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 *);
59 static int nfs_do_create(struct inode *dir, struct dentry *dentry,
60 umode_t mode, int open_flags);
62 const struct file_operations nfs_dir_operations = {
63 .llseek = nfs_llseek_dir,
64 .read = generic_read_dir,
65 .iterate_shared = nfs_readdir,
67 .release = nfs_closedir,
68 .fsync = nfs_fsync_dir,
71 const struct address_space_operations nfs_dir_aops = {
72 .free_folio = nfs_readdir_clear_array,
75 #define NFS_INIT_DTSIZE PAGE_SIZE
77 static struct nfs_open_dir_context *
78 alloc_nfs_open_dir_context(struct inode *dir)
80 struct nfs_inode *nfsi = NFS_I(dir);
81 struct nfs_open_dir_context *ctx;
83 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
85 ctx->attr_gencount = nfsi->attr_gencount;
86 ctx->dtsize = NFS_INIT_DTSIZE;
87 spin_lock(&dir->i_lock);
88 if (list_empty(&nfsi->open_files) &&
89 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
90 nfs_set_cache_invalid(dir,
91 NFS_INO_INVALID_DATA |
92 NFS_INO_REVAL_FORCED);
93 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
94 memcpy(ctx->verf, nfsi->cookieverf, sizeof(ctx->verf));
95 spin_unlock(&dir->i_lock);
98 return ERR_PTR(-ENOMEM);
101 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
103 spin_lock(&dir->i_lock);
104 list_del_rcu(&ctx->list);
105 spin_unlock(&dir->i_lock);
106 kfree_rcu(ctx, rcu_head);
113 nfs_opendir(struct inode *inode, struct file *filp)
116 struct nfs_open_dir_context *ctx;
118 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
120 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
122 ctx = alloc_nfs_open_dir_context(inode);
127 filp->private_data = ctx;
133 nfs_closedir(struct inode *inode, struct file *filp)
135 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
139 struct nfs_cache_array_entry {
143 unsigned int name_len;
144 unsigned char d_type;
147 struct nfs_cache_array {
151 unsigned char folio_full : 1,
153 cookies_are_ordered : 1;
154 struct nfs_cache_array_entry array[];
157 struct nfs_readdir_descriptor {
160 struct dir_context *ctx;
162 pgoff_t folio_index_max;
165 loff_t current_index;
167 __be32 verf[NFS_DIR_VERIFIER_SIZE];
168 unsigned long dir_verifier;
169 unsigned long timestamp;
170 unsigned long gencount;
171 unsigned long attr_gencount;
172 unsigned int cache_entry_index;
173 unsigned int buffer_fills;
181 static void nfs_set_dtsize(struct nfs_readdir_descriptor *desc, unsigned int sz)
183 struct nfs_server *server = NFS_SERVER(file_inode(desc->file));
184 unsigned int maxsize = server->dtsize;
188 if (sz < NFS_MIN_FILE_IO_SIZE)
189 sz = NFS_MIN_FILE_IO_SIZE;
193 static void nfs_shrink_dtsize(struct nfs_readdir_descriptor *desc)
195 nfs_set_dtsize(desc, desc->dtsize >> 1);
198 static void nfs_grow_dtsize(struct nfs_readdir_descriptor *desc)
200 nfs_set_dtsize(desc, desc->dtsize << 1);
203 static void nfs_readdir_folio_init_array(struct folio *folio, u64 last_cookie,
206 struct nfs_cache_array *array;
208 array = kmap_local_folio(folio, 0);
209 array->change_attr = change_attr;
210 array->last_cookie = last_cookie;
212 array->folio_full = 0;
213 array->folio_is_eof = 0;
214 array->cookies_are_ordered = 1;
219 * we are freeing strings created by nfs_add_to_readdir_array()
221 static void nfs_readdir_clear_array(struct folio *folio)
223 struct nfs_cache_array *array;
226 array = kmap_local_folio(folio, 0);
227 for (i = 0; i < array->size; i++)
228 kfree(array->array[i].name);
233 static void nfs_readdir_folio_reinit_array(struct folio *folio, u64 last_cookie,
236 nfs_readdir_clear_array(folio);
237 nfs_readdir_folio_init_array(folio, last_cookie, change_attr);
240 static struct folio *
241 nfs_readdir_folio_array_alloc(u64 last_cookie, gfp_t gfp_flags)
243 struct folio *folio = folio_alloc(gfp_flags, 0);
245 nfs_readdir_folio_init_array(folio, last_cookie, 0);
249 static void nfs_readdir_folio_array_free(struct folio *folio)
252 nfs_readdir_clear_array(folio);
257 static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array)
259 return array->size == 0 ? array->last_cookie : array->array[0].cookie;
262 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
264 array->folio_is_eof = 1;
265 array->folio_full = 1;
268 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
270 return array->folio_full;
274 * the caller is responsible for freeing qstr.name
275 * when called by nfs_readdir_add_to_array, the strings will be freed in
276 * nfs_clear_readdir_array()
278 static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
280 const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
283 * Avoid a kmemleak false positive. The pointer to the name is stored
284 * in a page cache page which kmemleak does not scan.
287 kmemleak_not_leak(ret);
291 static size_t nfs_readdir_array_maxentries(void)
293 return (PAGE_SIZE - sizeof(struct nfs_cache_array)) /
294 sizeof(struct nfs_cache_array_entry);
298 * Check that the next array entry lies entirely within the page bounds
300 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
302 if (array->folio_full)
304 if (array->size == nfs_readdir_array_maxentries()) {
305 array->folio_full = 1;
311 static int nfs_readdir_folio_array_append(struct folio *folio,
312 const struct nfs_entry *entry,
315 struct nfs_cache_array *array;
316 struct nfs_cache_array_entry *cache_entry;
320 name = nfs_readdir_copy_name(entry->name, entry->len);
322 array = kmap_local_folio(folio, 0);
325 ret = nfs_readdir_array_can_expand(array);
331 cache_entry = &array->array[array->size];
332 cache_entry->cookie = array->last_cookie;
333 cache_entry->ino = entry->ino;
334 cache_entry->d_type = entry->d_type;
335 cache_entry->name_len = entry->len;
336 cache_entry->name = name;
337 array->last_cookie = entry->cookie;
338 if (array->last_cookie <= cache_entry->cookie)
339 array->cookies_are_ordered = 0;
342 nfs_readdir_array_set_eof(array);
344 *cookie = array->last_cookie;
349 #define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14)
351 * Hash algorithm allowing content addressible access to sequences
352 * of directory cookies. Content is addressed by the value of the
353 * cookie index of the first readdir entry in a page.
355 * We select only the first 18 bits to avoid issues with excessive
356 * memory use for the page cache XArray. 18 bits should allow the caching
357 * of 262144 pages of sequences of readdir entries. Since each page holds
358 * 127 readdir entries for a typical 64-bit system, that works out to a
359 * cache of ~ 33 million entries per directory.
