xref: /linux/fs/fcntl.c (revision 5bb6ba448fe3598a7668838942db1f008beb581b)
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/fdtable.h>
16 #include <linux/capability.h>
17 #include <linux/dnotify.h>
18 #include <linux/slab.h>
19 #include <linux/module.h>
20 #include <linux/pipe_fs_i.h>
21 #include <linux/security.h>
22 #include <linux/ptrace.h>
23 #include <linux/signal.h>
24 #include <linux/rcupdate.h>
25 #include <linux/pid_namespace.h>
26 #include <linux/user_namespace.h>
27 #include <linux/memfd.h>
28 #include <linux/compat.h>
29 #include <linux/mount.h>
30 #include <linux/rw_hint.h>
31 
32 #include <linux/poll.h>
33 #include <asm/siginfo.h>
34 #include <linux/uaccess.h>
35 
36 #include "internal.h"
37 
38 #define SETFL_MASK (O_APPEND | O_NONBLOCK | O_NDELAY | O_DIRECT | O_NOATIME)
39 
40 static int setfl(int fd, struct file * filp, unsigned int arg)
41 {
42 	struct inode * inode = file_inode(filp);
43 	int error = 0;
44 
45 	/*
46 	 * O_APPEND cannot be cleared if the file is marked as append-only
47 	 * and the file is open for write.
48 	 */
49 	if (((arg ^ filp->f_flags) & O_APPEND) && IS_APPEND(inode))
50 		return -EPERM;
51 
52 	/* O_NOATIME can only be set by the owner or superuser */
53 	if ((arg & O_NOATIME) && !(filp->f_flags & O_NOATIME))
54 		if (!inode_owner_or_capable(file_mnt_idmap(filp), inode))
55 			return -EPERM;
56 
57 	/* required for strict SunOS emulation */
58 	if (O_NONBLOCK != O_NDELAY)
59 	       if (arg & O_NDELAY)
60 		   arg |= O_NONBLOCK;
61 
62 	/* Pipe packetized mode is controlled by O_DIRECT flag */
63 	if (!S_ISFIFO(inode->i_mode) &&
64 	    (arg & O_DIRECT) &&
65 	    !(filp->f_mode & FMODE_CAN_ODIRECT))
66 		return -EINVAL;
67 
68 	if (filp->f_op->check_flags)
69 		error = filp->f_op->check_flags(arg);
70 	if (error)
71 		return error;
72 
73 	/*
74 	 * ->fasync() is responsible for setting the FASYNC bit.
75 	 */
76 	if (((arg ^ filp->f_flags) & FASYNC) && filp->f_op->fasync) {
77 		error = filp->f_op->fasync(fd, filp, (arg & FASYNC) != 0);
78 		if (error < 0)
79 			goto out;
80 		if (error > 0)
81 			error = 0;
82 	}
83 	spin_lock(&filp->f_lock);
84 	filp->f_flags = (arg & SETFL_MASK) | (filp->f_flags & ~SETFL_MASK);
85 	filp->f_iocb_flags = iocb_flags(filp);
86 	spin_unlock(&filp->f_lock);
87 
88  out:
89 	return error;
90 }
91 
92 /*
93  * Allocate an file->f_owner struct if it doesn't exist, handling racing
94  * allocations correctly.
95  */
96 int file_f_owner_allocate(struct file *file)
97 {
98 	struct fown_struct *f_owner;
99 
100 	f_owner = file_f_owner(file);
101 	if (f_owner)
102 		return 0;
103 
104 	f_owner = kzalloc(sizeof(struct fown_struct), GFP_KERNEL);
105 	if (!f_owner)
106 		return -ENOMEM;
107 
108 	rwlock_init(&f_owner->lock);
109 	f_owner->file = file;
110 	/* If someone else raced us, drop our allocation. */
111 	if (unlikely(cmpxchg(&file->f_owner, NULL, f_owner)))
112 		kfree(f_owner);
113 	return 0;
114 }
115 EXPORT_SYMBOL(file_f_owner_allocate);
116 
117 void file_f_owner_release(struct file *file)
118 {
119 	struct fown_struct *f_owner;
120 
121 	f_owner = file_f_owner(file);
122 	if (f_owner) {
123 		put_pid(f_owner->pid);
124 		kfree(f_owner);
125 	}
126 }
127 
128 void __f_setown(struct file *filp, struct pid *pid, enum pid_type type,
129 		int force)
130 {
131 	struct fown_struct *f_owner;
132 
133 	f_owner = file_f_owner(filp);
134 	if (WARN_ON_ONCE(!f_owner))
135 		return;
136 
137 	write_lock_irq(&f_owner->lock);
138 	if (force || !f_owner->pid) {
139 		put_pid(f_owner->pid);
140 		f_owner->pid = get_pid(pid);
141 		f_owner->pid_type = type;
142 
143 		if (pid) {
144 			const struct cred *cred = current_cred();
145 			security_file_set_fowner(filp);
146 			f_owner->uid = cred->uid;
147 			f_owner->euid = cred->euid;
148 		}
149 	}
150 	write_unlock_irq(&f_owner->lock);
151 }
