]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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> | |
1da177e4 LT |
13 | #include <linux/fs.h> |
14 | #include <linux/security.h> | |
15 | #include <linux/eventpoll.h> | |
ab2af1f5 | 16 | #include <linux/rcupdate.h> |
1da177e4 | 17 | #include <linux/mount.h> |
16f7e0fe | 18 | #include <linux/capability.h> |
1da177e4 | 19 | #include <linux/cdev.h> |
0eeca283 | 20 | #include <linux/fsnotify.h> |
529bf6be DS |
21 | #include <linux/sysctl.h> |
22 | #include <linux/percpu_counter.h> | |
23 | ||
24 | #include <asm/atomic.h> | |
1da177e4 LT |
25 | |
26 | /* sysctl tunables... */ | |
27 | struct files_stat_struct files_stat = { | |
28 | .max_files = NR_FILE | |
29 | }; | |
30 | ||
1da177e4 | 31 | /* public. Not pretty! */ |
529bf6be | 32 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock); |
1da177e4 | 33 | |
529bf6be | 34 | static struct percpu_counter nr_files __cacheline_aligned_in_smp; |
1da177e4 | 35 | |
529bf6be | 36 | static inline void file_free_rcu(struct rcu_head *head) |
1da177e4 | 37 | { |
529bf6be DS |
38 | struct file *f = container_of(head, struct file, f_u.fu_rcuhead); |
39 | kmem_cache_free(filp_cachep, f); | |
1da177e4 LT |
40 | } |
41 | ||
529bf6be | 42 | static inline void file_free(struct file *f) |
1da177e4 | 43 | { |
529bf6be DS |
44 | percpu_counter_dec(&nr_files); |
45 | call_rcu(&f->f_u.fu_rcuhead, file_free_rcu); | |
1da177e4 LT |
46 | } |
47 | ||
529bf6be DS |
48 | /* |
49 | * Return the total number of open files in the system | |
50 | */ | |
51 | static int get_nr_files(void) | |
1da177e4 | 52 | { |
529bf6be | 53 | return percpu_counter_read_positive(&nr_files); |
1da177e4 LT |
54 | } |
55 | ||
529bf6be DS |
56 | /* |
57 | * Return the maximum number of open files in the system | |
58 | */ | |
59 | int get_max_files(void) | |
ab2af1f5 | 60 | { |
529bf6be | 61 | return files_stat.max_files; |
ab2af1f5 | 62 | } |
529bf6be DS |
63 | EXPORT_SYMBOL_GPL(get_max_files); |
64 | ||
65 | /* | |
66 | * Handle nr_files sysctl | |
67 | */ | |
68 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
69 | int proc_nr_files(ctl_table *table, int write, struct file *filp, | |
70 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
71 | { | |
72 | files_stat.nr_files = get_nr_files(); | |
73 | return proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
74 | } | |
75 | #else | |
76 | int proc_nr_files(ctl_table *table, int write, struct file *filp, | |
77 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
78 | { | |
79 | return -ENOSYS; | |
80 | } | |
81 | #endif | |
ab2af1f5 | 82 | |
1da177e4 LT |
83 | /* Find an unused file structure and return a pointer to it. |
84 | * Returns NULL, if there are no more free file structures or | |
85 | * we run out of memory. | |
430e285e DH |
86 | * |
87 | * Be very careful using this. You are responsible for | |
88 | * getting write access to any mount that you might assign | |
89 | * to this filp, if it is opened for write. If this is not | |
90 | * done, you will imbalance int the mount's writer count | |
91 | * and a warning at __fput() time. | |
1da177e4 LT |
92 | */ |
93 | struct file *get_empty_filp(void) | |
94 | { | |
5a6b7951 | 95 | struct task_struct *tsk; |
af4d2ecb | 96 | static int old_max; |
1da177e4 LT |
97 | struct file * f; |
98 | ||
99 | /* | |
100 | * Privileged users can go above max_files | |
101 | */ | |
529bf6be DS |
102 | if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) { |
103 | /* | |
104 | * percpu_counters are inaccurate. Do an expensive check before | |
