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->d_name, fhandle, fattr, label);
1168 if (ret == -ESTALE || ret == -ENOENT)
1173 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1175 if (nfs_refresh_inode(inode, fattr) < 0)
1178 nfs_setsecurity(inode, fattr, label);
1179 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1181 /* set a readdirplus hint that we had a cache miss */
1182 nfs_force_use_readdirplus(dir);
1185 nfs_free_fattr(fattr);
1186 nfs_free_fhandle(fhandle);
1187 nfs4_label_free(label);
1188 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1192 * This is called every time the dcache has a lookup hit,
1193 * and we should check whether we can really trust that
1196 * NOTE! The hit can be a negative hit too, don't assume
1199 * If the parent directory is seen to have changed, we throw out the
1200 * cached dentry and do a new lookup.
1203 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1206 struct inode *inode;
1209 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1210 inode = d_inode(dentry);
1213 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1215 if (is_bad_inode(inode)) {
1216 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1221 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1222 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1224 /* Force a full look up iff the parent directory has changed */
1225 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1226 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1227 error = nfs_lookup_verify_inode(inode, flags);
1229 if (error == -ESTALE)
1230 nfs_zap_caches(dir);
1233 nfs_advise_use_readdirplus(dir);
1237 if (flags & LOOKUP_RCU)
1240 if (NFS_STALE(inode))
1243 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1244 error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1245 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1248 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1250 if (flags & LOOKUP_RCU)
1252 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1256 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1257 int (*reval)(struct inode *, struct dentry *, unsigned int))
1259 struct dentry *parent;
1263 if (flags & LOOKUP_RCU) {
1264 parent = READ_ONCE(dentry->d_parent);
1265 dir = d_inode_rcu(parent);
1268 ret = reval(dir, dentry, flags);
1269 if (parent != READ_ONCE(dentry->d_parent))
1272 parent = dget_parent(dentry);
1273 ret = reval(d_inode(parent), dentry, flags);
1279 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1281 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1285 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1286 * when we don't really care about the dentry name. This is called when a
1287 * pathwalk ends on a dentry that was not found via a normal lookup in the
1288 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1290 * In this situation, we just want to verify that the inode itself is OK
1291 * since the dentry might have changed on the server.
1293 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1295 struct inode *inode = d_inode(dentry);
1299 * I believe we can only get a negative dentry here in the case of a
1300 * procfs-style symlink. Just assume it's correct for now, but we may
1301 * eventually need to do something more here.
1304 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1309 if (is_bad_inode(inode)) {
1310 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1315 error = nfs_lookup_verify_inode(inode, flags);
1316 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1317 __func__, inode->i_ino, error ? "invalid" : "valid");
1322 * This is called from dput() when d_count is going to 0.
1324 static int nfs_dentry_delete(const struct dentry *dentry)
1326 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1327 dentry, dentry->d_flags);
1329 /* Unhash any dentry with a stale inode */
1330 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1333 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1334 /* Unhash it, so that ->d_iput() would be called */
1337 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1338 /* Unhash it, so that ancestors of killed async unlink
1339 * files will be cleaned up during umount */
1346 /* Ensure that we revalidate inode->i_nlink */
1347 static void nfs_drop_nlink(struct inode *inode)
1349 spin_lock(&inode->i_lock);
1350 /* drop the inode if we're reasonably sure this is the last link */
1351 if (inode->i_nlink > 0)
1353 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1354 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE
1355 | NFS_INO_INVALID_CTIME
1356 | NFS_INO_INVALID_OTHER
1357 | NFS_INO_REVAL_FORCED;
1358 spin_unlock(&inode->i_lock);
1362 * Called when the dentry loses inode.
1363 * We use it to clean up silly-renamed files.
