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
)
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
)
61 if (filp
->f_op
->check_flags
)
62 error
= filp
->f_op
->check_flags(arg
);
67 * ->fasync() is responsible for setting the FASYNC bit.
69 if (((arg
^ filp
->f_flags
) & FASYNC
) && filp
->f_op
->fasync
) {
70 error
= filp
->f_op
->fasync(fd
, filp
, (arg
& FASYNC
) != 0);
76 spin_lock(&filp
->f_lock
);
77 filp
->f_flags
= (arg
& SETFL_MASK
) | (filp
->f_flags
& ~SETFL_MASK
);
78 spin_unlock(&filp
->f_lock
);
84 static void f_modown(struct file
*filp
, struct pid
*pid
, enum pid_type type
,
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
;
94 const struct cred
*cred
= current_cred();
95 filp
->f_owner
.uid
= cred
->uid
;
96 filp
->f_owner
.euid
= cred
->euid
;
99 write_unlock_irq(&filp
->f_owner
.lock
);
102 void __f_setown(struct file
*filp
, struct pid
*pid
, enum pid_type type
,
105 security_file_set_fowner(filp
);
106 f_modown(filp
, pid
, type
, force
);
108 EXPORT_SYMBOL(__f_setown
);
110 void f_setown(struct file
*filp
, unsigned long arg
, int force
)
117 /* avoid overflow below */
125 pid
= find_vpid(who
);
126 __f_setown(filp
, pid
, type
, force
);
129 EXPORT_SYMBOL(f_setown
);
131 void f_delown(struct file
*filp
)
133 f_modown(filp
, NULL
, PIDTYPE_PID
, 1);
136 pid_t
f_getown(struct file
*filp
)
139 read_lock(&filp
->f_owner
.lock
);
140 pid
= pid_vnr(filp
->f_owner
.pid
);
141 if (filp
->f_owner
.pid_type
== PIDTYPE_PGID
)
143 read_unlock(&filp
->f_owner
.lock
);
147 static int f_setown_ex(struct file
*filp
, unsigned long arg
)
149 struct f_owner_ex __user
*owner_p
= (void __user
*)arg
;
150 struct f_owner_ex owner
;
155 ret
= copy_from_user(&owner
, owner_p
, sizeof(owner
));
159 switch (owner
.type
) {
177 pid
= find_vpid(owner
.pid
);
178 if (owner
.pid
&& !pid
)
181 __f_setown(filp
, pid
, type
, 1);
187 static int f_getown_ex(struct file
*filp
, unsigned long arg
)
189 struct f_owner_ex __user
*owner_p
= (void __user
*)arg
;
190 struct f_owner_ex owner
;
193 read_lock(&filp
->f_owner
.lock
);
194 owner
.pid
= pid_vnr(filp
->f_owner
.pid
);
195 switch (filp
->f_owner
.pid_type
) {
197 owner
.type
= F_OWNER_TID
;
201 owner
.type
= F_OWNER_PID
;
205 owner
.type
= F_OWNER_PGRP
;
213 read_unlock(&filp
->f_owner
.lock
);
216 ret
= copy_to_user(owner_p
, &owner
, sizeof(owner
));
223 #ifdef CONFIG_CHECKPOINT_RESTORE
224 static int f_getowner_uids(struct file
*filp
, unsigned long arg
)
226 struct user_namespace
*user_ns
= current_user_ns();
227 uid_t __user
*dst
= (void __user
*)arg
;
231 read_lock(&filp
->f_owner
.lock
);
232 src
[0] = from_kuid(user_ns
, filp
->f_owner
.uid
);
233 src
[1] = from_kuid(user_ns
, filp
->f_owner
.euid
);
234 read_unlock(&filp
->f_owner
.lock
);
236 err
= put_user(src
[0], &dst
[0]);
237 err
|= put_user(src
[1], &dst
[1]);
242 static int f_getowner_uids(struct file
*filp
, unsigned long arg
)
248 static long do_fcntl(int fd
, unsigned int cmd
, unsigned long arg
,
255 err
= f_dupfd(arg
, filp
, 0);
257 case F_DUPFD_CLOEXEC
:
258 err
= f_dupfd(arg
, filp
, O_CLOEXEC
);
261 err
= get_close_on_exec(fd
) ? FD_CLOEXEC
: 0;
265 set_close_on_exec(fd
, arg
& FD_CLOEXEC
);
271 err
= setfl(fd
, filp
, arg
);
273 #if BITS_PER_LONG != 32
274 /* 32-bit arches must use fcntl64() */
278 err
= fcntl_getlk(filp
, cmd
, (struct flock __user
*) arg
);
280 #if BITS_PER_LONG != 32
281 /* 32-bit arches must use fcntl64() */
288 err
= fcntl_setlk(fd
, filp
, cmd
, (struct flock __user
*) arg
);
292 * XXX If f_owner is a process group, the
293 * negative return value will get converted
294 * into an error. Oops. If we keep the
295 * current syscall conventions, the only way
296 * to fix this will be in libc.
