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_shared = 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)
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;
82 spin_lock(&dir->i_lock);
83 if (list_empty(&nfsi->open_files) &&
84 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
85 nfs_set_cache_invalid(dir,
86 NFS_INO_INVALID_DATA |
87 NFS_INO_REVAL_FORCED);
88 list_add(&ctx->list, &nfsi->open_files);
89 clear_bit(NFS_INO_FORCE_READDIR, &nfsi->flags);
90 spin_unlock(&dir->i_lock);
93 return ERR_PTR(-ENOMEM);
96 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
98 spin_lock(&dir->i_lock);
100 spin_unlock(&dir->i_lock);
108 nfs_opendir(struct inode *inode, struct file *filp)
111 struct nfs_open_dir_context *ctx;
113 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
115 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
117 ctx = alloc_nfs_open_dir_context(inode);
122 filp->private_data = ctx;
128 nfs_closedir(struct inode *inode, struct file *filp)
130 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
134 struct nfs_cache_array_entry {
138 unsigned int name_len;
139 unsigned char d_type;
142 struct nfs_cache_array {
145 unsigned char page_full : 1,
147 cookies_are_ordered : 1;
148 struct nfs_cache_array_entry array[];
151 struct nfs_readdir_descriptor {
154 struct dir_context *ctx;
159 loff_t current_index;
162 __be32 verf[NFS_DIR_VERIFIER_SIZE];
163 unsigned long dir_verifier;
164 unsigned long timestamp;
165 unsigned long gencount;
166 unsigned long attr_gencount;
167 unsigned int cache_entry_index;
173 static void nfs_readdir_array_init(struct nfs_cache_array *array)
175 memset(array, 0, sizeof(struct nfs_cache_array));
178 static void nfs_readdir_page_init_array(struct page *page, u64 last_cookie)
180 struct nfs_cache_array *array;
182 array = kmap_atomic(page);
183 nfs_readdir_array_init(array);
184 array->last_cookie = last_cookie;
185 array->cookies_are_ordered = 1;
186 kunmap_atomic(array);
190 * we are freeing strings created by nfs_add_to_readdir_array()
193 void nfs_readdir_clear_array(struct page *page)
195 struct nfs_cache_array *array;
198 array = kmap_atomic(page);
199 for (i = 0; i < array->size; i++)
200 kfree(array->array[i].name);
201 nfs_readdir_array_init(array);
202 kunmap_atomic(array);
206 nfs_readdir_page_array_alloc(u64 last_cookie, gfp_t gfp_flags)
208 struct page *page = alloc_page(gfp_flags);
210 nfs_readdir_page_init_array(page, last_cookie);
214 static void nfs_readdir_page_array_free(struct page *page)
217 nfs_readdir_clear_array(page);
222 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
224 array->page_is_eof = 1;
225 array->page_full = 1;
228 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
230 return array->page_full;
234 * the caller is responsible for freeing qstr.name
235 * when called by nfs_readdir_add_to_array, the strings will be freed in
236 * nfs_clear_readdir_array()
238 static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
240 const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
243 * Avoid a kmemleak false positive. The pointer to the name is stored
244 * in a page cache page which kmemleak does not scan.
247 kmemleak_not_leak(ret);
252 * Check that the next array entry lies entirely within the page bounds
254 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
256 struct nfs_cache_array_entry *cache_entry;
258 if (array->page_full)
260 cache_entry = &array->array[array->size + 1];
261 if ((char *)cache_entry - (char *)array > PAGE_SIZE) {
262 array->page_full = 1;
269 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
271 struct nfs_cache_array *array;
272 struct nfs_cache_array_entry *cache_entry;
276 name = nfs_readdir_copy_name(entry->name, entry->len);
280 array = kmap_atomic(page);
281 ret = nfs_readdir_array_can_expand(array);
287 cache_entry = &array->array[array->size];
288 cache_entry->cookie = entry->prev_cookie;
289 cache_entry->ino = entry->ino;
290 cache_entry->d_type = entry->d_type;
291 cache_entry->name_len = entry->len;
292 cache_entry->name = name;
293 array->last_cookie = entry->cookie;
294 if (array->last_cookie <= cache_entry->cookie)
295 array->cookies_are_ordered = 0;
298 nfs_readdir_array_set_eof(array);
300 kunmap_atomic(array);
304 static struct page *nfs_readdir_page_get_locked(struct address_space *mapping,
305 pgoff_t index, u64 last_cookie)
309 page = grab_cache_page(mapping, index);
310 if (page && !PageUptodate(page)) {
311 nfs_readdir_page_init_array(page, last_cookie);
312 if (invalidate_inode_pages2_range(mapping, index + 1, -1) < 0)
313 nfs_zap_mapping(mapping->host, mapping);
314 SetPageUptodate(page);
320 static u64 nfs_readdir_page_last_cookie(struct page *page)
322 struct nfs_cache_array *array;
325 array = kmap_atomic(page);
326 ret = array->last_cookie;
327 kunmap_atomic(array);
331 static bool nfs_readdir_page_needs_filling(struct page *page)
333 struct nfs_cache_array *array;
336 array = kmap_atomic(page);
337 ret = !nfs_readdir_array_is_full(array);
338 kunmap_atomic(array);
342 static void nfs_readdir_page_set_eof(struct page *page)
344 struct nfs_cache_array *array;
346 array = kmap_atomic(page);
347 nfs_readdir_array_set_eof(array);
348 kunmap_atomic(array);
351 static void nfs_readdir_page_unlock_and_put(struct page *page)
357 static struct page *nfs_readdir_page_get_next(struct address_space *mapping,
358 pgoff_t index, u64 cookie)
362 page = nfs_readdir_page_get_locked(mapping, index, cookie);
364 if (nfs_readdir_page_last_cookie(page) == cookie)
366 nfs_readdir_page_unlock_and_put(page);
372 int is_32bit_api(void)
375 return in_compat_syscall();
377 return (BITS_PER_LONG == 32);
382 bool nfs_readdir_use_cookie(const struct file *filp)
384 if ((filp->f_mode & FMODE_32BITHASH) ||
385 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
390 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
391 struct nfs_readdir_descriptor *desc)
393 loff_t diff = desc->ctx->pos - desc->current_index;
398 if (diff >= array->size) {
399 if (array->page_is_eof)
404 index = (unsigned int)diff;
405 desc->dir_cookie = array->array[index].cookie;
406 desc->cache_entry_index = index;
414 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
416 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
419 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
422 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
425 if (!array->cookies_are_ordered)
427 /* Optimisation for monotonically increasing cookies */
428 if (cookie >= array->last_cookie)
430 if (array->size && cookie < array->array[0].cookie)
435 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
436 struct nfs_readdir_descriptor *desc)
440 int status = -EAGAIN;
442 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
445 for (i = 0; i < array->size; i++) {
446 if (array->array[i].cookie == desc->dir_cookie) {
447 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
449 new_pos = desc->current_index + i;
450 if (desc->attr_gencount != nfsi->attr_gencount ||
451 !nfs_readdir_inode_mapping_valid(nfsi)) {
453 desc->attr_gencount = nfsi->attr_gencount;
454 } else if (new_pos < desc->prev_index) {
456 && desc->dup_cookie == desc->dir_cookie) {
457 if (printk_ratelimit()) {
458 pr_notice("NFS: directory %pD2 contains a readdir loop."
