4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/syscalls.h>
8 #include <linux/init.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>
27 #include <asm/siginfo.h>
28 #include <asm/uaccess.h>
30 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
32 static int setfl(int fd
, struct file
* filp
, unsigned long arg
)
34 struct inode
* inode
= file_inode(filp
);
38 * O_APPEND cannot be cleared if the file is marked as append-only
39 * and the file is open for write.
41 if (((arg
^ filp
->f_flags
) & O_APPEND
) && IS_APPEND(inode
))
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
))
49 /* required for strict SunOS emulation */
50 if (O_NONBLOCK
!= O_NDELAY
)
55 if (!filp
->f_mapping
|| !filp
->f_mapping
->a_ops
||
56 !filp
->f_mapping
->a_ops
->direct_IO
)
60 if (filp
->f_op
->check_flags
)
61 error
= filp
->f_op
->check_flags(arg
);
66 * ->fasync() is responsible for setting the FASYNC bit.
68 if (((arg
^ filp
->f_flags
) & FASYNC
) && filp
->f_op
->fasync
) {
69 error
= filp
->f_op
->fasync(fd
, filp
, (arg
& FASYNC
) != 0);
75 spin_lock(&filp
->f_lock
);
76 filp
->f_flags
= (arg
& SETFL_MASK
) | (filp
->f_flags
& ~SETFL_MASK
);
77 spin_unlock(&filp
->f_lock
);
83 static void f_modown(struct file
*filp
, struct pid
*pid
, enum pid_type type
,
86 write_lock_irq(&filp
->f_owner
.lock
);
87 if (force
|| !filp
->f_owner
.pid
) {
88 put_pid(filp
->f_owner
.pid
);
89 filp
->f_owner
.pid
= get_pid(pid
);
90 filp
->f_owner
.pid_type
= type
;
93 const struct cred
*cred
= current_cred();
94 filp
->f_owner
.uid
= cred
->uid
;
95 filp
->f_owner
.euid
= cred
->euid
;
98 write_unlock_irq(&filp
->f_owner
.lock
);
101 void __f_setown(struct file
*filp
, struct pid
*pid
, enum pid_type type
,
104 security_file_set_fowner(filp
);
105 f_modown(filp
, pid
, type
, force
);
107 EXPORT_SYMBOL(__f_setown
);
109 void f_setown(struct file
*filp
, unsigned long arg
, int force
)
120 pid
= find_vpid(who
);
121 __f_setown(filp
, pid
, type
, force
);
124 EXPORT_SYMBOL(f_setown
);
126 void f_delown(struct file
*filp
)
128 f_modown(filp
, NULL
, PIDTYPE_PID
, 1);
131 pid_t
f_getown(struct file
*filp
)
134 read_lock(&filp
->f_owner
.lock
);
135 pid
= pid_vnr(filp
->f_owner
.pid
);
136 if (filp
->f_owner
.pid_type
== PIDTYPE_PGID
)
138 read_unlock(&filp
->f_owner
.lock
);
142 static int f_setown_ex(struct file
*filp
, unsigned long arg
)
144 struct f_owner_ex __user
*owner_p
= (void __user
*)arg
;
145 struct f_owner_ex owner
;
150 ret
= copy_from_user(&owner
, owner_p
, sizeof(owner
));
154 switch (owner
.type
) {
172 pid
= find_vpid(owner
.pid
);
173 if (owner
.pid
&& !pid
)
176 __f_setown(filp
, pid
, type
, 1);
182 static int f_getown_ex(struct file
*filp
, unsigned long arg
)
184 struct f_owner_ex __user
*owner_p
= (void __user
*)arg
;
185 struct f_owner_ex owner
;
188 read_lock(&filp
->f_owner
.lock
);
189 owner
.pid
= pid_vnr(filp
->f_owner
.pid
);
190 switch (filp
->f_owner
.pid_type
) {
192 owner
.type
= F_OWNER_TID
;
196 owner
.type
= F_OWNER_PID
;
200 owner
.type
= F_OWNER_PGRP
;
208 read_unlock(&filp
->f_owner
.lock
);
211 ret
= copy_to_user(owner_p
, &owner
, sizeof(owner
));
218 #ifdef CONFIG_CHECKPOINT_RESTORE
219 static int f_getowner_uids(struct file
*filp
, unsigned long arg
)
221 struct user_namespace
*user_ns
= current_user_ns();
222 uid_t __user
*dst
= (void __user
*)arg
;
226 read_lock(&filp
->f_owner
.lock
);
227 src
[0] = from_kuid(user_ns
, filp
->f_owner
.uid
);
228 src
[1] = from_kuid(user_ns
, filp
->f_owner
.euid
