xref: /freebsd/sys/compat/linuxkpi/common/src/linux_compat.c (revision b685454a1107a512e105ceb09b5ff04dadc04d1a)
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/fcntl.h>
48 #include <sys/file.h>
49 #include <sys/filio.h>
50 #include <sys/rwlock.h>
51 #include <sys/mman.h>
52 #include <sys/stack.h>
53 #include <sys/user.h>
54 
55 #include <vm/vm.h>
56 #include <vm/pmap.h>
57 #include <vm/vm_object.h>
58 #include <vm/vm_page.h>
59 #include <vm/vm_pager.h>
60 
61 #include <machine/stdarg.h>
62 
63 #if defined(__i386__) || defined(__amd64__)
64 #include <machine/md_var.h>
65 #endif
66 
67 #include <linux/kobject.h>
68 #include <linux/device.h>
69 #include <linux/slab.h>
70 #include <linux/module.h>
71 #include <linux/moduleparam.h>
72 #include <linux/cdev.h>
73 #include <linux/file.h>
74 #include <linux/sysfs.h>
75 #include <linux/mm.h>
76 #include <linux/io.h>
77 #include <linux/vmalloc.h>
78 #include <linux/netdevice.h>
79 #include <linux/timer.h>
80 #include <linux/interrupt.h>
81 #include <linux/uaccess.h>
82 #include <linux/list.h>
83 #include <linux/kthread.h>
84 #include <linux/kernel.h>
85 #include <linux/compat.h>
86 #include <linux/poll.h>
87 #include <linux/smp.h>
88 
89 #if defined(__i386__) || defined(__amd64__)
90 #include <asm/smp.h>
91 #endif
92 
93 SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW, 0, "LinuxKPI parameters");
94 
95 int linuxkpi_debug;
96 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN,
97     &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable.");
98 
99 MALLOC_DEFINE(M_KMALLOC, "linux", "Linux kmalloc compat");
100 
101 #include <linux/rbtree.h>
102 /* Undo Linux compat changes. */
103 #undef RB_ROOT
104 #undef file
105 #undef cdev
106 #define	RB_ROOT(head)	(head)->rbh_root
107 
108 static void linux_cdev_deref(struct linux_cdev *ldev);
109 static struct vm_area_struct *linux_cdev_handle_find(void *handle);
110 
111 struct kobject linux_class_root;
112 struct device linux_root_device;
113 struct class linux_class_misc;
114 struct list_head pci_drivers;
115 struct list_head pci_devices;
116 spinlock_t pci_lock;
117 
118 unsigned long linux_timer_hz_mask;
119 
120 int
121 panic_cmp(struct rb_node *one, struct rb_node *two)
122 {
123 	panic("no cmp");
124 }
125 
126 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp);
127 
128 int
129 kobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list args)
130 {
131 	va_list tmp_va;
132 	int len;
133 	char *old;
134 	char *name;
135 	char dummy;
136 
137 	old = kobj->name;
138 
139 	if (old && fmt == NULL)
140 		return (0);
141 
142 	/* compute length of string */
143 	va_copy(tmp_va, args);
144 	len = vsnprintf(&dummy, 0, fmt, tmp_va);
145 	va_end(tmp_va);
146 
147 	/* account for zero termination */
148 	len++;
149 
150 	/* check for error */
151 	if (len < 1)
152 		return (-EINVAL);
153 
154 	/* allocate memory for string */
155 	name = kzalloc(len, GFP_KERNEL);
156 	if (name == NULL)
157 		return (-ENOMEM);
158 	vsnprintf(name, len, fmt, args);
159 	kobj->name = name;
160 
161 	/* free old string */
162 	kfree(old);
163 
164 	/* filter new string */
165 	for (; *name != '\0'; name++)
166 		if (*name == '/')
167 			*name = '!';
168 	return (0);
169 }
170 
171 int
172 kobject_set_name(struct kobject *kobj, const char *fmt, ...)
173 {
174 	va_list args;
175 	int error;
176 
177 	va_start(args, fmt);
178 	error = kobject_set_name_vargs(kobj, fmt, args);
179 	va_end(args);
180 
181 	return (error);
182 }
183 
184 static int
185 kobject_add_complete(struct kobject *kobj, struct kobject *parent)
186 {
187 	const struct kobj_type *t;
188 	int error;
189 
190 	kobj->parent = parent;
191 	error = sysfs_create_dir(kobj);
192 	if (error == 0 && kobj->ktype && kobj->ktype->default_attrs) {
193 		struct attribute **attr;
194 		t = kobj->ktype;
195 
196 		for (attr = t->default_attrs; *attr != NULL; attr++) {
197 			error = sysfs_create_file(kobj, *attr);
198 			if (error)
199 				break;
200 		}
201 		if (error)
202 			sysfs_remove_dir(kobj);
203 
204 	}
205 	return (error);
206 }
207 
208 int
209 kobject_add(struct kobject *kobj, struct kobject *parent, const char *fmt, ...)
210 {
211 	va_list args;
212 	int error;
213 
214 	va_start(args, fmt);
215 	error = kobject_set_name_vargs(kobj, fmt, args);
216 	va_end(args);
217 	if (error)
218 		return (error);
219 
220 	return kobject_add_complete(kobj, parent);
221 }
222 
223 void
224 linux_kobject_release(struct kref *kref)
225 {
226 	struct kobject *kobj;
227 	char *name;
228 
229 	kobj = container_of(kref, struct kobject, kref);
230 	sysfs_remove_dir(kobj);
231 	name = kobj->name;
232 	if (kobj->ktype && kobj->ktype->release)
233 		kobj->ktype->release(kobj);
234 	kfree(name);
235 }
236 
237 static void
238 linux_kobject_kfree(struct kobject *kobj)
239 {
240 	kfree(kobj);
241 }
242 
243 static void
244 linux_kobject_kfree_name(struct kobject *kobj)
245 {
246 	if (kobj) {
247 		kfree(kobj->name);
248 	}
249 }
250 
251 const struct kobj_type linux_kfree_type = {
252 	.release = linux_kobject_kfree
253 };
254 
255 static void
256 linux_device_release(struct device *dev)
257 {
258 	pr_debug("linux_device_release: %s\n", dev_name(dev));
259 	kfree(dev);
260 }
261 
262 static ssize_t
263 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf)
264 {
265 	struct class_attribute *dattr;
266 	ssize_t error;
267 
268 	dattr = container_of(attr, struct class_attribute, attr);
269 	error = -EIO;
270 	if (dattr->show)
271 		error = dattr->show(container_of(kobj, struct class, kobj),
272 		    dattr, buf);
273 	return (error);
274 }
275 
276 static ssize_t
277 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf,
278     size_t count)
279 {
280 	struct class_attribute *dattr;
281 	ssize_t error;
282 
283 	dattr = container_of(attr, struct class_attribute, attr);
284 	error = -EIO;
285 	if (dattr->store)
286 		error = dattr->store(container_of(kobj, struct class, kobj),
287 		    dattr, buf, count);
288 	return (error);
289 }
290 
291 static void
292 linux_class_release(struct kobject *kobj)
293 {
294 	struct class *class;
295 
296 	class = container_of(kobj, struct class, kobj);
297 	if (class->class_release)
298 		class->class_release(class);
299 }
300 
301 static const struct sysfs_ops linux_class_sysfs = {
302 	.show  = linux_class_show,
303 	.store = linux_class_store,
304 };
305 
306 const struct kobj_type linux_class_ktype = {
307 	.release = linux_class_release,
308 	.sysfs_ops = &linux_class_sysfs
309 };
310 
311 static void
312 linux_dev_release(struct kobject *kobj)
313 {
314 	struct device *dev;
315 
316 	dev = container_of(kobj, struct device, kobj);
317 	/* This is the precedence defined by linux. */
318 	if (dev->release)
319 		dev->release(dev);
320 	else if (dev->class && dev->class->dev_release)
321 		dev->class->dev_release(dev);
322 }
323 
324 static ssize_t
325 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf)
326 {
327 	struct device_attribute *dattr;
328 	ssize_t error;
329 
330 	dattr = container_of(attr, struct device_attribute, attr);
331 	error = -EIO;
332 	if (dattr->show)
333 		error = dattr->show(container_of(kobj, struct device, kobj),
334 		    dattr, buf);
335 	return (error);
336 }
337 
338 static ssize_t
339 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf,
340     size_t count)
341 {
342 	struct device_attribute *dattr;
343 	ssize_t error;
344 
345 	dattr = container_of(attr, struct device_attribute, attr);
346 	error = -EIO;
347 	if (dattr->store)
348 		error = dattr->store(container_of(kobj, struct device, kobj),
349 		    dattr, buf, count);
350 	return (error);
351 }
352 
353 static const struct sysfs_ops linux_dev_sysfs = {
354 	.show  = linux_dev_show,
355 	.store = linux_dev_store,
356 };
357 
358 const struct kobj_type linux_dev_ktype = {
359 	.release = linux_dev_release,
360 	.sysfs_ops = &linux_dev_sysfs
361 };
362 
363 struct device *
364 device_create(struct class *class, struct device *parent, dev_t devt,
365     void *drvdata, const char *fmt, ...)
366 {
367 	struct device *dev;
368 	va_list args;
369 
370 	dev = kzalloc(sizeof(*dev), M_WAITOK);
371 	dev->parent = parent;
372 	dev->class = class;
373 	dev->devt = devt;
374 	dev->driver_data = drvdata;
375 	dev->release = linux_device_release;
376 	va_start(args, fmt);
377 	kobject_set_name_vargs(&dev->kobj, fmt, args);
378 	va_end(args);
379 	device_register(dev);
380 
381 	return (dev);
382 }
383 
384 int
385 kobject_init_and_add(struct kobject *kobj, const struct kobj_type *ktype,
386     struct kobject *parent, const char *fmt, ...)
