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 if (unlikely(oldfd
== newfd
))
141 spin_lock(&files
->file_lock
);
142 if (!(file
= fcheck(oldfd
)))
144 if (newfd
>= current
->signal
->rlim
[RLIMIT_NOFILE
].rlim_cur
)
146 get_file(file
); /* We are now finished with oldfd */
148 err
= expand_files(files
, newfd
);
152 /* To avoid races with open() and dup(), we will mark the fd as
153 * in-use in the open-file bitmap throughout the entire dup2()
154 * process. This is quite safe: do_close() uses the fd array
155 * entry, not the bitmap, to decide what work needs to be
157 /* Doesn't work. open() might be there first. --AV */
159 /* Yes. It's a race. In user space. Nothing sane to do */
161 fdt
= files_fdtable(files
);
162 tofree
= fdt
->fd
[newfd
];
163 if (!tofree
&& FD_ISSET(newfd
, fdt
->open_fds
))
166 rcu_assign_pointer(fdt
->fd
[newfd
], file
);
167 FD_SET(newfd
, fdt
->open_fds
);
168 if (flags
& O_CLOEXEC
)
169 FD_SET(newfd
, fdt
->close_on_exec
);
171 FD_CLR(newfd
, fdt
->close_on_exec
);
172 spin_unlock(&files
->file_lock
);
175 filp_close(tofree
, files
);
180 spin_unlock(&files
->file_lock
);
184 spin_unlock(&files
->file_lock
);
189 asmlinkage
long sys_dup2(unsigned int oldfd
, unsigned int newfd
)
191 if (unlikely(newfd
== oldfd
)) { /* corner case */
192 struct files_struct
*files
= current
->files
;
194 if (!fcheck_files(files
, oldfd
))
199 return sys_dup3(oldfd
, newfd
, 0);
202 asmlinkage
long sys_dup(unsigned int fildes
)
205 struct file
* file
= fget(fildes
);
208 ret
= dupfd(file
, 0, 0);
212 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
214 static int setfl(int fd
, struct file
* filp
, unsigned long arg
)
216 struct inode
* inode
= filp
->f_path
.dentry
->d_inode
;
220 * O_APPEND cannot be cleared if the file is marked as append-only
221 * and the file is open for write.
223 if (((arg
^ filp
->f_flags
) & O_APPEND
) && IS_APPEND(inode
))
226 /* O_NOATIME can only be set by the owner or superuser */
227 if ((arg
& O_NOATIME
) && !(filp
->f_flags
& O_NOATIME
))
228 if (!is_owner_or_cap(inode
))
231 /* required for strict SunOS emulation */
232 if (O_NONBLOCK
!= O_NDELAY
)
236 if (arg
& O_DIRECT
) {
237 if (!filp
->f_mapping
|| !filp
->f_mapping
->a_ops
||
238 !filp
->f_mapping
->a_ops
->direct_IO
)
242 if (filp
->f_op
&& filp
->f_op
->check_flags
)
243 error
= filp
->f_op
->check_flags(arg
);
247 if ((arg
^ filp
->f_flags
) & FASYNC
) {
248 if (filp
->f_op
&& filp
->f_op
->fasync
) {
249 error
= filp
->f_op
->fasync(fd
, filp
, (arg
& FASYNC
) != 0);
255 filp
->f_flags
= (arg
& SETFL_MASK
) | (filp
->f_flags
& ~SETFL_MASK
);
260 static void f_modown(struct file
*filp
, struct pid
*pid
, enum pid_type type
,
261 uid_t uid
, uid_t euid
, int force
)
263 write_lock_irq(&filp
->f_owner
.lock
);
264 if (force
|| !filp
->f_owner
.pid
) {
265 put_pid(filp
->f_owner
.pid
);
266 filp
->f_owner
.pid
= get_pid(pid
);
267 filp
->f_owner
.pid_type
= type
;
268 filp
->f_owner
.uid
= uid
;
269 filp
->f_owner
.euid
= euid
;
271 write_unlock_irq(&filp
->f_owner
.lock
);
274 int __f_setown(struct file
*filp
, struct pid
*pid
, enum pid_type type
,
279 err
= security_file_set_fowner(filp
);
283 f_modown(filp
, pid
, type
, current
->uid
, current
->euid
, force