);
229 read_unlock(&filp
->f_owner
.lock
);
231 err
= put_user(src
[0], &dst
[0]);
232 err
|= put_user(src
[1], &dst
[1]);
237 static int f_getowner_uids(struct file
*filp
, unsigned long arg
)
243 static long do_fcntl(int fd
, unsigned int cmd
, unsigned long arg
,
250 err
= f_dupfd(arg
, filp
, 0);
252 case F_DUPFD_CLOEXEC
:
253 err
= f_dupfd(arg
, filp
, O_CLOEXEC
);
256 err
= get_close_on_exec(fd
) ? FD_CLOEXEC
: 0;
260 set_close_on_exec(fd
, arg
& FD_CLOEXEC
);
266 err
= setfl(fd
, filp
, arg
);
268 #if BITS_PER_LONG != 32
269 /* 32-bit arches must use fcntl64() */
273 err
= fcntl_getlk(filp
, cmd
, (struct flock __user
*) arg
);
275 #if BITS_PER_LONG != 32
276 /* 32-bit arches must use fcntl64() */
283 err
= fcntl_setlk(fd
, filp
, cmd
, (struct flock __user
*) arg
);
287 * XXX If f_owner is a process group, the
288 * negative return value will get converted
289 * into an error. Oops. If we keep the
290 * current syscall conventions, the only way
291 * to fix this will be in libc.
293 err
= f_getown(filp
);
294 force_successful_syscall_return();
297 f_setown(filp
, arg
, 1);
301 err
= f_getown_ex(filp
, arg
);
304 err
= f_setown_ex(filp
, arg
);
306 case F_GETOWNER_UIDS
:
307 err
= f_getowner_uids(filp
, arg
);
310 err
= filp
->f_owner
.signum
;
313 /* arg == 0 restores default behaviour. */
314 if (!valid_signal(arg
)) {
318 filp
->f_owner
.signum
= arg
;
321 err
= fcntl_getlease(filp
);
324 err
= fcntl_setlease(fd
, filp
, arg
);
327 err
= fcntl_dirnotify(fd
, filp
, arg
);
331 err
= pipe_fcntl(filp
, cmd
, arg
);
335 err
= shmem_fcntl(filp
, cmd
, arg
);
343 static int check_fcntl_cmd(unsigned cmd
)
347 case F_DUPFD_CLOEXEC
:
356 SYSCALL_DEFINE3(fcntl
, unsigned int, fd
, unsigned int, cmd
, unsigned long, arg
)
358 struct fd f
= fdget_raw(fd
);
364 if (unlikely(f
.file
->f_mode
& FMODE_PATH
)) {
365 if (!check_fcntl_cmd(cmd
))
369 err
= security_file_fcntl(f
.file
, cmd
, arg
);
371 err
= do_fcntl(fd
, cmd
, arg
, f
.file
);
379 #if BITS_PER_LONG == 32
380 SYSCALL_DEFINE3(fcntl64
, unsigned int, fd
, unsigned int, cmd
,
383 struct fd f
= fdget_raw(fd
);
389 if (unlikely(f
.file
->f_mode
& FMODE_PATH
)) {
390 if (!check_fcntl_cmd(cmd
))
394 err
= security_file_fcntl(f
.file
, cmd
, arg
);
401 err
= fcntl_getlk64(f
.file
, cmd
, (struct flock64 __user
*) arg
);
407 err
= fcntl_setlk64(fd
, f
.file
, cmd
,
408 (struct flock64 __user
*) arg
);
411 err
= do_fcntl(fd
, cmd
, arg
, f
.file
);
421 /* Table to convert sigio signal codes into poll band bitmaps */
423 static const long band_table
[NSIGPOLL
] = {
424 POLLIN
| POLLRDNORM
, /* POLL_IN */
425 POLLOUT
| POLLWRNORM
| POLLWRBAND
, /* POLL_OUT */
426 POLLIN
| POLLRDNORM
| POLLMSG
, /* POLL_MSG */
427 POLLERR
, /* POLL_ERR */
428 POLLPRI
| POLLRDBAND
, /* POLL_PRI */
429 POLLHUP
| POLLERR
/* POLL_HUP */
432 static inline int sigio_perm(struct task_struct
*p
,
433 struct fown_struct
*fown
, int sig
)
435 const struct cred
*cred
;
439 cred
= __task_cred(p
);
440 ret
= ((uid_eq(fown
->euid
, GLOBAL_ROOT_UID
) ||
441 uid_eq(fown
->euid
, cred
->suid
) || uid_eq(fown
->euid
, cred
->uid
) ||
442 uid_eq(fown
->uid
, cred
->suid
) || uid_eq(fown
->uid
, cred
->uid
)) &&
443 !security_file_send_sigiotask(p
, fown
, sig
));
448 static void send_sigio_to_task(struct task_struct
*p
,
449 struct fown_struct
*fown
,
450 int fd
, int reason
, int group
)
453 * F_SETSIG can change ->signum lockless in parallel, make
454 * sure we read it once and use the same value throughout.