361 static pgoff_t nfs_readdir_folio_cookie_hash(u64 cookie)
365 return hash_64(cookie, 18);
368 static bool nfs_readdir_folio_validate(struct folio *folio, u64 last_cookie,
371 struct nfs_cache_array *array = kmap_local_folio(folio, 0);
374 if (array->change_attr != change_attr)
376 if (nfs_readdir_array_index_cookie(array) != last_cookie)
382 static void nfs_readdir_folio_unlock_and_put(struct folio *folio)
388 static void nfs_readdir_folio_init_and_validate(struct folio *folio, u64 cookie,
391 if (folio_test_uptodate(folio)) {
392 if (nfs_readdir_folio_validate(folio, cookie, change_attr))
394 nfs_readdir_clear_array(folio);
396 nfs_readdir_folio_init_array(folio, cookie, change_attr);
397 folio_mark_uptodate(folio);
400 static struct folio *nfs_readdir_folio_get_locked(struct address_space *mapping,
401 u64 cookie, u64 change_attr)
403 pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
406 folio = filemap_grab_folio(mapping, index);
409 nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
413 static u64 nfs_readdir_folio_last_cookie(struct folio *folio)
415 struct nfs_cache_array *array;
418 array = kmap_local_folio(folio, 0);
419 ret = array->last_cookie;
424 static bool nfs_readdir_folio_needs_filling(struct folio *folio)
426 struct nfs_cache_array *array;
429 array = kmap_local_folio(folio, 0);
430 ret = !nfs_readdir_array_is_full(array);
435 static void nfs_readdir_folio_set_eof(struct folio *folio)
437 struct nfs_cache_array *array;
439 array = kmap_local_folio(folio, 0);
440 nfs_readdir_array_set_eof(array);
444 static struct folio *nfs_readdir_folio_get_next(struct address_space *mapping,
445 u64 cookie, u64 change_attr)
447 pgoff_t index = nfs_readdir_folio_cookie_hash(cookie);
450 folio = __filemap_get_folio(mapping, index,
451 FGP_LOCK|FGP_CREAT|FGP_NOFS|FGP_NOWAIT,
452 mapping_gfp_mask(mapping));
455 nfs_readdir_folio_init_and_validate(folio, cookie, change_attr);
456 if (nfs_readdir_folio_last_cookie(folio) != cookie)
457 nfs_readdir_folio_reinit_array(folio, cookie, change_attr);
462 int is_32bit_api(void)
465 return in_compat_syscall();
467 return (BITS_PER_LONG == 32);
472 bool nfs_readdir_use_cookie(const struct file *filp)
474 if ((filp->f_mode & FMODE_32BITHASH) ||
475 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
480 static void nfs_readdir_seek_next_array(struct nfs_cache_array *array,
481 struct nfs_readdir_descriptor *desc)
483 if (array->folio_full) {
484 desc->last_cookie = array->last_cookie;
485 desc->current_index += array->size;
486 desc->cache_entry_index = 0;
489 desc->last_cookie = nfs_readdir_array_index_cookie(array);
492 static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc)
494 desc->current_index = 0;
495 desc->last_cookie = 0;
496 desc->folio_index = 0;
499 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
500 struct nfs_readdir_descriptor *desc)
502 loff_t diff = desc->ctx->pos - desc->current_index;
507 if (diff >= array->size) {
508 if (array->folio_is_eof)
510 nfs_readdir_seek_next_array(array, desc);
514 index = (unsigned int)diff;
515 desc->dir_cookie = array->array[index].cookie;
516 desc->cache_entry_index = index;
523 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
526 if (!array->cookies_are_ordered)
528 /* Optimisation for monotonically increasing cookies */
529 if (cookie >= array->last_cookie)
531 if (array->size && cookie < array->array[0].cookie)
536 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
537 struct nfs_readdir_descriptor *desc)
540 int status = -EAGAIN;
542 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
545 for (i = 0; i < array->size; i++) {
546 if (array->array[i].cookie == desc->dir_cookie) {
547 if (nfs_readdir_use_cookie(desc->file))
548 desc->ctx->pos = desc->dir_cookie;
550 desc->ctx->pos = desc->current_index + i;
551 desc->cache_entry_index = i;
556 if (array->folio_is_eof) {
557 status = -EBADCOOKIE;
558 if (desc->dir_cookie == array->last_cookie)
561 nfs_readdir_seek_next_array(array, desc);
565 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
567 struct nfs_cache_array *array;
570 array = kmap_local_folio(desc->folio, 0);
572 if (desc->dir_cookie == 0)
573 status = nfs_readdir_search_for_pos(array, desc);
575 status = nfs_readdir_search_for_cookie(array, desc);
581 /* Fill a page with xdr information before transferring to the cache page */
582 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
583 __be32 *verf, u64 cookie,
584 struct page **pages, size_t bufsize,
587 struct inode *inode = file_inode(desc->file);
588 struct nfs_readdir_arg arg = {
589 .dentry = file_dentry(desc->file),
590 .cred = desc->file->f_cred,
597 struct nfs_readdir_res res = {
600 unsigned long timestamp, gencount;
605 gencount = nfs_inc_attr_generation_counter();
606 desc->dir_verifier = nfs_save_change_attribute(inode);
607 error = NFS_PROTO(inode)->readdir(&arg, &res);
609 /* We requested READDIRPLUS, but the server doesn't grok it */
610 if (error == -ENOTSUPP && desc->plus) {
611 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
612 desc->plus = arg.plus = false;
617 desc->timestamp = timestamp;
618 desc->gencount = gencount;
623 static int xdr_decode(struct nfs_readdir_descriptor *desc,
624 struct nfs_entry *entry, struct xdr_stream *xdr)
626 struct inode *inode = file_inode(desc->file);
629 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
632 entry->fattr->time_start = desc->timestamp;
633 entry->fattr->gencount = desc->gencount;
637 /* Match file and dirent using either filehandle or fileid
638 * Note: caller is responsible for checking the fsid
641 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
644 struct nfs_inode *nfsi;
646 if (d_really_is_negative(dentry))
649 inode = d_inode(dentry);
650 if (is_bad_inode(inode) || NFS_STALE(inode))
654 if (entry->fattr->fileid != nfsi->fileid)
656 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
661 #define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL)
663 static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx,
664 unsigned int cache_hits,
665 unsigned int cache_misses)
667 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
670 cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD)
676 * This function is called by the getattr code to request the
677 * use of readdirplus to accelerate any future lookups in the same
680 void nfs_readdir_record_entry_cache_hit(struct inode *dir)
682 struct nfs_inode *nfsi = NFS_I(dir);
683 struct nfs_open_dir_context *ctx;
685 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
686 S_ISDIR(dir->i_mode)) {
688 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
689 atomic_inc(&ctx->cache_hits);
695 * This function is mainly for use by nfs_getattr().
697 * If this is an 'ls -l', we want to force use of readdirplus.
699 void nfs_readdir_record_entry_cache_miss(struct inode *dir)
701 struct nfs_inode *nfsi = NFS_I(dir);
702 struct nfs_open_dir_context *ctx;
704 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
705 S_ISDIR(dir->i_mode)) {
707 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
708 atomic_inc(&ctx->cache_misses);
713 static void nfs_lookup_advise_force_readdirplus(struct inode *dir,
716 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
718 if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL))
720 nfs_readdir_record_entry_cache_miss(dir);
724 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
725 unsigned long dir_verifier)
727 struct qstr filename = QSTR_INIT(entry->name, entry->len);
728 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
729 struct dentry *dentry;
730 struct dentry *alias;
734 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
736 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
738 if (filename.len == 0)
740 /* Validate that the name doesn't contain any illegal '\0' */
741 if (strnlen(filename.name, filename.len) != filename.len)
744 if (strnchr(filename.name, filename.len, '/'))
746 if (filename.name[0] == '.') {
747 if (filename.len == 1)
749 if (filename.len == 2 && filename.name[1] == '.')
752 filename.hash = full_name_hash(parent, filename.name, filename.len);
754 dentry = d_lookup(parent, &filename);
757 dentry = d_alloc_parallel(parent, &filename, &wq);
761 if (!d_in_lookup(dentry)) {
762 /* Is there a mountpoint here? If so, just exit */
763 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
764 &entry->fattr->fsid))
766 if (nfs_same_file(dentry, entry)) {
767 if (!entry->fh->size)
769 nfs_set_verifier(dentry, dir_verifier);
770 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
772 nfs_setsecurity(d_inode(dentry), entry->fattr);
773 trace_nfs_readdir_lookup_revalidate(d_inode(parent),
777 trace_nfs_readdir_lookup_revalidate_failed(
778 d_inode(parent), dentry, 0);
779 d_invalidate(dentry);
785 if (!entry->fh->size) {
786 d_lookup_done(dentry);
790 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
791 alias = d_splice_alias(inode, dentry);
792 d_lookup_done(dentry);
799 nfs_set_verifier(dentry, dir_verifier);
800 trace_nfs_readdir_lookup(d_inode(parent), dentry, 0);
805 static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc,
806 struct nfs_entry *entry,
807 struct xdr_stream *stream)
811 if (entry->fattr->label)
812 entry->fattr->label->len = NFS4_MAXLABELLEN;
813 ret = xdr_decode(desc, entry, stream);
814 if (ret || !desc->plus)
816 nfs_prime_dcache(file_dentry(desc->file), entry, desc->dir_verifier);
820 /* Perform conversion from xdr to cache array */
821 static int nfs_readdir_folio_filler(struct nfs_readdir_descriptor *desc,
822 struct nfs_entry *entry,
823 struct page **xdr_pages, unsigned int buflen,
824 struct folio **arrays, size_t narrays,
827 struct address_space *mapping = desc->file->f_mapping;
828 struct folio *new, *folio = *arrays;
829 struct xdr_stream stream;
830 struct page *scratch;
835 scratch = alloc_page(GFP_KERNEL);
839 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
840 xdr_set_scratch_page(&stream, scratch);
843 status = nfs_readdir_entry_decode(desc, entry, &stream);
847 status = nfs_readdir_folio_array_append(folio, entry, &cookie);
848 if (status != -ENOSPC)
851 if (folio->mapping != mapping) {
854 new = nfs_readdir_folio_array_alloc(cookie, GFP_KERNEL);
858 *arrays = folio = new;
860 new = nfs_readdir_folio_get_next(mapping, cookie,
864 if (folio != *arrays)
865 nfs_readdir_folio_unlock_and_put(folio);
868 desc->folio_index_max++;
869 status = nfs_readdir_folio_array_append(folio, entry, &cookie);
870 } while (!status && !entry->eof);
876 nfs_readdir_folio_set_eof(folio);
885 while (!nfs_readdir_entry_decode(desc, entry, &stream))
889 if (folio != *arrays)
890 nfs_readdir_folio_unlock_and_put(folio);
896 static void nfs_readdir_free_pages(struct page **pages, size_t npages)
899 put_page(pages[npages]);
904 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
905 * to nfs_readdir_free_pages()
907 static struct page **nfs_readdir_alloc_pages(size_t npages)
912 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
915 for (i = 0; i < npages; i++) {
916 struct page *page = alloc_page(GFP_KERNEL);
924 nfs_readdir_free_pages(pages, i);
928 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
929 __be32 *verf_arg, __be32 *verf_res,
930 struct folio **arrays, size_t narrays)
934 struct folio *folio = *arrays;
935 struct nfs_entry *entry;
937 struct inode *inode = file_inode(desc->file);
938 unsigned int dtsize = desc->dtsize;
940 int status = -ENOMEM;
942 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
945 entry->cookie = nfs_readdir_folio_last_cookie(folio);
946 entry->fh = nfs_alloc_fhandle();
947 entry->fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
948 entry->server = NFS_SERVER(inode);
949 if (entry->fh == NULL || entry->fattr == NULL)
952 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
953 pages = nfs_readdir_alloc_pages(array_size);
957 change_attr = inode_peek_iversion_raw(inode);
958 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, pages,
965 status = nfs_readdir_folio_filler(desc, entry, pages, pglen,
966 arrays, narrays, change_attr);
968 nfs_readdir_folio_set_eof(folio);
969 desc->buffer_fills++;
972 nfs_readdir_free_pages(pages, array_size);
974 nfs_free_fattr(entry->fattr);
975 nfs_free_fhandle(entry->fh);
980 static void nfs_readdir_folio_put(struct nfs_readdir_descriptor *desc)
982 folio_put(desc->folio);
987 nfs_readdir_folio_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
989 folio_unlock(desc->folio);
990 nfs_readdir_folio_put(desc);
993 static struct folio *
994 nfs_readdir_folio_get_cached(struct nfs_readdir_descriptor *desc)
996 struct address_space *mapping = desc->file->f_mapping;
997 u64 change_attr = inode_peek_iversion_raw(mapping->host);
998 u64 cookie = desc->last_cookie;
1001 folio = nfs_readdir_folio_get_locked(mapping, cookie, change_attr);
1004 if (desc->clear_cache && !nfs_readdir_folio_needs_filling(folio))
1005 nfs_readdir_folio_reinit_array(folio, cookie, change_attr);
1010 * Returns 0 if desc->dir_cookie was found on page desc->page_index
1011 * and locks the page to prevent removal from the page cache.
