1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1992 Rick Sladkey
7 * nfs directory handling functions
9 * 10 Apr 1996 Added silly rename for unlink --okir
10 * 28 Sep 1996 Improved directory cache --okir
11 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
12 * Re-implemented silly rename for unlink, newly implemented
13 * silly rename for nfs_rename() following the suggestions
14 * of Olaf Kirch (okir) found in this file.
15 * Following Linus comments on my original hack, this version
16 * depends only on the dcache stuff and doesn't touch the inode
17 * layer (iput() and friends).
18 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
21 #include <linux/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_free_folio(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_free_folio,
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 page_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 page_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_page_init_array(struct page *page, u64 last_cookie,
204 struct nfs_cache_array *array;
206 array = kmap_atomic(page);
207 array->change_attr = change_attr;
208 array->last_cookie = last_cookie;
210 array->page_full = 0;
211 array->page_is_eof = 0;
212 array->cookies_are_ordered = 1;
213 kunmap_atomic(array);
217 * we are freeing strings created by nfs_add_to_readdir_array()
219 static void nfs_readdir_clear_array(struct page *page)
221 struct nfs_cache_array *array;
224 array = kmap_atomic(page);
225 for (i = 0; i < array->size; i++)
226 kfree(array->array[i].name);
228 kunmap_atomic(array);
231 static void nfs_readdir_free_folio(struct folio *folio)
233 nfs_readdir_clear_array(&folio->page);
236 static void nfs_readdir_page_reinit_array(struct page *page, u64 last_cookie,
239 nfs_readdir_clear_array(page);
240 nfs_readdir_page_init_array(page, last_cookie, change_attr);
244 nfs_readdir_page_array_alloc(u64 last_cookie, gfp_t gfp_flags)
246 struct page *page = alloc_page(gfp_flags);
248 nfs_readdir_page_init_array(page, last_cookie, 0);
252 static void nfs_readdir_page_array_free(struct page *page)
255 nfs_readdir_clear_array(page);
260 static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array)
262 return array->size == 0 ? array->last_cookie : array->array[0].cookie;
265 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
267 array->page_is_eof = 1;
268 array->page_full = 1;
271 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
273 return array->page_full;
277 * the caller is responsible for freeing qstr.name
278 * when called by nfs_readdir_add_to_array, the strings will be freed in
279 * nfs_clear_readdir_array()
281 static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
283 const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
286 * Avoid a kmemleak false positive. The pointer to the name is stored
287 * in a page cache page which kmemleak does not scan.
290 kmemleak_not_leak(ret);
294 static size_t nfs_readdir_array_maxentries(void)
296 return (PAGE_SIZE - sizeof(struct nfs_cache_array)) /
297 sizeof(struct nfs_cache_array_entry);
301 * Check that the next array entry lies entirely within the page bounds
303 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
305 if (array->page_full)
307 if (array->size == nfs_readdir_array_maxentries()) {
308 array->page_full = 1;
314 static int nfs_readdir_page_array_append(struct page *page,
315 const struct nfs_entry *entry,
318 struct nfs_cache_array *array;
319 struct nfs_cache_array_entry *cache_entry;
323 name = nfs_readdir_copy_name(entry->name, entry->len);
325 array = kmap_atomic(page);
328 ret = nfs_readdir_array_can_expand(array);
334 cache_entry = &array->array[array->size];
335 cache_entry->cookie = array->last_cookie;
336 cache_entry->ino = entry->ino;
337 cache_entry->d_type = entry->d_type;
338 cache_entry->name_len = entry->len;
339 cache_entry->name = name;
340 array->last_cookie = entry->cookie;
341 if (array->last_cookie <= cache_entry->cookie)
342 array->cookies_are_ordered = 0;
345 nfs_readdir_array_set_eof(array);
347 *cookie = array->last_cookie;
348 kunmap_atomic(array);
352 #define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14)
354 * Hash algorithm allowing content addressible access to sequences
355 * of directory cookies. Content is addressed by the value of the
356 * cookie index of the first readdir entry in a page.
358 * We select only the first 18 bits to avoid issues with excessive
359 * memory use for the page cache XArray. 18 bits should allow the caching
360 * of 262144 pages of sequences of readdir entries. Since each page holds
361 * 127 readdir entries for a typical 64-bit system, that works out to a
362 * cache of ~ 33 million entries per directory.
364 static pgoff_t nfs_readdir_page_cookie_hash(u64 cookie)
368 return hash_64(cookie, 18);
371 static bool nfs_readdir_page_validate(struct page *page, u64 last_cookie,
374 struct nfs_cache_array *array = kmap_atomic(page);
377 if (array->change_attr != change_attr)
379 if (nfs_readdir_array_index_cookie(array) != last_cookie)
381 kunmap_atomic(array);
385 static void nfs_readdir_page_unlock_and_put(struct page *page)
391 static void nfs_readdir_page_init_and_validate(struct page *page, u64 cookie,
394 if (PageUptodate(page)) {
395 if (nfs_readdir_page_validate(page, cookie, change_attr))
397 nfs_readdir_clear_array(page);
399 nfs_readdir_page_init_array(page, cookie, change_attr);
400 SetPageUptodate(page);
403 static struct page *nfs_readdir_page_get_locked(struct address_space *mapping,
404 u64 cookie, u64 change_attr)
406 pgoff_t index = nfs_readdir_page_cookie_hash(cookie);
409 page = grab_cache_page(mapping, index);
412 nfs_readdir_page_init_and_validate(page, cookie, change_attr);
416 static u64 nfs_readdir_page_last_cookie(struct page *page)
418 struct nfs_cache_array *array;
421 array = kmap_atomic(page);
422 ret = array->last_cookie;
423 kunmap_atomic(array);
427 static bool nfs_readdir_page_needs_filling(struct page *page)
429 struct nfs_cache_array *array;
432 array = kmap_atomic(page);
433 ret = !nfs_readdir_array_is_full(array);
434 kunmap_atomic(array);
438 static void nfs_readdir_page_set_eof(struct page *page)
440 struct nfs_cache_array *array;
442 array = kmap_atomic(page);
443 nfs_readdir_array_set_eof(array);
444 kunmap_atomic(array);
447 static struct page *nfs_readdir_page_get_next(struct address_space *mapping,
448 u64 cookie, u64 change_attr)
450 pgoff_t index = nfs_readdir_page_cookie_hash(cookie);
453 page = grab_cache_page_nowait(mapping, index);
456 nfs_readdir_page_init_and_validate(page, cookie, change_attr);
457 if (nfs_readdir_page_last_cookie(page) != cookie)
458 nfs_readdir_page_reinit_array(page, cookie, change_attr);
463 int is_32bit_api(void)
466 return in_compat_syscall();
468 return (BITS_PER_LONG == 32);
473 bool nfs_readdir_use_cookie(const struct file *filp)
475 if ((filp->f_mode & FMODE_32BITHASH) ||
476 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
481 static void nfs_readdir_seek_next_array(struct nfs_cache_array *array,
482 struct nfs_readdir_descriptor *desc)
484 if (array->page_full) {
485 desc->last_cookie = array->last_cookie;
486 desc->current_index += array->size;
487 desc->cache_entry_index = 0;
490 desc->last_cookie = nfs_readdir_array_index_cookie(array);
493 static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc)
495 desc->current_index = 0;
496 desc->last_cookie = 0;
497 desc->page_index = 0;
500 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
501 struct nfs_readdir_descriptor *desc)
503 loff_t diff = desc->ctx->pos - desc->current_index;
508 if (diff >= array->size) {
509 if (array->page_is_eof)
511 nfs_readdir_seek_next_array(array, desc);
515 index = (unsigned int)diff;
516 desc->dir_cookie = array->array[index].cookie;
517 desc->cache_entry_index = index;
524 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
527 if (!array->cookies_are_ordered)
529 /* Optimisation for monotonically increasing cookies */
530 if (cookie >= array->last_cookie)
532 if (array->size && cookie < array->array[0].cookie)
537 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
538 struct nfs_readdir_descriptor *desc)
541 int status = -EAGAIN;
543 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
546 for (i = 0; i < array->size; i++) {
547 if (array->array[i].cookie == desc->dir_cookie) {
548 if (nfs_readdir_use_cookie(desc->file))
549 desc->ctx->pos = desc->dir_cookie;
551 desc->ctx->pos = desc->current_index + i;
552 desc->cache_entry_index = i;
557 if (array->page_is_eof) {
558 status = -EBADCOOKIE;
559 if (desc->dir_cookie == array->last_cookie)
562 nfs_readdir_seek_next_array(array, desc);
566 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
568 struct nfs_cache_array *array;
571 array = kmap_atomic(desc->page);
573 if (desc->dir_cookie == 0)
