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 __FBSDID("$FreeBSD$");
32
33 #include "opt_stack.h"
34
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/malloc.h>
38 #include <sys/kernel.h>
39 #include <sys/sysctl.h>
40 #include <sys/proc.h>
41 #include <sys/sglist.h>
42 #include <sys/sleepqueue.h>
43 #include <sys/refcount.h>
44 #include <sys/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/bus.h>
47 #include <sys/eventhandler.h>
48 #include <sys/fcntl.h>
49 #include <sys/file.h>
50 #include <sys/filio.h>
51 #include <sys/rwlock.h>
52 #include <sys/mman.h>
53 #include <sys/stack.h>
54 #include <sys/sysent.h>
55 #include <sys/time.h>
56 #include <sys/user.h>
57
58 #include <vm/vm.h>
59 #include <vm/pmap.h>
60 #include <vm/vm_object.h>
61 #include <vm/vm_page.h>
62 #include <vm/vm_pager.h>
63
64 #include <machine/stdarg.h>
65
66 #if defined(__i386__) || defined(__amd64__)
67 #include <machine/md_var.h>
68 #endif
69
70 #include <linux/kobject.h>
71 #include <linux/cpu.h>
72 #include <linux/device.h>
73 #include <linux/slab.h>
74 #include <linux/module.h>
75 #include <linux/moduleparam.h>
76 #include <linux/cdev.h>
77 #include <linux/file.h>
78 #include <linux/sysfs.h>
79 #include <linux/mm.h>
80 #include <linux/io.h>
81 #include <linux/vmalloc.h>
82 #include <linux/netdevice.h>
83 #include <linux/timer.h>
84 #include <linux/interrupt.h>
85 #include <linux/uaccess.h>
86 #include <linux/list.h>
87 #include <linux/kthread.h>
88 #include <linux/kernel.h>
89 #include <linux/compat.h>
90 #include <linux/io-mapping.h>
91 #include <linux/poll.h>
92 #include <linux/smp.h>
93 #include <linux/wait_bit.h>
94 #include <linux/rcupdate.h>
95 #include <linux/interval_tree.h>
96 #include <linux/interval_tree_generic.h>
97
98 #if defined(__i386__) || defined(__amd64__)
99 #include <asm/smp.h>
100 #include <asm/processor.h>
101 #endif
102
103 SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
104 "LinuxKPI parameters");
105
106 int linuxkpi_debug;
107 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN,
108 &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable.");
109
110 int linuxkpi_warn_dump_stack = 0;
111 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, warn_dump_stack, CTLFLAG_RWTUN,
112 &linuxkpi_warn_dump_stack, 0,
113 "Set to enable stack traces from WARN_ON(). Clear to disable.");
114
115 static struct timeval lkpi_net_lastlog;
116 static int lkpi_net_curpps;
117 static int lkpi_net_maxpps = 99;
118 SYSCTL_INT(_compat_linuxkpi, OID_AUTO, net_ratelimit, CTLFLAG_RWTUN,
119 &lkpi_net_maxpps, 0, "Limit number of LinuxKPI net messages per second.");
120
121 MALLOC_DEFINE(M_KMALLOC, "lkpikmalloc", "Linux kmalloc compat");
122
123 #include <linux/rbtree.h>
124 /* Undo Linux compat changes. */
125 #undef RB_ROOT
126 #undef file
127 #undef cdev
128 #define RB_ROOT(head) (head)->rbh_root
129
130 static void linux_destroy_dev(struct linux_cdev *);
131 static void linux_cdev_deref(struct linux_cdev *ldev);
132 static struct vm_area_struct *linux_cdev_handle_find(void *handle);
133
134 cpumask_t cpu_online_mask;
135 static cpumask_t static_single_cpu_mask[MAXCPU];
136 struct kobject linux_class_root;
137 struct device linux_root_device;
138 struct class linux_class_misc;
139 struct list_head pci_drivers;
140 struct list_head pci_devices;
141 spinlock_t pci_lock;
142
143 unsigned long linux_timer_hz_mask;
144
145 wait_queue_head_t linux_bit_waitq;
146 wait_queue_head_t linux_var_waitq;
147
148 int
149 panic_cmp(struct rb_node *one, struct rb_node *two)
150 {
151 panic("no cmp");
152 }
153
154 RB_GENERATE(linux_root, rb_node, __entry, panic_cmp);
155
156 #define START(node) ((node)->start)
157 #define LAST(node) ((node)->last)
158
159 INTERVAL_TREE_DEFINE(struct interval_tree_node, rb, unsigned long,, START,
160 LAST,, lkpi_interval_tree)
161
162 struct kobject *
163 kobject_create(void)
164 {
165 struct kobject *kobj;
166
167 kobj = kzalloc(sizeof(*kobj), GFP_KERNEL);
168 if (kobj == NULL)
169 return (NULL);
170 kobject_init(kobj, &linux_kfree_type);
171
172 return (kobj);
173 }
174
175
176 int
177 kobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list args)
178 {
179 va_list tmp_va;
180 int len;
181 char *old;
182 char *name;
183 char dummy;
184
185 old = kobj->name;
186
187 if (old && fmt == NULL)
188 return (0);
189
190 /* compute length of string */
191 va_copy(tmp_va, args);
192 len = vsnprintf(&dummy, 0, fmt, tmp_va);
193 va_end(tmp_va);
194
195 /* account for zero termination */
196 len++;
197
198 /* check for error */
199 if (len < 1)
200 return (-EINVAL);
201
202 /* allocate memory for string */
203 name = kzalloc(len, GFP_KERNEL);
204 if (name == NULL)
205 return (-ENOMEM);
206 vsnprintf(name, len, fmt, args);
207 kobj->name = name;
208
209 /* free old string */
210 kfree(old);
211
212 /* filter new string */
213 for (; *name != '\0'; name++)
214 if (*name == '/')
215 *name = '!';
216 return (0);
217 }
218
219 int
220 kobject_set_name(struct kobject *kobj, const char *fmt, ...)
221 {
222 va_list args;
223 int error;
224
225 va_start(args, fmt);
226 error = kobject_set_name_vargs(kobj, fmt, args);
227 va_end(args);
228
229 return (error);
230 }
231
232 static int
233 kobject_add_complete(struct kobject *kobj, struct kobject *parent)
234 {
235 const struct kobj_type *t;
236 int error;
237
238 kobj->parent = parent;
239 error = sysfs_create_dir(kobj);
240 if (error == 0 && kobj->ktype && kobj->ktype->default_attrs) {
241 struct attribute **attr;
242 t = kobj->ktype;
243
244 for (attr = t->default_attrs; *attr != NULL; attr++) {
245 error = sysfs_create_file(kobj, *attr);
246 if (error)
247 break;
248 }
249 if (error)
250 sysfs_remove_dir(kobj);
251 }
252 return (error);
253 }
254
255 int
256 kobject_add(struct kobject *kobj, struct kobject *parent, const char *fmt, ...)
257 {
258 va_list args;
259 int error;
260
261 va_start(args, fmt);
262 error = kobject_set_name_vargs(kobj, fmt, args);
263 va_end(args);
264 if (error)
265 return (error);
266
267 return kobject_add_complete(kobj, parent);
268 }
269
270 void
271 linux_kobject_release(struct kref *kref)
272 {
273 struct kobject *kobj;
274 char *name;
275
276 kobj = container_of(kref, struct kobject, kref);
277 sysfs_remove_dir(kobj);
278 name = kobj->name;
279 if (kobj->ktype && kobj->ktype->release)
280 kobj->ktype->release(kobj);
281 kfree(name);
282 }
283
284 static void
285 linux_kobject_kfree(struct kobject *kobj)
286 {
287 kfree(kobj);
288 }
289
290 static void
291 linux_kobject_kfree_name(struct kobject *kobj)
292 {
293 if (kobj) {
294 kfree(kobj->name);
295 }
296 }
297
298 const struct kobj_type linux_kfree_type = {
299 .release = linux_kobject_kfree
300 };
301
302 static ssize_t
303 lkpi_kobj_attr_show(struct kobject *kobj, struct attribute *attr, char *buf)
304 {
305 struct kobj_attribute *ka =
306 container_of(attr, struct kobj_attribute, attr);
307
308 if (ka->show == NULL)
309 return (-EIO);
310
311 return (ka->show(kobj, ka, buf));
312 }
313
314 static ssize_t
315 lkpi_kobj_attr_store(struct kobject *kobj, struct attribute *attr,
316 const char *buf, size_t count)
317 {
318 struct kobj_attribute *ka =
319 container_of(attr, struct kobj_attribute, attr);
320
321 if (ka->store == NULL)
322 return (-EIO);
323
324 return (ka->store(kobj, ka, buf, count));
325 }
326
327 const struct sysfs_ops kobj_sysfs_ops = {
328 .show = lkpi_kobj_attr_show,
329 .store = lkpi_kobj_attr_store,
330 };
331
332 static void
333 linux_device_release(struct device *dev)
334 {
335 pr_debug("linux_device_release: %s\n", dev_name(dev));
336 kfree(dev);
337 }
338
339 static ssize_t
340 linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf)
341 {
342 struct class_attribute *dattr;
343 ssize_t error;
344
345 dattr = container_of(attr, struct class_attribute, attr);
346 error = -EIO;
347 if (dattr->show)
348 error = dattr->show(container_of(kobj, struct class, kobj),
349 dattr, buf);
350 return (error);
351 }
352
353 static ssize_t
354 linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf,
355 size_t count)
356 {
357 struct class_attribute *dattr;
358 ssize_t error;
359
360 dattr = container_of(attr, struct class_attribute, attr);
361 error = -EIO;
362 if (dattr->store)
363 error = dattr->store(container_of(kobj, struct class, kobj),
364 dattr, buf, count);
365 return (error);
366 }
367
368 static void
369 linux_class_release(struct kobject *kobj)
370 {
371 struct class *class;
372
373 class = container_of(kobj, struct class, kobj);
374 if (class->class_release)
375 class->class_release(class);
376 }
377
378 static const struct sysfs_ops linux_class_sysfs = {
379 .show = linux_class_show,
380 .store = linux_class_store,
381 };
382
383 const struct kobj_type linux_class_ktype = {
384 .release = linux_class_release,
385 .sysfs_ops = &linux_class_sysfs
386 };
387
388 static void
389 linux_dev_release(struct kobject *kobj)
390 {
391 struct device *dev;
392
393 dev = container_of(kobj, struct device, kobj);
394 /* This is the precedence defined by linux. */
395 if (dev->release)
396 dev->release(dev);
397 else if (dev->class && dev->class->dev_release)
398 dev->class->dev_release(dev);
399 }
400
401 static ssize_t
402 linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf)
403 {
404 struct device_attribute *dattr;
405 ssize_t error;
406
407 dattr = container_of(attr, struct device_attribute, attr);
408 error = -EIO;
409 if (dattr->show)
410 error = dattr->show(container_of(kobj, struct device, kobj),
411 dattr, buf);
412 return (error);
413 }
414
415 static ssize_t
416 linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf,
417 size_t count)
418 {
419 struct device_attribute *dattr;
420 ssize_t error;
421
422 dattr = container_of(attr, struct device_attribute, attr);
423 error = -EIO;
424 if (dattr->store)
425 error = dattr->store(container_of(kobj, struct device, kobj),
426 dattr, buf, count);
427 return (error);
428 }
429
430 static const struct sysfs_ops linux_dev_sysfs = {
431 .show = linux_dev_show,
432 .store = linux_dev_store,
433 };
434
435 const struct kobj_type linux_dev_ktype = {
436 .release = linux_dev_release,
437 .sysfs_ops = &linux_dev_sysfs
438 };
439
440 struct device *
441 device_create(struct class *class, struct device *parent, dev_t devt,
442 void *drvdata, const char *fmt, ...)
