xref: /freebsd/sys/compat/linuxkpi/common/src/linux_compat.c (revision 76cc4c20473495c839c4df9cfe0bf46632738f03)
1 /*-
2  * Copyright (c) 2010 Isilon Systems, Inc.
3  * Copyright (c) 2010 iX Systems, Inc.
4  * Copyright (c) 2010 Panasas, Inc.
5  * Copyright (c) 2013-2018 Mellanox Technologies, Ltd.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice unmodified, this list of conditions, and the following
13  *    disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28  */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD$");
32 
33 #include "opt_stack.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/malloc.h>
38 #include <sys/kernel.h>
39 #include <sys/sysctl.h>
40 #include <sys/proc.h>
41 #include <sys/sglist.h>
42 #include <sys/sleepqueue.h>
43 #include <sys/refcount.h>
44 #include <sys/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/bus.h>
47 #include <sys/eventhandler.h>
48 #include <sys/fcntl.h>
49 #include <sys/file.h>
50 #include <sys/filio.h>
51 #include <sys/rwlock.h>
52 #include <sys/mman.h>
53 #include <sys/stack.h>
54 #include <sys/user.h>
55 
56 #include <vm/vm.h>
57 #include <vm/pmap.h>
58 #include <vm/vm_object.h>
59 #include <vm/vm_page.h>
60 #include <vm/vm_pager.h>
61 
62 #include <machine/stdarg.h>
63 
64 #if defined(__i386__) || defined(__amd64__)
65 #include <machine/md_var.h>
66 #endif
67 
68 #include <linux/kobject.h>
69 #include <linux/device.h>
70 #include <linux/slab.h>
71 #include <linux/module.h>
72 #include <linux/moduleparam.h>
73 #include <linux/cdev.h>
74 #include <linux/file.h>
75 #include <linux/sysfs.h>
76 #include <linux/mm.h>
77 #include <linux/io.h>
78 #include <linux/vmalloc.h>
79 #include <linux/netdevice.h>
80 #include <linux/timer.h>
81 #include <linux/interrupt.h>
82 #include <linux/uaccess.h>
83 #include <linux/list.h>
84 #include <linux/kthread.h>
85 #include <linux/kernel.h>
86 #include <linux/compat.h>
87 #include <linux/poll.h>
88 #include <linux/smp.h>
89 #include <linux/wait_bit.h>
90 
91 #if defined(__i386__) || defined(__amd64__)
92 #include <asm/smp.h>
93 #endif
94 
95 SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
96     "LinuxKPI parameters");
97 
98 int linuxkpi_debug;
99 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN,
100     &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable.");
101 
102 MALLOC_DEFINE(M_KMALLOC, "linux", "Linux kmalloc compat");
103 
104 #include <linux/rbtree.h>
105 /* Undo Linux compat changes. */
106 #undef RB_ROOT
107 #undef file
108 #undef cdev
109 #define	RB_ROOT(head)	(head)->rbh_root
110 
111 static void linux_cdev_deref(struct linux_cdev *ldev);
112 static struct vm_area_struct *linux_cdev_handle_find(void *handle);
113 
114 struct kobject linux_class_root;
115 struct device linux_root_device;
116 struct class linux_class_misc;
117 struct list_head pci_drivers;
118 struct list_head pci_devices;
119 spinlock_t pci_lock;
120 
121 unsigned long linux_timer_hz_mask;
122 
123 wait_queue_head_t linux_bit_waitq;
124 wait_queue_head_t linux_var_waitq;
125 
126 int
127 panic_cmp(struct rb_node *one, struct rb_node *two)
128 {
129 	panic("no cmp");
130 }
131 
132 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp);
133 
134 int
135 kobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list args)
136 {
137 	va_list tmp_va;
138 	int len;
139 	char *old;
140 	char *name;
141 	char dummy;
142 
143 	old = kobj->name;
144 
145 	if (old && fmt == NULL)
146 		return (0);
147 
148 	/* compute length of string */
149 	va_copy(tmp_va, args);
150 	len = vsnprintf(&dummy, 0, fmt, tmp_va);
151 	va_end(tmp_va);
152 
153 	/* account for zero termination */
154 	len++;
155 
156 	/* check for error */
157 	if (len < 1)
158 		return (-EINVAL);
159 
160 	/* allocate memory for string */
161 	name = kzalloc(len, GFP_KERNEL);
162 	if (name == NULL)
163 		return (-ENOMEM);
164 	vsnprintf(name, len, fmt, args);
165 	kobj->name = name;
166 
167 	/* free old string */
168 	kfree(old);
169 
170 	/* filter new string */
171 	for (; *name != '\0'; name++)
172 		if (*name == '/')
173 			*name = '!';
174 	return (0);
175 }
176 
177 int
178 kobject_set_name(struct kobject *kobj, const char *fmt, ...)
179 {
180 	va_list args;
181 	int error;
182 
183 	va_start(args, fmt);
184 	error = kobject_set_name_vargs(kobj, fmt, args);
185 	va_end(args);
186 
187 	return (error);
188 }
189 
190 static int
191 kobject_add_complete(struct kobject *kobj, struct kobject *parent)
192 {
193 	const struct kobj_type *t;
194 	int error;
195 
196 	kobj->parent = parent;
197 	error = sysfs_create_dir(kobj);
198 	if (error == 0 && kobj->ktype && kobj->ktype->default_attrs) {
199 		struct attribute **attr;
200 		t = kobj->ktype;
201 
202 		for (attr = t->default_attrs; *attr != NULL; attr++) {
203 			error = sysfs_create_file(kobj, *attr);
204 			if (error)
205 				break;
206 		}
207 		if (error)
208 			sysfs_remove_dir(kobj);
209 
210 	}
211 	return (error);
212 }
213 
214 int
215 kobject_add(struct kobject *kobj, struct kobject *parent, const char *fmt, ...)
216 {
217 	va_list args;
218 	int error;
219 
220 	va_start(args, fmt);
221 	error = kobject_set_name_vargs(kobj, fmt, args);
222 	va_end(args);
223 	if (error)
224 		return (error);
225 
226 	return kobject_add_complete(kobj, parent);
227 }
228 
229 void
230 linux_kobject_release(struct kref *kref)
231 {
232 	struct kobject *kobj;
233 	char *name;
234 
235 	kobj = container_of(kref, struct kobject, kref);
236 	sysfs_remove_dir(kobj);
237 	name = kobj->name;
238 	if (kobj->ktype && kobj->ktype->release)
239 		kobj->ktype->release(kobj);
240 	kfree(name);
241 }
242 
243 static void
244 linux_kobject_kfree(struct kobject *kobj)
245 {
246 	kfree(kobj);
247 }
248 
249 static void
250 linux_kobject_kfree_name(struct kobject *kobj)
251 {
252 	if (kobj) {
253 		kfree(kobj->name);
254 	}
255 }
256 
257 const struct kobj_type linux_kfree_type = {
258 	.release = linux_kobject_kfree
259 };
260 
261 static void
262 linux_device_release(struct device *dev)
263 {
264 	pr_debug("linux_device_release: %s\n", dev_name(dev));
265 	kfree(dev);
266 }
267 
268 static ssize_t
269 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf)
270 {
271 	struct class_attribute *dattr;
272 	ssize_t error;
273 
274 	dattr = container_of(attr, struct class_attribute, attr);
275 	error = -EIO;
276 	if (dattr->show)
277 		error = dattr->show(container_of(kobj, struct class, kobj),
278 		    dattr, buf);
279 	return (error);
280 }
281 
282 static ssize_t
283 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf,
284     size_t count)
285 {
286 	struct class_attribute *dattr;
287 	ssize_t error;
288 
289 	dattr = container_of(attr, struct class_attribute, attr);
290 	error = -EIO;
291 	if (dattr->store)
292 		error = dattr->store(container_of(kobj, struct class, kobj),
293 		    dattr, buf, count);
294 	return (error);
295 }
296 
297 static void
298 linux_class_release(struct kobject *kobj)
299 {
300 	struct class *class;
301 
302 	class = container_of(kobj, struct class, kobj);
303 	if (class->class_release)
304 		class->class_release(class);
305 }
306 
307 static const struct sysfs_ops linux_class_sysfs = {
308 	.show  = linux_class_show,
309 	.store = linux_class_store,
310 };
311 
312 const struct kobj_type linux_class_ktype = {
313 	.release = linux_class_release,
314 	.sysfs_ops = &linux_class_sysfs
315 };
316 
317 static void
318 linux_dev_release(struct kobject *kobj)
319 {
320 	struct device *dev;
321 
322 	dev = container_of(kobj, struct device, kobj);
323 	/* This is the precedence defined by linux. */
324 	if (dev->release)
325 		dev->release(dev);
326 	else if (dev->class && dev->class->dev_release)
327 		dev->class->dev_release(dev);
328 }
329 
330 static ssize_t
331 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf)
332 {
333 	struct device_attribute *dattr;
334 	ssize_t error;
335 
336 	dattr = container_of(attr, struct device_attribute, attr);
337 	error = -EIO;
338 	if (dattr->show)
339 		error = dattr->show(container_of(kobj, struct device, kobj),
340 		    dattr, buf);
341 	return (error);
342 }
343 
344 static ssize_t
345 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf,
346     size_t count)
347 {
348 	struct device_attribute *dattr;
349 	ssize_t error;
350 
351 	dattr = container_of(attr, struct device_attribute, attr);
352 	error = -EIO;
353 	if (dattr->store)
354 		error = dattr->store(container_of(kobj, struct device, kobj),
355 		    dattr, buf, count);
356 	return (error);
357 }
358 
359 static const struct sysfs_ops linux_dev_sysfs = {
360 	.show  = linux_dev_show,
361 	.store = linux_dev_store,
362 };
363 
364 const struct kobj_type linux_dev_ktype = {
365 	.release = linux_dev_release,
366 	.sysfs_ops = &linux_dev_sysfs
367 };
368 
369 struct device *
370 device_create(struct class *class, struct device *parent, dev_t devt,
371     void *drvdata, const char *fmt, ...)
