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;
81 spin_lock(&dir->i_lock);
82 if (list_empty(&nfsi->open_files) &&
83 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
84 nfsi->cache_validity |= NFS_INO_INVALID_DATA |
86 list_add(&ctx->list, &nfsi->open_files);
87 spin_unlock(&dir->i_lock);
90 return ERR_PTR(-ENOMEM);
93 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
95 spin_lock(&dir->i_lock);
97 spin_unlock(&dir->i_lock);
105 nfs_opendir(struct inode *inode, struct file *filp)
108 struct nfs_open_dir_context *ctx;
110 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
112 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
114 ctx = alloc_nfs_open_dir_context(inode);
119 filp->private_data = ctx;
125 nfs_closedir(struct inode *inode, struct file *filp)
127 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
131 struct nfs_cache_array_entry {
135 unsigned int name_len;
136 unsigned char d_type;
139 struct nfs_cache_array {
142 unsigned char page_full : 1,
144 cookies_are_ordered : 1;
145 struct nfs_cache_array_entry array[];
148 struct nfs_readdir_descriptor {
151 struct dir_context *ctx;
156 loff_t current_index;
159 __be32 verf[NFS_DIR_VERIFIER_SIZE];
160 unsigned long dir_verifier;
161 unsigned long timestamp;
162 unsigned long gencount;
163 unsigned long attr_gencount;
164 unsigned int cache_entry_index;
170 static void nfs_readdir_array_init(struct nfs_cache_array *array)
172 memset(array, 0, sizeof(struct nfs_cache_array));
175 static void nfs_readdir_page_init_array(struct page *page, u64 last_cookie)
177 struct nfs_cache_array *array;
179 array = kmap_atomic(page);
180 nfs_readdir_array_init(array);
181 array->last_cookie = last_cookie;
182 array->cookies_are_ordered = 1;
183 kunmap_atomic(array);
187 * we are freeing strings created by nfs_add_to_readdir_array()
190 void nfs_readdir_clear_array(struct page *page)
192 struct nfs_cache_array *array;
195 array = kmap_atomic(page);
196 for (i = 0; i < array->size; i++)
197 kfree(array->array[i].name);
198 nfs_readdir_array_init(array);
199 kunmap_atomic(array);
203 nfs_readdir_page_array_alloc(u64 last_cookie, gfp_t gfp_flags)
205 struct page *page = alloc_page(gfp_flags);
207 nfs_readdir_page_init_array(page, last_cookie);
211 static void nfs_readdir_page_array_free(struct page *page)
214 nfs_readdir_clear_array(page);
219 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
221 array->page_is_eof = 1;
222 array->page_full = 1;
225 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
227 return array->page_full;
231 * the caller is responsible for freeing qstr.name
232 * when called by nfs_readdir_add_to_array, the strings will be freed in
233 * nfs_clear_readdir_array()
235 static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
237 const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
240 * Avoid a kmemleak false positive. The pointer to the name is stored
241 * in a page cache page which kmemleak does not scan.
244 kmemleak_not_leak(ret);
249 * Check that the next array entry lies entirely within the page bounds
251 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
253 struct nfs_cache_array_entry *cache_entry;
255 if (array->page_full)
257 cache_entry = &array->array[array->size + 1];
258 if ((char *)cache_entry - (char *)array > PAGE_SIZE) {
259 array->page_full = 1;
266 int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
268 struct nfs_cache_array *array;
269 struct nfs_cache_array_entry *cache_entry;
273 name = nfs_readdir_copy_name(entry->name, entry->len);
277 array = kmap_atomic(page);
278 ret = nfs_readdir_array_can_expand(array);
284 cache_entry = &array->array[array->size];
285 cache_entry->cookie = entry->prev_cookie;
286 cache_entry->ino = entry->ino;
287 cache_entry->d_type = entry->d_type;
288 cache_entry->name_len = entry->len;
289 cache_entry->name = name;
290 array->last_cookie = entry->cookie;
291 if (array->last_cookie <= cache_entry->cookie)
292 array->cookies_are_ordered = 0;
295 nfs_readdir_array_set_eof(array);
297 kunmap_atomic(array);
301 static struct page *nfs_readdir_page_get_locked(struct address_space *mapping,
302 pgoff_t index, u64 last_cookie)
306 page = grab_cache_page(mapping, index);
307 if (page && !PageUptodate(page)) {
308 nfs_readdir_page_init_array(page, last_cookie);
309 if (invalidate_inode_pages2_range(mapping, index + 1, -1) < 0)
310 nfs_zap_mapping(mapping->host, mapping);
311 SetPageUptodate(page);
317 static u64 nfs_readdir_page_last_cookie(struct page *page)
319 struct nfs_cache_array *array;
322 array = kmap_atomic(page);
323 ret = array->last_cookie;
324 kunmap_atomic(array);
328 static bool nfs_readdir_page_needs_filling(struct page *page)
330 struct nfs_cache_array *array;
333 array = kmap_atomic(page);
334 ret = !nfs_readdir_array_is_full(array);
335 kunmap_atomic(array);
339 static void nfs_readdir_page_set_eof(struct page *page)
341 struct nfs_cache_array *array;
343 array = kmap_atomic(page);
344 nfs_readdir_array_set_eof(array);
345 kunmap_atomic(array);
348 static void nfs_readdir_page_unlock_and_put(struct page *page)
354 static struct page *nfs_readdir_page_get_next(struct address_space *mapping,
355 pgoff_t index, u64 cookie)
359 page = nfs_readdir_page_get_locked(mapping, index, cookie);
361 if (nfs_readdir_page_last_cookie(page) == cookie)
363 nfs_readdir_page_unlock_and_put(page);
369 int is_32bit_api(void)
372 return in_compat_syscall();
374 return (BITS_PER_LONG == 32);
379 bool nfs_readdir_use_cookie(const struct file *filp)
381 if ((filp->f_mode & FMODE_32BITHASH) ||
382 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
387 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
388 struct nfs_readdir_descriptor *desc)
390 loff_t diff = desc->ctx->pos - desc->current_index;
395 if (diff >= array->size) {
396 if (array->page_is_eof)
401 index = (unsigned int)diff;
402 desc->dir_cookie = array->array[index].cookie;
403 desc->cache_entry_index = index;
411 nfs_readdir_inode_mapping_valid(struct nfs_inode *nfsi)
413 if (nfsi->cache_validity & (NFS_INO_INVALID_ATTR|NFS_INO_INVALID_DATA))
416 return !test_bit(NFS_INO_INVALIDATING, &nfsi->flags);
419 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
422 if (!array->cookies_are_ordered)
424 /* Optimisation for monotonically increasing cookies */
425 if (cookie >= array->last_cookie)
427 if (array->size && cookie < array->array[0].cookie)
432 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
433 struct nfs_readdir_descriptor *desc)
437 int status = -EAGAIN;
439 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
442 for (i = 0; i < array->size; i++) {
443 if (array->array[i].cookie == desc->dir_cookie) {
444 struct nfs_inode *nfsi = NFS_I(file_inode(desc->file));
446 new_pos = desc->current_index + i;
447 if (desc->attr_gencount != nfsi->attr_gencount ||
448 !nfs_readdir_inode_mapping_valid(nfsi)) {
450 desc->attr_gencount = nfsi->attr_gencount;
451 } else if (new_pos < desc->prev_index) {
453 && desc->dup_cookie == desc->dir_cookie) {
454 if (printk_ratelimit()) {
455 pr_notice("NFS: directory %pD2 contains a readdir loop."