456 int signum
= ACCESS_ONCE(fown
->signum
);
458 if (!sigio_perm(p
, fown
, signum
))
464 /* Queue a rt signal with the appropriate fd as its
465 value. We use SI_SIGIO as the source, not
466 SI_KERNEL, since kernel signals always get
467 delivered even if we can't queue. Failure to
468 queue in this case _should_ be reported; we fall
469 back to SIGIO in that case. --sct */
470 si
.si_signo
= signum
;
473 /* Make sure we are called with one of the POLL_*
474 reasons, otherwise we could leak kernel stack into
476 BUG_ON((reason
& __SI_MASK
) != __SI_POLL
);
477 if (reason
- POLL_IN
>= NSIGPOLL
)
480 si
.si_band
= band_table
[reason
- POLL_IN
];
482 if (!do_send_sig_info(signum
, &si
, p
, group
))
484 /* fall-through: fall back on the old plain SIGIO signal */
486 do_send_sig_info(SIGIO
, SEND_SIG_PRIV
, p
, group
);
490 void send_sigio(struct fown_struct
*fown
, int fd
, int band
)
492 struct task_struct
*p
;
497 read_lock(&fown
->lock
);
499 type
= fown
->pid_type
;
500 if (type
== PIDTYPE_MAX
) {
507 goto out_unlock_fown
;
509 read_lock(&tasklist_lock
);
510 do_each_pid_task(pid
, type
, p
) {
511 send_sigio_to_task(p
, fown
, fd
, band
, group
);
512 } while_each_pid_task(pid
, type
, p
);
513 read_unlock(&tasklist_lock
);
515 read_unlock(&fown
->lock
);
518 static void send_sigurg_to_task(struct task_struct
*p
,
519 struct fown_struct
*fown
, int group
)
521 if (sigio_perm(p
, fown
, SIGURG
))
522 do_send_sig_info(SIGURG
, SEND_SIG_PRIV
, p
, group
);
525 int send_sigurg(struct fown_struct
*fown
)
527 struct task_struct
*p
;
533 read_lock(&fown
->lock
);
535 type
= fown
->pid_type
;
536 if (type
== PIDTYPE_MAX
) {
543 goto out_unlock_fown
;
547 read_lock(&tasklist_lock
);
548 do_each_pid_task(pid
, type
, p
) {
549 send_sigurg_to_task(p
, fown
, group
);
550 } while_each_pid_task(pid
, type
, p
);
551 read_unlock(&tasklist_lock
);
553 read_unlock(&fown
->lock
);
557 static DEFINE_SPINLOCK(fasync_lock
);
558 static struct kmem_cache
*fasync_cache __read_mostly
;
560 static void fasync_free_rcu(struct rcu_head
*head
)
562 kmem_cache_free(fasync_cache
,
563 container_of(head
, struct fasync_struct
, fa_rcu
));
567 * Remove a fasync entry. If successfully removed, return
568 * positive and clear the FASYNC flag. If no entry exists,
569 * do nothing and return 0.
571 * NOTE! It is very important that the FASYNC flag always
572 * match the state "is the filp on a fasync list".
575 int fasync_remove_entry(struct file
*filp
, struct fasync_struct
**fapp
)
577 struct fasync_struct
*fa
, **fp
;
580 spin_lock(&filp
->f_lock
);
581 spin_lock(&fasync_lock
);
582 for (fp
= fapp
; (fa
= *fp
) != NULL
; fp
= &fa
->fa_next
) {
583 if (fa
->fa_file
!= filp
)
586 spin_lock_irq(&fa
->fa_lock
);
588 spin_unlock_irq(&fa
->fa_lock
);
591 call_rcu(&fa
->fa_rcu
, fasync_free_rcu
);
592 filp
->f_flags
&= ~FASYNC
;
596 spin_unlock(&fasync_lock
);
597 spin_unlock(&filp
->f_lock
);
601 struct fasync_struct
*fasync_alloc(void)
603 return kmem_cache_alloc(fasync_cache
, GFP_KERNEL
);
607 * NOTE! This can be used only for unused fasync entries:
608 * entries that actually got inserted on the fasync list
609 * need to be released by rcu - see fasync_remove_entry.
