ARC: export "abort" for modules
[linux/fpc-iii.git] / fs / fcntl.c
blobec03cf620fd7a4a2253633573e4f08bc05a3a478
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/pipe_fs_i.h>
18 #include <linux/security.h>
19 #include <linux/ptrace.h>
20 #include <linux/signal.h>
21 #include <linux/rcupdate.h>
22 #include <linux/pid_namespace.h>
23 #include <linux/user_namespace.h>
24 #include <linux/shmem_fs.h>
26 #include <asm/poll.h>
27 #include <asm/siginfo.h>
28 #include <asm/uaccess.h>
30 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
32 static int setfl(int fd, struct file * filp, unsigned long arg)
34 struct inode * inode = file_inode(filp);
35 int error = 0;
38 * O_APPEND cannot be cleared if the file is marked as append-only
39 * and the file is open for write.
41 if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
42 return -EPERM;
44 /* O_NOATIME can only be set by the owner or superuser */
45 if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
46 if (!inode_owner_or_capable(inode))
47 return -EPERM;
49 /* required for strict SunOS emulation */
50 if (O_NONBLOCK != O_NDELAY)
51 if (arg & O_NDELAY)
52 arg |= O_NONBLOCK;
54 /* Pipe packetized mode is controlled by O_DIRECT flag */
55 if (!S_ISFIFO(filp->f_inode->i_mode) && (arg & O_DIRECT)) {
56 if (!filp->f_mapping || !filp->f_mapping->a_ops ||
57 !filp->f_mapping->a_ops->direct_IO)
58 return -EINVAL;
61 if (filp->f_op->check_flags)
62 error = filp->f_op->check_flags(arg);
63 if (error)
64 return error;
67 * ->fasync() is responsible for setting the FASYNC bit.
69 if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
70 error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
71 if (error < 0)
72 goto out;
73 if (error > 0)
74 error = 0;
76 spin_lock(&filp->f_lock);
77 filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
78 spin_unlock(&filp->f_lock);
80 out:
81 return error;
84 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
85 int force)
87 write_lock_irq(&filp->f_owner.lock);
88 if (force || !filp->f_owner.pid) {
89 put_pid(filp->f_owner.pid);
90 filp->f_owner.pid = get_pid(pid);
91 filp->f_owner.pid_type = type;
93 if (pid) {
94 const struct cred *cred = current_cred();
95 filp->f_owner.uid = cred->uid;
96 filp->f_owner.euid = cred->euid;
99 write_unlock_irq(&filp->f_owner.lock);
102 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
103 int force)
105 security_file_set_fowner(filp);
106 f_modown(filp, pid, type, force);
108 EXPORT_SYMBOL(__f_setown);
110 void f_setown(struct file *filp, unsigned long arg, int force)
112 enum pid_type type;
113 struct pid *pid;
114 int who = arg;
115 type = PIDTYPE_PID;
116 if (who < 0) {
117 /* avoid overflow below */
118 if (who == INT_MIN)
119 return;
121 type = PIDTYPE_PGID;
122 who = -who;
124 rcu_read_lock();
125 pid = find_vpid(who);
126 __f_setown(filp, pid, type, force);
127 rcu_read_unlock();
129 EXPORT_SYMBOL(f_setown);
131 void f_delown(struct file *filp)
133 f_modown(filp, NULL, PIDTYPE_PID, 1);
136 pid_t f_getown(struct file *filp)
138 pid_t pid;
139 read_lock(&filp->f_owner.lock);
140 pid = pid_vnr(filp->f_owner.pid);
141 if (filp->f_owner.pid_type == PIDTYPE_PGID)
142 pid = -pid;
143 read_unlock(&filp->f_owner.lock);
144 return pid;
