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);
52 asmlinkage long sys_dup3(unsigned int oldfd, unsigned int newfd, int flags)
55 struct file * file, *tofree;
56 struct files_struct * files = current->files;
59 if ((flags & ~O_CLOEXEC) != 0)
62 if (unlikely(oldfd == newfd))
65 spin_lock(&files->file_lock);
66 if (!(file = fcheck(oldfd)))
68 get_file(file); /* We are now finished with oldfd */
70 err = expand_files(files, newfd);
71 if (unlikely(err < 0)) {
77 /* To avoid races with open() and dup(), we will mark the fd as
78 * in-use in the open-file bitmap throughout the entire dup2()
79 * process. This is quite safe: do_close() uses the fd array
80 * entry, not the bitmap, to decide what work needs to be
82 /* Doesn't work. open() might be there first. --AV */
84 /* Yes. It's a race. In user space. Nothing sane to do */
86 fdt = files_fdtable(files);
87 tofree = fdt->fd[newfd];
88 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
91 rcu_assign_pointer(fdt->fd[newfd], file);
92 FD_SET(newfd, fdt->open_fds);
93 if (flags & O_CLOEXEC)
94 FD_SET(newfd, fdt->close_on_exec);
96 FD_CLR(newfd, fdt->close_on_exec);
97 spin_unlock(&files->file_lock);
100 filp_close(tofree, files);
105 spin_unlock(&files->file_lock);
109 spin_unlock(&files->file_lock);
114 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
116 if (unlikely(newfd == oldfd)) { /* corner case */
117 struct files_struct *files = current->files;
119 if (!fcheck_files(files, oldfd))
124 return sys_dup3(oldfd, newfd, 0);
127 asmlinkage long sys_dup(unsigned int fildes)
130 struct file *file = fget(fildes);
133 ret = get_unused_fd();
135 fd_install(ret, file);
142 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
144 static int setfl(int fd, struct file * filp, unsigned long arg)
146 struct inode * inode = filp->f_path.dentry->d_inode;
150 * O_APPEND cannot be cleared if the file is marked as append-only
151 * and the file is open for write.
153 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
156 /* O_NOATIME can only be set by the owner or superuser */
157 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
158 if (!is_owner_or_cap(inode))
161 /* required for strict SunOS emulation */
162 if (O_NONBLOCK != O_NDELAY)
166 if (arg & O_DIRECT) {
167 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
168 !filp->f_mapping->a_ops->direct_IO)
172 if (filp->f_op && filp->f_op->check_flags)
173 error = filp->f_op->check_flags(arg);
177 if ((arg ^ filp->f_flags) & FASYNC) {
178 if (filp->f_op && filp->f_op->fasync) {
179 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
185 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
190 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
191 uid_t uid, uid_t euid, int force)
193 write_lock_irq(&filp->f_owner.lock);
194 if (force || !filp->f_owner.pid) {
195 put_pid(filp->f_owner.pid);
196 filp->f_owner.pid = get_pid(pid);
197 filp->f_owner.pid_type = type;
198 filp->f_owner.uid = uid;
199 filp->f_owner.euid = euid;
201 write_unlock_irq(&filp->f_owner.lock);
204 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
209 err = security_file_set_fowner(filp);
213 f_modown(filp, pid, type, current->uid, current->euid, force);
216 EXPORT_SYMBOL(__f_setown);
218 int f_setown(struct file *filp, unsigned long arg, int force)
230 pid = find_vpid(who);
231 result = __f_setown(filp, pid, type, force);
235 EXPORT_SYMBOL(f_setown);
237 void f_delown(struct file *filp)
239 f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
242 pid_t f_getown(struct file *filp)
245 read_lock(&filp->f_owner.lock);
246 pid = pid_vnr(filp->f_owner.pid);
247 if (filp->f_owner.pid_type == PIDTYPE_PGID)
249 read_unlock(&filp->f_owner.lock);
253 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
260 case F_DUPFD_CLOEXEC:
261 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
263 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
266 fd_install(err, filp);
270 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
274 set_close_on_exec(fd, arg & FD_CLOEXEC);
280 err = setfl(fd, filp, arg);
283 err = fcntl_getlk(filp, (struct flock __user *) arg);
287 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
291 * XXX If f_owner is a process group, the
292 * negative return value will get converted
293 * into an error. Oops. If we keep the
294 * current syscall conventions, the only way
295 * to fix this will be in libc.