152 EXPORT_SYMBOL(__f_setown);
153 
154 int f_setown(struct file *filp, int who, int force)
155 {
156 	enum pid_type type;
157 	struct pid *pid = NULL;
158 	int ret = 0;
159 
160 	might_sleep();
161 
162 	type = PIDTYPE_TGID;
163 	if (who < 0) {
164 		/* avoid overflow below */
165 		if (who == INT_MIN)
166 			return -EINVAL;
167 
168 		type = PIDTYPE_PGID;
169 		who = -who;
170 	}
171 
172 	ret = file_f_owner_allocate(filp);
173 	if (ret)
174 		return ret;
175 
176 	rcu_read_lock();
177 	if (who) {
178 		pid = find_vpid(who);
179 		if (!pid)
180 			ret = -ESRCH;
181 	}
182 
183 	if (!ret)
184 		__f_setown(filp, pid, type, force);
185 	rcu_read_unlock();
186 
187 	return ret;
188 }
189 EXPORT_SYMBOL(f_setown);
190 
191 void f_delown(struct file *filp)
192 {
193 	__f_setown(filp, NULL, PIDTYPE_TGID, 1);
194 }
195 
196 pid_t f_getown(struct file *filp)
197 {
198 	pid_t pid = 0;
199 	struct fown_struct *f_owner;
200 
201 	f_owner = file_f_owner(filp);
202 	if (!f_owner)
203 		return pid;
204 
205 	read_lock_irq(&f_owner->lock);
206 	rcu_read_lock();
207 	if (pid_task(f_owner->pid, f_owner->pid_type)) {
208 		pid = pid_vnr(f_owner->pid);
209 		if (f_owner->pid_type == PIDTYPE_PGID)
210 			pid = -pid;
211 	}
212 	rcu_read_unlock();
213 	read_unlock_irq(&f_owner->lock);
214 	return pid;
215 }
216 
217 static int f_setown_ex(struct file *filp, unsigned long arg)
218 {
219 	struct f_owner_ex __user *owner_p = (void __user *)arg;
220 	struct f_owner_ex owner;
221 	struct pid *pid;
222 	int type;
223 	int ret;
224 
225 	ret = copy_from_user(&owner, owner_p, sizeof(owner));
226 	if (ret)
227 		return -EFAULT;
228 
229 	switch (owner.type) {
230 	case F_OWNER_TID:
231 		type = PIDTYPE_PID;
232 		break;
233 
234 	case F_OWNER_PID:
235 		type = PIDTYPE_TGID;
236 		break;
237 
238 	case F_OWNER_PGRP:
239 		type = PIDTYPE_PGID;
240 		break;
241 
242 	default:
243 		return -EINVAL;
244 	}
245 
246 	ret = file_f_owner_allocate(filp);
247 	if (ret)
248 		return ret;
249 
250 	rcu_read_lock();
251 	pid = find_vpid(owner.pid);
252 	if (owner.pid && !pid)
253 		ret = -ESRCH;
254 	else
255 		 __f_setown(filp, pid, type, 1);
256 	rcu_read_unlock();
257 
258 	return ret;
259 }
260 
261 static int f_getown_ex(struct file *filp, unsigned long arg)
262 {
263 	struct f_owner_ex __user *owner_p = (void __user *)arg;
264 	struct f_owner_ex owner = {};
265 	int ret = 0;
266 	struct fown_struct *f_owner;
267 	enum pid_type pid_type = PIDTYPE_PID;
268 
269 	f_owner = file_f_owner(filp);
270 	if (f_owner) {
271 		read_lock_irq(&f_owner->lock);
272 		rcu_read_lock();
273 		if (pid_task(f_owner->pid, f_owner->pid_type))
274 			owner.pid = pid_vnr(f_owner->pid);
275 		rcu_read_unlock();
276 		pid_type = f_owner->pid_type;
277 	}
278 
279 	switch (pid_type) {
280 	case PIDTYPE_PID:
281 		owner.type = F_OWNER_TID;
282 		break;
283 
284 	case PIDTYPE_TGID:
285 		owner.type = F_OWNER_PID;
286 		break;
287 
288 	case PIDTYPE_PGID:
289 		owner.type = F_OWNER_PGRP;
290 		break;
291 
292 	default:
293 		WARN_ON(1);
294 		ret = -EINVAL;
295 		break;
296 	}
297 	if (f_owner)
298 		read_unlock_irq(&f_owner->lock);
299 
300 	if (!ret) {
301 		ret = copy_to_user(owner_p, &owner, sizeof(owner));
302 		if (ret)
303 			ret = -EFAULT;
304 	}
305 	return ret;
306 }
307 
308 #ifdef CONFIG_CHECKPOINT_RESTORE
309 static int f_getowner_uids(struct file *filp, unsigned long arg)
310 {
311 	struct user_namespace *user_ns = current_user_ns();
312 	struct fown_struct *f_owner;
313 	uid_t __user *dst = (void __user *)arg;
314 	uid_t src[2] = {0, 0};
315 	int err;
316 
317 	f_owner = file_f_owner(filp);
318 	if (f_owner) {
319 		read_lock_irq(&f_owner->lock);
320 		src[0] = from_kuid(user_ns, f_owner->uid);
321 		src[1] = from_kuid(user_ns, f_owner->euid);
322 		read_unlock_irq(&f_owner->lock);
323 	}
324 
325 	err  = put_user(src[0], &dst[0]);
