drivers/xen/privcmd.c

   1 /******************************************************************************
   2  * privcmd.c
   3  *
   4  * Interface to privileged domain-0 commands.
   5  *
   6  * Copyright (c) 2002-2004, K A Fraser, B Dragovic
   7  */
   8
   9 #define pr_fmt(fmt) "xen:" KBUILD_MODNAME ": " fmt
  10
  11 #include <linux/kernel.h>
  12 #include <linux/module.h>
  13 #include <linux/sched.h>
  14 #include <linux/slab.h>
  15 #include <linux/string.h>
  16 #include <linux/errno.h>
  17 #include <linux/mm.h>
  18 #include <linux/mman.h>
  19 #include <linux/uaccess.h>
  20 #include <linux/swap.h>
  21 #include <linux/highmem.h>
  22 #include <linux/pagemap.h>
  23 #include <linux/seq_file.h>
  24 #include <linux/miscdevice.h>
  25 #include <linux/moduleparam.h>
  26
  27 #include <asm/pgalloc.h>
  28 #include <asm/pgtable.h>
  29 #include <asm/tlb.h>
  30 #include <asm/xen/hypervisor.h>
  31 #include <asm/xen/hypercall.h>
  32
  33 #include <xen/xen.h>
  34 #include <xen/privcmd.h>
  35 #include <xen/interface/xen.h>
  36 #include <xen/interface/hvm/dm_op.h>
  37 #include <xen/features.h>
  38 #include <xen/page.h>
  39 #include <xen/xen-ops.h>
  40 #include <xen/balloon.h>
  41
  42 #include "privcmd.h"
  43
  44 MODULE_LICENSE("GPL");
  45
  46 #define PRIV_VMA_LOCKED ((void *)1)
  47
  48 static unsigned int privcmd_dm_op_max_num = 16;
  49 module_param_named(dm_op_max_nr_bufs, privcmd_dm_op_max_num, uint, 0644);
  50 MODULE_PARM_DESC(dm_op_max_nr_bufs,
  51                  "Maximum number of buffers per dm_op hypercall");
  52
  53 static unsigned int privcmd_dm_op_buf_max_size = 4096;
  54 module_param_named(dm_op_buf_max_size, privcmd_dm_op_buf_max_size, uint,
  55                    0644);
  56 MODULE_PARM_DESC(dm_op_buf_max_size,
  57                  "Maximum size of a dm_op hypercall buffer");
  58
  59 struct privcmd_data {
  60         domid_t domid;
  61 };
  62
  63 static int privcmd_vma_range_is_mapped(
  64                struct vm_area_struct *vma,
  65                unsigned long addr,
  66                unsigned long nr_pages);
  67
  68 static long privcmd_ioctl_hypercall(struct file *file, void __user *udata)
  69 {
  70         struct privcmd_data *data = file->private_data;
  71         struct privcmd_hypercall hypercall;
  72         long ret;
  73
  74         /* Disallow arbitrary hypercalls if restricted */
  75         if (data->domid != DOMID_INVALID)
  76                 return -EPERM;
  77
  78         if (copy_from_user(&hypercall, udata, sizeof(hypercall)))
  79                 return -EFAULT;
  80
  81         xen_preemptible_hcall_begin();
  82         ret = privcmd_call(hypercall.op,
  83                            hypercall.arg[0], hypercall.arg[1],
  84                            hypercall.arg[2], hypercall.arg[3],
  85                            hypercall.arg[4]);
  86         xen_preemptible_hcall_end();
  87
  88         return ret;
  89 }
  90
  91 static void free_page_list(struct list_head *pages)
  92 {
  93         struct page *p, *n;
  94
  95         list_for_each_entry_safe(p, n, pages, lru)
  96                 __free_page(p);
  97
  98         INIT_LIST_HEAD(pages);
  99 }
 100
 101 /*
 102  * Given an array of items in userspace, return a list of pages
 103  * containing the data.  If copying fails, either because of memory
 104  * allocation failure or a problem reading user memory, return an
 105  * error code; its up to the caller to dispose of any partial list.
