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>
24 #include <asm/siginfo.h>
25 #include <asm/uaccess.h>
27 void set_close_on_exec(unsigned int fd, int flag)
29 struct files_struct *files = current->files;
31 spin_lock(&files->file_lock);
32 fdt = files_fdtable(files);
34 FD_SET(fd, fdt->close_on_exec);
36 FD_CLR(fd, fdt->close_on_exec);
37 spin_unlock(&files->file_lock);
40 static int get_close_on_exec(unsigned int fd)
42 struct files_struct *files = current->files;
46 fdt = files_fdtable(files);
47 res = FD_ISSET(fd, fdt->close_on_exec);
53 * locate_fd finds a free file descriptor in the open_fds fdset,
54 * expanding the fd arrays if necessary. Must be called with the
55 * file_lock held for write.
58 static int locate_fd(unsigned int orig_start, int cloexec)
60 struct files_struct *files = current->files;
66 spin_lock(&files->file_lock);
68 fdt = files_fdtable(files);
70 * Someone might have closed fd's in the range
71 * orig_start..fdt->next_fd
74 if (start < files->next_fd)
75 start = files->next_fd;
78 if (start < fdt->max_fds)
79 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
82 error = expand_files(files, newfd);
87 * If we needed to expand the fs array we
88 * might have blocked - try again.
93 if (start <= files->next_fd)
94 files->next_fd = newfd + 1;
96 FD_SET(newfd, fdt->open_fds);
98 FD_SET(newfd, fdt->close_on_exec);
100 FD_CLR(newfd, fdt->close_on_exec);
104 spin_unlock(&files->file_lock);
108 static int dupfd(struct file *file, unsigned int start, int cloexec)
110 int fd = locate_fd(start, cloexec);
112 fd_install(fd, file);
119 asmlinkage long sys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
122 struct file * file, *tofree;
123 struct files_struct * files = current->files;
126 if ((flags & ~O_CLOEXEC) != 0)
129 if (unlikely(oldfd == newfd))
132 spin_lock(&files->file_lock);
133 if (!(file = fcheck(oldfd)))
135 get_file(file); /* We are now finished with oldfd */
137 err = expand_files(files, newfd);
138 if (unlikely(err < 0)) {
144 /* To avoid races with open() and dup(), we will mark the fd as
145 * in-use in the open-file bitmap throughout the entire dup2()
146 * process. This is quite safe: do_close() uses the fd array
147 * entry, not the bitmap, to decide what work needs to be
149 /* Doesn't work. open() might be there first. --AV */
151 /* Yes. It's a race. In user space. Nothing sane to do */
153 fdt = files_fdtable(files);
154 tofree = fdt->fd[newfd];
155 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
158 rcu_assign_pointer(fdt->fd[newfd], file);
159 FD_SET(newfd, fdt->open_fds);
160 if (flags & O_CLOEXEC)
161 FD_SET(newfd, fdt->close_on_exec);
163 FD_CLR(newfd, fdt->close_on_exec);
164 spin_unlock(&files->file_lock);
167 filp_close(tofree, files);
172 spin_unlock(&files->file_lock);
176 spin_unlock(&files->file_lock);
181 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
183 if (unlikely(newfd == oldfd)) { /* corner case */
184 struct files_struct *files = current->files;
186 if (!fcheck_files(files, oldfd))
191 return sys_dup3(oldfd, newfd, 0);
194 asmlinkage long sys_dup(unsigned int fildes)
197 struct file * file = fget(fildes);
200 ret = dupfd(file, 0, 0);
204 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
206 static int setfl(int fd, struct file * filp, unsigned long arg)
208 struct inode * inode = filp->f_path.dentry->d_inode;
212 * O_APPEND cannot be cleared if the file is marked as append-only
213 * and the file is open for write.
215 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
218 /* O_NOATIME can only be set by the owner or superuser */
219 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
220 if (!is_owner_or_cap(inode))
223 /* required for strict SunOS emulation */
224 if (O_NONBLOCK != O_NDELAY)
228 if (arg & O_DIRECT) {
229 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
230 !filp->f_mapping->a_ops->direct_IO)
234 if (filp->f_op && filp->f_op->check_flags)
235 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);
252 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
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 put_pid(filp->f_owner.pid);
258 filp->f_owner.pid = get_pid(pid);
259 filp->f_owner.pid_type = type;
260 filp->f_owner.uid = uid;
261 filp->f_owner.euid = euid;
263 write_unlock_irq(&filp->f_owner.lock);
266 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
271 err = security_file_set_fowner(filp);
275 f_modown(filp, pid, type, current->uid, current->euid, force);
278 EXPORT_SYMBOL(__f_setown);
280 int f_setown(struct file *filp, unsigned long arg, int force)
292 pid = find_vpid(who);
293 result = __f_setown(filp, pid, type, force);
297 EXPORT_SYMBOL(f_setown);
299 void f_delown(struct file *filp)
301 f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
304 pid_t f_getown(struct file *filp)
307 read_lock(&filp->f_owner.lock);
308 pid = pid_vnr(filp->f_owner.pid);
309 if (filp->f_owner.pid_type == PIDTYPE_PGID)
311 read_unlock(&filp->f_owner.lock);
315 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
322 case F_DUPFD_CLOEXEC:
323 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
326 err = dupfd(filp, arg, cmd == F_DUPFD_CLOEXEC);
329 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
333 set_close_on_exec(fd, arg & FD_CLOEXEC);
339 err = setfl(fd, filp, arg);
342 err = fcntl_getlk(filp, (struct flock __user *) arg);
346 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
350 * XXX If f_owner is a process group, the
351 * negative return value will get converted
352 * into an error. Oops. If we keep the
353 * current syscall conventions, the only way
354 * to fix this will be in libc.
