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/xxhash.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 page*);
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 .freepage = nfs_readdir_clear_array,
73 #define NFS_INIT_DTSIZE PAGE_SIZE
75 static struct nfs_open_dir_context *
76 alloc_nfs_open_dir_context(struct inode *dir)
78 struct nfs_inode *nfsi = NFS_I(dir);
79 struct nfs_open_dir_context *ctx;
81 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL_ACCOUNT);
83 ctx->attr_gencount = nfsi->attr_gencount;
84 ctx->dtsize = NFS_INIT_DTSIZE;
85 spin_lock(&dir->i_lock);
86 if (list_empty(&nfsi->open_files) &&
87 (nfsi->cache_validity & NFS_INO_DATA_INVAL_DEFER))
88 nfs_set_cache_invalid(dir,
89 NFS_INO_INVALID_DATA |
90 NFS_INO_REVAL_FORCED);
91 list_add_tail_rcu(&ctx->list, &nfsi->open_files);
92 memcpy(ctx->verf, nfsi->cookieverf, sizeof(ctx->verf));
93 spin_unlock(&dir->i_lock);
96 return ERR_PTR(-ENOMEM);
99 static void put_nfs_open_dir_context(struct inode *dir, struct nfs_open_dir_context *ctx)
101 spin_lock(&dir->i_lock);
102 list_del_rcu(&ctx->list);
103 spin_unlock(&dir->i_lock);
104 kfree_rcu(ctx, rcu_head);
111 nfs_opendir(struct inode *inode, struct file *filp)
114 struct nfs_open_dir_context *ctx;
116 dfprintk(FILE, "NFS: open dir(%pD2)\n", filp);
118 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
120 ctx = alloc_nfs_open_dir_context(inode);
125 filp->private_data = ctx;
131 nfs_closedir(struct inode *inode, struct file *filp)
133 put_nfs_open_dir_context(file_inode(filp), filp->private_data);
137 struct nfs_cache_array_entry {
141 unsigned int name_len;
142 unsigned char d_type;
145 struct nfs_cache_array {
149 unsigned char 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_page_reinit_array(struct page *page, u64 last_cookie,
234 nfs_readdir_clear_array(page);
235 nfs_readdir_page_init_array(page, last_cookie, change_attr);
239 nfs_readdir_page_array_alloc(u64 last_cookie, gfp_t gfp_flags)
241 struct page *page = alloc_page(gfp_flags);
243 nfs_readdir_page_init_array(page, last_cookie, 0);
247 static void nfs_readdir_page_array_free(struct page *page)
250 nfs_readdir_clear_array(page);
255 static u64 nfs_readdir_array_index_cookie(struct nfs_cache_array *array)
257 return array->size == 0 ? array->last_cookie : array->array[0].cookie;
260 static void nfs_readdir_array_set_eof(struct nfs_cache_array *array)
262 array->page_is_eof = 1;
263 array->page_full = 1;
266 static bool nfs_readdir_array_is_full(struct nfs_cache_array *array)
268 return array->page_full;
272 * the caller is responsible for freeing qstr.name
273 * when called by nfs_readdir_add_to_array, the strings will be freed in
274 * nfs_clear_readdir_array()
276 static const char *nfs_readdir_copy_name(const char *name, unsigned int len)
278 const char *ret = kmemdup_nul(name, len, GFP_KERNEL);
281 * Avoid a kmemleak false positive. The pointer to the name is stored
282 * in a page cache page which kmemleak does not scan.
285 kmemleak_not_leak(ret);
289 static size_t nfs_readdir_array_maxentries(void)
291 return (PAGE_SIZE - sizeof(struct nfs_cache_array)) /
292 sizeof(struct nfs_cache_array_entry);
296 * Check that the next array entry lies entirely within the page bounds
298 static int nfs_readdir_array_can_expand(struct nfs_cache_array *array)
300 if (array->page_full)
302 if (array->size == nfs_readdir_array_maxentries()) {
303 array->page_full = 1;
309 static int nfs_readdir_page_array_append(struct page *page,
310 const struct nfs_entry *entry,
313 struct nfs_cache_array *array;
314 struct nfs_cache_array_entry *cache_entry;
318 name = nfs_readdir_copy_name(entry->name, entry->len);
320 array = kmap_atomic(page);
323 ret = nfs_readdir_array_can_expand(array);
329 cache_entry = &array->array[array->size];
330 cache_entry->cookie = array->last_cookie;
331 cache_entry->ino = entry->ino;
332 cache_entry->d_type = entry->d_type;
333 cache_entry->name_len = entry->len;
334 cache_entry->name = name;
335 array->last_cookie = entry->cookie;
336 if (array->last_cookie <= cache_entry->cookie)
337 array->cookies_are_ordered = 0;
340 nfs_readdir_array_set_eof(array);
342 *cookie = array->last_cookie;
343 kunmap_atomic(array);
347 #define NFS_READDIR_COOKIE_MASK (U32_MAX >> 14)
349 * Hash algorithm allowing content addressible access to sequences
350 * of directory cookies. Content is addressed by the value of the
351 * cookie index of the first readdir entry in a page.
353 * The xxhash algorithm is chosen because it is fast, and is supposed
354 * to result in a decent flat distribution of hashes.
356 * We then select only the first 18 bits to avoid issues with excessive
357 * memory use for the page cache XArray. 18 bits should allow the caching
358 * of 262144 pages of sequences of readdir entries. Since each page holds
359 * 127 readdir entries for a typical 64-bit system, that works out to a
360 * cache of ~ 33 million entries per directory.
362 static pgoff_t nfs_readdir_page_cookie_hash(u64 cookie)
366 return xxhash(&cookie, sizeof(cookie), 0) & NFS_READDIR_COOKIE_MASK;
369 static bool nfs_readdir_page_validate(struct page *page, u64 last_cookie,
372 struct nfs_cache_array *array = kmap_atomic(page);
375 if (array->change_attr != change_attr)
377 if (nfs_readdir_array_index_cookie(array) != last_cookie)
379 kunmap_atomic(array);
383 static void nfs_readdir_page_unlock_and_put(struct page *page)
389 static void nfs_readdir_page_init_and_validate(struct page *page, u64 cookie,
392 if (PageUptodate(page)) {
393 if (nfs_readdir_page_validate(page, cookie, change_attr))
395 nfs_readdir_clear_array(page);
397 nfs_readdir_page_init_array(page, cookie, change_attr);
398 SetPageUptodate(page);
401 static struct page *nfs_readdir_page_get_locked(struct address_space *mapping,
402 u64 cookie, u64 change_attr)
404 pgoff_t index = nfs_readdir_page_cookie_hash(cookie);
407 page = grab_cache_page(mapping, index);
410 nfs_readdir_page_init_and_validate(page, cookie, change_attr);
414 static u64 nfs_readdir_page_last_cookie(struct page *page)
416 struct nfs_cache_array *array;
419 array = kmap_atomic(page);
420 ret = array->last_cookie;
421 kunmap_atomic(array);
425 static bool nfs_readdir_page_needs_filling(struct page *page)
427 struct nfs_cache_array *array;
430 array = kmap_atomic(page);
431 ret = !nfs_readdir_array_is_full(array);
432 kunmap_atomic(array);
436 static void nfs_readdir_page_set_eof(struct page *page)
438 struct nfs_cache_array *array;
440 array = kmap_atomic(page);
441 nfs_readdir_array_set_eof(array);
442 kunmap_atomic(array);
445 static struct page *nfs_readdir_page_get_next(struct address_space *mapping,
446 u64 cookie, u64 change_attr)
448 pgoff_t index = nfs_readdir_page_cookie_hash(cookie);
451 page = grab_cache_page_nowait(mapping, index);
454 nfs_readdir_page_init_and_validate(page, cookie, change_attr);
455 if (nfs_readdir_page_last_cookie(page) != cookie)
456 nfs_readdir_page_reinit_array(page, cookie, change_attr);
461 int is_32bit_api(void)
464 return in_compat_syscall();
466 return (BITS_PER_LONG == 32);
471 bool nfs_readdir_use_cookie(const struct file *filp)
473 if ((filp->f_mode & FMODE_32BITHASH) ||
474 (!(filp->f_mode & FMODE_64BITHASH) && is_32bit_api()))
479 static void nfs_readdir_seek_next_array(struct nfs_cache_array *array,
480 struct nfs_readdir_descriptor *desc)
482 if (array->page_full) {
483 desc->last_cookie = array->last_cookie;
484 desc->current_index += array->size;
485 desc->cache_entry_index = 0;
488 desc->last_cookie = nfs_readdir_array_index_cookie(array);
491 static void nfs_readdir_rewind_search(struct nfs_readdir_descriptor *desc)
493 desc->current_index = 0;
494 desc->last_cookie = 0;
495 desc->page_index = 0;
498 static int nfs_readdir_search_for_pos(struct nfs_cache_array *array,
499 struct nfs_readdir_descriptor *desc)
501 loff_t diff = desc->ctx->pos - desc->current_index;
506 if (diff >= array->size) {
507 if (array->page_is_eof)
509 nfs_readdir_seek_next_array(array, desc);
513 index = (unsigned int)diff;
514 desc->dir_cookie = array->array[index].cookie;
515 desc->cache_entry_index = index;
522 static bool nfs_readdir_array_cookie_in_range(struct nfs_cache_array *array,
525 if (!array->cookies_are_ordered)
527 /* Optimisation for monotonically increasing cookies */
528 if (cookie >= array->last_cookie)
530 if (array->size && cookie < array->array[0].cookie)
535 static int nfs_readdir_search_for_cookie(struct nfs_cache_array *array,
536 struct nfs_readdir_descriptor *desc)
539 int status = -EAGAIN;
541 if (!nfs_readdir_array_cookie_in_range(array, desc->dir_cookie))
544 for (i = 0; i < array->size; i++) {
545 if (array->array[i].cookie == desc->dir_cookie) {
546 if (nfs_readdir_use_cookie(desc->file))
547 desc->ctx->pos = desc->dir_cookie;
549 desc->ctx->pos = desc->current_index + i;
550 desc->cache_entry_index = i;
555 if (array->page_is_eof) {
556 status = -EBADCOOKIE;
557 if (desc->dir_cookie == array->last_cookie)
560 nfs_readdir_seek_next_array(array, desc);
564 static int nfs_readdir_search_array(struct nfs_readdir_descriptor *desc)
566 struct nfs_cache_array *array;
569 array = kmap_atomic(desc->page);
571 if (desc->dir_cookie == 0)
572 status = nfs_readdir_search_for_pos(array, desc);
574 status = nfs_readdir_search_for_cookie(array, desc);
576 kunmap_atomic(array);
580 /* Fill a page with xdr information before transferring to the cache page */
581 static int nfs_readdir_xdr_filler(struct nfs_readdir_descriptor *desc,
582 __be32 *verf, u64 cookie,
583 struct page **pages, size_t bufsize,
586 struct inode *inode = file_inode(desc->file);
587 struct nfs_readdir_arg arg = {
588 .dentry = file_dentry(desc->file),
589 .cred = desc->file->f_cred,
596 struct nfs_readdir_res res = {
599 unsigned long timestamp, gencount;
604 gencount = nfs_inc_attr_generation_counter();
605 desc->dir_verifier = nfs_save_change_attribute(inode);
606 error = NFS_PROTO(inode)->readdir(&arg, &res);
608 /* We requested READDIRPLUS, but the server doesn't grok it */
609 if (error == -ENOTSUPP && desc->plus) {
610 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
611 desc->plus = arg.plus = false;
616 desc->timestamp = timestamp;
617 desc->gencount = gencount;
622 static int xdr_decode(struct nfs_readdir_descriptor *desc,
623 struct nfs_entry *entry, struct xdr_stream *xdr)
625 struct inode *inode = file_inode(desc->file);
628 error = NFS_PROTO(inode)->decode_dirent(xdr, entry, desc->plus);
631 entry->fattr->time_start = desc->timestamp;
632 entry->fattr->gencount = desc->gencount;
636 /* Match file and dirent using either filehandle or fileid
637 * Note: caller is responsible for checking the fsid
640 int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
643 struct nfs_inode *nfsi;
645 if (d_really_is_negative(dentry))
648 inode = d_inode(dentry);
649 if (is_bad_inode(inode) || NFS_STALE(inode))
653 if (entry->fattr->fileid != nfsi->fileid)
655 if (entry->fh->size && nfs_compare_fh(entry->fh, &nfsi->fh) != 0)
660 #define NFS_READDIR_CACHE_USAGE_THRESHOLD (8UL)
662 static bool nfs_use_readdirplus(struct inode *dir, struct dir_context *ctx,
663 unsigned int cache_hits,
664 unsigned int cache_misses)
666 if (!nfs_server_capable(dir, NFS_CAP_READDIRPLUS))
669 cache_hits + cache_misses > NFS_READDIR_CACHE_USAGE_THRESHOLD)
675 * This function is called by the getattr code to request the
676 * use of readdirplus to accelerate any future lookups in the same
679 void nfs_readdir_record_entry_cache_hit(struct inode *dir)
681 struct nfs_inode *nfsi = NFS_I(dir);
682 struct nfs_open_dir_context *ctx;
684 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
685 S_ISDIR(dir->i_mode)) {
687 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
688 atomic_inc(&ctx->cache_hits);
694 * This function is mainly for use by nfs_getattr().
