1 /****************************************************************************** 2 * gntdev.c 3 * 4 * Device for accessing (in user-space) pages that have been granted by other 5 * domains. 6 * 7 * Copyright (c) 2006-2007, D G Murray. 8 * (c) 2009 Gerd Hoffmann <kraxel@redhat.com> 9 * 10 * This program is distributed in the hope that it will be useful, 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 13 * GNU General Public License for more details. 14 * 15 * You should have received a copy of the GNU General Public License 16 * along with this program; if not, write to the Free Software 17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA 18 */ 19 20 #undef DEBUG 21 22 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt 23 24 #include <linux/module.h> 25 #include <linux/kernel.h> 26 #include <linux/init.h> 27 #include <linux/miscdevice.h> 28 #include <linux/fs.h> 29 #include <linux/mm.h> 30 #include <linux/mman.h> 31 #include <linux/mmu_notifier.h> 32 #include <linux/types.h> 33 #include <linux/uaccess.h> 34 #include <linux/sched.h> 35 #include <linux/spinlock.h> 36 #include <linux/slab.h> 37 #include <linux/highmem.h> 38 39 #include <xen/xen.h> 40 #include <xen/grant_table.h> 41 #include <xen/balloon.h> 42 #include <xen/gntdev.h> 43 #include <xen/events.h> 44 #include <asm/xen/hypervisor.h> 45 #include <asm/xen/hypercall.h> 46 #include <asm/xen/page.h> 47 48 MODULE_LICENSE("GPL"); 49 MODULE_AUTHOR("Derek G. Murray <Derek.Murray@cl.cam.ac.uk>, " 50 "Gerd Hoffmann <kraxel@redhat.com>"); 51 MODULE_DESCRIPTION("User-space granted page access driver"); 52 53 static int limit = 1024*1024; 54 module_param(limit, int, 0644); 55 MODULE_PARM_DESC(limit, "Maximum number of grants that may be mapped by " 56 "the gntdev device"); 57 58 static atomic_t pages_mapped = ATOMIC_INIT(0); 59 60 static int use_ptemod; 61 #define populate_freeable_maps use_ptemod 62 63 struct gntdev_priv { 64 /* maps with visible offsets in the file descriptor */ 65 struct list_head maps; 66 /* maps that are not visible; will be freed on munmap. 67 * Only populated if populate_freeable_maps == 1 */ 68 struct list_head freeable_maps; 69 /* lock protects maps and freeable_maps */ 70 struct mutex lock; 71 struct mm_struct *mm; 72 struct mmu_notifier mn; 73 }; 74 75 struct unmap_notify { 76 int flags; 77 /* Address relative to the start of the grant_map */ 78 int addr; 79 int event; 80 }; 81 82 struct grant_map { 83 struct list_head next; 84 struct vm_area_struct *vma; 85 int index; 86 int count; 87 int flags; 88 atomic_t users; 89 struct unmap_notify notify; 90 struct ioctl_gntdev_grant_ref *grants; 91 struct gnttab_map_grant_ref *map_ops; 92 struct gnttab_unmap_grant_ref *unmap_ops; 93 struct gnttab_map_grant_ref *kmap_ops; 94 struct gnttab_unmap_grant_ref *kunmap_ops; 95 struct page **pages; 96 unsigned long pages_vm_start; 97 }; 98 99 static int unmap_grant_pages(struct grant_map *map, int offset, int pages); 100 101 /* ------------------------------------------------------------------ */ 102 103 static void gntdev_print_maps(struct gntdev_priv *priv, 104 char *text, int text_index) 105 { 106 #ifdef DEBUG 107 struct grant_map *map; 108 109 