456 "Please contact your server vendor. "
457 "The file: %s has duplicate cookie %llu\n",
458 desc->file, array->array[i].name, desc->dir_cookie);
463 desc->dup_cookie = desc->dir_cookie;
466 if (nfs_readdir_use_cookie(desc->file))
467 desc->ctx->pos = desc->dir_cookie;
469 desc->ctx->pos = new_pos;
470 desc->prev_index = new_pos;
471 desc->cache_entry_index = i;
476 if (array->page_is_eof) {
477 status = -EBADCOOKIE;
478 if (desc->dir_cookie == array->last_cookie)
485 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
487 struct nfs_cache_array *array;
490 array = kmap_atomic(desc->page);
492 if (desc->dir_cookie == 0)
493 status = nfs_readdir_search_for_pos(array, desc);
495 status = nfs_readdir_search_for_cookie(array, desc);
497 if (status == -EAGAIN) {
498 desc->last_cookie = array->last_cookie;
499 desc->current_index += array->size;
502 kunmap_atomic(array);
506 /* Fill a page with xdr information before transferring to the cache page */
507 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
508 __be32 *verf, u64 cookie,
509 struct page **pages, size_t bufsize,
512 struct inode *inode = file_inode(desc->file);
513 struct nfs_readdir_arg arg = {
514 .dentry = file_dentry(desc->file),
515 .cred = desc->file->f_cred,
522 struct nfs_readdir_res res = {
525 unsigned long timestamp, gencount;
530 gencount = nfs_inc_attr_generation_counter();
531 desc->dir_verifier = nfs_save_change_attribute(inode);
532 error = NFS_PROTO(inode)->readdir(&arg, &res);
534 /* We requested READDIRPLUS, but the server doesn't grok it */
535 if (error == -ENOTSUPP && desc->plus) {
536 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
537 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
538 desc->plus = arg.plus = false;
543 desc->timestamp = timestamp;
544 desc->gencount = gencount;
549 static int xdr_decode(struct nfs_readdir_descriptor *desc,
550 struct nfs_entry *entry, struct xdr_stream *xdr)
552 struct inode *inode = file_inode(desc->file);
555 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
558 entry->fattr->time_start = desc->timestamp;
559 entry->fattr->gencount = desc->gencount;
563 /* Match file and dirent using either filehandle or fileid
564 * Note: caller is responsible for checking the fsid
567 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
570 struct nfs_inode *nfsi;
572 if (d_really_is_negative(dentry))
575 inode = d_inode(dentry);
576 if (is_bad_inode(inode) || NFS_STALE(inode))
580 if (entry->fattr->fileid != nfsi->fileid)
582 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
588 bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx)
590 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
592 if (test_and_clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(dir)->flags))
600 * This function is called by the lookup and getattr code to request the
601 * use of readdirplus to accelerate any future lookups in the same
604 void nfs_advise_use_readdirplus(struct inode *dir)
606 struct nfs_inode *nfsi = NFS_I(dir);
608 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
609 !list_empty(&nfsi->open_files))
610 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
614 * This function is mainly for use by nfs_getattr().
616 * If this is an 'ls -l', we want to force use of readdirplus.
617 * Do this by checking if there is an active file descriptor
618 * and calling nfs_advise_use_readdirplus, then forcing a
621 void nfs_force_use_readdirplus(struct inode *dir)
623 struct nfs_inode *nfsi = NFS_I(dir);
625 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
626 !list_empty(&nfsi->open_files)) {
627 set_bit(NFS_INO_ADVISE_RDPLUS, &nfsi->flags);
628 invalidate_mapping_pages(dir->i_mapping,
629 nfsi->page_index + 1, -1);
634 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
635 unsigned long dir_verifier)
637 struct qstr filename = QSTR_INIT(entry->name, entry->len);
638 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
639 struct dentry *dentry;
640 struct dentry *alias;
644 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
646 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
648 if (filename.len == 0)
650 /* Validate that the name doesn't contain any illegal '\0' */
651 if (strnlen(filename.name, filename.len) != filename.len)
654 if (strnchr(filename.name, filename.len, '/'))
656 if (filename.name[0] == '.') {
657 if (filename.len == 1)
659 if (filename.len == 2 && filename.name[1] == '.')
662 filename.hash = full_name_hash(parent, filename.name, filename.len);
664 dentry = d_lookup(parent, &filename);
667 dentry = d_alloc_parallel(parent, &filename, &wq);
671 if (!d_in_lookup(dentry)) {
672 /* Is there a mountpoint here? If so, just exit */
673 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
674 &entry->fattr->fsid))
676 if (nfs_same_file(dentry, entry)) {
677 if (!entry->fh->size)
679 nfs_set_verifier(dentry, dir_verifier);
680 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
682 nfs_setsecurity(d_inode(dentry), entry->fattr, entry->label);
685 d_invalidate(dentry);
691 if (!entry->fh->size) {
692 d_lookup_done(dentry);
696 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr, entry->label);
697 alias = d_splice_alias(inode, dentry);
698 d_lookup_done(dentry);
705 nfs_set_verifier(dentry, dir_verifier);
710 /* Perform conversion from xdr to cache array */
711 static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc,
712 struct nfs_entry *entry,
713 struct page **xdr_pages,
715 struct page **arrays,
718 struct address_space *mapping = desc->file->f_mapping;
719 struct xdr_stream stream;
721 struct page *scratch, *new, *page = *arrays;
724 scratch = alloc_page(GFP_KERNEL);
728 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
729 xdr_set_scratch_page(&stream, scratch);
733 entry->label->len = NFS4_MAXLABELLEN;
735 status = xdr_decode(desc, entry, &stream);
740 nfs_prime_dcache(file_dentry(desc->file), entry,
743 status = nfs_readdir_add_to_array(entry, page);
744 if (status != -ENOSPC)
747 if (page->mapping != mapping) {
750 new = nfs_readdir_page_array_alloc(entry->prev_cookie,
755 *arrays = page = new;
757 new = nfs_readdir_page_get_next(mapping,
763 nfs_readdir_page_unlock_and_put(page);
766 status = nfs_readdir_add_to_array(entry, page);
767 } while (!status && !entry->eof);
772 nfs_readdir_page_set_eof(page);
783 nfs_readdir_page_unlock_and_put(page);
789 static void nfs_readdir_free_pages(struct page **pages, size_t npages)
792 put_page(pages[npages]);
797 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
798 * to nfs_readdir_free_pages()
800 static struct page **nfs_readdir_alloc_pages(size_t npages)
805 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
808 for (i = 0; i < npages; i++) {
809 struct page *page = alloc_page(GFP_KERNEL);
817 nfs_readdir_free_pages(pages, i);
821 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
822 __be32 *verf_arg, __be32 *verf_res,
823 struct page **arrays, size_t narrays)
826 struct page *page = *arrays;
827 struct nfs_entry *entry;
829 struct inode *inode = file_inode(desc->file);
830 size_t dtsize = NFS_SERVER(inode)->dtsize;
831 int status = -ENOMEM;
833 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
836 entry->cookie = nfs_readdir_page_last_cookie(page);
837 entry->fh = nfs_alloc_fhandle();
838 entry->fattr = nfs_alloc_fattr();
839 entry->server = NFS_SERVER(inode);
840 if (entry->fh == NULL || entry->fattr == NULL)
843 entry->label = nfs4_label_alloc(NFS_SERVER(inode), GFP_NOWAIT);
844 if (IS_ERR(entry->label)) {
845 status = PTR_ERR(entry->label);
849 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
850 pages = nfs_readdir_alloc_pages(array_size);
852 goto out_release_label;
856 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie,
864 nfs_readdir_page_set_eof(page);
868 status = nfs_readdir_page_filler(desc, entry, pages, pglen,
870 } while (!status && nfs_readdir_page_needs_filling(page));
872 nfs_readdir_free_pages(pages, array_size);
874 nfs4_label_free(entry->label);
876 nfs_free_fattr(entry->fattr);
877 nfs_free_fhandle(entry->fh);
882 static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc)
884 put_page(desc->page);
889 nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
891 unlock_page(desc->page);
892 nfs_readdir_page_put(desc);
896 nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc)
898 return nfs_readdir_page_get_locked(desc->file->f_mapping,
904 * Returns 0 if desc->dir_cookie was found on page desc->page_index
905 * and locks the page to prevent removal from the page cache.
