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
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/compat.h>
22 #include <linux/module.h>
23 #include <linux/time.h>
24 #include <linux/errno.h>
25 #include <linux/stat.h>
26 #include <linux/fcntl.h>
27 #include <linux/string.h>
28 #include <linux/kernel.h>
29 #include <linux/slab.h>
31 #include <linux/sunrpc/clnt.h>
32 #include <linux/nfs_fs.h>
33 #include <linux/nfs_mount.h>
34 #include <linux/pagemap.h>
35 #include <linux/pagevec.h>
36 #include <linux/namei.h>
37 #include <linux/mount.h>
38 #include <linux/swap.h>
39 #include <linux/sched.h>
40 #include <linux/kmemleak.h>
41 #include <linux/xattr.h>
42 #include <linux/hash.h>
44 #include "delegation.h"
51 /* #define NFS_DEBUG_VERBOSE 1 */
53 static int nfs_opendir(struct inode *, struct file *);
54 static int nfs_closedir(struct inode *, struct file *);
55 static int nfs_readdir(struct file *, struct dir_context *);
56 static int nfs_fsync_dir(struct file *, loff_t, loff_t, int);
57 static loff_t nfs_llseek_dir(struct file *, loff_t, int);
58 static void nfs_readdir_clear_array(struct folio *);
60 const struct file_operations nfs_dir_operations = {
61 .llseek = nfs_llseek_dir,
62 .read = generic_read_dir,
63 .iterate_shared = nfs_readdir,
65 .release = nfs_closedir,
66 .fsync = nfs_fsync_dir,
69 const struct address_space_operations nfs_dir_aops = {
70 .free_folio = nfs_readdir_clear_array,
73 #define NFS_INIT_DTSIZE PAGE_SIZE
75 static struct nfs_open_dir_context *
76 alloc_nfs_open_dir_context(struct inode *dir)
78 struct nfs_inode *nfsi = NFS_I(dir);
79 struct nfs_open_dir_context *ctx;
81 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
83 ctx->attr_gencount = nfsi->attr_gencount;
84 ctx->dtsize = NFS_INIT_DTSIZE;
85 spin_lock(&dir->i_lock);
86 if (list_empty(&nfsi->open_files) &&
87 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
88 nfs_set_cache_invalid(dir,
89 NFS_INO_INVALID_DATA |
90 NFS_INO_REVAL_FORCED);
91 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
92 memcpy(ctx->verf, nfsi->cookieverf, sizeof(ctx->verf));
93 spin_unlock(&dir->i_lock);
96 return ERR_PTR(-ENOMEM);
99 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
101 spin_lock(&dir->i_lock);
102 list_del_rcu(&ctx->list);
103 spin_unlock(&dir->i_lock);
104 kfree_rcu(ctx, rcu_head);
111 nfs_opendir(struct inode *inode, struct file *filp)
114 struct nfs_open_dir_context *ctx;
116 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
118 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
120 ctx = alloc_nfs_open_dir_context(inode);
125 filp->private_data = ctx;
131 nfs_closedir(struct inode *inode, struct file *filp)
133 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
137 struct nfs_cache_array_entry {
141 unsigned int name_len;
142 unsigned char d_type;
145 struct nfs_cache_array {
149 unsigned char folio_full : 1,
151 cookies_are_ordered : 1;
152 struct nfs_cache_array_entry array[];
155 struct nfs_readdir_descriptor {
158 struct dir_context *ctx;
160 pgoff_t folio_index_max;
163 loff_t current_index;
165 __be32 verf[NFS_DIR_VERIFIER_SIZE];
166 unsigned long dir_verifier;
167 unsigned long timestamp;
168 unsigned long gencount;
169 unsigned long attr_gencount;
170 unsigned int cache_entry_index;
171 unsigned int buffer_fills;
179 static void nfs_set_dtsize(struct nfs_readdir_descriptor *desc, unsigned int sz)
181 struct nfs_server *server = NFS_SERVER(file_inode(desc->file));
182 unsigned int maxsize = server->dtsize;
186 if (sz < NFS_MIN_FILE_IO_SIZE)
187 sz = NFS_MIN_FILE_IO_SIZE;
191 static void nfs_shrink_dtsize(struct nfs_readdir_descriptor *desc)
193 nfs_set_dtsize(desc, desc->dtsize >> 1);
196 static void nfs_grow_dtsize(struct nfs_readdir_descriptor *desc)
198 nfs_set_dtsize(desc, desc->dtsize << 1);
201 static void nfs_readdir_folio_init_array(struct folio *folio, u64 last_cookie,
204 struct nfs_cache_array *array;
206 array = kmap_local_folio(folio, 0);
207 array->change_attr = change_attr;
208 array->last_cookie = last_cookie;
210 array->folio_full = 0;
211 array->folio_is_eof = 0;
212 array->cookies_are_ordered = 1;
217 * we are freeing strings created by nfs_add_to_readdir_array()
219 static void nfs_readdir_clear_array(struct folio *folio)
221 struct nfs_cache_array *array;
224 array = kmap_local_folio(folio, 0);
225 for (i = 0; i < array->size; i++)
226 kfree(array->array[i].name);
231 static void nfs_readdir_folio_reinit_array(struct folio *folio, u64 last_cookie,
234 nfs_readdir_clear_array(folio);
235 nfs_readdir_folio_init_array(folio, last_cookie, change_attr);
238 static struct folio *
239 nfs_readdir_folio_array_alloc(u64 last_cookie, gfp_t gfp_flags)
241 struct folio *folio = folio_alloc(gfp_flags, 0);
243 nfs_readdir_folio_init_array(folio, last_cookie, 0);
247 static void nfs_readdir_folio_array_free(struct folio *folio)
250 nfs_readdir_clear_array(folio);
255 static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array)
257 return array->size == 0 ? array->last_cookie : array->array[0].cookie;
260 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
262 array->folio_is_eof = 1;
263 array->folio_full = 1;
266 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
268 return array->folio_full;
272 * the caller is responsible for freeing qstr.name
273 * when called by nfs_readdir_add_to_array, the strings will be freed in
274 * nfs_clear_readdir_array()
276 static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
278 const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
281 * Avoid a kmemleak false positive. The pointer to the name is stored
282 * in a page cache page which kmemleak does not scan.
285 kmemleak_not_leak(ret);
289 static size_t nfs_readdir_array_maxentries(void)
291 return (PAGE_SIZE - sizeof(struct nfs_cache_array)) /
292 sizeof(struct nfs_cache_array_entry);
296 * Check that the next array entry lies entirely within the page bounds
298 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
300 if (array->folio_full)
302 if (array->size == nfs_readdir_array_maxentries()) {
303 array->folio_full = 1;
309 static int nfs_readdir_folio_array_append(struct folio *folio,
310 const struct nfs_entry *entry,
313 struct nfs_cache_array *array;
314 struct nfs_cache_array_entry *cache_entry;
318 name = nfs_readdir_copy_name(entry->name, entry->len);
320 array = kmap_local_folio(folio, 0);
323 ret = nfs_readdir_array_can_expand(array);
329 cache_entry = &array->array[array->size];
330 cache_entry->cookie = array->last_cookie;
331 cache_entry->ino = entry->ino;
332 cache_entry->d_type = entry->d_type;
333 cache_entry->name_len = entry->len;
334 cache_entry->name = name;
335 array->last_cookie = entry->cookie;
336 if (array->last_cookie <= cache_entry->cookie)
337 array->cookies_are_ordered = 0;
340 nfs_readdir_array_set_eof(array);
342 *cookie = array->last_cookie;
347 #define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14)
349 * Hash algorithm allowing content addressible access to sequences
350 * of directory cookies. Content is addressed by the value of the
351 * cookie index of the first readdir entry in a page.
353 * We select only the first 18 bits to avoid issues with excessive
354 * memory use for the page cache XArray. 18 bits should allow the caching
355 * of 262144 pages of sequences of readdir entries. Since each page holds
356 * 127 readdir entries for a typical 64-bit system, that works out to a
357 * cache of ~ 33 million entries per directory.
