xref: /freebsd/sys/compat/linuxkpi/common/src/linux_compat.c (revision 5fad251f05634e3220d36d51f624d77b41bc01c1)
1 /*-
2  * Copyright (c) 2010 Isilon Systems, Inc.
3  * Copyright (c) 2010 iX Systems, Inc.
4  * Copyright (c) 2010 Panasas, Inc.
5  * Copyright (c) 2013-2021 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 #include "opt_global.h"
32 #include "opt_stack.h"
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/malloc.h>
37 #include <sys/kernel.h>
38 #include <sys/sysctl.h>
39 #include <sys/proc.h>
40 #include <sys/sglist.h>
41 #include <sys/sleepqueue.h>
42 #include <sys/refcount.h>
43 #include <sys/lock.h>
44 #include <sys/mutex.h>
45 #include <sys/bus.h>
46 #include <sys/eventhandler.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/stdarg.h>
54 #include <sys/syscall.h>
55 #include <sys/sysent.h>
56 #include <sys/time.h>
57 #include <sys/user.h>
58 
59 #include <vm/vm.h>
60 #include <vm/pmap.h>
61 #include <vm/vm_object.h>
62 #include <vm/vm_page.h>
63 #include <vm/vm_pager.h>
64 #include <vm/vm_radix.h>
65 
66 #if defined(__i386__) || defined(__amd64__)
67 #include <machine/cputypes.h>
68 #include <machine/md_var.h>
69 #endif
70 
71 #include <linux/kobject.h>
72 #include <linux/cpu.h>
73 #include <linux/device.h>
74 #include <linux/slab.h>
75 #include <linux/module.h>
76 #include <linux/moduleparam.h>
77 #include <linux/cdev.h>
78 #include <linux/file.h>
79 #include <linux/fs.h>
80 #include <linux/sysfs.h>
81 #include <linux/mm.h>
82 #include <linux/io.h>
83 #include <linux/vmalloc.h>
84 #include <linux/netdevice.h>
85 #include <linux/timer.h>
86 #include <linux/interrupt.h>
87 #include <linux/uaccess.h>
88 #include <linux/utsname.h>
89 #include <linux/list.h>
90 #include <linux/kthread.h>
91 #include <linux/kernel.h>
92 #include <linux/compat.h>
93 #include <linux/io-mapping.h>
94 #include <linux/poll.h>
95 #include <linux/smp.h>
96 #include <linux/wait_bit.h>
97 #include <linux/rcupdate.h>
98 #include <linux/interval_tree.h>
99 #include <linux/interval_tree_generic.h>
100 #include <linux/printk.h>
101 #include <linux/seq_file.h>
102 #include <linux/uuid.h>
103 
104 #if defined(__i386__) || defined(__amd64__)
105 #include <asm/smp.h>
106 #include <asm/processor.h>
107 #endif
108 
109 #include <xen/xen.h>
110 #ifdef XENHVM
111 #undef xen_pv_domain
112 #undef xen_initial_domain
113 /* xen/xen-os.h redefines __must_check */
114 #undef __must_check
115 #include <xen/xen-os.h>
116 #endif
117 
118 SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
119     "LinuxKPI parameters");
120 
121 int linuxkpi_debug;
122 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN,
123     &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable.");
124 
125 int linuxkpi_rcu_debug;
126 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, rcu_debug, CTLFLAG_RWTUN,
127     &linuxkpi_rcu_debug, 0, "Set to enable RCU warning. Clear to disable.");
128 
129 int linuxkpi_warn_dump_stack = 0;
130 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, warn_dump_stack, CTLFLAG_RWTUN,
131     &linuxkpi_warn_dump_stack, 0,
132     "Set to enable stack traces from WARN_ON(). Clear to disable.");
133 
134 static struct timeval lkpi_net_lastlog;
135 static int lkpi_net_curpps;
136 static int lkpi_net_maxpps = 99;
137 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, net_ratelimit, CTLFLAG_RWTUN,
138     &lkpi_net_maxpps, 0, "Limit number of LinuxKPI net messages per second.");
139 
140 MALLOC_DEFINE(M_KMALLOC, "lkpikmalloc", "Linux kmalloc compat");
141 
142 #include <linux/rbtree.h>
143 /* Undo Linux compat changes. */
144 #undef RB_ROOT
145 #undef file
146 #undef cdev
147 #define	RB_ROOT(head)	(head)->rbh_root
148 
149 static void linux_destroy_dev(struct linux_cdev *);
150 static void linux_cdev_deref(struct linux_cdev *ldev);
151 static struct vm_area_struct *linux_cdev_handle_find(void *handle);
152 
153 cpumask_t cpu_online_mask;
154 static cpumask_t **static_single_cpu_mask;
155 static cpumask_t *static_single_cpu_mask_lcs;
156 struct kobject linux_class_root;
157 struct device linux_root_device;
158 struct class linux_class_misc;
159 struct list_head pci_drivers;
160 struct list_head pci_devices;
161 spinlock_t pci_lock;
162 struct uts_namespace init_uts_ns;
163 
164 unsigned long linux_timer_hz_mask;
165 
166 wait_queue_head_t linux_bit_waitq;
167 wait_queue_head_t linux_var_waitq;
168 
169 const guid_t guid_null;
170 
171 enum system_states system_state = SYSTEM_RUNNING;
172 
173 struct task_struct *
174 __lkpi_current(void)
175 {
176 	struct thread *td;
177 
178 	td = curthread;
179 	linux_set_current(td);
180 	return ((struct task_struct *)td->td_lkpi_task);
181 }
182 
183 int
184 panic_cmp(struct rb_node *one, struct rb_node *two)
185 {
186 	panic("no cmp");
187 }
188 
189 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp);
190 
191 #define	START(node)	((node)->start)
192 #define	LAST(node)	((node)->last)
193 
194 INTERVAL_TREE_DEFINE(struct interval_tree_node, rb, unsigned long,, START,
195     LAST,, lkpi_interval_tree)
196 
197 static void
198 linux_device_release(struct device *dev)
199 {
200 	pr_debug("linux_device_release: %s\n", dev_name(dev));
201 	kfree(dev);
202 }
203 
204 static ssize_t
205 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf)
206 {
207 	struct class_attribute *dattr;
208 	ssize_t error;
209 
210 	dattr = container_of(attr, struct class_attribute, attr);
211 	error = -EIO;
212 	if (dattr->show)
213 		error = dattr->show(container_of(kobj, struct class, kobj),
214 		    dattr, buf);
215 	return (error);
216 }
217 
218 static ssize_t
219 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf,
220     size_t count)
221 {
222 	struct class_attribute *dattr;
223 	ssize_t error;
224 
225 	dattr = container_of(attr, struct class_attribute, attr);
226 	error = -EIO;
227 	if (dattr->store)
228 		error = dattr->store(container_of(kobj, struct class, kobj),
229 		    dattr, buf, count);
230 	return (error);
231 }
232 
233 static void
234 linux_class_release(struct kobject *kobj)
235 {
236 	struct class *class;
237 
238 	class = container_of(kobj, struct class, kobj);
239 	if (class->class_release)
240 		class->class_release(class);
241 }
242 
243 static const struct sysfs_ops linux_class_sysfs = {
244 	.show  = linux_class_show,
245 	.store = linux_class_store,
246 };
247 
248 const struct kobj_type linux_class_ktype = {
249 	.release = linux_class_release,
250 	.sysfs_ops = &linux_class_sysfs
251 };
252 
253 static void
254 linux_dev_release(struct kobject *kobj)
255 {
256 	struct device *dev;
257 
258 	dev = container_of(kobj, struct device, kobj);
259 	/* This is the precedence defined by linux. */
260 	if (dev->release)
261 		dev->release(dev);
262 	else if (dev->class && dev->class->dev_release)
263 		dev->class->dev_release(dev);
264 }
265 
266 static ssize_t
267 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf)
268 {
269 	struct device_attribute *dattr;
270 	ssize_t error;
271 
272 	dattr = container_of(attr, struct device_attribute, attr);
273 	error = -EIO;
274 	if (dattr->show)
275 		error = dattr->show(container_of(kobj, struct device, kobj),
276 		    dattr, buf);
277 	return (error);
278 }
279 
280 static ssize_t
281 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf,
282     size_t count)
283 {
284 	struct device_attribute *dattr;
285 	ssize_t error;
286 
287 	dattr = container_of(attr, struct device_attribute, attr);
288 	error = -EIO;
289 	if (dattr->store)
290 		error = dattr->store(container_of(kobj, struct device, kobj),
291 		    dattr, buf, count);
292 	return (error);
293 }
294 
295 static const struct sysfs_ops linux_dev_sysfs = {
296 	.show  = linux_dev_show,
297 	.store = linux_dev_store,
298 };
299 
300 const struct kobj_type linux_dev_ktype = {
301 	.release = linux_dev_release,
302 	.sysfs_ops = &linux_dev_sysfs
303 };
304 
305 struct device *
306 device_create(struct class *class, struct device *parent, dev_t devt,
307     void *drvdata, const char *fmt, ...)
308 {
309 	struct device *dev;
310 	va_list args;
311 
312 	dev = kzalloc(sizeof(*dev), M_WAITOK);
313 	dev->parent = parent;
314 	dev->class = class;
315 	dev->devt = devt;
316 	dev->driver_data = drvdata;
317 	dev->release = linux_device_release;
318 	va_start(args, fmt);
319 	kobject_set_name_vargs(&dev->kobj, fmt, args);
320 	va_end(args);
321 	device_register(dev);
322 
323 	return (dev);
324 }
325 
326 struct device *
327 device_create_groups_vargs(struct class *class, struct device *parent,
328     dev_t devt, void *drvdata, const struct attribute_group **groups,
329     const char *fmt, va_list args)
330 {
331 	struct device *dev = NULL;
332 	int retval = -ENODEV;
333 
334 	if (class == NULL || IS_ERR(class))
335 		goto error;
336 
337 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
338 	if (!dev) {
339 		retval = -ENOMEM;
340 		goto error;
341 	}
342 
343 	dev->devt = devt;
344 	dev->class = class;
345 	dev->parent = parent;
346 	dev->groups = groups;
347 	dev->release = device_create_release;
348 	/* device_initialize() needs the class and parent to be set */
349 	device_initialize(dev);
350 	dev_set_drvdata(dev, drvdata);
351 
352 	retval = kobject_set_name_vargs(&dev->kobj, fmt, args);
353 	if (retval)
354 		goto error;
355 
356 	retval = device_add(dev);
357 	if (retval)
358 		goto error;
359 
360 	return dev;
361 
362 error:
363 	put_device(dev);
364 	return ERR_PTR(retval);
365 }
366 
367 struct class *
368 lkpi_class_create(const char *name)
369 {
370 	struct class *class;
371 	int error;
372 
373 	class = kzalloc(sizeof(*class), M_WAITOK);
374 	class->name = name;
375 	class->class_release = linux_class_kfree;
376 	error = class_register(class);
377 	if (error) {
378 		kfree(class);
379 		return (NULL);
380 	}
381 
382 	return (class);
383 }
384 
385 static void
386 linux_kq_lock(void *arg)
387 {
388 	spinlock_t *s = arg;
389 
390 	spin_lock(s);
391 }
392 static void
393 linux_kq_unlock(void *arg)
394 {
395 	spinlock_t *s = arg;
396 
397 	spin_unlock(s);
398 }
399 
400 static void
401 linux_kq_assert_lock(void *arg, int what)
402 {
403 #ifdef INVARIANTS
404 	spinlock_t *s = arg;
405 
406 	if (what == LA_LOCKED)
407 		mtx_assert(s, MA_OWNED);
408 	else
409 		mtx_assert(s, MA_NOTOWNED);
410 #endif
411 }
412 
413 static void
414 linux_file_kqfilter_poll(struct linux_file *, int);
415 
416 struct linux_file *
417 linux_file_alloc(void)
418 {
419 	struct linux_file *filp;
420 
421 	filp = kzalloc(sizeof(*filp), GFP_KERNEL);
422 
423 	/* set initial refcount */
424 	filp->f_count = 1;
425 
426 	/* setup fields needed by kqueue support */
427 	spin_lock_init(&filp->f_kqlock);
428 	knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock,
429 	    linux_kq_lock, linux_kq_unlock, linux_kq_assert_lock);
430 
431 	return (filp);
432 }
433 
434 void
435 linux_file_free(struct linux_file *filp)
436 {
437 	if (filp->_file == NULL) {
438 		if (filp->f_op != NULL && filp->f_op->release != NULL)
439 			filp->f_op->release(filp->f_vnode, filp);
440 		if (filp->f_shmem != NULL)
441 			vm_object_deallocate(filp->f_shmem);
442 		kfree_rcu(filp, rcu);
443 	} else {
444 		/*
445 		 * The close method of the character device or file
446 		 * will free the linux_file structure:
447 		 */
448 		_fdrop(filp->_file, curthread);
449 	}
450 }
451 
452 struct linux_cdev *
453 cdev_alloc(void)
454 {
455 	struct linux_cdev *cdev;
456 
457 	cdev = kzalloc(sizeof(struct linux_cdev), M_WAITOK);
458 	kobject_init(&cdev->kobj, &linux_cdev_ktype);
459 	cdev->refs = 1;
460 	return (cdev);
461 }
462 
463 static int
464 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
465     vm_page_t *mres)
466 {
467 	struct vm_area_struct *vmap;
468 
469 	vmap = linux_cdev_handle_find(vm_obj->handle);
470 
471 	MPASS(vmap != NULL);
472 	MPASS(vmap->vm_private_data == vm_obj->handle);
473 
474 	if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) {
475 		vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset;
476 		vm_page_t page;
477 
478 		if (((*mres)->flags & PG_FICTITIOUS) != 0) {
479 			/*
480 			 * If the passed in result page is a fake
481 			 * page, update it with the new physical
482 			 * address.