443 {
444 struct device *dev;
445 va_list args;
446
447 dev = kzalloc(sizeof(*dev), M_WAITOK);
448 dev->parent = parent;
449 dev->class = class;
450 dev->devt = devt;
451 dev->driver_data = drvdata;
452 dev->release = linux_device_release;
453 va_start(args, fmt);
454 kobject_set_name_vargs(&dev->kobj, fmt, args);
455 va_end(args);
456 device_register(dev);
457
458 return (dev);
459 }
460
461 struct device *
462 device_create_groups_vargs(struct class *class, struct device *parent,
463 dev_t devt, void *drvdata, const struct attribute_group **groups,
464 const char *fmt, va_list args)
465 {
466 struct device *dev = NULL;
467 int retval = -ENODEV;
468
469 if (class == NULL || IS_ERR(class))
470 goto error;
471
472 dev = kzalloc(sizeof(*dev), GFP_KERNEL);
473 if (!dev) {
474 retval = -ENOMEM;
475 goto error;
476 }
477
478 dev->devt = devt;
479 dev->class = class;
480 dev->parent = parent;
481 dev->groups = groups;
482 dev->release = device_create_release;
483 /* device_initialize() needs the class and parent to be set */
484 device_initialize(dev);
485 dev_set_drvdata(dev, drvdata);
486
487 retval = kobject_set_name_vargs(&dev->kobj, fmt, args);
488 if (retval)
489 goto error;
490
491 retval = device_add(dev);
492 if (retval)
493 goto error;
494
495 return dev;
496
497 error:
498 put_device(dev);
499 return ERR_PTR(retval);
500 }
501
502 struct class *
503 class_create(struct module *owner, const char *name)
504 {
505 struct class *class;
506 int error;
507
508 class = kzalloc(sizeof(*class), M_WAITOK);
509 class->owner = owner;
510 class->name = name;
511 class->class_release = linux_class_kfree;
512 error = class_register(class);
513 if (error) {
514 kfree(class);
515 return (NULL);
516 }
517
518 return (class);
519 }
520
521 int
522 kobject_init_and_add(struct kobject *kobj, const struct kobj_type *ktype,
523 struct kobject *parent, const char *fmt, ...)
524 {
525 va_list args;
526 int error;
527
528 kobject_init(kobj, ktype);
529 kobj->ktype = ktype;
530 kobj->parent = parent;
531 kobj->name = NULL;
532
533 va_start(args, fmt);
534 error = kobject_set_name_vargs(kobj, fmt, args);
535 va_end(args);
536 if (error)
537 return (error);
538 return kobject_add_complete(kobj, parent);
539 }
540
541 static void
542 linux_kq_lock(void *arg)
543 {
544 spinlock_t *s = arg;
545
546 spin_lock(s);
547 }
548 static void
549 linux_kq_unlock(void *arg)
550 {
551 spinlock_t *s = arg;
552
553 spin_unlock(s);
554 }
555
556 static void
557 linux_kq_assert_lock(void *arg, int what)
558 {
559 #ifdef INVARIANTS
560 spinlock_t *s = arg;
561
562 if (what == LA_LOCKED)
563 mtx_assert(&s->m, MA_OWNED);
564 else
565 mtx_assert(&s->m, MA_NOTOWNED);
566 #endif
567 }
568
569 static void
570 linux_file_kqfilter_poll(struct linux_file *, int);
571
572 struct linux_file *
573 linux_file_alloc(void)
574 {
575 struct linux_file *filp;
576
577 filp = kzalloc(sizeof(*filp), GFP_KERNEL);
578
579 /* set initial refcount */
580 filp->f_count = 1;
581
582 /* setup fields needed by kqueue support */
583 spin_lock_init(&filp->f_kqlock);
584 knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock,
585 linux_kq_lock, linux_kq_unlock, linux_kq_assert_lock);
586
587 return (filp);
588 }
589
590 void
591 linux_file_free(struct linux_file *filp)
592 {
593 if (filp->_file == NULL) {
594 if (filp->f_op != NULL && filp->f_op->release != NULL)
595 filp->f_op->release(filp->f_vnode, filp);
596 if (filp->f_shmem != NULL)
597 vm_object_deallocate(filp->f_shmem);
598 kfree_rcu(filp, rcu);
599 } else {
600 /*
601 * The close method of the character device or file
602 * will free the linux_file structure:
603 */
604 _fdrop(filp->_file, curthread);
605 }
606 }
607
608 struct linux_cdev *
609 cdev_alloc(void)
610 {
611 struct linux_cdev *cdev;
612
613 cdev = kzalloc(sizeof(struct linux_cdev), M_WAITOK);
614 kobject_init(&cdev->kobj, &linux_cdev_ktype);
615 cdev->refs = 1;
616 return (cdev);
617 }
618
619 static int
620 linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot,
621 vm_page_t *mres)
622 {
623 struct vm_area_struct *vmap;
624
625 vmap = linux_cdev_handle_find(vm_obj->handle);
626
627 MPASS(vmap != NULL);
628 MPASS(vmap->vm_private_data == vm_obj->handle);
629
630 if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) {
631 vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset;
632 vm_page_t page;
633
634 if (((*mres)->flags & PG_FICTITIOUS) != 0) {
635 /*
636 * If the passed in result page is a fake
637 * page, update it with the new physical
638 * address.
639 */
640 page = *mres;
641 vm_page_updatefake(page, paddr, vm_obj->memattr);
642 } else {
643 /*
644 * Replace the passed in "mres" page with our
645 * own fake page and free up the all of the
646 * original pages.
647 */
648 VM_OBJECT_WUNLOCK(vm_obj);
649 page = vm_page_getfake(paddr, vm_obj->memattr);
650 VM_OBJECT_WLOCK(vm_obj);
651
652 vm_page_replace(page, vm_obj, (*mres)->pindex, *mres);
653 *mres = page;
654 }
655 vm_page_valid(page);
656 return (VM_PAGER_OK);
657 }
658 return (VM_PAGER_FAIL);
659 }
660
661 static int
662 linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type,
663 vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last)
664 {
665 struct vm_area_struct *vmap;
666 int err;
667
668 /* get VM area structure */
669 vmap = linux_cdev_handle_find(vm_obj->handle);
670 MPASS(vmap != NULL);
671 MPASS(vmap->vm_private_data == vm_obj->handle);
672
673 VM_OBJECT_WUNLOCK(vm_obj);
674
675 linux_set_current(curthread);
676
677 down_write(&vmap->vm_mm->mmap_sem);
678 if (unlikely(vmap->vm_ops == NULL)) {
679 err = VM_FAULT_SIGBUS;
680 } else {
681 struct vm_fault vmf;
682
683 /* fill out VM fault structure */
684 vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx);
685 vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0;
686 vmf.pgoff = 0;
687 vmf.page = NULL;
688 vmf.vma = vmap;
689
690 vmap->vm_pfn_count = 0;
691 vmap->vm_pfn_pcount = &vmap->vm_pfn_count;
692 vmap->vm_obj = vm_obj;
693
694 err = vmap->vm_ops->fault(&vmf);
695
696 while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) {
697 kern_yield(PRI_USER);
698 err = vmap->vm_ops->fault(&vmf);
699 }
700 }
701
702 /* translate return code */
703 switch (err) {
704 case VM_FAULT_OOM:
705 err = VM_PAGER_AGAIN;
706 break;
707 case VM_FAULT_SIGBUS:
708 err = VM_PAGER_BAD;
709 break;
710 case VM_FAULT_NOPAGE:
711 /*
712 * By contract the fault handler will return having
713 * busied all the pages itself. If pidx is already
714 * found in the object, it will simply xbusy the first
715 * page and return with vm_pfn_count set to 1.
716 */
717 *first = vmap->vm_pfn_first;
718 *last = *first + vmap->vm_pfn_count - 1;
719 err = VM_PAGER_OK;
720 break;
721 default:
722 err = VM_PAGER_ERROR;
723 break;
724 }
725 up_write(&vmap->vm_mm->mmap_sem);
726 VM_OBJECT_WLOCK(vm_obj);
727 return (err);
728 }
729
730 static struct rwlock linux_vma_lock;
731 static TAILQ_HEAD(, vm_area_struct) linux_vma_head =
732 TAILQ_HEAD_INITIALIZER(linux_vma_head);
733
734 static void
735 linux_cdev_handle_free(struct vm_area_struct *vmap)
736 {
737 /* Drop reference on vm_file */
738 if (vmap->vm_file != NULL)
739 fput(vmap->vm_file);
740
741 /* Drop reference on mm_struct */
742 mmput(vmap->vm_mm);
743
744 kfree(vmap);
745 }
746
747 static void
748 linux_cdev_handle_remove(struct vm_area_struct *vmap)
749 {
750 rw_wlock(&linux_vma_lock);
751 TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry);
752 rw_wunlock(&linux_vma_lock);
753 }
754
755 static struct vm_area_struct *
756 linux_cdev_handle_find(void *handle)
757 {
758 struct vm_area_struct *vmap;
759
760 rw_rlock(&linux_vma_lock);
761 TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) {
762 if (vmap->vm_private_data == handle)
763 break;
764 }
765 rw_runlock(&linux_vma_lock);
766 return (vmap);
767 }
768
769 static int
770 linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot,
771 vm_ooffset_t foff, struct ucred *cred, u_short *color)
772 {
773
774 MPASS(linux_cdev_handle_find(handle) != NULL);
775 *color = 0;
776 return (0);
777 }
778
779 static void
780 linux_cdev_pager_dtor(void *handle)
781 {
782 const struct vm_operations_struct *vm_ops;
783 struct vm_area_struct *vmap;
784
785 vmap = linux_cdev_handle_find(handle);
786 MPASS(vmap != NULL);
787
788 /*
789 * Remove handle before calling close operation to prevent
790 * other threads from reusing the handle pointer.
