xref: /linux/fs/fcntl.c (revision be5498cac2ddb112c5bd7433d5e834a1a2493427)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *  linux/fs/fcntl.c
4  *
5  *  Copyright (C) 1991, 1992  Linus Torvalds
6  */
7 
8 #include <linux/syscalls.h>
9 #include <linux/init.h>
10 #include <linux/mm.h>
11 #include <linux/sched/task.h>
12 #include <linux/fs.h>
13 #include <linux/filelock.h>
14 #include <linux/file.h>
15 #include <linux/capability.h>
16 #include <linux/dnotify.h>
17 #include <linux/slab.h>
18 #include <linux/module.h>
19 #include <linux/pipe_fs_i.h>
20 #include <linux/security.h>
21 #include <linux/ptrace.h>
22 #include <linux/signal.h>
23 #include <linux/rcupdate.h>
24 #include <linux/pid_namespace.h>
25 #include <linux/user_namespace.h>
26 #include <linux/memfd.h>
27 #include <linux/compat.h>
28 #include <linux/mount.h>
29 #include <linux/rw_hint.h>
30 
31 #include <linux/poll.h>
32 #include <asm/siginfo.h>
33 #include <linux/uaccess.h>
34 
35 #include "internal.h"
36 
37 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
38 
39 static int setfl(int fd, struct file * filp, unsigned int arg)
40 {
41 	struct inode * inode = file_inode(filp);
42 	int error = 0;
43 
44 	/*
45 	 * O_APPEND cannot be cleared if the file is marked as append-only
46 	 * and the file is open for write.
47 	 */
48 	if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
49 		return -EPERM;
50 
51 	/* O_NOATIME can only be set by the owner or superuser */
52 	if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
53 		if (!inode_owner_or_capable(file_mnt_idmap(filp), inode))
54 			return -EPERM;
55 
56 	/* required for strict SunOS emulation */
57 	if (O_NONBLOCK != O_NDELAY)
58 	       if (arg & O_NDELAY)
59 		   arg |= O_NONBLOCK;
60 
61 	/* Pipe packetized mode is controlled by O_DIRECT flag */
62 	if (!S_ISFIFO(inode->i_mode) &&
63 	    (arg & O_DIRECT) &&
64 	    !(filp->f_mode & FMODE_CAN_ODIRECT))
65 		return -EINVAL;
66 
67 	if (filp->f_op->check_flags)
68 		error = filp->f_op->check_flags(arg);
69 	if (error)
70 		return error;
71 
72 	/*
73 	 * ->fasync() is responsible for setting the FASYNC bit.
74 	 */
75 	if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
76 		error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
77 		if (error < 0)
78 			goto out;
79 		if (error > 0)
80 			error = 0;
81 	}
82 	spin_lock(&filp->f_lock);
83 	filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
84 	filp->f_iocb_flags = iocb_flags(filp);
85 	spin_unlock(&filp->f_lock);
86 
87  out:
88 	return error;
89 }
90 
91 /*
92  * Allocate an file->f_owner struct if it doesn't exist, handling racing
93  * allocations correctly.
94  */
95 int file_f_owner_allocate(struct file *file)
96 {
97 	struct fown_struct *f_owner;
98 
99 	f_owner = file_f_owner(file);
100 	if (f_owner)
101 		return 0;
102 
103 	f_owner = kzalloc(sizeof(struct fown_struct), GFP_KERNEL);
104 	if (!f_owner)
105 		return -ENOMEM;
106 
107 	rwlock_init(&f_owner->lock);
108 	f_owner->file = file;
109 	/* If someone else raced us, drop our allocation. */
110 	if (unlikely(cmpxchg(&file->f_owner, NULL, f_owner)))
111 		kfree(f_owner);
112 	return 0;
113 }
114 EXPORT_SYMBOL(file_f_owner_allocate);
115 
116 void file_f_owner_release(struct file *file)
117 {
118 	struct fown_struct *f_owner;
119 
120 	f_owner = file_f_owner(file);
121 	if (f_owner) {
122 		put_pid(f_owner->pid);
123 		kfree(f_owner);
124 	}
125 }
126 
