linux_compat.c revision 358700
1/*- 2 * Copyright (c) 2010 Isilon Systems, Inc. 3 * Copyright (c) 2010 iX Systems, Inc. 4 * Copyright (c) 2010 Panasas, Inc. 5 * Copyright (c) 2013-2018 Mellanox Technologies, Ltd. 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice unmodified, this list of conditions, and the following 13 * disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30#include <sys/cdefs.h> 31__FBSDID("$FreeBSD: stable/11/sys/compat/linuxkpi/common/src/linux_compat.c 358700 2020-03-06 13:58:08Z hselasky $"); 32 33#include "opt_stack.h" 34 35#include <sys/param.h> 36#include <sys/systm.h> 37#include <sys/malloc.h> 38#include <sys/kernel.h> 39#include <sys/sysctl.h> 40#include <sys/proc.h> 41#include <sys/sglist.h> 42#include <sys/sleepqueue.h> 43#include <sys/refcount.h> 44#include <sys/lock.h> 45#include <sys/mutex.h> 46#include <sys/bus.h> 47#include <sys/fcntl.h> 48#include <sys/file.h> 49#include <sys/filio.h> 50#include <sys/rwlock.h> 51#include <sys/mman.h> 52#include <sys/stack.h> 53#include <sys/user.h> 54 55#include <vm/vm.h> 56#include <vm/pmap.h> 57#include <vm/vm_object.h> 58#include <vm/vm_page.h> 59#include <vm/vm_pager.h> 60 61#include <machine/stdarg.h> 62 63#if defined(__i386__) || defined(__amd64__) 64#include <machine/md_var.h> 65#endif 66 67#include <linux/kobject.h> 68#include <linux/device.h> 69#include <linux/slab.h> 70#include <linux/module.h> 71#include <linux/moduleparam.h> 72#include <linux/cdev.h> 73#include <linux/file.h> 74#include <linux/sysfs.h> 75#include <linux/mm.h> 76#include <linux/io.h> 77#include <linux/vmalloc.h> 78#include <linux/netdevice.h> 79#include <linux/timer.h> 80#include <linux/interrupt.h> 81#include <linux/uaccess.h> 82#include <linux/list.h> 83#include <linux/kthread.h> 84#include <linux/kernel.h> 85#include <linux/compat.h> 86#include <linux/poll.h> 87#include <linux/smp.h> 88 89#if defined(__i386__) || defined(__amd64__) 90#include <asm/smp.h> 91#endif 92 93SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW, 0, "LinuxKPI parameters"); 94 95int linuxkpi_debug; 96SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN, 97 &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable."); 98 99MALLOC_DEFINE(M_KMALLOC, "linux", "Linux kmalloc compat"); 100 101#include <linux/rbtree.h> 102/* Undo Linux compat changes. */ 103#undef RB_ROOT 104#undef file 105#undef cdev 106#define RB_ROOT(head) (head)->rbh_root 107 108static void linux_cdev_deref(struct linux_cdev *ldev); 109static struct vm_area_struct *linux_cdev_handle_find(void *handle); 110 111struct kobject linux_class_root; 112struct device linux_root_device; 113struct class linux_class_misc; 114struct list_head pci_drivers; 115struct list_head pci_devices; 116spinlock_t pci_lock; 117 118unsigned long linux_timer_hz_mask; 119 120int 121panic_cmp(struct rb_node *one, struct rb_node *two) 122{ 123 panic("no cmp"); 124} 125 126RB_GENERATE(linux_root, rb_node, __entry, panic_cmp); 127 128int 129kobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list args) 130{ 131 va_list tmp_va; 132 int len; 133 char *old; 134 char *name; 135 char dummy; 136 137 old = kobj->name; 138 139 if (old && fmt == NULL) 140 return (0); 141 142 /* compute length of string */ 143 va_copy(tmp_va, args); 144 len = vsnprintf(&dummy, 0, fmt, tmp_va); 145 va_end(tmp_va); 146 147 /* account for zero termination */ 148 len++; 149 150 /* check for error */ 151 if (len < 1) 152 return (-EINVAL); 153 154 /* allocate memory for string */ 155 name = kzalloc(len, GFP_KERNEL); 156 if (name == NULL) 157 return (-ENOMEM); 158 vsnprintf(name, len, fmt, args); 159 kobj->name = name; 160 161 /* free old string */ 162 kfree(old); 163 164 /* filter new string */ 165 for (; *name != '\0'; name++) 166 if (*name == '/') 167 *name = '!'; 168 return (0); 169} 170 171int 172kobject_set_name(struct kobject *kobj, const char *fmt, ...) 173{ 174 va_list args; 175 int error; 176 177 va_start(args, fmt); 178 error = kobject_set_name_vargs(kobj, fmt, args); 179 va_end(args); 180 181 return (error); 182} 183 184static int 185kobject_add_complete(struct kobject *kobj, struct kobject *parent) 186{ 187 const struct kobj_type *t; 188 int error; 189 190 kobj->parent = parent; 191 error = sysfs_create_dir(kobj); 192 if (error == 0 && kobj->ktype && kobj->ktype->default_attrs) { 193 struct attribute **attr; 194 t = kobj->ktype; 195 196 for (attr = t->default_attrs; *attr != NULL; attr++) { 197 error = sysfs_create_file(kobj, *attr); 198 if (error) 199 break; 200 } 201 if (error) 202 sysfs_remove_dir(kobj); 203 204 } 205 return (error); 206} 207 208int 209kobject_add(struct kobject *kobj, struct kobject *parent, const char *fmt, ...) 210{ 211 va_list args; 212 int error; 213 214 va_start(args, fmt); 215 error = kobject_set_name_vargs(kobj, fmt, args); 216 va_end(args); 217 if (error) 218 return (error); 219 220 return kobject_add_complete(kobj, parent); 221} 222 223void 224linux_kobject_release(struct kref *kref) 225{ 226 struct kobject *kobj; 227 char *name; 228 229 kobj = container_of(kref, struct kobject, kref); 230 sysfs_remove_dir(kobj); 231 name = kobj->name; 232 if (kobj->ktype && kobj->ktype->release) 233 kobj->ktype->release(kobj); 234 kfree(name); 235} 236 237static void 238linux_kobject_kfree(struct kobject *kobj) 239{ 240 kfree(kobj); 241} 242 243static void 244linux_kobject_kfree_name(struct kobject *kobj) 245{ 246 if (kobj) { 247 kfree(kobj->name); 248 } 249} 250 251const struct kobj_type linux_kfree_type = { 252 .release = linux_kobject_kfree 253}; 254 255static void 256linux_device_release(struct device *dev) 257{ 258 pr_debug("linux_device_release: %s\n", dev_name(dev)); 259 kfree(dev); 260} 261 262static ssize_t 263linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf) 264{ 265 struct class_attribute *dattr; 266 ssize_t error; 267 268 dattr = container_of(attr, struct class_attribute, attr); 269 error = -EIO; 270 if (dattr->show) 271 error = dattr->show(container_of(kobj, struct class, kobj), 272 dattr, buf); 273 return (error); 274} 275 276static ssize_t 277linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf, 278 size_t count) 279{ 280 struct class_attribute *dattr; 281 ssize_t error; 282 283 dattr = container_of(attr, struct class_attribute, attr); 284 error = -EIO; 285 if (dattr->store) 286 error = dattr->store(container_of(kobj, struct class, kobj), 287 dattr, buf, count); 288 return (error); 289} 290 291static void 292linux_class_release(struct kobject *kobj) 293{ 294 struct class *class; 295 296 class = container_of(kobj, struct class, kobj); 297 if (class->class_release) 298 class->class_release(class); 299} 300 301static const struct sysfs_ops linux_class_sysfs = { 302 .show = linux_class_show, 303 .store = linux_class_store, 304}; 305 306const struct kobj_type linux_class_ktype = { 307 .release = linux_class_release, 308 .sysfs_ops = &linux_class_sysfs 309}; 310 311static void 312linux_dev_release(struct kobject *kobj) 313{ 314 struct device *dev; 315 316 dev = container_of(kobj, struct device, kobj); 317 /* This is the precedence defined by linux. */ 318 if (dev->release) 319 dev->release(dev); 320 else if (dev->class && dev->class->dev_release) 321 dev->class->dev_release(dev); 322} 323 324static ssize_t 325linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf) 326{ 327 struct device_attribute *dattr; 328 ssize_t error; 329 330 dattr = container_of(attr, struct device_attribute, attr); 331 error = -EIO; 332 if (dattr->show) 333 error = dattr->show(container_of(kobj, struct device, kobj), 334 dattr, buf); 335 return (error); 336} 337 338static ssize_t 339linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf, 340 size_t count) 341{ 342 struct device_attribute *dattr; 343 ssize_t error; 344 345 dattr = container_of(attr, struct device_attribute, attr); 346 error = -EIO; 347 if (dattr->store) 348 error = dattr->store(container_of(kobj, struct device, kobj), 349 dattr, buf, count); 350 return (error); 351} 352 353static const struct sysfs_ops linux_dev_sysfs = { 354 .show = linux_dev_show, 355 .store = linux_dev_store, 356}; 357 358const struct kobj_type linux_dev_ktype = { 359 .release = linux_dev_release, 360 .sysfs_ops = &linux_dev_sysfs 361}; 362 363struct device * 364device_create(struct class *class, struct device *parent, dev_t devt, 365 void *drvdata, const char *fmt, ...) 