1/*- 2 * Copyright (c) 2010 Isilon Systems, Inc. 3 * Copyright (c) 2010 iX Systems, Inc. 4 * Copyright (c) 2010 Panasas, Inc. 5 * Copyright (c) 2013-2018 Mellanox Technologies, Ltd. 6 * All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice unmodified, this list of conditions, and the following 13 * disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 21 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 22 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 23 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 27 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30#include <sys/cdefs.h> 31__FBSDID("$FreeBSD$"); 32 33#include "opt_stack.h" 34 35#include <sys/param.h> 36#include <sys/systm.h> 37#include <sys/malloc.h> 38#include <sys/kernel.h> 39#include <sys/sysctl.h> 40#include <sys/proc.h> 41#include <sys/sglist.h> 42#include <sys/sleepqueue.h> 43#include <sys/refcount.h> 44#include <sys/lock.h> 45#include <sys/mutex.h> 46#include <sys/bus.h> 47#include <sys/fcntl.h> 48#include <sys/file.h> 49#include <sys/filio.h> 50#include <sys/rwlock.h> 51#include <sys/mman.h> 52#include <sys/stack.h> 53#include <sys/user.h> 54 55#include <vm/vm.h> 56#include <vm/pmap.h> 57#include <vm/vm_object.h> 58#include <vm/vm_page.h> 59#include <vm/vm_pager.h> 60 61#include <machine/stdarg.h> 62 63#if defined(__i386__) || defined(__amd64__) 64#include <machine/md_var.h> 65#endif 66 67#include <linux/kobject.h> 68#include <linux/device.h> 69#include <linux/slab.h> 70#include <linux/module.h> 71#include <linux/moduleparam.h> 72#include <linux/cdev.h> 73#include <linux/file.h> 74#include <linux/sysfs.h> 75#include <linux/mm.h> 76#include <linux/io.h> 77#include <linux/vmalloc.h> 78#include <linux/netdevice.h> 79#include <linux/timer.h> 80#include <linux/interrupt.h> 81#include <linux/uaccess.h> 82#include <linux/list.h> 83#include <linux/kthread.h> 84#include <linux/kernel.h> 85#include <linux/compat.h> 86#include <linux/poll.h> 87#include <linux/smp.h> 88#include <linux/wait_bit.h> 89 90#if defined(__i386__) || defined(__amd64__) 91#include <asm/smp.h> 92#endif 93 94SYSCTL_NODE(_compat, OID_AUTO, linuxkpi, CTLFLAG_RW, 0, "LinuxKPI parameters"); 95 96int linuxkpi_debug; 97SYSCTL_INT(_compat_linuxkpi, OID_AUTO, debug, CTLFLAG_RWTUN, 98 &linuxkpi_debug, 0, "Set to enable pr_debug() prints. Clear to disable."); 99 100MALLOC_DEFINE(M_KMALLOC, "linux", "Linux kmalloc compat"); 101 102#include <linux/rbtree.h> 103/* Undo Linux compat changes. */ 104#undef RB_ROOT 105#undef file 106#undef cdev 107#define RB_ROOT(head) (head)->rbh_root 108 109static void linux_cdev_deref(struct linux_cdev *ldev); 110static struct vm_area_struct *linux_cdev_handle_find(void *handle); 111 112struct kobject linux_class_root; 113struct device linux_root_device; 114struct class linux_class_misc; 115struct list_head pci_drivers; 116struct list_head pci_devices; 117spinlock_t pci_lock; 118 119unsigned long linux_timer_hz_mask; 120 121wait_queue_head_t linux_bit_waitq; 122wait_queue_head_t linux_var_waitq; 123 124int 125panic_cmp(struct rb_node *one, struct rb_node *two) 126{ 127 panic("no cmp"); 128} 129 130RB_GENERATE(linux_root, rb_node, __entry, panic_cmp); 131 132int 133kobject_set_name_vargs(struct kobject *kobj, const char *fmt, va_list args) 134{ 135 va_list tmp_va; 136 int len; 137 char *old; 138 char *name; 139 char dummy; 140 141 old = kobj->name; 142 143 if (old && fmt == NULL) 144 return (0); 145 146 /* compute length of string */ 147 va_copy(tmp_va, args); 148 len = vsnprintf(&dummy, 0, fmt, tmp_va); 149 va_end(tmp_va); 150 151 /* account for zero termination */ 152 len++; 153 154 /* check for error */ 155 if (len < 1) 156 return (-EINVAL); 157 158 /* allocate memory for string */ 159 name = kzalloc(len, GFP_KERNEL); 160 if (name == NULL) 161 return (-ENOMEM); 162 vsnprintf(name, len, fmt, args); 163 kobj->name = name; 164 165 /* free old string */ 166 kfree(old); 167 168 /* filter new string */ 169 for (; *name != '\0'; name++) 170 if (*name == '/') 171 *name = '!'; 172 return (0); 173} 174 175int 176kobject_set_name(struct kobject *kobj, const char *fmt, ...) 177{ 178 va_list args; 179 int error; 180 181 va_start(args, fmt); 182 error = kobject_set_name_vargs(kobj, fmt, args); 183 va_end(args); 184 185 return (error); 186} 187 188static int 189kobject_add_complete(struct kobject *kobj, struct kobject *parent) 190{ 191 const struct kobj_type *t; 192 int error; 193 194 kobj->parent = parent; 195 error = sysfs_create_dir(kobj); 196 if (error == 0 && kobj->ktype && kobj->ktype->default_attrs) { 197 struct attribute **attr; 198 t = kobj->ktype; 199 200 for (attr = t->default_attrs; *attr != NULL; attr++) { 201 error = sysfs_create_file(kobj, *attr); 202 if (error) 203 break; 204 } 205 if (error) 206 sysfs_remove_dir(kobj); 207 208 } 209 return (error); 210} 211 212int 213kobject_add(struct kobject *kobj, struct kobject *parent, const char *fmt, ...) 214{ 215 va_list args; 216 int error; 217 218 va_start(args, fmt); 219 error = kobject_set_name_vargs(kobj, fmt, args); 220 va_end(args); 221 if (error) 222 return (error); 223 224 return kobject_add_complete(kobj, parent); 225} 226 227void 228linux_kobject_release(struct kref *kref) 229{ 230 struct kobject *kobj; 231 char *name; 232 233 kobj = container_of(kref, struct kobject, kref); 234 sysfs_remove_dir(kobj); 235 name = kobj->name; 236 if (kobj->ktype && kobj->ktype->release) 237 kobj->ktype->release(kobj); 238 kfree(name); 239} 240 241static void 242linux_kobject_kfree(struct kobject *kobj) 243{ 244 kfree(kobj); 245} 246 247static void 248linux_kobject_kfree_name(struct kobject *kobj) 249{ 250 if (kobj) { 251 kfree(kobj->name); 252 } 253} 254 255const struct kobj_type linux_kfree_type = { 256 .release = linux_kobject_kfree 257}; 258 259static void 260linux_device_release(struct device *dev) 261{ 262 pr_debug("linux_device_release: %s\n", dev_name(dev)); 263 kfree(dev); 264} 265 266static ssize_t 267linux_class_show(struct kobject *kobj, struct attribute *attr, char *buf) 268{ 269 struct class_attribute *dattr; 270 ssize_t error; 271 272 dattr = container_of(attr, struct class_attribute, attr); 273 error = -EIO; 274 if (dattr->show) 275 error = dattr->show(container_of(kobj, struct class, kobj), 276 dattr, buf); 277 return (error); 278} 279 280static ssize_t 281linux_class_store(struct kobject *kobj, struct attribute *attr, const char *buf, 282 size_t count) 283{ 284 struct class_attribute *dattr; 285 ssize_t error; 286 287 dattr = container_of(attr, struct class_attribute, attr); 288 error = -EIO; 289 if (dattr->store) 290 error = dattr->store(container_of(kobj, struct class, kobj), 291 dattr, buf, count); 292 return (error); 293} 294 295static void 296linux_class_release(struct kobject *kobj) 297{ 298 struct class *class; 299 300 class = container_of(kobj, struct class, kobj); 301 if (class->class_release) 302 class->class_release(class); 303} 304 305static const struct sysfs_ops linux_class_sysfs = { 306 .show = linux_class_show, 307 .store = linux_class_store, 308}; 309 310const struct kobj_type linux_class_ktype = { 311 .release = linux_class_release, 312 .sysfs_ops = &linux_class_sysfs 313}; 314 315static void 316linux_dev_release(struct kobject *kobj) 317{ 318 struct device *dev; 319 320 dev = container_of(kobj, struct device, kobj); 321 /* This is the precedence defined by linux. */ 322 if (dev->release) 323 dev->release(dev); 324 else if (dev->class && dev->class->dev_release) 325 dev->class->dev_release(dev); 326} 327 328static ssize_t 329linux_dev_show(struct kobject *kobj, struct attribute *attr, char *buf) 330{ 331 struct device_attribute *dattr; 332 ssize_t error; 333 334 dattr = container_of(attr, struct device_attribute, attr); 335 error = -EIO; 336 if (dattr->show) 337 error = dattr->show(container_of(kobj, struct device, kobj), 338 dattr, buf); 339 return (error); 340} 341 342static ssize_t 343linux_dev_store(struct kobject *kobj, struct attribute *attr, const char *buf, 344 size_t count) 345{ 346 struct device_attribute *dattr; 347 ssize_t error; 348 349 dattr = container_of(attr, struct device_attribute, attr); 350 error = -EIO; 351 if (dattr->store) 352 error = dattr->store(container_of(kobj, struct device, kobj), 353 dattr, buf, count); 354 return (error); 355} 356 357static const struct sysfs_ops linux_dev_sysfs = { 358 .show = linux_dev_show, 359 .store = linux_dev_store, 360}; 361 362const struct kobj_type linux_dev_ktype = { 363 .release = linux_dev_release, 364 .sysfs_ops = &linux_dev_sysfs 365}; 366 367struct device * 368device_create(struct class *class, struct device *parent, dev_t devt, 369 void *drvdata, const char *fmt, ...) 