326 	err |= put_user(src[1], &dst[1]);
327 
328 	return err;
329 }
330 #else
331 static int f_getowner_uids(struct file *filp, unsigned long arg)
332 {
333 	return -EINVAL;
334 }
335 #endif
336 
337 static bool rw_hint_valid(u64 hint)
338 {
339 	BUILD_BUG_ON(WRITE_LIFE_NOT_SET != RWH_WRITE_LIFE_NOT_SET);
340 	BUILD_BUG_ON(WRITE_LIFE_NONE != RWH_WRITE_LIFE_NONE);
341 	BUILD_BUG_ON(WRITE_LIFE_SHORT != RWH_WRITE_LIFE_SHORT);
342 	BUILD_BUG_ON(WRITE_LIFE_MEDIUM != RWH_WRITE_LIFE_MEDIUM);
343 	BUILD_BUG_ON(WRITE_LIFE_LONG != RWH_WRITE_LIFE_LONG);
344 	BUILD_BUG_ON(WRITE_LIFE_EXTREME != RWH_WRITE_LIFE_EXTREME);
345 
346 	switch (hint) {
347 	case RWH_WRITE_LIFE_NOT_SET:
348 	case RWH_WRITE_LIFE_NONE:
349 	case RWH_WRITE_LIFE_SHORT:
350 	case RWH_WRITE_LIFE_MEDIUM:
351 	case RWH_WRITE_LIFE_LONG:
352 	case RWH_WRITE_LIFE_EXTREME:
353 		return true;
354 	default:
355 		return false;
356 	}
357 }
358 
359 static long fcntl_get_rw_hint(struct file *file, unsigned int cmd,
360 			      unsigned long arg)
361 {
362 	struct inode *inode = file_inode(file);
363 	u64 __user *argp = (u64 __user *)arg;
364 	u64 hint = READ_ONCE(inode->i_write_hint);
365 
366 	if (copy_to_user(argp, &hint, sizeof(*argp)))
367 		return -EFAULT;
368 	return 0;
369 }
370 
371 static long fcntl_set_rw_hint(struct file *file, unsigned int cmd,
372 			      unsigned long arg)
373 {
374 	struct inode *inode = file_inode(file);
375 	u64 __user *argp = (u64 __user *)arg;
376 	u64 hint;
377 
378 	if (copy_from_user(&hint, argp, sizeof(hint)))
379 		return -EFAULT;
380 	if (!rw_hint_valid(hint))
381 		return -EINVAL;
382 
383 	WRITE_ONCE(inode->i_write_hint, hint);
384 
385 	/*
386 	 * file->f_mapping->host may differ from inode. As an example,
387 	 * blkdev_open() modifies file->f_mapping.
388 	 */
389 	if (file->f_mapping->host != inode)
390 		WRITE_ONCE(file->f_mapping->host->i_write_hint, hint);
391 
392 	return 0;
393 }
394 
395 /* Is the file descriptor a dup of the file? */
396 static long f_dupfd_query(int fd, struct file *filp)
397 {
398 	CLASS(fd_raw, f)(fd);
399 
400 	if (fd_empty(f))
401 		return -EBADF;
402 
403 	/*
404 	 * We can do the 'fdput()' immediately, as the only thing that
405 	 * matters is the pointer value which isn't changed by the fdput.
406 	 *
407 	 * Technically we didn't need a ref at all, and 'fdget()' was
408 	 * overkill, but given our lockless file pointer lookup, the
409 	 * alternatives are complicated.
410 	 */
411 	return fd_file(f) == filp;
412 }
413 
414 /* Let the caller figure out whether a given file was just created. */
415 static long f_created_query(const struct file *filp)
416 {
417 	return !!(filp->f_mode & FMODE_CREATED);
418 }
419 
420 static int f_owner_sig(struct file *filp, int signum, bool setsig)
421 {
422 	int ret = 0;
423 	struct fown_struct *f_owner;
424 
425 	might_sleep();
426 
427 	if (setsig) {
428 		if (!valid_signal(signum))
429 			return -EINVAL;
430 
431 		ret = file_f_owner_allocate(filp);
432 		if (ret)
433 			return ret;
434 	}
435 
436 	f_owner = file_f_owner(filp);
437 	if (setsig)
438 		f_owner->signum = signum;
439 	else if (f_owner)
440 		ret = f_owner->signum;
441 	return ret;
442 }
443 
444 static long do_fcntl(int fd, unsigned int cmd, unsigned long arg,
445 		struct file *filp)
446 {
447 	void __user *argp = (void __user *)arg;
448 	int argi = (int)arg;
449 	struct flock flock;
450 	long err = -EINVAL;
451 
452 	switch (cmd) {
453 	case F_CREATED_QUERY:
454 		err = f_created_query(filp);
455 		break;
456 	case F_DUPFD:
457 		err = f_dupfd(argi, filp, 0);
458 		break;
459 	case F_DUPFD_CLOEXEC:
460 		err = f_dupfd(argi, filp, O_CLOEXEC);
461 		break;
462 	case F_DUPFD_QUERY:
463 		err = f_dupfd_query(argi, filp);
464 		break;
465 	case F_GETFD:
466 		err = get_close_on_exec(fd) ? FD_CLOEXEC : 0;
467 		break;
