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 | 22 | #include <linux/sysctl.h> |
6416ccb7 | 23 | #include <linux/lglock.h> |
529bf6be | 24 | #include <linux/percpu_counter.h> |
6416ccb7 | 25 | #include <linux/percpu.h> |
4a9d4b02 AV |
26 | #include <linux/hardirq.h> |
27 | #include <linux/task_work.h> | |
0552f879 | 28 | #include <linux/ima.h> |
529bf6be | 29 | |
60063497 | 30 | #include <linux/atomic.h> |
1da177e4 | 31 | |
e81e3f4d EP |
32 | #include "internal.h" |
33 | ||
6fa3eb70 S |
34 | #define FILE_OVER_MAX |
35 | #ifdef FILE_OVER_MAX | |
36 | extern void fd_show_open_files(pid_t pid, struct files_struct *files, struct fdtable *fdt); | |
37 | #endif | |
38 | ||
1da177e4 LT |
39 | /* sysctl tunables... */ |
40 | struct files_stat_struct files_stat = { | |
41 | .max_files = NR_FILE | |
42 | }; | |
43 | ||
b6b3fdea ED |
44 | /* SLAB cache for file structures */ |
45 | static struct kmem_cache *filp_cachep __read_mostly; | |
46 | ||
529bf6be | 47 | static struct percpu_counter nr_files __cacheline_aligned_in_smp; |
1da177e4 | 48 | |
5c33b183 | 49 | static void file_free_rcu(struct rcu_head *head) |
1da177e4 | 50 | { |
d76b0d9b DH |
51 | struct file *f = container_of(head, struct file, f_u.fu_rcuhead); |
52 | ||
53 | put_cred(f->f_cred); | |
529bf6be | 54 | kmem_cache_free(filp_cachep, f); |
1da177e4 LT |
55 | } |
56 | ||
529bf6be | 57 | static inline void file_free(struct file *f) |
1da177e4 | 58 | { |
529bf6be | 59 | percpu_counter_dec(&nr_files); |
ad775f5a | 60 | file_check_state(f); |
529bf6be | 61 | call_rcu(&f->f_u.fu_rcuhead, file_free_rcu); |
1da177e4 LT |
62 | } |
63 | ||
529bf6be DS |
64 | /* |
65 | * Return the total number of open files in the system | |
66 | */ | |
518de9b3 | 67 | static long get_nr_files(void) |
1da177e4 | 68 | { |
529bf6be | 69 | return percpu_counter_read_positive(&nr_files); |
1da177e4 LT |
70 | } |
71 | ||
529bf6be DS |
72 | /* |
73 | * Return the maximum number of open files in the system | |
74 | */ | |
518de9b3 | 75 | unsigned long get_max_files(void) |
ab2af1f5 | 76 | { |
529bf6be | 77 | return files_stat.max_files; |
ab2af1f5 | 78 | } |
529bf6be DS |
79 | EXPORT_SYMBOL_GPL(get_max_files); |
80 | ||
81 | /* | |
82 | * Handle nr_files sysctl | |
83 | */ | |
84 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
8d65af78 | 85 | int proc_nr_files(ctl_table *table, int write, |
529bf6be DS |
86 | void __user *buffer, size_t *lenp, loff_t *ppos) |
87 | { | |
88 | files_stat.nr_files = get_nr_files(); | |
518de9b3 | 89 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
529bf6be DS |
90 | } |
91 | #else | |
8d65af78 | 92 | int proc_nr_files(ctl_table *table, int write, |
529bf6be DS |
93 | void __user *buffer, size_t *lenp, loff_t *ppos) |
94 | { | |
95 | return -ENOSYS; | |
96 | } | |
97 | #endif | |
ab2af1f5 | 98 | |
1da177e4 | 99 | /* Find an unused file structure and return a pointer to it. |
1afc99be AV |
100 | * Returns an error pointer if some error happend e.g. we over file |
101 | * structures limit, run out of memory or operation is not permitted. | |
430e285e DH |
102 | * |
103 | * Be very careful using this. You are responsible for | |
104 | * getting write access to any mount that you might assign | |
105 | * to this filp, if it is opened for write. If this is not | |
106 | * done, you will imbalance int the mount's writer count | |
107 | * and a warning at __fput() time. | |
1da177e4 LT |
108 | */ |
109 | struct file *get_empty_filp(void) | |
110 | { | |
86a264ab | 111 | const struct cred *cred = current_cred(); |
518de9b3 | 112 | static long old_max; |
