xref: /linux/fs/fcntl.c (revision eae3df7e82318d798f45dedf111e241805ec7a4a)
1 /*
2  *  linux/fs/fcntl.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 #include <linux/syscalls.h>
8 #include <linux/init.h>
9 #include <linux/mm.h>
10 #include <linux/sched/task.h>
11 #include <linux/fs.h>
12 #include <linux/file.h>
13 #include <linux/fdtable.h>
14 #include <linux/capability.h>
15 #include <linux/dnotify.h>
16 #include <linux/slab.h>
17 #include <linux/module.h>
18 #include <linux/pipe_fs_i.h>
19 #include <linux/security.h>
20 #include <linux/ptrace.h>
21 #include <linux/signal.h>
22 #include <linux/rcupdate.h>
23 #include <linux/pid_namespace.h>
24 #include <linux/user_namespace.h>
25 #include <linux/shmem_fs.h>
26 #include <linux/compat.h>
27 
28 #include <asm/poll.h>
29 #include <asm/siginfo.h>
30 #include <linux/uaccess.h>
31 
32 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
33 
34 static int setfl(int fd, struct file * filp, unsigned long arg)
35 {
36 	struct inode * inode = file_inode(filp);
37 	int error = 0;
38 
39 	/*
40 	 * O_APPEND cannot be cleared if the file is marked as append-only
41 	 * and the file is open for write.
42 	 */
43 	if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
44 		return -EPERM;
45 
46 	/* O_NOATIME can only be set by the owner or superuser */
47 	if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
48 		if (!inode_owner_or_capable(inode))
49 			return -EPERM;
50 
51 	/* required for strict SunOS emulation */
52 	if (O_NONBLOCK != O_NDELAY)
53 	       if (arg & O_NDELAY)
54 		   arg |= O_NONBLOCK;
55 
56 	/* Pipe packetized mode is controlled by O_DIRECT flag */
57 	if (!S_ISFIFO(inode->i_mode) && (arg & O_DIRECT)) {
58 		if (!filp->f_mapping || !filp->f_mapping->a_ops ||
59 			!filp->f_mapping->a_ops->direct_IO)
60 				return -EINVAL;
61 	}
62 
63 	if (filp->f_op->check_flags)
64 		error = filp->f_op->check_flags(arg);
65 	if (error)
66 		return error;
67 
68 	/*
69 	 * ->fasync() is responsible for setting the FASYNC bit.
70 	 */
71 	if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
72 		error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
73 		if (error < 0)
74 			goto out;
75 		if (error > 0)
76 			error = 0;
77 	}
78 	spin_lock(&filp->f_lock);
79 	filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
80 	spin_unlock(&filp->f_lock);
81 
82  out:
83 	return error;
84 }
85 
86 static void f_modown(struct file *filp, struct pid *pid, enum pid_type type,
87                      int force)
88 {
89 	write_lock_irq(&filp->f_owner.lock);
90 	if (force || !filp->f_owner.pid) {
91 		put_pid(filp->f_owner.pid);
92 		filp->f_owner.pid = get_pid(pid);
93 		filp->f_owner.pid_type = type;
94 
95 		if (pid) {
96 			const struct cred *cred = current_cred();
97 			filp->f_owner.uid = cred->uid;
98 			filp->f_owner.euid = cred->euid;
99 		}
100 	}
101 	write_unlock_irq(&filp->f_owner.lock);
102 }
103 
104 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
105 		int force)
106 {
107 	security_file_set_fowner(filp);
108 	f_modown(filp, pid, type, force);
109 }
110 EXPORT_SYMBOL(__f_setown);
111 
112 int f_setown(struct file *filp, unsigned long arg, int force)