483 			 */
484 			page = *mres;
485 			vm_page_updatefake(page, paddr, vm_obj->memattr);
486 		} else {
487 			/*
488 			 * Replace the passed in "mres" page with our
489 			 * own fake page and free up the all of the
490 			 * original pages.
491 			 */
492 			VM_OBJECT_WUNLOCK(vm_obj);
493 			page = vm_page_getfake(paddr, vm_obj->memattr);
494 			VM_OBJECT_WLOCK(vm_obj);
495 
496 			vm_page_replace(page, vm_obj, (*mres)->pindex, *mres);
497 			*mres = page;
498 		}
499 		vm_page_valid(page);
500 		return (VM_PAGER_OK);
501 	}
502 	return (VM_PAGER_FAIL);
503 }
504 
505 static int
506 linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type,
507     vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
508 {
509 	struct vm_area_struct *vmap;
510 	int err;
511 
512 	/* get VM area structure */
513 	vmap = linux_cdev_handle_find(vm_obj->handle);
514 	MPASS(vmap != NULL);
515 	MPASS(vmap->vm_private_data == vm_obj->handle);
516 
517 	VM_OBJECT_WUNLOCK(vm_obj);
518 
519 	linux_set_current(curthread);
520 
521 	down_write(&vmap->vm_mm->mmap_sem);
522 	if (unlikely(vmap->vm_ops == NULL)) {
523 		err = VM_FAULT_SIGBUS;
524 	} else {
525 		struct vm_fault vmf;
526 
527 		/* fill out VM fault structure */
528 		vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx);
529 		vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
530 		vmf.pgoff = 0;
531 		vmf.page = NULL;
532 		vmf.vma = vmap;
533 
534 		vmap->vm_pfn_count = 0;
535 		vmap->vm_pfn_pcount = &vmap->vm_pfn_count;
536 		vmap->vm_obj = vm_obj;
537 
538 		err = vmap->vm_ops->fault(&vmf);
539 
540 		while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) {
541 			kern_yield(PRI_USER);
542 			err = vmap->vm_ops->fault(&vmf);
543 		}
544 	}
545 
546 	/* translate return code */
547 	switch (err) {
548 	case VM_FAULT_OOM:
549 		err = VM_PAGER_AGAIN;
550 		break;
551 	case VM_FAULT_SIGBUS:
552 		err = VM_PAGER_BAD;
553 		break;
554 	case VM_FAULT_NOPAGE:
555 		/*
556 		 * By contract the fault handler will return having
557 		 * busied all the pages itself. If pidx is already
558 		 * found in the object, it will simply xbusy the first
559 		 * page and return with vm_pfn_count set to 1.
560 		 */
561 		*first = vmap->vm_pfn_first;
562 		*last = *first + vmap->vm_pfn_count - 1;
563 		MPASS(pidx >= *first);
564 		MPASS(pidx <= *last);
565 		err = VM_PAGER_OK;
566 		break;
567 	default:
568 		err = VM_PAGER_ERROR;
569 		break;
570 	}
571 	up_write(&vmap->vm_mm->mmap_sem);
572 	VM_OBJECT_WLOCK(vm_obj);
573 	return (err);
574 }
575 
576 static struct rwlock linux_vma_lock;
577 static TAILQ_HEAD(, vm_area_struct) linux_vma_head =
578     TAILQ_HEAD_INITIALIZER(linux_vma_head);
579 
580 static void
581 linux_cdev_handle_free(struct vm_area_struct *vmap)
582 {
583 	/* Drop reference on vm_file */
584 	if (vmap->vm_file != NULL)
585 		fput(vmap->vm_file);
586 
587 	/* Drop reference on mm_struct */
588 	mmput(vmap->vm_mm);
589 
590 	kfree(vmap);
591 }
592 
593 static void
594 linux_cdev_handle_remove(struct vm_area_struct *vmap)
595 {
596 	rw_wlock(&linux_vma_lock);
597 	TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry);
598 	rw_wunlock(&linux_vma_lock);
599 }
600 
601 static struct vm_area_struct *
602 linux_cdev_handle_find(void *handle)
603 {
604 	struct vm_area_struct *vmap;
605 
606 	rw_rlock(&linux_vma_lock);
607 	TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) {
608 		if (vmap->vm_private_data == handle)
609 			break;
610 	}
611 	rw_runlock(&linux_vma_lock);
612 	return (vmap);
613 }
614 
615 static int
616 linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
617 		      vm_ooffset_t foff, struct ucred *cred, u_short *color)
618 {
619 
620 	MPASS(linux_cdev_handle_find(handle) != NULL);
621 	*color = 0;
622 	return (0);
623 }
624 
625 static void
626 linux_cdev_pager_dtor(void *handle)
627 {
628 	const struct vm_operations_struct *vm_ops;
629 	struct vm_area_struct *vmap;
630 
631 	vmap = linux_cdev_handle_find(handle);
632 	MPASS(vmap != NULL);
633 
634 	/*
635 	 * Remove handle before calling close operation to prevent
636 	 * other threads from reusing the handle pointer.
637 	 */
638 	linux_cdev_handle_remove(vmap);
639 
640 	down_write(&vmap->vm_mm->mmap_sem);
641 	vm_ops = vmap->vm_ops;
642 	if (likely(vm_ops != NULL))
643 		vm_ops->close(vmap);
644 	up_write(&vmap->vm_mm->mmap_sem);
645 
646 	linux_cdev_handle_free(vmap);
647 }
648 
649 static struct cdev_pager_ops linux_cdev_pager_ops[2] = {
650   {
651 	/* OBJT_MGTDEVICE */
652 	.cdev_pg_populate	= linux_cdev_pager_populate,
653 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
654 	.cdev_pg_dtor	= linux_cdev_pager_dtor
655   },
656   {
657 	/* OBJT_DEVICE */
658 	.cdev_pg_fault	= linux_cdev_pager_fault,
659 	.cdev_pg_ctor	= linux_cdev_pager_ctor,
660 	.cdev_pg_dtor	= linux_cdev_pager_dtor
661   },
662 };
663 
664 int
665 zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
666     unsigned long size)
667 {
668 	struct pctrie_iter pages;
669 	vm_object_t obj;
670 	vm_page_t m;
671 
672 	obj = vma->vm_obj;
673 	if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0)
674 		return (-ENOTSUP);
675 	VM_OBJECT_RLOCK(obj);
676 	vm_page_iter_limit_init(&pages, obj, OFF_TO_IDX(address + size));
677 	VM_RADIX_FOREACH_FROM(m, &pages, OFF_TO_IDX(address))
678 		pmap_remove_all(m);
679 	VM_OBJECT_RUNLOCK(obj);
680 	return (0);
681 }
682 
683 void
684 vma_set_file(struct vm_area_struct *vma, struct linux_file *file)
685 {
686 	struct linux_file *tmp;
687 
688 	/* Changing an anonymous vma with this is illegal */
689 	get_file(file);
690 	tmp = vma->vm_file;
691 	vma->vm_file = file;
692 	fput(tmp);
693 }
694 
695 static struct file_operations dummy_ldev_ops = {
696 	/* XXXKIB */
697 };
698 
699 static struct linux_cdev dummy_ldev = {
700 	.ops = &dummy_ldev_ops,
701 };
702 
703 #define	LDEV_SI_DTR	0x0001
704 #define	LDEV_SI_REF	0x0002
705 
706 static void
707 linux_get_fop(struct linux_file *filp, const struct file_operations **fop,
708     struct linux_cdev **dev)
709 {
710 	struct linux_cdev *ldev;
711 	u_int siref;
712 
713 	ldev = filp->f_cdev;
714 	*fop = filp->f_op;
715 	if (ldev != NULL) {
716 		if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
717 			refcount_acquire(&ldev->refs);
718 		} else {
719 			for (siref = ldev->siref;;) {
720 				if ((siref & LDEV_SI_DTR) != 0) {
721 					ldev = &dummy_ldev;
722 					*fop = ldev->ops;
723 					siref = ldev->siref;
724 					MPASS((ldev->siref & LDEV_SI_DTR) == 0);
725 				} else if (atomic_fcmpset_int(&ldev->siref,
726 				    &siref, siref + LDEV_SI_REF)) {
727 					break;
728 				}
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 	if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
742 		linux_cdev_deref(ldev);
743 	} else {
744 		MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
745 		MPASS((ldev->siref & ~LDEV_SI_DTR) != 0);
746 		atomic_subtract_int(&ldev->siref, LDEV_SI_REF);
747 	}
748 }
749 
750 #define	OPW(fp,td,code) ({			\
751 	struct file *__fpop;			\
752 	__typeof(code) __retval;		\
753 						\
754 	__fpop = (td)->td_fpop;			\
755 	(td)->td_fpop = (fp);			\
756 	__retval = (code);			\
757 	(td)->td_fpop = __fpop;			\
758 	__retval;				\
759 })
760 
761 static int
762 linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td,
763     struct file *file)
764 {
765 	struct linux_cdev *ldev;
766 	struct linux_file *filp;
767 	const struct file_operations *fop;
768 	int error;
769 
770 	ldev = dev->si_drv1;
771 
772 	filp = linux_file_alloc();
773 	filp->f_dentry = &filp->f_dentry_store;
774 	filp->f_op = ldev->ops;
775 	filp->f_mode = file->f_flag;
776 	filp->f_flags = file->f_flag;
777 	filp->f_vnode = file->f_vnode;
778 	filp->_file = file;
779 	refcount_acquire(&ldev->refs);
780 	filp->f_cdev = ldev;
781 
782 	linux_set_current(td);
783 	linux_get_fop(filp, &fop, &ldev);
784 
785 	if (fop->open != NULL) {
786 		error = -fop->open(file->f_vnode, filp);
787 		if (error != 0) {
788 			linux_drop_fop(ldev);
789 			linux_cdev_deref(filp->f_cdev);
790 			kfree(filp);
791 			return (error);
792 		}
793 	}
794 
795 	/* hold on to the vnode - used for fstat() */
796 	vref(filp->f_vnode);
797 
798 	/* release the file from devfs */
799 	finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops);
800 	linux_drop_fop(ldev);
801 	return (ENXIO);
802 }
803 
804 #define	LINUX_IOCTL_MIN_PTR 0x10000UL
805 #define	LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX)
806 
807 static inline int
808 linux_remap_address(void **uaddr, size_t len)
809 {
810 	uintptr_t uaddr_val = (uintptr_t)(*uaddr);
811 
812 	if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR &&
813 	    uaddr_val < LINUX_IOCTL_MAX_PTR)) {
814 		struct task_struct *pts = current;
815 		if (pts == NULL) {
816 			*uaddr = NULL;
817 			return (1);
818 		}
819 
820 		/* compute data offset */
821 		uaddr_val -= LINUX_IOCTL_MIN_PTR;
822 
823 		/* check that length is within bounds */
824 		if ((len > IOCPARM_MAX) ||
825 		    (uaddr_val + len) > pts->bsd_ioctl_len) {
826 			*uaddr = NULL;
827 			return (1);
828 		}
829 
830 		/* re-add kernel buffer address */
831 		uaddr_val += (uintptr_t)pts->bsd_ioctl_data;
832 
833 		/* update address location */
834 		*uaddr = (void *)uaddr_val;
835 		return (1);
836 	}
837 	return (0);
838 }
839 
840 int
841 linux_copyin(const void *uaddr, void *kaddr, size_t len)
842 {
843 	if (linux_remap_address(__DECONST(void **, &uaddr), len)) {
844 		if (uaddr == NULL)
845 			return (-EFAULT);
846 		memcpy(kaddr, uaddr, len);
847 		return (0);
848 	}
849 	return (-copyin(uaddr, kaddr, len));
850 }
851 
852 int
853 linux_copyout(const void *kaddr, void *uaddr, size_t len)
854 {
855 	if (linux_remap_address(&uaddr, len)) {
856 		if (uaddr == NULL)
857 			return (-EFAULT);
858 		memcpy(uaddr, kaddr, len);
859 		return (0);
860 	}
861 	return (-copyout(kaddr, uaddr, len));
862 }
863 
864 size_t
865 linux_clear_user(void *_uaddr, size_t _len)
866 {
867 	uint8_t *uaddr = _uaddr;
868 	size_t len = _len;
869 
870 	/* make sure uaddr is aligned before going into the fast loop */
871 	while (((uintptr_t)uaddr & 7) != 0 && len > 7) {
872 		if (subyte(uaddr, 0))
873 			return (_len);
874 		uaddr++;
875 		len--;
876 	}
877 
878 	/* zero 8 bytes at a time */
879 	while (len > 7) {
880 #ifdef __LP64__
881 		if (suword64(uaddr, 0))
882 			return (_len);
883 #else
884 		if (suword32(uaddr, 0))
885 			return (_len);
886 		if (suword32(uaddr + 4, 0))
887 			return (_len);
888 #endif
889 		uaddr += 8;
890 		len -= 8;
891 	}
892 
893 	/* zero fill end, if any */
894 	while (len > 0) {
895 		if (subyte(uaddr, 0))
896 			return (_len);
897 		uaddr++;
898 		len--;
899 	}
900 	return (0);
901 }
902 
903 int
904 linux_access_ok(const void *uaddr, size_t len)
905 {
906 	uintptr_t saddr;
907 	uintptr_t eaddr;
908 
909 	/* get start and end address */
910 	saddr = (uintptr_t)uaddr;
911 	eaddr = (uintptr_t)uaddr + len;
912 
913 	/* verify addresses are valid for userspace */
914 	return ((saddr == eaddr) ||
915 	    (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS));
916 }
917 
918 /*
919  * This function should return either EINTR or ERESTART depending on
920  * the signal type sent to this thread:
921  */
922 static int
923 linux_get_error(struct task_struct *task, int error)
924 {
925 	/* check for signal type interrupt code */
926 	if (error == EINTR || error == ERESTARTSYS || error == ERESTART) {
927 		error = -linux_schedule_get_interrupt_value(task);
928 		if (error == 0)
929 			error = EINTR;
930 	}
931 	return (error);
932 }
933 
934 static int
935 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
936     const struct file_operations *fop, u_long cmd, caddr_t data,
937     struct thread *td)
938 {
939 	struct task_struct *task = current;
940 	unsigned size;
941 	int error;
942 	bool direct;
943 
944 	size = IOCPARM_LEN(cmd);
945 	/* refer to logic in sys_ioctl() */
946 	direct = false;
947 	if (size > 0) {
948 		/*
949 		 * Setup hint for linux_copyin() and linux_copyout().