791 */
792 linux_cdev_handle_remove(vmap);
793
794 down_write(&vmap->vm_mm->mmap_sem);
795 vm_ops = vmap->vm_ops;
796 if (likely(vm_ops != NULL))
797 vm_ops->close(vmap);
798 up_write(&vmap->vm_mm->mmap_sem);
799
800 linux_cdev_handle_free(vmap);
801 }
802
803 static struct cdev_pager_ops linux_cdev_pager_ops[2] = {
804 {
805 /* OBJT_MGTDEVICE */
806 .cdev_pg_populate = linux_cdev_pager_populate,
807 .cdev_pg_ctor = linux_cdev_pager_ctor,
808 .cdev_pg_dtor = linux_cdev_pager_dtor
809 },
810 {
811 /* OBJT_DEVICE */
812 .cdev_pg_fault = linux_cdev_pager_fault,
813 .cdev_pg_ctor = linux_cdev_pager_ctor,
814 .cdev_pg_dtor = linux_cdev_pager_dtor
815 },
816 };
817
818 int
819 zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
820 unsigned long size)
821 {
822 vm_object_t obj;
823 vm_page_t m;
824
825 obj = vma->vm_obj;
826 if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0)
827 return (-ENOTSUP);
828 VM_OBJECT_RLOCK(obj);
829 for (m = vm_page_find_least(obj, OFF_TO_IDX(address));
830 m != NULL && m->pindex < OFF_TO_IDX(address + size);
831 m = TAILQ_NEXT(m, listq))
832 pmap_remove_all(m);
833 VM_OBJECT_RUNLOCK(obj);
834 return (0);
835 }
836
837 void
838 vma_set_file(struct vm_area_struct *vma, struct linux_file *file)
839 {
840 struct linux_file *tmp;
841
842 /* Changing an anonymous vma with this is illegal */
843 get_file(file);
844 tmp = vma->vm_file;
845 vma->vm_file = file;
846 fput(tmp);
847 }
848
849 static struct file_operations dummy_ldev_ops = {
850 /* XXXKIB */
851 };
852
853 static struct linux_cdev dummy_ldev = {
854 .ops = &dummy_ldev_ops,
855 };
856
857 #define LDEV_SI_DTR 0x0001
858 #define LDEV_SI_REF 0x0002
859
860 static void
861 linux_get_fop(struct linux_file *filp, const struct file_operations **fop,
862 struct linux_cdev **dev)
863 {
864 struct linux_cdev *ldev;
865 u_int siref;
866
867 ldev = filp->f_cdev;
868 *fop = filp->f_op;
869 if (ldev != NULL) {
870 if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
871 refcount_acquire(&ldev->refs);
872 } else {
873 for (siref = ldev->siref;;) {
874 if ((siref & LDEV_SI_DTR) != 0) {
875 ldev = &dummy_ldev;
876 *fop = ldev->ops;
877 siref = ldev->siref;
878 MPASS((ldev->siref & LDEV_SI_DTR) == 0);
879 } else if (atomic_fcmpset_int(&ldev->siref,
880 &siref, siref + LDEV_SI_REF)) {
881 break;
882 }
883 }
884 }
885 }
886 *dev = ldev;
887 }
888
889 static void
890 linux_drop_fop(struct linux_cdev *ldev)
891 {
892
893 if (ldev == NULL)
894 return;
895 if (ldev->kobj.ktype == &linux_cdev_static_ktype) {
896 linux_cdev_deref(ldev);
897 } else {
898 MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
899 MPASS((ldev->siref & ~LDEV_SI_DTR) != 0);
900 atomic_subtract_int(&ldev->siref, LDEV_SI_REF);
901 }
902 }
903
904 #define OPW(fp,td,code) ({ \
905 struct file *__fpop; \
906 __typeof(code) __retval; \
907 \
908 __fpop = (td)->td_fpop; \
909 (td)->td_fpop = (fp); \
910 __retval = (code); \
911 (td)->td_fpop = __fpop; \
912 __retval; \
913 })
914
915 static int
916 linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td,
917 struct file *file)
918 {
919 struct linux_cdev *ldev;
920 struct linux_file *filp;
921 const struct file_operations *fop;
922 int error;
923
924 ldev = dev->si_drv1;
925
926 filp = linux_file_alloc();
927 filp->f_dentry = &filp->f_dentry_store;
928 filp->f_op = ldev->ops;
929 filp->f_mode = file->f_flag;
930 filp->f_flags = file->f_flag;
931 filp->f_vnode = file->f_vnode;
932 filp->_file = file;
933 refcount_acquire(&ldev->refs);
934 filp->f_cdev = ldev;
935
936 linux_set_current(td);
937 linux_get_fop(filp, &fop, &ldev);
938
939 if (fop->open != NULL) {
940 error = -fop->open(file->f_vnode, filp);
941 if (error != 0) {
942 linux_drop_fop(ldev);
943 linux_cdev_deref(filp->f_cdev);
944 kfree(filp);
945 return (error);
946 }
947 }
948
949 /* hold on to the vnode - used for fstat() */
950 vhold(filp->f_vnode);
951
952 /* release the file from devfs */
953 finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops);
954 linux_drop_fop(ldev);
955 return (ENXIO);
956 }
957
958 #define LINUX_IOCTL_MIN_PTR 0x10000UL
959 #define LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX)
960
961 static inline int
962 linux_remap_address(void **uaddr, size_t len)
963 {
964 uintptr_t uaddr_val = (uintptr_t)(*uaddr);
965
966 if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR &&
967 uaddr_val < LINUX_IOCTL_MAX_PTR)) {
968 struct task_struct *pts = current;
969 if (pts == NULL) {
970 *uaddr = NULL;
971 return (1);
972 }
973
974 /* compute data offset */
975 uaddr_val -= LINUX_IOCTL_MIN_PTR;
976
977 /* check that length is within bounds */
978 if ((len > IOCPARM_MAX) ||
979 (uaddr_val + len) > pts->bsd_ioctl_len) {
980 *uaddr = NULL;
981 return (1);
982 }
983
984 /* re-add kernel buffer address */
985 uaddr_val += (uintptr_t)pts->bsd_ioctl_data;
986
987 /* update address location */
988 *uaddr = (void *)uaddr_val;
989 return (1);
990 }
991 return (0);
992 }
993
994 int
995 linux_copyin(const void *uaddr, void *kaddr, size_t len)
996 {
997 if (linux_remap_address(__DECONST(void **, &uaddr), len)) {
998 if (uaddr == NULL)
999 return (-EFAULT);
1000 memcpy(kaddr, uaddr, len);
1001 return (0);
1002 }
1003 return (-copyin(uaddr, kaddr, len));
1004 }
1005
1006 int
1007 linux_copyout(const void *kaddr, void *uaddr, size_t len)
1008 {
1009 if (linux_remap_address(&uaddr, len)) {
1010 if (uaddr == NULL)
1011 return (-EFAULT);
1012 memcpy(uaddr, kaddr, len);
1013 return (0);
1014 }
1015 return (-copyout(kaddr, uaddr, len));
1016 }
1017
1018 size_t
1019 linux_clear_user(void *_uaddr, size_t _len)
1020 {
1021 uint8_t *uaddr = _uaddr;
1022 size_t len = _len;
1023
1024 /* make sure uaddr is aligned before going into the fast loop */
1025 while (((uintptr_t)uaddr & 7) != 0 && len > 7) {
1026 if (subyte(uaddr, 0))
1027 return (_len);
1028 uaddr++;
1029 len--;
1030 }
1031
1032 /* zero 8 bytes at a time */
1033 while (len > 7) {
1034 #ifdef __LP64__
1035 if (suword64(uaddr, 0))
1036 return (_len);
1037 #else
1038 if (suword32(uaddr, 0))
1039 return (_len);
1040 if (suword32(uaddr + 4, 0))
1041 return (_len);
1042 #endif
1043 uaddr += 8;
1044 len -= 8;
1045 }
1046
1047 /* zero fill end, if any */
1048 while (len > 0) {
1049 if (subyte(uaddr, 0))
1050 return (_len);
1051 uaddr++;
1052 len--;
1053 }
1054 return (0);
1055 }
1056
1057 int
1058 linux_access_ok(const void *uaddr, size_t len)
1059 {
1060 uintptr_t saddr;
1061 uintptr_t eaddr;
1062
1063 /* get start and end address */
1064 saddr = (uintptr_t)uaddr;
1065 eaddr = (uintptr_t)uaddr + len;
1066
1067 /* verify addresses are valid for userspace */
1068 return ((saddr == eaddr) ||
1069 (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS));
1070 }
1071
1072 /*
1073 * This function should return either EINTR or ERESTART depending on
1074 * the signal type sent to this thread:
1075 */
1076 static int
1077 linux_get_error(struct task_struct *task, int error)
1078 {
1079 /* check for signal type interrupt code */
1080 if (error == EINTR || error == ERESTARTSYS || error == ERESTART) {
1081 error = -linux_schedule_get_interrupt_value(task);
1082 if (error == 0)
1083 error = EINTR;
1084 }
1085 return (error);
1086 }
1087
1088 static int
1089 linux_file_ioctl_sub(struct file *fp, struct linux_file *filp,
1090 const struct file_operations *fop, u_long cmd, caddr_t data,
1091 struct thread *td)
1092 {
1093 struct task_struct *task = current;
1094 unsigned size;
1095 int error;
1096
1097 size = IOCPARM_LEN(cmd);
1098 /* refer to logic in sys_ioctl() */
1099 if (size > 0) {
1100 /*
1101 * Setup hint for linux_copyin() and linux_copyout().
1102 *
1103 * Background: Linux code expects a user-space address
1104 * while FreeBSD supplies a kernel-space address.