370{ 371 struct device *dev; 372 va_list args; 373 374 dev = kzalloc(sizeof(*dev), M_WAITOK); 375 dev->parent = parent; 376 dev->class = class; 377 dev->devt = devt; 378 dev->driver_data = drvdata; 379 dev->release = linux_device_release; 380 va_start(args, fmt); 381 kobject_set_name_vargs(&dev->kobj, fmt, args); 382 va_end(args); 383 device_register(dev); 384 385 return (dev); 386} 387 388int 389kobject_init_and_add(struct kobject *kobj, const struct kobj_type *ktype, 390 struct kobject *parent, const char *fmt, ...) 391{ 392 va_list args; 393 int error; 394 395 kobject_init(kobj, ktype); 396 kobj->ktype = ktype; 397 kobj->parent = parent; 398 kobj->name = NULL; 399 400 va_start(args, fmt); 401 error = kobject_set_name_vargs(kobj, fmt, args); 402 va_end(args); 403 if (error) 404 return (error); 405 return kobject_add_complete(kobj, parent); 406} 407 408static void 409linux_kq_lock(void *arg) 410{ 411 spinlock_t *s = arg; 412 413 spin_lock(s); 414} 415static void 416linux_kq_unlock(void *arg) 417{ 418 spinlock_t *s = arg; 419 420 spin_unlock(s); 421} 422 423static void 424linux_kq_lock_owned(void *arg) 425{ 426#ifdef INVARIANTS 427 spinlock_t *s = arg; 428 429 mtx_assert(&s->m, MA_OWNED); 430#endif 431} 432 433static void 434linux_kq_lock_unowned(void *arg) 435{ 436#ifdef INVARIANTS 437 spinlock_t *s = arg; 438 439 mtx_assert(&s->m, MA_NOTOWNED); 440#endif 441} 442 443static void 444linux_file_kqfilter_poll(struct linux_file *, int); 445 446struct linux_file * 447linux_file_alloc(void) 448{ 449 struct linux_file *filp; 450 451 filp = kzalloc(sizeof(*filp), GFP_KERNEL); 452 453 /* set initial refcount */ 454 filp->f_count = 1; 455 456 /* setup fields needed by kqueue support */ 457 spin_lock_init(&filp->f_kqlock); 458 knlist_init(&filp->f_selinfo.si_note, &filp->f_kqlock, 459 linux_kq_lock, linux_kq_unlock, 460 linux_kq_lock_owned, linux_kq_lock_unowned); 461 462 return (filp); 463} 464 465void 466linux_file_free(struct linux_file *filp) 467{ 468 if (filp->_file == NULL) { 469 if (filp->f_shmem != NULL) 470 vm_object_deallocate(filp->f_shmem); 471 kfree(filp); 472 } else { 473 /* 474 * The close method of the character device or file 475 * will free the linux_file structure: 476 */ 477 _fdrop(filp->_file, curthread); 478 } 479} 480 481static int 482linux_cdev_pager_fault(vm_object_t vm_obj, vm_ooffset_t offset, int prot, 483 vm_page_t *mres) 484{ 485 struct vm_area_struct *vmap; 486 487 vmap = linux_cdev_handle_find(vm_obj->handle); 488 489 MPASS(vmap != NULL); 490 MPASS(vmap->vm_private_data == vm_obj->handle); 491 492 if (likely(vmap->vm_ops != NULL && offset < vmap->vm_len)) { 493 vm_paddr_t paddr = IDX_TO_OFF(vmap->vm_pfn) + offset; 494 vm_page_t page; 495 496 if (((*mres)->flags & PG_FICTITIOUS) != 0) { 497 /* 498 * If the passed in result page is a fake 499 * page, update it with the new physical 500 * address. 501 */ 502 page = *mres; 503 vm_page_updatefake(page, paddr, vm_obj->memattr); 504 } else { 505 /* 506 * Replace the passed in "mres" page with our 507 * own fake page and free up the all of the 508 * original pages. 509 */ 510 VM_OBJECT_WUNLOCK(vm_obj); 511 page = vm_page_getfake(paddr, vm_obj->memattr); 512 VM_OBJECT_WLOCK(vm_obj); 513 514 vm_page_replace_checked(page, vm_obj, 515 (*mres)->pindex, *mres); 516 517 vm_page_lock(*mres); 518 vm_page_free(*mres); 519 vm_page_unlock(*mres); 520 *mres = page; 521 } 522 page->valid = VM_PAGE_BITS_ALL; 523 return (VM_PAGER_OK); 524 } 525 return (VM_PAGER_FAIL); 526} 527 528static int 529linux_cdev_pager_populate(vm_object_t vm_obj, vm_pindex_t pidx, int fault_type, 530 vm_prot_t max_prot, vm_pindex_t *first, vm_pindex_t *last) 531{ 532 struct vm_area_struct *vmap; 533 int err; 534 535 /* get VM area structure */ 536 vmap = linux_cdev_handle_find(vm_obj->handle); 537 MPASS(vmap != NULL); 538 MPASS(vmap->vm_private_data == vm_obj->handle); 539 540 VM_OBJECT_WUNLOCK(vm_obj); 541 542 linux_set_current(curthread); 543 544 down_write(&vmap->vm_mm->mmap_sem); 545 if (unlikely(vmap->vm_ops == NULL)) { 546 err = VM_FAULT_SIGBUS; 547 } else { 548 struct vm_fault vmf; 549 550 /* fill out VM fault structure */ 551 vmf.virtual_address = (void *)(uintptr_t)IDX_TO_OFF(pidx); 552 vmf.flags = (fault_type & VM_PROT_WRITE) ? FAULT_FLAG_WRITE : 0; 553 vmf.pgoff = 0; 554 vmf.page = NULL; 555 vmf.vma = vmap; 556 557 vmap->vm_pfn_count = 0; 558 vmap->vm_pfn_pcount = &vmap->vm_pfn_count; 559 vmap->vm_obj = vm_obj; 560 561 err = vmap->vm_ops->fault(vmap, &vmf); 562 563 while (vmap->vm_pfn_count == 0 && err == VM_FAULT_NOPAGE) { 564 kern_yield(PRI_USER); 565 err = vmap->vm_ops->fault(vmap, &vmf); 566 } 567 } 568 569 /* translate return code */ 570 switch (err) { 571 case VM_FAULT_OOM: 572 err = VM_PAGER_AGAIN; 573 break; 574 case VM_FAULT_SIGBUS: 575 err = VM_PAGER_BAD; 576 break; 577 case VM_FAULT_NOPAGE: 578 /* 579 * By contract the fault handler will return having 580 * busied all the pages itself. If pidx is already 581 * found in the object, it will simply xbusy the first 582 * page and return with vm_pfn_count set to 1. 583 */ 584 *first = vmap->vm_pfn_first; 585 *last = *first + vmap->vm_pfn_count - 1; 586 err = VM_PAGER_OK; 587 break; 588 default: 589 err = VM_PAGER_ERROR; 590 break; 591 } 592 up_write(&vmap->vm_mm->mmap_sem); 593 VM_OBJECT_WLOCK(vm_obj); 594 return (err); 595} 596 597static struct rwlock linux_vma_lock; 598static TAILQ_HEAD(, vm_area_struct) linux_vma_head = 599 TAILQ_HEAD_INITIALIZER(linux_vma_head); 600 601static void 602linux_cdev_handle_free(struct vm_area_struct *vmap) 603{ 604 /* Drop reference on vm_file */ 605 if (vmap->vm_file != NULL) 606 fput(vmap->vm_file); 607 608 /* Drop reference on mm_struct */ 609 mmput(vmap->vm_mm); 610 611 kfree(vmap); 612} 613 614static void 615linux_cdev_handle_remove(struct vm_area_struct *vmap) 616{ 617 rw_wlock(&linux_vma_lock); 618 TAILQ_REMOVE(&linux_vma_head, vmap, vm_entry); 619 rw_wunlock(&linux_vma_lock); 620} 621 622static struct vm_area_struct * 623linux_cdev_handle_find(void *handle) 624{ 625 struct vm_area_struct *vmap; 626 627 rw_rlock(&linux_vma_lock); 628 TAILQ_FOREACH(vmap, &linux_vma_head, vm_entry) { 629 if (vmap->vm_private_data == handle) 630 break; 631 } 632 rw_runlock(&linux_vma_lock); 633 return (vmap); 634} 635 636static int 637linux_cdev_pager_ctor(void *handle, vm_ooffset_t size, vm_prot_t prot, 638 vm_ooffset_t foff, struct ucred *cred, u_short *color) 639{ 640 641 MPASS(linux_cdev_handle_find(handle) != NULL); 642 *color = 0; 643 return (0); 644} 645 646static void 647linux_cdev_pager_dtor(void *handle) 648{ 649 const struct vm_operations_struct *vm_ops; 650 struct vm_area_struct *vmap; 651 652 vmap = linux_cdev_handle_find(handle); 653 MPASS(vmap != NULL); 654 655 /* 656 * Remove handle before calling close operation to prevent 657 * other threads from reusing the handle pointer. 