372 {
373 	struct device *dev;
374 	va_list args;
375 
376 	dev = kzalloc(sizeof(*dev), M_WAITOK);
377 	dev->parent = parent;
378 	dev->class = class;
379 	dev->devt = devt;
380 	dev->driver_data = drvdata;
381 	dev->release = linux_device_release;
382 	va_start(args, fmt);
383 	kobject_set_name_vargs(&dev->kobj, fmt, args);
384 	va_end(args);
385 	device_register(dev);
386 
387 	return (dev);
388 }
389 
390 int
391 kobject_init_and_add(struct kobject *kobj, const struct kobj_type *ktype,
392     struct kobject *parent, const char *fmt, ...)
393 {
394 	va_list args;
395 	int error;
396 
397 	kobject_init(kobj, ktype);
398 	kobj->ktype = ktype;
399 	kobj->parent = parent;
400 	kobj->name = NULL;
401 
402 	va_start(args, fmt);
403 	error = kobject_set_name_vargs(kobj, fmt, args);
404 	va_end(args);
405 	if (error)
406 		return (error);
407 	return kobject_add_complete(kobj, parent);
408 }
409 
410 static void
411 linux_kq_lock(void *arg)
412 {
413 	spinlock_t *s = arg;
414 
415 	spin_lock(s);
416 }
417 static void
418 linux_kq_unlock(void *arg)
419 {
420 	spinlock_t *s = arg;
421 
422 	spin_unlock(s);
423 }
424 
425 static void
426 linux_kq_lock_owned(void *arg)
427 {
428 #ifdef INVARIANTS
429 	spinlock_t *s = arg;
430 
431 	mtx_assert(&s->m, MA_OWNED);
432 #endif
433 }
434 
435 static void
436 linux_kq_lock_unowned(void *arg)
437 {
438 #ifdef INVARIANTS
439 	spinlock_t *s = arg;
440 
441 	mtx_assert(&s->m, MA_NOTOWNED);
442 #endif
443 }
444 
445 static void
446 linux_file_kqfilter_poll(struct linux_file *, int);
447 
448 struct linux_file *
449 linux_file_alloc(void)
450 {
451 	struct linux_file *filp;
452 
453 	filp = kzalloc(sizeof(*filp), GFP_KERNEL);
454 
455 	/* set initial refcount */
456 	filp->f_count = 1;
457 
458 	/* setup fields needed by kqueue support */
459 	spin_lock_init(&filp->f_kqlock);
460 	knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock,
461 	    linux_kq_lock, linux_kq_unlock,
462 	    linux_kq_lock_owned, linux_kq_lock_unowned);
463 
464 	return (filp);
465 }
466 
467 void
468 linux_file_free(struct linux_file *filp)
469 {
470 	if (filp->_file == NULL) {
471 		if (filp->f_shmem != NULL)
472 			vm_object_deallocate(filp->f_shmem);
473 		kfree(filp);
474 	} else {
475 		/*
476 		 * The close method of the character device or file
477 		 * will free the linux_file structure:
478 		 */
479 		_fdrop(filp->_file, curthread);
480 	}
481 }
482 
483 static int
484 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
485     vm_page_t *mres)
486 {
487 	struct vm_area_struct *vmap;
488 
489 	vmap = linux_cdev_handle_find(vm_obj->handle);
490 
491 	MPASS(vmap != NULL);
492 	MPASS(vmap->vm_private_data == vm_obj->handle);
493 
494 	if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) {
495 		vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset;
496 		vm_page_t page;
497 
498 		if (((*mres)->flags & PG_FICTITIOUS) != 0) {
499 			/*
500 			 * If the passed in result page is a fake
501 			 * page, update it with the new physical
502 			 * address.
503 			 */
504 			page = *mres;
505 			vm_page_updatefake(page, paddr, vm_obj->memattr);
506 		} else {
507 			/*
508 			 * Replace the passed in "mres" page with our
509 			 * own fake page and free up the all of the
510 			 * original pages.
511 			 */
512 			VM_OBJECT_WUNLOCK(vm_obj);
513 			page = vm_page_getfake(paddr, vm_obj->memattr);
514 			VM_OBJECT_WLOCK(vm_obj);
515 
516 			vm_page_replace(page, vm_obj, (*mres)->pindex, *mres);
517 			*mres = page;
518 		}
519 		vm_page_valid(page);
520 		return (VM_PAGER_OK);
521 	}
522 	return (VM_PAGER_FAIL);
523 }
524 
525 static int
526 linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type,
527     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
528 {
529 	struct vm_area_struct *vmap;
530 	int err;
531 
532 	/* get VM area structure */
533 	vmap = linux_cdev_handle_find(vm_obj->handle);
534 	MPASS(vmap != NULL);
535 	MPASS(vmap->vm_private_data == vm_obj->handle);
536 
537 	VM_OBJECT_WUNLOCK(vm_obj);
538 
539 	linux_set_current(curthread);
540 
541 	down_write(&vmap->vm_mm->mmap_sem);
542 	if (unlikely(vmap->vm_ops == NULL)) {
543 		err = VM_FAULT_SIGBUS;
544 	} else {
545 		struct vm_fault vmf;
546 
547 		/* fill out VM fault structure */
548 		vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx);
549 		vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
550 		vmf.pgoff = 0;
551 		vmf.page = NULL;
552 		vmf.vma = vmap;
553 
554 		vmap->vm_pfn_count = 0;
555 		vmap->vm_pfn_pcount = &vmap->vm_pfn_count;
556 		vmap->vm_obj = vm_obj;
557 
558 		err = vmap->vm_ops->fault(vmap, &vmf);
559 
560 		while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) {
561 			kern_yield(PRI_USER);
562 			err = vmap->vm_ops->fault(vmap, &vmf);
563 		}
564 	}
565 
566 	/* translate return code */
567 	switch (err) {
568 	case VM_FAULT_OOM:
569 		err = VM_PAGER_AGAIN;
570 		break;
571 	case VM_FAULT_SIGBUS:
572 		err = VM_PAGER_BAD;
573 		break;
574 	case VM_FAULT_NOPAGE:
575 		/*
576 		 * By contract the fault handler will return having
577 		 * busied all the pages itself. If pidx is already
578 		 * found in the object, it will simply xbusy the first
579 		 * page and return with vm_pfn_count set to 1.
580 		 */
581 		*first = vmap->vm_pfn_first;
582 		*last = *first + vmap->vm_pfn_count - 1;
583 		err = VM_PAGER_OK;
584 		break;
585 	default:
586 		err = VM_PAGER_ERROR;
587 		break;
588 	}
589 	up_write(&vmap->vm_mm->mmap_sem);
590 	VM_OBJECT_WLOCK(vm_obj);
591 	return (err);
592 }
593 
594 static struct rwlock linux_vma_lock;
595 static TAILQ_HEAD(, vm_area_struct) linux_vma_head =
596     TAILQ_HEAD_INITIALIZER(linux_vma_head);
597 
598 static void
599 linux_cdev_handle_free(struct vm_area_struct *vmap)
600 {
601 	/* Drop reference on vm_file */
602 	if (vmap->vm_file != NULL)
603 		fput(vmap->vm_file);
604 
605 	/* Drop reference on mm_struct */
606 	mmput(vmap->vm_mm);
607 
608 	kfree(vmap);
609 }
610 
611 static void
612 linux_cdev_handle_remove(struct vm_area_struct *vmap)
613 {
614 	rw_wlock(&linux_vma_lock);
615 	TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry);
616 	rw_wunlock(&linux_vma_lock);
617 }
618 
619 static struct vm_area_struct *
620 linux_cdev_handle_find(void *handle)
621 {
622 	struct vm_area_struct *vmap;
623 
624 	rw_rlock(&linux_vma_lock);
625 	TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) {
626 		if (vmap->vm_private_data == handle)
627 			break;
628 	}
629 	rw_runlock(&linux_vma_lock);
630 	return (vmap);
631 }
632 
633 static int
634 linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
635 		      vm_ooffset_t foff, struct ucred *cred, u_short *color)
636 {
637 
638 	MPASS(linux_cdev_handle_find(handle) != NULL);
639 	*color = 0;
640 	return (0);
641 }
642 
643 static void
644 linux_cdev_pager_dtor(void *handle)
645 {
646 	const struct vm_operations_struct *vm_ops;
647 	struct vm_area_struct *vmap;
648 
649 	vmap = linux_cdev_handle_find(handle);
650 	MPASS(vmap != NULL);
651 
652 	/*
653 	 * Remove handle before calling close operation to prevent
654 	 * other threads from reusing the handle pointer.