1afc99be AV |
113 | struct file *f; |
114 | int error; | |
1da177e4 LT |
115 | |
116 | /* | |
117 | * Privileged users can go above max_files | |
118 | */ | |
529bf6be DS |
119 | if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) { |
120 | /* | |
121 | * percpu_counters are inaccurate. Do an expensive check before | |
122 | * we go and fail. | |
123 | */ | |
52d9f3b4 | 124 | if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files) |
529bf6be DS |
125 | goto over; |
126 | } | |
af4d2ecb | 127 | |
4975e45f | 128 | f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL); |
1afc99be AV |
129 | if (unlikely(!f)) |
130 | return ERR_PTR(-ENOMEM); | |
af4d2ecb | 131 | |
529bf6be | 132 | percpu_counter_inc(&nr_files); |
78d29788 | 133 | f->f_cred = get_cred(cred); |
1afc99be AV |
134 | error = security_file_alloc(f); |
135 | if (unlikely(error)) { | |
136 | file_free(f); | |
137 | return ERR_PTR(error); | |
138 | } | |
1da177e4 | 139 | |
516e0cc5 | 140 | atomic_long_set(&f->f_count, 1); |
af4d2ecb | 141 | rwlock_init(&f->f_owner.lock); |
68499914 | 142 | spin_lock_init(&f->f_lock); |
5a6b7951 | 143 | eventpoll_init_file(f); |
af4d2ecb | 144 | /* f->f_version: 0 */ |
af4d2ecb KK |
145 | return f; |
146 | ||
147 | over: | |
1da177e4 | 148 | /* Ran out of filps - report that */ |
529bf6be | 149 | if (get_nr_files() > old_max) { |
6fa3eb70 S |
150 | #ifdef FILE_OVER_MAX |
151 | static int fd_dump_all_files = 0; | |
152 | if(!fd_dump_all_files) { | |
153 | struct task_struct *p; | |
154 | pr_debug("[FD_LEAK](PID:%d)files %ld over old_max:%ld", current->pid, get_nr_files(), old_max); | |
155 | for_each_process(p) { | |
156 | pid_t pid = p->pid; | |
157 | struct files_struct *files = p->files; | |
158 | struct fdtable *fdt = files_fdtable(files); | |
159 | if(files && fdt) { | |
160 | fd_show_open_files(pid, files, fdt); | |
161 | } | |
162 | } | |
163 | fd_dump_all_files = 0x1; | |
164 | } | |
165 | #endif | |
518de9b3 | 166 | pr_info("VFS: file-max limit %lu reached\n", get_max_files()); |
529bf6be | 167 | old_max = get_nr_files(); |
1da177e4 | 168 | } |
1afc99be | 169 | return ERR_PTR(-ENFILE); |
1da177e4 LT |
170 | } |
171 | ||
ce8d2cdf DH |
172 | /** |
173 | * alloc_file - allocate and initialize a 'struct file' | |
174 | * @mnt: the vfsmount on which the file will reside | |
175 | * @dentry: the dentry representing the new file | |
176 | * @mode: the mode with which the new file will be opened | |
177 | * @fop: the 'struct file_operations' for the new file | |
178 | * | |
179 | * Use this instead of get_empty_filp() to get a new | |
180 | * 'struct file'. Do so because of the same initialization | |
181 | * pitfalls reasons listed for init_file(). This is a | |
182 | * preferred interface to using init_file(). | |
183 | * | |
184 | * If all the callers of init_file() are eliminated, its | |
185 | * code should be moved into this function. | |
186 | */ | |
2c48b9c4 AV |
187 | struct file *alloc_file(struct path *path, fmode_t mode, |
188 | const struct file_operations *fop) | |
ce8d2cdf DH |
189 | { |
190 | struct file *file; | |
ce8d2cdf DH |
191 | |
192 | file = get_empty_filp(); | |
1afc99be | 193 | if (IS_ERR(file)) |
39b65252 | 194 | return file; |
ce8d2cdf | 195 | |
2c48b9c4 | 196 | file->f_path = *path; |
dd37978c | 197 | file->f_inode = path->dentry->d_inode; |
2c48b9c4 | 198 | file->f_mapping = path->dentry->d_inode->i_mapping; |
ce8d2cdf DH |
199 | file->f_mode = mode; |
200 | file->f_op = fop; | |
4a3fd211 DH |
201 | |
202 | /* | |
203 | * These mounts don't really matter in practice | |