356 err = f_getown(filp);
357 force_successful_syscall_return();
360 err = f_setown(filp, arg, 1);
363 err = filp->f_owner.signum;
366 /* arg == 0 restores default behaviour. */
367 if (!valid_signal(arg)) {
371 filp->f_owner.signum = arg;
374 err = fcntl_getlease(filp);
377 err = fcntl_setlease(fd, filp, arg);
380 err = fcntl_dirnotify(fd, filp, arg);
388 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
397 err = security_file_fcntl(filp, cmd, arg);
403 err = do_fcntl(fd, cmd, arg, filp);
410 #if BITS_PER_LONG == 32
411 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
421 err = security_file_fcntl(filp, cmd, arg);
430 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
434 err = fcntl_setlk64(fd, filp, cmd,
435 (struct flock64 __user *) arg);
438 err = do_fcntl(fd, cmd, arg, filp);
447 /* Table to convert sigio signal codes into poll band bitmaps */
449 static const long band_table[NSIGPOLL] = {
450 POLLIN | POLLRDNORM, /* POLL_IN */
451 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
452 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
453 POLLERR, /* POLL_ERR */
454 POLLPRI | POLLRDBAND, /* POLL_PRI */
455 POLLHUP | POLLERR /* POLL_HUP */
458 static inline int sigio_perm(struct task_struct *p,
459 struct fown_struct *fown, int sig)
461 return (((fown->euid == 0) ||
462 (fown->euid == p->suid) || (fown->euid == p->uid) ||
463 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
464 !security_file_send_sigiotask(p, fown, sig));
467 static void send_sigio_to_task(struct task_struct *p,
468 struct fown_struct *fown,
472 if (!sigio_perm(p, fown, fown->signum))
475 switch (fown->signum) {
478 /* Queue a rt signal with the appropriate fd as its
479 value. We use SI_SIGIO as the source, not
480 SI_KERNEL, since kernel signals always get
481 delivered even if we can't queue. Failure to
482 queue in this case _should_ be reported; we fall
483 back to SIGIO in that case. --sct */
484 si.si_signo = fown->signum;
487 /* Make sure we are called with one of the POLL_*
488 reasons, otherwise we could leak kernel stack into
490 BUG_ON((reason & __SI_MASK) != __SI_POLL);
491 if (reason - POLL_IN >= NSIGPOLL)
494 si.si_band = band_table[reason - POLL_IN];
496 if (!group_send_sig_info(fown->signum, &si, p))
498 /* fall-through: fall back on the old plain SIGIO signal */
500 group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
504 void send_sigio(struct fown_struct *fown, int fd, int band)
506 struct task_struct *p;
510 read_lock(&fown->lock);
511 type = fown->pid_type;
514 goto out_unlock_fown;
516 read_lock(&tasklist_lock);
517 do_each_pid_task(pid, type, p) {
518 send_sigio_to_task(p, fown, fd, band);
519 } while_each_pid_task(pid, type, p);
520 read_unlock(&tasklist_lock);
522 read_unlock(&fown->lock);
525 static void send_sigurg_to_task(struct task_struct *p,
526 struct fown_struct *fown)
528 if (sigio_perm(p, fown, SIGURG))
529 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
532 int send_sigurg(struct fown_struct *fown)
534 struct task_struct *p;
539 read_lock(&fown->lock);
540 type = fown->pid_type;
543 goto out_unlock_fown;
547 read_lock(&tasklist_lock);
548 do_each_pid_task(pid, type, p) {
549 send_sigurg_to_task(p, fown);
550 } while_each_pid_task(pid, type, p);
551 read_unlock(&tasklist_lock);
553 read_unlock(&fown->lock);
557 static DEFINE_RWLOCK(fasync_lock);
558 static struct kmem_cache *fasync_cache __read_mostly;
561 * fasync_helper() is used by some character device drivers (mainly mice)
562 * to set up the fasync queue. It returns negative on error, 0 if it did
563 * no changes and positive if it added/deleted the entry.
565 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
567 struct fasync_struct *fa, **fp;
568 struct fasync_struct *new = NULL;
572 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
576 write_lock_irq(&fasync_lock);
577 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
578 if (fa->fa_file == filp) {
581 kmem_cache_free(fasync_cache, new);
584 kmem_cache_free(fasync_cache, fa);
592 new->magic = FASYNC_MAGIC;
595 new->fa_next = *fapp;
600 write_unlock_irq(&fasync_lock);
604 EXPORT_SYMBOL(fasync_helper);
606 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
609 struct fown_struct * fown;
610 if (fa->magic != FASYNC_MAGIC) {
611 printk(KERN_ERR "kill_fasync: bad magic number in "
615 fown = &fa->fa_file->f_owner;
616 /* Don't send SIGURG to processes which have not set a
617 queued signum: SIGURG has its own default signalling
619 if (!(sig == SIGURG && fown->signum == 0))
620 send_sigio(fown, fa->fa_fd, band);
625 EXPORT_SYMBOL(__kill_fasync);
627 void kill_fasync(struct fasync_struct **fp, int sig, int band)
629 /* First a quick test without locking: usually
633 read_lock(&fasync_lock);
634 /* reread *fp after obtaining the lock */
635 __kill_fasync(*fp, sig, band);
636 read_unlock(&fasync_lock);
639 EXPORT_SYMBOL(kill_fasync);
641 static int __init fasync_init(void)
643 fasync_cache = kmem_cache_create("fasync_cache",
644 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
648 module_init(fasync_init)