658 */ 659 linux_cdev_handle_remove(vmap); 660 661 down_write(&vmap->vm_mm->mmap_sem); 662 vm_ops = vmap->vm_ops; 663 if (likely(vm_ops != NULL)) 664 vm_ops->close(vmap); 665 up_write(&vmap->vm_mm->mmap_sem); 666 667 linux_cdev_handle_free(vmap); 668} 669 670static struct cdev_pager_ops linux_cdev_pager_ops[2] = { 671 { 672 /* OBJT_MGTDEVICE */ 673 .cdev_pg_populate = linux_cdev_pager_populate, 674 .cdev_pg_ctor = linux_cdev_pager_ctor, 675 .cdev_pg_dtor = linux_cdev_pager_dtor 676 }, 677 { 678 /* OBJT_DEVICE */ 679 .cdev_pg_fault = linux_cdev_pager_fault, 680 .cdev_pg_ctor = linux_cdev_pager_ctor, 681 .cdev_pg_dtor = linux_cdev_pager_dtor 682 }, 683}; 684 685int 686zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, 687 unsigned long size) 688{ 689 vm_object_t obj; 690 vm_page_t m; 691 692 obj = vma->vm_obj; 693 if (obj == NULL || (obj->flags & OBJ_UNMANAGED) != 0) 694 return (-ENOTSUP); 695 VM_OBJECT_RLOCK(obj); 696 for (m = vm_page_find_least(obj, OFF_TO_IDX(address)); 697 m != NULL && m->pindex < OFF_TO_IDX(address + size); 698 m = TAILQ_NEXT(m, listq)) 699 pmap_remove_all(m); 700 VM_OBJECT_RUNLOCK(obj); 701 return (0); 702} 703 704static struct file_operations dummy_ldev_ops = { 705 /* XXXKIB */ 706}; 707 708static struct linux_cdev dummy_ldev = { 709 .ops = &dummy_ldev_ops, 710}; 711 712#define LDEV_SI_DTR 0x0001 713#define LDEV_SI_REF 0x0002 714 715static void 716linux_get_fop(struct linux_file *filp, const struct file_operations **fop, 717 struct linux_cdev **dev) 718{ 719 struct linux_cdev *ldev; 720 u_int siref; 721 722 ldev = filp->f_cdev; 723 *fop = filp->f_op; 724 if (ldev != NULL) { 725 if (ldev->kobj.ktype == &linux_cdev_static_ktype) { 726 refcount_acquire(&ldev->refs); 727 } else { 728 for (siref = ldev->siref;;) { 729 if ((siref & LDEV_SI_DTR) != 0) { 730 ldev = &dummy_ldev; 731 *fop = ldev->ops; 732 siref = ldev->siref; 733 MPASS((ldev->siref & LDEV_SI_DTR) == 0); 734 } else if (atomic_fcmpset_int(&ldev->siref, 735 &siref, siref + LDEV_SI_REF)) { 736 break; 737 } 738 } 739 } 740 } 741 *dev = ldev; 742} 743 744static void 745linux_drop_fop(struct linux_cdev *ldev) 746{ 747 748 if (ldev == NULL) 749 return; 750 if (ldev->kobj.ktype == &linux_cdev_static_ktype) { 751 linux_cdev_deref(ldev); 752 } else { 753 MPASS(ldev->kobj.ktype == &linux_cdev_ktype); 754 MPASS((ldev->siref & ~LDEV_SI_DTR) != 0); 755 atomic_subtract_int(&ldev->siref, LDEV_SI_REF); 756 } 757} 758 759#define OPW(fp,td,code) ({ \ 760 struct file *__fpop; \ 761 __typeof(code) __retval; \ 762 \ 763 __fpop = (td)->td_fpop; \ 764 (td)->td_fpop = (fp); \ 765 __retval = (code); \ 766 (td)->td_fpop = __fpop; \ 767 __retval; \ 768}) 769 770static int 771linux_dev_fdopen(struct cdev *dev, int fflags, struct thread *td, 772 struct file *file) 773{ 774 struct linux_cdev *ldev; 775 struct linux_file *filp; 776 const struct file_operations *fop; 777 int error; 778 779 ldev = dev->si_drv1; 780 781 filp = linux_file_alloc(); 782 filp->f_dentry = &filp->f_dentry_store; 783 filp->f_op = ldev->ops; 784 filp->f_mode = file->f_flag; 785 filp->f_flags = file->f_flag; 786 filp->f_vnode = file->f_vnode; 787 filp->_file = file; 788 refcount_acquire(&ldev->refs); 789 filp->f_cdev = ldev; 790 791 linux_set_current(td); 792 linux_get_fop(filp, &fop, &ldev); 793 794 if (fop->open != NULL) { 795 error = -fop->open(file->f_vnode, filp); 796 if (error != 0) { 797 linux_drop_fop(ldev); 798 linux_cdev_deref(filp->f_cdev); 799 kfree(filp); 800 return (error); 801 } 802 } 803 804 /* hold on to the vnode - used for fstat() */ 805 vhold(filp->f_vnode); 806 807 /* release the file from devfs */ 808 finit(file, filp->f_mode, DTYPE_DEV, filp, &linuxfileops); 809 linux_drop_fop(ldev); 810 return (ENXIO); 811} 812 813#define LINUX_IOCTL_MIN_PTR 0x10000UL 814#define LINUX_IOCTL_MAX_PTR (LINUX_IOCTL_MIN_PTR + IOCPARM_MAX) 815 816static inline int 817linux_remap_address(void **uaddr, size_t len) 818{ 819 uintptr_t uaddr_val = (uintptr_t)(*uaddr); 820 821 if (unlikely(uaddr_val >= LINUX_IOCTL_MIN_PTR && 822 uaddr_val < LINUX_IOCTL_MAX_PTR)) { 823 struct task_struct *pts = current; 824 if (pts == NULL) { 825 *uaddr = NULL; 826 return (1); 827 } 828 829 /* compute data offset */ 830 uaddr_val -= LINUX_IOCTL_MIN_PTR; 831 832 /* check that length is within bounds */ 833 if ((len > IOCPARM_MAX) || 834 (uaddr_val + len) > pts->bsd_ioctl_len) { 835 *uaddr = NULL; 836 return (1); 837 } 838 839 /* re-add kernel buffer address */ 840 uaddr_val += (uintptr_t)pts->bsd_ioctl_data; 841 842 /* update address location */ 843 *uaddr = (void *)uaddr_val; 844 return (1); 845 } 846 return (0); 847} 848 849int 850linux_copyin(const void *uaddr, void *kaddr, size_t len) 851{ 852 if (linux_remap_address(__DECONST(void **, &uaddr), len)) { 853 if (uaddr == NULL) 854 return (-EFAULT); 855 memcpy(kaddr, uaddr, len); 856 return (0); 857 } 858 return (-copyin(uaddr, kaddr, len)); 859} 860 861int 862linux_copyout(const void *kaddr, void *uaddr, size_t len) 863{ 864 if (linux_remap_address(&uaddr, len)) { 865 if (uaddr == NULL) 866 return (-EFAULT); 867 memcpy(uaddr, kaddr, len); 868 return (0); 869 } 870 return (-copyout(kaddr, uaddr, len)); 871} 872 873size_t 874linux_clear_user(void *_uaddr, size_t _len) 875{ 876 uint8_t *uaddr = _uaddr; 877 size_t len = _len; 878 879 /* make sure uaddr is aligned before going into the fast loop */ 880 while (((uintptr_t)uaddr & 7) != 0 && len > 7) { 881 if (subyte(uaddr, 0)) 882 return (_len); 883 uaddr++; 884 len--; 885 } 886 887 /* zero 8 bytes at a time */ 888 while (len > 7) { 889#ifdef __LP64__ 890 if (suword64(uaddr, 0)) 891 return (_len); 892#else 893 if (suword32(uaddr, 0)) 894 return (_len); 895 if (suword32(uaddr + 4, 0)) 896 return (_len); 897#endif 898 uaddr += 8; 899 len -= 8; 900 } 901 902 /* zero fill end, if any */ 903 while (len > 0) { 904 if (subyte(uaddr, 0)) 905 return (_len); 906 uaddr++; 907 len--; 908 } 909 return (0); 910} 911 912int 913linux_access_ok(const void *uaddr, size_t len) 914{ 915 uintptr_t saddr; 916 uintptr_t eaddr; 917 918 /* get start and end address */ 919 saddr = (uintptr_t)uaddr; 920 eaddr = (uintptr_t)uaddr + len; 921 922 /* verify addresses are valid for userspace */ 923 return ((saddr == eaddr) || 924 (eaddr > saddr && eaddr <= VM_MAXUSER_ADDRESS)); 925} 926 927/* 928 * This function should return either EINTR or ERESTART depending on 929 * the signal type sent to this thread: 930 */ 931static int 932linux_get_error(struct task_struct *task, int error) 933{ 934 /* check for signal type interrupt code */ 935 if (error == EINTR || error == ERESTARTSYS || error == ERESTART) { 936 error = -linux_schedule_get_interrupt_value(task); 937 if (error == 0) 938 error = EINTR; 939 } 940 return (error); 941} 942 943static int 944linux_file_ioctl_sub(struct file *fp, struct linux_file *filp, 945 const struct file_operations *fop, u_long cmd, caddr_t data, 946 struct thread *td) 947{ 948 struct task_struct *task = current; 949 unsigned size; 950 int error; 951 952 size = IOCPARM_LEN(cmd); 953 /* refer to logic in sys_ioctl() */ 954 if (size > 0) { 955 /* 956 * Setup hint for linux_copyin() and linux_copyout(). 957 * 958 * Background: Linux code expects a user-space address 959 * while FreeBSD supplies a kernel-space address. 960 */ 961 task->bsd_ioctl_data = data; 962 task->bsd_ioctl_len = size; 963 data = (void *)LINUX_IOCTL_MIN_PTR; 964 } else { 965 /* fetch user-space pointer */ 966 data = *(void **)data; 967 } 968#if defined(__amd64__) 969 if (td->td_proc->p_elf_machine == EM_386) { 970 /* try the compat IOCTL handler first */ 971 if (fop->compat_ioctl != NULL) { 972 error = -OPW(fp, td, fop->compat_ioctl(filp, 973 cmd, (u_long)data)); 974 } else { 975 error = ENOTTY; 976 } 977 978 /* fallback to the regular IOCTL handler, if any */ 979 if (error == ENOTTY && fop->unlocked_ioctl != NULL) { 980 error = -OPW(fp, td, fop->unlocked_ioctl(filp, 981 cmd, (u_long)data)); 982 } 983 } else 984#endif 985 { 986 if (fop->unlocked_ioctl != NULL) { 987 error = -OPW(fp, td, fop->unlocked_ioctl(filp, 988 cmd, (u_long)data)); 989 } else { 990 error = ENOTTY; 991 } 992 } 993 if (size > 0) { 994 task->bsd_ioctl_data = NULL; 995 task->bsd_ioctl_len = 0; 996 } 997 998 if (error == EWOULDBLOCK) { 999 /* update kqfilter status, if any */ 1000 linux_file_kqfilter_poll(filp, 1001 LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE); 1002 } else { 1003 error = linux_get_error(task, error); 1004 } 1005 return (error); 1006} 1007 1008#define LINUX_POLL_TABLE_NORMAL ((poll_table *)1) 1009 1010/* 1011 * This function atomically updates the poll wakeup state and returns 1012 * the previous state at the time of update. 