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>
21 #include <linux/vs_base.h>
22 #include <linux/vs_limit.h>
25 #include <asm/siginfo.h>
26 #include <asm/uaccess.h>
28 void fastcall set_close_on_exec(unsigned int fd, int flag)
30 struct files_struct *files = current->files;
32 spin_lock(&files->file_lock);
33 fdt = files_fdtable(files);
35 FD_SET(fd, fdt->close_on_exec);
37 FD_CLR(fd, fdt->close_on_exec);
38 spin_unlock(&files->file_lock);
41 static int get_close_on_exec(unsigned int fd)
43 struct files_struct *files = current->files;
47 fdt = files_fdtable(files);
48 res = FD_ISSET(fd, fdt->close_on_exec);
54 * locate_fd finds a free file descriptor in the open_fds fdset,
55 * expanding the fd arrays if necessary. Must be called with the
56 * file_lock held for write.
59 static int locate_fd(struct files_struct *files,
60 struct file *file, unsigned int orig_start)
68 if (orig_start >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
72 fdt = files_fdtable(files);
74 * Someone might have closed fd's in the range
75 * orig_start..fdt->next_fd
78 if (start < files->next_fd)
79 start = files->next_fd;
82 if (start < fdt->max_fdset) {
83 newfd = find_next_zero_bit(fdt->open_fds->fds_bits,
84 fdt->max_fdset, start);
88 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
90 if (!vx_files_avail(1))
93 error = expand_files(files, newfd);
98 * If we needed to expand the fs array we
99 * might have blocked - try again.
105 * We reacquired files_lock, so we are safe as long as
106 * we reacquire the fdtable pointer and use it while holding
107 * the lock, no one can free it during that time.
109 if (start <= files->next_fd)
110 files->next_fd = newfd + 1;
118 int dupfd(struct file *file, unsigned int start)
120 struct files_struct * files = current->files;
124 spin_lock(&files->file_lock);
125 fd = locate_fd(files, file, start);
127 /* locate_fd() may have expanded fdtable, load the ptr */
128 fdt = files_fdtable(files);
129 FD_SET(fd, fdt->open_fds);
130 FD_CLR(fd, fdt->close_on_exec);
131 spin_unlock(&files->file_lock);
133 fd_install(fd, file);
135 spin_unlock(&files->file_lock);
142 EXPORT_SYMBOL_GPL(dupfd);
144 asmlinkage long sys_dup2(unsigned int oldfd, unsigned int newfd)
147 struct file * file, *tofree;
148 struct files_struct * files = current->files;
151 spin_lock(&files->file_lock);
152 if (!(file = fcheck(oldfd)))
158 if (newfd >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
160 get_file(file); /* We are now finished with oldfd */
162 err = expand_files(files, newfd);
166 /* To avoid races with open() and dup(), we will mark the fd as
167 * in-use in the open-file bitmap throughout the entire dup2()
168 * process. This is quite safe: do_close() uses the fd array
169 * entry, not the bitmap, to decide what work needs to be
171 /* Doesn't work. open() might be there first. --AV */
173 /* Yes. It's a race. In user space. Nothing sane to do */
175 fdt = files_fdtable(files);
176 tofree = fdt->fd[newfd];
177 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
180 rcu_assign_pointer(fdt->fd[newfd], file);
181 FD_SET(newfd, fdt->open_fds);
182 FD_CLR(newfd, fdt->close_on_exec);
183 spin_unlock(&files->file_lock);
186 filp_close(tofree, files);
188 vx_openfd_inc(newfd); /* fd was unused */
194 spin_unlock(&files->file_lock);
198 spin_unlock(&files->file_lock);
203 asmlinkage long sys_dup(unsigned int fildes)
206 struct file * file = fget(fildes);
209 ret = dupfd(file, 0);
213 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | FASYNC | O_DIRECT | O_NOATIME)
215 static int setfl(int fd, struct file * filp, unsigned long arg)
217 struct inode * inode = filp->f_dentry->d_inode;
221 * O_APPEND cannot be cleared if the file is marked as append-only
222 * and the file is open for write.
