]>
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 | |
b6b3fdea ED |
35 | /* SLAB cache for file structures */ |
36 | static struct kmem_cache *filp_cachep __read_mostly; | |
37 | ||
529bf6be | 38 | static struct percpu_counter nr_files __cacheline_aligned_in_smp; |
1da177e4 | 39 | |
529bf6be | 40 | static inline void file_free_rcu(struct rcu_head *head) |
1da177e4 | 41 | { |
d76b0d9b DH |
42 | struct file *f = container_of(head, struct file, f_u.fu_rcuhead); |
43 | ||
44 | put_cred(f->f_cred); | |
529bf6be | 45 | kmem_cache_free(filp_cachep, f); |
1da177e4 LT |
46 | } |
47 | ||
529bf6be | 48 | static inline void file_free(struct file *f) |
1da177e4 | 49 | { |
529bf6be | 50 | percpu_counter_dec(&nr_files); |
ad775f5a | 51 | file_check_state(f); |
529bf6be | 52 | call_rcu(&f->f_u.fu_rcuhead, file_free_rcu); |
1da177e4 LT |
53 | } |
54 | ||
529bf6be DS |
55 | /* |
56 | * Return the total number of open files in the system | |
57 | */ | |
58 | static int get_nr_files(void) | |
1da177e4 | 59 | { |
529bf6be | 60 | return percpu_counter_read_positive(&nr_files); |
1da177e4 LT |
61 | } |
62 | ||
529bf6be DS |
63 | /* |
64 | * Return the maximum number of open files in the system | |
65 | */ | |
66 | int get_max_files(void) | |
ab2af1f5 | 67 | { |
529bf6be | 68 | return files_stat.max_files; |
ab2af1f5 | 69 | } |
529bf6be DS |
70 | EXPORT_SYMBOL_GPL(get_max_files); |
71 | ||
72 | /* | |
73 | * Handle nr_files sysctl | |
74 | */ | |
75 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
8d65af78 | 76 | int proc_nr_files(ctl_table *table, int write, |
529bf6be DS |
77 | void __user *buffer, size_t *lenp, loff_t *ppos) |
78 | { | |
79 | files_stat.nr_files = get_nr_files(); | |
8d65af78 | 80 | return proc_dointvec(table, write, buffer, lenp, ppos); |
529bf6be DS |
81 | } |
82 | #else | |
8d65af78 | 83 | int proc_nr_files(ctl_table *table, int write, |
529bf6be DS |
84 | void __user *buffer, size_t *lenp, loff_t *ppos) |
85 | { | |
86 | return -ENOSYS; | |
87 | } | |
88 | #endif | |
ab2af1f5 | 89 | |
1da177e4 LT |
90 | /* Find an unused file structure and return a pointer to it. |
91 | * Returns NULL, if there are no more free file structures or | |
92 | * we run out of memory. | |
430e285e DH |
93 | * |
94 | * Be very careful using this. You are responsible for | |
95 | * getting write access to any mount that you might assign | |
96 | * to this filp, if it is opened for write. If this is not | |
97 | * done, you will imbalance int the mount's writer count | |
98 | * and a warning at __fput() time. | |
1da177e4 LT |
99 | */ |
100 | struct file *get_empty_filp(void) | |
101 | { | |
86a264ab | 102 | const struct cred *cred = current_cred(); |
af4d2ecb | 103 | static int old_max; |
1da177e4 LT |
104 | struct file * f; |
105 | ||
106 | /* | |
107 | * Privileged users can go above max_files | |
108 | */ | |
529bf6be DS |
109 | if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) { |
110 | /* | |
111 | * percpu_counters are inaccurate. Do an expensive check before | |
112 | * we go and fail. | |
113 | */ | |
52d9f3b4 | 114 | if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files) |
529bf6be DS |
115 | goto over; |
116 | } | |
af4d2ecb | 117 | |
4975e45f | 118 | f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL); |
af4d2ecb KK |
119 | if (f == NULL) |
120 | goto fail; | |
121 | ||
529bf6be | 122 | percpu_counter_inc(&nr_files); |
af4d2ecb KK |
123 | if (security_file_alloc(f)) |
124 | goto fail_sec; | |
1da177e4 | 125 | |
5a6b7951 | 126 | INIT_LIST_HEAD(&f->f_u.fu_list); |
516e0cc5 | 127 | atomic_long_set(&f->f_count, 1); |
