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> | |
6146f0d5 | 16 | #include <linux/ima.h> |
1da177e4 | 17 | #include <linux/eventpoll.h> |
ab2af1f5 | 18 | #include <linux/rcupdate.h> |
1da177e4 | 19 | #include <linux/mount.h> |
16f7e0fe | 20 | #include <linux/capability.h> |
1da177e4 | 21 | #include <linux/cdev.h> |
0eeca283 | 22 | #include <linux/fsnotify.h> |
529bf6be DS |
23 | #include <linux/sysctl.h> |
24 | #include <linux/percpu_counter.h> | |
25 | ||
26 | #include <asm/atomic.h> | |
1da177e4 LT |
27 | |
28 | /* sysctl tunables... */ | |
29 | struct files_stat_struct files_stat = { | |
30 | .max_files = NR_FILE | |
31 | }; | |
32 | ||
1da177e4 | 33 | /* public. Not pretty! */ |
529bf6be | 34 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(files_lock); |
1da177e4 | 35 | |
b6b3fdea ED |
36 | /* SLAB cache for file structures */ |
37 | static struct kmem_cache *filp_cachep __read_mostly; | |
38 | ||
529bf6be | 39 | static struct percpu_counter nr_files __cacheline_aligned_in_smp; |
1da177e4 | 40 | |
529bf6be | 41 | static inline void file_free_rcu(struct rcu_head *head) |
1da177e4 | 42 | { |
d76b0d9b DH |
43 | struct file *f = container_of(head, struct file, f_u.fu_rcuhead); |
44 | ||
45 | put_cred(f->f_cred); | |
529bf6be | 46 | kmem_cache_free(filp_cachep, f); |
1da177e4 LT |
47 | } |
48 | ||
529bf6be | 49 | static inline void file_free(struct file *f) |
1da177e4 | 50 | { |
529bf6be | 51 | percpu_counter_dec(&nr_files); |
ad775f5a | 52 | file_check_state(f); |
529bf6be | 53 | call_rcu(&f->f_u.fu_rcuhead, file_free_rcu); |
1da177e4 LT |
54 | } |
55 | ||
529bf6be DS |
56 | /* |
57 | * Return the total number of open files in the system | |
58 | */ | |
59 | static int get_nr_files(void) | |
1da177e4 | 60 | { |
529bf6be | 61 | return percpu_counter_read_positive(&nr_files); |
1da177e4 LT |
62 | } |
63 | ||
529bf6be DS |
64 | /* |
65 | * Return the maximum number of open files in the system | |
66 | */ | |
67 | int get_max_files(void) | |
ab2af1f5 | 68 | { |
529bf6be | 69 | return files_stat.max_files; |
ab2af1f5 | 70 | } |
529bf6be DS |
71 | EXPORT_SYMBOL_GPL(get_max_files); |
72 | ||
73 | /* | |
74 | * Handle nr_files sysctl | |
75 | */ | |
76 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
77 | int proc_nr_files(ctl_table *table, int write, struct file *filp, | |
78 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
79 | { | |
80 | files_stat.nr_files = get_nr_files(); | |
81 | return proc_dointvec(table, write, filp, buffer, lenp, ppos); | |
82 | } | |
83 | #else | |
84 | int proc_nr_files(ctl_table *table, int write, struct file *filp, | |
85 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
86 | { | |
87 | return -ENOSYS; | |
88 | } | |
89 | #endif | |
ab2af1f5 | 90 | |
1da177e4 LT |
91 | /* Find an unused file structure and return a pointer to it. |
92 | * Returns NULL, if there are no more free file structures or | |
93 | * we run out of memory. | |
430e285e DH |
94 | * |
95 | * Be very careful using this. You are responsible for | |
96 | * getting write access to any mount that you might assign | |
97 | * to this filp, if it is opened for write. If this is not | |
98 | * done, you will imbalance int the mount's writer count | |
99 | * and a warning at __fput() time. | |