1013 */ 1014static uint8_t 1015linux_poll_wakeup_state(atomic_t *v, const uint8_t *pstate) 1016{ 1017 int c, old; 1018 1019 c = v->counter; 1020 1021 while ((old = atomic_cmpxchg(v, c, pstate[c])) != c) 1022 c = old; 1023 1024 return (c); 1025} 1026 1027 1028static int 1029linux_poll_wakeup_callback(wait_queue_t *wq, unsigned int wq_state, int flags, void *key) 1030{ 1031 static const uint8_t state[LINUX_FWQ_STATE_MAX] = { 1032 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */ 1033 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */ 1034 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_READY, 1035 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_READY, /* NOP */ 1036 }; 1037 struct linux_file *filp = container_of(wq, struct linux_file, f_wait_queue.wq); 1038 1039 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { 1040 case LINUX_FWQ_STATE_QUEUED: 1041 linux_poll_wakeup(filp); 1042 return (1); 1043 default: 1044 return (0); 1045 } 1046} 1047 1048void 1049linux_poll_wait(struct linux_file *filp, wait_queue_head_t *wqh, poll_table *p) 1050{ 1051 static const uint8_t state[LINUX_FWQ_STATE_MAX] = { 1052 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_NOT_READY, 1053 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_NOT_READY, /* NOP */ 1054 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_QUEUED, /* NOP */ 1055 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_QUEUED, 1056 }; 1057 1058 /* check if we are called inside the select system call */ 1059 if (p == LINUX_POLL_TABLE_NORMAL) 1060 selrecord(curthread, &filp->f_selinfo); 1061 1062 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { 1063 case LINUX_FWQ_STATE_INIT: 1064 /* NOTE: file handles can only belong to one wait-queue */ 1065 filp->f_wait_queue.wqh = wqh; 1066 filp->f_wait_queue.wq.func = &linux_poll_wakeup_callback; 1067 add_wait_queue(wqh, &filp->f_wait_queue.wq); 1068 atomic_set(&filp->f_wait_queue.state, LINUX_FWQ_STATE_QUEUED); 1069 break; 1070 default: 1071 break; 1072 } 1073} 1074 1075static void 1076linux_poll_wait_dequeue(struct linux_file *filp) 1077{ 1078 static const uint8_t state[LINUX_FWQ_STATE_MAX] = { 1079 [LINUX_FWQ_STATE_INIT] = LINUX_FWQ_STATE_INIT, /* NOP */ 1080 [LINUX_FWQ_STATE_NOT_READY] = LINUX_FWQ_STATE_INIT, 1081 [LINUX_FWQ_STATE_QUEUED] = LINUX_FWQ_STATE_INIT, 1082 [LINUX_FWQ_STATE_READY] = LINUX_FWQ_STATE_INIT, 1083 }; 1084 1085 seldrain(&filp->f_selinfo); 1086 1087 switch (linux_poll_wakeup_state(&filp->f_wait_queue.state, state)) { 1088 case LINUX_FWQ_STATE_NOT_READY: 1089 case LINUX_FWQ_STATE_QUEUED: 1090 case LINUX_FWQ_STATE_READY: 1091 remove_wait_queue(filp->f_wait_queue.wqh, &filp->f_wait_queue.wq); 1092 break; 1093 default: 1094 break; 1095 } 1096} 1097 1098void 1099linux_poll_wakeup(struct linux_file *filp) 1100{ 1101 /* this function should be NULL-safe */ 1102 if (filp == NULL) 1103 return; 1104 1105 selwakeup(&filp->f_selinfo); 1106 1107 spin_lock(&filp->f_kqlock); 1108 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ | 1109 LINUX_KQ_FLAG_NEED_WRITE; 1110 1111 /* make sure the "knote" gets woken up */ 1112 KNOTE_LOCKED(&filp->f_selinfo.si_note, 1); 1113 spin_unlock(&filp->f_kqlock); 1114} 1115 1116static void 1117linux_file_kqfilter_detach(struct knote *kn) 1118{ 1119 struct linux_file *filp = kn->kn_hook; 1120 1121 spin_lock(&filp->f_kqlock); 1122 knlist_remove(&filp->f_selinfo.si_note, kn, 1); 1123 spin_unlock(&filp->f_kqlock); 1124} 1125 1126static int 1127linux_file_kqfilter_read_event(struct knote *kn, long hint) 1128{ 1129 struct linux_file *filp = kn->kn_hook; 1130 1131 mtx_assert(&filp->f_kqlock.m, MA_OWNED); 1132 1133 return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_READ) ? 1 : 0); 1134} 1135 1136static int 1137linux_file_kqfilter_write_event(struct knote *kn, long hint) 1138{ 1139 struct linux_file *filp = kn->kn_hook; 1140 1141 mtx_assert(&filp->f_kqlock.m, MA_OWNED); 1142 1143 return ((filp->f_kqflags & LINUX_KQ_FLAG_NEED_WRITE) ? 1 : 0); 1144} 1145 1146static struct filterops linux_dev_kqfiltops_read = { 1147 .f_isfd = 1, 1148 .f_detach = linux_file_kqfilter_detach, 1149 .f_event = linux_file_kqfilter_read_event, 1150}; 1151 1152static struct filterops linux_dev_kqfiltops_write = { 1153 .f_isfd = 1, 1154 .f_detach = linux_file_kqfilter_detach, 1155 .f_event = linux_file_kqfilter_write_event, 1156}; 1157 1158static void 1159linux_file_kqfilter_poll(struct linux_file *filp, int kqflags) 1160{ 1161 struct thread *td; 1162 const struct file_operations *fop; 1163 struct linux_cdev *ldev; 1164 int temp; 1165 1166 if ((filp->f_kqflags & kqflags) == 0) 1167 return; 1168 1169 td = curthread; 1170 1171 linux_get_fop(filp, &fop, &ldev); 1172 /* get the latest polling state */ 1173 temp = OPW(filp->_file, td, fop->poll(filp, NULL)); 1174 linux_drop_fop(ldev); 1175 1176 spin_lock(&filp->f_kqlock); 1177 /* clear kqflags */ 1178 filp->f_kqflags &= ~(LINUX_KQ_FLAG_NEED_READ | 1179 LINUX_KQ_FLAG_NEED_WRITE); 1180 /* update kqflags */ 1181 if ((temp & (POLLIN | POLLOUT)) != 0) { 1182 if ((temp & POLLIN) != 0) 1183 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_READ; 1184 if ((temp & POLLOUT) != 0) 1185 filp->f_kqflags |= LINUX_KQ_FLAG_NEED_WRITE; 1186 1187 /* make sure the "knote" gets woken up */ 1188 KNOTE_LOCKED(&filp->f_selinfo.si_note, 0); 1189 } 1190 spin_unlock(&filp->f_kqlock); 1191} 1192 1193static int 1194linux_file_kqfilter(struct file *file, struct knote *kn) 1195{ 1196 struct linux_file *filp; 1197 struct thread *td; 1198 int error; 1199 1200 td = curthread; 1201 filp = (struct linux_file *)file->f_data; 1202 filp->f_flags = file->f_flag; 1203 if (filp->f_op->poll == NULL) 1204 return (EINVAL); 1205 1206 spin_lock(&filp->f_kqlock); 1207 switch (kn->kn_filter) { 1208 case EVFILT_READ: 1209 filp->f_kqflags |= LINUX_KQ_FLAG_HAS_READ; 1210 kn->kn_fop = &linux_dev_kqfiltops_read; 1211 kn->kn_hook = filp; 1212 knlist_add(&filp->f_selinfo.si_note, kn, 1); 1213 error = 0; 1214 break; 1215 case EVFILT_WRITE: 1216 filp->f_kqflags |= LINUX_KQ_FLAG_HAS_WRITE; 1217 kn->kn_fop = &linux_dev_kqfiltops_write; 1218 kn->kn_hook = filp; 1219 knlist_add(&filp->f_selinfo.si_note, kn, 1); 1220 error = 0; 1221 break; 1222 default: 1223 error = EINVAL; 1224 break; 1225 } 1226 spin_unlock(&filp->f_kqlock); 1227 1228 if (error == 0) { 1229 linux_set_current(td); 1230 1231 /* update kqfilter status, if any */ 1232 linux_file_kqfilter_poll(filp, 1233 LINUX_KQ_FLAG_HAS_READ | LINUX_KQ_FLAG_HAS_WRITE); 1234 } 1235 return (error); 1236} 1237 1238static int 1239linux_file_mmap_single(struct file *fp, const struct file_operations *fop, 1240 vm_ooffset_t *offset, vm_size_t size, struct vm_object **object, 1241 int nprot, struct thread *td) 1242{ 1243 struct task_struct *task; 1244 struct vm_area_struct *vmap; 1245 struct mm_struct *mm; 1246 struct linux_file *filp; 1247 vm_memattr_t attr; 1248 int error; 1249 1250 filp = (struct linux_file *)fp->f_data; 1251 filp->f_flags = fp->f_flag; 1252 1253 if (fop->mmap == NULL) 1254 return (EOPNOTSUPP); 1255 1256 linux_set_current(td); 1257 1258 /* 1259 * The same VM object might be shared by multiple processes 1260 * and the mm_struct is usually freed when a process exits. 