950 		 *
951 		 * Background: Linux kernel code expects to operate on
952 		 * userspace addresses, but FreeBSD's kern_ioctl()
953 		 * will generally provide a kernel address.  For the
954 		 * native process ABI, where we know how to find the
955 		 * original address, we reach directly into the system
956 		 * call args to get it.  Then, if the Linux driver
957 		 * copied out to that address, we copy the whole block
958 		 * back into the kernel buffer allocated by
959 		 * kern_ioctl() so that kern_ioctl() itself doesn't
960 		 * clobber the driver's data.
961 		 *
962 		 * Otherwise, fall back to the LINUX_IOCTL_MIN_PTR
963 		 * hack.
964 		 */
965 		task->bsd_ioctl_data = data;
966 		task->bsd_ioctl_len = size;
967 		if ((td->td_pflags & TDP_KTHREAD) == 0 &&
968 		    SV_PROC_ABI(td->td_proc) == SV_ABI_FREEBSD &&
969 		    td->td_sa.code == SYS_ioctl) {
970 			direct = true;
971 			data = (void *)(uintptr_t)td->td_sa.args[2];
972 		} else {
973 			data = (void *)LINUX_IOCTL_MIN_PTR;
974 		}
975 	} else {
976 		/* fetch user-space pointer */
977 		data = *(void **)data;
978 	}
979 #ifdef COMPAT_FREEBSD32
980 	if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
981 		/* try the compat IOCTL handler first */
982 		if (fop->compat_ioctl != NULL) {
983 			error = -OPW(fp, td, fop->compat_ioctl(filp,
984 			    cmd, (u_long)data));
985 		} else {
986 			error = ENOTTY;
987 		}
988 
989 		/* fallback to the regular IOCTL handler, if any */
990 		if (error == ENOTTY && fop->unlocked_ioctl != NULL) {
991 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
992 			    cmd, (u_long)data));
993 		}
994 	} else
995 #endif
996 	{
997 		if (fop->unlocked_ioctl != NULL) {
998 			error = -OPW(fp, td, fop->unlocked_ioctl(filp,
999 			    cmd, (u_long)data));
1000 		} else {
1001 			error = ENOTTY;
1002 		}
1003 	}
1004 	if (error == 0 && size > 0 && (cmd & IOC_OUT) != 0 && direct) {
1005 		void *xdata;
1006 		int error1;
1007 
1008 		/*
1009 		 * Ensure that the copyout in sys_generic.c copies
1010 		 * over the data which is possibly modified by the
1011 		 * driver.  A possible error from the copyin() is
1012 		 * ignored since it is formally possible for the memory
1013 		 * to become unaccessible in the meantime.  Do the copying
1014 		 * through the intermediate buffer instead of copying
1015 		 * directly to bsd_ioctl_data, to ensure atomicity of
1016 		 * the change with respect to the error.
1017 		 */
1018 		xdata = malloc(size, M_TEMP, M_WAITOK);
1019 		error1 = copyin(data, xdata, size);
1020 		if (error1 == 0)
1021 			memcpy(task->bsd_ioctl_data, xdata, size);
1022 		free(xdata, M_TEMP);
1023 	}
1024 	if (size > 0) {
1025 		task->bsd_ioctl_data = NULL;
1026 		task->bsd_ioctl_len = 0;
1027 	}
1028 	if (error == EWOULDBLOCK) {
1029 		/* update kqfilter status, if any */
1030 		linux_file_kqfilter_poll(filp,
1031 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1032 	} else {
1033 		error = linux_get_error(task, error);
1034 	}
1035 	return (error);
1036 }
1037 
1038 #define	LINUX_POLL_TABLE_NORMAL ((poll_table *)1)
1039 
1040 /*
1041  * This function atomically updates the poll wakeup state and returns
1042  * the previous state at the time of update.
1043  */
1044 static uint8_t
1045 linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate)
1046 {
1047 	int c, old;
1048 
1049 	c = v->counter;
1050 
1051 	while ((old = atomic_cmpxchg(v, c, pstate[c])) != c)
1052 		c = old;
1053 
1054 	return (c);
1055 }
1056 
1057 static int
1058 linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key)
1059 {
1060 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1061 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */
1062 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1063 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY,
1064 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */
1065 	};
1066 	struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq);
1067 
1068 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1069 	case LINUX_FWQ_STATE_QUEUED:
1070 		linux_poll_wakeup(filp);
1071 		return (1);
1072 	default:
1073 		return (0);
1074 	}
1075 }
1076 
1077 void
1078 linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p)
1079 {
1080 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1081 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY,
1082 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1083 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */
1084 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED,
1085 	};
1086 
1087 	/* check if we are called inside the select system call */
1088 	if (p == LINUX_POLL_TABLE_NORMAL)
1089 		selrecord(curthread, &filp->f_selinfo);
1090 
1091 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1092 	case LINUX_FWQ_STATE_INIT:
1093 		/* NOTE: file handles can only belong to one wait-queue */
1094 		filp->f_wait_queue.wqh = wqh;
1095 		filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback;
1096 		add_wait_queue(wqh, &filp->f_wait_queue.wq);
1097 		atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED);
1098 		break;
1099 	default:
1100 		break;
1101 	}
1102 }
1103 
1104 static void
1105 linux_poll_wait_dequeue(struct linux_file *filp)
1106 {
1107 	static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1108 		[LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT,	/* NOP */
1109 		[LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT,
1110 		[LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT,
1111 		[LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT,
1112 	};
1113 
1114 	seldrain(&filp->f_selinfo);
1115 
1116 	switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1117 	case LINUX_FWQ_STATE_NOT_READY:
1118 	case LINUX_FWQ_STATE_QUEUED:
1119 	case LINUX_FWQ_STATE_READY:
1120 		remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq);
1121 		break;
1122 	default:
1123 		break;
1124 	}
1125 }
1126 
1127 void
1128 linux_poll_wakeup(struct linux_file *filp)
1129 {
1130 	/* this function should be NULL-safe */
1131 	if (filp == NULL)
1132 		return;
1133 
1134 	selwakeup(&filp->f_selinfo);
1135 
1136 	spin_lock(&filp->f_kqlock);
1137 	filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ |
1138 	    LINUX_KQ_FLAG_NEED_WRITE;
1139 
1140 	/* make sure the "knote" gets woken up */
1141 	KNOTE_LOCKED(&filp->f_selinfo.si_note, 1);
1142 	spin_unlock(&filp->f_kqlock);
1143 }
1144 
1145 static struct linux_file *
1146 __get_file_rcu(struct linux_file **f)
1147 {
1148 	struct linux_file *file1, *file2;
1149 
1150 	file1 = READ_ONCE(*f);
1151 	if (file1 == NULL)
1152 		return (NULL);
1153 
1154 	if (!refcount_acquire_if_not_zero(
1155 	    file1->_file == NULL ? &file1->f_count : &file1->_file->f_count))
1156 		return (ERR_PTR(-EAGAIN));
1157 
1158 	file2 = READ_ONCE(*f);
1159 	if (file2 == file1)
1160 		return (file2);
1161 
1162 	fput(file1);
1163 	return (ERR_PTR(-EAGAIN));
1164 }
1165 
1166 struct linux_file *
1167 linux_get_file_rcu(struct linux_file **f)
1168 {
1169 	struct linux_file *file1;
1170 
1171 	for (;;) {
1172 		file1 = __get_file_rcu(f);
1173 		if (file1 == NULL)
1174 			return (NULL);
1175 
1176 		if (IS_ERR(file1))
1177 			continue;
1178 
1179 		return (file1);
1180 	}
1181 }
1182 
1183 struct linux_file *
1184 get_file_active(struct linux_file **f)
1185 {
1186 	struct linux_file *file1;
1187 
1188 	rcu_read_lock();
1189 	file1 = __get_file_rcu(f);
1190 	rcu_read_unlock();
1191 	if (IS_ERR(file1))
1192 		file1 = NULL;
1193 
1194 	return (file1);
1195 }
1196 
1197 static void
1198 linux_file_kqfilter_detach(struct knote *kn)
1199 {
1200 	struct linux_file *filp = kn->kn_hook;
1201 
1202 	spin_lock(&filp->f_kqlock);
1203 	knlist_remove(&filp->f_selinfo.si_note, kn, 1);
1204 	spin_unlock(&filp->f_kqlock);
1205 }
1206 
1207 static int
1208 linux_file_kqfilter_read_event(struct knote *kn, long hint)
1209 {
1210 	struct linux_file *filp = kn->kn_hook;
1211 
1212 	mtx_assert(&filp->f_kqlock, MA_OWNED);
1213 
1214 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0);
1215 }
1216 
1217 static int
1218 linux_file_kqfilter_write_event(struct knote *kn, long hint)
1219 {
1220 	struct linux_file *filp = kn->kn_hook;
1221 
1222 	mtx_assert(&filp->f_kqlock, MA_OWNED);
1223 
1224 	return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0);
1225 }
1226 
1227 static const struct filterops linux_dev_kqfiltops_read = {
1228 	.f_isfd = 1,
1229 	.f_detach = linux_file_kqfilter_detach,
1230 	.f_event = linux_file_kqfilter_read_event,
1231 	.f_copy = knote_triv_copy,
1232 };
1233 
1234 static const struct filterops linux_dev_kqfiltops_write = {
1235 	.f_isfd = 1,
1236 	.f_detach = linux_file_kqfilter_detach,
1237 	.f_event = linux_file_kqfilter_write_event,
1238 	.f_copy = knote_triv_copy,
1239 };
1240 
1241 static void
1242 linux_file_kqfilter_poll(struct linux_file *filp, int kqflags)
1243 {
1244 	struct thread *td;
1245 	const struct file_operations *fop;
1246 	struct linux_cdev *ldev;
1247 	int temp;
1248 
1249 	if ((filp->f_kqflags & kqflags) == 0)
1250 		return;
1251 
1252 	td = curthread;
1253 
1254 	linux_get_fop(filp, &fop, &ldev);
1255 	/* get the latest polling state */
1256 	temp = OPW(filp->_file, td, fop->poll(filp, NULL));
1257 	linux_drop_fop(ldev);
1258 
1259 	spin_lock(&filp->f_kqlock);
1260 	/* clear kqflags */
1261 	filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ |
1262 	    LINUX_KQ_FLAG_NEED_WRITE);
1263 	/* update kqflags */
1264 	if ((temp & (POLLIN | POLLOUT)) != 0) {
1265 		if ((temp & POLLIN) != 0)
1266 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ;
1267 		if ((temp & POLLOUT) != 0)
1268 			filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE;
1269 
1270 		/* make sure the "knote" gets woken up */
1271 		KNOTE_LOCKED(&filp->f_selinfo.si_note, 0);
1272 	}
1273 	spin_unlock(&filp->f_kqlock);
1274 }
1275 
1276 static int
1277 linux_file_kqfilter(struct file *file, struct knote *kn)
1278 {
1279 	struct linux_file *filp;
1280 	struct thread *td;
1281 	int error;
1282 
1283 	td = curthread;
1284 	filp = (struct linux_file *)file->f_data;
1285 	filp->f_flags = file->f_flag;
1286 	if (filp->f_op->poll == NULL)
1287 		return (EINVAL);
1288 
1289 	spin_lock(&filp->f_kqlock);
1290 	switch (kn->kn_filter) {
1291 	case EVFILT_READ:
1292 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ;
1293 		kn->kn_fop = &linux_dev_kqfiltops_read;
1294 		kn->kn_hook = filp;
1295 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1296 		error = 0;
1297 		break;
1298 	case EVFILT_WRITE:
1299 		filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE;
1300 		kn->kn_fop = &linux_dev_kqfiltops_write;
1301 		kn->kn_hook = filp;
1302 		knlist_add(&filp->f_selinfo.si_note, kn, 1);
1303 		error = 0;
1304 		break;
1305 	default:
1306 		error = EINVAL;
1307 		break;
1308 	}
1309 	spin_unlock(&filp->f_kqlock);
1310 
1311 	if (error == 0) {
1312 		linux_set_current(td);
1313 
1314 		/* update kqfilter status, if any */
1315 		linux_file_kqfilter_poll(filp,
1316 		    LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1317 	}
1318 	return (error);
1319 }
1320 
1321 static int
1322 linux_file_mmap_single(struct file *fp, const struct file_operations *fop,
1323     vm_ooffset_t *offset, vm_size_t size, struct vm_object **object,
1324     int nprot, bool is_shared, struct thread *td)
1325 {
1326 	struct task_struct *task;
1327 	struct vm_area_struct *vmap;
1328 	struct mm_struct *mm;
1329 	struct linux_file *filp;
1330 	vm_memattr_t attr;
1331 	int error;
1332 
1333 	filp = (struct linux_file *)fp->f_data;
1334 	filp->f_flags = fp->f_flag;
1335 
1336 	if (fop->mmap == NULL)
1337 		return (EOPNOTSUPP);
1338 
1339 	linux_set_current(td);
1340 
1341 	/*
1342 	 * The same VM object might be shared by multiple processes
1343 	 * and the mm_struct is usually freed when a process exits.
