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>
22 #include <linux/smp_lock.h>
25 #include <asm/siginfo.h>
26 #include <asm/uaccess.h>
28 void set_close_on_exec(unsigned int fd
, int flag
)
30 struct files_struct
*files
= current
->files
;
32 spin_lock(&files
->file_lock
);
33 fdt
= files_fdtable(files
);
35 FD_SET(fd
, fdt
->close_on_exec
);
37 FD_CLR(fd
, fdt
->close_on_exec
);
38 spin_unlock(&files
->file_lock
);
41 static int get_close_on_exec(unsigned int fd
)
43 struct files_struct
*files
= current
->files
;
47 fdt
= files_fdtable(files
);
48 res
= FD_ISSET(fd
, fdt
->close_on_exec
);
53 SYSCALL_DEFINE3(dup3
, unsigned int, oldfd
, unsigned int, newfd
, int, flags
)
56 struct file
* file
, *tofree
;
57 struct files_struct
* files
= current
->files
;
60 if ((flags
& ~O_CLOEXEC
) != 0)
63 if (unlikely(oldfd
== newfd
))
66 spin_lock(&files
->file_lock
);
67 err
= expand_files(files
, newfd
);
71 if (unlikely(err
< 0)) {
77 * We need to detect attempts to do dup2() over allocated but still
78 * not finished descriptor. NB: OpenBSD avoids that at the price of
79 * extra work in their equivalent of fget() - they insert struct
80 * file immediately after grabbing descriptor, mark it larval if
81 * more work (e.g. actual opening) is needed and make sure that
82 * fget() treats larval files as absent. Potentially interesting,
83 * but while extra work in fget() is trivial, locking implications
84 * and amount of surgery on open()-related paths in VFS are not.
85 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
86 * deadlocks in rather amusing ways, AFAICS. All of that is out of
87 * scope of POSIX or SUS, since neither considers shared descriptor
88 * tables and this condition does not arise without those.
91 fdt
= files_fdtable(files
);
92 tofree
= fdt
->fd
[newfd
];
93 if (!tofree
&& FD_ISSET(newfd
, fdt
->open_fds
))
96 rcu_assign_pointer(fdt
->fd
[newfd
], file
);
97 FD_SET(newfd
, fdt
->open_fds
);
98 if (flags
& O_CLOEXEC
)
99 FD_SET(newfd
, fdt
->close_on_exec
);
101 FD_CLR(newfd
, fdt
->close_on_exec
);
102 spin_unlock(&files
->file_lock
);
105 filp_close(tofree
, files
);
112 spin_unlock(&files
->file_lock
);
116 SYSCALL_DEFINE2(dup2
, unsigned int, oldfd
, unsigned int, newfd
)
118 if (unlikely(newfd
== oldfd
)) { /* corner case */
119 struct files_struct
*files
= current
->files
;
121 if (!fcheck_files(files
, oldfd
))
126 return sys_dup3(oldfd
, newfd
, 0);
129 SYSCALL_DEFINE1(dup
, unsigned int, fildes
)
132 struct file
*file
= fget(fildes
);
135 ret
= get_unused_fd();
137 fd_install(ret
, file
);
144 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
146 static int setfl(int fd
, struct file
* filp
, unsigned long arg
)
148 struct inode
* inode
= filp
->f_path
.dentry
->d_inode
;
152 * O_APPEND cannot be cleared if the file is marked as append-only
153 * and the file is open for write.
155 if (((arg
^ filp
->f_flags
) & O_APPEND
) && IS_APPEND(inode
))
158 /* O_NOATIME can only be set by the owner or superuser */
159 if ((arg
& O_NOATIME
) && !(filp
->f_flags
& O_NOATIME
))
160 if (!is_owner_or_cap(inode
))
163 /* required for strict SunOS emulation */
164 if (O_NONBLOCK
!= O_NDELAY
)
168 if (arg
& O_DIRECT
) {
169 if (!filp
->f_mapping
|| !filp
->f_mapping
->a_ops
||
170 !filp
->f_mapping
->a_ops
->direct_IO
)
174 if (filp
->f_op
&& filp
->f_op
->check_flags
)
175 error
= filp
->f_op
->check_flags(arg
);
180 * We still need a lock here for now to keep multiple FASYNC calls
181 * from racing with each other.
