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
, struct pid
*pid
, enum pid_type type
,
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 put_pid(filp
->f_owner
.pid
);
259 filp
->f_owner
.pid
= get_pid(pid
);
260 filp
->f_owner
.pid_type
= type
;
261 filp
->f_owner
.uid
= uid
;
262 filp
->f_owner
.euid
= euid
;
264 write_unlock_irq(&filp
->f_owner
.lock
);
267 int __f_setown(struct file
*filp
, struct pid
*pid
, enum pid_type type
,
272 err
= security_file_set_fowner(filp
);
276 f_modown(filp
, pid
, type
, current
->uid
, current
->euid
, force
);
279 EXPORT_SYMBOL(__f_setown
);
281 int f_setown(struct file
*filp
, unsigned long arg
, int force
)
294 result
= __f_setown(filp
, pid
, type
, force
);
298 EXPORT_SYMBOL(f_setown
);
300 void f_delown(struct file
*filp
)
302 f_modown(filp
, NULL
, PIDTYPE_PID
, 0, 0, 1);
305 pid_t
f_getown(struct file
*filp
)
308 pid
= pid_nr(filp
->f_owner
.pid
);
309 if (filp
->f_owner
.pid_type
== PIDTYPE_PGID
)
314 static long do_fcntl(int fd
, unsigned int cmd
, unsigned long arg
,
322 err
= dupfd(filp
, arg
);
325 err
= get_close_on_exec(fd
) ? FD_CLOEXEC
: 0;
329 set_close_on_exec(fd
, arg
& FD_CLOEXEC
);
335 err
= setfl(fd
, filp
, arg
);
338 err
= fcntl_getlk(filp
, (struct flock __user
*) arg
);
342 err
= fcntl_setlk(fd
, filp
, cmd
, (struct flock __user
*) arg
);
346 * XXX If f_owner is a process group, the
347 * negative return value will get converted
348 * into an error. Oops. If we keep the
349 * current syscall conventions, the only way
350 * to fix this will be in libc.
352 err
= f_getown(filp
);
353 force_successful_syscall_return();
356 err
= f_setown(filp
, arg
, 1);
359 err
= filp
->f_owner
.signum
;
362 /* arg == 0 restores default behaviour. */
363 if (!valid_signal(arg
)) {
367 filp
->f_owner
.signum
= arg
;
370 err
= fcntl_getlease(filp
);
373 err
= fcntl_setlease(fd
, filp
, arg
);
376 err
= fcntl_dirnotify(fd
, filp
, arg
);
384 asmlinkage
long sys_fcntl(unsigned int fd
, unsigned int cmd
, unsigned long arg
)
393 err
= security_file_fcntl(filp
, cmd
, arg
);
399 err
= do_fcntl(fd
, cmd
, arg
, filp
);
406 #if BITS_PER_LONG == 32
407 asmlinkage
long sys_fcntl64(unsigned int fd
, unsigned int cmd
, unsigned long arg
)
417 err
= security_file_fcntl(filp
, cmd
, arg
);
426 err
= fcntl_getlk64(filp
, (struct flock64 __user
*) arg
);
430 err
= fcntl_setlk64(fd
, filp
, cmd
,
431 (struct flock64 __user
*) arg
);
434 err
= do_fcntl(fd
, cmd
, arg
, filp
);
443 /* Table to convert sigio signal codes into poll band bitmaps */
445 static const long band_table
[NSIGPOLL
] = {
446 POLLIN
| POLLRDNORM
, /* POLL_IN */
447 POLLOUT
| POLLWRNORM
| POLLWRBAND
, /* POLL_OUT */
448 POLLIN
| POLLRDNORM
| POLLMSG
, /* POLL_MSG */
449 POLLERR
, /* POLL_ERR */
450 POLLPRI
| POLLRDBAND
, /* POLL_PRI */
451 POLLHUP
| POLLERR
/* POLL_HUP */
454 static inline int sigio_perm(struct task_struct
*p
,
455 struct fown_struct
*fown
, int sig
)
457 return (((fown
->euid
== 0) ||
458 (fown
->euid
== p
->suid
) || (fown
->euid
== p
->uid
) ||
459 (fown
->uid
== p
->suid
) || (fown
->uid
== p
->uid
)) &&
460 !security_file_send_sigiotask(p
, fown
, sig
));
463 static void send_sigio_to_task(struct task_struct
*p
,
464 struct fown_struct
*fown
,
468 if (!sigio_perm(p
, fown
, fown
->signum
))
471 switch (fown
->signum
) {
474 /* Queue a rt signal with the appropriate fd as its
475 value. We use SI_SIGIO as the source, not
476 SI_KERNEL, since kernel signals always get
