tracing: add sched_set_prio tracepoint
[deliverable/linux.git] / fs / fcntl.c
1 /*
2 * linux/fs/fcntl.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
6
7 #include <linux/syscalls.h>
8 #include <linux/init.h>
9 #include <linux/mm.h>
10 #include <linux/fs.h>
11 #include <linux/file.h>
12 #include <linux/fdtable.h>
13 #include <linux/capability.h>
14 #include <linux/dnotify.h>
15 #include <linux/slab.h>
16 #include <linux/module.h>
17 #include <linux/pipe_fs_i.h>
18 #include <linux/security.h>
19 #include <linux/ptrace.h>
20 #include <linux/signal.h>
21 #include <linux/rcupdate.h>
22 #include <linux/pid_namespace.h>
23 #include <linux/user_namespace.h>
24 #include <linux/shmem_fs.h>
25
26 #include <asm/poll.h>
27 #include <asm/siginfo.h>
28 #include <asm/uaccess.h>
29
30 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
31
32 static int setfl(int fd, struct file * filp, unsigned long arg)
33 {
34 struct inode * inode = file_inode(filp);
35 int error = 0;
36
37 /*
38 * O_APPEND cannot be cleared if the file is marked as append-only
39 * and the file is open for write.
40 */
41 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
42 return -EPERM;
43
44 /* O_NOATIME can only be set by the owner or superuser */
45 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
46 if (!inode_owner_or_capable(inode))
47 return -EPERM;
48
49 /* required for strict SunOS emulation */
50 if (O_NONBLOCK != O_NDELAY)
51 if (arg & O_NDELAY)
52 arg |= O_NONBLOCK;
53
54 /* Pipe packetized mode is controlled by O_DIRECT flag */
55 if (!S_ISFIFO(filp->f_inode->i_mode) && (arg & O_DIRECT)) {
56 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
57 !filp->f_mapping->a_ops->direct_IO)
58 return -EINVAL;
59 }
60
61 if (filp->f_op->check_flags)
62 error = filp->f_op->check_flags(arg);
63 if (error)
64 return error;
65
66 /*
67 * ->fasync() is responsible for setting the FASYNC bit.
68 */
69 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
70 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
71 if (error < 0)
72 goto out;
73 if (error > 0)
74 error = 0;
75 }
76 spin_lock(&filp->f_lock);
77 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
78 spin_unlock(&filp->f_lock);
79
80 out:
81 return error;
82 }
83
84 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
85 int force)
86 {
87 write_lock_irq(&filp->f_owner.lock);
88 if (force || !filp->f_owner.pid) {
89 put_pid(filp->f_owner.pid);
90 filp->f_owner.pid = get_pid(pid);
91 filp->f_owner.pid_type = type;
92
93 if (pid) {
94 const struct cred *cred = current_cred();
95 filp->f_owner.uid = cred->uid;
96 filp->f_owner.euid = cred->euid;
97 }
98 }
99 write_unlock_irq(&filp->f_owner.lock);
100 }
101
102 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
103 int force)
104 {
105 security_file_set_fowner(filp);
106 f_modown(filp, pid, type, force);
107 }
108 EXPORT_SYMBOL(__f_setown);
109
110 void f_setown(struct file *filp, unsigned long arg, int force)
111 {
112 enum pid_type type;
113 struct pid *pid;
114 int who = arg;
115 type = PIDTYPE_PID;
116 if (who < 0) {
117 type = PIDTYPE_PGID;
118 who = -who;
119 }
120 rcu_read_lock();
121 pid = find_vpid(who);
122 __f_setown(filp, pid, type, force);
123 rcu_read_unlock();
124 }
125 EXPORT_SYMBOL(f_setown);
126
127 void f_delown(struct file *filp)
128 {
129 f_modown(filp, NULL, PIDTYPE_PID, 1);
130 }
131
132 pid_t f_getown(struct file *filp)
133 {
134 pid_t pid;
135 read_lock(&filp->f_owner.lock);
136 pid = pid_vnr(filp->f_owner.pid);
137 if (filp->f_owner.pid_type == PIDTYPE_PGID)