1365 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1367 if (S_ISDIR(inode->i_mode))
1368 /* drop any readdir cache as it could easily be old */
1369 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1371 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1372 nfs_complete_unlink(dentry, inode);
1373 nfs_drop_nlink(inode);
1378 static void nfs_d_release(struct dentry *dentry)
1380 /* free cached devname value, if it survived that far */
1381 if (unlikely(dentry->d_fsdata)) {
1382 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1385 kfree(dentry->d_fsdata);
1389 const struct dentry_operations nfs_dentry_operations = {
1390 .d_revalidate = nfs_lookup_revalidate,
1391 .d_weak_revalidate = nfs_weak_revalidate,
1392 .d_delete = nfs_dentry_delete,
1393 .d_iput = nfs_dentry_iput,
1394 .d_automount = nfs_d_automount,
1395 .d_release = nfs_d_release,
1397 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1399 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1402 struct inode *inode = NULL;
1403 struct nfs_fh *fhandle = NULL;
1404 struct nfs_fattr *fattr = NULL;
1405 struct nfs4_label *label = NULL;
1408 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1409 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1411 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1412 return ERR_PTR(-ENAMETOOLONG);
1415 * If we're doing an exclusive create, optimize away the lookup
1416 * but don't hash the dentry.
1418 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1421 res = ERR_PTR(-ENOMEM);
1422 fhandle = nfs_alloc_fhandle();
1423 fattr = nfs_alloc_fattr();
1424 if (fhandle == NULL || fattr == NULL)
1427 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1431 trace_nfs_lookup_enter(dir, dentry, flags);
1432 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, label);
1433 if (error == -ENOENT)
1436 res = ERR_PTR(error);
1439 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1440 res = ERR_CAST(inode);
1444 /* Notify readdir to use READDIRPLUS */
1445 nfs_force_use_readdirplus(dir);
1448 res = d_splice_alias(inode, dentry);
1454 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1456 trace_nfs_lookup_exit(dir, dentry, flags, error);
1457 nfs4_label_free(label);
1459 nfs_free_fattr(fattr);
1460 nfs_free_fhandle(fhandle);
1463 EXPORT_SYMBOL_GPL(nfs_lookup);
1465 #if IS_ENABLED(CONFIG_NFS_V4)
1466 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1468 const struct dentry_operations nfs4_dentry_operations = {
1469 .d_revalidate = nfs4_lookup_revalidate,
1470 .d_weak_revalidate = nfs_weak_revalidate,
1471 .d_delete = nfs_dentry_delete,
1472 .d_iput = nfs_dentry_iput,
1473 .d_automount = nfs_d_automount,
1474 .d_release = nfs_d_release,
1476 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1478 static fmode_t flags_to_mode(int flags)
1480 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1481 if ((flags & O_ACCMODE) != O_WRONLY)
1483 if ((flags & O_ACCMODE) != O_RDONLY)
1488 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1490 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1493 static int do_open(struct inode *inode, struct file *filp)
1495 nfs_fscache_open_file(inode, filp);
1499 static int nfs_finish_open(struct nfs_open_context *ctx,
1500 struct dentry *dentry,
1501 struct file *file, unsigned open_flags)
1505 err = finish_open(file, dentry, do_open);
1508 if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1509 nfs_file_set_open_context(file, ctx);
1516 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1517 struct file *file, unsigned open_flags,
1520 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1521 struct nfs_open_context *ctx;
1523 struct iattr attr = { .ia_valid = ATTR_OPEN };
1524 struct inode *inode;
1525 unsigned int lookup_flags = 0;
1526 bool switched = false;
1530 /* Expect a negative dentry */
1531 BUG_ON(d_inode(dentry));
1533 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1534 dir->i_sb->s_id, dir->i_ino, dentry);
1536 err = nfs_check_flags(open_flags);
1540 /* NFS only supports OPEN on regular files */
1541 if ((open_flags & O_DIRECTORY)) {
1542 if (!d_in_lookup(dentry)) {
