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 DECLARE_LGLOCK(files_lglock);
38 DEFINE_LGLOCK(files_lglock);
40 /* SLAB cache for file structures */
41 static struct kmem_cache *filp_cachep __read_mostly;
43 static struct percpu_counter nr_files __cacheline_aligned_in_smp;
45 static inline void file_free_rcu(struct rcu_head *head)
47 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
50 kmem_cache_free(filp_cachep, f);
53 static inline void file_free(struct file *f)
55 percpu_counter_dec(&nr_files);
57 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
61 * Return the total number of open files in the system
63 static long get_nr_files(void)
65 return percpu_counter_read_positive(&nr_files);
69 * Return the maximum number of open files in the system
71 unsigned long get_max_files(void)
73 return files_stat.max_files;
75 EXPORT_SYMBOL_GPL(get_max_files);
78 * Handle nr_files sysctl
80 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
81 int proc_nr_files(ctl_table *table, int write,
82 void __user *buffer, size_t *lenp, loff_t *ppos)
84 files_stat.nr_files = get_nr_files();
85 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
88 int proc_nr_files(ctl_table *table, int write,
89 void __user *buffer, size_t *lenp, loff_t *ppos)
95 /* Find an unused file structure and return a pointer to it.
96 * Returns NULL, if there are no more free file structures or
97 * we run out of memory.
99 * Be very careful using this. You are responsible for
100 * getting write access to any mount that you might assign
101 * to this filp, if it is opened for write. If this is not
102 * done, you will imbalance int the mount's writer count
103 * and a warning at __fput() time.
105 struct file *get_empty_filp(void)
107 const struct cred *cred = current_cred();
112 * Privileged users can go above max_files
114 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
116 * percpu_counters are inaccurate. Do an expensive check before
119 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
123 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
127 percpu_counter_inc(&nr_files);
128 f->f_cred = get_cred(cred);
129 if (security_file_alloc(f))
132 INIT_LIST_HEAD(&f->f_u.fu_list);
133 atomic_long_set(&f->f_count, 1);
134 rwlock_init(&f->f_owner.lock);
135 spin_lock_init(&f->f_lock);
136 eventpoll_init_file(f);
137 /* f->f_version: 0 */
141 /* Ran out of filps - report that */
142 if (get_nr_files() > old_max) {
143 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
144 old_max = get_nr_files();
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 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)
220 file_release_write(file);
222 EXPORT_SYMBOL_GPL(drop_file_write_access);
224 /* the real guts of fput() - releasing the last reference to file
226 static void __fput(struct file *file)
228 struct dentry *dentry = file->f_path.dentry;
229 struct vfsmount *mnt = file->f_path.mnt;
230 struct inode *inode = dentry->d_inode;
234 fsnotify_close(file);
236 * The function eventpoll_release() should be the first called
237 * in the file cleanup chain.
239 eventpoll_release(file);
240 locks_remove_flock(file);
242 if (unlikely(file->f_flags & FASYNC)) {
243 if (file->f_op && file->f_op->fasync)
244 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);
250 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
251 !(file->f_mode & FMODE_PATH))) {
252 cdev_put(inode->i_cdev);
254 fops_put(file->f_op);
255 put_pid(file->f_owner.pid);
256 file_sb_list_del(file);
257 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
258 i_readcount_dec(inode);
259 if (file->f_mode & FMODE_WRITE)
260 drop_file_write_access(file);
261 file->f_path.dentry = NULL;
262 file->f_path.mnt = NULL;
268 void fput(struct file *file)
270 if (atomic_long_dec_and_test(&file->f_count))
276 struct file *fget(unsigned int fd)
279 struct files_struct *files = current->files;
282 file = fcheck_files(files, fd);
284 /* File object ref couldn't be taken */
285 if (file->f_mode & FMODE_PATH ||
286 !atomic_long_inc_not_zero(&file->f_count))
296 struct file *fget_raw(unsigned int fd)
299 struct files_struct *files = current->files;
302 file = fcheck_files(files, fd);
304 /* File object ref couldn't be taken */
305 if (!atomic_long_inc_not_zero(&file->f_count))
313 EXPORT_SYMBOL(fget_raw);
316 * Lightweight file lookup - no refcnt increment if fd table isn't shared.
318 * You can use this instead of fget if you satisfy all of the following
320 * 1) You must call fput_light before exiting the syscall and returning control
321 * to userspace (i.e. you cannot remember the returned struct file * after
322 * returning to userspace).
323 * 2) You must not call filp_close on the returned struct file * in between
324 * calls to fget_light and fput_light.