105 | * we go and fail. | |
106 | */ | |
52d9f3b4 | 107 | if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files) |
529bf6be DS |
108 | goto over; |
109 | } | |
af4d2ecb | 110 | |
4975e45f | 111 | f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL); |
af4d2ecb KK |
112 | if (f == NULL) |
113 | goto fail; | |
114 | ||
529bf6be | 115 | percpu_counter_inc(&nr_files); |
af4d2ecb KK |
116 | if (security_file_alloc(f)) |
117 | goto fail_sec; | |
1da177e4 | 118 | |
5a6b7951 BL |
119 | tsk = current; |
120 | INIT_LIST_HEAD(&f->f_u.fu_list); | |
af4d2ecb | 121 | atomic_set(&f->f_count, 1); |
af4d2ecb | 122 | rwlock_init(&f->f_owner.lock); |
5a6b7951 BL |
123 | f->f_uid = tsk->fsuid; |
124 | f->f_gid = tsk->fsgid; | |
125 | eventpoll_init_file(f); | |
af4d2ecb | 126 | /* f->f_version: 0 */ |
af4d2ecb KK |
127 | return f; |
128 | ||
129 | over: | |
1da177e4 | 130 | /* Ran out of filps - report that */ |
529bf6be | 131 | if (get_nr_files() > old_max) { |
1da177e4 | 132 | printk(KERN_INFO "VFS: file-max limit %d reached\n", |
529bf6be DS |
133 | get_max_files()); |
134 | old_max = get_nr_files(); | |
1da177e4 | 135 | } |
af4d2ecb KK |
136 | goto fail; |
137 | ||
138 | fail_sec: | |
139 | file_free(f); | |
1da177e4 LT |
140 | fail: |
141 | return NULL; | |
142 | } | |
143 | ||
144 | EXPORT_SYMBOL(get_empty_filp); | |
145 | ||
ce8d2cdf DH |
146 | /** |
147 | * alloc_file - allocate and initialize a 'struct file' | |
148 | * @mnt: the vfsmount on which the file will reside | |
149 | * @dentry: the dentry representing the new file | |
150 | * @mode: the mode with which the new file will be opened | |
151 | * @fop: the 'struct file_operations' for the new file | |
152 | * | |
153 | * Use this instead of get_empty_filp() to get a new | |
154 | * 'struct file'. Do so because of the same initialization | |
155 | * pitfalls reasons listed for init_file(). This is a | |
156 | * preferred interface to using init_file(). | |
157 | * | |
158 | * If all the callers of init_file() are eliminated, its | |
159 | * code should be moved into this function. | |
160 | */ | |
161 | struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry, | |
162 | mode_t mode, const struct file_operations *fop) | |
163 | { | |
164 | struct file *file; | |
165 | struct path; | |
166 | ||
167 | file = get_empty_filp(); | |
168 | if (!file) | |
169 | return NULL; | |
170 | ||
171 | init_file(file, mnt, dentry, mode, fop); | |
172 | return file; | |
173 | } | |
174 | EXPORT_SYMBOL(alloc_file); | |
175 | ||
176 | /** | |
177 | * init_file - initialize a 'struct file' | |
178 | * @file: the already allocated 'struct file' to initialized | |
179 | * @mnt: the vfsmount on which the file resides | |
180 | * @dentry: the dentry representing this file | |
181 | * @mode: the mode the file is opened with | |
182 | * @fop: the 'struct file_operations' for this file | |
183 | * | |
184 | * Use this instead of setting the members directly. Doing so | |
185 | * avoids making mistakes like forgetting the mntget() or | |
186 | * forgetting to take a write on the mnt. | |
187 | * | |
188 | * Note: This is a crappy interface. It is here to make | |
189 | * merging with the existing users of get_empty_filp() | |
190 | * who have complex failure logic easier. All users | |
191 | * of this should be moving to alloc_file(). | |
192 | */ | |
193 | int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry, | |
194 | mode_t mode, const struct file_operations *fop) | |
195 | { | |
196 | int error = 0; | |
197 | file->f_path.dentry = dentry; | |