1544 * Hashed negative dentry with O_DIRECTORY: dentry was
1545 * revalidated and is fine, no need to perform lookup
1550 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1554 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1555 return -ENAMETOOLONG;
1557 if (open_flags & O_CREAT) {
1558 struct nfs_server *server = NFS_SERVER(dir);
1560 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1561 mode &= ~current_umask();
1563 attr.ia_valid |= ATTR_MODE;
1564 attr.ia_mode = mode;
1566 if (open_flags & O_TRUNC) {
1567 attr.ia_valid |= ATTR_SIZE;
1571 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1574 dentry = d_alloc_parallel(dentry->d_parent,
1575 &dentry->d_name, &wq);
1577 return PTR_ERR(dentry);
1578 if (unlikely(!d_in_lookup(dentry)))
1579 return finish_no_open(file, dentry);
1582 ctx = create_nfs_open_context(dentry, open_flags, file);
1587 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1588 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1590 file->f_mode |= FMODE_CREATED;
1591 if (IS_ERR(inode)) {
1592 err = PTR_ERR(inode);
1593 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1594 put_nfs_open_context(ctx);
1598 d_splice_alias(NULL, dentry);
1599 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1605 if (!(open_flags & O_NOFOLLOW))
1615 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1616 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1617 put_nfs_open_context(ctx);
1619 if (unlikely(switched)) {
1620 d_lookup_done(dentry);
1626 res = nfs_lookup(dir, dentry, lookup_flags);
1628 d_lookup_done(dentry);
1635 return PTR_ERR(res);
1636 return finish_no_open(file, res);
1638 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1641 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1644 struct inode *inode;
1646 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1648 if (d_mountpoint(dentry))
1651 inode = d_inode(dentry);
1653 /* We can't create new files in nfs_open_revalidate(), so we
1654 * optimize away revalidation of negative dentries.
1659 if (NFS_PROTO(dir)->have_delegation(inode, FMODE_READ))
1660 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1662 /* NFS only supports OPEN on regular files */
1663 if (!S_ISREG(inode->i_mode))
1666 /* We cannot do exclusive creation on a positive dentry */
1667 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
1670 /* Check if the directory changed */
1671 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
1674 /* Let f_op->open() actually open (and revalidate) the file */
1677 if (flags & LOOKUP_RCU)
1679 return nfs_lookup_revalidate_dentry(dir, dentry, inode);
1682 return nfs_do_lookup_revalidate(dir, dentry, flags);
1685 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1687 return __nfs_lookup_revalidate(dentry, flags,
1688 nfs4_do_lookup_revalidate);
1691 #endif /* CONFIG_NFSV4 */
1694 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
1695 struct nfs_fattr *fattr,
1696 struct nfs4_label *label)
1698 struct dentry *parent = dget_parent(dentry);
1699 struct inode *dir = d_inode(parent);
1700 struct inode *inode;
1706 if (fhandle->size == 0) {
1707 error = NFS_PROTO(dir)->lookup(dir, &dentry->d_name, fhandle, fattr, NULL);
1711 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1712 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1713 struct nfs_server *server = NFS_SB(dentry->d_sb);
1714 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
1719 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1720 d = d_splice_alias(inode, dentry);
1725 nfs_mark_for_revalidate(dir);
1729 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
1732 * Code common to create, mkdir, and mknod.
1734 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1735 struct nfs_fattr *fattr,
1736 struct nfs4_label *label)
1740 d = nfs_add_or_obtain(dentry, fhandle, fattr, label);
1744 /* Callers don't care */
1748 EXPORT_SYMBOL_GPL(nfs_instantiate);
1751 * Following a failed create operation, we drop the dentry rather
1752 * than retain a negative dentry. This avoids a problem in the event
1753 * that the operation succeeded on the server, but an error in the
1754 * reply path made it appear to have failed.
