xref: /freebsd/sys/compat/linuxkpi/common/src/linux_compat.c (revision b9f654b163bce26de79705e77b872427c9f2afa1)
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 
90 #if defined(__i386__) || defined(__amd64__)
91 #include <asm/smp.h>
92 #endif
93 
94 SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW, 0, "LinuxKPI parameters");
95 
96 int linuxkpi_debug;
97 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN,
98     &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable.");
99 
100 MALLOC_DEFINE(M_KMALLOC, "linux", "Linux kmalloc compat");
101 
102 #include <linux/rbtree.h>
103 /* Undo Linux compat changes. */
104 #undef RB_ROOT
105 #undef file
106 #undef cdev
107 #define	RB_ROOT(head)	(head)->rbh_root
108 
109 static void linux_cdev_deref(struct linux_cdev *ldev);
110 static struct vm_area_struct *linux_cdev_handle_find(void *handle);
111 
112 struct kobject linux_class_root;
113 struct device linux_root_device;
114 struct class linux_class_misc;
115 struct list_head pci_drivers;
116 struct list_head pci_devices;
117 spinlock_t pci_lock;
118 
119 unsigned long linux_timer_hz_mask;
120 
121 int
122 panic_cmp(struct rb_node *one, struct rb_node *two)
123 {
124 	panic("no cmp");
125 }
126 
127 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp);
128 
129 int
130 kobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list args)
131 {
132 	va_list tmp_va;
133 	int len;
134 	char *old;
135 	char *name;
136 	char dummy;
137 
138 	old = kobj->name;
139 
140 	if (old && fmt == NULL)
141 		return (0);
142 
143 	/* compute length of string */
144 	va_copy(tmp_va, args);
145 	len = vsnprintf(&dummy, 0, fmt, tmp_va);
146 	va_end(tmp_va);
147 
148 	/* account for zero termination */
149 	len++;
150 
151 	/* check for error */
152 	if (len < 1)
153 		return (-EINVAL);
154 
155 	/* allocate memory for string */
156 	name = kzalloc(len, GFP_KERNEL);
157 	if (name == NULL)
158 		return (-ENOMEM);
159 	vsnprintf(name, len, fmt, args);
160 	kobj->name = name;
161 
162 	/* free old string */
163 	kfree(old);
164 
165 	/* filter new string */
166 	for (; *name != '\0'; name++)
167 		if (*name == '/')
168 			*name = '!';
169 	return (0);
170 }
171 
172 int
173 kobject_set_name(struct kobject *kobj, const char *fmt, ...)
174 {
175 	va_list args;
176 	int error;
177 
178 	va_start(args, fmt);
179 	error = kobject_set_name_vargs(kobj, fmt, args);
180 	va_end(args);
181 
182 	return (error);
183 }
184 
185 static int
186 kobject_add_complete(struct kobject *kobj, struct kobject *parent)
187 {
188 	const struct kobj_type *t;
189 	int error;
190 
191 	kobj->parent = parent;
192 	error = sysfs_create_dir(kobj);
193 	if (error == 0 && kobj->ktype && kobj->ktype->default_attrs) {
194 		struct attribute **attr;
195 		t = kobj->ktype;
196 
197 		for (attr = t->default_attrs; *attr != NULL; attr++) {
198 			error = sysfs_create_file(kobj, *attr);
199 			if (error)
200 				break;
201 		}
202 		if (error)
203 			sysfs_remove_dir(kobj);
204 
205 	}
206 	return (error);
207 }
208 
209 int
210 kobject_add(struct kobject *kobj, struct kobject *parent, const char *fmt, ...)
211 {
212 	va_list args;
213 	int error;
214 
215 	va_start(args, fmt);
216 	error = kobject_set_name_vargs(kobj, fmt, args);
217 	va_end(args);
218 	if (error)
219 		return (error);
220 
221 	return kobject_add_complete(kobj, parent);
222 }
223 
224 void
225 linux_kobject_release(struct kref *kref)
226 {
227 	struct kobject *kobj;
228 	char *name;
229 
230 	kobj = container_of(kref, struct kobject, kref);
231 	sysfs_remove_dir(kobj);
232 	name = kobj->name;
233 	if (kobj->ktype && kobj->ktype->release)
234 		kobj->ktype->release(kobj);
235 	kfree(name);
236 }
237 
238 static void
239 linux_kobject_kfree(struct kobject *kobj)
240 {
241 	kfree(kobj);
242 }
243 
244 static void
245 linux_kobject_kfree_name(struct kobject *kobj)
246 {
247 	if (kobj) {
248 		kfree(kobj->name);
249 	}
250 }
251 
252 const struct kobj_type linux_kfree_type = {
253 	.release = linux_kobject_kfree
254 };
255 
256 static void
257 linux_device_release(struct device *dev)
258 {
259 	pr_debug("linux_device_release: %s\n", dev_name(dev));
260 	kfree(dev);
261 }
262 
263 static ssize_t
264 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf)
265 {
266 	struct class_attribute *dattr;
267 	ssize_t error;
268 
269 	dattr = container_of(attr, struct class_attribute, attr);
270 	error = -EIO;
271 	if (dattr->show)
272 		error = dattr->show(container_of(kobj, struct class, kobj),
273 		    dattr, buf);
274 	return (error);
275 }
276 
277 static ssize_t
278 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf,
279     size_t count)
280 {
281 	struct class_attribute *dattr;
282 	ssize_t error;
283 
284 	dattr = container_of(attr, struct class_attribute, attr);
285 	error = -EIO;
286 	if (dattr->store)
287 		error = dattr->store(container_of(kobj, struct class, kobj),
288 		    dattr, buf, count);
289 	return (error);
290 }
291 
292 static void
293 linux_class_release(struct kobject *kobj)