468 	case F_SETFD:
469 		err = 0;
470 		set_close_on_exec(fd, argi & FD_CLOEXEC);
471 		break;
472 	case F_GETFL:
473 		err = filp->f_flags;
474 		break;
475 	case F_SETFL:
476 		err = setfl(fd, filp, argi);
477 		break;
478 #if BITS_PER_LONG != 32
479 	/* 32-bit arches must use fcntl64() */
480 	case F_OFD_GETLK:
481 #endif
482 	case F_GETLK:
483 		if (copy_from_user(&flock, argp, sizeof(flock)))
484 			return -EFAULT;
485 		err = fcntl_getlk(filp, cmd, &flock);
486 		if (!err && copy_to_user(argp, &flock, sizeof(flock)))
487 			return -EFAULT;
488 		break;
489 #if BITS_PER_LONG != 32
490 	/* 32-bit arches must use fcntl64() */
491 	case F_OFD_SETLK:
492 	case F_OFD_SETLKW:
493 		fallthrough;
494 #endif
495 	case F_SETLK:
496 	case F_SETLKW:
497 		if (copy_from_user(&flock, argp, sizeof(flock)))
498 			return -EFAULT;
499 		err = fcntl_setlk(fd, filp, cmd, &flock);
500 		break;
501 	case F_GETOWN:
502 		/*
503 		 * XXX If f_owner is a process group, the
504 		 * negative return value will get converted
505 		 * into an error.  Oops.  If we keep the
506 		 * current syscall conventions, the only way
507 		 * to fix this will be in libc.
508 		 */
509 		err = f_getown(filp);
510 		force_successful_syscall_return();
511 		break;
512 	case F_SETOWN:
513 		err = f_setown(filp, argi, 1);
514 		break;
515 	case F_GETOWN_EX:
516 		err = f_getown_ex(filp, arg);
517 		break;
518 	case F_SETOWN_EX:
519 		err = f_setown_ex(filp, arg);
520 		break;
521 	case F_GETOWNER_UIDS:
522 		err = f_getowner_uids(filp, arg);
523 		break;
524 	case F_GETSIG:
525 		err = f_owner_sig(filp, 0, false);
526 		break;
527 	case F_SETSIG:
528 		err = f_owner_sig(filp, argi, true);
529 		break;
530 	case F_GETLEASE:
531 		err = fcntl_getlease(filp);
532 		break;
533 	case F_SETLEASE:
534 		err = fcntl_setlease(fd, filp, argi);
535 		break;
536 	case F_NOTIFY:
537 		err = fcntl_dirnotify(fd, filp, argi);
538 		break;
539 	case F_SETPIPE_SZ:
540 	case F_GETPIPE_SZ:
541 		err = pipe_fcntl(filp, cmd, argi);
542 		break;
543 	case F_ADD_SEALS:
544 	case F_GET_SEALS:
545 		err = memfd_fcntl(filp, cmd, argi);
546 		break;
547 	case F_GET_RW_HINT:
548 		err = fcntl_get_rw_hint(filp, cmd, arg);
549 		break;
550 	case F_SET_RW_HINT:
551 		err = fcntl_set_rw_hint(filp, cmd, arg);
552 		break;
553 	default:
554 		break;
555 	}
556 	return err;
557 }
558 
559 static int check_fcntl_cmd(unsigned cmd)
560 {
561 	switch (cmd) {
562 	case F_CREATED_QUERY:
563 	case F_DUPFD:
564 	case F_DUPFD_CLOEXEC:
565 	case F_DUPFD_QUERY:
566 	case F_GETFD:
567 	case F_SETFD:
568 	case F_GETFL:
569 		return 1;
570 	}
571 	return 0;
572 }
573 
574 SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd, unsigned long, arg)
575 {
576 	struct fd f = fdget_raw(fd);
577 	long err = -EBADF;
578 
579 	if (!fd_file(f))
580 		goto out;
581 
582 	if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
583 		if (!check_fcntl_cmd(cmd))
584 			goto out1;
585 	}
586 
587 	err = security_file_fcntl(fd_file(f), cmd, arg);
588 	if (!err)
589 		err = do_fcntl(fd, cmd, arg, fd_file(f));
590 
591 out1:
592  	fdput(f);
593 out:
594 	return err;
595 }
596 
597 #if BITS_PER_LONG == 32
598 SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
599 		unsigned long, arg)
600 {
601 	void __user *argp = (void __user *)arg;
602 	struct fd f = fdget_raw(fd);
603 	struct flock64 flock;
604 	long err = -EBADF;
605 
606 	if (!fd_file(f))
607 		goto out;
608 
609 	if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
610 		if (!check_fcntl_cmd(cmd))
611 			goto out1;
612 	}
613 
614 	err = security_file_fcntl(fd_file(f), cmd, arg);
615 	if (err)
616 		goto out1;
617 
618 	switch (cmd) {
619 	case F_GETLK64:
620 	case F_OFD_GETLK:
621 		err = -EFAULT;
622 		if (copy_from_user(&flock, argp, sizeof(flock)))
623 			break;
624 		err = fcntl_getlk64(fd_file(f), cmd, &flock);
