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_shared = 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 {
152 struct nfs_cache_array_entry array[0];
155 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
159 struct dir_context *ctx;
160 unsigned long page_index;
163 loff_t current_index;
164 decode_dirent_t decode;
166 unsigned long timestamp;
167 unsigned long gencount;
168 unsigned int cache_entry_index;
171 } nfs_readdir_descriptor_t;
174 * The caller is responsible for calling nfs_readdir_release_array(page)
177 struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
181 return ERR_PTR(-EIO);
184 return ERR_PTR(-ENOMEM);
189 void nfs_readdir_release_array(struct page *page)
195 * we are freeing strings created by nfs_add_to_readdir_array()
198 void nfs_readdir_clear_array(struct page *page)
200 struct nfs_cache_array *array;
203 array = kmap_atomic(page);
204 if (atomic_dec_and_test(&array->refcount))
205 for (i = 0; i < array->size; i++)
206 kfree(array->array[i].string.name);
207 kunmap_atomic(array);
210 static bool grab_page(struct page *page)
212 struct nfs_cache_array *array = kmap_atomic(page);
213 bool res = atomic_inc_not_zero(&array->refcount);
214 kunmap_atomic(array);
219 * the caller is responsible for freeing qstr.name
220 * when called by nfs_readdir_add_to_array, the strings will be freed in
221 * nfs_clear_readdir_array()
224 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
227 string->name = kmemdup(name, len, GFP_KERNEL);
228 if (string->name == NULL)
231 * Avoid a kmemleak false positive. The pointer to the name is stored
232 * in a page cache page which kmemleak does not scan.
234 kmemleak_not_leak(string->name);
235 string->hash = full_name_hash(NULL, name, len);
240 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
242 struct nfs_cache_array *array = nfs_readdir_get_array(page);
243 struct nfs_cache_array_entry *cache_entry;
247 return PTR_ERR(array);
249 cache_entry = &array->array[array->size];
251 /* Check that this entry lies within the page bounds */
253 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
256 cache_entry->cookie = entry->prev_cookie;
257 cache_entry->ino = entry->ino;
258 cache_entry->d_type = entry->d_type;
259 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
262 array->last_cookie = entry->cookie;
265 array->eof_index = array->size;
267 nfs_readdir_release_array(page);
272 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
274 loff_t diff = desc->ctx->pos - desc->current_index;
279 if (diff >= array->size) {
280 if (array->eof_index >= 0)
285 index = (unsigned int)diff;
286 *desc->dir_cookie = array->array[index].cookie;
287 desc->cache_entry_index = index;
295 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
297 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
300 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
304 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
308 int status = -EAGAIN;
310 for (i = 0; i < array->size; i++) {
311 if (array->array[i].cookie == *desc->dir_cookie) {
312 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
313 struct nfs_open_dir_context *ctx = desc->file->private_data;
315 new_pos = desc->current_index + i;
316 if (ctx->attr_gencount != nfsi->attr_gencount ||
317 !nfs_readdir_inode_mapping_valid(nfsi)) {
319 ctx->attr_gencount = nfsi->attr_gencount;
320 } else if (new_pos < desc->ctx->pos) {
322 && ctx->dup_cookie == *desc->dir_cookie) {
323 if (printk_ratelimit()) {
324 pr_notice("NFS: directory %pD2 contains a readdir loop."
325 "Please contact your server vendor. "
326 "The file: %.*s has duplicate cookie %llu\n",
327 desc->file, array->array[i].string.len,
328 array->array[i].string.name, *desc->dir_cookie);
333 ctx->dup_cookie = *desc->dir_cookie;
336 desc->ctx->pos = new_pos;
337 desc->cache_entry_index = i;
341 if (array->eof_index >= 0) {
342 status = -EBADCOOKIE;
343 if (*desc->dir_cookie == array->last_cookie)
351 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
353 struct nfs_cache_array *array;
356 array = nfs_readdir_get_array(desc->page);
358 status = PTR_ERR(array);
362 if (*desc->dir_cookie == 0)
363 status = nfs_readdir_search_for_pos(array, desc);
365 status = nfs_readdir_search_for_cookie(array, desc);
367 if (status == -EAGAIN) {
368 desc->last_cookie = array->last_cookie;
369 desc->current_index += array->size;
372 nfs_readdir_release_array(desc->page);
377 /* Fill a page with xdr information before transferring to the cache page */
379 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
380 struct nfs_entry *entry, struct file *file, struct inode *inode)
382 struct nfs_open_dir_context *ctx = file->private_data;
383 struct rpc_cred *cred = ctx->cred;
384 unsigned long timestamp, gencount;
389 gencount = nfs_inc_attr_generation_counter();
390 error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages,
391 NFS_SERVER(inode)->dtsize, desc->plus);
393 /* We requested READDIRPLUS, but the server doesn't grok it */
394 if (error == -ENOTSUPP && desc->plus) {
395 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
396 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
402 desc->timestamp = timestamp;
403 desc->gencount = gencount;
408 static int xdr_decode(nfs_readdir_descriptor_t *desc,
409 struct nfs_entry *entry, struct xdr_stream *xdr)
413 error = desc->decode(xdr, entry, desc->plus);
416 entry->fattr->time_start = desc->timestamp;
417 entry->fattr->gencount = desc->gencount;
421 /* Match file and dirent using either filehandle or fileid
422 * Note: caller is responsible for checking the fsid
425 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
428 struct nfs_inode *nfsi;
430 if (d_really_is_negative(dentry))
433 inode = d_inode(dentry);
434 if (is_bad_inode(inode) || NFS_STALE(inode))
438 if (entry->fattr->fileid != nfsi->fileid)
440 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
446 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
448 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
450 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
458 * This function is called by the lookup and getattr code to request the
459 * use of readdirplus to accelerate any future lookups in the same
462 void nfs_advise_use_readdirplus(struct inode *dir)
464 struct nfs_inode *nfsi = NFS_I(dir);
466 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
467 !list_empty(&nfsi->open_files))
468 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
472 * This function is mainly for use by nfs_getattr().
474 * If this is an 'ls -l', we want to force use of readdirplus.
