Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/fs/file_table.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992 Linus Torvalds | |
6 | * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu) | |
7 | */ | |
8 | ||
9 | #include <linux/string.h> | |
10 | #include <linux/slab.h> | |
11 | #include <linux/file.h> | |
9f3acc31 | 12 | #include <linux/fdtable.h> |
1da177e4 LT |
13 | #include <linux/init.h> |
14 | #include <linux/module.h> | |
1da177e4 LT |
15 | #include <linux/fs.h> |
16 | #include <linux/security.h> | |
5b825c3a | 17 | #include <linux/cred.h> |
1da177e4 | 18 | #include <linux/eventpoll.h> |
ab2af1f5 | 19 | #include <linux/rcupdate.h> |
1da177e4 | 20 | #include <linux/mount.h> |
16f7e0fe | 21 | #include <linux/capability.h> |
1da177e4 | 22 | #include <linux/cdev.h> |
0eeca283 | 23 | #include <linux/fsnotify.h> |
529bf6be DS |
24 | #include <linux/sysctl.h> |
25 | #include <linux/percpu_counter.h> | |
6416ccb7 | 26 | #include <linux/percpu.h> |
4a9d4b02 | 27 | #include <linux/task_work.h> |
0552f879 | 28 | #include <linux/ima.h> |
4248b0da | 29 | #include <linux/swap.h> |
a3580ac9 | 30 | #include <linux/kmemleak.h> |
529bf6be | 31 | |
60063497 | 32 | #include <linux/atomic.h> |
1da177e4 | 33 | |
e81e3f4d EP |
34 | #include "internal.h" |
35 | ||
1da177e4 | 36 | /* sysctl tunables... */ |
204d5a24 | 37 | static struct files_stat_struct files_stat = { |
1da177e4 LT |
38 | .max_files = NR_FILE |
39 | }; | |
40 | ||
b6b3fdea ED |
41 | /* SLAB cache for file structures */ |
42 | static struct kmem_cache *filp_cachep __read_mostly; | |
43 | ||
529bf6be | 44 | static struct percpu_counter nr_files __cacheline_aligned_in_smp; |
1da177e4 | 45 | |
5c33b183 | 46 | static void file_free_rcu(struct rcu_head *head) |
1da177e4 | 47 | { |
e87f2c26 | 48 | struct file *f = container_of(head, struct file, f_rcuhead); |
d76b0d9b DH |
49 | |
50 | put_cred(f->f_cred); | |
529bf6be | 51 | kmem_cache_free(filp_cachep, f); |
1da177e4 LT |
52 | } |
53 | ||
529bf6be | 54 | static inline void file_free(struct file *f) |
1da177e4 | 55 | { |
e8cff84f | 56 | security_file_free(f); |
d3b1084d MS |
57 | if (!(f->f_mode & FMODE_NOACCOUNT)) |
58 | percpu_counter_dec(&nr_files); | |
e87f2c26 | 59 | call_rcu(&f->f_rcuhead, file_free_rcu); |
1da177e4 LT |
60 | } |
61 | ||
529bf6be DS |
62 | /* |
63 | * Return the total number of open files in the system | |
64 | */ | |
518de9b3 | 65 | static long get_nr_files(void) |
1da177e4 | 66 | { |
529bf6be | 67 | return percpu_counter_read_positive(&nr_files); |
1da177e4 LT |
68 | } |
69 | ||
529bf6be DS |
70 | /* |
71 | * Return the maximum number of open files in the system | |
72 | */ | |
518de9b3 | 73 | unsigned long get_max_files(void) |
ab2af1f5 | 74 | { |
529bf6be | 75 | return files_stat.max_files; |
ab2af1f5 | 76 | } |
529bf6be DS |
77 | EXPORT_SYMBOL_GPL(get_max_files); |
78 | ||
204d5a24 LC |
79 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) |
80 | ||
529bf6be DS |
81 | /* |
82 | * Handle nr_files sysctl | |
83 | */ | |
204d5a24 LC |
84 | static int proc_nr_files(struct ctl_table *table, int write, void *buffer, |
85 | size_t *lenp, loff_t *ppos) | |
529bf6be DS |
86 | { |
87 | files_stat.nr_files = get_nr_files(); | |
518de9b3 | 88 | return proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
529bf6be | 89 | } |
204d5a24 LC |
90 | |
91 | static struct ctl_table fs_stat_sysctls[] = { | |
92 | { | |