459 "Please contact your server vendor. "
460 "The file: %s has duplicate cookie %llu\n",
461 desc->file, array->array[i].name, desc->dir_cookie);
466 desc->dup_cookie = desc->dir_cookie;
469 if (nfs_readdir_use_cookie(desc->file))
470 desc->ctx->pos = desc->dir_cookie;
472 desc->ctx->pos = new_pos;
473 desc->prev_index = new_pos;
474 desc->cache_entry_index = i;
479 if (array->page_is_eof) {
480 status = -EBADCOOKIE;
481 if (desc->dir_cookie == array->last_cookie)
488 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
490 struct nfs_cache_array *array;
493 array = kmap_atomic(desc->page);
495 if (desc->dir_cookie == 0)
496 status = nfs_readdir_search_for_pos(array, desc);
498 status = nfs_readdir_search_for_cookie(array, desc);
500 if (status == -EAGAIN) {
501 desc->last_cookie = array->last_cookie;
502 desc->current_index += array->size;
505 kunmap_atomic(array);
509 /* Fill a page with xdr information before transferring to the cache page */
510 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
511 __be32 *verf, u64 cookie,
512 struct page **pages, size_t bufsize,
515 struct inode *inode = file_inode(desc->file);
516 struct nfs_readdir_arg arg = {
517 .dentry = file_dentry(desc->file),
518 .cred = desc->file->f_cred,
525 struct nfs_readdir_res res = {
528 unsigned long timestamp, gencount;
533 gencount = nfs_inc_attr_generation_counter();
534 desc->dir_verifier = nfs_save_change_attribute(inode);
535 error = NFS_PROTO(inode)->readdir(&arg, &res);
537 /* We requested READDIRPLUS, but the server doesn't grok it */
538 if (error == -ENOTSUPP && desc->plus) {
539 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
540 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
541 desc->plus = arg.plus = false;
546 desc->timestamp = timestamp;
547 desc->gencount = gencount;
552 static int xdr_decode(struct nfs_readdir_descriptor *desc,
553 struct nfs_entry *entry, struct xdr_stream *xdr)
555 struct inode *inode = file_inode(desc->file);
558 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
561 entry->fattr->time_start = desc->timestamp;
562 entry->fattr->gencount = desc->gencount;
566 /* Match file and dirent using either filehandle or fileid
567 * Note: caller is responsible for checking the fsid
570 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
573 struct nfs_inode *nfsi;
575 if (d_really_is_negative(dentry))
578 inode = d_inode(dentry);
579 if (is_bad_inode(inode) || NFS_STALE(inode))
583 if (entry->fattr->fileid != nfsi->fileid)
585 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
591 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
593 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
595 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
603 * This function is called by the lookup and getattr code to request the
604 * use of readdirplus to accelerate any future lookups in the same
607 void nfs_advise_use_readdirplus(struct inode *dir)
609 struct nfs_inode *nfsi = NFS_I(dir);
611 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
612 !list_empty(&nfsi->open_files))
613 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
617 * This function is mainly for use by nfs_getattr().
619 * If this is an 'ls -l', we want to force use of readdirplus.
620 * Do this by checking if there is an active file descriptor
621 * and calling nfs_advise_use_readdirplus, then forcing a
624 void nfs_force_use_readdirplus(struct inode *dir)
626 struct nfs_inode *nfsi = NFS_I(dir);
628 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
629 !list_empty(&nfsi->open_files)) {
630 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
631 set_bit(NFS_INO_FORCE_READDIR, &nfsi->flags);
636 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
637 unsigned long dir_verifier)
639 struct qstr filename = QSTR_INIT(entry->name, entry->len);
640 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
641 struct dentry *dentry;
642 struct dentry *alias;
646 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
648 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
650 if (filename.len == 0)
652 /* Validate that the name doesn't contain any illegal '\0' */
653 if (strnlen(filename.name, filename.len) != filename.len)
656 if (strnchr(filename.name, filename.len, '/'))
658 if (filename.name[0] == '.') {
659 if (filename.len == 1)
661 if (filename.len == 2 && filename.name[1] == '.')
664 filename.hash = full_name_hash(parent, filename.name, filename.len);
666 dentry = d_lookup(parent, &filename);
669 dentry = d_alloc_parallel(parent, &filename, &wq);
673 if (!d_in_lookup(dentry)) {
674 /* Is there a mountpoint here? If so, just exit */
675 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
676 &entry->fattr->fsid))
678 if (nfs_same_file(dentry, entry)) {
679 if (!entry->fh->size)
681 nfs_set_verifier(dentry, dir_verifier);
682 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
684 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
687 d_invalidate(dentry);
693 if (!entry->fh->size) {
694 d_lookup_done(dentry);
698 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
699 alias = d_splice_alias(inode, dentry);
700 d_lookup_done(dentry);
707 nfs_set_verifier(dentry, dir_verifier);
712 /* Perform conversion from xdr to cache array */
713 static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc,
714 struct nfs_entry *entry,
715 struct page **xdr_pages,
717 struct page **arrays,
720 struct address_space *mapping = desc->file->f_mapping;
721 struct xdr_stream stream;
723 struct page *scratch, *new, *page = *arrays;
726 scratch = alloc_page(GFP_KERNEL);
730 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
731 xdr_set_scratch_page(&stream, scratch);
735 entry->label->len = NFS4_MAXLABELLEN;
737 status = xdr_decode(desc, entry, &stream);
742 nfs_prime_dcache(file_dentry(desc->file), entry,
745 status = nfs_readdir_add_to_array(entry, page);
746 if (status != -ENOSPC)
749 if (page->mapping != mapping) {
752 new = nfs_readdir_page_array_alloc(entry->prev_cookie,
757 *arrays = page = new;
759 new = nfs_readdir_page_get_next(mapping,
765 nfs_readdir_page_unlock_and_put(page);
768 status = nfs_readdir_add_to_array(entry, page);
769 } while (!status && !entry->eof);
774 nfs_readdir_page_set_eof(page);
785 nfs_readdir_page_unlock_and_put(page);
791 static void nfs_readdir_free_pages(struct page **pages, size_t npages)
794 put_page(pages[npages]);
799 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
800 * to nfs_readdir_free_pages()
802 static struct page **nfs_readdir_alloc_pages(size_t npages)
807 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
810 for (i = 0; i < npages; i++) {
811 struct page *page = alloc_page(GFP_KERNEL);
819 nfs_readdir_free_pages(pages, i);
823 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
824 __be32 *verf_arg, __be32 *verf_res,
825 struct page **arrays, size_t narrays)
828 struct page *page = *arrays;
829 struct nfs_entry *entry;
831 struct inode *inode = file_inode(desc->file);
832 size_t dtsize = NFS_SERVER(inode)->dtsize;
833 int status = -ENOMEM;
835 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
838 entry->cookie = nfs_readdir_page_last_cookie(page);
839 entry->fh = nfs_alloc_fhandle();
840 entry->fattr = nfs_alloc_fattr();
841 entry->server = NFS_SERVER(inode);
842 if (entry->fh == NULL || entry->fattr == NULL)
845 entry->label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
846 if (IS_ERR(entry->label)) {
847 status = PTR_ERR(entry->label);
851 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
852 pages = nfs_readdir_alloc_pages(array_size);
854 goto out_release_label;
858 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie,
866 nfs_readdir_page_set_eof(page);
872 status = nfs_readdir_page_filler(desc, entry, pages, pglen,
874 } while (!status && nfs_readdir_page_needs_filling(page) &&
877 nfs_readdir_free_pages(pages, array_size);
879 nfs4_label_free(entry->label);
881 nfs_free_fattr(entry->fattr);
882 nfs_free_fhandle(entry->fh);
887 static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc)
889 put_page(desc->page);
894 nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
896 unlock_page(desc->page);
897 nfs_readdir_page_put(desc);
901 nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc)
903 return nfs_readdir_page_get_locked(desc->file->f_mapping,
909 * Returns 0 if desc->dir_cookie was found on page desc->page_index
910 * and locks the page to prevent removal from the page cache.