1261 * 1262 * The atomic reference below makes sure the mm_struct is 1263 * available as long as the vmap is in the linux_vma_head. 1264 */ 1265 task = current; 1266 mm = task->mm; 1267 if (atomic_inc_not_zero(&mm->mm_users) == 0) 1268 return (EINVAL); 1269 1270 vmap = kzalloc(sizeof(*vmap), GFP_KERNEL); 1271 vmap->vm_start = 0; 1272 vmap->vm_end = size; 1273 vmap->vm_pgoff = *offset / PAGE_SIZE; 1274 vmap->vm_pfn = 0; 1275 vmap->vm_flags = vmap->vm_page_prot = (nprot & VM_PROT_ALL); 1276 vmap->vm_ops = NULL; 1277 vmap->vm_file = get_file(filp); 1278 vmap->vm_mm = mm; 1279 1280 if (unlikely(down_write_killable(&vmap->vm_mm->mmap_sem))) { 1281 error = linux_get_error(task, EINTR); 1282 } else { 1283 error = -OPW(fp, td, fop->mmap(filp, vmap)); 1284 error = linux_get_error(task, error); 1285 up_write(&vmap->vm_mm->mmap_sem); 1286 } 1287 1288 if (error != 0) { 1289 linux_cdev_handle_free(vmap); 1290 return (error); 1291 } 1292 1293 attr = pgprot2cachemode(vmap->vm_page_prot); 1294 1295 if (vmap->vm_ops != NULL) { 1296 struct vm_area_struct *ptr; 1297 void *vm_private_data; 1298 bool vm_no_fault; 1299 1300 if (vmap->vm_ops->open == NULL || 1301 vmap->vm_ops->close == NULL || 1302 vmap->vm_private_data == NULL) { 1303 /* free allocated VM area struct */ 1304 linux_cdev_handle_free(vmap); 1305 return (EINVAL); 1306 } 1307 1308 vm_private_data = vmap->vm_private_data; 1309 1310 rw_wlock(&linux_vma_lock); 1311 TAILQ_FOREACH(ptr, &linux_vma_head, vm_entry) { 1312 if (ptr->vm_private_data == vm_private_data) 1313 break; 1314 } 1315 /* check if there is an existing VM area struct */ 1316 if (ptr != NULL) { 1317 /* check if the VM area structure is invalid */ 1318 if (ptr->vm_ops == NULL || 1319 ptr->vm_ops->open == NULL || 1320 ptr->vm_ops->close == NULL) { 1321 error = ESTALE; 1322 vm_no_fault = 1; 1323 } else { 1324 error = EEXIST; 1325 vm_no_fault = (ptr->vm_ops->fault == NULL); 1326 } 1327 } else { 1328 /* insert VM area structure into list */ 1329 TAILQ_INSERT_TAIL(&linux_vma_head, vmap, vm_entry); 1330 error = 0; 1331 vm_no_fault = (vmap->vm_ops->fault == NULL); 1332 } 1333 rw_wunlock(&linux_vma_lock); 1334 1335 if (error != 0) { 1336 /* free allocated VM area struct */ 1337 linux_cdev_handle_free(vmap); 1338 /* check for stale VM area struct */ 1339 if (error != EEXIST) 1340 return (error); 1341 } 1342 1343 /* check if there is no fault handler */ 1344 if (vm_no_fault) { 1345 *object = cdev_pager_allocate(vm_private_data, OBJT_DEVICE, 1346 &linux_cdev_pager_ops[1], size, nprot, *offset, 1347 td->td_ucred); 1348 } else { 1349 *object = cdev_pager_allocate(vm_private_data, OBJT_MGTDEVICE, 1350 &linux_cdev_pager_ops[0], size, nprot, *offset, 1351 td->td_ucred); 1352 } 1353 1354 /* check if allocating the VM object failed */ 1355 if (*object == NULL) { 1356 if (error == 0) { 1357 /* remove VM area struct from list */ 1358 linux_cdev_handle_remove(vmap); 1359 /* free allocated VM area struct */ 1360 linux_cdev_handle_free(vmap); 1361 } 1362 return (EINVAL); 1363 } 1364 } else { 1365 struct sglist *sg; 1366 1367 sg = sglist_alloc(1, M_WAITOK); 1368 sglist_append_phys(sg, 1369 (vm_paddr_t)vmap->vm_pfn << PAGE_SHIFT, vmap->vm_len); 1370 1371 *object = vm_pager_allocate(OBJT_SG, sg, vmap->vm_len, 1372 nprot, 0, td->td_ucred); 1373 1374 linux_cdev_handle_free(vmap); 1375 1376 if (*object == NULL) { 1377 sglist_free(sg); 1378 return (EINVAL); 1379 } 1380 } 1381 1382 if (attr != VM_MEMATTR_DEFAULT) { 1383 VM_OBJECT_WLOCK(*object); 1384 vm_object_set_memattr(*object, attr); 1385 VM_OBJECT_WUNLOCK(*object); 1386 } 1387 *offset = 0; 1388 return (0); 1389} 1390 1391struct cdevsw linuxcdevsw = { 1392 .d_version = D_VERSION, 1393 .d_fdopen = linux_dev_fdopen, 1394 .d_name = "lkpidev", 1395}; 1396 1397static int 1398linux_file_read(struct file *file, struct uio *uio, struct ucred *active_cred, 1399 int flags, struct thread *td) 1400{ 1401 struct linux_file *filp; 1402 const struct file_operations *fop; 1403 struct linux_cdev *ldev; 1404 ssize_t bytes; 1405 int error; 1406 1407 error = 0; 1408 filp = (struct linux_file *)file->f_data; 1409 filp->f_flags = file->f_flag; 1410 /* XXX no support for I/O vectors currently */ 1411 if (uio->uio_iovcnt != 1) 1412 return (EOPNOTSUPP); 1413 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1414 return (EINVAL); 1415 linux_set_current(td); 1416 linux_get_fop(filp, &fop, &ldev); 1417 if (fop->read != NULL) { 1418 bytes = OPW(file, td, fop->read(filp, 1419 uio->uio_iov->iov_base, 1420 uio->uio_iov->iov_len, &uio->uio_offset)); 1421 if (bytes >= 0) { 1422 uio->uio_iov->iov_base = 1423 ((uint8_t *)uio->uio_iov->iov_base) + bytes; 1424 uio->uio_iov->iov_len -= bytes; 1425 uio->uio_resid -= bytes; 1426 } else { 1427 error = linux_get_error(current, -bytes); 1428 } 1429 } else 1430 error = ENXIO; 1431 1432 /* update kqfilter status, if any */ 1433 linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_READ); 1434 linux_drop_fop(ldev); 1435 1436 return (error); 1437} 1438 1439static int 1440linux_file_write(struct file *file, struct uio *uio, struct ucred *active_cred, 1441 int flags, struct thread *td) 1442{ 1443 struct linux_file *filp; 1444 const struct file_operations *fop; 1445 struct linux_cdev *ldev; 1446 ssize_t bytes; 1447 int error; 1448 1449 filp = (struct linux_file *)file->f_data; 1450 filp->f_flags = file->f_flag; 1451 /* XXX no support for I/O vectors currently */ 1452 if (uio->uio_iovcnt != 1) 1453 return (EOPNOTSUPP); 1454 if (uio->uio_resid > DEVFS_IOSIZE_MAX) 1455 return (EINVAL); 1456 linux_set_current(td); 1457 linux_get_fop(filp, &fop, &ldev); 1458 if (fop->write != NULL) { 1459 bytes = OPW(file, td, fop->write(filp, 1460 uio->uio_iov->iov_base, 1461 uio->uio_iov->iov_len, &uio->uio_offset)); 1462 if (bytes >= 0) { 1463 uio->uio_iov->iov_base = 1464 ((uint8_t *)uio->uio_iov->iov_base) + bytes; 1465 uio->uio_iov->iov_len -= bytes; 1466 uio->uio_resid -= bytes; 1467 error = 0; 1468 } else { 1469 error = linux_get_error(current, -bytes); 1470 } 1471 } else 1472 error = ENXIO; 1473 1474 /* update kqfilter status, if any */ 1475 linux_file_kqfilter_poll(filp, LINUX_KQ_FLAG_HAS_WRITE); 1476 1477 linux_drop_fop(ldev); 1478 1479 return (error); 1480} 1481 1482static int 1483linux_file_poll(struct file *file, int events, struct ucred *active_cred, 1484 struct thread *td) 1485{ 1486 struct linux_file *filp; 1487 const struct file_operations *fop; 1488 struct linux_cdev *ldev; 1489 int revents; 1490 1491 filp = (struct linux_file *)file->f_data; 1492 filp->f_flags = file->f_flag; 1493 linux_set_current(td); 1494 linux_get_fop(filp, &fop, &ldev); 1495 if (fop->poll != NULL) { 1496 revents = OPW(file, td, fop->poll(filp, 1497 LINUX_POLL_TABLE_NORMAL)) & events; 1498 } else { 1499 revents = 0; 1500 } 1501 linux_drop_fop(ldev); 1502 return (revents); 1503} 1504 1505static int 1506linux_file_close(struct file *file, struct thread *td) 1507{ 1508 struct linux_file *filp; 1509 int (*release)(struct inode *, struct linux_file *); 1510 const struct file_operations *fop; 1511 struct linux_cdev *ldev; 1512 int error; 1513 1514 filp = (struct linux_file *)file->f_data; 1515 1516 KASSERT(file_count(filp) == 0, 1517 ("File refcount(%d) is not zero", file_count(filp))); 1518 1519 if (td == NULL) 1520 td = curthread; 1521 1522 error = 0; 1523 filp->f_flags = file->f_flag; 1524 linux_set_current(td); 1525 linux_poll_wait_dequeue(filp); 1526 linux_get_fop(filp, &fop, &ldev); 1527 /* 1528 * Always use the real release function, if any, to avoid 1529 * leaking device resources: 1530 */ 1531 release = filp->f_op->release; 1532 if (release != NULL) 1533 error = -OPW(file, td, release(filp->f_vnode, filp)); 1534 funsetown(&filp->f_sigio); 1535 if (filp->f_vnode != NULL) 1536 vdrop(filp->f_vnode); 1537 linux_drop_fop(ldev); 1538 ldev = filp->f_cdev; 1539 if (ldev != NULL) 1540 linux_cdev_deref(ldev); 1541 kfree(filp); 1542 1543 return (error); 1544} 1545 1546static int 1547linux_file_ioctl(struct file *fp, u_long cmd, void *data, struct ucred *cred, 1548 struct thread *td) 1549{ 1550 struct linux_file *filp; 1551 const struct file_operations *fop; 1552 struct linux_cdev *ldev; 1553 struct fiodgname_arg *fgn; 1554 const char *p; 1555 int error, i; 1556 1557 error = 0; 1558 filp = (struct linux_file *)fp->f_data; 1559 filp->f_flags = fp->f_flag; 1560 linux_get_fop(filp, &fop, &ldev); 1561 1562 linux_set_current(td); 1563 switch (cmd) { 1564 case FIONBIO: 1565 break; 1566 case FIOASYNC: 1567 if (fop->fasync == NULL) 1568 break; 1569 error = -OPW(fp, td, fop->fasync(0, filp, fp->f_flag & FASYNC)); 1570 break; 1571 case FIOSETOWN: 1572 error = fsetown(*(int *)data, &filp->f_sigio); 1573 if (error == 0) { 1574 if (fop->fasync == NULL) 1575 break; 1576 error = -OPW(fp, td, fop->fasync(0, filp, 1577 fp->f_flag & FASYNC)); 1578 } 1579 break; 1580 case FIOGETOWN: 1581 *(int *)data = fgetown(&filp->f_sigio); 1582 break; 1583 case FIODGNAME: 1584 if (filp->f_cdev == NULL || filp->f_cdev->cdev == NULL) { 1585 error = ENXIO; 1586 break; 1587 } 1588 fgn = data; 1589 p = devtoname(filp->f_cdev->cdev); 1590 i = strlen(p) + 1; 1591 if (i > fgn->len) { 1592 error = EINVAL; 1593 break; 1594 } 1595 error = copyout(p, fgn->buf, i); 1596 break; 1597 default: 1598 error = linux_file_ioctl_sub(fp, filp, fop, cmd, data, td); 1599 break; 1600 } 1601 linux_drop_fop(ldev); 1602 return (error); 1603} 1604 1605static int 1606linux_file_mmap_sub(struct thread *td, vm_size_t objsize, vm_prot_t prot, 1607 vm_prot_t *maxprotp, int *flagsp, struct file *fp, 1608 vm_ooffset_t *foff, const struct file_operations *fop, vm_object_t *objp) 1609{ 1610 /* 1611 * Character devices do not provide private mappings 1612 * of any kind: 1613 */ 1614 if ((*maxprotp & VM_PROT_WRITE) == 0 && 1615 (prot & VM_PROT_WRITE) != 0) 1616 return (EACCES); 1617 if ((*flagsp & (MAP_PRIVATE | MAP_COPY)) != 0) 1618 return (EINVAL); 1619 1620 return (linux_file_mmap_single(fp, fop, foff, objsize, objp, 1621 (int)prot, td)); 1622} 1623 1624static int 1625linux_file_mmap(struct file *fp, vm_map_t map, vm_offset_t *addr, vm_size_t size, 1626 vm_prot_t prot, vm_prot_t cap_maxprot, int flags, vm_ooffset_t foff, 1627 struct thread *td) 1628{ 1629 struct linux_file *filp; 1630 const struct file_operations *fop; 1631 struct linux_cdev *ldev; 1632 struct mount *mp; 1633 struct vnode *vp; 1634 vm_object_t object; 1635 vm_prot_t maxprot; 1636 int error; 1637 1638 filp = (struct linux_file *)fp->f_data; 1639 1640 vp = filp->f_vnode; 1641 if (vp == NULL) 1642 return (EOPNOTSUPP); 1643 1644 /* 1645 * Ensure that file and memory protections are 1646 * compatible. 1647 */ 1648 mp = vp->v_mount; 1649 if (mp != NULL && (mp->mnt_flag & MNT_NOEXEC) != 0) { 1650 maxprot = VM_PROT_NONE; 1651 if ((prot & VM_PROT_EXECUTE) != 0) 1652 return (EACCES); 1653 } else 1654 maxprot = VM_PROT_EXECUTE; 1655 if ((fp->f_flag & FREAD) != 0) 1656 maxprot |= VM_PROT_READ; 1657 else if ((prot & VM_PROT_READ) != 0) 1658 return (EACCES); 1659 1660 /* 1661 * If we are sharing potential changes via MAP_SHARED and we 1662 * are trying to get write permission although we opened it 1663 * without asking for it, bail out. 1664 * 1665 * Note that most character devices always share mappings. 1666 * 1667 * Rely on linux_file_mmap_sub() to fail invalid MAP_PRIVATE 1668 * requests rather than doing it here. 1669 */ 1670 if ((flags & MAP_SHARED) != 0) { 1671 if ((fp->f_flag & FWRITE) != 0) 1672 maxprot |= VM_PROT_WRITE; 1673 else if ((prot & VM_PROT_WRITE) != 0) 1674 return (EACCES); 1675 } 1676 maxprot &= cap_maxprot; 1677 1678 linux_get_fop(filp, &fop, &ldev); 1679 error = linux_file_mmap_sub(td, size, prot, &maxprot, &flags, fp, 1680 &foff, fop, &object); 1681 if (error != 0) 1682 goto out; 1683 1684 error = vm_mmap_object(map, addr, size, prot, maxprot, flags, object, 1685 foff, FALSE, td); 1686 if (error != 0) 1687 vm_object_deallocate(object); 1688out: 1689 linux_drop_fop(ldev); 1690 return (error); 1691} 1692 1693static int 1694linux_file_stat(struct file *fp, struct stat *sb, struct ucred *active_cred, 1695 struct thread *td) 1696{ 1697 struct linux_file *filp; 1698 struct vnode *vp; 1699 int error; 1700 1701 filp = (struct linux_file *)fp->f_data; 1702 if (filp->f_vnode == NULL) 1703 return (EOPNOTSUPP); 1704 1705 vp = filp->f_vnode; 1706 1707 vn_lock(vp, LK_SHARED | LK_RETRY); 1708 error = vn_stat(vp, sb, td->td_ucred, NOCRED, td); 1709 VOP_UNLOCK(vp, 0); 1710 1711 return (error); 1712} 1713 1714static int 1715linux_file_fill_kinfo(struct file *fp, struct kinfo_file *kif, 1716 struct filedesc *fdp) 1717{ 1718 struct linux_file *filp; 1719 struct vnode *vp; 1720 int error; 1721 1722 filp = fp->f_data; 1723 vp = filp->f_vnode; 1724 if (vp == NULL) { 1725 error = 0; 1726 kif->kf_type = KF_TYPE_DEV; 1727 } else { 1728 vref(vp); 1729 FILEDESC_SUNLOCK(fdp); 1730 error = vn_fill_kinfo_vnode(vp, kif); 1731 vrele(vp); 1732 kif->kf_type = KF_TYPE_VNODE; 1733 FILEDESC_SLOCK(fdp); 1734 } 1735 return (error); 1736} 1737 1738unsigned int 1739linux_iminor(struct inode *inode) 1740{ 1741 struct linux_cdev *ldev; 1742 1743 if (inode == NULL || inode->v_rdev == NULL || 1744 inode->v_rdev->si_devsw != &linuxcdevsw) 1745 return (-1U); 1746 ldev = inode->v_rdev->si_drv1; 1747 if (ldev == NULL) 1748 return (-1U); 1749 1750 return (minor(ldev->dev)); 1751} 1752 1753struct fileops linuxfileops = { 1754 .fo_read = linux_file_read, 1755 .fo_write = linux_file_write, 1756 .fo_truncate = invfo_truncate, 1757 .fo_kqfilter = linux_file_kqfilter, 1758 .fo_stat = linux_file_stat, 1759 .fo_fill_kinfo = linux_file_fill_kinfo, 1760 .fo_poll = linux_file_poll, 1761 .fo_close = linux_file_close, 1762 .fo_ioctl = linux_file_ioctl, 1763 .fo_mmap = linux_file_mmap, 1764 .fo_chmod = invfo_chmod, 1765 .fo_chown = invfo_chown, 1766 .fo_sendfile = invfo_sendfile, 1767 .fo_flags = DFLAG_PASSABLE, 1768}; 1769 1770/* 1771 * Hash of vmmap addresses. This is infrequently accessed and does not 1772 * need to be particularly large. This is done because we must store the 1773 * caller's idea of the map size to properly unmap. 1774 */ 1775struct vmmap { 1776 LIST_ENTRY(vmmap) vm_next; 1777 void *vm_addr; 1778 unsigned long vm_size; 1779}; 1780 1781struct vmmaphd { 1782 struct vmmap *lh_first; 1783}; 1784#define VMMAP_HASH_SIZE 64 1785#define VMMAP_HASH_MASK (VMMAP_HASH_SIZE - 1) 1786#define VM_HASH(addr) ((uintptr_t)(addr) >> PAGE_SHIFT) & VMMAP_HASH_MASK 1787static struct vmmaphd vmmaphead[VMMAP_HASH_SIZE]; 1788static struct mtx vmmaplock; 1789 1790static void 1791vmmap_add(void *addr, unsigned long size) 1792{ 1793 struct vmmap *vmmap; 1794 1795 vmmap = kmalloc(sizeof(*vmmap), GFP_KERNEL); 1796 mtx_lock(&vmmaplock); 1797 vmmap->vm_size = size; 1798 vmmap->vm_addr = addr; 1799 LIST_INSERT_HEAD(&vmmaphead[VM_HASH(addr)], vmmap, vm_next); 1800 mtx_unlock(&vmmaplock); 1801} 1802 1803static struct vmmap * 1804vmmap_remove(void *addr) 1805{ 1806 struct vmmap *vmmap; 1807 1808 mtx_lock(&vmmaplock); 1809 LIST_FOREACH(vmmap, &vmmaphead[VM_HASH(addr)], vm_next) 1810 if (vmmap->vm_addr == addr) 1811 break; 1812 if (vmmap) 1813 LIST_REMOVE(vmmap, vm_next); 1814 mtx_unlock(&vmmaplock); 1815 1816 return (vmmap); 1817} 1818 1819#if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) 1820void * 1821_ioremap_attr(vm_paddr_t phys_addr, unsigned long size, int attr) 1822{ 1823 void *addr; 1824 1825 addr = pmap_mapdev_attr(phys_addr, size, attr); 1826 if (addr == NULL) 1827 return (NULL); 1828 vmmap_add(addr, size); 1829 1830 return (addr); 1831} 1832#endif 1833 1834void 1835iounmap(void *addr) 1836{ 1837 struct vmmap *vmmap; 1838 1839 vmmap = vmmap_remove(addr); 1840 if (vmmap == NULL) 1841 return; 1842#if defined(__i386__) || defined(__amd64__) || defined(__powerpc__) || defined(__aarch64__) 1843 pmap_unmapdev((vm_offset_t)addr, vmmap->vm_size); 1844#endif 1845 kfree(vmmap); 1846} 1847 1848 1849void * 1850vmap(struct page **pages, unsigned int count, unsigned long flags, int prot) 1851{ 1852 vm_offset_t off; 1853 size_t size; 1854 1855 size = count * PAGE_SIZE; 1856 off = kva_alloc(size); 1857 if (off == 0) 1858 return (NULL); 1859 vmmap_add((void *)off, size); 1860 pmap_qenter(off, pages, count); 1861 1862 return ((void *)off); 1863} 1864 1865void 1866vunmap(void *addr) 1867{ 1868 struct vmmap *vmmap; 1869 1870 vmmap = vmmap_remove(addr); 1871 if (vmmap == NULL) 1872 return; 1873 pmap_qremove((vm_offset_t)addr, vmmap->vm_size / PAGE_SIZE); 1874 kva_free((vm_offset_t)addr, vmmap->vm_size); 1875 kfree(vmmap); 1876} 1877 1878char * 1879kvasprintf(gfp_t gfp, const char *fmt, va_list ap) 1880{ 1881 unsigned int len; 1882 char *p; 1883 va_list aq; 1884 1885 va_copy(aq, ap); 1886 len = vsnprintf(NULL, 0, fmt, aq); 1887 va_end(aq); 1888 1889 p = kmalloc(len + 1, gfp); 1890 if (p != NULL) 1891 vsnprintf(p, len + 1, fmt, ap); 1892 1893 return (p); 1894} 1895 1896char * 1897kasprintf(gfp_t gfp, const char *fmt, ...) 