138 pid = -pid;
139 read_unlock(&filp->f_owner.lock);
140 return pid;
141 }
142
143 static int f_setown_ex(struct file *filp, unsigned long arg)
144 {
145 struct f_owner_ex __user *owner_p = (void __user *)arg;
146 struct f_owner_ex owner;
147 struct pid *pid;
148 int type;
149 int ret;
150
151 ret = copy_from_user(&owner, owner_p, sizeof(owner));
152 if (ret)
153 return -EFAULT;
154
155 switch (owner.type) {
156 case F_OWNER_TID:
157 type = PIDTYPE_MAX;
158 break;
159
160 case F_OWNER_PID:
161 type = PIDTYPE_PID;
162 break;
163
164 case F_OWNER_PGRP:
165 type = PIDTYPE_PGID;
166 break;
167
168 default:
169 return -EINVAL;
170 }
171
172 rcu_read_lock();
173 pid = find_vpid(owner.pid);
174 if (owner.pid && !pid)
175 ret = -ESRCH;
176 else
177 __f_setown(filp, pid, type, 1);
178 rcu_read_unlock();
179
180 return ret;
181 }
182
183 static int f_getown_ex(struct file *filp, unsigned long arg)
184 {
185 struct f_owner_ex __user *owner_p = (void __user *)arg;
186 struct f_owner_ex owner;
187 int ret = 0;
188
189 read_lock(&filp->f_owner.lock);
190 owner.pid = pid_vnr(filp->f_owner.pid);
191 switch (filp->f_owner.pid_type) {
192 case PIDTYPE_MAX:
193 owner.type = F_OWNER_TID;
194 break;
195
196 case PIDTYPE_PID:
197 owner.type = F_OWNER_PID;
198 break;
199
200 case PIDTYPE_PGID:
201 owner.type = F_OWNER_PGRP;
202 break;
203
204 default:
205 WARN_ON(1);
206 ret = -EINVAL;
207 break;
208 }
209 read_unlock(&filp->f_owner.lock);
210
211 if (!ret) {
212 ret = copy_to_user(owner_p, &owner, sizeof(owner));
213 if (ret)
214 ret = -EFAULT;
215 }
216 return ret;
217 }
218
219 #ifdef CONFIG_CHECKPOINT_RESTORE
220 static int f_getowner_uids(struct file *filp, unsigned long arg)
221 {
222 struct user_namespace *user_ns = current_user_ns();
223 uid_t __user *dst = (void __user *)arg;
224 uid_t src[2];
225 int err;
226
227 read_lock(&filp->f_owner.lock);
228 src[0] = from_kuid(user_ns, filp->f_owner.uid);
229 src[1] = from_kuid(user_ns, filp->f_owner.euid);
230 read_unlock(&filp->f_owner.lock);
231
232 err = put_user(src[0], &dst[0]);
233 err |= put_user(src[1], &dst[1]);
234
235 return err;
236 }
237 #else
238 static int f_getowner_uids(struct file *filp, unsigned long arg)
239 {
240 return -EINVAL;
241 }
242 #endif
243
244 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
245 struct file *filp)
246 {
247 long err = -EINVAL;
248
249 switch (cmd) {
250 case F_DUPFD:
251 err = f_dupfd(arg, filp, 0);
252 break;
253 case F_DUPFD_CLOEXEC:
254 err = f_dupfd(arg, filp, O_CLOEXEC);
255 break;
256 case F_GETFD:
257 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
258 break;
259 case F_SETFD:
260 err = 0;
261 set_close_on_exec(fd, arg & FD_CLOEXEC);
262 break;
263 case F_GETFL:
264 err = filp->f_flags;
265 break;
266 case F_SETFL:
267 err = setfl(fd, filp, arg);
268 break;
269 #if BITS_PER_LONG != 32
270 /* 32-bit arches must use fcntl64() */
271 case F_OFD_GETLK:
272 #endif
273 case F_GETLK:
274 err = fcntl_getlk(filp, cmd, (struct flock __user *) arg);
275 break;
276 #if BITS_PER_LONG != 32
277 /* 32-bit arches must use fcntl64() */
278 case F_OFD_SETLK:
279 case F_OFD_SETLKW:
280 #endif
281 /* Fallthrough */
282 case F_SETLK:
283 case F_SETLKW:
284 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
285 break;
286 case F_GETOWN:
287 /*
288 * XXX If f_owner is a process group, the
289 * negative return value will get converted
290 * into an error. Oops. If we keep the
291 * current syscall conventions, the only way
292 * to fix this will be in libc.
