1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1992 Rick Sladkey
7 * nfs directory handling functions
9 * 10 Apr 1996 Added silly rename for unlink --okir
10 * 28 Sep 1996 Improved directory cache --okir
11 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
12 * Re-implemented silly rename for unlink, newly implemented
13 * silly rename for nfs_rename() following the suggestions
14 * of Olaf Kirch (okir) found in this file.
15 * Following Linus comments on my original hack, this version
16 * depends only on the dcache stuff and doesn't touch the inode
17 * layer (iput() and friends).
18 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
21 #include <linux/module.h>
22 #include <linux/time.h>
23 #include <linux/errno.h>
24 #include <linux/stat.h>
25 #include <linux/fcntl.h>
26 #include <linux/string.h>
27 #include <linux/kernel.h>
28 #include <linux/slab.h>
30 #include <linux/sunrpc/clnt.h>
31 #include <linux/nfs_fs.h>
32 #include <linux/nfs_mount.h>
33 #include <linux/pagemap.h>
34 #include <linux/pagevec.h>
35 #include <linux/namei.h>
36 #include <linux/mount.h>
37 #include <linux/swap.h>
38 #include <linux/sched.h>
39 #include <linux/kmemleak.h>
40 #include <linux/xattr.h>
42 #include "delegation.h"
49 /* #define NFS_DEBUG_VERBOSE 1 */
51 static int nfs_opendir(struct inode *, struct file *);
52 static int nfs_closedir(struct inode *, struct file *);
53 static int nfs_readdir(struct file *, struct dir_context *);
54 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
55 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
56 static void nfs_readdir_clear_array(struct page*);
58 const struct file_operations nfs_dir_operations = {
59 .llseek = nfs_llseek_dir,
60 .read = generic_read_dir,
61 .iterate = nfs_readdir,
63 .release = nfs_closedir,
64 .fsync = nfs_fsync_dir,
67 const struct address_space_operations nfs_dir_aops = {
68 .freepage = nfs_readdir_clear_array,
71 static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct inode *dir, const struct cred *cred)
73 struct nfs_inode *nfsi = NFS_I(dir);
74 struct nfs_open_dir_context *ctx;
75 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
78 ctx->attr_gencount = nfsi->attr_gencount;
81 ctx->cred = get_cred(cred);
82 spin_lock(&dir->i_lock);
83 if (list_empty(&nfsi->open_files) &&
84 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
85 nfsi->cache_validity |= NFS_INO_INVALID_DATA |
87 list_add(&ctx->list, &nfsi->open_files);
88 spin_unlock(&dir->i_lock);
91 return ERR_PTR(-ENOMEM);
94 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
96 spin_lock(&dir->i_lock);
98 spin_unlock(&dir->i_lock);
107 nfs_opendir(struct inode *inode, struct file *filp)
110 struct nfs_open_dir_context *ctx;
112 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
114 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
116 ctx = alloc_nfs_open_dir_context(inode, current_cred());
121 filp->private_data = ctx;
127 nfs_closedir(struct inode *inode, struct file *filp)
129 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
133 struct nfs_cache_array_entry {
137 unsigned char d_type;
140 struct nfs_cache_array {
144 struct nfs_cache_array_entry array[0];
147 typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, bool);
151 struct dir_context *ctx;
152 unsigned long page_index;
155 loff_t current_index;
156 decode_dirent_t decode;
158 unsigned long timestamp;
159 unsigned long gencount;
160 unsigned int cache_entry_index;
163 } nfs_readdir_descriptor_t;
166 void nfs_readdir_init_array(struct page *page)
168 struct nfs_cache_array *array;
170 array = kmap_atomic(page);
171 memset(array, 0, sizeof(struct nfs_cache_array));
172 array->eof_index = -1;
173 kunmap_atomic(array);
177 * we are freeing strings created by nfs_add_to_readdir_array()
180 void nfs_readdir_clear_array(struct page *page)
182 struct nfs_cache_array *array;
185 array = kmap_atomic(page);
186 for (i = 0; i < array->size; i++)
187 kfree(array->array[i].string.name);
189 kunmap_atomic(array);
193 * the caller is responsible for freeing qstr.name
194 * when called by nfs_readdir_add_to_array, the strings will be freed in
195 * nfs_clear_readdir_array()
198 int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
201 string->name = kmemdup(name, len, GFP_KERNEL);
202 if (string->name == NULL)
205 * Avoid a kmemleak false positive. The pointer to the name is stored
206 * in a page cache page which kmemleak does not scan.
208 kmemleak_not_leak(string->name);
209 string->hash = full_name_hash(NULL, name, len);
214 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
216 struct nfs_cache_array *array = kmap(page);
217 struct nfs_cache_array_entry *cache_entry;
220 cache_entry = &array->array[array->size];
222 /* Check that this entry lies within the page bounds */
224 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
227 cache_entry->cookie = entry->prev_cookie;
228 cache_entry->ino = entry->ino;
229 cache_entry->d_type = entry->d_type;
230 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
233 array->last_cookie = entry->cookie;
236 array->eof_index = array->size;
243 int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
245 loff_t diff = desc->ctx->pos - desc->current_index;
250 if (diff >= array->size) {
251 if (array->eof_index >= 0)
256 index = (unsigned int)diff;
257 *desc->dir_cookie = array->array[index].cookie;
258 desc->cache_entry_index = index;
266 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
268 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
271 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
275 int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
279 int status = -EAGAIN;
281 for (i = 0; i < array->size; i++) {
282 if (array->array[i].cookie == *desc->dir_cookie) {
283 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
284 struct nfs_open_dir_context *ctx = desc->file->private_data;
286 new_pos = desc->current_index + i;
287 if (ctx->attr_gencount != nfsi->attr_gencount ||
288 !nfs_readdir_inode_mapping_valid(nfsi)) {
290 ctx->attr_gencount = nfsi->attr_gencount;
291 } else if (new_pos < desc->ctx->pos) {
293 && ctx->dup_cookie == *desc->dir_cookie) {
294 if (printk_ratelimit()) {
295 pr_notice("NFS: directory %pD2 contains a readdir loop."