696 * If this is an 'ls -l', we want to force use of readdirplus.
698 void nfs_readdir_record_entry_cache_miss(struct inode *dir)
700 struct nfs_inode *nfsi = NFS_I(dir);
701 struct nfs_open_dir_context *ctx;
703 if (nfs_server_capable(dir, NFS_CAP_READDIRPLUS) &&
704 S_ISDIR(dir->i_mode)) {
706 list_for_each_entry_rcu (ctx, &nfsi->open_files, list)
707 atomic_inc(&ctx->cache_misses);
712 static void nfs_lookup_advise_force_readdirplus(struct inode *dir,
715 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
717 if (flags & (LOOKUP_EXCL | LOOKUP_PARENT | LOOKUP_REVAL))
719 nfs_readdir_record_entry_cache_miss(dir);
723 void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry,
724 unsigned long dir_verifier)
726 struct qstr filename = QSTR_INIT(entry->name, entry->len);
727 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
728 struct dentry *dentry;
729 struct dentry *alias;
733 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FILEID))
735 if (!(entry->fattr->valid & NFS_ATTR_FATTR_FSID))
737 if (filename.len == 0)
739 /* Validate that the name doesn't contain any illegal '\0' */
740 if (strnlen(filename.name, filename.len) != filename.len)
743 if (strnchr(filename.name, filename.len, '/'))
745 if (filename.name[0] == '.') {
746 if (filename.len == 1)
748 if (filename.len == 2 && filename.name[1] == '.')
751 filename.hash = full_name_hash(parent, filename.name, filename.len);
753 dentry = d_lookup(parent, &filename);
756 dentry = d_alloc_parallel(parent, &filename, &wq);
760 if (!d_in_lookup(dentry)) {
761 /* Is there a mountpoint here? If so, just exit */
762 if (!nfs_fsid_equal(&NFS_SB(dentry->d_sb)->fsid,
763 &entry->fattr->fsid))
765 if (nfs_same_file(dentry, entry)) {
766 if (!entry->fh->size)
768 nfs_set_verifier(dentry, dir_verifier);
769 status = nfs_refresh_inode(d_inode(dentry), entry->fattr);
771 nfs_setsecurity(d_inode(dentry), entry->fattr);
772 trace_nfs_readdir_lookup_revalidate(d_inode(parent),
776 trace_nfs_readdir_lookup_revalidate_failed(
777 d_inode(parent), dentry, 0);
778 d_invalidate(dentry);
784 if (!entry->fh->size) {
785 d_lookup_done(dentry);
789 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
790 alias = d_splice_alias(inode, dentry);
791 d_lookup_done(dentry);
798 nfs_set_verifier(dentry, dir_verifier);
799 trace_nfs_readdir_lookup(d_inode(parent), dentry, 0);
804 static int nfs_readdir_entry_decode(struct nfs_readdir_descriptor *desc,
805 struct nfs_entry *entry,
806 struct xdr_stream *stream)
810 if (entry->fattr->label)
811 entry->fattr->label->len = NFS4_MAXLABELLEN;
812 ret = xdr_decode(desc, entry, stream);
813 if (ret || !desc->plus)
815 nfs_prime_dcache(file_dentry(desc->file), entry, desc->dir_verifier);
819 /* Perform conversion from xdr to cache array */
820 static int nfs_readdir_page_filler(struct nfs_readdir_descriptor *desc,
821 struct nfs_entry *entry,
822 struct page **xdr_pages, unsigned int buflen,
823 struct page **arrays, size_t narrays,
826 struct address_space *mapping = desc->file->f_mapping;
827 struct xdr_stream stream;
829 struct page *scratch, *new, *page = *arrays;
833 scratch = alloc_page(GFP_KERNEL);
837 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
838 xdr_set_scratch_page(&stream, scratch);
841 status = nfs_readdir_entry_decode(desc, entry, &stream);
845 status = nfs_readdir_page_array_append(page, entry, &cookie);
846 if (status != -ENOSPC)
849 if (page->mapping != mapping) {
852 new = nfs_readdir_page_array_alloc(cookie, GFP_KERNEL);
856 *arrays = page = new;
858 new = nfs_readdir_page_get_next(mapping, cookie,
863 nfs_readdir_page_unlock_and_put(page);
866 desc->page_index_max++;
867 status = nfs_readdir_page_array_append(page, entry, &cookie);
868 } while (!status && !entry->eof);
874 nfs_readdir_page_set_eof(page);
883 while (!nfs_readdir_entry_decode(desc, entry, &stream))
888 nfs_readdir_page_unlock_and_put(page);
894 static void nfs_readdir_free_pages(struct page **pages, size_t npages)
897 put_page(pages[npages]);
902 * nfs_readdir_alloc_pages() will allocate pages that must be freed with a call
903 * to nfs_readdir_free_pages()
905 static struct page **nfs_readdir_alloc_pages(size_t npages)
910 pages = kmalloc_array(npages, sizeof(*pages), GFP_KERNEL);
913 for (i = 0; i < npages; i++) {
914 struct page *page = alloc_page(GFP_KERNEL);
922 nfs_readdir_free_pages(pages, i);
926 static int nfs_readdir_xdr_to_array(struct nfs_readdir_descriptor *desc,
927 __be32 *verf_arg, __be32 *verf_res,
928 struct page **arrays, size_t narrays)
932 struct page *page = *arrays;
933 struct nfs_entry *entry;
935 struct inode *inode = file_inode(desc->file);
936 unsigned int dtsize = desc->dtsize;
938 int status = -ENOMEM;
940 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
943 entry->cookie = nfs_readdir_page_last_cookie(page);
944 entry->fh = nfs_alloc_fhandle();
945 entry->fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
946 entry->server = NFS_SERVER(inode);
947 if (entry->fh == NULL || entry->fattr == NULL)
950 array_size = (dtsize + PAGE_SIZE - 1) >> PAGE_SHIFT;
951 pages = nfs_readdir_alloc_pages(array_size);
955 change_attr = inode_peek_iversion_raw(inode);
956 status = nfs_readdir_xdr_filler(desc, verf_arg, entry->cookie, pages,
963 status = nfs_readdir_page_filler(desc, entry, pages, pglen,
964 arrays, narrays, change_attr);
966 nfs_readdir_page_set_eof(page);
967 desc->buffer_fills++;
970 nfs_readdir_free_pages(pages, array_size);
972 nfs_free_fattr(entry->fattr);
973 nfs_free_fhandle(entry->fh);
978 static void nfs_readdir_page_put(struct nfs_readdir_descriptor *desc)
980 put_page(desc->page);
985 nfs_readdir_page_unlock_and_put_cached(struct nfs_readdir_descriptor *desc)
987 unlock_page(desc->page);
988 nfs_readdir_page_put(desc);
992 nfs_readdir_page_get_cached(struct nfs_readdir_descriptor *desc)
994 struct address_space *mapping = desc->file->f_mapping;
995 u64 change_attr = inode_peek_iversion_raw(mapping->host);
996 u64 cookie = desc->last_cookie;
999 page = nfs_readdir_page_get_locked(mapping, cookie, change_attr);
1002 if (desc->clear_cache && !nfs_readdir_page_needs_filling(page))
1003 nfs_readdir_page_reinit_array(page, cookie, change_attr);
1008 * Returns 0 if desc->dir_cookie was found on page desc->page_index
1009 * and locks the page to prevent removal from the page cache.
