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/capability.h>
13 #include <linux/dnotify.h>
14 #include <linux/smp_lock.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>
23 #include <asm/siginfo.h>
24 #include <asm/uaccess.h>
26 void fastcall set_close_on_exec(unsigned int fd, int flag)
28 struct files_struct *files = current->files;
30 spin_lock(&files->file_lock);
31 fdt = files_fdtable(files);
33 FD_SET(fd, fdt->close_on_exec);
35 FD_CLR(fd, fdt->close_on_exec);
36 spin_unlock(&files->file_lock);
39 static int get_close_on_exec(unsigned int fd)
41 struct files_struct *files = current->files;
45 fdt = files_fdtable(files);
46 res = FD_ISSET(fd, fdt->close_on_exec);
52 * locate_fd finds a free file descriptor in the open_fds fdset,
53 * expanding the fd arrays if necessary. Must be called with the
54 * file_lock held for write.
57 static int locate_fd(struct files_struct *files,
58 struct file *file, unsigned int orig_start)
66 if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
70 fdt = files_fdtable(files);
72 * Someone might have closed fd's in the range
73 * orig_start..fdt->next_fd
76 if (start < files->next_fd)
77 start = files->next_fd;
80 if (start < fdt->max_fds)
81 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
85 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
88 error = expand_files(files, newfd);
93 * If we needed to expand the fs array we
94 * might have blocked - try again.
100 * We reacquired files_lock, so we are safe as long as
101 * we reacquire the fdtable pointer and use it while holding
102 * the lock, no one can free it during that time.
104 if (start <= files->next_fd)
105 files->next_fd = newfd + 1;
113 static int dupfd(struct file *file, unsigned int start, int cloexec)
115 struct files_struct * files = current->files;
119 spin_lock(&files->file_lock);
120 fd = locate_fd(files, file, start);
122 /* locate_fd() may have expanded fdtable, load the ptr */
123 fdt = files_fdtable(files);
124 FD_SET(fd, fdt->open_fds);
126 FD_SET(fd, fdt->close_on_exec);
128 FD_CLR(fd, fdt->close_on_exec);
129 spin_unlock(&files->file_lock);
130 fd_install(fd, file);
132 spin_unlock(&files->file_lock);
139 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
142 struct file * file, *tofree;
143 struct files_struct * files = current->files;
146 spin_lock(&files->file_lock);
147 if (!(file = fcheck(oldfd)))
153 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
155 get_file(file); /* We are now finished with oldfd */
157 err = expand_files(files, newfd);
161 /* To avoid races with open() and dup(), we will mark the fd as
162 * in-use in the open-file bitmap throughout the entire dup2()
163 * process. This is quite safe: do_close() uses the fd array
164 * entry, not the bitmap, to decide what work needs to be
166 /* Doesn't work. open() might be there first. --AV */
168 /* Yes. It's a race. In user space. Nothing sane to do */
170 fdt = files_fdtable(files);
171 tofree = fdt->fd[newfd];
172 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
175 rcu_assign_pointer(fdt->fd[newfd], file);
176 FD_SET(newfd, fdt->open_fds);
177 FD_CLR(newfd, fdt->close_on_exec);
178 spin_unlock(&files->file_lock);
181 filp_close(tofree, files);
186 spin_unlock(&files->file_lock);
190 spin_unlock(&files->file_lock);
195 asmlinkage long sys_dup(unsigned int fildes)
198 struct file * file = fget(fildes);
201 ret = dupfd(file, 0, 0);
205 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
207 static int setfl(int fd, struct file * filp, unsigned long arg)
209 struct inode * inode = filp->f_path.dentry->d_inode;
213 * O_APPEND cannot be cleared if the file is marked as append-only
214 * and the file is open for write.