1013 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
1015 struct inode *inode = file_inode(desc->file);
1016 struct nfs_inode *nfsi = NFS_I(inode);
1017 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1020 desc->folio = nfs_readdir_folio_get_cached(desc);
1023 if (nfs_readdir_folio_needs_filling(desc->folio)) {
1024 /* Grow the dtsize if we had to go back for more pages */
1025 if (desc->folio_index == desc->folio_index_max)
1026 nfs_grow_dtsize(desc);
1027 desc->folio_index_max = desc->folio_index;
1028 trace_nfs_readdir_cache_fill(desc->file, nfsi->cookieverf,
1030 desc->folio->index, desc->dtsize);
1031 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
1034 nfs_readdir_folio_unlock_and_put_cached(desc);
1035 trace_nfs_readdir_cache_fill_done(inode, res);
1036 if (res == -EBADCOOKIE || res == -ENOTSYNC) {
1037 invalidate_inode_pages2(desc->file->f_mapping);
1038 nfs_readdir_rewind_search(desc);
1039 trace_nfs_readdir_invalidate_cache_range(
1040 inode, 0, MAX_LFS_FILESIZE);
1046 * Set the cookie verifier if the page cache was empty
1048 if (desc->last_cookie == 0 &&
1049 memcmp(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf))) {
1050 memcpy(nfsi->cookieverf, verf,
1051 sizeof(nfsi->cookieverf));
1052 invalidate_inode_pages2_range(desc->file->f_mapping, 1,
1054 trace_nfs_readdir_invalidate_cache_range(
1055 inode, 1, MAX_LFS_FILESIZE);
1057 desc->clear_cache = false;
1059 res = nfs_readdir_search_array(desc);
1062 nfs_readdir_folio_unlock_and_put_cached(desc);
1066 /* Search for desc->dir_cookie from the beginning of the page cache */
1067 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
1072 res = find_and_lock_cache_page(desc);
1073 } while (res == -EAGAIN);
1077 #define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1080 * Once we've found the start of the dirent within a page: fill 'er up...
1082 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
1085 struct file *file = desc->file;
1086 struct nfs_cache_array *array;
1088 bool first_emit = !desc->dir_cookie;
1090 array = kmap_local_folio(desc->folio, 0);
1091 for (i = desc->cache_entry_index; i < array->size; i++) {
1092 struct nfs_cache_array_entry *ent;
1095 * nfs_readdir_handle_cache_misses return force clear at
1096 * (cache_misses > NFS_READDIR_CACHE_MISS_THRESHOLD) for
1097 * readdir heuristic, NFS_READDIR_CACHE_MISS_THRESHOLD + 1
1098 * entries need be emitted here.
1100 if (first_emit && i > NFS_READDIR_CACHE_MISS_THRESHOLD + 2) {
1105 ent = &array->array[i];
1106 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
1107 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
1111 memcpy(desc->verf, verf, sizeof(desc->verf));
1112 if (i == array->size - 1) {
1113 desc->dir_cookie = array->last_cookie;
1114 nfs_readdir_seek_next_array(array, desc);
1116 desc->dir_cookie = array->array[i + 1].cookie;
1117 desc->last_cookie = array->array[0].cookie;
1119 if (nfs_readdir_use_cookie(file))
1120 desc->ctx->pos = desc->dir_cookie;
1124 if (array->folio_is_eof)
1125 desc->eof = !desc->eob;
1127 kunmap_local(array);
1128 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1129 (unsigned long long)desc->dir_cookie);
1133 * If we cannot find a cookie in our cache, we suspect that this is
1134 * because it points to a deleted file, so we ask the server to return
1135 * whatever it thinks is the next entry. We then feed this to filldir.
1136 * If all goes well, we should then be able to find our way round the
1137 * cache on the next call to readdir_search_pagecache();
1139 * NOTE: we cannot add the anonymous page to the pagecache because
1140 * the data it contains might not be page aligned. Besides,
1141 * we should already have a complete representation of the
1142 * directory in the page cache by the time we get here.
1144 static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1146 struct folio **arrays;
1148 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1149 int status = -ENOMEM;
1151 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1152 (unsigned long long)desc->dir_cookie);
1154 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1157 arrays[0] = nfs_readdir_folio_array_alloc(desc->dir_cookie, GFP_KERNEL);
1161 desc->folio_index = 0;
1162 desc->cache_entry_index = 0;
1163 desc->last_cookie = desc->dir_cookie;
1164 desc->folio_index_max = 0;
1166 trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie,
1169 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1171 trace_nfs_readdir_uncached_done(file_inode(desc->file), status);
1175 for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1176 desc->folio = arrays[i];
1177 nfs_do_filldir(desc, verf);
1182 * Grow the dtsize if we have to go back for more pages,
1183 * or shrink it if we're reading too many.
1187 nfs_grow_dtsize(desc);
1188 else if (desc->buffer_fills == 1 &&
1189 i < (desc->folio_index_max >> 1))
1190 nfs_shrink_dtsize(desc);
1193 for (i = 0; i < sz && arrays[i]; i++)
1194 nfs_readdir_folio_array_free(arrays[i]);
1196 if (!nfs_readdir_use_cookie(desc->file))
1197 nfs_readdir_rewind_search(desc);
1198 desc->folio_index_max = -1;
1200 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1204 static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1205 struct nfs_readdir_descriptor *desc,
1206 unsigned int cache_misses,
1209 if (desc->ctx->pos == 0 || !desc->plus)
1211 if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1213 trace_nfs_readdir_force_readdirplus(inode);
1217 /* The file offset position represents the dirent entry number. A
1218 last cookie cache takes care of the common case of reading the
1221 static int nfs_readdir(struct file *file, struct dir_context *ctx)
1223 struct dentry *dentry = file_dentry(file);
1224 struct inode *inode = d_inode(dentry);
1225 struct nfs_inode *nfsi = NFS_I(inode);
1226 struct nfs_open_dir_context *dir_ctx = file->private_data;
1227 struct nfs_readdir_descriptor *desc;
1228 unsigned int cache_hits, cache_misses;
1232 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1233 file, (long long)ctx->pos);
1234 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1237 * ctx->pos points to the dirent entry number.
1238 * *desc->dir_cookie has the cookie for the next entry. We have
1239 * to either find the entry with the appropriate number or
1240 * revalidate the cookie.