366{ 367 struct device *dev; 368 va_list args; 369 370 dev = kzalloc(sizeof(*dev), M_WAITOK); 371 dev->parent = parent; 372 dev->class = class; 373 dev->devt = devt; 374 dev->driver_data = drvdata; 375 dev->release = linux_device_release; 376 va_start(args, fmt); 377 kobject_set_name_vargs(&dev->kobj, fmt, args); 378 va_end(args); 379 device_register(dev); 380 381 return (dev); 382} 383 384int 385kobject_init_and_add(struct kobject *kobj, const struct kobj_type *ktype, 386 struct kobject *parent, const char *fmt, ...) 387{ 388 va_list args; 389 int error; 390 391 kobject_init(kobj, ktype); 392 kobj->ktype = ktype; 393 kobj->parent = parent; 394 kobj->name = NULL; 395 396 va_start(args, fmt); 397 error = kobject_set_name_vargs(kobj, fmt, args); 398 va_end(args); 399 if (error) 400 return (error); 401 return kobject_add_complete(kobj, parent); 402} 403 404static void 405linux_kq_lock(void *arg) 406{ 407 spinlock_t *s = arg; 408 409 spin_lock(s); 410} 411static void 412linux_kq_unlock(void *arg) 413{ 414 spinlock_t *s = arg; 415 416 spin_unlock(s); 417} 418 419static void 420linux_kq_lock_owned(void *arg) 421{ 422#ifdef INVARIANTS 423 spinlock_t *s = arg; 424 425 mtx_assert(&s->m, MA_OWNED); 426#endif 427} 428 429static void 430linux_kq_lock_unowned(void *arg) 431{ 432#ifdef INVARIANTS 433 spinlock_t *s = arg; 434 435 mtx_assert(&s->m, MA_NOTOWNED); 436#endif 437} 438 439static void 440linux_file_kqfilter_poll(struct linux_file *, int); 441 442struct linux_file * 443linux_file_alloc(void) 444{ 445 struct linux_file *filp; 446 447 filp = kzalloc(sizeof(*filp), GFP_KERNEL); 448 449 /* set initial refcount */ 450 filp->f_count = 1; 451 452 /* setup fields needed by kqueue support */ 453 spin_lock_init(&filp->f_kqlock); 454 knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock, 455 linux_kq_lock, linux_kq_unlock, 456 linux_kq_lock_owned, linux_kq_lock_unowned); 457 458 return (filp); 459} 460 461void 462linux_file_free(struct linux_file *filp) 463{ 464 if (filp->_file == NULL) { 465 if (filp->f_shmem != NULL) 466 vm_object_deallocate(filp->f_shmem); 467 kfree(filp); 468 } else { 469 /* 470 * The close method of the character device or file 471 * will free the linux_file structure: 472 */ 473 _fdrop(filp->_file, curthread); 474 } 475} 476 477static int 478linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot, 479 vm_page_t *mres) 480{ 481 struct vm_area_struct *vmap; 482 483 vmap = linux_cdev_handle_find(vm_obj->handle); 484 485 MPASS(vmap != NULL); 486 MPASS(vmap->vm_private_data == vm_obj->handle); 487 488 if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) { 489 vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset; 490 vm_page_t page; 491 492 if (((*mres)->flags & PG_FICTITIOUS) != 0) { 493 /* 494 * If the passed in result page is a fake 495 * page, update it with the new physical 496 * address. 497 */ 498 page = *mres; 499 vm_page_updatefake(page, paddr, vm_obj->memattr); 500 } else { 501 /* 502 * Replace the passed in "mres" page with our 503 * own fake page and free up the all of the 504 * original pages. 505 */ 506 VM_OBJECT_WUNLOCK(vm_obj); 507 page = vm_page_getfake(paddr, vm_obj->memattr); 508 VM_OBJECT_WLOCK(vm_obj); 509 510 vm_page_replace_checked(page, vm_obj, 511 (*mres)->pindex, *mres); 512 513 vm_page_lock(*mres); 514 vm_page_free(*mres); 515 vm_page_unlock(*mres); 516 *mres = page; 517 } 518 page->valid = VM_PAGE_BITS_ALL; 519 return (VM_PAGER_OK); 520 } 521 return (VM_PAGER_FAIL); 522} 523 524static int 525linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type, 526 vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last) 527{ 528 struct vm_area_struct *vmap; 529 int err; 530 531 linux_set_current(curthread); 532 533 /* get VM area structure */ 534 vmap = linux_cdev_handle_find(vm_obj->handle); 535 MPASS(vmap != NULL); 536 MPASS(vmap->vm_private_data == vm_obj->handle); 537 538 VM_OBJECT_WUNLOCK(vm_obj); 539 540 down_write(&vmap->vm_mm->mmap_sem); 541 if (unlikely(vmap->vm_ops == NULL)) { 542 err = VM_FAULT_SIGBUS; 543 } else { 544 struct vm_fault vmf; 545 546 /* fill out VM fault structure */ 547 vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx); 548 vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0; 549 vmf.pgoff = 0; 550 vmf.page = NULL; 551 vmf.vma = vmap; 552 553 vmap->vm_pfn_count = 0; 554 vmap->vm_pfn_pcount = &vmap->vm_pfn_count; 555 vmap->vm_obj = vm_obj; 556 557 err = vmap->vm_ops->fault(vmap, &vmf); 558 559 while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) { 560 kern_yield(PRI_USER); 561 err = vmap->vm_ops->fault(vmap, &vmf); 562 } 563 } 564 565 /* translate return code */ 566 switch (err) { 567 case VM_FAULT_OOM: 568 err = VM_PAGER_AGAIN; 569 break; 570 case VM_FAULT_SIGBUS: 571 err = VM_PAGER_BAD; 572 break; 573 case VM_FAULT_NOPAGE: 574 /* 575 * By contract the fault handler will return having 576 * busied all the pages itself. If pidx is already 577 * found in the object, it will simply xbusy the first 578 * page and return with vm_pfn_count set to 1. 579 */ 580 *first = vmap->vm_pfn_first; 581 *last = *first + vmap->vm_pfn_count - 1; 582 err = VM_PAGER_OK; 583 break; 584 default: 585 err = VM_PAGER_ERROR; 586 break; 587 } 588 up_write(&vmap->vm_mm->mmap_sem); 589 VM_OBJECT_WLOCK(vm_obj); 590 return (err); 591} 592 593static struct rwlock linux_vma_lock; 594static TAILQ_HEAD(, vm_area_struct) linux_vma_head = 595 TAILQ_HEAD_INITIALIZER(linux_vma_head); 596 597static void 598linux_cdev_handle_free(struct vm_area_struct *vmap) 599{ 600 /* Drop reference on vm_file */ 601 if (vmap->vm_file != NULL) 602 fput(vmap->vm_file); 603 604 /* Drop reference on mm_struct */ 605 mmput(vmap->vm_mm); 606 607 kfree(vmap); 608} 609 610static void 611linux_cdev_handle_remove(struct vm_area_struct *vmap) 612{ 613 rw_wlock(&linux_vma_lock); 614 TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry); 615 rw_wunlock(&linux_vma_lock); 616} 617 618static struct vm_area_struct * 619linux_cdev_handle_find(void *handle) 620{ 621 struct vm_area_struct *vmap; 622 623 rw_rlock(&linux_vma_lock); 624 TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) { 625 if (vmap->vm_private_data == handle) 626 break; 627 } 628 rw_runlock(&linux_vma_lock); 629 return (vmap); 630} 631 632static int 633linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, 634 vm_ooffset_t foff, struct ucred *cred, u_short *color) 635{ 636 637 MPASS(linux_cdev_handle_find(handle) != NULL); 638 *color = 0; 639 return (0); 640} 641 642static void 643linux_cdev_pager_dtor(void *handle) 644{ 645 const struct vm_operations_struct *vm_ops; 646 struct vm_area_struct *vmap; 647 648 vmap = linux_cdev_handle_find(handle); 649 MPASS(vmap != NULL); 650 651 /* 652 * Remove handle before calling close operation to prevent 653 * other threads from reusing the handle pointer. 