359 static pgoff_t nfs_readdir_folio_cookie_hash(u64 cookie)
363 return hash_64(cookie, 18);
366 static bool nfs_readdir_folio_validate(struct folio *folio, u64 last_cookie,
369 struct nfs_cache_array *array = kmap_local_folio(folio, 0);
372 if (array->change_attr != change_attr)
374 if (nfs_readdir_array_index_cookie(array) != last_cookie)
380 static void nfs_readdir_folio_unlock_and_put(struct folio *folio)
386 static void nfs_readdir_folio_init_and_validate(struct folio *folio, u64 cookie,
389 if (folio_test_uptodate(folio)) {
390 if (nfs_readdir_folio_validate(folio, cookie, change_attr))
392 nfs_readdir_clear_array(folio);
394 nfs_readdir_folio_init_array(folio, cookie, change_attr);
395 folio_mark_uptodate(folio);
398 static struct folio *nfs_readdir_folio_get_locked(struct address_space *mapping,
399 u64 cookie, u64 change_attr)
401 pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
404 folio = filemap_grab_folio(mapping, index);
407 nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
411 static u64 nfs_readdir_folio_last_cookie(struct folio *folio)
413 struct nfs_cache_array *array;
416 array = kmap_local_folio(folio, 0);
417 ret = array->last_cookie;
422 static bool nfs_readdir_folio_needs_filling(struct folio *folio)
424 struct nfs_cache_array *array;
427 array = kmap_local_folio(folio, 0);
428 ret = !nfs_readdir_array_is_full(array);
433 static void nfs_readdir_folio_set_eof(struct folio *folio)
435 struct nfs_cache_array *array;
437 array = kmap_local_folio(folio, 0);
438 nfs_readdir_array_set_eof(array);
442 static struct folio *nfs_readdir_folio_get_next(struct address_space *mapping,
443 u64 cookie, u64 change_attr)
445 pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
448 folio = __filemap_get_folio(mapping, index,
449 FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
450 mapping_gfp_mask(mapping));
453 nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
454 if (nfs_readdir_folio_last_cookie(folio) != cookie)
455 nfs_readdir_folio_reinit_array(folio, cookie, change_attr);
460 int is_32bit_api(void)
463 return in_compat_syscall();
465 return (BITS_PER_LONG == 32);
470 bool nfs_readdir_use_cookie(const struct file *filp)
472 if ((filp->f_mode & FMODE_32BITHASH) ||
473 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
478 static void nfs_readdir_seek_next_array(struct nfs_cache_array *array,
479 struct nfs_readdir_descriptor *desc)
481 if (array->folio_full) {
482 desc->last_cookie = array->last_cookie;
483 desc->current_index += array->size;
484 desc->cache_entry_index = 0;
487 desc->last_cookie = nfs_readdir_array_index_cookie(array);
490 static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc)
492 desc->current_index = 0;
493 desc->last_cookie = 0;
494 desc->folio_index = 0;
497 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
498 struct nfs_readdir_descriptor *desc)
500 loff_t diff = desc->ctx->pos - desc->current_index;
505 if (diff >= array->size) {
506 if (array->folio_is_eof)
508 nfs_readdir_seek_next_array(array, desc);
512 index = (unsigned int)diff;
513 desc->dir_cookie = array->array[index].cookie;
514 desc->cache_entry_index = index;
521 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
524 if (!array->cookies_are_ordered)
526 /* Optimisation for monotonically increasing cookies */
527 if (cookie >= array->last_cookie)
529 if (array->size && cookie < array->array[0].cookie)
534 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
535 struct nfs_readdir_descriptor *desc)
538 int status = -EAGAIN;
540 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
543 for (i = 0; i < array->size; i++) {
544 if (array->array[i].cookie == desc->dir_cookie) {
545 if (nfs_readdir_use_cookie(desc->file))
546 desc->ctx->pos = desc->dir_cookie;
548 desc->ctx->pos = desc->current_index + i;
549 desc->cache_entry_index = i;
554 if (array->folio_is_eof) {
555 status = -EBADCOOKIE;
556 if (desc->dir_cookie == array->last_cookie)
559 nfs_readdir_seek_next_array(array, desc);
563 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
565 struct nfs_cache_array *array;
568 array = kmap_local_folio(desc->folio, 0);
570 if (desc->dir_cookie == 0)
571 status = nfs_readdir_search_for_pos(array, desc);
573 status = nfs_readdir_search_for_cookie(array, desc);
579 /* Fill a page with xdr information before transferring to the cache page */
580 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
581 __be32 *verf, u64 cookie,
582 struct page **pages, size_t bufsize,
585 struct inode *inode = file_inode(desc->file);
586 struct nfs_readdir_arg arg = {
587 .dentry = file_dentry(desc->file),
588 .cred = desc->file->f_cred,
595 struct nfs_readdir_res res = {
598 unsigned long timestamp, gencount;
603 gencount = nfs_inc_attr_generation_counter();
604 desc->dir_verifier = nfs_save_change_attribute(inode);
605 error = NFS_PROTO(inode)->readdir(&arg, &res);
607 /* We requested READDIRPLUS, but the server doesn't grok it */
608 if (error == -ENOTSUPP && desc->plus) {
609 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
610 desc->plus = arg.plus = false;
615 desc->timestamp = timestamp;
616 desc->gencount = gencount;
621 static int xdr_decode(struct nfs_readdir_descriptor *desc,
622 struct nfs_entry *entry, struct xdr_stream *xdr)
624 struct inode *inode = file_inode(desc->file);
627 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
630 entry->fattr->time_start = desc->timestamp;
631 entry->fattr->gencount = desc->gencount;
635 /* Match file and dirent using either filehandle or fileid
636 * Note: caller is responsible for checking the fsid
639 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
642 struct nfs_inode *nfsi;
644 if (d_really_is_negative(dentry))
647 inode = d_inode(dentry);
648 if (is_bad_inode(inode) || NFS_STALE(inode))
652 if (entry->fattr->fileid != nfsi->fileid)
654 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
659 #define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL)
661 static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx,
662 unsigned int cache_hits,
663 unsigned int cache_misses)
665 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
668 cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD)
674 * This function is called by the getattr code to request the
675 * use of readdirplus to accelerate any future lookups in the same
678 void nfs_readdir_record_entry_cache_hit(struct inode *dir)
680 struct nfs_inode *nfsi = NFS_I(dir);
681 struct nfs_open_dir_context *ctx;
683 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
684 S_ISDIR(dir->i_mode)) {
686 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
687 atomic_inc(&ctx->cache_hits);
693 * This function is mainly for use by nfs_getattr().
695 * If this is an 'ls -l', we want to force use of readdirplus.
697 void nfs_readdir_record_entry_cache_miss(struct inode *dir)
699 struct nfs_inode *nfsi = NFS_I(dir);
700 struct nfs_open_dir_context *ctx;
702 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
703 S_ISDIR(dir->i_mode)) {
705 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
706 atomic_inc(&ctx->cache_misses);
711 static void nfs_lookup_advise_force_readdirplus(struct inode *dir,
714 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
716 if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL))
718 nfs_readdir_record_entry_cache_miss(dir);
722 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
723 unsigned long dir_verifier)
725 struct qstr filename = QSTR_INIT(entry->name, entry->len);
726 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
727 struct dentry *dentry;
728 struct dentry *alias;
732 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
734 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
736 if (filename.len == 0)
738 /* Validate that the name doesn't contain any illegal '\0' */
739 if (strnlen(filename.name, filename.len) != filename.len)
742 if (strnchr(filename.name, filename.len, '/'))
744 if (filename.name[0] == '.') {
745 if (filename.len == 1)
747 if (filename.len == 2 && filename.name[1] == '.')