af4d2ecb | 128 | rwlock_init(&f->f_owner.lock); |
d76b0d9b | 129 | f->f_cred = get_cred(cred); |
68499914 | 130 | spin_lock_init(&f->f_lock); |
5a6b7951 | 131 | eventpoll_init_file(f); |
af4d2ecb | 132 | /* f->f_version: 0 */ |
af4d2ecb KK |
133 | return f; |
134 | ||
135 | over: | |
1da177e4 | 136 | /* Ran out of filps - report that */ |
529bf6be | 137 | if (get_nr_files() > old_max) { |
1da177e4 | 138 | printk(KERN_INFO "VFS: file-max limit %d reached\n", |
529bf6be DS |
139 | get_max_files()); |
140 | old_max = get_nr_files(); | |
1da177e4 | 141 | } |
af4d2ecb KK |
142 | goto fail; |
143 | ||
144 | fail_sec: | |
145 | file_free(f); | |
1da177e4 LT |
146 | fail: |
147 | return NULL; | |
148 | } | |
149 | ||
ce8d2cdf DH |
150 | /** |
151 | * alloc_file - allocate and initialize a 'struct file' | |
152 | * @mnt: the vfsmount on which the file will reside | |
153 | * @dentry: the dentry representing the new file | |
154 | * @mode: the mode with which the new file will be opened | |
155 | * @fop: the 'struct file_operations' for the new file | |
156 | * | |
157 | * Use this instead of get_empty_filp() to get a new | |
158 | * 'struct file'. Do so because of the same initialization | |
159 | * pitfalls reasons listed for init_file(). This is a | |
160 | * preferred interface to using init_file(). | |
161 | * | |
162 | * If all the callers of init_file() are eliminated, its | |
163 | * code should be moved into this function. | |
164 | */ | |
165 | struct file *alloc_file(struct vfsmount *mnt, struct dentry *dentry, | |
aeb5d727 | 166 | fmode_t mode, const struct file_operations *fop) |
ce8d2cdf DH |
167 | { |
168 | struct file *file; | |
ce8d2cdf DH |
169 | |
170 | file = get_empty_filp(); | |
171 | if (!file) | |
172 | return NULL; | |
173 | ||
ce8d2cdf DH |
174 | file->f_path.dentry = dentry; |
175 | file->f_path.mnt = mntget(mnt); | |
176 | file->f_mapping = dentry->d_inode->i_mapping; | |
177 | file->f_mode = mode; | |
178 | file->f_op = fop; | |
4a3fd211 DH |
179 | |
180 | /* | |
181 | * These mounts don't really matter in practice | |
182 | * for r/o bind mounts. They aren't userspace- | |
183 | * visible. We do this for consistency, and so | |
184 | * that we can do debugging checks at __fput() | |
185 | */ | |
186 | if ((mode & FMODE_WRITE) && !special_file(dentry->d_inode->i_mode)) { | |
3d1e4631 | 187 | int error = 0; |
ad775f5a | 188 | file_take_write(file); |
96029c4e | 189 | error = mnt_clone_write(mnt); |
4a3fd211 DH |
190 | WARN_ON(error); |
191 | } | |
3d1e4631 | 192 | return file; |
ce8d2cdf | 193 | } |
ce8d2cdf | 194 | |
fc9b52cd | 195 | void fput(struct file *file) |
1da177e4 | 196 | { |
516e0cc5 | 197 | if (atomic_long_dec_and_test(&file->f_count)) |
1da177e4 LT |
198 | __fput(file); |
199 | } | |
200 | ||
201 | EXPORT_SYMBOL(fput); | |
202 | ||
aceaf78d DH |
203 | /** |
204 | * drop_file_write_access - give up ability to write to a file | |
205 | * @file: the file to which we will stop writing | |
206 | * | |
207 | * This is a central place which will give up the ability | |
208 | * to write to @file, along with access to write through | |
209 | * its vfsmount. | |
210 | */ | |
211 | void drop_file_write_access(struct file *file) | |
212 | { | |
4a3fd211 | 213 | struct vfsmount *mnt = file->f_path.mnt; |
aceaf78d DH |
214 | struct dentry *dentry = file->f_path.dentry; |
215 | struct inode *inode = dentry->d_inode; | |
216 | ||
217 | put_write_access(inode); | |
ad775f5a DH |
218 | |
219 | if (special_file(inode->i_mode)) | |
220 | return; | |
221 | if (file_check_writeable(file) != 0) | |
