4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/init.h>
10 #include <linux/file.h>
11 #include <linux/dnotify.h>
12 #include <linux/smp_lock.h>
13 #include <linux/slab.h>
14 #include <linux/module.h>
15 #include <linux/security.h>
16 #include <linux/ptrace.h>
19 #include <asm/siginfo.h>
20 #include <asm/uaccess.h>
22 void fastcall
set_close_on_exec(unsigned int fd
, int flag
)
24 struct files_struct
*files
= current
->files
;
25 spin_lock(&files
->file_lock
);
27 FD_SET(fd
, files
->close_on_exec
);
29 FD_CLR(fd
, files
->close_on_exec
);
30 spin_unlock(&files
->file_lock
);
33 static inline int get_close_on_exec(unsigned int fd
)
35 struct files_struct
*files
= current
->files
;
37 spin_lock(&files
->file_lock
);
38 res
= FD_ISSET(fd
, files
->close_on_exec
);
39 spin_unlock(&files
->file_lock
);
44 /* Expand files. Return <0 on error; 0 nothing done; 1 files expanded,
45 * we may have blocked.
47 * Should be called with the files->file_lock spinlock held for write.
49 static int expand_files(struct files_struct
*files
, int nr
)
53 printk (KERN_ERR
"%s %d: nr = %d\n", __FUNCTION__
, current
->pid
, nr
);
56 if (nr
>= files
->max_fdset
) {
58 if ((err
= expand_fdset(files
, nr
)))
61 if (nr
>= files
->max_fds
) {
63 if ((err
= expand_fd_array(files
, nr
)))
70 printk (KERN_ERR
"%s %d: return %d\n", __FUNCTION__
, current
->pid
, err
);
76 * locate_fd finds a free file descriptor in the open_fds fdset,
77 * expanding the fd arrays if necessary. Must be called with the
78 * file_lock held for write.
81 static int locate_fd(struct files_struct
*files
,
82 struct file
*file
, unsigned int orig_start
)
89 if (orig_start
>= current
->rlim
[RLIMIT_NOFILE
].rlim_cur
)
94 * Someone might have closed fd's in the range
95 * orig_start..files->next_fd
98 if (start
< files
->next_fd
)
99 start
= files
->next_fd
;
102 if (start
< files
->max_fdset
) {
103 newfd
= find_next_zero_bit(files
->open_fds
->fds_bits
,
104 files
->max_fdset
, start
);
108 if (newfd
>= current
->rlim
[RLIMIT_NOFILE
].rlim_cur
)
111 error
= expand_files(files
, newfd
);
116 * If we needed to expand the fs array we
117 * might have blocked - try again.
122 if (start
<= files
->next_fd
)
123 files
->next_fd
= newfd
+ 1;
131 static int dupfd(struct file
*file
, unsigned int start
)
133 struct files_struct
* files
= current
->files
;
136 spin_lock(&files
->file_lock
);
137 fd
= locate_fd(files
, file
, start
);
139 FD_SET(fd
, files
->open_fds
);
140 FD_CLR(fd
, files
->close_on_exec
);
141 spin_unlock(&files
->file_lock
);
142 fd_install(fd
, file
);
144 spin_unlock(&files
->file_lock
);
151 asmlinkage
long sys_dup2(unsigned int oldfd
, unsigned int newfd
)
154 struct file
* file
, *tofree
;
155 struct files_struct
* files
= current
->files
;
157 spin_lock(&files
->file_lock
);
158 if (!(file
= fcheck(oldfd
)))
164 if (newfd
>= current
->rlim
[RLIMIT_NOFILE
].rlim_cur
)
166 get_file(file
); /* We are now finished with oldfd */
168 err
= expand_files(files
, newfd
);
172 /* To avoid races with open() and dup(), we will mark the fd as
173 * in-use in the open-file bitmap throughout the entire dup2()
174 * process. This is quite safe: do_close() uses the fd array
175 * entry, not the bitmap, to decide what work needs to be
177 /* Doesn't work. open() might be there first. --AV */
179 /* Yes. It's a race. In user space. Nothing sane to do */
181 tofree
= files
->fd
[newfd
];
182 if (!tofree
&& FD_ISSET(newfd
, files
->open_fds
))
185 files
->fd
[newfd
] = file
;
186 FD_SET(newfd
, files
->open_fds
);
187 FD_CLR(newfd
, files
->close_on_exec
);
188 spin_unlock(&files
->file_lock
);
191 filp_close(tofree
, files
);
196 spin_unlock(&files
->file_lock
);