475 * Do this by checking if there is an active file descriptor
476 * and calling nfs_advise_use_readdirplus, then forcing a
479 void nfs_force_use_readdirplus(struct inode *dir)
481 struct nfs_inode *nfsi = NFS_I(dir);
483 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
484 !list_empty(&nfsi->open_files)) {
485 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
486 invalidate_mapping_pages(dir->i_mapping, 0, -1);
491 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
493 struct qstr filename = QSTR_INIT(entry->name, entry->len);
494 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
495 struct dentry *dentry;
496 struct dentry *alias;
497 struct inode *dir = d_inode(parent);
501 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
503 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
505 if (filename.len == 0)
507 /* Validate that the name doesn't contain any illegal '\0' */
508 if (strnlen(filename.name, filename.len) != filename.len)
511 if (strnchr(filename.name, filename.len, '/'))
513 if (filename.name[0] == '.') {
514 if (filename.len == 1)
516 if (filename.len == 2 && filename.name[1] == '.')
519 filename.hash = full_name_hash(parent, filename.name, filename.len);
521 dentry = d_lookup(parent, &filename);
524 dentry = d_alloc_parallel(parent, &filename, &wq);
528 if (!d_in_lookup(dentry)) {
529 /* Is there a mountpoint here? If so, just exit */
530 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
531 &entry->fattr->fsid))
533 if (nfs_same_file(dentry, entry)) {
534 if (!entry->fh->size)
536 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
537 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
539 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
542 d_invalidate(dentry);
548 if (!entry->fh->size) {
549 d_lookup_done(dentry);
553 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
554 alias = d_splice_alias(inode, dentry);
555 d_lookup_done(dentry);
562 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
567 /* Perform conversion from xdr to cache array */
569 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
570 struct page **xdr_pages, struct page *page, unsigned int buflen)
572 struct xdr_stream stream;
574 struct page *scratch;
575 struct nfs_cache_array *array;
576 unsigned int count = 0;
579 scratch = alloc_page(GFP_KERNEL);
586 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
587 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
590 status = xdr_decode(desc, entry, &stream);
592 if (status == -EAGAIN)
600 nfs_prime_dcache(file_dentry(desc->file), entry);
602 status = nfs_readdir_add_to_array(entry, page);
605 } while (!entry->eof);
608 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
609 array = nfs_readdir_get_array(page);
610 if (!IS_ERR(array)) {
611 array->eof_index = array->size;
613 nfs_readdir_release_array(page);
615 status = PTR_ERR(array);
623 void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
626 for (i = 0; i < npages; i++)
631 * nfs_readdir_large_page will allocate pages that must be freed with a call
632 * to nfs_readdir_free_pagearray
635 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
639 for (i = 0; i < npages; i++) {
640 struct page *page = alloc_page(GFP_KERNEL);
648 nfs_readdir_free_pages(pages, i);
653 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
655 struct page *pages[NFS_MAX_READDIR_PAGES];
656 struct nfs_entry entry;
657 struct file *file = desc->file;
658 struct nfs_cache_array *array;
659 int status = -ENOMEM;
660 unsigned int array_size = ARRAY_SIZE(pages);
662 entry.prev_cookie = 0;
663 entry.cookie = desc->last_cookie;
665 entry.fh = nfs_alloc_fhandle();
666 entry.fattr = nfs_alloc_fattr();
667 entry.server = NFS_SERVER(inode);
668 if (entry.fh == NULL || entry.fattr == NULL)
671 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
672 if (IS_ERR(entry.label)) {
673 status = PTR_ERR(entry.label);
677 array = nfs_readdir_get_array(page);
679 status = PTR_ERR(array);
682 memset(array, 0, sizeof(struct nfs_cache_array));
683 atomic_set(&array->refcount, 1);
684 array->eof_index = -1;
686 status = nfs_readdir_alloc_pages(pages, array_size);
688 goto out_release_array;
691 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
696 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
698 if (status == -ENOSPC)
702 } while (array->eof_index < 0);
704 nfs_readdir_free_pages(pages, array_size);
706 nfs_readdir_release_array(page);
708 nfs4_label_free(entry.label);
710 nfs_free_fattr(entry.fattr);
711 nfs_free_fhandle(entry.fh);
716 * Now we cache directories properly, by converting xdr information
717 * to an array that can be used for lookups later. This results in
718 * fewer cache pages, since we can store more information on each page.
719 * We only need to convert from xdr once so future lookups are much simpler
722 int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
724 struct inode *inode = file_inode(desc->file);
727 ret = nfs_readdir_xdr_to_array(desc, page, inode);
730 SetPageUptodate(page);
732 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
733 /* Should never happen */
734 nfs_zap_mapping(inode, inode->i_mapping);
744 void cache_page_release(nfs_readdir_descriptor_t *desc)
746 nfs_readdir_clear_array(desc->page);
747 put_page(desc->page);
752 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
757 page = read_cache_page(desc->file->f_mapping,
758 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
759 if (IS_ERR(page) || grab_page(page))
767 * Returns 0 if desc->dir_cookie was found on page desc->page_index
770 int find_cache_page(nfs_readdir_descriptor_t *desc)
774 desc->page = get_cache_page(desc);
775 if (IS_ERR(desc->page))
776 return PTR_ERR(desc->page);
778 res = nfs_readdir_search_array(desc);
780 cache_page_release(desc);
784 /* Search for desc->dir_cookie from the beginning of the page cache */
786 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
790 if (desc->page_index == 0) {
791 desc->current_index = 0;
792 desc->last_cookie = 0;
795 res = find_cache_page(desc);
796 } while (res == -EAGAIN);
801 * Once we've found the start of the dirent within a page: fill 'er up...
804 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
806 struct file *file = desc->file;
809 struct nfs_cache_array *array = NULL;
810 struct nfs_open_dir_context *ctx = file->private_data;
812 array = nfs_readdir_get_array(desc->page);
814 res = PTR_ERR(array);
818 for (i = desc->cache_entry_index; i < array->size; i++) {
819 struct nfs_cache_array_entry *ent;
821 ent = &array->array[i];
822 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
823 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
828 if (i < (array->size-1))
829 *desc->dir_cookie = array->array[i+1].cookie;
831 *desc->dir_cookie = array->last_cookie;
835 if (array->eof_index >= 0)
838 nfs_readdir_release_array(desc->page);
840 cache_page_release(desc);
841 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
842 (unsigned long long)*desc->dir_cookie, res);
847 * If we cannot find a cookie in our cache, we suspect that this is
848 * because it points to a deleted file, so we ask the server to return
849 * whatever it thinks is the next entry. We then feed this to filldir.