127 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
128 		int force)
129 {
130 	struct fown_struct *f_owner;
131 
132 	f_owner = file_f_owner(filp);
133 	if (WARN_ON_ONCE(!f_owner))
134 		return;
135 
136 	write_lock_irq(&f_owner->lock);
137 	if (force || !f_owner->pid) {
138 		put_pid(f_owner->pid);
139 		f_owner->pid = get_pid(pid);
140 		f_owner->pid_type = type;
141 
142 		if (pid) {
143 			const struct cred *cred = current_cred();
144 			security_file_set_fowner(filp);
145 			f_owner->uid = cred->uid;
146 			f_owner->euid = cred->euid;
147 		}
148 	}
149 	write_unlock_irq(&f_owner->lock);
150 }
151 EXPORT_SYMBOL(__f_setown);
152 
153 int f_setown(struct file *filp, int who, int force)
154 {
155 	enum pid_type type;
156 	struct pid *pid = NULL;
157 	int ret = 0;
158 
159 	might_sleep();
160 
161 	type = PIDTYPE_TGID;
162 	if (who < 0) {
163 		/* avoid overflow below */
164 		if (who == INT_MIN)
165 			return -EINVAL;
166 
167 		type = PIDTYPE_PGID;
168 		who = -who;
169 	}
170 
171 	ret = file_f_owner_allocate(filp);
172 	if (ret)
173 		return ret;
174 
175 	rcu_read_lock();
176 	if (who) {
177 		pid = find_vpid(who);
178 		if (!pid)
179 			ret = -ESRCH;
180 	}
181 
182 	if (!ret)
183 		__f_setown(filp, pid, type, force);
184 	rcu_read_unlock();
185 
186 	return ret;
187 }
188 EXPORT_SYMBOL(f_setown);
189 
190 void f_delown(struct file *filp)
191 {
192 	__f_setown(filp, NULL, PIDTYPE_TGID, 1);
193 }
194 
195 pid_t f_getown(struct file *filp)
196 {
197 	pid_t pid = 0;
198 	struct fown_struct *f_owner;
199 
200 	f_owner = file_f_owner(filp);
201 	if (!f_owner)
202 		return pid;
203 
204 	read_lock_irq(&f_owner->lock);
205 	rcu_read_lock();
206 	if (pid_task(f_owner->pid, f_owner->pid_type)) {
207 		pid = pid_vnr(f_owner->pid);
208 		if (f_owner->pid_type == PIDTYPE_PGID)
209 			pid = -pid;
210 	}
211 	rcu_read_unlock();
212 	read_unlock_irq(&f_owner->lock);
213 	return pid;
214 }
215 
216 static int f_setown_ex(struct file *filp, unsigned long arg)
217 {
218 	struct f_owner_ex __user *owner_p = (void __user *)arg;
219 	struct f_owner_ex owner;
220 	struct pid *pid;
221 	int type;
222 	int ret;
223 
224 	ret = copy_from_user(&owner, owner_p, sizeof(owner));
225 	if (ret)
226 		return -EFAULT;
227 
228 	switch (owner.type) {
229 	case F_OWNER_TID:
230 		type = PIDTYPE_PID;
231 		break;
232 
233 	case F_OWNER_PID:
234 		type = PIDTYPE_TGID;
235 		break;
236 
237 	case F_OWNER_PGRP:
238 		type = PIDTYPE_PGID;
239 		break;
240 
241 	default:
242 		return -EINVAL;
243 	}
244 
245 	ret = file_f_owner_allocate(filp);
246 	if (ret)
247 		return ret;
248 
249 	rcu_read_lock();
250 	pid = find_vpid(owner.pid);
251 	if (owner.pid && !pid)
252 		ret = -ESRCH;
253 	else
254 		 __f_setown(filp, pid, type, 1);
255 	rcu_read_unlock();
256 
257 	return ret;
258 }
259 
260 static int f_getown_ex(struct file *filp, unsigned long arg)
261 {
262 	struct f_owner_ex __user *owner_p = (void __user *)arg;
263 	struct f_owner_ex owner = {};
264 	int ret = 0;
265 	struct fown_struct *f_owner;
266 	enum pid_type pid_type = PIDTYPE_PID;
267 
268 	f_owner = file_f_owner(filp);
269 	if (f_owner) {
270 		read_lock_irq(&f_owner->lock);
271 		rcu_read_lock();
272 		if (pid_task(f_owner->pid, f_owner->pid_type))
273 			owner.pid = pid_vnr(f_owner->pid);
274 		rcu_read_unlock();
275 		pid_type = f_owner->pid_type;
276 	}
277 
278 	switch (pid_type) {
279 	case PIDTYPE_PID:
280 		owner.type = F_OWNER_TID;