1756 int nfs_create(struct inode *dir, struct dentry *dentry,
1757 umode_t mode, bool excl)
1760 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
1763 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
1764 dir->i_sb->s_id, dir->i_ino, dentry);
1766 attr.ia_mode = mode;
1767 attr.ia_valid = ATTR_MODE;
1769 trace_nfs_create_enter(dir, dentry, open_flags);
1770 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
1771 trace_nfs_create_exit(dir, dentry, open_flags, error);
1779 EXPORT_SYMBOL_GPL(nfs_create);
1782 * See comments for nfs_proc_create regarding failed operations.
1785 nfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t rdev)
1790 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
1791 dir->i_sb->s_id, dir->i_ino, dentry);
1793 attr.ia_mode = mode;
1794 attr.ia_valid = ATTR_MODE;
1796 trace_nfs_mknod_enter(dir, dentry);
1797 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
1798 trace_nfs_mknod_exit(dir, dentry, status);
1806 EXPORT_SYMBOL_GPL(nfs_mknod);
1809 * See comments for nfs_proc_create regarding failed operations.
1811 int nfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1816 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
1817 dir->i_sb->s_id, dir->i_ino, dentry);
1819 attr.ia_valid = ATTR_MODE;
1820 attr.ia_mode = mode | S_IFDIR;
1822 trace_nfs_mkdir_enter(dir, dentry);
1823 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
1824 trace_nfs_mkdir_exit(dir, dentry, error);
1832 EXPORT_SYMBOL_GPL(nfs_mkdir);
1834 static void nfs_dentry_handle_enoent(struct dentry *dentry)
1836 if (simple_positive(dentry))
1840 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1844 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
1845 dir->i_sb->s_id, dir->i_ino, dentry);
1847 trace_nfs_rmdir_enter(dir, dentry);
1848 if (d_really_is_positive(dentry)) {
1849 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1850 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1851 /* Ensure the VFS deletes this inode */
1854 clear_nlink(d_inode(dentry));
1857 nfs_dentry_handle_enoent(dentry);
1859 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
1861 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1862 trace_nfs_rmdir_exit(dir, dentry, error);
1866 EXPORT_SYMBOL_GPL(nfs_rmdir);
1869 * Remove a file after making sure there are no pending writes,
1870 * and after checking that the file has only one user.
1872 * We invalidate the attribute cache and free the inode prior to the operation
1873 * to avoid possible races if the server reuses the inode.
1875 static int nfs_safe_remove(struct dentry *dentry)
1877 struct inode *dir = d_inode(dentry->d_parent);
1878 struct inode *inode = d_inode(dentry);
1881 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
1883 /* If the dentry was sillyrenamed, we simply call d_delete() */
1884 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1889 trace_nfs_remove_enter(dir, dentry);
1890 if (inode != NULL) {
1891 error = NFS_PROTO(dir)->remove(dir, dentry);
1893 nfs_drop_nlink(inode);
1895 error = NFS_PROTO(dir)->remove(dir, dentry);
1896 if (error == -ENOENT)
1897 nfs_dentry_handle_enoent(dentry);
1898 trace_nfs_remove_exit(dir, dentry, error);
1903 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1904 * belongs to an active ".nfs..." file and we return -EBUSY.
1906 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1908 int nfs_unlink(struct inode *dir, struct dentry *dentry)
1911 int need_rehash = 0;
1913 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
1914 dir->i_ino, dentry);
1916 trace_nfs_unlink_enter(dir, dentry);
1917 spin_lock(&dentry->d_lock);
1918 if (d_count(dentry) > 1) {
1919 spin_unlock(&dentry->d_lock);
1920 /* Start asynchronous writeout of the inode */
1921 write_inode_now(d_inode(dentry), 0);
1922 error = nfs_sillyrename(dir, dentry);
1925 if (!d_unhashed(dentry)) {
1929 spin_unlock(&dentry->d_lock);
1930 error = nfs_safe_remove(dentry);
1931 if (!error || error == -ENOENT) {
1932 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1933 } else if (need_rehash)
1936 trace_nfs_unlink_exit(dir, dentry, error);
1939 EXPORT_SYMBOL_GPL(nfs_unlink);
1942 * To create a symbolic link, most file systems instantiate a new inode,
1943 * add a page to it containing the path, then write it out to the disk
1944 * using prepare_write/commit_write.
