xref: /freebsd/sys/compat/linuxkpi/common/src/linux_compat.c (revision 332a6001389f2b7e547fe24f02568a32f993bfa2)
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/time.h>
55 #include <sys/user.h>
56 
57 #include <vm/vm.h>
58 #include <vm/pmap.h>
59 #include <vm/vm_object.h>
60 #include <vm/vm_page.h>
61 #include <vm/vm_pager.h>
62 
63 #include <machine/stdarg.h>
64 
65 #if defined(__i386__) || defined(__amd64__)
66 #include <machine/md_var.h>
67 #endif
68 
69 #include <linux/kobject.h>
70 #include <linux/device.h>
71 #include <linux/slab.h>
72 #include <linux/module.h>
73 #include <linux/moduleparam.h>
74 #include <linux/cdev.h>
75 #include <linux/file.h>
76 #include <linux/sysfs.h>
77 #include <linux/mm.h>
78 #include <linux/io.h>
79 #include <linux/vmalloc.h>
80 #include <linux/netdevice.h>
81 #include <linux/timer.h>
82 #include <linux/interrupt.h>
83 #include <linux/uaccess.h>
84 #include <linux/list.h>
85 #include <linux/kthread.h>
86 #include <linux/kernel.h>
87 #include <linux/compat.h>
88 #include <linux/poll.h>
89 #include <linux/smp.h>
90 #include <linux/wait_bit.h>
91 
92 #if defined(__i386__) || defined(__amd64__)
93 #include <asm/smp.h>
94 #endif
95 
96 SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
97     "LinuxKPI parameters");
98 
99 int linuxkpi_debug;
100 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN,
101     &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable.");
102 
103 static struct timeval lkpi_net_lastlog;
104 static int lkpi_net_curpps;
105 static int lkpi_net_maxpps = 99;
106 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, net_ratelimit, CTLFLAG_RWTUN,
107     &lkpi_net_maxpps, 0, "Limit number of LinuxKPI net messages per second.");
108 
109 MALLOC_DEFINE(M_KMALLOC, "linux", "Linux kmalloc compat");
110 
111 #include <linux/rbtree.h>
112 /* Undo Linux compat changes. */
113 #undef RB_ROOT
114 #undef file
115 #undef cdev
116 #define	RB_ROOT(head)	(head)->rbh_root
117 
118 static void linux_cdev_deref(struct linux_cdev *ldev);
119 static struct vm_area_struct *linux_cdev_handle_find(void *handle);
120 
121 struct kobject linux_class_root;
122 struct device linux_root_device;
123 struct class linux_class_misc;
124 struct list_head pci_drivers;
125 struct list_head pci_devices;
126 spinlock_t pci_lock;
127 
128 unsigned long linux_timer_hz_mask;
129 
130 wait_queue_head_t linux_bit_waitq;
131 wait_queue_head_t linux_var_waitq;
132 
133 int
134 panic_cmp(struct rb_node *one, struct rb_node *two)
135 {
136 	panic("no cmp");
137 }
138 
139 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp);
140 
141 int
142 kobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list args)
143 {
144 	va_list tmp_va;
145 	int len;
146 	char *old;
147 	char *name;
148 	char dummy;
149 
150 	old = kobj->name;
151 
152 	if (old && fmt == NULL)
153 		return (0);
154 
155 	/* compute length of string */
156 	va_copy(tmp_va, args);
157 	len = vsnprintf(&dummy, 0, fmt, tmp_va);
158 	va_end(tmp_va);
159 
160 	/* account for zero termination */
161 	len++;
162 
163 	/* check for error */
164 	if (len < 1)
165 		return (-EINVAL);
166 
167 	/* allocate memory for string */
168 	name = kzalloc(len, GFP_KERNEL);
169 	if (name == NULL)
170 		return (-ENOMEM);
171 	vsnprintf(name, len, fmt, args);
172 	kobj->name = name;
173 
174 	/* free old string */
175 	kfree(old);
176 
177 	/* filter new string */
178 	for (; *name != '\0'; name++)
179 		if (*name == '/')
180 			*name = '!';
181 	return (0);
182 }
183 
184 int
185 kobject_set_name(struct kobject *kobj, const char *fmt, ...)
186 {
187 	va_list args;
188 	int error;
189 
190 	va_start(args, fmt);
191 	error = kobject_set_name_vargs(kobj, fmt, args);
192 	va_end(args);
193 
194 	return (error);
195 }
196 
197 static int
198 kobject_add_complete(struct kobject *kobj, struct kobject *parent)
199 {
200 	const struct kobj_type *t;
201 	int error;
202 
203 	kobj->parent = parent;
204 	error = sysfs_create_dir(kobj);
205 	if (error == 0 && kobj->ktype && kobj->ktype->default_attrs) {
206 		struct attribute **attr;
207 		t = kobj->ktype;
208 
209 		for (attr = t->default_attrs; *attr != NULL; attr++) {
210 			error = sysfs_create_file(kobj, *attr);
211 			if (error)
212 				break;
213 		}
214 		if (error)
215 			sysfs_remove_dir(kobj);
216 	}
217 	return (error);
218 }
219 
220 int
221 kobject_add(struct kobject *kobj, struct kobject *parent, const char *fmt, ...)
222 {
223 	va_list args;
224 	int error;
225 
226 	va_start(args, fmt);
227 	error = kobject_set_name_vargs(kobj, fmt, args);
228 	va_end(args);
229 	if (error)
230 		return (error);
231 
232 	return kobject_add_complete(kobj, parent);
233 }
234 
235 void
236 linux_kobject_release(struct kref *kref)
237 {
238 	struct kobject *kobj;
239 	char *name;
240 
241 	kobj = container_of(kref, struct kobject, kref);
242 	sysfs_remove_dir(kobj);
243 	name = kobj->name;
244 	if (kobj->ktype && kobj->ktype->release)
245 		kobj->ktype->release(kobj);
246 	kfree(name);
247 }
248 
249 static void
250 linux_kobject_kfree(struct kobject *kobj)
251 {
252 	kfree(kobj);
253 }
254 
255 static void
256 linux_kobject_kfree_name(struct kobject *kobj)
257 {
258 	if (kobj) {
259 		kfree(kobj->name);
260 	}
261 }
262 
263 const struct kobj_type linux_kfree_type = {
264 	.release = linux_kobject_kfree
265 };
266 
267 static void
268 linux_device_release(struct device *dev)
269 {
270 	pr_debug("linux_device_release: %s\n", dev_name(dev));
271 	kfree(dev);
272 }
273 
274 static ssize_t
275 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf)
276 {
277 	struct class_attribute *dattr;
278 	ssize_t error;
279 
280 	dattr = container_of(attr, struct class_attribute, attr);
281 	error = -EIO;
282 	if (dattr->show)
283 		error = dattr->show(container_of(kobj, struct class, kobj),
284 		    dattr, buf);
285 	return (error);
286 }
287 
288 static ssize_t
289 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf,
290     size_t count)
291 {
292 	struct class_attribute *dattr;
293 	ssize_t error;
294 
295 	dattr = container_of(attr, struct class_attribute, attr);
296 	error = -EIO;
297 	if (dattr->store)
298 		error = dattr->store(container_of(kobj, struct class, kobj),
299 		    dattr, buf, count);
300 	return (error);
301 }
302 
303 static void
304 linux_class_release(struct kobject *kobj)
305 {
306 	struct class *class;
307 
308 	class = container_of(kobj, struct class, kobj);
309 	if (class->class_release)
310 		class->class_release(class);
311 }
312 
313 static const struct sysfs_ops linux_class_sysfs = {
314 	.show  = linux_class_show,
315 	.store = linux_class_store,
316 };
317 
318 const struct kobj_type linux_class_ktype = {
319 	.release = linux_class_release,
320 	.sysfs_ops = &linux_class_sysfs
321 };
322 
323 static void
324 linux_dev_release(struct kobject *kobj)
325 {
326 	struct device *dev;
327 
328 	dev = container_of(kobj, struct device, kobj);
329 	/* This is the precedence defined by linux. */
330 	if (dev->release)
331 		dev->release(dev);
332 	else if (dev->class && dev->class->dev_release)
333 		dev->class->dev_release(dev);
334 }
335 
336 static ssize_t
337 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf)
338 {
339 	struct device_attribute *dattr;
340 	ssize_t error;
341 
342 	dattr = container_of(attr, struct device_attribute, attr);
343 	error = -EIO;
344 	if (dattr->show)
345 		error = dattr->show(container_of(kobj, struct device, kobj),
346 		    dattr, buf);
347 	return (error);
348 }
349 
350 static ssize_t
351 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf,
352     size_t count)
353 {
354 	struct device_attribute *dattr;
355 	ssize_t error;
356 
357 	dattr = container_of(attr, struct device_attribute, attr);
358 	error = -EIO;
359 	if (dattr->store)
360 		error = dattr->store(container_of(kobj, struct device, kobj),
361 		    dattr, buf, count);
362 	return (error);
363 }
364 
365 static const struct sysfs_ops linux_dev_sysfs = {
366 	.show  = linux_dev_show,
367 	.store = linux_dev_store,
368 };
369 
370 const struct kobj_type linux_dev_ktype = {
371 	.release = linux_dev_release,
372 	.sysfs_ops = &linux_dev_sysfs
373 };
374 
375 struct device *
376 device_create(struct class *class, struct device *parent, dev_t devt,
377     void *drvdata, const char *fmt, ...)
378 {
379 	struct device *dev;
380 	va_list args;
381 
382 	dev = kzalloc(sizeof(*dev), M_WAITOK);
383 	dev->parent = parent;
384 	dev->class = class;
385 	dev->devt = devt;
386 	dev->driver_data = drvdata;
387 	dev->release = linux_device_release;
388 	va_start(args, fmt);
389 	kobject_set_name_vargs(&dev->kobj, fmt, args);
390 	va_end(args);
391 	device_register(dev);
392 
393 	return (dev);
394 }
395 
396 int
397 kobject_init_and_add(struct kobject *kobj, const struct kobj_type *ktype,
398     struct kobject *parent, const char *fmt, ...)