611 void fasync_free(struct fasync_struct
*new)
613 kmem_cache_free(fasync_cache
, new);
617 * Insert a new entry into the fasync list. Return the pointer to the
618 * old one if we didn't use the new one.
620 * NOTE! It is very important that the FASYNC flag always
621 * match the state "is the filp on a fasync list".
623 struct fasync_struct
*fasync_insert_entry(int fd
, struct file
*filp
, struct fasync_struct
**fapp
, struct fasync_struct
*new)
625 struct fasync_struct
*fa
, **fp
;
627 spin_lock(&filp
->f_lock
);
628 spin_lock(&fasync_lock
);
629 for (fp
= fapp
; (fa
= *fp
) != NULL
; fp
= &fa
->fa_next
) {
630 if (fa
->fa_file
!= filp
)
633 spin_lock_irq(&fa
->fa_lock
);
635 spin_unlock_irq(&fa
->fa_lock
);
639 spin_lock_init(&new->fa_lock
);
640 new->magic
= FASYNC_MAGIC
;
643 new->fa_next
= *fapp
;
644 rcu_assign_pointer(*fapp
, new);
645 filp
->f_flags
|= FASYNC
;
648 spin_unlock(&fasync_lock
);
649 spin_unlock(&filp
->f_lock
);
654 * Add a fasync entry. Return negative on error, positive if
655 * added, and zero if did nothing but change an existing one.
657 static int fasync_add_entry(int fd
, struct file
*filp
, struct fasync_struct
**fapp
)
659 struct fasync_struct
*new;
661 new = fasync_alloc();
666 * fasync_insert_entry() returns the old (update) entry if
669 * So free the (unused) new entry and return 0 to let the
670 * caller know that we didn't add any new fasync entries.
672 if (fasync_insert_entry(fd
, filp
, fapp
, new)) {
681 * fasync_helper() is used by almost all character device drivers
682 * to set up the fasync queue, and for regular files by the file
683 * lease code. It returns negative on error, 0 if it did no changes
684 * and positive if it added/deleted the entry.
686 int fasync_helper(int fd
, struct file
* filp
, int on
, struct fasync_struct
**fapp
)
689 return fasync_remove_entry(filp
, fapp
);
690 return fasync_add_entry(fd
, filp
, fapp
);
693 EXPORT_SYMBOL(fasync_helper
);
696 * rcu_read_lock() is held
698 static void kill_fasync_rcu(struct fasync_struct
*fa
, int sig
, int band
)
701 struct fown_struct
*fown
;
704 if (fa
->magic
!= FASYNC_MAGIC
) {
705 printk(KERN_ERR
"kill_fasync: bad magic number in "
709 spin_lock_irqsave(&fa
->fa_lock
, flags
);
711 fown
= &fa
->fa_file
->f_owner
;
712 /* Don't send SIGURG to processes which have not set a
713 queued signum: SIGURG has its own default signalling
715 if (!(sig
== SIGURG
&& fown
->signum
== 0))
716 send_sigio(fown
, fa
->fa_fd
, band
);
718 spin_unlock_irqrestore(&fa
->fa_lock
, flags
);
719 fa
= rcu_dereference(fa
->fa_next
);
723 void kill_fasync(struct fasync_struct
**fp
, int sig
, int band
)
725 /* First a quick test without locking: usually
730 kill_fasync_rcu(rcu_dereference(*fp
), sig
, band
);
734 EXPORT_SYMBOL(kill_fasync
);
736 static int __init
fcntl_init(void)
739 * Please add new bits here to ensure allocation uniqueness.
740 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
741 * is defined as O_NONBLOCK on some platforms and not on others.
743 BUILD_BUG_ON(20 - 1 /* for O_RDONLY being 0 */ != HWEIGHT32(
744 O_RDONLY
| O_WRONLY
| O_RDWR
|
745 O_CREAT
| O_EXCL
| O_NOCTTY
|
746 O_TRUNC
| O_APPEND
| /* O_NONBLOCK | */
747 __O_SYNC
| O_DSYNC
| FASYNC
|
748 O_DIRECT
| O_LARGEFILE
| O_DIRECTORY
|
749 O_NOFOLLOW
| O_NOATIME
| O_CLOEXEC
|
750 __FMODE_EXEC
| O_PATH
| __O_TMPFILE
753 fasync_cache
= kmem_cache_create("fasync_cache",
754 sizeof(struct fasync_struct
), 0, SLAB_PANIC
, NULL
);
758 module_init(fcntl_init
)