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>
22 #include <linux/smp_lock.h>
25 #include <asm/siginfo.h>
26 #include <asm/uaccess.h>
28 void set_close_on_exec(unsigned int fd, int flag)
30 struct files_struct *files = current->files;
32 spin_lock(&files->file_lock);
33 fdt = files_fdtable(files);
35 FD_SET(fd, fdt->close_on_exec);
37 FD_CLR(fd, fdt->close_on_exec);
38 spin_unlock(&files->file_lock);
41 static int get_close_on_exec(unsigned int fd)
43 struct files_struct *files = current->files;
47 fdt = files_fdtable(files);
48 res = FD_ISSET(fd, fdt->close_on_exec);
53 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
56 struct file * file, *tofree;
57 struct files_struct * files = current->files;
60 if ((flags & ~O_CLOEXEC) != 0)
63 if (unlikely(oldfd == newfd))
66 spin_lock(&files->file_lock);
67 err = expand_files(files, newfd);
71 if (unlikely(err < 0)) {
77 * We need to detect attempts to do dup2() over allocated but still
78 * not finished descriptor. NB: OpenBSD avoids that at the price of
79 * extra work in their equivalent of fget() - they insert struct
80 * file immediately after grabbing descriptor, mark it larval if
81 * more work (e.g. actual opening) is needed and make sure that
82 * fget() treats larval files as absent. Potentially interesting,
83 * but while extra work in fget() is trivial, locking implications
84 * and amount of surgery on open()-related paths in VFS are not.
85 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
86 * deadlocks in rather amusing ways, AFAICS. All of that is out of
87 * scope of POSIX or SUS, since neither considers shared descriptor
88 * tables and this condition does not arise without those.
91 fdt = files_fdtable(files);
92 tofree = fdt->fd[newfd];
93 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
96 rcu_assign_pointer(fdt->fd[newfd], file);
97 FD_SET(newfd, fdt->open_fds);
98 if (flags & O_CLOEXEC)
99 FD_SET(newfd, fdt->close_on_exec);
101 FD_CLR(newfd, fdt->close_on_exec);
102 spin_unlock(&files->file_lock);
105 filp_close(tofree, files);
112 spin_unlock(&files->file_lock);
116 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
118 if (unlikely(newfd == oldfd)) { /* corner case */
119 struct files_struct *files = current->files;
123 if (!fcheck_files(files, oldfd))
128 return sys_dup3(oldfd, newfd, 0);
131 SYSCALL_DEFINE1(dup, unsigned int, fildes)
134 struct file *file = fget(fildes);
137 ret = get_unused_fd();
139 fd_install(ret, file);
146 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
148 static int setfl(int fd, struct file * filp, unsigned long arg)
150 struct inode * inode = filp->f_path.dentry->d_inode;
154 * O_APPEND cannot be cleared if the file is marked as append-only
155 * and the file is open for write.
157 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
160 /* O_NOATIME can only be set by the owner or superuser */
161 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
162 if (!is_owner_or_cap(inode))
165 /* required for strict SunOS emulation */
166 if (O_NONBLOCK != O_NDELAY)
170 if (arg & O_DIRECT) {
171 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
172 !filp->f_mapping->a_ops->direct_IO)
176 if (filp->f_op && filp->f_op->check_flags)
177 error = filp->f_op->check_flags(arg);
182 * ->fasync() is responsible for setting the FASYNC bit.
184 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
185 filp->f_op->fasync) {
186 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
192 spin_lock(&filp->f_lock);
193 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
194 spin_unlock(&filp->f_lock);
200 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
201 uid_t uid, uid_t euid, int force)
203 write_lock_irq(&filp->f_owner.lock);
204 if (force || !filp->f_owner.pid) {
205 put_pid(filp->f_owner.pid);
206 filp->f_owner.pid = get_pid(pid);
207 filp->f_owner.pid_type = type;
208 filp->f_owner.uid = uid;
209 filp->f_owner.euid = euid;
211 write_unlock_irq(&filp->f_owner.lock);
214 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
217 const struct cred *cred = current_cred();
220 err = security_file_set_fowner(filp);
224 f_modown(filp, pid, type, cred->uid, cred->euid, force);
227 EXPORT_SYMBOL(__f_setown);
229 int f_setown(struct file *filp, unsigned long arg, int force)
241 pid = find_vpid(who);
242 result = __f_setown(filp, pid, type, force);
246 EXPORT_SYMBOL(f_setown);
248 void f_delown(struct file *filp)
250 f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
253 pid_t f_getown(struct file *filp)
256 read_lock(&filp->f_owner.lock);
257 pid = pid_vnr(filp->f_owner.pid);
258 if (filp->f_owner.pid_type == PIDTYPE_PGID)
260 read_unlock(&filp->f_owner.lock);
264 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
271 case F_DUPFD_CLOEXEC:
272 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
274 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
277 fd_install(err, filp);
281 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
285 set_close_on_exec(fd, arg & FD_CLOEXEC);
291 err = setfl(fd, filp, arg);
294 err = fcntl_getlk(filp, (struct flock __user *) arg);
298 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
302 * XXX If f_owner is a process group, the
303 * negative return value will get converted
304 * into an error. Oops. If we keep the
305 * current syscall conventions, the only way
306 * to fix this will be in libc.