655 	 */
656 	linux_cdev_handle_remove(vmap);
657 
658 	down_write(&vmap->vm_mm->mmap_sem);
659 	vm_ops = vmap->vm_ops;
660 	if (likely(vm_ops != NULL))
661 		vm_ops->close(vmap);
662 	up_write(&vmap->vm_mm->mmap_sem);
663 
664 	linux_cdev_handle_free(vmap);
665 }
666 
667 static struct cdev_pager_ops linux_cdev_pager_ops[2] = {
668   {
669 	/* OBJT_MGTDEVICE */
670 	.cdev_pg_populate	= linux_cdev_pager_populate,
671 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
672 	.cdev_pg_dtor	= linux_cdev_pager_dtor
673   },
674   {
675 	/* OBJT_DEVICE */
676 	.cdev_pg_fault	= linux_cdev_pager_fault,
677 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
678 	.cdev_pg_dtor	= linux_cdev_pager_dtor
679   },
680 };
681 
682 int
683 zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
684     unsigned long size)
685 {
686 	vm_object_t obj;
687 	vm_page_t m;
688 
689 	obj = vma->vm_obj;
690 	if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0)
691 		return (-ENOTSUP);
692 	VM_OBJECT_RLOCK(obj);
693 	for (m = vm_page_find_least(obj, OFF_TO_IDX(address));
694 	    m != NULL && m->pindex < OFF_TO_IDX(address + size);
695 	    m = TAILQ_NEXT(m, listq))
696 		pmap_remove_all(m);
697 	VM_OBJECT_RUNLOCK(obj);
698 	return (0);
699 }
700 
701 static struct file_operations dummy_ldev_ops = {
702 	/* XXXKIB */
703 };
704 
705 static struct linux_cdev dummy_ldev = {
706 	.ops = &dummy_ldev_ops,
707 };
708 
709 #define	LDEV_SI_DTR	0x0001
710 #define	LDEV_SI_REF	0x0002
711 
712 static void
713 linux_get_fop(struct linux_file *filp, const struct file_operations **fop,
714     struct linux_cdev **dev)
715 {
716 	struct linux_cdev *ldev;
717 	u_int siref;
718 
719 	ldev = filp->f_cdev;
720 	*fop = filp->f_op;
721 	if (ldev != NULL) {
722 		for (siref = ldev->siref;;) {
723 			if ((siref & LDEV_SI_DTR) != 0) {
724 				ldev = &dummy_ldev;
725 				siref = ldev->siref;
726 				*fop = ldev->ops;
727 				MPASS((ldev->siref & LDEV_SI_DTR) == 0);
728 			} else if (atomic_fcmpset_int(&ldev->siref, &siref,
729 			    siref + LDEV_SI_REF)) {
730 				break;
731 			}
732 		}
733 	}
734 	*dev = ldev;
735 }
736 
737 static void
738 linux_drop_fop(struct linux_cdev *ldev)
739 {
740 
741 	if (ldev == NULL)
742 		return;
743 	MPASS((ldev->siref & ~LDEV_SI_DTR) != 0);
744 	atomic_subtract_int(&ldev->siref, LDEV_SI_REF);
745 }
746 
747 #define	OPW(fp,td,code) ({			\
748 	struct file *__fpop;			\
749 	__typeof(code) __retval;		\
750 						\
751 	__fpop = (td)->td_fpop;			\
752 	(td)->td_fpop = (fp);			\
753 	__retval = (code);			\
754 	(td)->td_fpop = __fpop;			\
755 	__retval;				\
756 })
757 
758 static int
759 linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td,
760     struct file *file)
761 {
762 	struct linux_cdev *ldev;
763 	struct linux_file *filp;
764 	const struct file_operations *fop;
765 	int error;
766 
767 	ldev = dev->si_drv1;
768 
769 	filp = linux_file_alloc();
770 	filp->f_dentry = &filp->f_dentry_store;
771 	filp->f_op = ldev->ops;
772 	filp->f_mode = file->f_flag;
773 	filp->f_flags = file->f_flag;
774 	filp->f_vnode = file->f_vnode;
775 	filp->_file = file;
776 	refcount_acquire(&ldev->refs);
777 	filp->f_cdev = ldev;
778 
779 	linux_set_current(td);
780 	linux_get_fop(filp, &fop, &ldev);
781 
782 	if (fop->open != NULL) {
783 		error = -fop->open(file->f_vnode, filp);
784 		if (error != 0) {
785 			linux_drop_fop(ldev);
786 			linux_cdev_deref(filp->f_cdev);
787 			kfree(filp);
788 			return (error);
789 		}
790 	}
791 
792 	/* hold on to the vnode - used for fstat() */
793 	vhold(filp->f_vnode);
794 
795 	/* release the file from devfs */
796 	finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops);
797 	linux_drop_fop(ldev);
798 	return (ENXIO);
799 }
800 
801 #define	LINUX_IOCTL_MIN_PTR 0x10000UL
802 #define	LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX)
803 
804 static inline int
805 linux_remap_address(void **uaddr, size_t len)
806 {
807 	uintptr_t uaddr_val = (uintptr_t)(*uaddr);
808 
809 	if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR &&
810 	    uaddr_val < LINUX_IOCTL_MAX_PTR)) {
811 		struct task_struct *pts = current;
812 		if (pts == NULL) {
813 			*uaddr = NULL;
814 			return (1);
815 		}
816 
817 		/* compute data offset */
818 		uaddr_val -= LINUX_IOCTL_MIN_PTR;
819 
820 		/* check that length is within bounds */
821 		if ((len > IOCPARM_MAX) ||
822 		    (uaddr_val + len) > pts->bsd_ioctl_len) {
823 			*uaddr = NULL;
824 			return (1);
825 		}
826 
827 		/* re-add kernel buffer address */
828 		uaddr_val += (uintptr_t)pts->bsd_ioctl_data;
829 
830 		/* update address location */
831 		*uaddr = (void *)uaddr_val;
832 		return (1);
833 	}
834 	return (0);
835 }
836 
837 int
838 linux_copyin(const void *uaddr, void *kaddr, size_t len)
839 {
840 	if (linux_remap_address(__DECONST(void **, &uaddr), len)) {
841 		if (uaddr == NULL)
842 			return (-EFAULT);
843 		memcpy(kaddr, uaddr, len);
844 		return (0);
845 	}
846 	return (-copyin(uaddr, kaddr, len));
847 }
848 
849 int
850 linux_copyout(const void *kaddr, void *uaddr, size_t len)
851 {
852 	if (linux_remap_address(&uaddr, len)) {
853 		if (uaddr == NULL)
854 			return (-EFAULT);
855 		memcpy(uaddr, kaddr, len);
856 		return (0);
857 	}
858 	return (-copyout(kaddr, uaddr, len));
859 }
860 
861 size_t
862 linux_clear_user(void *_uaddr, size_t _len)
863 {
864 	uint8_t *uaddr = _uaddr;
865 	size_t len = _len;
866 
867 	/* make sure uaddr is aligned before going into the fast loop */
868 	while (((uintptr_t)uaddr & 7) != 0 && len > 7) {
869 		if (subyte(uaddr, 0))
870 			return (_len);
871 		uaddr++;
872 		len--;
873 	}
874 
875 	/* zero 8 bytes at a time */
876 	while (len > 7) {
877 #ifdef __LP64__
878 		if (suword64(uaddr, 0))
879 			return (_len);
880 #else
881 		if (suword32(uaddr, 0))
882 			return (_len);
883 		if (suword32(uaddr + 4, 0))
884 			return (_len);
885 #endif
886 		uaddr += 8;
887 		len -= 8;
888 	}
889 
890 	/* zero fill end, if any */
891 	while (len > 0) {
892 		if (subyte(uaddr, 0))
893 			return (_len);
894 		uaddr++;
895 		len--;
896 	}
897 	return (0);
898 }
899 
900 int
901 linux_access_ok(const void *uaddr, size_t len)
902 {
903 	uintptr_t saddr;
904 	uintptr_t eaddr;
905 
906 	/* get start and end address */
907 	saddr = (uintptr_t)uaddr;
908 	eaddr = (uintptr_t)uaddr + len;
909 
910 	/* verify addresses are valid for userspace */
911 	return ((saddr == eaddr) ||
912 	    (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS));
913 }
914 
915 /*
916  * This function should return either EINTR or ERESTART depending on
917  * the signal type sent to this thread:
918  */
919 static int
920 linux_get_error(struct task_struct *task, int error)
921 {
922 	/* check for signal type interrupt code */
923 	if (error == EINTR || error == ERESTARTSYS || error == ERESTART) {
924 		error = -linux_schedule_get_interrupt_value(task);
925 		if (error == 0)
926 			error = EINTR;
927 	}
928 	return (error);
929 }
930 
931 static int
932 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
933     const struct file_operations *fop, u_long cmd, caddr_t data,
934     struct thread *td)
935 {
936 	struct task_struct *task = current;
937 	unsigned size;
938 	int error;
939 
940 	size = IOCPARM_LEN(cmd);
941 	/* refer to logic in sys_ioctl() */
942 	if (size > 0) {
943 		/*
944 		 * Setup hint for linux_copyin() and linux_copyout().
945 		 *
946 		 * Background: Linux code expects a user-space address
947 		 * while FreeBSD supplies a kernel-space address.
948 		 */
949 		task->bsd_ioctl_data = data;
950 		task->bsd_ioctl_len = size;
951 		data = (void *)LINUX_IOCTL_MIN_PTR;
952 	} else {
953 		/* fetch user-space pointer */
954 		data = *(void **)data;
955 	}
956 #if defined(__amd64__)
957 	if (td->td_proc->p_elf_machine == EM_386) {
958 		/* try the compat IOCTL handler first */
959 		if (fop->compat_ioctl != NULL) {
960 			error = -OPW(fp, td, fop->compat_ioctl(filp,
961 			    cmd, (u_long)data));
962 		} else {
963 			error = ENOTTY;
964 		}
965 
966 		/* fallback to the regular IOCTL handler, if any */
967 		if (error == ENOTTY && fop->unlocked_ioctl != NULL) {
968 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
969 			    cmd, (u_long)data));
970 		}
971 	} else
972 #endif
973 	{
974 		if (fop->unlocked_ioctl != NULL) {
975 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
976 			    cmd, (u_long)data));
977 		} else {
978 			error = ENOTTY;
979 		}
980 	}
981 	if (size > 0) {
982 		task->bsd_ioctl_data = NULL;
983 		task->bsd_ioctl_len = 0;
984 	}
985 
986 	if (error == EWOULDBLOCK) {
987 		/* update kqfilter status, if any */
988 		linux_file_kqfilter_poll(filp,
989 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
990 	} else {
991 		error = linux_get_error(task, error);
992 	}
993 	return (error);
994 }
995 
996 #define	LINUX_POLL_TABLE_NORMAL ((poll_table *)1)
997 
998 /*
999  * This function atomically updates the poll wakeup state and returns
1000  * the previous state at the time of update.
