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)
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 /* required for strict SunOS emulation */
227 if (O_NONBLOCK
!= O_NDELAY
)
231 if (arg
& O_DIRECT
) {
232 if (!filp
->f_mapping
|| !filp
->f_mapping
->a_ops
||
233 !filp
->f_mapping
->a_ops
->direct_IO
)
238 if ((arg
^ filp
->f_flags
) & FASYNC
) {
239 if (filp
->f_op
&& filp
->f_op
->fasync
) {
240 error
= filp
->f_op
->fasync(fd
, filp
, (arg
& FASYNC
) != 0);
246 filp
->f_flags
= (arg
& SETFL_MASK
) | (filp
->f_flags
& ~SETFL_MASK
);
252 static void f_modown(struct file
*filp
, unsigned long pid
,
253 uid_t uid
, uid_t euid
, int force
)
255 write_lock_irq(&filp
->f_owner
.lock
);
256 if (force
|| !filp
->f_owner
.pid
) {
257 filp
->f_owner
.pid
= pid
;
258 filp
->f_owner
.uid
= uid
;
259 filp
->f_owner
.euid
= euid
;
261 write_unlock_irq(&filp
->f_owner
.lock
);
264 int f_setown(struct file
*filp
, unsigned long arg
, int force
)
268 err
= security_file_set_fowner(filp
);
272 f_modown(filp
, arg
, current
->uid
, current
->euid
, force
);
276 EXPORT_SYMBOL(f_setown
);
278 void f_delown(struct file
*filp
)
280 f_modown(filp
, 0, 0, 0, 1);
283 EXPORT_SYMBOL(f_delown
);
285 static long do_fcntl(unsigned int fd
, unsigned int cmd
,
286 unsigned long arg
, struct file
* filp
)
293 err
= dupfd(filp
, arg
);
296 err
= get_close_on_exec(fd
) ? FD_CLOEXEC
: 0;
300 set_close_on_exec(fd
, arg
& FD_CLOEXEC
);
306 err
= setfl(fd
, filp
, arg
);
309 err
= fcntl_getlk(filp
, (struct flock __user
*) arg
);
313 err
= fcntl_setlk(filp
, cmd
, (struct flock __user
*) arg
);
317 * XXX If f_owner is a process group, the
318 * negative return value will get converted
319 * into an error. Oops. If we keep the
320 * current syscall conventions, the only way
321 * to fix this will be in libc.
323 err
= filp
->f_owner
.pid
;
324 force_successful_syscall_return();
327 err
= f_setown(filp
, arg
, 1);
330 err
= filp
->f_owner
.signum
;
333 /* arg == 0 restores default behaviour. */
334 if (arg
< 0 || arg
> _NSIG
) {
338 filp
->f_owner
.signum
= arg
;
341 err
= fcntl_getlease(filp
);
344 err
= fcntl_setlease(fd
, filp
, arg
);
347 err
= fcntl_dirnotify(fd
, filp
, arg
);
356 asmlinkage
long sys_fcntl(unsigned int fd
, unsigned int cmd
, unsigned long arg
)
365 err
= security_file_fcntl(filp
, cmd
, arg
);
371 err
= do_fcntl(fd
, cmd
, arg
, filp
);
378 #if BITS_PER_LONG == 32
379 asmlinkage
long sys_fcntl64(unsigned int fd
, unsigned int cmd
, unsigned long arg
)
389 err
= security_file_fcntl(filp
, cmd
, arg
);
398 err
= fcntl_getlk64(filp
, (struct flock64 __user
*) arg
);
402 err
= fcntl_setlk64(filp
, cmd
, (struct flock64 __user
*) arg
);
405 err
= do_fcntl(fd
, cmd
, arg
, filp
);
414 /* Table to convert sigio signal codes into poll band bitmaps */
416 static long band_table
[NSIGPOLL
] = {
417 POLLIN
| POLLRDNORM
, /* POLL_IN */
418 POLLOUT
| POLLWRNORM
| POLLWRBAND
, /* POLL_OUT */
419 POLLIN
| POLLRDNORM
| POLLMSG
, /* POLL_MSG */
420 POLLERR
, /* POLL_ERR */
421 POLLPRI
| POLLRDBAND
, /* POLL_PRI */
422 POLLHUP
| POLLERR
/* POLL_HUP */
425 static inline int sigio_perm(struct task_struct
*p
,
426 struct fown_struct
*fown
)
428 return ((fown
->euid
== 0) ||
429 (fown
->euid
== p
->suid
) || (fown
->euid
== p
->uid
) ||
430 (fown
->uid
== p
->suid
) || (fown
->uid
== p
->uid
));
433 static void send_sigio_to_task(struct task_struct
*p
,
434 struct fown_struct
*fown
,
438 if (!sigio_perm(p
, fown
))
441 if (security_file_send_sigiotask(p
, fown
, fd
, reason
))
444 switch (fown
->signum
) {
447 /* Queue a rt signal with the appropriate fd as its