 106  */
 107 static int gather_array(struct list_head *pagelist,
 108                         unsigned nelem, size_t size,
 109                         const void __user *data)
 110 {
 111         unsigned pageidx;
 112         void *pagedata;
 113         int ret;
 114
 115         if (size > PAGE_SIZE)
 116                 return 0;
 117
 118         pageidx = PAGE_SIZE;
 119         pagedata = NULL;        /* quiet, gcc */
 120         while (nelem--) {
 121                 if (pageidx > PAGE_SIZE-size) {
 122                         struct page *page = alloc_page(GFP_KERNEL);
 123
 124                         ret = -ENOMEM;
 125                         if (page == NULL)
 126                                 goto fail;
 127
 128                         pagedata = page_address(page);
 129
 130                         list_add_tail(&page->lru, pagelist);
 131                         pageidx = 0;
 132                 }
 133
 134                 ret = -EFAULT;
 135                 if (copy_from_user(pagedata + pageidx, data, size))
 136                         goto fail;
 137
 138                 data += size;
 139                 pageidx += size;
 140         }
 141
 142         ret = 0;
 143
 144 fail:
 145         return ret;
 146 }
 147
 148 /*
 149  * Call function "fn" on each element of the array fragmented
 150  * over a list of pages.
 151  */
 152 static int traverse_pages(unsigned nelem, size_t size,
 153                           struct list_head *pos,
 154                           int (*fn)(void *data, void *state),
 155                           void *state)
 156 {
 157         void *pagedata;
 158         unsigned pageidx;
 159         int ret = 0;
 160
 161         BUG_ON(size > PAGE_SIZE);
 162
 163         pageidx = PAGE_SIZE;
 164         pagedata = NULL;        /* hush, gcc */
 165
 166         while (nelem--) {
 167                 if (pageidx > PAGE_SIZE-size) {
 168                         struct page *page;
 169                         pos = pos->next;
 170                         page = list_entry(pos, struct page, lru);
 171                         pagedata = page_address(page);
 172                         pageidx = 0;
 173                 }
 174
 175                 ret = (*fn)(pagedata + pageidx, state);
 176                 if (ret)
 177                         break;
 178                 pageidx += size;
 179         }
 180
 181         return ret;
 182 }
 183
 184 /*
 185  * Similar to traverse_pages, but use each page as a "block" of
 186  * data to be processed as one unit.
 187  */
 188 static int traverse_pages_block(unsigned nelem, size_t size,
 189                                 struct list_head *pos,
 190                                 int (*fn)(void *data, int nr, void *state),
 191                                 void *state)
 192 {
 193         void *pagedata;
 194         unsigned pageidx;
 195         int ret = 0;
 196
 197         BUG_ON(size > PAGE_SIZE);
 198
 199         pageidx = PAGE_SIZE;
 200
 201         while (nelem) {
 202                 int nr = (PAGE_SIZE/size);
 203                 struct page *page;
 204                 if (nr > nelem)
 205                         nr = nelem;
 206                 pos = pos->next;
 207                 page = list_entry(pos, struct page, lru);
 208                 pagedata = page_address(page);
 209                 ret = (*fn)(pagedata, nr, state);
 210                 if (ret)
 211                         break;
 212                 nelem -= nr;
 213         }
 214
 215         return ret;
 216 }
 217
 218 struct mmap_gfn_state {
 219         unsigned long va;
 220         struct vm_area_struct *vma;
 221         domid_t domain;
 222 };
 223
 224 static int mmap_gfn_range(void *data, void *state)
 225 {
 226         struct privcmd_mmap_entry *msg = data;
 227         struct mmap_gfn_state *st = state;
 228         struct vm_area_struct *vma = st->vma;
 229         int rc;
 230