477 delivered even if we can't queue. Failure to
478 queue in this case _should_ be reported; we fall
479 back to SIGIO in that case. --sct */
480 si
.si_signo
= fown
->signum
;
483 /* Make sure we are called with one of the POLL_*
484 reasons, otherwise we could leak kernel stack into
486 BUG_ON((reason
& __SI_MASK
) != __SI_POLL
);
487 if (reason
- POLL_IN
>= NSIGPOLL
)
490 si
.si_band
= band_table
[reason
- POLL_IN
];
492 if (!group_send_sig_info(fown
->signum
, &si
, p
))
494 /* fall-through: fall back on the old plain SIGIO signal */
496 group_send_sig_info(SIGIO
, SEND_SIG_PRIV
, p
);
500 void send_sigio(struct fown_struct
*fown
, int fd
, int band
)
502 struct task_struct
*p
;
506 read_lock(&fown
->lock
);
507 type
= fown
->pid_type
;
510 goto out_unlock_fown
;
512 read_lock(&tasklist_lock
);
513 do_each_pid_task(pid
, type
, p
) {
514 send_sigio_to_task(p
, fown
, fd
, band
);
515 } while_each_pid_task(pid
, type
, p
);
516 read_unlock(&tasklist_lock
);
518 read_unlock(&fown
->lock
);
521 static void send_sigurg_to_task(struct task_struct
*p
,
522 struct fown_struct
*fown
)
524 if (sigio_perm(p
, fown
, SIGURG
))
525 group_send_sig_info(SIGURG
, SEND_SIG_PRIV
, p
);
528 int send_sigurg(struct fown_struct
*fown
)
530 struct task_struct
*p
;
535 read_lock(&fown
->lock
);
536 type
= fown
->pid_type
;
539 goto out_unlock_fown
;
543 read_lock(&tasklist_lock
);
544 do_each_pid_task(pid
, type
, p
) {
545 send_sigurg_to_task(p
, fown
);
546 } while_each_pid_task(pid
, type
, p
);
547 read_unlock(&tasklist_lock
);
549 read_unlock(&fown
->lock
);
553 static DEFINE_RWLOCK(fasync_lock
);
554 static kmem_cache_t
*fasync_cache __read_mostly
;
557 * fasync_helper() is used by some character device drivers (mainly mice)
558 * to set up the fasync queue. It returns negative on error, 0 if it did
559 * no changes and positive if it added/deleted the entry.
561 int fasync_helper(int fd
, struct file
* filp
, int on
, struct fasync_struct
**fapp
)
563 struct fasync_struct
*fa
, **fp
;
564 struct fasync_struct
*new = NULL
;
568 new = kmem_cache_alloc(fasync_cache
, SLAB_KERNEL
);
572 write_lock_irq(&fasync_lock
);
573 for (fp
= fapp
; (fa
= *fp
) != NULL
; fp
= &fa
->fa_next
) {
574 if (fa
->fa_file
== filp
) {
577 kmem_cache_free(fasync_cache
, new);
580 kmem_cache_free(fasync_cache
, fa
);
588 new->magic
= FASYNC_MAGIC
;
591 new->fa_next
= *fapp
;
596 write_unlock_irq(&fasync_lock
);
600 EXPORT_SYMBOL(fasync_helper
);
602 void __kill_fasync(struct fasync_struct
*fa
, int sig
, int band
)
605 struct fown_struct
* fown
;
606 if (fa
->magic
!= FASYNC_MAGIC
) {
607 printk(KERN_ERR
"kill_fasync: bad magic number in "
611 fown
= &fa
->fa_file
->f_owner
;
612 /* Don't send SIGURG to processes which have not set a
613 queued signum: SIGURG has its own default signalling
615 if (!(sig
== SIGURG
&& fown
->signum
== 0))
616 send_sigio(fown
, fa
->fa_fd
, band
);
621 EXPORT_SYMBOL(__kill_fasync
);
623 void kill_fasync(struct fasync_struct
**fp
, int sig
, int band
)
625 /* First a quick test without locking: usually
629 read_lock(&fasync_lock
);
630 /* reread *fp after obtaining the lock */
631 __kill_fasync(*fp
, sig
, band
);
632 read_unlock(&fasync_lock
);
635 EXPORT_SYMBOL(kill_fasync
);
637 static int __init
fasync_init(void)
639 fasync_cache
= kmem_cache_create("fasync_cache",
640 sizeof(struct fasync_struct
), 0, SLAB_PANIC
, NULL
, NULL
);
644 module_init(fasync_init
)