907 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
909 struct inode *inode = file_inode(desc->file);
910 struct nfs_inode *nfsi = NFS_I(inode);
911 __be32 verf[NFS_DIR_VERIFIER_SIZE];
914 desc->page = nfs_readdir_page_get_cached(desc);
917 if (nfs_readdir_page_needs_filling(desc->page)) {
918 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
921 nfs_readdir_page_unlock_and_put_cached(desc);
922 if (res == -EBADCOOKIE || res == -ENOTSYNC) {
923 invalidate_inode_pages2(desc->file->f_mapping);
924 desc->page_index = 0;
929 memcpy(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf));
931 res = nfs_readdir_search_array(desc);
933 nfsi->page_index = desc->page_index;
936 nfs_readdir_page_unlock_and_put_cached(desc);
940 static bool nfs_readdir_dont_search_cache(struct nfs_readdir_descriptor *desc)
942 struct address_space *mapping = desc->file->f_mapping;
943 struct inode *dir = file_inode(desc->file);
944 unsigned int dtsize = NFS_SERVER(dir)->dtsize;
945 loff_t size = i_size_read(dir);
948 * Default to uncached readdir if the page cache is empty, and
949 * we're looking for a non-zero cookie in a large directory.
951 return desc->dir_cookie != 0 && mapping->nrpages == 0 && size > dtsize;
954 /* Search for desc->dir_cookie from the beginning of the page cache */
955 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
959 if (nfs_readdir_dont_search_cache(desc))
963 if (desc->page_index == 0) {
964 desc->current_index = 0;
965 desc->prev_index = 0;
966 desc->last_cookie = 0;
968 res = find_and_lock_cache_page(desc);
969 } while (res == -EAGAIN);
974 * Once we've found the start of the dirent within a page: fill 'er up...
976 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc)
978 struct file *file = desc->file;
979 struct nfs_inode *nfsi = NFS_I(file_inode(file));
980 struct nfs_cache_array *array;
983 array = kmap(desc->page);
984 for (i = desc->cache_entry_index; i < array->size; i++) {
985 struct nfs_cache_array_entry *ent;
987 ent = &array->array[i];
988 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
989 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
993 memcpy(desc->verf, nfsi->cookieverf, sizeof(desc->verf));
994 if (i < (array->size-1))
995 desc->dir_cookie = array->array[i+1].cookie;
997 desc->dir_cookie = array->last_cookie;
998 if (nfs_readdir_use_cookie(file))
999 desc->ctx->pos = desc->dir_cookie;
1002 if (desc->duped != 0)
1005 if (array->page_is_eof)
1009 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1010 (unsigned long long)desc->dir_cookie);
1014 * If we cannot find a cookie in our cache, we suspect that this is
1015 * because it points to a deleted file, so we ask the server to return
1016 * whatever it thinks is the next entry. We then feed this to filldir.
1017 * If all goes well, we should then be able to find our way round the
1018 * cache on the next call to readdir_search_pagecache();
1020 * NOTE: we cannot add the anonymous page to the pagecache because
1021 * the data it contains might not be page aligned. Besides,
1022 * we should already have a complete representation of the
1023 * directory in the page cache by the time we get here.
1025 static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1027 struct page **arrays;
1029 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1030 int status = -ENOMEM;
1032 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1033 (unsigned long long)desc->dir_cookie);
1035 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1038 arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL);
1042 desc->page_index = 0;
1043 desc->last_cookie = desc->dir_cookie;
1046 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1048 for (i = 0; !desc->eof && i < sz && arrays[i]; i++) {
1049 desc->page = arrays[i];
1050 nfs_do_filldir(desc);
1055 for (i = 0; i < sz && arrays[i]; i++)
1056 nfs_readdir_page_array_free(arrays[i]);
1059 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1063 /* The file offset position represents the dirent entry number. A
1064 last cookie cache takes care of the common case of reading the
1067 static int nfs_readdir(struct file *file, struct dir_context *ctx)
1069 struct dentry *dentry = file_dentry(file);
1070 struct inode *inode = d_inode(dentry);
1071 struct nfs_open_dir_context *dir_ctx = file->private_data;
1072 struct nfs_readdir_descriptor *desc;
1075 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1076 file, (long long)ctx->pos);
1077 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1080 * ctx->pos points to the dirent entry number.
1081 * *desc->dir_cookie has the cookie for the next entry. We have
1082 * to either find the entry with the appropriate number or
1083 * revalidate the cookie.
1085 if (ctx->pos == 0 || nfs_attribute_cache_expired(inode)) {
1086 res = nfs_revalidate_mapping(inode, file->f_mapping);
1092 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1097 desc->plus = nfs_use_readdirplus(inode, ctx);
1099 spin_lock(&file->f_lock);
1100 desc->dir_cookie = dir_ctx->dir_cookie;
1101 desc->dup_cookie = dir_ctx->dup_cookie;
1102 desc->duped = dir_ctx->duped;
1103 desc->attr_gencount = dir_ctx->attr_gencount;
1104 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1105 spin_unlock(&file->f_lock);
1108 res = readdir_search_pagecache(desc);
1110 if (res == -EBADCOOKIE) {
1112 /* This means either end of directory */
1113 if (desc->dir_cookie && !desc->eof) {
1114 /* Or that the server has 'lost' a cookie */
1115 res = uncached_readdir(desc);
1118 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1123 if (res == -ETOOSMALL && desc->plus) {
1124 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
1125 nfs_zap_caches(inode);
1126 desc->page_index = 0;
1134 nfs_do_filldir(desc);
1135 nfs_readdir_page_unlock_and_put_cached(desc);
1136 } while (!desc->eof);
1138 spin_lock(&file->f_lock);
1139 dir_ctx->dir_cookie = desc->dir_cookie;
1140 dir_ctx->dup_cookie = desc->dup_cookie;
1141 dir_ctx->duped = desc->duped;
1142 dir_ctx->attr_gencount = desc->attr_gencount;
1143 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1144 spin_unlock(&file->f_lock);
1149 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1153 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1155 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1157 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1158 filp, offset, whence);
1166 spin_lock(&filp->f_lock);
1171 spin_lock(&filp->f_lock);
1172 offset += filp->f_pos;
1174 spin_unlock(&filp->f_lock);
1178 if (offset != filp->f_pos) {
1179 filp->f_pos = offset;
1180 if (nfs_readdir_use_cookie(filp))
1181 dir_ctx->dir_cookie = offset;
1183 dir_ctx->dir_cookie = 0;
1185 memset(dir_ctx->verf, 0, sizeof(dir_ctx->verf));
1188 spin_unlock(&filp->f_lock);
1193 * All directory operations under NFS are synchronous, so fsync()
1194 * is a dummy operation.
1196 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1199 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1201 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1206 * nfs_force_lookup_revalidate - Mark the directory as having changed
1207 * @dir: pointer to directory inode
1209 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1210 * full lookup on all child dentries of 'dir' whenever a change occurs
1211 * on the server that might have invalidated our dcache.