294 {
295 	struct class *class;
296 
297 	class = container_of(kobj, struct class, kobj);
298 	if (class->class_release)
299 		class->class_release(class);
300 }
301 
302 static const struct sysfs_ops linux_class_sysfs = {
303 	.show  = linux_class_show,
304 	.store = linux_class_store,
305 };
306 
307 const struct kobj_type linux_class_ktype = {
308 	.release = linux_class_release,
309 	.sysfs_ops = &linux_class_sysfs
310 };
311 
312 static void
313 linux_dev_release(struct kobject *kobj)
314 {
315 	struct device *dev;
316 
317 	dev = container_of(kobj, struct device, kobj);
318 	/* This is the precedence defined by linux. */
319 	if (dev->release)
320 		dev->release(dev);
321 	else if (dev->class && dev->class->dev_release)
322 		dev->class->dev_release(dev);
323 }
324 
325 static ssize_t
326 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf)
327 {
328 	struct device_attribute *dattr;
329 	ssize_t error;
330 
331 	dattr = container_of(attr, struct device_attribute, attr);
332 	error = -EIO;
333 	if (dattr->show)
334 		error = dattr->show(container_of(kobj, struct device, kobj),
335 		    dattr, buf);
336 	return (error);
337 }
338 
339 static ssize_t
340 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf,
341     size_t count)
342 {
343 	struct device_attribute *dattr;
344 	ssize_t error;
345 
346 	dattr = container_of(attr, struct device_attribute, attr);
347 	error = -EIO;
348 	if (dattr->store)
349 		error = dattr->store(container_of(kobj, struct device, kobj),
350 		    dattr, buf, count);
351 	return (error);
352 }
353 
354 static const struct sysfs_ops linux_dev_sysfs = {
355 	.show  = linux_dev_show,
356 	.store = linux_dev_store,
357 };
358 
359 const struct kobj_type linux_dev_ktype = {
360 	.release = linux_dev_release,
361 	.sysfs_ops = &linux_dev_sysfs
362 };
363 
364 struct device *
365 device_create(struct class *class, struct device *parent, dev_t devt,
366     void *drvdata, const char *fmt, ...)
367 {
368 	struct device *dev;
369 	va_list args;
370 
371 	dev = kzalloc(sizeof(*dev), M_WAITOK);
372 	dev->parent = parent;
373 	dev->class = class;
374 	dev->devt = devt;
375 	dev->driver_data = drvdata;
376 	dev->release = linux_device_release;
377 	va_start(args, fmt);
378 	kobject_set_name_vargs(&dev->kobj, fmt, args);
379 	va_end(args);
380 	device_register(dev);
381 
382 	return (dev);
383 }
384 
385 int
386 kobject_init_and_add(struct kobject *kobj, const struct kobj_type *ktype,
387     struct kobject *parent, const char *fmt, ...)
388 {
389 	va_list args;
390 	int error;
391 
392 	kobject_init(kobj, ktype);
393 	kobj->ktype = ktype;
394 	kobj->parent = parent;
395 	kobj->name = NULL;
396 
397 	va_start(args, fmt);
398 	error = kobject_set_name_vargs(kobj, fmt, args);
399 	va_end(args);
400 	if (error)
401 		return (error);
402 	return kobject_add_complete(kobj, parent);
403 }
404 
405 static void
406 linux_kq_lock(void *arg)
407 {
408 	spinlock_t *s = arg;
409 
410 	spin_lock(s);
411 }
412 static void
413 linux_kq_unlock(void *arg)
414 {
415 	spinlock_t *s = arg;
416 
417 	spin_unlock(s);
418 }
419 
420 static void
421 linux_kq_lock_owned(void *arg)
422 {
423 #ifdef INVARIANTS
424 	spinlock_t *s = arg;
425 
426 	mtx_assert(&s->m, MA_OWNED);
427 #endif
428 }
429 
430 static void
431 linux_kq_lock_unowned(void *arg)
432 {
433 #ifdef INVARIANTS
434 	spinlock_t *s = arg;
435 
436 	mtx_assert(&s->m, MA_NOTOWNED);
437 #endif
438 }
439 
440 static void
441 linux_file_kqfilter_poll(struct linux_file *, int);
442 
443 struct linux_file *
444 linux_file_alloc(void)
445 {
446 	struct linux_file *filp;
447 
448 	filp = kzalloc(sizeof(*filp), GFP_KERNEL);
449 
450 	/* set initial refcount */
451 	filp->f_count = 1;
452 
453 	/* setup fields needed by kqueue support */
454 	spin_lock_init(&filp->f_kqlock);
455 	knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock,
456 	    linux_kq_lock, linux_kq_unlock,
457 	    linux_kq_lock_owned, linux_kq_lock_unowned);
458 
459 	return (filp);
460 }
461 
462 void
463 linux_file_free(struct linux_file *filp)
464 {
465 	if (filp->_file == NULL) {
466 		if (filp->f_shmem != NULL)
467 			vm_object_deallocate(filp->f_shmem);
468 		kfree(filp);
469 	} else {
470 		/*
471 		 * The close method of the character device or file
472 		 * will free the linux_file structure:
473 		 */
474 		_fdrop(filp->_file, curthread);
475 	}
476 }
477 
478 static int
479 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
480     vm_page_t *mres)
481 {
482 	struct vm_area_struct *vmap;
483 
484 	vmap = linux_cdev_handle_find(vm_obj->handle);
485 
486 	MPASS(vmap != NULL);
487 	MPASS(vmap->vm_private_data == vm_obj->handle);
488 
489 	if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) {
490 		vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset;
491 		vm_page_t page;
492 
493 		if (((*mres)->flags & PG_FICTITIOUS) != 0) {
494 			/*
495 			 * If the passed in result page is a fake
496 			 * page, update it with the new physical
497 			 * address.
