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> | |
9f3acc31 | 11 | #include <linux/fdtable.h> |
1da177e4 LT |
12 | #include <linux/init.h> |
13 | #include <linux/module.h> | |
1da177e4 LT |
14 | #include <linux/fs.h> |
15 | #include <linux/security.h> | |
16 | #include <linux/eventpoll.h> | |
ab2af1f5 | 17 | #include <linux/rcupdate.h> |
1da177e4 | 18 | #include <linux/mount.h> |
16f7e0fe | 19 | #include <linux/capability.h> |
1da177e4 | 20 | #include <linux/cdev.h> |
0eeca283 | 21 | #include <linux/fsnotify.h> |
529bf6be DS |
22 | #include <linux/sysctl.h> |
23 | #include <linux/percpu_counter.h> | |
24 | ||
25 | #include <asm/atomic.h> | |
1da177e4 LT |
26 | |
27 | /* sysctl tunables... */ | |
28 | struct files_stat_struct files_stat = { | |
29 | .max_files = NR_FILE | |
30 | }; | |
31 | ||
1da177e4 | 32 | /* public. Not pretty! */ |
529bf6be | 33 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock); |
1da177e4 | 34 | |
529bf6be | 35 | static struct percpu_counter nr_files __cacheline_aligned_in_smp; |
1da177e4 | 36 | |
529bf6be | 37 | static inline void file_free_rcu(struct rcu_head *head) |
1da177e4 | 38 | { |
529bf6be DS |
39 | struct file *f = container_of(head, struct file, f_u.fu_rcuhead); |
40 | kmem_cache_free(filp_cachep, f); | |
1da177e4 LT |
41 | } |
42 | ||
529bf6be | 43 | static inline void file_free(struct file *f) |
1da177e4 | 44 | { |
529bf6be | 45 | percpu_counter_dec(&nr_files); |
ad775f5a | 46 | file_check_state(f); |
529bf6be | 47 | call_rcu(&f->f_u.fu_rcuhead, file_free_rcu); |
1da177e4 LT |
48 | } |
49 | ||
529bf6be DS |
50 | /* |
51 | * Return the total number of open files in the system | |
52 | */ | |
53 | static int get_nr_files(void) | |
1da177e4 | 54 | { |
529bf6be | 55 | return percpu_counter_read_positive(&nr_files); |
1da177e4 LT |
56 | } |
57 | ||
529bf6be DS |
58 | /* |
59 | * Return the maximum number of open files in the system | |
60 | */ | |
61 | int get_max_files(void) | |
ab2af1f5 | 62 | { |
529bf6be | 63 | return files_stat.max_files; |
ab2af1f5 | 64 | } |
529bf6be DS |
65 | EXPORT_SYMBOL_GPL(get_max_files); |
66 | ||
67 | /* | |
68 | * Handle nr_files sysctl | |
69 | */ | |
70 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
71 | int proc_nr_files(ctl_table *table, int write, struct file *filp, | |
72 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
73 | { | |
74 | files_stat.nr_files = get_nr_files(); | |
75 | return proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
76 | } | |
77 | #else | |
78 | int proc_nr_files(ctl_table *table, int write, struct file *filp, | |
79 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
80 | { | |
81 | return -ENOSYS; | |
82 | } | |
83 | #endif | |
ab2af1f5 | 84 | |
1da177e4 LT |
85 | /* Find an unused file structure and return a pointer to it. |
86 | * Returns NULL, if there are no more free file structures or | |
87 | * we run out of memory. | |
430e285e DH |
88 | * |
89 | * Be very careful using this. You are responsible for | |
90 | * getting write access to any mount that you might assign | |
91 | * to this filp, if it is opened for write. If this is not | |
92 | * done, you will imbalance int the mount's writer count | |
93 | * and a warning at __fput() time. | |
1da177e4 LT |
94 | */ |
95 | struct file *get_empty_filp(void) | |
96 | { | |
86a264ab | 97 | const struct cred *cred = current_cred(); |
af4d2ecb | 98 | static int old_max; |
1da177e4 LT |
99 | struct file * f; |
100 | ||
101 | /* | |
102 | * Privileged users can go above max_files | |
103 | */ | |
529bf6be DS |
104 | if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) { |
105 | /* | |
106 | * percpu_counters are inaccurate. Do an expensive check before | |
