu_head);
2091 smp_mb__before_atomic();
2092 atomic_long_dec(&nfs_access_nr_entries);
2093 smp_mb__after_atomic();
2096 static void nfs_access_free_list(struct list_head *head)
2098 struct nfs_access_entry *cache;
2100 while (!list_empty(head)) {
2101 cache = list_entry(head->next, struct nfs_access_entry, lru);
2102 list_del(&cache->lru);
2103 nfs_access_free_entry(cache);
2107 static unsigned long
2108 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2111 struct nfs_inode *nfsi, *next;
2112 struct nfs_access_entry *cache;
2115 spin_lock(&nfs_access_lru_lock);
2116 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2117 struct inode *inode;
2119 if (nr_to_scan-- == 0)
2121 inode = &nfsi->vfs_inode;
2122 spin_lock(&inode->i_lock);
2123 if (list_empty(&nfsi->access_cache_entry_lru))
2124 goto remove_lru_entry;
2125 cache = list_entry(nfsi->access_cache_entry_lru.next,
2126 struct nfs_access_entry, lru);
2127 list_move(&cache->lru, &head);
2128 rb_erase(&cache->rb_node, &nfsi->access_cache);
2130 if (!list_empty(&nfsi->access_cache_entry_lru))
2131 list_move_tail(&nfsi->access_cache_inode_lru,
2132 &nfs_access_lru_list);
2135 list_del_init(&nfsi->access_cache_inode_lru);
2136 smp_mb__before_atomic();
2137 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2138 smp_mb__after_atomic();
2140 spin_unlock(&inode->i_lock);
2142 spin_unlock(&nfs_access_lru_lock);
2143 nfs_access_free_list(&head);
2148 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2150 int nr_to_scan = sc->nr_to_scan;
2151 gfp_t gfp_mask = sc->gfp_mask;
2153 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2155 return nfs_do_access_cache_scan(nr_to_scan);
2160 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2162 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2166 nfs_access_cache_enforce_limit(void)
2168 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2170 unsigned int nr_to_scan;
2172 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2175 diff = nr_entries - nfs_access_max_cachesize;
2176 if (diff < nr_to_scan)
2178 nfs_do_access_cache_scan(nr_to_scan);
2181 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2183 struct rb_root *root_node = &nfsi->access_cache;
2185 struct nfs_access_entry *entry;
2187 /* Unhook entries from the cache */
2188 while ((n = rb_first(root_node)) != NULL) {
2189 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2190 rb_erase(n, root_node);
2191 list_move(&entry->lru, head);
2193 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2196 void nfs_access_zap_cache(struct inode *inode)
2200 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2202 /* Remove from global LRU init */
2203 spin_lock(&nfs_access_lru_lock);
2204 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2205 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2207 spin_lock(&inode->i_lock);
2208 __nfs_access_zap_cache(NFS_I(inode), &head);
2209 spin_unlock(&inode->i_lock);
2210 spin_unlock(&nfs_access_lru_lock);
2211 nfs_access_free_list(&head);
2213 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2215 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2217 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2218 struct nfs_access_entry *entry;
2221 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2223 if (cred < entry->cred)
2225 else if (cred > entry->cred)
2233 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res, bool may_block)
2235 struct nfs_inode *nfsi = NFS_I(inode);
2236 struct nfs_access_entry *cache;
2240 spin_lock(&inode->i_lock);
2242 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2244 cache = nfs_access_search_rbtree(inode, cred);
2248 /* Found an entry, is our attribute cache valid? */
2249 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2256 spin_unlock(&inode->i_lock);
2257 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2260 spin_lock(&inode->i_lock);
2263 res->cred = cache->cred;
2264 res->mask = cache->mask;
2265 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2268 spin_unlock(&inode->i_lock);
2271 spin_unlock(&inode->i_lock);
2272 nfs_access_zap_cache(inode);
2276 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2278 /* Only check the most recently returned cache entry,
2279 * but do it without locking.
2281 struct nfs_inode *nfsi = NFS_I(inode);
2282 struct nfs_access_entry *cache;
2284 struct list_head *lh;
2287 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2289 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2290 cache = list_entry(lh, struct nfs_access_entry, lru);
2291 if (lh == &nfsi->access_cache_entry_lru ||
2292 cred != cache->cred)
2296 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2298 res->cred = cache->cred;
2299 res->mask = cache->mask;
2306 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2308 struct nfs_inode *nfsi = NFS_I(inode);
2309 struct rb_root *root_node = &nfsi->access_cache;
2310 struct rb_node **p = &root_node->rb_node;
2311 struct rb_node *parent = NULL;
2312 struct nfs_access_entry *entry;
2314 spin_lock(&inode->i_lock);
2315 while (*p != NULL) {
2317 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2319 if (set->cred < entry->cred)
2320 p = &parent->rb_left;
2321 else if (set->cred > entry->cred)
2322 p = &parent->rb_right;
2326 rb_link_node(&set->rb_node, parent, p);
2327 rb_insert_color(&set->rb_node, root_node);
2328 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2329 spin_unlock(&inode->i_lock);
2332 rb_replace_node(parent, &set->rb_node, root_node);
2333 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2334 list_del(&entry->lru);
2335 spin_unlock(&inode->i_lock);
2336 nfs_access_free_entry(entry);
2339 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2341 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2344 RB_CLEAR_NODE(&cache->rb_node);
2345 cache->cred = get_rpccred(set->cred);
2346 cache->mask = set->mask;
2348 /* The above field assignments must be visible
2349 * before this item appears on the lru. We cannot easily
2350 * use rcu_assign_pointer, so just force the memory barrier.