1001  */
1002 static uint8_t
1003 linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate)
1004 {
1005 	int c, old;
1006 
1007 	c = v->counter;
1008 
1009 	while ((old = atomic_cmpxchg(v, c, pstate[c])) != c)
1010 		c = old;
1011 
1012 	return (c);
1013 }
1014 
1015 
1016 static int
1017 linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key)
1018 {
1019 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1020 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */
1021 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1022 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY,
1023 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */
1024 	};
1025 	struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq);
1026 
1027 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1028 	case LINUX_FWQ_STATE_QUEUED:
1029 		linux_poll_wakeup(filp);
1030 		return (1);
1031 	default:
1032 		return (0);
1033 	}
1034 }
1035 
1036 void
1037 linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p)
1038 {
1039 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1040 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY,
1041 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1042 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */
1043 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED,
1044 	};
1045 
1046 	/* check if we are called inside the select system call */
1047 	if (p == LINUX_POLL_TABLE_NORMAL)
1048 		selrecord(curthread, &filp->f_selinfo);
1049 
1050 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1051 	case LINUX_FWQ_STATE_INIT:
1052 		/* NOTE: file handles can only belong to one wait-queue */
1053 		filp->f_wait_queue.wqh = wqh;
1054 		filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback;
1055 		add_wait_queue(wqh, &filp->f_wait_queue.wq);
1056 		atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED);
1057 		break;
1058 	default:
1059 		break;
1060 	}
1061 }
1062 
1063 static void
1064 linux_poll_wait_dequeue(struct linux_file *filp)
1065 {
1066 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1067 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT,	/* NOP */
1068 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT,
1069 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT,
1070 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT,
1071 	};
1072 
1073 	seldrain(&filp->f_selinfo);
1074 
1075 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1076 	case LINUX_FWQ_STATE_NOT_READY:
1077 	case LINUX_FWQ_STATE_QUEUED:
1078 	case LINUX_FWQ_STATE_READY:
1079 		remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq);
1080 		break;
1081 	default:
1082 		break;
1083 	}
1084 }
1085 
1086 void
1087 linux_poll_wakeup(struct linux_file *filp)
1088 {
1089 	/* this function should be NULL-safe */
1090 	if (filp == NULL)
1091 		return;
1092 
1093 	selwakeup(&filp->f_selinfo);
1094 
1095 	spin_lock(&filp->f_kqlock);
1096 	filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ |
1097 	    LINUX_KQ_FLAG_NEED_WRITE;
1098 
1099 	/* make sure the "knote" gets woken up */
1100 	KNOTE_LOCKED(&filp->f_selinfo.si_note, 1);
1101 	spin_unlock(&filp->f_kqlock);
1102 }
1103 
1104 static void
1105 linux_file_kqfilter_detach(struct knote *kn)
1106 {
1107 	struct linux_file *filp = kn->kn_hook;
1108 
1109 	spin_lock(&filp->f_kqlock);
1110 	knlist_remove(&filp->f_selinfo.si_note, kn, 1);
1111 	spin_unlock(&filp->f_kqlock);
1112 }
1113 
1114 static int
1115 linux_file_kqfilter_read_event(struct knote *kn, long hint)
1116 {
1117 	struct linux_file *filp = kn->kn_hook;
1118 
1119 	mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1120 
1121 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0);
1122 }
1123 
1124 static int
1125 linux_file_kqfilter_write_event(struct knote *kn, long hint)
1126 {
1127 	struct linux_file *filp = kn->kn_hook;
1128 
1129 	mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1130 
1131 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0);
1132 }
1133 
1134 static struct filterops linux_dev_kqfiltops_read = {
1135 	.f_isfd = 1,
1136 	.f_detach = linux_file_kqfilter_detach,
1137 	.f_event = linux_file_kqfilter_read_event,
1138 };
1139 
1140 static struct filterops linux_dev_kqfiltops_write = {
1141 	.f_isfd = 1,
1142 	.f_detach = linux_file_kqfilter_detach,
1143 	.f_event = linux_file_kqfilter_write_event,
1144 };
1145 
1146 static void
1147 linux_file_kqfilter_poll(struct linux_file *filp, int kqflags)
1148 {
1149 	struct thread *td;
1150 	const struct file_operations *fop;
1151 	struct linux_cdev *ldev;
1152 	int temp;
1153 
1154 	if ((filp->f_kqflags & kqflags) == 0)
1155 		return;
1156 
1157 	td = curthread;
1158 
1159 	linux_get_fop(filp, &fop, &ldev);
1160 	/* get the latest polling state */
1161 	temp = OPW(filp->_file, td, fop->poll(filp, NULL));
1162 	linux_drop_fop(ldev);
1163 
1164 	spin_lock(&filp->f_kqlock);
1165 	/* clear kqflags */
1166 	filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ |
1167 	    LINUX_KQ_FLAG_NEED_WRITE);
1168 	/* update kqflags */
1169 	if ((temp & (POLLIN | POLLOUT)) != 0) {
1170 		if ((temp & POLLIN) != 0)
1171 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ;
1172 		if ((temp & POLLOUT) != 0)
1173 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE;
1174 
1175 		/* make sure the "knote" gets woken up */
1176 		KNOTE_LOCKED(&filp->f_selinfo.si_note, 0);
1177 	}
1178 	spin_unlock(&filp->f_kqlock);
1179 }
1180 
1181 static int
1182 linux_file_kqfilter(struct file *file, struct knote *kn)
1183 {
1184 	struct linux_file *filp;
1185 	struct thread *td;
1186 	int error;
1187 
1188 	td = curthread;
1189 	filp = (struct linux_file *)file->f_data;
1190 	filp->f_flags = file->f_flag;
1191 	if (filp->f_op->poll == NULL)
1192 		return (EINVAL);
1193 
1194 	spin_lock(&filp->f_kqlock);
1195 	switch (kn->kn_filter) {
1196 	case EVFILT_READ:
1197 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ;
1198 		kn->kn_fop = &linux_dev_kqfiltops_read;
1199 		kn->kn_hook = filp;
1200 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1201 		error = 0;
1202 		break;
1203 	case EVFILT_WRITE:
1204 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE;
1205 		kn->kn_fop = &linux_dev_kqfiltops_write;
1206 		kn->kn_hook = filp;
1207 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1208 		error = 0;
1209 		break;
1210 	default:
1211 		error = EINVAL;
1212 		break;
1213 	}
1214 	spin_unlock(&filp->f_kqlock);
1215 
1216 	if (error == 0) {
1217 		linux_set_current(td);
1218 
1219 		/* update kqfilter status, if any */
1220 		linux_file_kqfilter_poll(filp,
1221 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1222 	}
1223 	return (error);
1224 }
1225 
1226 static int
1227 linux_file_mmap_single(struct file *fp, const struct file_operations *fop,
1228     vm_ooffset_t *offset, vm_size_t size, struct vm_object **object,
1229     int nprot, struct thread *td)
1230 {
1231 	struct task_struct *task;
1232 	struct vm_area_struct *vmap;
1233 	struct mm_struct *mm;
1234 	struct linux_file *filp;
1235 	vm_memattr_t attr;
1236 	int error;
1237 
1238 	filp = (struct linux_file *)fp->f_data;
1239 	filp->f_flags = fp->f_flag;
1240 
1241 	if (fop->mmap == NULL)
1242 		return (EOPNOTSUPP);
1243 
1244 	linux_set_current(td);
1245 
1246 	/*
1247 	 * The same VM object might be shared by multiple processes
1248 	 * and the mm_struct is usually freed when a process exits.
1249 	 *
1250 	 * The atomic reference below makes sure the mm_struct is
1251 	 * available as long as the vmap is in the linux_vma_head.
