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 static 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 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
158 struct file * file, *tofree;
159 struct files_struct * files = current->files;
161 spin_lock(&files->file_lock);
162 if (!(file = fcheck(oldfd)))
168 if (newfd >= current->rlim[RLIMIT_NOFILE].rlim_cur)
170 get_file(file); /* We are now finished with oldfd */
172 err = expand_files(files, newfd);
176 /* To avoid races with open() and dup(), we will mark the fd as
177 * in-use in the open-file bitmap throughout the entire dup2()
178 * process. This is quite safe: do_close() uses the fd array
179 * entry, not the bitmap, to decide what work needs to be
181 /* Doesn't work. open() might be there first. --AV */
183 /* Yes. It's a race. In user space. Nothing sane to do */
185 tofree = files->fd[newfd];
186 if (!tofree && FD_ISSET(newfd, files->open_fds))
189 files->fd[newfd] = file;
190 FD_SET(newfd, files->open_fds);
191 FD_CLR(newfd, files->close_on_exec);
192 spin_unlock(&files->file_lock);
193 vx_openfd_inc(newfd);
196 filp_close(tofree, files);
201 spin_unlock(&files->file_lock);
205 spin_unlock(&files->file_lock);
210 asmlinkage long sys_dup(unsigned int fildes)
213 struct file * file = fget(fildes);
216 ret = dupfd(file, 0);
220 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT)
222 static int setfl(int fd, struct file * filp, unsigned long arg)
224 struct inode * inode = filp->f_dentry->d_inode;
227 /* O_APPEND cannot be cleared if the file is marked as append-only */
228 if (!(arg & O_APPEND) && IS_APPEND(inode))
231 /* required for strict SunOS emulation */
232 if (O_NONBLOCK != O_NDELAY)
236 if (arg & O_DIRECT) {
237 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
238 !filp->f_mapping->a_ops->direct_IO)
243 if ((arg ^ filp->f_flags) & FASYNC) {
244 if (filp->f_op && filp->f_op->fasync) {
245 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
251 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
257 static void f_modown(struct file *filp, unsigned long pid,
258 uid_t uid, uid_t euid, int force)
260 write_lock_irq(&filp->f_owner.lock);
261 if (force || !filp->f_owner.pid) {
262 filp->f_owner.pid = pid;
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, unsigned long arg, int force)
273 err = security_file_set_fowner(filp);
277 f_modown(filp, arg, current->uid, current->euid, force);
281 EXPORT_SYMBOL(f_setown);
283 void f_delown(struct file *filp)
285 f_modown(filp, 0, 0, 0, 1);
288 EXPORT_SYMBOL(f_delown);
290 long generic_file_fcntl(int fd, unsigned int cmd,
291 unsigned long arg, struct file *filp)
298 err = dupfd(filp, arg);
301 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
305 set_close_on_exec(fd, arg & FD_CLOEXEC);
311 err = setfl(fd, filp, arg);
314 err = fcntl_getlk(filp, (struct flock __user *) arg);
318 err = fcntl_setlk(filp, cmd, (struct flock __user *) arg);
322 * XXX If f_owner is a process group, the
323 * negative return value will get converted
324 * into an error. Oops. If we keep the
325 * current syscall conventions, the only way
326 * to fix this will be in libc.
328 err = filp->f_owner.pid;
329 force_successful_syscall_return();
332 err = f_setown(filp, arg, 1);
335 err = filp->f_owner.signum;
338 /* arg == 0 restores default behaviour. */
339 if (arg < 0 || arg > _NSIG) {
343 filp->f_owner.signum = arg;
346 err = fcntl_getlease(filp);
349 err = fcntl_setlease(fd, filp, arg);
352 err = fcntl_dirnotify(fd, filp, arg);
359 EXPORT_SYMBOL(generic_file_fcntl);
361 static long do_fcntl(int fd, unsigned int cmd,
362 unsigned long arg, struct file *filp)
364 if (filp->f_op && filp->f_op->fcntl)
365 return filp->f_op->fcntl(fd, cmd, arg, filp);
366 return generic_file_fcntl(fd, cmd, arg, filp);
369 asmlinkage long sys_fcntl(int fd, unsigned int cmd, unsigned long arg)
378 err = security_file_fcntl(filp, cmd, arg);
384 err = do_fcntl(fd, cmd, arg, filp);
391 #if BITS_PER_LONG == 32
392 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
402 err = security_file_fcntl(filp, cmd, arg);
411 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
415 err = fcntl_setlk64(filp, cmd, (struct flock64 __user *) arg);
418 err = do_fcntl(fd, cmd, arg, filp);