850 * If all goes well, we should then be able to find our way round the
851 * cache on the next call to readdir_search_pagecache();
853 * NOTE: we cannot add the anonymous page to the pagecache because
854 * the data it contains might not be page aligned. Besides,
855 * we should already have a complete representation of the
856 * directory in the page cache by the time we get here.
859 int uncached_readdir(nfs_readdir_descriptor_t *desc)
861 struct page *page = NULL;
863 struct inode *inode = file_inode(desc->file);
864 struct nfs_open_dir_context *ctx = desc->file->private_data;
866 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
867 (unsigned long long)*desc->dir_cookie);
869 page = alloc_page(GFP_HIGHUSER);
875 desc->page_index = 0;
876 desc->last_cookie = *desc->dir_cookie;
880 status = nfs_readdir_xdr_to_array(desc, page, inode);
884 status = nfs_do_filldir(desc);
887 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
891 cache_page_release(desc);
895 /* The file offset position represents the dirent entry number. A
896 last cookie cache takes care of the common case of reading the
899 static int nfs_readdir(struct file *file, struct dir_context *ctx)
901 struct dentry *dentry = file_dentry(file);
902 struct inode *inode = d_inode(dentry);
903 nfs_readdir_descriptor_t my_desc,
905 struct nfs_open_dir_context *dir_ctx = file->private_data;
908 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
909 file, (long long)ctx->pos);
910 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
913 * ctx->pos points to the dirent entry number.
914 * *desc->dir_cookie has the cookie for the next entry. We have
915 * to either find the entry with the appropriate number or
916 * revalidate the cookie.
918 memset(desc, 0, sizeof(*desc));
922 desc->dir_cookie = &dir_ctx->dir_cookie;
923 desc->decode = NFS_PROTO(inode)->decode_dirent;
924 desc->plus = nfs_use_readdirplus(inode, ctx) ? 1 : 0;
926 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
927 res = nfs_revalidate_mapping(inode, file->f_mapping);
932 res = readdir_search_pagecache(desc);
934 if (res == -EBADCOOKIE) {
936 /* This means either end of directory */
937 if (*desc->dir_cookie && desc->eof == 0) {
938 /* Or that the server has 'lost' a cookie */
939 res = uncached_readdir(desc);
945 if (res == -ETOOSMALL && desc->plus) {
946 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
947 nfs_zap_caches(inode);
948 desc->page_index = 0;
956 res = nfs_do_filldir(desc);
959 } while (!desc->eof);
963 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
967 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
969 struct nfs_open_dir_context *dir_ctx = filp->private_data;
971 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
972 filp, offset, whence);
976 offset += filp->f_pos;
983 if (offset != filp->f_pos) {
984 filp->f_pos = offset;
985 dir_ctx->dir_cookie = 0;
992 * All directory operations under NFS are synchronous, so fsync()
993 * is a dummy operation.
995 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
998 struct inode *inode = file_inode(filp);
1000 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1003 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
1004 inode_unlock(inode);
1009 * nfs_force_lookup_revalidate - Mark the directory as having changed
1010 * @dir - pointer to directory inode
1012 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1013 * full lookup on all child dentries of 'dir' whenever a change occurs
1014 * on the server that might have invalidated our dcache.
1016 * The caller should be holding dir->i_lock
1018 void nfs_force_lookup_revalidate(struct inode *dir)
1020 NFS_I(dir)->cache_change_attribute++;
1022 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1025 * A check for whether or not the parent directory has changed.
1026 * In the case it has, we assume that the dentries are untrustworthy
1027 * and may need to be looked up again.
1028 * If rcu_walk prevents us from performing a full check, return 0.
1030 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1033 if (IS_ROOT(dentry))
1035 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1037 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1039 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1040 if (nfs_mapping_need_revalidate_inode(dir)) {
1043 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1046 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1052 * Use intent information to check whether or not we're going to do
1053 * an O_EXCL create using this path component.
1055 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1057 if (NFS_PROTO(dir)->version == 2)
1059 return flags & LOOKUP_EXCL;
1063 * Inode and filehandle revalidation for lookups.
1065 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1066 * or if the intent information indicates that we're about to open this
1067 * particular file and the "nocto" mount flag is not set.
1071 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1073 struct nfs_server *server = NFS_SERVER(inode);
1076 if (IS_AUTOMOUNT(inode))
1078 /* VFS wants an on-the-wire revalidation */
1079 if (flags & LOOKUP_REVAL)
1081 /* This is an open(2) */
1082 if ((flags & LOOKUP_OPEN) && !(server->flags & NFS_MOUNT_NOCTO) &&
1083 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)))
1086 return (inode->i_nlink == 0) ? -ENOENT : 0;
1088 if (flags & LOOKUP_RCU)
1090 ret = __nfs_revalidate_inode(server, inode);
1097 * We judge how long we want to trust negative
1098 * dentries by looking at the parent inode mtime.
1100 * If parent mtime has changed, we revalidate, else we wait for a
1101 * period corresponding to the parent's attribute cache timeout value.
1103 * If LOOKUP_RCU prevents us from performing a full check, return 1
1104 * suggesting a reval is needed.
1107 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1110 /* Don't revalidate a negative dentry if we're creating a new file */
1111 if (flags & LOOKUP_CREATE)
1113 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1115 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1119 * This is called every time the dcache has a lookup hit,
1120 * and we should check whether we can really trust that
1123 * NOTE! The hit can be a negative hit too, don't assume
1126 * If the parent directory is seen to have changed, we throw out the
1127 * cached dentry and do a new lookup.
