]> git.proxmox.com Git - mirror_ubuntu-eoan-kernel.git/blame - fs/file_table.c
9p: if v9fs_fid_lookup() gets to asking server, it'd better have hashed dentry
[mirror_ubuntu-eoan-kernel.git] / fs / file_table.c
CommitLineData
1da177e4
LT
1/*
2 * linux/fs/file_table.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu)
6 */
7
8#include <linux/string.h>
9#include <linux/slab.h>
10#include <linux/file.h>
9f3acc31 11#include <linux/fdtable.h>
1da177e4
LT
12#include <linux/init.h>
13#include <linux/module.h>
1da177e4
LT
14#include <linux/fs.h>
15#include <linux/security.h>
16#include <linux/eventpoll.h>
ab2af1f5 17#include <linux/rcupdate.h>
1da177e4 18#include <linux/mount.h>
16f7e0fe 19#include <linux/capability.h>
1da177e4 20#include <linux/cdev.h>
0eeca283 21#include <linux/fsnotify.h>
529bf6be 22#include <linux/sysctl.h>
6416ccb7 23#include <linux/lglock.h>
529bf6be 24#include <linux/percpu_counter.h>
6416ccb7 25#include <linux/percpu.h>
4a9d4b02
AV
26#include <linux/hardirq.h>
27#include <linux/task_work.h>
0552f879 28#include <linux/ima.h>
529bf6be 29
60063497 30#include <linux/atomic.h>
1da177e4 31
e81e3f4d
EP
32#include "internal.h"
33
1da177e4
LT
34/* sysctl tunables... */
35struct files_stat_struct files_stat = {
36 .max_files = NR_FILE
37};
38
4b2c551f 39DEFINE_STATIC_LGLOCK(files_lglock);
1da177e4 40
b6b3fdea
ED
41/* SLAB cache for file structures */
42static struct kmem_cache *filp_cachep __read_mostly;
43
529bf6be 44static struct percpu_counter nr_files __cacheline_aligned_in_smp;
1da177e4 45
5c33b183 46static void file_free_rcu(struct rcu_head *head)
1da177e4 47{
d76b0d9b
DH
48 struct file *f = container_of(head, struct file, f_u.fu_rcuhead);
49
50 put_cred(f->f_cred);
529bf6be 51 kmem_cache_free(filp_cachep, f);
1da177e4
LT
52}
53
529bf6be 54static inline void file_free(struct file *f)
1da177e4 55{
529bf6be 56 percpu_counter_dec(&nr_files);
ad775f5a 57 file_check_state(f);
529bf6be 58 call_rcu(&f->f_u.fu_rcuhead, file_free_rcu);
1da177e4
LT
59}
60
529bf6be
DS
61/*
62 * Return the total number of open files in the system
63 */
518de9b3 64static long get_nr_files(void)
1da177e4 65{
529bf6be 66 return percpu_counter_read_positive(&nr_files);
1da177e4
LT
67}
68
529bf6be
DS
69/*
70 * Return the maximum number of open files in the system
71 */
518de9b3 72unsigned long get_max_files(void)
ab2af1f5 73{
529bf6be 74 return files_stat.max_files;
ab2af1f5 75}
529bf6be
DS
76EXPORT_SYMBOL_GPL(get_max_files);
77
78/*
79 * Handle nr_files sysctl
80 */
81#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
8d65af78 82int proc_nr_files(ctl_table *table, int write,
529bf6be
DS
83 void __user *buffer, size_t *lenp, loff_t *ppos)
84{
85 files_stat.nr_files = get_nr_files();
518de9b3 86 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
529bf6be
DS
87}
88#else
8d65af78 89int proc_nr_files(ctl_table *table, int write,
529bf6be
DS
90 void __user *buffer, size_t *lenp, loff_t *ppos)
91{
92 return -ENOSYS;
93}
94#endif
ab2af1f5 95
1da177e4 96/* Find an unused file structure and return a pointer to it.
1afc99be
AV
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.
430e285e
DH
99 *
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.