912 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
914 struct inode *inode = file_inode(desc->file);
915 struct nfs_inode *nfsi = NFS_I(inode);
916 __be32 verf[NFS_DIR_VERIFIER_SIZE];
919 desc->page = nfs_readdir_page_get_cached(desc);
922 if (nfs_readdir_page_needs_filling(desc->page)) {
923 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
926 nfs_readdir_page_unlock_and_put_cached(desc);
927 if (res == -EBADCOOKIE || res == -ENOTSYNC) {
928 invalidate_inode_pages2(desc->file->f_mapping);
929 desc->page_index = 0;
935 * Set the cookie verifier if the page cache was empty
937 if (desc->page_index == 0)
938 memcpy(nfsi->cookieverf, verf,
939 sizeof(nfsi->cookieverf));
941 res = nfs_readdir_search_array(desc);
944 nfs_readdir_page_unlock_and_put_cached(desc);
948 static bool nfs_readdir_dont_search_cache(struct nfs_readdir_descriptor *desc)
950 struct address_space *mapping = desc->file->f_mapping;
951 struct inode *dir = file_inode(desc->file);
952 unsigned int dtsize = NFS_SERVER(dir)->dtsize;
953 loff_t size = i_size_read(dir);
956 * Default to uncached readdir if the page cache is empty, and
957 * we're looking for a non-zero cookie in a large directory.
959 return desc->dir_cookie != 0 && mapping->nrpages == 0 && size > dtsize;
962 /* Search for desc->dir_cookie from the beginning of the page cache */
963 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
967 if (nfs_readdir_dont_search_cache(desc))
971 if (desc->page_index == 0) {
972 desc->current_index = 0;
973 desc->prev_index = 0;
974 desc->last_cookie = 0;
976 res = find_and_lock_cache_page(desc);
977 } while (res == -EAGAIN);
982 * Once we've found the start of the dirent within a page: fill 'er up...
984 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
987 struct file *file = desc->file;
988 struct nfs_cache_array *array;
991 array = kmap(desc->page);
992 for (i = desc->cache_entry_index; i < array->size; i++) {
993 struct nfs_cache_array_entry *ent;
995 ent = &array->array[i];
996 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
997 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
1001 memcpy(desc->verf, verf, sizeof(desc->verf));
1002 if (i < (array->size-1))
1003 desc->dir_cookie = array->array[i+1].cookie;
1005 desc->dir_cookie = array->last_cookie;
1006 if (nfs_readdir_use_cookie(file))
1007 desc->ctx->pos = desc->dir_cookie;
1010 if (desc->duped != 0)
1013 if (array->page_is_eof)
1017 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1018 (unsigned long long)desc->dir_cookie);
1022 * If we cannot find a cookie in our cache, we suspect that this is
1023 * because it points to a deleted file, so we ask the server to return
1024 * whatever it thinks is the next entry. We then feed this to filldir.
1025 * If all goes well, we should then be able to find our way round the
1026 * cache on the next call to readdir_search_pagecache();
1028 * NOTE: we cannot add the anonymous page to the pagecache because
1029 * the data it contains might not be page aligned. Besides,
1030 * we should already have a complete representation of the
1031 * directory in the page cache by the time we get here.
1033 static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1035 struct page **arrays;
1037 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1038 int status = -ENOMEM;
1040 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1041 (unsigned long long)desc->dir_cookie);
1043 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1046 arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL);
1050 desc->page_index = 0;
1051 desc->cache_entry_index = 0;
1052 desc->last_cookie = desc->dir_cookie;
1055 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1057 for (i = 0; !desc->eof && i < sz && arrays[i]; i++) {
1058 desc->page = arrays[i];
1059 nfs_do_filldir(desc, verf);
1064 for (i = 0; i < sz && arrays[i]; i++)
1065 nfs_readdir_page_array_free(arrays[i]);
1068 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1072 /* The file offset position represents the dirent entry number. A
1073 last cookie cache takes care of the common case of reading the
1076 static int nfs_readdir(struct file *file, struct dir_context *ctx)
1078 struct dentry *dentry = file_dentry(file);
1079 struct inode *inode = d_inode(dentry);
1080 struct nfs_inode *nfsi = NFS_I(inode);
1081 struct nfs_open_dir_context *dir_ctx = file->private_data;
1082 struct nfs_readdir_descriptor *desc;
1086 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1087 file, (long long)ctx->pos);
1088 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1091 * ctx->pos points to the dirent entry number.
1092 * *desc->dir_cookie has the cookie for the next entry. We have
1093 * to either find the entry with the appropriate number or
1094 * revalidate the cookie.
1096 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) {
1097 res = nfs_revalidate_mapping(inode, file->f_mapping);
1103 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1108 desc->plus = nfs_use_readdirplus(inode, ctx);
1110 spin_lock(&file->f_lock);
1111 desc->dir_cookie = dir_ctx->dir_cookie;
1112 desc->dup_cookie = dir_ctx->dup_cookie;
1113 desc->duped = dir_ctx->duped;
1114 page_index = dir_ctx->page_index;
1115 desc->attr_gencount = dir_ctx->attr_gencount;
1116 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1117 spin_unlock(&file->f_lock);
1119 if (test_and_clear_bit(NFS_INO_FORCE_READDIR, &nfsi->flags) &&
1120 list_is_singular(&nfsi->open_files))
1121 invalidate_mapping_pages(inode->i_mapping, page_index + 1, -1);
1124 res = readdir_search_pagecache(desc);
1126 if (res == -EBADCOOKIE) {
1128 /* This means either end of directory */
1129 if (desc->dir_cookie && !desc->eof) {
1130 /* Or that the server has 'lost' a cookie */
1131 res = uncached_readdir(desc);
1134 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1139 if (res == -ETOOSMALL && desc->plus) {
1140 clear_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
1141 nfs_zap_caches(inode);
1142 desc->page_index = 0;
1150 nfs_do_filldir(desc, nfsi->cookieverf);
1151 nfs_readdir_page_unlock_and_put_cached(desc);
1152 } while (!desc->eof);
1154 spin_lock(&file->f_lock);
1155 dir_ctx->dir_cookie = desc->dir_cookie;
1156 dir_ctx->dup_cookie = desc->dup_cookie;
1157 dir_ctx->duped = desc->duped;
1158 dir_ctx->attr_gencount = desc->attr_gencount;
1159 dir_ctx->page_index = desc->page_index;
1160 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1161 spin_unlock(&file->f_lock);
1166 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1170 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1172 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1174 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1175 filp, offset, whence);
1183 spin_lock(&filp->f_lock);
1188 spin_lock(&filp->f_lock);
1189 offset += filp->f_pos;
1191 spin_unlock(&filp->f_lock);
1195 if (offset != filp->f_pos) {
1196 filp->f_pos = offset;
1197 if (nfs_readdir_use_cookie(filp))
1198 dir_ctx->dir_cookie = offset;
1200 dir_ctx->dir_cookie = 0;
1202 memset(dir_ctx->verf, 0, sizeof(dir_ctx->verf));
1205 spin_unlock(&filp->f_lock);
1210 * All directory operations under NFS are synchronous, so fsync()
1211 * is a dummy operation.
