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/security.h>
18 #include <linux/ptrace.h>
19 #include <linux/signal.h>
20 #include <linux/rcupdate.h>
21 #include <linux/pid_namespace.h>
24 #include <asm/siginfo.h>
25 #include <asm/uaccess.h>
27 void set_close_on_exec(unsigned int fd
, int flag
)
29 struct files_struct
*files
= current
->files
;
31 spin_lock(&files
->file_lock
);
32 fdt
= files_fdtable(files
);
34 FD_SET(fd
, fdt
->close_on_exec
);
36 FD_CLR(fd
, fdt
->close_on_exec
);
37 spin_unlock(&files
->file_lock
);
40 static int get_close_on_exec(unsigned int fd
)
42 struct files_struct
*files
= current
->files
;
46 fdt
= files_fdtable(files
);
47 res
= FD_ISSET(fd
, fdt
->close_on_exec
);
53 * locate_fd finds a free file descriptor in the open_fds fdset,
54 * expanding the fd arrays if necessary. Must be called with the
55 * file_lock held for write.
58 static int locate_fd(unsigned int orig_start
, int cloexec
)
60 struct files_struct
*files
= current
->files
;
66 spin_lock(&files
->file_lock
);
69 if (orig_start
>= current
->signal
->rlim
[RLIMIT_NOFILE
].rlim_cur
)
73 fdt
= files_fdtable(files
);
75 * Someone might have closed fd's in the range
76 * orig_start..fdt->next_fd
79 if (start
< files
->next_fd
)
80 start
= files
->next_fd
;
83 if (start
< fdt
->max_fds
)
84 newfd
= find_next_zero_bit(fdt
->open_fds
->fds_bits
,
88 if (newfd
>= current
->signal
->rlim
[RLIMIT_NOFILE
].rlim_cur
)
91 error
= expand_files(files
, newfd
);
96 * If we needed to expand the fs array we
97 * might have blocked - try again.
102 if (start
<= files
->next_fd
)
103 files
->next_fd
= newfd
+ 1;
105 FD_SET(newfd
, fdt
->open_fds
);
107 FD_SET(newfd
, fdt
->close_on_exec
);
109 FD_CLR(newfd
, fdt
->close_on_exec
);
113 spin_unlock(&files
->file_lock
);
117 static int dupfd(struct file
*file
, unsigned int start
, int cloexec
)
119 int fd
= locate_fd(start
, cloexec
);
121 fd_install(fd
, file
);
128 asmlinkage
long sys_dup3(unsigned int oldfd
, unsigned int newfd
, int flags
)
131 struct file
* file
, *tofree
;
132 struct files_struct
* files
= current
->files
;
135 if ((flags
& ~O_CLOEXEC
) != 0)
138 spin_lock(&files
->file_lock
);
139 if (!(file
= fcheck(oldfd
)))
142 if (unlikely(newfd
== oldfd
)) {
143 if (flags
& O_CLOEXEC
) {
144 fdt
= files_fdtable(files
);
145 FD_SET(newfd
, fdt
->close_on_exec
);
150 if (newfd
>= current
->signal
->rlim
[RLIMIT_NOFILE
].rlim_cur
)
152 get_file(file
); /* We are now finished with oldfd */
154 err
= expand_files(files
, newfd
);
158 /* To avoid races with open() and dup(), we will mark the fd as
159 * in-use in the open-file bitmap throughout the entire dup2()
160 * process. This is quite safe: do_close() uses the fd array
161 * entry, not the bitmap, to decide what work needs to be
163 /* Doesn't work. open() might be there first. --AV */
165 /* Yes. It's a race. In user space. Nothing sane to do */
167 fdt
= files_fdtable(files
);
168 tofree
= fdt
->fd
[newfd
];
169 if (!tofree
&& FD_ISSET(newfd
, fdt
->open_fds
))
172 rcu_assign_pointer(fdt
->fd
[newfd
], file
);
173 FD_SET(newfd
, fdt
->open_fds
);
174 if (flags
& O_CLOEXEC
)
175 FD_SET(newfd
, fdt
->close_on_exec
);
177 FD_CLR(newfd
, fdt
->close_on_exec
);
178 spin_unlock(&files
->file_lock
);
181 filp_close(tofree
, files
);
186 spin_unlock(&files
->file_lock
);
190 spin_unlock(&files
->file_lock
);
195 asmlinkage
long sys_dup2(unsigned int oldfd
, unsigned int newfd
)
197 return sys_dup3(oldfd
, newfd
, 0);
200 asmlinkage
long sys_dup(unsigned int fildes
)
203 struct file
* file
= fget(fildes
);
206 ret
= dupfd(file
, 0, 0);
210 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
212 static int setfl(int fd
, struct file
* filp
, unsigned long arg
)
214 struct inode
* inode
= filp
->f_path
.dentry
->d_inode
;
218 * O_APPEND cannot be cleared if the file is marked as append-only
219 * and the file is open for write.