654 */ 655 linux_cdev_handle_remove(vmap); 656 657 down_write(&vmap->vm_mm->mmap_sem); 658 vm_ops = vmap->vm_ops; 659 if (likely(vm_ops != NULL)) 660 vm_ops->close(vmap); 661 up_write(&vmap->vm_mm->mmap_sem); 662 663 linux_cdev_handle_free(vmap); 664} 665 666static struct cdev_pager_ops linux_cdev_pager_ops[2] = { 667 { 668 /* OBJT_MGTDEVICE */ 669 .cdev_pg_populate = linux_cdev_pager_populate, 670 .cdev_pg_ctor = linux_cdev_pager_ctor, 671 .cdev_pg_dtor = linux_cdev_pager_dtor 672 }, 673 { 674 /* OBJT_DEVICE */ 675 .cdev_pg_fault = linux_cdev_pager_fault, 676 .cdev_pg_ctor = linux_cdev_pager_ctor, 677 .cdev_pg_dtor = linux_cdev_pager_dtor 678 }, 679}; 680 681int 682zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 683 unsigned long size) 684{ 685 vm_object_t obj; 686 vm_page_t m; 687 688 obj = vma->vm_obj; 689 if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0) 690 return (-ENOTSUP); 691 VM_OBJECT_RLOCK(obj); 692 for (m = vm_page_find_least(obj, OFF_TO_IDX(address)); 693 m != NULL && m->pindex < OFF_TO_IDX(address + size); 694 m = TAILQ_NEXT(m, listq)) 695 pmap_remove_all(m); 696 VM_OBJECT_RUNLOCK(obj); 697 return (0); 698} 699 700static struct file_operations dummy_ldev_ops = { 701 /* XXXKIB */ 702}; 703 704static struct linux_cdev dummy_ldev = { 705 .ops = &dummy_ldev_ops, 706}; 707 708#define LDEV_SI_DTR 0x0001 709#define LDEV_SI_REF 0x0002 710 711static void 712linux_get_fop(struct linux_file *filp, const struct file_operations **fop, 713 struct linux_cdev **dev) 714{ 715 struct linux_cdev *ldev; 716 u_int siref; 717 718 ldev = filp->f_cdev; 719 *fop = filp->f_op; 720 if (ldev != NULL) { 721 for (siref = ldev->siref;;) { 722 if ((siref & LDEV_SI_DTR) != 0) { 723 ldev = &dummy_ldev; 724 siref = ldev->siref; 725 *fop = ldev->ops; 726 MPASS((ldev->siref & LDEV_SI_DTR) == 0); 727 } else if (atomic_fcmpset_int(&ldev->siref, &siref, 728 siref + LDEV_SI_REF)) { 729 break; 730 } 731 } 732 } 733 *dev = ldev; 734} 735 736static void 737linux_drop_fop(struct linux_cdev *ldev) 738{ 739 740 if (ldev == NULL) 741 return; 742 MPASS((ldev->siref & ~LDEV_SI_DTR) != 0); 743 atomic_subtract_int(&ldev->siref, LDEV_SI_REF); 744} 745 746#define OPW(fp,td,code) ({ \ 747 struct file *__fpop; \ 748 __typeof(code) __retval; \ 749 \ 750 __fpop = (td)->td_fpop; \ 751 (td)->td_fpop = (fp); \ 752 __retval = (code); \ 753 (td)->td_fpop = __fpop; \ 754 __retval; \ 755}) 756 757static int 758linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td, 759 struct file *file) 760{ 761 struct linux_cdev *ldev; 762 struct linux_file *filp; 763 const struct file_operations *fop; 764 int error; 765 766 ldev = dev->si_drv1; 767 768 filp = linux_file_alloc(); 769 filp->f_dentry = &filp->f_dentry_store; 770 filp->f_op = ldev->ops; 771 filp->f_mode = file->f_flag; 772 filp->f_flags = file->f_flag; 773 filp->f_vnode = file->f_vnode; 774 filp->_file = file; 775 refcount_acquire(&ldev->refs); 776 filp->f_cdev = ldev; 777 778 linux_set_current(td); 779 linux_get_fop(filp, &fop, &ldev); 780 781 if (fop->open != NULL) { 782 error = -fop->open(file->f_vnode, filp); 783 if (error != 0) { 784 linux_drop_fop(ldev); 785 linux_cdev_deref(filp->f_cdev); 786 kfree(filp); 787 return (error); 788 } 789 } 790 791 /* hold on to the vnode - used for fstat() */ 792 vhold(filp->f_vnode); 793 794 /* release the file from devfs */ 795 finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops); 796 linux_drop_fop(ldev); 797 return (ENXIO); 798} 799 800#define LINUX_IOCTL_MIN_PTR 0x10000UL 801#define LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX) 802 803static inline int 804linux_remap_address(void **uaddr, size_t len) 805{ 806 uintptr_t uaddr_val = (uintptr_t)(*uaddr); 807 808 if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR && 809 uaddr_val < LINUX_IOCTL_MAX_PTR)) { 810 struct task_struct *pts = current; 811 if (pts == NULL) { 812 *uaddr = NULL; 813 return (1); 814 } 815 816 /* compute data offset */ 817 uaddr_val -= LINUX_IOCTL_MIN_PTR; 818 819 /* check that length is within bounds */ 820 if ((len > IOCPARM_MAX) || 821 (uaddr_val + len) > pts->bsd_ioctl_len) { 822 *uaddr = NULL; 823 return (1); 824 } 825 826 /* re-add kernel buffer address */ 827 uaddr_val += (uintptr_t)pts->bsd_ioctl_data; 828 829 /* update address location */ 830 *uaddr = (void *)uaddr_val; 831 return (1); 832 } 833 return (0); 834} 835 836int 837linux_copyin(const void *uaddr, void *kaddr, size_t len) 838{ 839 if (linux_remap_address(__DECONST(void **, &uaddr), len)) { 840 if (uaddr == NULL) 841 return (-EFAULT); 842 memcpy(kaddr, uaddr, len); 843 return (0); 844 } 845 return (-copyin(uaddr, kaddr, len)); 846} 847 848int 849linux_copyout(const void *kaddr, void *uaddr, size_t len) 850{ 851 if (linux_remap_address(&uaddr, len)) { 852 if (uaddr == NULL) 853 return (-EFAULT); 854 memcpy(uaddr, kaddr, len); 855 return (0); 856 } 857 return (-copyout(kaddr, uaddr, len)); 858} 859 860size_t 861linux_clear_user(void *_uaddr, size_t _len) 862{ 863 uint8_t *uaddr = _uaddr; 864 size_t len = _len; 865 866 /* make sure uaddr is aligned before going into the fast loop */ 867 while (((uintptr_t)uaddr & 7) != 0 && len > 7) { 868 if (subyte(uaddr, 0)) 869 return (_len); 870 uaddr++; 871 len--; 872 } 873 874 /* zero 8 bytes at a time */ 875 while (len > 7) { 876#ifdef __LP64__ 877 if (suword64(uaddr, 0)) 878 return (_len); 879#else 880 if (suword32(uaddr, 0)) 881 return (_len); 882 if (suword32(uaddr + 4, 0)) 883 return (_len); 884#endif 885 uaddr += 8; 886 len -= 8; 887 } 888 889 /* zero fill end, if any */ 890 while (len > 0) { 891 if (subyte(uaddr, 0)) 892 return (_len); 893 uaddr++; 894 len--; 895 } 896 return (0); 897} 898 899int 900linux_access_ok(int rw, const void *uaddr, size_t len) 901{ 902 uintptr_t saddr; 903 uintptr_t eaddr; 904 905 /* get start and end address */ 906 saddr = (uintptr_t)uaddr; 907 eaddr = (uintptr_t)uaddr + len; 908 909 /* verify addresses are valid for userspace */ 910 return ((saddr == eaddr) || 911 (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS)); 912} 913 914/* 915 * This function should return either EINTR or ERESTART depending on 916 * the signal type sent to this thread: 917 */ 918static int 919linux_get_error(struct task_struct *task, int error) 920{ 921 /* check for signal type interrupt code */ 922 if (error == EINTR || error == ERESTARTSYS || error == ERESTART) { 923 error = -linux_schedule_get_interrupt_value(task); 924 if (error == 0) 925 error = EINTR; 926 } 927 return (error); 928} 929 930static int 931linux_file_ioctl_sub(struct file *fp, struct linux_file *filp, 932 const struct file_operations *fop, u_long cmd, caddr_t data, 933 struct thread *td) 934{ 935 struct task_struct *task = current; 936 unsigned size; 937 int error; 938 939 size = IOCPARM_LEN(cmd); 940 /* refer to logic in sys_ioctl() */ 941 if (size > 0) { 942 /* 943 * Setup hint for linux_copyin() and linux_copyout(). 944 * 945 * Background: Linux code expects a user-space address 946 * while FreeBSD supplies a kernel-space address. 947 */ 948 task->bsd_ioctl_data = data; 949 task->bsd_ioctl_len = size; 950 data = (void *)LINUX_IOCTL_MIN_PTR; 951 } else { 952 /* fetch user-space pointer */ 953 data = *(void **)data; 954 } 955#if defined(__amd64__) 956 if (td->td_proc->p_elf_machine == EM_386) { 957 /* try the compat IOCTL handler first */ 958 if (fop->compat_ioctl != NULL) { 959 error = -OPW(fp, td, fop->compat_ioctl(filp, 960 cmd, (u_long)data)); 961 } else { 962 error = ENOTTY; 963 } 964 965 /* fallback to the regular IOCTL handler, if any */ 966 if (error == ENOTTY && fop->unlocked_ioctl != NULL) { 967 error = -OPW(fp, td, fop->unlocked_ioctl(filp, 968 cmd, (u_long)data)); 969 } 970 } else 971#endif 972 { 973 if (fop->unlocked_ioctl != NULL) { 974 error = -OPW(fp, td, fop->unlocked_ioctl(filp, 975 cmd, (u_long)data)); 976 } else { 977 error = ENOTTY; 978 } 979 } 980 if (size > 0) { 981 task->bsd_ioctl_data = NULL; 982 task->bsd_ioctl_len = 0; 983 } 984 985 if (error == EWOULDBLOCK) { 986 /* update kqfilter status, if any */ 987 linux_file_kqfilter_poll(filp, 988 LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE); 989 } else { 990 error = linux_get_error(task, error); 991 } 992 return (error); 993} 994 995#define LINUX_POLL_TABLE_NORMAL ((poll_table *)1) 996 997/* 998 * This function atomically updates the poll wakeup state and returns 999 * the previous state at the time of update. 