1242 nfs_revalidate_mapping(inode, file->f_mapping);
1245 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1250 desc->folio_index_max = -1;
1252 spin_lock(&file->f_lock);
1253 desc->dir_cookie = dir_ctx->dir_cookie;
1254 desc->folio_index = dir_ctx->page_index;
1255 desc->last_cookie = dir_ctx->last_cookie;
1256 desc->attr_gencount = dir_ctx->attr_gencount;
1257 desc->eof = dir_ctx->eof;
1258 nfs_set_dtsize(desc, dir_ctx->dtsize);
1259 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1260 cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0);
1261 cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0);
1262 force_clear = dir_ctx->force_clear;
1263 spin_unlock(&file->f_lock);
1270 desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses);
1271 force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1273 desc->clear_cache = force_clear;
1276 res = readdir_search_pagecache(desc);
1278 if (res == -EBADCOOKIE) {
1280 /* This means either end of directory */
1281 if (desc->dir_cookie && !desc->eof) {
1282 /* Or that the server has 'lost' a cookie */
1283 res = uncached_readdir(desc);
1286 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1291 if (res == -ETOOSMALL && desc->plus) {
1292 nfs_zap_caches(inode);
1300 nfs_do_filldir(desc, nfsi->cookieverf);
1301 nfs_readdir_folio_unlock_and_put_cached(desc);
1302 if (desc->folio_index == desc->folio_index_max)
1303 desc->clear_cache = force_clear;
1304 } while (!desc->eob && !desc->eof);
1306 spin_lock(&file->f_lock);
1307 dir_ctx->dir_cookie = desc->dir_cookie;
1308 dir_ctx->last_cookie = desc->last_cookie;
1309 dir_ctx->attr_gencount = desc->attr_gencount;
1310 dir_ctx->page_index = desc->folio_index;
1311 dir_ctx->force_clear = force_clear;
1312 dir_ctx->eof = desc->eof;
1313 dir_ctx->dtsize = desc->dtsize;
1314 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1315 spin_unlock(&file->f_lock);
1320 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1324 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1326 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1328 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1329 filp, offset, whence);
1337 spin_lock(&filp->f_lock);
1342 spin_lock(&filp->f_lock);
1343 offset += filp->f_pos;
1345 spin_unlock(&filp->f_lock);
1349 if (offset != filp->f_pos) {
1350 filp->f_pos = offset;
1351 dir_ctx->page_index = 0;
1352 if (!nfs_readdir_use_cookie(filp)) {
1353 dir_ctx->dir_cookie = 0;
1354 dir_ctx->last_cookie = 0;
1356 dir_ctx->dir_cookie = offset;
1357 dir_ctx->last_cookie = offset;
1359 dir_ctx->eof = false;
1361 spin_unlock(&filp->f_lock);
1366 * All directory operations under NFS are synchronous, so fsync()
1367 * is a dummy operation.
1369 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1372 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1374 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1379 * nfs_force_lookup_revalidate - Mark the directory as having changed
1380 * @dir: pointer to directory inode
1382 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1383 * full lookup on all child dentries of 'dir' whenever a change occurs
1384 * on the server that might have invalidated our dcache.
1386 * Note that we reserve bit '0' as a tag to let us know when a dentry
1387 * was revalidated while holding a delegation on its inode.
1389 * The caller should be holding dir->i_lock
1391 void nfs_force_lookup_revalidate(struct inode *dir)
1393 NFS_I(dir)->cache_change_attribute += 2;
1395 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1398 * nfs_verify_change_attribute - Detects NFS remote directory changes
1399 * @dir: pointer to parent directory inode
1400 * @verf: previously saved change attribute
1402 * Return "false" if the verifiers doesn't match the change attribute.
1403 * This would usually indicate that the directory contents have changed on
1404 * the server, and that any dentries need revalidating.
1406 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1408 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1411 static void nfs_set_verifier_delegated(unsigned long *verf)
1416 #if IS_ENABLED(CONFIG_NFS_V4)
1417 static void nfs_unset_verifier_delegated(unsigned long *verf)
1421 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1423 static bool nfs_test_verifier_delegated(unsigned long verf)
1428 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1430 return nfs_test_verifier_delegated(dentry->d_time);
1433 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1435 struct inode *inode = d_inode(dentry);
1436 struct inode *dir = d_inode_rcu(dentry->d_parent);
1438 if (!dir || !nfs_verify_change_attribute(dir, verf))
1440 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1441 nfs_set_verifier_delegated(&verf);
1442 dentry->d_time = verf;
1446 * nfs_set_verifier - save a parent directory verifier in the dentry
1447 * @dentry: pointer to dentry
1448 * @verf: verifier to save
1450 * Saves the parent directory verifier in @dentry. If the inode has
1451 * a delegation, we also tag the dentry as having been revalidated
1452 * while holding a delegation so that we know we don't have to
1453 * look it up again after a directory change.
1455 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1458 spin_lock(&dentry->d_lock);
1459 nfs_set_verifier_locked(dentry, verf);
1460 spin_unlock(&dentry->d_lock);
1462 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1464 #if IS_ENABLED(CONFIG_NFS_V4)
1466 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1467 * @inode: pointer to inode
1469 * Iterates through the dentries in the inode alias list and clears
1470 * the tag used to indicate that the dentry has been revalidated
1471 * while holding a delegation.
1472 * This function is intended for use when the delegation is being
1473 * returned or revoked.
1475 void nfs_clear_verifier_delegated(struct inode *inode)
1477 struct dentry *alias;
1481 spin_lock(&inode->i_lock);
1482 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1483 spin_lock(&alias->d_lock);
1484 nfs_unset_verifier_delegated(&alias->d_time);
1485 spin_unlock(&alias->d_lock);
1487 spin_unlock(&inode->i_lock);
1489 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1490 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1492 static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1494 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) &&
1495 d_really_is_negative(dentry))
1496 return dentry->d_time == inode_peek_iversion_raw(dir);
1497 return nfs_verify_change_attribute(dir, dentry->d_time);
1501 * A check for whether or not the parent directory has changed.
1502 * In the case it has, we assume that the dentries are untrustworthy
1503 * and may need to be looked up again.
1504 * If rcu_walk prevents us from performing a full check, return 0.
1506 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1509 if (IS_ROOT(dentry))
1511 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1513 if (!nfs_dentry_verify_change(dir, dentry))
1515 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1516 if (nfs_mapping_need_revalidate_inode(dir)) {
1519 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1522 if (!nfs_dentry_verify_change(dir, dentry))
1528 * Use intent information to check whether or not we're going to do
1529 * an O_EXCL create using this path component.
1531 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1533 if (NFS_PROTO(dir)->version == 2)
1535 return flags & LOOKUP_EXCL;
1539 * Inode and filehandle revalidation for lookups.
1541 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1542 * or if the intent information indicates that we're about to open this
1543 * particular file and the "nocto" mount flag is not set.
1547 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1549 struct nfs_server *server = NFS_SERVER(inode);
1552 if (IS_AUTOMOUNT(inode))
1555 if (flags & LOOKUP_OPEN) {
1556 switch (inode->i_mode & S_IFMT) {
1558 /* A NFSv4 OPEN will revalidate later */
1559 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1563 if (server->flags & NFS_MOUNT_NOCTO)
1565 /* NFS close-to-open cache consistency validation */
1570 /* VFS wants an on-the-wire revalidation */
1571 if (flags & LOOKUP_REVAL)
1574 if (inode->i_nlink > 0 ||
1575 (inode->i_nlink == 0 &&
1576 test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1581 if (flags & LOOKUP_RCU)
1583 ret = __nfs_revalidate_inode(server, inode);
1589 static void nfs_mark_dir_for_revalidate(struct inode *inode)
1591 spin_lock(&inode->i_lock);
1592 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1593 spin_unlock(&inode->i_lock);
1597 * We judge how long we want to trust negative
1598 * dentries by looking at the parent inode mtime.
1600 * If parent mtime has changed, we revalidate, else we wait for a
1601 * period corresponding to the parent's attribute cache timeout value.
1603 * If LOOKUP_RCU prevents us from performing a full check, return 1
1604 * suggesting a reval is needed.
1606 * Note that when creating a new file, or looking up a rename target,
1607 * then it shouldn't be necessary to revalidate a negative dentry.
1610 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1613 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1615 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1617 /* Case insensitive server? Revalidate negative dentries */
1618 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1620 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1624 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1625 struct inode *inode, int error)
1632 * We can't d_drop the root of a disconnected tree:
1633 * its d_hash is on the s_anon list and d_drop() would hide
1634 * it from shrink_dcache_for_unmount(), leading to busy
1635 * inodes on unmount and further oopses.
1637 if (inode && IS_ROOT(dentry))
1641 trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error);
1646 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1650 if (nfs_neg_need_reval(dir, dentry, flags)) {
1651 if (flags & LOOKUP_RCU)
1655 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1659 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1660 struct inode *inode)
1662 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1663 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1666 static int nfs_lookup_revalidate_dentry(struct inode *dir,
1667 struct dentry *dentry,
1668 struct inode *inode, unsigned int flags)
1670 struct nfs_fh *fhandle;
1671 struct nfs_fattr *fattr;
1672 unsigned long dir_verifier;
1675 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1678 fhandle = nfs_alloc_fhandle();
1679 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1680 if (fhandle == NULL || fattr == NULL)
1683 dir_verifier = nfs_save_change_attribute(dir);
1684 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1692 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1698 /* Request help from readdirplus */
1699 nfs_lookup_advise_force_readdirplus(dir, flags);
1702 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1704 if (nfs_refresh_inode(inode, fattr) < 0)
1707 nfs_setsecurity(inode, fattr);
1708 nfs_set_verifier(dentry, dir_verifier);
1712 nfs_free_fattr(fattr);
1713 nfs_free_fhandle(fhandle);
1716 * If the lookup failed despite the dentry change attribute being
1717 * a match, then we should revalidate the directory cache.
