]>
Commit | Line | Data |
---|---|---|
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/fdtable.h> | |
12 | #include <linux/init.h> | |
13 | #include <linux/module.h> | |
14 | #include <linux/fs.h> | |
15 | #include <linux/security.h> | |
16 | #include <linux/eventpoll.h> | |
17 | #include <linux/rcupdate.h> | |
18 | #include <linux/mount.h> | |
19 | #include <linux/capability.h> | |
20 | #include <linux/cdev.h> | |
21 | #include <linux/fsnotify.h> | |
22 | #include <linux/sysctl.h> | |
23 | #include <linux/lglock.h> | |
24 | #include <linux/percpu_counter.h> | |
25 | #include <linux/percpu.h> | |
26 | #include <linux/ima.h> | |
27 | ||
28 | #include <linux/atomic.h> | |
29 | ||
30 | #include "internal.h" | |
31 | ||
32 | /* sysctl tunables... */ | |
33 | struct files_stat_struct files_stat = { | |
34 | .max_files = NR_FILE | |
35 | }; | |
36 | ||
37 | DEFINE_LGLOCK(files_lglock); | |
38 | ||
39 | /* SLAB cache for file structures */ | |
40 | static struct kmem_cache *filp_cachep __read_mostly; | |
41 | ||
42 | static struct percpu_counter nr_files __cacheline_aligned_in_smp; | |
43 | ||
44 | static inline void file_free_rcu(struct rcu_head *head) | |
45 | { | |
46 | struct file *f = container_of(head, struct file, f_u.fu_rcuhead); | |
47 | ||
48 | put_cred(f->f_cred); | |
49 | kmem_cache_free(filp_cachep, f); | |
50 | } | |
51 | ||
52 | static inline void file_free(struct file *f) | |
53 | { | |
54 | percpu_counter_dec(&nr_files); | |
55 | file_check_state(f); | |
56 | call_rcu(&f->f_u.fu_rcuhead, file_free_rcu); | |
57 | } | |
58 | ||
59 | /* | |
60 | * Return the total number of open files in the system | |
61 | */ | |
62 | static long get_nr_files(void) | |
63 | { | |
64 | return percpu_counter_read_positive(&nr_files); | |
65 | } | |
66 | ||
67 | /* | |
68 | * Return the maximum number of open files in the system | |
69 | */ | |
70 | unsigned long get_max_files(void) | |
71 | { | |
72 | return files_stat.max_files; | |
73 | } | |
74 | EXPORT_SYMBOL_GPL(get_max_files); | |
75 | ||
76 | /* | |
77 | * Handle nr_files sysctl | |
78 | */ | |
79 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
80 | int proc_nr_files(ctl_table *table, int write, | |
81 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
82 | { | |
83 | files_stat.nr_files = get_nr_files(); | |
84 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); | |
85 | } | |
86 | #else | |
87 | int proc_nr_files(ctl_table *table, int write, | |
88 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
89 | { | |
90 | return -ENOSYS; | |
91 | } | |
92 | #endif | |
93 | ||
94 | /* Find an unused file structure and return a pointer to it. | |
95 | * Returns NULL, if there are no more free file structures or | |
96 | * we run out of memory. | |
97 | * | |
98 | * Be very careful using this. You are responsible for | |
99 | * getting write access to any mount that you might assign | |
100 | * to this filp, if it is opened for write. If this is not | |
101 | * done, you will imbalance int the mount's writer count | |
102 | * and a warning at __fput() time. | |
103 | */ | |
104 | struct file *get_empty_filp(void) | |
105 | { | |
106 | const struct cred *cred = current_cred(); | |
107 | static long old_max; | |
108 | struct file * f; | |
109 | ||
110 | /* | |
111 | * Privileged users can go above max_files | |
112 | */ | |
113 | if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) { | |
114 | /* | |
115 | * percpu_counters are inaccurate. Do an expensive check before | |
116 | * we go and fail. | |
117 | */ | |
118 | if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files) | |
119 | goto over; | |
120 | } | |
121 | ||
122 | f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL); | |
123 | if (f == NULL) | |
124 | goto fail; | |
125 | ||
126 | percpu_counter_inc(&nr_files); | |
127 | f->f_cred = get_cred(cred); | |
128 | if (security_file_alloc(f)) | |
129 | goto fail_sec; | |
130 | ||
131 | INIT_LIST_HEAD(&f->f_u.fu_list); | |
132 | atomic_long_set(&f->f_count, 1); | |
133 | rwlock_init(&f->f_owner.lock); | |
134 | spin_lock_init(&f->f_lock); | |