107 | * we go and fail. | |
108 | */ | |
52d9f3b4 | 109 | if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files) |
529bf6be DS |
110 | goto over; |
111 | } | |
af4d2ecb | 112 | |
4975e45f | 113 | f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL); |
af4d2ecb KK |
114 | if (f == NULL) |
115 | goto fail; | |
116 | ||
529bf6be | 117 | percpu_counter_inc(&nr_files); |
af4d2ecb KK |
118 | if (security_file_alloc(f)) |
119 | goto fail_sec; | |
1da177e4 | 120 | |
5a6b7951 | 121 | INIT_LIST_HEAD(&f->f_u.fu_list); |
516e0cc5 | 122 | atomic_long_set(&f->f_count, 1); |
af4d2ecb | 123 | rwlock_init(&f->f_owner.lock); |
86a264ab DH |
124 | f->f_uid = cred->fsuid; |
125 | f->f_gid = cred->fsgid; | |
5a6b7951 | 126 | eventpoll_init_file(f); |
af4d2ecb | 127 | /* f->f_version: 0 */ |
af4d2ecb KK |
128 | return f; |
129 | ||
130 | over: | |
1da177e4 | 131 | /* Ran out of filps - report that */ |
529bf6be | 132 | if (get_nr_files() > old_max) { |
1da177e4 | 133 | printk(KERN_INFO "VFS: file-max limit %d reached\n", |
529bf6be DS |
134 | get_max_files()); |
135 | old_max = get_nr_files(); | |
1da177e4 | 136 | } |
af4d2ecb KK |
137 | goto fail; |
138 | ||
139 | fail_sec: | |
140 | file_free(f); | |
1da177e4 LT |
141 | fail: |
142 | return NULL; | |
143 | } | |
144 | ||
145 | EXPORT_SYMBOL(get_empty_filp); | |
146 | ||
ce8d2cdf DH |
147 | /** |
148 | * alloc_file - allocate and initialize a 'struct file' | |
149 | * @mnt: the vfsmount on which the file will reside | |
150 | * @dentry: the dentry representing the new file | |
151 | * @mode: the mode with which the new file will be opened | |
152 | * @fop: the 'struct file_operations' for the new file | |
153 | * | |
154 | * Use this instead of get_empty_filp() to get a new | |
155 | * 'struct file'. Do so because of the same initialization | |
156 | * pitfalls reasons listed for init_file(). This is a | |
157 | * preferred interface to using init_file(). | |
158 | * | |
159 | * If all the callers of init_file() are eliminated, its | |
160 | * code should be moved into this function. | |
161 | */ | |
162 | struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry, | |
aeb5d727 | 163 | fmode_t mode, const struct file_operations *fop) |
ce8d2cdf DH |
164 | { |
165 | struct file *file; | |
166 | struct path; | |
167 | ||
168 | file = get_empty_filp(); | |
169 | if (!file) | |
170 | return NULL; | |
171 | ||
172 | init_file(file, mnt, dentry, mode, fop); | |
173 | return file; | |
174 | } | |
175 | EXPORT_SYMBOL(alloc_file); | |
176 | ||
177 | /** | |
178 | * init_file - initialize a 'struct file' | |
179 | * @file: the already allocated 'struct file' to initialized | |
180 | * @mnt: the vfsmount on which the file resides | |
181 | * @dentry: the dentry representing this file | |
182 | * @mode: the mode the file is opened with | |
183 | * @fop: the 'struct file_operations' for this file | |
184 | * | |
185 | * Use this instead of setting the members directly. Doing so | |
186 | * avoids making mistakes like forgetting the mntget() or | |
187 | * forgetting to take a write on the mnt. | |
188 | * | |
189 | * Note: This is a crappy interface. It is here to make | |
190 | * merging with the existing users of get_empty_filp() | |
191 | * who have complex failure logic easier. All users | |
192 | * of this should be moving to alloc_file(). | |
193 | */ | |
194 | int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry, | |
aeb5d727 | 195 | fmode_t mode, const struct file_operations *fop) |
ce8d2cdf DH |
196 | { |
197 | int error = 0; | |
198 | file->f_path.dentry = dentry; | |
199 | file->f_path.mnt = mntget(mnt); | |
200 | file->f_mapping = dentry->d_inode->i_mapping; | |