1da177e4
LT
105 */
106struct file *get_empty_filp(void)
107{
86a264ab 108 const struct cred *cred = current_cred();
518de9b3 109 static long old_max;
1afc99be
AV
110 struct file *f;
111 int error;
1da177e4
LT
112
113 /*
114 * Privileged users can go above max_files
115 */
529bf6be
DS
116 if (get_nr_files() >= files_stat.max_files && !capable(CAP_SYS_ADMIN)) {
117 /*
118 * percpu_counters are inaccurate. Do an expensive check before
119 * we go and fail.
120 */
52d9f3b4 121 if (percpu_counter_sum_positive(&nr_files) >= files_stat.max_files)
529bf6be
DS
122 goto over;
123 }
af4d2ecb 124
4975e45f 125 f = kmem_cache_zalloc(filp_cachep, GFP_KERNEL);
1afc99be
AV
126 if (unlikely(!f))
127 return ERR_PTR(-ENOMEM);
af4d2ecb 128
529bf6be 129 percpu_counter_inc(&nr_files);
78d29788 130 f->f_cred = get_cred(cred);
1afc99be
AV
131 error = security_file_alloc(f);
132 if (unlikely(error)) {
133 file_free(f);
134 return ERR_PTR(error);
135 }
1da177e4 136
5a6b7951 137 INIT_LIST_HEAD(&f->f_u.fu_list);
516e0cc5 138 atomic_long_set(&f->f_count, 1);
af4d2ecb 139 rwlock_init(&f->f_owner.lock);
68499914 140 spin_lock_init(&f->f_lock);
5a6b7951 141 eventpoll_init_file(f);
af4d2ecb 142 /* f->f_version: 0 */
af4d2ecb
KK
143 return f;
144
145over:
1da177e4 146 /* Ran out of filps - report that */
529bf6be 147 if (get_nr_files() > old_max) {
518de9b3 148 pr_info("VFS: file-max limit %lu reached\n", get_max_files());
529bf6be 149 old_max = get_nr_files();
1da177e4 150 }
1afc99be 151 return ERR_PTR(-ENFILE);
1da177e4
LT
152}
153
ce8d2cdf
DH
154/**
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
160 *
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().
165 *
166 * If all the callers of init_file() are eliminated, its
167 * code should be moved into this function.
168 */
2c48b9c4
AV
169struct file *alloc_file(struct path *path, fmode_t mode,
170 const struct file_operations *fop)
ce8d2cdf
DH
171{
172 struct file *file;
ce8d2cdf
DH
173
174 file = get_empty_filp();
1afc99be 175 if (IS_ERR(file))
39b65252 176 return file;
ce8d2cdf 177
2c48b9c4
AV
178 file->f_path = *path;
179 file->f_mapping = path->dentry->d_inode->i_mapping;
ce8d2cdf
DH
180 file->f_mode = mode;
181 file->f_op = fop;
4a3fd211
DH
182
183 /*
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()
188 */
2c48b9c4 189 if ((mode & FMODE_WRITE) && !special_file(path->dentry->d_inode->i_mode)) {
ad775f5a 190 file_take_write(file);
385e3ed4 191 WARN_ON(mnt_clone_write(path->mnt));
4a3fd211 192 }
890275b5
MZ
193 if ((mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
194 i_readcount_inc(path->dentry->d_inode);
3d1e4631 195 return file;
ce8d2cdf 196}
73efc468 197EXPORT_SYMBOL(alloc_file);
ce8d2cdf 198
aceaf78d
DH
199/**
200 * drop_file_write_access - give up ability to write to a file
201 * @file: the file to which we will stop writing
202 *
203 * This is a central place which will give up the ability
204 * to write to @file, along with access to write through
205 * its vfsmount.
