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/capability.h>
13 #include <linux/dnotify.h>
14 #include <linux/smp_lock.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>
23 #include <asm/siginfo.h>
24 #include <asm/uaccess.h>
26 void fastcall
set_close_on_exec(unsigned int fd
, int flag
)
28 struct files_struct
*files
= current
->files
;
30 spin_lock(&files
->file_lock
);
31 fdt
= files_fdtable(files
);
33 FD_SET(fd
, fdt
->close_on_exec
);
35 FD_CLR(fd
, fdt
->close_on_exec
);
36 spin_unlock(&files
->file_lock
);
39 static int get_close_on_exec(unsigned int fd
)
41 struct files_struct
*files
= current
->files
;
45 fdt
= files_fdtable(files
);
46 res
= FD_ISSET(fd
, fdt
->close_on_exec
);
52 * locate_fd finds a free file descriptor in the open_fds fdset,
53 * expanding the fd arrays if necessary. Must be called with the
54 * file_lock held for write.
57 static int locate_fd(struct files_struct
*files
,
58 struct file
*file
, unsigned int orig_start
)
66 if (orig_start
>= current
->signal
->rlim
[RLIMIT_NOFILE
].rlim_cur
)
70 fdt
= files_fdtable(files
);
72 * Someone might have closed fd's in the range
73 * orig_start..fdt->next_fd
76 if (start
< files
->next_fd
)
77 start
= files
->next_fd
;
80 if (start
< fdt
->max_fdset
) {
81 newfd
= find_next_zero_bit(fdt
->open_fds
->fds_bits
,
82 fdt
->max_fdset
, start
);
86 if (newfd
>= current
->signal
->rlim
[RLIMIT_NOFILE
].rlim_cur
)
89 error
= expand_files(files
, newfd
);
94 * If we needed to expand the fs array we
95 * might have blocked - try again.
101 * We reacquired files_lock, so we are safe as long as
102 * we reacquire the fdtable pointer and use it while holding
103 * the lock, no one can free it during that time.
105 if (start
<= files
->next_fd
)
106 files
->next_fd
= newfd
+ 1;
114 static int dupfd(struct file
*file
, unsigned int start
)
116 struct files_struct
* files
= current
->files
;
120 spin_lock(&files
->file_lock
);
121 fd
= locate_fd(files
, file
, start
);
123 /* locate_fd() may have expanded fdtable, load the ptr */
124 fdt
= files_fdtable(files
);
125 FD_SET(fd
, fdt
->open_fds
);
126 FD_CLR(fd
, fdt
->close_on_exec
);
127 spin_unlock(&files
->file_lock
);
128 fd_install(fd
, file
);
130 spin_unlock(&files
->file_lock
);
137 asmlinkage
long sys_dup2(unsigned int oldfd
, unsigned int newfd
)
140 struct file
* file
, *tofree
;
141 struct files_struct
* files
= current
->files
;
144 spin_lock(&files
->file_lock
);
145 if (!(file
= fcheck(oldfd
)))
151 if (newfd
>= current
->signal
->rlim
[RLIMIT_NOFILE
].rlim_cur
)
153 get_file(file
); /* We are now finished with oldfd */
155 err
= expand_files(files
, newfd
);
159 /* To avoid races with open() and dup(), we will mark the fd as
160 * in-use in the open-file bitmap throughout the entire dup2()
161 * process. This is quite safe: do_close() uses the fd array
162 * entry, not the bitmap, to decide what work needs to be
164 /* Doesn't work. open() might be there first. --AV */
166 /* Yes. It's a race. In user space. Nothing sane to do */
168 fdt
= files_fdtable(files
);
169 tofree
= fdt
->fd
[newfd
];
170 if (!tofree
&& FD_ISSET(newfd
, fdt
->open_fds
))
173 rcu_assign_pointer(fdt
->fd
[newfd
], file
);
174 FD_SET(newfd
, fdt
->open_fds
);
175 FD_CLR(newfd
, fdt
->close_on_exec
);
176 spin_unlock(&files
->file_lock
);
179 filp_close(tofree
, files
);
184 spin_unlock(&files
->file_lock
);
188 spin_unlock(&files
->file_lock
);
193 asmlinkage
long sys_dup(unsigned int fildes
)
196 struct file
* file
= fget(fildes
);
199 ret
= dupfd(file
, 0);
203 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
205 static int setfl(int fd
, struct file
* filp
, unsigned long arg
)
207 struct inode
* inode
= filp
->f_dentry
->d_inode
;
211 * O_APPEND cannot be cleared if the file is marked as append-only
212 * and the file is open for write.
