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/capability.h>
13 #include <linux/dnotify.h>
14 #include <linux/smp_lock.h>
15 #include <linux/slab.h>
16 #include <linux/module.h>
17 #include <linux/security.h>
18 #include <linux/ptrace.h>
19 #include <linux/signal.h>
20 #include <linux/rcupdate.h>
23 #include <asm/siginfo.h>
24 #include <asm/uaccess.h>
26 void fastcall set_close_on_exec(unsigned int fd, int flag)
28 struct files_struct *files = current->files;
30 spin_lock(&files->file_lock);
31 fdt = files_fdtable(files);
33 FD_SET(fd, fdt->close_on_exec);
35 FD_CLR(fd, fdt->close_on_exec);
36 spin_unlock(&files->file_lock);
39 static int get_close_on_exec(unsigned int fd)
41 struct files_struct *files = current->files;
45 fdt = files_fdtable(files);
46 res = FD_ISSET(fd, fdt->close_on_exec);
52 * locate_fd finds a free file descriptor in the open_fds fdset,
53 * expanding the fd arrays if necessary. Must be called with the
54 * file_lock held for write.
57 static int locate_fd(struct files_struct *files,
58 struct file *file, unsigned int orig_start)
66 if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
70 fdt = files_fdtable(files);
72 * Someone might have closed fd's in the range
73 * orig_start..fdt->next_fd
76 if (start < fdt->next_fd)
80 if (start < fdt->max_fdset) {
81 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
82 fdt->max_fdset, start);
86 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
89 error = expand_files(files, newfd);
94 * If we needed to expand the fs array we
95 * might have blocked - try again.
101 * We reacquired files_lock, so we are safe as long as
102 * we reacquire the fdtable pointer and use it while holding
103 * the lock, no one can free it during that time.
105 fdt = files_fdtable(files);
106 if (start <= fdt->next_fd)
107 fdt->next_fd = newfd + 1;
115 static int dupfd(struct file *file, unsigned int start)
117 struct files_struct * files = current->files;
121 spin_lock(&files->file_lock);
122 fd = locate_fd(files, file, start);
124 /* locate_fd() may have expanded fdtable, load the ptr */
125 fdt = files_fdtable(files);
126 FD_SET(fd, fdt->open_fds);
127 FD_CLR(fd, fdt->close_on_exec);
128 spin_unlock(&files->file_lock);
129 fd_install(fd, file);
131 spin_unlock(&files->file_lock);
138 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
141 struct file * file, *tofree;
142 struct files_struct * files = current->files;
145 spin_lock(&files->file_lock);
146 if (!(file = fcheck(oldfd)))
152 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
154 get_file(file); /* We are now finished with oldfd */
156 err = expand_files(files, newfd);
160 /* To avoid races with open() and dup(), we will mark the fd as
161 * in-use in the open-file bitmap throughout the entire dup2()
162 * process. This is quite safe: do_close() uses the fd array
163 * entry, not the bitmap, to decide what work needs to be
165 /* Doesn't work. open() might be there first. --AV */
167 /* Yes. It's a race. In user space. Nothing sane to do */
169 fdt = files_fdtable(files);
170 tofree = fdt->fd[newfd];
171 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
174 rcu_assign_pointer(fdt->fd[newfd], file);
175 FD_SET(newfd, fdt->open_fds);
176 FD_CLR(newfd, fdt->close_on_exec);
177 spin_unlock(&files->file_lock);
180 filp_close(tofree, files);
185 spin_unlock(&files->file_lock);
189 spin_unlock(&files->file_lock);
194 asmlinkage long sys_dup(unsigned int fildes)
197 struct file * file = fget(fildes);
200 ret = dupfd(file, 0);
204 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
206 static int setfl(int fd, struct file * filp, unsigned long arg)
208 struct inode * inode = filp->f_dentry->d_inode;
212 * O_APPEND cannot be cleared if the file is marked as append-only
213 * and the file is open for write.
215 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
218 /* O_NOATIME can only be set by the owner or superuser */
219 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
220 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
223 /* required for strict SunOS emulation */
224 if (O_NONBLOCK != O_NDELAY)
228 if (arg & O_DIRECT) {
229 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
230 !filp->f_mapping->a_ops->direct_IO)
234 if (filp->f_op && filp->f_op->check_flags)
235 error = filp->f_op->check_flags(arg);
240 if ((arg ^ filp->f_flags) & FASYNC) {
241 if (filp->f_op && filp->f_op->fasync) {
242 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
248 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
254 static void f_modown(struct file *filp, unsigned long pid,
255 uid_t uid, uid_t euid, int force)
257 write_lock_irq(&filp->f_owner.lock);
258 if (force || !filp->f_owner.pid) {
259 filp->f_owner.pid = pid;
260 filp->f_owner.uid = uid;
261 filp->f_owner.euid = euid;
263 write_unlock_irq(&filp->f_owner.lock);
266 int f_setown(struct file *filp, unsigned long arg, int force)
270 err = security_file_set_fowner(filp);
274 f_modown(filp, arg, current->uid, current->euid, force);
278 EXPORT_SYMBOL(f_setown);
280 void f_delown(struct file *filp)
282 f_modown(filp, 0, 0, 0, 1);
285 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
293 err = dupfd(filp, arg);
296 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
300 set_close_on_exec(fd, arg & FD_CLOEXEC);
306 err = setfl(fd, filp, arg);
309 err = fcntl_getlk(filp, (struct flock __user *) arg);
313 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
317 * XXX If f_owner is a process group, the
318 * negative return value will get converted
319 * into an error. Oops. If we keep the
320 * current syscall conventions, the only way
321 * to fix this will be in libc.