1213 * Note that we reserve bit '0' as a tag to let us know when a dentry
1214 * was revalidated while holding a delegation on its inode.
1216 * The caller should be holding dir->i_lock
1218 void nfs_force_lookup_revalidate(struct inode *dir)
1220 NFS_I(dir)->cache_change_attribute += 2;
1222 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1225 * nfs_verify_change_attribute - Detects NFS remote directory changes
1226 * @dir: pointer to parent directory inode
1227 * @verf: previously saved change attribute
1229 * Return "false" if the verifiers doesn't match the change attribute.
1230 * This would usually indicate that the directory contents have changed on
1231 * the server, and that any dentries need revalidating.
1233 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1235 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1238 static void nfs_set_verifier_delegated(unsigned long *verf)
1243 #if IS_ENABLED(CONFIG_NFS_V4)
1244 static void nfs_unset_verifier_delegated(unsigned long *verf)
1248 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1250 static bool nfs_test_verifier_delegated(unsigned long verf)
1255 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1257 return nfs_test_verifier_delegated(dentry->d_time);
1260 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1262 struct inode *inode = d_inode(dentry);
1264 if (!nfs_verifier_is_delegated(dentry) &&
1265 !nfs_verify_change_attribute(d_inode(dentry->d_parent), verf))
1267 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1268 nfs_set_verifier_delegated(&verf);
1270 dentry->d_time = verf;
1274 * nfs_set_verifier - save a parent directory verifier in the dentry
1275 * @dentry: pointer to dentry
1276 * @verf: verifier to save
1278 * Saves the parent directory verifier in @dentry. If the inode has
1279 * a delegation, we also tag the dentry as having been revalidated
1280 * while holding a delegation so that we know we don't have to
1281 * look it up again after a directory change.
1283 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1286 spin_lock(&dentry->d_lock);
1287 nfs_set_verifier_locked(dentry, verf);
1288 spin_unlock(&dentry->d_lock);
1290 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1292 #if IS_ENABLED(CONFIG_NFS_V4)
1294 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1295 * @inode: pointer to inode
1297 * Iterates through the dentries in the inode alias list and clears
1298 * the tag used to indicate that the dentry has been revalidated
1299 * while holding a delegation.
1300 * This function is intended for use when the delegation is being
1301 * returned or revoked.
1303 void nfs_clear_verifier_delegated(struct inode *inode)
1305 struct dentry *alias;
1309 spin_lock(&inode->i_lock);
1310 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1311 spin_lock(&alias->d_lock);
1312 nfs_unset_verifier_delegated(&alias->d_time);
1313 spin_unlock(&alias->d_lock);
1315 spin_unlock(&inode->i_lock);
1317 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1318 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1321 * A check for whether or not the parent directory has changed.
1322 * In the case it has, we assume that the dentries are untrustworthy
1323 * and may need to be looked up again.
1324 * If rcu_walk prevents us from performing a full check, return 0.
1326 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1329 if (IS_ROOT(dentry))
1331 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1333 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1335 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1336 if (nfs_mapping_need_revalidate_inode(dir)) {
1339 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1342 if (!nfs_verify_change_attribute(dir, dentry->d_time))
1348 * Use intent information to check whether or not we're going to do
1349 * an O_EXCL create using this path component.
1351 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1353 if (NFS_PROTO(dir)->version == 2)
1355 return flags & LOOKUP_EXCL;
1359 * Inode and filehandle revalidation for lookups.
1361 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1362 * or if the intent information indicates that we're about to open this
1363 * particular file and the "nocto" mount flag is not set.
1367 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1369 struct nfs_server *server = NFS_SERVER(inode);
1372 if (IS_AUTOMOUNT(inode))
1375 if (flags & LOOKUP_OPEN) {
1376 switch (inode->i_mode & S_IFMT) {
1378 /* A NFSv4 OPEN will revalidate later */
1379 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1383 if (server->flags & NFS_MOUNT_NOCTO)
1385 /* NFS close-to-open cache consistency validation */
1390 /* VFS wants an on-the-wire revalidation */
1391 if (flags & LOOKUP_REVAL)
1394 return (inode->i_nlink == 0) ? -ESTALE : 0;
1396 if (flags & LOOKUP_RCU)
1398 ret = __nfs_revalidate_inode(server, inode);
1405 * We judge how long we want to trust negative
1406 * dentries by looking at the parent inode mtime.
1408 * If parent mtime has changed, we revalidate, else we wait for a
1409 * period corresponding to the parent's attribute cache timeout value.
1411 * If LOOKUP_RCU prevents us from performing a full check, return 1
1412 * suggesting a reval is needed.
1414 * Note that when creating a new file, or looking up a rename target,
1415 * then it shouldn't be necessary to revalidate a negative dentry.
1418 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1421 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1423 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1425 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1429 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1430 struct inode *inode, int error)
1434 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is valid\n",
1438 nfs_mark_for_revalidate(dir);
1439 if (inode && S_ISDIR(inode->i_mode)) {
1440 /* Purge readdir caches. */
1441 nfs_zap_caches(inode);
1443 * We can't d_drop the root of a disconnected tree:
1444 * its d_hash is on the s_anon list and d_drop() would hide
1445 * it from shrink_dcache_for_unmount(), leading to busy
1446 * inodes on unmount and further oopses.
1448 if (IS_ROOT(dentry))
1451 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) is invalid\n",
1455 dfprintk(LOOKUPCACHE, "NFS: %s(%pd2) lookup returned error %d\n",
1456 __func__, dentry, error);
1461 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1465 if (nfs_neg_need_reval(dir, dentry, flags)) {
1466 if (flags & LOOKUP_RCU)
1470 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1474 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1475 struct inode *inode)
1477 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1478 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1482 nfs_lookup_revalidate_dentry(struct inode *dir, struct dentry *dentry,
1483 struct inode *inode)
1485 struct nfs_fh *fhandle;
1486 struct nfs_fattr *fattr;
1487 struct nfs4_label *label;
1488 unsigned long dir_verifier;
1492 fhandle = nfs_alloc_fhandle();
1493 fattr = nfs_alloc_fattr();
1494 label = nfs4_label_alloc(NFS_SERVER(inode), GFP_KERNEL);
1495 if (fhandle == NULL || fattr == NULL || IS_ERR(label))
1498 dir_verifier = nfs_save_change_attribute(dir);
1499 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1507 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1513 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1515 if (nfs_refresh_inode(inode, fattr) < 0)
1518 nfs_setsecurity(inode, fattr, label);
1519 nfs_set_verifier(dentry, dir_verifier);
1521 /* set a readdirplus hint that we had a cache miss */
1522 nfs_force_use_readdirplus(dir);
1525 nfs_free_fattr(fattr);
1526 nfs_free_fhandle(fhandle);
1527 nfs4_label_free(label);
1528 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1532 * This is called every time the dcache has a lookup hit,
1533 * and we should check whether we can really trust that
1536 * NOTE! The hit can be a negative hit too, don't assume
1539 * If the parent directory is seen to have changed, we throw out the
1540 * cached dentry and do a new lookup.