 231         /* Do not allow range to wrap the address space. */
 232         if ((msg->npages > (LONG_MAX >> PAGE_SHIFT)) ||
 233             ((unsigned long)(msg->npages << PAGE_SHIFT) >= -st->va))
 234                 return -EINVAL;
 235
 236         /* Range chunks must be contiguous in va space. */
 237         if ((msg->va != st->va) ||
 238             ((msg->va+(msg->npages<<PAGE_SHIFT)) > vma->vm_end))
 239                 return -EINVAL;
 240
 241         rc = xen_remap_domain_gfn_range(vma,
 242                                         msg->va & PAGE_MASK,
 243                                         msg->mfn, msg->npages,
 244                                         vma->vm_page_prot,
 245                                         st->domain, NULL);
 246         if (rc < 0)
 247                 return rc;
 248
 249         st->va += msg->npages << PAGE_SHIFT;
 250
 251         return 0;
 252 }
 253
 254 static long privcmd_ioctl_mmap(struct file *file, void __user *udata)
 255 {
 256         struct privcmd_data *data = file->private_data;
 257         struct privcmd_mmap mmapcmd;
 258         struct mm_struct *mm = current->mm;
 259         struct vm_area_struct *vma;
 260         int rc;
 261         LIST_HEAD(pagelist);
 262         struct mmap_gfn_state state;
 263
 264         /* We only support privcmd_ioctl_mmap_batch for auto translated. */
 265         if (xen_feature(XENFEAT_auto_translated_physmap))
 266                 return -ENOSYS;
 267
 268         if (copy_from_user(&mmapcmd, udata, sizeof(mmapcmd)))
 269                 return -EFAULT;
 270
 271         /* If restriction is in place, check the domid matches */
 272         if (data->domid != DOMID_INVALID && data->domid != mmapcmd.dom)
 273                 return -EPERM;
 274
 275         rc = gather_array(&pagelist,
 276                           mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 277                           mmapcmd.entry);
 278
 279         if (rc || list_empty(&pagelist))
 280                 goto out;
 281
 282         down_write(&mm->mmap_sem);
 283
 284         {
 285                 struct page *page = list_first_entry(&pagelist,
 286                                                      struct page, lru);
 287                 struct privcmd_mmap_entry *msg = page_address(page);
 288
 289                 vma = find_vma(mm, msg->va);
 290                 rc = -EINVAL;
 291
 292                 if (!vma || (msg->va != vma->vm_start) || vma->vm_private_data)
 293                         goto out_up;
 294                 vma->vm_private_data = PRIV_VMA_LOCKED;
 295         }
 296
 297         state.va = vma->vm_start;
 298         state.vma = vma;
 299         state.domain = mmapcmd.dom;
 300
 301         rc = traverse_pages(mmapcmd.num, sizeof(struct privcmd_mmap_entry),
 302                             &pagelist,
 303                             mmap_gfn_range, &state);
 304
 305
 306 out_up:
 307         up_write(&mm->mmap_sem);
 308
 309 out:
 310         free_page_list(&pagelist);
 311
 312         return rc;
 313 }
 314
 315 struct mmap_batch_state {
 316         domid_t domain;
 317         unsigned long va;
 318         struct vm_area_struct *vma;
 319         int index;
 320         /* A tristate:
 321          *      0 for no errors
 322          *      1 if at least one error has happened (and no
 323          *          -ENOENT errors have happened)
 324          *      -ENOENT if at least 1 -ENOENT has happened.
 325          */
 326         int global_error;
 327         int version;
 328
 329         /* User-space gfn array to store errors in the second pass for V1. */
 330         xen_pfn_t __user *user_gfn;
 331         /* User-space int array to store errors in the second pass for V2. */
 332         int __user *user_err;
 333 };
 334
 335 /* auto translated dom0 note: if domU being created is PV, then gfn is
 336  * mfn(addr on bus). If it's auto xlated, then gfn is pfn (input to HAP).