323 err = filp->f_owner.pid;
324 force_successful_syscall_return();
327 err = f_setown(filp, arg, 1);
330 err = filp->f_owner.signum;
333 /* arg == 0 restores default behaviour. */
334 if (!valid_signal(arg)) {
338 filp->f_owner.signum = arg;
341 err = fcntl_getlease(filp);
344 err = fcntl_setlease(fd, filp, arg);
347 err = fcntl_dirnotify(fd, filp, arg);
355 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
364 err = security_file_fcntl(filp, cmd, arg);
370 err = do_fcntl(fd, cmd, arg, filp);
377 #if BITS_PER_LONG == 32
378 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
388 err = security_file_fcntl(filp, cmd, arg);
397 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
401 err = fcntl_setlk64(fd, filp, cmd,
402 (struct flock64 __user *) arg);
405 err = do_fcntl(fd, cmd, arg, filp);
414 /* Table to convert sigio signal codes into poll band bitmaps */
416 static long band_table[NSIGPOLL] = {
417 POLLIN | POLLRDNORM, /* POLL_IN */
418 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
419 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
420 POLLERR, /* POLL_ERR */
421 POLLPRI | POLLRDBAND, /* POLL_PRI */
422 POLLHUP | POLLERR /* POLL_HUP */
425 static inline int sigio_perm(struct task_struct *p,
426 struct fown_struct *fown, int sig)
428 return (((fown->euid == 0) ||
429 (fown->euid == p->suid) || (fown->euid == p->uid) ||
430 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
431 !security_file_send_sigiotask(p, fown, sig));
434 static void send_sigio_to_task(struct task_struct *p,
435 struct fown_struct *fown,
439 if (!sigio_perm(p, fown, fown->signum))
442 switch (fown->signum) {
445 /* Queue a rt signal with the appropriate fd as its
446 value. We use SI_SIGIO as the source, not
447 SI_KERNEL, since kernel signals always get
448 delivered even if we can't queue. Failure to
449 queue in this case _should_ be reported; we fall
450 back to SIGIO in that case. --sct */
451 si.si_signo = fown->signum;
454 /* Make sure we are called with one of the POLL_*
455 reasons, otherwise we could leak kernel stack into
457 if ((reason & __SI_MASK) != __SI_POLL)
459 if (reason - POLL_IN >= NSIGPOLL)
462 si.si_band = band_table[reason - POLL_IN];
464 if (!group_send_sig_info(fown->signum, &si, p))
466 /* fall-through: fall back on the old plain SIGIO signal */
468 group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
472 void send_sigio(struct fown_struct *fown, int fd, int band)
474 struct task_struct *p;
477 read_lock(&fown->lock);
480 goto out_unlock_fown;
482 read_lock(&tasklist_lock);
484 p = find_task_by_pid(pid);
486 send_sigio_to_task(p, fown, fd, band);
489 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
490 send_sigio_to_task(p, fown, fd, band);
491 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
493 read_unlock(&tasklist_lock);
495 read_unlock(&fown->lock);
498 static void send_sigurg_to_task(struct task_struct *p,
499 struct fown_struct *fown)
501 if (sigio_perm(p, fown, SIGURG))
502 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
505 int send_sigurg(struct fown_struct *fown)
507 struct task_struct *p;
510 read_lock(&fown->lock);
513 goto out_unlock_fown;
517 read_lock(&tasklist_lock);
519 p = find_task_by_pid(pid);
521 send_sigurg_to_task(p, fown);
524 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
525 send_sigurg_to_task(p, fown);
526 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
528 read_unlock(&tasklist_lock);
530 read_unlock(&fown->lock);
534 static DEFINE_RWLOCK(fasync_lock);
535 static kmem_cache_t *fasync_cache;
538 * fasync_helper() is used by some character device drivers (mainly mice)
539 * to set up the fasync queue. It returns negative on error, 0 if it did
540 * no changes and positive if it added/deleted the entry.
542 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
544 struct fasync_struct *fa, **fp;
545 struct fasync_struct *new = NULL;
549 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
553 write_lock_irq(&fasync_lock);
554 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
555 if (fa->fa_file == filp) {
558 kmem_cache_free(fasync_cache, new);
561 kmem_cache_free(fasync_cache, fa);
569 new->magic = FASYNC_MAGIC;
572 new->fa_next = *fapp;
577 write_unlock_irq(&fasync_lock);
581 EXPORT_SYMBOL(fasync_helper);
583 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
586 struct fown_struct * fown;
587 if (fa->magic != FASYNC_MAGIC) {
588 printk(KERN_ERR "kill_fasync: bad magic number in "
592 fown = &fa->fa_file->f_owner;
593 /* Don't send SIGURG to processes which have not set a
594 queued signum: SIGURG has its own default signalling
596 if (!(sig == SIGURG && fown->signum == 0))
597 send_sigio(fown, fa->fa_fd, band);
602 EXPORT_SYMBOL(__kill_fasync);
604 void kill_fasync(struct fasync_struct **fp, int sig, int band)
606 /* First a quick test without locking: usually
610 read_lock(&fasync_lock);
611 /* reread *fp after obtaining the lock */
612 __kill_fasync(*fp, sig, band);
613 read_unlock(&fasync_lock);
616 EXPORT_SYMBOL(kill_fasync);
618 static int __init fasync_init(void)
620 fasync_cache = kmem_cache_create("fasync_cache",
621 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
625 module_init(fasync_init)