113 {
114 	enum pid_type type;
115 	struct pid *pid = NULL;
116 	int who = arg, ret = 0;
117 
118 	type = PIDTYPE_PID;
119 	if (who < 0) {
120 		/* avoid overflow below */
121 		if (who == INT_MIN)
122 			return -EINVAL;
123 
124 		type = PIDTYPE_PGID;
125 		who = -who;
126 	}
127 
128 	rcu_read_lock();
129 	if (who) {
130 		pid = find_vpid(who);
131 		if (!pid)
132 			ret = -ESRCH;
133 	}
134 
135 	if (!ret)
136 		__f_setown(filp, pid, type, force);
137 	rcu_read_unlock();
138 
139 	return ret;
140 }
141 EXPORT_SYMBOL(f_setown);
142 
143 void f_delown(struct file *filp)
144 {
145 	f_modown(filp, NULL, PIDTYPE_PID, 1);
146 }
147 
148 pid_t f_getown(struct file *filp)
149 {
150 	pid_t pid;
151 	read_lock(&filp->f_owner.lock);
152 	pid = pid_vnr(filp->f_owner.pid);
153 	if (filp->f_owner.pid_type == PIDTYPE_PGID)
154 		pid = -pid;
155 	read_unlock(&filp->f_owner.lock);
156 	return pid;
157 }
158 
159 static int f_setown_ex(struct file *filp, unsigned long arg)
160 {
161 	struct f_owner_ex __user *owner_p = (void __user *)arg;
162 	struct f_owner_ex owner;
163 	struct pid *pid;
164 	int type;
165 	int ret;
166 
167 	ret = copy_from_user(&owner, owner_p, sizeof(owner));
168 	if (ret)
169 		return -EFAULT;
170 
171 	switch (owner.type) {
172 	case F_OWNER_TID:
173 		type = PIDTYPE_MAX;
174 		break;
175 
176 	case F_OWNER_PID:
177 		type = PIDTYPE_PID;
178 		break;
179 
180 	case F_OWNER_PGRP:
181 		type = PIDTYPE_PGID;
182 		break;
183 
184 	default:
185 		return -EINVAL;
186 	}
187 
188 	rcu_read_lock();
189 	pid = find_vpid(owner.pid);
190 	if (owner.pid && !pid)
191 		ret = -ESRCH;
192 	else
193 		 __f_setown(filp, pid, type, 1);
194 	rcu_read_unlock();
195 
196 	return ret;
197 }
198 
199 static int f_getown_ex(struct file *filp, unsigned long arg)
200 {
201 	struct f_owner_ex __user *owner_p = (void __user *)arg;
202 	struct f_owner_ex owner;
203 	int ret = 0;
204 
205 	read_lock(&filp->f_owner.lock);
206 	owner.pid = pid_vnr(filp->f_owner.pid);
207 	switch (filp->f_owner.pid_type) {
208 	case PIDTYPE_MAX:
209 		owner.type = F_OWNER_TID;
210 		break;
211 
212 	case PIDTYPE_PID:
213 		owner.type = F_OWNER_PID;
214 		break;
215 
216 	case PIDTYPE_PGID:
217 		owner.type = F_OWNER_PGRP;
218 		break;
219 
220 	default:
221 		WARN_ON(1);
222 		ret = -EINVAL;
223 		break;
224 	}
225 	read_unlock(&filp->f_owner.lock);
226 
227 	if (!ret) {
228 		ret = copy_to_user(owner_p, &owner, sizeof(owner));
229 		if (ret)
230 			ret = -EFAULT;
231 	}
232 	return ret;
233 }
234 
235 #ifdef CONFIG_CHECKPOINT_RESTORE
236 static int f_getowner_uids(struct file *filp, unsigned long arg)
237 {
238 	struct user_namespace *user_ns = current_user_ns();
239 	uid_t __user *dst = (void __user *)arg;
240 	uid_t src[2];
241 	int err;
242 
243 	read_lock(&filp->f_owner.lock);
244 	src[0] = from_kuid(user_ns, filp->f_owner.uid);
245 	src[1] = from_kuid(user_ns, filp->f_owner.euid);
246 	read_unlock(&filp->f_owner.lock);
247 
248 	err  = put_user(src[0], &dst[0]);
249 	err |= put_user(src[1], &dst[1]);
250 
251 	return err;
252 }
253 #else
254 static int f_getowner_uids(struct file *filp, unsigned long arg)
255 {
256 	return -EINVAL;
