4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/syscalls.h>
8 #include <linux/init.h>
10 #include <linux/sched/task.h>
12 #include <linux/file.h>
13 #include <linux/fdtable.h>
14 #include <linux/capability.h>
15 #include <linux/dnotify.h>
16 #include <linux/slab.h>
17 #include <linux/module.h>
18 #include <linux/pipe_fs_i.h>
19 #include <linux/security.h>
20 #include <linux/ptrace.h>
21 #include <linux/signal.h>
22 #include <linux/rcupdate.h>
23 #include <linux/pid_namespace.h>
24 #include <linux/user_namespace.h>
25 #include <linux/shmem_fs.h>
28 #include <asm/siginfo.h>
29 #include <linux/uaccess.h>
31 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
33 static int setfl(int fd
, struct file
* filp
, unsigned long arg
)
35 struct inode
* inode
= file_inode(filp
);
39 * O_APPEND cannot be cleared if the file is marked as append-only
40 * and the file is open for write.
42 if (((arg
^ filp
->f_flags
) & O_APPEND
) && IS_APPEND(inode
))
45 /* O_NOATIME can only be set by the owner or superuser */
46 if ((arg
& O_NOATIME
) && !(filp
->f_flags
& O_NOATIME
))
47 if (!inode_owner_or_capable(inode
))
50 /* required for strict SunOS emulation */
51 if (O_NONBLOCK
!= O_NDELAY
)
55 /* Pipe packetized mode is controlled by O_DIRECT flag */
56 if (!S_ISFIFO(inode
->i_mode
) && (arg
& O_DIRECT
)) {
57 if (!filp
->f_mapping
|| !filp
->f_mapping
->a_ops
||
58 !filp
->f_mapping
->a_ops
->direct_IO
)
62 if (filp
->f_op
->check_flags
)
63 error
= filp
->f_op
->check_flags(arg
);
68 * ->fasync() is responsible for setting the FASYNC bit.
70 if (((arg
^ filp
->f_flags
) & FASYNC
) && filp
->f_op
->fasync
) {
71 error
= filp
->f_op
->fasync(fd
, filp
, (arg
& FASYNC
) != 0);
77 spin_lock(&filp
->f_lock
);
78 filp
->f_flags
= (arg
& SETFL_MASK
) | (filp
->f_flags
& ~SETFL_MASK
);
79 spin_unlock(&filp
->f_lock
);
85 static void f_modown(struct file
*filp
, struct pid
*pid
, enum pid_type type
,
88 write_lock_irq(&filp
->f_owner
.lock
);
89 if (force
|| !filp
->f_owner
.pid
) {
90 put_pid(filp
->f_owner
.pid
);
91 filp
->f_owner
.pid
= get_pid(pid
);
92 filp
->f_owner
.pid_type
= type
;
95 const struct cred
*cred
= current_cred();
96 filp
->f_owner
.uid
= cred
->uid
;
97 filp
->f_owner
.euid
= cred
->euid
;
100 write_unlock_irq(&filp
->f_owner
.lock
);
103 void __f_setown(struct file
*filp
, struct pid
*pid
, enum pid_type type
,
106 security_file_set_fowner(filp
);
107 f_modown(filp
, pid
, type
, force
);
109 EXPORT_SYMBOL(__f_setown
);
111 void f_setown(struct file
*filp
, unsigned long arg
, int force
)
122 pid
= find_vpid(who
);
123 __f_setown(filp
, pid
, type
, force
);
126 EXPORT_SYMBOL(f_setown
);
128 void f_delown(struct file
*filp
)
130 f_modown(filp
, NULL
, PIDTYPE_PID
, 1);
133 pid_t
f_getown(struct file
*filp
)
136 read_lock(&filp
->f_owner
.lock
);
137 pid
= pid_vnr(filp
->f_owner
.pid
);
138 if (filp
->f_owner
.pid_type
== PIDTYPE_PGID
)
140 read_unlock(&filp
->f_owner
.lock
);
144 static int f_setown_ex(struct file
*filp
, unsigned long arg
)
146 struct f_owner_ex __user
*owner_p
= (void __user
*)arg
;
147 struct f_owner_ex owner
;
152 ret
= copy_from_user(&owner
, owner_p
, sizeof(owner
));
156 switch (owner
.type
) {
174 pid
= find_vpid(owner
.pid
);
175 if (owner
.pid
&& !pid
)
178 __f_setown(filp
, pid
, type
, 1);
184 static int f_getown_ex(struct file
*filp
, unsigned long arg
)
186 struct f_owner_ex __user
*owner_p
= (void __user
*)arg
;
187 struct f_owner_ex owner
;
190 read_lock(&filp
->f_owner
.lock
);
191 owner
.pid
= pid_vnr(filp
->f_owner
.pid
);
192 switch (filp
->f_owner
.pid_type
) {
194 owner
.type
= F_OWNER_TID
;
198 owner
.type
= F_OWNER_PID
;
202 owner
.type
= F_OWNER_PGRP
;
210 read_unlock(&filp
->f_owner
.lock
);
213 ret
= copy_to_user(owner_p
, &owner
, sizeof(owner
));
220 #ifdef CONFIG_CHECKPOINT_RESTORE