93 | .procname = "file-nr", | |
94 | .data = &files_stat, | |
95 | .maxlen = sizeof(files_stat), | |
96 | .mode = 0444, | |
97 | .proc_handler = proc_nr_files, | |
98 | }, | |
99 | { | |
100 | .procname = "file-max", | |
101 | .data = &files_stat.max_files, | |
102 | .maxlen = sizeof(files_stat.max_files), | |
103 | .mode = 0644, | |
104 | .proc_handler = proc_doulongvec_minmax, | |
105 | .extra1 = SYSCTL_LONG_ZERO, | |
106 | .extra2 = SYSCTL_LONG_MAX, | |
107 | }, | |
108 | { | |
109 | .procname = "nr_open", | |
110 | .data = &sysctl_nr_open, | |
111 | .maxlen = sizeof(unsigned int), | |
112 | .mode = 0644, | |
113 | .proc_handler = proc_dointvec_minmax, | |
114 | .extra1 = &sysctl_nr_open_min, | |
115 | .extra2 = &sysctl_nr_open_max, | |
116 | }, | |
117 | { } | |
118 | }; | |
119 | ||
120 | static int __init init_fs_stat_sysctls(void) | |
529bf6be | 121 | { |
204d5a24 | 122 | register_sysctl_init("fs", fs_stat_sysctls); |
a3580ac9 LC |
123 | if (IS_ENABLED(CONFIG_BINFMT_MISC)) { |
124 | struct ctl_table_header *hdr; | |
125 | hdr = register_sysctl_mount_point("fs/binfmt_misc"); | |
126 | kmemleak_not_leak(hdr); | |
127 | } | |
204d5a24 | 128 | return 0; |
529bf6be | 129 | } |
204d5a24 | 130 | fs_initcall(init_fs_stat_sysctls); |
529bf6be | 131 | #endif |
ab2af1f5 | 132 | |
d3b1084d MS |
133 | static struct file *__alloc_file(int flags, const struct cred *cred) |
134 | { | |
135 | struct file *f; | |
136 | int error; | |
137 | ||
138 | f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL); | |
139 | if (unlikely(!f)) | |
140 | return ERR_PTR(-ENOMEM); | |
141 | ||
142 | f->f_cred = get_cred(cred); | |
143 | error = security_file_alloc(f); | |
144 | if (unlikely(error)) { | |
e87f2c26 | 145 | file_free_rcu(&f->f_rcuhead); |
d3b1084d MS |
146 | return ERR_PTR(error); |
147 | } | |
148 | ||
149 | atomic_long_set(&f->f_count, 1); | |
150 | rwlock_init(&f->f_owner.lock); | |
151 | spin_lock_init(&f->f_lock); | |
152 | mutex_init(&f->f_pos_lock); | |
d3b1084d MS |
153 | f->f_flags = flags; |
154 | f->f_mode = OPEN_FMODE(flags); | |
155 | /* f->f_version: 0 */ | |
156 | ||
157 | return f; | |
158 | } | |
159 | ||
1da177e4 | 160 | /* Find an unused file structure and return a pointer to it. |
1afc99be AV |
161 | * Returns an error pointer if some error happend e.g. we over file |
162 | * structures limit, run out of memory or operation is not permitted. | |
430e285e DH |
163 | * |
164 | * Be very careful using this. You are responsible for | |
165 | * getting write access to any mount that you might assign | |
166 | * to this filp, if it is opened for write. If this is not | |
167 | * done, you will imbalance int the mount's writer count | |
168 | * and a warning at __fput() time. | |
1da177e4 | 169 | */ |
ea73ea72 | 170 | struct file *alloc_empty_file(int flags, const struct cred *cred) |
1da177e4 | 171 | { |
518de9b3 | 172 | static long old_max; |
1afc99be | 173 | struct file *f; |
1da177e4 LT |
174 | |
175 | /* | |
176 | * Privileged users can go above max_files | |
177 | */ | |
529bf6be DS |
178 | if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) { |
179 | /* | |
180 | * percpu_counters are inaccurate. Do an expensive check before | |
181 | * we go and fail. | |
182 | */ | |
52d9f3b4 | 183 | if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files) |
529bf6be DS |
184 | goto over; |
185 | } | |
af4d2ecb | 186 | |
d3b1084d MS |
187 | f = __alloc_file(flags, cred); |