257 }
258 #endif
259 
260 static bool rw_hint_valid(enum rw_hint hint)
261 {
262 	switch (hint) {
263 	case RWF_WRITE_LIFE_NOT_SET:
264 	case RWH_WRITE_LIFE_NONE:
265 	case RWH_WRITE_LIFE_SHORT:
266 	case RWH_WRITE_LIFE_MEDIUM:
267 	case RWH_WRITE_LIFE_LONG:
268 	case RWH_WRITE_LIFE_EXTREME:
269 		return true;
270 	default:
271 		return false;
272 	}
273 }
274 
275 static long fcntl_rw_hint(struct file *file, unsigned int cmd,
276 			  unsigned long arg)
277 {
278 	struct inode *inode = file_inode(file);
279 	u64 *argp = (u64 __user *)arg;
280 	enum rw_hint hint;
281 	u64 h;
282 
283 	switch (cmd) {
284 	case F_GET_FILE_RW_HINT:
285 		h = file_write_hint(file);
286 		if (copy_to_user(argp, &h, sizeof(*argp)))
287 			return -EFAULT;
288 		return 0;
289 	case F_SET_FILE_RW_HINT:
290 		if (copy_from_user(&h, argp, sizeof(h)))
291 			return -EFAULT;
292 		hint = (enum rw_hint) h;
293 		if (!rw_hint_valid(hint))
294 			return -EINVAL;
295 
296 		spin_lock(&file->f_lock);
297 		file->f_write_hint = hint;
298 		spin_unlock(&file->f_lock);
299 		return 0;
300 	case F_GET_RW_HINT:
301 		h = inode->i_write_hint;
302 		if (copy_to_user(argp, &h, sizeof(*argp)))
303 			return -EFAULT;
304 		return 0;
305 	case F_SET_RW_HINT:
306 		if (copy_from_user(&h, argp, sizeof(h)))
307 			return -EFAULT;
308 		hint = (enum rw_hint) h;
309 		if (!rw_hint_valid(hint))
310 			return -EINVAL;
311 
312 		inode_lock(inode);
313 		inode->i_write_hint = hint;
314 		inode_unlock(inode);
315 		return 0;
316 	default:
317 		return -EINVAL;
318 	}
319 }
320 
321 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
322 		struct file *filp)
323 {
324 	void __user *argp = (void __user *)arg;
325 	struct flock flock;
326 	long err = -EINVAL;
327 
328 	switch (cmd) {
329 	case F_DUPFD:
330 		err = f_dupfd(arg, filp, 0);
331 		break;
332 	case F_DUPFD_CLOEXEC:
333 		err = f_dupfd(arg, filp, O_CLOEXEC);
334 		break;
335 	case F_GETFD:
336 		err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
337 		break;
338 	case F_SETFD:
339 		err = 0;
340 		set_close_on_exec(fd, arg & FD_CLOEXEC);
341 		break;
342 	case F_GETFL:
343 		err = filp->f_flags;
344 		break;
345 	case F_SETFL:
346 		err = setfl(fd, filp, arg);
347 		break;
348 #if BITS_PER_LONG != 32
349 	/* 32-bit arches must use fcntl64() */
350 	case F_OFD_GETLK:
351 #endif
352 	case F_GETLK:
353 		if (copy_from_user(&flock, argp, sizeof(flock)))
354 			return -EFAULT;
355 		err = fcntl_getlk(filp, cmd, &flock);
356 		if (!err && copy_to_user(argp, &flock, sizeof(flock)))
357 			return -EFAULT;
358 		break;
359 #if BITS_PER_LONG != 32
360 	/* 32-bit arches must use fcntl64() */
361 	case F_OFD_SETLK:
362 	case F_OFD_SETLKW:
363 #endif
364 		/* Fallthrough */
365 	case F_SETLK:
366 	case F_SETLKW:
367 		if (copy_from_user(&flock, argp, sizeof(flock)))
368 			return -EFAULT;
369 		err = fcntl_setlk(fd, filp, cmd, &flock);
370 		break;
371 	case F_GETOWN:
372 		/*
373 		 * XXX If f_owner is a process group, the
374 		 * negative return value will get converted
375 		 * into an error.  Oops.  If we keep the
376 		 * current syscall conventions, the only way
377 		 * to fix this will be in libc.