221 static int f_getowner_uids(struct file
*filp
, unsigned long arg
)
223 struct user_namespace
*user_ns
= current_user_ns();
224 uid_t __user
*dst
= (void __user
*)arg
;
228 read_lock(&filp
->f_owner
.lock
);
229 src
[0] = from_kuid(user_ns
, filp
->f_owner
.uid
);
230 src
[1] = from_kuid(user_ns
, filp
->f_owner
.euid
);
231 read_unlock(&filp
->f_owner
.lock
);
233 err
= put_user(src
[0], &dst
[0]);
234 err
|= put_user(src
[1], &dst
[1]);
239 static int f_getowner_uids(struct file
*filp
, unsigned long arg
)
245 static long do_fcntl(int fd
, unsigned int cmd
, unsigned long arg
,
252 err
= f_dupfd(arg
, filp
, 0);
254 case F_DUPFD_CLOEXEC
:
255 err
= f_dupfd(arg
, filp
, O_CLOEXEC
);
258 err
= get_close_on_exec(fd
) ? FD_CLOEXEC
: 0;
262 set_close_on_exec(fd
, arg
& FD_CLOEXEC
);
268 err
= setfl(fd
, filp
, arg
);
270 #if BITS_PER_LONG != 32
271 /* 32-bit arches must use fcntl64() */
275 err
= fcntl_getlk(filp
, cmd
, (struct flock __user
*) arg
);
277 #if BITS_PER_LONG != 32
278 /* 32-bit arches must use fcntl64() */
285 err
= fcntl_setlk(fd
, filp
, cmd
, (struct flock __user
*) arg
);
289 * XXX If f_owner is a process group, the
290 * negative return value will get converted
291 * into an error. Oops. If we keep the
292 * current syscall conventions, the only way
293 * to fix this will be in libc.
295 err
= f_getown(filp
);
296 force_successful_syscall_return();
299 f_setown(filp
, arg
, 1);
303 err
= f_getown_ex(filp
, arg
);
306 err
= f_setown_ex(filp
, arg
);
308 case F_GETOWNER_UIDS
:
309 err
= f_getowner_uids(filp
, arg
);
312 err
= filp
->f_owner
.signum
;
315 /* arg == 0 restores default behaviour. */
316 if (!valid_signal(arg
)) {
320 filp
->f_owner
.signum
= arg
;
323 err
= fcntl_getlease(filp
);
326 err
= fcntl_setlease(fd
, filp
, arg
);
329 err
= fcntl_dirnotify(fd
, filp
, arg
);
333 err
= pipe_fcntl(filp
, cmd
, arg
);
337 err
= shmem_fcntl(filp
, cmd
, arg
);
345 static int check_fcntl_cmd(unsigned cmd
)
349 case F_DUPFD_CLOEXEC
:
358 SYSCALL_DEFINE3(fcntl
, unsigned int, fd
, unsigned int, cmd
, unsigned long, arg
)
360 struct fd f
= fdget_raw(fd
);
366 if (unlikely(f
.file
->f_mode
& FMODE_PATH
)) {
367 if (!check_fcntl_cmd(cmd
))
371 err
= security_file_fcntl(f
.file
, cmd
, arg
);
373 err
= do_fcntl(fd
, cmd
, arg
, f
.file
);
381 #if BITS_PER_LONG == 32
382 SYSCALL_DEFINE3(fcntl64
, unsigned int, fd
, unsigned int, cmd
,
385 struct fd f
= fdget_raw(fd
);
391 if (unlikely(f
.file
->f_mode
& FMODE_PATH
)) {
392 if (!check_fcntl_cmd(cmd
))
396 err
= security_file_fcntl(f
.file
, cmd
, arg
);
403 err
= fcntl_getlk64(f
.file
, cmd
, (struct flock64 __user
*) arg
);
409 err
= fcntl_setlk64(fd
, f
.file
, cmd
,
410 (struct flock64 __user
*) arg
);
413 err
= do_fcntl(fd
, cmd
, arg
, f
.file
);
423 /* Table to convert sigio signal codes into poll band bitmaps */
425 static const long band_table
[NSIGPOLL
] = {
426 POLLIN
| POLLRDNORM
, /* POLL_IN */
427 POLLOUT
| POLLWRNORM
| POLLWRBAND
, /* POLL_OUT */
428 POLLIN
| POLLRDNORM
| POLLMSG
, /* POLL_MSG */
429 POLLERR
, /* POLL_ERR */
430 POLLPRI
| POLLRDBAND
, /* POLL_PRI */
431 POLLHUP
| POLLERR
/* POLL_HUP */
434 static inline int sigio_perm(struct task_struct
*p
,
435 struct fown_struct
*fown
, int sig
)
437 const struct cred
*cred
;
441 cred
= __task_cred(p
);
442 ret
= ((uid_eq(fown
->euid
, GLOBAL_ROOT_UID
) ||
443 uid_eq(fown
->euid
, cred
->suid
) || uid_eq(fown
->euid
, cred
->uid
) ||
444 uid_eq(fown
->uid
, cred
->suid
) || uid_eq(fown
->uid
, cred
->uid
)) &&
445 !security_file_send_sigiotask(p
, fown
, sig
));
450 static void send_sigio_to_task(struct task_struct
*p
,
451 struct fown_struct
*fown
,
452 int fd
, int reason
, int group
)
455 * F_SETSIG can change ->signum lockless in parallel, make
456 * sure we read it once and use the same value throughout.