298 err
= f_getown(filp
);
299 force_successful_syscall_return();
302 f_setown(filp
, arg
, 1);
306 err
= f_getown_ex(filp
, arg
);
309 err
= f_setown_ex(filp
, arg
);
311 case F_GETOWNER_UIDS
:
312 err
= f_getowner_uids(filp
, arg
);
315 err
= filp
->f_owner
.signum
;
318 /* arg == 0 restores default behaviour. */
319 if (!valid_signal(arg
)) {
323 filp
->f_owner
.signum
= arg
;
326 err
= fcntl_getlease(filp
);
329 err
= fcntl_setlease(fd
, filp
, arg
);
332 err
= fcntl_dirnotify(fd
, filp
, arg
);
336 err
= pipe_fcntl(filp
, cmd
, arg
);
340 err
= shmem_fcntl(filp
, cmd
, arg
);
348 static int check_fcntl_cmd(unsigned cmd
)
352 case F_DUPFD_CLOEXEC
:
361 SYSCALL_DEFINE3(fcntl
, unsigned int, fd
, unsigned int, cmd
, unsigned long, arg
)
363 struct fd f
= fdget_raw(fd
);
369 if (unlikely(f
.file
->f_mode
& FMODE_PATH
)) {
370 if (!check_fcntl_cmd(cmd
))
374 err
= security_file_fcntl(f
.file
, cmd
, arg
);
376 err
= do_fcntl(fd
, cmd
, arg
, f
.file
);
384 #if BITS_PER_LONG == 32
385 SYSCALL_DEFINE3(fcntl64
, unsigned int, fd
, unsigned int, cmd
,
388 struct fd f
= fdget_raw(fd
);
394 if (unlikely(f
.file
->f_mode
& FMODE_PATH
)) {
395 if (!check_fcntl_cmd(cmd
))
399 err
= security_file_fcntl(f
.file
, cmd
, arg
);
406 err
= fcntl_getlk64(f
.file
, cmd
, (struct flock64 __user
*) arg
);
412 err
= fcntl_setlk64(fd
, f
.file
, cmd
,
413 (struct flock64 __user
*) arg
);
416 err
= do_fcntl(fd
, cmd
, arg
, f
.file
);
426 /* Table to convert sigio signal codes into poll band bitmaps */
428 static const long band_table
[NSIGPOLL
] = {
429 POLLIN
| POLLRDNORM
, /* POLL_IN */
430 POLLOUT
| POLLWRNORM
| POLLWRBAND
, /* POLL_OUT */
431 POLLIN
| POLLRDNORM
| POLLMSG
, /* POLL_MSG */
432 POLLERR
, /* POLL_ERR */
433 POLLPRI
| POLLRDBAND
, /* POLL_PRI */
434 POLLHUP
| POLLERR
/* POLL_HUP */
437 static inline int sigio_perm(struct task_struct
*p
,
438 struct fown_struct
*fown
, int sig
)
440 const struct cred
*cred
;
444 cred
= __task_cred(p
);
445 ret
= ((uid_eq(fown
->euid
, GLOBAL_ROOT_UID
) ||
446 uid_eq(fown
->euid
, cred
->suid
) || uid_eq(fown
->euid
, cred
->uid
) ||
447 uid_eq(fown
->uid
, cred
->suid
) || uid_eq(fown
->uid
, cred
->uid
)) &&
448 !security_file_send_sigiotask(p
, fown
, sig
));
453 static void send_sigio_to_task(struct task_struct
*p
,
454 struct fown_struct
*fown
,
455 int fd
, int reason
, int group
)
458 * F_SETSIG can change ->signum lockless in parallel, make
459 * sure we read it once and use the same value throughout.
