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> | |
6416ccb7 | 24 | #include <linux/percpu.h> |
4a9d4b02 AV |
25 | #include <linux/hardirq.h> |
26 | #include <linux/task_work.h> | |
0552f879 | 27 | #include <linux/ima.h> |
529bf6be | 28 | |
60063497 | 29 | #include <linux/atomic.h> |
1da177e4 | 30 | |
e81e3f4d EP |
31 | #include "internal.h" |
32 | ||
1da177e4 LT |
33 | /* sysctl tunables... */ |
34 | struct files_stat_struct files_stat = { | |
35 | .max_files = NR_FILE | |
36 | }; | |
37 | ||
b6b3fdea ED |
38 | /* SLAB cache for file structures */ |
39 | static struct kmem_cache *filp_cachep __read_mostly; | |
40 | ||
529bf6be | 41 | static struct percpu_counter nr_files __cacheline_aligned_in_smp; |
1da177e4 | 42 | |
5c33b183 | 43 | static void file_free_rcu(struct rcu_head *head) |
1da177e4 | 44 | { |
d76b0d9b DH |
45 | struct file *f = container_of(head, struct file, f_u.fu_rcuhead); |
46 | ||
47 | put_cred(f->f_cred); | |
529bf6be | 48 | kmem_cache_free(filp_cachep, f); |
1da177e4 LT |
49 | } |
50 | ||
529bf6be | 51 | static inline void file_free(struct file *f) |
1da177e4 | 52 | { |
529bf6be DS |
53 | percpu_counter_dec(&nr_files); |
54 | call_rcu(&f->f_u.fu_rcuhead, file_free_rcu); | |
1da177e4 LT |
55 | } |
56 | ||
529bf6be DS |
57 | /* |
58 | * Return the total number of open files in the system | |
59 | */ | |
518de9b3 | 60 | static long get_nr_files(void) |
1da177e4 | 61 | { |
529bf6be | 62 | return percpu_counter_read_positive(&nr_files); |
1da177e4 LT |
63 | } |
64 | ||
529bf6be DS |
65 | /* |
66 | * Return the maximum number of open files in the system | |
67 | */ | |
518de9b3 | 68 | unsigned long get_max_files(void) |
ab2af1f5 | 69 | { |
529bf6be | 70 | return files_stat.max_files; |
ab2af1f5 | 71 | } |
529bf6be DS |
72 | EXPORT_SYMBOL_GPL(get_max_files); |
73 | ||
74 | /* | |
75 | * Handle nr_files sysctl | |
76 | */ | |
77 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
1f7e0616 | 78 | int proc_nr_files(struct ctl_table *table, int write, |
529bf6be DS |
79 | void __user *buffer, size_t *lenp, loff_t *ppos) |
80 | { | |
81 | files_stat.nr_files = get_nr_files(); | |
518de9b3 | 82 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
529bf6be DS |
83 | } |
84 | #else | |
1f7e0616 | 85 | int proc_nr_files(struct ctl_table *table, int write, |
529bf6be DS |
86 | void __user *buffer, size_t *lenp, loff_t *ppos) |
87 | { | |
88 | return -ENOSYS; | |
89 | } | |
90 | #endif | |
ab2af1f5 | 91 | |
1da177e4 | 92 | /* Find an unused file structure and return a pointer to it. |
1afc99be AV |
93 | * Returns an error pointer if some error happend e.g. we over file |
94 | * structures limit, run out of memory or operation is not permitted. | |
430e285e DH |
95 | * |
96 | * Be very careful using this. You are responsible for | |
97 | * getting write access to any mount that you might assign | |
98 | * to this filp, if it is opened for write. If this is not | |
99 | * done, you will imbalance int the mount's writer count | |
100 | * and a warning at __fput() time. | |
1da177e4 LT |
101 | */ |
102 | struct file *get_empty_filp(void) | |
103 | { | |
86a264ab | 104 | const struct cred *cred = current_cred(); |
518de9b3 | 105 | static long old_max; |
1afc99be AV |
106 | struct file *f; |
107 | int error; | |
1da177e4 LT |
108 | |
109 | /* | |
110 | * Privileged users can go above max_files | |
111 | */ | |
529bf6be DS |
112 | if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) { |
113 | /* | |
114 | * percpu_counters are inaccurate. Do an expensive check before | |
115 | * we go and fail. | |
116 | */ | |
52d9f3b4 | 117 | if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files) |
529bf6be DS |
118 | goto over; |
119 | } | |
af4d2ecb | 120 | |
4975e45f | 121 | f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL); |
1afc99be AV |
122 | if (unlikely(!f)) |
123 | return ERR_PTR(-ENOMEM); | |
af4d2ecb | 124 | |
529bf6be | 125 | percpu_counter_inc(&nr_files); |
78d29788 | 126 | f->f_cred = get_cred(cred); |
1afc99be AV |
127 | error = security_file_alloc(f); |