387 {
388 	va_list args;
389 	int error;
390 
391 	kobject_init(kobj, ktype);
392 	kobj->ktype = ktype;
393 	kobj->parent = parent;
394 	kobj->name = NULL;
395 
396 	va_start(args, fmt);
397 	error = kobject_set_name_vargs(kobj, fmt, args);
398 	va_end(args);
399 	if (error)
400 		return (error);
401 	return kobject_add_complete(kobj, parent);
402 }
403 
404 static void
405 linux_kq_lock(void *arg)
406 {
407 	spinlock_t *s = arg;
408 
409 	spin_lock(s);
410 }
411 static void
412 linux_kq_unlock(void *arg)
413 {
414 	spinlock_t *s = arg;
415 
416 	spin_unlock(s);
417 }
418 
419 static void
420 linux_kq_lock_owned(void *arg)
421 {
422 #ifdef INVARIANTS
423 	spinlock_t *s = arg;
424 
425 	mtx_assert(&s->m, MA_OWNED);
426 #endif
427 }
428 
429 static void
430 linux_kq_lock_unowned(void *arg)
431 {
432 #ifdef INVARIANTS
433 	spinlock_t *s = arg;
434 
435 	mtx_assert(&s->m, MA_NOTOWNED);
436 #endif
437 }
438 
439 static void
440 linux_file_kqfilter_poll(struct linux_file *, int);
441 
442 struct linux_file *
443 linux_file_alloc(void)
444 {
445 	struct linux_file *filp;
446 
447 	filp = kzalloc(sizeof(*filp), GFP_KERNEL);
448 
449 	/* set initial refcount */
450 	filp->f_count = 1;
451 
452 	/* setup fields needed by kqueue support */
453 	spin_lock_init(&filp->f_kqlock);
454 	knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock,
455 	    linux_kq_lock, linux_kq_unlock,
456 	    linux_kq_lock_owned, linux_kq_lock_unowned);
457 
458 	return (filp);
459 }
460 
461 void
462 linux_file_free(struct linux_file *filp)
463 {
464 	if (filp->_file == NULL) {
465 		if (filp->f_shmem != NULL)
466 			vm_object_deallocate(filp->f_shmem);
467 		kfree(filp);
468 	} else {
469 		/*
470 		 * The close method of the character device or file
471 		 * will free the linux_file structure:
472 		 */
473 		_fdrop(filp->_file, curthread);
474 	}
475 }
476 
477 static int
478 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
479     vm_page_t *mres)
480 {
481 	struct vm_area_struct *vmap;
482 
483 	vmap = linux_cdev_handle_find(vm_obj->handle);
484 
485 	MPASS(vmap != NULL);
486 	MPASS(vmap->vm_private_data == vm_obj->handle);
487 
488 	if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) {
489 		vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset;
490 		vm_page_t page;
491 
492 		if (((*mres)->flags & PG_FICTITIOUS) != 0) {
493 			/*
494 			 * If the passed in result page is a fake
495 			 * page, update it with the new physical
496 			 * address.
497 			 */
498 			page = *mres;
499 			vm_page_updatefake(page, paddr, vm_obj->memattr);
500 		} else {
501 			/*
502 			 * Replace the passed in "mres" page with our
503 			 * own fake page and free up the all of the
504 			 * original pages.
505 			 */
506 			VM_OBJECT_WUNLOCK(vm_obj);
507 			page = vm_page_getfake(paddr, vm_obj->memattr);
508 			VM_OBJECT_WLOCK(vm_obj);
509 
510 			vm_page_replace_checked(page, vm_obj,
511 			    (*mres)->pindex, *mres);
512 
513 			vm_page_lock(*mres);
514 			vm_page_free(*mres);
515 			vm_page_unlock(*mres);
516 			*mres = page;
517 		}
518 		page->valid = VM_PAGE_BITS_ALL;
519 		return (VM_PAGER_OK);
520 	}
521 	return (VM_PAGER_FAIL);
522 }
523 
524 static int
525 linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type,
526     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
527 {
528 	struct vm_area_struct *vmap;
529 	int err;
530 
531 	linux_set_current(curthread);
532 
533 	/* get VM area structure */
534 	vmap = linux_cdev_handle_find(vm_obj->handle);
535 	MPASS(vmap != NULL);
536 	MPASS(vmap->vm_private_data == vm_obj->handle);
537 
538 	VM_OBJECT_WUNLOCK(vm_obj);
539 
540 	down_write(&vmap->vm_mm->mmap_sem);
541 	if (unlikely(vmap->vm_ops == NULL)) {
542 		err = VM_FAULT_SIGBUS;
543 	} else {
544 		struct vm_fault vmf;
545 
546 		/* fill out VM fault structure */
547 		vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx);
548 		vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
549 		vmf.pgoff = 0;
550 		vmf.page = NULL;
551 		vmf.vma = vmap;
552 
553 		vmap->vm_pfn_count = 0;
554 		vmap->vm_pfn_pcount = &vmap->vm_pfn_count;
555 		vmap->vm_obj = vm_obj;
556 
557 		err = vmap->vm_ops->fault(vmap, &vmf);
558 
559 		while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) {
560 			kern_yield(PRI_USER);
561 			err = vmap->vm_ops->fault(vmap, &vmf);
562 		}
563 	}
564 
565 	/* translate return code */
566 	switch (err) {
567 	case VM_FAULT_OOM:
568 		err = VM_PAGER_AGAIN;
569 		break;
570 	case VM_FAULT_SIGBUS:
571 		err = VM_PAGER_BAD;
572 		break;
573 	case VM_FAULT_NOPAGE:
574 		/*
575 		 * By contract the fault handler will return having
576 		 * busied all the pages itself. If pidx is already
577 		 * found in the object, it will simply xbusy the first
578 		 * page and return with vm_pfn_count set to 1.
579 		 */
580 		*first = vmap->vm_pfn_first;
581 		*last = *first + vmap->vm_pfn_count - 1;
582 		err = VM_PAGER_OK;
583 		break;
584 	default:
585 		err = VM_PAGER_ERROR;
586 		break;
587 	}
588 	up_write(&vmap->vm_mm->mmap_sem);
589 	VM_OBJECT_WLOCK(vm_obj);
590 	return (err);
591 }
592 
593 static struct rwlock linux_vma_lock;
594 static TAILQ_HEAD(, vm_area_struct) linux_vma_head =
595     TAILQ_HEAD_INITIALIZER(linux_vma_head);
596 
597 static void
598 linux_cdev_handle_free(struct vm_area_struct *vmap)
599 {
600 	/* Drop reference on vm_file */
601 	if (vmap->vm_file != NULL)
602 		fput(vmap->vm_file);
603 
604 	/* Drop reference on mm_struct */
605 	mmput(vmap->vm_mm);
606 
607 	kfree(vmap);
608 }
609 
610 static void
611 linux_cdev_handle_remove(struct vm_area_struct *vmap)
612 {
613 	rw_wlock(&linux_vma_lock);
614 	TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry);
615 	rw_wunlock(&linux_vma_lock);
616 }
617 
618 static struct vm_area_struct *
619 linux_cdev_handle_find(void *handle)
620 {
621 	struct vm_area_struct *vmap;
622 
623 	rw_rlock(&linux_vma_lock);
624 	TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) {
625 		if (vmap->vm_private_data == handle)
626 			break;
627 	}
628 	rw_runlock(&linux_vma_lock);
629 	return (vmap);
630 }
631 
632 static int
633 linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
634 		      vm_ooffset_t foff, struct ucred *cred, u_short *color)
635 {
636 
637 	MPASS(linux_cdev_handle_find(handle) != NULL);
638 	*color = 0;
639 	return (0);
640 }
641 
642 static void
643 linux_cdev_pager_dtor(void *handle)
644 {
645 	const struct vm_operations_struct *vm_ops;
646 	struct vm_area_struct *vmap;
647 
648 	vmap = linux_cdev_handle_find(handle);
649 	MPASS(vmap != NULL);
650 
651 	/*
652 	 * Remove handle before calling close operation to prevent
653 	 * other threads from reusing the handle pointer.