1213 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1216 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1218 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1223 * nfs_force_lookup_revalidate - Mark the directory as having changed
1224 * @dir: pointer to directory inode
1226 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1227 * full lookup on all child dentries of 'dir' whenever a change occurs
1228 * on the server that might have invalidated our dcache.
1230 * Note that we reserve bit '0' as a tag to let us know when a dentry
1231 * was revalidated while holding a delegation on its inode.
1233 * The caller should be holding dir->i_lock
1235 void nfs_force_lookup_revalidate(struct inode *dir)
1237 NFS_I(dir)->cache_change_attribute += 2;
1239 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1242 * nfs_verify_change_attribute - Detects NFS remote directory changes
1243 * @dir: pointer to parent directory inode
1244 * @verf: previously saved change attribute
1246 * Return "false" if the verifiers doesn't match the change attribute.
1247 * This would usually indicate that the directory contents have changed on
1248 * the server, and that any dentries need revalidating.
1250 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1252 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1255 static void nfs_set_verifier_delegated(unsigned long *verf)
1260 #if IS_ENABLED(CONFIG_NFS_V4)
1261 static void nfs_unset_verifier_delegated(unsigned long *verf)
1265 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1267 static bool nfs_test_verifier_delegated(unsigned long verf)
1272 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1274 return nfs_test_verifier_delegated(dentry->d_time);
1277 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1279 struct inode *inode = d_inode(dentry);
1280 struct inode *dir = d_inode(dentry->d_parent);
1282 if (!nfs_verify_change_attribute(dir, verf))
1284 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1285 nfs_set_verifier_delegated(&verf);
1286 dentry->d_time = verf;
1290 * nfs_set_verifier - save a parent directory verifier in the dentry
1291 * @dentry: pointer to dentry
1292 * @verf: verifier to save
1294 * Saves the parent directory verifier in @dentry. If the inode has
1295 * a delegation, we also tag the dentry as having been revalidated
1296 * while holding a delegation so that we know we don't have to
1297 * look it up again after a directory change.
1299 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1302 spin_lock(&dentry->d_lock);
1303 nfs_set_verifier_locked(dentry, verf);
1304 spin_unlock(&dentry->d_lock);
1306 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1308 #if IS_ENABLED(CONFIG_NFS_V4)
1310 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1311 * @inode: pointer to inode
1313 * Iterates through the dentries in the inode alias list and clears
1314 * the tag used to indicate that the dentry has been revalidated
1315 * while holding a delegation.
1316 * This function is intended for use when the delegation is being
1317 * returned or revoked.
1319 void nfs_clear_verifier_delegated(struct inode *inode)
1321 struct dentry *alias;
1325 spin_lock(&inode->i_lock);
1326 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1327 spin_lock(&alias->d_lock);
1328 nfs_unset_verifier_delegated(&alias->d_time);
1329 spin_unlock(&alias->d_lock);
1331 spin_unlock(&inode->i_lock);
1333 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1334 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1337 * A check for whether or not the parent directory has changed.
1338 * In the case it has, we assume that the dentries are untrustworthy
1339 * and may need to be looked up again.
1340 * If rcu_walk prevents us from performing a full check, return 0.
1342 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1345 if (IS_ROOT(dentry))
1347 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1349 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1351 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1352 if (nfs_mapping_need_revalidate_inode(dir)) {
1355 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1358 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1364 * Use intent information to check whether or not we're going to do
1365 * an O_EXCL create using this path component.
1367 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1369 if (NFS_PROTO(dir)->version == 2)
1371 return flags & LOOKUP_EXCL;
1375 * Inode and filehandle revalidation for lookups.
1377 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1378 * or if the intent information indicates that we're about to open this
1379 * particular file and the "nocto" mount flag is not set.
1383 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1385 struct nfs_server *server = NFS_SERVER(inode);
1388 if (IS_AUTOMOUNT(inode))
1391 if (flags & LOOKUP_OPEN) {
1392 switch (inode->i_mode & S_IFMT) {
1394 /* A NFSv4 OPEN will revalidate later */
1395 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1399 if (server->flags & NFS_MOUNT_NOCTO)
1401 /* NFS close-to-open cache consistency validation */
1406 /* VFS wants an on-the-wire revalidation */
1407 if (flags & LOOKUP_REVAL)
1410 return (inode->i_nlink == 0) ? -ESTALE : 0;
1412 if (flags & LOOKUP_RCU)
1414 ret = __nfs_revalidate_inode(server, inode);
1420 static void nfs_mark_dir_for_revalidate(struct inode *inode)
1422 spin_lock(&inode->i_lock);
1423 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1424 spin_unlock(&inode->i_lock);
1428 * We judge how long we want to trust negative
1429 * dentries by looking at the parent inode mtime.
1431 * If parent mtime has changed, we revalidate, else we wait for a
1432 * period corresponding to the parent's attribute cache timeout value.
1434 * If LOOKUP_RCU prevents us from performing a full check, return 1
1435 * suggesting a reval is needed.
1437 * Note that when creating a new file, or looking up a rename target,
1438 * then it shouldn't be necessary to revalidate a negative dentry.
1441 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1444 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1446 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1448 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1452 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1453 struct inode *inode, int error)
1457 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1462 * We can't d_drop the root of a disconnected tree:
1463 * its d_hash is on the s_anon list and d_drop() would hide
1464 * it from shrink_dcache_for_unmount(), leading to busy
1465 * inodes on unmount and further oopses.
1467 if (inode && IS_ROOT(dentry))
1469 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1473 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1474 __func__, dentry, error);
1479 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1483 if (nfs_neg_need_reval(dir, dentry, flags)) {
1484 if (flags & LOOKUP_RCU)
1488 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1492 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1493 struct inode *inode)
1495 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1496 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1500 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry,
1501 struct inode *inode)
1503 struct nfs_fh *fhandle;
1504 struct nfs_fattr *fattr;
1505 struct nfs4_label *label;
1506 unsigned long dir_verifier;
1510 fhandle = nfs_alloc_fhandle();
1511 fattr = nfs_alloc_fattr();
1512 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1513 if (fhandle == NULL || fattr == NULL || IS_ERR(label))
1516 dir_verifier = nfs_save_change_attribute(dir);
1517 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1525 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1531 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1533 if (nfs_refresh_inode(inode, fattr) < 0)
1536 nfs_setsecurity(inode, fattr, label);
1537 nfs_set_verifier(dentry, dir_verifier);
1539 /* set a readdirplus hint that we had a cache miss */
1540 nfs_force_use_readdirplus(dir);
1543 nfs_free_fattr(fattr);
1544 nfs_free_fhandle(fhandle);
1545 nfs4_label_free(label);
1548 * If the lookup failed despite the dentry change attribute being
1549 * a match, then we should revalidate the directory cache.
1551 if (!ret && nfs_verify_change_attribute(dir, dentry->d_time))
1552 nfs_mark_dir_for_revalidate(dir);
1553 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1557 * This is called every time the dcache has a lookup hit,
1558 * and we should check whether we can really trust that
1561 * NOTE! The hit can be a negative hit too, don't assume
1564 * If the parent directory is seen to have changed, we throw out the
1565 * cached dentry and do a new lookup.