296 "Please contact your server vendor. "
297 "The file: %.*s has duplicate cookie %llu\n",
298 desc->file, array->array[i].string.len,
299 array->array[i].string.name, *desc->dir_cookie);
304 ctx->dup_cookie = *desc->dir_cookie;
307 desc->ctx->pos = new_pos;
308 desc->cache_entry_index = i;
312 if (array->eof_index >= 0) {
313 status = -EBADCOOKIE;
314 if (*desc->dir_cookie == array->last_cookie)
322 int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
324 struct nfs_cache_array *array;
327 array = kmap(desc->page);
329 if (*desc->dir_cookie == 0)
330 status = nfs_readdir_search_for_pos(array, desc);
332 status = nfs_readdir_search_for_cookie(array, desc);
334 if (status == -EAGAIN) {
335 desc->last_cookie = array->last_cookie;
336 desc->current_index += array->size;
343 /* Fill a page with xdr information before transferring to the cache page */
345 int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
346 struct nfs_entry *entry, struct file *file, struct inode *inode)
348 struct nfs_open_dir_context *ctx = file->private_data;
349 const struct cred *cred = ctx->cred;
350 unsigned long timestamp, gencount;
355 gencount = nfs_inc_attr_generation_counter();
356 error = NFS_PROTO(inode)->readdir(file_dentry(file), cred, entry->cookie, pages,
357 NFS_SERVER(inode)->dtsize, desc->plus);
359 /* We requested READDIRPLUS, but the server doesn't grok it */
360 if (error == -ENOTSUPP && desc->plus) {
361 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
362 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
368 desc->timestamp = timestamp;
369 desc->gencount = gencount;
374 static int xdr_decode(nfs_readdir_descriptor_t *desc,
375 struct nfs_entry *entry, struct xdr_stream *xdr)
379 error = desc->decode(xdr, entry, desc->plus);
382 entry->fattr->time_start = desc->timestamp;
383 entry->fattr->gencount = desc->gencount;
387 /* Match file and dirent using either filehandle or fileid
388 * Note: caller is responsible for checking the fsid
391 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
394 struct nfs_inode *nfsi;
396 if (d_really_is_negative(dentry))
399 inode = d_inode(dentry);
400 if (is_bad_inode(inode) || NFS_STALE(inode))
404 if (entry->fattr->fileid != nfsi->fileid)
406 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
412 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
414 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
416 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
424 * This function is called by the lookup and getattr code to request the
425 * use of readdirplus to accelerate any future lookups in the same
428 void nfs_advise_use_readdirplus(struct inode *dir)
430 struct nfs_inode *nfsi = NFS_I(dir);
432 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
433 !list_empty(&nfsi->open_files))
434 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
438 * This function is mainly for use by nfs_getattr().
440 * If this is an 'ls -l', we want to force use of readdirplus.
441 * Do this by checking if there is an active file descriptor
442 * and calling nfs_advise_use_readdirplus, then forcing a
445 void nfs_force_use_readdirplus(struct inode *dir)
447 struct nfs_inode *nfsi = NFS_I(dir);
449 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
450 !list_empty(&nfsi->open_files)) {
451 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
452 invalidate_mapping_pages(dir->i_mapping, 0, -1);
457 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
459 struct qstr filename = QSTR_INIT(entry->name, entry->len);
460 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
461 struct dentry *dentry;
462 struct dentry *alias;
463 struct inode *dir = d_inode(parent);
467 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
469 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
471 if (filename.len == 0)
473 /* Validate that the name doesn't contain any illegal '\0' */
474 if (strnlen(filename.name, filename.len) != filename.len)
477 if (strnchr(filename.name, filename.len, '/'))
479 if (filename.name[0] == '.') {
480 if (filename.len == 1)
482 if (filename.len == 2 && filename.name[1] == '.')
485 filename.hash = full_name_hash(parent, filename.name, filename.len);
487 dentry = d_lookup(parent, &filename);
490 dentry = d_alloc_parallel(parent, &filename, &wq);
494 if (!d_in_lookup(dentry)) {
495 /* Is there a mountpoint here? If so, just exit */
496 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
497 &entry->fattr->fsid))
499 if (nfs_same_file(dentry, entry)) {
500 if (!entry->fh->size)
502 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
503 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
505 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
508 d_invalidate(dentry);
514 if (!entry->fh->size) {
515 d_lookup_done(dentry);
519 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
520 alias = d_splice_alias(inode, dentry);
521 d_lookup_done(dentry);
528 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
533 /* Perform conversion from xdr to cache array */
535 int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
536 struct page **xdr_pages, struct page *page, unsigned int buflen)
538 struct xdr_stream stream;
540 struct page *scratch;
541 struct nfs_cache_array *array;
542 unsigned int count = 0;
545 scratch = alloc_page(GFP_KERNEL);
552 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
553 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
556 status = xdr_decode(desc, entry, &stream);
558 if (status == -EAGAIN)
566 nfs_prime_dcache(file_dentry(desc->file), entry);
568 status = nfs_readdir_add_to_array(entry, page);
571 } while (!entry->eof);
574 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
576 array->eof_index = array->size;
586 void nfs_readdir_free_pages(struct page **pages, unsigned int npages)
589 for (i = 0; i < npages; i++)
594 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
595 * to nfs_readdir_free_pages()
598 int nfs_readdir_alloc_pages(struct page **pages, unsigned int npages)
602 for (i = 0; i < npages; i++) {
603 struct page *page = alloc_page(GFP_KERNEL);
611 nfs_readdir_free_pages(pages, i);
616 int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
618 struct page *pages[NFS_MAX_READDIR_PAGES];
619 struct nfs_entry entry;
620 struct file *file = desc->file;
621 struct nfs_cache_array *array;
622 int status = -ENOMEM;
623 unsigned int array_size = ARRAY_SIZE(pages);
625 nfs_readdir_init_array(page);
627 entry.prev_cookie = 0;
628 entry.cookie = desc->last_cookie;
630 entry.fh = nfs_alloc_fhandle();
631 entry.fattr = nfs_alloc_fattr();
632 entry.server = NFS_SERVER(inode);
633 if (entry.fh == NULL || entry.fattr == NULL)
636 entry.label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
637 if (IS_ERR(entry.label)) {
638 status = PTR_ERR(entry.label);
644 status = nfs_readdir_alloc_pages(pages, array_size);
646 goto out_release_array;
649 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
654 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
656 if (status == -ENOSPC)
660 } while (array->eof_index < 0);
662 nfs_readdir_free_pages(pages, array_size);
665 nfs4_label_free(entry.label);
667 nfs_free_fattr(entry.fattr);
668 nfs_free_fhandle(entry.fh);
673 * Now we cache directories properly, by converting xdr information
674 * to an array that can be used for lookups later. This results in
675 * fewer cache pages, since we can store more information on each page.
