2 * linux/fs/file_table.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
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>
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/percpu_counter.h>
24 #include <linux/percpu.h>
25 #include <linux/hardirq.h>
26 #include <linux/task_work.h>
27 #include <linux/ima.h>
28 #include <linux/swap.h>
30 #include <linux/atomic.h>
34 /* sysctl tunables... */
35 struct files_stat_struct files_stat = {
39 /* SLAB cache for file structures */
40 static struct kmem_cache *filp_cachep __read_mostly;
42 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
44 static void file_free_rcu(struct rcu_head *head)
46 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
49 kmem_cache_free(filp_cachep, f);
52 static inline void file_free(struct file *f)
54 percpu_counter_dec(&nr_files);
55 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
59 * Return the total number of open files in the system
61 static long get_nr_files(void)
63 return percpu_counter_read_positive(&nr_files);
67 * Return the maximum number of open files in the system
69 unsigned long get_max_files(void)
71 return files_stat.max_files;
73 EXPORT_SYMBOL_GPL(get_max_files);
76 * Handle nr_files sysctl
78 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
79 int proc_nr_files(struct ctl_table *table, int write,
80 void __user *buffer, size_t *lenp, loff_t *ppos)
82 files_stat.nr_files = get_nr_files();
83 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
86 int proc_nr_files(struct ctl_table *table, int write,
87 void __user *buffer, size_t *lenp, loff_t *ppos)
93 /* Find an unused file structure and return a pointer to it.
94 * Returns an error pointer if some error happend e.g. we over file
95 * structures limit, run out of memory or operation is not permitted.
97 * Be very careful using this. You are responsible for
98 * getting write access to any mount that you might assign
99 * to this filp, if it is opened for write. If this is not
100 * done, you will imbalance int the mount's writer count
101 * and a warning at __fput() time.
103 struct file *get_empty_filp(void)
105 const struct cred *cred = current_cred();
111 * Privileged users can go above max_files
113 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
115 * percpu_counters are inaccurate. Do an expensive check before
118 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
122 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
124 return ERR_PTR(-ENOMEM);
126 percpu_counter_inc(&nr_files);
127 f->f_cred = get_cred(cred);
128 error = security_file_alloc(f);
129 if (unlikely(error)) {
131 return ERR_PTR(error);
134 atomic_long_set(&f->f_count, 1);
135 rwlock_init(&f->f_owner.lock);
136 spin_lock_init(&f->f_lock);
137 mutex_init(&f->f_pos_lock);
138 eventpoll_init_file(f);
139 /* f->f_version: 0 */
143 /* Ran out of filps - report that */
144 if (get_nr_files() > old_max) {
145 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
146 old_max = get_nr_files();
148 return ERR_PTR(-ENFILE);
152 * alloc_file - allocate and initialize a 'struct file'
154 * @path: the (dentry, vfsmount) pair for the new file
155 * @mode: the mode with which the new file will be opened
156 * @fop: the 'struct file_operations' for the new file
158 struct file *alloc_file(struct path *path, fmode_t mode,
159 const struct file_operations *fop)
163 file = get_empty_filp();
167 file->f_path = *path;
168 file->f_inode = path->dentry->d_inode;
169 file->f_mapping = path->dentry->d_inode->i_mapping;
170 if ((mode & FMODE_READ) &&
171 likely(fop->read || fop->read_iter))
172 mode |= FMODE_CAN_READ;
173 if ((mode & FMODE_WRITE) &&
174 likely(fop->write || fop->write_iter))
175 mode |= FMODE_CAN_WRITE;
178 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
179 i_readcount_inc(path->dentry->d_inode);
182 EXPORT_SYMBOL(alloc_file);
184 /* the real guts of fput() - releasing the last reference to file
186 static void __fput(struct file *file)
188 struct dentry *dentry = file->f_path.dentry;
189 struct vfsmount *mnt = file->f_path.mnt;
190 struct inode *inode = file->f_inode;
194 fsnotify_close(file);
196 * The function eventpoll_release() should be the first called
197 * in the file cleanup chain.
199 eventpoll_release(file);
200 locks_remove_file(file);
202 if (unlikely(file->f_flags & FASYNC)) {
203 if (file->f_op->fasync)
204 file->f_op->fasync(-1, file, 0);
207 if (file->f_op->release)
208 file->f_op->release(inode, file);
209 security_file_free(file);
210 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
211 !(file->f_mode & FMODE_PATH))) {
212 cdev_put(inode->i_cdev);
214 fops_put(file->f_op);
215 put_pid(file->f_owner.pid);
216 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
217 i_readcount_dec(inode);
218 if (file->f_mode & FMODE_WRITER) {
219 put_write_access(inode);
220 __mnt_drop_write(mnt);
222 file->f_path.dentry = NULL;
223 file->f_path.mnt = NULL;
224 file->f_inode = NULL;
230 static LLIST_HEAD(global_fput_list);
231 static void global_fput(struct work_struct *unused)
233 struct llist_node *node = llist_del_all(&global_fput_list);
234 struct llist_node *next;
236 for (; node; node = next) {
237 next = llist_next(node);
238 __fput(llist_entry(node, struct file, f_u.fu_llist));
242 static void ____fput(struct callback_head *work)
244 __fput(container_of(work, struct file, f_u.fu_rcuhead));
247 static DECLARE_DELAYED_WORK(global_fput_work, global_fput);
250 * fput_global_flush - ensure that all global_fput work is complete
252 * If a kernel thread really needs to have the final fput() it has done to
253 * complete, call this. One of the main users is the boot - we *do* need to
254 * make sure our writes to binaries on initramfs has not left us with opened
255 * struct file waiting for __fput() - execve() won't work without that.