216 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
219 /* O_NOATIME can only be set by the owner or superuser */
220 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
221 if (!is_owner_or_cap(inode))
224 /* required for strict SunOS emulation */
225 if (O_NONBLOCK != O_NDELAY)
229 if (arg & O_DIRECT) {
230 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
231 !filp->f_mapping->a_ops->direct_IO)
235 if (filp->f_op && filp->f_op->check_flags)
236 error = filp->f_op->check_flags(arg);
241 if ((arg ^ filp->f_flags) & FASYNC) {
242 if (filp->f_op && filp->f_op->fasync) {
243 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
249 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
255 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
256 uid_t uid, uid_t euid, int force)
258 write_lock_irq(&filp->f_owner.lock);
259 if (force || !filp->f_owner.pid) {
260 put_pid(filp->f_owner.pid);
261 filp->f_owner.pid = get_pid(pid);
262 filp->f_owner.pid_type = type;
263 filp->f_owner.uid = uid;
264 filp->f_owner.euid = euid;
266 write_unlock_irq(&filp->f_owner.lock);
269 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
274 err = security_file_set_fowner(filp);
278 f_modown(filp, pid, type, current->uid, current->euid, force);
281 EXPORT_SYMBOL(__f_setown);
283 int f_setown(struct file *filp, unsigned long arg, int force)
296 result = __f_setown(filp, pid, type, force);
300 EXPORT_SYMBOL(f_setown);
302 void f_delown(struct file *filp)
304 f_modown(filp, NULL, PIDTYPE_PID, 0, 0, 1);
307 pid_t f_getown(struct file *filp)
310 read_lock(&filp->f_owner.lock);
311 pid = pid_nr(filp->f_owner.pid);
312 if (filp->f_owner.pid_type == PIDTYPE_PGID)
314 read_unlock(&filp->f_owner.lock);
318 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
325 case F_DUPFD_CLOEXEC:
327 err = dupfd(filp, arg, cmd == F_DUPFD_CLOEXEC);
330 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
334 set_close_on_exec(fd, arg & FD_CLOEXEC);
340 err = setfl(fd, filp, arg);
343 err = fcntl_getlk(filp, (struct flock __user *) arg);
347 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
351 * XXX If f_owner is a process group, the
352 * negative return value will get converted
353 * into an error. Oops. If we keep the
354 * current syscall conventions, the only way
355 * to fix this will be in libc.
357 err = f_getown(filp);
358 force_successful_syscall_return();
361 err = f_setown(filp, arg, 1);
364 err = filp->f_owner.signum;
367 /* arg == 0 restores default behaviour. */
368 if (!valid_signal(arg)) {
372 filp->f_owner.signum = arg;
375 err = fcntl_getlease(filp);
378 err = fcntl_setlease(fd, filp, arg);
381 err = fcntl_dirnotify(fd, filp, arg);
389 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
398 err = security_file_fcntl(filp, cmd, arg);
404 err = do_fcntl(fd, cmd, arg, filp);
411 #if BITS_PER_LONG == 32
412 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
422 err = security_file_fcntl(filp, cmd, arg);
431 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
435 err = fcntl_setlk64(fd, filp, cmd,
436 (struct flock64 __user *) arg);
439 err = do_fcntl(fd, cmd, arg, filp);
448 /* Table to convert sigio signal codes into poll band bitmaps */
450 static const long band_table[NSIGPOLL] = {
451 POLLIN | POLLRDNORM, /* POLL_IN */
452 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
453 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
454 POLLERR, /* POLL_ERR */
455 POLLPRI | POLLRDBAND, /* POLL_PRI */
456 POLLHUP | POLLERR /* POLL_HUP */
459 static inline int sigio_perm(struct task_struct *p,
460 struct fown_struct *fown, int sig)
462 return (((fown->euid == 0) ||
463 (fown->euid == p->suid) || (fown->euid == p->uid) ||