281 		break;
282 
283 	case PIDTYPE_TGID:
284 		owner.type = F_OWNER_PID;
285 		break;
286 
287 	case PIDTYPE_PGID:
288 		owner.type = F_OWNER_PGRP;
289 		break;
290 
291 	default:
292 		WARN_ON(1);
293 		ret = -EINVAL;
294 		break;
295 	}
296 	if (f_owner)
297 		read_unlock_irq(&f_owner->lock);
298 
299 	if (!ret) {
300 		ret = copy_to_user(owner_p, &owner, sizeof(owner));
301 		if (ret)
302 			ret = -EFAULT;
303 	}
304 	return ret;
305 }
306 
307 #ifdef CONFIG_CHECKPOINT_RESTORE
308 static int f_getowner_uids(struct file *filp, unsigned long arg)
309 {
310 	struct user_namespace *user_ns = current_user_ns();
311 	struct fown_struct *f_owner;
312 	uid_t __user *dst = (void __user *)arg;
313 	uid_t src[2] = {0, 0};
314 	int err;
315 
316 	f_owner = file_f_owner(filp);
317 	if (f_owner) {
318 		read_lock_irq(&f_owner->lock);
319 		src[0] = from_kuid(user_ns, f_owner->uid);
320 		src[1] = from_kuid(user_ns, f_owner->euid);
321 		read_unlock_irq(&f_owner->lock);
322 	}
323 
324 	err  = put_user(src[0], &dst[0]);
325 	err |= put_user(src[1], &dst[1]);
326 
327 	return err;
328 }
329 #else
330 static int f_getowner_uids(struct file *filp, unsigned long arg)
331 {
332 	return -EINVAL;
333 }
334 #endif
335 
336 static bool rw_hint_valid(u64 hint)
337 {
338 	BUILD_BUG_ON(WRITE_LIFE_NOT_SET != RWH_WRITE_LIFE_NOT_SET);
339 	BUILD_BUG_ON(WRITE_LIFE_NONE != RWH_WRITE_LIFE_NONE);
340 	BUILD_BUG_ON(WRITE_LIFE_SHORT != RWH_WRITE_LIFE_SHORT);
341 	BUILD_BUG_ON(WRITE_LIFE_MEDIUM != RWH_WRITE_LIFE_MEDIUM);
342 	BUILD_BUG_ON(WRITE_LIFE_LONG != RWH_WRITE_LIFE_LONG);
343 	BUILD_BUG_ON(WRITE_LIFE_EXTREME != RWH_WRITE_LIFE_EXTREME);
344 
345 	switch (hint) {
346 	case RWH_WRITE_LIFE_NOT_SET:
347 	case RWH_WRITE_LIFE_NONE:
348 	case RWH_WRITE_LIFE_SHORT:
349 	case RWH_WRITE_LIFE_MEDIUM:
350 	case RWH_WRITE_LIFE_LONG:
351 	case RWH_WRITE_LIFE_EXTREME:
352 		return true;
353 	default:
354 		return false;
355 	}
356 }
357 
358 static long fcntl_get_rw_hint(struct file *file, unsigned int cmd,
359 			      unsigned long arg)
360 {
361 	struct inode *inode = file_inode(file);
362 	u64 __user *argp = (u64 __user *)arg;
363 	u64 hint = READ_ONCE(inode->i_write_hint);
364 
365 	if (copy_to_user(argp, &hint, sizeof(*argp)))
366 		return -EFAULT;
367 	return 0;
368 }
369 
370 static long fcntl_set_rw_hint(struct file *file, unsigned int cmd,
371 			      unsigned long arg)
372 {
373 	struct inode *inode = file_inode(file);
374 	u64 __user *argp = (u64 __user *)arg;
375 	u64 hint;
376 
377 	if (copy_from_user(&hint, argp, sizeof(hint)))
378 		return -EFAULT;
379 	if (!rw_hint_valid(hint))
380 		return -EINVAL;
381 
382 	WRITE_ONCE(inode->i_write_hint, hint);
383 
384 	/*
385 	 * file->f_mapping->host may differ from inode. As an example,
386 	 * blkdev_open() modifies file->f_mapping.
387 	 */
388 	if (file->f_mapping->host != inode)
389 		WRITE_ONCE(file->f_mapping->host->i_write_hint, hint);
390 
391 	return 0;
392 }
393 
394 /* Is the file descriptor a dup of the file? */
395 static long f_dupfd_query(int fd, struct file *filp)
396 {
397 	CLASS(fd_raw, f)(fd);
398 
399 	/*
400 	 * We can do the 'fdput()' immediately, as the only thing that
401 	 * matters is the pointer value which isn't changed by the fdput.
402 	 *
403 	 * Technically we didn't need a ref at all, and 'fdget()' was
404 	 * overkill, but given our lockless file pointer lookup, the
405 	 * alternatives are complicated.
