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/lglock.h>
24 #include <linux/percpu_counter.h>
25 #include <linux/percpu.h>
26 #include <linux/hardirq.h>
27 #include <linux/task_work.h>
28 #include <linux/ima.h>
30 #include <linux/atomic.h>
34 /* sysctl tunables... */
35 struct files_stat_struct files_stat = {
39 DEFINE_STATIC_LGLOCK(files_lglock);
41 /* SLAB cache for file structures */
42 static struct kmem_cache *filp_cachep __read_mostly;
44 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
46 static void file_free_rcu(struct rcu_head *head)
48 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
51 kmem_cache_free(filp_cachep, f);
54 static inline void file_free(struct file *f)
56 percpu_counter_dec(&nr_files);
58 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
62 * Return the total number of open files in the system
64 static long get_nr_files(void)
66 return percpu_counter_read_positive(&nr_files);
70 * Return the maximum number of open files in the system
72 unsigned long get_max_files(void)
74 return files_stat.max_files;
76 EXPORT_SYMBOL_GPL(get_max_files);
79 * Handle nr_files sysctl
81 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
82 int proc_nr_files(ctl_table *table, int write,
83 void __user *buffer, size_t *lenp, loff_t *ppos)
85 files_stat.nr_files = get_nr_files();
86 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
89 int proc_nr_files(ctl_table *table, int write,
90 void __user *buffer, size_t *lenp, loff_t *ppos)
96 /* Find an unused file structure and return a pointer to it.
97 * Returns an error pointer if some error happend e.g. we over file
98 * structures limit, run out of memory or operation is not permitted.
100 * Be very careful using this. You are responsible for
101 * getting write access to any mount that you might assign
102 * to this filp, if it is opened for write. If this is not
103 * done, you will imbalance int the mount's writer count
104 * and a warning at __fput() time.
106 struct file *get_empty_filp(void)
108 const struct cred *cred = current_cred();
114 * Privileged users can go above max_files
116 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
118 * percpu_counters are inaccurate. Do an expensive check before
121 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
125 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
127 return ERR_PTR(-ENOMEM);
129 percpu_counter_inc(&nr_files);
130 f->f_cred = get_cred(cred);
131 error = security_file_alloc(f);
132 if (unlikely(error)) {
134 return ERR_PTR(error);
137 INIT_LIST_HEAD(&f->f_u.fu_list);
138 atomic_long_set(&f->f_count, 1);
139 rwlock_init(&f->f_owner.lock);
140 spin_lock_init(&f->f_lock);
141 eventpoll_init_file(f);
142 /* f->f_version: 0 */
146 /* Ran out of filps - report that */
147 if (get_nr_files() > old_max) {
148 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
149 old_max = get_nr_files();
151 return ERR_PTR(-ENFILE);
155 * alloc_file - allocate and initialize a 'struct file'
156 * @mnt: the vfsmount on which the file will reside
157 * @dentry: the dentry representing the new file
158 * @mode: the mode with which the new file will be opened
159 * @fop: the 'struct file_operations' for the new file
161 * Use this instead of get_empty_filp() to get a new
162 * 'struct file'. Do so because of the same initialization
163 * pitfalls reasons listed for init_file(). This is a
164 * preferred interface to using init_file().
166 * If all the callers of init_file() are eliminated, its
167 * code should be moved into this function.