464 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
465 !security_file_send_sigiotask(p, fown, sig));
468 static void send_sigio_to_task(struct task_struct *p,
469 struct fown_struct *fown,
473 if (!sigio_perm(p, fown, fown->signum))
476 switch (fown->signum) {
479 /* Queue a rt signal with the appropriate fd as its
480 value. We use SI_SIGIO as the source, not
481 SI_KERNEL, since kernel signals always get
482 delivered even if we can't queue. Failure to
483 queue in this case _should_ be reported; we fall
484 back to SIGIO in that case. --sct */
485 si.si_signo = fown->signum;
488 /* Make sure we are called with one of the POLL_*
489 reasons, otherwise we could leak kernel stack into
491 BUG_ON((reason & __SI_MASK) != __SI_POLL);
492 if (reason - POLL_IN >= NSIGPOLL)
495 si.si_band = band_table[reason - POLL_IN];
497 if (!group_send_sig_info(fown->signum, &si, p))
499 /* fall-through: fall back on the old plain SIGIO signal */
501 group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
505 void send_sigio(struct fown_struct *fown, int fd, int band)
507 struct task_struct *p;
511 read_lock(&fown->lock);
512 type = fown->pid_type;
515 goto out_unlock_fown;
517 read_lock(&tasklist_lock);
518 do_each_pid_task(pid, type, p) {
519 send_sigio_to_task(p, fown, fd, band);
520 } while_each_pid_task(pid, type, p);
521 read_unlock(&tasklist_lock);
523 read_unlock(&fown->lock);
526 static void send_sigurg_to_task(struct task_struct *p,
527 struct fown_struct *fown)
529 if (sigio_perm(p, fown, SIGURG))
530 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
533 int send_sigurg(struct fown_struct *fown)
535 struct task_struct *p;
540 read_lock(&fown->lock);
541 type = fown->pid_type;
544 goto out_unlock_fown;
548 read_lock(&tasklist_lock);
549 do_each_pid_task(pid, type, p) {
550 send_sigurg_to_task(p, fown);
551 } while_each_pid_task(pid, type, p);
552 read_unlock(&tasklist_lock);
554 read_unlock(&fown->lock);
558 static DEFINE_RWLOCK(fasync_lock);
559 static struct kmem_cache *fasync_cache __read_mostly;
562 * fasync_helper() is used by some character device drivers (mainly mice)
563 * to set up the fasync queue. It returns negative on error, 0 if it did
564 * no changes and positive if it added/deleted the entry.
566 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
568 struct fasync_struct *fa, **fp;
569 struct fasync_struct *new = NULL;
573 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
577 write_lock_irq(&fasync_lock);
578 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
579 if (fa->fa_file == filp) {
582 kmem_cache_free(fasync_cache, new);
585 kmem_cache_free(fasync_cache, fa);
593 new->magic = FASYNC_MAGIC;
596 new->fa_next = *fapp;
601 write_unlock_irq(&fasync_lock);
605 EXPORT_SYMBOL(fasync_helper);
607 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
610 struct fown_struct * fown;
611 if (fa->magic != FASYNC_MAGIC) {
612 printk(KERN_ERR "kill_fasync: bad magic number in "
616 fown = &fa->fa_file->f_owner;
617 /* Don't send SIGURG to processes which have not set a
618 queued signum: SIGURG has its own default signalling
620 if (!(sig == SIGURG && fown->signum == 0))
621 send_sigio(fown, fa->fa_fd, band);
626 EXPORT_SYMBOL(__kill_fasync);
628 void kill_fasync(struct fasync_struct **fp, int sig, int band)
630 /* First a quick test without locking: usually
634 read_lock(&fasync_lock);
635 /* reread *fp after obtaining the lock */
636 __kill_fasync(*fp, sig, band);
637 read_unlock(&fasync_lock);
640 EXPORT_SYMBOL(kill_fasync);
642 static int __init fasync_init(void)
644 fasync_cache = kmem_cache_create("fasync_cache",
645 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
649 module_init(fasync_init)