222 | return; | |
223 | mnt_drop_write(mnt); | |
224 | file_release_write(file); | |
aceaf78d DH |
225 | } |
226 | EXPORT_SYMBOL_GPL(drop_file_write_access); | |
227 | ||
1da177e4 LT |
228 | /* __fput is called from task context when aio completion releases the last |
229 | * last use of a struct file *. Do not use otherwise. | |
230 | */ | |
fc9b52cd | 231 | void __fput(struct file *file) |
1da177e4 | 232 | { |
0f7fc9e4 JJS |
233 | struct dentry *dentry = file->f_path.dentry; |
234 | struct vfsmount *mnt = file->f_path.mnt; | |
1da177e4 LT |
235 | struct inode *inode = dentry->d_inode; |
236 | ||
237 | might_sleep(); | |
0eeca283 RL |
238 | |
239 | fsnotify_close(file); | |
1da177e4 LT |
240 | /* |
241 | * The function eventpoll_release() should be the first called | |
242 | * in the file cleanup chain. | |
243 | */ | |
244 | eventpoll_release(file); | |
245 | locks_remove_flock(file); | |
246 | ||
233e70f4 AV |
247 | if (unlikely(file->f_flags & FASYNC)) { |
248 | if (file->f_op && file->f_op->fasync) | |
249 | file->f_op->fasync(-1, file, 0); | |
250 | } | |
1da177e4 LT |
251 | if (file->f_op && file->f_op->release) |
252 | file->f_op->release(inode, file); | |
253 | security_file_free(file); | |
577c4eb0 | 254 | if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL)) |
1da177e4 LT |
255 | cdev_put(inode->i_cdev); |
256 | fops_put(file->f_op); | |
609d7fa9 | 257 | put_pid(file->f_owner.pid); |
1da177e4 | 258 | file_kill(file); |
aceaf78d DH |
259 | if (file->f_mode & FMODE_WRITE) |
260 | drop_file_write_access(file); | |
0f7fc9e4 JJS |
261 | file->f_path.dentry = NULL; |
262 | file->f_path.mnt = NULL; | |
1da177e4 LT |
263 | file_free(file); |
264 | dput(dentry); | |
265 | mntput(mnt); | |
266 | } | |
267 | ||
fc9b52cd | 268 | struct file *fget(unsigned int fd) |
1da177e4 LT |
269 | { |
270 | struct file *file; | |
271 | struct files_struct *files = current->files; | |
272 | ||
ab2af1f5 | 273 | rcu_read_lock(); |
1da177e4 | 274 | file = fcheck_files(files, fd); |
ab2af1f5 | 275 | if (file) { |
516e0cc5 | 276 | if (!atomic_long_inc_not_zero(&file->f_count)) { |
ab2af1f5 DS |
277 | /* File object ref couldn't be taken */ |
278 | rcu_read_unlock(); | |
279 | return NULL; | |
280 | } | |
281 | } | |
282 | rcu_read_unlock(); | |
283 | ||
1da177e4 LT |
284 | return file; |
285 | } | |
286 | ||
287 | EXPORT_SYMBOL(fget); | |
288 | ||
289 | /* | |
290 | * Lightweight file lookup - no refcnt increment if fd table isn't shared. | |
291 | * You can use this only if it is guranteed that the current task already | |
292 | * holds a refcnt to that file. That check has to be done at fget() only | |
293 | * and a flag is returned to be passed to the corresponding fput_light(). | |
294 | * There must not be a cloning between an fget_light/fput_light pair. | |
295 | */ | |
fc9b52cd | 296 | struct file *fget_light(unsigned int fd, int *fput_needed) |
1da177e4 LT |
297 | { |
298 | struct file *file; | |
299 | struct files_struct *files = current->files; | |
300 | ||
301 | *fput_needed = 0; | |
302 | if (likely((atomic_read(&files->count) == 1))) { | |
303 | file = fcheck_files(files, fd); | |
304 | } else { | |
ab2af1f5 | 305 | rcu_read_lock(); |
1da177e4 LT |
306 | file = fcheck_files(files, fd); |
307 | if (file) { | |
516e0cc5 | 308 | if (atomic_long_inc_not_zero(&file->f_count)) |
ab2af1f5 DS |
309 | *fput_needed = 1; |
310 | else | |
311 | /* Didn't get the reference, someone's freed */ | |
312 | file = NULL; | |
1da177e4 | 313 | } |
ab2af1f5 | 314 | rcu_read_unlock(); |
1da177e4 | 315 | } |
ab2af1f5 | 316 | |