169 struct file *alloc_file(struct path *path, fmode_t mode,
170 const struct file_operations *fop)
174 file = get_empty_filp();
178 file->f_path = *path;
179 file->f_mapping = path->dentry->d_inode->i_mapping;
184 * These mounts don't really matter in practice
185 * for r/o bind mounts. They aren't userspace-
186 * visible. We do this for consistency, and so
187 * that we can do debugging checks at __fput()
189 if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
190 file_take_write(file);
191 WARN_ON(mnt_clone_write(path->mnt));
193 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
194 i_readcount_inc(path->dentry->d_inode);
197 EXPORT_SYMBOL(alloc_file);
200 * drop_file_write_access - give up ability to write to a file
201 * @file: the file to which we will stop writing
203 * This is a central place which will give up the ability
204 * to write to @file, along with access to write through
207 static void drop_file_write_access(struct file *file)
209 struct vfsmount *mnt = file->f_path.mnt;
210 struct dentry *dentry = file->f_path.dentry;
211 struct inode *inode = dentry->d_inode;
213 put_write_access(inode);
215 if (special_file(inode->i_mode))
217 if (file_check_writeable(file) != 0)
219 __mnt_drop_write(mnt);
220 file_release_write(file);
223 /* the real guts of fput() - releasing the last reference to file
225 static void __fput(struct file *file)
227 struct dentry *dentry = file->f_path.dentry;
228 struct vfsmount *mnt = file->f_path.mnt;
229 struct inode *inode = dentry->d_inode;
233 fsnotify_close(file);
235 * The function eventpoll_release() should be the first called
236 * in the file cleanup chain.
238 eventpoll_release(file);
239 locks_remove_flock(file);
241 if (unlikely(file->f_flags & FASYNC)) {
242 if (file->f_op && file->f_op->fasync)
243 file->f_op->fasync(-1, file, 0);
246 if (file->f_op && file->f_op->release)
247 file->f_op->release(inode, file);
248 security_file_free(file);
249 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
250 !(file->f_mode & FMODE_PATH))) {
251 cdev_put(inode->i_cdev);
253 fops_put(file->f_op);
254 put_pid(file->f_owner.pid);
255 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
256 i_readcount_dec(inode);
257 if (file->f_mode & FMODE_WRITE)
258 drop_file_write_access(file);
259 file->f_path.dentry = NULL;
260 file->f_path.mnt = NULL;
266 static DEFINE_SPINLOCK(delayed_fput_lock);
267 static LIST_HEAD(delayed_fput_list);
268 static void delayed_fput(struct work_struct *unused)
271 spin_lock_irq(&delayed_fput_lock);
272 list_splice_init(&delayed_fput_list, &head);
273 spin_unlock_irq(&delayed_fput_lock);
274 while (!list_empty(&head)) {
275 struct file *f = list_first_entry(&head, struct file, f_u.fu_list);
276 list_del_init(&f->f_u.fu_list);
281 static void ____fput(struct callback_head *work)
283 __fput(container_of(work, struct file, f_u.fu_rcuhead));
287 * If kernel thread really needs to have the final fput() it has done
288 * to complete, call this. The only user right now is the boot - we
289 * *do* need to make sure our writes to binaries on initramfs has
290 * not left us with opened struct file waiting for __fput() - execve()
291 * won't work without that. Please, don't add more callers without
292 * very good reasons; in particular, never call that with locks
293 * held and never call that from a thread that might need to do
294 * some work on any kind of umount.
296 void flush_delayed_fput(void)
301 static DECLARE_WORK(delayed_fput_work, delayed_fput);
303 void fput(struct file *file)
305 if (atomic_long_dec_and_test(&file->f_count)) {
306 struct task_struct *task = current;
307 file_sb_list_del(file);
308 if (unlikely(in_interrupt() || task->flags & PF_KTHREAD)) {
310 spin_lock_irqsave(&delayed_fput_lock, flags);
311 list_add(&file->f_u.fu_list, &delayed_fput_list);
312 schedule_work(&delayed_fput_work);
313 spin_unlock_irqrestore(&delayed_fput_lock, flags);
316 init_task_work(&file->f_u.fu_rcuhead, ____fput);
317 task_work_add(task, &file->f_u.fu_rcuhead, true);
322 * synchronous analog of fput(); for kernel threads that might be needed
323 * in some umount() (and thus can't use flush_delayed_fput() without
324 * risking deadlocks), need to wait for completion of __fput() and know
325 * for this specific struct file it won't involve anything that would
326 * need them. Use only if you really need it - at the very least,
327 * don't blindly convert fput() by kernel thread to that.