399 {
400 	va_list args;
401 	int error;
402 
403 	kobject_init(kobj, ktype);
404 	kobj->ktype = ktype;
405 	kobj->parent = parent;
406 	kobj->name = NULL;
407 
408 	va_start(args, fmt);
409 	error = kobject_set_name_vargs(kobj, fmt, args);
410 	va_end(args);
411 	if (error)
412 		return (error);
413 	return kobject_add_complete(kobj, parent);
414 }
415 
416 static void
417 linux_kq_lock(void *arg)
418 {
419 	spinlock_t *s = arg;
420 
421 	spin_lock(s);
422 }
423 static void
424 linux_kq_unlock(void *arg)
425 {
426 	spinlock_t *s = arg;
427 
428 	spin_unlock(s);
429 }
430 
431 static void
432 linux_kq_assert_lock(void *arg, int what)
433 {
434 #ifdef INVARIANTS
435 	spinlock_t *s = arg;
436 
437 	if (what == LA_LOCKED)
438 		mtx_assert(&s->m, MA_OWNED);
439 	else
440 		mtx_assert(&s->m, MA_NOTOWNED);
441 #endif
442 }
443 
444 static void
445 linux_file_kqfilter_poll(struct linux_file *, int);
446 
447 struct linux_file *
448 linux_file_alloc(void)
449 {
450 	struct linux_file *filp;
451 
452 	filp = kzalloc(sizeof(*filp), GFP_KERNEL);
453 
454 	/* set initial refcount */
455 	filp->f_count = 1;
456 
457 	/* setup fields needed by kqueue support */
458 	spin_lock_init(&filp->f_kqlock);
459 	knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock,
460 	    linux_kq_lock, linux_kq_unlock, linux_kq_assert_lock);
461 
462 	return (filp);
463 }
464 
465 void
466 linux_file_free(struct linux_file *filp)
467 {
468 	if (filp->_file == NULL) {
469 		if (filp->f_shmem != NULL)
470 			vm_object_deallocate(filp->f_shmem);
471 		kfree(filp);
472 	} else {
473 		/*
474 		 * The close method of the character device or file
475 		 * will free the linux_file structure:
476 		 */
477 		_fdrop(filp->_file, curthread);
478 	}
479 }
480 
481 static int
482 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
483     vm_page_t *mres)
484 {
485 	struct vm_area_struct *vmap;
486 
487 	vmap = linux_cdev_handle_find(vm_obj->handle);
488 
489 	MPASS(vmap != NULL);
490 	MPASS(vmap->vm_private_data == vm_obj->handle);
491 
492 	if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) {
493 		vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset;
494 		vm_page_t page;
495 
496 		if (((*mres)->flags & PG_FICTITIOUS) != 0) {
497 			/*
498 			 * If the passed in result page is a fake
499 			 * page, update it with the new physical
500 			 * address.
501 			 */
502 			page = *mres;
503 			vm_page_updatefake(page, paddr, vm_obj->memattr);
504 		} else {
505 			/*
506 			 * Replace the passed in "mres" page with our
507 			 * own fake page and free up the all of the
508 			 * original pages.
509 			 */
510 			VM_OBJECT_WUNLOCK(vm_obj);
511 			page = vm_page_getfake(paddr, vm_obj->memattr);
512 			VM_OBJECT_WLOCK(vm_obj);
513 
514 			vm_page_replace(page, vm_obj, (*mres)->pindex, *mres);
515 			*mres = page;
516 		}
517 		vm_page_valid(page);
518 		return (VM_PAGER_OK);
519 	}
520 	return (VM_PAGER_FAIL);
521 }
522 
523 static int
524 linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type,
525     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
526 {
527 	struct vm_area_struct *vmap;
528 	int err;
529 
530 	/* get VM area structure */
531 	vmap = linux_cdev_handle_find(vm_obj->handle);
532 	MPASS(vmap != NULL);
533 	MPASS(vmap->vm_private_data == vm_obj->handle);
534 
535 	VM_OBJECT_WUNLOCK(vm_obj);
536 
537 	linux_set_current(curthread);
538 
539 	down_write(&vmap->vm_mm->mmap_sem);
540 	if (unlikely(vmap->vm_ops == NULL)) {
541 		err = VM_FAULT_SIGBUS;
542 	} else {
543 		struct vm_fault vmf;
544 
545 		/* fill out VM fault structure */
546 		vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx);
547 		vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
548 		vmf.pgoff = 0;
549 		vmf.page = NULL;
550 		vmf.vma = vmap;
551 
552 		vmap->vm_pfn_count = 0;
553 		vmap->vm_pfn_pcount = &vmap->vm_pfn_count;
554 		vmap->vm_obj = vm_obj;
555 
556 		err = vmap->vm_ops->fault(vmap, &vmf);
557 
558 		while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) {
559 			kern_yield(PRI_USER);
560 			err = vmap->vm_ops->fault(vmap, &vmf);
561 		}
562 	}
563 
564 	/* translate return code */
565 	switch (err) {
566 	case VM_FAULT_OOM:
567 		err = VM_PAGER_AGAIN;
568 		break;
569 	case VM_FAULT_SIGBUS:
570 		err = VM_PAGER_BAD;
571 		break;
572 	case VM_FAULT_NOPAGE:
573 		/*
574 		 * By contract the fault handler will return having
575 		 * busied all the pages itself. If pidx is already
576 		 * found in the object, it will simply xbusy the first
577 		 * page and return with vm_pfn_count set to 1.
578 		 */
579 		*first = vmap->vm_pfn_first;
580 		*last = *first + vmap->vm_pfn_count - 1;
581 		err = VM_PAGER_OK;
582 		break;
583 	default:
584 		err = VM_PAGER_ERROR;
585 		break;
586 	}
587 	up_write(&vmap->vm_mm->mmap_sem);
588 	VM_OBJECT_WLOCK(vm_obj);
589 	return (err);
590 }
591 
592 static struct rwlock linux_vma_lock;
593 static TAILQ_HEAD(, vm_area_struct) linux_vma_head =
594     TAILQ_HEAD_INITIALIZER(linux_vma_head);
595 
596 static void
597 linux_cdev_handle_free(struct vm_area_struct *vmap)
598 {
599 	/* Drop reference on vm_file */
600 	if (vmap->vm_file != NULL)
601 		fput(vmap->vm_file);
602 
603 	/* Drop reference on mm_struct */
604 	mmput(vmap->vm_mm);
605 
606 	kfree(vmap);
607 }
608 
609 static void
610 linux_cdev_handle_remove(struct vm_area_struct *vmap)
611 {
612 	rw_wlock(&linux_vma_lock);
613 	TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry);
614 	rw_wunlock(&linux_vma_lock);
615 }
616 
617 static struct vm_area_struct *
618 linux_cdev_handle_find(void *handle)
619 {
620 	struct vm_area_struct *vmap;
621 
622 	rw_rlock(&linux_vma_lock);
623 	TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) {
624 		if (vmap->vm_private_data == handle)
625 			break;
626 	}
627 	rw_runlock(&linux_vma_lock);
628 	return (vmap);
629 }
630 
631 static int
632 linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
633 		      vm_ooffset_t foff, struct ucred *cred, u_short *color)
634 {
635 
636 	MPASS(linux_cdev_handle_find(handle) != NULL);
637 	*color = 0;
638 	return (0);
639 }
640 
641 static void
642 linux_cdev_pager_dtor(void *handle)
643 {
644 	const struct vm_operations_struct *vm_ops;
645 	struct vm_area_struct *vmap;
646 
647 	vmap = linux_cdev_handle_find(handle);
648 	MPASS(vmap != NULL);
649 
650 	/*
651 	 * Remove handle before calling close operation to prevent
652 	 * other threads from reusing the handle pointer.