147 static int f_setown_ex(struct file *filp, unsigned long arg)
149 struct f_owner_ex __user *owner_p = (void __user *)arg;
150 struct f_owner_ex owner;
151 struct pid *pid;
152 int type;
153 int ret;
155 ret = copy_from_user(&owner, owner_p, sizeof(owner));
156 if (ret)
157 return -EFAULT;
159 switch (owner.type) {
160 case F_OWNER_TID:
161 type = PIDTYPE_MAX;
162 break;
164 case F_OWNER_PID:
165 type = PIDTYPE_PID;
166 break;
168 case F_OWNER_PGRP:
169 type = PIDTYPE_PGID;
170 break;
172 default:
173 return -EINVAL;
176 rcu_read_lock();
177 pid = find_vpid(owner.pid);
178 if (owner.pid && !pid)
179 ret = -ESRCH;
180 else
181 __f_setown(filp, pid, type, 1);
182 rcu_read_unlock();
184 return ret;
187 static int f_getown_ex(struct file *filp, unsigned long arg)
189 struct f_owner_ex __user *owner_p = (void __user *)arg;
190 struct f_owner_ex owner;
191 int ret = 0;
193 read_lock(&filp->f_owner.lock);
194 owner.pid = pid_vnr(filp->f_owner.pid);
195 switch (filp->f_owner.pid_type) {
196 case PIDTYPE_MAX:
197 owner.type = F_OWNER_TID;
198 break;
200 case PIDTYPE_PID:
201 owner.type = F_OWNER_PID;
202 break;
204 case PIDTYPE_PGID:
205 owner.type = F_OWNER_PGRP;
206 break;
208 default:
209 WARN_ON(1);
210 ret = -EINVAL;
211 break;
213 read_unlock(&filp->f_owner.lock);
215 if (!ret) {
216 ret = copy_to_user(owner_p, &owner, sizeof(owner));
217 if (ret)
218 ret = -EFAULT;
220 return ret;
223 #ifdef CONFIG_CHECKPOINT_RESTORE
224 static int f_getowner_uids(struct file *filp, unsigned long arg)
226 struct user_namespace *user_ns = current_user_ns();
227 uid_t __user *dst = (void __user *)arg;
228 uid_t src[2];
229 int err;
231 read_lock(&filp->f_owner.lock);
232 src[0] = from_kuid(user_ns, filp->f_owner.uid);
233 src[1] = from_kuid(user_ns, filp->f_owner.euid);
234 read_unlock(&filp->f_owner.lock);
236 err = put_user(src[0], &dst[0]);
237 err |= put_user(src[1], &dst[1]);
239 return err;
241 #else
242 static int f_getowner_uids(struct file *filp, unsigned long arg)
244 return -EINVAL;
246 #endif
248 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
249 struct file *filp)
251 long err = -EINVAL;
253 switch (cmd) {
254 case F_DUPFD:
255 err = f_dupfd(arg, filp, 0);
256 break;
257 case F_DUPFD_CLOEXEC:
258 err = f_dupfd(arg, filp, O_CLOEXEC);
259 break;
260 case F_GETFD:
261 err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
262 break;
263 case F_SETFD:
264 err = 0;
265 set_close_on_exec(fd, arg & FD_CLOEXEC);
266 break;
267 case F_GETFL:
268 err = filp->f_flags;
269 break;
270 case F_SETFL:
271 err = setfl(fd, filp, arg);
272 break;
273 #if BITS_PER_LONG != 32
274 /* 32-bit arches must use fcntl64() */
275 case F_OFD_GETLK:
276 #endif
277 case F_GETLK:
278 err = fcntl_getlk(filp, cmd, (struct flock __user *) arg);
279 break;
280 #if BITS_PER_LONG != 32
281 /* 32-bit arches must use fcntl64() */
282 case F_OFD_SETLK:
283 case F_OFD_SETLKW:
284 #endif
285 /* Fallthrough */
286 case F_SETLK:
287 case F_SETLKW:
288 err = fcntl_setlk(fd, filp, cmd, (struct flock __user *) arg);
289 break;
290 case F_GETOWN:
292 * XXX If f_owner is a process group, the
293 * negative return value will get converted
294 * into an error. Oops. If we keep the
295 * current syscall conventions, the only way
296 * to fix this will be in libc.