200 spin_unlock(&files
->file_lock
);
205 asmlinkage
long sys_dup(unsigned int fildes
)
208 struct file
* file
= fget(fildes
);
211 ret
= dupfd(file
, 0);
215 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
217 static int setfl(int fd
, struct file
* filp
, unsigned long arg
)
219 struct inode
* inode
= filp
->f_dentry
->d_inode
;
222 /* O_APPEND cannot be cleared if the file is marked as append-only */
223 if (!(arg
& 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 (current
->fsuid
!= inode
->i_uid
&& !capable(CAP_FOWNER
))
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
);
248 if ((arg
^ filp
->f_flags
) & FASYNC
) {
249 if (filp
->f_op
&& filp
->f_op
->fasync
) {
250 error
= filp
->f_op
->fasync(fd
, filp
, (arg
& FASYNC
) != 0);
256 filp
->f_flags
= (arg
& SETFL_MASK
) | (filp
->f_flags
& ~SETFL_MASK
);
262 static void f_modown(struct file
*filp
, unsigned long pid
,
263 uid_t uid
, uid_t euid
, int force
)
265 write_lock_irq(&filp
->f_owner
.lock
);
266 if (force
|| !filp
->f_owner
.pid
) {
267 filp
->f_owner
.pid
= pid
;
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
, unsigned long arg
, int force
)
278 err
= security_file_set_fowner(filp
);
282 f_modown(filp
, arg
, current
->uid
, current
->euid
, force
);
286 EXPORT_SYMBOL(f_setown
);
288 void f_delown(struct file
*filp
)
290 f_modown(filp
, 0, 0, 0, 1);
293 EXPORT_SYMBOL(f_delown
);
295 static long do_fcntl(int fd
, unsigned int cmd
, unsigned long arg
,
303 err
= dupfd(filp
, arg
);
306 err
= get_close_on_exec(fd
) ? FD_CLOEXEC
: 0;
310 set_close_on_exec(fd
, arg
& FD_CLOEXEC
);
316 err
= setfl(fd
, filp
, arg
);
319 err
= fcntl_getlk(filp
, (struct flock __user
*) arg
);
323 err
= fcntl_setlk(filp
, cmd
, (struct flock __user
*) arg
);
327 * XXX If f_owner is a process group, the
328 * negative return value will get converted
329 * into an error. Oops. If we keep the
330 * current syscall conventions, the only way
331 * to fix this will be in libc.
333 err
= filp
->f_owner
.pid
;
334 force_successful_syscall_return();
337 err
= f_setown(filp
, arg
, 1);
340 err
= filp
->f_owner
.signum
;
343 /* arg == 0 restores default behaviour. */
344 if (arg
< 0 || arg
> _NSIG
) {
348 filp
->f_owner
.signum
= arg
;
351 err
= fcntl_getlease(filp
);
354 err
= fcntl_setlease(fd
, filp
, arg
);
357 err
= fcntl_dirnotify(fd
, filp
, arg
);
365 asmlinkage
long sys_fcntl(int fd
, unsigned int cmd
, unsigned long arg
)
374 err
= security_file_fcntl(filp
, cmd
, arg
);
380 err
= do_fcntl(fd
, cmd
, arg
, filp
);
387 #if BITS_PER_LONG == 32
388 asmlinkage
long sys_fcntl64(unsigned int fd
, unsigned int cmd
, unsigned long arg
)
398 err
= security_file_fcntl(filp
, cmd
, arg
);
407 err
= fcntl_getlk64(filp
, (struct flock64 __user
*) arg
);
411 err
= fcntl_setlk64(filp
, cmd
, (struct flock64 __user
*) arg
);
414 err
= do_fcntl(fd
, cmd
, arg
, filp
);
423 /* Table to convert sigio signal codes into poll band bitmaps */
425 static long band_table
[NSIGPOLL
] = {
426 POLLIN
| POLLRDNORM
, /* POLL_IN */
427 POLLOUT
| POLLWRNORM
| POLLWRBAND
, /* POLL_OUT */
428 POLLIN
| POLLRDNORM
| POLLMSG
, /* POLL_MSG */
429 POLLERR
, /* POLL_ERR */
430 POLLPRI
| POLLRDBAND
, /* POLL_PRI */
431 POLLHUP
| POLLERR
/* POLL_HUP */
434 static inline int sigio_perm(struct task_struct
*p
,
435 struct fown_struct
*fown
)
437 return ((fown
->euid
== 0) ||
438 (fown
->euid
== p
->suid
) || (fown
->euid
== p
->uid
) ||
439 (fown
->uid
== p
->suid
) || (fown
->uid
== p
->uid
));
442 static void send_sigio_to_task(struct task_struct
*p
,
443 struct fown_struct
*fown
,
447 if (!sigio_perm(p
, fown
))
450 if (security_file_send_sigiotask(p
, fown
, fd
, reason
))
453 switch (fown