 337  */
 338 static int mmap_batch_fn(void *data, int nr, void *state)
 339 {
 340         xen_pfn_t *gfnp = data;
 341         struct mmap_batch_state *st = state;
 342         struct vm_area_struct *vma = st->vma;
 343         struct page **pages = vma->vm_private_data;
 344         struct page **cur_pages = NULL;
 345         int ret;
 346
 347         if (xen_feature(XENFEAT_auto_translated_physmap))
 348                 cur_pages = &pages[st->index];
 349
 350         BUG_ON(nr < 0);
 351         ret = xen_remap_domain_gfn_array(st->vma, st->va & PAGE_MASK, gfnp, nr,
 352                                          (int *)gfnp, st->vma->vm_page_prot,
 353                                          st->domain, cur_pages);
 354
 355         /* Adjust the global_error? */
 356         if (ret != nr) {
 357                 if (ret == -ENOENT)
 358                         st->global_error = -ENOENT;
 359                 else {
 360                         /* Record that at least one error has happened. */
 361                         if (st->global_error == 0)
 362                                 st->global_error = 1;
 363                 }
 364         }
 365         st->va += XEN_PAGE_SIZE * nr;
 366         st->index += nr / XEN_PFN_PER_PAGE;
 367
 368         return 0;
 369 }
 370
 371 static int mmap_return_error(int err, struct mmap_batch_state *st)
 372 {
 373         int ret;
 374
 375         if (st->version == 1) {
 376                 if (err) {
 377                         xen_pfn_t gfn;
 378
 379                         ret = get_user(gfn, st->user_gfn);
 380                         if (ret < 0)
 381                                 return ret;
 382                         /*
 383                          * V1 encodes the error codes in the 32bit top
 384                          * nibble of the gfn (with its known
 385                          * limitations vis-a-vis 64 bit callers).
 386                          */
 387                         gfn |= (err == -ENOENT) ?
 388                                 PRIVCMD_MMAPBATCH_PAGED_ERROR :
 389                                 PRIVCMD_MMAPBATCH_MFN_ERROR;
 390                         return __put_user(gfn, st->user_gfn++);
 391                 } else
 392                         st->user_gfn++;
 393         } else { /* st->version == 2 */
 394                 if (err)
 395                         return __put_user(err, st->user_err++);
 396                 else
 397                         st->user_err++;
 398         }
 399
 400         return 0;
 401 }
 402
 403 static int mmap_return_errors(void *data, int nr, void *state)
 404 {
 405         struct mmap_batch_state *st = state;
 406         int *errs = data;
 407         int i;
 408         int ret;
 409
 410         for (i = 0; i < nr; i++) {
 411                 ret = mmap_return_error(errs[i], st);
 412                 if (ret < 0)
 413                         return ret;
 414         }
 415         return 0;
 416 }
 417
 418 /* Allocate pfns that are then mapped with gfns from foreign domid. Update
 419  * the vma with the page info to use later.
 420  * Returns: 0 if success, otherwise -errno
 421  */
 422 static int alloc_empty_pages(struct vm_area_struct *vma, int numpgs)
 423 {
 424         int rc;
 425         struct page **pages;
 426
 427         pages = kcalloc(numpgs, sizeof(pages[0]), GFP_KERNEL);
 428         if (pages == NULL)
 429                 return -ENOMEM;
 430
 431         rc = alloc_xenballooned_pages(numpgs, pages);
 432         if (rc != 0) {
 433                 pr_warn("%s Could not alloc %d pfns rc:%d\n", __func__,
 434                         numpgs, rc);
 435                 kfree(pages);
 436                 return -ENOMEM;
 437         }
 438         BUG_ON(vma->vm_private_data != NULL);
 439         vma->vm_private_data = pages;
 440
 441         return 0;