128 | if (unlikely(error)) { | |
129 | file_free(f); | |
130 | return ERR_PTR(error); | |
131 | } | |
1da177e4 | 132 | |
516e0cc5 | 133 | atomic_long_set(&f->f_count, 1); |
af4d2ecb | 134 | rwlock_init(&f->f_owner.lock); |
68499914 | 135 | spin_lock_init(&f->f_lock); |
9c225f26 | 136 | mutex_init(&f->f_pos_lock); |
5a6b7951 | 137 | eventpoll_init_file(f); |
af4d2ecb | 138 | /* f->f_version: 0 */ |
af4d2ecb KK |
139 | return f; |
140 | ||
141 | over: | |
1da177e4 | 142 | /* Ran out of filps - report that */ |
529bf6be | 143 | if (get_nr_files() > old_max) { |
518de9b3 | 144 | pr_info("VFS: file-max limit %lu reached\n", get_max_files()); |
529bf6be | 145 | old_max = get_nr_files(); |
1da177e4 | 146 | } |
1afc99be | 147 | return ERR_PTR(-ENFILE); |
1da177e4 LT |
148 | } |
149 | ||
ce8d2cdf DH |
150 | /** |
151 | * alloc_file - allocate and initialize a 'struct file' | |
a457606a EB |
152 | * |
153 | * @path: the (dentry, vfsmount) pair for the new file | |
ce8d2cdf DH |
154 | * @mode: the mode with which the new file will be opened |
155 | * @fop: the 'struct file_operations' for the new file | |
ce8d2cdf | 156 | */ |
2c48b9c4 AV |
157 | struct file *alloc_file(struct path *path, fmode_t mode, |
158 | const struct file_operations *fop) | |
ce8d2cdf DH |
159 | { |
160 | struct file *file; | |
ce8d2cdf DH |
161 | |
162 | file = get_empty_filp(); | |
1afc99be | 163 | if (IS_ERR(file)) |
39b65252 | 164 | return file; |
ce8d2cdf | 165 | |
2c48b9c4 | 166 | file->f_path = *path; |
dd37978c | 167 | file->f_inode = path->dentry->d_inode; |
2c48b9c4 | 168 | file->f_mapping = path->dentry->d_inode->i_mapping; |
293bc982 | 169 | if ((mode & FMODE_READ) && |
84363182 | 170 | likely(fop->read || fop->read_iter)) |
7f7f25e8 | 171 | mode |= FMODE_CAN_READ; |
293bc982 | 172 | if ((mode & FMODE_WRITE) && |
84363182 | 173 | likely(fop->write || fop->write_iter)) |
7f7f25e8 | 174 | mode |= FMODE_CAN_WRITE; |
ce8d2cdf DH |
175 | file->f_mode = mode; |
176 | file->f_op = fop; | |
890275b5 MZ |
177 | if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) |
178 | i_readcount_inc(path->dentry->d_inode); | |
3d1e4631 | 179 | return file; |
ce8d2cdf | 180 | } |
73efc468 | 181 | EXPORT_SYMBOL(alloc_file); |
ce8d2cdf | 182 | |
d7065da0 | 183 | /* the real guts of fput() - releasing the last reference to file |
1da177e4 | 184 | */ |
d7065da0 | 185 | static void __fput(struct file *file) |
1da177e4 | 186 | { |
0f7fc9e4 JJS |
187 | struct dentry *dentry = file->f_path.dentry; |
188 | struct vfsmount *mnt = file->f_path.mnt; | |
c77cecee | 189 | struct inode *inode = file->f_inode; |
1da177e4 LT |
190 | |
191 | might_sleep(); | |
0eeca283 RL |
192 | |
193 | fsnotify_close(file); | |
1da177e4 LT |
194 | /* |
195 | * The function eventpoll_release() should be the first called | |
196 | * in the file cleanup chain. | |
197 | */ | |
198 | eventpoll_release(file); | |
78ed8a13 | 199 | locks_remove_file(file); |
1da177e4 | 200 | |
233e70f4 | 201 | if (unlikely(file->f_flags & FASYNC)) { |
72c2d531 | 202 | if (file->f_op->fasync) |
233e70f4 AV |
203 | file->f_op->fasync(-1, file, 0); |
204 | } | |
4199d35c | 205 | ima_file_free(file); |
72c2d531 | 206 | if (file->f_op->release) |
1da177e4 LT |
207 | file->f_op->release(inode, file); |
208 | security_file_free(file); | |
60ed8cf7 MS |
209 | if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL && |
210 | !(file->f_mode & FMODE_PATH))) { | |
1da177e4 | 211 | cdev_put(inode->i_cdev); |
60ed8cf7 | 212 | } |
1da177e4 | 213 | fops_put(file->f_op); |
609d7fa9 | 214 | put_pid(file->f_owner.pid); |
890275b5 MZ |
215 | if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) |
216 | i_readcount_dec(inode); | |
83f936c7 AV |
217 | if (file->f_mode & FMODE_WRITER) { |
218 | put_write_access(inode); | |
219 | __mnt_drop_write(mnt); | |
220 | } | |
0f7fc9e4 JJS |
221 | file->f_path.dentry = NULL; |
222 | file->f_path.mnt = NULL; | |
dd37978c | 223 | file->f_inode = NULL; |
1da177e4 LT |
224 | file_free(file); |
225 | dput(dentry); | |
226 | mntput(mnt); | |
227 | } | |
228 | ||