1946 * Unfortunately the NFS client can't create the in-core inode first
1947 * because it needs a file handle to create an in-core inode (see
1948 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1949 * symlink request has completed on the server.
1951 * So instead we allocate a raw page, copy the symname into it, then do
1952 * the SYMLINK request with the page as the buffer. If it succeeds, we
1953 * now have a new file handle and can instantiate an in-core NFS inode
1954 * and move the raw page into its mapping.
1956 int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1961 unsigned int pathlen = strlen(symname);
1964 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
1965 dir->i_ino, dentry, symname);
1967 if (pathlen > PAGE_SIZE)
1968 return -ENAMETOOLONG;
1970 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1971 attr.ia_valid = ATTR_MODE;
1973 page = alloc_page(GFP_USER);
1977 kaddr = page_address(page);
1978 memcpy(kaddr, symname, pathlen);
1979 if (pathlen < PAGE_SIZE)
1980 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1982 trace_nfs_symlink_enter(dir, dentry);
1983 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
1984 trace_nfs_symlink_exit(dir, dentry, error);
1986 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
1987 dir->i_sb->s_id, dir->i_ino,
1988 dentry, symname, error);
1995 * No big deal if we can't add this page to the page cache here.
1996 * READLINK will get the missing page from the server if needed.
1998 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2000 SetPageUptodate(page);
2003 * add_to_page_cache_lru() grabs an extra page refcount.
2004 * Drop it here to avoid leaking this page later.
2012 EXPORT_SYMBOL_GPL(nfs_symlink);
2015 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2017 struct inode *inode = d_inode(old_dentry);
2020 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2021 old_dentry, dentry);
2023 trace_nfs_link_enter(inode, dir, dentry);
2025 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2028 d_add(dentry, inode);
2030 trace_nfs_link_exit(inode, dir, dentry, error);
2033 EXPORT_SYMBOL_GPL(nfs_link);
2037 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2038 * different file handle for the same inode after a rename (e.g. when
2039 * moving to a different directory). A fail-safe method to do so would
2040 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2041 * rename the old file using the sillyrename stuff. This way, the original
2042 * file in old_dir will go away when the last process iput()s the inode.
2046 * It actually works quite well. One needs to have the possibility for
2047 * at least one ".nfs..." file in each directory the file ever gets
2048 * moved or linked to which happens automagically with the new
2049 * implementation that only depends on the dcache stuff instead of
2050 * using the inode layer
2052 * Unfortunately, things are a little more complicated than indicated
2053 * above. For a cross-directory move, we want to make sure we can get
2054 * rid of the old inode after the operation. This means there must be
2055 * no pending writes (if it's a file), and the use count must be 1.
2056 * If these conditions are met, we can drop the dentries before doing
2059 int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2060 struct inode *new_dir, struct dentry *new_dentry,
2063 struct inode *old_inode = d_inode(old_dentry);
2064 struct inode *new_inode = d_inode(new_dentry);
2065 struct dentry *dentry = NULL, *rehash = NULL;
2066 struct rpc_task *task;
2072 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2073 old_dentry, new_dentry,
2074 d_count(new_dentry));
2076 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2078 * For non-directories, check whether the target is busy and if so,
2079 * make a copy of the dentry and then do a silly-rename. If the
2080 * silly-rename succeeds, the copied dentry is hashed and becomes
2083 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2085 * To prevent any new references to the target during the
2086 * rename, we unhash the dentry in advance.