1568 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1571 struct inode *inode;
1574 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1575 inode = d_inode(dentry);
1578 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1580 if (is_bad_inode(inode)) {
1581 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1586 if (nfs_verifier_is_delegated(dentry))
1587 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1589 /* Force a full look up iff the parent directory has changed */
1590 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1591 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1592 error = nfs_lookup_verify_inode(inode, flags);
1594 if (error == -ESTALE)
1595 nfs_mark_dir_for_revalidate(dir);
1598 nfs_advise_use_readdirplus(dir);
1602 if (flags & LOOKUP_RCU)
1605 if (NFS_STALE(inode))
1608 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1609 error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1610 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1613 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1615 if (flags & LOOKUP_RCU)
1617 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1621 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1622 int (*reval)(struct inode *, struct dentry *, unsigned int))
1624 struct dentry *parent;
1628 if (flags & LOOKUP_RCU) {
1629 parent = READ_ONCE(dentry->d_parent);
1630 dir = d_inode_rcu(parent);
1633 ret = reval(dir, dentry, flags);
1634 if (parent != READ_ONCE(dentry->d_parent))
1637 parent = dget_parent(dentry);
1638 ret = reval(d_inode(parent), dentry, flags);
1644 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1646 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1650 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1651 * when we don't really care about the dentry name. This is called when a
1652 * pathwalk ends on a dentry that was not found via a normal lookup in the
1653 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1655 * In this situation, we just want to verify that the inode itself is OK
1656 * since the dentry might have changed on the server.
1658 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1660 struct inode *inode = d_inode(dentry);
1664 * I believe we can only get a negative dentry here in the case of a
1665 * procfs-style symlink. Just assume it's correct for now, but we may
1666 * eventually need to do something more here.
1669 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1674 if (is_bad_inode(inode)) {
1675 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1680 error = nfs_lookup_verify_inode(inode, flags);
1681 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1682 __func__, inode->i_ino, error ? "invalid" : "valid");
1687 * This is called from dput() when d_count is going to 0.
1689 static int nfs_dentry_delete(const struct dentry *dentry)
1691 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1692 dentry, dentry->d_flags);
1694 /* Unhash any dentry with a stale inode */
1695 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1698 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1699 /* Unhash it, so that ->d_iput() would be called */
1702 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1703 /* Unhash it, so that ancestors of killed async unlink
1704 * files will be cleaned up during umount */
1711 /* Ensure that we revalidate inode->i_nlink */
1712 static void nfs_drop_nlink(struct inode *inode)
1714 spin_lock(&inode->i_lock);
1715 /* drop the inode if we're reasonably sure this is the last link */
1716 if (inode->i_nlink > 0)
1718 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1719 nfs_set_cache_invalid(
1720 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1721 NFS_INO_INVALID_NLINK);
1722 spin_unlock(&inode->i_lock);
1726 * Called when the dentry loses inode.
1727 * We use it to clean up silly-renamed files.
1729 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1731 if (S_ISDIR(inode->i_mode))
1732 /* drop any readdir cache as it could easily be old */
1733 nfs_set_cache_invalid(inode, NFS_INO_INVALID_DATA);
1735 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1736 nfs_complete_unlink(dentry, inode);
1737 nfs_drop_nlink(inode);
1742 static void nfs_d_release(struct dentry *dentry)
1744 /* free cached devname value, if it survived that far */
1745 if (unlikely(dentry->d_fsdata)) {
1746 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1749 kfree(dentry->d_fsdata);
1753 const struct dentry_operations nfs_dentry_operations = {
1754 .d_revalidate = nfs_lookup_revalidate,
1755 .d_weak_revalidate = nfs_weak_revalidate,
1756 .d_delete = nfs_dentry_delete,
1757 .d_iput = nfs_dentry_iput,
1758 .d_automount = nfs_d_automount,
1759 .d_release = nfs_d_release,
1761 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1763 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1766 struct inode *inode = NULL;
1767 struct nfs_fh *fhandle = NULL;
1768 struct nfs_fattr *fattr = NULL;
1769 struct nfs4_label *label = NULL;
1770 unsigned long dir_verifier;
1773 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1774 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1776 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1777 return ERR_PTR(-ENAMETOOLONG);
1780 * If we're doing an exclusive create, optimize away the lookup
1781 * but don't hash the dentry.
1783 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1786 res = ERR_PTR(-ENOMEM);
1787 fhandle = nfs_alloc_fhandle();
1788 fattr = nfs_alloc_fattr();
1789 if (fhandle == NULL || fattr == NULL)
1792 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1796 dir_verifier = nfs_save_change_attribute(dir);
1797 trace_nfs_lookup_enter(dir, dentry, flags);
1798 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1799 if (error == -ENOENT)
1802 res = ERR_PTR(error);
1805 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1806 res = ERR_CAST(inode);
1810 /* Notify readdir to use READDIRPLUS */
1811 nfs_force_use_readdirplus(dir);
1814 res = d_splice_alias(inode, dentry);
1820 nfs_set_verifier(dentry, dir_verifier);
1822 trace_nfs_lookup_exit(dir, dentry, flags, error);
1823 nfs4_label_free(label);
1825 nfs_free_fattr(fattr);
1826 nfs_free_fhandle(fhandle);
1829 EXPORT_SYMBOL_GPL(nfs_lookup);
1831 #if IS_ENABLED(CONFIG_NFS_V4)
1832 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1834 const struct dentry_operations nfs4_dentry_operations = {
1835 .d_revalidate = nfs4_lookup_revalidate,
1836 .d_weak_revalidate = nfs_weak_revalidate,
1837 .d_delete = nfs_dentry_delete,
1838 .d_iput = nfs_dentry_iput,
1839 .d_automount = nfs_d_automount,
1840 .d_release = nfs_d_release,
1842 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1844 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1846 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1849 static int do_open(struct inode *inode, struct file *filp)
1851 nfs_fscache_open_file(inode, filp);
1855 static int nfs_finish_open(struct nfs_open_context *ctx,
1856 struct dentry *dentry,
1857 struct file *file, unsigned open_flags)
1861 err = finish_open(file, dentry, do_open);
1864 if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1865 nfs_file_set_open_context(file, ctx);
1872 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1873 struct file *file, unsigned open_flags,
1876 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1877 struct nfs_open_context *ctx;
1879 struct iattr attr = { .ia_valid = ATTR_OPEN };
1880 struct inode *inode;
1881 unsigned int lookup_flags = 0;
1882 bool switched = false;
1886 /* Expect a negative dentry */
1887 BUG_ON(d_inode(dentry));
1889 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1890 dir->i_sb->s_id, dir->i_ino, dentry);
1892 err = nfs_check_flags(open_flags);
1896 /* NFS only supports OPEN on regular files */
1897 if ((open_flags & O_DIRECTORY)) {
1898 if (!d_in_lookup(dentry)) {
1900 * Hashed negative dentry with O_DIRECTORY: dentry was
1901 * revalidated and is fine, no need to perform lookup
1906 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1910 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1911 return -ENAMETOOLONG;
1913 if (open_flags & O_CREAT) {
1914 struct nfs_server *server = NFS_SERVER(dir);
1916 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1917 mode &= ~current_umask();
1919 attr.ia_valid |= ATTR_MODE;
1920 attr.ia_mode = mode;
1922 if (open_flags & O_TRUNC) {
1923 attr.ia_valid |= ATTR_SIZE;
1927 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1930 dentry = d_alloc_parallel(dentry->d_parent,
1931 &dentry->d_name, &wq);
1933 return PTR_ERR(dentry);
1934 if (unlikely(!d_in_lookup(dentry)))
1935 return finish_no_open(file, dentry);
1938 ctx = create_nfs_open_context(dentry, open_flags, file);
1943 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1944 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1946 file->f_mode |= FMODE_CREATED;
1947 if (IS_ERR(inode)) {
1948 err = PTR_ERR(inode);
1949 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1950 put_nfs_open_context(ctx);
1954 d_splice_alias(NULL, dentry);
1955 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1961 if (!(open_flags & O_NOFOLLOW))
1971 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1972 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1973 put_nfs_open_context(ctx);
1975 if (unlikely(switched)) {
1976 d_lookup_done(dentry);
1982 res = nfs_lookup(dir, dentry, lookup_flags);
1984 inode = d_inode(dentry);
1985 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
1986 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
1987 res = ERR_PTR(-ENOTDIR);
1988 else if (inode && S_ISREG(inode->i_mode))
1989 res = ERR_PTR(-EOPENSTALE);
1990 } else if (!IS_ERR(res)) {
1991 inode = d_inode(res);
1992 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
1993 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
1995 res = ERR_PTR(-ENOTDIR);
1996 } else if (inode && S_ISREG(inode->i_mode)) {
1998 res = ERR_PTR(-EOPENSTALE);
2002 d_lookup_done(dentry);
2009 return PTR_ERR(res);
2010 return finish_no_open(file, res);
2012 EXPORT_SYMBOL_GPL(nfs_atomic_open);
2015 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2018 struct inode *inode;
2020 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2022 if (d_mountpoint(dentry))
2025 inode = d_inode(dentry);
2027 /* We can't create new files in nfs_open_revalidate(), so we
2028 * optimize away revalidation of negative dentries.