pr_debug("%s: maps list (priv %p)\n", __func__, priv); 110 list_for_each_entry(map, &priv->maps, next) 111 pr_debug(" index %2d, count %2d %s\n", 112 map->index, map->count, 113 map->index == text_index && text ? text : ""); 114 #endif 115 } 116 117 static void gntdev_free_map(struct grant_map *map) 118 { 119 if (map == NULL) 120 return; 121 122 if (map->pages) 123 gnttab_free_pages(map->count, map->pages); 124 kfree(map->pages); 125 kfree(map->grants); 126 kfree(map->map_ops); 127 kfree(map->unmap_ops); 128 kfree(map->kmap_ops); 129 kfree(map->kunmap_ops); 130 kfree(map); 131 } 132 133 static struct grant_map *gntdev_alloc_map(struct gntdev_priv *priv, int count) 134 { 135 struct grant_map *add; 136 int i; 137 138 add = kzalloc(sizeof(struct grant_map), GFP_KERNEL); 139 if (NULL == add) 140 return NULL; 141 142 add->grants = kcalloc(count, sizeof(add->grants[0]), GFP_KERNEL); 143 add->map_ops = kcalloc(count, sizeof(add->map_ops[0]), GFP_KERNEL); 144 add->unmap_ops = kcalloc(count, sizeof(add->unmap_ops[0]), GFP_KERNEL); 145 add->kmap_ops = kcalloc(count, sizeof(add->kmap_ops[0]), GFP_KERNEL); 146 add->kunmap_ops = kcalloc(count, sizeof(add->kunmap_ops[0]), GFP_KERNEL); 147 add->pages = kcalloc(count, sizeof(add->pages[0]), GFP_KERNEL); 148 if (NULL == add->grants || 149 NULL == add->map_ops || 150 NULL == add->unmap_ops || 151 NULL == add->kmap_ops || 152 NULL == add->kunmap_ops || 153 NULL == add->pages) 154 goto err; 155 156 if (gnttab_alloc_pages(count, add->pages)) 157 goto err; 158 159 for (i = 0; i < count; i++) { 160 add->map_ops[i].handle = -1; 161 add->unmap_ops[i].handle = -1; 162 add->kmap_ops[i].handle = -1; 163 add->kunmap_ops[i].handle = -1; 164 } 165 166 add->index = 0; 167 add->count = count; 168 atomic_set(&add->users, 1); 169 170 return add; 171 172 err: 173 gntdev_free_map(add); 174 return NULL; 175 } 176 177 static void gntdev_add_map(struct gntdev_priv *priv, struct grant_map *add) 178 { 179 struct grant_map *map; 180 181 list_for_each_entry(map, &priv->maps, next) { 182 if (add->index + add->count < map->index) { 183 list_add_tail(&add->next, &map->next); 184 goto done; 185 } 186 add->index = map->index + map->count; 187 } 188 list_add_tail(&add->next, &priv->maps); 189 190 done: 191 gntdev_print_maps(priv, "[new]", add->index); 192 } 193 194 static struct grant_map *gntdev_find_map_index(struct gntdev_priv *priv, 195 int index, int count) 196 { 197 struct grant_map *map; 198 199 list_for_each_entry(map, &priv->maps, next) { 200 if (map->index != index) 201 continue; 202 if (count && map->count != count) 203 continue; 204 return map; 205 } 206 return NULL; 207 } 208 209 static void gntdev_put_map(struct gntdev_priv *priv, struct grant_map *map) 210 { 211 if (!map) 212 return; 213 214 if (!atomic_dec_and_test(&map->users)) 215 return; 216 217 atomic_sub(map->count, &pages_mapped); 218 219 if (map->notify.flags & UNMAP_NOTIFY_SEND_EVENT) { 220 notify_remote_via_evtchn(map->notify.event); 221 evtchn_put(map->notify.event); 222 } 223 224 if (populate_freeable_maps && priv) { 225 mutex_lock(&priv->lock); 226 list_del(&map->next); 227 mutex_unlock(&priv->lock); 228 } 229 230 if (map->pages && !use_ptemod) 231 unmap_grant_pages(map, 0, map->count); 232 gntdev_free_map(map); 233 } 234 235 /* ------------------------------------------------------------------ */ 236 237 static int find_grant_ptes(pte_t *pte, pgtable_t token, 238 unsigned long addr, void *data) 239 { 240 struct grant_map *map = data; 241 unsigned int pgnr = (addr - map->vma->vm_start) >> PAGE_SHIFT; 242 int flags = map->flags | GNTMAP_application_map | GNTMAP_contains_pte; 243 u64 pte_maddr; 244 245 BUG_ON(pgnr >= map->count); 246 pte_maddr = arbitrary_virt_to_machine(pte).maddr; 247 248 /* 249 * Set the PTE as special to force get_user_pages_fast() fall 250 * back to the slow path. If this is not supported as part of 251 * the grant map, it will be done afterwards. 252 */ 253 if (xen_feature(XENFEAT_gnttab_map_avail_bits)) 254 flags |= (1 << _GNTMAP_guest_avail0); 255 256 gnttab_set_map_op(&map->map_ops[pgnr], pte_maddr, flags, 257 map->grants[pgnr].ref, 258 map->grants[pgnr].domid); 259 gnttab_set_unmap_op(&map->unmap_ops[pgnr], pte_maddr, flags, 260 -1 /* handle */); 261 return 0; 262 } 263 264 #ifdef CONFIG_X86 265 static int set_grant_ptes_as_special(pte_t *pte, pgtable_t token, 266 unsigned long addr, void *data) 267 { 268 set_pte_at(current->mm, addr, pte, pte_mkspecial(*pte)); 269 return 0; 270 } 271 #endif 272 273 static int map_grant_pages(struct grant_map *map) 274 { 275 int i, err = 0; 276 277 if (!use_ptemod) { 278 /* Note: it could already be mapped */ 279 if (map->map_ops[0].handle != -1) 280 return 0; 281 for (i = 0; i < map->count; i++) { 282 unsigned long addr = (unsigned long) 283 pfn_to_kaddr(page_to_pfn(map->pages[i])); 284 gnttab_set_map_op(&map->map_ops[i], addr, map->flags, 285 map->grants[i].ref, 286 map->grants[i].domid); 287 gnttab_set_unmap_op(&map->unmap_ops[i], addr, 288 map->flags, -1 /* handle */); 289 } 290 } else { 291 /* 292 * Setup the map_ops corresponding to the pte entries pointing 293 * to the kernel linear addresses of the struct pages. 294 * These ptes are completely different from the user ptes dealt 295 * with find_grant_ptes. 296 */ 297 for (i = 0; i < map->count; i++) { 298 unsigned long address = (unsigned long) 299 pfn_to_kaddr(page_to_pfn(map->pages[i])); 300 BUG_ON(PageHighMem(map->pages[i])); 301 302 gnttab_set_map_op(&map->kmap_ops[i], address, 303 map->flags | GNTMAP_host_map, 304 map->grants[i].ref, 305 map->grants[i].domid); 306 gnttab_set_unmap_op(&map->kunmap_ops[i], address, 307 map->flags | GNTMAP_host_map, -1); 308 } 309 } 310 311 pr_debug("map %d+%d\n", map->index, map->count); 312 err = gnttab_map_refs(map->map_ops, use_ptemod ? map->kmap_ops : NULL, 313 map->pages, map->count); 314 if (err) 315 return err; 316 317 for (i = 0; i < map->count; i++) { 318 if (map->map_ops[i].status) { 319 err = -EINVAL; 320 continue; 321 } 322 323 map->unmap_ops[i].handle = map->map_ops[i].handle; 324 if (use_ptemod) 325 map->kunmap_ops[i].handle = map->kmap_ops[i].handle; 326 } 327 return err; 328 } 329 330 static int __unmap_grant_pages(struct grant_map *map, int offset, int pages) 331 { 332 int i, err = 0; 333 struct gntab_unmap_queue_data unmap_data; 334 335 if (map->notify.flags & UNMAP_NOTIFY_CLEAR_BYTE) { 