406 	 */
407 	return fd_file(f) == filp;
408 }
409 
410 /* Let the caller figure out whether a given file was just created. */
411 static long f_created_query(const struct file *filp)
412 {
413 	return !!(filp->f_mode & FMODE_CREATED);
414 }
415 
416 static int f_owner_sig(struct file *filp, int signum, bool setsig)
417 {
418 	int ret = 0;
419 	struct fown_struct *f_owner;
420 
421 	might_sleep();
422 
423 	if (setsig) {
424 		if (!valid_signal(signum))
425 			return -EINVAL;
426 
427 		ret = file_f_owner_allocate(filp);
428 		if (ret)
429 			return ret;
430 	}
431 
432 	f_owner = file_f_owner(filp);
433 	if (setsig)
434 		f_owner->signum = signum;
435 	else if (f_owner)
436 		ret = f_owner->signum;
437 	return ret;
438 }
439 
440 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
441 		struct file *filp)
442 {
443 	void __user *argp = (void __user *)arg;
444 	int argi = (int)arg;
445 	struct flock flock;
446 	long err = -EINVAL;
447 
448 	switch (cmd) {
449 	case F_CREATED_QUERY:
450 		err = f_created_query(filp);
451 		break;
452 	case F_DUPFD:
453 		err = f_dupfd(argi, filp, 0);
454 		break;
455 	case F_DUPFD_CLOEXEC:
456 		err = f_dupfd(argi, filp, O_CLOEXEC);
457 		break;
458 	case F_DUPFD_QUERY:
459 		err = f_dupfd_query(argi, filp);
460 		break;
461 	case F_GETFD:
462 		err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
463 		break;
464 	case F_SETFD:
465 		err = 0;
466 		set_close_on_exec(fd, argi & FD_CLOEXEC);
467 		break;
468 	case F_GETFL:
469 		err = filp->f_flags;
470 		break;
471 	case F_SETFL:
472 		err = setfl(fd, filp, argi);
473 		break;
474 #if BITS_PER_LONG != 32
475 	/* 32-bit arches must use fcntl64() */
476 	case F_OFD_GETLK:
477 #endif
478 	case F_GETLK:
479 		if (copy_from_user(&flock, argp, sizeof(flock)))
480 			return -EFAULT;
481 		err = fcntl_getlk(filp, cmd, &flock);
482 		if (!err && copy_to_user(argp, &flock, sizeof(flock)))
483 			return -EFAULT;
484 		break;
485 #if BITS_PER_LONG != 32
486 	/* 32-bit arches must use fcntl64() */
487 	case F_OFD_SETLK:
488 	case F_OFD_SETLKW:
489 		fallthrough;
490 #endif
491 	case F_SETLK:
492 	case F_SETLKW:
493 		if (copy_from_user(&flock, argp, sizeof(flock)))
494 			return -EFAULT;
495 		err = fcntl_setlk(fd, filp, cmd, &flock);
496 		break;
497 	case F_GETOWN:
498 		/*
499 		 * XXX If f_owner is a process group, the
500 		 * negative return value will get converted
501 		 * into an error.  Oops.  If we keep the
502 		 * current syscall conventions, the only way
503 		 * to fix this will be in libc.
504 		 */
505 		err = f_getown(filp);
506 		force_successful_syscall_return();
507 		break;
508 	case F_SETOWN:
509 		err = f_setown(filp, argi, 1);
510 		break;
511 	case F_GETOWN_EX:
512 		err = f_getown_ex(filp, arg);
513 		break;
514 	case F_SETOWN_EX:
515 		err = f_setown_ex(filp, arg);
516 		break;
517 	case F_GETOWNER_UIDS:
518 		err = f_getowner_uids(filp, arg);
519 		break;
520 	case F_GETSIG:
521 		err = f_owner_sig(filp, 0, false);
522 		break;
523 	case F_SETSIG:
524 		err = f_owner_sig(filp, argi, true);
525 		break;
526 	case F_GETLEASE:
527 		err = fcntl_getlease(filp);
528 		break;
529 	case F_SETLEASE:
530 		err = fcntl_setlease(fd, filp, argi);
531 		break;
532 	case F_NOTIFY:
533 		err = fcntl_dirnotify(fd, filp, argi);
534 		break;
535 	case F_SETPIPE_SZ:
536 	case F_GETPIPE_SZ:
537 		err = pipe_fcntl(filp, cmd, argi);
538 		break;
539 	case F_ADD_SEALS:
540 	case F_GET_SEALS:
541 		err = memfd_fcntl(filp, cmd, argi);
542 		break;
543 	case F_GET_RW_HINT:
544 		err = fcntl_get_rw_hint(filp, cmd, arg);
545 		break;
546 	case F_SET_RW_HINT:
547 		err = fcntl_set_rw_hint(filp, cmd, arg);
548 		break;
549 	default:
550 		break;
551 	}
552 	return err;
553 }
554 
555 static int check_fcntl_cmd(unsigned cmd)
556 {
557 	switch (cmd) {
558 	case F_CREATED_QUERY:
559 	case F_DUPFD:
560 	case F_DUPFD_CLOEXEC:
561 	case F_DUPFD_QUERY:
562 	case F_GETFD:
563 	case F_SETFD:
564 	case F_GETFL:
565 		return 1;
566 	}
567 	return 0;
568 }
569 
570 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
571 {
572 	struct fd f = fdget_raw(fd);
573 	long err = -EBADF;
574 
575 	if (!fd_file(f))
576 		goto out;
577 
578 	if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
579 		if (!check_fcntl_cmd(cmd))
580 			goto out1;
581 	}
582 
583 	err = security_file_fcntl(fd_file(f), cmd, arg);
584 	if (!err)
585 		err = do_fcntl(fd, cmd, arg, fd_file(f));
586 
587 out1:
588  	fdput(f);
589 out:
590 	return err;
591 }
592 
593 #if BITS_PER_LONG == 32
594 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
595 		unsigned long, arg)
596 {
597 	void __user *argp = (void __user *)arg;
598 	struct fd f = fdget_raw(fd);
599 	struct flock64 flock;
600 	long err = -EBADF;
601 
602 	if (!fd_file(f))
603 		goto out;
604 
605 	if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
606 		if (!check_fcntl_cmd(cmd))
607 			goto out1;
608 	}
609 
610 	err = security_file_fcntl(fd_file(f), cmd, arg);
611 	if (err)
612 		goto out1;
613 
614 	switch (cmd) {
615 	case F_GETLK64:
616 	case F_OFD_GETLK:
617 		err = -EFAULT;
618 		if (copy_from_user(&flock, argp, sizeof(flock)))
619 			break;
620 		err = fcntl_getlk64(fd_file(f), cmd, &flock);
621 		if (!err && copy_to_user(argp, &flock, sizeof(flock)))
622 			err = -EFAULT;
623 		break;
624 	case F_SETLK64:
625 	case F_SETLKW64:
626 	case F_OFD_SETLK:
627 	case F_OFD_SETLKW:
628 		err = -EFAULT;
629 		if (copy_from_user(&flock, argp, sizeof(flock)))
630 			break;
631 		err = fcntl_setlk64(fd, fd_file(f), cmd, &flock);
632 		break;
633 	default:
634 		err = do_fcntl(fd, cmd, arg, fd_file(f));
635 		break;
636 	}
637 out1:
638 	fdput(f);
639 out:
640 	return err;
641 }
642 #endif
643 
644 #ifdef CONFIG_COMPAT
645 /* careful - don't use anywhere else */
646 #define copy_flock_fields(dst, src)		\
647 	(dst)->l_type = (src)->l_type;		\
648 	(dst)->l_whence = (src)->l_whence;	\
649 	(dst)->l_start = (src)->l_start;	\
650 	(dst)->l_len = (src)->l_len;		\
651 	(dst)->l_pid = (src)->l_pid;
652 
653 static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
654 {
655 	struct compat_flock fl;
656 
657 	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
658 		return -EFAULT;
659 	copy_flock_fields(kfl, &fl);
660 	return 0;
661 }
662 
663 static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
664 {
665 	struct compat_flock64 fl;
666 
667 	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
668 		return -EFAULT;
669 	copy_flock_fields(kfl, &fl);
670 	return 0;
671 }
672 
673 static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
674 {
675 	struct compat_flock fl;
676 
677 	memset(&fl, 0, sizeof(struct compat_flock));
678 	copy_flock_fields(&fl, kfl);
679 	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
680 		return -EFAULT;
681 	return 0;
682 }
683 
684 static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
685 {
686 	struct compat_flock64 fl;
687 
688 	BUILD_BUG_ON(sizeof(kfl->l_start) > sizeof(ufl->l_start));
689 	BUILD_BUG_ON(sizeof(kfl->l_len) > sizeof(ufl->l_len));
690 
691 	memset(&fl, 0, sizeof(struct compat_flock64));
692 	copy_flock_fields(&fl, kfl);
693 	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
694 		return -EFAULT;
695 	return 0;
696 }
697 #undef copy_flock_fields
698 
699 static unsigned int
700 convert_fcntl_cmd(unsigned int cmd)
701 {
702 	switch (cmd) {
703 	case F_GETLK64:
704 		return F_GETLK;
705 	case F_SETLK64:
706 		return F_SETLK;
707 	case F_SETLKW64:
708 		return F_SETLKW;
709 	}
710 
711 	return cmd;
712 }
713 
714 /*
715  * GETLK was successful and we need to return the data, but it needs to fit in
716  * the compat structure.