325 * 3) You must not clone the current task in between the calls to fget_light
328 * The fput_needed flag returned by fget_light should be passed to the
329 * corresponding fput_light.
331 struct file *fget_light(unsigned int fd, int *fput_needed)
334 struct files_struct *files = current->files;
337 if (atomic_read(&files->count) == 1) {
338 file = fcheck_files(files, fd);
339 if (file && (file->f_mode & FMODE_PATH))
343 file = fcheck_files(files, fd);
345 if (!(file->f_mode & FMODE_PATH) &&
346 atomic_long_inc_not_zero(&file->f_count))
349 /* Didn't get the reference, someone's freed */
358 struct file *fget_raw_light(unsigned int fd, int *fput_needed)
361 struct files_struct *files = current->files;
364 if (atomic_read(&files->count) == 1) {
365 file = fcheck_files(files, fd);
368 file = fcheck_files(files, fd);
370 if (atomic_long_inc_not_zero(&file->f_count))
373 /* Didn't get the reference, someone's freed */
382 void put_filp(struct file *file)
384 if (atomic_long_dec_and_test(&file->f_count)) {
385 security_file_free(file);
386 file_sb_list_del(file);
391 static inline int file_list_cpu(struct file *file)
394 return file->f_sb_list_cpu;
396 return smp_processor_id();
400 /* helper for file_sb_list_add to reduce ifdefs */
401 static inline void __file_sb_list_add(struct file *file, struct super_block *sb)
403 struct list_head *list;
406 cpu = smp_processor_id();
407 file->f_sb_list_cpu = cpu;
408 list = per_cpu_ptr(sb->s_files, cpu);
412 list_add(&file->f_u.fu_list, list);
416 * file_sb_list_add - add a file to the sb's file list
418 * @sb: sb to add it to
420 * Use this function to associate a file with the superblock of the inode it
423 void file_sb_list_add(struct file *file, struct super_block *sb)
425 lg_local_lock(files_lglock);
426 __file_sb_list_add(file, sb);
427 lg_local_unlock(files_lglock);
431 * file_sb_list_del - remove a file from the sb's file list
432 * @file: file to remove
433 * @sb: sb to remove it from
435 * Use this function to remove a file from its superblock.
437 void file_sb_list_del(struct file *file)
439 if (!list_empty(&file->f_u.fu_list)) {
440 lg_local_lock_cpu(files_lglock, file_list_cpu(file));
441 list_del_init(&file->f_u.fu_list);
442 lg_local_unlock_cpu(files_lglock, file_list_cpu(file));
449 * These macros iterate all files on all CPUs for a given superblock.
450 * files_lglock must be held globally.
452 #define do_file_list_for_each_entry(__sb, __file) \
455 for_each_possible_cpu(i) { \
456 struct list_head *list; \
457 list = per_cpu_ptr((__sb)->s_files, i); \
458 list_for_each_entry((__file), list, f_u.fu_list)
460 #define while_file_list_for_each_entry \
466 #define do_file_list_for_each_entry(__sb, __file) \
468 struct list_head *list; \
469 list = &(sb)->s_files; \
470 list_for_each_entry((__file), list, f_u.fu_list)
472 #define while_file_list_for_each_entry \
478 * mark_files_ro - mark all files read-only
479 * @sb: superblock in question
481 * All files are marked read-only. We don't care about pending
482 * delete files so this should be used in 'force' mode only.
484 void mark_files_ro(struct super_block *sb)
489 lg_global_lock(files_lglock);
490 do_file_list_for_each_entry(sb, f) {
491 struct vfsmount *mnt;
492 if (!S_ISREG(f->f_path.dentry->d_inode->i_mode))
496 if (!(f->f_mode & FMODE_WRITE))
498 spin_lock(&f->f_lock);
499 f->f_mode &= ~FMODE_WRITE;
500 spin_unlock(&f->f_lock);
501 if (file_check_writeable(f) != 0)
503 file_release_write(f);
504 mnt = mntget(f->f_path.mnt);
505 /* This can sleep, so we can't hold the spinlock. */
506 lg_global_unlock(files_lglock);
510 } while_file_list_for_each_entry;
511 lg_global_unlock(files_lglock);
514 void __init files_init(unsigned long mempages)
518 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
519 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
522 * One file with associated inode and dcache is very roughly 1K.
523 * Per default don't use more than 10% of our memory for files.
526 n = (mempages * (PAGE_SIZE / 1024)) / 10;
527 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
529 lg_lock_init(files_lglock);
530 percpu_counter_init(&nr_files, 0);