blob: 1f4625749038ca0074978bead538c21e8c7acd21 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/nfs/dir.c
3 *
4 * Copyright (C) 1992 Rick Sladkey
5 *
6 * nfs directory handling functions
7 *
8 * 10 Apr 1996 Added silly rename for unlink --okir
9 * 28 Sep 1996 Improved directory cache --okir
10 * 23 Aug 1997 Claus Heine claus@momo.math.rwth-aachen.de
11 * Re-implemented silly rename for unlink, newly implemented
12 * silly rename for nfs_rename() following the suggestions
13 * of Olaf Kirch (okir) found in this file.
14 * Following Linus comments on my original hack, this version
15 * depends only on the dcache stuff and doesn't touch the inode
16 * layer (iput() and friends).
17 * 6 Jun 1999 Cache readdir lookups in the page cache. -DaveM
18 */
19
20#include <linux/time.h>
21#include <linux/errno.h>
22#include <linux/stat.h>
23#include <linux/fcntl.h>
24#include <linux/string.h>
25#include <linux/kernel.h>
26#include <linux/slab.h>
27#include <linux/mm.h>
28#include <linux/sunrpc/clnt.h>
29#include <linux/nfs_fs.h>
30#include <linux/nfs_mount.h>
31#include <linux/pagemap.h>
Chuck Lever873101b2006-08-22 20:06:23 -040032#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/namei.h>
David Howells54ceac42006-08-22 20:06:13 -040034#include <linux/mount.h>
Alexey Dobriyane8edc6e2007-05-21 01:22:52 +040035#include <linux/sched.h>
Catalin Marinas04e4bd12010-11-11 12:53:47 +000036#include <linux/kmemleak.h>
Aneesh Kumar K.V64c2ce82010-12-09 11:35:25 +000037#include <linux/xattr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39#include "delegation.h"
Chuck Lever91d5b472006-03-20 13:44:14 -050040#include "iostat.h"
Adrian Bunk4c30d562007-11-21 15:04:31 -080041#include "internal.h"
Trond Myklebustcd9a1c02010-09-17 10:56:50 -040042#include "fscache.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070043
Linus Torvalds1da177e2005-04-16 15:20:36 -070044/* #define NFS_DEBUG_VERBOSE 1 */
45
46static int nfs_opendir(struct inode *, struct file *);
Bryan Schumaker480c2002011-03-23 14:48:29 -040047static int nfs_closedir(struct inode *, struct file *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070048static int nfs_readdir(struct file *, void *, filldir_t);
49static struct dentry *nfs_lookup(struct inode *, struct dentry *, struct nameidata *);
50static int nfs_create(struct inode *, struct dentry *, int, struct nameidata *);
51static int nfs_mkdir(struct inode *, struct dentry *, int);
52static int nfs_rmdir(struct inode *, struct dentry *);
53static int nfs_unlink(struct inode *, struct dentry *);
54static int nfs_symlink(struct inode *, struct dentry *, const char *);
55static int nfs_link(struct dentry *, struct inode *, struct dentry *);
56static int nfs_mknod(struct inode *, struct dentry *, int, dev_t);
57static int nfs_rename(struct inode *, struct dentry *,
58 struct inode *, struct dentry *);
Christoph Hellwig7ea80852010-05-26 17:53:25 +020059static int nfs_fsync_dir(struct file *, int);
Trond Myklebustf0dd2132005-06-22 17:16:29 +000060static loff_t nfs_llseek_dir(struct file *, loff_t, int);
Trond Myklebust11de3b12010-12-01 14:17:06 -050061static void nfs_readdir_clear_array(struct page*);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -080063const struct file_operations nfs_dir_operations = {
Trond Myklebustf0dd2132005-06-22 17:16:29 +000064 .llseek = nfs_llseek_dir,
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 .read = generic_read_dir,
66 .readdir = nfs_readdir,
67 .open = nfs_opendir,
Bryan Schumaker480c2002011-03-23 14:48:29 -040068 .release = nfs_closedir,
Linus Torvalds1da177e2005-04-16 15:20:36 -070069 .fsync = nfs_fsync_dir,
70};
71
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080072const struct inode_operations nfs_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -070073 .create = nfs_create,
74 .lookup = nfs_lookup,
75 .link = nfs_link,
76 .unlink = nfs_unlink,
77 .symlink = nfs_symlink,
78 .mkdir = nfs_mkdir,
79 .rmdir = nfs_rmdir,
80 .mknod = nfs_mknod,
81 .rename = nfs_rename,
82 .permission = nfs_permission,
83 .getattr = nfs_getattr,
84 .setattr = nfs_setattr,
85};
86
Trond Myklebust11de3b12010-12-01 14:17:06 -050087const struct address_space_operations nfs_dir_aops = {
88 .freepage = nfs_readdir_clear_array,
Bryan Schumakerd1bacf92010-09-24 14:48:42 -040089};
90
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000091#ifdef CONFIG_NFS_V3
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -080092const struct inode_operations nfs3_dir_inode_operations = {
Andreas Gruenbacherb7fa0552005-06-22 17:16:27 +000093 .create = nfs_create,
94 .lookup = nfs_lookup,
95 .link = nfs_link,
96 .unlink = nfs_unlink,
97 .symlink = nfs_symlink,
98 .mkdir = nfs_mkdir,
99 .rmdir = nfs_rmdir,
100 .mknod = nfs_mknod,
101 .rename = nfs_rename,
102 .permission = nfs_permission,
103 .getattr = nfs_getattr,
104 .setattr = nfs_setattr,
105 .listxattr = nfs3_listxattr,
106 .getxattr = nfs3_getxattr,
107 .setxattr = nfs3_setxattr,
108 .removexattr = nfs3_removexattr,
109};
110#endif /* CONFIG_NFS_V3 */
111
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112#ifdef CONFIG_NFS_V4
113
114static struct dentry *nfs_atomic_lookup(struct inode *, struct dentry *, struct nameidata *);
Trond Myklebustc0204fd2010-09-17 10:56:51 -0400115static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode, struct nameidata *nd);
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800116const struct inode_operations nfs4_dir_inode_operations = {
Trond Myklebustc0204fd2010-09-17 10:56:51 -0400117 .create = nfs_open_create,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 .lookup = nfs_atomic_lookup,
119 .link = nfs_link,
120 .unlink = nfs_unlink,
121 .symlink = nfs_symlink,
122 .mkdir = nfs_mkdir,
123 .rmdir = nfs_rmdir,
124 .mknod = nfs_mknod,
125 .rename = nfs_rename,
126 .permission = nfs_permission,
127 .getattr = nfs_getattr,
128 .setattr = nfs_setattr,
Aneesh Kumar K.V64c2ce82010-12-09 11:35:25 +0000129 .getxattr = generic_getxattr,
130 .setxattr = generic_setxattr,
131 .listxattr = generic_listxattr,
132 .removexattr = generic_removexattr,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133};
134
135#endif /* CONFIG_NFS_V4 */
136
Bryan Schumaker480c2002011-03-23 14:48:29 -0400137static struct nfs_open_dir_context *alloc_nfs_open_dir_context(struct rpc_cred *cred)
138{
139 struct nfs_open_dir_context *ctx;
140 ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
141 if (ctx != NULL) {
Bryan Schumaker8ef2ce32011-03-23 15:04:31 -0400142 ctx->duped = 0;
Bryan Schumaker480c2002011-03-23 14:48:29 -0400143 ctx->dir_cookie = 0;
Bryan Schumaker8ef2ce32011-03-23 15:04:31 -0400144 ctx->dup_cookie = 0;
Bryan Schumaker480c2002011-03-23 14:48:29 -0400145 ctx->cred = get_rpccred(cred);
146 } else
147 ctx = ERR_PTR(-ENOMEM);
148 return ctx;
149}
150
151static void put_nfs_open_dir_context(struct nfs_open_dir_context *ctx)
152{
153 put_rpccred(ctx->cred);
154 kfree(ctx);
155}
156
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157/*
158 * Open file
159 */
160static int
161nfs_opendir(struct inode *inode, struct file *filp)
162{
Bryan Schumaker480c2002011-03-23 14:48:29 -0400163 int res = 0;
164 struct nfs_open_dir_context *ctx;
165 struct rpc_cred *cred;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166
Chuck Lever6da24bc2008-06-11 17:55:58 -0400167 dfprintk(FILE, "NFS: open dir(%s/%s)\n",
Chuck Levercc0dd2d2008-06-11 17:55:42 -0400168 filp->f_path.dentry->d_parent->d_name.name,
169 filp->f_path.dentry->d_name.name);
170
171 nfs_inc_stats(inode, NFSIOS_VFSOPEN);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500172
Bryan Schumaker480c2002011-03-23 14:48:29 -0400173 cred = rpc_lookup_cred();
174 if (IS_ERR(cred))
175 return PTR_ERR(cred);
176 ctx = alloc_nfs_open_dir_context(cred);
177 if (IS_ERR(ctx)) {
178 res = PTR_ERR(ctx);
179 goto out;
180 }
181 filp->private_data = ctx;
Neil Brownf5a73672010-08-10 10:20:05 -0400182 if (filp->f_path.dentry == filp->f_path.mnt->mnt_root) {
183 /* This is a mountpoint, so d_revalidate will never
184 * have been called, so we need to refresh the
185 * inode (for close-open consistency) ourselves.
186 */
187 __nfs_revalidate_inode(NFS_SERVER(inode), inode);
188 }
Bryan Schumaker480c2002011-03-23 14:48:29 -0400189out:
190 put_rpccred(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 return res;
192}
193
Bryan Schumaker480c2002011-03-23 14:48:29 -0400194static int
195nfs_closedir(struct inode *inode, struct file *filp)
196{
197 put_nfs_open_dir_context(filp->private_data);
198 return 0;
199}
200
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400201struct nfs_cache_array_entry {
202 u64 cookie;
203 u64 ino;
204 struct qstr string;
Trond Myklebust0b26a0b2010-11-20 14:26:44 -0500205 unsigned char d_type;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400206};
207
208struct nfs_cache_array {
209 unsigned int size;
210 int eof_index;
211 u64 last_cookie;
212 struct nfs_cache_array_entry array[0];
213};
214
Chuck Lever573c4e12010-12-14 14:58:11 +0000215typedef int (*decode_dirent_t)(struct xdr_stream *, struct nfs_entry *, int);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216typedef struct {
217 struct file *file;
218 struct page *page;
219 unsigned long page_index;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000220 u64 *dir_cookie;
Trond Myklebust0aded702010-11-30 21:56:32 -0500221 u64 last_cookie;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000222 loff_t current_index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 decode_dirent_t decode;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400224
Neil Brown1f4eab72007-04-16 09:35:27 +1000225 unsigned long timestamp;
Trond Myklebust4704f0e2008-10-14 19:16:07 -0400226 unsigned long gencount;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400227 unsigned int cache_entry_index;
228 unsigned int plus:1;
229 unsigned int eof:1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230} nfs_readdir_descriptor_t;
231
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400232/*
233 * The caller is responsible for calling nfs_readdir_release_array(page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 */
235static
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400236struct nfs_cache_array *nfs_readdir_get_array(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237{
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500238 void *ptr;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400239 if (page == NULL)
240 return ERR_PTR(-EIO);
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500241 ptr = kmap(page);
242 if (ptr == NULL)
243 return ERR_PTR(-ENOMEM);
244 return ptr;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400245}
246
247static
248void nfs_readdir_release_array(struct page *page)
249{
250 kunmap(page);
251}
252
253/*
254 * we are freeing strings created by nfs_add_to_readdir_array()
255 */
256static
Trond Myklebust11de3b12010-12-01 14:17:06 -0500257void nfs_readdir_clear_array(struct page *page)
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400258{
Trond Myklebust11de3b12010-12-01 14:17:06 -0500259 struct nfs_cache_array *array;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400260 int i;
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500261
Trond Myklebust11de3b12010-12-01 14:17:06 -0500262 array = kmap_atomic(page, KM_USER0);
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400263 for (i = 0; i < array->size; i++)
264 kfree(array->array[i].string.name);
Trond Myklebust11de3b12010-12-01 14:17:06 -0500265 kunmap_atomic(array, KM_USER0);
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400266}
267
268/*
269 * the caller is responsible for freeing qstr.name
270 * when called by nfs_readdir_add_to_array, the strings will be freed in
271 * nfs_clear_readdir_array()
272 */
273static
Trond Myklebust4a201d62010-10-23 14:53:23 -0400274int nfs_readdir_make_qstr(struct qstr *string, const char *name, unsigned int len)
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400275{
276 string->len = len;
277 string->name = kmemdup(name, len, GFP_KERNEL);
Trond Myklebust4a201d62010-10-23 14:53:23 -0400278 if (string->name == NULL)
279 return -ENOMEM;
Catalin Marinas04e4bd12010-11-11 12:53:47 +0000280 /*
281 * Avoid a kmemleak false positive. The pointer to the name is stored
282 * in a page cache page which kmemleak does not scan.