717  * l_start shouldn't be too big, unless the original start + end is greater than
718  * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return
719  * -EOVERFLOW in that case.  l_len could be too big, in which case we just
720  * truncate it, and only allow the app to see that part of the conflicting lock
721  * that might make sense to it anyway
722  */
723 static int fixup_compat_flock(struct flock *flock)
724 {
725 	if (flock->l_start > COMPAT_OFF_T_MAX)
726 		return -EOVERFLOW;
727 	if (flock->l_len > COMPAT_OFF_T_MAX)
728 		flock->l_len = COMPAT_OFF_T_MAX;
729 	return 0;
730 }
731 
732 static long do_compat_fcntl64(unsigned int fd, unsigned int cmd,
733 			     compat_ulong_t arg)
734 {
735 	struct fd f = fdget_raw(fd);
736 	struct flock flock;
737 	long err = -EBADF;
738 
739 	if (!fd_file(f))
740 		return err;
741 
742 	if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
743 		if (!check_fcntl_cmd(cmd))
744 			goto out_put;
745 	}
746 
747 	err = security_file_fcntl(fd_file(f), cmd, arg);
748 	if (err)
749 		goto out_put;
750 
751 	switch (cmd) {
752 	case F_GETLK:
753 		err = get_compat_flock(&flock, compat_ptr(arg));
754 		if (err)
755 			break;
756 		err = fcntl_getlk(fd_file(f), convert_fcntl_cmd(cmd), &flock);
757 		if (err)
758 			break;
759 		err = fixup_compat_flock(&flock);
760 		if (!err)
761 			err = put_compat_flock(&flock, compat_ptr(arg));
762 		break;
763 	case F_GETLK64:
764 	case F_OFD_GETLK:
765 		err = get_compat_flock64(&flock, compat_ptr(arg));
766 		if (err)
767 			break;
768 		err = fcntl_getlk(fd_file(f), convert_fcntl_cmd(cmd), &flock);
769 		if (!err)
770 			err = put_compat_flock64(&flock, compat_ptr(arg));
771 		break;
772 	case F_SETLK:
773 	case F_SETLKW:
774 		err = get_compat_flock(&flock, compat_ptr(arg));
775 		if (err)
776 			break;
777 		err = fcntl_setlk(fd, fd_file(f), convert_fcntl_cmd(cmd), &flock);
778 		break;
779 	case F_SETLK64:
780 	case F_SETLKW64:
781 	case F_OFD_SETLK:
782 	case F_OFD_SETLKW:
783 		err = get_compat_flock64(&flock, compat_ptr(arg));
784 		if (err)
785 			break;
786 		err = fcntl_setlk(fd, fd_file(f), convert_fcntl_cmd(cmd), &flock);
787 		break;
788 	default:
789 		err = do_fcntl(fd, cmd, arg, fd_file(f));
790 		break;
791 	}
792 out_put:
793 	fdput(f);
794 	return err;
795 }
796 
797 COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
798 		       compat_ulong_t, arg)
799 {
800 	return do_compat_fcntl64(fd, cmd, arg);
801 }
802 
803 COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
804 		       compat_ulong_t, arg)
805 {
806 	switch (cmd) {
807 	case F_GETLK64:
808 	case F_SETLK64:
809 	case F_SETLKW64:
810 	case F_OFD_GETLK:
811 	case F_OFD_SETLK:
812 	case F_OFD_SETLKW:
813 		return -EINVAL;
814 	}
815 	return do_compat_fcntl64(fd, cmd, arg);
816 }
817 #endif
818 
819 /* Table to convert sigio signal codes into poll band bitmaps */
820 
821 static const __poll_t band_table[NSIGPOLL] = {
822 	EPOLLIN | EPOLLRDNORM,			/* POLL_IN */
823 	EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND,	/* POLL_OUT */
824 	EPOLLIN | EPOLLRDNORM | EPOLLMSG,		/* POLL_MSG */
825 	EPOLLERR,				/* POLL_ERR */
826 	EPOLLPRI | EPOLLRDBAND,			/* POLL_PRI */
827 	EPOLLHUP | EPOLLERR			/* POLL_HUP */
828 };
829 
830 static inline int sigio_perm(struct task_struct *p,
831                              struct fown_struct *fown, int sig)
832 {
833 	const struct cred *cred;
834 	int ret;
835 
836 	rcu_read_lock();
837 	cred = __task_cred(p);
838 	ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
839 		uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
840 		uid_eq(fown->uid,  cred->suid) || uid_eq(fown->uid,  cred->uid)) &&
841 	       !security_file_send_sigiotask(p, fown, sig));
842 	rcu_read_unlock();
843 	return ret;
844 }
845 
846 static void send_sigio_to_task(struct task_struct *p,
847 			       struct fown_struct *fown,
848 			       int fd, int reason, enum pid_type type)
849 {
850 	/*
851 	 * F_SETSIG can change ->signum lockless in parallel, make
852 	 * sure we read it once and use the same value throughout.
853 	 */
854 	int signum = READ_ONCE(fown->signum);
855 
856 	if (!sigio_perm(p, fown, signum))
857 		return;
858 
859 	switch (signum) {
860 		default: {
861 			kernel_siginfo_t si;
862 
863 			/* Queue a rt signal with the appropriate fd as its
864 			   value.  We use SI_SIGIO as the source, not
865 			   SI_KERNEL, since kernel signals always get
866 			   delivered even if we can't queue.  Failure to
867 			   queue in this case _should_ be reported; we fall
868 			   back to SIGIO in that case. --sct */
869 			clear_siginfo(&si);
870 			si.si_signo = signum;
871 			si.si_errno = 0;
872 		        si.si_code  = reason;
873 			/*
874 			 * Posix definies POLL_IN and friends to be signal
875 			 * specific si_codes for SIG_POLL.  Linux extended
876 			 * these si_codes to other signals in a way that is
877 			 * ambiguous if other signals also have signal
878 			 * specific si_codes.  In that case use SI_SIGIO instead
879 			 * to remove the ambiguity.