198 | file->f_path.mnt = mntget(mnt); | |
199 | file->f_mapping = dentry->d_inode->i_mapping; | |
200 | file->f_mode = mode; | |
201 | file->f_op = fop; | |
4a3fd211 DH |
202 | |
203 | /* | |
204 | * These mounts don't really matter in practice | |
205 | * for r/o bind mounts. They aren't userspace- | |
206 | * visible. We do this for consistency, and so | |
207 | * that we can do debugging checks at __fput() | |
208 | */ | |
209 | if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) { | |
210 | error = mnt_want_write(mnt); | |
211 | WARN_ON(error); | |
212 | } | |
ce8d2cdf DH |
213 | return error; |
214 | } | |
215 | EXPORT_SYMBOL(init_file); | |
216 | ||
fc9b52cd | 217 | void fput(struct file *file) |
1da177e4 | 218 | { |
095975da | 219 | if (atomic_dec_and_test(&file->f_count)) |
1da177e4 LT |
220 | __fput(file); |
221 | } | |
222 | ||
223 | EXPORT_SYMBOL(fput); | |
224 | ||
aceaf78d DH |
225 | /** |
226 | * drop_file_write_access - give up ability to write to a file | |
227 | * @file: the file to which we will stop writing | |
228 | * | |
229 | * This is a central place which will give up the ability | |
230 | * to write to @file, along with access to write through | |
231 | * its vfsmount. | |
232 | */ | |
233 | void drop_file_write_access(struct file *file) | |
234 | { | |
4a3fd211 | 235 | struct vfsmount *mnt = file->f_path.mnt; |
aceaf78d DH |
236 | struct dentry *dentry = file->f_path.dentry; |
237 | struct inode *inode = dentry->d_inode; | |
238 | ||
239 | put_write_access(inode); | |
4a3fd211 DH |
240 | if (!special_file(inode->i_mode)) |
241 | mnt_drop_write(mnt); | |
aceaf78d DH |
242 | } |
243 | EXPORT_SYMBOL_GPL(drop_file_write_access); | |
244 | ||
1da177e4 LT |
245 | /* __fput is called from task context when aio completion releases the last |
246 | * last use of a struct file *. Do not use otherwise. | |
247 | */ | |
fc9b52cd | 248 | void __fput(struct file *file) |
1da177e4 | 249 | { |
0f7fc9e4 JJS |
250 | struct dentry *dentry = file->f_path.dentry; |
251 | struct vfsmount *mnt = file->f_path.mnt; | |
1da177e4 LT |
252 | struct inode *inode = dentry->d_inode; |
253 | ||
254 | might_sleep(); | |
0eeca283 RL |
255 | |
256 | fsnotify_close(file); | |
1da177e4 LT |
257 | /* |
258 | * The function eventpoll_release() should be the first called | |
259 | * in the file cleanup chain. | |
260 | */ | |
261 | eventpoll_release(file); | |
262 | locks_remove_flock(file); | |
263 | ||
264 | if (file->f_op && file->f_op->release) | |
265 | file->f_op->release(inode, file); | |
266 | security_file_free(file); | |
577c4eb0 | 267 | if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL)) |
1da177e4 LT |
268 | cdev_put(inode->i_cdev); |
269 | fops_put(file->f_op); | |
609d7fa9 | 270 | put_pid(file->f_owner.pid); |
1da177e4 | 271 | file_kill(file); |
aceaf78d DH |
272 | if (file->f_mode & FMODE_WRITE) |
273 | drop_file_write_access(file); | |
0f7fc9e4 JJS |
274 | file->f_path.dentry = NULL; |
275 | file->f_path.mnt = NULL; | |
1da177e4 LT |
276 | file_free(file); |
277 | dput(dentry); | |
278 | mntput(mnt); | |
279 | } | |
280 | ||
fc9b52cd | 281 | struct file *fget(unsigned int fd) |
1da177e4 LT |
282 | { |
283 | struct file *file; | |
284 | struct files_struct *files = current->files; | |
285 | ||
ab2af1f5 | 286 | rcu_read_lock(); |
1da177e4 | 287 | file = fcheck_files(files, fd); |
ab2af1f5 | 288 | if (file) { |
095975da | 289 | if (!atomic_inc_not_zero(&file->f_count)) { |
ab2af1f5 DS |
290 | /* File object ref couldn't be taken */ |
291 | rcu_read_unlock(); | |