283 */
284 kmemleak_not_leak(string->name);
Trond Myklebust4a201d62010-10-23 14:53:23 -0400285 string->hash = full_name_hash(name, len);
286 return 0;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400287}
288
289static
290int nfs_readdir_add_to_array(struct nfs_entry *entry, struct page *page)
291{
292 struct nfs_cache_array *array = nfs_readdir_get_array(page);
Trond Myklebust4a201d62010-10-23 14:53:23 -0400293 struct nfs_cache_array_entry *cache_entry;
294 int ret;
295
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400296 if (IS_ERR(array))
297 return PTR_ERR(array);
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400298
Trond Myklebust4a201d62010-10-23 14:53:23 -0400299 cache_entry = &array->array[array->size];
Trond Myklebust30200932010-11-20 15:18:22 -0500300
301 /* Check that this entry lies within the page bounds */
302 ret = -ENOSPC;
303 if ((char *)&cache_entry[1] - (char *)page_address(page) > PAGE_SIZE)
304 goto out;
305
Trond Myklebust4a201d62010-10-23 14:53:23 -0400306 cache_entry->cookie = entry->prev_cookie;
307 cache_entry->ino = entry->ino;
Trond Myklebust0b26a0b2010-11-20 14:26:44 -0500308 cache_entry->d_type = entry->d_type;
Trond Myklebust4a201d62010-10-23 14:53:23 -0400309 ret = nfs_readdir_make_qstr(&cache_entry->string, entry->name, entry->len);
310 if (ret)
311 goto out;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400312 array->last_cookie = entry->cookie;
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500313 array->size++;
Trond Myklebust47c716c2010-12-07 12:44:56 -0500314 if (entry->eof != 0)
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400315 array->eof_index = array->size;
Trond Myklebust4a201d62010-10-23 14:53:23 -0400316out:
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400317 nfs_readdir_release_array(page);
Trond Myklebust4a201d62010-10-23 14:53:23 -0400318 return ret;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400319}
320
321static
322int nfs_readdir_search_for_pos(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
323{
324 loff_t diff = desc->file->f_pos - desc->current_index;
325 unsigned int index;
Bryan Schumaker8ef2ce32011-03-23 15:04:31 -0400326 struct nfs_open_dir_context *ctx = desc->file->private_data;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400327
328 if (diff < 0)
329 goto out_eof;
330 if (diff >= array->size) {
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500331 if (array->eof_index >= 0)
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400332 goto out_eof;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400333 return -EAGAIN;
334 }
335
336 index = (unsigned int)diff;
337 *desc->dir_cookie = array->array[index].cookie;
338 desc->cache_entry_index = index;
Bryan Schumaker8ef2ce32011-03-23 15:04:31 -0400339 ctx->duped = 0;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400340 return 0;
341out_eof:
342 desc->eof = 1;
343 return -EBADCOOKIE;
344}
345
346static
347int nfs_readdir_search_for_cookie(struct nfs_cache_array *array, nfs_readdir_descriptor_t *desc)
348{
349 int i;
Bryan Schumaker8ef2ce32011-03-23 15:04:31 -0400350 loff_t new_pos;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400351 int status = -EAGAIN;
Bryan Schumaker8ef2ce32011-03-23 15:04:31 -0400352 struct nfs_open_dir_context *ctx = desc->file->private_data;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400353
354 for (i = 0; i < array->size; i++) {
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400355 if (array->array[i].cookie == *desc->dir_cookie) {
Bryan Schumaker8ef2ce32011-03-23 15:04:31 -0400356 new_pos = desc->current_index + i;
357 if (new_pos < desc->file->f_pos) {
358 ctx->dup_cookie = *desc->dir_cookie;
359 ctx->duped = 1;
360 }
361 desc->file->f_pos = new_pos;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400362 desc->cache_entry_index = i;
Trond Myklebust47c716c2010-12-07 12:44:56 -0500363 return 0;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400364 }
365 }
Trond Myklebust47c716c2010-12-07 12:44:56 -0500366 if (array->eof_index >= 0) {
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500367 status = -EBADCOOKIE;
Trond Myklebust18fb5fe2010-12-07 12:41:58 -0500368 if (*desc->dir_cookie == array->last_cookie)
369 desc->eof = 1;
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500370 }
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400371 return status;
372}
373
374static
375int nfs_readdir_search_array(nfs_readdir_descriptor_t *desc)
376{
377 struct nfs_cache_array *array;
Trond Myklebust47c716c2010-12-07 12:44:56 -0500378 int status;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400379
380 array = nfs_readdir_get_array(desc->page);
381 if (IS_ERR(array)) {
382 status = PTR_ERR(array);
383 goto out;
384 }
385
386 if (*desc->dir_cookie == 0)
387 status = nfs_readdir_search_for_pos(array, desc);
388 else
389 status = nfs_readdir_search_for_cookie(array, desc);
390
Trond Myklebust47c716c2010-12-07 12:44:56 -0500391 if (status == -EAGAIN) {
Trond Myklebust0aded702010-11-30 21:56:32 -0500392 desc->last_cookie = array->last_cookie;
Trond Myklebuste47c0852011-03-23 08:43:09 -0400393 desc->current_index += array->size;
Trond Myklebust47c716c2010-12-07 12:44:56 -0500394 desc->page_index++;
395 }
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400396 nfs_readdir_release_array(desc->page);
397out:
398 return status;
399}
400
401/* Fill a page with xdr information before transferring to the cache page */
402static
Bryan Schumaker56e4ebf2010-10-20 15:44:37 -0400403int nfs_readdir_xdr_filler(struct page **pages, nfs_readdir_descriptor_t *desc,
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400404 struct nfs_entry *entry, struct file *file, struct inode *inode)
405{
Bryan Schumaker480c2002011-03-23 14:48:29 -0400406 struct nfs_open_dir_context *ctx = file->private_data;
407 struct rpc_cred *cred = ctx->cred;
Trond Myklebust4704f0e2008-10-14 19:16:07 -0400408 unsigned long timestamp, gencount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 int error;
410
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 again:
412 timestamp = jiffies;
Trond Myklebust4704f0e2008-10-14 19:16:07 -0400413 gencount = nfs_inc_attr_generation_counter();
Bryan Schumaker56e4ebf2010-10-20 15:44:37 -0400414 error = NFS_PROTO(inode)->readdir(file->f_path.dentry, cred, entry->cookie, pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 NFS_SERVER(inode)->dtsize, desc->plus);
416 if (error < 0) {
417 /* We requested READDIRPLUS, but the server doesn't grok it */
418 if (error == -ENOTSUPP && desc->plus) {
419 NFS_SERVER(inode)->caps &= ~NFS_CAP_READDIRPLUS;
Benny Halevy3a10c302008-01-23 08:58:59 +0200420 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 desc->plus = 0;
422 goto again;
423 }
424 goto error;
425 }
Neil Brown1f4eab72007-04-16 09:35:27 +1000426 desc->timestamp = timestamp;
Trond Myklebust4704f0e2008-10-14 19:16:07 -0400427 desc->gencount = gencount;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400428error:
429 return error;
430}
431
Chuck Lever573c4e12010-12-14 14:58:11 +0000432static int xdr_decode(nfs_readdir_descriptor_t *desc,
433 struct nfs_entry *entry, struct xdr_stream *xdr)
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400434{
Chuck Lever573c4e12010-12-14 14:58:11 +0000435 int error;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400436
Chuck Lever573c4e12010-12-14 14:58:11 +0000437 error = desc->decode(xdr, entry, desc->plus);
438 if (error)
439 return error;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400440 entry->fattr->time_start = desc->timestamp;
441 entry->fattr->gencount = desc->gencount;
442 return 0;
443}
444
Bryan Schumakerd39ab9d2010-09-24 18:50:01 -0400445static
446int nfs_same_file(struct dentry *dentry, struct nfs_entry *entry)
447{
Bryan Schumakerd39ab9d2010-09-24 18:50:01 -0400448 if (dentry->d_inode == NULL)
449 goto different;
Trond Myklebust37a09f02010-11-30 12:42:34 -0500450 if (nfs_compare_fh(entry->fh, NFS_FH(dentry->d_inode)) != 0)
Bryan Schumakerd39ab9d2010-09-24 18:50:01 -0400451 goto different;
452 return 1;
453different:
454 return 0;
455}
456
457static
458void nfs_prime_dcache(struct dentry *parent, struct nfs_entry *entry)
459{
Trond Myklebust4a201d62010-10-23 14:53:23 -0400460 struct qstr filename = {
461 .len = entry->len,
462 .name = entry->name,
463 };
464 struct dentry *dentry;
465 struct dentry *alias;
Bryan Schumakerd39ab9d2010-09-24 18:50:01 -0400466 struct inode *dir = parent->d_inode;
467 struct inode *inode;
468
Trond Myklebust4a201d62010-10-23 14:53:23 -0400469 if (filename.name[0] == '.') {
470 if (filename.len == 1)
471 return;
472 if (filename.len == 2 && filename.name[1] == '.')
473 return;
474 }
475 filename.hash = full_name_hash(filename.name, filename.len);
Bryan Schumakerd39ab9d2010-09-24 18:50:01 -0400476
Trond Myklebust4a201d62010-10-23 14:53:23 -0400477 dentry = d_lookup(parent, &filename);
Bryan Schumakerd39ab9d2010-09-24 18:50:01 -0400478 if (dentry != NULL) {
479 if (nfs_same_file(dentry, entry)) {
480 nfs_refresh_inode(dentry->d_inode, entry->fattr);
481 goto out;
482 } else {
483 d_drop(dentry);
484 dput(dentry);
485 }
486 }
487
488 dentry = d_alloc(parent, &filename);
Trond Myklebust4a201d62010-10-23 14:53:23 -0400489 if (dentry == NULL)
490 return;
491
Bryan Schumakerd39ab9d2010-09-24 18:50:01 -0400492 inode = nfs_fhget(dentry->d_sb, entry->fh, entry->fattr);
493 if (IS_ERR(inode))
494 goto out;
495
496 alias = d_materialise_unique(dentry, inode);
497 if (IS_ERR(alias))
498 goto out;
499 else if (alias) {
500 nfs_set_verifier(alias, nfs_save_change_attribute(dir));
501 dput(alias);
502 } else
503 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
504
505out:
506 dput(dentry);
Bryan Schumakerd39ab9d2010-09-24 18:50:01 -0400507}
508
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400509/* Perform conversion from xdr to cache array */
510static
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500511int nfs_readdir_page_filler(nfs_readdir_descriptor_t *desc, struct nfs_entry *entry,
Trond Myklebust66502392011-01-08 17:45:38 -0500512 struct page **xdr_pages, struct page *page, unsigned int buflen)
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400513{
Bryan Schumakerbabddc72010-10-20 15:44:29 -0400514 struct xdr_stream stream;
Benny Halevyf7da7a12011-05-19 14:16:47 -0400515 struct xdr_buf buf;
Trond Myklebust66502392011-01-08 17:45:38 -0500516 struct page *scratch;
Bryan Schumaker99424382010-10-21 16:33:16 -0400517 struct nfs_cache_array *array;
Trond Myklebust5c346852010-11-20 12:43:45 -0500518 unsigned int count = 0;
519 int status;
Bryan Schumakerbabddc72010-10-20 15:44:29 -0400520
Trond Myklebust66502392011-01-08 17:45:38 -0500521 scratch = alloc_page(GFP_KERNEL);
522 if (scratch == NULL)
523 return -ENOMEM;
Bryan Schumakerbabddc72010-10-20 15:44:29 -0400524
Benny Halevyf7da7a12011-05-19 14:16:47 -0400525 xdr_init_decode_pages(&stream, &buf, xdr_pages, buflen);
Trond Myklebust66502392011-01-08 17:45:38 -0500526 xdr_set_scratch_buffer(&stream, page_address(scratch), PAGE_SIZE);
Bryan Schumaker99424382010-10-21 16:33:16 -0400527
528 do {
529 status = xdr_decode(desc, entry, &stream);
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500530 if (status != 0) {
531 if (status == -EAGAIN)
532 status = 0;
Bryan Schumaker99424382010-10-21 16:33:16 -0400533 break;
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500534 }
Bryan Schumaker99424382010-10-21 16:33:16 -0400535
Trond Myklebust5c346852010-11-20 12:43:45 -0500536 count++;
537
Trond Myklebust47c716c2010-12-07 12:44:56 -0500538 if (desc->plus != 0)
Bryan Schumakerd39ab9d2010-09-24 18:50:01 -0400539 nfs_prime_dcache(desc->file->f_path.dentry, entry);
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500540
541 status = nfs_readdir_add_to_array(entry, page);
542 if (status != 0)
543 break;
Bryan Schumaker99424382010-10-21 16:33:16 -0400544 } while (!entry->eof);
545
Trond Myklebust47c716c2010-12-07 12:44:56 -0500546 if (count == 0 || (status == -EBADCOOKIE && entry->eof != 0)) {
Bryan Schumaker99424382010-10-21 16:33:16 -0400547 array = nfs_readdir_get_array(page);
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500548 if (!IS_ERR(array)) {
549 array->eof_index = array->size;
550 status = 0;
551 nfs_readdir_release_array(page);
Trond Myklebust5c346852010-11-20 12:43:45 -0500552 } else
553 status = PTR_ERR(array);
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400554 }
Trond Myklebust66502392011-01-08 17:45:38 -0500555
556 put_page(scratch);
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500557 return status;
Bryan Schumaker56e4ebf2010-10-20 15:44:37 -0400558}
559
560static
561void nfs_readdir_free_pagearray(struct page **pages, unsigned int npages)
562{
563 unsigned int i;
564 for (i = 0; i < npages; i++)
565 put_page(pages[i]);
566}
567
568static
569void nfs_readdir_free_large_page(void *ptr, struct page **pages,
570 unsigned int npages)
571{
Bryan Schumaker56e4ebf2010-10-20 15:44:37 -0400572 nfs_readdir_free_pagearray(pages, npages);
573}
574
575/*
576 * nfs_readdir_large_page will allocate pages that must be freed with a call
577 * to nfs_readdir_free_large_page
578 */
579static
Trond Myklebust66502392011-01-08 17:45:38 -0500580int nfs_readdir_large_page(struct page **pages, unsigned int npages)
Bryan Schumaker56e4ebf2010-10-20 15:44:37 -0400581{
Bryan Schumaker56e4ebf2010-10-20 15:44:37 -0400582 unsigned int i;
583
584 for (i = 0; i < npages; i++) {
585 struct page *page = alloc_page(GFP_KERNEL);
586 if (page == NULL)
587 goto out_freepages;
588 pages[i] = page;
589 }
Trond Myklebust66502392011-01-08 17:45:38 -0500590 return 0;
Bryan Schumaker56e4ebf2010-10-20 15:44:37 -0400591
Bryan Schumaker56e4ebf2010-10-20 15:44:37 -0400592out_freepages:
593 nfs_readdir_free_pagearray(pages, i);
Trond Myklebust66502392011-01-08 17:45:38 -0500594 return -ENOMEM;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400595}
596
597static
598int nfs_readdir_xdr_to_array(nfs_readdir_descriptor_t *desc, struct page *page, struct inode *inode)
599{
Bryan Schumaker56e4ebf2010-10-20 15:44:37 -0400600 struct page *pages[NFS_MAX_READDIR_PAGES];
601 void *pages_ptr = NULL;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400602 struct nfs_entry entry;
603 struct file *file = desc->file;
604 struct nfs_cache_array *array;
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500605 int status = -ENOMEM;
Bryan Schumaker56e4ebf2010-10-20 15:44:37 -0400606 unsigned int array_size = ARRAY_SIZE(pages);
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400607
608 entry.prev_cookie = 0;
Trond Myklebust0aded702010-11-30 21:56:32 -0500609 entry.cookie = desc->last_cookie;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400610 entry.eof = 0;
611 entry.fh = nfs_alloc_fhandle();
612 entry.fattr = nfs_alloc_fattr();
Chuck Lever573c4e12010-12-14 14:58:11 +0000613 entry.server = NFS_SERVER(inode);
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400614 if (entry.fh == NULL || entry.fattr == NULL)
615 goto out;
616
617 array = nfs_readdir_get_array(page);
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500618 if (IS_ERR(array)) {
619 status = PTR_ERR(array);
620 goto out;
621 }
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400622 memset(array, 0, sizeof(struct nfs_cache_array));
623 array->eof_index = -1;
624
Trond Myklebust66502392011-01-08 17:45:38 -0500625 status = nfs_readdir_large_page(pages, array_size);
626 if (status < 0)
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400627 goto out_release_array;
628 do {
Trond Myklebustac396122010-11-15 20:26:22 -0500629 unsigned int pglen;
Bryan Schumaker56e4ebf2010-10-20 15:44:37 -0400630 status = nfs_readdir_xdr_filler(pages, desc, &entry, file, inode);
Bryan Schumakerbabddc72010-10-20 15:44:29 -0400631
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400632 if (status < 0)
633 break;
Trond Myklebustac396122010-11-15 20:26:22 -0500634 pglen = status;
Trond Myklebust66502392011-01-08 17:45:38 -0500635 status = nfs_readdir_page_filler(desc, &entry, pages, page, pglen);
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500636 if (status < 0) {
637 if (status == -ENOSPC)
638 status = 0;
639 break;
640 }
641 } while (array->eof_index < 0);
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400642
Bryan Schumaker56e4ebf2010-10-20 15:44:37 -0400643 nfs_readdir_free_large_page(pages_ptr, pages, array_size);
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400644out_release_array:
645 nfs_readdir_release_array(page);
646out:
647 nfs_free_fattr(entry.fattr);
648 nfs_free_fhandle(entry.fh);
649 return status;
650}
651
652/*
653 * Now we cache directories properly, by converting xdr information
654 * to an array that can be used for lookups later. This results in
655 * fewer cache pages, since we can store more information on each page.