378 		 */
379 		err = f_getown(filp);
380 		force_successful_syscall_return();
381 		break;
382 	case F_SETOWN:
383 		err = f_setown(filp, arg, 1);
384 		break;
385 	case F_GETOWN_EX:
386 		err = f_getown_ex(filp, arg);
387 		break;
388 	case F_SETOWN_EX:
389 		err = f_setown_ex(filp, arg);
390 		break;
391 	case F_GETOWNER_UIDS:
392 		err = f_getowner_uids(filp, arg);
393 		break;
394 	case F_GETSIG:
395 		err = filp->f_owner.signum;
396 		break;
397 	case F_SETSIG:
398 		/* arg == 0 restores default behaviour. */
399 		if (!valid_signal(arg)) {
400 			break;
401 		}
402 		err = 0;
403 		filp->f_owner.signum = arg;
404 		break;
405 	case F_GETLEASE:
406 		err = fcntl_getlease(filp);
407 		break;
408 	case F_SETLEASE:
409 		err = fcntl_setlease(fd, filp, arg);
410 		break;
411 	case F_NOTIFY:
412 		err = fcntl_dirnotify(fd, filp, arg);
413 		break;
414 	case F_SETPIPE_SZ:
415 	case F_GETPIPE_SZ:
416 		err = pipe_fcntl(filp, cmd, arg);
417 		break;
418 	case F_ADD_SEALS:
419 	case F_GET_SEALS:
420 		err = shmem_fcntl(filp, cmd, arg);
421 		break;
422 	case F_GET_RW_HINT:
423 	case F_SET_RW_HINT:
424 	case F_GET_FILE_RW_HINT:
425 	case F_SET_FILE_RW_HINT:
426 		err = fcntl_rw_hint(filp, cmd, arg);
427 		break;
428 	default:
429 		break;
430 	}
431 	return err;
432 }
433 
434 static int check_fcntl_cmd(unsigned cmd)
435 {
436 	switch (cmd) {
437 	case F_DUPFD:
438 	case F_DUPFD_CLOEXEC:
439 	case F_GETFD:
440 	case F_SETFD:
441 	case F_GETFL:
442 		return 1;
443 	}
444 	return 0;
445 }
446 
447 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
448 {
449 	struct fd f = fdget_raw(fd);
450 	long err = -EBADF;
451 
452 	if (!f.file)
453 		goto out;
454 
455 	if (unlikely(f.file->f_mode & FMODE_PATH)) {
456 		if (!check_fcntl_cmd(cmd))
457 			goto out1;
458 	}
459 
460 	err = security_file_fcntl(f.file, cmd, arg);
461 	if (!err)
462 		err = do_fcntl(fd, cmd, arg, f.file);
463 
464 out1:
465  	fdput(f);
466 out:
467 	return err;
468 }
469 
470 #if BITS_PER_LONG == 32
471 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
472 		unsigned long, arg)
473 {
474 	void __user *argp = (void __user *)arg;
475 	struct fd f = fdget_raw(fd);
476 	struct flock64 flock;
477 	long err = -EBADF;
478 
479 	if (!f.file)
480 		goto out;
481 
482 	if (unlikely(f.file->f_mode & FMODE_PATH)) {
483 		if (!check_fcntl_cmd(cmd))
484 			goto out1;
485 	}
486 
487 	err = security_file_fcntl(f.file, cmd, arg);
488 	if (err)
489 		goto out1;
490 
491 	switch (cmd) {
492 	case F_GETLK64:
493 	case F_OFD_GETLK:
494 		err = -EFAULT;
495 		if (copy_from_user(&flock, argp, sizeof(flock)))
496 			break;
497 		err = fcntl_getlk64(f.file, cmd, &flock);
498 		if (!err && copy_to_user(argp, &flock, sizeof(flock)))
499 			err = -EFAULT;
500 		break;
501 	case F_SETLK64:
502 	case F_SETLKW64:
503 	case F_OFD_SETLK:
504 	case F_OFD_SETLKW:
505 		err = -EFAULT;
506 		if (copy_from_user(&flock, argp, sizeof(flock)))
507 			break;
508 		err = fcntl_setlk64(fd, f.file, cmd, &flock);
509 		break;
510 	default:
511 		err = do_fcntl(fd, cmd, arg, f.file);
512 		break;
513 	}
514 out1:
515 	fdput(f);
516 out:
517 	return err;
518 }
519 #endif
520 
521 #ifdef CONFIG_COMPAT
522 /* careful - don't use anywhere else */
523 #define copy_flock_fields(dst, src)		\