676 * We only need to convert from xdr once so future lookups are much simpler
679 int nfs_readdir_filler(void *data, struct page* page)
681 nfs_readdir_descriptor_t *desc = data;
682 struct inode *inode = file_inode(desc->file);
685 ret = nfs_readdir_xdr_to_array(desc, page, inode);
688 SetPageUptodate(page);
690 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
691 /* Should never happen */
692 nfs_zap_mapping(inode, inode->i_mapping);
697 nfs_readdir_clear_array(page);
703 void cache_page_release(nfs_readdir_descriptor_t *desc)
705 put_page(desc->page);
710 struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
712 return read_cache_page(desc->file->f_mapping, desc->page_index,
713 nfs_readdir_filler, desc);
717 * Returns 0 if desc->dir_cookie was found on page desc->page_index
718 * and locks the page to prevent removal from the page cache.
721 int find_and_lock_cache_page(nfs_readdir_descriptor_t *desc)
725 desc->page = get_cache_page(desc);
726 if (IS_ERR(desc->page))
727 return PTR_ERR(desc->page);
728 res = lock_page_killable(desc->page);
732 if (desc->page->mapping != NULL) {
733 res = nfs_readdir_search_array(desc);
737 unlock_page(desc->page);
739 cache_page_release(desc);
743 /* Search for desc->dir_cookie from the beginning of the page cache */
745 int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
749 if (desc->page_index == 0) {
750 desc->current_index = 0;
751 desc->last_cookie = 0;
754 res = find_and_lock_cache_page(desc);
755 } while (res == -EAGAIN);
760 * Once we've found the start of the dirent within a page: fill 'er up...
763 int nfs_do_filldir(nfs_readdir_descriptor_t *desc)
765 struct file *file = desc->file;
768 struct nfs_cache_array *array = NULL;
769 struct nfs_open_dir_context *ctx = file->private_data;
771 array = kmap(desc->page);
772 for (i = desc->cache_entry_index; i < array->size; i++) {
773 struct nfs_cache_array_entry *ent;
775 ent = &array->array[i];
776 if (!dir_emit(desc->ctx, ent->string.name, ent->string.len,
777 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
782 if (i < (array->size-1))
783 *desc->dir_cookie = array->array[i+1].cookie;
785 *desc->dir_cookie = array->last_cookie;
789 if (array->eof_index >= 0)
793 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
794 (unsigned long long)*desc->dir_cookie, res);
799 * If we cannot find a cookie in our cache, we suspect that this is
800 * because it points to a deleted file, so we ask the server to return
801 * whatever it thinks is the next entry. We then feed this to filldir.
802 * If all goes well, we should then be able to find our way round the
803 * cache on the next call to readdir_search_pagecache();
805 * NOTE: we cannot add the anonymous page to the pagecache because
806 * the data it contains might not be page aligned. Besides,
807 * we should already have a complete representation of the
808 * directory in the page cache by the time we get here.
811 int uncached_readdir(nfs_readdir_descriptor_t *desc)
813 struct page *page = NULL;
815 struct inode *inode = file_inode(desc->file);
816 struct nfs_open_dir_context *ctx = desc->file->private_data;
818 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
819 (unsigned long long)*desc->dir_cookie);
821 page = alloc_page(GFP_HIGHUSER);
827 desc->page_index = 0;
828 desc->last_cookie = *desc->dir_cookie;
832 status = nfs_readdir_xdr_to_array(desc, page, inode);
836 status = nfs_do_filldir(desc);
839 nfs_readdir_clear_array(desc->page);
840 cache_page_release(desc);
842 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
847 /* The file offset position represents the dirent entry number. A
848 last cookie cache takes care of the common case of reading the
851 static int nfs_readdir(struct file *file, struct dir_context *ctx)
853 struct dentry *dentry = file_dentry(file);
854 struct inode *inode = d_inode(dentry);
855 nfs_readdir_descriptor_t my_desc,
857 struct nfs_open_dir_context *dir_ctx = file->private_data;
860 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
861 file, (long long)ctx->pos);
862 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
865 * ctx->pos points to the dirent entry number.
866 * *desc->dir_cookie has the cookie for the next entry. We have
867 * to either find the entry with the appropriate number or
868 * revalidate the cookie.
870 memset(desc, 0, sizeof(*desc));
874 desc->dir_cookie = &dir_ctx->dir_cookie;
875 desc->decode = NFS_PROTO(inode)->decode_dirent;
876 desc->plus = nfs_use_readdirplus(inode, ctx);
878 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode))
879 res = nfs_revalidate_mapping(inode, file->f_mapping);
884 res = readdir_search_pagecache(desc);
886 if (res == -EBADCOOKIE) {
888 /* This means either end of directory */
889 if (*desc->dir_cookie && !desc->eof) {
890 /* Or that the server has 'lost' a cookie */
891 res = uncached_readdir(desc);
897 if (res == -ETOOSMALL && desc->plus) {
898 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
899 nfs_zap_caches(inode);
900 desc->page_index = 0;
908 res = nfs_do_filldir(desc);
909 unlock_page(desc->page);
910 cache_page_release(desc);
913 } while (!desc->eof);
917 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
921 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
923 struct inode *inode = file_inode(filp);
924 struct nfs_open_dir_context *dir_ctx = filp->private_data;
926 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
927 filp, offset, whence);
941 offset += filp->f_pos;
947 if (offset != filp->f_pos) {
948 filp->f_pos = offset;
949 dir_ctx->dir_cookie = 0;
957 * All directory operations under NFS are synchronous, so fsync()
958 * is a dummy operation.
960 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
963 struct inode *inode = file_inode(filp);
965 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
968 nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
974 * nfs_force_lookup_revalidate - Mark the directory as having changed
975 * @dir: pointer to directory inode
977 * This forces the revalidation code in nfs_lookup_revalidate() to do a
978 * full lookup on all child dentries of 'dir' whenever a change occurs
979 * on the server that might have invalidated our dcache.
981 * The caller should be holding dir->i_lock
983 void nfs_force_lookup_revalidate(struct inode *dir)
985 NFS_I(dir)->cache_change_attribute++;
987 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
990 * A check for whether or not the parent directory has changed.
991 * In the case it has, we assume that the dentries are untrustworthy
992 * and may need to be looked up again.
993 * If rcu_walk prevents us from performing a full check, return 0.
995 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1000 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1002 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1004 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1005 if (nfs_mapping_need_revalidate_inode(dir)) {
1008 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1011 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1017 * Use intent information to check whether or not we're going to do
1018 * an O_EXCL create using this path component.
1020 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1022 if (NFS_PROTO(dir)->version == 2)
1024 return flags & LOOKUP_EXCL;
1028 * Inode and filehandle revalidation for lookups.