336 int pgno = (map->notify.addr >> PAGE_SHIFT); 337 if (pgno >= offset && pgno < offset + pages) { 338 /* No need for kmap, pages are in lowmem */ 339 uint8_t *tmp = pfn_to_kaddr(page_to_pfn(map->pages[pgno])); 340 tmp[map->notify.addr & (PAGE_SIZE-1)] = 0; 341 map->notify.flags &= ~UNMAP_NOTIFY_CLEAR_BYTE; 342 } 343 } 344 345 unmap_data.unmap_ops = map->unmap_ops + offset; 346 unmap_data.kunmap_ops = use_ptemod ? map->kunmap_ops + offset : NULL; 347 unmap_data.pages = map->pages + offset; 348 unmap_data.count = pages; 349 350 err = gnttab_unmap_refs_sync(&unmap_data); 351 if (err) 352 return err; 353 354 for (i = 0; i < pages; i++) { 355 if (map->unmap_ops[offset+i].status) 356 err = -EINVAL; 357 pr_debug("unmap handle=%d st=%d\n", 358 map->unmap_ops[offset+i].handle, 359 map->unmap_ops[offset+i].status); 360 map->unmap_ops[offset+i].handle = -1; 361 } 362 return err; 363 } 364 365 static int unmap_grant_pages(struct grant_map *map, int offset, int pages) 366 { 367 int range, err = 0; 368 369 pr_debug("unmap %d+%d [%d+%d]\n", map->index, map->count, offset, pages); 370 371 /* It is possible the requested range will have a "hole" where we 372 * already unmapped some of the grants. Only unmap valid ranges. 373 */ 374 while (pages && !err) { 375 while (pages && map->unmap_ops[offset].handle == -1) { 376 offset++; 377 pages--; 378 } 379 range = 0; 380 while (range < pages) { 381 if (map->unmap_ops[offset+range].handle == -1) { 382 range--; 383 break; 384 } 385 range++; 386 } 387 err = __unmap_grant_pages(map, offset, range); 388 offset += range; 389 pages -= range; 390 } 391 392 return err; 393 } 394 395 /* ------------------------------------------------------------------ */ 396 397 static void gntdev_vma_open(struct vm_area_struct *vma) 398 { 399 struct grant_map *map = vma->vm_private_data; 400 401 pr_debug("gntdev_vma_open %p\n", vma); 402 atomic_inc(&map->users); 403 } 404 405 static void gntdev_vma_close(struct vm_area_struct *vma) 406 { 407 struct grant_map *map = vma->vm_private_data; 408 struct file *file = vma->vm_file; 409 struct gntdev_priv *priv = file->private_data; 410 411 pr_debug("gntdev_vma_close %p\n", vma); 412 if (use_ptemod) { 413 /* It is possible that an mmu notifier could be running 414 * concurrently, so take priv->lock to ensure that the vma won't 415 * vanishing during the unmap_grant_pages call, since we will 416 * spin here until that completes. Such a concurrent call will 417 * not do any unmapping, since that has been done prior to 418 * closing the vma, but it may still iterate the unmap_ops list. 419 */ 420 mutex_lock(&priv->lock); 421 map->vma = NULL; 422 mutex_unlock(&priv->lock); 423 } 424 vma->vm_private_data = NULL; 425 gntdev_put_map(priv, map); 426 } 427 428 static struct page *gntdev_vma_find_special_page(struct vm_area_struct *vma, 429 unsigned long addr) 430 { 431 struct grant_map *map = vma->vm_private_data; 432 433 return map->pages[(addr - map->pages_vm_start) >> PAGE_SHIFT]; 434 } 435 436 static struct vm_operations_struct gntdev_vmops = { 437 .open = gntdev_vma_open, 438 .close = gntdev_vma_close, 439 .find_special_page = gntdev_vma_find_special_page, 440 }; 