308 err = f_getown(filp);
309 force_successful_syscall_return();
312 err = f_setown(filp, arg, 1);
315 err = filp->f_owner.signum;
318 /* arg == 0 restores default behaviour. */
319 if (!valid_signal(arg)) {
323 filp->f_owner.signum = arg;
326 err = fcntl_getlease(filp);
329 err = fcntl_setlease(fd, filp, arg);
332 err = fcntl_dirnotify(fd, filp, arg);
340 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
349 err = security_file_fcntl(filp, cmd, arg);
355 err = do_fcntl(fd, cmd, arg, filp);
362 #if BITS_PER_LONG == 32
363 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
374 err = security_file_fcntl(filp, cmd, arg);
383 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
387 err = fcntl_setlk64(fd, filp, cmd,
388 (struct flock64 __user *) arg);
391 err = do_fcntl(fd, cmd, arg, filp);
400 /* Table to convert sigio signal codes into poll band bitmaps */
402 static const long band_table[NSIGPOLL] = {
403 POLLIN | POLLRDNORM, /* POLL_IN */
404 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
405 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
406 POLLERR, /* POLL_ERR */
407 POLLPRI | POLLRDBAND, /* POLL_PRI */
408 POLLHUP | POLLERR /* POLL_HUP */
411 static inline int sigio_perm(struct task_struct *p,
412 struct fown_struct *fown, int sig)
414 const struct cred *cred;
418 cred = __task_cred(p);
419 ret = ((fown->euid == 0 ||
420 fown->euid == cred->suid || fown->euid == cred->uid ||
421 fown->uid == cred->suid || fown->uid == cred->uid) &&
422 !security_file_send_sigiotask(p, fown, sig));
427 static void send_sigio_to_task(struct task_struct *p,
428 struct fown_struct *fown,
432 if (!sigio_perm(p, fown, fown->signum))
435 switch (fown->signum) {
438 /* Queue a rt signal with the appropriate fd as its
439 value. We use SI_SIGIO as the source, not
440 SI_KERNEL, since kernel signals always get
441 delivered even if we can't queue. Failure to
442 queue in this case _should_ be reported; we fall
443 back to SIGIO in that case. --sct */
444 si.si_signo = fown->signum;
447 /* Make sure we are called with one of the POLL_*
448 reasons, otherwise we could leak kernel stack into
450 BUG_ON((reason & __SI_MASK) != __SI_POLL);
451 if (reason - POLL_IN >= NSIGPOLL)
454 si.si_band = band_table[reason - POLL_IN];
456 if (!group_send_sig_info(fown->signum, &si, p))
458 /* fall-through: fall back on the old plain SIGIO signal */
460 group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
464 void send_sigio(struct fown_struct *fown, int fd, int band)
466 struct task_struct *p;
470 read_lock(&fown->lock);
471 type = fown->pid_type;
474 goto out_unlock_fown;
476 read_lock(&tasklist_lock);
477 do_each_pid_task(pid, type, p) {
478 send_sigio_to_task(p, fown, fd, band);
479 } while_each_pid_task(pid, type, p);
480 read_unlock(&tasklist_lock);
482 read_unlock(&fown->lock);
485 static void send_sigurg_to_task(struct task_struct *p,
486 struct fown_struct *fown)
488 if (sigio_perm(p, fown, SIGURG))
489 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
492 int send_sigurg(struct fown_struct *fown)
494 struct task_struct *p;
499 read_lock(&fown->lock);
500 type = fown->pid_type;
503 goto out_unlock_fown;
507 read_lock(&tasklist_lock);
508 do_each_pid_task(pid, type, p) {
509 send_sigurg_to_task(p, fown);
510 } while_each_pid_task(pid, type, p);
511 read_unlock(&tasklist_lock);
513 read_unlock(&fown->lock);
517 static DEFINE_RWLOCK(fasync_lock);
518 static struct kmem_cache *fasync_cache __read_mostly;
521 * fasync_helper() is used by almost all character device drivers
522 * to set up the fasync queue. It returns negative on error, 0 if it did
523 * no changes and positive if it added/deleted the entry.
525 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
527 struct fasync_struct *fa, **fp;
528 struct fasync_struct *new = NULL;
532 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
538 * We need to take f_lock first since it's not an IRQ-safe
541 spin_lock(&filp->f_lock);
542 write_lock_irq(&fasync_lock);
543 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
544 if (fa->fa_file == filp) {
547 kmem_cache_free(fasync_cache, new);
550 kmem_cache_free(fasync_cache, fa);
558 new->magic = FASYNC_MAGIC;
561 new->fa_next = *fapp;
567 filp->f_flags |= FASYNC;
569 filp->f_flags &= ~FASYNC;
570 write_unlock_irq(&fasync_lock);
571 spin_unlock(&filp->f_lock);
575 EXPORT_SYMBOL(fasync_helper);
577 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
580 struct fown_struct * fown;
581 if (fa->magic != FASYNC_MAGIC) {
582 printk(KERN_ERR "kill_fasync: bad magic number in "
586 fown = &fa->fa_file->f_owner;
587 /* Don't send SIGURG to processes which have not set a
588 queued signum: SIGURG has its own default signalling
590 if (!(sig == SIGURG && fown->signum == 0))
591 send_sigio(fown, fa->fa_fd, band);
596 EXPORT_SYMBOL(__kill_fasync);
598 void kill_fasync(struct fasync_struct **fp, int sig, int band)
600 /* First a quick test without locking: usually
604 read_lock(&fasync_lock);
605 /* reread *fp after obtaining the lock */
606 __kill_fasync(*fp, sig, band);
607 read_unlock(&fasync_lock);
610 EXPORT_SYMBOL(kill_fasync);
612 static int __init fasync_init(void)
614 fasync_cache = kmem_cache_create("fasync_cache",
615 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
619 module_init(fasync_init)