2033 if (nfs_verifier_is_delegated(dentry))
2034 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2036 /* NFS only supports OPEN on regular files */
2037 if (!S_ISREG(inode->i_mode))
2040 /* We cannot do exclusive creation on a positive dentry */
2041 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2044 /* Check if the directory changed */
2045 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2048 /* Let f_op->open() actually open (and revalidate) the file */
2051 if (flags & LOOKUP_RCU)
2053 return nfs_lookup_revalidate_dentry(dir, dentry, inode);
2056 return nfs_do_lookup_revalidate(dir, dentry, flags);
2059 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2061 return __nfs_lookup_revalidate(dentry, flags,
2062 nfs4_do_lookup_revalidate);
2065 #endif /* CONFIG_NFSV4 */
2068 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2069 struct nfs_fattr *fattr,
2070 struct nfs4_label *label)
2072 struct dentry *parent = dget_parent(dentry);
2073 struct inode *dir = d_inode(parent);
2074 struct inode *inode;
2080 if (fhandle->size == 0) {
2081 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL);
2085 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2086 if (!(fattr->valid & NFS_ATTR_FATTR)) {
2087 struct nfs_server *server = NFS_SB(dentry->d_sb);
2088 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2093 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
2094 d = d_splice_alias(inode, dentry);
2102 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2105 * Code common to create, mkdir, and mknod.
2107 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2108 struct nfs_fattr *fattr,
2109 struct nfs4_label *label)
2113 d = nfs_add_or_obtain(dentry, fhandle, fattr, label);
2117 /* Callers don't care */
2121 EXPORT_SYMBOL_GPL(nfs_instantiate);
2124 * Following a failed create operation, we drop the dentry rather
2125 * than retain a negative dentry. This avoids a problem in the event
2126 * that the operation succeeded on the server, but an error in the
2127 * reply path made it appear to have failed.
2129 int nfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2130 struct dentry *dentry, umode_t mode, bool excl)
2133 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2136 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2137 dir->i_sb->s_id, dir->i_ino, dentry);
2139 attr.ia_mode = mode;
2140 attr.ia_valid = ATTR_MODE;
2142 trace_nfs_create_enter(dir, dentry, open_flags);
2143 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2144 trace_nfs_create_exit(dir, dentry, open_flags, error);
2152 EXPORT_SYMBOL_GPL(nfs_create);
2155 * See comments for nfs_proc_create regarding failed operations.
2158 nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2159 struct dentry *dentry, umode_t mode, dev_t rdev)
2164 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2165 dir->i_sb->s_id, dir->i_ino, dentry);
2167 attr.ia_mode = mode;
2168 attr.ia_valid = ATTR_MODE;
2170 trace_nfs_mknod_enter(dir, dentry);
2171 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2172 trace_nfs_mknod_exit(dir, dentry, status);
2180 EXPORT_SYMBOL_GPL(nfs_mknod);
2183 * See comments for nfs_proc_create regarding failed operations.
2185 int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2186 struct dentry *dentry, umode_t mode)
2191 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2192 dir->i_sb->s_id, dir->i_ino, dentry);
2194 attr.ia_valid = ATTR_MODE;
2195 attr.ia_mode = mode | S_IFDIR;
2197 trace_nfs_mkdir_enter(dir, dentry);
2198 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2199 trace_nfs_mkdir_exit(dir, dentry, error);
2207 EXPORT_SYMBOL_GPL(nfs_mkdir);
2209 static void nfs_dentry_handle_enoent(struct dentry *dentry)
2211 if (simple_positive(dentry))
2215 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2219 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2220 dir->i_sb->s_id, dir->i_ino, dentry);
2222 trace_nfs_rmdir_enter(dir, dentry);
2223 if (d_really_is_positive(dentry)) {
2224 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2225 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2226 /* Ensure the VFS deletes this inode */
2229 clear_nlink(d_inode(dentry));
2232 nfs_dentry_handle_enoent(dentry);
2234 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2236 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2237 trace_nfs_rmdir_exit(dir, dentry, error);
2241 EXPORT_SYMBOL_GPL(nfs_rmdir);
2244 * Remove a file after making sure there are no pending writes,
2245 * and after checking that the file has only one user.
2247 * We invalidate the attribute cache and free the inode prior to the operation
2248 * to avoid possible races if the server reuses the inode.
2250 static int nfs_safe_remove(struct dentry *dentry)
2252 struct inode *dir = d_inode(dentry->d_parent);
2253 struct inode *inode = d_inode(dentry);
2256 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2258 /* If the dentry was sillyrenamed, we simply call d_delete() */
2259 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2264 trace_nfs_remove_enter(dir, dentry);
2265 if (inode != NULL) {
2266 error = NFS_PROTO(dir)->remove(dir, dentry);
2268 nfs_drop_nlink(inode);
2270 error = NFS_PROTO(dir)->remove(dir, dentry);
2271 if (error == -ENOENT)
2272 nfs_dentry_handle_enoent(dentry);
2273 trace_nfs_remove_exit(dir, dentry, error);
2278 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2279 * belongs to an active ".nfs..." file and we return -EBUSY.
2281 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2283 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2286 int need_rehash = 0;
2288 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2289 dir->i_ino, dentry);
2291 trace_nfs_unlink_enter(dir, dentry);
2292 spin_lock(&dentry->d_lock);
2293 if (d_count(dentry) > 1) {
2294 spin_unlock(&dentry->d_lock);
2295 /* Start asynchronous writeout of the inode */
2296 write_inode_now(d_inode(dentry), 0);
2297 error = nfs_sillyrename(dir, dentry);
2300 if (!d_unhashed(dentry)) {
2304 spin_unlock(&dentry->d_lock);
2305 error = nfs_safe_remove(dentry);
2306 if (!error || error == -ENOENT) {
2307 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2308 } else if (need_rehash)
2311 trace_nfs_unlink_exit(dir, dentry, error);
2314 EXPORT_SYMBOL_GPL(nfs_unlink);
2317 * To create a symbolic link, most file systems instantiate a new inode,
2318 * add a page to it containing the path, then write it out to the disk
2319 * using prepare_write/commit_write.