458 int signum
= ACCESS_ONCE(fown
->signum
);
460 if (!sigio_perm(p
, fown
, signum
))
466 /* Queue a rt signal with the appropriate fd as its
467 value. We use SI_SIGIO as the source, not
468 SI_KERNEL, since kernel signals always get
469 delivered even if we can't queue. Failure to
470 queue in this case _should_ be reported; we fall
471 back to SIGIO in that case. --sct */
472 si
.si_signo
= signum
;
475 /* Make sure we are called with one of the POLL_*
476 reasons, otherwise we could leak kernel stack into
478 BUG_ON((reason
& __SI_MASK
) != __SI_POLL
);
479 if (reason
- POLL_IN
>= NSIGPOLL
)
482 si
.si_band
= band_table
[reason
- POLL_IN
];
484 if (!do_send_sig_info(signum
, &si
, p
, group
))
486 /* fall-through: fall back on the old plain SIGIO signal */
488 do_send_sig_info(SIGIO
, SEND_SIG_PRIV
, p
, group
);
492 void send_sigio(struct fown_struct
*fown
, int fd
, int band
)
494 struct task_struct
*p
;
499 read_lock(&fown
->lock
);
501 type
= fown
->pid_type
;
502 if (type
== PIDTYPE_MAX
) {
509 goto out_unlock_fown
;
511 read_lock(&tasklist_lock
);
512 do_each_pid_task(pid
, type
, p
) {
513 send_sigio_to_task(p
, fown
, fd
, band
, group
);
514 } while_each_pid_task(pid
, type
, p
);
515 read_unlock(&tasklist_lock
);
517 read_unlock(&fown
->lock
);
520 static void send_sigurg_to_task(struct task_struct
*p
,
521 struct fown_struct
*fown
, int group
)
523 if (sigio_perm(p
, fown
, SIGURG
))
524 do_send_sig_info(SIGURG
, SEND_SIG_PRIV
, p
, group
);
527 int send_sigurg(struct fown_struct
*fown
)
529 struct task_struct
*p
;
535 read_lock(&fown
->lock
);
537 type
= fown
->pid_type
;
538 if (type
== PIDTYPE_MAX
) {
545 goto out_unlock_fown
;
549 read_lock(&tasklist_lock
);
550 do_each_pid_task(pid
, type
, p
) {
551 send_sigurg_to_task(p
, fown
, group
);
552 } while_each_pid_task(pid
, type
, p
);
553 read_unlock(&tasklist_lock
);
555 read_unlock(&fown
->lock
);
559 static DEFINE_SPINLOCK(fasync_lock
);
560 static struct kmem_cache
*fasync_cache __read_mostly
;
562 static void fasync_free_rcu(struct rcu_head
*head
)
564 kmem_cache_free(fasync_cache
,
565 container_of(head
, struct fasync_struct
, fa_rcu
));
569 * Remove a fasync entry. If successfully removed, return
570 * positive and clear the FASYNC flag. If no entry exists,
571 * do nothing and return 0.
573 * NOTE! It is very important that the FASYNC flag always
574 * match the state "is the filp on a fasync list".