524 	(dst)->l_type = (src)->l_type;		\
525 	(dst)->l_whence = (src)->l_whence;	\
526 	(dst)->l_start = (src)->l_start;	\
527 	(dst)->l_len = (src)->l_len;		\
528 	(dst)->l_pid = (src)->l_pid;
529 
530 static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
531 {
532 	struct compat_flock fl;
533 
534 	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
535 		return -EFAULT;
536 	copy_flock_fields(kfl, &fl);
537 	return 0;
538 }
539 
540 static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
541 {
542 	struct compat_flock64 fl;
543 
544 	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
545 		return -EFAULT;
546 	copy_flock_fields(kfl, &fl);
547 	return 0;
548 }
549 
550 static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
551 {
552 	struct compat_flock fl;
553 
554 	memset(&fl, 0, sizeof(struct compat_flock));
555 	copy_flock_fields(&fl, kfl);
556 	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
557 		return -EFAULT;
558 	return 0;
559 }
560 
561 static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
562 {
563 	struct compat_flock64 fl;
564 
565 	memset(&fl, 0, sizeof(struct compat_flock64));
566 	copy_flock_fields(&fl, kfl);
567 	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
568 		return -EFAULT;
569 	return 0;
570 }
571 #undef copy_flock_fields
572 
573 static unsigned int
574 convert_fcntl_cmd(unsigned int cmd)
575 {
576 	switch (cmd) {
577 	case F_GETLK64:
578 		return F_GETLK;
579 	case F_SETLK64:
580 		return F_SETLK;
581 	case F_SETLKW64:
582 		return F_SETLKW;
583 	}
584 
585 	return cmd;
586 }
587 
588 /*
589  * GETLK was successful and we need to return the data, but it needs to fit in
590  * the compat structure.
591  * l_start shouldn't be too big, unless the original start + end is greater than
592  * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return
593  * -EOVERFLOW in that case.  l_len could be too big, in which case we just
594  * truncate it, and only allow the app to see that part of the conflicting lock
595  * that might make sense to it anyway
596  */
597 static int fixup_compat_flock(struct flock *flock)
598 {
599 	if (flock->l_start > COMPAT_OFF_T_MAX)
600 		return -EOVERFLOW;
601 	if (flock->l_len > COMPAT_OFF_T_MAX)
602 		flock->l_len = COMPAT_OFF_T_MAX;
603 	return 0;
604 }
605 
606 COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
607 		       compat_ulong_t, arg)
608 {
609 	struct fd f = fdget_raw(fd);
610 	struct flock flock;
611 	long err = -EBADF;
612 
613 	if (!f.file)
614 		return err;
615 
616 	if (unlikely(f.file->f_mode & FMODE_PATH)) {
617 		if (!check_fcntl_cmd(cmd))
618 			goto out_put;
619 	}
620 
621 	err = security_file_fcntl(f.file, cmd, arg);
622 	if (err)
623 		goto out_put;
624 
625 	switch (cmd) {
626 	case F_GETLK:
627 		err = get_compat_flock(&flock, compat_ptr(arg));
628 		if (err)
629 			break;
630 		err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
631 		if (err)
632 			break;
633 		err = fixup_compat_flock(&flock);
634 		if (err)
635 			return err;
636 		err = put_compat_flock(&flock, compat_ptr(arg));
637 		break;
638 	case F_GETLK64:
639 	case F_OFD_GETLK:
640 		err = get_compat_flock64(&flock, compat_ptr(arg));
641 		if (err)
642 			break;
643 		err = fcntl_getlk(f.file, convert_fcntl_cmd(cmd), &flock);