4f5e65a1 | 229 | static LLIST_HEAD(delayed_fput_list); |
4a9d4b02 AV |
230 | static void delayed_fput(struct work_struct *unused) |
231 | { | |
4f5e65a1 ON |
232 | struct llist_node *node = llist_del_all(&delayed_fput_list); |
233 | struct llist_node *next; | |
234 | ||
235 | for (; node; node = next) { | |
236 | next = llist_next(node); | |
237 | __fput(llist_entry(node, struct file, f_u.fu_llist)); | |
4a9d4b02 AV |
238 | } |
239 | } | |
240 | ||
241 | static void ____fput(struct callback_head *work) | |
242 | { | |
243 | __fput(container_of(work, struct file, f_u.fu_rcuhead)); | |
244 | } | |
245 | ||
246 | /* | |
247 | * If kernel thread really needs to have the final fput() it has done | |
248 | * to complete, call this. The only user right now is the boot - we | |
249 | * *do* need to make sure our writes to binaries on initramfs has | |
250 | * not left us with opened struct file waiting for __fput() - execve() | |
251 | * won't work without that. Please, don't add more callers without | |
252 | * very good reasons; in particular, never call that with locks | |
253 | * held and never call that from a thread that might need to do | |
254 | * some work on any kind of umount. | |
255 | */ | |
256 | void flush_delayed_fput(void) | |
257 | { | |
258 | delayed_fput(NULL); | |
259 | } | |
260 | ||
c7314d74 | 261 | static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput); |
4a9d4b02 | 262 | |
d7065da0 AV |
263 | void fput(struct file *file) |
264 | { | |
4a9d4b02 AV |
265 | if (atomic_long_dec_and_test(&file->f_count)) { |
266 | struct task_struct *task = current; | |
e7b2c406 | 267 | |
e7b2c406 ON |
268 | if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) { |
269 | init_task_work(&file->f_u.fu_rcuhead, ____fput); | |
270 | if (!task_work_add(task, &file->f_u.fu_rcuhead, true)) | |
271 | return; | |
64372501 AM |
272 | /* |
273 | * After this task has run exit_task_work(), | |
be49b30a | 274 | * task_work_add() will fail. Fall through to delayed |
64372501 AM |
275 | * fput to avoid leaking *file. |
276 | */ | |
4a9d4b02 | 277 | } |
4f5e65a1 ON |
278 | |
279 | if (llist_add(&file->f_u.fu_llist, &delayed_fput_list)) | |
c7314d74 | 280 | schedule_delayed_work(&delayed_fput_work, 1); |
4a9d4b02 AV |
281 | } |
282 | } | |
283 | ||
284 | /* | |
285 | * synchronous analog of fput(); for kernel threads that might be needed | |
286 | * in some umount() (and thus can't use flush_delayed_fput() without | |
287 | * risking deadlocks), need to wait for completion of __fput() and know | |
288 | * for this specific struct file it won't involve anything that would | |
289 | * need them. Use only if you really need it - at the very least, | |
290 | * don't blindly convert fput() by kernel thread to that. | |
291 | */ | |
292 | void __fput_sync(struct file *file) | |
293 | { | |
294 | if (atomic_long_dec_and_test(&file->f_count)) { | |
295 | struct task_struct *task = current; | |
4a9d4b02 | 296 | BUG_ON(!(task->flags & PF_KTHREAD)); |
d7065da0 | 297 | __fput(file); |
4a9d4b02 | 298 | } |
d7065da0 AV |
299 | } |
300 | ||
301 | EXPORT_SYMBOL(fput); | |
302 | ||
1da177e4 LT |
303 | void put_filp(struct file *file) |
304 | { | |
516e0cc5 | 305 | if (atomic_long_dec_and_test(&file->f_count)) { |
1da177e4 | 306 | security_file_free(file); |
1da177e4 LT |
307 | file_free(file); |
308 | } | |
309 | } | |
310 | ||
1da177e4 LT |
311 | void __init files_init(unsigned long mempages) |
312 | { | |
518de9b3 | 313 | unsigned long n; |
b6b3fdea ED |
314 | |
315 | filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0, | |
316 | SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); | |
317 | ||
318 | /* | |
319 | * One file with associated inode and dcache is very roughly 1K. | |
1da177e4 LT |
320 | * Per default don't use more than 10% of our memory for files. |
321 | */ | |
322 | ||
323 | n = (mempages * (PAGE_SIZE / 1024)) / 10; | |
518de9b3 | 324 | files_stat.max_files = max_t(unsigned long, n, NR_FILE); |
908c7f19 | 325 | percpu_counter_init(&nr_files, 0, GFP_KERNEL); |
1da177e4 | 326 | } |