 442 }
 443
 444 static const struct vm_operations_struct privcmd_vm_ops;
 445
 446 static long privcmd_ioctl_mmap_batch(
 447         struct file *file, void __user *udata, int version)
 448 {
 449         struct privcmd_data *data = file->private_data;
 450         int ret;
 451         struct privcmd_mmapbatch_v2 m;
 452         struct mm_struct *mm = current->mm;
 453         struct vm_area_struct *vma;
 454         unsigned long nr_pages;
 455         LIST_HEAD(pagelist);
 456         struct mmap_batch_state state;
 457
 458         switch (version) {
 459         case 1:
 460                 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch)))
 461                         return -EFAULT;
 462                 /* Returns per-frame error in m.arr. */
 463                 m.err = NULL;
 464                 if (!access_ok(VERIFY_WRITE, m.arr, m.num * sizeof(*m.arr)))
 465                         return -EFAULT;
 466                 break;
 467         case 2:
 468                 if (copy_from_user(&m, udata, sizeof(struct privcmd_mmapbatch_v2)))
 469                         return -EFAULT;
 470                 /* Returns per-frame error code in m.err. */
 471                 if (!access_ok(VERIFY_WRITE, m.err, m.num * (sizeof(*m.err))))
 472                         return -EFAULT;
 473                 break;
 474         default:
 475                 return -EINVAL;
 476         }
 477
 478         /* If restriction is in place, check the domid matches */
 479         if (data->domid != DOMID_INVALID && data->domid != m.dom)
 480                 return -EPERM;
 481
 482         nr_pages = DIV_ROUND_UP(m.num, XEN_PFN_PER_PAGE);
 483         if ((m.num <= 0) || (nr_pages > (LONG_MAX >> PAGE_SHIFT)))
 484                 return -EINVAL;
 485
 486         ret = gather_array(&pagelist, m.num, sizeof(xen_pfn_t), m.arr);
 487
 488         if (ret)
 489                 goto out;
 490         if (list_empty(&pagelist)) {
 491                 ret = -EINVAL;
 492                 goto out;
 493         }
 494
 495         if (version == 2) {
 496                 /* Zero error array now to only copy back actual errors. */
 497                 if (clear_user(m.err, sizeof(int) * m.num)) {
 498                         ret = -EFAULT;
 499                         goto out;
 500                 }
 501         }
 502
 503         down_write(&mm->mmap_sem);
 504
 505         vma = find_vma(mm, m.addr);
 506         if (!vma ||
 507             vma->vm_ops != &privcmd_vm_ops) {
 508                 ret = -EINVAL;
 509                 goto out_unlock;
 510         }
 511
 512         /*
 513          * Caller must either:
 514          *
 515          * Map the whole VMA range, which will also allocate all the
 516          * pages required for the auto_translated_physmap case.
 517          *
 518          * Or
 519          *
 520          * Map unmapped holes left from a previous map attempt (e.g.,
 521          * because those foreign frames were previously paged out).
 522          */
 523         if (vma->vm_private_data == NULL) {
 524                 if (m.addr != vma->vm_start ||
 525                     m.addr + (nr_pages << PAGE_SHIFT) != vma->vm_end) {
 526                         ret = -EINVAL;
 527                         goto out_unlock;
 528                 }
 529                 if (xen_feature(XENFEAT_auto_translated_physmap)) {
 530                         ret = alloc_empty_pages(vma, nr_pages);
 531                         if (ret < 0)
 532                                 goto out_unlock;
 533                 } else
 534                         vma->vm_private_data = PRIV_VMA_LOCKED;
 535         } else {
 536                 if (m.addr < vma->vm_start ||
 537                     m.addr + (nr_pages << PAGE_SHIFT) > vma->vm_end) {
 538                         ret = -EINVAL;
 539                         goto out_unlock;