292 | return NULL; | |
293 | } | |
294 | } | |
295 | rcu_read_unlock(); | |
296 | ||
1da177e4 LT |
297 | return file; |
298 | } | |
299 | ||
300 | EXPORT_SYMBOL(fget); | |
301 | ||
302 | /* | |
303 | * Lightweight file lookup - no refcnt increment if fd table isn't shared. | |
304 | * You can use this only if it is guranteed that the current task already | |
305 | * holds a refcnt to that file. That check has to be done at fget() only | |
306 | * and a flag is returned to be passed to the corresponding fput_light(). | |
307 | * There must not be a cloning between an fget_light/fput_light pair. | |
308 | */ | |
fc9b52cd | 309 | struct file *fget_light(unsigned int fd, int *fput_needed) |
1da177e4 LT |
310 | { |
311 | struct file *file; | |
312 | struct files_struct *files = current->files; | |
313 | ||
314 | *fput_needed = 0; | |
315 | if (likely((atomic_read(&files->count) == 1))) { | |
316 | file = fcheck_files(files, fd); | |
317 | } else { | |
ab2af1f5 | 318 | rcu_read_lock(); |
1da177e4 LT |
319 | file = fcheck_files(files, fd); |
320 | if (file) { | |
095975da | 321 | if (atomic_inc_not_zero(&file->f_count)) |
ab2af1f5 DS |
322 | *fput_needed = 1; |
323 | else | |
324 | /* Didn't get the reference, someone's freed */ | |
325 | file = NULL; | |
1da177e4 | 326 | } |
ab2af1f5 | 327 | rcu_read_unlock(); |
1da177e4 | 328 | } |
ab2af1f5 | 329 | |
1da177e4 LT |
330 | return file; |
331 | } | |
332 | ||
333 | ||
334 | void put_filp(struct file *file) | |
335 | { | |
095975da | 336 | if (atomic_dec_and_test(&file->f_count)) { |
1da177e4 LT |
337 | security_file_free(file); |
338 | file_kill(file); | |
339 | file_free(file); | |
340 | } | |
341 | } | |
342 | ||
343 | void file_move(struct file *file, struct list_head *list) | |
344 | { | |
345 | if (!list) | |
346 | return; | |
347 | file_list_lock(); | |
2f512016 | 348 | list_move(&file->f_u.fu_list, list); |
1da177e4 LT |
349 | file_list_unlock(); |
350 | } | |
351 | ||
352 | void file_kill(struct file *file) | |
353 | { | |
2f512016 | 354 | if (!list_empty(&file->f_u.fu_list)) { |
1da177e4 | 355 | file_list_lock(); |
2f512016 | 356 | list_del_init(&file->f_u.fu_list); |
1da177e4 LT |
357 | file_list_unlock(); |
358 | } | |
359 | } | |
360 | ||
361 | int fs_may_remount_ro(struct super_block *sb) | |
362 | { | |
cfdaf9e5 | 363 | struct file *file; |
1da177e4 LT |
364 | |
365 | /* Check that no files are currently opened for writing. */ | |
366 | file_list_lock(); | |
cfdaf9e5 | 367 | list_for_each_entry(file, &sb->s_files, f_u.fu_list) { |
0f7fc9e4 | 368 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
369 | |
370 | /* File with pending delete? */ | |
371 | if (inode->i_nlink == 0) | |
372 | goto too_bad; | |
373 | ||
374 | /* Writeable file? */ | |
375 | if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE)) | |
376 | goto too_bad; | |
377 | } | |
378 | file_list_unlock(); | |
379 | return 1; /* Tis' cool bro. */ | |
380 | too_bad: | |
381 | file_list_unlock(); | |
382 | return 0; | |
383 | } | |
384 | ||
385 | void __init files_init(unsigned long mempages) | |
386 | { | |
387 | int n; | |
388 | /* One file with associated inode and dcache is very roughly 1K. | |
389 | * Per default don't use more than 10% of our memory for files. | |
390 | */ | |
391 | ||
392 | n = (mempages * (PAGE_SIZE / 1024)) / 10; | |
393 | files_stat.max_files = n; | |
394 | if (files_stat.max_files < NR_FILE) | |
395 | files_stat.max_files = NR_FILE; | |
ab2af1f5 | 396 | files_defer_init(); |
0216bfcf | 397 | percpu_counter_init(&nr_files, 0); |
1da177e4 | 398 | } |