2088 if (!d_unhashed(new_dentry)) {
2090 rehash = new_dentry;
2093 if (d_count(new_dentry) > 2) {
2096 /* copy the target dentry's name */
2097 dentry = d_alloc(new_dentry->d_parent,
2098 &new_dentry->d_name);
2102 /* silly-rename the existing target ... */
2103 err = nfs_sillyrename(new_dir, new_dentry);
2107 new_dentry = dentry;
2113 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2115 error = PTR_ERR(task);
2119 error = rpc_wait_for_completion_task(task);
2121 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2122 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2125 error = task->tk_status;
2127 /* Ensure the inode attributes are revalidated */
2129 spin_lock(&old_inode->i_lock);
2130 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2131 NFS_I(old_inode)->cache_validity |= NFS_INO_INVALID_CHANGE
2132 | NFS_INO_INVALID_CTIME
2133 | NFS_INO_REVAL_FORCED;
2134 spin_unlock(&old_inode->i_lock);
2139 trace_nfs_rename_exit(old_dir, old_dentry,
2140 new_dir, new_dentry, error);
2142 if (new_inode != NULL)
2143 nfs_drop_nlink(new_inode);
2145 * The d_move() should be here instead of in an async RPC completion
2146 * handler because we need the proper locks to move the dentry. If
2147 * we're interrupted by a signal, the async RPC completion handler
2148 * should mark the directories for revalidation.
2150 d_move(old_dentry, new_dentry);
2151 nfs_set_verifier(old_dentry,
2152 nfs_save_change_attribute(new_dir));
2153 } else if (error == -ENOENT)
2154 nfs_dentry_handle_enoent(old_dentry);
2156 /* new dentry created? */
2161 EXPORT_SYMBOL_GPL(nfs_rename);
2163 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2164 static LIST_HEAD(nfs_access_lru_list);
2165 static atomic_long_t nfs_access_nr_entries;
2167 static unsigned long nfs_access_max_cachesize = ULONG_MAX;
2168 module_param(nfs_access_max_cachesize, ulong, 0644);
2169 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2171 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2173 put_cred(entry->cred);
2174 kfree_rcu(entry, rcu_head);
2175 smp_mb__before_atomic();
2176 atomic_long_dec(&nfs_access_nr_entries);
2177 smp_mb__after_atomic();
2180 static void nfs_access_free_list(struct list_head *head)
2182 struct nfs_access_entry *cache;
2184 while (!list_empty(head)) {
2185 cache = list_entry(head->next, struct nfs_access_entry, lru);
2186 list_del(&cache->lru);
2187 nfs_access_free_entry(cache);
2191 static unsigned long
2192 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2195 struct nfs_inode *nfsi, *next;
2196 struct nfs_access_entry *cache;
2199 spin_lock(&nfs_access_lru_lock);
2200 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2201 struct inode *inode;
2203 if (nr_to_scan-- == 0)
2205 inode = &nfsi->vfs_inode;
2206 spin_lock(&inode->i_lock);
2207 if (list_empty(&nfsi->access_cache_entry_lru))
2208 goto remove_lru_entry;
2209 cache = list_entry(nfsi->access_cache_entry_lru.next,
2210 struct nfs_access_entry, lru);
2211 list_move(&cache->lru, &head);
2212 rb_erase(&cache->rb_node, &nfsi->access_cache);
2214 if (!list_empty(&nfsi->access_cache_entry_lru))
2215 list_move_tail(&nfsi->access_cache_inode_lru,
2216 &nfs_access_lru_list);
2219 list_del_init(&nfsi->access_cache_inode_lru);
2220 smp_mb__before_atomic();
2221 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2222 smp_mb__after_atomic();
2224 spin_unlock(&inode->i_lock);
2226 spin_unlock(&nfs_access_lru_lock);
2227 nfs_access_free_list(&head);
2232 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2234 int nr_to_scan = sc->nr_to_scan;
2235 gfp_t gfp_mask = sc->gfp_mask;
2237 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2239 return nfs_do_access_cache_scan(nr_to_scan);
2244 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2246 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2250 nfs_access_cache_enforce_limit(void)
2252 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2254 unsigned int nr_to_scan;
2256 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2259 diff = nr_entries - nfs_access_max_cachesize;