574 status = nfs_readdir_search_for_pos(array, desc);
576 status = nfs_readdir_search_for_cookie(array, desc);
578 kunmap_atomic(array);
582 /* Fill a page with xdr information before transferring to the cache page */
583 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
584 __be32 *verf, u64 cookie,
585 struct page **pages, size_t bufsize,
588 struct inode *inode = file_inode(desc->file);
589 struct nfs_readdir_arg arg = {
590 .dentry = file_dentry(desc->file),
591 .cred = desc->file->f_cred,
598 struct nfs_readdir_res res = {
601 unsigned long timestamp, gencount;
606 gencount = nfs_inc_attr_generation_counter();
607 desc->dir_verifier = nfs_save_change_attribute(inode);
608 error = NFS_PROTO(inode)->readdir(&arg, &res);
610 /* We requested READDIRPLUS, but the server doesn't grok it */
611 if (error == -ENOTSUPP && desc->plus) {
612 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
613 desc->plus = arg.plus = false;
618 desc->timestamp = timestamp;
619 desc->gencount = gencount;
624 static int xdr_decode(struct nfs_readdir_descriptor *desc,
625 struct nfs_entry *entry, struct xdr_stream *xdr)
627 struct inode *inode = file_inode(desc->file);
630 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
633 entry->fattr->time_start = desc->timestamp;
634 entry->fattr->gencount = desc->gencount;
638 /* Match file and dirent using either filehandle or fileid
639 * Note: caller is responsible for checking the fsid
642 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
645 struct nfs_inode *nfsi;
647 if (d_really_is_negative(dentry))
650 inode = d_inode(dentry);
651 if (is_bad_inode(inode) || NFS_STALE(inode))
655 if (entry->fattr->fileid != nfsi->fileid)
657 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
662 #define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL)
664 static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx,
665 unsigned int cache_hits,
666 unsigned int cache_misses)
668 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
671 cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD)
677 * This function is called by the getattr code to request the
678 * use of readdirplus to accelerate any future lookups in the same
681 void nfs_readdir_record_entry_cache_hit(struct inode *dir)
683 struct nfs_inode *nfsi = NFS_I(dir);
684 struct nfs_open_dir_context *ctx;
686 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
687 S_ISDIR(dir->i_mode)) {
689 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
690 atomic_inc(&ctx->cache_hits);
696 * This function is mainly for use by nfs_getattr().
698 * If this is an 'ls -l', we want to force use of readdirplus.
700 void nfs_readdir_record_entry_cache_miss(struct inode *dir)
702 struct nfs_inode *nfsi = NFS_I(dir);
703 struct nfs_open_dir_context *ctx;
705 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
706 S_ISDIR(dir->i_mode)) {
708 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
709 atomic_inc(&ctx->cache_misses);
714 static void nfs_lookup_advise_force_readdirplus(struct inode *dir,
717 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
719 if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL))
721 nfs_readdir_record_entry_cache_miss(dir);
725 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
726 unsigned long dir_verifier)
728 struct qstr filename = QSTR_INIT(entry->name, entry->len);
729 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
730 struct dentry *dentry;
731 struct dentry *alias;
735 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
737 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
739 if (filename.len == 0)
741 /* Validate that the name doesn't contain any illegal '\0' */
742 if (strnlen(filename.name, filename.len) != filename.len)
745 if (strnchr(filename.name, filename.len, '/'))
747 if (filename.name[0] == '.') {
748 if (filename.len == 1)
750 if (filename.len == 2 && filename.name[1] == '.')
753 filename.hash = full_name_hash(parent, filename.name, filename.len);
755 dentry = d_lookup(parent, &filename);
758 dentry = d_alloc_parallel(parent, &filename, &wq);
762 if (!d_in_lookup(dentry)) {
763 /* Is there a mountpoint here? If so, just exit */
764 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
765 &entry->fattr->fsid))
767 if (nfs_same_file(dentry, entry)) {
768 if (!entry->fh->size)
770 nfs_set_verifier(dentry, dir_verifier);
771 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
773 nfs_setsecurity(d_inode(dentry), entry->fattr);
774 trace_nfs_readdir_lookup_revalidate(d_inode(parent),
778 trace_nfs_readdir_lookup_revalidate_failed(
779 d_inode(parent), dentry, 0);
780 d_invalidate(dentry);
786 if (!entry->fh->size) {
787 d_lookup_done(dentry);
791 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
792 alias = d_splice_alias(inode, dentry);
793 d_lookup_done(dentry);
800 nfs_set_verifier(dentry, dir_verifier);
801 trace_nfs_readdir_lookup(d_inode(parent), dentry, 0);
806 static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc,
807 struct nfs_entry *entry,
808 struct xdr_stream *stream)
812 if (entry->fattr->label)
813 entry->fattr->label->len = NFS4_MAXLABELLEN;
814 ret = xdr_decode(desc, entry, stream);
815 if (ret || !desc->plus)
817 nfs_prime_dcache(file_dentry(desc->file), entry, desc->dir_verifier);
821 /* Perform conversion from xdr to cache array */
822 static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc,
823 struct nfs_entry *entry,
824 struct page **xdr_pages, unsigned int buflen,
825 struct page **arrays, size_t narrays,
828 struct address_space *mapping = desc->file->f_mapping;
829 struct xdr_stream stream;
831 struct page *scratch, *new, *page = *arrays;
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_page_array_append(page, entry, &cookie);
848 if (status != -ENOSPC)
851 if (page->mapping != mapping) {
854 new = nfs_readdir_page_array_alloc(cookie, GFP_KERNEL);
858 *arrays = page = new;
860 new = nfs_readdir_page_get_next(mapping, cookie,
865 nfs_readdir_page_unlock_and_put(page);
868 desc->page_index_max++;
869 status = nfs_readdir_page_array_append(page, entry, &cookie);
870 } while (!status && !entry->eof);
876 nfs_readdir_page_set_eof(page);
885 while (!nfs_readdir_entry_decode(desc, entry, &stream))
890 nfs_readdir_page_unlock_and_put(page);
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 page **arrays, size_t narrays)
934 struct page *page = *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_page_last_cookie(page);
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_page_filler(desc, entry, pages, pglen,
966 arrays, narrays, change_attr);
968 nfs_readdir_page_set_eof(page);
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_page_put(struct nfs_readdir_descriptor *desc)
982 put_page(desc->page);
987 nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
989 unlock_page(desc->page);
990 nfs_readdir_page_put(desc);
994 nfs_readdir_page_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 page = nfs_readdir_page_get_locked(mapping, cookie, change_attr);
1004 if (desc->clear_cache && !nfs_readdir_page_needs_filling(page))
1005 nfs_readdir_page_reinit_array(page, 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->page = nfs_readdir_page_get_cached(desc);
1023 if (nfs_readdir_page_needs_filling(desc->page)) {
1024 /* Grow the dtsize if we had to go back for more pages */
1025 if (desc->page_index == desc->page_index_max)
1026 nfs_grow_dtsize(desc);
1027 desc->page_index_max = desc->page_index;
1028 trace_nfs_readdir_cache_fill(desc->file, nfsi->cookieverf,
1030 desc->page->index, desc->dtsize);
1031 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
1034 nfs_readdir_page_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_page_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);
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;
1087 array = kmap(desc->page);
1088 for (i = desc->cache_entry_index; i < array->size; i++) {
1089 struct nfs_cache_array_entry *ent;
1091 ent = &array->array[i];
1092 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
1093 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
1097 memcpy(desc->verf, verf, sizeof(desc->verf));
1098 if (i == array->size - 1) {
1099 desc->dir_cookie = array->last_cookie;
1100 nfs_readdir_seek_next_array(array, desc);
1102 desc->dir_cookie = array->array[i + 1].cookie;
1103 desc->last_cookie = array->array[0].cookie;
1105 if (nfs_readdir_use_cookie(file))
1106 desc->ctx->pos = desc->dir_cookie;
1110 if (array->page_is_eof)
1111 desc->eof = !desc->eob;
1114 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1115 (unsigned long long)desc->dir_cookie);
1119 * If we cannot find a cookie in our cache, we suspect that this is
1120 * because it points to a deleted file, so we ask the server to return
1121 * whatever it thinks is the next entry. We then feed this to filldir.