206 */
b57ce969 207static void drop_file_write_access(struct file *file)
aceaf78d 208{
4a3fd211 209 struct vfsmount *mnt = file->f_path.mnt;
aceaf78d
DH
210 struct dentry *dentry = file->f_path.dentry;
211 struct inode *inode = dentry->d_inode;
212
213 put_write_access(inode);
ad775f5a
DH
214
215 if (special_file(inode->i_mode))
216 return;
217 if (file_check_writeable(file) != 0)
218 return;
eb04c282 219 __mnt_drop_write(mnt);
ad775f5a 220 file_release_write(file);
aceaf78d 221}
aceaf78d 222
d7065da0 223/* the real guts of fput() - releasing the last reference to file
1da177e4 224 */
d7065da0 225static void __fput(struct file *file)
1da177e4 226{
0f7fc9e4
JJS
227 struct dentry *dentry = file->f_path.dentry;
228 struct vfsmount *mnt = file->f_path.mnt;
1da177e4
LT
229 struct inode *inode = dentry->d_inode;
230
231 might_sleep();
0eeca283
RL
232
233 fsnotify_close(file);
1da177e4
LT
234 /*
235 * The function eventpoll_release() should be the first called
236 * in the file cleanup chain.
237 */
238 eventpoll_release(file);
239 locks_remove_flock(file);
240
233e70f4
AV
241 if (unlikely(file->f_flags & FASYNC)) {
242 if (file->f_op && file->f_op->fasync)
243 file->f_op->fasync(-1, file, 0);
244 }
4199d35c 245 ima_file_free(file);
1da177e4
LT
246 if (file->f_op && file->f_op->release)
247 file->f_op->release(inode, file);
248 security_file_free(file);
60ed8cf7
MS
249 if (unlikely(S_ISCHR(inode->i_mode) && inode->i_cdev != NULL &&
250 !(file->f_mode & FMODE_PATH))) {
1da177e4 251 cdev_put(inode->i_cdev);
60ed8cf7 252 }
1da177e4 253 fops_put(file->f_op);
609d7fa9 254 put_pid(file->f_owner.pid);
890275b5
MZ
255 if ((file->f_mode & (FMODE_READ | FMODE_WRITE)) == FMODE_READ)
256 i_readcount_dec(inode);
aceaf78d
DH
257 if (file->f_mode & FMODE_WRITE)
258 drop_file_write_access(file);
0f7fc9e4
JJS
259 file->f_path.dentry = NULL;
260 file->f_path.mnt = NULL;
1da177e4
LT
261 file_free(file);
262 dput(dentry);
263 mntput(mnt);
264}
265
4a9d4b02
AV
266static DEFINE_SPINLOCK(delayed_fput_lock);
267static LIST_HEAD(delayed_fput_list);
268static void delayed_fput(struct work_struct *unused)
269{
270 LIST_HEAD(head);
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);
277 __fput(f);
278 }
279}
280
281static void ____fput(struct callback_head *work)
282{
283 __fput(container_of(work, struct file, f_u.fu_rcuhead));
284}
285
286/*
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.
295 */
296void flush_delayed_fput(void)
297{
298 delayed_fput(NULL);
299}
300
301static DECLARE_WORK(delayed_fput_work, delayed_fput);
302
d7065da0
AV
303void fput(struct file *file)
304{
4a9d4b02
AV
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)) {
309 unsigned long flags;
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);
314 return;
315 }
316 init_task_work(&file->f_u.fu_rcuhead, ____fput);
317 task_work_add(task, &file->f_u.fu_rcuhead, true);
318 }
319}
320
321/*
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.
328 */
329void __fput_sync(struct file *file)
330{
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));
d7065da0 335 __fput(file);
4a9d4b02 336 }
d7065da0
AV
337}
338
339EXPORT_SYMBOL(fput);
340
1da177e4
LT
341void put_filp(struct file *file)
342{
516e0cc5 343 if (atomic_long_dec_and_test(&file->f_count)) {
1da177e4 344 security_file_free(file);
ee2ffa0d 345 file_sb_list_del(file);
1da177e4
LT
346 file_free(file);
347 }
348}
349
6416ccb7
NP
350static inline int file_list_cpu(struct file *file)
351{
352#ifdef CONFIG_SMP
353 return file->f_sb_list_cpu;
354#else
355 return smp_processor_id();
356#endif
357}
358
359/* helper for file_sb_list_add to reduce ifdefs */
360static inline void __file_sb_list_add(struct file *file, struct super_block *sb)
361{
362 struct list_head *list;
363#ifdef CONFIG_SMP
364 int cpu;
365 cpu = smp_processor_id();
366 file->f_sb_list_cpu = cpu;
367 list = per_cpu_ptr(sb->s_files, cpu);
368#else
369 list = &sb->s_files;
370#endif
371 list_add(&file->f_u.fu_list, list);
372}
373
374/**
375 * file_sb_list_add - add a file to the sb's file list
376 * @file: file to add
377 * @sb: sb to add it to
378 *
379 * Use this function to associate a file with the superblock of the inode it
380 * refers to.