644 		if (err)
645 			break;
646 		err = fixup_compat_flock(&flock);
647 		if (err)
648 			return err;
649 		err = put_compat_flock64(&flock, compat_ptr(arg));
650 		break;
651 	case F_SETLK:
652 	case F_SETLKW:
653 		err = get_compat_flock(&flock, compat_ptr(arg));
654 		if (err)
655 			break;
656 		err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
657 		break;
658 	case F_SETLK64:
659 	case F_SETLKW64:
660 	case F_OFD_SETLK:
661 	case F_OFD_SETLKW:
662 		err = get_compat_flock64(&flock, compat_ptr(arg));
663 		if (err)
664 			break;
665 		err = fcntl_setlk(fd, f.file, convert_fcntl_cmd(cmd), &flock);
666 		break;
667 	default:
668 		err = do_fcntl(fd, cmd, arg, f.file);
669 		break;
670 	}
671 out_put:
672 	fdput(f);
673 	return err;
674 }
675 
676 COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
677 		       compat_ulong_t, arg)
678 {
679 	switch (cmd) {
680 	case F_GETLK64:
681 	case F_SETLK64:
682 	case F_SETLKW64:
683 	case F_OFD_GETLK:
684 	case F_OFD_SETLK:
685 	case F_OFD_SETLKW:
686 		return -EINVAL;
687 	}
688 	return compat_sys_fcntl64(fd, cmd, arg);
689 }
690 #endif
691 
692 /* Table to convert sigio signal codes into poll band bitmaps */
693 
694 static const long band_table[NSIGPOLL] = {
695 	POLLIN | POLLRDNORM,			/* POLL_IN */
696 	POLLOUT | POLLWRNORM | POLLWRBAND,	/* POLL_OUT */
697 	POLLIN | POLLRDNORM | POLLMSG,		/* POLL_MSG */
698 	POLLERR,				/* POLL_ERR */
699 	POLLPRI | POLLRDBAND,			/* POLL_PRI */
700 	POLLHUP | POLLERR			/* POLL_HUP */
701 };
702 
703 static inline int sigio_perm(struct task_struct *p,
704                              struct fown_struct *fown, int sig)
705 {
706 	const struct cred *cred;
707 	int ret;
708 
709 	rcu_read_lock();
710 	cred = __task_cred(p);
711 	ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
712 		uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
713 		uid_eq(fown->uid,  cred->suid) || uid_eq(fown->uid,  cred->uid)) &&
714 	       !security_file_send_sigiotask(p, fown, sig));
715 	rcu_read_unlock();
716 	return ret;
717 }
718 
719 static void send_sigio_to_task(struct task_struct *p,
720 			       struct fown_struct *fown,
721 			       int fd, int reason, int group)
722 {
723 	/*
724 	 * F_SETSIG can change ->signum lockless in parallel, make
725 	 * sure we read it once and use the same value throughout.
726 	 */
727 	int signum = ACCESS_ONCE(fown->signum);
728 
729 	if (!sigio_perm(p, fown, signum))
730 		return;
731 
732 	switch (signum) {
733 		siginfo_t si;
734 		default:
735 			/* Queue a rt signal with the appropriate fd as its
736 			   value.  We use SI_SIGIO as the source, not
737 			   SI_KERNEL, since kernel signals always get
738 			   delivered even if we can't queue.  Failure to
739 			   queue in this case _should_ be reported; we fall
740 			   back to SIGIO in that case. --sct */
741 			si.si_signo = signum;
742 			si.si_errno = 0;
743 		        si.si_code  = reason;
744 			/*
745 			 * Posix definies POLL_IN and friends to be signal
746 			 * specific si_codes for SIG_POLL.  Linux extended
747 			 * these si_codes to other signals in a way that is
748 			 * ambiguous if other signals also have signal
749 			 * specific si_codes.  In that case use SI_SIGIO instead
750 			 * to remove the ambiguity.