1122 * If all goes well, we should then be able to find our way round the
1123 * cache on the next call to readdir_search_pagecache();
1125 * NOTE: we cannot add the anonymous page to the pagecache because
1126 * the data it contains might not be page aligned. Besides,
1127 * we should already have a complete representation of the
1128 * directory in the page cache by the time we get here.
1130 static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1132 struct page **arrays;
1134 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1135 int status = -ENOMEM;
1137 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1138 (unsigned long long)desc->dir_cookie);
1140 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1143 arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL);
1147 desc->page_index = 0;
1148 desc->cache_entry_index = 0;
1149 desc->last_cookie = desc->dir_cookie;
1150 desc->page_index_max = 0;
1152 trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie,
1155 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1157 trace_nfs_readdir_uncached_done(file_inode(desc->file), status);
1161 for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1162 desc->page = arrays[i];
1163 nfs_do_filldir(desc, verf);
1168 * Grow the dtsize if we have to go back for more pages,
1169 * or shrink it if we're reading too many.
1173 nfs_grow_dtsize(desc);
1174 else if (desc->buffer_fills == 1 &&
1175 i < (desc->page_index_max >> 1))
1176 nfs_shrink_dtsize(desc);
1179 for (i = 0; i < sz && arrays[i]; i++)
1180 nfs_readdir_page_array_free(arrays[i]);
1182 if (!nfs_readdir_use_cookie(desc->file))
1183 nfs_readdir_rewind_search(desc);
1184 desc->page_index_max = -1;
1186 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1190 #define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1192 static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1193 struct nfs_readdir_descriptor *desc,
1194 unsigned int cache_misses,
1197 if (desc->ctx->pos == 0 || !desc->plus)
1199 if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1201 trace_nfs_readdir_force_readdirplus(inode);
1205 /* The file offset position represents the dirent entry number. A
1206 last cookie cache takes care of the common case of reading the
1209 static int nfs_readdir(struct file *file, struct dir_context *ctx)
1211 struct dentry *dentry = file_dentry(file);
1212 struct inode *inode = d_inode(dentry);
1213 struct nfs_inode *nfsi = NFS_I(inode);
1214 struct nfs_open_dir_context *dir_ctx = file->private_data;
1215 struct nfs_readdir_descriptor *desc;
1216 unsigned int cache_hits, cache_misses;
1220 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1221 file, (long long)ctx->pos);
1222 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1225 * ctx->pos points to the dirent entry number.
1226 * *desc->dir_cookie has the cookie for the next entry. We have
1227 * to either find the entry with the appropriate number or
1228 * revalidate the cookie.
1230 nfs_revalidate_mapping(inode, file->f_mapping);
1233 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1238 desc->page_index_max = -1;
1240 spin_lock(&file->f_lock);
1241 desc->dir_cookie = dir_ctx->dir_cookie;
1242 desc->page_index = dir_ctx->page_index;
1243 desc->last_cookie = dir_ctx->last_cookie;
1244 desc->attr_gencount = dir_ctx->attr_gencount;
1245 desc->eof = dir_ctx->eof;
1246 nfs_set_dtsize(desc, dir_ctx->dtsize);
1247 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1248 cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0);
1249 cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0);
1250 force_clear = dir_ctx->force_clear;
1251 spin_unlock(&file->f_lock);
1258 desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses);
1259 force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1261 desc->clear_cache = force_clear;
1264 res = readdir_search_pagecache(desc);
1266 if (res == -EBADCOOKIE) {
1268 /* This means either end of directory */
1269 if (desc->dir_cookie && !desc->eof) {
1270 /* Or that the server has 'lost' a cookie */
1271 res = uncached_readdir(desc);
1274 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1279 if (res == -ETOOSMALL && desc->plus) {
1280 nfs_zap_caches(inode);
1288 nfs_do_filldir(desc, nfsi->cookieverf);
1289 nfs_readdir_page_unlock_and_put_cached(desc);
1290 if (desc->page_index == desc->page_index_max)
1291 desc->clear_cache = force_clear;
1292 } while (!desc->eob && !desc->eof);
1294 spin_lock(&file->f_lock);
1295 dir_ctx->dir_cookie = desc->dir_cookie;
1296 dir_ctx->last_cookie = desc->last_cookie;
1297 dir_ctx->attr_gencount = desc->attr_gencount;
1298 dir_ctx->page_index = desc->page_index;
1299 dir_ctx->force_clear = force_clear;
1300 dir_ctx->eof = desc->eof;
1301 dir_ctx->dtsize = desc->dtsize;
1302 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1303 spin_unlock(&file->f_lock);
1308 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1312 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1314 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1316 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1317 filp, offset, whence);
1325 spin_lock(&filp->f_lock);
1330 spin_lock(&filp->f_lock);
1331 offset += filp->f_pos;
1333 spin_unlock(&filp->f_lock);
1337 if (offset != filp->f_pos) {
1338 filp->f_pos = offset;
1339 dir_ctx->page_index = 0;
1340 if (!nfs_readdir_use_cookie(filp)) {
1341 dir_ctx->dir_cookie = 0;
1342 dir_ctx->last_cookie = 0;
1344 dir_ctx->dir_cookie = offset;
1345 dir_ctx->last_cookie = offset;
1347 dir_ctx->eof = false;
1349 spin_unlock(&filp->f_lock);
1354 * All directory operations under NFS are synchronous, so fsync()
1355 * is a dummy operation.
1357 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1360 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1362 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1367 * nfs_force_lookup_revalidate - Mark the directory as having changed
1368 * @dir: pointer to directory inode
1370 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1371 * full lookup on all child dentries of 'dir' whenever a change occurs
1372 * on the server that might have invalidated our dcache.
1374 * Note that we reserve bit '0' as a tag to let us know when a dentry
1375 * was revalidated while holding a delegation on its inode.
1377 * The caller should be holding dir->i_lock
1379 void nfs_force_lookup_revalidate(struct inode *dir)
1381 NFS_I(dir)->cache_change_attribute += 2;
1383 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1386 * nfs_verify_change_attribute - Detects NFS remote directory changes
1387 * @dir: pointer to parent directory inode
1388 * @verf: previously saved change attribute
1390 * Return "false" if the verifiers doesn't match the change attribute.