1000 */ 1001static uint8_t 1002linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate) 1003{ 1004 int c, old; 1005 1006 c = v->counter; 1007 1008 while ((old = atomic_cmpxchg(v, c, pstate[c])) != c) 1009 c = old; 1010 1011 return (c); 1012} 1013 1014 1015static int 1016linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key) 1017{ 1018 static const uint8_t state[LINUX_FWQ_STATE_MAX] = { 1019 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */ 1020 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */ 1021 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY, 1022 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */ 1023 }; 1024 struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq); 1025 1026 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { 1027 case LINUX_FWQ_STATE_QUEUED: 1028 linux_poll_wakeup(filp); 1029 return (1); 1030 default: 1031 return (0); 1032 } 1033} 1034 1035void 1036linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p) 1037{ 1038 static const uint8_t state[LINUX_FWQ_STATE_MAX] = { 1039 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY, 1040 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */ 1041 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */ 1042 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED, 1043 }; 1044 1045 /* check if we are called inside the select system call */ 1046 if (p == LINUX_POLL_TABLE_NORMAL) 1047 selrecord(curthread, &filp->f_selinfo); 1048 1049 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { 1050 case LINUX_FWQ_STATE_INIT: 1051 /* NOTE: file handles can only belong to one wait-queue */ 1052 filp->f_wait_queue.wqh = wqh; 1053 filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback; 1054 add_wait_queue(wqh, &filp->f_wait_queue.wq); 1055 atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED); 1056 break; 1057 default: 1058 break; 1059 } 1060} 1061 1062static void 1063linux_poll_wait_dequeue(struct linux_file *filp) 1064{ 1065 static const uint8_t state[LINUX_FWQ_STATE_MAX] = { 1066 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */ 1067 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT, 1068 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT, 1069 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT, 1070 }; 1071 1072 seldrain(&filp->f_selinfo); 1073 1074 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { 1075 case LINUX_FWQ_STATE_NOT_READY: 1076 case LINUX_FWQ_STATE_QUEUED: 1077 case LINUX_FWQ_STATE_READY: 1078 remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq); 1079 break; 1080 default: 1081 break; 1082 } 1083} 1084 1085void 1086linux_poll_wakeup(struct linux_file *filp) 1087{ 1088 /* this function should be NULL-safe */ 1089 if (filp == NULL) 1090 return; 1091 1092 selwakeup(&filp->f_selinfo); 1093 1094 spin_lock(&filp->f_kqlock); 1095 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ | 1096 LINUX_KQ_FLAG_NEED_WRITE; 1097 1098 /* make sure the "knote" gets woken up */ 1099 KNOTE_LOCKED(&filp->f_selinfo.si_note, 1); 1100 spin_unlock(&filp->f_kqlock); 1101} 1102 1103static void 1104linux_file_kqfilter_detach(struct knote *kn) 1105{ 1106 struct linux_file *filp = kn->kn_hook; 1107 1108 spin_lock(&filp->f_kqlock); 1109 knlist_remove(&filp->f_selinfo.si_note, kn, 1); 1110 spin_unlock(&filp->f_kqlock); 1111} 1112 1113static int 1114linux_file_kqfilter_read_event(struct knote *kn, long hint) 1115{ 1116 struct linux_file *filp = kn->kn_hook; 1117 1118 mtx_assert(&filp->f_kqlock.m, MA_OWNED); 1119 1120 return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0); 1121} 1122 1123static int 1124linux_file_kqfilter_write_event(struct knote *kn, long hint) 1125{ 1126 struct linux_file *filp = kn->kn_hook; 1127 1128 mtx_assert(&filp->f_kqlock.m, MA_OWNED); 1129 1130 return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0); 1131} 1132 1133static struct filterops linux_dev_kqfiltops_read = { 1134 .f_isfd = 1, 1135 .f_detach = linux_file_kqfilter_detach, 1136 .f_event = linux_file_kqfilter_read_event, 1137}; 1138 1139static struct filterops linux_dev_kqfiltops_write = { 1140 .f_isfd = 1, 1141 .f_detach = linux_file_kqfilter_detach, 1142 .f_event = linux_file_kqfilter_write_event, 1143}; 1144 1145static void 1146linux_file_kqfilter_poll(struct linux_file *filp, int kqflags) 1147{ 1148 struct thread *td; 1149 const struct file_operations *fop; 1150 struct linux_cdev *ldev; 1151 int temp; 1152 1153 if ((filp->f_kqflags & kqflags) == 0) 1154 return; 1155 1156 td = curthread; 1157 1158 linux_get_fop(filp, &fop, &ldev); 1159 /* get the latest polling state */ 1160 temp = OPW(filp->_file, td, fop->poll(filp, NULL)); 1161 linux_drop_fop(ldev); 1162 1163 spin_lock(&filp->f_kqlock); 1164 /* clear kqflags */ 1165 filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ | 1166 LINUX_KQ_FLAG_NEED_WRITE); 1167 /* update kqflags */ 1168 if ((temp & (POLLIN | POLLOUT)) != 0) { 1169 if ((temp & POLLIN) != 0) 1170 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ; 1171 if ((temp & POLLOUT) != 0) 1172 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE; 1173 1174 /* make sure the "knote" gets woken up */ 1175 KNOTE_LOCKED(&filp->f_selinfo.si_note, 0); 1176 } 1177 spin_unlock(&filp->f_kqlock); 1178} 1179 1180static int 1181linux_file_kqfilter(struct file *file, struct knote *kn) 1182{ 1183 struct linux_file *filp; 1184 struct thread *td; 1185 int error; 1186 1187 td = curthread; 1188 filp = (struct linux_file *)file->f_data; 1189 filp->f_flags = file->f_flag; 1190 if (filp->f_op->poll == NULL) 1191 return (EINVAL); 1192 1193 spin_lock(&filp->f_kqlock); 1194 switch (kn->kn_filter) { 1195 case EVFILT_READ: 1196 filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ; 1197 kn->kn_fop = &linux_dev_kqfiltops_read; 1198 kn->kn_hook = filp; 1199 knlist_add(&filp->f_selinfo.si_note, kn, 1); 1200 error = 0; 1201 break; 1202 case EVFILT_WRITE: 1203 filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE; 1204 kn->kn_fop = &linux_dev_kqfiltops_write; 1205 kn->kn_hook = filp; 1206 knlist_add(&filp->f_selinfo.si_note, kn, 1); 1207 error = 0; 1208 break; 1209 default: 1210 error = EINVAL; 1211 break; 1212 } 1213 spin_unlock(&filp->f_kqlock); 1214 1215 if (error == 0) { 1216 linux_set_current(td); 1217 1218 /* update kqfilter status, if any */ 1219 linux_file_kqfilter_poll(filp, 1220 LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE); 1221 } 1222 return (error); 1223} 1224 1225static int 1226linux_file_mmap_single(struct file *fp, const struct file_operations *fop, 1227 vm_ooffset_t *offset, vm_size_t size, struct vm_object **object, 1228 int nprot, struct thread *td) 1229{ 1230 struct task_struct *task; 1231 struct vm_area_struct *vmap; 1232 struct mm_struct *mm; 1233 struct linux_file *filp; 1234 vm_memattr_t attr; 1235 int error; 1236 1237 filp = (struct linux_file *)fp->f_data; 1238 filp->f_flags = fp->f_flag; 1239 1240 if (fop->mmap == NULL) 1241 return (EOPNOTSUPP); 1242 1243 linux_set_current(td); 1244 1245 /* 1246 * The same VM object might be shared by multiple processes 1247 * and the mm_struct is usually freed when a process exits. 