498 			 */
499 			page = *mres;
500 			vm_page_updatefake(page, paddr, vm_obj->memattr);
501 		} else {
502 			/*
503 			 * Replace the passed in "mres" page with our
504 			 * own fake page and free up the all of the
505 			 * original pages.
506 			 */
507 			VM_OBJECT_WUNLOCK(vm_obj);
508 			page = vm_page_getfake(paddr, vm_obj->memattr);
509 			VM_OBJECT_WLOCK(vm_obj);
510 
511 			vm_page_replace_checked(page, vm_obj,
512 			    (*mres)->pindex, *mres);
513 
514 			vm_page_lock(*mres);
515 			vm_page_free(*mres);
516 			vm_page_unlock(*mres);
517 			*mres = page;
518 		}
519 		page->valid = VM_PAGE_BITS_ALL;
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 	linux_set_current(curthread);
533 
534 	/* get VM area structure */
535 	vmap = linux_cdev_handle_find(vm_obj->handle);
536 	MPASS(vmap != NULL);
537 	MPASS(vmap->vm_private_data == vm_obj->handle);
538 
539 	VM_OBJECT_WUNLOCK(vm_obj);
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 	const struct file_operations *fop;
1498 	struct linux_cdev *ldev;
1499 	int error;
1500 
1501 	filp = (struct linux_file *)file->f_data;
1502 
1503 	KASSERT(file_count(filp) == 0,
1504 	    ("File refcount(%d) is not zero", file_count(filp)));
1505 
1506 	error = 0;
1507 	filp->f_flags = file->f_flag;
1508 	linux_set_current(td);
1509 	linux_poll_wait_dequeue(filp);
1510 	linux_get_fop(filp, &fop, &ldev);
1511 	if (fop->release != NULL)
1512 		error = -OPW(file, td, fop->release(filp->f_vnode, filp));
1513 	funsetown(&filp->f_sigio);
1514 	if (filp->f_vnode != NULL)
1515 		vdrop(filp->f_vnode);
1516 	linux_drop_fop(ldev);
1517 	if (filp->f_cdev != NULL)
1518 		linux_cdev_deref(filp->f_cdev);
1519 	kfree(filp);
1520 
1521 	return (error);
1522 }
1523 
1524 static int
1525 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
1526     struct thread *td)
1527 {
1528 	struct linux_file *filp;
1529 	const struct file_operations *fop;
1530 	struct linux_cdev *ldev;
1531 	int error;
1532 
1533 	error = 0;
1534 	filp = (struct linux_file *)fp->f_data;
1535 	filp->f_flags = fp->f_flag;
1536 	linux_get_fop(filp, &fop, &ldev);
1537 
1538 	linux_set_current(td);
1539 	switch (cmd) {
1540 	case FIONBIO:
1541 		break;
1542 	case FIOASYNC:
1543 		if (fop->fasync == NULL)
1544 			break;
1545 		error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC));
1546 		break;
1547 	case FIOSETOWN:
1548 		error = fsetown(*(int *)data, &filp->f_sigio);
1549 		if (error == 0) {
1550 			if (fop->fasync == NULL)
1551 				break;
1552 			error = -OPW(fp, td, fop->fasync(0, filp,
1553 			    fp->f_flag & FASYNC));
1554 		}
1555 		break;
1556 	case FIOGETOWN:
1557 		*(int *)data = fgetown(&filp->f_sigio);
1558 		break;
1559 	default:
1560 		error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td);
1561 		break;
1562 	}
1563 	linux_drop_fop(ldev);
1564 	return (error);
1565 }
1566 
1567 static int
1568 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1569     vm_prot_t *maxprotp, int *flagsp, struct file *fp,
1570     vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp)
1571 {
1572 	/*
1573 	 * Character devices do not provide private mappings
1574 	 * of any kind:
1575 	 */
1576 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1577 	    (prot & VM_PROT_WRITE) != 0)
1578 		return (EACCES);
1579 	if ((*flagsp & (MAP_PRIVATE | MAP_COPY)) != 0)
1580 		return (EINVAL);
1581 
1582 	return (linux_file_mmap_single(fp, fop, foff, objsize, objp,
1583 	    (int)prot, td));
1584 }
1585 
1586 static int
1587 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1588     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1589     struct thread *td)
1590 {
1591 	struct linux_file *filp;
1592 	const struct file_operations *fop;
1593 	struct linux_cdev *ldev;
1594 	struct mount *mp;
1595 	struct vnode *vp;
1596 	vm_object_t object;
1597 	vm_prot_t maxprot;
1598 	int error;
1599 
1600 	filp = (struct linux_file *)fp->f_data;
1601 
1602 	vp = filp->f_vnode;
1603 	if (vp == NULL)
1604 		return (EOPNOTSUPP);
1605 
1606 	/*
1607 	 * Ensure that file and memory protections are
1608 	 * compatible.