1391 * This would usually indicate that the directory contents have changed on
1392 * the server, and that any dentries need revalidating.
1394 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1396 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1399 static void nfs_set_verifier_delegated(unsigned long *verf)
1404 #if IS_ENABLED(CONFIG_NFS_V4)
1405 static void nfs_unset_verifier_delegated(unsigned long *verf)
1409 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1411 static bool nfs_test_verifier_delegated(unsigned long verf)
1416 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1418 return nfs_test_verifier_delegated(dentry->d_time);
1421 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1423 struct inode *inode = d_inode(dentry);
1424 struct inode *dir = d_inode(dentry->d_parent);
1426 if (!nfs_verify_change_attribute(dir, verf))
1428 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1429 nfs_set_verifier_delegated(&verf);
1430 dentry->d_time = verf;
1434 * nfs_set_verifier - save a parent directory verifier in the dentry
1435 * @dentry: pointer to dentry
1436 * @verf: verifier to save
1438 * Saves the parent directory verifier in @dentry. If the inode has
1439 * a delegation, we also tag the dentry as having been revalidated
1440 * while holding a delegation so that we know we don't have to
1441 * look it up again after a directory change.
1443 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1446 spin_lock(&dentry->d_lock);
1447 nfs_set_verifier_locked(dentry, verf);
1448 spin_unlock(&dentry->d_lock);
1450 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1452 #if IS_ENABLED(CONFIG_NFS_V4)
1454 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1455 * @inode: pointer to inode
1457 * Iterates through the dentries in the inode alias list and clears
1458 * the tag used to indicate that the dentry has been revalidated
1459 * while holding a delegation.
1460 * This function is intended for use when the delegation is being
1461 * returned or revoked.
1463 void nfs_clear_verifier_delegated(struct inode *inode)
1465 struct dentry *alias;
1469 spin_lock(&inode->i_lock);
1470 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1471 spin_lock(&alias->d_lock);
1472 nfs_unset_verifier_delegated(&alias->d_time);
1473 spin_unlock(&alias->d_lock);
1475 spin_unlock(&inode->i_lock);
1477 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1478 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1480 static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1482 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) &&
1483 d_really_is_negative(dentry))
1484 return dentry->d_time == inode_peek_iversion_raw(dir);
1485 return nfs_verify_change_attribute(dir, dentry->d_time);
1489 * A check for whether or not the parent directory has changed.
1490 * In the case it has, we assume that the dentries are untrustworthy
1491 * and may need to be looked up again.
1492 * If rcu_walk prevents us from performing a full check, return 0.
1494 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1497 if (IS_ROOT(dentry))
1499 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1501 if (!nfs_dentry_verify_change(dir, dentry))
1503 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1504 if (nfs_mapping_need_revalidate_inode(dir)) {
1507 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1510 if (!nfs_dentry_verify_change(dir, dentry))
1516 * Use intent information to check whether or not we're going to do
1517 * an O_EXCL create using this path component.
1519 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1521 if (NFS_PROTO(dir)->version == 2)
1523 return flags & LOOKUP_EXCL;
1527 * Inode and filehandle revalidation for lookups.
1529 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1530 * or if the intent information indicates that we're about to open this
1531 * particular file and the "nocto" mount flag is not set.
1535 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1537 struct nfs_server *server = NFS_SERVER(inode);
1540 if (IS_AUTOMOUNT(inode))
1543 if (flags & LOOKUP_OPEN) {
1544 switch (inode->i_mode & S_IFMT) {
1546 /* A NFSv4 OPEN will revalidate later */
1547 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1551 if (server->flags & NFS_MOUNT_NOCTO)
1553 /* NFS close-to-open cache consistency validation */
1558 /* VFS wants an on-the-wire revalidation */
1559 if (flags & LOOKUP_REVAL)
1562 if (inode->i_nlink > 0 ||
1563 (inode->i_nlink == 0 &&
1564 test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1569 if (flags & LOOKUP_RCU)
1571 ret = __nfs_revalidate_inode(server, inode);
1577 static void nfs_mark_dir_for_revalidate(struct inode *inode)
1579 spin_lock(&inode->i_lock);
1580 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1581 spin_unlock(&inode->i_lock);
1585 * We judge how long we want to trust negative
1586 * dentries by looking at the parent inode mtime.
1588 * If parent mtime has changed, we revalidate, else we wait for a
1589 * period corresponding to the parent's attribute cache timeout value.
1591 * If LOOKUP_RCU prevents us from performing a full check, return 1
1592 * suggesting a reval is needed.
1594 * Note that when creating a new file, or looking up a rename target,
1595 * then it shouldn't be necessary to revalidate a negative dentry.
1598 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1601 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1603 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1605 /* Case insensitive server? Revalidate negative dentries */
1606 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1608 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1612 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1613 struct inode *inode, int error)
1620 * We can't d_drop the root of a disconnected tree:
1621 * its d_hash is on the s_anon list and d_drop() would hide
1622 * it from shrink_dcache_for_unmount(), leading to busy
1623 * inodes on unmount and further oopses.
1625 if (inode && IS_ROOT(dentry))
1629 trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error);
1634 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1638 if (nfs_neg_need_reval(dir, dentry, flags)) {
1639 if (flags & LOOKUP_RCU)
1643 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1647 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1648 struct inode *inode)
1650 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1651 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1654 static int nfs_lookup_revalidate_dentry(struct inode *dir,
1655 struct dentry *dentry,
1656 struct inode *inode, unsigned int flags)
1658 struct nfs_fh *fhandle;
1659 struct nfs_fattr *fattr;
1660 unsigned long dir_verifier;
1663 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1666 fhandle = nfs_alloc_fhandle();
1667 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1668 if (fhandle == NULL || fattr == NULL)
1671 dir_verifier = nfs_save_change_attribute(dir);
1672 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1680 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1686 /* Request help from readdirplus */
1687 nfs_lookup_advise_force_readdirplus(dir, flags);
1690 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1692 if (nfs_refresh_inode(inode, fattr) < 0)
1695 nfs_setsecurity(inode, fattr);
1696 nfs_set_verifier(dentry, dir_verifier);
1700 nfs_free_fattr(fattr);
1701 nfs_free_fhandle(fhandle);
1704 * If the lookup failed despite the dentry change attribute being
1705 * a match, then we should revalidate the directory cache.
1707 if (!ret && nfs_dentry_verify_change(dir, dentry))
1708 nfs_mark_dir_for_revalidate(dir);
1709 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1713 * This is called every time the dcache has a lookup hit,
1714 * and we should check whether we can really trust that
1717 * NOTE! The hit can be a negative hit too, don't assume
1720 * If the parent directory is seen to have changed, we throw out the
1721 * cached dentry and do a new lookup.
1724 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1727 struct inode *inode;
1730 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1731 inode = d_inode(dentry);
1734 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1736 if (is_bad_inode(inode)) {
1737 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1742 if (nfs_verifier_is_delegated(dentry))
1743 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1745 /* Force a full look up iff the parent directory has changed */
1746 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1747 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1748 error = nfs_lookup_verify_inode(inode, flags);
1750 if (error == -ESTALE)
1751 nfs_mark_dir_for_revalidate(dir);
1757 if (flags & LOOKUP_RCU)
1760 if (NFS_STALE(inode))
1763 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
1765 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1767 if (flags & LOOKUP_RCU)
1769 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1773 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1774 int (*reval)(struct inode *, struct dentry *, unsigned int))
1776 struct dentry *parent;
1780 if (flags & LOOKUP_RCU) {
1781 parent = READ_ONCE(dentry->d_parent);
1782 dir = d_inode_rcu(parent);
1785 ret = reval(dir, dentry, flags);
1786 if (parent != READ_ONCE(dentry->d_parent))
1789 parent = dget_parent(dentry);
1790 ret = reval(d_inode(parent), dentry, flags);
1796 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1798 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1802 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1803 * when we don't really care about the dentry name. This is called when a
1804 * pathwalk ends on a dentry that was not found via a normal lookup in the
1805 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1807 * In this situation, we just want to verify that the inode itself is OK
1808 * since the dentry might have changed on the server.