1129 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1132 struct inode *inode;
1133 struct dentry *parent;
1134 struct nfs_fh *fhandle = NULL;
1135 struct nfs_fattr *fattr = NULL;
1136 struct nfs4_label *label = NULL;
1139 if (flags & LOOKUP_RCU) {
1140 parent = ACCESS_ONCE(dentry->d_parent);
1141 dir = d_inode_rcu(parent);
1145 parent = dget_parent(dentry);
1146 dir = d_inode(parent);
1148 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1149 inode = d_inode(dentry);
1152 if (nfs_neg_need_reval(dir, dentry, flags)) {
1153 if (flags & LOOKUP_RCU)
1160 if (is_bad_inode(inode)) {
1161 if (flags & LOOKUP_RCU)
1163 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1168 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1169 goto out_set_verifier;
1171 /* Force a full look up iff the parent directory has changed */
1172 if (!nfs_is_exclusive_create(dir, flags) &&
1173 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1175 if (nfs_lookup_verify_inode(inode, flags)) {
1176 if (flags & LOOKUP_RCU)
1178 goto out_zap_parent;
1180 nfs_advise_use_readdirplus(dir);
1184 if (flags & LOOKUP_RCU)
1187 if (NFS_STALE(inode))
1191 fhandle = nfs_alloc_fhandle();
1192 fattr = nfs_alloc_fattr();
1193 if (fhandle == NULL || fattr == NULL)
1196 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
1200 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1201 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1202 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1205 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1207 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
1210 nfs_setsecurity(inode, fattr, label);
1212 nfs_free_fattr(fattr);
1213 nfs_free_fhandle(fhandle);
1214 nfs4_label_free(label);
1216 /* set a readdirplus hint that we had a cache miss */
1217 nfs_force_use_readdirplus(dir);
1220 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1222 if (flags & LOOKUP_RCU) {
1223 if (parent != ACCESS_ONCE(dentry->d_parent))
1227 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1231 nfs_zap_caches(dir);
1233 WARN_ON(flags & LOOKUP_RCU);
1234 nfs_free_fattr(fattr);
1235 nfs_free_fhandle(fhandle);
1236 nfs4_label_free(label);
1237 nfs_mark_for_revalidate(dir);
1238 if (inode && S_ISDIR(inode->i_mode)) {
1239 /* Purge readdir caches. */
1240 nfs_zap_caches(inode);
1242 * We can't d_drop the root of a disconnected tree:
1243 * its d_hash is on the s_anon list and d_drop() would hide
1244 * it from shrink_dcache_for_unmount(), leading to busy
1245 * inodes on unmount and further oopses.
1247 if (IS_ROOT(dentry))
1251 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1255 WARN_ON(flags & LOOKUP_RCU);
1256 nfs_free_fattr(fattr);
1257 nfs_free_fhandle(fhandle);
1258 nfs4_label_free(label);
1260 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1261 __func__, dentry, error);
1266 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1267 * when we don't really care about the dentry name. This is called when a
1268 * pathwalk ends on a dentry that was not found via a normal lookup in the
1269 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1271 * In this situation, we just want to verify that the inode itself is OK
1272 * since the dentry might have changed on the server.
1274 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1276 struct inode *inode = d_inode(dentry);
1280 * I believe we can only get a negative dentry here in the case of a
1281 * procfs-style symlink. Just assume it's correct for now, but we may
1282 * eventually need to do something more here.
1285 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1290 if (is_bad_inode(inode)) {
1291 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1296 if (nfs_mapping_need_revalidate_inode(inode))
1297 error = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
1298 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1299 __func__, inode->i_ino, error ? "invalid" : "valid");
1304 * This is called from dput() when d_count is going to 0.
1306 static int nfs_dentry_delete(const struct dentry *dentry)
1308 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1309 dentry, dentry->d_flags);
1311 /* Unhash any dentry with a stale inode */
1312 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1315 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1316 /* Unhash it, so that ->d_iput() would be called */
1319 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1320 /* Unhash it, so that ancestors of killed async unlink
1321 * files will be cleaned up during umount */
1328 /* Ensure that we revalidate inode->i_nlink */
1329 static void nfs_drop_nlink(struct inode *inode)
1331 spin_lock(&inode->i_lock);
1332 /* drop the inode if we're reasonably sure this is the last link */
1333 if (inode->i_nlink == 1)
1335 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_ATTR;
1336 spin_unlock(&inode->i_lock);
1340 * Called when the dentry loses inode.
1341 * We use it to clean up silly-renamed files.
1343 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1345 if (S_ISDIR(inode->i_mode))
1346 /* drop any readdir cache as it could easily be old */
1347 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1349 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1350 nfs_complete_unlink(dentry, inode);
1351 nfs_drop_nlink(inode);
1356 static void nfs_d_release(struct dentry *dentry)
1358 /* free cached devname value, if it survived that far */
1359 if (unlikely(dentry->d_fsdata)) {
1360 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1363 kfree(dentry->d_fsdata);
1367 const struct dentry_operations nfs_dentry_operations = {
1368 .d_revalidate = nfs_lookup_revalidate,
1369 .d_weak_revalidate = nfs_weak_revalidate,
1370 .d_delete = nfs_dentry_delete,
1371 .d_iput = nfs_dentry_iput,
1372 .d_automount = nfs_d_automount,
1373 .d_release = nfs_d_release,
1375 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1377 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1380 struct inode *inode = NULL;
1381 struct nfs_fh *fhandle = NULL;
1382 struct nfs_fattr *fattr = NULL;
1383 struct nfs4_label *label = NULL;
1386 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1387 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1389 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1390 return ERR_PTR(-ENAMETOOLONG);
1393 * If we're doing an exclusive create, optimize away the lookup
1394 * but don't hash the dentry.