1252 	 */
1253 	task = current;
1254 	mm = task->mm;
1255 	if (atomic_inc_not_zero(&mm->mm_users) == 0)
1256 		return (EINVAL);
1257 
1258 	vmap = kzalloc(sizeof(*vmap), GFP_KERNEL);
1259 	vmap->vm_start = 0;
1260 	vmap->vm_end = size;
1261 	vmap->vm_pgoff = *offset / PAGE_SIZE;
1262 	vmap->vm_pfn = 0;
1263 	vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL);
1264 	vmap->vm_ops = NULL;
1265 	vmap->vm_file = get_file(filp);
1266 	vmap->vm_mm = mm;
1267 
1268 	if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) {
1269 		error = linux_get_error(task, EINTR);
1270 	} else {
1271 		error = -OPW(fp, td, fop->mmap(filp, vmap));
1272 		error = linux_get_error(task, error);
1273 		up_write(&vmap->vm_mm->mmap_sem);
1274 	}
1275 
1276 	if (error != 0) {
1277 		linux_cdev_handle_free(vmap);
1278 		return (error);
1279 	}
1280 
1281 	attr = pgprot2cachemode(vmap->vm_page_prot);
1282 
1283 	if (vmap->vm_ops != NULL) {
1284 		struct vm_area_struct *ptr;
1285 		void *vm_private_data;
1286 		bool vm_no_fault;
1287 
1288 		if (vmap->vm_ops->open == NULL ||
1289 		    vmap->vm_ops->close == NULL ||
1290 		    vmap->vm_private_data == NULL) {
1291 			/* free allocated VM area struct */
1292 			linux_cdev_handle_free(vmap);
1293 			return (EINVAL);
1294 		}
1295 
1296 		vm_private_data = vmap->vm_private_data;
1297 
1298 		rw_wlock(&linux_vma_lock);
1299 		TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) {
1300 			if (ptr->vm_private_data == vm_private_data)
1301 				break;
1302 		}
1303 		/* check if there is an existing VM area struct */
1304 		if (ptr != NULL) {
1305 			/* check if the VM area structure is invalid */
1306 			if (ptr->vm_ops == NULL ||
1307 			    ptr->vm_ops->open == NULL ||
1308 			    ptr->vm_ops->close == NULL) {
1309 				error = ESTALE;
1310 				vm_no_fault = 1;
1311 			} else {
1312 				error = EEXIST;
1313 				vm_no_fault = (ptr->vm_ops->fault == NULL);
1314 			}
1315 		} else {
1316 			/* insert VM area structure into list */
1317 			TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry);
1318 			error = 0;
1319 			vm_no_fault = (vmap->vm_ops->fault == NULL);
1320 		}
1321 		rw_wunlock(&linux_vma_lock);
1322 
1323 		if (error != 0) {
1324 			/* free allocated VM area struct */
1325 			linux_cdev_handle_free(vmap);
1326 			/* check for stale VM area struct */
1327 			if (error != EEXIST)
1328 				return (error);
1329 		}
1330 
1331 		/* check if there is no fault handler */
1332 		if (vm_no_fault) {
1333 			*object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE,
1334 			    &linux_cdev_pager_ops[1], size, nprot, *offset,
1335 			    td->td_ucred);
1336 		} else {
1337 			*object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE,
1338 			    &linux_cdev_pager_ops[0], size, nprot, *offset,
1339 			    td->td_ucred);
1340 		}
1341 
1342 		/* check if allocating the VM object failed */
1343 		if (*object == NULL) {
1344 			if (error == 0) {
1345 				/* remove VM area struct from list */
1346 				linux_cdev_handle_remove(vmap);
1347 				/* free allocated VM area struct */
1348 				linux_cdev_handle_free(vmap);
1349 			}
1350 			return (EINVAL);
1351 		}
1352 	} else {
1353 		struct sglist *sg;
1354 
1355 		sg = sglist_alloc(1, M_WAITOK);
1356 		sglist_append_phys(sg,
1357 		    (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len);
1358 
1359 		*object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len,
1360 		    nprot, 0, td->td_ucred);
1361 
1362 		linux_cdev_handle_free(vmap);
1363 
1364 		if (*object == NULL) {
1365 			sglist_free(sg);
1366 			return (EINVAL);
1367 		}
1368 	}
1369 
1370 	if (attr != VM_MEMATTR_DEFAULT) {
1371 		VM_OBJECT_WLOCK(*object);
1372 		vm_object_set_memattr(*object, attr);
1373 		VM_OBJECT_WUNLOCK(*object);
1374 	}
1375 	*offset = 0;
1376 	return (0);
1377 }
1378 
1379 struct cdevsw linuxcdevsw = {
1380 	.d_version = D_VERSION,
1381 	.d_fdopen = linux_dev_fdopen,
1382 	.d_name = "lkpidev",
1383 };
1384 
1385 static int
1386 linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
1387     int flags, struct thread *td)
1388 {
1389 	struct linux_file *filp;
1390 	const struct file_operations *fop;
1391 	struct linux_cdev *ldev;
1392 	ssize_t bytes;
1393 	int error;
1394 
1395 	error = 0;
1396 	filp = (struct linux_file *)file->f_data;
1397 	filp->f_flags = file->f_flag;
1398 	/* XXX no support for I/O vectors currently */
1399 	if (uio->uio_iovcnt != 1)
1400 		return (EOPNOTSUPP);
1401 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1402 		return (EINVAL);
1403 	linux_set_current(td);
1404 	linux_get_fop(filp, &fop, &ldev);
1405 	if (fop->read != NULL) {
1406 		bytes = OPW(file, td, fop->read(filp,
1407 		    uio->uio_iov->iov_base,
1408 		    uio->uio_iov->iov_len, &uio->uio_offset));
1409 		if (bytes >= 0) {
1410 			uio->uio_iov->iov_base =
1411 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1412 			uio->uio_iov->iov_len -= bytes;
1413 			uio->uio_resid -= bytes;
1414 		} else {
1415 			error = linux_get_error(current, -bytes);
1416 		}
1417 	} else
1418 		error = ENXIO;
1419 
1420 	/* update kqfilter status, if any */
1421 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ);
1422 	linux_drop_fop(ldev);
1423 
1424 	return (error);
1425 }
1426 
1427 static int
1428 linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred,
1429     int flags, struct thread *td)
1430 {
1431 	struct linux_file *filp;
1432 	const struct file_operations *fop;
1433 	struct linux_cdev *ldev;
1434 	ssize_t bytes;
1435 	int error;
1436 
1437 	filp = (struct linux_file *)file->f_data;
1438 	filp->f_flags = file->f_flag;
1439 	/* XXX no support for I/O vectors currently */
1440 	if (uio->uio_iovcnt != 1)
1441 		return (EOPNOTSUPP);
1442 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1443 		return (EINVAL);
1444 	linux_set_current(td);
1445 	linux_get_fop(filp, &fop, &ldev);
1446 	if (fop->write != NULL) {
1447 		bytes = OPW(file, td, fop->write(filp,
1448 		    uio->uio_iov->iov_base,
1449 		    uio->uio_iov->iov_len, &uio->uio_offset));
1450 		if (bytes >= 0) {
1451 			uio->uio_iov->iov_base =
1452 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1453 			uio->uio_iov->iov_len -= bytes;
1454 			uio->uio_resid -= bytes;
1455 			error = 0;
1456 		} else {
1457 			error = linux_get_error(current, -bytes);
1458 		}
1459 	} else
1460 		error = ENXIO;
1461 
1462 	/* update kqfilter status, if any */
1463 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE);
1464 
1465 	linux_drop_fop(ldev);
1466 
1467 	return (error);
1468 }
1469 
1470 static int
1471 linux_file_poll(struct file *file, int events, struct ucred *active_cred,
1472     struct thread *td)
1473 {
1474 	struct linux_file *filp;
1475 	const struct file_operations *fop;
1476 	struct linux_cdev *ldev;
1477 	int revents;
1478 
1479 	filp = (struct linux_file *)file->f_data;
1480 	filp->f_flags = file->f_flag;
1481 	linux_set_current(td);
1482 	linux_get_fop(filp, &fop, &ldev);
1483 	if (fop->poll != NULL) {
1484 		revents = OPW(file, td, fop->poll(filp,
1485 		    LINUX_POLL_TABLE_NORMAL)) & events;
1486 	} else {
1487 		revents = 0;
1488 	}
1489 	linux_drop_fop(ldev);
1490 	return (revents);
1491 }
1492 
1493 static int
1494 linux_file_close(struct file *file, struct thread *td)
1495 {
1496 	struct linux_file *filp;
1497 	int (*release)(struct inode *, struct linux_file *);
1498 	const struct file_operations *fop;
1499 	struct linux_cdev *ldev;
1500 	int error;
1501 
1502 	filp = (struct linux_file *)file->f_data;
1503 
1504 	KASSERT(file_count(filp) == 0,
1505 	    ("File refcount(%d) is not zero", file_count(filp)));
1506 
1507 	if (td == NULL)
1508 		td = curthread;
1509 
1510 	error = 0;
1511 	filp->f_flags = file->f_flag;
1512 	linux_set_current(td);
1513 	linux_poll_wait_dequeue(filp);
1514 	linux_get_fop(filp, &fop, &ldev);
1515 	/*
1516 	 * Always use the real release function, if any, to avoid
1517 	 * leaking device resources:
1518 	 */
1519 	release = filp->f_op->release;
1520 	if (release != NULL)
1521 		error = -OPW(file, td, release(filp->f_vnode, filp));
1522 	funsetown(&filp->f_sigio);
1523 	if (filp->f_vnode != NULL)
1524 		vdrop(filp->f_vnode);
1525 	linux_drop_fop(ldev);
1526 	if (filp->f_cdev != NULL)
1527 		linux_cdev_deref(filp->f_cdev);
1528 	kfree(filp);
1529 
1530 	return (error);
1531 }
1532 
1533 static int
1534 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
1535     struct thread *td)
1536 {
1537 	struct linux_file *filp;
1538 	const struct file_operations *fop;
1539 	struct linux_cdev *ldev;
1540 	struct fiodgname_arg *fgn;
1541 	const char *p;
1542 	int error, i;
1543 
1544 	error = 0;
1545 	filp = (struct linux_file *)fp->f_data;
1546 	filp->f_flags = fp->f_flag;
1547 	linux_get_fop(filp, &fop, &ldev);
1548 
1549 	linux_set_current(td);
1550 	switch (cmd) {
1551 	case FIONBIO:
1552 		break;
1553 	case FIOASYNC:
1554 		if (fop->fasync == NULL)
1555 			break;
1556 		error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC));
1557 		break;
1558 	case FIOSETOWN:
1559 		error = fsetown(*(int *)data, &filp->f_sigio);
1560 		if (error == 0) {
1561 			if (fop->fasync == NULL)
1562 				break;
1563 			error = -OPW(fp, td, fop->fasync(0, filp,
1564 			    fp->f_flag & FASYNC));
1565 		}
1566 		break;
1567 	case FIOGETOWN:
1568 		*(int *)data = fgetown(&filp->f_sigio);
1569 		break;
1570 	case FIODGNAME:
1571 #ifdef	COMPAT_FREEBSD32
1572 	case FIODGNAME_32:
1573 #endif
1574 		if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) {
1575 			error = ENXIO;
1576 			break;
1577 		}
1578 		fgn = data;
1579 		p = devtoname(filp->f_cdev->cdev);
1580 		i = strlen(p) + 1;
1581 		if (i > fgn->len) {
1582 			error = EINVAL;
1583 			break;
1584 		}
1585 		error = copyout(p, fiodgname_buf_get_ptr(fgn, cmd), i);
1586 		break;
1587 	default:
1588 		error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td);
1589 		break;
1590 	}
1591 	linux_drop_fop(ldev);
1592 	return (error);
1593 }
1594 
1595 static int
1596 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1597     vm_prot_t *maxprotp, int *flagsp, struct file *fp,
1598     vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp)
1599 {
1600 	/*
1601 	 * Character devices do not provide private mappings
1602 	 * of any kind:
1603 	 */
1604 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1605 	    (prot & VM_PROT_WRITE) != 0)
1606 		return (EACCES);
1607 	if ((*flagsp & (MAP_PRIVATE | MAP_COPY)) != 0)
1608 		return (EINVAL);
1609 
1610 	return (linux_file_mmap_single(fp, fop, foff, objsize, objp,
1611 	    (int)prot, td));
1612 }
1613 
1614 static int
1615 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1616     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1617     struct thread *td)
1618 {
1619 	struct linux_file *filp;
1620 	const struct file_operations *fop;
1621 	struct linux_cdev *ldev;
1622 	struct mount *mp;
1623 	struct vnode *vp;
1624 	vm_object_t object;
1625 	vm_prot_t maxprot;
1626 	int error;
1627 
1628 	filp = (struct linux_file *)fp->f_data;
1629 
1630 	vp = filp->f_vnode;
1631 	if (vp == NULL)
1632 		return (EOPNOTSUPP);
1633 
1634 	/*
1635 	 * Ensure that file and memory protections are
1636 	 * compatible.
