Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
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>
12 : #include <linux/fdtable.h>
13 : #include <linux/init.h>
14 : #include <linux/module.h>
15 : #include <linux/fs.h>
16 : #include <linux/security.h>
17 : #include <linux/cred.h>
18 : #include <linux/eventpoll.h>
19 : #include <linux/rcupdate.h>
20 : #include <linux/mount.h>
21 : #include <linux/capability.h>
22 : #include <linux/cdev.h>
23 : #include <linux/fsnotify.h>
24 : #include <linux/sysctl.h>
25 : #include <linux/percpu_counter.h>
26 : #include <linux/percpu.h>
27 : #include <linux/task_work.h>
28 : #include <linux/ima.h>
29 : #include <linux/swap.h>
30 : #include <linux/kmemleak.h>
31 :
32 : #include <linux/atomic.h>
33 :
34 : #include "internal.h"
35 :
36 : /* sysctl tunables... */
37 : static struct files_stat_struct files_stat = {
38 : .max_files = NR_FILE
39 : };
40 :
41 : /* SLAB cache for file structures */
42 : static struct kmem_cache *filp_cachep __read_mostly;
43 :
44 : static struct percpu_counter nr_files __cacheline_aligned_in_smp;
45 :
46 0 : static void file_free_rcu(struct rcu_head *head)
47 : {
48 0 : struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
49 :
50 0 : put_cred(f->f_cred);
51 0 : kmem_cache_free(filp_cachep, f);
52 0 : }
53 :
54 0 : static inline void file_free(struct file *f)
55 : {
56 0 : security_file_free(f);
57 0 : if (!(f->f_mode & FMODE_NOACCOUNT))
58 : percpu_counter_dec(&nr_files);
59 0 : call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
60 0 : }
61 :
62 : /*
63 : * Return the total number of open files in the system
64 : */
65 : static long get_nr_files(void)
66 : {
67 0 : return percpu_counter_read_positive(&nr_files);
68 : }
69 :
70 : /*
71 : * Return the maximum number of open files in the system
72 : */
73 0 : unsigned long get_max_files(void)
74 : {
75 0 : return files_stat.max_files;
76 : }
77 : EXPORT_SYMBOL_GPL(get_max_files);
78 :
79 : #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
80 :
81 : /*
82 : * Handle nr_files sysctl
83 : */
84 0 : static int proc_nr_files(struct ctl_table *table, int write, void *buffer,
85 : size_t *lenp, loff_t *ppos)
86 : {
87 0 : files_stat.nr_files = get_nr_files();
88 0 : return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
89 : }
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 1 : static int __init init_fs_stat_sysctls(void)
121 : {
122 1 : register_sysctl_init("fs", fs_stat_sysctls);
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 : }
128 1 : return 0;
129 : }
130 : fs_initcall(init_fs_stat_sysctls);
131 : #endif
132 :
133 0 : static struct file *__alloc_file(int flags, const struct cred *cred)
134 : {
135 : struct file *f;
136 : int error;
137 :
138 0 : f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
139 0 : if (unlikely(!f))
140 : return ERR_PTR(-ENOMEM);
141 :
142 0 : f->f_cred = get_cred(cred);
143 0 : error = security_file_alloc(f);
144 : if (unlikely(error)) {
145 : file_free_rcu(&f->f_u.fu_rcuhead);
146 : return ERR_PTR(error);
147 : }
148 :
149 0 : atomic_long_set(&f->f_count, 1);
150 : rwlock_init(&f->f_owner.lock);
151 0 : spin_lock_init(&f->f_lock);
152 0 : mutex_init(&f->f_pos_lock);
153 0 : f->f_flags = flags;
154 0 : f->f_mode = OPEN_FMODE(flags);
155 : /* f->f_version: 0 */
156 :
157 0 : return f;
158 : }
159 :
160 : /* Find an unused file structure and return a pointer to it.
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.
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.
