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