329 void __fput_sync(struct file *file)
331 if (atomic_long_dec_and_test(&file->f_count)) {
332 struct task_struct *task = current;
333 file_sb_list_del(file);
334 BUG_ON(!(task->flags & PF_KTHREAD));
341 void put_filp(struct file *file)
343 if (atomic_long_dec_and_test(&file->f_count)) {
344 security_file_free(file);
345 file_sb_list_del(file);
350 static inline int file_list_cpu(struct file *file)
353 return file->f_sb_list_cpu;
355 return smp_processor_id();
359 /* helper for file_sb_list_add to reduce ifdefs */
360 static inline void __file_sb_list_add(struct file *file, struct super_block *sb)
362 struct list_head *list;
365 cpu = smp_processor_id();
366 file->f_sb_list_cpu = cpu;
367 list = per_cpu_ptr(sb->s_files, cpu);
371 list_add(&file->f_u.fu_list, list);
375 * file_sb_list_add - add a file to the sb's file list
377 * @sb: sb to add it to
379 * Use this function to associate a file with the superblock of the inode it
382 void file_sb_list_add(struct file *file, struct super_block *sb)
384 lg_local_lock(&files_lglock);
385 __file_sb_list_add(file, sb);
386 lg_local_unlock(&files_lglock);
390 * file_sb_list_del - remove a file from the sb's file list
391 * @file: file to remove
392 * @sb: sb to remove it from
394 * Use this function to remove a file from its superblock.
396 void file_sb_list_del(struct file *file)
398 if (!list_empty(&file->f_u.fu_list)) {
399 lg_local_lock_cpu(&files_lglock, file_list_cpu(file));
400 list_del_init(&file->f_u.fu_list);
401 lg_local_unlock_cpu(&files_lglock, file_list_cpu(file));
408 * These macros iterate all files on all CPUs for a given superblock.
409 * files_lglock must be held globally.
411 #define do_file_list_for_each_entry(__sb, __file) \
414 for_each_possible_cpu(i) { \
415 struct list_head *list; \
416 list = per_cpu_ptr((__sb)->s_files, i); \
417 list_for_each_entry((__file), list, f_u.fu_list)
419 #define while_file_list_for_each_entry \
425 #define do_file_list_for_each_entry(__sb, __file) \
427 struct list_head *list; \
428 list = &(sb)->s_files; \
429 list_for_each_entry((__file), list, f_u.fu_list)
431 #define while_file_list_for_each_entry \
437 * mark_files_ro - mark all files read-only
438 * @sb: superblock in question
440 * All files are marked read-only. We don't care about pending
441 * delete files so this should be used in 'force' mode only.
443 void mark_files_ro(struct super_block *sb)
447 lg_global_lock(&files_lglock);
448 do_file_list_for_each_entry(sb, f) {
449 if (!S_ISREG(file_inode(f)->i_mode))
453 if (!(f->f_mode & FMODE_WRITE))
455 spin_lock(&f->f_lock);
456 f->f_mode &= ~FMODE_WRITE;
457 spin_unlock(&f->f_lock);
458 if (file_check_writeable(f) != 0)
460 __mnt_drop_write(f->f_path.mnt);
461 file_release_write(f);
462 } while_file_list_for_each_entry;
463 lg_global_unlock(&files_lglock);
466 void __init files_init(unsigned long mempages)
470 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
471 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
474 * One file with associated inode and dcache is very roughly 1K.
475 * Per default don't use more than 10% of our memory for files.
478 n = (mempages * (PAGE_SIZE / 1024)) / 10;
479 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
481 lg_lock_init(&files_lglock, "files_lglock");
482 percpu_counter_init(&nr_files, 0);