1719 if (!ret && nfs_dentry_verify_change(dir, dentry))
1720 nfs_mark_dir_for_revalidate(dir);
1721 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1725 * This is called every time the dcache has a lookup hit,
1726 * and we should check whether we can really trust that
1729 * NOTE! The hit can be a negative hit too, don't assume
1732 * If the parent directory is seen to have changed, we throw out the
1733 * cached dentry and do a new lookup.
1736 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1739 struct inode *inode;
1742 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1743 inode = d_inode(dentry);
1746 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1748 if (is_bad_inode(inode)) {
1749 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1754 if ((flags & LOOKUP_RENAME_TARGET) && d_count(dentry) < 2 &&
1755 nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1758 if (nfs_verifier_is_delegated(dentry))
1759 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1761 /* Force a full look up iff the parent directory has changed */
1762 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1763 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1764 error = nfs_lookup_verify_inode(inode, flags);
1766 if (error == -ESTALE)
1767 nfs_mark_dir_for_revalidate(dir);
1773 if (flags & LOOKUP_RCU)
1776 if (NFS_STALE(inode))
1779 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
1781 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1783 if (flags & LOOKUP_RCU)
1785 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1789 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1790 int (*reval)(struct inode *, struct dentry *, unsigned int))
1792 struct dentry *parent;
1796 if (flags & LOOKUP_RCU) {
1797 if (dentry->d_fsdata == NFS_FSDATA_BLOCKED)
1799 parent = READ_ONCE(dentry->d_parent);
1800 dir = d_inode_rcu(parent);
1803 ret = reval(dir, dentry, flags);
1804 if (parent != READ_ONCE(dentry->d_parent))
1807 /* Wait for unlink to complete */
1808 wait_var_event(&dentry->d_fsdata,
1809 dentry->d_fsdata != NFS_FSDATA_BLOCKED);
1810 parent = dget_parent(dentry);
1811 ret = reval(d_inode(parent), dentry, flags);
1817 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1819 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1823 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1824 * when we don't really care about the dentry name. This is called when a
1825 * pathwalk ends on a dentry that was not found via a normal lookup in the
1826 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1828 * In this situation, we just want to verify that the inode itself is OK
1829 * since the dentry might have changed on the server.
1831 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1833 struct inode *inode = d_inode(dentry);
1837 * I believe we can only get a negative dentry here in the case of a
1838 * procfs-style symlink. Just assume it's correct for now, but we may
1839 * eventually need to do something more here.
1842 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1847 if (is_bad_inode(inode)) {
1848 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1853 error = nfs_lookup_verify_inode(inode, flags);
1854 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1855 __func__, inode->i_ino, error ? "invalid" : "valid");
1860 * This is called from dput() when d_count is going to 0.
1862 static int nfs_dentry_delete(const struct dentry *dentry)
1864 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1865 dentry, dentry->d_flags);
1867 /* Unhash any dentry with a stale inode */
1868 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1871 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1872 /* Unhash it, so that ->d_iput() would be called */
1875 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1876 /* Unhash it, so that ancestors of killed async unlink
1877 * files will be cleaned up during umount */
1884 /* Ensure that we revalidate inode->i_nlink */
1885 static void nfs_drop_nlink(struct inode *inode)
1887 spin_lock(&inode->i_lock);
1888 /* drop the inode if we're reasonably sure this is the last link */
1889 if (inode->i_nlink > 0)
1891 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1892 nfs_set_cache_invalid(
1893 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1894 NFS_INO_INVALID_NLINK);
1895 spin_unlock(&inode->i_lock);
1899 * Called when the dentry loses inode.
1900 * We use it to clean up silly-renamed files.
1902 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1904 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1905 nfs_complete_unlink(dentry, inode);
1906 nfs_drop_nlink(inode);
1911 static void nfs_d_release(struct dentry *dentry)
1913 /* free cached devname value, if it survived that far */
1914 if (unlikely(dentry->d_fsdata)) {
1915 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1918 kfree(dentry->d_fsdata);
1922 const struct dentry_operations nfs_dentry_operations = {
1923 .d_revalidate = nfs_lookup_revalidate,
1924 .d_weak_revalidate = nfs_weak_revalidate,
1925 .d_delete = nfs_dentry_delete,
1926 .d_iput = nfs_dentry_iput,
1927 .d_automount = nfs_d_automount,
1928 .d_release = nfs_d_release,
1930 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1932 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1935 struct inode *inode = NULL;
1936 struct nfs_fh *fhandle = NULL;
1937 struct nfs_fattr *fattr = NULL;
1938 unsigned long dir_verifier;
1941 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1942 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1944 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1945 return ERR_PTR(-ENAMETOOLONG);
1948 * If we're doing an exclusive create, optimize away the lookup
1949 * but don't hash the dentry.
1951 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1954 res = ERR_PTR(-ENOMEM);
1955 fhandle = nfs_alloc_fhandle();
1956 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir));
1957 if (fhandle == NULL || fattr == NULL)
1960 dir_verifier = nfs_save_change_attribute(dir);
1961 trace_nfs_lookup_enter(dir, dentry, flags);
1962 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1963 if (error == -ENOENT) {
1964 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1965 dir_verifier = inode_peek_iversion_raw(dir);
1969 res = ERR_PTR(error);
1972 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1973 res = ERR_CAST(inode);
1977 /* Notify readdir to use READDIRPLUS */
1978 nfs_lookup_advise_force_readdirplus(dir, flags);
1981 res = d_splice_alias(inode, dentry);
1987 nfs_set_verifier(dentry, dir_verifier);
1989 trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res));
1990 nfs_free_fattr(fattr);
1991 nfs_free_fhandle(fhandle);
1994 EXPORT_SYMBOL_GPL(nfs_lookup);
1996 void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
1998 /* Case insensitive server? Revalidate dentries */
1999 if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
2000 d_prune_aliases(inode);
2002 EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
2004 #if IS_ENABLED(CONFIG_NFS_V4)
2005 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
2007 const struct dentry_operations nfs4_dentry_operations = {
2008 .d_revalidate = nfs4_lookup_revalidate,
2009 .d_weak_revalidate = nfs_weak_revalidate,
2010 .d_delete = nfs_dentry_delete,
2011 .d_iput = nfs_dentry_iput,
2012 .d_automount = nfs_d_automount,
2013 .d_release = nfs_d_release,
2015 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
2017 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
2019 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
2022 static int do_open(struct inode *inode, struct file *filp)
2024 nfs_fscache_open_file(inode, filp);
2028 static int nfs_finish_open(struct nfs_open_context *ctx,
2029 struct dentry *dentry,
2030 struct file *file, unsigned open_flags)
2034 err = finish_open(file, dentry, do_open);
2037 if (S_ISREG(file_inode(file)->i_mode))
2038 nfs_file_set_open_context(file, ctx);
2045 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2046 struct file *file, unsigned open_flags,
2049 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2050 struct nfs_open_context *ctx;
2052 struct iattr attr = { .ia_valid = ATTR_OPEN };
2053 struct inode *inode;
2054 unsigned int lookup_flags = 0;
2055 unsigned long dir_verifier;
2056 bool switched = false;
2060 /* Expect a negative dentry */
2061 BUG_ON(d_inode(dentry));
2063 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2064 dir->i_sb->s_id, dir->i_ino, dentry);
2066 err = nfs_check_flags(open_flags);
2070 /* NFS only supports OPEN on regular files */
2071 if ((open_flags & O_DIRECTORY)) {
2072 if (!d_in_lookup(dentry)) {
2074 * Hashed negative dentry with O_DIRECTORY: dentry was
2075 * revalidated and is fine, no need to perform lookup
2080 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2084 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
2085 return -ENAMETOOLONG;
2087 if (open_flags & O_CREAT) {
2088 struct nfs_server *server = NFS_SERVER(dir);
2090 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2091 mode &= ~current_umask();
2093 attr.ia_valid |= ATTR_MODE;
2094 attr.ia_mode = mode;
2096 if (open_flags & O_TRUNC) {
2097 attr.ia_valid |= ATTR_SIZE;
2101 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2104 dentry = d_alloc_parallel(dentry->d_parent,
2105 &dentry->d_name, &wq);
2107 return PTR_ERR(dentry);
2108 if (unlikely(!d_in_lookup(dentry)))
2109 return finish_no_open(file, dentry);
2112 ctx = create_nfs_open_context(dentry, open_flags, file);
2117 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
2118 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
2120 file->f_mode |= FMODE_CREATED;
2121 if (IS_ERR(inode)) {
2122 err = PTR_ERR(inode);
2123 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2124 put_nfs_open_context(ctx);
2128 d_splice_alias(NULL, dentry);
2129 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
2130 dir_verifier = inode_peek_iversion_raw(dir);
2132 dir_verifier = nfs_save_change_attribute(dir);
2133 nfs_set_verifier(dentry, dir_verifier);
2139 if (!(open_flags & O_NOFOLLOW))
2148 file->f_mode |= FMODE_CAN_ODIRECT;
2150 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
2151 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2152 put_nfs_open_context(ctx);
2154 if (unlikely(switched)) {
2155 d_lookup_done(dentry);
2161 res = nfs_lookup(dir, dentry, lookup_flags);
2163 inode = d_inode(dentry);
2164 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2165 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2166 res = ERR_PTR(-ENOTDIR);
2167 else if (inode && S_ISREG(inode->i_mode))
2168 res = ERR_PTR(-EOPENSTALE);
2169 } else if (!IS_ERR(res)) {
2170 inode = d_inode(res);
2171 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2172 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2174 res = ERR_PTR(-ENOTDIR);
2175 } else if (inode && S_ISREG(inode->i_mode)) {
2177 res = ERR_PTR(-EOPENSTALE);
2181 d_lookup_done(dentry);
2188 return PTR_ERR(res);
2189 return finish_no_open(file, res);
2191 EXPORT_SYMBOL_GPL(nfs_atomic_open);
2194 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2197 struct inode *inode;
2199 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
2201 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2203 if (d_mountpoint(dentry))
2206 inode = d_inode(dentry);
2208 /* We can't create new files in nfs_open_revalidate(), so we
2209 * optimize away revalidation of negative dentries.