750 filename.hash = full_name_hash(parent, filename.name, filename.len);
752 dentry = d_lookup(parent, &filename);
755 dentry = d_alloc_parallel(parent, &filename, &wq);
759 if (!d_in_lookup(dentry)) {
760 /* Is there a mountpoint here? If so, just exit */
761 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
762 &entry->fattr->fsid))
764 if (nfs_same_file(dentry, entry)) {
765 if (!entry->fh->size)
767 nfs_set_verifier(dentry, dir_verifier);
768 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
770 nfs_setsecurity(d_inode(dentry), entry->fattr);
771 trace_nfs_readdir_lookup_revalidate(d_inode(parent),
775 trace_nfs_readdir_lookup_revalidate_failed(
776 d_inode(parent), dentry, 0);
777 d_invalidate(dentry);
783 if (!entry->fh->size) {
784 d_lookup_done(dentry);
788 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
789 alias = d_splice_alias(inode, dentry);
790 d_lookup_done(dentry);
797 nfs_set_verifier(dentry, dir_verifier);
798 trace_nfs_readdir_lookup(d_inode(parent), dentry, 0);
803 static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc,
804 struct nfs_entry *entry,
805 struct xdr_stream *stream)
809 if (entry->fattr->label)
810 entry->fattr->label->len = NFS4_MAXLABELLEN;
811 ret = xdr_decode(desc, entry, stream);
812 if (ret || !desc->plus)
814 nfs_prime_dcache(file_dentry(desc->file), entry, desc->dir_verifier);
818 /* Perform conversion from xdr to cache array */
819 static int nfs_readdir_folio_filler(struct nfs_readdir_descriptor *desc,
820 struct nfs_entry *entry,
821 struct page **xdr_pages, unsigned int buflen,
822 struct folio **arrays, size_t narrays,
825 struct address_space *mapping = desc->file->f_mapping;
826 struct folio *new, *folio = *arrays;
827 struct xdr_stream stream;
828 struct page *scratch;
833 scratch = alloc_page(GFP_KERNEL);
837 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
838 xdr_set_scratch_page(&stream, scratch);
841 status = nfs_readdir_entry_decode(desc, entry, &stream);
845 status = nfs_readdir_folio_array_append(folio, entry, &cookie);
846 if (status != -ENOSPC)
849 if (folio->mapping != mapping) {
852 new = nfs_readdir_folio_array_alloc(cookie, GFP_KERNEL);
856 *arrays = folio = new;
858 new = nfs_readdir_folio_get_next(mapping, cookie,
862 if (folio != *arrays)
863 nfs_readdir_folio_unlock_and_put(folio);
866 desc->folio_index_max++;
867 status = nfs_readdir_folio_array_append(folio, entry, &cookie);
868 } while (!status && !entry->eof);
874 nfs_readdir_folio_set_eof(folio);
883 while (!nfs_readdir_entry_decode(desc, entry, &stream))
887 if (folio != *arrays)
888 nfs_readdir_folio_unlock_and_put(folio);
894 static void nfs_readdir_free_pages(struct page **pages, size_t npages)
897 put_page(pages[npages]);
902 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
903 * to nfs_readdir_free_pages()
905 static struct page **nfs_readdir_alloc_pages(size_t npages)
910 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
913 for (i = 0; i < npages; i++) {
914 struct page *page = alloc_page(GFP_KERNEL);
922 nfs_readdir_free_pages(pages, i);
926 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
927 __be32 *verf_arg, __be32 *verf_res,
928 struct folio **arrays, size_t narrays)
932 struct folio *folio = *arrays;
933 struct nfs_entry *entry;
935 struct inode *inode = file_inode(desc->file);
936 unsigned int dtsize = desc->dtsize;
938 int status = -ENOMEM;
940 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
943 entry->cookie = nfs_readdir_folio_last_cookie(folio);
944 entry->fh = nfs_alloc_fhandle();
945 entry->fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
946 entry->server = NFS_SERVER(inode);
947 if (entry->fh == NULL || entry->fattr == NULL)
950 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
951 pages = nfs_readdir_alloc_pages(array_size);
955 change_attr = inode_peek_iversion_raw(inode);
956 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, pages,
963 status = nfs_readdir_folio_filler(desc, entry, pages, pglen,
964 arrays, narrays, change_attr);
966 nfs_readdir_folio_set_eof(folio);
967 desc->buffer_fills++;
970 nfs_readdir_free_pages(pages, array_size);
972 nfs_free_fattr(entry->fattr);
973 nfs_free_fhandle(entry->fh);
978 static void nfs_readdir_folio_put(struct nfs_readdir_descriptor *desc)
980 folio_put(desc->folio);
985 nfs_readdir_folio_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
987 folio_unlock(desc->folio);
988 nfs_readdir_folio_put(desc);
991 static struct folio *
992 nfs_readdir_folio_get_cached(struct nfs_readdir_descriptor *desc)
994 struct address_space *mapping = desc->file->f_mapping;
995 u64 change_attr = inode_peek_iversion_raw(mapping->host);
996 u64 cookie = desc->last_cookie;
999 folio = nfs_readdir_folio_get_locked(mapping, cookie, change_attr);
1002 if (desc->clear_cache && !nfs_readdir_folio_needs_filling(folio))
1003 nfs_readdir_folio_reinit_array(folio, cookie, change_attr);
1008 * Returns 0 if desc->dir_cookie was found on page desc->page_index
1009 * and locks the page to prevent removal from the page cache.
1011 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
1013 struct inode *inode = file_inode(desc->file);
1014 struct nfs_inode *nfsi = NFS_I(inode);
1015 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1018 desc->folio = nfs_readdir_folio_get_cached(desc);
1021 if (nfs_readdir_folio_needs_filling(desc->folio)) {
1022 /* Grow the dtsize if we had to go back for more pages */
1023 if (desc->folio_index == desc->folio_index_max)
1024 nfs_grow_dtsize(desc);
1025 desc->folio_index_max = desc->folio_index;
1026 trace_nfs_readdir_cache_fill(desc->file, nfsi->cookieverf,
1028 desc->folio->index, desc->dtsize);
1029 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
1032 nfs_readdir_folio_unlock_and_put_cached(desc);
1033 trace_nfs_readdir_cache_fill_done(inode, res);
1034 if (res == -EBADCOOKIE || res == -ENOTSYNC) {
1035 invalidate_inode_pages2(desc->file->f_mapping);
1036 nfs_readdir_rewind_search(desc);
1037 trace_nfs_readdir_invalidate_cache_range(
1038 inode, 0, MAX_LFS_FILESIZE);
1044 * Set the cookie verifier if the page cache was empty
1046 if (desc->last_cookie == 0 &&
1047 memcmp(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf))) {
1048 memcpy(nfsi->cookieverf, verf,
1049 sizeof(nfsi->cookieverf));
1050 invalidate_inode_pages2_range(desc->file->f_mapping, 1,
1052 trace_nfs_readdir_invalidate_cache_range(
1053 inode, 1, MAX_LFS_FILESIZE);
1055 desc->clear_cache = false;
1057 res = nfs_readdir_search_array(desc);
1060 nfs_readdir_folio_unlock_and_put_cached(desc);
1064 /* Search for desc->dir_cookie from the beginning of the page cache */
1065 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
1070 res = find_and_lock_cache_page(desc);
1071 } while (res == -EAGAIN);
1075 #define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1078 * Once we've found the start of the dirent within a page: fill 'er up...
1080 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
1083 struct file *file = desc->file;
1084 struct nfs_cache_array *array;
1086 bool first_emit = !desc->dir_cookie;
1088 array = kmap_local_folio(desc->folio, 0);
1089 for (i = desc->cache_entry_index; i < array->size; i++) {
1090 struct nfs_cache_array_entry *ent;
1092 ent = &array->array[i];
1093 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
1094 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
1098 memcpy(desc->verf, verf, sizeof(desc->verf));
1099 if (i == array->size - 1) {
1100 desc->dir_cookie = array->last_cookie;
1101 nfs_readdir_seek_next_array(array, desc);
1103 desc->dir_cookie = array->array[i + 1].cookie;
1104 desc->last_cookie = array->array[0].cookie;
1106 if (nfs_readdir_use_cookie(file))
1107 desc->ctx->pos = desc->dir_cookie;
1110 if (first_emit && i > NFS_READDIR_CACHE_MISS_THRESHOLD + 1) {
1115 if (array->folio_is_eof)
1116 desc->eof = !desc->eob;
1118 kunmap_local(array);
1119 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1120 (unsigned long long)desc->dir_cookie);
1124 * If we cannot find a cookie in our cache, we suspect that this is
1125 * because it points to a deleted file, so we ask the server to return
1126 * whatever it thinks is the next entry. We then feed this to filldir.
1127 * If all goes well, we should then be able to find our way round the
1128 * cache on the next call to readdir_search_pagecache();
1130 * NOTE: we cannot add the anonymous page to the pagecache because
1131 * the data it contains might not be page aligned. Besides,
1132 * we should already have a complete representation of the
1133 * directory in the page cache by the time we get here.
1135 static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1137 struct folio **arrays;
1139 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1140 int status = -ENOMEM;
1142 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1143 (unsigned long long)desc->dir_cookie);
1145 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1148 arrays[0] = nfs_readdir_folio_array_alloc(desc->dir_cookie, GFP_KERNEL);
1152 desc->folio_index = 0;
1153 desc->cache_entry_index = 0;
1154 desc->last_cookie = desc->dir_cookie;
1155 desc->folio_index_max = 0;
1157 trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie,
1160 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1162 trace_nfs_readdir_uncached_done(file_inode(desc->file), status);
1166 for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1167 desc->folio = arrays[i];
1168 nfs_do_filldir(desc, verf);
1173 * Grow the dtsize if we have to go back for more pages,
1174 * or shrink it if we're reading too many.
1178 nfs_grow_dtsize(desc);
1179 else if (desc->buffer_fills == 1 &&
1180 i < (desc->folio_index_max >> 1))
1181 nfs_shrink_dtsize(desc);
1184 for (i = 0; i < sz && arrays[i]; i++)
1185 nfs_readdir_folio_array_free(arrays[i]);
1187 if (!nfs_readdir_use_cookie(desc->file))
1188 nfs_readdir_rewind_search(desc);
1189 desc->folio_index_max = -1;
1191 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1195 static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1196 struct nfs_readdir_descriptor *desc,
1197 unsigned int cache_misses,
1200 if (desc->ctx->pos == 0 || !desc->plus)
1202 if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1204 trace_nfs_readdir_force_readdirplus(inode);
1208 /* The file offset position represents the dirent entry number. A
1209 last cookie cache takes care of the common case of reading the
1212 static int nfs_readdir(struct file *file, struct dir_context *ctx)
1214 struct dentry *dentry = file_dentry(file);
1215 struct inode *inode = d_inode(dentry);
1216 struct nfs_inode *nfsi = NFS_I(inode);
1217 struct nfs_open_dir_context *dir_ctx = file->private_data;
1218 struct nfs_readdir_descriptor *desc;
1219 unsigned int cache_hits, cache_misses;
1223 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1224 file, (long long)ctx->pos);
1225 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1228 * ctx->pos points to the dirent entry number.
