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 SYSCALL_DEFINE3(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 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
117 if (unlikely(newfd == oldfd)) { /* corner case */
118 struct files_struct *files = current->files;
122 if (!fcheck_files(files, oldfd))
127 return sys_dup3(oldfd, newfd, 0);
130 SYSCALL_DEFINE1(dup, unsigned int, fildes)
133 struct file *file = fget(fildes);
136 ret = get_unused_fd();
138 fd_install(ret, file);
145 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
147 static int setfl(int fd, struct file * filp, unsigned long arg)
149 struct inode * inode = filp->f_path.dentry->d_inode;
153 * O_APPEND cannot be cleared if the file is marked as append-only
154 * and the file is open for write.
156 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
159 /* O_NOATIME can only be set by the owner or superuser */
160 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
161 if (!is_owner_or_cap(inode))
164 /* required for strict SunOS emulation */
165 if (O_NONBLOCK != O_NDELAY)
169 if (arg & O_DIRECT) {
170 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
171 !filp->f_mapping->a_ops->direct_IO)
175 if (filp->f_op && filp->f_op->check_flags)
176 error = filp->f_op->check_flags(arg);
181 * ->fasync() is responsible for setting the FASYNC bit.
183 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
184 filp->f_op->fasync) {
185 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
191 spin_lock(&filp->f_lock);
192 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
193 spin_unlock(&filp->f_lock);
199 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
202 write_lock_irq(&filp->f_owner.lock);
203 if (force || !filp->f_owner.pid) {
204 put_pid(filp->f_owner.pid);
205 filp->f_owner.pid = get_pid(pid);
206 filp->f_owner.pid_type = type;
209 const struct cred *cred = current_cred();
210 filp->f_owner.uid = cred->uid;
211 filp->f_owner.euid = cred->euid;
214 write_unlock_irq(&filp->f_owner.lock);
217 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
222 err = security_file_set_fowner(filp);
226 f_modown(filp, pid, type, force);
229 EXPORT_SYMBOL(__f_setown);
231 int f_setown(struct file *filp, unsigned long arg, int force)
243 pid = find_vpid(who);
244 result = __f_setown(filp, pid, type, force);
248 EXPORT_SYMBOL(f_setown);
250 void f_delown(struct file *filp)
252 f_modown(filp, NULL, PIDTYPE_PID, 1);
255 pid_t f_getown(struct file *filp)
258 read_lock(&filp->f_owner.lock);
259 pid = pid_vnr(filp->f_owner.pid);
260 if (filp->f_owner.pid_type == PIDTYPE_PGID)
262 read_unlock(&filp->f_owner.lock);
266 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
273 case F_DUPFD_CLOEXEC:
274 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
276 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
279 fd_install(err, filp);
283 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
287 set_close_on_exec(fd, arg & FD_CLOEXEC);
293 err = setfl(fd, filp, arg);
296 err = fcntl_getlk(filp, (struct flock __user *) arg);
300 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
304 * XXX If f_owner is a process group, the
305 * negative return value will get converted
306 * into an error. Oops. If we keep the
307 * current syscall conventions, the only way
308 * to fix this will be in libc.
310 err = f_getown(filp);
311 force_successful_syscall_return();
314 err = f_setown(filp, arg, 1);
317 err = filp->f_owner.signum;
320 /* arg == 0 restores default behaviour. */
321 if (!valid_signal(arg)) {
325 filp->f_owner.signum = arg;
328 err = fcntl_getlease(filp);
331 err = fcntl_setlease(fd, filp, arg);
334 err = fcntl_dirnotify(fd, filp, arg);
342 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
351 err = security_file_fcntl(filp, cmd, arg);
357 err = do_fcntl(fd, cmd, arg, filp);
364 #if BITS_PER_LONG == 32
365 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
376 err = security_file_fcntl(filp, cmd, arg);
385 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
389 err = fcntl_setlk64(fd, filp, cmd,
390 (struct flock64 __user *) arg);
393 err = do_fcntl(fd, cmd, arg, filp);
402 /* Table to convert sigio signal codes into poll band bitmaps */
404 static const long band_table[NSIGPOLL] = {
405 POLLIN | POLLRDNORM, /* POLL_IN */
406 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
407 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
408 POLLERR, /* POLL_ERR */
409 POLLPRI | POLLRDBAND, /* POLL_PRI */
410 POLLHUP | POLLERR /* POLL_HUP */
413 static inline int sigio_perm(struct task_struct *p,
414 struct fown_struct *fown, int sig)
416 const struct cred *cred;
420 cred = __task_cred(p);
421 ret = ((fown->euid == 0 ||
422 fown->euid == cred->suid || fown->euid == cred->uid ||
423 fown->uid == cred->suid || fown->uid == cred->uid) &&
424 !security_file_send_sigiotask(p, fown, sig));
429 static void send_sigio_to_task(struct task_struct *p,
430 struct fown_struct *fown,
435 * F_SETSIG can change ->signum lockless in parallel, make
436 * sure we read it once and use the same value throughout.