2260 if (diff < nr_to_scan)
2262 nfs_do_access_cache_scan(nr_to_scan);
2265 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2267 struct rb_root *root_node = &nfsi->access_cache;
2269 struct nfs_access_entry *entry;
2271 /* Unhook entries from the cache */
2272 while ((n = rb_first(root_node)) != NULL) {
2273 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2274 rb_erase(n, root_node);
2275 list_move(&entry->lru, head);
2277 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2280 void nfs_access_zap_cache(struct inode *inode)
2284 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2286 /* Remove from global LRU init */
2287 spin_lock(&nfs_access_lru_lock);
2288 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2289 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2291 spin_lock(&inode->i_lock);
2292 __nfs_access_zap_cache(NFS_I(inode), &head);
2293 spin_unlock(&inode->i_lock);
2294 spin_unlock(&nfs_access_lru_lock);
2295 nfs_access_free_list(&head);
2297 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2299 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2301 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2304 struct nfs_access_entry *entry =
2305 rb_entry(n, struct nfs_access_entry, rb_node);
2306 int cmp = cred_fscmp(cred, entry->cred);
2318 static int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block)
2320 struct nfs_inode *nfsi = NFS_I(inode);
2321 struct nfs_access_entry *cache;
2325 spin_lock(&inode->i_lock);
2327 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2329 cache = nfs_access_search_rbtree(inode, cred);
2333 /* Found an entry, is our attribute cache valid? */
2334 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2341 spin_unlock(&inode->i_lock);
2342 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2345 spin_lock(&inode->i_lock);
2348 res->cred = cache->cred;
2349 res->mask = cache->mask;
2350 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2353 spin_unlock(&inode->i_lock);
2356 spin_unlock(&inode->i_lock);
2357 nfs_access_zap_cache(inode);
2361 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res)
2363 /* Only check the most recently returned cache entry,
2364 * but do it without locking.
2366 struct nfs_inode *nfsi = NFS_I(inode);
2367 struct nfs_access_entry *cache;
2369 struct list_head *lh;
2372 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2374 lh = rcu_dereference(nfsi->access_cache_entry_lru.prev);
2375 cache = list_entry(lh, struct nfs_access_entry, lru);
2376 if (lh == &nfsi->access_cache_entry_lru ||
2377 cred != cache->cred)
2381 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2383 res->cred = cache->cred;
2384 res->mask = cache->mask;
2391 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2393 struct nfs_inode *nfsi = NFS_I(inode);
2394 struct rb_root *root_node = &nfsi->access_cache;
2395 struct rb_node **p = &root_node->rb_node;
2396 struct rb_node *parent = NULL;
2397 struct nfs_access_entry *entry;
2400 spin_lock(&inode->i_lock);
2401 while (*p != NULL) {
2403 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2404 cmp = cred_fscmp(set->cred, entry->cred);
2407 p = &parent->rb_left;
2409 p = &parent->rb_right;
2413 rb_link_node(&set->rb_node, parent, p);
2414 rb_insert_color(&set->rb_node, root_node);
2415 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2416 spin_unlock(&inode->i_lock);
2419 rb_replace_node(parent, &set->rb_node, root_node);
2420 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2421 list_del(&entry->lru);
2422 spin_unlock(&inode->i_lock);
2423 nfs_access_free_entry(entry);
2426 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2428 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2431 RB_CLEAR_NODE(&cache->rb_node);
2432 cache->cred = get_cred(set->cred);
2433 cache->mask = set->mask;
2435 /* The above field assignments must be visible
2436 * before this item appears on the lru. We cannot easily
2437 * use rcu_assign_pointer, so just force the memory barrier.