441 442 /* ------------------------------------------------------------------ */ 443 444 static void unmap_if_in_range(struct grant_map *map, 445 unsigned long start, unsigned long end) 446 { 447 unsigned long mstart, mend; 448 int err; 449 450 if (!map->vma) 451 return; 452 if (map->vma->vm_start >= end) 453 return; 454 if (map->vma->vm_end <= start) 455 return; 456 mstart = max(start, map->vma->vm_start); 457 mend = min(end, map->vma->vm_end); 458 pr_debug("map %d+%d (%lx %lx), range %lx %lx, mrange %lx %lx\n", 459 map->index, map->count, 460 map->vma->vm_start, map->vma->vm_end, 461 start, end, mstart, mend); 462 err = unmap_grant_pages(map, 463 (mstart - map->vma->vm_start) >> PAGE_SHIFT, 464 (mend - mstart) >> PAGE_SHIFT); 465 WARN_ON(err); 466 } 467 468 static void mn_invl_range_start(struct mmu_notifier *mn, 469 struct mm_struct *mm, 470 unsigned long start, unsigned long end) 471 { 472 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn); 473 struct grant_map *map; 474 475 mutex_lock(&priv->lock); 476 list_for_each_entry(map, &priv->maps, next) { 477 unmap_if_in_range(map, start, end); 478 } 479 list_for_each_entry(map, &priv->freeable_maps, next) { 480 unmap_if_in_range(map, start, end); 481 } 482 mutex_unlock(&priv->lock); 483 } 484 485 static void mn_invl_page(struct mmu_notifier *mn, 486 struct mm_struct *mm, 487 unsigned long address) 488 { 489 mn_invl_range_start(mn, mm, address, address + PAGE_SIZE); 490 } 491 492 static void mn_release(struct mmu_notifier *mn, 493 struct mm_struct *mm) 494 { 495 struct gntdev_priv *priv = container_of(mn, struct gntdev_priv, mn); 496 struct grant_map *map; 497 int err; 498 499 mutex_lock(&priv->lock); 500 list_for_each_entry(map, &priv->maps, next) { 501 if (!map->vma) 502 continue; 503 pr_debug("map %d+%d (%lx %lx)\n", 504 map->index, map->count, 505 map->vma->vm_start, map->vma->vm_end); 506 err = unmap_grant_pages(map, /* offset */ 0, map->count); 507 WARN_ON(err); 508 } 509 list_for_each_entry(map, &priv->freeable_maps, next) { 510 if (!map->vma) 511 continue; 512 pr_debug("map %d+%d (%lx %lx)\n", 513 map->index, map->count, 514 map->vma->vm_start, map->vma->vm_end); 515 err = unmap_grant_pages(map, /* offset */ 0, map->count); 516 WARN_ON(err); 517 } 518 mutex_unlock(&priv->lock); 519 } 520 521 static struct mmu_notifier_ops gntdev_mmu_ops = { 522 .release = mn_release, 523 .invalidate_page = mn_invl_page, 524 .invalidate_range_start = mn_invl_range_start, 525 }; 526 527 /* ------------------------------------------------------------------ */ 528 529 static int gntdev_open(struct inode *inode, struct file *flip) 530 { 531 struct gntdev_priv *priv; 532 int ret = 0; 533 534 priv = kzalloc(sizeof(*priv), GFP_KERNEL); 535 if (!priv) 536 return -ENOMEM; 537 538 INIT_LIST_HEAD(&priv->maps); 539 INIT_LIST_HEAD(&priv->freeable_maps); 540 mutex_init(&priv->lock); 541 542 if (use_ptemod) { 543 priv->mm = get_task_mm(current); 544 if (!priv->mm) { 545 kfree(priv); 546 return -ENOMEM; 547 } 548 priv->mn.ops = &gntdev_mmu_ops; 549 ret = mmu_notifier_register(&priv->mn, priv->mm); 550 mmput(priv->mm); 551 } 552 553 if (ret) { 554 kfree(priv); 555 return ret; 556 } 557 558 