1637 	 */
1638 	mp = vp->v_mount;
1639 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1640 		maxprot = VM_PROT_NONE;
1641 		if ((prot & VM_PROT_EXECUTE) != 0)
1642 			return (EACCES);
1643 	} else
1644 		maxprot = VM_PROT_EXECUTE;
1645 	if ((fp->f_flag & FREAD) != 0)
1646 		maxprot |= VM_PROT_READ;
1647 	else if ((prot & VM_PROT_READ) != 0)
1648 		return (EACCES);
1649 
1650 	/*
1651 	 * If we are sharing potential changes via MAP_SHARED and we
1652 	 * are trying to get write permission although we opened it
1653 	 * without asking for it, bail out.
1654 	 *
1655 	 * Note that most character devices always share mappings.
1656 	 *
1657 	 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE
1658 	 * requests rather than doing it here.
1659 	 */
1660 	if ((flags & MAP_SHARED) != 0) {
1661 		if ((fp->f_flag & FWRITE) != 0)
1662 			maxprot |= VM_PROT_WRITE;
1663 		else if ((prot & VM_PROT_WRITE) != 0)
1664 			return (EACCES);
1665 	}
1666 	maxprot &= cap_maxprot;
1667 
1668 	linux_get_fop(filp, &fop, &ldev);
1669 	error = linux_file_mmap_sub(td, size, prot, &maxprot, &flags, fp,
1670 	    &foff, fop, &object);
1671 	if (error != 0)
1672 		goto out;
1673 
1674 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1675 	    foff, FALSE, td);
1676 	if (error != 0)
1677 		vm_object_deallocate(object);
1678 out:
1679 	linux_drop_fop(ldev);
1680 	return (error);
1681 }
1682 
1683 static int
1684 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
1685     struct thread *td)
1686 {
1687 	struct linux_file *filp;
1688 	struct vnode *vp;
1689 	int error;
1690 
1691 	filp = (struct linux_file *)fp->f_data;
1692 	if (filp->f_vnode == NULL)
1693 		return (EOPNOTSUPP);
1694 
1695 	vp = filp->f_vnode;
1696 
1697 	vn_lock(vp, LK_SHARED | LK_RETRY);
1698 	error = VOP_STAT(vp, sb, td->td_ucred, NOCRED, td);
1699 	VOP_UNLOCK(vp);
1700 
1701 	return (error);
1702 }
1703 
1704 static int
1705 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1706     struct filedesc *fdp)
1707 {
1708 	struct linux_file *filp;
1709 	struct vnode *vp;
1710 	int error;
1711 
1712 	filp = fp->f_data;
1713 	vp = filp->f_vnode;
1714 	if (vp == NULL) {
1715 		error = 0;
1716 		kif->kf_type = KF_TYPE_DEV;
1717 	} else {
1718 		vref(vp);
1719 		FILEDESC_SUNLOCK(fdp);
1720 		error = vn_fill_kinfo_vnode(vp, kif);
1721 		vrele(vp);
1722 		kif->kf_type = KF_TYPE_VNODE;
1723 		FILEDESC_SLOCK(fdp);
1724 	}
1725 	return (error);
1726 }
1727 
1728 unsigned int
1729 linux_iminor(struct inode *inode)
1730 {
1731 	struct linux_cdev *ldev;
1732 
1733 	if (inode == NULL || inode->v_rdev == NULL ||
1734 	    inode->v_rdev->si_devsw != &linuxcdevsw)
1735 		return (-1U);
1736 	ldev = inode->v_rdev->si_drv1;
1737 	if (ldev == NULL)
1738 		return (-1U);
1739 
1740 	return (minor(ldev->dev));
1741 }
1742 
1743 struct fileops linuxfileops = {
1744 	.fo_read = linux_file_read,
1745 	.fo_write = linux_file_write,
1746 	.fo_truncate = invfo_truncate,
1747 	.fo_kqfilter = linux_file_kqfilter,
1748 	.fo_stat = linux_file_stat,
1749 	.fo_fill_kinfo = linux_file_fill_kinfo,
1750 	.fo_poll = linux_file_poll,
1751 	.fo_close = linux_file_close,
1752 	.fo_ioctl = linux_file_ioctl,
1753 	.fo_mmap = linux_file_mmap,
1754 	.fo_chmod = invfo_chmod,
1755 	.fo_chown = invfo_chown,
1756 	.fo_sendfile = invfo_sendfile,
1757 	.fo_flags = DFLAG_PASSABLE,
1758 };
1759 
1760 /*
1761  * Hash of vmmap addresses.  This is infrequently accessed and does not
1762  * need to be particularly large.  This is done because we must store the
1763  * caller's idea of the map size to properly unmap.
1764  */
1765 struct vmmap {
1766 	LIST_ENTRY(vmmap)	vm_next;
1767 	void 			*vm_addr;
1768 	unsigned long		vm_size;
1769 };
1770 
1771 struct vmmaphd {
1772 	struct vmmap *lh_first;
1773 };
1774 #define	VMMAP_HASH_SIZE	64
1775 #define	VMMAP_HASH_MASK	(VMMAP_HASH_SIZE - 1)
1776 #define	VM_HASH(addr)	((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK
1777 static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE];
1778 static struct mtx vmmaplock;
1779 
1780 static void
1781 vmmap_add(void *addr, unsigned long size)
1782 {
1783 	struct vmmap *vmmap;
1784 
1785 	vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL);
1786 	mtx_lock(&vmmaplock);
1787 	vmmap->vm_size = size;
1788 	vmmap->vm_addr = addr;
1789 	LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next);
1790 	mtx_unlock(&vmmaplock);
1791 }
1792 
1793 static struct vmmap *
1794 vmmap_remove(void *addr)
1795 {
1796 	struct vmmap *vmmap;
1797 
1798 	mtx_lock(&vmmaplock);
1799 	LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next)
1800 		if (vmmap->vm_addr == addr)
1801 			break;
1802 	if (vmmap)
1803 		LIST_REMOVE(vmmap, vm_next);
1804 	mtx_unlock(&vmmaplock);
1805 
1806 	return (vmmap);
1807 }
1808 
1809 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__)
1810 void *
1811 _ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr)
1812 {
1813 	void *addr;
1814 
1815 	addr = pmap_mapdev_attr(phys_addr, size, attr);
1816 	if (addr == NULL)
1817 		return (NULL);
1818 	vmmap_add(addr, size);
1819 
1820 	return (addr);
1821 }
1822 #endif
1823 
1824 void
1825 iounmap(void *addr)
1826 {
1827 	struct vmmap *vmmap;
1828 
1829 	vmmap = vmmap_remove(addr);
1830 	if (vmmap == NULL)
1831 		return;
1832 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__)
1833 	pmap_unmapdev((vm_offset_t)addr, vmmap->vm_size);
1834 #endif
1835 	kfree(vmmap);
1836 }
1837 
1838 
1839 void *
1840 vmap(struct page **pages, unsigned int count, unsigned long flags, int prot)
1841 {
1842 	vm_offset_t off;
1843 	size_t size;
1844 
1845 	size = count * PAGE_SIZE;
1846 	off = kva_alloc(size);
1847 	if (off == 0)
1848 		return (NULL);
1849 	vmmap_add((void *)off, size);
1850 	pmap_qenter(off, pages, count);
1851 
1852 	return ((void *)off);
1853 }
1854 
1855 void
1856 vunmap(void *addr)
1857 {
1858 	struct vmmap *vmmap;
1859 
1860 	vmmap = vmmap_remove(addr);
1861 	if (vmmap == NULL)
1862 		return;
1863 	pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE);
1864 	kva_free((vm_offset_t)addr, vmmap->vm_size);
1865 	kfree(vmmap);
1866 }
1867 
1868 char *
1869 kvasprintf(gfp_t gfp, const char *fmt, va_list ap)
1870 {
1871 	unsigned int len;
1872 	char *p;
1873 	va_list aq;
1874 
1875 	va_copy(aq, ap);
1876 	len = vsnprintf(NULL, 0, fmt, aq);
1877 	va_end(aq);
1878 
1879 	p = kmalloc(len + 1, gfp);
1880 	if (p != NULL)
1881 		vsnprintf(p, len + 1, fmt, ap);
1882 
1883 	return (p);
1884 }
1885 
1886 char *
1887 kasprintf(gfp_t gfp, const char *fmt, ...)