1810 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1812 struct inode *inode = d_inode(dentry);
1816 * I believe we can only get a negative dentry here in the case of a
1817 * procfs-style symlink. Just assume it's correct for now, but we may
1818 * eventually need to do something more here.
1821 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1826 if (is_bad_inode(inode)) {
1827 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1832 error = nfs_lookup_verify_inode(inode, flags);
1833 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1834 __func__, inode->i_ino, error ? "invalid" : "valid");
1839 * This is called from dput() when d_count is going to 0.
1841 static int nfs_dentry_delete(const struct dentry *dentry)
1843 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1844 dentry, dentry->d_flags);
1846 /* Unhash any dentry with a stale inode */
1847 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1850 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1851 /* Unhash it, so that ->d_iput() would be called */
1854 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1855 /* Unhash it, so that ancestors of killed async unlink
1856 * files will be cleaned up during umount */
1863 /* Ensure that we revalidate inode->i_nlink */
1864 static void nfs_drop_nlink(struct inode *inode)
1866 spin_lock(&inode->i_lock);
1867 /* drop the inode if we're reasonably sure this is the last link */
1868 if (inode->i_nlink > 0)
1870 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1871 nfs_set_cache_invalid(
1872 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1873 NFS_INO_INVALID_NLINK);
1874 spin_unlock(&inode->i_lock);
1878 * Called when the dentry loses inode.
1879 * We use it to clean up silly-renamed files.
1881 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1883 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1884 nfs_complete_unlink(dentry, inode);
1885 nfs_drop_nlink(inode);
1890 static void nfs_d_release(struct dentry *dentry)
1892 /* free cached devname value, if it survived that far */
1893 if (unlikely(dentry->d_fsdata)) {
1894 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1897 kfree(dentry->d_fsdata);
1901 const struct dentry_operations nfs_dentry_operations = {
1902 .d_revalidate = nfs_lookup_revalidate,
1903 .d_weak_revalidate = nfs_weak_revalidate,
1904 .d_delete = nfs_dentry_delete,
1905 .d_iput = nfs_dentry_iput,
1906 .d_automount = nfs_d_automount,
1907 .d_release = nfs_d_release,
1909 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1911 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1914 struct inode *inode = NULL;
1915 struct nfs_fh *fhandle = NULL;
1916 struct nfs_fattr *fattr = NULL;
1917 unsigned long dir_verifier;
1920 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1921 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1923 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1924 return ERR_PTR(-ENAMETOOLONG);
1927 * If we're doing an exclusive create, optimize away the lookup
1928 * but don't hash the dentry.
1930 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1933 res = ERR_PTR(-ENOMEM);
1934 fhandle = nfs_alloc_fhandle();
1935 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir));
1936 if (fhandle == NULL || fattr == NULL)
1939 dir_verifier = nfs_save_change_attribute(dir);
1940 trace_nfs_lookup_enter(dir, dentry, flags);
1941 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1942 if (error == -ENOENT) {
1943 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1944 dir_verifier = inode_peek_iversion_raw(dir);
1948 res = ERR_PTR(error);
1951 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1952 res = ERR_CAST(inode);
1956 /* Notify readdir to use READDIRPLUS */
1957 nfs_lookup_advise_force_readdirplus(dir, flags);
1960 res = d_splice_alias(inode, dentry);
1966 nfs_set_verifier(dentry, dir_verifier);
1968 trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res));
1969 nfs_free_fattr(fattr);
1970 nfs_free_fhandle(fhandle);
1973 EXPORT_SYMBOL_GPL(nfs_lookup);
1975 void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
1977 /* Case insensitive server? Revalidate dentries */
1978 if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
1979 d_prune_aliases(inode);
1981 EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
1983 #if IS_ENABLED(CONFIG_NFS_V4)
1984 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1986 const struct dentry_operations nfs4_dentry_operations = {
1987 .d_revalidate = nfs4_lookup_revalidate,
1988 .d_weak_revalidate = nfs_weak_revalidate,
1989 .d_delete = nfs_dentry_delete,
1990 .d_iput = nfs_dentry_iput,
1991 .d_automount = nfs_d_automount,
1992 .d_release = nfs_d_release,
1994 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1996 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
1998 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
2001 static int do_open(struct inode *inode, struct file *filp)
2003 nfs_fscache_open_file(inode, filp);
2007 static int nfs_finish_open(struct nfs_open_context *ctx,
2008 struct dentry *dentry,
2009 struct file *file, unsigned open_flags)
2013 err = finish_open(file, dentry, do_open);
2016 if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
2017 nfs_file_set_open_context(file, ctx);
2024 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2025 struct file *file, unsigned open_flags,
2028 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2029 struct nfs_open_context *ctx;
2031 struct iattr attr = { .ia_valid = ATTR_OPEN };
2032 struct inode *inode;
2033 unsigned int lookup_flags = 0;
2034 unsigned long dir_verifier;
2035 bool switched = false;
2039 /* Expect a negative dentry */
2040 BUG_ON(d_inode(dentry));
2042 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2043 dir->i_sb->s_id, dir->i_ino, dentry);
2045 err = nfs_check_flags(open_flags);
2049 /* NFS only supports OPEN on regular files */
2050 if ((open_flags & O_DIRECTORY)) {
2051 if (!d_in_lookup(dentry)) {
2053 * Hashed negative dentry with O_DIRECTORY: dentry was
2054 * revalidated and is fine, no need to perform lookup
2059 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2063 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
2064 return -ENAMETOOLONG;
2066 if (open_flags & O_CREAT) {
2067 struct nfs_server *server = NFS_SERVER(dir);
2069 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2070 mode &= ~current_umask();
2072 attr.ia_valid |= ATTR_MODE;
2073 attr.ia_mode = mode;
2075 if (open_flags & O_TRUNC) {
2076 attr.ia_valid |= ATTR_SIZE;
2080 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2083 dentry = d_alloc_parallel(dentry->d_parent,
2084 &dentry->d_name, &wq);
2086 return PTR_ERR(dentry);
2087 if (unlikely(!d_in_lookup(dentry)))
2088 return finish_no_open(file, dentry);
2091 ctx = create_nfs_open_context(dentry, open_flags, file);
2096 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
2097 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
2099 file->f_mode |= FMODE_CREATED;
2100 if (IS_ERR(inode)) {
2101 err = PTR_ERR(inode);
2102 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2103 put_nfs_open_context(ctx);
2107 d_splice_alias(NULL, dentry);
2108 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
2109 dir_verifier = inode_peek_iversion_raw(dir);
2111 dir_verifier = nfs_save_change_attribute(dir);
2112 nfs_set_verifier(dentry, dir_verifier);
2118 if (!(open_flags & O_NOFOLLOW))
2127 file->f_mode |= FMODE_CAN_ODIRECT;
2129 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
2130 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2131 put_nfs_open_context(ctx);
2133 if (unlikely(switched)) {
2134 d_lookup_done(dentry);
2140 res = nfs_lookup(dir, dentry, lookup_flags);
2142 inode = d_inode(dentry);
2143 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2144 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2145 res = ERR_PTR(-ENOTDIR);
2146 else if (inode && S_ISREG(inode->i_mode))
2147 res = ERR_PTR(-EOPENSTALE);
2148 } else if (!IS_ERR(res)) {
2149 inode = d_inode(res);
2150 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2151 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2153 res = ERR_PTR(-ENOTDIR);
2154 } else if (inode && S_ISREG(inode->i_mode)) {
2156 res = ERR_PTR(-EOPENSTALE);
2160 d_lookup_done(dentry);
2167 return PTR_ERR(res);
2168 return finish_no_open(file, res);
2170 EXPORT_SYMBOL_GPL(nfs_atomic_open);
2173 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2176 struct inode *inode;
2178 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2180 if (d_mountpoint(dentry))
2183 inode = d_inode(dentry);
2185 /* We can't create new files in nfs_open_revalidate(), so we
2186 * optimize away revalidation of negative dentries.