1030 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1031 * or if the intent information indicates that we're about to open this
1032 * particular file and the "nocto" mount flag is not set.
1036 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1038 struct nfs_server *server = NFS_SERVER(inode);
1041 if (IS_AUTOMOUNT(inode))
1044 if (flags & LOOKUP_OPEN) {
1045 switch (inode->i_mode & S_IFMT) {
1047 /* A NFSv4 OPEN will revalidate later */
1048 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1052 if (server->flags & NFS_MOUNT_NOCTO)
1054 /* NFS close-to-open cache consistency validation */
1059 /* VFS wants an on-the-wire revalidation */
1060 if (flags & LOOKUP_REVAL)
1063 return (inode->i_nlink == 0) ? -ESTALE : 0;
1065 if (flags & LOOKUP_RCU)
1067 ret = __nfs_revalidate_inode(server, inode);
1074 * We judge how long we want to trust negative
1075 * dentries by looking at the parent inode mtime.
1077 * If parent mtime has changed, we revalidate, else we wait for a
1078 * period corresponding to the parent's attribute cache timeout value.
1080 * If LOOKUP_RCU prevents us from performing a full check, return 1
1081 * suggesting a reval is needed.
1083 * Note that when creating a new file, or looking up a rename target,
1084 * then it shouldn't be necessary to revalidate a negative dentry.
1087 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1090 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1092 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1094 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1098 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1099 struct inode *inode, int error)
1103 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1107 nfs_mark_for_revalidate(dir);
1108 if (inode && S_ISDIR(inode->i_mode)) {
1109 /* Purge readdir caches. */
1110 nfs_zap_caches(inode);
1112 * We can't d_drop the root of a disconnected tree:
1113 * its d_hash is on the s_anon list and d_drop() would hide
1114 * it from shrink_dcache_for_unmount(), leading to busy
1115 * inodes on unmount and further oopses.
1117 if (IS_ROOT(dentry))
1120 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1124 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1125 __func__, dentry, error);
1130 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1134 if (nfs_neg_need_reval(dir, dentry, flags)) {
1135 if (flags & LOOKUP_RCU)
1139 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1143 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1144 struct inode *inode)
1146 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1147 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1151 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry,
1152 struct inode *inode)
1154 struct nfs_fh *fhandle;
1155 struct nfs_fattr *fattr;
1156 struct nfs4_label *label;
1160 fhandle = nfs_alloc_fhandle();
1161 fattr = nfs_alloc_fattr();
1162 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1163 if (fhandle == NULL || fattr == NULL || IS_ERR(label))
1166 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1174 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1180 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1182 if (nfs_refresh_inode(inode, fattr) < 0)
1185 nfs_setsecurity(inode, fattr, label);
1186 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1188 /* set a readdirplus hint that we had a cache miss */
1189 nfs_force_use_readdirplus(dir);
1192 nfs_free_fattr(fattr);
1193 nfs_free_fhandle(fhandle);
1194 nfs4_label_free(label);
1195 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1199 * This is called every time the dcache has a lookup hit,
1200 * and we should check whether we can really trust that
1203 * NOTE! The hit can be a negative hit too, don't assume
1206 * If the parent directory is seen to have changed, we throw out the
1207 * cached dentry and do a new lookup.
1210 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1213 struct inode *inode;
1216 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1217 inode = d_inode(dentry);
1220 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1222 if (is_bad_inode(inode)) {
1223 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1228 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1229 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1231 /* Force a full look up iff the parent directory has changed */
1232 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1233 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1234 error = nfs_lookup_verify_inode(inode, flags);
1236 if (error == -ESTALE)
1237 nfs_zap_caches(dir);
1240 nfs_advise_use_readdirplus(dir);
1244 if (flags & LOOKUP_RCU)
1247 if (NFS_STALE(inode))
1250 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1251 error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1252 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1255 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1257 if (flags & LOOKUP_RCU)
1259 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1263 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1264 int (*reval)(struct inode *, struct dentry *, unsigned int))
1266 struct dentry *parent;
1270 if (flags & LOOKUP_RCU) {
1271 parent = READ_ONCE(dentry->d_parent);
1272 dir = d_inode_rcu(parent);
1275 ret = reval(dir, dentry, flags);
1276 if (parent != READ_ONCE(dentry->d_parent))
1279 parent = dget_parent(dentry);
1280 ret = reval(d_inode(parent), dentry, flags);
1286 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1288 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1292 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1293 * when we don't really care about the dentry name. This is called when a
1294 * pathwalk ends on a dentry that was not found via a normal lookup in the
1295 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1297 * In this situation, we just want to verify that the inode itself is OK
1298 * since the dentry might have changed on the server.
1300 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1302 struct inode *inode = d_inode(dentry);
1306 * I believe we can only get a negative dentry here in the case of a
1307 * procfs-style symlink. Just assume it's correct for now, but we may
1308 * eventually need to do something more here.
1311 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1316 if (is_bad_inode(inode)) {
1317 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1322 error = nfs_lookup_verify_inode(inode, flags);
1323 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1324 __func__, inode->i_ino, error ? "invalid" : "valid");
1329 * This is called from dput() when d_count is going to 0.
1331 static int nfs_dentry_delete(const struct dentry *dentry)
1333 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1334 dentry, dentry->d_flags);
1336 /* Unhash any dentry with a stale inode */
1337 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1340 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1341 /* Unhash it, so that ->d_iput() would be called */
1344 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1345 /* Unhash it, so that ancestors of killed async unlink
1346 * files will be cleaned up during umount */
1353 /* Ensure that we revalidate inode->i_nlink */
1354 static void nfs_drop_nlink(struct inode *inode)
1356 spin_lock(&inode->i_lock);
1357 /* drop the inode if we're reasonably sure this is the last link */
1358 if (inode->i_nlink > 0)
1360 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1361 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE
1362 | NFS_INO_INVALID_CTIME
1363 | NFS_INO_INVALID_OTHER
1364 | NFS_INO_REVAL_FORCED;
1365 spin_unlock(&inode->i_lock);
1369 * Called when the dentry loses inode.
1370 * We use it to clean up silly-renamed files.