1898{ 1899 va_list ap; 1900 char *p; 1901 1902 va_start(ap, fmt); 1903 p = kvasprintf(gfp, fmt, ap); 1904 va_end(ap); 1905 1906 return (p); 1907} 1908 1909static void 1910linux_timer_callback_wrapper(void *context) 1911{ 1912 struct timer_list *timer; 1913 1914 timer = context; 1915 1916 if (linux_set_current_flags(curthread, M_NOWAIT)) { 1917 /* try again later */ 1918 callout_reset(&timer->callout, 1, 1919 &linux_timer_callback_wrapper, timer); 1920 return; 1921 } 1922 1923 timer->function(timer->data); 1924} 1925 1926int 1927mod_timer(struct timer_list *timer, int expires) 1928{ 1929 int ret; 1930 1931 timer->expires = expires; 1932 ret = callout_reset(&timer->callout, 1933 linux_timer_jiffies_until(expires), 1934 &linux_timer_callback_wrapper, timer); 1935 1936 MPASS(ret == 0 || ret == 1); 1937 1938 return (ret == 1); 1939} 1940 1941void 1942add_timer(struct timer_list *timer) 1943{ 1944 1945 callout_reset(&timer->callout, 1946 linux_timer_jiffies_until(timer->expires), 1947 &linux_timer_callback_wrapper, timer); 1948} 1949 1950void 1951add_timer_on(struct timer_list *timer, int cpu) 1952{ 1953 1954 callout_reset_on(&timer->callout, 1955 linux_timer_jiffies_until(timer->expires), 1956 &linux_timer_callback_wrapper, timer, cpu); 1957} 1958 1959int 1960del_timer(struct timer_list *timer) 1961{ 1962 1963 if (callout_stop(&(timer)->callout) == -1) 1964 return (0); 1965 return (1); 1966} 1967 1968int 1969del_timer_sync(struct timer_list *timer) 1970{ 1971 1972 if (callout_drain(&(timer)->callout) == -1) 1973 return (0); 1974 return (1); 1975} 1976 1977/* greatest common divisor, Euclid equation */ 1978static uint64_t 1979lkpi_gcd_64(uint64_t a, uint64_t b) 1980{ 1981 uint64_t an; 1982 uint64_t bn; 1983 1984 while (b != 0) { 1985 an = b; 1986 bn = a % b; 1987 a = an; 1988 b = bn; 1989 } 1990 return (a); 1991} 1992 1993uint64_t lkpi_nsec2hz_rem; 1994uint64_t lkpi_nsec2hz_div = 1000000000ULL; 1995uint64_t lkpi_nsec2hz_max; 1996 1997uint64_t lkpi_usec2hz_rem; 1998uint64_t lkpi_usec2hz_div = 1000000ULL; 1999uint64_t lkpi_usec2hz_max; 2000 2001uint64_t lkpi_msec2hz_rem; 2002uint64_t lkpi_msec2hz_div = 1000ULL; 2003uint64_t lkpi_msec2hz_max; 2004 2005static void 2006linux_timer_init(void *arg) 2007{ 2008 uint64_t gcd; 2009 2010 /* 2011 * Compute an internal HZ value which can divide 2**32 to 2012 * avoid timer rounding problems when the tick value wraps 2013 * around 2**32: 2014 */ 2015 linux_timer_hz_mask = 1; 2016 while (linux_timer_hz_mask < (unsigned long)hz) 2017 linux_timer_hz_mask *= 2; 2018 linux_timer_hz_mask--; 2019 2020 /* compute some internal constants */ 2021 2022 lkpi_nsec2hz_rem = hz; 2023 lkpi_usec2hz_rem = hz; 2024 lkpi_msec2hz_rem = hz; 2025 2026 gcd = lkpi_gcd_64(lkpi_nsec2hz_rem, lkpi_nsec2hz_div); 2027 lkpi_nsec2hz_rem /= gcd; 2028 lkpi_nsec2hz_div /= gcd; 2029 lkpi_nsec2hz_max = -1ULL / lkpi_nsec2hz_rem; 2030 2031 gcd = lkpi_gcd_64(lkpi_usec2hz_rem, lkpi_usec2hz_div); 2032 lkpi_usec2hz_rem /= gcd; 2033 lkpi_usec2hz_div /= gcd; 2034 lkpi_usec2hz_max = -1ULL / lkpi_usec2hz_rem; 2035 2036 gcd = lkpi_gcd_64(lkpi_msec2hz_rem, lkpi_msec2hz_div); 2037 lkpi_msec2hz_rem /= gcd; 2038 lkpi_msec2hz_div /= gcd; 2039 lkpi_msec2hz_max = -1ULL / lkpi_msec2hz_rem; 2040} 2041SYSINIT(linux_timer, SI_SUB_DRIVERS, SI_ORDER_FIRST, linux_timer_init, NULL); 2042 2043void 2044linux_complete_common(struct completion *c, int all) 2045{ 2046 int wakeup_swapper; 2047 2048 sleepq_lock(c); 2049 if (all) { 2050 c->done = UINT_MAX; 2051 wakeup_swapper = sleepq_broadcast(c, SLEEPQ_SLEEP, 0, 0); 2052 } else { 2053 if (c->done != UINT_MAX) 2054 c->done++; 2055 wakeup_swapper = sleepq_signal(c, SLEEPQ_SLEEP, 0, 0); 2056 } 2057 sleepq_release(c); 2058 if (wakeup_swapper) 2059 kick_proc0(); 2060} 2061 2062/* 2063 * Indefinite wait for done != 0 with or without signals. 2064 */ 2065int 2066linux_wait_for_common(struct completion *c, int flags) 2067{ 2068 struct task_struct *task; 2069 int error; 2070 2071 if (SCHEDULER_STOPPED()) 2072 return (0); 2073 2074 task = current; 2075 2076 if (flags != 0) 2077 flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP; 2078 else 2079 flags = SLEEPQ_SLEEP; 2080 error = 0; 2081 for (;;) { 2082 sleepq_lock(c); 2083 if (c->done) 2084 break; 2085 sleepq_add(c, NULL, "completion", flags, 0); 2086 if (flags & SLEEPQ_INTERRUPTIBLE) { 2087 DROP_GIANT(); 2088 error = -sleepq_wait_sig(c, 0); 2089 PICKUP_GIANT(); 2090 if (error != 0) { 2091 linux_schedule_save_interrupt_value(task, error); 2092 error = -ERESTARTSYS; 2093 goto intr; 2094 } 2095 } else { 2096 DROP_GIANT(); 2097 sleepq_wait(c, 0); 2098 PICKUP_GIANT(); 2099 } 2100 } 2101 if (c->done != UINT_MAX) 2102 c->done--; 2103 sleepq_release(c); 2104 2105intr: 2106 return (error); 2107} 2108 2109/* 2110 * Time limited wait for done != 0 with or without signals. 2111 */ 2112int 2113linux_wait_for_timeout_common(struct completion *c, int timeout, int flags) 2114{ 2115 struct task_struct *task; 2116 int end = jiffies + timeout; 2117 int error; 2118 2119 if (SCHEDULER_STOPPED()) 2120 return (0); 2121 2122 task = current; 2123 2124 if (flags != 0) 2125 flags = SLEEPQ_INTERRUPTIBLE | SLEEPQ_SLEEP; 2126 else 2127 flags = SLEEPQ_SLEEP; 2128 2129 for (;;) { 2130 sleepq_lock(c); 2131 if (c->done) 2132 break; 2133 sleepq_add(c, NULL, "completion", flags, 0); 2134 sleepq_set_timeout(c, linux_timer_jiffies_until(end)); 2135 2136 DROP_GIANT(); 2137 if (flags & SLEEPQ_INTERRUPTIBLE) 2138 error = -sleepq_timedwait_sig(c, 0); 2139 else 2140 error = -sleepq_timedwait(c, 0); 2141 PICKUP_GIANT(); 2142 2143 if (error != 0) { 2144 /* check for timeout */ 2145 if (error == -EWOULDBLOCK) { 2146 error = 0; /* timeout */ 2147 } else { 2148 /* signal happened */ 2149 linux_schedule_save_interrupt_value(task, error); 2150 error = -ERESTARTSYS; 2151 } 2152 goto done; 2153 } 2154 } 2155 if (c->done != UINT_MAX) 2156 c->done--; 2157 sleepq_release(c); 2158 2159 /* return how many jiffies are left */ 2160 error = linux_timer_jiffies_until(end); 2161done: 2162 return (error); 2163} 2164 2165int 2166linux_try_wait_for_completion(struct completion *c) 2167{ 2168 int isdone; 2169 2170 sleepq_lock(c); 2171 isdone = (c->done != 0); 2172 if (c->done != 0 && c->done != UINT_MAX) 2173 c->done--; 2174 sleepq_release(c); 2175 return (isdone); 2176} 2177 2178int 2179linux_completion_done(struct completion *c) 2180{ 2181 int isdone; 2182 2183 sleepq_lock(c); 2184 isdone = (c->done != 0); 2185 sleepq_release(c); 2186 return (isdone); 2187} 2188 2189static void 2190linux_cdev_deref(struct linux_cdev *ldev) 2191{ 2192 if (refcount_release(&ldev->refs) && 2193 ldev->kobj.ktype == &linux_cdev_ktype) 2194 kfree(ldev); 2195} 2196 2197static void 2198linux_cdev_release(struct kobject *kobj) 2199{ 2200 struct linux_cdev *cdev; 2201 struct kobject *parent; 2202 2203 cdev = container_of(kobj, struct linux_cdev, kobj); 2204 parent = kobj->parent; 2205 linux_destroy_dev(cdev); 2206 linux_cdev_deref(cdev); 2207 kobject_put(parent); 2208} 2209 2210static void 2211linux_cdev_static_release(struct kobject *kobj) 2212{ 2213 struct cdev *cdev; 2214 struct linux_cdev *ldev; 2215 2216 ldev = container_of(kobj, struct linux_cdev, kobj); 2217 cdev = ldev->cdev; 2218 if (cdev != NULL) { 2219 destroy_dev(cdev); 2220 ldev->cdev = NULL; 2221 } 2222 kobject_put(kobj->parent); 2223} 2224 2225void 2226linux_destroy_dev(struct linux_cdev *ldev) 2227{ 2228 2229 if (ldev->cdev == NULL) 2230 return; 2231 2232 MPASS((ldev->siref & LDEV_SI_DTR) == 0); 2233 MPASS(ldev->kobj.ktype == &linux_cdev_ktype); 2234 2235 atomic_set_int(&ldev->siref, LDEV_SI_DTR); 2236 while ((atomic_load_int(&ldev->siref) & ~LDEV_SI_DTR) != 0) 2237 pause("ldevdtr", hz / 4); 2238 2239 destroy_dev(ldev->cdev); 2240 ldev->cdev = NULL; 2241} 2242 2243const struct kobj_type linux_cdev_ktype = { 2244 .release = linux_cdev_release, 2245}; 2246 2247const struct kobj_type linux_cdev_static_ktype = { 2248 .release = linux_cdev_static_release, 2249}; 2250 2251static void 2252linux_handle_ifnet_link_event(void *arg, struct ifnet *ifp, int linkstate) 2253{ 2254 struct notifier_block *nb; 2255 2256 nb = arg; 2257 if (linkstate == LINK_STATE_UP) 2258 nb->notifier_call(nb, NETDEV_UP, ifp); 2259 else 2260 nb->notifier_call(nb, NETDEV_DOWN, ifp); 2261} 2262 2263static void 2264linux_handle_ifnet_arrival_event(void *arg, struct ifnet *ifp) 2265{ 2266 struct notifier_block *nb; 2267 2268 nb = arg; 2269 nb->notifier_call(nb, NETDEV_REGISTER, ifp); 2270} 2271 2272static void 2273linux_handle_ifnet_departure_event(void *arg, struct ifnet *ifp) 2274{ 2275 struct notifier_block *nb; 2276 2277 nb = arg; 2278 