188 | if (!IS_ERR(f)) | |
189 | percpu_counter_inc(&nr_files); | |
1da177e4 | 190 | |
af4d2ecb KK |
191 | return f; |
192 | ||
193 | over: | |
1da177e4 | 194 | /* Ran out of filps - report that */ |
529bf6be | 195 | if (get_nr_files() > old_max) { |
518de9b3 | 196 | pr_info("VFS: file-max limit %lu reached\n", get_max_files()); |
529bf6be | 197 | old_max = get_nr_files(); |
1da177e4 | 198 | } |
1afc99be | 199 | return ERR_PTR(-ENFILE); |
1da177e4 LT |
200 | } |
201 | ||
d3b1084d MS |
202 | /* |
203 | * Variant of alloc_empty_file() that doesn't check and modify nr_files. | |
204 | * | |
205 | * Should not be used unless there's a very good reason to do so. | |
206 | */ | |
207 | struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred) | |
208 | { | |
209 | struct file *f = __alloc_file(flags, cred); | |
210 | ||
211 | if (!IS_ERR(f)) | |
212 | f->f_mode |= FMODE_NOACCOUNT; | |
213 | ||
214 | return f; | |
215 | } | |
216 | ||
ce8d2cdf DH |
217 | /** |
218 | * alloc_file - allocate and initialize a 'struct file' | |
a457606a EB |
219 | * |
220 | * @path: the (dentry, vfsmount) pair for the new file | |
c9c554f2 | 221 | * @flags: O_... flags with which the new file will be opened |
ce8d2cdf | 222 | * @fop: the 'struct file_operations' for the new file |
ce8d2cdf | 223 | */ |
ee1904ba | 224 | static struct file *alloc_file(const struct path *path, int flags, |
2c48b9c4 | 225 | const struct file_operations *fop) |
ce8d2cdf DH |
226 | { |
227 | struct file *file; | |
ce8d2cdf | 228 | |
ea73ea72 | 229 | file = alloc_empty_file(flags, current_cred()); |
1afc99be | 230 | if (IS_ERR(file)) |
39b65252 | 231 | return file; |
ce8d2cdf | 232 | |
2c48b9c4 | 233 | file->f_path = *path; |
dd37978c | 234 | file->f_inode = path->dentry->d_inode; |
2c48b9c4 | 235 | file->f_mapping = path->dentry->d_inode->i_mapping; |
5660e13d | 236 | file->f_wb_err = filemap_sample_wb_err(file->f_mapping); |
735e4ae5 | 237 | file->f_sb_err = file_sample_sb_err(file); |
868941b1 | 238 | if (fop->llseek) |
e7478158 | 239 | file->f_mode |= FMODE_LSEEK; |
c9c554f2 | 240 | if ((file->f_mode & FMODE_READ) && |
84363182 | 241 | likely(fop->read || fop->read_iter)) |
c9c554f2 AV |
242 | file->f_mode |= FMODE_CAN_READ; |
243 | if ((file->f_mode & FMODE_WRITE) && | |
84363182 | 244 | likely(fop->write || fop->write_iter)) |
c9c554f2 | 245 | file->f_mode |= FMODE_CAN_WRITE; |
164f4064 | 246 | file->f_iocb_flags = iocb_flags(file); |
f5d11409 | 247 | file->f_mode |= FMODE_OPENED; |
ce8d2cdf | 248 | file->f_op = fop; |
c9c554f2 | 249 | if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) |
890275b5 | 250 | i_readcount_inc(path->dentry->d_inode); |
3d1e4631 | 251 | return file; |
ce8d2cdf | 252 | } |
ce8d2cdf | 253 | |
d93aa9d8 AV |
254 | struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt, |
255 | const char *name, int flags, | |
256 | const struct file_operations *fops) | |
257 | { | |
258 | static const struct dentry_operations anon_ops = { | |
259 | .d_dname = simple_dname | |
260 | }; | |
261 | struct qstr this = QSTR_INIT(name, strlen(name)); | |
262 | struct path path; | |
263 | struct file *file; | |
264 | ||
265 | path.dentry = d_alloc_pseudo(mnt->mnt_sb, &this); | |
266 | if (!path.dentry) | |
267 | return ERR_PTR(-ENOMEM); | |
268 | if (!mnt->mnt_sb->s_d_op) | |
269 | d_set_d_op(path.dentry, &anon_ops); | |
270 | path.mnt = mntget(mnt); | |