2214 if (nfs_verifier_is_delegated(dentry))
2215 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2217 /* NFS only supports OPEN on regular files */
2218 if (!S_ISREG(inode->i_mode))
2221 /* We cannot do exclusive creation on a positive dentry */
2222 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2225 /* Check if the directory changed */
2226 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2229 /* Let f_op->open() actually open (and revalidate) the file */
2232 if (flags & LOOKUP_RCU)
2234 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
2237 return nfs_do_lookup_revalidate(dir, dentry, flags);
2240 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2242 return __nfs_lookup_revalidate(dentry, flags,
2243 nfs4_do_lookup_revalidate);
2246 #endif /* CONFIG_NFSV4 */
2248 int nfs_atomic_open_v23(struct inode *dir, struct dentry *dentry,
2249 struct file *file, unsigned int open_flags,
2253 /* Same as look+open from lookup_open(), but with different O_TRUNC
2258 if (open_flags & O_CREAT) {
2259 file->f_mode |= FMODE_CREATED;
2260 error = nfs_do_create(dir, dentry, mode, open_flags);
2263 return finish_open(file, dentry, NULL);
2264 } else if (d_in_lookup(dentry)) {
2265 /* The only flags nfs_lookup considers are
2266 * LOOKUP_EXCL and LOOKUP_RENAME_TARGET, and
2267 * we want those to be zero so the lookup isn't skipped.
2269 struct dentry *res = nfs_lookup(dir, dentry, 0);
2271 d_lookup_done(dentry);
2272 if (unlikely(res)) {
2274 return PTR_ERR(res);
2275 return finish_no_open(file, res);
2278 return finish_no_open(file, NULL);
2281 EXPORT_SYMBOL_GPL(nfs_atomic_open_v23);
2284 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2285 struct nfs_fattr *fattr)
2287 struct dentry *parent = dget_parent(dentry);
2288 struct inode *dir = d_inode(parent);
2289 struct inode *inode;
2295 if (fhandle->size == 0) {
2296 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
2300 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2301 if (!(fattr->valid & NFS_ATTR_FATTR)) {
2302 struct nfs_server *server = NFS_SB(dentry->d_sb);
2303 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2308 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2309 d = d_splice_alias(inode, dentry);
2317 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2320 * Code common to create, mkdir, and mknod.
2322 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2323 struct nfs_fattr *fattr)
2327 d = nfs_add_or_obtain(dentry, fhandle, fattr);
2331 /* Callers don't care */
2335 EXPORT_SYMBOL_GPL(nfs_instantiate);
2338 * Following a failed create operation, we drop the dentry rather
2339 * than retain a negative dentry. This avoids a problem in the event
2340 * that the operation succeeded on the server, but an error in the
2341 * reply path made it appear to have failed.
2343 static int nfs_do_create(struct inode *dir, struct dentry *dentry,
2344 umode_t mode, int open_flags)
2349 open_flags |= O_CREAT;
2351 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2352 dir->i_sb->s_id, dir->i_ino, dentry);
2354 attr.ia_mode = mode;
2355 attr.ia_valid = ATTR_MODE;
2356 if (open_flags & O_TRUNC) {
2358 attr.ia_valid |= ATTR_SIZE;
2361 trace_nfs_create_enter(dir, dentry, open_flags);
2362 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2363 trace_nfs_create_exit(dir, dentry, open_flags, error);
2372 int nfs_create(struct mnt_idmap *idmap, struct inode *dir,
2373 struct dentry *dentry, umode_t mode, bool excl)
2375 return nfs_do_create(dir, dentry, mode, excl ? O_EXCL : 0);
2377 EXPORT_SYMBOL_GPL(nfs_create);
2380 * See comments for nfs_proc_create regarding failed operations.
2383 nfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
2384 struct dentry *dentry, umode_t mode, dev_t rdev)
2389 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2390 dir->i_sb->s_id, dir->i_ino, dentry);
2392 attr.ia_mode = mode;
2393 attr.ia_valid = ATTR_MODE;
2395 trace_nfs_mknod_enter(dir, dentry);
2396 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2397 trace_nfs_mknod_exit(dir, dentry, status);
2405 EXPORT_SYMBOL_GPL(nfs_mknod);
2408 * See comments for nfs_proc_create regarding failed operations.
2410 int nfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
2411 struct dentry *dentry, umode_t mode)
2416 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2417 dir->i_sb->s_id, dir->i_ino, dentry);
2419 attr.ia_valid = ATTR_MODE;
2420 attr.ia_mode = mode | S_IFDIR;
2422 trace_nfs_mkdir_enter(dir, dentry);
2423 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2424 trace_nfs_mkdir_exit(dir, dentry, error);
2432 EXPORT_SYMBOL_GPL(nfs_mkdir);
2434 static void nfs_dentry_handle_enoent(struct dentry *dentry)
2436 if (simple_positive(dentry))
2440 static void nfs_dentry_remove_handle_error(struct inode *dir,
2441 struct dentry *dentry, int error)
2445 if (d_really_is_positive(dentry))
2447 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2450 nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2451 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2455 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2459 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2460 dir->i_sb->s_id, dir->i_ino, dentry);
2462 trace_nfs_rmdir_enter(dir, dentry);
2463 if (d_really_is_positive(dentry)) {
2464 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2465 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2466 /* Ensure the VFS deletes this inode */
2469 clear_nlink(d_inode(dentry));
2472 nfs_dentry_handle_enoent(dentry);
2474 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2476 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2477 nfs_dentry_remove_handle_error(dir, dentry, error);
2478 trace_nfs_rmdir_exit(dir, dentry, error);
2482 EXPORT_SYMBOL_GPL(nfs_rmdir);
2485 * Remove a file after making sure there are no pending writes,
2486 * and after checking that the file has only one user.
2488 * We invalidate the attribute cache and free the inode prior to the operation
2489 * to avoid possible races if the server reuses the inode.
2491 static int nfs_safe_remove(struct dentry *dentry)
2493 struct inode *dir = d_inode(dentry->d_parent);
2494 struct inode *inode = d_inode(dentry);
2497 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2499 /* If the dentry was sillyrenamed, we simply call d_delete() */
2500 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2505 trace_nfs_remove_enter(dir, dentry);
2506 if (inode != NULL) {
2507 error = NFS_PROTO(dir)->remove(dir, dentry);
2509 nfs_drop_nlink(inode);
2511 error = NFS_PROTO(dir)->remove(dir, dentry);
2512 if (error == -ENOENT)
2513 nfs_dentry_handle_enoent(dentry);
2514 trace_nfs_remove_exit(dir, dentry, error);
2519 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2520 * belongs to an active ".nfs..." file and we return -EBUSY.
2522 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2524 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2528 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2529 dir->i_ino, dentry);
2531 trace_nfs_unlink_enter(dir, dentry);
2532 spin_lock(&dentry->d_lock);
2533 if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2534 &NFS_I(d_inode(dentry))->flags)) {
2535 spin_unlock(&dentry->d_lock);
2536 /* Start asynchronous writeout of the inode */
2537 write_inode_now(d_inode(dentry), 0);
2538 error = nfs_sillyrename(dir, dentry);
2541 /* We must prevent any concurrent open until the unlink
2542 * completes. ->d_revalidate will wait for ->d_fsdata
2543 * to clear. We set it here to ensure no lookup succeeds until
2544 * the unlink is complete on the server.
2547 if (WARN_ON(dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2548 WARN_ON(dentry->d_fsdata == NFS_FSDATA_BLOCKED)) {
2549 spin_unlock(&dentry->d_lock);
2553 kfree(dentry->d_fsdata);
2554 dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2556 spin_unlock(&dentry->d_lock);
2557 error = nfs_safe_remove(dentry);
2558 nfs_dentry_remove_handle_error(dir, dentry, error);
2559 dentry->d_fsdata = NULL;
2560 wake_up_var(&dentry->d_fsdata);
2562 trace_nfs_unlink_exit(dir, dentry, error);
2565 EXPORT_SYMBOL_GPL(nfs_unlink);
2568 * To create a symbolic link, most file systems instantiate a new inode,
2569 * add a page to it containing the path, then write it out to the disk
2570 * using prepare_write/commit_write.