flip->private_data = priv; 559 pr_debug("priv %p\n", priv); 560 561 return 0; 562 } 563 564 static int gntdev_release(struct inode *inode, struct file *flip) 565 { 566 struct gntdev_priv *priv = flip->private_data; 567 struct grant_map *map; 568 569 pr_debug("priv %p\n", priv); 570 571 while (!list_empty(&priv->maps)) { 572 map = list_entry(priv->maps.next, struct grant_map, next); 573 list_del(&map->next); 574 gntdev_put_map(NULL /* already removed */, map); 575 } 576 WARN_ON(!list_empty(&priv->freeable_maps)); 577 578 if (use_ptemod) 579 mmu_notifier_unregister(&priv->mn, priv->mm); 580 kfree(priv); 581 return 0; 582 } 583 584 static long gntdev_ioctl_map_grant_ref(struct gntdev_priv *priv, 585 struct ioctl_gntdev_map_grant_ref __user *u) 586 { 587 struct ioctl_gntdev_map_grant_ref op; 588 struct grant_map *map; 589 int err; 590 591 if (copy_from_user(&op, u, sizeof(op)) != 0) 592 return -EFAULT; 593 pr_debug("priv %p, add %d\n", priv, op.count); 594 if (unlikely(op.count <= 0)) 595 return -EINVAL; 596 597 err = -ENOMEM; 598 map = gntdev_alloc_map(priv, op.count); 599 if (!map) 600 return err; 601 602 if (unlikely(atomic_add_return(op.count, &pages_mapped) > limit)) { 603 pr_debug("can't map: over limit\n"); 604 gntdev_put_map(NULL, map); 605 return err; 606 } 607 608 if (copy_from_user(map->grants, &u->refs, 609 sizeof(map->grants[0]) * op.count) != 0) { 610 gntdev_put_map(NULL, map); 611 return -EFAULT; 612 } 613 614 mutex_lock(&priv->lock); 615 gntdev_add_map(priv, map); 616 op.index = map->index << PAGE_SHIFT; 617 mutex_unlock(&priv->lock); 618 619 if (copy_to_user(u, &op, sizeof(op)) != 0) 620 return -EFAULT; 621 622 return 0; 623 } 624 625 static long gntdev_ioctl_unmap_grant_ref(struct gntdev_priv *priv, 626 struct ioctl_gntdev_unmap_grant_ref __user *u) 627 { 628 struct ioctl_gntdev_unmap_grant_ref op; 629 struct grant_map *map; 630 int err = -ENOENT; 631 632 if (copy_from_user(&op, u, sizeof(op)) != 0) 633 return -EFAULT; 634 pr_debug("priv %p, del %d+%d\n", priv, (int)op.index, (int)op.count); 635 636 mutex_lock(&priv->lock); 637 map = gntdev_find_map_index(priv, op.index >> PAGE_SHIFT, op.count); 638 if (map) { 639 list_del(&map->next); 640 if (populate_freeable_maps) 641 list_add_tail(&map->next, &priv->freeable_maps); 642 err = 0; 643 } 644 mutex_unlock(&priv->lock); 645 if (map) 646 gntdev_put_map(priv, map); 647 return err; 648 } 649 650 static long gntdev_ioctl_get_offset_for_vaddr(struct gntdev_priv *priv, 651 struct ioctl_gntdev_get_offset_for_vaddr __user *u) 652 { 653 struct ioctl_gntdev_get_offset_for_vaddr op; 654 struct vm_area_struct *vma; 655 struct grant_map *map; 656 int rv = -EINVAL; 657 658 if (copy_from_user(&op, u, sizeof(op)) != 0) 659 return -EFAULT; 660 pr_debug("priv %p, offset for vaddr %lx\n", priv, (unsigned long)op.vaddr); 661 662 down_read(¤t->mm->mmap_sem); 663 vma = find_vma(current->mm, op.vaddr); 664 if (!vma || vma->vm_ops != &gntdev_vmops) 665 goto out_unlock; 666 667 map = vma->vm_private_data; 668 if (!map) 669 goto out_unlock; 670 671 op.offset = map->index << PAGE_SHIFT; 672 op.count = map->count; 673 rv = 0; 674 675 out_unlock: 676 up_read(¤t->mm->mmap_sem); 677 678 