2191 if (nfs_verifier_is_delegated(dentry))
2192 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2194 /* NFS only supports OPEN on regular files */
2195 if (!S_ISREG(inode->i_mode))
2198 /* We cannot do exclusive creation on a positive dentry */
2199 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2202 /* Check if the directory changed */
2203 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2206 /* Let f_op->open() actually open (and revalidate) the file */
2209 if (flags & LOOKUP_RCU)
2211 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
2214 return nfs_do_lookup_revalidate(dir, dentry, flags);
2217 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2219 return __nfs_lookup_revalidate(dentry, flags,
2220 nfs4_do_lookup_revalidate);
2223 #endif /* CONFIG_NFSV4 */
2226 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2227 struct nfs_fattr *fattr)
2229 struct dentry *parent = dget_parent(dentry);
2230 struct inode *dir = d_inode(parent);
2231 struct inode *inode;
2237 if (fhandle->size == 0) {
2238 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
2242 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2243 if (!(fattr->valid & NFS_ATTR_FATTR)) {
2244 struct nfs_server *server = NFS_SB(dentry->d_sb);
2245 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2250 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2251 d = d_splice_alias(inode, dentry);
2259 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2262 * Code common to create, mkdir, and mknod.
2264 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2265 struct nfs_fattr *fattr)
2269 d = nfs_add_or_obtain(dentry, fhandle, fattr);
2273 /* Callers don't care */
2277 EXPORT_SYMBOL_GPL(nfs_instantiate);
2280 * Following a failed create operation, we drop the dentry rather
2281 * than retain a negative dentry. This avoids a problem in the event
2282 * that the operation succeeded on the server, but an error in the
2283 * reply path made it appear to have failed.
2285 int nfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2286 struct dentry *dentry, umode_t mode, bool excl)
2289 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2292 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2293 dir->i_sb->s_id, dir->i_ino, dentry);
2295 attr.ia_mode = mode;
2296 attr.ia_valid = ATTR_MODE;
2298 trace_nfs_create_enter(dir, dentry, open_flags);
2299 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2300 trace_nfs_create_exit(dir, dentry, open_flags, error);
2308 EXPORT_SYMBOL_GPL(nfs_create);
2311 * See comments for nfs_proc_create regarding failed operations.
2314 nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2315 struct dentry *dentry, umode_t mode, dev_t rdev)
2320 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2321 dir->i_sb->s_id, dir->i_ino, dentry);
2323 attr.ia_mode = mode;
2324 attr.ia_valid = ATTR_MODE;
2326 trace_nfs_mknod_enter(dir, dentry);
2327 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2328 trace_nfs_mknod_exit(dir, dentry, status);
2336 EXPORT_SYMBOL_GPL(nfs_mknod);
2339 * See comments for nfs_proc_create regarding failed operations.
2341 int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2342 struct dentry *dentry, umode_t mode)
2347 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2348 dir->i_sb->s_id, dir->i_ino, dentry);
2350 attr.ia_valid = ATTR_MODE;
2351 attr.ia_mode = mode | S_IFDIR;
2353 trace_nfs_mkdir_enter(dir, dentry);
2354 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2355 trace_nfs_mkdir_exit(dir, dentry, error);
2363 EXPORT_SYMBOL_GPL(nfs_mkdir);
2365 static void nfs_dentry_handle_enoent(struct dentry *dentry)
2367 if (simple_positive(dentry))
2371 static void nfs_dentry_remove_handle_error(struct inode *dir,
2372 struct dentry *dentry, int error)
2377 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2380 nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2381 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2385 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2389 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2390 dir->i_sb->s_id, dir->i_ino, dentry);
2392 trace_nfs_rmdir_enter(dir, dentry);
2393 if (d_really_is_positive(dentry)) {
2394 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2395 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2396 /* Ensure the VFS deletes this inode */
2399 clear_nlink(d_inode(dentry));
2402 nfs_dentry_handle_enoent(dentry);
2404 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2406 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2407 nfs_dentry_remove_handle_error(dir, dentry, error);
2408 trace_nfs_rmdir_exit(dir, dentry, error);
2412 EXPORT_SYMBOL_GPL(nfs_rmdir);
2415 * Remove a file after making sure there are no pending writes,
2416 * and after checking that the file has only one user.
2418 * We invalidate the attribute cache and free the inode prior to the operation
2419 * to avoid possible races if the server reuses the inode.
2421 static int nfs_safe_remove(struct dentry *dentry)
2423 struct inode *dir = d_inode(dentry->d_parent);
2424 struct inode *inode = d_inode(dentry);
2427 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2429 /* If the dentry was sillyrenamed, we simply call d_delete() */
2430 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2435 trace_nfs_remove_enter(dir, dentry);
2436 if (inode != NULL) {
2437 error = NFS_PROTO(dir)->remove(dir, dentry);
2439 nfs_drop_nlink(inode);
2441 error = NFS_PROTO(dir)->remove(dir, dentry);
2442 if (error == -ENOENT)
2443 nfs_dentry_handle_enoent(dentry);
2444 trace_nfs_remove_exit(dir, dentry, error);
2449 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2450 * belongs to an active ".nfs..." file and we return -EBUSY.
2452 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2454 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2457 int need_rehash = 0;
2459 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2460 dir->i_ino, dentry);
2462 trace_nfs_unlink_enter(dir, dentry);
2463 spin_lock(&dentry->d_lock);
2464 if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2465 &NFS_I(d_inode(dentry))->flags)) {
2466 spin_unlock(&dentry->d_lock);
2467 /* Start asynchronous writeout of the inode */
2468 write_inode_now(d_inode(dentry), 0);
2469 error = nfs_sillyrename(dir, dentry);
2472 if (!d_unhashed(dentry)) {
2476 spin_unlock(&dentry->d_lock);
2477 error = nfs_safe_remove(dentry);
2478 nfs_dentry_remove_handle_error(dir, dentry, error);
2482 trace_nfs_unlink_exit(dir, dentry, error);
2485 EXPORT_SYMBOL_GPL(nfs_unlink);
2488 * To create a symbolic link, most file systems instantiate a new inode,
2489 * add a page to it containing the path, then write it out to the disk
2490 * using prepare_write/commit_write.
2492 * Unfortunately the NFS client can't create the in-core inode first
2493 * because it needs a file handle to create an in-core inode (see
2494 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2495 * symlink request has completed on the server.
2497 * So instead we allocate a raw page, copy the symname into it, then do
2498 * the SYMLINK request with the page as the buffer. If it succeeds, we
2499 * now have a new file handle and can instantiate an in-core NFS inode
2500 * and move the raw page into its mapping.
