blob: 03d83cb686b12598191d97edd55779315d74dc8a [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/file_table.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
6 */
7
8#include <linux/string.h>
9#include <linux/slab.h>
10#include <linux/file.h>
11#include <linux/init.h>
12#include <linux/module.h>
13#include <linux/smp_lock.h>
14#include <linux/fs.h>
15#include <linux/security.h>
16#include <linux/eventpoll.h>
17#include <linux/mount.h>
18#include <linux/cdev.h>
19
20/* sysctl tunables... */
21struct files_stat_struct files_stat = {
22 .max_files = NR_FILE
23};
24
25EXPORT_SYMBOL(files_stat); /* Needed by unix.o */
26
27/* public. Not pretty! */
28 __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock);
29
30static DEFINE_SPINLOCK(filp_count_lock);
31
32/* slab constructors and destructors are called from arbitrary
33 * context and must be fully threaded - use a local spinlock
34 * to protect files_stat.nr_files
35 */
36void filp_ctor(void * objp, struct kmem_cache_s *cachep, unsigned long cflags)
37{
38 if ((cflags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
39 SLAB_CTOR_CONSTRUCTOR) {
40 unsigned long flags;
41 spin_lock_irqsave(&filp_count_lock, flags);
42 files_stat.nr_files++;
43 spin_unlock_irqrestore(&filp_count_lock, flags);
44 }
45}
46
47void filp_dtor(void * objp, struct kmem_cache_s *cachep, unsigned long dflags)
48{
49 unsigned long flags;
50 spin_lock_irqsave(&filp_count_lock, flags);
51 files_stat.nr_files--;
52 spin_unlock_irqrestore(&filp_count_lock, flags);
53}
54
55static inline void file_free(struct file *f)
56{
57 kmem_cache_free(filp_cachep, f);
58}
59
60/* Find an unused file structure and return a pointer to it.
61 * Returns NULL, if there are no more free file structures or
62 * we run out of memory.
63 */
64struct file *get_empty_filp(void)
65{
66static int old_max;
67 struct file * f;
68
69 /*
70 * Privileged users can go above max_files
71 */
72 if (files_stat.nr_files < files_stat.max_files ||
73 capable(CAP_SYS_ADMIN)) {
74 f = kmem_cache_alloc(filp_cachep, GFP_KERNEL);
75 if (f) {
76 memset(f, 0, sizeof(*f));
77 if (security_file_alloc(f)) {
78 file_free(f);
79 goto fail;
80 }
81 eventpoll_init_file(f);
82 atomic_set(&f->f_count, 1);
83 f->f_uid = current->fsuid;
84 f->f_gid = current->fsgid;
85 rwlock_init(&f->f_owner.lock);
86 /* f->f_version: 0 */
87 INIT_LIST_HEAD(&f->f_list);
88 f->f_maxcount = INT_MAX;
89 return f;
90 }
91 }
92
93 /* Ran out of filps - report that */
94 if (files_stat.max_files >= old_max) {
95 printk(KERN_INFO "VFS: file-max limit %d reached\n",
96 files_stat.max_files);
97 old_max = files_stat.max_files;
98 } else {
99 /* Big problems... */
100 printk(KERN_WARNING "VFS: filp allocation failed\n");
101 }
102fail:
103 return NULL;
104}
105
106EXPORT_SYMBOL(get_empty_filp);
107
108void fastcall fput(struct file *file)
109{
110 if (atomic_dec_and_test(&file->f_count))
111 __fput(file);
112}
113
114EXPORT_SYMBOL(fput);
115
116/* __fput is called from task context when aio completion releases the last
117 * last use of a struct file *. Do not use otherwise.
118 */
119void fastcall __fput(struct file *file)
120{
121 struct dentry *dentry = file->f_dentry;
122 struct vfsmount *mnt = file->f_vfsmnt;
123 struct inode *inode = dentry->d_inode;
124
125 might_sleep();
126 /*
127 * The function eventpoll_release() should be the first called
128 * in the file cleanup chain.
129 */
130 eventpoll_release(file);
131 locks_remove_flock(file);
132
133 if (file->f_op && file->f_op->release)
134 file->f_op->release(inode, file);
135 security_file_free(file);
136 if (unlikely(inode->i_cdev != NULL))
137 cdev_put(inode->i_cdev);
138 fops_put(file->f_op);
139 if (file->f_mode & FMODE_WRITE)
140 put_write_access(inode);
141 file_kill(file);
142 file->f_dentry = NULL;
143 file->f_vfsmnt = NULL;
144 file_free(file);
145 dput(dentry);
146 mntput(mnt);
147}
148
149struct file fastcall *fget(unsigned int fd)
150{
151 struct file *file;
152 struct files_struct *files = current->files;
153
154 spin_lock(&files->file_lock);
155 file = fcheck_files(files, fd);
156 if (file)
157 get_file(file);
158 spin_unlock(&files->file_lock);
159 return file;
160}
161
162EXPORT_SYMBOL(fget);
163
164/*
165 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
166 * You can use this only if it is guranteed that the current task already
167 * holds a refcnt to that file. That check has to be done at fget() only
168 * and a flag is returned to be passed to the corresponding fput_light().
169 * There must not be a cloning between an fget_light/fput_light pair.
170 */
171struct file fastcall *fget_light(unsigned int fd, int *fput_needed)
172{
173 struct file *file;
174 struct files_struct *files = current->files;
175
176 *fput_needed = 0;
177 if (likely((atomic_read(&files->count) == 1))) {
178 file = fcheck_files(files, fd);
179 } else {
180 spin_lock(&files->file_lock);
181 file = fcheck_files(files, fd);
182 if (file) {
183 get_file(file);
184 *fput_needed = 1;
185 }
186 spin_unlock(&files->file_lock);
187 }
188 return file;
189}
190
191
192void put_filp(struct file *file)
193{
194 if (atomic_dec_and_test(&file->f_count)) {
195 security_file_free(file);
196 file_kill(file);
197 file_free(file);
198 }
199}
200
201void file_move(struct file *file, struct list_head *list)
202{
203 if (!list)
204 return;
205 file_list_lock();
206 list_move(&file->f_list, list);
207 file_list_unlock();
208}
209
210void file_kill(struct file *file)
211{
212 if (!list_empty(&file->f_list)) {
213 file_list_lock();
214 list_del_init(&file->f_list);
215 file_list_unlock();
216 }
217}
218
219int fs_may_remount_ro(struct super_block *sb)
220{
221 struct list_head *p;
222
223 /* Check that no files are currently opened for writing. */
224 file_list_lock();
225 list_for_each(p, &sb->s_files) {
226 struct file *file = list_entry(p, struct file, f_list);
227 struct inode *inode = file->f_dentry->d_inode;
228
229 /* File with pending delete? */
230 if (inode->i_nlink == 0)
231 goto too_bad;
232
233 /* Writeable file? */
234 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
235 goto too_bad;
236 }
237 file_list_unlock();
238 return 1; /* Tis' cool bro. */
239too_bad:
240 file_list_unlock();
241 return 0;
242}
243
244void __init files_init(unsigned long mempages)
245{
246 int n;
247 /* One file with associated inode and dcache is very roughly 1K.
248 * Per default don't use more than 10% of our memory for files.
249 */
250
251 n = (mempages * (PAGE_SIZE / 1024)) / 10;
252 files_stat.max_files = n;
253 if (files_stat.max_files < NR_FILE)
254 files_stat.max_files = NR_FILE;
255}