184 if ((arg
^ filp
->f_flags
) & FASYNC
) {
185 if (filp
->f_op
&& filp
->f_op
->fasync
) {
186 error
= filp
->f_op
->fasync(fd
, filp
, (arg
& FASYNC
) != 0);
192 filp
->f_flags
= (arg
& SETFL_MASK
) | (filp
->f_flags
& ~SETFL_MASK
);
198 static void f_modown(struct file
*filp
, struct pid
*pid
, enum pid_type type
,
199 uid_t uid
, uid_t euid
, int force
)
201 write_lock_irq(&filp
->f_owner
.lock
);
202 if (force
|| !filp
->f_owner
.pid
) {
203 put_pid(filp
->f_owner
.pid
);
204 filp
->f_owner
.pid
= get_pid(pid
);
205 filp
->f_owner
.pid_type
= type
;
206 filp
->f_owner
.uid
= uid
;
207 filp
->f_owner
.euid
= euid
;
209 write_unlock_irq(&filp
->f_owner
.lock
);
212 int __f_setown(struct file
*filp
, struct pid
*pid
, enum pid_type type
,
215 const struct cred
*cred
= current_cred();
218 err
= security_file_set_fowner(filp
);
222 f_modown(filp
, pid
, type
, cred
->uid
, cred
->euid
, force
);
225 EXPORT_SYMBOL(__f_setown
);
227 int f_setown(struct file
*filp
, unsigned long arg
, int force
)
239 pid
= find_vpid(who
);
240 result
= __f_setown(filp
, pid
, type
, force
);
244 EXPORT_SYMBOL(f_setown
);
246 void f_delown(struct file
*filp
)
248 f_modown(filp
, NULL
, PIDTYPE_PID
, 0, 0, 1);
251 pid_t
f_getown(struct file
*filp
)
254 read_lock(&filp
->f_owner
.lock
);
255 pid
= pid_vnr(filp
->f_owner
.pid
);
256 if (filp
->f_owner
.pid_type
== PIDTYPE_PGID
)
258 read_unlock(&filp
->f_owner
.lock
);
262 static long do_fcntl(int fd
, unsigned int cmd
, unsigned long arg
,
269 case F_DUPFD_CLOEXEC
:
270 if (arg
>= current
->signal
->rlim
[RLIMIT_NOFILE
].rlim_cur
)
272 err
= alloc_fd(arg
, cmd
== F_DUPFD_CLOEXEC
? O_CLOEXEC
: 0);
275 fd_install(err
, filp
);
279 err
= get_close_on_exec(fd
) ? FD_CLOEXEC
: 0;
283 set_close_on_exec(fd
, arg
& FD_CLOEXEC
);
289 err
= setfl(fd
, filp
, arg
);
292 err
= fcntl_getlk(filp
, (struct flock __user
*) arg
);
296 err
= fcntl_setlk(fd
, filp
, cmd
, (struct flock __user
*) arg
);
300 * XXX If f_owner is a process group, the
301 * negative return value will get converted
302 * into an error. Oops. If we keep the
303 * current syscall conventions, the only way
304 * to fix this will be in libc.
306 err
= f_getown(filp
);
307 force_successful_syscall_return();
310 err
= f_setown(filp
, arg
, 1);
313 err
= filp
->f_owner
.signum
;
316 /* arg == 0 restores default behaviour. */
317 if (!valid_signal(arg
)) {
321 filp
->f_owner
.signum
= arg
;
324 err
= fcntl_getlease(filp
);
327 err
= fcntl_setlease(fd
, filp
, arg
);
330 err
= fcntl_dirnotify(fd
, filp
, arg
);
338 SYSCALL_DEFINE3(fcntl
, unsigned int, fd
, unsigned int, cmd
, unsigned long, arg
)
347 err
= security_file_fcntl(filp
, cmd
, arg
);
353 err
= do_fcntl(fd
, cmd
, arg
, filp
);
360 #if BITS_PER_LONG == 32
361 SYSCALL_DEFINE3(fcntl64
, unsigned int, fd
, unsigned int, cmd
,
372 err
= security_file_fcntl(filp
, cmd
, arg
);
381 err
= fcntl_getlk64(filp
, (struct flock64 __user
*) arg
);
385 err
= fcntl_setlk64(fd
, filp
, cmd
,
386 (struct flock64 __user
*) arg
);
389 err
= do_fcntl(fd
, cmd
, arg
, filp
);
398 /* Table to convert sigio signal codes into poll band bitmaps */
400 static const long band_table
[NSIGPOLL
] = {
401 POLLIN
| POLLRDNORM
, /* POLL_IN */
402 POLLOUT
| POLLWRNORM
| POLLWRBAND
, /* POLL_OUT */
403 POLLIN
| POLLRDNORM
| POLLMSG
, /* POLL_MSG */
404 POLLERR
, /* POLL_ERR */
405 POLLPRI
| POLLRDBAND
, /* POLL_PRI */
406 POLLHUP
| POLLERR
/* POLL_HUP */
409 static inline int sigio_perm(struct task_struct
*p
,
410 struct fown_struct
*fown
, int sig
)
412 const struct cred
*cred
;
416 cred
= __task_cred(p
);
417 ret
= ((fown
->euid
== 0 ||
418 fown
->euid
== cred
->suid
|| fown
->euid
== cred
->uid
||
419 fown
->uid
== cred
->suid
|| fown
->uid
== cred
->uid
) &&
420 !security_file_send_sigiotask(p