1248 * 1249 * The atomic reference below makes sure the mm_struct is 1250 * available as long as the vmap is in the linux_vma_head. 1251 */ 1252 task = current; 1253 mm = task->mm; 1254 if (atomic_inc_not_zero(&mm->mm_users) == 0) 1255 return (EINVAL); 1256 1257 vmap = kzalloc(sizeof(*vmap), GFP_KERNEL); 1258 vmap->vm_start = 0; 1259 vmap->vm_end = size; 1260 vmap->vm_pgoff = *offset / PAGE_SIZE; 1261 vmap->vm_pfn = 0; 1262 vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL); 1263 vmap->vm_ops = NULL; 1264 vmap->vm_file = get_file(filp); 1265 vmap->vm_mm = mm; 1266 1267 if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) { 1268 error = linux_get_error(task, EINTR); 1269 } else { 1270 error = -OPW(fp, td, fop->mmap(filp, vmap)); 1271 error = linux_get_error(task, error); 1272 up_write(&vmap->vm_mm->mmap_sem); 1273 } 1274 1275 if (error != 0) { 1276 linux_cdev_handle_free(vmap); 1277 return (error); 1278 } 1279 1280 attr = pgprot2cachemode(vmap->vm_page_prot); 1281 1282 if (vmap->vm_ops != NULL) { 1283 struct vm_area_struct *ptr; 1284 void *vm_private_data; 1285 bool vm_no_fault; 1286 1287 if (vmap->vm_ops->open == NULL || 1288 vmap->vm_ops->close == NULL || 1289 vmap->vm_private_data == NULL) { 1290 /* free allocated VM area struct */ 1291 linux_cdev_handle_free(vmap); 1292 return (EINVAL); 1293 } 1294 1295 vm_private_data = vmap->vm_private_data; 1296 1297 rw_wlock(&linux_vma_lock); 1298 TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) { 1299 if (ptr->vm_private_data == vm_private_data) 1300 break; 1301 } 1302 /* check if there is an existing VM area struct */ 1303 if (ptr != NULL) { 1304 /* check if the VM area structure is invalid */ 1305 if (ptr->vm_ops == NULL || 1306 ptr->vm_ops->open == NULL || 1307 ptr->vm_ops->close == NULL) { 1308 error = ESTALE; 1309 vm_no_fault = 1; 1310 } else { 1311 error = EEXIST; 1312 vm_no_fault = (ptr->vm_ops->fault == NULL); 1313 } 1314 } else { 1315 /* insert VM area structure into list */ 1316 TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry); 1317 error = 0; 1318 vm_no_fault = (vmap->vm_ops->fault == NULL); 1319 } 1320 rw_wunlock(&linux_vma_lock); 1321 1322 if (error != 0) { 1323 /* free allocated VM area struct */ 1324 linux_cdev_handle_free(vmap); 1325 /* check for stale VM area struct */ 1326 if (error != EEXIST) 1327 return (error); 1328 } 1329 1330 /* check if there is no fault handler */ 1331 if (vm_no_fault) { 1332 *object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE, 1333 &linux_cdev_pager_ops[1], size, nprot, *offset, 1334 td->td_ucred); 1335 } else { 1336 *object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE, 1337 &linux_cdev_pager_ops[0], size, nprot, *offset, 1338 td->td_ucred); 1339 } 1340 1341 /* check if allocating the VM object failed */ 1342 if (*object == NULL) { 1343 if (error == 0) { 1344 /* remove VM area struct from list */ 1345 linux_cdev_handle_remove(vmap); 1346 /* free allocated VM area struct */ 1347 linux_cdev_handle_free(vmap); 1348 } 1349 return (EINVAL); 1350 } 1351 } else { 1352 struct sglist *sg; 1353 1354 sg = sglist_alloc(1, M_WAITOK); 1355 sglist_append_phys(sg, 1356 (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len); 1357 1358 *object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len, 1359 nprot, 0, td->td_ucred); 1360 1361 linux_cdev_handle_free(vmap); 1362 1363 if (*object == NULL) { 1364 sglist_free(sg); 1365 return (EINVAL); 1366 } 1367 } 1368 1369 if (attr != VM_MEMATTR_DEFAULT) { 1370 VM_OBJECT_WLOCK(*object); 1371 vm_object_set_memattr(*object, attr); 1372 VM_OBJECT_WUNLOCK(*object); 1373 } 1374 *offset = 0; 1375 return (0); 1376} 1377 1378struct cdevsw linuxcdevsw = { 1379 .d_version = D_VERSION, 1380 .d_fdopen = linux_dev_fdopen, 1381 .d_name = "lkpidev", 1382}; 1383 1384static int 1385linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred, 1386 int flags, struct thread *td) 1387{ 1388 struct linux_file *filp; 1389 const struct file_operations *fop; 1390 struct linux_cdev *ldev; 1391 ssize_t bytes; 1392 int error; 1393 1394 error = 0; 1395 filp = (struct linux_file *)file->f_data; 1396 filp->f_flags = file->f_flag; 1397 /* XXX no support for I/O vectors currently */ 1398 if (uio->uio_iovcnt != 1) 1399 return (EOPNOTSUPP); 1400 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1401 return (EINVAL); 1402 linux_set_current(td); 1403 linux_get_fop(filp, &fop, &ldev); 1404 if (fop->read != NULL) { 1405 bytes = OPW(file, td, fop->read(filp, 1406 uio->uio_iov->iov_base, 1407 uio->uio_iov->iov_len, &uio->uio_offset)); 1408 if (bytes >= 0) { 1409 uio->uio_iov->iov_base = 1410 ((uint8_t *)uio->uio_iov->iov_base) + bytes; 1411 uio->uio_iov->iov_len -= bytes; 1412 uio->uio_resid -= bytes; 1413 } else { 1414 error = linux_get_error(current, -bytes); 1415 } 1416 } else 1417 error = ENXIO; 1418 1419 /* update kqfilter status, if any */ 1420 linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ); 1421 linux_drop_fop(ldev); 1422 1423 return (error); 1424} 1425 1426static int 1427linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred, 1428 int flags, struct thread *td) 1429{ 1430 struct linux_file *filp; 1431 const struct file_operations *fop; 1432 struct linux_cdev *ldev; 1433 ssize_t bytes; 1434 int error; 1435 1436 filp = (struct linux_file *)file->f_data; 1437 filp->f_flags = file->f_flag; 1438 /* XXX no support for I/O vectors currently */ 1439 if (uio->uio_iovcnt != 1) 1440 return (EOPNOTSUPP); 1441 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1442 return (EINVAL); 1443 linux_set_current(td); 1444 linux_get_fop(filp, &fop, &ldev); 1445 if (fop->write != NULL) { 1446 bytes = OPW(file, td, fop->write(filp, 1447 uio->uio_iov->iov_base, 1448 uio->uio_iov->iov_len, &uio->uio_offset)); 1449 if (bytes >= 0) { 1450 uio->uio_iov->iov_base = 1451 ((uint8_t *)uio->uio_iov->iov_base) + bytes; 1452 uio->uio_iov->iov_len -= bytes; 1453 uio->uio_resid -= bytes; 1454 error = 0; 1455 } else { 1456 error = linux_get_error(current, -bytes); 1457 } 1458 } else 1459 error = ENXIO; 1460 1461 /* update kqfilter status, if any */ 1462 linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE); 1463 1464 linux_drop_fop(ldev); 1465 1466 return (error); 1467} 1468 1469static int 1470linux_file_poll(struct file *file, int events, struct ucred *active_cred, 1471 struct thread *td) 1472{ 1473 struct linux_file *filp; 1474 const struct file_operations *fop; 1475 struct linux_cdev *ldev; 1476 int revents; 1477 1478 filp = (struct linux_file *)file->f_data; 1479 filp->f_flags = file->f_flag; 1480 linux_set_current(td); 1481 linux_get_fop(filp, &fop, &ldev); 1482 if (fop->poll != NULL) { 1483 revents = OPW(file, td, fop->poll(filp, 1484 LINUX_POLL_TABLE_NORMAL)) & events; 1485 } else { 1486 revents = 0; 1487 } 1488 linux_drop_fop(ldev); 1489 return (revents); 1490} 1491 1492static int 1493linux_file_close(struct file *file, struct thread *td) 1494{ 1495 struct linux_file *filp; 1496 const struct file_operations *fop; 1497 struct linux_cdev *ldev; 1498 int error; 1499 1500 filp = (struct linux_file *)file->f_data; 1501 1502 KASSERT(file_count(filp) == 0, 1503 ("File refcount(%d) is not zero", file_count(filp))); 1504 1505 if (td == NULL) 1506 td = curthread; 1507 1508 error = 0; 1509 filp->f_flags = file->f_flag; 1510 linux_set_current(td); 1511 linux_poll_wait_dequeue(filp); 1512 linux_get_fop(filp, &fop, &ldev); 1513 if (fop->release != NULL) 1514 error = -OPW(file, td, fop->release(filp->f_vnode, filp)); 1515 funsetown(&filp->f_sigio); 1516 if (filp->f_vnode != NULL) 1517 vdrop(filp->f_vnode); 1518 linux_drop_fop(ldev); 1519 if (filp->f_cdev != NULL) 1520 linux_cdev_deref(filp->f_cdev); 1521 kfree(filp); 1522 1523 return (error); 1524} 1525 1526static int 1527linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred, 1528 struct thread *td) 1529{ 1530 struct linux_file *filp; 1531 const struct file_operations *fop; 1532 struct linux_cdev *ldev; 1533 struct fiodgname_arg *fgn; 1534 const char *p; 1535 int error, i; 1536 1537 error = 0; 1538 filp = (struct linux_file *)fp->f_data; 1539 filp->f_flags = fp->f_flag; 1540 linux_get_fop(filp, &fop, &ldev); 1541 1542 linux_set_current(td); 1543 switch (cmd) { 1544 case FIONBIO: 1545 break; 1546 case FIOASYNC: 1547 if (fop->fasync == NULL) 1548 break; 1549 error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC)); 1550 break; 1551 case FIOSETOWN: 1552 error = fsetown(*(int *)data, &filp->f_sigio); 1553 if (error == 0) { 1554 if (fop->fasync == NULL) 1555 break; 1556 error = -OPW(fp, td, fop->fasync(0, filp, 1557 fp->f_flag & FASYNC)); 1558 } 1559 break; 1560 case FIOGETOWN: 1561 *(int *)data = fgetown(&filp->f_sigio); 1562 break; 1563 case FIODGNAME: 1564 if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) { 1565 error = ENXIO; 1566 break; 1567 } 1568 fgn = data; 1569 p = devtoname(filp->f_cdev->cdev); 1570 i = strlen(p) + 1; 1571 if (i > fgn->len) { 1572 error = EINVAL; 1573 break; 1574 } 1575 error = copyout(p, fgn->buf, i); 1576 break; 1577 default: 1578 error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td); 1579 break; 1580 } 1581 linux_drop_fop(ldev); 1582 return (error); 1583} 1584 1585static int 1586linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot, 1587 vm_prot_t *maxprotp, int *flagsp, struct file *fp, 1588 vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp) 1589{ 1590 /* 1591 * Character devices do not provide private mappings 1592 * of any kind: 1593 */ 1594 if ((*maxprotp & VM_PROT_WRITE) == 0 && 1595 (prot & VM_PROT_WRITE) != 0) 1596 return (EACCES); 1597 if ((*flagsp & (MAP_PRIVATE | MAP_COPY)) != 0) 1598 return (EINVAL); 1599 1600 return (linux_file_mmap_single(fp, fop, foff, objsize, objp, 1601 (int)prot, td)); 1602} 1603 1604static int 1605linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 1606 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 1607 struct thread *td) 1608{ 1609 struct linux_file *filp; 1610 const struct file_operations *fop; 1611 struct linux_cdev *ldev; 1612 struct mount *mp; 1613 struct vnode *vp; 1614 vm_object_t object; 1615 vm_prot_t maxprot; 1616 int error; 1617 1618 filp = (struct linux_file *)fp->f_data; 1619 1620 vp = filp->f_vnode; 1621 if (vp == NULL) 1622 return (EOPNOTSUPP); 1623 1624 /* 1625 * Ensure that file and memory protections are 1626 * compatible. 1627 */ 1628 mp = vp->v_mount; 1629 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { 1630 maxprot = VM_PROT_NONE; 1631 if ((prot & VM_PROT_EXECUTE) != 0) 1632 return (EACCES); 1633 } else 1634 maxprot = VM_PROT_EXECUTE; 1635 if ((fp->f_flag & FREAD) != 0) 1636 maxprot |= VM_PROT_READ; 1637 else if ((prot & VM_PROT_READ) != 0) 1638 return (EACCES); 1639 1640 /* 1641 * If we are sharing potential changes via MAP_SHARED and we 1642 * are trying to get write permission although we opened it 1643 * without asking for it, bail out. 1644 * 1645 * Note that most character devices always share mappings. 1646 * 1647 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE 1648 * requests rather than doing it here. 1649 */ 1650 if ((flags & MAP_SHARED) != 0) { 1651 if ((fp->f_flag & FWRITE) != 0) 1652 maxprot |= VM_PROT_WRITE; 1653 else if ((prot & VM_PROT_WRITE) != 0) 1654 return (EACCES); 1655 } 1656 maxprot &= cap_maxprot; 1657 1658 linux_get_fop(filp, &fop, &ldev); 1659 error = linux_file_mmap_sub(td, size, prot, &maxprot, &flags, fp, 1660 &foff, fop, &object); 1661 if (error != 0) 1662 goto out; 1663 1664 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1665 foff, FALSE, td); 1666 if (error != 0) 1667 vm_object_deallocate(object); 1668out: 1669 linux_drop_fop(ldev); 1670 return (error); 1671} 1672 1673static int 1674linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 1675 struct thread *td) 1676{ 1677 struct linux_file *filp; 1678 struct vnode *vp; 1679 int error; 1680 1681 filp = (struct linux_file *)fp->f_data; 1682 if (filp->f_vnode == NULL) 1683 return (EOPNOTSUPP); 1684 1685 vp = filp->f_vnode; 1686 1687 vn_lock(vp, LK_SHARED | LK_RETRY); 1688 error = vn_stat(vp, sb, td->td_ucred, NOCRED, td); 1689 VOP_UNLOCK(vp, 0); 1690 1691 return (error); 1692} 1693 1694static int 1695linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif, 1696 struct filedesc *fdp) 1697{ 1698 struct linux_file *filp; 1699 struct vnode *vp; 1700 int error; 1701 1702 filp = fp->f_data; 1703 vp = filp->f_vnode; 1704 if (vp == NULL) { 1705 error = 0; 1706 kif->kf_type = KF_TYPE_DEV; 1707 } else { 1708 vref(vp); 1709 FILEDESC_SUNLOCK(fdp); 1710 error = vn_fill_kinfo_vnode(vp, kif); 1711 vrele(vp); 1712 kif->kf_type = KF_TYPE_VNODE; 1713 FILEDESC_SLOCK(fdp); 1714 } 1715 return (error); 1716} 1717 1718unsigned int 1719linux_iminor(struct inode *inode) 1720{ 1721 struct linux_cdev *ldev; 1722 1723 if (inode == NULL || inode->v_rdev == NULL || 1724 inode->v_rdev->si_devsw != &linuxcdevsw) 1725 return (-1U); 1726 ldev = inode->v_rdev->si_drv1; 1727 if (ldev == NULL) 1728 return (-1U); 1729 1730 return (minor(ldev->dev)); 1731} 1732 1733struct fileops linuxfileops = { 1734 .fo_read = linux_file_read, 1735 .fo_write = linux_file_write, 1736 .fo_truncate = invfo_truncate, 1737 .fo_kqfilter = linux_file_kqfilter, 1738 .fo_stat = linux_file_stat, 1739 .fo_fill_kinfo = linux_file_fill_kinfo, 1740 .fo_poll = linux_file_poll, 1741 .fo_close = linux_file_close, 1742 .fo_ioctl = linux_file_ioctl, 1743 .fo_mmap = linux_file_mmap, 1744 .fo_chmod = invfo_chmod, 1745 .fo_chown = invfo_chown, 1746 .fo_sendfile = invfo_sendfile, 1747 .fo_flags = DFLAG_PASSABLE, 1748}; 1749 1750/* 1751 * Hash of vmmap addresses. This is infrequently accessed and does not 1752 * need to be particularly large. This is done because we must store the 1753 * caller's idea of the map size to properly unmap. 1754 */ 1755struct vmmap { 1756 LIST_ENTRY(vmmap) vm_next; 1757 void *vm_addr; 1758 unsigned long vm_size; 1759}; 1760 1761struct vmmaphd { 1762 struct vmmap *lh_first; 1763}; 1764#define VMMAP_HASH_SIZE 64 1765#define VMMAP_HASH_MASK (VMMAP_HASH_SIZE - 1) 1766#define VM_HASH(addr) ((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK 1767static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE]; 1768static struct mtx vmmaplock; 1769 1770static void 1771vmmap_add(void *addr, unsigned long size) 1772{ 1773 struct vmmap *vmmap; 1774 1775 vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL); 1776 mtx_lock(&vmmaplock); 1777 vmmap->vm_size = size; 1778 vmmap->vm_addr = addr; 1779 LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next); 1780 mtx_unlock(&vmmaplock); 1781} 1782 1783static struct vmmap * 1784vmmap_remove(void *addr) 1785{ 1786 struct vmmap *vmmap; 1787 1788 mtx_lock(&vmmaplock); 1789 LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next) 1790 if (vmmap->vm_addr == addr) 1791 break; 1792 if (vmmap) 1793 LIST_REMOVE(vmmap, vm_next); 1794 mtx_unlock(&vmmaplock); 1795 1796 return (vmmap); 1797} 1798 1799#if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) 1800void * 1801_ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr) 1802{ 1803 void *addr; 1804 1805 addr = pmap_mapdev_attr(phys_addr, size, attr); 1806 if (addr == NULL) 1807 return (NULL); 1808 vmmap_add(addr, size); 1809 1810 return (addr); 1811} 1812#endif 1813 1814void 1815iounmap(void *addr) 1816{ 1817 struct vmmap *vmmap; 1818 1819 vmmap = vmmap_remove(addr); 1820 if (vmmap == NULL) 1821 return; 1822#if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) 1823 pmap_unmapdev((vm_offset_t)addr, vmmap->vm_size); 1824#endif 1825 kfree(vmmap); 1826} 1827 1828 1829void * 1830vmap(struct page **pages, unsigned int count, unsigned long flags, int prot) 1831{ 1832 vm_offset_t off; 1833 size_t size; 1834 1835 size = count * PAGE_SIZE; 1836 off = kva_alloc(size); 1837 if (off == 0) 1838 return (NULL); 1839 vmmap_add((void *)off, size); 1840 pmap_qenter(off, pages, count); 1841 1842 return ((void *)off); 1843} 1844 1845void 1846vunmap(void *addr) 1847{ 1848 struct vmmap *vmmap; 1849 1850 vmmap = vmmap_remove(addr); 1851 if (vmmap == NULL) 1852 return; 1853 pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE); 1854 kva_free((vm_offset_t)addr, vmmap->vm_size); 1855 kfree(vmmap); 1856} 1857 1858char * 1859kvasprintf(gfp_t gfp, const char *fmt, va_list ap) 1860{ 1861 unsigned int len; 1862 char *p; 1863 va_list aq; 1864 1865 va_copy(aq, ap); 1866 len = vsnprintf(NULL, 0, fmt, aq); 1867 va_end(aq); 1868 1869 p = kmalloc(len + 1, gfp); 1870 if (p != NULL) 1871 vsnprintf(p, len + 1, fmt, ap); 1872 1873 return (p); 1874} 1875 1876char * 1877kasprintf(gfp_t gfp, const char *fmt, ...) 