1344 	 *
1345 	 * The atomic reference below makes sure the mm_struct is
1346 	 * available as long as the vmap is in the linux_vma_head.
1347 	 */
1348 	task = current;
1349 	mm = task->mm;
1350 	if (atomic_inc_not_zero(&mm->mm_users) == 0)
1351 		return (EINVAL);
1352 
1353 	vmap = kzalloc(sizeof(*vmap), GFP_KERNEL);
1354 	vmap->vm_start = 0;
1355 	vmap->vm_end = size;
1356 	vmap->vm_pgoff = *offset / PAGE_SIZE;
1357 	vmap->vm_pfn = 0;
1358 	vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL);
1359 	if (is_shared)
1360 		vmap->vm_flags |= VM_SHARED;
1361 	vmap->vm_ops = NULL;
1362 	vmap->vm_file = get_file(filp);
1363 	vmap->vm_mm = mm;
1364 
1365 	if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) {
1366 		error = linux_get_error(task, EINTR);
1367 	} else {
1368 		error = -OPW(fp, td, fop->mmap(filp, vmap));
1369 		error = linux_get_error(task, error);
1370 		up_write(&vmap->vm_mm->mmap_sem);
1371 	}
1372 
1373 	if (error != 0) {
1374 		linux_cdev_handle_free(vmap);
1375 		return (error);
1376 	}
1377 
1378 	attr = pgprot2cachemode(vmap->vm_page_prot);
1379 
1380 	if (vmap->vm_ops != NULL) {
1381 		struct vm_area_struct *ptr;
1382 		void *vm_private_data;
1383 		bool vm_no_fault;
1384 
1385 		if (vmap->vm_ops->open == NULL ||
1386 		    vmap->vm_ops->close == NULL ||
1387 		    vmap->vm_private_data == NULL) {
1388 			/* free allocated VM area struct */
1389 			linux_cdev_handle_free(vmap);
1390 			return (EINVAL);
1391 		}
1392 
1393 		vm_private_data = vmap->vm_private_data;
1394 
1395 		rw_wlock(&linux_vma_lock);
1396 		TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) {
1397 			if (ptr->vm_private_data == vm_private_data)
1398 				break;
1399 		}
1400 		/* check if there is an existing VM area struct */
1401 		if (ptr != NULL) {
1402 			/* check if the VM area structure is invalid */
1403 			if (ptr->vm_ops == NULL ||
1404 			    ptr->vm_ops->open == NULL ||
1405 			    ptr->vm_ops->close == NULL) {
1406 				error = ESTALE;
1407 				vm_no_fault = 1;
1408 			} else {
1409 				if (ptr->vm_start == vmap->vm_start &&
1410 				    ptr->vm_end <= vmap->vm_end) {
1411 					/*
1412 					 * Userspace wants to grow an existing
1413 					 * mapping. We already have a
1414 					 * `vm_object_t' for this mapping. We
1415 					 * just need to update the `struct
1416 					 * vm_area_struct` to have the correct
1417 					 * end address.
1418 					 */
1419 					ptr->vm_end = vmap->vm_end;
1420 				}
1421 
1422 				error = EEXIST;
1423 				vm_no_fault = (ptr->vm_ops->fault == NULL);
1424 			}
1425 		} else {
1426 			/* insert VM area structure into list */
1427 			TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry);
1428 			error = 0;
1429 			vm_no_fault = (vmap->vm_ops->fault == NULL);
1430 		}
1431 		rw_wunlock(&linux_vma_lock);
1432 
1433 		if (error != 0) {
1434 			/* free allocated VM area struct */
1435 			linux_cdev_handle_free(vmap);
1436 			/* check for stale VM area struct */
1437 			if (error != EEXIST)
1438 				return (error);
1439 		}
1440 
1441 		/* check if there is no fault handler */
1442 		if (vm_no_fault) {
1443 			*object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE,
1444 			    &linux_cdev_pager_ops[1], size, nprot, *offset,
1445 			    td->td_ucred);
1446 		} else {
1447 			*object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE,
1448 			    &linux_cdev_pager_ops[0], size, nprot, *offset,
1449 			    td->td_ucred);
1450 		}
1451 
1452 		/* check if allocating the VM object failed */
1453 		if (*object == NULL) {
1454 			if (error == 0) {
1455 				/* remove VM area struct from list */
1456 				linux_cdev_handle_remove(vmap);
1457 				/* free allocated VM area struct */
1458 				linux_cdev_handle_free(vmap);
1459 			}
1460 			return (EINVAL);
1461 		}
1462 	} else {
1463 		struct sglist *sg;
1464 
1465 		sg = sglist_alloc(1, M_WAITOK);
1466 		sglist_append_phys(sg,
1467 		    (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len);
1468 
1469 		*object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len,
1470 		    nprot, 0, td->td_ucred);
1471 
1472 		linux_cdev_handle_free(vmap);
1473 
1474 		if (*object == NULL) {
1475 			sglist_free(sg);
1476 			return (EINVAL);
1477 		}
1478 	}
1479 
1480 	if (attr != VM_MEMATTR_DEFAULT) {
1481 		VM_OBJECT_WLOCK(*object);
1482 		vm_object_set_memattr(*object, attr);
1483 		VM_OBJECT_WUNLOCK(*object);
1484 	}
1485 	*offset = 0;
1486 	return (0);
1487 }
1488 
1489 struct cdevsw linuxcdevsw = {
1490 	.d_version = D_VERSION,
1491 	.d_fdopen = linux_dev_fdopen,
1492 	.d_name = "lkpidev",
1493 };
1494 
1495 static int
1496 linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
1497     int flags, struct thread *td)
1498 {
1499 	struct linux_file *filp;
1500 	const struct file_operations *fop;
1501 	struct linux_cdev *ldev;
1502 	ssize_t bytes;
1503 	int error;
1504 
1505 	error = 0;
1506 	filp = (struct linux_file *)file->f_data;
1507 	filp->f_flags = file->f_flag;
1508 	/* XXX no support for I/O vectors currently */
1509 	if (uio->uio_iovcnt != 1)
1510 		return (EOPNOTSUPP);
1511 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1512 		return (EINVAL);
1513 	linux_set_current(td);
1514 	linux_get_fop(filp, &fop, &ldev);
1515 	if (fop->read != NULL) {
1516 		bytes = OPW(file, td, fop->read(filp,
1517 		    uio->uio_iov->iov_base,
1518 		    uio->uio_iov->iov_len, &uio->uio_offset));
1519 		if (bytes >= 0) {
1520 			uio->uio_iov->iov_base =
1521 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1522 			uio->uio_iov->iov_len -= bytes;
1523 			uio->uio_resid -= bytes;
1524 		} else {
1525 			error = linux_get_error(current, -bytes);
1526 		}
1527 	} else
1528 		error = ENXIO;
1529 
1530 	/* update kqfilter status, if any */
1531 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ);
1532 	linux_drop_fop(ldev);
1533 
1534 	return (error);
1535 }
1536 
1537 static int
1538 linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred,
1539     int flags, struct thread *td)
1540 {
1541 	struct linux_file *filp;
1542 	const struct file_operations *fop;
1543 	struct linux_cdev *ldev;
1544 	ssize_t bytes;
1545 	int error;
1546 
1547 	filp = (struct linux_file *)file->f_data;
1548 	filp->f_flags = file->f_flag;
1549 	/* XXX no support for I/O vectors currently */
1550 	if (uio->uio_iovcnt != 1)
1551 		return (EOPNOTSUPP);
1552 	if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1553 		return (EINVAL);
1554 	linux_set_current(td);
1555 	linux_get_fop(filp, &fop, &ldev);
1556 	if (fop->write != NULL) {
1557 		bytes = OPW(file, td, fop->write(filp,
1558 		    uio->uio_iov->iov_base,
1559 		    uio->uio_iov->iov_len, &uio->uio_offset));
1560 		if (bytes >= 0) {
1561 			uio->uio_iov->iov_base =
1562 			    ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1563 			uio->uio_iov->iov_len -= bytes;
1564 			uio->uio_resid -= bytes;
1565 			error = 0;
1566 		} else {
1567 			error = linux_get_error(current, -bytes);
1568 		}
1569 	} else
1570 		error = ENXIO;
1571 
1572 	/* update kqfilter status, if any */
1573 	linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE);
1574 
1575 	linux_drop_fop(ldev);
1576 
1577 	return (error);
1578 }
1579 
1580 static int
1581 linux_file_poll(struct file *file, int events, struct ucred *active_cred,
1582     struct thread *td)
1583 {
1584 	struct linux_file *filp;
1585 	const struct file_operations *fop;
1586 	struct linux_cdev *ldev;
1587 	int revents;
1588 
1589 	filp = (struct linux_file *)file->f_data;
1590 	filp->f_flags = file->f_flag;
1591 	linux_set_current(td);
1592 	linux_get_fop(filp, &fop, &ldev);
1593 	if (fop->poll != NULL) {
1594 		revents = OPW(file, td, fop->poll(filp,
1595 		    LINUX_POLL_TABLE_NORMAL)) & events;
1596 	} else {
1597 		revents = 0;
1598 	}
1599 	linux_drop_fop(ldev);
1600 	return (revents);
1601 }
1602 
1603 static int
1604 linux_file_close(struct file *file, struct thread *td)
1605 {
1606 	struct linux_file *filp;
1607 	int (*release)(struct inode *, struct linux_file *);
1608 	const struct file_operations *fop;
1609 	struct linux_cdev *ldev;
1610 	int error;
1611 
1612 	filp = (struct linux_file *)file->f_data;
1613 
1614 	KASSERT(file_count(filp) == 0,
1615 	    ("File refcount(%d) is not zero", file_count(filp)));
1616 
1617 	if (td == NULL)
1618 		td = curthread;
1619 
1620 	error = 0;
1621 	filp->f_flags = file->f_flag;
1622 	linux_set_current(td);
1623 	linux_poll_wait_dequeue(filp);
1624 	linux_get_fop(filp, &fop, &ldev);
1625 	/*
1626 	 * Always use the real release function, if any, to avoid