1543 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1546 struct inode *inode;
1549 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1550 inode = d_inode(dentry);
1553 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1555 if (is_bad_inode(inode)) {
1556 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1561 if (nfs_verifier_is_delegated(dentry))
1562 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1564 /* Force a full look up iff the parent directory has changed */
1565 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1566 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1567 error = nfs_lookup_verify_inode(inode, flags);
1569 if (error == -ESTALE)
1570 nfs_zap_caches(dir);
1573 nfs_advise_use_readdirplus(dir);
1577 if (flags & LOOKUP_RCU)
1580 if (NFS_STALE(inode))
1583 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1584 error = nfs_lookup_revalidate_dentry(dir, dentry, inode);
1585 trace_nfs_lookup_revalidate_exit(dir, dentry, flags, error);
1588 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1590 if (flags & LOOKUP_RCU)
1592 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1596 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1597 int (*reval)(struct inode *, struct dentry *, unsigned int))
1599 struct dentry *parent;
1603 if (flags & LOOKUP_RCU) {
1604 parent = READ_ONCE(dentry->d_parent);
1605 dir = d_inode_rcu(parent);
1608 ret = reval(dir, dentry, flags);
1609 if (parent != READ_ONCE(dentry->d_parent))
1612 parent = dget_parent(dentry);
1613 ret = reval(d_inode(parent), dentry, flags);
1619 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1621 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1625 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1626 * when we don't really care about the dentry name. This is called when a
1627 * pathwalk ends on a dentry that was not found via a normal lookup in the
1628 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1630 * In this situation, we just want to verify that the inode itself is OK
1631 * since the dentry might have changed on the server.
1633 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1635 struct inode *inode = d_inode(dentry);
1639 * I believe we can only get a negative dentry here in the case of a
1640 * procfs-style symlink. Just assume it's correct for now, but we may
1641 * eventually need to do something more here.
1644 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1649 if (is_bad_inode(inode)) {
1650 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1655 error = nfs_lookup_verify_inode(inode, flags);
1656 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1657 __func__, inode->i_ino, error ? "invalid" : "valid");
1662 * This is called from dput() when d_count is going to 0.
1664 static int nfs_dentry_delete(const struct dentry *dentry)
1666 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1667 dentry, dentry->d_flags);
1669 /* Unhash any dentry with a stale inode */
1670 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1673 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1674 /* Unhash it, so that ->d_iput() would be called */
1677 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1678 /* Unhash it, so that ancestors of killed async unlink
1679 * files will be cleaned up during umount */
1686 /* Ensure that we revalidate inode->i_nlink */
1687 static void nfs_drop_nlink(struct inode *inode)
1689 spin_lock(&inode->i_lock);
1690 /* drop the inode if we're reasonably sure this is the last link */
1691 if (inode->i_nlink > 0)
1693 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1694 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_CHANGE
1695 | NFS_INO_INVALID_CTIME
1696 | NFS_INO_INVALID_OTHER
1697 | NFS_INO_REVAL_FORCED;
1698 spin_unlock(&inode->i_lock);
1702 * Called when the dentry loses inode.
1703 * We use it to clean up silly-renamed files.
1705 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1707 if (S_ISDIR(inode->i_mode))
1708 /* drop any readdir cache as it could easily be old */
1709 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1711 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1712 nfs_complete_unlink(dentry, inode);
1713 nfs_drop_nlink(inode);
1718 static void nfs_d_release(struct dentry *dentry)
1720 /* free cached devname value, if it survived that far */
1721 if (unlikely(dentry->d_fsdata)) {
1722 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1725 kfree(dentry->d_fsdata);
1729 const struct dentry_operations nfs_dentry_operations = {
1730 .d_revalidate = nfs_lookup_revalidate,
1731 .d_weak_revalidate = nfs_weak_revalidate,
1732 .d_delete = nfs_dentry_delete,
1733 .d_iput = nfs_dentry_iput,
1734 .d_automount = nfs_d_automount,
1735 .d_release = nfs_d_release,
1737 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1739 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1742 struct inode *inode = NULL;
1743 struct nfs_fh *fhandle = NULL;
1744 struct nfs_fattr *fattr = NULL;
1745 struct nfs4_label *label = NULL;
1746 unsigned long dir_verifier;
1749 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1750 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1752 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1753 return ERR_PTR(-ENAMETOOLONG);
1756 * If we're doing an exclusive create, optimize away the lookup
1757 * but don't hash the dentry.
1759 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1762 res = ERR_PTR(-ENOMEM);
1763 fhandle = nfs_alloc_fhandle();
1764 fattr = nfs_alloc_fattr();
1765 if (fhandle == NULL || fattr == NULL)
1768 label = nfs4_label_alloc(NFS_SERVER(dir), GFP_NOWAIT);
1772 dir_verifier = nfs_save_change_attribute(dir);
1773 trace_nfs_lookup_enter(dir, dentry, flags);
1774 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, label);
1775 if (error == -ENOENT)
1778 res = ERR_PTR(error);
1781 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
1782 res = ERR_CAST(inode);
1786 /* Notify readdir to use READDIRPLUS */
1787 nfs_force_use_readdirplus(dir);
1790 res = d_splice_alias(inode, dentry);
1796 nfs_set_verifier(dentry, dir_verifier);
1798 trace_nfs_lookup_exit(dir, dentry, flags, error);
1799 nfs4_label_free(label);
1801 nfs_free_fattr(fattr);
1802 nfs_free_fhandle(fhandle);
1805 EXPORT_SYMBOL_GPL(nfs_lookup);
1807 #if IS_ENABLED(CONFIG_NFS_V4)
1808 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1810 const struct dentry_operations nfs4_dentry_operations = {
1811 .d_revalidate = nfs4_lookup_revalidate,
1812 .d_weak_revalidate = nfs_weak_revalidate,
1813 .d_delete = nfs_dentry_delete,
1814 .d_iput = nfs_dentry_iput,
1815 .d_automount = nfs_d_automount,
1816 .d_release = nfs_d_release,
1818 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1820 static fmode_t flags_to_mode(int flags)
1822 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1823 if ((flags & O_ACCMODE) != O_WRONLY)
1825 if ((flags & O_ACCMODE) != O_RDONLY)
1830 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1832 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
1835 static int do_open(struct inode *inode, struct file *filp)
1837 nfs_fscache_open_file(inode, filp);
1841 static int nfs_finish_open(struct nfs_open_context *ctx,
1842 struct dentry *dentry,
1843 struct file *file, unsigned open_flags)
1847 err = finish_open(file, dentry, do_open);
1850 if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
1851 nfs_file_set_open_context(file, ctx);
1858 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
1859 struct file *file, unsigned open_flags,
1862 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1863 struct nfs_open_context *ctx;
1865 struct iattr attr = { .ia_valid = ATTR_OPEN };
1866 struct inode *inode;
1867 unsigned int lookup_flags = 0;
1868 bool switched = false;
1872 /* Expect a negative dentry */
1873 BUG_ON(d_inode(dentry));
1875 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
1876 dir->i_sb->s_id, dir->i_ino, dentry);
1878 err = nfs_check_flags(open_flags);
1882 /* NFS only supports OPEN on regular files */
1883 if ((open_flags & O_DIRECTORY)) {
1884 if (!d_in_lookup(dentry)) {
1886 * Hashed negative dentry with O_DIRECTORY: dentry was
1887 * revalidated and is fine, no need to perform lookup
1892 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
1896 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1897 return -ENAMETOOLONG;
1899 if (open_flags & O_CREAT) {
1900 struct nfs_server *server = NFS_SERVER(dir);
1902 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
1903 mode &= ~current_umask();
1905 attr.ia_valid |= ATTR_MODE;
1906 attr.ia_mode = mode;
1908 if (open_flags & O_TRUNC) {
1909 attr.ia_valid |= ATTR_SIZE;
1913 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
1916 dentry = d_alloc_parallel(dentry->d_parent,
1917 &dentry->d_name, &wq);
1919 return PTR_ERR(dentry);
1920 if (unlikely(!d_in_lookup(dentry)))
1921 return finish_no_open(file, dentry);
1924 ctx = create_nfs_open_context(dentry, open_flags, file);
1929 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
1930 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
1932 file->f_mode |= FMODE_CREATED;
1933 if (IS_ERR(inode)) {
1934 err = PTR_ERR(inode);
1935 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1936 put_nfs_open_context(ctx);
1940 d_splice_alias(NULL, dentry);
1941 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1947 if (!(open_flags & O_NOFOLLOW))
1957 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
1958 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
1959 put_nfs_open_context(ctx);
1961 if (unlikely(switched)) {
1962 d_lookup_done(dentry);
1968 res = nfs_lookup(dir, dentry, lookup_flags);
1970 d_lookup_done(dentry);
1977 return PTR_ERR(res);
1978 return finish_no_open(file, res);
1980 EXPORT_SYMBOL_GPL(nfs_atomic_open);
1983 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1986 struct inode *inode;
1988 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
1990 if (d_mountpoint(dentry))
1993 inode = d_inode(dentry);
1995 /* We can't create new files in nfs_open_revalidate(), so we
1996 * optimize away revalidation of negative dentries.