1011 static int find_and_lock_cache_page(struct nfs_readdir_descriptor *desc)
1013 struct inode *inode = file_inode(desc->file);
1014 struct nfs_inode *nfsi = NFS_I(inode);
1015 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1018 desc->page = nfs_readdir_page_get_cached(desc);
1021 if (nfs_readdir_page_needs_filling(desc->page)) {
1022 /* Grow the dtsize if we had to go back for more pages */
1023 if (desc->page_index == desc->page_index_max)
1024 nfs_grow_dtsize(desc);
1025 desc->page_index_max = desc->page_index;
1026 trace_nfs_readdir_cache_fill(desc->file, nfsi->cookieverf,
1028 desc->page->index, desc->dtsize);
1029 res = nfs_readdir_xdr_to_array(desc, nfsi->cookieverf, verf,
1032 nfs_readdir_page_unlock_and_put_cached(desc);
1033 trace_nfs_readdir_cache_fill_done(inode, res);
1034 if (res == -EBADCOOKIE || res == -ENOTSYNC) {
1035 invalidate_inode_pages2(desc->file->f_mapping);
1036 nfs_readdir_rewind_search(desc);
1037 trace_nfs_readdir_invalidate_cache_range(
1038 inode, 0, MAX_LFS_FILESIZE);
1044 * Set the cookie verifier if the page cache was empty
1046 if (desc->last_cookie == 0 &&
1047 memcmp(nfsi->cookieverf, verf, sizeof(nfsi->cookieverf))) {
1048 memcpy(nfsi->cookieverf, verf,
1049 sizeof(nfsi->cookieverf));
1050 invalidate_inode_pages2_range(desc->file->f_mapping, 1,
1052 trace_nfs_readdir_invalidate_cache_range(
1053 inode, 1, MAX_LFS_FILESIZE);
1055 desc->clear_cache = false;
1057 res = nfs_readdir_search_array(desc);
1060 nfs_readdir_page_unlock_and_put_cached(desc);
1064 /* Search for desc->dir_cookie from the beginning of the page cache */
1065 static int readdir_search_pagecache(struct nfs_readdir_descriptor *desc)
1070 res = find_and_lock_cache_page(desc);
1071 } while (res == -EAGAIN);
1076 * Once we've found the start of the dirent within a page: fill 'er up...
1078 static void nfs_do_filldir(struct nfs_readdir_descriptor *desc,
1081 struct file *file = desc->file;
1082 struct nfs_cache_array *array;
1085 array = kmap(desc->page);
1086 for (i = desc->cache_entry_index; i < array->size; i++) {
1087 struct nfs_cache_array_entry *ent;
1089 ent = &array->array[i];
1090 if (!dir_emit(desc->ctx, ent->name, ent->name_len,
1091 nfs_compat_user_ino64(ent->ino), ent->d_type)) {
1095 memcpy(desc->verf, verf, sizeof(desc->verf));
1096 if (i == array->size - 1) {
1097 desc->dir_cookie = array->last_cookie;
1098 nfs_readdir_seek_next_array(array, desc);
1100 desc->dir_cookie = array->array[i + 1].cookie;
1101 desc->last_cookie = array->array[0].cookie;
1103 if (nfs_readdir_use_cookie(file))
1104 desc->ctx->pos = desc->dir_cookie;
1108 if (array->page_is_eof)
1109 desc->eof = !desc->eob;
1112 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %llu\n",
1113 (unsigned long long)desc->dir_cookie);
1117 * If we cannot find a cookie in our cache, we suspect that this is
1118 * because it points to a deleted file, so we ask the server to return
1119 * whatever it thinks is the next entry. We then feed this to filldir.
1120 * If all goes well, we should then be able to find our way round the
1121 * cache on the next call to readdir_search_pagecache();
1123 * NOTE: we cannot add the anonymous page to the pagecache because
1124 * the data it contains might not be page aligned. Besides,
1125 * we should already have a complete representation of the
1126 * directory in the page cache by the time we get here.
1128 static int uncached_readdir(struct nfs_readdir_descriptor *desc)
1130 struct page **arrays;
1132 __be32 verf[NFS_DIR_VERIFIER_SIZE];
1133 int status = -ENOMEM;
1135 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %llu\n",
1136 (unsigned long long)desc->dir_cookie);
1138 arrays = kcalloc(sz, sizeof(*arrays), GFP_KERNEL);
1141 arrays[0] = nfs_readdir_page_array_alloc(desc->dir_cookie, GFP_KERNEL);
1145 desc->page_index = 0;
1146 desc->cache_entry_index = 0;
1147 desc->last_cookie = desc->dir_cookie;
1148 desc->page_index_max = 0;
1150 trace_nfs_readdir_uncached(desc->file, desc->verf, desc->last_cookie,
1153 status = nfs_readdir_xdr_to_array(desc, desc->verf, verf, arrays, sz);
1155 trace_nfs_readdir_uncached_done(file_inode(desc->file), status);
1159 for (i = 0; !desc->eob && i < sz && arrays[i]; i++) {
1160 desc->page = arrays[i];
1161 nfs_do_filldir(desc, verf);
1166 * Grow the dtsize if we have to go back for more pages,
1167 * or shrink it if we're reading too many.
1171 nfs_grow_dtsize(desc);
1172 else if (desc->buffer_fills == 1 &&
1173 i < (desc->page_index_max >> 1))
1174 nfs_shrink_dtsize(desc);
1177 for (i = 0; i < sz && arrays[i]; i++)
1178 nfs_readdir_page_array_free(arrays[i]);
1180 if (!nfs_readdir_use_cookie(desc->file))
1181 nfs_readdir_rewind_search(desc);
1182 desc->page_index_max = -1;
1184 dfprintk(DIRCACHE, "NFS: %s: returns %d\n", __func__, status);
1188 #define NFS_READDIR_CACHE_MISS_THRESHOLD (16UL)
1190 static bool nfs_readdir_handle_cache_misses(struct inode *inode,
1191 struct nfs_readdir_descriptor *desc,
1192 unsigned int cache_misses,
1195 if (desc->ctx->pos == 0 || !desc->plus)
1197 if (cache_misses <= NFS_READDIR_CACHE_MISS_THRESHOLD && !force_clear)
1199 trace_nfs_readdir_force_readdirplus(inode);
1203 /* The file offset position represents the dirent entry number. A
1204 last cookie cache takes care of the common case of reading the
1207 static int nfs_readdir(struct file *file, struct dir_context *ctx)
1209 struct dentry *dentry = file_dentry(file);
1210 struct inode *inode = d_inode(dentry);
1211 struct nfs_inode *nfsi = NFS_I(inode);
1212 struct nfs_open_dir_context *dir_ctx = file->private_data;
1213 struct nfs_readdir_descriptor *desc;
1214 unsigned int cache_hits, cache_misses;
1218 dfprintk(FILE, "NFS: readdir(%pD2) starting at cookie %llu\n",
1219 file, (long long)ctx->pos);
1220 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
1223 * ctx->pos points to the dirent entry number.
1224 * *desc->dir_cookie has the cookie for the next entry. We have
1225 * to either find the entry with the appropriate number or
1226 * revalidate the cookie.
1228 nfs_revalidate_mapping(inode, file->f_mapping);
1231 desc = kzalloc(sizeof(*desc), GFP_KERNEL);
1236 desc->page_index_max = -1;
1238 spin_lock(&file->f_lock);
1239 desc->dir_cookie = dir_ctx->dir_cookie;
1240 desc->page_index = dir_ctx->page_index;
1241 desc->last_cookie = dir_ctx->last_cookie;
1242 desc->attr_gencount = dir_ctx->attr_gencount;
1243 desc->eof = dir_ctx->eof;
1244 nfs_set_dtsize(desc, dir_ctx->dtsize);
1245 memcpy(desc->verf, dir_ctx->verf, sizeof(desc->verf));
1246 cache_hits = atomic_xchg(&dir_ctx->cache_hits, 0);
1247 cache_misses = atomic_xchg(&dir_ctx->cache_misses, 0);
1248 force_clear = dir_ctx->force_clear;
1249 spin_unlock(&file->f_lock);
1256 desc->plus = nfs_use_readdirplus(inode, ctx, cache_hits, cache_misses);
1257 force_clear = nfs_readdir_handle_cache_misses(inode, desc, cache_misses,
1259 desc->clear_cache = force_clear;
1262 res = readdir_search_pagecache(desc);
1264 if (res == -EBADCOOKIE) {
1266 /* This means either end of directory */
1267 if (desc->dir_cookie && !desc->eof) {
1268 /* Or that the server has 'lost' a cookie */
1269 res = uncached_readdir(desc);
1272 if (res == -EBADCOOKIE || res == -ENOTSYNC)
1277 if (res == -ETOOSMALL && desc->plus) {
1278 nfs_zap_caches(inode);
1286 nfs_do_filldir(desc, nfsi->cookieverf);
1287 nfs_readdir_page_unlock_and_put_cached(desc);
1288 if (desc->page_index == desc->page_index_max)
1289 desc->clear_cache = force_clear;
1290 } while (!desc->eob && !desc->eof);
1292 spin_lock(&file->f_lock);
1293 dir_ctx->dir_cookie = desc->dir_cookie;
1294 dir_ctx->last_cookie = desc->last_cookie;
1295 dir_ctx->attr_gencount = desc->attr_gencount;
1296 dir_ctx->page_index = desc->page_index;
1297 dir_ctx->force_clear = force_clear;
1298 dir_ctx->eof = desc->eof;
1299 dir_ctx->dtsize = desc->dtsize;
1300 memcpy(dir_ctx->verf, desc->verf, sizeof(dir_ctx->verf));
1301 spin_unlock(&file->f_lock);
1306 dfprintk(FILE, "NFS: readdir(%pD2) returns %d\n", file, res);
1310 static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int whence)
1312 struct nfs_open_dir_context *dir_ctx = filp->private_data;
1314 dfprintk(FILE, "NFS: llseek dir(%pD2, %lld, %d)\n",
1315 filp, offset, whence);
1323 spin_lock(&filp->f_lock);
1328 spin_lock(&filp->f_lock);
1329 offset += filp->f_pos;
1331 spin_unlock(&filp->f_lock);
1335 if (offset != filp->f_pos) {
1336 filp->f_pos = offset;
1337 dir_ctx->page_index = 0;
1338 if (!nfs_readdir_use_cookie(filp)) {
1339 dir_ctx->dir_cookie = 0;
1340 dir_ctx->last_cookie = 0;
1342 dir_ctx->dir_cookie = offset;
1343 dir_ctx->last_cookie = offset;
1345 dir_ctx->eof = false;
1347 spin_unlock(&filp->f_lock);
1352 * All directory operations under NFS are synchronous, so fsync()
1353 * is a dummy operation.