201 | file->f_mode = mode; | |
202 | file->f_op = fop; | |
4a3fd211 DH |
203 | |
204 | /* | |
205 | * These mounts don't really matter in practice | |
206 | * for r/o bind mounts. They aren't userspace- | |
207 | * visible. We do this for consistency, and so | |
208 | * that we can do debugging checks at __fput() | |
209 | */ | |
210 | if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) { | |
ad775f5a | 211 | file_take_write(file); |
4a3fd211 DH |
212 | error = mnt_want_write(mnt); |
213 | WARN_ON(error); | |
214 | } | |
ce8d2cdf DH |
215 | return error; |
216 | } | |
217 | EXPORT_SYMBOL(init_file); | |
218 | ||
fc9b52cd | 219 | void fput(struct file *file) |
1da177e4 | 220 | { |
516e0cc5 | 221 | if (atomic_long_dec_and_test(&file->f_count)) |
1da177e4 LT |
222 | __fput(file); |
223 | } | |
224 | ||
225 | EXPORT_SYMBOL(fput); | |
226 | ||
aceaf78d DH |
227 | /** |
228 | * drop_file_write_access - give up ability to write to a file | |
229 | * @file: the file to which we will stop writing | |
230 | * | |
231 | * This is a central place which will give up the ability | |
232 | * to write to @file, along with access to write through | |
233 | * its vfsmount. | |
234 | */ | |
235 | void drop_file_write_access(struct file *file) | |
236 | { | |
4a3fd211 | 237 | struct vfsmount *mnt = file->f_path.mnt; |
aceaf78d DH |
238 | struct dentry *dentry = file->f_path.dentry; |
239 | struct inode *inode = dentry->d_inode; | |
240 | ||
241 | put_write_access(inode); | |
ad775f5a DH |
242 | |
243 | if (special_file(inode->i_mode)) | |
244 | return; | |
245 | if (file_check_writeable(file) != 0) | |
246 | return; | |
247 | mnt_drop_write(mnt); | |
248 | file_release_write(file); | |
aceaf78d DH |
249 | } |
250 | EXPORT_SYMBOL_GPL(drop_file_write_access); | |
251 | ||
1da177e4 LT |
252 | /* __fput is called from task context when aio completion releases the last |
253 | * last use of a struct file *. Do not use otherwise. | |
254 | */ | |
fc9b52cd | 255 | void __fput(struct file *file) |
1da177e4 | 256 | { |
0f7fc9e4 JJS |
257 | struct dentry *dentry = file->f_path.dentry; |
258 | struct vfsmount *mnt = file->f_path.mnt; | |
1da177e4 LT |
259 | struct inode *inode = dentry->d_inode; |
260 | ||
261 | might_sleep(); | |
0eeca283 RL |
262 | |
263 | fsnotify_close(file); | |
1da177e4 LT |
264 | /* |
265 | * The function eventpoll_release() should be the first called | |
266 | * in the file cleanup chain. | |
267 | */ | |
268 | eventpoll_release(file); | |
269 | locks_remove_flock(file); | |
270 | ||
233e70f4 AV |
271 | if (unlikely(file->f_flags & FASYNC)) { |
272 | if (file->f_op && file->f_op->fasync) | |
273 | file->f_op->fasync(-1, file, 0); | |
274 | } | |
1da177e4 LT |
275 | if (file->f_op && file->f_op->release) |
276 | file->f_op->release(inode, file); | |
277 | security_file_free(file); | |
577c4eb0 | 278 | if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL)) |
1da177e4 LT |
279 | cdev_put(inode->i_cdev); |
280 | fops_put(file->f_op); | |
609d7fa9 | 281 | put_pid(file->f_owner.pid); |
1da177e4 | 282 | file_kill(file); |
aceaf78d DH |
283 | if (file->f_mode & FMODE_WRITE) |
284 | drop_file_write_access(file); | |
0f7fc9e4 JJS |
285 | file->f_path.dentry = NULL; |
286 | file->f_path.mnt = NULL; | |
1da177e4 LT |
287 | file_free(file); |
288 | dput(dentry); | |
289 | mntput(mnt); | |
290 | } | |
291 | ||
fc9b52cd | 292 | struct file *fget(unsigned int fd) |
1da177e4 LT |
293 | { |
294 | struct file *file; | |
295 | struct files_struct *files = current->files; | |
296 | ||
ab2af1f5 | 297 | rcu_read_lock(); |
1da177e4 | 298 | file = fcheck_files(files, fd); |
ab2af1f5 | 299 | if (file) { |
516e0cc5 | 300 | if (!atomic_long_inc_not_zero(&file->f_count)) { |