653 	 */
654 	linux_cdev_handle_remove(vmap);
655 
656 	down_write(&vmap->vm_mm->mmap_sem);
657 	vm_ops = vmap->vm_ops;
658 	if (likely(vm_ops != NULL))
659 		vm_ops->close(vmap);
660 	up_write(&vmap->vm_mm->mmap_sem);
661 
662 	linux_cdev_handle_free(vmap);
663 }
664 
665 static struct cdev_pager_ops linux_cdev_pager_ops[2] = {
666   {
667 	/* OBJT_MGTDEVICE */
668 	.cdev_pg_populate	= linux_cdev_pager_populate,
669 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
670 	.cdev_pg_dtor	= linux_cdev_pager_dtor
671   },
672   {
673 	/* OBJT_DEVICE */
674 	.cdev_pg_fault	= linux_cdev_pager_fault,
675 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
676 	.cdev_pg_dtor	= linux_cdev_pager_dtor
677   },
678 };
679 
680 int
681 zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
682     unsigned long size)
683 {
684 	vm_object_t obj;
685 	vm_page_t m;
686 
687 	obj = vma->vm_obj;
688 	if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0)
689 		return (-ENOTSUP);
690 	VM_OBJECT_RLOCK(obj);
691 	for (m = vm_page_find_least(obj, OFF_TO_IDX(address));
692 	    m != NULL && m->pindex < OFF_TO_IDX(address + size);
693 	    m = TAILQ_NEXT(m, listq))
694 		pmap_remove_all(m);
695 	VM_OBJECT_RUNLOCK(obj);
696 	return (0);
697 }
698 
699 static struct file_operations dummy_ldev_ops = {
700 	/* XXXKIB */
701 };
702 
703 static struct linux_cdev dummy_ldev = {
704 	.ops = &dummy_ldev_ops,
705 };
706 
707 #define	LDEV_SI_DTR	0x0001
708 #define	LDEV_SI_REF	0x0002
709 
710 static void
711 linux_get_fop(struct linux_file *filp, const struct file_operations **fop,
712     struct linux_cdev **dev)
713 {
714 	struct linux_cdev *ldev;
715 	u_int siref;
716 
717 	ldev = filp->f_cdev;
718 	*fop = filp->f_op;
719 	if (ldev != NULL) {
720 		for (siref = ldev->siref;;) {
721 			if ((siref & LDEV_SI_DTR) != 0) {
722 				ldev = &dummy_ldev;
723 				siref = ldev->siref;
724 				*fop = ldev->ops;
725 				MPASS((ldev->siref & LDEV_SI_DTR) == 0);
726 			} else if (atomic_fcmpset_int(&ldev->siref, &siref,
727 			    siref + LDEV_SI_REF)) {
728 				break;
729 			}
730 		}
731 	}
732 	*dev = ldev;
733 }
734 
735 static void
736 linux_drop_fop(struct linux_cdev *ldev)
737 {
738 
739 	if (ldev == NULL)
740 		return;
741 	MPASS((ldev->siref & ~LDEV_SI_DTR) != 0);
742 	atomic_subtract_int(&ldev->siref, LDEV_SI_REF);
743 }
744 
745 #define	OPW(fp,td,code) ({			\
746 	struct file *__fpop;			\
747 	__typeof(code) __retval;		\
748 						\
749 	__fpop = (td)->td_fpop;			\
750 	(td)->td_fpop = (fp);			\
751 	__retval = (code);			\
752 	(td)->td_fpop = __fpop;			\
753 	__retval;				\
754 })
755 
756 static int
757 linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td,
758     struct file *file)
759 {
760 	struct linux_cdev *ldev;
761 	struct linux_file *filp;
762 	const struct file_operations *fop;
763 	int error;
764 
765 	ldev = dev->si_drv1;
766 
767 	filp = linux_file_alloc();
768 	filp->f_dentry = &filp->f_dentry_store;
769 	filp->f_op = ldev->ops;
770 	filp->f_mode = file->f_flag;
771 	filp->f_flags = file->f_flag;
772 	filp->f_vnode = file->f_vnode;
773 	filp->_file = file;
774 	refcount_acquire(&ldev->refs);
775 	filp->f_cdev = ldev;
776 
777 	linux_set_current(td);
778 	linux_get_fop(filp, &fop, &ldev);
779 
780 	if (fop->open != NULL) {
781 		error = -fop->open(file->f_vnode, filp);
782 		if (error != 0) {
783 			linux_drop_fop(ldev);
784 			linux_cdev_deref(filp->f_cdev);
785 			kfree(filp);
786 			return (error);
787 		}
788 	}
789 
790 	/* hold on to the vnode - used for fstat() */
791 	vhold(filp->f_vnode);
792 
793 	/* release the file from devfs */
794 	finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops);
795 	linux_drop_fop(ldev);
796 	return (ENXIO);
797 }
798 
799 #define	LINUX_IOCTL_MIN_PTR 0x10000UL
800 #define	LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX)
801 
802 static inline int
803 linux_remap_address(void **uaddr, size_t len)
804 {
805 	uintptr_t uaddr_val = (uintptr_t)(*uaddr);
806 
807 	if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR &&
808 	    uaddr_val < LINUX_IOCTL_MAX_PTR)) {
809 		struct task_struct *pts = current;
810 		if (pts == NULL) {
811 			*uaddr = NULL;
812 			return (1);
813 		}
814 
815 		/* compute data offset */
816 		uaddr_val -= LINUX_IOCTL_MIN_PTR;
817 
818 		/* check that length is within bounds */
819 		if ((len > IOCPARM_MAX) ||
820 		    (uaddr_val + len) > pts->bsd_ioctl_len) {
821 			*uaddr = NULL;
822 			return (1);
823 		}
824 
825 		/* re-add kernel buffer address */
826 		uaddr_val += (uintptr_t)pts->bsd_ioctl_data;
827 
828 		/* update address location */
829 		*uaddr = (void *)uaddr_val;
830 		return (1);
831 	}
832 	return (0);
833 }
834 
835 int
836 linux_copyin(const void *uaddr, void *kaddr, size_t len)
837 {
838 	if (linux_remap_address(__DECONST(void **, &uaddr), len)) {
839 		if (uaddr == NULL)
840 			return (-EFAULT);
841 		memcpy(kaddr, uaddr, len);
842 		return (0);
843 	}
844 	return (-copyin(uaddr, kaddr, len));
845 }
846 
847 int
848 linux_copyout(const void *kaddr, void *uaddr, size_t len)
849 {
850 	if (linux_remap_address(&uaddr, len)) {
851 		if (uaddr == NULL)
852 			return (-EFAULT);
853 		memcpy(uaddr, kaddr, len);
854 		return (0);
855 	}
856 	return (-copyout(kaddr, uaddr, len));
857 }
858 
859 size_t
860 linux_clear_user(void *_uaddr, size_t _len)
861 {
862 	uint8_t *uaddr = _uaddr;
863 	size_t len = _len;
864 
865 	/* make sure uaddr is aligned before going into the fast loop */
866 	while (((uintptr_t)uaddr & 7) != 0 && len > 7) {
867 		if (subyte(uaddr, 0))
868 			return (_len);
869 		uaddr++;
870 		len--;
871 	}
872 
873 	/* zero 8 bytes at a time */
874 	while (len > 7) {
875 #ifdef __LP64__
876 		if (suword64(uaddr, 0))
877 			return (_len);
878 #else
879 		if (suword32(uaddr, 0))
880 			return (_len);
881 		if (suword32(uaddr + 4, 0))
882 			return (_len);
883 #endif
884 		uaddr += 8;
885 		len -= 8;
886 	}
887 
888 	/* zero fill end, if any */
889 	while (len > 0) {
890 		if (subyte(uaddr, 0))
891 			return (_len);
892 		uaddr++;
893 		len--;
894 	}
895 	return (0);
896 }
897 
898 int
899 linux_access_ok(const void *uaddr, size_t len)
900 {
901 	uintptr_t saddr;
902 	uintptr_t eaddr;
903 
904 	/* get start and end address */
905 	saddr = (uintptr_t)uaddr;
906 	eaddr = (uintptr_t)uaddr + len;
907 
908 	/* verify addresses are valid for userspace */
909 	return ((saddr == eaddr) ||
910 	    (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS));
911 }
912 
913 /*
914  * This function should return either EINTR or ERESTART depending on
915  * the signal type sent to this thread:
916  */
917 static int
918 linux_get_error(struct task_struct *task, int error)
919 {
920 	/* check for signal type interrupt code */
921 	if (error == EINTR || error == ERESTARTSYS || error == ERESTART) {
922 		error = -linux_schedule_get_interrupt_value(task);
923 		if (error == 0)
924 			error = EINTR;
925 	}
926 	return (error);
927 }
928 
929 static int
930 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
931     const struct file_operations *fop, u_long cmd, caddr_t data,
932     struct thread *td)
933 {
934 	struct task_struct *task = current;
935 	unsigned size;
936 	int error;
937 
938 	size = IOCPARM_LEN(cmd);
939 	/* refer to logic in sys_ioctl() */
940 	if (size > 0) {
941 		/*
942 		 * Setup hint for linux_copyin() and linux_copyout().
943 		 *
944 		 * Background: Linux code expects a user-space address
945 		 * while FreeBSD supplies a kernel-space address.
946 		 */
947 		task->bsd_ioctl_data = data;
948 		task->bsd_ioctl_len = size;
949 		data = (void *)LINUX_IOCTL_MIN_PTR;
950 	} else {
951 		/* fetch user-space pointer */
952 		data = *(void **)data;
953 	}
954 #if defined(__amd64__)
955 	if (td->td_proc->p_elf_machine == EM_386) {
956 		/* try the compat IOCTL handler first */
957 		if (fop->compat_ioctl != NULL) {
958 			error = -OPW(fp, td, fop->compat_ioctl(filp,
959 			    cmd, (u_long)data));
960 		} else {
961 			error = ENOTTY;
962 		}
963 
964 		/* fallback to the regular IOCTL handler, if any */
965 		if (error == ENOTTY && fop->unlocked_ioctl != NULL) {
966 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
967 			    cmd, (u_long)data));
968 		}
969 	} else
970 #endif
971 	{
972 		if (fop->unlocked_ioctl != NULL) {
973 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
974 			    cmd, (u_long)data));
975 		} else {
976 			error = ENOTTY;
977 		}
978 	}
979 	if (size > 0) {
980 		task->bsd_ioctl_data = NULL;
981 		task->bsd_ioctl_len = 0;
982 	}
983 
984 	if (error == EWOULDBLOCK) {
985 		/* update kqfilter status, if any */
986 		linux_file_kqfilter_poll(filp,
987 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
988 	} else {
989 		error = linux_get_error(task, error);
990 	}
991 	return (error);
992 }
993 
994 #define	LINUX_POLL_TABLE_NORMAL ((poll_table *)1)
995 
996 /*
997  * This function atomically updates the poll wakeup state and returns
998  * the previous state at the time of update.