1627 	 * leaking device resources:
1628 	 */
1629 	release = filp->f_op->release;
1630 	if (release != NULL)
1631 		error = -OPW(file, td, release(filp->f_vnode, filp));
1632 	funsetown(&filp->f_sigio);
1633 	if (filp->f_vnode != NULL)
1634 		vrele(filp->f_vnode);
1635 	linux_drop_fop(ldev);
1636 	ldev = filp->f_cdev;
1637 	if (ldev != NULL)
1638 		linux_cdev_deref(ldev);
1639 	linux_synchronize_rcu(RCU_TYPE_REGULAR);
1640 	kfree(filp);
1641 
1642 	return (error);
1643 }
1644 
1645 static int
1646 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
1647     struct thread *td)
1648 {
1649 	struct linux_file *filp;
1650 	const struct file_operations *fop;
1651 	struct linux_cdev *ldev;
1652 	struct fiodgname_arg *fgn;
1653 	const char *p;
1654 	int error, i;
1655 
1656 	error = 0;
1657 	filp = (struct linux_file *)fp->f_data;
1658 	filp->f_flags = fp->f_flag;
1659 	linux_get_fop(filp, &fop, &ldev);
1660 
1661 	linux_set_current(td);
1662 	switch (cmd) {
1663 	case FIONBIO:
1664 		break;
1665 	case FIOASYNC:
1666 		if (fop->fasync == NULL)
1667 			break;
1668 		error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC));
1669 		break;
1670 	case FIOSETOWN:
1671 		error = fsetown(*(int *)data, &filp->f_sigio);
1672 		if (error == 0) {
1673 			if (fop->fasync == NULL)
1674 				break;
1675 			error = -OPW(fp, td, fop->fasync(0, filp,
1676 			    fp->f_flag & FASYNC));
1677 		}
1678 		break;
1679 	case FIOGETOWN:
1680 		*(int *)data = fgetown(&filp->f_sigio);
1681 		break;
1682 	case FIODGNAME:
1683 #ifdef	COMPAT_FREEBSD32
1684 	case FIODGNAME_32:
1685 #endif
1686 		if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) {
1687 			error = ENXIO;
1688 			break;
1689 		}
1690 		fgn = data;
1691 		p = devtoname(filp->f_cdev->cdev);
1692 		i = strlen(p) + 1;
1693 		if (i > fgn->len) {
1694 			error = EINVAL;
1695 			break;
1696 		}
1697 		error = copyout(p, fiodgname_buf_get_ptr(fgn, cmd), i);
1698 		break;
1699 	default:
1700 		error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td);
1701 		break;
1702 	}
1703 	linux_drop_fop(ldev);
1704 	return (error);
1705 }
1706 
1707 static int
1708 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1709     vm_prot_t maxprot, int flags, struct file *fp,
1710     vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp)
1711 {
1712 	/*
1713 	 * Character devices do not provide private mappings
1714 	 * of any kind:
1715 	 */
1716 	if ((maxprot & VM_PROT_WRITE) == 0 &&
1717 	    (prot & VM_PROT_WRITE) != 0)
1718 		return (EACCES);
1719 	if ((flags & (MAP_PRIVATE | MAP_COPY)) != 0)
1720 		return (EINVAL);
1721 
1722 	return (linux_file_mmap_single(fp, fop, foff, objsize, objp,
1723 	    (int)prot, (flags & MAP_SHARED) ? true : false, td));
1724 }
1725 
1726 static int
1727 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1728     vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1729     struct thread *td)
1730 {
1731 	struct linux_file *filp;
1732 	const struct file_operations *fop;
1733 	struct linux_cdev *ldev;
1734 	struct mount *mp;
1735 	struct vnode *vp;
1736 	vm_object_t object;
1737 	vm_prot_t maxprot;
1738 	int error;
1739 
1740 	filp = (struct linux_file *)fp->f_data;
1741 
1742 	vp = filp->f_vnode;
1743 	if (vp == NULL)
1744 		return (EOPNOTSUPP);
1745 
1746 	/*
1747 	 * Ensure that file and memory protections are
1748 	 * compatible.
1749 	 */
1750 	mp = vp->v_mount;
1751 	if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1752 		maxprot = VM_PROT_NONE;
1753 		if ((prot & VM_PROT_EXECUTE) != 0)
1754 			return (EACCES);
1755 	} else
1756 		maxprot = VM_PROT_EXECUTE;
1757 	if ((fp->f_flag & FREAD) != 0)
1758 		maxprot |= VM_PROT_READ;
1759 	else if ((prot & VM_PROT_READ) != 0)
1760 		return (EACCES);
1761 
1762 	/*
1763 	 * If we are sharing potential changes via MAP_SHARED and we
1764 	 * are trying to get write permission although we opened it
1765 	 * without asking for it, bail out.
1766 	 *
1767 	 * Note that most character devices always share mappings.
1768 	 *
1769 	 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE
1770 	 * requests rather than doing it here.
1771 	 */
1772 	if ((flags & MAP_SHARED) != 0) {
1773 		if ((fp->f_flag & FWRITE) != 0)
1774 			maxprot |= VM_PROT_WRITE;
1775 		else if ((prot & VM_PROT_WRITE) != 0)
1776 			return (EACCES);
1777 	}
1778 	maxprot &= cap_maxprot;
1779 
1780 	linux_get_fop(filp, &fop, &ldev);
1781 	error = linux_file_mmap_sub(td, size, prot, maxprot, flags, fp,
1782 	    &foff, fop, &object);
1783 	if (error != 0)
1784 		goto out;
1785 
1786 	error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1787 	    foff, FALSE, td);
1788 	if (error != 0)
1789 		vm_object_deallocate(object);
1790 out:
1791 	linux_drop_fop(ldev);
1792 	return (error);
1793 }
1794 
1795 static int
1796 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
1797 {
1798 	struct linux_file *filp;
1799 	struct vnode *vp;
1800 	int error;
1801 
1802 	filp = (struct linux_file *)fp->f_data;
1803 	if (filp->f_vnode == NULL)
1804 		return (EOPNOTSUPP);
1805 
1806 	vp = filp->f_vnode;
1807 
1808 	vn_lock(vp, LK_SHARED | LK_RETRY);
1809 	error = VOP_STAT(vp, sb, curthread->td_ucred, NOCRED);
1810 	VOP_UNLOCK(vp);
1811 
1812 	return (error);
1813 }
1814 
1815 static int
1816 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1817     struct filedesc *fdp)
1818 {
1819 	struct linux_file *filp;
1820 	struct vnode *vp;
1821 	int error;
1822 
1823 	filp = fp->f_data;
1824 	vp = filp->f_vnode;
1825 	if (vp == NULL) {
1826 		error = 0;
1827 		kif->kf_type = KF_TYPE_DEV;
1828 	} else {
1829 		vref(vp);
1830 		FILEDESC_SUNLOCK(fdp);
1831 		error = vn_fill_kinfo_vnode(vp, kif);
1832 		vrele(vp);
1833 		kif->kf_type = KF_TYPE_VNODE;
1834 		FILEDESC_SLOCK(fdp);
1835 	}
1836 	return (error);
1837 }
1838 
1839 unsigned int
1840 linux_iminor(struct inode *inode)
1841 {
1842 	struct linux_cdev *ldev;
1843 
1844 	if (inode == NULL || inode->v_rdev == NULL ||
1845 	    inode->v_rdev->si_devsw != &linuxcdevsw)
1846 		return (-1U);
1847 	ldev = inode->v_rdev->si_drv1;
1848 	if (ldev == NULL)
1849 		return (-1U);
1850 
1851 	return (minor(ldev->dev));
1852 }
1853 
1854 static int
1855 linux_file_kcmp(struct file *fp1, struct file *fp2, struct thread *td)
1856 {
1857 	struct linux_file *filp1, *filp2;
1858 
1859 	if (fp2->f_type != DTYPE_DEV)
1860 		return (3);
1861 
1862 	filp1 = fp1->f_data;
1863 	filp2 = fp2->f_data;
1864 	return (kcmp_cmp((uintptr_t)filp1->f_cdev, (uintptr_t)filp2->f_cdev));
1865 }
1866 
1867 const struct fileops linuxfileops = {
1868 	.fo_read = linux_file_read,
1869 	.fo_write = linux_file_write,
1870 	.fo_truncate = invfo_truncate,
1871 	.fo_kqfilter = linux_file_kqfilter,
1872 	.fo_stat = linux_file_stat,
1873 	.fo_fill_kinfo = linux_file_fill_kinfo,
1874 	.fo_poll = linux_file_poll,
1875 	.fo_close = linux_file_close,
1876 	.fo_ioctl = linux_file_ioctl,
1877 	.fo_mmap = linux_file_mmap,
1878 	.fo_chmod = invfo_chmod,
1879 	.fo_chown = invfo_chown,
1880 	.fo_sendfile = invfo_sendfile,
1881 	.fo_cmp = linux_file_kcmp,
1882 	.fo_flags = DFLAG_PASSABLE,
1883 };
1884 
1885 static char *
1886 devm_kvasprintf(struct device *dev, gfp_t gfp, const char *fmt, va_list ap)
1887 {
1888 	unsigned int len;
1889 	char *p;
1890 	va_list aq;
1891 
1892 	va_copy(aq, ap);
1893 	len = vsnprintf(NULL, 0, fmt, aq);
1894 	va_end(aq);
1895 
1896 	if (dev != NULL)
1897 		p = devm_kmalloc(dev, len + 1, gfp);
1898 	else
1899 		p = kmalloc(len + 1, gfp);
1900 	if (p != NULL)
1901 		vsnprintf(p, len + 1, fmt, ap);
1902 
1903 	return (p);
1904 }
1905 
1906 char *
1907 kvasprintf(gfp_t gfp, const char *fmt, va_list ap)
1908 {
1909 
1910 	return (devm_kvasprintf(NULL, gfp, fmt, ap));
1911 }
1912 
1913 char *
1914 lkpi_devm_kasprintf(struct device *dev, gfp_t gfp, const char *fmt, ...)
1915 {
1916 	va_list ap;
1917 	char *p;
1918 
1919 	va_start(ap, fmt);
1920 	p = devm_kvasprintf(dev, gfp, fmt, ap);
1921 	va_end(ap);
1922 
1923 	return (p);
1924 }
1925 
1926 char *
1927 kasprintf(gfp_t gfp, const char *fmt, ...)
1928 {
1929 	va_list ap;
1930 	char *p;
1931 
1932 	va_start(ap, fmt);
1933 	p = kvasprintf(gfp, fmt, ap);
1934 	va_end(ap);
1935 
1936 	return (p);
1937 }
1938 
1939 int
1940 __lkpi_hexdump_printf(void *arg1 __unused, const char *fmt, ...)
1941 {
1942 	va_list ap;
1943 	int result;
1944 
1945 	va_start(ap, fmt);
1946 	result = vprintf(fmt, ap);
1947 	va_end(ap);
1948 	return (result);
1949 }
1950 
1951 int
1952 __lkpi_hexdump_sbuf_printf(void *arg1, const char *fmt, ...)