221 if (((arg
^ filp
->f_flags
) & O_APPEND
) && IS_APPEND(inode
))
224 /* O_NOATIME can only be set by the owner or superuser */
225 if ((arg
& O_NOATIME
) && !(filp
->f_flags
& O_NOATIME
))
226 if (!is_owner_or_cap(inode
))
229 /* required for strict SunOS emulation */
230 if (O_NONBLOCK
!= O_NDELAY
)
234 if (arg
& O_DIRECT
) {
235 if (!filp
->f_mapping
|| !filp
->f_mapping
->a_ops
||
236 !filp
->f_mapping
->a_ops
->direct_IO
)
240 if (filp
->f_op
&& filp
->f_op
->check_flags
)
241 error
= filp
->f_op
->check_flags(arg
);
245 if ((arg
^ filp
->f_flags
) & FASYNC
) {
246 if (filp
->f_op
&& filp
->f_op
->fasync
) {
247 error
= filp
->f_op
->fasync(fd
, filp
, (arg
& FASYNC
) != 0);
253 filp
->f_flags
= (arg
& SETFL_MASK
) | (filp
->f_flags
& ~SETFL_MASK
);
258 static void f_modown(struct file
*filp
, struct pid
*pid
, enum pid_type type
,
259 uid_t uid
, uid_t euid
, int force
)
261 write_lock_irq(&filp
->f_owner
.lock
);
262 if (force
|| !filp
->f_owner
.pid
) {
263 put_pid(filp
->f_owner
.pid
);
264 filp
->f_owner
.pid
= get_pid(pid
);
265 filp
->f_owner
.pid_type
= type
;
266 filp
->f_owner
.uid
= uid
;
267 filp
->f_owner
.euid
= euid
;
269 write_unlock_irq(&filp
->f_owner
.lock
);
272 int __f_setown(struct file
*filp
, struct pid
*pid
, enum pid_type type
,
277 err
= security_file_set_fowner(filp
);
281 f_modown(filp
, pid
, type
, current
->uid
, current
->euid
, force
);
284 EXPORT_SYMBOL(__f_setown
);
286 int f_setown(struct file
*filp
, unsigned long arg
, int force
)
298 pid
= find_vpid(who
);
299 result
= __f_setown(filp
, pid
, type
, force
);
303 EXPORT_SYMBOL(f_setown
);
305 void f_delown(struct file
*filp
)
307 f_modown(filp
, NULL
, PIDTYPE_PID
, 0, 0, 1);
310 pid_t
f_getown(struct file
*filp
)
313 read_lock(&filp
->f_owner
.lock
);
314 pid
= pid_vnr(filp
->f_owner
.pid
);
315 if (filp
->f_owner
.pid_type
== PIDTYPE_PGID
)
317 read_unlock(&filp
->f_owner
.lock
);
321 static long do_fcntl(int fd
, unsigned int cmd
, unsigned long arg
,
328 case F_DUPFD_CLOEXEC
:
330 err
= dupfd(filp
, arg
, cmd
== F_DUPFD_CLOEXEC
);
333 err
= get_close_on_exec(fd
) ? FD_CLOEXEC
: 0;
337 set_close_on_exec(fd
, arg
& FD_CLOEXEC
);
343 err
= setfl(fd
, filp
, arg
);
346 err
= fcntl_getlk(filp
, (struct flock __user
*) arg
);
350 err
= fcntl_setlk(fd
, filp
, cmd
, (struct flock __user
*) arg
);
354 * XXX If f_owner is a process group, the
355 * negative return value will get converted
356 * into an error. Oops. If we keep the
357 * current syscall conventions, the only way
358 * to fix this will be in libc.