999  */
1000 static uint8_t
1001 linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate)
1002 {
1003 	int c, old;
1004 
1005 	c = v->counter;
1006 
1007 	while ((old = atomic_cmpxchg(v, c, pstate[c])) != c)
1008 		c = old;
1009 
1010 	return (c);
1011 }
1012 
1013 static int
1014 linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key)
1015 {
1016 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1017 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */
1018 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1019 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY,
1020 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */
1021 	};
1022 	struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq);
1023 
1024 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1025 	case LINUX_FWQ_STATE_QUEUED:
1026 		linux_poll_wakeup(filp);
1027 		return (1);
1028 	default:
1029 		return (0);
1030 	}
1031 }
1032 
1033 void
1034 linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p)
1035 {
1036 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1037 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY,
1038 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1039 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */
1040 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED,
1041 	};
1042 
1043 	/* check if we are called inside the select system call */
1044 	if (p == LINUX_POLL_TABLE_NORMAL)
1045 		selrecord(curthread, &filp->f_selinfo);
1046 
1047 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1048 	case LINUX_FWQ_STATE_INIT:
1049 		/* NOTE: file handles can only belong to one wait-queue */
1050 		filp->f_wait_queue.wqh = wqh;
1051 		filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback;
1052 		add_wait_queue(wqh, &filp->f_wait_queue.wq);
1053 		atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED);
1054 		break;
1055 	default:
1056 		break;
1057 	}
1058 }
1059 
1060 static void
1061 linux_poll_wait_dequeue(struct linux_file *filp)
1062 {
1063 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1064 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT,	/* NOP */
1065 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT,
1066 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT,
1067 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT,
1068 	};
1069 
1070 	seldrain(&filp->f_selinfo);
1071 
1072 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1073 	case LINUX_FWQ_STATE_NOT_READY:
1074 	case LINUX_FWQ_STATE_QUEUED:
1075 	case LINUX_FWQ_STATE_READY:
1076 		remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq);
1077 		break;
1078 	default:
1079 		break;
1080 	}
1081 }
1082 
1083 void
1084 linux_poll_wakeup(struct linux_file *filp)
1085 {
1086 	/* this function should be NULL-safe */
1087 	if (filp == NULL)
1088 		return;
1089 
1090 	selwakeup(&filp->f_selinfo);
1091 
1092 	spin_lock(&filp->f_kqlock);
1093 	filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ |
1094 	    LINUX_KQ_FLAG_NEED_WRITE;
1095 
1096 	/* make sure the "knote" gets woken up */
1097 	KNOTE_LOCKED(&filp->f_selinfo.si_note, 1);
1098 	spin_unlock(&filp->f_kqlock);
1099 }
1100 
1101 static void
1102 linux_file_kqfilter_detach(struct knote *kn)
1103 {
1104 	struct linux_file *filp = kn->kn_hook;
1105 
1106 	spin_lock(&filp->f_kqlock);
1107 	knlist_remove(&filp->f_selinfo.si_note, kn, 1);
1108 	spin_unlock(&filp->f_kqlock);
1109 }
1110 
1111 static int
1112 linux_file_kqfilter_read_event(struct knote *kn, long hint)
1113 {
1114 	struct linux_file *filp = kn->kn_hook;
1115 
1116 	mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1117 
1118 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0);
1119 }
1120 
1121 static int
1122 linux_file_kqfilter_write_event(struct knote *kn, long hint)
1123 {
1124 	struct linux_file *filp = kn->kn_hook;
1125 
1126 	mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1127 
1128 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0);
1129 }
1130 
1131 static struct filterops linux_dev_kqfiltops_read = {
1132 	.f_isfd = 1,
1133 	.f_detach = linux_file_kqfilter_detach,
1134 	.f_event = linux_file_kqfilter_read_event,
1135 };
1136 
1137 static struct filterops linux_dev_kqfiltops_write = {
1138 	.f_isfd = 1,
1139 	.f_detach = linux_file_kqfilter_detach,
1140 	.f_event = linux_file_kqfilter_write_event,
1141 };
1142 
1143 static void
1144 linux_file_kqfilter_poll(struct linux_file *filp, int kqflags)
1145 {
1146 	struct thread *td;
1147 	const struct file_operations *fop;
1148 	struct linux_cdev *ldev;
1149 	int temp;
1150 
1151 	if ((filp->f_kqflags & kqflags) == 0)
1152 		return;
1153 
1154 	td = curthread;
1155 
1156 	linux_get_fop(filp, &fop, &ldev);
1157 	/* get the latest polling state */
1158 	temp = OPW(filp->_file, td, fop->poll(filp, NULL));
1159 	linux_drop_fop(ldev);
1160 
1161 	spin_lock(&filp->f_kqlock);
1162 	/* clear kqflags */
1163 	filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ |
1164 	    LINUX_KQ_FLAG_NEED_WRITE);
1165 	/* update kqflags */
1166 	if ((temp & (POLLIN | POLLOUT)) != 0) {
1167 		if ((temp & POLLIN) != 0)
1168 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ;
1169 		if ((temp & POLLOUT) != 0)
1170 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE;
1171 
1172 		/* make sure the "knote" gets woken up */
1173 		KNOTE_LOCKED(&filp->f_selinfo.si_note, 0);
1174 	}
1175 	spin_unlock(&filp->f_kqlock);
1176 }
1177 
1178 static int
1179 linux_file_kqfilter(struct file *file, struct knote *kn)
1180 {
1181 	struct linux_file *filp;
1182 	struct thread *td;
1183 	int error;
1184 
1185 	td = curthread;
1186 	filp = (struct linux_file *)file->f_data;
1187 	filp->f_flags = file->f_flag;
1188 	if (filp->f_op->poll == NULL)
1189 		return (EINVAL);
1190 
1191 	spin_lock(&filp->f_kqlock);
1192 	switch (kn->kn_filter) {
1193 	case EVFILT_READ:
1194 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ;
1195 		kn->kn_fop = &linux_dev_kqfiltops_read;
1196 		kn->kn_hook = filp;
1197 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1198 		error = 0;
1199 		break;
1200 	case EVFILT_WRITE:
1201 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE;
1202 		kn->kn_fop = &linux_dev_kqfiltops_write;
1203 		kn->kn_hook = filp;
1204 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1205 		error = 0;
1206 		break;
1207 	default:
1208 		error = EINVAL;
1209 		break;
1210 	}
1211 	spin_unlock(&filp->f_kqlock);
1212 
1213 	if (error == 0) {
1214 		linux_set_current(td);
1215 
1216 		/* update kqfilter status, if any */
1217 		linux_file_kqfilter_poll(filp,
1218 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1219 	}
1220 	return (error);
1221 }
1222 
1223 static int
1224 linux_file_mmap_single(struct file *fp, const struct file_operations *fop,
1225     vm_ooffset_t *offset, vm_size_t size, struct vm_object **object,
1226     int nprot, struct thread *td)
1227 {
1228 	struct task_struct *task;
1229 	struct vm_area_struct *vmap;
1230 	struct mm_struct *mm;
1231 	struct linux_file *filp;
1232 	vm_memattr_t attr;
1233 	int error;
1234 
1235 	filp = (struct linux_file *)fp->f_data;
1236 	filp->f_flags = fp->f_flag;
1237 
1238 	if (fop->mmap == NULL)
1239 		return (EOPNOTSUPP);
1240 
1241 	linux_set_current(td);
1242 
1243 	/*
1244 	 * The same VM object might be shared by multiple processes
1245 	 * and the mm_struct is usually freed when a process exits.
1246 	 *
1247 	 * The atomic reference below makes sure the mm_struct is
1248 	 * available as long as the vmap is in the linux_vma_head.
1249 	 */
1250 	task = current;
1251 	mm = task->mm;
1252 	if (atomic_inc_not_zero(&mm->mm_users) == 0)
1253 		return (EINVAL);
1254 
1255 	vmap = kzalloc(sizeof(*vmap), GFP_KERNEL);
1256 	vmap->vm_start = 0;
1257 	vmap->vm_end = size;
1258 	vmap->vm_pgoff = *offset / PAGE_SIZE;
1259 	vmap->vm_pfn = 0;
1260 	vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL);
1261 	vmap->vm_ops = NULL;
1262 	vmap->vm_file = get_file(filp);
1263 	vmap->vm_mm = mm;
1264 
1265 	if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) {
1266 		error = linux_get_error(task, EINTR);
1267 	} else {
1268 		error = -OPW(fp, td, fop->mmap(filp, vmap));
1269 		error = linux_get_error(task, error);
1270 		up_write(&vmap->vm_mm->mmap_sem);
1271 	}
1272 
1273 	if (error != 0) {
1274 		linux_cdev_handle_free(vmap);
1275 		return (error);
1276 	}
1277 
1278 	attr = pgprot2cachemode(vmap->vm_page_prot);
1279 
1280 	if (vmap->vm_ops != NULL) {
1281 		struct vm_area_struct *ptr;
1282 		void *vm_private_data;
1283 		bool vm_no_fault;
1284 
1285 		if (vmap->vm_ops->open == NULL ||
1286 		    vmap->vm_ops->close == NULL ||
1287 		    vmap->vm_private_data == NULL) {
1288 			/* free allocated VM area struct */
1289 			linux_cdev_handle_free(vmap);
1290 			return (EINVAL);
1291 		}
1292 
1293 		vm_private_data = vmap->vm_private_data;
1294 
1295 		rw_wlock(&linux_vma_lock);
1296 		TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) {
1297 			if (ptr->vm_private_data == vm_private_data)
1298 				break;
1299 		}
1300 		/* check if there is an existing VM area struct */
1301 		if (ptr != NULL) {
1302 			/* check if the VM area structure is invalid */
1303 			if (ptr->vm_ops == NULL ||
1304 			    ptr->vm_ops->open == NULL ||
1305 			    ptr->vm_ops->close == NULL) {
1306 				error = ESTALE;
1307 				vm_no_fault = 1;
1308 			} else {
1309 				error = EEXIST;
1310 				vm_no_fault = (ptr->vm_ops->fault == NULL);
1311 			}
1312 		} else {
1313 			/* insert VM area structure into list */
1314 			TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry);
1315 			error = 0;
1316 			vm_no_fault = (vmap->vm_ops->fault == NULL);
1317 		}
1318 		rw_wunlock(&linux_vma_lock);
1319 
1320 		if (error != 0) {