448 value. We use SI_SIGIO as the source, not
449 SI_KERNEL, since kernel signals always get
450 delivered even if we can't queue. Failure to
451 queue in this case _should_ be reported; we fall
452 back to SIGIO in that case. --sct */
453 si
.si_signo
= fown
->signum
;
456 /* Make sure we are called with one of the POLL_*
457 reasons, otherwise we could leak kernel stack into
459 if ((reason
& __SI_MASK
) != __SI_POLL
)
461 if (reason
- POLL_IN
>= NSIGPOLL
)
464 si
.si_band
= band_table
[reason
- POLL_IN
];
466 if (!send_sig_info(fown
->signum
, &si
, p
))
468 /* fall-through: fall back on the old plain SIGIO signal */
470 send_group_sig_info(SIGIO
, SEND_SIG_PRIV
, p
);
474 void send_sigio(struct fown_struct
*fown
, int fd
, int band
)
476 struct task_struct
*p
;
479 read_lock(&fown
->lock
);
482 goto out_unlock_fown
;
484 read_lock(&tasklist_lock
);
486 p
= find_task_by_pid(pid
);
488 send_sigio_to_task(p
, fown
, fd
, band
);
493 for_each_task_pid(-pid
, PIDTYPE_PGID
, p
, l
, pidptr
) {
494 send_sigio_to_task(p
, fown
, fd
, band
);
497 read_unlock(&tasklist_lock
);
499 read_unlock(&fown
->lock
);
502 static void send_sigurg_to_task(struct task_struct
*p
,
503 struct fown_struct
*fown
)
505 if (sigio_perm(p
, fown
))
506 send_group_sig_info(SIGURG
, SEND_SIG_PRIV
, p
);
509 int send_sigurg(struct fown_struct
*fown
)
511 struct task_struct
*p
;
514 read_lock(&fown
->lock
);
517 goto out_unlock_fown
;
521 read_lock(&tasklist_lock
);
523 p
= find_task_by_pid(pid
);
525 send_sigurg_to_task(p
, fown
);
530 for_each_task_pid(-pid
, PIDTYPE_PGID
, p
, l
, pidptr
) {
531 send_sigurg_to_task(p
, fown
);
534 read_unlock(&tasklist_lock
);
536 read_unlock(&fown
->lock
);
540 static rwlock_t fasync_lock
= RW_LOCK_UNLOCKED
;
541 static kmem_cache_t
*fasync_cache
;
544 * fasync_helper() is used by some character device drivers (mainly mice)
545 * to set up the fasync queue. It returns negative on error, 0 if it did
546 * no changes and positive if it added/deleted the entry.
548 int fasync_helper(int fd
, struct file
* filp
, int on
, struct fasync_struct
**fapp
)
550 struct fasync_struct
*fa
, **fp
;
551 struct fasync_struct
*new = NULL
;
555 new = kmem_cache_alloc(fasync_cache
, SLAB_KERNEL
);
559 write_lock_irq(&fasync_lock
);
560 for (fp
= fapp
; (fa
= *fp
) != NULL
; fp
= &fa
->fa_next
) {
561 if (fa
->fa_file
== filp
) {
564 kmem_cache_free(fasync_cache
, new);
567 kmem_cache_free(fasync_cache
, fa
);
575 new->magic
= FASYNC_MAGIC
;
578 new->fa_next
= *fapp
;
583 write_unlock_irq(&fasync_lock
);
587 EXPORT_SYMBOL(fasync_helper
);
589 void __kill_fasync(struct fasync_struct
*fa
, int sig
, int band
)
592 struct fown_struct
* fown
;
593 if (fa
->magic
!= FASYNC_MAGIC
) {
594 printk(KERN_ERR
"kill_fasync: bad magic number in "
598 fown
= &fa
->fa_file
->f_owner
;
599 /* Don't send SIGURG to processes which have not set a
600 queued signum: SIGURG has its own default signalling
602 if (!(sig
== SIGURG
&& fown
->signum
== 0))
603 send_sigio(fown
, fa
->fa_fd
, band
);
608 EXPORT_SYMBOL(__kill_fasync
);
610 void kill_fasync(struct fasync_struct
**fp
, int sig
, int band
)
612 /* First a quick test without locking: usually
616 read_lock(&fasync_lock
);
617 /* reread *fp after obtaining the lock */
618 __kill_fasync(*fp
, sig
, band
);
619 read_unlock(&fasync_lock
);
623 EXPORT_SYMBOL(kill_fasync
);
625 static int __init
fasync_init(void)
627 fasync_cache
= kmem_cache_create("fasync_cache",
628 sizeof(struct fasync_struct
), 0, 0, NULL
, NULL
);
630 panic("cannot create fasync slab cache");
634 module_init(fasync_init
)