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 err = expand_files(files, newfd);
70 if (unlikely(err < 0)) {
76 * We need to detect attempts to do dup2() over allocated but still
77 * not finished descriptor. NB: OpenBSD avoids that at the price of
78 * extra work in their equivalent of fget() - they insert struct
79 * file immediately after grabbing descriptor, mark it larval if
80 * more work (e.g. actual opening) is needed and make sure that
81 * fget() treats larval files as absent. Potentially interesting,
82 * but while extra work in fget() is trivial, locking implications
83 * and amount of surgery on open()-related paths in VFS are not.
84 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
85 * deadlocks in rather amusing ways, AFAICS. All of that is out of
86 * scope of POSIX or SUS, since neither considers shared descriptor
87 * tables and this condition does not arise without those.
90 fdt = files_fdtable(files);
91 tofree = fdt->fd[newfd];
92 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
95 rcu_assign_pointer(fdt->fd[newfd], file);
96 FD_SET(newfd, fdt->open_fds);
97 if (flags & O_CLOEXEC)
98 FD_SET(newfd, fdt->close_on_exec);
100 FD_CLR(newfd, fdt->close_on_exec);
101 spin_unlock(&files->file_lock);
104 filp_close(tofree, files);
111 spin_unlock(&files->file_lock);
115 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
117 if (unlikely(newfd == oldfd)) { /* corner case */
118 struct files_struct *files = current->files;
120 if (!fcheck_files(files, oldfd))
125 return sys_dup3(oldfd, newfd, 0);
128 asmlinkage long sys_dup(unsigned int fildes)
131 struct file *file = fget(fildes);
134 ret = get_unused_fd();
136 fd_install(ret, file);
143 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
145 static int setfl(int fd, struct file * filp, unsigned long arg)
147 struct inode * inode = filp->f_path.dentry->d_inode;
151 * O_APPEND cannot be cleared if the file is marked as append-only
152 * and the file is open for write.
154 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
157 /* O_NOATIME can only be set by the owner or superuser */
158 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
159 if (!is_owner_or_cap(inode))
162 /* required for strict SunOS emulation */
163 if (O_NONBLOCK != O_NDELAY)
167 if (arg & O_DIRECT) {
168 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
169 !filp->f_mapping->a_ops->direct_IO)
173 if (filp->f_op && filp->f_op->check_flags)
174 error = filp->f_op->check_flags(arg);
178 if ((arg ^ filp->f_flags) & FASYNC) {
179 if (filp->f_op && filp->f_op->fasync) {
180 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
186 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
191 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
192 uid_t uid, uid_t euid, int force)
194 write_lock_irq(&filp->f_owner.lock);
195 if (force || !filp->f_owner.pid) {
196 put_pid(filp->f_owner.pid);
197 filp->f_owner.pid = get_pid(pid);
198 filp->f_owner.pid_type = type;
199 filp->f_owner.uid = uid;
200 filp->f_owner.euid = euid;
202 write_unlock_irq(&filp->f_owner.lock);
205 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
210 err = security_file_set_fowner(filp);
214 f_modown(filp, pid, type, current->uid, current->euid, force);
217 EXPORT_SYMBOL(__f_setown);
219 int f_setown(struct file *filp, unsigned long arg, int force)
231 pid = find_vpid(who);
232 result = __f_setown(filp, pid, type, force);
236 EXPORT_SYMBOL(f_setown);
238 void f_delown(struct file *filp)
240 f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
243 pid_t f_getown(struct file *filp)
246 read_lock(&filp->f_owner.lock);
247 pid = pid_vnr(filp->f_owner.pid);
248 if (filp->f_owner.pid_type == PIDTYPE_PGID)
250 read_unlock(&filp->f_owner.lock);
254 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
261 case F_DUPFD_CLOEXEC:
262 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
264 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
267 fd_install(err, filp);
271 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
275 set_close_on_exec(fd, arg & FD_CLOEXEC);
281 err = setfl(fd, filp, arg);
284 err = fcntl_getlk(filp, (struct flock __user *) arg);
288 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
292 * XXX If f_owner is a process group, the
293 * negative return value will get converted
294 * into an error. Oops. If we keep the
295 * current syscall conventions, the only way
296 * to fix this will be in libc.