656 * We only need to convert from xdr once so future lookups are much simpler
657 */
658static
659int nfs_readdir_filler(nfs_readdir_descriptor_t *desc, struct page* page)
660{
661 struct inode *inode = desc->file->f_path.dentry->d_inode;
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500662 int ret;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400663
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500664 ret = nfs_readdir_xdr_to_array(desc, page, inode);
665 if (ret < 0)
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400666 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 SetPageUptodate(page);
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400668
Trond Myklebust2aac05a2008-07-07 13:26:10 -0400669 if (invalidate_inode_pages2_range(inode->i_mapping, page->index + 1, -1) < 0) {
Trond Myklebustcd9ae2b2006-10-19 23:28:40 -0700670 /* Should never happen */
671 nfs_zap_mapping(inode, inode->i_mapping);
672 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 unlock_page(page);
674 return 0;
675 error:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676 unlock_page(page);
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500677 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678}
679
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400680static
681void cache_page_release(nfs_readdir_descriptor_t *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682{
Trond Myklebust11de3b12010-12-01 14:17:06 -0500683 if (!desc->page->mapping)
684 nfs_readdir_clear_array(desc->page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 page_cache_release(desc->page);
686 desc->page = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687}
688
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400689static
690struct page *get_cache_page(nfs_readdir_descriptor_t *desc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691{
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500692 return read_cache_page(desc->file->f_path.dentry->d_inode->i_mapping,
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400693 desc->page_index, (filler_t *)nfs_readdir_filler, desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694}
695
696/*
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400697 * Returns 0 if desc->dir_cookie was found on page desc->page_index
Olivier Galibert00a92642005-06-22 17:16:29 +0000698 */
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400699static
700int find_cache_page(nfs_readdir_descriptor_t *desc)
Olivier Galibert00a92642005-06-22 17:16:29 +0000701{
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400702 int res;
Olivier Galibert00a92642005-06-22 17:16:29 +0000703
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400704 desc->page = get_cache_page(desc);
705 if (IS_ERR(desc->page))
706 return PTR_ERR(desc->page);
Olivier Galibert00a92642005-06-22 17:16:29 +0000707
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400708 res = nfs_readdir_search_array(desc);
Trond Myklebust47c716c2010-12-07 12:44:56 -0500709 if (res != 0)
710 cache_page_release(desc);
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400711 return res;
Olivier Galibert00a92642005-06-22 17:16:29 +0000712}
713
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400714/* Search for desc->dir_cookie from the beginning of the page cache */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715static inline
716int readdir_search_pagecache(nfs_readdir_descriptor_t *desc)
717{
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500718 int res;
Olivier Galibert00a92642005-06-22 17:16:29 +0000719
Trond Myklebust0aded702010-11-30 21:56:32 -0500720 if (desc->page_index == 0) {
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500721 desc->current_index = 0;
Trond Myklebust0aded702010-11-30 21:56:32 -0500722 desc->last_cookie = 0;
723 }
Trond Myklebust47c716c2010-12-07 12:44:56 -0500724 do {
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400725 res = find_cache_page(desc);
Trond Myklebust47c716c2010-12-07 12:44:56 -0500726 } while (res == -EAGAIN);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 return res;
728}
729
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730/*
731 * Once we've found the start of the dirent within a page: fill 'er up...
732 */
733static
734int nfs_do_filldir(nfs_readdir_descriptor_t *desc, void *dirent,
735 filldir_t filldir)
736{
737 struct file *file = desc->file;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400738 int i = 0;
739 int res = 0;
740 struct nfs_cache_array *array = NULL;
Bryan Schumaker8ef2ce32011-03-23 15:04:31 -0400741 struct nfs_open_dir_context *ctx = file->private_data;
742
743 if (ctx->duped != 0 && ctx->dup_cookie == *desc->dir_cookie) {
744 if (printk_ratelimit()) {
745 pr_notice("NFS: directory %s/%s contains a readdir loop. "
746 "Please contact your server vendor. "
747 "Offending cookie: %llu\n",
748 file->f_dentry->d_parent->d_name.name,
749 file->f_dentry->d_name.name,
750 *desc->dir_cookie);
751 }
752 res = -ELOOP;
753 goto out;
754 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400756 array = nfs_readdir_get_array(desc->page);
Trond Myklebuste7c58e92010-11-20 13:22:24 -0500757 if (IS_ERR(array)) {
758 res = PTR_ERR(array);
759 goto out;
760 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400762 for (i = desc->cache_entry_index; i < array->size; i++) {
Trond Myklebustece0b422010-11-20 13:55:33 -0500763 struct nfs_cache_array_entry *ent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Trond Myklebustece0b422010-11-20 13:55:33 -0500765 ent = &array->array[i];
766 if (filldir(dirent, ent->string.name, ent->string.len,
Trond Myklebust0b26a0b2010-11-20 14:26:44 -0500767 file->f_pos, nfs_compat_user_ino64(ent->ino),
768 ent->d_type) < 0) {
Trond Myklebustece0b422010-11-20 13:55:33 -0500769 desc->eof = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 break;
Trond Myklebustece0b422010-11-20 13:55:33 -0500771 }
Olivier Galibert00a92642005-06-22 17:16:29 +0000772 file->f_pos++;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400773 if (i < (array->size-1))
774 *desc->dir_cookie = array->array[i+1].cookie;
775 else
776 *desc->dir_cookie = array->last_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 }
Trond Myklebust47c716c2010-12-07 12:44:56 -0500778 if (array->eof_index >= 0)
Trond Myklebust8cd51a02010-11-15 20:26:22 -0500779 desc->eof = 1;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400780
781 nfs_readdir_release_array(desc->page);
Trond Myklebuste7c58e92010-11-20 13:22:24 -0500782out:
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400783 cache_page_release(desc);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500784 dfprintk(DIRCACHE, "NFS: nfs_do_filldir() filling ended @ cookie %Lu; returning = %d\n",
785 (unsigned long long)*desc->dir_cookie, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 return res;
787}
788
789/*
790 * If we cannot find a cookie in our cache, we suspect that this is
791 * because it points to a deleted file, so we ask the server to return
792 * whatever it thinks is the next entry. We then feed this to filldir.
793 * If all goes well, we should then be able to find our way round the
794 * cache on the next call to readdir_search_pagecache();
795 *
796 * NOTE: we cannot add the anonymous page to the pagecache because
797 * the data it contains might not be page aligned. Besides,
798 * we should already have a complete representation of the
799 * directory in the page cache by the time we get here.
800 */
801static inline
802int uncached_readdir(nfs_readdir_descriptor_t *desc, void *dirent,
803 filldir_t filldir)
804{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700805 struct page *page = NULL;
806 int status;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400807 struct inode *inode = desc->file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500809 dfprintk(DIRCACHE, "NFS: uncached_readdir() searching for cookie %Lu\n",
810 (unsigned long long)*desc->dir_cookie);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700811
812 page = alloc_page(GFP_HIGHUSER);
813 if (!page) {
814 status = -ENOMEM;
815 goto out;
816 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
Trond Myklebust7a8e1dc2010-11-20 13:24:46 -0500818 desc->page_index = 0;
Trond Myklebust0aded702010-11-30 21:56:32 -0500819 desc->last_cookie = *desc->dir_cookie;
Trond Myklebust7a8e1dc2010-11-20 13:24:46 -0500820 desc->page = page;
821
Trond Myklebust85f86072010-11-20 13:24:49 -0500822 status = nfs_readdir_xdr_to_array(desc, page, inode);
823 if (status < 0)
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400824 goto out_release;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400825
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 status = nfs_do_filldir(desc, dirent, filldir);
827
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 out:
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500829 dfprintk(DIRCACHE, "NFS: %s: returns %d\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -0700830 __func__, status);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700831 return status;
832 out_release:
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400833 cache_page_release(desc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 goto out;
835}
836
Olivier Galibert00a92642005-06-22 17:16:29 +0000837/* The file offset position represents the dirent entry number. A
838 last cookie cache takes care of the common case of reading the
839 whole directory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 */
841static int nfs_readdir(struct file *filp, void *dirent, filldir_t filldir)
842{
Josef "Jeff" Sipek01cce932006-12-08 02:36:40 -0800843 struct dentry *dentry = filp->f_path.dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 struct inode *inode = dentry->d_inode;
845 nfs_readdir_descriptor_t my_desc,
846 *desc = &my_desc;
Bryan Schumaker480c2002011-03-23 14:48:29 -0400847 struct nfs_open_dir_context *dir_ctx = filp->private_data;
Trond Myklebust47c716c2010-12-07 12:44:56 -0500848 int res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
Chuck Lever6da24bc2008-06-11 17:55:58 -0400850 dfprintk(FILE, "NFS: readdir(%s/%s) starting at cookie %llu\n",
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500851 dentry->d_parent->d_name.name, dentry->d_name.name,
852 (long long)filp->f_pos);
Chuck Lever91d5b472006-03-20 13:44:14 -0500853 nfs_inc_stats(inode, NFSIOS_VFSGETDENTS);
854
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855 /*
Olivier Galibert00a92642005-06-22 17:16:29 +0000856 * filp->f_pos points to the dirent entry number.