1953 {
1954 	va_list ap;
1955 	int result;
1956 
1957 	va_start(ap, fmt);
1958 	result = sbuf_vprintf(arg1, fmt, ap);
1959 	va_end(ap);
1960 	return (result);
1961 }
1962 
1963 void
1964 lkpi_hex_dump(int(*_fpf)(void *, const char *, ...), void *arg1,
1965     const char *level, const char *prefix_str,
1966     const int prefix_type, const int rowsize, const int groupsize,
1967     const void *buf, size_t len, const bool ascii, const bool trailing_newline)
1968 {
1969 	typedef const struct { long long value; } __packed *print_64p_t;
1970 	typedef const struct { uint32_t value; } __packed *print_32p_t;
1971 	typedef const struct { uint16_t value; } __packed *print_16p_t;
1972 	const void *buf_old = buf;
1973 	int row, linelen, ret;
1974 
1975 	while (len > 0) {
1976 		linelen = 0;
1977 		if (level != NULL) {
1978 			ret = _fpf(arg1, "%s", level);
1979 			if (ret < 0)
1980 				break;
1981 			linelen += ret;
1982 		}
1983 		if (prefix_str != NULL) {
1984 			ret = _fpf(
1985 			    arg1, "%s%s", linelen ? " " : "", prefix_str);
1986 			if (ret < 0)
1987 				break;
1988 			linelen += ret;
1989 		}
1990 
1991 		switch (prefix_type) {
1992 		case DUMP_PREFIX_ADDRESS:
1993 			ret = _fpf(
1994 			    arg1, "%s[%p]", linelen ? " " : "", buf);
1995 			if (ret < 0)
1996 				return;
1997 			linelen += ret;
1998 			break;
1999 		case DUMP_PREFIX_OFFSET:
2000 			ret = _fpf(
2001 			    arg1, "%s[%#tx]", linelen ? " " : "",
2002 			    ((const char *)buf - (const char *)buf_old));
2003 			if (ret < 0)
2004 				return;
2005 			linelen += ret;
2006 			break;
2007 		default:
2008 			break;
2009 		}
2010 		for (row = 0; row != rowsize; row++) {
2011 			if (groupsize == 8 && len > 7) {
2012 				ret = _fpf(
2013 				    arg1, "%s%016llx", linelen ? " " : "",
2014 				    ((print_64p_t)buf)->value);
2015 				if (ret < 0)
2016 					return;
2017 				linelen += ret;
2018 				buf = (const uint8_t *)buf + 8;
2019 				len -= 8;
2020 			} else if (groupsize == 4 && len > 3) {
2021 				ret = _fpf(
2022 				    arg1, "%s%08x", linelen ? " " : "",
2023 				    ((print_32p_t)buf)->value);
2024 				if (ret < 0)
2025 					return;
2026 				linelen += ret;
2027 				buf = (const uint8_t *)buf + 4;
2028 				len -= 4;
2029 			} else if (groupsize == 2 && len > 1) {
2030 				ret = _fpf(
2031 				    arg1, "%s%04x", linelen ? " " : "",
2032 				    ((print_16p_t)buf)->value);
2033 				if (ret < 0)
2034 					return;
2035 				linelen += ret;
2036 				buf = (const uint8_t *)buf + 2;
2037 				len -= 2;
2038 			} else if (len > 0) {
2039 				ret = _fpf(
2040 				    arg1, "%s%02x", linelen ? " " : "",
2041 				    *(const uint8_t *)buf);
2042 				if (ret < 0)
2043 					return;
2044 				linelen += ret;
2045 				buf = (const uint8_t *)buf + 1;
2046 				len--;
2047 			} else {
2048 				break;
2049 			}
2050 		}
2051 		if (len > 0 && trailing_newline) {
2052 			ret = _fpf(arg1, "\n");
2053 			if (ret < 0)
2054 				break;
2055 		}
2056 	}
2057 }
2058 
2059 struct hdtb_context {
2060 	char	*linebuf;
2061 	size_t	 linebuflen;
2062 	int	 written;
2063 };
2064 
2065 static int
2066 hdtb_cb(void *arg, const char *format, ...)
2067 {
2068 	struct hdtb_context *context;
2069 	int written;
2070 	va_list args;
2071 
2072 	context = arg;
2073 
2074 	va_start(args, format);
2075 	written = vsnprintf(
2076 	    context->linebuf, context->linebuflen, format, args);
2077 	va_end(args);
2078 
2079 	if (written < 0)
2080 		return (written);
2081 
2082 	/*
2083 	 * Linux' hex_dump_to_buffer() function has the same behaviour as
2084 	 * snprintf() basically. Therefore, it returns the number of bytes it
2085 	 * would have written if the destination buffer was large enough.
2086 	 *
2087 	 * If the destination buffer was exhausted, lkpi_hex_dump() will
2088 	 * continue to call this callback but it will only compute the bytes it
2089 	 * would have written but write nothing to that buffer.
2090 	 */
2091 	context->written += written;
2092 
2093 	if (written < context->linebuflen) {
2094 		context->linebuf += written;
2095 		context->linebuflen -= written;
2096 	} else {
2097 		context->linebuf += context->linebuflen;
2098 		context->linebuflen = 0;
2099 	}
2100 
2101 	return (written);
2102 }
2103 
2104 int
2105 lkpi_hex_dump_to_buffer(const void *buf, size_t len, int rowsize,
2106     int groupsize, char *linebuf, size_t linebuflen, bool ascii)
2107 {
2108 	int written;
2109 	struct hdtb_context context;
2110 
2111 	context.linebuf = linebuf;
2112 	context.linebuflen = linebuflen;
2113 	context.written = 0;
2114 
2115 	if (rowsize != 16 && rowsize != 32)
2116 		rowsize = 16;
2117 
2118 	len = min(len, rowsize);
2119 
2120 	lkpi_hex_dump(
2121 	    hdtb_cb, &context, NULL, NULL, DUMP_PREFIX_NONE,
2122 	    rowsize, groupsize, buf, len, ascii, false);
2123 
2124 	written = context.written;
2125 
2126 	return (written);
2127 }
2128 
2129 static void
2130 linux_timer_callback_wrapper(void *context)
2131 {
2132 	struct timer_list *timer;
2133 
2134 	timer = context;
2135 
2136 	/* the timer is about to be shutdown permanently */
2137 	if (timer->function == NULL)
2138 		return;
2139 
2140 	if (linux_set_current_flags(curthread, M_NOWAIT)) {
2141 		/* try again later */
2142 		callout_reset(&timer->callout, 1,
2143 		    &linux_timer_callback_wrapper, timer);
2144 		return;
2145 	}
2146 
2147 	timer->function(timer->data);
2148 }
2149 
2150 static int
2151 linux_timer_jiffies_until(unsigned long expires)
2152 {
2153 	unsigned long delta = expires - jiffies;
2154 
2155 	/*
2156 	 * Guard against already expired values and make sure that the value can
2157 	 * be used as a tick count, rather than a jiffies count.
2158 	 */
2159 	if ((long)delta < 1)
2160 		delta = 1;
2161 	else if (delta > INT_MAX)
2162 		delta = INT_MAX;
2163 	return ((int)delta);
2164 }
2165 
2166 int
2167 mod_timer(struct timer_list *timer, unsigned long expires)
2168 {
2169 	int ret;
2170 
2171 	timer->expires = expires;
2172 	ret = callout_reset(&timer->callout,
2173 	    linux_timer_jiffies_until(expires),
2174 	    &linux_timer_callback_wrapper, timer);
2175 
2176 	MPASS(ret == 0 || ret == 1);
2177 
2178 	return (ret == 1);
2179 }
2180 
2181 void
2182 add_timer(struct timer_list *timer)
2183 {
2184 
2185 	callout_reset(&timer->callout,
2186 	    linux_timer_jiffies_until(timer->expires),
2187 	    &linux_timer_callback_wrapper, timer);
2188 }
2189 
2190 void
2191 add_timer_on(struct timer_list *timer, int cpu)
2192 {
2193 
2194 	callout_reset_on(&timer->callout,
2195 	    linux_timer_jiffies_until(timer->expires),
2196 	    &linux_timer_callback_wrapper, timer, cpu);
2197 }
2198 
2199 int
2200 timer_delete(struct timer_list *timer)
2201 {
2202 
2203 	if (callout_stop(&(timer)->callout) == -1)
2204 		return (0);
2205 	return (1);
2206 }
2207 
2208 int
2209 timer_delete_sync(struct timer_list *timer)
2210 {
2211 
2212 	if (callout_drain(&(timer)->callout) == -1)
2213 		return (0);
2214 	return (1);
2215 }
2216 
2217 int
2218 timer_shutdown_sync(struct timer_list *timer)
2219 {
2220 
2221 	timer->function = NULL;
2222 	return (del_timer_sync(timer));
2223 }
2224 
2225 /* greatest common divisor, Euclid equation */
2226 static uint64_t
2227 lkpi_gcd_64(uint64_t a, uint64_t b)
2228 {
2229 	uint64_t an;
2230 	uint64_t bn;
2231 
2232 	while (b != 0) {
2233 		an = b;
2234 		bn = a % b;
2235 		a = an;
2236 		b = bn;
2237 	}
2238 	return (a);
2239 }
2240 
2241 uint64_t lkpi_nsec2hz_rem;
2242 uint64_t lkpi_nsec2hz_div = 1000000000ULL;
2243 uint64_t lkpi_nsec2hz_max;
2244 
2245 uint64_t lkpi_usec2hz_rem;
2246 uint64_t lkpi_usec2hz_div = 1000000ULL;
2247 uint64_t lkpi_usec2hz_max;
2248 
2249 uint64_t lkpi_msec2hz_rem;
2250 uint64_t lkpi_msec2hz_div = 1000ULL;
2251 uint64_t lkpi_msec2hz_max;
2252 
2253 static void
2254 linux_timer_init(void *arg)
2255 {
2256 	uint64_t gcd;
2257 
2258 	/*
2259 	 * Compute an internal HZ value which can divide 2**32 to
2260 	 * avoid timer rounding problems when the tick value wraps
2261 	 * around 2**32:
2262 	 */
2263 	linux_timer_hz_mask = 1;
2264 	while (linux_timer_hz_mask < (unsigned long)hz)
2265 		linux_timer_hz_mask *= 2;
2266 	linux_timer_hz_mask--;
2267 
2268 	/* compute some internal constants */
2269 
2270 	lkpi_nsec2hz_rem = hz;
2271 	lkpi_usec2hz_rem = hz;
2272 	lkpi_msec2hz_rem = hz;
2273 
2274 	gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div);
2275 	lkpi_nsec2hz_rem /= gcd;
2276 	lkpi_nsec2hz_div /= gcd;
2277 	lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem;
2278 
2279 	gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div);
2280 	lkpi_usec2hz_rem /= gcd;
2281 	lkpi_usec2hz_div /= gcd;
2282 	lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem;
2283 
2284 	gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div);
2285 	lkpi_msec2hz_rem /= gcd;
2286 	lkpi_msec2hz_div /= gcd;
2287 	lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem;
2288 }
2289 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL);
2290 
2291 void
2292 linux_complete_common(struct completion *c, int all)
2293 {
2294 	sleepq_lock(c);
2295 	if (all) {
2296 		c->done = UINT_MAX;
2297 		sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0);
2298 	} else {
2299 		if (c->done != UINT_MAX)
2300 			c->done++;
2301 		sleepq_signal(c, SLEEPQ_SLEEP, 0, 0);
2302 	}
2303 	sleepq_release(c);
2304 }
2305 
2306 /*
2307  * Indefinite wait for done != 0 with or without signals.
2308  */
2309 int
2310 linux_wait_for_common(struct completion *c, int flags)
2311 {
2312 	struct task_struct *task;
2313 	int error;
2314 
2315 	if (SCHEDULER_STOPPED())
2316 		return (0);
2317 
2318 	task = current;
2319 
2320 	if (flags != 0)
2321 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2322 	else
2323 		flags = SLEEPQ_SLEEP;
2324 	error = 0;
2325 	for (;;) {
2326 		sleepq_lock(c);
2327 		if (c->done)
2328 			break;
2329 		sleepq_add(c, NULL, "completion", flags, 0);
2330 		if (flags & SLEEPQ_INTERRUPTIBLE) {
2331 			DROP_GIANT();
2332 			error = -sleepq_wait_sig(c, 0);
2333 			PICKUP_GIANT();
2334 			if (error != 0) {
2335 				linux_schedule_save_interrupt_value(task, error);
2336 				error = -ERESTARTSYS;
2337 				goto intr;
2338 			}
2339 		} else {
2340 			DROP_GIANT();
2341 			sleepq_wait(c, 0);
2342 			PICKUP_GIANT();
2343 		}
2344 	}
2345 	if (c->done != UINT_MAX)
2346 		c->done--;
2347 	sleepq_release(c);
2348 
2349 intr:
2350 	return (error);
2351 }
2352 
2353 /*
2354  * Time limited wait for done != 0 with or without signals.