1105 */
1106 task->bsd_ioctl_data = data;
1107 task->bsd_ioctl_len = size;
1108 data = (void *)LINUX_IOCTL_MIN_PTR;
1109 } else {
1110 /* fetch user-space pointer */
1111 data = *(void **)data;
1112 }
1113 #ifdef COMPAT_FREEBSD32
1114 if (SV_PROC_FLAG(td->td_proc, SV_ILP32)) {
1115 /* try the compat IOCTL handler first */
1116 if (fop->compat_ioctl != NULL) {
1117 error = -OPW(fp, td, fop->compat_ioctl(filp,
1118 cmd, (u_long)data));
1119 } else {
1120 error = ENOTTY;
1121 }
1122
1123 /* fallback to the regular IOCTL handler, if any */
1124 if (error == ENOTTY && fop->unlocked_ioctl != NULL) {
1125 error = -OPW(fp, td, fop->unlocked_ioctl(filp,
1126 cmd, (u_long)data));
1127 }
1128 } else
1129 #endif
1130 {
1131 if (fop->unlocked_ioctl != NULL) {
1132 error = -OPW(fp, td, fop->unlocked_ioctl(filp,
1133 cmd, (u_long)data));
1134 } else {
1135 error = ENOTTY;
1136 }
1137 }
1138 if (size > 0) {
1139 task->bsd_ioctl_data = NULL;
1140 task->bsd_ioctl_len = 0;
1141 }
1142
1143 if (error == EWOULDBLOCK) {
1144 /* update kqfilter status, if any */
1145 linux_file_kqfilter_poll(filp,
1146 LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1147 } else {
1148 error = linux_get_error(task, error);
1149 }
1150 return (error);
1151 }
1152
1153 #define LINUX_POLL_TABLE_NORMAL ((poll_table *)1)
1154
1155 /*
1156 * This function atomically updates the poll wakeup state and returns
1157 * the previous state at the time of update.
1158 */
1159 static uint8_t
1160 linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate)
1161 {
1162 int c, old;
1163
1164 c = v->counter;
1165
1166 while ((old = atomic_cmpxchg(v, c, pstate[c])) != c)
1167 c = old;
1168
1169 return (c);
1170 }
1171
1172 static int
1173 linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key)
1174 {
1175 static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1176 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */
1177 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1178 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY,
1179 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */
1180 };
1181 struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq);
1182
1183 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1184 case LINUX_FWQ_STATE_QUEUED:
1185 linux_poll_wakeup(filp);
1186 return (1);
1187 default:
1188 return (0);
1189 }
1190 }
1191
1192 void
1193 linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p)
1194 {
1195 static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1196 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY,
1197 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */
1198 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */
1199 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED,
1200 };
1201
1202 /* check if we are called inside the select system call */
1203 if (p == LINUX_POLL_TABLE_NORMAL)
1204 selrecord(curthread, &filp->f_selinfo);
1205
1206 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1207 case LINUX_FWQ_STATE_INIT:
1208 /* NOTE: file handles can only belong to one wait-queue */
1209 filp->f_wait_queue.wqh = wqh;
1210 filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback;
1211 add_wait_queue(wqh, &filp->f_wait_queue.wq);
1212 atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED);
1213 break;
1214 default:
1215 break;
1216 }
1217 }
1218
1219 static void
1220 linux_poll_wait_dequeue(struct linux_file *filp)
1221 {
1222 static const uint8_t state[LINUX_FWQ_STATE_MAX] = {
1223 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */
1224 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT,
1225 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT,
1226 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT,
1227 };
1228
1229 seldrain(&filp->f_selinfo);
1230
1231 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) {
1232 case LINUX_FWQ_STATE_NOT_READY:
1233 case LINUX_FWQ_STATE_QUEUED:
1234 case LINUX_FWQ_STATE_READY:
1235 remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq);
1236 break;
1237 default:
1238 break;
1239 }
1240 }
1241
1242 void
1243 linux_poll_wakeup(struct linux_file *filp)
1244 {
1245 /* this function should be NULL-safe */
1246 if (filp == NULL)
1247 return;
1248
1249 selwakeup(&filp->f_selinfo);
1250
1251 spin_lock(&filp->f_kqlock);
1252 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ |
1253 LINUX_KQ_FLAG_NEED_WRITE;
1254
1255 /* make sure the "knote" gets woken up */
1256 KNOTE_LOCKED(&filp->f_selinfo.si_note, 1);
1257 spin_unlock(&filp->f_kqlock);
1258 }
1259
1260 static void
1261 linux_file_kqfilter_detach(struct knote *kn)
1262 {
1263 struct linux_file *filp = kn->kn_hook;
1264
1265 spin_lock(&filp->f_kqlock);
1266 knlist_remove(&filp->f_selinfo.si_note, kn, 1);
1267 spin_unlock(&filp->f_kqlock);
1268 }
1269
1270 static int
1271 linux_file_kqfilter_read_event(struct knote *kn, long hint)
1272 {
1273 struct linux_file *filp = kn->kn_hook;
1274
1275 mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1276
1277 return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0);
1278 }
1279
1280 static int
1281 linux_file_kqfilter_write_event(struct knote *kn, long hint)
1282 {
1283 struct linux_file *filp = kn->kn_hook;
1284
1285 mtx_assert(&filp->f_kqlock.m, MA_OWNED);
1286
1287 return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0);
1288 }
1289
1290 static struct filterops linux_dev_kqfiltops_read = {
1291 .f_isfd = 1,
1292 .f_detach = linux_file_kqfilter_detach,
1293 .f_event = linux_file_kqfilter_read_event,
1294 };
1295
1296 static struct filterops linux_dev_kqfiltops_write = {
1297 .f_isfd = 1,
1298 .f_detach = linux_file_kqfilter_detach,
1299 .f_event = linux_file_kqfilter_write_event,
1300 };
1301
1302 static void
1303 linux_file_kqfilter_poll(struct linux_file *filp, int kqflags)
1304 {
1305 struct thread *td;
1306 const struct file_operations *fop;
1307 struct linux_cdev *ldev;
1308 int temp;
1309
1310 if ((filp->f_kqflags & kqflags) == 0)
1311 return;
1312
1313 td = curthread;
1314
1315 linux_get_fop(filp, &fop, &ldev);
1316 /* get the latest polling state */
1317 temp = OPW(filp->_file, td, fop->poll(filp, NULL));
1318 linux_drop_fop(ldev);
1319
1320 spin_lock(&filp->f_kqlock);
1321 /* clear kqflags */
1322 filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ |
1323 LINUX_KQ_FLAG_NEED_WRITE);
1324 /* update kqflags */
1325 if ((temp & (POLLIN | POLLOUT)) != 0) {
1326 if ((temp & POLLIN) != 0)
1327 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ;
1328 if ((temp & POLLOUT) != 0)
1329 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE;
1330
1331 /* make sure the "knote" gets woken up */
1332 KNOTE_LOCKED(&filp->f_selinfo.si_note, 0);
1333 }
1334 spin_unlock(&filp->f_kqlock);
1335 }
1336
1337 static int
1338 linux_file_kqfilter(struct file *file, struct knote *kn)
1339 {
1340 struct linux_file *filp;
1341 struct thread *td;
1342 int error;
1343
1344 td = curthread;
1345 filp = (struct linux_file *)file->f_data;
1346 filp->f_flags = file->f_flag;
1347 if (filp->f_op->poll == NULL)
1348 return (EINVAL);
1349
1350 spin_lock(&filp->f_kqlock);
1351 switch (kn->kn_filter) {
1352 case EVFILT_READ:
1353 filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ;
1354 kn->kn_fop = &linux_dev_kqfiltops_read;
1355 kn->kn_hook = filp;
1356 knlist_add(&filp->f_selinfo.si_note, kn, 1);
1357 error = 0;
1358 break;
1359 case EVFILT_WRITE:
1360 filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE;
1361 kn->kn_fop = &linux_dev_kqfiltops_write;
1362 kn->kn_hook = filp;
1363 knlist_add(&filp->f_selinfo.si_note, kn, 1);
1364 error = 0;
1365 break;
1366 default:
1367 error = EINVAL;
1368 break;
1369 }
1370 spin_unlock(&filp->f_kqlock);
1371
1372 if (error == 0) {
1373 linux_set_current(td);
1374
1375 /* update kqfilter status, if any */
1376 linux_file_kqfilter_poll(filp,
1377 LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE);
1378 }
1379 return (error);
1380 }
1381
1382 static int
1383 linux_file_mmap_single(struct file *fp, const struct file_operations *fop,
1384 vm_ooffset_t *offset, vm_size_t size, struct vm_object **object,
1385 int nprot, bool is_shared, struct thread *td)
1386 {
1387 struct task_struct *task;
1388 struct vm_area_struct *vmap;
1389 struct mm_struct *mm;
1390 struct linux_file *filp;
1391 vm_memattr_t attr;
1392 int error;
1393
1394 filp = (struct linux_file *)fp->f_data;
1395 filp->f_flags = fp->f_flag;
1396
1397 if (fop->mmap == NULL)
1398 return (EOPNOTSUPP);
1399
1400 linux_set_current(td);
1401
1402 /*
1403 * The same VM object might be shared by multiple processes
1404 * and the mm_struct is usually freed when a process exits.
1405 *
1406 * The atomic reference below makes sure the mm_struct is
1407 * available as long as the vmap is in the linux_vma_head.