1355 static int nfs_fsync_dir(struct file *filp, loff_t start, loff_t end,
1358 dfprintk(FILE, "NFS: fsync dir(%pD2) datasync %d\n", filp, datasync);
1360 nfs_inc_stats(file_inode(filp), NFSIOS_VFSFSYNC);
1365 * nfs_force_lookup_revalidate - Mark the directory as having changed
1366 * @dir: pointer to directory inode
1368 * This forces the revalidation code in nfs_lookup_revalidate() to do a
1369 * full lookup on all child dentries of 'dir' whenever a change occurs
1370 * on the server that might have invalidated our dcache.
1372 * Note that we reserve bit '0' as a tag to let us know when a dentry
1373 * was revalidated while holding a delegation on its inode.
1375 * The caller should be holding dir->i_lock
1377 void nfs_force_lookup_revalidate(struct inode *dir)
1379 NFS_I(dir)->cache_change_attribute += 2;
1381 EXPORT_SYMBOL_GPL(nfs_force_lookup_revalidate);
1384 * nfs_verify_change_attribute - Detects NFS remote directory changes
1385 * @dir: pointer to parent directory inode
1386 * @verf: previously saved change attribute
1388 * Return "false" if the verifiers doesn't match the change attribute.
1389 * This would usually indicate that the directory contents have changed on
1390 * the server, and that any dentries need revalidating.
1392 static bool nfs_verify_change_attribute(struct inode *dir, unsigned long verf)
1394 return (verf & ~1UL) == nfs_save_change_attribute(dir);
1397 static void nfs_set_verifier_delegated(unsigned long *verf)
1402 #if IS_ENABLED(CONFIG_NFS_V4)
1403 static void nfs_unset_verifier_delegated(unsigned long *verf)
1407 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1409 static bool nfs_test_verifier_delegated(unsigned long verf)
1414 static bool nfs_verifier_is_delegated(struct dentry *dentry)
1416 return nfs_test_verifier_delegated(dentry->d_time);
1419 static void nfs_set_verifier_locked(struct dentry *dentry, unsigned long verf)
1421 struct inode *inode = d_inode(dentry);
1422 struct inode *dir = d_inode(dentry->d_parent);
1424 if (!nfs_verify_change_attribute(dir, verf))
1426 if (inode && NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
1427 nfs_set_verifier_delegated(&verf);
1428 dentry->d_time = verf;
1432 * nfs_set_verifier - save a parent directory verifier in the dentry
1433 * @dentry: pointer to dentry
1434 * @verf: verifier to save
1436 * Saves the parent directory verifier in @dentry. If the inode has
1437 * a delegation, we also tag the dentry as having been revalidated
1438 * while holding a delegation so that we know we don't have to
1439 * look it up again after a directory change.
1441 void nfs_set_verifier(struct dentry *dentry, unsigned long verf)
1444 spin_lock(&dentry->d_lock);
1445 nfs_set_verifier_locked(dentry, verf);
1446 spin_unlock(&dentry->d_lock);
1448 EXPORT_SYMBOL_GPL(nfs_set_verifier);
1450 #if IS_ENABLED(CONFIG_NFS_V4)
1452 * nfs_clear_verifier_delegated - clear the dir verifier delegation tag
1453 * @inode: pointer to inode
1455 * Iterates through the dentries in the inode alias list and clears
1456 * the tag used to indicate that the dentry has been revalidated
1457 * while holding a delegation.
1458 * This function is intended for use when the delegation is being
1459 * returned or revoked.
1461 void nfs_clear_verifier_delegated(struct inode *inode)
1463 struct dentry *alias;
1467 spin_lock(&inode->i_lock);
1468 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1469 spin_lock(&alias->d_lock);
1470 nfs_unset_verifier_delegated(&alias->d_time);
1471 spin_unlock(&alias->d_lock);
1473 spin_unlock(&inode->i_lock);
1475 EXPORT_SYMBOL_GPL(nfs_clear_verifier_delegated);
1476 #endif /* IS_ENABLED(CONFIG_NFS_V4) */
1478 static int nfs_dentry_verify_change(struct inode *dir, struct dentry *dentry)
1480 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE) &&
1481 d_really_is_negative(dentry))
1482 return dentry->d_time == inode_peek_iversion_raw(dir);
1483 return nfs_verify_change_attribute(dir, dentry->d_time);
1487 * A check for whether or not the parent directory has changed.
1488 * In the case it has, we assume that the dentries are untrustworthy
1489 * and may need to be looked up again.
1490 * If rcu_walk prevents us from performing a full check, return 0.
1492 static int nfs_check_verifier(struct inode *dir, struct dentry *dentry,
1495 if (IS_ROOT(dentry))
1497 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
1499 if (!nfs_dentry_verify_change(dir, dentry))
1501 /* Revalidate nfsi->cache_change_attribute before we declare a match */
1502 if (nfs_mapping_need_revalidate_inode(dir)) {
1505 if (__nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
1508 if (!nfs_dentry_verify_change(dir, dentry))
1514 * Use intent information to check whether or not we're going to do
1515 * an O_EXCL create using this path component.
1517 static int nfs_is_exclusive_create(struct inode *dir, unsigned int flags)
1519 if (NFS_PROTO(dir)->version == 2)
1521 return flags & LOOKUP_EXCL;
1525 * Inode and filehandle revalidation for lookups.
1527 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1528 * or if the intent information indicates that we're about to open this
1529 * particular file and the "nocto" mount flag is not set.
1533 int nfs_lookup_verify_inode(struct inode *inode, unsigned int flags)
1535 struct nfs_server *server = NFS_SERVER(inode);
1538 if (IS_AUTOMOUNT(inode))
1541 if (flags & LOOKUP_OPEN) {
1542 switch (inode->i_mode & S_IFMT) {
1544 /* A NFSv4 OPEN will revalidate later */
1545 if (server->caps & NFS_CAP_ATOMIC_OPEN)
1549 if (server->flags & NFS_MOUNT_NOCTO)
1551 /* NFS close-to-open cache consistency validation */
1556 /* VFS wants an on-the-wire revalidation */
1557 if (flags & LOOKUP_REVAL)
1560 if (inode->i_nlink > 0 ||
1561 (inode->i_nlink == 0 &&
1562 test_bit(NFS_INO_PRESERVE_UNLINKED, &NFS_I(inode)->flags)))
1567 if (flags & LOOKUP_RCU)
1569 ret = __nfs_revalidate_inode(server, inode);
1575 static void nfs_mark_dir_for_revalidate(struct inode *inode)
1577 spin_lock(&inode->i_lock);
1578 nfs_set_cache_invalid(inode, NFS_INO_INVALID_CHANGE);
1579 spin_unlock(&inode->i_lock);
1583 * We judge how long we want to trust negative
1584 * dentries by looking at the parent inode mtime.
1586 * If parent mtime has changed, we revalidate, else we wait for a
1587 * period corresponding to the parent's attribute cache timeout value.
1589 * If LOOKUP_RCU prevents us from performing a full check, return 1
1590 * suggesting a reval is needed.
1592 * Note that when creating a new file, or looking up a rename target,
1593 * then it shouldn't be necessary to revalidate a negative dentry.
1596 int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1599 if (flags & (LOOKUP_CREATE | LOOKUP_RENAME_TARGET))
1601 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1603 /* Case insensitive server? Revalidate negative dentries */
1604 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1606 return !nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU);
1610 nfs_lookup_revalidate_done(struct inode *dir, struct dentry *dentry,
1611 struct inode *inode, int error)
1618 * We can't d_drop the root of a disconnected tree:
1619 * its d_hash is on the s_anon list and d_drop() would hide
1620 * it from shrink_dcache_for_unmount(), leading to busy
1621 * inodes on unmount and further oopses.
1623 if (inode && IS_ROOT(dentry))
1627 trace_nfs_lookup_revalidate_exit(dir, dentry, 0, error);
1632 nfs_lookup_revalidate_negative(struct inode *dir, struct dentry *dentry,
1636 if (nfs_neg_need_reval(dir, dentry, flags)) {
1637 if (flags & LOOKUP_RCU)
1641 return nfs_lookup_revalidate_done(dir, dentry, NULL, ret);
1645 nfs_lookup_revalidate_delegated(struct inode *dir, struct dentry *dentry,
1646 struct inode *inode)
1648 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1649 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1652 static int nfs_lookup_revalidate_dentry(struct inode *dir,
1653 struct dentry *dentry,
1654 struct inode *inode, unsigned int flags)
1656 struct nfs_fh *fhandle;
1657 struct nfs_fattr *fattr;
1658 unsigned long dir_verifier;
1661 trace_nfs_lookup_revalidate_enter(dir, dentry, flags);
1664 fhandle = nfs_alloc_fhandle();
1665 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(inode));
1666 if (fhandle == NULL || fattr == NULL)
1669 dir_verifier = nfs_save_change_attribute(dir);
1670 ret = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1678 if (NFS_SERVER(inode)->flags & NFS_MOUNT_SOFTREVAL)
1684 /* Request help from readdirplus */
1685 nfs_lookup_advise_force_readdirplus(dir, flags);
1688 if (nfs_compare_fh(NFS_FH(inode), fhandle))
1690 if (nfs_refresh_inode(inode, fattr) < 0)
1693 nfs_setsecurity(inode, fattr);
1694 nfs_set_verifier(dentry, dir_verifier);
1698 nfs_free_fattr(fattr);
1699 nfs_free_fhandle(fhandle);
1702 * If the lookup failed despite the dentry change attribute being
1703 * a match, then we should revalidate the directory cache.
1705 if (!ret && nfs_dentry_verify_change(dir, dentry))
1706 nfs_mark_dir_for_revalidate(dir);
1707 return nfs_lookup_revalidate_done(dir, dentry, inode, ret);
1711 * This is called every time the dcache has a lookup hit,
1712 * and we should check whether we can really trust that
1715 * NOTE! The hit can be a negative hit too, don't assume
1718 * If the parent directory is seen to have changed, we throw out the
1719 * cached dentry and do a new lookup.