1609 	 */
1610 	mp = vp->v_mount;
1611 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1612 		maxprot = VM_PROT_NONE;
1613 		if ((prot & VM_PROT_EXECUTE) != 0)
1614 			return (EACCES);
1615 	} else
1616 		maxprot = VM_PROT_EXECUTE;
1617 	if ((fp->f_flag & FREAD) != 0)
1618 		maxprot |= VM_PROT_READ;
1619 	else if ((prot & VM_PROT_READ) != 0)
1620 		return (EACCES);
1621 
1622 	/*
1623 	 * If we are sharing potential changes via MAP_SHARED and we
1624 	 * are trying to get write permission although we opened it
1625 	 * without asking for it, bail out.
1626 	 *
1627 	 * Note that most character devices always share mappings.
1628 	 *
1629 	 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE
1630 	 * requests rather than doing it here.
1631 	 */
1632 	if ((flags & MAP_SHARED) != 0) {
1633 		if ((fp->f_flag & FWRITE) != 0)
1634 			maxprot |= VM_PROT_WRITE;
1635 		else if ((prot & VM_PROT_WRITE) != 0)
1636 			return (EACCES);
1637 	}
1638 	maxprot &= cap_maxprot;
1639 
1640 	linux_get_fop(filp, &fop, &ldev);
1641 	error = linux_file_mmap_sub(td, size, prot, &maxprot, &flags, fp,
1642 	    &foff, fop, &object);
1643 	if (error != 0)
1644 		goto out;
1645 
1646 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1647 	    foff, FALSE, td);
1648 	if (error != 0)
1649 		vm_object_deallocate(object);
1650 out:
1651 	linux_drop_fop(ldev);
1652 	return (error);
1653 }
1654 
1655 static int
1656 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
1657     struct thread *td)
1658 {
1659 	struct linux_file *filp;
1660 	struct vnode *vp;
1661 	int error;
1662 
1663 	filp = (struct linux_file *)fp->f_data;
1664 	if (filp->f_vnode == NULL)
1665 		return (EOPNOTSUPP);
1666 
1667 	vp = filp->f_vnode;
1668 
1669 	vn_lock(vp, LK_SHARED | LK_RETRY);
1670 	error = vn_stat(vp, sb, td->td_ucred, NOCRED, td);
1671 	VOP_UNLOCK(vp, 0);
1672 
1673 	return (error);
1674 }
1675 
1676 static int
1677 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1678     struct filedesc *fdp)
1679 {
1680 	struct linux_file *filp;
1681 	struct vnode *vp;
1682 	int error;
1683 
1684 	filp = fp->f_data;
1685 	vp = filp->f_vnode;
1686 	if (vp == NULL) {
1687 		error = 0;
1688 		kif->kf_type = KF_TYPE_DEV;
1689 	} else {
1690 		vref(vp);
1691 		FILEDESC_SUNLOCK(fdp);
1692 		error = vn_fill_kinfo_vnode(vp, kif);
1693 		vrele(vp);
1694 		kif->kf_type = KF_TYPE_VNODE;
1695 		FILEDESC_SLOCK(fdp);
1696 	}
1697 	return (error);
1698 }
1699 
1700 unsigned int
1701 linux_iminor(struct inode *inode)
1702 {
1703 	struct linux_cdev *ldev;
1704 
1705 	if (inode == NULL || inode->v_rdev == NULL ||
1706 	    inode->v_rdev->si_devsw != &linuxcdevsw)
1707 		return (-1U);
1708 	ldev = inode->v_rdev->si_drv1;
1709 	if (ldev == NULL)
1710 		return (-1U);
1711 
1712 	return (minor(ldev->dev));
1713 }
1714 
1715 struct fileops linuxfileops = {
1716 	.fo_read = linux_file_read,
1717 	.fo_write = linux_file_write,
1718 	.fo_truncate = invfo_truncate,
1719 	.fo_kqfilter = linux_file_kqfilter,
1720 	.fo_stat = linux_file_stat,
1721 	.fo_fill_kinfo = linux_file_fill_kinfo,
1722 	.fo_poll = linux_file_poll,
1723 	.fo_close = linux_file_close,
1724 	.fo_ioctl = linux_file_ioctl,
1725 	.fo_mmap = linux_file_mmap,
1726 	.fo_chmod = invfo_chmod,
1727 	.fo_chown = invfo_chown,
1728 	.fo_sendfile = invfo_sendfile,
1729 	.fo_flags = DFLAG_PASSABLE,
1730 };
1731 
1732 /*
1733  * Hash of vmmap addresses.  This is infrequently accessed and does not
1734  * need to be particularly large.  This is done because we must store the
1735  * caller's idea of the map size to properly unmap.
