percpu: add __percpu for sparse
[linux-2.6/kvm.git] / fs / fcntl.c
blob5ef953e6f908ea2d6f68061310a106fd0179d15a
1 /*
2 * linux/fs/fcntl.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 */
7 #include <linux/syscalls.h>
8 #include <linux/init.h>
9 #include <linux/mm.h>
10 #include <linux/fs.h>
11 #include <linux/file.h>
12 #include <linux/fdtable.h>
13 #include <linux/capability.h>
14 #include <linux/dnotify.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/pid_namespace.h>
23 #include <asm/poll.h>
24 #include <asm/siginfo.h>
25 #include <asm/uaccess.h>
27 void set_close_on_exec(unsigned int fd, int flag)
29 struct files_struct *files = current->files;
30 struct fdtable *fdt;
31 spin_lock(&files->file_lock);
32 fdt = files_fdtable(files);
33 if (flag)
34 FD_SET(fd, fdt->close_on_exec);
35 else
36 FD_CLR(fd, fdt->close_on_exec);
37 spin_unlock(&files->file_lock);
40 static int get_close_on_exec(unsigned int fd)
42 struct files_struct *files = current->files;
43 struct fdtable *fdt;
44 int res;
45 rcu_read_lock();
46 fdt = files_fdtable(files);
47 res = FD_ISSET(fd, fdt->close_on_exec);
48 rcu_read_unlock();
49 return res;
52 SYSCALL_DEFINE3(dup3, unsigned int, oldfd, unsigned int, newfd, int, flags)
54 int err = -EBADF;
55 struct file * file, *tofree;
56 struct files_struct * files = current->files;
57 struct fdtable *fdt;
59 if ((flags & ~O_CLOEXEC) != 0)
60 return -EINVAL;
62 if (unlikely(oldfd == newfd))
63 return -EINVAL;
65 spin_lock(&files->file_lock);
66 err = expand_files(files, newfd);
67 file = fcheck(oldfd);
68 if (unlikely(!file))
69 goto Ebadf;
70 if (unlikely(err < 0)) {
71 if (err == -EMFILE)
72 goto Ebadf;
73 goto out_unlock;
76 * We need to detect attempts to do dup2() over allocated but still
77 * not finished descriptor. NB: OpenBSD avoids that at the price of
78 * extra work in their equivalent of fget() - they insert struct
79 * file immediately after grabbing descriptor, mark it larval if
80 * more work (e.g. actual opening) is needed and make sure that
81 * fget() treats larval files as absent. Potentially interesting,
82 * but while extra work in fget() is trivial, locking implications
83 * and amount of surgery on open()-related paths in VFS are not.
84 * FreeBSD fails with -EBADF in the same situation, NetBSD "solution"
85 * deadlocks in rather amusing ways, AFAICS. All of that is out of
86 * scope of POSIX or SUS, since neither considers shared descriptor
87 * tables and this condition does not arise without those.
89 err = -EBUSY;
90 fdt = files_fdtable(files);
91 tofree = fdt->fd[newfd];
92 if (!tofree && FD_ISSET(newfd, fdt->open_fds))
93 goto out_unlock;
94 get_file(file);
95 rcu_assign_pointer(fdt->fd[newfd], file);
96 FD_SET(newfd, fdt->open_fds);
97 if (flags & O_CLOEXEC)
98 FD_SET(newfd, fdt->close_on_exec);
99 else
100 FD_CLR(newfd, fdt->close_on_exec);
101 spin_unlock(&files->file_lock);
103 if (tofree)
104 filp_close(tofree, files);
106 return newfd;
108 Ebadf:
109 err = -EBADF;
110 out_unlock:
111 spin_unlock(&files->file_lock);
112 return err;
115 SYSCALL_DEFINE2(dup2, unsigned int, oldfd, unsigned int, newfd)
117 if (unlikely(newfd == oldfd)) { /* corner case */
118 struct files_struct *files = current->files;
119 int retval = oldfd;
121 rcu_read_lock();
122 if (!fcheck_files(files, oldfd))
123 retval = -EBADF;
124 rcu_read_unlock();
125 return retval;
127 return sys_dup3(oldfd, newfd, 0);
130 SYSCALL_DEFINE1(dup, unsigned int, fildes)
132 int ret = -EBADF;
133 struct file *file = fget(fildes);
135 if (file) {
136 ret = get_unused_fd();
137 if (ret >= 0)
138 fd_install(ret, file);
139 else
140 fput(file);
142 return ret;
145 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
147 static int setfl(int fd, struct file * filp, unsigned long arg)
149 struct inode * inode = filp->f_path.dentry->d_inode;
150 int error = 0;
153 * O_APPEND cannot be cleared if the file is marked as append-only
154 * and the file is open for write.