1722 nfs_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
1725 struct inode *inode;
1728 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
1729 inode = d_inode(dentry);
1732 return nfs_lookup_revalidate_negative(dir, dentry, flags);
1734 if (is_bad_inode(inode)) {
1735 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1740 if (nfs_verifier_is_delegated(dentry))
1741 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
1743 /* Force a full look up iff the parent directory has changed */
1744 if (!(flags & (LOOKUP_EXCL | LOOKUP_REVAL)) &&
1745 nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU)) {
1746 error = nfs_lookup_verify_inode(inode, flags);
1748 if (error == -ESTALE)
1749 nfs_mark_dir_for_revalidate(dir);
1755 if (flags & LOOKUP_RCU)
1758 if (NFS_STALE(inode))
1761 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
1763 return nfs_lookup_revalidate_done(dir, dentry, inode, 1);
1765 if (flags & LOOKUP_RCU)
1767 return nfs_lookup_revalidate_done(dir, dentry, inode, 0);
1771 __nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags,
1772 int (*reval)(struct inode *, struct dentry *, unsigned int))
1774 struct dentry *parent;
1778 if (flags & LOOKUP_RCU) {
1779 parent = READ_ONCE(dentry->d_parent);
1780 dir = d_inode_rcu(parent);
1783 ret = reval(dir, dentry, flags);
1784 if (parent != READ_ONCE(dentry->d_parent))
1787 parent = dget_parent(dentry);
1788 ret = reval(d_inode(parent), dentry, flags);
1794 static int nfs_lookup_revalidate(struct dentry *dentry, unsigned int flags)
1796 return __nfs_lookup_revalidate(dentry, flags, nfs_do_lookup_revalidate);
1800 * A weaker form of d_revalidate for revalidating just the d_inode(dentry)
1801 * when we don't really care about the dentry name. This is called when a
1802 * pathwalk ends on a dentry that was not found via a normal lookup in the
1803 * parent dir (e.g.: ".", "..", procfs symlinks or mountpoint traversals).
1805 * In this situation, we just want to verify that the inode itself is OK
1806 * since the dentry might have changed on the server.
1808 static int nfs_weak_revalidate(struct dentry *dentry, unsigned int flags)
1810 struct inode *inode = d_inode(dentry);
1814 * I believe we can only get a negative dentry here in the case of a
1815 * procfs-style symlink. Just assume it's correct for now, but we may
1816 * eventually need to do something more here.
1819 dfprintk(LOOKUPCACHE, "%s: %pd2 has negative inode\n",
1824 if (is_bad_inode(inode)) {
1825 dfprintk(LOOKUPCACHE, "%s: %pd2 has dud inode\n",
1830 error = nfs_lookup_verify_inode(inode, flags);
1831 dfprintk(LOOKUPCACHE, "NFS: %s: inode %lu is %s\n",
1832 __func__, inode->i_ino, error ? "invalid" : "valid");
1837 * This is called from dput() when d_count is going to 0.
1839 static int nfs_dentry_delete(const struct dentry *dentry)
1841 dfprintk(VFS, "NFS: dentry_delete(%pd2, %x)\n",
1842 dentry, dentry->d_flags);
1844 /* Unhash any dentry with a stale inode */
1845 if (d_really_is_positive(dentry) && NFS_STALE(d_inode(dentry)))
1848 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1849 /* Unhash it, so that ->d_iput() would be called */
1852 if (!(dentry->d_sb->s_flags & SB_ACTIVE)) {
1853 /* Unhash it, so that ancestors of killed async unlink
1854 * files will be cleaned up during umount */
1861 /* Ensure that we revalidate inode->i_nlink */
1862 static void nfs_drop_nlink(struct inode *inode)
1864 spin_lock(&inode->i_lock);
1865 /* drop the inode if we're reasonably sure this is the last link */
1866 if (inode->i_nlink > 0)
1868 NFS_I(inode)->attr_gencount = nfs_inc_attr_generation_counter();
1869 nfs_set_cache_invalid(
1870 inode, NFS_INO_INVALID_CHANGE | NFS_INO_INVALID_CTIME |
1871 NFS_INO_INVALID_NLINK);
1872 spin_unlock(&inode->i_lock);
1876 * Called when the dentry loses inode.
1877 * We use it to clean up silly-renamed files.
1879 static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1881 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1882 nfs_complete_unlink(dentry, inode);
1883 nfs_drop_nlink(inode);
1888 static void nfs_d_release(struct dentry *dentry)
1890 /* free cached devname value, if it survived that far */
1891 if (unlikely(dentry->d_fsdata)) {
1892 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1895 kfree(dentry->d_fsdata);
1899 const struct dentry_operations nfs_dentry_operations = {
1900 .d_revalidate = nfs_lookup_revalidate,
1901 .d_weak_revalidate = nfs_weak_revalidate,
1902 .d_delete = nfs_dentry_delete,
1903 .d_iput = nfs_dentry_iput,
1904 .d_automount = nfs_d_automount,
1905 .d_release = nfs_d_release,
1907 EXPORT_SYMBOL_GPL(nfs_dentry_operations);
1909 struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, unsigned int flags)
1912 struct inode *inode = NULL;
1913 struct nfs_fh *fhandle = NULL;
1914 struct nfs_fattr *fattr = NULL;
1915 unsigned long dir_verifier;
1918 dfprintk(VFS, "NFS: lookup(%pd2)\n", dentry);
1919 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
1921 if (unlikely(dentry->d_name.len > NFS_SERVER(dir)->namelen))
1922 return ERR_PTR(-ENAMETOOLONG);
1925 * If we're doing an exclusive create, optimize away the lookup
1926 * but don't hash the dentry.
1928 if (nfs_is_exclusive_create(dir, flags) || flags & LOOKUP_RENAME_TARGET)
1931 res = ERR_PTR(-ENOMEM);
1932 fhandle = nfs_alloc_fhandle();
1933 fattr = nfs_alloc_fattr_with_label(NFS_SERVER(dir));
1934 if (fhandle == NULL || fattr == NULL)
1937 dir_verifier = nfs_save_change_attribute(dir);
1938 trace_nfs_lookup_enter(dir, dentry, flags);
1939 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
1940 if (error == -ENOENT) {
1941 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
1942 dir_verifier = inode_peek_iversion_raw(dir);
1946 res = ERR_PTR(error);
1949 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
1950 res = ERR_CAST(inode);
1954 /* Notify readdir to use READDIRPLUS */
1955 nfs_lookup_advise_force_readdirplus(dir, flags);
1958 res = d_splice_alias(inode, dentry);
1964 nfs_set_verifier(dentry, dir_verifier);
1966 trace_nfs_lookup_exit(dir, dentry, flags, PTR_ERR_OR_ZERO(res));
1967 nfs_free_fattr(fattr);
1968 nfs_free_fhandle(fhandle);
1971 EXPORT_SYMBOL_GPL(nfs_lookup);
1973 void nfs_d_prune_case_insensitive_aliases(struct inode *inode)
1975 /* Case insensitive server? Revalidate dentries */
1976 if (inode && nfs_server_capable(inode, NFS_CAP_CASE_INSENSITIVE))
1977 d_prune_aliases(inode);
1979 EXPORT_SYMBOL_GPL(nfs_d_prune_case_insensitive_aliases);
1981 #if IS_ENABLED(CONFIG_NFS_V4)
1982 static int nfs4_lookup_revalidate(struct dentry *, unsigned int);
1984 const struct dentry_operations nfs4_dentry_operations = {
1985 .d_revalidate = nfs4_lookup_revalidate,
1986 .d_weak_revalidate = nfs_weak_revalidate,
1987 .d_delete = nfs_dentry_delete,
1988 .d_iput = nfs_dentry_iput,
1989 .d_automount = nfs_d_automount,
1990 .d_release = nfs_d_release,
1992 EXPORT_SYMBOL_GPL(nfs4_dentry_operations);
1994 static fmode_t flags_to_mode(int flags)
1996 fmode_t res = (__force fmode_t)flags & FMODE_EXEC;
1997 if ((flags & O_ACCMODE) != O_WRONLY)
1999 if ((flags & O_ACCMODE) != O_RDONLY)
2004 static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags, struct file *filp)
2006 return alloc_nfs_open_context(dentry, flags_to_mode(open_flags), filp);
2009 static int do_open(struct inode *inode, struct file *filp)
2011 nfs_fscache_open_file(inode, filp);
2015 static int nfs_finish_open(struct nfs_open_context *ctx,
2016 struct dentry *dentry,
2017 struct file *file, unsigned open_flags)
2021 err = finish_open(file, dentry, do_open);
2024 if (S_ISREG(file->f_path.dentry->d_inode->i_mode))
2025 nfs_file_set_open_context(file, ctx);
2032 int nfs_atomic_open(struct inode *dir, struct dentry *dentry,
2033 struct file *file, unsigned open_flags,
2036 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
2037 struct nfs_open_context *ctx;
2039 struct iattr attr = { .ia_valid = ATTR_OPEN };
2040 struct inode *inode;
2041 unsigned int lookup_flags = 0;
2042 unsigned long dir_verifier;
2043 bool switched = false;
2047 /* Expect a negative dentry */
2048 BUG_ON(d_inode(dentry));
2050 dfprintk(VFS, "NFS: atomic_open(%s/%lu), %pd\n",
2051 dir->i_sb->s_id, dir->i_ino, dentry);
2053 err = nfs_check_flags(open_flags);
2057 /* NFS only supports OPEN on regular files */
2058 if ((open_flags & O_DIRECTORY)) {
2059 if (!d_in_lookup(dentry)) {
2061 * Hashed negative dentry with O_DIRECTORY: dentry was
2062 * revalidated and is fine, no need to perform lookup
2067 lookup_flags = LOOKUP_OPEN|LOOKUP_DIRECTORY;
2071 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
2072 return -ENAMETOOLONG;
2074 if (open_flags & O_CREAT) {
2075 struct nfs_server *server = NFS_SERVER(dir);
2077 if (!(server->attr_bitmask[2] & FATTR4_WORD2_MODE_UMASK))
2078 mode &= ~current_umask();
2080 attr.ia_valid |= ATTR_MODE;
2081 attr.ia_mode = mode;
2083 if (open_flags & O_TRUNC) {
2084 attr.ia_valid |= ATTR_SIZE;
2088 if (!(open_flags & O_CREAT) && !d_in_lookup(dentry)) {
2091 dentry = d_alloc_parallel(dentry->d_parent,
2092 &dentry->d_name, &wq);
2094 return PTR_ERR(dentry);
2095 if (unlikely(!d_in_lookup(dentry)))
2096 return finish_no_open(file, dentry);
2099 ctx = create_nfs_open_context(dentry, open_flags, file);
2104 trace_nfs_atomic_open_enter(dir, ctx, open_flags);
2105 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr, &created);
2107 file->f_mode |= FMODE_CREATED;
2108 if (IS_ERR(inode)) {
2109 err = PTR_ERR(inode);
2110 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2111 put_nfs_open_context(ctx);
2115 d_splice_alias(NULL, dentry);
2116 if (nfs_server_capable(dir, NFS_CAP_CASE_INSENSITIVE))
2117 dir_verifier = inode_peek_iversion_raw(dir);
2119 dir_verifier = nfs_save_change_attribute(dir);
2120 nfs_set_verifier(dentry, dir_verifier);
2126 if (!(open_flags & O_NOFOLLOW))
2136 err = nfs_finish_open(ctx, ctx->dentry, file, open_flags);
2137 trace_nfs_atomic_open_exit(dir, ctx, open_flags, err);
2138 put_nfs_open_context(ctx);
2140 if (unlikely(switched)) {
2141 d_lookup_done(dentry);
2147 res = nfs_lookup(dir, dentry, lookup_flags);
2149 inode = d_inode(dentry);
2150 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2151 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode)))
2152 res = ERR_PTR(-ENOTDIR);
2153 else if (inode && S_ISREG(inode->i_mode))
2154 res = ERR_PTR(-EOPENSTALE);
2155 } else if (!IS_ERR(res)) {
2156 inode = d_inode(res);
2157 if ((lookup_flags & LOOKUP_DIRECTORY) && inode &&
2158 !(S_ISDIR(inode->i_mode) || S_ISLNK(inode->i_mode))) {
2160 res = ERR_PTR(-ENOTDIR);
2161 } else if (inode && S_ISREG(inode->i_mode)) {
2163 res = ERR_PTR(-EOPENSTALE);
2167 d_lookup_done(dentry);
2174 return PTR_ERR(res);
2175 return finish_no_open(file, res);
2177 EXPORT_SYMBOL_GPL(nfs_atomic_open);
2180 nfs4_do_lookup_revalidate(struct inode *dir, struct dentry *dentry,
2183 struct inode *inode;
2185 if (!(flags & LOOKUP_OPEN) || (flags & LOOKUP_DIRECTORY))
2187 if (d_mountpoint(dentry))
2190 inode = d_inode(dentry);
2192 /* We can't create new files in nfs_open_revalidate(), so we
2193 * optimize away revalidation of negative dentries.