2502 int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
2503 struct dentry *dentry, const char *symname)
2508 unsigned int pathlen = strlen(symname);
2511 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2512 dir->i_ino, dentry, symname);
2514 if (pathlen > PAGE_SIZE)
2515 return -ENAMETOOLONG;
2517 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2518 attr.ia_valid = ATTR_MODE;
2520 page = alloc_page(GFP_USER);
2524 kaddr = page_address(page);
2525 memcpy(kaddr, symname, pathlen);
2526 if (pathlen < PAGE_SIZE)
2527 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2529 trace_nfs_symlink_enter(dir, dentry);
2530 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2531 trace_nfs_symlink_exit(dir, dentry, error);
2533 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2534 dir->i_sb->s_id, dir->i_ino,
2535 dentry, symname, error);
2541 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2544 * No big deal if we can't add this page to the page cache here.
2545 * READLINK will get the missing page from the server if needed.
2547 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2549 SetPageUptodate(page);
2552 * add_to_page_cache_lru() grabs an extra page refcount.
2553 * Drop it here to avoid leaking this page later.
2561 EXPORT_SYMBOL_GPL(nfs_symlink);
2564 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2566 struct inode *inode = d_inode(old_dentry);
2569 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2570 old_dentry, dentry);
2572 trace_nfs_link_enter(inode, dir, dentry);
2574 if (S_ISREG(inode->i_mode))
2575 nfs_sync_inode(inode);
2576 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2578 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2580 d_add(dentry, inode);
2582 trace_nfs_link_exit(inode, dir, dentry, error);
2585 EXPORT_SYMBOL_GPL(nfs_link);
2589 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2590 * different file handle for the same inode after a rename (e.g. when
2591 * moving to a different directory). A fail-safe method to do so would
2592 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2593 * rename the old file using the sillyrename stuff. This way, the original
2594 * file in old_dir will go away when the last process iput()s the inode.
2598 * It actually works quite well. One needs to have the possibility for
2599 * at least one ".nfs..." file in each directory the file ever gets
2600 * moved or linked to which happens automagically with the new
2601 * implementation that only depends on the dcache stuff instead of
2602 * using the inode layer
2604 * Unfortunately, things are a little more complicated than indicated
2605 * above. For a cross-directory move, we want to make sure we can get
2606 * rid of the old inode after the operation. This means there must be
2607 * no pending writes (if it's a file), and the use count must be 1.
2608 * If these conditions are met, we can drop the dentries before doing
2611 int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
2612 struct dentry *old_dentry, struct inode *new_dir,
2613 struct dentry *new_dentry, unsigned int flags)
2615 struct inode *old_inode = d_inode(old_dentry);
2616 struct inode *new_inode = d_inode(new_dentry);
2617 struct dentry *dentry = NULL, *rehash = NULL;
2618 struct rpc_task *task;
2624 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2625 old_dentry, new_dentry,
2626 d_count(new_dentry));
2628 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2630 * For non-directories, check whether the target is busy and if so,
2631 * make a copy of the dentry and then do a silly-rename. If the
2632 * silly-rename succeeds, the copied dentry is hashed and becomes
2635 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2637 * To prevent any new references to the target during the
2638 * rename, we unhash the dentry in advance.
2640 if (!d_unhashed(new_dentry)) {
2642 rehash = new_dentry;
2645 if (d_count(new_dentry) > 2) {
2648 /* copy the target dentry's name */
2649 dentry = d_alloc(new_dentry->d_parent,
2650 &new_dentry->d_name);
2654 /* silly-rename the existing target ... */
2655 err = nfs_sillyrename(new_dir, new_dentry);
2659 new_dentry = dentry;
2665 if (S_ISREG(old_inode->i_mode))
2666 nfs_sync_inode(old_inode);
2667 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2669 error = PTR_ERR(task);
2673 error = rpc_wait_for_completion_task(task);
2675 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2676 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2679 error = task->tk_status;
2681 /* Ensure the inode attributes are revalidated */
2683 spin_lock(&old_inode->i_lock);
2684 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2685 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2686 NFS_INO_INVALID_CTIME |
2687 NFS_INO_REVAL_FORCED);
2688 spin_unlock(&old_inode->i_lock);
2693 trace_nfs_rename_exit(old_dir, old_dentry,
2694 new_dir, new_dentry, error);
2696 if (new_inode != NULL)
2697 nfs_drop_nlink(new_inode);
2699 * The d_move() should be here instead of in an async RPC completion
2700 * handler because we need the proper locks to move the dentry. If
2701 * we're interrupted by a signal, the async RPC completion handler
2702 * should mark the directories for revalidation.
2704 d_move(old_dentry, new_dentry);
2705 nfs_set_verifier(old_dentry,
2706 nfs_save_change_attribute(new_dir));
2707 } else if (error == -ENOENT)
2708 nfs_dentry_handle_enoent(old_dentry);
2710 /* new dentry created? */
2715 EXPORT_SYMBOL_GPL(nfs_rename);
2717 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2718 static LIST_HEAD(nfs_access_lru_list);
2719 static atomic_long_t nfs_access_nr_entries;
2721 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2722 module_param(nfs_access_max_cachesize, ulong, 0644);
2723 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2725 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2727 put_group_info(entry->group_info);
2728 kfree_rcu(entry, rcu_head);
2729 smp_mb__before_atomic();
2730 atomic_long_dec(&nfs_access_nr_entries);
2731 smp_mb__after_atomic();
2734 static void nfs_access_free_list(struct list_head *head)
2736 struct nfs_access_entry *cache;
2738 while (!list_empty(head)) {
2739 cache = list_entry(head->next, struct nfs_access_entry, lru);
2740 list_del(&cache->lru);
2741 nfs_access_free_entry(cache);
2745 static unsigned long
2746 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2749 struct nfs_inode *nfsi, *next;
2750 struct nfs_access_entry *cache;
2753 spin_lock(&nfs_access_lru_lock);
2754 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2755 struct inode *inode;
2757 if (nr_to_scan-- == 0)
2759 inode = &nfsi->vfs_inode;
2760 spin_lock(&inode->i_lock);
2761 if (list_empty(&nfsi->access_cache_entry_lru))
2762 goto remove_lru_entry;
2763 cache = list_entry(nfsi->access_cache_entry_lru.next,
2764 struct nfs_access_entry, lru);
2765 list_move(&cache->lru, &head);
2766 rb_erase(&cache->rb_node, &nfsi->access_cache);
2768 if (!list_empty(&nfsi->access_cache_entry_lru))
2769 list_move_tail(&nfsi->access_cache_inode_lru,
2770 &nfs_access_lru_list);
2773 list_del_init(&nfsi->access_cache_inode_lru);
2774 smp_mb__before_atomic();
2775 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2776 smp_mb__after_atomic();
2778 spin_unlock(&inode->i_lock);
2780 spin_unlock(&nfs_access_lru_lock);
2781 nfs_access_free_list(&head);
2786 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2788 int nr_to_scan = sc->nr_to_scan;
2789 gfp_t gfp_mask = sc->gfp_mask;
2791 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2793 return nfs_do_access_cache_scan(nr_to_scan);
2798 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2800 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2804 nfs_access_cache_enforce_limit(void)
2806 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2808 unsigned int nr_to_scan;
2810 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2813 diff = nr_entries - nfs_access_max_cachesize;
2814 if (diff < nr_to_scan)
2816 nfs_do_access_cache_scan(nr_to_scan);
2819 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2821 struct rb_root *root_node = &nfsi->access_cache;
2823 struct nfs_access_entry *entry;
2825 /* Unhook entries from the cache */
2826 while ((n = rb_first(root_node)) != NULL) {
2827 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2828 rb_erase(n, root_node);
2829 list_move(&entry->lru, head);
2831 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2834 void nfs_access_zap_cache(struct inode *inode)
2838 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2840 /* Remove from global LRU init */
2841 spin_lock(&nfs_access_lru_lock);
2842 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2843 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2845 spin_lock(&inode->i_lock);
2846 __nfs_access_zap_cache(NFS_I(inode), &head);
2847 spin_unlock(&inode->i_lock);
2848 spin_unlock(&nfs_access_lru_lock);
2849 nfs_access_free_list(&head);
2851 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2853 static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2855 struct group_info *ga, *gb;
2858 if (uid_lt(a->fsuid, b->fsuid))
2860 if (uid_gt(a->fsuid, b->fsuid))
2863 if (gid_lt(a->fsgid, b->fsgid))
2865 if (gid_gt(a->fsgid, b->fsgid))
2876 if (ga->ngroups < gb->ngroups)
2878 if (ga->ngroups > gb->ngroups)
2881 for (g = 0; g < ga->ngroups; g++) {
2882 if (gid_lt(ga->gid[g], gb->gid[g]))
2884 if (gid_gt(ga->gid[g], gb->gid[g]))
2890 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2892 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2895 struct nfs_access_entry *entry =
2896 rb_entry(n, struct nfs_access_entry, rb_node);
2897 int cmp = access_cmp(cred, entry);
2909 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
2911 struct nfs_inode *nfsi = NFS_I(inode);
2912 struct nfs_access_entry *cache;
2916 spin_lock(&inode->i_lock);
2918 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2920 cache = nfs_access_search_rbtree(inode, cred);
2924 /* Found an entry, is our attribute cache valid? */
2925 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2932 spin_unlock(&inode->i_lock);
2933 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2936 spin_lock(&inode->i_lock);
2939 *mask = cache->mask;
2940 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2943 spin_unlock(&inode->i_lock);
2946 spin_unlock(&inode->i_lock);
2947 nfs_access_zap_cache(inode);
2951 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
2953 /* Only check the most recently returned cache entry,
2954 * but do it without locking.