1321 			/* free allocated VM area struct */
1322 			linux_cdev_handle_free(vmap);
1323 			/* check for stale VM area struct */
1324 			if (error != EEXIST)
1325 				return (error);
1326 		}
1327 
1328 		/* check if there is no fault handler */
1329 		if (vm_no_fault) {
1330 			*object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE,
1331 			    &linux_cdev_pager_ops[1], size, nprot, *offset,
1332 			    td->td_ucred);
1333 		} else {
1334 			*object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE,
1335 			    &linux_cdev_pager_ops[0], size, nprot, *offset,
1336 			    td->td_ucred);
1337 		}
1338 
1339 		/* check if allocating the VM object failed */
1340 		if (*object == NULL) {
1341 			if (error == 0) {
1342 				/* remove VM area struct from list */
1343 				linux_cdev_handle_remove(vmap);
1344 				/* free allocated VM area struct */
1345 				linux_cdev_handle_free(vmap);
1346 			}
1347 			return (EINVAL);
1348 		}
1349 	} else {
1350 		struct sglist *sg;
1351 
1352 		sg = sglist_alloc(1, M_WAITOK);
1353 		sglist_append_phys(sg,
1354 		    (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len);
1355 
1356 		*object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len,
1357 		    nprot, 0, td->td_ucred);
1358 
1359 		linux_cdev_handle_free(vmap);
1360 
1361 		if (*object == NULL) {
1362 			sglist_free(sg);
1363 			return (EINVAL);
1364 		}
1365 	}
1366 
1367 	if (attr != VM_MEMATTR_DEFAULT) {
1368 		VM_OBJECT_WLOCK(*object);
1369 		vm_object_set_memattr(*object, attr);
1370 		VM_OBJECT_WUNLOCK(*object);
1371 	}
1372 	*offset = 0;
1373 	return (0);
1374 }
1375 
1376 struct cdevsw linuxcdevsw = {
1377 	.d_version = D_VERSION,
1378 	.d_fdopen = linux_dev_fdopen,
1379 	.d_name = "lkpidev",
1380 };
1381 
1382 static int
1383 linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
1384     int flags, struct thread *td)
1385 {
1386 	struct linux_file *filp;
1387 	const struct file_operations *fop;
1388 	struct linux_cdev *ldev;
1389 	ssize_t bytes;
1390 	int error;
1391 
1392 	error = 0;
1393 	filp = (struct linux_file *)file->f_data;
1394 	filp->f_flags = file->f_flag;
1395 	/* XXX no support for I/O vectors currently */
1396 	if (uio->uio_iovcnt != 1)
1397 		return (EOPNOTSUPP);
1398 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1399 		return (EINVAL);
1400 	linux_set_current(td);
1401 	linux_get_fop(filp, &fop, &ldev);
1402 	if (fop->read != NULL) {
1403 		bytes = OPW(file, td, fop->read(filp,
1404 		    uio->uio_iov->iov_base,
1405 		    uio->uio_iov->iov_len, &uio->uio_offset));
1406 		if (bytes >= 0) {
1407 			uio->uio_iov->iov_base =
1408 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1409 			uio->uio_iov->iov_len -= bytes;
1410 			uio->uio_resid -= bytes;
1411 		} else {
1412 			error = linux_get_error(current, -bytes);
1413 		}
1414 	} else
1415 		error = ENXIO;
1416 
1417 	/* update kqfilter status, if any */
1418 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ);
1419 	linux_drop_fop(ldev);
1420 
1421 	return (error);
1422 }
1423 
1424 static int
1425 linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred,
1426     int flags, struct thread *td)
1427 {
1428 	struct linux_file *filp;
1429 	const struct file_operations *fop;
1430 	struct linux_cdev *ldev;
1431 	ssize_t bytes;
1432 	int error;
1433 
1434 	filp = (struct linux_file *)file->f_data;
1435 	filp->f_flags = file->f_flag;
1436 	/* XXX no support for I/O vectors currently */
1437 	if (uio->uio_iovcnt != 1)
1438 		return (EOPNOTSUPP);
1439 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1440 		return (EINVAL);
1441 	linux_set_current(td);
1442 	linux_get_fop(filp, &fop, &ldev);
1443 	if (fop->write != NULL) {
1444 		bytes = OPW(file, td, fop->write(filp,
1445 		    uio->uio_iov->iov_base,
1446 		    uio->uio_iov->iov_len, &uio->uio_offset));
1447 		if (bytes >= 0) {
1448 			uio->uio_iov->iov_base =
1449 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1450 			uio->uio_iov->iov_len -= bytes;
1451 			uio->uio_resid -= bytes;
1452 			error = 0;
1453 		} else {
1454 			error = linux_get_error(current, -bytes);
1455 		}
1456 	} else
1457 		error = ENXIO;
1458 
1459 	/* update kqfilter status, if any */
1460 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE);
1461 
1462 	linux_drop_fop(ldev);
1463 
1464 	return (error);
1465 }
1466 
1467 static int
1468 linux_file_poll(struct file *file, int events, struct ucred *active_cred,
1469     struct thread *td)
1470 {
1471 	struct linux_file *filp;
1472 	const struct file_operations *fop;
1473 	struct linux_cdev *ldev;
1474 	int revents;
1475 
1476 	filp = (struct linux_file *)file->f_data;
1477 	filp->f_flags = file->f_flag;
1478 	linux_set_current(td);
1479 	linux_get_fop(filp, &fop, &ldev);
1480 	if (fop->poll != NULL) {
1481 		revents = OPW(file, td, fop->poll(filp,
1482 		    LINUX_POLL_TABLE_NORMAL)) & events;
1483 	} else {
1484 		revents = 0;
1485 	}
1486 	linux_drop_fop(ldev);
1487 	return (revents);
1488 }
1489 
1490 static int
1491 linux_file_close(struct file *file, struct thread *td)
1492 {
1493 	struct linux_file *filp;
1494 	int (*release)(struct inode *, struct linux_file *);
1495 	const struct file_operations *fop;
1496 	struct linux_cdev *ldev;
1497 	int error;
1498 
1499 	filp = (struct linux_file *)file->f_data;
1500 
1501 	KASSERT(file_count(filp) == 0,
1502 	    ("File refcount(%d) is not zero", file_count(filp)));
1503 
1504 	if (td == NULL)
1505 		td = curthread;
1506 
1507 	error = 0;
1508 	filp->f_flags = file->f_flag;
1509 	linux_set_current(td);
1510 	linux_poll_wait_dequeue(filp);
1511 	linux_get_fop(filp, &fop, &ldev);
1512 	/*
1513 	 * Always use the real release function, if any, to avoid
1514 	 * leaking device resources:
1515 	 */
1516 	release = filp->f_op->release;
1517 	if (release != NULL)
1518 		error = -OPW(file, td, release(filp->f_vnode, filp));
1519 	funsetown(&filp->f_sigio);
1520 	if (filp->f_vnode != NULL)
1521 		vdrop(filp->f_vnode);
1522 	linux_drop_fop(ldev);
1523 	if (filp->f_cdev != NULL)
1524 		linux_cdev_deref(filp->f_cdev);
1525 	kfree(filp);
1526 
1527 	return (error);
1528 }
1529 
1530 static int
1531 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
1532     struct thread *td)
1533 {
1534 	struct linux_file *filp;
1535 	const struct file_operations *fop;
1536 	struct linux_cdev *ldev;
1537 	struct fiodgname_arg *fgn;
1538 	const char *p;
1539 	int error, i;
1540 
1541 	error = 0;
1542 	filp = (struct linux_file *)fp->f_data;
1543 	filp->f_flags = fp->f_flag;
1544 	linux_get_fop(filp, &fop, &ldev);
1545 
1546 	linux_set_current(td);
1547 	switch (cmd) {
1548 	case FIONBIO:
1549 		break;
1550 	case FIOASYNC:
1551 		if (fop->fasync == NULL)
1552 			break;
1553 		error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC));
1554 		break;
1555 	case FIOSETOWN:
1556 		error = fsetown(*(int *)data, &filp->f_sigio);
1557 		if (error == 0) {
1558 			if (fop->fasync == NULL)
1559 				break;
1560 			error = -OPW(fp, td, fop->fasync(0, filp,
1561 			    fp->f_flag & FASYNC));
1562 		}
1563 		break;
1564 	case FIOGETOWN:
1565 		*(int *)data = fgetown(&filp->f_sigio);
1566 		break;
1567 	case FIODGNAME:
1568 #ifdef	COMPAT_FREEBSD32
1569 	case FIODGNAME_32:
1570 #endif
1571 		if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) {
1572 			error = ENXIO;
1573 			break;
1574 		}
1575 		fgn = data;
1576 		p = devtoname(filp->f_cdev->cdev);
1577 		i = strlen(p) + 1;
1578 		if (i > fgn->len) {
1579 			error = EINVAL;
1580 			break;
1581 		}
1582 		error = copyout(p, fiodgname_buf_get_ptr(fgn, cmd), i);
1583 		break;
1584 	default:
1585 		error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td);
1586 		break;
1587 	}
1588 	linux_drop_fop(ldev);
1589 	return (error);
1590 }
1591 
1592 static int
1593 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1594     vm_prot_t *maxprotp, int *flagsp, struct file *fp,
1595     vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp)
1596 {
1597 	/*
1598 	 * Character devices do not provide private mappings
1599 	 * of any kind:
1600 	 */
1601 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1602 	    (prot & VM_PROT_WRITE) != 0)
1603 		return (EACCES);
1604 	if ((*flagsp & (MAP_PRIVATE | MAP_COPY)) != 0)
1605 		return (EINVAL);
1606 
1607 	return (linux_file_mmap_single(fp, fop, foff, objsize, objp,
1608 	    (int)prot, td));
1609 }
1610 
1611 static int
1612 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1613     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1614     struct thread *td)
1615 {
1616 	struct linux_file *filp;
1617 	const struct file_operations *fop;
1618 	struct linux_cdev *ldev;
1619 	struct mount *mp;
1620 	struct vnode *vp;
1621 	vm_object_t object;
1622 	vm_prot_t maxprot;
1623 	int error;
1624 
1625 	filp = (struct linux_file *)fp->f_data;
1626 
1627 	vp = filp->f_vnode;
1628 	if (vp == NULL)
1629 		return (EOPNOTSUPP);
1630 
1631 	/*
1632 	 * Ensure that file and memory protections are
1633 	 * compatible.
1634 	 */
1635 	mp = vp->v_mount;
1636 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1637 		maxprot = VM_PROT_NONE;
1638 		if ((prot & VM_PROT_EXECUTE) != 0)
1639 			return (EACCES);
1640 	} else
1641 		maxprot = VM_PROT_EXECUTE;
1642 	if ((fp->f_flag & FREAD) != 0)
1643 		maxprot |= VM_PROT_READ;
1644 	else if ((prot & VM_PROT_READ) != 0)
1645 		return (EACCES);
1646 
1647 	/*
1648 	 * If we are sharing potential changes via MAP_SHARED and we
1649 	 * are trying to get write permission although we opened it
1650 	 * without asking for it, bail out.
1651 	 *
1652 	 * Note that most character devices always share mappings.
1653 	 *
1654 	 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE
1655 	 * requests rather than doing it here.