625 		if (!err && copy_to_user(argp, &flock, sizeof(flock)))
626 			err = -EFAULT;
627 		break;
628 	case F_SETLK64:
629 	case F_SETLKW64:
630 	case F_OFD_SETLK:
631 	case F_OFD_SETLKW:
632 		err = -EFAULT;
633 		if (copy_from_user(&flock, argp, sizeof(flock)))
634 			break;
635 		err = fcntl_setlk64(fd, fd_file(f), cmd, &flock);
636 		break;
637 	default:
638 		err = do_fcntl(fd, cmd, arg, fd_file(f));
639 		break;
640 	}
641 out1:
642 	fdput(f);
643 out:
644 	return err;
645 }
646 #endif
647 
648 #ifdef CONFIG_COMPAT
649 /* careful - don't use anywhere else */
650 #define copy_flock_fields(dst, src)		\
651 	(dst)->l_type = (src)->l_type;		\
652 	(dst)->l_whence = (src)->l_whence;	\
653 	(dst)->l_start = (src)->l_start;	\
654 	(dst)->l_len = (src)->l_len;		\
655 	(dst)->l_pid = (src)->l_pid;
656 
657 static int get_compat_flock(struct flock *kfl, const struct compat_flock __user *ufl)
658 {
659 	struct compat_flock fl;
660 
661 	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock)))
662 		return -EFAULT;
663 	copy_flock_fields(kfl, &fl);
664 	return 0;
665 }
666 
667 static int get_compat_flock64(struct flock *kfl, const struct compat_flock64 __user *ufl)
668 {
669 	struct compat_flock64 fl;
670 
671 	if (copy_from_user(&fl, ufl, sizeof(struct compat_flock64)))
672 		return -EFAULT;
673 	copy_flock_fields(kfl, &fl);
674 	return 0;
675 }
676 
677 static int put_compat_flock(const struct flock *kfl, struct compat_flock __user *ufl)
678 {
679 	struct compat_flock fl;
680 
681 	memset(&fl, 0, sizeof(struct compat_flock));
682 	copy_flock_fields(&fl, kfl);
683 	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock)))
684 		return -EFAULT;
685 	return 0;
686 }
687 
688 static int put_compat_flock64(const struct flock *kfl, struct compat_flock64 __user *ufl)
689 {
690 	struct compat_flock64 fl;
691 
692 	BUILD_BUG_ON(sizeof(kfl->l_start) > sizeof(ufl->l_start));
693 	BUILD_BUG_ON(sizeof(kfl->l_len) > sizeof(ufl->l_len));
694 
695 	memset(&fl, 0, sizeof(struct compat_flock64));
696 	copy_flock_fields(&fl, kfl);
697 	if (copy_to_user(ufl, &fl, sizeof(struct compat_flock64)))
698 		return -EFAULT;
699 	return 0;
700 }
701 #undef copy_flock_fields
702 
703 static unsigned int
704 convert_fcntl_cmd(unsigned int cmd)
705 {
706 	switch (cmd) {
707 	case F_GETLK64:
708 		return F_GETLK;
709 	case F_SETLK64:
710 		return F_SETLK;
711 	case F_SETLKW64:
712 		return F_SETLKW;
713 	}
714 
715 	return cmd;
716 }
717 
718 /*
719  * GETLK was successful and we need to return the data, but it needs to fit in
720  * the compat structure.
721  * l_start shouldn't be too big, unless the original start + end is greater than
722  * COMPAT_OFF_T_MAX, in which case the app was asking for trouble, so we return
723  * -EOVERFLOW in that case.  l_len could be too big, in which case we just
724  * truncate it, and only allow the app to see that part of the conflicting lock
725  * that might make sense to it anyway
726  */
727 static int fixup_compat_flock(struct flock *flock)
728 {
729 	if (flock->l_start > COMPAT_OFF_T_MAX)
730 		return -EOVERFLOW;
731 	if (flock->l_len > COMPAT_OFF_T_MAX)
732 		flock->l_len = COMPAT_OFF_T_MAX;
733 	return 0;
734 }
735 
736 static long do_compat_fcntl64(unsigned int fd, unsigned int cmd,
737 			     compat_ulong_t arg)
738 {
739 	struct fd f = fdget_raw(fd);
740 	struct flock flock;
741 	long err = -EBADF;
742 
743 	if (!fd_file(f))
744 		return err;
745 
746 	if (unlikely(fd_file(f)->f_mode & FMODE_PATH)) {
747 		if (!check_fcntl_cmd(cmd))
748 			goto out_put;
749 	}
750 
751 	err = security_file_fcntl(fd_file(f), cmd, arg);
752 	if (err)
753 		goto out_put;
754 
755 	switch (cmd) {
756 	case F_GETLK:
757 		err = get_compat_flock(&flock, compat_ptr(arg));
758 		if (err)
759 			break;
760 		err = fcntl_getlk(fd_file(f), convert_fcntl_cmd(cmd), &flock);