1da177e4 LT |
317 | return file; |
318 | } | |
319 | ||
320 | ||
321 | void put_filp(struct file *file) | |
322 | { | |
516e0cc5 | 323 | if (atomic_long_dec_and_test(&file->f_count)) { |
1da177e4 LT |
324 | security_file_free(file); |
325 | file_kill(file); | |
326 | file_free(file); | |
327 | } | |
328 | } | |
329 | ||
330 | void file_move(struct file *file, struct list_head *list) | |
331 | { | |
332 | if (!list) | |
333 | return; | |
334 | file_list_lock(); | |
2f512016 | 335 | list_move(&file->f_u.fu_list, list); |
1da177e4 LT |
336 | file_list_unlock(); |
337 | } | |
338 | ||
339 | void file_kill(struct file *file) | |
340 | { | |
2f512016 | 341 | if (!list_empty(&file->f_u.fu_list)) { |
1da177e4 | 342 | file_list_lock(); |
2f512016 | 343 | list_del_init(&file->f_u.fu_list); |
1da177e4 LT |
344 | file_list_unlock(); |
345 | } | |
346 | } | |
347 | ||
348 | int fs_may_remount_ro(struct super_block *sb) | |
349 | { | |
cfdaf9e5 | 350 | struct file *file; |
1da177e4 LT |
351 | |
352 | /* Check that no files are currently opened for writing. */ | |
353 | file_list_lock(); | |
cfdaf9e5 | 354 | list_for_each_entry(file, &sb->s_files, f_u.fu_list) { |
0f7fc9e4 | 355 | struct inode *inode = file->f_path.dentry->d_inode; |
1da177e4 LT |
356 | |
357 | /* File with pending delete? */ | |
358 | if (inode->i_nlink == 0) | |
359 | goto too_bad; | |
360 | ||
361 | /* Writeable file? */ | |
362 | if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE)) | |
363 | goto too_bad; | |
364 | } | |
365 | file_list_unlock(); | |
366 | return 1; /* Tis' cool bro. */ | |
367 | too_bad: | |
368 | file_list_unlock(); | |
369 | return 0; | |
370 | } | |
371 | ||
864d7c4c | 372 | /** |
373 | * mark_files_ro - mark all files read-only | |
374 | * @sb: superblock in question | |
375 | * | |
376 | * All files are marked read-only. We don't care about pending | |
377 | * delete files so this should be used in 'force' mode only. | |
378 | */ | |
379 | void mark_files_ro(struct super_block *sb) | |
380 | { | |
381 | struct file *f; | |
382 | ||
383 | retry: | |
384 | file_list_lock(); | |
385 | list_for_each_entry(f, &sb->s_files, f_u.fu_list) { | |
386 | struct vfsmount *mnt; | |
387 | if (!S_ISREG(f->f_path.dentry->d_inode->i_mode)) | |
388 | continue; | |
389 | if (!file_count(f)) | |
390 | continue; | |
391 | if (!(f->f_mode & FMODE_WRITE)) | |
392 | continue; | |
393 | f->f_mode &= ~FMODE_WRITE; | |
394 | if (file_check_writeable(f) != 0) | |
395 | continue; | |
396 | file_release_write(f); | |
397 | mnt = mntget(f->f_path.mnt); | |
398 | file_list_unlock(); | |
399 | /* | |
400 | * This can sleep, so we can't hold | |
401 | * the file_list_lock() spinlock. | |
402 | */ | |
403 | mnt_drop_write(mnt); | |
404 | mntput(mnt); | |
405 | goto retry; | |
406 | } | |
407 | file_list_unlock(); | |
408 | } | |
409 | ||
1da177e4 LT |
410 | void __init files_init(unsigned long mempages) |
411 | { | |
412 | int n; | |
b6b3fdea ED |
413 | |
414 | filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0, | |
415 | SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
416 | ||
417 | /* | |
418 | * One file with associated inode and dcache is very roughly 1K. | |
1da177e4 LT |
419 | * Per default don't use more than 10% of our memory for files. |
420 | */ | |
421 | ||
422 | n = (mempages * (PAGE_SIZE / 1024)) / 10; | |
423 | files_stat.max_files = n; | |
424 | if (files_stat.max_files < NR_FILE) | |
425 | files_stat.max_files = NR_FILE; | |
ab2af1f5 | 426 | files_defer_init(); |
0216bfcf | 427 | percpu_counter_init(&nr_files, 0); |
1da177e4 | 428 | } |