204 | * for r/o bind mounts. They aren't userspace- | |
205 | * visible. We do this for consistency, and so | |
206 | * that we can do debugging checks at __fput() | |
207 | */ | |
2c48b9c4 | 208 | if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) { |
ad775f5a | 209 | file_take_write(file); |
385e3ed4 | 210 | WARN_ON(mnt_clone_write(path->mnt)); |
4a3fd211 | 211 | } |
890275b5 MZ |
212 | if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) |
213 | i_readcount_inc(path->dentry->d_inode); | |
3d1e4631 | 214 | return file; |
ce8d2cdf | 215 | } |
73efc468 | 216 | EXPORT_SYMBOL(alloc_file); |
ce8d2cdf | 217 | |
aceaf78d DH |
218 | /** |
219 | * drop_file_write_access - give up ability to write to a file | |
220 | * @file: the file to which we will stop writing | |
221 | * | |
222 | * This is a central place which will give up the ability | |
223 | * to write to @file, along with access to write through | |
224 | * its vfsmount. | |
225 | */ | |
b57ce969 | 226 | static void drop_file_write_access(struct file *file) |
aceaf78d | 227 | { |
4a3fd211 | 228 | struct vfsmount *mnt = file->f_path.mnt; |
aceaf78d DH |
229 | struct dentry *dentry = file->f_path.dentry; |
230 | struct inode *inode = dentry->d_inode; | |
231 | ||
ad775f5a DH |
232 | if (special_file(inode->i_mode)) |
233 | return; | |
68ad89c7 AV |
234 | |
235 | put_write_access(inode); | |
ad775f5a DH |
236 | if (file_check_writeable(file) != 0) |
237 | return; | |
eb04c282 | 238 | __mnt_drop_write(mnt); |
ad775f5a | 239 | file_release_write(file); |
aceaf78d | 240 | } |
aceaf78d | 241 | |
d7065da0 | 242 | /* the real guts of fput() - releasing the last reference to file |
1da177e4 | 243 | */ |
d7065da0 | 244 | static void __fput(struct file *file) |
1da177e4 | 245 | { |
0f7fc9e4 JJS |
246 | struct dentry *dentry = file->f_path.dentry; |
247 | struct vfsmount *mnt = file->f_path.mnt; | |
1da177e4 LT |
248 | struct inode *inode = dentry->d_inode; |
249 | ||
250 | might_sleep(); | |
0eeca283 RL |
251 | |
252 | fsnotify_close(file); | |
1da177e4 LT |
253 | /* |
254 | * The function eventpoll_release() should be the first called | |
255 | * in the file cleanup chain. | |
256 | */ | |
257 | eventpoll_release(file); | |
258 | locks_remove_flock(file); | |
259 | ||
233e70f4 AV |
260 | if (unlikely(file->f_flags & FASYNC)) { |
261 | if (file->f_op && file->f_op->fasync) | |
262 | file->f_op->fasync(-1, file, 0); | |
263 | } | |
4199d35c | 264 | ima_file_free(file); |
1da177e4 LT |
265 | if (file->f_op && file->f_op->release) |
266 | file->f_op->release(inode, file); | |
267 | security_file_free(file); | |
60ed8cf7 MS |
268 | if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL && |
269 | !(file->f_mode & FMODE_PATH))) { | |
1da177e4 | 270 | cdev_put(inode->i_cdev); |
60ed8cf7 | 271 | } |
1da177e4 | 272 | fops_put(file->f_op); |
609d7fa9 | 273 | put_pid(file->f_owner.pid); |
890275b5 MZ |
274 | if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) |
275 | i_readcount_dec(inode); | |
aceaf78d DH |
276 | if (file->f_mode & FMODE_WRITE) |
277 | drop_file_write_access(file); | |
0f7fc9e4 JJS |
278 | file->f_path.dentry = NULL; |
279 | file->f_path.mnt = NULL; | |
dd37978c | 280 | file->f_inode = NULL; |
1da177e4 LT |
281 | file_free(file); |
282 | dput(dentry); | |
283 | mntput(mnt); | |
284 | } | |
285 | ||
2a7abb5a | 286 | static LLIST_HEAD(delayed_fput_list); |
4a9d4b02 AV |
287 | static void delayed_fput(struct work_struct *unused) |
288 | { | |
2a7abb5a ON |
289 | struct llist_node *node = llist_del_all(&delayed_fput_list); |