360 err
= f_getown(filp
);
361 force_successful_syscall_return();
364 err
= f_setown(filp
, arg
, 1);
367 err
= filp
->f_owner
.signum
;
370 /* arg == 0 restores default behaviour. */
371 if (!valid_signal(arg
)) {
375 filp
->f_owner
.signum
= arg
;
378 err
= fcntl_getlease(filp
);
381 err
= fcntl_setlease(fd
, filp
, arg
);
384 err
= fcntl_dirnotify(fd
, filp
, arg
);
392 asmlinkage
long sys_fcntl(unsigned int fd
, unsigned int cmd
, unsigned long arg
)
401 err
= security_file_fcntl(filp
, cmd
, arg
);
407 err
= do_fcntl(fd
, cmd
, arg
, filp
);
414 #if BITS_PER_LONG == 32
415 asmlinkage
long sys_fcntl64(unsigned int fd
, unsigned int cmd
, unsigned long arg
)
425 err
= security_file_fcntl(filp
, cmd
, arg
);
434 err
= fcntl_getlk64(filp
, (struct flock64 __user
*) arg
);
438 err
= fcntl_setlk64(fd
, filp
, cmd
,
439 (struct flock64 __user
*) arg
);
442 err
= do_fcntl(fd
, cmd
, arg
, filp
);
451 /* Table to convert sigio signal codes into poll band bitmaps */
453 static const long band_table
[NSIGPOLL
] = {
454 POLLIN
| POLLRDNORM
, /* POLL_IN */
455 POLLOUT
| POLLWRNORM
| POLLWRBAND
, /* POLL_OUT */
456 POLLIN
| POLLRDNORM
| POLLMSG
, /* POLL_MSG */
457 POLLERR
, /* POLL_ERR */
458 POLLPRI
| POLLRDBAND
, /* POLL_PRI */
459 POLLHUP
| POLLERR
/* POLL_HUP */
462 static inline int sigio_perm(struct task_struct
*p
,
463 struct fown_struct
*fown
, int sig
)
465 return (((fown
->euid
== 0) ||
466 (fown
->euid
== p
->suid
) || (fown
->euid
== p
->uid
) ||
467 (fown
->uid
== p
->suid
) || (fown
->uid
== p
->uid
)) &&
468 !security_file_send_sigiotask(p
, fown
, sig
));
471 static void send_sigio_to_task(struct task_struct
*p
,
472 struct fown_struct
*fown
,
476 if (!sigio_perm(p
, fown
, fown
->signum
))
479 switch (fown
->signum
) {
482 /* Queue a rt signal with the appropriate fd as its
483 value. We use SI_SIGIO as the source, not
484 SI_KERNEL, since kernel signals always get
485 delivered even if we can't queue. Failure to
486 queue in this case _should_ be reported; we fall
487 back to SIGIO in that case. --sct */
488 si
.si_signo
= fown
->signum
;
491 /* Make sure we are called with one of the POLL_*
492 reasons, otherwise we could leak kernel stack into
494 BUG_ON((reason
& __SI_MASK
) != __SI_POLL
);
495 if (reason
- POLL_IN
>= NSIGPOLL
)
498 si
.si_band
= band_table
[reason
- POLL_IN
];
500 if (!group_send_sig_info(fown
->signum
, &si
, p
))
502 /* fall-through: fall back on the old plain SIGIO signal */
504 group_send_sig_info(SIGIO
, SEND_SIG_PRIV
, p
);
508 void send_sigio(struct fown_struct
*fown
, int fd
, int band
)
510 struct task_struct
*p
;
514 read_lock(&fown
->lock
);
515 type
= fown
->pid_type
;