);
286 EXPORT_SYMBOL(__f_setown
);
288 int f_setown(struct file
*filp
, unsigned long arg
, int force
)
300 pid
= find_vpid(who
);
301 result
= __f_setown(filp
, pid
, type
, force
);
305 EXPORT_SYMBOL(f_setown
);
307 void f_delown(struct file
*filp
)
309 f_modown(filp
, NULL
, PIDTYPE_PID
, 0, 0, 1);
312 pid_t
f_getown(struct file
*filp
)
315 read_lock(&filp
->f_owner
.lock
);
316 pid
= pid_vnr(filp
->f_owner
.pid
);
317 if (filp
->f_owner
.pid_type
== PIDTYPE_PGID
)
319 read_unlock(&filp
->f_owner
.lock
);
323 static long do_fcntl(int fd
, unsigned int cmd
, unsigned long arg
,
330 case F_DUPFD_CLOEXEC
:
332 err
= dupfd(filp
, arg
, cmd
== F_DUPFD_CLOEXEC
);
335 err
= get_close_on_exec(fd
) ? FD_CLOEXEC
: 0;
339 set_close_on_exec(fd
, arg
& FD_CLOEXEC
);
345 err
= setfl(fd
, filp
, arg
);
348 err
= fcntl_getlk(filp
, (struct flock __user
*) arg
);
352 err
= fcntl_setlk(fd
, filp
, cmd
, (struct flock __user
*) arg
);
356 * XXX If f_owner is a process group, the
357 * negative return value will get converted
358 * into an error. Oops. If we keep the
359 * current syscall conventions, the only way
360 * to fix this will be in libc.
362 err
= f_getown(filp
);
363 force_successful_syscall_return();
366 err
= f_setown(filp
, arg
, 1);
369 err
= filp
->f_owner
.signum
;
372 /* arg == 0 restores default behaviour. */
373 if (!valid_signal(arg
)) {
377 filp
->f_owner
.signum
= arg
;
380 err
= fcntl_getlease(filp
);
383 err
= fcntl_setlease(fd
, filp
, arg
);
386 err
= fcntl_dirnotify(fd
, filp
, arg
);
394 asmlinkage
long sys_fcntl(unsigned int fd
, unsigned int cmd
, unsigned long arg
)
403 err
= security_file_fcntl(filp
, cmd
, arg
);
409 err
= do_fcntl(fd
, cmd
, arg
, filp
);
416 #if BITS_PER_LONG == 32
417 asmlinkage
long sys_fcntl64(unsigned int fd
, unsigned int cmd
, unsigned long arg
)
427 err
= security_file_fcntl(filp
, cmd
, arg
);
436 err
= fcntl_getlk64(filp
, (struct flock64 __user
*) arg
);
440 err
= fcntl_setlk64(fd
, filp
, cmd
,
441 (struct flock64 __user
*) arg
);
444 err
= do_fcntl(fd
, cmd
, arg
, filp
);
453 /* Table to convert sigio signal codes into poll band bitmaps */
455 static const long band_table
[NSIGPOLL
] = {
456 POLLIN
| POLLRDNORM
, /* POLL_IN */
457 POLLOUT
| POLLWRNORM
| POLLWRBAND
, /* POLL_OUT */
458 POLLIN
| POLLRDNORM
| POLLMSG
, /* POLL_MSG */
459 POLLERR
, /* POLL_ERR */
460 POLLPRI
| POLLRDBAND
, /* POLL_PRI */
461 POLLHUP
| POLLERR
/* POLL_HUP */
464 static inline int sigio_perm(struct task_struct
*p
,
465 struct fown_struct
*fown
, int sig
)
467 return (((fown
->euid
== 0) ||
468 (fown
->euid
== p
->suid
) || (fown
->euid
== p
->uid
) ||
469 (fown
->uid
== p
->suid
) || (fown
->uid
== p
->uid
)) &&
470 !security_file_send_sigiotask(p
, fown
, sig
));
473 static void send_sigio_to_task(struct task_struct
*p
,
474 struct fown_struct
*fown
,
478 if (!sigio_perm(p
, fown
, fown
->signum
))
481 switch (fown
->signum
) {
484 /* Queue a rt signal with the appropriate fd as its
485 value. We use SI_SIGIO as the source, not
486 SI_KERNEL, since kernel signals always get
487 delivered even if we can't queue. Failure to
488 queue in this case _should_ be reported; we fall
489 back to SIGIO in that case. --sct */
490 si
.si_signo
= fown
->signum
;