271 | d_instantiate(path.dentry, inode); | |
b6509f6a | 272 | file = alloc_file(&path, flags, fops); |
d93aa9d8 AV |
273 | if (IS_ERR(file)) { |
274 | ihold(inode); | |
275 | path_put(&path); | |
276 | } | |
277 | return file; | |
278 | } | |
279 | EXPORT_SYMBOL(alloc_file_pseudo); | |
280 | ||
183266f2 AV |
281 | struct file *alloc_file_clone(struct file *base, int flags, |
282 | const struct file_operations *fops) | |
283 | { | |
284 | struct file *f = alloc_file(&base->f_path, flags, fops); | |
285 | if (!IS_ERR(f)) { | |
286 | path_get(&f->f_path); | |
287 | f->f_mapping = base->f_mapping; | |
288 | } | |
289 | return f; | |
290 | } | |
291 | ||
d7065da0 | 292 | /* the real guts of fput() - releasing the last reference to file |
1da177e4 | 293 | */ |
d7065da0 | 294 | static void __fput(struct file *file) |
1da177e4 | 295 | { |
0f7fc9e4 JJS |
296 | struct dentry *dentry = file->f_path.dentry; |
297 | struct vfsmount *mnt = file->f_path.mnt; | |
c77cecee | 298 | struct inode *inode = file->f_inode; |
a07b2000 | 299 | fmode_t mode = file->f_mode; |
1da177e4 | 300 | |
4d27f326 AV |
301 | if (unlikely(!(file->f_mode & FMODE_OPENED))) |
302 | goto out; | |
303 | ||
1da177e4 | 304 | might_sleep(); |
0eeca283 RL |
305 | |
306 | fsnotify_close(file); | |
1da177e4 LT |
307 | /* |
308 | * The function eventpoll_release() should be the first called | |
309 | * in the file cleanup chain. | |
310 | */ | |
311 | eventpoll_release(file); | |
78ed8a13 | 312 | locks_remove_file(file); |
1da177e4 | 313 | |
bb02b186 | 314 | ima_file_free(file); |
233e70f4 | 315 | if (unlikely(file->f_flags & FASYNC)) { |
72c2d531 | 316 | if (file->f_op->fasync) |
233e70f4 AV |
317 | file->f_op->fasync(-1, file, 0); |
318 | } | |
72c2d531 | 319 | if (file->f_op->release) |
1da177e4 | 320 | file->f_op->release(inode, file); |
60ed8cf7 | 321 | if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL && |
a07b2000 | 322 | !(mode & FMODE_PATH))) { |
1da177e4 | 323 | cdev_put(inode->i_cdev); |
60ed8cf7 | 324 | } |
1da177e4 | 325 | fops_put(file->f_op); |
609d7fa9 | 326 | put_pid(file->f_owner.pid); |
a07b2000 | 327 | if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ) |
890275b5 | 328 | i_readcount_dec(inode); |
a07b2000 | 329 | if (mode & FMODE_WRITER) { |
83f936c7 AV |
330 | put_write_access(inode); |
331 | __mnt_drop_write(mnt); | |
332 | } | |
1da177e4 | 333 | dput(dentry); |
a07b2000 AV |
334 | if (unlikely(mode & FMODE_NEED_UNMOUNT)) |
335 | dissolve_on_fput(mnt); | |
1da177e4 | 336 | mntput(mnt); |
4d27f326 AV |
337 | out: |
338 | file_free(file); | |
1da177e4 LT |
339 | } |
340 | ||
4f5e65a1 | 341 | static LLIST_HEAD(delayed_fput_list); |
4a9d4b02 AV |
342 | static void delayed_fput(struct work_struct *unused) |
343 | { | |
4f5e65a1 | 344 | struct llist_node *node = llist_del_all(&delayed_fput_list); |
b9ea557e | 345 | struct file *f, *t; |
4f5e65a1 | 346 | |
e87f2c26 | 347 | llist_for_each_entry_safe(f, t, node, f_llist) |
b9ea557e | 348 | __fput(f); |
4a9d4b02 AV |
349 | } |
350 | ||
351 | static void ____fput(struct callback_head *work) | |
352 | { | |
e87f2c26 | 353 | __fput(container_of(work, struct file, f_rcuhead)); |
4a9d4b02 AV |
354 | } |
355 | ||
356 | /* | |
357 | * If kernel thread really needs to have the final fput() it has done | |
358 | * to complete, call this. The only user right now is the boot - we | |
359 | * *do* need to make sure our writes to binaries on initramfs has | |