1229 * *desc->dir_cookie has the cookie for the next entry. We have
1230 * to either find the entry with the appropriate number or
1231 * revalidate the cookie.
1233 nfs_revalidate_mapping(inode, file->f_mapping);
1236 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1241 desc->folio_index_max = -1;
1243 spin_lock(&file->f_lock);
1244 desc->dir_cookie = dir_ctx->dir_cookie;
1245 desc->folio_index = dir_ctx->page_index;
1246 desc->last_cookie = dir_ctx->last_cookie;
1247 desc->attr_gencount = dir_ctx->attr_gencount;
1248 desc->eof = dir_ctx->eof;
1249 nfs_set_dtsize(desc, dir_ctx->dtsize);
1250 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1251 cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0);
1252 cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0);
1253 force_clear = dir_ctx->force_clear;
1254 spin_unlock(&file->f_lock);
1261 desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses);
1262 force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1264 desc->clear_cache = force_clear;
1267 res = readdir_search_pagecache(desc);
1269 if (res == -EBADCOOKIE) {
1271 /* This means either end of directory */
1272 if (desc->dir_cookie && !desc->eof) {
1273 /* Or that the server has 'lost' a cookie */
1274 res = uncached_readdir(desc);
1277 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1282 if (res == -ETOOSMALL && desc->plus) {
1283 nfs_zap_caches(inode);
1291 nfs_do_filldir(desc, nfsi->cookieverf);
1292 nfs_readdir_folio_unlock_and_put_cached(desc);
1293 if (desc->folio_index == desc->folio_index_max)
1294 desc->clear_cache = force_clear;
1295 } while (!desc->eob && !desc->eof);
1297 spin_lock(&file->f_lock);
1298 dir_ctx->dir_cookie = desc->dir_cookie;
1299 dir_ctx->last_cookie = desc->last_cookie;
1300 dir_ctx->attr_gencount = desc->attr_gencount;
1301 dir_ctx->page_index = desc->folio_index;
1302 dir_ctx->force_clear = force_clear;
1303 dir_ctx->eof = desc->eof;
1304 dir_ctx->dtsize = desc->dtsize;
1305 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1306 spin_unlock(&file->f_lock);
1311 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1315 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1317 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1319 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1320 filp, offset, whence);
1328 spin_lock(&filp->f_lock);
1333 spin_lock(&filp->f_lock);
1334 offset += filp->f_pos;
1336 spin_unlock(&filp->f_lock);
1340 if (offset != filp->f_pos) {
1341 filp->f_pos = offset;
1342 dir_ctx->page_index = 0;
1343 if (!nfs_readdir_use_cookie(filp)) {
1344 dir_ctx->dir_cookie = 0;
1345 dir_ctx->last_cookie = 0;
1347 dir_ctx->dir_cookie = offset;
1348 dir_ctx->last_cookie = offset;
1350 dir_ctx->eof = false;
1352 spin_unlock(&filp->f_lock);
1357 * All directory operations under NFS are synchronous, so fsync()
1358 * is a dummy operation.
1360 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1363 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1365 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1370 * nfs_force_lookup_revalidate - Mark the directory as having changed
1371 * @dir: pointer to directory inode
1373 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1374 * full lookup on all child dentries of 'dir' whenever a change occurs
1375 * on the server that might have invalidated our dcache.
1377 * Note that we reserve bit '0' as a tag to let us know when a dentry
1378 * was revalidated while holding a delegation on its inode.
1380 * The caller should be holding dir->i_lock
1382 void nfs_force_lookup_revalidate(struct inode *dir)
1384 NFS_I(dir)->cache_change_attribute += 2;
1386 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1389 * nfs_verify_change_attribute - Detects NFS remote directory changes
1390 * @dir: pointer to parent directory inode
1391 * @verf: previously saved change attribute
1393 * Return "false" if the verifiers doesn't match the change attribute.
1394 * This would usually indicate that the directory contents have changed on
1395 * the server, and that any dentries need revalidating.
1397 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1399 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1402 static void nfs_set_verifier_delegated(unsigned long *verf)
1407 #if IS_ENABLED(CONFIG_NFS_V4)
1408 static void nfs_unset_verifier_delegated(unsigned long *verf)
1412 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1414 static bool nfs_test_verifier_delegated(unsigned long verf)
1419 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1421 return nfs_test_verifier_delegated(dentry->d_time);
1424 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1426 struct inode *inode = d_inode(dentry);
1427 struct inode *dir = d_inode(dentry->d_parent);
1429 if (!nfs_verify_change_attribute(dir, verf))
1431 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1432 nfs_set_verifier_delegated(&verf);
1433 dentry->d_time = verf;
1437 * nfs_set_verifier - save a parent directory verifier in the dentry
1438 * @dentry: pointer to dentry
1439 * @verf: verifier to save
1441 * Saves the parent directory verifier in @dentry. If the inode has
1442 * a delegation, we also tag the dentry as having been revalidated
1443 * while holding a delegation so that we know we don't have to
1444 * look it up again after a directory change.
1446 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1449 spin_lock(&dentry->d_lock);
1450 nfs_set_verifier_locked(dentry, verf);
1451 spin_unlock(&dentry->d_lock);
1453 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1455 #if IS_ENABLED(CONFIG_NFS_V4)
1457 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1458 * @inode: pointer to inode
1460 * Iterates through the dentries in the inode alias list and clears
1461 * the tag used to indicate that the dentry has been revalidated
1462 * while holding a delegation.
1463 * This function is intended for use when the delegation is being
1464 * returned or revoked.
1466 void nfs_clear_verifier_delegated(struct inode *inode)
1468 struct dentry *alias;
1472 spin_lock(&inode->i_lock);
1473 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1474 spin_lock(&alias->d_lock);
1475 nfs_unset_verifier_delegated(&alias->d_time);
1476 spin_unlock(&alias->d_lock);
1478 spin_unlock(&inode->i_lock);
1480 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1481 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1483 static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1485 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) &&
1486 d_really_is_negative(dentry))
1487 return dentry->d_time == inode_peek_iversion_raw(dir);
1488 return nfs_verify_change_attribute(dir, dentry->d_time);
1492 * A check for whether or not the parent directory has changed.
1493 * In the case it has, we assume that the dentries are untrustworthy
1494 * and may need to be looked up again.
1495 * If rcu_walk prevents us from performing a full check, return 0.
1497 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1500 if (IS_ROOT(dentry))
1502 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1504 if (!nfs_dentry_verify_change(dir, dentry))
1506 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1507 if (nfs_mapping_need_revalidate_inode(dir)) {
1510 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1513 if (!nfs_dentry_verify_change(dir, dentry))
1519 * Use intent information to check whether or not we're going to do
1520 * an O_EXCL create using this path component.
1522 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1524 if (NFS_PROTO(dir)->version == 2)
1526 return flags & LOOKUP_EXCL;
1530 * Inode and filehandle revalidation for lookups.
1532 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1533 * or if the intent information indicates that we're about to open this
1534 * particular file and the "nocto" mount flag is not set.
1538 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1540 struct nfs_server *server = NFS_SERVER(inode);
1543 if (IS_AUTOMOUNT(inode))
1546 if (flags & LOOKUP_OPEN) {
1547 switch (inode->i_mode & S_IFMT) {
1549 /* A NFSv4 OPEN will revalidate later */
1550 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1554 if (server->flags & NFS_MOUNT_NOCTO)
1556 /* NFS close-to-open cache consistency validation */
1561 /* VFS wants an on-the-wire revalidation */
1562 if (flags & LOOKUP_REVAL)
1565 if (inode->i_nlink > 0 ||
1566 (inode->i_nlink == 0 &&
1567 test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1572 if (flags & LOOKUP_RCU)
1574 ret = __nfs_revalidate_inode(server, inode);
1580 static void nfs_mark_dir_for_revalidate(struct inode *inode)
1582 spin_lock(&inode->i_lock);
1583 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1584 spin_unlock(&inode->i_lock);
1588 * We judge how long we want to trust negative
1589 * dentries by looking at the parent inode mtime.
1591 * If parent mtime has changed, we revalidate, else we wait for a
1592 * period corresponding to the parent's attribute cache timeout value.
1594 * If LOOKUP_RCU prevents us from performing a full check, return 1
1595 * suggesting a reval is needed.
1597 * Note that when creating a new file, or looking up a rename target,
1598 * then it shouldn't be necessary to revalidate a negative dentry.
1601 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1604 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1606 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1608 /* Case insensitive server? Revalidate negative dentries */
1609 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1611 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1615 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1616 struct inode *inode, int error)
1623 * We can't d_drop the root of a disconnected tree:
1624 * its d_hash is on the s_anon list and d_drop() would hide
1625 * it from shrink_dcache_for_unmount(), leading to busy
1626 * inodes on unmount and further oopses.