293 */
294 err = f_getown(filp);
295 force_successful_syscall_return();
296 break;
297 case F_SETOWN:
298 f_setown(filp, arg, 1);
299 err = 0;
300 break;
301 case F_GETOWN_EX:
302 err = f_getown_ex(filp, arg);
303 break;
304 case F_SETOWN_EX:
305 err = f_setown_ex(filp, arg);
306 break;
307 case F_GETOWNER_UIDS:
308 err = f_getowner_uids(filp, arg);
309 break;
310 case F_GETSIG:
311 err = filp->f_owner.signum;
312 break;
313 case F_SETSIG:
314 /* arg == 0 restores default behaviour. */
315 if (!valid_signal(arg)) {
316 break;
317 }
318 err = 0;
319 filp->f_owner.signum = arg;
320 break;
321 case F_GETLEASE:
322 err = fcntl_getlease(filp);
323 break;
324 case F_SETLEASE:
325 err = fcntl_setlease(fd, filp, arg);
326 break;
327 case F_NOTIFY:
328 err = fcntl_dirnotify(fd, filp, arg);
329 break;
330 case F_SETPIPE_SZ:
331 case F_GETPIPE_SZ:
332 err = pipe_fcntl(filp, cmd, arg);
333 break;
334 case F_ADD_SEALS:
335 case F_GET_SEALS:
336 err = shmem_fcntl(filp, cmd, arg);
337 break;
338 default:
339 break;
340 }
341 return err;
342 }
343
344 static int check_fcntl_cmd(unsigned cmd)
345 {
346 switch (cmd) {
347 case F_DUPFD:
348 case F_DUPFD_CLOEXEC:
349 case F_GETFD:
350 case F_SETFD:
351 case F_GETFL:
352 return 1;
353 }
354 return 0;
355 }
356
357 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
358 {
359 struct fd f = fdget_raw(fd);
360 long err = -EBADF;
361
362 if (!f.file)
363 goto out;
364
365 if (unlikely(f.file->f_mode & FMODE_PATH)) {
366 if (!check_fcntl_cmd(cmd))
367 goto out1;
368 }
369
370 err = security_file_fcntl(f.file, cmd, arg);
371 if (!err)
372 err = do_fcntl(fd, cmd, arg, f.file);
373
374 out1:
375 fdput(f);
376 out:
377 return err;
378 }
379
380 #if BITS_PER_LONG == 32
381 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
382 unsigned long, arg)
383 {
384 struct fd f = fdget_raw(fd);
385 long err = -EBADF;
386
387 if (!f.file)
388 goto out;
389
390 if (unlikely(f.file->f_mode & FMODE_PATH)) {
391 if (!check_fcntl_cmd(cmd))
392 goto out1;
393 }
394
395 err = security_file_fcntl(f.file, cmd, arg);
396 if (err)
397 goto out1;
398
399 switch (cmd) {
400 case F_GETLK64:
401 case F_OFD_GETLK:
402 err = fcntl_getlk64(f.file, cmd, (struct flock64 __user *) arg);
403 break;
404 case F_SETLK64:
405 case F_SETLKW64:
406 case F_OFD_SETLK:
407 case F_OFD_SETLKW:
408 err = fcntl_setlk64(fd, f.file, cmd,
409 (struct flock64 __user *) arg);
410 break;
411 default:
412 err = do_fcntl(fd, cmd, arg, f.file);
413 break;
414 }
415 out1:
416 fdput(f);
417 out:
418 return err;
419 }
420 #endif
421
422 /* Table to convert sigio signal codes into poll band bitmaps */
423
424 static const long band_table[NSIGPOLL] = {
425 POLLIN | POLLRDNORM, /* POLL_IN */
426 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
427 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
428 POLLERR, /* POLL_ERR */
429 POLLPRI | POLLRDBAND, /* POLL_PRI */
430 POLLHUP | POLLERR /* POLL_HUP */
431 };
432
433 static inline int sigio_perm(struct task_struct *p,
434 struct fown_struct *fown, int sig)
435 {
436 const struct cred *cred;
437 int ret;
438
439 rcu_read_lock();
440 cred = __task_cred(p);
441 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
442 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
443 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
444 !security_file_send_sigiotask(p, fown, sig));
445 rcu_read_unlock();
446 return ret;
447 }
448
449 static void send_sigio_to_task(struct task_struct *p,
450 struct fown_struct *fown,
451 int fd, int reason, int group)
452 {
453 /*
454 * F_SETSIG can change ->signum lockless in parallel, make
455 * sure we read it once and use the same value throughout.