654 	 */
655 	linux_cdev_handle_remove(vmap);
656 
657 	down_write(&vmap->vm_mm->mmap_sem);
658 	vm_ops = vmap->vm_ops;
659 	if (likely(vm_ops != NULL))
660 		vm_ops->close(vmap);
661 	up_write(&vmap->vm_mm->mmap_sem);
662 
663 	linux_cdev_handle_free(vmap);
664 }
665 
666 static struct cdev_pager_ops linux_cdev_pager_ops[2] = {
667   {
668 	/* OBJT_MGTDEVICE */
669 	.cdev_pg_populate	= linux_cdev_pager_populate,
670 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
671 	.cdev_pg_dtor	= linux_cdev_pager_dtor
672   },
673   {
674 	/* OBJT_DEVICE */
675 	.cdev_pg_fault	= linux_cdev_pager_fault,
676 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
677 	.cdev_pg_dtor	= linux_cdev_pager_dtor
678   },
679 };
680 
681 int
682 zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
683     unsigned long size)
684 {
685 	vm_object_t obj;
686 	vm_page_t m;
687 
688 	obj = vma->vm_obj;
689 	if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0)
690 		return (-ENOTSUP);
691 	VM_OBJECT_RLOCK(obj);
692 	for (m = vm_page_find_least(obj, OFF_TO_IDX(address));
693 	    m != NULL && m->pindex < OFF_TO_IDX(address + size);
694 	    m = TAILQ_NEXT(m, listq))
695 		pmap_remove_all(m);
696 	VM_OBJECT_RUNLOCK(obj);
697 	return (0);
698 }
699 
700 static struct file_operations dummy_ldev_ops = {
701 	/* XXXKIB */
702 };
703 
704 static struct linux_cdev dummy_ldev = {
705 	.ops = &dummy_ldev_ops,
706 };
707 
708 #define	LDEV_SI_DTR	0x0001
709 #define	LDEV_SI_REF	0x0002
710 
711 static void
712 linux_get_fop(struct linux_file *filp, const struct file_operations **fop,
713     struct linux_cdev **dev)
714 {
715 	struct linux_cdev *ldev;
716 	u_int siref;
717 
718 	ldev = filp->f_cdev;
719 	*fop = filp->f_op;
720 	if (ldev != NULL) {
721 		for (siref = ldev->siref;;) {
722 			if ((siref & LDEV_SI_DTR) != 0) {
723 				ldev = &dummy_ldev;
724 				siref = ldev->siref;
725 				*fop = ldev->ops;
726 				MPASS((ldev->siref & LDEV_SI_DTR) == 0);
727 			} else if (atomic_fcmpset_int(&ldev->siref, &siref,
728 			    siref + LDEV_SI_REF)) {
729 				break;
730 			}
731 		}
732 	}
733 	*dev = ldev;
734 }
735 
736 static void
737 linux_drop_fop(struct linux_cdev *ldev)
738 {
739 
740 	if (ldev == NULL)
741 		return;
742 	MPASS((ldev->siref & ~LDEV_SI_DTR) != 0);
743 	atomic_subtract_int(&ldev->siref, LDEV_SI_REF);
744 }
745 
746 #define	OPW(fp,td,code) ({			\
747 	struct file *__fpop;			\
748 	__typeof(code) __retval;		\
749 						\
750 	__fpop = (td)->td_fpop;			\
751 	(td)->td_fpop = (fp);			\
752 	__retval = (code);			\
753 	(td)->td_fpop = __fpop;			\
754 	__retval;				\
755 })
756 
757 static int
758 linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td,
759     struct file *file)
760 {
761 	struct linux_cdev *ldev;
762 	struct linux_file *filp;
763 	const struct file_operations *fop;
764 	int error;
765 
766 	ldev = dev->si_drv1;
767 
768 	filp = linux_file_alloc();
769 	filp->f_dentry = &filp->f_dentry_store;
770 	filp->f_op = ldev->ops;
771 	filp->f_mode = file->f_flag;
772 	filp->f_flags = file->f_flag;
773 	filp->f_vnode = file->f_vnode;
774 	filp->_file = file;
775 	refcount_acquire(&ldev->refs);
776 	filp->f_cdev = ldev;
777 
778 	linux_set_current(td);
779 	linux_get_fop(filp, &fop, &ldev);
780 
781 	if (fop->open != NULL) {
782 		error = -fop->open(file->f_vnode, filp);
783 		if (error != 0) {
784 			linux_drop_fop(ldev);
785 			linux_cdev_deref(filp->f_cdev);
786 			kfree(filp);
787 			return (error);
788 		}
789 	}
790 
791 	/* hold on to the vnode - used for fstat() */
792 	vhold(filp->f_vnode);
793 
794 	/* release the file from devfs */
795 	finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops);
796 	linux_drop_fop(ldev);
797 	return (ENXIO);
798 }
799 
800 #define	LINUX_IOCTL_MIN_PTR 0x10000UL
801 #define	LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX)
802 
803 static inline int
804 linux_remap_address(void **uaddr, size_t len)
805 {
806 	uintptr_t uaddr_val = (uintptr_t)(*uaddr);
807 
808 	if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR &&
809 	    uaddr_val < LINUX_IOCTL_MAX_PTR)) {
810 		struct task_struct *pts = current;
811 		if (pts == NULL) {
812 			*uaddr = NULL;
813 			return (1);
814 		}
815 
816 		/* compute data offset */
817 		uaddr_val -= LINUX_IOCTL_MIN_PTR;
818 
819 		/* check that length is within bounds */
820 		if ((len > IOCPARM_MAX) ||
821 		    (uaddr_val + len) > pts->bsd_ioctl_len) {
822 			*uaddr = NULL;
823 			return (1);
824 		}
825 
826 		/* re-add kernel buffer address */
827 		uaddr_val += (uintptr_t)pts->bsd_ioctl_data;
828 
829 		/* update address location */
830 		*uaddr = (void *)uaddr_val;
831 		return (1);
832 	}
833 	return (0);
834 }
835 
836 int
837 linux_copyin(const void *uaddr, void *kaddr, size_t len)
838 {
839 	if (linux_remap_address(__DECONST(void **, &uaddr), len)) {
840 		if (uaddr == NULL)
841 			return (-EFAULT);
842 		memcpy(kaddr, uaddr, len);
843 		return (0);
844 	}
845 	return (-copyin(uaddr, kaddr, len));
846 }
847 
848 int
849 linux_copyout(const void *kaddr, void *uaddr, size_t len)
850 {
851 	if (linux_remap_address(&uaddr, len)) {
852 		if (uaddr == NULL)
853 			return (-EFAULT);
854 		memcpy(uaddr, kaddr, len);
855 		return (0);
856 	}
857 	return (-copyout(kaddr, uaddr, len));
858 }
859 
860 size_t
861 linux_clear_user(void *_uaddr, size_t _len)
862 {
863 	uint8_t *uaddr = _uaddr;
864 	size_t len = _len;
865 
866 	/* make sure uaddr is aligned before going into the fast loop */
867 	while (((uintptr_t)uaddr & 7) != 0 && len > 7) {
868 		if (subyte(uaddr, 0))
869 			return (_len);
870 		uaddr++;
871 		len--;
872 	}
873 
874 	/* zero 8 bytes at a time */
875 	while (len > 7) {
876 #ifdef __LP64__
877 		if (suword64(uaddr, 0))
878 			return (_len);
879 #else
880 		if (suword32(uaddr, 0))
881 			return (_len);
882 		if (suword32(uaddr + 4, 0))
883 			return (_len);
884 #endif
885 		uaddr += 8;
886 		len -= 8;
887 	}
888 
889 	/* zero fill end, if any */
890 	while (len > 0) {
891 		if (subyte(uaddr, 0))
892 			return (_len);
893 		uaddr++;
894 		len--;
895 	}
896 	return (0);
897 }
898 
899 int
900 linux_access_ok(int rw, const void *uaddr, size_t len)
901 {
902 	uintptr_t saddr;
903 	uintptr_t eaddr;
904 
905 	/* get start and end address */
906 	saddr = (uintptr_t)uaddr;
907 	eaddr = (uintptr_t)uaddr + len;
908 
909 	/* verify addresses are valid for userspace */
910 	return ((saddr == eaddr) ||
911 	    (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS));
912 }
913 
914 /*
915  * This function should return either EINTR or ERESTART depending on
916  * the signal type sent to this thread:
917  */
918 static int
919 linux_get_error(struct task_struct *task, int error)
920 {
921 	/* check for signal type interrupt code */
922 	if (error == EINTR || error == ERESTARTSYS || error == ERESTART) {
923 		error = -linux_schedule_get_interrupt_value(task);
924 		if (error == 0)
925 			error = EINTR;
926 	}
927 	return (error);
928 }
929 
930 static int
931 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
932     const struct file_operations *fop, u_long cmd, caddr_t data,
933     struct thread *td)
934 {
935 	struct task_struct *task = current;
936 	unsigned size;
937 	int error;
938 
939 	size = IOCPARM_LEN(cmd);
940 	/* refer to logic in sys_ioctl() */
941 	if (size > 0) {
942 		/*
943 		 * Setup hint for linux_copyin() and linux_copyout().
944 		 *
945 		 * Background: Linux code expects a user-space address
946 		 * while FreeBSD supplies a kernel-space address.
947 		 */
948 		task->bsd_ioctl_data = data;
949 		task->bsd_ioctl_len = size;
950 		data = (void *)LINUX_IOCTL_MIN_PTR;
951 	} else {
952 		/* fetch user-space pointer */
953 		data = *(void **)data;
954 	}
955 #if defined(__amd64__)
956 	if (td->td_proc->p_elf_machine == EM_386) {
957 		/* try the compat IOCTL handler first */
958 		if (fop->compat_ioctl != NULL) {
959 			error = -OPW(fp, td, fop->compat_ioctl(filp,
960 			    cmd, (u_long)data));
961 		} else {
962 			error = ENOTTY;
963 		}
964 
965 		/* fallback to the regular IOCTL handler, if any */
966 		if (error == ENOTTY && fop->unlocked_ioctl != NULL) {
967 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
968 			    cmd, (u_long)data));
969 		}
970 	} else
971 #endif
972 	{
973 		if (fop->unlocked_ioctl != NULL) {
974 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
975 			    cmd, (u_long)data));
976 		} else {
977 			error = ENOTTY;
978 		}
979 	}
980 	if (size > 0) {
981 		task->bsd_ioctl_data = NULL;
982 		task->bsd_ioctl_len = 0;
983 	}
984 
985 	if (error == EWOULDBLOCK) {
986 		/* update kqfilter status, if any */
987 		linux_file_kqfilter_poll(filp,
988 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
989 	} else {
990 		error = linux_get_error(task, error);
991 	}
992 	return (error);
993 }
994 
995 #define	LINUX_POLL_TABLE_NORMAL ((poll_table *)1)
996 
997 /*
998  * This function atomically updates the poll wakeup state and returns
999  * the previous state at the time of update.