->signum
) {
456 /* Queue a rt signal with the appropriate fd as its
457 value. We use SI_SIGIO as the source, not
458 SI_KERNEL, since kernel signals always get
459 delivered even if we can't queue. Failure to
460 queue in this case _should_ be reported; we fall
461 back to SIGIO in that case. --sct */
462 si
.si_signo
= fown
->signum
;
465 /* Make sure we are called with one of the POLL_*
466 reasons, otherwise we could leak kernel stack into
468 if ((reason
& __SI_MASK
) != __SI_POLL
)
470 if (reason
- POLL_IN
>= NSIGPOLL
)
473 si
.si_band
= band_table
[reason
- POLL_IN
];
475 if (!send_sig_info(fown
->signum
, &si
, p
))
477 /* fall-through: fall back on the old plain SIGIO signal */
479 send_group_sig_info(SIGIO
, SEND_SIG_PRIV
, p
);
483 void send_sigio(struct fown_struct
*fown
, int fd
, int band
)
485 struct task_struct
*p
;
488 read_lock(&fown
->lock
);
491 goto out_unlock_fown
;
493 read_lock(&tasklist_lock
);
495 p
= find_task_by_pid(pid
);
497 send_sigio_to_task(p
, fown
, fd
, band
);
500 do_each_task_pid(-pid
, PIDTYPE_PGID
, p
) {
501 send_sigio_to_task(p
, fown
, fd
, band
);
502 } while_each_task_pid(-pid
, PIDTYPE_PGID
, p
);
504 read_unlock(&tasklist_lock
);
506 read_unlock(&fown
->lock
);
509 static void send_sigurg_to_task(struct task_struct
*p
,
510 struct fown_struct
*fown
)
512 if (sigio_perm(p
, fown
))
513 send_group_sig_info(SIGURG
, SEND_SIG_PRIV
, p
);
516 int send_sigurg(struct fown_struct
*fown
)
518 struct task_struct
*p
;
521 read_lock(&fown
->lock
);
524 goto out_unlock_fown
;
528 read_lock(&tasklist_lock
);
530 p
= find_task_by_pid(pid
);
532 send_sigurg_to_task(p
, fown
);
535 do_each_task_pid(-pid
, PIDTYPE_PGID
, p
) {
536 send_sigurg_to_task(p
, fown
);
537 } while_each_task_pid(-pid
, PIDTYPE_PGID
, p
);
539 read_unlock(&tasklist_lock
);
541 read_unlock(&fown
->lock
);
545 static rwlock_t fasync_lock
= RW_LOCK_UNLOCKED
;
546 static kmem_cache_t
*fasync_cache
;
549 * fasync_helper() is used by some character device drivers (mainly mice)
550 * to set up the fasync queue. It returns negative on error, 0 if it did
551 * no changes and positive if it added/deleted the entry.
553 int fasync_helper(int fd
, struct file
* filp
, int on
, struct fasync_struct
**fapp
)
555 struct fasync_struct
*fa
, **fp
;
556 struct fasync_struct
*new = NULL
;
560 new = kmem_cache_alloc(fasync_cache
, SLAB_KERNEL
);
564 write_lock_irq(&fasync_lock
);
565 for (fp
= fapp
; (fa
= *fp
) != NULL
; fp
= &fa
->fa_next
) {
566 if (fa
->fa_file
== filp
) {
569 kmem_cache_free(fasync_cache
, new);
572 kmem_cache_free(fasync_cache
, fa
);
580 new->magic
= FASYNC_MAGIC
;
583 new->fa_next
= *fapp
;
588 write_unlock_irq(&fasync_lock
);
592 EXPORT_SYMBOL(fasync_helper
);
594 void __kill_fasync(struct fasync_struct
*fa
, int sig
, int band
)
597 struct fown_struct
* fown
;
598 if (fa
->magic
!= FASYNC_MAGIC
) {
599 printk(KERN_ERR
"kill_fasync: bad magic number in "
603 fown
= &fa
->fa_file
->f_owner
;
604 /* Don't send SIGURG to processes which have not set a
605 queued signum: SIGURG has its own default signalling
607 if (!(sig
== SIGURG
&& fown
->signum
== 0))
608 send_sigio(fown
, fa
->fa_fd
, band
);
613 EXPORT_SYMBOL(__kill_fasync
);
615 void kill_fasync(struct fasync_struct
**fp
, int sig
, int band
)
617 /* First a quick test without locking: usually
621 read_lock(&fasync_lock
);
622 /* reread *fp after obtaining the lock */
623 __kill_fasync(*fp
, sig
, band
);
624 read_unlock(&fasync_lock
);
627 EXPORT_SYMBOL(kill_fasync
);
629 static int __init
fasync_init(void)
631 fasync_cache
= kmem_cache_create("fasync_cache",
632 sizeof(struct fasync_struct
), 0, SLAB_PANIC
, NULL
, NULL
);
636 module_init(fasync_init
)