456 */
457 int signum = ACCESS_ONCE(fown->signum);
458
459 if (!sigio_perm(p, fown, signum))
460 return;
461
462 switch (signum) {
463 siginfo_t si;
464 default:
465 /* Queue a rt signal with the appropriate fd as its
466 value. We use SI_SIGIO as the source, not
467 SI_KERNEL, since kernel signals always get
468 delivered even if we can't queue. Failure to
469 queue in this case _should_ be reported; we fall
470 back to SIGIO in that case. --sct */
471 si.si_signo = signum;
472 si.si_errno = 0;
473 si.si_code = reason;
474 /* Make sure we are called with one of the POLL_*
475 reasons, otherwise we could leak kernel stack into
476 userspace. */
477 BUG_ON((reason & __SI_MASK) != __SI_POLL);
478 if (reason - POLL_IN >= NSIGPOLL)
479 si.si_band = ~0L;
480 else
481 si.si_band = band_table[reason - POLL_IN];
482 si.si_fd = fd;
483 if (!do_send_sig_info(signum, &si, p, group))
484 break;
485 /* fall-through: fall back on the old plain SIGIO signal */
486 case 0:
487 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
488 }
489 }
490
491 void send_sigio(struct fown_struct *fown, int fd, int band)
492 {
493 struct task_struct *p;
494 enum pid_type type;
495 struct pid *pid;
496 int group = 1;
497
498 read_lock(&fown->lock);
499
500 type = fown->pid_type;
501 if (type == PIDTYPE_MAX) {
502 group = 0;
503 type = PIDTYPE_PID;
504 }
505
506 pid = fown->pid;
507 if (!pid)
508 goto out_unlock_fown;
509
510 read_lock(&tasklist_lock);
511 do_each_pid_task(pid, type, p) {
512 send_sigio_to_task(p, fown, fd, band, group);
513 } while_each_pid_task(pid, type, p);
514 read_unlock(&tasklist_lock);
515 out_unlock_fown:
516 read_unlock(&fown->lock);
517 }
518
519 static void send_sigurg_to_task(struct task_struct *p,
520 struct fown_struct *fown, int group)
521 {
522 if (sigio_perm(p, fown, SIGURG))
523 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
524 }
525
526 int send_sigurg(struct fown_struct *fown)
527 {
528 struct task_struct *p;
529 enum pid_type type;
530 struct pid *pid;
531 int group = 1;
532 int ret = 0;
533
534 read_lock(&fown->lock);
535
536 type = fown->pid_type;
537 if (type == PIDTYPE_MAX) {
538 group = 0;
539 type = PIDTYPE_PID;
540 }
541
542 pid = fown->pid;
543 if (!pid)
544 goto out_unlock_fown;
545
546 ret = 1;
547
548 read_lock(&tasklist_lock);
549 do_each_pid_task(pid, type, p) {
550 send_sigurg_to_task(p, fown, group);
551 } while_each_pid_task(pid, type, p);
552 read_unlock(&tasklist_lock);
553 out_unlock_fown:
554 read_unlock(&fown->lock);
555 return ret;
556 }
557
558 static DEFINE_SPINLOCK(fasync_lock);
559 static struct kmem_cache *fasync_cache __read_mostly;
560
561 static void fasync_free_rcu(struct rcu_head *head)
562 {
563 kmem_cache_free(fasync_cache,
564 container_of(head, struct fasync_struct, fa_rcu));
565 }
566
567 /*
568 * Remove a fasync entry. If successfully removed, return
569 * positive and clear the FASYNC flag. If no entry exists,
570 * do nothing and return 0.
571 *
572 * NOTE! It is very important that the FASYNC flag always
573 * match the state "is the filp on a fasync list".
574 *
575 */
576 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
577 {
578 struct fasync_struct *fa, **fp;
579 int result = 0;
580
581 spin_lock(&filp->f_lock);
582 spin_lock(&fasync_lock);
583 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
584 if (fa->fa_file != filp)
585 continue;
586
587 spin_lock_irq(&fa->fa_lock);
588 fa->fa_file = NULL;
589 spin_unlock_irq(&fa->fa_lock);
590
591 *fp = fa->fa_next;
592 call_rcu(&fa->fa_rcu, fasync_free_rcu);
593 filp->f_flags &= ~FASYNC;
594 result = 1;
595 break;
596 }
597 spin_unlock(&fasync_lock);
598 spin_unlock(&filp->f_lock);
599 return result;
600 }
601
602 struct fasync_struct *fasync_alloc(void)
603 {
604 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
605 }
606
607 /*
608 * NOTE! This can be used only for unused fasync entries:
609 * entries that actually got inserted on the fasync list
610 * need to be released by rcu - see fasync_remove_entry.