, fown
, sig
));
425 static void send_sigio_to_task(struct task_struct
*p
,
426 struct fown_struct
*fown
,
430 if (!sigio_perm(p
, fown
, fown
->signum
))
433 switch (fown
->signum
) {
436 /* Queue a rt signal with the appropriate fd as its
437 value. We use SI_SIGIO as the source, not
438 SI_KERNEL, since kernel signals always get
439 delivered even if we can't queue. Failure to
440 queue in this case _should_ be reported; we fall
441 back to SIGIO in that case. --sct */
442 si
.si_signo
= fown
->signum
;
445 /* Make sure we are called with one of the POLL_*
446 reasons, otherwise we could leak kernel stack into
448 BUG_ON((reason
& __SI_MASK
) != __SI_POLL
);
449 if (reason
- POLL_IN
>= NSIGPOLL
)
452 si
.si_band
= band_table
[reason
- POLL_IN
];
454 if (!group_send_sig_info(fown
->signum
, &si
, p
))
456 /* fall-through: fall back on the old plain SIGIO signal */
458 group_send_sig_info(SIGIO
, SEND_SIG_PRIV
, p
);
462 void send_sigio(struct fown_struct
*fown
, int fd
, int band
)
464 struct task_struct
*p
;
468 read_lock(&fown
->lock
);
469 type
= fown
->pid_type
;
472 goto out_unlock_fown
;
474 read_lock(&tasklist_lock
);
475 do_each_pid_task(pid
, type
, p
) {
476 send_sigio_to_task(p
, fown
, fd
, band
);
477 } while_each_pid_task(pid
, type
, p
);
478 read_unlock(&tasklist_lock
);
480 read_unlock(&fown
->lock
);
483 static void send_sigurg_to_task(struct task_struct
*p
,
484 struct fown_struct
*fown
)
486 if (sigio_perm(p
, fown
, SIGURG
))
487 group_send_sig_info(SIGURG
, SEND_SIG_PRIV
, p
);
490 int send_sigurg(struct fown_struct
*fown
)
492 struct task_struct
*p
;
497 read_lock(&fown
->lock
);
498 type
= fown
->pid_type
;
501 goto out_unlock_fown
;
505 read_lock(&tasklist_lock
);
506 do_each_pid_task(pid
, type
, p
) {
507 send_sigurg_to_task(p
, fown
);
508 } while_each_pid_task(pid
, type
, p
);
509 read_unlock(&tasklist_lock
);
511 read_unlock(&fown
->lock
);
515 static DEFINE_RWLOCK(fasync_lock
);
516 static struct kmem_cache
*fasync_cache __read_mostly
;
519 * fasync_helper() is used by some character device drivers (mainly mice)
520 * to set up the fasync queue. It returns negative on error, 0 if it did
521 * no changes and positive if it added/deleted the entry.
523 int fasync_helper(int fd
, struct file
* filp
, int on
, struct fasync_struct
**fapp
)
525 struct fasync_struct
*fa
, **fp
;
526 struct fasync_struct
*new = NULL
;
530 new = kmem_cache_alloc(fasync_cache
, GFP_KERNEL
);
534 write_lock_irq(&fasync_lock
);
535 for (fp
= fapp
; (fa
= *fp
) != NULL
; fp
= &fa
->fa_next
) {
536 if (fa
->fa_file
== filp
) {
539 kmem_cache_free(fasync_cache
, new);
542 kmem_cache_free(fasync_cache
, fa
);
550 new->magic
= FASYNC_MAGIC
;
553 new->fa_next
= *fapp
;
558 write_unlock_irq(&fasync_lock
);
562 EXPORT_SYMBOL(fasync_helper
);
564 void __kill_fasync(struct fasync_struct
*fa
, int sig
, int band
)
567 struct fown_struct
* fown
;
568 if (fa
->magic
!= FASYNC_MAGIC
) {
569 printk(KERN_ERR
"kill_fasync: bad magic number in "
573 fown
= &fa
->fa_file
->f_owner
;
574 /* Don't send SIGURG to processes which have not set a
575 queued signum: SIGURG has its own default signalling
577 if (!(sig
== SIGURG
&& fown
->signum
== 0))
578 send_sigio(fown
, fa
->fa_fd
, band
);
583 EXPORT_SYMBOL(__kill_fasync
);
585 void kill_fasync(struct fasync_struct
**fp
, int sig
, int band
)
587 /* First a quick test without locking: usually
591 read_lock(&fasync_lock
);
592 /* reread *fp after obtaining the lock */
593 __kill_fasync(*fp
, sig
, band
);
594 read_unlock(&fasync_lock
);
597 EXPORT_SYMBOL(kill_fasync
);
599 static int __init
fasync_init(void)
601 fasync_cache
= kmem_cache_create("fasync_cache",
602 sizeof(struct fasync_struct
), 0, SLAB_PANIC
, NULL
);
606 module_init(fasync_init
)