1656 	 */
1657 	if ((flags & MAP_SHARED) != 0) {
1658 		if ((fp->f_flag & FWRITE) != 0)
1659 			maxprot |= VM_PROT_WRITE;
1660 		else if ((prot & VM_PROT_WRITE) != 0)
1661 			return (EACCES);
1662 	}
1663 	maxprot &= cap_maxprot;
1664 
1665 	linux_get_fop(filp, &fop, &ldev);
1666 	error = linux_file_mmap_sub(td, size, prot, &maxprot, &flags, fp,
1667 	    &foff, fop, &object);
1668 	if (error != 0)
1669 		goto out;
1670 
1671 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1672 	    foff, FALSE, td);
1673 	if (error != 0)
1674 		vm_object_deallocate(object);
1675 out:
1676 	linux_drop_fop(ldev);
1677 	return (error);
1678 }
1679 
1680 static int
1681 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
1682     struct thread *td)
1683 {
1684 	struct linux_file *filp;
1685 	struct vnode *vp;
1686 	int error;
1687 
1688 	filp = (struct linux_file *)fp->f_data;
1689 	if (filp->f_vnode == NULL)
1690 		return (EOPNOTSUPP);
1691 
1692 	vp = filp->f_vnode;
1693 
1694 	vn_lock(vp, LK_SHARED | LK_RETRY);
1695 	error = VOP_STAT(vp, sb, td->td_ucred, NOCRED, td);
1696 	VOP_UNLOCK(vp);
1697 
1698 	return (error);
1699 }
1700 
1701 static int
1702 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1703     struct filedesc *fdp)
1704 {
1705 	struct linux_file *filp;
1706 	struct vnode *vp;
1707 	int error;
1708 
1709 	filp = fp->f_data;
1710 	vp = filp->f_vnode;
1711 	if (vp == NULL) {
1712 		error = 0;
1713 		kif->kf_type = KF_TYPE_DEV;
1714 	} else {
1715 		vref(vp);
1716 		FILEDESC_SUNLOCK(fdp);
1717 		error = vn_fill_kinfo_vnode(vp, kif);
1718 		vrele(vp);
1719 		kif->kf_type = KF_TYPE_VNODE;
1720 		FILEDESC_SLOCK(fdp);
1721 	}
1722 	return (error);
1723 }
1724 
1725 unsigned int
1726 linux_iminor(struct inode *inode)
1727 {
1728 	struct linux_cdev *ldev;
1729 
1730 	if (inode == NULL || inode->v_rdev == NULL ||
1731 	    inode->v_rdev->si_devsw != &linuxcdevsw)
1732 		return (-1U);
1733 	ldev = inode->v_rdev->si_drv1;
1734 	if (ldev == NULL)
1735 		return (-1U);
1736 
1737 	return (minor(ldev->dev));
1738 }
1739 
1740 struct fileops linuxfileops = {
1741 	.fo_read = linux_file_read,
1742 	.fo_write = linux_file_write,
1743 	.fo_truncate = invfo_truncate,
1744 	.fo_kqfilter = linux_file_kqfilter,
1745 	.fo_stat = linux_file_stat,
1746 	.fo_fill_kinfo = linux_file_fill_kinfo,
1747 	.fo_poll = linux_file_poll,
1748 	.fo_close = linux_file_close,
1749 	.fo_ioctl = linux_file_ioctl,
1750 	.fo_mmap = linux_file_mmap,
1751 	.fo_chmod = invfo_chmod,
1752 	.fo_chown = invfo_chown,
1753 	.fo_sendfile = invfo_sendfile,
1754 	.fo_flags = DFLAG_PASSABLE,
1755 };
1756 
1757 /*
1758  * Hash of vmmap addresses.  This is infrequently accessed and does not
1759  * need to be particularly large.  This is done because we must store the
1760  * caller's idea of the map size to properly unmap.
1761  */
1762 struct vmmap {
1763 	LIST_ENTRY(vmmap)	vm_next;
1764 	void 			*vm_addr;
1765 	unsigned long		vm_size;
1766 };
1767 
1768 struct vmmaphd {
1769 	struct vmmap *lh_first;
1770 };
1771 #define	VMMAP_HASH_SIZE	64
1772 #define	VMMAP_HASH_MASK	(VMMAP_HASH_SIZE - 1)
1773 #define	VM_HASH(addr)	((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK
1774 static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE];
1775 static struct mtx vmmaplock;
1776 
1777 static void
1778 vmmap_add(void *addr, unsigned long size)
1779 {
1780 	struct vmmap *vmmap;
1781 
1782 	vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL);
1783 	mtx_lock(&vmmaplock);
1784 	vmmap->vm_size = size;
1785 	vmmap->vm_addr = addr;
1786 	LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next);
1787 	mtx_unlock(&vmmaplock);
1788 }
1789 
1790 static struct vmmap *
1791 vmmap_remove(void *addr)
1792 {
1793 	struct vmmap *vmmap;
1794 
1795 	mtx_lock(&vmmaplock);
1796 	LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next)
1797 		if (vmmap->vm_addr == addr)
1798 			break;
1799 	if (vmmap)
1800 		LIST_REMOVE(vmmap, vm_next);
1801 	mtx_unlock(&vmmaplock);
1802 
1803 	return (vmmap);
1804 }
1805 
1806 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__)
1807 void *
1808 _ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr)
1809 {
1810 	void *addr;
1811 
1812 	addr = pmap_mapdev_attr(phys_addr, size, attr);
1813 	if (addr == NULL)
1814 		return (NULL);
1815 	vmmap_add(addr, size);
1816 
1817 	return (addr);
1818 }
1819 #endif
1820 
1821 void
1822 iounmap(void *addr)
1823 {
1824 	struct vmmap *vmmap;
1825 
1826 	vmmap = vmmap_remove(addr);
1827 	if (vmmap == NULL)
1828 		return;
1829 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__)
1830 	pmap_unmapdev((vm_offset_t)addr, vmmap->vm_size);
1831 #endif
1832 	kfree(vmmap);
1833 }
1834 
1835 void *
1836 vmap(struct page **pages, unsigned int count, unsigned long flags, int prot)
1837 {
1838 	vm_offset_t off;
1839 	size_t size;
1840 
1841 	size = count * PAGE_SIZE;
1842 	off = kva_alloc(size);
1843 	if (off == 0)
1844 		return (NULL);
1845 	vmmap_add((void *)off, size);
1846 	pmap_qenter(off, pages, count);
1847 
1848 	return ((void *)off);
1849 }
1850 
1851 void
1852 vunmap(void *addr)
1853 {
1854 	struct vmmap *vmmap;
1855 
1856 	vmmap = vmmap_remove(addr);
1857 	if (vmmap == NULL)
1858 		return;
1859 	pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE);
1860 	kva_free((vm_offset_t)addr, vmmap->vm_size);
1861 	kfree(vmmap);
1862 }
1863 
1864 static char *
1865 devm_kvasprintf(struct device *dev, gfp_t gfp, const char *fmt, va_list ap)
1866 {
1867 	unsigned int len;
1868 	char *p;
1869 	va_list aq;
1870 
1871 	va_copy(aq, ap);
1872 	len = vsnprintf(NULL, 0, fmt, aq);
1873 	va_end(aq);
1874 
1875 	if (dev != NULL)
1876 		p = devm_kmalloc(dev, len + 1, gfp);
1877 	else
1878 		p = kmalloc(len + 1, gfp);
1879 	if (p != NULL)
1880 		vsnprintf(p, len + 1, fmt, ap);
1881 
1882 	return (p);
1883 }
1884 
1885 char *
1886 kvasprintf(gfp_t gfp, const char *fmt, va_list ap)
1887 {
1888 
1889 	return (devm_kvasprintf(NULL, gfp, fmt, ap));
1890 }
1891 
1892 char *
1893 lkpi_devm_kasprintf(struct device *dev, gfp_t gfp, const char *fmt, ...)
1894 {
1895 	va_list ap;
1896 	char *p;
1897 
1898 	va_start(ap, fmt);
1899 	p = devm_kvasprintf(dev, gfp, fmt, ap);
1900 	va_end(ap);
1901 
1902 	return (p);
1903 }
1904 
1905 char *
1906 kasprintf(gfp_t gfp, const char *fmt, ...)