298 err = f_getown(filp);
299 force_successful_syscall_return();
300 break;
301 case F_SETOWN:
302 f_setown(filp, arg, 1);
303 err = 0;
304 break;
305 case F_GETOWN_EX:
306 err = f_getown_ex(filp, arg);
307 break;
308 case F_SETOWN_EX:
309 err = f_setown_ex(filp, arg);
310 break;
311 case F_GETOWNER_UIDS:
312 err = f_getowner_uids(filp, arg);
313 break;
314 case F_GETSIG:
315 err = filp->f_owner.signum;
316 break;
317 case F_SETSIG:
318 /* arg == 0 restores default behaviour. */
319 if (!valid_signal(arg)) {
320 break;
322 err = 0;
323 filp->f_owner.signum = arg;
324 break;
325 case F_GETLEASE:
326 err = fcntl_getlease(filp);
327 break;
328 case F_SETLEASE:
329 err = fcntl_setlease(fd, filp, arg);
330 break;
331 case F_NOTIFY:
332 err = fcntl_dirnotify(fd, filp, arg);
333 break;
334 case F_SETPIPE_SZ:
335 case F_GETPIPE_SZ:
336 err = pipe_fcntl(filp, cmd, arg);
337 break;
338 case F_ADD_SEALS:
339 case F_GET_SEALS:
340 err = shmem_fcntl(filp, cmd, arg);
341 break;
342 default:
343 break;
345 return err;
348 static int check_fcntl_cmd(unsigned cmd)
350 switch (cmd) {
351 case F_DUPFD:
352 case F_DUPFD_CLOEXEC:
353 case F_GETFD:
354 case F_SETFD:
355 case F_GETFL:
356 return 1;
358 return 0;
361 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
363 struct fd f = fdget_raw(fd);
364 long err = -EBADF;
366 if (!f.file)
367 goto out;
369 if (unlikely(f.file->f_mode & FMODE_PATH)) {
370 if (!check_fcntl_cmd(cmd))
371 goto out1;
374 err = security_file_fcntl(f.file, cmd, arg);
375 if (!err)
376 err = do_fcntl(fd, cmd, arg, f.file);
378 out1:
379 fdput(f);
380 out:
381 return err;
384 #if BITS_PER_LONG == 32
385 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
386 unsigned long, arg)
388 struct fd f = fdget_raw(fd);
389 long err = -EBADF;
391 if (!f.file)
392 goto out;
394 if (unlikely(f.file->f_mode & FMODE_PATH)) {
395 if (!check_fcntl_cmd(cmd))
396 goto out1;
399 err = security_file_fcntl(f.file, cmd, arg);
400 if (err)
401 goto out1;
403 switch (cmd) {
404 case F_GETLK64:
405 case F_OFD_GETLK:
406 err = fcntl_getlk64(f.file, cmd, (struct flock64 __user *) arg);
407 break;
408 case F_SETLK64:
409 case F_SETLKW64:
410 case F_OFD_SETLK:
411 case F_OFD_SETLKW:
412 err = fcntl_setlk64(fd, f.file, cmd,
413 (struct flock64 __user *) arg);
414 break;
415 default:
416 err = do_fcntl(fd, cmd, arg, f.file);
417 break;
419 out1:
420 fdput(f);
421 out:
422 return err;
424 #endif
426 /* Table to convert sigio signal codes into poll band bitmaps */
428 static const long band_table[NSIGPOLL] = {
429 POLLIN | POLLRDNORM, /* POLL_IN */
430 POLLOUT | POLLWRNORM | POLLWRBAND, /* POLL_OUT */
431 POLLIN | POLLRDNORM | POLLMSG, /* POLL_MSG */
432 POLLERR, /* POLL_ERR */
433 POLLPRI | POLLRDBAND, /* POLL_PRI */
434 POLLHUP | POLLERR /* POLL_HUP */
437 static inline int sigio_perm(struct task_struct *p,
438 struct fown_struct *fown, int sig)
440 const struct cred *cred;
441 int ret;
443 rcu_read_lock();
444 cred = __task_cred(p);
445 ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
446 uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
447 uid_eq(fown->uid, cred->suid) || uid_eq(fown->uid, cred->uid)) &&
448 !security_file_send_sigiotask(p, fown, sig));
449 rcu_read_unlock();
450 return ret;
453 static void send_sigio_to_task(struct task_struct *p,
454 struct fown_struct *fown,
455 int fd, int reason, int group)
458 * F_SETSIG can change ->signum lockless in parallel, make
459 * sure we read it once and use the same value throughout.
