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>
21 #include <asm/siginfo.h>
22 #include <asm/uaccess.h>
24 void fastcall set_close_on_exec(unsigned int fd, int flag)
26 struct files_struct *files = current->files;
27 spin_lock(&files->file_lock);
29 FD_SET(fd, files->close_on_exec);
31 FD_CLR(fd, files->close_on_exec);
32 spin_unlock(&files->file_lock);
35 static inline int get_close_on_exec(unsigned int fd)
37 struct files_struct *files = current->files;
39 spin_lock(&files->file_lock);
40 res = FD_ISSET(fd, files->close_on_exec);
41 spin_unlock(&files->file_lock);
46 * locate_fd finds a free file descriptor in the open_fds fdset,
47 * expanding the fd arrays if necessary. Must be called with the
48 * file_lock held for write.
51 static int locate_fd(struct files_struct *files,
52 struct file *file, unsigned int orig_start)
59 if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
64 * Someone might have closed fd's in the range
65 * orig_start..files->next_fd
68 if (start < files->next_fd)
69 start = files->next_fd;
72 if (start < files->max_fdset) {
73 newfd = find_next_zero_bit(files->open_fds->fds_bits,
74 files->max_fdset, start);
78 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
81 error = expand_files(files, newfd);
86 * If we needed to expand the fs array we
87 * might have blocked - try again.
92 if (start <= files->next_fd)
93 files->next_fd = newfd + 1;
101 static int dupfd(struct file *file, unsigned int start)
103 struct files_struct * files = current->files;
106 spin_lock(&files->file_lock);
107 fd = locate_fd(files, file, start);
109 FD_SET(fd, files->open_fds);
110 FD_CLR(fd, files->close_on_exec);
111 spin_unlock(&files->file_lock);
112 fd_install(fd, file);
114 spin_unlock(&files->file_lock);
121 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
124 struct file * file, *tofree;
125 struct files_struct * files = current->files;
127 spin_lock(&files->file_lock);
128 if (!(file = fcheck(oldfd)))
134 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
136 get_file(file); /* We are now finished with oldfd */
138 err = expand_files(files, newfd);
142 /* To avoid races with open() and dup(), we will mark the fd as
143 * in-use in the open-file bitmap throughout the entire dup2()
144 * process. This is quite safe: do_close() uses the fd array
145 * entry, not the bitmap, to decide what work needs to be
147 /* Doesn't work. open() might be there first. --AV */
149 /* Yes. It's a race. In user space. Nothing sane to do */
151 tofree = files->fd[newfd];
152 if (!tofree && FD_ISSET(newfd, files->open_fds))
155 files->fd[newfd] = file;
156 FD_SET(newfd, files->open_fds);
157 FD_CLR(newfd, files->close_on_exec);
158 spin_unlock(&files->file_lock);
161 filp_close(tofree, files);
166 spin_unlock(&files->file_lock);
170 spin_unlock(&files->file_lock);
175 asmlinkage long sys_dup(unsigned int fildes)
178 struct file * file = fget(fildes);
181 ret = dupfd(file, 0);
185 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
187 static int setfl(int fd, struct file * filp, unsigned long arg)
189 struct inode * inode = filp->f_dentry->d_inode;
192 /* O_APPEND cannot be cleared if the file is marked as append-only */
193 if (!(arg & O_APPEND) && IS_APPEND(inode))
196 /* O_NOATIME can only be set by the owner or superuser */
197 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
198 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
201 /* required for strict SunOS emulation */
202 if (O_NONBLOCK != O_NDELAY)
206 if (arg & O_DIRECT) {
207 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
208 !filp->f_mapping->a_ops->direct_IO)
212 if (filp->f_op && filp->f_op->check_flags)
213 error = filp->f_op->check_flags(arg);
218 if ((arg ^ filp->f_flags) & FASYNC) {
219 if (filp->f_op && filp->f_op->fasync) {
220 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
226 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
232 static void f_modown(struct file *filp, unsigned long pid,
233 uid_t uid, uid_t euid, int force)
235 write_lock_irq(&filp->f_owner.lock);
236 if (force || !filp->f_owner.pid) {
237 filp->f_owner.pid = pid;
238 filp->f_owner.uid = uid;
239 filp->f_owner.euid = euid;
241 write_unlock_irq(&filp->f_owner.lock);
244 int f_setown(struct file *filp, unsigned long arg, int force)
248 err = security_file_set_fowner(filp);
252 f_modown(filp, arg, current->uid, current->euid, force);
256 EXPORT_SYMBOL(f_setown);
258 void f_delown(struct file *filp)
260 f_modown(filp, 0, 0, 0, 1);
263 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
271 err = dupfd(filp, arg);
274 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
278 set_close_on_exec(fd, arg & FD_CLOEXEC);
284 err = setfl(fd, filp, arg);
287 err = fcntl_getlk(filp, (struct flock __user *) arg);
291 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
295 * XXX If f_owner is a process group, the
296 * negative return value will get converted
297 * into an error. Oops. If we keep the
298 * current syscall conventions, the only way
299 * to fix this will be in libc.