1628 if (inode && IS_ROOT(dentry))
1632 trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error);
1637 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1641 if (nfs_neg_need_reval(dir, dentry, flags)) {
1642 if (flags & LOOKUP_RCU)
1646 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1650 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1651 struct inode *inode)
1653 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1654 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1657 static int nfs_lookup_revalidate_dentry(struct inode *dir,
1658 struct dentry *dentry,
1659 struct inode *inode, unsigned int flags)
1661 struct nfs_fh *fhandle;
1662 struct nfs_fattr *fattr;
1663 unsigned long dir_verifier;
1666 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1669 fhandle = nfs_alloc_fhandle();
1670 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1671 if (fhandle == NULL || fattr == NULL)
1674 dir_verifier = nfs_save_change_attribute(dir);
1675 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1683 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1689 /* Request help from readdirplus */
1690 nfs_lookup_advise_force_readdirplus(dir, flags);
1693 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1695 if (nfs_refresh_inode(inode, fattr) < 0)
1698 nfs_setsecurity(inode, fattr);
1699 nfs_set_verifier(dentry, dir_verifier);
1703 nfs_free_fattr(fattr);
1704 nfs_free_fhandle(fhandle);
1707 * If the lookup failed despite the dentry change attribute being
1708 * a match, then we should revalidate the directory cache.
1710 if (!ret && nfs_dentry_verify_change(dir, dentry))
1711 nfs_mark_dir_for_revalidate(dir);
1712 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1716 * This is called every time the dcache has a lookup hit,
1717 * and we should check whether we can really trust that
1720 * NOTE! The hit can be a negative hit too, don't assume
1723 * If the parent directory is seen to have changed, we throw out the
1724 * cached dentry and do a new lookup.
1727 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1730 struct inode *inode;
1733 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1734 inode = d_inode(dentry);
1737 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1739 if (is_bad_inode(inode)) {
1740 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1745 if ((flags & LOOKUP_RENAME_TARGET) && d_count(dentry) < 2 &&
1746 nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1749 if (nfs_verifier_is_delegated(dentry))
1750 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1752 /* Force a full look up iff the parent directory has changed */
1753 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1754 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1755 error = nfs_lookup_verify_inode(inode, flags);
1757 if (error == -ESTALE)
1758 nfs_mark_dir_for_revalidate(dir);
1764 if (flags & LOOKUP_RCU)
1767 if (NFS_STALE(inode))
1770 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
1772 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1774 if (flags & LOOKUP_RCU)
1776 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1780 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1781 int (*reval)(struct inode *, struct dentry *, unsigned int))
1783 struct dentry *parent;
1787 if (flags & LOOKUP_RCU) {
1788 if (dentry->d_fsdata == NFS_FSDATA_BLOCKED)
1790 parent = READ_ONCE(dentry->d_parent);
1791 dir = d_inode_rcu(parent);
1794 ret = reval(dir, dentry, flags);
1795 if (parent != READ_ONCE(dentry->d_parent))
1798 /* Wait for unlink to complete */
1799 wait_var_event(&dentry->d_fsdata,
1800 dentry->d_fsdata != NFS_FSDATA_BLOCKED);
1801 parent = dget_parent(dentry);
1802 ret = reval(d_inode(parent), dentry, flags);
1808 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1810 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1814 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1815 * when we don't really care about the dentry name. This is called when a
1816 * pathwalk ends on a dentry that was not found via a normal lookup in the
1817 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1819 * In this situation, we just want to verify that the inode itself is OK
1820 * since the dentry might have changed on the server.
1822 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1824 struct inode *inode = d_inode(dentry);
1828 * I believe we can only get a negative dentry here in the case of a
1829 * procfs-style symlink. Just assume it's correct for now, but we may
1830 * eventually need to do something more here.
1833 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1838 if (is_bad_inode(inode)) {
1839 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1844 error = nfs_lookup_verify_inode(inode, flags);
1845 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1846 __func__, inode->i_ino, error ? "invalid" : "valid");
1851 * This is called from dput() when d_count is going to 0.
1853 static int nfs_dentry_delete(const struct dentry *dentry)
1855 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1856 dentry, dentry->d_flags);
1858 /* Unhash any dentry with a stale inode */
1859 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1862 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1863 /* Unhash it, so that ->d_iput() would be called */
1866 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1867 /* Unhash it, so that ancestors of killed async unlink
1868 * files will be cleaned up during umount */
1875 /* Ensure that we revalidate inode->i_nlink */
1876 static void nfs_drop_nlink(struct inode *inode)
1878 spin_lock(&inode->i_lock);
1879 /* drop the inode if we're reasonably sure this is the last link */
1880 if (inode->i_nlink > 0)
1882 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1883 nfs_set_cache_invalid(
1884 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1885 NFS_INO_INVALID_NLINK);
1886 spin_unlock(&inode->i_lock);
1890 * Called when the dentry loses inode.
1891 * We use it to clean up silly-renamed files.
1893 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1895 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1896 nfs_complete_unlink(dentry, inode);
1897 nfs_drop_nlink(inode);
1902 static void nfs_d_release(struct dentry *dentry)
1904 /* free cached devname value, if it survived that far */
1905 if (unlikely(dentry->d_fsdata)) {
1906 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1909 kfree(dentry->d_fsdata);
1913 const struct dentry_operations nfs_dentry_operations = {
1914 .d_revalidate = nfs_lookup_revalidate,
1915 .d_weak_revalidate = nfs_weak_revalidate,
1916 .d_delete = nfs_dentry_delete,
1917 .d_iput = nfs_dentry_iput,
1918 .d_automount = nfs_d_automount,
1919 .d_release = nfs_d_release,
1921 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1923 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1926 struct inode *inode = NULL;
1927 struct nfs_fh *fhandle = NULL;
1928 struct nfs_fattr *fattr = NULL;
1929 unsigned long dir_verifier;
1932 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1933 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1935 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1936 return ERR_PTR(-ENAMETOOLONG);
1939 * If we're doing an exclusive create, optimize away the lookup
1940 * but don't hash the dentry.
1942 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1945 res = ERR_PTR(-ENOMEM);
1946 fhandle = nfs_alloc_fhandle();
1947 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir));
1948 if (fhandle == NULL || fattr == NULL)
1951 dir_verifier = nfs_save_change_attribute(dir);
1952 trace_nfs_lookup_enter(dir, dentry, flags);
1953 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1954 if (error == -ENOENT) {
1955 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1956 dir_verifier = inode_peek_iversion_raw(dir);
1960 res = ERR_PTR(error);
1963 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1964 res = ERR_CAST(inode);
1968 /* Notify readdir to use READDIRPLUS */
1969 nfs_lookup_advise_force_readdirplus(dir, flags);
1972 res = d_splice_alias(inode, dentry);
1978 nfs_set_verifier(dentry, dir_verifier);
1980 trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res));
1981 nfs_free_fattr(fattr);
1982 nfs_free_fhandle(fhandle);
1985 EXPORT_SYMBOL_GPL(nfs_lookup);
1987 void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
1989 /* Case insensitive server? Revalidate dentries */
1990 if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
1991 d_prune_aliases(inode);
1993 EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
1995 #if IS_ENABLED(CONFIG_NFS_V4)
1996 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1998 const struct dentry_operations nfs4_dentry_operations = {
1999 .d_revalidate = nfs4_lookup_revalidate,
2000 .d_weak_revalidate = nfs_weak_revalidate,
2001 .d_delete = nfs_dentry_delete,
2002 .d_iput = nfs_dentry_iput,
2003 .d_automount = nfs_d_automount,
2004 .d_release = nfs_d_release,
2006 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
2008 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
2010 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
2013 static int do_open(struct inode *inode, struct file *filp)
2015 nfs_fscache_open_file(inode, filp);
2019 static int nfs_finish_open(struct nfs_open_context *ctx,
2020 struct dentry *dentry,
2021 struct file *file, unsigned open_flags)
2025 err = finish_open(file, dentry, do_open);
2028 if (S_ISREG(file_inode(file)->i_mode))
2029 nfs_file_set_open_context(file, ctx);
2036 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2037 struct file *file, unsigned open_flags,
2040 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2041 struct nfs_open_context *ctx;
2043 struct iattr attr = { .ia_valid = ATTR_OPEN };
2044 struct inode *inode;
2045 unsigned int lookup_flags = 0;
2046 unsigned long dir_verifier;
2047 bool switched = false;
2051 /* Expect a negative dentry */
2052 BUG_ON(d_inode(dentry));
2054 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2055 dir->i_sb->s_id, dir->i_ino, dentry);
2057 err = nfs_check_flags(open_flags);
2061 /* NFS only supports OPEN on regular files */
2062 if ((open_flags & O_DIRECTORY)) {
2063 if (!d_in_lookup(dentry)) {
2065 * Hashed negative dentry with O_DIRECTORY: dentry was
2066 * revalidated and is fine, no need to perform lookup
2071 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2075 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
2076 return -ENAMETOOLONG;
2078 if (open_flags & O_CREAT) {
2079 struct nfs_server *server = NFS_SERVER(dir);
2081 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2082 mode &= ~current_umask();
2084 attr.ia_valid |= ATTR_MODE;
2085 attr.ia_mode = mode;
2087 if (open_flags & O_TRUNC) {
2088 attr.ia_valid |= ATTR_SIZE;
2092 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2095 dentry = d_alloc_parallel(dentry->d_parent,
2096 &dentry->d_name, &wq);
2098 return PTR_ERR(dentry);
2099 if (unlikely(!d_in_lookup(dentry)))
2100 return finish_no_open(file, dentry);
2103 ctx = create_nfs_open_context(dentry, open_flags, file);
2108 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
2109 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
2111 file->f_mode |= FMODE_CREATED;
2112 if (IS_ERR(inode)) {
2113 err = PTR_ERR(inode);
2114 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2115 put_nfs_open_context(ctx);
2119 d_splice_alias(NULL, dentry);
2120 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
2121 dir_verifier = inode_peek_iversion_raw(dir);
2123 dir_verifier = nfs_save_change_attribute(dir);
2124 nfs_set_verifier(dentry, dir_verifier);
2130 if (!(open_flags & O_NOFOLLOW))
2139 file->f_mode |= FMODE_CAN_ODIRECT;
2141 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
2142 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2143 put_nfs_open_context(ctx);
2145 if (unlikely(switched)) {
2146 d_lookup_done(dentry);
2152 res = nfs_lookup(dir, dentry, lookup_flags);
2154 inode = d_inode(dentry);
2155 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2156 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2157 res = ERR_PTR(-ENOTDIR);
2158 else if (inode && S_ISREG(inode->i_mode))
2159 res = ERR_PTR(-EOPENSTALE);
2160 } else if (!IS_ERR(res)) {
2161 inode = d_inode(res);
2162 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2163 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2165 res = ERR_PTR(-ENOTDIR);
2166 } else if (inode && S_ISREG(inode->i_mode)) {
2168 res = ERR_PTR(-EOPENSTALE);
2172 d_lookup_done(dentry);
2179 return PTR_ERR(res);
2180 return finish_no_open(file, res);
2182 EXPORT_SYMBOL_GPL(nfs_atomic_open);
2185 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2188 struct inode *inode;
2190 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2192 if (d_mountpoint(dentry))
2195 inode = d_inode(dentry);
2197 /* We can't create new files in nfs_open_revalidate(), so we
2198 * optimize away revalidation of negative dentries.