135 | eventpoll_init_file(f); | |
136 | /* f->f_version: 0 */ | |
137 | return f; | |
138 | ||
139 | over: | |
140 | /* Ran out of filps - report that */ | |
141 | if (get_nr_files() > old_max) { | |
142 | pr_info("VFS: file-max limit %lu reached\n", get_max_files()); | |
143 | old_max = get_nr_files(); | |
144 | } | |
145 | goto fail; | |
146 | ||
147 | fail_sec: | |
148 | file_free(f); | |
149 | fail: | |
150 | return NULL; | |
151 | } | |
152 | ||
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 path *path, fmode_t mode, | |
169 | const struct file_operations *fop) | |
170 | { | |
171 | struct file *file; | |
172 | ||
173 | file = get_empty_filp(); | |
174 | if (!file) | |
175 | return NULL; | |
176 | ||
177 | file->f_path = *path; | |
178 | file->f_mapping = path->dentry->d_inode->i_mapping; | |
179 | file->f_mode = mode; | |
180 | file->f_op = fop; | |
181 | ||
182 | /* | |
183 | * These mounts don't really matter in practice | |
184 | * for r/o bind mounts. They aren't userspace- | |
185 | * visible. We do this for consistency, and so | |
186 | * that we can do debugging checks at __fput() | |
187 | */ | |
188 | if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) { | |
189 | file_take_write(file); | |
190 | WARN_ON(mnt_clone_write(path->mnt)); | |
191 | } | |
192 | if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) | |
193 | i_readcount_inc(path->dentry->d_inode); | |
194 | return file; | |
195 | } | |
196 | EXPORT_SYMBOL(alloc_file); | |
197 | ||
198 | /** | |
199 | * drop_file_write_access - give up ability to write to a file | |
200 | * @file: the file to which we will stop writing | |
201 | * | |
202 | * This is a central place which will give up the ability | |
203 | * to write to @file, along with access to write through | |
204 | * its vfsmount. | |
205 | */ | |
206 | static void drop_file_write_access(struct file *file) | |
207 | { | |
208 | struct vfsmount *mnt = file->f_path.mnt; | |
209 | struct dentry *dentry = file->f_path.dentry; | |
210 | struct inode *inode = dentry->d_inode; | |
211 | ||
212 | put_write_access(inode); | |
213 | ||
214 | if (special_file(inode->i_mode)) | |
215 | return; | |
216 | if (file_check_writeable(file) != 0) | |
217 | return; | |
218 | mnt_drop_write(mnt); | |
219 | file_release_write(file); | |
220 | } | |
221 | ||
222 | /* the real guts of fput() - releasing the last reference to file | |
223 | */ | |
224 | static void __fput(struct file *file) | |
225 | { | |
226 | struct dentry *dentry = file->f_path.dentry; | |
227 | struct vfsmount *mnt = file->f_path.mnt; | |
228 | struct inode *inode = dentry->d_inode; | |
229 | ||
230 | might_sleep(); | |
231 | ||
232 | fsnotify_close(file); | |
233 | /* | |
234 | * The function eventpoll_release() should be the first called | |
235 | * in the file cleanup chain. | |
236 | */ | |
237 | eventpoll_release(file); | |
238 | locks_remove_flock(file); | |
239 | ||
240 | if (unlikely(file->f_flags & FASYNC)) { | |
241 | if (file->f_op && file->f_op->fasync) | |
242 | file->f_op->fasync(-1, file, 0); | |
243 | } | |
244 | if (file->f_op && file->f_op->release) | |
245 | file->f_op->release(inode, file); | |
246 | security_file_free(file); | |
247 | ima_file_free(file); | |
248 | if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL && | |
249 | !(file->f_mode & FMODE_PATH))) { | |
250 | cdev_put(inode->i_cdev); | |
251 | } | |
252 | fops_put(file->f_op); | |
253 | put_pid(file->f_owner.pid); | |
254 | file_sb_list_del(file); | |
255 | if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) | |
256 | i_readcount_dec(inode); | |
257 | if (file->f_mode & FMODE_WRITE) | |
258 | drop_file_write_access(file); | |
259 | file->f_path.dentry = NULL; | |
260 | file->f_path.mnt = NULL; | |
261 | file_free(file); | |
262 | dput(dentry); | |
263 | mntput(mnt); | |
264 | } | |
265 | ||
266 | void fput(struct file *file) | |
267 | { | |
268 | if (atomic_long_dec_and_test(&file->f_count)) | |
269 | __fput(file); | |
270 | } | |
271 | ||
272 | EXPORT_SYMBOL(fput); | |
273 | ||
274 | struct file *fget(unsigned int fd) | |
275 | { | |