1da177e4 LT |
100 | */ |
101 | struct file *get_empty_filp(void) | |
102 | { | |
86a264ab | 103 | const struct cred *cred = current_cred(); |
af4d2ecb | 104 | static int old_max; |
1da177e4 LT |
105 | struct file * f; |
106 | ||
107 | /* | |
108 | * Privileged users can go above max_files | |
109 | */ | |
529bf6be DS |
110 | if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) { |
111 | /* | |
112 | * percpu_counters are inaccurate. Do an expensive check before | |
113 | * we go and fail. | |
114 | */ | |
52d9f3b4 | 115 | if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files) |
529bf6be DS |
116 | goto over; |
117 | } | |
af4d2ecb | 118 | |
4975e45f | 119 | f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL); |
af4d2ecb KK |
120 | if (f == NULL) |
121 | goto fail; | |
122 | ||
529bf6be | 123 | percpu_counter_inc(&nr_files); |
af4d2ecb KK |
124 | if (security_file_alloc(f)) |
125 | goto fail_sec; | |
1da177e4 | 126 | |
5a6b7951 | 127 | INIT_LIST_HEAD(&f->f_u.fu_list); |
516e0cc5 | 128 | atomic_long_set(&f->f_count, 1); |
af4d2ecb | 129 | rwlock_init(&f->f_owner.lock); |
d76b0d9b | 130 | f->f_cred = get_cred(cred); |
68499914 | 131 | spin_lock_init(&f->f_lock); |
5a6b7951 | 132 | eventpoll_init_file(f); |
af4d2ecb | 133 | /* f->f_version: 0 */ |
af4d2ecb KK |
134 | return f; |
135 | ||
136 | over: | |
1da177e4 | 137 | /* Ran out of filps - report that */ |
529bf6be | 138 | if (get_nr_files() > old_max) { |
1da177e4 | 139 | printk(KERN_INFO "VFS: file-max limit %d reached\n", |
529bf6be DS |
140 | get_max_files()); |
141 | old_max = get_nr_files(); | |
1da177e4 | 142 | } |
af4d2ecb KK |
143 | goto fail; |
144 | ||
145 | fail_sec: | |
146 | file_free(f); | |
1da177e4 LT |
147 | fail: |
148 | return NULL; | |
149 | } | |
150 | ||
151 | EXPORT_SYMBOL(get_empty_filp); | |
152 | ||
ce8d2cdf DH |
153 | /** |
154 | * alloc_file - allocate and initialize a 'struct file' | |
155 | * @mnt: the vfsmount on which the file will reside | |
156 | * @dentry: the dentry representing the new file | |
157 | * @mode: the mode with which the new file will be opened | |
158 | * @fop: the 'struct file_operations' for the new file | |
159 | * | |
160 | * Use this instead of get_empty_filp() to get a new | |
161 | * 'struct file'. Do so because of the same initialization | |
162 | * pitfalls reasons listed for init_file(). This is a | |
163 | * preferred interface to using init_file(). | |
164 | * | |
165 | * If all the callers of init_file() are eliminated, its | |
166 | * code should be moved into this function. | |
167 | */ | |
168 | struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry, | |
aeb5d727 | 169 | fmode_t mode, const struct file_operations *fop) |
ce8d2cdf DH |
170 | { |
171 | struct file *file; | |
ce8d2cdf DH |
172 | |
173 | file = get_empty_filp(); | |
174 | if (!file) | |
175 | return NULL; | |
176 | ||
177 | init_file(file, mnt, dentry, mode, fop); | |
178 | return file; | |
179 | } | |
180 | EXPORT_SYMBOL(alloc_file); | |
181 | ||
182 | /** | |
183 | * init_file - initialize a 'struct file' | |
184 | * @file: the already allocated 'struct file' to initialized | |
185 | * @mnt: the vfsmount on which the file resides | |
186 | * @dentry: the dentry representing this file | |