1396 if (nfs_is_exclusive_create(dir, flags))
1399 res = ERR_PTR(-ENOMEM);
1400 fhandle = nfs_alloc_fhandle();
1401 fattr = nfs_alloc_fattr();
1402 if (fhandle == NULL || fattr == NULL)
1405 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1409 trace_nfs_lookup_enter(dir, dentry, flags);
1410 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1411 if (error == -ENOENT)
1414 res = ERR_PTR(error);
1417 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1418 res = ERR_CAST(inode);
1422 /* Notify readdir to use READDIRPLUS */
1423 nfs_force_use_readdirplus(dir);
1426 res = d_splice_alias(inode, dentry);
1432 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1434 trace_nfs_lookup_exit(dir, dentry, flags, error);
1435 nfs4_label_free(label);
1437 nfs_free_fattr(fattr);
1438 nfs_free_fhandle(fhandle);
1441 EXPORT_SYMBOL_GPL(nfs_lookup);
1443 #if IS_ENABLED(CONFIG_NFS_V4)
1444 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1446 const struct dentry_operations nfs4_dentry_operations = {
1447 .d_revalidate = nfs4_lookup_revalidate,
1448 .d_delete = nfs_dentry_delete,
1449 .d_iput = nfs_dentry_iput,
1450 .d_automount = nfs_d_automount,
1451 .d_release = nfs_d_release,
1453 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1455 static fmode_t flags_to_mode(int flags)
1457 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1458 if ((flags & O_ACCMODE) != O_WRONLY)
1460 if ((flags & O_ACCMODE) != O_RDONLY)
1465 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1467 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1470 static int do_open(struct inode *inode, struct file *filp)
1472 nfs_fscache_open_file(inode, filp);
1476 static int nfs_finish_open(struct nfs_open_context *ctx,
1477 struct dentry *dentry,
1478 struct file *file, unsigned open_flags,
1483 err = finish_open(file, dentry, do_open, opened);
1486 nfs_file_set_open_context(file, ctx);
1492 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1493 struct file *file, unsigned open_flags,
1494 umode_t mode, int *opened)
1496 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1497 struct nfs_open_context *ctx;
1499 struct iattr attr = { .ia_valid = ATTR_OPEN };
1500 struct inode *inode;
1501 unsigned int lookup_flags = 0;
1502 bool switched = false;
1505 /* Expect a negative dentry */
1506 BUG_ON(d_inode(dentry));
1508 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1509 dir->i_sb->s_id, dir->i_ino, dentry);
1511 err = nfs_check_flags(open_flags);
1515 /* NFS only supports OPEN on regular files */
1516 if ((open_flags & O_DIRECTORY)) {
1517 if (!d_in_lookup(dentry)) {
1519 * Hashed negative dentry with O_DIRECTORY: dentry was
1520 * revalidated and is fine, no need to perform lookup
1525 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1529 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1530 return -ENAMETOOLONG;
1532 if (open_flags & O_CREAT) {
1533 struct nfs_server *server = NFS_SERVER(dir);
1535 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1536 mode &= ~current_umask();
1538 attr.ia_valid |= ATTR_MODE;
1539 attr.ia_mode = mode;
1541 if (open_flags & O_TRUNC) {
1542 attr.ia_valid |= ATTR_SIZE;
1546 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1549 dentry = d_alloc_parallel(dentry->d_parent,
1550 &dentry->d_name, &wq);
1552 return PTR_ERR(dentry);
1553 if (unlikely(!d_in_lookup(dentry)))
1554 return finish_no_open(file, dentry);
1557 ctx = create_nfs_open_context(dentry, open_flags, file);
1562 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1563 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, opened);
1564 if (IS_ERR(inode)) {
1565 err = PTR_ERR(inode);
1566 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1567 put_nfs_open_context(ctx);
1571 d_add(dentry, NULL);
1572 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1578 if (!(open_flags & O_NOFOLLOW))
1588 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags, opened);
1589 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1590 put_nfs_open_context(ctx);
1592 if (unlikely(switched)) {
1593 d_lookup_done(dentry);
1599 res = nfs_lookup(dir, dentry, lookup_flags);
1601 d_lookup_done(dentry);
1608 return PTR_ERR(res);
1609 return finish_no_open(file, res);
1611 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1613 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1615 struct inode *inode;
1618 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1620 if (d_mountpoint(dentry))
1622 if (NFS_SB(dentry->d_sb)->caps & NFS_CAP_ATOMIC_OPEN_V1)
1625 inode = d_inode(dentry);
1627 /* We can't create new files in nfs_open_revalidate(), so we
1628 * optimize away revalidation of negative dentries.
1630 if (inode == NULL) {
1631 struct dentry *parent;
1634 if (flags & LOOKUP_RCU) {
1635 parent = ACCESS_ONCE(dentry->d_parent);
1636 dir = d_inode_rcu(parent);
1640 parent = dget_parent(dentry);
1641 dir = d_inode(parent);
1643 if (!nfs_neg_need_reval(dir, dentry, flags))
1645 else if (flags & LOOKUP_RCU)
1647 if (!(flags & LOOKUP_RCU))
1649 else if (parent != ACCESS_ONCE(dentry->d_parent))
1654 /* NFS only supports OPEN on regular files */
1655 if (!S_ISREG(inode->i_mode))
1657 /* We cannot do exclusive creation on a positive dentry */
1658 if (flags & LOOKUP_EXCL)
1661 /* Let f_op->open() actually open (and revalidate) the file */
1668 return nfs_lookup_revalidate(dentry, flags);
1671 #endif /* CONFIG_NFSV4 */
1674 * Code common to create, mkdir, and mknod.
1676 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1677 struct nfs_fattr *fattr,
1678 struct nfs4_label *label)
1680 struct dentry *parent = dget_parent(dentry);
1681 struct inode *dir = d_inode(parent);
1682 struct inode *inode;
1683 int error = -EACCES;
1687 /* We may have been initialized further down */
1688 if (d_really_is_positive(dentry))
1690 if (fhandle->size == 0) {
1691 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1695 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1696 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1697 struct nfs_server *server = NFS_SB(dentry->d_sb);
1698 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr, NULL);
1702 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1703 error = PTR_ERR(inode);
1706 d_add(dentry, inode);
1711 nfs_mark_for_revalidate(dir);
1715 EXPORT_SYMBOL_GPL(nfs_instantiate);
1718 * Following a failed create operation, we drop the dentry rather
1719 * than retain a negative dentry. This avoids a problem in the event
1720 * that the operation succeeded on the server, but an error in the
1721 * reply path made it appear to have failed.
1723 int nfs_create(struct inode *dir, struct dentry *dentry,
1724 umode_t mode, bool excl)
1727 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1730 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1731 dir->i_sb->s_id, dir->i_ino, dentry);
1733 attr.ia_mode = mode;
1734 attr.ia_valid = ATTR_MODE;
1736 trace_nfs_create_enter(dir, dentry, open_flags);
1737 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1738 trace_nfs_create_exit(dir, dentry, open_flags, error);
1746 EXPORT_SYMBOL_GPL(nfs_create);
1749 * See comments for nfs_proc_create regarding failed operations.