1736  */
1737 struct vmmap {
1738 	LIST_ENTRY(vmmap)	vm_next;
1739 	void 			*vm_addr;
1740 	unsigned long		vm_size;
1741 };
1742 
1743 struct vmmaphd {
1744 	struct vmmap *lh_first;
1745 };
1746 #define	VMMAP_HASH_SIZE	64
1747 #define	VMMAP_HASH_MASK	(VMMAP_HASH_SIZE - 1)
1748 #define	VM_HASH(addr)	((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK
1749 static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE];
1750 static struct mtx vmmaplock;
1751 
1752 static void
1753 vmmap_add(void *addr, unsigned long size)
1754 {
1755 	struct vmmap *vmmap;
1756 
1757 	vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL);
1758 	mtx_lock(&vmmaplock);
1759 	vmmap->vm_size = size;
1760 	vmmap->vm_addr = addr;
1761 	LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next);
1762 	mtx_unlock(&vmmaplock);
1763 }
1764 
1765 static struct vmmap *
1766 vmmap_remove(void *addr)
1767 {
1768 	struct vmmap *vmmap;
1769 
1770 	mtx_lock(&vmmaplock);
1771 	LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next)
1772 		if (vmmap->vm_addr == addr)
1773 			break;
1774 	if (vmmap)
1775 		LIST_REMOVE(vmmap, vm_next);
1776 	mtx_unlock(&vmmaplock);
1777 
1778 	return (vmmap);
1779 }
1780 
1781 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__)
1782 void *
1783 _ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr)
1784 {
1785 	void *addr;
1786 
1787 	addr = pmap_mapdev_attr(phys_addr, size, attr);
1788 	if (addr == NULL)
1789 		return (NULL);
1790 	vmmap_add(addr, size);
1791 
1792 	return (addr);
1793 }
1794 #endif
1795 
1796 void
1797 iounmap(void *addr)
1798 {
1799 	struct vmmap *vmmap;
1800 
1801 	vmmap = vmmap_remove(addr);
1802 	if (vmmap == NULL)
1803 		return;
1804 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__)
1805 	pmap_unmapdev((vm_offset_t)addr, vmmap->vm_size);
1806 #endif
1807 	kfree(vmmap);
1808 }
1809 
1810 
1811 void *
1812 vmap(struct page **pages, unsigned int count, unsigned long flags, int prot)
1813 {
1814 	vm_offset_t off;
1815 	size_t size;
1816 
1817 	size = count * PAGE_SIZE;
1818 	off = kva_alloc(size);
1819 	if (off == 0)
1820 		return (NULL);
1821 	vmmap_add((void *)off, size);
1822 	pmap_qenter(off, pages, count);
1823 
1824 	return ((void *)off);
1825 }
1826 
1827 void
1828 vunmap(void *addr)
1829 {
1830 	struct vmmap *vmmap;
1831 
1832 	vmmap = vmmap_remove(addr);
1833 	if (vmmap == NULL)
1834 		return;
1835 	pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE);
1836 	kva_free((vm_offset_t)addr, vmmap->vm_size);
1837 	kfree(vmmap);
1838 }
1839 
1840 char *
1841 kvasprintf(gfp_t gfp, const char *fmt, va_list ap)
1842 {
1843 	unsigned int len;
1844 	char *p;
1845 	va_list aq;
1846 
1847 	va_copy(aq, ap);
1848 	len = vsnprintf(NULL, 0, fmt, aq);
1849 	va_end(aq);
1850 
1851 	p = kmalloc(len + 1, gfp);
1852 	if (p != NULL)
1853 		vsnprintf(p, len + 1, fmt, ap);
1854 
1855 	return (p);
1856 }
1857 
1858 char *
1859 kasprintf(gfp_t gfp, const char *fmt, ...)
1860 {
1861 	va_list ap;
1862 	char *p;
1863 
1864 	va_start(ap, fmt);
1865 	p = kvasprintf(gfp, fmt, ap);
1866 	va_end(ap);
1867 
1868 	return (p);
1869 }
1870 
1871 static void
1872 linux_timer_callback_wrapper(void *context)
1873 {
1874 	struct timer_list *timer;
1875 
1876 	linux_set_current(curthread);
1877 
1878 	timer = context;
1879 	timer->function(timer->data);
1880 }
1881 
1882 void
1883 mod_timer(struct timer_list *timer, int expires)
1884 {
1885 
1886 	timer->expires = expires;
1887 	callout_reset(&timer->callout,
1888 	    linux_timer_jiffies_until(expires),
1889 	    &linux_timer_callback_wrapper, timer);
1890 }
1891 
1892 void
1893 add_timer(struct timer_list *timer)
1894 {
1895 
1896 	callout_reset(&timer->callout,
1897 	    linux_timer_jiffies_until(timer->expires),
1898 	    &linux_timer_callback_wrapper, timer);
1899 }
1900 
1901 void
1902 add_timer_on(struct timer_list *timer, int cpu)
1903 {
1904 
1905 	callout_reset_on(&timer->callout,
1906 	    linux_timer_jiffies_until(timer->expires),
1907 	    &linux_timer_callback_wrapper, timer, cpu);
1908 }
1909 
1910 int
1911 del_timer(struct timer_list *timer)
1912 {
1913 
1914 	if (callout_stop(&(timer)->callout) == -1)
1915 		return (0);
1916 	return (1);
1917 }
1918 
1919 static void
1920 linux_timer_init(void *arg)
1921 {
1922 
1923 	/*
1924 	 * Compute an internal HZ value which can divide 2**32 to
1925 	 * avoid timer rounding problems when the tick value wraps
1926 	 * around 2**32:
1927 	 */
1928 	linux_timer_hz_mask = 1;
1929 	while (linux_timer_hz_mask < (unsigned long)hz)
1930 		linux_timer_hz_mask *= 2;
1931 	linux_timer_hz_mask--;
1932 }
1933 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL);
1934 
1935 void
1936 linux_complete_common(struct completion *c, int all)
1937 {
1938 	int wakeup_swapper;
1939 
1940 	sleepq_lock(c);
1941 	if (all) {
1942 		c->done = UINT_MAX;
1943 		wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0);
1944 	} else {
1945 		if (c->done != UINT_MAX)
1946 			c->done++;
1947 		wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0);
1948 	}
1949 	sleepq_release(c);
1950 	if (wakeup_swapper)
1951 		kick_proc0();
1952 }
1953 
1954 /*
1955  * Indefinite wait for done != 0 with or without signals.