1372 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1374 if (S_ISDIR(inode->i_mode))
1375 /* drop any readdir cache as it could easily be old */
1376 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1378 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1379 nfs_complete_unlink(dentry, inode);
1380 nfs_drop_nlink(inode);
1385 static void nfs_d_release(struct dentry *dentry)
1387 /* free cached devname value, if it survived that far */
1388 if (unlikely(dentry->d_fsdata)) {
1389 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1392 kfree(dentry->d_fsdata);
1396 const struct dentry_operations nfs_dentry_operations = {
1397 .d_revalidate = nfs_lookup_revalidate,
1398 .d_weak_revalidate = nfs_weak_revalidate,
1399 .d_delete = nfs_dentry_delete,
1400 .d_iput = nfs_dentry_iput,
1401 .d_automount = nfs_d_automount,
1402 .d_release = nfs_d_release,
1404 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1406 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1409 struct inode *inode = NULL;
1410 struct nfs_fh *fhandle = NULL;
1411 struct nfs_fattr *fattr = NULL;
1412 struct nfs4_label *label = NULL;
1415 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1416 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1418 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1419 return ERR_PTR(-ENAMETOOLONG);
1422 * If we're doing an exclusive create, optimize away the lookup
1423 * but don't hash the dentry.
1425 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1428 res = ERR_PTR(-ENOMEM);
1429 fhandle = nfs_alloc_fhandle();
1430 fattr = nfs_alloc_fattr();
1431 if (fhandle == NULL || fattr == NULL)
1434 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1438 trace_nfs_lookup_enter(dir, dentry, flags);
1439 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1440 if (error == -ENOENT)
1443 res = ERR_PTR(error);
1446 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1447 res = ERR_CAST(inode);
1451 /* Notify readdir to use READDIRPLUS */
1452 nfs_force_use_readdirplus(dir);
1455 res = d_splice_alias(inode, dentry);
1461 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1463 trace_nfs_lookup_exit(dir, dentry, flags, error);
1464 nfs4_label_free(label);
1466 nfs_free_fattr(fattr);
1467 nfs_free_fhandle(fhandle);
1470 EXPORT_SYMBOL_GPL(nfs_lookup);
1472 #if IS_ENABLED(CONFIG_NFS_V4)
1473 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1475 const struct dentry_operations nfs4_dentry_operations = {
1476 .d_revalidate = nfs4_lookup_revalidate,
1477 .d_weak_revalidate = nfs_weak_revalidate,
1478 .d_delete = nfs_dentry_delete,
1479 .d_iput = nfs_dentry_iput,
1480 .d_automount = nfs_d_automount,
1481 .d_release = nfs_d_release,
1483 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1485 static fmode_t flags_to_mode(int flags)
1487 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1488 if ((flags & O_ACCMODE) != O_WRONLY)
1490 if ((flags & O_ACCMODE) != O_RDONLY)
1495 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1497 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1500 static int do_open(struct inode *inode, struct file *filp)
1502 nfs_fscache_open_file(inode, filp);
1506 static int nfs_finish_open(struct nfs_open_context *ctx,
1507 struct dentry *dentry,
1508 struct file *file, unsigned open_flags)
1512 err = finish_open(file, dentry, do_open);
1515 if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1516 nfs_file_set_open_context(file, ctx);
1523 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1524 struct file *file, unsigned open_flags,
1527 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1528 struct nfs_open_context *ctx;
1530 struct iattr attr = { .ia_valid = ATTR_OPEN };
1531 struct inode *inode;
1532 unsigned int lookup_flags = 0;
1533 bool switched = false;
1537 /* Expect a negative dentry */
1538 BUG_ON(d_inode(dentry));
1540 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1541 dir->i_sb->s_id, dir->i_ino, dentry);
1543 err = nfs_check_flags(open_flags);
1547 /* NFS only supports OPEN on regular files */
1548 if ((open_flags & O_DIRECTORY)) {
1549 if (!d_in_lookup(dentry)) {
1551 * Hashed negative dentry with O_DIRECTORY: dentry was
1552 * revalidated and is fine, no need to perform lookup
1557 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1561 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1562 return -ENAMETOOLONG;
1564 if (open_flags & O_CREAT) {
1565 struct nfs_server *server = NFS_SERVER(dir);
1567 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1568 mode &= ~current_umask();
1570 attr.ia_valid |= ATTR_MODE;
1571 attr.ia_mode = mode;
1573 if (open_flags & O_TRUNC) {
1574 attr.ia_valid |= ATTR_SIZE;
1578 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1581 dentry = d_alloc_parallel(dentry->d_parent,
1582 &dentry->d_name, &wq);
1584 return PTR_ERR(dentry);
1585 if (unlikely(!d_in_lookup(dentry)))
1586 return finish_no_open(file, dentry);
1589 ctx = create_nfs_open_context(dentry, open_flags, file);
1594 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1595 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1597 file->f_mode |= FMODE_CREATED;
1598 if (IS_ERR(inode)) {
1599 err = PTR_ERR(inode);
1600 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1601 put_nfs_open_context(ctx);
1605 d_splice_alias(NULL, dentry);
1606 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1612 if (!(open_flags & O_NOFOLLOW))
1622 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1623 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1624 put_nfs_open_context(ctx);
1626 if (unlikely(switched)) {
1627 d_lookup_done(dentry);
1633 res = nfs_lookup(dir, dentry, lookup_flags);
1635 d_lookup_done(dentry);
1642 return PTR_ERR(res);
1643 return finish_no_open(file, res);
1645 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1648 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1651 struct inode *inode;
1653 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1655 if (d_mountpoint(dentry))
1658 inode = d_inode(dentry);
1660 /* We can't create new files in nfs_open_revalidate(), so we
1661 * optimize away revalidation of negative dentries.
1666 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1667 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1669 /* NFS only supports OPEN on regular files */
1670 if (!S_ISREG(inode->i_mode))
1673 /* We cannot do exclusive creation on a positive dentry */
1674 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
1677 /* Check if the directory changed */
1678 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
1681 /* Let f_op->open() actually open (and revalidate) the file */
1684 if (flags & LOOKUP_RCU)
1686 return nfs_lookup_revalidate_dentry(dir, dentry, inode);
1689 return nfs_do_lookup_revalidate(dir, dentry, flags);
1692 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1694 return __nfs_lookup_revalidate(dentry, flags,
1695 nfs4_do_lookup_revalidate);
1698 #endif /* CONFIG_NFSV4 */
1701 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
1702 struct nfs_fattr *fattr,
1703 struct nfs4_label *label)
1705 struct dentry *parent = dget_parent(dentry);
1706 struct inode *dir = d_inode(parent);
1707 struct inode *inode;
1713 if (fhandle->size == 0) {
1714 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL);
1718 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1719 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1720 struct nfs_server *server = NFS_SB(dentry->d_sb);
1721 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
1726 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1727 d = d_splice_alias(inode, dentry);
1732 nfs_mark_for_revalidate(dir);
1736 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
1739 * Code common to create, mkdir, and mknod.