214 if (((arg
^ filp
->f_flags
) & O_APPEND
) && IS_APPEND(inode
))
217 /* O_NOATIME can only be set by the owner or superuser */
218 if ((arg
& O_NOATIME
) && !(filp
->f_flags
& O_NOATIME
))
219 if (current
->fsuid
!= inode
->i_uid
&& !capable(CAP_FOWNER
))
222 /* required for strict SunOS emulation */
223 if (O_NONBLOCK
!= O_NDELAY
)
227 if (arg
& O_DIRECT
) {
228 if (!filp
->f_mapping
|| !filp
->f_mapping
->a_ops
||
229 !filp
->f_mapping
->a_ops
->direct_IO
)
233 if (filp
->f_op
&& filp
->f_op
->check_flags
)
234 error
= filp
->f_op
->check_flags(arg
);
239 if ((arg
^ filp
->f_flags
) & FASYNC
) {
240 if (filp
->f_op
&& filp
->f_op
->fasync
) {
241 error
= filp
->f_op
->fasync(fd
, filp
, (arg
& FASYNC
) != 0);
247 filp
->f_flags
= (arg
& SETFL_MASK
) | (filp
->f_flags
& ~SETFL_MASK
);
253 static void f_modown(struct file
*filp
, unsigned long pid
,
254 uid_t uid
, uid_t euid
, int force
)
256 write_lock_irq(&filp
->f_owner
.lock
);
257 if (force
|| !filp
->f_owner
.pid
) {
258 filp
->f_owner
.pid
= pid
;
259 filp
->f_owner
.uid
= uid
;
260 filp
->f_owner
.euid
= euid
;
262 write_unlock_irq(&filp
->f_owner
.lock
);
265 int f_setown(struct file
*filp
, unsigned long arg
, int force
)
269 err
= security_file_set_fowner(filp
);
273 f_modown(filp
, arg
, current
->uid
, current
->euid
, force
);
277 EXPORT_SYMBOL(f_setown
);
279 void f_delown(struct file
*filp
)
281 f_modown(filp
, 0, 0, 0, 1);
284 static long do_fcntl(int fd
, unsigned int cmd
, unsigned long arg
,
292 err
= dupfd(filp
, arg
);
295 err
= get_close_on_exec(fd
) ? FD_CLOEXEC
: 0;
299 set_close_on_exec(fd
, arg
& FD_CLOEXEC
);
305 err
= setfl(fd
, filp
, arg
);
308 err
= fcntl_getlk(filp
, (struct flock __user
*) arg
);
312 err
= fcntl_setlk(fd
, filp
, cmd
, (struct flock __user
*) arg
);
316 * XXX If f_owner is a process group, the
317 * negative return value will get converted
318 * into an error. Oops. If we keep the
319 * current syscall conventions, the only way
320 * to fix this will be in libc.
322 err
= filp
->f_owner
.pid
;
323 force_successful_syscall_return();
326 err
= f_setown(filp
, arg
, 1);
329 err
= filp
->f_owner
.signum
;
332 /* arg == 0 restores default behaviour. */
333 if (!valid_signal(arg
)) {
337 filp
->f_owner
.signum
= arg
;
340 err
= fcntl_getlease(filp
);
343 err
= fcntl_setlease(fd
, filp
, arg
);
346 err
= fcntl_dirnotify(fd
, filp
, arg
);
354 asmlinkage
long sys_fcntl(unsigned int fd
, unsigned int cmd
, unsigned long arg
)
363 err
= security_file_fcntl(filp
, cmd
, arg
);
369 err
= do_fcntl(fd
, cmd
, arg
, filp
);
376 #if BITS_PER_LONG == 32
377 asmlinkage
long sys_fcntl64(unsigned int fd
, unsigned int cmd
, unsigned long arg
)
387 err
= security_file_fcntl(filp
, cmd
, arg
);
396 err
= fcntl_getlk64(filp
, (struct flock64 __user
*) arg
);
400 err
= fcntl_setlk64(fd
, filp
, cmd
,
401 (struct flock64 __user
*) arg
);
404 err
= do_fcntl(fd
, cmd
, arg
, filp
);
413 /* Table to convert sigio signal codes into poll band bitmaps */
415 static const long band_table
[NSIGPOLL
] = {
416 POLLIN
| POLLRDNORM
, /* POLL_IN */
417 POLLOUT
| POLLWRNORM
| POLLWRBAND
, /* POLL_OUT */
418 POLLIN
| POLLRDNORM
| POLLMSG
, /* POLL_MSG */
419 POLLERR
, /* POLL_ERR */
420 POLLPRI
| POLLRDBAND
, /* POLL_PRI */
421 POLLHUP
| POLLERR
/* POLL_HUP */
424 static inline int sigio_perm(struct task_struct
*p
,
425 struct fown_struct
*fown
, int sig
)
427 return (((fown
->euid
== 0) ||
428 (fown
->euid
== p
->suid
) || (fown
->euid
== p
->uid
) ||
429 (fown
->uid
== p
->suid
) || (fown
->uid
== p
->uid
)) &&
430 !security_file_send_sigiotask(p
, fown
, sig
));
433 static void send_sigio_to_task(struct task_struct
*p
,
434 struct fown_struct
*fown
,
438 if (!sigio_perm(p
, fown
, fown