224 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
227 /* O_NOATIME can only be set by the owner or superuser */
228 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
229 if (current->fsuid != inode->i_uid && !capable(CAP_FOWNER))
232 /* required for strict SunOS emulation */
233 if (O_NONBLOCK != O_NDELAY)
237 if (arg & O_DIRECT) {
238 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
239 !filp->f_mapping->a_ops->direct_IO)
243 if (filp->f_op && filp->f_op->check_flags)
244 error = filp->f_op->check_flags(arg);
249 if ((arg ^ filp->f_flags) & FASYNC) {
250 if (filp->f_op && filp->f_op->fasync) {
251 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
257 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
263 static void f_modown(struct file *filp, unsigned long pid,
264 uid_t uid, uid_t euid, int force)
266 write_lock_irq(&filp->f_owner.lock);
267 if (force || !filp->f_owner.pid) {
268 filp->f_owner.pid = pid;
269 filp->f_owner.uid = uid;
270 filp->f_owner.euid = euid;
272 write_unlock_irq(&filp->f_owner.lock);
275 int f_setown(struct file *filp, unsigned long arg, int force)
279 err = security_file_set_fowner(filp);
283 f_modown(filp, arg, current->uid, current->euid, force);
287 EXPORT_SYMBOL(f_setown);
289 void f_delown(struct file *filp)
291 f_modown(filp, 0, 0, 0, 1);
294 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
302 err = dupfd(filp, arg);
305 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
309 set_close_on_exec(fd, arg & FD_CLOEXEC);
315 err = setfl(fd, filp, arg);
318 err = fcntl_getlk(filp, (struct flock __user *) arg);
322 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
326 * XXX If f_owner is a process group, the
327 * negative return value will get converted
328 * into an error. Oops. If we keep the
329 * current syscall conventions, the only way
330 * to fix this will be in libc.
332 err = filp->f_owner.pid;
333 force_successful_syscall_return();
336 err = f_setown(filp, arg, 1);
339 err = filp->f_owner.signum;
342 /* arg == 0 restores default behaviour. */
343 if (!valid_signal(arg)) {
347 filp->f_owner.signum = arg;
350 err = fcntl_getlease(filp);
353 err = fcntl_setlease(fd, filp, arg);
356 err = fcntl_dirnotify(fd, filp, arg);
364 asmlinkage long sys_fcntl(unsigned int fd, unsigned int cmd, unsigned long arg)
373 err = security_file_fcntl(filp, cmd, arg);
379 err = do_fcntl(fd, cmd, arg, filp);
386 #if BITS_PER_LONG == 32
387 asmlinkage long sys_fcntl64(unsigned int fd, unsigned int cmd, unsigned long arg)
397 err = security_file_fcntl(filp, cmd, arg);
406 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
410 err = fcntl_setlk64(fd, filp, cmd,
411 (struct flock64 __user *) arg);
414 err = do_fcntl(fd, cmd, arg, filp);
423 /* Table to convert sigio signal codes into poll band bitmaps */
425 static const long band_table[NSIGPOLL] = {
426 POLLIN | POLLRDNORM, /* POLL_IN */
427 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
428 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
429 POLLERR, /* POLL_ERR */
430 POLLPRI | POLLRDBAND, /* POLL_PRI */
431 POLLHUP | POLLERR /* POLL_HUP */
434 static inline int sigio_perm(struct task_struct *p,
435 struct fown_struct *fown, int sig)
437 return (((fown->euid == 0) ||
438 (fown->euid == p->suid) || (fown->euid == p->uid) ||
439 (fown->uid == p->suid) || (fown->uid == p->uid)) &&
440 !security_file_send_sigiotask(p, fown, sig));
443 static void send_sigio_to_task(struct task_struct *p,
444 struct fown_struct *fown,
448 if (!sigio_perm(p, fown, fown->signum))
451 switch (fown->signum) {