2198 if (nfs_verifier_is_delegated(dentry))
2199 return nfs_lookup_revalidate_delegated(dir, dentry, inode);
2201 /* NFS only supports OPEN on regular files */
2202 if (!S_ISREG(inode->i_mode))
2205 /* We cannot do exclusive creation on a positive dentry */
2206 if (flags & (LOOKUP_EXCL | LOOKUP_REVAL))
2209 /* Check if the directory changed */
2210 if (!nfs_check_verifier(dir, dentry, flags & LOOKUP_RCU))
2213 /* Let f_op->open() actually open (and revalidate) the file */
2216 if (flags & LOOKUP_RCU)
2218 return nfs_lookup_revalidate_dentry(dir, dentry, inode, flags);
2221 return nfs_do_lookup_revalidate(dir, dentry, flags);
2224 static int nfs4_lookup_revalidate(struct dentry *dentry, unsigned int flags)
2226 return __nfs_lookup_revalidate(dentry, flags,
2227 nfs4_do_lookup_revalidate);
2230 #endif /* CONFIG_NFSV4 */
2233 nfs_add_or_obtain(struct dentry *dentry, struct nfs_fh *fhandle,
2234 struct nfs_fattr *fattr)
2236 struct dentry *parent = dget_parent(dentry);
2237 struct inode *dir = d_inode(parent);
2238 struct inode *inode;
2244 if (fhandle->size == 0) {
2245 error = NFS_PROTO(dir)->lookup(dir, dentry, fhandle, fattr);
2249 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2250 if (!(fattr->valid & NFS_ATTR_FATTR)) {
2251 struct nfs_server *server = NFS_SB(dentry->d_sb);
2252 error = server->nfs_client->rpc_ops->getattr(server, fhandle,
2257 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
2258 d = d_splice_alias(inode, dentry);
2266 EXPORT_SYMBOL_GPL(nfs_add_or_obtain);
2269 * Code common to create, mkdir, and mknod.
2271 int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
2272 struct nfs_fattr *fattr)
2276 d = nfs_add_or_obtain(dentry, fhandle, fattr);
2280 /* Callers don't care */
2284 EXPORT_SYMBOL_GPL(nfs_instantiate);
2287 * Following a failed create operation, we drop the dentry rather
2288 * than retain a negative dentry. This avoids a problem in the event
2289 * that the operation succeeded on the server, but an error in the
2290 * reply path made it appear to have failed.
2292 int nfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2293 struct dentry *dentry, umode_t mode, bool excl)
2296 int open_flags = excl ? O_CREAT | O_EXCL : O_CREAT;
2299 dfprintk(VFS, "NFS: create(%s/%lu), %pd\n",
2300 dir->i_sb->s_id, dir->i_ino, dentry);
2302 attr.ia_mode = mode;
2303 attr.ia_valid = ATTR_MODE;
2305 trace_nfs_create_enter(dir, dentry, open_flags);
2306 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags);
2307 trace_nfs_create_exit(dir, dentry, open_flags, error);
2315 EXPORT_SYMBOL_GPL(nfs_create);
2318 * See comments for nfs_proc_create regarding failed operations.
2321 nfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
2322 struct dentry *dentry, umode_t mode, dev_t rdev)
2327 dfprintk(VFS, "NFS: mknod(%s/%lu), %pd\n",
2328 dir->i_sb->s_id, dir->i_ino, dentry);
2330 attr.ia_mode = mode;
2331 attr.ia_valid = ATTR_MODE;
2333 trace_nfs_mknod_enter(dir, dentry);
2334 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
2335 trace_nfs_mknod_exit(dir, dentry, status);
2343 EXPORT_SYMBOL_GPL(nfs_mknod);
2346 * See comments for nfs_proc_create regarding failed operations.
2348 int nfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
2349 struct dentry *dentry, umode_t mode)
2354 dfprintk(VFS, "NFS: mkdir(%s/%lu), %pd\n",
2355 dir->i_sb->s_id, dir->i_ino, dentry);
2357 attr.ia_valid = ATTR_MODE;
2358 attr.ia_mode = mode | S_IFDIR;
2360 trace_nfs_mkdir_enter(dir, dentry);
2361 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
2362 trace_nfs_mkdir_exit(dir, dentry, error);
2370 EXPORT_SYMBOL_GPL(nfs_mkdir);
2372 static void nfs_dentry_handle_enoent(struct dentry *dentry)
2374 if (simple_positive(dentry))
2378 static void nfs_dentry_remove_handle_error(struct inode *dir,
2379 struct dentry *dentry, int error)
2384 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2387 nfs_d_prune_case_insensitive_aliases(d_inode(dentry));
2388 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2392 int nfs_rmdir(struct inode *dir, struct dentry *dentry)
2396 dfprintk(VFS, "NFS: rmdir(%s/%lu), %pd\n",
2397 dir->i_sb->s_id, dir->i_ino, dentry);
2399 trace_nfs_rmdir_enter(dir, dentry);
2400 if (d_really_is_positive(dentry)) {
2401 down_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2402 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2403 /* Ensure the VFS deletes this inode */
2406 clear_nlink(d_inode(dentry));
2409 nfs_dentry_handle_enoent(dentry);
2411 up_write(&NFS_I(d_inode(dentry))->rmdir_sem);
2413 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
2414 nfs_dentry_remove_handle_error(dir, dentry, error);
2415 trace_nfs_rmdir_exit(dir, dentry, error);
2419 EXPORT_SYMBOL_GPL(nfs_rmdir);
2422 * Remove a file after making sure there are no pending writes,
2423 * and after checking that the file has only one user.
2425 * We invalidate the attribute cache and free the inode prior to the operation
2426 * to avoid possible races if the server reuses the inode.
2428 static int nfs_safe_remove(struct dentry *dentry)
2430 struct inode *dir = d_inode(dentry->d_parent);
2431 struct inode *inode = d_inode(dentry);
2434 dfprintk(VFS, "NFS: safe_remove(%pd2)\n", dentry);
2436 /* If the dentry was sillyrenamed, we simply call d_delete() */
2437 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
2442 trace_nfs_remove_enter(dir, dentry);
2443 if (inode != NULL) {
2444 error = NFS_PROTO(dir)->remove(dir, dentry);
2446 nfs_drop_nlink(inode);
2448 error = NFS_PROTO(dir)->remove(dir, dentry);
2449 if (error == -ENOENT)
2450 nfs_dentry_handle_enoent(dentry);
2451 trace_nfs_remove_exit(dir, dentry, error);
2456 /* We do silly rename. In case sillyrename() returns -EBUSY, the inode
2457 * belongs to an active ".nfs..." file and we return -EBUSY.
2459 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
2461 int nfs_unlink(struct inode *dir, struct dentry *dentry)
2464 int need_rehash = 0;
2466 dfprintk(VFS, "NFS: unlink(%s/%lu, %pd)\n", dir->i_sb->s_id,
2467 dir->i_ino, dentry);
2469 trace_nfs_unlink_enter(dir, dentry);
2470 spin_lock(&dentry->d_lock);
2471 if (d_count(dentry) > 1 && !test_bit(NFS_INO_PRESERVE_UNLINKED,
2472 &NFS_I(d_inode(dentry))->flags)) {
2473 spin_unlock(&dentry->d_lock);
2474 /* Start asynchronous writeout of the inode */
2475 write_inode_now(d_inode(dentry), 0);
2476 error = nfs_sillyrename(dir, dentry);
2479 if (!d_unhashed(dentry)) {
2483 spin_unlock(&dentry->d_lock);
2484 error = nfs_safe_remove(dentry);
2485 nfs_dentry_remove_handle_error(dir, dentry, error);
2489 trace_nfs_unlink_exit(dir, dentry, error);
2492 EXPORT_SYMBOL_GPL(nfs_unlink);
2495 * To create a symbolic link, most file systems instantiate a new inode,
2496 * add a page to it containing the path, then write it out to the disk
2497 * using prepare_write/commit_write.