1000  */
1001 static uint8_t
1002 linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate)
1003 {
1004 	int c, old;
1005 
1006 	c = v->counter;
1007 
1008 	while ((old = atomic_cmpxchg(v, c, pstate[c])) != c)
1009 		c = old;
1010 
1011 	return (c);
1012 }
1013 
1014 
1015 static int
1016 linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key)
1017 {
1018 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1019 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */
1020 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1021 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY,
1022 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */
1023 	};
1024 	struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq);
1025 
1026 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1027 	case LINUX_FWQ_STATE_QUEUED:
1028 		linux_poll_wakeup(filp);
1029 		return (1);
1030 	default:
1031 		return (0);
1032 	}
1033 }
1034 
1035 void
1036 linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p)
1037 {
1038 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1039 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY,
1040 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1041 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */
1042 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED,
1043 	};
1044 
1045 	/* check if we are called inside the select system call */
1046 	if (p == LINUX_POLL_TABLE_NORMAL)
1047 		selrecord(curthread, &filp->f_selinfo);
1048 
1049 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1050 	case LINUX_FWQ_STATE_INIT:
1051 		/* NOTE: file handles can only belong to one wait-queue */
1052 		filp->f_wait_queue.wqh = wqh;
1053 		filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback;
1054 		add_wait_queue(wqh, &filp->f_wait_queue.wq);
1055 		atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED);
1056 		break;
1057 	default:
1058 		break;
1059 	}
1060 }
1061 
1062 static void
1063 linux_poll_wait_dequeue(struct linux_file *filp)
1064 {
1065 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1066 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT,	/* NOP */
1067 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT,
1068 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT,
1069 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT,
1070 	};
1071 
1072 	seldrain(&filp->f_selinfo);
1073 
1074 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1075 	case LINUX_FWQ_STATE_NOT_READY:
1076 	case LINUX_FWQ_STATE_QUEUED:
1077 	case LINUX_FWQ_STATE_READY:
1078 		remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq);
1079 		break;
1080 	default:
1081 		break;
1082 	}
1083 }
1084 
1085 void
1086 linux_poll_wakeup(struct linux_file *filp)
1087 {
1088 	/* this function should be NULL-safe */
1089 	if (filp == NULL)
1090 		return;
1091 
1092 	selwakeup(&filp->f_selinfo);
1093 
1094 	spin_lock(&filp->f_kqlock);
1095 	filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ |
1096 	    LINUX_KQ_FLAG_NEED_WRITE;
1097 
1098 	/* make sure the "knote" gets woken up */
1099 	KNOTE_LOCKED(&filp->f_selinfo.si_note, 1);
1100 	spin_unlock(&filp->f_kqlock);
1101 }
1102 
1103 static void
1104 linux_file_kqfilter_detach(struct knote *kn)
1105 {
1106 	struct linux_file *filp = kn->kn_hook;
1107 
1108 	spin_lock(&filp->f_kqlock);
1109 	knlist_remove(&filp->f_selinfo.si_note, kn, 1);
1110 	spin_unlock(&filp->f_kqlock);
1111 }
1112 
1113 static int
1114 linux_file_kqfilter_read_event(struct knote *kn, long hint)
1115 {
1116 	struct linux_file *filp = kn->kn_hook;
1117 
1118 	mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1119 
1120 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0);
1121 }
1122 
1123 static int
1124 linux_file_kqfilter_write_event(struct knote *kn, long hint)
1125 {
1126 	struct linux_file *filp = kn->kn_hook;
1127 
1128 	mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1129 
1130 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0);
1131 }
1132 
1133 static struct filterops linux_dev_kqfiltops_read = {
1134 	.f_isfd = 1,
1135 	.f_detach = linux_file_kqfilter_detach,
1136 	.f_event = linux_file_kqfilter_read_event,
1137 };
1138 
1139 static struct filterops linux_dev_kqfiltops_write = {
1140 	.f_isfd = 1,
1141 	.f_detach = linux_file_kqfilter_detach,
1142 	.f_event = linux_file_kqfilter_write_event,
1143 };
1144 
1145 static void
1146 linux_file_kqfilter_poll(struct linux_file *filp, int kqflags)
1147 {
1148 	struct thread *td;
1149 	const struct file_operations *fop;
1150 	struct linux_cdev *ldev;
1151 	int temp;
1152 
1153 	if ((filp->f_kqflags & kqflags) == 0)
1154 		return;
1155 
1156 	td = curthread;
1157 
1158 	linux_get_fop(filp, &fop, &ldev);
1159 	/* get the latest polling state */
1160 	temp = OPW(filp->_file, td, fop->poll(filp, NULL));
1161 	linux_drop_fop(ldev);
1162 
1163 	spin_lock(&filp->f_kqlock);
1164 	/* clear kqflags */
1165 	filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ |
1166 	    LINUX_KQ_FLAG_NEED_WRITE);
1167 	/* update kqflags */
1168 	if ((temp & (POLLIN | POLLOUT)) != 0) {
1169 		if ((temp & POLLIN) != 0)
1170 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ;
1171 		if ((temp & POLLOUT) != 0)
1172 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE;
1173 
1174 		/* make sure the "knote" gets woken up */
1175 		KNOTE_LOCKED(&filp->f_selinfo.si_note, 0);
1176 	}
1177 	spin_unlock(&filp->f_kqlock);
1178 }
1179 
1180 static int
1181 linux_file_kqfilter(struct file *file, struct knote *kn)
1182 {
1183 	struct linux_file *filp;
1184 	struct thread *td;
1185 	int error;
1186 
1187 	td = curthread;
1188 	filp = (struct linux_file *)file->f_data;
1189 	filp->f_flags = file->f_flag;
1190 	if (filp->f_op->poll == NULL)
1191 		return (EINVAL);
1192 
1193 	spin_lock(&filp->f_kqlock);
1194 	switch (kn->kn_filter) {
1195 	case EVFILT_READ:
1196 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ;
1197 		kn->kn_fop = &linux_dev_kqfiltops_read;
1198 		kn->kn_hook = filp;
1199 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1200 		error = 0;
1201 		break;
1202 	case EVFILT_WRITE:
1203 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE;
1204 		kn->kn_fop = &linux_dev_kqfiltops_write;
1205 		kn->kn_hook = filp;
1206 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1207 		error = 0;
1208 		break;
1209 	default:
1210 		error = EINVAL;
1211 		break;
1212 	}
1213 	spin_unlock(&filp->f_kqlock);
1214 
1215 	if (error == 0) {
1216 		linux_set_current(td);
1217 
1218 		/* update kqfilter status, if any */
1219 		linux_file_kqfilter_poll(filp,
1220 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1221 	}
1222 	return (error);
1223 }
1224 
1225 static int
1226 linux_file_mmap_single(struct file *fp, const struct file_operations *fop,
1227     vm_ooffset_t *offset, vm_size_t size, struct vm_object **object,
1228     int nprot, struct thread *td)
1229 {
1230 	struct task_struct *task;
1231 	struct vm_area_struct *vmap;
1232 	struct mm_struct *mm;
1233 	struct linux_file *filp;
1234 	vm_memattr_t attr;
1235 	int error;
1236 
1237 	filp = (struct linux_file *)fp->f_data;
1238 	filp->f_flags = fp->f_flag;
1239 
1240 	if (fop->mmap == NULL)
1241 		return (EOPNOTSUPP);
1242 
1243 	linux_set_current(td);
1244 
1245 	/*
1246 	 * The same VM object might be shared by multiple processes
1247 	 * and the mm_struct is usually freed when a process exits.
1248 	 *
1249 	 * The atomic reference below makes sure the mm_struct is
1250 	 * available as long as the vmap is in the linux_vma_head.