nb->notifier_call(nb, NETDEV_UNREGISTER, ifp); 2279} 2280 2281static void 2282linux_handle_iflladdr_event(void *arg, struct ifnet *ifp) 2283{ 2284 struct notifier_block *nb; 2285 2286 nb = arg; 2287 nb->notifier_call(nb, NETDEV_CHANGEADDR, ifp); 2288} 2289 2290static void 2291linux_handle_ifaddr_event(void *arg, struct ifnet *ifp) 2292{ 2293 struct notifier_block *nb; 2294 2295 nb = arg; 2296 nb->notifier_call(nb, NETDEV_CHANGEIFADDR, ifp); 2297} 2298 2299int 2300register_netdevice_notifier(struct notifier_block *nb) 2301{ 2302 2303 nb->tags[NETDEV_UP] = EVENTHANDLER_REGISTER( 2304 ifnet_link_event, linux_handle_ifnet_link_event, nb, 0); 2305 nb->tags[NETDEV_REGISTER] = EVENTHANDLER_REGISTER( 2306 ifnet_arrival_event, linux_handle_ifnet_arrival_event, nb, 0); 2307 nb->tags[NETDEV_UNREGISTER] = EVENTHANDLER_REGISTER( 2308 ifnet_departure_event, linux_handle_ifnet_departure_event, nb, 0); 2309 nb->tags[NETDEV_CHANGEADDR] = EVENTHANDLER_REGISTER( 2310 iflladdr_event, linux_handle_iflladdr_event, nb, 0); 2311 2312 return (0); 2313} 2314 2315int 2316register_inetaddr_notifier(struct notifier_block *nb) 2317{ 2318 2319 nb->tags[NETDEV_CHANGEIFADDR] = EVENTHANDLER_REGISTER( 2320 ifaddr_event, linux_handle_ifaddr_event, nb, 0); 2321 return (0); 2322} 2323 2324int 2325unregister_netdevice_notifier(struct notifier_block *nb) 2326{ 2327 2328 EVENTHANDLER_DEREGISTER(ifnet_link_event, 2329 nb->tags[NETDEV_UP]); 2330 EVENTHANDLER_DEREGISTER(ifnet_arrival_event, 2331 nb->tags[NETDEV_REGISTER]); 2332 EVENTHANDLER_DEREGISTER(ifnet_departure_event, 2333 nb->tags[NETDEV_UNREGISTER]); 2334 EVENTHANDLER_DEREGISTER(iflladdr_event, 2335 nb->tags[NETDEV_CHANGEADDR]); 2336 2337 return (0); 2338} 2339 2340int 2341unregister_inetaddr_notifier(struct notifier_block *nb) 2342{ 2343 2344 EVENTHANDLER_DEREGISTER(ifaddr_event, 2345 nb->tags[NETDEV_CHANGEIFADDR]); 2346 2347 return (0); 2348} 2349 2350struct list_sort_thunk { 2351 int (*cmp)(void *, struct list_head *, struct list_head *); 2352 void *priv; 2353}; 2354 2355static inline int 2356linux_le_cmp(void *priv, const void *d1, const void *d2) 2357{ 2358 struct list_head *le1, *le2; 2359 struct list_sort_thunk *thunk; 2360 2361 thunk = priv; 2362 le1 = *(__DECONST(struct list_head **, d1)); 2363 le2 = *(__DECONST(struct list_head **, d2)); 2364 return ((thunk->cmp)(thunk->priv, le1, le2)); 2365} 2366 2367void 2368list_sort(void *priv, struct list_head *head, int (*cmp)(void *priv, 2369 struct list_head *a, struct list_head *b)) 2370{ 2371 struct list_sort_thunk thunk; 2372 struct list_head **ar, *le; 2373 size_t count, i; 2374 2375 count = 0; 2376 list_for_each(le, head) 2377 count++; 2378 ar = malloc(sizeof(struct list_head *) * count, M_KMALLOC, M_WAITOK); 2379 i = 0; 2380 list_for_each(le, head) 2381 ar[i++] = le; 2382 thunk.cmp = cmp; 2383 thunk.priv = priv; 2384 qsort_r(ar, count, sizeof(struct list_head *), &thunk, linux_le_cmp); 2385 INIT_LIST_HEAD(head); 2386 for (i = 0; i < count; i++) 2387 list_add_tail(ar[i], head); 2388 free(ar, M_KMALLOC); 2389} 2390 2391void 2392linux_irq_handler(void *ent) 2393{ 2394 struct irq_ent *irqe; 2395 2396 if (linux_set_current_flags(curthread, M_NOWAIT)) 2397 return; 2398 2399 irqe = ent; 2400 irqe->handler(irqe->irq, irqe->arg); 2401} 2402 2403#if defined(__i386__) || defined(__amd64__) 2404int 2405linux_wbinvd_on_all_cpus(void) 2406{ 2407 2408 pmap_invalidate_cache(); 2409 return (0); 2410} 2411#endif 2412 2413int 2414linux_on_each_cpu(void callback(void *), void *data) 2415{ 2416 2417 smp_rendezvous(smp_no_rendezvous_barrier, callback, 2418 smp_no_rendezvous_barrier, data); 2419 return (0); 2420} 2421 2422int 2423linux_in_atomic(void) 2424{ 2425 2426 return ((curthread->td_pflags & TDP_NOFAULTING) != 0); 2427} 2428 2429struct linux_cdev * 2430linux_find_cdev(const char *name, unsigned major, unsigned minor) 2431{ 2432 dev_t dev = MKDEV(major, minor); 2433 struct cdev *cdev; 2434 2435 dev_lock(); 2436 LIST_FOREACH(cdev, &linuxcdevsw.d_devs, si_list) { 2437 struct linux_cdev *ldev = cdev->si_drv1; 2438 if (ldev->dev == dev && 2439 strcmp(kobject_name(&ldev->kobj), name) == 0) { 2440 break; 2441 } 2442 } 2443 dev_unlock(); 2444 2445 return (cdev != NULL ? cdev->si_drv1 : NULL); 2446} 2447 2448int 2449__register_chrdev(unsigned int major, unsigned int baseminor, 2450 unsigned int count, const char *name, 2451 const struct file_operations *fops) 2452{ 2453 struct linux_cdev *cdev; 2454 int ret = 0; 2455 int i; 2456 2457 for (i = baseminor; i < baseminor + count; i++) { 2458 cdev = cdev_alloc(); 2459 cdev->ops = fops; 2460 kobject_set_name(&cdev->kobj, name); 2461 2462 ret = cdev_add(cdev, makedev(major, i), 1); 2463 if (ret != 0) 2464 break; 2465 } 2466 return (ret); 2467} 2468 2469int 2470__register_chrdev_p(unsigned int major, unsigned int baseminor, 2471 unsigned int count, const char *name, 2472 const struct file_operations *fops, uid_t uid, 2473 gid_t gid, int mode) 2474{ 2475 struct linux_cdev *cdev; 2476 int ret = 0; 2477 int i; 2478 2479 for (i = baseminor; i < baseminor + count; i++) { 2480 cdev = cdev_alloc(); 2481 cdev->ops = fops; 2482 kobject_set_name(&cdev->kobj, name); 2483 2484 ret = cdev_add_ext(cdev, makedev(major, i), uid, gid, mode); 2485 if (ret != 0) 2486 break; 2487 } 2488 return (ret); 2489} 2490 2491void 2492__unregister_chrdev(unsigned int major, unsigned int baseminor, 2493 unsigned int count, const char *name) 2494{ 2495 struct linux_cdev *cdevp; 2496 int i; 2497 2498 for (i = baseminor; i < baseminor + count; i++) { 2499 cdevp = linux_find_cdev(name, major, i); 2500 if (cdevp != NULL) 2501 cdev_del(cdevp); 2502 } 2503} 2504 2505void 2506linux_dump_stack(void) 2507{ 2508#ifdef STACK 2509 struct stack st; 2510 2511 stack_zero(&st); 2512 stack_save(&st); 2513 stack_print(&st); 2514#endif 2515} 2516 2517#if defined(__i386__) || defined(__amd64__) 2518bool linux_cpu_has_clflush; 2519#endif 2520 2521static void 2522linux_compat_init(void *arg) 2523{ 2524 struct sysctl_oid *rootoid; 2525 int i; 2526 2527#if defined(__i386__) || defined(__amd64__) 2528 linux_cpu_has_clflush = (cpu_feature & CPUID_CLFSH); 2529#endif 2530 rw_init(&linux_vma_lock, "lkpi-vma-lock"); 2531 2532 rootoid = SYSCTL_ADD_ROOT_NODE(NULL, 2533 OID_AUTO, "sys", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "sys"); 2534 kobject_init(&linux_class_root, &linux_class_ktype); 2535 kobject_set_name(&linux_class_root, "class"); 2536 linux_class_root.oidp = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(rootoid), 2537 OID_AUTO, "class", CTLFLAG_RD|CTLFLAG_MPSAFE, NULL, "class"); 2538 kobject_init(&linux_root_device.kobj, &linux_dev_ktype); 2539 kobject_set_name(&linux_root_device.kobj, "device"); 2540 linux_root_device.kobj.oidp = SYSCTL_ADD_NODE(NULL, 2541 SYSCTL_CHILDREN(rootoid), OID_AUTO, "device", CTLFLAG_RD, NULL, 2542 "device"); 2543 linux_root_device.bsddev = root_bus; 2544 linux_class_misc.name = "misc"; 2545 class_register(&linux_class_misc); 2546 INIT_LIST_HEAD(&pci_drivers); 2547 INIT_LIST_HEAD(&pci_devices); 2548 spin_lock_init(&pci_lock); 2549 mtx_init(&vmmaplock, "IO Map lock", NULL, MTX_DEF); 2550 for (i = 0; i < VMMAP_HASH_SIZE; i++) 2551 LIST_INIT(&vmmaphead[i]); 2552 init_waitqueue_head(&linux_bit_waitq); 2553 init_waitqueue_head(&linux_var_waitq); 2554} 2555SYSINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_init, NULL); 2556 2557static void 2558linux_compat_uninit(void *arg) 2559{ 2560 linux_kobject_kfree_name(&linux_class_root); 2561 linux_kobject_kfree_name(&linux_root_device.kobj); 2562 linux_kobject_kfree_name(&linux_class_misc.kobj); 2563 2564 mtx_destroy(&vmmaplock); 2565 spin_lock_destroy(&pci_lock); 2566 rw_destroy(&linux_vma_lock); 2567} 2568SYSUNINIT(linux_compat, SI_SUB_DRIVERS, SI_ORDER_SECOND, linux_compat_uninit, NULL); 2569 2570/* 2571 * NOTE: Linux frequently uses "unsigned long" for pointer to integer 2572 * conversion and vice versa, where in FreeBSD "uintptr_t" would be 2573 * used. Assert these types have the same size, else some parts of the 2574 * LinuxKPI may not work like expected: 2575 */ 2576CTASSERT(sizeof(unsigned long) == sizeof(uintptr_t)); 2577