461 int signum
= ACCESS_ONCE(fown
->signum
);
463 if (!sigio_perm(p
, fown
, signum
))
469 /* Queue a rt signal with the appropriate fd as its
470 value. We use SI_SIGIO as the source, not
471 SI_KERNEL, since kernel signals always get
472 delivered even if we can't queue. Failure to
473 queue in this case _should_ be reported; we fall
474 back to SIGIO in that case. --sct */
475 si
.si_signo
= signum
;
478 /* Make sure we are called with one of the POLL_*
479 reasons, otherwise we could leak kernel stack into
481 BUG_ON((reason
& __SI_MASK
) != __SI_POLL
);
482 if (reason
- POLL_IN
>= NSIGPOLL
)
485 si
.si_band
= band_table
[reason
- POLL_IN
];
487 if (!do_send_sig_info(signum
, &si
, p
, group
))
489 /* fall-through: fall back on the old plain SIGIO signal */
491 do_send_sig_info(SIGIO
, SEND_SIG_PRIV
, p
, group
);
495 void send_sigio(struct fown_struct
*fown
, int fd
, int band
)
497 struct task_struct
*p
;
502 read_lock(&fown
->lock
);
504 type
= fown
->pid_type
;
505 if (type
== PIDTYPE_MAX
) {
512 goto out_unlock_fown
;
514 read_lock(&tasklist_lock
);
515 do_each_pid_task(pid
, type
, p
) {
516 send_sigio_to_task(p
, fown
, fd
, band
, group
);
517 } while_each_pid_task(pid
, type
, p
);
518 read_unlock(&tasklist_lock
);
520 read_unlock(&fown
->lock
);
523 static void send_sigurg_to_task(struct task_struct
*p
,
524 struct fown_struct
*fown
, int group
)
526 if (sigio_perm(p
, fown
, SIGURG
))
527 do_send_sig_info(SIGURG
, SEND_SIG_PRIV
, p
, group
);
530 int send_sigurg(struct fown_struct
*fown
)
532 struct task_struct
*p
;
538 read_lock(&fown
->lock
);
540 type
= fown
->pid_type
;
541 if (type
== PIDTYPE_MAX
) {
548 goto out_unlock_fown
;
552 read_lock(&tasklist_lock
);
553 do_each_pid_task(pid
, type
, p
) {
554 send_sigurg_to_task(p
, fown
, group
);
555 } while_each_pid_task(pid
, type
, p
);
556 read_unlock(&tasklist_lock
);
558 read_unlock(&fown
->lock
);
562 static DEFINE_SPINLOCK(fasync_lock
);
563 static struct kmem_cache
*fasync_cache __read_mostly
;
565 static void fasync_free_rcu(struct rcu_head
*head
)
567 kmem_cache_free(fasync_cache
,
568 container_of(head
, struct fasync_struct
, fa_rcu
));
572 * Remove a fasync entry. If successfully removed, return
573 * positive and clear the FASYNC flag. If no entry exists,
574 * do nothing and return 0.
576 * NOTE! It is very important that the FASYNC flag always
577 * match the state "is the filp on a fasync list".
580 int fasync_remove_entry(struct file
*filp
, struct fasync_struct
**fapp
)
582 struct fasync_struct
*fa
, **fp
;
585 spin_lock(&filp
->f_lock
);
586 spin_lock(&fasync_lock
);
587 for (fp
= fapp
; (fa
= *fp
) != NULL
; fp
= &fa
->fa_next
) {
588 if (fa
->fa_file
!= filp
)
591 spin_lock_irq(&fa
->fa_lock
);
593 spin_unlock_irq(&fa
->fa_lock
);
596 call_rcu(&fa
->fa_rcu
, fasync_free_rcu
);
597 filp
->f_flags
&= ~FASYNC
;
601 spin_unlock(&fasync_lock
);
602 spin_unlock(&filp
->f_lock
);
606 struct fasync_struct
*fasync_alloc(void)
608 return kmem_cache_alloc(fasync_cache
, GFP_KERNEL
);
612 * NOTE! This can be used only for unused fasync entries:
613 * entries that actually got inserted on the fasync list
614 * need to be released by rcu - see fasync_remove_entry.