1956  */
1957 int
1958 linux_wait_for_common(struct completion *c, int flags)
1959 {
1960 	struct task_struct *task;
1961 	int error;
1962 
1963 	if (SCHEDULER_STOPPED())
1964 		return (0);
1965 
1966 	task = current;
1967 
1968 	if (flags != 0)
1969 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
1970 	else
1971 		flags = SLEEPQ_SLEEP;
1972 	error = 0;
1973 	for (;;) {
1974 		sleepq_lock(c);
1975 		if (c->done)
1976 			break;
1977 		sleepq_add(c, NULL, "completion", flags, 0);
1978 		if (flags & SLEEPQ_INTERRUPTIBLE) {
1979 			DROP_GIANT();
1980 			error = -sleepq_wait_sig(c, 0);
1981 			PICKUP_GIANT();
1982 			if (error != 0) {
1983 				linux_schedule_save_interrupt_value(task, error);
1984 				error = -ERESTARTSYS;
1985 				goto intr;
1986 			}
1987 		} else {
1988 			DROP_GIANT();
1989 			sleepq_wait(c, 0);
1990 			PICKUP_GIANT();
1991 		}
1992 	}
1993 	if (c->done != UINT_MAX)
1994 		c->done--;
1995 	sleepq_release(c);
1996 
1997 intr:
1998 	return (error);
1999 }
2000 
2001 /*
2002  * Time limited wait for done != 0 with or without signals.
2003  */
2004 int
2005 linux_wait_for_timeout_common(struct completion *c, int timeout, int flags)
2006 {
2007 	struct task_struct *task;
2008 	int end = jiffies + timeout;
2009 	int error;
2010 
2011 	if (SCHEDULER_STOPPED())
2012 		return (0);
2013 
2014 	task = current;
2015 
2016 	if (flags != 0)
2017 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2018 	else
2019 		flags = SLEEPQ_SLEEP;
2020 
2021 	for (;;) {
2022 		sleepq_lock(c);
2023 		if (c->done)
2024 			break;
2025 		sleepq_add(c, NULL, "completion", flags, 0);
2026 		sleepq_set_timeout(c, linux_timer_jiffies_until(end));
2027 
2028 		DROP_GIANT();
2029 		if (flags & SLEEPQ_INTERRUPTIBLE)
2030 			error = -sleepq_timedwait_sig(c, 0);
2031 		else
2032 			error = -sleepq_timedwait(c, 0);
2033 		PICKUP_GIANT();
2034 
2035 		if (error != 0) {
2036 			/* check for timeout */
2037 			if (error == -EWOULDBLOCK) {
2038 				error = 0;	/* timeout */
2039 			} else {
2040 				/* signal happened */
2041 				linux_schedule_save_interrupt_value(task, error);
2042 				error = -ERESTARTSYS;
2043 			}
2044 			goto done;
2045 		}
2046 	}
2047 	if (c->done != UINT_MAX)
2048 		c->done--;
2049 	sleepq_release(c);
2050 
2051 	/* return how many jiffies are left */
2052 	error = linux_timer_jiffies_until(end);
2053 done:
2054 	return (error);
2055 }
2056 
2057 int
2058 linux_try_wait_for_completion(struct completion *c)
2059 {
2060 	int isdone;
2061 
2062 	sleepq_lock(c);
2063 	isdone = (c->done != 0);
2064 	if (c->done != 0 && c->done != UINT_MAX)
2065 		c->done--;
2066 	sleepq_release(c);
2067 	return (isdone);
2068 }
2069 
2070 int
2071 linux_completion_done(struct completion *c)
2072 {
2073 	int isdone;
2074 
2075 	sleepq_lock(c);
2076 	isdone = (c->done != 0);
2077 	sleepq_release(c);
2078 	return (isdone);
2079 }
2080 
2081 static void
2082 linux_cdev_deref(struct linux_cdev *ldev)
2083 {
2084 
2085 	if (refcount_release(&ldev->refs))
2086 		kfree(ldev);
2087 }
2088 
2089 static void
2090 linux_cdev_release(struct kobject *kobj)
2091 {
2092 	struct linux_cdev *cdev;
2093 	struct kobject *parent;
2094 
2095 	cdev = container_of(kobj, struct linux_cdev, kobj);
2096 	parent = kobj->parent;
2097 	linux_destroy_dev(cdev);
2098 	linux_cdev_deref(cdev);
2099 	kobject_put(parent);
2100 }
2101 
2102 static void
2103 linux_cdev_static_release(struct kobject *kobj)
2104 {
2105 	struct linux_cdev *cdev;
2106 	struct kobject *parent;
2107 
2108 	cdev = container_of(kobj, struct linux_cdev, kobj);
2109 	parent = kobj->parent;
2110 	linux_destroy_dev(cdev);
2111 	kobject_put(parent);
2112 }
2113 
2114 void
2115 linux_destroy_dev(struct linux_cdev *ldev)
2116 {
2117 
2118 	if (ldev->cdev == NULL)
2119 		return;
2120 
2121 	MPASS((ldev->siref & LDEV_SI_DTR) == 0);
2122 	atomic_set_int(&ldev->siref, LDEV_SI_DTR);
2123 	while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0)
2124 		pause("ldevdtr", hz / 4);
2125 
2126 	destroy_dev(ldev->cdev);
2127 	ldev->cdev = NULL;
2128 }
2129 
2130 const struct kobj_type linux_cdev_ktype = {
2131 	.release = linux_cdev_release,
2132 };
2133 
2134 const struct kobj_type linux_cdev_static_ktype = {
2135 	.release = linux_cdev_static_release,
2136 };
2137 
2138 static void
2139 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate)
2140 {
2141 	struct notifier_block *nb;
2142 
2143 	nb = arg;
2144 	if (linkstate == LINK_STATE_UP)