 540                 }
 541                 if (privcmd_vma_range_is_mapped(vma, m.addr, nr_pages)) {
 542                         ret = -EINVAL;
 543                         goto out_unlock;
 544                 }
 545         }
 546
 547         state.domain        = m.dom;
 548         state.vma           = vma;
 549         state.va            = m.addr;
 550         state.index         = 0;
 551         state.global_error  = 0;
 552         state.version       = version;
 553
 554         BUILD_BUG_ON(((PAGE_SIZE / sizeof(xen_pfn_t)) % XEN_PFN_PER_PAGE) != 0);
 555         /* mmap_batch_fn guarantees ret == 0 */
 556         BUG_ON(traverse_pages_block(m.num, sizeof(xen_pfn_t),
 557                                     &pagelist, mmap_batch_fn, &state));
 558
 559         up_write(&mm->mmap_sem);
 560
 561         if (state.global_error) {
 562                 /* Write back errors in second pass. */
 563                 state.user_gfn = (xen_pfn_t *)m.arr;
 564                 state.user_err = m.err;
 565                 ret = traverse_pages_block(m.num, sizeof(xen_pfn_t),
 566                                            &pagelist, mmap_return_errors, &state);
 567         } else
 568                 ret = 0;
 569
 570         /* If we have not had any EFAULT-like global errors then set the global
 571          * error to -ENOENT if necessary. */
 572         if ((ret == 0) && (state.global_error == -ENOENT))
 573                 ret = -ENOENT;
 574
 575 out:
 576         free_page_list(&pagelist);
 577         return ret;
 578
 579 out_unlock:
 580         up_write(&mm->mmap_sem);
 581         goto out;
 582 }
 583
 584 static int lock_pages(
 585         struct privcmd_dm_op_buf kbufs[], unsigned int num,
 586         struct page *pages[], unsigned int nr_pages)
 587 {
 588         unsigned int i;
 589
 590         for (i = 0; i < num; i++) {
 591                 unsigned int requested;
 592                 int pinned;
 593
 594                 requested = DIV_ROUND_UP(
 595                         offset_in_page(kbufs[i].uptr) + kbufs[i].size,
 596                         PAGE_SIZE);
 597                 if (requested > nr_pages)
 598                         return -ENOSPC;
 599
 600                 pinned = get_user_pages_fast(
 601                         (unsigned long) kbufs[i].uptr,
 602                         requested, FOLL_WRITE, pages);
 603                 if (pinned < 0)
 604                         return pinned;
 605
 606                 nr_pages -= pinned;
 607                 pages += pinned;
 608         }
 609
 610         return 0;
 611 }
 612
 613 static void unlock_pages(struct page *pages[], unsigned int nr_pages)
 614 {
 615         unsigned int i;
 616
 617         if (!pages)
 618                 return;
 619
 620         for (i = 0; i < nr_pages; i++) {
 621                 if (pages[i])
 622                         put_page(pages[i]);
 623         }
 624 }
 625
 626 static long privcmd_ioctl_dm_op(struct file *file, void __user *udata)
 627 {
 628         struct privcmd_data *data = file->private_data;
 629         struct privcmd_dm_op kdata;
 630         struct privcmd_dm_op_buf *kbufs;
 631         unsigned int nr_pages = 0;
 632         struct page **pages = NULL;
 633         struct xen_dm_op_buf *xbufs = NULL;
 634         unsigned int i;
 635         long rc;
 636
 637         if (copy_from_user(&kdata, udata, sizeof(kdata)))
 638                 return -EFAULT;
 639
 640         /* If restriction is in place, check the domid matches */
 641         if (data->domid != DOMID_INVALID && data->domid != kdata.dom)
 642                 return -EPERM;
 643
 644         if (kdata.num == 0)
 645                 return 0;
 646
 647         if (kdata.num > privcmd_dm_op_max_num)
 648                 return -E2BIG;
 649
 650         kbufs = kcalloc(kdata.num, sizeof(*kbufs), GFP_KERNEL);