2355  */
2356 unsigned long
2357 linux_wait_for_timeout_common(struct completion *c, unsigned long timeout,
2358     int flags)
2359 {
2360 	struct task_struct *task;
2361 	unsigned long end = jiffies + timeout, error;
2362 
2363 	if (SCHEDULER_STOPPED())
2364 		return (0);
2365 
2366 	task = current;
2367 
2368 	if (flags != 0)
2369 		flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2370 	else
2371 		flags = SLEEPQ_SLEEP;
2372 
2373 	for (;;) {
2374 		sleepq_lock(c);
2375 		if (c->done)
2376 			break;
2377 		sleepq_add(c, NULL, "completion", flags, 0);
2378 		sleepq_set_timeout(c, linux_timer_jiffies_until(end));
2379 
2380 		DROP_GIANT();
2381 		if (flags & SLEEPQ_INTERRUPTIBLE)
2382 			error = -sleepq_timedwait_sig(c, 0);
2383 		else
2384 			error = -sleepq_timedwait(c, 0);
2385 		PICKUP_GIANT();
2386 
2387 		if (error != 0) {
2388 			/* check for timeout */
2389 			if (error == -EWOULDBLOCK) {
2390 				error = 0;	/* timeout */
2391 			} else {
2392 				/* signal happened */
2393 				linux_schedule_save_interrupt_value(task, error);
2394 				error = -ERESTARTSYS;
2395 			}
2396 			goto done;
2397 		}
2398 	}
2399 	if (c->done != UINT_MAX)
2400 		c->done--;
2401 	sleepq_release(c);
2402 
2403 	/* return how many jiffies are left */
2404 	error = linux_timer_jiffies_until(end);
2405 done:
2406 	return (error);
2407 }
2408 
2409 int
2410 linux_try_wait_for_completion(struct completion *c)
2411 {
2412 	int isdone;
2413 
2414 	sleepq_lock(c);
2415 	isdone = (c->done != 0);
2416 	if (c->done != 0 && c->done != UINT_MAX)
2417 		c->done--;
2418 	sleepq_release(c);
2419 	return (isdone);
2420 }
2421 
2422 int
2423 linux_completion_done(struct completion *c)
2424 {
2425 	int isdone;
2426 
2427 	sleepq_lock(c);
2428 	isdone = (c->done != 0);
2429 	sleepq_release(c);
2430 	return (isdone);
2431 }
2432 
2433 static void
2434 linux_cdev_deref(struct linux_cdev *ldev)
2435 {
2436 	if (refcount_release(&ldev->refs) &&
2437 	    ldev->kobj.ktype == &linux_cdev_ktype)
2438 		kfree(ldev);
2439 }
2440 
2441 static void
2442 linux_cdev_release(struct kobject *kobj)
2443 {
2444 	struct linux_cdev *cdev;
2445 	struct kobject *parent;
2446 
2447 	cdev = container_of(kobj, struct linux_cdev, kobj);
2448 	parent = kobj->parent;
2449 	linux_destroy_dev(cdev);
2450 	linux_cdev_deref(cdev);
2451 	kobject_put(parent);
2452 }
2453 
2454 static void
2455 linux_cdev_static_release(struct kobject *kobj)
2456 {
2457 	struct cdev *cdev;
2458 	struct linux_cdev *ldev;
2459 
2460 	ldev = container_of(kobj, struct linux_cdev, kobj);
2461 	cdev = ldev->cdev;
2462 	if (cdev != NULL) {
2463 		destroy_dev(cdev);
2464 		ldev->cdev = NULL;
2465 	}
2466 	kobject_put(kobj->parent);
2467 }
2468 
2469 int
2470 linux_cdev_device_add(struct linux_cdev *ldev, struct device *dev)
2471 {
2472 	int ret;
2473 
2474 	if (dev->devt != 0) {
2475 		/* Set parent kernel object. */
2476 		ldev->kobj.parent = &dev->kobj;
2477 
2478 		/*
2479 		 * Unlike Linux we require the kobject of the
2480 		 * character device structure to have a valid name
2481 		 * before calling this function:
2482 		 */
2483 		if (ldev->kobj.name == NULL)
2484 			return (-EINVAL);
2485 
2486 		ret = cdev_add(ldev, dev->devt, 1);
2487 		if (ret)
2488 			return (ret);
2489 	}
2490 	ret = device_add(dev);
2491 	if (ret != 0 && dev->devt != 0)
2492 		cdev_del(ldev);
2493 	return (ret);
2494 }
2495 
2496 void
2497 linux_cdev_device_del(struct linux_cdev *ldev, struct device *dev)
2498 {
2499 	device_del(dev);
2500 
2501 	if (dev->devt != 0)
2502 		cdev_del(ldev);
2503 }
2504 
2505 static void
2506 linux_destroy_dev(struct linux_cdev *ldev)
2507 {
2508 
2509 	if (ldev->cdev == NULL)
2510 		return;
2511 
2512 	MPASS((ldev->siref & LDEV_SI_DTR) == 0);
2513 	MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
2514 
2515 	atomic_set_int(&ldev->siref, LDEV_SI_DTR);
2516 	while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0)
2517 		pause("ldevdtr", hz / 4);
2518 
2519 	destroy_dev(ldev->cdev);
2520 	ldev->cdev = NULL;
2521 }
2522 
2523 const struct kobj_type linux_cdev_ktype = {
2524 	.release = linux_cdev_release,
2525 };
2526 
2527 const struct kobj_type linux_cdev_static_ktype = {
2528 	.release = linux_cdev_static_release,
2529 };
2530 
2531 static void
2532 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate)
2533 {
2534 	struct notifier_block *nb;
2535 	struct netdev_notifier_info ni;
2536 
2537 	nb = arg;
2538 	ni.ifp = ifp;
2539 	ni.dev = (struct net_device *)ifp;
2540 	if (linkstate == LINK_STATE_UP)
2541 		nb->notifier_call(nb, NETDEV_UP, &ni);
2542 	else
2543 		nb->notifier_call(nb, NETDEV_DOWN, &ni);
2544 }
2545 
2546 static void
2547 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp)
2548 {
2549 	struct notifier_block *nb;
2550 	struct netdev_notifier_info ni;
2551 
2552 	nb = arg;
2553 	ni.ifp = ifp;
2554 	ni.dev = (struct net_device *)ifp;
2555 	nb->notifier_call(nb, NETDEV_REGISTER, &ni);
2556 }
2557 
2558 static void
2559 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp)
2560 {
2561 	struct notifier_block *nb;
2562 	struct netdev_notifier_info ni;
2563 
2564 	nb = arg;
2565 	ni.ifp = ifp;
2566 	ni.dev = (struct net_device *)ifp;
2567 	nb->notifier_call(nb, NETDEV_UNREGISTER, &ni);
2568 }
2569 
2570 static void
2571 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp)
2572 {
2573 	struct notifier_block *nb;
2574 	struct netdev_notifier_info ni;
2575 
2576 	nb = arg;
2577 	ni.ifp = ifp;
2578 	ni.dev = (struct net_device *)ifp;
2579 	nb->notifier_call(nb, NETDEV_CHANGEADDR, &ni);
2580 }
2581 
2582 static void
2583 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp)
2584 {
2585 	struct notifier_block *nb;
2586 	struct netdev_notifier_info ni;
2587 
2588 	nb = arg;
2589 	ni.ifp = ifp;
2590 	ni.dev = (struct net_device *)ifp;
2591 	nb->notifier_call(nb, NETDEV_CHANGEIFADDR, &ni);
2592 }
2593 
2594 int
2595 register_netdevice_notifier(struct notifier_block *nb)
2596 {
2597 
2598 	nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER(
2599 	    ifnet_link_event, linux_handle_ifnet_link_event, nb, 0);
2600 	nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER(
2601 	    ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0);
2602 	nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER(
2603 	    ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0);
2604 	nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER(
2605 	    iflladdr_event, linux_handle_iflladdr_event, nb, 0);
2606 
2607 	return (0);
2608 }
2609 
2610 int
2611 register_inetaddr_notifier(struct notifier_block *nb)
2612 {
2613 
2614 	nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER(
2615 	    ifaddr_event, linux_handle_ifaddr_event, nb, 0);
2616 	return (0);
2617 }
2618 
2619 int
2620 unregister_netdevice_notifier(struct notifier_block *nb)
2621 {
2622 
2623 	EVENTHANDLER_DEREGISTER(ifnet_link_event,
2624 	    nb->tags[NETDEV_UP]);
2625 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event,
2626 	    nb->tags[NETDEV_REGISTER]);
2627 	EVENTHANDLER_DEREGISTER(ifnet_departure_event,
2628 	    nb->tags[NETDEV_UNREGISTER]);
2629 	EVENTHANDLER_DEREGISTER(iflladdr_event,
2630 	    nb->tags[NETDEV_CHANGEADDR]);
2631 
2632 	return (0);
2633 }
2634 
2635 int
2636 unregister_inetaddr_notifier(struct notifier_block *nb)
2637 {
2638 
2639 	EVENTHANDLER_DEREGISTER(ifaddr_event,
2640 	    nb->tags[NETDEV_CHANGEIFADDR]);
2641 
2642 	return (0);
2643 }
2644 
2645 struct list_sort_thunk {
2646 	int (*cmp)(void *, struct list_head *, struct list_head *);
2647 	void *priv;
2648 };
2649 
2650 static inline int
2651 linux_le_cmp(const void *d1, const void *d2, void *priv)
2652 {
2653 	struct list_head *le1, *le2;
2654 	struct list_sort_thunk *thunk;
2655 
2656 	thunk = priv;
2657 	le1 = *(__DECONST(struct list_head **, d1));
2658 	le2 = *(__DECONST(struct list_head **, d2));
2659 	return ((thunk->cmp)(thunk->priv, le1, le2));
2660 }
2661 
2662 void
2663 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv,
2664     struct list_head *a, struct list_head *b))
2665 {
2666 	struct list_sort_thunk thunk;
2667 	struct list_head **ar, *le;
2668 	size_t count, i;
2669 
2670 	count = 0;
2671 	list_for_each(le, head)
2672 		count++;
2673 	ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK);
2674 	i = 0;
2675 	list_for_each(le, head)
2676 		ar[i++] = le;
2677 	thunk.cmp = cmp;
2678 	thunk.priv = priv;
2679 	qsort_r(ar, count, sizeof(struct list_head *), linux_le_cmp, &thunk);
2680 	INIT_LIST_HEAD(head);
2681 	for (i = 0; i < count; i++)
2682 		list_add_tail(ar[i], head);
2683 	free(ar, M_KMALLOC);
2684 }
2685 
2686 #if defined(__i386__) || defined(__amd64__)
2687 int
2688 linux_wbinvd_on_all_cpus(void)
2689 {
2690 
2691 	pmap_invalidate_cache();
2692 	return (0);
2693 }
2694 #endif
2695 
2696 int
2697 linux_on_each_cpu(void callback(void *), void *data)
2698 {
2699 
2700 	smp_rendezvous(smp_no_rendezvous_barrier, callback,
2701 	    smp_no_rendezvous_barrier, data);
2702 	return (0);
2703 }
2704 
2705 int
2706 linux_in_atomic(void)
2707 {
2708 
2709 	return ((curthread->td_pflags & TDP_NOFAULTING) != 0);
2710 }
2711 
2712 struct linux_cdev *
2713 linux_find_cdev(const char *name, unsigned major, unsigned minor)
2714 {
2715 	dev_t dev = MKDEV(major, minor);
2716 	struct cdev *cdev;
2717 
2718 	dev_lock();
2719 	LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) {
2720 		struct linux_cdev *ldev = cdev->si_drv1;
2721 		if (ldev->dev == dev &&
2722 		    strcmp(kobject_name(&ldev->kobj), name) == 0) {
2723 			break;
2724 		}
2725 	}
2726 	dev_unlock();
2727 
2728 	return (cdev != NULL ? cdev->si_drv1 : NULL);
2729 }
2730 
2731 int
2732 __register_chrdev(unsigned int major, unsigned int baseminor,
2733     unsigned int count, const char *name,
2734     const struct file_operations *fops)
2735 {
2736 	struct linux_cdev *cdev;
2737 	int ret = 0;
2738 	int i;
2739 
2740 	for (i = baseminor; i < baseminor + count; i++) {
2741 		cdev = cdev_alloc();
2742 		cdev->ops = fops;
2743 		kobject_set_name(&cdev->kobj, name);
2744 
2745 		ret = cdev_add(cdev, makedev(major, i), 1);
2746 		if (ret != 0)
2747 			break;
2748 	}
2749 	return (ret);
2750 }
2751 
2752 int
2753 __register_chrdev_p(unsigned int major, unsigned int baseminor,
2754     unsigned int count, const char *name,
2755     const struct file_operations *fops, uid_t uid,
2756     gid_t gid, int mode)
2757 {
2758 	struct linux_cdev *cdev;
2759 	int ret = 0;
2760 	int i;
2761 
2762 	for (i = baseminor; i < baseminor + count; i++) {
2763 		cdev = cdev_alloc();
2764 		cdev->ops = fops;
2765 		kobject_set_name(&cdev->kobj, name);
2766 
2767 		ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode);
2768 		if (ret != 0)
2769 			break;
2770 	}
2771 	return (ret);
2772 }
2773 
2774 void
2775 __unregister_chrdev(unsigned int major, unsigned int baseminor,
2776     unsigned int count, const char *name)
2777 {
2778 	struct linux_cdev *cdevp;
2779 	int i;
2780 
2781 	for (i = baseminor; i < baseminor + count; i++) {
2782 		cdevp = linux_find_cdev(name, major, i);
2783 		if (cdevp != NULL)
2784 			cdev_del(cdevp);
2785 	}
2786 }
2787 
2788 void
2789 linux_dump_stack(void)
2790 {
2791 #ifdef STACK
2792 	struct stack st;
2793 
2794 	stack_save(&st);
2795 	stack_print(&st);
2796 #endif
2797 }
2798 
2799 int
2800 linuxkpi_net_ratelimit(void)
2801 {
2802 
2803 	return (ppsratecheck(&lkpi_net_lastlog, &lkpi_net_curpps,
2804 	   lkpi_net_maxpps));
2805 }
2806 
2807 struct io_mapping *
2808 io_mapping_create_wc(resource_size_t base, unsigned long size)
2809 {
2810 	struct io_mapping *mapping;
2811 
2812 	mapping = kmalloc(sizeof(*mapping), GFP_KERNEL);
2813 	if (mapping == NULL)
2814 		return (NULL);
2815 	return (io_mapping_init_wc(mapping, base, size));
2816 }
2817 
2818 /* We likely want a linuxkpi_device.c at some point. */
2819 bool
2820 device_can_wakeup(struct device *dev)
2821 {
2822 
2823 	if (dev == NULL)
2824 		return (false);
2825 	/*
2826 	 * XXX-BZ iwlwifi queries it as part of enabling WoWLAN.