1888 {
1889 	va_list ap;
1890 	char *p;
1891 
1892 	va_start(ap, fmt);
1893 	p = kvasprintf(gfp, fmt, ap);
1894 	va_end(ap);
1895 
1896 	return (p);
1897 }
1898 
1899 static void
1900 linux_timer_callback_wrapper(void *context)
1901 {
1902 	struct timer_list *timer;
1903 
1904 	linux_set_current(curthread);
1905 
1906 	timer = context;
1907 	timer->function(timer->data);
1908 }
1909 
1910 int
1911 mod_timer(struct timer_list *timer, int expires)
1912 {
1913 	int ret;
1914 
1915 	timer->expires = expires;
1916 	ret = callout_reset(&timer->callout,
1917 	    linux_timer_jiffies_until(expires),
1918 	    &linux_timer_callback_wrapper, timer);
1919 
1920 	MPASS(ret == 0 || ret == 1);
1921 
1922 	return (ret == 1);
1923 }
1924 
1925 void
1926 add_timer(struct timer_list *timer)
1927 {
1928 
1929 	callout_reset(&timer->callout,
1930 	    linux_timer_jiffies_until(timer->expires),
1931 	    &linux_timer_callback_wrapper, timer);
1932 }
1933 
1934 void
1935 add_timer_on(struct timer_list *timer, int cpu)
1936 {
1937 
1938 	callout_reset_on(&timer->callout,
1939 	    linux_timer_jiffies_until(timer->expires),
1940 	    &linux_timer_callback_wrapper, timer, cpu);
1941 }
1942 
1943 int
1944 del_timer(struct timer_list *timer)
1945 {
1946 
1947 	if (callout_stop(&(timer)->callout) == -1)
1948 		return (0);
1949 	return (1);
1950 }
1951 
1952 int
1953 del_timer_sync(struct timer_list *timer)
1954 {
1955 
1956 	if (callout_drain(&(timer)->callout) == -1)
1957 		return (0);
1958 	return (1);
1959 }
1960 
1961 /* greatest common divisor, Euclid equation */
1962 static uint64_t
1963 lkpi_gcd_64(uint64_t a, uint64_t b)
1964 {
1965 	uint64_t an;
1966 	uint64_t bn;
1967 
1968 	while (b != 0) {
1969 		an = b;
1970 		bn = a % b;
1971 		a = an;
1972 		b = bn;
1973 	}
1974 	return (a);
1975 }
1976 
1977 uint64_t lkpi_nsec2hz_rem;
1978 uint64_t lkpi_nsec2hz_div = 1000000000ULL;
1979 uint64_t lkpi_nsec2hz_max;
1980 
1981 uint64_t lkpi_usec2hz_rem;
1982 uint64_t lkpi_usec2hz_div = 1000000ULL;
1983 uint64_t lkpi_usec2hz_max;
1984 
1985 uint64_t lkpi_msec2hz_rem;
1986 uint64_t lkpi_msec2hz_div = 1000ULL;
1987 uint64_t lkpi_msec2hz_max;
1988 
1989 static void
1990 linux_timer_init(void *arg)
1991 {
1992 	uint64_t gcd;
1993 
1994 	/*
1995 	 * Compute an internal HZ value which can divide 2**32 to
1996 	 * avoid timer rounding problems when the tick value wraps
1997 	 * around 2**32:
1998 	 */
1999 	linux_timer_hz_mask = 1;
2000 	while (linux_timer_hz_mask < (unsigned long)hz)
2001 		linux_timer_hz_mask *= 2;
2002 	linux_timer_hz_mask--;
2003 
2004 	/* compute some internal constants */
2005 
2006 	lkpi_nsec2hz_rem = hz;
2007 	lkpi_usec2hz_rem = hz;
2008 	lkpi_msec2hz_rem = hz;
2009 
2010 	gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div);
2011 	lkpi_nsec2hz_rem /= gcd;
2012 	lkpi_nsec2hz_div /= gcd;
2013 	lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem;
2014 
2015 	gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div);
2016 	lkpi_usec2hz_rem /= gcd;
2017 	lkpi_usec2hz_div /= gcd;
2018 	lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem;
2019 
2020 	gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div);
2021 	lkpi_msec2hz_rem /= gcd;
2022 	lkpi_msec2hz_div /= gcd;
2023 	lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem;
2024 }
2025 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL);
2026 
2027 void
2028 linux_complete_common(struct completion *c, int all)
2029 {
2030 	int wakeup_swapper;
2031 
2032 	sleepq_lock(c);
2033 	if (all) {
2034 		c->done = UINT_MAX;
2035 		wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0);
2036 	} else {
2037 		if (c->done != UINT_MAX)
2038 			c->done++;
2039 		wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0);
2040 	}
2041 	sleepq_release(c);
2042 	if (wakeup_swapper)
2043 		kick_proc0();
2044 }
2045 
2046 /*
2047  * Indefinite wait for done != 0 with or without signals.
2048  */
2049 int
2050 linux_wait_for_common(struct completion *c, int flags)
2051 {
2052 	struct task_struct *task;
2053 	int error;
2054 
2055 	if (SCHEDULER_STOPPED())
2056 		return (0);
2057 
2058 	task = current;
2059 
2060 	if (flags != 0)
2061 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2062 	else
2063 		flags = SLEEPQ_SLEEP;
2064 	error = 0;
2065 	for (;;) {
2066 		sleepq_lock(c);
2067 		if (c->done)
2068 			break;
2069 		sleepq_add(c, NULL, "completion", flags, 0);
2070 		if (flags & SLEEPQ_INTERRUPTIBLE) {
2071 			DROP_GIANT();
2072 			error = -sleepq_wait_sig(c, 0);
2073 			PICKUP_GIANT();
2074 			if (error != 0) {
2075 				linux_schedule_save_interrupt_value(task, error);
2076 				error = -ERESTARTSYS;
2077 				goto intr;
2078 			}
2079 		} else {
2080 			DROP_GIANT();
2081 			sleepq_wait(c, 0);
2082 			PICKUP_GIANT();
2083 		}
2084 	}
2085 	if (c->done != UINT_MAX)
2086 		c->done--;
2087 	sleepq_release(c);
2088 
2089 intr:
2090 	return (error);
2091 }
2092 
2093 /*
2094  * Time limited wait for done != 0 with or without signals.
2095  */
2096 int
2097 linux_wait_for_timeout_common(struct completion *c, int timeout, int flags)
2098 {
2099 	struct task_struct *task;
2100 	int end = jiffies + timeout;
2101 	int error;
2102 
2103 	if (SCHEDULER_STOPPED())
2104 		return (0);
2105 
2106 	task = current;
2107 
2108 	if (flags != 0)
2109 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2110 	else
2111 		flags = SLEEPQ_SLEEP;
2112 
2113 	for (;;) {
2114 		sleepq_lock(c);
2115 		if (c->done)
2116 			break;
2117 		sleepq_add(c, NULL, "completion", flags, 0);
2118 		sleepq_set_timeout(c, linux_timer_jiffies_until(end));
2119 
2120 		DROP_GIANT();
2121 		if (flags & SLEEPQ_INTERRUPTIBLE)
2122 			error = -sleepq_timedwait_sig(c, 0);
2123 		else
2124 			error = -sleepq_timedwait(c, 0);
2125 		PICKUP_GIANT();
2126 
2127 		if (error != 0) {
2128 			/* check for timeout */
2129 			if (error == -EWOULDBLOCK) {
2130 				error = 0;	/* timeout */
2131 			} else {
2132 				/* signal happened */
2133 				linux_schedule_save_interrupt_value(task, error);
2134 				error = -ERESTARTSYS;
2135 			}
2136 			goto done;
2137 		}
2138 	}
2139 	if (c->done != UINT_MAX)
2140 		c->done--;
2141 	sleepq_release(c);
2142 
2143 	/* return how many jiffies are left */
2144 	error = linux_timer_jiffies_until(end);
2145 done:
2146 	return (error);
2147 }
2148 
2149 int
2150 linux_try_wait_for_completion(struct completion *c)
2151 {
2152 	int isdone;
2153 
2154 	sleepq_lock(c);
2155 	isdone = (c->done != 0);
2156 	if (c->done != 0 && c->done != UINT_MAX)
2157 		c->done--;
2158 	sleepq_release(c);
2159 	return (isdone);
2160 }
2161 
2162 int
2163 linux_completion_done(struct completion *c)
2164 {
2165 	int isdone;
2166 
2167 	sleepq_lock(c);
2168 	isdone = (c->done != 0);
2169 	sleepq_release(c);
2170 	return (isdone);
2171 }
2172 
2173 static void
2174 linux_cdev_deref(struct linux_cdev *ldev)
2175 {
2176 
2177 	if (refcount_release(&ldev->refs))
2178 		kfree(ldev);
2179 }
2180 
2181 static void
2182 linux_cdev_release(struct kobject *kobj)
2183 {
2184 	struct linux_cdev *cdev;
2185 	struct kobject *parent;
2186 
2187 	cdev = container_of(kobj, struct linux_cdev, kobj);
2188 	parent = kobj->parent;
2189 	linux_destroy_dev(cdev);
2190 	linux_cdev_deref(cdev);
2191 	kobject_put(parent);
2192 }
2193 
2194 static void
2195 linux_cdev_static_release(struct kobject *kobj)
2196 {
2197 	struct linux_cdev *cdev;
2198 	struct kobject *parent;
2199 
2200 	cdev = container_of(kobj, struct linux_cdev, kobj);
2201 	parent = kobj->parent;
2202 	linux_destroy_dev(cdev);
2203 	kobject_put(parent);
2204 }
2205 
2206 void
2207 linux_destroy_dev(struct linux_cdev *ldev)
2208 {
2209 
2210 	if (ldev->cdev == NULL)
2211 		return;
2212 
2213 	MPASS((ldev->siref & LDEV_SI_DTR) == 0);
2214 	atomic_set_int(&ldev->siref, LDEV_SI_DTR);
2215 	while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0)
2216 		pause("ldevdtr", hz / 4);
2217 
2218 	destroy_dev(ldev->cdev);
2219 	ldev->cdev = NULL;
2220 }
2221 
2222 const struct kobj_type linux_cdev_ktype = {
2223 	.release = linux_cdev_release,
2224 };
2225 
2226 const struct kobj_type linux_cdev_static_ktype = {
2227 	.release = linux_cdev_static_release,
2228 };
2229 
2230 static void
2231 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate)
2232 {
2233 	struct notifier_block *nb;
2234 
2235 	nb = arg;
2236 	if (linkstate == LINK_STATE_UP)
2237 		nb->notifier_call(nb, NETDEV_UP, ifp);
2238 	else
2239 		nb->notifier_call(nb, NETDEV_DOWN, ifp);
2240 }
2241 
2242 static void
2243 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp)
2244 {
2245 	struct notifier_block *nb;
2246 
2247 	nb = arg;
2248 	nb->notifier_call(nb, NETDEV_REGISTER, ifp);