1878{ 1879 va_list ap; 1880 char *p; 1881 1882 va_start(ap, fmt); 1883 p = kvasprintf(gfp, fmt, ap); 1884 va_end(ap); 1885 1886 return (p); 1887} 1888 1889static void 1890linux_timer_callback_wrapper(void *context) 1891{ 1892 struct timer_list *timer; 1893 1894 linux_set_current(curthread); 1895 1896 timer = context; 1897 timer->function(timer->data); 1898} 1899 1900void 1901mod_timer(struct timer_list *timer, int expires) 1902{ 1903 1904 timer->expires = expires; 1905 callout_reset(&timer->callout, 1906 linux_timer_jiffies_until(expires), 1907 &linux_timer_callback_wrapper, timer); 1908} 1909 1910void 1911add_timer(struct timer_list *timer) 1912{ 1913 1914 callout_reset(&timer->callout, 1915 linux_timer_jiffies_until(timer->expires), 1916 &linux_timer_callback_wrapper, timer); 1917} 1918 1919void 1920add_timer_on(struct timer_list *timer, int cpu) 1921{ 1922 1923 callout_reset_on(&timer->callout, 1924 linux_timer_jiffies_until(timer->expires), 1925 &linux_timer_callback_wrapper, timer, cpu); 1926} 1927 1928/* greatest common divisor, Euclid equation */ 1929static uint64_t 1930lkpi_gcd_64(uint64_t a, uint64_t b) 1931{ 1932 uint64_t an; 1933 uint64_t bn; 1934 1935 while (b != 0) { 1936 an = b; 1937 bn = a % b; 1938 a = an; 1939 b = bn; 1940 } 1941 return (a); 1942} 1943 1944uint64_t lkpi_nsec2hz_rem; 1945uint64_t lkpi_nsec2hz_div = 1000000000ULL; 1946uint64_t lkpi_nsec2hz_max; 1947 1948uint64_t lkpi_usec2hz_rem; 1949uint64_t lkpi_usec2hz_div = 1000000ULL; 1950uint64_t lkpi_usec2hz_max; 1951 1952uint64_t lkpi_msec2hz_rem; 1953uint64_t lkpi_msec2hz_div = 1000ULL; 1954uint64_t lkpi_msec2hz_max; 1955 1956static void 1957linux_timer_init(void *arg) 1958{ 1959 uint64_t gcd; 1960 1961 /* 1962 * Compute an internal HZ value which can divide 2**32 to 1963 * avoid timer rounding problems when the tick value wraps 1964 * around 2**32: 1965 */ 1966 linux_timer_hz_mask = 1; 1967 while (linux_timer_hz_mask < (unsigned long)hz) 1968 linux_timer_hz_mask *= 2; 1969 linux_timer_hz_mask--; 1970 1971 /* compute some internal constants */ 1972 1973 lkpi_nsec2hz_rem = hz; 1974 lkpi_usec2hz_rem = hz; 1975 lkpi_msec2hz_rem = hz; 1976 1977 gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div); 1978 lkpi_nsec2hz_rem /= gcd; 1979 lkpi_nsec2hz_div /= gcd; 1980 lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem; 1981 1982 gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div); 1983 lkpi_usec2hz_rem /= gcd; 1984 lkpi_usec2hz_div /= gcd; 1985 lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem; 1986 1987 gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div); 1988 lkpi_msec2hz_rem /= gcd; 1989 lkpi_msec2hz_div /= gcd; 1990 lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem; 1991} 1992SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL); 1993 1994void 1995linux_complete_common(struct completion *c, int all) 1996{ 1997 int wakeup_swapper; 1998 1999 sleepq_lock(c); 2000 if (all) { 2001 c->done = UINT_MAX; 2002 wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0); 2003 } else { 2004 if (c->done != UINT_MAX) 2005 c->done++; 2006 wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0); 2007 } 2008 sleepq_release(c); 2009 if (wakeup_swapper) 2010 kick_proc0(); 2011} 2012 2013/* 2014 * Indefinite wait for done != 0 with or without signals. 2015 */ 2016int 2017linux_wait_for_common(struct completion *c, int flags) 2018{ 2019 struct task_struct *task; 2020 int error; 2021 2022 if (SCHEDULER_STOPPED()) 2023 return (0); 2024 2025 task = current; 2026 2027 if (flags != 0) 2028 flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP; 2029 else 2030 flags = SLEEPQ_SLEEP; 2031 error = 0; 2032 for (;;) { 2033 sleepq_lock(c); 2034 if (c->done) 2035 break; 2036 sleepq_add(c, NULL, "completion", flags, 0); 2037 if (flags & SLEEPQ_INTERRUPTIBLE) { 2038 DROP_GIANT(); 2039 error = -sleepq_wait_sig(c, 0); 2040 PICKUP_GIANT(); 2041 if (error != 0) { 2042 linux_schedule_save_interrupt_value(task, error); 2043 error = -ERESTARTSYS; 2044 goto intr; 2045 } 2046 } else { 2047 DROP_GIANT(); 2048 sleepq_wait(c, 0); 2049 PICKUP_GIANT(); 2050 } 2051 } 2052 if (c->done != UINT_MAX) 2053 c->done--; 2054 sleepq_release(c); 2055 2056intr: 2057 return (error); 2058} 2059 2060/* 2061 * Time limited wait for done != 0 with or without signals. 2062 */ 2063int 2064linux_wait_for_timeout_common(struct completion *c, int timeout, int flags) 2065{ 2066 struct task_struct *task; 2067 int end = jiffies + timeout; 2068 int error; 2069 2070 if (SCHEDULER_STOPPED()) 2071 return (0); 2072 2073 task = current; 2074 2075 if (flags != 0) 2076 flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP; 2077 else 2078 flags = SLEEPQ_SLEEP; 2079 2080 for (;;) { 2081 sleepq_lock(c); 2082 if (c->done) 2083 break; 2084 sleepq_add(c, NULL, "completion", flags, 0); 2085 sleepq_set_timeout(c, linux_timer_jiffies_until(end)); 2086 2087 DROP_GIANT(); 2088 if (flags & SLEEPQ_INTERRUPTIBLE) 2089 error = -sleepq_timedwait_sig(c, 0); 2090 else 2091 error = -sleepq_timedwait(c, 0); 2092 PICKUP_GIANT(); 2093 2094 if (error != 0) { 2095 /* check for timeout */ 2096 if (error == -EWOULDBLOCK) { 2097 error = 0; /* timeout */ 2098 } else { 2099 /* signal happened */ 2100 linux_schedule_save_interrupt_value(task, error); 2101 error = -ERESTARTSYS; 2102 } 2103 goto done; 2104 } 2105 } 2106 if (c->done != UINT_MAX) 2107 c->done--; 2108 sleepq_release(c); 2109 2110 /* return how many jiffies are left */ 2111 error = linux_timer_jiffies_until(end); 2112done: 2113 return (error); 2114} 2115 2116int 2117linux_try_wait_for_completion(struct completion *c) 2118{ 2119 int isdone; 2120 2121 sleepq_lock(c); 2122 isdone = (c->done != 0); 2123 if (c->done != 0 && c->done != UINT_MAX) 2124 c->done--; 2125 sleepq_release(c); 2126 return (isdone); 2127} 2128 2129int 2130linux_completion_done(struct completion *c) 2131{ 2132 int isdone; 2133 2134 sleepq_lock(c); 2135 isdone = (c->done != 0); 2136 sleepq_release(c); 2137 return (isdone); 2138} 2139 2140static void 2141linux_cdev_deref(struct linux_cdev *ldev) 2142{ 2143 2144 if (refcount_release(&ldev->refs)) 2145 kfree(ldev); 2146} 2147 2148static void 2149linux_cdev_release(struct kobject *kobj) 2150{ 2151 struct linux_cdev *cdev; 2152 struct kobject *parent; 2153 2154 cdev = container_of(kobj, struct linux_cdev, kobj); 2155 parent = kobj->parent; 2156 linux_destroy_dev(cdev); 2157 linux_cdev_deref(cdev); 2158 kobject_put(parent); 2159} 2160 2161static void 2162linux_cdev_static_release(struct kobject *kobj) 2163{ 2164 struct linux_cdev *cdev; 2165 struct kobject *parent; 2166 2167 cdev = container_of(kobj, struct linux_cdev, kobj); 2168 parent = kobj->parent; 2169 linux_destroy_dev(cdev); 2170 kobject_put(parent); 2171} 2172 2173void 2174linux_destroy_dev(struct linux_cdev *ldev) 2175{ 2176 2177 if (ldev->cdev == NULL) 2178 return; 2179 2180 MPASS((ldev->siref & LDEV_SI_DTR) == 0); 2181 atomic_set_int(&ldev->siref, LDEV_SI_DTR); 2182 while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0) 2183 pause("ldevdtr", hz / 4); 2184 2185 destroy_dev(ldev->cdev); 2186 ldev->cdev = NULL; 2187} 2188 2189const struct kobj_type linux_cdev_ktype = { 2190 .release = linux_cdev_release, 2191}; 2192 2193const struct kobj_type linux_cdev_static_ktype = { 2194 .release = linux_cdev_static_release, 2195}; 2196 2197static void 2198linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate) 2199{ 2200 struct notifier_block *nb; 2201 2202 nb = arg; 2203 if (linkstate == LINK_STATE_UP) 2204 nb->notifier_call(nb, NETDEV_UP, ifp); 2205 else 2206 nb->notifier_call(nb, NETDEV_DOWN, ifp); 2207} 2208 2209static void 2210linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp) 2211{ 2212 struct notifier_block *nb; 2213 2214 nb = arg; 2215 nb->notifier_call(nb, NETDEV_REGISTER, ifp); 2216} 2217 2218static void 2219linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp) 2220{ 2221 struct notifier_block *nb; 2222 2223 nb = arg; 2224 nb->notifier_call(nb, NETDEV_UNREGISTER, ifp); 2225} 2226 