577 int fasync_remove_entry(struct file
*filp
, struct fasync_struct
**fapp
)
579 struct fasync_struct
*fa
, **fp
;
582 spin_lock(&filp
->f_lock
);
583 spin_lock(&fasync_lock
);
584 for (fp
= fapp
; (fa
= *fp
) != NULL
; fp
= &fa
->fa_next
) {
585 if (fa
->fa_file
!= filp
)
588 spin_lock_irq(&fa
->fa_lock
);
590 spin_unlock_irq(&fa
->fa_lock
);
593 call_rcu(&fa
->fa_rcu
, fasync_free_rcu
);
594 filp
->f_flags
&= ~FASYNC
;
598 spin_unlock(&fasync_lock
);
599 spin_unlock(&filp
->f_lock
);
603 struct fasync_struct
*fasync_alloc(void)
605 return kmem_cache_alloc(fasync_cache
, GFP_KERNEL
);
609 * NOTE! This can be used only for unused fasync entries:
610 * entries that actually got inserted on the fasync list
611 * need to be released by rcu - see fasync_remove_entry.
613 void fasync_free(struct fasync_struct
*new)
615 kmem_cache_free(fasync_cache
, new);
619 * Insert a new entry into the fasync list. Return the pointer to the
620 * old one if we didn't use the new one.
622 * NOTE! It is very important that the FASYNC flag always
623 * match the state "is the filp on a fasync list".
625 struct fasync_struct
*fasync_insert_entry(int fd
, struct file
*filp
, struct fasync_struct
**fapp
, struct fasync_struct
*new)
627 struct fasync_struct
*fa
, **fp
;
629 spin_lock(&filp
->f_lock
);
630 spin_lock(&fasync_lock
);
631 for (fp
= fapp
; (fa
= *fp
) != NULL
; fp
= &fa
->fa_next
) {
632 if (fa
->fa_file
!= filp
)
635 spin_lock_irq(&fa
->fa_lock
);
637 spin_unlock_irq(&fa
->fa_lock
);
641 spin_lock_init(&new->fa_lock
);
642 new->magic
= FASYNC_MAGIC
;
645 new->fa_next
= *fapp
;
646 rcu_assign_pointer(*fapp
, new);
647 filp
->f_flags
|= FASYNC
;
650 spin_unlock(&fasync_lock
);
651 spin_unlock(&filp
->f_lock
);
656 * Add a fasync entry. Return negative on error, positive if
657 * added, and zero if did nothing but change an existing one.
659 static int fasync_add_entry(int fd
, struct file
*filp
, struct fasync_struct
**fapp
)
661 struct fasync_struct
*new;
663 new = fasync_alloc();
668 * fasync_insert_entry() returns the old (update) entry if
671 * So free the (unused) new entry and return 0 to let the
672 * caller know that we didn't add any new fasync entries.
674 if (fasync_insert_entry(fd
, filp
, fapp
, new)) {
683 * fasync_helper() is used by almost all character device drivers
684 * to set up the fasync queue, and for regular files by the file
685 * lease code. It returns negative on error, 0 if it did no changes
686 * and positive if it added/deleted the entry.
688 int fasync_helper(int fd
, struct file
* filp
, int on
, struct fasync_struct
**fapp
)
691 return fasync_remove_entry(filp
, fapp
);
692 return fasync_add_entry(fd
, filp
, fapp
);
695 EXPORT_SYMBOL(fasync_helper
);
698 * rcu_read_lock() is held
700 static void kill_fasync_rcu(struct fasync_struct
*fa
, int sig
, int band
)
703 struct fown_struct
*fown
;
706 if (fa
->magic
!= FASYNC_MAGIC
) {
707 printk(KERN_ERR
"kill_fasync: bad magic number in "
711 spin_lock_irqsave(&fa
->fa_lock
, flags
);
713 fown
= &fa
->fa_file
->f_owner
;
714 /* Don't send SIGURG to processes which have not set a
715 queued signum: SIGURG has its own default signalling
717 if (!(sig
== SIGURG
&& fown
->signum
== 0))
718 send_sigio(fown
, fa
->fa_fd
, band
);
720 spin_unlock_irqrestore(&fa
->fa_lock
, flags
);
721 fa
= rcu_dereference(fa
->fa_next
);
725 void kill_fasync(struct fasync_struct
**fp
, int sig
, int band
)
727 /* First a quick test without locking: usually
732 kill_fasync_rcu(rcu_dereference(*fp
), sig
, band
);
736 EXPORT_SYMBOL(kill_fasync
);
738 static int __init
fcntl_init(void)
741 * Please add new bits here to ensure allocation uniqueness.
742 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
743 * is defined as O_NONBLOCK on some platforms and not on others.
745 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
747 (VALID_OPEN_FLAGS
& ~(O_NONBLOCK
| O_NDELAY
)) |
748 __FMODE_EXEC
| __FMODE_NONOTIFY
));
750 fasync_cache
= kmem_cache_create("fasync_cache",
751 sizeof(struct fasync_struct
), 0, SLAB_PANIC
, NULL
);
755 module_init(fcntl_init
)