381 */
ee2ffa0d 382void file_sb_list_add(struct file *file, struct super_block *sb)
1da177e4 383{
962830df 384 lg_local_lock(&files_lglock);
6416ccb7 385 __file_sb_list_add(file, sb);
962830df 386 lg_local_unlock(&files_lglock);
1da177e4
LT
387}
388
6416ccb7
NP
389/**
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
393 *
394 * Use this function to remove a file from its superblock.
395 */
ee2ffa0d 396void file_sb_list_del(struct file *file)
1da177e4 397{
2f512016 398 if (!list_empty(&file->f_u.fu_list)) {
962830df 399 lg_local_lock_cpu(&files_lglock, file_list_cpu(file));
2f512016 400 list_del_init(&file->f_u.fu_list);
962830df 401 lg_local_unlock_cpu(&files_lglock, file_list_cpu(file));
1da177e4
LT
402 }
403}
404
6416ccb7
NP
405#ifdef CONFIG_SMP
406
407/*
408 * These macros iterate all files on all CPUs for a given superblock.
409 * files_lglock must be held globally.
410 */
411#define do_file_list_for_each_entry(__sb, __file) \
412{ \
413 int i; \
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)
418
419#define while_file_list_for_each_entry \
420 } \
421}
422
423#else
424
425#define do_file_list_for_each_entry(__sb, __file) \
426{ \
427 struct list_head *list; \
428 list = &(sb)->s_files; \
429 list_for_each_entry((__file), list, f_u.fu_list)
430
431#define while_file_list_for_each_entry \
432}
433
434#endif
435
864d7c4c 436/**
437 * mark_files_ro - mark all files read-only
438 * @sb: superblock in question
439 *
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.
442 */
443void mark_files_ro(struct super_block *sb)
444{
445 struct file *f;
446
962830df 447 lg_global_lock(&files_lglock);
6416ccb7 448 do_file_list_for_each_entry(sb, f) {
496ad9aa 449 if (!S_ISREG(file_inode(f)->i_mode))
864d7c4c 450 continue;
451 if (!file_count(f))
452 continue;
453 if (!(f->f_mode & FMODE_WRITE))
454 continue;
42e49608 455 spin_lock(&f->f_lock);
864d7c4c 456 f->f_mode &= ~FMODE_WRITE;
42e49608 457 spin_unlock(&f->f_lock);
864d7c4c 458 if (file_check_writeable(f) != 0)
459 continue;
72651cac 460 __mnt_drop_write(f->f_path.mnt);
864d7c4c 461 file_release_write(f);
6416ccb7 462 } while_file_list_for_each_entry;
962830df 463 lg_global_unlock(&files_lglock);
864d7c4c 464}
465
1da177e4
LT
466void __init files_init(unsigned long mempages)
467{
518de9b3 468 unsigned long n;
b6b3fdea
ED
469
470 filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0,
471 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
472
473 /*
474 * One file with associated inode and dcache is very roughly 1K.
1da177e4
LT
475 * Per default don't use more than 10% of our memory for files.
476 */
477
478 n = (mempages * (PAGE_SIZE / 1024)) / 10;
518de9b3 479 files_stat.max_files = max_t(unsigned long, n, NR_FILE);
ab2af1f5 480 files_defer_init();
962830df 481 lg_lock_init(&files_lglock, "files_lglock");
0216bfcf 482 percpu_counter_init(&nr_files, 0);
1da177e4 483}