2572 * Unfortunately the NFS client can't create the in-core inode first
2573 * because it needs a file handle to create an in-core inode (see
2574 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2575 * symlink request has completed on the server.
2577 * So instead we allocate a raw page, copy the symname into it, then do
2578 * the SYMLINK request with the page as the buffer. If it succeeds, we
2579 * now have a new file handle and can instantiate an in-core NFS inode
2580 * and move the raw page into its mapping.
2582 int nfs_symlink(struct mnt_idmap *idmap, struct inode *dir,
2583 struct dentry *dentry, const char *symname)
2585 struct folio *folio;
2588 unsigned int pathlen = strlen(symname);
2591 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2592 dir->i_ino, dentry, symname);
2594 if (pathlen > PAGE_SIZE)
2595 return -ENAMETOOLONG;
2597 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2598 attr.ia_valid = ATTR_MODE;
2600 folio = folio_alloc(GFP_USER, 0);
2604 kaddr = folio_address(folio);
2605 memcpy(kaddr, symname, pathlen);
2606 if (pathlen < PAGE_SIZE)
2607 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2609 trace_nfs_symlink_enter(dir, dentry);
2610 error = NFS_PROTO(dir)->symlink(dir, dentry, folio, pathlen, &attr);
2611 trace_nfs_symlink_exit(dir, dentry, error);
2613 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2614 dir->i_sb->s_id, dir->i_ino,
2615 dentry, symname, error);
2621 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2624 * No big deal if we can't add this page to the page cache here.
2625 * READLINK will get the missing page from the server if needed.
2627 if (filemap_add_folio(d_inode(dentry)->i_mapping, folio, 0,
2629 folio_mark_uptodate(folio);
2630 folio_unlock(folio);
2636 EXPORT_SYMBOL_GPL(nfs_symlink);
2639 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2641 struct inode *inode = d_inode(old_dentry);
2644 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2645 old_dentry, dentry);
2647 trace_nfs_link_enter(inode, dir, dentry);
2649 if (S_ISREG(inode->i_mode))
2650 nfs_sync_inode(inode);
2651 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2653 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2655 d_add(dentry, inode);
2657 trace_nfs_link_exit(inode, dir, dentry, error);
2660 EXPORT_SYMBOL_GPL(nfs_link);
2663 nfs_unblock_rename(struct rpc_task *task, struct nfs_renamedata *data)
2665 struct dentry *new_dentry = data->new_dentry;
2667 new_dentry->d_fsdata = NULL;
2668 wake_up_var(&new_dentry->d_fsdata);
2673 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2674 * different file handle for the same inode after a rename (e.g. when
2675 * moving to a different directory). A fail-safe method to do so would
2676 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2677 * rename the old file using the sillyrename stuff. This way, the original
2678 * file in old_dir will go away when the last process iput()s the inode.
2682 * It actually works quite well. One needs to have the possibility for
2683 * at least one ".nfs..." file in each directory the file ever gets
2684 * moved or linked to which happens automagically with the new
2685 * implementation that only depends on the dcache stuff instead of
2686 * using the inode layer
2688 * Unfortunately, things are a little more complicated than indicated
2689 * above. For a cross-directory move, we want to make sure we can get
2690 * rid of the old inode after the operation. This means there must be
2691 * no pending writes (if it's a file), and the use count must be 1.
2692 * If these conditions are met, we can drop the dentries before doing
2695 int nfs_rename(struct mnt_idmap *idmap, struct inode *old_dir,
2696 struct dentry *old_dentry, struct inode *new_dir,
2697 struct dentry *new_dentry, unsigned int flags)
2699 struct inode *old_inode = d_inode(old_dentry);
2700 struct inode *new_inode = d_inode(new_dentry);
2701 struct dentry *dentry = NULL;
2702 struct rpc_task *task;
2703 bool must_unblock = false;
2709 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2710 old_dentry, new_dentry,
2711 d_count(new_dentry));
2713 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2715 * For non-directories, check whether the target is busy and if so,
2716 * make a copy of the dentry and then do a silly-rename. If the
2717 * silly-rename succeeds, the copied dentry is hashed and becomes
2720 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2721 /* We must prevent any concurrent open until the unlink
2722 * completes. ->d_revalidate will wait for ->d_fsdata
2723 * to clear. We set it here to ensure no lookup succeeds until
2724 * the unlink is complete on the server.
2727 if (WARN_ON(new_dentry->d_flags & DCACHE_NFSFS_RENAMED) ||
2728 WARN_ON(new_dentry->d_fsdata == NFS_FSDATA_BLOCKED))
2730 if (new_dentry->d_fsdata) {
2732 kfree(new_dentry->d_fsdata);
2733 new_dentry->d_fsdata = NULL;
2736 spin_lock(&new_dentry->d_lock);
2737 if (d_count(new_dentry) > 2) {
2740 spin_unlock(&new_dentry->d_lock);
2742 /* copy the target dentry's name */
2743 dentry = d_alloc(new_dentry->d_parent,
2744 &new_dentry->d_name);
2748 /* silly-rename the existing target ... */
2749 err = nfs_sillyrename(new_dir, new_dentry);
2753 new_dentry = dentry;
2756 new_dentry->d_fsdata = NFS_FSDATA_BLOCKED;
2757 must_unblock = true;
2758 spin_unlock(&new_dentry->d_lock);
2763 if (S_ISREG(old_inode->i_mode))
2764 nfs_sync_inode(old_inode);
2765 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry,
2766 must_unblock ? nfs_unblock_rename : NULL);
2768 error = PTR_ERR(task);
2772 error = rpc_wait_for_completion_task(task);
2774 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2775 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2778 error = task->tk_status;
2780 /* Ensure the inode attributes are revalidated */
2782 spin_lock(&old_inode->i_lock);
2783 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2784 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2785 NFS_INO_INVALID_CTIME |
2786 NFS_INO_REVAL_FORCED);
2787 spin_unlock(&old_inode->i_lock);
2790 trace_nfs_rename_exit(old_dir, old_dentry,
2791 new_dir, new_dentry, error);
2793 if (new_inode != NULL)
2794 nfs_drop_nlink(new_inode);
2796 * The d_move() should be here instead of in an async RPC completion
2797 * handler because we need the proper locks to move the dentry. If
2798 * we're interrupted by a signal, the async RPC completion handler
2799 * should mark the directories for revalidation.