611 */
612 void fasync_free(struct fasync_struct *new)
613 {
614 kmem_cache_free(fasync_cache, new);
615 }
616
617 /*
618 * Insert a new entry into the fasync list. Return the pointer to the
619 * old one if we didn't use the new one.
620 *
621 * NOTE! It is very important that the FASYNC flag always
622 * match the state "is the filp on a fasync list".
623 */
624 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
625 {
626 struct fasync_struct *fa, **fp;
627
628 spin_lock(&filp->f_lock);
629 spin_lock(&fasync_lock);
630 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
631 if (fa->fa_file != filp)
632 continue;
633
634 spin_lock_irq(&fa->fa_lock);
635 fa->fa_fd = fd;
636 spin_unlock_irq(&fa->fa_lock);
637 goto out;
638 }
639
640 spin_lock_init(&new->fa_lock);
641 new->magic = FASYNC_MAGIC;
642 new->fa_file = filp;
643 new->fa_fd = fd;
644 new->fa_next = *fapp;
645 rcu_assign_pointer(*fapp, new);
646 filp->f_flags |= FASYNC;
647
648 out:
649 spin_unlock(&fasync_lock);
650 spin_unlock(&filp->f_lock);
651 return fa;
652 }
653
654 /*
655 * Add a fasync entry. Return negative on error, positive if
656 * added, and zero if did nothing but change an existing one.
657 */
658 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
659 {
660 struct fasync_struct *new;
661
662 new = fasync_alloc();
663 if (!new)
664 return -ENOMEM;
665
666 /*
667 * fasync_insert_entry() returns the old (update) entry if
668 * it existed.
669 *
670 * So free the (unused) new entry and return 0 to let the
671 * caller know that we didn't add any new fasync entries.
672 */
673 if (fasync_insert_entry(fd, filp, fapp, new)) {
674 fasync_free(new);
675 return 0;
676 }
677
678 return 1;
679 }
680
681 /*
682 * fasync_helper() is used by almost all character device drivers
683 * to set up the fasync queue, and for regular files by the file
684 * lease code. It returns negative on error, 0 if it did no changes
685 * and positive if it added/deleted the entry.
686 */
687 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
688 {
689 if (!on)
690 return fasync_remove_entry(filp, fapp);
691 return fasync_add_entry(fd, filp, fapp);
692 }
693
694 EXPORT_SYMBOL(fasync_helper);
695
696 /*
697 * rcu_read_lock() is held
698 */
699 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
700 {
701 while (fa) {
702 struct fown_struct *fown;
703 unsigned long flags;
704
705 if (fa->magic != FASYNC_MAGIC) {
706 printk(KERN_ERR "kill_fasync: bad magic number in "
707 "fasync_struct!\n");
708 return;
709 }
710 spin_lock_irqsave(&fa->fa_lock, flags);
711 if (fa->fa_file) {
712 fown = &fa->fa_file->f_owner;
713 /* Don't send SIGURG to processes which have not set a
714 queued signum: SIGURG has its own default signalling
715 mechanism. */
716 if (!(sig == SIGURG && fown->signum == 0))
717 send_sigio(fown, fa->fa_fd, band);
718 }
719 spin_unlock_irqrestore(&fa->fa_lock, flags);
720 fa = rcu_dereference(fa->fa_next);
721 }
722 }
723
724 void kill_fasync(struct fasync_struct **fp, int sig, int band)
725 {
726 /* First a quick test without locking: usually
727 * the list is empty.
728 */
729 if (*fp) {
730 rcu_read_lock();
731 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
732 rcu_read_unlock();
733 }
734 }
735 EXPORT_SYMBOL(kill_fasync);
736
737 static int __init fcntl_init(void)
738 {
739 /*
740 * Please add new bits here to ensure allocation uniqueness.
741 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
742 * is defined as O_NONBLOCK on some platforms and not on others.
743 */
744 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
745 O_RDONLY | O_WRONLY | O_RDWR |
746 O_CREAT | O_EXCL | O_NOCTTY |
747 O_TRUNC | O_APPEND | /* O_NONBLOCK | */
748 __O_SYNC | O_DSYNC | FASYNC |
749 O_DIRECT | O_LARGEFILE | O_DIRECTORY |
750 O_NOFOLLOW | O_NOATIME | O_CLOEXEC |
751 __FMODE_EXEC | O_PATH | __O_TMPFILE |
752 __FMODE_NONOTIFY
753 ));
754
755 fasync_cache = kmem_cache_create("fasync_cache",
756 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
757 return 0;
758 }
759
760 module_init(fcntl_init)
This page took 0.045087 seconds and 5 git commands to generate.