761 		if (err)
762 			break;
763 		err = fixup_compat_flock(&flock);
764 		if (!err)
765 			err = put_compat_flock(&flock, compat_ptr(arg));
766 		break;
767 	case F_GETLK64:
768 	case F_OFD_GETLK:
769 		err = get_compat_flock64(&flock, compat_ptr(arg));
770 		if (err)
771 			break;
772 		err = fcntl_getlk(fd_file(f), convert_fcntl_cmd(cmd), &flock);
773 		if (!err)
774 			err = put_compat_flock64(&flock, compat_ptr(arg));
775 		break;
776 	case F_SETLK:
777 	case F_SETLKW:
778 		err = get_compat_flock(&flock, compat_ptr(arg));
779 		if (err)
780 			break;
781 		err = fcntl_setlk(fd, fd_file(f), convert_fcntl_cmd(cmd), &flock);
782 		break;
783 	case F_SETLK64:
784 	case F_SETLKW64:
785 	case F_OFD_SETLK:
786 	case F_OFD_SETLKW:
787 		err = get_compat_flock64(&flock, compat_ptr(arg));
788 		if (err)
789 			break;
790 		err = fcntl_setlk(fd, fd_file(f), convert_fcntl_cmd(cmd), &flock);
791 		break;
792 	default:
793 		err = do_fcntl(fd, cmd, arg, fd_file(f));
794 		break;
795 	}
796 out_put:
797 	fdput(f);
798 	return err;
799 }
800 
801 COMPAT_SYSCALL_DEFINE3(fcntl64, unsigned int, fd, unsigned int, cmd,
802 		       compat_ulong_t, arg)
803 {
804 	return do_compat_fcntl64(fd, cmd, arg);
805 }
806 
807 COMPAT_SYSCALL_DEFINE3(fcntl, unsigned int, fd, unsigned int, cmd,
808 		       compat_ulong_t, arg)
809 {
810 	switch (cmd) {
811 	case F_GETLK64:
812 	case F_SETLK64:
813 	case F_SETLKW64:
814 	case F_OFD_GETLK:
815 	case F_OFD_SETLK:
816 	case F_OFD_SETLKW:
817 		return -EINVAL;
818 	}
819 	return do_compat_fcntl64(fd, cmd, arg);
820 }
821 #endif
822 
823 /* Table to convert sigio signal codes into poll band bitmaps */
824 
825 static const __poll_t band_table[NSIGPOLL] = {
826 	EPOLLIN | EPOLLRDNORM,			/* POLL_IN */
827 	EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND,	/* POLL_OUT */
828 	EPOLLIN | EPOLLRDNORM | EPOLLMSG,		/* POLL_MSG */
829 	EPOLLERR,				/* POLL_ERR */
830 	EPOLLPRI | EPOLLRDBAND,			/* POLL_PRI */
831 	EPOLLHUP | EPOLLERR			/* POLL_HUP */
832 };
833 
834 static inline int sigio_perm(struct task_struct *p,
835                              struct fown_struct *fown, int sig)
836 {
837 	const struct cred *cred;
838 	int ret;
839 
840 	rcu_read_lock();
841 	cred = __task_cred(p);
842 	ret = ((uid_eq(fown->euid, GLOBAL_ROOT_UID) ||
843 		uid_eq(fown->euid, cred->suid) || uid_eq(fown->euid, cred->uid) ||
844 		uid_eq(fown->uid,  cred->suid) || uid_eq(fown->uid,  cred->uid)) &&
845 	       !security_file_send_sigiotask(p, fown, sig));
846 	rcu_read_unlock();
847 	return ret;
848 }
849 
850 static void send_sigio_to_task(struct task_struct *p,
851 			       struct fown_struct *fown,
852 			       int fd, int reason, enum pid_type type)
853 {
854 	/*
855 	 * F_SETSIG can change ->signum lockless in parallel, make
856 	 * sure we read it once and use the same value throughout.
857 	 */
858 	int signum = READ_ONCE(fown->signum);
859 
860 	if (!sigio_perm(p, fown, signum))
861 		return;
862 
863 	switch (signum) {
864 		default: {
865 			kernel_siginfo_t si;
866 
867 			/* Queue a rt signal with the appropriate fd as its
868 			   value.  We use SI_SIGIO as the source, not
869 			   SI_KERNEL, since kernel signals always get
870 			   delivered even if we can't queue.  Failure to
871 			   queue in this case _should_ be reported; we fall
872 			   back to SIGIO in that case. --sct */
873 			clear_siginfo(&si);
874 			si.si_signo = signum;
875 			si.si_errno = 0;
876 		        si.si_code  = reason;
877 			/*
878 			 * Posix definies POLL_IN and friends to be signal
879 			 * specific si_codes for SIG_POLL.  Linux extended
880 			 * these si_codes to other signals in a way that is
881 			 * ambiguous if other signals also have signal
882 			 * specific si_codes.  In that case use SI_SIGIO instead
883 			 * to remove the ambiguity.