360 | * not left us with opened struct file waiting for __fput() - execve() | |
361 | * won't work without that. Please, don't add more callers without | |
362 | * very good reasons; in particular, never call that with locks | |
363 | * held and never call that from a thread that might need to do | |
364 | * some work on any kind of umount. | |
365 | */ | |
366 | void flush_delayed_fput(void) | |
367 | { | |
368 | delayed_fput(NULL); | |
369 | } | |
7239a40c | 370 | EXPORT_SYMBOL_GPL(flush_delayed_fput); |
4a9d4b02 | 371 | |
c7314d74 | 372 | static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput); |
4a9d4b02 | 373 | |
81132a39 | 374 | void fput(struct file *file) |
d7065da0 | 375 | { |
81132a39 | 376 | if (atomic_long_dec_and_test(&file->f_count)) { |
4a9d4b02 | 377 | struct task_struct *task = current; |
e7b2c406 | 378 | |
e7b2c406 | 379 | if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) { |
e87f2c26 AV |
380 | init_task_work(&file->f_rcuhead, ____fput); |
381 | if (!task_work_add(task, &file->f_rcuhead, TWA_RESUME)) | |
e7b2c406 | 382 | return; |
64372501 AM |
383 | /* |
384 | * After this task has run exit_task_work(), | |
be49b30a | 385 | * task_work_add() will fail. Fall through to delayed |
64372501 AM |
386 | * fput to avoid leaking *file. |
387 | */ | |
4a9d4b02 | 388 | } |
4f5e65a1 | 389 | |
e87f2c26 | 390 | if (llist_add(&file->f_llist, &delayed_fput_list)) |
c7314d74 | 391 | schedule_delayed_work(&delayed_fput_work, 1); |
4a9d4b02 AV |
392 | } |
393 | } | |
394 | ||
395 | /* | |
396 | * synchronous analog of fput(); for kernel threads that might be needed | |
397 | * in some umount() (and thus can't use flush_delayed_fput() without | |
398 | * risking deadlocks), need to wait for completion of __fput() and know | |
399 | * for this specific struct file it won't involve anything that would | |
400 | * need them. Use only if you really need it - at the very least, | |
401 | * don't blindly convert fput() by kernel thread to that. | |
402 | */ | |
403 | void __fput_sync(struct file *file) | |
404 | { | |
405 | if (atomic_long_dec_and_test(&file->f_count)) { | |
406 | struct task_struct *task = current; | |
4a9d4b02 | 407 | BUG_ON(!(task->flags & PF_KTHREAD)); |
d7065da0 | 408 | __fput(file); |
4a9d4b02 | 409 | } |
d7065da0 AV |
410 | } |
411 | ||
412 | EXPORT_SYMBOL(fput); | |
f0043206 | 413 | EXPORT_SYMBOL(__fput_sync); |
d7065da0 | 414 | |
4248b0da | 415 | void __init files_init(void) |
b9ea557e | 416 | { |
b6b3fdea | 417 | filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0, |
f3f7c093 | 418 | SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, NULL); |
4248b0da MG |
419 | percpu_counter_init(&nr_files, 0, GFP_KERNEL); |
420 | } | |
b6b3fdea | 421 | |
4248b0da MG |
422 | /* |
423 | * One file with associated inode and dcache is very roughly 1K. Per default | |
424 | * do not use more than 10% of our memory for files. | |
425 | */ | |
426 | void __init files_maxfiles_init(void) | |
427 | { | |
428 | unsigned long n; | |
ca79b0c2 | 429 | unsigned long nr_pages = totalram_pages(); |
3d6357de | 430 | unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2; |
4248b0da | 431 | |
3d6357de AK |
432 | memreserve = min(memreserve, nr_pages - 1); |
433 | n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10; | |
1da177e4 | 434 | |
518de9b3 | 435 | files_stat.max_files = max_t(unsigned long, n, NR_FILE); |
b9ea557e | 436 | } |