518 goto out_unlock_fown
;
520 read_lock(&tasklist_lock
);
521 do_each_pid_task(pid
, type
, p
) {
522 send_sigio_to_task(p
, fown
, fd
, band
);
523 } while_each_pid_task(pid
, type
, p
);
524 read_unlock(&tasklist_lock
);
526 read_unlock(&fown
->lock
);
529 static void send_sigurg_to_task(struct task_struct
*p
,
530 struct fown_struct
*fown
)
532 if (sigio_perm(p
, fown
, SIGURG
))
533 group_send_sig_info(SIGURG
, SEND_SIG_PRIV
, p
);
536 int send_sigurg(struct fown_struct
*fown
)
538 struct task_struct
*p
;
543 read_lock(&fown
->lock
);
544 type
= fown
->pid_type
;
547 goto out_unlock_fown
;
551 read_lock(&tasklist_lock
);
552 do_each_pid_task(pid
, type
, p
) {
553 send_sigurg_to_task(p
, fown
);
554 } while_each_pid_task(pid
, type
, p
);
555 read_unlock(&tasklist_lock
);
557 read_unlock(&fown
->lock
);
561 static DEFINE_RWLOCK(fasync_lock
);
562 static struct kmem_cache
*fasync_cache __read_mostly
;
565 * fasync_helper() is used by some character device drivers (mainly mice)
566 * to set up the fasync queue. It returns negative on error, 0 if it did
567 * no changes and positive if it added/deleted the entry.
569 int fasync_helper(int fd
, struct file
* filp
, int on
, struct fasync_struct
**fapp
)
571 struct fasync_struct
*fa
, **fp
;
572 struct fasync_struct
*new = NULL
;
576 new = kmem_cache_alloc(fasync_cache
, GFP_KERNEL
);
580 write_lock_irq(&fasync_lock
);
581 for (fp
= fapp
; (fa
= *fp
) != NULL
; fp
= &fa
->fa_next
) {
582 if (fa
->fa_file
== filp
) {
585 kmem_cache_free(fasync_cache
, new);
588 kmem_cache_free(fasync_cache
, fa
);
596 new->magic
= FASYNC_MAGIC
;
599 new->fa_next
= *fapp
;
604 write_unlock_irq(&fasync_lock
);
608 EXPORT_SYMBOL(fasync_helper
);
610 void __kill_fasync(struct fasync_struct
*fa
, int sig
, int band
)
613 struct fown_struct
* fown
;
614 if (fa
->magic
!= FASYNC_MAGIC
) {
615 printk(KERN_ERR
"kill_fasync: bad magic number in "
619 fown
= &fa
->fa_file
->f_owner
;
620 /* Don't send SIGURG to processes which have not set a
621 queued signum: SIGURG has its own default signalling
623 if (!(sig
== SIGURG
&& fown
->signum
== 0))
624 send_sigio(fown
, fa
->fa_fd
, band
);
629 EXPORT_SYMBOL(__kill_fasync
);
631 void kill_fasync(struct fasync_struct
**fp
, int sig
, int band
)
633 /* First a quick test without locking: usually
637 read_lock(&fasync_lock
);
638 /* reread *fp after obtaining the lock */
639 __kill_fasync(*fp
, sig
, band
);
640 read_unlock(&fasync_lock
);
643 EXPORT_SYMBOL(kill_fasync
);
645 static int __init
fasync_init(void)
647 fasync_cache
= kmem_cache_create("fasync_cache",
648 sizeof(struct fasync_struct
), 0, SLAB_PANIC
, NULL
);
652 module_init(fasync_init
)