2956 struct nfs_inode *nfsi = NFS_I(inode);
2957 struct nfs_access_entry *cache;
2959 struct list_head *lh;
2962 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2964 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
2965 cache = list_entry(lh, struct nfs_access_entry, lru);
2966 if (lh == &nfsi->access_cache_entry_lru ||
2967 access_cmp(cred, cache) != 0)
2971 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2973 *mask = cache->mask;
2980 int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
2981 u32 *mask, bool may_block)
2985 status = nfs_access_get_cached_rcu(inode, cred, mask);
2987 status = nfs_access_get_cached_locked(inode, cred, mask,
2992 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
2994 static void nfs_access_add_rbtree(struct inode *inode,
2995 struct nfs_access_entry *set,
2996 const struct cred *cred)
2998 struct nfs_inode *nfsi = NFS_I(inode);
2999 struct rb_root *root_node = &nfsi->access_cache;
3000 struct rb_node **p = &root_node->rb_node;
3001 struct rb_node *parent = NULL;
3002 struct nfs_access_entry *entry;
3005 spin_lock(&inode->i_lock);
3006 while (*p != NULL) {
3008 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3009 cmp = access_cmp(cred, entry);
3012 p = &parent->rb_left;
3014 p = &parent->rb_right;
3018 rb_link_node(&set->rb_node, parent, p);
3019 rb_insert_color(&set->rb_node, root_node);
3020 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3021 spin_unlock(&inode->i_lock);
3024 rb_replace_node(parent, &set->rb_node, root_node);
3025 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3026 list_del(&entry->lru);
3027 spin_unlock(&inode->i_lock);
3028 nfs_access_free_entry(entry);
3031 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3032 const struct cred *cred)
3034 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
3037 RB_CLEAR_NODE(&cache->rb_node);
3038 cache->fsuid = cred->fsuid;
3039 cache->fsgid = cred->fsgid;
3040 cache->group_info = get_group_info(cred->group_info);
3041 cache->mask = set->mask;
3043 /* The above field assignments must be visible
3044 * before this item appears on the lru. We cannot easily
3045 * use rcu_assign_pointer, so just force the memory barrier.
3048 nfs_access_add_rbtree(inode, cache, cred);
3050 /* Update accounting */
3051 smp_mb__before_atomic();
3052 atomic_long_inc(&nfs_access_nr_entries);
3053 smp_mb__after_atomic();
3055 /* Add inode to global LRU list */
3056 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3057 spin_lock(&nfs_access_lru_lock);
3058 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
3059 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
3060 &nfs_access_lru_list);
3061 spin_unlock(&nfs_access_lru_lock);
3063 nfs_access_cache_enforce_limit();
3065 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3067 #define NFS_MAY_READ (NFS_ACCESS_READ)
3068 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3069 NFS_ACCESS_EXTEND | \
3071 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3073 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3074 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3075 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3077 nfs_access_calc_mask(u32 access_result, umode_t umode)
3081 if (access_result & NFS_MAY_READ)
3083 if (S_ISDIR(umode)) {
3084 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3086 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3088 } else if (S_ISREG(umode)) {
3089 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3091 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3093 } else if (access_result & NFS_MAY_WRITE)
3098 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3100 entry->mask = access_result;
3102 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3104 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3106 struct nfs_access_entry cache;
3107 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3108 int cache_mask = -1;
3111 trace_nfs_access_enter(inode);
3113 status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3122 * Determine which access bits we want to ask for...
3124 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3125 nfs_access_xattr_mask(NFS_SERVER(inode));
3126 if (S_ISDIR(inode->i_mode))
3127 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3129 cache.mask |= NFS_ACCESS_EXECUTE;
3130 status = NFS_PROTO(inode)->access(inode, &cache, cred);
3132 if (status == -ESTALE) {
3133 if (!S_ISDIR(inode->i_mode))
3134 nfs_set_inode_stale(inode);
3136 nfs_zap_caches(inode);
3140 nfs_access_add_cache(inode, &cache, cred);
3142 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
3143 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3146 trace_nfs_access_exit(inode, mask, cache_mask, status);
3150 static int nfs_open_permission_mask(int openflags)
3154 if (openflags & __FMODE_EXEC) {
3155 /* ONLY check exec rights */
3158 if ((openflags & O_ACCMODE) != O_WRONLY)
3160 if ((openflags & O_ACCMODE) != O_RDONLY)
3167 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3169 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
3171 EXPORT_SYMBOL_GPL(nfs_may_open);
3173 static int nfs_execute_ok(struct inode *inode, int mask)
3175 struct nfs_server *server = NFS_SERVER(inode);
3178 if (S_ISDIR(inode->i_mode))
3180 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3181 if (mask & MAY_NOT_BLOCK)
3183 ret = __nfs_revalidate_inode(server, inode);
3185 if (ret == 0 && !execute_ok(inode))
3190 int nfs_permission(struct user_namespace *mnt_userns,
3191 struct inode *inode,
3194 const struct cred *cred = current_cred();
3197 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
3199 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3201 /* Is this sys_access() ? */
3202 if (mask & (MAY_ACCESS | MAY_CHDIR))
3205 switch (inode->i_mode & S_IFMT) {
3209 if ((mask & MAY_OPEN) &&
3210 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3215 * Optimize away all write operations, since the server
3216 * will check permissions when we perform the op.
3218 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3223 if (!NFS_PROTO(inode)->access)
3226 res = nfs_do_access(inode, cred, mask);
3228 if (!res && (mask & MAY_EXEC))
3229 res = nfs_execute_ok(inode, mask);
3231 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3232 inode->i_sb->s_id, inode->i_ino, mask, res);
3235 if (mask & MAY_NOT_BLOCK)
3238 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3239 NFS_INO_INVALID_OTHER);
3241 res = generic_permission(&init_user_ns, inode, mask);
3244 EXPORT_SYMBOL_GPL(nfs_permission);