2203 if (nfs_verifier_is_delegated(dentry))
2204 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2206 /* NFS only supports OPEN on regular files */
2207 if (!S_ISREG(inode->i_mode))
2210 /* We cannot do exclusive creation on a positive dentry */
2211 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2214 /* Check if the directory changed */
2215 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2218 /* Let f_op->open() actually open (and revalidate) the file */
2221 if (flags & LOOKUP_RCU)
2223 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
2226 return nfs_do_lookup_revalidate(dir, dentry, flags);
2229 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2231 return __nfs_lookup_revalidate(dentry, flags,
2232 nfs4_do_lookup_revalidate);
2235 #endif /* CONFIG_NFSV4 */
2238 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2239 struct nfs_fattr *fattr)
2241 struct dentry *parent = dget_parent(dentry);
2242 struct inode *dir = d_inode(parent);
2243 struct inode *inode;
2249 if (fhandle->size == 0) {
2250 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
2254 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2255 if (!(fattr->valid & NFS_ATTR_FATTR)) {
2256 struct nfs_server *server = NFS_SB(dentry->d_sb);
2257 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2262 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2263 d = d_splice_alias(inode, dentry);
2271 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2274 * Code common to create, mkdir, and mknod.
2276 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2277 struct nfs_fattr *fattr)
2281 d = nfs_add_or_obtain(dentry, fhandle, fattr);
2285 /* Callers don't care */
2289 EXPORT_SYMBOL_GPL(nfs_instantiate);
2292 * Following a failed create operation, we drop the dentry rather
2293 * than retain a negative dentry. This avoids a problem in the event
2294 * that the operation succeeded on the server, but an error in the
2295 * reply path made it appear to have failed.
2297 int nfs_create(struct mnt_idmap *idmap, struct inode *dir,
2298 struct dentry *dentry, umode_t mode, bool excl)
2301 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2304 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2305 dir->i_sb->s_id, dir->i_ino, dentry);
2307 attr.ia_mode = mode;
2308 attr.ia_valid = ATTR_MODE;
2310 trace_nfs_create_enter(dir, dentry, open_flags);
2311 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2312 trace_nfs_create_exit(dir, dentry, open_flags, error);
2320 EXPORT_SYMBOL_GPL(nfs_create);
2323 * See comments for nfs_proc_create regarding failed operations.
2326 nfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
2327 struct dentry *dentry, umode_t mode, dev_t rdev)
2332 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2333 dir->i_sb->s_id, dir->i_ino, dentry);
2335 attr.ia_mode = mode;
2336 attr.ia_valid = ATTR_MODE;
2338 trace_nfs_mknod_enter(dir, dentry);
2339 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2340 trace_nfs_mknod_exit(dir, dentry, status);
2348 EXPORT_SYMBOL_GPL(nfs_mknod);
2351 * See comments for nfs_proc_create regarding failed operations.
2353 int nfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2354 struct dentry *dentry, umode_t mode)
2359 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2360 dir->i_sb->s_id, dir->i_ino, dentry);
2362 attr.ia_valid = ATTR_MODE;
2363 attr.ia_mode = mode | S_IFDIR;
2365 trace_nfs_mkdir_enter(dir, dentry);
2366 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2367 trace_nfs_mkdir_exit(dir, dentry, error);
2375 EXPORT_SYMBOL_GPL(nfs_mkdir);
2377 static void nfs_dentry_handle_enoent(struct dentry *dentry)
2379 if (simple_positive(dentry))
2383 static void nfs_dentry_remove_handle_error(struct inode *dir,
2384 struct dentry *dentry, int error)
2388 if (d_really_is_positive(dentry))
2390 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2393 nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2394 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2398 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2402 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2403 dir->i_sb->s_id, dir->i_ino, dentry);
2405 trace_nfs_rmdir_enter(dir, dentry);
2406 if (d_really_is_positive(dentry)) {
2407 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2408 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2409 /* Ensure the VFS deletes this inode */
2412 clear_nlink(d_inode(dentry));
2415 nfs_dentry_handle_enoent(dentry);
2417 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2419 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2420 nfs_dentry_remove_handle_error(dir, dentry, error);
2421 trace_nfs_rmdir_exit(dir, dentry, error);
2425 EXPORT_SYMBOL_GPL(nfs_rmdir);
2428 * Remove a file after making sure there are no pending writes,
2429 * and after checking that the file has only one user.
2431 * We invalidate the attribute cache and free the inode prior to the operation
2432 * to avoid possible races if the server reuses the inode.
2434 static int nfs_safe_remove(struct dentry *dentry)
2436 struct inode *dir = d_inode(dentry->d_parent);
2437 struct inode *inode = d_inode(dentry);
2440 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2442 /* If the dentry was sillyrenamed, we simply call d_delete() */
2443 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2448 trace_nfs_remove_enter(dir, dentry);
2449 if (inode != NULL) {
2450 error = NFS_PROTO(dir)->remove(dir, dentry);
2452 nfs_drop_nlink(inode);
2454 error = NFS_PROTO(dir)->remove(dir, dentry);
2455 if (error == -ENOENT)
2456 nfs_dentry_handle_enoent(dentry);
2457 trace_nfs_remove_exit(dir, dentry, error);
2462 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2463 * belongs to an active ".nfs..." file and we return -EBUSY.
2465 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2467 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2471 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2472 dir->i_ino, dentry);
2474 trace_nfs_unlink_enter(dir, dentry);
2475 spin_lock(&dentry->d_lock);
2476 if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2477 &NFS_I(d_inode(dentry))->flags)) {
2478 spin_unlock(&dentry->d_lock);
2479 /* Start asynchronous writeout of the inode */
2480 write_inode_now(d_inode(dentry), 0);
2481 error = nfs_sillyrename(dir, dentry);
2484 /* We must prevent any concurrent open until the unlink
2485 * completes. ->d_revalidate will wait for ->d_fsdata
2486 * to clear. We set it here to ensure no lookup succeeds until
2487 * the unlink is complete on the server.
2490 if (WARN_ON(dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2491 WARN_ON(dentry->d_fsdata == NFS_FSDATA_BLOCKED)) {
2492 spin_unlock(&dentry->d_lock);
2496 kfree(dentry->d_fsdata);
2497 dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2499 spin_unlock(&dentry->d_lock);
2500 error = nfs_safe_remove(dentry);
2501 nfs_dentry_remove_handle_error(dir, dentry, error);
2502 dentry->d_fsdata = NULL;
2503 wake_up_var(&dentry->d_fsdata);
2505 trace_nfs_unlink_exit(dir, dentry, error);
2508 EXPORT_SYMBOL_GPL(nfs_unlink);
2511 * To create a symbolic link, most file systems instantiate a new inode,
2512 * add a page to it containing the path, then write it out to the disk
2513 * using prepare_write/commit_write.