156 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
157 return -EPERM;
159 /* O_NOATIME can only be set by the owner or superuser */
160 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
161 if (!is_owner_or_cap(inode))
162 return -EPERM;
164 /* required for strict SunOS emulation */
165 if (O_NONBLOCK != O_NDELAY)
166 if (arg & O_NDELAY)
167 arg |= O_NONBLOCK;
169 if (arg & O_DIRECT) {
170 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
171 !filp->f_mapping->a_ops->direct_IO)
172 return -EINVAL;
175 if (filp->f_op && filp->f_op->check_flags)
176 error = filp->f_op->check_flags(arg);
177 if (error)
178 return error;
181 * ->fasync() is responsible for setting the FASYNC bit.
183 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op &&
184 filp->f_op->fasync) {
185 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
186 if (error < 0)
187 goto out;
188 if (error > 0)
189 error = 0;
191 spin_lock(&filp->f_lock);
192 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
193 spin_unlock(&filp->f_lock);
195 out:
196 return error;
199 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
200 int force)
202 unsigned long flags;
204 write_lock_irqsave(&filp->f_owner.lock, flags);
205 if (force || !filp->f_owner.pid) {
206 put_pid(filp->f_owner.pid);
207 filp->f_owner.pid = get_pid(pid);
208 filp->f_owner.pid_type = type;
210 if (pid) {
211 const struct cred *cred = current_cred();
212 filp->f_owner.uid = cred->uid;
213 filp->f_owner.euid = cred->euid;
216 write_unlock_irqrestore(&filp->f_owner.lock, flags);
219 int __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
220 int force)
222 int err;
224 err = security_file_set_fowner(filp);
225 if (err)
226 return err;
228 f_modown(filp, pid, type, force);
229 return 0;
231 EXPORT_SYMBOL(__f_setown);
233 int f_setown(struct file *filp, unsigned long arg, int force)
235 enum pid_type type;
236 struct pid *pid;
237 int who = arg;
238 int result;
239 type = PIDTYPE_PID;
240 if (who < 0) {
241 type = PIDTYPE_PGID;
242 who = -who;
244 rcu_read_lock();
245 pid = find_vpid(who);
246 result = __f_setown(filp, pid, type, force);
247 rcu_read_unlock();
248 return result;
250 EXPORT_SYMBOL(f_setown);
252 void f_delown(struct file *filp)
254 f_modown(filp, NULL, PIDTYPE_PID, 1);
257 pid_t f_getown(struct file *filp)
259 pid_t pid;
260 read_lock(&filp->f_owner.lock);
261 pid = pid_vnr(filp->f_owner.pid);
262 if (filp->f_owner.pid_type == PIDTYPE_PGID)
263 pid = -pid;
264 read_unlock(&filp->f_owner.lock);
265 return pid;
268 static int f_setown_ex(struct file *filp, unsigned long arg)
270 struct f_owner_ex * __user owner_p = (void * __user)arg;
271 struct f_owner_ex owner;
272 struct pid *pid;
273 int type;
274 int ret;
276 ret = copy_from_user(&owner, owner_p, sizeof(owner));
277 if (ret)
278 return ret;
280 switch (owner.type) {
281 case F_OWNER_TID:
282 type = PIDTYPE_MAX;
283 break;
285 case F_OWNER_PID:
286 type = PIDTYPE_PID;
287 break;
289 case F_OWNER_PGRP:
290 type = PIDTYPE_PGID;
291 break;
293 default:
294 return -EINVAL;
297 rcu_read_lock();
298 pid = find_vpid(owner.pid);
299 if (owner.pid && !pid)
300 ret = -ESRCH;
301 else
302 ret = __f_setown(filp, pid, type, 1);
303 rcu_read_unlock();
305 return ret;
308 static int f_getown_ex(struct file *filp, unsigned long arg)
310 struct f_owner_ex * __user owner_p = (void * __user)arg;
311 struct f_owner_ex owner;
312 int ret = 0;
314 read_lock(&filp->f_owner.lock);
315 owner.pid = pid_vnr(filp->f_owner.pid);