1251 	 */
1252 	task = current;
1253 	mm = task->mm;
1254 	if (atomic_inc_not_zero(&mm->mm_users) == 0)
1255 		return (EINVAL);
1256 
1257 	vmap = kzalloc(sizeof(*vmap), GFP_KERNEL);
1258 	vmap->vm_start = 0;
1259 	vmap->vm_end = size;
1260 	vmap->vm_pgoff = *offset / PAGE_SIZE;
1261 	vmap->vm_pfn = 0;
1262 	vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL);
1263 	vmap->vm_ops = NULL;
1264 	vmap->vm_file = get_file(filp);
1265 	vmap->vm_mm = mm;
1266 
1267 	if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) {
1268 		error = linux_get_error(task, EINTR);
1269 	} else {
1270 		error = -OPW(fp, td, fop->mmap(filp, vmap));
1271 		error = linux_get_error(task, error);
1272 		up_write(&vmap->vm_mm->mmap_sem);
1273 	}
1274 
1275 	if (error != 0) {
1276 		linux_cdev_handle_free(vmap);
1277 		return (error);
1278 	}
1279 
1280 	attr = pgprot2cachemode(vmap->vm_page_prot);
1281 
1282 	if (vmap->vm_ops != NULL) {
1283 		struct vm_area_struct *ptr;
1284 		void *vm_private_data;
1285 		bool vm_no_fault;
1286 
1287 		if (vmap->vm_ops->open == NULL ||
1288 		    vmap->vm_ops->close == NULL ||
1289 		    vmap->vm_private_data == NULL) {
1290 			/* free allocated VM area struct */
1291 			linux_cdev_handle_free(vmap);
1292 			return (EINVAL);
1293 		}
1294 
1295 		vm_private_data = vmap->vm_private_data;
1296 
1297 		rw_wlock(&linux_vma_lock);
1298 		TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) {
1299 			if (ptr->vm_private_data == vm_private_data)
1300 				break;
1301 		}
1302 		/* check if there is an existing VM area struct */
1303 		if (ptr != NULL) {
1304 			/* check if the VM area structure is invalid */
1305 			if (ptr->vm_ops == NULL ||
1306 			    ptr->vm_ops->open == NULL ||
1307 			    ptr->vm_ops->close == NULL) {
1308 				error = ESTALE;
1309 				vm_no_fault = 1;
1310 			} else {
1311 				error = EEXIST;
1312 				vm_no_fault = (ptr->vm_ops->fault == NULL);
1313 			}
1314 		} else {
1315 			/* insert VM area structure into list */
1316 			TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry);
1317 			error = 0;
1318 			vm_no_fault = (vmap->vm_ops->fault == NULL);
1319 		}
1320 		rw_wunlock(&linux_vma_lock);
1321 
1322 		if (error != 0) {
1323 			/* free allocated VM area struct */
1324 			linux_cdev_handle_free(vmap);
1325 			/* check for stale VM area struct */
1326 			if (error != EEXIST)
1327 				return (error);
1328 		}
1329 
1330 		/* check if there is no fault handler */
1331 		if (vm_no_fault) {
1332 			*object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE,
1333 			    &linux_cdev_pager_ops[1], size, nprot, *offset,
1334 			    td->td_ucred);
1335 		} else {
1336 			*object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE,
1337 			    &linux_cdev_pager_ops[0], size, nprot, *offset,
1338 			    td->td_ucred);
1339 		}
1340 
1341 		/* check if allocating the VM object failed */
1342 		if (*object == NULL) {
1343 			if (error == 0) {
1344 				/* remove VM area struct from list */
1345 				linux_cdev_handle_remove(vmap);
1346 				/* free allocated VM area struct */
1347 				linux_cdev_handle_free(vmap);
1348 			}
1349 			return (EINVAL);
1350 		}
1351 	} else {
1352 		struct sglist *sg;
1353 
1354 		sg = sglist_alloc(1, M_WAITOK);
1355 		sglist_append_phys(sg,
1356 		    (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len);
1357 
1358 		*object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len,
1359 		    nprot, 0, td->td_ucred);
1360 
1361 		linux_cdev_handle_free(vmap);
1362 
1363 		if (*object == NULL) {
1364 			sglist_free(sg);
1365 			return (EINVAL);
1366 		}
1367 	}
1368 
1369 	if (attr != VM_MEMATTR_DEFAULT) {
1370 		VM_OBJECT_WLOCK(*object);
1371 		vm_object_set_memattr(*object, attr);
1372 		VM_OBJECT_WUNLOCK(*object);
1373 	}
1374 	*offset = 0;
1375 	return (0);
1376 }
1377 
1378 struct cdevsw linuxcdevsw = {
1379 	.d_version = D_VERSION,
1380 	.d_fdopen = linux_dev_fdopen,
1381 	.d_name = "lkpidev",
1382 };
1383 
1384 static int
1385 linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
1386     int flags, struct thread *td)
1387 {
1388 	struct linux_file *filp;
1389 	const struct file_operations *fop;
1390 	struct linux_cdev *ldev;
1391 	ssize_t bytes;
1392 	int error;
1393 
1394 	error = 0;
1395 	filp = (struct linux_file *)file->f_data;
1396 	filp->f_flags = file->f_flag;
1397 	/* XXX no support for I/O vectors currently */
1398 	if (uio->uio_iovcnt != 1)
1399 		return (EOPNOTSUPP);
1400 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1401 		return (EINVAL);
1402 	linux_set_current(td);
1403 	linux_get_fop(filp, &fop, &ldev);
1404 	if (fop->read != NULL) {
1405 		bytes = OPW(file, td, fop->read(filp,
1406 		    uio->uio_iov->iov_base,
1407 		    uio->uio_iov->iov_len, &uio->uio_offset));
1408 		if (bytes >= 0) {
1409 			uio->uio_iov->iov_base =
1410 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1411 			uio->uio_iov->iov_len -= bytes;
1412 			uio->uio_resid -= bytes;
1413 		} else {
1414 			error = linux_get_error(current, -bytes);
1415 		}
1416 	} else
1417 		error = ENXIO;
1418 
1419 	/* update kqfilter status, if any */
1420 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ);
1421 	linux_drop_fop(ldev);
1422 
1423 	return (error);
1424 }
1425 
1426 static int
1427 linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred,
1428     int flags, struct thread *td)
1429 {
1430 	struct linux_file *filp;
1431 	const struct file_operations *fop;
1432 	struct linux_cdev *ldev;
1433 	ssize_t bytes;
1434 	int error;
1435 
1436 	filp = (struct linux_file *)file->f_data;
1437 	filp->f_flags = file->f_flag;
1438 	/* XXX no support for I/O vectors currently */
1439 	if (uio->uio_iovcnt != 1)
1440 		return (EOPNOTSUPP);
1441 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1442 		return (EINVAL);
1443 	linux_set_current(td);
1444 	linux_get_fop(filp, &fop, &ldev);
1445 	if (fop->write != NULL) {
1446 		bytes = OPW(file, td, fop->write(filp,
1447 		    uio->uio_iov->iov_base,
1448 		    uio->uio_iov->iov_len, &uio->uio_offset));
1449 		if (bytes >= 0) {
1450 			uio->uio_iov->iov_base =
1451 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1452 			uio->uio_iov->iov_len -= bytes;
1453 			uio->uio_resid -= bytes;
1454 			error = 0;
1455 		} else {
1456 			error = linux_get_error(current, -bytes);
1457 		}
1458 	} else
1459 		error = ENXIO;
1460 
1461 	/* update kqfilter status, if any */
1462 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE);
1463 
1464 	linux_drop_fop(ldev);
1465 
1466 	return (error);
1467 }
1468 
1469 static int
1470 linux_file_poll(struct file *file, int events, struct ucred *active_cred,
1471     struct thread *td)
1472 {
1473 	struct linux_file *filp;
1474 	const struct file_operations *fop;
1475 	struct linux_cdev *ldev;
1476 	int revents;
1477 
1478 	filp = (struct linux_file *)file->f_data;
1479 	filp->f_flags = file->f_flag;
1480 	linux_set_current(td);
1481 	linux_get_fop(filp, &fop, &ldev);
1482 	if (fop->poll != NULL) {
1483 		revents = OPW(file, td, fop->poll(filp,
1484 		    LINUX_POLL_TABLE_NORMAL)) & events;
1485 	} else {
1486 		revents = 0;
1487 	}
1488 	linux_drop_fop(ldev);
1489 	return (revents);
1490 }
1491 
1492 static int
1493 linux_file_close(struct file *file, struct thread *td)
1494 {
1495 	struct linux_file *filp;
1496 	const struct file_operations *fop;
1497 	struct linux_cdev *ldev;
1498 	int error;
1499 
1500 	filp = (struct linux_file *)file->f_data;
1501 
1502 	KASSERT(file_count(filp) == 0,
1503 	    ("File refcount(%d) is not zero", file_count(filp)));
1504 
1505 	error = 0;
1506 	filp->f_flags = file->f_flag;
1507 	linux_set_current(td);
1508 	linux_poll_wait_dequeue(filp);
1509 	linux_get_fop(filp, &fop, &ldev);
1510 	if (fop->release != NULL)
1511 		error = -OPW(file, td, fop->release(filp->f_vnode, filp));
1512 	funsetown(&filp->f_sigio);
1513 	if (filp->f_vnode != NULL)
1514 		vdrop(filp->f_vnode);
1515 	linux_drop_fop(ldev);
1516 	if (filp->f_cdev != NULL)
1517 		linux_cdev_deref(filp->f_cdev);
1518 	kfree(filp);
1519 
1520 	return (error);
1521 }
1522 
1523 static int
1524 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
1525     struct thread *td)
1526 {
1527 	struct linux_file *filp;
1528 	const struct file_operations *fop;
1529 	struct linux_cdev *ldev;
1530 	int error;
1531 
1532 	error = 0;
1533 	filp = (struct linux_file *)fp->f_data;
1534 	filp->f_flags = fp->f_flag;
1535 	linux_get_fop(filp, &fop, &ldev);
1536 
1537 	linux_set_current(td);
1538 	switch (cmd) {
1539 	case FIONBIO:
1540 		break;
1541 	case FIOASYNC:
1542 		if (fop->fasync == NULL)
1543 			break;
1544 		error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC));
1545 		break;
1546 	case FIOSETOWN:
1547 		error = fsetown(*(int *)data, &filp->f_sigio);
1548 		if (error == 0) {
1549 			if (fop->fasync == NULL)
1550 				break;
1551 			error = -OPW(fp, td, fop->fasync(0, filp,
1552 			    fp->f_flag & FASYNC));
1553 		}
1554 		break;
1555 	case FIOGETOWN:
1556 		*(int *)data = fgetown(&filp->f_sigio);
1557 		break;
1558 	default:
1559 		error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td);
1560 		break;
1561 	}
1562 	linux_drop_fop(ldev);
1563 	return (error);
1564 }
1565 
1566 static int
1567 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1568     vm_prot_t *maxprotp, int *flagsp, struct file *fp,
1569     vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp)
1570 {
1571 	/*
1572 	 * Character devices do not provide private mappings
1573 	 * of any kind:
1574 	 */
1575 	if ((*maxprotp & VM_PROT_WRITE) == 0 &&
1576 	    (prot & VM_PROT_WRITE) != 0)
1577 		return (EACCES);
1578 	if ((*flagsp & (MAP_PRIVATE | MAP_COPY)) != 0)
1579 		return (EINVAL);
1580 
1581 	return (linux_file_mmap_single(fp, fop, foff, objsize, objp,
1582 	    (int)prot, td));
1583 }
1584 
1585 static int
1586 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1587     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1588     struct thread *td)
1589 {
1590 	struct linux_file *filp;
1591 	const struct file_operations *fop;
1592 	struct linux_cdev *ldev;
1593 	struct mount *mp;
1594 	struct vnode *vp;
1595 	vm_object_t object;
1596 	vm_prot_t maxprot;
1597 	int error;
1598 
1599 	filp = (struct linux_file *)fp->f_data;
1600 
1601 	vp = filp->f_vnode;
1602 	if (vp == NULL)
1603 		return (EOPNOTSUPP);
1604 
1605 	/*
1606 	 * Ensure that file and memory protections are
1607 	 * compatible.