2145 		nb->notifier_call(nb, NETDEV_UP, ifp);
2146 	else
2147 		nb->notifier_call(nb, NETDEV_DOWN, ifp);
2148 }
2149 
2150 static void
2151 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp)
2152 {
2153 	struct notifier_block *nb;
2154 
2155 	nb = arg;
2156 	nb->notifier_call(nb, NETDEV_REGISTER, ifp);
2157 }
2158 
2159 static void
2160 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp)
2161 {
2162 	struct notifier_block *nb;
2163 
2164 	nb = arg;
2165 	nb->notifier_call(nb, NETDEV_UNREGISTER, ifp);
2166 }
2167 
2168 static void
2169 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp)
2170 {
2171 	struct notifier_block *nb;
2172 
2173 	nb = arg;
2174 	nb->notifier_call(nb, NETDEV_CHANGEADDR, ifp);
2175 }
2176 
2177 static void
2178 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp)
2179 {
2180 	struct notifier_block *nb;
2181 
2182 	nb = arg;
2183 	nb->notifier_call(nb, NETDEV_CHANGEIFADDR, ifp);
2184 }
2185 
2186 int
2187 register_netdevice_notifier(struct notifier_block *nb)
2188 {
2189 
2190 	nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER(
2191 	    ifnet_link_event, linux_handle_ifnet_link_event, nb, 0);
2192 	nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER(
2193 	    ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0);
2194 	nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER(
2195 	    ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0);
2196 	nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER(
2197 	    iflladdr_event, linux_handle_iflladdr_event, nb, 0);
2198 
2199 	return (0);
2200 }
2201 
2202 int
2203 register_inetaddr_notifier(struct notifier_block *nb)
2204 {
2205 
2206 	nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER(
2207 	    ifaddr_event, linux_handle_ifaddr_event, nb, 0);
2208 	return (0);
2209 }
2210 
2211 int
2212 unregister_netdevice_notifier(struct notifier_block *nb)
2213 {
2214 
2215 	EVENTHANDLER_DEREGISTER(ifnet_link_event,
2216 	    nb->tags[NETDEV_UP]);
2217 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event,
2218 	    nb->tags[NETDEV_REGISTER]);
2219 	EVENTHANDLER_DEREGISTER(ifnet_departure_event,
2220 	    nb->tags[NETDEV_UNREGISTER]);
2221 	EVENTHANDLER_DEREGISTER(iflladdr_event,
2222 	    nb->tags[NETDEV_CHANGEADDR]);
2223 
2224 	return (0);
2225 }
2226 
2227 int
2228 unregister_inetaddr_notifier(struct notifier_block *nb)
2229 {
2230 
2231 	EVENTHANDLER_DEREGISTER(ifaddr_event,
2232 	    nb->tags[NETDEV_CHANGEIFADDR]);
2233 
2234 	return (0);
2235 }
2236 
2237 struct list_sort_thunk {
2238 	int (*cmp)(void *, struct list_head *, struct list_head *);
2239 	void *priv;
2240 };
2241 
2242 static inline int
2243 linux_le_cmp(void *priv, const void *d1, const void *d2)
2244 {
2245 	struct list_head *le1, *le2;
2246 	struct list_sort_thunk *thunk;
2247 
2248 	thunk = priv;
2249 	le1 = *(__DECONST(struct list_head **, d1));
2250 	le2 = *(__DECONST(struct list_head **, d2));
2251 	return ((thunk->cmp)(thunk->priv, le1, le2));
2252 }
2253 
2254 void
2255 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv,
2256     struct list_head *a, struct list_head *b))
2257 {
2258 	struct list_sort_thunk thunk;
2259 	struct list_head **ar, *le;
2260 	size_t count, i;
2261 
2262 	count = 0;
2263 	list_for_each(le, head)
2264 		count++;
2265 	ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK);
2266 	i = 0;
2267 	list_for_each(le, head)
2268 		ar[i++] = le;
2269 	thunk.cmp = cmp;
2270 	thunk.priv = priv;
2271 	qsort_r(ar, count, sizeof(struct list_head *), &thunk, linux_le_cmp);
2272 	INIT_LIST_HEAD(head);
2273 	for (i = 0; i < count; i++)
2274 		list_add_tail(ar[i], head);
2275 	free(ar, M_KMALLOC);
2276 }
2277 
2278 void
2279 linux_irq_handler(void *ent)
2280 {
2281 	struct irq_ent *irqe;
2282 
2283 	linux_set_current(curthread);
2284 
2285 	irqe = ent;
2286 	irqe->handler(irqe->irq, irqe->arg);
2287 }
2288 
2289 #if defined(__i386__) || defined(__amd64__)
2290 int
2291 linux_wbinvd_on_all_cpus(void)
2292 {
2293 
2294 	pmap_invalidate_cache();
2295 	return (0);
2296 }
2297 #endif
2298 
2299 int
2300 linux_on_each_cpu(void callback(void *), void *data)
2301 {
2302 
2303 	smp_rendezvous(smp_no_rendezvous_barrier, callback,
2304 	    smp_no_rendezvous_barrier, data);
2305 	return (0);
2306 }
2307 
2308 int
2309 linux_in_atomic(void)
2310 {
2311 
2312 	return ((curthread->td_pflags & TDP_NOFAULTING) != 0);
2313 }
2314 