2321 * Unfortunately the NFS client can't create the in-core inode first
2322 * because it needs a file handle to create an in-core inode (see
2323 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2324 * symlink request has completed on the server.
2326 * So instead we allocate a raw page, copy the symname into it, then do
2327 * the SYMLINK request with the page as the buffer. If it succeeds, we
2328 * now have a new file handle and can instantiate an in-core NFS inode
2329 * and move the raw page into its mapping.
2331 int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
2332 struct dentry *dentry, const char *symname)
2337 unsigned int pathlen = strlen(symname);
2340 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2341 dir->i_ino, dentry, symname);
2343 if (pathlen > PAGE_SIZE)
2344 return -ENAMETOOLONG;
2346 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2347 attr.ia_valid = ATTR_MODE;
2349 page = alloc_page(GFP_USER);
2353 kaddr = page_address(page);
2354 memcpy(kaddr, symname, pathlen);
2355 if (pathlen < PAGE_SIZE)
2356 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2358 trace_nfs_symlink_enter(dir, dentry);
2359 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2360 trace_nfs_symlink_exit(dir, dentry, error);
2362 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2363 dir->i_sb->s_id, dir->i_ino,
2364 dentry, symname, error);
2371 * No big deal if we can't add this page to the page cache here.
2372 * READLINK will get the missing page from the server if needed.
2374 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2376 SetPageUptodate(page);
2379 * add_to_page_cache_lru() grabs an extra page refcount.
2380 * Drop it here to avoid leaking this page later.
2388 EXPORT_SYMBOL_GPL(nfs_symlink);
2391 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2393 struct inode *inode = d_inode(old_dentry);
2396 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2397 old_dentry, dentry);
2399 trace_nfs_link_enter(inode, dir, dentry);
2401 if (S_ISREG(inode->i_mode))
2402 nfs_sync_inode(inode);
2403 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2406 d_add(dentry, inode);
2408 trace_nfs_link_exit(inode, dir, dentry, error);
2411 EXPORT_SYMBOL_GPL(nfs_link);
2415 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2416 * different file handle for the same inode after a rename (e.g. when
2417 * moving to a different directory). A fail-safe method to do so would
2418 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2419 * rename the old file using the sillyrename stuff. This way, the original
2420 * file in old_dir will go away when the last process iput()s the inode.
2424 * It actually works quite well. One needs to have the possibility for
2425 * at least one ".nfs..." file in each directory the file ever gets
2426 * moved or linked to which happens automagically with the new
2427 * implementation that only depends on the dcache stuff instead of
2428 * using the inode layer
2430 * Unfortunately, things are a little more complicated than indicated
2431 * above. For a cross-directory move, we want to make sure we can get
2432 * rid of the old inode after the operation. This means there must be
2433 * no pending writes (if it's a file), and the use count must be 1.
2434 * If these conditions are met, we can drop the dentries before doing
2437 int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
2438 struct dentry *old_dentry, struct inode *new_dir,
2439 struct dentry *new_dentry, unsigned int flags)
2441 struct inode *old_inode = d_inode(old_dentry);
2442 struct inode *new_inode = d_inode(new_dentry);
2443 struct dentry *dentry = NULL, *rehash = NULL;
2444 struct rpc_task *task;
2450 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2451 old_dentry, new_dentry,
2452 d_count(new_dentry));
2454 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2456 * For non-directories, check whether the target is busy and if so,
2457 * make a copy of the dentry and then do a silly-rename. If the
2458 * silly-rename succeeds, the copied dentry is hashed and becomes
2461 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2463 * To prevent any new references to the target during the
2464 * rename, we unhash the dentry in advance.
2466 if (!d_unhashed(new_dentry)) {
2468 rehash = new_dentry;
2471 if (d_count(new_dentry) > 2) {
2474 /* copy the target dentry's name */
2475 dentry = d_alloc(new_dentry->d_parent,
2476 &new_dentry->d_name);
2480 /* silly-rename the existing target ... */
2481 err = nfs_sillyrename(new_dir, new_dentry);
2485 new_dentry = dentry;
2491 if (S_ISREG(old_inode->i_mode))
2492 nfs_sync_inode(old_inode);
2493 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2495 error = PTR_ERR(task);
2499 error = rpc_wait_for_completion_task(task);
2501 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2502 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2505 error = task->tk_status;
2507 /* Ensure the inode attributes are revalidated */
2509 spin_lock(&old_inode->i_lock);
2510 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2511 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2512 NFS_INO_INVALID_CTIME |
2513 NFS_INO_REVAL_FORCED);
2514 spin_unlock(&old_inode->i_lock);
2519 trace_nfs_rename_exit(old_dir, old_dentry,
2520 new_dir, new_dentry, error);
2522 if (new_inode != NULL)
2523 nfs_drop_nlink(new_inode);
2525 * The d_move() should be here instead of in an async RPC completion
2526 * handler because we need the proper locks to move the dentry. If
2527 * we're interrupted by a signal, the async RPC completion handler
2528 * should mark the directories for revalidation.
2530 d_move(old_dentry, new_dentry);
2531 nfs_set_verifier(old_dentry,
2532 nfs_save_change_attribute(new_dir));
2533 } else if (error == -ENOENT)
2534 nfs_dentry_handle_enoent(old_dentry);
2536 /* new dentry created? */
2541 EXPORT_SYMBOL_GPL(nfs_rename);
2543 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2544 static LIST_HEAD(nfs_access_lru_list);
2545 static atomic_long_t nfs_access_nr_entries;
2547 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2548 module_param(nfs_access_max_cachesize, ulong, 0644);
2549 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2551 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2553 put_cred(entry->cred);
2554 kfree_rcu(entry, rcu_head);
2555 smp_mb__before_atomic();
2556 atomic_long_dec(&nfs_access_nr_entries);
2557 smp_mb__after_atomic();
2560 static void nfs_access_free_list(struct list_head *head)
2562 struct nfs_access_entry *cache;
2564 while (!list_empty(head)) {
2565 cache = list_entry(head->next, struct nfs_access_entry, lru);
2566 list_del(&cache->lru);
2567 nfs_access_free_entry(cache);
2571 static unsigned long
2572 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2575 struct nfs_inode *nfsi, *next;
2576 struct nfs_access_entry *cache;
2579 spin_lock(&nfs_access_lru_lock);
2580 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2581 struct inode *inode;
2583 if (nr_to_scan-- == 0)
2585 inode = &nfsi->vfs_inode;
2586 spin_lock(&inode->i_lock);
2587 if (list_empty(&nfsi->access_cache_entry_lru))
2588 goto remove_lru_entry;
2589 cache = list_entry(nfsi->access_cache_entry_lru.next,
2590 struct nfs_access_entry, lru);
2591 list_move(&cache->lru, &head);
2592 rb_erase(&cache->rb_node, &nfsi->access_cache);
2594 if (!list_empty(&nfsi->access_cache_entry_lru))
2595 list_move_tail(&nfsi->access_cache_inode_lru,
2596 &nfs_access_lru_list);
2599 list_del_init(&nfsi->access_cache_inode_lru);
2600 smp_mb__before_atomic();
2601 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2602 smp_mb__after_atomic();
2604 spin_unlock(&inode->i_lock);
2606 spin_unlock(&nfs_access_lru_lock);
2607 nfs_access_free_list(&head);
2612 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2614 int nr_to_scan = sc->nr_to_scan;
2615 gfp_t gfp_mask = sc->gfp_mask;
2617 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2619 return nfs_do_access_cache_scan(nr_to_scan);
2624 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2626 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2630 nfs_access_cache_enforce_limit(void)
2632 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2634 unsigned int nr_to_scan;
2636 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2639 diff = nr_entries - nfs_access_max_cachesize;
2640 if (diff < nr_to_scan)
2642 nfs_do_access_cache_scan(nr_to_scan);
2645 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2647 struct rb_root *root_node = &nfsi->access_cache;
2649 struct nfs_access_entry *entry;
2651 /* Unhook entries from the cache */
2652 while ((n = rb_first(root_node)) != NULL) {
2653 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2654 rb_erase(n, root_node);
2655 list_move(&entry->lru, head);
2657 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2660 void nfs_access_zap_cache(struct inode *inode)
2664 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2666 /* Remove from global LRU init */
2667 spin_lock(&nfs_access_lru_lock);
2668 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2669 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2671 spin_lock(&inode->i_lock);
2672 __nfs_access_zap_cache(NFS_I(inode), &head);
2673 spin_unlock(&inode->i_lock);
2674 spin_unlock(&nfs_access_lru_lock);
2675 nfs_access_free_list(&head);
2677 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2679 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2681 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2684 struct nfs_access_entry *entry =
2685 rb_entry(n, struct nfs_access_entry, rb_node);
2686 int cmp = cred_fscmp(cred, entry->cred);
2698 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
2700 struct nfs_inode *nfsi = NFS_I(inode);
2701 struct nfs_access_entry *cache;
2705 spin_lock(&inode->i_lock);
2707 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2709 cache = nfs_access_search_rbtree(inode, cred);
2713 /* Found an entry, is our attribute cache valid? */
2714 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2721 spin_unlock(&inode->i_lock);
2722 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2725 spin_lock(&inode->i_lock);
2728 *mask = cache->mask;
2729 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2732 spin_unlock(&inode->i_lock);
2735 spin_unlock(&inode->i_lock);
2736 nfs_access_zap_cache(inode);
2740 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
2742 /* Only check the most recently returned cache entry,
2743 * but do it without locking.