884 			 */
885 			if ((signum != SIGPOLL) && sig_specific_sicodes(signum))
886 				si.si_code = SI_SIGIO;
887 
888 			/* Make sure we are called with one of the POLL_*
889 			   reasons, otherwise we could leak kernel stack into
890 			   userspace.  */
891 			BUG_ON((reason < POLL_IN) || ((reason - POLL_IN) >= NSIGPOLL));
892 			if (reason - POLL_IN >= NSIGPOLL)
893 				si.si_band  = ~0L;
894 			else
895 				si.si_band = mangle_poll(band_table[reason - POLL_IN]);
896 			si.si_fd    = fd;
897 			if (!do_send_sig_info(signum, &si, p, type))
898 				break;
899 		}
900 			fallthrough;	/* fall back on the old plain SIGIO signal */
901 		case 0:
902 			do_send_sig_info(SIGIO, SEND_SIG_PRIV, p, type);
903 	}
904 }
905 
906 void send_sigio(struct fown_struct *fown, int fd, int band)
907 {
908 	struct task_struct *p;
909 	enum pid_type type;
910 	unsigned long flags;
911 	struct pid *pid;
912 
913 	read_lock_irqsave(&fown->lock, flags);
914 
915 	type = fown->pid_type;
916 	pid = fown->pid;
917 	if (!pid)
918 		goto out_unlock_fown;
919 
920 	if (type <= PIDTYPE_TGID) {
921 		rcu_read_lock();
922 		p = pid_task(pid, PIDTYPE_PID);
923 		if (p)
924 			send_sigio_to_task(p, fown, fd, band, type);
925 		rcu_read_unlock();
926 	} else {
927 		read_lock(&tasklist_lock);
928 		do_each_pid_task(pid, type, p) {
929 			send_sigio_to_task(p, fown, fd, band, type);
930 		} while_each_pid_task(pid, type, p);
931 		read_unlock(&tasklist_lock);
932 	}
933  out_unlock_fown:
934 	read_unlock_irqrestore(&fown->lock, flags);
935 }
936 
937 static void send_sigurg_to_task(struct task_struct *p,
938 				struct fown_struct *fown, enum pid_type type)
939 {
940 	if (sigio_perm(p, fown, SIGURG))
941 		do_send_sig_info(SIGURG, SEND_SIG_PRIV, p, type);
942 }
943 
944 int send_sigurg(struct file *file)
945 {
946 	struct fown_struct *fown;
947 	struct task_struct *p;
948 	enum pid_type type;
949 	struct pid *pid;
950 	unsigned long flags;
951 	int ret = 0;
952 
953 	fown = file_f_owner(file);
954 	if (!fown)
955 		return 0;
956 
957 	read_lock_irqsave(&fown->lock, flags);
958 
959 	type = fown->pid_type;
960 	pid = fown->pid;
961 	if (!pid)
962 		goto out_unlock_fown;
963 
964 	ret = 1;
965 
966 	if (type <= PIDTYPE_TGID) {
967 		rcu_read_lock();
968 		p = pid_task(pid, PIDTYPE_PID);
969 		if (p)
970 			send_sigurg_to_task(p, fown, type);
971 		rcu_read_unlock();
972 	} else {
973 		read_lock(&tasklist_lock);
974 		do_each_pid_task(pid, type, p) {
975 			send_sigurg_to_task(p, fown, type);
976 		} while_each_pid_task(pid, type, p);
977 		read_unlock(&tasklist_lock);
978 	}
979  out_unlock_fown:
980 	read_unlock_irqrestore(&fown->lock, flags);
981 	return ret;
982 }
983 
984 static DEFINE_SPINLOCK(fasync_lock);
985 static struct kmem_cache *fasync_cache __ro_after_init;
986 
987 /*
988  * Remove a fasync entry. If successfully removed, return
989  * positive and clear the FASYNC flag. If no entry exists,
990  * do nothing and return 0.
991  *
992  * NOTE! It is very important that the FASYNC flag always
993  * match the state "is the filp on a fasync list".
994  *
995  */
996 int fasync_remove_entry(struct file *filp, struct fasync_struct **fapp)
997 {
998 	struct fasync_struct *fa, **fp;
999 	int result = 0;
1000 
1001 	spin_lock(&filp->f_lock);
1002 	spin_lock(&fasync_lock);
1003 	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
1004 		if (fa->fa_file != filp)
1005 			continue;
1006 
1007 		write_lock_irq(&fa->fa_lock);
1008 		fa->fa_file = NULL;
1009 		write_unlock_irq(&fa->fa_lock);
1010 
1011 		*fp = fa->fa_next;
1012 		kfree_rcu(fa, fa_rcu);
1013 		filp->f_flags &= ~FASYNC;
1014 		result = 1;
1015 		break;
1016 	}
1017 	spin_unlock(&fasync_lock);
1018 	spin_unlock(&filp->f_lock);
1019 	return result;
1020 }
1021 
1022 struct fasync_struct *fasync_alloc(void)
1023 {
1024 	return kmem_cache_alloc(fasync_cache, GFP_KERNEL);
1025 }
1026 
1027 /*
1028  * NOTE! This can be used only for unused fasync entries:
1029  * entries that actually got inserted on the fasync list
1030  * need to be released by rcu - see fasync_remove_entry.
1031  */
1032 void fasync_free(struct fasync_struct *new)
1033 {
1034 	kmem_cache_free(fasync_cache, new);
1035 }
1036 
1037 /*
1038  * Insert a new entry into the fasync list.  Return the pointer to the
1039  * old one if we didn't use the new one.
1040  *
1041  * NOTE! It is very important that the FASYNC flag always
1042  * match the state "is the filp on a fasync list".