316 switch (filp->f_owner.pid_type) {
317 case PIDTYPE_MAX:
318 owner.type = F_OWNER_TID;
319 break;
321 case PIDTYPE_PID:
322 owner.type = F_OWNER_PID;
323 break;
325 case PIDTYPE_PGID:
326 owner.type = F_OWNER_PGRP;
327 break;
329 default:
330 WARN_ON(1);
331 ret = -EINVAL;
332 break;
334 read_unlock(&filp->f_owner.lock);
336 if (!ret)
337 ret = copy_to_user(owner_p, &owner, sizeof(owner));
338 return ret;
341 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
342 struct file *filp)
344 long err = -EINVAL;
346 switch (cmd) {
347 case F_DUPFD:
348 case F_DUPFD_CLOEXEC:
349 if (arg >= current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
350 break;
351 err = alloc_fd(arg, cmd == F_DUPFD_CLOEXEC ? O_CLOEXEC : 0);
352 if (err >= 0) {
353 get_file(filp);
354 fd_install(err, filp);
356 break;
357 case F_GETFD:
358 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
359 break;
360 case F_SETFD:
361 err = 0;
362 set_close_on_exec(fd, arg & FD_CLOEXEC);
363 break;
364 case F_GETFL:
365 err = filp->f_flags;
366 break;
367 case F_SETFL:
368 err = setfl(fd, filp, arg);
369 break;
370 case F_GETLK:
371 err = fcntl_getlk(filp, (struct flock __user *) arg);
372 break;
373 case F_SETLK:
374 case F_SETLKW:
375 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
376 break;
377 case F_GETOWN:
379 * XXX If f_owner is a process group, the
380 * negative return value will get converted
381 * into an error. Oops. If we keep the
382 * current syscall conventions, the only way
383 * to fix this will be in libc.
385 err = f_getown(filp);
386 force_successful_syscall_return();
387 break;
388 case F_SETOWN:
389 err = f_setown(filp, arg, 1);
390 break;
391 case F_GETOWN_EX:
392 err = f_getown_ex(filp, arg);
393 break;
394 case F_SETOWN_EX:
395 err = f_setown_ex(filp, arg);
396 break;
397 case F_GETSIG:
398 err = filp->f_owner.signum;
399 break;
400 case F_SETSIG:
401 /* arg == 0 restores default behaviour. */
402 if (!valid_signal(arg)) {
403 break;
405 err = 0;
406 filp->f_owner.signum = arg;
407 break;
408 case F_GETLEASE:
409 err = fcntl_getlease(filp);
410 break;
411 case F_SETLEASE:
412 err = fcntl_setlease(fd, filp, arg);
413 break;
414 case F_NOTIFY:
415 err = fcntl_dirnotify(fd, filp, arg);
416 break;
417 default:
418 break;
420 return err;
423 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
425 struct file *filp;
426 long err = -EBADF;
428 filp = fget(fd);
429 if (!filp)
430 goto out;
432 err = security_file_fcntl(filp, cmd, arg);
433 if (err) {
434 fput(filp);
435 return err;
438 err = do_fcntl(fd, cmd, arg, filp);
440 fput(filp);
441 out:
442 return err;
445 #if BITS_PER_LONG == 32
446 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
447 unsigned long, arg)
449 struct file * filp;
450 long err;
452 err = -EBADF;
453 filp = fget(fd);
454 if (!filp)
455 goto out;
457 err = security_file_fcntl(filp, cmd, arg);
458 if (err) {
459 fput(filp);
460 return err;
462 err = -EBADF;
464 switch (cmd) {
465 case F_GETLK64:
466 err = fcntl_getlk64(filp, (struct flock64 __user *) arg);
467 break;
468 case F_SETLK64:
469 case F_SETLKW64:
470 err = fcntl_setlk64(fd, filp, cmd,
471 (struct flock64 __user *) arg);
472 break;
473 default:
474 err = do_fcntl(fd, cmd, arg, filp);
475 break;
477 fput(filp);
478 out:
479 return err;
481 #endif
483 /* Table to convert sigio signal codes into poll band bitmaps */
485 static const long band_table[NSIGPOLL] = {
486 POLLIN | POLLRDNORM, /* POLL_IN */
487 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