2249 }
2250 
2251 static void
2252 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp)
2253 {
2254 	struct notifier_block *nb;
2255 
2256 	nb = arg;
2257 	nb->notifier_call(nb, NETDEV_UNREGISTER, ifp);
2258 }
2259 
2260 static void
2261 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp)
2262 {
2263 	struct notifier_block *nb;
2264 
2265 	nb = arg;
2266 	nb->notifier_call(nb, NETDEV_CHANGEADDR, ifp);
2267 }
2268 
2269 static void
2270 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp)
2271 {
2272 	struct notifier_block *nb;
2273 
2274 	nb = arg;
2275 	nb->notifier_call(nb, NETDEV_CHANGEIFADDR, ifp);
2276 }
2277 
2278 int
2279 register_netdevice_notifier(struct notifier_block *nb)
2280 {
2281 
2282 	nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER(
2283 	    ifnet_link_event, linux_handle_ifnet_link_event, nb, 0);
2284 	nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER(
2285 	    ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0);
2286 	nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER(
2287 	    ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0);
2288 	nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER(
2289 	    iflladdr_event, linux_handle_iflladdr_event, nb, 0);
2290 
2291 	return (0);
2292 }
2293 
2294 int
2295 register_inetaddr_notifier(struct notifier_block *nb)
2296 {
2297 
2298 	nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER(
2299 	    ifaddr_event, linux_handle_ifaddr_event, nb, 0);
2300 	return (0);
2301 }
2302 
2303 int
2304 unregister_netdevice_notifier(struct notifier_block *nb)
2305 {
2306 
2307 	EVENTHANDLER_DEREGISTER(ifnet_link_event,
2308 	    nb->tags[NETDEV_UP]);
2309 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event,
2310 	    nb->tags[NETDEV_REGISTER]);
2311 	EVENTHANDLER_DEREGISTER(ifnet_departure_event,
2312 	    nb->tags[NETDEV_UNREGISTER]);
2313 	EVENTHANDLER_DEREGISTER(iflladdr_event,
2314 	    nb->tags[NETDEV_CHANGEADDR]);
2315 
2316 	return (0);
2317 }
2318 
2319 int
2320 unregister_inetaddr_notifier(struct notifier_block *nb)
2321 {
2322 
2323 	EVENTHANDLER_DEREGISTER(ifaddr_event,
2324 	    nb->tags[NETDEV_CHANGEIFADDR]);
2325 
2326 	return (0);
2327 }
2328 
2329 struct list_sort_thunk {
2330 	int (*cmp)(void *, struct list_head *, struct list_head *);
2331 	void *priv;
2332 };
2333 
2334 static inline int
2335 linux_le_cmp(void *priv, const void *d1, const void *d2)
2336 {
2337 	struct list_head *le1, *le2;
2338 	struct list_sort_thunk *thunk;
2339 
2340 	thunk = priv;
2341 	le1 = *(__DECONST(struct list_head **, d1));
2342 	le2 = *(__DECONST(struct list_head **, d2));
2343 	return ((thunk->cmp)(thunk->priv, le1, le2));
2344 }
2345 
2346 void
2347 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv,
2348     struct list_head *a, struct list_head *b))
2349 {
2350 	struct list_sort_thunk thunk;
2351 	struct list_head **ar, *le;
2352 	size_t count, i;
2353 
2354 	count = 0;
2355 	list_for_each(le, head)
2356 		count++;
2357 	ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK);
2358 	i = 0;
2359 	list_for_each(le, head)
2360 		ar[i++] = le;
2361 	thunk.cmp = cmp;
2362 	thunk.priv = priv;
2363 	qsort_r(ar, count, sizeof(struct list_head *), &thunk, linux_le_cmp);
2364 	INIT_LIST_HEAD(head);
2365 	for (i = 0; i < count; i++)
2366 		list_add_tail(ar[i], head);
2367 	free(ar, M_KMALLOC);
2368 }
2369 
2370 void
2371 linux_irq_handler(void *ent)
2372 {
2373 	struct irq_ent *irqe;
2374 
2375 	linux_set_current(curthread);
2376 
2377 	irqe = ent;
2378 	irqe->handler(irqe->irq, irqe->arg);
2379 }
2380 
2381 #if defined(__i386__) || defined(__amd64__)
2382 int
2383 linux_wbinvd_on_all_cpus(void)
2384 {
2385 
2386 	pmap_invalidate_cache();
2387 	return (0);
2388 }
2389 #endif
2390 
2391 int
2392 linux_on_each_cpu(void callback(void *), void *data)
2393 {
2394 
2395 	smp_rendezvous(smp_no_rendezvous_barrier, callback,
2396 	    smp_no_rendezvous_barrier, data);
2397 	return (0);
2398 }
2399 
2400 int
2401 linux_in_atomic(void)
2402 {
2403 
2404 	return ((curthread->td_pflags & TDP_NOFAULTING) != 0);
2405 }
2406 
2407 struct linux_cdev *
2408 linux_find_cdev(const char *name, unsigned major, unsigned minor)
2409 {
2410 	dev_t dev = MKDEV(major, minor);
2411 	struct cdev *cdev;
2412 
2413 	dev_lock();
2414 	LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) {
2415 		struct linux_cdev *ldev = cdev->si_drv1;
2416 		if (ldev->dev == dev &&
2417 		    strcmp(kobject_name(&ldev->kobj), name) == 0) {
2418 			break;
2419 		}
2420 	}
2421 	dev_unlock();
2422 
2423 	return (cdev != NULL ? cdev->si_drv1 : NULL);
2424 }
2425 
2426 int
2427 __register_chrdev(unsigned int major, unsigned int baseminor,
2428     unsigned int count, const char *name,
2429     const struct file_operations *fops)
2430 {
2431 	struct linux_cdev *cdev;
2432 	int ret = 0;
2433 	int i;
2434 
2435 	for (i = baseminor; i < baseminor + count; i++) {
2436 		cdev = cdev_alloc();
2437 		cdev->ops = fops;
2438 		kobject_set_name(&cdev->kobj, name);
2439 
2440 		ret = cdev_add(cdev, makedev(major, i), 1);
2441 		if (ret != 0)
2442 			break;
2443 	}
2444 	return (ret);
2445 }
2446 
2447 int
2448 __register_chrdev_p(unsigned int major, unsigned int baseminor,
2449     unsigned int count, const char *name,
2450     const struct file_operations *fops, uid_t uid,
2451     gid_t gid, int mode)
2452 {
2453 	struct linux_cdev *cdev;
2454 	int ret = 0;
2455 	int i;
2456 
2457 	for (i = baseminor; i < baseminor + count; i++) {
2458 		cdev = cdev_alloc();
2459 		cdev->ops = fops;
2460 		kobject_set_name(&cdev->kobj, name);
2461 
2462 		ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode);
2463 		if (ret != 0)
2464 			break;
2465 	}
2466 	return (ret);
2467 }
2468 
2469 void
2470 __unregister_chrdev(unsigned int major, unsigned int baseminor,
2471     unsigned int count, const char *name)
2472 {
2473 	struct linux_cdev *cdevp;
2474 	int i;
2475 
2476 	for (i = baseminor; i < baseminor + count; i++) {
2477 		cdevp = linux_find_cdev(name, major, i);
2478 		if (cdevp != NULL)
2479 			cdev_del(cdevp);
2480 	}
2481 }
2482 
2483 void
2484 linux_dump_stack(void)
2485 {
2486 #ifdef STACK
2487 	struct stack st;
2488 
2489 	stack_zero(&st);
2490 	stack_save(&st);
2491 	stack_print(&st);
2492 #endif
2493 }
2494 
2495 #if defined(__i386__) || defined(__amd64__)
2496 bool linux_cpu_has_clflush;
2497 #endif
2498 
2499 static void
2500 linux_compat_init(void *arg)
2501 {
2502 	struct sysctl_oid *rootoid;
2503 	int i;
2504 
2505 #if defined(__i386__) || defined(__amd64__)
2506 	linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH);
2507 #endif
2508 	rw_init(&linux_vma_lock, "lkpi-vma-lock");
2509 
2510 	rootoid = SYSCTL_ADD_ROOT_NODE(NULL,
2511 	    OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys");
2512 	kobject_init(&linux_class_root, &linux_class_ktype);
2513 	kobject_set_name(&linux_class_root, "class");
2514 	linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid),
2515 	    OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class");
2516 	kobject_init(&linux_root_device.kobj, &linux_dev_ktype);
2517 	kobject_set_name(&linux_root_device.kobj, "device");
2518 	linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL,
2519 	    SYSCTL_CHILDREN(rootoid), OID_AUTO, "device",
2520 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "device");
2521 	linux_root_device.bsddev = root_bus;
2522 	linux_class_misc.name = "misc";
2523 	class_register(&linux_class_misc);
2524 	INIT_LIST_HEAD(&pci_drivers);
2525 	INIT_LIST_HEAD(&pci_devices);
2526 	spin_lock_init(&pci_lock);
2527 	mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF);
2528 	for (i = 0; i < VMMAP_HASH_SIZE; i++)
2529 		LIST_INIT(&vmmaphead[i]);
2530 	init_waitqueue_head(&linux_bit_waitq);
2531 	init_waitqueue_head(&linux_var_waitq);
2532 }
2533 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL);
2534 
2535 static void
2536 linux_compat_uninit(void *arg)
2537 {
2538 	linux_kobject_kfree_name(&linux_class_root);
2539 	linux_kobject_kfree_name(&linux_root_device.kobj);
2540 	linux_kobject_kfree_name(&linux_class_misc.kobj);
2541 
2542 	mtx_destroy(&vmmaplock);
2543 	spin_lock_destroy(&pci_lock);
2544 	rw_destroy(&linux_vma_lock);
2545 }
2546 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL);
2547 
2548 /*
2549  * NOTE: Linux frequently uses "unsigned long" for pointer to integer
2550  * conversion and vice versa, where in FreeBSD "uintptr_t" would be
2551  * used. Assert these types have the same size, else some parts of the
2552  * LinuxKPI may not work like expected:
2553  */
2554 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t));
2555