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);
69 if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
73 fdt = files_fdtable(files);
75 * Someone might have closed fd's in the range
76 * orig_start..fdt->next_fd
79 if (start < files->next_fd)
80 start = files->next_fd;
83 if (start < fdt->max_fds)
84 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
88 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
91 error = expand_files(files, newfd);
96 * If we needed to expand the fs array we
97 * might have blocked - try again.
102 if (start <= files->next_fd)
103 files->next_fd = newfd + 1;
105 FD_SET(newfd, fdt->open_fds);
107 FD_SET(newfd, fdt->close_on_exec);
109 FD_CLR(newfd, fdt->close_on_exec);
113 spin_unlock(&files->file_lock);
117 static int dupfd(struct file *file, unsigned int start, int cloexec)
119 int fd = locate_fd(start, cloexec);
121 fd_install(fd, file);
128 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
131 struct file * file, *tofree;
132 struct files_struct * files = current->files;
135 spin_lock(&files->file_lock);
136 if (!(file = fcheck(oldfd)))
142 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
144 get_file(file); /* We are now finished with oldfd */
146 err = expand_files(files, newfd);
150 /* To avoid races with open() and dup(), we will mark the fd as
151 * in-use in the open-file bitmap throughout the entire dup2()
152 * process. This is quite safe: do_close() uses the fd array
153 * entry, not the bitmap, to decide what work needs to be
155 /* Doesn't work. open() might be there first. --AV */
157 /* Yes. It's a race. In user space. Nothing sane to do */
159 fdt = files_fdtable(files);
160 tofree = fdt->fd[newfd];
161 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
164 rcu_assign_pointer(fdt->fd[newfd], file);
165 FD_SET(newfd, fdt->open_fds);
166 FD_CLR(newfd, fdt->close_on_exec);
167 spin_unlock(&files->file_lock);
170 filp_close(tofree, files);
175 spin_unlock(&files->file_lock);
179 spin_unlock(&files->file_lock);
184 asmlinkage long sys_dup(unsigned int fildes)
187 struct file * file = fget(fildes);
190 ret = dupfd(file, 0, 0);
194 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
196 static int setfl(int fd, struct file * filp, unsigned long arg)
198 struct inode * inode = filp->f_path.dentry->d_inode;
202 * O_APPEND cannot be cleared if the file is marked as append-only
203 * and the file is open for write.
205 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
208 /* O_NOATIME can only be set by the owner or superuser */
209 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
210 if (!is_owner_or_cap(inode))
213 /* required for strict SunOS emulation */
214 if (O_NONBLOCK != O_NDELAY)
218 if (arg & O_DIRECT) {
219 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
220 !filp->f_mapping->a_ops->direct_IO)
224 if (filp->f_op && filp->f_op->check_flags)
225 error = filp->f_op->check_flags(arg);
229 if ((arg ^ filp->f_flags) & FASYNC) {
230 if (filp->f_op && filp->f_op->fasync) {
231 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
237 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
242 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
243 uid_t uid, uid_t euid, int force)
245 write_lock_irq(&filp->f_owner.lock);
246 if (force || !filp->f_owner.pid) {
247 put_pid(filp->f_owner.pid);
248 filp->f_owner.pid = get_pid(pid);
249 filp->f_owner.pid_type = type;
250 filp->f_owner.uid = uid;
251 filp->f_owner.euid = euid;
253 write_unlock_irq(&filp->f_owner.lock);
256 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
261 err = security_file_set_fowner(filp);
265 f_modown(filp, pid, type, current->uid, current->euid, force);
268 EXPORT_SYMBOL(__f_setown);
270 int f_setown(struct file *filp, unsigned long arg, int force)
282 pid = find_vpid(who);
283 result = __f_setown(filp, pid, type, force);
287 EXPORT_SYMBOL(f_setown);
289 void f_delown(struct file *filp)
291 f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
294 pid_t f_getown(struct file *filp)
297 read_lock(&filp->f_owner.lock);
298 pid = pid_vnr(filp->f_owner.pid);
299 if (filp->f_owner.pid_type == PIDTYPE_PGID)
301 read_unlock(&filp->f_owner.lock);
305 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
312 case F_DUPFD_CLOEXEC:
314 err = dupfd(filp, arg, cmd == F_DUPFD_CLOEXEC);
317 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
321 set_close_on_exec(fd, arg & FD_CLOEXEC);
327 err = setfl(fd, filp, arg);
330 err = fcntl_getlk(filp, (struct flock __user *) arg);
334 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
338 * XXX If f_owner is a process group, the
339 * negative return value will get converted
340 * into an error. Oops. If we keep the
341 * current syscall conventions, the only way
342 * to fix this will be in libc.