1608 	 */
1609 	mp = vp->v_mount;
1610 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1611 		maxprot = VM_PROT_NONE;
1612 		if ((prot & VM_PROT_EXECUTE) != 0)
1613 			return (EACCES);
1614 	} else
1615 		maxprot = VM_PROT_EXECUTE;
1616 	if ((fp->f_flag & FREAD) != 0)
1617 		maxprot |= VM_PROT_READ;
1618 	else if ((prot & VM_PROT_READ) != 0)
1619 		return (EACCES);
1620 
1621 	/*
1622 	 * If we are sharing potential changes via MAP_SHARED and we
1623 	 * are trying to get write permission although we opened it
1624 	 * without asking for it, bail out.
1625 	 *
1626 	 * Note that most character devices always share mappings.
1627 	 *
1628 	 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE
1629 	 * requests rather than doing it here.
1630 	 */
1631 	if ((flags & MAP_SHARED) != 0) {
1632 		if ((fp->f_flag & FWRITE) != 0)
1633 			maxprot |= VM_PROT_WRITE;
1634 		else if ((prot & VM_PROT_WRITE) != 0)
1635 			return (EACCES);
1636 	}
1637 	maxprot &= cap_maxprot;
1638 
1639 	linux_get_fop(filp, &fop, &ldev);
1640 	error = linux_file_mmap_sub(td, size, prot, &maxprot, &flags, fp,
1641 	    &foff, fop, &object);
1642 	if (error != 0)
1643 		goto out;
1644 
1645 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1646 	    foff, FALSE, td);
1647 	if (error != 0)
1648 		vm_object_deallocate(object);
1649 out:
1650 	linux_drop_fop(ldev);
1651 	return (error);
1652 }
1653 
1654 static int
1655 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred,
1656     struct thread *td)
1657 {
1658 	struct linux_file *filp;
1659 	struct vnode *vp;
1660 	int error;
1661 
1662 	filp = (struct linux_file *)fp->f_data;
1663 	if (filp->f_vnode == NULL)
1664 		return (EOPNOTSUPP);
1665 
1666 	vp = filp->f_vnode;
1667 
1668 	vn_lock(vp, LK_SHARED | LK_RETRY);
1669 	error = vn_stat(vp, sb, td->td_ucred, NOCRED, td);
1670 	VOP_UNLOCK(vp, 0);
1671 
1672 	return (error);
1673 }
1674 
1675 static int
1676 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1677     struct filedesc *fdp)
1678 {
1679 	struct linux_file *filp;
1680 	struct vnode *vp;
1681 	int error;
1682 
1683 	filp = fp->f_data;
1684 	vp = filp->f_vnode;
1685 	if (vp == NULL) {
1686 		error = 0;
1687 		kif->kf_type = KF_TYPE_DEV;
1688 	} else {
1689 		vref(vp);
1690 		FILEDESC_SUNLOCK(fdp);
1691 		error = vn_fill_kinfo_vnode(vp, kif);
1692 		vrele(vp);
1693 		kif->kf_type = KF_TYPE_VNODE;
1694 		FILEDESC_SLOCK(fdp);
1695 	}
1696 	return (error);
1697 }
1698 
1699 unsigned int
1700 linux_iminor(struct inode *inode)
1701 {
1702 	struct linux_cdev *ldev;
1703 
1704 	if (inode == NULL || inode->v_rdev == NULL ||
1705 	    inode->v_rdev->si_devsw != &linuxcdevsw)
1706 		return (-1U);
1707 	ldev = inode->v_rdev->si_drv1;
1708 	if (ldev == NULL)
1709 		return (-1U);
1710 
1711 	return (minor(ldev->dev));
1712 }
1713 
1714 struct fileops linuxfileops = {
1715 	.fo_read = linux_file_read,
1716 	.fo_write = linux_file_write,
1717 	.fo_truncate = invfo_truncate,
1718 	.fo_kqfilter = linux_file_kqfilter,
1719 	.fo_stat = linux_file_stat,
1720 	.fo_fill_kinfo = linux_file_fill_kinfo,
1721 	.fo_poll = linux_file_poll,
1722 	.fo_close = linux_file_close,
1723 	.fo_ioctl = linux_file_ioctl,
1724 	.fo_mmap = linux_file_mmap,
1725 	.fo_chmod = invfo_chmod,
1726 	.fo_chown = invfo_chown,
1727 	.fo_sendfile = invfo_sendfile,
1728 	.fo_flags = DFLAG_PASSABLE,
1729 };
1730 
1731 /*
1732  * Hash of vmmap addresses.  This is infrequently accessed and does not
1733  * need to be particularly large.  This is done because we must store the
1734  * caller's idea of the map size to properly unmap.
1735  */
1736 struct vmmap {
1737 	LIST_ENTRY(vmmap)	vm_next;
1738 	void 			*vm_addr;
1739 	unsigned long		vm_size;
1740 };
1741 
1742 struct vmmaphd {
1743 	struct vmmap *lh_first;
1744 };
1745 #define	VMMAP_HASH_SIZE	64
1746 #define	VMMAP_HASH_MASK	(VMMAP_HASH_SIZE - 1)
1747 #define	VM_HASH(addr)	((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK
1748 static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE];
1749 static struct mtx vmmaplock;
1750 
1751 static void
1752 vmmap_add(void *addr, unsigned long size)
1753 {
1754 	struct vmmap *vmmap;
1755 
1756 	vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL);
1757 	mtx_lock(&vmmaplock);
1758 	vmmap->vm_size = size;
1759 	vmmap->vm_addr = addr;
1760 	LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next);
1761 	mtx_unlock(&vmmaplock);
1762 }
1763 
1764 static struct vmmap *
1765 vmmap_remove(void *addr)
1766 {
1767 	struct vmmap *vmmap;
1768 
1769 	mtx_lock(&vmmaplock);
1770 	LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next)
1771 		if (vmmap->vm_addr == addr)
1772 			break;
1773 	if (vmmap)
1774 		LIST_REMOVE(vmmap, vm_next);
1775 	mtx_unlock(&vmmaplock);
1776 
1777 	return (vmmap);
1778 }
1779 
1780 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__)
1781 void *
1782 _ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr)
1783 {
1784 	void *addr;
1785 
1786 	addr = pmap_mapdev_attr(phys_addr, size, attr);
1787 	if (addr == NULL)
1788 		return (NULL);
1789 	vmmap_add(addr, size);
1790 
1791 	return (addr);
1792 }
1793 #endif
1794 
1795 void
1796 iounmap(void *addr)
1797 {
1798 	struct vmmap *vmmap;
1799 
1800 	vmmap = vmmap_remove(addr);
1801 	if (vmmap == NULL)
1802 		return;
1803 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__)
1804 	pmap_unmapdev((vm_offset_t)addr, vmmap->vm_size);
1805 #endif
1806 	kfree(vmmap);
1807 }
1808 
1809 
1810 void *
1811 vmap(struct page **pages, unsigned int count, unsigned long flags, int prot)
1812 {
1813 	vm_offset_t off;
1814 	size_t size;
1815 
1816 	size = count * PAGE_SIZE;
1817 	off = kva_alloc(size);
1818 	if (off == 0)
1819 		return (NULL);
1820 	vmmap_add((void *)off, size);
1821 	pmap_qenter(off, pages, count);
1822 
1823 	return ((void *)off);
1824 }
1825 
1826 void
1827 vunmap(void *addr)
1828 {
1829 	struct vmmap *vmmap;
1830 
1831 	vmmap = vmmap_remove(addr);
1832 	if (vmmap == NULL)
1833 		return;
1834 	pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE);
1835 	kva_free((vm_offset_t)addr, vmmap->vm_size);
1836 	kfree(vmmap);
1837 }
1838 
1839 char *
1840 kvasprintf(gfp_t gfp, const char *fmt, va_list ap)
1841 {
1842 	unsigned int len;
1843 	char *p;
1844 	va_list aq;
1845 
1846 	va_copy(aq, ap);
1847 	len = vsnprintf(NULL, 0, fmt, aq);
1848 	va_end(aq);
1849 
1850 	p = kmalloc(len + 1, gfp);
1851 	if (p != NULL)
1852 		vsnprintf(p, len + 1, fmt, ap);
1853 
1854 	return (p);
1855 }
1856 
1857 char *
1858 kasprintf(gfp_t gfp, const char *fmt, ...)