301 err = filp->f_owner.pid;
302 force_successful_syscall_return();
305 err = f_setown(filp, arg, 1);
308 err = filp->f_owner.signum;
311 /* arg == 0 restores default behaviour. */
312 if (!valid_signal(arg)) {
316 filp->f_owner.signum = arg;
319 err = fcntl_getlease(filp);
322 err = fcntl_setlease(fd, filp, arg);
325 err = fcntl_dirnotify(fd, filp, arg);
333 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
342 err = security_file_fcntl(filp, cmd, arg);
348 err = do_fcntl(fd, cmd, arg, filp);
355 #if BITS_PER_LONG == 32
356 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
366 err = security_file_fcntl(filp, cmd, arg);
375 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
379 err = fcntl_setlk64(fd, filp, cmd,
380 (struct flock64 __user *) arg);
383 err = do_fcntl(fd, cmd, arg, filp);
392 /* Table to convert sigio signal codes into poll band bitmaps */
394 static long band_table[NSIGPOLL] = {
395 POLLIN | POLLRDNORM, /* POLL_IN */
396 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
397 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
398 POLLERR, /* POLL_ERR */
399 POLLPRI | POLLRDBAND, /* POLL_PRI */
400 POLLHUP | POLLERR /* POLL_HUP */
403 static inline int sigio_perm(struct task_struct *p,
404 struct fown_struct *fown, int sig)
406 return (((fown->euid == 0) ||
407 (fown->euid == p->suid) || (fown->euid == p->uid) ||
408 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
409 !security_file_send_sigiotask(p, fown, sig));
412 static void send_sigio_to_task(struct task_struct *p,
413 struct fown_struct *fown,
417 if (!sigio_perm(p, fown, fown->signum))
420 switch (fown->signum) {
423 /* Queue a rt signal with the appropriate fd as its
424 value. We use SI_SIGIO as the source, not
425 SI_KERNEL, since kernel signals always get
426 delivered even if we can't queue. Failure to
427 queue in this case _should_ be reported; we fall
428 back to SIGIO in that case. --sct */
429 si.si_signo = fown->signum;
432 /* Make sure we are called with one of the POLL_*
433 reasons, otherwise we could leak kernel stack into
435 if ((reason & __SI_MASK) != __SI_POLL)
437 if (reason - POLL_IN >= NSIGPOLL)
440 si.si_band = band_table[reason - POLL_IN];
442 if (!send_group_sig_info(fown->signum, &si, p))
444 /* fall-through: fall back on the old plain SIGIO signal */
446 send_group_sig_info(SIGIO, SEND_SIG_PRIV, p);
450 void send_sigio(struct fown_struct *fown, int fd, int band)
452 struct task_struct *p;
455 read_lock(&fown->lock);
458 goto out_unlock_fown;
460 read_lock(&tasklist_lock);
462 p = find_task_by_pid(pid);
464 send_sigio_to_task(p, fown, fd, band);
467 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
468 send_sigio_to_task(p, fown, fd, band);
469 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
471 read_unlock(&tasklist_lock);
473 read_unlock(&fown->lock);
476 static void send_sigurg_to_task(struct task_struct *p,
477 struct fown_struct *fown)
479 if (sigio_perm(p, fown, SIGURG))
480 send_group_sig_info(SIGURG, SEND_SIG_PRIV, p);
483 int send_sigurg(struct fown_struct *fown)
485 struct task_struct *p;
488 read_lock(&fown->lock);
491 goto out_unlock_fown;
495 read_lock(&tasklist_lock);
497 p = find_task_by_pid(pid);
499 send_sigurg_to_task(p, fown);
502 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
503 send_sigurg_to_task(p, fown);
504 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
506 read_unlock(&tasklist_lock);
508 read_unlock(&fown->lock);
512 static DEFINE_RWLOCK(fasync_lock);
513 static kmem_cache_t *fasync_cache;
516 * fasync_helper() is used by some character device drivers (mainly mice)
517 * to set up the fasync queue. It returns negative on error, 0 if it did
518 * no changes and positive if it added/deleted the entry.
520 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
522 struct fasync_struct *fa, **fp;
523 struct fasync_struct *new = NULL;
527 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
531 write_lock_irq(&fasync_lock);
532 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
533 if (fa->fa_file == filp) {
536 kmem_cache_free(fasync_cache, new);
539 kmem_cache_free(fasync_cache, fa);
547 new->magic = FASYNC_MAGIC;
550 new->fa_next = *fapp;
555 write_unlock_irq(&fasync_lock);
559 EXPORT_SYMBOL(fasync_helper);
561 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
564 struct fown_struct * fown;
565 if (fa->magic != FASYNC_MAGIC) {
566 printk(KERN_ERR "kill_fasync: bad magic number in "
570 fown = &fa->fa_file->f_owner;
571 /* Don't send SIGURG to processes which have not set a
572 queued signum: SIGURG has its own default signalling
574 if (!(sig == SIGURG && fown->signum == 0))
575 send_sigio(fown, fa->fa_fd, band);
580 EXPORT_SYMBOL(__kill_fasync);
582 void kill_fasync(struct fasync_struct **fp, int sig, int band)
584 /* First a quick test without locking: usually
588 read_lock(&fasync_lock);
589 /* reread *fp after obtaining the lock */
590 __kill_fasync(*fp, sig, band);
591 read_unlock(&fasync_lock);
594 EXPORT_SYMBOL(kill_fasync);
596 static int __init fasync_init(void)
598 fasync_cache = kmem_cache_create("fasync_cache",
599 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
603 module_init(fasync_init)