if (rv == 0 && copy_to_user(u, &op, sizeof(op)) != 0) 679 return -EFAULT; 680 return rv; 681 } 682 683 static long gntdev_ioctl_notify(struct gntdev_priv *priv, void __user *u) 684 { 685 struct ioctl_gntdev_unmap_notify op; 686 struct grant_map *map; 687 int rc; 688 int out_flags; 689 unsigned int out_event; 690 691 if (copy_from_user(&op, u, sizeof(op))) 692 return -EFAULT; 693 694 if (op.action & ~(UNMAP_NOTIFY_CLEAR_BYTE|UNMAP_NOTIFY_SEND_EVENT)) 695 return -EINVAL; 696 697 /* We need to grab a reference to the event channel we are going to use 698 * to send the notify before releasing the reference we may already have 699 * (if someone has called this ioctl twice). This is required so that 700 * it is possible to change the clear_byte part of the notification 701 * without disturbing the event channel part, which may now be the last 702 * reference to that event channel. 703 */ 704 if (op.action & UNMAP_NOTIFY_SEND_EVENT) { 705 if (evtchn_get(op.event_channel_port)) 706 return -EINVAL; 707 } 708 709 out_flags = op.action; 710 out_event = op.event_channel_port; 711 712 mutex_lock(&priv->lock); 713 714 list_for_each_entry(map, &priv->maps, next) { 715 uint64_t begin = map->index << PAGE_SHIFT; 716 uint64_t end = (map->index + map->count) << PAGE_SHIFT; 717 if (op.index >= begin && op.index < end) 718 goto found; 719 } 720 rc = -ENOENT; 721 goto unlock_out; 722 723 found: 724 if ((op.action & UNMAP_NOTIFY_CLEAR_BYTE) && 725 (map->flags & GNTMAP_readonly)) { 726 rc = -EINVAL; 727 goto unlock_out; 728 } 729 730 out_flags = map->notify.flags; 731 out_event = map->notify.event; 732 733 map->notify.flags = op.action; 734 map->notify.addr = op.index - (map->index << PAGE_SHIFT); 735 map->notify.event = op.event_channel_port; 736 737 rc = 0; 738 739 unlock_out: 740 mutex_unlock(&priv->lock); 741 742 /* Drop the reference to the event channel we did not save in the map */ 743 if (out_flags & UNMAP_NOTIFY_SEND_EVENT) 744 evtchn_put(out_event); 745 746 return rc; 747 } 748 749 static long gntdev_ioctl(struct file *flip, 750 unsigned int cmd, unsigned long arg) 751 { 752 struct gntdev_priv *priv = flip->private_data; 753 void __user *ptr = (void __user *)arg; 754 755 switch (cmd) { 756 case IOCTL_GNTDEV_MAP_GRANT_REF: 757 return gntdev_ioctl_map_grant_ref(priv, ptr); 758 759 case IOCTL_GNTDEV_UNMAP_GRANT_REF: 760 return gntdev_ioctl_unmap_grant_ref(priv, ptr); 761 762 case IOCTL_GNTDEV_GET_OFFSET_FOR_VADDR: 763 return gntdev_ioctl_get_offset_for_vaddr(priv, ptr); 764 765 case IOCTL_GNTDEV_SET_UNMAP_NOTIFY: 766 return gntdev_ioctl_notify(priv, ptr); 767 768 default: 769 pr_debug("priv %p, unknown cmd %x\n", priv, cmd); 770 return -ENOIOCTLCMD; 771 } 772 773 return 0; 774 } 775 776 static int gntdev_mmap(struct file *flip, struct vm_area_struct *vma) 777 { 778 struct gntdev_priv *priv = flip->private_data; 779 int index = vma->vm_pgoff; 780 int count = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; 781 struct grant_map *map; 782 int i, err = -EINVAL; 783 784 if ((vma->vm_flags & VM_WRITE) && !