187 | * @mode: the mode the file is opened with | |
188 | * @fop: the 'struct file_operations' for this file | |
189 | * | |
190 | * Use this instead of setting the members directly. Doing so | |
191 | * avoids making mistakes like forgetting the mntget() or | |
192 | * forgetting to take a write on the mnt. | |
193 | * | |
194 | * Note: This is a crappy interface. It is here to make | |
195 | * merging with the existing users of get_empty_filp() | |
196 | * who have complex failure logic easier. All users | |
197 | * of this should be moving to alloc_file(). | |
198 | */ | |
199 | int init_file(struct file *file, struct vfsmount *mnt, struct dentry *dentry, | |
aeb5d727 | 200 | fmode_t mode, const struct file_operations *fop) |
ce8d2cdf DH |
201 | { |
202 | int error = 0; | |
203 | file->f_path.dentry = dentry; | |
204 | file->f_path.mnt = mntget(mnt); | |
205 | file->f_mapping = dentry->d_inode->i_mapping; | |
206 | file->f_mode = mode; | |
207 | file->f_op = fop; | |
4a3fd211 DH |
208 | |
209 | /* | |
210 | * These mounts don't really matter in practice | |
211 | * for r/o bind mounts. They aren't userspace- | |
212 | * visible. We do this for consistency, and so | |
213 | * that we can do debugging checks at __fput() | |
214 | */ | |
215 | if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) { | |
ad775f5a | 216 | file_take_write(file); |
96029c4e | 217 | error = mnt_clone_write(mnt); |
4a3fd211 DH |
218 | WARN_ON(error); |
219 | } | |
ce8d2cdf DH |
220 | return error; |
221 | } | |
222 | EXPORT_SYMBOL(init_file); | |
223 | ||
fc9b52cd | 224 | void fput(struct file *file) |
1da177e4 | 225 | { |
516e0cc5 | 226 | if (atomic_long_dec_and_test(&file->f_count)) |
1da177e4 LT |
227 | __fput(file); |
228 | } | |
229 | ||
230 | EXPORT_SYMBOL(fput); | |
231 | ||
aceaf78d DH |
232 | /** |
233 | * drop_file_write_access - give up ability to write to a file | |
234 | * @file: the file to which we will stop writing | |
235 | * | |
236 | * This is a central place which will give up the ability | |
237 | * to write to @file, along with access to write through | |
238 | * its vfsmount. | |
239 | */ | |
240 | void drop_file_write_access(struct file *file) | |
241 | { | |
4a3fd211 | 242 | struct vfsmount *mnt = file->f_path.mnt; |
aceaf78d DH |
243 | struct dentry *dentry = file->f_path.dentry; |
244 | struct inode *inode = dentry->d_inode; | |
245 | ||
246 | put_write_access(inode); | |
ad775f5a DH |
247 | |
248 | if (special_file(inode->i_mode)) | |
249 | return; | |
250 | if (file_check_writeable(file) != 0) | |
251 | return; | |
252 | mnt_drop_write(mnt); | |
253 | file_release_write(file); | |
aceaf78d DH |
254 | } |
255 | EXPORT_SYMBOL_GPL(drop_file_write_access); | |
256 | ||
1da177e4 LT |
257 | /* __fput is called from task context when aio completion releases the last |
258 | * last use of a struct file *. Do not use otherwise. | |
259 | */ | |
fc9b52cd | 260 | void __fput(struct file *file) |
1da177e4 | 261 | { |
0f7fc9e4 JJS |
262 | struct dentry *dentry = file->f_path.dentry; |
263 | struct vfsmount *mnt = file->f_path.mnt; | |
1da177e4 LT |
264 | struct inode *inode = dentry->d_inode; |
265 | ||
266 | might_sleep(); | |
0eeca283 RL |
267 | |
268 | fsnotify_close(file); | |