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000857 * *desc->dir_cookie has the cookie for the next entry. We have
Olivier Galibert00a92642005-06-22 17:16:29 +0000858 * to either find the entry with the appropriate number or
859 * revalidate the cookie.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 */
861 memset(desc, 0, sizeof(*desc));
862
863 desc->file = filp;
Bryan Schumaker480c2002011-03-23 14:48:29 -0400864 desc->dir_cookie = &dir_ctx->dir_cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 desc->decode = NFS_PROTO(inode)->decode_dirent;
866 desc->plus = NFS_USE_READDIRPLUS(inode);
867
Trond Myklebust565277f2007-10-15 18:17:53 -0400868 nfs_block_sillyrename(dentry);
Trond Myklebust1cda7072010-02-19 17:03:30 -0800869 res = nfs_revalidate_mapping(inode, filp->f_mapping);
Trond Myklebustfccca7f2008-01-26 17:37:47 -0500870 if (res < 0)
871 goto out;
872
Trond Myklebust47c716c2010-12-07 12:44:56 -0500873 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874 res = readdir_search_pagecache(desc);
Olivier Galibert00a92642005-06-22 17:16:29 +0000875
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876 if (res == -EBADCOOKIE) {
Trond Myklebustece0b422010-11-20 13:55:33 -0500877 res = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 /* This means either end of directory */
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400879 if (*desc->dir_cookie && desc->eof == 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 /* Or that the server has 'lost' a cookie */
881 res = uncached_readdir(desc, dirent, filldir);
Trond Myklebustece0b422010-11-20 13:55:33 -0500882 if (res == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 continue;
884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 break;
886 }
887 if (res == -ETOOSMALL && desc->plus) {
Benny Halevy3a10c302008-01-23 08:58:59 +0200888 clear_bit(NFS_INO_ADVISE_RDPLUS, &NFS_I(inode)->flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 nfs_zap_caches(inode);
Trond Myklebustbaf57a02010-09-24 18:49:43 -0400890 desc->page_index = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 desc->plus = 0;
Bryan Schumakerd1bacf92010-09-24 14:48:42 -0400892 desc->eof = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893 continue;
894 }
895 if (res < 0)
896 break;
897
898 res = nfs_do_filldir(desc, dirent, filldir);
Trond Myklebustece0b422010-11-20 13:55:33 -0500899 if (res < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700900 break;
Trond Myklebust47c716c2010-12-07 12:44:56 -0500901 } while (!desc->eof);
Trond Myklebustfccca7f2008-01-26 17:37:47 -0500902out:
Trond Myklebust565277f2007-10-15 18:17:53 -0400903 nfs_unblock_sillyrename(dentry);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500904 if (res > 0)
905 res = 0;
Trond Myklebustaa49b4c2010-04-16 16:22:49 -0400906 dfprintk(FILE, "NFS: readdir(%s/%s) returns %d\n",
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500907 dentry->d_parent->d_name.name, dentry->d_name.name,
908 res);
909 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910}
911
Trond Myklebust10afec92007-05-14 17:16:04 -0400912static loff_t nfs_llseek_dir(struct file *filp, loff_t offset, int origin)
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000913{
Chuck Leverb84e06c2008-06-11 17:55:34 -0400914 struct dentry *dentry = filp->f_path.dentry;
915 struct inode *inode = dentry->d_inode;
Bryan Schumaker480c2002011-03-23 14:48:29 -0400916 struct nfs_open_dir_context *dir_ctx = filp->private_data;
Chuck Leverb84e06c2008-06-11 17:55:34 -0400917
Chuck Lever6da24bc2008-06-11 17:55:58 -0400918 dfprintk(FILE, "NFS: llseek dir(%s/%s, %lld, %d)\n",
Chuck Leverb84e06c2008-06-11 17:55:34 -0400919 dentry->d_parent->d_name.name,
920 dentry->d_name.name,
921 offset, origin);
922
923 mutex_lock(&inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000924 switch (origin) {
925 case 1:
926 offset += filp->f_pos;
927 case 0:
928 if (offset >= 0)
929 break;
930 default:
931 offset = -EINVAL;
932 goto out;
933 }
934 if (offset != filp->f_pos) {
935 filp->f_pos = offset;
Bryan Schumaker480c2002011-03-23 14:48:29 -0400936 dir_ctx->dir_cookie = 0;
Bryan Schumaker8ef2ce32011-03-23 15:04:31 -0400937 dir_ctx->duped = 0;
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000938 }
939out:
Chuck Leverb84e06c2008-06-11 17:55:34 -0400940 mutex_unlock(&inode->i_mutex);
Trond Myklebustf0dd2132005-06-22 17:16:29 +0000941 return offset;
942}
943
Linus Torvalds1da177e2005-04-16 15:20:36 -0700944/*
945 * All directory operations under NFS are synchronous, so fsync()
946 * is a dummy operation.
947 */
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200948static int nfs_fsync_dir(struct file *filp, int datasync)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949{
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200950 struct dentry *dentry = filp->f_path.dentry;
951
Chuck Lever6da24bc2008-06-11 17:55:58 -0400952 dfprintk(FILE, "NFS: fsync dir(%s/%s) datasync %d\n",
Chuck Lever1e7cb3d2006-03-20 13:44:24 -0500953 dentry->d_parent->d_name.name, dentry->d_name.name,
954 datasync);
955
Chuck Lever54917782008-05-27 16:29:07 -0400956 nfs_inc_stats(dentry->d_inode, NFSIOS_VFSFSYNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 return 0;
958}
959
Trond Myklebustbfc69a42007-10-15 18:18:29 -0400960/**
961 * nfs_force_lookup_revalidate - Mark the directory as having changed
962 * @dir - pointer to directory inode
963 *
964 * This forces the revalidation code in nfs_lookup_revalidate() to do a
965 * full lookup on all child dentries of 'dir' whenever a change occurs
966 * on the server that might have invalidated our dcache.
967 *
968 * The caller should be holding dir->i_lock
969 */
970void nfs_force_lookup_revalidate(struct inode *dir)
971{
Trond Myklebust011935a2008-10-14 19:24:50 -0400972 NFS_I(dir)->cache_change_attribute++;
Trond Myklebustbfc69a42007-10-15 18:18:29 -0400973}
974
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975/*
976 * A check for whether or not the parent directory has changed.
977 * In the case it has, we assume that the dentries are untrustworthy
978 * and may need to be looked up again.
979 */
Trond Myklebustc79ba782007-01-31 08:16:24 -0500980static int nfs_check_verifier(struct inode *dir, struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981{
982 if (IS_ROOT(dentry))
983 return 1;
Trond Myklebust4eec9522008-07-15 17:58:13 -0400984 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONE)
985 return 0;
Trond Myklebustf2c77f42007-10-02 12:54:39 -0400986 if (!nfs_verify_change_attribute(dir, dentry->d_time))
987 return 0;
988 /* Revalidate nfsi->cache_change_attribute before we declare a match */
989 if (nfs_revalidate_inode(NFS_SERVER(dir), dir) < 0)
990 return 0;
991 if (!nfs_verify_change_attribute(dir, dentry->d_time))
992 return 0;
993 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994}
995
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996/*
Trond Myklebust1d6757f2005-06-07 18:37:01 -0400997 * Return the intent data that applies to this particular path component
998 *
999 * Note that the current set of intents only apply to the very last
1000 * component of the path.
1001 * We check for this using LOOKUP_CONTINUE and LOOKUP_PARENT.
1002 */
Nick Piggin34286d62011-01-07 17:49:57 +11001003static inline unsigned int nfs_lookup_check_intent(struct nameidata *nd,
1004 unsigned int mask)
Trond Myklebust1d6757f2005-06-07 18:37:01 -04001005{
1006 if (nd->flags & (LOOKUP_CONTINUE|LOOKUP_PARENT))
1007 return 0;
1008 return nd->flags & mask;
1009}
1010
1011/*
Trond Myklebusta12802c2007-10-02 19:13:04 -04001012 * Use intent information to check whether or not we're going to do
1013 * an O_EXCL create using this path component.
1014 */
1015static int nfs_is_exclusive_create(struct inode *dir, struct nameidata *nd)
1016{
1017 if (NFS_PROTO(dir)->version == 2)
1018 return 0;
Al Viro35165862008-08-05 03:00:49 -04001019 return nd && nfs_lookup_check_intent(nd, LOOKUP_EXCL);
Trond Myklebusta12802c2007-10-02 19:13:04 -04001020}
1021
1022/*
Trond Myklebust1d6757f2005-06-07 18:37:01 -04001023 * Inode and filehandle revalidation for lookups.
1024 *
1025 * We force revalidation in the cases where the VFS sets LOOKUP_REVAL,
1026 * or if the intent information indicates that we're about to open this
1027 * particular file and the "nocto" mount flag is not set.
1028 *
1029 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030static inline
1031int nfs_lookup_verify_inode(struct inode *inode, struct nameidata *nd)
1032{
1033 struct nfs_server *server = NFS_SERVER(inode);
1034
David Howells36d43a42011-01-14 18:45:42 +00001035 if (IS_AUTOMOUNT(inode))
Trond Myklebust4e99a1f2008-03-06 12:34:59 -05001036 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 if (nd != NULL) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 /* VFS wants an on-the-wire revalidation */
Trond Myklebust1d6757f2005-06-07 18:37:01 -04001039 if (nd->flags & LOOKUP_REVAL)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 goto out_force;
1041 /* This is an open(2) */
Trond Myklebust1d6757f2005-06-07 18:37:01 -04001042 if (nfs_lookup_check_intent(nd, LOOKUP_OPEN) != 0 &&
Trond Myklebust4e0641a2006-07-05 13:05:13 -04001043 !(server->flags & NFS_MOUNT_NOCTO) &&
1044 (S_ISREG(inode->i_mode) ||
1045 S_ISDIR(inode->i_mode)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 goto out_force;
Trond Myklebust4f48af42007-10-02 23:13:32 -04001047 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 }
1049 return nfs_revalidate_inode(server, inode);
1050out_force:
1051 return __nfs_revalidate_inode(server, inode);
1052}
1053
1054/*
1055 * We judge how long we want to trust negative
1056 * dentries by looking at the parent inode mtime.
1057 *
1058 * If parent mtime has changed, we revalidate, else we wait for a
1059 * period corresponding to the parent's attribute cache timeout value.
1060 */
1061static inline
1062int nfs_neg_need_reval(struct inode *dir, struct dentry *dentry,
1063 struct nameidata *nd)
1064{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 /* Don't revalidate a negative dentry if we're creating a new file */
Trond Myklebust1d6757f2005-06-07 18:37:01 -04001066 if (nd != NULL && nfs_lookup_check_intent(nd, LOOKUP_CREATE) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067 return 0;
Trond Myklebust4eec9522008-07-15 17:58:13 -04001068 if (NFS_SERVER(dir)->flags & NFS_MOUNT_LOOKUP_CACHE_NONEG)
1069 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 return !nfs_check_verifier(dir, dentry);
1071}
1072
1073/*
1074 * This is called every time the dcache has a lookup hit,
1075 * and we should check whether we can really trust that
1076 * lookup.
1077 *
1078 * NOTE! The hit can be a negative hit too, don't assume
1079 * we have an inode!
1080 *
1081 * If the parent directory is seen to have changed, we throw out the
1082 * cached dentry and do a new lookup.
1083 */
Nick Piggin34286d62011-01-07 17:49:57 +11001084static int nfs_lookup_revalidate(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085{
1086 struct inode *dir;
1087 struct inode *inode;
1088 struct dentry *parent;
Trond Myklebuste1fb4d02010-04-16 16:22:47 -04001089 struct nfs_fh *fhandle = NULL;
1090 struct nfs_fattr *fattr = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
Nick Piggin34286d62011-01-07 17:49:57 +11001093 if (nd->flags & LOOKUP_RCU)
1094 return -ECHILD;
1095
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 parent = dget_parent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 dir = parent->d_inode;
Chuck Lever91d5b472006-03-20 13:44:14 -05001098 nfs_inc_stats(dir, NFSIOS_DENTRYREVALIDATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 inode = dentry->d_inode;
1100
1101 if (!inode) {
1102 if (nfs_neg_need_reval(dir, dentry, nd))
1103 goto out_bad;
1104 goto out_valid;
1105 }
1106
1107 if (is_bad_inode(inode)) {
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001108 dfprintk(LOOKUPCACHE, "%s: %s/%s has dud inode\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -07001109 __func__, dentry->d_parent->d_name.name,
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001110 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 goto out_bad;
1112 }
1113
Trond Myklebust15860ab2008-12-23 15:21:54 -05001114 if (nfs_have_delegation(inode, FMODE_READ))
1115 goto out_set_verifier;
1116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 /* Force a full look up iff the parent directory has changed */
Trond Myklebusta12802c2007-10-02 19:13:04 -04001118 if (!nfs_is_exclusive_create(dir, nd) && nfs_check_verifier(dir, dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 if (nfs_lookup_verify_inode(inode, nd))
1120 goto out_zap_parent;
1121 goto out_valid;
1122 }
1123
1124 if (NFS_STALE(inode))
1125 goto out_bad;
1126
Trond Myklebuste1fb4d02010-04-16 16:22:47 -04001127 error = -ENOMEM;
1128 fhandle = nfs_alloc_fhandle();
1129 fattr = nfs_alloc_fattr();
1130 if (fhandle == NULL || fattr == NULL)
1131 goto out_error;
1132
Bryan Schumaker7c513052011-03-24 17:12:24 +00001133 error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 if (error)
1135 goto out_bad;
Trond Myklebuste1fb4d02010-04-16 16:22:47 -04001136 if (nfs_compare_fh(NFS_FH(inode), fhandle))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 goto out_bad;
Trond Myklebuste1fb4d02010-04-16 16:22:47 -04001138 if ((error = nfs_refresh_inode(inode, fattr)) != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 goto out_bad;
1140
Trond Myklebuste1fb4d02010-04-16 16:22:47 -04001141 nfs_free_fattr(fattr);
1142 nfs_free_fhandle(fhandle);
Trond Myklebust15860ab2008-12-23 15:21:54 -05001143out_set_verifier:
Trond Myklebustcf8ba452007-10-01 13:46:53 -04001144 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 out_valid:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001147 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is valid\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -07001148 __func__, dentry->d_parent->d_name.name,
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001149 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 return 1;
1151out_zap_parent:
1152 nfs_zap_caches(dir);
1153 out_bad:
Trond Myklebusta1643a92007-09-29 17:25:43 -04001154 nfs_mark_for_revalidate(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 if (inode && S_ISDIR(inode->i_mode)) {
1156 /* Purge readdir caches. */
1157 nfs_zap_caches(inode);
1158 /* If we have submounts, don't unhash ! */
1159 if (have_submounts(dentry))
1160 goto out_valid;
Al Virod9e80b72010-04-29 03:10:43 +01001161 if (dentry->d_flags & DCACHE_DISCONNECTED)
1162 goto out_valid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 shrink_dcache_parent(dentry);
1164 }
1165 d_drop(dentry);
Trond Myklebuste1fb4d02010-04-16 16:22:47 -04001166 nfs_free_fattr(fattr);
1167 nfs_free_fhandle(fhandle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001168 dput(parent);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001169 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) is invalid\n",
Harvey Harrison3110ff82008-05-02 13:42:44 -07001170 __func__, dentry->d_parent->d_name.name,
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001171 dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001172 return 0;
Trond Myklebuste1fb4d02010-04-16 16:22:47 -04001173out_error:
1174 nfs_free_fattr(fattr);
1175 nfs_free_fhandle(fhandle);
1176 dput(parent);
1177 dfprintk(LOOKUPCACHE, "NFS: %s(%s/%s) lookup returned error %d\n",
1178 __func__, dentry->d_parent->d_name.name,
1179 dentry->d_name.name, error);
1180 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
1182
1183/*
1184 * This is called from dput() when d_count is going to 0.