751 			 */
752 			if ((signum != SIGPOLL) && sig_specific_sicodes(signum))
753 				si.si_code = SI_SIGIO;
754 
755 			/* Make sure we are called with one of the POLL_*
756 			   reasons, otherwise we could leak kernel stack into
757 			   userspace.  */
758 			BUG_ON((reason < POLL_IN) || ((reason - POLL_IN) >= NSIGPOLL));
759 			if (reason - POLL_IN >= NSIGPOLL)
760 				si.si_band  = ~0L;
761 			else
762 				si.si_band = band_table[reason - POLL_IN];
763 			si.si_fd    = fd;
764 			if (!do_send_sig_info(signum, &si, p, group))
765 				break;
766 		/* fall-through: fall back on the old plain SIGIO signal */
767 		case 0:
768 			do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, group);
769 	}
770 }
771 
772 void send_sigio(struct fown_struct *fown, int fd, int band)
773 {
774 	struct task_struct *p;
775 	enum pid_type type;
776 	struct pid *pid;
777 	int group = 1;
778 
779 	read_lock(&fown->lock);
780 
781 	type = fown->pid_type;
782 	if (type == PIDTYPE_MAX) {
783 		group = 0;
784 		type = PIDTYPE_PID;
785 	}
786 
787 	pid = fown->pid;
788 	if (!pid)
789 		goto out_unlock_fown;
790 
791 	read_lock(&tasklist_lock);
792 	do_each_pid_task(pid, type, p) {
793 		send_sigio_to_task(p, fown, fd, band, group);
794 	} while_each_pid_task(pid, type, p);
795 	read_unlock(&tasklist_lock);
796  out_unlock_fown:
797 	read_unlock(&fown->lock);
798 }
799 
800 static void send_sigurg_to_task(struct task_struct *p,
801 				struct fown_struct *fown, int group)
802 {
803 	if (sigio_perm(p, fown, SIGURG))
804 		do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, group);
805 }
806 
807 int send_sigurg(struct fown_struct *fown)
808 {
809 	struct task_struct *p;
810 	enum pid_type type;
811 	struct pid *pid;
812 	int group = 1;
813 	int ret = 0;
814 
815 	read_lock(&fown->lock);
816 
817 	type = fown->pid_type;
818 	if (type == PIDTYPE_MAX) {
819 		group = 0;
820 		type = PIDTYPE_PID;
821 	}
822 
823 	pid = fown->pid;
824 	if (!pid)
825 		goto out_unlock_fown;
826 
827 	ret = 1;
828 
829 	read_lock(&tasklist_lock);
830 	do_each_pid_task(pid, type, p) {
831 		send_sigurg_to_task(p, fown, group);
832 	} while_each_pid_task(pid, type, p);
833 	read_unlock(&tasklist_lock);
834  out_unlock_fown:
835 	read_unlock(&fown->lock);
836 	return ret;
837 }
838 
839 static DEFINE_SPINLOCK(fasync_lock);
840 static struct kmem_cache *fasync_cache __read_mostly;
841 
842 static void fasync_free_rcu(struct rcu_head *head)
843 {
844 	kmem_cache_free(fasync_cache,
845 			container_of(head, struct fasync_struct, fa_rcu));
846 }
847 
848 /*
849  * Remove a fasync entry. If successfully removed, return
850  * positive and clear the FASYNC flag. If no entry exists,
851  * do nothing and return 0.
852  *
853  * NOTE! It is very important that the FASYNC flag always
854  * match the state "is the filp on a fasync list".
855  *
856  */
857 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
858 {
859 	struct fasync_struct *fa, **fp;
860 	int result = 0;
861 
862 	spin_lock(&filp->f_lock);
863 	spin_lock(&fasync_lock);
864 	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
865 		if (fa->fa_file != filp)
866 			continue;
867 
868 		spin_lock_irq(&fa->fa_lock);
869 		fa->fa_file = NULL;
870 		spin_unlock_irq(&fa->fa_lock);
871 
872 		*fp = fa->fa_next;
873 		call_rcu(&fa->fa_rcu, fasync_free_rcu);
874 		filp->f_flags &= ~FASYNC;
875 		result = 1;
876 		break;
877 	}
878 	spin_unlock(&fasync_lock);
879 	spin_unlock(&filp->f_lock);
880 	return result;
881 }
882 
883 struct fasync_struct *fasync_alloc(void)
884 {
885 	return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
886 }
887 
888 /*
889  * NOTE! This can be used only for unused fasync entries:
890  * entries that actually got inserted on the fasync list
891  * need to be released by rcu - see fasync_remove_entry.
892  */
893 void fasync_free(struct fasync_struct *new)
894 {
895 	kmem_cache_free(fasync_cache, new);
896 }
897 
898 /*
899  * Insert a new entry into the fasync list.  Return the pointer to the
900  * old one if we didn't use the new one.
901  *
902  * NOTE! It is very important that the FASYNC flag always
903  * match the state "is the filp on a fasync list".