880 			 */
881 			if ((signum != SIGPOLL) && sig_specific_sicodes(signum))
882 				si.si_code = SI_SIGIO;
883 
884 			/* Make sure we are called with one of the POLL_*
885 			   reasons, otherwise we could leak kernel stack into
886 			   userspace.  */
887 			BUG_ON((reason < POLL_IN) || ((reason - POLL_IN) >= NSIGPOLL));
888 			if (reason - POLL_IN >= NSIGPOLL)
889 				si.si_band  = ~0L;
890 			else
891 				si.si_band = mangle_poll(band_table[reason - POLL_IN]);
892 			si.si_fd    = fd;
893 			if (!do_send_sig_info(signum, &si, p, type))
894 				break;
895 		}
896 			fallthrough;	/* fall back on the old plain SIGIO signal */
897 		case 0:
898 			do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, type);
899 	}
900 }
901 
902 void send_sigio(struct fown_struct *fown, int fd, int band)
903 {
904 	struct task_struct *p;
905 	enum pid_type type;
906 	unsigned long flags;
907 	struct pid *pid;
908 
909 	read_lock_irqsave(&fown->lock, flags);
910 
911 	type = fown->pid_type;
912 	pid = fown->pid;
913 	if (!pid)
914 		goto out_unlock_fown;
915 
916 	if (type <= PIDTYPE_TGID) {
917 		rcu_read_lock();
918 		p = pid_task(pid, PIDTYPE_PID);
919 		if (p)
920 			send_sigio_to_task(p, fown, fd, band, type);
921 		rcu_read_unlock();
922 	} else {
923 		read_lock(&tasklist_lock);
924 		do_each_pid_task(pid, type, p) {
925 			send_sigio_to_task(p, fown, fd, band, type);
926 		} while_each_pid_task(pid, type, p);
927 		read_unlock(&tasklist_lock);
928 	}
929  out_unlock_fown:
930 	read_unlock_irqrestore(&fown->lock, flags);
931 }
932 
933 static void send_sigurg_to_task(struct task_struct *p,
934 				struct fown_struct *fown, enum pid_type type)
935 {
936 	if (sigio_perm(p, fown, SIGURG))
937 		do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, type);
938 }
939 
940 int send_sigurg(struct file *file)
941 {
942 	struct fown_struct *fown;
943 	struct task_struct *p;
944 	enum pid_type type;
945 	struct pid *pid;
946 	unsigned long flags;
947 	int ret = 0;
948 
949 	fown = file_f_owner(file);
950 	if (!fown)
951 		return 0;
952 
953 	read_lock_irqsave(&fown->lock, flags);
954 
955 	type = fown->pid_type;
956 	pid = fown->pid;
957 	if (!pid)
958 		goto out_unlock_fown;
959 
960 	ret = 1;
961 
962 	if (type <= PIDTYPE_TGID) {
963 		rcu_read_lock();
964 		p = pid_task(pid, PIDTYPE_PID);
965 		if (p)
966 			send_sigurg_to_task(p, fown, type);
967 		rcu_read_unlock();
968 	} else {
969 		read_lock(&tasklist_lock);
970 		do_each_pid_task(pid, type, p) {
971 			send_sigurg_to_task(p, fown, type);
972 		} while_each_pid_task(pid, type, p);
973 		read_unlock(&tasklist_lock);
974 	}
975  out_unlock_fown:
976 	read_unlock_irqrestore(&fown->lock, flags);
977 	return ret;
978 }
979 
980 static DEFINE_SPINLOCK(fasync_lock);
981 static struct kmem_cache *fasync_cache __ro_after_init;
982 
983 /*
984  * Remove a fasync entry. If successfully removed, return
985  * positive and clear the FASYNC flag. If no entry exists,
986  * do nothing and return 0.
987  *
988  * NOTE! It is very important that the FASYNC flag always
989  * match the state "is the filp on a fasync list".
990  *
991  */
992 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
993 {
994 	struct fasync_struct *fa, **fp;
995 	int result = 0;
996 
997 	spin_lock(&filp->f_lock);
998 	spin_lock(&fasync_lock);
999 	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
1000 		if (fa->fa_file != filp)
1001 			continue;
1002 
1003 		write_lock_irq(&fa->fa_lock);
1004 		fa->fa_file = NULL;
1005 		write_unlock_irq(&fa->fa_lock);
1006 
1007 		*fp = fa->fa_next;
1008 		kfree_rcu(fa, fa_rcu);
1009 		filp->f_flags &= ~FASYNC;
1010 		result = 1;
1011 		break;
1012 	}
1013 	spin_unlock(&fasync_lock);
1014 	spin_unlock(&filp->f_lock);
1015 	return result;
1016 }
1017 
1018 struct fasync_struct *fasync_alloc(void)
1019 {
1020 	return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
1021 }
1022 
1023 /*
1024  * NOTE! This can be used only for unused fasync entries:
1025  * entries that actually got inserted on the fasync list
1026  * need to be released by rcu - see fasync_remove_entry.