1859 {
1860 	va_list ap;
1861 	char *p;
1862 
1863 	va_start(ap, fmt);
1864 	p = kvasprintf(gfp, fmt, ap);
1865 	va_end(ap);
1866 
1867 	return (p);
1868 }
1869 
1870 static void
1871 linux_timer_callback_wrapper(void *context)
1872 {
1873 	struct timer_list *timer;
1874 
1875 	linux_set_current(curthread);
1876 
1877 	timer = context;
1878 	timer->function(timer->data);
1879 }
1880 
1881 void
1882 mod_timer(struct timer_list *timer, int expires)
1883 {
1884 
1885 	timer->expires = expires;
1886 	callout_reset(&timer->callout,
1887 	    linux_timer_jiffies_until(expires),
1888 	    &linux_timer_callback_wrapper, timer);
1889 }
1890 
1891 void
1892 add_timer(struct timer_list *timer)
1893 {
1894 
1895 	callout_reset(&timer->callout,
1896 	    linux_timer_jiffies_until(timer->expires),
1897 	    &linux_timer_callback_wrapper, timer);
1898 }
1899 
1900 void
1901 add_timer_on(struct timer_list *timer, int cpu)
1902 {
1903 
1904 	callout_reset_on(&timer->callout,
1905 	    linux_timer_jiffies_until(timer->expires),
1906 	    &linux_timer_callback_wrapper, timer, cpu);
1907 }
1908 
1909 int
1910 del_timer(struct timer_list *timer)
1911 {
1912 
1913 	if (callout_stop(&(timer)->callout) == -1)
1914 		return (0);
1915 	return (1);
1916 }
1917 
1918 static void
1919 linux_timer_init(void *arg)
1920 {
1921 
1922 	/*
1923 	 * Compute an internal HZ value which can divide 2**32 to
1924 	 * avoid timer rounding problems when the tick value wraps
1925 	 * around 2**32:
1926 	 */
1927 	linux_timer_hz_mask = 1;
1928 	while (linux_timer_hz_mask < (unsigned long)hz)
1929 		linux_timer_hz_mask *= 2;
1930 	linux_timer_hz_mask--;
1931 }
1932 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL);
1933 
1934 void
1935 linux_complete_common(struct completion *c, int all)
1936 {
1937 	int wakeup_swapper;
1938 
1939 	sleepq_lock(c);
1940 	if (all) {
1941 		c->done = UINT_MAX;
1942 		wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0);
1943 	} else {
1944 		if (c->done != UINT_MAX)
1945 			c->done++;
1946 		wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0);
1947 	}
1948 	sleepq_release(c);
1949 	if (wakeup_swapper)
1950 		kick_proc0();
1951 }
1952 
1953 /*
1954  * Indefinite wait for done != 0 with or without signals.
1955  */
1956 int
1957 linux_wait_for_common(struct completion *c, int flags)
1958 {
1959 	struct task_struct *task;
1960 	int error;
1961 
1962 	if (SCHEDULER_STOPPED())
1963 		return (0);
1964 
1965 	task = current;
1966 
1967 	if (flags != 0)
1968 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
1969 	else
1970 		flags = SLEEPQ_SLEEP;
1971 	error = 0;
1972 	for (;;) {
1973 		sleepq_lock(c);
1974 		if (c->done)
1975 			break;
1976 		sleepq_add(c, NULL, "completion", flags, 0);
1977 		if (flags & SLEEPQ_INTERRUPTIBLE) {
1978 			DROP_GIANT();
1979 			error = -sleepq_wait_sig(c, 0);
1980 			PICKUP_GIANT();
1981 			if (error != 0) {
1982 				linux_schedule_save_interrupt_value(task, error);
1983 				error = -ERESTARTSYS;
1984 				goto intr;
1985 			}
1986 		} else {
1987 			DROP_GIANT();
1988 			sleepq_wait(c, 0);
1989 			PICKUP_GIANT();
1990 		}
1991 	}
1992 	if (c->done != UINT_MAX)
1993 		c->done--;
1994 	sleepq_release(c);
1995 
1996 intr:
1997 	return (error);
1998 }
1999 
2000 /*
2001  * Time limited wait for done != 0 with or without signals.
2002  */
2003 int
2004 linux_wait_for_timeout_common(struct completion *c, int timeout, int flags)
2005 {
2006 	struct task_struct *task;
2007 	int end = jiffies + timeout;
2008 	int error;
2009 
2010 	if (SCHEDULER_STOPPED())
2011 		return (0);
2012 
2013 	task = current;
2014 
2015 	if (flags != 0)
2016 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2017 	else
2018 		flags = SLEEPQ_SLEEP;
2019 
2020 	for (;;) {
2021 		sleepq_lock(c);
2022 		if (c->done)
2023 			break;
2024 		sleepq_add(c, NULL, "completion", flags, 0);
2025 		sleepq_set_timeout(c, linux_timer_jiffies_until(end));
2026 
2027 		DROP_GIANT();
2028 		if (flags & SLEEPQ_INTERRUPTIBLE)
2029 			error = -sleepq_timedwait_sig(c, 0);
2030 		else
2031 			error = -sleepq_timedwait(c, 0);
2032 		PICKUP_GIANT();
2033 
2034 		if (error != 0) {
2035 			/* check for timeout */
2036 			if (error == -EWOULDBLOCK) {
2037 				error = 0;	/* timeout */
2038 			} else {
2039 				/* signal happened */
2040 				linux_schedule_save_interrupt_value(task, error);
2041 				error = -ERESTARTSYS;
2042 			}
2043 			goto done;
2044 		}
2045 	}
2046 	if (c->done != UINT_MAX)
2047 		c->done--;
2048 	sleepq_release(c);
2049 
2050 	/* return how many jiffies are left */
2051 	error = linux_timer_jiffies_until(end);
2052 done:
2053 	return (error);
2054 }
2055 
2056 int
2057 linux_try_wait_for_completion(struct completion *c)
2058 {
2059 	int isdone;
2060 
2061 	sleepq_lock(c);
2062 	isdone = (c->done != 0);
2063 	if (c->done != 0 && c->done != UINT_MAX)
2064 		c->done--;
2065 	sleepq_release(c);
2066 	return (isdone);
2067 }
2068 
2069 int
2070 linux_completion_done(struct completion *c)
2071 {
2072 	int isdone;
2073 
2074 	sleepq_lock(c);
2075 	isdone = (c->done != 0);
2076 	sleepq_release(c);
2077 	return (isdone);
2078 }
2079 
2080 static void
2081 linux_cdev_deref(struct linux_cdev *ldev)
2082 {
2083 
2084 	if (refcount_release(&ldev->refs))
2085 		kfree(ldev);
2086 }
2087 
2088 static void
2089 linux_cdev_release(struct kobject *kobj)
2090 {
2091 	struct linux_cdev *cdev;
2092 	struct kobject *parent;
2093 
2094 	cdev = container_of(kobj, struct linux_cdev, kobj);
2095 	parent = kobj->parent;
2096 	linux_destroy_dev(cdev);
2097 	linux_cdev_deref(cdev);
2098 	kobject_put(parent);
2099 }
2100 
2101 static void
2102 linux_cdev_static_release(struct kobject *kobj)
2103 {
2104 	struct linux_cdev *cdev;
2105 	struct kobject *parent;
2106 
2107 	cdev = container_of(kobj, struct linux_cdev, kobj);
2108 	parent = kobj->parent;
2109 	linux_destroy_dev(cdev);
2110 	kobject_put(parent);
2111 }
2112 
2113 void
2114 linux_destroy_dev(struct linux_cdev *ldev)
2115 {
2116 
2117 	if (ldev->cdev == NULL)
2118 		return;
2119 
2120 	MPASS((ldev->siref & LDEV_SI_DTR) == 0);
2121 	atomic_set_int(&ldev->siref, LDEV_SI_DTR);
2122 	while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0)
2123 		pause("ldevdtr", hz / 4);
2124 
2125 	destroy_dev(ldev->cdev);
2126 	ldev->cdev = NULL;
2127 }
2128 
2129 const struct kobj_type linux_cdev_ktype = {
2130 	.release = linux_cdev_release,
2131 };
2132 
2133 const struct kobj_type linux_cdev_static_ktype = {
2134 	.release = linux_cdev_static_release,
2135 };
2136 
2137 static void
2138 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate)
2139 {
2140 	struct notifier_block *nb;
2141 
2142 	nb = arg;
2143 	if (linkstate == LINK_STATE_UP)
2144 		nb->notifier_call(nb, NETDEV_UP, ifp);
2145 	else
2146 		nb->notifier_call(nb, NETDEV_DOWN, ifp);
2147 }
2148 
2149 static void
2150 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp)
2151 {
2152 	struct notifier_block *nb;
2153 
2154 	nb = arg;
2155 	nb->notifier_call(nb, NETDEV_REGISTER, ifp);
2156 }
2157 
2158 static void
2159 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp)
2160 {
2161 	struct notifier_block *nb;
2162 
2163 	nb = arg;
2164 	nb->notifier_call(nb, NETDEV_UNREGISTER, ifp);
2165 }
2166 
2167 static void
2168 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp)
2169 {
2170 	struct notifier_block *nb;
2171 
2172 	nb = arg;
2173 	nb->notifier_call(nb, NETDEV_CHANGEADDR, ifp);
2174 }
2175 
2176 static void
2177 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp)
2178 {
2179 	struct notifier_block *nb;
2180 
2181 	nb = arg;
2182 	nb->notifier_call(nb, NETDEV_CHANGEIFADDR, ifp);
2183 }
2184 
2185 int
2186 register_netdevice_notifier(struct notifier_block *nb)
2187 {
2188 
2189 	nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER(