1408 */
1409 task = current;
1410 mm = task->mm;
1411 if (atomic_inc_not_zero(&mm->mm_users) == 0)
1412 return (EINVAL);
1413
1414 vmap = kzalloc(sizeof(*vmap), GFP_KERNEL);
1415 vmap->vm_start = 0;
1416 vmap->vm_end = size;
1417 vmap->vm_pgoff = *offset / PAGE_SIZE;
1418 vmap->vm_pfn = 0;
1419 vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL);
1420 if (is_shared)
1421 vmap->vm_flags |= VM_SHARED;
1422 vmap->vm_ops = NULL;
1423 vmap->vm_file = get_file(filp);
1424 vmap->vm_mm = mm;
1425
1426 if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) {
1427 error = linux_get_error(task, EINTR);
1428 } else {
1429 error = -OPW(fp, td, fop->mmap(filp, vmap));
1430 error = linux_get_error(task, error);
1431 up_write(&vmap->vm_mm->mmap_sem);
1432 }
1433
1434 if (error != 0) {
1435 linux_cdev_handle_free(vmap);
1436 return (error);
1437 }
1438
1439 attr = pgprot2cachemode(vmap->vm_page_prot);
1440
1441 if (vmap->vm_ops != NULL) {
1442 struct vm_area_struct *ptr;
1443 void *vm_private_data;
1444 bool vm_no_fault;
1445
1446 if (vmap->vm_ops->open == NULL ||
1447 vmap->vm_ops->close == NULL ||
1448 vmap->vm_private_data == NULL) {
1449 /* free allocated VM area struct */
1450 linux_cdev_handle_free(vmap);
1451 return (EINVAL);
1452 }
1453
1454 vm_private_data = vmap->vm_private_data;
1455
1456 rw_wlock(&linux_vma_lock);
1457 TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) {
1458 if (ptr->vm_private_data == vm_private_data)
1459 break;
1460 }
1461 /* check if there is an existing VM area struct */
1462 if (ptr != NULL) {
1463 /* check if the VM area structure is invalid */
1464 if (ptr->vm_ops == NULL ||
1465 ptr->vm_ops->open == NULL ||
1466 ptr->vm_ops->close == NULL) {
1467 error = ESTALE;
1468 vm_no_fault = 1;
1469 } else {
1470 error = EEXIST;
1471 vm_no_fault = (ptr->vm_ops->fault == NULL);
1472 }
1473 } else {
1474 /* insert VM area structure into list */
1475 TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry);
1476 error = 0;
1477 vm_no_fault = (vmap->vm_ops->fault == NULL);
1478 }
1479 rw_wunlock(&linux_vma_lock);
1480
1481 if (error != 0) {
1482 /* free allocated VM area struct */
1483 linux_cdev_handle_free(vmap);
1484 /* check for stale VM area struct */
1485 if (error != EEXIST)
1486 return (error);
1487 }
1488
1489 /* check if there is no fault handler */
1490 if (vm_no_fault) {
1491 *object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE,
1492 &linux_cdev_pager_ops[1], size, nprot, *offset,
1493 td->td_ucred);
1494 } else {
1495 *object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE,
1496 &linux_cdev_pager_ops[0], size, nprot, *offset,
1497 td->td_ucred);
1498 }
1499
1500 /* check if allocating the VM object failed */
1501 if (*object == NULL) {
1502 if (error == 0) {
1503 /* remove VM area struct from list */
1504 linux_cdev_handle_remove(vmap);
1505 /* free allocated VM area struct */
1506 linux_cdev_handle_free(vmap);
1507 }
1508 return (EINVAL);
1509 }
1510 } else {
1511 struct sglist *sg;
1512
1513 sg = sglist_alloc(1, M_WAITOK);
1514 sglist_append_phys(sg,
1515 (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len);
1516
1517 *object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len,
1518 nprot, 0, td->td_ucred);
1519
1520 linux_cdev_handle_free(vmap);
1521
1522 if (*object == NULL) {
1523 sglist_free(sg);
1524 return (EINVAL);
1525 }
1526 }
1527
1528 if (attr != VM_MEMATTR_DEFAULT) {
1529 VM_OBJECT_WLOCK(*object);
1530 vm_object_set_memattr(*object, attr);
1531 VM_OBJECT_WUNLOCK(*object);
1532 }
1533 *offset = 0;
1534 return (0);
1535 }
1536
1537 struct cdevsw linuxcdevsw = {
1538 .d_version = D_VERSION,
1539 .d_fdopen = linux_dev_fdopen,
1540 .d_name = "lkpidev",
1541 };
1542
1543 static int
1544 linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred,
1545 int flags, struct thread *td)
1546 {
1547 struct linux_file *filp;
1548 const struct file_operations *fop;
1549 struct linux_cdev *ldev;
1550 ssize_t bytes;
1551 int error;
1552
1553 error = 0;
1554 filp = (struct linux_file *)file->f_data;
1555 filp->f_flags = file->f_flag;
1556 /* XXX no support for I/O vectors currently */
1557 if (uio->uio_iovcnt != 1)
1558 return (EOPNOTSUPP);
1559 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1560 return (EINVAL);
1561 linux_set_current(td);
1562 linux_get_fop(filp, &fop, &ldev);
1563 if (fop->read != NULL) {
1564 bytes = OPW(file, td, fop->read(filp,
1565 uio->uio_iov->iov_base,
1566 uio->uio_iov->iov_len, &uio->uio_offset));
1567 if (bytes >= 0) {
1568 uio->uio_iov->iov_base =
1569 ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1570 uio->uio_iov->iov_len -= bytes;
1571 uio->uio_resid -= bytes;
1572 } else {
1573 error = linux_get_error(current, -bytes);
1574 }
1575 } else
1576 error = ENXIO;
1577
1578 /* update kqfilter status, if any */
1579 linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ);
1580 linux_drop_fop(ldev);
1581
1582 return (error);
1583 }
1584
1585 static int
1586 linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred,
1587 int flags, struct thread *td)
1588 {
1589 struct linux_file *filp;
1590 const struct file_operations *fop;
1591 struct linux_cdev *ldev;
1592 ssize_t bytes;
1593 int error;
1594
1595 filp = (struct linux_file *)file->f_data;
1596 filp->f_flags = file->f_flag;
1597 /* XXX no support for I/O vectors currently */
1598 if (uio->uio_iovcnt != 1)
1599 return (EOPNOTSUPP);
1600 if (uio->uio_resid > DEVFS_IOSIZE_MAX)
1601 return (EINVAL);
1602 linux_set_current(td);
1603 linux_get_fop(filp, &fop, &ldev);
1604 if (fop->write != NULL) {
1605 bytes = OPW(file, td, fop->write(filp,
1606 uio->uio_iov->iov_base,
1607 uio->uio_iov->iov_len, &uio->uio_offset));
1608 if (bytes >= 0) {
1609 uio->uio_iov->iov_base =
1610 ((uint8_t *)uio->uio_iov->iov_base) + bytes;
1611 uio->uio_iov->iov_len -= bytes;
1612 uio->uio_resid -= bytes;
1613 error = 0;
1614 } else {
1615 error = linux_get_error(current, -bytes);
1616 }
1617 } else
1618 error = ENXIO;
1619
1620 /* update kqfilter status, if any */
1621 linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE);
1622
1623 linux_drop_fop(ldev);
1624
1625 return (error);
1626 }
1627
1628 static int
1629 linux_file_poll(struct file *file, int events, struct ucred *active_cred,
1630 struct thread *td)
1631 {
1632 struct linux_file *filp;
1633 const struct file_operations *fop;
1634 struct linux_cdev *ldev;
1635 int revents;
1636
1637 filp = (struct linux_file *)file->f_data;
1638 filp->f_flags = file->f_flag;
1639 linux_set_current(td);
1640 linux_get_fop(filp, &fop, &ldev);
1641 if (fop->poll != NULL) {
1642 revents = OPW(file, td, fop->poll(filp,
1643 LINUX_POLL_TABLE_NORMAL)) & events;
1644 } else {
1645 revents = 0;
1646 }
1647 linux_drop_fop(ldev);
1648 return (revents);
1649 }
1650
1651 static int
1652 linux_file_close(struct file *file, struct thread *td)
1653 {
1654 struct linux_file *filp;
1655 int (*release)(struct inode *, struct linux_file *);
1656 const struct file_operations *fop;
1657 struct linux_cdev *ldev;
1658 int error;
1659
1660 filp = (struct linux_file *)file->f_data;
1661
1662 KASSERT(file_count(filp) == 0,
1663 ("File refcount(%d) is not zero", file_count(filp)));
1664
1665 if (td == NULL)
1666 td = curthread;
1667
1668 error = 0;
1669 filp->f_flags = file->f_flag;
1670 linux_set_current(td);
1671 linux_poll_wait_dequeue(filp);
1672 linux_get_fop(filp, &fop, &ldev);
1673 /*
1674 * Always use the real release function, if any, to avoid
1675 * leaking device resources:
1676 */
1677 release = filp->f_op->release;
1678 if (release != NULL)
1679 error = -OPW(file, td, release(filp->f_vnode, filp));
1680 funsetown(&filp->f_sigio);
1681 if (filp->f_vnode != NULL)
1682 vdrop(filp->f_vnode);
1683 linux_drop_fop(ldev);
1684 ldev = filp->f_cdev;
1685 if (ldev != NULL)
1686 linux_cdev_deref(ldev);
1687 linux_synchronize_rcu(RCU_TYPE_REGULAR);
1688 kfree(filp);
1689
1690 return (error);
1691 }
1692
1693 static int
1694 linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred,
1695 struct thread *td)
1696 {
1697 struct linux_file *filp;
1698 const struct file_operations *fop;
1699 struct linux_cdev *ldev;
1700 struct fiodgname_arg *fgn;
1701 const char *p;
1702 int error, i;
1703
1704 error = 0;
1705 filp = (struct linux_file *)fp->f_data;
1706 filp->f_flags = fp->f_flag;
1707 linux_get_fop(filp, &fop, &ldev);
1708
1709 linux_set_current(td);
1710 switch (cmd) {
1711 case FIONBIO:
1712 break;
1713 case FIOASYNC:
1714 if (fop->fasync == NULL)
1715 break;
1716 error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC));
1717 break;
1718 case FIOSETOWN:
1719 error = fsetown(*(int *)data, &filp->f_sigio);
1720 if (error == 0) {
1721 if (fop->fasync == NULL)
1722 break;
1723 error = -OPW(fp, td, fop->fasync(0, filp,
1724 fp->f_flag & FASYNC));
1725 }
1726 break;
1727 case FIOGETOWN:
1728 *(int *)data = fgetown(&filp->f_sigio);
1729 break;
1730 case FIODGNAME:
1731 #ifdef COMPAT_FREEBSD32
1732 case FIODGNAME_32:
1733 #endif
1734 if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) {
1735 error = ENXIO;
1736 break;
1737 }
1738 fgn = data;
1739 p = devtoname(filp->f_cdev->cdev);
1740 i = strlen(p) + 1;
1741 if (i > fgn->len) {
1742 error = EINVAL;
1743 break;
1744 }
1745 error = copyout(p, fiodgname_buf_get_ptr(fgn, cmd), i);
1746 break;
1747 default:
1748 error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td);
1749 break;
1750 }
1751 linux_drop_fop(ldev);
1752 return (error);
1753 }
1754
1755 static int
1756 linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot,
1757 vm_prot_t maxprot, int flags, struct file *fp,
1758 vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp)
1759 {
1760 /*
1761 * Character devices do not provide private mappings
1762 * of any kind:
1763 */
1764 if ((maxprot & VM_PROT_WRITE) == 0 &&
1765 (prot & VM_PROT_WRITE) != 0)
1766 return (EACCES);
1767 if ((flags & (MAP_PRIVATE | MAP_COPY)) != 0)
1768 return (EINVAL);
1769
1770 return (linux_file_mmap_single(fp, fop, foff, objsize, objp,
1771 (int)prot, (flags & MAP_SHARED) ? true : false, td));
1772 }
1773
1774 static int
1775 linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size,
1776 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff,
1777 struct thread *td)
1778 {
1779 struct linux_file *filp;
1780 const struct file_operations *fop;
1781 struct linux_cdev *ldev;
1782 struct mount *mp;
1783 struct vnode *vp;
1784 vm_object_t object;
1785 vm_prot_t maxprot;
1786 int error;
1787
1788 filp = (struct linux_file *)fp->f_data;
1789
1790 vp = filp->f_vnode;
1791 if (vp == NULL)
1792 return (EOPNOTSUPP);
1793
1794 /*
1795 * Ensure that file and memory protections are
1796 * compatible.