488 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
489 POLLERR, /* POLL_ERR */
490 POLLPRI | POLLRDBAND, /* POLL_PRI */
491 POLLHUP | POLLERR /* POLL_HUP */
494 static inline int sigio_perm(struct task_struct *p,
495 struct fown_struct *fown, int sig)
497 const struct cred *cred;
498 int ret;
500 rcu_read_lock();
501 cred = __task_cred(p);
502 ret = ((fown->euid == 0 ||
503 fown->euid == cred->suid || fown->euid == cred->uid ||
504 fown->uid == cred->suid || fown->uid == cred->uid) &&
505 !security_file_send_sigiotask(p, fown, sig));
506 rcu_read_unlock();
507 return ret;
510 static void send_sigio_to_task(struct task_struct *p,
511 struct fown_struct *fown,
512 int fd, int reason, int group)
515 * F_SETSIG can change ->signum lockless in parallel, make
516 * sure we read it once and use the same value throughout.
518 int signum = ACCESS_ONCE(fown->signum);
520 if (!sigio_perm(p, fown, signum))
521 return;
523 switch (signum) {
524 siginfo_t si;
525 default:
526 /* Queue a rt signal with the appropriate fd as its
527 value. We use SI_SIGIO as the source, not
528 SI_KERNEL, since kernel signals always get
529 delivered even if we can't queue. Failure to
530 queue in this case _should_ be reported; we fall
531 back to SIGIO in that case. --sct */
532 si.si_signo = signum;
533 si.si_errno = 0;
534 si.si_code = reason;
535 /* Make sure we are called with one of the POLL_*
536 reasons, otherwise we could leak kernel stack into
537 userspace. */
538 BUG_ON((reason & __SI_MASK) != __SI_POLL);
539 if (reason - POLL_IN >= NSIGPOLL)
540 si.si_band = ~0L;
541 else
542 si.si_band = band_table[reason - POLL_IN];
543 si.si_fd = fd;
544 if (!do_send_sig_info(signum, &si, p, group))
545 break;
546 /* fall-through: fall back on the old plain SIGIO signal */
547 case 0:
548 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
552 void send_sigio(struct fown_struct *fown, int fd, int band)
554 struct task_struct *p;
555 enum pid_type type;
556 struct pid *pid;
557 int group = 1;
559 read_lock(&fown->lock);
561 type = fown->pid_type;
562 if (type == PIDTYPE_MAX) {
563 group = 0;
564 type = PIDTYPE_PID;
567 pid = fown->pid;
568 if (!pid)
569 goto out_unlock_fown;
571 read_lock(&tasklist_lock);
572 do_each_pid_task(pid, type, p) {
573 send_sigio_to_task(p, fown, fd, band, group);
574 } while_each_pid_task(pid, type, p);
575 read_unlock(&tasklist_lock);
576 out_unlock_fown:
577 read_unlock(&fown->lock);
580 static void send_sigurg_to_task(struct task_struct *p,
581 struct fown_struct *fown, int group)
583 if (sigio_perm(p, fown, SIGURG))
584 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
587 int send_sigurg(struct fown_struct *fown)
589 struct task_struct *p;
590 enum pid_type type;
591 struct pid *pid;
592 int group = 1;
593 int ret = 0;
595 read_lock(&fown->lock);
597 type = fown->pid_type;
598 if (type == PIDTYPE_MAX) {
599 group = 0;
600 type = PIDTYPE_PID;
603 pid = fown->pid;
604 if (!pid)
605 goto out_unlock_fown;
607 ret = 1;
609 read_lock(&tasklist_lock);
610 do_each_pid_task(pid, type, p) {
611 send_sigurg_to_task(p, fown, group);
612 } while_each_pid_task(pid, type, p);
613 read_unlock(&tasklist_lock);
614 out_unlock_fown:
615 read_unlock(&fown->lock);
616 return ret;
619 static DEFINE_RWLOCK(fasync_lock);
620 static struct kmem_cache *fasync_cache __read_mostly;
623 * Remove a fasync entry. If successfully removed, return
624 * positive and clear the FASYNC flag. If no entry exists,
625 * do nothing and return 0.