2001 if (nfs_verifier_is_delegated(dentry))
2002 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2004 /* NFS only supports OPEN on regular files */
2005 if (!S_ISREG(inode->i_mode))
2008 /* We cannot do exclusive creation on a positive dentry */
2009 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2012 /* Check if the directory changed */
2013 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2016 /* Let f_op->open() actually open (and revalidate) the file */
2019 if (flags & LOOKUP_RCU)
2021 return nfs_lookup_revalidate_dentry(dir, dentry, inode);
2024 return nfs_do_lookup_revalidate(dir, dentry, flags);
2027 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2029 return __nfs_lookup_revalidate(dentry, flags,
2030 nfs4_do_lookup_revalidate);
2033 #endif /* CONFIG_NFSV4 */
2036 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2037 struct nfs_fattr *fattr,
2038 struct nfs4_label *label)
2040 struct dentry *parent = dget_parent(dentry);
2041 struct inode *dir = d_inode(parent);
2042 struct inode *inode;
2048 if (fhandle->size == 0) {
2049 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr, NULL);
2053 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2054 if (!(fattr->valid & NFS_ATTR_FATTR)) {
2055 struct nfs_server *server = NFS_SB(dentry->d_sb);
2056 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2061 inode = nfs_fhget(dentry->d_sb, fhandle, fattr, label);
2062 d = d_splice_alias(inode, dentry);
2067 nfs_mark_for_revalidate(dir);
2071 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2074 * Code common to create, mkdir, and mknod.
2076 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2077 struct nfs_fattr *fattr,
2078 struct nfs4_label *label)
2082 d = nfs_add_or_obtain(dentry, fhandle, fattr, label);
2086 /* Callers don't care */
2090 EXPORT_SYMBOL_GPL(nfs_instantiate);
2093 * Following a failed create operation, we drop the dentry rather
2094 * than retain a negative dentry. This avoids a problem in the event
2095 * that the operation succeeded on the server, but an error in the
2096 * reply path made it appear to have failed.
2098 int nfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2099 struct dentry *dentry, umode_t mode, bool excl)
2102 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2105 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2106 dir->i_sb->s_id, dir->i_ino, dentry);
2108 attr.ia_mode = mode;
2109 attr.ia_valid = ATTR_MODE;
2111 trace_nfs_create_enter(dir, dentry, open_flags);
2112 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2113 trace_nfs_create_exit(dir, dentry, open_flags, error);
2121 EXPORT_SYMBOL_GPL(nfs_create);
2124 * See comments for nfs_proc_create regarding failed operations.
2127 nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2128 struct dentry *dentry, umode_t mode, dev_t rdev)
2133 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2134 dir->i_sb->s_id, dir->i_ino, dentry);
2136 attr.ia_mode = mode;
2137 attr.ia_valid = ATTR_MODE;
2139 trace_nfs_mknod_enter(dir, dentry);
2140 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2141 trace_nfs_mknod_exit(dir, dentry, status);
2149 EXPORT_SYMBOL_GPL(nfs_mknod);
2152 * See comments for nfs_proc_create regarding failed operations.
2154 int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2155 struct dentry *dentry, umode_t mode)
2160 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2161 dir->i_sb->s_id, dir->i_ino, dentry);
2163 attr.ia_valid = ATTR_MODE;
2164 attr.ia_mode = mode | S_IFDIR;
2166 trace_nfs_mkdir_enter(dir, dentry);
2167 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2168 trace_nfs_mkdir_exit(dir, dentry, error);
2176 EXPORT_SYMBOL_GPL(nfs_mkdir);
2178 static void nfs_dentry_handle_enoent(struct dentry *dentry)
2180 if (simple_positive(dentry))
2184 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2188 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2189 dir->i_sb->s_id, dir->i_ino, dentry);
2191 trace_nfs_rmdir_enter(dir, dentry);
2192 if (d_really_is_positive(dentry)) {
2193 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2194 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2195 /* Ensure the VFS deletes this inode */
2198 clear_nlink(d_inode(dentry));
2201 nfs_dentry_handle_enoent(dentry);
2203 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2205 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2206 trace_nfs_rmdir_exit(dir, dentry, error);
2210 EXPORT_SYMBOL_GPL(nfs_rmdir);
2213 * Remove a file after making sure there are no pending writes,
2214 * and after checking that the file has only one user.
2216 * We invalidate the attribute cache and free the inode prior to the operation
2217 * to avoid possible races if the server reuses the inode.
2219 static int nfs_safe_remove(struct dentry *dentry)
2221 struct inode *dir = d_inode(dentry->d_parent);
2222 struct inode *inode = d_inode(dentry);
2225 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2227 /* If the dentry was sillyrenamed, we simply call d_delete() */
2228 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2233 trace_nfs_remove_enter(dir, dentry);
2234 if (inode != NULL) {
2235 error = NFS_PROTO(dir)->remove(dir, dentry);
2237 nfs_drop_nlink(inode);
2239 error = NFS_PROTO(dir)->remove(dir, dentry);
2240 if (error == -ENOENT)
2241 nfs_dentry_handle_enoent(dentry);
2242 trace_nfs_remove_exit(dir, dentry, error);
2247 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2248 * belongs to an active ".nfs..." file and we return -EBUSY.
2250 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2252 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2255 int need_rehash = 0;
2257 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2258 dir->i_ino, dentry);
2260 trace_nfs_unlink_enter(dir, dentry);
2261 spin_lock(&dentry->d_lock);
2262 if (d_count(dentry) > 1) {
2263 spin_unlock(&dentry->d_lock);
2264 /* Start asynchronous writeout of the inode */
2265 write_inode_now(d_inode(dentry), 0);
2266 error = nfs_sillyrename(dir, dentry);
2269 if (!d_unhashed(dentry)) {
2273 spin_unlock(&dentry->d_lock);
2274 error = nfs_safe_remove(dentry);
2275 if (!error || error == -ENOENT) {
2276 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2277 } else if (need_rehash)
2280 trace_nfs_unlink_exit(dir, dentry, error);
2283 EXPORT_SYMBOL_GPL(nfs_unlink);
2286 * To create a symbolic link, most file systems instantiate a new inode,
2287 * add a page to it containing the path, then write it out to the disk
2288 * using prepare_write/commit_write.
2290 * Unfortunately the NFS client can't create the in-core inode first
2291 * because it needs a file handle to create an in-core inode (see
2292 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2293 * symlink request has completed on the server.