454 /* Queue a rt signal with the appropriate fd as its
455 value. We use SI_SIGIO as the source, not
456 SI_KERNEL, since kernel signals always get
457 delivered even if we can't queue. Failure to
458 queue in this case _should_ be reported; we fall
459 back to SIGIO in that case. --sct */
460 si.si_signo = fown->signum;
463 /* Make sure we are called with one of the POLL_*
464 reasons, otherwise we could leak kernel stack into
466 BUG_ON((reason & __SI_MASK) != __SI_POLL);
467 if (reason - POLL_IN >= NSIGPOLL)
470 si.si_band = band_table[reason - POLL_IN];
472 if (!group_send_sig_info(fown->signum, &si, p))
474 /* fall-through: fall back on the old plain SIGIO signal */
476 group_send_sig_info(SIGIO, SEND_SIG_PRIV, p);
480 void send_sigio(struct fown_struct *fown, int fd, int band)
482 struct task_struct *p;
485 read_lock(&fown->lock);
488 goto out_unlock_fown;
490 read_lock(&tasklist_lock);
492 p = find_task_by_real_pid(pid);
494 send_sigio_to_task(p, fown, fd, band);
497 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
498 send_sigio_to_task(p, fown, fd, band);
499 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
501 read_unlock(&tasklist_lock);
503 read_unlock(&fown->lock);
506 static void send_sigurg_to_task(struct task_struct *p,
507 struct fown_struct *fown)
509 if (sigio_perm(p, fown, SIGURG))
510 group_send_sig_info(SIGURG, SEND_SIG_PRIV, p);
513 int send_sigurg(struct fown_struct *fown)
515 struct task_struct *p;
518 read_lock(&fown->lock);
521 goto out_unlock_fown;
525 read_lock(&tasklist_lock);
527 p = find_task_by_real_pid(pid);
529 send_sigurg_to_task(p, fown);
532 do_each_task_pid(-pid, PIDTYPE_PGID, p) {
533 send_sigurg_to_task(p, fown);
534 } while_each_task_pid(-pid, PIDTYPE_PGID, p);
536 read_unlock(&tasklist_lock);
538 read_unlock(&fown->lock);
542 static DEFINE_RWLOCK(fasync_lock);
543 static kmem_cache_t *fasync_cache __read_mostly;
546 * fasync_helper() is used by some character device drivers (mainly mice)
547 * to set up the fasync queue. It returns negative on error, 0 if it did
548 * no changes and positive if it added/deleted the entry.
550 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
552 struct fasync_struct *fa, **fp;
553 struct fasync_struct *new = NULL;
557 new = kmem_cache_alloc(fasync_cache, SLAB_KERNEL);
561 write_lock_irq(&fasync_lock);
562 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
563 if (fa->fa_file == filp) {
566 kmem_cache_free(fasync_cache, new);
569 kmem_cache_free(fasync_cache, fa);
577 new->magic = FASYNC_MAGIC;
580 new->fa_next = *fapp;
585 write_unlock_irq(&fasync_lock);
589 EXPORT_SYMBOL(fasync_helper);
591 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
594 struct fown_struct * fown;
595 if (fa->magic != FASYNC_MAGIC) {
596 printk(KERN_ERR "kill_fasync: bad magic number in "
600 fown = &fa->fa_file->f_owner;
601 /* Don't send SIGURG to processes which have not set a
602 queued signum: SIGURG has its own default signalling
604 if (!(sig == SIGURG && fown->signum == 0))
605 send_sigio(fown, fa->fa_fd, band);
610 EXPORT_SYMBOL(__kill_fasync);
612 void kill_fasync(struct fasync_struct **fp, int sig, int band)
614 /* First a quick test without locking: usually
618 read_lock(&fasync_lock);
619 /* reread *fp after obtaining the lock */
620 __kill_fasync(*fp, sig, band);
621 read_unlock(&fasync_lock);
624 EXPORT_SYMBOL(kill_fasync);
626 static int __init fasync_init(void)
628 fasync_cache = kmem_cache_create("fasync_cache",
629 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL, NULL);
633 module_init(fasync_init)