461 int signum = ACCESS_ONCE(fown->signum);
463 if (!sigio_perm(p, fown, signum))
464 return;
466 switch (signum) {
467 siginfo_t si;
468 default:
469 /* Queue a rt signal with the appropriate fd as its
470 value. We use SI_SIGIO as the source, not
471 SI_KERNEL, since kernel signals always get
472 delivered even if we can't queue. Failure to
473 queue in this case _should_ be reported; we fall
474 back to SIGIO in that case. --sct */
475 si.si_signo = signum;
476 si.si_errno = 0;
477 si.si_code = reason;
478 /* Make sure we are called with one of the POLL_*
479 reasons, otherwise we could leak kernel stack into
480 userspace. */
481 BUG_ON((reason & __SI_MASK) != __SI_POLL);
482 if (reason - POLL_IN >= NSIGPOLL)
483 si.si_band = ~0L;
484 else
485 si.si_band = band_table[reason - POLL_IN];
486 si.si_fd = fd;
487 if (!do_send_sig_info(signum, &si, p, group))
488 break;
489 /* fall-through: fall back on the old plain SIGIO signal */
490 case 0:
491 do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
495 void send_sigio(struct fown_struct *fown, int fd, int band)
497 struct task_struct *p;
498 enum pid_type type;
499 struct pid *pid;
500 int group = 1;
502 read_lock(&fown->lock);
504 type = fown->pid_type;
505 if (type == PIDTYPE_MAX) {
506 group = 0;
507 type = PIDTYPE_PID;
510 pid = fown->pid;
511 if (!pid)
512 goto out_unlock_fown;
514 read_lock(&tasklist_lock);
515 do_each_pid_task(pid, type, p) {
516 send_sigio_to_task(p, fown, fd, band, group);
517 } while_each_pid_task(pid, type, p);
518 read_unlock(&tasklist_lock);
519 out_unlock_fown:
520 read_unlock(&fown->lock);
523 static void send_sigurg_to_task(struct task_struct *p,
524 struct fown_struct *fown, int group)
526 if (sigio_perm(p, fown, SIGURG))
527 do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
530 int send_sigurg(struct fown_struct *fown)
532 struct task_struct *p;
533 enum pid_type type;
534 struct pid *pid;
535 int group = 1;
536 int ret = 0;
538 read_lock(&fown->lock);
540 type = fown->pid_type;
541 if (type == PIDTYPE_MAX) {
542 group = 0;
543 type = PIDTYPE_PID;
546 pid = fown->pid;
547 if (!pid)
548 goto out_unlock_fown;
550 ret = 1;
552 read_lock(&tasklist_lock);
553 do_each_pid_task(pid, type, p) {
554 send_sigurg_to_task(p, fown, group);
555 } while_each_pid_task(pid, type, p);
556 read_unlock(&tasklist_lock);
557 out_unlock_fown:
558 read_unlock(&fown->lock);
559 return ret;
562 static DEFINE_SPINLOCK(fasync_lock);
563 static struct kmem_cache *fasync_cache __read_mostly;
565 static void fasync_free_rcu(struct rcu_head *head)
567 kmem_cache_free(fasync_cache,
568 container_of(head, struct fasync_struct, fa_rcu));
572 * Remove a fasync entry. If successfully removed, return
573 * positive and clear the FASYNC flag. If no entry exists,
574 * do nothing and return 0.
576 * NOTE! It is very important that the FASYNC flag always
577 * match the state "is the filp on a fasync list".
580 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
582 struct fasync_struct *fa, **fp;
583 int result = 0;
585 spin_lock(&filp->f_lock);
586 spin_lock(&fasync_lock);
587 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
588 if (fa->fa_file != filp)
589 continue;
591 spin_lock_irq(&fa->fa_lock);
592 fa->fa_file = NULL;
593 spin_unlock_irq(&fa->fa_lock);
595 *fp = fa->fa_next;
596 call_rcu(&fa->fa_rcu, fasync_free_rcu);
597 filp->f_flags &= ~FASYNC;
598 result = 1;
599 break;
601 spin_unlock(&fasync_lock);
602 spin_unlock(&filp->f_lock);
603 return result;
606 struct fasync_struct *fasync_alloc(void)
608 return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
612 * NOTE! This can be used only for unused fasync entries:
613 * entries that actually got inserted on the fasync list
614 * need to be released by rcu - see fasync_remove_entry.