1185 */
Nick Pigginfe15ce42011-01-07 17:49:23 +11001186static int nfs_dentry_delete(const struct dentry *dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001187{
1188 dfprintk(VFS, "NFS: dentry_delete(%s/%s, %x)\n",
1189 dentry->d_parent->d_name.name, dentry->d_name.name,
1190 dentry->d_flags);
1191
Trond Myklebust77f11192008-01-28 19:43:19 -05001192 /* Unhash any dentry with a stale inode */
1193 if (dentry->d_inode != NULL && NFS_STALE(dentry->d_inode))
1194 return 1;
1195
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1197 /* Unhash it, so that ->d_iput() would be called */
1198 return 1;
1199 }
1200 if (!(dentry->d_sb->s_flags & MS_ACTIVE)) {
1201 /* Unhash it, so that ancestors of killed async unlink
1202 * files will be cleaned up during umount */
1203 return 1;
1204 }
1205 return 0;
1206
1207}
1208
Trond Myklebust1b83d702008-06-11 15:44:04 -04001209static void nfs_drop_nlink(struct inode *inode)
1210{
1211 spin_lock(&inode->i_lock);
1212 if (inode->i_nlink > 0)
1213 drop_nlink(inode);
1214 spin_unlock(&inode->i_lock);
1215}
1216
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217/*
1218 * Called when the dentry loses inode.
1219 * We use it to clean up silly-renamed files.
1220 */
1221static void nfs_dentry_iput(struct dentry *dentry, struct inode *inode)
1222{
Neil Brown83672d32007-02-26 12:48:25 +11001223 if (S_ISDIR(inode->i_mode))
1224 /* drop any readdir cache as it could easily be old */
1225 NFS_I(inode)->cache_validity |= NFS_INO_INVALID_DATA;
1226
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001228 drop_nlink(inode);
Trond Myklebuste4eff1a2007-07-14 15:39:58 -04001229 nfs_complete_unlink(dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 iput(inode);
1232}
1233
Al Virob1942c52011-03-16 05:44:14 -04001234static void nfs_d_release(struct dentry *dentry)
1235{
1236 /* free cached devname value, if it survived that far */
1237 if (unlikely(dentry->d_fsdata)) {
1238 if (dentry->d_flags & DCACHE_NFSFS_RENAMED)
1239 WARN_ON(1);
1240 else
1241 kfree(dentry->d_fsdata);
1242 }
1243}
1244
Al Virof786aa92009-02-20 05:51:22 +00001245const struct dentry_operations nfs_dentry_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 .d_revalidate = nfs_lookup_revalidate,
1247 .d_delete = nfs_dentry_delete,
1248 .d_iput = nfs_dentry_iput,
David Howells36d43a42011-01-14 18:45:42 +00001249 .d_automount = nfs_d_automount,
Al Virob1942c52011-03-16 05:44:14 -04001250 .d_release = nfs_d_release,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251};
1252
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253static struct dentry *nfs_lookup(struct inode *dir, struct dentry * dentry, struct nameidata *nd)
1254{
1255 struct dentry *res;
Trond Myklebust565277f2007-10-15 18:17:53 -04001256 struct dentry *parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257 struct inode *inode = NULL;
Trond Myklebuste1fb4d02010-04-16 16:22:47 -04001258 struct nfs_fh *fhandle = NULL;
1259 struct nfs_fattr *fattr = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261
1262 dfprintk(VFS, "NFS: lookup(%s/%s)\n",
1263 dentry->d_parent->d_name.name, dentry->d_name.name);
Chuck Lever91d5b472006-03-20 13:44:14 -05001264 nfs_inc_stats(dir, NFSIOS_VFSLOOKUP);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265
1266 res = ERR_PTR(-ENAMETOOLONG);
1267 if (dentry->d_name.len > NFS_SERVER(dir)->namelen)
1268 goto out;
1269
Trond Myklebustfd684072006-09-05 12:27:44 -04001270 /*
1271 * If we're doing an exclusive create, optimize away the lookup
1272 * but don't hash the dentry.
1273 */
1274 if (nfs_is_exclusive_create(dir, nd)) {
1275 d_instantiate(dentry, NULL);
1276 res = NULL;
Trond Myklebustfc0f6842008-06-11 15:44:20 -04001277 goto out;
Trond Myklebustfd684072006-09-05 12:27:44 -04001278 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279
Trond Myklebuste1fb4d02010-04-16 16:22:47 -04001280 res = ERR_PTR(-ENOMEM);
1281 fhandle = nfs_alloc_fhandle();
1282 fattr = nfs_alloc_fattr();
1283 if (fhandle == NULL || fattr == NULL)
1284 goto out;
1285
Trond Myklebust565277f2007-10-15 18:17:53 -04001286 parent = dentry->d_parent;
1287 /* Protect against concurrent sillydeletes */
1288 nfs_block_sillyrename(parent);
Bryan Schumaker7c513052011-03-24 17:12:24 +00001289 error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 if (error == -ENOENT)
1291 goto no_entry;
1292 if (error < 0) {
1293 res = ERR_PTR(error);
Trond Myklebust565277f2007-10-15 18:17:53 -04001294 goto out_unblock_sillyrename;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 }
Trond Myklebuste1fb4d02010-04-16 16:22:47 -04001296 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
Namhyung Kimbf0c84f2010-12-28 17:02:46 +00001297 res = ERR_CAST(inode);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001298 if (IS_ERR(res))
Trond Myklebust565277f2007-10-15 18:17:53 -04001299 goto out_unblock_sillyrename;
David Howells54ceac42006-08-22 20:06:13 -04001300
Linus Torvalds1da177e2005-04-16 15:20:36 -07001301no_entry:
David Howells54ceac42006-08-22 20:06:13 -04001302 res = d_materialise_unique(dentry, inode);
Trond Myklebust9eaef272006-10-21 10:24:20 -07001303 if (res != NULL) {
1304 if (IS_ERR(res))
Trond Myklebust565277f2007-10-15 18:17:53 -04001305 goto out_unblock_sillyrename;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306 dentry = res;
Trond Myklebust9eaef272006-10-21 10:24:20 -07001307 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Trond Myklebust565277f2007-10-15 18:17:53 -04001309out_unblock_sillyrename:
1310 nfs_unblock_sillyrename(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311out:
Trond Myklebuste1fb4d02010-04-16 16:22:47 -04001312 nfs_free_fattr(fattr);
1313 nfs_free_fhandle(fhandle);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 return res;
1315}
1316
1317#ifdef CONFIG_NFS_V4
1318static int nfs_open_revalidate(struct dentry *, struct nameidata *);
1319
Al Virof786aa92009-02-20 05:51:22 +00001320const struct dentry_operations nfs4_dentry_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 .d_revalidate = nfs_open_revalidate,
1322 .d_delete = nfs_dentry_delete,
1323 .d_iput = nfs_dentry_iput,
David Howells36d43a42011-01-14 18:45:42 +00001324 .d_automount = nfs_d_automount,
Al Virob1942c52011-03-16 05:44:14 -04001325 .d_release = nfs_d_release,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326};
1327
Trond Myklebust1d6757f2005-06-07 18:37:01 -04001328/*
1329 * Use intent information to determine whether we need to substitute
1330 * the NFSv4-style stateful OPEN for the LOOKUP call
1331 */
Trond Myklebust5584c302008-12-23 15:21:54 -05001332static int is_atomic_open(struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333{
Trond Myklebust1d6757f2005-06-07 18:37:01 -04001334 if (nd == NULL || nfs_lookup_check_intent(nd, LOOKUP_OPEN) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 return 0;
1336 /* NFS does not (yet) have a stateful open for directories */
1337 if (nd->flags & LOOKUP_DIRECTORY)
1338 return 0;
1339 /* Are we trying to write to a read only partition? */
Dave Hansen2c463e92008-02-15 14:37:56 -08001340 if (__mnt_is_readonly(nd->path.mnt) &&
1341 (nd->intent.open.flags & (O_CREAT|O_TRUNC|FMODE_WRITE)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 return 0;
1343 return 1;
1344}
1345
Al Viro51141592011-06-22 18:47:28 -04001346static struct nfs_open_context *create_nfs_open_context(struct dentry *dentry, int open_flags)
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001347{
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001348 struct nfs_open_context *ctx;
1349 struct rpc_cred *cred;
Al Viro51141592011-06-22 18:47:28 -04001350 fmode_t fmode = open_flags & (FMODE_READ | FMODE_WRITE | FMODE_EXEC);
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001351
1352 cred = rpc_lookup_cred();
1353 if (IS_ERR(cred))
1354 return ERR_CAST(cred);
Al Viro3d4ff432011-06-22 18:40:12 -04001355 ctx = alloc_nfs_open_context(dentry, cred, fmode);
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001356 put_rpccred(cred);
1357 if (ctx == NULL)
1358 return ERR_PTR(-ENOMEM);
1359 return ctx;
1360}
1361
1362static int do_open(struct inode *inode, struct file *filp)
1363{
1364 nfs_fscache_set_inode_cookie(inode, filp);
1365 return 0;
1366}
1367
1368static int nfs_intent_set_file(struct nameidata *nd, struct nfs_open_context *ctx)
1369{
1370 struct file *filp;
1371 int ret = 0;
1372
1373 /* If the open_intent is for execute, we have an extra check to make */
1374 if (ctx->mode & FMODE_EXEC) {
Al Viro3d4ff432011-06-22 18:40:12 -04001375 ret = nfs_may_open(ctx->dentry->d_inode,
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001376 ctx->cred,
1377 nd->intent.open.flags);
1378 if (ret < 0)
1379 goto out;
1380 }
Al Viro3d4ff432011-06-22 18:40:12 -04001381 filp = lookup_instantiate_filp(nd, ctx->dentry, do_open);
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001382 if (IS_ERR(filp))
1383 ret = PTR_ERR(filp);
1384 else
1385 nfs_file_set_open_context(filp, ctx);
1386out:
1387 put_nfs_open_context(ctx);
1388 return ret;
1389}
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391static struct dentry *nfs_atomic_lookup(struct inode *dir, struct dentry *dentry, struct nameidata *nd)
1392{
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001393 struct nfs_open_context *ctx;
1394 struct iattr attr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 struct dentry *res = NULL;
Trond Myklebustf46e0bd2010-09-17 10:56:50 -04001396 struct inode *inode;
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001397 int open_flags;
Trond Myklebust898f6352010-10-23 11:24:25 -04001398 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001400 dfprintk(VFS, "NFS: atomic_lookup(%s/%ld), %s\n",
1401 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1402
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 /* Check that we are indeed trying to open this file */
Trond Myklebust5584c302008-12-23 15:21:54 -05001404 if (!is_atomic_open(nd))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 goto no_open;
1406
1407 if (dentry->d_name.len > NFS_SERVER(dir)->namelen) {
1408 res = ERR_PTR(-ENAMETOOLONG);
1409 goto out;
1410 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411
Trond Myklebustd4d9cdc2007-10-02 18:38:53 -04001412 /* Let vfs_create() deal with O_EXCL. Instantiate, but don't hash
1413 * the dentry. */
Al Viro35165862008-08-05 03:00:49 -04001414 if (nd->flags & LOOKUP_EXCL) {
Trond Myklebustd4d9cdc2007-10-02 18:38:53 -04001415 d_instantiate(dentry, NULL);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001416 goto out;
1417 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
Al Viro51141592011-06-22 18:47:28 -04001419 open_flags = nd->intent.open.flags;
1420
1421 ctx = create_nfs_open_context(dentry, open_flags);
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001422 res = ERR_CAST(ctx);
1423 if (IS_ERR(ctx))
1424 goto out;
1425
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001426 if (nd->flags & LOOKUP_CREATE) {
1427 attr.ia_mode = nd->intent.open.create_mode;
1428 attr.ia_valid = ATTR_MODE;
Aneesh Kumar K.Va8a5da92010-12-09 11:35:14 +00001429 attr.ia_mode &= ~current_umask();
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001430 } else {
Trond Myklebust898f6352010-10-23 11:24:25 -04001431 open_flags &= ~(O_EXCL | O_CREAT);
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001432 attr.ia_valid = 0;
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001433 }
1434
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 /* Open the file on the server */
Trond Myklebustf46e0bd2010-09-17 10:56:50 -04001436 nfs_block_sillyrename(dentry->d_parent);
Trond Myklebust2b484292010-09-17 10:56:51 -04001437 inode = NFS_PROTO(dir)->open_context(dir, ctx, open_flags, &attr);
Trond Myklebustf46e0bd2010-09-17 10:56:50 -04001438 if (IS_ERR(inode)) {
1439 nfs_unblock_sillyrename(dentry->d_parent);
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001440 put_nfs_open_context(ctx);
Trond Myklebustf46e0bd2010-09-17 10:56:50 -04001441 switch (PTR_ERR(inode)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 /* Make a negative dentry */
1443 case -ENOENT:
Trond Myklebustf46e0bd2010-09-17 10:56:50 -04001444 d_add(dentry, NULL);
Trond Myklebust02a913a2005-10-18 14:20:17 -07001445 res = NULL;
1446 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 /* This turned out not to be a regular file */
Trond Myklebust6f926b52005-10-18 14:20:18 -07001448 case -ENOTDIR:
1449 goto no_open;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450 case -ELOOP:
1451 if (!(nd->intent.open.flags & O_NOFOLLOW))
1452 goto no_open;
Trond Myklebust23ebbd92010-11-03 10:24:16 -04001453 /* case -EISDIR: */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 /* case -EINVAL: */
1455 default:
Trond Myklebustf46e0bd2010-09-17 10:56:50 -04001456 res = ERR_CAST(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457 goto out;
1458 }
Trond Myklebustcd9a1c02010-09-17 10:56:50 -04001459 }
Trond Myklebustf46e0bd2010-09-17 10:56:50 -04001460 res = d_add_unique(dentry, inode);
Trond Myklebust898f6352010-10-23 11:24:25 -04001461 nfs_unblock_sillyrename(dentry->d_parent);
Trond Myklebustf46e0bd2010-09-17 10:56:50 -04001462 if (res != NULL) {
Al Viro3d4ff432011-06-22 18:40:12 -04001463 dput(ctx->dentry);
1464 ctx->dentry = dget(res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465 dentry = res;
Trond Myklebustf46e0bd2010-09-17 10:56:50 -04001466 }
Trond Myklebust898f6352010-10-23 11:24:25 -04001467 err = nfs_intent_set_file(nd, ctx);
1468 if (err < 0) {
1469 if (res != NULL)
1470 dput(res);
1471 return ERR_PTR(err);
1472 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473out:
Trond Myklebustf46e0bd2010-09-17 10:56:50 -04001474 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001475 return res;
1476no_open:
1477 return nfs_lookup(dir, dentry, nd);
1478}
1479
1480static int nfs_open_revalidate(struct dentry *dentry, struct nameidata *nd)
1481{
1482 struct dentry *parent = NULL;
Nick Piggin657e94b2011-01-14 02:48:39 +00001483 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 struct inode *dir;
Trond Myklebustb8d4cad2010-09-17 10:56:51 -04001485 struct nfs_open_context *ctx;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 int openflags, ret = 0;
1487
Nick Piggin657e94b2011-01-14 02:48:39 +00001488 if (nd->flags & LOOKUP_RCU)
1489 return -ECHILD;
1490
1491 inode = dentry->d_inode;
Trond Myklebust1f063d22010-04-22 15:35:55 -04001492 if (!is_atomic_open(nd) || d_mountpoint(dentry))
Trond Myklebust5584c302008-12-23 15:21:54 -05001493 goto no_open;
Trond Myklebust2b484292010-09-17 10:56:51 -04001494
Linus Torvalds1da177e2005-04-16 15:20:36 -07001495 parent = dget_parent(dentry);
1496 dir = parent->d_inode;
Trond Myklebust2b484292010-09-17 10:56:51 -04001497
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498 /* We can't create new files in nfs_open_revalidate(), so we
1499 * optimize away revalidation of negative dentries.