2515 * Unfortunately the NFS client can't create the in-core inode first
2516 * because it needs a file handle to create an in-core inode (see
2517 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2518 * symlink request has completed on the server.
2520 * So instead we allocate a raw page, copy the symname into it, then do
2521 * the SYMLINK request with the page as the buffer. If it succeeds, we
2522 * now have a new file handle and can instantiate an in-core NFS inode
2523 * and move the raw page into its mapping.
2525 int nfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
2526 struct dentry *dentry, const char *symname)
2531 unsigned int pathlen = strlen(symname);
2534 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2535 dir->i_ino, dentry, symname);
2537 if (pathlen > PAGE_SIZE)
2538 return -ENAMETOOLONG;
2540 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2541 attr.ia_valid = ATTR_MODE;
2543 page = alloc_page(GFP_USER);
2547 kaddr = page_address(page);
2548 memcpy(kaddr, symname, pathlen);
2549 if (pathlen < PAGE_SIZE)
2550 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2552 trace_nfs_symlink_enter(dir, dentry);
2553 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2554 trace_nfs_symlink_exit(dir, dentry, error);
2556 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2557 dir->i_sb->s_id, dir->i_ino,
2558 dentry, symname, error);
2564 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2567 * No big deal if we can't add this page to the page cache here.
2568 * READLINK will get the missing page from the server if needed.
2570 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2572 SetPageUptodate(page);
2575 * add_to_page_cache_lru() grabs an extra page refcount.
2576 * Drop it here to avoid leaking this page later.
2584 EXPORT_SYMBOL_GPL(nfs_symlink);
2587 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2589 struct inode *inode = d_inode(old_dentry);
2592 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2593 old_dentry, dentry);
2595 trace_nfs_link_enter(inode, dir, dentry);
2597 if (S_ISREG(inode->i_mode))
2598 nfs_sync_inode(inode);
2599 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2601 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2603 d_add(dentry, inode);
2605 trace_nfs_link_exit(inode, dir, dentry, error);
2608 EXPORT_SYMBOL_GPL(nfs_link);
2611 nfs_unblock_rename(struct rpc_task *task, struct nfs_renamedata *data)
2613 struct dentry *new_dentry = data->new_dentry;
2615 new_dentry->d_fsdata = NULL;
2616 wake_up_var(&new_dentry->d_fsdata);
2621 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2622 * different file handle for the same inode after a rename (e.g. when
2623 * moving to a different directory). A fail-safe method to do so would
2624 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2625 * rename the old file using the sillyrename stuff. This way, the original
2626 * file in old_dir will go away when the last process iput()s the inode.
2630 * It actually works quite well. One needs to have the possibility for
2631 * at least one ".nfs..." file in each directory the file ever gets
2632 * moved or linked to which happens automagically with the new
2633 * implementation that only depends on the dcache stuff instead of
2634 * using the inode layer
2636 * Unfortunately, things are a little more complicated than indicated
2637 * above. For a cross-directory move, we want to make sure we can get
2638 * rid of the old inode after the operation. This means there must be
2639 * no pending writes (if it's a file), and the use count must be 1.
2640 * If these conditions are met, we can drop the dentries before doing
2643 int nfs_rename(struct mnt_idmap *idmap, struct inode *old_dir,
2644 struct dentry *old_dentry, struct inode *new_dir,
2645 struct dentry *new_dentry, unsigned int flags)
2647 struct inode *old_inode = d_inode(old_dentry);
2648 struct inode *new_inode = d_inode(new_dentry);
2649 struct dentry *dentry = NULL;
2650 struct rpc_task *task;
2651 bool must_unblock = false;
2657 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2658 old_dentry, new_dentry,
2659 d_count(new_dentry));
2661 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2663 * For non-directories, check whether the target is busy and if so,
2664 * make a copy of the dentry and then do a silly-rename. If the
2665 * silly-rename succeeds, the copied dentry is hashed and becomes
2668 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2669 /* We must prevent any concurrent open until the unlink
2670 * completes. ->d_revalidate will wait for ->d_fsdata
2671 * to clear. We set it here to ensure no lookup succeeds until
2672 * the unlink is complete on the server.
2675 if (WARN_ON(new_dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2676 WARN_ON(new_dentry->d_fsdata == NFS_FSDATA_BLOCKED))
2678 if (new_dentry->d_fsdata) {
2680 kfree(new_dentry->d_fsdata);
2681 new_dentry->d_fsdata = NULL;
2684 spin_lock(&new_dentry->d_lock);
2685 if (d_count(new_dentry) > 2) {
2688 spin_unlock(&new_dentry->d_lock);
2690 /* copy the target dentry's name */
2691 dentry = d_alloc(new_dentry->d_parent,
2692 &new_dentry->d_name);
2696 /* silly-rename the existing target ... */
2697 err = nfs_sillyrename(new_dir, new_dentry);
2701 new_dentry = dentry;
2704 new_dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2705 must_unblock = true;
2706 spin_unlock(&new_dentry->d_lock);
2711 if (S_ISREG(old_inode->i_mode))
2712 nfs_sync_inode(old_inode);
2713 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry,
2714 must_unblock ? nfs_unblock_rename : NULL);
2716 error = PTR_ERR(task);
2720 error = rpc_wait_for_completion_task(task);
2722 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2723 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2726 error = task->tk_status;
2728 /* Ensure the inode attributes are revalidated */
2730 spin_lock(&old_inode->i_lock);
2731 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2732 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2733 NFS_INO_INVALID_CTIME |
2734 NFS_INO_REVAL_FORCED);
2735 spin_unlock(&old_inode->i_lock);
2738 trace_nfs_rename_exit(old_dir, old_dentry,
2739 new_dir, new_dentry, error);
2741 if (new_inode != NULL)
2742 nfs_drop_nlink(new_inode);
2744 * The d_move() should be here instead of in an async RPC completion
2745 * handler because we need the proper locks to move the dentry. If
2746 * we're interrupted by a signal, the async RPC completion handler
2747 * should mark the directories for revalidation.
2749 d_move(old_dentry, new_dentry);
2750 nfs_set_verifier(old_dentry,
2751 nfs_save_change_attribute(new_dir));
2752 } else if (error == -ENOENT)
2753 nfs_dentry_handle_enoent(old_dentry);
2755 /* new dentry created? */
2760 EXPORT_SYMBOL_GPL(nfs_rename);
2762 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2763 static LIST_HEAD(nfs_access_lru_list);
2764 static atomic_long_t nfs_access_nr_entries;
2766 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2767 module_param(nfs_access_max_cachesize, ulong, 0644);
2768 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2770 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2772 put_group_info(entry->group_info);
2773 kfree_rcu(entry, rcu_head);
2774 smp_mb__before_atomic();
2775 atomic_long_dec(&nfs_access_nr_entries);
2776 smp_mb__after_atomic();
2779 static void nfs_access_free_list(struct list_head *head)
2781 struct nfs_access_entry *cache;
2783 while (!list_empty(head)) {
2784 cache = list_entry(head->next, struct nfs_access_entry, lru);
2785 list_del(&cache->lru);
2786 nfs_access_free_entry(cache);
2790 static unsigned long
2791 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2794 struct nfs_inode *nfsi, *next;
2795 struct nfs_access_entry *cache;
2798 spin_lock(&nfs_access_lru_lock);
2799 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2800 struct inode *inode;
2802 if (nr_to_scan-- == 0)
2804 inode = &nfsi->vfs_inode;
2805 spin_lock(&inode->i_lock);
2806 if (list_empty(&nfsi->access_cache_entry_lru))
2807 goto remove_lru_entry;
2808 cache = list_entry(nfsi->access_cache_entry_lru.next,
2809 struct nfs_access_entry, lru);
2810 list_move(&cache->lru, &head);
2811 rb_erase(&cache->rb_node, &nfsi->access_cache);
2813 if (!list_empty(&nfsi->access_cache_entry_lru))
2814 list_move_tail(&nfsi->access_cache_inode_lru,
2815 &nfs_access_lru_list);
2818 list_del_init(&nfsi->access_cache_inode_lru);
2819 smp_mb__before_atomic();
2820 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2821 smp_mb__after_atomic();
2823 spin_unlock(&inode->i_lock);
2825 spin_unlock(&nfs_access_lru_lock);
2826 nfs_access_free_list(&head);
2831 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2833 int nr_to_scan = sc->nr_to_scan;
2834 gfp_t gfp_mask = sc->gfp_mask;
2836 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2838 return nfs_do_access_cache_scan(nr_to_scan);
2843 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2845 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2849 nfs_access_cache_enforce_limit(void)
2851 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2853 unsigned int nr_to_scan;
2855 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2858 diff = nr_entries - nfs_access_max_cachesize;
2859 if (diff < nr_to_scan)
2861 nfs_do_access_cache_scan(nr_to_scan);
2864 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2866 struct rb_root *root_node = &nfsi->access_cache;
2868 struct nfs_access_entry *entry;