1907 {
1908 	va_list ap;
1909 	char *p;
1910 
1911 	va_start(ap, fmt);
1912 	p = kvasprintf(gfp, fmt, ap);
1913 	va_end(ap);
1914 
1915 	return (p);
1916 }
1917 
1918 static void
1919 linux_timer_callback_wrapper(void *context)
1920 {
1921 	struct timer_list *timer;
1922 
1923 	timer = context;
1924 
1925 	if (linux_set_current_flags(curthread, M_NOWAIT)) {
1926 		/* try again later */
1927 		callout_reset(&timer->callout, 1,
1928 		    &linux_timer_callback_wrapper, timer);
1929 		return;
1930 	}
1931 
1932 	timer->function(timer->data);
1933 }
1934 
1935 int
1936 mod_timer(struct timer_list *timer, int expires)
1937 {
1938 	int ret;
1939 
1940 	timer->expires = expires;
1941 	ret = callout_reset(&timer->callout,
1942 	    linux_timer_jiffies_until(expires),
1943 	    &linux_timer_callback_wrapper, timer);
1944 
1945 	MPASS(ret == 0 || ret == 1);
1946 
1947 	return (ret == 1);
1948 }
1949 
1950 void
1951 add_timer(struct timer_list *timer)
1952 {
1953 
1954 	callout_reset(&timer->callout,
1955 	    linux_timer_jiffies_until(timer->expires),
1956 	    &linux_timer_callback_wrapper, timer);
1957 }
1958 
1959 void
1960 add_timer_on(struct timer_list *timer, int cpu)
1961 {
1962 
1963 	callout_reset_on(&timer->callout,
1964 	    linux_timer_jiffies_until(timer->expires),
1965 	    &linux_timer_callback_wrapper, timer, cpu);
1966 }
1967 
1968 int
1969 del_timer(struct timer_list *timer)
1970 {
1971 
1972 	if (callout_stop(&(timer)->callout) == -1)
1973 		return (0);
1974 	return (1);
1975 }
1976 
1977 int
1978 del_timer_sync(struct timer_list *timer)
1979 {
1980 
1981 	if (callout_drain(&(timer)->callout) == -1)
1982 		return (0);
1983 	return (1);
1984 }
1985 
1986 /* greatest common divisor, Euclid equation */
1987 static uint64_t
1988 lkpi_gcd_64(uint64_t a, uint64_t b)
1989 {
1990 	uint64_t an;
1991 	uint64_t bn;
1992 
1993 	while (b != 0) {
1994 		an = b;
1995 		bn = a % b;
1996 		a = an;
1997 		b = bn;
1998 	}
1999 	return (a);
2000 }
2001 
2002 uint64_t lkpi_nsec2hz_rem;
2003 uint64_t lkpi_nsec2hz_div = 1000000000ULL;
2004 uint64_t lkpi_nsec2hz_max;
2005 
2006 uint64_t lkpi_usec2hz_rem;
2007 uint64_t lkpi_usec2hz_div = 1000000ULL;
2008 uint64_t lkpi_usec2hz_max;
2009 
2010 uint64_t lkpi_msec2hz_rem;
2011 uint64_t lkpi_msec2hz_div = 1000ULL;
2012 uint64_t lkpi_msec2hz_max;
2013 
2014 static void
2015 linux_timer_init(void *arg)
2016 {
2017 	uint64_t gcd;
2018 
2019 	/*
2020 	 * Compute an internal HZ value which can divide 2**32 to
2021 	 * avoid timer rounding problems when the tick value wraps
2022 	 * around 2**32:
2023 	 */
2024 	linux_timer_hz_mask = 1;
2025 	while (linux_timer_hz_mask < (unsigned long)hz)
2026 		linux_timer_hz_mask *= 2;
2027 	linux_timer_hz_mask--;
2028 
2029 	/* compute some internal constants */
2030 
2031 	lkpi_nsec2hz_rem = hz;
2032 	lkpi_usec2hz_rem = hz;
2033 	lkpi_msec2hz_rem = hz;
2034 
2035 	gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div);
2036 	lkpi_nsec2hz_rem /= gcd;
2037 	lkpi_nsec2hz_div /= gcd;
2038 	lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem;
2039 
2040 	gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div);
2041 	lkpi_usec2hz_rem /= gcd;
2042 	lkpi_usec2hz_div /= gcd;
2043 	lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem;
2044 
2045 	gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div);
2046 	lkpi_msec2hz_rem /= gcd;
2047 	lkpi_msec2hz_div /= gcd;
2048 	lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem;
2049 }
2050 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL);
2051 
2052 void
2053 linux_complete_common(struct completion *c, int all)
2054 {
2055 	int wakeup_swapper;
2056 
2057 	sleepq_lock(c);
2058 	if (all) {
2059 		c->done = UINT_MAX;
2060 		wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0);
2061 	} else {
2062 		if (c->done != UINT_MAX)
2063 			c->done++;
2064 		wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0);
2065 	}
2066 	sleepq_release(c);
2067 	if (wakeup_swapper)
2068 		kick_proc0();
2069 }
2070 
2071 /*
2072  * Indefinite wait for done != 0 with or without signals.
2073  */
2074 int
2075 linux_wait_for_common(struct completion *c, int flags)
2076 {
2077 	struct task_struct *task;
2078 	int error;
2079 
2080 	if (SCHEDULER_STOPPED())
2081 		return (0);
2082 
2083 	task = current;
2084 
2085 	if (flags != 0)
2086 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2087 	else
2088 		flags = SLEEPQ_SLEEP;
2089 	error = 0;
2090 	for (;;) {
2091 		sleepq_lock(c);
2092 		if (c->done)
2093 			break;
2094 		sleepq_add(c, NULL, "completion", flags, 0);
2095 		if (flags & SLEEPQ_INTERRUPTIBLE) {
2096 			DROP_GIANT();
2097 			error = -sleepq_wait_sig(c, 0);
2098 			PICKUP_GIANT();
2099 			if (error != 0) {
2100 				linux_schedule_save_interrupt_value(task, error);
2101 				error = -ERESTARTSYS;
2102 				goto intr;
2103 			}
2104 		} else {
2105 			DROP_GIANT();
2106 			sleepq_wait(c, 0);
2107 			PICKUP_GIANT();
2108 		}
2109 	}
2110 	if (c->done != UINT_MAX)
2111 		c->done--;
2112 	sleepq_release(c);
2113 
2114 intr:
2115 	return (error);
2116 }
2117 
2118 /*
2119  * Time limited wait for done != 0 with or without signals.
2120  */
2121 int
2122 linux_wait_for_timeout_common(struct completion *c, int timeout, int flags)
2123 {
2124 	struct task_struct *task;
2125 	int end = jiffies + timeout;
2126 	int error;
2127 
2128 	if (SCHEDULER_STOPPED())
2129 		return (0);
2130 
2131 	task = current;
2132 
2133 	if (flags != 0)
2134 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2135 	else
2136 		flags = SLEEPQ_SLEEP;
2137 
2138 	for (;;) {
2139 		sleepq_lock(c);
2140 		if (c->done)
2141 			break;
2142 		sleepq_add(c, NULL, "completion", flags, 0);
2143 		sleepq_set_timeout(c, linux_timer_jiffies_until(end));
2144 
2145 		DROP_GIANT();
2146 		if (flags & SLEEPQ_INTERRUPTIBLE)
2147 			error = -sleepq_timedwait_sig(c, 0);
2148 		else
2149 			error = -sleepq_timedwait(c, 0);
2150 		PICKUP_GIANT();
2151 
2152 		if (error != 0) {
2153 			/* check for timeout */
2154 			if (error == -EWOULDBLOCK) {
2155 				error = 0;	/* timeout */
2156 			} else {
2157 				/* signal happened */
2158 				linux_schedule_save_interrupt_value(task, error);
2159 				error = -ERESTARTSYS;
2160 			}
2161 			goto done;
2162 		}
2163 	}
2164 	if (c->done != UINT_MAX)
2165 		c->done--;
2166 	sleepq_release(c);
2167 
2168 	/* return how many jiffies are left */
2169 	error = linux_timer_jiffies_until(end);
2170 done:
2171 	return (error);
2172 }
2173 
2174 int
2175 linux_try_wait_for_completion(struct completion *c)
2176 {
2177 	int isdone;
2178 
2179 	sleepq_lock(c);
2180 	isdone = (c->done != 0);
2181 	if (c->done != 0 && c->done != UINT_MAX)
2182 		c->done--;
2183 	sleepq_release(c);
2184 	return (isdone);
2185 }
2186 
2187 int
2188 linux_completion_done(struct completion *c)
2189 {
2190 	int isdone;
2191 
2192 	sleepq_lock(c);
2193 	isdone = (c->done != 0);
2194 	sleepq_release(c);
2195 	return (isdone);
2196 }
2197 
2198 static void
2199 linux_cdev_deref(struct linux_cdev *ldev)
2200 {
2201 
2202 	if (refcount_release(&ldev->refs))
2203 		kfree(ldev);
2204 }
2205 
2206 static void
2207 linux_cdev_release(struct kobject *kobj)
2208 {
2209 	struct linux_cdev *cdev;
2210 	struct kobject *parent;
2211 
2212 	cdev = container_of(kobj, struct linux_cdev, kobj);
2213 	parent = kobj->parent;
2214 	linux_destroy_dev(cdev);
2215 	linux_cdev_deref(cdev);
2216 	kobject_put(parent);
2217 }
2218 
2219 static void
2220 linux_cdev_static_release(struct kobject *kobj)
2221 {
2222 	struct linux_cdev *cdev;
2223 	struct kobject *parent;
2224 
2225 	cdev = container_of(kobj, struct linux_cdev, kobj);
2226 	parent = kobj->parent;
2227 	linux_destroy_dev(cdev);
2228 	kobject_put(parent);
2229 }
2230 
2231 void
2232 linux_destroy_dev(struct linux_cdev *ldev)
2233 {
2234 
2235 	if (ldev->cdev == NULL)
2236 		return;
2237 
2238 	MPASS((ldev->siref & LDEV_SI_DTR) == 0);
2239 	atomic_set_int(&ldev->siref, LDEV_SI_DTR);
2240 	while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0)
2241 		pause("ldevdtr", hz / 4);
2242 
2243 	destroy_dev(ldev->cdev);
2244 	ldev->cdev = NULL;
2245 }
2246 
2247 const struct kobj_type linux_cdev_ktype = {
2248 	.release = linux_cdev_release,
2249 };
2250 
2251 const struct kobj_type linux_cdev_static_ktype = {
2252 	.release = linux_cdev_static_release,
2253 };
2254 
2255 static void
2256 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate)
2257 {
2258 	struct notifier_block *nb;
2259 	struct netdev_notifier_info ni;
2260 
2261 	nb = arg;
2262 	ni.dev = (struct net_device *)ifp;
2263 	if (linkstate == LINK_STATE_UP)
2264 		nb->notifier_call(nb, NETDEV_UP, &ni);
2265 	else
2266 		nb->notifier_call(nb, NETDEV_DOWN, &ni);
2267 }
2268 
2269 static void
2270 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp)
2271 {
2272 	struct notifier_block *nb;
2273 	struct netdev_notifier_info ni;
2274 
2275 	nb = arg;
2276 	ni.dev = (struct net_device *)ifp;
2277 	nb->notifier_call(nb, NETDEV_REGISTER, &ni);
2278 }
2279 
2280 static void
