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/ima.h>
28 #include <linux/atomic.h>
32 /* sysctl tunables... */
33 struct files_stat_struct files_stat = {
37 DEFINE_LGLOCK(files_lglock);
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 inline 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);
56 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
60 * Return the total number of open files in the system
62 static long get_nr_files(void)
64 return percpu_counter_read_positive(&nr_files);
68 * Return the maximum number of open files in the system
70 unsigned long get_max_files(void)
72 return files_stat.max_files;
74 EXPORT_SYMBOL_GPL(get_max_files);
77 * Handle nr_files sysctl
79 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
80 int proc_nr_files(ctl_table *table, int write,
81 void __user *buffer, size_t *lenp, loff_t *ppos)
83 files_stat.nr_files = get_nr_files();
84 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
87 int proc_nr_files(ctl_table *table, int write,
88 void __user *buffer, size_t *lenp, loff_t *ppos)
94 /* Find an unused file structure and return a pointer to it.
95 * Returns NULL, if there are no more free file structures or
96 * we run out of memory.
98 * Be very careful using this. You are responsible for
99 * getting write access to any mount that you might assign
100 * to this filp, if it is opened for write. If this is not
101 * done, you will imbalance int the mount's writer count
102 * and a warning at __fput() time.
104 struct file *get_empty_filp(void)
106 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);
126 percpu_counter_inc(&nr_files);
127 f->f_cred = get_cred(cred);
128 if (security_file_alloc(f))
131 INIT_LIST_HEAD(&f->f_u.fu_list);
132 atomic_long_set(&f->f_count, 1);
133 rwlock_init(&f->f_owner.lock);
134 spin_lock_init(&f->f_lock);
135 eventpoll_init_file(f);
136 /* f->f_version: 0 */
140 /* Ran out of filps - report that */
141 if (get_nr_files() > old_max) {
142 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
143 old_max = get_nr_files();
154 * alloc_file - allocate and initialize a 'struct file'
155 * @mnt: the vfsmount on which the file will reside
156 * @dentry: the dentry representing the new file
157 * @mode: the mode with which the new file will be opened
158 * @fop: the 'struct file_operations' for the new file
160 * Use this instead of get_empty_filp() to get a new
161 * 'struct file'. Do so because of the same initialization
162 * pitfalls reasons listed for init_file(). This is a
163 * preferred interface to using init_file().
165 * If all the callers of init_file() are eliminated, its
166 * code should be moved into this function.
168 struct file *alloc_file(struct path *path, fmode_t mode,
169 const struct file_operations *fop)
173 file = get_empty_filp();
177 file->f_path = *path;
178 file->f_mapping = path->dentry->d_inode->i_mapping;
183 * These mounts don't really matter in practice
184 * for r/o bind mounts. They aren't userspace-
185 * visible. We do this for consistency, and so
186 * that we can do debugging checks at __fput()
188 if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
189 file_take_write(file);
190 WARN_ON(mnt_clone_write(path->mnt));
192 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
193 i_readcount_inc(path->dentry->d_inode);
196 EXPORT_SYMBOL(alloc_file);
199 * drop_file_write_access - give up ability to write to a file
200 * @file: the file to which we will stop writing
202 * This is a central place which will give up the ability
203 * to write to @file, along with access to write through
206 void drop_file_write_access(struct file *file)
208 struct vfsmount *mnt = file->f_path.mnt;
209 struct dentry *dentry = file->f_path.dentry;
210 struct inode *inode = dentry->d_inode;
212 put_write_access(inode);
214 if (special_file(inode->i_mode))
216 if (file_check_writeable(file) != 0)
219 file_release_write(file);
221 EXPORT_SYMBOL_GPL(drop_file_write_access);
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);
245 if (file->f_op && file->f_op->release)
246 file->f_op->release(inode, file);
247 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 file_sb_list_del(file);
256 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
257 i_readcount_dec(inode);
258 if (file->f_mode & FMODE_WRITE)
259 drop_file_write_access(file);
260 file->f_path.dentry = NULL;
261 file->f_path.mnt = NULL;
267 void fput(struct file *file)
269 if (atomic_long_dec_and_test(&file->f_count))
275 struct file *fget(unsigned int fd)
278 struct files_struct *files = current->files;
281 file = fcheck_files(files, fd);
283 /* File object ref couldn't be taken */
284 if (file->f_mode & FMODE_PATH ||
285 !atomic_long_inc_not_zero(&file->f_count))
295 struct file *fget_raw(unsigned int fd)
298 struct files_struct *files = current->files;
301 file = fcheck_files(files, fd);
303 /* File object ref couldn't be taken */
304 if (!atomic_long_inc_not_zero(&file->f_count))
312 EXPORT_SYMBOL(fget_raw);
315 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
317 * You can use this instead of fget if you satisfy all of the following
319 * 1) You must call fput_light before exiting the syscall and returning control
320 * to userspace (i.e. you cannot remember the returned struct file * after
321 * returning to userspace).
322 * 2) You must not call filp_close on the returned struct file * in between
323 * calls to fget_light and fput_light.
