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>
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 * locate_fd finds a free file descriptor in the open_fds fdset,
46 * expanding the fd arrays if necessary. Must be called with the
47 * file_lock held for write.
50 static int locate_fd(struct files_struct *files,
51 struct file *file, unsigned int orig_start)
58 if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
63 * Someone might have closed fd's in the range
64 * orig_start..files->next_fd
67 if (start < files->next_fd)
68 start = files->next_fd;
71 if (start < files->max_fdset) {
72 newfd = find_next_zero_bit(files->open_fds->fds_bits,
73 files->max_fdset, start);
77 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
80 error = expand_files(files, newfd);
85 * If we needed to expand the fs array we
86 * might have blocked - try again.
91 if (start <= files->next_fd)
92 files->next_fd = newfd + 1;
100 static int dupfd(struct file *file, unsigned int start)
102 struct files_struct * files = current->files;
105 spin_lock(&files->file_lock);
106 fd = locate_fd(files, file, start);
108 FD_SET(fd, files->open_fds);
109 FD_CLR(fd, files->close_on_exec);
110 spin_unlock(&files->file_lock);
111 fd_install(fd, file);
113 spin_unlock(&files->file_lock);
120 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
123 struct file * file, *tofree;
124 struct files_struct * files = current->files;
126 spin_lock(&files->file_lock);
127 if (!(file = fcheck(oldfd)))
133 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
135 get_file(file); /* We are now finished with oldfd */
137 err = expand_files(files, newfd);
141 /* To avoid races with open() and dup(), we will mark the fd as
142 * in-use in the open-file bitmap throughout the entire dup2()
143 * process. This is quite safe: do_close() uses the fd array
144 * entry, not the bitmap, to decide what work needs to be
146 /* Doesn't work. open() might be there first. --AV */
148 /* Yes. It's a race. In user space. Nothing sane to do */
150 tofree = files->fd[newfd];
151 if (!tofree && FD_ISSET(newfd, files->open_fds))
154 files->fd[newfd] = file;
155 FD_SET(newfd, files->open_fds);
156 FD_CLR(newfd, files->close_on_exec);
157 spin_unlock(&files->file_lock);
160 filp_close(tofree, files);
165 spin_unlock(&files->file_lock);
169 spin_unlock(&files->file_lock);
174 asmlinkage long sys_dup(unsigned int fildes)
177 struct file * file = fget(fildes);
180 ret = dupfd(file, 0);
184 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
186 static int setfl(int fd, struct file * filp, unsigned long arg)
188 struct inode * inode = filp->f_dentry->d_inode;
191 /* O_APPEND cannot be cleared if the file is marked as append-only */
192 if (!(arg & O_APPEND) && IS_APPEND(inode))
195 /* O_NOATIME can only be set by the owner or superuser */
196 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
197 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
200 /* required for strict SunOS emulation */
201 if (O_NONBLOCK != O_NDELAY)
205 if (arg & O_DIRECT) {
206 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
207 !filp->f_mapping->a_ops->direct_IO)
211 if (filp->f_op && filp->f_op->check_flags)
212 error = filp->f_op->check_flags(arg);
217 if ((arg ^ filp->f_flags) & FASYNC) {
218 if (filp->f_op && filp->f_op->fasync) {
219 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
225 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
231 static void f_modown(struct file *filp, unsigned long pid,
232 uid_t uid, uid_t euid, int force)
234 write_lock_irq(&filp->f_owner.lock);
235 if (force || !filp->f_owner.pid) {
236 filp->f_owner.pid = pid;
237 filp->f_owner.uid = uid;
238 filp->f_owner.euid = euid;
240 write_unlock_irq(&filp->f_owner.lock);
243 int f_setown(struct file *filp, unsigned long arg, int force)
247 err = security_file_set_fowner(filp);
251 f_modown(filp, arg, current->uid, current->euid, force);
255 EXPORT_SYMBOL(f_setown);
257 void f_delown(struct file *filp)
259 f_modown(filp, 0, 0, 0, 1);
262 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
270 err = dupfd(filp, arg);
273 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
277 set_close_on_exec(fd, arg & FD_CLOEXEC);
283 err = setfl(fd, filp, arg);
286 err = fcntl_getlk(filp, (struct flock __user *) arg);
290 err = fcntl_setlk(filp, cmd, (struct flock __user *) arg);
294 * XXX If f_owner is a process group, the
295 * negative return value will get converted
296 * into an error. Oops. If we keep the
297 * current syscall conventions, the only way
298 * to fix this will be in libc.