493 /* Make sure we are called with one of the POLL_*
494 reasons, otherwise we could leak kernel stack into
496 BUG_ON((reason
& __SI_MASK
) != __SI_POLL
);
497 if (reason
- POLL_IN
>= NSIGPOLL
)
500 si
.si_band
= band_table
[reason
- POLL_IN
];
502 if (!group_send_sig_info(fown
->signum
, &si
, p
))
504 /* fall-through: fall back on the old plain SIGIO signal */
506 group_send_sig_info(SIGIO
, SEND_SIG_PRIV
, p
);
510 void send_sigio(struct fown_struct
*fown
, int fd
, int band
)
512 struct task_struct
*p
;
516 read_lock(&fown
->lock
);
517 type
= fown
->pid_type
;
520 goto out_unlock_fown
;
522 read_lock(&tasklist_lock
);
523 do_each_pid_task(pid
, type
, p
) {
524 send_sigio_to_task(p
, fown
, fd
, band
);
525 } while_each_pid_task(pid
, type
, p
);
526 read_unlock(&tasklist_lock
);
528 read_unlock(&fown
->lock
);
531 static void send_sigurg_to_task(struct task_struct
*p
,
532 struct fown_struct
*fown
)
534 if (sigio_perm(p
, fown
, SIGURG
))
535 group_send_sig_info(SIGURG
, SEND_SIG_PRIV
, p
);
538 int send_sigurg(struct fown_struct
*fown
)
540 struct task_struct
*p
;
545 read_lock(&fown
->lock
);
546 type
= fown
->pid_type
;
549 goto out_unlock_fown
;
553 read_lock(&tasklist_lock
);
554 do_each_pid_task(pid
, type
, p
) {
555 send_sigurg_to_task(p
, fown
);
556 } while_each_pid_task(pid
, type
, p
);
557 read_unlock(&tasklist_lock
);
559 read_unlock(&fown
->lock
);
563 static DEFINE_RWLOCK(fasync_lock
);
564 static struct kmem_cache
*fasync_cache __read_mostly
;
567 * fasync_helper() is used by some character device drivers (mainly mice)
568 * to set up the fasync queue. It returns negative on error, 0 if it did
569 * no changes and positive if it added/deleted the entry.
571 int fasync_helper(int fd
, struct file
* filp
, int on
, struct fasync_struct
**fapp
)
573 struct fasync_struct
*fa
, **fp
;
574 struct fasync_struct
*new = NULL
;
578 new = kmem_cache_alloc(fasync_cache
, GFP_KERNEL
);
582 write_lock_irq(&fasync_lock
);
583 for (fp
= fapp
; (fa
= *fp
) != NULL
; fp
= &fa
->fa_next
) {
584 if (fa
->fa_file
== filp
) {
587 kmem_cache_free(fasync_cache
, new);
590 kmem_cache_free(fasync_cache
, fa
);
598 new->magic
= FASYNC_MAGIC
;
601 new->fa_next
= *fapp
;
606 write_unlock_irq(&fasync_lock
);
610 EXPORT_SYMBOL(fasync_helper
);
612 void __kill_fasync(struct fasync_struct
*fa
, int sig
, int band
)
615 struct fown_struct
* fown
;
616 if (fa
->magic
!= FASYNC_MAGIC
) {
617 printk(KERN_ERR
"kill_fasync: bad magic number in "
621 fown
= &fa
->fa_file
->f_owner
;
622 /* Don't send SIGURG to processes which have not set a
623 queued signum: SIGURG has its own default signalling
625 if (!(sig
== SIGURG
&& fown
->signum
== 0))
626 send_sigio(fown
, fa
->fa_fd
, band
);
631 EXPORT_SYMBOL(__kill_fasync
);
633 void kill_fasync(struct fasync_struct
**fp
, int sig
, int band
)
635 /* First a quick test without locking: usually
639 read_lock(&fasync_lock
);
640 /* reread *fp after obtaining the lock */
641 __kill_fasync(*fp
, sig
, band
);
642 read_unlock(&fasync_lock
);
645 EXPORT_SYMBOL(kill_fasync
);
647 static int __init
fasync_init(void)
649 fasync_cache
= kmem_cache_create("fasync_cache",
650 sizeof(struct fasync_struct
), 0, SLAB_PANIC
, NULL
);
654 module_init(fasync_init
)