1500 */
Trond Myklebust216d5d062007-10-01 20:10:12 -04001501 if (inode == NULL) {
1502 if (!nfs_neg_need_reval(dir, dentry, nd))
1503 ret = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 goto out;
Trond Myklebust216d5d062007-10-01 20:10:12 -04001505 }
1506
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 /* NFS only supports OPEN on regular files */
1508 if (!S_ISREG(inode->i_mode))
Trond Myklebust5584c302008-12-23 15:21:54 -05001509 goto no_open_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 openflags = nd->intent.open.flags;
1511 /* We cannot do exclusive creation on a positive dentry */
1512 if ((openflags & (O_CREAT|O_EXCL)) == (O_CREAT|O_EXCL))
Trond Myklebust5584c302008-12-23 15:21:54 -05001513 goto no_open_dput;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 /* We can't create new files, or truncate existing ones here */
Trond Myklebust0a377cf2010-08-18 09:25:42 -04001515 openflags &= ~(O_CREAT|O_EXCL|O_TRUNC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516
Al Viro51141592011-06-22 18:47:28 -04001517 ctx = create_nfs_open_context(dentry, openflags);
Trond Myklebustb8d4cad2010-09-17 10:56:51 -04001518 ret = PTR_ERR(ctx);
1519 if (IS_ERR(ctx))
1520 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521 /*
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001522 * Note: we're not holding inode->i_mutex and so may be racing with
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 * operations that change the directory. We therefore save the
1524 * change attribute *before* we do the RPC call.
1525 */
Trond Myklebust2b484292010-09-17 10:56:51 -04001526 inode = NFS_PROTO(dir)->open_context(dir, ctx, openflags, NULL);
Trond Myklebust535918f2010-09-17 10:56:51 -04001527 if (IS_ERR(inode)) {
1528 ret = PTR_ERR(inode);
1529 switch (ret) {
1530 case -EPERM:
1531 case -EACCES:
1532 case -EDQUOT:
1533 case -ENOSPC:
1534 case -EROFS:
1535 goto out_put_ctx;
1536 default:
1537 goto out_drop;
1538 }
1539 }
1540 iput(inode);
Trond Myklebust898f6352010-10-23 11:24:25 -04001541 if (inode != dentry->d_inode)
Trond Myklebust535918f2010-09-17 10:56:51 -04001542 goto out_drop;
Trond Myklebust898f6352010-10-23 11:24:25 -04001543
1544 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1545 ret = nfs_intent_set_file(nd, ctx);
1546 if (ret >= 0)
1547 ret = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548out:
1549 dput(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 return ret;
Trond Myklebust535918f2010-09-17 10:56:51 -04001551out_drop:
1552 d_drop(dentry);
1553 ret = 0;
1554out_put_ctx:
1555 put_nfs_open_context(ctx);
1556 goto out;
1557
Trond Myklebust5584c302008-12-23 15:21:54 -05001558no_open_dput:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 dput(parent);
Trond Myklebust5584c302008-12-23 15:21:54 -05001560no_open:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001561 return nfs_lookup_revalidate(dentry, nd);
1562}
Trond Myklebustc0204fd2010-09-17 10:56:51 -04001563
1564static int nfs_open_create(struct inode *dir, struct dentry *dentry, int mode,
1565 struct nameidata *nd)
1566{
1567 struct nfs_open_context *ctx = NULL;
1568 struct iattr attr;
1569 int error;
Al Virof7c85862011-06-22 18:53:18 -04001570 int open_flags = O_CREAT|O_EXCL|FMODE_READ;
Trond Myklebustc0204fd2010-09-17 10:56:51 -04001571
1572 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1573 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
1574
1575 attr.ia_mode = mode;
1576 attr.ia_valid = ATTR_MODE;
1577
Al Virof7c85862011-06-22 18:53:18 -04001578 if (nd && (nd->flags & LOOKUP_OPEN) != 0)
Trond Myklebustc0204fd2010-09-17 10:56:51 -04001579 open_flags = nd->intent.open.flags;
1580
Al Virof7c85862011-06-22 18:53:18 -04001581 ctx = create_nfs_open_context(dentry, open_flags);
1582 error = PTR_ERR(ctx);
1583 if (IS_ERR(ctx))
1584 goto out_err_drop;
Trond Myklebustc0204fd2010-09-17 10:56:51 -04001585
1586 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, ctx);
1587 if (error != 0)
1588 goto out_put_ctx;
Al Virof7c85862011-06-22 18:53:18 -04001589 if (nd && (nd->flags & LOOKUP_OPEN) != 0) {
Trond Myklebust898f6352010-10-23 11:24:25 -04001590 error = nfs_intent_set_file(nd, ctx);
1591 if (error < 0)
1592 goto out_err;
Al Virof7c85862011-06-22 18:53:18 -04001593 } else {
1594 put_nfs_open_context(ctx);
Trond Myklebust898f6352010-10-23 11:24:25 -04001595 }
Trond Myklebustc0204fd2010-09-17 10:56:51 -04001596 return 0;
1597out_put_ctx:
Al Virof7c85862011-06-22 18:53:18 -04001598 put_nfs_open_context(ctx);
Trond Myklebust898f6352010-10-23 11:24:25 -04001599out_err_drop:
Trond Myklebustc0204fd2010-09-17 10:56:51 -04001600 d_drop(dentry);
Trond Myklebust898f6352010-10-23 11:24:25 -04001601out_err:
Trond Myklebustc0204fd2010-09-17 10:56:51 -04001602 return error;
1603}
1604
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605#endif /* CONFIG_NFSV4 */
1606
Linus Torvalds1da177e2005-04-16 15:20:36 -07001607/*
1608 * Code common to create, mkdir, and mknod.
1609 */
1610int nfs_instantiate(struct dentry *dentry, struct nfs_fh *fhandle,
1611 struct nfs_fattr *fattr)
1612{
Trond Myklebustfab728e2007-09-29 17:41:33 -04001613 struct dentry *parent = dget_parent(dentry);
1614 struct inode *dir = parent->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001615 struct inode *inode;
1616 int error = -EACCES;
1617
Trond Myklebustfab728e2007-09-29 17:41:33 -04001618 d_drop(dentry);
1619
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620 /* We may have been initialized further down */
1621 if (dentry->d_inode)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001622 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 if (fhandle->size == 0) {
Bryan Schumaker7c513052011-03-24 17:12:24 +00001624 error = NFS_PROTO(dir)->lookup(NFS_SERVER(dir)->client, dir, &dentry->d_name, fhandle, fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625 if (error)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001626 goto out_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 }
Trond Myklebust5724ab32007-10-01 21:51:38 -04001628 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629 if (!(fattr->valid & NFS_ATTR_FATTR)) {
1630 struct nfs_server *server = NFS_SB(dentry->d_sb);
David Howells8fa5c002006-08-22 20:06:12 -04001631 error = server->nfs_client->rpc_ops->getattr(server, fhandle, fattr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 if (error < 0)
Trond Myklebustfab728e2007-09-29 17:41:33 -04001633 goto out_error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001635 inode = nfs_fhget(dentry->d_sb, fhandle, fattr);
Trond Myklebust03f28e32006-03-20 13:44:48 -05001636 error = PTR_ERR(inode);
1637 if (IS_ERR(inode))
Trond Myklebustfab728e2007-09-29 17:41:33 -04001638 goto out_error;
1639 d_add(dentry, inode);
1640out:
1641 dput(parent);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642 return 0;
Trond Myklebustfab728e2007-09-29 17:41:33 -04001643out_error:
1644 nfs_mark_for_revalidate(dir);
1645 dput(parent);
1646 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647}
1648
1649/*
1650 * Following a failed create operation, we drop the dentry rather
1651 * than retain a negative dentry. This avoids a problem in the event
1652 * that the operation succeeded on the server, but an error in the
1653 * reply path made it appear to have failed.
1654 */
1655static int nfs_create(struct inode *dir, struct dentry *dentry, int mode,
1656 struct nameidata *nd)
1657{
1658 struct iattr attr;
1659 int error;
Al Virof7c85862011-06-22 18:53:18 -04001660 int open_flags = O_CREAT|O_EXCL|FMODE_READ;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001662 dfprintk(VFS, "NFS: create(%s/%ld), %s\n",
1663 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
1665 attr.ia_mode = mode;
1666 attr.ia_valid = ATTR_MODE;
1667
Al Virof7c85862011-06-22 18:53:18 -04001668 if (nd && (nd->flags & LOOKUP_OPEN) != 0)
Trond Myklebust8a0eebf2011-01-13 14:15:50 -05001669 open_flags = nd->intent.open.flags;
1670
1671 error = NFS_PROTO(dir)->create(dir, dentry, &attr, open_flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672 if (error != 0)
1673 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 return 0;
1675out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001676 d_drop(dentry);
1677 return error;
1678}
1679
1680/*
1681 * See comments for nfs_proc_create regarding failed operations.
1682 */
1683static int
1684nfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t rdev)
1685{
1686 struct iattr attr;
1687 int status;
1688
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001689 dfprintk(VFS, "NFS: mknod(%s/%ld), %s\n",
1690 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691
1692 if (!new_valid_dev(rdev))
1693 return -EINVAL;
1694
1695 attr.ia_mode = mode;
1696 attr.ia_valid = ATTR_MODE;
1697
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698 status = NFS_PROTO(dir)->mknod(dir, dentry, &attr, rdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 if (status != 0)
1700 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701 return 0;
1702out_err:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 d_drop(dentry);
1704 return status;
1705}
1706
1707/*
1708 * See comments for nfs_proc_create regarding failed operations.
1709 */
1710static int nfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1711{
1712 struct iattr attr;
1713 int error;
1714
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001715 dfprintk(VFS, "NFS: mkdir(%s/%ld), %s\n",
1716 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717
1718 attr.ia_valid = ATTR_MODE;
1719 attr.ia_mode = mode | S_IFDIR;
1720
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 error = NFS_PROTO(dir)->mkdir(dir, dentry, &attr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 if (error != 0)
1723 goto out_err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 return 0;
1725out_err:
1726 d_drop(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 return error;
1728}
1729
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001730static void nfs_dentry_handle_enoent(struct dentry *dentry)
1731{
1732 if (dentry->d_inode != NULL && !d_unhashed(dentry))
1733 d_delete(dentry);
1734}
1735
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736static int nfs_rmdir(struct inode *dir, struct dentry *dentry)
1737{
1738 int error;
1739
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05001740 dfprintk(VFS, "NFS: rmdir(%s/%ld), %s\n",
1741 dir->i_sb->s_id, dir->i_ino, dentry->d_name.name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 error = NFS_PROTO(dir)->rmdir(dir, &dentry->d_name);
1744 /* Ensure the VFS deletes this inode */
1745 if (error == 0 && dentry->d_inode != NULL)
Dave Hansence71ec32006-09-30 23:29:06 -07001746 clear_nlink(dentry->d_inode);
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001747 else if (error == -ENOENT)
1748 nfs_dentry_handle_enoent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749
1750 return error;
1751}
1752
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753/*
1754 * Remove a file after making sure there are no pending writes,
1755 * and after checking that the file has only one user.
1756 *
1757 * We invalidate the attribute cache and free the inode prior to the operation
1758 * to avoid possible races if the server reuses the inode.