2353 nfs_access_add_rbtree(inode, cache);
2355 /* Update accounting */
2356 smp_mb__before_atomic();
2357 atomic_long_inc(&nfs_access_nr_entries);
2358 smp_mb__after_atomic();
2360 /* Add inode to global LRU list */
2361 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2362 spin_lock(&nfs_access_lru_lock);
2363 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2364 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2365 &nfs_access_lru_list);
2366 spin_unlock(&nfs_access_lru_lock);
2368 nfs_access_cache_enforce_limit();
2370 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2372 #define NFS_MAY_READ (NFS4_ACCESS_READ)
2373 #define NFS_MAY_WRITE (NFS4_ACCESS_MODIFY | \
2374 NFS4_ACCESS_EXTEND | \
2376 #define NFS_FILE_MAY_WRITE (NFS4_ACCESS_MODIFY | \
2378 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2379 #define NFS_MAY_LOOKUP (NFS4_ACCESS_LOOKUP)
2380 #define NFS_MAY_EXECUTE (NFS4_ACCESS_EXECUTE)
2382 nfs_access_calc_mask(u32 access_result, umode_t umode)
2386 if (access_result & NFS_MAY_READ)
2388 if (S_ISDIR(umode)) {
2389 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2391 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2393 } else if (S_ISREG(umode)) {
2394 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2396 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2398 } else if (access_result & NFS_MAY_WRITE)
2403 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2405 entry->mask = access_result;
2407 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2409 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2411 struct nfs_access_entry cache;
2412 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2416 trace_nfs_access_enter(inode);
2418 status = nfs_access_get_cached_rcu(inode, cred, &cache);
2420 status = nfs_access_get_cached(inode, cred, &cache, may_block);
2428 /* Be clever: ask server to check for all possible rights */
2429 cache.mask = NFS_MAY_LOOKUP | NFS_MAY_EXECUTE
2430 | NFS_MAY_WRITE | NFS_MAY_READ;
2432 status = NFS_PROTO(inode)->access(inode, &cache);
2434 if (status == -ESTALE) {
2435 nfs_zap_caches(inode);
2436 if (!S_ISDIR(inode->i_mode))
2437 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2441 nfs_access_add_cache(inode, &cache);
2443 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2444 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2447 trace_nfs_access_exit(inode, status);
2451 static int nfs_open_permission_mask(int openflags)
2455 if (openflags & __FMODE_EXEC) {
2456 /* ONLY check exec rights */
2459 if ((openflags & O_ACCMODE) != O_WRONLY)
2461 if ((openflags & O_ACCMODE) != O_RDONLY)
2468 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2470 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2472 EXPORT_SYMBOL_GPL(nfs_may_open);
2474 static int nfs_execute_ok(struct inode *inode, int mask)
2476 struct nfs_server *server = NFS_SERVER(inode);
2479 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) {
2480 if (mask & MAY_NOT_BLOCK)
2482 ret = __nfs_revalidate_inode(server, inode);
2484 if (ret == 0 && !execute_ok(inode))
2489 int nfs_permission(struct inode *inode, int mask)
2491 struct rpc_cred *cred;
2494 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2496 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2498 /* Is this sys_access() ? */
2499 if (mask & (MAY_ACCESS | MAY_CHDIR))
2502 switch (inode->i_mode & S_IFMT) {
2506 if ((mask & MAY_OPEN) &&
2507 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2512 * Optimize away all write operations, since the server
2513 * will check permissions when we perform the op.
2515 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2520 if (!NFS_PROTO(inode)->access)
2523 /* Always try fast lookups first */
2525 cred = rpc_lookup_cred_nonblock();
2527 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2529 res = PTR_ERR(cred);
2531 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2532 /* Fast lookup failed, try the slow way */
2533 cred = rpc_lookup_cred();
2534 if (!IS_ERR(cred)) {
2535 res = nfs_do_access(inode, cred, mask);
2538 res = PTR_ERR(cred);
2541 if (!res && (mask & MAY_EXEC))
2542 res = nfs_execute_ok(inode, mask);
2544 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2545 inode->i_sb->s_id, inode->i_ino, mask, res);
2548 if (mask & MAY_NOT_BLOCK)
2551 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2553 res = generic_permission(inode, mask);
2556 EXPORT_SYMBOL_GPL(nfs_permission);
2560 * version-control: t
2561 * kept-new-versions: 5