257 * Please, don't add more callers without very good reasons; in particular,
258 * never call that with locks held and never from a thread that might need to
259 * do some work on any kind of umount.
261 void fput_global_flush(void)
263 flush_delayed_work(&global_fput_work);
265 EXPORT_SYMBOL(fput_global_flush);
268 * fput - put a struct file reference
269 * @file: file of which to put the reference
271 * This function decrements the reference count for the struct file reference,
272 * and queues it up for destruction if the count goes to zero. In the case of
273 * most tasks we queue it to the task_work infrastructure, which will be run
274 * just before the task returns back to userspace. kthreads however never
275 * return to userspace, so for those we add them to a global list and schedule
276 * a delayed workqueue job to do the final cleanup work.
278 * Why not just do it synchronously? __fput can involve taking locks of all
279 * sorts, and doing it synchronously means that the callers must take extra care
280 * not to deadlock. That can be very difficult to ensure, so by deferring it
281 * until just before return to userland or to the workqueue, we sidestep that
282 * nastiness. Also, __fput can be quite stack intensive, so doing a final fput
283 * has the possibility of blowing up if we don't take steps to ensure that we
284 * have enough stack space to make it work.
286 void fput(struct file *file)
288 if (atomic_long_dec_and_test(&file->f_count)) {
289 struct task_struct *task = current;
291 if (likely(!in_interrupt() && !(task->flags & PF_KTHREAD))) {
292 init_task_work(&file->f_u.fu_rcuhead, ____fput);
293 if (!task_work_add(task, &file->f_u.fu_rcuhead, true))
296 * After this task has run exit_task_work(),
297 * task_work_add() will fail. Fall through to delayed
298 * fput to avoid leaking *file.
302 if (llist_add(&file->f_u.fu_llist, &global_fput_list))
303 schedule_delayed_work(&global_fput_work, 1);
309 * fput_global - do an fput without using task_work
310 * @file: file of which to put the reference
312 * When fput is called in the context of a userland process, it'll queue the
313 * actual work (__fput()) to be done just before returning to userland. In some
314 * cases however, we need to ensure that the __fput runs before that point.
316 * There is no safe way to flush work that has been queued via task_work_add
317 * however, so to do this we borrow the global_fput infrastructure that
318 * kthreads use. The userland process can use fput_global() on one or more
319 * struct files and then call fput_global_flush() to ensure that they are
320 * completely closed before proceeding.
322 * The main user is nfsd, which uses this to ensure that all cached but
323 * otherwise unused files are closed to allow a userland request for a lease
326 * Returns true if the final fput was done, false otherwise. The caller can
327 * use this to determine whether to fput_global_flush afterward.
329 bool fput_global(struct file *file)
331 if (atomic_long_dec_and_test(&file->f_count)) {
332 if (llist_add(&file->f_u.fu_llist, &global_fput_list))
333 schedule_delayed_work(&global_fput_work, 1);
338 EXPORT_SYMBOL(fput_global);
341 * synchronous analog of fput(); for kernel threads that might be needed
342 * in some umount() (and thus can't use fput_global_flush() without
343 * risking deadlocks), need to wait for completion of __fput() and know
344 * for this specific struct file it won't involve anything that would
345 * need them. Use only if you really need it - at the very least,
346 * don't blindly convert fput() by kernel thread to that.
348 void __fput_sync(struct file *file)
350 if (atomic_long_dec_and_test(&file->f_count)) {
351 struct task_struct *task = current;
352 BUG_ON(!(task->flags & PF_KTHREAD));
358 void put_filp(struct file *file)
360 if (atomic_long_dec_and_test(&file->f_count)) {
361 security_file_free(file);
366 void __init files_init(void)
368 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
369 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
370 percpu_counter_init(&nr_files, 0, GFP_KERNEL);
374 * One file with associated inode and dcache is very roughly 1K. Per default
375 * do not use more than 10% of our memory for files.
377 void __init files_maxfiles_init(void)
380 unsigned long memreserve = (totalram_pages - nr_free_pages()) * 3/2;
382 memreserve = min(memreserve, totalram_pages - 1);
383 n = ((totalram_pages - memreserve) * (PAGE_SIZE / 1024)) / 10;
385 files_stat.max_files = max_t(unsigned long, n, NR_FILE);