169 : */
170 0 : struct file *alloc_empty_file(int flags, const struct cred *cred)
171 : {
172 : static long old_max;
173 : struct file *f;
174 :
175 : /*
176 : * Privileged users can go above max_files
177 : */
178 0 : 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 : */
183 0 : if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
184 : goto over;
185 : }
186 :
187 0 : f = __alloc_file(flags, cred);
188 0 : if (!IS_ERR(f))
189 : percpu_counter_inc(&nr_files);
190 :
191 : return f;
192 :
193 : over:
194 : /* Ran out of filps - report that */
195 0 : if (get_nr_files() > old_max) {
196 0 : pr_info("VFS: file-max limit %lu reached\n", get_max_files());
197 0 : old_max = get_nr_files();
198 : }
199 : return ERR_PTR(-ENFILE);
200 : }
201 :
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 0 : struct file *alloc_empty_file_noaccount(int flags, const struct cred *cred)
208 : {
209 0 : struct file *f = __alloc_file(flags, cred);
210 :
211 0 : if (!IS_ERR(f))
212 0 : f->f_mode |= FMODE_NOACCOUNT;
213 :
214 0 : return f;
215 : }
216 :
217 : /**
218 : * alloc_file - allocate and initialize a 'struct file'
219 : *
220 : * @path: the (dentry, vfsmount) pair for the new file
221 : * @flags: O_... flags with which the new file will be opened
222 : * @fop: the 'struct file_operations' for the new file
223 : */
224 0 : static struct file *alloc_file(const struct path *path, int flags,
225 : const struct file_operations *fop)
226 : {
227 : struct file *file;
228 :
229 0 : file = alloc_empty_file(flags, current_cred());
230 0 : if (IS_ERR(file))
231 : return file;
232 :
233 0 : file->f_path = *path;
234 0 : file->f_inode = path->dentry->d_inode;
235 0 : file->f_mapping = path->dentry->d_inode->i_mapping;
236 0 : file->f_wb_err = filemap_sample_wb_err(file->f_mapping);
237 0 : file->f_sb_err = file_sample_sb_err(file);
238 0 : if ((file->f_mode & FMODE_READ) &&
239 0 : likely(fop->read || fop->read_iter))
240 0 : file->f_mode |= FMODE_CAN_READ;
241 0 : if ((file->f_mode & FMODE_WRITE) &&
242 0 : likely(fop->write || fop->write_iter))
243 0 : file->f_mode |= FMODE_CAN_WRITE;
244 0 : file->f_mode |= FMODE_OPENED;
245 0 : file->f_op = fop;
246 0 : if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
247 0 : i_readcount_inc(path->dentry->d_inode);
248 : return file;
249 : }
250 :
251 0 : struct file *alloc_file_pseudo(struct inode *inode, struct vfsmount *mnt,
252 : const char *name, int flags,
253 : const struct file_operations *fops)
254 : {
255 : static const struct dentry_operations anon_ops = {
256 : .d_dname = simple_dname
257 : };
258 0 : struct qstr this = QSTR_INIT(name, strlen(name));
259 : struct path path;
260 : struct file *file;
261 :
262 0 : path.dentry = d_alloc_pseudo(mnt->mnt_sb, &this);
263 0 : if (!path.dentry)
264 : return ERR_PTR(-ENOMEM);
265 0 : if (!mnt->mnt_sb->s_d_op)
266 0 : d_set_d_op(path.dentry, &anon_ops);
267 0 : path.mnt = mntget(mnt);
268 0 : d_instantiate(path.dentry, inode);
269 0 : file = alloc_file(&path, flags, fops);
270 0 : if (IS_ERR(file)) {
271 0 : ihold(inode);
272 0 : path_put(&path);
273 : }
274 : return file;
275 : }
276 : EXPORT_SYMBOL(alloc_file_pseudo);
277 :
278 0 : struct file *alloc_file_clone(struct file *base, int flags,
279 : const struct file_operations *fops)
280 : {
281 0 : struct file *f = alloc_file(&base->f_path, flags, fops);
282 0 : if (!IS_ERR(f)) {
283 0 : path_get(&f->f_path);
284 0 : f->f_mapping = base->f_mapping;
285 : }
286 0 : return f;
287 : }
288 :
289 : /* the real guts of fput() - releasing the last reference to file
290 : */
291 0 : static void __fput(struct file *file)
292 : {
293 0 : struct dentry *dentry = file->f_path.dentry;
294 0 : struct vfsmount *mnt = file->f_path.mnt;
295 0 : struct inode *inode = file->f_inode;
296 0 : fmode_t mode = file->f_mode;
297 :
298 0 : if (unlikely(!(file->f_mode & FMODE_OPENED)))
299 : goto out;
300 :
301 : might_sleep();
302 :
303 0 : fsnotify_close(file);
304 : /*
305 : * The function eventpoll_release() should be the first called
306 : * in the file cleanup chain.
