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/dnotify.h>
13 #include <linux/smp_lock.h>
14 #include <linux/slab.h>
15 #include <linux/module.h>
16 #include <linux/security.h>
17 #include <linux/ptrace.h>
18 #include <linux/signal.h>
19 #include <linux/vs_limit.h>
22 #include <asm/siginfo.h>
23 #include <asm/uaccess.h>
25 void fastcall set_close_on_exec(unsigned int fd, int flag)
27 struct files_struct *files = current->files;
28 spin_lock(&files->file_lock);
30 FD_SET(fd, files->close_on_exec);
32 FD_CLR(fd, files->close_on_exec);
33 spin_unlock(&files->file_lock);
36 static inline int get_close_on_exec(unsigned int fd)
38 struct files_struct *files = current->files;
40 spin_lock(&files->file_lock);
41 res = FD_ISSET(fd, files->close_on_exec);
42 spin_unlock(&files->file_lock);
47 * locate_fd finds a free file descriptor in the open_fds fdset,
48 * expanding the fd arrays if necessary. Must be called with the
49 * file_lock held for write.
52 static int locate_fd(struct files_struct *files,
53 struct file *file, unsigned int orig_start)
60 if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
65 * Someone might have closed fd's in the range
66 * orig_start..files->next_fd
69 if (start < files->next_fd)
70 start = files->next_fd;
73 if (start < files->max_fdset) {
74 newfd = find_next_zero_bit(files->open_fds->fds_bits,
75 files->max_fdset, start);
79 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
81 if (!vx_files_avail(1))
84 error = expand_files(files, newfd);
89 * If we needed to expand the fs array we
90 * might have blocked - try again.
95 if (start <= files->next_fd)
96 files->next_fd = newfd + 1;
104 static int dupfd(struct file *file, unsigned int start)
106 struct files_struct * files = current->files;
109 spin_lock(&files->file_lock);
110 fd = locate_fd(files, file, start);
112 FD_SET(fd, files->open_fds);
113 FD_CLR(fd, files->close_on_exec);
114 spin_unlock(&files->file_lock);
116 fd_install(fd, file);
118 spin_unlock(&files->file_lock);
125 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
128 struct file * file, *tofree;
129 struct files_struct * files = current->files;
131 spin_lock(&files->file_lock);
132 if (!(file = fcheck(oldfd)))
138 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
140 get_file(file); /* We are now finished with oldfd */
142 err = expand_files(files, newfd);
146 /* To avoid races with open() and dup(), we will mark the fd as
147 * in-use in the open-file bitmap throughout the entire dup2()
148 * process. This is quite safe: do_close() uses the fd array
149 * entry, not the bitmap, to decide what work needs to be
151 /* Doesn't work. open() might be there first. --AV */
153 /* Yes. It's a race. In user space. Nothing sane to do */
155 tofree = files->fd[newfd];
156 if (!tofree && FD_ISSET(newfd, files->open_fds))
159 files->fd[newfd] = file;
160 FD_SET(newfd, files->open_fds);
161 FD_CLR(newfd, files->close_on_exec);
162 spin_unlock(&files->file_lock);
165 filp_close(tofree, files);
167 vx_openfd_inc(newfd); /* fd was unused */
173 spin_unlock(&files->file_lock);
177 spin_unlock(&files->file_lock);
182 asmlinkage long sys_dup(unsigned int fildes)
185 struct file * file = fget(fildes);
188 ret = dupfd(file, 0);
192 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
194 static int setfl(int fd, struct file * filp, unsigned long arg)
196 struct inode * inode = filp->f_dentry->d_inode;
199 /* O_APPEND cannot be cleared if the file is marked as append-only */
200 if (!(arg & O_APPEND) && IS_APPEND(inode))
203 /* O_NOATIME can only be set by the owner or superuser */
204 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
205 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
208 /* required for strict SunOS emulation */
209 if (O_NONBLOCK != O_NDELAY)
213 if (arg & O_DIRECT) {
214 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
215 !filp->f_mapping->a_ops->direct_IO)
219 if (filp->f_op && filp->f_op->check_flags)
220 error = filp->f_op->check_flags(arg);
225 if ((arg ^ filp->f_flags) & FASYNC) {
226 if (filp->f_op && filp->f_op->fasync) {
227 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
233 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
239 static void f_modown(struct file *filp, unsigned long pid,
240 uid_t uid, uid_t euid, int force)
242 write_lock_irq(&filp->f_owner.lock);
243 if (force || !filp->f_owner.pid) {
244 filp->f_owner.pid = pid;
245 filp->f_owner.uid = uid;
246 filp->f_owner.euid = euid;
248 write_unlock_irq(&filp->f_owner.lock);
251 int f_setown(struct file *filp, unsigned long arg, int force)
255 err = security_file_set_fowner(filp);
259 f_modown(filp, arg, current->uid, current->euid, force);
263 EXPORT_SYMBOL(f_setown);
265 void f_delown(struct file *filp)
267 f_modown(filp, 0, 0, 0, 1);
270 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
278 err = dupfd(filp, arg);