 651         if (!kbufs)
 652                 return -ENOMEM;
 653
 654         if (copy_from_user(kbufs, kdata.ubufs,
 655                            sizeof(*kbufs) * kdata.num)) {
 656                 rc = -EFAULT;
 657                 goto out;
 658         }
 659
 660         for (i = 0; i < kdata.num; i++) {
 661                 if (kbufs[i].size > privcmd_dm_op_buf_max_size) {
 662                         rc = -E2BIG;
 663                         goto out;
 664                 }
 665
 666                 if (!access_ok(VERIFY_WRITE, kbufs[i].uptr,
 667                                kbufs[i].size)) {
 668                         rc = -EFAULT;
 669                         goto out;
 670                 }
 671
 672                 nr_pages += DIV_ROUND_UP(
 673                         offset_in_page(kbufs[i].uptr) + kbufs[i].size,
 674                         PAGE_SIZE);
 675         }
 676
 677         pages = kcalloc(nr_pages, sizeof(*pages), GFP_KERNEL);
 678         if (!pages) {
 679                 rc = -ENOMEM;
 680                 goto out;
 681         }
 682
 683         xbufs = kcalloc(kdata.num, sizeof(*xbufs), GFP_KERNEL);
 684         if (!xbufs) {
 685                 rc = -ENOMEM;
 686                 goto out;
 687         }
 688
 689         rc = lock_pages(kbufs, kdata.num, pages, nr_pages);
 690         if (rc)
 691                 goto out;
 692
 693         for (i = 0; i < kdata.num; i++) {
 694                 set_xen_guest_handle(xbufs[i].h, kbufs[i].uptr);
 695                 xbufs[i].size = kbufs[i].size;
 696         }
 697
 698         xen_preemptible_hcall_begin();
 699         rc = HYPERVISOR_dm_op(kdata.dom, kdata.num, xbufs);
 700         xen_preemptible_hcall_end();
 701
 702 out:
 703         unlock_pages(pages, nr_pages);
 704         kfree(xbufs);
 705         kfree(pages);
 706         kfree(kbufs);
 707
 708         return rc;
 709 }
 710
 711 static long privcmd_ioctl_restrict(struct file *file, void __user *udata)
 712 {
 713         struct privcmd_data *data = file->private_data;
 714         domid_t dom;
 715
 716         if (copy_from_user(&dom, udata, sizeof(dom)))
 717                 return -EFAULT;
 718
 719         /* Set restriction to the specified domain, or check it matches */
 720         if (data->domid == DOMID_INVALID)
 721                 data->domid = dom;
 722         else if (data->domid != dom)
 723                 return -EINVAL;
 724
 725         return 0;
 726 }
 727
 728 static long privcmd_ioctl(struct file *file,
 729                           unsigned int cmd, unsigned long data)
 730 {
 731         int ret = -ENOTTY;
 732         void __user *udata = (void __user *) data;
 733
 734         switch (cmd) {
 735         case IOCTL_PRIVCMD_HYPERCALL:
 736                 ret = privcmd_ioctl_hypercall(file, udata);
 737                 break;
 738
 739         case IOCTL_PRIVCMD_MMAP:
 740                 ret = privcmd_ioctl_mmap(file, udata);
 741                 break;
 742
 743         case IOCTL_PRIVCMD_MMAPBATCH:
 744                 ret = privcmd_ioctl_mmap_batch(file, udata, 1);
 745                 break;
 746
 747         case IOCTL_PRIVCMD_MMAPBATCH_V2:
 748                 ret = privcmd_ioctl_mmap_batch(file, udata, 2);
 749                 break;
 750
 751         case IOCTL_PRIVCMD_DM_OP:
 752                 ret = privcmd_ioctl_dm_op(file, udata);
 753                 break;
 754
 755         case IOCTL_PRIVCMD_RESTRICT:
 756                 ret = privcmd_ioctl_restrict(file, udata);
 757                 break;
 758
 759         default:
 760                 break;
 761         }
 762
 763         return ret;
 764 }
 765
 766 static int privcmd_open(struct inode *ino, struct file *file)
 767 {
 768         struct privcmd_data *data = kzalloc(sizeof(*data), GFP_KERNEL);
 769
 770         if (!data)
