4 * Copyright (C) 1991, 1992 Linus Torvalds
7 #include <linux/init.h>
10 #include <linux/file.h>
11 #include <linux/dnotify.h>
12 #include <linux/smp_lock.h>
13 #include <linux/slab.h>
14 #include <linux/module.h>
15 #include <linux/security.h>
16 #include <linux/ptrace.h>
17 #include <linux/vs_limit.h>
20 #include <asm/siginfo.h>
21 #include <asm/uaccess.h>
23 void fastcall set_close_on_exec(unsigned int fd, int flag)
25 struct files_struct *files = current->files;
26 spin_lock(&files->file_lock);
28 FD_SET(fd, files->close_on_exec);
30 FD_CLR(fd, files->close_on_exec);
31 spin_unlock(&files->file_lock);
34 static inline int get_close_on_exec(unsigned int fd)
36 struct files_struct *files = current->files;
38 spin_lock(&files->file_lock);
39 res = FD_ISSET(fd, files->close_on_exec);
40 spin_unlock(&files->file_lock);
45 /* Expand files. Return <0 on error; 0 nothing done; 1 files expanded,
46 * we may have blocked.
48 * Should be called with the files->file_lock spinlock held for write.
50 static int expand_files(struct files_struct *files, int nr)
54 printk (KERN_ERR "%s %d: nr = %d\n", __FUNCTION__, current->pid, nr);
57 if (nr >= files->max_fdset) {
59 if ((err = expand_fdset(files, nr)))
62 if (nr >= files->max_fds) {
64 if ((err = expand_fd_array(files, nr)))
71 printk (KERN_ERR "%s %d: return %d\n", __FUNCTION__, current->pid, err);
77 * locate_fd finds a free file descriptor in the open_fds fdset,
78 * expanding the fd arrays if necessary. Must be called with the
79 * file_lock held for write.
82 static int locate_fd(struct files_struct *files,
83 struct file *file, unsigned int orig_start)
90 if (orig_start >= current->rlim[RLIMIT_NOFILE].rlim_cur)
95 * Someone might have closed fd's in the range
96 * orig_start..files->next_fd
99 if (start < files->next_fd)
100 start = files->next_fd;
103 if (start < files->max_fdset) {
104 newfd = find_next_zero_bit(files->open_fds->fds_bits,
105 files->max_fdset, start);
109 if (newfd >= current->rlim[RLIMIT_NOFILE].rlim_cur)
111 if (!vx_files_avail(1))
114 error = expand_files(files, newfd);
119 * If we needed to expand the fs array we
120 * might have blocked - try again.
125 if (start <= files->next_fd)
126 files->next_fd = newfd + 1;
134 int dupfd(struct file *file, unsigned int start)
136 struct files_struct * files = current->files;
139 spin_lock(&files->file_lock);
140 fd = locate_fd(files, file, start);
142 FD_SET(fd, files->open_fds);
143 FD_CLR(fd, files->close_on_exec);
144 spin_unlock(&files->file_lock);
146 fd_install(fd, file);
148 spin_unlock(&files->file_lock);
155 EXPORT_SYMBOL_GPL(dupfd);
157 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
160 struct file * file, *tofree;
161 struct files_struct * files = current->files;
163 spin_lock(&files->file_lock);
164 if (!(file = fcheck(oldfd)))
170 if (newfd >= current->rlim[RLIMIT_NOFILE].rlim_cur)
172 get_file(file); /* We are now finished with oldfd */
174 err = expand_files(files, newfd);
178 /* To avoid races with open() and dup(), we will mark the fd as
179 * in-use in the open-file bitmap throughout the entire dup2()
180 * process. This is quite safe: do_close() uses the fd array
181 * entry, not the bitmap, to decide what work needs to be
183 /* Doesn't work. open() might be there first. --AV */
185 /* Yes. It's a race. In user space. Nothing sane to do */
187 tofree = files->fd[newfd];
188 if (!tofree && FD_ISSET(newfd, files->open_fds))
191 files->fd[newfd] = file;
192 FD_SET(newfd, files->open_fds);
193 FD_CLR(newfd, files->close_on_exec);
194 spin_unlock(&files->file_lock);
195 vx_openfd_inc(newfd);
198 filp_close(tofree, files);
203 spin_unlock(&files->file_lock);