1759 */
1760static int nfs_safe_remove(struct dentry *dentry)
1761{
1762 struct inode *dir = dentry->d_parent->d_inode;
1763 struct inode *inode = dentry->d_inode;
1764 int error = -EBUSY;
1765
1766 dfprintk(VFS, "NFS: safe_remove(%s/%s)\n",
1767 dentry->d_parent->d_name.name, dentry->d_name.name);
1768
1769 /* If the dentry was sillyrenamed, we simply call d_delete() */
1770 if (dentry->d_flags & DCACHE_NFSFS_RENAMED) {
1771 error = 0;
1772 goto out;
1773 }
1774
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 if (inode != NULL) {
Trond Myklebustcae7a072005-10-18 14:20:19 -07001776 nfs_inode_return_delegation(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
1778 /* The VFS may want to delete this inode */
1779 if (error == 0)
Trond Myklebust1b83d702008-06-11 15:44:04 -04001780 nfs_drop_nlink(inode);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001781 nfs_mark_for_revalidate(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 } else
1783 error = NFS_PROTO(dir)->remove(dir, &dentry->d_name);
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001784 if (error == -ENOENT)
1785 nfs_dentry_handle_enoent(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786out:
1787 return error;
1788}
1789
1790/* We do silly rename. In case sillyrename() returns -EBUSY, the inode
1791 * belongs to an active ".nfs..." file and we return -EBUSY.
1792 *
1793 * If sillyrename() returns 0, we do nothing, otherwise we unlink.
1794 */
1795static int nfs_unlink(struct inode *dir, struct dentry *dentry)
1796{
1797 int error;
1798 int need_rehash = 0;
1799
1800 dfprintk(VFS, "NFS: unlink(%s/%ld, %s)\n", dir->i_sb->s_id,
1801 dir->i_ino, dentry->d_name.name);
1802
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 spin_lock(&dentry->d_lock);
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001804 if (dentry->d_count > 1) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 spin_unlock(&dentry->d_lock);
Trond Myklebustccfeb502007-01-13 02:28:12 -05001806 /* Start asynchronous writeout of the inode */
1807 write_inode_now(dentry->d_inode, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 error = nfs_sillyrename(dir, dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 return error;
1810 }
1811 if (!d_unhashed(dentry)) {
1812 __d_drop(dentry);
1813 need_rehash = 1;
1814 }
1815 spin_unlock(&dentry->d_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816 error = nfs_safe_remove(dentry);
Trond Myklebustd45b9d82008-01-28 19:43:18 -05001817 if (!error || error == -ENOENT) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 nfs_set_verifier(dentry, nfs_save_change_attribute(dir));
1819 } else if (need_rehash)
1820 d_rehash(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 return error;
1822}
1823
Chuck Lever873101b2006-08-22 20:06:23 -04001824/*
1825 * To create a symbolic link, most file systems instantiate a new inode,
1826 * add a page to it containing the path, then write it out to the disk
1827 * using prepare_write/commit_write.
1828 *
1829 * Unfortunately the NFS client can't create the in-core inode first
1830 * because it needs a file handle to create an in-core inode (see
1831 * fs/nfs/inode.c:nfs_fhget). We only have a file handle *after* the
1832 * symlink request has completed on the server.
1833 *
1834 * So instead we allocate a raw page, copy the symname into it, then do
1835 * the SYMLINK request with the page as the buffer. If it succeeds, we
1836 * now have a new file handle and can instantiate an in-core NFS inode
1837 * and move the raw page into its mapping.
1838 */
1839static int nfs_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840{
Chuck Lever873101b2006-08-22 20:06:23 -04001841 struct pagevec lru_pvec;
1842 struct page *page;
1843 char *kaddr;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 struct iattr attr;
Chuck Lever873101b2006-08-22 20:06:23 -04001845 unsigned int pathlen = strlen(symname);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 int error;
1847
1848 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s)\n", dir->i_sb->s_id,
1849 dir->i_ino, dentry->d_name.name, symname);
1850
Chuck Lever873101b2006-08-22 20:06:23 -04001851 if (pathlen > PAGE_SIZE)
1852 return -ENAMETOOLONG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853
Chuck Lever873101b2006-08-22 20:06:23 -04001854 attr.ia_mode = S_IFLNK | S_IRWXUGO;
1855 attr.ia_valid = ATTR_MODE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856
Jeff Layton83d93f22007-06-07 09:58:08 -04001857 page = alloc_page(GFP_HIGHUSER);
Trond Myklebust76566992008-06-11 15:44:22 -04001858 if (!page)
Chuck Lever873101b2006-08-22 20:06:23 -04001859 return -ENOMEM;
Chuck Lever873101b2006-08-22 20:06:23 -04001860
1861 kaddr = kmap_atomic(page, KM_USER0);
1862 memcpy(kaddr, symname, pathlen);
1863 if (pathlen < PAGE_SIZE)
1864 memset(kaddr + pathlen, 0, PAGE_SIZE - pathlen);
1865 kunmap_atomic(kaddr, KM_USER0);
1866
Chuck Lever94a6d752006-08-22 20:06:23 -04001867 error = NFS_PROTO(dir)->symlink(dir, dentry, page, pathlen, &attr);
Chuck Lever873101b2006-08-22 20:06:23 -04001868 if (error != 0) {
1869 dfprintk(VFS, "NFS: symlink(%s/%ld, %s, %s) error %d\n",
1870 dir->i_sb->s_id, dir->i_ino,
1871 dentry->d_name.name, symname, error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 d_drop(dentry);
Chuck Lever873101b2006-08-22 20:06:23 -04001873 __free_page(page);
Chuck Lever873101b2006-08-22 20:06:23 -04001874 return error;
1875 }
1876
1877 /*
1878 * No big deal if we can't add this page to the page cache here.
1879 * READLINK will get the missing page from the server if needed.
1880 */
1881 pagevec_init(&lru_pvec, 0);
1882 if (!add_to_page_cache(page, dentry->d_inode->i_mapping, 0,
1883 GFP_KERNEL)) {
Chuck Lever39cf8a12006-10-19 23:28:41 -07001884 pagevec_add(&lru_pvec, page);
Rik van Riel4f98a2f2008-10-18 20:26:32 -07001885 pagevec_lru_add_file(&lru_pvec);
Chuck Lever873101b2006-08-22 20:06:23 -04001886 SetPageUptodate(page);
1887 unlock_page(page);
1888 } else
1889 __free_page(page);
1890
Chuck Lever873101b2006-08-22 20:06:23 -04001891 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892}
1893
1894static int
1895nfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1896{
1897 struct inode *inode = old_dentry->d_inode;
1898 int error;
1899
1900 dfprintk(VFS, "NFS: link(%s/%s -> %s/%s)\n",
1901 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1902 dentry->d_parent->d_name.name, dentry->d_name.name);
1903
Trond Myklebust9a3936a2009-10-26 08:09:46 -04001904 nfs_inode_return_delegation(inode);
1905
Trond Myklebust9697d232007-10-02 21:58:05 -04001906 d_drop(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907 error = NFS_PROTO(dir)->link(inode, dir, &dentry->d_name);
Trond Myklebustcf809552005-10-27 22:12:42 -04001908 if (error == 0) {
Al Viro7de9c6ee2010-10-23 11:11:40 -04001909 ihold(inode);
Trond Myklebust9697d232007-10-02 21:58:05 -04001910 d_add(dentry, inode);
Trond Myklebustcf809552005-10-27 22:12:42 -04001911 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912 return error;
1913}
1914
1915/*
1916 * RENAME
1917 * FIXME: Some nfsds, like the Linux user space nfsd, may generate a
1918 * different file handle for the same inode after a rename (e.g. when
1919 * moving to a different directory). A fail-safe method to do so would
1920 * be to look up old_dir/old_name, create a link to new_dir/new_name and
1921 * rename the old file using the sillyrename stuff. This way, the original
1922 * file in old_dir will go away when the last process iput()s the inode.
1923 *
1924 * FIXED.
1925 *
1926 * It actually works quite well. One needs to have the possibility for
1927 * at least one ".nfs..." file in each directory the file ever gets
1928 * moved or linked to which happens automagically with the new
1929 * implementation that only depends on the dcache stuff instead of
1930 * using the inode layer
1931 *
1932 * Unfortunately, things are a little more complicated than indicated
1933 * above. For a cross-directory move, we want to make sure we can get
1934 * rid of the old inode after the operation. This means there must be
1935 * no pending writes (if it's a file), and the use count must be 1.
1936 * If these conditions are met, we can drop the dentries before doing
1937 * the rename.
1938 */
1939static int nfs_rename(struct inode *old_dir, struct dentry *old_dentry,
1940 struct inode *new_dir, struct dentry *new_dentry)
1941{
1942 struct inode *old_inode = old_dentry->d_inode;
1943 struct inode *new_inode = new_dentry->d_inode;
1944 struct dentry *dentry = NULL, *rehash = NULL;
1945 int error = -EBUSY;
1946
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 dfprintk(VFS, "NFS: rename(%s/%s -> %s/%s, ct=%d)\n",
1948 old_dentry->d_parent->d_name.name, old_dentry->d_name.name,
1949 new_dentry->d_parent->d_name.name, new_dentry->d_name.name,
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001950 new_dentry->d_count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951
1952 /*
Miklos Szeredi28f79a12009-12-03 15:58:56 -05001953 * For non-directories, check whether the target is busy and if so,
1954 * make a copy of the dentry and then do a silly-rename. If the
1955 * silly-rename succeeds, the copied dentry is hashed and becomes
1956 * the new target.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 */
Miklos Szeredi27226102009-12-03 15:58:56 -05001958 if (new_inode && !S_ISDIR(new_inode->i_mode)) {
1959 /*
1960 * To prevent any new references to the target during the
1961 * rename, we unhash the dentry in advance.
1962 */
1963 if (!d_unhashed(new_dentry)) {
1964 d_drop(new_dentry);
1965 rehash = new_dentry;
1966 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967
Nick Pigginb7ab39f2011-01-07 17:49:32 +11001968 if (new_dentry->d_count > 2) {
Miklos Szeredi27226102009-12-03 15:58:56 -05001969 int err;
1970
1971 /* copy the target dentry's name */
1972 dentry = d_alloc(new_dentry->d_parent,
1973 &new_dentry->d_name);
1974 if (!dentry)
1975 goto out;
1976
1977 /* silly-rename the existing target ... */
1978 err = nfs_sillyrename(new_dir, new_dentry);
Miklos Szeredi24e93022009-12-03 15:58:56 -05001979 if (err)
Miklos Szeredi27226102009-12-03 15:58:56 -05001980 goto out;
Miklos Szeredi24e93022009-12-03 15:58:56 -05001981
1982 new_dentry = dentry;
OGAWA Hirofumi56335932010-01-06 18:48:26 -05001983 rehash = NULL;
Miklos Szeredi24e93022009-12-03 15:58:56 -05001984 new_inode = NULL;
Miklos Szeredi27226102009-12-03 15:58:56 -05001985 }
Trond Myklebustb1e4adf2009-03-19 15:35:49 -04001986 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001987
Trond Myklebustcae7a072005-10-18 14:20:19 -07001988 nfs_inode_return_delegation(old_inode);
Trond Myklebustb1e4adf2009-03-19 15:35:49 -04001989 if (new_inode != NULL)
Trond Myklebust24174112006-01-03 09:55:33 +01001990 nfs_inode_return_delegation(new_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 error = NFS_PROTO(old_dir)->rename(old_dir, &old_dentry->d_name,
1993 new_dir, &new_dentry->d_name);
Trond Myklebust5ba7cc42005-12-03 15:20:17 -05001994 nfs_mark_for_revalidate(old_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995out:
1996 if (rehash)
1997 d_rehash(rehash);
1998 if (!error) {
Trond Myklebustb1e4adf2009-03-19 15:35:49 -04001999 if (new_inode != NULL)
2000 nfs_drop_nlink(new_inode);
Mark Fasheh349457c2006-09-08 14:22:21 -07002001 d_move(old_dentry, new_dentry);
Chuck Lever8fb559f2007-09-24 15:40:16 -04002002 nfs_set_verifier(new_dentry,
2003 nfs_save_change_attribute(new_dir));
Trond Myklebustd45b9d82008-01-28 19:43:18 -05002004 } else if (error == -ENOENT)
2005 nfs_dentry_handle_enoent(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006
2007 /* new dentry created? */
2008 if (dentry)
2009 dput(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 return error;
2011}
2012
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04002013static DEFINE_SPINLOCK(nfs_access_lru_lock);
2014static LIST_HEAD(nfs_access_lru_list);
2015static atomic_long_t nfs_access_nr_entries;
2016
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002017static void nfs_access_free_entry(struct nfs_access_entry *entry)
2018{
2019 put_rpccred(entry->cred);
2020 kfree(entry);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04002021 smp_mb__before_atomic_dec();
2022 atomic_long_dec(&nfs_access_nr_entries);
2023 smp_mb__after_atomic_dec();
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002024}
2025
Trond Myklebust1a81bb82010-05-13 12:51:06 -04002026static void nfs_access_free_list(struct list_head *head)
2027{
2028 struct nfs_access_entry *cache;
2029
2030 while (!list_empty(head)) {
2031 cache = list_entry(head->next, struct nfs_access_entry, lru);
2032 list_del(&cache->lru);
2033 nfs_access_free_entry(cache);
2034 }
2035}
2036
Ying Han1495f232011-05-24 17:12:27 -07002037int nfs_access_cache_shrinker(struct shrinker *shrink,
2038 struct shrink_control *sc)
Trond Myklebust979df722006-07-25 11:28:19 -04002039{
2040 LIST_HEAD(head);
Trond Myklebustaa510da2010-09-29 15:11:56 -04002041 struct nfs_inode *nfsi, *next;
Trond Myklebust979df722006-07-25 11:28:19 -04002042 struct nfs_access_entry *cache;
Ying Han1495f232011-05-24 17:12:27 -07002043 int nr_to_scan = sc->nr_to_scan;
2044 gfp_t gfp_mask = sc->gfp_mask;