344 err = f_getown(filp);
345 force_successful_syscall_return();
348 err = f_setown(filp, arg, 1);
351 err = filp->f_owner.signum;
354 /* arg == 0 restores default behaviour. */
355 if (!valid_signal(arg)) {
359 filp->f_owner.signum = arg;
362 err = fcntl_getlease(filp);
365 err = fcntl_setlease(fd, filp, arg);
368 err = fcntl_dirnotify(fd, filp, arg);
376 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
385 err = security_file_fcntl(filp, cmd, arg);
391 err = do_fcntl(fd, cmd, arg, filp);
398 #if BITS_PER_LONG == 32
399 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
409 err = security_file_fcntl(filp, cmd, arg);
418 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
422 err = fcntl_setlk64(fd, filp, cmd,
423 (struct flock64 __user *) arg);
426 err = do_fcntl(fd, cmd, arg, filp);
435 /* Table to convert sigio signal codes into poll band bitmaps */
437 static const long band_table[NSIGPOLL] = {
438 POLLIN | POLLRDNORM, /* POLL_IN */
439 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
440 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
441 POLLERR, /* POLL_ERR */
442 POLLPRI | POLLRDBAND, /* POLL_PRI */
443 POLLHUP | POLLERR /* POLL_HUP */
446 static inline int sigio_perm(struct task_struct *p,
447 struct fown_struct *fown, int sig)
449 return (((fown->euid == 0) ||
450 (fown->euid == p->suid) || (fown->euid == p->uid) ||
451 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
452 !security_file_send_sigiotask(p, fown, sig));
455 static void send_sigio_to_task(struct task_struct *p,
456 struct fown_struct *fown,
460 if (!sigio_perm(p, fown, fown->signum))
463 switch (fown->signum) {
466 /* Queue a rt signal with the appropriate fd as its
467 value. We use SI_SIGIO as the source, not
468 SI_KERNEL, since kernel signals always get
469 delivered even if we can't queue. Failure to
470 queue in this case _should_ be reported; we fall
471 back to SIGIO in that case. --sct */
472 si.si_signo = fown->signum;
475 /* Make sure we are called with one of the POLL_*
476 reasons, otherwise we could leak kernel stack into
478 BUG_ON((reason & __SI_MASK) != __SI_POLL);
479 if (reason - POLL_IN >= NSIGPOLL)
482 si.si_band = band_table[reason - POLL_IN];
484 if (!group_send_sig_info(fown->signum, &si, p))
486 /* fall-through: fall back on the old plain SIGIO signal */
488 group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
492 void send_sigio(struct fown_struct *fown, int fd, int band)
494 struct task_struct *p;
498 read_lock(&fown->lock);
499 type = fown->pid_type;
502 goto out_unlock_fown;
504 read_lock(&tasklist_lock);
505 do_each_pid_task(pid, type, p) {
506 send_sigio_to_task(p, fown, fd, band);
507 } while_each_pid_task(pid, type, p);
508 read_unlock(&tasklist_lock);
510 read_unlock(&fown->lock);
513 static void send_sigurg_to_task(struct task_struct *p,
514 struct fown_struct *fown)
516 if (sigio_perm(p, fown, SIGURG))
517 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
520 int send_sigurg(struct fown_struct *fown)
522 struct task_struct *p;
527 read_lock(&fown->lock);
528 type = fown->pid_type;
531 goto out_unlock_fown;
535 read_lock(&tasklist_lock);
536 do_each_pid_task(pid, type, p) {
537 send_sigurg_to_task(p, fown);
538 } while_each_pid_task(pid, type, p);
539 read_unlock(&tasklist_lock);
541 read_unlock(&fown->lock);
545 static DEFINE_RWLOCK(fasync_lock);
546 static struct kmem_cache *fasync_cache __read_mostly;
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, GFP_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);
636 module_init(fasync_init)