2281 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp)
2282 {
2283 	struct notifier_block *nb;
2284 	struct netdev_notifier_info ni;
2285 
2286 	nb = arg;
2287 	ni.dev = (struct net_device *)ifp;
2288 	nb->notifier_call(nb, NETDEV_UNREGISTER, &ni);
2289 }
2290 
2291 static void
2292 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp)
2293 {
2294 	struct notifier_block *nb;
2295 	struct netdev_notifier_info ni;
2296 
2297 	nb = arg;
2298 	ni.dev = (struct net_device *)ifp;
2299 	nb->notifier_call(nb, NETDEV_CHANGEADDR, &ni);
2300 }
2301 
2302 static void
2303 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp)
2304 {
2305 	struct notifier_block *nb;
2306 	struct netdev_notifier_info ni;
2307 
2308 	nb = arg;
2309 	ni.dev = (struct net_device *)ifp;
2310 	nb->notifier_call(nb, NETDEV_CHANGEIFADDR, &ni);
2311 }
2312 
2313 int
2314 register_netdevice_notifier(struct notifier_block *nb)
2315 {
2316 
2317 	nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER(
2318 	    ifnet_link_event, linux_handle_ifnet_link_event, nb, 0);
2319 	nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER(
2320 	    ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0);
2321 	nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER(
2322 	    ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0);
2323 	nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER(
2324 	    iflladdr_event, linux_handle_iflladdr_event, nb, 0);
2325 
2326 	return (0);
2327 }
2328 
2329 int
2330 register_inetaddr_notifier(struct notifier_block *nb)
2331 {
2332 
2333 	nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER(
2334 	    ifaddr_event, linux_handle_ifaddr_event, nb, 0);
2335 	return (0);
2336 }
2337 
2338 int
2339 unregister_netdevice_notifier(struct notifier_block *nb)
2340 {
2341 
2342 	EVENTHANDLER_DEREGISTER(ifnet_link_event,
2343 	    nb->tags[NETDEV_UP]);
2344 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event,
2345 	    nb->tags[NETDEV_REGISTER]);
2346 	EVENTHANDLER_DEREGISTER(ifnet_departure_event,
2347 	    nb->tags[NETDEV_UNREGISTER]);
2348 	EVENTHANDLER_DEREGISTER(iflladdr_event,
2349 	    nb->tags[NETDEV_CHANGEADDR]);
2350 
2351 	return (0);
2352 }
2353 
2354 int
2355 unregister_inetaddr_notifier(struct notifier_block *nb)
2356 {
2357 
2358 	EVENTHANDLER_DEREGISTER(ifaddr_event,
2359 	    nb->tags[NETDEV_CHANGEIFADDR]);
2360 
2361 	return (0);
2362 }
2363 
2364 struct list_sort_thunk {
2365 	int (*cmp)(void *, struct list_head *, struct list_head *);
2366 	void *priv;
2367 };
2368 
2369 static inline int
2370 linux_le_cmp(void *priv, const void *d1, const void *d2)
2371 {
2372 	struct list_head *le1, *le2;
2373 	struct list_sort_thunk *thunk;
2374 
2375 	thunk = priv;
2376 	le1 = *(__DECONST(struct list_head **, d1));
2377 	le2 = *(__DECONST(struct list_head **, d2));
2378 	return ((thunk->cmp)(thunk->priv, le1, le2));
2379 }
2380 
2381 void
2382 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv,
2383     struct list_head *a, struct list_head *b))
2384 {
2385 	struct list_sort_thunk thunk;
2386 	struct list_head **ar, *le;
2387 	size_t count, i;
2388 
2389 	count = 0;
2390 	list_for_each(le, head)
2391 		count++;
2392 	ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK);
2393 	i = 0;
2394 	list_for_each(le, head)
2395 		ar[i++] = le;
2396 	thunk.cmp = cmp;
2397 	thunk.priv = priv;
2398 	qsort_r(ar, count, sizeof(struct list_head *), &thunk, linux_le_cmp);
2399 	INIT_LIST_HEAD(head);
2400 	for (i = 0; i < count; i++)
2401 		list_add_tail(ar[i], head);
2402 	free(ar, M_KMALLOC);
2403 }
2404 
2405 void
2406 linux_irq_handler(void *ent)
2407 {
2408 	struct irq_ent *irqe;
2409 
2410 	if (linux_set_current_flags(curthread, M_NOWAIT))
2411 		return;
2412 
2413 	irqe = ent;
2414 	irqe->handler(irqe->irq, irqe->arg);
2415 }
2416 
2417 #if defined(__i386__) || defined(__amd64__)
2418 int
2419 linux_wbinvd_on_all_cpus(void)
2420 {
2421 
2422 	pmap_invalidate_cache();
2423 	return (0);
2424 }
2425 #endif
2426 
2427 int
2428 linux_on_each_cpu(void callback(void *), void *data)
2429 {
2430 
2431 	smp_rendezvous(smp_no_rendezvous_barrier, callback,
2432 	    smp_no_rendezvous_barrier, data);
2433 	return (0);
2434 }
2435 
2436 int
2437 linux_in_atomic(void)
2438 {
2439 
2440 	return ((curthread->td_pflags & TDP_NOFAULTING) != 0);
2441 }
2442 
2443 struct linux_cdev *
2444 linux_find_cdev(const char *name, unsigned major, unsigned minor)
2445 {
2446 	dev_t dev = MKDEV(major, minor);
2447 	struct cdev *cdev;
2448 
2449 	dev_lock();
2450 	LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) {
2451 		struct linux_cdev *ldev = cdev->si_drv1;
2452 		if (ldev->dev == dev &&
2453 		    strcmp(kobject_name(&ldev->kobj), name) == 0) {
2454 			break;
2455 		}
2456 	}
2457 	dev_unlock();
2458 
2459 	return (cdev != NULL ? cdev->si_drv1 : NULL);
2460 }
2461 
2462 int
2463 __register_chrdev(unsigned int major, unsigned int baseminor,
2464     unsigned int count, const char *name,
2465     const struct file_operations *fops)
2466 {
2467 	struct linux_cdev *cdev;
2468 	int ret = 0;
2469 	int i;
2470 
2471 	for (i = baseminor; i < baseminor + count; i++) {
2472 		cdev = cdev_alloc();
2473 		cdev->ops = fops;
2474 		kobject_set_name(&cdev->kobj, name);
2475 
2476 		ret = cdev_add(cdev, makedev(major, i), 1);
2477 		if (ret != 0)
2478 			break;
2479 	}
2480 	return (ret);
2481 }
2482 
2483 int
2484 __register_chrdev_p(unsigned int major, unsigned int baseminor,
2485     unsigned int count, const char *name,
2486     const struct file_operations *fops, uid_t uid,
2487     gid_t gid, int mode)
2488 {
2489 	struct linux_cdev *cdev;
2490 	int ret = 0;
2491 	int i;
2492 
2493 	for (i = baseminor; i < baseminor + count; i++) {
2494 		cdev = cdev_alloc();
2495 		cdev->ops = fops;
2496 		kobject_set_name(&cdev->kobj, name);
2497 
2498 		ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode);
2499 		if (ret != 0)
2500 			break;
2501 	}
2502 	return (ret);
2503 }
2504 
2505 void
2506 __unregister_chrdev(unsigned int major, unsigned int baseminor,
2507     unsigned int count, const char *name)
2508 {
2509 	struct linux_cdev *cdevp;
2510 	int i;
2511 
2512 	for (i = baseminor; i < baseminor + count; i++) {
2513 		cdevp = linux_find_cdev(name, major, i);
2514 		if (cdevp != NULL)
2515 			cdev_del(cdevp);
2516 	}
2517 }
2518 
2519 void
2520 linux_dump_stack(void)
2521 {
2522 #ifdef STACK
2523 	struct stack st;
2524 
2525 	stack_zero(&st);
2526 	stack_save(&st);
2527 	stack_print(&st);
2528 #endif
2529 }
2530 
2531 int
2532 linuxkpi_net_ratelimit(void)
2533 {
2534 
2535 	return (ppsratecheck(&lkpi_net_lastlog, &lkpi_net_curpps,
2536 	   lkpi_net_maxpps));
2537 }
2538 
2539 #if defined(__i386__) || defined(__amd64__)
2540 bool linux_cpu_has_clflush;
2541 #endif
2542 
2543 static void
2544 linux_compat_init(void *arg)
2545 {
2546 	struct sysctl_oid *rootoid;
2547 	int i;
2548 
2549 #if defined(__i386__) || defined(__amd64__)
2550 	linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH);
2551 #endif
2552 	rw_init(&linux_vma_lock, "lkpi-vma-lock");
2553 
2554 	rootoid = SYSCTL_ADD_ROOT_NODE(NULL,
2555 	    OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys");
2556 	kobject_init(&linux_class_root, &linux_class_ktype);
2557 	kobject_set_name(&linux_class_root, "class");
2558 	linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid),
2559 	    OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class");
2560 	kobject_init(&linux_root_device.kobj, &linux_dev_ktype);
2561 	kobject_set_name(&linux_root_device.kobj, "device");
2562 	linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL,
2563 	    SYSCTL_CHILDREN(rootoid), OID_AUTO, "device",
2564 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "device");
2565 	linux_root_device.bsddev = root_bus;
2566 	linux_class_misc.name = "misc";
2567 	class_register(&linux_class_misc);
2568 	INIT_LIST_HEAD(&pci_drivers);
2569 	INIT_LIST_HEAD(&pci_devices);
2570 	spin_lock_init(&pci_lock);
2571 	mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF);
2572 	for (i = 0; i < VMMAP_HASH_SIZE; i++)
2573 		LIST_INIT(&vmmaphead[i]);
2574 	init_waitqueue_head(&linux_bit_waitq);
2575 	init_waitqueue_head(&linux_var_waitq);
2576 }
2577 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL);
2578 
2579 static void
2580 linux_compat_uninit(void *arg)
2581 {
2582 	linux_kobject_kfree_name(&linux_class_root);
2583 	linux_kobject_kfree_name(&linux_root_device.kobj);
2584 	linux_kobject_kfree_name(&linux_class_misc.kobj);
2585 
2586 	mtx_destroy(&vmmaplock);
2587 	spin_lock_destroy(&pci_lock);
2588 	rw_destroy(&linux_vma_lock);
2589 }
2590 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL);
2591 
2592 /*
2593  * NOTE: Linux frequently uses "unsigned long" for pointer to integer
2594  * conversion and vice versa, where in FreeBSD "uintptr_t" would be
2595  * used. Assert these types have the same size, else some parts of the
2596  * LinuxKPI may not work like expected:
2597  */
2598 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t));
2599