2499 * Unfortunately the NFS client can't create the in-core inode first
2500 * because it needs a file handle to create an in-core inode (see
2501 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
2502 * symlink request has completed on the server.
2504 * So instead we allocate a raw page, copy the symname into it, then do
2505 * the SYMLINK request with the page as the buffer. If it succeeds, we
2506 * now have a new file handle and can instantiate an in-core NFS inode
2507 * and move the raw page into its mapping.
2509 int nfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
2510 struct dentry *dentry, const char *symname)
2515 unsigned int pathlen = strlen(symname);
2518 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s)\n", dir->i_sb->s_id,
2519 dir->i_ino, dentry, symname);
2521 if (pathlen > PAGE_SIZE)
2522 return -ENAMETOOLONG;
2524 attr.ia_mode = S_IFLNK | S_IRWXUGO;
2525 attr.ia_valid = ATTR_MODE;
2527 page = alloc_page(GFP_USER);
2531 kaddr = page_address(page);
2532 memcpy(kaddr, symname, pathlen);
2533 if (pathlen < PAGE_SIZE)
2534 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
2536 trace_nfs_symlink_enter(dir, dentry);
2537 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
2538 trace_nfs_symlink_exit(dir, dentry, error);
2540 dfprintk(VFS, "NFS: symlink(%s/%lu, %pd, %s) error %d\n",
2541 dir->i_sb->s_id, dir->i_ino,
2542 dentry, symname, error);
2548 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2551 * No big deal if we can't add this page to the page cache here.
2552 * READLINK will get the missing page from the server if needed.
2554 if (!add_to_page_cache_lru(page, d_inode(dentry)->i_mapping, 0,
2556 SetPageUptodate(page);
2559 * add_to_page_cache_lru() grabs an extra page refcount.
2560 * Drop it here to avoid leaking this page later.
2568 EXPORT_SYMBOL_GPL(nfs_symlink);
2571 nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2573 struct inode *inode = d_inode(old_dentry);
2576 dfprintk(VFS, "NFS: link(%pd2 -> %pd2)\n",
2577 old_dentry, dentry);
2579 trace_nfs_link_enter(inode, dir, dentry);
2581 if (S_ISREG(inode->i_mode))
2582 nfs_sync_inode(inode);
2583 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
2585 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
2587 d_add(dentry, inode);
2589 trace_nfs_link_exit(inode, dir, dentry, error);
2592 EXPORT_SYMBOL_GPL(nfs_link);
2596 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
2597 * different file handle for the same inode after a rename (e.g. when
2598 * moving to a different directory). A fail-safe method to do so would
2599 * be to look up old_dir/old_name, create a link to new_dir/new_name and
2600 * rename the old file using the sillyrename stuff. This way, the original
2601 * file in old_dir will go away when the last process iput()s the inode.
2605 * It actually works quite well. One needs to have the possibility for
2606 * at least one ".nfs..." file in each directory the file ever gets
2607 * moved or linked to which happens automagically with the new
2608 * implementation that only depends on the dcache stuff instead of
2609 * using the inode layer
2611 * Unfortunately, things are a little more complicated than indicated
2612 * above. For a cross-directory move, we want to make sure we can get
2613 * rid of the old inode after the operation. This means there must be
2614 * no pending writes (if it's a file), and the use count must be 1.
2615 * If these conditions are met, we can drop the dentries before doing
2618 int nfs_rename(struct user_namespace *mnt_userns, struct inode *old_dir,
2619 struct dentry *old_dentry, struct inode *new_dir,
2620 struct dentry *new_dentry, unsigned int flags)
2622 struct inode *old_inode = d_inode(old_dentry);
2623 struct inode *new_inode = d_inode(new_dentry);
2624 struct dentry *dentry = NULL, *rehash = NULL;
2625 struct rpc_task *task;
2631 dfprintk(VFS, "NFS: rename(%pd2 -> %pd2, ct=%d)\n",
2632 old_dentry, new_dentry,
2633 d_count(new_dentry));
2635 trace_nfs_rename_enter(old_dir, old_dentry, new_dir, new_dentry);
2637 * For non-directories, check whether the target is busy and if so,
2638 * make a copy of the dentry and then do a silly-rename. If the
2639 * silly-rename succeeds, the copied dentry is hashed and becomes
2642 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
2644 * To prevent any new references to the target during the
2645 * rename, we unhash the dentry in advance.
2647 if (!d_unhashed(new_dentry)) {
2649 rehash = new_dentry;
2652 if (d_count(new_dentry) > 2) {
2655 /* copy the target dentry's name */
2656 dentry = d_alloc(new_dentry->d_parent,
2657 &new_dentry->d_name);
2661 /* silly-rename the existing target ... */
2662 err = nfs_sillyrename(new_dir, new_dentry);
2666 new_dentry = dentry;
2672 if (S_ISREG(old_inode->i_mode))
2673 nfs_sync_inode(old_inode);
2674 task = nfs_async_rename(old_dir, new_dir, old_dentry, new_dentry, NULL);
2676 error = PTR_ERR(task);
2680 error = rpc_wait_for_completion_task(task);
2682 ((struct nfs_renamedata *)task->tk_calldata)->cancelled = 1;
2683 /* Paired with the atomic_dec_and_test() barrier in rpc_do_put_task() */
2686 error = task->tk_status;
2688 /* Ensure the inode attributes are revalidated */
2690 spin_lock(&old_inode->i_lock);
2691 NFS_I(old_inode)->attr_gencount = nfs_inc_attr_generation_counter();
2692 nfs_set_cache_invalid(old_inode, NFS_INO_INVALID_CHANGE |
2693 NFS_INO_INVALID_CTIME |
2694 NFS_INO_REVAL_FORCED);
2695 spin_unlock(&old_inode->i_lock);
2700 trace_nfs_rename_exit(old_dir, old_dentry,
2701 new_dir, new_dentry, error);
2703 if (new_inode != NULL)
2704 nfs_drop_nlink(new_inode);
2706 * The d_move() should be here instead of in an async RPC completion
2707 * handler because we need the proper locks to move the dentry. If
2708 * we're interrupted by a signal, the async RPC completion handler
2709 * should mark the directories for revalidation.
2711 d_move(old_dentry, new_dentry);
2712 nfs_set_verifier(old_dentry,
2713 nfs_save_change_attribute(new_dir));
2714 } else if (error == -ENOENT)
2715 nfs_dentry_handle_enoent(old_dentry);
2717 /* new dentry created? */
2722 EXPORT_SYMBOL_GPL(nfs_rename);
2724 static DEFINE_SPINLOCK(nfs_access_lru_lock);
2725 static LIST_HEAD(nfs_access_lru_list);
2726 static atomic_long_t nfs_access_nr_entries;
2728 static unsigned long nfs_access_max_cachesize = 4*1024*1024;
2729 module_param(nfs_access_max_cachesize, ulong, 0644);
2730 MODULE_PARM_DESC(nfs_access_max_cachesize, "NFS access maximum total cache length");
2732 static void nfs_access_free_entry(struct nfs_access_entry *entry)
2734 put_group_info(entry->group_info);
2735 kfree_rcu(entry, rcu_head);
2736 smp_mb__before_atomic();
2737 atomic_long_dec(&nfs_access_nr_entries);
2738 smp_mb__after_atomic();
2741 static void nfs_access_free_list(struct list_head *head)
2743 struct nfs_access_entry *cache;
2745 while (!list_empty(head)) {
2746 cache = list_entry(head->next, struct nfs_access_entry, lru);
2747 list_del(&cache->lru);
2748 nfs_access_free_entry(cache);
2752 static unsigned long
2753 nfs_do_access_cache_scan(unsigned int nr_to_scan)
2756 struct nfs_inode *nfsi, *next;
2757 struct nfs_access_entry *cache;
2760 spin_lock(&nfs_access_lru_lock);
2761 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
2762 struct inode *inode;
2764 if (nr_to_scan-- == 0)
2766 inode = &nfsi->vfs_inode;
2767 spin_lock(&inode->i_lock);
2768 if (list_empty(&nfsi->access_cache_entry_lru))
2769 goto remove_lru_entry;
2770 cache = list_entry(nfsi->access_cache_entry_lru.next,
2771 struct nfs_access_entry, lru);
2772 list_move(&cache->lru, &head);
2773 rb_erase(&cache->rb_node, &nfsi->access_cache);
2775 if (!list_empty(&nfsi->access_cache_entry_lru))
2776 list_move_tail(&nfsi->access_cache_inode_lru,
2777 &nfs_access_lru_list);
2780 list_del_init(&nfsi->access_cache_inode_lru);
2781 smp_mb__before_atomic();
2782 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
2783 smp_mb__after_atomic();
2785 spin_unlock(&inode->i_lock);
2787 spin_unlock(&nfs_access_lru_lock);
2788 nfs_access_free_list(&head);
2793 nfs_access_cache_scan(struct shrinker *shrink, struct shrink_control *sc)
2795 int nr_to_scan = sc->nr_to_scan;
2796 gfp_t gfp_mask = sc->gfp_mask;
2798 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2800 return nfs_do_access_cache_scan(nr_to_scan);
2805 nfs_access_cache_count(struct shrinker *shrink, struct shrink_control *sc)
2807 return vfs_pressure_ratio(atomic_long_read(&nfs_access_nr_entries));
2811 nfs_access_cache_enforce_limit(void)
2813 long nr_entries = atomic_long_read(&nfs_access_nr_entries);
2815 unsigned int nr_to_scan;
2817 if (nr_entries < 0 || nr_entries <= nfs_access_max_cachesize)
2820 diff = nr_entries - nfs_access_max_cachesize;
2821 if (diff < nr_to_scan)
2823 nfs_do_access_cache_scan(nr_to_scan);
2826 static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
2828 struct rb_root *root_node = &nfsi->access_cache;
2830 struct nfs_access_entry *entry;
2832 /* Unhook entries from the cache */
2833 while ((n = rb_first(root_node)) != NULL) {
2834 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2835 rb_erase(n, root_node);
2836 list_move(&entry->lru, head);
2838 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
2841 void nfs_access_zap_cache(struct inode *inode)
2845 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2847 /* Remove from global LRU init */
2848 spin_lock(&nfs_access_lru_lock);
2849 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2850 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
2852 spin_lock(&inode->i_lock);
2853 __nfs_access_zap_cache(NFS_I(inode), &head);
2854 spin_unlock(&inode->i_lock);
2855 spin_unlock(&nfs_access_lru_lock);
2856 nfs_access_free_list(&head);
2858 EXPORT_SYMBOL_GPL(nfs_access_zap_cache);
2860 static int access_cmp(const struct cred *a, const struct nfs_access_entry *b)
2862 struct group_info *ga, *gb;
2865 if (uid_lt(a->fsuid, b->fsuid))
2867 if (uid_gt(a->fsuid, b->fsuid))
2870 if (gid_lt(a->fsgid, b->fsgid))
2872 if (gid_gt(a->fsgid, b->fsgid))
2883 if (ga->ngroups < gb->ngroups)
2885 if (ga->ngroups > gb->ngroups)
2888 for (g = 0; g < ga->ngroups; g++) {
2889 if (gid_lt(ga->gid[g], gb->gid[g]))
2891 if (gid_gt(ga->gid[g], gb->gid[g]))
2897 static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, const struct cred *cred)
2899 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2902 struct nfs_access_entry *entry =
2903 rb_entry(n, struct nfs_access_entry, rb_node);
2904 int cmp = access_cmp(cred, entry);
2916 static int nfs_access_get_cached_locked(struct inode *inode, const struct cred *cred, u32 *mask, bool may_block)
2918 struct nfs_inode *nfsi = NFS_I(inode);
2919 struct nfs_access_entry *cache;
2923 spin_lock(&inode->i_lock);
2925 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2927 cache = nfs_access_search_rbtree(inode, cred);
2931 /* Found an entry, is our attribute cache valid? */
2932 if (!nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2939 spin_unlock(&inode->i_lock);
2940 err = __nfs_revalidate_inode(NFS_SERVER(inode), inode);
2943 spin_lock(&inode->i_lock);
2946 *mask = cache->mask;
2947 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
2950 spin_unlock(&inode->i_lock);
2953 spin_unlock(&inode->i_lock);
2954 nfs_access_zap_cache(inode);
2958 static int nfs_access_get_cached_rcu(struct inode *inode, const struct cred *cred, u32 *mask)
2960 /* Only check the most recently returned cache entry,
2961 * but do it without locking.