1752 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1757 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1758 dir->i_sb->s_id, dir->i_ino, dentry);
1760 attr.ia_mode = mode;
1761 attr.ia_valid = ATTR_MODE;
1763 trace_nfs_mknod_enter(dir, dentry);
1764 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1765 trace_nfs_mknod_exit(dir, dentry, status);
1773 EXPORT_SYMBOL_GPL(nfs_mknod);
1776 * See comments for nfs_proc_create regarding failed operations.
1778 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1783 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1784 dir->i_sb->s_id, dir->i_ino, dentry);
1786 attr.ia_valid = ATTR_MODE;
1787 attr.ia_mode = mode | S_IFDIR;
1789 trace_nfs_mkdir_enter(dir, dentry);
1790 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1791 trace_nfs_mkdir_exit(dir, dentry, error);
1799 EXPORT_SYMBOL_GPL(nfs_mkdir);
1801 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1803 if (simple_positive(dentry))
1807 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1811 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1812 dir->i_sb->s_id, dir->i_ino, dentry);
1814 trace_nfs_rmdir_enter(dir, dentry);
1815 if (d_really_is_positive(dentry)) {
1816 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1817 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1818 /* Ensure the VFS deletes this inode */
1821 clear_nlink(d_inode(dentry));
1824 nfs_dentry_handle_enoent(dentry);
1826 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1828 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1829 trace_nfs_rmdir_exit(dir, dentry, error);
1833 EXPORT_SYMBOL_GPL(nfs_rmdir);
1836 * Remove a file after making sure there are no pending writes,
1837 * and after checking that the file has only one user.
1839 * We invalidate the attribute cache and free the inode prior to the operation
1840 * to avoid possible races if the server reuses the inode.
1842 static int nfs_safe_remove(struct dentry *dentry)
1844 struct inode *dir = d_inode(dentry->d_parent);
1845 struct inode *inode = d_inode(dentry);
1848 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1850 /* If the dentry was sillyrenamed, we simply call d_delete() */
1851 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1856 trace_nfs_remove_enter(dir, dentry);
1857 if (inode != NULL) {
1858 NFS_PROTO(inode)->return_delegation(inode);
1859 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1861 nfs_drop_nlink(inode);
1863 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1864 if (error == -ENOENT)
1865 nfs_dentry_handle_enoent(dentry);
1866 trace_nfs_remove_exit(dir, dentry, error);
1871 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1872 * belongs to an active ".nfs..." file and we return -EBUSY.
1874 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1876 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1879 int need_rehash = 0;
1881 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1882 dir->i_ino, dentry);
1884 trace_nfs_unlink_enter(dir, dentry);
1885 spin_lock(&dentry->d_lock);
1886 if (d_count(dentry) > 1) {
1887 spin_unlock(&dentry->d_lock);
1888 /* Start asynchronous writeout of the inode */
1889 write_inode_now(d_inode(dentry), 0);
1890 error = nfs_sillyrename(dir, dentry);
1893 if (!d_unhashed(dentry)) {
1897 spin_unlock(&dentry->d_lock);
1898 error = nfs_safe_remove(dentry);
1899 if (!error || error == -ENOENT) {
1900 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1901 } else if (need_rehash)
1904 trace_nfs_unlink_exit(dir, dentry, error);
1907 EXPORT_SYMBOL_GPL(nfs_unlink);
1910 * To create a symbolic link, most file systems instantiate a new inode,
1911 * add a page to it containing the path, then write it out to the disk
1912 * using prepare_write/commit_write.
1914 * Unfortunately the NFS client can't create the in-core inode first
1915 * because it needs a file handle to create an in-core inode (see
1916 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1917 * symlink request has completed on the server.
1919 * So instead we allocate a raw page, copy the symname into it, then do
1920 * the SYMLINK request with the page as the buffer. If it succeeds, we
1921 * now have a new file handle and can instantiate an in-core NFS inode
1922 * and move the raw page into its mapping.
1924 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1929 unsigned int pathlen = strlen(symname);
1932 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1933 dir->i_ino, dentry, symname);
1935 if (pathlen > PAGE_SIZE)
1936 return -ENAMETOOLONG;
1938 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1939 attr.ia_valid = ATTR_MODE;
1941 page = alloc_page(GFP_USER);
1945 kaddr = page_address(page);
1946 memcpy(kaddr, symname, pathlen);
1947 if (pathlen < PAGE_SIZE)
1948 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1950 trace_nfs_symlink_enter(dir, dentry);
1951 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1952 trace_nfs_symlink_exit(dir, dentry, error);
1954 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1955 dir->i_sb->s_id, dir->i_ino,
1956 dentry, symname, error);
1963 * No big deal if we can't add this page to the page cache here.
1964 * READLINK will get the missing page from the server if needed.
1966 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
1968 SetPageUptodate(page);
1971 * add_to_page_cache_lru() grabs an extra page refcount.
1972 * Drop it here to avoid leaking this page later.
1980 EXPORT_SYMBOL_GPL(nfs_symlink);
1983 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1985 struct inode *inode = d_inode(old_dentry);
1988 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
1989 old_dentry, dentry);
1991 trace_nfs_link_enter(inode, dir, dentry);
1992 NFS_PROTO(inode)->return_delegation(inode);
1995 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
1998 d_add(dentry, inode);
2000 trace_nfs_link_exit(inode, dir, dentry, error);
2003 EXPORT_SYMBOL_GPL(nfs_link);
2006 nfs_complete_rename(struct rpc_task *task, struct nfs_renamedata *data)
2008 struct dentry *old_dentry = data->old_dentry;
2009 struct dentry *new_dentry = data->new_dentry;
2010 struct inode *old_inode = d_inode(old_dentry);
2011 struct inode *new_inode = d_inode(new_dentry);
2013 nfs_mark_for_revalidate(old_inode);
2015 switch (task->tk_status) {
2017 if (new_inode != NULL)
2018 nfs_drop_nlink(new_inode);
2019 d_move(old_dentry, new_dentry);
2020 nfs_set_verifier(new_dentry,
2021 nfs_save_change_attribute(data->new_dir));
2024 nfs_dentry_handle_enoent(old_dentry);
2030 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2031 * different file handle for the same inode after a rename (e.g. when
2032 * moving to a different directory). A fail-safe method to do so would
2033 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2034 * rename the old file using the sillyrename stuff. This way, the original
2035 * file in old_dir will go away when the last process iput()s the inode.