(vma->vm_flags & VM_SHARED)) 785 return -EINVAL; 786 787 pr_debug("map %d+%d at %lx (pgoff %lx)\n", 788 index, count, vma->vm_start, vma->vm_pgoff); 789 790 mutex_lock(&priv->lock); 791 map = gntdev_find_map_index(priv, index, count); 792 if (!map) 793 goto unlock_out; 794 if (use_ptemod && map->vma) 795 goto unlock_out; 796 if (use_ptemod && priv->mm != vma->vm_mm) { 797 pr_warn("Huh? Other mm?\n"); 798 goto unlock_out; 799 } 800 801 atomic_inc(&map->users); 802 803 vma->vm_ops = &gntdev_vmops; 804 805 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP; 806 807 if (use_ptemod) 808 vma->vm_flags |= VM_DONTCOPY; 809 810 vma->vm_private_data = map; 811 812 if (use_ptemod) 813 map->vma = vma; 814 815 if (map->flags) { 816 if ((vma->vm_flags & VM_WRITE) && 817 (map->flags & GNTMAP_readonly)) 818 goto out_unlock_put; 819 } else { 820 map->flags = GNTMAP_host_map; 821 if (!(vma->vm_flags & VM_WRITE)) 822 map->flags |= GNTMAP_readonly; 823 } 824 825 mutex_unlock(&priv->lock); 826 827 if (use_ptemod) { 828 err = apply_to_page_range(vma->vm_mm, vma->vm_start, 829 vma->vm_end - vma->vm_start, 830 find_grant_ptes, map); 831 if (err) { 832 pr_warn("find_grant_ptes() failure.\n"); 833 goto out_put_map; 834 } 835 } 836 837 err = map_grant_pages(map); 838 if (err) 839 goto out_put_map; 840 841 if (!use_ptemod) { 842 for (i = 0; i < count; i++) { 843 err = vm_insert_page(vma, vma->vm_start + i*PAGE_SIZE, 844 map->pages[i]); 845 if (err) 846 goto out_put_map; 847 } 848 } else { 849 #ifdef CONFIG_X86 850 /* 851 * If the PTEs were not made special by the grant map 852 * hypercall, do so here. 853 * 854 * This is racy since the mapping is already visible 855 * to userspace but userspace should be well-behaved 856 * enough to not touch it until the mmap() call 857 * returns. 858 */ 859 if (!xen_feature(XENFEAT_gnttab_map_avail_bits)) { 860 apply_to_page_range(vma->vm_mm, vma->vm_start, 861 vma->vm_end - vma->vm_start, 862 set_grant_ptes_as_special, NULL); 863 } 864 #endif 865 map->pages_vm_start = vma->vm_start; 866 } 867 868 return 0; 869 870 unlock_out: 871 mutex_unlock(&priv->lock); 872 return err; 873 874 out_unlock_put: 875 mutex_unlock(&priv->lock); 876 out_put_map: 877 if (use_ptemod) 878 map->vma = NULL; 879 gntdev_put_map(priv, map); 880 return err; 881 } 882 883 static const struct file_operations gntdev_fops = { 884 .owner = THIS_MODULE, 885 .open = gntdev_open, 886 .release = gntdev_release, 887 .mmap = gntdev_mmap, 888 .unlocked_ioctl = gntdev_ioctl 889 }; 890 891 static struct miscdevice gntdev_miscdev = { 892 .minor = MISC_DYNAMIC_MINOR, 893 .name = "xen/gntdev", 894 .fops = &gntdev_fops, 895 }; 896 897 /* ------------------------------------------------------------------ */ 898 899 static int __init gntdev_init(void) 900 { 901 int err; 902 903 if (!xen_domain()) 904 return -ENODEV; 905 906 use_ptemod = !xen_feature(XENFEAT_auto_translated_physmap); 907 908 err = misc_register(&gntdev_miscdev); 909 if (err != 0) { 910 pr_err("Could not register gntdev device\n"); 911 return err; 912 } 913 return 0; 914 } 915 916 static void __exit gntdev_exit(void) 917 { 918 misc_deregister(&gntdev_miscdev); 919 } 920 921 module_init(gntdev_init); 922 module_exit(gntdev_exit); 923 924 /* ------------------------------------------------------------------ */ 925