2295 * So instead we allocate a raw page, copy the symname into it, then do
2296 * the SYMLINK request with the page as the buffer. If it succeeds, we
2297 * now have a new file handle and can instantiate an in-core NFS inode
2298 * and move the raw page into its mapping.
2300 int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
2301 struct dentry *dentry, const char *symname)
2306 unsigned int pathlen = strlen(symname);
2309 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2310 dir->i_ino, dentry, symname);
2312 if (pathlen > PAGE_SIZE)
2313 return -ENAMETOOLONG;
2315 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2316 attr.ia_valid = ATTR_MODE;
2318 page = alloc_page(GFP_USER);
2322 kaddr = page_address(page);
2323 memcpy(kaddr, symname, pathlen);
2324 if (pathlen < PAGE_SIZE)
2325 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2327 trace_nfs_symlink_enter(dir, dentry);
2328 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2329 trace_nfs_symlink_exit(dir, dentry, error);
2331 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2332 dir->i_sb->s_id, dir->i_ino,
2333 dentry, symname, error);
2340 * No big deal if we can't add this page to the page cache here.
2341 * READLINK will get the missing page from the server if needed.
2343 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2345 SetPageUptodate(page);
2348 * add_to_page_cache_lru() grabs an extra page refcount.
2349 * Drop it here to avoid leaking this page later.
2357 EXPORT_SYMBOL_GPL(nfs_symlink);
2360 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2362 struct inode *inode = d_inode(old_dentry);
2365 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2366 old_dentry, dentry);
2368 trace_nfs_link_enter(inode, dir, dentry);
2370 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2373 d_add(dentry, inode);
2375 trace_nfs_link_exit(inode, dir, dentry, error);
2378 EXPORT_SYMBOL_GPL(nfs_link);
2382 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2383 * different file handle for the same inode after a rename (e.g. when
2384 * moving to a different directory). A fail-safe method to do so would
2385 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2386 * rename the old file using the sillyrename stuff. This way, the original
2387 * file in old_dir will go away when the last process iput()s the inode.
2391 * It actually works quite well. One needs to have the possibility for
2392 * at least one ".nfs..." file in each directory the file ever gets
2393 * moved or linked to which happens automagically with the new
2394 * implementation that only depends on the dcache stuff instead of
2395 * using the inode layer
2397 * Unfortunately, things are a little more complicated than indicated
2398 * above. For a cross-directory move, we want to make sure we can get
2399 * rid of the old inode after the operation. This means there must be
2400 * no pending writes (if it's a file), and the use count must be 1.
2401 * If these conditions are met, we can drop the dentries before doing
2404 int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
2405 struct dentry *old_dentry, struct inode *new_dir,
2406 struct dentry *new_dentry, unsigned int flags)
2408 struct inode *old_inode = d_inode(old_dentry);
2409 struct inode *new_inode = d_inode(new_dentry);
2410 struct dentry *dentry = NULL, *rehash = NULL;
2411 struct rpc_task *task;
2417 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2418 old_dentry, new_dentry,
2419 d_count(new_dentry));
2421 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2423 * For non-directories, check whether the target is busy and if so,
2424 * make a copy of the dentry and then do a silly-rename. If the
2425 * silly-rename succeeds, the copied dentry is hashed and becomes
2428 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2430 * To prevent any new references to the target during the
2431 * rename, we unhash the dentry in advance.
2433 if (!d_unhashed(new_dentry)) {
2435 rehash = new_dentry;
2438 if (d_count(new_dentry) > 2) {
2441 /* copy the target dentry's name */
2442 dentry = d_alloc(new_dentry->d_parent,
2443 &new_dentry->d_name);
2447 /* silly-rename the existing target ... */
2448 err = nfs_sillyrename(new_dir, new_dentry);
2452 new_dentry = dentry;
2458 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2460 error = PTR_ERR(task);
2464 error = rpc_wait_for_completion_task(task);
2466 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2467 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2470 error = task->tk_status;
2472 /* Ensure the inode attributes are revalidated */
2474 spin_lock(&old_inode->i_lock);
2475 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2476 NFS_I(old_inode)->cache_validity |= NFS_INO_INVALID_CHANGE
2477 | NFS_INO_INVALID_CTIME
2478 | NFS_INO_REVAL_FORCED;
2479 spin_unlock(&old_inode->i_lock);
2484 trace_nfs_rename_exit(old_dir, old_dentry,
2485 new_dir, new_dentry, error);
2487 if (new_inode != NULL)
2488 nfs_drop_nlink(new_inode);
2490 * The d_move() should be here instead of in an async RPC completion
2491 * handler because we need the proper locks to move the dentry. If
2492 * we're interrupted by a signal, the async RPC completion handler
2493 * should mark the directories for revalidation.
2495 d_move(old_dentry, new_dentry);
2496 nfs_set_verifier(old_dentry,
2497 nfs_save_change_attribute(new_dir));
2498 } else if (error == -ENOENT)
2499 nfs_dentry_handle_enoent(old_dentry);
2501 /* new dentry created? */
2506 EXPORT_SYMBOL_GPL(nfs_rename);
2508 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2509 static LIST_HEAD(nfs_access_lru_list);
2510 static atomic_long_t nfs_access_nr_entries;
2512 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2513 module_param(nfs_access_max_cachesize, ulong, 0644);
2514 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2516 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2518 put_cred(entry->cred);
2519 kfree_rcu(entry, rcu_head);
2520 smp_mb__before_atomic();
2521 atomic_long_dec(&nfs_access_nr_entries);
2522 smp_mb__after_atomic();
2525 static void nfs_access_free_list(struct list_head *head)
2527 struct nfs_access_entry *cache;
2529 while (!list_empty(head)) {
2530 cache = list_entry(head->next, struct nfs_access_entry, lru);
2531 list_del(&cache->lru);
2532 nfs_access_free_entry(cache);
2536 static unsigned long
2537 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2540 struct nfs_inode *nfsi, *next;
2541 struct nfs_access_entry *cache;
2544 spin_lock(&nfs_access_lru_lock);
2545 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2546 struct inode *inode;
2548 if (nr_to_scan-- == 0)
2550 inode = &nfsi->vfs_inode;
2551 spin_lock(&inode->i_lock);
2552 if (list_empty(&nfsi->access_cache_entry_lru))
2553 goto remove_lru_entry;
2554 cache = list_entry(nfsi->access_cache_entry_lru.next,
2555 struct nfs_access_entry, lru);
2556 list_move(&cache->lru, &head);
2557 rb_erase(&cache->rb_node, &nfsi->access_cache);
2559 if (!list_empty(&nfsi->access_cache_entry_lru))
2560 list_move_tail(&nfsi->access_cache_inode_lru,
2561 &nfs_access_lru_list);
2564 list_del_init(&nfsi->access_cache_inode_lru);
2565 smp_mb__before_atomic();
2566 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2567 smp_mb__after_atomic();
2569 spin_unlock(&inode->i_lock);
2571 spin_unlock(&nfs_access_lru_lock);
2572 nfs_access_free_list(&head);
2577 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2579 int nr_to_scan = sc->nr_to_scan;
2580 gfp_t gfp_mask = sc->gfp_mask;
2582 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2584 return nfs_do_access_cache_scan(nr_to_scan);
2589 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2591 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2595 nfs_access_cache_enforce_limit(void)
2597 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2599 unsigned int nr_to_scan;
2601 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2604 diff = nr_entries - nfs_access_max_cachesize;
2605 if (diff < nr_to_scan)
2607 nfs_do_access_cache_scan(nr_to_scan);
2610 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2612 struct rb_root *root_node = &nfsi->access_cache;
2614 struct nfs_access_entry *entry;
2616 /* Unhook entries from the cache */
2617 while ((n = rb_first(root_node)) != NULL) {
2618 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2619 rb_erase(n, root_node);
2620 list_move(&entry->lru, head);
2622 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2625 void nfs_access_zap_cache(struct inode *inode)
2629 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2631 /* Remove from global LRU init */
2632 spin_lock(&nfs_access_lru_lock);
2633 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2634 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2636 spin_lock(&inode->i_lock);
2637 __nfs_access_zap_cache(NFS_I(inode), &head);
2638 spin_unlock(&inode->i_lock);
2639 spin_unlock(&nfs_access_lru_lock);
2640 nfs_access_free_list(&head);
2642 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2644 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2646 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2649 struct nfs_access_entry *entry =
2650 rb_entry(n, struct nfs_access_entry, rb_node);
2651 int cmp = cred_fscmp(cred, entry->cred);
2663 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res, bool may_block)
2665 struct nfs_inode *nfsi = NFS_I(inode);
2666 struct nfs_access_entry *cache;
2670 spin_lock(&inode->i_lock);
2672 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2674 cache = nfs_access_search_rbtree(inode, cred);
2678 /* Found an entry, is our attribute cache valid? */
2679 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2686 spin_unlock(&inode->i_lock);
2687 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2690 spin_lock(&inode->i_lock);
2693 res->cred = cache->cred;
2694 res->mask = cache->mask;
2695 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2698 spin_unlock(&inode->i_lock);
2701 spin_unlock(&inode->i_lock);
2702 nfs_access_zap_cache(inode);
2706 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, struct nfs_access_entry *res)
2708 /* Only check the most recently returned cache entry,
2709 * but do it without locking.