2039 * It actually works quite well. One needs to have the possibility for
2040 * at least one ".nfs..." file in each directory the file ever gets
2041 * moved or linked to which happens automagically with the new
2042 * implementation that only depends on the dcache stuff instead of
2043 * using the inode layer
2045 * Unfortunately, things are a little more complicated than indicated
2046 * above. For a cross-directory move, we want to make sure we can get
2047 * rid of the old inode after the operation. This means there must be
2048 * no pending writes (if it's a file), and the use count must be 1.
2049 * If these conditions are met, we can drop the dentries before doing
2052 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2053 struct inode *new_dir, struct dentry *new_dentry,
2056 struct inode *old_inode = d_inode(old_dentry);
2057 struct inode *new_inode = d_inode(new_dentry);
2058 struct dentry *dentry = NULL, *rehash = NULL;
2059 struct rpc_task *task;
2065 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2066 old_dentry, new_dentry,
2067 d_count(new_dentry));
2069 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2071 * For non-directories, check whether the target is busy and if so,
2072 * make a copy of the dentry and then do a silly-rename. If the
2073 * silly-rename succeeds, the copied dentry is hashed and becomes
2076 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2078 * To prevent any new references to the target during the
2079 * rename, we unhash the dentry in advance.
2081 if (!d_unhashed(new_dentry)) {
2083 rehash = new_dentry;
2086 if (d_count(new_dentry) > 2) {
2089 /* copy the target dentry's name */
2090 dentry = d_alloc(new_dentry->d_parent,
2091 &new_dentry->d_name);
2095 /* silly-rename the existing target ... */
2096 err = nfs_sillyrename(new_dir, new_dentry);
2100 new_dentry = dentry;
2106 NFS_PROTO(old_inode)->return_delegation(old_inode);
2107 if (new_inode != NULL)
2108 NFS_PROTO(new_inode)->return_delegation(new_inode);
2110 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry,
2111 nfs_complete_rename);
2113 error = PTR_ERR(task);
2117 error = rpc_wait_for_completion_task(task);
2119 error = task->tk_status;
2124 trace_nfs_rename_exit(old_dir, old_dentry,
2125 new_dir, new_dentry, error);
2126 /* new dentry created? */
2131 EXPORT_SYMBOL_GPL(nfs_rename);
2133 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2134 static LIST_HEAD(nfs_access_lru_list);
2135 static atomic_long_t nfs_access_nr_entries;
2137 static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2138 module_param(nfs_access_max_cachesize, ulong, 0644);
2139 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2141 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2143 put_rpccred(entry->cred);
2144 kfree_rcu(entry, rcu_head);
2145 smp_mb__before_atomic();
2146 atomic_long_dec(&nfs_access_nr_entries);
2147 smp_mb__after_atomic();
2150 static void nfs_access_free_list(struct list_head *head)
2152 struct nfs_access_entry *cache;
2154 while (!list_empty(head)) {
2155 cache = list_entry(head->next, struct nfs_access_entry, lru);
2156 list_del(&cache->lru);
2157 nfs_access_free_entry(cache);
2161 static unsigned long
2162 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2165 struct nfs_inode *nfsi, *next;
2166 struct nfs_access_entry *cache;
2169 spin_lock(&nfs_access_lru_lock);
2170 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2171 struct inode *inode;
2173 if (nr_to_scan-- == 0)
2175 inode = &nfsi->vfs_inode;
2176 spin_lock(&inode->i_lock);
2177 if (list_empty(&nfsi->access_cache_entry_lru))
2178 goto remove_lru_entry;
2179 cache = list_entry(nfsi->access_cache_entry_lru.next,
2180 struct nfs_access_entry, lru);
2181 list_move(&cache->lru, &head);
2182 rb_erase(&cache->rb_node, &nfsi->access_cache);
2184 if (!list_empty(&nfsi->access_cache_entry_lru))
2185 list_move_tail(&nfsi->access_cache_inode_lru,
2186 &nfs_access_lru_list);
2189 list_del_init(&nfsi->access_cache_inode_lru);
2190 smp_mb__before_atomic();
2191 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2192 smp_mb__after_atomic();
2194 spin_unlock(&inode->i_lock);
2196 spin_unlock(&nfs_access_lru_lock);
2197 nfs_access_free_list(&head);
2202 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2204 int nr_to_scan = sc->nr_to_scan;
2205 gfp_t gfp_mask = sc->gfp_mask;
2207 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2209 return nfs_do_access_cache_scan(nr_to_scan);
2214 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2216 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2220 nfs_access_cache_enforce_limit(void)
2222 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2224 unsigned int nr_to_scan;
2226 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2229 diff = nr_entries - nfs_access_max_cachesize;
2230 if (diff < nr_to_scan)
2232 nfs_do_access_cache_scan(nr_to_scan);
2235 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2237 struct rb_root *root_node = &nfsi->access_cache;
2239 struct nfs_access_entry *entry;
2241 /* Unhook entries from the cache */
2242 while ((n = rb_first(root_node)) != NULL) {
2243 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2244 rb_erase(n, root_node);
2245 list_move(&entry->lru, head);
2247 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2250 void nfs_access_zap_cache(struct inode *inode)
2254 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2256 /* Remove from global LRU init */
2257 spin_lock(&nfs_access_lru_lock);
2258 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2259 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2261 spin_lock(&inode->i_lock);
2262 __nfs_access_zap_cache(NFS_I(inode), &head);
2263 spin_unlock(&inode->i_lock);
2264 spin_unlock(&nfs_access_lru_lock);
2265 nfs_access_free_list(&head);
2267 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2269 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2271 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2272 struct nfs_access_entry *entry;
2275 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2277 if (cred < entry->cred)
2279 else if (cred > entry->cred)
2287 static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res, bool may_block)
2289 struct nfs_inode *nfsi = NFS_I(inode);
2290 struct nfs_access_entry *cache;
2294 spin_lock(&inode->i_lock);
2296 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2298 cache = nfs_access_search_rbtree(inode, cred);
2302 /* Found an entry, is our attribute cache valid? */
2303 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2310 spin_unlock(&inode->i_lock);
2311 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2314 spin_lock(&inode->i_lock);
2317 res->jiffies = cache->jiffies;
2318 res->cred = cache->cred;
2319 res->mask = cache->mask;
2320 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2323 spin_unlock(&inode->i_lock);
2326 spin_unlock(&inode->i_lock);
2327 nfs_access_zap_cache(inode);
2331 static int nfs_access_get_cached_rcu(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
2333 /* Only check the most recently returned cache entry,
2334 * but do it without locking.