324 * 3) You must not clone the current task in between the calls to fget_light
327 * The fput_needed flag returned by fget_light should be passed to the
328 * corresponding fput_light.
330 struct file *fget_light(unsigned int fd, int *fput_needed)
333 struct files_struct *files = current->files;
336 if (atomic_read(&files->count) == 1) {
337 file = fcheck_files(files, fd);
338 if (file && (file->f_mode & FMODE_PATH))
342 file = fcheck_files(files, fd);
344 if (!(file->f_mode & FMODE_PATH) &&
345 atomic_long_inc_not_zero(&file->f_count))
348 /* Didn't get the reference, someone's freed */
357 struct file *fget_raw_light(unsigned int fd, int *fput_needed)
360 struct files_struct *files = current->files;
363 if (atomic_read(&files->count) == 1) {
364 file = fcheck_files(files, fd);
367 file = fcheck_files(files, fd);
369 if (atomic_long_inc_not_zero(&file->f_count))
372 /* Didn't get the reference, someone's freed */
381 void put_filp(struct file *file)
383 if (atomic_long_dec_and_test(&file->f_count)) {
384 security_file_free(file);
385 file_sb_list_del(file);
390 static inline int file_list_cpu(struct file *file)
393 return file->f_sb_list_cpu;
395 return smp_processor_id();
399 /* helper for file_sb_list_add to reduce ifdefs */
400 static inline void __file_sb_list_add(struct file *file, struct super_block *sb)
402 struct list_head *list;
405 cpu = smp_processor_id();
406 file->f_sb_list_cpu = cpu;
407 list = per_cpu_ptr(sb->s_files, cpu);
411 list_add(&file->f_u.fu_list, list);
415 * file_sb_list_add - add a file to the sb's file list
417 * @sb: sb to add it to
419 * Use this function to associate a file with the superblock of the inode it
422 void file_sb_list_add(struct file *file, struct super_block *sb)
424 lg_local_lock(&files_lglock);
425 __file_sb_list_add(file, sb);
426 lg_local_unlock(&files_lglock);
430 * file_sb_list_del - remove a file from the sb's file list
431 * @file: file to remove
432 * @sb: sb to remove it from
434 * Use this function to remove a file from its superblock.
436 void file_sb_list_del(struct file *file)
438 if (!list_empty(&file->f_u.fu_list)) {
439 lg_local_lock_cpu(&files_lglock, file_list_cpu(file));
440 list_del_init(&file->f_u.fu_list);
441 lg_local_unlock_cpu(&files_lglock, file_list_cpu(file));
448 * These macros iterate all files on all CPUs for a given superblock.
449 * files_lglock must be held globally.
451 #define do_file_list_for_each_entry(__sb, __file) \
454 for_each_possible_cpu(i) { \
455 struct list_head *list; \
456 list = per_cpu_ptr((__sb)->s_files, i); \
457 list_for_each_entry((__file), list, f_u.fu_list)
459 #define while_file_list_for_each_entry \
465 #define do_file_list_for_each_entry(__sb, __file) \
467 struct list_head *list; \
468 list = &(sb)->s_files; \
469 list_for_each_entry((__file), list, f_u.fu_list)
471 #define while_file_list_for_each_entry \
476 int fs_may_remount_ro(struct super_block *sb)
479 /* Check that no files are currently opened for writing. */
480 lg_global_lock(&files_lglock);
481 do_file_list_for_each_entry(sb, file) {
482 struct inode *inode = file->f_path.dentry->d_inode;
484 /* File with pending delete? */
485 if (inode->i_nlink == 0)
488 /* Writeable file? */
489 if (S_ISREG(inode->i_mode) && (file->f_mode & FMODE_WRITE))
491 } while_file_list_for_each_entry;
492 lg_global_unlock(&files_lglock);
493 return 1; /* Tis' cool bro. */
495 lg_global_unlock(&files_lglock);
500 * mark_files_ro - mark all files read-only
501 * @sb: superblock in question
503 * All files are marked read-only. We don't care about pending
504 * delete files so this should be used in 'force' mode only.
506 void mark_files_ro(struct super_block *sb)
511 lg_global_lock(&files_lglock);
512 do_file_list_for_each_entry(sb, f) {
513 struct vfsmount *mnt;
514 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
518 if (!(f->f_mode & FMODE_WRITE))
520 spin_lock(&f->f_lock);
521 f->f_mode &= ~FMODE_WRITE;
522 spin_unlock(&f->f_lock);
523 if (file_check_writeable(f) != 0)
525 file_release_write(f);
526 mnt = mntget(f->f_path.mnt);
527 /* This can sleep, so we can't hold the spinlock. */
528 lg_global_unlock(&files_lglock);
532 } while_file_list_for_each_entry;
533 lg_global_unlock(&files_lglock);
536 void __init files_init(unsigned long mempages)
540 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
541 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
544 * One file with associated inode and dcache is very roughly 1K.
545 * Per default don't use more than 10% of our memory for files.
548 n = (mempages * (PAGE_SIZE / 1024)) / 10;
549 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
551 lg_lock_init(&files_lglock, "files_lglock");
552 percpu_counter_init(&nr_files, 0);