1797 */
1798 mp = vp->v_mount;
1799 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) {
1800 maxprot = VM_PROT_NONE;
1801 if ((prot & VM_PROT_EXECUTE) != 0)
1802 return (EACCES);
1803 } else
1804 maxprot = VM_PROT_EXECUTE;
1805 if ((fp->f_flag & FREAD) != 0)
1806 maxprot |= VM_PROT_READ;
1807 else if ((prot & VM_PROT_READ) != 0)
1808 return (EACCES);
1809
1810 /*
1811 * If we are sharing potential changes via MAP_SHARED and we
1812 * are trying to get write permission although we opened it
1813 * without asking for it, bail out.
1814 *
1815 * Note that most character devices always share mappings.
1816 *
1817 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE
1818 * requests rather than doing it here.
1819 */
1820 if ((flags & MAP_SHARED) != 0) {
1821 if ((fp->f_flag & FWRITE) != 0)
1822 maxprot |= VM_PROT_WRITE;
1823 else if ((prot & VM_PROT_WRITE) != 0)
1824 return (EACCES);
1825 }
1826 maxprot &= cap_maxprot;
1827
1828 linux_get_fop(filp, &fop, &ldev);
1829 error = linux_file_mmap_sub(td, size, prot, maxprot, flags, fp,
1830 &foff, fop, &object);
1831 if (error != 0)
1832 goto out;
1833
1834 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object,
1835 foff, FALSE, td);
1836 if (error != 0)
1837 vm_object_deallocate(object);
1838 out:
1839 linux_drop_fop(ldev);
1840 return (error);
1841 }
1842
1843 static int
1844 linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred)
1845 {
1846 struct linux_file *filp;
1847 struct vnode *vp;
1848 int error;
1849
1850 filp = (struct linux_file *)fp->f_data;
1851 if (filp->f_vnode == NULL)
1852 return (EOPNOTSUPP);
1853
1854 vp = filp->f_vnode;
1855
1856 vn_lock(vp, LK_SHARED | LK_RETRY);
1857 error = VOP_STAT(vp, sb, curthread->td_ucred, NOCRED);
1858 VOP_UNLOCK(vp);
1859
1860 return (error);
1861 }
1862
1863 static int
1864 linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif,
1865 struct filedesc *fdp)
1866 {
1867 struct linux_file *filp;
1868 struct vnode *vp;
1869 int error;
1870
1871 filp = fp->f_data;
1872 vp = filp->f_vnode;
1873 if (vp == NULL) {
1874 error = 0;
1875 kif->kf_type = KF_TYPE_DEV;
1876 } else {
1877 vref(vp);
1878 FILEDESC_SUNLOCK(fdp);
1879 error = vn_fill_kinfo_vnode(vp, kif);
1880 vrele(vp);
1881 kif->kf_type = KF_TYPE_VNODE;
1882 FILEDESC_SLOCK(fdp);
1883 }
1884 return (error);
1885 }
1886
1887 unsigned int
1888 linux_iminor(struct inode *inode)
1889 {
1890 struct linux_cdev *ldev;
1891
1892 if (inode == NULL || inode->v_rdev == NULL ||
1893 inode->v_rdev->si_devsw != &linuxcdevsw)
1894 return (-1U);
1895 ldev = inode->v_rdev->si_drv1;
1896 if (ldev == NULL)
1897 return (-1U);
1898
1899 return (minor(ldev->dev));
1900 }
1901
1902 struct fileops linuxfileops = {
1903 .fo_read = linux_file_read,
1904 .fo_write = linux_file_write,
1905 .fo_truncate = invfo_truncate,
1906 .fo_kqfilter = linux_file_kqfilter,
1907 .fo_stat = linux_file_stat,
1908 .fo_fill_kinfo = linux_file_fill_kinfo,
1909 .fo_poll = linux_file_poll,
1910 .fo_close = linux_file_close,
1911 .fo_ioctl = linux_file_ioctl,
1912 .fo_mmap = linux_file_mmap,
1913 .fo_chmod = invfo_chmod,
1914 .fo_chown = invfo_chown,
1915 .fo_sendfile = invfo_sendfile,
1916 .fo_flags = DFLAG_PASSABLE,
1917 };
1918
1919 /*
1920 * Hash of vmmap addresses. This is infrequently accessed and does not
1921 * need to be particularly large. This is done because we must store the
1922 * caller's idea of the map size to properly unmap.
1923 */
1924 struct vmmap {
1925 LIST_ENTRY(vmmap) vm_next;
1926 void *vm_addr;
1927 unsigned long vm_size;
1928 };
1929
1930 struct vmmaphd {
1931 struct vmmap *lh_first;
1932 };
1933 #define VMMAP_HASH_SIZE 64
1934 #define VMMAP_HASH_MASK (VMMAP_HASH_SIZE - 1)
1935 #define VM_HASH(addr) ((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK
1936 static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE];
1937 static struct mtx vmmaplock;
1938
1939 static void
1940 vmmap_add(void *addr, unsigned long size)
1941 {
1942 struct vmmap *vmmap;
1943
1944 vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL);
1945 mtx_lock(&vmmaplock);
1946 vmmap->vm_size = size;
1947 vmmap->vm_addr = addr;
1948 LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next);
1949 mtx_unlock(&vmmaplock);
1950 }
1951
1952 static struct vmmap *
1953 vmmap_remove(void *addr)
1954 {
1955 struct vmmap *vmmap;
1956
1957 mtx_lock(&vmmaplock);
1958 LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next)
1959 if (vmmap->vm_addr == addr)
1960 break;
1961 if (vmmap)
1962 LIST_REMOVE(vmmap, vm_next);
1963 mtx_unlock(&vmmaplock);
1964
1965 return (vmmap);
1966 }
1967
1968 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) || defined(__riscv)
1969 void *
1970 _ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr)
1971 {
1972 void *addr;
1973
1974 addr = pmap_mapdev_attr(phys_addr, size, attr);
1975 if (addr == NULL)
1976 return (NULL);
1977 vmmap_add(addr, size);
1978
1979 return (addr);
1980 }
1981 #endif
1982
1983 void
1984 iounmap(void *addr)
1985 {
1986 struct vmmap *vmmap;
1987
1988 vmmap = vmmap_remove(addr);
1989 if (vmmap == NULL)
1990 return;
1991 #if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) || defined(__riscv)
1992 pmap_unmapdev(addr, vmmap->vm_size);
1993 #endif
1994 kfree(vmmap);
1995 }
1996
1997 void *
1998 vmap(struct page **pages, unsigned int count, unsigned long flags, int prot)
1999 {
2000 vm_offset_t off;
2001 size_t size;
2002
2003 size = count * PAGE_SIZE;
2004 off = kva_alloc(size);
2005 if (off == 0)
2006 return (NULL);
2007 vmmap_add((void *)off, size);
2008 pmap_qenter(off, pages, count);
2009
2010 return ((void *)off);
2011 }
2012
2013 void
2014 vunmap(void *addr)
2015 {
2016 struct vmmap *vmmap;
2017
2018 vmmap = vmmap_remove(addr);
2019 if (vmmap == NULL)
2020 return;
2021 pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE);
2022 kva_free((vm_offset_t)addr, vmmap->vm_size);
2023 kfree(vmmap);
2024 }
2025
2026 static char *
2027 devm_kvasprintf(struct device *dev, gfp_t gfp, const char *fmt, va_list ap)
2028 {
2029 unsigned int len;
2030 char *p;
2031 va_list aq;
2032
2033 va_copy(aq, ap);
2034 len = vsnprintf(NULL, 0, fmt, aq);
2035 va_end(aq);
2036
2037 if (dev != NULL)
2038 p = devm_kmalloc(dev, len + 1, gfp);
2039 else
2040 p = kmalloc(len + 1, gfp);
2041 if (p != NULL)
2042 vsnprintf(p, len + 1, fmt, ap);
2043
2044 return (p);
2045 }
2046
2047 char *
2048 kvasprintf(gfp_t gfp, const char *fmt, va_list ap)
2049 {
2050
2051 return (devm_kvasprintf(NULL, gfp, fmt, ap));
2052 }
2053
2054 char *
2055 lkpi_devm_kasprintf(struct device *dev, gfp_t gfp, const char *fmt, ...)
2056 {
2057 va_list ap;
2058 char *p;
2059
2060 va_start(ap, fmt);
2061 p = devm_kvasprintf(dev, gfp, fmt, ap);
2062 va_end(ap);
2063
2064 return (p);
2065 }
2066
2067 char *
2068 kasprintf(gfp_t gfp, const char *fmt, ...)
2069 {
2070 va_list ap;
2071 char *p;
2072
2073 va_start(ap, fmt);
2074 p = kvasprintf(gfp, fmt, ap);
2075 va_end(ap);
2076
2077 return (p);
2078 }
2079
2080 static void
2081 linux_timer_callback_wrapper(void *context)
2082 {
2083 struct timer_list *timer;
2084
2085 timer = context;
2086
2087 if (linux_set_current_flags(curthread, M_NOWAIT)) {
2088 /* try again later */
2089 callout_reset(&timer->callout, 1,
2090 &linux_timer_callback_wrapper, timer);
2091 return;
2092 }
2093
2094 timer->function(timer->data);
2095 }
2096
2097 int
2098 mod_timer(struct timer_list *timer, int expires)
2099 {
2100 int ret;
2101
2102 timer->expires = expires;
2103 ret = callout_reset(&timer->callout,
2104 linux_timer_jiffies_until(expires),
2105 &linux_timer_callback_wrapper, timer);
2106
2107 MPASS(ret == 0 || ret == 1);
2108
2109 return (ret == 1);
2110 }
2111
2112 void
2113 add_timer(struct timer_list *timer)
2114 {
2115
2116 callout_reset(&timer->callout,
2117 linux_timer_jiffies_until(timer->expires),
2118 &linux_timer_callback_wrapper, timer);
2119 }
2120
2121 void
2122 add_timer_on(struct timer_list *timer, int cpu)
2123 {
2124
2125 callout_reset_on(&timer->callout,
2126 linux_timer_jiffies_until(timer->expires),
2127 &linux_timer_callback_wrapper, timer, cpu);
2128 }
2129
2130 int
2131 del_timer(struct timer_list *timer)
2132 {
2133
2134 if (callout_stop(&(timer)->callout) == -1)
2135 return (0);
2136 return (1);
2137 }
2138
2139 int
2140 del_timer_sync(struct timer_list *timer)
2141 {
2142
2143 if (callout_drain(&(timer)->callout) == -1)
2144 return (0);
2145 return (1);
2146 }
2147
2148 /* greatest common divisor, Euclid equation */
2149 static uint64_t
2150 lkpi_gcd_64(uint64_t a, uint64_t b)
2151 {
2152 uint64_t an;
2153 uint64_t bn;
2154
2155 while (b != 0) {
2156 an = b;
2157 bn = a % b;
2158 a = an;
2159 b = bn;
2160 }
2161 return (a);
2162 }
2163
2164 uint64_t lkpi_nsec2hz_rem;
2165 uint64_t lkpi_nsec2hz_div = 1000000000ULL;
2166 uint64_t lkpi_nsec2hz_max;
2167
2168 uint64_t lkpi_usec2hz_rem;
2169 uint64_t lkpi_usec2hz_div = 1000000ULL;
2170 uint64_t lkpi_usec2hz_max;
2171
2172 uint64_t lkpi_msec2hz_rem;
2173 uint64_t lkpi_msec2hz_div = 1000ULL;
2174 uint64_t lkpi_msec2hz_max;
2175
2176 static void
2177 linux_timer_init(void *arg)
2178 {
2179 uint64_t gcd;
2180
2181 /*
2182 * Compute an internal HZ value which can divide 2**32 to
2183 * avoid timer rounding problems when the tick value wraps
2184 * around 2**32:
2185 */
2186 linux_timer_hz_mask = 1;
2187 while (linux_timer_hz_mask < (unsigned long)hz)
2188 linux_timer_hz_mask *= 2;
2189 linux_timer_hz_mask--;
2190
2191 /* compute some internal constants */
2192
2193 lkpi_nsec2hz_rem = hz;
2194 lkpi_usec2hz_rem = hz;
2195 lkpi_msec2hz_rem = hz;
2196
2197 gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div);
2198 lkpi_nsec2hz_rem /= gcd;
2199 lkpi_nsec2hz_div /= gcd;
2200 lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem;
2201
2202 gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div);
2203 lkpi_usec2hz_rem /= gcd;
2204 lkpi_usec2hz_div /= gcd;
2205 lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem;
2206
2207 gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div);
2208 lkpi_msec2hz_rem /= gcd;
2209 lkpi_msec2hz_div /= gcd;
2210 lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem;
2211 }
2212 SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL);
2213
2214 void
2215 linux_complete_common(struct completion *c, int all)
2216 {
2217 int wakeup_swapper;
2218
2219 sleepq_lock(c);
2220 if (all) {
2221 c->done = UINT_MAX;
2222 wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0);
2223 } else {
2224 if (c->done != UINT_MAX)
2225 c->done++;
2226 wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0);
2227 }
2228 sleepq_release(c);
2229 if (wakeup_swapper)
2230 kick_proc0();
2231 }
2232
2233 /*
2234 * Indefinite wait for done != 0 with or without signals.