438 int signum = ACCESS_ONCE(fown->signum);
440 if (!sigio_perm(p, fown, signum))
446 /* Queue a rt signal with the appropriate fd as its
447 value. We use SI_SIGIO as the source, not
448 SI_KERNEL, since kernel signals always get
449 delivered even if we can't queue. Failure to
450 queue in this case _should_ be reported; we fall
451 back to SIGIO in that case. --sct */
452 si.si_signo = signum;
455 /* Make sure we are called with one of the POLL_*
456 reasons, otherwise we could leak kernel stack into
458 BUG_ON((reason & __SI_MASK) != __SI_POLL);
459 if (reason - POLL_IN >= NSIGPOLL)
462 si.si_band = band_table[reason - POLL_IN];
464 if (!group_send_sig_info(signum, &si, p))
466 /* fall-through: fall back on the old plain SIGIO signal */
468 group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
472 void send_sigio(struct fown_struct *fown, int fd, int band)
474 struct task_struct *p;
478 read_lock(&fown->lock);
479 type = fown->pid_type;
482 goto out_unlock_fown;
484 read_lock(&tasklist_lock);
485 do_each_pid_task(pid, type, p) {
486 send_sigio_to_task(p, fown, fd, band);
487 } while_each_pid_task(pid, type, p);
488 read_unlock(&tasklist_lock);
490 read_unlock(&fown->lock);
493 static void send_sigurg_to_task(struct task_struct *p,
494 struct fown_struct *fown)
496 if (sigio_perm(p, fown, SIGURG))
497 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
500 int send_sigurg(struct fown_struct *fown)
502 struct task_struct *p;
507 read_lock(&fown->lock);
508 type = fown->pid_type;
511 goto out_unlock_fown;
515 read_lock(&tasklist_lock);
516 do_each_pid_task(pid, type, p) {
517 send_sigurg_to_task(p, fown);
518 } while_each_pid_task(pid, type, p);
519 read_unlock(&tasklist_lock);
521 read_unlock(&fown->lock);
525 static DEFINE_RWLOCK(fasync_lock);
526 static struct kmem_cache *fasync_cache __read_mostly;
529 * fasync_helper() is used by almost all character device drivers
530 * to set up the fasync queue. It returns negative on error, 0 if it did
531 * no changes and positive if it added/deleted the entry.
533 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
535 struct fasync_struct *fa, **fp;
536 struct fasync_struct *new = NULL;
540 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
546 * We need to take f_lock first since it's not an IRQ-safe
549 spin_lock(&filp->f_lock);
550 write_lock_irq(&fasync_lock);
551 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
552 if (fa->fa_file == filp) {
555 kmem_cache_free(fasync_cache, new);
558 kmem_cache_free(fasync_cache, fa);
566 new->magic = FASYNC_MAGIC;
569 new->fa_next = *fapp;
575 filp->f_flags |= FASYNC;
577 filp->f_flags &= ~FASYNC;
578 write_unlock_irq(&fasync_lock);
579 spin_unlock(&filp->f_lock);
583 EXPORT_SYMBOL(fasync_helper);
585 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
588 struct fown_struct * fown;
589 if (fa->magic != FASYNC_MAGIC) {
590 printk(KERN_ERR "kill_fasync: bad magic number in "
594 fown = &fa->fa_file->f_owner;
595 /* Don't send SIGURG to processes which have not set a
596 queued signum: SIGURG has its own default signalling
598 if (!(sig == SIGURG && fown->signum == 0))
599 send_sigio(fown, fa->fa_fd, band);
604 EXPORT_SYMBOL(__kill_fasync);
606 void kill_fasync(struct fasync_struct **fp, int sig, int band)
608 /* First a quick test without locking: usually
612 read_lock(&fasync_lock);
613 /* reread *fp after obtaining the lock */
614 __kill_fasync(*fp, sig, band);
615 read_unlock(&fasync_lock);
618 EXPORT_SYMBOL(kill_fasync);
620 static int __init fasync_init(void)
622 fasync_cache = kmem_cache_create("fasync_cache",
623 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
627 module_init(fasync_init)