2870 /* Unhook entries from the cache */
2871 while ((n = rb_first(root_node)) != NULL) {
2872 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2873 rb_erase(n, root_node);
2874 list_move(&entry->lru, head);
2876 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2879 void nfs_access_zap_cache(struct inode *inode)
2883 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2885 /* Remove from global LRU init */
2886 spin_lock(&nfs_access_lru_lock);
2887 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2888 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2890 spin_lock(&inode->i_lock);
2891 __nfs_access_zap_cache(NFS_I(inode), &head);
2892 spin_unlock(&inode->i_lock);
2893 spin_unlock(&nfs_access_lru_lock);
2894 nfs_access_free_list(&head);
2896 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2898 static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2900 struct group_info *ga, *gb;
2903 if (uid_lt(a->fsuid, b->fsuid))
2905 if (uid_gt(a->fsuid, b->fsuid))
2908 if (gid_lt(a->fsgid, b->fsgid))
2910 if (gid_gt(a->fsgid, b->fsgid))
2921 if (ga->ngroups < gb->ngroups)
2923 if (ga->ngroups > gb->ngroups)
2926 for (g = 0; g < ga->ngroups; g++) {
2927 if (gid_lt(ga->gid[g], gb->gid[g]))
2929 if (gid_gt(ga->gid[g], gb->gid[g]))
2935 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2937 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2940 struct nfs_access_entry *entry =
2941 rb_entry(n, struct nfs_access_entry, rb_node);
2942 int cmp = access_cmp(cred, entry);
2954 static u64 nfs_access_login_time(const struct task_struct *task,
2955 const struct cred *cred)
2957 const struct task_struct *parent;
2958 const struct cred *pcred;
2963 parent = rcu_dereference(task->real_parent);
2964 pcred = rcu_dereference(parent->cred);
2965 if (parent == task || cred_fscmp(pcred, cred) != 0)
2969 ret = task->start_time;
2974 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
2976 struct nfs_inode *nfsi = NFS_I(inode);
2977 u64 login_time = nfs_access_login_time(current, cred);
2978 struct nfs_access_entry *cache;
2982 spin_lock(&inode->i_lock);
2984 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2986 cache = nfs_access_search_rbtree(inode, cred);
2990 /* Found an entry, is our attribute cache valid? */
2991 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2998 spin_unlock(&inode->i_lock);
2999 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
3002 spin_lock(&inode->i_lock);
3006 if ((s64)(login_time - cache->timestamp) > 0)
3008 *mask = cache->mask;
3009 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
3012 spin_unlock(&inode->i_lock);
3015 spin_unlock(&inode->i_lock);
3016 nfs_access_zap_cache(inode);
3020 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
3022 /* Only check the most recently returned cache entry,
3023 * but do it without locking.
3025 struct nfs_inode *nfsi = NFS_I(inode);
3026 u64 login_time = nfs_access_login_time(current, cred);
3027 struct nfs_access_entry *cache;
3029 struct list_head *lh;
3032 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3034 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
3035 cache = list_entry(lh, struct nfs_access_entry, lru);
3036 if (lh == &nfsi->access_cache_entry_lru ||
3037 access_cmp(cred, cache) != 0)
3041 if ((s64)(login_time - cache->timestamp) > 0)
3043 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3045 *mask = cache->mask;
3052 int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
3053 u32 *mask, bool may_block)
3057 status = nfs_access_get_cached_rcu(inode, cred, mask);
3059 status = nfs_access_get_cached_locked(inode, cred, mask,
3064 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
3066 static void nfs_access_add_rbtree(struct inode *inode,
3067 struct nfs_access_entry *set,
3068 const struct cred *cred)
3070 struct nfs_inode *nfsi = NFS_I(inode);
3071 struct rb_root *root_node = &nfsi->access_cache;
3072 struct rb_node **p = &root_node->rb_node;
3073 struct rb_node *parent = NULL;
3074 struct nfs_access_entry *entry;
3077 spin_lock(&inode->i_lock);
3078 while (*p != NULL) {
3080 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3081 cmp = access_cmp(cred, entry);
3084 p = &parent->rb_left;
3086 p = &parent->rb_right;
3090 rb_link_node(&set->rb_node, parent, p);
3091 rb_insert_color(&set->rb_node, root_node);
3092 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3093 spin_unlock(&inode->i_lock);
3096 rb_replace_node(parent, &set->rb_node, root_node);
3097 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3098 list_del(&entry->lru);
3099 spin_unlock(&inode->i_lock);
3100 nfs_access_free_entry(entry);
3103 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3104 const struct cred *cred)
3106 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
3109 RB_CLEAR_NODE(&cache->rb_node);
3110 cache->fsuid = cred->fsuid;
3111 cache->fsgid = cred->fsgid;
3112 cache->group_info = get_group_info(cred->group_info);
3113 cache->mask = set->mask;
3114 cache->timestamp = ktime_get_ns();
3116 /* The above field assignments must be visible
3117 * before this item appears on the lru. We cannot easily
3118 * use rcu_assign_pointer, so just force the memory barrier.
3121 nfs_access_add_rbtree(inode, cache, cred);
3123 /* Update accounting */
3124 smp_mb__before_atomic();
3125 atomic_long_inc(&nfs_access_nr_entries);
3126 smp_mb__after_atomic();
3128 /* Add inode to global LRU list */
3129 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3130 spin_lock(&nfs_access_lru_lock);
3131 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
3132 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
3133 &nfs_access_lru_list);
3134 spin_unlock(&nfs_access_lru_lock);
3136 nfs_access_cache_enforce_limit();
3138 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3140 #define NFS_MAY_READ (NFS_ACCESS_READ)
3141 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3142 NFS_ACCESS_EXTEND | \
3144 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3146 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3147 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3148 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3150 nfs_access_calc_mask(u32 access_result, umode_t umode)
3154 if (access_result & NFS_MAY_READ)
3156 if (S_ISDIR(umode)) {
3157 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3159 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3161 } else if (S_ISREG(umode)) {
3162 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3164 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3166 } else if (access_result & NFS_MAY_WRITE)
3171 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3173 entry->mask = access_result;
3175 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3177 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3179 struct nfs_access_entry cache;
3180 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3181 int cache_mask = -1;
3184 trace_nfs_access_enter(inode);
3186 status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3195 * Determine which access bits we want to ask for...
3197 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3198 nfs_access_xattr_mask(NFS_SERVER(inode));
3199 if (S_ISDIR(inode->i_mode))
3200 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3202 cache.mask |= NFS_ACCESS_EXECUTE;
3203 status = NFS_PROTO(inode)->access(inode, &cache, cred);
3205 if (status == -ESTALE) {
3206 if (!S_ISDIR(inode->i_mode))
3207 nfs_set_inode_stale(inode);
3209 nfs_zap_caches(inode);
3213 nfs_access_add_cache(inode, &cache, cred);
3215 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
3216 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3219 trace_nfs_access_exit(inode, mask, cache_mask, status);
3223 static int nfs_open_permission_mask(int openflags)
3227 if (openflags & __FMODE_EXEC) {
3228 /* ONLY check exec rights */
3231 if ((openflags & O_ACCMODE) != O_WRONLY)
3233 if ((openflags & O_ACCMODE) != O_RDONLY)
3240 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3242 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
3244 EXPORT_SYMBOL_GPL(nfs_may_open);
3246 static int nfs_execute_ok(struct inode *inode, int mask)
3248 struct nfs_server *server = NFS_SERVER(inode);
3251 if (S_ISDIR(inode->i_mode))
3253 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3254 if (mask & MAY_NOT_BLOCK)
3256 ret = __nfs_revalidate_inode(server, inode);
3258 if (ret == 0 && !execute_ok(inode))
3263 int nfs_permission(struct mnt_idmap *idmap,
3264 struct inode *inode,
3267 const struct cred *cred = current_cred();
3270 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
3272 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3274 /* Is this sys_access() ? */
3275 if (mask & (MAY_ACCESS | MAY_CHDIR))
3278 switch (inode->i_mode & S_IFMT) {
3282 if ((mask & MAY_OPEN) &&
3283 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3288 * Optimize away all write operations, since the server
3289 * will check permissions when we perform the op.
3291 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3296 if (!NFS_PROTO(inode)->access)
3299 res = nfs_do_access(inode, cred, mask);
3301 if (!res && (mask & MAY_EXEC))
3302 res = nfs_execute_ok(inode, mask);
3304 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3305 inode->i_sb->s_id, inode->i_ino, mask, res);
3308 if (mask & MAY_NOT_BLOCK)
3311 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3312 NFS_INO_INVALID_OTHER);
3314 res = generic_permission(&nop_mnt_idmap, inode, mask);
3317 EXPORT_SYMBOL_GPL(nfs_permission);