->signum
))
441 switch (fown
->signum
) {
444 /* Queue a rt signal with the appropriate fd as its
445 value. We use SI_SIGIO as the source, not
446 SI_KERNEL, since kernel signals always get
447 delivered even if we can't queue. Failure to
448 queue in this case _should_ be reported; we fall
449 back to SIGIO in that case. --sct */
450 si
.si_signo
= fown
->signum
;
453 /* Make sure we are called with one of the POLL_*
454 reasons, otherwise we could leak kernel stack into
456 if ((reason
& __SI_MASK
) != __SI_POLL
)
458 if (reason
- POLL_IN
>= NSIGPOLL
)
461 si
.si_band
= band_table
[reason
- POLL_IN
];
463 if (!group_send_sig_info(fown
->signum
, &si
, p
))
465 /* fall-through: fall back on the old plain SIGIO signal */
467 group_send_sig_info(SIGIO
, SEND_SIG_PRIV
, p
);
471 void send_sigio(struct fown_struct
*fown
, int fd
, int band
)
473 struct task_struct
*p
;
476 read_lock(&fown
->lock
);
479 goto out_unlock_fown
;
481 read_lock(&tasklist_lock
);
483 p
= find_task_by_pid(pid
);
485 send_sigio_to_task(p
, fown
, fd
, band
);
488 do_each_task_pid(-pid
, PIDTYPE_PGID
, p
) {
489 send_sigio_to_task(p
, fown
, fd
, band
);
490 } while_each_task_pid(-pid
, PIDTYPE_PGID
, p
);
492 read_unlock(&tasklist_lock
);
494 read_unlock(&fown
->lock
);
497 static void send_sigurg_to_task(struct task_struct
*p
,
498 struct fown_struct
*fown
)
500 if (sigio_perm(p
, fown
, SIGURG
))
501 group_send_sig_info(SIGURG
, SEND_SIG_PRIV
, p
);
504 int send_sigurg(struct fown_struct
*fown
)
506 struct task_struct
*p
;
509 read_lock(&fown
->lock
);
512 goto out_unlock_fown
;
516 read_lock(&tasklist_lock
);
518 p
= find_task_by_pid(pid
);
520 send_sigurg_to_task(p
, fown
);
523 do_each_task_pid(-pid
, PIDTYPE_PGID
, p
) {
524 send_sigurg_to_task(p
, fown
);
525 } while_each_task_pid(-pid
, PIDTYPE_PGID
, p
);
527 read_unlock(&tasklist_lock
);
529 read_unlock(&fown
->lock
);
533 static DEFINE_RWLOCK(fasync_lock
);
534 static kmem_cache_t
*fasync_cache __read_mostly
;
537 * fasync_helper() is used by some character device drivers (mainly mice)
538 * to set up the fasync queue. It returns negative on error, 0 if it did
539 * no changes and positive if it added/deleted the entry.
541 int fasync_helper(int fd
, struct file
* filp
, int on
, struct fasync_struct
**fapp
)
543 struct fasync_struct
*fa
, **fp
;
544 struct fasync_struct
*new = NULL
;
548 new = kmem_cache_alloc(fasync_cache
, SLAB_KERNEL
);
552 write_lock_irq(&fasync_lock
);
553 for (fp
= fapp
; (fa
= *fp
) != NULL
; fp
= &fa
->fa_next
) {
554 if (fa
->fa_file
== filp
) {
557 kmem_cache_free(fasync_cache
, new);
560 kmem_cache_free(fasync_cache
, fa
);
568 new->magic
= FASYNC_MAGIC
;
571 new->fa_next
= *fapp
;
576 write_unlock_irq(&fasync_lock
);
580 EXPORT_SYMBOL(fasync_helper
);
582 void __kill_fasync(struct fasync_struct
*fa
, int sig
, int band
)
585 struct fown_struct
* fown
;
586 if (fa
->magic
!= FASYNC_MAGIC
) {
587 printk(KERN_ERR
"kill_fasync: bad magic number in "
591 fown
= &fa
->fa_file
->f_owner
;
592 /* Don't send SIGURG to processes which have not set a
593 queued signum: SIGURG has its own default signalling
595 if (!(sig
== SIGURG
&& fown
->signum
== 0))
596 send_sigio(fown
, fa
->fa_fd
, band
);
601 EXPORT_SYMBOL(__kill_fasync
);
603 void kill_fasync(struct fasync_struct
**fp
, int sig
, int band
)
605 /* First a quick test without locking: usually
609 read_lock(&fasync_lock
);
610 /* reread *fp after obtaining the lock */
611 __kill_fasync(*fp
, sig
, band
);
612 read_unlock(&fasync_lock
);
615 EXPORT_SYMBOL(kill_fasync
);
617 static int __init
fasync_init(void)
619 fasync_cache
= kmem_cache_create("fasync_cache",
620 sizeof(struct fasync_struct
), 0, SLAB_PANIC
, NULL
, NULL
);
624 module_init(fasync_init
)