1da177e4 LT |
269 | /* |
270 | * The function eventpoll_release() should be the first called | |
271 | * in the file cleanup chain. | |
272 | */ | |
273 | eventpoll_release(file); | |
274 | locks_remove_flock(file); | |
275 | ||
233e70f4 AV |
276 | if (unlikely(file->f_flags & FASYNC)) { |
277 | if (file->f_op && file->f_op->fasync) | |
278 | file->f_op->fasync(-1, file, 0); | |
279 | } | |
1da177e4 LT |
280 | if (file->f_op && file->f_op->release) |
281 | file->f_op->release(inode, file); | |
282 | security_file_free(file); | |
6146f0d5 | 283 | ima_file_free(file); |
577c4eb0 | 284 | if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL)) |
1da177e4 LT |
285 | cdev_put(inode->i_cdev); |
286 | fops_put(file->f_op); | |
609d7fa9 | 287 | put_pid(file->f_owner.pid); |
1da177e4 | 288 | file_kill(file); |
aceaf78d DH |
289 | if (file->f_mode & FMODE_WRITE) |
290 | drop_file_write_access(file); | |
0f7fc9e4 JJS |
291 | file->f_path.dentry = NULL; |
292 | file->f_path.mnt = NULL; | |
1da177e4 LT |
293 | file_free(file); |
294 | dput(dentry); | |
295 | mntput(mnt); | |
296 | } | |
297 | ||
fc9b52cd | 298 | struct file *fget(unsigned int fd) |
1da177e4 LT |
299 | { |
300 | struct file *file; | |
301 | struct files_struct *files = current->files; | |
302 | ||
ab2af1f5 | 303 | rcu_read_lock(); |
1da177e4 | 304 | file = fcheck_files(files, fd); |
ab2af1f5 | 305 | if (file) { |
516e0cc5 | 306 | if (!atomic_long_inc_not_zero(&file->f_count)) { |
ab2af1f5 DS |
307 | /* File object ref couldn't be taken */ |
308 | rcu_read_unlock(); | |
309 | return NULL; | |
310 | } | |
311 | } | |
312 | rcu_read_unlock(); | |
313 | ||
1da177e4 LT |
314 | return file; |
315 | } | |
316 | ||
317 | EXPORT_SYMBOL(fget); | |
318 | ||
319 | /* | |
320 | * Lightweight file lookup - no refcnt increment if fd table isn't shared. | |
321 | * You can use this only if it is guranteed that the current task already | |
322 | * holds a refcnt to that file. That check has to be done at fget() only | |
323 | * and a flag is returned to be passed to the corresponding fput_light(). | |
324 | * There must not be a cloning between an fget_light/fput_light pair. | |
325 | */ | |
fc9b52cd | 326 | struct file *fget_light(unsigned int fd, int *fput_needed) |
1da177e4 LT |
327 | { |
328 | struct file *file; | |
329 | struct files_struct *files = current->files; | |
330 | ||
331 | *fput_needed = 0; | |
332 | if (likely((atomic_read(&files->count) == 1))) { | |
333 | file = fcheck_files(files, fd); | |
334 | } else { | |
ab2af1f5 | 335 | rcu_read_lock(); |
1da177e4 LT |
336 | file = fcheck_files(files, fd); |
337 | if (file) { | |
516e0cc5 | 338 | if (atomic_long_inc_not_zero(&file->f_count)) |
ab2af1f5 DS |
339 | *fput_needed = 1; |
340 | else | |
341 | /* Didn't get the reference, someone's freed */ | |
342 | file = NULL; | |
1da177e4 | 343 | } |
ab2af1f5 | 344 | rcu_read_unlock(); |
1da177e4 | 345 | } |
ab2af1f5 | 346 | |
1da177e4 LT |
347 | return file; |
348 | } | |
349 | ||
350 | ||
351 | void put_filp(struct file *file) | |
352 | { | |
516e0cc5 | 353 | if (atomic_long_dec_and_test(&file->f_count)) { |
1da177e4 LT |
354 | security_file_free(file); |
355 | file_kill(file); | |
356 | file_free(file); | |
357 | } | |
358 | } | |