ab2af1f5 DS |
301 | /* File object ref couldn't be taken */ |
302 | rcu_read_unlock(); | |
303 | return NULL; | |
304 | } | |
305 | } | |
306 | rcu_read_unlock(); | |
307 | ||
1da177e4 LT |
308 | return file; |
309 | } | |
310 | ||
311 | EXPORT_SYMBOL(fget); | |
312 | ||
313 | /* | |
314 | * Lightweight file lookup - no refcnt increment if fd table isn't shared. | |
315 | * You can use this only if it is guranteed that the current task already | |
316 | * holds a refcnt to that file. That check has to be done at fget() only | |
317 | * and a flag is returned to be passed to the corresponding fput_light(). | |
318 | * There must not be a cloning between an fget_light/fput_light pair. | |
319 | */ | |
fc9b52cd | 320 | struct file *fget_light(unsigned int fd, int *fput_needed) |
1da177e4 LT |
321 | { |
322 | struct file *file; | |
323 | struct files_struct *files = current->files; | |
324 | ||
325 | *fput_needed = 0; | |
326 | if (likely((atomic_read(&files->count) == 1))) { | |
327 | file = fcheck_files(files, fd); | |
328 | } else { | |
ab2af1f5 | 329 | rcu_read_lock(); |
1da177e4 LT |
330 | file = fcheck_files(files, fd); |
331 | if (file) { | |
516e0cc5 | 332 | if (atomic_long_inc_not_zero(&file->f_count)) |
ab2af1f5 DS |
333 | *fput_needed = 1; |
334 | else | |
335 | /* Didn't get the reference, someone's freed */ | |
336 | file = NULL; | |
1da177e4 | 337 | } |
ab2af1f5 | 338 | rcu_read_unlock(); |
1da177e4 | 339 | } |
ab2af1f5 | 340 | |
1da177e4 LT |
341 | return file; |
342 | } | |
343 | ||
344 | ||
345 | void put_filp(struct file *file) | |
346 | { | |
516e0cc5 | 347 | if (atomic_long_dec_and_test(&file->f_count)) { |
1da177e4 LT |
348 | security_file_free(file); |
349 | file_kill(file); | |
350 | file_free(file); | |
351 | } | |
352 | } | |
353 | ||
354 | void file_move(struct file *file, struct list_head *list) | |
355 | { | |
356 | if (!list) | |
357 | return; | |
358 | file_list_lock(); | |
2f512016 | 359 | list_move(&file->f_u.fu_list, list); |
1da177e4 LT |
360 | file_list_unlock(); |
361 | } | |
362 | ||
363 | void file_kill(struct file *file) | |
364 | { | |
2f512016 | 365 | if (!list_empty(&file->f_u.fu_list)) { |
1da177e4 | 366 | file_list_lock(); |
2f512016 | 367 | list_del_init(&file->f_u.fu_list); |
1da177e4 LT |
368 | file_list_unlock(); |
369 | } | |
370 | } | |
371 | ||
372 | int fs_may_remount_ro(struct super_block *sb) | |
373 | { | |
cfdaf9e5 | 374 | struct file *file; |
1da177e4 LT |
375 | |
376 | /* Check that no files are currently opened for writing. */ | |
377 | file_list_lock(); | |
cfdaf9e5 | 378 | list_for_each_entry(file, &sb->s_files, f_u.fu_list) { |
0f7fc9e4 | 379 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
380 | |
381 | /* File with pending delete? */ | |
382 | if (inode->i_nlink == 0) | |
383 | goto too_bad; | |
384 | ||
385 | /* Writeable file? */ | |
386 | if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE)) | |
387 | goto too_bad; | |
388 | } | |
389 | file_list_unlock(); | |
390 | return 1; /* Tis' cool bro. */ | |
391 | too_bad: | |
392 | file_list_unlock(); | |
393 | return 0; | |
394 | } | |
395 | ||
396 | void __init files_init(unsigned long mempages) | |
397 | { | |
398 | int n; | |
399 | /* One file with associated inode and dcache is very roughly 1K. | |
400 | * Per default don't use more than 10% of our memory for files. | |
401 | */ | |
402 | ||
403 | n = (mempages * (PAGE_SIZE / 1024)) / 10; | |
404 | files_stat.max_files = n; | |
405 | if (files_stat.max_files < NR_FILE) | |
406 | files_stat.max_files = NR_FILE; | |
ab2af1f5 | 407 | files_defer_init(); |
0216bfcf | 408 | percpu_counter_init(&nr_files, 0); |
1da177e4 | 409 | } |