290 | struct llist_node *next; | |
291 | ||
292 | for (; node; node = next) { | |
293 | next = llist_next(node); | |
294 | __fput(llist_entry(node, struct file, f_u.fu_llist)); | |
4a9d4b02 AV |
295 | } |
296 | } | |
297 | ||
298 | static void ____fput(struct callback_head *work) | |
299 | { | |
300 | __fput(container_of(work, struct file, f_u.fu_rcuhead)); | |
301 | } | |
302 | ||
303 | /* | |
304 | * If kernel thread really needs to have the final fput() it has done | |
305 | * to complete, call this. The only user right now is the boot - we | |
306 | * *do* need to make sure our writes to binaries on initramfs has | |
307 | * not left us with opened struct file waiting for __fput() - execve() | |
308 | * won't work without that. Please, don't add more callers without | |
309 | * very good reasons; in particular, never call that with locks | |
310 | * held and never call that from a thread that might need to do | |
311 | * some work on any kind of umount. | |
312 | */ | |
313 | void flush_delayed_fput(void) | |
314 | { | |
315 | delayed_fput(NULL); | |
316 | } | |
317 | ||
318 | static DECLARE_WORK(delayed_fput_work, delayed_fput); | |
319 | ||
d7065da0 AV |
320 | void fput(struct file *file) |
321 | { | |
4a9d4b02 AV |
322 | if (atomic_long_dec_and_test(&file->f_count)) { |
323 | struct task_struct *task = current; | |
e7b2c406 | 324 | |
e7b2c406 ON |
325 | if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) { |
326 | init_task_work(&file->f_u.fu_rcuhead, ____fput); | |
327 | if (!task_work_add(task, &file->f_u.fu_rcuhead, true)) | |
328 | return; | |
4a9d4b02 | 329 | } |
2a7abb5a ON |
330 | |
331 | if (llist_add(&file->f_u.fu_llist, &delayed_fput_list)) | |
332 | schedule_work(&delayed_fput_work); | |
4a9d4b02 AV |
333 | } |
334 | } | |
335 | ||
336 | /* | |
337 | * synchronous analog of fput(); for kernel threads that might be needed | |
338 | * in some umount() (and thus can't use flush_delayed_fput() without | |
339 | * risking deadlocks), need to wait for completion of __fput() and know | |
340 | * for this specific struct file it won't involve anything that would | |
341 | * need them. Use only if you really need it - at the very least, | |
342 | * don't blindly convert fput() by kernel thread to that. | |
343 | */ | |
344 | void __fput_sync(struct file *file) | |
345 | { | |
346 | if (atomic_long_dec_and_test(&file->f_count)) { | |
347 | struct task_struct *task = current; | |
4a9d4b02 | 348 | BUG_ON(!(task->flags & PF_KTHREAD)); |
d7065da0 | 349 | __fput(file); |
4a9d4b02 | 350 | } |
d7065da0 AV |
351 | } |
352 | ||
353 | EXPORT_SYMBOL(fput); | |
354 | ||
1da177e4 LT |
355 | void put_filp(struct file *file) |
356 | { | |
516e0cc5 | 357 | if (atomic_long_dec_and_test(&file->f_count)) { |
1da177e4 | 358 | security_file_free(file); |
1da177e4 LT |
359 | file_free(file); |
360 | } | |
361 | } | |
362 | ||
1da177e4 LT |
363 | void __init files_init(unsigned long mempages) |
364 | { | |
518de9b3 | 365 | unsigned long n; |
b6b3fdea ED |
366 | |
367 | filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0, | |
368 | SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
369 | ||
370 | /* | |
371 | * One file with associated inode and dcache is very roughly 1K. | |
1da177e4 LT |
372 | * Per default don't use more than 10% of our memory for files. |
373 | */ | |
374 | ||
375 | n = (mempages * (PAGE_SIZE / 1024)) / 10; | |
518de9b3 | 376 | files_stat.max_files = max_t(unsigned long, n, NR_FILE); |
ab2af1f5 | 377 | files_defer_init(); |
0216bfcf | 378 | percpu_counter_init(&nr_files, 0); |
1da177e4 | 379 | } |