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(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 = filp->f_owner.pid;
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 asmlinkage long sys_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 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
373 err = security_file_fcntl(filp, cmd, arg);
382 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
386 err = fcntl_setlk64(filp, cmd, (struct flock64 __user *) arg);
389 err = do_fcntl(fd, cmd, arg, filp);
398 /* Table to convert sigio signal codes into poll band bitmaps */
400 static long band_table[NSIGPOLL] = {
401 POLLIN | POLLRDNORM, /* POLL_IN */
402 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
403 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
404 POLLERR, /* POLL_ERR */
405 POLLPRI | POLLRDBAND, /* POLL_PRI */
406 POLLHUP | POLLERR /* POLL_HUP */
409 static inline int sigio_perm(struct task_struct *p,
410 struct fown_struct *fown, int sig)
412 return (((fown->euid == 0) ||
413 (fown->euid == p->suid) || (fown->euid == p->uid) ||
414 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
415 !security_file_send_sigiotask(p, fown, sig));
418 static void send_sigio_to_task(struct task_struct *p,
419 struct fown_struct *fown,
423 if (!sigio_perm(p, fown, fown->signum))
426 switch (fown->signum) {
429 /* Queue a rt signal with the appropriate fd as its
430 value. We use SI_SIGIO as the source, not
431 SI_KERNEL, since kernel signals always get
432 delivered even if we can't queue. Failure to
433 queue in this case _should_ be reported; we fall
434 back to SIGIO in that case. --sct */
435 si.si_signo = fown->signum;
438 /* Make sure we are called with one of the POLL_*
439 reasons, otherwise we could leak kernel stack into
441 if ((reason & __SI_MASK) != __SI_POLL)
443 if (reason - POLL_IN >= NSIGPOLL)
446 si.si_band = band_table[reason - POLL_IN];
448 if (!send_group_sig_info(fown->signum, &si, p))
450 /* fall-through: fall back on the old plain SIGIO signal */
452 send_group_sig_info(SIGIO, SEND_SIG_PRIV, p);
456 void send_sigio(struct fown_struct *fown, int fd, int band)
458 struct task_struct *p;
461 read_lock(&fown->lock);
464 goto out_unlock_fown;
466 read_lock(&tasklist_lock);
468 p = find_task_by_real_pid(pid);
470 send_sigio_to_task(p, fown, fd, band);
473 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
474 send_sigio_to_task(p, fown, fd, band);
475 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
477 read_unlock(&tasklist_lock);
479 read_unlock(&fown->lock);
482 static void send_sigurg_to_task(struct task_struct *p,
483 struct fown_struct *fown)
485 if (sigio_perm(p, fown, SIGURG))
486 send_group_sig_info(SIGURG, SEND_SIG_PRIV, p);
489 int send_sigurg(struct fown_struct *fown)
491 struct task_struct *p;
494 read_lock(&fown->lock);
497 goto out_unlock_fown;
501 read_lock(&tasklist_lock);
503 p = find_task_by_real_pid(pid);
505 send_sigurg_to_task(p, fown);
508 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
509 send_sigurg_to_task(p, fown);
510 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
512 read_unlock(&tasklist_lock);
514 read_unlock(&fown->lock);
518 static DEFINE_RWLOCK(fasync_lock);
519 static kmem_cache_t *fasync_cache;
522 * fasync_helper() is used by some character device drivers (mainly mice)
523 * to set up the fasync queue. It returns negative on error, 0 if it did
524 * no changes and positive if it added/deleted the entry.
526 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
528 struct fasync_struct *fa, **fp;
529 struct fasync_struct *new = NULL;
533 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
537 write_lock_irq(&fasync_lock);
538 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
539 if (fa->fa_file == filp) {
542 kmem_cache_free(fasync_cache, new);
545 kmem_cache_free(fasync_cache, fa);
553 new->magic = FASYNC_MAGIC;
556 new->fa_next = *fapp;
561 write_unlock_irq(&fasync_lock);
565 EXPORT_SYMBOL(fasync_helper);
567 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
570 struct fown_struct * fown;
571 if (fa->magic != FASYNC_MAGIC) {
572 printk(KERN_ERR "kill_fasync: bad magic number in "
576 fown = &fa->fa_file->f_owner;
577 /* Don't send SIGURG to processes which have not set a
578 queued signum: SIGURG has its own default signalling
580 if (!(sig == SIGURG && fown->signum == 0))
581 send_sigio(fown, fa->fa_fd, band);
586 EXPORT_SYMBOL(__kill_fasync);
588 void kill_fasync(struct fasync_struct **fp, int sig, int band)
590 /* First a quick test without locking: usually
594 read_lock(&fasync_lock);
595 /* reread *fp after obtaining the lock */
596 __kill_fasync(*fp, sig, band);
597 read_unlock(&fasync_lock);
600 EXPORT_SYMBOL(kill_fasync);
602 static int __init fasync_init(void)
604 fasync_cache = kmem_cache_create("fasync_cache",
605 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
609 module_init(fasync_init)