904  */
905 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
906 {
907         struct fasync_struct *fa, **fp;
908 
909 	spin_lock(&filp->f_lock);
910 	spin_lock(&fasync_lock);
911 	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
912 		if (fa->fa_file != filp)
913 			continue;
914 
915 		spin_lock_irq(&fa->fa_lock);
916 		fa->fa_fd = fd;
917 		spin_unlock_irq(&fa->fa_lock);
918 		goto out;
919 	}
920 
921 	spin_lock_init(&new->fa_lock);
922 	new->magic = FASYNC_MAGIC;
923 	new->fa_file = filp;
924 	new->fa_fd = fd;
925 	new->fa_next = *fapp;
926 	rcu_assign_pointer(*fapp, new);
927 	filp->f_flags |= FASYNC;
928 
929 out:
930 	spin_unlock(&fasync_lock);
931 	spin_unlock(&filp->f_lock);
932 	return fa;
933 }
934 
935 /*
936  * Add a fasync entry. Return negative on error, positive if
937  * added, and zero if did nothing but change an existing one.
938  */
939 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
940 {
941 	struct fasync_struct *new;
942 
943 	new = fasync_alloc();
944 	if (!new)
945 		return -ENOMEM;
946 
947 	/*
948 	 * fasync_insert_entry() returns the old (update) entry if
949 	 * it existed.
950 	 *
951 	 * So free the (unused) new entry and return 0 to let the
952 	 * caller know that we didn't add any new fasync entries.
953 	 */
954 	if (fasync_insert_entry(fd, filp, fapp, new)) {
955 		fasync_free(new);
956 		return 0;
957 	}
958 
959 	return 1;
960 }
961 
962 /*
963  * fasync_helper() is used by almost all character device drivers
964  * to set up the fasync queue, and for regular files by the file
965  * lease code. It returns negative on error, 0 if it did no changes
966  * and positive if it added/deleted the entry.
967  */
968 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
969 {
970 	if (!on)
971 		return fasync_remove_entry(filp, fapp);
972 	return fasync_add_entry(fd, filp, fapp);
973 }
974 
975 EXPORT_SYMBOL(fasync_helper);
976 
977 /*
978  * rcu_read_lock() is held
979  */
980 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
981 {
982 	while (fa) {
983 		struct fown_struct *fown;
984 		unsigned long flags;
985 
986 		if (fa->magic != FASYNC_MAGIC) {
987 			printk(KERN_ERR "kill_fasync: bad magic number in "
988 			       "fasync_struct!\n");
989 			return;
990 		}
991 		spin_lock_irqsave(&fa->fa_lock, flags);
992 		if (fa->fa_file) {
993 			fown = &fa->fa_file->f_owner;
994 			/* Don't send SIGURG to processes which have not set a
995 			   queued signum: SIGURG has its own default signalling
996 			   mechanism. */
997 			if (!(sig == SIGURG && fown->signum == 0))
998 				send_sigio(fown, fa->fa_fd, band);
999 		}
1000 		spin_unlock_irqrestore(&fa->fa_lock, flags);
1001 		fa = rcu_dereference(fa->fa_next);
1002 	}
1003 }
1004 
1005 void kill_fasync(struct fasync_struct **fp, int sig, int band)
1006 {
1007 	/* First a quick test without locking: usually
1008 	 * the list is empty.
1009 	 */
1010 	if (*fp) {
1011 		rcu_read_lock();
1012 		kill_fasync_rcu(rcu_dereference(*fp), sig, band);
1013 		rcu_read_unlock();
1014 	}
1015 }
1016 EXPORT_SYMBOL(kill_fasync);
1017 
1018 static int __init fcntl_init(void)
1019 {
1020 	/*
1021 	 * Please add new bits here to ensure allocation uniqueness.
1022 	 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
1023 	 * is defined as O_NONBLOCK on some platforms and not on others.
1024 	 */
1025 	BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
1026 		HWEIGHT32(
1027 			(VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
1028 			__FMODE_EXEC | __FMODE_NONOTIFY));
1029 
1030 	fasync_cache = kmem_cache_create("fasync_cache",
1031 		sizeof(struct fasync_struct), 0, SLAB_PANIC, NULL);
1032 	return 0;
1033 }
1034 
1035 module_init(fcntl_init)
1036