359 | ||
360 | void file_move(struct file *file, struct list_head *list) | |
361 | { | |
362 | if (!list) | |
363 | return; | |
364 | file_list_lock(); | |
2f512016 | 365 | list_move(&file->f_u.fu_list, list); |
1da177e4 LT |
366 | file_list_unlock(); |
367 | } | |
368 | ||
369 | void file_kill(struct file *file) | |
370 | { | |
2f512016 | 371 | if (!list_empty(&file->f_u.fu_list)) { |
1da177e4 | 372 | file_list_lock(); |
2f512016 | 373 | list_del_init(&file->f_u.fu_list); |
1da177e4 LT |
374 | file_list_unlock(); |
375 | } | |
376 | } | |
377 | ||
378 | int fs_may_remount_ro(struct super_block *sb) | |
379 | { | |
cfdaf9e5 | 380 | struct file *file; |
1da177e4 LT |
381 | |
382 | /* Check that no files are currently opened for writing. */ | |
383 | file_list_lock(); | |
cfdaf9e5 | 384 | list_for_each_entry(file, &sb->s_files, f_u.fu_list) { |
0f7fc9e4 | 385 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
386 | |
387 | /* File with pending delete? */ | |
388 | if (inode->i_nlink == 0) | |
389 | goto too_bad; | |
390 | ||
391 | /* Writeable file? */ | |
392 | if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE)) | |
393 | goto too_bad; | |
394 | } | |
395 | file_list_unlock(); | |
396 | return 1; /* Tis' cool bro. */ | |
397 | too_bad: | |
398 | file_list_unlock(); | |
399 | return 0; | |
400 | } | |
401 | ||
864d7c4c | 402 | /** |
403 | * mark_files_ro - mark all files read-only | |
404 | * @sb: superblock in question | |
405 | * | |
406 | * All files are marked read-only. We don't care about pending | |
407 | * delete files so this should be used in 'force' mode only. | |
408 | */ | |
409 | void mark_files_ro(struct super_block *sb) | |
410 | { | |
411 | struct file *f; | |
412 | ||
413 | retry: | |
414 | file_list_lock(); | |
415 | list_for_each_entry(f, &sb->s_files, f_u.fu_list) { | |
416 | struct vfsmount *mnt; | |
417 | if (!S_ISREG(f->f_path.dentry->d_inode->i_mode)) | |
418 | continue; | |
419 | if (!file_count(f)) | |
420 | continue; | |
421 | if (!(f->f_mode & FMODE_WRITE)) | |
422 | continue; | |
423 | f->f_mode &= ~FMODE_WRITE; | |
424 | if (file_check_writeable(f) != 0) | |
425 | continue; | |
426 | file_release_write(f); | |
427 | mnt = mntget(f->f_path.mnt); | |
428 | file_list_unlock(); | |
429 | /* | |
430 | * This can sleep, so we can't hold | |
431 | * the file_list_lock() spinlock. | |
432 | */ | |
433 | mnt_drop_write(mnt); | |
434 | mntput(mnt); | |
435 | goto retry; | |
436 | } | |
437 | file_list_unlock(); | |
438 | } | |
439 | ||
1da177e4 LT |
440 | void __init files_init(unsigned long mempages) |
441 | { | |
442 | int n; | |
b6b3fdea ED |
443 | |
444 | filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0, | |
445 | SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
446 | ||
447 | /* | |
448 | * One file with associated inode and dcache is very roughly 1K. | |
1da177e4 LT |
449 | * Per default don't use more than 10% of our memory for files. |
450 | */ | |
451 | ||
452 | n = (mempages * (PAGE_SIZE / 1024)) / 10; | |
453 | files_stat.max_files = n; | |
454 | if (files_stat.max_files < NR_FILE) | |
455 | files_stat.max_files = NR_FILE; | |
ab2af1f5 | 456 | files_defer_init(); |
0216bfcf | 457 | percpu_counter_init(&nr_files, 0); |
1da177e4 | 458 | } |