2336 struct nfs_inode *nfsi = NFS_I(inode);
2337 struct nfs_access_entry *cache;
2339 struct list_head *lh;
2342 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2344 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2345 cache = list_entry(lh, struct nfs_access_entry, lru);
2346 if (lh == &nfsi->access_cache_entry_lru ||
2347 cred != cache->cred)
2351 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2353 res->jiffies = cache->jiffies;
2354 res->cred = cache->cred;
2355 res->mask = cache->mask;
2362 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2364 struct nfs_inode *nfsi = NFS_I(inode);
2365 struct rb_root *root_node = &nfsi->access_cache;
2366 struct rb_node **p = &root_node->rb_node;
2367 struct rb_node *parent = NULL;
2368 struct nfs_access_entry *entry;
2370 spin_lock(&inode->i_lock);
2371 while (*p != NULL) {
2373 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2375 if (set->cred < entry->cred)
2376 p = &parent->rb_left;
2377 else if (set->cred > entry->cred)
2378 p = &parent->rb_right;
2382 rb_link_node(&set->rb_node, parent, p);
2383 rb_insert_color(&set->rb_node, root_node);
2384 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2385 spin_unlock(&inode->i_lock);
2388 rb_replace_node(parent, &set->rb_node, root_node);
2389 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2390 list_del(&entry->lru);
2391 spin_unlock(&inode->i_lock);
2392 nfs_access_free_entry(entry);
2395 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2397 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2400 RB_CLEAR_NODE(&cache->rb_node);
2401 cache->jiffies = set->jiffies;
2402 cache->cred = get_rpccred(set->cred);
2403 cache->mask = set->mask;
2405 /* The above field assignments must be visible
2406 * before this item appears on the lru. We cannot easily
2407 * use rcu_assign_pointer, so just force the memory barrier.
2410 nfs_access_add_rbtree(inode, cache);
2412 /* Update accounting */
2413 smp_mb__before_atomic();
2414 atomic_long_inc(&nfs_access_nr_entries);
2415 smp_mb__after_atomic();
2417 /* Add inode to global LRU list */
2418 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2419 spin_lock(&nfs_access_lru_lock);
2420 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2421 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2422 &nfs_access_lru_list);
2423 spin_unlock(&nfs_access_lru_lock);
2425 nfs_access_cache_enforce_limit();
2427 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2429 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2432 if (access_result & NFS4_ACCESS_READ)
2433 entry->mask |= MAY_READ;
2435 (NFS4_ACCESS_MODIFY | NFS4_ACCESS_EXTEND | NFS4_ACCESS_DELETE))
2436 entry->mask |= MAY_WRITE;
2437 if (access_result & (NFS4_ACCESS_LOOKUP|NFS4_ACCESS_EXECUTE))
2438 entry->mask |= MAY_EXEC;
2440 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2442 static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2444 struct nfs_access_entry cache;
2445 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2448 trace_nfs_access_enter(inode);
2450 status = nfs_access_get_cached_rcu(inode, cred, &cache);
2452 status = nfs_access_get_cached(inode, cred, &cache, may_block);
2460 /* Be clever: ask server to check for all possible rights */
2461 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2463 cache.jiffies = jiffies;
2464 status = NFS_PROTO(inode)->access(inode, &cache);
2466 if (status == -ESTALE) {
2467 nfs_zap_caches(inode);
2468 if (!S_ISDIR(inode->i_mode))
2469 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2473 nfs_access_add_cache(inode, &cache);
2475 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2478 trace_nfs_access_exit(inode, status);
2482 static int nfs_open_permission_mask(int openflags)
2486 if (openflags & __FMODE_EXEC) {
2487 /* ONLY check exec rights */
2490 if ((openflags & O_ACCMODE) != O_WRONLY)
2492 if ((openflags & O_ACCMODE) != O_RDONLY)
2499 int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2501 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2503 EXPORT_SYMBOL_GPL(nfs_may_open);
2505 static int nfs_execute_ok(struct inode *inode, int mask)
2507 struct nfs_server *server = NFS_SERVER(inode);
2510 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS)) {
2511 if (mask & MAY_NOT_BLOCK)
2513 ret = __nfs_revalidate_inode(server, inode);
2515 if (ret == 0 && !execute_ok(inode))
2520 int nfs_permission(struct inode *inode, int mask)
2522 struct rpc_cred *cred;
2525 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2527 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2529 /* Is this sys_access() ? */
2530 if (mask & (MAY_ACCESS | MAY_CHDIR))
2533 switch (inode->i_mode & S_IFMT) {
2537 if ((mask & MAY_OPEN) &&
2538 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2543 * Optimize away all write operations, since the server
2544 * will check permissions when we perform the op.
2546 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2551 if (!NFS_PROTO(inode)->access)
2554 /* Always try fast lookups first */
2556 cred = rpc_lookup_cred_nonblock();
2558 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2560 res = PTR_ERR(cred);
2562 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2563 /* Fast lookup failed, try the slow way */
2564 cred = rpc_lookup_cred();
2565 if (!IS_ERR(cred)) {
2566 res = nfs_do_access(inode, cred, mask);
2569 res = PTR_ERR(cred);
2572 if (!res && (mask & MAY_EXEC))
2573 res = nfs_execute_ok(inode, mask);
2575 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2576 inode->i_sb->s_id, inode->i_ino, mask, res);
2579 if (mask & MAY_NOT_BLOCK)
2582 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2584 res = generic_permission(inode, mask);
2587 EXPORT_SYMBOL_GPL(nfs_permission);
2591 * version-control: t
2592 * kept-new-versions: 5