427 /* Table to convert sigio signal codes into poll band bitmaps */
429 static long band_table[NSIGPOLL] = {
430 POLLIN | POLLRDNORM, /* POLL_IN */
431 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
432 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
433 POLLERR, /* POLL_ERR */
434 POLLPRI | POLLRDBAND, /* POLL_PRI */
435 POLLHUP | POLLERR /* POLL_HUP */
438 static inline int sigio_perm(struct task_struct *p,
439 struct fown_struct *fown)
441 return ((fown->euid == 0) ||
442 (fown->euid == p->suid) || (fown->euid == p->uid) ||
443 (fown->uid == p->suid) || (fown->uid == p->uid));
446 static void send_sigio_to_task(struct task_struct *p,
447 struct fown_struct *fown,
451 if (!sigio_perm(p, fown))
454 if (security_file_send_sigiotask(p, fown, fd, reason))
457 switch (fown->signum) {
460 /* Queue a rt signal with the appropriate fd as its
461 value. We use SI_SIGIO as the source, not
462 SI_KERNEL, since kernel signals always get
463 delivered even if we can't queue. Failure to
464 queue in this case _should_ be reported; we fall
465 back to SIGIO in that case. --sct */
466 si.si_signo = fown->signum;
469 /* Make sure we are called with one of the POLL_*
470 reasons, otherwise we could leak kernel stack into
472 if ((reason & __SI_MASK) != __SI_POLL)
474 if (reason - POLL_IN >= NSIGPOLL)
477 si.si_band = band_table[reason - POLL_IN];
479 if (!send_sig_info(fown->signum, &si, p))
481 /* fall-through: fall back on the old plain SIGIO signal */
483 send_group_sig_info(SIGIO, SEND_SIG_PRIV, p);
487 void send_sigio(struct fown_struct *fown, int fd, int band)
489 struct task_struct *p;
492 read_lock(&fown->lock);
495 goto out_unlock_fown;
497 read_lock(&tasklist_lock);
499 p = find_task_by_pid(pid);
501 send_sigio_to_task(p, fown, fd, band);
506 for_each_task_pid(-pid, PIDTYPE_PGID, p, l, pidptr) {
507 send_sigio_to_task(p, fown, fd, band);
510 read_unlock(&tasklist_lock);
512 read_unlock(&fown->lock);
515 static void send_sigurg_to_task(struct task_struct *p,
516 struct fown_struct *fown)
518 if (sigio_perm(p, fown))
519 send_group_sig_info(SIGURG, SEND_SIG_PRIV, p);
522 int send_sigurg(struct fown_struct *fown)
524 struct task_struct *p;
527 read_lock(&fown->lock);
530 goto out_unlock_fown;
534 read_lock(&tasklist_lock);
536 p = find_task_by_pid(pid);
538 send_sigurg_to_task(p, fown);
543 for_each_task_pid(-pid, PIDTYPE_PGID, p, l, pidptr) {
544 send_sigurg_to_task(p, fown);
547 read_unlock(&tasklist_lock);
549 read_unlock(&fown->lock);
553 static rwlock_t fasync_lock = RW_LOCK_UNLOCKED;
554 static kmem_cache_t *fasync_cache;
557 * fasync_helper() is used by some character device drivers (mainly mice)
558 * to set up the fasync queue. It returns negative on error, 0 if it did
559 * no changes and positive if it added/deleted the entry.
561 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
563 struct fasync_struct *fa, **fp;
564 struct fasync_struct *new = NULL;
568 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
572 write_lock_irq(&fasync_lock);
573 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
574 if (fa->fa_file == filp) {
577 kmem_cache_free(fasync_cache, new);
580 kmem_cache_free(fasync_cache, fa);
588 new->magic = FASYNC_MAGIC;
591 new->fa_next = *fapp;
596 write_unlock_irq(&fasync_lock);
600 EXPORT_SYMBOL(fasync_helper);
602 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
605 struct fown_struct * fown;
606 if (fa->magic != FASYNC_MAGIC) {
607 printk(KERN_ERR "kill_fasync: bad magic number in "
611 fown = &fa->fa_file->f_owner;
612 /* Don't send SIGURG to processes which have not set a
613 queued signum: SIGURG has its own default signalling
615 if (!(sig == SIGURG && fown->signum == 0))
616 send_sigio(fown, fa->fa_fd, band);
621 EXPORT_SYMBOL(__kill_fasync);
623 void kill_fasync(struct fasync_struct **fp, int sig, int band)
625 /* First a quick test without locking: usually
629 read_lock(&fasync_lock);
630 /* reread *fp after obtaining the lock */
631 __kill_fasync(*fp, sig, band);
632 read_unlock(&fasync_lock);
635 EXPORT_SYMBOL(kill_fasync);
637 static int __init fasync_init(void)
639 fasync_cache = kmem_cache_create("fasync_cache",
640 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
644 module_init(fasync_init)