297 err = f_getown(filp);
298 force_successful_syscall_return();
301 err = f_setown(filp, arg, 1);
304 err = filp->f_owner.signum;
307 /* arg == 0 restores default behaviour. */
308 if (!valid_signal(arg)) {
312 filp->f_owner.signum = arg;
315 err = fcntl_getlease(filp);
318 err = fcntl_setlease(fd, filp, arg);
321 err = fcntl_dirnotify(fd, filp, arg);
329 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
338 err = security_file_fcntl(filp, cmd, arg);
344 err = do_fcntl(fd, cmd, arg, filp);
351 #if BITS_PER_LONG == 32
352 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
362 err = security_file_fcntl(filp, cmd, arg);
371 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
375 err = fcntl_setlk64(fd, filp, cmd,
376 (struct flock64 __user *) arg);
379 err = do_fcntl(fd, cmd, arg, filp);
388 /* Table to convert sigio signal codes into poll band bitmaps */
390 static const long band_table[NSIGPOLL] = {
391 POLLIN | POLLRDNORM, /* POLL_IN */
392 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
393 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
394 POLLERR, /* POLL_ERR */
395 POLLPRI | POLLRDBAND, /* POLL_PRI */
396 POLLHUP | POLLERR /* POLL_HUP */
399 static inline int sigio_perm(struct task_struct *p,
400 struct fown_struct *fown, int sig)
402 return (((fown->euid == 0) ||
403 (fown->euid == p->suid) || (fown->euid == p->uid) ||
404 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
405 !security_file_send_sigiotask(p, fown, sig));
408 static void send_sigio_to_task(struct task_struct *p,
409 struct fown_struct *fown,
413 if (!sigio_perm(p, fown, fown->signum))
416 switch (fown->signum) {
419 /* Queue a rt signal with the appropriate fd as its
420 value. We use SI_SIGIO as the source, not
421 SI_KERNEL, since kernel signals always get
422 delivered even if we can't queue. Failure to
423 queue in this case _should_ be reported; we fall
424 back to SIGIO in that case. --sct */
425 si.si_signo = fown->signum;
428 /* Make sure we are called with one of the POLL_*
429 reasons, otherwise we could leak kernel stack into
431 BUG_ON((reason & __SI_MASK) != __SI_POLL);
432 if (reason - POLL_IN >= NSIGPOLL)
435 si.si_band = band_table[reason - POLL_IN];
437 if (!group_send_sig_info(fown->signum, &si, p))
439 /* fall-through: fall back on the old plain SIGIO signal */
441 group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
445 void send_sigio(struct fown_struct *fown, int fd, int band)
447 struct task_struct *p;
451 read_lock(&fown->lock);
452 type = fown->pid_type;
455 goto out_unlock_fown;
457 read_lock(&tasklist_lock);
458 do_each_pid_task(pid, type, p) {
459 send_sigio_to_task(p, fown, fd, band);
460 } while_each_pid_task(pid, type, p);
461 read_unlock(&tasklist_lock);
463 read_unlock(&fown->lock);
466 static void send_sigurg_to_task(struct task_struct *p,
467 struct fown_struct *fown)
469 if (sigio_perm(p, fown, SIGURG))
470 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
473 int send_sigurg(struct fown_struct *fown)
475 struct task_struct *p;
480 read_lock(&fown->lock);
481 type = fown->pid_type;
484 goto out_unlock_fown;
488 read_lock(&tasklist_lock);
489 do_each_pid_task(pid, type, p) {
490 send_sigurg_to_task(p, fown);
491 } while_each_pid_task(pid, type, p);
492 read_unlock(&tasklist_lock);
494 read_unlock(&fown->lock);
498 static DEFINE_RWLOCK(fasync_lock);
499 static struct kmem_cache *fasync_cache __read_mostly;
502 * fasync_helper() is used by some character device drivers (mainly mice)
503 * to set up the fasync queue. It returns negative on error, 0 if it did
504 * no changes and positive if it added/deleted the entry.
506 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
508 struct fasync_struct *fa, **fp;
509 struct fasync_struct *new = NULL;
513 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
517 write_lock_irq(&fasync_lock);
518 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
519 if (fa->fa_file == filp) {
522 kmem_cache_free(fasync_cache, new);
525 kmem_cache_free(fasync_cache, fa);
533 new->magic = FASYNC_MAGIC;
536 new->fa_next = *fapp;
541 write_unlock_irq(&fasync_lock);
545 EXPORT_SYMBOL(fasync_helper);
547 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
550 struct fown_struct * fown;
551 if (fa->magic != FASYNC_MAGIC) {
552 printk(KERN_ERR "kill_fasync: bad magic number in "
556 fown = &fa->fa_file->f_owner;
557 /* Don't send SIGURG to processes which have not set a
558 queued signum: SIGURG has its own default signalling
560 if (!(sig == SIGURG && fown->signum == 0))
561 send_sigio(fown, fa->fa_fd, band);
566 EXPORT_SYMBOL(__kill_fasync);
568 void kill_fasync(struct fasync_struct **fp, int sig, int band)
570 /* First a quick test without locking: usually
574 read_lock(&fasync_lock);
575 /* reread *fp after obtaining the lock */
576 __kill_fasync(*fp, sig, band);
577 read_unlock(&fasync_lock);
580 EXPORT_SYMBOL(kill_fasync);
582 static int __init fasync_init(void)
584 fasync_cache = kmem_cache_create("fasync_cache",
585 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
589 module_init(fasync_init)