276 | struct file *file; | |
277 | struct files_struct *files = current->files; | |
278 | ||
279 | rcu_read_lock(); | |
280 | file = fcheck_files(files, fd); | |
281 | if (file) { | |
282 | /* File object ref couldn't be taken */ | |
283 | if (file->f_mode & FMODE_PATH || | |
284 | !atomic_long_inc_not_zero(&file->f_count)) | |
285 | file = NULL; | |
286 | } | |
287 | rcu_read_unlock(); | |
288 | ||
289 | return file; | |
290 | } | |
291 | ||
292 | EXPORT_SYMBOL(fget); | |
293 | ||
294 | struct file *fget_raw(unsigned int fd) | |
295 | { | |
296 | struct file *file; | |
297 | struct files_struct *files = current->files; | |
298 | ||
299 | rcu_read_lock(); | |
300 | file = fcheck_files(files, fd); | |
301 | if (file) { | |
302 | /* File object ref couldn't be taken */ | |
303 | if (!atomic_long_inc_not_zero(&file->f_count)) | |
304 | file = NULL; | |
305 | } | |
306 | rcu_read_unlock(); | |
307 | ||
308 | return file; | |
309 | } | |
310 | ||
311 | EXPORT_SYMBOL(fget_raw); | |
312 | ||
313 | /* | |
314 | * Lightweight file lookup - no refcnt increment if fd table isn't shared. | |
315 | * | |
316 | * You can use this instead of fget if you satisfy all of the following | |
317 | * conditions: | |
318 | * 1) You must call fput_light before exiting the syscall and returning control | |
319 | * to userspace (i.e. you cannot remember the returned struct file * after | |
320 | * returning to userspace). | |
321 | * 2) You must not call filp_close on the returned struct file * in between | |
322 | * calls to fget_light and fput_light. | |
323 | * 3) You must not clone the current task in between the calls to fget_light | |
324 | * and fput_light. | |
325 | * | |
326 | * The fput_needed flag returned by fget_light should be passed to the | |
327 | * corresponding fput_light. | |
328 | */ | |
329 | struct file *fget_light(unsigned int fd, int *fput_needed) | |
330 | { | |
331 | struct file *file; | |
332 | struct files_struct *files = current->files; | |
333 | ||
334 | *fput_needed = 0; | |
335 | if (atomic_read(&files->count) == 1) { | |
336 | file = fcheck_files(files, fd); | |
337 | if (file && (file->f_mode & FMODE_PATH)) | |
338 | file = NULL; | |
339 | } else { | |
340 | rcu_read_lock(); | |
341 | file = fcheck_files(files, fd); | |
342 | if (file) { | |
343 | if (!(file->f_mode & FMODE_PATH) && | |
344 | atomic_long_inc_not_zero(&file->f_count)) | |
345 | *fput_needed = 1; | |
346 | else | |
347 | /* Didn't get the reference, someone's freed */ | |
348 | file = NULL; | |
349 | } | |
350 | rcu_read_unlock(); | |
351 | } | |
352 | ||
353 | return file; | |
354 | } | |
355 | ||
356 | struct file *fget_raw_light(unsigned int fd, int *fput_needed) | |
357 | { | |
358 | struct file *file; | |
359 | struct files_struct *files = current->files; | |
360 | ||
361 | *fput_needed = 0; | |
362 | if (atomic_read(&files->count) == 1) { | |
363 | file = fcheck_files(files, fd); | |
364 | } else { | |
365 | rcu_read_lock(); | |
366 | file = fcheck_files(files, fd); | |
367 | if (file) { | |
368 | if (atomic_long_inc_not_zero(&file->f_count)) | |
369 | *fput_needed = 1; | |
370 | else | |
371 | /* Didn't get the reference, someone's freed */ | |
372 | file = NULL; | |
373 | } | |
374 | rcu_read_unlock(); | |
375 | } | |
376 | ||
377 | return file; | |
378 | } | |
379 | ||
380 | void put_filp(struct file *file) | |
381 | { | |
382 | if (atomic_long_dec_and_test(&file->f_count)) { | |
383 | security_file_free(file); | |
384 | file_sb_list_del(file); | |
385 | file_free(file); | |
386 | } | |
387 | } | |
388 | ||
389 | static inline int file_list_cpu(struct file *file) | |
390 | { | |
391 | #ifdef CONFIG_SMP | |
392 | return file->f_sb_list_cpu; | |
393 | #else | |
394 | return smp_processor_id(); | |
395 | #endif | |
396 | } | |
397 | ||
398 | /* helper for file_sb_list_add to reduce ifdefs */ | |
399 | static inline void __file_sb_list_add(struct file *file, struct super_block *sb) | |
400 | { | |
401 | struct list_head *list; | |
402 | #ifdef CONFIG_SMP | |
403 | int cpu; | |
404 | cpu = smp_processor_id(); | |
405 | file->f_sb_list_cpu = cpu; | |