1043  */
1044 struct fasync_struct *fasync_insert_entry(int fd, struct file *filp, struct fasync_struct **fapp, struct fasync_struct *new)
1045 {
1046         struct fasync_struct *fa, **fp;
1047 
1048 	spin_lock(&filp->f_lock);
1049 	spin_lock(&fasync_lock);
1050 	for (fp = fapp; (fa = *fp) != NULL; fp = &fa->fa_next) {
1051 		if (fa->fa_file != filp)
1052 			continue;
1053 
1054 		write_lock_irq(&fa->fa_lock);
1055 		fa->fa_fd = fd;
1056 		write_unlock_irq(&fa->fa_lock);
1057 		goto out;
1058 	}
1059 
1060 	rwlock_init(&new->fa_lock);
1061 	new->magic = FASYNC_MAGIC;
1062 	new->fa_file = filp;
1063 	new->fa_fd = fd;
1064 	new->fa_next = *fapp;
1065 	rcu_assign_pointer(*fapp, new);
1066 	filp->f_flags |= FASYNC;
1067 
1068 out:
1069 	spin_unlock(&fasync_lock);
1070 	spin_unlock(&filp->f_lock);
1071 	return fa;
1072 }
1073 
1074 /*
1075  * Add a fasync entry. Return negative on error, positive if
1076  * added, and zero if did nothing but change an existing one.
1077  */
1078 static int fasync_add_entry(int fd, struct file *filp, struct fasync_struct **fapp)
1079 {
1080 	struct fasync_struct *new;
1081 
1082 	new = fasync_alloc();
1083 	if (!new)
1084 		return -ENOMEM;
1085 
1086 	/*
1087 	 * fasync_insert_entry() returns the old (update) entry if
1088 	 * it existed.
1089 	 *
1090 	 * So free the (unused) new entry and return 0 to let the
1091 	 * caller know that we didn't add any new fasync entries.
1092 	 */
1093 	if (fasync_insert_entry(fd, filp, fapp, new)) {
1094 		fasync_free(new);
1095 		return 0;
1096 	}
1097 
1098 	return 1;
1099 }
1100 
1101 /*
1102  * fasync_helper() is used by almost all character device drivers
1103  * to set up the fasync queue, and for regular files by the file
1104  * lease code. It returns negative on error, 0 if it did no changes
1105  * and positive if it added/deleted the entry.
1106  */
1107 int fasync_helper(int fd, struct file * filp, int on, struct fasync_struct **fapp)
1108 {
1109 	if (!on)
1110 		return fasync_remove_entry(filp, fapp);
1111 	return fasync_add_entry(fd, filp, fapp);
1112 }
1113 
1114 EXPORT_SYMBOL(fasync_helper);
1115 
1116 /*
1117  * rcu_read_lock() is held
1118  */
1119 static void kill_fasync_rcu(struct fasync_struct *fa, int sig, int band)
1120 {
1121 	while (fa) {
1122 		struct fown_struct *fown;
1123 		unsigned long flags;
1124 
1125 		if (fa->magic != FASYNC_MAGIC) {
1126 			printk(KERN_ERR "kill_fasync: bad magic number in "
1127 			       "fasync_struct!\n");
1128 			return;
1129 		}
1130 		read_lock_irqsave(&fa->fa_lock, flags);
1131 		if (fa->fa_file) {
1132 			fown = file_f_owner(fa->fa_file);
1133 			if (!fown)
1134 				goto next;
1135 			/* Don't send SIGURG to processes which have not set a
1136 			   queued signum: SIGURG has its own default signalling
1137 			   mechanism. */
1138 			if (!(sig == SIGURG && fown->signum == 0))
1139 				send_sigio(fown, fa->fa_fd, band);
1140 		}
1141 next:
1142 		read_unlock_irqrestore(&fa->fa_lock, flags);
1143 		fa = rcu_dereference(fa->fa_next);
1144 	}
1145 }
1146 
1147 void kill_fasync(struct fasync_struct **fp, int sig, int band)
1148 {
1149 	/* First a quick test without locking: usually
1150 	 * the list is empty.
1151 	 */
1152 	if (*fp) {
1153 		rcu_read_lock();
1154 		kill_fasync_rcu(rcu_dereference(*fp), sig, band);
1155 		rcu_read_unlock();
1156 	}
1157 }
1158 EXPORT_SYMBOL(kill_fasync);
1159 
1160 static int __init fcntl_init(void)
1161 {
1162 	/*
1163 	 * Please add new bits here to ensure allocation uniqueness.
1164 	 * Exceptions: O_NONBLOCK is a two bit define on parisc; O_NDELAY
1165 	 * is defined as O_NONBLOCK on some platforms and not on others.
1166 	 */
1167 	BUILD_BUG_ON(21 - 1 /* for O_RDONLY being 0 */ !=
1168 		HWEIGHT32(
1169 			(VALID_OPEN_FLAGS & ~(O_NONBLOCK | O_NDELAY)) |
1170 			__FMODE_EXEC | __FMODE_NONOTIFY));
1171 
1172 	fasync_cache = kmem_cache_create("fasync_cache",
1173 					 sizeof(struct fasync_struct), 0,
1174 					 SLAB_PANIC | SLAB_ACCOUNT, NULL);
1175 	return 0;
1176 }
1177 
1178 module_init(fcntl_init)
1179