 771                 return -ENOMEM;
 772
 773         /* DOMID_INVALID implies no restriction */
 774         data->domid = DOMID_INVALID;
 775
 776         file->private_data = data;
 777         return 0;
 778 }
 779
 780 static int privcmd_release(struct inode *ino, struct file *file)
 781 {
 782         struct privcmd_data *data = file->private_data;
 783
 784         kfree(data);
 785         return 0;
 786 }
 787
 788 static void privcmd_close(struct vm_area_struct *vma)
 789 {
 790         struct page **pages = vma->vm_private_data;
 791         int numpgs = vma_pages(vma);
 792         int numgfns = (vma->vm_end - vma->vm_start) >> XEN_PAGE_SHIFT;
 793         int rc;
 794
 795         if (!xen_feature(XENFEAT_auto_translated_physmap) || !numpgs || !pages)
 796                 return;
 797
 798         rc = xen_unmap_domain_gfn_range(vma, numgfns, pages);
 799         if (rc == 0)
 800                 free_xenballooned_pages(numpgs, pages);
 801         else
 802                 pr_crit("unable to unmap MFN range: leaking %d pages. rc=%d\n",
 803                         numpgs, rc);
 804         kfree(pages);
 805 }
 806
 807 static int privcmd_fault(struct vm_fault *vmf)
 808 {
 809         printk(KERN_DEBUG "privcmd_fault: vma=%p %lx-%lx, pgoff=%lx, uv=%p\n",
 810                vmf->vma, vmf->vma->vm_start, vmf->vma->vm_end,
 811                vmf->pgoff, (void *)vmf->address);
 812
 813         return VM_FAULT_SIGBUS;
 814 }
 815
 816 static const struct vm_operations_struct privcmd_vm_ops = {
 817         .close = privcmd_close,
 818         .fault = privcmd_fault
 819 };
 820
 821 static int privcmd_mmap(struct file *file, struct vm_area_struct *vma)
 822 {
 823         /* DONTCOPY is essential for Xen because copy_page_range doesn't know
 824          * how to recreate these mappings */
 825         vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTCOPY |
 826                          VM_DONTEXPAND | VM_DONTDUMP;
 827         vma->vm_ops = &privcmd_vm_ops;
 828         vma->vm_private_data = NULL;
 829
 830         return 0;
 831 }
 832
 833 /*
 834  * For MMAPBATCH*. This allows asserting the singleshot mapping
 835  * on a per pfn/pte basis. Mapping calls that fail with ENOENT
 836  * can be then retried until success.
 837  */
 838 static int is_mapped_fn(pte_t *pte, struct page *pmd_page,
 839                         unsigned long addr, void *data)
 840 {
 841         return pte_none(*pte) ? 0 : -EBUSY;
 842 }
 843
 844 static int privcmd_vma_range_is_mapped(
 845                    struct vm_area_struct *vma,
 846                    unsigned long addr,
 847                    unsigned long nr_pages)
 848 {
 849         return apply_to_page_range(vma->vm_mm, addr, nr_pages << PAGE_SHIFT,
 850                                    is_mapped_fn, NULL) != 0;
 851 }
 852
 853 const struct file_operations xen_privcmd_fops = {
 854         .owner = THIS_MODULE,
 855         .unlocked_ioctl = privcmd_ioctl,
 856         .open = privcmd_open,
 857         .release = privcmd_release,
 858         .mmap = privcmd_mmap,
 859 };
 860 EXPORT_SYMBOL_GPL(xen_privcmd_fops);
 861
 862 static struct miscdevice privcmd_dev = {
 863         .minor = MISC_DYNAMIC_MINOR,
 864         .name = "xen/privcmd",
 865         .fops = &xen_privcmd_fops,
 866 };
 867
 868 static int __init privcmd_init(void)
 869 {
 870         int err;
 871
 872         if (!xen_domain())
 873                 return -ENODEV;
 874
 875         err = misc_register(&privcmd_dev);
 876         if (err != 0) {
 877                 pr_err("Could not register Xen privcmd device\n");
 878                 return err;
 879         }
 880         return 0;
 881 }
 882
 883 static void __exit privcmd_exit(void)
 884 {
 885         misc_deregister(&privcmd_dev);
 886 }
 887
 888 module_init(privcmd_init);
 889 module_exit(privcmd_exit);