Trond Myklebust979df722006-07-25 11:28:19 -04002045
Trond Myklebust61d5eb22010-05-13 12:51:06 -04002046 if ((gfp_mask & GFP_KERNEL) != GFP_KERNEL)
2047 return (nr_to_scan == 0) ? 0 : -1;
Trond Myklebust9c7e7e22010-05-13 12:51:06 -04002048
Trond Myklebusta50f7952007-06-05 19:23:43 -04002049 spin_lock(&nfs_access_lru_lock);
Trond Myklebustaa510da2010-09-29 15:11:56 -04002050 list_for_each_entry_safe(nfsi, next, &nfs_access_lru_list, access_cache_inode_lru) {
Trond Myklebust979df722006-07-25 11:28:19 -04002051 struct inode *inode;
2052
2053 if (nr_to_scan-- == 0)
2054 break;
Trond Myklebust9c7e7e22010-05-13 12:51:06 -04002055 inode = &nfsi->vfs_inode;
Trond Myklebust979df722006-07-25 11:28:19 -04002056 spin_lock(&inode->i_lock);
2057 if (list_empty(&nfsi->access_cache_entry_lru))
2058 goto remove_lru_entry;
2059 cache = list_entry(nfsi->access_cache_entry_lru.next,
2060 struct nfs_access_entry, lru);
2061 list_move(&cache->lru, &head);
2062 rb_erase(&cache->rb_node, &nfsi->access_cache);
2063 if (!list_empty(&nfsi->access_cache_entry_lru))
2064 list_move_tail(&nfsi->access_cache_inode_lru,
2065 &nfs_access_lru_list);
2066 else {
2067remove_lru_entry:
2068 list_del_init(&nfsi->access_cache_inode_lru);
Trond Myklebust9c7e7e22010-05-13 12:51:06 -04002069 smp_mb__before_clear_bit();
Trond Myklebust979df722006-07-25 11:28:19 -04002070 clear_bit(NFS_INO_ACL_LRU_SET, &nfsi->flags);
Trond Myklebust9c7e7e22010-05-13 12:51:06 -04002071 smp_mb__after_clear_bit();
Trond Myklebust979df722006-07-25 11:28:19 -04002072 }
Trond Myklebust59844a92010-05-26 08:42:24 -04002073 spin_unlock(&inode->i_lock);
Trond Myklebust979df722006-07-25 11:28:19 -04002074 }
2075 spin_unlock(&nfs_access_lru_lock);
Trond Myklebust1a81bb82010-05-13 12:51:06 -04002076 nfs_access_free_list(&head);
Trond Myklebust979df722006-07-25 11:28:19 -04002077 return (atomic_long_read(&nfs_access_nr_entries) / 100) * sysctl_vfs_cache_pressure;
2078}
2079
Trond Myklebust1a81bb82010-05-13 12:51:06 -04002080static void __nfs_access_zap_cache(struct nfs_inode *nfsi, struct list_head *head)
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002081{
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002082 struct rb_root *root_node = &nfsi->access_cache;
Trond Myklebust1a81bb82010-05-13 12:51:06 -04002083 struct rb_node *n;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002084 struct nfs_access_entry *entry;
2085
2086 /* Unhook entries from the cache */
2087 while ((n = rb_first(root_node)) != NULL) {
2088 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2089 rb_erase(n, root_node);
Trond Myklebust1a81bb82010-05-13 12:51:06 -04002090 list_move(&entry->lru, head);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002091 }
2092 nfsi->cache_validity &= ~NFS_INO_INVALID_ACCESS;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002093}
2094
2095void nfs_access_zap_cache(struct inode *inode)
2096{
Trond Myklebust1a81bb82010-05-13 12:51:06 -04002097 LIST_HEAD(head);
2098
2099 if (test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags) == 0)
2100 return;
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04002101 /* Remove from global LRU init */
Trond Myklebust1a81bb82010-05-13 12:51:06 -04002102 spin_lock(&nfs_access_lru_lock);
2103 if (test_and_clear_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04002104 list_del_init(&NFS_I(inode)->access_cache_inode_lru);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04002105
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002106 spin_lock(&inode->i_lock);
Trond Myklebust1a81bb82010-05-13 12:51:06 -04002107 __nfs_access_zap_cache(NFS_I(inode), &head);
2108 spin_unlock(&inode->i_lock);
2109 spin_unlock(&nfs_access_lru_lock);
2110 nfs_access_free_list(&head);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002111}
2112
2113static struct nfs_access_entry *nfs_access_search_rbtree(struct inode *inode, struct rpc_cred *cred)
2114{
2115 struct rb_node *n = NFS_I(inode)->access_cache.rb_node;
2116 struct nfs_access_entry *entry;
2117
2118 while (n != NULL) {
2119 entry = rb_entry(n, struct nfs_access_entry, rb_node);
2120
2121 if (cred < entry->cred)
2122 n = n->rb_left;
2123 else if (cred > entry->cred)
2124 n = n->rb_right;
2125 else
2126 return entry;
2127 }
2128 return NULL;
2129}
2130
Trond Myklebustaf22f942007-08-10 17:45:10 -04002131static int nfs_access_get_cached(struct inode *inode, struct rpc_cred *cred, struct nfs_access_entry *res)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002132{
Chuck Lever55296802005-08-18 11:24:09 -07002133 struct nfs_inode *nfsi = NFS_I(inode);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002134 struct nfs_access_entry *cache;
2135 int err = -ENOENT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002136
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002137 spin_lock(&inode->i_lock);
2138 if (nfsi->cache_validity & NFS_INO_INVALID_ACCESS)
2139 goto out_zap;
2140 cache = nfs_access_search_rbtree(inode, cred);
2141 if (cache == NULL)
2142 goto out;
Trond Myklebustb4d23142010-03-10 15:21:44 -05002143 if (!nfs_have_delegated_attributes(inode) &&
Peter Staubach64672d52008-12-23 15:21:56 -05002144 !time_in_range_open(jiffies, cache->jiffies, cache->jiffies + nfsi->attrtimeo))
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002145 goto out_stale;
2146 res->jiffies = cache->jiffies;
2147 res->cred = cache->cred;
2148 res->mask = cache->mask;
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04002149 list_move_tail(&cache->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002150 err = 0;
2151out:
2152 spin_unlock(&inode->i_lock);
2153 return err;
2154out_stale:
2155 rb_erase(&cache->rb_node, &nfsi->access_cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04002156 list_del(&cache->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002157 spin_unlock(&inode->i_lock);
2158 nfs_access_free_entry(cache);
2159 return -ENOENT;
2160out_zap:
Trond Myklebust1a81bb82010-05-13 12:51:06 -04002161 spin_unlock(&inode->i_lock);
2162 nfs_access_zap_cache(inode);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002163 return -ENOENT;
2164}
2165
2166static void nfs_access_add_rbtree(struct inode *inode, struct nfs_access_entry *set)
2167{
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04002168 struct nfs_inode *nfsi = NFS_I(inode);
2169 struct rb_root *root_node = &nfsi->access_cache;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002170 struct rb_node **p = &root_node->rb_node;
2171 struct rb_node *parent = NULL;
2172 struct nfs_access_entry *entry;
2173
2174 spin_lock(&inode->i_lock);
2175 while (*p != NULL) {
2176 parent = *p;
2177 entry = rb_entry(parent, struct nfs_access_entry, rb_node);
2178
2179 if (set->cred < entry->cred)
2180 p = &parent->rb_left;
2181 else if (set->cred > entry->cred)
2182 p = &parent->rb_right;
2183 else
2184 goto found;
2185 }
2186 rb_link_node(&set->rb_node, parent, p);
2187 rb_insert_color(&set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04002188 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002189 spin_unlock(&inode->i_lock);
2190 return;
2191found:
2192 rb_replace_node(parent, &set->rb_node, root_node);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04002193 list_add_tail(&set->lru, &nfsi->access_cache_entry_lru);
2194 list_del(&entry->lru);
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002195 spin_unlock(&inode->i_lock);
2196 nfs_access_free_entry(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197}
2198
Trond Myklebustaf22f942007-08-10 17:45:10 -04002199static void nfs_access_add_cache(struct inode *inode, struct nfs_access_entry *set)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200{
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002201 struct nfs_access_entry *cache = kmalloc(sizeof(*cache), GFP_KERNEL);
2202 if (cache == NULL)
2203 return;
2204 RB_CLEAR_NODE(&cache->rb_node);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 cache->jiffies = set->jiffies;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002206 cache->cred = get_rpccred(set->cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 cache->mask = set->mask;
Trond Myklebust1c3c07e2006-07-25 11:28:18 -04002208
2209 nfs_access_add_rbtree(inode, cache);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04002210
2211 /* Update accounting */
2212 smp_mb__before_atomic_inc();
2213 atomic_long_inc(&nfs_access_nr_entries);
2214 smp_mb__after_atomic_inc();
2215
2216 /* Add inode to global LRU list */
Trond Myklebust1a81bb82010-05-13 12:51:06 -04002217 if (!test_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags)) {
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04002218 spin_lock(&nfs_access_lru_lock);
Trond Myklebust1a81bb82010-05-13 12:51:06 -04002219 if (!test_and_set_bit(NFS_INO_ACL_LRU_SET, &NFS_I(inode)->flags))
2220 list_add_tail(&NFS_I(inode)->access_cache_inode_lru,
2221 &nfs_access_lru_list);
Trond Myklebustcfcea3e2006-07-25 11:28:18 -04002222 spin_unlock(&nfs_access_lru_lock);
2223 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224}
2225
2226static int nfs_do_access(struct inode *inode, struct rpc_cred *cred, int mask)
2227{
2228 struct nfs_access_entry cache;
2229 int status;
2230
2231 status = nfs_access_get_cached(inode, cred, &cache);
2232 if (status == 0)
2233 goto out;
2234
2235 /* Be clever: ask server to check for all possible rights */
2236 cache.mask = MAY_EXEC | MAY_WRITE | MAY_READ;
2237 cache.cred = cred;
2238 cache.jiffies = jiffies;
2239 status = NFS_PROTO(inode)->access(inode, &cache);
Suresh Jayaramana71ee332009-03-10 20:33:21 -04002240 if (status != 0) {
2241 if (status == -ESTALE) {
2242 nfs_zap_caches(inode);
2243 if (!S_ISDIR(inode->i_mode))
2244 set_bit(NFS_INO_STALE, &NFS_I(inode)->flags);
2245 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 return status;
Suresh Jayaramana71ee332009-03-10 20:33:21 -04002247 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 nfs_access_add_cache(inode, &cache);
2249out:
Al Viroe6305c42008-07-15 21:03:57 -04002250 if ((mask & ~cache.mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251 return 0;
2252 return -EACCES;
2253}
2254
Trond Myklebustaf22f942007-08-10 17:45:10 -04002255static int nfs_open_permission_mask(int openflags)
2256{
2257 int mask = 0;
2258
2259 if (openflags & FMODE_READ)
2260 mask |= MAY_READ;
2261 if (openflags & FMODE_WRITE)
2262 mask |= MAY_WRITE;
2263 if (openflags & FMODE_EXEC)
2264 mask |= MAY_EXEC;
2265 return mask;
2266}
2267
2268int nfs_may_open(struct inode *inode, struct rpc_cred *cred, int openflags)
2269{
2270 return nfs_do_access(inode, cred, nfs_open_permission_mask(openflags));
2271}
2272
Al Viro10556cb2011-06-20 19:28:19 -04002273int nfs_permission(struct inode *inode, int mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274{
2275 struct rpc_cred *cred;
2276 int res = 0;
2277
Al Viro10556cb2011-06-20 19:28:19 -04002278 if (mask & MAY_NOT_BLOCK)
Nick Pigginb74c79e2011-01-07 17:49:58 +11002279 return -ECHILD;
2280
Chuck Lever91d5b472006-03-20 13:44:14 -05002281 nfs_inc_stats(inode, NFSIOS_VFSACCESS);
2282
Al Viroe6305c42008-07-15 21:03:57 -04002283 if ((mask & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284 goto out;
2285 /* Is this sys_access() ? */
Eric Paris9cfcac82010-07-23 11:43:51 -04002286 if (mask & (MAY_ACCESS | MAY_CHDIR))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 goto force_lookup;
2288
2289 switch (inode->i_mode & S_IFMT) {
2290 case S_IFLNK:
2291 goto out;
2292 case S_IFREG:
2293 /* NFSv4 has atomic_open... */
2294 if (nfs_server_capable(inode, NFS_CAP_ATOMIC_OPEN)
Frank Filz7ee2cb72009-05-18 17:41:40 -04002295 && (mask & MAY_OPEN)
2296 && !(mask & MAY_EXEC))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 goto out;
2298 break;
2299 case S_IFDIR:
2300 /*
2301 * Optimize away all write operations, since the server
2302 * will check permissions when we perform the op.
2303 */
2304 if ((mask & MAY_WRITE) && !(mask & MAY_READ))
2305 goto out;
2306 }
2307
2308force_lookup:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 if (!NFS_PROTO(inode)->access)
2310 goto out_notsup;
2311
Trond Myklebust98a8e322008-03-12 12:25:28 -04002312 cred = rpc_lookup_cred();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 if (!IS_ERR(cred)) {
2314 res = nfs_do_access(inode, cred, mask);
2315 put_rpccred(cred);
2316 } else
2317 res = PTR_ERR(cred);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318out:
Miklos Szeredif696a362008-07-31 13:41:58 +02002319 if (!res && (mask & MAY_EXEC) && !execute_ok(inode))
2320 res = -EACCES;
2321
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05002322 dfprintk(VFS, "NFS: permission(%s/%ld), mask=0x%x, res=%d\n",
2323 inode->i_sb->s_id, inode->i_ino, mask, res);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002324 return res;
2325out_notsup:
2326 res = nfs_revalidate_inode(NFS_SERVER(inode), inode);
2327 if (res == 0)
Al Viro2830ba72011-06-20 19:16:29 -04002328 res = generic_permission(inode, mask);
Chuck Lever1e7cb3d2006-03-20 13:44:24 -05002329 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330}
2331
2332/*
2333 * Local variables:
2334 * version-control: t
2335 * kept-new-versions: 5
2336 * End:
2337 */