2711 struct nfs_inode *nfsi = NFS_I(inode);
2712 struct nfs_access_entry *cache;
2714 struct list_head *lh;
2717 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2719 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
2720 cache = list_entry(lh, struct nfs_access_entry, lru);
2721 if (lh == &nfsi->access_cache_entry_lru ||
2722 cred_fscmp(cred, cache->cred) != 0)
2726 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2728 res->cred = cache->cred;
2729 res->mask = cache->mask;
2736 int nfs_access_get_cached(struct inode *inode, const struct cred *cred, struct
2737 nfs_access_entry *res, bool may_block)
2741 status = nfs_access_get_cached_rcu(inode, cred, res);
2743 status = nfs_access_get_cached_locked(inode, cred, res,
2748 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
2750 static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2752 struct nfs_inode *nfsi = NFS_I(inode);
2753 struct rb_root *root_node = &nfsi->access_cache;
2754 struct rb_node **p = &root_node->rb_node;
2755 struct rb_node *parent = NULL;
2756 struct nfs_access_entry *entry;
2759 spin_lock(&inode->i_lock);
2760 while (*p != NULL) {
2762 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2763 cmp = cred_fscmp(set->cred, entry->cred);
2766 p = &parent->rb_left;
2768 p = &parent->rb_right;
2772 rb_link_node(&set->rb_node, parent, p);
2773 rb_insert_color(&set->rb_node, root_node);
2774 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2775 spin_unlock(&inode->i_lock);
2778 rb_replace_node(parent, &set->rb_node, root_node);
2779 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2780 list_del(&entry->lru);
2781 spin_unlock(&inode->i_lock);
2782 nfs_access_free_entry(entry);
2785 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
2787 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2790 RB_CLEAR_NODE(&cache->rb_node);
2791 cache->cred = get_cred(set->cred);
2792 cache->mask = set->mask;
2794 /* The above field assignments must be visible
2795 * before this item appears on the lru. We cannot easily
2796 * use rcu_assign_pointer, so just force the memory barrier.
2799 nfs_access_add_rbtree(inode, cache);
2801 /* Update accounting */
2802 smp_mb__before_atomic();
2803 atomic_long_inc(&nfs_access_nr_entries);
2804 smp_mb__after_atomic();
2806 /* Add inode to global LRU list */
2807 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
2808 spin_lock(&nfs_access_lru_lock);
2809 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2810 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2811 &nfs_access_lru_list);
2812 spin_unlock(&nfs_access_lru_lock);
2814 nfs_access_cache_enforce_limit();
2816 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
2818 #define NFS_MAY_READ (NFS_ACCESS_READ)
2819 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
2820 NFS_ACCESS_EXTEND | \
2822 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
2824 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
2825 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
2826 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
2828 nfs_access_calc_mask(u32 access_result, umode_t umode)
2832 if (access_result & NFS_MAY_READ)
2834 if (S_ISDIR(umode)) {
2835 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
2837 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
2839 } else if (S_ISREG(umode)) {
2840 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
2842 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
2844 } else if (access_result & NFS_MAY_WRITE)
2849 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
2851 entry->mask = access_result;
2853 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
2855 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
2857 struct nfs_access_entry cache;
2858 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
2859 int cache_mask = -1;
2862 trace_nfs_access_enter(inode);
2864 status = nfs_access_get_cached(inode, cred, &cache, may_block);
2873 * Determine which access bits we want to ask for...
2875 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND;
2876 if (nfs_server_capable(inode, NFS_CAP_XATTR)) {
2877 cache.mask |= NFS_ACCESS_XAREAD | NFS_ACCESS_XAWRITE |
2880 if (S_ISDIR(inode->i_mode))
2881 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
2883 cache.mask |= NFS_ACCESS_EXECUTE;
2885 status = NFS_PROTO(inode)->access(inode, &cache);
2887 if (status == -ESTALE) {
2888 if (!S_ISDIR(inode->i_mode))
2889 nfs_set_inode_stale(inode);
2891 nfs_zap_caches(inode);
2895 nfs_access_add_cache(inode, &cache);
2897 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
2898 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
2901 trace_nfs_access_exit(inode, mask, cache_mask, status);
2905 static int nfs_open_permission_mask(int openflags)
2909 if (openflags & __FMODE_EXEC) {
2910 /* ONLY check exec rights */
2913 if ((openflags & O_ACCMODE) != O_WRONLY)
2915 if ((openflags & O_ACCMODE) != O_RDONLY)
2922 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
2924 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2926 EXPORT_SYMBOL_GPL(nfs_may_open);
2928 static int nfs_execute_ok(struct inode *inode, int mask)
2930 struct nfs_server *server = NFS_SERVER(inode);
2933 if (S_ISDIR(inode->i_mode))
2935 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_OTHER)) {
2936 if (mask & MAY_NOT_BLOCK)
2938 ret = __nfs_revalidate_inode(server, inode);
2940 if (ret == 0 && !execute_ok(inode))
2945 int nfs_permission(struct user_namespace *mnt_userns,
2946 struct inode *inode,
2949 const struct cred *cred = current_cred();
2952 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2954 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
2956 /* Is this sys_access() ? */
2957 if (mask & (MAY_ACCESS | MAY_CHDIR))
2960 switch (inode->i_mode & S_IFMT) {
2964 if ((mask & MAY_OPEN) &&
2965 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
2970 * Optimize away all write operations, since the server
2971 * will check permissions when we perform the op.
2973 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2978 if (!NFS_PROTO(inode)->access)
2981 res = nfs_do_access(inode, cred, mask);
2983 if (!res && (mask & MAY_EXEC))
2984 res = nfs_execute_ok(inode, mask);
2986 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
2987 inode->i_sb->s_id, inode->i_ino, mask, res);
2990 if (mask & MAY_NOT_BLOCK)
2993 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2995 res = generic_permission(&init_user_ns, inode, mask);
2998 EXPORT_SYMBOL_GPL(nfs_permission);
3002 * version-control: t
3003 * kept-new-versions: 5