207 spin_unlock(&files->file_lock);
212 asmlinkage long sys_dup(unsigned int fildes)
215 struct file * file = fget(fildes);
218 ret = dupfd(file, 0);
222 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
224 static int setfl(int fd, struct file * filp, unsigned long arg)
226 struct inode * inode = filp->f_dentry->d_inode;
229 /* O_APPEND cannot be cleared if the file is marked as append-only */
230 if (!(arg & O_APPEND) && IS_APPEND(inode))
233 /* O_NOATIME can only be set by the owner or superuser */
234 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
235 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
238 /* required for strict SunOS emulation */
239 if (O_NONBLOCK != O_NDELAY)
243 if (arg & O_DIRECT) {
244 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
245 !filp->f_mapping->a_ops->direct_IO)
249 if (filp->f_op && filp->f_op->check_flags)
250 error = filp->f_op->check_flags(arg);
255 if ((arg ^ filp->f_flags) & FASYNC) {
256 if (filp->f_op && filp->f_op->fasync) {
257 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
263 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
269 static void f_modown(struct file *filp, unsigned long pid,
270 uid_t uid, uid_t euid, int force)
272 write_lock_irq(&filp->f_owner.lock);
273 if (force || !filp->f_owner.pid) {
274 filp->f_owner.pid = pid;
275 filp->f_owner.uid = uid;
276 filp->f_owner.euid = euid;
278 write_unlock_irq(&filp->f_owner.lock);
281 int f_setown(struct file *filp, unsigned long arg, int force)
285 err = security_file_set_fowner(filp);
289 f_modown(filp, arg, current->uid, current->euid, force);
293 EXPORT_SYMBOL(f_setown);
295 void f_delown(struct file *filp)
297 f_modown(filp, 0, 0, 0, 1);
300 EXPORT_SYMBOL(f_delown);
302 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
310 err = dupfd(filp, arg);
313 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
317 set_close_on_exec(fd, arg & FD_CLOEXEC);
323 err = setfl(fd, filp, arg);
326 err = fcntl_getlk(filp, (struct flock __user *) arg);
330 err = fcntl_setlk(filp, cmd, (struct flock __user *) arg);
334 * XXX If f_owner is a process group, the
335 * negative return value will get converted
336 * into an error. Oops. If we keep the
337 * current syscall conventions, the only way
338 * to fix this will be in libc.
340 err = filp->f_owner.pid;
341 force_successful_syscall_return();
344 err = f_setown(filp, arg, 1);
347 err = filp->f_owner.signum;
350 /* arg == 0 restores default behaviour. */
351 if (arg < 0 || arg > _NSIG) {
355 filp->f_owner.signum = arg;
358 err = fcntl_getlease(filp);
361 err = fcntl_setlease(fd, filp, arg);
364 err = fcntl_dirnotify(fd, filp, arg);
372 asmlinkage long sys_fcntl(int fd, unsigned int cmd, unsigned long arg)
381 err = security_file_fcntl(filp, cmd, arg);
387 err = do_fcntl(fd, cmd, arg, filp);
394 #if BITS_PER_LONG == 32
395 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
405 err = security_file_fcntl(filp, cmd, arg);
414 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
418 err = fcntl_setlk64(filp, cmd, (struct flock64 __user *) arg);
421 err = do_fcntl(fd, cmd, arg, filp);
430 /* Table to convert sigio signal codes into poll band bitmaps */
432 static long band_table[NSIGPOLL] = {
433 POLLIN | POLLRDNORM, /* POLL_IN */
434 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
435 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
436 POLLERR, /* POLL_ERR */
437 POLLPRI | POLLRDBAND, /* POLL_PRI */
438 POLLHUP | POLLERR /* POLL_HUP */
441 static inline int sigio_perm(struct task_struct *p,
442 struct fown_struct *fown)
444 return ((fown->euid == 0) ||
445 (fown->euid == p->suid) || (fown->euid == p->uid) ||