2235 */
2236 int
2237 linux_wait_for_common(struct completion *c, int flags)
2238 {
2239 struct task_struct *task;
2240 int error;
2241
2242 if (SCHEDULER_STOPPED())
2243 return (0);
2244
2245 task = current;
2246
2247 if (flags != 0)
2248 flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2249 else
2250 flags = SLEEPQ_SLEEP;
2251 error = 0;
2252 for (;;) {
2253 sleepq_lock(c);
2254 if (c->done)
2255 break;
2256 sleepq_add(c, NULL, "completion", flags, 0);
2257 if (flags & SLEEPQ_INTERRUPTIBLE) {
2258 DROP_GIANT();
2259 error = -sleepq_wait_sig(c, 0);
2260 PICKUP_GIANT();
2261 if (error != 0) {
2262 linux_schedule_save_interrupt_value(task, error);
2263 error = -ERESTARTSYS;
2264 goto intr;
2265 }
2266 } else {
2267 DROP_GIANT();
2268 sleepq_wait(c, 0);
2269 PICKUP_GIANT();
2270 }
2271 }
2272 if (c->done != UINT_MAX)
2273 c->done--;
2274 sleepq_release(c);
2275
2276 intr:
2277 return (error);
2278 }
2279
2280 /*
2281 * Time limited wait for done != 0 with or without signals.
2282 */
2283 int
2284 linux_wait_for_timeout_common(struct completion *c, int timeout, int flags)
2285 {
2286 struct task_struct *task;
2287 int end = jiffies + timeout;
2288 int error;
2289
2290 if (SCHEDULER_STOPPED())
2291 return (0);
2292
2293 task = current;
2294
2295 if (flags != 0)
2296 flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP;
2297 else
2298 flags = SLEEPQ_SLEEP;
2299
2300 for (;;) {
2301 sleepq_lock(c);
2302 if (c->done)
2303 break;
2304 sleepq_add(c, NULL, "completion", flags, 0);
2305 sleepq_set_timeout(c, linux_timer_jiffies_until(end));
2306
2307 DROP_GIANT();
2308 if (flags & SLEEPQ_INTERRUPTIBLE)
2309 error = -sleepq_timedwait_sig(c, 0);
2310 else
2311 error = -sleepq_timedwait(c, 0);
2312 PICKUP_GIANT();
2313
2314 if (error != 0) {
2315 /* check for timeout */
2316 if (error == -EWOULDBLOCK) {
2317 error = 0; /* timeout */
2318 } else {
2319 /* signal happened */
2320 linux_schedule_save_interrupt_value(task, error);
2321 error = -ERESTARTSYS;
2322 }
2323 goto done;
2324 }
2325 }
2326 if (c->done != UINT_MAX)
2327 c->done--;
2328 sleepq_release(c);
2329
2330 /* return how many jiffies are left */
2331 error = linux_timer_jiffies_until(end);
2332 done:
2333 return (error);
2334 }
2335
2336 int
2337 linux_try_wait_for_completion(struct completion *c)
2338 {
2339 int isdone;
2340
2341 sleepq_lock(c);
2342 isdone = (c->done != 0);
2343 if (c->done != 0 && c->done != UINT_MAX)
2344 c->done--;
2345 sleepq_release(c);
2346 return (isdone);
2347 }
2348
2349 int
2350 linux_completion_done(struct completion *c)
2351 {
2352 int isdone;
2353
2354 sleepq_lock(c);
2355 isdone = (c->done != 0);
2356 sleepq_release(c);
2357 return (isdone);
2358 }
2359
2360 static void
2361 linux_cdev_deref(struct linux_cdev *ldev)
2362 {
2363 if (refcount_release(&ldev->refs) &&
2364 ldev->kobj.ktype == &linux_cdev_ktype)
2365 kfree(ldev);
2366 }
2367
2368 static void
2369 linux_cdev_release(struct kobject *kobj)
2370 {
2371 struct linux_cdev *cdev;
2372 struct kobject *parent;
2373
2374 cdev = container_of(kobj, struct linux_cdev, kobj);
2375 parent = kobj->parent;
2376 linux_destroy_dev(cdev);
2377 linux_cdev_deref(cdev);
2378 kobject_put(parent);
2379 }
2380
2381 static void
2382 linux_cdev_static_release(struct kobject *kobj)
2383 {
2384 struct cdev *cdev;
2385 struct linux_cdev *ldev;
2386
2387 ldev = container_of(kobj, struct linux_cdev, kobj);
2388 cdev = ldev->cdev;
2389 if (cdev != NULL) {
2390 destroy_dev(cdev);
2391 ldev->cdev = NULL;
2392 }
2393 kobject_put(kobj->parent);
2394 }
2395
2396 int
2397 linux_cdev_device_add(struct linux_cdev *ldev, struct device *dev)
2398 {
2399 int ret;
2400
2401 if (dev->devt != 0) {
2402 /* Set parent kernel object. */
2403 ldev->kobj.parent = &dev->kobj;
2404
2405 /*
2406 * Unlike Linux we require the kobject of the
2407 * character device structure to have a valid name
2408 * before calling this function:
2409 */
2410 if (ldev->kobj.name == NULL)
2411 return (-EINVAL);
2412
2413 ret = cdev_add(ldev, dev->devt, 1);
2414 if (ret)
2415 return (ret);
2416 }
2417 ret = device_add(dev);
2418 if (ret != 0 && dev->devt != 0)
2419 cdev_del(ldev);
2420 return (ret);
2421 }
2422
2423 void
2424 linux_cdev_device_del(struct linux_cdev *ldev, struct device *dev)
2425 {
2426 device_del(dev);
2427
2428 if (dev->devt != 0)
2429 cdev_del(ldev);
2430 }
2431
2432 static void
2433 linux_destroy_dev(struct linux_cdev *ldev)
2434 {
2435
2436 if (ldev->cdev == NULL)
2437 return;
2438
2439 MPASS((ldev->siref & LDEV_SI_DTR) == 0);
2440 MPASS(ldev->kobj.ktype == &linux_cdev_ktype);
2441
2442 atomic_set_int(&ldev->siref, LDEV_SI_DTR);
2443 while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0)
2444 pause("ldevdtr", hz / 4);
2445
2446 destroy_dev(ldev->cdev);
2447 ldev->cdev = NULL;
2448 }
2449
2450 const struct kobj_type linux_cdev_ktype = {
2451 .release = linux_cdev_release,
2452 };
2453
2454 const struct kobj_type linux_cdev_static_ktype = {
2455 .release = linux_cdev_static_release,
2456 };
2457
2458 static void
2459 linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate)
2460 {
2461 struct notifier_block *nb;
2462 struct netdev_notifier_info ni;
2463
2464 nb = arg;
2465 ni.ifp = ifp;
2466 ni.dev = (struct net_device *)ifp;
2467 if (linkstate == LINK_STATE_UP)
2468 nb->notifier_call(nb, NETDEV_UP, &ni);
2469 else
2470 nb->notifier_call(nb, NETDEV_DOWN, &ni);
2471 }
2472
2473 static void
2474 linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp)
2475 {
2476 struct notifier_block *nb;
2477 struct netdev_notifier_info ni;
2478
2479 nb = arg;
2480 ni.ifp = ifp;
2481 ni.dev = (struct net_device *)ifp;
2482 nb->notifier_call(nb, NETDEV_REGISTER, &ni);
2483 }
2484
2485 static void
2486 linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp)
2487 {
2488 struct notifier_block *nb;
2489 struct netdev_notifier_info ni;
2490
2491 nb = arg;
2492 ni.ifp = ifp;
2493 ni.dev = (struct net_device *)ifp;
2494 nb->notifier_call(nb, NETDEV_UNREGISTER, &ni);
2495 }
2496
2497 static void
2498 linux_handle_iflladdr_event(void *arg, struct ifnet *ifp)
2499 {
2500 struct notifier_block *nb;
2501 struct netdev_notifier_info ni;
2502
2503 nb = arg;
2504 ni.ifp = ifp;
2505 ni.dev = (struct net_device *)ifp;
2506 nb->notifier_call(nb, NETDEV_CHANGEADDR, &ni);
2507 }
2508
2509 static void
2510 linux_handle_ifaddr_event(void *arg, struct ifnet *ifp)
2511 {
2512 struct notifier_block *nb;
2513 struct netdev_notifier_info ni;
2514
2515 nb = arg;
2516 ni.ifp = ifp;
2517 ni.dev = (struct net_device *)ifp;
2518 nb->notifier_call(nb, NETDEV_CHANGEIFADDR, &ni);
2519 }
2520
2521 int
2522 register_netdevice_notifier(struct notifier_block *nb)
2523 {
2524
2525 nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER(
2526 ifnet_link_event, linux_handle_ifnet_link_event, nb, 0);
2527 nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER(
2528 ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0);
2529 nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER(
2530 ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0);
2531 nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER(
2532 iflladdr_event, linux_handle_iflladdr_event, nb, 0);
2533
2534 return (0);
2535 }
2536
2537 int
2538 register_inetaddr_notifier(struct notifier_block *nb)
2539 {
2540
2541 nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER(
2542 ifaddr_event, linux_handle_ifaddr_event, nb, 0);
2543 return (0);
2544 }
2545
2546 int
2547 unregister_netdevice_notifier(struct notifier_block *nb)
2548 {
2549
2550 EVENTHANDLER_DEREGISTER(ifnet_link_event,
2551 nb->tags[NETDEV_UP]);
2552 EVENTHANDLER_DEREGISTER(ifnet_arrival_event,
2553 nb->tags[NETDEV_REGISTER]);
2554 EVENTHANDLER_DEREGISTER(ifnet_departure_event,
2555 nb->tags[NETDEV_UNREGISTER]);