1741 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1742 struct nfs_fattr *fattr,
1743 struct nfs4_label *label)
1747 d = nfs_add_or_obtain(dentry, fhandle, fattr, label);
1751 /* Callers don't care */
1755 EXPORT_SYMBOL_GPL(nfs_instantiate);
1758 * Following a failed create operation, we drop the dentry rather
1759 * than retain a negative dentry. This avoids a problem in the event
1760 * that the operation succeeded on the server, but an error in the
1761 * reply path made it appear to have failed.
1763 int nfs_create(struct inode *dir, struct dentry *dentry,
1764 umode_t mode, bool excl)
1767 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1770 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1771 dir->i_sb->s_id, dir->i_ino, dentry);
1773 attr.ia_mode = mode;
1774 attr.ia_valid = ATTR_MODE;
1776 trace_nfs_create_enter(dir, dentry, open_flags);
1777 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1778 trace_nfs_create_exit(dir, dentry, open_flags, error);
1786 EXPORT_SYMBOL_GPL(nfs_create);
1789 * See comments for nfs_proc_create regarding failed operations.
1792 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1797 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1798 dir->i_sb->s_id, dir->i_ino, dentry);
1800 attr.ia_mode = mode;
1801 attr.ia_valid = ATTR_MODE;
1803 trace_nfs_mknod_enter(dir, dentry);
1804 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1805 trace_nfs_mknod_exit(dir, dentry, status);
1813 EXPORT_SYMBOL_GPL(nfs_mknod);
1816 * See comments for nfs_proc_create regarding failed operations.
1818 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1823 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1824 dir->i_sb->s_id, dir->i_ino, dentry);
1826 attr.ia_valid = ATTR_MODE;
1827 attr.ia_mode = mode | S_IFDIR;
1829 trace_nfs_mkdir_enter(dir, dentry);
1830 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1831 trace_nfs_mkdir_exit(dir, dentry, error);
1839 EXPORT_SYMBOL_GPL(nfs_mkdir);
1841 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1843 if (simple_positive(dentry))
1847 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1851 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1852 dir->i_sb->s_id, dir->i_ino, dentry);
1854 trace_nfs_rmdir_enter(dir, dentry);
1855 if (d_really_is_positive(dentry)) {
1856 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1857 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1858 /* Ensure the VFS deletes this inode */
1861 clear_nlink(d_inode(dentry));
1864 nfs_dentry_handle_enoent(dentry);
1866 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1868 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1869 trace_nfs_rmdir_exit(dir, dentry, error);
1873 EXPORT_SYMBOL_GPL(nfs_rmdir);
1876 * Remove a file after making sure there are no pending writes,
1877 * and after checking that the file has only one user.
1879 * We invalidate the attribute cache and free the inode prior to the operation
1880 * to avoid possible races if the server reuses the inode.
1882 static int nfs_safe_remove(struct dentry *dentry)
1884 struct inode *dir = d_inode(dentry->d_parent);
1885 struct inode *inode = d_inode(dentry);
1888 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1890 /* If the dentry was sillyrenamed, we simply call d_delete() */
1891 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1896 trace_nfs_remove_enter(dir, dentry);
1897 if (inode != NULL) {
1898 error = NFS_PROTO(dir)->remove(dir, dentry);
1900 nfs_drop_nlink(inode);
1902 error = NFS_PROTO(dir)->remove(dir, dentry);
1903 if (error == -ENOENT)
1904 nfs_dentry_handle_enoent(dentry);
1905 trace_nfs_remove_exit(dir, dentry, error);
1910 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1911 * belongs to an active ".nfs..." file and we return -EBUSY.
1913 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1915 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1918 int need_rehash = 0;
1920 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1921 dir->i_ino, dentry);
1923 trace_nfs_unlink_enter(dir, dentry);
1924 spin_lock(&dentry->d_lock);
1925 if (d_count(dentry) > 1) {
1926 spin_unlock(&dentry->d_lock);
1927 /* Start asynchronous writeout of the inode */
1928 write_inode_now(d_inode(dentry), 0);
1929 error = nfs_sillyrename(dir, dentry);
1932 if (!d_unhashed(dentry)) {
1936 spin_unlock(&dentry->d_lock);
1937 error = nfs_safe_remove(dentry);
1938 if (!error || error == -ENOENT) {
1939 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1940 } else if (need_rehash)
1943 trace_nfs_unlink_exit(dir, dentry, error);
1946 EXPORT_SYMBOL_GPL(nfs_unlink);
1949 * To create a symbolic link, most file systems instantiate a new inode,
1950 * add a page to it containing the path, then write it out to the disk
1951 * using prepare_write/commit_write.
1953 * Unfortunately the NFS client can't create the in-core inode first
1954 * because it needs a file handle to create an in-core inode (see
1955 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1956 * symlink request has completed on the server.
1958 * So instead we allocate a raw page, copy the symname into it, then do
1959 * the SYMLINK request with the page as the buffer. If it succeeds, we
1960 * now have a new file handle and can instantiate an in-core NFS inode
1961 * and move the raw page into its mapping.
1963 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1968 unsigned int pathlen = strlen(symname);
1971 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1972 dir->i_ino, dentry, symname);
1974 if (pathlen > PAGE_SIZE)
1975 return -ENAMETOOLONG;
1977 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1978 attr.ia_valid = ATTR_MODE;
1980 page = alloc_page(GFP_USER);
1984 kaddr = page_address(page);
1985 memcpy(kaddr, symname, pathlen);
1986 if (pathlen < PAGE_SIZE)
1987 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1989 trace_nfs_symlink_enter(dir, dentry);
1990 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1991 trace_nfs_symlink_exit(dir, dentry, error);
1993 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1994 dir->i_sb->s_id, dir->i_ino,
1995 dentry, symname, error);
2002 * No big deal if we can't add this page to the page cache here.
2003 * READLINK will get the missing page from the server if needed.
2005 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2007 SetPageUptodate(page);
2010 * add_to_page_cache_lru() grabs an extra page refcount.
2011 * Drop it here to avoid leaking this page later.
2019 EXPORT_SYMBOL_GPL(nfs_symlink);
2022 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2024 struct inode *inode = d_inode(old_dentry);
2027 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2028 old_dentry, dentry);
2030 trace_nfs_link_enter(inode, dir, dentry);
2032 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2035 d_add(dentry, inode);
2037 trace_nfs_link_exit(inode, dir, dentry, error);
2040 EXPORT_SYMBOL_GPL(nfs_link);
2044 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2045 * different file handle for the same inode after a rename (e.g. when
2046 * moving to a different directory). A fail-safe method to do so would
2047 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2048 * rename the old file using the sillyrename stuff. This way, the original
2049 * file in old_dir will go away when the last process iput()s the inode.