2963 struct nfs_inode *nfsi = NFS_I(inode);
2964 struct nfs_access_entry *cache;
2966 struct list_head *lh;
2969 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2971 lh = rcu_dereference(list_tail_rcu(&nfsi->access_cache_entry_lru));
2972 cache = list_entry(lh, struct nfs_access_entry, lru);
2973 if (lh == &nfsi->access_cache_entry_lru ||
2974 access_cmp(cred, cache) != 0)
2978 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_ACCESS))
2980 *mask = cache->mask;
2987 int nfs_access_get_cached(struct inode *inode, const struct cred *cred,
2988 u32 *mask, bool may_block)
2992 status = nfs_access_get_cached_rcu(inode, cred, mask);
2994 status = nfs_access_get_cached_locked(inode, cred, mask,
2999 EXPORT_SYMBOL_GPL(nfs_access_get_cached);
3001 static void nfs_access_add_rbtree(struct inode *inode,
3002 struct nfs_access_entry *set,
3003 const struct cred *cred)
3005 struct nfs_inode *nfsi = NFS_I(inode);
3006 struct rb_root *root_node = &nfsi->access_cache;
3007 struct rb_node **p = &root_node->rb_node;
3008 struct rb_node *parent = NULL;
3009 struct nfs_access_entry *entry;
3012 spin_lock(&inode->i_lock);
3013 while (*p != NULL) {
3015 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
3016 cmp = access_cmp(cred, entry);
3019 p = &parent->rb_left;
3021 p = &parent->rb_right;
3025 rb_link_node(&set->rb_node, parent, p);
3026 rb_insert_color(&set->rb_node, root_node);
3027 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3028 spin_unlock(&inode->i_lock);
3031 rb_replace_node(parent, &set->rb_node, root_node);
3032 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
3033 list_del(&entry->lru);
3034 spin_unlock(&inode->i_lock);
3035 nfs_access_free_entry(entry);
3038 void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set,
3039 const struct cred *cred)
3041 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
3044 RB_CLEAR_NODE(&cache->rb_node);
3045 cache->fsuid = cred->fsuid;
3046 cache->fsgid = cred->fsgid;
3047 cache->group_info = get_group_info(cred->group_info);
3048 cache->mask = set->mask;
3050 /* The above field assignments must be visible
3051 * before this item appears on the lru. We cannot easily
3052 * use rcu_assign_pointer, so just force the memory barrier.
3055 nfs_access_add_rbtree(inode, cache, cred);
3057 /* Update accounting */
3058 smp_mb__before_atomic();
3059 atomic_long_inc(&nfs_access_nr_entries);
3060 smp_mb__after_atomic();
3062 /* Add inode to global LRU list */
3063 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
3064 spin_lock(&nfs_access_lru_lock);
3065 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
3066 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
3067 &nfs_access_lru_list);
3068 spin_unlock(&nfs_access_lru_lock);
3070 nfs_access_cache_enforce_limit();
3072 EXPORT_SYMBOL_GPL(nfs_access_add_cache);
3074 #define NFS_MAY_READ (NFS_ACCESS_READ)
3075 #define NFS_MAY_WRITE (NFS_ACCESS_MODIFY | \
3076 NFS_ACCESS_EXTEND | \
3078 #define NFS_FILE_MAY_WRITE (NFS_ACCESS_MODIFY | \
3080 #define NFS_DIR_MAY_WRITE NFS_MAY_WRITE
3081 #define NFS_MAY_LOOKUP (NFS_ACCESS_LOOKUP)
3082 #define NFS_MAY_EXECUTE (NFS_ACCESS_EXECUTE)
3084 nfs_access_calc_mask(u32 access_result, umode_t umode)
3088 if (access_result & NFS_MAY_READ)
3090 if (S_ISDIR(umode)) {
3091 if ((access_result & NFS_DIR_MAY_WRITE) == NFS_DIR_MAY_WRITE)
3093 if ((access_result & NFS_MAY_LOOKUP) == NFS_MAY_LOOKUP)
3095 } else if (S_ISREG(umode)) {
3096 if ((access_result & NFS_FILE_MAY_WRITE) == NFS_FILE_MAY_WRITE)
3098 if ((access_result & NFS_MAY_EXECUTE) == NFS_MAY_EXECUTE)
3100 } else if (access_result & NFS_MAY_WRITE)
3105 void nfs_access_set_mask(struct nfs_access_entry *entry, u32 access_result)
3107 entry->mask = access_result;
3109 EXPORT_SYMBOL_GPL(nfs_access_set_mask);
3111 static int nfs_do_access(struct inode *inode, const struct cred *cred, int mask)
3113 struct nfs_access_entry cache;
3114 bool may_block = (mask & MAY_NOT_BLOCK) == 0;
3115 int cache_mask = -1;
3118 trace_nfs_access_enter(inode);
3120 status = nfs_access_get_cached(inode, cred, &cache.mask, may_block);
3129 * Determine which access bits we want to ask for...
3131 cache.mask = NFS_ACCESS_READ | NFS_ACCESS_MODIFY | NFS_ACCESS_EXTEND |
3132 nfs_access_xattr_mask(NFS_SERVER(inode));
3133 if (S_ISDIR(inode->i_mode))
3134 cache.mask |= NFS_ACCESS_DELETE | NFS_ACCESS_LOOKUP;
3136 cache.mask |= NFS_ACCESS_EXECUTE;
3137 status = NFS_PROTO(inode)->access(inode, &cache, cred);
3139 if (status == -ESTALE) {
3140 if (!S_ISDIR(inode->i_mode))
3141 nfs_set_inode_stale(inode);
3143 nfs_zap_caches(inode);
3147 nfs_access_add_cache(inode, &cache, cred);
3149 cache_mask = nfs_access_calc_mask(cache.mask, inode->i_mode);
3150 if ((mask & ~cache_mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) != 0)
3153 trace_nfs_access_exit(inode, mask, cache_mask, status);
3157 static int nfs_open_permission_mask(int openflags)
3161 if (openflags & __FMODE_EXEC) {
3162 /* ONLY check exec rights */
3165 if ((openflags & O_ACCMODE) != O_WRONLY)
3167 if ((openflags & O_ACCMODE) != O_RDONLY)
3174 int nfs_may_open(struct inode *inode, const struct cred *cred, int openflags)
3176 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
3178 EXPORT_SYMBOL_GPL(nfs_may_open);
3180 static int nfs_execute_ok(struct inode *inode, int mask)
3182 struct nfs_server *server = NFS_SERVER(inode);
3185 if (S_ISDIR(inode->i_mode))
3187 if (nfs_check_cache_invalid(inode, NFS_INO_INVALID_MODE)) {
3188 if (mask & MAY_NOT_BLOCK)
3190 ret = __nfs_revalidate_inode(server, inode);
3192 if (ret == 0 && !execute_ok(inode))
3197 int nfs_permission(struct user_namespace *mnt_userns,
3198 struct inode *inode,
3201 const struct cred *cred = current_cred();
3204 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
3206 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
3208 /* Is this sys_access() ? */
3209 if (mask & (MAY_ACCESS | MAY_CHDIR))
3212 switch (inode->i_mode & S_IFMT) {
3216 if ((mask & MAY_OPEN) &&
3217 nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN))
3222 * Optimize away all write operations, since the server
3223 * will check permissions when we perform the op.
3225 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
3230 if (!NFS_PROTO(inode)->access)
3233 res = nfs_do_access(inode, cred, mask);
3235 if (!res && (mask & MAY_EXEC))
3236 res = nfs_execute_ok(inode, mask);
3238 dfprintk(VFS, "NFS: permission(%s/%lu), mask=0x%x, res=%d\n",
3239 inode->i_sb->s_id, inode->i_ino, mask, res);
3242 if (mask & MAY_NOT_BLOCK)
3245 res = nfs_revalidate_inode(inode, NFS_INO_INVALID_MODE |
3246 NFS_INO_INVALID_OTHER);
3248 res = generic_permission(&init_user_ns, inode, mask);
3251 EXPORT_SYMBOL_GPL(nfs_permission);