2556 EVENTHANDLER_DEREGISTER(iflladdr_event,
2557 nb->tags[NETDEV_CHANGEADDR]);
2558
2559 return (0);
2560 }
2561
2562 int
2563 unregister_inetaddr_notifier(struct notifier_block *nb)
2564 {
2565
2566 EVENTHANDLER_DEREGISTER(ifaddr_event,
2567 nb->tags[NETDEV_CHANGEIFADDR]);
2568
2569 return (0);
2570 }
2571
2572 struct list_sort_thunk {
2573 int (*cmp)(void *, struct list_head *, struct list_head *);
2574 void *priv;
2575 };
2576
2577 static inline int
2578 linux_le_cmp(const void *d1, const void *d2, void *priv)
2579 {
2580 struct list_head *le1, *le2;
2581 struct list_sort_thunk *thunk;
2582
2583 thunk = priv;
2584 le1 = *(__DECONST(struct list_head **, d1));
2585 le2 = *(__DECONST(struct list_head **, d2));
2586 return ((thunk->cmp)(thunk->priv, le1, le2));
2587 }
2588
2589 void
2590 list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv,
2591 struct list_head *a, struct list_head *b))
2592 {
2593 struct list_sort_thunk thunk;
2594 struct list_head **ar, *le;
2595 size_t count, i;
2596
2597 count = 0;
2598 list_for_each(le, head)
2599 count++;
2600 ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK);
2601 i = 0;
2602 list_for_each(le, head)
2603 ar[i++] = le;
2604 thunk.cmp = cmp;
2605 thunk.priv = priv;
2606 qsort_r(ar, count, sizeof(struct list_head *), linux_le_cmp, &thunk);
2607 INIT_LIST_HEAD(head);
2608 for (i = 0; i < count; i++)
2609 list_add_tail(ar[i], head);
2610 free(ar, M_KMALLOC);
2611 }
2612
2613 #if defined(__i386__) || defined(__amd64__)
2614 int
2615 linux_wbinvd_on_all_cpus(void)
2616 {
2617
2618 pmap_invalidate_cache();
2619 return (0);
2620 }
2621 #endif
2622
2623 int
2624 linux_on_each_cpu(void callback(void *), void *data)
2625 {
2626
2627 smp_rendezvous(smp_no_rendezvous_barrier, callback,
2628 smp_no_rendezvous_barrier, data);
2629 return (0);
2630 }
2631
2632 int
2633 linux_in_atomic(void)
2634 {
2635
2636 return ((curthread->td_pflags & TDP_NOFAULTING) != 0);
2637 }
2638
2639 struct linux_cdev *
2640 linux_find_cdev(const char *name, unsigned major, unsigned minor)
2641 {
2642 dev_t dev = MKDEV(major, minor);
2643 struct cdev *cdev;
2644
2645 dev_lock();
2646 LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) {
2647 struct linux_cdev *ldev = cdev->si_drv1;
2648 if (ldev->dev == dev &&
2649 strcmp(kobject_name(&ldev->kobj), name) == 0) {
2650 break;
2651 }
2652 }
2653 dev_unlock();
2654
2655 return (cdev != NULL ? cdev->si_drv1 : NULL);
2656 }
2657
2658 int
2659 __register_chrdev(unsigned int major, unsigned int baseminor,
2660 unsigned int count, const char *name,
2661 const struct file_operations *fops)
2662 {
2663 struct linux_cdev *cdev;
2664 int ret = 0;
2665 int i;
2666
2667 for (i = baseminor; i < baseminor + count; i++) {
2668 cdev = cdev_alloc();
2669 cdev->ops = fops;
2670 kobject_set_name(&cdev->kobj, name);
2671
2672 ret = cdev_add(cdev, makedev(major, i), 1);
2673 if (ret != 0)
2674 break;
2675 }
2676 return (ret);
2677 }
2678
2679 int
2680 __register_chrdev_p(unsigned int major, unsigned int baseminor,
2681 unsigned int count, const char *name,
2682 const struct file_operations *fops, uid_t uid,
2683 gid_t gid, int mode)
2684 {
2685 struct linux_cdev *cdev;
2686 int ret = 0;
2687 int i;
2688
2689 for (i = baseminor; i < baseminor + count; i++) {
2690 cdev = cdev_alloc();
2691 cdev->ops = fops;
2692 kobject_set_name(&cdev->kobj, name);
2693
2694 ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode);
2695 if (ret != 0)
2696 break;
2697 }
2698 return (ret);
2699 }
2700
2701 void
2702 __unregister_chrdev(unsigned int major, unsigned int baseminor,
2703 unsigned int count, const char *name)
2704 {
2705 struct linux_cdev *cdevp;
2706 int i;
2707
2708 for (i = baseminor; i < baseminor + count; i++) {
2709 cdevp = linux_find_cdev(name, major, i);
2710 if (cdevp != NULL)
2711 cdev_del(cdevp);
2712 }
2713 }
2714
2715 void
2716 linux_dump_stack(void)
2717 {
2718 #ifdef STACK
2719 struct stack st;
2720
2721 stack_save(&st);
2722 stack_print(&st);
2723 #endif
2724 }
2725
2726 int
2727 linuxkpi_net_ratelimit(void)
2728 {
2729
2730 return (ppsratecheck(&lkpi_net_lastlog, &lkpi_net_curpps,
2731 lkpi_net_maxpps));
2732 }
2733
2734 struct io_mapping *
2735 io_mapping_create_wc(resource_size_t base, unsigned long size)
2736 {
2737 struct io_mapping *mapping;
2738
2739 mapping = kmalloc(sizeof(*mapping), GFP_KERNEL);
2740 if (mapping == NULL)
2741 return (NULL);
2742 return (io_mapping_init_wc(mapping, base, size));
2743 }
2744
2745 #if defined(__i386__) || defined(__amd64__)
2746 bool linux_cpu_has_clflush;
2747 struct cpuinfo_x86 boot_cpu_data;
2748 #endif
2749
2750 cpumask_t *
2751 lkpi_get_static_single_cpu_mask(int cpuid)
2752 {
2753
2754 KASSERT((cpuid >= 0 && cpuid < MAXCPU), ("%s: invalid cpuid %d\n",
2755 __func__, cpuid));
2756
2757 return (&static_single_cpu_mask[cpuid]);
2758 }
2759
2760 static void
2761 linux_compat_init(void *arg)
2762 {
2763 struct sysctl_oid *rootoid;
2764 int i;
2765
2766 #if defined(__i386__) || defined(__amd64__)
2767 linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH);
2768 boot_cpu_data.x86_clflush_size = cpu_clflush_line_size;
2769 boot_cpu_data.x86_max_cores = mp_ncpus;
2770 boot_cpu_data.x86 = ((cpu_id & 0xf0000) >> 12) | ((cpu_id & 0xf0) >> 4);
2771 #endif
2772 rw_init(&linux_vma_lock, "lkpi-vma-lock");
2773
2774 rootoid = SYSCTL_ADD_ROOT_NODE(NULL,
2775 OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys");
2776 kobject_init(&linux_class_root, &linux_class_ktype);
2777 kobject_set_name(&linux_class_root, "class");
2778 linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid),
2779 OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class");
2780 kobject_init(&linux_root_device.kobj, &linux_dev_ktype);
2781 kobject_set_name(&linux_root_device.kobj, "device");
2782 linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL,
2783 SYSCTL_CHILDREN(rootoid), OID_AUTO, "device",
2784 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "device");
2785 linux_root_device.bsddev = root_bus;
2786 linux_class_misc.name = "misc";
2787 class_register(&linux_class_misc);
2788 INIT_LIST_HEAD(&pci_drivers);
2789 INIT_LIST_HEAD(&pci_devices);
2790 spin_lock_init(&pci_lock);
2791 mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF);
2792 for (i = 0; i < VMMAP_HASH_SIZE; i++)
2793 LIST_INIT(&vmmaphead[i]);
2794 init_waitqueue_head(&linux_bit_waitq);
2795 init_waitqueue_head(&linux_var_waitq);
2796
2797 CPU_COPY(&all_cpus, &cpu_online_mask);
2798 /*
2799 * Generate a single-CPU cpumask_t for each CPU (possibly) in the system.
2800 * CPUs are indexed from 0..(MAXCPU-1). The entry for cpuid 0 will only
2801 * have itself in the cpumask, cupid 1 only itself on entry 1, and so on.
2802 * This is used by cpumask_of() (and possibly others in the future) for,
2803 * e.g., drivers to pass hints to irq_set_affinity_hint().
2804 */
2805 for (i = 0; i < MAXCPU; i++)
2806 CPU_SET(i, &static_single_cpu_mask[i]);
2807 }
2808 SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL);
2809
2810 static void
2811 linux_compat_uninit(void *arg)
2812 {
2813 linux_kobject_kfree_name(&linux_class_root);
2814 linux_kobject_kfree_name(&linux_root_device.kobj);
2815 linux_kobject_kfree_name(&linux_class_misc.kobj);
2816
2817 mtx_destroy(&vmmaplock);
2818 spin_lock_destroy(&pci_lock);
2819 rw_destroy(&linux_vma_lock);
2820 }
2821 SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL);
2822
2823 /*
2824 * NOTE: Linux frequently uses "unsigned long" for pointer to integer
2825 * conversion and vice versa, where in FreeBSD "uintptr_t" would be
2826 * used. Assert these types have the same size, else some parts of the
2827 * LinuxKPI may not work like expected:
2828 */
2829 CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t));
Cache object: 6edf31dd206787e019d632aebc7de5c5
|