2053 * It actually works quite well. One needs to have the possibility for
2054 * at least one ".nfs..." file in each directory the file ever gets
2055 * moved or linked to which happens automagically with the new
2056 * implementation that only depends on the dcache stuff instead of
2057 * using the inode layer
2059 * Unfortunately, things are a little more complicated than indicated
2060 * above. For a cross-directory move, we want to make sure we can get
2061 * rid of the old inode after the operation. This means there must be
2062 * no pending writes (if it's a file), and the use count must be 1.
2063 * If these conditions are met, we can drop the dentries before doing
2066 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2067 struct inode *new_dir, struct dentry *new_dentry,
2070 struct inode *old_inode = d_inode(old_dentry);
2071 struct inode *new_inode = d_inode(new_dentry);
2072 struct dentry *dentry = NULL, *rehash = NULL;
2073 struct rpc_task *task;
2079 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2080 old_dentry, new_dentry,
2081 d_count(new_dentry));
2083 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2085 * For non-directories, check whether the target is busy and if so,
2086 * make a copy of the dentry and then do a silly-rename. If the
2087 * silly-rename succeeds, the copied dentry is hashed and becomes
2090 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2092 * To prevent any new references to the target during the
2093 * rename, we unhash the dentry in advance.
2095 if (!d_unhashed(new_dentry)) {
2097 rehash = new_dentry;
2100 if (d_count(new_dentry) > 2) {
2103 /* copy the target dentry's name */
2104 dentry = d_alloc(new_dentry->d_parent,
2105 &new_dentry->d_name);
2109 /* silly-rename the existing target ... */
2110 err = nfs_sillyrename(new_dir, new_dentry);
2114 new_dentry = dentry;
2120 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2122 error = PTR_ERR(task);
2126 error = rpc_wait_for_completion_task(task);
2128 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2129 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2132 error = task->tk_status;
2134 /* Ensure the inode attributes are revalidated */
2136 spin_lock(&old_inode->i_lock);
2137 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2138 NFS_I(old_inode)->cache_validity |= NFS_INO_INVALID_CHANGE
2139 | NFS_INO_INVALID_CTIME
2140 | NFS_INO_REVAL_FORCED;
2141 spin_unlock(&old_inode->i_lock);
2146 trace_nfs_rename_exit(old_dir, old_dentry,
2147 new_dir, new_dentry, error);
2149 if (new_inode != NULL)
2150 nfs_drop_nlink(new_inode);
2152 * The d_move() should be here instead of in an async RPC completion
2153 * handler because we need the proper locks to move the dentry. If
2154 * we're interrupted by a signal, the async RPC completion handler
2155 * should mark the directories for revalidation.
2157 d_move(old_dentry, new_dentry);
2158 nfs_set_verifier(old_dentry,
2159 nfs_save_change_attribute(new_dir));
2160 } else if (error == -ENOENT)
2161 nfs_dentry_handle_enoent(old_dentry);
2163 /* new dentry created? */
2168 EXPORT_SYMBOL_GPL(nfs_rename);
2170 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2171 static LIST_HEAD(nfs_access_lru_list);
2172 static atomic_long_t nfs_access_nr_entries;
2174 static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2175 module_param(nfs_access_max_cachesize, ulong, 0644);
2176 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2178 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2180 put_cred(entry->cred);
2181 kfree_rcu(entry, rcu_head);
2182 smp_mb__before_atomic();
2183 atomic_long_dec(&nfs_access_nr_entries);
2184 smp_mb__after_atomic();
2187 static void nfs_access_free_list(struct list_head *head)
2189 struct nfs_access_entry *cache;
2191 while (!list_empty(head)) {
2192 cache = list_entry(head->next, struct nfs_access_entry, lru);
2193 list_del(&cache->lru);
2194 nfs_access_free_entry(cache);
2198 static unsigned long
2199 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2202 struct nfs_inode *nfsi, *next;
2203 struct nfs_access_entry *cache;
2206 spin_lock(&nfs_access_lru_lock);
2207 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2208 struct inode *inode;
2210 if (nr_to_scan-- == 0)
2212 inode = &nfsi->vfs_inode;
2213 spin_lock(&inode->i_lock);
2214 if (list_empty(&nfsi->access_cache_entry_lru))
2215 goto remove_lru_entry;
2216 cache = list_entry(nfsi->access_cache_entry_lru.next,
2217 struct nfs_access_entry, lru);
2218 list_move(&cache->lru, &head);
2219 rb_erase(&cache->rb_node, &nfsi->access_cache);
2221 if (!list_empty(&nfsi->access_cache_entry_lru))
2222 list_move_tail(&nfsi->access_cache_inode_lru,
2223 &nfs_access_lru_list);
2226 list_del_init(&nfsi->access_cache_inode_lru);
2227 smp_mb__before_atomic();
2228 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2229 smp_mb__after_atomic();
2231 spin_unlock(&inode->i_lock);
2233 spin_unlock(&nfs_access_lru_lock);
2234 nfs_access_free_list(&head);
2239 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2241 int nr_to_scan = sc->nr_to_scan;
2242 gfp_t gfp_mask = sc->gfp_mask;
2244 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2246 return nfs_do_access_cache_scan(nr_to_scan);
2251 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2253 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2257 nfs_access_cache_enforce_limit(void)
2259 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2261 unsigned int nr_to_scan;
2263 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2266 diff = nr_entries - nfs_access_max_cachesize;
2267 if (diff < nr_to_scan)
2269 nfs_do_access_cache_scan(nr_to_scan);
2272 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2274 struct rb_root *root_node = &nfsi->access_cache;
2276 struct nfs_access_entry *entry;
2278 /* Unhook entries from the cache */
2279 while ((n = rb_first(root_node)) != NULL) {
2280 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2281 rb_erase(n, root_node);
2282 list_move(&entry->lru, head);
2284 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2287 void nfs_access_zap_cache(struct inode *inode)
2291 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2293 /* Remove from global LRU init */
2294 spin_lock(&nfs_access_lru_lock);
2295 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2296 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2298 spin_lock(&inode->i_lock);
2299 __nfs_access_zap_cache(NFS_I(inode), &head);
2300 spin_unlock(&inode->i_lock);
2301 spin_unlock(&nfs_access_lru_lock);
2302 nfs_access_free_list(&head);
2304 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2306 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2308 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2311 struct nfs_access_entry *entry =
2312 rb_entry(n, struct nfs_access_entry, rb_node);
2313 int cmp = cred_fscmp(cred, entry->cred);
2325 static int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block)
2327 struct nfs_inode *nfsi = NFS_I(inode);
2328 struct nfs_access_entry *cache;
2332 spin_lock(&inode->i_lock);
2334 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2336 cache = nfs_access_search_rbtree(inode, cred);
2340 /* Found an entry, is our attribute cache valid? */
2341 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2348 spin_unlock(&inode->i_lock);
2349 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2352 spin_lock(&inode->i_lock);
2355 res->cred = cache->cred;
2356 res->mask = cache->mask;
2357 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2360 spin_unlock(&inode->i_lock);
2363 spin_unlock(&inode->i_lock);
2364 nfs_access_zap_cache(inode);
2368 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res)
2370 /* Only check the most recently returned cache entry,
2371 * but do it without locking.