2745 struct nfs_inode *nfsi = NFS_I(inode);
2746 struct nfs_access_entry *cache;
2748 struct list_head *lh;
2751 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2753 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
2754 cache = list_entry(lh, struct nfs_access_entry, lru);
2755 if (lh == &nfsi->access_cache_entry_lru ||
2756 cred_fscmp(cred, cache->cred) != 0)
2760 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2762 *mask = cache->mask;
2769 int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
2770 u32 *mask, bool may_block)
2774 status = nfs_access_get_cached_rcu(inode, cred, mask);
2776 status = nfs_access_get_cached_locked(inode, cred, mask,
2781 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
2783 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2785 struct nfs_inode *nfsi = NFS_I(inode);
2786 struct rb_root *root_node = &nfsi->access_cache;
2787 struct rb_node **p = &root_node->rb_node;
2788 struct rb_node *parent = NULL;
2789 struct nfs_access_entry *entry;
2792 spin_lock(&inode->i_lock);
2793 while (*p != NULL) {
2795 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2796 cmp = cred_fscmp(set->cred, entry->cred);
2799 p = &parent->rb_left;
2801 p = &parent->rb_right;
2805 rb_link_node(&set->rb_node, parent, p);
2806 rb_insert_color(&set->rb_node, root_node);
2807 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2808 spin_unlock(&inode->i_lock);
2811 rb_replace_node(parent, &set->rb_node, root_node);
2812 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2813 list_del(&entry->lru);
2814 spin_unlock(&inode->i_lock);
2815 nfs_access_free_entry(entry);
2818 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2820 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2823 RB_CLEAR_NODE(&cache->rb_node);
2824 cache->cred = get_cred(set->cred);
2825 cache->mask = set->mask;
2827 /* The above field assignments must be visible
2828 * before this item appears on the lru. We cannot easily
2829 * use rcu_assign_pointer, so just force the memory barrier.
2832 nfs_access_add_rbtree(inode, cache);
2834 /* Update accounting */
2835 smp_mb__before_atomic();
2836 atomic_long_inc(&nfs_access_nr_entries);
2837 smp_mb__after_atomic();
2839 /* Add inode to global LRU list */
2840 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2841 spin_lock(&nfs_access_lru_lock);
2842 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2843 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2844 &nfs_access_lru_list);
2845 spin_unlock(&nfs_access_lru_lock);
2847 nfs_access_cache_enforce_limit();
2849 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2851 #define NFS_MAY_READ (NFS_ACCESS_READ)
2852 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2853 NFS_ACCESS_EXTEND | \
2855 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2857 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2858 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2859 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2861 nfs_access_calc_mask(u32 access_result, umode_t umode)
2865 if (access_result & NFS_MAY_READ)
2867 if (S_ISDIR(umode)) {
2868 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2870 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2872 } else if (S_ISREG(umode)) {
2873 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2875 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2877 } else if (access_result & NFS_MAY_WRITE)
2882 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2884 entry->mask = access_result;
2886 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2888 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
2890 struct nfs_access_entry cache;
2891 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2892 int cache_mask = -1;
2895 trace_nfs_access_enter(inode);
2897 status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
2906 * Determine which access bits we want to ask for...
2908 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2909 if (nfs_server_capable(inode, NFS_CAP_XATTR)) {
2910 cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE |
2913 if (S_ISDIR(inode->i_mode))
2914 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2916 cache.mask |= NFS_ACCESS_EXECUTE;
2918 status = NFS_PROTO(inode)->access(inode, &cache);
2920 if (status == -ESTALE) {
2921 if (!S_ISDIR(inode->i_mode))
2922 nfs_set_inode_stale(inode);
2924 nfs_zap_caches(inode);
2928 nfs_access_add_cache(inode, &cache);
2930 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2931 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2934 trace_nfs_access_exit(inode, mask, cache_mask, status);
2938 static int nfs_open_permission_mask(int openflags)
2942 if (openflags & __FMODE_EXEC) {
2943 /* ONLY check exec rights */
2946 if ((openflags & O_ACCMODE) != O_WRONLY)
2948 if ((openflags & O_ACCMODE) != O_RDONLY)
2955 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
2957 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2959 EXPORT_SYMBOL_GPL(nfs_may_open);
2961 static int nfs_execute_ok(struct inode *inode, int mask)
2963 struct nfs_server *server = NFS_SERVER(inode);
2966 if (S_ISDIR(inode->i_mode))
2968 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
2969 if (mask & MAY_NOT_BLOCK)
2971 ret = __nfs_revalidate_inode(server, inode);
2973 if (ret == 0 && !execute_ok(inode))
2978 int nfs_permission(struct user_namespace *mnt_userns,
2979 struct inode *inode,
2982 const struct cred *cred = current_cred();
2985 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2987 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2989 /* Is this sys_access() ? */
2990 if (mask & (MAY_ACCESS | MAY_CHDIR))
2993 switch (inode->i_mode & S_IFMT) {
2997 if ((mask & MAY_OPEN) &&
2998 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3003 * Optimize away all write operations, since the server
3004 * will check permissions when we perform the op.
3006 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3011 if (!NFS_PROTO(inode)->access)
3014 res = nfs_do_access(inode, cred, mask);
3016 if (!res && (mask & MAY_EXEC))
3017 res = nfs_execute_ok(inode, mask);
3019 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3020 inode->i_sb->s_id, inode->i_ino, mask, res);
3023 if (mask & MAY_NOT_BLOCK)
3026 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3027 NFS_INO_INVALID_OTHER);
3029 res = generic_permission(&init_user_ns, inode, mask);
3032 EXPORT_SYMBOL_GPL(nfs_permission);