627 * NOTE! It is very important that the FASYNC flag always
628 * match the state "is the filp on a fasync list".
630 * We always take the 'filp->f_lock', in since fasync_lock
631 * needs to be irq-safe.
633 static int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
635 struct fasync_struct *fa, **fp;
636 int result = 0;
638 spin_lock(&filp->f_lock);
639 write_lock_irq(&fasync_lock);
640 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
641 if (fa->fa_file != filp)
642 continue;
643 *fp = fa->fa_next;
644 kmem_cache_free(fasync_cache, fa);
645 filp->f_flags &= ~FASYNC;
646 result = 1;
647 break;
649 write_unlock_irq(&fasync_lock);
650 spin_unlock(&filp->f_lock);
651 return result;
655 * Add a fasync entry. Return negative on error, positive if
656 * added, and zero if did nothing but change an existing one.
658 * NOTE! It is very important that the FASYNC flag always
659 * match the state "is the filp on a fasync list".
661 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
663 struct fasync_struct *new, *fa, **fp;
664 int result = 0;
666 new = kmem_cache_alloc(fasync_cache, GFP_KERNEL);
667 if (!new)
668 return -ENOMEM;
670 spin_lock(&filp->f_lock);
671 write_lock_irq(&fasync_lock);
672 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
673 if (fa->fa_file != filp)
674 continue;
675 fa->fa_fd = fd;
676 kmem_cache_free(fasync_cache, new);
677 goto out;
680 new->magic = FASYNC_MAGIC;
681 new->fa_file = filp;
682 new->fa_fd = fd;
683 new->fa_next = *fapp;
684 *fapp = new;
685 result = 1;
686 filp->f_flags |= FASYNC;
688 out:
689 write_unlock_irq(&fasync_lock);
690 spin_unlock(&filp->f_lock);
691 return result;
695 * fasync_helper() is used by almost all character device drivers
696 * to set up the fasync queue, and for regular files by the file
697 * lease code. It returns negative on error, 0 if it did no changes
698 * and positive if it added/deleted the entry.
700 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
702 if (!on)
703 return fasync_remove_entry(filp, fapp);
704 return fasync_add_entry(fd, filp, fapp);
707 EXPORT_SYMBOL(fasync_helper);
709 void __kill_fasync(struct fasync_struct *fa, int sig, int band)
711 while (fa) {
712 struct fown_struct * fown;
713 if (fa->magic != FASYNC_MAGIC) {
714 printk(KERN_ERR "kill_fasync: bad magic number in "
715 "fasync_struct!\n");
716 return;
718 fown = &fa->fa_file->f_owner;
719 /* Don't send SIGURG to processes which have not set a
720 queued signum: SIGURG has its own default signalling
721 mechanism. */
722 if (!(sig == SIGURG && fown->signum == 0))
723 send_sigio(fown, fa->fa_fd, band);
724 fa = fa->fa_next;
728 EXPORT_SYMBOL(__kill_fasync);
730 void kill_fasync(struct fasync_struct **fp, int sig, int band)
732 /* First a quick test without locking: usually
733 * the list is empty.
735 if (*fp) {
736 read_lock(&fasync_lock);
737 /* reread *fp after obtaining the lock */
738 __kill_fasync(*fp, sig, band);
739 read_unlock(&fasync_lock);
742 EXPORT_SYMBOL(kill_fasync);
744 static int __init fasync_init(void)
746 fasync_cache = kmem_cache_create("fasync_cache",
747 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
748 return 0;
751 module_init(fasync_init)