2227static void 2228linux_handle_iflladdr_event(void *arg, struct ifnet *ifp) 2229{ 2230 struct notifier_block *nb; 2231 2232 nb = arg; 2233 nb->notifier_call(nb, NETDEV_CHANGEADDR, ifp); 2234} 2235 2236static void 2237linux_handle_ifaddr_event(void *arg, struct ifnet *ifp) 2238{ 2239 struct notifier_block *nb; 2240 2241 nb = arg; 2242 nb->notifier_call(nb, NETDEV_CHANGEIFADDR, ifp); 2243} 2244 2245int 2246register_netdevice_notifier(struct notifier_block *nb) 2247{ 2248 2249 nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER( 2250 ifnet_link_event, linux_handle_ifnet_link_event, nb, 0); 2251 nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER( 2252 ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0); 2253 nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER( 2254 ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0); 2255 nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER( 2256 iflladdr_event, linux_handle_iflladdr_event, nb, 0); 2257 2258 return (0); 2259} 2260 2261int 2262register_inetaddr_notifier(struct notifier_block *nb) 2263{ 2264 2265 nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER( 2266 ifaddr_event, linux_handle_ifaddr_event, nb, 0); 2267 return (0); 2268} 2269 2270int 2271unregister_netdevice_notifier(struct notifier_block *nb) 2272{ 2273 2274 EVENTHANDLER_DEREGISTER(ifnet_link_event, 2275 nb->tags[NETDEV_UP]); 2276 EVENTHANDLER_DEREGISTER(ifnet_arrival_event, 2277 nb->tags[NETDEV_REGISTER]); 2278 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 2279 nb->tags[NETDEV_UNREGISTER]); 2280 EVENTHANDLER_DEREGISTER(iflladdr_event, 2281 nb->tags[NETDEV_CHANGEADDR]); 2282 2283 return (0); 2284} 2285 2286int 2287unregister_inetaddr_notifier(struct notifier_block *nb) 2288{ 2289 2290 EVENTHANDLER_DEREGISTER(ifaddr_event, 2291 nb->tags[NETDEV_CHANGEIFADDR]); 2292 2293 return (0); 2294} 2295 2296struct list_sort_thunk { 2297 int (*cmp)(void *, struct list_head *, struct list_head *); 2298 void *priv; 2299}; 2300 2301static inline int 2302linux_le_cmp(void *priv, const void *d1, const void *d2) 2303{ 2304 struct list_head *le1, *le2; 2305 struct list_sort_thunk *thunk; 2306 2307 thunk = priv; 2308 le1 = *(__DECONST(struct list_head **, d1)); 2309 le2 = *(__DECONST(struct list_head **, d2)); 2310 return ((thunk->cmp)(thunk->priv, le1, le2)); 2311} 2312 2313void 2314list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv, 2315 struct list_head *a, struct list_head *b)) 2316{ 2317 struct list_sort_thunk thunk; 2318 struct list_head **ar, *le; 2319 size_t count, i; 2320 2321 count = 0; 2322 list_for_each(le, head) 2323 count++; 2324 ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK); 2325 i = 0; 2326 list_for_each(le, head) 2327 ar[i++] = le; 2328 thunk.cmp = cmp; 2329 thunk.priv = priv; 2330 qsort_r(ar, count, sizeof(struct list_head *), &thunk, linux_le_cmp); 2331 INIT_LIST_HEAD(head); 2332 for (i = 0; i < count; i++) 2333 list_add_tail(ar[i], head); 2334 free(ar, M_KMALLOC); 2335} 2336 2337void 2338linux_irq_handler(void *ent) 2339{ 2340 struct irq_ent *irqe; 2341 2342 linux_set_current(curthread); 2343 2344 irqe = ent; 2345 irqe->handler(irqe->irq, irqe->arg); 2346} 2347 2348#if defined(__i386__) || defined(__amd64__) 2349int 2350linux_wbinvd_on_all_cpus(void) 2351{ 2352 2353 pmap_invalidate_cache(); 2354 return (0); 2355} 2356#endif 2357 2358int 2359linux_on_each_cpu(void callback(void *), void *data) 2360{ 2361 2362 smp_rendezvous(smp_no_rendezvous_barrier, callback, 2363 smp_no_rendezvous_barrier, data); 2364 return (0); 2365} 2366 2367int 2368linux_in_atomic(void) 2369{ 2370 2371 return ((curthread->td_pflags & TDP_NOFAULTING) != 0); 2372} 2373 2374struct linux_cdev * 2375linux_find_cdev(const char *name, unsigned major, unsigned minor) 2376{ 2377 dev_t dev = MKDEV(major, minor); 2378 struct cdev *cdev; 2379 2380 dev_lock(); 2381 LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) { 2382 struct linux_cdev *ldev = cdev->si_drv1; 2383 if (ldev->dev == dev && 2384 strcmp(kobject_name(&ldev->kobj), name) == 0) { 2385 break; 2386 } 2387 } 2388 dev_unlock(); 2389 2390 return (cdev != NULL ? cdev->si_drv1 : NULL); 2391} 2392 2393int 2394__register_chrdev(unsigned int major, unsigned int baseminor, 2395 unsigned int count, const char *name, 2396 const struct file_operations *fops) 2397{ 2398 struct linux_cdev *cdev; 2399 int ret = 0; 2400 int i; 2401 2402 for (i = baseminor; i < baseminor + count; i++) { 2403 cdev = cdev_alloc(); 2404 cdev->ops = fops; 2405 kobject_set_name(&cdev->kobj, name); 2406 2407 ret = cdev_add(cdev, makedev(major, i), 1); 2408 if (ret != 0) 2409 break; 2410 } 2411 return (ret); 2412} 2413 2414int 2415__register_chrdev_p(unsigned int major, unsigned int baseminor, 2416 unsigned int count, const char *name, 2417 const struct file_operations *fops, uid_t uid, 2418 gid_t gid, int mode) 2419{ 2420 struct linux_cdev *cdev; 2421 int ret = 0; 2422 int i; 2423 2424 for (i = baseminor; i < baseminor + count; i++) { 2425 cdev = cdev_alloc(); 2426 cdev->ops = fops; 2427 kobject_set_name(&cdev->kobj, name); 2428 2429 ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode); 2430 if (ret != 0) 2431 break; 2432 } 2433 return (ret); 2434} 2435 2436void 2437__unregister_chrdev(unsigned int major, unsigned int baseminor, 2438 unsigned int count, const char *name) 2439{ 2440 struct linux_cdev *cdevp; 2441 int i; 2442 2443 for (i = baseminor; i < baseminor + count; i++) { 2444 cdevp = linux_find_cdev(name, major, i); 2445 if (cdevp != NULL) 2446 cdev_del(cdevp); 2447 } 2448} 2449 2450void 2451linux_dump_stack(void) 2452{ 2453#ifdef STACK 2454 struct stack st; 2455 2456 stack_zero(&st); 2457 stack_save(&st); 2458 stack_print(&st); 2459#endif 2460} 2461 2462#if defined(__i386__) || defined(__amd64__) 2463bool linux_cpu_has_clflush; 2464#endif 2465 2466static void 2467linux_compat_init(void *arg) 2468{ 2469 struct sysctl_oid *rootoid; 2470 int i; 2471 2472#if defined(__i386__) || defined(__amd64__) 2473 linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH); 2474#endif 2475 rw_init(&linux_vma_lock, "lkpi-vma-lock"); 2476 2477 rootoid = SYSCTL_ADD_ROOT_NODE(NULL, 2478 OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys"); 2479 kobject_init(&linux_class_root, &linux_class_ktype); 2480 kobject_set_name(&linux_class_root, "class"); 2481 linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid), 2482 OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class"); 2483 kobject_init(&linux_root_device.kobj, &linux_dev_ktype); 2484 kobject_set_name(&linux_root_device.kobj, "device"); 2485 linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL, 2486 SYSCTL_CHILDREN(rootoid), OID_AUTO, "device", CTLFLAG_RD, NULL, 2487 "device"); 2488 linux_root_device.bsddev = root_bus; 2489 linux_class_misc.name = "misc"; 2490 class_register(&linux_class_misc); 2491 INIT_LIST_HEAD(&pci_drivers); 2492 INIT_LIST_HEAD(&pci_devices); 2493 spin_lock_init(&pci_lock); 2494 mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF); 2495 for (i = 0; i < VMMAP_HASH_SIZE; i++) 2496 LIST_INIT(&vmmaphead[i]); 2497} 2498SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL); 2499 2500static void 2501linux_compat_uninit(void *arg) 2502{ 2503 linux_kobject_kfree_name(&linux_class_root); 2504 linux_kobject_kfree_name(&linux_root_device.kobj); 2505 linux_kobject_kfree_name(&linux_class_misc.kobj); 2506 2507 mtx_destroy(&vmmaplock); 2508 spin_lock_destroy(&pci_lock); 2509 rw_destroy(&linux_vma_lock); 2510} 2511SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL); 2512 2513/* 2514 * NOTE: Linux frequently uses "unsigned long" for pointer to integer 2515 * conversion and vice versa, where in FreeBSD "uintptr_t" would be 2516 * used. Assert these types have the same size, else some parts of the 2517 * LinuxKPI may not work like expected: 2518 */ 2519CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t)); 2520