446 (fown->uid == p->suid) || (fown->uid == p->uid));
449 static void send_sigio_to_task(struct task_struct *p,
450 struct fown_struct *fown,
454 if (!sigio_perm(p, fown))
457 if (security_file_send_sigiotask(p, fown, fd, reason))
460 switch (fown->signum) {
463 /* Queue a rt signal with the appropriate fd as its
464 value. We use SI_SIGIO as the source, not
465 SI_KERNEL, since kernel signals always get
466 delivered even if we can't queue. Failure to
467 queue in this case _should_ be reported; we fall
468 back to SIGIO in that case. --sct */
469 si.si_signo = fown->signum;
472 /* Make sure we are called with one of the POLL_*
473 reasons, otherwise we could leak kernel stack into
475 if ((reason & __SI_MASK) != __SI_POLL)
477 if (reason - POLL_IN >= NSIGPOLL)
480 si.si_band = band_table[reason - POLL_IN];
482 if (!send_sig_info(fown->signum, &si, p))
484 /* fall-through: fall back on the old plain SIGIO signal */
486 send_group_sig_info(SIGIO, SEND_SIG_PRIV, p);
490 void send_sigio(struct fown_struct *fown, int fd, int band)
492 struct task_struct *p;
495 read_lock(&fown->lock);
498 goto out_unlock_fown;
500 read_lock(&tasklist_lock);
502 p = find_task_by_pid(pid);
504 send_sigio_to_task(p, fown, fd, band);
509 for_each_task_pid(-pid, PIDTYPE_PGID, p, l, pidptr) {
510 send_sigio_to_task(p, fown, fd, band);
513 read_unlock(&tasklist_lock);
515 read_unlock(&fown->lock);
518 static void send_sigurg_to_task(struct task_struct *p,
519 struct fown_struct *fown)
521 if (sigio_perm(p, fown))
522 send_group_sig_info(SIGURG, SEND_SIG_PRIV, p);
525 int send_sigurg(struct fown_struct *fown)
527 struct task_struct *p;
530 read_lock(&fown->lock);
533 goto out_unlock_fown;
537 read_lock(&tasklist_lock);
539 p = find_task_by_pid(pid);
541 send_sigurg_to_task(p, fown);
546 for_each_task_pid(-pid, PIDTYPE_PGID, p, l, pidptr) {
547 send_sigurg_to_task(p, fown);
550 read_unlock(&tasklist_lock);
552 read_unlock(&fown->lock);
556 static rwlock_t fasync_lock = RW_LOCK_UNLOCKED;
557 static kmem_cache_t *fasync_cache;
560 * fasync_helper() is used by some character device drivers (mainly mice)
561 * to set up the fasync queue. It returns negative on error, 0 if it did
562 * no changes and positive if it added/deleted the entry.
564 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
566 struct fasync_struct *fa, **fp;
567 struct fasync_struct *new = NULL;
571 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
575 write_lock_irq(&fasync_lock);
576 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
577 if (fa->fa_file == filp) {
580 kmem_cache_free(fasync_cache, new);
583 kmem_cache_free(fasync_cache, fa);
591 new->magic = FASYNC_MAGIC;
594 new->fa_next = *fapp;
599 write_unlock_irq(&fasync_lock);
603 EXPORT_SYMBOL(fasync_helper);
605 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
608 struct fown_struct * fown;
609 if (fa->magic != FASYNC_MAGIC) {
610 printk(KERN_ERR "kill_fasync: bad magic number in "
614 fown = &fa->fa_file->f_owner;
615 /* Don't send SIGURG to processes which have not set a
616 queued signum: SIGURG has its own default signalling
618 if (!(sig == SIGURG && fown->signum == 0))
619 send_sigio(fown, fa->fa_fd, band);
624 EXPORT_SYMBOL(__kill_fasync);
626 void kill_fasync(struct fasync_struct **fp, int sig, int band)
628 /* First a quick test without locking: usually
632 read_lock(&fasync_lock);
633 /* reread *fp after obtaining the lock */
634 __kill_fasync(*fp, sig, band);
635 read_unlock(&fasync_lock);
638 EXPORT_SYMBOL(kill_fasync);
640 static int __init fasync_init(void)
642 fasync_cache = kmem_cache_create("fasync_cache",
643 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
647 module_init(fasync_init)