300 err = filp->f_owner.pid;
301 force_successful_syscall_return();
304 err = f_setown(filp, arg, 1);
307 err = filp->f_owner.signum;
310 /* arg == 0 restores default behaviour. */
311 if (arg < 0 || arg > _NSIG) {
315 filp->f_owner.signum = arg;
318 err = fcntl_getlease(filp);
321 err = fcntl_setlease(fd, filp, arg);
324 err = fcntl_dirnotify(fd, filp, arg);
332 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
341 err = security_file_fcntl(filp, cmd, arg);
347 err = do_fcntl(fd, cmd, arg, filp);
354 #if BITS_PER_LONG == 32
355 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
365 err = security_file_fcntl(filp, cmd, arg);
374 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
378 err = fcntl_setlk64(filp, cmd, (struct flock64 __user *) arg);
381 err = do_fcntl(fd, cmd, arg, filp);
390 /* Table to convert sigio signal codes into poll band bitmaps */
392 static long band_table[NSIGPOLL] = {
393 POLLIN | POLLRDNORM, /* POLL_IN */
394 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
395 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
396 POLLERR, /* POLL_ERR */
397 POLLPRI | POLLRDBAND, /* POLL_PRI */
398 POLLHUP | POLLERR /* POLL_HUP */
401 static inline int sigio_perm(struct task_struct *p,
402 struct fown_struct *fown, int sig)
404 return (((fown->euid == 0) ||
405 (fown->euid == p->suid) || (fown->euid == p->uid) ||
406 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
407 !security_file_send_sigiotask(p, fown, sig));
410 static void send_sigio_to_task(struct task_struct *p,
411 struct fown_struct *fown,
415 if (!sigio_perm(p, fown, fown->signum))
418 switch (fown->signum) {
421 /* Queue a rt signal with the appropriate fd as its
422 value. We use SI_SIGIO as the source, not
423 SI_KERNEL, since kernel signals always get
424 delivered even if we can't queue. Failure to
425 queue in this case _should_ be reported; we fall
426 back to SIGIO in that case. --sct */
427 si.si_signo = fown->signum;
430 /* Make sure we are called with one of the POLL_*
431 reasons, otherwise we could leak kernel stack into
433 if ((reason & __SI_MASK) != __SI_POLL)
435 if (reason - POLL_IN >= NSIGPOLL)
438 si.si_band = band_table[reason - POLL_IN];
440 if (!send_sig_info(fown->signum, &si, p))
442 /* fall-through: fall back on the old plain SIGIO signal */
444 send_group_sig_info(SIGIO, SEND_SIG_PRIV, p);
448 void send_sigio(struct fown_struct *fown, int fd, int band)
450 struct task_struct *p;
453 read_lock(&fown->lock);
456 goto out_unlock_fown;
458 read_lock(&tasklist_lock);
460 p = find_task_by_pid(pid);
462 send_sigio_to_task(p, fown, fd, band);
465 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
466 send_sigio_to_task(p, fown, fd, band);
467 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
469 read_unlock(&tasklist_lock);
471 read_unlock(&fown->lock);
474 static void send_sigurg_to_task(struct task_struct *p,
475 struct fown_struct *fown)
477 if (sigio_perm(p, fown, SIGURG))
478 send_group_sig_info(SIGURG, SEND_SIG_PRIV, p);
481 int send_sigurg(struct fown_struct *fown)
483 struct task_struct *p;
486 read_lock(&fown->lock);
489 goto out_unlock_fown;
493 read_lock(&tasklist_lock);
495 p = find_task_by_pid(pid);
497 send_sigurg_to_task(p, fown);
500 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
501 send_sigurg_to_task(p, fown);
502 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
504 read_unlock(&tasklist_lock);
506 read_unlock(&fown->lock);
510 static DEFINE_RWLOCK(fasync_lock);
511 static kmem_cache_t *fasync_cache;
514 * fasync_helper() is used by some character device drivers (mainly mice)
515 * to set up the fasync queue. It returns negative on error, 0 if it did
516 * no changes and positive if it added/deleted the entry.
518 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
520 struct fasync_struct *fa, **fp;
521 struct fasync_struct *new = NULL;
525 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
529 write_lock_irq(&fasync_lock);
530 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
531 if (fa->fa_file == filp) {
534 kmem_cache_free(fasync_cache, new);
537 kmem_cache_free(fasync_cache, fa);
545 new->magic = FASYNC_MAGIC;
548 new->fa_next = *fapp;
553 write_unlock_irq(&fasync_lock);
557 EXPORT_SYMBOL(fasync_helper);
559 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
562 struct fown_struct * fown;
563 if (fa->magic != FASYNC_MAGIC) {
564 printk(KERN_ERR "kill_fasync: bad magic number in "
568 fown = &fa->fa_file->f_owner;
569 /* Don't send SIGURG to processes which have not set a
570 queued signum: SIGURG has its own default signalling
572 if (!(sig == SIGURG && fown->signum == 0))
573 send_sigio(fown, fa->fa_fd, band);
578 EXPORT_SYMBOL(__kill_fasync);
580 void kill_fasync(struct fasync_struct **fp, int sig, int band)
582 /* First a quick test without locking: usually
586 read_lock(&fasync_lock);
587 /* reread *fp after obtaining the lock */
588 __kill_fasync(*fp, sig, band);
589 read_unlock(&fasync_lock);
592 EXPORT_SYMBOL(kill_fasync);
594 static int __init fasync_init(void)
596 fasync_cache = kmem_cache_create("fasync_cache",
597 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
601 module_init(fasync_init)