2801 d_move(old_dentry, new_dentry);
2802 nfs_set_verifier(old_dentry,
2803 nfs_save_change_attribute(new_dir));
2804 } else if (error == -ENOENT)
2805 nfs_dentry_handle_enoent(old_dentry);
2807 /* new dentry created? */
2812 EXPORT_SYMBOL_GPL(nfs_rename);
2814 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2815 static LIST_HEAD(nfs_access_lru_list);
2816 static atomic_long_t nfs_access_nr_entries;
2818 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2819 module_param(nfs_access_max_cachesize, ulong, 0644);
2820 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2822 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2824 put_group_info(entry->group_info);
2825 kfree_rcu(entry, rcu_head);
2826 smp_mb__before_atomic();
2827 atomic_long_dec(&nfs_access_nr_entries);
2828 smp_mb__after_atomic();
2831 static void nfs_access_free_list(struct list_head *head)
2833 struct nfs_access_entry *cache;
2835 while (!list_empty(head)) {
2836 cache = list_entry(head->next, struct nfs_access_entry, lru);
2837 list_del(&cache->lru);
2838 nfs_access_free_entry(cache);
2842 static unsigned long
2843 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2846 struct nfs_inode *nfsi, *next;
2847 struct nfs_access_entry *cache;
2850 spin_lock(&nfs_access_lru_lock);
2851 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2852 struct inode *inode;
2854 if (nr_to_scan-- == 0)
2856 inode = &nfsi->vfs_inode;
2857 spin_lock(&inode->i_lock);
2858 if (list_empty(&nfsi->access_cache_entry_lru))
2859 goto remove_lru_entry;
2860 cache = list_entry(nfsi->access_cache_entry_lru.next,
2861 struct nfs_access_entry, lru);
2862 list_move(&cache->lru, &head);
2863 rb_erase(&cache->rb_node, &nfsi->access_cache);
2865 if (!list_empty(&nfsi->access_cache_entry_lru))
2866 list_move_tail(&nfsi->access_cache_inode_lru,
2867 &nfs_access_lru_list);
2870 list_del_init(&nfsi->access_cache_inode_lru);
2871 smp_mb__before_atomic();
2872 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2873 smp_mb__after_atomic();
2875 spin_unlock(&inode->i_lock);
2877 spin_unlock(&nfs_access_lru_lock);
2878 nfs_access_free_list(&head);
2883 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2885 int nr_to_scan = sc->nr_to_scan;
2886 gfp_t gfp_mask = sc->gfp_mask;
2888 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2890 return nfs_do_access_cache_scan(nr_to_scan);
2895 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2897 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2901 nfs_access_cache_enforce_limit(void)
2903 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2905 unsigned int nr_to_scan;
2907 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2910 diff = nr_entries - nfs_access_max_cachesize;
2911 if (diff < nr_to_scan)
2913 nfs_do_access_cache_scan(nr_to_scan);
2916 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2918 struct rb_root *root_node = &nfsi->access_cache;
2920 struct nfs_access_entry *entry;
2922 /* Unhook entries from the cache */
2923 while ((n = rb_first(root_node)) != NULL) {
2924 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2925 rb_erase(n, root_node);
2926 list_move(&entry->lru, head);
2928 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2931 void nfs_access_zap_cache(struct inode *inode)
2935 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2937 /* Remove from global LRU init */
2938 spin_lock(&nfs_access_lru_lock);
2939 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2940 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2942 spin_lock(&inode->i_lock);
2943 __nfs_access_zap_cache(NFS_I(inode), &head);
2944 spin_unlock(&inode->i_lock);
2945 spin_unlock(&nfs_access_lru_lock);
2946 nfs_access_free_list(&head);
2948 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2950 static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2952 struct group_info *ga, *gb;
2955 if (uid_lt(a->fsuid, b->fsuid))
2957 if (uid_gt(a->fsuid, b->fsuid))
2960 if (gid_lt(a->fsgid, b->fsgid))
2962 if (gid_gt(a->fsgid, b->fsgid))
2973 if (ga->ngroups < gb->ngroups)
2975 if (ga->ngroups > gb->ngroups)
2978 for (g = 0; g < ga->ngroups; g++) {
2979 if (gid_lt(ga->gid[g], gb->gid[g]))
2981 if (gid_gt(ga->gid[g], gb->gid[g]))
2987 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2989 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2992 struct nfs_access_entry *entry =
2993 rb_entry(n, struct nfs_access_entry, rb_node);
2994 int cmp = access_cmp(cred, entry);
3006 static u64 nfs_access_login_time(const struct task_struct *task,
3007 const struct cred *cred)
3009 const struct task_struct *parent;
3010 const struct cred *pcred;
3015 parent = rcu_dereference(task->real_parent);
3016 pcred = __task_cred(parent);
3017 if (parent == task || cred_fscmp(pcred, cred) != 0)
3021 ret = task->start_time;
3026 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
3028 struct nfs_inode *nfsi = NFS_I(inode);
3029 u64 login_time = nfs_access_login_time(current, cred);
3030 struct nfs_access_entry *cache;
3034 spin_lock(&inode->i_lock);
3036 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3038 cache = nfs_access_search_rbtree(inode, cred);
3042 /* Found an entry, is our attribute cache valid? */
3043 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3050 spin_unlock(&inode->i_lock);
3051 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
3054 spin_lock(&inode->i_lock);
3058 if ((s64)(login_time - cache->timestamp) > 0)
3060 *mask = cache->mask;
3061 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
3064 spin_unlock(&inode->i_lock);
3067 spin_unlock(&inode->i_lock);
3068 nfs_access_zap_cache(inode);
3072 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
3074 /* Only check the most recently returned cache entry,
3075 * but do it without locking.
3077 struct nfs_inode *nfsi = NFS_I(inode);
3078 u64 login_time = nfs_access_login_time(current, cred);
3079 struct nfs_access_entry *cache;
3081 struct list_head *lh;
3084 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
3086 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
3087 cache = list_entry(lh, struct nfs_access_entry, lru);
3088 if (lh == &nfsi->access_cache_entry_lru ||
3089 access_cmp(cred, cache) != 0)
3093 if ((s64)(login_time - cache->timestamp) > 0)
3095 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
3097 *mask = cache->mask;
3104 int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
3105 u32 *mask, bool may_block)
3109 status = nfs_access_get_cached_rcu(inode, cred, mask);
3111 status = nfs_access_get_cached_locked(inode, cred, mask,
3116 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
3118 static void nfs_access_add_rbtree(struct inode *inode,
3119 struct nfs_access_entry *set,
3120 const struct cred *cred)
3122 struct nfs_inode *nfsi = NFS_I(inode);
3123 struct rb_root *root_node = &nfsi->access_cache;
3124 struct rb_node **p = &root_node->rb_node;
3125 struct rb_node *parent = NULL;
3126 struct nfs_access_entry *entry;
3129 spin_lock(&inode->i_lock);
3130 while (*p != NULL) {
3132 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3133 cmp = access_cmp(cred, entry);
3136 p = &parent->rb_left;
3138 p = &parent->rb_right;
3142 rb_link_node(&set->rb_node, parent, p);
3143 rb_insert_color(&set->rb_node, root_node);
3144 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3145 spin_unlock(&inode->i_lock);
3148 rb_replace_node(parent, &set->rb_node, root_node);
3149 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3150 list_del(&entry->lru);
3151 spin_unlock(&inode->i_lock);
3152 nfs_access_free_entry(entry);
3155 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3156 const struct cred *cred)
3158 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
3161 RB_CLEAR_NODE(&cache->rb_node);
3162 cache->fsuid = cred->fsuid;
3163 cache->fsgid = cred->fsgid;
3164 cache->group_info = get_group_info(cred->group_info);
3165 cache->mask = set->mask;
3166 cache->timestamp = ktime_get_ns();
3168 /* The above field assignments must be visible
3169 * before this item appears on the lru. We cannot easily
3170 * use rcu_assign_pointer, so just force the memory barrier.
3173 nfs_access_add_rbtree(inode, cache, cred);
3175 /* Update accounting */
3176 smp_mb__before_atomic();
3177 atomic_long_inc(&nfs_access_nr_entries);
3178 smp_mb__after_atomic();
3180 /* Add inode to global LRU list */
3181 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3182 spin_lock(&nfs_access_lru_lock);
3183 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
3184 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
3185 &nfs_access_lru_list);
3186 spin_unlock(&nfs_access_lru_lock);
3188 nfs_access_cache_enforce_limit();
3190 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3192 #define NFS_MAY_READ (NFS_ACCESS_READ)
3193 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3194 NFS_ACCESS_EXTEND | \
3196 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3198 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3199 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3200 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3202 nfs_access_calc_mask(u32 access_result, umode_t umode)
3206 if (access_result & NFS_MAY_READ)
3208 if (S_ISDIR(umode)) {
3209 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3211 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3213 } else if (S_ISREG(umode)) {
3214 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3216 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3218 } else if (access_result & NFS_MAY_WRITE)
3223 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3225 entry->mask = access_result;
3227 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3229 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3231 struct nfs_access_entry cache;
3232 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3233 int cache_mask = -1;
3236 trace_nfs_access_enter(inode);
3238 status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3247 * Determine which access bits we want to ask for...
3249 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3250 nfs_access_xattr_mask(NFS_SERVER(inode));
3251 if (S_ISDIR(inode->i_mode))
3252 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3254 cache.mask |= NFS_ACCESS_EXECUTE;
3255 status = NFS_PROTO(inode)->access(inode, &cache, cred);
3257 if (status == -ESTALE) {
3258 if (!S_ISDIR(inode->i_mode))
3259 nfs_set_inode_stale(inode);
3261 nfs_zap_caches(inode);
3265 nfs_access_add_cache(inode, &cache, cred);
3267 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
3268 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3271 trace_nfs_access_exit(inode, mask, cache_mask, status);
3275 static int nfs_open_permission_mask(int openflags)
3279 if (openflags & __FMODE_EXEC) {
3280 /* ONLY check exec rights */
3283 if ((openflags & O_ACCMODE) != O_WRONLY)
3285 if ((openflags & O_ACCMODE) != O_RDONLY)
3292 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3294 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
3296 EXPORT_SYMBOL_GPL(nfs_may_open);
3298 static int nfs_execute_ok(struct inode *inode, int mask)
3300 struct nfs_server *server = NFS_SERVER(inode);
3303 if (S_ISDIR(inode->i_mode))
3305 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3306 if (mask & MAY_NOT_BLOCK)
3308 ret = __nfs_revalidate_inode(server, inode);
3310 if (ret == 0 && !execute_ok(inode))
3315 int nfs_permission(struct mnt_idmap *idmap,
3316 struct inode *inode,
3319 const struct cred *cred = current_cred();
3322 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
3324 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3326 /* Is this sys_access() ? */
3327 if (mask & (MAY_ACCESS | MAY_CHDIR))
3330 switch (inode->i_mode & S_IFMT) {
3334 if ((mask & MAY_OPEN) &&
3335 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3340 * Optimize away all write operations, since the server
3341 * will check permissions when we perform the op.
3343 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3348 if (!NFS_PROTO(inode)->access)
3351 res = nfs_do_access(inode, cred, mask);
3353 if (!res && (mask & MAY_EXEC))
3354 res = nfs_execute_ok(inode, mask);
3356 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3357 inode->i_sb->s_id, inode->i_ino, mask, res);
3360 if (mask & MAY_NOT_BLOCK)
3363 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3364 NFS_INO_INVALID_OTHER);
3366 res = generic_permission(&nop_mnt_idmap, inode, mask);
3369 EXPORT_SYMBOL_GPL(nfs_permission);