616 void fasync_free(struct fasync_struct *new)
618 kmem_cache_free(fasync_cache, new);
622 * Insert a new entry into the fasync list. Return the pointer to the
623 * old one if we didn't use the new one.
625 * NOTE! It is very important that the FASYNC flag always
626 * match the state "is the filp on a fasync list".
628 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
630 struct fasync_struct *fa, **fp;
632 spin_lock(&filp->f_lock);
633 spin_lock(&fasync_lock);
634 for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
635 if (fa->fa_file != filp)
636 continue;
638 spin_lock_irq(&fa->fa_lock);
639 fa->fa_fd = fd;
640 spin_unlock_irq(&fa->fa_lock);
641 goto out;
644 spin_lock_init(&new->fa_lock);
645 new->magic = FASYNC_MAGIC;
646 new->fa_file = filp;
647 new->fa_fd = fd;
648 new->fa_next = *fapp;
649 rcu_assign_pointer(*fapp, new);
650 filp->f_flags |= FASYNC;
652 out:
653 spin_unlock(&fasync_lock);
654 spin_unlock(&filp->f_lock);
655 return fa;
659 * Add a fasync entry. Return negative on error, positive if
660 * added, and zero if did nothing but change an existing one.
662 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
664 struct fasync_struct *new;
666 new = fasync_alloc();
667 if (!new)
668 return -ENOMEM;
671 * fasync_insert_entry() returns the old (update) entry if
672 * it existed.
674 * So free the (unused) new entry and return 0 to let the
675 * caller know that we didn't add any new fasync entries.
677 if (fasync_insert_entry(fd, filp, fapp, new)) {
678 fasync_free(new);
679 return 0;
682 return 1;
686 * fasync_helper() is used by almost all character device drivers
687 * to set up the fasync queue, and for regular files by the file
688 * lease code. It returns negative on error, 0 if it did no changes
689 * and positive if it added/deleted the entry.
691 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
693 if (!on)
694 return fasync_remove_entry(filp, fapp);
695 return fasync_add_entry(fd, filp, fapp);
698 EXPORT_SYMBOL(fasync_helper);
701 * rcu_read_lock() is held
703 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
705 while (fa) {
706 struct fown_struct *fown;
707 unsigned long flags;
709 if (fa->magic != FASYNC_MAGIC) {
710 printk(KERN_ERR "kill_fasync: bad magic number in "
711 "fasync_struct!\n");
712 return;
714 spin_lock_irqsave(&fa->fa_lock, flags);
715 if (fa->fa_file) {
716 fown = &fa->fa_file->f_owner;
717 /* Don't send SIGURG to processes which have not set a
718 queued signum: SIGURG has its own default signalling
719 mechanism. */
720 if (!(sig == SIGURG && fown->signum == 0))
721 send_sigio(fown, fa->fa_fd, band);
723 spin_unlock_irqrestore(&fa->fa_lock, flags);
724 fa = rcu_dereference(fa->fa_next);
728 void kill_fasync(struct fasync_struct **fp, int sig, int band)
730 /* First a quick test without locking: usually
731 * the list is empty.
733 if (*fp) {
734 rcu_read_lock();
735 kill_fasync_rcu(rcu_dereference(*fp), sig, band);
736 rcu_read_unlock();
739 EXPORT_SYMBOL(kill_fasync);
741 static int __init fcntl_init(void)
744 * Please add new bits here to ensure allocation uniqueness.
745 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
746 * is defined as O_NONBLOCK on some platforms and not on others.
748 BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
749 HWEIGHT32(
750 (VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
751 __FMODE_EXEC | __FMODE_NONOTIFY));
753 fasync_cache = kmem_cache_create("fasync_cache",
754 sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
755 return 0;
758 module_init(fcntl_init)