406 | list = per_cpu_ptr(sb->s_files, cpu); | |
407 | #else | |
408 | list = &sb->s_files; | |
409 | #endif | |
410 | list_add(&file->f_u.fu_list, list); | |
411 | } | |
412 | ||
413 | /** | |
414 | * file_sb_list_add - add a file to the sb's file list | |
415 | * @file: file to add | |
416 | * @sb: sb to add it to | |
417 | * | |
418 | * Use this function to associate a file with the superblock of the inode it | |
419 | * refers to. | |
420 | */ | |
421 | void file_sb_list_add(struct file *file, struct super_block *sb) | |
422 | { | |
423 | lg_local_lock(&files_lglock); | |
424 | __file_sb_list_add(file, sb); | |
425 | lg_local_unlock(&files_lglock); | |
426 | } | |
427 | ||
428 | /** | |
429 | * file_sb_list_del - remove a file from the sb's file list | |
430 | * @file: file to remove | |
431 | * @sb: sb to remove it from | |
432 | * | |
433 | * Use this function to remove a file from its superblock. | |
434 | */ | |
435 | void file_sb_list_del(struct file *file) | |
436 | { | |
437 | if (!list_empty(&file->f_u.fu_list)) { | |
438 | lg_local_lock_cpu(&files_lglock, file_list_cpu(file)); | |
439 | list_del_init(&file->f_u.fu_list); | |
440 | lg_local_unlock_cpu(&files_lglock, file_list_cpu(file)); | |
441 | } | |
442 | } | |
443 | ||
444 | #ifdef CONFIG_SMP | |
445 | ||
446 | /* | |
447 | * These macros iterate all files on all CPUs for a given superblock. | |
448 | * files_lglock must be held globally. | |
449 | */ | |
450 | #define do_file_list_for_each_entry(__sb, __file) \ | |
451 | { \ | |
452 | int i; \ | |
453 | for_each_possible_cpu(i) { \ | |
454 | struct list_head *list; \ | |
455 | list = per_cpu_ptr((__sb)->s_files, i); \ | |
456 | list_for_each_entry((__file), list, f_u.fu_list) | |
457 | ||
458 | #define while_file_list_for_each_entry \ | |
459 | } \ | |
460 | } | |
461 | ||
462 | #else | |
463 | ||
464 | #define do_file_list_for_each_entry(__sb, __file) \ | |
465 | { \ | |
466 | struct list_head *list; \ | |
467 | list = &(sb)->s_files; \ | |
468 | list_for_each_entry((__file), list, f_u.fu_list) | |
469 | ||
470 | #define while_file_list_for_each_entry \ | |
471 | } | |
472 | ||
473 | #endif | |
474 | ||
475 | /** | |
476 | * mark_files_ro - mark all files read-only | |
477 | * @sb: superblock in question | |
478 | * | |
479 | * All files are marked read-only. We don't care about pending | |
480 | * delete files so this should be used in 'force' mode only. | |
481 | */ | |
482 | void mark_files_ro(struct super_block *sb) | |
483 | { | |
484 | struct file *f; | |
485 | ||
486 | retry: | |
487 | lg_global_lock(&files_lglock); | |
488 | do_file_list_for_each_entry(sb, f) { | |
489 | struct vfsmount *mnt; | |
490 | if (!S_ISREG(f->f_path.dentry->d_inode->i_mode)) | |
491 | continue; | |
492 | if (!file_count(f)) | |
493 | continue; | |
494 | if (!(f->f_mode & FMODE_WRITE)) | |
495 | continue; | |
496 | spin_lock(&f->f_lock); | |
497 | f->f_mode &= ~FMODE_WRITE; | |
498 | spin_unlock(&f->f_lock); | |
499 | if (file_check_writeable(f) != 0) | |
500 | continue; | |
501 | file_release_write(f); | |
502 | mnt = mntget(f->f_path.mnt); | |
503 | /* This can sleep, so we can't hold the spinlock. */ | |
504 | lg_global_unlock(&files_lglock); | |
505 | mnt_drop_write(mnt); | |
506 | mntput(mnt); | |
507 | goto retry; | |
508 | } while_file_list_for_each_entry; | |
509 | lg_global_unlock(&files_lglock); | |
510 | } | |
511 | ||
512 | void __init files_init(unsigned long mempages) | |
513 | { | |
514 | unsigned long n; | |
515 | ||
516 | filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0, | |
517 | SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
518 | ||
519 | /* | |
520 | * One file with associated inode and dcache is very roughly 1K. | |
521 | * Per default don't use more than 10% of our memory for files. | |
522 | */ | |
523 | ||
524 | n = (mempages * (PAGE_SIZE / 1024)) / 10; | |
525 | files_stat.max_files = max_t(unsigned long, n, NR_FILE); | |
526 | files_defer_init(); | |
527 | lg_lock_init(&files_lglock, "files_lglock"); | |
528 | percpu_counter_init(&nr_files, 0); | |
529 | } |