2315 struct linux_cdev *
2316 linux_find_cdev(const char *name, unsigned major, unsigned minor)
2317 {
2318 	dev_t dev = MKDEV(major, minor);
2319 	struct cdev *cdev;
2320 
2321 	dev_lock();
2322 	LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) {
2323 		struct linux_cdev *ldev = cdev->si_drv1;
2324 		if (ldev->dev == dev &&
2325 		    strcmp(kobject_name(&ldev->kobj), name) == 0) {
2326 			break;
2327 		}
2328 	}
2329 	dev_unlock();
2330 
2331 	return (cdev != NULL ? cdev->si_drv1 : NULL);
2332 }
2333 
2334 int
2335 __register_chrdev(unsigned int major, unsigned int baseminor,
2336     unsigned int count, const char *name,
2337     const struct file_operations *fops)
2338 {
2339 	struct linux_cdev *cdev;
2340 	int ret = 0;
2341 	int i;
2342 
2343 	for (i = baseminor; i < baseminor + count; i++) {
2344 		cdev = cdev_alloc();
2345 		cdev->ops = fops;
2346 		kobject_set_name(&cdev->kobj, name);
2347 
2348 		ret = cdev_add(cdev, makedev(major, i), 1);
2349 		if (ret != 0)
2350 			break;
2351 	}
2352 	return (ret);
2353 }
2354 
2355 int
2356 __register_chrdev_p(unsigned int major, unsigned int baseminor,
2357     unsigned int count, const char *name,
2358     const struct file_operations *fops, uid_t uid,
2359     gid_t gid, int mode)
2360 {
2361 	struct linux_cdev *cdev;
2362 	int ret = 0;
2363 	int i;
2364 
2365 	for (i = baseminor; i < baseminor + count; i++) {
2366 		cdev = cdev_alloc();
2367 		cdev->ops = fops;
2368 		kobject_set_name(&cdev->kobj, name);
2369 
2370 		ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode);
2371 		if (ret != 0)
2372 			break;
2373 	}
2374 	return (ret);
2375 }
2376 
2377 void
2378 __unregister_chrdev(unsigned int major, unsigned int baseminor,
2379     unsigned int count, const char *name)
2380 {
2381 	struct linux_cdev *cdevp;
2382 	int i;
2383 
2384 	for (i = baseminor; i < baseminor + count; i++) {
2385 		cdevp = linux_find_cdev(name, major, i);
2386 		if (cdevp != NULL)
2387 			cdev_del(cdevp);
2388 	}
2389 }
2390 
2391 void
2392 linux_dump_stack(void)
2393 {
2394 #ifdef STACK
2395 	struct stack st;
2396 
2397 	stack_zero(&st);
2398 	stack_save(&st);
2399 	stack_print(&st);
2400 #endif
2401 }
2402 
2403 #if defined(__i386__) || defined(__amd64__)
2404 bool linux_cpu_has_clflush;
2405 #endif
2406 
2407 static void
2408 linux_compat_init(void *arg)
2409 {
2410 	struct sysctl_oid *rootoid;
2411 	int i;
2412 
2413 #if defined(__i386__) || defined(__amd64__)
2414 	linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH);
2415 #endif
2416 	rw_init(&linux_vma_lock, "lkpi-vma-lock");
2417 
2418 	rootoid = SYSCTL_ADD_ROOT_NODE(NULL,
2419 	    OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys");
2420 	kobject_init(&linux_class_root, &linux_class_ktype);
2421 	kobject_set_name(&linux_class_root, "class");
2422 	linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid),
2423 	    OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class");
2424 	kobject_init(&linux_root_device.kobj, &linux_dev_ktype);
2425 	kobject_set_name(&linux_root_device.kobj, "device");
2426 	linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL,
2427 	    SYSCTL_CHILDREN(rootoid), OID_AUTO, "device", CTLFLAG_RD, NULL,
2428 	    "device");
2429 	linux_root_device.bsddev = root_bus;
2430 	linux_class_misc.name = "misc";
2431 	class_register(&linux_class_misc);
2432 	INIT_LIST_HEAD(&pci_drivers);
2433 	INIT_LIST_HEAD(&pci_devices);
2434 	spin_lock_init(&pci_lock);
2435 	mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF);
2436 	for (i = 0; i < VMMAP_HASH_SIZE; i++)
2437 		LIST_INIT(&vmmaphead[i]);
2438 }
2439 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL);
2440 
2441 static void
2442 linux_compat_uninit(void *arg)
2443 {
2444 	linux_kobject_kfree_name(&linux_class_root);
2445 	linux_kobject_kfree_name(&linux_root_device.kobj);
2446 	linux_kobject_kfree_name(&linux_class_misc.kobj);
2447 
2448 	mtx_destroy(&vmmaplock);
2449 	spin_lock_destroy(&pci_lock);
2450 	rw_destroy(&linux_vma_lock);
2451 }
2452 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL);
2453 
2454 /*
2455  * NOTE: Linux frequently uses "unsigned long" for pointer to integer
2456  * conversion and vice versa, where in FreeBSD "uintptr_t" would be
2457  * used. Assert these types have the same size, else some parts of the
2458  * LinuxKPI may not work like expected:
2459  */
2460 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t));
2461