298 err = f_getown(filp);
299 force_successful_syscall_return();
302 err = f_setown(filp, arg, 1);
305 err = filp->f_owner.signum;
308 /* arg == 0 restores default behaviour. */
309 if (!valid_signal(arg)) {
313 filp->f_owner.signum = arg;
316 err = fcntl_getlease(filp);
319 err = fcntl_setlease(fd, filp, arg);
322 err = fcntl_dirnotify(fd, filp, arg);
330 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
339 err = security_file_fcntl(filp, cmd, arg);
345 err = do_fcntl(fd, cmd, arg, filp);
352 #if BITS_PER_LONG == 32
353 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
363 err = security_file_fcntl(filp, cmd, arg);
372 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
376 err = fcntl_setlk64(fd, filp, cmd,
377 (struct flock64 __user *) arg);
380 err = do_fcntl(fd, cmd, arg, filp);
389 /* Table to convert sigio signal codes into poll band bitmaps */
391 static const long band_table[NSIGPOLL] = {
392 POLLIN | POLLRDNORM, /* POLL_IN */
393 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
394 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
395 POLLERR, /* POLL_ERR */
396 POLLPRI | POLLRDBAND, /* POLL_PRI */
397 POLLHUP | POLLERR /* POLL_HUP */
400 static inline int sigio_perm(struct task_struct *p,
401 struct fown_struct *fown, int sig)
403 return (((fown->euid == 0) ||
404 (fown->euid == p->suid) || (fown->euid == p->uid) ||
405 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
406 !security_file_send_sigiotask(p, fown, sig));
409 static void send_sigio_to_task(struct task_struct *p,
410 struct fown_struct *fown,
414 if (!sigio_perm(p, fown, fown->signum))
417 switch (fown->signum) {
420 /* Queue a rt signal with the appropriate fd as its
421 value. We use SI_SIGIO as the source, not
422 SI_KERNEL, since kernel signals always get
423 delivered even if we can't queue. Failure to
424 queue in this case _should_ be reported; we fall
425 back to SIGIO in that case. --sct */
426 si.si_signo = fown->signum;
429 /* Make sure we are called with one of the POLL_*
430 reasons, otherwise we could leak kernel stack into
432 BUG_ON((reason & __SI_MASK) != __SI_POLL);
433 if (reason - POLL_IN >= NSIGPOLL)
436 si.si_band = band_table[reason - POLL_IN];
438 if (!group_send_sig_info(fown->signum, &si, p))
440 /* fall-through: fall back on the old plain SIGIO signal */
442 group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
446 void send_sigio(struct fown_struct *fown, int fd, int band)
448 struct task_struct *p;
452 read_lock(&fown->lock);
453 type = fown->pid_type;
456 goto out_unlock_fown;
458 read_lock(&tasklist_lock);
459 do_each_pid_task(pid, type, p) {
460 send_sigio_to_task(p, fown, fd, band);
461 } while_each_pid_task(pid, type, p);
462 read_unlock(&tasklist_lock);
464 read_unlock(&fown->lock);
467 static void send_sigurg_to_task(struct task_struct *p,
468 struct fown_struct *fown)
470 if (sigio_perm(p, fown, SIGURG))
471 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
474 int send_sigurg(struct fown_struct *fown)
476 struct task_struct *p;
481 read_lock(&fown->lock);
482 type = fown->pid_type;
485 goto out_unlock_fown;
489 read_lock(&tasklist_lock);
490 do_each_pid_task(pid, type, p) {
491 send_sigurg_to_task(p, fown);
492 } while_each_pid_task(pid, type, p);
493 read_unlock(&tasklist_lock);
495 read_unlock(&fown->lock);
499 static DEFINE_RWLOCK(fasync_lock);
500 static struct kmem_cache *fasync_cache __read_mostly;
503 * fasync_helper() is used by some character device drivers (mainly mice)
504 * to set up the fasync queue. It returns negative on error, 0 if it did
505 * no changes and positive if it added/deleted the entry.
507 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
509 struct fasync_struct *fa, **fp;
510 struct fasync_struct *new = NULL;
514 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
518 write_lock_irq(&fasync_lock);
519 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
520 if (fa->fa_file == filp) {
523 kmem_cache_free(fasync_cache, new);
526 kmem_cache_free(fasync_cache, fa);
534 new->magic = FASYNC_MAGIC;
537 new->fa_next = *fapp;
542 write_unlock_irq(&fasync_lock);
546 EXPORT_SYMBOL(fasync_helper);
548 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
551 struct fown_struct * fown;
552 if (fa->magic != FASYNC_MAGIC) {
553 printk(KERN_ERR "kill_fasync: bad magic number in "
557 fown = &fa->fa_file->f_owner;
558 /* Don't send SIGURG to processes which have not set a
559 queued signum: SIGURG has its own default signalling
561 if (!(sig == SIGURG && fown->signum == 0))
562 send_sigio(fown, fa->fa_fd, band);
567 EXPORT_SYMBOL(__kill_fasync);
569 void kill_fasync(struct fasync_struct **fp, int sig, int band)
571 /* First a quick test without locking: usually
575 read_lock(&fasync_lock);
576 /* reread *fp after obtaining the lock */
577 __kill_fasync(*fp, sig, band);
578 read_unlock(&fasync_lock);
581 EXPORT_SYMBOL(kill_fasync);
583 static int __init fasync_init(void)
585 fasync_cache = kmem_cache_create("fasync_cache",
586 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
590 module_init(fasync_init)