2827 	 * Normally this would be based on a bool in dev->power.XXX.
2828 	 * Check such as PCI PCIM_PCAP_*PME.  We have no way to enable this yet.
2829 	 * We may get away by directly calling into bsddev for as long as
2830 	 * we can assume PCI only avoiding changing struct device breaking KBI.
2831 	 */
2832 	pr_debug("%s:%d: not enabled; see comment.\n", __func__, __LINE__);
2833 	return (false);
2834 }
2835 
2836 static void
2837 devm_device_group_remove(struct device *dev, void *p)
2838 {
2839 	const struct attribute_group **dr = p;
2840 	const struct attribute_group *group = *dr;
2841 
2842 	sysfs_remove_group(&dev->kobj, group);
2843 }
2844 
2845 int
2846 lkpi_devm_device_add_group(struct device *dev,
2847     const struct attribute_group *group)
2848 {
2849 	const struct attribute_group **dr;
2850 	int ret;
2851 
2852 	dr = devres_alloc(devm_device_group_remove, sizeof(*dr), GFP_KERNEL);
2853 	if (dr == NULL)
2854 		return (-ENOMEM);
2855 
2856 	ret = sysfs_create_group(&dev->kobj, group);
2857 	if (ret == 0) {
2858 		*dr = group;
2859 		devres_add(dev, dr);
2860 	} else
2861 		devres_free(dr);
2862 
2863 	return (ret);
2864 }
2865 
2866 #if defined(__i386__) || defined(__amd64__)
2867 bool linux_cpu_has_clflush;
2868 struct cpuinfo_x86 boot_cpu_data;
2869 struct cpuinfo_x86 *__cpu_data;
2870 #endif
2871 
2872 cpumask_t *
2873 lkpi_get_static_single_cpu_mask(int cpuid)
2874 {
2875 
2876 	KASSERT((cpuid >= 0 && cpuid <= mp_maxid), ("%s: invalid cpuid %d\n",
2877 	    __func__, cpuid));
2878 	KASSERT(!CPU_ABSENT(cpuid), ("%s: cpu with cpuid %d is absent\n",
2879 	    __func__, cpuid));
2880 
2881 	return (static_single_cpu_mask[cpuid]);
2882 }
2883 
2884 bool
2885 lkpi_xen_initial_domain(void)
2886 {
2887 #ifdef XENHVM
2888 	return (xen_initial_domain());
2889 #else
2890 	return (false);
2891 #endif
2892 }
2893 
2894 bool
2895 lkpi_xen_pv_domain(void)
2896 {
2897 #ifdef XENHVM
2898 	return (xen_pv_domain());
2899 #else
2900 	return (false);
2901 #endif
2902 }
2903 
2904 static void
2905 linux_compat_init(void *arg)
2906 {
2907 	struct sysctl_oid *rootoid;
2908 	int i;
2909 
2910 #if defined(__i386__) || defined(__amd64__)
2911 	static const uint32_t x86_vendors[X86_VENDOR_NUM] = {
2912 		[X86_VENDOR_INTEL] = CPU_VENDOR_INTEL,
2913 		[X86_VENDOR_CYRIX] = CPU_VENDOR_CYRIX,
2914 		[X86_VENDOR_AMD] = CPU_VENDOR_AMD,
2915 		[X86_VENDOR_UMC] = CPU_VENDOR_UMC,
2916 		[X86_VENDOR_CENTAUR] = CPU_VENDOR_CENTAUR,
2917 		[X86_VENDOR_TRANSMETA] = CPU_VENDOR_TRANSMETA,
2918 		[X86_VENDOR_NSC] = CPU_VENDOR_NSC,
2919 		[X86_VENDOR_HYGON] = CPU_VENDOR_HYGON,
2920 	};
2921 	uint8_t x86_vendor = X86_VENDOR_UNKNOWN;
2922 
2923 	for (i = 0; i < X86_VENDOR_NUM; i++) {
2924 		if (cpu_vendor_id != 0 && cpu_vendor_id == x86_vendors[i]) {
2925 			x86_vendor = i;
2926 			break;
2927 		}
2928 	}
2929 	linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH);
2930 	boot_cpu_data.x86_clflush_size = cpu_clflush_line_size;
2931 	boot_cpu_data.x86_max_cores = mp_ncpus;
2932 	boot_cpu_data.x86 = CPUID_TO_FAMILY(cpu_id);
2933 	boot_cpu_data.x86_model = CPUID_TO_MODEL(cpu_id);
2934 	boot_cpu_data.x86_vendor = x86_vendor;
2935 	boot_cpu_data.x86_stepping = CPUID_TO_STEPPING(cpu_id);
2936 
2937 	__cpu_data = kmalloc_array(mp_maxid + 1,
2938 	    sizeof(*__cpu_data), M_WAITOK | M_ZERO);
2939 	CPU_FOREACH(i) {
2940 		__cpu_data[i].x86_clflush_size = cpu_clflush_line_size;
2941 		__cpu_data[i].x86_max_cores = mp_ncpus;
2942 		__cpu_data[i].x86 = CPUID_TO_FAMILY(cpu_id);
2943 		__cpu_data[i].x86_model = CPUID_TO_MODEL(cpu_id);
2944 		__cpu_data[i].x86_vendor = x86_vendor;
2945 	}
2946 #endif
2947 	rw_init(&linux_vma_lock, "lkpi-vma-lock");
2948 
2949 	rootoid = SYSCTL_ADD_ROOT_NODE(NULL,
2950 	    OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys");
2951 	kobject_init(&linux_class_root, &linux_class_ktype);
2952 	kobject_set_name(&linux_class_root, "class");
2953 	linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid),
2954 	    OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class");
2955 	kobject_init(&linux_root_device.kobj, &linux_dev_ktype);
2956 	kobject_set_name(&linux_root_device.kobj, "device");
2957 	linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL,
2958 	    SYSCTL_CHILDREN(rootoid), OID_AUTO, "device",
2959 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "device");
2960 	linux_root_device.bsddev = root_bus;
2961 	linux_class_misc.name = "misc";
2962 	class_register(&linux_class_misc);
2963 	INIT_LIST_HEAD(&pci_drivers);
2964 	INIT_LIST_HEAD(&pci_devices);
2965 	spin_lock_init(&pci_lock);
2966 	init_waitqueue_head(&linux_bit_waitq);
2967 	init_waitqueue_head(&linux_var_waitq);
2968 
2969 	CPU_COPY(&all_cpus, &cpu_online_mask);
2970 	/*
2971 	 * Generate a single-CPU cpumask_t for each CPU (possibly) in the system.
2972 	 * CPUs are indexed from 0..(mp_maxid).  The entry for cpuid 0 will only
2973 	 * have itself in the cpumask, cupid 1 only itself on entry 1, and so on.
2974 	 * This is used by cpumask_of() (and possibly others in the future) for,
2975 	 * e.g., drivers to pass hints to irq_set_affinity_hint().
2976 	 */
2977 	static_single_cpu_mask = kmalloc_array(mp_maxid + 1,
2978 	    sizeof(static_single_cpu_mask), M_WAITOK | M_ZERO);
2979 
2980 	/*
2981 	 * When the number of CPUs reach a threshold, we start to save memory
2982 	 * given the sets are static by overlapping those having their single
2983 	 * bit set at same position in a bitset word.  Asymptotically, this
2984 	 * regular scheme is in O(n²) whereas the overlapping one is in O(n)
2985 	 * only with n being the maximum number of CPUs, so the gain will become
2986 	 * huge quite quickly.  The threshold for 64-bit architectures is 128
2987 	 * CPUs.
2988 	 */
2989 	if (mp_ncpus < (2 * _BITSET_BITS)) {
2990 		cpumask_t *sscm_ptr;
2991 
2992 		/*
2993 		 * This represents 'mp_ncpus * __bitset_words(CPU_SETSIZE) *
2994 		 * (_BITSET_BITS / 8)' bytes (for comparison with the
2995 		 * overlapping scheme).
2996 		 */
2997 		static_single_cpu_mask_lcs = kmalloc_array(mp_ncpus,
2998 		    sizeof(*static_single_cpu_mask_lcs),
2999 		    M_WAITOK | M_ZERO);
3000 
3001 		sscm_ptr = static_single_cpu_mask_lcs;
3002 		CPU_FOREACH(i) {
3003 			static_single_cpu_mask[i] = sscm_ptr++;
3004 			CPU_SET(i, static_single_cpu_mask[i]);
3005 		}
3006 	} else {
3007 		/* Pointer to a bitset word. */
3008 		__typeof(((cpuset_t *)NULL)->__bits[0]) *bwp;
3009 
3010 		/*
3011 		 * Allocate memory for (static) spans of 'cpumask_t' ('cpuset_t'
3012 		 * really) with a single bit set that can be reused for all
3013 		 * single CPU masks by making them start at different offsets.
3014 		 * We need '__bitset_words(CPU_SETSIZE) - 1' bitset words before
3015 		 * the word having its single bit set, and the same amount
3016 		 * after.
3017 		 */
3018 		static_single_cpu_mask_lcs = mallocarray(_BITSET_BITS,
3019 		    (2 * __bitset_words(CPU_SETSIZE) - 1) * (_BITSET_BITS / 8),
3020 		    M_KMALLOC, M_WAITOK | M_ZERO);
3021 
3022 		/*
3023 		 * We rely below on cpuset_t and the bitset generic
3024 		 * implementation assigning words in the '__bits' array in the
3025 		 * same order of bits (i.e., little-endian ordering, not to be
3026 		 * confused with machine endianness, which concerns bits in
3027 		 * words and other integers).  This is an imperfect test, but it
3028 		 * will detect a change to big-endian ordering.
3029 		 */
3030 		_Static_assert(
3031 		    __bitset_word(_BITSET_BITS + 1, _BITSET_BITS) == 1,
3032 		    "Assumes a bitset implementation that is little-endian "
3033 		    "on its words");
3034 
3035 		/* Initialize the single bit of each static span. */
3036 		bwp = (__typeof(bwp))static_single_cpu_mask_lcs +
3037 		    (__bitset_words(CPU_SETSIZE) - 1);
3038 		for (i = 0; i < _BITSET_BITS; i++) {
3039 			CPU_SET(i, (cpuset_t *)bwp);
3040 			bwp += (2 * __bitset_words(CPU_SETSIZE) - 1);
3041 		}
3042 
3043 		/*
3044 		 * Finally set all CPU masks to the proper word in their
3045 		 * relevant span.
3046 		 */
3047 		CPU_FOREACH(i) {
3048 			bwp = (__typeof(bwp))static_single_cpu_mask_lcs;
3049 			/* Find the non-zero word of the relevant span. */
3050 			bwp += (2 * __bitset_words(CPU_SETSIZE) - 1) *
3051 			    (i % _BITSET_BITS) +
3052 			    __bitset_words(CPU_SETSIZE) - 1;
3053 			/* Shift to find the CPU mask start. */
3054 			bwp -= (i / _BITSET_BITS);
3055 			static_single_cpu_mask[i] = (cpuset_t *)bwp;
3056 		}
3057 	}
3058 
3059 	strlcpy(init_uts_ns.name.release, osrelease, sizeof(init_uts_ns.name.release));
3060 }
3061 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL);
3062 
3063 static void
3064 linux_compat_uninit(void *arg)
3065 {
3066 	linux_kobject_kfree_name(&linux_class_root);
3067 	linux_kobject_kfree_name(&linux_root_device.kobj);
3068 	linux_kobject_kfree_name(&linux_class_misc.kobj);
3069 
3070 	free(static_single_cpu_mask_lcs, M_KMALLOC);
3071 	free(static_single_cpu_mask, M_KMALLOC);
3072 #if defined(__i386__) || defined(__amd64__)
3073 	free(__cpu_data, M_KMALLOC);
3074 #endif
3075 
3076 	spin_lock_destroy(&pci_lock);
3077 	rw_destroy(&linux_vma_lock);
3078 }
3079 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL);
3080 
3081 /*
3082  * NOTE: Linux frequently uses "unsigned long" for pointer to integer
3083  * conversion and vice versa, where in FreeBSD "uintptr_t" would be
3084  * used. Assert these types have the same size, else some parts of the
3085  * LinuxKPI may not work like expected:
3086  */
3087 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t));
3088