1027  */
1028 void fasync_free(struct fasync_struct *new)
1029 {
1030 	kmem_cache_free(fasync_cache, new);
1031 }
1032 
1033 /*
1034  * Insert a new entry into the fasync list.  Return the pointer to the
1035  * old one if we didn't use the new one.
1036  *
1037  * NOTE! It is very important that the FASYNC flag always
1038  * match the state "is the filp on a fasync list".
1039  */
1040 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
1041 {
1042         struct fasync_struct *fa, **fp;
1043 
1044 	spin_lock(&filp->f_lock);
1045 	spin_lock(&fasync_lock);
1046 	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
1047 		if (fa->fa_file != filp)
1048 			continue;
1049 
1050 		write_lock_irq(&fa->fa_lock);
1051 		fa->fa_fd = fd;
1052 		write_unlock_irq(&fa->fa_lock);
1053 		goto out;
1054 	}
1055 
1056 	rwlock_init(&new->fa_lock);
1057 	new->magic = FASYNC_MAGIC;
1058 	new->fa_file = filp;
1059 	new->fa_fd = fd;
1060 	new->fa_next = *fapp;
1061 	rcu_assign_pointer(*fapp, new);
1062 	filp->f_flags |= FASYNC;
1063 
1064 out:
1065 	spin_unlock(&fasync_lock);
1066 	spin_unlock(&filp->f_lock);
1067 	return fa;
1068 }
1069 
1070 /*
1071  * Add a fasync entry. Return negative on error, positive if
1072  * added, and zero if did nothing but change an existing one.
1073  */
1074 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
1075 {
1076 	struct fasync_struct *new;
1077 
1078 	new = fasync_alloc();
1079 	if (!new)
1080 		return -ENOMEM;
1081 
1082 	/*
1083 	 * fasync_insert_entry() returns the old (update) entry if
1084 	 * it existed.
1085 	 *
1086 	 * So free the (unused) new entry and return 0 to let the
1087 	 * caller know that we didn't add any new fasync entries.
1088 	 */
1089 	if (fasync_insert_entry(fd, filp, fapp, new)) {
1090 		fasync_free(new);
1091 		return 0;
1092 	}
1093 
1094 	return 1;
1095 }
1096 
1097 /*
1098  * fasync_helper() is used by almost all character device drivers
1099  * to set up the fasync queue, and for regular files by the file
1100  * lease code. It returns negative on error, 0 if it did no changes
1101  * and positive if it added/deleted the entry.
1102  */
1103 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
1104 {
1105 	if (!on)
1106 		return fasync_remove_entry(filp, fapp);
1107 	return fasync_add_entry(fd, filp, fapp);
1108 }
1109 
1110 EXPORT_SYMBOL(fasync_helper);
1111 
1112 /*
1113  * rcu_read_lock() is held
1114  */
1115 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
1116 {
1117 	while (fa) {
1118 		struct fown_struct *fown;
1119 		unsigned long flags;
1120 
1121 		if (fa->magic != FASYNC_MAGIC) {
1122 			printk(KERN_ERR "kill_fasync: bad magic number in "
1123 			       "fasync_struct!\n");
1124 			return;
1125 		}
1126 		read_lock_irqsave(&fa->fa_lock, flags);
1127 		if (fa->fa_file) {
1128 			fown = file_f_owner(fa->fa_file);
1129 			if (!fown)
1130 				goto next;
1131 			/* Don't send SIGURG to processes which have not set a
1132 			   queued signum: SIGURG has its own default signalling
1133 			   mechanism. */
1134 			if (!(sig == SIGURG && fown->signum == 0))
1135 				send_sigio(fown, fa->fa_fd, band);
1136 		}
1137 next:
1138 		read_unlock_irqrestore(&fa->fa_lock, flags);
1139 		fa = rcu_dereference(fa->fa_next);
1140 	}
1141 }
1142 
1143 void kill_fasync(struct fasync_struct **fp, int sig, int band)
1144 {
1145 	/* First a quick test without locking: usually
1146 	 * the list is empty.
1147 	 */
1148 	if (*fp) {
1149 		rcu_read_lock();
1150 		kill_fasync_rcu(rcu_dereference(*fp), sig, band);
1151 		rcu_read_unlock();
1152 	}
1153 }
1154 EXPORT_SYMBOL(kill_fasync);
1155 
1156 static int __init fcntl_init(void)
1157 {
1158 	/*
1159 	 * Please add new bits here to ensure allocation uniqueness.
1160 	 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
1161 	 * is defined as O_NONBLOCK on some platforms and not on others.
1162 	 */
1163 	BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
1164 		HWEIGHT32(
1165 			(VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
1166 			__FMODE_EXEC | __FMODE_NONOTIFY));
1167 
1168 	fasync_cache = kmem_cache_create("fasync_cache",
1169 					 sizeof(struct fasync_struct), 0,
1170 					 SLAB_PANIC | SLAB_ACCOUNT, NULL);
1171 	return 0;
1172 }
1173 
1174 module_init(fcntl_init)
1175