2373 struct nfs_inode *nfsi = NFS_I(inode);
2374 struct nfs_access_entry *cache;
2376 struct list_head *lh;
2379 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2381 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2382 cache = list_entry(lh, struct nfs_access_entry, lru);
2383 if (lh == &nfsi->access_cache_entry_lru ||
2384 cred_fscmp(cred, cache->cred) != 0)
2388 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2390 res->cred = cache->cred;
2391 res->mask = cache->mask;
2398 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2400 struct nfs_inode *nfsi = NFS_I(inode);
2401 struct rb_root *root_node = &nfsi->access_cache;
2402 struct rb_node **p = &root_node->rb_node;
2403 struct rb_node *parent = NULL;
2404 struct nfs_access_entry *entry;
2407 spin_lock(&inode->i_lock);
2408 while (*p != NULL) {
2410 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2411 cmp = cred_fscmp(set->cred, entry->cred);
2414 p = &parent->rb_left;
2416 p = &parent->rb_right;
2420 rb_link_node(&set->rb_node, parent, p);
2421 rb_insert_color(&set->rb_node, root_node);
2422 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2423 spin_unlock(&inode->i_lock);
2426 rb_replace_node(parent, &set->rb_node, root_node);
2427 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2428 list_del(&entry->lru);
2429 spin_unlock(&inode->i_lock);
2430 nfs_access_free_entry(entry);
2433 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2435 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2438 RB_CLEAR_NODE(&cache->rb_node);
2439 cache->cred = get_cred(set->cred);
2440 cache->mask = set->mask;
2442 /* The above field assignments must be visible
2443 * before this item appears on the lru. We cannot easily
2444 * use rcu_assign_pointer, so just force the memory barrier.
2447 nfs_access_add_rbtree(inode, cache);
2449 /* Update accounting */
2450 smp_mb__before_atomic();
2451 atomic_long_inc(&nfs_access_nr_entries);
2452 smp_mb__after_atomic();
2454 /* Add inode to global LRU list */
2455 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2456 spin_lock(&nfs_access_lru_lock);
2457 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2458 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2459 &nfs_access_lru_list);
2460 spin_unlock(&nfs_access_lru_lock);
2462 nfs_access_cache_enforce_limit();
2464 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2466 #define NFS_MAY_READ (NFS_ACCESS_READ)
2467 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2468 NFS_ACCESS_EXTEND | \
2470 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2472 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2473 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2474 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2476 nfs_access_calc_mask(u32 access_result, umode_t umode)
2480 if (access_result & NFS_MAY_READ)
2482 if (S_ISDIR(umode)) {
2483 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2485 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2487 } else if (S_ISREG(umode)) {
2488 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2490 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2492 } else if (access_result & NFS_MAY_WRITE)
2497 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2499 entry->mask = access_result;
2501 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2503 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
2505 struct nfs_access_entry cache;
2506 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2507 int cache_mask = -1;
2510 trace_nfs_access_enter(inode);
2512 status = nfs_access_get_cached_rcu(inode, cred, &cache);
2514 status = nfs_access_get_cached(inode, cred, &cache, may_block);
2523 * Determine which access bits we want to ask for...
2525 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2526 if (S_ISDIR(inode->i_mode))
2527 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2529 cache.mask |= NFS_ACCESS_EXECUTE;
2531 status = NFS_PROTO(inode)->access(inode, &cache);
2533 if (status == -ESTALE) {
2534 nfs_zap_caches(inode);
2535 if (!S_ISDIR(inode->i_mode))
2536 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2540 nfs_access_add_cache(inode, &cache);
2542 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2543 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2546 trace_nfs_access_exit(inode, mask, cache_mask, status);
2550 static int nfs_open_permission_mask(int openflags)
2554 if (openflags & __FMODE_EXEC) {
2555 /* ONLY check exec rights */
2558 if ((openflags & O_ACCMODE) != O_WRONLY)
2560 if ((openflags & O_ACCMODE) != O_RDONLY)
2567 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
2569 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2571 EXPORT_SYMBOL_GPL(nfs_may_open);
2573 static int nfs_execute_ok(struct inode *inode, int mask)
2575 struct nfs_server *server = NFS_SERVER(inode);
2578 if (S_ISDIR(inode->i_mode))
2580 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) {
2581 if (mask & MAY_NOT_BLOCK)
2583 ret = __nfs_revalidate_inode(server, inode);
2585 if (ret == 0 && !execute_ok(inode))
2590 int nfs_permission(struct inode *inode, int mask)
2592 const struct cred *cred = current_cred();
2595 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2597 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2599 /* Is this sys_access() ? */
2600 if (mask & (MAY_ACCESS | MAY_CHDIR))
2603 switch (inode->i_mode & S_IFMT) {
2607 if ((mask & MAY_OPEN) &&
2608 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2613 * Optimize away all write operations, since the server
2614 * will check permissions when we perform the op.
2616 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2621 if (!NFS_PROTO(inode)->access)
2624 /* Always try fast lookups first */
2626 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2628 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2629 /* Fast lookup failed, try the slow way */
2630 res = nfs_do_access(inode, cred, mask);
2633 if (!res && (mask & MAY_EXEC))
2634 res = nfs_execute_ok(inode, mask);
2636 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2637 inode->i_sb->s_id, inode->i_ino, mask, res);
2640 if (mask & MAY_NOT_BLOCK)
2643 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2645 res = generic_permission(inode, mask);
2648 EXPORT_SYMBOL_GPL(nfs_permission);
2652 * version-control: t
2653 * kept-new-versions: 5