616 void fasync_free(struct fasync_struct
*new)
618 kmem_cache_free(fasync_cache
, new);
622 * Insert a new entry into the fasync list. Return the pointer to the
623 * old one if we didn't use the new one.
625 * NOTE! It is very important that the FASYNC flag always
626 * match the state "is the filp on a fasync list".
628 struct fasync_struct
*fasync_insert_entry(int fd
, struct file
*filp
, struct fasync_struct
**fapp
, struct fasync_struct
*new)
630 struct fasync_struct
*fa
, **fp
;
632 spin_lock(&filp
->f_lock
);
633 spin_lock(&fasync_lock
);
634 for (fp
= fapp
; (fa
= *fp
) != NULL
; fp
= &fa
->fa_next
) {
635 if (fa
->fa_file
!= filp
)
638 spin_lock_irq(&fa
->fa_lock
);
640 spin_unlock_irq(&fa
->fa_lock
);
644 spin_lock_init(&new->fa_lock
);
645 new->magic
= FASYNC_MAGIC
;
648 new->fa_next
= *fapp
;
649 rcu_assign_pointer(*fapp
, new);
650 filp
->f_flags
|= FASYNC
;
653 spin_unlock(&fasync_lock
);
654 spin_unlock(&filp
->f_lock
);
659 * Add a fasync entry. Return negative on error, positive if
660 * added, and zero if did nothing but change an existing one.
662 static int fasync_add_entry(int fd
, struct file
*filp
, struct fasync_struct
**fapp
)
664 struct fasync_struct
*new;
666 new = fasync_alloc();
671 * fasync_insert_entry() returns the old (update) entry if
674 * So free the (unused) new entry and return 0 to let the
675 * caller know that we didn't add any new fasync entries.
677 if (fasync_insert_entry(fd
, filp
, fapp
, new)) {
686 * fasync_helper() is used by almost all character device drivers
687 * to set up the fasync queue, and for regular files by the file
688 * lease code. It returns negative on error, 0 if it did no changes
689 * and positive if it added/deleted the entry.
691 int fasync_helper(int fd
, struct file
* filp
, int on
, struct fasync_struct
**fapp
)
694 return fasync_remove_entry(filp
, fapp
);
695 return fasync_add_entry(fd
, filp
, fapp
);
698 EXPORT_SYMBOL(fasync_helper
);
701 * rcu_read_lock() is held
703 static void kill_fasync_rcu(struct fasync_struct
*fa
, int sig
, int band
)
706 struct fown_struct
*fown
;
709 if (fa
->magic
!= FASYNC_MAGIC
) {
710 printk(KERN_ERR
"kill_fasync: bad magic number in "
714 spin_lock_irqsave(&fa
->fa_lock
, flags
);
716 fown
= &fa
->fa_file
->f_owner
;
717 /* Don't send SIGURG to processes which have not set a
718 queued signum: SIGURG has its own default signalling
720 if (!(sig
== SIGURG
&& fown
->signum
== 0))
721 send_sigio(fown
, fa
->fa_fd
, band
);
723 spin_unlock_irqrestore(&fa
->fa_lock
, flags
);
724 fa
= rcu_dereference(fa
->fa_next
);
728 void kill_fasync(struct fasync_struct
**fp
, int sig
, int band
)
730 /* First a quick test without locking: usually
735 kill_fasync_rcu(rcu_dereference(*fp
), sig
, band
);
739 EXPORT_SYMBOL(kill_fasync
);
741 static int __init
fcntl_init(void)
744 * Please add new bits here to ensure allocation uniqueness.
745 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
746 * is defined as O_NONBLOCK on some platforms and not on others.
748 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
750 (VALID_OPEN_FLAGS
& ~(O_NONBLOCK
| O_NDELAY
)) |
751 __FMODE_EXEC
| __FMODE_NONOTIFY
));
753 fasync_cache
= kmem_cache_create("fasync_cache",
754 sizeof(struct fasync_struct
), 0, SLAB_PANIC
, NULL
);
758 module_init(fcntl_init
)