2440 nfs_access_add_rbtree(inode, cache);
2442 /* Update accounting */
2443 smp_mb__before_atomic();
2444 atomic_long_inc(&nfs_access_nr_entries);
2445 smp_mb__after_atomic();
2447 /* Add inode to global LRU list */
2448 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2449 spin_lock(&nfs_access_lru_lock);
2450 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2451 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2452 &nfs_access_lru_list);
2453 spin_unlock(&nfs_access_lru_lock);
2455 nfs_access_cache_enforce_limit();
2457 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2459 #define NFS_MAY_READ (NFS_ACCESS_READ)
2460 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2461 NFS_ACCESS_EXTEND | \
2463 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2465 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2466 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2467 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2469 nfs_access_calc_mask(u32 access_result, umode_t umode)
2473 if (access_result & NFS_MAY_READ)
2475 if (S_ISDIR(umode)) {
2476 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2478 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2480 } else if (S_ISREG(umode)) {
2481 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2483 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2485 } else if (access_result & NFS_MAY_WRITE)
2490 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2492 entry->mask = access_result;
2494 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2496 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
2498 struct nfs_access_entry cache;
2499 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2503 trace_nfs_access_enter(inode);
2505 status = nfs_access_get_cached_rcu(inode, cred, &cache);
2507 status = nfs_access_get_cached(inode, cred, &cache, may_block);
2516 * Determine which access bits we want to ask for...
2518 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2519 if (S_ISDIR(inode->i_mode))
2520 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2522 cache.mask |= NFS_ACCESS_EXECUTE;
2524 status = NFS_PROTO(inode)->access(inode, &cache);
2526 if (status == -ESTALE) {
2527 nfs_zap_caches(inode);
2528 if (!S_ISDIR(inode->i_mode))
2529 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2533 nfs_access_add_cache(inode, &cache);
2535 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2536 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2539 trace_nfs_access_exit(inode, status);
2543 static int nfs_open_permission_mask(int openflags)
2547 if (openflags & __FMODE_EXEC) {
2548 /* ONLY check exec rights */
2551 if ((openflags & O_ACCMODE) != O_WRONLY)
2553 if ((openflags & O_ACCMODE) != O_RDONLY)
2560 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
2562 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2564 EXPORT_SYMBOL_GPL(nfs_may_open);
2566 static int nfs_execute_ok(struct inode *inode, int mask)
2568 struct nfs_server *server = NFS_SERVER(inode);
2571 if (S_ISDIR(inode->i_mode))
2573 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) {
2574 if (mask & MAY_NOT_BLOCK)
2576 ret = __nfs_revalidate_inode(server, inode);
2578 if (ret == 0 && !execute_ok(inode))
2583 int nfs_permission(struct inode *inode, int mask)
2585 const struct cred *cred = current_cred();
2588 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2590 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2592 /* Is this sys_access() ? */
2593 if (mask & (MAY_ACCESS | MAY_CHDIR))
2596 switch (inode->i_mode & S_IFMT) {
2600 if ((mask & MAY_OPEN) &&
2601 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2606 * Optimize away all write operations, since the server
2607 * will check permissions when we perform the op.
2609 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2614 if (!NFS_PROTO(inode)->access)
2617 /* Always try fast lookups first */
2619 res = nfs_do_access(inode, cred, mask|MAY_NOT_BLOCK);
2621 if (res == -ECHILD && !(mask & MAY_NOT_BLOCK)) {
2622 /* Fast lookup failed, try the slow way */
2623 res = nfs_do_access(inode, cred, mask);
2626 if (!res && (mask & MAY_EXEC))
2627 res = nfs_execute_ok(inode, mask);
2629 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2630 inode->i_sb->s_id, inode->i_ino, mask, res);
2633 if (mask & MAY_NOT_BLOCK)
2636 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2638 res = generic_permission(inode, mask);
2641 EXPORT_SYMBOL_GPL(nfs_permission);
2645 * version-control: t
2646 * kept-new-versions: 5