2190 	    ifnet_link_event, linux_handle_ifnet_link_event, nb, 0);
2191 	nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER(
2192 	    ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0);
2193 	nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER(
2194 	    ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0);
2195 	nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER(
2196 	    iflladdr_event, linux_handle_iflladdr_event, nb, 0);
2197 
2198 	return (0);
2199 }
2200 
2201 int
2202 register_inetaddr_notifier(struct notifier_block *nb)
2203 {
2204 
2205 	nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER(
2206 	    ifaddr_event, linux_handle_ifaddr_event, nb, 0);
2207 	return (0);
2208 }
2209 
2210 int
2211 unregister_netdevice_notifier(struct notifier_block *nb)
2212 {
2213 
2214 	EVENTHANDLER_DEREGISTER(ifnet_link_event,
2215 	    nb->tags[NETDEV_UP]);
2216 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event,
2217 	    nb->tags[NETDEV_REGISTER]);
2218 	EVENTHANDLER_DEREGISTER(ifnet_departure_event,
2219 	    nb->tags[NETDEV_UNREGISTER]);
2220 	EVENTHANDLER_DEREGISTER(iflladdr_event,
2221 	    nb->tags[NETDEV_CHANGEADDR]);
2222 
2223 	return (0);
2224 }
2225 
2226 int
2227 unregister_inetaddr_notifier(struct notifier_block *nb)
2228 {
2229 
2230 	EVENTHANDLER_DEREGISTER(ifaddr_event,
2231 	    nb->tags[NETDEV_CHANGEIFADDR]);
2232 
2233 	return (0);
2234 }
2235 
2236 struct list_sort_thunk {
2237 	int (*cmp)(void *, struct list_head *, struct list_head *);
2238 	void *priv;
2239 };
2240 
2241 static inline int
2242 linux_le_cmp(void *priv, const void *d1, const void *d2)
2243 {
2244 	struct list_head *le1, *le2;
2245 	struct list_sort_thunk *thunk;
2246 
2247 	thunk = priv;
2248 	le1 = *(__DECONST(struct list_head **, d1));
2249 	le2 = *(__DECONST(struct list_head **, d2));
2250 	return ((thunk->cmp)(thunk->priv, le1, le2));
2251 }
2252 
2253 void
2254 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv,
2255     struct list_head *a, struct list_head *b))
2256 {
2257 	struct list_sort_thunk thunk;
2258 	struct list_head **ar, *le;
2259 	size_t count, i;
2260 
2261 	count = 0;
2262 	list_for_each(le, head)
2263 		count++;
2264 	ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK);
2265 	i = 0;
2266 	list_for_each(le, head)
2267 		ar[i++] = le;
2268 	thunk.cmp = cmp;
2269 	thunk.priv = priv;
2270 	qsort_r(ar, count, sizeof(struct list_head *), &thunk, linux_le_cmp);
2271 	INIT_LIST_HEAD(head);
2272 	for (i = 0; i < count; i++)
2273 		list_add_tail(ar[i], head);
2274 	free(ar, M_KMALLOC);
2275 }
2276 
2277 void
2278 linux_irq_handler(void *ent)
2279 {
2280 	struct irq_ent *irqe;
2281 
2282 	linux_set_current(curthread);
2283 
2284 	irqe = ent;
2285 	irqe->handler(irqe->irq, irqe->arg);
2286 }
2287 
2288 #if defined(__i386__) || defined(__amd64__)
2289 int
2290 linux_wbinvd_on_all_cpus(void)
2291 {
2292 
2293 	pmap_invalidate_cache();
2294 	return (0);
2295 }
2296 #endif
2297 
2298 int
2299 linux_on_each_cpu(void callback(void *), void *data)
2300 {
2301 
2302 	smp_rendezvous(smp_no_rendezvous_barrier, callback,
2303 	    smp_no_rendezvous_barrier, data);
2304 	return (0);
2305 }
2306 
2307 int
2308 linux_in_atomic(void)
2309 {
2310 
2311 	return ((curthread->td_pflags & TDP_NOFAULTING) != 0);
2312 }
2313 
2314 struct linux_cdev *
2315 linux_find_cdev(const char *name, unsigned major, unsigned minor)
2316 {
2317 	dev_t dev = MKDEV(major, minor);
2318 	struct cdev *cdev;
2319 
2320 	dev_lock();
2321 	LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) {
2322 		struct linux_cdev *ldev = cdev->si_drv1;
2323 		if (ldev->dev == dev &&
2324 		    strcmp(kobject_name(&ldev->kobj), name) == 0) {
2325 			break;
2326 		}
2327 	}
2328 	dev_unlock();
2329 
2330 	return (cdev != NULL ? cdev->si_drv1 : NULL);
2331 }
2332 
2333 int
2334 __register_chrdev(unsigned int major, unsigned int baseminor,
2335     unsigned int count, const char *name,
2336     const struct file_operations *fops)
2337 {
2338 	struct linux_cdev *cdev;
2339 	int ret = 0;
2340 	int i;
2341 
2342 	for (i = baseminor; i < baseminor + count; i++) {
2343 		cdev = cdev_alloc();
2344 		cdev->ops = fops;
2345 		kobject_set_name(&cdev->kobj, name);
2346 
2347 		ret = cdev_add(cdev, makedev(major, i), 1);
2348 		if (ret != 0)
2349 			break;
2350 	}
2351 	return (ret);
2352 }
2353 
2354 int
2355 __register_chrdev_p(unsigned int major, unsigned int baseminor,
2356     unsigned int count, const char *name,
2357     const struct file_operations *fops, uid_t uid,
2358     gid_t gid, int mode)
2359 {
2360 	struct linux_cdev *cdev;
2361 	int ret = 0;
2362 	int i;
2363 
2364 	for (i = baseminor; i < baseminor + count; i++) {
2365 		cdev = cdev_alloc();
2366 		cdev->ops = fops;
2367 		kobject_set_name(&cdev->kobj, name);
2368 
2369 		ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode);
2370 		if (ret != 0)
2371 			break;
2372 	}
2373 	return (ret);
2374 }
2375 
2376 void
2377 __unregister_chrdev(unsigned int major, unsigned int baseminor,
2378     unsigned int count, const char *name)
2379 {
2380 	struct linux_cdev *cdevp;
2381 	int i;
2382 
2383 	for (i = baseminor; i < baseminor + count; i++) {
2384 		cdevp = linux_find_cdev(name, major, i);
2385 		if (cdevp != NULL)
2386 			cdev_del(cdevp);
2387 	}
2388 }
2389 
2390 void
2391 linux_dump_stack(void)
2392 {
2393 #ifdef STACK
2394 	struct stack st;
2395 
2396 	stack_zero(&st);
2397 	stack_save(&st);
2398 	stack_print(&st);
2399 #endif
2400 }
2401 
2402 #if defined(__i386__) || defined(__amd64__)
2403 bool linux_cpu_has_clflush;
2404 #endif
2405 
2406 static void
2407 linux_compat_init(void *arg)
2408 {
2409 	struct sysctl_oid *rootoid;
2410 	int i;
2411 
2412 #if defined(__i386__) || defined(__amd64__)
2413 	linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH);
2414 #endif
2415 	rw_init(&linux_vma_lock, "lkpi-vma-lock");
2416 
2417 	rootoid = SYSCTL_ADD_ROOT_NODE(NULL,
2418 	    OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys");
2419 	kobject_init(&linux_class_root, &linux_class_ktype);
2420 	kobject_set_name(&linux_class_root, "class");
2421 	linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid),
2422 	    OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class");
2423 	kobject_init(&linux_root_device.kobj, &linux_dev_ktype);
2424 	kobject_set_name(&linux_root_device.kobj, "device");
2425 	linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL,
2426 	    SYSCTL_CHILDREN(rootoid), OID_AUTO, "device", CTLFLAG_RD, NULL,
2427 	    "device");
2428 	linux_root_device.bsddev = root_bus;
2429 	linux_class_misc.name = "misc";
2430 	class_register(&linux_class_misc);
2431 	INIT_LIST_HEAD(&pci_drivers);
2432 	INIT_LIST_HEAD(&pci_devices);
2433 	spin_lock_init(&pci_lock);
2434 	mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF);
2435 	for (i = 0; i < VMMAP_HASH_SIZE; i++)
2436 		LIST_INIT(&vmmaphead[i]);
2437 }
2438 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL);
2439 
2440 static void
2441 linux_compat_uninit(void *arg)
2442 {
2443 	linux_kobject_kfree_name(&linux_class_root);
2444 	linux_kobject_kfree_name(&linux_root_device.kobj);
2445 	linux_kobject_kfree_name(&linux_class_misc.kobj);
2446 
2447 	mtx_destroy(&vmmaplock);
2448 	spin_lock_destroy(&pci_lock);
2449 	rw_destroy(&linux_vma_lock);
2450 }
2451 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL);
2452 
2453 /*
2454  * NOTE: Linux frequently uses "unsigned long" for pointer to integer
2455  * conversion and vice versa, where in FreeBSD "uintptr_t" would be
2456  * used. Assert these types have the same size, else some parts of the
2457  * LinuxKPI may not work like expected:
2458  */
2459 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t));
2460