307 : */
308 0 : eventpoll_release(file);
309 0 : locks_remove_file(file);
310 :
311 0 : ima_file_free(file);
312 0 : if (unlikely(file->f_flags & FASYNC)) {
313 0 : if (file->f_op->fasync)
314 0 : file->f_op->fasync(-1, file, 0);
315 : }
316 0 : if (file->f_op->release)
317 0 : file->f_op->release(inode, file);
318 0 : if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
319 : !(mode & FMODE_PATH))) {
320 0 : cdev_put(inode->i_cdev);
321 : }
322 0 : fops_put(file->f_op);
323 0 : put_pid(file->f_owner.pid);
324 0 : if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
325 0 : i_readcount_dec(inode);
326 0 : if (mode & FMODE_WRITER) {
327 0 : put_write_access(inode);
328 0 : __mnt_drop_write(mnt);
329 : }
330 0 : dput(dentry);
331 0 : if (unlikely(mode & FMODE_NEED_UNMOUNT))
332 0 : dissolve_on_fput(mnt);
333 0 : mntput(mnt);
334 : out:
335 0 : file_free(file);
336 0 : }
337 :
338 : static LLIST_HEAD(delayed_fput_list);
339 0 : static void delayed_fput(struct work_struct *unused)
340 : {
341 0 : struct llist_node *node = llist_del_all(&delayed_fput_list);
342 : struct file *f, *t;
343 :
344 0 : llist_for_each_entry_safe(f, t, node, f_u.fu_llist)
345 0 : __fput(f);
346 0 : }
347 :
348 0 : static void ____fput(struct callback_head *work)
349 : {
350 0 : __fput(container_of(work, struct file, f_u.fu_rcuhead));
351 0 : }
352 :
353 : /*
354 : * If kernel thread really needs to have the final fput() it has done
355 : * to complete, call this. The only user right now is the boot - we
356 : * *do* need to make sure our writes to binaries on initramfs has
357 : * not left us with opened struct file waiting for __fput() - execve()
358 : * won't work without that. Please, don't add more callers without
359 : * very good reasons; in particular, never call that with locks
360 : * held and never call that from a thread that might need to do
361 : * some work on any kind of umount.
362 : */
363 0 : void flush_delayed_fput(void)
364 : {
365 0 : delayed_fput(NULL);
366 0 : }
367 : EXPORT_SYMBOL_GPL(flush_delayed_fput);
368 :
369 : static DECLARE_DELAYED_WORK(delayed_fput_work, delayed_fput);
370 :
371 0 : void fput_many(struct file *file, unsigned int refs)
372 : {
373 0 : if (atomic_long_sub_and_test(refs, &file->f_count)) {
374 0 : struct task_struct *task = current;
375 :
376 0 : if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
377 0 : init_task_work(&file->f_u.fu_rcuhead, ____fput);
378 0 : if (!task_work_add(task, &file->f_u.fu_rcuhead, TWA_RESUME))
379 : return;
380 : /*
381 : * After this task has run exit_task_work(),
382 : * task_work_add() will fail. Fall through to delayed
383 : * fput to avoid leaking *file.
384 : */
385 : }
386 :
387 0 : if (llist_add(&file->f_u.fu_llist, &delayed_fput_list))
388 : schedule_delayed_work(&delayed_fput_work, 1);
389 : }
390 : }
391 :
392 0 : void fput(struct file *file)
393 : {
394 0 : fput_many(file, 1);
395 0 : }
396 :
397 : /*
398 : * synchronous analog of fput(); for kernel threads that might be needed
399 : * in some umount() (and thus can't use flush_delayed_fput() without
400 : * risking deadlocks), need to wait for completion of __fput() and know
401 : * for this specific struct file it won't involve anything that would
402 : * need them. Use only if you really need it - at the very least,
403 : * don't blindly convert fput() by kernel thread to that.
404 : */
405 0 : void __fput_sync(struct file *file)
406 : {
407 0 : if (atomic_long_dec_and_test(&file->f_count)) {
408 0 : struct task_struct *task = current;
409 0 : BUG_ON(!(task->flags & PF_KTHREAD));
410 0 : __fput(file);
411 : }
412 0 : }
413 :
414 : EXPORT_SYMBOL(fput);
415 : EXPORT_SYMBOL(__fput_sync);
416 :
417 1 : void __init files_init(void)
418 : {
419 1 : filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
420 : SLAB_HWCACHE_ALIGN | SLAB_PANIC | SLAB_ACCOUNT, NULL);
421 2 : percpu_counter_init(&nr_files, 0, GFP_KERNEL);
422 1 : }
423 :
424 : /*
425 : * One file with associated inode and dcache is very roughly 1K. Per default
426 : * do not use more than 10% of our memory for files.
427 : */
428 1 : void __init files_maxfiles_init(void)
429 : {
430 : unsigned long n;
431 1 : unsigned long nr_pages = totalram_pages();
432 1 : unsigned long memreserve = (nr_pages - nr_free_pages()) * 3/2;
433 :
434 1 : memreserve = min(memreserve, nr_pages - 1);
435 1 : n = ((nr_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
436 :
437 1 : files_stat.max_files = max_t(unsigned long, n, NR_FILE);
438 1 : }
|
